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/compat.h>
19 #include <linux/kernel.h>
20 #include <linux/sched/signal.h>
21 #include <linux/sched/rt.h>
22 #include <linux/syscalls.h>
23 #include <linux/export.h>
24 #include <linux/slab.h>
25 #include <linux/poll.h>
26 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
30 #include <linux/rcupdate.h>
31 #include <linux/hrtimer.h>
32 #include <linux/freezer.h>
33 #include <net/busy_poll.h>
34 #include <linux/vmalloc.h>
36 #include <linux/uaccess.h>
40 * Estimate expected accuracy in ns from a timeval.
42 * After quite a bit of churning around, we've settled on
43 * a simple thing of taking 0.1% of the timeout as the
44 * slack, with a cap of 100 msec.
45 * "nice" tasks get a 0.5% slack instead.
47 * Consider this comment an open invitation to come up with even
51 #define MAX_SLACK (100 * NSEC_PER_MSEC)
53 static long __estimate_accuracy(struct timespec64
*tv
)
61 if (task_nice(current
) > 0)
62 divfactor
= divfactor
/ 5;
64 if (tv
->tv_sec
> MAX_SLACK
/ (NSEC_PER_SEC
/divfactor
))
67 slack
= tv
->tv_nsec
/ divfactor
;
68 slack
+= tv
->tv_sec
* (NSEC_PER_SEC
/divfactor
);
70 if (slack
> MAX_SLACK
)
76 u64
select_estimate_accuracy(struct timespec64
*tv
)
79 struct timespec64 now
;
82 * Realtime tasks get a slack of 0 for obvious reasons.
89 now
= timespec64_sub(*tv
, now
);
90 ret
= __estimate_accuracy(&now
);
91 if (ret
< current
->timer_slack_ns
)
92 return current
->timer_slack_ns
;
98 struct poll_table_page
{
99 struct poll_table_page
* next
;
100 struct poll_table_entry
* entry
;
101 struct poll_table_entry entries
[];
104 #define POLL_TABLE_FULL(table) \
105 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
108 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
109 * I have rewritten this, taking some shortcuts: This code may not be easy to
110 * follow, but it should be free of race-conditions, and it's practical. If you
111 * understand what I'm doing here, then you understand how the linux
112 * sleep/wakeup mechanism works.
114 * Two very simple procedures, poll_wait() and poll_freewait() make all the
115 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
116 * as all select/poll functions have to call it to add an entry to the
119 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
122 void poll_initwait(struct poll_wqueues
*pwq
)
124 init_poll_funcptr(&pwq
->pt
, __pollwait
);
125 pwq
->polling_task
= current
;
129 pwq
->inline_index
= 0;
131 EXPORT_SYMBOL(poll_initwait
);
133 static void free_poll_entry(struct poll_table_entry
*entry
)
135 remove_wait_queue(entry
->wait_address
, &entry
->wait
);
139 void poll_freewait(struct poll_wqueues
*pwq
)
141 struct poll_table_page
* p
= pwq
->table
;
143 for (i
= 0; i
< pwq
->inline_index
; i
++)
144 free_poll_entry(pwq
->inline_entries
+ i
);
146 struct poll_table_entry
* entry
;
147 struct poll_table_page
*old
;
152 free_poll_entry(entry
);
153 } while (entry
> p
->entries
);
156 free_page((unsigned long) old
);
159 EXPORT_SYMBOL(poll_freewait
);
161 static struct poll_table_entry
*poll_get_entry(struct poll_wqueues
*p
)
163 struct poll_table_page
*table
= p
->table
;
165 if (p
->inline_index
< N_INLINE_POLL_ENTRIES
)
166 return p
->inline_entries
+ p
->inline_index
++;
168 if (!table
|| POLL_TABLE_FULL(table
)) {
169 struct poll_table_page
*new_table
;
171 new_table
= (struct poll_table_page
*) __get_free_page(GFP_KERNEL
);
176 new_table
->entry
= new_table
->entries
;
177 new_table
->next
= table
;
178 p
->table
= new_table
;
182 return table
->entry
++;
185 static int __pollwake(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
187 struct poll_wqueues
*pwq
= wait
->private;
188 DECLARE_WAITQUEUE(dummy_wait
, pwq
->polling_task
);
191 * Although this function is called under waitqueue lock, LOCK
192 * doesn't imply write barrier and the users expect write
193 * barrier semantics on wakeup functions. The following
194 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
195 * and is paired with smp_store_mb() in poll_schedule_timeout.
201 * Perform the default wake up operation using a dummy
204 * TODO: This is hacky but there currently is no interface to
205 * pass in @sync. @sync is scheduled to be removed and once
206 * that happens, wake_up_process() can be used directly.
208 return default_wake_function(&dummy_wait
, mode
, sync
, key
);
211 static int pollwake(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
213 struct poll_table_entry
*entry
;
215 entry
= container_of(wait
, struct poll_table_entry
, wait
);
216 if (key
&& !(key_to_poll(key
) & entry
->key
))
218 return __pollwake(wait
, mode
, sync
, key
);
221 /* Add a new entry */
222 static void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
,
225 struct poll_wqueues
*pwq
= container_of(p
, struct poll_wqueues
, pt
);
226 struct poll_table_entry
*entry
= poll_get_entry(pwq
);
229 entry
->filp
= get_file(filp
);
230 entry
->wait_address
= wait_address
;
231 entry
->key
= p
->_key
;
232 init_waitqueue_func_entry(&entry
->wait
, pollwake
);
233 entry
->wait
.private = pwq
;
234 add_wait_queue(wait_address
, &entry
->wait
);
237 static int poll_schedule_timeout(struct poll_wqueues
*pwq
, int state
,
238 ktime_t
*expires
, unsigned long slack
)
242 set_current_state(state
);
244 rc
= schedule_hrtimeout_range(expires
, slack
, HRTIMER_MODE_ABS
);
245 __set_current_state(TASK_RUNNING
);
248 * Prepare for the next iteration.
250 * The following smp_store_mb() serves two purposes. First, it's
251 * the counterpart rmb of the wmb in pollwake() such that data
252 * written before wake up is always visible after wake up.
253 * Second, the full barrier guarantees that triggered clearing
254 * doesn't pass event check of the next iteration. Note that
255 * this problem doesn't exist for the first iteration as
256 * add_wait_queue() has full barrier semantics.
258 smp_store_mb(pwq
->triggered
, 0);
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 enum poll_time_type
{
298 static int poll_select_finish(struct timespec64
*end_time
,
300 enum poll_time_type pt_type
, int ret
)
302 struct timespec64 rts
;
304 restore_saved_sigmask_unless(ret
== -ERESTARTNOHAND
);
309 if (current
->personality
& STICKY_TIMEOUTS
)
312 /* No update for zero timeout */
313 if (!end_time
->tv_sec
&& !end_time
->tv_nsec
)
316 ktime_get_ts64(&rts
);
317 rts
= timespec64_sub(*end_time
, rts
);
319 rts
.tv_sec
= rts
.tv_nsec
= 0;
325 struct __kernel_old_timeval rtv
;
327 if (sizeof(rtv
) > sizeof(rtv
.tv_sec
) + sizeof(rtv
.tv_usec
))
328 memset(&rtv
, 0, sizeof(rtv
));
329 rtv
.tv_sec
= rts
.tv_sec
;
330 rtv
.tv_usec
= rts
.tv_nsec
/ NSEC_PER_USEC
;
331 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
337 struct old_timeval32 rtv
;
339 rtv
.tv_sec
= rts
.tv_sec
;
340 rtv
.tv_usec
= rts
.tv_nsec
/ NSEC_PER_USEC
;
341 if (!copy_to_user(p
, &rtv
, sizeof(rtv
)))
346 if (!put_timespec64(&rts
, p
))
349 case PT_OLD_TIMESPEC
:
350 if (!put_old_timespec32(&rts
, p
))
357 * If an application puts its timeval in read-only memory, we
358 * don't want the Linux-specific update to the timeval to
359 * cause a fault after the select has completed
360 * successfully. However, because we're not updating the
361 * timeval, we can't restart the system call.
365 if (ret
== -ERESTARTNOHAND
)
371 * Scalable version of the fd_set.
375 unsigned long *in
, *out
, *ex
;
376 unsigned long *res_in
, *res_out
, *res_ex
;
380 * How many longwords for "nr" bits?
382 #define FDS_BITPERLONG (8*sizeof(long))
383 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
384 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
387 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
390 int get_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
394 return copy_from_user(fdset
, ufdset
, nr
) ? -EFAULT
: 0;
396 memset(fdset
, 0, nr
);
400 static inline unsigned long __must_check
401 set_fd_set(unsigned long nr
, void __user
*ufdset
, unsigned long *fdset
)
404 return __copy_to_user(ufdset
, fdset
, FDS_BYTES(nr
));
409 void zero_fd_set(unsigned long nr
, unsigned long *fdset
)
411 memset(fdset
, 0, FDS_BYTES(nr
));
414 #define FDS_IN(fds, n) (fds->in + n)
415 #define FDS_OUT(fds, n) (fds->out + n)
416 #define FDS_EX(fds, n) (fds->ex + n)
418 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
420 static int max_select_fd(unsigned long n
, fd_set_bits
*fds
)
422 unsigned long *open_fds
;
427 /* handle last in-complete long-word first */
428 set
= ~(~0UL << (n
& (BITS_PER_LONG
-1)));
430 fdt
= files_fdtable(current
->files
);
431 open_fds
= fdt
->open_fds
+ n
;
436 if (!(set
& ~*open_fds
))
447 if (set
& ~*open_fds
)
456 max
+= n
* BITS_PER_LONG
;
462 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR |\
464 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\
466 #define POLLEX_SET (EPOLLPRI | EPOLLNVAL)
468 static inline void wait_key_set(poll_table
*wait
, unsigned long in
,
469 unsigned long out
, unsigned long bit
,
472 wait
->_key
= POLLEX_SET
| ll_flag
;
474 wait
->_key
|= POLLIN_SET
;
476 wait
->_key
|= POLLOUT_SET
;
479 static int do_select(int n
, fd_set_bits
*fds
, struct timespec64
*end_time
)
481 ktime_t expire
, *to
= NULL
;
482 struct poll_wqueues table
;
484 int retval
, i
, timed_out
= 0;
486 __poll_t busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
487 unsigned long busy_start
= 0;
490 retval
= max_select_fd(n
, fds
);
497 poll_initwait(&table
);
499 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
504 if (end_time
&& !timed_out
)
505 slack
= select_estimate_accuracy(end_time
);
509 unsigned long *rinp
, *routp
, *rexp
, *inp
, *outp
, *exp
;
510 bool can_busy_loop
= false;
512 inp
= fds
->in
; outp
= fds
->out
; exp
= fds
->ex
;
513 rinp
= fds
->res_in
; routp
= fds
->res_out
; rexp
= fds
->res_ex
;
515 for (i
= 0; i
< n
; ++rinp
, ++routp
, ++rexp
) {
516 unsigned long in
, out
, ex
, all_bits
, bit
= 1, j
;
517 unsigned long res_in
= 0, res_out
= 0, res_ex
= 0;
520 in
= *inp
++; out
= *outp
++; ex
= *exp
++;
521 all_bits
= in
| out
| ex
;
527 for (j
= 0; j
< BITS_PER_LONG
; ++j
, ++i
, bit
<<= 1) {
531 if (!(bit
& all_bits
))
536 wait_key_set(wait
, in
, out
, bit
,
538 mask
= vfs_poll(f
.file
, wait
);
542 if ((mask
& POLLIN_SET
) && (in
& bit
)) {
547 if ((mask
& POLLOUT_SET
) && (out
& bit
)) {
552 if ((mask
& POLLEX_SET
) && (ex
& bit
)) {
557 /* got something, stop busy polling */
559 can_busy_loop
= false;
563 * only remember a returned
564 * POLL_BUSY_LOOP if we asked for it
566 } else if (busy_flag
& mask
)
567 can_busy_loop
= true;
579 if (retval
|| timed_out
|| signal_pending(current
))
582 retval
= table
.error
;
586 /* only if found POLL_BUSY_LOOP sockets && not out of time */
587 if (can_busy_loop
&& !need_resched()) {
589 busy_start
= busy_loop_current_time();
592 if (!busy_loop_timeout(busy_start
))
598 * If this is the first loop and we have a timeout
599 * given, then we convert to ktime_t and set the to
600 * pointer to the expiry value.
602 if (end_time
&& !to
) {
603 expire
= timespec64_to_ktime(*end_time
);
607 if (!poll_schedule_timeout(&table
, TASK_INTERRUPTIBLE
,
612 poll_freewait(&table
);
618 * We can actually return ERESTARTSYS instead of EINTR, but I'd
619 * like to be certain this leads to no problems. So I return
620 * EINTR just for safety.
622 * Update: ERESTARTSYS breaks at least the xview clock binary, so
623 * I'm trying ERESTARTNOHAND which restart only when you want to.
625 int core_sys_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
626 fd_set __user
*exp
, struct timespec64
*end_time
)
631 size_t size
, alloc_size
;
633 /* Allocate small arguments on the stack to save memory and be faster */
634 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
640 /* max_fds can increase, so grab it once to avoid race */
642 fdt
= files_fdtable(current
->files
);
643 max_fds
= fdt
->max_fds
;
649 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
650 * since we used fdset we need to allocate memory in units of
655 if (size
> sizeof(stack_fds
) / 6) {
656 /* Not enough space in on-stack array; must use kmalloc */
658 if (size
> (SIZE_MAX
/ 6))
661 alloc_size
= 6 * size
;
662 bits
= kvmalloc(alloc_size
, GFP_KERNEL
);
667 fds
.out
= bits
+ size
;
668 fds
.ex
= bits
+ 2*size
;
669 fds
.res_in
= bits
+ 3*size
;
670 fds
.res_out
= bits
+ 4*size
;
671 fds
.res_ex
= bits
+ 5*size
;
673 if ((ret
= get_fd_set(n
, inp
, fds
.in
)) ||
674 (ret
= get_fd_set(n
, outp
, fds
.out
)) ||
675 (ret
= get_fd_set(n
, exp
, fds
.ex
)))
677 zero_fd_set(n
, fds
.res_in
);
678 zero_fd_set(n
, fds
.res_out
);
679 zero_fd_set(n
, fds
.res_ex
);
681 ret
= do_select(n
, &fds
, end_time
);
686 ret
= -ERESTARTNOHAND
;
687 if (signal_pending(current
))
692 if (set_fd_set(n
, inp
, fds
.res_in
) ||
693 set_fd_set(n
, outp
, fds
.res_out
) ||
694 set_fd_set(n
, exp
, fds
.res_ex
))
698 if (bits
!= stack_fds
)
704 static int kern_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
705 fd_set __user
*exp
, struct __kernel_old_timeval __user
*tvp
)
707 struct timespec64 end_time
, *to
= NULL
;
708 struct __kernel_old_timeval tv
;
712 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
716 if (poll_select_set_timeout(to
,
717 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
718 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
722 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
723 return poll_select_finish(&end_time
, tvp
, PT_TIMEVAL
, ret
);
726 SYSCALL_DEFINE5(select
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
727 fd_set __user
*, exp
, struct __kernel_old_timeval __user
*, tvp
)
729 return kern_select(n
, inp
, outp
, exp
, tvp
);
732 static long do_pselect(int n
, fd_set __user
*inp
, fd_set __user
*outp
,
733 fd_set __user
*exp
, void __user
*tsp
,
734 const sigset_t __user
*sigmask
, size_t sigsetsize
,
735 enum poll_time_type type
)
737 struct timespec64 ts
, end_time
, *to
= NULL
;
743 if (get_timespec64(&ts
, tsp
))
746 case PT_OLD_TIMESPEC
:
747 if (get_old_timespec32(&ts
, tsp
))
755 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
759 ret
= set_user_sigmask(sigmask
, sigsetsize
);
763 ret
= core_sys_select(n
, inp
, outp
, exp
, to
);
764 return poll_select_finish(&end_time
, tsp
, type
, ret
);
768 * Most architectures can't handle 7-argument syscalls. So we provide a
769 * 6-argument version where the sixth argument is a pointer to a structure
770 * which has a pointer to the sigset_t itself followed by a size_t containing
773 struct sigset_argpack
{
778 static inline int get_sigset_argpack(struct sigset_argpack
*to
,
779 struct sigset_argpack __user
*from
)
781 // the path is hot enough for overhead of copy_from_user() to matter
783 if (!user_read_access_begin(from
, sizeof(*from
)))
785 unsafe_get_user(to
->p
, &from
->p
, Efault
);
786 unsafe_get_user(to
->size
, &from
->size
, Efault
);
787 user_read_access_end();
795 SYSCALL_DEFINE6(pselect6
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
796 fd_set __user
*, exp
, struct __kernel_timespec __user
*, tsp
,
799 struct sigset_argpack x
= {NULL
, 0};
801 if (get_sigset_argpack(&x
, sig
))
804 return do_pselect(n
, inp
, outp
, exp
, tsp
, x
.p
, x
.size
, PT_TIMESPEC
);
807 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
809 SYSCALL_DEFINE6(pselect6_time32
, int, n
, fd_set __user
*, inp
, fd_set __user
*, outp
,
810 fd_set __user
*, exp
, struct old_timespec32 __user
*, tsp
,
813 struct sigset_argpack x
= {NULL
, 0};
815 if (get_sigset_argpack(&x
, sig
))
818 return do_pselect(n
, inp
, outp
, exp
, tsp
, x
.p
, x
.size
, PT_OLD_TIMESPEC
);
823 #ifdef __ARCH_WANT_SYS_OLD_SELECT
824 struct sel_arg_struct
{
826 fd_set __user
*inp
, *outp
, *exp
;
827 struct __kernel_old_timeval __user
*tvp
;
830 SYSCALL_DEFINE1(old_select
, struct sel_arg_struct __user
*, arg
)
832 struct sel_arg_struct a
;
834 if (copy_from_user(&a
, arg
, sizeof(a
)))
836 return kern_select(a
.n
, a
.inp
, a
.outp
, a
.exp
, a
.tvp
);
841 struct poll_list
*next
;
843 struct pollfd entries
[];
846 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
849 * Fish for pollable events on the pollfd->fd file descriptor. We're only
850 * interested in events matching the pollfd->events mask, and the result
851 * matching that mask is both recorded in pollfd->revents and returned. The
852 * pwait poll_table will be used by the fd-provided poll handler for waiting,
853 * if pwait->_qproc is non-NULL.
855 static inline __poll_t
do_pollfd(struct pollfd
*pollfd
, poll_table
*pwait
,
860 __poll_t mask
= 0, filter
;
870 /* userland u16 ->events contains POLL... bitmap */
871 filter
= demangle_poll(pollfd
->events
) | EPOLLERR
| EPOLLHUP
;
872 pwait
->_key
= filter
| busy_flag
;
873 mask
= vfs_poll(f
.file
, pwait
);
874 if (mask
& busy_flag
)
875 *can_busy_poll
= true;
876 mask
&= filter
; /* Mask out unneeded events. */
880 /* ... and so does ->revents */
881 pollfd
->revents
= mangle_poll(mask
);
885 static int do_poll(struct poll_list
*list
, struct poll_wqueues
*wait
,
886 struct timespec64
*end_time
)
888 poll_table
* pt
= &wait
->pt
;
889 ktime_t expire
, *to
= NULL
;
890 int timed_out
= 0, count
= 0;
892 __poll_t busy_flag
= net_busy_loop_on() ? POLL_BUSY_LOOP
: 0;
893 unsigned long busy_start
= 0;
895 /* Optimise the no-wait case */
896 if (end_time
&& !end_time
->tv_sec
&& !end_time
->tv_nsec
) {
901 if (end_time
&& !timed_out
)
902 slack
= select_estimate_accuracy(end_time
);
905 struct poll_list
*walk
;
906 bool can_busy_loop
= false;
908 for (walk
= list
; walk
!= NULL
; walk
= walk
->next
) {
909 struct pollfd
* pfd
, * pfd_end
;
912 pfd_end
= pfd
+ walk
->len
;
913 for (; pfd
!= pfd_end
; pfd
++) {
915 * Fish for events. If we found one, record it
916 * and kill poll_table->_qproc, so we don't
917 * needlessly register any other waiters after
918 * this. They'll get immediately deregistered
919 * when we break out and return.
921 if (do_pollfd(pfd
, pt
, &can_busy_loop
,
925 /* found something, stop busy polling */
927 can_busy_loop
= false;
932 * All waiters have already been registered, so don't provide
933 * a poll_table->_qproc to them on the next loop iteration.
938 if (signal_pending(current
))
939 count
= -ERESTARTNOHAND
;
941 if (count
|| timed_out
)
944 /* only if found POLL_BUSY_LOOP sockets && not out of time */
945 if (can_busy_loop
&& !need_resched()) {
947 busy_start
= busy_loop_current_time();
950 if (!busy_loop_timeout(busy_start
))
956 * If this is the first loop and we have a timeout
957 * given, then we convert to ktime_t and set the to
958 * pointer to the expiry value.
960 if (end_time
&& !to
) {
961 expire
= timespec64_to_ktime(*end_time
);
965 if (!poll_schedule_timeout(wait
, TASK_INTERRUPTIBLE
, to
, slack
))
971 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
972 sizeof(struct pollfd))
974 static int do_sys_poll(struct pollfd __user
*ufds
, unsigned int nfds
,
975 struct timespec64
*end_time
)
977 struct poll_wqueues table
;
978 int err
= -EFAULT
, fdcount
, len
;
979 /* Allocate small arguments on the stack to save memory and be
980 faster - use long to make sure the buffer is aligned properly
981 on 64 bit archs to avoid unaligned access */
982 long stack_pps
[POLL_STACK_ALLOC
/sizeof(long)];
983 struct poll_list
*const head
= (struct poll_list
*)stack_pps
;
984 struct poll_list
*walk
= head
;
985 unsigned long todo
= nfds
;
987 if (nfds
> rlimit(RLIMIT_NOFILE
))
990 len
= min_t(unsigned int, nfds
, N_STACK_PPS
);
997 if (copy_from_user(walk
->entries
, ufds
+ nfds
-todo
,
998 sizeof(struct pollfd
) * walk
->len
))
1005 len
= min(todo
, POLLFD_PER_PAGE
);
1006 walk
= walk
->next
= kmalloc(struct_size(walk
, entries
, len
),
1014 poll_initwait(&table
);
1015 fdcount
= do_poll(head
, &table
, end_time
);
1016 poll_freewait(&table
);
1018 if (!user_write_access_begin(ufds
, nfds
* sizeof(*ufds
)))
1021 for (walk
= head
; walk
; walk
= walk
->next
) {
1022 struct pollfd
*fds
= walk
->entries
;
1025 for (j
= walk
->len
; j
; fds
++, ufds
++, j
--)
1026 unsafe_put_user(fds
->revents
, &ufds
->revents
, Efault
);
1028 user_write_access_end();
1034 struct poll_list
*pos
= walk
;
1042 user_write_access_end();
1047 static long do_restart_poll(struct restart_block
*restart_block
)
1049 struct pollfd __user
*ufds
= restart_block
->poll
.ufds
;
1050 int nfds
= restart_block
->poll
.nfds
;
1051 struct timespec64
*to
= NULL
, end_time
;
1054 if (restart_block
->poll
.has_timeout
) {
1055 end_time
.tv_sec
= restart_block
->poll
.tv_sec
;
1056 end_time
.tv_nsec
= restart_block
->poll
.tv_nsec
;
1060 ret
= do_sys_poll(ufds
, nfds
, to
);
1062 if (ret
== -ERESTARTNOHAND
)
1063 ret
= set_restart_fn(restart_block
, do_restart_poll
);
1068 SYSCALL_DEFINE3(poll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1071 struct timespec64 end_time
, *to
= NULL
;
1074 if (timeout_msecs
>= 0) {
1076 poll_select_set_timeout(to
, timeout_msecs
/ MSEC_PER_SEC
,
1077 NSEC_PER_MSEC
* (timeout_msecs
% MSEC_PER_SEC
));
1080 ret
= do_sys_poll(ufds
, nfds
, to
);
1082 if (ret
== -ERESTARTNOHAND
) {
1083 struct restart_block
*restart_block
;
1085 restart_block
= ¤t
->restart_block
;
1086 restart_block
->poll
.ufds
= ufds
;
1087 restart_block
->poll
.nfds
= nfds
;
1089 if (timeout_msecs
>= 0) {
1090 restart_block
->poll
.tv_sec
= end_time
.tv_sec
;
1091 restart_block
->poll
.tv_nsec
= end_time
.tv_nsec
;
1092 restart_block
->poll
.has_timeout
= 1;
1094 restart_block
->poll
.has_timeout
= 0;
1096 ret
= set_restart_fn(restart_block
, do_restart_poll
);
1101 SYSCALL_DEFINE5(ppoll
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1102 struct __kernel_timespec __user
*, tsp
, const sigset_t __user
*, sigmask
,
1105 struct timespec64 ts
, end_time
, *to
= NULL
;
1109 if (get_timespec64(&ts
, tsp
))
1113 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1117 ret
= set_user_sigmask(sigmask
, sigsetsize
);
1121 ret
= do_sys_poll(ufds
, nfds
, to
);
1122 return poll_select_finish(&end_time
, tsp
, PT_TIMESPEC
, ret
);
1125 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1127 SYSCALL_DEFINE5(ppoll_time32
, struct pollfd __user
*, ufds
, unsigned int, nfds
,
1128 struct old_timespec32 __user
*, tsp
, const sigset_t __user
*, sigmask
,
1131 struct timespec64 ts
, end_time
, *to
= NULL
;
1135 if (get_old_timespec32(&ts
, tsp
))
1139 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1143 ret
= set_user_sigmask(sigmask
, sigsetsize
);
1147 ret
= do_sys_poll(ufds
, nfds
, to
);
1148 return poll_select_finish(&end_time
, tsp
, PT_OLD_TIMESPEC
, ret
);
1152 #ifdef CONFIG_COMPAT
1153 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1156 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1157 * 64-bit unsigned longs.
1160 int compat_get_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1161 unsigned long *fdset
)
1164 return compat_get_bitmap(fdset
, ufdset
, nr
);
1166 zero_fd_set(nr
, fdset
);
1172 int compat_set_fd_set(unsigned long nr
, compat_ulong_t __user
*ufdset
,
1173 unsigned long *fdset
)
1177 return compat_put_bitmap(ufdset
, fdset
, nr
);
1182 * This is a virtual copy of sys_select from fs/select.c and probably
1183 * should be compared to it from time to time
1187 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1188 * like to be certain this leads to no problems. So I return
1189 * EINTR just for safety.
1191 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1192 * I'm trying ERESTARTNOHAND which restart only when you want to.
1194 static int compat_core_sys_select(int n
, compat_ulong_t __user
*inp
,
1195 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1196 struct timespec64
*end_time
)
1200 int size
, max_fds
, ret
= -EINVAL
;
1201 struct fdtable
*fdt
;
1202 long stack_fds
[SELECT_STACK_ALLOC
/sizeof(long)];
1207 /* max_fds can increase, so grab it once to avoid race */
1209 fdt
= files_fdtable(current
->files
);
1210 max_fds
= fdt
->max_fds
;
1216 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1217 * since we used fdset we need to allocate memory in units of
1220 size
= FDS_BYTES(n
);
1222 if (size
> sizeof(stack_fds
) / 6) {
1223 bits
= kmalloc_array(6, size
, GFP_KERNEL
);
1228 fds
.in
= (unsigned long *) bits
;
1229 fds
.out
= (unsigned long *) (bits
+ size
);
1230 fds
.ex
= (unsigned long *) (bits
+ 2*size
);
1231 fds
.res_in
= (unsigned long *) (bits
+ 3*size
);
1232 fds
.res_out
= (unsigned long *) (bits
+ 4*size
);
1233 fds
.res_ex
= (unsigned long *) (bits
+ 5*size
);
1235 if ((ret
= compat_get_fd_set(n
, inp
, fds
.in
)) ||
1236 (ret
= compat_get_fd_set(n
, outp
, fds
.out
)) ||
1237 (ret
= compat_get_fd_set(n
, exp
, fds
.ex
)))
1239 zero_fd_set(n
, fds
.res_in
);
1240 zero_fd_set(n
, fds
.res_out
);
1241 zero_fd_set(n
, fds
.res_ex
);
1243 ret
= do_select(n
, &fds
, end_time
);
1248 ret
= -ERESTARTNOHAND
;
1249 if (signal_pending(current
))
1254 if (compat_set_fd_set(n
, inp
, fds
.res_in
) ||
1255 compat_set_fd_set(n
, outp
, fds
.res_out
) ||
1256 compat_set_fd_set(n
, exp
, fds
.res_ex
))
1259 if (bits
!= stack_fds
)
1265 static int do_compat_select(int n
, compat_ulong_t __user
*inp
,
1266 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1267 struct old_timeval32 __user
*tvp
)
1269 struct timespec64 end_time
, *to
= NULL
;
1270 struct old_timeval32 tv
;
1274 if (copy_from_user(&tv
, tvp
, sizeof(tv
)))
1278 if (poll_select_set_timeout(to
,
1279 tv
.tv_sec
+ (tv
.tv_usec
/ USEC_PER_SEC
),
1280 (tv
.tv_usec
% USEC_PER_SEC
) * NSEC_PER_USEC
))
1284 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1285 return poll_select_finish(&end_time
, tvp
, PT_OLD_TIMEVAL
, ret
);
1288 COMPAT_SYSCALL_DEFINE5(select
, int, n
, compat_ulong_t __user
*, inp
,
1289 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1290 struct old_timeval32 __user
*, tvp
)
1292 return do_compat_select(n
, inp
, outp
, exp
, tvp
);
1295 struct compat_sel_arg_struct
{
1303 COMPAT_SYSCALL_DEFINE1(old_select
, struct compat_sel_arg_struct __user
*, arg
)
1305 struct compat_sel_arg_struct a
;
1307 if (copy_from_user(&a
, arg
, sizeof(a
)))
1309 return do_compat_select(a
.n
, compat_ptr(a
.inp
), compat_ptr(a
.outp
),
1310 compat_ptr(a
.exp
), compat_ptr(a
.tvp
));
1313 static long do_compat_pselect(int n
, compat_ulong_t __user
*inp
,
1314 compat_ulong_t __user
*outp
, compat_ulong_t __user
*exp
,
1315 void __user
*tsp
, compat_sigset_t __user
*sigmask
,
1316 compat_size_t sigsetsize
, enum poll_time_type type
)
1318 struct timespec64 ts
, end_time
, *to
= NULL
;
1323 case PT_OLD_TIMESPEC
:
1324 if (get_old_timespec32(&ts
, tsp
))
1328 if (get_timespec64(&ts
, tsp
))
1336 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1340 ret
= set_compat_user_sigmask(sigmask
, sigsetsize
);
1344 ret
= compat_core_sys_select(n
, inp
, outp
, exp
, to
);
1345 return poll_select_finish(&end_time
, tsp
, type
, ret
);
1348 struct compat_sigset_argpack
{
1352 static inline int get_compat_sigset_argpack(struct compat_sigset_argpack
*to
,
1353 struct compat_sigset_argpack __user
*from
)
1356 if (!user_read_access_begin(from
, sizeof(*from
)))
1358 unsafe_get_user(to
->p
, &from
->p
, Efault
);
1359 unsafe_get_user(to
->size
, &from
->size
, Efault
);
1360 user_read_access_end();
1368 COMPAT_SYSCALL_DEFINE6(pselect6_time64
, int, n
, compat_ulong_t __user
*, inp
,
1369 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1370 struct __kernel_timespec __user
*, tsp
, void __user
*, sig
)
1372 struct compat_sigset_argpack x
= {0, 0};
1374 if (get_compat_sigset_argpack(&x
, sig
))
1377 return do_compat_pselect(n
, inp
, outp
, exp
, tsp
, compat_ptr(x
.p
),
1378 x
.size
, PT_TIMESPEC
);
1381 #if defined(CONFIG_COMPAT_32BIT_TIME)
1383 COMPAT_SYSCALL_DEFINE6(pselect6_time32
, int, n
, compat_ulong_t __user
*, inp
,
1384 compat_ulong_t __user
*, outp
, compat_ulong_t __user
*, exp
,
1385 struct old_timespec32 __user
*, tsp
, void __user
*, sig
)
1387 struct compat_sigset_argpack x
= {0, 0};
1389 if (get_compat_sigset_argpack(&x
, sig
))
1392 return do_compat_pselect(n
, inp
, outp
, exp
, tsp
, compat_ptr(x
.p
),
1393 x
.size
, PT_OLD_TIMESPEC
);
1398 #if defined(CONFIG_COMPAT_32BIT_TIME)
1399 COMPAT_SYSCALL_DEFINE5(ppoll_time32
, struct pollfd __user
*, ufds
,
1400 unsigned int, nfds
, struct old_timespec32 __user
*, tsp
,
1401 const compat_sigset_t __user
*, sigmask
, compat_size_t
, sigsetsize
)
1403 struct timespec64 ts
, end_time
, *to
= NULL
;
1407 if (get_old_timespec32(&ts
, tsp
))
1411 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1415 ret
= set_compat_user_sigmask(sigmask
, sigsetsize
);
1419 ret
= do_sys_poll(ufds
, nfds
, to
);
1420 return poll_select_finish(&end_time
, tsp
, PT_OLD_TIMESPEC
, ret
);
1424 /* New compat syscall for 64 bit time_t*/
1425 COMPAT_SYSCALL_DEFINE5(ppoll_time64
, struct pollfd __user
*, ufds
,
1426 unsigned int, nfds
, struct __kernel_timespec __user
*, tsp
,
1427 const compat_sigset_t __user
*, sigmask
, compat_size_t
, sigsetsize
)
1429 struct timespec64 ts
, end_time
, *to
= NULL
;
1433 if (get_timespec64(&ts
, tsp
))
1437 if (poll_select_set_timeout(to
, ts
.tv_sec
, ts
.tv_nsec
))
1441 ret
= set_compat_user_sigmask(sigmask
, sigsetsize
);
1445 ret
= do_sys_poll(ufds
, nfds
, to
);
1446 return poll_select_finish(&end_time
, tsp
, PT_TIMESPEC
, ret
);