realpath(3) is used a lot e.g., by clang and is a major source of getcwd
and fstatat calls. This can be done more efficiently in the kernel.
This works by performing a regular lookup while saving the name and found
parent directory. If the terminal vnode is a directory we can resolve it using
usual means. Otherwise we can use the name saved by lookup and resolve the
parent.
See the review for sample syscall counts.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23574
lld 10.0 seems to generate this relocation for rdtsc_mb() ifunc in our libc.
Reported, reviewed, and tested by: dim (amd64, previous version)
Discussed with: emaste
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D23652
This allows for rtld to not issue two sigprocmask(2) syscalls for each
symbol binding operation in single-threaded processes. Rtld needs to
block signals as part of locking to ensure signal safety of the bind
process, because signal handlers might need to lazily resolve symbol
references.
As result, number of syscalls issued on startup by simple programs not
using libthr, is typically reduced 2x. For instance, for hello world,
I see:
non-sigfastblock
# (truss ./hello > /dev/null) |& wc -l
63
sigfastblock
# (truss ./hello > /dev/null) |& wc -l
37
Tested by: pho
Disscussed with: cem, emaste, jilles
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D12773
They were introduced to take care of ifunc, but right now no architecture
provides ifunc'ed variants. Since rtld uses memset extensively this results in
a pessmization. Should someone want to use ifunc here they should provide a
mandatory symbol (e.g., rtld_memset).
See the review for profiling data.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23176
Move all MD statements into $MACHINE_ARCH/Makefile.inc.
Unconditionally apply version script to rtld, the interpreter is not
functional without it for long time.
Reviewed by: brooks, emaste
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D23083
The path is already absolute.
Noted and reviewed by: rstone
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D23121
instead of killing the process. The same behaviour of terminating
image activation if the $ORIGIN cannot be resolved for the main
object, is kept.
Reported by: Greg V <greg@unrelenting.technology>
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D23053
We know the binary relative name and can reliably calculate cwd path.
Because realpath(3) was already linked into ld-elf.so.1, reuse it
there to resolve dots and dotdots making the path more canonical.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D23014
binpath local was changed from char array to a char pointer, update
strlcpy/strlcat uses.
Reported by: Coverity through vangyzen
CID: 1412239 and 1412240
Reviewed by: emaste, imp, vangyzen
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D23090
When rtld is directly executed with arguments, it has to move the
program arguments, environment and elf aux data up a few slots to
remove its own arguments before the process being executed sees
them. When copying the environment, rtld was incorrectly testing
whether the location about to be written to currently contained
NULL, when was supposed to check whether it had just copied the
NULL terminator of the environment string. This had the result
that the ELF aux data was mostly treated as environment variables,
and rtld would quickly crash when it tried to access required
ELF aux data that it didn't think was present.
Differential Revision: https://reviews.freebsd.org/D23008
Reviewed by: kib
MFC after: 1 month
When activated in direct exec mode, kernel-provided AT_EXECPATH points
to the interpreter. We need to recalculate auxv to point to the
string with the path to the executable which is actually executed.
The somewhat problematic case is when the executable path is relative
and either $PATH use is not enabled or it contains '/' so $PATH search
is not performed. In this case resulting AT_EXECPATH is relative, I
might fix this later.
Reported and reviewed by: rstone
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D22894
Summary:
As a transition aide, implement an alternative elfN_freebsd_fixup which
is called for old powerpc binaries. Similarly, add a translation to rtld to
convert old values to new ones (as expected by a new rtld).
Translation of old<->new values is incomplete, but sufficient to allow an
installworld of a new userspace from an old one when a new kernel is running.
Test Plan:
Someone needs to see how a new kernel/rtld/libc works with an old
binary. If if works we can probalby ship this. If not we probalby need
some more compat bits.
Submitted by: brooks
Reviewed by: jhibbits
Differential Revision: https://reviews.freebsd.org/D20799
As PowerPC is moving to clang, we can finally start taking advantage of
IFUNC.
Implement the MD parts of IFUNC handling for rtld.
Currently, it is necessary to look for R_PPC_IRELATIVE in the PLT in
addition to RELA. This is an ABI violation, but LLD9 has some .iplt bugs
that require this as a workaround.
Reviewed by: jhibbits
Differential Revision: https://reviews.freebsd.org/D22855
As PowerPC is moving to clang, we can finally start taking advantage of
IFUNC.
Implement the MD parts of IFUNC handling for rtld.
Currently, it is necessary to look for R_PPC_IRELATIVE in the PLT in
addition to RELA. This is an ABI violation, but LLD9 has some .iplt bugs
that require this as a workaround.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D22789
When a PT_LOAD segment has a zero p_filesz, skip the data mmap, as mmapping
zero bytes from a file is an error.
A PT_LOAD with zero p_filesz is legal (but somewhat uncommon due to segment
merging in modern linkers, as it is more efficient to merge .data and .bss
by just extending p_memsz in the previous segment, assuming compatible
page protection.)
This was seen on ports/graphics/glew on a powerpc64 ELFv2 experimental
build.
Submitted by: Alfredo Dal'Ava Junior <alfredo.junior@eldorado.org.br>
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D22634
TP points to the start of the TLS block after the tcb, but
Obj_Entry.tlsoffset includes the tcb, so subtract the size of the tcb
to compute the offset relative to TP.
This is identical to the same fixes for powerpc in r339072 and r342671.
Reviewed by: James Clarke
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D22661
Four bytes of padding are needed in the regular powerpc case to bring the
stack frame size up to a multiple of 16 bytes to meet ABI requirements.
Fixes odd hangs I was encountering during testing.
Summary:
We need to save off the full 64-bit register, not just the low 32 bits,
of all registers getting saved off in _rtld_bind_start. Additionally,
we need to save off the other SPE registers (SPEFSCR and accumulator),
so that their program state is not affected by the PLT resolver.
Reviewed by: bdragon
Differential Revision: https://reviews.freebsd.org/D22520
Alter bsd.compat.mk to set MACHINE and MACHINE_ARCH when included
directly so MD paths in Makefiles work. In the process centralize
setting them in LIBCOMPATWMAKEENV.
Alter .PATH and CFLAGS settings in work when the Makefile is included.
While here only support LIB32 on supported platforms rather than always
enabling it and requiring users of MK_LIB32 to filter based
TARGET/MACHINE_ARCH.
The net effect of this change is to make Makefile.libcompat only build
compatability libraries.
Changes relative to r354449:
Correct detection of the compiler type when bsd.compat.mk is used
outside Makefile.libcompat. Previously it always matched the clang
case.
Set LDFLAGS including the linker emulation for mips where -m32 seems to
be insufficent.
Reviewed by: imp, kib (origional version in r354449)
Obtained from: CheriBSD (conceptually)
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D22251
Alter bsd.compat.mk to set MACHINE and MACHINE_ARCH when included
directly so MD paths in Makefiles work. In the process centralize
setting them in LIBCOMPATWMAKEENV.
Alter .PATH and CFLAGS settings in work when the Makefile is included.
While here only support LIB32 on supported platforms rather than always
enabling it and requiring users of MK_LIB32 to filter based
TARGET/MACHINE_ARCH.
The net effect of this change is to make Makefile.libcompat only build
compatability libraries.
Reviewed by: imp, kib
Obtained from: CheriBSD (conceptually)
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D22251
After the aux vector is moved, it is necessary to re-digest aux_info so the
pointers are updated to the new locations.
This was causing thread creation to fail on powerpc64 when using direct
execution due to a nonsense value being read for aux_info[AT_STACKPROT].
Approved by: jhibbits (mentor)
Differential Revision: https://reviews.freebsd.org/D21656
In the past, this allocator seems to have allocated things larger than
a page seperately. Much of this code was removed at some point (perhaps
along with sbrk() used) so remove the rest. Instead, keep allocating in
power-of-two bins up to FIRST_BUCKET_SIZE << (NBUCKETS - 1). If we want
something more efficent, we should use a fancier allocator.
While here, remove some vestages of sbrk() use. Most importantly, don't
try to page align the pagepool since it's always page aligned by mmap().
Reviewed by: kib
Obtained from: CheriBSD
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D21453
In Seventh Edition UNIX, the last pointer passed to free() was
guaranteed to not actually have been freed allowing memory to be
"compacted" via the following pattern:
free(foo);
foo = realloc(foo, newsize);
Further, Andrew Koenig reports in "C Traps and Pitfalls" that the
original realloc() implementation required this pattern.
The C standard is clear that this is Undefined Behavior. Modern
allocators don't support it and no portable code could rely on it so
remove this support.
Note: the removed implementation contains an off-by-one error and if
an item isn't found on the freelist, then twice as much memory as the
largest possible allocation will be copied.
Reviewed by: kib, imp
Obtained from: CheriBSD
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D21296
Instead of restoring the saved values of argc, argv and envp,
these must be loaded from the stack that _rtld() modifies.
This fixes rtld direct exec mode.
E.g.: /libexec/ld-elf.so.1 /bin/ls
Reviewed by: jhibbits
Differential Revision: https://reviews.freebsd.org/D21131
First, amd64 version of the script cannot work at least due to the
wrong architecture specification. Second, kernel can activate shared
objects for long time, due to PIE support.
It seems the intent was to allow ld-elf.so.1 to be build and used as
an executable. Since we have direct exec mode implemented for dso
ld-elf.so.1, the non-functional and commented out scripts can be
finally removed.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
I found this on one of the CheriBSD Jenkins builders. Using
beforelinking instead of ${PROG} should fix the dependency for the
DEBUG_FILES case.
Reviewed by: brooks
Currently RTLD is linked against libc_nossp_pic which means that any libc
symbol used in rtld can pull in a lot of depedencies. This was causing
symbol such as __libc_interposing and all the pthread stubs to be included
in RTLD even though they are not required. It turns out most of these
dependencies can easily be avoided by providing overrides inside of rtld.
This change is motivated by CHERI, where we have an experimental ABI that
requires additional relocation processing to allow the use of function
pointers inside of rtld. Instead of adding this self-relocation code to
RTLD I attempted to remove most function pointers from RTLD and discovered
that most of them came from the libc dependencies instead of being actually
used inside rtld.
A nice side-effect of this change is that rtld is now 22% smaller on amd64.
text data bss dec hex filename
0x21eb6 0xce0 0xe60 145910 239f6 /home/alr48/ld-elf-x86.before.so.1
0x1a6ed 0x728 0xdd8 113645 1bbed /home/alr48/ld-elf-x86.after.so.1
The number of R_X86_64_RELATIVE relocations that need to be processed on
startup has also gone down from 368 to 187 (almost 50% less).
Reviewed By: kib
Differential Revision: https://reviews.freebsd.org/D20663
Summary:
PowerPC has two PLT models: BSS-PLT and Secure-PLT. BSS-PLT uses runtime
code generation to generate the PLT stubs. Secure-PLT was introduced with
GCC 4.1 and Binutils 2.17 (base has GCC 4.2.1 and Binutils 2.17), and is a
more secure PLT format, using a read-only linkage table, with the dynamic
linker populating a non-executable index table.
This is the libc, rtld, and kernel support only. The toolchain and build
parts will be updated separately.
Reviewed By: nwhitehorn, bdragon, pfg
Differential Revision: https://reviews.freebsd.org/D20598
MFC after: 1 month
Use roundup2() and rounddown2() instead of inlining them.
Get rid of the fd local variable, use literal -1 for the mmap argument.
Use MAP_FAILED as mmap(2) failure indicator.
After that, apply some style.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
kern_execve() locks text vnode exclusive to be able to set and clear
VV_TEXT flag. VV_TEXT is mutually exclusive with the v_writecount > 0
condition.
The change removes VV_TEXT, replacing it with the condition
v_writecount <= -1, and puts v_writecount under the vnode interlock.
Each text reference decrements v_writecount. To clear the text
reference when the segment is unmapped, it is recorded in the
vm_map_entry backed by the text file as MAP_ENTRY_VN_TEXT flag, and
v_writecount is incremented on the map entry removal
The operations like VOP_ADD_WRITECOUNT() and VOP_SET_TEXT() check that
v_writecount does not contradict the desired change. vn_writecheck()
is now racy and its use was eliminated everywhere except access.
Atomic check for writeability and increment of v_writecount is
performed by the VOP. vn_truncate() now increments v_writecount
around VOP_SETATTR() call, lack of which is arguably a bug on its own.
nullfs bypasses v_writecount to the lower vnode always, so nullfs
vnode has its own v_writecount correct, and lower vnode gets all
references, since object->handle is always lower vnode.
On the text vnode' vm object dealloc, the v_writecount value is reset
to zero, and deadfs vop_unset_text short-circuit the operation.
Reclamation of lowervp always reclaims all nullfs vnodes referencing
lowervp first, so no stray references are left.
Reviewed by: markj, trasz
Tested by: mjg, pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 month
Differential revision: https://reviews.freebsd.org/D19923
- Remove dead and most likely rotten MALLOC_DEBUG, MSTAT, and RCHECK options.
- Remove unused headers.
- Remove one case of undefined behavior where left shift could overflow.
It is impossible on practice for rtld and libthr consumer.
PR: 237577
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Since inits for the main binary are run from rtld (for some time), the
rtld_exit atexit(3) handler, which is passed from rtld to the program
entry and installed by csu, is installed after any atexit(3) handlers
installed by main binary constructors. This means that rtld_exit() is
fired before main binary handlers.
Typical C++ static constructors are executed from init (either binary
or libs) but use atexit(3) to ensure that destructors are called in
the right order, independent of the linking order. Also, C++
libraries finalizers call __cxa_finalize(3) to flush library'
atexit(3) entries. Since atexit(3) entry is cleared after being run,
this would be mostly innocent, except that, atexit(rtld_exit) done
after main binary constructors, makes destructors from libraries
executed before destructors for main.
Fix by reordering atexit(rtld_exit) before inits for main binary, same
as it happened when inits were called by csu. Do it using new private
libc symbol with pre-defined ABI.
Reported. tested, and reviewed by: kan
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
This causes some increase of the dynamic linker size, but benefits of
avoiding compiling private copy or the linker when debugging is
required. definitely worth it.
The dbg() calls can be compiled out by defining LD_NO_DEBUG symbol.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
If dso uses initial exec TLS mode, rtld tries to allocate TLS in
static space. If there is no space left, the dlopen(3) fails. If space
if allocated, initial content from PT_TLS segment is distributed to
all threads' pcbs, which was missed and caused un-initialized TLS
segment for such dso after dlopen(3).
The mode is auto-detected either due to the relocation used, or if the
DF_STATIC_TLS dynamic flag is set. In the later case, the TLS segment
is tried to allocate earlier, which increases chance of the dlopen(3)
to succeed. LLD was recently fixed to properly emit the flag, ld.bdf
did it always.
Initial test by: dumbbell
Tested by: emaste (amd64), ian (arm)
Tested by: Gerald Aryeetey <aryeeteygerald_rogers.com> (arm64)
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D19072
allocate_tls_offset returns true on success. The same issue existed
on arm and was fixed in r345693.
PR: 236880
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
allocate_tls_offset returns true on success. This still needs more
testing and review, but this change is consistent with other archs.
PR: 236880
Reported by: Andrew Gierth <andrew@tao11.riddles.org.uk>
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
r345620 by kib@ fixed the rtld issue that caused a crash at startup
during resolution of libc's ifuncs with BIND_NOW.
PR: 233333
Sponsored by: The FreeBSD Foundation
