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/signal.h>
19 #include <linux/sched/rt.h>
20 #include <linux/syscalls.h>
21 #include <linux/export.h>
22 #include <linux/slab.h>
23 #include <linux/poll.h>
24 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
25 #include <linux/file.h>
26 #include <linux/fdtable.h>
28 #include <linux/rcupdate.h>
29 #include <linux/hrtimer.h>
30 #include <linux/freezer.h>
31 #include <net/busy_poll.h>
32 #include <linux/vmalloc.h>
34 #include <linux/uaccess.h>
38 * Estimate expected accuracy in ns from a timeval.
40 * After quite a bit of churning around, we've settled on
41 * a simple thing of taking 0.1% of the timeout as the
42 * slack, with a cap of 100 msec.
43 * "nice" tasks get a 0.5% slack instead.
45 * Consider this comment an open invitation to come up with even
49 #define MAX_SLACK (100 * NSEC_PER_MSEC)
51 static long __estimate_accuracy(struct timespec64
*tv
)
59 if (task_nice(current
) > 0)
60 divfactor
= divfactor
/ 5;
62 if (tv
->tv_sec
> MAX_SLACK
/ (NSEC_PER_SEC
/divfactor
))
65 slack
= tv
->tv_nsec
/ divfactor
;
66 slack
+= tv
->tv_sec
* (NSEC_PER_SEC
/divfactor
);
68 if (slack
> MAX_SLACK
)
74 u64
select_estimate_accuracy(struct timespec64
*tv
)
77 struct timespec64 now
;
80 * Realtime tasks get a slack of 0 for obvious reasons.
87 now
= timespec64_sub(*tv
, now
);
88 ret
= __estimate_accuracy(&now
);
89 if (ret
< current
->timer_slack_ns
)
90 return current
->timer_slack_ns
;
96 struct poll_table_page
{
97 struct poll_table_page
* next
;
98 struct poll_table_entry
* entry
;
99 struct poll_table_entry entries
[0];
102 #define POLL_TABLE_FULL(table) \
103 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
106 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
107 * I have rewritten this, taking some shortcuts: This code may not be easy to
108 * follow, but it should be free of race-conditions, and it's practical. If you
109 * understand what I'm doing here, then you understand how the linux
110 * sleep/wakeup mechanism works.
112 * Two very simple procedures, poll_wait() and poll_freewait() make all the
113 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
114 * as all select/poll functions have to call it to add an entry to the
117 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
120 void poll_initwait(struct poll_wqueues
*pwq
)
122 init_poll_funcptr(&pwq
->pt
, __pollwait
);
123 pwq
->polling_task
= current
;
127 pwq
->inline_index
= 0;
129 EXPORT_SYMBOL(poll_initwait
);
131 static void free_poll_entry(struct poll_table_entry
*entry
)
133 remove_wait_queue(entry
->wait_address
, &entry
->wait
);
137 void poll_freewait(struct poll_wqueues
*pwq
)
139 struct poll_table_page
* p
= pwq
->table
;
141 for (i
= 0; i
< pwq
->inline_index
; i
++)
142 free_poll_entry(pwq
->inline_entries
+ i
);
144 struct poll_table_entry
* entry
;
145 struct poll_table_page
*old
;
150 free_poll_entry(entry
);
151 } while (entry
> p
->entries
);
154 free_page((unsigned long) old
);
157 EXPORT_SYMBOL(poll_freewait
);
159 static struct poll_table_entry
*poll_get_entry(struct poll_wqueues
*p
)
161 struct poll_table_page
*table
= p
->table
;
163 if (p
->inline_index
< N_INLINE_POLL_ENTRIES
)
164 return p
->inline_entries
+ p
->inline_index
++;
166 if (!table
|| POLL_TABLE_FULL(table
)) {
167 struct poll_table_page
*new_table
;
169 new_table
= (struct poll_table_page
*) __get_free_page(GFP_KERNEL
);
174 new_table
->entry
= new_table
->entries
;
175 new_table
->next
= table
;
176 p
->table
= new_table
;
180 return table
->entry
++;
183 static int __pollwake(wait_queue_t
*wait
, unsigned mode
, int sync
, void *key
)
185 struct poll_wqueues
*pwq
= wait
->private;
186 DECLARE_WAITQUEUE(dummy_wait
, pwq
->polling_task
);
189 * Although this function is called under waitqueue lock, LOCK
190 * doesn't imply write barrier and the users expect write
191 * barrier semantics on wakeup functions. The following
192 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
193 * and is paired with smp_store_mb() in poll_schedule_timeout.
199 * Perform the default wake up operation using a dummy
202 * TODO: This is hacky but there currently is no interface to
203 * pass in @sync. @sync is scheduled to be removed and once
204 * that happens, wake_up_process() can be used directly.
206 return default_wake_function(&dummy_wait
, mode
, sync
, key
);
209 static int pollwake(wait_queue_t
*wait
, unsigned mode
, int sync
, void *key
)
211 struct poll_table_entry
*entry
;
213 entry
= container_of(wait
, struct poll_table_entry
, wait
);
214 if (key
&& !((unsigned long)key
& entry
->key
))
216 return __pollwake(wait
, mode
, sync
, key
);
219 /* Add a new entry */
220 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
223 struct poll_wqueues
*pwq
= container_of(p
, struct poll_wqueues
, pt
);
224 struct poll_table_entry
*entry
= poll_get_entry(pwq
);
227 entry
->filp
= get_file(filp
);
228 entry
->wait_address
= wait_address
;
229 entry
->key
= p
->_key
;
230 init_waitqueue_func_entry(&entry
->wait
, pollwake
);
231 entry
->wait
.private = pwq
;
232 add_wait_queue(wait_address
, &entry
->wait
);
235 int poll_schedule_timeout(struct poll_wqueues
*pwq
, int state
,
236 ktime_t
*expires
, unsigned long slack
)
240 set_current_state(state
);
242 rc
= schedule_hrtimeout_range(expires
, slack
, HRTIMER_MODE_ABS
);
243 __set_current_state(TASK_RUNNING
);
246 * Prepare for the next iteration.
248 * The following smp_store_mb() serves two purposes. First, it's
249 * the counterpart rmb of the wmb in pollwake() such that data
250 * written before wake up is always visible after wake up.
251 * Second, the full barrier guarantees that triggered clearing
252 * doesn't pass event check of the next iteration. Note that
253 * this problem doesn't exist for the first iteration as
254 * add_wait_queue() has full barrier semantics.
256 smp_store_mb(pwq
->triggered
, 0);
260 EXPORT_SYMBOL(poll_schedule_timeout
);
263 * poll_select_set_timeout - helper function to setup the timeout value
264 * @to: pointer to timespec64 variable for the final timeout
265 * @sec: seconds (from user space)
266 * @nsec: nanoseconds (from user space)
268 * Note, we do not use a timespec for the user space value here, That
269 * way we can use the function for timeval and compat interfaces as well.
271 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
273 int poll_select_set_timeout(struct timespec64
*to
, time64_t sec
, long nsec
)
275 struct timespec64 ts
= {.tv_sec
= sec
, .tv_nsec
= nsec
};
277 if (!timespec64_valid(&ts
))
280 /* Optimize for the zero timeout value here */
282 to
->tv_sec
= to
->tv_nsec
= 0;
285 *to
= timespec64_add_safe(*to
, ts
);
290 static int poll_select_copy_remaining(struct timespec64
*end_time
,
292 int timeval
, int ret
)
294 struct timespec64 rts64
;
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(&rts64
);
309 rts64
= timespec64_sub(*end_time
, rts64
);
310 if (rts64
.tv_sec
< 0)
311 rts64
.tv_sec
= rts64
.tv_nsec
= 0;
313 rts
= timespec64_to_timespec(rts64
);
316 if (sizeof(rtv
) > sizeof(rtv
.tv_sec
) + sizeof(rtv
.tv_usec
))
317 memset(&rtv
, 0, sizeof(rtv
));
318 rtv
.tv_sec
= rts64
.tv_sec
;
319 rtv
.tv_usec
= rts64
.tv_nsec
/ NSEC_PER_USEC
;
321 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
324 } else if (!copy_to_user(p
, &rts
, sizeof(rts
)))
328 * If an application puts its timeval in read-only memory, we
329 * don't want the Linux-specific update to the timeval to
330 * cause a fault after the select has completed
331 * successfully. However, because we're not updating the
332 * timeval, we can't restart the system call.
336 if (ret
== -ERESTARTNOHAND
)
342 * Scalable version of the fd_set.
346 unsigned long *in
, *out
, *ex
;
347 unsigned long *res_in
, *res_out
, *res_ex
;
351 * How many longwords for "nr" bits?
353 #define FDS_BITPERLONG (8*sizeof(long))
354 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
355 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
358 * We do a VERIFY_WRITE here even though we are only reading this time:
359 * we'll write to it eventually..
361 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
364 int get_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
368 return copy_from_user(fdset
, ufdset
, nr
) ? -EFAULT
: 0;
370 memset(fdset
, 0, nr
);
374 static inline unsigned long __must_check
375 set_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
378 return __copy_to_user(ufdset
, fdset
, FDS_BYTES(nr
));
383 void zero_fd_set(unsigned long nr
, unsigned long *fdset
)
385 memset(fdset
, 0, FDS_BYTES(nr
));
388 #define FDS_IN(fds, n) (fds->in + n)
389 #define FDS_OUT(fds, n) (fds->out + n)
390 #define FDS_EX(fds, n) (fds->ex + n)
392 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
394 static int max_select_fd(unsigned long n
, fd_set_bits
*fds
)
396 unsigned long *open_fds
;
401 /* handle last in-complete long-word first */
402 set
= ~(~0UL << (n
& (BITS_PER_LONG
-1)));
404 fdt
= files_fdtable(current
->files
);
405 open_fds
= fdt
->open_fds
+ n
;
410 if (!(set
& ~*open_fds
))
421 if (set
& ~*open_fds
)
430 max
+= n
* BITS_PER_LONG
;
436 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
437 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
438 #define POLLEX_SET (POLLPRI)
440 static inline void wait_key_set(poll_table
*wait
, unsigned long in
,
441 unsigned long out
, unsigned long bit
,
442 unsigned int ll_flag
)
444 wait
->_key
= POLLEX_SET
| ll_flag
;
446 wait
->_key
|= POLLIN_SET
;
448 wait
->_key
|= POLLOUT_SET
;
451 static int do_select(int n
, fd_set_bits
*fds
, struct timespec64
*end_time
)
453 ktime_t expire
, *to
= NULL
;
454 struct poll_wqueues table
;
456 int retval
, i
, timed_out
= 0;
458 unsigned int busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
459 unsigned long busy_start
= 0;
462 retval
= max_select_fd(n
, fds
);
469 poll_initwait(&table
);
471 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
476 if (end_time
&& !timed_out
)
477 slack
= select_estimate_accuracy(end_time
);
481 unsigned long *rinp
, *routp
, *rexp
, *inp
, *outp
, *exp
;
482 bool can_busy_loop
= false;
484 inp
= fds
->in
; outp
= fds
->out
; exp
= fds
->ex
;
485 rinp
= fds
->res_in
; routp
= fds
->res_out
; rexp
= fds
->res_ex
;
487 for (i
= 0; i
< n
; ++rinp
, ++routp
, ++rexp
) {
488 unsigned long in
, out
, ex
, all_bits
, bit
= 1, mask
, j
;
489 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
= kmalloc(alloc_size
, GFP_KERNEL
|__GFP_NOWARN
);
637 if (!bits
&& alloc_size
> PAGE_SIZE
)
638 bits
= vmalloc(alloc_size
);
644 fds
.out
= bits
+ size
;
645 fds
.ex
= bits
+ 2*size
;
646 fds
.res_in
= bits
+ 3*size
;
647 fds
.res_out
= bits
+ 4*size
;
648 fds
.res_ex
= bits
+ 5*size
;
650 if ((ret
= get_fd_set(n
, inp
, fds
.in
)) ||
651 (ret
= get_fd_set(n
, outp
, fds
.out
)) ||
652 (ret
= get_fd_set(n
, exp
, fds
.ex
)))
654 zero_fd_set(n
, fds
.res_in
);
655 zero_fd_set(n
, fds
.res_out
);
656 zero_fd_set(n
, fds
.res_ex
);
658 ret
= do_select(n
, &fds
, end_time
);
663 ret
= -ERESTARTNOHAND
;
664 if (signal_pending(current
))
669 if (set_fd_set(n
, inp
, fds
.res_in
) ||
670 set_fd_set(n
, outp
, fds
.res_out
) ||
671 set_fd_set(n
, exp
, fds
.res_ex
))
675 if (bits
!= stack_fds
)
681 SYSCALL_DEFINE5(select
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
682 fd_set __user
*, exp
, struct timeval __user
*, tvp
)
684 struct timespec64 end_time
, *to
= NULL
;
689 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
693 if (poll_select_set_timeout(to
,
694 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
695 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
699 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
700 ret
= poll_select_copy_remaining(&end_time
, tvp
, 1, ret
);
705 static long do_pselect(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
706 fd_set __user
*exp
, struct timespec __user
*tsp
,
707 const sigset_t __user
*sigmask
, size_t sigsetsize
)
709 sigset_t ksigmask
, sigsaved
;
711 struct timespec64 ts64
, end_time
, *to
= NULL
;
715 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
717 ts64
= timespec_to_timespec64(ts
);
720 if (poll_select_set_timeout(to
, ts64
.tv_sec
, ts64
.tv_nsec
))
725 /* XXX: Don't preclude handling different sized sigset_t's. */
726 if (sigsetsize
!= sizeof(sigset_t
))
728 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
731 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
732 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
735 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
736 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
738 if (ret
== -ERESTARTNOHAND
) {
740 * Don't restore the signal mask yet. Let do_signal() deliver
741 * the signal on the way back to userspace, before the signal
745 memcpy(¤t
->saved_sigmask
, &sigsaved
,
747 set_restore_sigmask();
750 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
756 * Most architectures can't handle 7-argument syscalls. So we provide a
757 * 6-argument version where the sixth argument is a pointer to a structure
758 * which has a pointer to the sigset_t itself followed by a size_t containing
761 SYSCALL_DEFINE6(pselect6
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
762 fd_set __user
*, exp
, struct timespec __user
*, tsp
,
765 size_t sigsetsize
= 0;
766 sigset_t __user
*up
= NULL
;
769 if (!access_ok(VERIFY_READ
, sig
, sizeof(void *)+sizeof(size_t))
770 || __get_user(up
, (sigset_t __user
* __user
*)sig
)
771 || __get_user(sigsetsize
,
772 (size_t __user
*)(sig
+sizeof(void *))))
776 return do_pselect(n
, inp
, outp
, exp
, tsp
, up
, sigsetsize
);
779 #ifdef __ARCH_WANT_SYS_OLD_SELECT
780 struct sel_arg_struct
{
782 fd_set __user
*inp
, *outp
, *exp
;
783 struct timeval __user
*tvp
;
786 SYSCALL_DEFINE1(old_select
, struct sel_arg_struct __user
*, arg
)
788 struct sel_arg_struct a
;
790 if (copy_from_user(&a
, arg
, sizeof(a
)))
792 return sys_select(a
.n
, a
.inp
, a
.outp
, a
.exp
, a
.tvp
);
797 struct poll_list
*next
;
799 struct pollfd entries
[0];
802 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
805 * Fish for pollable events on the pollfd->fd file descriptor. We're only
806 * interested in events matching the pollfd->events mask, and the result
807 * matching that mask is both recorded in pollfd->revents and returned. The
808 * pwait poll_table will be used by the fd-provided poll handler for waiting,
809 * if pwait->_qproc is non-NULL.
811 static inline unsigned int do_pollfd(struct pollfd
*pollfd
, poll_table
*pwait
,
813 unsigned int busy_flag
)
821 struct fd f
= fdget(fd
);
824 mask
= DEFAULT_POLLMASK
;
825 if (f
.file
->f_op
->poll
) {
826 pwait
->_key
= pollfd
->events
|POLLERR
|POLLHUP
;
827 pwait
->_key
|= busy_flag
;
828 mask
= f
.file
->f_op
->poll(f
.file
, pwait
);
829 if (mask
& busy_flag
)
830 *can_busy_poll
= true;
832 /* Mask out unneeded events. */
833 mask
&= pollfd
->events
| POLLERR
| POLLHUP
;
837 pollfd
->revents
= mask
;
842 static int do_poll(struct poll_list
*list
, struct poll_wqueues
*wait
,
843 struct timespec64
*end_time
)
845 poll_table
* pt
= &wait
->pt
;
846 ktime_t expire
, *to
= NULL
;
847 int timed_out
= 0, count
= 0;
849 unsigned int busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
850 unsigned long busy_start
= 0;
852 /* Optimise the no-wait case */
853 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
858 if (end_time
&& !timed_out
)
859 slack
= select_estimate_accuracy(end_time
);
862 struct poll_list
*walk
;
863 bool can_busy_loop
= false;
865 for (walk
= list
; walk
!= NULL
; walk
= walk
->next
) {
866 struct pollfd
* pfd
, * pfd_end
;
869 pfd_end
= pfd
+ walk
->len
;
870 for (; pfd
!= pfd_end
; pfd
++) {
872 * Fish for events. If we found one, record it
873 * and kill poll_table->_qproc, so we don't
874 * needlessly register any other waiters after
875 * this. They'll get immediately deregistered
876 * when we break out and return.
878 if (do_pollfd(pfd
, pt
, &can_busy_loop
,
882 /* found something, stop busy polling */
884 can_busy_loop
= false;
889 * All waiters have already been registered, so don't provide
890 * a poll_table->_qproc to them on the next loop iteration.
895 if (signal_pending(current
))
898 if (count
|| timed_out
)
901 /* only if found POLL_BUSY_LOOP sockets && not out of time */
902 if (can_busy_loop
&& !need_resched()) {
904 busy_start
= busy_loop_current_time();
907 if (!busy_loop_timeout(busy_start
))
913 * If this is the first loop and we have a timeout
914 * given, then we convert to ktime_t and set the to
915 * pointer to the expiry value.
917 if (end_time
&& !to
) {
918 expire
= timespec64_to_ktime(*end_time
);
922 if (!poll_schedule_timeout(wait
, TASK_INTERRUPTIBLE
, to
, slack
))
928 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
929 sizeof(struct pollfd))
931 static int do_sys_poll(struct pollfd __user
*ufds
, unsigned int nfds
,
932 struct timespec64
*end_time
)
934 struct poll_wqueues table
;
935 int err
= -EFAULT
, fdcount
, len
, size
;
936 /* Allocate small arguments on the stack to save memory and be
937 faster - use long to make sure the buffer is aligned properly
938 on 64 bit archs to avoid unaligned access */
939 long stack_pps
[POLL_STACK_ALLOC
/sizeof(long)];
940 struct poll_list
*const head
= (struct poll_list
*)stack_pps
;
941 struct poll_list
*walk
= head
;
942 unsigned long todo
= nfds
;
944 if (nfds
> rlimit(RLIMIT_NOFILE
))
947 len
= min_t(unsigned int, nfds
, N_STACK_PPS
);
954 if (copy_from_user(walk
->entries
, ufds
+ nfds
-todo
,
955 sizeof(struct pollfd
) * walk
->len
))
962 len
= min(todo
, POLLFD_PER_PAGE
);
963 size
= sizeof(struct poll_list
) + sizeof(struct pollfd
) * len
;
964 walk
= walk
->next
= kmalloc(size
, GFP_KERNEL
);
971 poll_initwait(&table
);
972 fdcount
= do_poll(head
, &table
, end_time
);
973 poll_freewait(&table
);
975 for (walk
= head
; walk
; walk
= walk
->next
) {
976 struct pollfd
*fds
= walk
->entries
;
979 for (j
= 0; j
< walk
->len
; j
++, ufds
++)
980 if (__put_user(fds
[j
].revents
, &ufds
->revents
))
988 struct poll_list
*pos
= walk
;
996 static long do_restart_poll(struct restart_block
*restart_block
)
998 struct pollfd __user
*ufds
= restart_block
->poll
.ufds
;
999 int nfds
= restart_block
->poll
.nfds
;
1000 struct timespec64
*to
= NULL
, end_time
;
1003 if (restart_block
->poll
.has_timeout
) {
1004 end_time
.tv_sec
= restart_block
->poll
.tv_sec
;
1005 end_time
.tv_nsec
= restart_block
->poll
.tv_nsec
;
1009 ret
= do_sys_poll(ufds
, nfds
, to
);
1011 if (ret
== -EINTR
) {
1012 restart_block
->fn
= do_restart_poll
;
1013 ret
= -ERESTART_RESTARTBLOCK
;
1018 SYSCALL_DEFINE3(poll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1021 struct timespec64 end_time
, *to
= NULL
;
1024 if (timeout_msecs
>= 0) {
1026 poll_select_set_timeout(to
, timeout_msecs
/ MSEC_PER_SEC
,
1027 NSEC_PER_MSEC
* (timeout_msecs
% MSEC_PER_SEC
));
1030 ret
= do_sys_poll(ufds
, nfds
, to
);
1032 if (ret
== -EINTR
) {
1033 struct restart_block
*restart_block
;
1035 restart_block
= ¤t
->restart_block
;
1036 restart_block
->fn
= do_restart_poll
;
1037 restart_block
->poll
.ufds
= ufds
;
1038 restart_block
->poll
.nfds
= nfds
;
1040 if (timeout_msecs
>= 0) {
1041 restart_block
->poll
.tv_sec
= end_time
.tv_sec
;
1042 restart_block
->poll
.tv_nsec
= end_time
.tv_nsec
;
1043 restart_block
->poll
.has_timeout
= 1;
1045 restart_block
->poll
.has_timeout
= 0;
1047 ret
= -ERESTART_RESTARTBLOCK
;
1052 SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1053 struct timespec __user
*, tsp
, const sigset_t __user
*, sigmask
,
1056 sigset_t ksigmask
, sigsaved
;
1058 struct timespec64 end_time
, *to
= NULL
;
1062 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
1066 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1071 /* XXX: Don't preclude handling different sized sigset_t's. */
1072 if (sigsetsize
!= sizeof(sigset_t
))
1074 if (copy_from_user(&ksigmask
, sigmask
, sizeof(ksigmask
)))
1077 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1078 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1081 ret
= do_sys_poll(ufds
, nfds
, to
);
1083 /* We can restart this syscall, usually */
1084 if (ret
== -EINTR
) {
1086 * Don't restore the signal mask yet. Let do_signal() deliver
1087 * the signal on the way back to userspace, before the signal
1091 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1093 set_restore_sigmask();
1095 ret
= -ERESTARTNOHAND
;
1097 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1099 ret
= poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
1104 #ifdef CONFIG_COMPAT
1105 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1108 int compat_poll_select_copy_remaining(struct timespec
*end_time
, void __user
*p
,
1109 int timeval
, int ret
)
1116 if (current
->personality
& STICKY_TIMEOUTS
)
1119 /* No update for zero timeout */
1120 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
1124 ts
= timespec_sub(*end_time
, ts
);
1126 ts
.tv_sec
= ts
.tv_nsec
= 0;
1129 struct compat_timeval rtv
;
1131 rtv
.tv_sec
= ts
.tv_sec
;
1132 rtv
.tv_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
1134 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
1137 struct compat_timespec rts
;
1139 rts
.tv_sec
= ts
.tv_sec
;
1140 rts
.tv_nsec
= ts
.tv_nsec
;
1142 if (!copy_to_user(p
, &rts
, sizeof(rts
)))
1146 * If an application puts its timeval in read-only memory, we
1147 * don't want the Linux-specific update to the timeval to
1148 * cause a fault after the select has completed
1149 * successfully. However, because we're not updating the
1150 * timeval, we can't restart the system call.
1154 if (ret
== -ERESTARTNOHAND
)
1160 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1161 * 64-bit unsigned longs.
1164 int compat_get_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1165 unsigned long *fdset
)
1167 nr
= DIV_ROUND_UP(nr
, __COMPAT_NFDBITS
);
1171 if (!access_ok(VERIFY_WRITE
, ufdset
, nr
*sizeof(compat_ulong_t
)))
1178 if (__get_user(l
, ufdset
) || __get_user(h
, ufdset
+1))
1181 *fdset
++ = h
<< 32 | l
;
1184 if (odd
&& __get_user(*fdset
, ufdset
))
1187 /* Tricky, must clear full unsigned long in the
1188 * kernel fdset at the end, this makes sure that
1191 memset(fdset
, 0, ((nr
+ 1) & ~1)*sizeof(compat_ulong_t
));
1197 int compat_set_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1198 unsigned long *fdset
)
1201 nr
= DIV_ROUND_UP(nr
, __COMPAT_NFDBITS
);
1212 if (__put_user(l
, ufdset
) || __put_user(h
, ufdset
+1))
1217 if (odd
&& __put_user(*fdset
, ufdset
))
1224 * This is a virtual copy of sys_select from fs/select.c and probably
1225 * should be compared to it from time to time
1229 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1230 * like to be certain this leads to no problems. So I return
1231 * EINTR just for safety.
1233 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1234 * I'm trying ERESTARTNOHAND which restart only when you want to.
1236 static int compat_core_sys_select(int n
, compat_ulong_t __user
*inp
,
1237 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1238 struct timespec
*end_time
)
1242 int size
, max_fds
, ret
= -EINVAL
;
1243 struct fdtable
*fdt
;
1244 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
1249 /* max_fds can increase, so grab it once to avoid race */
1251 fdt
= files_fdtable(current
->files
);
1252 max_fds
= fdt
->max_fds
;
1258 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1259 * since we used fdset we need to allocate memory in units of
1262 size
= FDS_BYTES(n
);
1264 if (size
> sizeof(stack_fds
) / 6) {
1265 bits
= kmalloc(6 * size
, GFP_KERNEL
);
1270 fds
.in
= (unsigned long *) bits
;
1271 fds
.out
= (unsigned long *) (bits
+ size
);
1272 fds
.ex
= (unsigned long *) (bits
+ 2*size
);
1273 fds
.res_in
= (unsigned long *) (bits
+ 3*size
);
1274 fds
.res_out
= (unsigned long *) (bits
+ 4*size
);
1275 fds
.res_ex
= (unsigned long *) (bits
+ 5*size
);
1277 if ((ret
= compat_get_fd_set(n
, inp
, fds
.in
)) ||
1278 (ret
= compat_get_fd_set(n
, outp
, fds
.out
)) ||
1279 (ret
= compat_get_fd_set(n
, exp
, fds
.ex
)))
1281 zero_fd_set(n
, fds
.res_in
);
1282 zero_fd_set(n
, fds
.res_out
);
1283 zero_fd_set(n
, fds
.res_ex
);
1285 ret
= do_select(n
, &fds
, end_time
);
1290 ret
= -ERESTARTNOHAND
;
1291 if (signal_pending(current
))
1296 if (compat_set_fd_set(n
, inp
, fds
.res_in
) ||
1297 compat_set_fd_set(n
, outp
, fds
.res_out
) ||
1298 compat_set_fd_set(n
, exp
, fds
.res_ex
))
1301 if (bits
!= stack_fds
)
1307 COMPAT_SYSCALL_DEFINE5(select
, int, n
, compat_ulong_t __user
*, inp
,
1308 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1309 struct compat_timeval __user
*, tvp
)
1311 struct timespec end_time
, *to
= NULL
;
1312 struct compat_timeval tv
;
1316 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
1320 if (poll_select_set_timeout(to
,
1321 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
1322 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
1326 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1327 ret
= compat_poll_select_copy_remaining(&end_time
, tvp
, 1, ret
);
1332 struct compat_sel_arg_struct
{
1340 COMPAT_SYSCALL_DEFINE1(old_select
, struct compat_sel_arg_struct __user
*, arg
)
1342 struct compat_sel_arg_struct a
;
1344 if (copy_from_user(&a
, arg
, sizeof(a
)))
1346 return compat_sys_select(a
.n
, compat_ptr(a
.inp
), compat_ptr(a
.outp
),
1347 compat_ptr(a
.exp
), compat_ptr(a
.tvp
));
1350 static long do_compat_pselect(int n
, compat_ulong_t __user
*inp
,
1351 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1352 struct compat_timespec __user
*tsp
, compat_sigset_t __user
*sigmask
,
1353 compat_size_t sigsetsize
)
1355 compat_sigset_t ss32
;
1356 sigset_t ksigmask
, sigsaved
;
1357 struct compat_timespec ts
;
1358 struct timespec end_time
, *to
= NULL
;
1362 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
1366 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1371 if (sigsetsize
!= sizeof(compat_sigset_t
))
1373 if (copy_from_user(&ss32
, sigmask
, sizeof(ss32
)))
1375 sigset_from_compat(&ksigmask
, &ss32
);
1377 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1378 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1381 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1382 ret
= compat_poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);
1384 if (ret
== -ERESTARTNOHAND
) {
1386 * Don't restore the signal mask yet. Let do_signal() deliver
1387 * the signal on the way back to userspace, before the signal
1391 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1393 set_restore_sigmask();
1396 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1401 COMPAT_SYSCALL_DEFINE6(pselect6
, int, n
, compat_ulong_t __user
*, inp
,
1402 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1403 struct compat_timespec __user
*, tsp
, void __user
*, sig
)
1405 compat_size_t sigsetsize
= 0;
1406 compat_uptr_t up
= 0;
1409 if (!access_ok(VERIFY_READ
, sig
,
1410 sizeof(compat_uptr_t
)+sizeof(compat_size_t
)) ||
1411 __get_user(up
, (compat_uptr_t __user
*)sig
) ||
1412 __get_user(sigsetsize
,
1413 (compat_size_t __user
*)(sig
+sizeof(up
))))
1416 return do_compat_pselect(n
, inp
, outp
, exp
, tsp
, compat_ptr(up
),
1420 COMPAT_SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
,
1421 unsigned int, nfds
, struct compat_timespec __user
*, tsp
,
1422 const compat_sigset_t __user
*, sigmask
, compat_size_t
, sigsetsize
)
1424 compat_sigset_t ss32
;
1425 sigset_t ksigmask
, sigsaved
;
1426 struct compat_timespec ts
;
1427 struct timespec end_time
, *to
= NULL
;
1431 if (copy_from_user(&ts
, tsp
, sizeof(ts
)))
1435 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1440 if (sigsetsize
!= sizeof(compat_sigset_t
))
1442 if (copy_from_user(&ss32
, sigmask
, sizeof(ss32
)))
1444 sigset_from_compat(&ksigmask
, &ss32
);
1446 sigdelsetmask(&ksigmask
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1447 sigprocmask(SIG_SETMASK
, &ksigmask
, &sigsaved
);
1450 ret
= do_sys_poll(ufds
, nfds
, to
);
1452 /* We can restart this syscall, usually */
1453 if (ret
== -EINTR
) {
1455 * Don't restore the signal mask yet. Let do_signal() deliver
1456 * the signal on the way back to userspace, before the signal
1460 memcpy(¤t
->saved_sigmask
, &sigsaved
,
1462 set_restore_sigmask();
1464 ret
= -ERESTARTNOHAND
;
1466 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1468 ret
= compat_poll_select_copy_remaining(&end_time
, tsp
, 0, ret
);