1 // SPDX-License-Identifier: GPL-2.0
3 * This file contains the procedures for the handling of select and poll
5 * Created for Linux based loosely upon Mathius Lattner's minix
6 * patches by Peter MacDonald. Heavily edited by Linus.
9 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
10 * flag set in its personality we do *not* modify the given timeout
11 * parameter to reflect time remaining.
14 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
15 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
18 #include <linux/kernel.h>
19 #include <linux/sched/signal.h>
20 #include <linux/sched/rt.h>
21 #include <linux/syscalls.h>
22 #include <linux/export.h>
23 #include <linux/slab.h>
24 #include <linux/poll.h>
25 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/rcupdate.h>
30 #include <linux/hrtimer.h>
31 #include <linux/freezer.h>
32 #include <net/busy_poll.h>
33 #include <linux/vmalloc.h>
35 #include <linux/uaccess.h>
39 * Estimate expected accuracy in ns from a timeval.
41 * After quite a bit of churning around, we've settled on
42 * a simple thing of taking 0.1% of the timeout as the
43 * slack, with a cap of 100 msec.
44 * "nice" tasks get a 0.5% slack instead.
46 * Consider this comment an open invitation to come up with even
50 #define MAX_SLACK (100 * NSEC_PER_MSEC)
52 static long __estimate_accuracy(struct timespec64 *tv)
60 if (task_nice(current) > 0)
61 divfactor = divfactor / 5;
63 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
66 slack = tv->tv_nsec / divfactor;
67 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
69 if (slack > MAX_SLACK)
75 u64 select_estimate_accuracy(struct timespec64 *tv)
78 struct timespec64 now;
81 * Realtime tasks get a slack of 0 for obvious reasons.
88 now = timespec64_sub(*tv, now);
89 ret = __estimate_accuracy(&now);
90 if (ret < current->timer_slack_ns)
91 return current->timer_slack_ns;
97 struct poll_table_page {
98 struct poll_table_page * next;
99 struct poll_table_entry * entry;
100 struct poll_table_entry entries[0];
103 #define POLL_TABLE_FULL(table) \
104 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
107 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
108 * I have rewritten this, taking some shortcuts: This code may not be easy to
109 * follow, but it should be free of race-conditions, and it's practical. If you
110 * understand what I'm doing here, then you understand how the linux
111 * sleep/wakeup mechanism works.
113 * Two very simple procedures, poll_wait() and poll_freewait() make all the
114 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
115 * as all select/poll functions have to call it to add an entry to the
118 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
121 void poll_initwait(struct poll_wqueues *pwq)
123 init_poll_funcptr(&pwq->pt, __pollwait);
124 pwq->polling_task = current;
128 pwq->inline_index = 0;
130 EXPORT_SYMBOL(poll_initwait);
132 static void free_poll_entry(struct poll_table_entry *entry)
134 remove_wait_queue(entry->wait_address, &entry->wait);
138 void poll_freewait(struct poll_wqueues *pwq)
140 struct poll_table_page * p = pwq->table;
142 for (i = 0; i < pwq->inline_index; i++)
143 free_poll_entry(pwq->inline_entries + i);
145 struct poll_table_entry * entry;
146 struct poll_table_page *old;
151 free_poll_entry(entry);
152 } while (entry > p->entries);
155 free_page((unsigned long) old);
158 EXPORT_SYMBOL(poll_freewait);
160 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
162 struct poll_table_page *table = p->table;
164 if (p->inline_index < N_INLINE_POLL_ENTRIES)
165 return p->inline_entries + p->inline_index++;
167 if (!table || POLL_TABLE_FULL(table)) {
168 struct poll_table_page *new_table;
170 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
175 new_table->entry = new_table->entries;
176 new_table->next = table;
177 p->table = new_table;
181 return table->entry++;
184 static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
186 struct poll_wqueues *pwq = wait->private;
187 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
190 * Although this function is called under waitqueue lock, LOCK
191 * doesn't imply write barrier and the users expect write
192 * barrier semantics on wakeup functions. The following
193 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
194 * and is paired with smp_store_mb() in poll_schedule_timeout.
200 * Perform the default wake up operation using a dummy
203 * TODO: This is hacky but there currently is no interface to
204 * pass in @sync. @sync is scheduled to be removed and once
205 * that happens, wake_up_process() can be used directly.
207 return default_wake_function(&dummy_wait, mode, sync, key);
210 static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
212 struct poll_table_entry *entry;
214 entry = container_of(wait, struct poll_table_entry, wait);
215 if (key && !((unsigned long)key & entry->key))
217 return __pollwake(wait, mode, sync, key);
220 /* Add a new entry */
221 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
224 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
225 struct poll_table_entry *entry = poll_get_entry(pwq);
228 entry->filp = get_file(filp);
229 entry->wait_address = wait_address;
230 entry->key = p->_key;
231 init_waitqueue_func_entry(&entry->wait, pollwake);
232 entry->wait.private = pwq;
233 add_wait_queue(wait_address, &entry->wait);
236 int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
237 ktime_t *expires, unsigned long slack)
241 set_current_state(state);
243 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
244 __set_current_state(TASK_RUNNING);
247 * Prepare for the next iteration.
249 * The following smp_store_mb() serves two purposes. First, it's
250 * the counterpart rmb of the wmb in pollwake() such that data
251 * written before wake up is always visible after wake up.
252 * Second, the full barrier guarantees that triggered clearing
253 * doesn't pass event check of the next iteration. Note that
254 * this problem doesn't exist for the first iteration as
255 * add_wait_queue() has full barrier semantics.
257 smp_store_mb(pwq->triggered, 0);
261 EXPORT_SYMBOL(poll_schedule_timeout);
264 * poll_select_set_timeout - helper function to setup the timeout value
265 * @to: pointer to timespec64 variable for the final timeout
266 * @sec: seconds (from user space)
267 * @nsec: nanoseconds (from user space)
269 * Note, we do not use a timespec for the user space value here, That
270 * way we can use the function for timeval and compat interfaces as well.
272 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
274 int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
276 struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
278 if (!timespec64_valid(&ts))
281 /* Optimize for the zero timeout value here */
283 to->tv_sec = to->tv_nsec = 0;
286 *to = timespec64_add_safe(*to, ts);
291 static int poll_select_copy_remaining(struct timespec64 *end_time,
293 int timeval, int ret)
295 struct timespec64 rts;
301 if (current->personality & STICKY_TIMEOUTS)
304 /* No update for zero timeout */
305 if (!end_time->tv_sec && !end_time->tv_nsec)
308 ktime_get_ts64(&rts);
309 rts = timespec64_sub(*end_time, rts);
311 rts.tv_sec = rts.tv_nsec = 0;
315 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
316 memset(&rtv, 0, sizeof(rtv));
317 rtv.tv_sec = rts.tv_sec;
318 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
320 if (!copy_to_user(p, &rtv, sizeof(rtv)))
323 } else if (!put_timespec64(&rts, p))
327 * If an application puts its timeval in read-only memory, we
328 * don't want the Linux-specific update to the timeval to
329 * cause a fault after the select has completed
330 * successfully. However, because we're not updating the
331 * timeval, we can't restart the system call.
335 if (ret == -ERESTARTNOHAND)
341 * Scalable version of the fd_set.
345 unsigned long *in, *out, *ex;
346 unsigned long *res_in, *res_out, *res_ex;
350 * How many longwords for "nr" bits?
352 #define FDS_BITPERLONG (8*sizeof(long))
353 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
354 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
357 * We do a VERIFY_WRITE here even though we are only reading this time:
358 * we'll write to it eventually..
360 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
363 int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
367 return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
369 memset(fdset, 0, nr);
373 static inline unsigned long __must_check
374 set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
377 return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
382 void zero_fd_set(unsigned long nr, unsigned long *fdset)
384 memset(fdset, 0, FDS_BYTES(nr));
387 #define FDS_IN(fds, n) (fds->in + n)
388 #define FDS_OUT(fds, n) (fds->out + n)
389 #define FDS_EX(fds, n) (fds->ex + n)
391 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
393 static int max_select_fd(unsigned long n, fd_set_bits *fds)
395 unsigned long *open_fds;
400 /* handle last in-complete long-word first */
401 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
403 fdt = files_fdtable(current->files);
404 open_fds = fdt->open_fds + n;
409 if (!(set & ~*open_fds))
420 if (set & ~*open_fds)
429 max += n * BITS_PER_LONG;
435 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
436 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
437 #define POLLEX_SET (POLLPRI)
439 static inline void wait_key_set(poll_table *wait, unsigned long in,
440 unsigned long out, unsigned long bit,
441 unsigned int ll_flag)
443 wait->_key = POLLEX_SET | ll_flag;
445 wait->_key |= POLLIN_SET;
447 wait->_key |= POLLOUT_SET;
450 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
452 ktime_t expire, *to = NULL;
453 struct poll_wqueues table;
455 int retval, i, timed_out = 0;
457 unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
458 unsigned long busy_start = 0;
461 retval = max_select_fd(n, fds);
468 poll_initwait(&table);
470 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
475 if (end_time && !timed_out)
476 slack = select_estimate_accuracy(end_time);
480 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
481 bool can_busy_loop = false;
483 inp = fds->in; outp = fds->out; exp = fds->ex;
484 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
486 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
487 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
488 unsigned long res_in = 0, res_out = 0, res_ex = 0;
490 in = *inp++; out = *outp++; ex = *exp++;
491 all_bits = in | out | ex;
497 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
501 if (!(bit & all_bits))
505 const struct file_operations *f_op;
507 mask = DEFAULT_POLLMASK;
509 wait_key_set(wait, in, out,
511 mask = (*f_op->poll)(f.file, wait);
514 if ((mask & POLLIN_SET) && (in & bit)) {
519 if ((mask & POLLOUT_SET) && (out & bit)) {
524 if ((mask & POLLEX_SET) && (ex & bit)) {
529 /* got something, stop busy polling */
531 can_busy_loop = false;
535 * only remember a returned
536 * POLL_BUSY_LOOP if we asked for it
538 } else if (busy_flag & mask)
539 can_busy_loop = true;
552 if (retval || timed_out || signal_pending(current))
555 retval = table.error;
559 /* only if found POLL_BUSY_LOOP sockets && not out of time */
560 if (can_busy_loop && !need_resched()) {
562 busy_start = busy_loop_current_time();
565 if (!busy_loop_timeout(busy_start))
571 * If this is the first loop and we have a timeout
572 * given, then we convert to ktime_t and set the to
573 * pointer to the expiry value.
575 if (end_time && !to) {
576 expire = timespec64_to_ktime(*end_time);
580 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
585 poll_freewait(&table);
591 * We can actually return ERESTARTSYS instead of EINTR, but I'd
592 * like to be certain this leads to no problems. So I return
593 * EINTR just for safety.
595 * Update: ERESTARTSYS breaks at least the xview clock binary, so
596 * I'm trying ERESTARTNOHAND which restart only when you want to.
598 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
599 fd_set __user *exp, struct timespec64 *end_time)
604 size_t size, alloc_size;
606 /* Allocate small arguments on the stack to save memory and be faster */
607 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
613 /* max_fds can increase, so grab it once to avoid race */
615 fdt = files_fdtable(current->files);
616 max_fds = fdt->max_fds;
622 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
623 * since we used fdset we need to allocate memory in units of
628 if (size > sizeof(stack_fds) / 6) {
629 /* Not enough space in on-stack array; must use kmalloc */
631 if (size > (SIZE_MAX / 6))
634 alloc_size = 6 * size;
635 bits = kvmalloc(alloc_size, GFP_KERNEL);
640 fds.out = bits + size;
641 fds.ex = bits + 2*size;
642 fds.res_in = bits + 3*size;
643 fds.res_out = bits + 4*size;
644 fds.res_ex = bits + 5*size;
646 if ((ret = get_fd_set(n, inp, fds.in)) ||
647 (ret = get_fd_set(n, outp, fds.out)) ||
648 (ret = get_fd_set(n, exp, fds.ex)))
650 zero_fd_set(n, fds.res_in);
651 zero_fd_set(n, fds.res_out);
652 zero_fd_set(n, fds.res_ex);
654 ret = do_select(n, &fds, end_time);
659 ret = -ERESTARTNOHAND;
660 if (signal_pending(current))
665 if (set_fd_set(n, inp, fds.res_in) ||
666 set_fd_set(n, outp, fds.res_out) ||
667 set_fd_set(n, exp, fds.res_ex))
671 if (bits != stack_fds)
677 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
678 fd_set __user *, exp, struct timeval __user *, tvp)
680 struct timespec64 end_time, *to = NULL;
685 if (copy_from_user(&tv, tvp, sizeof(tv)))
689 if (poll_select_set_timeout(to,
690 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
691 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
695 ret = core_sys_select(n, inp, outp, exp, to);
696 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
701 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
702 fd_set __user *exp, struct timespec __user *tsp,
703 const sigset_t __user *sigmask, size_t sigsetsize)
705 sigset_t ksigmask, sigsaved;
706 struct timespec64 ts, end_time, *to = NULL;
710 if (get_timespec64(&ts, tsp))
714 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
719 /* XXX: Don't preclude handling different sized sigset_t's. */
720 if (sigsetsize != sizeof(sigset_t))
722 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
725 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
726 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
729 ret = core_sys_select(n, inp, outp, exp, to);
730 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
732 if (ret == -ERESTARTNOHAND) {
734 * Don't restore the signal mask yet. Let do_signal() deliver
735 * the signal on the way back to userspace, before the signal
739 memcpy(¤t->saved_sigmask, &sigsaved,
741 set_restore_sigmask();
744 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
750 * Most architectures can't handle 7-argument syscalls. So we provide a
751 * 6-argument version where the sixth argument is a pointer to a structure
752 * which has a pointer to the sigset_t itself followed by a size_t containing
755 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
756 fd_set __user *, exp, struct timespec __user *, tsp,
759 size_t sigsetsize = 0;
760 sigset_t __user *up = NULL;
763 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
764 || __get_user(up, (sigset_t __user * __user *)sig)
765 || __get_user(sigsetsize,
766 (size_t __user *)(sig+sizeof(void *))))
770 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
773 #ifdef __ARCH_WANT_SYS_OLD_SELECT
774 struct sel_arg_struct {
776 fd_set __user *inp, *outp, *exp;
777 struct timeval __user *tvp;
780 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
782 struct sel_arg_struct a;
784 if (copy_from_user(&a, arg, sizeof(a)))
786 return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
791 struct poll_list *next;
793 struct pollfd entries[0];
796 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
799 * Fish for pollable events on the pollfd->fd file descriptor. We're only
800 * interested in events matching the pollfd->events mask, and the result
801 * matching that mask is both recorded in pollfd->revents and returned. The
802 * pwait poll_table will be used by the fd-provided poll handler for waiting,
803 * if pwait->_qproc is non-NULL.
805 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait,
807 unsigned int busy_flag)
815 struct fd f = fdget(fd);
818 mask = DEFAULT_POLLMASK;
819 if (f.file->f_op->poll) {
820 pwait->_key = pollfd->events|POLLERR|POLLHUP;
821 pwait->_key |= busy_flag;
822 mask = f.file->f_op->poll(f.file, pwait);
823 if (mask & busy_flag)
824 *can_busy_poll = true;
826 /* Mask out unneeded events. */
827 mask &= pollfd->events | POLLERR | POLLHUP;
831 pollfd->revents = mask;
836 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
837 struct timespec64 *end_time)
839 poll_table* pt = &wait->pt;
840 ktime_t expire, *to = NULL;
841 int timed_out = 0, count = 0;
843 unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
844 unsigned long busy_start = 0;
846 /* Optimise the no-wait case */
847 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
852 if (end_time && !timed_out)
853 slack = select_estimate_accuracy(end_time);
856 struct poll_list *walk;
857 bool can_busy_loop = false;
859 for (walk = list; walk != NULL; walk = walk->next) {
860 struct pollfd * pfd, * pfd_end;
863 pfd_end = pfd + walk->len;
864 for (; pfd != pfd_end; pfd++) {
866 * Fish for events. If we found one, record it
867 * and kill poll_table->_qproc, so we don't
868 * needlessly register any other waiters after
869 * this. They'll get immediately deregistered
870 * when we break out and return.
872 if (do_pollfd(pfd, pt, &can_busy_loop,
876 /* found something, stop busy polling */
878 can_busy_loop = false;
883 * All waiters have already been registered, so don't provide
884 * a poll_table->_qproc to them on the next loop iteration.
889 if (signal_pending(current))
892 if (count || timed_out)
895 /* only if found POLL_BUSY_LOOP sockets && not out of time */
896 if (can_busy_loop && !need_resched()) {
898 busy_start = busy_loop_current_time();
901 if (!busy_loop_timeout(busy_start))
907 * If this is the first loop and we have a timeout
908 * given, then we convert to ktime_t and set the to
909 * pointer to the expiry value.
911 if (end_time && !to) {
912 expire = timespec64_to_ktime(*end_time);
916 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
922 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
923 sizeof(struct pollfd))
925 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
926 struct timespec64 *end_time)
928 struct poll_wqueues table;
929 int err = -EFAULT, fdcount, len, size;
930 /* Allocate small arguments on the stack to save memory and be
931 faster - use long to make sure the buffer is aligned properly
932 on 64 bit archs to avoid unaligned access */
933 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
934 struct poll_list *const head = (struct poll_list *)stack_pps;
935 struct poll_list *walk = head;
936 unsigned long todo = nfds;
938 if (nfds > rlimit(RLIMIT_NOFILE))
941 len = min_t(unsigned int, nfds, N_STACK_PPS);
948 if (copy_from_user(walk->entries, ufds + nfds-todo,
949 sizeof(struct pollfd) * walk->len))
956 len = min(todo, POLLFD_PER_PAGE);
957 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
958 walk = walk->next = kmalloc(size, GFP_KERNEL);
965 poll_initwait(&table);
966 fdcount = do_poll(head, &table, end_time);
967 poll_freewait(&table);
969 for (walk = head; walk; walk = walk->next) {
970 struct pollfd *fds = walk->entries;
973 for (j = 0; j < walk->len; j++, ufds++)
974 if (__put_user(fds[j].revents, &ufds->revents))
982 struct poll_list *pos = walk;
990 static long do_restart_poll(struct restart_block *restart_block)
992 struct pollfd __user *ufds = restart_block->poll.ufds;
993 int nfds = restart_block->poll.nfds;
994 struct timespec64 *to = NULL, end_time;
997 if (restart_block->poll.has_timeout) {
998 end_time.tv_sec = restart_block->poll.tv_sec;
999 end_time.tv_nsec = restart_block->poll.tv_nsec;
1003 ret = do_sys_poll(ufds, nfds, to);
1005 if (ret == -EINTR) {
1006 restart_block->fn = do_restart_poll;
1007 ret = -ERESTART_RESTARTBLOCK;
1012 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1015 struct timespec64 end_time, *to = NULL;
1018 if (timeout_msecs >= 0) {
1020 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1021 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1024 ret = do_sys_poll(ufds, nfds, to);
1026 if (ret == -EINTR) {
1027 struct restart_block *restart_block;
1029 restart_block = ¤t->restart_block;
1030 restart_block->fn = do_restart_poll;
1031 restart_block->poll.ufds = ufds;
1032 restart_block->poll.nfds = nfds;
1034 if (timeout_msecs >= 0) {
1035 restart_block->poll.tv_sec = end_time.tv_sec;
1036 restart_block->poll.tv_nsec = end_time.tv_nsec;
1037 restart_block->poll.has_timeout = 1;
1039 restart_block->poll.has_timeout = 0;
1041 ret = -ERESTART_RESTARTBLOCK;
1046 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1047 struct timespec __user *, tsp, const sigset_t __user *, sigmask,
1050 sigset_t ksigmask, sigsaved;
1051 struct timespec64 ts, end_time, *to = NULL;
1055 if (get_timespec64(&ts, tsp))
1059 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1064 /* XXX: Don't preclude handling different sized sigset_t's. */
1065 if (sigsetsize != sizeof(sigset_t))
1067 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1070 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1071 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1074 ret = do_sys_poll(ufds, nfds, to);
1076 /* We can restart this syscall, usually */
1077 if (ret == -EINTR) {
1079 * Don't restore the signal mask yet. Let do_signal() deliver
1080 * the signal on the way back to userspace, before the signal
1084 memcpy(¤t->saved_sigmask, &sigsaved,
1086 set_restore_sigmask();
1088 ret = -ERESTARTNOHAND;
1090 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1092 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1097 #ifdef CONFIG_COMPAT
1098 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1101 int compat_poll_select_copy_remaining(struct timespec64 *end_time, void __user *p,
1102 int timeval, int ret)
1104 struct timespec64 ts;
1109 if (current->personality & STICKY_TIMEOUTS)
1112 /* No update for zero timeout */
1113 if (!end_time->tv_sec && !end_time->tv_nsec)
1116 ktime_get_ts64(&ts);
1117 ts = timespec64_sub(*end_time, ts);
1119 ts.tv_sec = ts.tv_nsec = 0;
1122 struct compat_timeval rtv;
1124 rtv.tv_sec = ts.tv_sec;
1125 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1127 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1130 if (!compat_put_timespec64(&ts, p))
1134 * If an application puts its timeval in read-only memory, we
1135 * don't want the Linux-specific update to the timeval to
1136 * cause a fault after the select has completed
1137 * successfully. However, because we're not updating the
1138 * timeval, we can't restart the system call.
1142 if (ret == -ERESTARTNOHAND)
1148 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1149 * 64-bit unsigned longs.
1152 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1153 unsigned long *fdset)
1156 return compat_get_bitmap(fdset, ufdset, nr);
1158 zero_fd_set(nr, fdset);
1164 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1165 unsigned long *fdset)
1169 return compat_put_bitmap(ufdset, fdset, nr);
1174 * This is a virtual copy of sys_select from fs/select.c and probably
1175 * should be compared to it from time to time
1179 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1180 * like to be certain this leads to no problems. So I return
1181 * EINTR just for safety.
1183 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1184 * I'm trying ERESTARTNOHAND which restart only when you want to.
1186 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1187 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1188 struct timespec64 *end_time)
1192 int size, max_fds, ret = -EINVAL;
1193 struct fdtable *fdt;
1194 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1199 /* max_fds can increase, so grab it once to avoid race */
1201 fdt = files_fdtable(current->files);
1202 max_fds = fdt->max_fds;
1208 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1209 * since we used fdset we need to allocate memory in units of
1212 size = FDS_BYTES(n);
1214 if (size > sizeof(stack_fds) / 6) {
1215 bits = kmalloc(6 * size, GFP_KERNEL);
1220 fds.in = (unsigned long *) bits;
1221 fds.out = (unsigned long *) (bits + size);
1222 fds.ex = (unsigned long *) (bits + 2*size);
1223 fds.res_in = (unsigned long *) (bits + 3*size);
1224 fds.res_out = (unsigned long *) (bits + 4*size);
1225 fds.res_ex = (unsigned long *) (bits + 5*size);
1227 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1228 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1229 (ret = compat_get_fd_set(n, exp, fds.ex)))
1231 zero_fd_set(n, fds.res_in);
1232 zero_fd_set(n, fds.res_out);
1233 zero_fd_set(n, fds.res_ex);
1235 ret = do_select(n, &fds, end_time);
1240 ret = -ERESTARTNOHAND;
1241 if (signal_pending(current))
1246 if (compat_set_fd_set(n, inp, fds.res_in) ||
1247 compat_set_fd_set(n, outp, fds.res_out) ||
1248 compat_set_fd_set(n, exp, fds.res_ex))
1251 if (bits != stack_fds)
1257 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1258 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1259 struct compat_timeval __user *, tvp)
1261 struct timespec64 end_time, *to = NULL;
1262 struct compat_timeval tv;
1266 if (copy_from_user(&tv, tvp, sizeof(tv)))
1270 if (poll_select_set_timeout(to,
1271 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1272 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1276 ret = compat_core_sys_select(n, inp, outp, exp, to);
1277 ret = compat_poll_select_copy_remaining(&end_time, tvp, 1, ret);
1282 struct compat_sel_arg_struct {
1290 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1292 struct compat_sel_arg_struct a;
1294 if (copy_from_user(&a, arg, sizeof(a)))
1296 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1297 compat_ptr(a.exp), compat_ptr(a.tvp));
1300 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1301 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1302 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1303 compat_size_t sigsetsize)
1305 sigset_t ksigmask, sigsaved;
1306 struct timespec64 ts, end_time, *to = NULL;
1310 if (compat_get_timespec64(&ts, tsp))
1314 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1319 if (sigsetsize != sizeof(compat_sigset_t))
1321 if (get_compat_sigset(&ksigmask, sigmask))
1324 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1325 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1328 ret = compat_core_sys_select(n, inp, outp, exp, to);
1329 ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret);
1331 if (ret == -ERESTARTNOHAND) {
1333 * Don't restore the signal mask yet. Let do_signal() deliver
1334 * the signal on the way back to userspace, before the signal
1338 memcpy(¤t->saved_sigmask, &sigsaved,
1340 set_restore_sigmask();
1343 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1348 COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp,
1349 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1350 struct compat_timespec __user *, tsp, void __user *, sig)
1352 compat_size_t sigsetsize = 0;
1353 compat_uptr_t up = 0;
1356 if (!access_ok(VERIFY_READ, sig,
1357 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1358 __get_user(up, (compat_uptr_t __user *)sig) ||
1359 __get_user(sigsetsize,
1360 (compat_size_t __user *)(sig+sizeof(up))))
1363 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1367 COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds,
1368 unsigned int, nfds, struct compat_timespec __user *, tsp,
1369 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1371 sigset_t ksigmask, sigsaved;
1372 struct timespec64 ts, end_time, *to = NULL;
1376 if (compat_get_timespec64(&ts, tsp))
1380 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1385 if (sigsetsize != sizeof(compat_sigset_t))
1387 if (get_compat_sigset(&ksigmask, sigmask))
1390 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1391 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1394 ret = do_sys_poll(ufds, nfds, to);
1396 /* We can restart this syscall, usually */
1397 if (ret == -EINTR) {
1399 * Don't restore the signal mask yet. Let do_signal() deliver
1400 * the signal on the way back to userspace, before the signal
1404 memcpy(¤t->saved_sigmask, &sigsaved,
1406 set_restore_sigmask();
1408 ret = -ERESTARTNOHAND;
1410 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1412 ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret);