d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
06786ccfb3
freebsd-skq/libexec/rtld-elf/rtld.c

3713 lines
96 KiB
C
Raw Normal View History

Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*-
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
* Copyright 1996, 1997, 1998, 1999, 2000 John D. Polstra.
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
* Copyright 2003 Alexander Kabaev <kan@FreeBSD.ORG>.
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
$Id$ -> $FreeBSD$
1999-08-28 00:22:10 +00:00
* $FreeBSD$
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
*/
/*
* Dynamic linker for ELF.
*
* John Polstra <jdp@polstra.com>.
*/
#ifndef __GNUC__
#error "GCC is needed to compile this file"
#endif
#include <sys/param.h>
If "dangerous" environment variables (LD_PRELOAD, LD_LIBMAP, LD_LIBMAP_DISABLE, LD_LIBRARY_PATH) are used, then make sure the libraries being loaded aren't on a noexec-mounted filesystem. This is a compromise position: I'm assuming that nobody will be silly enough to set the noexec mount flag on part of the default library path, in order to avoid adding extra overhead into the common case (where those environment variables aren't used). Discussed with: csjp, secteam MFC after: 1 week
2005-03-24 10:12:29 +00:00
#include <sys/mount.h>
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
#include <sys/mman.h>
When checking to see if a shared object is already loaded, look for a device/inode match if no pathname match is found.
1999-08-30 01:50:41 +00:00
#include <sys/stat.h>
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
#include <sys/sysctl.h>
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
#include <sys/uio.h>
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
#include <sys/utsname.h>
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
#include <sys/ktrace.h>
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
#include <dlfcn.h>
#include <err.h>
#include <errno.h>
#include <fcntl.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include "debug.h"
#include "rtld.h"
Dynamic object dependency mapping: libmap. This is an optional feature, disabled by default. This will be useful to people testing the various POSIX threading libraries under -CURRENT but can easily serve other needs.
2003-04-07 16:21:26 +00:00
#include "libmap.h"
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
#include "rtld_tls.h"
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
#ifndef COMPAT_32BIT
As long threatened, stage 2 of making a dynamically-linked root a reality. Install rtld into /libexec.
2003-08-17 08:06:00 +00:00
#define PATH_RTLD "/libexec/ld-elf.so.1"
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
#else
#define PATH_RTLD "/libexec/ld-elf32.so.1"
#endif
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/* Types. */
typedef void (*func_ptr_type)();
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
typedef void * (*path_enum_proc) (const char *path, size_t len, void *arg);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
/*
* This structure provides a reentrant way to keep a list of objects and
* check which ones have already been processed in some way.
*/
typedef struct Struct_DoneList {
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
const Obj_Entry **objs; /* Array of object pointers */
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
unsigned int num_alloc; /* Allocated size of the array */
unsigned int num_used; /* Number of array slots used */
} DoneList;
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Function declarations.
*/
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
static const char *basename(const char *);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
static void die(void) __dead2;
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
static void digest_dynamic1(Obj_Entry *, int, const Elf_Dyn **,
const Elf_Dyn **);
static void digest_dynamic2(Obj_Entry *, const Elf_Dyn *, const Elf_Dyn *);
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
static void digest_dynamic(Obj_Entry *, int);
Change many asserts into normal errors. They were all for conditions caused by invalid shared objects rather than by internal errors. Enable format string mismatch checking for _rtld_error().
1999-07-18 00:02:19 +00:00
static Obj_Entry *digest_phdr(const Elf_Phdr *, int, caddr_t, const char *);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static Obj_Entry *dlcheck(void *);
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
static Obj_Entry *do_load_object(int, const char *, char *, struct stat *, int);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
static int do_search_info(const Obj_Entry *obj, int, struct dl_serinfo *);
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
static bool donelist_check(DoneList *, const Obj_Entry *);
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
static void errmsg_restore(char *);
static char *errmsg_save(void);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
static void *fill_search_info(const char *, size_t, void *);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static char *find_library(const char *, const Obj_Entry *);
Implement ldconfig functionality for ELF. The hints are stored in a different file than the a.out hints, namely, "/var/run/ld-elf.so.hints". These hints consist only of the directory search path. There is no hash table as in the a.out hints, because ELF doesn't have to search for the file with the highest minor version number. (It doesn't have minor version numbers at all.) A single run of ldconfig updates either the a.out hints or the ELF hints, but not both. The set of hints to process is selected in the usual way, via /etc/objformat, or ${OBJFORMAT}, or the "-aout" or "-elf" command line option. The rationale is that you probably want to search different directories for ELF than for a.out. "ldconfig -r" is faked up to produce output like we are used to, except that for ELF there are no minor version numbers. This should enable "ldconfig -r" to be used for checking LIB_DEPENDS in ports even for ELF. I implemented the ELF functionality in a new source file, with an eye toward eliminating the a.out code entirely at some point in the future.
1998-09-05 03:31:00 +00:00
static const char *gethints(void);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static void init_dag(Obj_Entry *);
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
static void init_dag1(Obj_Entry *, Obj_Entry *, DoneList *);
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
static void init_rtld(caddr_t, Elf_Auxinfo **);
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
static void initlist_add_neededs(Needed_Entry *, Objlist *);
static void initlist_add_objects(Obj_Entry *, Obj_Entry **, Objlist *);
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
static void linkmap_add(Obj_Entry *);
static void linkmap_delete(Obj_Entry *);
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
static int load_needed_objects(Obj_Entry *, int);
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
static int load_preload_objects(void);
Implement RTLD_NOLOAD flag for dlopen(3). Requested and tested by: jkim Reviewed by: kan Approved by: re (kensmith)
2009-07-17 19:45:42 +00:00
static Obj_Entry *load_object(const char *, const Obj_Entry *, int);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static Obj_Entry *obj_from_addr(const void *);
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
static void objlist_call_fini(Objlist *, Obj_Entry *, int *);
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
static void objlist_call_init(Objlist *, int *);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
static void objlist_clear(Objlist *);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static Objlist_Entry *objlist_find(Objlist *, const Obj_Entry *);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
static void objlist_init(Objlist *);
static void objlist_push_head(Objlist *, Obj_Entry *);
static void objlist_push_tail(Objlist *, Obj_Entry *);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static void objlist_remove(Objlist *, Obj_Entry *);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
static void *path_enumerate(const char *, path_enum_proc, void *);
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
static int relocate_objects(Obj_Entry *, bool, Obj_Entry *);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
static int rtld_dirname(const char *, char *);
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
static int rtld_dirname_abs(const char *, char *);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static void rtld_exit(void);
static char *search_library_path(const char *, const char *);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
static const void **get_program_var_addr(const char *);
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
static void set_program_var(const char *, const void *);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
static const Elf_Sym *symlook_default(const char *, unsigned long,
const Obj_Entry *, const Obj_Entry **, const Ver_Entry *, int);
When looking up the symbol by dlsym, look it not only in the object given as dso handle, but also in the implicit dependencies of that dso. Also, const-ify the read-only parameter objlist of symlook_list. Reported by: "Simon 'corecode' Schubert" <corecode at fs ei tum de> Approved by: kan (mentor) X-MFC-After: 6.2
2006-09-08 14:59:54 +00:00
static const Elf_Sym *symlook_list(const char *, unsigned long, const Objlist *,
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
const Obj_Entry **, const Ver_Entry *, int, DoneList *);
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
static const Elf_Sym *symlook_needed(const char *, unsigned long,
const Needed_Entry *, const Obj_Entry **, const Ver_Entry *,
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
int, DoneList *);
static void trace_loaded_objects(Obj_Entry *);
Do not remove object from the lists at the unref_dag() stage. Introduce a new unlink_object() function and call it in unload_object() instead. Removing the object in unref_dag() is too early, rtld calls _fini() function after that and shared objects might fail resolve their own symbols.
2003-02-17 20:58:27 +00:00
static void unlink_object(Obj_Entry *);
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
static void unload_object(Obj_Entry *);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static void unref_dag(Obj_Entry *);
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
static void ref_dag(Obj_Entry *);
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
static int origin_subst_one(char **, const char *, const char *,
const char *, char *);
static char *origin_subst(const char *, const char *);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
static int rtld_verify_versions(const Objlist *);
static int rtld_verify_object_versions(Obj_Entry *);
static void object_add_name(Obj_Entry *, const char *);
static int object_match_name(const Obj_Entry *, const char *);
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
static void ld_utrace_log(int, void *, void *, size_t, int, const char *);
Introduce implementation-private rtld interface _rtld_addr_phdr, which fills struct dl_phdr_info for the shared object that contains the specified address, if any. Idea and reviewed by: kan MFC after: 3 weeks
2010-08-23 15:27:03 +00:00
static void rtld_fill_dl_phdr_info(const Obj_Entry *obj,
struct dl_phdr_info *phdr_info);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
void r_debug_state(struct r_debug *, struct link_map *);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Data declarations.
*/
static char *error_message; /* Message for dlerror(), or NULL */
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
struct r_debug r_debug; /* for GDB; */
Use the environment variable LD_LIBMAP_DISABLE to disable libmap.conf(5) functionality.
2003-05-31 14:45:11 +00:00
static bool libmap_disable; /* Disable libmap */
Description from Dan: Another handy libmap patch. Lets you do stuff like this: LD_LIBMAP="libpthread.so.1=libthr.so.1" mythreadedapp If you already have a program-specific override in libmap.conf, note that you must use a program-specific override in LD_LIBMAP: LD_LIBMAP="[mythreadedapp],libpthread.so.1=libthr.so.1" mythreadedapp PR: bin/74471 Submitted by: Dan Nelson <dnelson AT allantgroup.com> MFC after: 2 weeks
2005-02-04 02:46:41 +00:00
static char *libmap_override; /* Maps to use in addition to libmap.conf */
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static bool trust; /* False for setuid and setgid programs */
If "dangerous" environment variables (LD_PRELOAD, LD_LIBMAP, LD_LIBMAP_DISABLE, LD_LIBRARY_PATH) are used, then make sure the libraries being loaded aren't on a noexec-mounted filesystem. This is a compromise position: I'm assuming that nobody will be silly enough to set the noexec mount flag on part of the default library path, in order to avoid adding extra overhead into the common case (where those environment variables aren't used). Discussed with: csjp, secteam MFC after: 1 week
2005-03-24 10:12:29 +00:00
static bool dangerous_ld_env; /* True if environment variables have been
used to affect the libraries loaded */
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static char *ld_bind_now; /* Environment variable for immediate binding */
static char *ld_debug; /* Environment variable for debugging */
static char *ld_library_path; /* Environment variable for search path */
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
static char *ld_preload; /* Environment variable for libraries to
load first */
Support for a new environment variable, LD_ELF_HINTS_PATH for overriding the rtld hints file. This environment variable would be unset if the process is considered as tainted with setuid/setgid. This feature gives a convenient way of using a custom set of shared library that is not located in the default location and switch back. Feature requested by: iXsystems Original patch by: John Hixson MFC after: 2 weeks
2009-03-23 16:49:00 +00:00
static char *ld_elf_hints_path; /* Environment variable for alternative hints path */
Add support for ldd.
1998-05-01 08:39:27 +00:00
static char *ld_tracing; /* Called from ldd to print libs */
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
static char *ld_utrace; /* Use utrace() to log events. */
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static Obj_Entry *obj_list; /* Head of linked list of shared objects */
static Obj_Entry **obj_tail; /* Link field of last object in list */
static Obj_Entry *obj_main; /* The main program shared object */
static Obj_Entry obj_rtld; /* The dynamic linker shared object */
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
static unsigned int obj_count; /* Number of objects in obj_list */
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
static unsigned int obj_loads; /* Number of objects in obj_list */
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static Objlist list_global = /* Objects dlopened with RTLD_GLOBAL */
STAILQ_HEAD_INITIALIZER(list_global);
static Objlist list_main = /* Objects loaded at program startup */
STAILQ_HEAD_INITIALIZER(list_main);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
static Objlist list_fini = /* Objects needing fini() calls */
STAILQ_HEAD_INITIALIZER(list_fini);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Check for undefined weak symbols during PLT binding on powerpc64, and do not attempt to copy NULL function descriptors. This fixes LD_BIND_NOW on powerpc64 after r211706. Reviewed by: kib
2010-09-12 17:04:51 +00:00
Elf_Sym sym_zero; /* For resolving undefined weak refs. */
Resolve undefined weak references to a value of 0. This solves the "__deregister_frame_info" problem that was seen when combining a program linked using the old gcc with shared libraries that were built using egcs.
1999-04-05 02:36:40 +00:00
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
#define GDB_STATE(s,m) r_debug.r_state = s; r_debug_state(&r_debug,m);
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
extern Elf_Dyn _DYNAMIC;
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
#pragma weak _DYNAMIC
Fix the problem that surfaced with the new binutils import on sparc64 (and that is for now being worked around by a binutils patch). The rtld code tested &_DYNAMIC against 0 to see whether rtld itself was built as PIC or not. While the sparc64 MD code did not rely on the preset value of the GOT slot for _DYNAMIC any more due to previous binutils changes, it still used to not be 0, so that this check did work. The new binutils do however initialize this slot with 0. As a consequence, rtld would not properly initialize itself and crash. Fix that by introducing a new macro, RTLD_IS_DYNAMIC, to take the role of this test. For sparc64, it is implemented using the rtld_dynamic() code that was already there. If an architecture does not provide its own implementation, we default to the old check. While being there, mark _DYNAMIC as a weak symbol in the sparc64 rtld_start.S. This is needed in the LDSCRIPT case, which is however not currently supported for want of an actual ldscript. Sanity checked with md5 on alpha, amd64, i386 and ia64.
2004-06-18 02:01:37 +00:00
#ifndef RTLD_IS_DYNAMIC
#define RTLD_IS_DYNAMIC() (&_DYNAMIC != NULL)
#endif
Resolve undefined weak references to a value of 0. This solves the "__deregister_frame_info" problem that was seen when combining a program linked using the old gcc with shared libraries that were built using egcs.
1999-04-05 02:36:40 +00:00
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
int osreldate, pagesize;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Global declarations normally provided by crt1. The dynamic linker is
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
* not built with crt1, so we have to provide them ourselves.
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
*/
char *__progname;
char **environ;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
/*
* Globals to control TLS allocation.
*/
size_t tls_last_offset; /* Static TLS offset of last module */
size_t tls_last_size; /* Static TLS size of last module */
size_t tls_static_space; /* Static TLS space allocated */
int tls_dtv_generation = 1; /* Used to detect when dtv size changes */
int tls_max_index = 1; /* Largest module index allocated */
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
/*
* Fill in a DoneList with an allocation large enough to hold all of
* the currently-loaded objects. Keep this as a macro since it calls
* alloca and we want that to occur within the scope of the caller.
*/
#define donelist_init(dlp) \
((dlp)->objs = alloca(obj_count * sizeof (dlp)->objs[0]), \
assert((dlp)->objs != NULL), \
(dlp)->num_alloc = obj_count, \
(dlp)->num_used = 0)
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
#define UTRACE_DLOPEN_START 1
#define UTRACE_DLOPEN_STOP 2
#define UTRACE_DLCLOSE_START 3
#define UTRACE_DLCLOSE_STOP 4
#define UTRACE_LOAD_OBJECT 5
#define UTRACE_UNLOAD_OBJECT 6
#define UTRACE_ADD_RUNDEP 7
#define UTRACE_PRELOAD_FINISHED 8
#define UTRACE_INIT_CALL 9
#define UTRACE_FINI_CALL 10
struct utrace_rtld {
char sig[4]; /* 'RTLD' */
int event;
void *handle;
void *mapbase; /* Used for 'parent' and 'init/fini' */
size_t mapsize;
int refcnt; /* Used for 'mode' */
char name[MAXPATHLEN];
};
#define LD_UTRACE(e, h, mb, ms, r, n) do { \
if (ld_utrace != NULL) \
ld_utrace_log(e, h, mb, ms, r, n); \
} while (0)
static void
ld_utrace_log(int event, void *handle, void *mapbase, size_t mapsize,
int refcnt, const char *name)
{
struct utrace_rtld ut;
ut.sig[0] = 'R';
ut.sig[1] = 'T';
ut.sig[2] = 'L';
ut.sig[3] = 'D';
ut.event = event;
ut.handle = handle;
ut.mapbase = mapbase;
ut.mapsize = mapsize;
ut.refcnt = refcnt;
bzero(ut.name, sizeof(ut.name));
if (name)
strlcpy(ut.name, name, sizeof(ut.name));
utrace(&ut, sizeof(ut));
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Main entry point for dynamic linking. The first argument is the
* stack pointer. The stack is expected to be laid out as described
* in the SVR4 ABI specification, Intel 386 Processor Supplement.
* Specifically, the stack pointer points to a word containing
* ARGC. Following that in the stack is a null-terminated sequence
* of pointers to argument strings. Then comes a null-terminated
* sequence of pointers to environment strings. Finally, there is a
* sequence of "auxiliary vector" entries.
*
* The second argument points to a place to store the dynamic linker's
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
* exit procedure pointer and the third to a place to store the main
* program's object.
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
*
* The return value is the main program's entry point.
*/
func_ptr_type
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
_rtld(Elf_Addr *sp, func_ptr_type *exit_proc, Obj_Entry **objp)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
Elf_Auxinfo *aux_info[AT_COUNT];
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
int i;
int argc;
char **argv;
char **env;
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
Elf_Auxinfo *aux;
Elf_Auxinfo *auxp;
Change many asserts into normal errors. They were all for conditions caused by invalid shared objects rather than by internal errors. Enable format string mismatch checking for _rtld_error().
1999-07-18 00:02:19 +00:00
const char *argv0;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Objlist_Entry *entry;
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
Obj_Entry *obj;
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
Obj_Entry **preload_tail;
Objlist initlist;
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
int lockstate;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* On entry, the dynamic linker itself has not been relocated yet.
* Be very careful not to reference any global data until after
* init_rtld has returned. It is OK to reference file-scope statics
* and string constants, and to call static and global functions.
*/
/* Find the auxiliary vector on the stack. */
argc = *sp++;
argv = (char **) sp;
sp += argc + 1; /* Skip over arguments and NULL terminator */
env = (char **) sp;
while (*sp++ != 0) /* Skip over environment, and NULL terminator */
;
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
aux = (Elf_Auxinfo *) sp;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/* Digest the auxiliary vector. */
for (i = 0; i < AT_COUNT; i++)
aux_info[i] = NULL;
for (auxp = aux; auxp->a_type != AT_NULL; auxp++) {
if (auxp->a_type < AT_COUNT)
aux_info[auxp->a_type] = auxp;
}
/* Initialize and relocate ourselves. */
assert(aux_info[AT_BASE] != NULL);
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
init_rtld((caddr_t) aux_info[AT_BASE]->a_un.a_ptr, aux_info);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
__progname = obj_rtld.path;
Change many asserts into normal errors. They were all for conditions caused by invalid shared objects rather than by internal errors. Enable format string mismatch checking for _rtld_error().
1999-07-18 00:02:19 +00:00
argv0 = argv[0] != NULL ? argv[0] : "(null)";
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
environ = env;
- use issetugid() - be paranoid about honoring LD_LIBMAP_DISABLE. Suggested by: rwatson
2003-05-31 15:24:29 +00:00
trust = !issetugid();
Use the environment variable LD_LIBMAP_DISABLE to disable libmap.conf(5) functionality.
2003-05-31 14:45:11 +00:00
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
ld_bind_now = getenv(LD_ "BIND_NOW");
In the event a process is tainted (setuid/setgid binaries), un-set any potentially dangerous environment variables all together. It should be noted that the run-time linker will not honnor these environment variables if the process is tainted currently. However, once a child of the tainted process calls setuid(2), it's status as being tainted (as defined by issetugid(2)) will be removed. This could be problematic because subsequent activations of the run-time linker could honnor these dangerous variables. This is more of an anti foot-shot mechanism, there is nothing I am aware of in base that does this, however there may be third party utilities which do, and there is no real negative impact of clearing these environment variables. Discussed on: secteam Reviewed by: cperciva PR: kern/109836 MFC after: 2 weeks
2007-05-17 18:00:27 +00:00
/*
* If the process is tainted, then we un-set the dangerous environment
* variables. The process will be marked as tainted until setuid(2)
* is called. If any child process calls setuid(2) we do not want any
* future processes to honor the potentially un-safe variables.
*/
if (!trust) {
Fix local root vulnerability. Security: Advisory will be coming soon. X-MFC-After: 30 seconds
2009-12-01 02:57:06 +00:00
if (unsetenv(LD_ "PRELOAD") || unsetenv(LD_ "LIBMAP") ||
unsetenv(LD_ "LIBRARY_PATH") || unsetenv(LD_ "LIBMAP_DISABLE") ||
unsetenv(LD_ "DEBUG") || unsetenv(LD_ "ELF_HINTS_PATH")) {
_rtld_error("environment corrupt; aborting");
die();
}
In the event a process is tainted (setuid/setgid binaries), un-set any potentially dangerous environment variables all together. It should be noted that the run-time linker will not honnor these environment variables if the process is tainted currently. However, once a child of the tainted process calls setuid(2), it's status as being tainted (as defined by issetugid(2)) will be removed. This could be problematic because subsequent activations of the run-time linker could honnor these dangerous variables. This is more of an anti foot-shot mechanism, there is nothing I am aware of in base that does this, however there may be third party utilities which do, and there is no real negative impact of clearing these environment variables. Discussed on: secteam Reviewed by: cperciva PR: kern/109836 MFC after: 2 weeks
2007-05-17 18:00:27 +00:00
}
ld_debug = getenv(LD_ "DEBUG");
libmap_disable = getenv(LD_ "LIBMAP_DISABLE") != NULL;
libmap_override = getenv(LD_ "LIBMAP");
ld_library_path = getenv(LD_ "LIBRARY_PATH");
ld_preload = getenv(LD_ "PRELOAD");
Support for a new environment variable, LD_ELF_HINTS_PATH for overriding the rtld hints file. This environment variable would be unset if the process is considered as tainted with setuid/setgid. This feature gives a convenient way of using a custom set of shared library that is not located in the default location and switch back. Feature requested by: iXsystems Original patch by: John Hixson MFC after: 2 weeks
2009-03-23 16:49:00 +00:00
ld_elf_hints_path = getenv(LD_ "ELF_HINTS_PATH");
In the event a process is tainted (setuid/setgid binaries), un-set any potentially dangerous environment variables all together. It should be noted that the run-time linker will not honnor these environment variables if the process is tainted currently. However, once a child of the tainted process calls setuid(2), it's status as being tainted (as defined by issetugid(2)) will be removed. This could be problematic because subsequent activations of the run-time linker could honnor these dangerous variables. This is more of an anti foot-shot mechanism, there is nothing I am aware of in base that does this, however there may be third party utilities which do, and there is no real negative impact of clearing these environment variables. Discussed on: secteam Reviewed by: cperciva PR: kern/109836 MFC after: 2 weeks
2007-05-17 18:00:27 +00:00
dangerous_ld_env = libmap_disable || (libmap_override != NULL) ||
Support for a new environment variable, LD_ELF_HINTS_PATH for overriding the rtld hints file. This environment variable would be unset if the process is considered as tainted with setuid/setgid. This feature gives a convenient way of using a custom set of shared library that is not located in the default location and switch back. Feature requested by: iXsystems Original patch by: John Hixson MFC after: 2 weeks
2009-03-23 16:49:00 +00:00
(ld_library_path != NULL) || (ld_preload != NULL) ||
(ld_elf_hints_path != NULL);
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
ld_tracing = getenv(LD_ "TRACE_LOADED_OBJECTS");
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
ld_utrace = getenv(LD_ "UTRACE");
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Support for a new environment variable, LD_ELF_HINTS_PATH for overriding the rtld hints file. This environment variable would be unset if the process is considered as tainted with setuid/setgid. This feature gives a convenient way of using a custom set of shared library that is not located in the default location and switch back. Feature requested by: iXsystems Original patch by: John Hixson MFC after: 2 weeks
2009-03-23 16:49:00 +00:00
if ((ld_elf_hints_path == NULL) || strlen(ld_elf_hints_path) == 0)
ld_elf_hints_path = _PATH_ELF_HINTS;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
if (ld_debug != NULL && *ld_debug != '\0')
debug = 1;
dbg("%s is initialized, base address = %p", __progname,
(caddr_t) aux_info[AT_BASE]->a_un.a_ptr);
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
dbg("RTLD dynamic = %p", obj_rtld.dynamic);
dbg("RTLD pltgot = %p", obj_rtld.pltgot);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Load the main program, or process its program header if it is
* already loaded.
*/
if (aux_info[AT_EXECFD] != NULL) { /* Load the main program. */
int fd = aux_info[AT_EXECFD]->a_un.a_val;
dbg("loading main program");
When checking to see if a shared object is already loaded, look for a device/inode match if no pathname match is found.
1999-08-30 01:50:41 +00:00
obj_main = map_object(fd, argv0, NULL);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
close(fd);
if (obj_main == NULL)
die();
} else { /* Main program already loaded. */
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
const Elf_Phdr *phdr;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
int phnum;
caddr_t entry;
dbg("processing main program's program header");
assert(aux_info[AT_PHDR] != NULL);
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
phdr = (const Elf_Phdr *) aux_info[AT_PHDR]->a_un.a_ptr;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
assert(aux_info[AT_PHNUM] != NULL);
phnum = aux_info[AT_PHNUM]->a_un.a_val;
assert(aux_info[AT_PHENT] != NULL);
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
assert(aux_info[AT_PHENT]->a_un.a_val == sizeof(Elf_Phdr));
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
assert(aux_info[AT_ENTRY] != NULL);
entry = (caddr_t) aux_info[AT_ENTRY]->a_un.a_ptr;
Change many asserts into normal errors. They were all for conditions caused by invalid shared objects rather than by internal errors. Enable format string mismatch checking for _rtld_error().
1999-07-18 00:02:19 +00:00
if ((obj_main = digest_phdr(phdr, phnum, entry, argv0)) == NULL)
die();
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
if (aux_info[AT_EXECPATH] != 0) {
char *kexecpath;
char buf[MAXPATHLEN];
kexecpath = aux_info[AT_EXECPATH]->a_un.a_ptr;
dbg("AT_EXECPATH %p %s", kexecpath, kexecpath);
if (kexecpath[0] == '/')
obj_main->path = kexecpath;
else if (getcwd(buf, sizeof(buf)) == NULL ||
strlcat(buf, "/", sizeof(buf)) >= sizeof(buf) ||
strlcat(buf, kexecpath, sizeof(buf)) >= sizeof(buf))
obj_main->path = xstrdup(argv0);
else
obj_main->path = xstrdup(buf);
} else {
dbg("No AT_EXECPATH");
obj_main->path = xstrdup(argv0);
}
dbg("obj_main path %s", obj_main->path);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
obj_main->mainprog = true;
Get the actual pathname of the dynamic linker from the executable's PT_INTERP program header entry, to ensure that gdb always finds the right dynamic linker. Use obj->relocbase to simplify a few calculations where appropriate.
1999-08-30 01:54:13 +00:00
/*
* Get the actual dynamic linker pathname from the executable if
* possible. (It should always be possible.) That ensures that
* gdb will find the right dynamic linker even if a non-standard
* one is being used.
*/
if (obj_main->interp != NULL &&
strcmp(obj_main->interp, obj_rtld.path) != 0) {
free(obj_rtld.path);
obj_rtld.path = xstrdup(obj_main->interp);
If we change obj_rtld.path after initialising __progname, make sure we change __progname to point at the new storage otherwise it ends up pointing at freed memory which leads to confusing garbled error messages.
2004-03-29 18:37:37 +00:00
__progname = obj_rtld.path;
Get the actual pathname of the dynamic linker from the executable's PT_INTERP program header entry, to ensure that gdb always finds the right dynamic linker. Use obj->relocbase to simplify a few calculations where appropriate.
1999-08-30 01:54:13 +00:00
}
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
digest_dynamic(obj_main, 0);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
linkmap_add(obj_main);
linkmap_add(&obj_rtld);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/* Link the main program into the list of objects. */
*obj_tail = obj_main;
obj_tail = &obj_main->next;
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
obj_count++;
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
obj_loads++;
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/* Make sure we don't call the main program's init and fini functions. */
Make NULL a (void*)0 whereever possible, and fix the warnings(-Werror) that this provokes. "Wherever possible" means "In the kernel OR NOT C++" (implying C). There are places where (void *) pointers are not valid, such as for function pointers, but in the special case of (void *)0, agreement settles on it being OK. Most of the fixes were NULL where an integer zero was needed; many of the fixes were NULL where ascii <nul> ('\0') was needed, and a few were just "other". Tested on: i386 sparc64
2004-03-05 08:10:19 +00:00
obj_main->init = obj_main->fini = (Elf_Addr)NULL;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Resolve undefined weak references to a value of 0. This solves the "__deregister_frame_info" problem that was seen when combining a program linked using the old gcc with shared libraries that were built using egcs.
1999-04-05 02:36:40 +00:00
/* Initialize a fake symbol for resolving undefined weak references. */
sym_zero.st_info = ELF_ST_INFO(STB_GLOBAL, STT_NOTYPE);
Fix handling of weak references to undefined symbols on ia64: o Set st_shndx for sym_zero to SHN_UNDEF instead of SHN_ABS. This gives us something to reliably test against. o For weak references to undefined sysmbols (as indicated by having st_shndx equals SHN_UNDEF) in the context of OPDs, the address of the OPD is to be zero, not the address of the function it contains. o For weak references to undefined symbols in all other cases (only DIR64LSB at this time), the actual relocated value is to be zero, not the value prior to relocating. Roughly speaking, weak references to undefined symbols are no-ops. Tested on: i386, ia64
2002-04-27 05:32:51 +00:00
sym_zero.st_shndx = SHN_UNDEF;
Calculate relocation base for the main object, and apply the relocation adjustment for all virtual addresses encoded into the ELF structures of it. PIE binary could and should be loaded at non-zero mapbase. For sym_zero pseudosymbol used as a return value from find_symdef() for undefined weak symbols, st_value also should be adjusted, since _rtld_bind corrects symbol values by relocbase. Discussed with: bz Reviewed by: kan Tested by: bz (i386, amd64), bsam (linux) MFC after: some time
2009-10-10 15:27:10 +00:00
sym_zero.st_value = -(uintptr_t)obj_main->relocbase;
Resolve undefined weak references to a value of 0. This solves the "__deregister_frame_info" problem that was seen when combining a program linked using the old gcc with shared libraries that were built using egcs.
1999-04-05 02:36:40 +00:00
Use the environment variable LD_LIBMAP_DISABLE to disable libmap.conf(5) functionality.
2003-05-31 14:45:11 +00:00
if (!libmap_disable)
Description from Dan: Another handy libmap patch. Lets you do stuff like this: LD_LIBMAP="libpthread.so.1=libthr.so.1" mythreadedapp If you already have a program-specific override in libmap.conf, note that you must use a program-specific override in LD_LIBMAP: LD_LIBMAP="[mythreadedapp],libpthread.so.1=libthr.so.1" mythreadedapp PR: bin/74471 Submitted by: Dan Nelson <dnelson AT allantgroup.com> MFC after: 2 weeks
2005-02-04 02:46:41 +00:00
libmap_disable = (bool)lm_init(libmap_override);
Dynamic object dependency mapping: libmap. This is an optional feature, disabled by default. This will be useful to people testing the various POSIX threading libraries under -CURRENT but can easily serve other needs.
2003-04-07 16:21:26 +00:00
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
dbg("loading LD_PRELOAD libraries");
if (load_preload_objects() == -1)
die();
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
preload_tail = obj_tail;
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
dbg("loading needed objects");
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
if (load_needed_objects(obj_main, 0) == -1)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
die();
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/* Make a list of all objects loaded at startup. */
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
for (obj = obj_list; obj != NULL; obj = obj->next) {
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_push_tail(&list_main, obj);
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
obj->refcount++;
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
dbg("checking for required versions");
if (rtld_verify_versions(&list_main) == -1 && !ld_tracing)
die();
Add support for ldd.
1998-05-01 08:39:27 +00:00
if (ld_tracing) { /* We're done */
trace_loaded_objects(obj_main);
exit(0);
}
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
if (getenv(LD_ "DUMP_REL_PRE") != NULL) {
Provide a mechanism for dumping relocation information. Setting the LD_DUMP_REL_PRE or LD_DUMP_REL_POST environment variables cause rtld-elf to output a table of all relocations. This is useful for debugging.
2003-06-19 03:55:38 +00:00
dump_relocations(obj_main);
exit (0);
}
Do the TLS offset allocations before relocations, as otherwise there can be overlap in the TLS offsets, if the relocations are done in a certain order. Approved by: dfr, grehan (mentor)
2004-11-02 09:42:21 +00:00
/* setup TLS for main thread */
dbg("initializing initial thread local storage");
STAILQ_FOREACH(entry, &list_main, link) {
/*
* Allocate all the initial objects out of the static TLS
* block even if they didn't ask for it.
*/
allocate_tls_offset(entry->obj);
}
allocate_initial_tls(obj_list);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
if (relocate_objects(obj_main,
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
ld_bind_now != NULL && *ld_bind_now != '\0', &obj_rtld) == -1)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
die();
dbg("doing copy relocations");
if (do_copy_relocations(obj_main) == -1)
die();
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
if (getenv(LD_ "DUMP_REL_POST") != NULL) {
Provide a mechanism for dumping relocation information. Setting the LD_DUMP_REL_PRE or LD_DUMP_REL_POST environment variables cause rtld-elf to output a table of all relocations. This is useful for debugging.
2003-06-19 03:55:38 +00:00
dump_relocations(obj_main);
exit (0);
}
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
dbg("initializing key program variables");
set_program_var("__progname", argv[0] != NULL ? basename(argv[0]) : "");
set_program_var("environ", env);
Provide the starting image with the pointer to ELF aux vector. It is written into the __elf_aux_vector variable, if the symbol is present. Idea from: kan Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:08:28 +00:00
set_program_var("__elf_aux_vector", aux);
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
dbg("initializing thread locks");
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
lockdflt_init();
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/* Make a list of init functions to call. */
objlist_init(&initlist);
initlist_add_objects(obj_list, preload_tail, &initlist);
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
r_debug_state(NULL, &obj_main->linkmap); /* say hello to gdb! */
Shake hands with GDB a little bit earlier so that it is possible to debug the init functions. Submitted by: dfr
1999-07-03 23:54:02 +00:00
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
lockstate = wlock_acquire(rtld_bind_lock);
Make sure internal rtld malloc routines are not called from unlocked contexts as rtld's malloc is not thread safe and is only supposed to be called with exclusive bind lock already held. The originating PR submitted a patch on top of different pre-requisite workaroud for unsafe dlopen calls, and the patch was midief slighlty to apply to stock sources for the purpose of this commit. Running rtld malloc from unlocked contexts is a bug on its own. PR: 126950 Submited by: Oleg Dolgov
2008-09-03 01:05:32 +00:00
objlist_call_init(&initlist, &lockstate);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_clear(&initlist);
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
wlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
dbg("transferring control to program entry point = %p", obj_main->entry);
/* Return the exit procedure and the program entry point. */
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
*exit_proc = rtld_exit;
*objp = obj_main;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return (func_ptr_type) obj_main->entry;
}
Fix a serious performance bug for large programs on the Alpha, discovered by Hidetoshi Shimokawa. Large programs need multiple GOTs. The lazy binding stub in the PLT can be reached from any of these GOTs, but the dynamic linker only has enough information to fix up the first GOT entry. Thus calls through the other GOTs went through the time-consuming lazy binding process on every call. This fix rewrites the PLT entries themselves to bypass the lazy binding. Tested by Hidetoshi Shimokawa and Steve Price. Reviewed by: Doug Rabson <dfr@freebsd.org>
1999-06-25 02:53:59 +00:00
Elf_Addr
Make our ELF64 type definitions match standards. In particular this means: o Remove Elf64_Quarter, o Redefine Elf64_Half to be 16-bit, o Redefine Elf64_Word to be 32-bit, o Add Elf64_Xword and Elf64_Sxword for 64-bit entities, o Use Elf_Size in MI code to abstract the difference between Elf32_Word and Elf64_Word. o Add Elf_Ssize as the signed counterpart of Elf_Size. MFC after: 2 weeks
2005-12-18 04:52:37 +00:00
_rtld_bind(Obj_Entry *obj, Elf_Size reloff)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
const Elf_Rel *rel;
const Elf_Sym *def;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
const Obj_Entry *defobj;
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
Elf_Addr *where;
Fix a serious performance bug for large programs on the Alpha, discovered by Hidetoshi Shimokawa. Large programs need multiple GOTs. The lazy binding stub in the PLT can be reached from any of these GOTs, but the dynamic linker only has enough information to fix up the first GOT entry. Thus calls through the other GOTs went through the time-consuming lazy binding process on every call. This fix rewrites the PLT entries themselves to bypass the lazy binding. Tested by Hidetoshi Shimokawa and Steve Price. Reviewed by: Doug Rabson <dfr@freebsd.org>
1999-06-25 02:53:59 +00:00
Elf_Addr target;
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
int lockstate;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
lockstate = rlock_acquire(rtld_bind_lock);
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
if (obj->pltrel)
rel = (const Elf_Rel *) ((caddr_t) obj->pltrel + reloff);
else
rel = (const Elf_Rel *) ((caddr_t) obj->pltrela + reloff);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
where = (Elf_Addr *) (obj->relocbase + rel->r_offset);
Performance improvements for the ELF dynamic linker. These particularly help programs which load many shared libraries with a lot of relocations. Large C++ programs such as are found in KDE are a prime example. While relocating a shared object, maintain a vector of symbols which have already been looked up, directly indexed by symbol number. Typically, symbols which are referenced by a relocation entry are referenced by many of them. This is the same optimization I made to the a.out dynamic linker in 1995 (rtld.c revision 1.30). Also, compare the first character of a sought-after symbol with its symbol table entry before calling strcmp(). On a PII/400 these changes reduce the start-up time of a typical KDE program from 833 msec (elapsed) to 370 msec. MFC after: 5 days
2001-05-05 23:21:05 +00:00
def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj, true, NULL);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
if (def == NULL)
die();
Fix a serious performance bug for large programs on the Alpha, discovered by Hidetoshi Shimokawa. Large programs need multiple GOTs. The lazy binding stub in the PLT can be reached from any of these GOTs, but the dynamic linker only has enough information to fix up the first GOT entry. Thus calls through the other GOTs went through the time-consuming lazy binding process on every call. This fix rewrites the PLT entries themselves to bypass the lazy binding. Tested by Hidetoshi Shimokawa and Steve Price. Reviewed by: Doug Rabson <dfr@freebsd.org>
1999-06-25 02:53:59 +00:00
target = (Elf_Addr)(defobj->relocbase + def->st_value);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
dbg("\"%s\" in \"%s\" ==> %p in \"%s\"",
defobj->strtab + def->st_name, basename(obj->path),
Fix a serious performance bug for large programs on the Alpha, discovered by Hidetoshi Shimokawa. Large programs need multiple GOTs. The lazy binding stub in the PLT can be reached from any of these GOTs, but the dynamic linker only has enough information to fix up the first GOT entry. Thus calls through the other GOTs went through the time-consuming lazy binding process on every call. This fix rewrites the PLT entries themselves to bypass the lazy binding. Tested by Hidetoshi Shimokawa and Steve Price. Reviewed by: Doug Rabson <dfr@freebsd.org>
1999-06-25 02:53:59 +00:00
(void *)target, basename(defobj->path));
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Add ia64 support. Various adjustments were made to existing targets to cope with a few interface changes required by the ia64. In particular, function pointers on ia64 need special treatment in rtld.
2001-10-15 18:48:42 +00:00
/*
* Write the new contents for the jmpslot. Note that depending on
* architecture, the value which we need to return back to the
* lazy binding trampoline may or may not be the target
* address. The value returned from reloc_jmpslot() is the value
* that the trampoline needs.
*/
Fix the handling of high PLT entries (> 32764) on sparc64. This requires additional arguments to reloc_jmpslot(), which is why MI code and MD code of other platforms had to be changed. Reviewed by: jake Approved by: re
2002-11-18 22:08:50 +00:00
target = reloc_jmpslot(where, target, defobj, obj, rel);
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return target;
}
/*
* Error reporting function. Use it like printf. If formats the message
* into a buffer, and sets things up so that the next call to dlerror()
* will return the message.
*/
void
_rtld_error(const char *fmt, ...)
{
static char buf[512];
va_list ap;
va_start(ap, fmt);
vsnprintf(buf, sizeof buf, fmt, ap);
error_message = buf;
va_end(ap);
}
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
/*
* Return a dynamically-allocated copy of the current error message, if any.
*/
static char *
errmsg_save(void)
{
return error_message == NULL ? NULL : xstrdup(error_message);
}
/*
* Restore the current error message from a copy which was previously saved
* by errmsg_save(). The copy is freed.
*/
static void
errmsg_restore(char *saved_msg)
{
if (saved_msg == NULL)
error_message = NULL;
else {
_rtld_error("%s", saved_msg);
free(saved_msg);
}
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static const char *
basename(const char *name)
{
const char *p = strrchr(name, '/');
return p != NULL ? p + 1 : name;
}
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
static struct utsname uts;
static int
origin_subst_one(char **res, const char *real, const char *kw, const char *subst,
char *may_free)
{
const char *p, *p1;
char *res1;
int subst_len;
int kw_len;
res1 = *res = NULL;
p = real;
subst_len = kw_len = 0;
for (;;) {
p1 = strstr(p, kw);
if (p1 != NULL) {
if (subst_len == 0) {
subst_len = strlen(subst);
kw_len = strlen(kw);
}
if (*res == NULL) {
*res = xmalloc(PATH_MAX);
res1 = *res;
}
if ((res1 - *res) + subst_len + (p1 - p) >= PATH_MAX) {
_rtld_error("Substitution of %s in %s cannot be performed",
kw, real);
if (may_free != NULL)
free(may_free);
free(res);
return (false);
}
memcpy(res1, p, p1 - p);
res1 += p1 - p;
memcpy(res1, subst, subst_len);
res1 += subst_len;
p = p1 + kw_len;
} else {
if (*res == NULL) {
if (may_free != NULL)
*res = may_free;
else
*res = xstrdup(real);
return (true);
}
*res1 = '\0';
if (may_free != NULL)
free(may_free);
if (strlcat(res1, p, PATH_MAX - (res1 - *res)) >= PATH_MAX) {
free(res);
return (false);
}
return (true);
}
}
}
static char *
origin_subst(const char *real, const char *origin_path)
{
char *res1, *res2, *res3, *res4;
if (uts.sysname[0] == '\0') {
if (uname(&uts) != 0) {
_rtld_error("utsname failed: %d", errno);
return (NULL);
}
}
if (!origin_subst_one(&res1, real, "$ORIGIN", origin_path, NULL) ||
!origin_subst_one(&res2, res1, "$OSNAME", uts.sysname, res1) ||
!origin_subst_one(&res3, res2, "$OSREL", uts.release, res2) ||
!origin_subst_one(&res4, res3, "$PLATFORM", uts.machine, res3))
return (NULL);
return (res4);
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
static void
die(void)
{
const char *msg = dlerror();
if (msg == NULL)
msg = "Fatal error";
errx(1, "%s", msg);
}
/*
* Process a shared object's DYNAMIC section, and save the important
* information in its Obj_Entry structure.
*/
static void
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
digest_dynamic1(Obj_Entry *obj, int early, const Elf_Dyn **dyn_rpath,
const Elf_Dyn **dyn_soname)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
const Elf_Dyn *dynp;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Needed_Entry **needed_tail = &obj->needed;
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
int plttype = DT_REL;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
*dyn_rpath = NULL;
*dyn_soname = NULL;
Initialise some uninitialised variables. Thanks to: valgrind
2003-12-31 15:10:41 +00:00
obj->bind_now = false;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
for (dynp = obj->dynamic; dynp->d_tag != DT_NULL; dynp++) {
switch (dynp->d_tag) {
case DT_REL:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
obj->rel = (const Elf_Rel *) (obj->relocbase + dynp->d_un.d_ptr);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_RELSZ:
obj->relsize = dynp->d_un.d_val;
break;
case DT_RELENT:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
assert(dynp->d_un.d_val == sizeof(Elf_Rel));
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_JMPREL:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
obj->pltrel = (const Elf_Rel *)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
(obj->relocbase + dynp->d_un.d_ptr);
break;
case DT_PLTRELSZ:
obj->pltrelsize = dynp->d_un.d_val;
break;
case DT_RELA:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
obj->rela = (const Elf_Rela *) (obj->relocbase + dynp->d_un.d_ptr);
break;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
case DT_RELASZ:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
obj->relasize = dynp->d_un.d_val;
break;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
case DT_RELAENT:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
assert(dynp->d_un.d_val == sizeof(Elf_Rela));
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_PLTREL:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
plttype = dynp->d_un.d_val;
assert(dynp->d_un.d_val == DT_REL || plttype == DT_RELA);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_SYMTAB:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
obj->symtab = (const Elf_Sym *)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
(obj->relocbase + dynp->d_un.d_ptr);
break;
case DT_SYMENT:
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
assert(dynp->d_un.d_val == sizeof(Elf_Sym));
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_STRTAB:
obj->strtab = (const char *) (obj->relocbase + dynp->d_un.d_ptr);
break;
case DT_STRSZ:
obj->strsize = dynp->d_un.d_val;
break;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
case DT_VERNEED:
obj->verneed = (const Elf_Verneed *) (obj->relocbase +
dynp->d_un.d_val);
break;
case DT_VERNEEDNUM:
obj->verneednum = dynp->d_un.d_val;
break;
case DT_VERDEF:
obj->verdef = (const Elf_Verdef *) (obj->relocbase +
dynp->d_un.d_val);
break;
case DT_VERDEFNUM:
obj->verdefnum = dynp->d_un.d_val;
break;
case DT_VERSYM:
obj->versyms = (const Elf_Versym *)(obj->relocbase +
dynp->d_un.d_val);
break;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
case DT_HASH:
{
Add ia64 support. Various adjustments were made to existing targets to cope with a few interface changes required by the ia64. In particular, function pointers on ia64 need special treatment in rtld.
2001-10-15 18:48:42 +00:00
const Elf_Hashelt *hashtab = (const Elf_Hashelt *)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
(obj->relocbase + dynp->d_un.d_ptr);
obj->nbuckets = hashtab[0];
obj->nchains = hashtab[1];
obj->buckets = hashtab + 2;
obj->chains = obj->buckets + obj->nbuckets;
}
break;
case DT_NEEDED:
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
if (!obj->rtld) {
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Needed_Entry *nep = NEW(Needed_Entry);
nep->name = dynp->d_un.d_val;
nep->obj = NULL;
nep->next = NULL;
*needed_tail = nep;
needed_tail = &nep->next;
}
break;
case DT_PLTGOT:
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
obj->pltgot = (Elf_Addr *) (obj->relocbase + dynp->d_un.d_ptr);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_TEXTREL:
obj->textrel = true;
break;
case DT_SYMBOLIC:
obj->symbolic = true;
break;
case DT_RPATH:
- Add support for DT_FLAGS. - Define various things from the most recent ELF spec.
2003-06-18 03:34:29 +00:00
case DT_RUNPATH: /* XXX: process separately */
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* We have to wait until later to process this, because we
* might not have gotten the address of the string table yet.
*/
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
*dyn_rpath = dynp;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_SONAME:
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
*dyn_soname = dynp;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_INIT:
Update rtld for the "new" ia64 ABI. In the old toolchain, the DT_INIT and DT_FINI tags pointed to fptr records. In 2.11.2, it points to the actuall address of the function. On IA64 you cannot just take an address of a function, store it in a function pointer variable and call it.. the function pointers point to a fptr data block that has the target gp and address in it. This is absolutely necessary for using the in-tree binutils toolchain, but (unfortunately) will not work with old shared libraries. Save your old ld-elf.so.1 if you want to use old ones still. Do not mix-and-match. This is a no-op change for i386 and alpha. Reviewed by: dfr
2001-10-29 10:10:10 +00:00
obj->init = (Elf_Addr) (obj->relocbase + dynp->d_un.d_ptr);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
case DT_FINI:
Update rtld for the "new" ia64 ABI. In the old toolchain, the DT_INIT and DT_FINI tags pointed to fptr records. In 2.11.2, it points to the actuall address of the function. On IA64 you cannot just take an address of a function, store it in a function pointer variable and call it.. the function pointers point to a fptr data block that has the target gp and address in it. This is absolutely necessary for using the in-tree binutils toolchain, but (unfortunately) will not work with old shared libraries. Save your old ld-elf.so.1 if you want to use old ones still. Do not mix-and-match. This is a no-op change for i386 and alpha. Reviewed by: dfr
2001-10-29 10:10:10 +00:00
obj->fini = (Elf_Addr) (obj->relocbase + dynp->d_un.d_ptr);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
/*
* Don't process DT_DEBUG on MIPS as the dynamic section
* is mapped read-only. DT_MIPS_RLD_MAP is used instead.
*/
#ifndef __mips__
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
case DT_DEBUG:
/* XXX - not implemented yet */
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
if (!early)
dbg("Filling in DT_DEBUG entry");
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
((Elf_Dyn*)dynp)->d_un.d_ptr = (Elf_Addr) &r_debug;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
#endif
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
- Add support for DT_FLAGS. - Define various things from the most recent ELF spec.
2003-06-18 03:34:29 +00:00
case DT_FLAGS:
Flag controlling origin expansion in DT_FLAGS is DF_ORIGIN, not DF_1_ORIGIN. Reviewed by: kan MFC after: 3 days
2009-11-26 13:55:49 +00:00
if ((dynp->d_un.d_val & DF_ORIGIN) && trust)
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
obj->z_origin = true;
- Add support for DT_FLAGS. - Define various things from the most recent ELF spec.
2003-06-18 03:34:29 +00:00
if (dynp->d_un.d_val & DF_SYMBOLIC)
obj->symbolic = true;
if (dynp->d_un.d_val & DF_TEXTREL)
obj->textrel = true;
if (dynp->d_un.d_val & DF_BIND_NOW)
obj->bind_now = true;
if (dynp->d_un.d_val & DF_STATIC_TLS)
;
break;
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
#ifdef __mips__
case DT_MIPS_LOCAL_GOTNO:
obj->local_gotno = dynp->d_un.d_val;
break;
case DT_MIPS_SYMTABNO:
obj->symtabno = dynp->d_un.d_val;
break;
case DT_MIPS_GOTSYM:
obj->gotsym = dynp->d_un.d_val;
break;
case DT_MIPS_RLD_MAP:
#ifdef notyet
if (!early)
dbg("Filling in DT_DEBUG entry");
((Elf_Dyn*)dynp)->d_un.d_ptr = (Elf_Addr) &r_debug;
#endif
break;
#endif
- Add support for DT_FLAGS. - Define various things from the most recent ELF spec.
2003-06-18 03:34:29 +00:00
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
case DT_FLAGS_1:
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
if (dynp->d_un.d_val & DF_1_NOOPEN)
obj->z_noopen = true;
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
if ((dynp->d_un.d_val & DF_1_ORIGIN) && trust)
obj->z_origin = true;
if (dynp->d_un.d_val & DF_1_GLOBAL)
/* XXX */;
if (dynp->d_un.d_val & DF_1_BIND_NOW)
obj->bind_now = true;
Implement support for RTLD_NODELETE flag for dlopen() and -z nodelete static linker option. Do it by incrementing reference count on the loaded object and its dependencies. Reviewed by: davidxu, kan
2009-03-30 08:47:28 +00:00
if (dynp->d_un.d_val & DF_1_NODELETE)
obj->z_nodelete = true;
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
break;
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
default:
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
if (!early) {
dbg("Ignoring d_tag %ld = %#lx", (long)dynp->d_tag,
(long)dynp->d_tag);
}
Change the warning about unrecognized entries in the dynamic table to a debug message which is disabled in production builds of the dynamic linker. The condition warned about is normally harmless. PR: bin/12849
1999-09-04 20:14:48 +00:00
break;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
}
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
obj->traced = false;
if (plttype == DT_RELA) {
obj->pltrela = (const Elf_Rela *) obj->pltrel;
obj->pltrel = NULL;
obj->pltrelasize = obj->pltrelsize;
obj->pltrelsize = 0;
}
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
}
static void
digest_dynamic2(Obj_Entry *obj, const Elf_Dyn *dyn_rpath,
const Elf_Dyn *dyn_soname)
{
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
if (obj->z_origin && obj->origin_path == NULL) {
obj->origin_path = xmalloc(PATH_MAX);
if (rtld_dirname_abs(obj->path, obj->origin_path) == -1)
die();
}
if (dyn_rpath != NULL) {
obj->rpath = (char *)obj->strtab + dyn_rpath->d_un.d_val;
if (obj->z_origin)
obj->rpath = origin_subst(obj->rpath, obj->origin_path);
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if (dyn_soname != NULL)
object_add_name(obj, obj->strtab + dyn_soname->d_un.d_val);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
static void
digest_dynamic(Obj_Entry *obj, int early)
{
const Elf_Dyn *dyn_rpath;
const Elf_Dyn *dyn_soname;
digest_dynamic1(obj, early, &dyn_rpath, &dyn_soname);
digest_dynamic2(obj, dyn_rpath, dyn_soname);
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Process a shared object's program header. This is used only for the
* main program, when the kernel has already loaded the main program
* into memory before calling the dynamic linker. It creates and
* returns an Obj_Entry structure.
*/
static Obj_Entry *
Change many asserts into normal errors. They were all for conditions caused by invalid shared objects rather than by internal errors. Enable format string mismatch checking for _rtld_error().
1999-07-18 00:02:19 +00:00
digest_phdr(const Elf_Phdr *phdr, int phnum, caddr_t entry, const char *path)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
Obj_Entry *obj;
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
const Elf_Phdr *phlimit = phdr + phnum;
const Elf_Phdr *ph;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
int nsegs = 0;
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
obj = obj_new();
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
for (ph = phdr; ph < phlimit; ph++) {
Calculate relocation base for the main object, and apply the relocation adjustment for all virtual addresses encoded into the ELF structures of it. PIE binary could and should be loaded at non-zero mapbase. For sym_zero pseudosymbol used as a return value from find_symdef() for undefined weak symbols, st_value also should be adjusted, since _rtld_bind corrects symbol values by relocbase. Discussed with: bz Reviewed by: kan Tested by: bz (i386, amd64), bsam (linux) MFC after: some time
2009-10-10 15:27:10 +00:00
if (ph->p_type != PT_PHDR)
continue;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Calculate relocation base for the main object, and apply the relocation adjustment for all virtual addresses encoded into the ELF structures of it. PIE binary could and should be loaded at non-zero mapbase. For sym_zero pseudosymbol used as a return value from find_symdef() for undefined weak symbols, st_value also should be adjusted, since _rtld_bind corrects symbol values by relocbase. Discussed with: bz Reviewed by: kan Tested by: bz (i386, amd64), bsam (linux) MFC after: some time
2009-10-10 15:27:10 +00:00
obj->phdr = phdr;
obj->phsize = ph->p_memsz;
obj->relocbase = (caddr_t)phdr - ph->p_vaddr;
break;
}
for (ph = phdr; ph < phlimit; ph++) {
switch (ph->p_type) {
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Get the actual pathname of the dynamic linker from the executable's PT_INTERP program header entry, to ensure that gdb always finds the right dynamic linker. Use obj->relocbase to simplify a few calculations where appropriate.
1999-08-30 01:54:13 +00:00
case PT_INTERP:
Calculate relocation base for the main object, and apply the relocation adjustment for all virtual addresses encoded into the ELF structures of it. PIE binary could and should be loaded at non-zero mapbase. For sym_zero pseudosymbol used as a return value from find_symdef() for undefined weak symbols, st_value also should be adjusted, since _rtld_bind corrects symbol values by relocbase. Discussed with: bz Reviewed by: kan Tested by: bz (i386, amd64), bsam (linux) MFC after: some time
2009-10-10 15:27:10 +00:00
obj->interp = (const char *)(ph->p_vaddr + obj->relocbase);
Get the actual pathname of the dynamic linker from the executable's PT_INTERP program header entry, to ensure that gdb always finds the right dynamic linker. Use obj->relocbase to simplify a few calculations where appropriate.
1999-08-30 01:54:13 +00:00
break;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
case PT_LOAD:
if (nsegs == 0) { /* First load segment */
obj->vaddrbase = trunc_page(ph->p_vaddr);
Calculate relocation base for the main object, and apply the relocation adjustment for all virtual addresses encoded into the ELF structures of it. PIE binary could and should be loaded at non-zero mapbase. For sym_zero pseudosymbol used as a return value from find_symdef() for undefined weak symbols, st_value also should be adjusted, since _rtld_bind corrects symbol values by relocbase. Discussed with: bz Reviewed by: kan Tested by: bz (i386, amd64), bsam (linux) MFC after: some time
2009-10-10 15:27:10 +00:00
obj->mapbase = obj->vaddrbase + obj->relocbase;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
obj->textsize = round_page(ph->p_vaddr + ph->p_memsz) -
obj->vaddrbase;
} else { /* Last load segment */
obj->mapsize = round_page(ph->p_vaddr + ph->p_memsz) -
obj->vaddrbase;
}
nsegs++;
break;
case PT_DYNAMIC:
Calculate relocation base for the main object, and apply the relocation adjustment for all virtual addresses encoded into the ELF structures of it. PIE binary could and should be loaded at non-zero mapbase. For sym_zero pseudosymbol used as a return value from find_symdef() for undefined weak symbols, st_value also should be adjusted, since _rtld_bind corrects symbol values by relocbase. Discussed with: bz Reviewed by: kan Tested by: bz (i386, amd64), bsam (linux) MFC after: some time
2009-10-10 15:27:10 +00:00
obj->dynamic = (const Elf_Dyn *)(ph->p_vaddr + obj->relocbase);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
break;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
case PT_TLS:
obj->tlsindex = 1;
obj->tlssize = ph->p_memsz;
obj->tlsalign = ph->p_align;
obj->tlsinitsize = ph->p_filesz;
Calculate relocation base for the main object, and apply the relocation adjustment for all virtual addresses encoded into the ELF structures of it. PIE binary could and should be loaded at non-zero mapbase. For sym_zero pseudosymbol used as a return value from find_symdef() for undefined weak symbols, st_value also should be adjusted, since _rtld_bind corrects symbol values by relocbase. Discussed with: bz Reviewed by: kan Tested by: bz (i386, amd64), bsam (linux) MFC after: some time
2009-10-10 15:27:10 +00:00
obj->tlsinit = (void*)(ph->p_vaddr + obj->relocbase);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
break;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
}
Put back a test for binaries with no PT_LOAD entries I over-jealosly removed in r1.69. Apploved by: re (rwatson)
2002-11-29 16:41:31 +00:00
if (nsegs < 1) {
_rtld_error("%s: too few PT_LOAD segments", path);
return NULL;
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
obj->entry = entry;
return obj;
}
static Obj_Entry *
dlcheck(void *handle)
{
Obj_Entry *obj;
for (obj = obj_list; obj != NULL; obj = obj->next)
if (obj == (Obj_Entry *) handle)
break;
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
if (obj == NULL || obj->refcount == 0 || obj->dl_refcount == 0) {
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
_rtld_error("Invalid shared object handle %p", handle);
return NULL;
}
return obj;
}
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
/*
* If the given object is already in the donelist, return true. Otherwise
* add the object to the list and return false.
*/
static bool
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
donelist_check(DoneList *dlp, const Obj_Entry *obj)
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
{
unsigned int i;
for (i = 0; i < dlp->num_used; i++)
if (dlp->objs[i] == obj)
return true;
/*
* Our donelist allocation should always be sufficient. But if
* our threads locking isn't working properly, more shared objects
* could have been loaded since we allocated the list. That should
* never happen, but we'll handle it properly just in case it does.
*/
if (dlp->num_used < dlp->num_alloc)
dlp->objs[dlp->num_used++] = obj;
return false;
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Hash function for symbol table lookup. Don't even think about changing
* this. It is specified by the System V ABI.
*/
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
unsigned long
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
elf_hash(const char *name)
{
const unsigned char *p = (const unsigned char *) name;
unsigned long h = 0;
unsigned long g;
while (*p != '\0') {
h = (h << 4) + *p++;
if ((g = h & 0xf0000000) != 0)
h ^= g >> 24;
h &= ~g;
}
return h;
}
/*
* Find the library with the given name, and return its full pathname.
* The returned string is dynamically allocated. Generates an error
* message and returns NULL if the library cannot be found.
*
* If the second argument is non-NULL, then it refers to an already-
* loaded shared object, whose library search path will be searched.
Implement ldconfig functionality for ELF. The hints are stored in a different file than the a.out hints, namely, "/var/run/ld-elf.so.hints". These hints consist only of the directory search path. There is no hash table as in the a.out hints, because ELF doesn't have to search for the file with the highest minor version number. (It doesn't have minor version numbers at all.) A single run of ldconfig updates either the a.out hints or the ELF hints, but not both. The set of hints to process is selected in the usual way, via /etc/objformat, or ${OBJFORMAT}, or the "-aout" or "-elf" command line option. The rationale is that you probably want to search different directories for ELF than for a.out. "ldconfig -r" is faked up to produce output like we are used to, except that for ELF there are no minor version numbers. This should enable "ldconfig -r" to be used for checking LIB_DEPENDS in ports even for ELF. I implemented the ELF functionality in a new source file, with an eye toward eliminating the a.out code entirely at some point in the future.
1998-09-05 03:31:00 +00:00
*
* The search order is:
* LD_LIBRARY_PATH
Sync comment with code's reality. MFC after: 1 week
2003-11-14 12:56:56 +00:00
* rpath in the referencing file
Implement ldconfig functionality for ELF. The hints are stored in a different file than the a.out hints, namely, "/var/run/ld-elf.so.hints". These hints consist only of the directory search path. There is no hash table as in the a.out hints, because ELF doesn't have to search for the file with the highest minor version number. (It doesn't have minor version numbers at all.) A single run of ldconfig updates either the a.out hints or the ELF hints, but not both. The set of hints to process is selected in the usual way, via /etc/objformat, or ${OBJFORMAT}, or the "-aout" or "-elf" command line option. The rationale is that you probably want to search different directories for ELF than for a.out. "ldconfig -r" is faked up to produce output like we are used to, except that for ELF there are no minor version numbers. This should enable "ldconfig -r" to be used for checking LIB_DEPENDS in ports even for ELF. I implemented the ELF functionality in a new source file, with an eye toward eliminating the a.out code entirely at some point in the future.
1998-09-05 03:31:00 +00:00
* ldconfig hints
Prepend /lib to the builtin library search path in rtld.
2003-08-17 07:55:17 +00:00
* /lib:/usr/lib
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
*/
static char *
Dynamic object dependency mapping: libmap. This is an optional feature, disabled by default. This will be useful to people testing the various POSIX threading libraries under -CURRENT but can easily serve other needs.
2003-04-07 16:21:26 +00:00
find_library(const char *xname, const Obj_Entry *refobj)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
char *pathname;
Dynamic object dependency mapping: libmap. This is an optional feature, disabled by default. This will be useful to people testing the various POSIX threading libraries under -CURRENT but can easily serve other needs.
2003-04-07 16:21:26 +00:00
char *name;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Dynamic object dependency mapping: libmap. This is an optional feature, disabled by default. This will be useful to people testing the various POSIX threading libraries under -CURRENT but can easily serve other needs.
2003-04-07 16:21:26 +00:00
if (strchr(xname, '/') != NULL) { /* Hard coded pathname */
if (xname[0] != '/' && !trust) {
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
_rtld_error("Absolute pathname required for shared object \"%s\"",
Dynamic object dependency mapping: libmap. This is an optional feature, disabled by default. This will be useful to people testing the various POSIX threading libraries under -CURRENT but can easily serve other needs.
2003-04-07 16:21:26 +00:00
xname);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return NULL;
}
Do not dereference NULL pointer. refobj is NULL for the objects that are preloaded. Reported and tested by: ed
2009-03-28 15:54:08 +00:00
if (refobj != NULL && refobj->z_origin)
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
return origin_subst(xname, refobj->origin_path);
else
return xstrdup(xname);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Avoid a NULL pointer dereference.
2003-06-18 16:17:13 +00:00
if (libmap_disable || (refobj == NULL) ||
(name = lm_find(refobj->path, xname)) == NULL)
Dynamic object dependency mapping: libmap. This is an optional feature, disabled by default. This will be useful to people testing the various POSIX threading libraries under -CURRENT but can easily serve other needs.
2003-04-07 16:21:26 +00:00
name = (char *)xname;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
dbg(" Searching for \"%s\"", name);
Change the library search order so that LD_LIBRARY_PATH overrides all others. PR: bin/28191 MFC after: 2 weeks
2002-01-25 16:35:43 +00:00
if ((pathname = search_library_path(name, ld_library_path)) != NULL ||
(refobj != NULL &&
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
(pathname = search_library_path(name, refobj->rpath)) != NULL) ||
The ELF specification says that the RPATH in the executable or shared object takes precedence over LD_LIBRARY_PATH. Make the dynamic linker do it that way.
1999-04-09 06:42:00 +00:00
(pathname = search_library_path(name, gethints())) != NULL ||
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
(pathname = search_library_path(name, STANDARD_LIBRARY_PATH)) != NULL)
return pathname;
give out a little more information in case of a missing dependency PR: 56549 Submitted by: edwin Reviewed by: joerg, ru Approved by: joerg MFC after: 2 weeks
2004-05-28 00:05:28 +00:00
if(refobj != NULL && refobj->path != NULL) {
_rtld_error("Shared object \"%s\" not found, required by \"%s\"",
name, basename(refobj->path));
} else {
_rtld_error("Shared object \"%s\" not found", name);
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return NULL;
}
/*
* Given a symbol number in a referencing object, find the corresponding
* definition of the symbol. Returns a pointer to the symbol, or NULL if
* no definition was found. Returns a pointer to the Obj_Entry of the
* defining object via the reference parameter DEFOBJ_OUT.
*/
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
const Elf_Sym *
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
find_symdef(unsigned long symnum, const Obj_Entry *refobj,
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
const Obj_Entry **defobj_out, int flags, SymCache *cache)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
const Elf_Sym *ref;
Simplify the logic in find_symdef().
1999-08-30 01:24:08 +00:00
const Elf_Sym *def;
const Obj_Entry *defobj;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
const Ver_Entry *ventry;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
const char *name;
unsigned long hash;
Performance improvements for the ELF dynamic linker. These particularly help programs which load many shared libraries with a lot of relocations. Large C++ programs such as are found in KDE are a prime example. While relocating a shared object, maintain a vector of symbols which have already been looked up, directly indexed by symbol number. Typically, symbols which are referenced by a relocation entry are referenced by many of them. This is the same optimization I made to the a.out dynamic linker in 1995 (rtld.c revision 1.30). Also, compare the first character of a sought-after symbol with its symbol table entry before calling strcmp(). On a PII/400 these changes reduce the start-up time of a typical KDE program from 833 msec (elapsed) to 370 msec. MFC after: 5 days
2001-05-05 23:21:05 +00:00
/*
* If we have already found this symbol, get the information from
* the cache.
*/
if (symnum >= refobj->nchains)
return NULL; /* Bad object */
if (cache != NULL && cache[symnum].sym != NULL) {
*defobj_out = cache[symnum].obj;
return cache[symnum].sym;
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
ref = refobj->symtab + symnum;
name = refobj->strtab + ref->st_name;
When honoring -Bsymbolic, still keep searching if only a weak definition was found in the referencing object.
1999-08-30 01:25:38 +00:00
defobj = NULL;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Don't do symbol lookups for local symbols. The symbol index in the relocation identifies the symbol to which we need to bind. This solves a problem seen on ia64 where the symbol hash table does not contain local symbols and thus resulted in unresolved symbols. Tested on: alpha, i386, ia64
2002-04-27 02:48:29 +00:00
/*
* We don't have to do a full scale lookup if the symbol is local.
* We know it will bind to the instance in this load module; to
* which we already have a pointer (ie ref). By not doing a lookup,
* we not only improve performance, but it also avoids unresolvable
* symbols when local symbols are not in the hash table. This has
* been seen with the ia64 toolchain.
*/
if (ELF_ST_BIND(ref->st_info) != STB_LOCAL) {
if (ELF_ST_TYPE(ref->st_info) == STT_SECTION) {
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
_rtld_error("%s: Bogus symbol table entry %lu", refobj->path,
symnum);
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
ventry = fetch_ventry(refobj, symnum);
Don't do symbol lookups for local symbols. The symbol index in the relocation identifies the symbol to which we need to bind. This solves a problem seen on ia64 where the symbol hash table does not contain local symbols and thus resulted in unresolved symbols. Tested on: alpha, i386, ia64
2002-04-27 02:48:29 +00:00
hash = elf_hash(name);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
def = symlook_default(name, hash, refobj, &defobj, ventry, flags);
Don't do symbol lookups for local symbols. The symbol index in the relocation identifies the symbol to which we need to bind. This solves a problem seen on ia64 where the symbol hash table does not contain local symbols and thus resulted in unresolved symbols. Tested on: alpha, i386, ia64
2002-04-27 02:48:29 +00:00
} else {
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
def = ref;
defobj = refobj;
Don't do symbol lookups for local symbols. The symbol index in the relocation identifies the symbol to which we need to bind. This solves a problem seen on ia64 where the symbol hash table does not contain local symbols and thus resulted in unresolved symbols. Tested on: alpha, i386, ia64
2002-04-27 02:48:29 +00:00
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Simplify the logic in find_symdef().
1999-08-30 01:24:08 +00:00
/*
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
* If we found no definition and the reference is weak, treat the
Simplify the logic in find_symdef().
1999-08-30 01:24:08 +00:00
* symbol as having the value zero.
*/
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
if (def == NULL && ELF_ST_BIND(ref->st_info) == STB_WEAK) {
def = &sym_zero;
defobj = obj_main;
Resolve undefined weak references to a value of 0. This solves the "__deregister_frame_info" problem that was seen when combining a program linked using the old gcc with shared libraries that were built using egcs.
1999-04-05 02:36:40 +00:00
}
Performance improvements for the ELF dynamic linker. These particularly help programs which load many shared libraries with a lot of relocations. Large C++ programs such as are found in KDE are a prime example. While relocating a shared object, maintain a vector of symbols which have already been looked up, directly indexed by symbol number. Typically, symbols which are referenced by a relocation entry are referenced by many of them. This is the same optimization I made to the a.out dynamic linker in 1995 (rtld.c revision 1.30). Also, compare the first character of a sought-after symbol with its symbol table entry before calling strcmp(). On a PII/400 these changes reduce the start-up time of a typical KDE program from 833 msec (elapsed) to 370 msec. MFC after: 5 days
2001-05-05 23:21:05 +00:00
if (def != NULL) {
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
*defobj_out = defobj;
Performance improvements for the ELF dynamic linker. These particularly help programs which load many shared libraries with a lot of relocations. Large C++ programs such as are found in KDE are a prime example. While relocating a shared object, maintain a vector of symbols which have already been looked up, directly indexed by symbol number. Typically, symbols which are referenced by a relocation entry are referenced by many of them. This is the same optimization I made to the a.out dynamic linker in 1995 (rtld.c revision 1.30). Also, compare the first character of a sought-after symbol with its symbol table entry before calling strcmp(). On a PII/400 these changes reduce the start-up time of a typical KDE program from 833 msec (elapsed) to 370 msec. MFC after: 5 days
2001-05-05 23:21:05 +00:00
/* Record the information in the cache to avoid subsequent lookups. */
if (cache != NULL) {
cache[symnum].sym = def;
cache[symnum].obj = defobj;
}
Add ia64 support. Various adjustments were made to existing targets to cope with a few interface changes required by the ia64. In particular, function pointers on ia64 need special treatment in rtld.
2001-10-15 18:48:42 +00:00
} else {
if (refobj != &obj_rtld)
_rtld_error("%s: Undefined symbol \"%s\"", refobj->path, name);
}
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
return def;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Implement ldconfig functionality for ELF. The hints are stored in a different file than the a.out hints, namely, "/var/run/ld-elf.so.hints". These hints consist only of the directory search path. There is no hash table as in the a.out hints, because ELF doesn't have to search for the file with the highest minor version number. (It doesn't have minor version numbers at all.) A single run of ldconfig updates either the a.out hints or the ELF hints, but not both. The set of hints to process is selected in the usual way, via /etc/objformat, or ${OBJFORMAT}, or the "-aout" or "-elf" command line option. The rationale is that you probably want to search different directories for ELF than for a.out. "ldconfig -r" is faked up to produce output like we are used to, except that for ELF there are no minor version numbers. This should enable "ldconfig -r" to be used for checking LIB_DEPENDS in ports even for ELF. I implemented the ELF functionality in a new source file, with an eye toward eliminating the a.out code entirely at some point in the future.
1998-09-05 03:31:00 +00:00
/*
* Return the search path from the ldconfig hints file, reading it if
* necessary. Returns NULL if there are problems with the hints file,
* or if the search path there is empty.
*/
static const char *
gethints(void)
{
static char *hints;
if (hints == NULL) {
int fd;
struct elfhints_hdr hdr;
char *p;
/* Keep from trying again in case the hints file is bad. */
hints = "";
Support for a new environment variable, LD_ELF_HINTS_PATH for overriding the rtld hints file. This environment variable would be unset if the process is considered as tainted with setuid/setgid. This feature gives a convenient way of using a custom set of shared library that is not located in the default location and switch back. Feature requested by: iXsystems Original patch by: John Hixson MFC after: 2 weeks
2009-03-23 16:49:00 +00:00
if ((fd = open(ld_elf_hints_path, O_RDONLY)) == -1)
Implement ldconfig functionality for ELF. The hints are stored in a different file than the a.out hints, namely, "/var/run/ld-elf.so.hints". These hints consist only of the directory search path. There is no hash table as in the a.out hints, because ELF doesn't have to search for the file with the highest minor version number. (It doesn't have minor version numbers at all.) A single run of ldconfig updates either the a.out hints or the ELF hints, but not both. The set of hints to process is selected in the usual way, via /etc/objformat, or ${OBJFORMAT}, or the "-aout" or "-elf" command line option. The rationale is that you probably want to search different directories for ELF than for a.out. "ldconfig -r" is faked up to produce output like we are used to, except that for ELF there are no minor version numbers. This should enable "ldconfig -r" to be used for checking LIB_DEPENDS in ports even for ELF. I implemented the ELF functionality in a new source file, with an eye toward eliminating the a.out code entirely at some point in the future.
1998-09-05 03:31:00 +00:00
return NULL;
if (read(fd, &hdr, sizeof hdr) != sizeof hdr ||
hdr.magic != ELFHINTS_MAGIC ||
hdr.version != 1) {
close(fd);
return NULL;
}
p = xmalloc(hdr.dirlistlen + 1);
if (lseek(fd, hdr.strtab + hdr.dirlist, SEEK_SET) == -1 ||
Fix signed/unsigned comparison warnings.
2003-05-04 00:56:00 +00:00
read(fd, p, hdr.dirlistlen + 1) != (ssize_t)hdr.dirlistlen + 1) {
Implement ldconfig functionality for ELF. The hints are stored in a different file than the a.out hints, namely, "/var/run/ld-elf.so.hints". These hints consist only of the directory search path. There is no hash table as in the a.out hints, because ELF doesn't have to search for the file with the highest minor version number. (It doesn't have minor version numbers at all.) A single run of ldconfig updates either the a.out hints or the ELF hints, but not both. The set of hints to process is selected in the usual way, via /etc/objformat, or ${OBJFORMAT}, or the "-aout" or "-elf" command line option. The rationale is that you probably want to search different directories for ELF than for a.out. "ldconfig -r" is faked up to produce output like we are used to, except that for ELF there are no minor version numbers. This should enable "ldconfig -r" to be used for checking LIB_DEPENDS in ports even for ELF. I implemented the ELF functionality in a new source file, with an eye toward eliminating the a.out code entirely at some point in the future.
1998-09-05 03:31:00 +00:00
free(p);
close(fd);
return NULL;
}
hints = p;
close(fd);
}
return hints[0] != '\0' ? hints : NULL;
}
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static void
init_dag(Obj_Entry *root)
{
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
DoneList donelist;
If dlopen() is called for the dso that has been already loaded as a dependency, then the dso never has its DAG initialized. Empty DAG makes ref_dag() call in dlopen() a nop, and the dso refcount is off by one. Initialize the DAG on the first dlopen() call, using a boolean flag to prevent double initialization. From the PR (edited): Assume we have a library liba.so, containing a function a(), and a library libb.so, containing function b(). liba.so needs functionality from libb.so, so liba.so links in libb.so. An application doesn't know about the relation between these libraries, but needs to call a() and b(). It dlopen()s liba.so and obtains a pointer to a(), then it dlopen()s libb.so and obtains a pointer to b(). As soon as the application doesn't need a() anymore, it dlclose()s liba.so. Expected result: the pointer to b() is still valid and can be called Actual result: the pointer to b() has become invalid, even though the application did not dlclose() the handle to libb.so. On calling b(), the application crashes with a segmentation fault. PR: misc/151861 Based on patch by: jh Reviewed by: kan Tested by: Arjan van Leeuwen <freebsd-maintainer opera com> MFC after: 1 week
2010-11-03 09:23:08 +00:00
if (root->dag_inited)
Fix style. MFC after: 6 days
2010-11-04 09:19:14 +00:00
return;
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
donelist_init(&donelist);
init_dag1(root, root, &donelist);
If dlopen() is called for the dso that has been already loaded as a dependency, then the dso never has its DAG initialized. Empty DAG makes ref_dag() call in dlopen() a nop, and the dso refcount is off by one. Initialize the DAG on the first dlopen() call, using a boolean flag to prevent double initialization. From the PR (edited): Assume we have a library liba.so, containing a function a(), and a library libb.so, containing function b(). liba.so needs functionality from libb.so, so liba.so links in libb.so. An application doesn't know about the relation between these libraries, but needs to call a() and b(). It dlopen()s liba.so and obtains a pointer to a(), then it dlopen()s libb.so and obtains a pointer to b(). As soon as the application doesn't need a() anymore, it dlclose()s liba.so. Expected result: the pointer to b() is still valid and can be called Actual result: the pointer to b() has become invalid, even though the application did not dlclose() the handle to libb.so. On calling b(), the application crashes with a segmentation fault. PR: misc/151861 Based on patch by: jh Reviewed by: kan Tested by: Arjan van Leeuwen <freebsd-maintainer opera com> MFC after: 1 week
2010-11-03 09:23:08 +00:00
root->dag_inited = true;
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
}
static void
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
init_dag1(Obj_Entry *root, Obj_Entry *obj, DoneList *dlp)
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
{
const Needed_Entry *needed;
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
if (donelist_check(dlp, obj))
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
return;
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_push_tail(&obj->dldags, root);
objlist_push_tail(&root->dagmembers, obj);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
for (needed = obj->needed; needed != NULL; needed = needed->next)
if (needed->obj != NULL)
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
init_dag1(root, needed->obj, dlp);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Initialize the dynamic linker. The argument is the address at which
* the dynamic linker has been mapped into memory. The primary task of
* this function is to relocate the dynamic linker.
*/
static void
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
init_rtld(caddr_t mapbase, Elf_Auxinfo **aux_info)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
Obj_Entry objtmp; /* Temporary rtld object */
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
const Elf_Dyn *dyn_rpath;
const Elf_Dyn *dyn_soname;
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
Make the pathname pointed to by the Obj_Entry structure for the dynamic linker itself dynamically allocated. All of them are supposed to be dynamically allocated, but we cheated before. It made gdb unhappy under some circumstances.
1998-09-15 21:07:52 +00:00
/*
* Conjure up an Obj_Entry structure for the dynamic linker.
*
Fix typo in comment. Submitted by: Christoph Mallon
2009-06-23 09:50:50 +00:00
* The "path" member can't be initialized yet because string constants
Fix a typo in the same comment, one line below. Submitted by: bf1783 googlemail com
2009-06-23 14:12:49 +00:00
* cannot yet be accessed. Below we will set it correctly.
Make the pathname pointed to by the Obj_Entry structure for the dynamic linker itself dynamically allocated. All of them are supposed to be dynamically allocated, but we cheated before. It made gdb unhappy under some circumstances.
1998-09-15 21:07:52 +00:00
*/
Initialise some uninitialised variables. Thanks to: valgrind
2003-12-31 15:10:41 +00:00
memset(&objtmp, 0, sizeof(objtmp));
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
objtmp.path = NULL;
objtmp.rtld = true;
objtmp.mapbase = mapbase;
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
#ifdef PIC
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
objtmp.relocbase = mapbase;
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
#endif
Fix the problem that surfaced with the new binutils import on sparc64 (and that is for now being worked around by a binutils patch). The rtld code tested &_DYNAMIC against 0 to see whether rtld itself was built as PIC or not. While the sparc64 MD code did not rely on the preset value of the GOT slot for _DYNAMIC any more due to previous binutils changes, it still used to not be 0, so that this check did work. The new binutils do however initialize this slot with 0. As a consequence, rtld would not properly initialize itself and crash. Fix that by introducing a new macro, RTLD_IS_DYNAMIC, to take the role of this test. For sparc64, it is implemented using the rtld_dynamic() code that was already there. If an architecture does not provide its own implementation, we default to the old check. While being there, mark _DYNAMIC as a weak symbol in the sparc64 rtld_start.S. This is needed in the LDSCRIPT case, which is however not currently supported for want of an actual ldscript. Sanity checked with md5 on alpha, amd64, i386 and ia64.
2004-06-18 02:01:37 +00:00
if (RTLD_IS_DYNAMIC()) {
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
objtmp.dynamic = rtld_dynamic(&objtmp);
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
digest_dynamic1(&objtmp, 1, &dyn_rpath, &dyn_soname);
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
assert(objtmp.needed == NULL);
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
#if !defined(__mips__)
/* MIPS and SH{3,5} have a bogus DT_TEXTREL. */
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
assert(!objtmp.textrel);
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
#endif
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
/*
* Temporarily put the dynamic linker entry into the object list, so
* that symbols can be found.
*/
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
relocate_objects(&objtmp, true, &objtmp);
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
/* Initialize the object list. */
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
obj_tail = &obj_list;
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
/* Now that non-local variables can be accesses, copy out obj_rtld. */
memcpy(&obj_rtld, &objtmp, sizeof(obj_rtld));
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
if (aux_info[AT_PAGESZ] != NULL)
pagesize = aux_info[AT_PAGESZ]->a_un.a_val;
if (aux_info[AT_OSRELDATE] != NULL)
osreldate = aux_info[AT_OSRELDATE]->a_un.a_val;
digest_dynamic2(&obj_rtld, dyn_rpath, dyn_soname);
Make the pathname pointed to by the Obj_Entry structure for the dynamic linker itself dynamically allocated. All of them are supposed to be dynamically allocated, but we cheated before. It made gdb unhappy under some circumstances.
1998-09-15 21:07:52 +00:00
/* Replace the path with a dynamically allocated copy. */
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
obj_rtld.path = xstrdup(PATH_RTLD);
Make the pathname pointed to by the Obj_Entry structure for the dynamic linker itself dynamically allocated. All of them are supposed to be dynamically allocated, but we cheated before. It made gdb unhappy under some circumstances.
1998-09-15 21:07:52 +00:00
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
r_debug.r_brk = r_debug_state;
r_debug.r_state = RT_CONSISTENT;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/*
* Add the init functions from a needed object list (and its recursive
* needed objects) to "list". This is not used directly; it is a helper
* function for initlist_add_objects(). The write lock must be held
* when this function is called.
*/
static void
initlist_add_neededs(Needed_Entry *needed, Objlist *list)
{
/* Recursively process the successor needed objects. */
if (needed->next != NULL)
initlist_add_neededs(needed->next, list);
/* Process the current needed object. */
if (needed->obj != NULL)
initlist_add_objects(needed->obj, &needed->obj->next, list);
}
/*
* Scan all of the DAGs rooted in the range of objects from "obj" to
* "tail" and add their init functions to "list". This recurses over
* the DAGs and ensure the proper init ordering such that each object's
* needed libraries are initialized before the object itself. At the
* same time, this function adds the objects to the global finalization
* list "list_fini" in the opposite order. The write lock must be
* held when this function is called.
*/
static void
initlist_add_objects(Obj_Entry *obj, Obj_Entry **tail, Objlist *list)
{
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
if (obj->init_scanned || obj->init_done)
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
return;
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
obj->init_scanned = true;
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/* Recursively process the successor objects. */
if (&obj->next != tail)
initlist_add_objects(obj->next, tail, list);
/* Recursively process the needed objects. */
if (obj->needed != NULL)
initlist_add_neededs(obj->needed, list);
/* Add the object to the init list. */
Make NULL a (void*)0 whereever possible, and fix the warnings(-Werror) that this provokes. "Wherever possible" means "In the kernel OR NOT C++" (implying C). There are places where (void *) pointers are not valid, such as for function pointers, but in the special case of (void *)0, agreement settles on it being OK. Most of the fixes were NULL where an integer zero was needed; many of the fixes were NULL where ascii <nul> ('\0') was needed, and a few were just "other". Tested on: i386 sparc64
2004-03-05 08:10:19 +00:00
if (obj->init != (Elf_Addr)NULL)
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_push_tail(list, obj);
/* Add the object to the global fini list in the reverse order. */
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
if (obj->fini != (Elf_Addr)NULL && !obj->on_fini_list) {
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_push_head(&list_fini, obj);
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
obj->on_fini_list = true;
}
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
}
Add ia64 support. Various adjustments were made to existing targets to cope with a few interface changes required by the ia64. In particular, function pointers on ia64 need special treatment in rtld.
2001-10-15 18:48:42 +00:00
#ifndef FPTR_TARGET
#define FPTR_TARGET(f) ((Elf_Addr) (f))
#endif
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Given a shared object, traverse its list of needed objects, and load
* each of them. Returns 0 on success. Generates an error message and
* returns -1 on failure.
*/
static int
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
load_needed_objects(Obj_Entry *first, int flags)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Implement support for RTLD_NODELETE flag for dlopen() and -z nodelete static linker option. Do it by incrementing reference count on the loaded object and its dependencies. Reviewed by: davidxu, kan
2009-03-30 08:47:28 +00:00
Obj_Entry *obj, *obj1;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
for (obj = first; obj != NULL; obj = obj->next) {
Needed_Entry *needed;
for (needed = obj->needed; needed != NULL; needed = needed->next) {
Implement RTLD_NOLOAD flag for dlopen(3). Requested and tested by: jkim Reviewed by: kan Approved by: re (kensmith)
2009-07-17 19:45:42 +00:00
obj1 = needed->obj = load_object(obj->strtab + needed->name, obj,
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
flags & ~RTLD_LO_NOLOAD);
Implement support for RTLD_NODELETE flag for dlopen() and -z nodelete static linker option. Do it by incrementing reference count on the loaded object and its dependencies. Reviewed by: davidxu, kan
2009-03-30 08:47:28 +00:00
if (obj1 == NULL && !ld_tracing)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return -1;
Implement support for RTLD_NODELETE flag for dlopen() and -z nodelete static linker option. Do it by incrementing reference count on the loaded object and its dependencies. Reviewed by: davidxu, kan
2009-03-30 08:47:28 +00:00
if (obj1 != NULL && obj1->z_nodelete && !obj1->ref_nodel) {
dbg("obj %s nodelete", obj1->path);
init_dag(obj1);
ref_dag(obj1);
obj1->ref_nodel = true;
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
}
return 0;
}
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
static int
load_preload_objects(void)
{
char *p = ld_preload;
Allow files in LD_PRELOAD to be separated by white space, like Solaris and Linux.
2000-01-22 22:20:05 +00:00
static const char delim[] = " \t:;";
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
if (p == NULL)
Make NULL a (void*)0 whereever possible, and fix the warnings(-Werror) that this provokes. "Wherever possible" means "In the kernel OR NOT C++" (implying C). There are places where (void *) pointers are not valid, such as for function pointers, but in the special case of (void *)0, agreement settles on it being OK. Most of the fixes were NULL where an integer zero was needed; many of the fixes were NULL where ascii <nul> ('\0') was needed, and a few were just "other". Tested on: i386 sparc64
2004-03-05 08:10:19 +00:00
return 0;
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
Allow files in LD_PRELOAD to be separated by white space, like Solaris and Linux.
2000-01-22 22:20:05 +00:00
p += strspn(p, delim);
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
while (*p != '\0') {
Allow files in LD_PRELOAD to be separated by white space, like Solaris and Linux.
2000-01-22 22:20:05 +00:00
size_t len = strcspn(p, delim);
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
char savech;
savech = p[len];
p[len] = '\0';
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
if (load_object(p, NULL, 0) == NULL)
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
return -1; /* XXX - cleanup */
p[len] = savech;
p += len;
Allow files in LD_PRELOAD to be separated by white space, like Solaris and Linux.
2000-01-22 22:20:05 +00:00
p += strspn(p, delim);
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
}
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_PRELOAD_FINISHED, NULL, NULL, 0, 0, NULL);
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
return 0;
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
* Load a shared object into memory, if it is not already loaded.
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
*
* Returns a pointer to the Obj_Entry for the object. Returns NULL
* on failure.
*/
static Obj_Entry *
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
load_object(const char *name, const Obj_Entry *refobj, int flags)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Obj_Entry *obj;
When checking to see if a shared object is already loaded, look for a device/inode match if no pathname match is found.
1999-08-30 01:50:41 +00:00
int fd = -1;
struct stat sb;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
char *path;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
for (obj = obj_list->next; obj != NULL; obj = obj->next)
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if (object_match_name(obj, name))
return obj;
path = find_library(name, refobj);
if (path == NULL)
return NULL;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
When checking to see if a shared object is already loaded, look for a device/inode match if no pathname match is found.
1999-08-30 01:50:41 +00:00
/*
* If we didn't find a match by pathname, open the file and check
* again by device and inode. This avoids false mismatches caused
* by multiple links or ".." in pathnames.
*
* To avoid a race, we open the file and use fstat() rather than
* using stat().
*/
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if ((fd = open(path, O_RDONLY)) == -1) {
_rtld_error("Cannot open \"%s\"", path);
free(path);
return NULL;
}
if (fstat(fd, &sb) == -1) {
_rtld_error("Cannot fstat \"%s\"", path);
close(fd);
free(path);
return NULL;
}
for (obj = obj_list->next; obj != NULL; obj = obj->next) {
if (obj->ino == sb.st_ino && obj->dev == sb.st_dev) {
When checking to see if a shared object is already loaded, look for a device/inode match if no pathname match is found.
1999-08-30 01:50:41 +00:00
close(fd);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
break;
When checking to see if a shared object is already loaded, look for a device/inode match if no pathname match is found.
1999-08-30 01:50:41 +00:00
}
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if (obj != NULL) {
object_add_name(obj, name);
free(path);
close(fd);
return obj;
}
Plug possible memory leak. Found by: Coverity MFC after: 2 weeks
2010-06-03 00:25:43 +00:00
if (flags & RTLD_LO_NOLOAD) {
free(path);
Implement RTLD_NOLOAD flag for dlopen(3). Requested and tested by: jkim Reviewed by: kan Approved by: re (kensmith)
2009-07-17 19:45:42 +00:00
return (NULL);
Plug possible memory leak. Found by: Coverity MFC after: 2 weeks
2010-06-03 00:25:43 +00:00
}
When checking to see if a shared object is already loaded, look for a device/inode match if no pathname match is found.
1999-08-30 01:50:41 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
/* First use of this object, so we must map it in */
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
obj = do_load_object(fd, name, path, &sb, flags);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if (obj == NULL)
free(path);
close(fd);
return obj;
}
static Obj_Entry *
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
do_load_object(int fd, const char *name, char *path, struct stat *sbp,
int flags)
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
{
Obj_Entry *obj;
struct statfs fs;
/*
* but first, make sure that environment variables haven't been
* used to circumvent the noexec flag on a filesystem.
*/
if (dangerous_ld_env) {
if (fstatfs(fd, &fs) != 0) {
_rtld_error("Cannot fstatfs \"%s\"", path);
If "dangerous" environment variables (LD_PRELOAD, LD_LIBMAP, LD_LIBMAP_DISABLE, LD_LIBRARY_PATH) are used, then make sure the libraries being loaded aren't on a noexec-mounted filesystem. This is a compromise position: I'm assuming that nobody will be silly enough to set the noexec mount flag on part of the default library path, in order to avoid adding extra overhead into the common case (where those environment variables aren't used). Discussed with: csjp, secteam MFC after: 1 week
2005-03-24 10:12:29 +00:00
return NULL;
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if (fs.f_flags & MNT_NOEXEC) {
_rtld_error("Cannot execute objects on %s\n", fs.f_mntonname);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return NULL;
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
}
dbg("loading \"%s\"", path);
obj = map_object(fd, path, sbp);
if (obj == NULL)
return NULL;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
object_add_name(obj, name);
obj->path = path;
digest_dynamic(obj, 0);
Allow to load not-openable dso when tracing. This fixes ldd on such dso or dso linked to non-openable object. Remove '\n' at the end of error message. End comments with dot. MFC after: 3 weeks (together with r199829)
2009-11-28 14:29:32 +00:00
if (obj->z_noopen && (flags & (RTLD_LO_DLOPEN | RTLD_LO_TRACE)) ==
RTLD_LO_DLOPEN) {
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
dbg("refusing to load non-loadable \"%s\"", obj->path);
Allow to load not-openable dso when tracing. This fixes ldd on such dso or dso linked to non-openable object. Remove '\n' at the end of error message. End comments with dot. MFC after: 3 weeks (together with r199829)
2009-11-28 14:29:32 +00:00
_rtld_error("Cannot dlopen non-loadable %s", obj->path);
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
munmap(obj->mapbase, obj->mapsize);
obj_free(obj);
return (NULL);
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
*obj_tail = obj;
obj_tail = &obj->next;
obj_count++;
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
obj_loads++;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
linkmap_add(obj); /* for GDB & dlinfo() */
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
dbg(" %p .. %p: %s", obj->mapbase,
obj->mapbase + obj->mapsize - 1, obj->path);
if (obj->textrel)
dbg(" WARNING: %s has impure text", obj->path);
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_LOAD_OBJECT, obj, obj->mapbase, obj->mapsize, 0,
obj->path);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return obj;
}
static Obj_Entry *
obj_from_addr(const void *addr)
{
Obj_Entry *obj;
for (obj = obj_list; obj != NULL; obj = obj->next) {
if (addr < (void *) obj->mapbase)
continue;
Do not depend on existence of _end symbol in obj_from_addr, use obj->mapbase and obj->mapsize instead. Prompted by: OpenOffice debugging session at last BSDCon.
2004-02-25 17:06:16 +00:00
if (addr < (void *) (obj->mapbase + obj->mapsize))
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return obj;
}
return NULL;
}
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/*
* Call the finalization functions for each of the objects in "list"
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
* belonging to the DAG of "root" and referenced once. If NULL "root"
* is specified, every finalization function will be called regardless
* of the reference count and the list elements won't be freed. All of
* the objects are expected to have non-NULL fini functions.
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
*/
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static void
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
objlist_call_fini(Objlist *list, Obj_Entry *root, int *lockstate)
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
{
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
Objlist_Entry *elm;
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
char *saved_msg;
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
assert(root == NULL || root->refcount == 1);
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
/*
* Preserve the current error message since a fini function might
* call into the dynamic linker and overwrite it.
*/
saved_msg = errmsg_save();
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
do {
STAILQ_FOREACH(elm, list, link) {
if (root != NULL && (elm->obj->refcount != 1 ||
objlist_find(&root->dagmembers, elm->obj) == NULL))
continue;
Update rtld for the "new" ia64 ABI. In the old toolchain, the DT_INIT and DT_FINI tags pointed to fptr records. In 2.11.2, it points to the actuall address of the function. On IA64 you cannot just take an address of a function, store it in a function pointer variable and call it.. the function pointers point to a fptr data block that has the target gp and address in it. This is absolutely necessary for using the in-tree binutils toolchain, but (unfortunately) will not work with old shared libraries. Save your old ld-elf.so.1 if you want to use old ones still. Do not mix-and-match. This is a no-op change for i386 and alpha. Reviewed by: dfr
2001-10-29 10:10:10 +00:00
dbg("calling fini function for %s at %p", elm->obj->path,
(void *)elm->obj->fini);
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_FINI_CALL, elm->obj, (void *)elm->obj->fini, 0, 0,
elm->obj->path);
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
/* Remove object from fini list to prevent recursive invocation. */
STAILQ_REMOVE(list, elm, Struct_Objlist_Entry, link);
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
/*
* XXX: If a dlopen() call references an object while the
* fini function is in progress, we might end up trying to
* unload the referenced object in dlclose() or the object
* won't be unloaded although its fini function has been
* called.
*/
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
wlock_release(rtld_bind_lock, *lockstate);
Update rtld for the "new" ia64 ABI. In the old toolchain, the DT_INIT and DT_FINI tags pointed to fptr records. In 2.11.2, it points to the actuall address of the function. On IA64 you cannot just take an address of a function, store it in a function pointer variable and call it.. the function pointers point to a fptr data block that has the target gp and address in it. This is absolutely necessary for using the in-tree binutils toolchain, but (unfortunately) will not work with old shared libraries. Save your old ld-elf.so.1 if you want to use old ones still. Do not mix-and-match. This is a no-op change for i386 and alpha. Reviewed by: dfr
2001-10-29 10:10:10 +00:00
call_initfini_pointer(elm->obj, elm->obj->fini);
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
*lockstate = wlock_acquire(rtld_bind_lock);
/* No need to free anything if process is going down. */
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
if (root != NULL)
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
free(elm);
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
/*
* We must restart the list traversal after every fini call
* because a dlclose() call from the fini function or from
* another thread might have modified the reference counts.
*/
break;
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
}
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
} while (elm != NULL);
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
errmsg_restore(saved_msg);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
}
/*
* Call the initialization functions for each of the objects in
* "list". All of the objects are expected to have non-NULL init
* functions.
*/
static void
Make sure internal rtld malloc routines are not called from unlocked contexts as rtld's malloc is not thread safe and is only supposed to be called with exclusive bind lock already held. The originating PR submitted a patch on top of different pre-requisite workaroud for unsafe dlopen calls, and the patch was midief slighlty to apply to stock sources for the purpose of this commit. Running rtld malloc from unlocked contexts is a bug on its own. PR: 126950 Submited by: Oleg Dolgov
2008-09-03 01:05:32 +00:00
objlist_call_init(Objlist *list, int *lockstate)
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
{
Objlist_Entry *elm;
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
Obj_Entry *obj;
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
char *saved_msg;
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
/*
* Clean init_scanned flag so that objects can be rechecked and
* possibly initialized earlier if any of vectors called below
* cause the change by using dlopen.
*/
for (obj = obj_list; obj != NULL; obj = obj->next)
obj->init_scanned = false;
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
/*
* Preserve the current error message since an init function might
* call into the dynamic linker and overwrite it.
*/
saved_msg = errmsg_save();
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
STAILQ_FOREACH(elm, list, link) {
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
if (elm->obj->init_done) /* Initialized early. */
continue;
Update rtld for the "new" ia64 ABI. In the old toolchain, the DT_INIT and DT_FINI tags pointed to fptr records. In 2.11.2, it points to the actuall address of the function. On IA64 you cannot just take an address of a function, store it in a function pointer variable and call it.. the function pointers point to a fptr data block that has the target gp and address in it. This is absolutely necessary for using the in-tree binutils toolchain, but (unfortunately) will not work with old shared libraries. Save your old ld-elf.so.1 if you want to use old ones still. Do not mix-and-match. This is a no-op change for i386 and alpha. Reviewed by: dfr
2001-10-29 10:10:10 +00:00
dbg("calling init function for %s at %p", elm->obj->path,
(void *)elm->obj->init);
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_INIT_CALL, elm->obj, (void *)elm->obj->init, 0, 0,
elm->obj->path);
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
/*
* Race: other thread might try to use this object before current
* one completes the initilization. Not much can be done here
* without better locking.
*/
elm->obj->init_done = true;
wlock_release(rtld_bind_lock, *lockstate);
Update rtld for the "new" ia64 ABI. In the old toolchain, the DT_INIT and DT_FINI tags pointed to fptr records. In 2.11.2, it points to the actuall address of the function. On IA64 you cannot just take an address of a function, store it in a function pointer variable and call it.. the function pointers point to a fptr data block that has the target gp and address in it. This is absolutely necessary for using the in-tree binutils toolchain, but (unfortunately) will not work with old shared libraries. Save your old ld-elf.so.1 if you want to use old ones still. Do not mix-and-match. This is a no-op change for i386 and alpha. Reviewed by: dfr
2001-10-29 10:10:10 +00:00
call_initfini_pointer(elm->obj, elm->obj->init);
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
*lockstate = wlock_acquire(rtld_bind_lock);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
}
Fix a bug in which a program called dlclose from a destructor and got an assert failure in the dynamic linker.
2001-01-05 04:36:17 +00:00
errmsg_restore(saved_msg);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
}
static void
objlist_clear(Objlist *list)
{
Objlist_Entry *elm;
while (!STAILQ_EMPTY(list)) {
elm = STAILQ_FIRST(list);
STAILQ_REMOVE_HEAD(list, link);
free(elm);
}
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
}
static Objlist_Entry *
objlist_find(Objlist *list, const Obj_Entry *obj)
{
Objlist_Entry *elm;
STAILQ_FOREACH(elm, list, link)
if (elm->obj == obj)
return elm;
return NULL;
}
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
static void
objlist_init(Objlist *list)
{
STAILQ_INIT(list);
}
static void
objlist_push_head(Objlist *list, Obj_Entry *obj)
{
Objlist_Entry *elm;
elm = NEW(Objlist_Entry);
elm->obj = obj;
STAILQ_INSERT_HEAD(list, elm, link);
}
static void
objlist_push_tail(Objlist *list, Obj_Entry *obj)
{
Objlist_Entry *elm;
elm = NEW(Objlist_Entry);
elm->obj = obj;
STAILQ_INSERT_TAIL(list, elm, link);
}
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static void
objlist_remove(Objlist *list, Obj_Entry *obj)
{
Objlist_Entry *elm;
if ((elm = objlist_find(list, obj)) != NULL) {
Back out the previous change to the queue(3) interface. It was not discussed and should probably not happen. Requested by: msmith and others
2000-05-26 02:09:24 +00:00
STAILQ_REMOVE(list, elm, Struct_Objlist_Entry, link);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
free(elm);
}
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Relocate newly-loaded shared objects. The argument is a pointer to
* the Obj_Entry for the first such object. All objects from the first
* to the end of the list of objects are relocated. Returns 0 on success,
* or -1 on failure.
*/
static int
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
relocate_objects(Obj_Entry *first, bool bind_now, Obj_Entry *rtldobj)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Obj_Entry *obj;
for (obj = first; obj != NULL; obj = obj->next) {
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
if (obj != rtldobj)
Make LD_PRELOAD work for ELF.
1998-09-22 02:09:56 +00:00
dbg("relocating \"%s\"", obj->path);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
if (obj->nbuckets == 0 || obj->nchains == 0 || obj->buckets == NULL ||
obj->symtab == NULL || obj->strtab == NULL) {
_rtld_error("%s: Shared object has no run-time symbol table",
obj->path);
return -1;
}
if (obj->textrel) {
/* There are relocations to the write-protected text segment. */
if (mprotect(obj->mapbase, obj->textsize,
PROT_READ|PROT_WRITE|PROT_EXEC) == -1) {
_rtld_error("%s: Cannot write-enable text segment: %s",
obj->path, strerror(errno));
return -1;
}
}
/* Process the non-PLT relocations. */
Minor changes to make this work on sparc64. Approved by: jdp Tested on: alpha, i386, sparc64
2002-04-02 02:19:02 +00:00
if (reloc_non_plt(obj, rtldobj))
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return -1;
if (obj->textrel) { /* Re-protected the text segment. */
if (mprotect(obj->mapbase, obj->textsize,
PROT_READ|PROT_EXEC) == -1) {
_rtld_error("%s: Cannot write-protect text segment: %s",
obj->path, strerror(errno));
return -1;
}
}
/* Process the PLT relocations. */
When a threads package registers locking methods with dllockinit(), figure out which shared object(s) contain the the locking methods and fully bind those objects as if they had been loaded with LD_BIND_NOW=1. The goal is to keep the locking methods from requiring any lazy binding. Otherwise infinite recursion occurs in _rtld_bind. This fixes the infinite recursion problem in the linuxthreads port.
2000-01-29 01:27:04 +00:00
if (reloc_plt(obj) == -1)
return -1;
/* Relocate the jump slots if we are doing immediate binding. */
- Add support for DT_FLAGS. - Define various things from the most recent ELF spec.
2003-06-18 03:34:29 +00:00
if (obj->bind_now || bind_now)
When a threads package registers locking methods with dllockinit(), figure out which shared object(s) contain the the locking methods and fully bind those objects as if they had been loaded with LD_BIND_NOW=1. The goal is to keep the locking methods from requiring any lazy binding. Otherwise infinite recursion occurs in _rtld_bind. This fixes the infinite recursion problem in the linuxthreads port.
2000-01-29 01:27:04 +00:00
if (reloc_jmpslots(obj) == -1)
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
return -1;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
When a threads package registers locking methods with dllockinit(), figure out which shared object(s) contain the the locking methods and fully bind those objects as if they had been loaded with LD_BIND_NOW=1. The goal is to keep the locking methods from requiring any lazy binding. Otherwise infinite recursion occurs in _rtld_bind. This fixes the infinite recursion problem in the linuxthreads port.
2000-01-29 01:27:04 +00:00
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
/*
* Set up the magic number and version in the Obj_Entry. These
* were checked in the crt1.o from the original ElfKit, so we
* set them for backward compatibility.
*/
obj->magic = RTLD_MAGIC;
obj->version = RTLD_VERSION;
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
/* Set the special PLT or GOT entries. */
init_pltgot(obj);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
return 0;
}
/*
* Cleanup procedure. It will be called (by the atexit mechanism) just
* before the process exits.
*/
static void
rtld_exit(void)
{
Make sure internal rtld malloc routines are not called from unlocked contexts as rtld's malloc is not thread safe and is only supposed to be called with exclusive bind lock already held. The originating PR submitted a patch on top of different pre-requisite workaroud for unsafe dlopen calls, and the patch was midief slighlty to apply to stock sources for the purpose of this commit. Running rtld malloc from unlocked contexts is a bug on its own. PR: 126950 Submited by: Oleg Dolgov
2008-09-03 01:05:32 +00:00
int lockstate;
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
Make sure internal rtld malloc routines are not called from unlocked contexts as rtld's malloc is not thread safe and is only supposed to be called with exclusive bind lock already held. The originating PR submitted a patch on top of different pre-requisite workaroud for unsafe dlopen calls, and the patch was midief slighlty to apply to stock sources for the purpose of this commit. Running rtld malloc from unlocked contexts is a bug on its own. PR: 126950 Submited by: Oleg Dolgov
2008-09-03 01:05:32 +00:00
lockstate = wlock_acquire(rtld_bind_lock);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
dbg("rtld_exit()");
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
objlist_call_fini(&list_fini, NULL, &lockstate);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/* No need to remove the items from the list, since we are exiting. */
Use the environment variable LD_LIBMAP_DISABLE to disable libmap.conf(5) functionality.
2003-05-31 14:45:11 +00:00
if (!libmap_disable)
lm_fini();
Make sure internal rtld malloc routines are not called from unlocked contexts as rtld's malloc is not thread safe and is only supposed to be called with exclusive bind lock already held. The originating PR submitted a patch on top of different pre-requisite workaroud for unsafe dlopen calls, and the patch was midief slighlty to apply to stock sources for the purpose of this commit. Running rtld malloc from unlocked contexts is a bug on its own. PR: 126950 Submited by: Oleg Dolgov
2008-09-03 01:05:32 +00:00
wlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
static void *
path_enumerate(const char *path, path_enum_proc callback, void *arg)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
#ifdef COMPAT_32BIT
const char *trans;
#endif
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
if (path == NULL)
return (NULL);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
path += strspn(path, ":;");
while (*path != '\0') {
size_t len;
char *res;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
len = strcspn(path, ":;");
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
#ifdef COMPAT_32BIT
trans = lm_findn(NULL, path, len);
if (trans)
res = callback(trans, strlen(trans), arg);
else
#endif
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
res = callback(path, len, arg);
if (res != NULL)
return (res);
path += len;
path += strspn(path, ":;");
}
return (NULL);
}
struct try_library_args {
const char *name;
size_t namelen;
char *buffer;
size_t buflen;
};
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
static void *
try_library_path(const char *dir, size_t dirlen, void *param)
{
struct try_library_args *arg;
arg = param;
if (*dir == '/' || trust) {
char *pathname;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
if (dirlen + 1 + arg->namelen + 1 > arg->buflen)
return (NULL);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
pathname = arg->buffer;
strncpy(pathname, dir, dirlen);
pathname[dirlen] = '/';
strcpy(pathname + dirlen + 1, arg->name);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
dbg(" Trying \"%s\"", pathname);
if (access(pathname, F_OK) == 0) { /* We found it */
pathname = xmalloc(dirlen + 1 + arg->namelen + 1);
strcpy(pathname, arg->buffer);
return (pathname);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
}
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
return (NULL);
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
static char *
search_library_path(const char *name, const char *path)
{
char *p;
struct try_library_args arg;
if (path == NULL)
return NULL;
arg.name = name;
arg.namelen = strlen(name);
arg.buffer = xmalloc(PATH_MAX);
arg.buflen = PATH_MAX;
p = path_enumerate(path, try_library_path, &arg);
free(arg.buffer);
return (p);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
int
dlclose(void *handle)
{
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
Obj_Entry *root;
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
int lockstate;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
lockstate = wlock_acquire(rtld_bind_lock);
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
root = dlcheck(handle);
if (root == NULL) {
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
wlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return -1;
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
}
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_DLCLOSE_START, handle, NULL, 0, root->dl_refcount,
root->path);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
/* Unreference the object and its dependencies. */
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
root->dl_refcount--;
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
if (root->refcount == 1) {
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
/*
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
* The object will be no longer referenced, so we must unload it.
Make sure internal rtld malloc routines are not called from unlocked contexts as rtld's malloc is not thread safe and is only supposed to be called with exclusive bind lock already held. The originating PR submitted a patch on top of different pre-requisite workaroud for unsafe dlopen calls, and the patch was midief slighlty to apply to stock sources for the purpose of this commit. Running rtld malloc from unlocked contexts is a bug on its own. PR: 126950 Submited by: Oleg Dolgov
2008-09-03 01:05:32 +00:00
* First, call the fini functions.
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
*/
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
objlist_call_fini(&list_fini, root, &lockstate);
unref_dag(root);
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
/* Finish cleaning up the newly-unreferenced objects. */
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
GDB_STATE(RT_DELETE,&root->linkmap);
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
unload_object(root);
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
GDB_STATE(RT_CONSISTENT,NULL);
If dlclose() is called recursively from a _fini() function, the inner dlclose() call may unload the object of the outer call prematurely because objects are unreferenced before _fini() calls. Fix this by unreferencing objects after calling objlist_call_fini() in dlclose(). Therefore objlist_call_fini() now calls the fini function if the reference count of an object is 1. In addition we must restart the list_fini traversal after every _fini() call because another dlclose() call might have modified the reference counts. Add an XXX comment to objlist_call_fini() about possible race with dlopen(). PR: 133246, 149464 Reviewed by: kan, kib
2010-12-16 16:56:44 +00:00
} else
unref_dag(root);
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_DLCLOSE_STOP, handle, NULL, 0, 0, NULL);
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
wlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return 0;
}
- Remove const'ness from dlerror(3) prototype, for consistency with POSIX. Approved by: cognet MFC after: 1 week
2010-03-24 15:59:51 +00:00
char *
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
dlerror(void)
{
char *msg = error_message;
error_message = NULL;
return msg;
}
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
/*
* This function is deprecated and has no effect.
*/
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
void
dllockinit(void *context,
void *(*lock_create)(void *context),
void (*rlock_acquire)(void *lock),
void (*wlock_acquire)(void *lock),
void (*lock_release)(void *lock),
void (*lock_destroy)(void *lock),
void (*context_destroy)(void *context))
{
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
static void *cur_context;
static void (*cur_context_destroy)(void *);
/* Just destroy the context from the previous call, if necessary. */
if (cur_context_destroy != NULL)
cur_context_destroy(cur_context);
cur_context = context;
cur_context_destroy = context_destroy;
When a threads package registers locking methods with dllockinit(), figure out which shared object(s) contain the the locking methods and fully bind those objects as if they had been loaded with LD_BIND_NOW=1. The goal is to keep the locking methods from requiring any lazy binding. Otherwise infinite recursion occurs in _rtld_bind. This fixes the infinite recursion problem in the linuxthreads port.
2000-01-29 01:27:04 +00:00
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
void *
dlopen(const char *name, int mode)
{
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
Obj_Entry **old_obj_tail;
Obj_Entry *obj;
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
Objlist initlist;
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
int result, lockstate, nodelete, lo_flags;
Allow ldd(1) be used on shared libraries in addition to executables.
2002-02-04 10:33:48 +00:00
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_DLOPEN_START, NULL, NULL, 0, mode, name);
Allow ldd(1) be used on shared libraries in addition to executables.
2002-02-04 10:33:48 +00:00
ld_tracing = (mode & RTLD_TRACE) == 0 ? NULL : "1";
if (ld_tracing != NULL)
environ = (char **)*get_program_var_addr("environ");
Implement support for RTLD_NODELETE flag for dlopen() and -z nodelete static linker option. Do it by incrementing reference count on the loaded object and its dependencies. Reviewed by: davidxu, kan
2009-03-30 08:47:28 +00:00
nodelete = mode & RTLD_NODELETE;
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
lo_flags = RTLD_LO_DLOPEN;
if (mode & RTLD_NOLOAD)
lo_flags |= RTLD_LO_NOLOAD;
Allow to load not-openable dso when tracing. This fixes ldd on such dso or dso linked to non-openable object. Remove '\n' at the end of error message. End comments with dot. MFC after: 3 weeks (together with r199829)
2009-11-28 14:29:32 +00:00
if (ld_tracing != NULL)
lo_flags |= RTLD_LO_TRACE;
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_init(&initlist);
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
lockstate = wlock_acquire(rtld_bind_lock);
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
GDB_STATE(RT_ADD,NULL);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
old_obj_tail = obj_tail;
obj = NULL;
Fix a reference counting problem when using dlopen(NULL, ...). PR: bin/12129
1999-06-25 04:50:06 +00:00
if (name == NULL) {
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
obj = obj_main;
Fix a reference counting problem when using dlopen(NULL, ...). PR: bin/12129
1999-06-25 04:50:06 +00:00
obj->refcount++;
} else {
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
obj = load_object(name, obj_main, lo_flags);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
if (obj) {
obj->dl_refcount++;
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
if (mode & RTLD_GLOBAL && objlist_find(&list_global, obj) == NULL)
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_push_tail(&list_global, obj);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
mode &= RTLD_MODEMASK;
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
if (*old_obj_tail != NULL) { /* We loaded something new. */
assert(*old_obj_tail == obj);
Implement rtld part of the support for -z nodlopen (see ld(1)). Reviewed by: kan MFC after: 3 weeks
2009-11-26 13:57:20 +00:00
result = load_needed_objects(obj, RTLD_LO_DLOPEN);
Initialize object dagmembers list before checking version dependencies.
2005-12-22 16:42:38 +00:00
init_dag(obj);
In r214728, if dlopen() is called for the object that has been already loaded as a dependency and marked -z nodlopen, object' DAG is already initialized by load_needed_objects(). Due to this, the init_dag() call from dlopen() does not increment refcount for the object [1]. Change init_dag() to not increment DAG refcount. Require explicit calls to ref_dag() to increment, and assert that ref_dag() and unref_dag() are called for root that has dag initialized. To fix the noted issue, unconditionally call both init_dag() and ref_dag() in dlopen() for the case when the object was already loaded, making it similar to the case of newly loaded object. Noted by: jh [1] Reviewed by: jh, kan MFC after: 6 days
2010-11-04 09:29:00 +00:00
ref_dag(obj);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if (result != -1)
result = rtld_verify_versions(&obj->dagmembers);
Fix a problem with RTLD_TRACE flag to dlopen(3), which sometimes can return even if there was no error occured (when trying to dlopen(3) object that already linked into executable which does dlopen(3) call). This is more proper fix for `ldd /usr/lib/libc.so' problem, because the new behaviour conforms to documentation. Remove workaround from ldd.c (rev.1.32). PR: 35099 Submitted by: Nathan Hawkins <utsl@quic.net> MFC after: 1 week
2002-10-19 10:18:29 +00:00
if (result != -1 && ld_tracing)
goto trace;
Allow ldd(1) be used on shared libraries in addition to executables.
2002-02-04 10:33:48 +00:00
if (result == -1 ||
Initialize object dagmembers list before checking version dependencies.
2005-12-22 16:42:38 +00:00
(relocate_objects(obj, mode == RTLD_NOW, &obj_rtld)) == -1) {
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
obj->dl_refcount--;
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
unref_dag(obj);
if (obj->refcount == 0)
unload_object(obj);
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
obj = NULL;
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
} else {
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
/* Make list of init functions to call. */
initlist_add_objects(obj, &obj->next, &initlist);
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
}
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
} else {
If dlopen() is called for the dso that has been already loaded as a dependency, then the dso never has its DAG initialized. Empty DAG makes ref_dag() call in dlopen() a nop, and the dso refcount is off by one. Initialize the DAG on the first dlopen() call, using a boolean flag to prevent double initialization. From the PR (edited): Assume we have a library liba.so, containing a function a(), and a library libb.so, containing function b(). liba.so needs functionality from libb.so, so liba.so links in libb.so. An application doesn't know about the relation between these libraries, but needs to call a() and b(). It dlopen()s liba.so and obtains a pointer to a(), then it dlopen()s libb.so and obtains a pointer to b(). As soon as the application doesn't need a() anymore, it dlclose()s liba.so. Expected result: the pointer to b() is still valid and can be called Actual result: the pointer to b() has become invalid, even though the application did not dlclose() the handle to libb.so. On calling b(), the application crashes with a segmentation fault. PR: misc/151861 Based on patch by: jh Reviewed by: kan Tested by: Arjan van Leeuwen <freebsd-maintainer opera com> MFC after: 1 week
2010-11-03 09:23:08 +00:00
/*
* Bump the reference counts for objects on this DAG. If
* this is the first dlopen() call for the object that was
* already loaded as a dependency, initialize the dag
* starting at it.
*/
In r214728, if dlopen() is called for the object that has been already loaded as a dependency and marked -z nodlopen, object' DAG is already initialized by load_needed_objects(). Due to this, the init_dag() call from dlopen() does not increment refcount for the object [1]. Change init_dag() to not increment DAG refcount. Require explicit calls to ref_dag() to increment, and assert that ref_dag() and unref_dag() are called for root that has dag initialized. To fix the noted issue, unconditionally call both init_dag() and ref_dag() in dlopen() for the case when the object was already loaded, making it similar to the case of newly loaded object. Noted by: jh [1] Reviewed by: jh, kan MFC after: 6 days
2010-11-04 09:29:00 +00:00
init_dag(obj);
ref_dag(obj);
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
if (ld_tracing)
goto trace;
}
Implement support for RTLD_NODELETE flag for dlopen() and -z nodelete static linker option. Do it by incrementing reference count on the loaded object and its dependencies. Reviewed by: davidxu, kan
2009-03-30 08:47:28 +00:00
if (obj != NULL && (nodelete || obj->z_nodelete) && !obj->ref_nodel) {
dbg("obj %s nodelete", obj->path);
ref_dag(obj);
obj->z_nodelete = obj->ref_nodel = true;
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_DLOPEN_STOP, obj, NULL, 0, obj ? obj->dl_refcount : 0,
name);
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
GDB_STATE(RT_CONSISTENT,obj ? &obj->linkmap : NULL);
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
Make sure internal rtld malloc routines are not called from unlocked contexts as rtld's malloc is not thread safe and is only supposed to be called with exclusive bind lock already held. The originating PR submitted a patch on top of different pre-requisite workaroud for unsafe dlopen calls, and the patch was midief slighlty to apply to stock sources for the purpose of this commit. Running rtld malloc from unlocked contexts is a bug on its own. PR: 126950 Submited by: Oleg Dolgov
2008-09-03 01:05:32 +00:00
/* Call the init functions. */
objlist_call_init(&initlist, &lockstate);
Revamp the code that calls shared libraries' init and fini functions. Formerly the init functions were called in the opposite of the order in which libraries were loaded, and libraries were loaded according to a breadth-first traversal of the dependency graph. That ordering came from SVR4.0, and it was easy to implement but not always sensible. Now we do a depth-first walk over the dependency graph and call the init functions in an order such that each shared object's needed objects are initialized before the shared object itself. At the same time we build a list of finalization (fini) functions in the opposite order, to guarantee correct C++ destructor ordering whenever possible. (It may not be possible if dlopen and dlclose are used in strange ways, but we come as close as one can come.) The need for this renovation has become apparent as more programs have started using multithreading. The multithreaded C library libc_r requires initialization, whereas the standard libc does not. Since virtually every other object depends on the C library, it is important that it get initialized first.
2000-07-26 04:24:40 +00:00
objlist_clear(&initlist);
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
wlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return obj;
Fix a problem with RTLD_TRACE flag to dlopen(3), which sometimes can return even if there was no error occured (when trying to dlopen(3) object that already linked into executable which does dlopen(3) call). This is more proper fix for `ldd /usr/lib/libc.so' problem, because the new behaviour conforms to documentation. Remove workaround from ldd.c (rev.1.32). PR: 35099 Submitted by: Nathan Hawkins <utsl@quic.net> MFC after: 1 week
2002-10-19 10:18:29 +00:00
trace:
trace_loaded_objects(obj);
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
wlock_release(rtld_bind_lock, lockstate);
Fix a problem with RTLD_TRACE flag to dlopen(3), which sometimes can return even if there was no error occured (when trying to dlopen(3) object that already linked into executable which does dlopen(3) call). This is more proper fix for `ldd /usr/lib/libc.so' problem, because the new behaviour conforms to documentation. Remove workaround from ldd.c (rev.1.32). PR: 35099 Submitted by: Nathan Hawkins <utsl@quic.net> MFC after: 1 week
2002-10-19 10:18:29 +00:00
exit(0);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
static void *
do_dlsym(void *handle, const char *name, void *retaddr, const Ver_Entry *ve,
int flags)
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
{
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
DoneList donelist;
const Obj_Entry *obj, *defobj;
Allow strong symbols to override weak ones for lookups done through dlsym with RTLD_NEXT/RTLD_SELF handles. Allow symbols from ld-elf.so to be located this way too. Based on report and original patch from sobomax@.
2008-10-10 00:16:32 +00:00
const Elf_Sym *def, *symp;
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
unsigned long hash;
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
int lockstate;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
hash = elf_hash(name);
Handle dlsym(NULL, ...) properly, by searching in the caller's shared object. Note, this searches _only_ that object, and not its needed objects, in accordance with the documentation. Also fix dlopen(NULL, ...) so that the executable's needed objects are searched as well as the executable itself.
1998-09-02 01:09:34 +00:00
def = NULL;
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
defobj = NULL;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
flags |= SYMLOOK_IN_PLT;
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
lockstate = rlock_acquire(rtld_bind_lock);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
if (handle == NULL || handle == RTLD_NEXT ||
handle == RTLD_DEFAULT || handle == RTLD_SELF) {
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
if ((obj = obj_from_addr(retaddr)) == NULL) {
_rtld_error("Cannot determine caller's shared object");
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return NULL;
}
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
if (handle == NULL) { /* Just the caller's shared object. */
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
def = symlook_obj(name, hash, obj, ve, flags);
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
defobj = obj;
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
} else if (handle == RTLD_NEXT || /* Objects after caller's */
handle == RTLD_SELF) { /* ... caller included */
if (handle == RTLD_NEXT)
obj = obj->next;
for (; obj != NULL; obj = obj->next) {
Allow strong symbols to override weak ones for lookups done through dlsym with RTLD_NEXT/RTLD_SELF handles. Allow symbols from ld-elf.so to be located this way too. Based on report and original patch from sobomax@.
2008-10-10 00:16:32 +00:00
if ((symp = symlook_obj(name, hash, obj, ve, flags)) != NULL) {
if (def == NULL || ELF_ST_BIND(symp->st_info) != STB_WEAK) {
def = symp;
defobj = obj;
if (ELF_ST_BIND(def->st_info) != STB_WEAK)
break;
}
}
}
/*
* Search the dynamic linker itself, and possibly resolve the
* symbol from there. This is how the application links to
Remove exports table. Export control by the version script is enough. Reviewed by: kan MFC after: 3 weeks
2010-08-24 13:01:14 +00:00
* dynamic linker services such as dlopen.
Allow strong symbols to override weak ones for lookups done through dlsym with RTLD_NEXT/RTLD_SELF handles. Allow symbols from ld-elf.so to be located this way too. Based on report and original patch from sobomax@.
2008-10-10 00:16:32 +00:00
*/
if (def == NULL || ELF_ST_BIND(def->st_info) == STB_WEAK) {
symp = symlook_obj(name, hash, &obj_rtld, ve, flags);
Remove exports table. Export control by the version script is enough. Reviewed by: kan MFC after: 3 weeks
2010-08-24 13:01:14 +00:00
if (symp != NULL) {
Allow strong symbols to override weak ones for lookups done through dlsym with RTLD_NEXT/RTLD_SELF handles. Allow symbols from ld-elf.so to be located this way too. Based on report and original patch from sobomax@.
2008-10-10 00:16:32 +00:00
def = symp;
defobj = &obj_rtld;
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
}
}
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
} else {
assert(handle == RTLD_DEFAULT);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
def = symlook_default(name, hash, obj, &defobj, ve, flags);
Handle dlsym(NULL, ...) properly, by searching in the caller's shared object. Note, this searches _only_ that object, and not its needed objects, in accordance with the documentation. Also fix dlopen(NULL, ...) so that the executable's needed objects are searched as well as the executable itself.
1998-09-02 01:09:34 +00:00
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
} else {
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
if ((obj = dlcheck(handle)) == NULL) {
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return NULL;
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
}
Handle dlsym(NULL, ...) properly, by searching in the caller's shared object. Note, this searches _only_ that object, and not its needed objects, in accordance with the documentation. Also fix dlopen(NULL, ...) so that the executable's needed objects are searched as well as the executable itself.
1998-09-02 01:09:34 +00:00
When looking up the symbol by dlsym, look it not only in the object given as dso handle, but also in the implicit dependencies of that dso. Also, const-ify the read-only parameter objlist of symlook_list. Reported by: "Simon 'corecode' Schubert" <corecode at fs ei tum de> Approved by: kan (mentor) X-MFC-After: 6.2
2006-09-08 14:59:54 +00:00
donelist_init(&donelist);
Handle dlsym(NULL, ...) properly, by searching in the caller's shared object. Note, this searches _only_ that object, and not its needed objects, in accordance with the documentation. Also fix dlopen(NULL, ...) so that the executable's needed objects are searched as well as the executable itself.
1998-09-02 01:09:34 +00:00
if (obj->mainprog) {
/* Search main program and all libraries loaded by it. */
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
def = symlook_list(name, hash, &list_main, &defobj, ve, flags,
&donelist);
FreeBSD returns main object handle from dlopen(NULL, ...) calls. dlsym seaches using this handle are expected to look for symbol definitions in all objects loaded at the program start time along with all objects currently in RTLD_GLOBAL scope. Discussed with: kib Reported by: Maho NAKATA MFC after: 2 weeks
2009-06-16 16:38:54 +00:00
/*
Allow order of initialization of loaded shared objects to be altered through their .init code. This might happen if init vector calls dlopen on its own and that dlopen causes some not yet initialized object to be initialized earlier as part of that dlopened DAG. Do not reset module reference counts to zero on final fini vector run when process is exiting. Just add an additional parameter to force fini vector invocation regardless of current reference count value if object was not destructed yet. This allows dlclose called from fini vector to proceed normally instead of failing with handle validation error. Reviewed by: kib Reported by: venki kaps
2009-06-20 14:16:41 +00:00
* We do not distinguish between 'main' object and global scope.
FreeBSD returns main object handle from dlopen(NULL, ...) calls. dlsym seaches using this handle are expected to look for symbol definitions in all objects loaded at the program start time along with all objects currently in RTLD_GLOBAL scope. Discussed with: kib Reported by: Maho NAKATA MFC after: 2 weeks
2009-06-16 16:38:54 +00:00
* If symbol is not defined by objects loaded at startup, continue
* search among dynamically loaded objects with RTLD_GLOBAL
* scope.
*/
if (def == NULL)
def = symlook_list(name, hash, &list_global, &defobj, ve,
flags, &donelist);
Handle dlsym(NULL, ...) properly, by searching in the caller's shared object. Note, this searches _only_ that object, and not its needed objects, in accordance with the documentation. Also fix dlopen(NULL, ...) so that the executable's needed objects are searched as well as the executable itself.
1998-09-02 01:09:34 +00:00
} else {
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
Needed_Entry fake;
When looking up the symbol by dlsym, look it not only in the object given as dso handle, but also in the implicit dependencies of that dso. Also, const-ify the read-only parameter objlist of symlook_list. Reported by: "Simon 'corecode' Schubert" <corecode at fs ei tum de> Approved by: kan (mentor) X-MFC-After: 6.2
2006-09-08 14:59:54 +00:00
/* Search the whole DAG rooted at the given object. */
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
fake.next = NULL;
fake.obj = (Obj_Entry *)obj;
fake.name = 0;
def = symlook_needed(name, hash, &fake, &defobj, ve, flags,
&donelist);
Handle dlsym(NULL, ...) properly, by searching in the caller's shared object. Note, this searches _only_ that object, and not its needed objects, in accordance with the documentation. Also fix dlopen(NULL, ...) so that the executable's needed objects are searched as well as the executable itself.
1998-09-02 01:09:34 +00:00
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
}
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
if (def != NULL) {
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
Add ia64 support. Various adjustments were made to existing targets to cope with a few interface changes required by the ia64. In particular, function pointers on ia64 need special treatment in rtld.
2001-10-15 18:48:42 +00:00
/*
* The value required by the caller is derived from the value
* of the symbol. For the ia64 architecture, we need to
* construct a function descriptor which the caller can use to
* call the function with the right 'gp' value. For other
* architectures and for non-functions, the value is simply
* the relocated value of the symbol.
*/
if (ELF_ST_TYPE(def->st_info) == STT_FUNC)
return make_function_pointer(def, defobj);
else
return defobj->relocbase + def->st_value;
Add a new function dllockinit() for registering thread locking functions to be used by the dynamic linker. This can be called by threads packages at start-up time. I will add the call to libc_r soon. Also add a default locking method that is used up until dllockinit() is called. The default method works by blocking SIGVTALRM, SIGPROF, and SIGALRM in critical sections. It is based on the observation that most user-space threads packages implement thread preemption with one of these signals (usually SIGVTALRM). The dynamic linker has never been reentrant, but it became less reentrant in revision 1.34 of "src/libexec/rtld-elf/rtld.c". Starting with that revision, multiple threads each doing lazy binding could interfere with each other. The usual symptom was that a symbol was falsely reported as undefined at start-up time. It was rare but not unseen. This commit fixes it.
1999-12-27 04:44:04 +00:00
}
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
_rtld_error("Undefined symbol \"%s\"", name);
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
Import the ELF dynamic linker. This is the ElfKit version with quite a few enhancements and bug fixes. There are still some known deficiencies, but it should be adequate to get us started with ELF. Submitted by: John Polstra <jdp@polstra.com>
1998-03-07 19:24:35 +00:00
return NULL;
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
void *
dlsym(void *handle, const char *name)
{
return do_dlsym(handle, name, __builtin_return_address(0), NULL,
SYMLOOK_DLSYM);
}
Allow the NULL, RTLD_SELF and RTLD_NEXT handles to work with dlfunc(3). dlfunc() called dlsym() to do the work, and dlsym() determines the dso that originating the call by the return address. Due to this, dlfunc() operated as if the caller is always the libc. To fix this, move the dlfunc() to rtld, where it can call the internal implementation of dlsym, and still correctly fetch return address. Provide usual weak stub for the symbol from libc for static binaries. dlfunc is put to FBSD_1.0 symver namespace in the ld.so export to override dlfunc@FBSD_1.0 weak symbol, exported by libc. Reported, analyzed and tested by: Tijl Coosemans <tijl ulyssis org> PR: standards/133339 Reviewed by: kan
2009-04-03 19:17:23 +00:00
dlfunc_t
dlfunc(void *handle, const char *name)
{
union {
void *d;
dlfunc_t f;
} rv;
rv.d = do_dlsym(handle, name, __builtin_return_address(0), NULL,
SYMLOOK_DLSYM);
return (rv.f);
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
void *
dlvsym(void *handle, const char *name, const char *version)
{
Ver_Entry ventry;
ventry.name = version;
ventry.file = NULL;
ventry.hash = elf_hash(version);
ventry.flags= 0;
return do_dlsym(handle, name, __builtin_return_address(0), &ventry,
SYMLOOK_DLSYM);
}
Introduce implementation-private rtld interface _rtld_addr_phdr, which fills struct dl_phdr_info for the shared object that contains the specified address, if any. Idea and reviewed by: kan MFC after: 3 weeks
2010-08-23 15:27:03 +00:00
int
_rtld_addr_phdr(const void *addr, struct dl_phdr_info *phdr_info)
{
const Obj_Entry *obj;
int lockstate;
lockstate = rlock_acquire(rtld_bind_lock);
obj = obj_from_addr(addr);
if (obj == NULL) {
_rtld_error("No shared object contains address");
rlock_release(rtld_bind_lock, lockstate);
return (0);
}
rtld_fill_dl_phdr_info(obj, phdr_info);
rlock_release(rtld_bind_lock, lockstate);
return (1);
}
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
int
dladdr(const void *addr, Dl_info *info)
{
const Obj_Entry *obj;
const Elf_Sym *def;
void *symbol_addr;
unsigned long symoffset;
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
int lockstate;
lockstate = rlock_acquire(rtld_bind_lock);
- Commit the correct dladdr() implementation. Reviewed by: jdp@FreeBSD.org <This is the version he reviewed!>
1999-03-24 23:47:29 +00:00
obj = obj_from_addr(addr);
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
if (obj == NULL) {
- Commit the correct dladdr() implementation. Reviewed by: jdp@FreeBSD.org <This is the version he reviewed!>
1999-03-24 23:47:29 +00:00
_rtld_error("No shared object contains address");
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
return 0;
}
info->dli_fname = obj->path;
- Commit the correct dladdr() implementation. Reviewed by: jdp@FreeBSD.org <This is the version he reviewed!>
1999-03-24 23:47:29 +00:00
info->dli_fbase = obj->mapbase;
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
info->dli_saddr = (void *)0;
info->dli_sname = NULL;
/*
Fix a couple of typos in comments.
1999-04-07 02:48:43 +00:00
* Walk the symbol list looking for the symbol whose address is
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
* closest to the address sent in.
*/
for (symoffset = 0; symoffset < obj->nchains; symoffset++) {
def = obj->symtab + symoffset;
- Commit the correct dladdr() implementation. Reviewed by: jdp@FreeBSD.org <This is the version he reviewed!>
1999-03-24 23:47:29 +00:00
/*
* For skip the symbol if st_shndx is either SHN_UNDEF or
* SHN_COMMON.
*/
if (def->st_shndx == SHN_UNDEF || def->st_shndx == SHN_COMMON)
continue;
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
/*
Fix a couple of typos in comments.
1999-04-07 02:48:43 +00:00
* If the symbol is greater than the specified address, or if it
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
* is further away from addr than the current nearest symbol,
* then reject it.
*/
- Commit the correct dladdr() implementation. Reviewed by: jdp@FreeBSD.org <This is the version he reviewed!>
1999-03-24 23:47:29 +00:00
symbol_addr = obj->relocbase + def->st_value;
if (symbol_addr > addr || symbol_addr < info->dli_saddr)
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
continue;
/* Update our idea of the nearest symbol. */
info->dli_sname = obj->strtab + def->st_name;
info->dli_saddr = symbol_addr;
/* Exact match? */
if (info->dli_saddr == addr)
break;
}
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
- Added dladdr(3) support. Reviewed by: jdp@FreeBSD.org
1999-03-24 23:37:35 +00:00
return 1;
}
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
int
dlinfo(void *handle, int request, void *p)
{
const Obj_Entry *obj;
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
int error, lockstate;
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
lockstate = rlock_acquire(rtld_bind_lock);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
if (handle == NULL || handle == RTLD_SELF) {
void *retaddr;
retaddr = __builtin_return_address(0); /* __GNUC__ only */
if ((obj = obj_from_addr(retaddr)) == NULL)
_rtld_error("Cannot determine caller's shared object");
} else
obj = dlcheck(handle);
if (obj == NULL) {
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
return (-1);
}
error = 0;
switch (request) {
case RTLD_DI_LINKMAP:
*((struct link_map const **)p) = &obj->linkmap;
break;
case RTLD_DI_ORIGIN:
error = rtld_dirname(obj->path, p);
break;
case RTLD_DI_SERINFOSIZE:
case RTLD_DI_SERINFO:
error = do_search_info(obj, request, (struct dl_serinfo *)p);
break;
default:
_rtld_error("Invalid request %d passed to dlinfo()", request);
error = -1;
}
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
rlock_release(rtld_bind_lock, lockstate);
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
return (error);
}
Introduce implementation-private rtld interface _rtld_addr_phdr, which fills struct dl_phdr_info for the shared object that contains the specified address, if any. Idea and reviewed by: kan MFC after: 3 weeks
2010-08-23 15:27:03 +00:00
static void
rtld_fill_dl_phdr_info(const Obj_Entry *obj, struct dl_phdr_info *phdr_info)
{
phdr_info->dlpi_addr = (Elf_Addr)obj->relocbase;
phdr_info->dlpi_name = STAILQ_FIRST(&obj->names) ?
STAILQ_FIRST(&obj->names)->name : obj->path;
phdr_info->dlpi_phdr = obj->phdr;
phdr_info->dlpi_phnum = obj->phsize / sizeof(obj->phdr[0]);
phdr_info->dlpi_tls_modid = obj->tlsindex;
phdr_info->dlpi_tls_data = obj->tlsinit;
phdr_info->dlpi_adds = obj_loads;
phdr_info->dlpi_subs = obj_loads - obj_count;
}
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
int
dl_iterate_phdr(__dl_iterate_hdr_callback callback, void *param)
{
struct dl_phdr_info phdr_info;
const Obj_Entry *obj;
Fix the problem with the C++ exception handling for the multithreaded programs. From the PR description: The gcc runtime's _Unwind_Find_FDE function, invoked during exception handling's stack unwinding, is not safe to execute from within multiple threads. FreeBSD' s dl_iterate_phdr() however permits multiple threads to pass through it though. The result is surprisingly reliable infinite looping of one or more threads if they just happen to be unwinding at the same time. Introduce the new lock that is write locked around the dl_iterate_pdr, thus providing required exclusion for the stack unwinders. PR: threads/123062 Submitted by: Andy Newman <an at atrn org> Reviewed by: kan MFC after: 2 weeks
2008-05-06 09:27:41 +00:00
int error, bind_lockstate, phdr_lockstate;
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
Fix the problem with the C++ exception handling for the multithreaded programs. From the PR description: The gcc runtime's _Unwind_Find_FDE function, invoked during exception handling's stack unwinding, is not safe to execute from within multiple threads. FreeBSD' s dl_iterate_phdr() however permits multiple threads to pass through it though. The result is surprisingly reliable infinite looping of one or more threads if they just happen to be unwinding at the same time. Introduce the new lock that is write locked around the dl_iterate_pdr, thus providing required exclusion for the stack unwinders. PR: threads/123062 Submitted by: Andy Newman <an at atrn org> Reviewed by: kan MFC after: 2 weeks
2008-05-06 09:27:41 +00:00
phdr_lockstate = wlock_acquire(rtld_phdr_lock);
bind_lockstate = rlock_acquire(rtld_bind_lock);
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
error = 0;
for (obj = obj_list; obj != NULL; obj = obj->next) {
Introduce implementation-private rtld interface _rtld_addr_phdr, which fills struct dl_phdr_info for the shared object that contains the specified address, if any. Idea and reviewed by: kan MFC after: 3 weeks
2010-08-23 15:27:03 +00:00
rtld_fill_dl_phdr_info(obj, &phdr_info);
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
if ((error = callback(&phdr_info, sizeof phdr_info, param)) != 0)
break;
}
Fix the problem with the C++ exception handling for the multithreaded programs. From the PR description: The gcc runtime's _Unwind_Find_FDE function, invoked during exception handling's stack unwinding, is not safe to execute from within multiple threads. FreeBSD' s dl_iterate_phdr() however permits multiple threads to pass through it though. The result is surprisingly reliable infinite looping of one or more threads if they just happen to be unwinding at the same time. Introduce the new lock that is write locked around the dl_iterate_pdr, thus providing required exclusion for the stack unwinders. PR: threads/123062 Submitted by: Andy Newman <an at atrn org> Reviewed by: kan MFC after: 2 weeks
2008-05-06 09:27:41 +00:00
rlock_release(rtld_bind_lock, bind_lockstate);
wlock_release(rtld_phdr_lock, phdr_lockstate);
Implement dl_iterate_phdr function. Convert boolean flags in internal Obj_Entry structure into bitfields. Properly check for loaded segment alignment in map_object.
2007-04-03 18:31:20 +00:00
return (error);
}
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
struct fill_search_info_args {
int request;
unsigned int flags;
Dl_serinfo *serinfo;
Dl_serpath *serpath;
char *strspace;
};
static void *
fill_search_info(const char *dir, size_t dirlen, void *param)
{
struct fill_search_info_args *arg;
arg = param;
if (arg->request == RTLD_DI_SERINFOSIZE) {
arg->serinfo->dls_cnt ++;
Fix a bug in dlinfo(RTLD_DI_SERINFOSIZE) requests. For each search path we included the length of the path in the returned size but not the length of the associated Dl_serpath structure. Without this fix, programs attempting to allocate a structure to hold the search path information would allocate too small of a buffer and rtld would overrun the buffer while filling it via a subsequent RTLD_DI_SERINFO request. Submitted by: "William K. Josephson" wkj at morphisms dot net Reviewed by: jdp MFC after: 2 weeks
2005-11-11 19:57:41 +00:00
arg->serinfo->dls_size += sizeof(Dl_serpath) + dirlen + 1;
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
} else {
struct dl_serpath *s_entry;
s_entry = arg->serpath;
s_entry->dls_name = arg->strspace;
s_entry->dls_flags = arg->flags;
strncpy(arg->strspace, dir, dirlen);
arg->strspace[dirlen] = '\0';
arg->strspace += dirlen + 1;
arg->serpath++;
}
return (NULL);
}
static int
do_search_info(const Obj_Entry *obj, int request, struct dl_serinfo *info)
{
struct dl_serinfo _info;
struct fill_search_info_args args;
args.request = RTLD_DI_SERINFOSIZE;
args.serinfo = &_info;
_info.dls_size = __offsetof(struct dl_serinfo, dls_serpath);
_info.dls_cnt = 0;
path_enumerate(ld_library_path, fill_search_info, &args);
path_enumerate(obj->rpath, fill_search_info, &args);
path_enumerate(gethints(), fill_search_info, &args);
path_enumerate(STANDARD_LIBRARY_PATH, fill_search_info, &args);
if (request == RTLD_DI_SERINFOSIZE) {
info->dls_size = _info.dls_size;
info->dls_cnt = _info.dls_cnt;
return (0);
}
if (info->dls_cnt != _info.dls_cnt || info->dls_size != _info.dls_size) {
_rtld_error("Uninitialized Dl_serinfo struct passed to dlinfo()");
return (-1);
}
args.request = RTLD_DI_SERINFO;
args.serinfo = info;
args.serpath = &info->dls_serpath[0];
args.strspace = (char *)&info->dls_serpath[_info.dls_cnt];
args.flags = LA_SER_LIBPATH;
if (path_enumerate(ld_library_path, fill_search_info, &args) != NULL)
return (-1);
args.flags = LA_SER_RUNPATH;
if (path_enumerate(obj->rpath, fill_search_info, &args) != NULL)
return (-1);
args.flags = LA_SER_CONFIG;
if (path_enumerate(gethints(), fill_search_info, &args) != NULL)
return (-1);
args.flags = LA_SER_DEFAULT;
if (path_enumerate(STANDARD_LIBRARY_PATH, fill_search_info, &args) != NULL)
return (-1);
return (0);
}
static int
rtld_dirname(const char *path, char *bname)
{
const char *endp;
/* Empty or NULL string gets treated as "." */
if (path == NULL || *path == '\0') {
bname[0] = '.';
Fix a typo in rtld_dirname.
2003-02-13 22:47:41 +00:00
bname[1] = '\0';
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
return (0);
}
/* Strip trailing slashes */
endp = path + strlen(path) - 1;
while (endp > path && *endp == '/')
endp--;
/* Find the start of the dir */
while (endp > path && *endp != '/')
endp--;
/* Either the dir is "/" or there are no slashes */
if (endp == path) {
bname[0] = *endp == '/' ? '/' : '.';
bname[1] = '\0';
return (0);
} else {
do {
endp--;
} while (endp > path && *endp == '/');
}
if (endp - path + 2 > PATH_MAX)
{
_rtld_error("Filename is too long: %s", path);
return(-1);
}
strncpy(bname, path, endp - path + 1);
bname[endp - path + 1] = '\0';
return (0);
}
Implement the dynamic string token substitution in the rpath and soneeded pathes. The $ORIGIN, $OSNAME, $OSREL and $PLATFORM tokens are supported. Enabling the substitution requires DF_ORIGIN flag in DT_FLAGS or DF_1_ORIGIN if DF_FLAGS_1, that may be set with -z origin gnu ld flag. Translation is unconditionally disabled for setuid/setgid processes. The $ORIGIN translation relies on the AT_EXECPATH auxinfo supplied by kernel. Requested by: maho Tested by: maho, pho Reviewed by: kan
2009-03-18 13:40:37 +00:00
static int
rtld_dirname_abs(const char *path, char *base)
{
char base_rel[PATH_MAX];
if (rtld_dirname(path, base) == -1)
return (-1);
if (base[0] == '/')
return (0);
if (getcwd(base_rel, sizeof(base_rel)) == NULL ||
strlcat(base_rel, "/", sizeof(base_rel)) >= sizeof(base_rel) ||
strlcat(base_rel, base, sizeof(base_rel)) >= sizeof(base_rel))
return (-1);
strcpy(base, base_rel);
return (0);
}
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
static void
linkmap_add(Obj_Entry *obj)
{
struct link_map *l = &obj->linkmap;
struct link_map *prev;
obj->linkmap.l_name = obj->path;
obj->linkmap.l_addr = obj->mapbase;
obj->linkmap.l_ld = obj->dynamic;
#ifdef __mips__
/* GDB needs load offset on MIPS to use the symbols */
obj->linkmap.l_offs = obj->relocbase;
#endif
if (r_debug.r_map == NULL) {
r_debug.r_map = l;
return;
}
Implement dlinfo() function. Introdice RTLD_SELF special handle and properly process it within dlsym() and dlinfo() functions. The intention is to improve our compatibility with Solaris and to make a Java port easier. Partially submitted by: phantom
2003-02-13 17:47:44 +00:00
Fix a bug that showed up when debugging dynamically linked programs. References from GDB to "printf" and various other functions would find the versions in the dynamic linker itself, rather than the versions in the program's libc. This fix moves the GDB link map entry for the dynamic linker to the end of the search list, where its symbols will be found only if they are not found anywhere else. It was suggested by Doug Rabson, though I implemented it a little differently. I personally would prefer to leave the dynamic linker's entry out of the GDB search list altogether. But Doug argues that it is handy there for such things as setting breakpoints on dlopen(). So it stays for now, at least. Note, if we ever integrate the dynamic linker with libc (which has several important benefits to recommend it), this whole problem goes away.
1998-09-16 02:54:08 +00:00
/*
* Scan to the end of the list, but not past the entry for the
* dynamic linker, which we want to keep at the very end.
*/
for (prev = r_debug.r_map;
prev->l_next != NULL && prev->l_next != &obj_rtld.linkmap;
prev = prev->l_next)
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
;
Fix a bug that showed up when debugging dynamically linked programs. References from GDB to "printf" and various other functions would find the versions in the dynamic linker itself, rather than the versions in the program's libc. This fix moves the GDB link map entry for the dynamic linker to the end of the search list, where its symbols will be found only if they are not found anywhere else. It was suggested by Doug Rabson, though I implemented it a little differently. I personally would prefer to leave the dynamic linker's entry out of the GDB search list altogether. But Doug argues that it is handy there for such things as setting breakpoints on dlopen(). So it stays for now, at least. Note, if we ever integrate the dynamic linker with libc (which has several important benefits to recommend it), this whole problem goes away.
1998-09-16 02:54:08 +00:00
/* Link in the new entry. */
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
l->l_prev = prev;
Fix a bug that showed up when debugging dynamically linked programs. References from GDB to "printf" and various other functions would find the versions in the dynamic linker itself, rather than the versions in the program's libc. This fix moves the GDB link map entry for the dynamic linker to the end of the search list, where its symbols will be found only if they are not found anywhere else. It was suggested by Doug Rabson, though I implemented it a little differently. I personally would prefer to leave the dynamic linker's entry out of the GDB search list altogether. But Doug argues that it is handy there for such things as setting breakpoints on dlopen(). So it stays for now, at least. Note, if we ever integrate the dynamic linker with libc (which has several important benefits to recommend it), this whole problem goes away.
1998-09-16 02:54:08 +00:00
l->l_next = prev->l_next;
if (l->l_next != NULL)
l->l_next->l_prev = l;
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
prev->l_next = l;
}
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
static void
linkmap_delete(Obj_Entry *obj)
Add GDB support. The method and some of the code came from NetBSD's elf runtime linker.
1998-04-30 07:48:02 +00:00
{
struct link_map *l = &obj->linkmap;
if (l->l_prev == NULL) {
if ((r_debug.r_map = l->l_next) != NULL)
l->l_next->l_prev = NULL;
return;
}
if ((l->l_prev->l_next = l->l_next) != NULL)
l->l_next->l_prev = l->l_prev;
}
Add support for ldd.
1998-05-01 08:39:27 +00:00
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
/*
* Function for the debugger to set a breakpoint on to gain control.
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
*
* The two parameters allow the debugger to easily find and determine
* what the runtime loader is doing and to whom it is doing it.
*
* When the loadhook trap is hit (r_debug_state, set at program
* initialization), the arguments can be found on the stack:
*
* +8 struct link_map *m
* +4 struct r_debug *rd
* +0 RetAddr
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
*/
void
Pass two pointer parameters to the r_debug_state() hook function, thus allowing a debugger or other trace tool to easily grab the addresses of the needed structures off the stack. This change is transparent to gdb, which locates the link_map list and transfers it to debugger memory for comparison purposes. A sample program will be committed showing how this can be used. Reviewed by: John Polstra <jdp@FreeBSD.org>
2000-08-26 05:13:29 +00:00
r_debug_state(struct r_debug* rd, struct link_map *m)
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
{
}
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
/*
Allow ldd(1) be used on shared libraries in addition to executables.
2002-02-04 10:33:48 +00:00
* Get address of the pointer variable in the main program.
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
*/
Allow ldd(1) be used on shared libraries in addition to executables.
2002-02-04 10:33:48 +00:00
static const void **
get_program_var_addr(const char *name)
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
{
const Obj_Entry *obj;
unsigned long hash;
hash = elf_hash(name);
for (obj = obj_main; obj != NULL; obj = obj->next) {
const Elf_Sym *def;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if ((def = symlook_obj(name, hash, obj, NULL, 0)) != NULL) {
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
const void **addr;
addr = (const void **)(obj->relocbase + def->st_value);
Allow ldd(1) be used on shared libraries in addition to executables.
2002-02-04 10:33:48 +00:00
return addr;
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
}
}
Allow ldd(1) be used on shared libraries in addition to executables.
2002-02-04 10:33:48 +00:00
return NULL;
}
/*
* Set a pointer variable in the main program to the given value. This
* is used to set key variables such as "environ" before any of the
* init functions are called.
*/
static void
set_program_var(const char *name, const void *value)
{
const void **addr;
if ((addr = get_program_var_addr(name)) != NULL) {
dbg("\"%s\": *%p <-- %p", name, addr, value);
*addr = value;
}
After relocating the main program, but before calling any of the _init() functions, initialize the global variables "__progname" and "environ". This makes it possible for the _init() functions to call things like getenv() and err().
1999-04-21 04:06:57 +00:00
}
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
/*
* Given a symbol name in a referencing object, find the corresponding
* definition of the symbol. Returns a pointer to the symbol, or NULL if
* no definition was found. Returns a pointer to the Obj_Entry of the
* defining object via the reference parameter DEFOBJ_OUT.
*/
static const Elf_Sym *
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
symlook_default(const char *name, unsigned long hash, const Obj_Entry *refobj,
const Obj_Entry **defobj_out, const Ver_Entry *ventry, int flags)
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
{
DoneList donelist;
const Elf_Sym *def;
const Elf_Sym *symp;
const Obj_Entry *obj;
const Obj_Entry *defobj;
const Objlist_Entry *elm;
def = NULL;
defobj = NULL;
donelist_init(&donelist);
/* Look first in the referencing object if linked symbolically. */
if (refobj->symbolic && !donelist_check(&donelist, refobj)) {
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
symp = symlook_obj(name, hash, refobj, ventry, flags);
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
if (symp != NULL) {
def = symp;
defobj = refobj;
}
}
/* Search all objects loaded at program start up. */
if (def == NULL || ELF_ST_BIND(def->st_info) == STB_WEAK) {
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
symp = symlook_list(name, hash, &list_main, &obj, ventry, flags,
&donelist);
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
if (symp != NULL &&
(def == NULL || ELF_ST_BIND(symp->st_info) != STB_WEAK)) {
def = symp;
defobj = obj;
}
}
Change the symbol lookup order to search RTLD_GLOBAL objects before referencing object's DAG. This makes it possible for C++ exceptions to work across shared libraries and brings us closer to the search order used by Solaris/Linux. Reviewed by: jdp Approved by: obrien MFC after: 1 month
2002-10-19 23:03:35 +00:00
/* Search all DAGs whose roots are RTLD_GLOBAL objects. */
STAILQ_FOREACH(elm, &list_global, link) {
if (def != NULL && ELF_ST_BIND(def->st_info) != STB_WEAK)
break;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
symp = symlook_list(name, hash, &elm->obj->dagmembers, &obj, ventry,
flags, &donelist);
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
if (symp != NULL &&
(def == NULL || ELF_ST_BIND(symp->st_info) != STB_WEAK)) {
def = symp;
defobj = obj;
}
}
Change the symbol lookup order to search RTLD_GLOBAL objects before referencing object's DAG. This makes it possible for C++ exceptions to work across shared libraries and brings us closer to the search order used by Solaris/Linux. Reviewed by: jdp Approved by: obrien MFC after: 1 month
2002-10-19 23:03:35 +00:00
/* Search all dlopened DAGs containing the referencing object. */
STAILQ_FOREACH(elm, &refobj->dldags, link) {
if (def != NULL && ELF_ST_BIND(def->st_info) != STB_WEAK)
break;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
symp = symlook_list(name, hash, &elm->obj->dagmembers, &obj, ventry,
flags, &donelist);
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
if (symp != NULL &&
(def == NULL || ELF_ST_BIND(symp->st_info) != STB_WEAK)) {
def = symp;
defobj = obj;
}
}
/*
* Search the dynamic linker itself, and possibly resolve the
* symbol from there. This is how the application links to
Remove exports table. Export control by the version script is enough. Reviewed by: kan MFC after: 3 weeks
2010-08-24 13:01:14 +00:00
* dynamic linker services such as dlopen.
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
*/
if (def == NULL || ELF_ST_BIND(def->st_info) == STB_WEAK) {
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
symp = symlook_obj(name, hash, &obj_rtld, ventry, flags);
Remove exports table. Export control by the version script is enough. Reviewed by: kan MFC after: 3 weeks
2010-08-24 13:01:14 +00:00
if (symp != NULL) {
Add support for dlsym(RTLD_DEFAULT, ...).
2000-09-19 04:27:16 +00:00
def = symp;
defobj = &obj_rtld;
}
}
if (def != NULL)
*defobj_out = defobj;
return def;
}
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
static const Elf_Sym *
When looking up the symbol by dlsym, look it not only in the object given as dso handle, but also in the implicit dependencies of that dso. Also, const-ify the read-only parameter objlist of symlook_list. Reported by: "Simon 'corecode' Schubert" <corecode at fs ei tum de> Approved by: kan (mentor) X-MFC-After: 6.2
2006-09-08 14:59:54 +00:00
symlook_list(const char *name, unsigned long hash, const Objlist *objlist,
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
const Obj_Entry **defobj_out, const Ver_Entry *ventry, int flags,
DoneList *dlp)
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
{
const Elf_Sym *symp;
const Elf_Sym *def;
const Obj_Entry *defobj;
const Objlist_Entry *elm;
def = NULL;
defobj = NULL;
STAILQ_FOREACH(elm, objlist, link) {
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
if (donelist_check(dlp, elm->obj))
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
continue;
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if ((symp = symlook_obj(name, hash, elm->obj, ventry, flags)) != NULL) {
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
if (def == NULL || ELF_ST_BIND(symp->st_info) != STB_WEAK) {
def = symp;
defobj = elm->obj;
if (ELF_ST_BIND(def->st_info) != STB_WEAK)
break;
}
}
}
if (def != NULL)
*defobj_out = defobj;
return def;
}
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
/*
* Search the symbol table of a shared object and all objects needed
* by it for a symbol of the given name. Search order is
* breadth-first. Returns a pointer to the symbol, or NULL if no
* definition was found.
*/
static const Elf_Sym *
symlook_needed(const char *name, unsigned long hash, const Needed_Entry *needed,
const Obj_Entry **defobj_out, const Ver_Entry *ventry, int flags,
DoneList *dlp)
{
const Elf_Sym *def, *def_w;
const Needed_Entry *n;
const Obj_Entry *obj, *defobj, *defobj1;
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
def = def_w = NULL;
defobj = NULL;
for (n = needed; n != NULL; n = n->next) {
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
if ((obj = n->obj) == NULL ||
donelist_check(dlp, obj) ||
(def = symlook_obj(name, hash, obj, ventry, flags)) == NULL)
continue;
defobj = obj;
if (ELF_ST_BIND(def->st_info) != STB_WEAK) {
*defobj_out = defobj;
return (def);
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
}
}
/*
* There we come when either symbol definition is not found in
* directly needed objects, or found symbol is weak.
*/
for (n = needed; n != NULL; n = n->next) {
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
if ((obj = n->obj) == NULL)
continue;
def_w = symlook_needed(name, hash, obj->needed, &defobj1,
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
ventry, flags, dlp);
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
if (def_w == NULL)
continue;
if (def == NULL || ELF_ST_BIND(def_w->st_info) != STB_WEAK) {
def = def_w;
defobj = defobj1;
}
if (ELF_ST_BIND(def_w->st_info) != STB_WEAK)
break;
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
}
if (def != NULL)
Clean up white spaces and fix style(9).
2006-09-19 16:48:08 +00:00
*defobj_out = defobj;
return (def);
Fix the buggy rev. 1.117. dagmembers are only initialized for dlopen'ed dso that are actually loading. If dso a.so depends on b.so, then dlsym with handle from dlopen("b.so") will fail unconditionally. Correct implementation shall use the Obj_Entry.needed list to walk dependencies DAG. Test provided by: jkim Tested (prev. version) by: jkim, Nicolas Blais <nb_root at videotron ca>, h.blanke at chello nl Pointy hat to: kib Approved by: kan (mentor)
2006-09-19 12:47:13 +00:00
}
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
/*
* Search the symbol table of a single shared object for a symbol of
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
* the given name and version, if requested. Returns a pointer to the
* symbol, or NULL if no definition was found.
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
*
* The symbol's hash value is passed in for efficiency reasons; that
* eliminates many recomputations of the hash value.
*/
Add alpha support. Submitted by: John Birrell <jb@cimlogic.com.au> (with extra hacks by me) Obtained from: Probably NetBSD
1998-09-04 19:03:57 +00:00
const Elf_Sym *
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
symlook_obj(const char *name, unsigned long hash, const Obj_Entry *obj,
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
const Ver_Entry *ventry, int flags)
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
{
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
unsigned long symnum;
const Elf_Sym *vsymp;
Elf_Versym verndx;
int vcount;
if (obj->buckets == NULL)
return NULL;
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
vsymp = NULL;
vcount = 0;
symnum = obj->buckets[hash % obj->nbuckets];
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
for (; symnum != STN_UNDEF; symnum = obj->chains[symnum]) {
const Elf_Sym *symp;
const char *strp;
if (symnum >= obj->nchains)
Change many asserts into normal errors. They were all for conditions caused by invalid shared objects rather than by internal errors. Enable format string mismatch checking for _rtld_error().
1999-07-18 00:02:19 +00:00
return NULL; /* Bad object */
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
symp = obj->symtab + symnum;
strp = obj->strtab + symp->st_name;
switch (ELF_ST_TYPE(symp->st_info)) {
case STT_FUNC:
case STT_NOTYPE:
case STT_OBJECT:
if (symp->st_value == 0)
continue;
/* fallthrough */
case STT_TLS:
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
if (symp->st_shndx != SHN_UNDEF)
break;
#ifndef __mips__
else if (((flags & SYMLOOK_IN_PLT) == 0) &&
(ELF_ST_TYPE(symp->st_info) == STT_FUNC))
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
break;
/* fallthrough */
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
#endif
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
default:
continue;
}
if (name[0] != strp[0] || strcmp(name, strp) != 0)
continue;
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
if (ventry == NULL) {
if (obj->versyms != NULL) {
verndx = VER_NDX(obj->versyms[symnum]);
if (verndx > obj->vernum) {
_rtld_error("%s: symbol %s references wrong version %d",
obj->path, obj->strtab + symnum, verndx);
continue;
}
Make lookups for relocations from old unversioned binaries return oldest versioned symbol available. Do not accept hidden symbols for all other versions. Use "<obj->path>: <error message>" for all error messages in new functions to make them more consistent.
2005-12-24 15:37:30 +00:00
/*
* If we are not called from dlsym (i.e. this is a normal
* relocation from unversioned binary, accept the symbol
* immediately if it happens to have first version after
* this shared object became versioned. Otherwise, if
* symbol is versioned and not hidden, remember it. If it
* is the only symbol with this name exported by the
* shared object, it will be returned as a match at the
* end of the function. If symbol is global (verndx < 2)
* accept it unconditionally.
*/
if ((flags & SYMLOOK_DLSYM) == 0 && verndx == VER_NDX_GIVEN)
return symp;
else if (verndx >= VER_NDX_GIVEN) {
if ((obj->versyms[symnum] & VER_NDX_HIDDEN) == 0) {
if (vsymp == NULL)
vsymp = symp;
vcount ++;
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
continue;
}
}
return symp;
} else {
if (obj->versyms == NULL) {
if (object_match_name(obj, ventry->name)) {
In symlook_obj(): fix _rtld_error output. MFC After: 2 weeks
2006-08-04 13:37:54 +00:00
_rtld_error("%s: object %s should provide version %s for "
"symbol %s", obj_rtld.path, obj->path, ventry->name,
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
obj->strtab + symnum);
continue;
}
} else {
verndx = VER_NDX(obj->versyms[symnum]);
if (verndx > obj->vernum) {
_rtld_error("%s: symbol %s references wrong version %d",
obj->path, obj->strtab + symnum, verndx);
continue;
}
if (obj->vertab[verndx].hash != ventry->hash ||
strcmp(obj->vertab[verndx].name, ventry->name)) {
/*
Make lookups for relocations from old unversioned binaries return oldest versioned symbol available. Do not accept hidden symbols for all other versions. Use "<obj->path>: <error message>" for all error messages in new functions to make them more consistent.
2005-12-24 15:37:30 +00:00
* Version does not match. Look if this is a global symbol
* and if it is not hidden. If global symbol (verndx < 2)
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
* is available, use it. Do not return symbol if we are
* called by dlvsym, because dlvsym looks for a specific
* version and default one is not what dlvsym wants.
*/
if ((flags & SYMLOOK_DLSYM) ||
(obj->versyms[symnum] & VER_NDX_HIDDEN) ||
(verndx >= VER_NDX_GIVEN))
continue;
}
}
return symp;
Eliminate all machine-dependent code from the main source body and the Makefile, and move it down into the architecture-specific subdirectories. Eliminate an asm() statement for the i386. Make the dynamic linker work if it is built as an executable instead of as a shared library. See i386/Makefile.inc to find out how to do it. Note, this change is not enabled and it might never be enabled. But it might be useful in the future. Building the dynamic linker as an executable should make it start up faster, because it won't have any relocations. But in practice I suspect the difference is negligible.
1999-04-09 00:28:43 +00:00
}
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
return (vcount == 1) ? vsymp : NULL;
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
}
static void
trace_loaded_objects(Obj_Entry *obj)
Add support for ldd.
1998-05-01 08:39:27 +00:00
{
Add support such that if LD_TRACE_LOADED_OBJECTS_ALL is defined to a non-empty string in the environment; we indicate which objects caused each object to be loaded. PR: 30908 Submitted-by: Mike Meyer <mwm@mired.org>
2002-02-17 07:04:32 +00:00
char *fmt1, *fmt2, *fmt, *main_local, *list_containers;
Add support for ldd.
1998-05-01 08:39:27 +00:00
int c;
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
if ((main_local = getenv(LD_ "TRACE_LOADED_OBJECTS_PROGNAME")) == NULL)
Add support for ldd.
1998-05-01 08:39:27 +00:00
main_local = "";
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
if ((fmt1 = getenv(LD_ "TRACE_LOADED_OBJECTS_FMT1")) == NULL)
Add support for ldd.
1998-05-01 08:39:27 +00:00
fmt1 = "\t%o => %p (%x)\n";
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
if ((fmt2 = getenv(LD_ "TRACE_LOADED_OBJECTS_FMT2")) == NULL)
Add support for ldd.
1998-05-01 08:39:27 +00:00
fmt2 = "\t%o (%x)\n";
Add initial support for compiling a special 32 bit version of ld-elf.so.1 on 64 bit systems. Most of this involves using alternate paths, environment variables and diagnostic messages. The build glue is seperate.
2004-03-21 01:21:26 +00:00
list_containers = getenv(LD_ "TRACE_LOADED_OBJECTS_ALL");
Add support such that if LD_TRACE_LOADED_OBJECTS_ALL is defined to a non-empty string in the environment; we indicate which objects caused each object to be loaded. PR: 30908 Submitted-by: Mike Meyer <mwm@mired.org>
2002-02-17 07:04:32 +00:00
Add support for ldd.
1998-05-01 08:39:27 +00:00
for (; obj; obj = obj->next) {
Needed_Entry *needed;
char *name, *path;
bool is_lib;
Add support such that if LD_TRACE_LOADED_OBJECTS_ALL is defined to a non-empty string in the environment; we indicate which objects caused each object to be loaded. PR: 30908 Submitted-by: Mike Meyer <mwm@mired.org>
2002-02-17 07:04:32 +00:00
if (list_containers && obj->needed != NULL)
printf("%s:\n", obj->path);
Add support for ldd.
1998-05-01 08:39:27 +00:00
for (needed = obj->needed; needed; needed = needed->next) {
Suppress duplicate entries in ldd output.
1998-09-02 02:51:12 +00:00
if (needed->obj != NULL) {
Add support such that if LD_TRACE_LOADED_OBJECTS_ALL is defined to a non-empty string in the environment; we indicate which objects caused each object to be loaded. PR: 30908 Submitted-by: Mike Meyer <mwm@mired.org>
2002-02-17 07:04:32 +00:00
if (needed->obj->traced && !list_containers)
Suppress duplicate entries in ldd output.
1998-09-02 02:51:12 +00:00
continue;
needed->obj->traced = true;
Add support for ldd.
1998-05-01 08:39:27 +00:00
path = needed->obj->path;
Suppress duplicate entries in ldd output.
1998-09-02 02:51:12 +00:00
} else
path = "not found";
name = (char *)obj->strtab + needed->name;
is_lib = strncmp(name, "lib", 3) == 0; /* XXX - bogus */
Add support for ldd.
1998-05-01 08:39:27 +00:00
fmt = is_lib ? fmt1 : fmt2;
while ((c = *fmt++) != '\0') {
switch (c) {
default:
putchar(c);
continue;
case '\\':
switch (c = *fmt) {
case '\0':
continue;
case 'n':
putchar('\n');
break;
case 't':
putchar('\t');
break;
}
break;
case '%':
switch (c = *fmt) {
case '\0':
continue;
case '%':
default:
putchar(c);
break;
case 'A':
printf("%s", main_local);
break;
case 'a':
printf("%s", obj_main->path);
break;
case 'o':
printf("%s", name);
break;
#if 0
case 'm':
printf("%d", sodp->sod_major);
break;
case 'n':
printf("%d", sodp->sod_minor);
break;
#endif
case 'p':
printf("%s", path);
break;
case 'x':
printf("%p", needed->obj ? needed->obj->mapbase : 0);
break;
}
break;
}
++fmt;
}
}
}
}
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
/*
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
* Unload a dlopened object and its dependencies from memory and from
* our data structures. It is assumed that the DAG rooted in the
* object has already been unreferenced, and that the object has a
* reference count of 0.
Revamp the symbol lookup algorithm to cope better with objects loaded separately by dlopen that have global symbols with identical names. Viewing each dlopened object as a DAG which is linked by its DT_NEEDED entries in the dynamic table, the search order is as follows: * If the referencing object was linked with -Bsymbolic, search it internally. * Search all dlopened DAGs containing the referencing object. * Search all objects loaded at program start up. * Search all objects which were dlopened() using the RTLD_GLOBAL flag (which is now supported too). The search terminates as soon as a strong definition is found. Lacking that, the first weak definition is used. These rules match those of Solaris, as best I could determine them from its vague manual pages and the results of experiments I performed. PR: misc/12438
1999-08-30 01:48:19 +00:00
*/
Fix bug: if a dlopen() failed (e.g., because of undefined symbols), the dynamic linker didn't clean up properly. A subsequent dlopen() of the same object would appear to succeed. Another excellent fix from Max Khon. PR: bin/12471 Submitted by: Max Khon <fjoe@iclub.nsu.ru>
1999-07-09 16:22:55 +00:00
static void
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
unload_object(Obj_Entry *root)
Fix bug: if a dlopen() failed (e.g., because of undefined symbols), the dynamic linker didn't clean up properly. A subsequent dlopen() of the same object would appear to succeed. Another excellent fix from Max Khon. PR: bin/12471 Submitted by: Max Khon <fjoe@iclub.nsu.ru>
1999-07-09 16:22:55 +00:00
{
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
Obj_Entry *obj;
Obj_Entry **linkp;
assert(root->refcount == 0);
Do not remove object from the lists at the unref_dag() stage. Introduce a new unlink_object() function and call it in unload_object() instead. Removing the object in unref_dag() is too early, rtld calls _fini() function after that and shared objects might fail resolve their own symbols.
2003-02-17 20:58:27 +00:00
/*
* Pass over the DAG removing unreferenced objects from
* appropriate lists.
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
*/
Do not remove object from the lists at the unref_dag() stage. Introduce a new unlink_object() function and call it in unload_object() instead. Removing the object in unref_dag() is too early, rtld calls _fini() function after that and shared objects might fail resolve their own symbols.
2003-02-17 20:58:27 +00:00
unlink_object(root);
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
/* Unmap all objects that are no longer referenced. */
linkp = &obj_list->next;
while ((obj = *linkp) != NULL) {
if (obj->refcount == 0) {
Add various utrace's for use with ktrace to the ELF runtime linker. To activate the traces, set the LD_UTRACE (or LD_32_UTRACE) environment variable. This also includes code in kdump(8) to parse the traces. Reviewed by: kan, jdp MFC after: 2 weeks
2007-01-09 17:50:05 +00:00
LD_UTRACE(UTRACE_UNLOAD_OBJECT, obj, obj->mapbase, obj->mapsize, 0,
obj->path);
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
dbg("unloading \"%s\"", obj->path);
munmap(obj->mapbase, obj->mapsize);
linkmap_delete(obj);
*linkp = obj->next;
Solve the dynamic linker's problems with multithreaded programs once and for all (I hope). Packages such as wine, JDK, and linuxthreads should no longer have any problems with re-entering the dynamic linker. This commit replaces the locking used in the dynamic linker with a new spinlock-based reader/writer lock implementation. Brian Fundakowski Feldman <green> argued for this from the very beginning, but it took me a long time to come around to his point of view. Spinlocks are the only kinds of locks that work with all thread packages. But on uniprocessor systems they can be inefficient, because while a contender for the lock is spinning the holder of the lock cannot make any progress toward releasing it. To alleviate this disadvantage I have borrowed a trick from Sleepycat's Berkeley DB implementation. When spinning for a lock, the requester does a nanosleep() call for 1 usec. each time around the loop. This will generally yield the CPU to other threads, allowing the lock holder to finish its business and release the lock. I chose 1 usec. as the minimum sleep which would with reasonable certainty not be rounded down to 0. The formerly machine-independent file "lockdflt.c" has been moved into the architecture-specific subdirectories by repository copy. It now contains the machine-dependent spinlocking code. For the spinlocks I used the very nifty "simple, non-scalable reader-preference lock" which I found at <http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html> on all CPUs except the 80386 (the specific CPU model, not the architecture). The 80386 CPU doesn't support the necessary "cmpxchg" instruction, so on that CPU a simple exclusive test-and-set lock is used instead. 80386 CPUs are detected at initialization time by trying to execute "cmpxchg" and catching the resulting SIGILL signal. To reduce contention for the locks, I have revamped a couple of key data structures, permitting all common operations to be done under non-exclusive (reader) locking. The only operations that require exclusive locking now are the rare intrusive operations such as dlopen() and dlclose(). The dllockinit() interface is now deprecated. It still exists, but only as a do-nothing stub. I plan to remove it as soon as is reasonably possible. (From the very beginning it was clearly labeled as experimental and subject to change.) As far as I know, only the linuxthreads port uses dllockinit(). This interface turned out to have several problems. As one example, when the dynamic linker called a client-supplied locking function, that function sometimes needed lazy binding, causing re-entry into the dynamic linker and a big looping mess. And in any case, it turned out to be too burdensome to require threads packages to register themselves with the dynamic linker.
2000-07-08 04:10:38 +00:00
obj_count--;
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
obj_free(obj);
} else
linkp = &obj->next;
Fix bug: if a dlopen() failed (e.g., because of undefined symbols), the dynamic linker didn't clean up properly. A subsequent dlopen() of the same object would appear to succeed. Another excellent fix from Max Khon. PR: bin/12471 Submitted by: Max Khon <fjoe@iclub.nsu.ru>
1999-07-09 16:22:55 +00:00
}
Revamp the mechanism for enumerating and calling shared objects' init and fini functions. Now the code is very careful to hold no locks when calling these functions. Thus the dynamic linker cannot be re-entered with a lock already held. Remove the tolerance for recursive locking that I added in revision 1.2 of dllockinit.c. Recursive locking shouldn't happen any more. Mozilla and JDK users: I'd appreciate confirmation that things still work right (or at least the same) with these changes.
2000-01-09 21:13:48 +00:00
obj_tail = linkp;
Fix bug: if a dlopen() failed (e.g., because of undefined symbols), the dynamic linker didn't clean up properly. A subsequent dlopen() of the same object would appear to succeed. Another excellent fix from Max Khon. PR: bin/12471 Submitted by: Max Khon <fjoe@iclub.nsu.ru>
1999-07-09 16:22:55 +00:00
}
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
static void
Do not remove object from the lists at the unref_dag() stage. Introduce a new unlink_object() function and call it in unload_object() instead. Removing the object in unref_dag() is too early, rtld calls _fini() function after that and shared objects might fail resolve their own symbols.
2003-02-17 20:58:27 +00:00
unlink_object(Obj_Entry *root)
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
{
When unloading dependencies make sure they are removed from all the associated lists: remove RTLD_GLOBAL objects from global objects list; remove the parent object from dldags list of its children. Previosly we were doing that only to the top-level object OF the DAG being unloaded and all its dependencies were ignored, leading to mysterious crashes later. Submitted by: peter (partially)
2003-02-10 23:15:07 +00:00
Objlist_Entry *elm;
Add a NULL pointer check whose absence could cause segmentation violations in certain obscure cases involving failed dlopens. Many thanks to Archie Cobbs for providing me with a good test case. Eliminate a block that existed only to localize a declaration.
1999-08-20 22:33:44 +00:00
When unloading dependencies make sure they are removed from all the associated lists: remove RTLD_GLOBAL objects from global objects list; remove the parent object from dldags list of its children. Previosly we were doing that only to the top-level object OF the DAG being unloaded and all its dependencies were ignored, leading to mysterious crashes later. Submitted by: peter (partially)
2003-02-10 23:15:07 +00:00
if (root->refcount == 0) {
/* Remove the object from the RTLD_GLOBAL list. */
objlist_remove(&list_global, root);
/* Remove the object from all objects' DAG lists. */
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
STAILQ_FOREACH(elm, &root->dagmembers, link) {
When unloading dependencies make sure they are removed from all the associated lists: remove RTLD_GLOBAL objects from global objects list; remove the parent object from dldags list of its children. Previosly we were doing that only to the top-level object OF the DAG being unloaded and all its dependencies were ignored, leading to mysterious crashes later. Submitted by: peter (partially)
2003-02-10 23:15:07 +00:00
objlist_remove(&elm->obj->dldags, root);
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
if (elm->obj != root)
unlink_object(elm->obj);
}
When unloading dependencies make sure they are removed from all the associated lists: remove RTLD_GLOBAL objects from global objects list; remove the parent object from dldags list of its children. Previosly we were doing that only to the top-level object OF the DAG being unloaded and all its dependencies were ignored, leading to mysterious crashes later. Submitted by: peter (partially)
2003-02-10 23:15:07 +00:00
}
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
}
Do not remove object from the lists at the unref_dag() stage. Introduce a new unlink_object() function and call it in unload_object() instead. Removing the object in unref_dag() is too early, rtld calls _fini() function after that and shared objects might fail resolve their own symbols.
2003-02-17 20:58:27 +00:00
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
static void
ref_dag(Obj_Entry *root)
{
Objlist_Entry *elm;
In r214728, if dlopen() is called for the object that has been already loaded as a dependency and marked -z nodlopen, object' DAG is already initialized by load_needed_objects(). Due to this, the init_dag() call from dlopen() does not increment refcount for the object [1]. Change init_dag() to not increment DAG refcount. Require explicit calls to ref_dag() to increment, and assert that ref_dag() and unref_dag() are called for root that has dag initialized. To fix the noted issue, unconditionally call both init_dag() and ref_dag() in dlopen() for the case when the object was already loaded, making it similar to the case of newly loaded object. Noted by: jh [1] Reviewed by: jh, kan MFC after: 6 days
2010-11-04 09:29:00 +00:00
assert(root->dag_inited);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
STAILQ_FOREACH(elm, &root->dagmembers, link)
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
elm->obj->refcount++;
Do not remove object from the lists at the unref_dag() stage. Introduce a new unlink_object() function and call it in unload_object() instead. Removing the object in unref_dag() is too early, rtld calls _fini() function after that and shared objects might fail resolve their own symbols.
2003-02-17 20:58:27 +00:00
}
static void
unref_dag(Obj_Entry *root)
{
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
Objlist_Entry *elm;
Do not remove object from the lists at the unref_dag() stage. Introduce a new unlink_object() function and call it in unload_object() instead. Removing the object in unref_dag() is too early, rtld calls _fini() function after that and shared objects might fail resolve their own symbols.
2003-02-17 20:58:27 +00:00
In r214728, if dlopen() is called for the object that has been already loaded as a dependency and marked -z nodlopen, object' DAG is already initialized by load_needed_objects(). Due to this, the init_dag() call from dlopen() does not increment refcount for the object [1]. Change init_dag() to not increment DAG refcount. Require explicit calls to ref_dag() to increment, and assert that ref_dag() and unref_dag() are called for root that has dag initialized. To fix the noted issue, unconditionally call both init_dag() and ref_dag() in dlopen() for the case when the object was already loaded, making it similar to the case of newly loaded object. Noted by: jh [1] Reviewed by: jh, kan MFC after: 6 days
2010-11-04 09:29:00 +00:00
assert(root->dag_inited);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
STAILQ_FOREACH(elm, &root->dagmembers, link)
Rethink the way we count module references. Simply following DT_NEEDED links is not flexible enough for cases where dynamically loaded modules form a dependency cycle. This should fix an infinite recursion problem encountered by Yahoo. Approved by: re (jhb)
2003-05-08 01:31:36 +00:00
elm->obj->refcount--;
Style fixes. If it seems like a lot of lines of changes, it's because I moved some functions. Mr. Tidy likes them to be in alphabetical order.
1998-09-02 02:00:20 +00:00
}
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
/*
* Common code for MD __tls_get_addr().
*/
void *
tls_get_addr_common(Elf_Addr** dtvp, int index, size_t offset)
{
Elf_Addr* dtv = *dtvp;
Add locking code for tls routines.
2005-03-20 23:28:25 +00:00
int lockstate;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
/* Check dtv generation in case new modules have arrived */
if (dtv[0] != tls_dtv_generation) {
Elf_Addr* newdtv;
int to_copy;
Add locking code for tls routines.
2005-03-20 23:28:25 +00:00
lockstate = wlock_acquire(rtld_bind_lock);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
newdtv = calloc(1, (tls_max_index + 2) * sizeof(Elf_Addr));
to_copy = dtv[1];
if (to_copy > tls_max_index)
to_copy = tls_max_index;
memcpy(&newdtv[2], &dtv[2], to_copy * sizeof(Elf_Addr));
newdtv[0] = tls_dtv_generation;
newdtv[1] = tls_max_index;
free(dtv);
Add locking code for tls routines.
2005-03-20 23:28:25 +00:00
wlock_release(rtld_bind_lock, lockstate);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
*dtvp = newdtv;
}
/* Dynamically allocate module TLS if necessary */
Add locking code for tls routines.
2005-03-20 23:28:25 +00:00
if (!dtv[index + 1]) {
/* Signal safe, wlock will block out signals. */
lockstate = wlock_acquire(rtld_bind_lock);
if (!dtv[index + 1])
dtv[index + 1] = (Elf_Addr)allocate_module_tls(index);
wlock_release(rtld_bind_lock, lockstate);
}
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
return (void*) (dtv[index + 1] + offset);
}
Add stubs for powerpc TLS. Submitted by: ssouhlal
2004-08-04 19:12:14 +00:00
/* XXX not sure what variants to use for arm. */
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
#if defined(__ia64__) || defined(__powerpc__)
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
/*
* Allocate Static TLS using the Variant I method.
*/
void *
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
allocate_tls(Obj_Entry *objs, void *oldtcb, size_t tcbsize, size_t tcbalign)
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
{
Obj_Entry *obj;
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
char *tcb;
Elf_Addr **tls;
Elf_Addr *dtv;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Elf_Addr addr;
int i;
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
if (oldtcb != NULL && tcbsize == TLS_TCB_SIZE)
return (oldtcb);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
assert(tcbsize >= TLS_TCB_SIZE);
tcb = calloc(1, tls_static_space - TLS_TCB_SIZE + tcbsize);
tls = (Elf_Addr **)(tcb + tcbsize - TLS_TCB_SIZE);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
if (oldtcb != NULL) {
memcpy(tls, oldtcb, tls_static_space);
free(oldtcb);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
/* Adjust the DTV. */
dtv = tls[0];
for (i = 0; i < dtv[1]; i++) {
if (dtv[i+2] >= (Elf_Addr)oldtcb &&
dtv[i+2] < (Elf_Addr)oldtcb + tls_static_space) {
dtv[i+2] = dtv[i+2] - (Elf_Addr)oldtcb + (Elf_Addr)tls;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
}
}
} else {
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
dtv = calloc(tls_max_index + 2, sizeof(Elf_Addr));
tls[0] = dtv;
dtv[0] = tls_dtv_generation;
dtv[1] = tls_max_index;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
for (obj = objs; obj; obj = obj->next) {
Improve TLS variant I: o Use obj->tlsinitsize to determine whether there's initialized data. o If obj->tlssize > obj->tlsinitsize, then bzero uninitialized data. o Don't exclude variant I from the work-around in free_tls_offset().
2010-02-16 02:48:11 +00:00
if (obj->tlsoffset > 0) {
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
addr = (Elf_Addr)tls + obj->tlsoffset;
Improve TLS variant I: o Use obj->tlsinitsize to determine whether there's initialized data. o If obj->tlssize > obj->tlsinitsize, then bzero uninitialized data. o Don't exclude variant I from the work-around in free_tls_offset().
2010-02-16 02:48:11 +00:00
if (obj->tlsinitsize > 0)
memcpy((void*) addr, obj->tlsinit, obj->tlsinitsize);
if (obj->tlssize > obj->tlsinitsize)
memset((void*) (addr + obj->tlsinitsize), 0,
obj->tlssize - obj->tlsinitsize);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
dtv[obj->tlsindex + 1] = addr;
}
}
}
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
return (tcb);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
}
void
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
free_tls(void *tcb, size_t tcbsize, size_t tcbalign)
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
{
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
Elf_Addr *dtv;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Elf_Addr tlsstart, tlsend;
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
int dtvsize, i;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
assert(tcbsize >= TLS_TCB_SIZE);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
tlsstart = (Elf_Addr)tcb + tcbsize - TLS_TCB_SIZE;
tlsend = tlsstart + tls_static_space;
dtv = *(Elf_Addr **)tlsstart;
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
dtvsize = dtv[1];
for (i = 0; i < dtvsize; i++) {
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
if (dtv[i+2] && (dtv[i+2] < tlsstart || dtv[i+2] >= tlsend)) {
free((void*)dtv[i+2]);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
}
}
Fix the variant I allocation for KSE: Allow a larger TCB and assume that the documented TCB is at the tail of the extended TCB. In other words, the base of the TCB has a negative offset from the TLS.
2006-09-01 06:13:16 +00:00
free(dtv);
free(tcb);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
}
#endif
Add stubs for TLS. Arbitraly choose the 2nd variant until I figure out which one I should use.
2004-09-23 23:04:52 +00:00
#if defined(__i386__) || defined(__amd64__) || defined(__sparc64__) || \
MFp4: Add mips support for dynamic linking. This code came from the merged mips2 and Juniper mips repositories. Warner Losh, Randall Seager, Oleksandr Tymoshenko and Olivier Houchard worked to merge, debug and integrate this code. This code may also contain code derived from NetBSD.
2008-04-04 20:59:26 +00:00
defined(__arm__) || defined(__mips__)
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
/*
* Allocate Static TLS using the Variant II method.
*/
void *
allocate_tls(Obj_Entry *objs, void *oldtls, size_t tcbsize, size_t tcbalign)
{
Obj_Entry *obj;
size_t size;
char *tls;
Elf_Addr *dtv, *olddtv;
Elf_Addr segbase, oldsegbase, addr;
int i;
size = round(tls_static_space, tcbalign);
assert(tcbsize >= 2*sizeof(Elf_Addr));
Er, forgot to clear tls space to zero for Variant II.
2006-03-28 06:14:22 +00:00
tls = calloc(1, size + tcbsize);
When allocating TLS and DTV, make sure that any unused slots in the DTV are initialised to zero. When freeing TLS, don't attempt to free DTV slots which were not used. Pointed out by: Joerg Sonnenberger X-MFC-After: After the branch, probably
2005-03-30 08:28:26 +00:00
dtv = calloc(1, (tls_max_index + 2) * sizeof(Elf_Addr));
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
segbase = (Elf_Addr)(tls + size);
((Elf_Addr*)segbase)[0] = segbase;
((Elf_Addr*)segbase)[1] = (Elf_Addr) dtv;
dtv[0] = tls_dtv_generation;
dtv[1] = tls_max_index;
if (oldtls) {
/*
* Copy the static TLS block over whole.
*/
oldsegbase = (Elf_Addr) oldtls;
memcpy((void *)(segbase - tls_static_space),
(const void *)(oldsegbase - tls_static_space),
tls_static_space);
/*
* If any dynamic TLS blocks have been created tls_get_addr(),
* move them over.
*/
olddtv = ((Elf_Addr**)oldsegbase)[1];
for (i = 0; i < olddtv[1]; i++) {
if (olddtv[i+2] < oldsegbase - size || olddtv[i+2] > oldsegbase) {
dtv[i+2] = olddtv[i+2];
olddtv[i+2] = 0;
}
}
/*
* We assume that this block was the one we created with
* allocate_initial_tls().
*/
free_tls(oldtls, 2*sizeof(Elf_Addr), sizeof(Elf_Addr));
} else {
for (obj = objs; obj; obj = obj->next) {
if (obj->tlsoffset) {
addr = segbase - obj->tlsoffset;
memset((void*) (addr + obj->tlsinitsize),
0, obj->tlssize - obj->tlsinitsize);
if (obj->tlsinit)
memcpy((void*) addr, obj->tlsinit, obj->tlsinitsize);
dtv[obj->tlsindex + 1] = addr;
}
}
}
return (void*) segbase;
}
void
free_tls(void *tls, size_t tcbsize, size_t tcbalign)
{
size_t size;
Elf_Addr* dtv;
int dtvsize, i;
Elf_Addr tlsstart, tlsend;
/*
* Figure out the size of the initial TLS block so that we can
* find stuff which ___tls_get_addr() allocated dynamically.
*/
size = round(tls_static_space, tcbalign);
dtv = ((Elf_Addr**)tls)[1];
dtvsize = dtv[1];
tlsend = (Elf_Addr) tls;
tlsstart = tlsend - size;
for (i = 0; i < dtvsize; i++) {
When allocating TLS and DTV, make sure that any unused slots in the DTV are initialised to zero. When freeing TLS, don't attempt to free DTV slots which were not used. Pointed out by: Joerg Sonnenberger X-MFC-After: After the branch, probably
2005-03-30 08:28:26 +00:00
if (dtv[i+2] && (dtv[i+2] < tlsstart || dtv[i+2] > tlsend)) {
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
free((void*) dtv[i+2]);
}
}
free((void*) tlsstart);
Fix a TLS memory leak. PR: threads/112297 MFC: 1 week
2007-05-05 08:44:59 +00:00
free((void*) dtv);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
}
#endif
Allow threading libraries to register their own locking implementation in case default one provided by rtld is not suitable. Consolidate various identical MD lock implementation into a single file using appropriate machine/atomic.h. Approved by: re (scottl)
2003-05-29 22:58:26 +00:00
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
/*
* Allocate TLS block for module with given index.
*/
void *
allocate_module_tls(int index)
{
Obj_Entry* obj;
char* p;
for (obj = obj_list; obj; obj = obj->next) {
if (obj->tlsindex == index)
break;
}
if (!obj) {
_rtld_error("Can't find module with TLS index %d", index);
die();
}
p = malloc(obj->tlssize);
Only use the cache after the early stage of loading. This is because calling mmap() etc. may use GOT which is not set up yet. Use calloc() instead of mmap() in cases where this was the case before (sparc64, powerpc, arm). Submitted by: Dimitry Andric (dimitry andric com) Reviewed by: kan Approved by: ed (mentor)
2010-05-18 08:55:23 +00:00
if (p == NULL) {
_rtld_error("Cannot allocate TLS block for index %d", index);
die();
}
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
memcpy(p, obj->tlsinit, obj->tlsinitsize);
memset(p + obj->tlsinitsize, 0, obj->tlssize - obj->tlsinitsize);
return p;
}
bool
allocate_tls_offset(Obj_Entry *obj)
{
size_t off;
if (obj->tls_done)
return true;
if (obj->tlssize == 0) {
obj->tls_done = true;
return true;
}
if (obj->tlsindex == 1)
off = calculate_first_tls_offset(obj->tlssize, obj->tlsalign);
else
off = calculate_tls_offset(tls_last_offset, tls_last_size,
obj->tlssize, obj->tlsalign);
/*
* If we have already fixed the size of the static TLS block, we
* must stay within that size. When allocating the static TLS, we
* leave a small amount of space spare to be used for dynamically
* loading modules which use static TLS.
*/
if (tls_static_space) {
if (calculate_tls_end(off, obj->tlssize) > tls_static_space)
return false;
}
tls_last_offset = obj->tlsoffset = off;
tls_last_size = obj->tlssize;
obj->tls_done = true;
return true;
}
Attempt to free any static TLS space used by a shared library when it is unloaded. This allows applications which load and unload libraries like libGL.so.1 several times to work properly. MFC after: 2 days
2005-02-27 12:55:40 +00:00
void
free_tls_offset(Obj_Entry *obj)
{
Improve TLS variant I: o Use obj->tlsinitsize to determine whether there's initialized data. o If obj->tlssize > obj->tlsinitsize, then bzero uninitialized data. o Don't exclude variant I from the work-around in free_tls_offset().
2010-02-16 02:48:11 +00:00
Attempt to free any static TLS space used by a shared library when it is unloaded. This allows applications which load and unload libraries like libGL.so.1 several times to work properly. MFC after: 2 days
2005-02-27 12:55:40 +00:00
/*
* If we were the last thing to allocate out of the static TLS
* block, we give our space back to the 'allocator'. This is a
* simplistic workaround to allow libGL.so.1 to be loaded and
Improve TLS variant I: o Use obj->tlsinitsize to determine whether there's initialized data. o If obj->tlssize > obj->tlsinitsize, then bzero uninitialized data. o Don't exclude variant I from the work-around in free_tls_offset().
2010-02-16 02:48:11 +00:00
* unloaded multiple times.
Attempt to free any static TLS space used by a shared library when it is unloaded. This allows applications which load and unload libraries like libGL.so.1 several times to work properly. MFC after: 2 days
2005-02-27 12:55:40 +00:00
*/
if (calculate_tls_end(obj->tlsoffset, obj->tlssize)
== calculate_tls_end(tls_last_offset, tls_last_size)) {
tls_last_offset -= obj->tlssize;
tls_last_size = 0;
}
}
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
void *
_rtld_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign)
{
Add locking code for tls routines.
2005-03-20 23:28:25 +00:00
void *ret;
int lockstate;
lockstate = wlock_acquire(rtld_bind_lock);
ret = allocate_tls(obj_list, oldtls, tcbsize, tcbalign);
wlock_release(rtld_bind_lock, lockstate);
return (ret);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
}
void
_rtld_free_tls(void *tcb, size_t tcbsize, size_t tcbalign)
{
Add locking code for tls routines.
2005-03-20 23:28:25 +00:00
int lockstate;
lockstate = wlock_acquire(rtld_bind_lock);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
free_tls(tcb, tcbsize, tcbalign);
Add locking code for tls routines.
2005-03-20 23:28:25 +00:00
wlock_release(rtld_bind_lock, lockstate);
Add support for Thread Local Storage.
2004-08-03 08:51:00 +00:00
}
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
static void
object_add_name(Obj_Entry *obj, const char *name)
{
Name_Entry *entry;
size_t len;
len = strlen(name);
entry = malloc(sizeof(Name_Entry) + len);
if (entry != NULL) {
strcpy(entry->name, name);
STAILQ_INSERT_TAIL(&obj->names, entry, link);
}
}
static int
object_match_name(const Obj_Entry *obj, const char *name)
{
Name_Entry *entry;
STAILQ_FOREACH(entry, &obj->names, link) {
if (strcmp(name, entry->name) == 0)
return (1);
}
return (0);
}
static Obj_Entry *
locate_dependency(const Obj_Entry *obj, const char *name)
{
const Objlist_Entry *entry;
const Needed_Entry *needed;
STAILQ_FOREACH(entry, &list_main, link) {
if (object_match_name(entry->obj, name))
return entry->obj;
}
for (needed = obj->needed; needed != NULL; needed = needed->next) {
if (needed->obj == NULL)
continue;
if (object_match_name(needed->obj, name))
return needed->obj;
}
Revert r209544. It papers over a binutils bug, and is not even a reliable fix for it. The bug occurs when using the --as-needed flag to ld in the presence of synthetic linker-generated symbols that reference symbols defined in linked-to shared libraries with versioned symbols. When the only symbols used from a library fall into this category, ld will drop the DT_NEEDED entry for it, but retain the versioning information. This bug is best fixed/hacked around in binutils, not in rtld. Discussed with: kan
2010-06-28 01:40:56 +00:00
_rtld_error("%s: Unexpected inconsistency: dependency %s not found",
obj->path, name);
die();
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
}
static int
check_object_provided_version(Obj_Entry *refobj, const Obj_Entry *depobj,
const Elf_Vernaux *vna)
{
const Elf_Verdef *vd;
const char *vername;
vername = refobj->strtab + vna->vna_name;
vd = depobj->verdef;
if (vd == NULL) {
Make lookups for relocations from old unversioned binaries return oldest versioned symbol available. Do not accept hidden symbols for all other versions. Use "<obj->path>: <error message>" for all error messages in new functions to make them more consistent.
2005-12-24 15:37:30 +00:00
_rtld_error("%s: version %s required by %s not defined",
depobj->path, vername, refobj->path);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
return (-1);
}
for (;;) {
if (vd->vd_version != VER_DEF_CURRENT) {
Make lookups for relocations from old unversioned binaries return oldest versioned symbol available. Do not accept hidden symbols for all other versions. Use "<obj->path>: <error message>" for all error messages in new functions to make them more consistent.
2005-12-24 15:37:30 +00:00
_rtld_error("%s: Unsupported version %d of Elf_Verdef entry",
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
depobj->path, vd->vd_version);
return (-1);
}
if (vna->vna_hash == vd->vd_hash) {
const Elf_Verdaux *aux = (const Elf_Verdaux *)
((char *)vd + vd->vd_aux);
if (strcmp(vername, depobj->strtab + aux->vda_name) == 0)
return (0);
}
if (vd->vd_next == 0)
break;
vd = (const Elf_Verdef *) ((char *)vd + vd->vd_next);
}
if (vna->vna_flags & VER_FLG_WEAK)
return (0);
Make lookups for relocations from old unversioned binaries return oldest versioned symbol available. Do not accept hidden symbols for all other versions. Use "<obj->path>: <error message>" for all error messages in new functions to make them more consistent.
2005-12-24 15:37:30 +00:00
_rtld_error("%s: version %s required by %s not found",
depobj->path, vername, refobj->path);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
return (-1);
}
static int
rtld_verify_object_versions(Obj_Entry *obj)
{
const Elf_Verneed *vn;
const Elf_Verdef *vd;
const Elf_Verdaux *vda;
const Elf_Vernaux *vna;
const Obj_Entry *depobj;
int maxvernum, vernum;
maxvernum = 0;
/*
* Walk over defined and required version records and figure out
* max index used by any of them. Do very basic sanity checking
* while there.
*/
vn = obj->verneed;
while (vn != NULL) {
if (vn->vn_version != VER_NEED_CURRENT) {
Make lookups for relocations from old unversioned binaries return oldest versioned symbol available. Do not accept hidden symbols for all other versions. Use "<obj->path>: <error message>" for all error messages in new functions to make them more consistent.
2005-12-24 15:37:30 +00:00
_rtld_error("%s: Unsupported version %d of Elf_Verneed entry",
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
obj->path, vn->vn_version);
return (-1);
}
vna = (const Elf_Vernaux *) ((char *)vn + vn->vn_aux);
for (;;) {
vernum = VER_NEED_IDX(vna->vna_other);
if (vernum > maxvernum)
maxvernum = vernum;
if (vna->vna_next == 0)
break;
vna = (const Elf_Vernaux *) ((char *)vna + vna->vna_next);
}
if (vn->vn_next == 0)
break;
vn = (const Elf_Verneed *) ((char *)vn + vn->vn_next);
}
vd = obj->verdef;
while (vd != NULL) {
if (vd->vd_version != VER_DEF_CURRENT) {
Make lookups for relocations from old unversioned binaries return oldest versioned symbol available. Do not accept hidden symbols for all other versions. Use "<obj->path>: <error message>" for all error messages in new functions to make them more consistent.
2005-12-24 15:37:30 +00:00
_rtld_error("%s: Unsupported version %d of Elf_Verdef entry",
obj->path, vd->vd_version);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
return (-1);
}
vernum = VER_DEF_IDX(vd->vd_ndx);
if (vernum > maxvernum)
maxvernum = vernum;
if (vd->vd_next == 0)
break;
vd = (const Elf_Verdef *) ((char *)vd + vd->vd_next);
}
if (maxvernum == 0)
return (0);
/*
* Store version information in array indexable by version index.
* Verify that object version requirements are satisfied along the
* way.
*/
obj->vernum = maxvernum + 1;
obj->vertab = calloc(obj->vernum, sizeof(Ver_Entry));
vd = obj->verdef;
while (vd != NULL) {
if ((vd->vd_flags & VER_FLG_BASE) == 0) {
vernum = VER_DEF_IDX(vd->vd_ndx);
assert(vernum <= maxvernum);
vda = (const Elf_Verdaux *)((char *)vd + vd->vd_aux);
obj->vertab[vernum].hash = vd->vd_hash;
obj->vertab[vernum].name = obj->strtab + vda->vda_name;
obj->vertab[vernum].file = NULL;
obj->vertab[vernum].flags = 0;
}
if (vd->vd_next == 0)
break;
vd = (const Elf_Verdef *) ((char *)vd + vd->vd_next);
}
vn = obj->verneed;
while (vn != NULL) {
depobj = locate_dependency(obj, obj->strtab + vn->vn_file);
vna = (const Elf_Vernaux *) ((char *)vn + vn->vn_aux);
for (;;) {
if (check_object_provided_version(obj, depobj, vna))
return (-1);
vernum = VER_NEED_IDX(vna->vna_other);
assert(vernum <= maxvernum);
obj->vertab[vernum].hash = vna->vna_hash;
obj->vertab[vernum].name = obj->strtab + vna->vna_name;
obj->vertab[vernum].file = obj->strtab + vn->vn_file;
obj->vertab[vernum].flags = (vna->vna_other & VER_NEED_HIDDEN) ?
VER_INFO_HIDDEN : 0;
if (vna->vna_next == 0)
break;
vna = (const Elf_Vernaux *) ((char *)vna + vna->vna_next);
}
if (vn->vn_next == 0)
break;
vn = (const Elf_Verneed *) ((char *)vn + vn->vn_next);
}
return 0;
}
static int
rtld_verify_versions(const Objlist *objlist)
{
Objlist_Entry *entry;
int rc;
rc = 0;
STAILQ_FOREACH(entry, objlist, link) {
/*
* Skip dummy objects or objects that have their version requirements
* already checked.
*/
if (entry->obj->strtab == NULL || entry->obj->vertab != NULL)
continue;
if (rtld_verify_object_versions(entry->obj) == -1) {
rc = -1;
if (ld_tracing == NULL)
break;
}
}
Catch up on rtld's special status. Since it does not appear on main object list, its versioning information needs to be examined separately. This hopefully fixes problems that people running with SYMVER_ENABLED are experiencing.
2007-04-07 23:17:00 +00:00
if (rc == 0 || ld_tracing != NULL)
rc = rtld_verify_object_versions(&obj_rtld);
Implement ELF symbol versioning using GNU semantics. This code aims to be compatible with symbol versioning support as implemented by GNU libc and documented by http://people.redhat.com/~drepper/symbol-versioning and LSB 3.0. Implement dlvsym() function to allow lookups for a specific version of a given symbol.
2005-12-18 19:43:33 +00:00
return rc;
}
const Ver_Entry *
fetch_ventry(const Obj_Entry *obj, unsigned long symnum)
{
Elf_Versym vernum;
if (obj->vertab) {
vernum = VER_NDX(obj->versyms[symnum]);
if (vernum >= obj->vernum) {
_rtld_error("%s: symbol %s has wrong verneed value %d",
obj->path, obj->strtab + symnum, vernum);
} else if (obj->vertab[vernum].hash != 0) {
return &obj->vertab[vernum];
}
}
return NULL;
}
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
On shared object unload, in __cxa_finalize, call and clear all installed atexit and __cxa_atexit handlers that are either installed by unloaded dso, or points to the functions provided by the dso. Use _rtld_addr_phdr to locate segment information from the address of private variable belonging to the dso, supplied by crtstuff.c. Provide utility function __elf_phdr_match_addr to do the match of address against dso executable segment. Call back into libthr from __cxa_finalize using weak __pthread_cxa_finalize symbol to remove any atfork handler which function points into unloaded object. The rtld needs private __pthread_cxa_finalize symbol to not require resolution of the weak undefined symbol at initialization time. This cannot work, since rtld is relocated before sym_zero is set up. Idea by: kan Reviewed by: kan (previous version) MFC after: 3 weeks
2010-08-23 15:38:02 +00:00
/*
* Overrides for libc_pic-provided functions.
*/
Use the newly provided aux vectors to get pagesize and osreldate information. Use local version of getpagesize(), rtld_getpagesize() in private allocator. Override the __getosreldate() previously fetched from libc_pic.a with local version that uses aux value if present. Note that __getosreldate() is used by rtld indirectly, by mmap(2) libc wrapper. To be able to utilize aux, split digest_dynamic() for use by init_rtld() into two parts, where the first one does not call malloc(), and the second part uses it. init_rtld() is able to initialize global variables before digest_dynamic2() calls. In particular, pagesize and osreldate are set up from the aux values. Now, rtld avoids (two) sysctl calls in startup. Tested by: marius (sparc64) MFC after: 1 month
2010-08-17 09:05:39 +00:00
int
__getosreldate(void)
{
size_t len;
int oid[2];
int error, osrel;
if (osreldate != 0)
return (osreldate);
oid[0] = CTL_KERN;
oid[1] = KERN_OSRELDATE;
osrel = 0;
len = sizeof(osrel);
error = sysctl(oid, 2, &osrel, &len, NULL, 0);
if (error == 0 && osrel > 0 && len == sizeof(osrel))
osreldate = osrel;
return (osreldate);
}
On shared object unload, in __cxa_finalize, call and clear all installed atexit and __cxa_atexit handlers that are either installed by unloaded dso, or points to the functions provided by the dso. Use _rtld_addr_phdr to locate segment information from the address of private variable belonging to the dso, supplied by crtstuff.c. Provide utility function __elf_phdr_match_addr to do the match of address against dso executable segment. Call back into libthr from __cxa_finalize using weak __pthread_cxa_finalize symbol to remove any atfork handler which function points into unloaded object. The rtld needs private __pthread_cxa_finalize symbol to not require resolution of the weak undefined symbol at initialization time. This cannot work, since rtld is relocated before sym_zero is set up. Idea by: kan Reviewed by: kan (previous version) MFC after: 3 weeks
2010-08-23 15:38:02 +00:00
/*
* No unresolved symbols for rtld.
*/
void
__pthread_cxa_finalize(struct dl_phdr_info *a)
{
}
Reference in New Issue Copy Permalink