]> git.ipfire.org Git - thirdparty/glibc.git/log
thirdparty/glibc.git
21 months agoresolv: Add the __ns_samebinaryname function
Florian Weimer [Tue, 30 Aug 2022 08:02:49 +0000 (10:02 +0200)] 
resolv: Add the __ns_samebinaryname function

During packet parsing, only the binary name is available.  If the name
equality check is performed before conversion to text, we can sometimes
skip the last step.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
21 months agoresolv: Add internal __res_binary_hnok function
Florian Weimer [Tue, 30 Aug 2022 08:02:49 +0000 (10:02 +0200)] 
resolv: Add internal __res_binary_hnok function

During package parsing, only the binary representation is available,
and it is convenient to check that directly for conformance with host
name requirements.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
21 months agoresolv: Add tst-resolv-aliases
Florian Weimer [Tue, 30 Aug 2022 08:02:49 +0000 (10:02 +0200)] 
resolv: Add tst-resolv-aliases

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
21 months agoresolv: Add tst-resolv-byaddr for testing reverse lookup
Florian Weimer [Tue, 30 Aug 2022 08:02:49 +0000 (10:02 +0200)] 
resolv: Add tst-resolv-byaddr for testing reverse lookup

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
21 months agoLoongArch: Use __builtin_{fmax,fmaxf,fmin,fminf} with GCC >= 13
Xi Ruoyao [Sat, 20 Aug 2022 08:43:44 +0000 (16:43 +0800)] 
LoongArch: Use __builtin_{fmax,fmaxf,fmin,fminf} with GCC >= 13

GCC 13 compiles these built-ins to {fmax,fmin}.{s/d} instruction, use
them instead of the generic implementation.

Link: https://gcc.gnu.org/r13-2085
Signed-off-by: Xi Ruoyao <xry111@xry111.site>
21 months agoLoongArch: Fix ptr mangling/demangling features.
caiyinyu [Wed, 17 Aug 2022 00:17:39 +0000 (08:17 +0800)] 
LoongArch: Fix ptr mangling/demangling features.

21 months agonscd: Fix netlink cache invalidation if epoll is used [BZ #29415]
Fabian Vogt [Wed, 27 Jul 2022 09:44:07 +0000 (11:44 +0200)] 
nscd: Fix netlink cache invalidation if epoll is used [BZ #29415]

Processes cache network interface information such as whether IPv4 or IPv6
are enabled. This is only checked again if the "netlink timestamp" provided
by nscd changed, which is triggered by netlink socket activity.

However, in the epoll handler for the netlink socket, it was missed to
assign the new timestamp to the nscd database. The handler for plain poll
did that properly, copy that over.

This bug caused that e.g. processes which started before network
configuration got unusuable addresses from getaddrinfo, like IPv6 only even
though only IPv4 is available:
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/1041

It's a bit hard to reproduce, so I verified this by checking the timestamp
on calls to __check_pf manually. Without this patch it's stuck at 1, now
it's increasing on network changes as expected.

Signed-off-by: Fabian Vogt <fvogt@suse.de>
21 months agoAdd test for bug 29530
Andreas Schwab [Mon, 29 Aug 2022 13:05:40 +0000 (15:05 +0200)] 
Add test for bug 29530

This tests for a bug that was introduced in commit edc1686af0 ("vfprintf:
Reuse work_buffer in group_number") and fixed as a side effect of commit
6caddd34bd ("Remove most vfprintf width/precision-dependent allocations
(bug 14231, bug 26211).").

21 months agoMakeconfig: Set pie-ccflag to -fPIE by default [BZ# 29514]
Richard Henderson [Tue, 23 Aug 2022 20:49:23 +0000 (13:49 -0700)] 
Makeconfig: Set pie-ccflag to -fPIE by default [BZ# 29514]

We should default to the larger code model, in order to support
larger applications built with -static -pie.  This should be
consistent with pic-ccflag, which defaults to -fPIC.

Remove the now redundant override from sysdeps/sparc/Makefile.
Note that -fno-pie and -fno-PIE have the same effect.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Tested-by: Carlos O'Donell <carlos@redhat.com>
21 months agohurd: Fix vm_size_t incoherencies
Samuel Thibault [Sun, 28 Aug 2022 23:42:47 +0000 (01:42 +0200)] 
hurd: Fix vm_size_t incoherencies

In gnumach, 3e1702a65fb3 ("add rpc_versions for vm types") changed the type
of vm_size_t, making it always a unsigned long. This made it incompatible on
x86 with size_t. Even if we may want to revert it to unsigned int, it's
better to fix the types of parameters according to the .defs files.

21 months agomach: Make xpg_strerror_r set a message on error
Samuel Thibault [Sat, 27 Aug 2022 12:46:23 +0000 (14:46 +0200)] 
mach: Make xpg_strerror_r set a message on error

posix advises to have strerror_r fill a message even when we are returning
an error.

This makes mach's xpg_strerror_r do this, like the generic version does.

Spotted by the libunistring testsuite test-strerror_r

21 months agomach: Fix incoherency between perror and strerror
Samuel Thibault [Sat, 27 Aug 2022 11:52:46 +0000 (13:52 +0200)] 
mach: Fix incoherency between perror and strerror

08d2024b4167 ("string: Simplify strerror_r") inadvertently made
__strerror_r print unknown error system in decimal while the original
code was printing it in hexadecimal. perror was kept printing in
hexadecimal in 725eeb4af14c ("string: Use tls-internal on strerror_l"),
let us keep both coherent.

This also fixes a duplicate ':'

Spotted by the libunistring testsuite test-perror2

21 months agoelf: Call __libc_early_init for reused namespaces (bug 29528)
Florian Weimer [Fri, 26 Aug 2022 19:15:43 +0000 (21:15 +0200)] 
elf: Call __libc_early_init for reused namespaces (bug 29528)

libc_map is never reset to NULL, neither during dlclose nor on a
dlopen call which reuses the namespace structure.  As a result, if a
namespace is reused, its libc is not initialized properly.  The most
visible result is a crash in the <ctype.h> functions.

To prevent similar bugs on namespace reuse from surfacing,
unconditionally initialize the chosen namespace to zero using memset.

21 months agocsu: Change start code license to have link exception
Szabolcs Nagy [Mon, 22 Aug 2022 13:21:53 +0000 (14:21 +0100)] 
csu: Change start code license to have link exception

The start code can get linked into dynamic linked executables where
LGPL would require shipping the source or linkable binaries when the
executable is distributed.

On some targets the license exception was missing in start.S (which
is compiled into crt1.o and Scrt1.o which may end up linked into PDE
and PIE binaries).

I did not review what other code may end up in executables, just
fixed the start.S license inconsistency across targets.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
21 months agos390: Move hwcaps/platform names out of _rtld_global_ro
Florian Weimer [Thu, 25 Aug 2022 14:34:20 +0000 (16:34 +0200)] 
s390: Move hwcaps/platform names out of _rtld_global_ro

Changes to these arrays are often backported to stable releases,
but additions to these arrays shift the offsets of the following
_rltd_global_ro members, thus breaking the GLIBC_PRIVATE ABI.

Obviously, this change is itself an internal ABI break, but at least
it will avoid further ABI breaks going forward.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
21 months agoRevert "Detect ld.so and libc.so version inconsistency during startup"
Florian Weimer [Thu, 25 Aug 2022 16:46:41 +0000 (18:46 +0200)] 
Revert "Detect ld.so and libc.so version inconsistency during startup"

This reverts commit 6f85dbf102ad7982409ba0fe96886caeb6389fef.

Once this change hits the release branches, it will require relinking
of all statically linked applications before static dlopen works
again, for the majority of updates on release branches: The NEWS file
is regularly updated with bug references, so the __libc_early_init
suffix changes, and static dlopen cannot find the function anymore.

While this ABI check is still technically correct (we do require
rebuilding & relinking after glibc updates to keep static dlopen
working), it is too drastic for stable release branches.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
21 months agoAdd NT_LOONGARCH_* from Linux 5.19 to elf.h
Joseph Myers [Wed, 24 Aug 2022 18:19:13 +0000 (18:19 +0000)] 
Add NT_LOONGARCH_* from Linux 5.19 to elf.h

Add the new NT_LOONGARCH_* constants from Linux 5.19 to glibc's elf.h.

Tested for x86_64.

21 months agoDetect ld.so and libc.so version inconsistency during startup
Florian Weimer [Wed, 24 Aug 2022 15:35:36 +0000 (17:35 +0200)] 
Detect ld.so and libc.so version inconsistency during startup

The files NEWS, include/link.h, and sysdeps/generic/ldsodefs.h
contribute to the version fingerprint used for detection.  The
fingerprint can be further refined using the --with-extra-version-id
configure argument.

_dl_call_libc_early_init is replaced with _dl_lookup_libc_early_init.
The new function is used store a pointer to libc.so's
__libc_early_init function in the libc_map_early_init member of the
ld.so namespace structure.  This function pointer can then be called
directly, so the separate invocation function is no longer needed.

The versioned symbol lookup needs the symbol versioning data
structures, so the initialization of libc_map and libc_map_early_init
is now done from _dl_check_map_versions, after this information
becomes available.  (_dl_map_object_from_fd does not set this up
in time, so the initialization code had to be moved from there.)
This means that the separate initialization code can be removed from
dl_main because _dl_check_map_versions covers all maps, including
the initial executable loaded by the kernel.  The lookup still happens
before relocation and the invocation of IFUNC resolvers, so IFUNC
resolvers are protected from ABI mismatch.

The __libc_early_init function pointer is not protected because
so little code runs between the pointer write and the invocation
(only dynamic linker code and IFUNC resolvers).

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
21 months agoMerge getopt patch from Gnulib
Paul Eggert [Wed, 24 Aug 2022 04:16:40 +0000 (21:16 -0700)] 
Merge getopt patch from Gnulib

* posix/getopt.c [!_LIBC]: Merge _WIN32 patch from Gnulib
so that these source files are identical.
This makes no difference for glibc.

21 months agoMerge _GL_UNUSED C23 patch from Gnulib
Paul Eggert [Wed, 24 Aug 2022 04:16:39 +0000 (21:16 -0700)] 
Merge _GL_UNUSED C23 patch from Gnulib

* posix/getopt.c (_getopt_initialize):
* sysdeps/posix/tempname.c (try_dir, try_nocreate):
Put _GL_UNUSED before args instead of after.
This makes no difference for glibc.
It is needed for Gnulib when being compiled on
non-GCC C23 compilers.

21 months agoLoongArch: Fix dl-machine.h code formatting.
Xi Ruoyao [Wed, 24 Aug 2022 01:27:04 +0000 (09:27 +0800)] 
LoongArch: Fix dl-machine.h code formatting.

No functional change.

21 months agoscripts/glibcelf.py: Add hashing support
Florian Weimer [Fri, 19 Aug 2022 09:16:32 +0000 (11:16 +0200)] 
scripts/glibcelf.py: Add hashing support

ELF and GNU hashes can now be computed using the elf_hash and
gnu_hash functions.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Tested-by: Carlos O'Donell <carlos@redhat.com>
21 months agohurd: Fix starting static binaries with stack protection enabled
Samuel Thibault [Mon, 22 Aug 2022 20:34:31 +0000 (22:34 +0200)] 
hurd: Fix starting static binaries with stack protection enabled

gcc introduces gs:0x14 accesses in most functions, so we need some tcbhead
to be ready very early during initialization.  This configures a static area
which can be referenced by various protected functions, until proper TLS is
set up.

21 months agohtl: Make pthread*_cond_timedwait register wref before releasing mutex
Samuel Thibault [Mon, 22 Aug 2022 20:27:24 +0000 (22:27 +0200)] 
htl: Make pthread*_cond_timedwait register wref before releasing mutex

Otherwise another thread could be rightly trying to destroy the condition,
see e.g. tst-cond20.

21 months agohtl: make __pthread_hurd_cond_timedwait_internal check mutex is held
Samuel Thibault [Mon, 22 Aug 2022 20:25:27 +0000 (22:25 +0200)] 
htl: make __pthread_hurd_cond_timedwait_internal check mutex is held

Like __pthread_cond_timedwait_internal already does.

21 months agoAdd AArch64 HWCAP2_* constants from Linux 5.19
Joseph Myers [Mon, 22 Aug 2022 14:59:39 +0000 (14:59 +0000)] 
Add AArch64 HWCAP2_* constants from Linux 5.19

Linux 5.19 adds more HWCAP2_* values for AArch64; add these to its
bits/hwcap.h header in glibc.

Tested with build-many-glibcs.py for aarch64-linux-gnu.

21 months agoAdd AGROUP from Linux 5.19 to sys/acct.h, remove Alpha version (bug 29502)
Joseph Myers [Mon, 22 Aug 2022 14:16:57 +0000 (14:16 +0000)] 
Add AGROUP from Linux 5.19 to sys/acct.h, remove Alpha version (bug 29502)

Linux 5.19 adds a new accounting flag AGROUP; add it to the
enumeration in sys/acct.h.

This shows up that the Alpha-specific variant of this header has a
different set of constants and struct acct, which appear to be the
constants and structure layout from Linux 2.0.  These were changed
some time between Linux 2.0 and Linux 2.2; I see no evidence of an
Alpha-specific layout or set of constants, but haven't checked the
detailed Linux kernel history between those versions.  Rather, it
looks like tha Alpha-specific header was originally needed because of
the use of types in the kernel structure (such as uid_t and gid_t)
that had different sizes on Alpha, and when glibc was updated for
changes to the structure and constants in the kernel

1998-10-02  Andreas Jaeger  <aj@arthur.rhein-neckar.de>

        * sysdeps/unix/sysv/linux/sys/acct.h: Bring in sync with current
        linux 2.1 version.

that simply omitted to do anything about the Alpha version.

Thus, remove the Alpha version in order to get the updated definitions
into use on Alpha, as I don't think the interfaces are actually
different for Alpha with any kernel version supported by glibc.

Tested for x86_64, and with build-many-glibcs.py for alpha-linux-gnu.

21 months agoalpha: Fix generic brk system call emulation in __brk_call (bug 29490)
Florian Weimer [Mon, 22 Aug 2022 09:04:47 +0000 (11:04 +0200)] 
alpha: Fix generic brk system call emulation in __brk_call (bug 29490)

The kernel special-cases the zero argument for alpha brk, and we can
use that to restore the generic Linux error handling behavior.

Fixes commit b57ab258c1140bc45464b4b9908713e3e0ee35aa ("Linux:
Introduce __brk_call for invoking the brk system call").

21 months agohurd: Assume non-suid during bootstrap
Samuel Thibault [Fri, 19 Aug 2022 00:16:54 +0000 (02:16 +0200)] 
hurd: Assume non-suid during bootstrap

We do not have a hurd data block only when bootstrapping the system, in
which case we don't have a notion of suid yet anyway.

This is needed, otherwise init_standard_fds would check that standard
file descriptors are allocated, which is meaningless during bootstrap.

21 months agoUse binutils 2.39 branch in build-many-glibcs.py
Joseph Myers [Thu, 18 Aug 2022 13:16:56 +0000 (13:16 +0000)] 
Use binutils 2.39 branch in build-many-glibcs.py

This patch makes build-many-glibcs.py use binutils 2.39 branch.

Tested with build-many-glibcs.py (compilers and glibcs builds).  Note:
binutils 2.39 shows the same failures for i686-linux-gnu-no-pie,
x86_64-linux-gnu-no-pie and x86_64-linux-gnu-x32-no-pie building the
glibc testsuite as binutils mainline does.

21 months agoS390: Fix werror=unused-variable in ifunc-impl-list.c.
Stefan Liebler [Thu, 18 Aug 2022 07:10:48 +0000 (09:10 +0200)] 
S390: Fix werror=unused-variable in ifunc-impl-list.c.

If the architecture level set is high enough, no IFUNCs are used at
all and the variable i would be unused.  Then the build fails with:
../sysdeps/s390/multiarch/ifunc-impl-list.c: In function ‘__libc_ifunc_impl_list’:
../sysdeps/s390/multiarch/ifunc-impl-list.c:76:10: error: unused variable ‘i’ [-Werror=unused-variable]
   76 |   size_t i = max;
      |          ^
cc1: all warnings being treated as errors

21 months agoEnsure calculations happen with desired rounding mode in y1lf128
Michael Hudson-Doyle [Thu, 11 Aug 2022 23:29:31 +0000 (11:29 +1200)] 
Ensure calculations happen with desired rounding mode in y1lf128

math/test-float128-y1 fails on x86_64 and ppc64el with gcc 12 and -O3,
because code inside a block guarded by SET_RESTORE_ROUNDL is being moved
after the rounding mode has been restored. Use math_force_eval to
prevent this (and insert some math_opt_barrier calls to prevent code
from being moved before the rounding mode is set).

Fixes #29463

Reviewed-By: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
21 months agolocaledata: Convert French language locales (fr_*) to UTF-8
Florian Weimer [Wed, 17 Aug 2022 07:10:20 +0000 (09:10 +0200)] 
localedata: Convert French language locales (fr_*) to UTF-8

21 months agoLinux: Fix enum fsconfig_command detection in <sys/mount.h>
Florian Weimer [Tue, 16 Aug 2022 07:25:23 +0000 (09:25 +0200)] 
Linux: Fix enum fsconfig_command detection in <sys/mount.h>

The #ifdef FSOPEN_CLOEXEC check did not work because the macro
was always defined in this header prior to the check, so that
the <linux/mount.h> contents did not matter.

Fixes commit 774058d72942249f71d74e7f2b639f77184160a6
("linux: Fix sys/mount.h usage with kernel headers").

21 months agoelf: Run tst-audit-tlsdesc, tst-audit-tlsdesc-dlopen everywhere
Florian Weimer [Tue, 16 Aug 2022 06:27:50 +0000 (08:27 +0200)] 
elf: Run tst-audit-tlsdesc, tst-audit-tlsdesc-dlopen everywhere

The test is valid for all TLS models, but we want to make a reasonable
effort to test the GNU2 model specifically.  For example, aarch64
defaults to GNU2, but does not have -mtls-dialect=gnu2, and the test
was not run there.

Suggested-by: Martin Coufal <mcoufal@redhat.com>
21 months agoMove ip_mreqn structure from Linux to generic
Samuel Thibault [Sat, 13 Aug 2022 21:02:51 +0000 (23:02 +0200)] 
Move ip_mreqn structure from Linux to generic

I.e. from sysdeps/unix/sysv/linux/bits/in.h to netinet/in.h

It is following both the BSD and Linux definitions.

Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agomalloc: Do not use MAP_NORESERVE to allocate heap segments
Florian Weimer [Mon, 15 Aug 2022 14:45:40 +0000 (16:45 +0200)] 
malloc: Do not use MAP_NORESERVE to allocate heap segments

Address space for heap segments is reserved in a mmap call with
MAP_ANONYMOUS | MAP_PRIVATE and protection flags PROT_NONE.  This
reservation does not count against the RSS limit of the process or
system.  Backing memory is allocated using mprotect in alloc_new_heap
and grow_heap, and at this point, the allocator expects the kernel
to provide memory (subject to memory overcommit).

The SIGSEGV that might generate due to MAP_NORESERVE (according to
the mmap manual page) does not seem to occur in practice, it's always
SIGKILL from the OOM killer.  Even if there is a way that SIGSEGV
could be generated, it is confusing to applications that this only
happens for secondary heaps, not for large mmap-based allocations,
and not for the main arena.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
21 months agoLinux: Terminate subprocess on late failure in tst-pidfd (bug 29485)
Florian Weimer [Mon, 15 Aug 2022 14:43:59 +0000 (16:43 +0200)] 
Linux: Terminate subprocess on late failure in tst-pidfd (bug 29485)

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
21 months agonon-linux: bits/in.h: Add more RFC options
Samuel Thibault [Mon, 15 Aug 2022 09:25:55 +0000 (11:25 +0200)] 
non-linux: bits/in.h: Add more RFC options

21 months agoarm: Remove nested functionf rom relocate_pc24
Adhemerval Zanella [Tue, 26 Jul 2022 13:34:01 +0000 (10:34 -0300)] 
arm: Remove nested functionf rom relocate_pc24

Checked on arm-linux-gnueabihf.

21 months agolinux: Fix sys/mount.h usage with kernel headers
Adhemerval Zanella [Wed, 10 Aug 2022 17:24:47 +0000 (14:24 -0300)] 
linux: Fix sys/mount.h usage with kernel headers

Now that kernel exports linux/mount.h and includes it on linux/fs.h,
its definitions might clash with glibc exports sys/mount.h.  To avoid
the need to rearrange the Linux header to be always after glibc one,
the glibc sys/mount.h is changed to:

  1. Undefine the macros also used as enum constants.  This covers prior
     inclusion of <linux/mount.h> (for instance MS_RDONLY).

  2. Include <linux/mount.h> based on the usual __has_include check
     (needs to use __has_include ("linux/mount.h") to paper over GCC
     bugs.

  3. Define enum fsconfig_command only if FSOPEN_CLOEXEC is not defined.
     (FSOPEN_CLOEXEC should be a very close proxy.)

  4. Define struct mount_attr if MOUNT_ATTR_SIZE_VER0 is not defined.
     (Added in the same commit on the Linux side.)

This patch also adds some tests to check if including linux/fs.h and
linux/mount.h after and before sys/mount.h does work.

Checked on x86_64-linux-gnu.

Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agolinux: Use compile_c_snippet to check linux/mount.h availability
Adhemerval Zanella [Wed, 10 Aug 2022 17:24:46 +0000 (14:24 -0300)] 
linux: Use compile_c_snippet to check linux/mount.h availability

Checked on x86_64-linux-gnu.

Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agolinux: Mimic kernel defition for BLOCK_SIZE
Adhemerval Zanella [Wed, 10 Aug 2022 17:24:45 +0000 (14:24 -0300)] 
linux: Mimic kernel defition for BLOCK_SIZE

To avoid possible warnings if the kernel header is included before
sys/mount.h.

Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agolinux: Use compile_c_snippet to check linux/pidfd.h availability
Adhemerval Zanella [Wed, 10 Aug 2022 17:24:44 +0000 (14:24 -0300)] 
linux: Use compile_c_snippet to check linux/pidfd.h availability

Instead of tying to a specific kernel version.

Checked on x86_64-linux-gnu.

Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agoglibcextract.py: Add compile_c_snippet
Adhemerval Zanella [Wed, 10 Aug 2022 19:24:06 +0000 (16:24 -0300)] 
glibcextract.py: Add compile_c_snippet

It might be used on tests to check if a snippet build with the provided
compiler and flags.

Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agoLoongArch: Add pointer mangling support.
caiyinyu [Wed, 10 Aug 2022 02:21:46 +0000 (10:21 +0800)] 
LoongArch: Add pointer mangling support.

21 months agoAArch64: Fix typo in sve configure check (BZ# 29394)
Wilco Dijkstra [Thu, 11 Aug 2022 16:52:00 +0000 (17:52 +0100)] 
AArch64: Fix typo in sve configure check (BZ# 29394)

Fix a typo in the SVE configure check. This fixes [BZ# 29394].

21 months agolibio: Improve performance of IO locks
Wilco Dijkstra [Thu, 11 Aug 2022 15:47:45 +0000 (16:47 +0100)] 
libio: Improve performance of IO locks

Improve performance of recursive IO locks by adding a fast path for
the single-threaded case. To reduce the number of memory accesses for
locking/unlocking, only increment the recursion counter if the lock
is already taken.

On Neoverse V1, a microbenchmark with many small freads improved by
2.9x. Multithreaded performance improved by 2%.

Reviewed-by: Cristian Rodríguez <crrodriguez@opensuse.org>
21 months agotst-process_madvise: Check process_madvise-syscall support.
Stefan Liebler [Thu, 11 Aug 2022 07:47:46 +0000 (09:47 +0200)] 
tst-process_madvise: Check process_madvise-syscall support.

So far this test checks if pidfd_open-syscall is supported,
which was introduced with linux 5.3.

The process_madvise-syscall was introduced with linux 5.10.
Thus you'll get FAILs if you are running a kernel in between.

This patch adds a check if the first process_madvise-syscall
returns ENOSYS and in this case will fail with UNSUPPORTED.
Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agoelf.h: Add ELFCOMPRESS_ZSTD
Fangrui Song [Thu, 11 Aug 2022 04:46:09 +0000 (21:46 -0700)] 
elf.h: Add ELFCOMPRESS_ZSTD

From the approved generic ABI proposal
https://groups.google.com/g/generic-abi/c/satyPkuMisk
("Add new ch_type value: ELFCOMPRESS_ZSTD").

Reviewed-by: Florian Weimer <fweimer@redhat.com>
21 months agoinet: Turn __ivaliduser into a compatibility symbol
Florian Weimer [Wed, 10 Aug 2022 06:35:41 +0000 (08:35 +0200)] 
inet: Turn __ivaliduser into a compatibility symbol

It is not declared in a header file, and as the comment indicates,
it is not expected to be used.

21 months agox86: Fix `#define STRCPY` guard in strcpy-sse2.S
Noah Goldstein [Mon, 8 Aug 2022 03:26:21 +0000 (11:26 +0800)] 
x86: Fix `#define STRCPY` guard in strcpy-sse2.S

`#ifndef STPCPY` is incorrect for checking if `STRCPY` is already
defined.  It doesn't end up mattering as the whole check is
guarded by `#if IS_IN (libc)` but is incorrect none the less.

21 months agoelf: Replace `strcpy` call with `memcpy` [BZ #29454]
Noah Goldstein [Mon, 8 Aug 2022 03:26:22 +0000 (11:26 +0800)] 
elf: Replace `strcpy` call with `memcpy` [BZ #29454]

GCC normally does this optimization for us in
strlen_pass::handle_builtin_strcpy but only for optimized
build. To avoid needing to include strcpy.S in the rtld build to
support the debug build, just do the optimization by hand.

22 months agosoft-fp: Add fixhf[uns][di|si] and float[uns][di|si]hf
Kito Cheng [Tue, 5 Jul 2022 09:46:31 +0000 (17:46 +0800)] 
soft-fp: Add fixhf[uns][di|si] and float[uns][di|si]hf

Add more IEEE half conversion routines:

- Convert 32-bit/64-bit integer to IEEE half.
- Convert IEEE half to 32-bit/64-bit integer.

They are required by RISC-V _Float16 support, tested with RISC-V GCC.

22 months agoi386: Use cmpl instead of cmp
Adhemerval Zanella [Tue, 26 Jul 2022 11:44:37 +0000 (08:44 -0300)] 
i386: Use cmpl instead of cmp

Clang cannot assemble cmp in the AT&T dialect mode.

22 months agoi386: Use fldt instead of fld on e_logl.S
Adhemerval Zanella [Tue, 26 Jul 2022 11:42:50 +0000 (08:42 -0300)] 
i386: Use fldt instead of fld on e_logl.S

Clang cannot assemble fldt in the AT&T dialect mode.

22 months agoi386: Replace movzx with movzbl
Fangrui Song [Thu, 4 Aug 2022 21:06:50 +0000 (14:06 -0700)] 
i386: Replace movzx with movzbl

Similar to 6720d36b6623c5e48c070d86acf61198b33e144e for x86-64.

Clang cannot assemble movzx in the AT&T dialect mode.  Change movzx to
movzbl, which follows the AT&T dialect and is used elsewhere in the
file.

22 months agodlfcn: Pass caller pointer to static dlopen implementation (bug 29446)
Florian Weimer [Thu, 4 Aug 2022 15:54:48 +0000 (17:54 +0200)] 
dlfcn: Pass caller pointer to static dlopen implementation (bug 29446)

Fixes commit 0c1c3a771eceec46e66ce1183cf988e2303bd373 ("dlfcn: Move
dlopen into libc").

22 months agomalloc: Correct the documentation of the top_pad default
Florian Weimer [Thu, 4 Aug 2022 08:24:47 +0000 (10:24 +0200)] 
malloc: Correct the documentation of the top_pad default

DEFAULT_TOP_PAD is defined as 131072 in
sysdeps/generic/malloc-machine.h.

22 months agoi386: Remove RELA support
Adhemerval Zanella [Tue, 26 Jul 2022 11:40:52 +0000 (08:40 -0300)] 
i386: Remove RELA support

Now that prelink is not support, there is no need to keep supporting
rela for non bootstrap.

22 months agoarm: Remove RELA support
Adhemerval Zanella [Tue, 26 Jul 2022 11:39:20 +0000 (08:39 -0300)] 
arm: Remove RELA support

Now that prelink is not support, there is no need to keep supporting
rela for non bootstrap.

22 months agoRemove ldd libc4 support
Adhemerval Zanella [Tue, 2 Aug 2022 14:29:22 +0000 (11:29 -0300)] 
Remove ldd libc4 support

The older libc versions are obsolete for over twenty years now.

22 months agoAssume only FLAG_ELF_LIBC6 suport
Lucas A. M. Magalhaes [Thu, 14 Jul 2022 17:37:22 +0000 (14:37 -0300)] 
Assume only FLAG_ELF_LIBC6 suport

The older libc versions are obsolete for over twenty years now.
This patch removes the special flags for libc5 and libc4 and assumes
that all libraries cached are libc6 compatible and use FLAG_ELF_LIBC6.

Checked with a build for all affected architectures.

Co-authored-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
22 months agoRemove left over LD_LIBRARY_VERSION usages
Adhemerval Zanella [Tue, 2 Aug 2022 13:08:57 +0000 (10:08 -0300)] 
Remove left over LD_LIBRARY_VERSION usages

The environment variable was removed by
d2db60d8d830ef68c8d20a77ac3572d610aa40b1.

22 months agoLinux: Remove exit system call from _exit
Florian Weimer [Thu, 4 Aug 2022 04:17:50 +0000 (06:17 +0200)] 
Linux: Remove exit system call from _exit

exit only terminates the current thread, not the whole process, so it
is the wrong fallback system call in this context.  All supported
Linux versions implement the exit_group system call anyway.

22 months agoLoongArch: Add vdso support for gettimeofday.
caiyinyu [Thu, 14 Jul 2022 00:54:04 +0000 (08:54 +0800)] 
LoongArch: Add vdso support for gettimeofday.

22 months agoUpdate kernel version to 5.19 in header constant tests
Joseph Myers [Wed, 3 Aug 2022 16:31:58 +0000 (16:31 +0000)] 
Update kernel version to 5.19 in header constant tests

This patch updates the kernel version in the tests tst-mman-consts.py,
tst-mount-consts.py and tst-pidfd-consts.py to 5.18.  (There are no
new constants covered by these tests in 5.19, or in 5.17 or 5.18 in
the case of tst-mount-consts.py that previously used version 5.16,
that need any other header changes.)

Tested with build-many-glibcs.py.

22 months agoassert: Do not use stderr in libc-internal assert
Florian Weimer [Wed, 3 Aug 2022 09:41:53 +0000 (11:41 +0200)] 
assert: Do not use stderr in libc-internal assert

Redirect internal assertion failures to __libc_assert_fail, based on
based on __libc_message, which writes directly to STDERR_FILENO
and calls abort.  Also disable message translation and reword the
error message slightly (adjusting stdlib/tst-bz20544 accordingly).

As a result of these changes, malloc no longer needs its own
redefinition of __assert_fail.

__libc_assert_fail needs to be stubbed out during rtld dependency
analysis because the rtld rebuilds turn __libc_assert_fail into
__assert_fail, which is unconditionally provided by elf/dl-minimal.c.

This change is not possible for the public assert macro and its
__assert_fail function because POSIX requires that the diagnostic
is written to stderr.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
22 months agonptl: Remove uses of assert_perror
Florian Weimer [Wed, 3 Aug 2022 09:41:53 +0000 (11:41 +0200)] 
nptl: Remove uses of assert_perror

__pthread_sigmask cannot actually fail with valid pointer arguments
(it would need a really broken seccomp filter), and we do not check
for errors elsewhere.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
22 months agostdio: Clean up __libc_message after unconditional abort
Florian Weimer [Wed, 3 Aug 2022 09:41:53 +0000 (11:41 +0200)] 
stdio: Clean up __libc_message after unconditional abort

Since commit ec2c1fcefb200c6cb7e09553f3c6af8815013d83 ("malloc:
Abort on heap corruption, without a backtrace [BZ #21754]"),
__libc_message always terminates the process.  Since commit
a289ea09ea843ced6e5277c2f2e63c357bc7f9a3 ("Do not print backtraces
on fatal glibc errors"), the backtrace facility has been removed.
Therefore, remove enum __libc_message_action and the action
argument of __libc_message, and mark __libc_message as _No_return.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
22 months agoUpdate syscall lists for Linux 5.19
Joseph Myers [Tue, 2 Aug 2022 21:05:07 +0000 (21:05 +0000)] 
Update syscall lists for Linux 5.19

Linux 5.19 has no new syscalls, but enables memfd_secret in the uapi
headers for RISC-V.  Update the version number in syscall-names.list
to reflect that it is still current for 5.19 and regenerate the
arch-syscall.h headers with build-many-glibcs.py update-syscalls.

Tested with build-many-glibcs.py.

22 months agoUse Linux 5.19 in build-many-glibcs.py
Joseph Myers [Tue, 2 Aug 2022 14:09:24 +0000 (14:09 +0000)] 
Use Linux 5.19 in build-many-glibcs.py

This patch makes build-many-glibcs.py use Linux 5.19.

Tested with build-many-glibcs.py (host-libraries, compilers and glibcs
builds).

22 months agosocket: Check lengths before advancing pointer in CMSG_NXTHDR
Arjun Shankar [Tue, 2 Aug 2022 09:10:25 +0000 (11:10 +0200)] 
socket: Check lengths before advancing pointer in CMSG_NXTHDR

The inline and library functions that the CMSG_NXTHDR macro may expand
to increment the pointer to the header before checking the stride of
the increment against available space.  Since C only allows incrementing
pointers to one past the end of an array, the increment must be done
after a length check.  This commit fixes that and includes a regression
test for CMSG_FIRSTHDR and CMSG_NXTHDR.

The Linux, Hurd, and generic headers are all changed.

Tested on Linux on armv7hl, i686, x86_64, aarch64, ppc64le, and s390x.

[BZ #28846]

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
22 months agoDon't use unsupported format string in ld.so (bug 29427)
Andreas Schwab [Mon, 1 Aug 2022 14:30:15 +0000 (16:30 +0200)] 
Don't use unsupported format string in ld.so (bug 29427)

The dynamic loader does not support printf format strings that contain a
literal field width or precision, they have to be specified indirectly.

22 months agohtl: Let pthread_self and cancellability called early
Samuel Thibault [Thu, 28 Jul 2022 20:01:49 +0000 (22:01 +0200)] 
htl: Let pthread_self and cancellability called early

When applications redirect some functions they might get called before
libpthread is fully initialized.  They may still expected pthread_self
and cancellable functions to work, so cope with such calls in that
situation.

22 months agostdlib: Simplify arc4random_uniform
Adhemerval Zanella [Thu, 28 Jul 2022 12:18:01 +0000 (09:18 -0300)] 
stdlib: Simplify arc4random_uniform

It uses the bitmask with rejection [1], which calculates a mask
being the lowest power of two bounding the request upper bound,
successively queries new random values, and rejects values
outside the requested range.

Performance-wise, there is no much gain in trying to conserve
bits since arc4random is wrapper on getrandom syscall.  It should
be cheaper to just query a uint32_t value.  The algorithm also
avoids modulo and divide operations, which might be costly
depending of the architecture.

[1] https://www.pcg-random.org/posts/bounded-rands.html

Reviewed-by: Yann Droneaud <ydroneaud@opteya.com>
22 months agomalloc: Use __getrandom_nocancel during tcache initiailization
Florian Weimer [Mon, 1 Aug 2022 13:49:07 +0000 (15:49 +0200)] 
malloc: Use __getrandom_nocancel during tcache initiailization

Cancellation currently cannot happen at this point because dlopen
as used by the unwind link always performs additional allocations
for libgcc_s.so.1, even if it has been loaded already as a dependency
of the main executable.  But it seems prudent not to rely on this
quirk.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
22 months agoRemove spurious references to _dl_open_hook
Florian Weimer [Mon, 1 Aug 2022 13:49:07 +0000 (15:49 +0200)] 
Remove spurious references to _dl_open_hook

_dl_open_hook was removed in commit 466c1ea15f461edb8e3ffaf5d86d708
("dlfcn: Rework static dlopen hooks").

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
22 months agowcsmbs: Add missing test-c8rtomb/test-mbrtoc8 dependency
H.J. Lu [Fri, 29 Jul 2022 17:50:56 +0000 (10:50 -0700)] 
wcsmbs: Add missing test-c8rtomb/test-mbrtoc8 dependency

Make test-c8rtomb.out and test-mbrtoc8.out depend on $(gen-locales) for

  xsetlocale (LC_ALL, "de_DE.UTF-8");
  xsetlocale (LC_ALL, "zh_HK.BIG5-HKSCS");

Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
22 months agostdlib: Suppress gcc diagnostic that char8_t is a keyword in C++20 in uchar.h.
Tom Honermann [Sun, 24 Jul 2022 05:11:43 +0000 (01:11 -0400)] 
stdlib: Suppress gcc diagnostic that char8_t is a keyword in C++20 in uchar.h.

gcc 13 issues the following diagnostic for the uchar.h header when the
-Wc++20-compat option is enabled in C++ modes that do not enable char8_t
as a builtin type (C++17 and earlier by default; subject to _GNU_SOURCE
and the gcc -f[no-]char8_t option).
  warning: identifier ‘char8_t’ is a keyword in C++20 [-Wc++20-compat]
This change modifies the uchar.h header to suppress the diagnostic through
the use of '#pragma GCC diagnostic' directives for gcc 10 and later (the
-Wc++20-compat option was added in gcc version 10).  Unfortunately, a bug
in gcc currently prevents those directives from having the intended effect
as reported at https://gcc.gnu.org/PR106423.  A patch for that issue has
been submitted and is available in the email thread archive linked below.
  https://gcc.gnu.org/pipermail/gcc-patches/2022-July/598736.html

22 months agoOpen master branch for glibc 2.37 development glibc-2.36.9000
Carlos O'Donell [Sat, 30 Jul 2022 19:34:11 +0000 (15:34 -0400)] 
Open master branch for glibc 2.37 development

22 months agoCreate ChangeLog.old/ChangeLog.25. glibc-2.36
Carlos O'Donell [Fri, 29 Jul 2022 22:03:09 +0000 (18:03 -0400)] 
Create ChangeLog.old/ChangeLog.25.

22 months agoPrepare for glibc 2.36 release.
Carlos O'Donell [Fri, 29 Jul 2022 21:59:01 +0000 (17:59 -0400)] 
Prepare for glibc 2.36 release.

Update version.h, and include/features.h.

22 months agoUpdate install.texi, and regenerate INSTALL.
Carlos O'Donell [Fri, 29 Jul 2022 21:51:16 +0000 (17:51 -0400)] 
Update install.texi, and regenerate INSTALL.

22 months agoUpdate NEWS bug list.
Carlos O'Donell [Fri, 29 Jul 2022 21:49:20 +0000 (17:49 -0400)] 
Update NEWS bug list.

22 months agoUpdate libc.pot for 2.36 release.
Carlos O'Donell [Fri, 29 Jul 2022 17:01:35 +0000 (13:01 -0400)] 
Update libc.pot for 2.36 release.

22 months agotst-pidfd.c: UNSUPPORTED if we get EPERM on valid pidfd_getfd call
Mark Wielaard [Sun, 26 Jun 2022 20:17:17 +0000 (22:17 +0200)] 
tst-pidfd.c: UNSUPPORTED if we get EPERM on valid pidfd_getfd call

pidfd_getfd can fail for a valid pidfd with errno EPERM for various
reasons in a restricted environment. Use FAIL_UNSUPPORTED in that case.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
22 months agostdlib: Tuned down tst-arc4random-thread internal parameters
Adhemerval Zanella [Wed, 27 Jul 2022 11:59:15 +0000 (08:59 -0300)] 
stdlib: Tuned down tst-arc4random-thread internal parameters

With new arc4random implementation, the internal parameters might
require a lot of runtime and/or trigger some contention on older
kernels (which might trigger spurious timeout failures).

Also, since we are now testing getrandom entropy instead of an
userspace RNG, there is no much need to extensive testing.

With this change the tst-arc4random-thread goes from about 1m to
5s on a Ryzen 9 with 5.15.0-41-generic.

Checked on x86_64-linux-gnu.

Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
22 months agoLoongArch: Add greg_t and gregset_t.
caiyinyu [Wed, 27 Jul 2022 19:25:40 +0000 (03:25 +0800)] 
LoongArch: Add greg_t and gregset_t.

22 months agoLoongArch: Fix VDSO_HASH and VDSO_NAME.
caiyinyu [Wed, 27 Jul 2022 18:36:52 +0000 (02:36 +0800)] 
LoongArch: Fix VDSO_HASH and VDSO_NAME.

22 months agoriscv: Update rv64 libm test ulps
Darius Rad [Tue, 26 Jul 2022 12:54:55 +0000 (08:54 -0400)] 
riscv: Update rv64 libm test ulps

Generated on a Microsemi Polarfire Icicle Kit running Linux version
5.15.32.  Same ULPs were also produced on QEMU 5.2.0 running Linux
5.18.0.

22 months agoriscv: Update nofpu libm test ulps
Darius Rad [Tue, 26 Jul 2022 12:54:38 +0000 (08:54 -0400)] 
riscv: Update nofpu libm test ulps

22 months agoarc4random: simplify design for better safety
Jason A. Donenfeld [Tue, 26 Jul 2022 19:58:22 +0000 (21:58 +0200)] 
arc4random: simplify design for better safety

Rather than buffering 16 MiB of entropy in userspace (by way of
chacha20), simply call getrandom() every time.

This approach is doubtlessly slower, for now, but trying to prematurely
optimize arc4random appears to be leading toward all sorts of nasty
properties and gotchas. Instead, this patch takes a much more
conservative approach. The interface is added as a basic loop wrapper
around getrandom(), and then later, the kernel and libc together can
work together on optimizing that.

This prevents numerous issues in which userspace is unaware of when it
really must throw away its buffer, since we avoid buffering all
together. Future improvements may include userspace learning more from
the kernel about when to do that, which might make these sorts of
chacha20-based optimizations more possible. The current heuristic of 16
MiB is meaningless garbage that doesn't correspond to anything the
kernel might know about. So for now, let's just do something
conservative that we know is correct and won't lead to cryptographic
issues for users of this function.

This patch might be considered along the lines of, "optimization is the
root of all evil," in that the much more complex implementation it
replaces moves too fast without considering security implications,
whereas the incremental approach done here is a much safer way of going
about things. Once this lands, we can take our time in optimizing this
properly using new interplay between the kernel and userspace.

getrandom(0) is used, since that's the one that ensures the bytes
returned are cryptographically secure. But on systems without it, we
fallback to using /dev/urandom. This is unfortunate because it means
opening a file descriptor, but there's not much of a choice. Secondly,
as part of the fallback, in order to get more or less the same
properties of getrandom(0), we poll on /dev/random, and if the poll
succeeds at least once, then we assume the RNG is initialized. This is a
rough approximation, as the ancient "non-blocking pool" initialized
after the "blocking pool", not before, and it may not port back to all
ancient kernels, though it does to all kernels supported by glibc
(≥3.2), so generally it's the best approximation we can do.

The motivation for including arc4random, in the first place, is to have
source-level compatibility with existing code. That means this patch
doesn't attempt to litigate the interface itself. It does, however,
choose a conservative approach for implementing it.

Cc: Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>
Cc: Florian Weimer <fweimer@redhat.com>
Cc: Cristian Rodríguez <crrodriguez@opensuse.org>
Cc: Paul Eggert <eggert@cs.ucla.edu>
Cc: Mark Harris <mark.hsj@gmail.com>
Cc: Eric Biggers <ebiggers@kernel.org>
Cc: linux-crypto@vger.kernel.org
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
22 months agoLoongArch: Update NEWS and README for the LoongArch port.
caiyinyu [Tue, 19 Jul 2022 01:22:10 +0000 (09:22 +0800)] 
LoongArch: Update NEWS and README for the LoongArch port.

22 months agoLoongArch: Update build-many-glibcs.py for the LoongArch Port.
caiyinyu [Tue, 19 Jul 2022 01:22:09 +0000 (09:22 +0800)] 
LoongArch: Update build-many-glibcs.py for the LoongArch Port.

22 months agoLoongArch: Hard Float Support
caiyinyu [Tue, 19 Jul 2022 01:22:08 +0000 (09:22 +0800)] 
LoongArch: Hard Float Support

22 months agoLoongArch: Build Infrastructure
caiyinyu [Tue, 19 Jul 2022 01:22:07 +0000 (09:22 +0800)] 
LoongArch: Build Infrastructure

22 months agoLoongArch: Add ABI Lists
caiyinyu [Tue, 19 Jul 2022 01:22:06 +0000 (09:22 +0800)] 
LoongArch: Add ABI Lists

22 months agoLoongArch: Linux ABI
caiyinyu [Tue, 19 Jul 2022 01:20:51 +0000 (09:20 +0800)] 
LoongArch: Linux ABI

22 months agoLoongArch: Linux Syscall Interface
caiyinyu [Tue, 19 Jul 2022 01:20:50 +0000 (09:20 +0800)] 
LoongArch: Linux Syscall Interface