Patch 600f00b "linux: Use long time_t for wait4/getrusage" introduced
two bugs:
- The usage32 struct was set if the wait4 syscall had an error.
- For 32-bit systems the usage struct was set even if it was specified
as NULL.
This patch fixes the two issues.
X32 has 32-bit long and pointer with 64-bit off_t. Since x32 psABI
requires that pointers passed in registers must be zero-extended to
64bit, x32 can share many syscall interfaces with LP64. When a LP64
syscall with long and unsigned long arguments is used for x32, these
arguments must be properly extended to 64-bit. Otherwise if the upper
32 bits of the register have undefined value, such a syscall will be
rejected by kernel.
Enforce zero-extension for pointers and array system call arguments.
For integer types, extend to int64_t (the full register) using a
regular cast, resulting in zero or sign extension based on the
signedness of the original type.
For
void *mmap(void *addr, size_t length, int prot, int flags,
int fd, off_t offset);
we now generate
0: 41 f7 c1 ff 0f 00 00 test $0xfff,%r9d
7: 75 1f jne 28 <__mmap64+0x28>
9: 48 63 d2 movslq %edx,%rdx
c: 89 f6 mov %esi,%esi
e: 4d 63 c0 movslq %r8d,%r8
11: 4c 63 d1 movslq %ecx,%r10
14: b8 09 00 00 40 mov $0x40000009,%eax
19: 0f 05 syscall
That is
1. addr is unchanged.
2. length is zero-extend to 64 bits.
3. prot is sign-extend to 64 bits.
4. flags is sign-extend to 64 bits.
5. fd is sign-extend to 64 bits.
6. offset is unchanged.
For int arguments, since kernel uses only the lower 32 bits and ignores
the upper 32 bits in 64-bit registers, these work correctly.
Tested on x86-64 and x32. There are no code changes on x86-64.
This patch updates the kernel version in the test tst-mman-consts.py
to 5.6. (There are no new constants covered by this test in 5.6 that
need any other header changes.)
Tested with build-many-glibcs.py.
Since GCC 6.2 or later is required to build glibc, remove build support
for GCC older than GCC 6.
Testd with GCC 6.4 and GCC 9.3.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
This patch provides new __mq_timedreceive_time64 explicit 64 bit function for
receiving messages with absolute timeout.
Moreover, a 32 bit version - __mq_timedreceive has been refactored to
internally use __mq_timedreceive_time64.
The __mq_timedreceive is now supposed to be used on systems still supporting 32
bit time (__TIMESIZE != 64) - hence the necessary conversion to 64 bit struct
__timespec64 from struct timespec.
The new mq_timedsend_time64 syscall available from Linux 5.1+ has been used,
when applicable.
As this wrapper function is also used internally in the glibc, to e.g. provide
mq_receive implementation, an explicit check for abs_timeout being NULL has been
added due to conversions between struct timespec and struct __timespec64.
Before this change the Linux kernel handled this NULL pointer.
Build tests:
- ./src/scripts/build-many-glibcs.py glibcs
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
Linux kernel, headers and minimal kernel version for glibc build test matrix:
- Linux v5.1 (with mq_timedreceive_time64) and glibc built with v5.1 as
minimal kernel version (--enable-kernel="5.1.0")
The __ASSUME_TIME64_SYSCALLS flag defined.
- Linux v5.1 and default minimal kernel version
The __ASSUME_TIME64_SYSCALLS not defined, but kernel supports
mq_timedreceive_time64 syscall.
- Linux v4.19 (no mq_timedreceive_time64 support) with default minimal kernel
version for contemporary glibc (3.2.0)
This kernel doesn't support mq_timedreceive_time64 syscall, so the fallback to
mq_timedreceive is tested.
Above tests were performed with Y2038 redirection applied as well as without
(so the __TIMESIZE != 64 execution path is checked as well).
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
This patch provides new __mq_timedsend_time64 explicit 64 bit function for
sending messages with absolute timeout.
Moreover, a 32 bit version - __mq_timedsend has been refactored to internally
use __mq_timedsend_time64.
The __mq_timedsend is now supposed to be used on systems still supporting 32
bit time (__TIMESIZE != 64) - hence the necessary conversion to 64 bit struct
__timespec64 from struct timespec.
The new __mq_timedsend_time64 syscall available from Linux 5.1+ has been used,
when applicable.
As this wrapper function is also used internally in the glibc, to e.g. provide
mq_send implementation, an explicit check for abs_timeout being NULL has been
added due to conversions between struct timespec and struct __timespec64.
Before this change the Linux kernel handled this NULL pointer.
Build tests:
- ./src/scripts/build-many-glibcs.py glibcs
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
Linux kernel, headers and minimal kernel version for glibc build test matrix:
- Linux v5.1 (with mq_timedsend_time64) and glibc built with v5.1 as a
minimal kernel version (--enable-kernel="5.1.0")
The __ASSUME_TIME64_SYSCALLS flag defined.
- Linux v5.1 and default minimal kernel version
The __ASSUME_TIME64_SYSCALLS not defined, but kernel supports
mq_timedsend_time64 syscall.
- Linux v4.19 (no mq_timedsend_time64 support) with default minimal kernel
version for contemporary glibc (3.2.0)
This kernel doesn't support mq_timedsend_time64 syscall, so the fallback to
mq_timedsend is tested.
Above tests were performed with Y2038 redirection applied as well as without
(so the __TIMESIZE != 64 execution path is checked as well).
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
check_consistency should be disabled for GCC 5 and above since there is
no fixed PIC register in GCC 5 and above. Check __GNUC_PREREQ (5,0)
instead OPTIMIZE_FOR_GCC_5 since OPTIMIZE_FOR_GCC_5 is false with
-fno-omit-frame-pointer.
Linux 5.6 has new openat2 and pidfd_getfd syscalls. This patch adds
them to syscall-names.list and regenerates the arch-syscall.h files.
Tested with build-many-glibcs.py.
All cancellable syscalls are done by C implementations, so there is no
no need to use a specialized implementation to optimize register usage.
It fixes BZ #25765.
Checked on x86_64-linux-gnu.
Now there is a generic __timeval32 and helpers we can use them for Alpha
instead of the Alpha specific ones.
Reviewed-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
The Linux kernel expects rusage to use a 32-bit time_t, even on archs
with a 64-bit time_t (like RV32). To address this let's convert
rusage to/from 32-bit and 64-bit to ensure the kernel always gets
a 32-bit time_t.
While we are converting these functions let's also convert them to be
the y2038 safe versions. This means there is a *64 function that is
called by a backwards compatible wrapper.
Reviewed-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
The Linux kernel expects itimerval to use a 32-bit time_t, even on archs
with a 64-bit time_t (like RV32). To address this let's convert
itimerval to/from 32-bit and 64-bit to ensure the kernel always gets
a 32-bit time_t.
While we are converting these functions let's also convert them to be
the y2038 safe versions. This means there is a *64 function that is
called by a backwards compatible wrapper.
Tested-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
On y2038 safe 32-bit systems the Linux kernel expects itimerval
and rusage to use a 32-bit time_t, even though the other time_t's
are 64-bit. There are currently no plans to make 64-bit time_t versions
of these structs.
There are also other occurrences where the time passed to the kernel via
timeval doesn't match the wordsize.
To handle these cases let's define a new macro
__KERNEL_OLD_TIMEVAL_MATCHES_TIMEVAL64. This macro specifies if the
kernel's old_timeval matches the new timeval64. This should be 1 for
64-bit architectures except for Alpha's osf syscalls. The define should
be 0 for 32-bit architectures and Alpha's osf syscalls.
Reviewed-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
This addresses an issue that is present mainly on SMP machines running
threaded code. In a typical indirect call or PLT import stub, the
target address is loaded first. Then the global pointer is loaded into
the PIC register in the delay slot of a branch to the target address.
During lazy binding, the target address is a trampoline which transfers
to _dl_runtime_resolve().
_dl_runtime_resolve() uses the relocation offset stored in the global
pointer and the linkage map stored in the trampoline to find the
relocation. Then, the function descriptor is updated.
In a multi-threaded application, it is possible for the global pointer
to be updated between the load of the target address and the global
pointer. When this happens, the relocation offset has been replaced
by the new global pointer. The function pointer has probably been
updated as well but there is no way to find the address of the function
descriptor and to transfer to the target. So, _dl_runtime_resolve()
typically crashes.
HP-UX addressed this problem by adding an extra pc-relative branch to
the trampoline. The descriptor is initially setup to point to the
branch. The branch then transfers to the trampoline. This allowed
the trampoline code to figure out which descriptor was being used
without any modification to user code. I didn't use this approach
as it is more complex and changes function pointer canonicalization.
The order of loading the target address and global pointer in
indirect calls was not consistent with the order used in import stubs.
In particular, $$dyncall and some inline versions of it loaded the
global pointer first. This was inconsistent with the global pointer
being updated first in dl-machine.h. Assuming the accesses are
ordered, we want elf_machine_fixup_plt() to store the global pointer
first and calls to load it last. Then, the global pointer will be
correct when the target function is entered.
However, just to make things more fun, HP added support for
out-of-order execution of accesses in PA 2.0. The accesses used by
calls are weakly ordered. So, it's possibly under some circumstances
that a function might be entered with the wrong global pointer.
However, HP uses weakly ordered accesses in 64-bit HP-UX, so I assume
that loading the global pointer in the delay slot of the branch must
work consistently.
The basic fix for the race is a combination of modifying user code to
preserve the address of the function descriptor in register %r22 and
setting the least-significant bit in the relocation offset. The
latter was suggested by Carlos as a way to distinguish relocation
offsets from global pointer values. Conventionally, %r22 is used
as the address of the function descriptor in calls to $$dyncall.
So, it wasn't hard to preserve the address in %r22.
I have updated gcc trunk and gcc-9 branch to not clobber %r22 in
$$dyncall and inline indirect calls. I have also modified the import
stubs in binutils trunk and the 2.33 branch to preserve %r22. This
required making the stubs one instruction longer but we save one
relocation. I also modified binutils to align the .plt section on
a 8-byte boundary. This allows descriptors to be updated atomically
with a floting-point store.
With these changes, _dl_runtime_resolve() can fallback to an alternate
mechanism to find the relocation offset when it has been clobbered.
There's just one additional instruction in the fast path. I tested
the fallback function, _dl_fix_reloc_arg(), by changing the branch to
always use the fallback. Old code still runs as it did before.
Fixes bug 23296.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
These functions are alpha specifc, rename them to be clear.
Let's also rename the header file from tv32-compat.h to
alpha-tv32-compat.h. This is to avoid conflicts with the one we will
introduce later.
Reviewed-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
During the conversion to support 64 bit time on some architectures with
__WORDSIZE == 32 && __TIMESIZE != 64 the libc_hidden_def attribute for
eligible functions was by mistake omitted.
This patch fixes this issue and exports (and allows using) those
functions when Y2038 support is enabled in glibc.
Since legacy bitmap doesn't cover jitted code generated by legacy JIT
engine, it isn't very useful. This patch removes ARCH_CET_LEGACY_BITMAP
and treats indirect branch tracking similar to shadow stack by removing
legacy bitmap support.
Tested on CET Linux/x86-64 and non-CET Linux/x86-64.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
This conversion patch for supporting 64 bit time for futimesat only differs
from the work performed for futimes (when providing __futimes64) with passing
also the file name (and path) to utimensat.
All the design and conversion decisions are exactly the same as for futimens
conversion.
This conversion patch for supporting 64 bit time for lutimes mostly differs from
the work performed for futimes (when providing __futimes64) with adding the
AT_SYMLINK_NOFOLLOW flag to utimensat.
It also supports passing file name instead of file descriptor number, but this
is not relevant for utimensat used to implement it.
All the design and conversion decisions are exactly the same as for futimens
conversion.
This patch provides new __futimes64 explicit 64 bit function for setting file's
64 bit attributes for access and modification time (by specifying file
descriptor number).
Internally, the __utimensat64_helper function is used. This patch is necessary
for having architectures with __WORDSIZE == 32 Y2038 safe.
Moreover, a 32 bit version - __futimes has been refactored to internally use
__futimes64.
The __futimes is now supposed to be used on systems still supporting 32
bit time (__TIMESIZE != 64) - hence the necessary conversion of struct timeval
to 64 bit struct __timeval64.
The check if struct timevals' usec fields are in the range between 0 and 1000000
has been removed as Linux kernel performs it internally in the implementation
of utimensat (the conversion between struct __timeval64 and __timespec64 is not
relevant for this particular check).
Last but not least, checks for tvp{64} not being NULL have been preserved from
the original code as some legacy user space programs may rely on it.
Build tests:
./src/scripts/build-many-glibcs.py glibcs
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
Above tests were performed with Y2038 redirection applied as well as without to
test the proper usage of both __futimes64 and __futimes.
It seems that some gcc versions might generates a stack frame for the
sigreturn stub requires on sparc signal handling. For instance:
$ cat test.c
#define _GNU_SOURCE
#include <sys/syscall.h>
__attribute__ ((__optimize__ ("-fno-stack-protector")))
void
__sigreturn_stub (void)
{
__asm__ ("mov %0, %%g1\n\t"
"ta 0x10\n\t"
: /* no outputs */
: "i" (SYS_rt_sigreturn));
}
$ gcc -v
[...]
gcc version 9.2.1 20200224 (Debian 9.2.1-30)
$ gcc -O2 -m64 test.c -S -o -
[...]
__sigreturn_stub:
save %sp, -176, %sp
#APP
! 9 "t.c" 1
mov 101, %g1
ta 0x10
! 0 "" 2
#NO_APP
.size __sigreturn_stub, .-__sigreturn_stub
As indicated by kernel developers [1], the sigreturn stub can not change
the register window or the stack pointer since the kernel has setup the
restore frame at a precise location relative to the stack pointer when
the stub is invoked.
I tried to play with some compiler flags and even with _Noreturn and
__builtin_unreachable after the asm does not help (and Sparc does not
support naked functions).
To avoid similar issues, as the stack-protector support also have
stumbled, this patch moves the implementation of the sigreturn stubs to
assembly.
Checked on sparcv9-linux-gnu and sparc64-linux-gnu with gcc 9.2.1
and gcc 7.5.0.
[1] https://lkml.org/lkml/2016/5/27/465
The kernel might not clear the padding value for the ipc_perm mode
fields in compat mode (32 bit running on a 64 bit kernel). It was
fixed on v4.14 when the ipc compat code was refactored to move
(commits 553f770ef71b, 469391684626, c0ebccb6fa1e).
Although it is most likely a kernel issue, it was shown only due
BZ#18231 fix which made all the SysVIPC mode_t 32-bit regardless of
the kABI.
This patch fixes it by explicitly zeroing the upper bits for such
cases. The __ASSUME_SYSVIPC_BROKEN_MODE_T case already handles
it with the shift.
(The aarch64 ipc_priv.h is superflous since
__ASSUME_SYSVIPC_DEFAULT_IPC_64 is now defined as default).
Checked on i686-linux-gnu on 3.10 and on 4.15 kernel.
fstatat64 depends on inlining to produce the desired __fxstatat64
call, which does not happen with -Os, leading to a link failure
with an undefined reference to fstatat64. __fxstatat64 has a macro
definition in include/sys/stat.h and thus avoids the problem.
This patch replaces auto generated wrapper (as described in
sysdeps/unix/sysv/linux/syscalls.list) for utime with one which adds extra
support for setting file's access and modification 64 bit time on machines
with __TIMESIZE != 64.
Internally, the __utimensat_time64 helper function is used. This patch is
necessary for having architectures with __WORDSIZE == 32 && __TIMESIZE != 64
Y2038 safe.
Moreover, a 32 bit version - __utime has been refactored to internally use
__utime64.
The __utime is now supposed to be used on systems still supporting 32
bit time (__TIMESIZE != 64) - hence the necessary conversion between struct
utimbuf and struct __utimbuf64.
Build tests:
./src/scripts/build-many-glibcs.py glibcs
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
Above tests were performed with Y2038 redirection applied as well as
without to test proper usage of both __utime64 and __utime.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
This patch provides new __utimes64 explicit 64 bit function for setting file's
64 bit attributes for access and modification time.
Internally, the __utimensat64_helper function is used. This patch is necessary
for having architectures with __WORDSIZE == 32 Y2038 safe.
Moreover, a 32 bit version - __utimes has been refactored to internally use
__utimes64.
The __utimes is now supposed to be used on systems still supporting 32
bit time (__TIMESIZE != 64) - hence the necessary conversion of struct
timeval to 64 bit struct __timeval64.
Build tests:
./src/scripts/build-many-glibcs.py glibcs
Run-time tests:
- Run specific tests on ARM/x86 32bit systems (qemu):
https://github.com/lmajewski/meta-y2038 and run tests:
https://github.com/lmajewski/y2038-tests/commits/master
Above tests were performed with Y2038 redirection applied as well as without
to test proper usage of both __utimes64 and __utimes.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Due to the built-in tables, __NR_vfork is always defined, so the
fork-based fallback code is never used.
(It appears that the vfork system call was wired up when the port was
contributed to the kernel.)
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Due to the built-in tables, __NR_getdents64 is always defined,
although it may not be supported at run time.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
With the built-in tables __NR_preadv2 and __NR_pwritev2 are always
defined.
The kernel has never defined __NR_preadv64v2 and __NR_pwritev64v2
and is unlikely to do so, given that the preadv2 and pwritev2 system
calls themselves are 64-bit.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Due to the built-in tables, __NR_rt_sigqueueinfo is always defined.
sysdeps/pthread/time_routines.c is not updated because it is shared with
Hurd.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
The names __NR_preadv64, __NR_pwritev64 appear to be a glibc invention.
With the built-in tables, __NR_preadv and __NR_pwritev are always defined.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Linux removed the last definitions of __NR_pread and __NR_pwrite
in commit 4ba66a9760722ccbb691b8f7116cad2f791cca7b, the removal
of the blackfin port. All architectures now define __NR_pread64 and
__NR_pwrite64 only.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
