Loading...
1/*
2 * This file contains the procedures for the handling of select and poll
3 *
4 * Created for Linux based loosely upon Mathius Lattner's minix
5 * patches by Peter MacDonald. Heavily edited by Linus.
6 *
7 * 4 February 1994
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.
11 *
12 * 24 January 2000
13 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
14 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
15 */
16
17#include <linux/kernel.h>
18#include <linux/sched.h>
19#include <linux/syscalls.h>
20#include <linux/export.h>
21#include <linux/slab.h>
22#include <linux/poll.h>
23#include <linux/personality.h> /* for STICKY_TIMEOUTS */
24#include <linux/file.h>
25#include <linux/fdtable.h>
26#include <linux/fs.h>
27#include <linux/rcupdate.h>
28#include <linux/hrtimer.h>
29#include <linux/sched/rt.h>
30#include <linux/freezer.h>
31#include <net/busy_poll.h>
32#include <linux/vmalloc.h>
33
34#include <linux/uaccess.h>
35
36
37/*
38 * Estimate expected accuracy in ns from a timeval.
39 *
40 * After quite a bit of churning around, we've settled on
41 * a simple thing of taking 0.1% of the timeout as the
42 * slack, with a cap of 100 msec.
43 * "nice" tasks get a 0.5% slack instead.
44 *
45 * Consider this comment an open invitation to come up with even
46 * better solutions..
47 */
48
49#define MAX_SLACK (100 * NSEC_PER_MSEC)
50
51static long __estimate_accuracy(struct timespec64 *tv)
52{
53 long slack;
54 int divfactor = 1000;
55
56 if (tv->tv_sec < 0)
57 return 0;
58
59 if (task_nice(current) > 0)
60 divfactor = divfactor / 5;
61
62 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
63 return MAX_SLACK;
64
65 slack = tv->tv_nsec / divfactor;
66 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
67
68 if (slack > MAX_SLACK)
69 return MAX_SLACK;
70
71 return slack;
72}
73
74u64 select_estimate_accuracy(struct timespec64 *tv)
75{
76 u64 ret;
77 struct timespec64 now;
78
79 /*
80 * Realtime tasks get a slack of 0 for obvious reasons.
81 */
82
83 if (rt_task(current))
84 return 0;
85
86 ktime_get_ts64(&now);
87 now = timespec64_sub(*tv, now);
88 ret = __estimate_accuracy(&now);
89 if (ret < current->timer_slack_ns)
90 return current->timer_slack_ns;
91 return ret;
92}
93
94
95
96struct poll_table_page {
97 struct poll_table_page * next;
98 struct poll_table_entry * entry;
99 struct poll_table_entry entries[0];
100};
101
102#define POLL_TABLE_FULL(table) \
103 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
104
105/*
106 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
107 * I have rewritten this, taking some shortcuts: This code may not be easy to
108 * follow, but it should be free of race-conditions, and it's practical. If you
109 * understand what I'm doing here, then you understand how the linux
110 * sleep/wakeup mechanism works.
111 *
112 * Two very simple procedures, poll_wait() and poll_freewait() make all the
113 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
114 * as all select/poll functions have to call it to add an entry to the
115 * poll table.
116 */
117static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
118 poll_table *p);
119
120void poll_initwait(struct poll_wqueues *pwq)
121{
122 init_poll_funcptr(&pwq->pt, __pollwait);
123 pwq->polling_task = current;
124 pwq->triggered = 0;
125 pwq->error = 0;
126 pwq->table = NULL;
127 pwq->inline_index = 0;
128}
129EXPORT_SYMBOL(poll_initwait);
130
131static void free_poll_entry(struct poll_table_entry *entry)
132{
133 remove_wait_queue(entry->wait_address, &entry->wait);
134 fput(entry->filp);
135}
136
137void poll_freewait(struct poll_wqueues *pwq)
138{
139 struct poll_table_page * p = pwq->table;
140 int i;
141 for (i = 0; i < pwq->inline_index; i++)
142 free_poll_entry(pwq->inline_entries + i);
143 while (p) {
144 struct poll_table_entry * entry;
145 struct poll_table_page *old;
146
147 entry = p->entry;
148 do {
149 entry--;
150 free_poll_entry(entry);
151 } while (entry > p->entries);
152 old = p;
153 p = p->next;
154 free_page((unsigned long) old);
155 }
156}
157EXPORT_SYMBOL(poll_freewait);
158
159static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
160{
161 struct poll_table_page *table = p->table;
162
163 if (p->inline_index < N_INLINE_POLL_ENTRIES)
164 return p->inline_entries + p->inline_index++;
165
166 if (!table || POLL_TABLE_FULL(table)) {
167 struct poll_table_page *new_table;
168
169 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
170 if (!new_table) {
171 p->error = -ENOMEM;
172 return NULL;
173 }
174 new_table->entry = new_table->entries;
175 new_table->next = table;
176 p->table = new_table;
177 table = new_table;
178 }
179
180 return table->entry++;
181}
182
183static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
184{
185 struct poll_wqueues *pwq = wait->private;
186 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
187
188 /*
189 * Although this function is called under waitqueue lock, LOCK
190 * doesn't imply write barrier and the users expect write
191 * barrier semantics on wakeup functions. The following
192 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
193 * and is paired with smp_store_mb() in poll_schedule_timeout.
194 */
195 smp_wmb();
196 pwq->triggered = 1;
197
198 /*
199 * Perform the default wake up operation using a dummy
200 * waitqueue.
201 *
202 * TODO: This is hacky but there currently is no interface to
203 * pass in @sync. @sync is scheduled to be removed and once
204 * that happens, wake_up_process() can be used directly.
205 */
206 return default_wake_function(&dummy_wait, mode, sync, key);
207}
208
209static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
210{
211 struct poll_table_entry *entry;
212
213 entry = container_of(wait, struct poll_table_entry, wait);
214 if (key && !((unsigned long)key & entry->key))
215 return 0;
216 return __pollwake(wait, mode, sync, key);
217}
218
219/* Add a new entry */
220static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
221 poll_table *p)
222{
223 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
224 struct poll_table_entry *entry = poll_get_entry(pwq);
225 if (!entry)
226 return;
227 entry->filp = get_file(filp);
228 entry->wait_address = wait_address;
229 entry->key = p->_key;
230 init_waitqueue_func_entry(&entry->wait, pollwake);
231 entry->wait.private = pwq;
232 add_wait_queue(wait_address, &entry->wait);
233}
234
235int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
236 ktime_t *expires, unsigned long slack)
237{
238 int rc = -EINTR;
239
240 set_current_state(state);
241 if (!pwq->triggered)
242 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
243 __set_current_state(TASK_RUNNING);
244
245 /*
246 * Prepare for the next iteration.
247 *
248 * The following smp_store_mb() serves two purposes. First, it's
249 * the counterpart rmb of the wmb in pollwake() such that data
250 * written before wake up is always visible after wake up.
251 * Second, the full barrier guarantees that triggered clearing
252 * doesn't pass event check of the next iteration. Note that
253 * this problem doesn't exist for the first iteration as
254 * add_wait_queue() has full barrier semantics.
255 */
256 smp_store_mb(pwq->triggered, 0);
257
258 return rc;
259}
260EXPORT_SYMBOL(poll_schedule_timeout);
261
262/**
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)
267 *
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.
270 *
271 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
272 */
273int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
274{
275 struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
276
277 if (!timespec64_valid(&ts))
278 return -EINVAL;
279
280 /* Optimize for the zero timeout value here */
281 if (!sec && !nsec) {
282 to->tv_sec = to->tv_nsec = 0;
283 } else {
284 ktime_get_ts64(to);
285 *to = timespec64_add_safe(*to, ts);
286 }
287 return 0;
288}
289
290static int poll_select_copy_remaining(struct timespec64 *end_time,
291 void __user *p,
292 int timeval, int ret)
293{
294 struct timespec64 rts64;
295 struct timespec rts;
296 struct timeval rtv;
297
298 if (!p)
299 return ret;
300
301 if (current->personality & STICKY_TIMEOUTS)
302 goto sticky;
303
304 /* No update for zero timeout */
305 if (!end_time->tv_sec && !end_time->tv_nsec)
306 return ret;
307
308 ktime_get_ts64(&rts64);
309 rts64 = timespec64_sub(*end_time, rts64);
310 if (rts64.tv_sec < 0)
311 rts64.tv_sec = rts64.tv_nsec = 0;
312
313 rts = timespec64_to_timespec(rts64);
314
315 if (timeval) {
316 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
317 memset(&rtv, 0, sizeof(rtv));
318 rtv.tv_sec = rts64.tv_sec;
319 rtv.tv_usec = rts64.tv_nsec / NSEC_PER_USEC;
320
321 if (!copy_to_user(p, &rtv, sizeof(rtv)))
322 return ret;
323
324 } else if (!copy_to_user(p, &rts, sizeof(rts)))
325 return ret;
326
327 /*
328 * If an application puts its timeval in read-only memory, we
329 * don't want the Linux-specific update to the timeval to
330 * cause a fault after the select has completed
331 * successfully. However, because we're not updating the
332 * timeval, we can't restart the system call.
333 */
334
335sticky:
336 if (ret == -ERESTARTNOHAND)
337 ret = -EINTR;
338 return ret;
339}
340
341#define FDS_IN(fds, n) (fds->in + n)
342#define FDS_OUT(fds, n) (fds->out + n)
343#define FDS_EX(fds, n) (fds->ex + n)
344
345#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
346
347static int max_select_fd(unsigned long n, fd_set_bits *fds)
348{
349 unsigned long *open_fds;
350 unsigned long set;
351 int max;
352 struct fdtable *fdt;
353
354 /* handle last in-complete long-word first */
355 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
356 n /= BITS_PER_LONG;
357 fdt = files_fdtable(current->files);
358 open_fds = fdt->open_fds + n;
359 max = 0;
360 if (set) {
361 set &= BITS(fds, n);
362 if (set) {
363 if (!(set & ~*open_fds))
364 goto get_max;
365 return -EBADF;
366 }
367 }
368 while (n) {
369 open_fds--;
370 n--;
371 set = BITS(fds, n);
372 if (!set)
373 continue;
374 if (set & ~*open_fds)
375 return -EBADF;
376 if (max)
377 continue;
378get_max:
379 do {
380 max++;
381 set >>= 1;
382 } while (set);
383 max += n * BITS_PER_LONG;
384 }
385
386 return max;
387}
388
389#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
390#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
391#define POLLEX_SET (POLLPRI)
392
393static inline void wait_key_set(poll_table *wait, unsigned long in,
394 unsigned long out, unsigned long bit,
395 unsigned int ll_flag)
396{
397 wait->_key = POLLEX_SET | ll_flag;
398 if (in & bit)
399 wait->_key |= POLLIN_SET;
400 if (out & bit)
401 wait->_key |= POLLOUT_SET;
402}
403
404int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
405{
406 ktime_t expire, *to = NULL;
407 struct poll_wqueues table;
408 poll_table *wait;
409 int retval, i, timed_out = 0;
410 u64 slack = 0;
411 unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
412 unsigned long busy_end = 0;
413
414 rcu_read_lock();
415 retval = max_select_fd(n, fds);
416 rcu_read_unlock();
417
418 if (retval < 0)
419 return retval;
420 n = retval;
421
422 poll_initwait(&table);
423 wait = &table.pt;
424 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
425 wait->_qproc = NULL;
426 timed_out = 1;
427 }
428
429 if (end_time && !timed_out)
430 slack = select_estimate_accuracy(end_time);
431
432 retval = 0;
433 for (;;) {
434 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
435 bool can_busy_loop = false;
436
437 inp = fds->in; outp = fds->out; exp = fds->ex;
438 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
439
440 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
441 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
442 unsigned long res_in = 0, res_out = 0, res_ex = 0;
443
444 in = *inp++; out = *outp++; ex = *exp++;
445 all_bits = in | out | ex;
446 if (all_bits == 0) {
447 i += BITS_PER_LONG;
448 continue;
449 }
450
451 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
452 struct fd f;
453 if (i >= n)
454 break;
455 if (!(bit & all_bits))
456 continue;
457 f = fdget(i);
458 if (f.file) {
459 const struct file_operations *f_op;
460 f_op = f.file->f_op;
461 mask = DEFAULT_POLLMASK;
462 if (f_op->poll) {
463 wait_key_set(wait, in, out,
464 bit, busy_flag);
465 mask = (*f_op->poll)(f.file, wait);
466 }
467 fdput(f);
468 if ((mask & POLLIN_SET) && (in & bit)) {
469 res_in |= bit;
470 retval++;
471 wait->_qproc = NULL;
472 }
473 if ((mask & POLLOUT_SET) && (out & bit)) {
474 res_out |= bit;
475 retval++;
476 wait->_qproc = NULL;
477 }
478 if ((mask & POLLEX_SET) && (ex & bit)) {
479 res_ex |= bit;
480 retval++;
481 wait->_qproc = NULL;
482 }
483 /* got something, stop busy polling */
484 if (retval) {
485 can_busy_loop = false;
486 busy_flag = 0;
487
488 /*
489 * only remember a returned
490 * POLL_BUSY_LOOP if we asked for it
491 */
492 } else if (busy_flag & mask)
493 can_busy_loop = true;
494
495 }
496 }
497 if (res_in)
498 *rinp = res_in;
499 if (res_out)
500 *routp = res_out;
501 if (res_ex)
502 *rexp = res_ex;
503 cond_resched();
504 }
505 wait->_qproc = NULL;
506 if (retval || timed_out || signal_pending(current))
507 break;
508 if (table.error) {
509 retval = table.error;
510 break;
511 }
512
513 /* only if found POLL_BUSY_LOOP sockets && not out of time */
514 if (can_busy_loop && !need_resched()) {
515 if (!busy_end) {
516 busy_end = busy_loop_end_time();
517 continue;
518 }
519 if (!busy_loop_timeout(busy_end))
520 continue;
521 }
522 busy_flag = 0;
523
524 /*
525 * If this is the first loop and we have a timeout
526 * given, then we convert to ktime_t and set the to
527 * pointer to the expiry value.
528 */
529 if (end_time && !to) {
530 expire = timespec64_to_ktime(*end_time);
531 to = &expire;
532 }
533
534 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
535 to, slack))
536 timed_out = 1;
537 }
538
539 poll_freewait(&table);
540
541 return retval;
542}
543
544/*
545 * We can actually return ERESTARTSYS instead of EINTR, but I'd
546 * like to be certain this leads to no problems. So I return
547 * EINTR just for safety.
548 *
549 * Update: ERESTARTSYS breaks at least the xview clock binary, so
550 * I'm trying ERESTARTNOHAND which restart only when you want to.
551 */
552int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
553 fd_set __user *exp, struct timespec64 *end_time)
554{
555 fd_set_bits fds;
556 void *bits;
557 int ret, max_fds;
558 size_t size, alloc_size;
559 struct fdtable *fdt;
560 /* Allocate small arguments on the stack to save memory and be faster */
561 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
562
563 ret = -EINVAL;
564 if (n < 0)
565 goto out_nofds;
566
567 /* max_fds can increase, so grab it once to avoid race */
568 rcu_read_lock();
569 fdt = files_fdtable(current->files);
570 max_fds = fdt->max_fds;
571 rcu_read_unlock();
572 if (n > max_fds)
573 n = max_fds;
574
575 /*
576 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
577 * since we used fdset we need to allocate memory in units of
578 * long-words.
579 */
580 size = FDS_BYTES(n);
581 bits = stack_fds;
582 if (size > sizeof(stack_fds) / 6) {
583 /* Not enough space in on-stack array; must use kmalloc */
584 ret = -ENOMEM;
585 if (size > (SIZE_MAX / 6))
586 goto out_nofds;
587
588 alloc_size = 6 * size;
589 bits = kmalloc(alloc_size, GFP_KERNEL|__GFP_NOWARN);
590 if (!bits && alloc_size > PAGE_SIZE)
591 bits = vmalloc(alloc_size);
592
593 if (!bits)
594 goto out_nofds;
595 }
596 fds.in = bits;
597 fds.out = bits + size;
598 fds.ex = bits + 2*size;
599 fds.res_in = bits + 3*size;
600 fds.res_out = bits + 4*size;
601 fds.res_ex = bits + 5*size;
602
603 if ((ret = get_fd_set(n, inp, fds.in)) ||
604 (ret = get_fd_set(n, outp, fds.out)) ||
605 (ret = get_fd_set(n, exp, fds.ex)))
606 goto out;
607 zero_fd_set(n, fds.res_in);
608 zero_fd_set(n, fds.res_out);
609 zero_fd_set(n, fds.res_ex);
610
611 ret = do_select(n, &fds, end_time);
612
613 if (ret < 0)
614 goto out;
615 if (!ret) {
616 ret = -ERESTARTNOHAND;
617 if (signal_pending(current))
618 goto out;
619 ret = 0;
620 }
621
622 if (set_fd_set(n, inp, fds.res_in) ||
623 set_fd_set(n, outp, fds.res_out) ||
624 set_fd_set(n, exp, fds.res_ex))
625 ret = -EFAULT;
626
627out:
628 if (bits != stack_fds)
629 kvfree(bits);
630out_nofds:
631 return ret;
632}
633
634SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
635 fd_set __user *, exp, struct timeval __user *, tvp)
636{
637 struct timespec64 end_time, *to = NULL;
638 struct timeval tv;
639 int ret;
640
641 if (tvp) {
642 if (copy_from_user(&tv, tvp, sizeof(tv)))
643 return -EFAULT;
644
645 to = &end_time;
646 if (poll_select_set_timeout(to,
647 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
648 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
649 return -EINVAL;
650 }
651
652 ret = core_sys_select(n, inp, outp, exp, to);
653 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
654
655 return ret;
656}
657
658static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
659 fd_set __user *exp, struct timespec __user *tsp,
660 const sigset_t __user *sigmask, size_t sigsetsize)
661{
662 sigset_t ksigmask, sigsaved;
663 struct timespec ts;
664 struct timespec64 ts64, end_time, *to = NULL;
665 int ret;
666
667 if (tsp) {
668 if (copy_from_user(&ts, tsp, sizeof(ts)))
669 return -EFAULT;
670 ts64 = timespec_to_timespec64(ts);
671
672 to = &end_time;
673 if (poll_select_set_timeout(to, ts64.tv_sec, ts64.tv_nsec))
674 return -EINVAL;
675 }
676
677 if (sigmask) {
678 /* XXX: Don't preclude handling different sized sigset_t's. */
679 if (sigsetsize != sizeof(sigset_t))
680 return -EINVAL;
681 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
682 return -EFAULT;
683
684 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
685 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
686 }
687
688 ret = core_sys_select(n, inp, outp, exp, to);
689 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
690
691 if (ret == -ERESTARTNOHAND) {
692 /*
693 * Don't restore the signal mask yet. Let do_signal() deliver
694 * the signal on the way back to userspace, before the signal
695 * mask is restored.
696 */
697 if (sigmask) {
698 memcpy(¤t->saved_sigmask, &sigsaved,
699 sizeof(sigsaved));
700 set_restore_sigmask();
701 }
702 } else if (sigmask)
703 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
704
705 return ret;
706}
707
708/*
709 * Most architectures can't handle 7-argument syscalls. So we provide a
710 * 6-argument version where the sixth argument is a pointer to a structure
711 * which has a pointer to the sigset_t itself followed by a size_t containing
712 * the sigset size.
713 */
714SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
715 fd_set __user *, exp, struct timespec __user *, tsp,
716 void __user *, sig)
717{
718 size_t sigsetsize = 0;
719 sigset_t __user *up = NULL;
720
721 if (sig) {
722 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
723 || __get_user(up, (sigset_t __user * __user *)sig)
724 || __get_user(sigsetsize,
725 (size_t __user *)(sig+sizeof(void *))))
726 return -EFAULT;
727 }
728
729 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
730}
731
732#ifdef __ARCH_WANT_SYS_OLD_SELECT
733struct sel_arg_struct {
734 unsigned long n;
735 fd_set __user *inp, *outp, *exp;
736 struct timeval __user *tvp;
737};
738
739SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
740{
741 struct sel_arg_struct a;
742
743 if (copy_from_user(&a, arg, sizeof(a)))
744 return -EFAULT;
745 return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
746}
747#endif
748
749struct poll_list {
750 struct poll_list *next;
751 int len;
752 struct pollfd entries[0];
753};
754
755#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
756
757/*
758 * Fish for pollable events on the pollfd->fd file descriptor. We're only
759 * interested in events matching the pollfd->events mask, and the result
760 * matching that mask is both recorded in pollfd->revents and returned. The
761 * pwait poll_table will be used by the fd-provided poll handler for waiting,
762 * if pwait->_qproc is non-NULL.
763 */
764static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait,
765 bool *can_busy_poll,
766 unsigned int busy_flag)
767{
768 unsigned int mask;
769 int fd;
770
771 mask = 0;
772 fd = pollfd->fd;
773 if (fd >= 0) {
774 struct fd f = fdget(fd);
775 mask = POLLNVAL;
776 if (f.file) {
777 mask = DEFAULT_POLLMASK;
778 if (f.file->f_op->poll) {
779 pwait->_key = pollfd->events|POLLERR|POLLHUP;
780 pwait->_key |= busy_flag;
781 mask = f.file->f_op->poll(f.file, pwait);
782 if (mask & busy_flag)
783 *can_busy_poll = true;
784 }
785 /* Mask out unneeded events. */
786 mask &= pollfd->events | POLLERR | POLLHUP;
787 fdput(f);
788 }
789 }
790 pollfd->revents = mask;
791
792 return mask;
793}
794
795static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
796 struct timespec64 *end_time)
797{
798 poll_table* pt = &wait->pt;
799 ktime_t expire, *to = NULL;
800 int timed_out = 0, count = 0;
801 u64 slack = 0;
802 unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
803 unsigned long busy_end = 0;
804
805 /* Optimise the no-wait case */
806 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
807 pt->_qproc = NULL;
808 timed_out = 1;
809 }
810
811 if (end_time && !timed_out)
812 slack = select_estimate_accuracy(end_time);
813
814 for (;;) {
815 struct poll_list *walk;
816 bool can_busy_loop = false;
817
818 for (walk = list; walk != NULL; walk = walk->next) {
819 struct pollfd * pfd, * pfd_end;
820
821 pfd = walk->entries;
822 pfd_end = pfd + walk->len;
823 for (; pfd != pfd_end; pfd++) {
824 /*
825 * Fish for events. If we found one, record it
826 * and kill poll_table->_qproc, so we don't
827 * needlessly register any other waiters after
828 * this. They'll get immediately deregistered
829 * when we break out and return.
830 */
831 if (do_pollfd(pfd, pt, &can_busy_loop,
832 busy_flag)) {
833 count++;
834 pt->_qproc = NULL;
835 /* found something, stop busy polling */
836 busy_flag = 0;
837 can_busy_loop = false;
838 }
839 }
840 }
841 /*
842 * All waiters have already been registered, so don't provide
843 * a poll_table->_qproc to them on the next loop iteration.
844 */
845 pt->_qproc = NULL;
846 if (!count) {
847 count = wait->error;
848 if (signal_pending(current))
849 count = -EINTR;
850 }
851 if (count || timed_out)
852 break;
853
854 /* only if found POLL_BUSY_LOOP sockets && not out of time */
855 if (can_busy_loop && !need_resched()) {
856 if (!busy_end) {
857 busy_end = busy_loop_end_time();
858 continue;
859 }
860 if (!busy_loop_timeout(busy_end))
861 continue;
862 }
863 busy_flag = 0;
864
865 /*
866 * If this is the first loop and we have a timeout
867 * given, then we convert to ktime_t and set the to
868 * pointer to the expiry value.
869 */
870 if (end_time && !to) {
871 expire = timespec64_to_ktime(*end_time);
872 to = &expire;
873 }
874
875 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
876 timed_out = 1;
877 }
878 return count;
879}
880
881#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
882 sizeof(struct pollfd))
883
884int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
885 struct timespec64 *end_time)
886{
887 struct poll_wqueues table;
888 int err = -EFAULT, fdcount, len, size;
889 /* Allocate small arguments on the stack to save memory and be
890 faster - use long to make sure the buffer is aligned properly
891 on 64 bit archs to avoid unaligned access */
892 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
893 struct poll_list *const head = (struct poll_list *)stack_pps;
894 struct poll_list *walk = head;
895 unsigned long todo = nfds;
896
897 if (nfds > rlimit(RLIMIT_NOFILE))
898 return -EINVAL;
899
900 len = min_t(unsigned int, nfds, N_STACK_PPS);
901 for (;;) {
902 walk->next = NULL;
903 walk->len = len;
904 if (!len)
905 break;
906
907 if (copy_from_user(walk->entries, ufds + nfds-todo,
908 sizeof(struct pollfd) * walk->len))
909 goto out_fds;
910
911 todo -= walk->len;
912 if (!todo)
913 break;
914
915 len = min(todo, POLLFD_PER_PAGE);
916 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
917 walk = walk->next = kmalloc(size, GFP_KERNEL);
918 if (!walk) {
919 err = -ENOMEM;
920 goto out_fds;
921 }
922 }
923
924 poll_initwait(&table);
925 fdcount = do_poll(head, &table, end_time);
926 poll_freewait(&table);
927
928 for (walk = head; walk; walk = walk->next) {
929 struct pollfd *fds = walk->entries;
930 int j;
931
932 for (j = 0; j < walk->len; j++, ufds++)
933 if (__put_user(fds[j].revents, &ufds->revents))
934 goto out_fds;
935 }
936
937 err = fdcount;
938out_fds:
939 walk = head->next;
940 while (walk) {
941 struct poll_list *pos = walk;
942 walk = walk->next;
943 kfree(pos);
944 }
945
946 return err;
947}
948
949static long do_restart_poll(struct restart_block *restart_block)
950{
951 struct pollfd __user *ufds = restart_block->poll.ufds;
952 int nfds = restart_block->poll.nfds;
953 struct timespec64 *to = NULL, end_time;
954 int ret;
955
956 if (restart_block->poll.has_timeout) {
957 end_time.tv_sec = restart_block->poll.tv_sec;
958 end_time.tv_nsec = restart_block->poll.tv_nsec;
959 to = &end_time;
960 }
961
962 ret = do_sys_poll(ufds, nfds, to);
963
964 if (ret == -EINTR) {
965 restart_block->fn = do_restart_poll;
966 ret = -ERESTART_RESTARTBLOCK;
967 }
968 return ret;
969}
970
971SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
972 int, timeout_msecs)
973{
974 struct timespec64 end_time, *to = NULL;
975 int ret;
976
977 if (timeout_msecs >= 0) {
978 to = &end_time;
979 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
980 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
981 }
982
983 ret = do_sys_poll(ufds, nfds, to);
984
985 if (ret == -EINTR) {
986 struct restart_block *restart_block;
987
988 restart_block = ¤t->restart_block;
989 restart_block->fn = do_restart_poll;
990 restart_block->poll.ufds = ufds;
991 restart_block->poll.nfds = nfds;
992
993 if (timeout_msecs >= 0) {
994 restart_block->poll.tv_sec = end_time.tv_sec;
995 restart_block->poll.tv_nsec = end_time.tv_nsec;
996 restart_block->poll.has_timeout = 1;
997 } else
998 restart_block->poll.has_timeout = 0;
999
1000 ret = -ERESTART_RESTARTBLOCK;
1001 }
1002 return ret;
1003}
1004
1005SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1006 struct timespec __user *, tsp, const sigset_t __user *, sigmask,
1007 size_t, sigsetsize)
1008{
1009 sigset_t ksigmask, sigsaved;
1010 struct timespec ts;
1011 struct timespec64 end_time, *to = NULL;
1012 int ret;
1013
1014 if (tsp) {
1015 if (copy_from_user(&ts, tsp, sizeof(ts)))
1016 return -EFAULT;
1017
1018 to = &end_time;
1019 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1020 return -EINVAL;
1021 }
1022
1023 if (sigmask) {
1024 /* XXX: Don't preclude handling different sized sigset_t's. */
1025 if (sigsetsize != sizeof(sigset_t))
1026 return -EINVAL;
1027 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1028 return -EFAULT;
1029
1030 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1031 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1032 }
1033
1034 ret = do_sys_poll(ufds, nfds, to);
1035
1036 /* We can restart this syscall, usually */
1037 if (ret == -EINTR) {
1038 /*
1039 * Don't restore the signal mask yet. Let do_signal() deliver
1040 * the signal on the way back to userspace, before the signal
1041 * mask is restored.
1042 */
1043 if (sigmask) {
1044 memcpy(¤t->saved_sigmask, &sigsaved,
1045 sizeof(sigsaved));
1046 set_restore_sigmask();
1047 }
1048 ret = -ERESTARTNOHAND;
1049 } else if (sigmask)
1050 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1051
1052 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1053
1054 return ret;
1055}
1/*
2 * This file contains the procedures for the handling of select and poll
3 *
4 * Created for Linux based loosely upon Mathius Lattner's minix
5 * patches by Peter MacDonald. Heavily edited by Linus.
6 *
7 * 4 February 1994
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.
11 *
12 * 24 January 2000
13 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
14 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
15 */
16
17#include <linux/kernel.h>
18#include <linux/sched.h>
19#include <linux/syscalls.h>
20#include <linux/export.h>
21#include <linux/slab.h>
22#include <linux/poll.h>
23#include <linux/personality.h> /* for STICKY_TIMEOUTS */
24#include <linux/file.h>
25#include <linux/fdtable.h>
26#include <linux/fs.h>
27#include <linux/rcupdate.h>
28#include <linux/hrtimer.h>
29
30#include <asm/uaccess.h>
31
32
33/*
34 * Estimate expected accuracy in ns from a timeval.
35 *
36 * After quite a bit of churning around, we've settled on
37 * a simple thing of taking 0.1% of the timeout as the
38 * slack, with a cap of 100 msec.
39 * "nice" tasks get a 0.5% slack instead.
40 *
41 * Consider this comment an open invitation to come up with even
42 * better solutions..
43 */
44
45#define MAX_SLACK (100 * NSEC_PER_MSEC)
46
47static long __estimate_accuracy(struct timespec *tv)
48{
49 long slack;
50 int divfactor = 1000;
51
52 if (tv->tv_sec < 0)
53 return 0;
54
55 if (task_nice(current) > 0)
56 divfactor = divfactor / 5;
57
58 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
59 return MAX_SLACK;
60
61 slack = tv->tv_nsec / divfactor;
62 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
63
64 if (slack > MAX_SLACK)
65 return MAX_SLACK;
66
67 return slack;
68}
69
70long select_estimate_accuracy(struct timespec *tv)
71{
72 unsigned long ret;
73 struct timespec now;
74
75 /*
76 * Realtime tasks get a slack of 0 for obvious reasons.
77 */
78
79 if (rt_task(current))
80 return 0;
81
82 ktime_get_ts(&now);
83 now = timespec_sub(*tv, now);
84 ret = __estimate_accuracy(&now);
85 if (ret < current->timer_slack_ns)
86 return current->timer_slack_ns;
87 return ret;
88}
89
90
91
92struct poll_table_page {
93 struct poll_table_page * next;
94 struct poll_table_entry * entry;
95 struct poll_table_entry entries[0];
96};
97
98#define POLL_TABLE_FULL(table) \
99 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
100
101/*
102 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
103 * I have rewritten this, taking some shortcuts: This code may not be easy to
104 * follow, but it should be free of race-conditions, and it's practical. If you
105 * understand what I'm doing here, then you understand how the linux
106 * sleep/wakeup mechanism works.
107 *
108 * Two very simple procedures, poll_wait() and poll_freewait() make all the
109 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
110 * as all select/poll functions have to call it to add an entry to the
111 * poll table.
112 */
113static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
114 poll_table *p);
115
116void poll_initwait(struct poll_wqueues *pwq)
117{
118 init_poll_funcptr(&pwq->pt, __pollwait);
119 pwq->polling_task = current;
120 pwq->triggered = 0;
121 pwq->error = 0;
122 pwq->table = NULL;
123 pwq->inline_index = 0;
124}
125EXPORT_SYMBOL(poll_initwait);
126
127static void free_poll_entry(struct poll_table_entry *entry)
128{
129 remove_wait_queue(entry->wait_address, &entry->wait);
130 fput(entry->filp);
131}
132
133void poll_freewait(struct poll_wqueues *pwq)
134{
135 struct poll_table_page * p = pwq->table;
136 int i;
137 for (i = 0; i < pwq->inline_index; i++)
138 free_poll_entry(pwq->inline_entries + i);
139 while (p) {
140 struct poll_table_entry * entry;
141 struct poll_table_page *old;
142
143 entry = p->entry;
144 do {
145 entry--;
146 free_poll_entry(entry);
147 } while (entry > p->entries);
148 old = p;
149 p = p->next;
150 free_page((unsigned long) old);
151 }
152}
153EXPORT_SYMBOL(poll_freewait);
154
155static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
156{
157 struct poll_table_page *table = p->table;
158
159 if (p->inline_index < N_INLINE_POLL_ENTRIES)
160 return p->inline_entries + p->inline_index++;
161
162 if (!table || POLL_TABLE_FULL(table)) {
163 struct poll_table_page *new_table;
164
165 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
166 if (!new_table) {
167 p->error = -ENOMEM;
168 return NULL;
169 }
170 new_table->entry = new_table->entries;
171 new_table->next = table;
172 p->table = new_table;
173 table = new_table;
174 }
175
176 return table->entry++;
177}
178
179static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
180{
181 struct poll_wqueues *pwq = wait->private;
182 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
183
184 /*
185 * Although this function is called under waitqueue lock, LOCK
186 * doesn't imply write barrier and the users expect write
187 * barrier semantics on wakeup functions. The following
188 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
189 * and is paired with set_mb() in poll_schedule_timeout.
190 */
191 smp_wmb();
192 pwq->triggered = 1;
193
194 /*
195 * Perform the default wake up operation using a dummy
196 * waitqueue.
197 *
198 * TODO: This is hacky but there currently is no interface to
199 * pass in @sync. @sync is scheduled to be removed and once
200 * that happens, wake_up_process() can be used directly.
201 */
202 return default_wake_function(&dummy_wait, mode, sync, key);
203}
204
205static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
206{
207 struct poll_table_entry *entry;
208
209 entry = container_of(wait, struct poll_table_entry, wait);
210 if (key && !((unsigned long)key & entry->key))
211 return 0;
212 return __pollwake(wait, mode, sync, key);
213}
214
215/* Add a new entry */
216static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
217 poll_table *p)
218{
219 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
220 struct poll_table_entry *entry = poll_get_entry(pwq);
221 if (!entry)
222 return;
223 get_file(filp);
224 entry->filp = filp;
225 entry->wait_address = wait_address;
226 entry->key = p->_key;
227 init_waitqueue_func_entry(&entry->wait, pollwake);
228 entry->wait.private = pwq;
229 add_wait_queue(wait_address, &entry->wait);
230}
231
232int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
233 ktime_t *expires, unsigned long slack)
234{
235 int rc = -EINTR;
236
237 set_current_state(state);
238 if (!pwq->triggered)
239 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
240 __set_current_state(TASK_RUNNING);
241
242 /*
243 * Prepare for the next iteration.
244 *
245 * The following set_mb() serves two purposes. First, it's
246 * the counterpart rmb of the wmb in pollwake() such that data
247 * written before wake up is always visible after wake up.
248 * Second, the full barrier guarantees that triggered clearing
249 * doesn't pass event check of the next iteration. Note that
250 * this problem doesn't exist for the first iteration as
251 * add_wait_queue() has full barrier semantics.
252 */
253 set_mb(pwq->triggered, 0);
254
255 return rc;
256}
257EXPORT_SYMBOL(poll_schedule_timeout);
258
259/**
260 * poll_select_set_timeout - helper function to setup the timeout value
261 * @to: pointer to timespec variable for the final timeout
262 * @sec: seconds (from user space)
263 * @nsec: nanoseconds (from user space)
264 *
265 * Note, we do not use a timespec for the user space value here, That
266 * way we can use the function for timeval and compat interfaces as well.
267 *
268 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
269 */
270int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
271{
272 struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
273
274 if (!timespec_valid(&ts))
275 return -EINVAL;
276
277 /* Optimize for the zero timeout value here */
278 if (!sec && !nsec) {
279 to->tv_sec = to->tv_nsec = 0;
280 } else {
281 ktime_get_ts(to);
282 *to = timespec_add_safe(*to, ts);
283 }
284 return 0;
285}
286
287static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
288 int timeval, int ret)
289{
290 struct timespec rts;
291 struct timeval rtv;
292
293 if (!p)
294 return ret;
295
296 if (current->personality & STICKY_TIMEOUTS)
297 goto sticky;
298
299 /* No update for zero timeout */
300 if (!end_time->tv_sec && !end_time->tv_nsec)
301 return ret;
302
303 ktime_get_ts(&rts);
304 rts = timespec_sub(*end_time, rts);
305 if (rts.tv_sec < 0)
306 rts.tv_sec = rts.tv_nsec = 0;
307
308 if (timeval) {
309 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
310 memset(&rtv, 0, sizeof(rtv));
311 rtv.tv_sec = rts.tv_sec;
312 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
313
314 if (!copy_to_user(p, &rtv, sizeof(rtv)))
315 return ret;
316
317 } else if (!copy_to_user(p, &rts, sizeof(rts)))
318 return ret;
319
320 /*
321 * If an application puts its timeval in read-only memory, we
322 * don't want the Linux-specific update to the timeval to
323 * cause a fault after the select has completed
324 * successfully. However, because we're not updating the
325 * timeval, we can't restart the system call.
326 */
327
328sticky:
329 if (ret == -ERESTARTNOHAND)
330 ret = -EINTR;
331 return ret;
332}
333
334#define FDS_IN(fds, n) (fds->in + n)
335#define FDS_OUT(fds, n) (fds->out + n)
336#define FDS_EX(fds, n) (fds->ex + n)
337
338#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
339
340static int max_select_fd(unsigned long n, fd_set_bits *fds)
341{
342 unsigned long *open_fds;
343 unsigned long set;
344 int max;
345 struct fdtable *fdt;
346
347 /* handle last in-complete long-word first */
348 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
349 n /= BITS_PER_LONG;
350 fdt = files_fdtable(current->files);
351 open_fds = fdt->open_fds + n;
352 max = 0;
353 if (set) {
354 set &= BITS(fds, n);
355 if (set) {
356 if (!(set & ~*open_fds))
357 goto get_max;
358 return -EBADF;
359 }
360 }
361 while (n) {
362 open_fds--;
363 n--;
364 set = BITS(fds, n);
365 if (!set)
366 continue;
367 if (set & ~*open_fds)
368 return -EBADF;
369 if (max)
370 continue;
371get_max:
372 do {
373 max++;
374 set >>= 1;
375 } while (set);
376 max += n * BITS_PER_LONG;
377 }
378
379 return max;
380}
381
382#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
383#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
384#define POLLEX_SET (POLLPRI)
385
386static inline void wait_key_set(poll_table *wait, unsigned long in,
387 unsigned long out, unsigned long bit)
388{
389 wait->_key = POLLEX_SET;
390 if (in & bit)
391 wait->_key |= POLLIN_SET;
392 if (out & bit)
393 wait->_key |= POLLOUT_SET;
394}
395
396int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
397{
398 ktime_t expire, *to = NULL;
399 struct poll_wqueues table;
400 poll_table *wait;
401 int retval, i, timed_out = 0;
402 unsigned long slack = 0;
403
404 rcu_read_lock();
405 retval = max_select_fd(n, fds);
406 rcu_read_unlock();
407
408 if (retval < 0)
409 return retval;
410 n = retval;
411
412 poll_initwait(&table);
413 wait = &table.pt;
414 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
415 wait->_qproc = NULL;
416 timed_out = 1;
417 }
418
419 if (end_time && !timed_out)
420 slack = select_estimate_accuracy(end_time);
421
422 retval = 0;
423 for (;;) {
424 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
425
426 inp = fds->in; outp = fds->out; exp = fds->ex;
427 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
428
429 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
430 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
431 unsigned long res_in = 0, res_out = 0, res_ex = 0;
432 const struct file_operations *f_op = NULL;
433 struct file *file = NULL;
434
435 in = *inp++; out = *outp++; ex = *exp++;
436 all_bits = in | out | ex;
437 if (all_bits == 0) {
438 i += BITS_PER_LONG;
439 continue;
440 }
441
442 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
443 int fput_needed;
444 if (i >= n)
445 break;
446 if (!(bit & all_bits))
447 continue;
448 file = fget_light(i, &fput_needed);
449 if (file) {
450 f_op = file->f_op;
451 mask = DEFAULT_POLLMASK;
452 if (f_op && f_op->poll) {
453 wait_key_set(wait, in, out, bit);
454 mask = (*f_op->poll)(file, wait);
455 }
456 fput_light(file, fput_needed);
457 if ((mask & POLLIN_SET) && (in & bit)) {
458 res_in |= bit;
459 retval++;
460 wait->_qproc = NULL;
461 }
462 if ((mask & POLLOUT_SET) && (out & bit)) {
463 res_out |= bit;
464 retval++;
465 wait->_qproc = NULL;
466 }
467 if ((mask & POLLEX_SET) && (ex & bit)) {
468 res_ex |= bit;
469 retval++;
470 wait->_qproc = NULL;
471 }
472 }
473 }
474 if (res_in)
475 *rinp = res_in;
476 if (res_out)
477 *routp = res_out;
478 if (res_ex)
479 *rexp = res_ex;
480 cond_resched();
481 }
482 wait->_qproc = NULL;
483 if (retval || timed_out || signal_pending(current))
484 break;
485 if (table.error) {
486 retval = table.error;
487 break;
488 }
489
490 /*
491 * If this is the first loop and we have a timeout
492 * given, then we convert to ktime_t and set the to
493 * pointer to the expiry value.
494 */
495 if (end_time && !to) {
496 expire = timespec_to_ktime(*end_time);
497 to = &expire;
498 }
499
500 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
501 to, slack))
502 timed_out = 1;
503 }
504
505 poll_freewait(&table);
506
507 return retval;
508}
509
510/*
511 * We can actually return ERESTARTSYS instead of EINTR, but I'd
512 * like to be certain this leads to no problems. So I return
513 * EINTR just for safety.
514 *
515 * Update: ERESTARTSYS breaks at least the xview clock binary, so
516 * I'm trying ERESTARTNOHAND which restart only when you want to.
517 */
518int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
519 fd_set __user *exp, struct timespec *end_time)
520{
521 fd_set_bits fds;
522 void *bits;
523 int ret, max_fds;
524 unsigned int size;
525 struct fdtable *fdt;
526 /* Allocate small arguments on the stack to save memory and be faster */
527 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
528
529 ret = -EINVAL;
530 if (n < 0)
531 goto out_nofds;
532
533 /* max_fds can increase, so grab it once to avoid race */
534 rcu_read_lock();
535 fdt = files_fdtable(current->files);
536 max_fds = fdt->max_fds;
537 rcu_read_unlock();
538 if (n > max_fds)
539 n = max_fds;
540
541 /*
542 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
543 * since we used fdset we need to allocate memory in units of
544 * long-words.
545 */
546 size = FDS_BYTES(n);
547 bits = stack_fds;
548 if (size > sizeof(stack_fds) / 6) {
549 /* Not enough space in on-stack array; must use kmalloc */
550 ret = -ENOMEM;
551 bits = kmalloc(6 * size, GFP_KERNEL);
552 if (!bits)
553 goto out_nofds;
554 }
555 fds.in = bits;
556 fds.out = bits + size;
557 fds.ex = bits + 2*size;
558 fds.res_in = bits + 3*size;
559 fds.res_out = bits + 4*size;
560 fds.res_ex = bits + 5*size;
561
562 if ((ret = get_fd_set(n, inp, fds.in)) ||
563 (ret = get_fd_set(n, outp, fds.out)) ||
564 (ret = get_fd_set(n, exp, fds.ex)))
565 goto out;
566 zero_fd_set(n, fds.res_in);
567 zero_fd_set(n, fds.res_out);
568 zero_fd_set(n, fds.res_ex);
569
570 ret = do_select(n, &fds, end_time);
571
572 if (ret < 0)
573 goto out;
574 if (!ret) {
575 ret = -ERESTARTNOHAND;
576 if (signal_pending(current))
577 goto out;
578 ret = 0;
579 }
580
581 if (set_fd_set(n, inp, fds.res_in) ||
582 set_fd_set(n, outp, fds.res_out) ||
583 set_fd_set(n, exp, fds.res_ex))
584 ret = -EFAULT;
585
586out:
587 if (bits != stack_fds)
588 kfree(bits);
589out_nofds:
590 return ret;
591}
592
593SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
594 fd_set __user *, exp, struct timeval __user *, tvp)
595{
596 struct timespec end_time, *to = NULL;
597 struct timeval tv;
598 int ret;
599
600 if (tvp) {
601 if (copy_from_user(&tv, tvp, sizeof(tv)))
602 return -EFAULT;
603
604 to = &end_time;
605 if (poll_select_set_timeout(to,
606 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
607 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
608 return -EINVAL;
609 }
610
611 ret = core_sys_select(n, inp, outp, exp, to);
612 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
613
614 return ret;
615}
616
617static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
618 fd_set __user *exp, struct timespec __user *tsp,
619 const sigset_t __user *sigmask, size_t sigsetsize)
620{
621 sigset_t ksigmask, sigsaved;
622 struct timespec ts, end_time, *to = NULL;
623 int ret;
624
625 if (tsp) {
626 if (copy_from_user(&ts, tsp, sizeof(ts)))
627 return -EFAULT;
628
629 to = &end_time;
630 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
631 return -EINVAL;
632 }
633
634 if (sigmask) {
635 /* XXX: Don't preclude handling different sized sigset_t's. */
636 if (sigsetsize != sizeof(sigset_t))
637 return -EINVAL;
638 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
639 return -EFAULT;
640
641 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
642 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
643 }
644
645 ret = core_sys_select(n, inp, outp, exp, to);
646 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
647
648 if (ret == -ERESTARTNOHAND) {
649 /*
650 * Don't restore the signal mask yet. Let do_signal() deliver
651 * the signal on the way back to userspace, before the signal
652 * mask is restored.
653 */
654 if (sigmask) {
655 memcpy(¤t->saved_sigmask, &sigsaved,
656 sizeof(sigsaved));
657 set_restore_sigmask();
658 }
659 } else if (sigmask)
660 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
661
662 return ret;
663}
664
665/*
666 * Most architectures can't handle 7-argument syscalls. So we provide a
667 * 6-argument version where the sixth argument is a pointer to a structure
668 * which has a pointer to the sigset_t itself followed by a size_t containing
669 * the sigset size.
670 */
671SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
672 fd_set __user *, exp, struct timespec __user *, tsp,
673 void __user *, sig)
674{
675 size_t sigsetsize = 0;
676 sigset_t __user *up = NULL;
677
678 if (sig) {
679 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
680 || __get_user(up, (sigset_t __user * __user *)sig)
681 || __get_user(sigsetsize,
682 (size_t __user *)(sig+sizeof(void *))))
683 return -EFAULT;
684 }
685
686 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
687}
688
689#ifdef __ARCH_WANT_SYS_OLD_SELECT
690struct sel_arg_struct {
691 unsigned long n;
692 fd_set __user *inp, *outp, *exp;
693 struct timeval __user *tvp;
694};
695
696SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
697{
698 struct sel_arg_struct a;
699
700 if (copy_from_user(&a, arg, sizeof(a)))
701 return -EFAULT;
702 return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
703}
704#endif
705
706struct poll_list {
707 struct poll_list *next;
708 int len;
709 struct pollfd entries[0];
710};
711
712#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
713
714/*
715 * Fish for pollable events on the pollfd->fd file descriptor. We're only
716 * interested in events matching the pollfd->events mask, and the result
717 * matching that mask is both recorded in pollfd->revents and returned. The
718 * pwait poll_table will be used by the fd-provided poll handler for waiting,
719 * if pwait->_qproc is non-NULL.
720 */
721static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
722{
723 unsigned int mask;
724 int fd;
725
726 mask = 0;
727 fd = pollfd->fd;
728 if (fd >= 0) {
729 int fput_needed;
730 struct file * file;
731
732 file = fget_light(fd, &fput_needed);
733 mask = POLLNVAL;
734 if (file != NULL) {
735 mask = DEFAULT_POLLMASK;
736 if (file->f_op && file->f_op->poll) {
737 pwait->_key = pollfd->events|POLLERR|POLLHUP;
738 mask = file->f_op->poll(file, pwait);
739 }
740 /* Mask out unneeded events. */
741 mask &= pollfd->events | POLLERR | POLLHUP;
742 fput_light(file, fput_needed);
743 }
744 }
745 pollfd->revents = mask;
746
747 return mask;
748}
749
750static int do_poll(unsigned int nfds, struct poll_list *list,
751 struct poll_wqueues *wait, struct timespec *end_time)
752{
753 poll_table* pt = &wait->pt;
754 ktime_t expire, *to = NULL;
755 int timed_out = 0, count = 0;
756 unsigned long slack = 0;
757
758 /* Optimise the no-wait case */
759 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
760 pt->_qproc = NULL;
761 timed_out = 1;
762 }
763
764 if (end_time && !timed_out)
765 slack = select_estimate_accuracy(end_time);
766
767 for (;;) {
768 struct poll_list *walk;
769
770 for (walk = list; walk != NULL; walk = walk->next) {
771 struct pollfd * pfd, * pfd_end;
772
773 pfd = walk->entries;
774 pfd_end = pfd + walk->len;
775 for (; pfd != pfd_end; pfd++) {
776 /*
777 * Fish for events. If we found one, record it
778 * and kill poll_table->_qproc, so we don't
779 * needlessly register any other waiters after
780 * this. They'll get immediately deregistered
781 * when we break out and return.
782 */
783 if (do_pollfd(pfd, pt)) {
784 count++;
785 pt->_qproc = NULL;
786 }
787 }
788 }
789 /*
790 * All waiters have already been registered, so don't provide
791 * a poll_table->_qproc to them on the next loop iteration.
792 */
793 pt->_qproc = NULL;
794 if (!count) {
795 count = wait->error;
796 if (signal_pending(current))
797 count = -EINTR;
798 }
799 if (count || timed_out)
800 break;
801
802 /*
803 * If this is the first loop and we have a timeout
804 * given, then we convert to ktime_t and set the to
805 * pointer to the expiry value.
806 */
807 if (end_time && !to) {
808 expire = timespec_to_ktime(*end_time);
809 to = &expire;
810 }
811
812 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
813 timed_out = 1;
814 }
815 return count;
816}
817
818#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
819 sizeof(struct pollfd))
820
821int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
822 struct timespec *end_time)
823{
824 struct poll_wqueues table;
825 int err = -EFAULT, fdcount, len, size;
826 /* Allocate small arguments on the stack to save memory and be
827 faster - use long to make sure the buffer is aligned properly
828 on 64 bit archs to avoid unaligned access */
829 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
830 struct poll_list *const head = (struct poll_list *)stack_pps;
831 struct poll_list *walk = head;
832 unsigned long todo = nfds;
833
834 if (nfds > rlimit(RLIMIT_NOFILE))
835 return -EINVAL;
836
837 len = min_t(unsigned int, nfds, N_STACK_PPS);
838 for (;;) {
839 walk->next = NULL;
840 walk->len = len;
841 if (!len)
842 break;
843
844 if (copy_from_user(walk->entries, ufds + nfds-todo,
845 sizeof(struct pollfd) * walk->len))
846 goto out_fds;
847
848 todo -= walk->len;
849 if (!todo)
850 break;
851
852 len = min(todo, POLLFD_PER_PAGE);
853 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
854 walk = walk->next = kmalloc(size, GFP_KERNEL);
855 if (!walk) {
856 err = -ENOMEM;
857 goto out_fds;
858 }
859 }
860
861 poll_initwait(&table);
862 fdcount = do_poll(nfds, head, &table, end_time);
863 poll_freewait(&table);
864
865 for (walk = head; walk; walk = walk->next) {
866 struct pollfd *fds = walk->entries;
867 int j;
868
869 for (j = 0; j < walk->len; j++, ufds++)
870 if (__put_user(fds[j].revents, &ufds->revents))
871 goto out_fds;
872 }
873
874 err = fdcount;
875out_fds:
876 walk = head->next;
877 while (walk) {
878 struct poll_list *pos = walk;
879 walk = walk->next;
880 kfree(pos);
881 }
882
883 return err;
884}
885
886static long do_restart_poll(struct restart_block *restart_block)
887{
888 struct pollfd __user *ufds = restart_block->poll.ufds;
889 int nfds = restart_block->poll.nfds;
890 struct timespec *to = NULL, end_time;
891 int ret;
892
893 if (restart_block->poll.has_timeout) {
894 end_time.tv_sec = restart_block->poll.tv_sec;
895 end_time.tv_nsec = restart_block->poll.tv_nsec;
896 to = &end_time;
897 }
898
899 ret = do_sys_poll(ufds, nfds, to);
900
901 if (ret == -EINTR) {
902 restart_block->fn = do_restart_poll;
903 ret = -ERESTART_RESTARTBLOCK;
904 }
905 return ret;
906}
907
908SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
909 int, timeout_msecs)
910{
911 struct timespec end_time, *to = NULL;
912 int ret;
913
914 if (timeout_msecs >= 0) {
915 to = &end_time;
916 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
917 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
918 }
919
920 ret = do_sys_poll(ufds, nfds, to);
921
922 if (ret == -EINTR) {
923 struct restart_block *restart_block;
924
925 restart_block = ¤t_thread_info()->restart_block;
926 restart_block->fn = do_restart_poll;
927 restart_block->poll.ufds = ufds;
928 restart_block->poll.nfds = nfds;
929
930 if (timeout_msecs >= 0) {
931 restart_block->poll.tv_sec = end_time.tv_sec;
932 restart_block->poll.tv_nsec = end_time.tv_nsec;
933 restart_block->poll.has_timeout = 1;
934 } else
935 restart_block->poll.has_timeout = 0;
936
937 ret = -ERESTART_RESTARTBLOCK;
938 }
939 return ret;
940}
941
942SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
943 struct timespec __user *, tsp, const sigset_t __user *, sigmask,
944 size_t, sigsetsize)
945{
946 sigset_t ksigmask, sigsaved;
947 struct timespec ts, end_time, *to = NULL;
948 int ret;
949
950 if (tsp) {
951 if (copy_from_user(&ts, tsp, sizeof(ts)))
952 return -EFAULT;
953
954 to = &end_time;
955 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
956 return -EINVAL;
957 }
958
959 if (sigmask) {
960 /* XXX: Don't preclude handling different sized sigset_t's. */
961 if (sigsetsize != sizeof(sigset_t))
962 return -EINVAL;
963 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
964 return -EFAULT;
965
966 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
967 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
968 }
969
970 ret = do_sys_poll(ufds, nfds, to);
971
972 /* We can restart this syscall, usually */
973 if (ret == -EINTR) {
974 /*
975 * Don't restore the signal mask yet. Let do_signal() deliver
976 * the signal on the way back to userspace, before the signal
977 * mask is restored.
978 */
979 if (sigmask) {
980 memcpy(¤t->saved_sigmask, &sigsaved,
981 sizeof(sigsaved));
982 set_restore_sigmask();
983 }
984 ret = -ERESTARTNOHAND;
985 } else if (sigmask)
986 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
987
988 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
989
990 return ret;
991}