2 * This file contains the procedures for the handling of select and poll
4 * Created for Linux based loosely upon Mathius Lattner's minix
5 * patches by Peter MacDonald. Heavily edited by Linus.
8 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
9 * flag set in its personality we do *not* modify the given timeout
10 * parameter to reflect time remaining.
13 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
14 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
17 #include <linux/kernel.h>
18 #include <linux/syscalls.h>
19 #include <linux/module.h>
20 #include <linux/slab.h>
21 #include <linux/poll.h>
22 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
23 #include <linux/file.h>
25 #include <linux/rcupdate.h>
27 #include <asm/uaccess.h>
29 struct poll_table_page {
30 struct poll_table_page * next;
31 struct poll_table_entry * entry;
32 struct poll_table_entry entries[0];
35 #define POLL_TABLE_FULL(table) \
36 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
39 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
40 * I have rewritten this, taking some shortcuts: This code may not be easy to
41 * follow, but it should be free of race-conditions, and it's practical. If you
42 * understand what I'm doing here, then you understand how the linux
43 * sleep/wakeup mechanism works.
45 * Two very simple procedures, poll_wait() and poll_freewait() make all the
46 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
47 * as all select/poll functions have to call it to add an entry to the
50 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
53 void poll_initwait(struct poll_wqueues *pwq)
55 init_poll_funcptr(&pwq->pt, __pollwait);
58 pwq->inline_index = 0;
61 EXPORT_SYMBOL(poll_initwait);
63 static void free_poll_entry(struct poll_table_entry *entry)
65 remove_wait_queue(entry->wait_address, &entry->wait);
69 void poll_freewait(struct poll_wqueues *pwq)
71 struct poll_table_page * p = pwq->table;
73 for (i = 0; i < pwq->inline_index; i++)
74 free_poll_entry(pwq->inline_entries + i);
76 struct poll_table_entry * entry;
77 struct poll_table_page *old;
82 free_poll_entry(entry);
83 } while (entry > p->entries);
86 free_page((unsigned long) old);
90 EXPORT_SYMBOL(poll_freewait);
92 static struct poll_table_entry *poll_get_entry(poll_table *_p)
94 struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt);
95 struct poll_table_page *table = p->table;
97 if (p->inline_index < N_INLINE_POLL_ENTRIES)
98 return p->inline_entries + p->inline_index++;
100 if (!table || POLL_TABLE_FULL(table)) {
101 struct poll_table_page *new_table;
103 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
106 __set_current_state(TASK_RUNNING);
109 new_table->entry = new_table->entries;
110 new_table->next = table;
111 p->table = new_table;
115 return table->entry++;
118 /* Add a new entry */
119 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
122 struct poll_table_entry *entry = poll_get_entry(p);
127 entry->wait_address = wait_address;
128 init_waitqueue_entry(&entry->wait, current);
129 add_wait_queue(wait_address, &entry->wait);
132 #define FDS_IN(fds, n) (fds->in + n)
133 #define FDS_OUT(fds, n) (fds->out + n)
134 #define FDS_EX(fds, n) (fds->ex + n)
136 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
138 static int max_select_fd(unsigned long n, fd_set_bits *fds)
140 unsigned long *open_fds;
145 /* handle last in-complete long-word first */
146 set = ~(~0UL << (n & (__NFDBITS-1)));
148 fdt = files_fdtable(current->files);
149 open_fds = fdt->open_fds->fds_bits+n;
154 if (!(set & ~*open_fds))
165 if (set & ~*open_fds)
174 max += n * __NFDBITS;
180 #define BIT(i) (1UL << ((i)&(__NFDBITS-1)))
181 #define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS)
182 #define ISSET(i,m) (((i)&*(m)) != 0)
183 #define SET(i,m) (*(m) |= (i))
185 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
186 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
187 #define POLLEX_SET (POLLPRI)
189 int do_select(int n, fd_set_bits *fds, s64 *timeout)
191 struct poll_wqueues table;
196 retval = max_select_fd(n, fds);
203 poll_initwait(&table);
209 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
212 set_current_state(TASK_INTERRUPTIBLE);
214 inp = fds->in; outp = fds->out; exp = fds->ex;
215 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
217 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
218 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
219 unsigned long res_in = 0, res_out = 0, res_ex = 0;
220 const struct file_operations *f_op = NULL;
221 struct file *file = NULL;
223 in = *inp++; out = *outp++; ex = *exp++;
224 all_bits = in | out | ex;
230 for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
234 if (!(bit & all_bits))
236 file = fget_light(i, &fput_needed);
239 mask = DEFAULT_POLLMASK;
240 if (f_op && f_op->poll)
241 mask = (*f_op->poll)(file, retval ? NULL : wait);
242 fput_light(file, fput_needed);
243 if ((mask & POLLIN_SET) && (in & bit)) {
247 if ((mask & POLLOUT_SET) && (out & bit)) {
251 if ((mask & POLLEX_SET) && (ex & bit)) {
266 if (retval || !*timeout || signal_pending(current))
269 retval = table.error;
274 /* Wait indefinitely */
275 __timeout = MAX_SCHEDULE_TIMEOUT;
276 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT - 1)) {
277 /* Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in a loop */
278 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
279 *timeout -= __timeout;
281 __timeout = *timeout;
284 __timeout = schedule_timeout(__timeout);
286 *timeout += __timeout;
288 __set_current_state(TASK_RUNNING);
290 poll_freewait(&table);
296 * We can actually return ERESTARTSYS instead of EINTR, but I'd
297 * like to be certain this leads to no problems. So I return
298 * EINTR just for safety.
300 * Update: ERESTARTSYS breaks at least the xview clock binary, so
301 * I'm trying ERESTARTNOHAND which restart only when you want to.
303 #define MAX_SELECT_SECONDS \
304 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
306 static int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
307 fd_set __user *exp, s64 *timeout)
314 /* Allocate small arguments on the stack to save memory and be faster */
315 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
321 /* max_fds can increase, so grab it once to avoid race */
323 fdt = files_fdtable(current->files);
324 max_fds = fdt->max_fds;
330 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
331 * since we used fdset we need to allocate memory in units of
336 if (size > sizeof(stack_fds) / 6) {
337 /* Not enough space in on-stack array; must use kmalloc */
339 bits = kmalloc(6 * size, GFP_KERNEL);
344 fds.out = bits + size;
345 fds.ex = bits + 2*size;
346 fds.res_in = bits + 3*size;
347 fds.res_out = bits + 4*size;
348 fds.res_ex = bits + 5*size;
350 if ((ret = get_fd_set(n, inp, fds.in)) ||
351 (ret = get_fd_set(n, outp, fds.out)) ||
352 (ret = get_fd_set(n, exp, fds.ex)))
354 zero_fd_set(n, fds.res_in);
355 zero_fd_set(n, fds.res_out);
356 zero_fd_set(n, fds.res_ex);
358 ret = do_select(n, &fds, timeout);
363 ret = -ERESTARTNOHAND;
364 if (signal_pending(current))
369 if (set_fd_set(n, inp, fds.res_in) ||
370 set_fd_set(n, outp, fds.res_out) ||
371 set_fd_set(n, exp, fds.res_ex))
375 if (bits != stack_fds)
381 asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
382 fd_set __user *exp, struct timeval __user *tvp)
389 if (copy_from_user(&tv, tvp, sizeof(tv)))
392 if (tv.tv_sec < 0 || tv.tv_usec < 0)
395 /* Cast to u64 to make GCC stop complaining */
396 if ((u64)tv.tv_sec >= (u64)MAX_INT64_SECONDS)
397 timeout = -1; /* infinite */
399 timeout = DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC/HZ);
400 timeout += tv.tv_sec * HZ;
404 ret = core_sys_select(n, inp, outp, exp, &timeout);
409 if (current->personality & STICKY_TIMEOUTS)
411 rtv.tv_usec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ));
412 rtv.tv_sec = timeout;
413 if (timeval_compare(&rtv, &tv) >= 0)
415 if (copy_to_user(tvp, &rtv, sizeof(rtv))) {
418 * If an application puts its timeval in read-only
419 * memory, we don't want the Linux-specific update to
420 * the timeval to cause a fault after the select has
421 * completed successfully. However, because we're not
422 * updating the timeval, we can't restart the system
425 if (ret == -ERESTARTNOHAND)
433 #ifdef TIF_RESTORE_SIGMASK
434 asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp,
435 fd_set __user *exp, struct timespec __user *tsp,
436 const sigset_t __user *sigmask, size_t sigsetsize)
438 s64 timeout = MAX_SCHEDULE_TIMEOUT;
439 sigset_t ksigmask, sigsaved;
444 if (copy_from_user(&ts, tsp, sizeof(ts)))
447 if (ts.tv_sec < 0 || ts.tv_nsec < 0)
450 /* Cast to u64 to make GCC stop complaining */
451 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
452 timeout = -1; /* infinite */
454 timeout = DIV_ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
455 timeout += ts.tv_sec * HZ;
460 /* XXX: Don't preclude handling different sized sigset_t's. */
461 if (sigsetsize != sizeof(sigset_t))
463 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
466 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
467 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
470 ret = core_sys_select(n, inp, outp, exp, &timeout);
475 if (current->personality & STICKY_TIMEOUTS)
477 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
479 rts.tv_sec = timeout;
480 if (timespec_compare(&rts, &ts) >= 0)
482 if (copy_to_user(tsp, &rts, sizeof(rts))) {
485 * If an application puts its timeval in read-only
486 * memory, we don't want the Linux-specific update to
487 * the timeval to cause a fault after the select has
488 * completed successfully. However, because we're not
489 * updating the timeval, we can't restart the system
492 if (ret == -ERESTARTNOHAND)
497 if (ret == -ERESTARTNOHAND) {
499 * Don't restore the signal mask yet. Let do_signal() deliver
500 * the signal on the way back to userspace, before the signal
504 memcpy(¤t->saved_sigmask, &sigsaved,
506 set_thread_flag(TIF_RESTORE_SIGMASK);
509 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
515 * Most architectures can't handle 7-argument syscalls. So we provide a
516 * 6-argument version where the sixth argument is a pointer to a structure
517 * which has a pointer to the sigset_t itself followed by a size_t containing
520 asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp,
521 fd_set __user *exp, struct timespec __user *tsp, void __user *sig)
523 size_t sigsetsize = 0;
524 sigset_t __user *up = NULL;
527 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
528 || __get_user(up, (sigset_t __user * __user *)sig)
529 || __get_user(sigsetsize,
530 (size_t __user *)(sig+sizeof(void *))))
534 return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize);
536 #endif /* TIF_RESTORE_SIGMASK */
539 struct poll_list *next;
541 struct pollfd entries[0];
544 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
547 * Fish for pollable events on the pollfd->fd file descriptor. We're only
548 * interested in events matching the pollfd->events mask, and the result
549 * matching that mask is both recorded in pollfd->revents and returned. The
550 * pwait poll_table will be used by the fd-provided poll handler for waiting,
553 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
564 file = fget_light(fd, &fput_needed);
567 mask = DEFAULT_POLLMASK;
568 if (file->f_op && file->f_op->poll)
569 mask = file->f_op->poll(file, pwait);
570 /* Mask out unneeded events. */
571 mask &= pollfd->events | POLLERR | POLLHUP;
572 fput_light(file, fput_needed);
575 pollfd->revents = mask;
580 static int do_poll(unsigned int nfds, struct poll_list *list,
581 struct poll_wqueues *wait, s64 *timeout)
584 poll_table* pt = &wait->pt;
586 /* Optimise the no-wait case */
591 struct poll_list *walk;
594 set_current_state(TASK_INTERRUPTIBLE);
595 for (walk = list; walk != NULL; walk = walk->next) {
596 struct pollfd * pfd, * pfd_end;
599 pfd_end = pfd + walk->len;
600 for (; pfd != pfd_end; pfd++) {
602 * Fish for events. If we found one, record it
603 * and kill the poll_table, so we don't
604 * needlessly register any other waiters after
605 * this. They'll get immediately deregistered
606 * when we break out and return.
608 if (do_pollfd(pfd, pt)) {
615 * All waiters have already been registered, so don't provide
616 * a poll_table to them on the next loop iteration.
619 if (count || !*timeout || signal_pending(current))
626 /* Wait indefinitely */
627 __timeout = MAX_SCHEDULE_TIMEOUT;
628 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT-1)) {
630 * Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in
633 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
634 *timeout -= __timeout;
636 __timeout = *timeout;
640 __timeout = schedule_timeout(__timeout);
642 *timeout += __timeout;
644 __set_current_state(TASK_RUNNING);
648 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
649 sizeof(struct pollfd))
651 int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, s64 *timeout)
653 struct poll_wqueues table;
656 struct poll_list *head;
657 struct poll_list *walk;
658 /* Allocate small arguments on the stack to save memory and be
659 faster - use long to make sure the buffer is aligned properly
660 on 64 bit archs to avoid unaligned access */
661 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
662 struct poll_list *stack_pp = NULL;
664 /* Do a sanity check on nfds ... */
665 if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
668 poll_initwait(&table);
675 struct poll_list *pp;
677 if (stack_pp == NULL)
680 num = POLLFD_PER_PAGE;
683 size = sizeof(struct poll_list) + sizeof(struct pollfd)*num;
685 stack_pp = pp = (struct poll_list *)stack_pps;
687 pp = kmalloc(size, GFP_KERNEL);
699 if (copy_from_user(pp->entries, ufds + nfds-i,
700 sizeof(struct pollfd)*num)) {
707 fdcount = do_poll(nfds, head, &table, timeout);
709 /* OK, now copy the revents fields back to user space. */
712 while(walk != NULL) {
713 struct pollfd *fds = walk->entries;
716 for (j=0; j < walk->len; j++, ufds++) {
717 if(__put_user(fds[j].revents, &ufds->revents))
723 if (!fdcount && signal_pending(current))
728 struct poll_list *pp = walk->next;
729 if (walk != stack_pp)
733 poll_freewait(&table);
737 asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds,
742 if (timeout_msecs > 0) {
744 /* We can only overflow if HZ > 1000 */
745 if (timeout_msecs / 1000 > (s64)0x7fffffffffffffffULL / (s64)HZ)
746 timeout_jiffies = -1;
749 timeout_jiffies = msecs_to_jiffies(timeout_msecs);
751 /* Infinite (< 0) or no (0) timeout */
752 timeout_jiffies = timeout_msecs;
755 return do_sys_poll(ufds, nfds, &timeout_jiffies);
758 #ifdef TIF_RESTORE_SIGMASK
759 asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds,
760 struct timespec __user *tsp, const sigset_t __user *sigmask,
763 sigset_t ksigmask, sigsaved;
769 if (copy_from_user(&ts, tsp, sizeof(ts)))
772 /* Cast to u64 to make GCC stop complaining */
773 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
774 timeout = -1; /* infinite */
776 timeout = DIV_ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
777 timeout += ts.tv_sec * HZ;
782 /* XXX: Don't preclude handling different sized sigset_t's. */
783 if (sigsetsize != sizeof(sigset_t))
785 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
788 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
789 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
792 ret = do_sys_poll(ufds, nfds, &timeout);
794 /* We can restart this syscall, usually */
797 * Don't restore the signal mask yet. Let do_signal() deliver
798 * the signal on the way back to userspace, before the signal
802 memcpy(¤t->saved_sigmask, &sigsaved,
804 set_thread_flag(TIF_RESTORE_SIGMASK);
806 ret = -ERESTARTNOHAND;
808 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
810 if (tsp && timeout >= 0) {
813 if (current->personality & STICKY_TIMEOUTS)
815 /* Yes, we know it's actually an s64, but it's also positive. */
816 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
818 rts.tv_sec = timeout;
819 if (timespec_compare(&rts, &ts) >= 0)
821 if (copy_to_user(tsp, &rts, sizeof(rts))) {
824 * If an application puts its timeval in read-only
825 * memory, we don't want the Linux-specific update to
826 * the timeval to cause a fault after the select has
827 * completed successfully. However, because we're not
828 * updating the timeval, we can't restart the system
831 if (ret == -ERESTARTNOHAND && timeout >= 0)
838 #endif /* TIF_RESTORE_SIGMASK */