blob: 79b876eb91dad25c91ea3715592c2b9f6655852b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Milind Arun Choudhary022a1692007-05-08 00:29:02 -070017#include <linux/kernel.h>
Alexey Dobriyana99bbaf2009-10-04 16:11:37 +040018#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/syscalls.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050020#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/poll.h>
23#include <linux/personality.h> /* for STICKY_TIMEOUTS */
24#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040025#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/fs.h>
Dipankar Sarmab8359962005-09-09 13:04:14 -070027#include <linux/rcupdate.h>
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -070028#include <linux/hrtimer.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060029#include <linux/sched/rt.h>
Eliezer Tamir2d48d672013-06-24 10:28:03 +030030#include <net/ll_poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/uaccess.h>
33
Arjan van de Ven90d6e242008-09-01 15:55:35 -070034
35/*
36 * Estimate expected accuracy in ns from a timeval.
37 *
38 * After quite a bit of churning around, we've settled on
39 * a simple thing of taking 0.1% of the timeout as the
40 * slack, with a cap of 100 msec.
41 * "nice" tasks get a 0.5% slack instead.
42 *
43 * Consider this comment an open invitation to come up with even
44 * better solutions..
45 */
46
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070047#define MAX_SLACK (100 * NSEC_PER_MSEC)
48
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070049static long __estimate_accuracy(struct timespec *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070050{
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070051 long slack;
Arjan van de Ven90d6e242008-09-01 15:55:35 -070052 int divfactor = 1000;
53
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070054 if (tv->tv_sec < 0)
55 return 0;
56
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070057 if (task_nice(current) > 0)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070058 divfactor = divfactor / 5;
59
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070060 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
61 return MAX_SLACK;
62
Arjan van de Ven90d6e242008-09-01 15:55:35 -070063 slack = tv->tv_nsec / divfactor;
64 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
65
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070066 if (slack > MAX_SLACK)
67 return MAX_SLACK;
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070068
Arjan van de Ven90d6e242008-09-01 15:55:35 -070069 return slack;
70}
71
Shawn Bohrer95aac7b2010-10-27 15:34:54 -070072long select_estimate_accuracy(struct timespec *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070073{
74 unsigned long ret;
75 struct timespec now;
76
77 /*
78 * Realtime tasks get a slack of 0 for obvious reasons.
79 */
80
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070081 if (rt_task(current))
Arjan van de Ven90d6e242008-09-01 15:55:35 -070082 return 0;
83
84 ktime_get_ts(&now);
85 now = timespec_sub(*tv, now);
86 ret = __estimate_accuracy(&now);
87 if (ret < current->timer_slack_ns)
88 return current->timer_slack_ns;
89 return ret;
90}
91
92
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094struct poll_table_page {
95 struct poll_table_page * next;
96 struct poll_table_entry * entry;
97 struct poll_table_entry entries[0];
98};
99
100#define POLL_TABLE_FULL(table) \
101 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
102
103/*
104 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
105 * I have rewritten this, taking some shortcuts: This code may not be easy to
106 * follow, but it should be free of race-conditions, and it's practical. If you
107 * understand what I'm doing here, then you understand how the linux
108 * sleep/wakeup mechanism works.
109 *
110 * Two very simple procedures, poll_wait() and poll_freewait() make all the
111 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
112 * as all select/poll functions have to call it to add an entry to the
113 * poll table.
114 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700115static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
116 poll_table *p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
118void poll_initwait(struct poll_wqueues *pwq)
119{
120 init_poll_funcptr(&pwq->pt, __pollwait);
Tejun Heo5f820f62009-01-06 14:40:59 -0800121 pwq->polling_task = current;
Guillaume Knispelb2add732009-08-15 19:30:24 +0200122 pwq->triggered = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 pwq->error = 0;
124 pwq->table = NULL;
Andi Kleen70674f92006-03-28 01:56:33 -0800125 pwq->inline_index = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127EXPORT_SYMBOL(poll_initwait);
128
Andi Kleen70674f92006-03-28 01:56:33 -0800129static void free_poll_entry(struct poll_table_entry *entry)
130{
WANG Congccf67802007-05-09 07:10:02 +0200131 remove_wait_queue(entry->wait_address, &entry->wait);
Andi Kleen70674f92006-03-28 01:56:33 -0800132 fput(entry->filp);
133}
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135void poll_freewait(struct poll_wqueues *pwq)
136{
137 struct poll_table_page * p = pwq->table;
Andi Kleen70674f92006-03-28 01:56:33 -0800138 int i;
139 for (i = 0; i < pwq->inline_index; i++)
140 free_poll_entry(pwq->inline_entries + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 while (p) {
142 struct poll_table_entry * entry;
143 struct poll_table_page *old;
144
145 entry = p->entry;
146 do {
147 entry--;
Andi Kleen70674f92006-03-28 01:56:33 -0800148 free_poll_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 } while (entry > p->entries);
150 old = p;
151 p = p->next;
152 free_page((unsigned long) old);
153 }
154}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155EXPORT_SYMBOL(poll_freewait);
156
Tejun Heo5f820f62009-01-06 14:40:59 -0800157static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 struct poll_table_page *table = p->table;
160
Andi Kleen70674f92006-03-28 01:56:33 -0800161 if (p->inline_index < N_INLINE_POLL_ENTRIES)
162 return p->inline_entries + p->inline_index++;
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 if (!table || POLL_TABLE_FULL(table)) {
165 struct poll_table_page *new_table;
166
167 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
168 if (!new_table) {
169 p->error = -ENOMEM;
Andi Kleen70674f92006-03-28 01:56:33 -0800170 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 }
172 new_table->entry = new_table->entries;
173 new_table->next = table;
174 p->table = new_table;
175 table = new_table;
176 }
177
Andi Kleen70674f92006-03-28 01:56:33 -0800178 return table->entry++;
179}
180
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700181static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
Tejun Heo5f820f62009-01-06 14:40:59 -0800182{
183 struct poll_wqueues *pwq = wait->private;
184 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
185
186 /*
187 * Although this function is called under waitqueue lock, LOCK
188 * doesn't imply write barrier and the users expect write
189 * barrier semantics on wakeup functions. The following
190 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
191 * and is paired with set_mb() in poll_schedule_timeout.
192 */
193 smp_wmb();
194 pwq->triggered = 1;
195
196 /*
197 * Perform the default wake up operation using a dummy
198 * waitqueue.
199 *
200 * TODO: This is hacky but there currently is no interface to
201 * pass in @sync. @sync is scheduled to be removed and once
202 * that happens, wake_up_process() can be used directly.
203 */
204 return default_wake_function(&dummy_wait, mode, sync, key);
205}
206
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700207static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
208{
209 struct poll_table_entry *entry;
210
211 entry = container_of(wait, struct poll_table_entry, wait);
212 if (key && !((unsigned long)key & entry->key))
213 return 0;
214 return __pollwake(wait, mode, sync, key);
215}
216
Andi Kleen70674f92006-03-28 01:56:33 -0800217/* Add a new entry */
218static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
219 poll_table *p)
220{
Tejun Heo5f820f62009-01-06 14:40:59 -0800221 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
222 struct poll_table_entry *entry = poll_get_entry(pwq);
Andi Kleen70674f92006-03-28 01:56:33 -0800223 if (!entry)
224 return;
Al Virocb0942b2012-08-27 14:48:26 -0400225 entry->filp = get_file(filp);
Andi Kleen70674f92006-03-28 01:56:33 -0800226 entry->wait_address = wait_address;
Hans Verkuil626cf232012-03-23 15:02:27 -0700227 entry->key = p->_key;
Tejun Heo5f820f62009-01-06 14:40:59 -0800228 init_waitqueue_func_entry(&entry->wait, pollwake);
229 entry->wait.private = pwq;
WANG Congccf67802007-05-09 07:10:02 +0200230 add_wait_queue(wait_address, &entry->wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
232
Tejun Heo5f820f62009-01-06 14:40:59 -0800233int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
234 ktime_t *expires, unsigned long slack)
235{
236 int rc = -EINTR;
237
238 set_current_state(state);
239 if (!pwq->triggered)
240 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
241 __set_current_state(TASK_RUNNING);
242
243 /*
244 * Prepare for the next iteration.
245 *
246 * The following set_mb() serves two purposes. First, it's
247 * the counterpart rmb of the wmb in pollwake() such that data
248 * written before wake up is always visible after wake up.
249 * Second, the full barrier guarantees that triggered clearing
250 * doesn't pass event check of the next iteration. Note that
251 * this problem doesn't exist for the first iteration as
252 * add_wait_queue() has full barrier semantics.
253 */
254 set_mb(pwq->triggered, 0);
255
256 return rc;
257}
258EXPORT_SYMBOL(poll_schedule_timeout);
259
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700260/**
261 * poll_select_set_timeout - helper function to setup the timeout value
262 * @to: pointer to timespec variable for the final timeout
263 * @sec: seconds (from user space)
264 * @nsec: nanoseconds (from user space)
265 *
266 * Note, we do not use a timespec for the user space value here, That
267 * way we can use the function for timeval and compat interfaces as well.
268 *
269 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
270 */
271int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
272{
273 struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
274
275 if (!timespec_valid(&ts))
276 return -EINVAL;
277
278 /* Optimize for the zero timeout value here */
279 if (!sec && !nsec) {
280 to->tv_sec = to->tv_nsec = 0;
281 } else {
282 ktime_get_ts(to);
283 *to = timespec_add_safe(*to, ts);
284 }
285 return 0;
286}
287
288static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
289 int timeval, int ret)
290{
291 struct timespec rts;
292 struct timeval rtv;
293
294 if (!p)
295 return ret;
296
297 if (current->personality & STICKY_TIMEOUTS)
298 goto sticky;
299
300 /* No update for zero timeout */
301 if (!end_time->tv_sec && !end_time->tv_nsec)
302 return ret;
303
304 ktime_get_ts(&rts);
305 rts = timespec_sub(*end_time, rts);
306 if (rts.tv_sec < 0)
307 rts.tv_sec = rts.tv_nsec = 0;
308
309 if (timeval) {
Vasiliy Kulikov65329bf2011-01-12 17:00:00 -0800310 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
311 memset(&rtv, 0, sizeof(rtv));
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700312 rtv.tv_sec = rts.tv_sec;
313 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
314
315 if (!copy_to_user(p, &rtv, sizeof(rtv)))
316 return ret;
317
318 } else if (!copy_to_user(p, &rts, sizeof(rts)))
319 return ret;
320
321 /*
322 * If an application puts its timeval in read-only memory, we
323 * don't want the Linux-specific update to the timeval to
324 * cause a fault after the select has completed
325 * successfully. However, because we're not updating the
326 * timeval, we can't restart the system call.
327 */
328
329sticky:
330 if (ret == -ERESTARTNOHAND)
331 ret = -EINTR;
332 return ret;
333}
334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335#define FDS_IN(fds, n) (fds->in + n)
336#define FDS_OUT(fds, n) (fds->out + n)
337#define FDS_EX(fds, n) (fds->ex + n)
338
339#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
340
341static int max_select_fd(unsigned long n, fd_set_bits *fds)
342{
343 unsigned long *open_fds;
344 unsigned long set;
345 int max;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700346 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 /* handle last in-complete long-word first */
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400349 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
350 n /= BITS_PER_LONG;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700351 fdt = files_fdtable(current->files);
David Howells1fd36ad2012-02-16 17:49:54 +0000352 open_fds = fdt->open_fds + n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 max = 0;
354 if (set) {
355 set &= BITS(fds, n);
356 if (set) {
357 if (!(set & ~*open_fds))
358 goto get_max;
359 return -EBADF;
360 }
361 }
362 while (n) {
363 open_fds--;
364 n--;
365 set = BITS(fds, n);
366 if (!set)
367 continue;
368 if (set & ~*open_fds)
369 return -EBADF;
370 if (max)
371 continue;
372get_max:
373 do {
374 max++;
375 set >>= 1;
376 } while (set);
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400377 max += n * BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 }
379
380 return max;
381}
382
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
384#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
385#define POLLEX_SET (POLLPRI)
386
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700387static inline void wait_key_set(poll_table *wait, unsigned long in,
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300388 unsigned long out, unsigned long bit,
389 unsigned int ll_flag)
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700390{
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300391 wait->_key = POLLEX_SET | ll_flag;
Hans Verkuil626cf232012-03-23 15:02:27 -0700392 if (in & bit)
393 wait->_key |= POLLIN_SET;
394 if (out & bit)
395 wait->_key |= POLLOUT_SET;
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700396}
397
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700398int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700400 ktime_t expire, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 struct poll_wqueues table;
402 poll_table *wait;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700403 int retval, i, timed_out = 0;
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700404 unsigned long slack = 0;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300405 unsigned int ll_flag = POLL_LL;
406 u64 ll_time = ll_end_time();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Dipankar Sarmab8359962005-09-09 13:04:14 -0700408 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 retval = max_select_fd(n, fds);
Dipankar Sarmab8359962005-09-09 13:04:14 -0700410 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412 if (retval < 0)
413 return retval;
414 n = retval;
415
416 poll_initwait(&table);
417 wait = &table.pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700418 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700419 wait->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700420 timed_out = 1;
421 }
422
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700423 if (end_time && !timed_out)
Andrew Morton231f3d32010-10-27 15:34:53 -0700424 slack = select_estimate_accuracy(end_time);
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 retval = 0;
427 for (;;) {
428 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300429 bool can_ll = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 inp = fds->in; outp = fds->out; exp = fds->ex;
432 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
433
434 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
435 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
436 unsigned long res_in = 0, res_out = 0, res_ex = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
438 in = *inp++; out = *outp++; ex = *exp++;
439 all_bits = in | out | ex;
440 if (all_bits == 0) {
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400441 i += BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 continue;
443 }
444
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400445 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
Al Viro2903ff02012-08-28 12:52:22 -0400446 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 if (i >= n)
448 break;
449 if (!(bit & all_bits))
450 continue;
Al Viro2903ff02012-08-28 12:52:22 -0400451 f = fdget(i);
452 if (f.file) {
453 const struct file_operations *f_op;
454 f_op = f.file->f_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 mask = DEFAULT_POLLMASK;
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700456 if (f_op && f_op->poll) {
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300457 wait_key_set(wait, in, out,
458 bit, ll_flag);
Al Viro2903ff02012-08-28 12:52:22 -0400459 mask = (*f_op->poll)(f.file, wait);
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700460 }
Al Viro2903ff02012-08-28 12:52:22 -0400461 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 if ((mask & POLLIN_SET) && (in & bit)) {
463 res_in |= bit;
464 retval++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700465 wait->_qproc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 }
467 if ((mask & POLLOUT_SET) && (out & bit)) {
468 res_out |= bit;
469 retval++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700470 wait->_qproc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 }
472 if ((mask & POLLEX_SET) && (ex & bit)) {
473 res_ex |= bit;
474 retval++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700475 wait->_qproc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300477 if (mask & POLL_LL)
478 can_ll = true;
479 /* got something, stop busy polling */
480 if (retval)
481 ll_flag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 }
484 if (res_in)
485 *rinp = res_in;
486 if (res_out)
487 *routp = res_out;
488 if (res_ex)
489 *rexp = res_ex;
Linus Torvalds55d85382008-06-22 12:23:15 -0700490 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 }
Hans Verkuil626cf232012-03-23 15:02:27 -0700492 wait->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700493 if (retval || timed_out || signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 break;
Pavel Machekf5264482008-04-21 22:15:06 +0000495 if (table.error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 retval = table.error;
497 break;
498 }
David Woodhouse9f729492006-01-18 17:44:05 -0800499
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300500 if (can_ll && can_poll_ll(ll_time))
501 continue;
502
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700503 /*
504 * If this is the first loop and we have a timeout
505 * given, then we convert to ktime_t and set the to
506 * pointer to the expiry value.
507 */
508 if (end_time && !to) {
509 expire = timespec_to_ktime(*end_time);
510 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800511 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700512
Tejun Heo5f820f62009-01-06 14:40:59 -0800513 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
514 to, slack))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700515 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
518 poll_freewait(&table);
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 return retval;
521}
522
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523/*
524 * We can actually return ERESTARTSYS instead of EINTR, but I'd
525 * like to be certain this leads to no problems. So I return
526 * EINTR just for safety.
527 *
528 * Update: ERESTARTSYS breaks at least the xview clock binary, so
529 * I'm trying ERESTARTNOHAND which restart only when you want to.
530 */
Al Viroa2dcb442008-04-23 14:05:15 -0400531int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700532 fd_set __user *exp, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
534 fd_set_bits fds;
Andrew Morton29ff2db2006-04-10 22:52:46 -0700535 void *bits;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800536 int ret, max_fds;
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700537 unsigned int size;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700538 struct fdtable *fdt;
Andi Kleen70674f92006-03-28 01:56:33 -0800539 /* Allocate small arguments on the stack to save memory and be faster */
Jes Sorensen30c14e42006-03-31 11:18:57 -0500540 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 ret = -EINVAL;
543 if (n < 0)
544 goto out_nofds;
545
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800546 /* max_fds can increase, so grab it once to avoid race */
Dipankar Sarmab8359962005-09-09 13:04:14 -0700547 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700548 fdt = files_fdtable(current->files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800549 max_fds = fdt->max_fds;
Dipankar Sarmab8359962005-09-09 13:04:14 -0700550 rcu_read_unlock();
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800551 if (n > max_fds)
552 n = max_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
554 /*
555 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
556 * since we used fdset we need to allocate memory in units of
557 * long-words.
558 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 size = FDS_BYTES(n);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700560 bits = stack_fds;
561 if (size > sizeof(stack_fds) / 6) {
562 /* Not enough space in on-stack array; must use kmalloc */
563 ret = -ENOMEM;
Andi Kleen70674f92006-03-28 01:56:33 -0800564 bits = kmalloc(6 * size, GFP_KERNEL);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700565 if (!bits)
566 goto out_nofds;
567 }
Andrew Morton29ff2db2006-04-10 22:52:46 -0700568 fds.in = bits;
569 fds.out = bits + size;
570 fds.ex = bits + 2*size;
571 fds.res_in = bits + 3*size;
572 fds.res_out = bits + 4*size;
573 fds.res_ex = bits + 5*size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 if ((ret = get_fd_set(n, inp, fds.in)) ||
576 (ret = get_fd_set(n, outp, fds.out)) ||
577 (ret = get_fd_set(n, exp, fds.ex)))
578 goto out;
579 zero_fd_set(n, fds.res_in);
580 zero_fd_set(n, fds.res_out);
581 zero_fd_set(n, fds.res_ex);
582
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700583 ret = do_select(n, &fds, end_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 if (ret < 0)
586 goto out;
587 if (!ret) {
588 ret = -ERESTARTNOHAND;
589 if (signal_pending(current))
590 goto out;
591 ret = 0;
592 }
593
594 if (set_fd_set(n, inp, fds.res_in) ||
595 set_fd_set(n, outp, fds.res_out) ||
596 set_fd_set(n, exp, fds.res_ex))
597 ret = -EFAULT;
598
599out:
Andi Kleen70674f92006-03-28 01:56:33 -0800600 if (bits != stack_fds)
601 kfree(bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602out_nofds:
603 return ret;
604}
605
Heiko Carstens5a8a82b2009-01-14 14:14:25 +0100606SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
607 fd_set __user *, exp, struct timeval __user *, tvp)
David Woodhouse9f729492006-01-18 17:44:05 -0800608{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700609 struct timespec end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800610 struct timeval tv;
611 int ret;
612
613 if (tvp) {
614 if (copy_from_user(&tv, tvp, sizeof(tv)))
615 return -EFAULT;
616
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700617 to = &end_time;
Arjan van de Ven4d36a9e2008-10-25 12:41:41 -0700618 if (poll_select_set_timeout(to,
619 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
620 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
David Woodhouse9f729492006-01-18 17:44:05 -0800621 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800622 }
623
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700624 ret = core_sys_select(n, inp, outp, exp, to);
625 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800626
627 return ret;
628}
629
Heiko Carstensc9da9f22009-01-14 14:13:57 +0100630static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
631 fd_set __user *exp, struct timespec __user *tsp,
632 const sigset_t __user *sigmask, size_t sigsetsize)
David Woodhouse9f729492006-01-18 17:44:05 -0800633{
David Woodhouse9f729492006-01-18 17:44:05 -0800634 sigset_t ksigmask, sigsaved;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700635 struct timespec ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800636 int ret;
637
638 if (tsp) {
639 if (copy_from_user(&ts, tsp, sizeof(ts)))
640 return -EFAULT;
641
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700642 to = &end_time;
643 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
David Woodhouse9f729492006-01-18 17:44:05 -0800644 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800645 }
646
647 if (sigmask) {
648 /* XXX: Don't preclude handling different sized sigset_t's. */
649 if (sigsetsize != sizeof(sigset_t))
650 return -EINVAL;
651 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
652 return -EFAULT;
653
654 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
655 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
656 }
657
Bernd Schmidt62568512009-01-13 22:14:48 +0100658 ret = core_sys_select(n, inp, outp, exp, to);
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700659 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800660
661 if (ret == -ERESTARTNOHAND) {
662 /*
663 * Don't restore the signal mask yet. Let do_signal() deliver
664 * the signal on the way back to userspace, before the signal
665 * mask is restored.
666 */
667 if (sigmask) {
668 memcpy(&current->saved_sigmask, &sigsaved,
669 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -0700670 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -0800671 }
672 } else if (sigmask)
673 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
674
675 return ret;
676}
677
678/*
679 * Most architectures can't handle 7-argument syscalls. So we provide a
680 * 6-argument version where the sixth argument is a pointer to a structure
681 * which has a pointer to the sigset_t itself followed by a size_t containing
682 * the sigset size.
683 */
Heiko Carstensd4e82042009-01-14 14:14:34 +0100684SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
685 fd_set __user *, exp, struct timespec __user *, tsp,
686 void __user *, sig)
David Woodhouse9f729492006-01-18 17:44:05 -0800687{
688 size_t sigsetsize = 0;
689 sigset_t __user *up = NULL;
690
691 if (sig) {
692 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
Al Viroe110ab92006-02-01 05:26:09 -0500693 || __get_user(up, (sigset_t __user * __user *)sig)
David Woodhouse9f729492006-01-18 17:44:05 -0800694 || __get_user(sigsetsize,
Al Viroe110ab92006-02-01 05:26:09 -0500695 (size_t __user *)(sig+sizeof(void *))))
David Woodhouse9f729492006-01-18 17:44:05 -0800696 return -EFAULT;
697 }
698
Heiko Carstensc9da9f22009-01-14 14:13:57 +0100699 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
David Woodhouse9f729492006-01-18 17:44:05 -0800700}
David Woodhouse9f729492006-01-18 17:44:05 -0800701
Christoph Hellwig5d0e5282010-03-10 15:21:13 -0800702#ifdef __ARCH_WANT_SYS_OLD_SELECT
703struct sel_arg_struct {
704 unsigned long n;
705 fd_set __user *inp, *outp, *exp;
706 struct timeval __user *tvp;
707};
708
709SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
710{
711 struct sel_arg_struct a;
712
713 if (copy_from_user(&a, arg, sizeof(a)))
714 return -EFAULT;
715 return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
716}
717#endif
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719struct poll_list {
720 struct poll_list *next;
721 int len;
722 struct pollfd entries[0];
723};
724
725#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
726
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700727/*
728 * Fish for pollable events on the pollfd->fd file descriptor. We're only
729 * interested in events matching the pollfd->events mask, and the result
730 * matching that mask is both recorded in pollfd->revents and returned. The
731 * pwait poll_table will be used by the fd-provided poll handler for waiting,
Hans Verkuil626cf232012-03-23 15:02:27 -0700732 * if pwait->_qproc is non-NULL.
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700733 */
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300734static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait,
735 bool *can_ll, unsigned int ll_flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736{
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700737 unsigned int mask;
738 int fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700740 mask = 0;
741 fd = pollfd->fd;
742 if (fd >= 0) {
Al Viro2903ff02012-08-28 12:52:22 -0400743 struct fd f = fdget(fd);
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700744 mask = POLLNVAL;
Al Viro2903ff02012-08-28 12:52:22 -0400745 if (f.file) {
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700746 mask = DEFAULT_POLLMASK;
Al Viro2903ff02012-08-28 12:52:22 -0400747 if (f.file->f_op && f.file->f_op->poll) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700748 pwait->_key = pollfd->events|POLLERR|POLLHUP;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300749 pwait->_key |= ll_flag;
Al Viro2903ff02012-08-28 12:52:22 -0400750 mask = f.file->f_op->poll(f.file, pwait);
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300751 if (mask & POLL_LL)
752 *can_ll = true;
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700753 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700754 /* Mask out unneeded events. */
755 mask &= pollfd->events | POLLERR | POLLHUP;
Al Viro2903ff02012-08-28 12:52:22 -0400756 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700759 pollfd->revents = mask;
760
761 return mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
764static int do_poll(unsigned int nfds, struct poll_list *list,
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700765 struct poll_wqueues *wait, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 poll_table* pt = &wait->pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700768 ktime_t expire, *to = NULL;
769 int timed_out = 0, count = 0;
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700770 unsigned long slack = 0;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300771 unsigned int ll_flag = POLL_LL;
772 u64 ll_time = ll_end_time();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
David Woodhouse9f729492006-01-18 17:44:05 -0800774 /* Optimise the no-wait case */
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700775 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700776 pt->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700777 timed_out = 1;
778 }
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700779
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700780 if (end_time && !timed_out)
Andrew Morton231f3d32010-10-27 15:34:53 -0700781 slack = select_estimate_accuracy(end_time);
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 for (;;) {
784 struct poll_list *walk;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300785 bool can_ll = false;
David Woodhouse9f729492006-01-18 17:44:05 -0800786
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700787 for (walk = list; walk != NULL; walk = walk->next) {
788 struct pollfd * pfd, * pfd_end;
789
790 pfd = walk->entries;
791 pfd_end = pfd + walk->len;
792 for (; pfd != pfd_end; pfd++) {
793 /*
794 * Fish for events. If we found one, record it
Hans Verkuil626cf232012-03-23 15:02:27 -0700795 * and kill poll_table->_qproc, so we don't
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700796 * needlessly register any other waiters after
797 * this. They'll get immediately deregistered
798 * when we break out and return.
799 */
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300800 if (do_pollfd(pfd, pt, &can_ll, ll_flag)) {
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700801 count++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700802 pt->_qproc = NULL;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300803 ll_flag = 0;
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700804 }
805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700807 /*
808 * All waiters have already been registered, so don't provide
Hans Verkuil626cf232012-03-23 15:02:27 -0700809 * a poll_table->_qproc to them on the next loop iteration.
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700810 */
Hans Verkuil626cf232012-03-23 15:02:27 -0700811 pt->_qproc = NULL;
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700812 if (!count) {
813 count = wait->error;
814 if (signal_pending(current))
815 count = -EINTR;
816 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700817 if (count || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 break;
David Woodhouse9f729492006-01-18 17:44:05 -0800819
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300820 if (can_ll && can_poll_ll(ll_time))
821 continue;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700822 /*
823 * If this is the first loop and we have a timeout
824 * given, then we convert to ktime_t and set the to
825 * pointer to the expiry value.
826 */
827 if (end_time && !to) {
828 expire = timespec_to_ktime(*end_time);
829 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800830 }
831
Tejun Heo5f820f62009-01-06 14:40:59 -0800832 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700833 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 return count;
836}
837
Andi Kleen70674f92006-03-28 01:56:33 -0800838#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
839 sizeof(struct pollfd))
840
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700841int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
842 struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
844 struct poll_wqueues table;
Oleg Nesterov252e5722007-10-16 23:26:17 -0700845 int err = -EFAULT, fdcount, len, size;
Jes Sorensen30c14e42006-03-31 11:18:57 -0500846 /* Allocate small arguments on the stack to save memory and be
847 faster - use long to make sure the buffer is aligned properly
848 on 64 bit archs to avoid unaligned access */
849 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
Oleg Nesterov252e5722007-10-16 23:26:17 -0700850 struct poll_list *const head = (struct poll_list *)stack_pps;
851 struct poll_list *walk = head;
852 unsigned long todo = nfds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
Jiri Slabyd554ed892010-03-05 13:42:42 -0800854 if (nfds > rlimit(RLIMIT_NOFILE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 return -EINVAL;
856
Oleg Nesterov252e5722007-10-16 23:26:17 -0700857 len = min_t(unsigned int, nfds, N_STACK_PPS);
858 for (;;) {
859 walk->next = NULL;
860 walk->len = len;
861 if (!len)
862 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Oleg Nesterov252e5722007-10-16 23:26:17 -0700864 if (copy_from_user(walk->entries, ufds + nfds-todo,
865 sizeof(struct pollfd) * walk->len))
866 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Oleg Nesterov252e5722007-10-16 23:26:17 -0700868 todo -= walk->len;
869 if (!todo)
870 break;
871
872 len = min(todo, POLLFD_PER_PAGE);
873 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
874 walk = walk->next = kmalloc(size, GFP_KERNEL);
875 if (!walk) {
876 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 goto out_fds;
878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
David Woodhouse9f729492006-01-18 17:44:05 -0800880
Oleg Nesterov252e5722007-10-16 23:26:17 -0700881 poll_initwait(&table);
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700882 fdcount = do_poll(nfds, head, &table, end_time);
Oleg Nesterov252e5722007-10-16 23:26:17 -0700883 poll_freewait(&table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Oleg Nesterov252e5722007-10-16 23:26:17 -0700885 for (walk = head; walk; walk = walk->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 struct pollfd *fds = walk->entries;
887 int j;
888
Oleg Nesterov252e5722007-10-16 23:26:17 -0700889 for (j = 0; j < walk->len; j++, ufds++)
890 if (__put_user(fds[j].revents, &ufds->revents))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 err = fdcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895out_fds:
Oleg Nesterov252e5722007-10-16 23:26:17 -0700896 walk = head->next;
897 while (walk) {
898 struct poll_list *pos = walk;
899 walk = walk->next;
900 kfree(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 return err;
904}
David Woodhouse9f729492006-01-18 17:44:05 -0800905
Chris Wright3075d9d2007-10-16 23:27:18 -0700906static long do_restart_poll(struct restart_block *restart_block)
907{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700908 struct pollfd __user *ufds = restart_block->poll.ufds;
909 int nfds = restart_block->poll.nfds;
910 struct timespec *to = NULL, end_time;
Chris Wright3075d9d2007-10-16 23:27:18 -0700911 int ret;
912
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700913 if (restart_block->poll.has_timeout) {
914 end_time.tv_sec = restart_block->poll.tv_sec;
915 end_time.tv_nsec = restart_block->poll.tv_nsec;
916 to = &end_time;
917 }
918
919 ret = do_sys_poll(ufds, nfds, to);
920
Chris Wright3075d9d2007-10-16 23:27:18 -0700921 if (ret == -EINTR) {
922 restart_block->fn = do_restart_poll;
Chris Wright3075d9d2007-10-16 23:27:18 -0700923 ret = -ERESTART_RESTARTBLOCK;
924 }
925 return ret;
926}
927
Heiko Carstens5a8a82b2009-01-14 14:14:25 +0100928SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
Linus Torvaldsfaf30902012-02-21 17:24:20 -0800929 int, timeout_msecs)
David Woodhouse9f729492006-01-18 17:44:05 -0800930{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700931 struct timespec end_time, *to = NULL;
Chris Wright3075d9d2007-10-16 23:27:18 -0700932 int ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800933
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700934 if (timeout_msecs >= 0) {
935 to = &end_time;
936 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
937 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
David Woodhouse9f729492006-01-18 17:44:05 -0800938 }
939
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700940 ret = do_sys_poll(ufds, nfds, to);
941
Chris Wright3075d9d2007-10-16 23:27:18 -0700942 if (ret == -EINTR) {
943 struct restart_block *restart_block;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700944
Chris Wright3075d9d2007-10-16 23:27:18 -0700945 restart_block = &current_thread_info()->restart_block;
946 restart_block->fn = do_restart_poll;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700947 restart_block->poll.ufds = ufds;
948 restart_block->poll.nfds = nfds;
949
950 if (timeout_msecs >= 0) {
951 restart_block->poll.tv_sec = end_time.tv_sec;
952 restart_block->poll.tv_nsec = end_time.tv_nsec;
953 restart_block->poll.has_timeout = 1;
954 } else
955 restart_block->poll.has_timeout = 0;
956
Chris Wright3075d9d2007-10-16 23:27:18 -0700957 ret = -ERESTART_RESTARTBLOCK;
958 }
959 return ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800960}
961
Heiko Carstensd4e82042009-01-14 14:14:34 +0100962SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
963 struct timespec __user *, tsp, const sigset_t __user *, sigmask,
964 size_t, sigsetsize)
David Woodhouse9f729492006-01-18 17:44:05 -0800965{
966 sigset_t ksigmask, sigsaved;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700967 struct timespec ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800968 int ret;
969
970 if (tsp) {
971 if (copy_from_user(&ts, tsp, sizeof(ts)))
972 return -EFAULT;
973
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700974 to = &end_time;
975 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
976 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800977 }
978
979 if (sigmask) {
980 /* XXX: Don't preclude handling different sized sigset_t's. */
981 if (sigsetsize != sizeof(sigset_t))
982 return -EINVAL;
983 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
984 return -EFAULT;
985
986 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
987 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
988 }
989
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700990 ret = do_sys_poll(ufds, nfds, to);
David Woodhouse9f729492006-01-18 17:44:05 -0800991
992 /* We can restart this syscall, usually */
993 if (ret == -EINTR) {
994 /*
995 * Don't restore the signal mask yet. Let do_signal() deliver
996 * the signal on the way back to userspace, before the signal
997 * mask is restored.
998 */
999 if (sigmask) {
1000 memcpy(&current->saved_sigmask, &sigsaved,
1001 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001002 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -08001003 }
1004 ret = -ERESTARTNOHAND;
1005 } else if (sigmask)
1006 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1007
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001008 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
David Woodhouse9f729492006-01-18 17:44:05 -08001009
1010 return ret;
1011}