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> |
1da177e4 LT |
18 | #include <linux/syscalls.h> |
19 | #include <linux/module.h> | |
20 | #include <linux/slab.h> | |
1da177e4 LT |
21 | #include <linux/poll.h> |
22 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ | |
23 | #include <linux/file.h> | |
9f3acc31 | 24 | #include <linux/fdtable.h> |
1da177e4 | 25 | #include <linux/fs.h> |
b835996f | 26 | #include <linux/rcupdate.h> |
8ff3e8e8 | 27 | #include <linux/hrtimer.h> |
1da177e4 LT |
28 | |
29 | #include <asm/uaccess.h> | |
30 | ||
90d6e24a AV |
31 | |
32 | /* | |
33 | * Estimate expected accuracy in ns from a timeval. | |
34 | * | |
35 | * After quite a bit of churning around, we've settled on | |
36 | * a simple thing of taking 0.1% of the timeout as the | |
37 | * slack, with a cap of 100 msec. | |
38 | * "nice" tasks get a 0.5% slack instead. | |
39 | * | |
40 | * Consider this comment an open invitation to come up with even | |
41 | * better solutions.. | |
42 | */ | |
43 | ||
5ae87e79 GK |
44 | #define MAX_SLACK (100 * NSEC_PER_MSEC) |
45 | ||
96d2ab48 | 46 | static long __estimate_accuracy(struct timespec *tv) |
90d6e24a | 47 | { |
96d2ab48 | 48 | long slack; |
90d6e24a AV |
49 | int divfactor = 1000; |
50 | ||
5ae87e79 GK |
51 | if (tv->tv_sec < 0) |
52 | return 0; | |
53 | ||
4ce105d3 | 54 | if (task_nice(current) > 0) |
90d6e24a AV |
55 | divfactor = divfactor / 5; |
56 | ||
5ae87e79 GK |
57 | if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) |
58 | return MAX_SLACK; | |
59 | ||
90d6e24a AV |
60 | slack = tv->tv_nsec / divfactor; |
61 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); | |
62 | ||
5ae87e79 GK |
63 | if (slack > MAX_SLACK) |
64 | return MAX_SLACK; | |
96d2ab48 | 65 | |
90d6e24a AV |
66 | return slack; |
67 | } | |
68 | ||
96d2ab48 | 69 | static long estimate_accuracy(struct timespec *tv) |
90d6e24a AV |
70 | { |
71 | unsigned long ret; | |
72 | struct timespec now; | |
73 | ||
74 | /* | |
75 | * Realtime tasks get a slack of 0 for obvious reasons. | |
76 | */ | |
77 | ||
4ce105d3 | 78 | if (rt_task(current)) |
90d6e24a AV |
79 | return 0; |
80 | ||
81 | ktime_get_ts(&now); | |
82 | now = timespec_sub(*tv, now); | |
83 | ret = __estimate_accuracy(&now); | |
84 | if (ret < current->timer_slack_ns) | |
85 | return current->timer_slack_ns; | |
86 | return ret; | |
87 | } | |
88 | ||
89 | ||
90 | ||
1da177e4 LT |
91 | struct poll_table_page { |
92 | struct poll_table_page * next; | |
93 | struct poll_table_entry * entry; | |
94 | struct poll_table_entry entries[0]; | |
95 | }; | |
96 | ||
97 | #define POLL_TABLE_FULL(table) \ | |
98 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
99 | ||
100 | /* | |
101 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
102 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
103 | * follow, but it should be free of race-conditions, and it's practical. If you | |
104 | * understand what I'm doing here, then you understand how the linux | |
105 | * sleep/wakeup mechanism works. | |
106 | * | |
107 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
108 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
109 | * as all select/poll functions have to call it to add an entry to the | |
110 | * poll table. | |
111 | */ | |
75c96f85 AB |
112 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
113 | poll_table *p); | |
1da177e4 LT |
114 | |
115 | void poll_initwait(struct poll_wqueues *pwq) | |
116 | { | |
117 | init_poll_funcptr(&pwq->pt, __pollwait); | |
5f820f64 | 118 | pwq->polling_task = current; |
b2add73d | 119 | pwq->triggered = 0; |
1da177e4 LT |
120 | pwq->error = 0; |
121 | pwq->table = NULL; | |
70674f95 | 122 | pwq->inline_index = 0; |
1da177e4 | 123 | } |
1da177e4 LT |
124 | EXPORT_SYMBOL(poll_initwait); |
125 | ||
70674f95 AK |
126 | static void free_poll_entry(struct poll_table_entry *entry) |
127 | { | |
ccf6780d | 128 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
129 | fput(entry->filp); |
130 | } | |
131 | ||
1da177e4 LT |
132 | void poll_freewait(struct poll_wqueues *pwq) |
133 | { | |
134 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
135 | int i; |
136 | for (i = 0; i < pwq->inline_index; i++) | |
137 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
138 | while (p) { |
139 | struct poll_table_entry * entry; | |
140 | struct poll_table_page *old; | |
141 | ||
142 | entry = p->entry; | |
143 | do { | |
144 | entry--; | |
70674f95 | 145 | free_poll_entry(entry); |
1da177e4 LT |
146 | } while (entry > p->entries); |
147 | old = p; | |
148 | p = p->next; | |
149 | free_page((unsigned long) old); | |
150 | } | |
151 | } | |
1da177e4 LT |
152 | EXPORT_SYMBOL(poll_freewait); |
153 | ||
5f820f64 | 154 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
1da177e4 | 155 | { |
1da177e4 LT |
156 | struct poll_table_page *table = p->table; |
157 | ||
70674f95 AK |
158 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
159 | return p->inline_entries + p->inline_index++; | |
160 | ||
1da177e4 LT |
161 | if (!table || POLL_TABLE_FULL(table)) { |
162 | struct poll_table_page *new_table; | |
163 | ||
164 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
165 | if (!new_table) { | |
166 | p->error = -ENOMEM; | |
70674f95 | 167 | return NULL; |
1da177e4 LT |
168 | } |
169 | new_table->entry = new_table->entries; | |
170 | new_table->next = table; | |
171 | p->table = new_table; | |
172 | table = new_table; | |
173 | } | |
174 | ||
70674f95 AK |
175 | return table->entry++; |
176 | } | |
177 | ||
4938d7e0 | 178 | static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) |
5f820f64 TH |
179 | { |
180 | struct poll_wqueues *pwq = wait->private; | |
181 | DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); | |
182 | ||
183 | /* | |
184 | * Although this function is called under waitqueue lock, LOCK | |
185 | * doesn't imply write barrier and the users expect write | |
186 | * barrier semantics on wakeup functions. The following | |
187 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
188 | * and is paired with set_mb() in poll_schedule_timeout. | |
189 | */ | |
190 | smp_wmb(); | |
191 | pwq->triggered = 1; | |
192 | ||
193 | /* | |
194 | * Perform the default wake up operation using a dummy | |
195 | * waitqueue. | |
196 | * | |
197 | * TODO: This is hacky but there currently is no interface to | |
198 | * pass in @sync. @sync is scheduled to be removed and once | |
199 | * that happens, wake_up_process() can be used directly. | |
200 | */ | |
201 | return default_wake_function(&dummy_wait, mode, sync, key); | |
202 | } | |
203 | ||
4938d7e0 ED |
204 | static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) |
205 | { | |
206 | struct poll_table_entry *entry; | |
207 | ||
208 | entry = container_of(wait, struct poll_table_entry, wait); | |
209 | if (key && !((unsigned long)key & entry->key)) | |
210 | return 0; | |
211 | return __pollwake(wait, mode, sync, key); | |
212 | } | |
213 | ||
70674f95 AK |
214 | /* Add a new entry */ |
215 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
216 | poll_table *p) | |
217 | { | |
5f820f64 TH |
218 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
219 | struct poll_table_entry *entry = poll_get_entry(pwq); | |
70674f95 AK |
220 | if (!entry) |
221 | return; | |
222 | get_file(filp); | |
223 | entry->filp = filp; | |
224 | entry->wait_address = wait_address; | |
4938d7e0 | 225 | entry->key = p->key; |
5f820f64 TH |
226 | init_waitqueue_func_entry(&entry->wait, pollwake); |
227 | entry->wait.private = pwq; | |
ccf6780d | 228 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
229 | } |
230 | ||
5f820f64 TH |
231 | int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
232 | ktime_t *expires, unsigned long slack) | |
233 | { | |
234 | int rc = -EINTR; | |
235 | ||
236 | set_current_state(state); | |
237 | if (!pwq->triggered) | |
238 | rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); | |
239 | __set_current_state(TASK_RUNNING); | |
240 | ||
241 | /* | |
242 | * Prepare for the next iteration. | |
243 | * | |
244 | * The following set_mb() serves two purposes. First, it's | |
245 | * the counterpart rmb of the wmb in pollwake() such that data | |
246 | * written before wake up is always visible after wake up. | |
247 | * Second, the full barrier guarantees that triggered clearing | |
248 | * doesn't pass event check of the next iteration. Note that | |
249 | * this problem doesn't exist for the first iteration as | |
250 | * add_wait_queue() has full barrier semantics. | |
251 | */ | |
252 | set_mb(pwq->triggered, 0); | |
253 | ||
254 | return rc; | |
255 | } | |
256 | EXPORT_SYMBOL(poll_schedule_timeout); | |
257 | ||
b773ad40 TG |
258 | /** |
259 | * poll_select_set_timeout - helper function to setup the timeout value | |
260 | * @to: pointer to timespec variable for the final timeout | |
261 | * @sec: seconds (from user space) | |
262 | * @nsec: nanoseconds (from user space) | |
263 | * | |
264 | * Note, we do not use a timespec for the user space value here, That | |
265 | * way we can use the function for timeval and compat interfaces as well. | |
266 | * | |
267 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
268 | */ | |
269 | int poll_select_set_timeout(struct timespec *to, long sec, long nsec) | |
270 | { | |
271 | struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec}; | |
272 | ||
273 | if (!timespec_valid(&ts)) | |
274 | return -EINVAL; | |
275 | ||
276 | /* Optimize for the zero timeout value here */ | |
277 | if (!sec && !nsec) { | |
278 | to->tv_sec = to->tv_nsec = 0; | |
279 | } else { | |
280 | ktime_get_ts(to); | |
281 | *to = timespec_add_safe(*to, ts); | |
282 | } | |
283 | return 0; | |
284 | } | |
285 | ||
286 | static int poll_select_copy_remaining(struct timespec *end_time, void __user *p, | |
287 | int timeval, int ret) | |
288 | { | |
289 | struct timespec rts; | |
290 | struct timeval rtv; | |
291 | ||
292 | if (!p) | |
293 | return ret; | |
294 | ||
295 | if (current->personality & STICKY_TIMEOUTS) | |
296 | goto sticky; | |
297 | ||
298 | /* No update for zero timeout */ | |
299 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
300 | return ret; | |
301 | ||
302 | ktime_get_ts(&rts); | |
303 | rts = timespec_sub(*end_time, rts); | |
304 | if (rts.tv_sec < 0) | |
305 | rts.tv_sec = rts.tv_nsec = 0; | |
306 | ||
307 | if (timeval) { | |
308 | rtv.tv_sec = rts.tv_sec; | |
309 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
310 | ||
311 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
312 | return ret; | |
313 | ||
314 | } else if (!copy_to_user(p, &rts, sizeof(rts))) | |
315 | return ret; | |
316 | ||
317 | /* | |
318 | * If an application puts its timeval in read-only memory, we | |
319 | * don't want the Linux-specific update to the timeval to | |
320 | * cause a fault after the select has completed | |
321 | * successfully. However, because we're not updating the | |
322 | * timeval, we can't restart the system call. | |
323 | */ | |
324 | ||
325 | sticky: | |
326 | if (ret == -ERESTARTNOHAND) | |
327 | ret = -EINTR; | |
328 | return ret; | |
329 | } | |
330 | ||
1da177e4 LT |
331 | #define FDS_IN(fds, n) (fds->in + n) |
332 | #define FDS_OUT(fds, n) (fds->out + n) | |
333 | #define FDS_EX(fds, n) (fds->ex + n) | |
334 | ||
335 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
336 | ||
337 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
338 | { | |
339 | unsigned long *open_fds; | |
340 | unsigned long set; | |
341 | int max; | |
badf1662 | 342 | struct fdtable *fdt; |
1da177e4 LT |
343 | |
344 | /* handle last in-complete long-word first */ | |
345 | set = ~(~0UL << (n & (__NFDBITS-1))); | |
346 | n /= __NFDBITS; | |
badf1662 DS |
347 | fdt = files_fdtable(current->files); |
348 | open_fds = fdt->open_fds->fds_bits+n; | |
1da177e4 LT |
349 | max = 0; |
350 | if (set) { | |
351 | set &= BITS(fds, n); | |
352 | if (set) { | |
353 | if (!(set & ~*open_fds)) | |
354 | goto get_max; | |
355 | return -EBADF; | |
356 | } | |
357 | } | |
358 | while (n) { | |
359 | open_fds--; | |
360 | n--; | |
361 | set = BITS(fds, n); | |
362 | if (!set) | |
363 | continue; | |
364 | if (set & ~*open_fds) | |
365 | return -EBADF; | |
366 | if (max) | |
367 | continue; | |
368 | get_max: | |
369 | do { | |
370 | max++; | |
371 | set >>= 1; | |
372 | } while (set); | |
373 | max += n * __NFDBITS; | |
374 | } | |
375 | ||
376 | return max; | |
377 | } | |
378 | ||
1da177e4 LT |
379 | #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) |
380 | #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) | |
381 | #define POLLEX_SET (POLLPRI) | |
382 | ||
4938d7e0 ED |
383 | static inline void wait_key_set(poll_table *wait, unsigned long in, |
384 | unsigned long out, unsigned long bit) | |
385 | { | |
386 | if (wait) { | |
387 | wait->key = POLLEX_SET; | |
388 | if (in & bit) | |
389 | wait->key |= POLLIN_SET; | |
390 | if (out & bit) | |
391 | wait->key |= POLLOUT_SET; | |
392 | } | |
393 | } | |
394 | ||
8ff3e8e8 | 395 | int do_select(int n, fd_set_bits *fds, struct timespec *end_time) |
1da177e4 | 396 | { |
8ff3e8e8 | 397 | ktime_t expire, *to = NULL; |
1da177e4 LT |
398 | struct poll_wqueues table; |
399 | poll_table *wait; | |
8ff3e8e8 | 400 | int retval, i, timed_out = 0; |
90d6e24a | 401 | unsigned long slack = 0; |
1da177e4 | 402 | |
b835996f | 403 | rcu_read_lock(); |
1da177e4 | 404 | retval = max_select_fd(n, fds); |
b835996f | 405 | rcu_read_unlock(); |
1da177e4 LT |
406 | |
407 | if (retval < 0) | |
408 | return retval; | |
409 | n = retval; | |
410 | ||
411 | poll_initwait(&table); | |
412 | wait = &table.pt; | |
8ff3e8e8 | 413 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
1da177e4 | 414 | wait = NULL; |
8ff3e8e8 AV |
415 | timed_out = 1; |
416 | } | |
417 | ||
96d2ab48 | 418 | if (end_time && !timed_out) |
90d6e24a AV |
419 | slack = estimate_accuracy(end_time); |
420 | ||
1da177e4 LT |
421 | retval = 0; |
422 | for (;;) { | |
423 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
424 | ||
1da177e4 LT |
425 | inp = fds->in; outp = fds->out; exp = fds->ex; |
426 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
427 | ||
428 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
429 | unsigned long in, out, ex, all_bits, bit = 1, mask, j; | |
430 | unsigned long res_in = 0, res_out = 0, res_ex = 0; | |
99ac48f5 | 431 | const struct file_operations *f_op = NULL; |
1da177e4 LT |
432 | struct file *file = NULL; |
433 | ||
434 | in = *inp++; out = *outp++; ex = *exp++; | |
435 | all_bits = in | out | ex; | |
436 | if (all_bits == 0) { | |
437 | i += __NFDBITS; | |
438 | continue; | |
439 | } | |
440 | ||
441 | for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) { | |
e4a1f129 | 442 | int fput_needed; |
1da177e4 LT |
443 | if (i >= n) |
444 | break; | |
445 | if (!(bit & all_bits)) | |
446 | continue; | |
e4a1f129 | 447 | file = fget_light(i, &fput_needed); |
1da177e4 LT |
448 | if (file) { |
449 | f_op = file->f_op; | |
450 | mask = DEFAULT_POLLMASK; | |
4938d7e0 ED |
451 | if (f_op && f_op->poll) { |
452 | wait_key_set(wait, in, out, bit); | |
453 | mask = (*f_op->poll)(file, wait); | |
454 | } | |
e4a1f129 | 455 | fput_light(file, fput_needed); |
1da177e4 LT |
456 | if ((mask & POLLIN_SET) && (in & bit)) { |
457 | res_in |= bit; | |
458 | retval++; | |
4938d7e0 | 459 | wait = NULL; |
1da177e4 LT |
460 | } |
461 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
462 | res_out |= bit; | |
463 | retval++; | |
4938d7e0 | 464 | wait = NULL; |
1da177e4 LT |
465 | } |
466 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
467 | res_ex |= bit; | |
468 | retval++; | |
4938d7e0 | 469 | wait = NULL; |
1da177e4 LT |
470 | } |
471 | } | |
1da177e4 LT |
472 | } |
473 | if (res_in) | |
474 | *rinp = res_in; | |
475 | if (res_out) | |
476 | *routp = res_out; | |
477 | if (res_ex) | |
478 | *rexp = res_ex; | |
55d85384 | 479 | cond_resched(); |
1da177e4 LT |
480 | } |
481 | wait = NULL; | |
8ff3e8e8 | 482 | if (retval || timed_out || signal_pending(current)) |
1da177e4 | 483 | break; |
f5264481 | 484 | if (table.error) { |
1da177e4 LT |
485 | retval = table.error; |
486 | break; | |
487 | } | |
9f72949f | 488 | |
8ff3e8e8 AV |
489 | /* |
490 | * If this is the first loop and we have a timeout | |
491 | * given, then we convert to ktime_t and set the to | |
492 | * pointer to the expiry value. | |
493 | */ | |
494 | if (end_time && !to) { | |
495 | expire = timespec_to_ktime(*end_time); | |
496 | to = &expire; | |
9f72949f | 497 | } |
8ff3e8e8 | 498 | |
5f820f64 TH |
499 | if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE, |
500 | to, slack)) | |
8ff3e8e8 | 501 | timed_out = 1; |
1da177e4 | 502 | } |
1da177e4 LT |
503 | |
504 | poll_freewait(&table); | |
505 | ||
1da177e4 LT |
506 | return retval; |
507 | } | |
508 | ||
1da177e4 LT |
509 | /* |
510 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
511 | * like to be certain this leads to no problems. So I return | |
512 | * EINTR just for safety. | |
513 | * | |
514 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
515 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
516 | */ | |
517 | #define MAX_SELECT_SECONDS \ | |
518 | ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1) | |
519 | ||
a2dcb44c | 520 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
8ff3e8e8 | 521 | fd_set __user *exp, struct timespec *end_time) |
1da177e4 LT |
522 | { |
523 | fd_set_bits fds; | |
29ff2db5 | 524 | void *bits; |
bbea9f69 | 525 | int ret, max_fds; |
b04eb6aa | 526 | unsigned int size; |
badf1662 | 527 | struct fdtable *fdt; |
70674f95 | 528 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 529 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 530 | |
1da177e4 LT |
531 | ret = -EINVAL; |
532 | if (n < 0) | |
533 | goto out_nofds; | |
534 | ||
bbea9f69 | 535 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 536 | rcu_read_lock(); |
badf1662 | 537 | fdt = files_fdtable(current->files); |
bbea9f69 | 538 | max_fds = fdt->max_fds; |
b835996f | 539 | rcu_read_unlock(); |
bbea9f69 VL |
540 | if (n > max_fds) |
541 | n = max_fds; | |
1da177e4 LT |
542 | |
543 | /* | |
544 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
545 | * since we used fdset we need to allocate memory in units of | |
546 | * long-words. | |
547 | */ | |
1da177e4 | 548 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
549 | bits = stack_fds; |
550 | if (size > sizeof(stack_fds) / 6) { | |
551 | /* Not enough space in on-stack array; must use kmalloc */ | |
552 | ret = -ENOMEM; | |
70674f95 | 553 | bits = kmalloc(6 * size, GFP_KERNEL); |
b04eb6aa MBJ |
554 | if (!bits) |
555 | goto out_nofds; | |
556 | } | |
29ff2db5 AM |
557 | fds.in = bits; |
558 | fds.out = bits + size; | |
559 | fds.ex = bits + 2*size; | |
560 | fds.res_in = bits + 3*size; | |
561 | fds.res_out = bits + 4*size; | |
562 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
563 | |
564 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
565 | (ret = get_fd_set(n, outp, fds.out)) || | |
566 | (ret = get_fd_set(n, exp, fds.ex))) | |
567 | goto out; | |
568 | zero_fd_set(n, fds.res_in); | |
569 | zero_fd_set(n, fds.res_out); | |
570 | zero_fd_set(n, fds.res_ex); | |
571 | ||
8ff3e8e8 | 572 | ret = do_select(n, &fds, end_time); |
1da177e4 LT |
573 | |
574 | if (ret < 0) | |
575 | goto out; | |
576 | if (!ret) { | |
577 | ret = -ERESTARTNOHAND; | |
578 | if (signal_pending(current)) | |
579 | goto out; | |
580 | ret = 0; | |
581 | } | |
582 | ||
583 | if (set_fd_set(n, inp, fds.res_in) || | |
584 | set_fd_set(n, outp, fds.res_out) || | |
585 | set_fd_set(n, exp, fds.res_ex)) | |
586 | ret = -EFAULT; | |
587 | ||
588 | out: | |
70674f95 AK |
589 | if (bits != stack_fds) |
590 | kfree(bits); | |
1da177e4 LT |
591 | out_nofds: |
592 | return ret; | |
593 | } | |
594 | ||
5a8a82b1 HC |
595 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
596 | fd_set __user *, exp, struct timeval __user *, tvp) | |
9f72949f | 597 | { |
8ff3e8e8 | 598 | struct timespec end_time, *to = NULL; |
9f72949f DW |
599 | struct timeval tv; |
600 | int ret; | |
601 | ||
602 | if (tvp) { | |
603 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
604 | return -EFAULT; | |
605 | ||
8ff3e8e8 | 606 | to = &end_time; |
4d36a9e6 AV |
607 | if (poll_select_set_timeout(to, |
608 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
609 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
9f72949f | 610 | return -EINVAL; |
9f72949f DW |
611 | } |
612 | ||
8ff3e8e8 AV |
613 | ret = core_sys_select(n, inp, outp, exp, to); |
614 | ret = poll_select_copy_remaining(&end_time, tvp, 1, ret); | |
9f72949f DW |
615 | |
616 | return ret; | |
617 | } | |
618 | ||
f3de272b | 619 | #ifdef HAVE_SET_RESTORE_SIGMASK |
c9da9f21 HC |
620 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
621 | fd_set __user *exp, struct timespec __user *tsp, | |
622 | const sigset_t __user *sigmask, size_t sigsetsize) | |
9f72949f | 623 | { |
9f72949f | 624 | sigset_t ksigmask, sigsaved; |
8ff3e8e8 | 625 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
626 | int ret; |
627 | ||
628 | if (tsp) { | |
629 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
630 | return -EFAULT; | |
631 | ||
8ff3e8e8 AV |
632 | to = &end_time; |
633 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
9f72949f | 634 | return -EINVAL; |
9f72949f DW |
635 | } |
636 | ||
637 | if (sigmask) { | |
638 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
639 | if (sigsetsize != sizeof(sigset_t)) | |
640 | return -EINVAL; | |
641 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
642 | return -EFAULT; | |
643 | ||
644 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
645 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
646 | } | |
647 | ||
62568510 | 648 | ret = core_sys_select(n, inp, outp, exp, to); |
8ff3e8e8 | 649 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
650 | |
651 | if (ret == -ERESTARTNOHAND) { | |
652 | /* | |
653 | * Don't restore the signal mask yet. Let do_signal() deliver | |
654 | * the signal on the way back to userspace, before the signal | |
655 | * mask is restored. | |
656 | */ | |
657 | if (sigmask) { | |
658 | memcpy(¤t->saved_sigmask, &sigsaved, | |
659 | sizeof(sigsaved)); | |
4e4c22c7 | 660 | set_restore_sigmask(); |
9f72949f DW |
661 | } |
662 | } else if (sigmask) | |
663 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
664 | ||
665 | return ret; | |
666 | } | |
667 | ||
668 | /* | |
669 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
670 | * 6-argument version where the sixth argument is a pointer to a structure | |
671 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
672 | * the sigset size. | |
673 | */ | |
d4e82042 HC |
674 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
675 | fd_set __user *, exp, struct timespec __user *, tsp, | |
676 | void __user *, sig) | |
9f72949f DW |
677 | { |
678 | size_t sigsetsize = 0; | |
679 | sigset_t __user *up = NULL; | |
680 | ||
681 | if (sig) { | |
682 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
e110ab94 | 683 | || __get_user(up, (sigset_t __user * __user *)sig) |
9f72949f | 684 | || __get_user(sigsetsize, |
e110ab94 | 685 | (size_t __user *)(sig+sizeof(void *)))) |
9f72949f DW |
686 | return -EFAULT; |
687 | } | |
688 | ||
c9da9f21 | 689 | return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize); |
9f72949f | 690 | } |
f3de272b | 691 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
9f72949f | 692 | |
1da177e4 LT |
693 | struct poll_list { |
694 | struct poll_list *next; | |
695 | int len; | |
696 | struct pollfd entries[0]; | |
697 | }; | |
698 | ||
699 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
700 | ||
4a4b69f7 VL |
701 | /* |
702 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
703 | * interested in events matching the pollfd->events mask, and the result | |
704 | * matching that mask is both recorded in pollfd->revents and returned. The | |
705 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
706 | * if non-NULL. | |
707 | */ | |
708 | static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait) | |
1da177e4 | 709 | { |
4a4b69f7 VL |
710 | unsigned int mask; |
711 | int fd; | |
712 | ||
713 | mask = 0; | |
714 | fd = pollfd->fd; | |
715 | if (fd >= 0) { | |
716 | int fput_needed; | |
717 | struct file * file; | |
718 | ||
719 | file = fget_light(fd, &fput_needed); | |
720 | mask = POLLNVAL; | |
721 | if (file != NULL) { | |
722 | mask = DEFAULT_POLLMASK; | |
4938d7e0 ED |
723 | if (file->f_op && file->f_op->poll) { |
724 | if (pwait) | |
725 | pwait->key = pollfd->events | | |
726 | POLLERR | POLLHUP; | |
4a4b69f7 | 727 | mask = file->f_op->poll(file, pwait); |
4938d7e0 | 728 | } |
4a4b69f7 VL |
729 | /* Mask out unneeded events. */ |
730 | mask &= pollfd->events | POLLERR | POLLHUP; | |
731 | fput_light(file, fput_needed); | |
1da177e4 | 732 | } |
1da177e4 | 733 | } |
4a4b69f7 VL |
734 | pollfd->revents = mask; |
735 | ||
736 | return mask; | |
1da177e4 LT |
737 | } |
738 | ||
739 | static int do_poll(unsigned int nfds, struct poll_list *list, | |
8ff3e8e8 | 740 | struct poll_wqueues *wait, struct timespec *end_time) |
1da177e4 | 741 | { |
1da177e4 | 742 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
743 | ktime_t expire, *to = NULL; |
744 | int timed_out = 0, count = 0; | |
90d6e24a | 745 | unsigned long slack = 0; |
1da177e4 | 746 | |
9f72949f | 747 | /* Optimise the no-wait case */ |
8ff3e8e8 | 748 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
1da177e4 | 749 | pt = NULL; |
8ff3e8e8 AV |
750 | timed_out = 1; |
751 | } | |
9bf084f7 | 752 | |
96d2ab48 | 753 | if (end_time && !timed_out) |
90d6e24a AV |
754 | slack = estimate_accuracy(end_time); |
755 | ||
1da177e4 LT |
756 | for (;;) { |
757 | struct poll_list *walk; | |
9f72949f | 758 | |
4a4b69f7 VL |
759 | for (walk = list; walk != NULL; walk = walk->next) { |
760 | struct pollfd * pfd, * pfd_end; | |
761 | ||
762 | pfd = walk->entries; | |
763 | pfd_end = pfd + walk->len; | |
764 | for (; pfd != pfd_end; pfd++) { | |
765 | /* | |
766 | * Fish for events. If we found one, record it | |
767 | * and kill the poll_table, so we don't | |
768 | * needlessly register any other waiters after | |
769 | * this. They'll get immediately deregistered | |
770 | * when we break out and return. | |
771 | */ | |
772 | if (do_pollfd(pfd, pt)) { | |
773 | count++; | |
774 | pt = NULL; | |
775 | } | |
776 | } | |
1da177e4 | 777 | } |
4a4b69f7 VL |
778 | /* |
779 | * All waiters have already been registered, so don't provide | |
780 | * a poll_table to them on the next loop iteration. | |
781 | */ | |
1da177e4 | 782 | pt = NULL; |
9bf084f7 ON |
783 | if (!count) { |
784 | count = wait->error; | |
785 | if (signal_pending(current)) | |
786 | count = -EINTR; | |
787 | } | |
8ff3e8e8 | 788 | if (count || timed_out) |
1da177e4 | 789 | break; |
9f72949f | 790 | |
8ff3e8e8 AV |
791 | /* |
792 | * If this is the first loop and we have a timeout | |
793 | * given, then we convert to ktime_t and set the to | |
794 | * pointer to the expiry value. | |
795 | */ | |
796 | if (end_time && !to) { | |
797 | expire = timespec_to_ktime(*end_time); | |
798 | to = &expire; | |
9f72949f DW |
799 | } |
800 | ||
5f820f64 | 801 | if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) |
8ff3e8e8 | 802 | timed_out = 1; |
1da177e4 | 803 | } |
1da177e4 LT |
804 | return count; |
805 | } | |
806 | ||
70674f95 AK |
807 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
808 | sizeof(struct pollfd)) | |
809 | ||
8ff3e8e8 AV |
810 | int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
811 | struct timespec *end_time) | |
1da177e4 LT |
812 | { |
813 | struct poll_wqueues table; | |
252e5725 | 814 | int err = -EFAULT, fdcount, len, size; |
30c14e40 JS |
815 | /* Allocate small arguments on the stack to save memory and be |
816 | faster - use long to make sure the buffer is aligned properly | |
817 | on 64 bit archs to avoid unaligned access */ | |
818 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
819 | struct poll_list *const head = (struct poll_list *)stack_pps; |
820 | struct poll_list *walk = head; | |
821 | unsigned long todo = nfds; | |
1da177e4 | 822 | |
4e6fd33b | 823 | if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur) |
1da177e4 LT |
824 | return -EINVAL; |
825 | ||
252e5725 ON |
826 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
827 | for (;;) { | |
828 | walk->next = NULL; | |
829 | walk->len = len; | |
830 | if (!len) | |
831 | break; | |
1da177e4 | 832 | |
252e5725 ON |
833 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
834 | sizeof(struct pollfd) * walk->len)) | |
835 | goto out_fds; | |
836 | ||
837 | todo -= walk->len; | |
838 | if (!todo) | |
839 | break; | |
1da177e4 | 840 | |
252e5725 ON |
841 | len = min(todo, POLLFD_PER_PAGE); |
842 | size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; | |
843 | walk = walk->next = kmalloc(size, GFP_KERNEL); | |
844 | if (!walk) { | |
845 | err = -ENOMEM; | |
1da177e4 LT |
846 | goto out_fds; |
847 | } | |
1da177e4 | 848 | } |
9f72949f | 849 | |
252e5725 | 850 | poll_initwait(&table); |
8ff3e8e8 | 851 | fdcount = do_poll(nfds, head, &table, end_time); |
252e5725 | 852 | poll_freewait(&table); |
1da177e4 | 853 | |
252e5725 | 854 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
855 | struct pollfd *fds = walk->entries; |
856 | int j; | |
857 | ||
252e5725 ON |
858 | for (j = 0; j < walk->len; j++, ufds++) |
859 | if (__put_user(fds[j].revents, &ufds->revents)) | |
1da177e4 | 860 | goto out_fds; |
1da177e4 | 861 | } |
252e5725 | 862 | |
1da177e4 | 863 | err = fdcount; |
1da177e4 | 864 | out_fds: |
252e5725 ON |
865 | walk = head->next; |
866 | while (walk) { | |
867 | struct poll_list *pos = walk; | |
868 | walk = walk->next; | |
869 | kfree(pos); | |
1da177e4 | 870 | } |
252e5725 | 871 | |
1da177e4 LT |
872 | return err; |
873 | } | |
9f72949f | 874 | |
3075d9da CW |
875 | static long do_restart_poll(struct restart_block *restart_block) |
876 | { | |
8ff3e8e8 AV |
877 | struct pollfd __user *ufds = restart_block->poll.ufds; |
878 | int nfds = restart_block->poll.nfds; | |
879 | struct timespec *to = NULL, end_time; | |
3075d9da CW |
880 | int ret; |
881 | ||
8ff3e8e8 AV |
882 | if (restart_block->poll.has_timeout) { |
883 | end_time.tv_sec = restart_block->poll.tv_sec; | |
884 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
885 | to = &end_time; | |
886 | } | |
887 | ||
888 | ret = do_sys_poll(ufds, nfds, to); | |
889 | ||
3075d9da CW |
890 | if (ret == -EINTR) { |
891 | restart_block->fn = do_restart_poll; | |
3075d9da CW |
892 | ret = -ERESTART_RESTARTBLOCK; |
893 | } | |
894 | return ret; | |
895 | } | |
896 | ||
5a8a82b1 HC |
897 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
898 | long, timeout_msecs) | |
9f72949f | 899 | { |
8ff3e8e8 | 900 | struct timespec end_time, *to = NULL; |
3075d9da | 901 | int ret; |
9f72949f | 902 | |
8ff3e8e8 AV |
903 | if (timeout_msecs >= 0) { |
904 | to = &end_time; | |
905 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
906 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
907 | } |
908 | ||
8ff3e8e8 AV |
909 | ret = do_sys_poll(ufds, nfds, to); |
910 | ||
3075d9da CW |
911 | if (ret == -EINTR) { |
912 | struct restart_block *restart_block; | |
8ff3e8e8 | 913 | |
3075d9da CW |
914 | restart_block = ¤t_thread_info()->restart_block; |
915 | restart_block->fn = do_restart_poll; | |
8ff3e8e8 AV |
916 | restart_block->poll.ufds = ufds; |
917 | restart_block->poll.nfds = nfds; | |
918 | ||
919 | if (timeout_msecs >= 0) { | |
920 | restart_block->poll.tv_sec = end_time.tv_sec; | |
921 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
922 | restart_block->poll.has_timeout = 1; | |
923 | } else | |
924 | restart_block->poll.has_timeout = 0; | |
925 | ||
3075d9da CW |
926 | ret = -ERESTART_RESTARTBLOCK; |
927 | } | |
928 | return ret; | |
9f72949f DW |
929 | } |
930 | ||
f3de272b | 931 | #ifdef HAVE_SET_RESTORE_SIGMASK |
d4e82042 HC |
932 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
933 | struct timespec __user *, tsp, const sigset_t __user *, sigmask, | |
934 | size_t, sigsetsize) | |
9f72949f DW |
935 | { |
936 | sigset_t ksigmask, sigsaved; | |
8ff3e8e8 | 937 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
938 | int ret; |
939 | ||
940 | if (tsp) { | |
941 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
942 | return -EFAULT; | |
943 | ||
8ff3e8e8 AV |
944 | to = &end_time; |
945 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
946 | return -EINVAL; | |
9f72949f DW |
947 | } |
948 | ||
949 | if (sigmask) { | |
950 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
951 | if (sigsetsize != sizeof(sigset_t)) | |
952 | return -EINVAL; | |
953 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
954 | return -EFAULT; | |
955 | ||
956 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
957 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
958 | } | |
959 | ||
8ff3e8e8 | 960 | ret = do_sys_poll(ufds, nfds, to); |
9f72949f DW |
961 | |
962 | /* We can restart this syscall, usually */ | |
963 | if (ret == -EINTR) { | |
964 | /* | |
965 | * Don't restore the signal mask yet. Let do_signal() deliver | |
966 | * the signal on the way back to userspace, before the signal | |
967 | * mask is restored. | |
968 | */ | |
969 | if (sigmask) { | |
970 | memcpy(¤t->saved_sigmask, &sigsaved, | |
971 | sizeof(sigsaved)); | |
4e4c22c7 | 972 | set_restore_sigmask(); |
9f72949f DW |
973 | } |
974 | ret = -ERESTARTNOHAND; | |
975 | } else if (sigmask) | |
976 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
977 | ||
8ff3e8e8 | 978 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
979 | |
980 | return ret; | |
981 | } | |
f3de272b | 982 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |