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/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 */
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
81 /*
82 * Realtime tasks get a slack of 0 for obvious reasons.
83 */
84
85 if (rt_task(current))
86 return 0;
87
88 ktime_get_ts64(&now);
89 now = timespec64_sub(*tv, now);
90 ret = __estimate_accuracy(&now);
91 if (ret < current->timer_slack_ns)
92 return current->timer_slack_ns;
93 return ret;
94}
95
96
97
98struct poll_table_page {
99 struct poll_table_page * next;
100 struct poll_table_entry * entry;
101 struct poll_table_entry entries[];
102};
103
104#define POLL_TABLE_FULL(table) \
105 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
106
107/*
108 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
109 * I have rewritten this, taking some shortcuts: This code may not be easy to
110 * follow, but it should be free of race-conditions, and it's practical. If you
111 * understand what I'm doing here, then you understand how the linux
112 * sleep/wakeup mechanism works.
113 *
114 * Two very simple procedures, poll_wait() and poll_freewait() make all the
115 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
116 * as all select/poll functions have to call it to add an entry to the
117 * poll table.
118 */
119static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
120 poll_table *p);
121
122void poll_initwait(struct poll_wqueues *pwq)
123{
124 init_poll_funcptr(&pwq->pt, __pollwait);
125 pwq->polling_task = current;
126 pwq->triggered = 0;
127 pwq->error = 0;
128 pwq->table = NULL;
129 pwq->inline_index = 0;
130}
131EXPORT_SYMBOL(poll_initwait);
132
133static void free_poll_entry(struct poll_table_entry *entry)
134{
135 remove_wait_queue(entry->wait_address, &entry->wait);
136 fput(entry->filp);
137}
138
139void poll_freewait(struct poll_wqueues *pwq)
140{
141 struct poll_table_page * p = pwq->table;
142 int i;
143 for (i = 0; i < pwq->inline_index; i++)
144 free_poll_entry(pwq->inline_entries + i);
145 while (p) {
146 struct poll_table_entry * entry;
147 struct poll_table_page *old;
148
149 entry = p->entry;
150 do {
151 entry--;
152 free_poll_entry(entry);
153 } while (entry > p->entries);
154 old = p;
155 p = p->next;
156 free_page((unsigned long) old);
157 }
158}
159EXPORT_SYMBOL(poll_freewait);
160
161static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
162{
163 struct poll_table_page *table = p->table;
164
165 if (p->inline_index < N_INLINE_POLL_ENTRIES)
166 return p->inline_entries + p->inline_index++;
167
168 if (!table || POLL_TABLE_FULL(table)) {
169 struct poll_table_page *new_table;
170
171 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
172 if (!new_table) {
173 p->error = -ENOMEM;
174 return NULL;
175 }
176 new_table->entry = new_table->entries;
177 new_table->next = table;
178 p->table = new_table;
179 table = new_table;
180 }
181
182 return table->entry++;
183}
184
185static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
186{
187 struct poll_wqueues *pwq = wait->private;
188 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
189
190 /*
191 * Although this function is called under waitqueue lock, LOCK
192 * doesn't imply write barrier and the users expect write
193 * barrier semantics on wakeup functions. The following
194 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
195 * and is paired with smp_store_mb() in poll_schedule_timeout.
196 */
197 smp_wmb();
198 pwq->triggered = 1;
199
200 /*
201 * Perform the default wake up operation using a dummy
202 * waitqueue.
203 *
204 * TODO: This is hacky but there currently is no interface to
205 * pass in @sync. @sync is scheduled to be removed and once
206 * that happens, wake_up_process() can be used directly.
207 */
208 return default_wake_function(&dummy_wait, mode, sync, key);
209}
210
211static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
212{
213 struct poll_table_entry *entry;
214
215 entry = container_of(wait, struct poll_table_entry, wait);
216 if (key && !(key_to_poll(key) & entry->key))
217 return 0;
218 return __pollwake(wait, mode, sync, key);
219}
220
221/* Add a new entry */
222static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
223 poll_table *p)
224{
225 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
226 struct poll_table_entry *entry = poll_get_entry(pwq);
227 if (!entry)
228 return;
229 entry->filp = get_file(filp);
230 entry->wait_address = wait_address;
231 entry->key = p->_key;
232 init_waitqueue_func_entry(&entry->wait, pollwake);
233 entry->wait.private = pwq;
234 add_wait_queue(wait_address, &entry->wait);
235}
236
237static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
238 ktime_t *expires, unsigned long slack)
239{
240 int rc = -EINTR;
241
242 set_current_state(state);
243 if (!pwq->triggered)
244 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
245 __set_current_state(TASK_RUNNING);
246
247 /*
248 * Prepare for the next iteration.
249 *
250 * The following smp_store_mb() serves two purposes. First, it's
251 * the counterpart rmb of the wmb in pollwake() such that data
252 * written before wake up is always visible after wake up.
253 * Second, the full barrier guarantees that triggered clearing
254 * doesn't pass event check of the next iteration. Note that
255 * this problem doesn't exist for the first iteration as
256 * add_wait_queue() has full barrier semantics.
257 */
258 smp_store_mb(pwq->triggered, 0);
259
260 return rc;
261}
262
263/**
264 * poll_select_set_timeout - helper function to setup the timeout value
265 * @to: pointer to timespec64 variable for the final timeout
266 * @sec: seconds (from user space)
267 * @nsec: nanoseconds (from user space)
268 *
269 * Note, we do not use a timespec for the user space value here, That
270 * way we can use the function for timeval and compat interfaces as well.
271 *
272 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
273 */
274int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
275{
276 struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
277
278 if (!timespec64_valid(&ts))
279 return -EINVAL;
280
281 /* Optimize for the zero timeout value here */
282 if (!sec && !nsec) {
283 to->tv_sec = to->tv_nsec = 0;
284 } else {
285 ktime_get_ts64(to);
286 *to = timespec64_add_safe(*to, ts);
287 }
288 return 0;
289}
290
291enum poll_time_type {
292 PT_TIMEVAL = 0,
293 PT_OLD_TIMEVAL = 1,
294 PT_TIMESPEC = 2,
295 PT_OLD_TIMESPEC = 3,
296};
297
298static int poll_select_finish(struct timespec64 *end_time,
299 void __user *p,
300 enum poll_time_type pt_type, int ret)
301{
302 struct timespec64 rts;
303
304 restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
305
306 if (!p)
307 return ret;
308
309 if (current->personality & STICKY_TIMEOUTS)
310 goto sticky;
311
312 /* No update for zero timeout */
313 if (!end_time->tv_sec && !end_time->tv_nsec)
314 return ret;
315
316 ktime_get_ts64(&rts);
317 rts = timespec64_sub(*end_time, rts);
318 if (rts.tv_sec < 0)
319 rts.tv_sec = rts.tv_nsec = 0;
320
321
322 switch (pt_type) {
323 case PT_TIMEVAL:
324 {
325 struct __kernel_old_timeval rtv;
326
327 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
328 memset(&rtv, 0, sizeof(rtv));
329 rtv.tv_sec = rts.tv_sec;
330 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
331 if (!copy_to_user(p, &rtv, sizeof(rtv)))
332 return ret;
333 }
334 break;
335 case PT_OLD_TIMEVAL:
336 {
337 struct old_timeval32 rtv;
338
339 rtv.tv_sec = rts.tv_sec;
340 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
341 if (!copy_to_user(p, &rtv, sizeof(rtv)))
342 return ret;
343 }
344 break;
345 case PT_TIMESPEC:
346 if (!put_timespec64(&rts, p))
347 return ret;
348 break;
349 case PT_OLD_TIMESPEC:
350 if (!put_old_timespec32(&rts, p))
351 return ret;
352 break;
353 default:
354 BUG();
355 }
356 /*
357 * If an application puts its timeval in read-only memory, we
358 * don't want the Linux-specific update to the timeval to
359 * cause a fault after the select has completed
360 * successfully. However, because we're not updating the
361 * timeval, we can't restart the system call.
362 */
363
364sticky:
365 if (ret == -ERESTARTNOHAND)
366 ret = -EINTR;
367 return ret;
368}
369
370/*
371 * Scalable version of the fd_set.
372 */
373
374typedef struct {
375 unsigned long *in, *out, *ex;
376 unsigned long *res_in, *res_out, *res_ex;
377} fd_set_bits;
378
379/*
380 * How many longwords for "nr" bits?
381 */
382#define FDS_BITPERLONG (8*sizeof(long))
383#define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
384#define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
385
386/*
387 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
388 */
389static inline
390int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
391{
392 nr = FDS_BYTES(nr);
393 if (ufdset)
394 return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
395
396 memset(fdset, 0, nr);
397 return 0;
398}
399
400static inline unsigned long __must_check
401set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
402{
403 if (ufdset)
404 return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
405 return 0;
406}
407
408static inline
409void zero_fd_set(unsigned long nr, unsigned long *fdset)
410{
411 memset(fdset, 0, FDS_BYTES(nr));
412}
413
414#define FDS_IN(fds, n) (fds->in + n)
415#define FDS_OUT(fds, n) (fds->out + n)
416#define FDS_EX(fds, n) (fds->ex + n)
417
418#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
419
420static int max_select_fd(unsigned long n, fd_set_bits *fds)
421{
422 unsigned long *open_fds;
423 unsigned long set;
424 int max;
425 struct fdtable *fdt;
426
427 /* handle last in-complete long-word first */
428 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
429 n /= BITS_PER_LONG;
430 fdt = files_fdtable(current->files);
431 open_fds = fdt->open_fds + n;
432 max = 0;
433 if (set) {
434 set &= BITS(fds, n);
435 if (set) {
436 if (!(set & ~*open_fds))
437 goto get_max;
438 return -EBADF;
439 }
440 }
441 while (n) {
442 open_fds--;
443 n--;
444 set = BITS(fds, n);
445 if (!set)
446 continue;
447 if (set & ~*open_fds)
448 return -EBADF;
449 if (max)
450 continue;
451get_max:
452 do {
453 max++;
454 set >>= 1;
455 } while (set);
456 max += n * BITS_PER_LONG;
457 }
458
459 return max;
460}
461
462#define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR |\
463 EPOLLNVAL)
464#define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\
465 EPOLLNVAL)
466#define POLLEX_SET (EPOLLPRI | EPOLLNVAL)
467
468static inline void wait_key_set(poll_table *wait, unsigned long in,
469 unsigned long out, unsigned long bit,
470 __poll_t ll_flag)
471{
472 wait->_key = POLLEX_SET | ll_flag;
473 if (in & bit)
474 wait->_key |= POLLIN_SET;
475 if (out & bit)
476 wait->_key |= POLLOUT_SET;
477}
478
479static noinline_for_stack int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
480{
481 ktime_t expire, *to = NULL;
482 struct poll_wqueues table;
483 poll_table *wait;
484 int retval, i, timed_out = 0;
485 u64 slack = 0;
486 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
487 unsigned long busy_start = 0;
488
489 rcu_read_lock();
490 retval = max_select_fd(n, fds);
491 rcu_read_unlock();
492
493 if (retval < 0)
494 return retval;
495 n = retval;
496
497 poll_initwait(&table);
498 wait = &table.pt;
499 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
500 wait->_qproc = NULL;
501 timed_out = 1;
502 }
503
504 if (end_time && !timed_out)
505 slack = select_estimate_accuracy(end_time);
506
507 retval = 0;
508 for (;;) {
509 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
510 bool can_busy_loop = false;
511
512 inp = fds->in; outp = fds->out; exp = fds->ex;
513 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
514
515 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
516 unsigned long in, out, ex, all_bits, bit = 1, j;
517 unsigned long res_in = 0, res_out = 0, res_ex = 0;
518 __poll_t mask;
519
520 in = *inp++; out = *outp++; ex = *exp++;
521 all_bits = in | out | ex;
522 if (all_bits == 0) {
523 i += BITS_PER_LONG;
524 continue;
525 }
526
527 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
528 struct fd f;
529 if (i >= n)
530 break;
531 if (!(bit & all_bits))
532 continue;
533 mask = EPOLLNVAL;
534 f = fdget(i);
535 if (f.file) {
536 wait_key_set(wait, in, out, bit,
537 busy_flag);
538 mask = vfs_poll(f.file, wait);
539
540 fdput(f);
541 }
542 if ((mask & POLLIN_SET) && (in & bit)) {
543 res_in |= bit;
544 retval++;
545 wait->_qproc = NULL;
546 }
547 if ((mask & POLLOUT_SET) && (out & bit)) {
548 res_out |= bit;
549 retval++;
550 wait->_qproc = NULL;
551 }
552 if ((mask & POLLEX_SET) && (ex & bit)) {
553 res_ex |= bit;
554 retval++;
555 wait->_qproc = NULL;
556 }
557 /* got something, stop busy polling */
558 if (retval) {
559 can_busy_loop = false;
560 busy_flag = 0;
561
562 /*
563 * only remember a returned
564 * POLL_BUSY_LOOP if we asked for it
565 */
566 } else if (busy_flag & mask)
567 can_busy_loop = true;
568
569 }
570 if (res_in)
571 *rinp = res_in;
572 if (res_out)
573 *routp = res_out;
574 if (res_ex)
575 *rexp = res_ex;
576 cond_resched();
577 }
578 wait->_qproc = NULL;
579 if (retval || timed_out || signal_pending(current))
580 break;
581 if (table.error) {
582 retval = table.error;
583 break;
584 }
585
586 /* only if found POLL_BUSY_LOOP sockets && not out of time */
587 if (can_busy_loop && !need_resched()) {
588 if (!busy_start) {
589 busy_start = busy_loop_current_time();
590 continue;
591 }
592 if (!busy_loop_timeout(busy_start))
593 continue;
594 }
595 busy_flag = 0;
596
597 /*
598 * If this is the first loop and we have a timeout
599 * given, then we convert to ktime_t and set the to
600 * pointer to the expiry value.
601 */
602 if (end_time && !to) {
603 expire = timespec64_to_ktime(*end_time);
604 to = &expire;
605 }
606
607 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
608 to, slack))
609 timed_out = 1;
610 }
611
612 poll_freewait(&table);
613
614 return retval;
615}
616
617/*
618 * We can actually return ERESTARTSYS instead of EINTR, but I'd
619 * like to be certain this leads to no problems. So I return
620 * EINTR just for safety.
621 *
622 * Update: ERESTARTSYS breaks at least the xview clock binary, so
623 * I'm trying ERESTARTNOHAND which restart only when you want to.
624 */
625int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
626 fd_set __user *exp, struct timespec64 *end_time)
627{
628 fd_set_bits fds;
629 void *bits;
630 int ret, max_fds;
631 size_t size, alloc_size;
632 struct fdtable *fdt;
633 /* Allocate small arguments on the stack to save memory and be faster */
634 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
635
636 ret = -EINVAL;
637 if (n < 0)
638 goto out_nofds;
639
640 /* max_fds can increase, so grab it once to avoid race */
641 rcu_read_lock();
642 fdt = files_fdtable(current->files);
643 max_fds = fdt->max_fds;
644 rcu_read_unlock();
645 if (n > max_fds)
646 n = max_fds;
647
648 /*
649 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
650 * since we used fdset we need to allocate memory in units of
651 * long-words.
652 */
653 size = FDS_BYTES(n);
654 bits = stack_fds;
655 if (size > sizeof(stack_fds) / 6) {
656 /* Not enough space in on-stack array; must use kmalloc */
657 ret = -ENOMEM;
658 if (size > (SIZE_MAX / 6))
659 goto out_nofds;
660
661 alloc_size = 6 * size;
662 bits = kvmalloc(alloc_size, GFP_KERNEL);
663 if (!bits)
664 goto out_nofds;
665 }
666 fds.in = bits;
667 fds.out = bits + size;
668 fds.ex = bits + 2*size;
669 fds.res_in = bits + 3*size;
670 fds.res_out = bits + 4*size;
671 fds.res_ex = bits + 5*size;
672
673 if ((ret = get_fd_set(n, inp, fds.in)) ||
674 (ret = get_fd_set(n, outp, fds.out)) ||
675 (ret = get_fd_set(n, exp, fds.ex)))
676 goto out;
677 zero_fd_set(n, fds.res_in);
678 zero_fd_set(n, fds.res_out);
679 zero_fd_set(n, fds.res_ex);
680
681 ret = do_select(n, &fds, end_time);
682
683 if (ret < 0)
684 goto out;
685 if (!ret) {
686 ret = -ERESTARTNOHAND;
687 if (signal_pending(current))
688 goto out;
689 ret = 0;
690 }
691
692 if (set_fd_set(n, inp, fds.res_in) ||
693 set_fd_set(n, outp, fds.res_out) ||
694 set_fd_set(n, exp, fds.res_ex))
695 ret = -EFAULT;
696
697out:
698 if (bits != stack_fds)
699 kvfree(bits);
700out_nofds:
701 return ret;
702}
703
704static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
705 fd_set __user *exp, struct __kernel_old_timeval __user *tvp)
706{
707 struct timespec64 end_time, *to = NULL;
708 struct __kernel_old_timeval tv;
709 int ret;
710
711 if (tvp) {
712 if (copy_from_user(&tv, tvp, sizeof(tv)))
713 return -EFAULT;
714
715 to = &end_time;
716 if (poll_select_set_timeout(to,
717 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
718 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
719 return -EINVAL;
720 }
721
722 ret = core_sys_select(n, inp, outp, exp, to);
723 return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
724}
725
726SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
727 fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp)
728{
729 return kern_select(n, inp, outp, exp, tvp);
730}
731
732static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
733 fd_set __user *exp, void __user *tsp,
734 const sigset_t __user *sigmask, size_t sigsetsize,
735 enum poll_time_type type)
736{
737 struct timespec64 ts, end_time, *to = NULL;
738 int ret;
739
740 if (tsp) {
741 switch (type) {
742 case PT_TIMESPEC:
743 if (get_timespec64(&ts, tsp))
744 return -EFAULT;
745 break;
746 case PT_OLD_TIMESPEC:
747 if (get_old_timespec32(&ts, tsp))
748 return -EFAULT;
749 break;
750 default:
751 BUG();
752 }
753
754 to = &end_time;
755 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
756 return -EINVAL;
757 }
758
759 ret = set_user_sigmask(sigmask, sigsetsize);
760 if (ret)
761 return ret;
762
763 ret = core_sys_select(n, inp, outp, exp, to);
764 return poll_select_finish(&end_time, tsp, type, ret);
765}
766
767/*
768 * Most architectures can't handle 7-argument syscalls. So we provide a
769 * 6-argument version where the sixth argument is a pointer to a structure
770 * which has a pointer to the sigset_t itself followed by a size_t containing
771 * the sigset size.
772 */
773struct sigset_argpack {
774 sigset_t __user *p;
775 size_t size;
776};
777
778static inline int get_sigset_argpack(struct sigset_argpack *to,
779 struct sigset_argpack __user *from)
780{
781 // the path is hot enough for overhead of copy_from_user() to matter
782 if (from) {
783 if (!user_read_access_begin(from, sizeof(*from)))
784 return -EFAULT;
785 unsafe_get_user(to->p, &from->p, Efault);
786 unsafe_get_user(to->size, &from->size, Efault);
787 user_read_access_end();
788 }
789 return 0;
790Efault:
791 user_access_end();
792 return -EFAULT;
793}
794
795SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
796 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
797 void __user *, sig)
798{
799 struct sigset_argpack x = {NULL, 0};
800
801 if (get_sigset_argpack(&x, sig))
802 return -EFAULT;
803
804 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
805}
806
807#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
808
809SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
810 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
811 void __user *, sig)
812{
813 struct sigset_argpack x = {NULL, 0};
814
815 if (get_sigset_argpack(&x, sig))
816 return -EFAULT;
817
818 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
819}
820
821#endif
822
823#ifdef __ARCH_WANT_SYS_OLD_SELECT
824struct sel_arg_struct {
825 unsigned long n;
826 fd_set __user *inp, *outp, *exp;
827 struct __kernel_old_timeval __user *tvp;
828};
829
830SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
831{
832 struct sel_arg_struct a;
833
834 if (copy_from_user(&a, arg, sizeof(a)))
835 return -EFAULT;
836 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
837}
838#endif
839
840struct poll_list {
841 struct poll_list *next;
842 unsigned int len;
843 struct pollfd entries[];
844};
845
846#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
847
848/*
849 * Fish for pollable events on the pollfd->fd file descriptor. We're only
850 * interested in events matching the pollfd->events mask, and the result
851 * matching that mask is both recorded in pollfd->revents and returned. The
852 * pwait poll_table will be used by the fd-provided poll handler for waiting,
853 * if pwait->_qproc is non-NULL.
854 */
855static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
856 bool *can_busy_poll,
857 __poll_t busy_flag)
858{
859 int fd = pollfd->fd;
860 __poll_t mask = 0, filter;
861 struct fd f;
862
863 if (fd < 0)
864 goto out;
865 mask = EPOLLNVAL;
866 f = fdget(fd);
867 if (!f.file)
868 goto out;
869
870 /* userland u16 ->events contains POLL... bitmap */
871 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
872 pwait->_key = filter | busy_flag;
873 mask = vfs_poll(f.file, pwait);
874 if (mask & busy_flag)
875 *can_busy_poll = true;
876 mask &= filter; /* Mask out unneeded events. */
877 fdput(f);
878
879out:
880 /* ... and so does ->revents */
881 pollfd->revents = mangle_poll(mask);
882 return mask;
883}
884
885static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
886 struct timespec64 *end_time)
887{
888 poll_table* pt = &wait->pt;
889 ktime_t expire, *to = NULL;
890 int timed_out = 0, count = 0;
891 u64 slack = 0;
892 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
893 unsigned long busy_start = 0;
894
895 /* Optimise the no-wait case */
896 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
897 pt->_qproc = NULL;
898 timed_out = 1;
899 }
900
901 if (end_time && !timed_out)
902 slack = select_estimate_accuracy(end_time);
903
904 for (;;) {
905 struct poll_list *walk;
906 bool can_busy_loop = false;
907
908 for (walk = list; walk != NULL; walk = walk->next) {
909 struct pollfd * pfd, * pfd_end;
910
911 pfd = walk->entries;
912 pfd_end = pfd + walk->len;
913 for (; pfd != pfd_end; pfd++) {
914 /*
915 * Fish for events. If we found one, record it
916 * and kill poll_table->_qproc, so we don't
917 * needlessly register any other waiters after
918 * this. They'll get immediately deregistered
919 * when we break out and return.
920 */
921 if (do_pollfd(pfd, pt, &can_busy_loop,
922 busy_flag)) {
923 count++;
924 pt->_qproc = NULL;
925 /* found something, stop busy polling */
926 busy_flag = 0;
927 can_busy_loop = false;
928 }
929 }
930 }
931 /*
932 * All waiters have already been registered, so don't provide
933 * a poll_table->_qproc to them on the next loop iteration.
934 */
935 pt->_qproc = NULL;
936 if (!count) {
937 count = wait->error;
938 if (signal_pending(current))
939 count = -ERESTARTNOHAND;
940 }
941 if (count || timed_out)
942 break;
943
944 /* only if found POLL_BUSY_LOOP sockets && not out of time */
945 if (can_busy_loop && !need_resched()) {
946 if (!busy_start) {
947 busy_start = busy_loop_current_time();
948 continue;
949 }
950 if (!busy_loop_timeout(busy_start))
951 continue;
952 }
953 busy_flag = 0;
954
955 /*
956 * If this is the first loop and we have a timeout
957 * given, then we convert to ktime_t and set the to
958 * pointer to the expiry value.
959 */
960 if (end_time && !to) {
961 expire = timespec64_to_ktime(*end_time);
962 to = &expire;
963 }
964
965 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
966 timed_out = 1;
967 }
968 return count;
969}
970
971#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
972 sizeof(struct pollfd))
973
974static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
975 struct timespec64 *end_time)
976{
977 struct poll_wqueues table;
978 int err = -EFAULT, fdcount;
979 /* Allocate small arguments on the stack to save memory and be
980 faster - use long to make sure the buffer is aligned properly
981 on 64 bit archs to avoid unaligned access */
982 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
983 struct poll_list *const head = (struct poll_list *)stack_pps;
984 struct poll_list *walk = head;
985 unsigned int todo = nfds;
986 unsigned int len;
987
988 if (nfds > rlimit(RLIMIT_NOFILE))
989 return -EINVAL;
990
991 len = min_t(unsigned int, nfds, N_STACK_PPS);
992 for (;;) {
993 walk->next = NULL;
994 walk->len = len;
995 if (!len)
996 break;
997
998 if (copy_from_user(walk->entries, ufds + nfds-todo,
999 sizeof(struct pollfd) * walk->len))
1000 goto out_fds;
1001
1002 if (walk->len >= todo)
1003 break;
1004 todo -= walk->len;
1005
1006 len = min(todo, POLLFD_PER_PAGE);
1007 walk = walk->next = kmalloc(struct_size(walk, entries, len),
1008 GFP_KERNEL);
1009 if (!walk) {
1010 err = -ENOMEM;
1011 goto out_fds;
1012 }
1013 }
1014
1015 poll_initwait(&table);
1016 fdcount = do_poll(head, &table, end_time);
1017 poll_freewait(&table);
1018
1019 if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1020 goto out_fds;
1021
1022 for (walk = head; walk; walk = walk->next) {
1023 struct pollfd *fds = walk->entries;
1024 unsigned int j;
1025
1026 for (j = walk->len; j; fds++, ufds++, j--)
1027 unsafe_put_user(fds->revents, &ufds->revents, Efault);
1028 }
1029 user_write_access_end();
1030
1031 err = fdcount;
1032out_fds:
1033 walk = head->next;
1034 while (walk) {
1035 struct poll_list *pos = walk;
1036 walk = walk->next;
1037 kfree(pos);
1038 }
1039
1040 return err;
1041
1042Efault:
1043 user_write_access_end();
1044 err = -EFAULT;
1045 goto out_fds;
1046}
1047
1048static long do_restart_poll(struct restart_block *restart_block)
1049{
1050 struct pollfd __user *ufds = restart_block->poll.ufds;
1051 int nfds = restart_block->poll.nfds;
1052 struct timespec64 *to = NULL, end_time;
1053 int ret;
1054
1055 if (restart_block->poll.has_timeout) {
1056 end_time.tv_sec = restart_block->poll.tv_sec;
1057 end_time.tv_nsec = restart_block->poll.tv_nsec;
1058 to = &end_time;
1059 }
1060
1061 ret = do_sys_poll(ufds, nfds, to);
1062
1063 if (ret == -ERESTARTNOHAND)
1064 ret = set_restart_fn(restart_block, do_restart_poll);
1065
1066 return ret;
1067}
1068
1069SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1070 int, timeout_msecs)
1071{
1072 struct timespec64 end_time, *to = NULL;
1073 int ret;
1074
1075 if (timeout_msecs >= 0) {
1076 to = &end_time;
1077 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1078 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1079 }
1080
1081 ret = do_sys_poll(ufds, nfds, to);
1082
1083 if (ret == -ERESTARTNOHAND) {
1084 struct restart_block *restart_block;
1085
1086 restart_block = ¤t->restart_block;
1087 restart_block->poll.ufds = ufds;
1088 restart_block->poll.nfds = nfds;
1089
1090 if (timeout_msecs >= 0) {
1091 restart_block->poll.tv_sec = end_time.tv_sec;
1092 restart_block->poll.tv_nsec = end_time.tv_nsec;
1093 restart_block->poll.has_timeout = 1;
1094 } else
1095 restart_block->poll.has_timeout = 0;
1096
1097 ret = set_restart_fn(restart_block, do_restart_poll);
1098 }
1099 return ret;
1100}
1101
1102SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1103 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1104 size_t, sigsetsize)
1105{
1106 struct timespec64 ts, end_time, *to = NULL;
1107 int ret;
1108
1109 if (tsp) {
1110 if (get_timespec64(&ts, tsp))
1111 return -EFAULT;
1112
1113 to = &end_time;
1114 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1115 return -EINVAL;
1116 }
1117
1118 ret = set_user_sigmask(sigmask, sigsetsize);
1119 if (ret)
1120 return ret;
1121
1122 ret = do_sys_poll(ufds, nfds, to);
1123 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1124}
1125
1126#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1127
1128SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1129 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1130 size_t, sigsetsize)
1131{
1132 struct timespec64 ts, end_time, *to = NULL;
1133 int ret;
1134
1135 if (tsp) {
1136 if (get_old_timespec32(&ts, tsp))
1137 return -EFAULT;
1138
1139 to = &end_time;
1140 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1141 return -EINVAL;
1142 }
1143
1144 ret = set_user_sigmask(sigmask, sigsetsize);
1145 if (ret)
1146 return ret;
1147
1148 ret = do_sys_poll(ufds, nfds, to);
1149 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1150}
1151#endif
1152
1153#ifdef CONFIG_COMPAT
1154#define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1155
1156/*
1157 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1158 * 64-bit unsigned longs.
1159 */
1160static
1161int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1162 unsigned long *fdset)
1163{
1164 if (ufdset) {
1165 return compat_get_bitmap(fdset, ufdset, nr);
1166 } else {
1167 zero_fd_set(nr, fdset);
1168 return 0;
1169 }
1170}
1171
1172static
1173int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1174 unsigned long *fdset)
1175{
1176 if (!ufdset)
1177 return 0;
1178 return compat_put_bitmap(ufdset, fdset, nr);
1179}
1180
1181
1182/*
1183 * This is a virtual copy of sys_select from fs/select.c and probably
1184 * should be compared to it from time to time
1185 */
1186
1187/*
1188 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1189 * like to be certain this leads to no problems. So I return
1190 * EINTR just for safety.
1191 *
1192 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1193 * I'm trying ERESTARTNOHAND which restart only when you want to.
1194 */
1195static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1196 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1197 struct timespec64 *end_time)
1198{
1199 fd_set_bits fds;
1200 void *bits;
1201 int size, max_fds, ret = -EINVAL;
1202 struct fdtable *fdt;
1203 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1204
1205 if (n < 0)
1206 goto out_nofds;
1207
1208 /* max_fds can increase, so grab it once to avoid race */
1209 rcu_read_lock();
1210 fdt = files_fdtable(current->files);
1211 max_fds = fdt->max_fds;
1212 rcu_read_unlock();
1213 if (n > max_fds)
1214 n = max_fds;
1215
1216 /*
1217 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1218 * since we used fdset we need to allocate memory in units of
1219 * long-words.
1220 */
1221 size = FDS_BYTES(n);
1222 bits = stack_fds;
1223 if (size > sizeof(stack_fds) / 6) {
1224 bits = kmalloc_array(6, size, GFP_KERNEL);
1225 ret = -ENOMEM;
1226 if (!bits)
1227 goto out_nofds;
1228 }
1229 fds.in = (unsigned long *) bits;
1230 fds.out = (unsigned long *) (bits + size);
1231 fds.ex = (unsigned long *) (bits + 2*size);
1232 fds.res_in = (unsigned long *) (bits + 3*size);
1233 fds.res_out = (unsigned long *) (bits + 4*size);
1234 fds.res_ex = (unsigned long *) (bits + 5*size);
1235
1236 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1237 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1238 (ret = compat_get_fd_set(n, exp, fds.ex)))
1239 goto out;
1240 zero_fd_set(n, fds.res_in);
1241 zero_fd_set(n, fds.res_out);
1242 zero_fd_set(n, fds.res_ex);
1243
1244 ret = do_select(n, &fds, end_time);
1245
1246 if (ret < 0)
1247 goto out;
1248 if (!ret) {
1249 ret = -ERESTARTNOHAND;
1250 if (signal_pending(current))
1251 goto out;
1252 ret = 0;
1253 }
1254
1255 if (compat_set_fd_set(n, inp, fds.res_in) ||
1256 compat_set_fd_set(n, outp, fds.res_out) ||
1257 compat_set_fd_set(n, exp, fds.res_ex))
1258 ret = -EFAULT;
1259out:
1260 if (bits != stack_fds)
1261 kfree(bits);
1262out_nofds:
1263 return ret;
1264}
1265
1266static int do_compat_select(int n, compat_ulong_t __user *inp,
1267 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1268 struct old_timeval32 __user *tvp)
1269{
1270 struct timespec64 end_time, *to = NULL;
1271 struct old_timeval32 tv;
1272 int ret;
1273
1274 if (tvp) {
1275 if (copy_from_user(&tv, tvp, sizeof(tv)))
1276 return -EFAULT;
1277
1278 to = &end_time;
1279 if (poll_select_set_timeout(to,
1280 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1281 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1282 return -EINVAL;
1283 }
1284
1285 ret = compat_core_sys_select(n, inp, outp, exp, to);
1286 return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1287}
1288
1289COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1290 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1291 struct old_timeval32 __user *, tvp)
1292{
1293 return do_compat_select(n, inp, outp, exp, tvp);
1294}
1295
1296struct compat_sel_arg_struct {
1297 compat_ulong_t n;
1298 compat_uptr_t inp;
1299 compat_uptr_t outp;
1300 compat_uptr_t exp;
1301 compat_uptr_t tvp;
1302};
1303
1304COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1305{
1306 struct compat_sel_arg_struct a;
1307
1308 if (copy_from_user(&a, arg, sizeof(a)))
1309 return -EFAULT;
1310 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1311 compat_ptr(a.exp), compat_ptr(a.tvp));
1312}
1313
1314static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1315 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1316 void __user *tsp, compat_sigset_t __user *sigmask,
1317 compat_size_t sigsetsize, enum poll_time_type type)
1318{
1319 struct timespec64 ts, end_time, *to = NULL;
1320 int ret;
1321
1322 if (tsp) {
1323 switch (type) {
1324 case PT_OLD_TIMESPEC:
1325 if (get_old_timespec32(&ts, tsp))
1326 return -EFAULT;
1327 break;
1328 case PT_TIMESPEC:
1329 if (get_timespec64(&ts, tsp))
1330 return -EFAULT;
1331 break;
1332 default:
1333 BUG();
1334 }
1335
1336 to = &end_time;
1337 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1338 return -EINVAL;
1339 }
1340
1341 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1342 if (ret)
1343 return ret;
1344
1345 ret = compat_core_sys_select(n, inp, outp, exp, to);
1346 return poll_select_finish(&end_time, tsp, type, ret);
1347}
1348
1349struct compat_sigset_argpack {
1350 compat_uptr_t p;
1351 compat_size_t size;
1352};
1353static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1354 struct compat_sigset_argpack __user *from)
1355{
1356 if (from) {
1357 if (!user_read_access_begin(from, sizeof(*from)))
1358 return -EFAULT;
1359 unsafe_get_user(to->p, &from->p, Efault);
1360 unsafe_get_user(to->size, &from->size, Efault);
1361 user_read_access_end();
1362 }
1363 return 0;
1364Efault:
1365 user_access_end();
1366 return -EFAULT;
1367}
1368
1369COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1370 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1371 struct __kernel_timespec __user *, tsp, void __user *, sig)
1372{
1373 struct compat_sigset_argpack x = {0, 0};
1374
1375 if (get_compat_sigset_argpack(&x, sig))
1376 return -EFAULT;
1377
1378 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1379 x.size, PT_TIMESPEC);
1380}
1381
1382#if defined(CONFIG_COMPAT_32BIT_TIME)
1383
1384COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1385 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1386 struct old_timespec32 __user *, tsp, void __user *, sig)
1387{
1388 struct compat_sigset_argpack x = {0, 0};
1389
1390 if (get_compat_sigset_argpack(&x, sig))
1391 return -EFAULT;
1392
1393 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1394 x.size, PT_OLD_TIMESPEC);
1395}
1396
1397#endif
1398
1399#if defined(CONFIG_COMPAT_32BIT_TIME)
1400COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1401 unsigned int, nfds, struct old_timespec32 __user *, tsp,
1402 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1403{
1404 struct timespec64 ts, end_time, *to = NULL;
1405 int ret;
1406
1407 if (tsp) {
1408 if (get_old_timespec32(&ts, tsp))
1409 return -EFAULT;
1410
1411 to = &end_time;
1412 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1413 return -EINVAL;
1414 }
1415
1416 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1417 if (ret)
1418 return ret;
1419
1420 ret = do_sys_poll(ufds, nfds, to);
1421 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1422}
1423#endif
1424
1425/* New compat syscall for 64 bit time_t*/
1426COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1427 unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1428 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1429{
1430 struct timespec64 ts, end_time, *to = NULL;
1431 int ret;
1432
1433 if (tsp) {
1434 if (get_timespec64(&ts, tsp))
1435 return -EFAULT;
1436
1437 to = &end_time;
1438 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1439 return -EINVAL;
1440 }
1441
1442 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1443 if (ret)
1444 return ret;
1445
1446 ret = do_sys_poll(ufds, nfds, to);
1447 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1448}
1449
1450#endif