freebsd-skq/sys/kern/link_elf_obj.c

1612 lines
40 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 1998-2000 Doug Rabson
* Copyright (c) 2004 Peter Wemm
* 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 AND CONTRIBUTORS ``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 OR CONTRIBUTORS 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.
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_ddb.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/mount.h>
#include <sys/proc.h>
#include <sys/namei.h>
#include <sys/fcntl.h>
#include <sys/vnode.h>
#include <sys/linker.h>
#include <machine/elf.h>
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
#include <net/vnet.h>
#include <security/mac/mac_framework.h>
1998-09-11 08:46:15 +00:00
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_object.h>
#include <vm/vm_kern.h>
#include <vm/vm_extern.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <sys/link_elf.h>
1998-09-11 08:46:15 +00:00
#ifdef DDB_CTF
#include <sys/zlib.h>
#endif
#include "linker_if.h"
typedef struct {
void *addr;
Elf_Off size;
int flags;
int sec; /* Original section */
char *name;
} Elf_progent;
typedef struct {
Elf_Rel *rel;
int nrel;
int sec;
} Elf_relent;
typedef struct {
Elf_Rela *rela;
int nrela;
int sec;
} Elf_relaent;
typedef struct elf_file {
struct linker_file lf; /* Common fields */
int preloaded;
caddr_t address; /* Relocation address */
vm_object_t object; /* VM object to hold file pages */
Elf_Shdr *e_shdr;
Elf_progent *progtab;
u_int nprogtab;
Elf_relaent *relatab;
u_int nrelatab;
Elf_relent *reltab;
int nreltab;
Elf_Sym *ddbsymtab; /* The symbol table we are using */
long ddbsymcnt; /* Number of symbols */
caddr_t ddbstrtab; /* String table */
long ddbstrcnt; /* number of bytes in string table */
caddr_t shstrtab; /* Section name string table */
long shstrcnt; /* number of bytes in string table */
caddr_t ctftab; /* CTF table */
long ctfcnt; /* number of bytes in CTF table */
caddr_t ctfoff; /* CTF offset table */
caddr_t typoff; /* Type offset table */
long typlen; /* Number of type entries. */
} *elf_file_t;
#include <kern/kern_ctf.c>
First round implementation of a fine grain enhanced module to module version dependency system. This isn't quite finished, but it is at a useful stage to do a functional checkpoint. Highlights: - version and dependency metadata is gathered via linker sets, so things are handled the same for static kernels and code built to live in a kld. - The dependencies are at module level (versus at file level). - Dependencies determine kld symbol search order - this means that you cannot link against symbols in another file unless you depend on it. This is so that you cannot accidently unload the target out from underneath the ones referencing it. - It is flexible enough that we can put tags in #include files and macros so that we can get decent hooks for enforcing recompiles on incompatable ABI changes. eg: if we change struct proc, we could force a recompile for all kld's that reference the proc struct. - Tangled dependency references at boot time are sorted. Files are relocated once all their dependencies are already relocated. Caveats: - Loader support is incomplete, but has been worked on seperately. - Actual enforcement of the version number tags is not active yet - just the module dependencies are live. The actual structure of versioning hasn't been agreed on yet. (eg: major.minor, or whatever) - There is some backwards compatability for old modules without metadata but I'm not sure how good it is. This is based on work originally done by Boris Popov (bp@freebsd.org), but I'm not sure he'd recognize much of it now. Don't blame him. :-) Also, ideas have been borrowed from Mike Smith.
2000-04-29 13:19:31 +00:00
static int link_elf_link_preload(linker_class_t cls,
const char *, linker_file_t *);
First round implementation of a fine grain enhanced module to module version dependency system. This isn't quite finished, but it is at a useful stage to do a functional checkpoint. Highlights: - version and dependency metadata is gathered via linker sets, so things are handled the same for static kernels and code built to live in a kld. - The dependencies are at module level (versus at file level). - Dependencies determine kld symbol search order - this means that you cannot link against symbols in another file unless you depend on it. This is so that you cannot accidently unload the target out from underneath the ones referencing it. - It is flexible enough that we can put tags in #include files and macros so that we can get decent hooks for enforcing recompiles on incompatable ABI changes. eg: if we change struct proc, we could force a recompile for all kld's that reference the proc struct. - Tangled dependency references at boot time are sorted. Files are relocated once all their dependencies are already relocated. Caveats: - Loader support is incomplete, but has been worked on seperately. - Actual enforcement of the version number tags is not active yet - just the module dependencies are live. The actual structure of versioning hasn't been agreed on yet. (eg: major.minor, or whatever) - There is some backwards compatability for old modules without metadata but I'm not sure how good it is. This is based on work originally done by Boris Popov (bp@freebsd.org), but I'm not sure he'd recognize much of it now. Don't blame him. :-) Also, ideas have been borrowed from Mike Smith.
2000-04-29 13:19:31 +00:00
static int link_elf_link_preload_finish(linker_file_t);
static int link_elf_load_file(linker_class_t, const char *, linker_file_t *);
static int link_elf_lookup_symbol(linker_file_t, const char *,
c_linker_sym_t *);
static int link_elf_symbol_values(linker_file_t, c_linker_sym_t,
linker_symval_t *);
static int link_elf_search_symbol(linker_file_t, caddr_t value,
c_linker_sym_t *sym, long *diffp);
static void link_elf_unload_file(linker_file_t);
static int link_elf_lookup_set(linker_file_t, const char *,
void ***, void ***, int *);
static int link_elf_each_function_name(linker_file_t,
int (*)(const char *, void *), void *);
static int link_elf_each_function_nameval(linker_file_t,
linker_function_nameval_callback_t,
void *);
static int link_elf_reloc_local(linker_file_t, bool);
static long link_elf_symtab_get(linker_file_t, const Elf_Sym **);
static long link_elf_strtab_get(linker_file_t, caddr_t *);
static int elf_obj_lookup(linker_file_t lf, Elf_Size symidx, int deps,
Elf_Addr *);
static kobj_method_t link_elf_methods[] = {
KOBJMETHOD(linker_lookup_symbol, link_elf_lookup_symbol),
KOBJMETHOD(linker_symbol_values, link_elf_symbol_values),
KOBJMETHOD(linker_search_symbol, link_elf_search_symbol),
KOBJMETHOD(linker_unload, link_elf_unload_file),
KOBJMETHOD(linker_load_file, link_elf_load_file),
KOBJMETHOD(linker_link_preload, link_elf_link_preload),
KOBJMETHOD(linker_link_preload_finish, link_elf_link_preload_finish),
KOBJMETHOD(linker_lookup_set, link_elf_lookup_set),
KOBJMETHOD(linker_each_function_name, link_elf_each_function_name),
KOBJMETHOD(linker_each_function_nameval, link_elf_each_function_nameval),
KOBJMETHOD(linker_ctf_get, link_elf_ctf_get),
KOBJMETHOD(linker_symtab_get, link_elf_symtab_get),
KOBJMETHOD(linker_strtab_get, link_elf_strtab_get),
{ 0, 0 }
};
static struct linker_class link_elf_class = {
#if ELF_TARG_CLASS == ELFCLASS32
"elf32_obj",
#else
"elf64_obj",
#endif
link_elf_methods, sizeof(struct elf_file)
};
static int relocate_file(elf_file_t ef);
static void elf_obj_cleanup_globals_cache(elf_file_t);
static void
link_elf_error(const char *filename, const char *s)
{
if (filename == NULL)
printf("kldload: %s\n", s);
else
printf("kldload: %s: %s\n", filename, s);
}
static void
link_elf_init(void *arg)
{
linker_add_class(&link_elf_class);
}
SYSINIT(link_elf_obj, SI_SUB_KLD, SI_ORDER_SECOND, link_elf_init, NULL);
static int
link_elf_link_preload(linker_class_t cls, const char *filename,
linker_file_t *result)
{
Elf_Ehdr *hdr;
Elf_Shdr *shdr;
Elf_Sym *es;
void *modptr, *baseptr, *sizeptr;
char *type;
elf_file_t ef;
linker_file_t lf;
Elf_Addr off;
int error, i, j, pb, ra, rl, shstrindex, symstrindex, symtabindex;
/* Look to see if we have the file preloaded */
modptr = preload_search_by_name(filename);
if (modptr == NULL)
return ENOENT;
type = (char *)preload_search_info(modptr, MODINFO_TYPE);
baseptr = preload_search_info(modptr, MODINFO_ADDR);
sizeptr = preload_search_info(modptr, MODINFO_SIZE);
hdr = (Elf_Ehdr *)preload_search_info(modptr, MODINFO_METADATA |
MODINFOMD_ELFHDR);
shdr = (Elf_Shdr *)preload_search_info(modptr, MODINFO_METADATA |
MODINFOMD_SHDR);
if (type == NULL || (strcmp(type, "elf" __XSTRING(__ELF_WORD_SIZE)
" obj module") != 0 &&
strcmp(type, "elf obj module") != 0)) {
return (EFTYPE);
}
if (baseptr == NULL || sizeptr == NULL || hdr == NULL ||
shdr == NULL)
return (EINVAL);
lf = linker_make_file(filename, &link_elf_class);
if (lf == NULL)
return (ENOMEM);
ef = (elf_file_t)lf;
ef->preloaded = 1;
ef->address = *(caddr_t *)baseptr;
lf->address = *(caddr_t *)baseptr;
lf->size = *(size_t *)sizeptr;
if (hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
hdr->e_ident[EI_VERSION] != EV_CURRENT ||
hdr->e_version != EV_CURRENT ||
hdr->e_type != ET_REL ||
hdr->e_machine != ELF_TARG_MACH) {
error = EFTYPE;
goto out;
}
ef->e_shdr = shdr;
/* Scan the section header for information and table sizing. */
symtabindex = -1;
symstrindex = -1;
for (i = 0; i < hdr->e_shnum; i++) {
switch (shdr[i].sh_type) {
case SHT_PROGBITS:
case SHT_NOBITS:
#ifdef __amd64__
case SHT_X86_64_UNWIND:
#endif
Require the SHF_ALLOC flag for program sections from kernel object modules. ELF object files can contain program sections which are not supposed to be loaded into memory (e.g. .comment). Normally the static linker uses these flags to decide which sections are allocated to loadable program segments in ELF binaries and shared objects (including kernels on all architectures and kernel modules on architectures other than amd64). Mapping ELF object files (such as amd64 kernel modules) into memory directly is a bit of a grey area. ELF object files are intended to be used as inputs to the static linker. As a result, there is not a standardized definition for what the memory layout of an ELF object should be (none of the section headers have valid virtual memory addresses for example). The kernel and loader were not checking the SHF_ALLOC flag but loading any program sections with certain types such as SHT_PROGBITS. As a result, the kernel and loader would load into RAM some sections that weren't marked with SHF_ALLOC such as .comment that are not loaded into RAM for kernel modules on other architectures (which are implemented as ELF shared objects). Aside from possibly requiring slightly more RAM to hold a kernel module this does not affect runtime correctness as the kernel relocates symbols based on the layout it uses. Debuggers such as gdb and lldb do not extract symbol tables from a running process or kernel. Instead, they replicate the memory layout of ELF executables and shared objects and use that to construct their own symbol tables. For executables and shared objects this works fine. For ELF objects the current logic in kgdb (and probably lldb based on a simple reading) assumes that only sections with SHF_ALLOC are memory resident when constructing a memory layout. If the debugger constructs a different memory layout than the kernel, then it will compute different addresses for symbols causing symbols in the debugger to appear to have the wrong values (though the kernel itself is working fine). The current port of mdb does not check SHF_ALLOC as it replicates the kernel's logic in its existing kernel support. The bfd linker sorts the sections in ELF object files such that all of the allocated sections (sections with SHF_ALLOCATED) are placed first followed by unallocated sections. As a result, when kgdb composed a memory layout using only the allocated sections, this layout happened to match the layout used by the kernel and loader. The lld linker does not sort the sections in ELF object files and mixed allocated and unallocated sections. This resulted in kgdb composing a different memory layout than the kernel and loader. We could either patch kgdb (and possibly in the future lldb) to use custom handling when generating memory layouts for kernel modules that are ELF objects, or we could change the kernel and loader to check SHF_ALLOCATED. I chose the latter as I feel we shouldn't be loading things into RAM that the module won't use. This should mostly be a NOP when linking with bfd but will allow the existing kgdb to work with amd64 kernel modules linked with lld. Note that we only require SHF_ALLOC for "program" sections for types like SHT_PROGBITS and SHT_NOBITS. Other section types such as symbol tables, string tables, and relocations must also be loaded and are not marked with SHF_ALLOC. Reported by: np Reviewed by: kib, emaste MFC after: 1 month Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D13926
2018-01-17 22:51:59 +00:00
/* Ignore sections not loaded by the loader. */
if (shdr[i].sh_addr == 0)
break;
ef->nprogtab++;
break;
case SHT_SYMTAB:
symtabindex = i;
symstrindex = shdr[i].sh_link;
break;
case SHT_REL:
/*
* Ignore relocation tables for sections not
* loaded by the loader.
*/
if (shdr[shdr[i].sh_info].sh_addr == 0)
break;
ef->nreltab++;
break;
case SHT_RELA:
if (shdr[shdr[i].sh_info].sh_addr == 0)
break;
ef->nrelatab++;
break;
}
}
shstrindex = hdr->e_shstrndx;
if (ef->nprogtab == 0 || symstrindex < 0 ||
symstrindex >= hdr->e_shnum ||
shdr[symstrindex].sh_type != SHT_STRTAB || shstrindex == 0 ||
shstrindex >= hdr->e_shnum ||
shdr[shstrindex].sh_type != SHT_STRTAB) {
printf("%s: bad/missing section headers\n", filename);
error = ENOEXEC;
goto out;
}
/* Allocate space for tracking the load chunks */
if (ef->nprogtab != 0)
ef->progtab = malloc(ef->nprogtab * sizeof(*ef->progtab),
M_LINKER, M_WAITOK | M_ZERO);
if (ef->nreltab != 0)
ef->reltab = malloc(ef->nreltab * sizeof(*ef->reltab),
M_LINKER, M_WAITOK | M_ZERO);
if (ef->nrelatab != 0)
ef->relatab = malloc(ef->nrelatab * sizeof(*ef->relatab),
M_LINKER, M_WAITOK | M_ZERO);
if ((ef->nprogtab != 0 && ef->progtab == NULL) ||
(ef->nreltab != 0 && ef->reltab == NULL) ||
(ef->nrelatab != 0 && ef->relatab == NULL)) {
error = ENOMEM;
goto out;
}
/* XXX, relocate the sh_addr fields saved by the loader. */
off = 0;
for (i = 0; i < hdr->e_shnum; i++) {
if (shdr[i].sh_addr != 0 && (off == 0 || shdr[i].sh_addr < off))
off = shdr[i].sh_addr;
}
for (i = 0; i < hdr->e_shnum; i++) {
if (shdr[i].sh_addr != 0)
shdr[i].sh_addr = shdr[i].sh_addr - off +
(Elf_Addr)ef->address;
}
ef->ddbsymcnt = shdr[symtabindex].sh_size / sizeof(Elf_Sym);
ef->ddbsymtab = (Elf_Sym *)shdr[symtabindex].sh_addr;
ef->ddbstrcnt = shdr[symstrindex].sh_size;
ef->ddbstrtab = (char *)shdr[symstrindex].sh_addr;
ef->shstrcnt = shdr[shstrindex].sh_size;
ef->shstrtab = (char *)shdr[shstrindex].sh_addr;
/* Now fill out progtab and the relocation tables. */
pb = 0;
rl = 0;
ra = 0;
for (i = 0; i < hdr->e_shnum; i++) {
switch (shdr[i].sh_type) {
case SHT_PROGBITS:
case SHT_NOBITS:
#ifdef __amd64__
case SHT_X86_64_UNWIND:
#endif
Require the SHF_ALLOC flag for program sections from kernel object modules. ELF object files can contain program sections which are not supposed to be loaded into memory (e.g. .comment). Normally the static linker uses these flags to decide which sections are allocated to loadable program segments in ELF binaries and shared objects (including kernels on all architectures and kernel modules on architectures other than amd64). Mapping ELF object files (such as amd64 kernel modules) into memory directly is a bit of a grey area. ELF object files are intended to be used as inputs to the static linker. As a result, there is not a standardized definition for what the memory layout of an ELF object should be (none of the section headers have valid virtual memory addresses for example). The kernel and loader were not checking the SHF_ALLOC flag but loading any program sections with certain types such as SHT_PROGBITS. As a result, the kernel and loader would load into RAM some sections that weren't marked with SHF_ALLOC such as .comment that are not loaded into RAM for kernel modules on other architectures (which are implemented as ELF shared objects). Aside from possibly requiring slightly more RAM to hold a kernel module this does not affect runtime correctness as the kernel relocates symbols based on the layout it uses. Debuggers such as gdb and lldb do not extract symbol tables from a running process or kernel. Instead, they replicate the memory layout of ELF executables and shared objects and use that to construct their own symbol tables. For executables and shared objects this works fine. For ELF objects the current logic in kgdb (and probably lldb based on a simple reading) assumes that only sections with SHF_ALLOC are memory resident when constructing a memory layout. If the debugger constructs a different memory layout than the kernel, then it will compute different addresses for symbols causing symbols in the debugger to appear to have the wrong values (though the kernel itself is working fine). The current port of mdb does not check SHF_ALLOC as it replicates the kernel's logic in its existing kernel support. The bfd linker sorts the sections in ELF object files such that all of the allocated sections (sections with SHF_ALLOCATED) are placed first followed by unallocated sections. As a result, when kgdb composed a memory layout using only the allocated sections, this layout happened to match the layout used by the kernel and loader. The lld linker does not sort the sections in ELF object files and mixed allocated and unallocated sections. This resulted in kgdb composing a different memory layout than the kernel and loader. We could either patch kgdb (and possibly in the future lldb) to use custom handling when generating memory layouts for kernel modules that are ELF objects, or we could change the kernel and loader to check SHF_ALLOCATED. I chose the latter as I feel we shouldn't be loading things into RAM that the module won't use. This should mostly be a NOP when linking with bfd but will allow the existing kgdb to work with amd64 kernel modules linked with lld. Note that we only require SHF_ALLOC for "program" sections for types like SHT_PROGBITS and SHT_NOBITS. Other section types such as symbol tables, string tables, and relocations must also be loaded and are not marked with SHF_ALLOC. Reported by: np Reviewed by: kib, emaste MFC after: 1 month Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D13926
2018-01-17 22:51:59 +00:00
if (shdr[i].sh_addr == 0)
break;
ef->progtab[pb].addr = (void *)shdr[i].sh_addr;
if (shdr[i].sh_type == SHT_PROGBITS)
ef->progtab[pb].name = "<<PROGBITS>>";
#ifdef __amd64__
else if (shdr[i].sh_type == SHT_X86_64_UNWIND)
ef->progtab[pb].name = "<<UNWIND>>";
#endif
else
ef->progtab[pb].name = "<<NOBITS>>";
ef->progtab[pb].size = shdr[i].sh_size;
ef->progtab[pb].sec = i;
if (ef->shstrtab && shdr[i].sh_name != 0)
ef->progtab[pb].name =
ef->shstrtab + shdr[i].sh_name;
if (ef->progtab[pb].name != NULL &&
!strcmp(ef->progtab[pb].name, DPCPU_SETNAME)) {
void *dpcpu;
dpcpu = dpcpu_alloc(shdr[i].sh_size);
if (dpcpu == NULL) {
printf("%s: pcpu module space is out "
"of space; cannot allocate %#jx "
"for %s\n", __func__,
(uintmax_t)shdr[i].sh_size,
filename);
error = ENOSPC;
goto out;
}
memcpy(dpcpu, ef->progtab[pb].addr,
ef->progtab[pb].size);
dpcpu_copy(dpcpu, shdr[i].sh_size);
ef->progtab[pb].addr = dpcpu;
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
#ifdef VIMAGE
} else if (ef->progtab[pb].name != NULL &&
!strcmp(ef->progtab[pb].name, VNET_SETNAME)) {
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
void *vnet_data;
vnet_data = vnet_data_alloc(shdr[i].sh_size);
if (vnet_data == NULL) {
printf("%s: vnet module space is out "
"of space; cannot allocate %#jx "
"for %s\n", __func__,
(uintmax_t)shdr[i].sh_size,
filename);
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
error = ENOSPC;
goto out;
}
memcpy(vnet_data, ef->progtab[pb].addr,
ef->progtab[pb].size);
vnet_data_copy(vnet_data, shdr[i].sh_size);
ef->progtab[pb].addr = vnet_data;
#endif
} else if (ef->progtab[pb].name != NULL &&
!strcmp(ef->progtab[pb].name, ".ctors")) {
lf->ctors_addr = ef->progtab[pb].addr;
lf->ctors_size = shdr[i].sh_size;
}
/* Update all symbol values with the offset. */
for (j = 0; j < ef->ddbsymcnt; j++) {
es = &ef->ddbsymtab[j];
if (es->st_shndx != i)
continue;
es->st_value += (Elf_Addr)ef->progtab[pb].addr;
}
pb++;
break;
case SHT_REL:
if (shdr[shdr[i].sh_info].sh_addr == 0)
break;
ef->reltab[rl].rel = (Elf_Rel *)shdr[i].sh_addr;
ef->reltab[rl].nrel = shdr[i].sh_size / sizeof(Elf_Rel);
ef->reltab[rl].sec = shdr[i].sh_info;
rl++;
break;
case SHT_RELA:
if (shdr[shdr[i].sh_info].sh_addr == 0)
break;
ef->relatab[ra].rela = (Elf_Rela *)shdr[i].sh_addr;
ef->relatab[ra].nrela =
shdr[i].sh_size / sizeof(Elf_Rela);
ef->relatab[ra].sec = shdr[i].sh_info;
ra++;
break;
}
}
if (pb != ef->nprogtab) {
printf("%s: lost progbits\n", filename);
error = ENOEXEC;
goto out;
}
if (rl != ef->nreltab) {
printf("%s: lost reltab\n", filename);
error = ENOEXEC;
goto out;
}
if (ra != ef->nrelatab) {
printf("%s: lost relatab\n", filename);
error = ENOEXEC;
goto out;
}
/* Local intra-module relocations */
error = link_elf_reloc_local(lf, false);
if (error != 0)
goto out;
*result = lf;
return (0);
out:
/* preload not done this way */
linker_file_unload(lf, LINKER_UNLOAD_FORCE);
return (error);
}
static void
link_elf_invoke_ctors(caddr_t addr, size_t size)
{
void (**ctor)(void);
size_t i, cnt;
if (addr == NULL || size == 0)
return;
cnt = size / sizeof(*ctor);
ctor = (void *)addr;
for (i = 0; i < cnt; i++) {
if (ctor[i] != NULL)
(*ctor[i])();
}
}
static int
link_elf_link_preload_finish(linker_file_t lf)
{
elf_file_t ef;
int error;
ef = (elf_file_t)lf;
error = relocate_file(ef);
if (error)
return (error);
/* Notify MD code that a module is being loaded. */
error = elf_cpu_load_file(lf);
if (error)
return (error);
#if defined(__i386__) || defined(__amd64__)
/* Now ifuncs. */
error = link_elf_reloc_local(lf, true);
if (error != 0)
return (error);
#endif
/* Invoke .ctors */
link_elf_invoke_ctors(lf->ctors_addr, lf->ctors_size);
return (0);
}
static int
link_elf_load_file(linker_class_t cls, const char *filename,
linker_file_t *result)
{
struct nameidata *nd;
struct thread *td = curthread; /* XXX */
Elf_Ehdr *hdr;
Elf_Shdr *shdr;
Elf_Sym *es;
int nbytes, i, j;
vm_offset_t mapbase;
size_t mapsize;
int error = 0;
ssize_t resid;
int flags;
elf_file_t ef;
linker_file_t lf;
int symtabindex;
int symstrindex;
int shstrindex;
int nsym;
int pb, rl, ra;
int alignmask;
shdr = NULL;
lf = NULL;
mapsize = 0;
hdr = NULL;
nd = malloc(sizeof(struct nameidata), M_TEMP, M_WAITOK);
NDINIT(nd, LOOKUP, FOLLOW, UIO_SYSSPACE, filename, td);
flags = FREAD;
error = vn_open(nd, &flags, 0, NULL);
if (error) {
free(nd, M_TEMP);
return error;
}
NDFREE(nd, NDF_ONLY_PNBUF);
if (nd->ni_vp->v_type != VREG) {
error = ENOEXEC;
goto out;
}
#ifdef MAC
error = mac_kld_check_load(td->td_ucred, nd->ni_vp);
if (error) {
goto out;
}
#endif
1998-09-11 08:46:15 +00:00
/* Read the elf header from the file. */
hdr = malloc(sizeof(*hdr), M_LINKER, M_WAITOK);
error = vn_rdwr(UIO_READ, nd->ni_vp, (void *)hdr, sizeof(*hdr), 0,
UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
&resid, td);
if (error)
goto out;
if (resid != 0){
error = ENOEXEC;
goto out;
}
if (!IS_ELF(*hdr)) {
error = ENOEXEC;
goto out;
}
if (hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS
|| hdr->e_ident[EI_DATA] != ELF_TARG_DATA) {
link_elf_error(filename, "Unsupported file layout");
error = ENOEXEC;
goto out;
}
if (hdr->e_ident[EI_VERSION] != EV_CURRENT
|| hdr->e_version != EV_CURRENT) {
link_elf_error(filename, "Unsupported file version");
error = ENOEXEC;
goto out;
}
if (hdr->e_type != ET_REL) {
error = ENOSYS;
goto out;
}
if (hdr->e_machine != ELF_TARG_MACH) {
link_elf_error(filename, "Unsupported machine");
error = ENOEXEC;
goto out;
}
First round implementation of a fine grain enhanced module to module version dependency system. This isn't quite finished, but it is at a useful stage to do a functional checkpoint. Highlights: - version and dependency metadata is gathered via linker sets, so things are handled the same for static kernels and code built to live in a kld. - The dependencies are at module level (versus at file level). - Dependencies determine kld symbol search order - this means that you cannot link against symbols in another file unless you depend on it. This is so that you cannot accidently unload the target out from underneath the ones referencing it. - It is flexible enough that we can put tags in #include files and macros so that we can get decent hooks for enforcing recompiles on incompatable ABI changes. eg: if we change struct proc, we could force a recompile for all kld's that reference the proc struct. - Tangled dependency references at boot time are sorted. Files are relocated once all their dependencies are already relocated. Caveats: - Loader support is incomplete, but has been worked on seperately. - Actual enforcement of the version number tags is not active yet - just the module dependencies are live. The actual structure of versioning hasn't been agreed on yet. (eg: major.minor, or whatever) - There is some backwards compatability for old modules without metadata but I'm not sure how good it is. This is based on work originally done by Boris Popov (bp@freebsd.org), but I'm not sure he'd recognize much of it now. Don't blame him. :-) Also, ideas have been borrowed from Mike Smith.
2000-04-29 13:19:31 +00:00
lf = linker_make_file(filename, &link_elf_class);
if (!lf) {
error = ENOMEM;
goto out;
}
ef = (elf_file_t) lf;
ef->nprogtab = 0;
ef->e_shdr = 0;
ef->nreltab = 0;
ef->nrelatab = 0;
/* Allocate and read in the section header */
nbytes = hdr->e_shnum * hdr->e_shentsize;
if (nbytes == 0 || hdr->e_shoff == 0 ||
hdr->e_shentsize != sizeof(Elf_Shdr)) {
error = ENOEXEC;
goto out;
}
shdr = malloc(nbytes, M_LINKER, M_WAITOK);
ef->e_shdr = shdr;
error = vn_rdwr(UIO_READ, nd->ni_vp, (caddr_t)shdr, nbytes,
hdr->e_shoff, UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
NOCRED, &resid, td);
First round implementation of a fine grain enhanced module to module version dependency system. This isn't quite finished, but it is at a useful stage to do a functional checkpoint. Highlights: - version and dependency metadata is gathered via linker sets, so things are handled the same for static kernels and code built to live in a kld. - The dependencies are at module level (versus at file level). - Dependencies determine kld symbol search order - this means that you cannot link against symbols in another file unless you depend on it. This is so that you cannot accidently unload the target out from underneath the ones referencing it. - It is flexible enough that we can put tags in #include files and macros so that we can get decent hooks for enforcing recompiles on incompatable ABI changes. eg: if we change struct proc, we could force a recompile for all kld's that reference the proc struct. - Tangled dependency references at boot time are sorted. Files are relocated once all their dependencies are already relocated. Caveats: - Loader support is incomplete, but has been worked on seperately. - Actual enforcement of the version number tags is not active yet - just the module dependencies are live. The actual structure of versioning hasn't been agreed on yet. (eg: major.minor, or whatever) - There is some backwards compatability for old modules without metadata but I'm not sure how good it is. This is based on work originally done by Boris Popov (bp@freebsd.org), but I'm not sure he'd recognize much of it now. Don't blame him. :-) Also, ideas have been borrowed from Mike Smith.
2000-04-29 13:19:31 +00:00
if (error)
goto out;
if (resid) {
error = ENOEXEC;
goto out;
}
/* Scan the section header for information and table sizing. */
nsym = 0;
symtabindex = -1;
symstrindex = -1;
for (i = 0; i < hdr->e_shnum; i++) {
if (shdr[i].sh_size == 0)
continue;
switch (shdr[i].sh_type) {
case SHT_PROGBITS:
case SHT_NOBITS:
#ifdef __amd64__
case SHT_X86_64_UNWIND:
#endif
Require the SHF_ALLOC flag for program sections from kernel object modules. ELF object files can contain program sections which are not supposed to be loaded into memory (e.g. .comment). Normally the static linker uses these flags to decide which sections are allocated to loadable program segments in ELF binaries and shared objects (including kernels on all architectures and kernel modules on architectures other than amd64). Mapping ELF object files (such as amd64 kernel modules) into memory directly is a bit of a grey area. ELF object files are intended to be used as inputs to the static linker. As a result, there is not a standardized definition for what the memory layout of an ELF object should be (none of the section headers have valid virtual memory addresses for example). The kernel and loader were not checking the SHF_ALLOC flag but loading any program sections with certain types such as SHT_PROGBITS. As a result, the kernel and loader would load into RAM some sections that weren't marked with SHF_ALLOC such as .comment that are not loaded into RAM for kernel modules on other architectures (which are implemented as ELF shared objects). Aside from possibly requiring slightly more RAM to hold a kernel module this does not affect runtime correctness as the kernel relocates symbols based on the layout it uses. Debuggers such as gdb and lldb do not extract symbol tables from a running process or kernel. Instead, they replicate the memory layout of ELF executables and shared objects and use that to construct their own symbol tables. For executables and shared objects this works fine. For ELF objects the current logic in kgdb (and probably lldb based on a simple reading) assumes that only sections with SHF_ALLOC are memory resident when constructing a memory layout. If the debugger constructs a different memory layout than the kernel, then it will compute different addresses for symbols causing symbols in the debugger to appear to have the wrong values (though the kernel itself is working fine). The current port of mdb does not check SHF_ALLOC as it replicates the kernel's logic in its existing kernel support. The bfd linker sorts the sections in ELF object files such that all of the allocated sections (sections with SHF_ALLOCATED) are placed first followed by unallocated sections. As a result, when kgdb composed a memory layout using only the allocated sections, this layout happened to match the layout used by the kernel and loader. The lld linker does not sort the sections in ELF object files and mixed allocated and unallocated sections. This resulted in kgdb composing a different memory layout than the kernel and loader. We could either patch kgdb (and possibly in the future lldb) to use custom handling when generating memory layouts for kernel modules that are ELF objects, or we could change the kernel and loader to check SHF_ALLOCATED. I chose the latter as I feel we shouldn't be loading things into RAM that the module won't use. This should mostly be a NOP when linking with bfd but will allow the existing kgdb to work with amd64 kernel modules linked with lld. Note that we only require SHF_ALLOC for "program" sections for types like SHT_PROGBITS and SHT_NOBITS. Other section types such as symbol tables, string tables, and relocations must also be loaded and are not marked with SHF_ALLOC. Reported by: np Reviewed by: kib, emaste MFC after: 1 month Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D13926
2018-01-17 22:51:59 +00:00
if ((shdr[i].sh_flags & SHF_ALLOC) == 0)
break;
ef->nprogtab++;
break;
case SHT_SYMTAB:
nsym++;
symtabindex = i;
symstrindex = shdr[i].sh_link;
break;
case SHT_REL:
/*
* Ignore relocation tables for unallocated
* sections.
*/
if ((shdr[shdr[i].sh_info].sh_flags & SHF_ALLOC) == 0)
break;
ef->nreltab++;
break;
case SHT_RELA:
if ((shdr[shdr[i].sh_info].sh_flags & SHF_ALLOC) == 0)
break;
ef->nrelatab++;
break;
case SHT_STRTAB:
break;
}
}
if (ef->nprogtab == 0) {
link_elf_error(filename, "file has no contents");
error = ENOEXEC;
goto out;
}
if (nsym != 1) {
/* Only allow one symbol table for now */
link_elf_error(filename,
"file must have exactly one symbol table");
error = ENOEXEC;
goto out;
}
if (symstrindex < 0 || symstrindex > hdr->e_shnum ||
shdr[symstrindex].sh_type != SHT_STRTAB) {
link_elf_error(filename, "file has invalid symbol strings");
error = ENOEXEC;
goto out;
}
/* Allocate space for tracking the load chunks */
if (ef->nprogtab != 0)
ef->progtab = malloc(ef->nprogtab * sizeof(*ef->progtab),
M_LINKER, M_WAITOK | M_ZERO);
if (ef->nreltab != 0)
ef->reltab = malloc(ef->nreltab * sizeof(*ef->reltab),
M_LINKER, M_WAITOK | M_ZERO);
if (ef->nrelatab != 0)
ef->relatab = malloc(ef->nrelatab * sizeof(*ef->relatab),
M_LINKER, M_WAITOK | M_ZERO);
if (symtabindex == -1) {
link_elf_error(filename, "lost symbol table index");
error = ENOEXEC;
goto out;
}
/* Allocate space for and load the symbol table */
ef->ddbsymcnt = shdr[symtabindex].sh_size / sizeof(Elf_Sym);
ef->ddbsymtab = malloc(shdr[symtabindex].sh_size, M_LINKER, M_WAITOK);
error = vn_rdwr(UIO_READ, nd->ni_vp, (void *)ef->ddbsymtab,
shdr[symtabindex].sh_size, shdr[symtabindex].sh_offset,
UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
&resid, td);
if (error)
goto out;
if (resid != 0){
error = EINVAL;
goto out;
}
if (symstrindex == -1) {
link_elf_error(filename, "lost symbol string index");
error = ENOEXEC;
goto out;
}
/* Allocate space for and load the symbol strings */
ef->ddbstrcnt = shdr[symstrindex].sh_size;
ef->ddbstrtab = malloc(shdr[symstrindex].sh_size, M_LINKER, M_WAITOK);
error = vn_rdwr(UIO_READ, nd->ni_vp, ef->ddbstrtab,
shdr[symstrindex].sh_size, shdr[symstrindex].sh_offset,
UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
&resid, td);
if (error)
goto out;
if (resid != 0){
error = EINVAL;
goto out;
}
/* Do we have a string table for the section names? */
shstrindex = -1;
if (hdr->e_shstrndx != 0 &&
shdr[hdr->e_shstrndx].sh_type == SHT_STRTAB) {
shstrindex = hdr->e_shstrndx;
ef->shstrcnt = shdr[shstrindex].sh_size;
ef->shstrtab = malloc(shdr[shstrindex].sh_size, M_LINKER,
M_WAITOK);
error = vn_rdwr(UIO_READ, nd->ni_vp, ef->shstrtab,
shdr[shstrindex].sh_size, shdr[shstrindex].sh_offset,
In order to better support flexible and extensible access control, make a series of modifications to the credential arguments relating to file read and write operations to cliarfy which credential is used for what: - Change fo_read() and fo_write() to accept "active_cred" instead of "cred", and change the semantics of consumers of fo_read() and fo_write() to pass the active credential of the thread requesting an operation rather than the cached file cred. The cached file cred is still available in fo_read() and fo_write() consumers via fp->f_cred. These changes largely in sys_generic.c. For each implementation of fo_read() and fo_write(), update cred usage to reflect this change and maintain current semantics: - badfo_readwrite() unchanged - kqueue_read/write() unchanged pipe_read/write() now authorize MAC using active_cred rather than td->td_ucred - soo_read/write() unchanged - vn_read/write() now authorize MAC using active_cred but VOP_READ/WRITE() with fp->f_cred Modify vn_rdwr() to accept two credential arguments instead of a single credential: active_cred and file_cred. Use active_cred for MAC authorization, and select a credential for use in VOP_READ/WRITE() based on whether file_cred is NULL or not. If file_cred is provided, authorize the VOP using that cred, otherwise the active credential, matching current semantics. Modify current vn_rdwr() consumers to pass a file_cred if used in the context of a struct file, and to always pass active_cred. When vn_rdwr() is used without a file_cred, pass NOCRED. These changes should maintain current semantics for read/write, but avoid a redundant passing of fp->f_cred, as well as making it more clear what the origin of each credential is in file descriptor read/write operations. Follow-up commits will make similar changes to other file descriptor operations, and modify the MAC framework to pass both credentials to MAC policy modules so they can implement either semantic for revocation. Obtained from: TrustedBSD Project Sponsored by: DARPA, NAI Labs
2002-08-15 20:55:08 +00:00
UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
&resid, td);
if (error)
goto out;
if (resid != 0){
error = EINVAL;
goto out;
}
1998-09-11 08:46:15 +00:00
}
/* Size up code/data(progbits) and bss(nobits). */
alignmask = 0;
for (i = 0; i < hdr->e_shnum; i++) {
if (shdr[i].sh_size == 0)
continue;
switch (shdr[i].sh_type) {
case SHT_PROGBITS:
case SHT_NOBITS:
#ifdef __amd64__
case SHT_X86_64_UNWIND:
#endif
Require the SHF_ALLOC flag for program sections from kernel object modules. ELF object files can contain program sections which are not supposed to be loaded into memory (e.g. .comment). Normally the static linker uses these flags to decide which sections are allocated to loadable program segments in ELF binaries and shared objects (including kernels on all architectures and kernel modules on architectures other than amd64). Mapping ELF object files (such as amd64 kernel modules) into memory directly is a bit of a grey area. ELF object files are intended to be used as inputs to the static linker. As a result, there is not a standardized definition for what the memory layout of an ELF object should be (none of the section headers have valid virtual memory addresses for example). The kernel and loader were not checking the SHF_ALLOC flag but loading any program sections with certain types such as SHT_PROGBITS. As a result, the kernel and loader would load into RAM some sections that weren't marked with SHF_ALLOC such as .comment that are not loaded into RAM for kernel modules on other architectures (which are implemented as ELF shared objects). Aside from possibly requiring slightly more RAM to hold a kernel module this does not affect runtime correctness as the kernel relocates symbols based on the layout it uses. Debuggers such as gdb and lldb do not extract symbol tables from a running process or kernel. Instead, they replicate the memory layout of ELF executables and shared objects and use that to construct their own symbol tables. For executables and shared objects this works fine. For ELF objects the current logic in kgdb (and probably lldb based on a simple reading) assumes that only sections with SHF_ALLOC are memory resident when constructing a memory layout. If the debugger constructs a different memory layout than the kernel, then it will compute different addresses for symbols causing symbols in the debugger to appear to have the wrong values (though the kernel itself is working fine). The current port of mdb does not check SHF_ALLOC as it replicates the kernel's logic in its existing kernel support. The bfd linker sorts the sections in ELF object files such that all of the allocated sections (sections with SHF_ALLOCATED) are placed first followed by unallocated sections. As a result, when kgdb composed a memory layout using only the allocated sections, this layout happened to match the layout used by the kernel and loader. The lld linker does not sort the sections in ELF object files and mixed allocated and unallocated sections. This resulted in kgdb composing a different memory layout than the kernel and loader. We could either patch kgdb (and possibly in the future lldb) to use custom handling when generating memory layouts for kernel modules that are ELF objects, or we could change the kernel and loader to check SHF_ALLOCATED. I chose the latter as I feel we shouldn't be loading things into RAM that the module won't use. This should mostly be a NOP when linking with bfd but will allow the existing kgdb to work with amd64 kernel modules linked with lld. Note that we only require SHF_ALLOC for "program" sections for types like SHT_PROGBITS and SHT_NOBITS. Other section types such as symbol tables, string tables, and relocations must also be loaded and are not marked with SHF_ALLOC. Reported by: np Reviewed by: kib, emaste MFC after: 1 month Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D13926
2018-01-17 22:51:59 +00:00
if ((shdr[i].sh_flags & SHF_ALLOC) == 0)
break;
alignmask = shdr[i].sh_addralign - 1;
mapsize += alignmask;
mapsize &= ~alignmask;
mapsize += shdr[i].sh_size;
break;
}
1998-09-11 08:46:15 +00:00
}
/*
* We know how much space we need for the text/data/bss/etc.
* This stuff needs to be in a single chunk so that profiling etc
* can get the bounds and gdb can associate offsets with modules
1998-09-11 08:46:15 +00:00
*/
ef->object = vm_object_allocate(OBJT_DEFAULT,
round_page(mapsize) >> PAGE_SHIFT);
if (ef->object == NULL) {
error = ENOMEM;
goto out;
}
ef->address = (caddr_t) vm_map_min(kernel_map);
/*
* In order to satisfy amd64's architectural requirements on the
* location of code and data in the kernel's address space, request a
* mapping that is above the kernel.
*/
#ifdef __amd64__
mapbase = KERNBASE;
#else
mapbase = VM_MIN_KERNEL_ADDRESS;
#endif
error = vm_map_find(kernel_map, ef->object, 0, &mapbase,
round_page(mapsize), 0, VMFS_OPTIMAL_SPACE, VM_PROT_ALL,
VM_PROT_ALL, 0);
if (error) {
vm_object_deallocate(ef->object);
ef->object = 0;
goto out;
}
/* Wire the pages */
2005-08-28 05:38:40 +00:00
error = vm_map_wire(kernel_map, mapbase,
mapbase + round_page(mapsize),
VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
2005-08-28 05:38:40 +00:00
if (error != KERN_SUCCESS) {
error = ENOMEM;
goto out;
}
/* Inform the kld system about the situation */
lf->address = ef->address = (caddr_t)mapbase;
lf->size = mapsize;
/*
* Now load code/data(progbits), zero bss(nobits), allocate space for
* and load relocs
*/
pb = 0;
rl = 0;
ra = 0;
alignmask = 0;
for (i = 0; i < hdr->e_shnum; i++) {
if (shdr[i].sh_size == 0)
continue;
switch (shdr[i].sh_type) {
case SHT_PROGBITS:
case SHT_NOBITS:
#ifdef __amd64__
case SHT_X86_64_UNWIND:
#endif
Require the SHF_ALLOC flag for program sections from kernel object modules. ELF object files can contain program sections which are not supposed to be loaded into memory (e.g. .comment). Normally the static linker uses these flags to decide which sections are allocated to loadable program segments in ELF binaries and shared objects (including kernels on all architectures and kernel modules on architectures other than amd64). Mapping ELF object files (such as amd64 kernel modules) into memory directly is a bit of a grey area. ELF object files are intended to be used as inputs to the static linker. As a result, there is not a standardized definition for what the memory layout of an ELF object should be (none of the section headers have valid virtual memory addresses for example). The kernel and loader were not checking the SHF_ALLOC flag but loading any program sections with certain types such as SHT_PROGBITS. As a result, the kernel and loader would load into RAM some sections that weren't marked with SHF_ALLOC such as .comment that are not loaded into RAM for kernel modules on other architectures (which are implemented as ELF shared objects). Aside from possibly requiring slightly more RAM to hold a kernel module this does not affect runtime correctness as the kernel relocates symbols based on the layout it uses. Debuggers such as gdb and lldb do not extract symbol tables from a running process or kernel. Instead, they replicate the memory layout of ELF executables and shared objects and use that to construct their own symbol tables. For executables and shared objects this works fine. For ELF objects the current logic in kgdb (and probably lldb based on a simple reading) assumes that only sections with SHF_ALLOC are memory resident when constructing a memory layout. If the debugger constructs a different memory layout than the kernel, then it will compute different addresses for symbols causing symbols in the debugger to appear to have the wrong values (though the kernel itself is working fine). The current port of mdb does not check SHF_ALLOC as it replicates the kernel's logic in its existing kernel support. The bfd linker sorts the sections in ELF object files such that all of the allocated sections (sections with SHF_ALLOCATED) are placed first followed by unallocated sections. As a result, when kgdb composed a memory layout using only the allocated sections, this layout happened to match the layout used by the kernel and loader. The lld linker does not sort the sections in ELF object files and mixed allocated and unallocated sections. This resulted in kgdb composing a different memory layout than the kernel and loader. We could either patch kgdb (and possibly in the future lldb) to use custom handling when generating memory layouts for kernel modules that are ELF objects, or we could change the kernel and loader to check SHF_ALLOCATED. I chose the latter as I feel we shouldn't be loading things into RAM that the module won't use. This should mostly be a NOP when linking with bfd but will allow the existing kgdb to work with amd64 kernel modules linked with lld. Note that we only require SHF_ALLOC for "program" sections for types like SHT_PROGBITS and SHT_NOBITS. Other section types such as symbol tables, string tables, and relocations must also be loaded and are not marked with SHF_ALLOC. Reported by: np Reviewed by: kib, emaste MFC after: 1 month Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D13926
2018-01-17 22:51:59 +00:00
if ((shdr[i].sh_flags & SHF_ALLOC) == 0)
break;
alignmask = shdr[i].sh_addralign - 1;
mapbase += alignmask;
mapbase &= ~alignmask;
if (ef->shstrtab != NULL && shdr[i].sh_name != 0) {
ef->progtab[pb].name =
ef->shstrtab + shdr[i].sh_name;
if (!strcmp(ef->progtab[pb].name, ".ctors")) {
lf->ctors_addr = (caddr_t)mapbase;
lf->ctors_size = shdr[i].sh_size;
}
} else if (shdr[i].sh_type == SHT_PROGBITS)
ef->progtab[pb].name = "<<PROGBITS>>";
#ifdef __amd64__
else if (shdr[i].sh_type == SHT_X86_64_UNWIND)
ef->progtab[pb].name = "<<UNWIND>>";
#endif
else
ef->progtab[pb].name = "<<NOBITS>>";
if (ef->progtab[pb].name != NULL &&
!strcmp(ef->progtab[pb].name, DPCPU_SETNAME)) {
ef->progtab[pb].addr =
dpcpu_alloc(shdr[i].sh_size);
if (ef->progtab[pb].addr == NULL) {
printf("%s: pcpu module space is out "
"of space; cannot allocate %#jx "
"for %s\n", __func__,
(uintmax_t)shdr[i].sh_size,
filename);
}
}
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
#ifdef VIMAGE
else if (ef->progtab[pb].name != NULL &&
!strcmp(ef->progtab[pb].name, VNET_SETNAME)) {
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
ef->progtab[pb].addr =
vnet_data_alloc(shdr[i].sh_size);
if (ef->progtab[pb].addr == NULL) {
printf("%s: vnet module space is out "
"of space; cannot allocate %#jx "
"for %s\n", __func__,
(uintmax_t)shdr[i].sh_size,
filename);
}
}
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
#endif
else
ef->progtab[pb].addr =
(void *)(uintptr_t)mapbase;
if (ef->progtab[pb].addr == NULL) {
error = ENOSPC;
goto out;
}
ef->progtab[pb].size = shdr[i].sh_size;
ef->progtab[pb].sec = i;
if (shdr[i].sh_type == SHT_PROGBITS
#ifdef __amd64__
|| shdr[i].sh_type == SHT_X86_64_UNWIND
#endif
) {
error = vn_rdwr(UIO_READ, nd->ni_vp,
ef->progtab[pb].addr,
shdr[i].sh_size, shdr[i].sh_offset,
UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
NOCRED, &resid, td);
if (error)
goto out;
if (resid != 0){
error = EINVAL;
goto out;
}
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
/* Initialize the per-cpu or vnet area. */
if (ef->progtab[pb].addr != (void *)mapbase &&
!strcmp(ef->progtab[pb].name, DPCPU_SETNAME))
dpcpu_copy(ef->progtab[pb].addr,
shdr[i].sh_size);
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
#ifdef VIMAGE
else if (ef->progtab[pb].addr !=
(void *)mapbase &&
!strcmp(ef->progtab[pb].name, VNET_SETNAME))
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
vnet_data_copy(ef->progtab[pb].addr,
shdr[i].sh_size);
#endif
} else
bzero(ef->progtab[pb].addr, shdr[i].sh_size);
/* Update all symbol values with the offset. */
for (j = 0; j < ef->ddbsymcnt; j++) {
es = &ef->ddbsymtab[j];
if (es->st_shndx != i)
continue;
es->st_value += (Elf_Addr)ef->progtab[pb].addr;
}
mapbase += shdr[i].sh_size;
pb++;
break;
case SHT_REL:
if ((shdr[shdr[i].sh_info].sh_flags & SHF_ALLOC) == 0)
break;
ef->reltab[rl].rel = malloc(shdr[i].sh_size, M_LINKER,
M_WAITOK);
ef->reltab[rl].nrel = shdr[i].sh_size / sizeof(Elf_Rel);
ef->reltab[rl].sec = shdr[i].sh_info;
error = vn_rdwr(UIO_READ, nd->ni_vp,
(void *)ef->reltab[rl].rel,
shdr[i].sh_size, shdr[i].sh_offset,
UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
&resid, td);
if (error)
goto out;
if (resid != 0){
error = EINVAL;
goto out;
}
rl++;
break;
case SHT_RELA:
if ((shdr[shdr[i].sh_info].sh_flags & SHF_ALLOC) == 0)
break;
ef->relatab[ra].rela = malloc(shdr[i].sh_size, M_LINKER,
M_WAITOK);
ef->relatab[ra].nrela =
shdr[i].sh_size / sizeof(Elf_Rela);
ef->relatab[ra].sec = shdr[i].sh_info;
error = vn_rdwr(UIO_READ, nd->ni_vp,
(void *)ef->relatab[ra].rela,
shdr[i].sh_size, shdr[i].sh_offset,
UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
&resid, td);
if (error)
goto out;
if (resid != 0){
error = EINVAL;
goto out;
}
ra++;
break;
}
}
if (pb != ef->nprogtab) {
link_elf_error(filename, "lost progbits");
error = ENOEXEC;
goto out;
}
if (rl != ef->nreltab) {
link_elf_error(filename, "lost reltab");
error = ENOEXEC;
goto out;
}
if (ra != ef->nrelatab) {
link_elf_error(filename, "lost relatab");
error = ENOEXEC;
goto out;
}
if (mapbase != (vm_offset_t)ef->address + mapsize) {
printf(
"%s: mapbase 0x%lx != address %p + mapsize 0x%lx (0x%lx)\n",
filename != NULL ? filename : "<none>",
(u_long)mapbase, ef->address, (u_long)mapsize,
(u_long)(vm_offset_t)ef->address + mapsize);
error = ENOMEM;
goto out;
}
/* Local intra-module relocations */
error = link_elf_reloc_local(lf, false);
if (error != 0)
goto out;
/* Pull in dependencies */
VOP_UNLOCK(nd->ni_vp, 0);
error = linker_load_dependencies(lf);
vn_lock(nd->ni_vp, LK_EXCLUSIVE | LK_RETRY);
if (error)
goto out;
/* External relocations */
error = relocate_file(ef);
if (error)
goto out;
/* Notify MD code that a module is being loaded. */
error = elf_cpu_load_file(lf);
if (error)
goto out;
#if defined(__i386__) || defined(__amd64__)
/* Now ifuncs. */
error = link_elf_reloc_local(lf, true);
if (error != 0)
goto out;
#endif
/* Invoke .ctors */
link_elf_invoke_ctors(lf->ctors_addr, lf->ctors_size);
*result = lf;
out:
VOP_UNLOCK(nd->ni_vp, 0);
vn_close(nd->ni_vp, FREAD, td->td_ucred, td);
free(nd, M_TEMP);
if (error && lf)
linker_file_unload(lf, LINKER_UNLOAD_FORCE);
free(hdr, M_LINKER);
return error;
}
static void
link_elf_unload_file(linker_file_t file)
{
elf_file_t ef = (elf_file_t) file;
u_int i;
/* Notify MD code that a module is being unloaded. */
elf_cpu_unload_file(file);
if (ef->progtab) {
for (i = 0; i < ef->nprogtab; i++) {
if (ef->progtab[i].size == 0)
continue;
if (ef->progtab[i].name == NULL)
continue;
if (!strcmp(ef->progtab[i].name, DPCPU_SETNAME))
dpcpu_free(ef->progtab[i].addr,
ef->progtab[i].size);
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
#ifdef VIMAGE
else if (!strcmp(ef->progtab[i].name, VNET_SETNAME))
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
vnet_data_free(ef->progtab[i].addr,
ef->progtab[i].size);
#endif
}
}
if (ef->preloaded) {
free(ef->reltab, M_LINKER);
free(ef->relatab, M_LINKER);
free(ef->progtab, M_LINKER);
free(ef->ctftab, M_LINKER);
free(ef->ctfoff, M_LINKER);
free(ef->typoff, M_LINKER);
if (file->pathname != NULL)
preload_delete_name(file->pathname);
return;
}
for (i = 0; i < ef->nreltab; i++)
free(ef->reltab[i].rel, M_LINKER);
for (i = 0; i < ef->nrelatab; i++)
free(ef->relatab[i].rela, M_LINKER);
free(ef->reltab, M_LINKER);
free(ef->relatab, M_LINKER);
free(ef->progtab, M_LINKER);
if (ef->object) {
vm_map_remove(kernel_map, (vm_offset_t) ef->address,
(vm_offset_t) ef->address +
(ef->object->size << PAGE_SHIFT));
}
free(ef->e_shdr, M_LINKER);
free(ef->ddbsymtab, M_LINKER);
free(ef->ddbstrtab, M_LINKER);
free(ef->shstrtab, M_LINKER);
free(ef->ctftab, M_LINKER);
free(ef->ctfoff, M_LINKER);
free(ef->typoff, M_LINKER);
}
1998-09-11 08:46:15 +00:00
static const char *
symbol_name(elf_file_t ef, Elf_Size r_info)
{
const Elf_Sym *ref;
if (ELF_R_SYM(r_info)) {
ref = ef->ddbsymtab + ELF_R_SYM(r_info);
return ef->ddbstrtab + ref->st_name;
} else
return NULL;
}
static Elf_Addr
findbase(elf_file_t ef, int sec)
{
int i;
Elf_Addr base = 0;
for (i = 0; i < ef->nprogtab; i++) {
if (sec == ef->progtab[i].sec) {
base = (Elf_Addr)ef->progtab[i].addr;
break;
}
}
return base;
}
static int
relocate_file(elf_file_t ef)
{
const Elf_Rel *rellim;
const Elf_Rel *rel;
const Elf_Rela *relalim;
const Elf_Rela *rela;
const char *symname;
const Elf_Sym *sym;
int i;
Elf_Size symidx;
Elf_Addr base;
/* Perform relocations without addend if there are any: */
for (i = 0; i < ef->nreltab; i++) {
rel = ef->reltab[i].rel;
if (rel == NULL) {
link_elf_error(ef->lf.filename, "lost a reltab!");
return (ENOEXEC);
}
rellim = rel + ef->reltab[i].nrel;
base = findbase(ef, ef->reltab[i].sec);
if (base == 0) {
link_elf_error(ef->lf.filename, "lost base for reltab");
return (ENOEXEC);
}
for ( ; rel < rellim; rel++) {
symidx = ELF_R_SYM(rel->r_info);
if (symidx >= ef->ddbsymcnt)
continue;
sym = ef->ddbsymtab + symidx;
/* Local relocs are already done */
if (ELF_ST_BIND(sym->st_info) == STB_LOCAL)
continue;
if (elf_reloc(&ef->lf, base, rel, ELF_RELOC_REL,
elf_obj_lookup)) {
symname = symbol_name(ef, rel->r_info);
printf("link_elf_obj: symbol %s undefined\n",
symname);
return (ENOENT);
}
}
1998-09-11 08:46:15 +00:00
}
/* Perform relocations with addend if there are any: */
for (i = 0; i < ef->nrelatab; i++) {
rela = ef->relatab[i].rela;
if (rela == NULL) {
link_elf_error(ef->lf.filename, "lost a relatab!");
return (ENOEXEC);
}
relalim = rela + ef->relatab[i].nrela;
base = findbase(ef, ef->relatab[i].sec);
if (base == 0) {
link_elf_error(ef->lf.filename,
"lost base for relatab");
return (ENOEXEC);
}
for ( ; rela < relalim; rela++) {
symidx = ELF_R_SYM(rela->r_info);
if (symidx >= ef->ddbsymcnt)
continue;
sym = ef->ddbsymtab + symidx;
/* Local relocs are already done */
if (ELF_ST_BIND(sym->st_info) == STB_LOCAL)
continue;
if (elf_reloc(&ef->lf, base, rela, ELF_RELOC_RELA,
elf_obj_lookup)) {
symname = symbol_name(ef, rela->r_info);
printf("link_elf_obj: symbol %s undefined\n",
symname);
return (ENOENT);
}
}
1998-09-11 08:46:15 +00:00
}
/*
* Only clean SHN_FBSD_CACHED for successful return. If we
* modified symbol table for the object but found an
* unresolved symbol, there is no reason to roll back.
*/
elf_obj_cleanup_globals_cache(ef);
return (0);
}
1998-09-11 08:46:15 +00:00
static int
link_elf_lookup_symbol(linker_file_t lf, const char *name, c_linker_sym_t *sym)
{
elf_file_t ef = (elf_file_t) lf;
const Elf_Sym *symp;
const char *strp;
int i;
for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
strp = ef->ddbstrtab + symp->st_name;
if (symp->st_shndx != SHN_UNDEF && strcmp(name, strp) == 0) {
*sym = (c_linker_sym_t) symp;
return 0;
}
}
return ENOENT;
}
static int
link_elf_symbol_values(linker_file_t lf, c_linker_sym_t sym,
linker_symval_t *symval)
{
elf_file_t ef;
const Elf_Sym *es;
caddr_t val;
ef = (elf_file_t) lf;
es = (const Elf_Sym*) sym;
val = (caddr_t)es->st_value;
if (es >= ef->ddbsymtab && es < (ef->ddbsymtab + ef->ddbsymcnt)) {
symval->name = ef->ddbstrtab + es->st_name;
val = (caddr_t)es->st_value;
if (ELF_ST_TYPE(es->st_info) == STT_GNU_IFUNC)
val = ((caddr_t (*)(void))val)();
symval->value = val;
symval->size = es->st_size;
return 0;
}
return ENOENT;
}
static int
link_elf_search_symbol(linker_file_t lf, caddr_t value,
c_linker_sym_t *sym, long *diffp)
{
elf_file_t ef = (elf_file_t) lf;
u_long off = (uintptr_t) (void *) value;
u_long diff = off;
u_long st_value;
const Elf_Sym *es;
const Elf_Sym *best = NULL;
int i;
for (i = 0, es = ef->ddbsymtab; i < ef->ddbsymcnt; i++, es++) {
if (es->st_name == 0)
continue;
st_value = es->st_value;
if (off >= st_value) {
if (off - st_value < diff) {
diff = off - st_value;
best = es;
if (diff == 0)
break;
} else if (off - st_value == diff) {
best = es;
}
}
}
if (best == NULL)
*diffp = off;
else
*diffp = diff;
*sym = (c_linker_sym_t) best;
return 0;
}
/*
* Look up a linker set on an ELF system.
*/
static int
link_elf_lookup_set(linker_file_t lf, const char *name,
void ***startp, void ***stopp, int *countp)
{
elf_file_t ef = (elf_file_t)lf;
void **start, **stop;
int i, count;
/* Relative to section number */
for (i = 0; i < ef->nprogtab; i++) {
if ((strncmp(ef->progtab[i].name, "set_", 4) == 0) &&
strcmp(ef->progtab[i].name + 4, name) == 0) {
start = (void **)ef->progtab[i].addr;
stop = (void **)((char *)ef->progtab[i].addr +
ef->progtab[i].size);
count = stop - start;
if (startp)
*startp = start;
if (stopp)
*stopp = stop;
if (countp)
*countp = count;
return (0);
}
}
return (ESRCH);
}
static int
link_elf_each_function_name(linker_file_t file,
int (*callback)(const char *, void *), void *opaque)
{
elf_file_t ef = (elf_file_t)file;
const Elf_Sym *symp;
int i, error;
/* Exhaustive search */
for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
if (symp->st_value != 0 &&
(ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC)) {
error = callback(ef->ddbstrtab + symp->st_name, opaque);
if (error)
return (error);
}
}
return (0);
}
static int
link_elf_each_function_nameval(linker_file_t file,
linker_function_nameval_callback_t callback, void *opaque)
{
linker_symval_t symval;
elf_file_t ef = (elf_file_t)file;
const Elf_Sym* symp;
int i, error;
/* Exhaustive search */
for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
if (symp->st_value != 0 &&
(ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC)) {
error = link_elf_symbol_values(file,
(c_linker_sym_t)symp, &symval);
if (error)
return (error);
error = callback(file, i, &symval, opaque);
if (error)
return (error);
}
}
return (0);
}
static void
elf_obj_cleanup_globals_cache(elf_file_t ef)
{
Elf_Sym *sym;
Elf_Size i;
for (i = 0; i < ef->ddbsymcnt; i++) {
sym = ef->ddbsymtab + i;
if (sym->st_shndx == SHN_FBSD_CACHED) {
sym->st_shndx = SHN_UNDEF;
sym->st_value = 0;
}
}
}
/*
* Symbol lookup function that can be used when the symbol index is known (ie
* in relocations). It uses the symbol index instead of doing a fully fledged
* hash table based lookup when such is valid. For example for local symbols.
* This is not only more efficient, it's also more correct. It's not always
* the case that the symbol can be found through the hash table.
*/
static int
elf_obj_lookup(linker_file_t lf, Elf_Size symidx, int deps, Elf_Addr *res)
{
elf_file_t ef = (elf_file_t)lf;
Elf_Sym *sym;
const char *symbol;
Elf_Addr res1;
/* Don't even try to lookup the symbol if the index is bogus. */
if (symidx >= ef->ddbsymcnt) {
*res = 0;
return (EINVAL);
}
sym = ef->ddbsymtab + symidx;
/* Quick answer if there is a definition included. */
if (sym->st_shndx != SHN_UNDEF) {
res1 = (Elf_Addr)sym->st_value;
if (ELF_ST_TYPE(sym->st_info) == STT_GNU_IFUNC)
res1 = ((Elf_Addr (*)(void))res1)();
*res = res1;
return (0);
}
/* If we get here, then it is undefined and needs a lookup. */
switch (ELF_ST_BIND(sym->st_info)) {
case STB_LOCAL:
/* Local, but undefined? huh? */
*res = 0;
return (EINVAL);
case STB_GLOBAL:
case STB_WEAK:
/* Relative to Data or Function name */
symbol = ef->ddbstrtab + sym->st_name;
/* Force a lookup failure if the symbol name is bogus. */
if (*symbol == 0) {
*res = 0;
return (EINVAL);
}
res1 = (Elf_Addr)linker_file_lookup_symbol(lf, symbol, deps);
/*
* Cache global lookups during module relocation. The failure
* case is particularly expensive for callers, who must scan
* through the entire globals table doing strcmp(). Cache to
* avoid doing such work repeatedly.
*
* After relocation is complete, undefined globals will be
* restored to SHN_UNDEF in elf_obj_cleanup_globals_cache(),
* above.
*/
if (res1 != 0) {
sym->st_shndx = SHN_FBSD_CACHED;
sym->st_value = res1;
*res = res1;
return (0);
} else if (ELF_ST_BIND(sym->st_info) == STB_WEAK) {
sym->st_value = 0;
*res = 0;
return (0);
}
return (EINVAL);
default:
return (EINVAL);
}
}
static void
link_elf_fix_link_set(elf_file_t ef)
{
static const char startn[] = "__start_";
static const char stopn[] = "__stop_";
Elf_Sym *sym;
const char *sym_name, *linkset_name;
Elf_Addr startp, stopp;
Elf_Size symidx;
int start, i;
startp = stopp = 0;
for (symidx = 1 /* zero entry is special */;
symidx < ef->ddbsymcnt; symidx++) {
sym = ef->ddbsymtab + symidx;
if (sym->st_shndx != SHN_UNDEF)
continue;
sym_name = ef->ddbstrtab + sym->st_name;
if (strncmp(sym_name, startn, sizeof(startn) - 1) == 0) {
start = 1;
linkset_name = sym_name + sizeof(startn) - 1;
}
else if (strncmp(sym_name, stopn, sizeof(stopn) - 1) == 0) {
start = 0;
linkset_name = sym_name + sizeof(stopn) - 1;
}
else
continue;
for (i = 0; i < ef->nprogtab; i++) {
if (strcmp(ef->progtab[i].name, linkset_name) == 0) {
startp = (Elf_Addr)ef->progtab[i].addr;
stopp = (Elf_Addr)(startp + ef->progtab[i].size);
break;
}
}
if (i == ef->nprogtab)
continue;
sym->st_value = start ? startp : stopp;
sym->st_shndx = i;
}
}
static int
link_elf_reloc_local(linker_file_t lf, bool ifuncs)
{
elf_file_t ef = (elf_file_t)lf;
const Elf_Rel *rellim;
const Elf_Rel *rel;
const Elf_Rela *relalim;
const Elf_Rela *rela;
const Elf_Sym *sym;
Elf_Addr base;
int i;
Elf_Size symidx;
link_elf_fix_link_set(ef);
/* Perform relocations without addend if there are any: */
for (i = 0; i < ef->nreltab; i++) {
rel = ef->reltab[i].rel;
if (rel == NULL) {
link_elf_error(ef->lf.filename, "lost a reltab");
return (ENOEXEC);
}
rellim = rel + ef->reltab[i].nrel;
base = findbase(ef, ef->reltab[i].sec);
if (base == 0) {
link_elf_error(ef->lf.filename, "lost base for reltab");
return (ENOEXEC);
}
for ( ; rel < rellim; rel++) {
symidx = ELF_R_SYM(rel->r_info);
if (symidx >= ef->ddbsymcnt)
continue;
sym = ef->ddbsymtab + symidx;
/* Only do local relocs */
if (ELF_ST_BIND(sym->st_info) != STB_LOCAL)
continue;
if ((ELF_ST_TYPE(sym->st_info) == STT_GNU_IFUNC ||
elf_is_ifunc_reloc(rel->r_info)) == ifuncs)
elf_reloc_local(lf, base, rel, ELF_RELOC_REL,
elf_obj_lookup);
}
}
/* Perform relocations with addend if there are any: */
for (i = 0; i < ef->nrelatab; i++) {
rela = ef->relatab[i].rela;
if (rela == NULL) {
link_elf_error(ef->lf.filename, "lost a relatab!");
return (ENOEXEC);
}
relalim = rela + ef->relatab[i].nrela;
base = findbase(ef, ef->relatab[i].sec);
if (base == 0) {
link_elf_error(ef->lf.filename, "lost base for reltab");
return (ENOEXEC);
}
for ( ; rela < relalim; rela++) {
symidx = ELF_R_SYM(rela->r_info);
if (symidx >= ef->ddbsymcnt)
continue;
sym = ef->ddbsymtab + symidx;
/* Only do local relocs */
if (ELF_ST_BIND(sym->st_info) != STB_LOCAL)
continue;
if ((ELF_ST_TYPE(sym->st_info) == STT_GNU_IFUNC ||
elf_is_ifunc_reloc(rela->r_info)) == ifuncs)
elf_reloc_local(lf, base, rela, ELF_RELOC_RELA,
elf_obj_lookup);
}
}
return (0);
}
static long
link_elf_symtab_get(linker_file_t lf, const Elf_Sym **symtab)
{
elf_file_t ef = (elf_file_t)lf;
*symtab = ef->ddbsymtab;
if (*symtab == NULL)
return (0);
return (ef->ddbsymcnt);
}
static long
link_elf_strtab_get(linker_file_t lf, caddr_t *strtab)
{
elf_file_t ef = (elf_file_t)lf;
*strtab = ef->ddbstrtab;
if (*strtab == NULL)
return (0);
return (ef->ddbstrcnt);
}