linux: Only use 64-bit syscall if required for select

For !__ASSUME_TIME64_SYSCALLS there is no need to issue a 64-bit syscall
if the provided timeout fits in a 32-bit one.  The 64-bit usage should
be rare since the timeout is a relative one.  This also avoids the need
to use supports_time64() (which breaks the usage case of live migration
like CRIU or similar).

It also fixes an issue on 32-bit select call for !__ASSUME_PSELECT
(microblase with older kernels only) where the expected timeout
is a 'struct timeval' instead of 'struct timespec'.

Checked on i686-linux-gnu on a 4.15 kernel and on a 5.11 kernel
(with and without --enable-kernel=5.1) and on x86_64-linux-gnu.

Reviewed-by: Lukasz Majewski <lukma@denx.de>
This commit is contained in:
Adhemerval Zanella 2021-06-15 21:00:50 -03:00
parent 91cf411ad3
commit 4c3df0eba5
4 changed files with 71 additions and 53 deletions

View File

@ -21,6 +21,11 @@ extern int __pselect32 (int __nfds, fd_set *__readfds,
const struct __timespec64 *__timeout,
const __sigset_t *__sigmask)
attribute_hidden;
extern int __select32 (int __nfds, fd_set *__readfds,
fd_set *__writefds, fd_set *__exceptfds,
const struct __timespec64 *ts64,
struct __timeval64 *timeout)
attribute_hidden;
extern int __select64 (int __nfds, fd_set *__readfds,
fd_set *__writefds, fd_set *__exceptfds,

View File

@ -169,5 +169,7 @@ $(objpfx)tst-allocate_once-mem.out: $(objpfx)tst-allocate_once.out
$(common-objpfx)malloc/mtrace $(objpfx)tst-allocate_once.mtrace > $@; \
$(evaluate-test)
$(objpfx)tst-select: $(librt)
$(objpfx)tst-select-time64: $(librt)
$(objpfx)tst-pselect: $(librt)
$(objpfx)tst-pselect-time64: $(librt)

View File

@ -17,6 +17,7 @@
<https://www.gnu.org/licenses/>. */
#include <errno.h>
#include <intprops.h>
#include <support/capture_subprocess.h>
#include <support/check.h>
#include <support/support.h>
@ -31,12 +32,6 @@ struct child_args
struct timeval tmo;
};
static void
alarm_handler (int signum)
{
/* Do nothing. */
}
static void
do_test_child (void *clousure)
{
@ -69,17 +64,20 @@ do_test_child (void *clousure)
static void
do_test_child_alarm (void *clousure)
{
struct sigaction act = { .sa_handler = alarm_handler };
xsigaction (SIGALRM, &act, NULL);
alarm (1);
struct child_args *args = (struct child_args *) clousure;
struct timeval tv = { .tv_sec = 10, .tv_usec = 0 };
support_create_timer (0, 100000000, false, NULL);
struct timeval tv = { .tv_sec = args->tmo.tv_sec, .tv_usec = 0 };
int r = select (0, NULL, NULL, NULL, &tv);
TEST_COMPARE (r, -1);
if (args->tmo.tv_sec > INT_MAX)
TEST_VERIFY (errno == EINTR || errno == EOVERFLOW);
else
{
TEST_COMPARE (errno, EINTR);
if (support_select_modifies_timeout ())
TEST_VERIFY (tv.tv_sec < 10);
TEST_VERIFY (tv.tv_sec < args->tmo.tv_sec);
}
}
static int
@ -121,13 +119,24 @@ do_test (void)
xclose (args.fds[0][0]);
xclose (args.fds[1][1]);
args.tmo = (struct timeval) { .tv_sec = 10, .tv_usec = 0 };
{
struct support_capture_subprocess result;
result = support_capture_subprocess (do_test_child_alarm, NULL);
result = support_capture_subprocess (do_test_child_alarm, &args);
support_capture_subprocess_check (&result, "tst-select-child", 0,
sc_allow_none);
}
args.tmo = (struct timeval) { .tv_sec = TYPE_MAXIMUM (time_t),
.tv_usec = 0 };
{
struct support_capture_subprocess result;
result = support_capture_subprocess (do_test_child_alarm, &args);
support_capture_subprocess_check (&result, "tst-select-child", 0,
sc_allow_none);
}
args.tmo = (struct timeval) { .tv_sec = 0, .tv_usec = 0 };
{
fd_set rfds;
FD_ZERO (&rfds);

View File

@ -21,7 +21,6 @@
#include <sys/select.h>
#include <errno.h>
#include <sysdep-cancel.h>
#include <time64-support.h>
/* Check the first NFDS descriptors each in READFDS (if not NULL) for read
readiness, in WRITEFDS (if not NULL) for write readiness, and in EXCEPTFDS
@ -65,53 +64,56 @@ __select64 (int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
#ifndef __NR_pselect6_time64
# define __NR_pselect6_time64 __NR_pselect6
#endif
int r;
if (supports_time64 ())
{
r = SYSCALL_CANCEL (pselect6_time64, nfds, readfds, writefds, exceptfds,
#ifdef __ASSUME_TIME64_SYSCALLS
int r = SYSCALL_CANCEL (pselect6_time64, nfds, readfds, writefds, exceptfds,
pts64, NULL);
/* Linux by default will update the timeout after a pselect6 syscall
(though the pselect() glibc call suppresses this behavior).
Since select() on Linux has the same behavior as the pselect6
syscall, we update the timeout here. */
if (r >= 0 || errno != ENOSYS)
{
if (timeout != NULL)
TIMESPEC_TO_TIMEVAL (timeout, pts64);
return r;
#else
bool need_time64 = timeout != NULL && !in_time_t_range (timeout->tv_sec);
if (need_time64)
{
int r = SYSCALL_CANCEL (pselect6_time64, nfds, readfds, writefds,
exceptfds, pts64, NULL);
if ((r >= 0 || errno != ENOSYS) && timeout != NULL)
{
TIMESPEC_TO_TIMEVAL (timeout, &ts64);
}
else
__set_errno (EOVERFLOW);
return r;
}
mark_time64_unsupported ();
}
#ifndef __ASSUME_TIME64_SYSCALLS
# ifdef __ASSUME_PSELECT
struct timespec ts32, *pts32 = NULL;
if (pts64 != NULL)
{
if (! in_time_t_range (pts64->tv_sec))
{
__set_errno (EINVAL);
return -1;
}
ts32.tv_sec = s;
ts32.tv_nsec = ns;
ts32.tv_sec = pts64->tv_sec;
ts32.tv_nsec = pts64->tv_nsec;
pts32 = &ts32;
}
# ifndef __ASSUME_PSELECT
# ifdef __NR__newselect
# undef __NR_select
# define __NR_select __NR__newselect
# endif
r = SYSCALL_CANCEL (select, nfds, readfds, writefds, exceptfds, pts32);
# else
r = SYSCALL_CANCEL (pselect6, nfds, readfds, writefds, exceptfds, pts32,
NULL);
# endif
if (timeout != NULL)
*timeout = valid_timespec_to_timeval64 (ts32);
#endif
int r = SYSCALL_CANCEL (pselect6, nfds, readfds, writefds, exceptfds, pts32,
NULL);
if (timeout != NULL)
TIMESPEC_TO_TIMEVAL (timeout, pts32);
return r;
# else
struct timeval tv32, *ptv32 = NULL;
if (pts64 != NULL)
{
tv32 = valid_timespec64_to_timeval (*pts64);
ptv32 = &tv32;
}
int r = SYSCALL_CANCEL (select, nfds, readfds, writefds, exceptfds, ptv32);
if (timeout != NULL)
*timeout = valid_timeval_to_timeval64 (tv32);
return r;
# endif /* __ASSUME_PSELECT */
#endif
}
#if __TIMESIZE != 64