platform/upstream/glibc.git
14 months agohurd: Fix x86_64 _hurd_tls_fork
Sergey Bugaev [Wed, 17 May 2023 19:14:33 +0000 (22:14 +0300)]
hurd: Fix x86_64 _hurd_tls_fork

It is illegal to call thread_get_state () on mach_thread_self (), so
this codepath cannot be used as-is to fork the calling thread's TLS.
Fortunately we can use THREAD_SELF (aka %fs:0x0) to find out the value
of our fs_base without calling into the kernel.

Fixes: f6cf701efc61c9ad910372bda14b9a235db310a8
"hurd: Implement TLS for x86_64"

Checked on x86_64-gnu: fork () now works!

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230517191436.73636-8-bugaevc@gmail.com>

14 months agohurd: Make sure to not use tcb->self
Sergey Bugaev [Wed, 17 May 2023 19:14:32 +0000 (22:14 +0300)]
hurd: Make sure to not use tcb->self

Unlike sigstate->thread, tcb->self did not hold a Mach port reference on
the thread port it names. This means that the port can be deallocated,
and the name reused for something else, without anyone noticing. Using
tcb->self will then lead to port use-after-free.

Fortunately nothing was accessing tcb->self, other than it being
intially set to then-valid thread port name upon TCB initialization. To
assert that this keeps being the case without altering TCB layout,
rename self -> self_do_not_use, and stop initializing it.

Also, do not (re-)allocate a whole separate and unused stack for the
main thread, and just exit __pthread_setup early in this case.

Found upon attempting to use tcb->self and getting unexpected crashes.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230517191436.73636-7-bugaevc@gmail.com>

14 months agohurd: Use __mach_setup_thread_call ()
Sergey Bugaev [Wed, 17 May 2023 19:14:31 +0000 (22:14 +0300)]
hurd: Use __mach_setup_thread_call ()

...instead of mach_setup_thread (), which is unsuitable for setting up
function calls.

Checked on x86_64-gnu: the signal thread no longer crashes upon trying
to process a message.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230517191436.73636-6-bugaevc@gmail.com>

14 months agomach: Add __mach_setup_thread_call ()
Sergey Bugaev [Wed, 17 May 2023 19:14:30 +0000 (22:14 +0300)]
mach: Add __mach_setup_thread_call ()

This is just like mach_setup_thread (), but it's suitable for making the
thread call a function correctly, as opposed to explicitly setting the
thread's stack and instruction pointers to the given values. Internally,
it uses MACHINE_THREAD_STATE_SETUP_CALL.

Unlike mach_setup_thread (), which is exported via mach.h for the
benefit of the Hurd exec server, __mach_setup_thread_call () is private
to glibc for the time being.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230517191436.73636-5-bugaevc@gmail.com>

14 months agohurd: Use MACHINE_THREAD_STATE_SETUP_CALL
Sergey Bugaev [Wed, 17 May 2023 19:14:29 +0000 (22:14 +0300)]
hurd: Use MACHINE_THREAD_STATE_SETUP_CALL

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230517191436.73636-4-bugaevc@gmail.com>

14 months agomach: Define MACHINE_THREAD_STATE_SETUP_CALL
Sergey Bugaev [Wed, 17 May 2023 19:14:28 +0000 (22:14 +0300)]
mach: Define MACHINE_THREAD_STATE_SETUP_CALL

The existing two macros, MACHINE_THREAD_STATE_SET_PC and
MACHINE_THREAD_STATE_SET_SP, can be used to set program counter and the
stack pointer registers in a machine-specific thread state structure.

Useful as it is, this may not be enough to set up the thread to make a
function call, because the machine-specific ABI may impose additional
requirements. In particular, x86_64 ABI requires that upon function
entry, the stack pointer is 8 less than 16-byte aligned (sp & 15 == 8).

To deal with this, introduce a new macro,
MACHINE_THREAD_STATE_SETUP_CALL (), which sets both stack and
instruction pointers, and also applies any machine-specific requirements
to make a valid function call. The default implementation simply
forwards to MACHINE_THREAD_STATE_SET_PC and MACHINE_THREAD_STATE_SET_SP,
but on x86_64 we additionally align the stack pointer.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230517191436.73636-3-bugaevc@gmail.com>

14 months agoUse TASK_THREAD_TIMES_INFO_COUNT when calling task_info with TASK_THREAD_TIMES_INFO
Flavio Cruz [Wed, 17 May 2023 02:59:24 +0000 (22:59 -0400)]
Use TASK_THREAD_TIMES_INFO_COUNT when calling task_info with TASK_THREAD_TIMES_INFO

This hasn't caused any problems yet but we are passing a pointer to struct
task_thread_times_info which can cause problems if we populate over the
existing size of the struct.
Message-Id: <ZGRDDNcOM2hA3CuT@jupiter.tail36e24.ts.net>

14 months agoargp: Reformat Makefile.
Carlos O'Donell [Tue, 16 May 2023 15:59:25 +0000 (11:59 -0400)]
argp: Reformat Makefile.

Reflow Makefile.
Sort using scripts/sort-makefile-lines.py.

Code generation is changed as routines are linked in sorted order
as expected.

No regressions on x86_64 and i686.

14 months agostdlib: Avoid undefined behavior in stdlib/tst-labs
Florian Weimer [Tue, 16 May 2023 21:26:40 +0000 (23:26 +0200)]
stdlib: Avoid undefined behavior in stdlib/tst-labs

The last loop could attempt to overflow beyond INT_MAX on 32-bit
architectures.

Also switch to GNU style.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agostdlib: Use long long int in stdlib/tst-llabs
Florian Weimer [Tue, 16 May 2023 20:54:25 +0000 (22:54 +0200)]
stdlib: Use long long int in stdlib/tst-llabs

And adjust for GNU style.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agoUpdate kernel version to 6.3 in header constant tests
Joseph Myers [Tue, 16 May 2023 23:15:13 +0000 (23:15 +0000)]
Update kernel version to 6.3 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 6.3.  (There are no new
constants covered by these tests in 6.3 that need any other header
changes.)

Tested with build-many-glibcs.py.

14 months agoi386: Use pthread_barrier for synchronization on tst-bz21269
DJ Delorie [Tue, 16 May 2023 02:50:35 +0000 (22:50 -0400)]
i386: Use pthread_barrier for synchronization on tst-bz21269

So I was able to reproduce the hangs in the original source, and debug
it, and fix it.  In doing so, I realized that we can't use anything
complex to trigger the thread because that "anything" might also cause
the expected segfault and force everything out of sync again.

Here's what I ended up with, and it doesn't seem to hang where the
original one hung quite often (in a tight while..end loop).  The key
changes are:

1. Calls to futex are error checked, with retries, to ensure that the
   futexes are actually doing what they're supposed to be doing.  In the
   original code, nearly every futex call returned an error.

2. The main loop has checks for whether the thread ran or not, and
   "unlocks" the thread if it didn't (this is how the original source
   hangs).

Note: the usleep() is not for timing purposes, but just to give the
kernel an excuse to run the other thread at that time.  The test will
not hang without it, but is more likely to test the right bugfix
if the usleep() is present.

14 months agostdlib: Add testcases for llabs(). (BZ #30263)
Joe Simmons-Talbott [Tue, 4 Apr 2023 13:57:47 +0000 (09:57 -0400)]
stdlib: Add testcases for llabs(). (BZ #30263)

Test minimum and maximum long long values, zero, 32bit crossover points, and
part of the range of long long values.  Use '-fno-builtin' to ensure we are
testing the implementation.

Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
14 months agostdlib: Add testcases for labs(). (BZ #30263)
Joe Simmons-Talbott [Tue, 4 Apr 2023 13:57:46 +0000 (09:57 -0400)]
stdlib: Add testcases for labs(). (BZ #30263)

Test minimum and maximum long values, zero, and part of the range
of long values.  Use '-fno-builtin' to ensure we are testing the
implementation.

Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
14 months agostdlib: Add testcases for abs(). (BZ #30263)
Joe Simmons-Talbott [Tue, 4 Apr 2023 13:57:45 +0000 (09:57 -0400)]
stdlib: Add testcases for abs(). (BZ #30263)

Test minimum and maximum int values, zero, and part of the range
of int values.  Use '-fno-builtin' to ensure we are testing the
implementation.

Reviewed-by: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
14 months agohurd: Fix computing user stack pointer
Sergey Bugaev [Mon, 15 May 2023 08:33:23 +0000 (11:33 +0300)]
hurd: Fix computing user stack pointer

Fixes b574ae0a2876ee94e4fe617f878407bf818c2df0
"hurd: Implement sigreturn for x86_64"

Checked on x86_64-gnu.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230515083323.1358039-5-bugaevc@gmail.com>

14 months agohurd: Fix sc_i386_thread_state layout
Sergey Bugaev [Mon, 15 May 2023 08:33:22 +0000 (11:33 +0300)]
hurd: Fix sc_i386_thread_state layout

The real i386_thread_state Mach structure has an alignment of 8 on
x86_64. However, in struct sigcontext, the compiler was packing sc_gs
(which is the first member of sc_i386_thread_state) into the same 8-byte
slot as sc_error; this resulted in the rest of sc_i386_thread_state
members having wrong offsets relative to each other, and the overall
sc_i386_thread_state layout mismatching that of i386_thread_state.

Fix this by explicitly adding the required padding members, and
statically asserting that this results in the desired alignment.

The same goes for sc_i386_float_state.

Checked on x86_64-gnu.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230515083323.1358039-4-bugaevc@gmail.com>

14 months agohurd: Align signal stack pointer after allocating stackframe
Sergey Bugaev [Mon, 15 May 2023 08:33:21 +0000 (11:33 +0300)]
hurd: Align signal stack pointer after allocating stackframe

sizeof (*stackframe) appears to be divisible by 16, but we should not
rely on that. So make sure to leave enough space for the stackframe
first, and then align the final pointer at 16 bytes.

Checked on x86_64-gnu.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230515083323.1358039-3-bugaevc@gmail.com>

14 months agohurd: Fix aligning signal stack pointer
Sergey Bugaev [Mon, 15 May 2023 08:33:20 +0000 (11:33 +0300)]
hurd: Fix aligning signal stack pointer

Fixes 60f9bf974694d50daf58d46347b06a5975ac5ddd
"hurd: Port trampoline.c to x86_64"

Checked on x86_64-gnu.

Reported-by: Bruno Haible <bruno@clisp.org>
Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230515083323.1358039-2-bugaevc@gmail.com>

14 months agowcsmbs: Reformat Makefile.
Carlos O'Donell [Fri, 12 May 2023 00:45:35 +0000 (20:45 -0400)]
wcsmbs: Reformat Makefile.

Reflow Makefile.
Sort using scripts/sort-makefile-lines.py.

No code generation changes observed in binary artifacts.
No regressions on x86_64 and i686.

14 months agolinux: Reformat Makefile.
Carlos O'Donell [Thu, 11 May 2023 19:01:18 +0000 (15:01 -0400)]
linux: Reformat Makefile.

Reflow Makefile.
Sort using scripts/sort-makefile-lines.py.

No code generation changes observed in binary artifacts.
No regressions on x86_64 and i686.

14 months agostdlib: Reformat Makefile.
Carlos O'Donell [Thu, 11 May 2023 18:35:55 +0000 (14:35 -0400)]
stdlib: Reformat Makefile.

Reflow Makefile.
Sort using scripts/sort-makefile-lines.py.

No code generation changes observed in binary artifacts.
No regressions on x86_64 and i686.

14 months agostdio-common: Reformat Makefile.
Carlos O'Donell [Thu, 11 May 2023 18:00:41 +0000 (14:00 -0400)]
stdio-common: Reformat Makefile.

Reflow Makefile.
Sort using scripts/sort-makefile-lines.py.

Code generation is changed as routines are linked in sorted order
as expected.

No regressions on x86_64 and i686.

14 months agosocket: Reformat Makefile.
Carlos O'Donell [Thu, 11 May 2023 17:14:51 +0000 (13:14 -0400)]
socket: Reformat Makefile.

Reflow Makefile.
Sort using scripts/sort-makefile-lines.py.

Code generation is changed as routines are linked in sorted order
as expected.

No regressions on x86_64 and i686.

14 months agomisc: Reformat Makefile.
Carlos O'Donell [Thu, 11 May 2023 11:04:11 +0000 (07:04 -0400)]
misc: Reformat Makefile.

Reflow Makefile.
Sort using scripts/sort-makefile-lines.py.

Code generation is changed as routines are linked in sorted order
as expected.

No regressions on x86_64 and i686.

14 months agodebug: Reformat Makefile.
Carlos O'Donell [Thu, 11 May 2023 10:46:27 +0000 (06:46 -0400)]
debug: Reformat Makefile.

Fix list terminator whitspace.
Sort using scripts/sort-makefile-lines.py.

No code generation changes observed in binary artifacts.
No regressions on x86_64 and i686.

14 months agoelf: Reformat Makefile.
Carlos O'Donell [Wed, 10 May 2023 17:40:17 +0000 (13:40 -0400)]
elf: Reformat Makefile.

Fix list terminator whitspace.
Sort using scripts/sort-makefile-lines.py.

No code generation changes observed in binary artifacts.
No regressions on x86_64 and i686.

14 months agolibio: Add __nonnull for FILE * arguments of fclose and freopen
Xi Ruoyao [Fri, 21 Apr 2023 06:21:33 +0000 (14:21 +0800)]
libio: Add __nonnull for FILE * arguments of fclose and freopen

Calling fclose or freopen with a null FILE * is undefined behavior, and
doing so in practice will cause a SIGSEGV.  So it seems suitable for
__nonnull.

This will help the compiler to warn for some buggy code, like
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109570.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
14 months agonss: Reconcile conditional declaration and use of `is_nscd'
Arjun Shankar [Thu, 11 May 2023 12:54:11 +0000 (14:54 +0200)]
nss: Reconcile conditional declaration and use of `is_nscd'

This change harmonizes the declaration and use of `is_nscd' and fixes a
build failure with the "--enable-static-nss --enable-nscd"
configuration options due to `is_nscd' being used undeclared.

The purpose of `is_nscd' is to avoid (nss <-> nscd) recursion in
dynamically linked libc (SHARED) that is nscd-aware (USE_NSCD), and so
its declaration and use should be guarded by the definition of those
macros.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
14 months agoUpdate syscall lists for Linux 6.3
Joseph Myers [Mon, 15 May 2023 22:26:56 +0000 (22:26 +0000)]
Update syscall lists for Linux 6.3

Linux 6.3 has no new syscalls.  Update the version number in
syscall-names.list to reflect that it is still current for 6.3.

Tested with build-many-glibcs.py.

14 months agohurd: rule out some mach headers when generating errno.h
Samuel Thibault [Wed, 10 May 2023 23:51:10 +0000 (01:51 +0200)]
hurd: rule out some mach headers when generating errno.h

While mach/kern_return.h happens to pull mach/machine/kern_return.h,
mach/machine/boolean.h, and mach/machine/vm_types.h (and realpath-ing them
exposes the machine-specific machine symlink content), those headers do not
actually define anything machine-specific for the content of errno.h.

So we can just rule out these machine-specific from the dependency
comment.

14 months agoStop checking if MiG supports retcode.
Flavio Cruz [Wed, 10 May 2023 05:20:34 +0000 (01:20 -0400)]
Stop checking if MiG supports retcode.

We already did the same change for Hurd
(https://git.savannah.gnu.org/cgit/hurd/hurd.git/commit/?id=ef5924402864ef049f40a39e73967628583bc1a4)

Due to MiG requiring the subsystem to be defined early in order to know the
size of a port, this was causing a division by zero error during ./configure.
We could have just move subsystem to the top of the snippet, however it is
simpler to just remove the check given that we have no plans to use some other
MiG anyway.

HAVE_MIG_RETCODE is removed completely since this will be a no-op either
way (compiling against old Hurd headers will work the same, new Hurd
headers will result in the same stubs since retcode is a no-op).
Message-Id: <ZFspor91aoMwbh9T@jupiter.tail36e24.ts.net>

14 months agoAdded Redirects to longdouble error functions [BZ #29033]
Sachin Monga [Wed, 10 May 2023 17:59:21 +0000 (13:59 -0400)]
Added Redirects to longdouble error functions [BZ #29033]

This patch redirects the error functions to the appropriate
longdouble variants which enables the compiler to optimize
for the abi ieeelongdouble.

Signed-off-by: Sachin Monga <smonga@linux.ibm.com>
14 months agonptl: Reformat Makefile.
Carlos O'Donell [Tue, 18 Apr 2023 15:02:09 +0000 (11:02 -0400)]
nptl: Reformat Makefile.

Reflow all long lines adding comment terminators.
Sort all reflowed text using scripts/sort-makefile-lines.py.

No code generation changes observed in binary artifacts.
No regressions on x86_64 and i686.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agoscripts: Add sort-makefile-lines.py to sort Makefile variables.
Carlos O'Donell [Tue, 18 Apr 2023 15:02:55 +0000 (11:02 -0400)]
scripts: Add sort-makefile-lines.py to sort Makefile variables.

The scripts/sort-makefile-lines.py script sorts Makefile variables
according to project expected order.

The script can be used like this:

$ scripts/sort-makefile-lines.py < elf/Makefile > elf/Makefile.tmp
$ mv elf/Makefile.tmp elf/Makefile

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agodlopen: skip debugger notification for DSO loaded from sprof (bug 30258)
Andreas Schwab [Thu, 23 Mar 2023 15:46:20 +0000 (16:46 +0100)]
dlopen: skip debugger notification for DSO loaded from sprof (bug 30258)

Avoid inconsistent state in the debugger interface.

14 months agoaligned_alloc: conform to C17
DJ Delorie [Tue, 21 Mar 2023 04:46:43 +0000 (00:46 -0400)]
aligned_alloc: conform to C17

This patch adds the strict checking for power-of-two alignments
in aligned_alloc(), and updates the manual accordingly.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>
14 months agotestsuite: stdlib/isomac.c: fix REQUIREMENTS
наб [Thu, 4 May 2023 20:25:57 +0000 (22:25 +0200)]
testsuite: stdlib/isomac.c: fix REQUIREMENTS

All of the mentioned variables are gone. gcc is just the default and
argv[1] can be used instead. /usr/include isn't hard-coded and you can
pass argv[2] with -I... to adjust.

Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
14 months agomanual: Remove unsupported line breaks in waiting-with-clock section
Florian Weimer [Mon, 8 May 2023 11:14:22 +0000 (13:14 +0200)]
manual: Remove unsupported line breaks in waiting-with-clock section

The argument to @deftypefun must be on a single line.
Also add the missing @safety for sem_clockwait.

Reported-by: Nilgün Belma Bugüner <nillguine@gmail.com>
14 months agoEnable new device_open_new RPC in libmachuser.
Flavio Cruz [Sun, 7 May 2023 17:13:23 +0000 (13:13 -0400)]
Enable new device_open_new RPC in libmachuser.

Message-Id: <ZFfcM3DP+x7VHpWA@jupiter.tail36e24.ts.net>

14 months agoRevert "riscv: Resolve symbols directly for symbols with STO_RISCV_VARIANT_CC."
Florian Weimer [Sun, 7 May 2023 12:16:03 +0000 (14:16 +0200)]
Revert "riscv: Resolve symbols directly for symbols with STO_RISCV_VARIANT_CC."

This reverts commit 117e8b341c5c0ace8d65feeef136fececb3fdc9c.

Reason for revert: Causes elf/tst-glibcelf and elf/tst-relro-*
to fail on all architectures.

14 months agoUpdate hurd/hurdselect.c to be more portable.
Flavio Cruz [Sat, 6 May 2023 20:55:10 +0000 (16:55 -0400)]
Update hurd/hurdselect.c to be more portable.

Summary of changes:
- Use BAD_TYPECHECK to perform type checking in a cleaner way.
  BAD_TYPECHECK is moved into sysdeps/mach/rpc.h to avoid duplication.
- Remove assertions for mach_msg_type_t since those won't work for
  x86_64.
- Update message structs to use mach_msg_type_t directly.
- Use designated initializers.
Message-Id: <ZFa+roan3ioo0ONM@jupiter.tail36e24.ts.net>

14 months agohurd: Fix ld.so name
Samuel Thibault [Sat, 6 May 2023 19:00:56 +0000 (21:00 +0200)]
hurd: Fix ld.so name

This was set to ld-x86-64.so.1 in gcc.

14 months agohurd: Add ioperm symbol on x86_64
Samuel Thibault [Sat, 6 May 2023 16:59:32 +0000 (18:59 +0200)]
hurd: Add ioperm symbol on x86_64

14 months agotime: Remove alloca() from getdate
Joe Simmons-Talbott [Wed, 3 May 2023 16:40:05 +0000 (12:40 -0400)]
time: Remove alloca() from getdate

Reduce the usage of alloca() to the bare minimum to avoid the potential
for stack overflow.  Use __strndup to simplify the code.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agoaarch64: More configure checks for libmvec
Szabolcs Nagy [Fri, 5 May 2023 08:01:58 +0000 (09:01 +0100)]
aarch64: More configure checks for libmvec

Check assembler and linker support too, not just SVE ACLE in the
compiler, since variant PCS requires at least binutils 2.32.1.

14 months agoaarch64: SVE ACLE configure test cleanups
Szabolcs Nagy [Thu, 4 May 2023 09:14:43 +0000 (10:14 +0100)]
aarch64: SVE ACLE configure test cleanups

Use more idiomatic configure test for better autoconf cache and logs.

14 months agohppa: Fix 'concurrency' typo in comment
Sam James [Fri, 5 May 2023 09:12:07 +0000 (10:12 +0100)]
hppa: Fix 'concurrency' typo in comment

Signed-off-by: Sam James <sam@gentoo.org>
14 months agoUpdate hurd/intr-msg.c to be more portable
Flavio Cruz [Thu, 4 May 2023 04:08:12 +0000 (00:08 -0400)]
Update hurd/intr-msg.c to be more portable

Summary of the changes:
- Introduce BAD_TYPECHECK from MiG to make it simpler to do type
  checking.
- Replace int type with mach_msg_type_t. This assumes that
  mach_msg_type_t is always the same size as int which is not true for
  x86_64.
- Calculate the size and align using PTR_ALIGN_UP, which is a bit
  cleaner and similar to what we do elsewhere.
- Define mach_msg_type_t to check using designated initializers.
Message-Id: <ZFMvrIkvoCSxqB/C@jupiter.tail36e24.ts.net>

14 months agoUpdate sysdeps/mach/hurd/ioctl.c to make it more portable
Flavio Cruz [Thu, 4 May 2023 04:06:46 +0000 (00:06 -0400)]
Update sysdeps/mach/hurd/ioctl.c to make it more portable

Summary of the changes:
- Update msg_align to use ALIGN_UP like we have done in previous
  patches. Use it below whenever necessary to avoid repeating the same
  alignment logic.
- Define BAD_TYPECHECK to make it easier to do type checking in a few
  places below.
- Update io2mach_type to use designated initializers.
- Make RetCodeType use mach_msg_type_t. mach_msg_type_t is 8 byte for
  x86_64, so this make it portable.
- Also call msg_align for _IOT_COUNT2/_IOT_TYPE2 since it is more
  correct.
Message-Id: <ZFMvVsuFKwIy2dUS@jupiter.tail36e24.ts.net>

14 months agoaarch64: fix SVE ACLE check for bootstrap glibc builds
Szabolcs Nagy [Thu, 4 May 2023 08:48:03 +0000 (09:48 +0100)]
aarch64: fix SVE ACLE check for bootstrap glibc builds

arm_sve.h depends on stdint.h but that relies on libc headers unless
compiled in freestanding mode.  Without this change a bootstrap glibc
build (that uses a compiler without installed libc headers) failed with

checking for availability of SVE ACLE... In file included from [...]/arm_sve.h:28,
                 from conftest.c:1:
[...]/stdint.h:9:16: fatal error: stdint.h: No such file or directory
    9 | # include_next <stdint.h>
      |                ^~~~~~~~~~
compilation terminated.
configure: error: mathvec is enabled but compiler does not have SVE ACLE. [...]

14 months agoEnable libmvec support for AArch64
Joe Ramsay [Wed, 12 Apr 2023 13:37:49 +0000 (14:37 +0100)]
Enable libmvec support for AArch64

This patch enables libmvec on AArch64. The proposed change is mainly
implementing build infrastructure to add the new routines to ABI,
tests and benchmarks. I have demonstrated how this all fits together
by adding implementations for vector cos, in both single and double
precision, targeting both Advanced SIMD and SVE.

The implementations of the routines themselves are just loops over the
scalar routine from libm for now, as we are more concerned with
getting the plumbing right at this point. We plan to contribute vector
routines from the Arm Optimized Routines repo that are compliant with
requirements described in the libmvec wiki.

Building libmvec requires minimum GCC 10 for SVE ACLE. To avoid raising
the minimum GCC by such a big jump, we allow users to disable libmvec
if their compiler is too old.

Note that at this point users have to manually call the vector math
functions. This seems to be acceptable to some downstream users.

Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
14 months agohurd: Enable x86_64 build script
Samuel Thibault [Tue, 2 May 2023 19:40:31 +0000 (21:40 +0200)]
hurd: Enable x86_64 build script

This now passes crossbuilds.

14 months agohurd 64bit: Make dev_t word type
Samuel Thibault [Tue, 2 May 2023 19:29:26 +0000 (21:29 +0200)]
hurd 64bit: Make dev_t word type

dev_t are 64bit on Linux ports, so better increase their size on 64bit
Hurd. It happens that this helps with BZ 23084 there: st_dev has type fsid_t
(quad) and is specified by POSIX to have type dev_t. Making dev_t 64bit
makes these match.

14 months agomalloc: Really fix tst-memalign-3 link against threads
Samuel Thibault [Tue, 2 May 2023 19:26:09 +0000 (21:26 +0200)]
malloc: Really fix tst-memalign-3 link against threads

All the tst malloc variants need the thread linking flags.

14 months agomalloc: Fix tst-memalign-3 link against threads
Samuel Thibault [Tue, 2 May 2023 19:07:31 +0000 (21:07 +0200)]
malloc: Fix tst-memalign-3 link against threads

14 months agoUse GCC 13 branch, Linux 6.3 in build-many-glibcs.py
Joseph Myers [Tue, 2 May 2023 13:57:58 +0000 (13:57 +0000)]
Use GCC 13 branch, Linux 6.3 in build-many-glibcs.py

This patch updates build-many-glibcs.py to use Linux 6.3 and GCC 13
branch by default.

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

14 months agoMark various cold functions as __COLD
Sergey Bugaev [Sat, 29 Apr 2023 13:12:23 +0000 (16:12 +0300)]
Mark various cold functions as __COLD

GCC docs explicitly list perror () as a good candidate for using
__attribute__ ((cold)). So apply __COLD to perror () and similar
functions.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131223.2507236-3-bugaevc@gmail.com>

14 months agoFix regex type usage
наб [Sun, 23 Apr 2023 00:55:05 +0000 (02:55 +0200)]
Fix regex type usage

include/regex.h had not been updated during the int -> Idx transition,
and the prototypes don't matched the definitions in regexec.c.

In regcomp.c, most interfaces were updated for Idx, except for two ones
guarded by #if _LIBC.

Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agohurd 64bit: Fix struct msqid_ds and shmid_ds fields
Samuel Thibault [Mon, 1 May 2023 12:51:22 +0000 (14:51 +0200)]
hurd 64bit: Fix struct msqid_ds and shmid_ds fields

The standards want msg_lspid/msg_lrpid/shm_cpid/shm_lpid to be pid_t, see BZ
23083 and 23085.

We can leave them __rpc_pid_t on i386 for ABI compatibility, but avoid
hitting the issue on 64bit.

14 months agohurd 64bit: Fix ipc_perm fields types
Samuel Thibault [Mon, 1 May 2023 12:40:12 +0000 (14:40 +0200)]
hurd 64bit: Fix ipc_perm fields types

The standards want uid/cuid to be uid_t, gid/cgid to be gid_t and mode to be
mode_t, see BZ 23082.

We can leave them short ints on i386 for ABI compatibility, but avoid
hitting the issue on 64bit.

bits/ipc.h ends up being exactly the same in sysdeps/gnu/ and
sysdeps/unix/sysv/linux/, so remove the latter.

14 months agohurd 64bit: Fix flock fields types
Samuel Thibault [Mon, 1 May 2023 12:35:32 +0000 (14:35 +0200)]
hurd 64bit: Fix flock fields types

The standards want l_type and l_whence to be short ints, see BZ 23081.

We can leave them ints on i386 for ABI compatibility, but avoid hitting the
issue on 64bit.

14 months agohurd 64bit: Add data for check-c++-types
Samuel Thibault [Mon, 1 May 2023 12:27:48 +0000 (14:27 +0200)]
hurd 64bit: Add data for check-c++-types

14 months agohurd 64bit: Fix pthread_t/thread_t type to long
Samuel Thibault [Mon, 1 May 2023 12:22:31 +0000 (14:22 +0200)]
hurd 64bit: Fix pthread_t/thread_t type to long

So that they can be trivially cast to pointer type, like with nptl.

14 months agosocket: Fix tst-cmsghdr-skeleton.c use of cmsg_len
Samuel Thibault [Mon, 1 May 2023 12:48:06 +0000 (14:48 +0200)]
socket: Fix tst-cmsghdr-skeleton.c use of cmsg_len

cmsg_len is supposed to be socklen_t according to standards, but it was made
size_t on Linux, see BZ 16919. For ports that have it socklen_t, SIZE_MAX is
too large. We can however explicitly cast it to the type of cmsg_len so it
will fit according to that type.

14 months agohurd 64bit: Add missing data file for check-localplt test
Samuel Thibault [Mon, 1 May 2023 11:38:57 +0000 (13:38 +0200)]
hurd 64bit: Add missing data file for check-localplt test

14 months agohurd 64bit: Add missing libanl
Samuel Thibault [Mon, 1 May 2023 11:35:54 +0000 (13:35 +0200)]
hurd 64bit: Add missing libanl

The move of libanl to libc was in glibc 2.34 for nptl only.

14 months agohurd: Also XFAIL missing SA_NOCLDWAIT on 64bit
Samuel Thibault [Mon, 1 May 2023 11:01:46 +0000 (13:01 +0200)]
hurd: Also XFAIL missing SA_NOCLDWAIT on 64bit

14 months agohurd: Fix tst-writev test
Samuel Thibault [Mon, 1 May 2023 11:00:50 +0000 (13:00 +0200)]
hurd: Fix tst-writev test

There is no compile-time IOV_MAX constraint on GNU/Hurd

14 months agonptl: move tst-x86-64-tls-1 to nptl-only tests
Samuel Thibault [Mon, 1 May 2023 10:59:12 +0000 (12:59 +0200)]
nptl: move tst-x86-64-tls-1 to nptl-only tests

It is essentially nptl-only.

14 months agohurd: Add expected abilist files for x86_64
Sergey Bugaev [Sat, 29 Apr 2023 20:18:20 +0000 (23:18 +0300)]
hurd: Add expected abilist files for x86_64

These were created by creating stub files, running 'make update-abi',
and reviewing the results.

Also, set baseline ABI to GLIBC_2.38, the (upcoming) first glibc
release to first have x86_64-gnu support.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
14 months agohurd: Replace reply port with a dead name on failed interruption
Sergey Bugaev [Sat, 29 Apr 2023 20:18:19 +0000 (23:18 +0300)]
hurd: Replace reply port with a dead name on failed interruption

If we're trying to interrupt an interruptible RPC, but the server fails
to respond to our __interrupt_operation () call, we instead destroy the
reply port we were expecting the reply to the RPC on.

Instead of deallocating the name completely, replace it with a dead
name, so the name won't get reused for some other right, and deallocate
it in _hurd_intr_rpc_mach_msg once we return from the signal handler.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429201822.2605207-4-bugaevc@gmail.com>

14 months agoDefine __mig_strlen to support dynamically sized strings in hurd RPCs
Flavio Cruz [Mon, 1 May 2023 00:09:49 +0000 (20:09 -0400)]
Define __mig_strlen to support dynamically sized strings in hurd RPCs

We make lib{mach,hurd}user.so only call __mig_strlen which can be
relocated before libc.so is relocated, similar to what is done with
__mig_memcpy.
Message-Id: <ZE8DTRDpY2hpPZlJ@jupiter.tail36e24.ts.net>

14 months agomach: Disable 32bit compatibility mode
Samuel Thibault [Mon, 1 May 2023 00:20:59 +0000 (02:20 +0200)]
mach: Disable 32bit compatibility mode

32bit compatibility is the default for now with 64bit gnumach, until 64bit
userland is ready.  Here we however want native 64bit userland anyway.

14 months agohurd: Make it possible to call memcpy very early
Sergey Bugaev [Sat, 29 Apr 2023 20:18:21 +0000 (23:18 +0300)]
hurd: Make it possible to call memcpy very early

Normally, in static builds, the first code that runs is _start, in e.g.
sysdeps/x86_64/start.S, which quickly calls __libc_start_main, passing
it the argv etc. Among the first things __libc_start_main does is
initializing the tunables (based on env), then CPU features, and then
calls _dl_relocate_static_pie (). Specifically, this runs ifunc
resolvers to pick, based on the CPU features discovered earlier, the
most suitable implementation of "string" functions such as memcpy.

Before that point, calling memcpy (or other ifunc-resolved functions)
will not work.

In the Hurd port, things are more complex. In order to get argv/env for
our process, glibc normally needs to do an RPC to the exec server,
unless our args/env are already located on the stack (which is what
happens to bootstrap processes spawned by GNU Mach). Fetching our
argv/env from the exec server has to be done before the call to
__libc_start_main, since we need to know what our argv/env are to pass
them to __libc_start_main.

On the other hand, the implementation of the RPC (and other initial
setup needed on the Hurd before __libc_start_main can be run) is not
very trivial. In particular, it may (and on x86_64, will) use memcpy.
But as described above, calling memcpy before __libc_start_main can not
work, since the GOT entry for it is not yet initialized at that point.

Work around this by pre-filling the GOT entry with the baseline version
of memcpy, __memcpy_sse2_unaligned. This makes it possible for early
calls to memcpy to just work. The initial value of the GOT entry is
unused on x86_64, and changing it won't interfere with the relocation
being performed later: once _dl_relocate_static_pie () is called, the
baseline version will get replaced with the most suitable one, and that
is what subsequent calls of memcpy are going to call.

Checked on x86_64-gnu.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429201822.2605207-6-bugaevc@gmail.com>

14 months agohurd: Implement longjmp for x86_64
Sergey Bugaev [Sat, 29 Apr 2023 20:18:18 +0000 (23:18 +0300)]
hurd: Implement longjmp for x86_64

Checked on x86_64-gnu.

[samuel.thibault@ens-lyon.org: Restored same comments as on i386]
Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429201822.2605207-3-bugaevc@gmail.com>

14 months agohurd: Implement sigreturn for x86_64
Sergey Bugaev [Sat, 29 Apr 2023 20:18:17 +0000 (23:18 +0300)]
hurd: Implement sigreturn for x86_64

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429201822.2605207-2-bugaevc@gmail.com>

14 months agoMake __mach_msg_destroy portable for x86_64
Flavio Cruz [Sun, 30 Apr 2023 03:02:20 +0000 (23:02 -0400)]
Make __mach_msg_destroy portable for x86_64

We need to align on uintptr_t to make this work for x86_64,
otherwise things will go wrong when RPCs return errors.
Message-Id: <ZE3aPH7uCEDti47H@jupiter.tail36e24.ts.net>

14 months agohurd: Mark error functions as __COLD
Sergey Bugaev [Sat, 29 Apr 2023 13:12:22 +0000 (16:12 +0300)]
hurd: Mark error functions as __COLD

This should hopefully hint the compiler that they are unlikely
to be called.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131223.2507236-2-bugaevc@gmail.com>

14 months agocdefs.h: Define __COLD
Sergey Bugaev [Thu, 27 Apr 2023 14:42:11 +0000 (17:42 +0300)]
cdefs.h: Define __COLD

This expands to __attribute__ ((cold)) when supported. It should be
used to mark up functions that are invoked rarely.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
14 months agohurd: Fix FS_RETRY_MAGICAL "machtype" handling
Sergey Bugaev [Sat, 29 Apr 2023 13:13:54 +0000 (16:13 +0300)]
hurd: Fix FS_RETRY_MAGICAL "machtype" handling

We need to set file_name, not update retryname. This is what the other
branches do.

Before this change, any attempt to access such a file would segfault due
to file_name being unset:

$ settrans -ac /tmp/my-machtype /hurd/magic machtype
$ cat /tmp/my-machtype
Segmentation fault

Checked on i686-gnu.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131354.2507443-7-bugaevc@gmail.com>

14 months agohurd: Respect existing FD_CLOEXEC in S_msg_set_fd
Sergey Bugaev [Sat, 29 Apr 2023 13:13:53 +0000 (16:13 +0300)]
hurd: Respect existing FD_CLOEXEC in S_msg_set_fd

If the process has set the close-on-exec flag for the file descriptor,
it expects the file descriptor to get closed on exec, even if we replace
what the file descriptor refers to.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131354.2507443-6-bugaevc@gmail.com>

14 months agohurd: Don't leak the auth port in msg* RPCs
Sergey Bugaev [Sat, 29 Apr 2023 13:13:52 +0000 (16:13 +0300)]
hurd: Don't leak the auth port in msg* RPCs

The leak can be easily reproduced (and observed) using the portinfo
tool:

$ portinfo -v $$ | grep task
    36: send task(1577)(self) (refs: 127)
$ portinfo -v $$ | grep task
    36: send task(1577)(self) (refs: 253)
$ portinfo -v $$ | grep task
    36: send task(1577)(self) (refs: 379)
$ portinfo -v $$ | grep task
    36: send task(1577)(self) (refs: 505)
$ portinfo -v $$ | grep task
    36: send task(1577)(self) (refs: 631)

Checked on i686-gnu.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131354.2507443-5-bugaevc@gmail.com>

14 months agohurd: Make _exit work during early boot-up
Sergey Bugaev [Sat, 29 Apr 2023 13:13:51 +0000 (16:13 +0300)]
hurd: Make _exit work during early boot-up

If any of the early boot-up tasks calls exit () or returns from main (),
terminate it properly instead of crashing on trying to dereference
_hurd_ports and getting forcibly terminated by the kernel.

We sadly cannot make the __USEPORT macro do the check for _hurd_ports
being unset, because it evaluates to the value of the expression
provided as the second argument, and that can be of any type; so there
is no single suitable fallback value for the macro to evaluate to in
case _hurd_ports is unset. Instead, each use site that wants to care for
this case will have to do its own checking.

Checked on x86_64-gnu.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131354.2507443-4-bugaevc@gmail.com>

14 months agohurd: Mark various conditions as unlikely
Sergey Bugaev [Sat, 29 Apr 2023 13:13:50 +0000 (16:13 +0300)]
hurd: Mark various conditions as unlikely

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131354.2507443-3-bugaevc@gmail.com>

14 months agohurd: Move libc_hidden_def's around
Sergey Bugaev [Sat, 29 Apr 2023 13:13:49 +0000 (16:13 +0300)]
hurd: Move libc_hidden_def's around

Each libc_hidden_def should be placed immediately next to its function,
not in some random unrelated place.

No functional change.

Fixes: 653d74f12abea144219af00400ed1f1ac5dfa79f
"hurd: Global signal disposition"

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131354.2507443-2-bugaevc@gmail.com>

14 months agohurd: Simplify _hurd_critical_section_lock a bit
Sergey Bugaev [Sat, 29 Apr 2023 13:13:48 +0000 (16:13 +0300)]
hurd: Simplify _hurd_critical_section_lock a bit

This block of code was doing exactly what _hurd_self_sigstate does; so
just call that and let it do its job.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230429131354.2507443-1-bugaevc@gmail.com>

14 months ago__check_pf: Add a cancellation cleanup handler [BZ #20975]
H.J. Lu [Thu, 27 Apr 2023 20:06:15 +0000 (13:06 -0700)]
__check_pf: Add a cancellation cleanup handler [BZ #20975]

There are reports for hang in __check_pf:

https://github.com/JoeDog/siege/issues/4

It is reproducible only under specific configurations:

1. Large number of cores (>= 64) and large number of threads (> 3X of
the number of cores) with long lived socket connection.
2. Low power (frequency) mode.
3. Power management is enabled.

While holding lock, __check_pf calls make_request which calls __sendto
and __recvmsg.  Since __sendto and __recvmsg are cancellation points,
lock held by __check_pf won't be released and can cause deadlock when
thread cancellation happens in __sendto or __recvmsg.  Add a cancellation
cleanup handler for __check_pf to unlock the lock when cancelled by
another thread.  This fixes BZ #20975 and the siege hang issue.

14 months agoRemap __GLIBC_FLT_EVAL_METHOD to 0 if __FLT_EVAL_METHOD__ is -1
Kito Cheng [Tue, 14 Mar 2023 15:19:48 +0000 (23:19 +0800)]
Remap __GLIBC_FLT_EVAL_METHOD to 0 if __FLT_EVAL_METHOD__ is -1

__GLIBC_FLT_EVAL_METHOD will effect the definition of float_t and
double_t, currently we'll set __GLIBC_FLT_EVAL_METHOD to 2 when
__FLT_EVAL_METHOD__ is -1, that means we'll define float_t and double_t
to long double.

However some target isn't natively (HW) support long double like AArch64 and
RISC-V, they defined long double as 128-bits IEEE 754 floating point type.

That means setting __GLIBC_FLT_EVAL_METHOD to 2 will cause very inefficient
code gen for those target who didn't provide native support for long
double, and that's violate the spirit float_t and double_t - most efficient
types at least as wide as float and double.

So this patch propose to remap __GLIBC_FLT_EVAL_METHOD to 0 rather than
2 when __FLT_EVAL_METHOD__ is -1, which means we'll use float/double
rather than long double for float_t and double_t.

Note: __FLT_EVAL_METHOD__ == -1 means the precision is indeterminable,
      which means compiler might using indeterminable precision during
      optimization/code gen, clang will set this value to -1 when fast
      math is enabled.

Note: Default definition float_t and double_t in current glibc:
      |  __GLIBC_FLT_EVAL_METHOD | float_t     | double_t
      |               0 or 16    | float       | double
      |               1          | double      | doulbe
      |               2          | long double | long double
      More complete list see math/math.h

Note: RISC-V has defined ISA extension to support 128-bits IEEE 754
      floating point operations, but only rare RISC-V core will implement that.

Related link:

[1] LLVM issue (__FLT_EVAL_METHOD__ is set to -1 with Ofast. #60781):
    https://github.com/llvm/llvm-project/issues/60781
[2] Last version of this patch: https://sourceware.org/pipermail/libc-alpha/2023-February/145622.html

Acked-by: Palmer Dabbelt <palmer@rivosinc.com> # RISC-V
Reviewed-by: Wilco Dijkstra  <Wilco.Dijkstra@arm.com>
Link: https://inbox.sourceware.org/libc-alpha/20230314151948.12892-1-kito.cheng@sifive.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
14 months agoriscv: Resolve symbols directly for symbols with STO_RISCV_VARIANT_CC.
Hsiangkai Wang [Tue, 14 Mar 2023 16:25:12 +0000 (00:25 +0800)]
riscv: Resolve symbols directly for symbols with STO_RISCV_VARIANT_CC.

In some cases, we do not want to go through the resolver for function
calls. For example, functions with vector arguments will use vector
registers to pass arguments. In the resolver, we do not save/restore the
vector argument registers for lazy binding efficiency. To avoid ruining
the vector arguments, functions with vector arguments will not go
through the resolver.

To achieve the goal, we will annotate the function symbols with
STO_RISCV_VARIANT_CC flag and add DT_RISCV_VARIANT_CC tag in the dynamic
section. In the first pass on PLT relocations, we do not set up to call
_dl_runtime_resolve. Instead, we resolve the functions directly.

Signed-off-by: Hsiangkai Wang <kai.wang@sifive.com>
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Reviewed-by: Palmer Dabbelt <palmer@rivosinc.com>
Acked-by: Palmer Dabbelt <palmer@rivosinc.com>
Link: https://inbox.sourceware.org/libc-alpha/20230314162512.35802-1-kito.cheng@sifive.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
14 months agoelf.h: add PT_GNU_SFRAME
Indu Bhagat [Fri, 28 Apr 2023 08:53:05 +0000 (10:53 +0200)]
elf.h: add PT_GNU_SFRAME

Support for SFrame format is available in Binutils 2.40.  The GNU ld merges
the input .sframe sections and creates an output .sframe section in a
segment PT_GNU_SFRAME.

14 months agoFix Hurd getcwd build with GCC >= 13
Joseph Myers [Wed, 26 Apr 2023 17:14:18 +0000 (17:14 +0000)]
Fix Hurd getcwd build with GCC >= 13

The build of glibc for i686-gnu has been failing for a while with GCC
mainline / GCC 13:

../sysdeps/mach/hurd/getcwd.c: In function '__hurd_canonicalize_directory_name_internal':
../sysdeps/mach/hurd/getcwd.c:242:48: error: pointer 'file_name' may be used after 'realloc' [-Werror=use-after-free]
  242 |                   file_namep = &buf[file_namep - file_name + size / 2];
      |                                     ~~~~~~~~~~~^~~~~~~~~~~
../sysdeps/mach/hurd/getcwd.c:236:25: note: call to 'realloc' here
  236 |                   buf = realloc (file_name, size);
      |                         ^~~~~~~~~~~~~~~~~~~~~~~~~

Fix by doing the subtraction before the reallocation.

Tested with build-many-glibcs.py for i686-gnu.

[samuel.thibault@ens-lyon.rg: Removed mention of this being a bug]

Message-Id: <18587337-7815-4056-ebd0-724df262d591@codesourcery.com>

14 months agoRegenerate sysdeps/mach/hurd/bits/errno.h
Joseph Myers [Wed, 26 Apr 2023 17:11:41 +0000 (17:11 +0000)]
Regenerate sysdeps/mach/hurd/bits/errno.h

This file was out of date, as shown by build-many-glibcs.py runs
resulting in a modified source directory.

14 months agolocale/programs/locarchive.c: Remove unnecessary check in add_locale_archive
Frédéric Bérat [Wed, 26 Apr 2023 12:30:58 +0000 (08:30 -0400)]
locale/programs/locarchive.c: Remove unnecessary check in add_locale_archive

Since asprintf is called "if (mask & XPG_NORM_CODESET)" there is no
point in checking the mask again within the asprintf call.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agomanual: document posix_openpt (bug 17010)
Gavin Smith [Wed, 12 Apr 2023 12:45:07 +0000 (13:45 +0100)]
manual: document posix_openpt (bug 17010)

State that getpt is similar to posix_openpt.  Use posix_openpt
instead of getpt in example.

Signed-off-by: Gavin Smith <gavinsmith0123@gmail.com>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
14 months agoif_index: Remove unneeded alloca.h include
Joe Simmons-Talbott [Tue, 25 Apr 2023 14:26:54 +0000 (10:26 -0400)]
if_index: Remove unneeded alloca.h include

Nothing is being used from this header.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agogethostid: Do not include alloca.h
Joe Simmons-Talbott [Mon, 24 Apr 2023 19:30:58 +0000 (15:30 -0400)]
gethostid: Do not include alloca.h

Nothing from alloca.h is being used here.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
14 months agosocket: Add a test for MSG_CMSG_CLOEXEC
Sergey Bugaev [Sun, 23 Apr 2023 16:05:48 +0000 (19:05 +0300)]
socket: Add a test for MSG_CMSG_CLOEXEC

This checks that:
* We can send and receive fds over Unix domain sockets using SCM_RIGHTS;
* msg_controllen, cmsg_level, cmsg_type, cmsg_len are all filled in
  correctly on receive;
* Most importantly, the received fd has or has not the close-on-exec
  flag set depending on whether we pass MSG_CMSG_CLOEXEC to recvmsg ().

Checked on i686-gnu and x86_64-linux-gnu.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230423160548.126576-4-bugaevc@gmail.com>

14 months agohurd: Do not take any flag from the CMSG_DATA
Samuel Thibault [Mon, 24 Apr 2023 22:14:58 +0000 (00:14 +0200)]
hurd: Do not take any flag from the CMSG_DATA

As fixed in 0822e3552a78 ("hurd: Don't pass FD_CLOEXEC in CMSG_DATA"),
senders currently don't have any flag to pass.  We shouldn't blindly take
random flags that senders could be erroneously giving us.

14 months agohurd: Implement MSG_CMSG_CLOEXEC
Sergey Bugaev [Sun, 23 Apr 2023 16:05:46 +0000 (19:05 +0300)]
hurd: Implement MSG_CMSG_CLOEXEC

This is a new flag that can be passed to recvmsg () to make it
atomically set the CLOEXEC flag on all the file descriptors received
using the SCM_RIGHTS mechanism. This is useful for all the same reasons
that the other XXX_CLOEXEC flags are useful: namely, it provides
atomicity with respect to another thread of the same process calling
(fork and then) exec at the same time.

This flag is already supported on Linux and FreeBSD. The flag's value,
0x40000, is choosen to match FreeBSD's.

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
Message-Id: <20230423160548.126576-2-bugaevc@gmail.com>