]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
022a1692 | 17 | #include <linux/kernel.h> |
a99bbaf5 | 18 | #include <linux/sched.h> |
1da177e4 | 19 | #include <linux/syscalls.h> |
630d9c47 | 20 | #include <linux/export.h> |
1da177e4 | 21 | #include <linux/slab.h> |
1da177e4 LT |
22 | #include <linux/poll.h> |
23 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ | |
24 | #include <linux/file.h> | |
9f3acc31 | 25 | #include <linux/fdtable.h> |
1da177e4 | 26 | #include <linux/fs.h> |
b835996f | 27 | #include <linux/rcupdate.h> |
8ff3e8e8 | 28 | #include <linux/hrtimer.h> |
8bd75c77 | 29 | #include <linux/sched/rt.h> |
9745cdb3 | 30 | #include <linux/freezer.h> |
076bb0c8 | 31 | #include <net/busy_poll.h> |
5f130705 | 32 | #include <linux/vmalloc.h> |
1da177e4 LT |
33 | |
34 | #include <asm/uaccess.h> | |
35 | ||
90d6e24a AV |
36 | |
37 | /* | |
38 | * Estimate expected accuracy in ns from a timeval. | |
39 | * | |
40 | * After quite a bit of churning around, we've settled on | |
41 | * a simple thing of taking 0.1% of the timeout as the | |
42 | * slack, with a cap of 100 msec. | |
43 | * "nice" tasks get a 0.5% slack instead. | |
44 | * | |
45 | * Consider this comment an open invitation to come up with even | |
46 | * better solutions.. | |
47 | */ | |
48 | ||
5ae87e79 GK |
49 | #define MAX_SLACK (100 * NSEC_PER_MSEC) |
50 | ||
96d2ab48 | 51 | static long __estimate_accuracy(struct timespec *tv) |
90d6e24a | 52 | { |
96d2ab48 | 53 | long slack; |
90d6e24a AV |
54 | int divfactor = 1000; |
55 | ||
5ae87e79 GK |
56 | if (tv->tv_sec < 0) |
57 | return 0; | |
58 | ||
4ce105d3 | 59 | if (task_nice(current) > 0) |
90d6e24a AV |
60 | divfactor = divfactor / 5; |
61 | ||
5ae87e79 GK |
62 | if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) |
63 | return MAX_SLACK; | |
64 | ||
90d6e24a AV |
65 | slack = tv->tv_nsec / divfactor; |
66 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); | |
67 | ||
5ae87e79 GK |
68 | if (slack > MAX_SLACK) |
69 | return MAX_SLACK; | |
96d2ab48 | 70 | |
90d6e24a AV |
71 | return slack; |
72 | } | |
73 | ||
95aac7b1 | 74 | long select_estimate_accuracy(struct timespec *tv) |
90d6e24a AV |
75 | { |
76 | unsigned long ret; | |
77 | struct timespec now; | |
78 | ||
79 | /* | |
80 | * Realtime tasks get a slack of 0 for obvious reasons. | |
81 | */ | |
82 | ||
4ce105d3 | 83 | if (rt_task(current)) |
90d6e24a AV |
84 | return 0; |
85 | ||
86 | ktime_get_ts(&now); | |
87 | now = timespec_sub(*tv, now); | |
88 | ret = __estimate_accuracy(&now); | |
89 | if (ret < current->timer_slack_ns) | |
90 | return current->timer_slack_ns; | |
91 | return ret; | |
92 | } | |
93 | ||
94 | ||
95 | ||
1da177e4 LT |
96 | struct poll_table_page { |
97 | struct poll_table_page * next; | |
98 | struct poll_table_entry * entry; | |
99 | struct poll_table_entry entries[0]; | |
100 | }; | |
101 | ||
102 | #define POLL_TABLE_FULL(table) \ | |
103 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
104 | ||
105 | /* | |
106 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
107 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
108 | * follow, but it should be free of race-conditions, and it's practical. If you | |
109 | * understand what I'm doing here, then you understand how the linux | |
110 | * sleep/wakeup mechanism works. | |
111 | * | |
112 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
113 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
114 | * as all select/poll functions have to call it to add an entry to the | |
115 | * poll table. | |
116 | */ | |
75c96f85 AB |
117 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
118 | poll_table *p); | |
1da177e4 LT |
119 | |
120 | void poll_initwait(struct poll_wqueues *pwq) | |
121 | { | |
122 | init_poll_funcptr(&pwq->pt, __pollwait); | |
5f820f64 | 123 | pwq->polling_task = current; |
b2add73d | 124 | pwq->triggered = 0; |
1da177e4 LT |
125 | pwq->error = 0; |
126 | pwq->table = NULL; | |
70674f95 | 127 | pwq->inline_index = 0; |
1da177e4 | 128 | } |
1da177e4 LT |
129 | EXPORT_SYMBOL(poll_initwait); |
130 | ||
70674f95 AK |
131 | static void free_poll_entry(struct poll_table_entry *entry) |
132 | { | |
ccf6780d | 133 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
134 | fput(entry->filp); |
135 | } | |
136 | ||
1da177e4 LT |
137 | void poll_freewait(struct poll_wqueues *pwq) |
138 | { | |
139 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
140 | int i; |
141 | for (i = 0; i < pwq->inline_index; i++) | |
142 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
143 | while (p) { |
144 | struct poll_table_entry * entry; | |
145 | struct poll_table_page *old; | |
146 | ||
147 | entry = p->entry; | |
148 | do { | |
149 | entry--; | |
70674f95 | 150 | free_poll_entry(entry); |
1da177e4 LT |
151 | } while (entry > p->entries); |
152 | old = p; | |
153 | p = p->next; | |
154 | free_page((unsigned long) old); | |
155 | } | |
156 | } | |
1da177e4 LT |
157 | EXPORT_SYMBOL(poll_freewait); |
158 | ||
5f820f64 | 159 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
1da177e4 | 160 | { |
1da177e4 LT |
161 | struct poll_table_page *table = p->table; |
162 | ||
70674f95 AK |
163 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
164 | return p->inline_entries + p->inline_index++; | |
165 | ||
1da177e4 LT |
166 | if (!table || POLL_TABLE_FULL(table)) { |
167 | struct poll_table_page *new_table; | |
168 | ||
169 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
170 | if (!new_table) { | |
171 | p->error = -ENOMEM; | |
70674f95 | 172 | return NULL; |
1da177e4 LT |
173 | } |
174 | new_table->entry = new_table->entries; | |
175 | new_table->next = table; | |
176 | p->table = new_table; | |
177 | table = new_table; | |
178 | } | |
179 | ||
70674f95 AK |
180 | return table->entry++; |
181 | } | |
182 | ||
4938d7e0 | 183 | static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) |
5f820f64 TH |
184 | { |
185 | struct poll_wqueues *pwq = wait->private; | |
186 | DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); | |
187 | ||
188 | /* | |
189 | * Although this function is called under waitqueue lock, LOCK | |
190 | * doesn't imply write barrier and the users expect write | |
191 | * barrier semantics on wakeup functions. The following | |
192 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
b92b8b35 | 193 | * and is paired with smp_store_mb() in poll_schedule_timeout. |
5f820f64 TH |
194 | */ |
195 | smp_wmb(); | |
196 | pwq->triggered = 1; | |
197 | ||
198 | /* | |
199 | * Perform the default wake up operation using a dummy | |
200 | * waitqueue. | |
201 | * | |
202 | * TODO: This is hacky but there currently is no interface to | |
203 | * pass in @sync. @sync is scheduled to be removed and once | |
204 | * that happens, wake_up_process() can be used directly. | |
205 | */ | |
206 | return default_wake_function(&dummy_wait, mode, sync, key); | |
207 | } | |
208 | ||
4938d7e0 ED |
209 | static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) |
210 | { | |
211 | struct poll_table_entry *entry; | |
212 | ||
213 | entry = container_of(wait, struct poll_table_entry, wait); | |
214 | if (key && !((unsigned long)key & entry->key)) | |
215 | return 0; | |
216 | return __pollwake(wait, mode, sync, key); | |
217 | } | |
218 | ||
70674f95 AK |
219 | /* Add a new entry */ |
220 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
221 | poll_table *p) | |
222 | { | |
5f820f64 TH |
223 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
224 | struct poll_table_entry *entry = poll_get_entry(pwq); | |
70674f95 AK |
225 | if (!entry) |
226 | return; | |
cb0942b8 | 227 | entry->filp = get_file(filp); |
70674f95 | 228 | entry->wait_address = wait_address; |
626cf236 | 229 | entry->key = p->_key; |
5f820f64 TH |
230 | init_waitqueue_func_entry(&entry->wait, pollwake); |
231 | entry->wait.private = pwq; | |
ccf6780d | 232 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
233 | } |
234 | ||
5f820f64 TH |
235 | int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
236 | ktime_t *expires, unsigned long slack) | |
237 | { | |
238 | int rc = -EINTR; | |
239 | ||
240 | set_current_state(state); | |
241 | if (!pwq->triggered) | |
59612d18 | 242 | rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); |
5f820f64 TH |
243 | __set_current_state(TASK_RUNNING); |
244 | ||
245 | /* | |
246 | * Prepare for the next iteration. | |
247 | * | |
b92b8b35 | 248 | * The following smp_store_mb() serves two purposes. First, it's |
5f820f64 TH |
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 | */ | |
b92b8b35 | 256 | smp_store_mb(pwq->triggered, 0); |
5f820f64 TH |
257 | |
258 | return rc; | |
259 | } | |
260 | EXPORT_SYMBOL(poll_schedule_timeout); | |
261 | ||
b773ad40 TG |
262 | /** |
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 | */ | |
273 | int 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 | ||
290 | static 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) { | |
65329bf4 VK |
312 | if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) |
313 | memset(&rtv, 0, sizeof(rtv)); | |
b773ad40 TG |
314 | 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 | ||
331 | sticky: | |
332 | if (ret == -ERESTARTNOHAND) | |
333 | ret = -EINTR; | |
334 | return ret; | |
335 | } | |
336 | ||
1da177e4 LT |
337 | #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 | ||
343 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
344 | { | |
345 | unsigned long *open_fds; | |
346 | unsigned long set; | |
347 | int max; | |
badf1662 | 348 | struct fdtable *fdt; |
1da177e4 LT |
349 | |
350 | /* handle last in-complete long-word first */ | |
8ded2bbc JB |
351 | set = ~(~0UL << (n & (BITS_PER_LONG-1))); |
352 | n /= BITS_PER_LONG; | |
badf1662 | 353 | fdt = files_fdtable(current->files); |
1fd36adc | 354 | open_fds = fdt->open_fds + n; |
1da177e4 LT |
355 | 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; | |
374 | get_max: | |
375 | do { | |
376 | max++; | |
377 | set >>= 1; | |
378 | } while (set); | |
8ded2bbc | 379 | max += n * BITS_PER_LONG; |
1da177e4 LT |
380 | } |
381 | ||
382 | return max; | |
383 | } | |
384 | ||
1da177e4 LT |
385 | #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) |
386 | #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) | |
387 | #define POLLEX_SET (POLLPRI) | |
388 | ||
4938d7e0 | 389 | static inline void wait_key_set(poll_table *wait, unsigned long in, |
2d48d67f ET |
390 | unsigned long out, unsigned long bit, |
391 | unsigned int ll_flag) | |
4938d7e0 | 392 | { |
2d48d67f | 393 | wait->_key = POLLEX_SET | ll_flag; |
626cf236 HV |
394 | if (in & bit) |
395 | wait->_key |= POLLIN_SET; | |
396 | if (out & bit) | |
397 | wait->_key |= POLLOUT_SET; | |
4938d7e0 ED |
398 | } |
399 | ||
8ff3e8e8 | 400 | int do_select(int n, fd_set_bits *fds, struct timespec *end_time) |
1da177e4 | 401 | { |
8ff3e8e8 | 402 | ktime_t expire, *to = NULL; |
1da177e4 LT |
403 | struct poll_wqueues table; |
404 | poll_table *wait; | |
8ff3e8e8 | 405 | int retval, i, timed_out = 0; |
90d6e24a | 406 | unsigned long slack = 0; |
cbf55001 | 407 | unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
76b1e9b9 | 408 | unsigned long busy_end = 0; |
1da177e4 | 409 | |
b835996f | 410 | rcu_read_lock(); |
1da177e4 | 411 | retval = max_select_fd(n, fds); |
b835996f | 412 | rcu_read_unlock(); |
1da177e4 LT |
413 | |
414 | if (retval < 0) | |
415 | return retval; | |
416 | n = retval; | |
417 | ||
418 | poll_initwait(&table); | |
419 | wait = &table.pt; | |
8ff3e8e8 | 420 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 421 | wait->_qproc = NULL; |
8ff3e8e8 AV |
422 | timed_out = 1; |
423 | } | |
424 | ||
96d2ab48 | 425 | if (end_time && !timed_out) |
231f3d39 | 426 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 427 | |
1da177e4 LT |
428 | retval = 0; |
429 | for (;;) { | |
430 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
cbf55001 | 431 | bool can_busy_loop = false; |
1da177e4 | 432 | |
1da177e4 LT |
433 | 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; | |
1da177e4 LT |
439 | |
440 | in = *inp++; out = *outp++; ex = *exp++; | |
441 | all_bits = in | out | ex; | |
442 | if (all_bits == 0) { | |
8ded2bbc | 443 | i += BITS_PER_LONG; |
1da177e4 LT |
444 | continue; |
445 | } | |
446 | ||
8ded2bbc | 447 | for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { |
2903ff01 | 448 | struct fd f; |
1da177e4 LT |
449 | if (i >= n) |
450 | break; | |
451 | if (!(bit & all_bits)) | |
452 | continue; | |
2903ff01 AV |
453 | f = fdget(i); |
454 | if (f.file) { | |
455 | const struct file_operations *f_op; | |
456 | f_op = f.file->f_op; | |
1da177e4 | 457 | mask = DEFAULT_POLLMASK; |
72c2d531 | 458 | if (f_op->poll) { |
2d48d67f | 459 | wait_key_set(wait, in, out, |
cbf55001 | 460 | bit, busy_flag); |
2903ff01 | 461 | mask = (*f_op->poll)(f.file, wait); |
4938d7e0 | 462 | } |
2903ff01 | 463 | fdput(f); |
1da177e4 LT |
464 | if ((mask & POLLIN_SET) && (in & bit)) { |
465 | res_in |= bit; | |
466 | retval++; | |
626cf236 | 467 | wait->_qproc = NULL; |
1da177e4 LT |
468 | } |
469 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
470 | res_out |= bit; | |
471 | retval++; | |
626cf236 | 472 | wait->_qproc = NULL; |
1da177e4 LT |
473 | } |
474 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
475 | res_ex |= bit; | |
476 | retval++; | |
626cf236 | 477 | wait->_qproc = NULL; |
1da177e4 | 478 | } |
2d48d67f | 479 | /* got something, stop busy polling */ |
cbf55001 ET |
480 | 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 | ||
1da177e4 | 491 | } |
1da177e4 LT |
492 | } |
493 | if (res_in) | |
494 | *rinp = res_in; | |
495 | if (res_out) | |
496 | *routp = res_out; | |
497 | if (res_ex) | |
498 | *rexp = res_ex; | |
55d85384 | 499 | cond_resched(); |
1da177e4 | 500 | } |
626cf236 | 501 | wait->_qproc = NULL; |
8ff3e8e8 | 502 | if (retval || timed_out || signal_pending(current)) |
1da177e4 | 503 | break; |
f5264481 | 504 | if (table.error) { |
1da177e4 LT |
505 | retval = table.error; |
506 | break; | |
507 | } | |
9f72949f | 508 | |
cbf55001 | 509 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 ET |
510 | 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; | |
2d48d67f | 519 | |
8ff3e8e8 AV |
520 | /* |
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; | |
9f72949f | 528 | } |
8ff3e8e8 | 529 | |
5f820f64 TH |
530 | if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE, |
531 | to, slack)) | |
8ff3e8e8 | 532 | timed_out = 1; |
1da177e4 | 533 | } |
1da177e4 LT |
534 | |
535 | poll_freewait(&table); | |
536 | ||
1da177e4 LT |
537 | return retval; |
538 | } | |
539 | ||
1da177e4 LT |
540 | /* |
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 | */ | |
a2dcb44c | 548 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
8ff3e8e8 | 549 | fd_set __user *exp, struct timespec *end_time) |
1da177e4 LT |
550 | { |
551 | fd_set_bits fds; | |
29ff2db5 | 552 | void *bits; |
bbea9f69 | 553 | int ret, max_fds; |
5f130705 | 554 | size_t size, alloc_size; |
badf1662 | 555 | struct fdtable *fdt; |
70674f95 | 556 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 557 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 558 | |
1da177e4 LT |
559 | ret = -EINVAL; |
560 | if (n < 0) | |
561 | goto out_nofds; | |
562 | ||
bbea9f69 | 563 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 564 | rcu_read_lock(); |
badf1662 | 565 | fdt = files_fdtable(current->files); |
bbea9f69 | 566 | max_fds = fdt->max_fds; |
b835996f | 567 | rcu_read_unlock(); |
bbea9f69 VL |
568 | if (n > max_fds) |
569 | n = max_fds; | |
1da177e4 LT |
570 | |
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 | */ | |
1da177e4 | 576 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
577 | bits = stack_fds; |
578 | if (size > sizeof(stack_fds) / 6) { | |
579 | /* Not enough space in on-stack array; must use kmalloc */ | |
580 | ret = -ENOMEM; | |
5f130705 VB |
581 | if (size > (SIZE_MAX / 6)) |
582 | goto out_nofds; | |
583 | ||
584 | alloc_size = 6 * size; | |
585 | bits = kmalloc(alloc_size, GFP_KERNEL|__GFP_NOWARN); | |
586 | if (!bits && alloc_size > PAGE_SIZE) | |
587 | bits = vmalloc(alloc_size); | |
588 | ||
b04eb6aa MBJ |
589 | if (!bits) |
590 | goto out_nofds; | |
591 | } | |
29ff2db5 AM |
592 | fds.in = bits; |
593 | fds.out = bits + size; | |
594 | fds.ex = bits + 2*size; | |
595 | fds.res_in = bits + 3*size; | |
596 | fds.res_out = bits + 4*size; | |
597 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
598 | |
599 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
600 | (ret = get_fd_set(n, outp, fds.out)) || | |
601 | (ret = get_fd_set(n, exp, fds.ex))) | |
602 | goto out; | |
603 | zero_fd_set(n, fds.res_in); | |
604 | zero_fd_set(n, fds.res_out); | |
605 | zero_fd_set(n, fds.res_ex); | |
606 | ||
8ff3e8e8 | 607 | ret = do_select(n, &fds, end_time); |
1da177e4 LT |
608 | |
609 | if (ret < 0) | |
610 | goto out; | |
611 | if (!ret) { | |
612 | ret = -ERESTARTNOHAND; | |
613 | if (signal_pending(current)) | |
614 | goto out; | |
615 | ret = 0; | |
616 | } | |
617 | ||
618 | if (set_fd_set(n, inp, fds.res_in) || | |
619 | set_fd_set(n, outp, fds.res_out) || | |
620 | set_fd_set(n, exp, fds.res_ex)) | |
621 | ret = -EFAULT; | |
622 | ||
623 | out: | |
70674f95 | 624 | if (bits != stack_fds) |
5f130705 | 625 | kvfree(bits); |
1da177e4 LT |
626 | out_nofds: |
627 | return ret; | |
628 | } | |
629 | ||
5a8a82b1 HC |
630 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
631 | fd_set __user *, exp, struct timeval __user *, tvp) | |
9f72949f | 632 | { |
8ff3e8e8 | 633 | struct timespec end_time, *to = NULL; |
9f72949f DW |
634 | struct timeval tv; |
635 | int ret; | |
636 | ||
637 | if (tvp) { | |
638 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
639 | return -EFAULT; | |
640 | ||
8ff3e8e8 | 641 | to = &end_time; |
4d36a9e6 AV |
642 | if (poll_select_set_timeout(to, |
643 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
644 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
9f72949f | 645 | return -EINVAL; |
9f72949f DW |
646 | } |
647 | ||
8ff3e8e8 AV |
648 | ret = core_sys_select(n, inp, outp, exp, to); |
649 | ret = poll_select_copy_remaining(&end_time, tvp, 1, ret); | |
9f72949f DW |
650 | |
651 | return ret; | |
652 | } | |
653 | ||
c9da9f21 HC |
654 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
655 | fd_set __user *exp, struct timespec __user *tsp, | |
656 | const sigset_t __user *sigmask, size_t sigsetsize) | |
9f72949f | 657 | { |
9f72949f | 658 | sigset_t ksigmask, sigsaved; |
8ff3e8e8 | 659 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
660 | int ret; |
661 | ||
662 | if (tsp) { | |
663 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
664 | return -EFAULT; | |
665 | ||
8ff3e8e8 AV |
666 | to = &end_time; |
667 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
9f72949f | 668 | return -EINVAL; |
9f72949f DW |
669 | } |
670 | ||
671 | if (sigmask) { | |
672 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
673 | if (sigsetsize != sizeof(sigset_t)) | |
674 | return -EINVAL; | |
675 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
676 | return -EFAULT; | |
677 | ||
678 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
679 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
680 | } | |
681 | ||
62568510 | 682 | ret = core_sys_select(n, inp, outp, exp, to); |
8ff3e8e8 | 683 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
684 | |
685 | if (ret == -ERESTARTNOHAND) { | |
686 | /* | |
687 | * Don't restore the signal mask yet. Let do_signal() deliver | |
688 | * the signal on the way back to userspace, before the signal | |
689 | * mask is restored. | |
690 | */ | |
691 | if (sigmask) { | |
692 | memcpy(¤t->saved_sigmask, &sigsaved, | |
693 | sizeof(sigsaved)); | |
4e4c22c7 | 694 | set_restore_sigmask(); |
9f72949f DW |
695 | } |
696 | } else if (sigmask) | |
697 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
698 | ||
699 | return ret; | |
700 | } | |
701 | ||
702 | /* | |
703 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
704 | * 6-argument version where the sixth argument is a pointer to a structure | |
705 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
706 | * the sigset size. | |
707 | */ | |
d4e82042 HC |
708 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
709 | fd_set __user *, exp, struct timespec __user *, tsp, | |
710 | void __user *, sig) | |
9f72949f DW |
711 | { |
712 | size_t sigsetsize = 0; | |
713 | sigset_t __user *up = NULL; | |
714 | ||
715 | if (sig) { | |
716 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
e110ab94 | 717 | || __get_user(up, (sigset_t __user * __user *)sig) |
9f72949f | 718 | || __get_user(sigsetsize, |
e110ab94 | 719 | (size_t __user *)(sig+sizeof(void *)))) |
9f72949f DW |
720 | return -EFAULT; |
721 | } | |
722 | ||
c9da9f21 | 723 | return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize); |
9f72949f | 724 | } |
9f72949f | 725 | |
5d0e5283 CH |
726 | #ifdef __ARCH_WANT_SYS_OLD_SELECT |
727 | struct sel_arg_struct { | |
728 | unsigned long n; | |
729 | fd_set __user *inp, *outp, *exp; | |
730 | struct timeval __user *tvp; | |
731 | }; | |
732 | ||
733 | SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) | |
734 | { | |
735 | struct sel_arg_struct a; | |
736 | ||
737 | if (copy_from_user(&a, arg, sizeof(a))) | |
738 | return -EFAULT; | |
739 | return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp); | |
740 | } | |
741 | #endif | |
742 | ||
1da177e4 LT |
743 | struct poll_list { |
744 | struct poll_list *next; | |
745 | int len; | |
746 | struct pollfd entries[0]; | |
747 | }; | |
748 | ||
749 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
750 | ||
4a4b69f7 VL |
751 | /* |
752 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
753 | * interested in events matching the pollfd->events mask, and the result | |
754 | * matching that mask is both recorded in pollfd->revents and returned. The | |
755 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
626cf236 | 756 | * if pwait->_qproc is non-NULL. |
4a4b69f7 | 757 | */ |
2d48d67f | 758 | static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait, |
cbf55001 ET |
759 | bool *can_busy_poll, |
760 | unsigned int busy_flag) | |
1da177e4 | 761 | { |
4a4b69f7 VL |
762 | unsigned int mask; |
763 | int fd; | |
764 | ||
765 | mask = 0; | |
766 | fd = pollfd->fd; | |
767 | if (fd >= 0) { | |
2903ff01 | 768 | struct fd f = fdget(fd); |
4a4b69f7 | 769 | mask = POLLNVAL; |
2903ff01 | 770 | if (f.file) { |
4a4b69f7 | 771 | mask = DEFAULT_POLLMASK; |
72c2d531 | 772 | if (f.file->f_op->poll) { |
626cf236 | 773 | pwait->_key = pollfd->events|POLLERR|POLLHUP; |
cbf55001 | 774 | pwait->_key |= busy_flag; |
2903ff01 | 775 | mask = f.file->f_op->poll(f.file, pwait); |
cbf55001 ET |
776 | if (mask & busy_flag) |
777 | *can_busy_poll = true; | |
4938d7e0 | 778 | } |
4a4b69f7 VL |
779 | /* Mask out unneeded events. */ |
780 | mask &= pollfd->events | POLLERR | POLLHUP; | |
2903ff01 | 781 | fdput(f); |
1da177e4 | 782 | } |
1da177e4 | 783 | } |
4a4b69f7 VL |
784 | pollfd->revents = mask; |
785 | ||
786 | return mask; | |
1da177e4 LT |
787 | } |
788 | ||
789 | static int do_poll(unsigned int nfds, struct poll_list *list, | |
8ff3e8e8 | 790 | struct poll_wqueues *wait, struct timespec *end_time) |
1da177e4 | 791 | { |
1da177e4 | 792 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
793 | ktime_t expire, *to = NULL; |
794 | int timed_out = 0, count = 0; | |
90d6e24a | 795 | unsigned long slack = 0; |
cbf55001 | 796 | unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
76b1e9b9 | 797 | unsigned long busy_end = 0; |
1da177e4 | 798 | |
9f72949f | 799 | /* Optimise the no-wait case */ |
8ff3e8e8 | 800 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 801 | pt->_qproc = NULL; |
8ff3e8e8 AV |
802 | timed_out = 1; |
803 | } | |
9bf084f7 | 804 | |
96d2ab48 | 805 | if (end_time && !timed_out) |
231f3d39 | 806 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 807 | |
1da177e4 LT |
808 | for (;;) { |
809 | struct poll_list *walk; | |
cbf55001 | 810 | bool can_busy_loop = false; |
9f72949f | 811 | |
4a4b69f7 VL |
812 | for (walk = list; walk != NULL; walk = walk->next) { |
813 | struct pollfd * pfd, * pfd_end; | |
814 | ||
815 | pfd = walk->entries; | |
816 | pfd_end = pfd + walk->len; | |
817 | for (; pfd != pfd_end; pfd++) { | |
818 | /* | |
819 | * Fish for events. If we found one, record it | |
626cf236 | 820 | * and kill poll_table->_qproc, so we don't |
4a4b69f7 VL |
821 | * needlessly register any other waiters after |
822 | * this. They'll get immediately deregistered | |
823 | * when we break out and return. | |
824 | */ | |
cbf55001 ET |
825 | if (do_pollfd(pfd, pt, &can_busy_loop, |
826 | busy_flag)) { | |
4a4b69f7 | 827 | count++; |
626cf236 | 828 | pt->_qproc = NULL; |
cbf55001 ET |
829 | /* found something, stop busy polling */ |
830 | busy_flag = 0; | |
831 | can_busy_loop = false; | |
4a4b69f7 VL |
832 | } |
833 | } | |
1da177e4 | 834 | } |
4a4b69f7 VL |
835 | /* |
836 | * All waiters have already been registered, so don't provide | |
626cf236 | 837 | * a poll_table->_qproc to them on the next loop iteration. |
4a4b69f7 | 838 | */ |
626cf236 | 839 | pt->_qproc = NULL; |
9bf084f7 ON |
840 | if (!count) { |
841 | count = wait->error; | |
842 | if (signal_pending(current)) | |
843 | count = -EINTR; | |
844 | } | |
8ff3e8e8 | 845 | if (count || timed_out) |
1da177e4 | 846 | break; |
9f72949f | 847 | |
cbf55001 | 848 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 ET |
849 | if (can_busy_loop && !need_resched()) { |
850 | if (!busy_end) { | |
851 | busy_end = busy_loop_end_time(); | |
852 | continue; | |
853 | } | |
854 | if (!busy_loop_timeout(busy_end)) | |
855 | continue; | |
856 | } | |
857 | busy_flag = 0; | |
91e2fd33 | 858 | |
8ff3e8e8 AV |
859 | /* |
860 | * If this is the first loop and we have a timeout | |
861 | * given, then we convert to ktime_t and set the to | |
862 | * pointer to the expiry value. | |
863 | */ | |
864 | if (end_time && !to) { | |
865 | expire = timespec_to_ktime(*end_time); | |
866 | to = &expire; | |
9f72949f DW |
867 | } |
868 | ||
5f820f64 | 869 | if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) |
8ff3e8e8 | 870 | timed_out = 1; |
1da177e4 | 871 | } |
1da177e4 LT |
872 | return count; |
873 | } | |
874 | ||
70674f95 AK |
875 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
876 | sizeof(struct pollfd)) | |
877 | ||
8ff3e8e8 AV |
878 | int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
879 | struct timespec *end_time) | |
1da177e4 LT |
880 | { |
881 | struct poll_wqueues table; | |
252e5725 | 882 | int err = -EFAULT, fdcount, len, size; |
30c14e40 JS |
883 | /* Allocate small arguments on the stack to save memory and be |
884 | faster - use long to make sure the buffer is aligned properly | |
885 | on 64 bit archs to avoid unaligned access */ | |
886 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
887 | struct poll_list *const head = (struct poll_list *)stack_pps; |
888 | struct poll_list *walk = head; | |
889 | unsigned long todo = nfds; | |
1da177e4 | 890 | |
d554ed89 | 891 | if (nfds > rlimit(RLIMIT_NOFILE)) |
1da177e4 LT |
892 | return -EINVAL; |
893 | ||
252e5725 ON |
894 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
895 | for (;;) { | |
896 | walk->next = NULL; | |
897 | walk->len = len; | |
898 | if (!len) | |
899 | break; | |
1da177e4 | 900 | |
252e5725 ON |
901 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
902 | sizeof(struct pollfd) * walk->len)) | |
903 | goto out_fds; | |
904 | ||
905 | todo -= walk->len; | |
906 | if (!todo) | |
907 | break; | |
1da177e4 | 908 | |
252e5725 ON |
909 | len = min(todo, POLLFD_PER_PAGE); |
910 | size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; | |
911 | walk = walk->next = kmalloc(size, GFP_KERNEL); | |
912 | if (!walk) { | |
913 | err = -ENOMEM; | |
1da177e4 LT |
914 | goto out_fds; |
915 | } | |
1da177e4 | 916 | } |
9f72949f | 917 | |
252e5725 | 918 | poll_initwait(&table); |
8ff3e8e8 | 919 | fdcount = do_poll(nfds, head, &table, end_time); |
252e5725 | 920 | poll_freewait(&table); |
1da177e4 | 921 | |
252e5725 | 922 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
923 | struct pollfd *fds = walk->entries; |
924 | int j; | |
925 | ||
252e5725 ON |
926 | for (j = 0; j < walk->len; j++, ufds++) |
927 | if (__put_user(fds[j].revents, &ufds->revents)) | |
1da177e4 | 928 | goto out_fds; |
1da177e4 | 929 | } |
252e5725 | 930 | |
1da177e4 | 931 | err = fdcount; |
1da177e4 | 932 | out_fds: |
252e5725 ON |
933 | walk = head->next; |
934 | while (walk) { | |
935 | struct poll_list *pos = walk; | |
936 | walk = walk->next; | |
937 | kfree(pos); | |
1da177e4 | 938 | } |
252e5725 | 939 | |
1da177e4 LT |
940 | return err; |
941 | } | |
9f72949f | 942 | |
3075d9da CW |
943 | static long do_restart_poll(struct restart_block *restart_block) |
944 | { | |
8ff3e8e8 AV |
945 | struct pollfd __user *ufds = restart_block->poll.ufds; |
946 | int nfds = restart_block->poll.nfds; | |
947 | struct timespec *to = NULL, end_time; | |
3075d9da CW |
948 | int ret; |
949 | ||
8ff3e8e8 AV |
950 | if (restart_block->poll.has_timeout) { |
951 | end_time.tv_sec = restart_block->poll.tv_sec; | |
952 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
953 | to = &end_time; | |
954 | } | |
955 | ||
956 | ret = do_sys_poll(ufds, nfds, to); | |
957 | ||
3075d9da CW |
958 | if (ret == -EINTR) { |
959 | restart_block->fn = do_restart_poll; | |
3075d9da CW |
960 | ret = -ERESTART_RESTARTBLOCK; |
961 | } | |
962 | return ret; | |
963 | } | |
964 | ||
5a8a82b1 | 965 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
faf30900 | 966 | int, timeout_msecs) |
9f72949f | 967 | { |
8ff3e8e8 | 968 | struct timespec end_time, *to = NULL; |
3075d9da | 969 | int ret; |
9f72949f | 970 | |
8ff3e8e8 AV |
971 | if (timeout_msecs >= 0) { |
972 | to = &end_time; | |
973 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
974 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
975 | } |
976 | ||
8ff3e8e8 AV |
977 | ret = do_sys_poll(ufds, nfds, to); |
978 | ||
3075d9da CW |
979 | if (ret == -EINTR) { |
980 | struct restart_block *restart_block; | |
8ff3e8e8 | 981 | |
f56141e3 | 982 | restart_block = ¤t->restart_block; |
3075d9da | 983 | restart_block->fn = do_restart_poll; |
8ff3e8e8 AV |
984 | restart_block->poll.ufds = ufds; |
985 | restart_block->poll.nfds = nfds; | |
986 | ||
987 | if (timeout_msecs >= 0) { | |
988 | restart_block->poll.tv_sec = end_time.tv_sec; | |
989 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
990 | restart_block->poll.has_timeout = 1; | |
991 | } else | |
992 | restart_block->poll.has_timeout = 0; | |
993 | ||
3075d9da CW |
994 | ret = -ERESTART_RESTARTBLOCK; |
995 | } | |
996 | return ret; | |
9f72949f DW |
997 | } |
998 | ||
d4e82042 HC |
999 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
1000 | struct timespec __user *, tsp, const sigset_t __user *, sigmask, | |
1001 | size_t, sigsetsize) | |
9f72949f DW |
1002 | { |
1003 | sigset_t ksigmask, sigsaved; | |
8ff3e8e8 | 1004 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
1005 | int ret; |
1006 | ||
1007 | if (tsp) { | |
1008 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
1009 | return -EFAULT; | |
1010 | ||
8ff3e8e8 AV |
1011 | to = &end_time; |
1012 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1013 | return -EINVAL; | |
9f72949f DW |
1014 | } |
1015 | ||
1016 | if (sigmask) { | |
1017 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
1018 | if (sigsetsize != sizeof(sigset_t)) | |
1019 | return -EINVAL; | |
1020 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1021 | return -EFAULT; | |
1022 | ||
1023 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
1024 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1025 | } | |
1026 | ||
8ff3e8e8 | 1027 | ret = do_sys_poll(ufds, nfds, to); |
9f72949f DW |
1028 | |
1029 | /* We can restart this syscall, usually */ | |
1030 | if (ret == -EINTR) { | |
1031 | /* | |
1032 | * Don't restore the signal mask yet. Let do_signal() deliver | |
1033 | * the signal on the way back to userspace, before the signal | |
1034 | * mask is restored. | |
1035 | */ | |
1036 | if (sigmask) { | |
1037 | memcpy(¤t->saved_sigmask, &sigsaved, | |
1038 | sizeof(sigsaved)); | |
4e4c22c7 | 1039 | set_restore_sigmask(); |
9f72949f DW |
1040 | } |
1041 | ret = -ERESTARTNOHAND; | |
1042 | } else if (sigmask) | |
1043 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1044 | ||
8ff3e8e8 | 1045 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
1046 | |
1047 | return ret; | |
1048 | } |