Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * This file contains the procedures for the handling of select and poll
4 *
5 * Created for Linux based loosely upon Mathius Lattner's minix
6 * patches by Peter MacDonald. Heavily edited by Linus.
7 *
8 * 4 February 1994
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.
12 *
13 * 24 January 2000
14 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
15 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
16 */
17
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>
29#include <linux/fs.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>
35
36#include <linux/uaccess.h>
37
38
39/*
40 * Estimate expected accuracy in ns from a timeval.
41 *
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.
46 *
47 * Consider this comment an open invitation to come up with even
48 * better solutions..
49 */
50
51#define MAX_SLACK (100 * NSEC_PER_MSEC)
52
53static long __estimate_accuracy(struct timespec64 *tv)
54{
55 long slack;
56 int divfactor = 1000;
57
58 if (tv->tv_sec < 0)
59 return 0;
60
61 if (task_nice(current) > 0)
62 divfactor = divfactor / 5;
63
64 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
65 return MAX_SLACK;
66
67 slack = tv->tv_nsec / divfactor;
68 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
69
70 if (slack > MAX_SLACK)
71 return MAX_SLACK;
72
73 return slack;
74}
75
76u64 select_estimate_accuracy(struct timespec64 *tv)
77{
78 u64 ret;
79 struct timespec64 now;
80 u64 slack = current->timer_slack_ns;
81
82 if (slack == 0)
83 return 0;
84
85 ktime_get_ts64(&now);
86 now = timespec64_sub(*tv, now);
87 ret = __estimate_accuracy(&now);
88 if (ret < slack)
89 return slack;
90 return ret;
91}
92
93
94
95struct poll_table_page {
96 struct poll_table_page * next;
97 struct poll_table_entry * entry;
98 struct poll_table_entry entries[];
99};
100
101#define POLL_TABLE_FULL(table) \
102 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
103
104/*
105 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
106 * I have rewritten this, taking some shortcuts: This code may not be easy to
107 * follow, but it should be free of race-conditions, and it's practical. If you
108 * understand what I'm doing here, then you understand how the linux
109 * sleep/wakeup mechanism works.
110 *
111 * Two very simple procedures, poll_wait() and poll_freewait() make all the
112 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
113 * as all select/poll functions have to call it to add an entry to the
114 * poll table.
115 */
116static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
117 poll_table *p);
118
119void poll_initwait(struct poll_wqueues *pwq)
120{
121 init_poll_funcptr(&pwq->pt, __pollwait);
122 pwq->polling_task = current;
123 pwq->triggered = 0;
124 pwq->error = 0;
125 pwq->table = NULL;
126 pwq->inline_index = 0;
127}
128EXPORT_SYMBOL(poll_initwait);
129
130static void free_poll_entry(struct poll_table_entry *entry)
131{
132 remove_wait_queue(entry->wait_address, &entry->wait);
133 fput(entry->filp);
134}
135
136void poll_freewait(struct poll_wqueues *pwq)
137{
138 struct poll_table_page * p = pwq->table;
139 int i;
140 for (i = 0; i < pwq->inline_index; i++)
141 free_poll_entry(pwq->inline_entries + i);
142 while (p) {
143 struct poll_table_entry * entry;
144 struct poll_table_page *old;
145
146 entry = p->entry;
147 do {
148 entry--;
149 free_poll_entry(entry);
150 } while (entry > p->entries);
151 old = p;
152 p = p->next;
153 free_page((unsigned long) old);
154 }
155}
156EXPORT_SYMBOL(poll_freewait);
157
158static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
159{
160 struct poll_table_page *table = p->table;
161
162 if (p->inline_index < N_INLINE_POLL_ENTRIES)
163 return p->inline_entries + p->inline_index++;
164
165 if (!table || POLL_TABLE_FULL(table)) {
166 struct poll_table_page *new_table;
167
168 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
169 if (!new_table) {
170 p->error = -ENOMEM;
171 return NULL;
172 }
173 new_table->entry = new_table->entries;
174 new_table->next = table;
175 p->table = new_table;
176 table = new_table;
177 }
178
179 return table->entry++;
180}
181
182static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
183{
184 struct poll_wqueues *pwq = wait->private;
185 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
186
187 /*
188 * Although this function is called under waitqueue lock, LOCK
189 * doesn't imply write barrier and the users expect write
190 * barrier semantics on wakeup functions. The following
191 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
192 * and is paired with smp_store_mb() in poll_schedule_timeout.
193 */
194 smp_wmb();
195 pwq->triggered = 1;
196
197 /*
198 * Perform the default wake up operation using a dummy
199 * waitqueue.
200 *
201 * TODO: This is hacky but there currently is no interface to
202 * pass in @sync. @sync is scheduled to be removed and once
203 * that happens, wake_up_process() can be used directly.
204 */
205 return default_wake_function(&dummy_wait, mode, sync, key);
206}
207
208static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
209{
210 struct poll_table_entry *entry;
211
212 entry = container_of(wait, struct poll_table_entry, wait);
213 if (key && !(key_to_poll(key) & entry->key))
214 return 0;
215 return __pollwake(wait, mode, sync, key);
216}
217
218/* Add a new entry */
219static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
220 poll_table *p)
221{
222 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
223 struct poll_table_entry *entry = poll_get_entry(pwq);
224 if (!entry)
225 return;
226 entry->filp = get_file(filp);
227 entry->wait_address = wait_address;
228 entry->key = p->_key;
229 init_waitqueue_func_entry(&entry->wait, pollwake);
230 entry->wait.private = pwq;
231 add_wait_queue(wait_address, &entry->wait);
232}
233
234static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
235 ktime_t *expires, unsigned long slack)
236{
237 int rc = -EINTR;
238
239 set_current_state(state);
240 if (!pwq->triggered)
241 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
242 __set_current_state(TASK_RUNNING);
243
244 /*
245 * Prepare for the next iteration.
246 *
247 * The following smp_store_mb() serves two purposes. First, it's
248 * the counterpart rmb of the wmb in pollwake() such that data
249 * written before wake up is always visible after wake up.
250 * Second, the full barrier guarantees that triggered clearing
251 * doesn't pass event check of the next iteration. Note that
252 * this problem doesn't exist for the first iteration as
253 * add_wait_queue() has full barrier semantics.
254 */
255 smp_store_mb(pwq->triggered, 0);
256
257 return rc;
258}
259
260/**
261 * poll_select_set_timeout - helper function to setup the timeout value
262 * @to: pointer to timespec64 variable for the final timeout
263 * @sec: seconds (from user space)
264 * @nsec: nanoseconds (from user space)
265 *
266 * Note, we do not use a timespec for the user space value here, That
267 * way we can use the function for timeval and compat interfaces as well.
268 *
269 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
270 */
271int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
272{
273 struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
274
275 if (!timespec64_valid(&ts))
276 return -EINVAL;
277
278 /* Optimize for the zero timeout value here */
279 if (!sec && !nsec) {
280 to->tv_sec = to->tv_nsec = 0;
281 } else {
282 ktime_get_ts64(to);
283 *to = timespec64_add_safe(*to, ts);
284 }
285 return 0;
286}
287
288enum poll_time_type {
289 PT_TIMEVAL = 0,
290 PT_OLD_TIMEVAL = 1,
291 PT_TIMESPEC = 2,
292 PT_OLD_TIMESPEC = 3,
293};
294
295static int poll_select_finish(struct timespec64 *end_time,
296 void __user *p,
297 enum poll_time_type pt_type, int ret)
298{
299 struct timespec64 rts;
300
301 restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
302
303 if (!p)
304 return ret;
305
306 if (current->personality & STICKY_TIMEOUTS)
307 goto sticky;
308
309 /* No update for zero timeout */
310 if (!end_time->tv_sec && !end_time->tv_nsec)
311 return ret;
312
313 ktime_get_ts64(&rts);
314 rts = timespec64_sub(*end_time, rts);
315 if (rts.tv_sec < 0)
316 rts.tv_sec = rts.tv_nsec = 0;
317
318
319 switch (pt_type) {
320 case PT_TIMEVAL:
321 {
322 struct __kernel_old_timeval rtv;
323
324 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
325 memset(&rtv, 0, sizeof(rtv));
326 rtv.tv_sec = rts.tv_sec;
327 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
328 if (!copy_to_user(p, &rtv, sizeof(rtv)))
329 return ret;
330 }
331 break;
332 case PT_OLD_TIMEVAL:
333 {
334 struct old_timeval32 rtv;
335
336 rtv.tv_sec = rts.tv_sec;
337 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
338 if (!copy_to_user(p, &rtv, sizeof(rtv)))
339 return ret;
340 }
341 break;
342 case PT_TIMESPEC:
343 if (!put_timespec64(&rts, p))
344 return ret;
345 break;
346 case PT_OLD_TIMESPEC:
347 if (!put_old_timespec32(&rts, p))
348 return ret;
349 break;
350 default:
351 BUG();
352 }
353 /*
354 * If an application puts its timeval in read-only memory, we
355 * don't want the Linux-specific update to the timeval to
356 * cause a fault after the select has completed
357 * successfully. However, because we're not updating the
358 * timeval, we can't restart the system call.
359 */
360
361sticky:
362 if (ret == -ERESTARTNOHAND)
363 ret = -EINTR;
364 return ret;
365}
366
367/*
368 * Scalable version of the fd_set.
369 */
370
371typedef struct {
372 unsigned long *in, *out, *ex;
373 unsigned long *res_in, *res_out, *res_ex;
374} fd_set_bits;
375
376/*
377 * How many longwords for "nr" bits?
378 */
379#define FDS_BITPERLONG (8*sizeof(long))
380#define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
381#define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
382
383/*
384 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
385 */
386static inline
387int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
388{
389 nr = FDS_BYTES(nr);
390 if (ufdset)
391 return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
392
393 memset(fdset, 0, nr);
394 return 0;
395}
396
397static inline unsigned long __must_check
398set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
399{
400 if (ufdset)
401 return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
402 return 0;
403}
404
405static inline
406void zero_fd_set(unsigned long nr, unsigned long *fdset)
407{
408 memset(fdset, 0, FDS_BYTES(nr));
409}
410
411#define FDS_IN(fds, n) (fds->in + n)
412#define FDS_OUT(fds, n) (fds->out + n)
413#define FDS_EX(fds, n) (fds->ex + n)
414
415#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
416
417static int max_select_fd(unsigned long n, fd_set_bits *fds)
418{
419 unsigned long *open_fds;
420 unsigned long set;
421 int max;
422 struct fdtable *fdt;
423
424 /* handle last in-complete long-word first */
425 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
426 n /= BITS_PER_LONG;
427 fdt = files_fdtable(current->files);
428 open_fds = fdt->open_fds + n;
429 max = 0;
430 if (set) {
431 set &= BITS(fds, n);
432 if (set) {
433 if (!(set & ~*open_fds))
434 goto get_max;
435 return -EBADF;
436 }
437 }
438 while (n) {
439 open_fds--;
440 n--;
441 set = BITS(fds, n);
442 if (!set)
443 continue;
444 if (set & ~*open_fds)
445 return -EBADF;
446 if (max)
447 continue;
448get_max:
449 do {
450 max++;
451 set >>= 1;
452 } while (set);
453 max += n * BITS_PER_LONG;
454 }
455
456 return max;
457}
458
459#define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR |\
460 EPOLLNVAL)
461#define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\
462 EPOLLNVAL)
463#define POLLEX_SET (EPOLLPRI | EPOLLNVAL)
464
465static inline __poll_t select_poll_one(int fd, poll_table *wait, unsigned long in,
466 unsigned long out, unsigned long bit,
467 __poll_t ll_flag)
468{
469 CLASS(fd, f)(fd);
470
471 if (fd_empty(f))
472 return EPOLLNVAL;
473
474 wait->_key = POLLEX_SET | ll_flag;
475 if (in & bit)
476 wait->_key |= POLLIN_SET;
477 if (out & bit)
478 wait->_key |= POLLOUT_SET;
479
480 return vfs_poll(fd_file(f), wait);
481}
482
483static noinline_for_stack int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
484{
485 ktime_t expire, *to = NULL;
486 struct poll_wqueues table;
487 poll_table *wait;
488 int retval, i, timed_out = 0;
489 u64 slack = 0;
490 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
491 unsigned long busy_start = 0;
492
493 rcu_read_lock();
494 retval = max_select_fd(n, fds);
495 rcu_read_unlock();
496
497 if (retval < 0)
498 return retval;
499 n = retval;
500
501 poll_initwait(&table);
502 wait = &table.pt;
503 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
504 wait->_qproc = NULL;
505 timed_out = 1;
506 }
507
508 if (end_time && !timed_out)
509 slack = select_estimate_accuracy(end_time);
510
511 retval = 0;
512 for (;;) {
513 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
514 bool can_busy_loop = false;
515
516 inp = fds->in; outp = fds->out; exp = fds->ex;
517 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
518
519 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
520 unsigned long in, out, ex, all_bits, bit = 1, j;
521 unsigned long res_in = 0, res_out = 0, res_ex = 0;
522 __poll_t mask;
523
524 in = *inp++; out = *outp++; ex = *exp++;
525 all_bits = in | out | ex;
526 if (all_bits == 0) {
527 i += BITS_PER_LONG;
528 continue;
529 }
530
531 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
532 if (i >= n)
533 break;
534 if (!(bit & all_bits))
535 continue;
536 mask = select_poll_one(i, wait, in, out, bit,
537 busy_flag);
538 if ((mask & POLLIN_SET) && (in & bit)) {
539 res_in |= bit;
540 retval++;
541 wait->_qproc = NULL;
542 }
543 if ((mask & POLLOUT_SET) && (out & bit)) {
544 res_out |= bit;
545 retval++;
546 wait->_qproc = NULL;
547 }
548 if ((mask & POLLEX_SET) && (ex & bit)) {
549 res_ex |= bit;
550 retval++;
551 wait->_qproc = NULL;
552 }
553 /* got something, stop busy polling */
554 if (retval) {
555 can_busy_loop = false;
556 busy_flag = 0;
557
558 /*
559 * only remember a returned
560 * POLL_BUSY_LOOP if we asked for it
561 */
562 } else if (busy_flag & mask)
563 can_busy_loop = true;
564
565 }
566 if (res_in)
567 *rinp = res_in;
568 if (res_out)
569 *routp = res_out;
570 if (res_ex)
571 *rexp = res_ex;
572 cond_resched();
573 }
574 wait->_qproc = NULL;
575 if (retval || timed_out || signal_pending(current))
576 break;
577 if (table.error) {
578 retval = table.error;
579 break;
580 }
581
582 /* only if found POLL_BUSY_LOOP sockets && not out of time */
583 if (can_busy_loop && !need_resched()) {
584 if (!busy_start) {
585 busy_start = busy_loop_current_time();
586 continue;
587 }
588 if (!busy_loop_timeout(busy_start))
589 continue;
590 }
591 busy_flag = 0;
592
593 /*
594 * If this is the first loop and we have a timeout
595 * given, then we convert to ktime_t and set the to
596 * pointer to the expiry value.
597 */
598 if (end_time && !to) {
599 expire = timespec64_to_ktime(*end_time);
600 to = &expire;
601 }
602
603 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
604 to, slack))
605 timed_out = 1;
606 }
607
608 poll_freewait(&table);
609
610 return retval;
611}
612
613/*
614 * We can actually return ERESTARTSYS instead of EINTR, but I'd
615 * like to be certain this leads to no problems. So I return
616 * EINTR just for safety.
617 *
618 * Update: ERESTARTSYS breaks at least the xview clock binary, so
619 * I'm trying ERESTARTNOHAND which restart only when you want to.
620 */
621int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
622 fd_set __user *exp, struct timespec64 *end_time)
623{
624 fd_set_bits fds;
625 void *bits;
626 int ret, max_fds;
627 size_t size, alloc_size;
628 struct fdtable *fdt;
629 /* Allocate small arguments on the stack to save memory and be faster */
630 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
631
632 ret = -EINVAL;
633 if (n < 0)
634 goto out_nofds;
635
636 /* max_fds can increase, so grab it once to avoid race */
637 rcu_read_lock();
638 fdt = files_fdtable(current->files);
639 max_fds = fdt->max_fds;
640 rcu_read_unlock();
641 if (n > max_fds)
642 n = max_fds;
643
644 /*
645 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
646 * since we used fdset we need to allocate memory in units of
647 * long-words.
648 */
649 size = FDS_BYTES(n);
650 bits = stack_fds;
651 if (size > sizeof(stack_fds) / 6) {
652 /* Not enough space in on-stack array; must use kmalloc */
653 ret = -ENOMEM;
654 if (size > (SIZE_MAX / 6))
655 goto out_nofds;
656
657 alloc_size = 6 * size;
658 bits = kvmalloc(alloc_size, GFP_KERNEL);
659 if (!bits)
660 goto out_nofds;
661 }
662 fds.in = bits;
663 fds.out = bits + size;
664 fds.ex = bits + 2*size;
665 fds.res_in = bits + 3*size;
666 fds.res_out = bits + 4*size;
667 fds.res_ex = bits + 5*size;
668
669 if ((ret = get_fd_set(n, inp, fds.in)) ||
670 (ret = get_fd_set(n, outp, fds.out)) ||
671 (ret = get_fd_set(n, exp, fds.ex)))
672 goto out;
673 zero_fd_set(n, fds.res_in);
674 zero_fd_set(n, fds.res_out);
675 zero_fd_set(n, fds.res_ex);
676
677 ret = do_select(n, &fds, end_time);
678
679 if (ret < 0)
680 goto out;
681 if (!ret) {
682 ret = -ERESTARTNOHAND;
683 if (signal_pending(current))
684 goto out;
685 ret = 0;
686 }
687
688 if (set_fd_set(n, inp, fds.res_in) ||
689 set_fd_set(n, outp, fds.res_out) ||
690 set_fd_set(n, exp, fds.res_ex))
691 ret = -EFAULT;
692
693out:
694 if (bits != stack_fds)
695 kvfree(bits);
696out_nofds:
697 return ret;
698}
699
700static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
701 fd_set __user *exp, struct __kernel_old_timeval __user *tvp)
702{
703 struct timespec64 end_time, *to = NULL;
704 struct __kernel_old_timeval tv;
705 int ret;
706
707 if (tvp) {
708 if (copy_from_user(&tv, tvp, sizeof(tv)))
709 return -EFAULT;
710
711 to = &end_time;
712 if (poll_select_set_timeout(to,
713 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
714 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
715 return -EINVAL;
716 }
717
718 ret = core_sys_select(n, inp, outp, exp, to);
719 return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
720}
721
722SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
723 fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp)
724{
725 return kern_select(n, inp, outp, exp, tvp);
726}
727
728static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
729 fd_set __user *exp, void __user *tsp,
730 const sigset_t __user *sigmask, size_t sigsetsize,
731 enum poll_time_type type)
732{
733 struct timespec64 ts, end_time, *to = NULL;
734 int ret;
735
736 if (tsp) {
737 switch (type) {
738 case PT_TIMESPEC:
739 if (get_timespec64(&ts, tsp))
740 return -EFAULT;
741 break;
742 case PT_OLD_TIMESPEC:
743 if (get_old_timespec32(&ts, tsp))
744 return -EFAULT;
745 break;
746 default:
747 BUG();
748 }
749
750 to = &end_time;
751 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
752 return -EINVAL;
753 }
754
755 ret = set_user_sigmask(sigmask, sigsetsize);
756 if (ret)
757 return ret;
758
759 ret = core_sys_select(n, inp, outp, exp, to);
760 return poll_select_finish(&end_time, tsp, type, ret);
761}
762
763/*
764 * Most architectures can't handle 7-argument syscalls. So we provide a
765 * 6-argument version where the sixth argument is a pointer to a structure
766 * which has a pointer to the sigset_t itself followed by a size_t containing
767 * the sigset size.
768 */
769struct sigset_argpack {
770 sigset_t __user *p;
771 size_t size;
772};
773
774static inline int get_sigset_argpack(struct sigset_argpack *to,
775 struct sigset_argpack __user *from)
776{
777 // the path is hot enough for overhead of copy_from_user() to matter
778 if (from) {
779 if (can_do_masked_user_access())
780 from = masked_user_access_begin(from);
781 else if (!user_read_access_begin(from, sizeof(*from)))
782 return -EFAULT;
783 unsafe_get_user(to->p, &from->p, Efault);
784 unsafe_get_user(to->size, &from->size, Efault);
785 user_read_access_end();
786 }
787 return 0;
788Efault:
789 user_read_access_end();
790 return -EFAULT;
791}
792
793SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
794 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
795 void __user *, sig)
796{
797 struct sigset_argpack x = {NULL, 0};
798
799 if (get_sigset_argpack(&x, sig))
800 return -EFAULT;
801
802 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
803}
804
805#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
806
807SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
808 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
809 void __user *, sig)
810{
811 struct sigset_argpack x = {NULL, 0};
812
813 if (get_sigset_argpack(&x, sig))
814 return -EFAULT;
815
816 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
817}
818
819#endif
820
821#ifdef __ARCH_WANT_SYS_OLD_SELECT
822struct sel_arg_struct {
823 unsigned long n;
824 fd_set __user *inp, *outp, *exp;
825 struct __kernel_old_timeval __user *tvp;
826};
827
828SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
829{
830 struct sel_arg_struct a;
831
832 if (copy_from_user(&a, arg, sizeof(a)))
833 return -EFAULT;
834 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
835}
836#endif
837
838struct poll_list {
839 struct poll_list *next;
840 unsigned int len;
841 struct pollfd entries[] __counted_by(len);
842};
843
844#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
845
846/*
847 * Fish for pollable events on the pollfd->fd file descriptor. We're only
848 * interested in events matching the pollfd->events mask, and the result
849 * matching that mask is both recorded in pollfd->revents and returned. The
850 * pwait poll_table will be used by the fd-provided poll handler for waiting,
851 * if pwait->_qproc is non-NULL.
852 */
853static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
854 bool *can_busy_poll,
855 __poll_t busy_flag)
856{
857 int fd = pollfd->fd;
858 __poll_t mask, filter;
859
860 if (fd < 0)
861 return 0;
862
863 CLASS(fd, f)(fd);
864 if (fd_empty(f))
865 return EPOLLNVAL;
866
867 /* userland u16 ->events contains POLL... bitmap */
868 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
869 pwait->_key = filter | busy_flag;
870 mask = vfs_poll(fd_file(f), pwait);
871 if (mask & busy_flag)
872 *can_busy_poll = true;
873 return mask & filter; /* Mask out unneeded events. */
874}
875
876static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
877 struct timespec64 *end_time)
878{
879 poll_table* pt = &wait->pt;
880 ktime_t expire, *to = NULL;
881 int timed_out = 0, count = 0;
882 u64 slack = 0;
883 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
884 unsigned long busy_start = 0;
885
886 /* Optimise the no-wait case */
887 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
888 pt->_qproc = NULL;
889 timed_out = 1;
890 }
891
892 if (end_time && !timed_out)
893 slack = select_estimate_accuracy(end_time);
894
895 for (;;) {
896 struct poll_list *walk;
897 bool can_busy_loop = false;
898
899 for (walk = list; walk != NULL; walk = walk->next) {
900 struct pollfd * pfd, * pfd_end;
901
902 pfd = walk->entries;
903 pfd_end = pfd + walk->len;
904 for (; pfd != pfd_end; pfd++) {
905 __poll_t mask;
906 /*
907 * Fish for events. If we found one, record it
908 * and kill poll_table->_qproc, so we don't
909 * needlessly register any other waiters after
910 * this. They'll get immediately deregistered
911 * when we break out and return.
912 */
913 mask = do_pollfd(pfd, pt, &can_busy_loop, busy_flag);
914 pfd->revents = mangle_poll(mask);
915 if (mask) {
916 count++;
917 pt->_qproc = NULL;
918 /* found something, stop busy polling */
919 busy_flag = 0;
920 can_busy_loop = false;
921 }
922 }
923 }
924 /*
925 * All waiters have already been registered, so don't provide
926 * a poll_table->_qproc to them on the next loop iteration.
927 */
928 pt->_qproc = NULL;
929 if (!count) {
930 count = wait->error;
931 if (signal_pending(current))
932 count = -ERESTARTNOHAND;
933 }
934 if (count || timed_out)
935 break;
936
937 /* only if found POLL_BUSY_LOOP sockets && not out of time */
938 if (can_busy_loop && !need_resched()) {
939 if (!busy_start) {
940 busy_start = busy_loop_current_time();
941 continue;
942 }
943 if (!busy_loop_timeout(busy_start))
944 continue;
945 }
946 busy_flag = 0;
947
948 /*
949 * If this is the first loop and we have a timeout
950 * given, then we convert to ktime_t and set the to
951 * pointer to the expiry value.
952 */
953 if (end_time && !to) {
954 expire = timespec64_to_ktime(*end_time);
955 to = &expire;
956 }
957
958 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
959 timed_out = 1;
960 }
961 return count;
962}
963
964#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
965 sizeof(struct pollfd))
966
967static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
968 struct timespec64 *end_time)
969{
970 struct poll_wqueues table;
971 int err = -EFAULT, fdcount;
972 /* Allocate small arguments on the stack to save memory and be
973 faster - use long to make sure the buffer is aligned properly
974 on 64 bit archs to avoid unaligned access */
975 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
976 struct poll_list *const head = (struct poll_list *)stack_pps;
977 struct poll_list *walk = head;
978 unsigned int todo = nfds;
979 unsigned int len;
980
981 if (nfds > rlimit(RLIMIT_NOFILE))
982 return -EINVAL;
983
984 len = min_t(unsigned int, nfds, N_STACK_PPS);
985 for (;;) {
986 walk->next = NULL;
987 walk->len = len;
988 if (!len)
989 break;
990
991 if (copy_from_user(walk->entries, ufds + nfds-todo,
992 sizeof(struct pollfd) * walk->len))
993 goto out_fds;
994
995 if (walk->len >= todo)
996 break;
997 todo -= walk->len;
998
999 len = min(todo, POLLFD_PER_PAGE);
1000 walk = walk->next = kmalloc(struct_size(walk, entries, len),
1001 GFP_KERNEL);
1002 if (!walk) {
1003 err = -ENOMEM;
1004 goto out_fds;
1005 }
1006 }
1007
1008 poll_initwait(&table);
1009 fdcount = do_poll(head, &table, end_time);
1010 poll_freewait(&table);
1011
1012 if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1013 goto out_fds;
1014
1015 for (walk = head; walk; walk = walk->next) {
1016 struct pollfd *fds = walk->entries;
1017 unsigned int j;
1018
1019 for (j = walk->len; j; fds++, ufds++, j--)
1020 unsafe_put_user(fds->revents, &ufds->revents, Efault);
1021 }
1022 user_write_access_end();
1023
1024 err = fdcount;
1025out_fds:
1026 walk = head->next;
1027 while (walk) {
1028 struct poll_list *pos = walk;
1029 walk = walk->next;
1030 kfree(pos);
1031 }
1032
1033 return err;
1034
1035Efault:
1036 user_write_access_end();
1037 err = -EFAULT;
1038 goto out_fds;
1039}
1040
1041static long do_restart_poll(struct restart_block *restart_block)
1042{
1043 struct pollfd __user *ufds = restart_block->poll.ufds;
1044 int nfds = restart_block->poll.nfds;
1045 struct timespec64 *to = NULL, end_time;
1046 int ret;
1047
1048 if (restart_block->poll.has_timeout) {
1049 end_time.tv_sec = restart_block->poll.tv_sec;
1050 end_time.tv_nsec = restart_block->poll.tv_nsec;
1051 to = &end_time;
1052 }
1053
1054 ret = do_sys_poll(ufds, nfds, to);
1055
1056 if (ret == -ERESTARTNOHAND)
1057 ret = set_restart_fn(restart_block, do_restart_poll);
1058
1059 return ret;
1060}
1061
1062SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1063 int, timeout_msecs)
1064{
1065 struct timespec64 end_time, *to = NULL;
1066 int ret;
1067
1068 if (timeout_msecs >= 0) {
1069 to = &end_time;
1070 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1071 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1072 }
1073
1074 ret = do_sys_poll(ufds, nfds, to);
1075
1076 if (ret == -ERESTARTNOHAND) {
1077 struct restart_block *restart_block;
1078
1079 restart_block = ¤t->restart_block;
1080 restart_block->poll.ufds = ufds;
1081 restart_block->poll.nfds = nfds;
1082
1083 if (timeout_msecs >= 0) {
1084 restart_block->poll.tv_sec = end_time.tv_sec;
1085 restart_block->poll.tv_nsec = end_time.tv_nsec;
1086 restart_block->poll.has_timeout = 1;
1087 } else
1088 restart_block->poll.has_timeout = 0;
1089
1090 ret = set_restart_fn(restart_block, do_restart_poll);
1091 }
1092 return ret;
1093}
1094
1095SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1096 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1097 size_t, sigsetsize)
1098{
1099 struct timespec64 ts, end_time, *to = NULL;
1100 int ret;
1101
1102 if (tsp) {
1103 if (get_timespec64(&ts, tsp))
1104 return -EFAULT;
1105
1106 to = &end_time;
1107 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1108 return -EINVAL;
1109 }
1110
1111 ret = set_user_sigmask(sigmask, sigsetsize);
1112 if (ret)
1113 return ret;
1114
1115 ret = do_sys_poll(ufds, nfds, to);
1116 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1117}
1118
1119#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1120
1121SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1122 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1123 size_t, sigsetsize)
1124{
1125 struct timespec64 ts, end_time, *to = NULL;
1126 int ret;
1127
1128 if (tsp) {
1129 if (get_old_timespec32(&ts, tsp))
1130 return -EFAULT;
1131
1132 to = &end_time;
1133 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1134 return -EINVAL;
1135 }
1136
1137 ret = set_user_sigmask(sigmask, sigsetsize);
1138 if (ret)
1139 return ret;
1140
1141 ret = do_sys_poll(ufds, nfds, to);
1142 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1143}
1144#endif
1145
1146#ifdef CONFIG_COMPAT
1147#define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1148
1149/*
1150 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1151 * 64-bit unsigned longs.
1152 */
1153static
1154int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1155 unsigned long *fdset)
1156{
1157 if (ufdset) {
1158 return compat_get_bitmap(fdset, ufdset, nr);
1159 } else {
1160 zero_fd_set(nr, fdset);
1161 return 0;
1162 }
1163}
1164
1165static
1166int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1167 unsigned long *fdset)
1168{
1169 if (!ufdset)
1170 return 0;
1171 return compat_put_bitmap(ufdset, fdset, nr);
1172}
1173
1174
1175/*
1176 * This is a virtual copy of sys_select from fs/select.c and probably
1177 * should be compared to it from time to time
1178 */
1179
1180/*
1181 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1182 * like to be certain this leads to no problems. So I return
1183 * EINTR just for safety.
1184 *
1185 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1186 * I'm trying ERESTARTNOHAND which restart only when you want to.
1187 */
1188static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1189 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1190 struct timespec64 *end_time)
1191{
1192 fd_set_bits fds;
1193 void *bits;
1194 int size, max_fds, ret = -EINVAL;
1195 struct fdtable *fdt;
1196 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1197
1198 if (n < 0)
1199 goto out_nofds;
1200
1201 /* max_fds can increase, so grab it once to avoid race */
1202 rcu_read_lock();
1203 fdt = files_fdtable(current->files);
1204 max_fds = fdt->max_fds;
1205 rcu_read_unlock();
1206 if (n > max_fds)
1207 n = max_fds;
1208
1209 /*
1210 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1211 * since we used fdset we need to allocate memory in units of
1212 * long-words.
1213 */
1214 size = FDS_BYTES(n);
1215 bits = stack_fds;
1216 if (size > sizeof(stack_fds) / 6) {
1217 bits = kmalloc_array(6, size, GFP_KERNEL);
1218 ret = -ENOMEM;
1219 if (!bits)
1220 goto out_nofds;
1221 }
1222 fds.in = (unsigned long *) bits;
1223 fds.out = (unsigned long *) (bits + size);
1224 fds.ex = (unsigned long *) (bits + 2*size);
1225 fds.res_in = (unsigned long *) (bits + 3*size);
1226 fds.res_out = (unsigned long *) (bits + 4*size);
1227 fds.res_ex = (unsigned long *) (bits + 5*size);
1228
1229 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1230 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1231 (ret = compat_get_fd_set(n, exp, fds.ex)))
1232 goto out;
1233 zero_fd_set(n, fds.res_in);
1234 zero_fd_set(n, fds.res_out);
1235 zero_fd_set(n, fds.res_ex);
1236
1237 ret = do_select(n, &fds, end_time);
1238
1239 if (ret < 0)
1240 goto out;
1241 if (!ret) {
1242 ret = -ERESTARTNOHAND;
1243 if (signal_pending(current))
1244 goto out;
1245 ret = 0;
1246 }
1247
1248 if (compat_set_fd_set(n, inp, fds.res_in) ||
1249 compat_set_fd_set(n, outp, fds.res_out) ||
1250 compat_set_fd_set(n, exp, fds.res_ex))
1251 ret = -EFAULT;
1252out:
1253 if (bits != stack_fds)
1254 kfree(bits);
1255out_nofds:
1256 return ret;
1257}
1258
1259static int do_compat_select(int n, compat_ulong_t __user *inp,
1260 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1261 struct old_timeval32 __user *tvp)
1262{
1263 struct timespec64 end_time, *to = NULL;
1264 struct old_timeval32 tv;
1265 int ret;
1266
1267 if (tvp) {
1268 if (copy_from_user(&tv, tvp, sizeof(tv)))
1269 return -EFAULT;
1270
1271 to = &end_time;
1272 if (poll_select_set_timeout(to,
1273 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1274 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1275 return -EINVAL;
1276 }
1277
1278 ret = compat_core_sys_select(n, inp, outp, exp, to);
1279 return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1280}
1281
1282COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1283 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1284 struct old_timeval32 __user *, tvp)
1285{
1286 return do_compat_select(n, inp, outp, exp, tvp);
1287}
1288
1289struct compat_sel_arg_struct {
1290 compat_ulong_t n;
1291 compat_uptr_t inp;
1292 compat_uptr_t outp;
1293 compat_uptr_t exp;
1294 compat_uptr_t tvp;
1295};
1296
1297COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1298{
1299 struct compat_sel_arg_struct a;
1300
1301 if (copy_from_user(&a, arg, sizeof(a)))
1302 return -EFAULT;
1303 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1304 compat_ptr(a.exp), compat_ptr(a.tvp));
1305}
1306
1307static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1308 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1309 void __user *tsp, compat_sigset_t __user *sigmask,
1310 compat_size_t sigsetsize, enum poll_time_type type)
1311{
1312 struct timespec64 ts, end_time, *to = NULL;
1313 int ret;
1314
1315 if (tsp) {
1316 switch (type) {
1317 case PT_OLD_TIMESPEC:
1318 if (get_old_timespec32(&ts, tsp))
1319 return -EFAULT;
1320 break;
1321 case PT_TIMESPEC:
1322 if (get_timespec64(&ts, tsp))
1323 return -EFAULT;
1324 break;
1325 default:
1326 BUG();
1327 }
1328
1329 to = &end_time;
1330 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1331 return -EINVAL;
1332 }
1333
1334 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1335 if (ret)
1336 return ret;
1337
1338 ret = compat_core_sys_select(n, inp, outp, exp, to);
1339 return poll_select_finish(&end_time, tsp, type, ret);
1340}
1341
1342struct compat_sigset_argpack {
1343 compat_uptr_t p;
1344 compat_size_t size;
1345};
1346static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1347 struct compat_sigset_argpack __user *from)
1348{
1349 if (from) {
1350 if (!user_read_access_begin(from, sizeof(*from)))
1351 return -EFAULT;
1352 unsafe_get_user(to->p, &from->p, Efault);
1353 unsafe_get_user(to->size, &from->size, Efault);
1354 user_read_access_end();
1355 }
1356 return 0;
1357Efault:
1358 user_read_access_end();
1359 return -EFAULT;
1360}
1361
1362COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1363 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1364 struct __kernel_timespec __user *, tsp, void __user *, sig)
1365{
1366 struct compat_sigset_argpack x = {0, 0};
1367
1368 if (get_compat_sigset_argpack(&x, sig))
1369 return -EFAULT;
1370
1371 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1372 x.size, PT_TIMESPEC);
1373}
1374
1375#if defined(CONFIG_COMPAT_32BIT_TIME)
1376
1377COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1378 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1379 struct old_timespec32 __user *, tsp, void __user *, sig)
1380{
1381 struct compat_sigset_argpack x = {0, 0};
1382
1383 if (get_compat_sigset_argpack(&x, sig))
1384 return -EFAULT;
1385
1386 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1387 x.size, PT_OLD_TIMESPEC);
1388}
1389
1390#endif
1391
1392#if defined(CONFIG_COMPAT_32BIT_TIME)
1393COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1394 unsigned int, nfds, struct old_timespec32 __user *, tsp,
1395 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1396{
1397 struct timespec64 ts, end_time, *to = NULL;
1398 int ret;
1399
1400 if (tsp) {
1401 if (get_old_timespec32(&ts, tsp))
1402 return -EFAULT;
1403
1404 to = &end_time;
1405 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1406 return -EINVAL;
1407 }
1408
1409 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1410 if (ret)
1411 return ret;
1412
1413 ret = do_sys_poll(ufds, nfds, to);
1414 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1415}
1416#endif
1417
1418/* New compat syscall for 64 bit time_t*/
1419COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1420 unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1421 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1422{
1423 struct timespec64 ts, end_time, *to = NULL;
1424 int ret;
1425
1426 if (tsp) {
1427 if (get_timespec64(&ts, tsp))
1428 return -EFAULT;
1429
1430 to = &end_time;
1431 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1432 return -EINVAL;
1433 }
1434
1435 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1436 if (ret)
1437 return ret;
1438
1439 ret = do_sys_poll(ufds, nfds, to);
1440 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1441}
1442
1443#endif
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/module.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 & (__NFDBITS-1)));
349 n /= __NFDBITS;
350 fdt = files_fdtable(current->files);
351 open_fds = fdt->open_fds->fds_bits+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 * __NFDBITS;
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 if (wait) {
390 wait->key = POLLEX_SET;
391 if (in & bit)
392 wait->key |= POLLIN_SET;
393 if (out & bit)
394 wait->key |= POLLOUT_SET;
395 }
396}
397
398int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
399{
400 ktime_t expire, *to = NULL;
401 struct poll_wqueues table;
402 poll_table *wait;
403 int retval, i, timed_out = 0;
404 unsigned long slack = 0;
405
406 rcu_read_lock();
407 retval = max_select_fd(n, fds);
408 rcu_read_unlock();
409
410 if (retval < 0)
411 return retval;
412 n = retval;
413
414 poll_initwait(&table);
415 wait = &table.pt;
416 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
417 wait = NULL;
418 timed_out = 1;
419 }
420
421 if (end_time && !timed_out)
422 slack = select_estimate_accuracy(end_time);
423
424 retval = 0;
425 for (;;) {
426 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
427
428 inp = fds->in; outp = fds->out; exp = fds->ex;
429 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
430
431 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
432 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
433 unsigned long res_in = 0, res_out = 0, res_ex = 0;
434 const struct file_operations *f_op = NULL;
435 struct file *file = NULL;
436
437 in = *inp++; out = *outp++; ex = *exp++;
438 all_bits = in | out | ex;
439 if (all_bits == 0) {
440 i += __NFDBITS;
441 continue;
442 }
443
444 for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
445 int fput_needed;
446 if (i >= n)
447 break;
448 if (!(bit & all_bits))
449 continue;
450 file = fget_light(i, &fput_needed);
451 if (file) {
452 f_op = file->f_op;
453 mask = DEFAULT_POLLMASK;
454 if (f_op && f_op->poll) {
455 wait_key_set(wait, in, out, bit);
456 mask = (*f_op->poll)(file, wait);
457 }
458 fput_light(file, fput_needed);
459 if ((mask & POLLIN_SET) && (in & bit)) {
460 res_in |= bit;
461 retval++;
462 wait = NULL;
463 }
464 if ((mask & POLLOUT_SET) && (out & bit)) {
465 res_out |= bit;
466 retval++;
467 wait = NULL;
468 }
469 if ((mask & POLLEX_SET) && (ex & bit)) {
470 res_ex |= bit;
471 retval++;
472 wait = NULL;
473 }
474 }
475 }
476 if (res_in)
477 *rinp = res_in;
478 if (res_out)
479 *routp = res_out;
480 if (res_ex)
481 *rexp = res_ex;
482 cond_resched();
483 }
484 wait = NULL;
485 if (retval || timed_out || signal_pending(current))
486 break;
487 if (table.error) {
488 retval = table.error;
489 break;
490 }
491
492 /*
493 * If this is the first loop and we have a timeout
494 * given, then we convert to ktime_t and set the to
495 * pointer to the expiry value.
496 */
497 if (end_time && !to) {
498 expire = timespec_to_ktime(*end_time);
499 to = &expire;
500 }
501
502 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
503 to, slack))
504 timed_out = 1;
505 }
506
507 poll_freewait(&table);
508
509 return retval;
510}
511
512/*
513 * We can actually return ERESTARTSYS instead of EINTR, but I'd
514 * like to be certain this leads to no problems. So I return
515 * EINTR just for safety.
516 *
517 * Update: ERESTARTSYS breaks at least the xview clock binary, so
518 * I'm trying ERESTARTNOHAND which restart only when you want to.
519 */
520int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
521 fd_set __user *exp, struct timespec *end_time)
522{
523 fd_set_bits fds;
524 void *bits;
525 int ret, max_fds;
526 unsigned int size;
527 struct fdtable *fdt;
528 /* Allocate small arguments on the stack to save memory and be faster */
529 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
530
531 ret = -EINVAL;
532 if (n < 0)
533 goto out_nofds;
534
535 /* max_fds can increase, so grab it once to avoid race */
536 rcu_read_lock();
537 fdt = files_fdtable(current->files);
538 max_fds = fdt->max_fds;
539 rcu_read_unlock();
540 if (n > max_fds)
541 n = max_fds;
542
543 /*
544 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
545 * since we used fdset we need to allocate memory in units of
546 * long-words.
547 */
548 size = FDS_BYTES(n);
549 bits = stack_fds;
550 if (size > sizeof(stack_fds) / 6) {
551 /* Not enough space in on-stack array; must use kmalloc */
552 ret = -ENOMEM;
553 bits = kmalloc(6 * size, GFP_KERNEL);
554 if (!bits)
555 goto out_nofds;
556 }
557 fds.in = bits;
558 fds.out = bits + size;
559 fds.ex = bits + 2*size;
560 fds.res_in = bits + 3*size;
561 fds.res_out = bits + 4*size;
562 fds.res_ex = bits + 5*size;
563
564 if ((ret = get_fd_set(n, inp, fds.in)) ||
565 (ret = get_fd_set(n, outp, fds.out)) ||
566 (ret = get_fd_set(n, exp, fds.ex)))
567 goto out;
568 zero_fd_set(n, fds.res_in);
569 zero_fd_set(n, fds.res_out);
570 zero_fd_set(n, fds.res_ex);
571
572 ret = do_select(n, &fds, end_time);
573
574 if (ret < 0)
575 goto out;
576 if (!ret) {
577 ret = -ERESTARTNOHAND;
578 if (signal_pending(current))
579 goto out;
580 ret = 0;
581 }
582
583 if (set_fd_set(n, inp, fds.res_in) ||
584 set_fd_set(n, outp, fds.res_out) ||
585 set_fd_set(n, exp, fds.res_ex))
586 ret = -EFAULT;
587
588out:
589 if (bits != stack_fds)
590 kfree(bits);
591out_nofds:
592 return ret;
593}
594
595SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
596 fd_set __user *, exp, struct timeval __user *, tvp)
597{
598 struct timespec end_time, *to = NULL;
599 struct timeval tv;
600 int ret;
601
602 if (tvp) {
603 if (copy_from_user(&tv, tvp, sizeof(tv)))
604 return -EFAULT;
605
606 to = &end_time;
607 if (poll_select_set_timeout(to,
608 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
609 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
610 return -EINVAL;
611 }
612
613 ret = core_sys_select(n, inp, outp, exp, to);
614 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
615
616 return ret;
617}
618
619#ifdef HAVE_SET_RESTORE_SIGMASK
620static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
621 fd_set __user *exp, struct timespec __user *tsp,
622 const sigset_t __user *sigmask, size_t sigsetsize)
623{
624 sigset_t ksigmask, sigsaved;
625 struct timespec ts, end_time, *to = NULL;
626 int ret;
627
628 if (tsp) {
629 if (copy_from_user(&ts, tsp, sizeof(ts)))
630 return -EFAULT;
631
632 to = &end_time;
633 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
634 return -EINVAL;
635 }
636
637 if (sigmask) {
638 /* XXX: Don't preclude handling different sized sigset_t's. */
639 if (sigsetsize != sizeof(sigset_t))
640 return -EINVAL;
641 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
642 return -EFAULT;
643
644 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
645 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
646 }
647
648 ret = core_sys_select(n, inp, outp, exp, to);
649 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
650
651 if (ret == -ERESTARTNOHAND) {
652 /*
653 * Don't restore the signal mask yet. Let do_signal() deliver
654 * the signal on the way back to userspace, before the signal
655 * mask is restored.
656 */
657 if (sigmask) {
658 memcpy(¤t->saved_sigmask, &sigsaved,
659 sizeof(sigsaved));
660 set_restore_sigmask();
661 }
662 } else if (sigmask)
663 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
664
665 return ret;
666}
667
668/*
669 * Most architectures can't handle 7-argument syscalls. So we provide a
670 * 6-argument version where the sixth argument is a pointer to a structure
671 * which has a pointer to the sigset_t itself followed by a size_t containing
672 * the sigset size.
673 */
674SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
675 fd_set __user *, exp, struct timespec __user *, tsp,
676 void __user *, sig)
677{
678 size_t sigsetsize = 0;
679 sigset_t __user *up = NULL;
680
681 if (sig) {
682 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
683 || __get_user(up, (sigset_t __user * __user *)sig)
684 || __get_user(sigsetsize,
685 (size_t __user *)(sig+sizeof(void *))))
686 return -EFAULT;
687 }
688
689 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
690}
691#endif /* HAVE_SET_RESTORE_SIGMASK */
692
693#ifdef __ARCH_WANT_SYS_OLD_SELECT
694struct sel_arg_struct {
695 unsigned long n;
696 fd_set __user *inp, *outp, *exp;
697 struct timeval __user *tvp;
698};
699
700SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
701{
702 struct sel_arg_struct a;
703
704 if (copy_from_user(&a, arg, sizeof(a)))
705 return -EFAULT;
706 return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
707}
708#endif
709
710struct poll_list {
711 struct poll_list *next;
712 int len;
713 struct pollfd entries[0];
714};
715
716#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
717
718/*
719 * Fish for pollable events on the pollfd->fd file descriptor. We're only
720 * interested in events matching the pollfd->events mask, and the result
721 * matching that mask is both recorded in pollfd->revents and returned. The
722 * pwait poll_table will be used by the fd-provided poll handler for waiting,
723 * if non-NULL.
724 */
725static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
726{
727 unsigned int mask;
728 int fd;
729
730 mask = 0;
731 fd = pollfd->fd;
732 if (fd >= 0) {
733 int fput_needed;
734 struct file * file;
735
736 file = fget_light(fd, &fput_needed);
737 mask = POLLNVAL;
738 if (file != NULL) {
739 mask = DEFAULT_POLLMASK;
740 if (file->f_op && file->f_op->poll) {
741 if (pwait)
742 pwait->key = pollfd->events |
743 POLLERR | POLLHUP;
744 mask = file->f_op->poll(file, pwait);
745 }
746 /* Mask out unneeded events. */
747 mask &= pollfd->events | POLLERR | POLLHUP;
748 fput_light(file, fput_needed);
749 }
750 }
751 pollfd->revents = mask;
752
753 return mask;
754}
755
756static int do_poll(unsigned int nfds, struct poll_list *list,
757 struct poll_wqueues *wait, struct timespec *end_time)
758{
759 poll_table* pt = &wait->pt;
760 ktime_t expire, *to = NULL;
761 int timed_out = 0, count = 0;
762 unsigned long slack = 0;
763
764 /* Optimise the no-wait case */
765 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
766 pt = NULL;
767 timed_out = 1;
768 }
769
770 if (end_time && !timed_out)
771 slack = select_estimate_accuracy(end_time);
772
773 for (;;) {
774 struct poll_list *walk;
775
776 for (walk = list; walk != NULL; walk = walk->next) {
777 struct pollfd * pfd, * pfd_end;
778
779 pfd = walk->entries;
780 pfd_end = pfd + walk->len;
781 for (; pfd != pfd_end; pfd++) {
782 /*
783 * Fish for events. If we found one, record it
784 * and kill the poll_table, so we don't
785 * needlessly register any other waiters after
786 * this. They'll get immediately deregistered
787 * when we break out and return.
788 */
789 if (do_pollfd(pfd, pt)) {
790 count++;
791 pt = NULL;
792 }
793 }
794 }
795 /*
796 * All waiters have already been registered, so don't provide
797 * a poll_table to them on the next loop iteration.
798 */
799 pt = NULL;
800 if (!count) {
801 count = wait->error;
802 if (signal_pending(current))
803 count = -EINTR;
804 }
805 if (count || timed_out)
806 break;
807
808 /*
809 * If this is the first loop and we have a timeout
810 * given, then we convert to ktime_t and set the to
811 * pointer to the expiry value.
812 */
813 if (end_time && !to) {
814 expire = timespec_to_ktime(*end_time);
815 to = &expire;
816 }
817
818 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
819 timed_out = 1;
820 }
821 return count;
822}
823
824#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
825 sizeof(struct pollfd))
826
827int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
828 struct timespec *end_time)
829{
830 struct poll_wqueues table;
831 int err = -EFAULT, fdcount, len, size;
832 /* Allocate small arguments on the stack to save memory and be
833 faster - use long to make sure the buffer is aligned properly
834 on 64 bit archs to avoid unaligned access */
835 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
836 struct poll_list *const head = (struct poll_list *)stack_pps;
837 struct poll_list *walk = head;
838 unsigned long todo = nfds;
839
840 if (nfds > rlimit(RLIMIT_NOFILE))
841 return -EINVAL;
842
843 len = min_t(unsigned int, nfds, N_STACK_PPS);
844 for (;;) {
845 walk->next = NULL;
846 walk->len = len;
847 if (!len)
848 break;
849
850 if (copy_from_user(walk->entries, ufds + nfds-todo,
851 sizeof(struct pollfd) * walk->len))
852 goto out_fds;
853
854 todo -= walk->len;
855 if (!todo)
856 break;
857
858 len = min(todo, POLLFD_PER_PAGE);
859 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
860 walk = walk->next = kmalloc(size, GFP_KERNEL);
861 if (!walk) {
862 err = -ENOMEM;
863 goto out_fds;
864 }
865 }
866
867 poll_initwait(&table);
868 fdcount = do_poll(nfds, head, &table, end_time);
869 poll_freewait(&table);
870
871 for (walk = head; walk; walk = walk->next) {
872 struct pollfd *fds = walk->entries;
873 int j;
874
875 for (j = 0; j < walk->len; j++, ufds++)
876 if (__put_user(fds[j].revents, &ufds->revents))
877 goto out_fds;
878 }
879
880 err = fdcount;
881out_fds:
882 walk = head->next;
883 while (walk) {
884 struct poll_list *pos = walk;
885 walk = walk->next;
886 kfree(pos);
887 }
888
889 return err;
890}
891
892static long do_restart_poll(struct restart_block *restart_block)
893{
894 struct pollfd __user *ufds = restart_block->poll.ufds;
895 int nfds = restart_block->poll.nfds;
896 struct timespec *to = NULL, end_time;
897 int ret;
898
899 if (restart_block->poll.has_timeout) {
900 end_time.tv_sec = restart_block->poll.tv_sec;
901 end_time.tv_nsec = restart_block->poll.tv_nsec;
902 to = &end_time;
903 }
904
905 ret = do_sys_poll(ufds, nfds, to);
906
907 if (ret == -EINTR) {
908 restart_block->fn = do_restart_poll;
909 ret = -ERESTART_RESTARTBLOCK;
910 }
911 return ret;
912}
913
914SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
915 long, timeout_msecs)
916{
917 struct timespec end_time, *to = NULL;
918 int ret;
919
920 if (timeout_msecs >= 0) {
921 to = &end_time;
922 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
923 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
924 }
925
926 ret = do_sys_poll(ufds, nfds, to);
927
928 if (ret == -EINTR) {
929 struct restart_block *restart_block;
930
931 restart_block = ¤t_thread_info()->restart_block;
932 restart_block->fn = do_restart_poll;
933 restart_block->poll.ufds = ufds;
934 restart_block->poll.nfds = nfds;
935
936 if (timeout_msecs >= 0) {
937 restart_block->poll.tv_sec = end_time.tv_sec;
938 restart_block->poll.tv_nsec = end_time.tv_nsec;
939 restart_block->poll.has_timeout = 1;
940 } else
941 restart_block->poll.has_timeout = 0;
942
943 ret = -ERESTART_RESTARTBLOCK;
944 }
945 return ret;
946}
947
948#ifdef HAVE_SET_RESTORE_SIGMASK
949SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
950 struct timespec __user *, tsp, const sigset_t __user *, sigmask,
951 size_t, sigsetsize)
952{
953 sigset_t ksigmask, sigsaved;
954 struct timespec ts, end_time, *to = NULL;
955 int ret;
956
957 if (tsp) {
958 if (copy_from_user(&ts, tsp, sizeof(ts)))
959 return -EFAULT;
960
961 to = &end_time;
962 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
963 return -EINVAL;
964 }
965
966 if (sigmask) {
967 /* XXX: Don't preclude handling different sized sigset_t's. */
968 if (sigsetsize != sizeof(sigset_t))
969 return -EINVAL;
970 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
971 return -EFAULT;
972
973 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
974 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
975 }
976
977 ret = do_sys_poll(ufds, nfds, to);
978
979 /* We can restart this syscall, usually */
980 if (ret == -EINTR) {
981 /*
982 * Don't restore the signal mask yet. Let do_signal() deliver
983 * the signal on the way back to userspace, before the signal
984 * mask is restored.
985 */
986 if (sigmask) {
987 memcpy(¤t->saved_sigmask, &sigsaved,
988 sizeof(sigsaved));
989 set_restore_sigmask();
990 }
991 ret = -ERESTARTNOHAND;
992 } else if (sigmask)
993 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
994
995 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
996
997 return ret;
998}
999#endif /* HAVE_SET_RESTORE_SIGMASK */