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
&& !(key_to_poll(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 rts
;
301 if (current
->personality
& STICKY_TIMEOUTS
)
304 /* No update for zero timeout */
305 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
308 ktime_get_ts64(&rts
);
309 rts
= timespec64_sub(*end_time
, rts
);
311 rts
.tv_sec
= rts
.tv_nsec
= 0;
315 if (sizeof(rtv
) > sizeof(rtv
.tv_sec
) + sizeof(rtv
.tv_usec
))
316 memset(&rtv
, 0, sizeof(rtv
));
317 rtv
.tv_sec
= rts
.tv_sec
;
318 rtv
.tv_usec
= rts
.tv_nsec
/ NSEC_PER_USEC
;
320 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
323 } else if (!put_timespec64(&rts
, p
))
327 * If an application puts its timeval in read-only memory, we
328 * don't want the Linux-specific update to the timeval to
329 * cause a fault after the select has completed
330 * successfully. However, because we're not updating the
331 * timeval, we can't restart the system call.
335 if (ret
== -ERESTARTNOHAND
)
341 * Scalable version of the fd_set.
345 unsigned long *in
, *out
, *ex
;
346 unsigned long *res_in
, *res_out
, *res_ex
;
350 * How many longwords for "nr" bits?
352 #define FDS_BITPERLONG (8*sizeof(long))
353 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
354 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
357 * We do a VERIFY_WRITE here even though we are only reading this time:
358 * we'll write to it eventually..
360 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
363 int get_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
367 return copy_from_user(fdset
, ufdset
, nr
) ? -EFAULT
: 0;
369 memset(fdset
, 0, nr
);
373 static inline unsigned long __must_check
374 set_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
377 return __copy_to_user(ufdset
, fdset
, FDS_BYTES(nr
));
382 void zero_fd_set(unsigned long nr
, unsigned long *fdset
)
384 memset(fdset
, 0, FDS_BYTES(nr
));
387 #define FDS_IN(fds, n) (fds->in + n)
388 #define FDS_OUT(fds, n) (fds->out + n)
389 #define FDS_EX(fds, n) (fds->ex + n)
391 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
393 static int max_select_fd(unsigned long n
, fd_set_bits
*fds
)
395 unsigned long *open_fds
;
400 /* handle last in-complete long-word first */
401 set
= ~(~0UL << (n
& (BITS_PER_LONG
-1)));
403 fdt
= files_fdtable(current
->files
);
404 open_fds
= fdt
->open_fds
+ n
;
409 if (!(set
& ~*open_fds
))
420 if (set
& ~*open_fds
)
429 max
+= n
* BITS_PER_LONG
;
435 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR)
436 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
437 #define POLLEX_SET (EPOLLPRI)
439 static inline void wait_key_set(poll_table
*wait
, unsigned long in
,
440 unsigned long out
, unsigned long bit
,
443 wait
->_key
= POLLEX_SET
| ll_flag
;
445 wait
->_key
|= POLLIN_SET
;
447 wait
->_key
|= POLLOUT_SET
;
450 static int do_select(int n
, fd_set_bits
*fds
, struct timespec64
*end_time
)
452 ktime_t expire
, *to
= NULL
;
453 struct poll_wqueues table
;
455 int retval
, i
, timed_out
= 0;
457 __poll_t busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
458 unsigned long busy_start
= 0;
461 retval
= max_select_fd(n
, fds
);
468 poll_initwait(&table
);
470 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
475 if (end_time
&& !timed_out
)
476 slack
= select_estimate_accuracy(end_time
);
480 unsigned long *rinp
, *routp
, *rexp
, *inp
, *outp
, *exp
;
481 bool can_busy_loop
= false;
483 inp
= fds
->in
; outp
= fds
->out
; exp
= fds
->ex
;
484 rinp
= fds
->res_in
; routp
= fds
->res_out
; rexp
= fds
->res_ex
;
486 for (i
= 0; i
< n
; ++rinp
, ++routp
, ++rexp
) {
487 unsigned long in
, out
, ex
, all_bits
, bit
= 1, j
;
488 unsigned long res_in
= 0, res_out
= 0, res_ex
= 0;
491 in
= *inp
++; out
= *outp
++; ex
= *exp
++;
492 all_bits
= in
| out
| ex
;
498 for (j
= 0; j
< BITS_PER_LONG
; ++j
, ++i
, bit
<<= 1) {
502 if (!(bit
& all_bits
))
506 const struct file_operations
*f_op
;
508 mask
= DEFAULT_POLLMASK
;
510 wait_key_set(wait
, in
, out
,
512 mask
= (*f_op
->poll
)(f
.file
, wait
);
515 if ((mask
& POLLIN_SET
) && (in
& bit
)) {
520 if ((mask
& POLLOUT_SET
) && (out
& bit
)) {
525 if ((mask
& POLLEX_SET
) && (ex
& bit
)) {
530 /* got something, stop busy polling */
532 can_busy_loop
= false;
536 * only remember a returned
537 * POLL_BUSY_LOOP if we asked for it
539 } else if (busy_flag
& mask
)
540 can_busy_loop
= true;
553 if (retval
|| timed_out
|| signal_pending(current
))
556 retval
= table
.error
;
560 /* only if found POLL_BUSY_LOOP sockets && not out of time */
561 if (can_busy_loop
&& !need_resched()) {
563 busy_start
= busy_loop_current_time();
566 if (!busy_loop_timeout(busy_start
))
572 * If this is the first loop and we have a timeout
573 * given, then we convert to ktime_t and set the to
574 * pointer to the expiry value.
576 if (end_time
&& !to
) {
577 expire
= timespec64_to_ktime(*end_time
);
581 if (!poll_schedule_timeout(&table
, TASK_INTERRUPTIBLE
,
586 poll_freewait(&table
);
592 * We can actually return ERESTARTSYS instead of EINTR, but I'd
593 * like to be certain this leads to no problems. So I return
594 * EINTR just for safety.
596 * Update: ERESTARTSYS breaks at least the xview clock binary, so
597 * I'm trying ERESTARTNOHAND which restart only when you want to.
599 int core_sys_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
600 fd_set __user
*exp
, struct timespec64
*end_time
)
605 size_t size
, alloc_size
;
607 /* Allocate small arguments on the stack to save memory and be faster */
608 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
614 /* max_fds can increase, so grab it once to avoid race */
616 fdt
= files_fdtable(current
->files
);
617 max_fds
= fdt
->max_fds
;
623 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
624 * since we used fdset we need to allocate memory in units of
629 if (size
> sizeof(stack_fds
) / 6) {
630 /* Not enough space in on-stack array; must use kmalloc */
632 if (size
> (SIZE_MAX
/ 6))
635 alloc_size
= 6 * size
;
636 bits
= kvmalloc(alloc_size
, GFP_KERNEL
);
641 fds
.out
= bits
+ size
;
642 fds
.ex
= bits
+ 2*size
;
643 fds
.res_in
= bits
+ 3*size
;
644 fds
.res_out
= bits
+ 4*size
;
645 fds
.res_ex
= bits
+ 5*size
;
647 if ((ret
= get_fd_set(n
, inp
, fds
.in
)) ||
648 (ret
= get_fd_set(n
, outp
, fds
.out
)) ||
649 (ret
= get_fd_set(n
, exp
, fds
.ex
)))
651 zero_fd_set(n
, fds
.res_in
);
652 zero_fd_set(n
, fds
.res_out
);
653 zero_fd_set(n
, fds
.res_ex
);
655 ret
= do_select(n
, &fds
, end_time
);
660 ret
= -ERESTARTNOHAND
;
661 if (signal_pending(current
))
666 if (set_fd_set(n
, inp
, fds
.res_in
) ||
667 set_fd_set(n
, outp
, fds
.res_out
) ||
668 set_fd_set(n
, exp
, fds
.res_ex
))
672 if (bits
!= stack_fds
)
678 static int kern_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
679 fd_set __user
*exp
, struct timeval __user
*tvp
)
681 struct timespec64 end_time
, *to
= NULL
;
686 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
690 if (poll_select_set_timeout(to
,
691 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
692 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
696 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
697 ret
= poll_select_copy_remaining(&end_time
, tvp
, 1, ret
);
702 SYSCALL_DEFINE5(select
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
703 fd_set __user
*, exp
, struct timeval __user
*, tvp
)
705 return kern_select(n
, inp
, outp
, exp
, tvp
);
708 static long do_pselect(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
709 fd_set __user
*exp
, struct timespec __user
*tsp
,
710 const sigset_t __user
*sigmask
, size_t sigsetsize
)
712 sigset_t ksigmask
, sigsaved
;
713 struct timespec64 ts
, end_time
, *to
= NULL
;
717 if (get_timespec64(&ts
, tsp
))
721 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
726 /* XXX: Don't preclude handling different sized sigset_t's. */
727 if (sigsetsize
!= sizeof(sigset_t
))
729 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
732 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
733 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
736 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
737 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
739 if (ret
== -ERESTARTNOHAND
) {
741 * Don't restore the signal mask yet. Let do_signal() deliver
742 * the signal on the way back to userspace, before the signal
746 memcpy(¤t
->saved_sigmask
, &sigsaved
,
748 set_restore_sigmask();
751 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
757 * Most architectures can't handle 7-argument syscalls. So we provide a
758 * 6-argument version where the sixth argument is a pointer to a structure
759 * which has a pointer to the sigset_t itself followed by a size_t containing
762 SYSCALL_DEFINE6(pselect6
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
763 fd_set __user
*, exp
, struct timespec __user
*, tsp
,
766 size_t sigsetsize
= 0;
767 sigset_t __user
*up
= NULL
;
770 if (!access_ok(VERIFY_READ
, sig
, sizeof(void *)+sizeof(size_t))
771 || __get_user(up
, (sigset_t __user
* __user
*)sig
)
772 || __get_user(sigsetsize
,
773 (size_t __user
*)(sig
+sizeof(void *))))
777 return do_pselect(n
, inp
, outp
, exp
, tsp
, up
, sigsetsize
);
780 #ifdef __ARCH_WANT_SYS_OLD_SELECT
781 struct sel_arg_struct
{
783 fd_set __user
*inp
, *outp
, *exp
;
784 struct timeval __user
*tvp
;
787 SYSCALL_DEFINE1(old_select
, struct sel_arg_struct __user
*, arg
)
789 struct sel_arg_struct a
;
791 if (copy_from_user(&a
, arg
, sizeof(a
)))
793 return kern_select(a
.n
, a
.inp
, a
.outp
, a
.exp
, a
.tvp
);
798 struct poll_list
*next
;
800 struct pollfd entries
[0];
803 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
806 * Fish for pollable events on the pollfd->fd file descriptor. We're only
807 * interested in events matching the pollfd->events mask, and the result
808 * matching that mask is both recorded in pollfd->revents and returned. The
809 * pwait poll_table will be used by the fd-provided poll handler for waiting,
810 * if pwait->_qproc is non-NULL.
812 static inline __poll_t
do_pollfd(struct pollfd
*pollfd
, poll_table
*pwait
,
822 struct fd f
= fdget(fd
);
825 /* userland u16 ->events contains POLL... bitmap */
826 __poll_t filter
= demangle_poll(pollfd
->events
) |
828 mask
= DEFAULT_POLLMASK
;
829 if (f
.file
->f_op
->poll
) {
830 pwait
->_key
= filter
;
831 pwait
->_key
|= busy_flag
;
832 mask
= f
.file
->f_op
->poll(f
.file
, pwait
);
833 if (mask
& busy_flag
)
834 *can_busy_poll
= true;
836 /* Mask out unneeded events. */
841 /* ... and so does ->revents */
842 pollfd
->revents
= mangle_poll(mask
);
847 static int do_poll(struct poll_list
*list
, struct poll_wqueues
*wait
,
848 struct timespec64
*end_time
)
850 poll_table
* pt
= &wait
->pt
;
851 ktime_t expire
, *to
= NULL
;
852 int timed_out
= 0, count
= 0;
854 __poll_t busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
855 unsigned long busy_start
= 0;
857 /* Optimise the no-wait case */
858 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
863 if (end_time
&& !timed_out
)
864 slack
= select_estimate_accuracy(end_time
);
867 struct poll_list
*walk
;
868 bool can_busy_loop
= false;
870 for (walk
= list
; walk
!= NULL
; walk
= walk
->next
) {
871 struct pollfd
* pfd
, * pfd_end
;
874 pfd_end
= pfd
+ walk
->len
;
875 for (; pfd
!= pfd_end
; pfd
++) {
877 * Fish for events. If we found one, record it
878 * and kill poll_table->_qproc, so we don't
879 * needlessly register any other waiters after
880 * this. They'll get immediately deregistered
881 * when we break out and return.
883 if (do_pollfd(pfd
, pt
, &can_busy_loop
,
887 /* found something, stop busy polling */
889 can_busy_loop
= false;
894 * All waiters have already been registered, so don't provide
895 * a poll_table->_qproc to them on the next loop iteration.
900 if (signal_pending(current
))
903 if (count
|| timed_out
)
906 /* only if found POLL_BUSY_LOOP sockets && not out of time */
907 if (can_busy_loop
&& !need_resched()) {
909 busy_start
= busy_loop_current_time();
912 if (!busy_loop_timeout(busy_start
))
918 * If this is the first loop and we have a timeout
919 * given, then we convert to ktime_t and set the to
920 * pointer to the expiry value.
922 if (end_time
&& !to
) {
923 expire
= timespec64_to_ktime(*end_time
);
927 if (!poll_schedule_timeout(wait
, TASK_INTERRUPTIBLE
, to
, slack
))
933 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
934 sizeof(struct pollfd))
936 static int do_sys_poll(struct pollfd __user
*ufds
, unsigned int nfds
,
937 struct timespec64
*end_time
)
939 struct poll_wqueues table
;
940 int err
= -EFAULT
, fdcount
, len
, size
;
941 /* Allocate small arguments on the stack to save memory and be
942 faster - use long to make sure the buffer is aligned properly
943 on 64 bit archs to avoid unaligned access */
944 long stack_pps
[POLL_STACK_ALLOC
/sizeof(long)];
945 struct poll_list
*const head
= (struct poll_list
*)stack_pps
;
946 struct poll_list
*walk
= head
;
947 unsigned long todo
= nfds
;
949 if (nfds
> rlimit(RLIMIT_NOFILE
))
952 len
= min_t(unsigned int, nfds
, N_STACK_PPS
);
959 if (copy_from_user(walk
->entries
, ufds
+ nfds
-todo
,
960 sizeof(struct pollfd
) * walk
->len
))
967 len
= min(todo
, POLLFD_PER_PAGE
);
968 size
= sizeof(struct poll_list
) + sizeof(struct pollfd
) * len
;
969 walk
= walk
->next
= kmalloc(size
, GFP_KERNEL
);
976 poll_initwait(&table
);
977 fdcount
= do_poll(head
, &table
, end_time
);
978 poll_freewait(&table
);
980 for (walk
= head
; walk
; walk
= walk
->next
) {
981 struct pollfd
*fds
= walk
->entries
;
984 for (j
= 0; j
< walk
->len
; j
++, ufds
++)
985 if (__put_user(fds
[j
].revents
, &ufds
->revents
))
993 struct poll_list
*pos
= walk
;
1001 static long do_restart_poll(struct restart_block
*restart_block
)
1003 struct pollfd __user
*ufds
= restart_block
->poll
.ufds
;
1004 int nfds
= restart_block
->poll
.nfds
;
1005 struct timespec64
*to
= NULL
, end_time
;
1008 if (restart_block
->poll
.has_timeout
) {
1009 end_time
.tv_sec
= restart_block
->poll
.tv_sec
;
1010 end_time
.tv_nsec
= restart_block
->poll
.tv_nsec
;
1014 ret
= do_sys_poll(ufds
, nfds
, to
);
1016 if (ret
== -EINTR
) {
1017 restart_block
->fn
= do_restart_poll
;
1018 ret
= -ERESTART_RESTARTBLOCK
;
1023 SYSCALL_DEFINE3(poll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1026 struct timespec64 end_time
, *to
= NULL
;
1029 if (timeout_msecs
>= 0) {
1031 poll_select_set_timeout(to
, timeout_msecs
/ MSEC_PER_SEC
,
1032 NSEC_PER_MSEC
* (timeout_msecs
% MSEC_PER_SEC
));
1035 ret
= do_sys_poll(ufds
, nfds
, to
);
1037 if (ret
== -EINTR
) {
1038 struct restart_block
*restart_block
;
1040 restart_block
= ¤t
->restart_block
;
1041 restart_block
->fn
= do_restart_poll
;
1042 restart_block
->poll
.ufds
= ufds
;
1043 restart_block
->poll
.nfds
= nfds
;
1045 if (timeout_msecs
>= 0) {
1046 restart_block
->poll
.tv_sec
= end_time
.tv_sec
;
1047 restart_block
->poll
.tv_nsec
= end_time
.tv_nsec
;
1048 restart_block
->poll
.has_timeout
= 1;
1050 restart_block
->poll
.has_timeout
= 0;
1052 ret
= -ERESTART_RESTARTBLOCK
;
1057 SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1058 struct timespec __user
*, tsp
, const sigset_t __user
*, sigmask
,
1061 sigset_t ksigmask
, sigsaved
;
1062 struct timespec64 ts
, end_time
, *to
= NULL
;
1066 if (get_timespec64(&ts
, tsp
))
1070 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1075 /* XXX: Don't preclude handling different sized sigset_t's. */
1076 if (sigsetsize
!= sizeof(sigset_t
))
1078 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
1081 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1082 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1085 ret
= do_sys_poll(ufds
, nfds
, to
);
1087 /* We can restart this syscall, usually */
1088 if (ret
== -EINTR
) {
1090 * Don't restore the signal mask yet. Let do_signal() deliver
1091 * the signal on the way back to userspace, before the signal
1095 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1097 set_restore_sigmask();
1099 ret
= -ERESTARTNOHAND
;
1101 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1103 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
1108 #ifdef CONFIG_COMPAT
1109 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1112 int compat_poll_select_copy_remaining(struct timespec64
*end_time
, void __user
*p
,
1113 int timeval
, int ret
)
1115 struct timespec64 ts
;
1120 if (current
->personality
& STICKY_TIMEOUTS
)
1123 /* No update for zero timeout */
1124 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
1127 ktime_get_ts64(&ts
);
1128 ts
= timespec64_sub(*end_time
, ts
);
1130 ts
.tv_sec
= ts
.tv_nsec
= 0;
1133 struct compat_timeval rtv
;
1135 rtv
.tv_sec
= ts
.tv_sec
;
1136 rtv
.tv_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
1138 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
1141 if (!compat_put_timespec64(&ts
, p
))
1145 * If an application puts its timeval in read-only memory, we
1146 * don't want the Linux-specific update to the timeval to
1147 * cause a fault after the select has completed
1148 * successfully. However, because we're not updating the
1149 * timeval, we can't restart the system call.
1153 if (ret
== -ERESTARTNOHAND
)
1159 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1160 * 64-bit unsigned longs.
1163 int compat_get_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1164 unsigned long *fdset
)
1167 return compat_get_bitmap(fdset
, ufdset
, nr
);
1169 zero_fd_set(nr
, fdset
);
1175 int compat_set_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1176 unsigned long *fdset
)
1180 return compat_put_bitmap(ufdset
, fdset
, nr
);
1185 * This is a virtual copy of sys_select from fs/select.c and probably
1186 * should be compared to it from time to time
1190 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1191 * like to be certain this leads to no problems. So I return
1192 * EINTR just for safety.
1194 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1195 * I'm trying ERESTARTNOHAND which restart only when you want to.
1197 static int compat_core_sys_select(int n
, compat_ulong_t __user
*inp
,
1198 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1199 struct timespec64
*end_time
)
1203 int size
, max_fds
, ret
= -EINVAL
;
1204 struct fdtable
*fdt
;
1205 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
1210 /* max_fds can increase, so grab it once to avoid race */
1212 fdt
= files_fdtable(current
->files
);
1213 max_fds
= fdt
->max_fds
;
1219 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1220 * since we used fdset we need to allocate memory in units of
1223 size
= FDS_BYTES(n
);
1225 if (size
> sizeof(stack_fds
) / 6) {
1226 bits
= kmalloc(6 * size
, GFP_KERNEL
);
1231 fds
.in
= (unsigned long *) bits
;
1232 fds
.out
= (unsigned long *) (bits
+ size
);
1233 fds
.ex
= (unsigned long *) (bits
+ 2*size
);
1234 fds
.res_in
= (unsigned long *) (bits
+ 3*size
);
1235 fds
.res_out
= (unsigned long *) (bits
+ 4*size
);
1236 fds
.res_ex
= (unsigned long *) (bits
+ 5*size
);
1238 if ((ret
= compat_get_fd_set(n
, inp
, fds
.in
)) ||
1239 (ret
= compat_get_fd_set(n
, outp
, fds
.out
)) ||
1240 (ret
= compat_get_fd_set(n
, exp
, fds
.ex
)))
1242 zero_fd_set(n
, fds
.res_in
);
1243 zero_fd_set(n
, fds
.res_out
);
1244 zero_fd_set(n
, fds
.res_ex
);
1246 ret
= do_select(n
, &fds
, end_time
);
1251 ret
= -ERESTARTNOHAND
;
1252 if (signal_pending(current
))
1257 if (compat_set_fd_set(n
, inp
, fds
.res_in
) ||
1258 compat_set_fd_set(n
, outp
, fds
.res_out
) ||
1259 compat_set_fd_set(n
, exp
, fds
.res_ex
))
1262 if (bits
!= stack_fds
)
1268 static int do_compat_select(int n
, compat_ulong_t __user
*inp
,
1269 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1270 struct compat_timeval __user
*tvp
)
1272 struct timespec64 end_time
, *to
= NULL
;
1273 struct compat_timeval tv
;
1277 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
1281 if (poll_select_set_timeout(to
,
1282 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
1283 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
1287 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1288 ret
= compat_poll_select_copy_remaining(&end_time
, tvp
, 1, ret
);
1293 COMPAT_SYSCALL_DEFINE5(select
, int, n
, compat_ulong_t __user
*, inp
,
1294 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1295 struct compat_timeval __user
*, tvp
)
1297 return do_compat_select(n
, inp
, outp
, exp
, tvp
);
1300 struct compat_sel_arg_struct
{
1308 COMPAT_SYSCALL_DEFINE1(old_select
, struct compat_sel_arg_struct __user
*, arg
)
1310 struct compat_sel_arg_struct a
;
1312 if (copy_from_user(&a
, arg
, sizeof(a
)))
1314 return do_compat_select(a
.n
, compat_ptr(a
.inp
), compat_ptr(a
.outp
),
1315 compat_ptr(a
.exp
), compat_ptr(a
.tvp
));
1318 static long do_compat_pselect(int n
, compat_ulong_t __user
*inp
,
1319 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1320 struct compat_timespec __user
*tsp
, compat_sigset_t __user
*sigmask
,
1321 compat_size_t sigsetsize
)
1323 sigset_t ksigmask
, sigsaved
;
1324 struct timespec64 ts
, end_time
, *to
= NULL
;
1328 if (compat_get_timespec64(&ts
, tsp
))
1332 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1337 if (sigsetsize
!= sizeof(compat_sigset_t
))
1339 if (get_compat_sigset(&ksigmask
, sigmask
))
1342 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1343 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1346 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1347 ret
= compat_poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
1349 if (ret
== -ERESTARTNOHAND
) {
1351 * Don't restore the signal mask yet. Let do_signal() deliver
1352 * the signal on the way back to userspace, before the signal
1356 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1358 set_restore_sigmask();
1361 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1366 COMPAT_SYSCALL_DEFINE6(pselect6
, int, n
, compat_ulong_t __user
*, inp
,
1367 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1368 struct compat_timespec __user
*, tsp
, void __user
*, sig
)
1370 compat_size_t sigsetsize
= 0;
1371 compat_uptr_t up
= 0;
1374 if (!access_ok(VERIFY_READ
, sig
,
1375 sizeof(compat_uptr_t
)+sizeof(compat_size_t
)) ||
1376 __get_user(up
, (compat_uptr_t __user
*)sig
) ||
1377 __get_user(sigsetsize
,
1378 (compat_size_t __user
*)(sig
+sizeof(up
))))
1381 return do_compat_pselect(n
, inp
, outp
, exp
, tsp
, compat_ptr(up
),
1385 COMPAT_SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
,
1386 unsigned int, nfds
, struct compat_timespec __user
*, tsp
,
1387 const compat_sigset_t __user
*, sigmask
, compat_size_t
, sigsetsize
)
1389 sigset_t ksigmask
, sigsaved
;
1390 struct timespec64 ts
, end_time
, *to
= NULL
;
1394 if (compat_get_timespec64(&ts
, tsp
))
1398 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1403 if (sigsetsize
!= sizeof(compat_sigset_t
))
1405 if (get_compat_sigset(&ksigmask
, sigmask
))
1408 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1409 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1412 ret
= do_sys_poll(ufds
, nfds
, to
);
1414 /* We can restart this syscall, usually */
1415 if (ret
== -EINTR
) {
1417 * Don't restore the signal mask yet. Let do_signal() deliver
1418 * the signal on the way back to userspace, before the signal
1422 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1424 set_restore_sigmask();
1426 ret
= -ERESTARTNOHAND
;
1428 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1430 ret
= compat_poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);