]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/select.c
2 * This file contains the procedures for the handling of select and poll
4 * Created for Linux based loosely upon Mathius Lattner's minix
5 * patches by Peter MacDonald. Heavily edited by Linus.
8 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
9 * flag set in its personality we do *not* modify the given timeout
10 * parameter to reflect time remaining.
13 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
14 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
17 #include <linux/kernel.h>
18 #include <linux/sched.h>
19 #include <linux/syscalls.h>
20 #include <linux/export.h>
21 #include <linux/slab.h>
22 #include <linux/poll.h>
23 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
27 #include <linux/rcupdate.h>
28 #include <linux/hrtimer.h>
29 #include <linux/sched/rt.h>
30 #include <net/ll_poll.h>
32 #include <asm/uaccess.h>
36 * Estimate expected accuracy in ns from a timeval.
38 * After quite a bit of churning around, we've settled on
39 * a simple thing of taking 0.1% of the timeout as the
40 * slack, with a cap of 100 msec.
41 * "nice" tasks get a 0.5% slack instead.
43 * Consider this comment an open invitation to come up with even
47 #define MAX_SLACK (100 * NSEC_PER_MSEC)
49 static long __estimate_accuracy(struct timespec
*tv
)
57 if (task_nice(current
) > 0)
58 divfactor
= divfactor
/ 5;
60 if (tv
->tv_sec
> MAX_SLACK
/ (NSEC_PER_SEC
/divfactor
))
63 slack
= tv
->tv_nsec
/ divfactor
;
64 slack
+= tv
->tv_sec
* (NSEC_PER_SEC
/divfactor
);
66 if (slack
> MAX_SLACK
)
72 long select_estimate_accuracy(struct timespec
*tv
)
78 * Realtime tasks get a slack of 0 for obvious reasons.
85 now
= timespec_sub(*tv
, now
);
86 ret
= __estimate_accuracy(&now
);
87 if (ret
< current
->timer_slack_ns
)
88 return current
->timer_slack_ns
;
94 struct poll_table_page
{
95 struct poll_table_page
* next
;
96 struct poll_table_entry
* entry
;
97 struct poll_table_entry entries
[0];
100 #define POLL_TABLE_FULL(table) \
101 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
104 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
105 * I have rewritten this, taking some shortcuts: This code may not be easy to
106 * follow, but it should be free of race-conditions, and it's practical. If you
107 * understand what I'm doing here, then you understand how the linux
108 * sleep/wakeup mechanism works.
110 * Two very simple procedures, poll_wait() and poll_freewait() make all the
111 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
112 * as all select/poll functions have to call it to add an entry to the
115 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
118 void poll_initwait(struct poll_wqueues
*pwq
)
120 init_poll_funcptr(&pwq
->pt
, __pollwait
);
121 pwq
->polling_task
= current
;
125 pwq
->inline_index
= 0;
127 EXPORT_SYMBOL(poll_initwait
);
129 static void free_poll_entry(struct poll_table_entry
*entry
)
131 remove_wait_queue(entry
->wait_address
, &entry
->wait
);
135 void poll_freewait(struct poll_wqueues
*pwq
)
137 struct poll_table_page
* p
= pwq
->table
;
139 for (i
= 0; i
< pwq
->inline_index
; i
++)
140 free_poll_entry(pwq
->inline_entries
+ i
);
142 struct poll_table_entry
* entry
;
143 struct poll_table_page
*old
;
148 free_poll_entry(entry
);
149 } while (entry
> p
->entries
);
152 free_page((unsigned long) old
);
155 EXPORT_SYMBOL(poll_freewait
);
157 static struct poll_table_entry
*poll_get_entry(struct poll_wqueues
*p
)
159 struct poll_table_page
*table
= p
->table
;
161 if (p
->inline_index
< N_INLINE_POLL_ENTRIES
)
162 return p
->inline_entries
+ p
->inline_index
++;
164 if (!table
|| POLL_TABLE_FULL(table
)) {
165 struct poll_table_page
*new_table
;
167 new_table
= (struct poll_table_page
*) __get_free_page(GFP_KERNEL
);
172 new_table
->entry
= new_table
->entries
;
173 new_table
->next
= table
;
174 p
->table
= new_table
;
178 return table
->entry
++;
181 static int __pollwake(wait_queue_t
*wait
, unsigned mode
, int sync
, void *key
)
183 struct poll_wqueues
*pwq
= wait
->private;
184 DECLARE_WAITQUEUE(dummy_wait
, pwq
->polling_task
);
187 * Although this function is called under waitqueue lock, LOCK
188 * doesn't imply write barrier and the users expect write
189 * barrier semantics on wakeup functions. The following
190 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
191 * and is paired with set_mb() in poll_schedule_timeout.
197 * Perform the default wake up operation using a dummy
200 * TODO: This is hacky but there currently is no interface to
201 * pass in @sync. @sync is scheduled to be removed and once
202 * that happens, wake_up_process() can be used directly.
204 return default_wake_function(&dummy_wait
, mode
, sync
, key
);
207 static int pollwake(wait_queue_t
*wait
, unsigned mode
, int sync
, void *key
)
209 struct poll_table_entry
*entry
;
211 entry
= container_of(wait
, struct poll_table_entry
, wait
);
212 if (key
&& !((unsigned long)key
& entry
->key
))
214 return __pollwake(wait
, mode
, sync
, key
);
217 /* Add a new entry */
218 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
221 struct poll_wqueues
*pwq
= container_of(p
, struct poll_wqueues
, pt
);
222 struct poll_table_entry
*entry
= poll_get_entry(pwq
);
225 entry
->filp
= get_file(filp
);
226 entry
->wait_address
= wait_address
;
227 entry
->key
= p
->_key
;
228 init_waitqueue_func_entry(&entry
->wait
, pollwake
);
229 entry
->wait
.private = pwq
;
230 add_wait_queue(wait_address
, &entry
->wait
);
233 int poll_schedule_timeout(struct poll_wqueues
*pwq
, int state
,
234 ktime_t
*expires
, unsigned long slack
)
238 set_current_state(state
);
240 rc
= schedule_hrtimeout_range(expires
, slack
, HRTIMER_MODE_ABS
);
241 __set_current_state(TASK_RUNNING
);
244 * Prepare for the next iteration.
246 * The following set_mb() serves two purposes. First, it's
247 * the counterpart rmb of the wmb in pollwake() such that data
248 * written before wake up is always visible after wake up.
249 * Second, the full barrier guarantees that triggered clearing
250 * doesn't pass event check of the next iteration. Note that
251 * this problem doesn't exist for the first iteration as
252 * add_wait_queue() has full barrier semantics.
254 set_mb(pwq
->triggered
, 0);
258 EXPORT_SYMBOL(poll_schedule_timeout
);
261 * poll_select_set_timeout - helper function to setup the timeout value
262 * @to: pointer to timespec variable for the final timeout
263 * @sec: seconds (from user space)
264 * @nsec: nanoseconds (from user space)
266 * Note, we do not use a timespec for the user space value here, That
267 * way we can use the function for timeval and compat interfaces as well.
269 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
271 int poll_select_set_timeout(struct timespec
*to
, long sec
, long nsec
)
273 struct timespec ts
= {.tv_sec
= sec
, .tv_nsec
= nsec
};
275 if (!timespec_valid(&ts
))
278 /* Optimize for the zero timeout value here */
280 to
->tv_sec
= to
->tv_nsec
= 0;
283 *to
= timespec_add_safe(*to
, ts
);
288 static int poll_select_copy_remaining(struct timespec
*end_time
, void __user
*p
,
289 int timeval
, int ret
)
297 if (current
->personality
& STICKY_TIMEOUTS
)
300 /* No update for zero timeout */
301 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
305 rts
= timespec_sub(*end_time
, rts
);
307 rts
.tv_sec
= rts
.tv_nsec
= 0;
310 if (sizeof(rtv
) > sizeof(rtv
.tv_sec
) + sizeof(rtv
.tv_usec
))
311 memset(&rtv
, 0, sizeof(rtv
));
312 rtv
.tv_sec
= rts
.tv_sec
;
313 rtv
.tv_usec
= rts
.tv_nsec
/ NSEC_PER_USEC
;
315 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
318 } else if (!copy_to_user(p
, &rts
, sizeof(rts
)))
322 * If an application puts its timeval in read-only memory, we
323 * don't want the Linux-specific update to the timeval to
324 * cause a fault after the select has completed
325 * successfully. However, because we're not updating the
326 * timeval, we can't restart the system call.
330 if (ret
== -ERESTARTNOHAND
)
335 #define FDS_IN(fds, n) (fds->in + n)
336 #define FDS_OUT(fds, n) (fds->out + n)
337 #define FDS_EX(fds, n) (fds->ex + n)
339 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
341 static int max_select_fd(unsigned long n
, fd_set_bits
*fds
)
343 unsigned long *open_fds
;
348 /* handle last in-complete long-word first */
349 set
= ~(~0UL << (n
& (BITS_PER_LONG
-1)));
351 fdt
= files_fdtable(current
->files
);
352 open_fds
= fdt
->open_fds
+ n
;
357 if (!(set
& ~*open_fds
))
368 if (set
& ~*open_fds
)
377 max
+= n
* BITS_PER_LONG
;
383 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
384 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
385 #define POLLEX_SET (POLLPRI)
387 static inline void wait_key_set(poll_table
*wait
, unsigned long in
,
388 unsigned long out
, unsigned long bit
,
389 unsigned int ll_flag
)
391 wait
->_key
= POLLEX_SET
| ll_flag
;
393 wait
->_key
|= POLLIN_SET
;
395 wait
->_key
|= POLLOUT_SET
;
398 int do_select(int n
, fd_set_bits
*fds
, struct timespec
*end_time
)
400 ktime_t expire
, *to
= NULL
;
401 struct poll_wqueues table
;
403 int retval
, i
, timed_out
= 0;
404 unsigned long slack
= 0;
405 unsigned int busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
406 u64 busy_start
= busy_loop_start_time(busy_flag
);
407 u64 busy_end
= busy_loop_end_time();
410 retval
= max_select_fd(n
, fds
);
417 poll_initwait(&table
);
419 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
424 if (end_time
&& !timed_out
)
425 slack
= select_estimate_accuracy(end_time
);
429 unsigned long *rinp
, *routp
, *rexp
, *inp
, *outp
, *exp
;
430 bool can_busy_loop
= false;
432 inp
= fds
->in
; outp
= fds
->out
; exp
= fds
->ex
;
433 rinp
= fds
->res_in
; routp
= fds
->res_out
; rexp
= fds
->res_ex
;
435 for (i
= 0; i
< n
; ++rinp
, ++routp
, ++rexp
) {
436 unsigned long in
, out
, ex
, all_bits
, bit
= 1, mask
, j
;
437 unsigned long res_in
= 0, res_out
= 0, res_ex
= 0;
439 in
= *inp
++; out
= *outp
++; ex
= *exp
++;
440 all_bits
= in
| out
| ex
;
446 for (j
= 0; j
< BITS_PER_LONG
; ++j
, ++i
, bit
<<= 1) {
450 if (!(bit
& all_bits
))
454 const struct file_operations
*f_op
;
456 mask
= DEFAULT_POLLMASK
;
457 if (f_op
&& f_op
->poll
) {
458 wait_key_set(wait
, in
, out
,
460 mask
= (*f_op
->poll
)(f
.file
, wait
);
463 if ((mask
& POLLIN_SET
) && (in
& bit
)) {
468 if ((mask
& POLLOUT_SET
) && (out
& bit
)) {
473 if ((mask
& POLLEX_SET
) && (ex
& bit
)) {
478 /* got something, stop busy polling */
480 can_busy_loop
= false;
484 * only remember a returned
485 * POLL_BUSY_LOOP if we asked for it
487 } else if (busy_flag
& mask
)
488 can_busy_loop
= true;
501 if (retval
|| timed_out
|| signal_pending(current
))
504 retval
= table
.error
;
508 /* only if found POLL_BUSY_LOOP sockets && not out of time */
509 if (!need_resched() && can_busy_loop
&&
510 busy_loop_range(busy_start
, busy_end
))
514 * If this is the first loop and we have a timeout
515 * given, then we convert to ktime_t and set the to
516 * pointer to the expiry value.
518 if (end_time
&& !to
) {
519 expire
= timespec_to_ktime(*end_time
);
523 if (!poll_schedule_timeout(&table
, TASK_INTERRUPTIBLE
,
528 poll_freewait(&table
);
534 * We can actually return ERESTARTSYS instead of EINTR, but I'd
535 * like to be certain this leads to no problems. So I return
536 * EINTR just for safety.
538 * Update: ERESTARTSYS breaks at least the xview clock binary, so
539 * I'm trying ERESTARTNOHAND which restart only when you want to.
541 int core_sys_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
542 fd_set __user
*exp
, struct timespec
*end_time
)
549 /* Allocate small arguments on the stack to save memory and be faster */
550 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
556 /* max_fds can increase, so grab it once to avoid race */
558 fdt
= files_fdtable(current
->files
);
559 max_fds
= fdt
->max_fds
;
565 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
566 * since we used fdset we need to allocate memory in units of
571 if (size
> sizeof(stack_fds
) / 6) {
572 /* Not enough space in on-stack array; must use kmalloc */
574 bits
= kmalloc(6 * size
, GFP_KERNEL
);
579 fds
.out
= bits
+ size
;
580 fds
.ex
= bits
+ 2*size
;
581 fds
.res_in
= bits
+ 3*size
;
582 fds
.res_out
= bits
+ 4*size
;
583 fds
.res_ex
= bits
+ 5*size
;
585 if ((ret
= get_fd_set(n
, inp
, fds
.in
)) ||
586 (ret
= get_fd_set(n
, outp
, fds
.out
)) ||
587 (ret
= get_fd_set(n
, exp
, fds
.ex
)))
589 zero_fd_set(n
, fds
.res_in
);
590 zero_fd_set(n
, fds
.res_out
);
591 zero_fd_set(n
, fds
.res_ex
);
593 ret
= do_select(n
, &fds
, end_time
);
598 ret
= -ERESTARTNOHAND
;
599 if (signal_pending(current
))
604 if (set_fd_set(n
, inp
, fds
.res_in
) ||
605 set_fd_set(n
, outp
, fds
.res_out
) ||
606 set_fd_set(n
, exp
, fds
.res_ex
))
610 if (bits
!= stack_fds
)
616 SYSCALL_DEFINE5(select
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
617 fd_set __user
*, exp
, struct timeval __user
*, tvp
)
619 struct timespec end_time
, *to
= NULL
;
624 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
628 if (poll_select_set_timeout(to
,
629 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
630 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
634 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
635 ret
= poll_select_copy_remaining(&end_time
, tvp
, 1, ret
);
640 static long do_pselect(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
641 fd_set __user
*exp
, struct timespec __user
*tsp
,
642 const sigset_t __user
*sigmask
, size_t sigsetsize
)
644 sigset_t ksigmask
, sigsaved
;
645 struct timespec ts
, end_time
, *to
= NULL
;
649 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
653 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
658 /* XXX: Don't preclude handling different sized sigset_t's. */
659 if (sigsetsize
!= sizeof(sigset_t
))
661 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
664 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
665 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
668 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
669 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
671 if (ret
== -ERESTARTNOHAND
) {
673 * Don't restore the signal mask yet. Let do_signal() deliver
674 * the signal on the way back to userspace, before the signal
678 memcpy(¤t
->saved_sigmask
, &sigsaved
,
680 set_restore_sigmask();
683 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
689 * Most architectures can't handle 7-argument syscalls. So we provide a
690 * 6-argument version where the sixth argument is a pointer to a structure
691 * which has a pointer to the sigset_t itself followed by a size_t containing
694 SYSCALL_DEFINE6(pselect6
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
695 fd_set __user
*, exp
, struct timespec __user
*, tsp
,
698 size_t sigsetsize
= 0;
699 sigset_t __user
*up
= NULL
;
702 if (!access_ok(VERIFY_READ
, sig
, sizeof(void *)+sizeof(size_t))
703 || __get_user(up
, (sigset_t __user
* __user
*)sig
)
704 || __get_user(sigsetsize
,
705 (size_t __user
*)(sig
+sizeof(void *))))
709 return do_pselect(n
, inp
, outp
, exp
, tsp
, up
, sigsetsize
);
712 #ifdef __ARCH_WANT_SYS_OLD_SELECT
713 struct sel_arg_struct
{
715 fd_set __user
*inp
, *outp
, *exp
;
716 struct timeval __user
*tvp
;
719 SYSCALL_DEFINE1(old_select
, struct sel_arg_struct __user
*, arg
)
721 struct sel_arg_struct a
;
723 if (copy_from_user(&a
, arg
, sizeof(a
)))
725 return sys_select(a
.n
, a
.inp
, a
.outp
, a
.exp
, a
.tvp
);
730 struct poll_list
*next
;
732 struct pollfd entries
[0];
735 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
738 * Fish for pollable events on the pollfd->fd file descriptor. We're only
739 * interested in events matching the pollfd->events mask, and the result
740 * matching that mask is both recorded in pollfd->revents and returned. The
741 * pwait poll_table will be used by the fd-provided poll handler for waiting,
742 * if pwait->_qproc is non-NULL.
744 static inline unsigned int do_pollfd(struct pollfd
*pollfd
, poll_table
*pwait
,
746 unsigned int busy_flag
)
754 struct fd f
= fdget(fd
);
757 mask
= DEFAULT_POLLMASK
;
758 if (f
.file
->f_op
&& f
.file
->f_op
->poll
) {
759 pwait
->_key
= pollfd
->events
|POLLERR
|POLLHUP
;
760 pwait
->_key
|= busy_flag
;
761 mask
= f
.file
->f_op
->poll(f
.file
, pwait
);
762 if (mask
& busy_flag
)
763 *can_busy_poll
= true;
765 /* Mask out unneeded events. */
766 mask
&= pollfd
->events
| POLLERR
| POLLHUP
;
770 pollfd
->revents
= mask
;
775 static int do_poll(unsigned int nfds
, struct poll_list
*list
,
776 struct poll_wqueues
*wait
, struct timespec
*end_time
)
778 poll_table
* pt
= &wait
->pt
;
779 ktime_t expire
, *to
= NULL
;
780 int timed_out
= 0, count
= 0;
781 unsigned long slack
= 0;
782 unsigned int busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
783 u64 busy_start
= busy_loop_start_time(busy_flag
);
784 u64 busy_end
= busy_loop_end_time();
787 /* Optimise the no-wait case */
788 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
793 if (end_time
&& !timed_out
)
794 slack
= select_estimate_accuracy(end_time
);
797 struct poll_list
*walk
;
798 bool can_busy_loop
= false;
800 for (walk
= list
; walk
!= NULL
; walk
= walk
->next
) {
801 struct pollfd
* pfd
, * pfd_end
;
804 pfd_end
= pfd
+ walk
->len
;
805 for (; pfd
!= pfd_end
; pfd
++) {
807 * Fish for events. If we found one, record it
808 * and kill poll_table->_qproc, so we don't
809 * needlessly register any other waiters after
810 * this. They'll get immediately deregistered
811 * when we break out and return.
813 if (do_pollfd(pfd
, pt
, &can_busy_loop
,
817 /* found something, stop busy polling */
819 can_busy_loop
= false;
824 * All waiters have already been registered, so don't provide
825 * a poll_table->_qproc to them on the next loop iteration.
830 if (signal_pending(current
))
833 if (count
|| timed_out
)
836 /* only if found POLL_BUSY_LOOP sockets && not out of time */
837 if (!need_resched() && can_busy_loop
&&
838 busy_loop_range(busy_start
, busy_end
))
842 * If this is the first loop and we have a timeout
843 * given, then we convert to ktime_t and set the to
844 * pointer to the expiry value.
846 if (end_time
&& !to
) {
847 expire
= timespec_to_ktime(*end_time
);
851 if (!poll_schedule_timeout(wait
, TASK_INTERRUPTIBLE
, to
, slack
))
857 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
858 sizeof(struct pollfd))
860 int do_sys_poll(struct pollfd __user
*ufds
, unsigned int nfds
,
861 struct timespec
*end_time
)
863 struct poll_wqueues table
;
864 int err
= -EFAULT
, fdcount
, len
, size
;
865 /* Allocate small arguments on the stack to save memory and be
866 faster - use long to make sure the buffer is aligned properly
867 on 64 bit archs to avoid unaligned access */
868 long stack_pps
[POLL_STACK_ALLOC
/sizeof(long)];
869 struct poll_list
*const head
= (struct poll_list
*)stack_pps
;
870 struct poll_list
*walk
= head
;
871 unsigned long todo
= nfds
;
873 if (nfds
> rlimit(RLIMIT_NOFILE
))
876 len
= min_t(unsigned int, nfds
, N_STACK_PPS
);
883 if (copy_from_user(walk
->entries
, ufds
+ nfds
-todo
,
884 sizeof(struct pollfd
) * walk
->len
))
891 len
= min(todo
, POLLFD_PER_PAGE
);
892 size
= sizeof(struct poll_list
) + sizeof(struct pollfd
) * len
;
893 walk
= walk
->next
= kmalloc(size
, GFP_KERNEL
);
900 poll_initwait(&table
);
901 fdcount
= do_poll(nfds
, head
, &table
, end_time
);
902 poll_freewait(&table
);
904 for (walk
= head
; walk
; walk
= walk
->next
) {
905 struct pollfd
*fds
= walk
->entries
;
908 for (j
= 0; j
< walk
->len
; j
++, ufds
++)
909 if (__put_user(fds
[j
].revents
, &ufds
->revents
))
917 struct poll_list
*pos
= walk
;
925 static long do_restart_poll(struct restart_block
*restart_block
)
927 struct pollfd __user
*ufds
= restart_block
->poll
.ufds
;
928 int nfds
= restart_block
->poll
.nfds
;
929 struct timespec
*to
= NULL
, end_time
;
932 if (restart_block
->poll
.has_timeout
) {
933 end_time
.tv_sec
= restart_block
->poll
.tv_sec
;
934 end_time
.tv_nsec
= restart_block
->poll
.tv_nsec
;
938 ret
= do_sys_poll(ufds
, nfds
, to
);
941 restart_block
->fn
= do_restart_poll
;
942 ret
= -ERESTART_RESTARTBLOCK
;
947 SYSCALL_DEFINE3(poll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
950 struct timespec end_time
, *to
= NULL
;
953 if (timeout_msecs
>= 0) {
955 poll_select_set_timeout(to
, timeout_msecs
/ MSEC_PER_SEC
,
956 NSEC_PER_MSEC
* (timeout_msecs
% MSEC_PER_SEC
));
959 ret
= do_sys_poll(ufds
, nfds
, to
);
962 struct restart_block
*restart_block
;
964 restart_block
= ¤t_thread_info()->restart_block
;
965 restart_block
->fn
= do_restart_poll
;
966 restart_block
->poll
.ufds
= ufds
;
967 restart_block
->poll
.nfds
= nfds
;
969 if (timeout_msecs
>= 0) {
970 restart_block
->poll
.tv_sec
= end_time
.tv_sec
;
971 restart_block
->poll
.tv_nsec
= end_time
.tv_nsec
;
972 restart_block
->poll
.has_timeout
= 1;
974 restart_block
->poll
.has_timeout
= 0;
976 ret
= -ERESTART_RESTARTBLOCK
;
981 SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
982 struct timespec __user
*, tsp
, const sigset_t __user
*, sigmask
,
985 sigset_t ksigmask
, sigsaved
;
986 struct timespec ts
, end_time
, *to
= NULL
;
990 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
994 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
999 /* XXX: Don't preclude handling different sized sigset_t's. */
1000 if (sigsetsize
!= sizeof(sigset_t
))
1002 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
1005 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1006 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1009 ret
= do_sys_poll(ufds
, nfds
, to
);
1011 /* We can restart this syscall, usually */
1012 if (ret
== -EINTR
) {
1014 * Don't restore the signal mask yet. Let do_signal() deliver
1015 * the signal on the way back to userspace, before the signal
1019 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1021 set_restore_sigmask();
1023 ret
= -ERESTARTNOHAND
;
1025 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1027 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);