blob: 55278a36a6582802edfd91600d311def49de17c0 [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>
Colin Cross9745cdb2013-05-06 23:50:17 +000030#include <linux/freezer.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +030031#include <net/busy_poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33#include <asm/uaccess.h>
34
Arjan van de Ven90d6e242008-09-01 15:55:35 -070035
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
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070048#define MAX_SLACK (100 * NSEC_PER_MSEC)
49
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070050static long __estimate_accuracy(struct timespec *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070051{
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070052 long slack;
Arjan van de Ven90d6e242008-09-01 15:55:35 -070053 int divfactor = 1000;
54
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070055 if (tv->tv_sec < 0)
56 return 0;
57
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070058 if (task_nice(current) > 0)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070059 divfactor = divfactor / 5;
60
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070061 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
62 return MAX_SLACK;
63
Arjan van de Ven90d6e242008-09-01 15:55:35 -070064 slack = tv->tv_nsec / divfactor;
65 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
66
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070067 if (slack > MAX_SLACK)
68 return MAX_SLACK;
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070069
Arjan van de Ven90d6e242008-09-01 15:55:35 -070070 return slack;
71}
72
Shawn Bohrer95aac7b2010-10-27 15:34:54 -070073long select_estimate_accuracy(struct timespec *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070074{
75 unsigned long ret;
76 struct timespec now;
77
78 /*
79 * Realtime tasks get a slack of 0 for obvious reasons.
80 */
81
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070082 if (rt_task(current))
Arjan van de Ven90d6e242008-09-01 15:55:35 -070083 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070095struct 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 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700116static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
117 poll_table *p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
119void poll_initwait(struct poll_wqueues *pwq)
120{
121 init_poll_funcptr(&pwq->pt, __pollwait);
Tejun Heo5f820f62009-01-06 14:40:59 -0800122 pwq->polling_task = current;
Guillaume Knispelb2add732009-08-15 19:30:24 +0200123 pwq->triggered = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 pwq->error = 0;
125 pwq->table = NULL;
Andi Kleen70674f92006-03-28 01:56:33 -0800126 pwq->inline_index = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128EXPORT_SYMBOL(poll_initwait);
129
Andi Kleen70674f92006-03-28 01:56:33 -0800130static void free_poll_entry(struct poll_table_entry *entry)
131{
WANG Congccf67802007-05-09 07:10:02 +0200132 remove_wait_queue(entry->wait_address, &entry->wait);
Andi Kleen70674f92006-03-28 01:56:33 -0800133 fput(entry->filp);
134}
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136void poll_freewait(struct poll_wqueues *pwq)
137{
138 struct poll_table_page * p = pwq->table;
Andi Kleen70674f92006-03-28 01:56:33 -0800139 int i;
140 for (i = 0; i < pwq->inline_index; i++)
141 free_poll_entry(pwq->inline_entries + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 while (p) {
143 struct poll_table_entry * entry;
144 struct poll_table_page *old;
145
146 entry = p->entry;
147 do {
148 entry--;
Andi Kleen70674f92006-03-28 01:56:33 -0800149 free_poll_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 } while (entry > p->entries);
151 old = p;
152 p = p->next;
153 free_page((unsigned long) old);
154 }
155}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156EXPORT_SYMBOL(poll_freewait);
157
Tejun Heo5f820f62009-01-06 14:40:59 -0800158static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 struct poll_table_page *table = p->table;
161
Andi Kleen70674f92006-03-28 01:56:33 -0800162 if (p->inline_index < N_INLINE_POLL_ENTRIES)
163 return p->inline_entries + p->inline_index++;
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 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;
Andi Kleen70674f92006-03-28 01:56:33 -0800171 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
173 new_table->entry = new_table->entries;
174 new_table->next = table;
175 p->table = new_table;
176 table = new_table;
177 }
178
Andi Kleen70674f92006-03-28 01:56:33 -0800179 return table->entry++;
180}
181
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700182static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
Tejun Heo5f820f62009-01-06 14:40:59 -0800183{
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 set_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
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700208static 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
Andi Kleen70674f92006-03-28 01:56:33 -0800218/* Add a new entry */
219static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
220 poll_table *p)
221{
Tejun Heo5f820f62009-01-06 14:40:59 -0800222 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
223 struct poll_table_entry *entry = poll_get_entry(pwq);
Andi Kleen70674f92006-03-28 01:56:33 -0800224 if (!entry)
225 return;
Al Virocb0942b2012-08-27 14:48:26 -0400226 entry->filp = get_file(filp);
Andi Kleen70674f92006-03-28 01:56:33 -0800227 entry->wait_address = wait_address;
Hans Verkuil626cf232012-03-23 15:02:27 -0700228 entry->key = p->_key;
Tejun Heo5f820f62009-01-06 14:40:59 -0800229 init_waitqueue_func_entry(&entry->wait, pollwake);
230 entry->wait.private = pwq;
WANG Congccf67802007-05-09 07:10:02 +0200231 add_wait_queue(wait_address, &entry->wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
Tejun Heo5f820f62009-01-06 14:40:59 -0800234int 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)
Colin Cross9745cdb2013-05-06 23:50:17 +0000241 rc = freezable_schedule_hrtimeout_range(expires, slack,
242 HRTIMER_MODE_ABS);
Tejun Heo5f820f62009-01-06 14:40:59 -0800243 __set_current_state(TASK_RUNNING);
244
245 /*
246 * Prepare for the next iteration.
247 *
248 * The following set_mb() serves two purposes. First, it's
249 * the counterpart rmb of the wmb in pollwake() such that data
250 * written before wake up is always visible after wake up.
251 * Second, the full barrier guarantees that triggered clearing
252 * doesn't pass event check of the next iteration. Note that
253 * this problem doesn't exist for the first iteration as
254 * add_wait_queue() has full barrier semantics.
255 */
256 set_mb(pwq->triggered, 0);
257
258 return rc;
259}
260EXPORT_SYMBOL(poll_schedule_timeout);
261
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700262/**
263 * poll_select_set_timeout - helper function to setup the timeout value
264 * @to: pointer to timespec variable for the final timeout
265 * @sec: seconds (from user space)
266 * @nsec: nanoseconds (from user space)
267 *
268 * Note, we do not use a timespec for the user space value here, That
269 * way we can use the function for timeval and compat interfaces as well.
270 *
271 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
272 */
273int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
274{
275 struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
276
277 if (!timespec_valid(&ts))
278 return -EINVAL;
279
280 /* Optimize for the zero timeout value here */
281 if (!sec && !nsec) {
282 to->tv_sec = to->tv_nsec = 0;
283 } else {
284 ktime_get_ts(to);
285 *to = timespec_add_safe(*to, ts);
286 }
287 return 0;
288}
289
290static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
291 int timeval, int ret)
292{
293 struct timespec rts;
294 struct timeval rtv;
295
296 if (!p)
297 return ret;
298
299 if (current->personality & STICKY_TIMEOUTS)
300 goto sticky;
301
302 /* No update for zero timeout */
303 if (!end_time->tv_sec && !end_time->tv_nsec)
304 return ret;
305
306 ktime_get_ts(&rts);
307 rts = timespec_sub(*end_time, rts);
308 if (rts.tv_sec < 0)
309 rts.tv_sec = rts.tv_nsec = 0;
310
311 if (timeval) {
Vasiliy Kulikov65329bf2011-01-12 17:00:00 -0800312 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
313 memset(&rtv, 0, sizeof(rtv));
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700314 rtv.tv_sec = rts.tv_sec;
315 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
316
317 if (!copy_to_user(p, &rtv, sizeof(rtv)))
318 return ret;
319
320 } else if (!copy_to_user(p, &rts, sizeof(rts)))
321 return ret;
322
323 /*
324 * If an application puts its timeval in read-only memory, we
325 * don't want the Linux-specific update to the timeval to
326 * cause a fault after the select has completed
327 * successfully. However, because we're not updating the
328 * timeval, we can't restart the system call.
329 */
330
331sticky:
332 if (ret == -ERESTARTNOHAND)
333 ret = -EINTR;
334 return ret;
335}
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337#define FDS_IN(fds, n) (fds->in + n)
338#define FDS_OUT(fds, n) (fds->out + n)
339#define FDS_EX(fds, n) (fds->ex + n)
340
341#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
342
343static int max_select_fd(unsigned long n, fd_set_bits *fds)
344{
345 unsigned long *open_fds;
346 unsigned long set;
347 int max;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700348 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350 /* handle last in-complete long-word first */
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400351 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
352 n /= BITS_PER_LONG;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700353 fdt = files_fdtable(current->files);
David Howells1fd36ad2012-02-16 17:49:54 +0000354 open_fds = fdt->open_fds + n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 max = 0;
356 if (set) {
357 set &= BITS(fds, n);
358 if (set) {
359 if (!(set & ~*open_fds))
360 goto get_max;
361 return -EBADF;
362 }
363 }
364 while (n) {
365 open_fds--;
366 n--;
367 set = BITS(fds, n);
368 if (!set)
369 continue;
370 if (set & ~*open_fds)
371 return -EBADF;
372 if (max)
373 continue;
374get_max:
375 do {
376 max++;
377 set >>= 1;
378 } while (set);
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400379 max += n * BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 }
381
382 return max;
383}
384
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
386#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
387#define POLLEX_SET (POLLPRI)
388
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700389static inline void wait_key_set(poll_table *wait, unsigned long in,
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300390 unsigned long out, unsigned long bit,
391 unsigned int ll_flag)
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700392{
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300393 wait->_key = POLLEX_SET | ll_flag;
Hans Verkuil626cf232012-03-23 15:02:27 -0700394 if (in & bit)
395 wait->_key |= POLLIN_SET;
396 if (out & bit)
397 wait->_key |= POLLOUT_SET;
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700398}
399
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700400int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700402 ktime_t expire, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 struct poll_wqueues table;
404 poll_table *wait;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700405 int retval, i, timed_out = 0;
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700406 unsigned long slack = 0;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300407 unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300408 unsigned long busy_end = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Dipankar Sarmab8359962005-09-09 13:04:14 -0700410 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 retval = max_select_fd(n, fds);
Dipankar Sarmab8359962005-09-09 13:04:14 -0700412 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
414 if (retval < 0)
415 return retval;
416 n = retval;
417
418 poll_initwait(&table);
419 wait = &table.pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700420 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700421 wait->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700422 timed_out = 1;
423 }
424
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700425 if (end_time && !timed_out)
Andrew Morton231f3d32010-10-27 15:34:53 -0700426 slack = select_estimate_accuracy(end_time);
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700427
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 retval = 0;
429 for (;;) {
430 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300431 bool can_busy_loop = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 inp = fds->in; outp = fds->out; exp = fds->ex;
434 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
435
436 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
437 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
438 unsigned long res_in = 0, res_out = 0, res_ex = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440 in = *inp++; out = *outp++; ex = *exp++;
441 all_bits = in | out | ex;
442 if (all_bits == 0) {
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400443 i += BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 continue;
445 }
446
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400447 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
Al Viro2903ff02012-08-28 12:52:22 -0400448 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 if (i >= n)
450 break;
451 if (!(bit & all_bits))
452 continue;
Al Viro2903ff02012-08-28 12:52:22 -0400453 f = fdget(i);
454 if (f.file) {
455 const struct file_operations *f_op;
456 f_op = f.file->f_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 mask = DEFAULT_POLLMASK;
Al Viro72c2d532013-09-22 16:27:52 -0400458 if (f_op->poll) {
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300459 wait_key_set(wait, in, out,
Eliezer Tamircbf55002013-07-08 16:20:34 +0300460 bit, busy_flag);
Al Viro2903ff02012-08-28 12:52:22 -0400461 mask = (*f_op->poll)(f.file, wait);
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700462 }
Al Viro2903ff02012-08-28 12:52:22 -0400463 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 if ((mask & POLLIN_SET) && (in & bit)) {
465 res_in |= bit;
466 retval++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700467 wait->_qproc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
469 if ((mask & POLLOUT_SET) && (out & bit)) {
470 res_out |= bit;
471 retval++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700472 wait->_qproc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474 if ((mask & POLLEX_SET) && (ex & bit)) {
475 res_ex |= bit;
476 retval++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700477 wait->_qproc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300479 /* got something, stop busy polling */
Eliezer Tamircbf55002013-07-08 16:20:34 +0300480 if (retval) {
481 can_busy_loop = false;
482 busy_flag = 0;
483
484 /*
485 * only remember a returned
486 * POLL_BUSY_LOOP if we asked for it
487 */
488 } else if (busy_flag & mask)
489 can_busy_loop = true;
490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 }
493 if (res_in)
494 *rinp = res_in;
495 if (res_out)
496 *routp = res_out;
497 if (res_ex)
498 *rexp = res_ex;
Linus Torvalds55d85382008-06-22 12:23:15 -0700499 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
Hans Verkuil626cf232012-03-23 15:02:27 -0700501 wait->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700502 if (retval || timed_out || signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 break;
Pavel Machekf5264482008-04-21 22:15:06 +0000504 if (table.error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 retval = table.error;
506 break;
507 }
David Woodhouse9f729492006-01-18 17:44:05 -0800508
Eliezer Tamircbf55002013-07-08 16:20:34 +0300509 /* only if found POLL_BUSY_LOOP sockets && not out of time */
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300510 if (can_busy_loop && !need_resched()) {
511 if (!busy_end) {
512 busy_end = busy_loop_end_time();
513 continue;
514 }
515 if (!busy_loop_timeout(busy_end))
516 continue;
517 }
518 busy_flag = 0;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300519
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700520 /*
521 * If this is the first loop and we have a timeout
522 * given, then we convert to ktime_t and set the to
523 * pointer to the expiry value.
524 */
525 if (end_time && !to) {
526 expire = timespec_to_ktime(*end_time);
527 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800528 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700529
Tejun Heo5f820f62009-01-06 14:40:59 -0800530 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
531 to, slack))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700532 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 poll_freewait(&table);
536
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return retval;
538}
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540/*
541 * We can actually return ERESTARTSYS instead of EINTR, but I'd
542 * like to be certain this leads to no problems. So I return
543 * EINTR just for safety.
544 *
545 * Update: ERESTARTSYS breaks at least the xview clock binary, so
546 * I'm trying ERESTARTNOHAND which restart only when you want to.
547 */
Al Viroa2dcb442008-04-23 14:05:15 -0400548int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700549 fd_set __user *exp, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
551 fd_set_bits fds;
Andrew Morton29ff2db2006-04-10 22:52:46 -0700552 void *bits;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800553 int ret, max_fds;
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700554 unsigned int size;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700555 struct fdtable *fdt;
Andi Kleen70674f92006-03-28 01:56:33 -0800556 /* Allocate small arguments on the stack to save memory and be faster */
Jes Sorensen30c14e42006-03-31 11:18:57 -0500557 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 ret = -EINVAL;
560 if (n < 0)
561 goto out_nofds;
562
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800563 /* max_fds can increase, so grab it once to avoid race */
Dipankar Sarmab8359962005-09-09 13:04:14 -0700564 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700565 fdt = files_fdtable(current->files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800566 max_fds = fdt->max_fds;
Dipankar Sarmab8359962005-09-09 13:04:14 -0700567 rcu_read_unlock();
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800568 if (n > max_fds)
569 n = max_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
571 /*
572 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
573 * since we used fdset we need to allocate memory in units of
574 * long-words.
575 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 size = FDS_BYTES(n);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700577 bits = stack_fds;
578 if (size > sizeof(stack_fds) / 6) {
579 /* Not enough space in on-stack array; must use kmalloc */
580 ret = -ENOMEM;
Andi Kleen70674f92006-03-28 01:56:33 -0800581 bits = kmalloc(6 * size, GFP_KERNEL);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700582 if (!bits)
583 goto out_nofds;
584 }
Andrew Morton29ff2db2006-04-10 22:52:46 -0700585 fds.in = bits;
586 fds.out = bits + size;
587 fds.ex = bits + 2*size;
588 fds.res_in = bits + 3*size;
589 fds.res_out = bits + 4*size;
590 fds.res_ex = bits + 5*size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 if ((ret = get_fd_set(n, inp, fds.in)) ||
593 (ret = get_fd_set(n, outp, fds.out)) ||
594 (ret = get_fd_set(n, exp, fds.ex)))
595 goto out;
596 zero_fd_set(n, fds.res_in);
597 zero_fd_set(n, fds.res_out);
598 zero_fd_set(n, fds.res_ex);
599
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700600 ret = do_select(n, &fds, end_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
602 if (ret < 0)
603 goto out;
604 if (!ret) {
605 ret = -ERESTARTNOHAND;
606 if (signal_pending(current))
607 goto out;
608 ret = 0;
609 }
610
611 if (set_fd_set(n, inp, fds.res_in) ||
612 set_fd_set(n, outp, fds.res_out) ||
613 set_fd_set(n, exp, fds.res_ex))
614 ret = -EFAULT;
615
616out:
Andi Kleen70674f92006-03-28 01:56:33 -0800617 if (bits != stack_fds)
618 kfree(bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619out_nofds:
620 return ret;
621}
622
Heiko Carstens5a8a82b2009-01-14 14:14:25 +0100623SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
624 fd_set __user *, exp, struct timeval __user *, tvp)
David Woodhouse9f729492006-01-18 17:44:05 -0800625{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700626 struct timespec end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800627 struct timeval tv;
628 int ret;
629
630 if (tvp) {
631 if (copy_from_user(&tv, tvp, sizeof(tv)))
632 return -EFAULT;
633
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700634 to = &end_time;
Arjan van de Ven4d36a9e2008-10-25 12:41:41 -0700635 if (poll_select_set_timeout(to,
636 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
637 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
David Woodhouse9f729492006-01-18 17:44:05 -0800638 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800639 }
640
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700641 ret = core_sys_select(n, inp, outp, exp, to);
642 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800643
644 return ret;
645}
646
Heiko Carstensc9da9f22009-01-14 14:13:57 +0100647static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
648 fd_set __user *exp, struct timespec __user *tsp,
649 const sigset_t __user *sigmask, size_t sigsetsize)
David Woodhouse9f729492006-01-18 17:44:05 -0800650{
David Woodhouse9f729492006-01-18 17:44:05 -0800651 sigset_t ksigmask, sigsaved;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700652 struct timespec ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800653 int ret;
654
655 if (tsp) {
656 if (copy_from_user(&ts, tsp, sizeof(ts)))
657 return -EFAULT;
658
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700659 to = &end_time;
660 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
David Woodhouse9f729492006-01-18 17:44:05 -0800661 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800662 }
663
664 if (sigmask) {
665 /* XXX: Don't preclude handling different sized sigset_t's. */
666 if (sigsetsize != sizeof(sigset_t))
667 return -EINVAL;
668 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
669 return -EFAULT;
670
671 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
672 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
673 }
674
Bernd Schmidt62568512009-01-13 22:14:48 +0100675 ret = core_sys_select(n, inp, outp, exp, to);
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700676 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800677
678 if (ret == -ERESTARTNOHAND) {
679 /*
680 * Don't restore the signal mask yet. Let do_signal() deliver
681 * the signal on the way back to userspace, before the signal
682 * mask is restored.
683 */
684 if (sigmask) {
685 memcpy(&current->saved_sigmask, &sigsaved,
686 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -0700687 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -0800688 }
689 } else if (sigmask)
690 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
691
692 return ret;
693}
694
695/*
696 * Most architectures can't handle 7-argument syscalls. So we provide a
697 * 6-argument version where the sixth argument is a pointer to a structure
698 * which has a pointer to the sigset_t itself followed by a size_t containing
699 * the sigset size.
700 */
Heiko Carstensd4e82042009-01-14 14:14:34 +0100701SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
702 fd_set __user *, exp, struct timespec __user *, tsp,
703 void __user *, sig)
David Woodhouse9f729492006-01-18 17:44:05 -0800704{
705 size_t sigsetsize = 0;
706 sigset_t __user *up = NULL;
707
708 if (sig) {
709 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
Al Viroe110ab92006-02-01 05:26:09 -0500710 || __get_user(up, (sigset_t __user * __user *)sig)
David Woodhouse9f729492006-01-18 17:44:05 -0800711 || __get_user(sigsetsize,
Al Viroe110ab92006-02-01 05:26:09 -0500712 (size_t __user *)(sig+sizeof(void *))))
David Woodhouse9f729492006-01-18 17:44:05 -0800713 return -EFAULT;
714 }
715
Heiko Carstensc9da9f22009-01-14 14:13:57 +0100716 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
David Woodhouse9f729492006-01-18 17:44:05 -0800717}
David Woodhouse9f729492006-01-18 17:44:05 -0800718
Christoph Hellwig5d0e5282010-03-10 15:21:13 -0800719#ifdef __ARCH_WANT_SYS_OLD_SELECT
720struct sel_arg_struct {
721 unsigned long n;
722 fd_set __user *inp, *outp, *exp;
723 struct timeval __user *tvp;
724};
725
726SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
727{
728 struct sel_arg_struct a;
729
730 if (copy_from_user(&a, arg, sizeof(a)))
731 return -EFAULT;
732 return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
733}
734#endif
735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736struct poll_list {
737 struct poll_list *next;
738 int len;
739 struct pollfd entries[0];
740};
741
742#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
743
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700744/*
745 * Fish for pollable events on the pollfd->fd file descriptor. We're only
746 * interested in events matching the pollfd->events mask, and the result
747 * matching that mask is both recorded in pollfd->revents and returned. The
748 * pwait poll_table will be used by the fd-provided poll handler for waiting,
Hans Verkuil626cf232012-03-23 15:02:27 -0700749 * if pwait->_qproc is non-NULL.
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700750 */
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300751static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait,
Eliezer Tamircbf55002013-07-08 16:20:34 +0300752 bool *can_busy_poll,
753 unsigned int busy_flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700755 unsigned int mask;
756 int fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700758 mask = 0;
759 fd = pollfd->fd;
760 if (fd >= 0) {
Al Viro2903ff02012-08-28 12:52:22 -0400761 struct fd f = fdget(fd);
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700762 mask = POLLNVAL;
Al Viro2903ff02012-08-28 12:52:22 -0400763 if (f.file) {
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700764 mask = DEFAULT_POLLMASK;
Al Viro72c2d532013-09-22 16:27:52 -0400765 if (f.file->f_op->poll) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700766 pwait->_key = pollfd->events|POLLERR|POLLHUP;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300767 pwait->_key |= busy_flag;
Al Viro2903ff02012-08-28 12:52:22 -0400768 mask = f.file->f_op->poll(f.file, pwait);
Eliezer Tamircbf55002013-07-08 16:20:34 +0300769 if (mask & busy_flag)
770 *can_busy_poll = true;
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700771 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700772 /* Mask out unneeded events. */
773 mask &= pollfd->events | POLLERR | POLLHUP;
Al Viro2903ff02012-08-28 12:52:22 -0400774 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700777 pollfd->revents = mask;
778
779 return mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
781
782static int do_poll(unsigned int nfds, struct poll_list *list,
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700783 struct poll_wqueues *wait, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 poll_table* pt = &wait->pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700786 ktime_t expire, *to = NULL;
787 int timed_out = 0, count = 0;
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700788 unsigned long slack = 0;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300789 unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300790 unsigned long busy_end = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
David Woodhouse9f729492006-01-18 17:44:05 -0800792 /* Optimise the no-wait case */
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700793 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700794 pt->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700795 timed_out = 1;
796 }
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700797
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700798 if (end_time && !timed_out)
Andrew Morton231f3d32010-10-27 15:34:53 -0700799 slack = select_estimate_accuracy(end_time);
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700800
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 for (;;) {
802 struct poll_list *walk;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300803 bool can_busy_loop = false;
David Woodhouse9f729492006-01-18 17:44:05 -0800804
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700805 for (walk = list; walk != NULL; walk = walk->next) {
806 struct pollfd * pfd, * pfd_end;
807
808 pfd = walk->entries;
809 pfd_end = pfd + walk->len;
810 for (; pfd != pfd_end; pfd++) {
811 /*
812 * Fish for events. If we found one, record it
Hans Verkuil626cf232012-03-23 15:02:27 -0700813 * and kill poll_table->_qproc, so we don't
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700814 * needlessly register any other waiters after
815 * this. They'll get immediately deregistered
816 * when we break out and return.
817 */
Eliezer Tamircbf55002013-07-08 16:20:34 +0300818 if (do_pollfd(pfd, pt, &can_busy_loop,
819 busy_flag)) {
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700820 count++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700821 pt->_qproc = NULL;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300822 /* found something, stop busy polling */
823 busy_flag = 0;
824 can_busy_loop = false;
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700825 }
826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700828 /*
829 * All waiters have already been registered, so don't provide
Hans Verkuil626cf232012-03-23 15:02:27 -0700830 * a poll_table->_qproc to them on the next loop iteration.
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700831 */
Hans Verkuil626cf232012-03-23 15:02:27 -0700832 pt->_qproc = NULL;
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700833 if (!count) {
834 count = wait->error;
835 if (signal_pending(current))
836 count = -EINTR;
837 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700838 if (count || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 break;
David Woodhouse9f729492006-01-18 17:44:05 -0800840
Eliezer Tamircbf55002013-07-08 16:20:34 +0300841 /* only if found POLL_BUSY_LOOP sockets && not out of time */
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300842 if (can_busy_loop && !need_resched()) {
843 if (!busy_end) {
844 busy_end = busy_loop_end_time();
845 continue;
846 }
847 if (!busy_loop_timeout(busy_end))
848 continue;
849 }
850 busy_flag = 0;
Eliezer Tamir91e2fd332013-06-28 15:59:35 +0300851
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700852 /*
853 * If this is the first loop and we have a timeout
854 * given, then we convert to ktime_t and set the to
855 * pointer to the expiry value.
856 */
857 if (end_time && !to) {
858 expire = timespec_to_ktime(*end_time);
859 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800860 }
861
Tejun Heo5f820f62009-01-06 14:40:59 -0800862 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700863 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 return count;
866}
867
Andi Kleen70674f92006-03-28 01:56:33 -0800868#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
869 sizeof(struct pollfd))
870
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700871int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
872 struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873{
874 struct poll_wqueues table;
Oleg Nesterov252e5722007-10-16 23:26:17 -0700875 int err = -EFAULT, fdcount, len, size;
Jes Sorensen30c14e42006-03-31 11:18:57 -0500876 /* Allocate small arguments on the stack to save memory and be
877 faster - use long to make sure the buffer is aligned properly
878 on 64 bit archs to avoid unaligned access */
879 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
Oleg Nesterov252e5722007-10-16 23:26:17 -0700880 struct poll_list *const head = (struct poll_list *)stack_pps;
881 struct poll_list *walk = head;
882 unsigned long todo = nfds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
Jiri Slabyd554ed892010-03-05 13:42:42 -0800884 if (nfds > rlimit(RLIMIT_NOFILE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 return -EINVAL;
886
Oleg Nesterov252e5722007-10-16 23:26:17 -0700887 len = min_t(unsigned int, nfds, N_STACK_PPS);
888 for (;;) {
889 walk->next = NULL;
890 walk->len = len;
891 if (!len)
892 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
Oleg Nesterov252e5722007-10-16 23:26:17 -0700894 if (copy_from_user(walk->entries, ufds + nfds-todo,
895 sizeof(struct pollfd) * walk->len))
896 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
Oleg Nesterov252e5722007-10-16 23:26:17 -0700898 todo -= walk->len;
899 if (!todo)
900 break;
901
902 len = min(todo, POLLFD_PER_PAGE);
903 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
904 walk = walk->next = kmalloc(size, GFP_KERNEL);
905 if (!walk) {
906 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 goto out_fds;
908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 }
David Woodhouse9f729492006-01-18 17:44:05 -0800910
Oleg Nesterov252e5722007-10-16 23:26:17 -0700911 poll_initwait(&table);
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700912 fdcount = do_poll(nfds, head, &table, end_time);
Oleg Nesterov252e5722007-10-16 23:26:17 -0700913 poll_freewait(&table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Oleg Nesterov252e5722007-10-16 23:26:17 -0700915 for (walk = head; walk; walk = walk->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 struct pollfd *fds = walk->entries;
917 int j;
918
Oleg Nesterov252e5722007-10-16 23:26:17 -0700919 for (j = 0; j < walk->len; j++, ufds++)
920 if (__put_user(fds[j].revents, &ufds->revents))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 err = fdcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925out_fds:
Oleg Nesterov252e5722007-10-16 23:26:17 -0700926 walk = head->next;
927 while (walk) {
928 struct poll_list *pos = walk;
929 walk = walk->next;
930 kfree(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 return err;
934}
David Woodhouse9f729492006-01-18 17:44:05 -0800935
Chris Wright3075d9d2007-10-16 23:27:18 -0700936static long do_restart_poll(struct restart_block *restart_block)
937{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700938 struct pollfd __user *ufds = restart_block->poll.ufds;
939 int nfds = restart_block->poll.nfds;
940 struct timespec *to = NULL, end_time;
Chris Wright3075d9d2007-10-16 23:27:18 -0700941 int ret;
942
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700943 if (restart_block->poll.has_timeout) {
944 end_time.tv_sec = restart_block->poll.tv_sec;
945 end_time.tv_nsec = restart_block->poll.tv_nsec;
946 to = &end_time;
947 }
948
949 ret = do_sys_poll(ufds, nfds, to);
950
Chris Wright3075d9d2007-10-16 23:27:18 -0700951 if (ret == -EINTR) {
952 restart_block->fn = do_restart_poll;
Chris Wright3075d9d2007-10-16 23:27:18 -0700953 ret = -ERESTART_RESTARTBLOCK;
954 }
955 return ret;
956}
957
Heiko Carstens5a8a82b2009-01-14 14:14:25 +0100958SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
Linus Torvaldsfaf30902012-02-21 17:24:20 -0800959 int, timeout_msecs)
David Woodhouse9f729492006-01-18 17:44:05 -0800960{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700961 struct timespec end_time, *to = NULL;
Chris Wright3075d9d2007-10-16 23:27:18 -0700962 int ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800963
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700964 if (timeout_msecs >= 0) {
965 to = &end_time;
966 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
967 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
David Woodhouse9f729492006-01-18 17:44:05 -0800968 }
969
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700970 ret = do_sys_poll(ufds, nfds, to);
971
Chris Wright3075d9d2007-10-16 23:27:18 -0700972 if (ret == -EINTR) {
973 struct restart_block *restart_block;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700974
Chris Wright3075d9d2007-10-16 23:27:18 -0700975 restart_block = &current_thread_info()->restart_block;
976 restart_block->fn = do_restart_poll;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700977 restart_block->poll.ufds = ufds;
978 restart_block->poll.nfds = nfds;
979
980 if (timeout_msecs >= 0) {
981 restart_block->poll.tv_sec = end_time.tv_sec;
982 restart_block->poll.tv_nsec = end_time.tv_nsec;
983 restart_block->poll.has_timeout = 1;
984 } else
985 restart_block->poll.has_timeout = 0;
986
Chris Wright3075d9d2007-10-16 23:27:18 -0700987 ret = -ERESTART_RESTARTBLOCK;
988 }
989 return ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800990}
991
Heiko Carstensd4e82042009-01-14 14:14:34 +0100992SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
993 struct timespec __user *, tsp, const sigset_t __user *, sigmask,
994 size_t, sigsetsize)
David Woodhouse9f729492006-01-18 17:44:05 -0800995{
996 sigset_t ksigmask, sigsaved;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700997 struct timespec ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800998 int ret;
999
1000 if (tsp) {
1001 if (copy_from_user(&ts, tsp, sizeof(ts)))
1002 return -EFAULT;
1003
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001004 to = &end_time;
1005 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1006 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -08001007 }
1008
1009 if (sigmask) {
1010 /* XXX: Don't preclude handling different sized sigset_t's. */
1011 if (sigsetsize != sizeof(sigset_t))
1012 return -EINVAL;
1013 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1014 return -EFAULT;
1015
1016 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1017 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1018 }
1019
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001020 ret = do_sys_poll(ufds, nfds, to);
David Woodhouse9f729492006-01-18 17:44:05 -08001021
1022 /* We can restart this syscall, usually */
1023 if (ret == -EINTR) {
1024 /*
1025 * Don't restore the signal mask yet. Let do_signal() deliver
1026 * the signal on the way back to userspace, before the signal
1027 * mask is restored.
1028 */
1029 if (sigmask) {
1030 memcpy(&current->saved_sigmask, &sigsaved,
1031 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -07001032 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -08001033 }
1034 ret = -ERESTARTNOHAND;
1035 } else if (sigmask)
1036 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1037
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001038 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
David Woodhouse9f729492006-01-18 17:44:05 -08001039
1040 return ret;
1041}