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_copy_remaining(struct timespec64 *end_time,
299 enum poll_time_type pt_type, int ret)
301 struct timespec64 rts;
306 if (current->personality & STICKY_TIMEOUTS)
309 /* No update for zero timeout */
310 if (!end_time->tv_sec && !end_time->tv_nsec)
313 ktime_get_ts64(&rts);
314 rts = timespec64_sub(*end_time, rts);
316 rts.tv_sec = rts.tv_nsec = 0;
324 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
325 memset(&rtv, 0, sizeof(rtv));
326 rtv.tv_sec = rts.tv_sec;
327 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
328 if (!copy_to_user(p, &rtv, sizeof(rtv)))
334 struct old_timeval32 rtv;
336 rtv.tv_sec = rts.tv_sec;
337 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
338 if (!copy_to_user(p, &rtv, sizeof(rtv)))
343 if (!put_timespec64(&rts, p))
346 case PT_OLD_TIMESPEC:
347 if (!put_old_timespec32(&rts, p))
354 * If an application puts its timeval in read-only memory, we
355 * don't want the Linux-specific update to the timeval to
356 * cause a fault after the select has completed
357 * successfully. However, because we're not updating the
358 * timeval, we can't restart the system call.
362 if (ret == -ERESTARTNOHAND)
368 * Scalable version of the fd_set.
372 unsigned long *in, *out, *ex;
373 unsigned long *res_in, *res_out, *res_ex;
377 * How many longwords for "nr" bits?
379 #define FDS_BITPERLONG (8*sizeof(long))
380 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
381 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
384 * We do a VERIFY_WRITE here even though we are only reading this time:
385 * we'll write to it eventually..
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)
463 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
464 #define POLLEX_SET (EPOLLPRI)
466 static inline void wait_key_set(poll_table *wait, unsigned long in,
467 unsigned long out, unsigned long bit,
470 wait->_key = POLLEX_SET | ll_flag;
472 wait->_key |= POLLIN_SET;
474 wait->_key |= POLLOUT_SET;
477 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
479 ktime_t expire, *to = NULL;
480 struct poll_wqueues table;
482 int retval, i, timed_out = 0;
484 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
485 unsigned long busy_start = 0;
488 retval = max_select_fd(n, fds);
495 poll_initwait(&table);
497 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
502 if (end_time && !timed_out)
503 slack = select_estimate_accuracy(end_time);
507 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
508 bool can_busy_loop = false;
510 inp = fds->in; outp = fds->out; exp = fds->ex;
511 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
513 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
514 unsigned long in, out, ex, all_bits, bit = 1, j;
515 unsigned long res_in = 0, res_out = 0, res_ex = 0;
518 in = *inp++; out = *outp++; ex = *exp++;
519 all_bits = in | out | ex;
525 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
529 if (!(bit & all_bits))
533 wait_key_set(wait, in, out, bit,
535 mask = vfs_poll(f.file, wait);
538 if ((mask & POLLIN_SET) && (in & bit)) {
543 if ((mask & POLLOUT_SET) && (out & bit)) {
548 if ((mask & POLLEX_SET) && (ex & bit)) {
553 /* got something, stop busy polling */
555 can_busy_loop = false;
559 * only remember a returned
560 * POLL_BUSY_LOOP if we asked for it
562 } else if (busy_flag & mask)
563 can_busy_loop = true;
576 if (retval || timed_out || signal_pending(current))
579 retval = table.error;
583 /* only if found POLL_BUSY_LOOP sockets && not out of time */
584 if (can_busy_loop && !need_resched()) {
586 busy_start = busy_loop_current_time();
589 if (!busy_loop_timeout(busy_start))
595 * If this is the first loop and we have a timeout
596 * given, then we convert to ktime_t and set the to
597 * pointer to the expiry value.
599 if (end_time && !to) {
600 expire = timespec64_to_ktime(*end_time);
604 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
609 poll_freewait(&table);
615 * We can actually return ERESTARTSYS instead of EINTR, but I'd
616 * like to be certain this leads to no problems. So I return
617 * EINTR just for safety.
619 * Update: ERESTARTSYS breaks at least the xview clock binary, so
620 * I'm trying ERESTARTNOHAND which restart only when you want to.
622 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
623 fd_set __user *exp, struct timespec64 *end_time)
628 size_t size, alloc_size;
630 /* Allocate small arguments on the stack to save memory and be faster */
631 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
637 /* max_fds can increase, so grab it once to avoid race */
639 fdt = files_fdtable(current->files);
640 max_fds = fdt->max_fds;
646 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
647 * since we used fdset we need to allocate memory in units of
652 if (size > sizeof(stack_fds) / 6) {
653 /* Not enough space in on-stack array; must use kmalloc */
655 if (size > (SIZE_MAX / 6))
658 alloc_size = 6 * size;
659 bits = kvmalloc(alloc_size, GFP_KERNEL);
664 fds.out = bits + size;
665 fds.ex = bits + 2*size;
666 fds.res_in = bits + 3*size;
667 fds.res_out = bits + 4*size;
668 fds.res_ex = bits + 5*size;
670 if ((ret = get_fd_set(n, inp, fds.in)) ||
671 (ret = get_fd_set(n, outp, fds.out)) ||
672 (ret = get_fd_set(n, exp, fds.ex)))
674 zero_fd_set(n, fds.res_in);
675 zero_fd_set(n, fds.res_out);
676 zero_fd_set(n, fds.res_ex);
678 ret = do_select(n, &fds, end_time);
683 ret = -ERESTARTNOHAND;
684 if (signal_pending(current))
689 if (set_fd_set(n, inp, fds.res_in) ||
690 set_fd_set(n, outp, fds.res_out) ||
691 set_fd_set(n, exp, fds.res_ex))
695 if (bits != stack_fds)
701 static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
702 fd_set __user *exp, struct timeval __user *tvp)
704 struct timespec64 end_time, *to = NULL;
709 if (copy_from_user(&tv, tvp, sizeof(tv)))
713 if (poll_select_set_timeout(to,
714 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
715 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
719 ret = core_sys_select(n, inp, outp, exp, to);
720 ret = poll_select_copy_remaining(&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 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 sigset_t ksigmask, sigsaved;
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, &ksigmask, &sigsaved, sigsetsize);
763 ret = core_sys_select(n, inp, outp, exp, to);
764 ret = poll_select_copy_remaining(&end_time, tsp, type, ret);
766 restore_user_sigmask(sigmask, &sigsaved);
772 * Most architectures can't handle 7-argument syscalls. So we provide a
773 * 6-argument version where the sixth argument is a pointer to a structure
774 * which has a pointer to the sigset_t itself followed by a size_t containing
777 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
778 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
781 size_t sigsetsize = 0;
782 sigset_t __user *up = NULL;
785 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
786 || __get_user(up, (sigset_t __user * __user *)sig)
787 || __get_user(sigsetsize,
788 (size_t __user *)(sig+sizeof(void *))))
792 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_TIMESPEC);
795 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
797 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
798 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
801 size_t sigsetsize = 0;
802 sigset_t __user *up = NULL;
805 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
806 || __get_user(up, (sigset_t __user * __user *)sig)
807 || __get_user(sigsetsize,
808 (size_t __user *)(sig+sizeof(void *))))
812 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_OLD_TIMESPEC);
817 #ifdef __ARCH_WANT_SYS_OLD_SELECT
818 struct sel_arg_struct {
820 fd_set __user *inp, *outp, *exp;
821 struct timeval __user *tvp;
824 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
826 struct sel_arg_struct a;
828 if (copy_from_user(&a, arg, sizeof(a)))
830 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
835 struct poll_list *next;
837 struct pollfd entries[0];
840 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
843 * Fish for pollable events on the pollfd->fd file descriptor. We're only
844 * interested in events matching the pollfd->events mask, and the result
845 * matching that mask is both recorded in pollfd->revents and returned. The
846 * pwait poll_table will be used by the fd-provided poll handler for waiting,
847 * if pwait->_qproc is non-NULL.
849 static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
854 __poll_t mask = 0, filter;
864 /* userland u16 ->events contains POLL... bitmap */
865 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
866 pwait->_key = filter | busy_flag;
867 mask = vfs_poll(f.file, pwait);
868 if (mask & busy_flag)
869 *can_busy_poll = true;
870 mask &= filter; /* Mask out unneeded events. */
874 /* ... and so does ->revents */
875 pollfd->revents = mangle_poll(mask);
879 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
880 struct timespec64 *end_time)
882 poll_table* pt = &wait->pt;
883 ktime_t expire, *to = NULL;
884 int timed_out = 0, count = 0;
886 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
887 unsigned long busy_start = 0;
889 /* Optimise the no-wait case */
890 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
895 if (end_time && !timed_out)
896 slack = select_estimate_accuracy(end_time);
899 struct poll_list *walk;
900 bool can_busy_loop = false;
902 for (walk = list; walk != NULL; walk = walk->next) {
903 struct pollfd * pfd, * pfd_end;
906 pfd_end = pfd + walk->len;
907 for (; pfd != pfd_end; pfd++) {
909 * Fish for events. If we found one, record it
910 * and kill poll_table->_qproc, so we don't
911 * needlessly register any other waiters after
912 * this. They'll get immediately deregistered
913 * when we break out and return.
915 if (do_pollfd(pfd, pt, &can_busy_loop,
919 /* found something, stop busy polling */
921 can_busy_loop = false;
926 * All waiters have already been registered, so don't provide
927 * a poll_table->_qproc to them on the next loop iteration.
932 if (signal_pending(current))
935 if (count || timed_out)
938 /* only if found POLL_BUSY_LOOP sockets && not out of time */
939 if (can_busy_loop && !need_resched()) {
941 busy_start = busy_loop_current_time();
944 if (!busy_loop_timeout(busy_start))
950 * If this is the first loop and we have a timeout
951 * given, then we convert to ktime_t and set the to
952 * pointer to the expiry value.
954 if (end_time && !to) {
955 expire = timespec64_to_ktime(*end_time);
959 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
965 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
966 sizeof(struct pollfd))
968 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
969 struct timespec64 *end_time)
971 struct poll_wqueues table;
972 int err = -EFAULT, fdcount, len, size;
973 /* Allocate small arguments on the stack to save memory and be
974 faster - use long to make sure the buffer is aligned properly
975 on 64 bit archs to avoid unaligned access */
976 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
977 struct poll_list *const head = (struct poll_list *)stack_pps;
978 struct poll_list *walk = head;
979 unsigned long todo = nfds;
981 if (nfds > rlimit(RLIMIT_NOFILE))
984 len = min_t(unsigned int, nfds, N_STACK_PPS);
991 if (copy_from_user(walk->entries, ufds + nfds-todo,
992 sizeof(struct pollfd) * walk->len))
999 len = min(todo, POLLFD_PER_PAGE);
1000 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
1001 walk = walk->next = kmalloc(size, GFP_KERNEL);
1008 poll_initwait(&table);
1009 fdcount = do_poll(head, &table, end_time);
1010 poll_freewait(&table);
1012 for (walk = head; walk; walk = walk->next) {
1013 struct pollfd *fds = walk->entries;
1016 for (j = 0; j < walk->len; j++, ufds++)
1017 if (__put_user(fds[j].revents, &ufds->revents))
1025 struct poll_list *pos = walk;
1033 static long do_restart_poll(struct restart_block *restart_block)
1035 struct pollfd __user *ufds = restart_block->poll.ufds;
1036 int nfds = restart_block->poll.nfds;
1037 struct timespec64 *to = NULL, end_time;
1040 if (restart_block->poll.has_timeout) {
1041 end_time.tv_sec = restart_block->poll.tv_sec;
1042 end_time.tv_nsec = restart_block->poll.tv_nsec;
1046 ret = do_sys_poll(ufds, nfds, to);
1048 if (ret == -EINTR) {
1049 restart_block->fn = do_restart_poll;
1050 ret = -ERESTART_RESTARTBLOCK;
1055 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1058 struct timespec64 end_time, *to = NULL;
1061 if (timeout_msecs >= 0) {
1063 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1064 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1067 ret = do_sys_poll(ufds, nfds, to);
1069 if (ret == -EINTR) {
1070 struct restart_block *restart_block;
1072 restart_block = ¤t->restart_block;
1073 restart_block->fn = do_restart_poll;
1074 restart_block->poll.ufds = ufds;
1075 restart_block->poll.nfds = nfds;
1077 if (timeout_msecs >= 0) {
1078 restart_block->poll.tv_sec = end_time.tv_sec;
1079 restart_block->poll.tv_nsec = end_time.tv_nsec;
1080 restart_block->poll.has_timeout = 1;
1082 restart_block->poll.has_timeout = 0;
1084 ret = -ERESTART_RESTARTBLOCK;
1089 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1090 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1093 sigset_t ksigmask, sigsaved;
1094 struct timespec64 ts, end_time, *to = NULL;
1098 if (get_timespec64(&ts, tsp))
1102 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1106 ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1110 ret = do_sys_poll(ufds, nfds, to);
1112 restore_user_sigmask(sigmask, &sigsaved);
1114 /* We can restart this syscall, usually */
1116 ret = -ERESTARTNOHAND;
1118 ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret);
1123 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1125 SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1126 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1129 sigset_t ksigmask, sigsaved;
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, &ksigmask, &sigsaved, sigsetsize);
1146 ret = do_sys_poll(ufds, nfds, to);
1148 restore_user_sigmask(sigmask, &sigsaved);
1150 /* We can restart this syscall, usually */
1152 ret = -ERESTARTNOHAND;
1154 ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1160 #ifdef CONFIG_COMPAT
1161 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1164 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1165 * 64-bit unsigned longs.
1168 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1169 unsigned long *fdset)
1172 return compat_get_bitmap(fdset, ufdset, nr);
1174 zero_fd_set(nr, fdset);
1180 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1181 unsigned long *fdset)
1185 return compat_put_bitmap(ufdset, fdset, nr);
1190 * This is a virtual copy of sys_select from fs/select.c and probably
1191 * should be compared to it from time to time
1195 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1196 * like to be certain this leads to no problems. So I return
1197 * EINTR just for safety.
1199 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1200 * I'm trying ERESTARTNOHAND which restart only when you want to.
1202 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1203 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1204 struct timespec64 *end_time)
1208 int size, max_fds, ret = -EINVAL;
1209 struct fdtable *fdt;
1210 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1215 /* max_fds can increase, so grab it once to avoid race */
1217 fdt = files_fdtable(current->files);
1218 max_fds = fdt->max_fds;
1224 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1225 * since we used fdset we need to allocate memory in units of
1228 size = FDS_BYTES(n);
1230 if (size > sizeof(stack_fds) / 6) {
1231 bits = kmalloc_array(6, size, GFP_KERNEL);
1236 fds.in = (unsigned long *) bits;
1237 fds.out = (unsigned long *) (bits + size);
1238 fds.ex = (unsigned long *) (bits + 2*size);
1239 fds.res_in = (unsigned long *) (bits + 3*size);
1240 fds.res_out = (unsigned long *) (bits + 4*size);
1241 fds.res_ex = (unsigned long *) (bits + 5*size);
1243 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1244 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1245 (ret = compat_get_fd_set(n, exp, fds.ex)))
1247 zero_fd_set(n, fds.res_in);
1248 zero_fd_set(n, fds.res_out);
1249 zero_fd_set(n, fds.res_ex);
1251 ret = do_select(n, &fds, end_time);
1256 ret = -ERESTARTNOHAND;
1257 if (signal_pending(current))
1262 if (compat_set_fd_set(n, inp, fds.res_in) ||
1263 compat_set_fd_set(n, outp, fds.res_out) ||
1264 compat_set_fd_set(n, exp, fds.res_ex))
1267 if (bits != stack_fds)
1273 static int do_compat_select(int n, compat_ulong_t __user *inp,
1274 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1275 struct old_timeval32 __user *tvp)
1277 struct timespec64 end_time, *to = NULL;
1278 struct old_timeval32 tv;
1282 if (copy_from_user(&tv, tvp, sizeof(tv)))
1286 if (poll_select_set_timeout(to,
1287 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1288 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1292 ret = compat_core_sys_select(n, inp, outp, exp, to);
1293 ret = poll_select_copy_remaining(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1298 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1299 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1300 struct old_timeval32 __user *, tvp)
1302 return do_compat_select(n, inp, outp, exp, tvp);
1305 struct compat_sel_arg_struct {
1313 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1315 struct compat_sel_arg_struct a;
1317 if (copy_from_user(&a, arg, sizeof(a)))
1319 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1320 compat_ptr(a.exp), compat_ptr(a.tvp));
1323 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1324 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1325 void __user *tsp, compat_sigset_t __user *sigmask,
1326 compat_size_t sigsetsize, enum poll_time_type type)
1328 sigset_t ksigmask, sigsaved;
1329 struct timespec64 ts, end_time, *to = NULL;
1334 case PT_OLD_TIMESPEC:
1335 if (get_old_timespec32(&ts, tsp))
1339 if (get_timespec64(&ts, tsp))
1347 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1351 ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1355 ret = compat_core_sys_select(n, inp, outp, exp, to);
1356 ret = poll_select_copy_remaining(&end_time, tsp, type, ret);
1358 restore_user_sigmask(sigmask, &sigsaved);
1363 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1364 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1365 struct __kernel_timespec __user *, tsp, void __user *, sig)
1367 compat_size_t sigsetsize = 0;
1368 compat_uptr_t up = 0;
1371 if (!access_ok(VERIFY_READ, sig,
1372 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1373 __get_user(up, (compat_uptr_t __user *)sig) ||
1374 __get_user(sigsetsize,
1375 (compat_size_t __user *)(sig+sizeof(up))))
1379 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1380 sigsetsize, PT_TIMESPEC);
1383 #if defined(CONFIG_COMPAT_32BIT_TIME)
1385 COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp,
1386 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1387 struct old_timespec32 __user *, tsp, void __user *, sig)
1389 compat_size_t sigsetsize = 0;
1390 compat_uptr_t up = 0;
1393 if (!access_ok(VERIFY_READ, sig,
1394 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1395 __get_user(up, (compat_uptr_t __user *)sig) ||
1396 __get_user(sigsetsize,
1397 (compat_size_t __user *)(sig+sizeof(up))))
1401 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1402 sigsetsize, PT_OLD_TIMESPEC);
1407 #if defined(CONFIG_COMPAT_32BIT_TIME)
1408 COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds,
1409 unsigned int, nfds, struct old_timespec32 __user *, tsp,
1410 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1412 sigset_t ksigmask, sigsaved;
1413 struct timespec64 ts, end_time, *to = NULL;
1417 if (get_old_timespec32(&ts, tsp))
1421 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1425 ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1429 ret = do_sys_poll(ufds, nfds, to);
1431 restore_user_sigmask(sigmask, &sigsaved);
1433 /* We can restart this syscall, usually */
1435 ret = -ERESTARTNOHAND;
1437 ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1443 /* New compat syscall for 64 bit time_t*/
1444 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1445 unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1446 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1448 sigset_t ksigmask, sigsaved;
1449 struct timespec64 ts, end_time, *to = NULL;
1453 if (get_timespec64(&ts, tsp))
1457 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1461 ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1465 ret = do_sys_poll(ufds, nfds, to);
1467 restore_user_sigmask(sigmask, &sigsaved);
1469 /* We can restart this syscall, usually */
1471 ret = -ERESTARTNOHAND;
1473 ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret);