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