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 static 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);
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 enum poll_time_type
{
297 static int poll_select_finish(struct timespec64
*end_time
,
299 enum poll_time_type pt_type
, int ret
)
301 struct timespec64 rts
;
303 restore_saved_sigmask_unless(ret
== -ERESTARTNOHAND
);
308 if (current
->personality
& STICKY_TIMEOUTS
)
311 /* No update for zero timeout */
312 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
315 ktime_get_ts64(&rts
);
316 rts
= timespec64_sub(*end_time
, rts
);
318 rts
.tv_sec
= rts
.tv_nsec
= 0;
324 struct __kernel_old_timeval rtv
;
326 if (sizeof(rtv
) > sizeof(rtv
.tv_sec
) + sizeof(rtv
.tv_usec
))
327 memset(&rtv
, 0, sizeof(rtv
));
328 rtv
.tv_sec
= rts
.tv_sec
;
329 rtv
.tv_usec
= rts
.tv_nsec
/ NSEC_PER_USEC
;
330 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
336 struct old_timeval32 rtv
;
338 rtv
.tv_sec
= rts
.tv_sec
;
339 rtv
.tv_usec
= rts
.tv_nsec
/ NSEC_PER_USEC
;
340 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
345 if (!put_timespec64(&rts
, p
))
348 case PT_OLD_TIMESPEC
:
349 if (!put_old_timespec32(&rts
, p
))
356 * If an application puts its timeval in read-only memory, we
357 * don't want the Linux-specific update to the timeval to
358 * cause a fault after the select has completed
359 * successfully. However, because we're not updating the
360 * timeval, we can't restart the system call.
364 if (ret
== -ERESTARTNOHAND
)
370 * Scalable version of the fd_set.
374 unsigned long *in
, *out
, *ex
;
375 unsigned long *res_in
, *res_out
, *res_ex
;
379 * How many longwords for "nr" bits?
381 #define FDS_BITPERLONG (8*sizeof(long))
382 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
383 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
386 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
389 int get_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
393 return copy_from_user(fdset
, ufdset
, nr
) ? -EFAULT
: 0;
395 memset(fdset
, 0, nr
);
399 static inline unsigned long __must_check
400 set_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
403 return __copy_to_user(ufdset
, fdset
, FDS_BYTES(nr
));
408 void zero_fd_set(unsigned long nr
, unsigned long *fdset
)
410 memset(fdset
, 0, FDS_BYTES(nr
));
413 #define FDS_IN(fds, n) (fds->in + n)
414 #define FDS_OUT(fds, n) (fds->out + n)
415 #define FDS_EX(fds, n) (fds->ex + n)
417 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
419 static int max_select_fd(unsigned long n
, fd_set_bits
*fds
)
421 unsigned long *open_fds
;
426 /* handle last in-complete long-word first */
427 set
= ~(~0UL << (n
& (BITS_PER_LONG
-1)));
429 fdt
= files_fdtable(current
->files
);
430 open_fds
= fdt
->open_fds
+ n
;
435 if (!(set
& ~*open_fds
))
446 if (set
& ~*open_fds
)
455 max
+= n
* BITS_PER_LONG
;
461 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR)
462 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
463 #define POLLEX_SET (EPOLLPRI)
465 static inline void wait_key_set(poll_table
*wait
, unsigned long in
,
466 unsigned long out
, unsigned long bit
,
469 wait
->_key
= POLLEX_SET
| ll_flag
;
471 wait
->_key
|= POLLIN_SET
;
473 wait
->_key
|= POLLOUT_SET
;
476 static int do_select(int n
, fd_set_bits
*fds
, struct timespec64
*end_time
)
478 ktime_t expire
, *to
= NULL
;
479 struct poll_wqueues table
;
481 int retval
, i
, timed_out
= 0;
483 __poll_t busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
484 unsigned long busy_start
= 0;
487 retval
= max_select_fd(n
, fds
);
494 poll_initwait(&table
);
496 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
501 if (end_time
&& !timed_out
)
502 slack
= select_estimate_accuracy(end_time
);
506 unsigned long *rinp
, *routp
, *rexp
, *inp
, *outp
, *exp
;
507 bool can_busy_loop
= false;
509 inp
= fds
->in
; outp
= fds
->out
; exp
= fds
->ex
;
510 rinp
= fds
->res_in
; routp
= fds
->res_out
; rexp
= fds
->res_ex
;
512 for (i
= 0; i
< n
; ++rinp
, ++routp
, ++rexp
) {
513 unsigned long in
, out
, ex
, all_bits
, bit
= 1, j
;
514 unsigned long res_in
= 0, res_out
= 0, res_ex
= 0;
517 in
= *inp
++; out
= *outp
++; ex
= *exp
++;
518 all_bits
= in
| out
| ex
;
524 for (j
= 0; j
< BITS_PER_LONG
; ++j
, ++i
, bit
<<= 1) {
528 if (!(bit
& all_bits
))
532 wait_key_set(wait
, in
, out
, bit
,
534 mask
= vfs_poll(f
.file
, wait
);
537 if ((mask
& POLLIN_SET
) && (in
& bit
)) {
542 if ((mask
& POLLOUT_SET
) && (out
& bit
)) {
547 if ((mask
& POLLEX_SET
) && (ex
& bit
)) {
552 /* got something, stop busy polling */
554 can_busy_loop
= false;
558 * only remember a returned
559 * POLL_BUSY_LOOP if we asked for it
561 } else if (busy_flag
& mask
)
562 can_busy_loop
= true;
575 if (retval
|| timed_out
|| signal_pending(current
))
578 retval
= table
.error
;
582 /* only if found POLL_BUSY_LOOP sockets && not out of time */
583 if (can_busy_loop
&& !need_resched()) {
585 busy_start
= busy_loop_current_time();
588 if (!busy_loop_timeout(busy_start
))
594 * If this is the first loop and we have a timeout
595 * given, then we convert to ktime_t and set the to
596 * pointer to the expiry value.
598 if (end_time
&& !to
) {
599 expire
= timespec64_to_ktime(*end_time
);
603 if (!poll_schedule_timeout(&table
, TASK_INTERRUPTIBLE
,
608 poll_freewait(&table
);
614 * We can actually return ERESTARTSYS instead of EINTR, but I'd
615 * like to be certain this leads to no problems. So I return
616 * EINTR just for safety.
618 * Update: ERESTARTSYS breaks at least the xview clock binary, so
619 * I'm trying ERESTARTNOHAND which restart only when you want to.
621 int core_sys_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
622 fd_set __user
*exp
, struct timespec64
*end_time
)
627 size_t size
, alloc_size
;
629 /* Allocate small arguments on the stack to save memory and be faster */
630 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
636 /* max_fds can increase, so grab it once to avoid race */
638 fdt
= files_fdtable(current
->files
);
639 max_fds
= fdt
->max_fds
;
645 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
646 * since we used fdset we need to allocate memory in units of
651 if (size
> sizeof(stack_fds
) / 6) {
652 /* Not enough space in on-stack array; must use kmalloc */
654 if (size
> (SIZE_MAX
/ 6))
657 alloc_size
= 6 * size
;
658 bits
= kvmalloc(alloc_size
, GFP_KERNEL
);
663 fds
.out
= bits
+ size
;
664 fds
.ex
= bits
+ 2*size
;
665 fds
.res_in
= bits
+ 3*size
;
666 fds
.res_out
= bits
+ 4*size
;
667 fds
.res_ex
= bits
+ 5*size
;
669 if ((ret
= get_fd_set(n
, inp
, fds
.in
)) ||
670 (ret
= get_fd_set(n
, outp
, fds
.out
)) ||
671 (ret
= get_fd_set(n
, exp
, fds
.ex
)))
673 zero_fd_set(n
, fds
.res_in
);
674 zero_fd_set(n
, fds
.res_out
);
675 zero_fd_set(n
, fds
.res_ex
);
677 ret
= do_select(n
, &fds
, end_time
);
682 ret
= -ERESTARTNOHAND
;
683 if (signal_pending(current
))
688 if (set_fd_set(n
, inp
, fds
.res_in
) ||
689 set_fd_set(n
, outp
, fds
.res_out
) ||
690 set_fd_set(n
, exp
, fds
.res_ex
))
694 if (bits
!= stack_fds
)
700 static int kern_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
701 fd_set __user
*exp
, struct __kernel_old_timeval __user
*tvp
)
703 struct timespec64 end_time
, *to
= NULL
;
704 struct __kernel_old_timeval tv
;
708 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
712 if (poll_select_set_timeout(to
,
713 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
714 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
718 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
719 return poll_select_finish(&end_time
, tvp
, PT_TIMEVAL
, ret
);
722 SYSCALL_DEFINE5(select
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
723 fd_set __user
*, exp
, struct __kernel_old_timeval __user
*, tvp
)
725 return kern_select(n
, inp
, outp
, exp
, tvp
);
728 static long do_pselect(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
729 fd_set __user
*exp
, void __user
*tsp
,
730 const sigset_t __user
*sigmask
, size_t sigsetsize
,
731 enum poll_time_type type
)
733 struct timespec64 ts
, end_time
, *to
= NULL
;
739 if (get_timespec64(&ts
, tsp
))
742 case PT_OLD_TIMESPEC
:
743 if (get_old_timespec32(&ts
, tsp
))
751 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
755 ret
= set_user_sigmask(sigmask
, sigsetsize
);
759 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
760 return poll_select_finish(&end_time
, tsp
, type
, ret
);
764 * Most architectures can't handle 7-argument syscalls. So we provide a
765 * 6-argument version where the sixth argument is a pointer to a structure
766 * which has a pointer to the sigset_t itself followed by a size_t containing
769 struct sigset_argpack
{
774 static inline int get_sigset_argpack(struct sigset_argpack
*to
,
775 struct sigset_argpack __user
*from
)
777 // the path is hot enough for overhead of copy_from_user() to matter
779 if (!user_read_access_begin(from
, sizeof(*from
)))
781 unsafe_get_user(to
->p
, &from
->p
, Efault
);
782 unsafe_get_user(to
->size
, &from
->size
, Efault
);
783 user_read_access_end();
791 SYSCALL_DEFINE6(pselect6
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
792 fd_set __user
*, exp
, struct __kernel_timespec __user
*, tsp
,
795 struct sigset_argpack x
= {NULL
, 0};
797 if (get_sigset_argpack(&x
, sig
))
800 return do_pselect(n
, inp
, outp
, exp
, tsp
, x
.p
, x
.size
, PT_TIMESPEC
);
803 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
805 SYSCALL_DEFINE6(pselect6_time32
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
806 fd_set __user
*, exp
, struct old_timespec32 __user
*, tsp
,
809 struct sigset_argpack x
= {NULL
, 0};
811 if (get_sigset_argpack(&x
, sig
))
814 return do_pselect(n
, inp
, outp
, exp
, tsp
, x
.p
, x
.size
, PT_OLD_TIMESPEC
);
819 #ifdef __ARCH_WANT_SYS_OLD_SELECT
820 struct sel_arg_struct
{
822 fd_set __user
*inp
, *outp
, *exp
;
823 struct __kernel_old_timeval __user
*tvp
;
826 SYSCALL_DEFINE1(old_select
, struct sel_arg_struct __user
*, arg
)
828 struct sel_arg_struct a
;
830 if (copy_from_user(&a
, arg
, sizeof(a
)))
832 return kern_select(a
.n
, a
.inp
, a
.outp
, a
.exp
, a
.tvp
);
837 struct poll_list
*next
;
839 struct pollfd entries
[0];
842 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
845 * Fish for pollable events on the pollfd->fd file descriptor. We're only
846 * interested in events matching the pollfd->events mask, and the result
847 * matching that mask is both recorded in pollfd->revents and returned. The
848 * pwait poll_table will be used by the fd-provided poll handler for waiting,
849 * if pwait->_qproc is non-NULL.
851 static inline __poll_t
do_pollfd(struct pollfd
*pollfd
, poll_table
*pwait
,
856 __poll_t mask
= 0, filter
;
866 /* userland u16 ->events contains POLL... bitmap */
867 filter
= demangle_poll(pollfd
->events
) | EPOLLERR
| EPOLLHUP
;
868 pwait
->_key
= filter
| busy_flag
;
869 mask
= vfs_poll(f
.file
, pwait
);
870 if (mask
& busy_flag
)
871 *can_busy_poll
= true;
872 mask
&= filter
; /* Mask out unneeded events. */
876 /* ... and so does ->revents */
877 pollfd
->revents
= mangle_poll(mask
);
881 static int do_poll(struct poll_list
*list
, struct poll_wqueues
*wait
,
882 struct timespec64
*end_time
)
884 poll_table
* pt
= &wait
->pt
;
885 ktime_t expire
, *to
= NULL
;
886 int timed_out
= 0, count
= 0;
888 __poll_t busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
889 unsigned long busy_start
= 0;
891 /* Optimise the no-wait case */
892 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
897 if (end_time
&& !timed_out
)
898 slack
= select_estimate_accuracy(end_time
);
901 struct poll_list
*walk
;
902 bool can_busy_loop
= false;
904 for (walk
= list
; walk
!= NULL
; walk
= walk
->next
) {
905 struct pollfd
* pfd
, * pfd_end
;
908 pfd_end
= pfd
+ walk
->len
;
909 for (; pfd
!= pfd_end
; pfd
++) {
911 * Fish for events. If we found one, record it
912 * and kill poll_table->_qproc, so we don't
913 * needlessly register any other waiters after
914 * this. They'll get immediately deregistered
915 * when we break out and return.
917 if (do_pollfd(pfd
, pt
, &can_busy_loop
,
921 /* found something, stop busy polling */
923 can_busy_loop
= false;
928 * All waiters have already been registered, so don't provide
929 * a poll_table->_qproc to them on the next loop iteration.
934 if (signal_pending(current
))
935 count
= -ERESTARTNOHAND
;
937 if (count
|| timed_out
)
940 /* only if found POLL_BUSY_LOOP sockets && not out of time */
941 if (can_busy_loop
&& !need_resched()) {
943 busy_start
= busy_loop_current_time();
946 if (!busy_loop_timeout(busy_start
))
952 * If this is the first loop and we have a timeout
953 * given, then we convert to ktime_t and set the to
954 * pointer to the expiry value.
956 if (end_time
&& !to
) {
957 expire
= timespec64_to_ktime(*end_time
);
961 if (!poll_schedule_timeout(wait
, TASK_INTERRUPTIBLE
, to
, slack
))
967 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
968 sizeof(struct pollfd))
970 static int do_sys_poll(struct pollfd __user
*ufds
, unsigned int nfds
,
971 struct timespec64
*end_time
)
973 struct poll_wqueues table
;
974 int err
= -EFAULT
, fdcount
, len
;
975 /* Allocate small arguments on the stack to save memory and be
976 faster - use long to make sure the buffer is aligned properly
977 on 64 bit archs to avoid unaligned access */
978 long stack_pps
[POLL_STACK_ALLOC
/sizeof(long)];
979 struct poll_list
*const head
= (struct poll_list
*)stack_pps
;
980 struct poll_list
*walk
= head
;
981 unsigned long todo
= nfds
;
983 if (nfds
> rlimit(RLIMIT_NOFILE
))
986 len
= min_t(unsigned int, nfds
, N_STACK_PPS
);
993 if (copy_from_user(walk
->entries
, ufds
+ nfds
-todo
,
994 sizeof(struct pollfd
) * walk
->len
))
1001 len
= min(todo
, POLLFD_PER_PAGE
);
1002 walk
= walk
->next
= kmalloc(struct_size(walk
, entries
, len
),
1010 poll_initwait(&table
);
1011 fdcount
= do_poll(head
, &table
, end_time
);
1012 poll_freewait(&table
);
1014 for (walk
= head
; walk
; walk
= walk
->next
) {
1015 struct pollfd
*fds
= walk
->entries
;
1018 for (j
= 0; j
< walk
->len
; j
++, ufds
++)
1019 if (__put_user(fds
[j
].revents
, &ufds
->revents
))
1027 struct poll_list
*pos
= walk
;
1035 static long do_restart_poll(struct restart_block
*restart_block
)
1037 struct pollfd __user
*ufds
= restart_block
->poll
.ufds
;
1038 int nfds
= restart_block
->poll
.nfds
;
1039 struct timespec64
*to
= NULL
, end_time
;
1042 if (restart_block
->poll
.has_timeout
) {
1043 end_time
.tv_sec
= restart_block
->poll
.tv_sec
;
1044 end_time
.tv_nsec
= restart_block
->poll
.tv_nsec
;
1048 ret
= do_sys_poll(ufds
, nfds
, to
);
1050 if (ret
== -ERESTARTNOHAND
) {
1051 restart_block
->fn
= do_restart_poll
;
1052 ret
= -ERESTART_RESTARTBLOCK
;
1057 SYSCALL_DEFINE3(poll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1060 struct timespec64 end_time
, *to
= NULL
;
1063 if (timeout_msecs
>= 0) {
1065 poll_select_set_timeout(to
, timeout_msecs
/ MSEC_PER_SEC
,
1066 NSEC_PER_MSEC
* (timeout_msecs
% MSEC_PER_SEC
));
1069 ret
= do_sys_poll(ufds
, nfds
, to
);
1071 if (ret
== -ERESTARTNOHAND
) {
1072 struct restart_block
*restart_block
;
1074 restart_block
= ¤t
->restart_block
;
1075 restart_block
->fn
= do_restart_poll
;
1076 restart_block
->poll
.ufds
= ufds
;
1077 restart_block
->poll
.nfds
= nfds
;
1079 if (timeout_msecs
>= 0) {
1080 restart_block
->poll
.tv_sec
= end_time
.tv_sec
;
1081 restart_block
->poll
.tv_nsec
= end_time
.tv_nsec
;
1082 restart_block
->poll
.has_timeout
= 1;
1084 restart_block
->poll
.has_timeout
= 0;
1086 ret
= -ERESTART_RESTARTBLOCK
;
1091 SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1092 struct __kernel_timespec __user
*, tsp
, const sigset_t __user
*, sigmask
,
1095 struct timespec64 ts
, end_time
, *to
= NULL
;
1099 if (get_timespec64(&ts
, tsp
))
1103 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1107 ret
= set_user_sigmask(sigmask
, sigsetsize
);
1111 ret
= do_sys_poll(ufds
, nfds
, to
);
1112 return poll_select_finish(&end_time
, tsp
, PT_TIMESPEC
, ret
);
1115 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1117 SYSCALL_DEFINE5(ppoll_time32
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1118 struct old_timespec32 __user
*, tsp
, const sigset_t __user
*, sigmask
,
1121 struct timespec64 ts
, end_time
, *to
= NULL
;
1125 if (get_old_timespec32(&ts
, tsp
))
1129 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1133 ret
= set_user_sigmask(sigmask
, sigsetsize
);
1137 ret
= do_sys_poll(ufds
, nfds
, to
);
1138 return poll_select_finish(&end_time
, tsp
, PT_OLD_TIMESPEC
, ret
);
1142 #ifdef CONFIG_COMPAT
1143 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1146 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1147 * 64-bit unsigned longs.
1150 int compat_get_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1151 unsigned long *fdset
)
1154 return compat_get_bitmap(fdset
, ufdset
, nr
);
1156 zero_fd_set(nr
, fdset
);
1162 int compat_set_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1163 unsigned long *fdset
)
1167 return compat_put_bitmap(ufdset
, fdset
, nr
);
1172 * This is a virtual copy of sys_select from fs/select.c and probably
1173 * should be compared to it from time to time
1177 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1178 * like to be certain this leads to no problems. So I return
1179 * EINTR just for safety.
1181 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1182 * I'm trying ERESTARTNOHAND which restart only when you want to.
1184 static int compat_core_sys_select(int n
, compat_ulong_t __user
*inp
,
1185 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1186 struct timespec64
*end_time
)
1190 int size
, max_fds
, ret
= -EINVAL
;
1191 struct fdtable
*fdt
;
1192 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
1197 /* max_fds can increase, so grab it once to avoid race */
1199 fdt
= files_fdtable(current
->files
);
1200 max_fds
= fdt
->max_fds
;
1206 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1207 * since we used fdset we need to allocate memory in units of
1210 size
= FDS_BYTES(n
);
1212 if (size
> sizeof(stack_fds
) / 6) {
1213 bits
= kmalloc_array(6, size
, GFP_KERNEL
);
1218 fds
.in
= (unsigned long *) bits
;
1219 fds
.out
= (unsigned long *) (bits
+ size
);
1220 fds
.ex
= (unsigned long *) (bits
+ 2*size
);
1221 fds
.res_in
= (unsigned long *) (bits
+ 3*size
);
1222 fds
.res_out
= (unsigned long *) (bits
+ 4*size
);
1223 fds
.res_ex
= (unsigned long *) (bits
+ 5*size
);
1225 if ((ret
= compat_get_fd_set(n
, inp
, fds
.in
)) ||
1226 (ret
= compat_get_fd_set(n
, outp
, fds
.out
)) ||
1227 (ret
= compat_get_fd_set(n
, exp
, fds
.ex
)))
1229 zero_fd_set(n
, fds
.res_in
);
1230 zero_fd_set(n
, fds
.res_out
);
1231 zero_fd_set(n
, fds
.res_ex
);
1233 ret
= do_select(n
, &fds
, end_time
);
1238 ret
= -ERESTARTNOHAND
;
1239 if (signal_pending(current
))
1244 if (compat_set_fd_set(n
, inp
, fds
.res_in
) ||
1245 compat_set_fd_set(n
, outp
, fds
.res_out
) ||
1246 compat_set_fd_set(n
, exp
, fds
.res_ex
))
1249 if (bits
!= stack_fds
)
1255 static int do_compat_select(int n
, compat_ulong_t __user
*inp
,
1256 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1257 struct old_timeval32 __user
*tvp
)
1259 struct timespec64 end_time
, *to
= NULL
;
1260 struct old_timeval32 tv
;
1264 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
1268 if (poll_select_set_timeout(to
,
1269 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
1270 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
1274 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1275 return poll_select_finish(&end_time
, tvp
, PT_OLD_TIMEVAL
, ret
);
1278 COMPAT_SYSCALL_DEFINE5(select
, int, n
, compat_ulong_t __user
*, inp
,
1279 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1280 struct old_timeval32 __user
*, tvp
)
1282 return do_compat_select(n
, inp
, outp
, exp
, tvp
);
1285 struct compat_sel_arg_struct
{
1293 COMPAT_SYSCALL_DEFINE1(old_select
, struct compat_sel_arg_struct __user
*, arg
)
1295 struct compat_sel_arg_struct a
;
1297 if (copy_from_user(&a
, arg
, sizeof(a
)))
1299 return do_compat_select(a
.n
, compat_ptr(a
.inp
), compat_ptr(a
.outp
),
1300 compat_ptr(a
.exp
), compat_ptr(a
.tvp
));
1303 static long do_compat_pselect(int n
, compat_ulong_t __user
*inp
,
1304 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1305 void __user
*tsp
, compat_sigset_t __user
*sigmask
,
1306 compat_size_t sigsetsize
, enum poll_time_type type
)
1308 struct timespec64 ts
, end_time
, *to
= NULL
;
1313 case PT_OLD_TIMESPEC
:
1314 if (get_old_timespec32(&ts
, tsp
))
1318 if (get_timespec64(&ts
, tsp
))
1326 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1330 ret
= set_compat_user_sigmask(sigmask
, sigsetsize
);
1334 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1335 return poll_select_finish(&end_time
, tsp
, type
, ret
);
1338 struct compat_sigset_argpack
{
1342 static inline int get_compat_sigset_argpack(struct compat_sigset_argpack
*to
,
1343 struct compat_sigset_argpack __user
*from
)
1346 if (!user_read_access_begin(from
, sizeof(*from
)))
1348 unsafe_get_user(to
->p
, &from
->p
, Efault
);
1349 unsafe_get_user(to
->size
, &from
->size
, Efault
);
1350 user_read_access_end();
1358 COMPAT_SYSCALL_DEFINE6(pselect6_time64
, int, n
, compat_ulong_t __user
*, inp
,
1359 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1360 struct __kernel_timespec __user
*, tsp
, void __user
*, sig
)
1362 struct compat_sigset_argpack x
= {0, 0};
1364 if (get_compat_sigset_argpack(&x
, sig
))
1367 return do_compat_pselect(n
, inp
, outp
, exp
, tsp
, compat_ptr(x
.p
),
1368 x
.size
, PT_TIMESPEC
);
1371 #if defined(CONFIG_COMPAT_32BIT_TIME)
1373 COMPAT_SYSCALL_DEFINE6(pselect6_time32
, int, n
, compat_ulong_t __user
*, inp
,
1374 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1375 struct old_timespec32 __user
*, tsp
, void __user
*, sig
)
1377 struct compat_sigset_argpack x
= {0, 0};
1379 if (get_compat_sigset_argpack(&x
, sig
))
1382 return do_compat_pselect(n
, inp
, outp
, exp
, tsp
, compat_ptr(x
.p
),
1383 x
.size
, PT_OLD_TIMESPEC
);
1388 #if defined(CONFIG_COMPAT_32BIT_TIME)
1389 COMPAT_SYSCALL_DEFINE5(ppoll_time32
, struct pollfd __user
*, ufds
,
1390 unsigned int, nfds
, struct old_timespec32 __user
*, tsp
,
1391 const compat_sigset_t __user
*, sigmask
, compat_size_t
, sigsetsize
)
1393 struct timespec64 ts
, end_time
, *to
= NULL
;
1397 if (get_old_timespec32(&ts
, tsp
))
1401 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1405 ret
= set_compat_user_sigmask(sigmask
, sigsetsize
);
1409 ret
= do_sys_poll(ufds
, nfds
, to
);
1410 return poll_select_finish(&end_time
, tsp
, PT_OLD_TIMESPEC
, ret
);
1414 /* New compat syscall for 64 bit time_t*/
1415 COMPAT_SYSCALL_DEFINE5(ppoll_time64
, struct pollfd __user
*, ufds
,
1416 unsigned int, nfds
, struct __kernel_timespec __user
*, tsp
,
1417 const compat_sigset_t __user
*, sigmask
, compat_size_t
, sigsetsize
)
1419 struct timespec64 ts
, end_time
, *to
= NULL
;
1423 if (get_timespec64(&ts
, tsp
))
1427 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1431 ret
= set_compat_user_sigmask(sigmask
, sigsetsize
);
1435 ret
= do_sys_poll(ufds
, nfds
, to
);
1436 return poll_select_finish(&end_time
, tsp
, PT_TIMESPEC
, ret
);