1 // SPDX-License-Identifier: GPL-2.0
3 * This file contains the procedures for the handling of select and poll
5 * Created for Linux based loosely upon Mathius Lattner's minix
6 * patches by Peter MacDonald. Heavily edited by Linus.
9 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
10 * flag set in its personality we do *not* modify the given timeout
11 * parameter to reflect time remaining.
14 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
15 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
18 #include <linux/kernel.h>
19 #include <linux/sched/signal.h>
20 #include <linux/sched/rt.h>
21 #include <linux/syscalls.h>
22 #include <linux/export.h>
23 #include <linux/slab.h>
24 #include <linux/poll.h>
25 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/rcupdate.h>
30 #include <linux/hrtimer.h>
31 #include <linux/freezer.h>
32 #include <net/busy_poll.h>
33 #include <linux/vmalloc.h>
35 #include <linux/uaccess.h>
39 * Estimate expected accuracy in ns from a timeval.
41 * After quite a bit of churning around, we've settled on
42 * a simple thing of taking 0.1% of the timeout as the
43 * slack, with a cap of 100 msec.
44 * "nice" tasks get a 0.5% slack instead.
46 * Consider this comment an open invitation to come up with even
50 #define MAX_SLACK (100 * NSEC_PER_MSEC)
52 static long __estimate_accuracy(struct timespec64
*tv
)
60 if (task_nice(current
) > 0)
61 divfactor
= divfactor
/ 5;
63 if (tv
->tv_sec
> MAX_SLACK
/ (NSEC_PER_SEC
/divfactor
))
66 slack
= tv
->tv_nsec
/ divfactor
;
67 slack
+= tv
->tv_sec
* (NSEC_PER_SEC
/divfactor
);
69 if (slack
> MAX_SLACK
)
75 u64
select_estimate_accuracy(struct timespec64
*tv
)
78 struct timespec64 now
;
81 * Realtime tasks get a slack of 0 for obvious reasons.
88 now
= timespec64_sub(*tv
, now
);
89 ret
= __estimate_accuracy(&now
);
90 if (ret
< current
->timer_slack_ns
)
91 return current
->timer_slack_ns
;
97 struct poll_table_page
{
98 struct poll_table_page
* next
;
99 struct poll_table_entry
* entry
;
100 struct poll_table_entry entries
[0];
103 #define POLL_TABLE_FULL(table) \
104 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
107 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
108 * I have rewritten this, taking some shortcuts: This code may not be easy to
109 * follow, but it should be free of race-conditions, and it's practical. If you
110 * understand what I'm doing here, then you understand how the linux
111 * sleep/wakeup mechanism works.
113 * Two very simple procedures, poll_wait() and poll_freewait() make all the
114 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
115 * as all select/poll functions have to call it to add an entry to the
118 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
121 void poll_initwait(struct poll_wqueues
*pwq
)
123 init_poll_funcptr(&pwq
->pt
, __pollwait
);
124 pwq
->polling_task
= current
;
128 pwq
->inline_index
= 0;
130 EXPORT_SYMBOL(poll_initwait
);
132 static void free_poll_entry(struct poll_table_entry
*entry
)
134 remove_wait_queue(entry
->wait_address
, &entry
->wait
);
138 void poll_freewait(struct poll_wqueues
*pwq
)
140 struct poll_table_page
* p
= pwq
->table
;
142 for (i
= 0; i
< pwq
->inline_index
; i
++)
143 free_poll_entry(pwq
->inline_entries
+ i
);
145 struct poll_table_entry
* entry
;
146 struct poll_table_page
*old
;
151 free_poll_entry(entry
);
152 } while (entry
> p
->entries
);
155 free_page((unsigned long) old
);
158 EXPORT_SYMBOL(poll_freewait
);
160 static struct poll_table_entry
*poll_get_entry(struct poll_wqueues
*p
)
162 struct poll_table_page
*table
= p
->table
;
164 if (p
->inline_index
< N_INLINE_POLL_ENTRIES
)
165 return p
->inline_entries
+ p
->inline_index
++;
167 if (!table
|| POLL_TABLE_FULL(table
)) {
168 struct poll_table_page
*new_table
;
170 new_table
= (struct poll_table_page
*) __get_free_page(GFP_KERNEL
);
175 new_table
->entry
= new_table
->entries
;
176 new_table
->next
= table
;
177 p
->table
= new_table
;
181 return table
->entry
++;
184 static int __pollwake(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
186 struct poll_wqueues
*pwq
= wait
->private;
187 DECLARE_WAITQUEUE(dummy_wait
, pwq
->polling_task
);
190 * Although this function is called under waitqueue lock, LOCK
191 * doesn't imply write barrier and the users expect write
192 * barrier semantics on wakeup functions. The following
193 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
194 * and is paired with smp_store_mb() in poll_schedule_timeout.
200 * Perform the default wake up operation using a dummy
203 * TODO: This is hacky but there currently is no interface to
204 * pass in @sync. @sync is scheduled to be removed and once
205 * that happens, wake_up_process() can be used directly.
207 return default_wake_function(&dummy_wait
, mode
, sync
, key
);
210 static int pollwake(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
212 struct poll_table_entry
*entry
;
214 entry
= container_of(wait
, struct poll_table_entry
, wait
);
215 if (key
&& !((unsigned long)key
& entry
->key
))
217 return __pollwake(wait
, mode
, sync
, key
);
220 /* Add a new entry */
221 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
224 struct poll_wqueues
*pwq
= container_of(p
, struct poll_wqueues
, pt
);
225 struct poll_table_entry
*entry
= poll_get_entry(pwq
);
228 entry
->filp
= get_file(filp
);
229 entry
->wait_address
= wait_address
;
230 entry
->key
= p
->_key
;
231 init_waitqueue_func_entry(&entry
->wait
, pollwake
);
232 entry
->wait
.private = pwq
;
233 add_wait_queue(wait_address
, &entry
->wait
);
236 int poll_schedule_timeout(struct poll_wqueues
*pwq
, int state
,
237 ktime_t
*expires
, unsigned long slack
)
241 set_current_state(state
);
243 rc
= schedule_hrtimeout_range(expires
, slack
, HRTIMER_MODE_ABS
);
244 __set_current_state(TASK_RUNNING
);
247 * Prepare for the next iteration.
249 * The following smp_store_mb() serves two purposes. First, it's
250 * the counterpart rmb of the wmb in pollwake() such that data
251 * written before wake up is always visible after wake up.
252 * Second, the full barrier guarantees that triggered clearing
253 * doesn't pass event check of the next iteration. Note that
254 * this problem doesn't exist for the first iteration as
255 * add_wait_queue() has full barrier semantics.
257 smp_store_mb(pwq
->triggered
, 0);
261 EXPORT_SYMBOL(poll_schedule_timeout
);
264 * poll_select_set_timeout - helper function to setup the timeout value
265 * @to: pointer to timespec64 variable for the final timeout
266 * @sec: seconds (from user space)
267 * @nsec: nanoseconds (from user space)
269 * Note, we do not use a timespec for the user space value here, That
270 * way we can use the function for timeval and compat interfaces as well.
272 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
274 int poll_select_set_timeout(struct timespec64
*to
, time64_t sec
, long nsec
)
276 struct timespec64 ts
= {.tv_sec
= sec
, .tv_nsec
= nsec
};
278 if (!timespec64_valid(&ts
))
281 /* Optimize for the zero timeout value here */
283 to
->tv_sec
= to
->tv_nsec
= 0;
286 *to
= timespec64_add_safe(*to
, ts
);
291 static int poll_select_copy_remaining(struct timespec64
*end_time
,
293 int timeval
, int ret
)
295 struct timespec64 rts64
;
302 if (current
->personality
& STICKY_TIMEOUTS
)
305 /* No update for zero timeout */
306 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
309 ktime_get_ts64(&rts64
);
310 rts64
= timespec64_sub(*end_time
, rts64
);
311 if (rts64
.tv_sec
< 0)
312 rts64
.tv_sec
= rts64
.tv_nsec
= 0;
314 rts
= timespec64_to_timespec(rts64
);
317 if (sizeof(rtv
) > sizeof(rtv
.tv_sec
) + sizeof(rtv
.tv_usec
))
318 memset(&rtv
, 0, sizeof(rtv
));
319 rtv
.tv_sec
= rts64
.tv_sec
;
320 rtv
.tv_usec
= rts64
.tv_nsec
/ NSEC_PER_USEC
;
322 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
325 } else if (!copy_to_user(p
, &rts
, sizeof(rts
)))
329 * If an application puts its timeval in read-only memory, we
330 * don't want the Linux-specific update to the timeval to
331 * cause a fault after the select has completed
332 * successfully. However, because we're not updating the
333 * timeval, we can't restart the system call.
337 if (ret
== -ERESTARTNOHAND
)
343 * Scalable version of the fd_set.
347 unsigned long *in
, *out
, *ex
;
348 unsigned long *res_in
, *res_out
, *res_ex
;
352 * How many longwords for "nr" bits?
354 #define FDS_BITPERLONG (8*sizeof(long))
355 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
356 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
359 * We do a VERIFY_WRITE here even though we are only reading this time:
360 * we'll write to it eventually..
362 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
365 int get_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
369 return copy_from_user(fdset
, ufdset
, nr
) ? -EFAULT
: 0;
371 memset(fdset
, 0, nr
);
375 static inline unsigned long __must_check
376 set_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
379 return __copy_to_user(ufdset
, fdset
, FDS_BYTES(nr
));
384 void zero_fd_set(unsigned long nr
, unsigned long *fdset
)
386 memset(fdset
, 0, FDS_BYTES(nr
));
389 #define FDS_IN(fds, n) (fds->in + n)
390 #define FDS_OUT(fds, n) (fds->out + n)
391 #define FDS_EX(fds, n) (fds->ex + n)
393 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
395 static int max_select_fd(unsigned long n
, fd_set_bits
*fds
)
397 unsigned long *open_fds
;
402 /* handle last in-complete long-word first */
403 set
= ~(~0UL << (n
& (BITS_PER_LONG
-1)));
405 fdt
= files_fdtable(current
->files
);
406 open_fds
= fdt
->open_fds
+ n
;
411 if (!(set
& ~*open_fds
))
422 if (set
& ~*open_fds
)
431 max
+= n
* BITS_PER_LONG
;
437 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
438 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
439 #define POLLEX_SET (POLLPRI)
441 static inline void wait_key_set(poll_table
*wait
, unsigned long in
,
442 unsigned long out
, unsigned long bit
,
443 unsigned int ll_flag
)
445 wait
->_key
= POLLEX_SET
| ll_flag
;
447 wait
->_key
|= POLLIN_SET
;
449 wait
->_key
|= POLLOUT_SET
;
452 static int do_select(int n
, fd_set_bits
*fds
, struct timespec64
*end_time
)
454 ktime_t expire
, *to
= NULL
;
455 struct poll_wqueues table
;
457 int retval
, i
, timed_out
= 0;
459 unsigned int busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
460 unsigned long busy_start
= 0;
463 retval
= max_select_fd(n
, fds
);
470 poll_initwait(&table
);
472 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
477 if (end_time
&& !timed_out
)
478 slack
= select_estimate_accuracy(end_time
);
482 unsigned long *rinp
, *routp
, *rexp
, *inp
, *outp
, *exp
;
483 bool can_busy_loop
= false;
485 inp
= fds
->in
; outp
= fds
->out
; exp
= fds
->ex
;
486 rinp
= fds
->res_in
; routp
= fds
->res_out
; rexp
= fds
->res_ex
;
488 for (i
= 0; i
< n
; ++rinp
, ++routp
, ++rexp
) {
489 unsigned long in
, out
, ex
, all_bits
, bit
= 1, mask
, j
;
490 unsigned long res_in
= 0, res_out
= 0, res_ex
= 0;
492 in
= *inp
++; out
= *outp
++; ex
= *exp
++;
493 all_bits
= in
| out
| ex
;
499 for (j
= 0; j
< BITS_PER_LONG
; ++j
, ++i
, bit
<<= 1) {
503 if (!(bit
& all_bits
))
507 const struct file_operations
*f_op
;
509 mask
= DEFAULT_POLLMASK
;
511 wait_key_set(wait
, in
, out
,
513 mask
= (*f_op
->poll
)(f
.file
, wait
);
516 if ((mask
& POLLIN_SET
) && (in
& bit
)) {
521 if ((mask
& POLLOUT_SET
) && (out
& bit
)) {
526 if ((mask
& POLLEX_SET
) && (ex
& bit
)) {
531 /* got something, stop busy polling */
533 can_busy_loop
= false;
537 * only remember a returned
538 * POLL_BUSY_LOOP if we asked for it
540 } else if (busy_flag
& mask
)
541 can_busy_loop
= true;
554 if (retval
|| timed_out
|| signal_pending(current
))
557 retval
= table
.error
;
561 /* only if found POLL_BUSY_LOOP sockets && not out of time */
562 if (can_busy_loop
&& !need_resched()) {
564 busy_start
= busy_loop_current_time();
567 if (!busy_loop_timeout(busy_start
))
573 * If this is the first loop and we have a timeout
574 * given, then we convert to ktime_t and set the to
575 * pointer to the expiry value.
577 if (end_time
&& !to
) {
578 expire
= timespec64_to_ktime(*end_time
);
582 if (!poll_schedule_timeout(&table
, TASK_INTERRUPTIBLE
,
587 poll_freewait(&table
);
593 * We can actually return ERESTARTSYS instead of EINTR, but I'd
594 * like to be certain this leads to no problems. So I return
595 * EINTR just for safety.
597 * Update: ERESTARTSYS breaks at least the xview clock binary, so
598 * I'm trying ERESTARTNOHAND which restart only when you want to.
600 int core_sys_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
601 fd_set __user
*exp
, struct timespec64
*end_time
)
606 size_t size
, alloc_size
;
608 /* Allocate small arguments on the stack to save memory and be faster */
609 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
615 /* max_fds can increase, so grab it once to avoid race */
617 fdt
= files_fdtable(current
->files
);
618 max_fds
= fdt
->max_fds
;
624 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
625 * since we used fdset we need to allocate memory in units of
630 if (size
> sizeof(stack_fds
) / 6) {
631 /* Not enough space in on-stack array; must use kmalloc */
633 if (size
> (SIZE_MAX
/ 6))
636 alloc_size
= 6 * size
;
637 bits
= kvmalloc(alloc_size
, GFP_KERNEL
);
642 fds
.out
= bits
+ size
;
643 fds
.ex
= bits
+ 2*size
;
644 fds
.res_in
= bits
+ 3*size
;
645 fds
.res_out
= bits
+ 4*size
;
646 fds
.res_ex
= bits
+ 5*size
;
648 if ((ret
= get_fd_set(n
, inp
, fds
.in
)) ||
649 (ret
= get_fd_set(n
, outp
, fds
.out
)) ||
650 (ret
= get_fd_set(n
, exp
, fds
.ex
)))
652 zero_fd_set(n
, fds
.res_in
);
653 zero_fd_set(n
, fds
.res_out
);
654 zero_fd_set(n
, fds
.res_ex
);
656 ret
= do_select(n
, &fds
, end_time
);
661 ret
= -ERESTARTNOHAND
;
662 if (signal_pending(current
))
667 if (set_fd_set(n
, inp
, fds
.res_in
) ||
668 set_fd_set(n
, outp
, fds
.res_out
) ||
669 set_fd_set(n
, exp
, fds
.res_ex
))
673 if (bits
!= stack_fds
)
679 SYSCALL_DEFINE5(select
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
680 fd_set __user
*, exp
, struct timeval __user
*, tvp
)
682 struct timespec64 end_time
, *to
= NULL
;
687 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
691 if (poll_select_set_timeout(to
,
692 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
693 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
697 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
698 ret
= poll_select_copy_remaining(&end_time
, tvp
, 1, ret
);
703 static long do_pselect(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
704 fd_set __user
*exp
, struct timespec __user
*tsp
,
705 const sigset_t __user
*sigmask
, size_t sigsetsize
)
707 sigset_t ksigmask
, sigsaved
;
709 struct timespec64 ts64
, end_time
, *to
= NULL
;
713 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
715 ts64
= timespec_to_timespec64(ts
);
718 if (poll_select_set_timeout(to
, ts64
.tv_sec
, ts64
.tv_nsec
))
723 /* XXX: Don't preclude handling different sized sigset_t's. */
724 if (sigsetsize
!= sizeof(sigset_t
))
726 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
729 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
730 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
733 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
734 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
736 if (ret
== -ERESTARTNOHAND
) {
738 * Don't restore the signal mask yet. Let do_signal() deliver
739 * the signal on the way back to userspace, before the signal
743 memcpy(¤t
->saved_sigmask
, &sigsaved
,
745 set_restore_sigmask();
748 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
754 * Most architectures can't handle 7-argument syscalls. So we provide a
755 * 6-argument version where the sixth argument is a pointer to a structure
756 * which has a pointer to the sigset_t itself followed by a size_t containing
759 SYSCALL_DEFINE6(pselect6
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
760 fd_set __user
*, exp
, struct timespec __user
*, tsp
,
763 size_t sigsetsize
= 0;
764 sigset_t __user
*up
= NULL
;
767 if (!access_ok(VERIFY_READ
, sig
, sizeof(void *)+sizeof(size_t))
768 || __get_user(up
, (sigset_t __user
* __user
*)sig
)
769 || __get_user(sigsetsize
,
770 (size_t __user
*)(sig
+sizeof(void *))))
774 return do_pselect(n
, inp
, outp
, exp
, tsp
, up
, sigsetsize
);
777 #ifdef __ARCH_WANT_SYS_OLD_SELECT
778 struct sel_arg_struct
{
780 fd_set __user
*inp
, *outp
, *exp
;
781 struct timeval __user
*tvp
;
784 SYSCALL_DEFINE1(old_select
, struct sel_arg_struct __user
*, arg
)
786 struct sel_arg_struct a
;
788 if (copy_from_user(&a
, arg
, sizeof(a
)))
790 return sys_select(a
.n
, a
.inp
, a
.outp
, a
.exp
, a
.tvp
);
795 struct poll_list
*next
;
797 struct pollfd entries
[0];
800 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
803 * Fish for pollable events on the pollfd->fd file descriptor. We're only
804 * interested in events matching the pollfd->events mask, and the result
805 * matching that mask is both recorded in pollfd->revents and returned. The
806 * pwait poll_table will be used by the fd-provided poll handler for waiting,
807 * if pwait->_qproc is non-NULL.
809 static inline unsigned int do_pollfd(struct pollfd
*pollfd
, poll_table
*pwait
,
811 unsigned int busy_flag
)
819 struct fd f
= fdget(fd
);
822 mask
= DEFAULT_POLLMASK
;
823 if (f
.file
->f_op
->poll
) {
824 pwait
->_key
= pollfd
->events
|POLLERR
|POLLHUP
;
825 pwait
->_key
|= busy_flag
;
826 mask
= f
.file
->f_op
->poll(f
.file
, pwait
);
827 if (mask
& busy_flag
)
828 *can_busy_poll
= true;
830 /* Mask out unneeded events. */
831 mask
&= pollfd
->events
| POLLERR
| POLLHUP
;
835 pollfd
->revents
= mask
;
840 static int do_poll(struct poll_list
*list
, struct poll_wqueues
*wait
,
841 struct timespec64
*end_time
)
843 poll_table
* pt
= &wait
->pt
;
844 ktime_t expire
, *to
= NULL
;
845 int timed_out
= 0, count
= 0;
847 unsigned int busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
848 unsigned long busy_start
= 0;
850 /* Optimise the no-wait case */
851 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
856 if (end_time
&& !timed_out
)
857 slack
= select_estimate_accuracy(end_time
);
860 struct poll_list
*walk
;
861 bool can_busy_loop
= false;
863 for (walk
= list
; walk
!= NULL
; walk
= walk
->next
) {
864 struct pollfd
* pfd
, * pfd_end
;
867 pfd_end
= pfd
+ walk
->len
;
868 for (; pfd
!= pfd_end
; pfd
++) {
870 * Fish for events. If we found one, record it
871 * and kill poll_table->_qproc, so we don't
872 * needlessly register any other waiters after
873 * this. They'll get immediately deregistered
874 * when we break out and return.
876 if (do_pollfd(pfd
, pt
, &can_busy_loop
,
880 /* found something, stop busy polling */
882 can_busy_loop
= false;
887 * All waiters have already been registered, so don't provide
888 * a poll_table->_qproc to them on the next loop iteration.
893 if (signal_pending(current
))
896 if (count
|| timed_out
)
899 /* only if found POLL_BUSY_LOOP sockets && not out of time */
900 if (can_busy_loop
&& !need_resched()) {
902 busy_start
= busy_loop_current_time();
905 if (!busy_loop_timeout(busy_start
))
911 * If this is the first loop and we have a timeout
912 * given, then we convert to ktime_t and set the to
913 * pointer to the expiry value.
915 if (end_time
&& !to
) {
916 expire
= timespec64_to_ktime(*end_time
);
920 if (!poll_schedule_timeout(wait
, TASK_INTERRUPTIBLE
, to
, slack
))
926 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
927 sizeof(struct pollfd))
929 static int do_sys_poll(struct pollfd __user
*ufds
, unsigned int nfds
,
930 struct timespec64
*end_time
)
932 struct poll_wqueues table
;
933 int err
= -EFAULT
, fdcount
, len
, size
;
934 /* Allocate small arguments on the stack to save memory and be
935 faster - use long to make sure the buffer is aligned properly
936 on 64 bit archs to avoid unaligned access */
937 long stack_pps
[POLL_STACK_ALLOC
/sizeof(long)];
938 struct poll_list
*const head
= (struct poll_list
*)stack_pps
;
939 struct poll_list
*walk
= head
;
940 unsigned long todo
= nfds
;
942 if (nfds
> rlimit(RLIMIT_NOFILE
))
945 len
= min_t(unsigned int, nfds
, N_STACK_PPS
);
952 if (copy_from_user(walk
->entries
, ufds
+ nfds
-todo
,
953 sizeof(struct pollfd
) * walk
->len
))
960 len
= min(todo
, POLLFD_PER_PAGE
);
961 size
= sizeof(struct poll_list
) + sizeof(struct pollfd
) * len
;
962 walk
= walk
->next
= kmalloc(size
, GFP_KERNEL
);
969 poll_initwait(&table
);
970 fdcount
= do_poll(head
, &table
, end_time
);
971 poll_freewait(&table
);
973 for (walk
= head
; walk
; walk
= walk
->next
) {
974 struct pollfd
*fds
= walk
->entries
;
977 for (j
= 0; j
< walk
->len
; j
++, ufds
++)
978 if (__put_user(fds
[j
].revents
, &ufds
->revents
))
986 struct poll_list
*pos
= walk
;
994 static long do_restart_poll(struct restart_block
*restart_block
)
996 struct pollfd __user
*ufds
= restart_block
->poll
.ufds
;
997 int nfds
= restart_block
->poll
.nfds
;
998 struct timespec64
*to
= NULL
, end_time
;
1001 if (restart_block
->poll
.has_timeout
) {
1002 end_time
.tv_sec
= restart_block
->poll
.tv_sec
;
1003 end_time
.tv_nsec
= restart_block
->poll
.tv_nsec
;
1007 ret
= do_sys_poll(ufds
, nfds
, to
);
1009 if (ret
== -EINTR
) {
1010 restart_block
->fn
= do_restart_poll
;
1011 ret
= -ERESTART_RESTARTBLOCK
;
1016 SYSCALL_DEFINE3(poll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1019 struct timespec64 end_time
, *to
= NULL
;
1022 if (timeout_msecs
>= 0) {
1024 poll_select_set_timeout(to
, timeout_msecs
/ MSEC_PER_SEC
,
1025 NSEC_PER_MSEC
* (timeout_msecs
% MSEC_PER_SEC
));
1028 ret
= do_sys_poll(ufds
, nfds
, to
);
1030 if (ret
== -EINTR
) {
1031 struct restart_block
*restart_block
;
1033 restart_block
= ¤t
->restart_block
;
1034 restart_block
->fn
= do_restart_poll
;
1035 restart_block
->poll
.ufds
= ufds
;
1036 restart_block
->poll
.nfds
= nfds
;
1038 if (timeout_msecs
>= 0) {
1039 restart_block
->poll
.tv_sec
= end_time
.tv_sec
;
1040 restart_block
->poll
.tv_nsec
= end_time
.tv_nsec
;
1041 restart_block
->poll
.has_timeout
= 1;
1043 restart_block
->poll
.has_timeout
= 0;
1045 ret
= -ERESTART_RESTARTBLOCK
;
1050 SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1051 struct timespec __user
*, tsp
, const sigset_t __user
*, sigmask
,
1054 sigset_t ksigmask
, sigsaved
;
1056 struct timespec64 end_time
, *to
= NULL
;
1060 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
1064 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1069 /* XXX: Don't preclude handling different sized sigset_t's. */
1070 if (sigsetsize
!= sizeof(sigset_t
))
1072 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
1075 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1076 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1079 ret
= do_sys_poll(ufds
, nfds
, to
);
1081 /* We can restart this syscall, usually */
1082 if (ret
== -EINTR
) {
1084 * Don't restore the signal mask yet. Let do_signal() deliver
1085 * the signal on the way back to userspace, before the signal
1089 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1091 set_restore_sigmask();
1093 ret
= -ERESTARTNOHAND
;
1095 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1097 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
1102 #ifdef CONFIG_COMPAT
1103 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1106 int compat_poll_select_copy_remaining(struct timespec
*end_time
, void __user
*p
,
1107 int timeval
, int ret
)
1114 if (current
->personality
& STICKY_TIMEOUTS
)
1117 /* No update for zero timeout */
1118 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
1122 ts
= timespec_sub(*end_time
, ts
);
1124 ts
.tv_sec
= ts
.tv_nsec
= 0;
1127 struct compat_timeval rtv
;
1129 rtv
.tv_sec
= ts
.tv_sec
;
1130 rtv
.tv_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
1132 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
1135 struct compat_timespec rts
;
1137 rts
.tv_sec
= ts
.tv_sec
;
1138 rts
.tv_nsec
= ts
.tv_nsec
;
1140 if (!copy_to_user(p
, &rts
, sizeof(rts
)))
1144 * If an application puts its timeval in read-only memory, we
1145 * don't want the Linux-specific update to the timeval to
1146 * cause a fault after the select has completed
1147 * successfully. However, because we're not updating the
1148 * timeval, we can't restart the system call.
1152 if (ret
== -ERESTARTNOHAND
)
1158 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1159 * 64-bit unsigned longs.
1162 int compat_get_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1163 unsigned long *fdset
)
1166 return compat_get_bitmap(fdset
, ufdset
, nr
);
1168 zero_fd_set(nr
, fdset
);
1174 int compat_set_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1175 unsigned long *fdset
)
1179 return compat_put_bitmap(ufdset
, fdset
, nr
);
1184 * This is a virtual copy of sys_select from fs/select.c and probably
1185 * should be compared to it from time to time
1189 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1190 * like to be certain this leads to no problems. So I return
1191 * EINTR just for safety.
1193 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1194 * I'm trying ERESTARTNOHAND which restart only when you want to.
1196 static int compat_core_sys_select(int n
, compat_ulong_t __user
*inp
,
1197 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1198 struct timespec
*end_time
)
1202 int size
, max_fds
, ret
= -EINVAL
;
1203 struct fdtable
*fdt
;
1204 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
1209 /* max_fds can increase, so grab it once to avoid race */
1211 fdt
= files_fdtable(current
->files
);
1212 max_fds
= fdt
->max_fds
;
1218 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1219 * since we used fdset we need to allocate memory in units of
1222 size
= FDS_BYTES(n
);
1224 if (size
> sizeof(stack_fds
) / 6) {
1225 bits
= kmalloc(6 * size
, GFP_KERNEL
);
1230 fds
.in
= (unsigned long *) bits
;
1231 fds
.out
= (unsigned long *) (bits
+ size
);
1232 fds
.ex
= (unsigned long *) (bits
+ 2*size
);
1233 fds
.res_in
= (unsigned long *) (bits
+ 3*size
);
1234 fds
.res_out
= (unsigned long *) (bits
+ 4*size
);
1235 fds
.res_ex
= (unsigned long *) (bits
+ 5*size
);
1237 if ((ret
= compat_get_fd_set(n
, inp
, fds
.in
)) ||
1238 (ret
= compat_get_fd_set(n
, outp
, fds
.out
)) ||
1239 (ret
= compat_get_fd_set(n
, exp
, fds
.ex
)))
1241 zero_fd_set(n
, fds
.res_in
);
1242 zero_fd_set(n
, fds
.res_out
);
1243 zero_fd_set(n
, fds
.res_ex
);
1245 ret
= do_select(n
, &fds
, end_time
);
1250 ret
= -ERESTARTNOHAND
;
1251 if (signal_pending(current
))
1256 if (compat_set_fd_set(n
, inp
, fds
.res_in
) ||
1257 compat_set_fd_set(n
, outp
, fds
.res_out
) ||
1258 compat_set_fd_set(n
, exp
, fds
.res_ex
))
1261 if (bits
!= stack_fds
)
1267 COMPAT_SYSCALL_DEFINE5(select
, int, n
, compat_ulong_t __user
*, inp
,
1268 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1269 struct compat_timeval __user
*, tvp
)
1271 struct timespec end_time
, *to
= NULL
;
1272 struct compat_timeval tv
;
1276 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
1280 if (poll_select_set_timeout(to
,
1281 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
1282 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
1286 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1287 ret
= compat_poll_select_copy_remaining(&end_time
, tvp
, 1, ret
);
1292 struct compat_sel_arg_struct
{
1300 COMPAT_SYSCALL_DEFINE1(old_select
, struct compat_sel_arg_struct __user
*, arg
)
1302 struct compat_sel_arg_struct a
;
1304 if (copy_from_user(&a
, arg
, sizeof(a
)))
1306 return compat_sys_select(a
.n
, compat_ptr(a
.inp
), compat_ptr(a
.outp
),
1307 compat_ptr(a
.exp
), compat_ptr(a
.tvp
));
1310 static long do_compat_pselect(int n
, compat_ulong_t __user
*inp
,
1311 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1312 struct compat_timespec __user
*tsp
, compat_sigset_t __user
*sigmask
,
1313 compat_size_t sigsetsize
)
1315 compat_sigset_t ss32
;
1316 sigset_t ksigmask
, sigsaved
;
1317 struct compat_timespec ts
;
1318 struct timespec end_time
, *to
= NULL
;
1322 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
1326 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1331 if (sigsetsize
!= sizeof(compat_sigset_t
))
1333 if (copy_from_user(&ss32
, sigmask
, sizeof(ss32
)))
1335 sigset_from_compat(&ksigmask
, &ss32
);
1337 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1338 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1341 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1342 ret
= compat_poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
1344 if (ret
== -ERESTARTNOHAND
) {
1346 * Don't restore the signal mask yet. Let do_signal() deliver
1347 * the signal on the way back to userspace, before the signal
1351 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1353 set_restore_sigmask();
1356 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1361 COMPAT_SYSCALL_DEFINE6(pselect6
, int, n
, compat_ulong_t __user
*, inp
,
1362 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1363 struct compat_timespec __user
*, tsp
, void __user
*, sig
)
1365 compat_size_t sigsetsize
= 0;
1366 compat_uptr_t up
= 0;
1369 if (!access_ok(VERIFY_READ
, sig
,
1370 sizeof(compat_uptr_t
)+sizeof(compat_size_t
)) ||
1371 __get_user(up
, (compat_uptr_t __user
*)sig
) ||
1372 __get_user(sigsetsize
,
1373 (compat_size_t __user
*)(sig
+sizeof(up
))))
1376 return do_compat_pselect(n
, inp
, outp
, exp
, tsp
, compat_ptr(up
),
1380 COMPAT_SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
,
1381 unsigned int, nfds
, struct compat_timespec __user
*, tsp
,
1382 const compat_sigset_t __user
*, sigmask
, compat_size_t
, sigsetsize
)
1384 compat_sigset_t ss32
;
1385 sigset_t ksigmask
, sigsaved
;
1386 struct compat_timespec ts
;
1387 struct timespec end_time
, *to
= NULL
;
1391 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
1395 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1400 if (sigsetsize
!= sizeof(compat_sigset_t
))
1402 if (copy_from_user(&ss32
, sigmask
, sizeof(ss32
)))
1404 sigset_from_compat(&ksigmask
, &ss32
);
1406 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1407 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1410 ret
= do_sys_poll(ufds
, nfds
, to
);
1412 /* We can restart this syscall, usually */
1413 if (ret
== -EINTR
) {
1415 * Don't restore the signal mask yet. Let do_signal() deliver
1416 * the signal on the way back to userspace, before the signal
1420 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1422 set_restore_sigmask();
1424 ret
= -ERESTARTNOHAND
;
1426 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1428 ret
= compat_poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);