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 noinline_for_stack 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 (can_do_masked_user_access())
784 from = masked_user_access_begin(from);
785 else if (!user_read_access_begin(from, sizeof(*from)))
787 unsafe_get_user(to->p, &from->p, Efault);
788 unsafe_get_user(to->size, &from->size, Efault);
789 user_read_access_end();
797 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
798 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
801 struct sigset_argpack x = {NULL, 0};
803 if (get_sigset_argpack(&x, sig))
806 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
809 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
811 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
812 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
815 struct sigset_argpack x = {NULL, 0};
817 if (get_sigset_argpack(&x, sig))
820 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
825 #ifdef __ARCH_WANT_SYS_OLD_SELECT
826 struct sel_arg_struct {
828 fd_set __user *inp, *outp, *exp;
829 struct __kernel_old_timeval __user *tvp;
832 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
834 struct sel_arg_struct a;
836 if (copy_from_user(&a, arg, sizeof(a)))
838 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
843 struct poll_list *next;
845 struct pollfd entries[];
848 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
851 * Fish for pollable events on the pollfd->fd file descriptor. We're only
852 * interested in events matching the pollfd->events mask, and the result
853 * matching that mask is both recorded in pollfd->revents and returned. The
854 * pwait poll_table will be used by the fd-provided poll handler for waiting,
855 * if pwait->_qproc is non-NULL.
857 static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
862 __poll_t mask = 0, filter;
872 /* userland u16 ->events contains POLL... bitmap */
873 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
874 pwait->_key = filter | busy_flag;
875 mask = vfs_poll(f.file, pwait);
876 if (mask & busy_flag)
877 *can_busy_poll = true;
878 mask &= filter; /* Mask out unneeded events. */
882 /* ... and so does ->revents */
883 pollfd->revents = mangle_poll(mask);
887 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
888 struct timespec64 *end_time)
890 poll_table* pt = &wait->pt;
891 ktime_t expire, *to = NULL;
892 int timed_out = 0, count = 0;
894 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
895 unsigned long busy_start = 0;
897 /* Optimise the no-wait case */
898 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
903 if (end_time && !timed_out)
904 slack = select_estimate_accuracy(end_time);
907 struct poll_list *walk;
908 bool can_busy_loop = false;
910 for (walk = list; walk != NULL; walk = walk->next) {
911 struct pollfd * pfd, * pfd_end;
914 pfd_end = pfd + walk->len;
915 for (; pfd != pfd_end; pfd++) {
917 * Fish for events. If we found one, record it
918 * and kill poll_table->_qproc, so we don't
919 * needlessly register any other waiters after
920 * this. They'll get immediately deregistered
921 * when we break out and return.
923 if (do_pollfd(pfd, pt, &can_busy_loop,
927 /* found something, stop busy polling */
929 can_busy_loop = false;
934 * All waiters have already been registered, so don't provide
935 * a poll_table->_qproc to them on the next loop iteration.
940 if (signal_pending(current))
941 count = -ERESTARTNOHAND;
943 if (count || timed_out)
946 /* only if found POLL_BUSY_LOOP sockets && not out of time */
947 if (can_busy_loop && !need_resched()) {
949 busy_start = busy_loop_current_time();
952 if (!busy_loop_timeout(busy_start))
958 * If this is the first loop and we have a timeout
959 * given, then we convert to ktime_t and set the to
960 * pointer to the expiry value.
962 if (end_time && !to) {
963 expire = timespec64_to_ktime(*end_time);
967 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
973 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
974 sizeof(struct pollfd))
976 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
977 struct timespec64 *end_time)
979 struct poll_wqueues table;
980 int err = -EFAULT, fdcount;
981 /* Allocate small arguments on the stack to save memory and be
982 faster - use long to make sure the buffer is aligned properly
983 on 64 bit archs to avoid unaligned access */
984 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
985 struct poll_list *const head = (struct poll_list *)stack_pps;
986 struct poll_list *walk = head;
987 unsigned int todo = nfds;
990 if (nfds > rlimit(RLIMIT_NOFILE))
993 len = min_t(unsigned int, nfds, N_STACK_PPS);
1000 if (copy_from_user(walk->entries, ufds + nfds-todo,
1001 sizeof(struct pollfd) * walk->len))
1004 if (walk->len >= todo)
1008 len = min(todo, POLLFD_PER_PAGE);
1009 walk = walk->next = kmalloc(struct_size(walk, entries, len),
1017 poll_initwait(&table);
1018 fdcount = do_poll(head, &table, end_time);
1019 poll_freewait(&table);
1021 if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1024 for (walk = head; walk; walk = walk->next) {
1025 struct pollfd *fds = walk->entries;
1028 for (j = walk->len; j; fds++, ufds++, j--)
1029 unsafe_put_user(fds->revents, &ufds->revents, Efault);
1031 user_write_access_end();
1037 struct poll_list *pos = walk;
1045 user_write_access_end();
1050 static long do_restart_poll(struct restart_block *restart_block)
1052 struct pollfd __user *ufds = restart_block->poll.ufds;
1053 int nfds = restart_block->poll.nfds;
1054 struct timespec64 *to = NULL, end_time;
1057 if (restart_block->poll.has_timeout) {
1058 end_time.tv_sec = restart_block->poll.tv_sec;
1059 end_time.tv_nsec = restart_block->poll.tv_nsec;
1063 ret = do_sys_poll(ufds, nfds, to);
1065 if (ret == -ERESTARTNOHAND)
1066 ret = set_restart_fn(restart_block, do_restart_poll);
1071 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1074 struct timespec64 end_time, *to = NULL;
1077 if (timeout_msecs >= 0) {
1079 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1080 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1083 ret = do_sys_poll(ufds, nfds, to);
1085 if (ret == -ERESTARTNOHAND) {
1086 struct restart_block *restart_block;
1088 restart_block = ¤t->restart_block;
1089 restart_block->poll.ufds = ufds;
1090 restart_block->poll.nfds = nfds;
1092 if (timeout_msecs >= 0) {
1093 restart_block->poll.tv_sec = end_time.tv_sec;
1094 restart_block->poll.tv_nsec = end_time.tv_nsec;
1095 restart_block->poll.has_timeout = 1;
1097 restart_block->poll.has_timeout = 0;
1099 ret = set_restart_fn(restart_block, do_restart_poll);
1104 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1105 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1108 struct timespec64 ts, end_time, *to = NULL;
1112 if (get_timespec64(&ts, tsp))
1116 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1120 ret = set_user_sigmask(sigmask, sigsetsize);
1124 ret = do_sys_poll(ufds, nfds, to);
1125 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1128 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1130 SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1131 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1134 struct timespec64 ts, end_time, *to = NULL;
1138 if (get_old_timespec32(&ts, tsp))
1142 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1146 ret = set_user_sigmask(sigmask, sigsetsize);
1150 ret = do_sys_poll(ufds, nfds, to);
1151 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1155 #ifdef CONFIG_COMPAT
1156 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1159 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1160 * 64-bit unsigned longs.
1163 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1164 unsigned long *fdset)
1167 return compat_get_bitmap(fdset, ufdset, nr);
1169 zero_fd_set(nr, fdset);
1175 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1176 unsigned long *fdset)
1180 return compat_put_bitmap(ufdset, fdset, nr);
1185 * This is a virtual copy of sys_select from fs/select.c and probably
1186 * should be compared to it from time to time
1190 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1191 * like to be certain this leads to no problems. So I return
1192 * EINTR just for safety.
1194 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1195 * I'm trying ERESTARTNOHAND which restart only when you want to.
1197 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1198 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1199 struct timespec64 *end_time)
1203 int size, max_fds, ret = -EINVAL;
1204 struct fdtable *fdt;
1205 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1210 /* max_fds can increase, so grab it once to avoid race */
1212 fdt = files_fdtable(current->files);
1213 max_fds = fdt->max_fds;
1219 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1220 * since we used fdset we need to allocate memory in units of
1223 size = FDS_BYTES(n);
1225 if (size > sizeof(stack_fds) / 6) {
1226 bits = kmalloc_array(6, size, GFP_KERNEL);
1231 fds.in = (unsigned long *) bits;
1232 fds.out = (unsigned long *) (bits + size);
1233 fds.ex = (unsigned long *) (bits + 2*size);
1234 fds.res_in = (unsigned long *) (bits + 3*size);
1235 fds.res_out = (unsigned long *) (bits + 4*size);
1236 fds.res_ex = (unsigned long *) (bits + 5*size);
1238 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1239 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1240 (ret = compat_get_fd_set(n, exp, fds.ex)))
1242 zero_fd_set(n, fds.res_in);
1243 zero_fd_set(n, fds.res_out);
1244 zero_fd_set(n, fds.res_ex);
1246 ret = do_select(n, &fds, end_time);
1251 ret = -ERESTARTNOHAND;
1252 if (signal_pending(current))
1257 if (compat_set_fd_set(n, inp, fds.res_in) ||
1258 compat_set_fd_set(n, outp, fds.res_out) ||
1259 compat_set_fd_set(n, exp, fds.res_ex))
1262 if (bits != stack_fds)
1268 static int do_compat_select(int n, compat_ulong_t __user *inp,
1269 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1270 struct old_timeval32 __user *tvp)
1272 struct timespec64 end_time, *to = NULL;
1273 struct old_timeval32 tv;
1277 if (copy_from_user(&tv, tvp, sizeof(tv)))
1281 if (poll_select_set_timeout(to,
1282 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1283 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1287 ret = compat_core_sys_select(n, inp, outp, exp, to);
1288 return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1291 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1292 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1293 struct old_timeval32 __user *, tvp)
1295 return do_compat_select(n, inp, outp, exp, tvp);
1298 struct compat_sel_arg_struct {
1306 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1308 struct compat_sel_arg_struct a;
1310 if (copy_from_user(&a, arg, sizeof(a)))
1312 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1313 compat_ptr(a.exp), compat_ptr(a.tvp));
1316 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1317 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1318 void __user *tsp, compat_sigset_t __user *sigmask,
1319 compat_size_t sigsetsize, enum poll_time_type type)
1321 struct timespec64 ts, end_time, *to = NULL;
1326 case PT_OLD_TIMESPEC:
1327 if (get_old_timespec32(&ts, tsp))
1331 if (get_timespec64(&ts, tsp))
1339 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1343 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1347 ret = compat_core_sys_select(n, inp, outp, exp, to);
1348 return poll_select_finish(&end_time, tsp, type, ret);
1351 struct compat_sigset_argpack {
1355 static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1356 struct compat_sigset_argpack __user *from)
1359 if (!user_read_access_begin(from, sizeof(*from)))
1361 unsafe_get_user(to->p, &from->p, Efault);
1362 unsafe_get_user(to->size, &from->size, Efault);
1363 user_read_access_end();
1371 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1372 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1373 struct __kernel_timespec __user *, tsp, void __user *, sig)
1375 struct compat_sigset_argpack x = {0, 0};
1377 if (get_compat_sigset_argpack(&x, sig))
1380 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1381 x.size, PT_TIMESPEC);
1384 #if defined(CONFIG_COMPAT_32BIT_TIME)
1386 COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1387 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1388 struct old_timespec32 __user *, tsp, void __user *, sig)
1390 struct compat_sigset_argpack x = {0, 0};
1392 if (get_compat_sigset_argpack(&x, sig))
1395 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1396 x.size, PT_OLD_TIMESPEC);
1401 #if defined(CONFIG_COMPAT_32BIT_TIME)
1402 COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1403 unsigned int, nfds, struct old_timespec32 __user *, tsp,
1404 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1406 struct timespec64 ts, end_time, *to = NULL;
1410 if (get_old_timespec32(&ts, tsp))
1414 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1418 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1422 ret = do_sys_poll(ufds, nfds, to);
1423 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1427 /* New compat syscall for 64 bit time_t*/
1428 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1429 unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1430 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1432 struct timespec64 ts, end_time, *to = NULL;
1436 if (get_timespec64(&ts, tsp))
1440 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1444 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1448 ret = do_sys_poll(ufds, nfds, to);
1449 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);