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[];
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 |\
463 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\
465 #define POLLEX_SET (EPOLLPRI | EPOLLNVAL)
467 static inline void wait_key_set(poll_table *wait, unsigned long in,
468 unsigned long out, unsigned long bit,
471 wait->_key = POLLEX_SET | ll_flag;
473 wait->_key |= POLLIN_SET;
475 wait->_key |= POLLOUT_SET;
478 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
480 ktime_t expire, *to = NULL;
481 struct poll_wqueues table;
483 int retval, i, timed_out = 0;
485 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
486 unsigned long busy_start = 0;
489 retval = max_select_fd(n, fds);
496 poll_initwait(&table);
498 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
503 if (end_time && !timed_out)
504 slack = select_estimate_accuracy(end_time);
508 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
509 bool can_busy_loop = false;
511 inp = fds->in; outp = fds->out; exp = fds->ex;
512 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
514 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
515 unsigned long in, out, ex, all_bits, bit = 1, j;
516 unsigned long res_in = 0, res_out = 0, res_ex = 0;
519 in = *inp++; out = *outp++; ex = *exp++;
520 all_bits = in | out | ex;
526 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
530 if (!(bit & all_bits))
535 wait_key_set(wait, in, out, bit,
537 mask = vfs_poll(f.file, wait);
541 if ((mask & POLLIN_SET) && (in & bit)) {
546 if ((mask & POLLOUT_SET) && (out & bit)) {
551 if ((mask & POLLEX_SET) && (ex & bit)) {
556 /* got something, stop busy polling */
558 can_busy_loop = false;
562 * only remember a returned
563 * POLL_BUSY_LOOP if we asked for it
565 } else if (busy_flag & mask)
566 can_busy_loop = true;
578 if (retval || timed_out || signal_pending(current))
581 retval = table.error;
585 /* only if found POLL_BUSY_LOOP sockets && not out of time */
586 if (can_busy_loop && !need_resched()) {
588 busy_start = busy_loop_current_time();
591 if (!busy_loop_timeout(busy_start))
597 * If this is the first loop and we have a timeout
598 * given, then we convert to ktime_t and set the to
599 * pointer to the expiry value.
601 if (end_time && !to) {
602 expire = timespec64_to_ktime(*end_time);
606 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
611 poll_freewait(&table);
617 * We can actually return ERESTARTSYS instead of EINTR, but I'd
618 * like to be certain this leads to no problems. So I return
619 * EINTR just for safety.
621 * Update: ERESTARTSYS breaks at least the xview clock binary, so
622 * I'm trying ERESTARTNOHAND which restart only when you want to.
624 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
625 fd_set __user *exp, struct timespec64 *end_time)
630 size_t size, alloc_size;
632 /* Allocate small arguments on the stack to save memory and be faster */
633 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
639 /* max_fds can increase, so grab it once to avoid race */
641 fdt = files_fdtable(current->files);
642 max_fds = fdt->max_fds;
648 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
649 * since we used fdset we need to allocate memory in units of
654 if (size > sizeof(stack_fds) / 6) {
655 /* Not enough space in on-stack array; must use kmalloc */
657 if (size > (SIZE_MAX / 6))
660 alloc_size = 6 * size;
661 bits = kvmalloc(alloc_size, GFP_KERNEL);
666 fds.out = bits + size;
667 fds.ex = bits + 2*size;
668 fds.res_in = bits + 3*size;
669 fds.res_out = bits + 4*size;
670 fds.res_ex = bits + 5*size;
672 if ((ret = get_fd_set(n, inp, fds.in)) ||
673 (ret = get_fd_set(n, outp, fds.out)) ||
674 (ret = get_fd_set(n, exp, fds.ex)))
676 zero_fd_set(n, fds.res_in);
677 zero_fd_set(n, fds.res_out);
678 zero_fd_set(n, fds.res_ex);
680 ret = do_select(n, &fds, end_time);
685 ret = -ERESTARTNOHAND;
686 if (signal_pending(current))
691 if (set_fd_set(n, inp, fds.res_in) ||
692 set_fd_set(n, outp, fds.res_out) ||
693 set_fd_set(n, exp, fds.res_ex))
697 if (bits != stack_fds)
703 static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
704 fd_set __user *exp, struct __kernel_old_timeval __user *tvp)
706 struct timespec64 end_time, *to = NULL;
707 struct __kernel_old_timeval tv;
711 if (copy_from_user(&tv, tvp, sizeof(tv)))
715 if (poll_select_set_timeout(to,
716 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
717 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
721 ret = core_sys_select(n, inp, outp, exp, to);
722 return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
725 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
726 fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp)
728 return kern_select(n, inp, outp, exp, tvp);
731 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
732 fd_set __user *exp, void __user *tsp,
733 const sigset_t __user *sigmask, size_t sigsetsize,
734 enum poll_time_type type)
736 struct timespec64 ts, end_time, *to = NULL;
742 if (get_timespec64(&ts, tsp))
745 case PT_OLD_TIMESPEC:
746 if (get_old_timespec32(&ts, tsp))
754 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
758 ret = set_user_sigmask(sigmask, sigsetsize);
762 ret = core_sys_select(n, inp, outp, exp, to);
763 return poll_select_finish(&end_time, tsp, type, ret);
767 * Most architectures can't handle 7-argument syscalls. So we provide a
768 * 6-argument version where the sixth argument is a pointer to a structure
769 * which has a pointer to the sigset_t itself followed by a size_t containing
772 struct sigset_argpack {
777 static inline int get_sigset_argpack(struct sigset_argpack *to,
778 struct sigset_argpack __user *from)
780 // the path is hot enough for overhead of copy_from_user() to matter
782 if (!user_read_access_begin(from, sizeof(*from)))
784 unsafe_get_user(to->p, &from->p, Efault);
785 unsafe_get_user(to->size, &from->size, Efault);
786 user_read_access_end();
794 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
795 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
798 struct sigset_argpack x = {NULL, 0};
800 if (get_sigset_argpack(&x, sig))
803 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
806 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
808 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
809 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
812 struct sigset_argpack x = {NULL, 0};
814 if (get_sigset_argpack(&x, sig))
817 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
822 #ifdef __ARCH_WANT_SYS_OLD_SELECT
823 struct sel_arg_struct {
825 fd_set __user *inp, *outp, *exp;
826 struct __kernel_old_timeval __user *tvp;
829 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
831 struct sel_arg_struct a;
833 if (copy_from_user(&a, arg, sizeof(a)))
835 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
840 struct poll_list *next;
842 struct pollfd entries[];
845 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
848 * Fish for pollable events on the pollfd->fd file descriptor. We're only
849 * interested in events matching the pollfd->events mask, and the result
850 * matching that mask is both recorded in pollfd->revents and returned. The
851 * pwait poll_table will be used by the fd-provided poll handler for waiting,
852 * if pwait->_qproc is non-NULL.
854 static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
859 __poll_t mask = 0, filter;
869 /* userland u16 ->events contains POLL... bitmap */
870 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
871 pwait->_key = filter | busy_flag;
872 mask = vfs_poll(f.file, pwait);
873 if (mask & busy_flag)
874 *can_busy_poll = true;
875 mask &= filter; /* Mask out unneeded events. */
879 /* ... and so does ->revents */
880 pollfd->revents = mangle_poll(mask);
884 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
885 struct timespec64 *end_time)
887 poll_table* pt = &wait->pt;
888 ktime_t expire, *to = NULL;
889 int timed_out = 0, count = 0;
891 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
892 unsigned long busy_start = 0;
894 /* Optimise the no-wait case */
895 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
900 if (end_time && !timed_out)
901 slack = select_estimate_accuracy(end_time);
904 struct poll_list *walk;
905 bool can_busy_loop = false;
907 for (walk = list; walk != NULL; walk = walk->next) {
908 struct pollfd * pfd, * pfd_end;
911 pfd_end = pfd + walk->len;
912 for (; pfd != pfd_end; pfd++) {
914 * Fish for events. If we found one, record it
915 * and kill poll_table->_qproc, so we don't
916 * needlessly register any other waiters after
917 * this. They'll get immediately deregistered
918 * when we break out and return.
920 if (do_pollfd(pfd, pt, &can_busy_loop,
924 /* found something, stop busy polling */
926 can_busy_loop = false;
931 * All waiters have already been registered, so don't provide
932 * a poll_table->_qproc to them on the next loop iteration.
937 if (signal_pending(current))
938 count = -ERESTARTNOHAND;
940 if (count || timed_out)
943 /* only if found POLL_BUSY_LOOP sockets && not out of time */
944 if (can_busy_loop && !need_resched()) {
946 busy_start = busy_loop_current_time();
949 if (!busy_loop_timeout(busy_start))
955 * If this is the first loop and we have a timeout
956 * given, then we convert to ktime_t and set the to
957 * pointer to the expiry value.
959 if (end_time && !to) {
960 expire = timespec64_to_ktime(*end_time);
964 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
970 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
971 sizeof(struct pollfd))
973 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
974 struct timespec64 *end_time)
976 struct poll_wqueues table;
977 int err = -EFAULT, fdcount, len;
978 /* Allocate small arguments on the stack to save memory and be
979 faster - use long to make sure the buffer is aligned properly
980 on 64 bit archs to avoid unaligned access */
981 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
982 struct poll_list *const head = (struct poll_list *)stack_pps;
983 struct poll_list *walk = head;
984 unsigned long todo = nfds;
986 if (nfds > rlimit(RLIMIT_NOFILE))
989 len = min_t(unsigned int, nfds, N_STACK_PPS);
996 if (copy_from_user(walk->entries, ufds + nfds-todo,
997 sizeof(struct pollfd) * walk->len))
1004 len = min(todo, POLLFD_PER_PAGE);
1005 walk = walk->next = kmalloc(struct_size(walk, entries, len),
1013 poll_initwait(&table);
1014 fdcount = do_poll(head, &table, end_time);
1015 poll_freewait(&table);
1017 if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1020 for (walk = head; walk; walk = walk->next) {
1021 struct pollfd *fds = walk->entries;
1024 for (j = walk->len; j; fds++, ufds++, j--)
1025 unsafe_put_user(fds->revents, &ufds->revents, Efault);
1027 user_write_access_end();
1033 struct poll_list *pos = walk;
1041 user_write_access_end();
1046 static long do_restart_poll(struct restart_block *restart_block)
1048 struct pollfd __user *ufds = restart_block->poll.ufds;
1049 int nfds = restart_block->poll.nfds;
1050 struct timespec64 *to = NULL, end_time;
1053 if (restart_block->poll.has_timeout) {
1054 end_time.tv_sec = restart_block->poll.tv_sec;
1055 end_time.tv_nsec = restart_block->poll.tv_nsec;
1059 ret = do_sys_poll(ufds, nfds, to);
1061 if (ret == -ERESTARTNOHAND)
1062 ret = set_restart_fn(restart_block, do_restart_poll);
1067 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1070 struct timespec64 end_time, *to = NULL;
1073 if (timeout_msecs >= 0) {
1075 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1076 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1079 ret = do_sys_poll(ufds, nfds, to);
1081 if (ret == -ERESTARTNOHAND) {
1082 struct restart_block *restart_block;
1084 restart_block = ¤t->restart_block;
1085 restart_block->poll.ufds = ufds;
1086 restart_block->poll.nfds = nfds;
1088 if (timeout_msecs >= 0) {
1089 restart_block->poll.tv_sec = end_time.tv_sec;
1090 restart_block->poll.tv_nsec = end_time.tv_nsec;
1091 restart_block->poll.has_timeout = 1;
1093 restart_block->poll.has_timeout = 0;
1095 ret = set_restart_fn(restart_block, do_restart_poll);
1100 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1101 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1104 struct timespec64 ts, end_time, *to = NULL;
1108 if (get_timespec64(&ts, tsp))
1112 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1116 ret = set_user_sigmask(sigmask, sigsetsize);
1120 ret = do_sys_poll(ufds, nfds, to);
1121 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1124 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1126 SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1127 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1130 struct timespec64 ts, end_time, *to = NULL;
1134 if (get_old_timespec32(&ts, tsp))
1138 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1142 ret = set_user_sigmask(sigmask, sigsetsize);
1146 ret = do_sys_poll(ufds, nfds, to);
1147 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1151 #ifdef CONFIG_COMPAT
1152 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1155 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1156 * 64-bit unsigned longs.
1159 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1160 unsigned long *fdset)
1163 return compat_get_bitmap(fdset, ufdset, nr);
1165 zero_fd_set(nr, fdset);
1171 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1172 unsigned long *fdset)
1176 return compat_put_bitmap(ufdset, fdset, nr);
1181 * This is a virtual copy of sys_select from fs/select.c and probably
1182 * should be compared to it from time to time
1186 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1187 * like to be certain this leads to no problems. So I return
1188 * EINTR just for safety.
1190 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1191 * I'm trying ERESTARTNOHAND which restart only when you want to.
1193 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1194 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1195 struct timespec64 *end_time)
1199 int size, max_fds, ret = -EINVAL;
1200 struct fdtable *fdt;
1201 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1206 /* max_fds can increase, so grab it once to avoid race */
1208 fdt = files_fdtable(current->files);
1209 max_fds = fdt->max_fds;
1215 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1216 * since we used fdset we need to allocate memory in units of
1219 size = FDS_BYTES(n);
1221 if (size > sizeof(stack_fds) / 6) {
1222 bits = kmalloc_array(6, size, GFP_KERNEL);
1227 fds.in = (unsigned long *) bits;
1228 fds.out = (unsigned long *) (bits + size);
1229 fds.ex = (unsigned long *) (bits + 2*size);
1230 fds.res_in = (unsigned long *) (bits + 3*size);
1231 fds.res_out = (unsigned long *) (bits + 4*size);
1232 fds.res_ex = (unsigned long *) (bits + 5*size);
1234 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1235 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1236 (ret = compat_get_fd_set(n, exp, fds.ex)))
1238 zero_fd_set(n, fds.res_in);
1239 zero_fd_set(n, fds.res_out);
1240 zero_fd_set(n, fds.res_ex);
1242 ret = do_select(n, &fds, end_time);
1247 ret = -ERESTARTNOHAND;
1248 if (signal_pending(current))
1253 if (compat_set_fd_set(n, inp, fds.res_in) ||
1254 compat_set_fd_set(n, outp, fds.res_out) ||
1255 compat_set_fd_set(n, exp, fds.res_ex))
1258 if (bits != stack_fds)
1264 static int do_compat_select(int n, compat_ulong_t __user *inp,
1265 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1266 struct old_timeval32 __user *tvp)
1268 struct timespec64 end_time, *to = NULL;
1269 struct old_timeval32 tv;
1273 if (copy_from_user(&tv, tvp, sizeof(tv)))
1277 if (poll_select_set_timeout(to,
1278 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1279 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1283 ret = compat_core_sys_select(n, inp, outp, exp, to);
1284 return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1287 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1288 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1289 struct old_timeval32 __user *, tvp)
1291 return do_compat_select(n, inp, outp, exp, tvp);
1294 struct compat_sel_arg_struct {
1302 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1304 struct compat_sel_arg_struct a;
1306 if (copy_from_user(&a, arg, sizeof(a)))
1308 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1309 compat_ptr(a.exp), compat_ptr(a.tvp));
1312 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1313 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1314 void __user *tsp, compat_sigset_t __user *sigmask,
1315 compat_size_t sigsetsize, enum poll_time_type type)
1317 struct timespec64 ts, end_time, *to = NULL;
1322 case PT_OLD_TIMESPEC:
1323 if (get_old_timespec32(&ts, tsp))
1327 if (get_timespec64(&ts, tsp))
1335 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1339 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1343 ret = compat_core_sys_select(n, inp, outp, exp, to);
1344 return poll_select_finish(&end_time, tsp, type, ret);
1347 struct compat_sigset_argpack {
1351 static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1352 struct compat_sigset_argpack __user *from)
1355 if (!user_read_access_begin(from, sizeof(*from)))
1357 unsafe_get_user(to->p, &from->p, Efault);
1358 unsafe_get_user(to->size, &from->size, Efault);
1359 user_read_access_end();
1367 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1368 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1369 struct __kernel_timespec __user *, tsp, void __user *, sig)
1371 struct compat_sigset_argpack x = {0, 0};
1373 if (get_compat_sigset_argpack(&x, sig))
1376 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1377 x.size, PT_TIMESPEC);
1380 #if defined(CONFIG_COMPAT_32BIT_TIME)
1382 COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1383 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1384 struct old_timespec32 __user *, tsp, void __user *, sig)
1386 struct compat_sigset_argpack x = {0, 0};
1388 if (get_compat_sigset_argpack(&x, sig))
1391 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1392 x.size, PT_OLD_TIMESPEC);
1397 #if defined(CONFIG_COMPAT_32BIT_TIME)
1398 COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1399 unsigned int, nfds, struct old_timespec32 __user *, tsp,
1400 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1402 struct timespec64 ts, end_time, *to = NULL;
1406 if (get_old_timespec32(&ts, tsp))
1410 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1414 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1418 ret = do_sys_poll(ufds, nfds, to);
1419 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1423 /* New compat syscall for 64 bit time_t*/
1424 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1425 unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1426 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1428 struct timespec64 ts, end_time, *to = NULL;
1432 if (get_timespec64(&ts, tsp))
1436 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1440 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1444 ret = do_sys_poll(ufds, nfds, to);
1445 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);