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/syscalls.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
20 #include <linux/smp_lock.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 #define ROUND_UP(x,y) (((x)+(y)-1)/(y))
30 #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)
32 struct poll_table_page {
33 struct poll_table_page * next;
34 struct poll_table_entry * entry;
35 struct poll_table_entry entries[0];
38 #define POLL_TABLE_FULL(table) \
39 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
42 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
43 * I have rewritten this, taking some shortcuts: This code may not be easy to
44 * follow, but it should be free of race-conditions, and it's practical. If you
45 * understand what I'm doing here, then you understand how the linux
46 * sleep/wakeup mechanism works.
48 * Two very simple procedures, poll_wait() and poll_freewait() make all the
49 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
50 * as all select/poll functions have to call it to add an entry to the
53 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
56 void poll_initwait(struct poll_wqueues *pwq)
58 init_poll_funcptr(&pwq->pt, __pollwait);
61 pwq->inline_index = 0;
64 EXPORT_SYMBOL(poll_initwait);
66 static void free_poll_entry(struct poll_table_entry *entry)
68 remove_wait_queue(entry->wait_address,&entry->wait);
72 void poll_freewait(struct poll_wqueues *pwq)
74 struct poll_table_page * p = pwq->table;
76 for (i = 0; i < pwq->inline_index; i++)
77 free_poll_entry(pwq->inline_entries + i);
79 struct poll_table_entry * entry;
80 struct poll_table_page *old;
85 free_poll_entry(entry);
86 } while (entry > p->entries);
89 free_page((unsigned long) old);
93 EXPORT_SYMBOL(poll_freewait);
95 static struct poll_table_entry *poll_get_entry(poll_table *_p)
97 struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt);
98 struct poll_table_page *table = p->table;
100 if (p->inline_index < N_INLINE_POLL_ENTRIES)
101 return p->inline_entries + p->inline_index++;
103 if (!table || POLL_TABLE_FULL(table)) {
104 struct poll_table_page *new_table;
106 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
109 __set_current_state(TASK_RUNNING);
112 new_table->entry = new_table->entries;
113 new_table->next = table;
114 p->table = new_table;
118 return table->entry++;
121 /* Add a new entry */
122 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
125 struct poll_table_entry *entry = poll_get_entry(p);
130 entry->wait_address = wait_address;
131 init_waitqueue_entry(&entry->wait, current);
132 add_wait_queue(wait_address,&entry->wait);
135 #define FDS_IN(fds, n) (fds->in + n)
136 #define FDS_OUT(fds, n) (fds->out + n)
137 #define FDS_EX(fds, n) (fds->ex + n)
139 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
141 static int max_select_fd(unsigned long n, fd_set_bits *fds)
143 unsigned long *open_fds;
148 /* handle last in-complete long-word first */
149 set = ~(~0UL << (n & (__NFDBITS-1)));
151 fdt = files_fdtable(current->files);
152 open_fds = fdt->open_fds->fds_bits+n;
157 if (!(set & ~*open_fds))
168 if (set & ~*open_fds)
177 max += n * __NFDBITS;
183 #define BIT(i) (1UL << ((i)&(__NFDBITS-1)))
184 #define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS)
185 #define ISSET(i,m) (((i)&*(m)) != 0)
186 #define SET(i,m) (*(m) |= (i))
188 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
189 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
190 #define POLLEX_SET (POLLPRI)
192 int do_select(int n, fd_set_bits *fds, s64 *timeout)
194 struct poll_wqueues table;
199 retval = max_select_fd(n, fds);
206 poll_initwait(&table);
212 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
215 set_current_state(TASK_INTERRUPTIBLE);
217 inp = fds->in; outp = fds->out; exp = fds->ex;
218 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
220 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
221 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
222 unsigned long res_in = 0, res_out = 0, res_ex = 0;
223 const struct file_operations *f_op = NULL;
224 struct file *file = NULL;
226 in = *inp++; out = *outp++; ex = *exp++;
227 all_bits = in | out | ex;
233 for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
237 if (!(bit & all_bits))
239 file = fget_light(i, &fput_needed);
242 mask = DEFAULT_POLLMASK;
243 if (f_op && f_op->poll)
244 mask = (*f_op->poll)(file, retval ? NULL : wait);
245 fput_light(file, fput_needed);
246 if ((mask & POLLIN_SET) && (in & bit)) {
250 if ((mask & POLLOUT_SET) && (out & bit)) {
254 if ((mask & POLLEX_SET) && (ex & bit)) {
269 if (retval || !*timeout || signal_pending(current))
272 retval = table.error;
277 /* Wait indefinitely */
278 __timeout = MAX_SCHEDULE_TIMEOUT;
279 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT - 1)) {
280 /* Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in a loop */
281 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
282 *timeout -= __timeout;
284 __timeout = *timeout;
287 __timeout = schedule_timeout(__timeout);
289 *timeout += __timeout;
291 __set_current_state(TASK_RUNNING);
293 poll_freewait(&table);
299 * We can actually return ERESTARTSYS instead of EINTR, but I'd
300 * like to be certain this leads to no problems. So I return
301 * EINTR just for safety.
303 * Update: ERESTARTSYS breaks at least the xview clock binary, so
304 * I'm trying ERESTARTNOHAND which restart only when you want to.
306 #define MAX_SELECT_SECONDS \
307 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
309 static int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
310 fd_set __user *exp, s64 *timeout)
317 /* Allocate small arguments on the stack to save memory and be faster */
318 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
324 /* max_fdset can increase, so grab it once to avoid race */
326 fdt = files_fdtable(current->files);
327 max_fdset = fdt->max_fdset;
333 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
334 * since we used fdset we need to allocate memory in units of
339 if (size > sizeof(stack_fds) / 6) {
340 /* Not enough space in on-stack array; must use kmalloc */
342 bits = kmalloc(6 * size, GFP_KERNEL);
347 fds.out = bits + size;
348 fds.ex = bits + 2*size;
349 fds.res_in = bits + 3*size;
350 fds.res_out = bits + 4*size;
351 fds.res_ex = bits + 5*size;
353 if ((ret = get_fd_set(n, inp, fds.in)) ||
354 (ret = get_fd_set(n, outp, fds.out)) ||
355 (ret = get_fd_set(n, exp, fds.ex)))
357 zero_fd_set(n, fds.res_in);
358 zero_fd_set(n, fds.res_out);
359 zero_fd_set(n, fds.res_ex);
361 ret = do_select(n, &fds, timeout);
366 ret = -ERESTARTNOHAND;
367 if (signal_pending(current))
372 if (set_fd_set(n, inp, fds.res_in) ||
373 set_fd_set(n, outp, fds.res_out) ||
374 set_fd_set(n, exp, fds.res_ex))
378 if (bits != stack_fds)
384 asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
385 fd_set __user *exp, struct timeval __user *tvp)
392 if (copy_from_user(&tv, tvp, sizeof(tv)))
395 if (tv.tv_sec < 0 || tv.tv_usec < 0)
398 /* Cast to u64 to make GCC stop complaining */
399 if ((u64)tv.tv_sec >= (u64)MAX_INT64_SECONDS)
400 timeout = -1; /* infinite */
402 timeout = ROUND_UP(tv.tv_usec, USEC_PER_SEC/HZ);
403 timeout += tv.tv_sec * HZ;
407 ret = core_sys_select(n, inp, outp, exp, &timeout);
412 if (current->personality & STICKY_TIMEOUTS)
414 rtv.tv_usec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ));
415 rtv.tv_sec = timeout;
416 if (timeval_compare(&rtv, &tv) >= 0)
418 if (copy_to_user(tvp, &rtv, sizeof(rtv))) {
421 * If an application puts its timeval in read-only
422 * memory, we don't want the Linux-specific update to
423 * the timeval to cause a fault after the select has
424 * completed successfully. However, because we're not
425 * updating the timeval, we can't restart the system
428 if (ret == -ERESTARTNOHAND)
436 #ifdef TIF_RESTORE_SIGMASK
437 asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp,
438 fd_set __user *exp, struct timespec __user *tsp,
439 const sigset_t __user *sigmask, size_t sigsetsize)
441 s64 timeout = MAX_SCHEDULE_TIMEOUT;
442 sigset_t ksigmask, sigsaved;
447 if (copy_from_user(&ts, tsp, sizeof(ts)))
450 if (ts.tv_sec < 0 || ts.tv_nsec < 0)
453 /* Cast to u64 to make GCC stop complaining */
454 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
455 timeout = -1; /* infinite */
457 timeout = ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
458 timeout += ts.tv_sec * HZ;
463 /* XXX: Don't preclude handling different sized sigset_t's. */
464 if (sigsetsize != sizeof(sigset_t))
466 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
469 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
470 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
473 ret = core_sys_select(n, inp, outp, exp, &timeout);
478 if (current->personality & STICKY_TIMEOUTS)
480 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
482 rts.tv_sec = timeout;
483 if (timespec_compare(&rts, &ts) >= 0)
485 if (copy_to_user(tsp, &rts, sizeof(rts))) {
488 * If an application puts its timeval in read-only
489 * memory, we don't want the Linux-specific update to
490 * the timeval to cause a fault after the select has
491 * completed successfully. However, because we're not
492 * updating the timeval, we can't restart the system
495 if (ret == -ERESTARTNOHAND)
500 if (ret == -ERESTARTNOHAND) {
502 * Don't restore the signal mask yet. Let do_signal() deliver
503 * the signal on the way back to userspace, before the signal
507 memcpy(¤t->saved_sigmask, &sigsaved,
509 set_thread_flag(TIF_RESTORE_SIGMASK);
512 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
518 * Most architectures can't handle 7-argument syscalls. So we provide a
519 * 6-argument version where the sixth argument is a pointer to a structure
520 * which has a pointer to the sigset_t itself followed by a size_t containing
523 asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp,
524 fd_set __user *exp, struct timespec __user *tsp, void __user *sig)
526 size_t sigsetsize = 0;
527 sigset_t __user *up = NULL;
530 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
531 || __get_user(up, (sigset_t __user * __user *)sig)
532 || __get_user(sigsetsize,
533 (size_t __user *)(sig+sizeof(void *))))
537 return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize);
539 #endif /* TIF_RESTORE_SIGMASK */
542 struct poll_list *next;
544 struct pollfd entries[0];
547 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
550 * Fish for pollable events on the pollfd->fd file descriptor. We're only
551 * interested in events matching the pollfd->events mask, and the result
552 * matching that mask is both recorded in pollfd->revents and returned. The
553 * pwait poll_table will be used by the fd-provided poll handler for waiting,
556 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
567 file = fget_light(fd, &fput_needed);
570 mask = DEFAULT_POLLMASK;
571 if (file->f_op && file->f_op->poll)
572 mask = file->f_op->poll(file, pwait);
573 /* Mask out unneeded events. */
574 mask &= pollfd->events | POLLERR | POLLHUP;
575 fput_light(file, fput_needed);
578 pollfd->revents = mask;
583 static int do_poll(unsigned int nfds, struct poll_list *list,
584 struct poll_wqueues *wait, s64 *timeout)
587 poll_table* pt = &wait->pt;
589 /* Optimise the no-wait case */
594 struct poll_list *walk;
597 set_current_state(TASK_INTERRUPTIBLE);
598 for (walk = list; walk != NULL; walk = walk->next) {
599 struct pollfd * pfd, * pfd_end;
602 pfd_end = pfd + walk->len;
603 for (; pfd != pfd_end; pfd++) {
605 * Fish for events. If we found one, record it
606 * and kill the poll_table, so we don't
607 * needlessly register any other waiters after
608 * this. They'll get immediately deregistered
609 * when we break out and return.
611 if (do_pollfd(pfd, pt)) {
618 * All waiters have already been registered, so don't provide
619 * a poll_table to them on the next loop iteration.
622 if (count || !*timeout || signal_pending(current))
629 /* Wait indefinitely */
630 __timeout = MAX_SCHEDULE_TIMEOUT;
631 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT-1)) {
633 * Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in
636 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
637 *timeout -= __timeout;
639 __timeout = *timeout;
643 __timeout = schedule_timeout(__timeout);
645 *timeout += __timeout;
647 __set_current_state(TASK_RUNNING);
651 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
652 sizeof(struct pollfd))
654 int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, s64 *timeout)
656 struct poll_wqueues table;
659 struct poll_list *head;
660 struct poll_list *walk;
661 /* Allocate small arguments on the stack to save memory and be
662 faster - use long to make sure the buffer is aligned properly
663 on 64 bit archs to avoid unaligned access */
664 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
665 struct poll_list *stack_pp = NULL;
667 /* Do a sanity check on nfds ... */
668 if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
671 poll_initwait(&table);
678 struct poll_list *pp;
680 if (stack_pp == NULL)
683 num = POLLFD_PER_PAGE;
686 size = sizeof(struct poll_list) + sizeof(struct pollfd)*num;
688 stack_pp = pp = (struct poll_list *)stack_pps;
690 pp = kmalloc(size, GFP_KERNEL);
702 if (copy_from_user(pp->entries, ufds + nfds-i,
703 sizeof(struct pollfd)*num)) {
710 fdcount = do_poll(nfds, head, &table, timeout);
712 /* OK, now copy the revents fields back to user space. */
715 while(walk != NULL) {
716 struct pollfd *fds = walk->entries;
719 for (j=0; j < walk->len; j++, ufds++) {
720 if(__put_user(fds[j].revents, &ufds->revents))
726 if (!fdcount && signal_pending(current))
731 struct poll_list *pp = walk->next;
732 if (walk != stack_pp)
736 poll_freewait(&table);
740 asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds,
745 if (timeout_msecs > 0) {
747 /* We can only overflow if HZ > 1000 */
748 if (timeout_msecs / 1000 > (s64)0x7fffffffffffffffULL / (s64)HZ)
749 timeout_jiffies = -1;
752 timeout_jiffies = msecs_to_jiffies(timeout_msecs);
754 /* Infinite (< 0) or no (0) timeout */
755 timeout_jiffies = timeout_msecs;
758 return do_sys_poll(ufds, nfds, &timeout_jiffies);
761 #ifdef TIF_RESTORE_SIGMASK
762 asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds,
763 struct timespec __user *tsp, const sigset_t __user *sigmask,
766 sigset_t ksigmask, sigsaved;
772 if (copy_from_user(&ts, tsp, sizeof(ts)))
775 /* Cast to u64 to make GCC stop complaining */
776 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
777 timeout = -1; /* infinite */
779 timeout = ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
780 timeout += ts.tv_sec * HZ;
785 /* XXX: Don't preclude handling different sized sigset_t's. */
786 if (sigsetsize != sizeof(sigset_t))
788 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
791 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
792 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
795 ret = do_sys_poll(ufds, nfds, &timeout);
797 /* We can restart this syscall, usually */
800 * Don't restore the signal mask yet. Let do_signal() deliver
801 * the signal on the way back to userspace, before the signal
805 memcpy(¤t->saved_sigmask, &sigsaved,
807 set_thread_flag(TIF_RESTORE_SIGMASK);
809 ret = -ERESTARTNOHAND;
811 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
813 if (tsp && timeout >= 0) {
816 if (current->personality & STICKY_TIMEOUTS)
818 /* Yes, we know it's actually an s64, but it's also positive. */
819 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
821 rts.tv_sec = timeout;
822 if (timespec_compare(&rts, &ts) >= 0)
824 if (copy_to_user(tsp, &rts, sizeof(rts))) {
827 * If an application puts its timeval in read-only
828 * memory, we don't want the Linux-specific update to
829 * the timeval to cause a fault after the select has
830 * completed successfully. However, because we're not
831 * updating the timeval, we can't restart the system
834 if (ret == -ERESTARTNOHAND && timeout >= 0)
841 #endif /* TIF_RESTORE_SIGMASK */