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> |
1da177e4 LT |
30 | |
31 | #include <asm/uaccess.h> | |
32 | ||
90d6e24a AV |
33 | |
34 | /* | |
35 | * Estimate expected accuracy in ns from a timeval. | |
36 | * | |
37 | * After quite a bit of churning around, we've settled on | |
38 | * a simple thing of taking 0.1% of the timeout as the | |
39 | * slack, with a cap of 100 msec. | |
40 | * "nice" tasks get a 0.5% slack instead. | |
41 | * | |
42 | * Consider this comment an open invitation to come up with even | |
43 | * better solutions.. | |
44 | */ | |
45 | ||
5ae87e79 GK |
46 | #define MAX_SLACK (100 * NSEC_PER_MSEC) |
47 | ||
96d2ab48 | 48 | static long __estimate_accuracy(struct timespec *tv) |
90d6e24a | 49 | { |
96d2ab48 | 50 | long slack; |
90d6e24a AV |
51 | int divfactor = 1000; |
52 | ||
5ae87e79 GK |
53 | if (tv->tv_sec < 0) |
54 | return 0; | |
55 | ||
4ce105d3 | 56 | if (task_nice(current) > 0) |
90d6e24a AV |
57 | divfactor = divfactor / 5; |
58 | ||
5ae87e79 GK |
59 | if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) |
60 | return MAX_SLACK; | |
61 | ||
90d6e24a AV |
62 | slack = tv->tv_nsec / divfactor; |
63 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); | |
64 | ||
5ae87e79 GK |
65 | if (slack > MAX_SLACK) |
66 | return MAX_SLACK; | |
96d2ab48 | 67 | |
90d6e24a AV |
68 | return slack; |
69 | } | |
70 | ||
95aac7b1 | 71 | long select_estimate_accuracy(struct timespec *tv) |
90d6e24a AV |
72 | { |
73 | unsigned long ret; | |
74 | struct timespec now; | |
75 | ||
76 | /* | |
77 | * Realtime tasks get a slack of 0 for obvious reasons. | |
78 | */ | |
79 | ||
4ce105d3 | 80 | if (rt_task(current)) |
90d6e24a AV |
81 | return 0; |
82 | ||
83 | ktime_get_ts(&now); | |
84 | now = timespec_sub(*tv, now); | |
85 | ret = __estimate_accuracy(&now); | |
86 | if (ret < current->timer_slack_ns) | |
87 | return current->timer_slack_ns; | |
88 | return ret; | |
89 | } | |
90 | ||
91 | ||
92 | ||
1da177e4 LT |
93 | struct poll_table_page { |
94 | struct poll_table_page * next; | |
95 | struct poll_table_entry * entry; | |
96 | struct poll_table_entry entries[0]; | |
97 | }; | |
98 | ||
99 | #define POLL_TABLE_FULL(table) \ | |
100 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
101 | ||
102 | /* | |
103 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
104 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
105 | * follow, but it should be free of race-conditions, and it's practical. If you | |
106 | * understand what I'm doing here, then you understand how the linux | |
107 | * sleep/wakeup mechanism works. | |
108 | * | |
109 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
110 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
111 | * as all select/poll functions have to call it to add an entry to the | |
112 | * poll table. | |
113 | */ | |
75c96f85 AB |
114 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
115 | poll_table *p); | |
1da177e4 LT |
116 | |
117 | void poll_initwait(struct poll_wqueues *pwq) | |
118 | { | |
119 | init_poll_funcptr(&pwq->pt, __pollwait); | |
5f820f64 | 120 | pwq->polling_task = current; |
b2add73d | 121 | pwq->triggered = 0; |
1da177e4 LT |
122 | pwq->error = 0; |
123 | pwq->table = NULL; | |
70674f95 | 124 | pwq->inline_index = 0; |
1da177e4 | 125 | } |
1da177e4 LT |
126 | EXPORT_SYMBOL(poll_initwait); |
127 | ||
70674f95 AK |
128 | static void free_poll_entry(struct poll_table_entry *entry) |
129 | { | |
ccf6780d | 130 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
131 | fput(entry->filp); |
132 | } | |
133 | ||
1da177e4 LT |
134 | void poll_freewait(struct poll_wqueues *pwq) |
135 | { | |
136 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
137 | int i; |
138 | for (i = 0; i < pwq->inline_index; i++) | |
139 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
140 | while (p) { |
141 | struct poll_table_entry * entry; | |
142 | struct poll_table_page *old; | |
143 | ||
144 | entry = p->entry; | |
145 | do { | |
146 | entry--; | |
70674f95 | 147 | free_poll_entry(entry); |
1da177e4 LT |
148 | } while (entry > p->entries); |
149 | old = p; | |
150 | p = p->next; | |
151 | free_page((unsigned long) old); | |
152 | } | |
153 | } | |
1da177e4 LT |
154 | EXPORT_SYMBOL(poll_freewait); |
155 | ||
5f820f64 | 156 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
1da177e4 | 157 | { |
1da177e4 LT |
158 | struct poll_table_page *table = p->table; |
159 | ||
70674f95 AK |
160 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
161 | return p->inline_entries + p->inline_index++; | |
162 | ||
1da177e4 LT |
163 | if (!table || POLL_TABLE_FULL(table)) { |
164 | struct poll_table_page *new_table; | |
165 | ||
166 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
167 | if (!new_table) { | |
168 | p->error = -ENOMEM; | |
70674f95 | 169 | return NULL; |
1da177e4 LT |
170 | } |
171 | new_table->entry = new_table->entries; | |
172 | new_table->next = table; | |
173 | p->table = new_table; | |
174 | table = new_table; | |
175 | } | |
176 | ||
70674f95 AK |
177 | return table->entry++; |
178 | } | |
179 | ||
4938d7e0 | 180 | static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) |
5f820f64 TH |
181 | { |
182 | struct poll_wqueues *pwq = wait->private; | |
183 | DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); | |
184 | ||
185 | /* | |
186 | * Although this function is called under waitqueue lock, LOCK | |
187 | * doesn't imply write barrier and the users expect write | |
188 | * barrier semantics on wakeup functions. The following | |
189 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
190 | * and is paired with set_mb() in poll_schedule_timeout. | |
191 | */ | |
192 | smp_wmb(); | |
193 | pwq->triggered = 1; | |
194 | ||
195 | /* | |
196 | * Perform the default wake up operation using a dummy | |
197 | * waitqueue. | |
198 | * | |
199 | * TODO: This is hacky but there currently is no interface to | |
200 | * pass in @sync. @sync is scheduled to be removed and once | |
201 | * that happens, wake_up_process() can be used directly. | |
202 | */ | |
203 | return default_wake_function(&dummy_wait, mode, sync, key); | |
204 | } | |
205 | ||
4938d7e0 ED |
206 | static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) |
207 | { | |
208 | struct poll_table_entry *entry; | |
209 | ||
210 | entry = container_of(wait, struct poll_table_entry, wait); | |
211 | if (key && !((unsigned long)key & entry->key)) | |
212 | return 0; | |
213 | return __pollwake(wait, mode, sync, key); | |
214 | } | |
215 | ||
70674f95 AK |
216 | /* Add a new entry */ |
217 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
218 | poll_table *p) | |
219 | { | |
5f820f64 TH |
220 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
221 | struct poll_table_entry *entry = poll_get_entry(pwq); | |
70674f95 AK |
222 | if (!entry) |
223 | return; | |
cb0942b8 | 224 | entry->filp = get_file(filp); |
70674f95 | 225 | entry->wait_address = wait_address; |
626cf236 | 226 | entry->key = p->_key; |
5f820f64 TH |
227 | init_waitqueue_func_entry(&entry->wait, pollwake); |
228 | entry->wait.private = pwq; | |
ccf6780d | 229 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
230 | } |
231 | ||
5f820f64 TH |
232 | int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
233 | ktime_t *expires, unsigned long slack) | |
234 | { | |
235 | int rc = -EINTR; | |
236 | ||
237 | set_current_state(state); | |
238 | if (!pwq->triggered) | |
239 | rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); | |
240 | __set_current_state(TASK_RUNNING); | |
241 | ||
242 | /* | |
243 | * Prepare for the next iteration. | |
244 | * | |
245 | * The following set_mb() serves two purposes. First, it's | |
246 | * the counterpart rmb of the wmb in pollwake() such that data | |
247 | * written before wake up is always visible after wake up. | |
248 | * Second, the full barrier guarantees that triggered clearing | |
249 | * doesn't pass event check of the next iteration. Note that | |
250 | * this problem doesn't exist for the first iteration as | |
251 | * add_wait_queue() has full barrier semantics. | |
252 | */ | |
253 | set_mb(pwq->triggered, 0); | |
254 | ||
255 | return rc; | |
256 | } | |
257 | EXPORT_SYMBOL(poll_schedule_timeout); | |
258 | ||
b773ad40 TG |
259 | /** |
260 | * poll_select_set_timeout - helper function to setup the timeout value | |
261 | * @to: pointer to timespec variable for the final timeout | |
262 | * @sec: seconds (from user space) | |
263 | * @nsec: nanoseconds (from user space) | |
264 | * | |
265 | * Note, we do not use a timespec for the user space value here, That | |
266 | * way we can use the function for timeval and compat interfaces as well. | |
267 | * | |
268 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
269 | */ | |
270 | int poll_select_set_timeout(struct timespec *to, long sec, long nsec) | |
271 | { | |
272 | struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec}; | |
273 | ||
274 | if (!timespec_valid(&ts)) | |
275 | return -EINVAL; | |
276 | ||
277 | /* Optimize for the zero timeout value here */ | |
278 | if (!sec && !nsec) { | |
279 | to->tv_sec = to->tv_nsec = 0; | |
280 | } else { | |
281 | ktime_get_ts(to); | |
282 | *to = timespec_add_safe(*to, ts); | |
283 | } | |
284 | return 0; | |
285 | } | |
286 | ||
287 | static int poll_select_copy_remaining(struct timespec *end_time, void __user *p, | |
288 | int timeval, int ret) | |
289 | { | |
290 | struct timespec rts; | |
291 | struct timeval rtv; | |
292 | ||
293 | if (!p) | |
294 | return ret; | |
295 | ||
296 | if (current->personality & STICKY_TIMEOUTS) | |
297 | goto sticky; | |
298 | ||
299 | /* No update for zero timeout */ | |
300 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
301 | return ret; | |
302 | ||
303 | ktime_get_ts(&rts); | |
304 | rts = timespec_sub(*end_time, rts); | |
305 | if (rts.tv_sec < 0) | |
306 | rts.tv_sec = rts.tv_nsec = 0; | |
307 | ||
308 | if (timeval) { | |
65329bf4 VK |
309 | if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) |
310 | memset(&rtv, 0, sizeof(rtv)); | |
b773ad40 TG |
311 | rtv.tv_sec = rts.tv_sec; |
312 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
313 | ||
314 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
315 | return ret; | |
316 | ||
317 | } else if (!copy_to_user(p, &rts, sizeof(rts))) | |
318 | return ret; | |
319 | ||
320 | /* | |
321 | * If an application puts its timeval in read-only memory, we | |
322 | * don't want the Linux-specific update to the timeval to | |
323 | * cause a fault after the select has completed | |
324 | * successfully. However, because we're not updating the | |
325 | * timeval, we can't restart the system call. | |
326 | */ | |
327 | ||
328 | sticky: | |
329 | if (ret == -ERESTARTNOHAND) | |
330 | ret = -EINTR; | |
331 | return ret; | |
332 | } | |
333 | ||
1da177e4 LT |
334 | #define FDS_IN(fds, n) (fds->in + n) |
335 | #define FDS_OUT(fds, n) (fds->out + n) | |
336 | #define FDS_EX(fds, n) (fds->ex + n) | |
337 | ||
338 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
339 | ||
340 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
341 | { | |
342 | unsigned long *open_fds; | |
343 | unsigned long set; | |
344 | int max; | |
badf1662 | 345 | struct fdtable *fdt; |
1da177e4 LT |
346 | |
347 | /* handle last in-complete long-word first */ | |
8ded2bbc JB |
348 | set = ~(~0UL << (n & (BITS_PER_LONG-1))); |
349 | n /= BITS_PER_LONG; | |
badf1662 | 350 | fdt = files_fdtable(current->files); |
1fd36adc | 351 | open_fds = fdt->open_fds + n; |
1da177e4 LT |
352 | max = 0; |
353 | if (set) { | |
354 | set &= BITS(fds, n); | |
355 | if (set) { | |
356 | if (!(set & ~*open_fds)) | |
357 | goto get_max; | |
358 | return -EBADF; | |
359 | } | |
360 | } | |
361 | while (n) { | |
362 | open_fds--; | |
363 | n--; | |
364 | set = BITS(fds, n); | |
365 | if (!set) | |
366 | continue; | |
367 | if (set & ~*open_fds) | |
368 | return -EBADF; | |
369 | if (max) | |
370 | continue; | |
371 | get_max: | |
372 | do { | |
373 | max++; | |
374 | set >>= 1; | |
375 | } while (set); | |
8ded2bbc | 376 | max += n * BITS_PER_LONG; |
1da177e4 LT |
377 | } |
378 | ||
379 | return max; | |
380 | } | |
381 | ||
1da177e4 LT |
382 | #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) |
383 | #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) | |
384 | #define POLLEX_SET (POLLPRI) | |
385 | ||
4938d7e0 ED |
386 | static inline void wait_key_set(poll_table *wait, unsigned long in, |
387 | unsigned long out, unsigned long bit) | |
388 | { | |
626cf236 HV |
389 | wait->_key = POLLEX_SET; |
390 | if (in & bit) | |
391 | wait->_key |= POLLIN_SET; | |
392 | if (out & bit) | |
393 | wait->_key |= POLLOUT_SET; | |
4938d7e0 ED |
394 | } |
395 | ||
8ff3e8e8 | 396 | int do_select(int n, fd_set_bits *fds, struct timespec *end_time) |
1da177e4 | 397 | { |
8ff3e8e8 | 398 | ktime_t expire, *to = NULL; |
1da177e4 LT |
399 | struct poll_wqueues table; |
400 | poll_table *wait; | |
8ff3e8e8 | 401 | int retval, i, timed_out = 0; |
90d6e24a | 402 | unsigned long slack = 0; |
1da177e4 | 403 | |
b835996f | 404 | rcu_read_lock(); |
1da177e4 | 405 | retval = max_select_fd(n, fds); |
b835996f | 406 | rcu_read_unlock(); |
1da177e4 LT |
407 | |
408 | if (retval < 0) | |
409 | return retval; | |
410 | n = retval; | |
411 | ||
412 | poll_initwait(&table); | |
413 | wait = &table.pt; | |
8ff3e8e8 | 414 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 415 | wait->_qproc = NULL; |
8ff3e8e8 AV |
416 | timed_out = 1; |
417 | } | |
418 | ||
96d2ab48 | 419 | if (end_time && !timed_out) |
231f3d39 | 420 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 421 | |
1da177e4 LT |
422 | retval = 0; |
423 | for (;;) { | |
424 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
425 | ||
1da177e4 LT |
426 | inp = fds->in; outp = fds->out; exp = fds->ex; |
427 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
428 | ||
429 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
430 | unsigned long in, out, ex, all_bits, bit = 1, mask, j; | |
431 | unsigned long res_in = 0, res_out = 0, res_ex = 0; | |
1da177e4 LT |
432 | |
433 | in = *inp++; out = *outp++; ex = *exp++; | |
434 | all_bits = in | out | ex; | |
435 | if (all_bits == 0) { | |
8ded2bbc | 436 | i += BITS_PER_LONG; |
1da177e4 LT |
437 | continue; |
438 | } | |
439 | ||
8ded2bbc | 440 | for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { |
2903ff01 | 441 | struct fd f; |
1da177e4 LT |
442 | if (i >= n) |
443 | break; | |
444 | if (!(bit & all_bits)) | |
445 | continue; | |
2903ff01 AV |
446 | f = fdget(i); |
447 | if (f.file) { | |
448 | const struct file_operations *f_op; | |
449 | f_op = f.file->f_op; | |
1da177e4 | 450 | mask = DEFAULT_POLLMASK; |
4938d7e0 ED |
451 | if (f_op && f_op->poll) { |
452 | wait_key_set(wait, in, out, bit); | |
2903ff01 | 453 | mask = (*f_op->poll)(f.file, wait); |
4938d7e0 | 454 | } |
2903ff01 | 455 | fdput(f); |
1da177e4 LT |
456 | if ((mask & POLLIN_SET) && (in & bit)) { |
457 | res_in |= bit; | |
458 | retval++; | |
626cf236 | 459 | wait->_qproc = NULL; |
1da177e4 LT |
460 | } |
461 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
462 | res_out |= bit; | |
463 | retval++; | |
626cf236 | 464 | wait->_qproc = NULL; |
1da177e4 LT |
465 | } |
466 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
467 | res_ex |= bit; | |
468 | retval++; | |
626cf236 | 469 | wait->_qproc = 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 | 480 | } |
626cf236 | 481 | wait->_qproc = 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 | */ | |
a2dcb44c | 517 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
8ff3e8e8 | 518 | fd_set __user *exp, struct timespec *end_time) |
1da177e4 LT |
519 | { |
520 | fd_set_bits fds; | |
29ff2db5 | 521 | void *bits; |
bbea9f69 | 522 | int ret, max_fds; |
b04eb6aa | 523 | unsigned int size; |
badf1662 | 524 | struct fdtable *fdt; |
70674f95 | 525 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 526 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 527 | |
1da177e4 LT |
528 | ret = -EINVAL; |
529 | if (n < 0) | |
530 | goto out_nofds; | |
531 | ||
bbea9f69 | 532 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 533 | rcu_read_lock(); |
badf1662 | 534 | fdt = files_fdtable(current->files); |
bbea9f69 | 535 | max_fds = fdt->max_fds; |
b835996f | 536 | rcu_read_unlock(); |
bbea9f69 VL |
537 | if (n > max_fds) |
538 | n = max_fds; | |
1da177e4 LT |
539 | |
540 | /* | |
541 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
542 | * since we used fdset we need to allocate memory in units of | |
543 | * long-words. | |
544 | */ | |
1da177e4 | 545 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
546 | bits = stack_fds; |
547 | if (size > sizeof(stack_fds) / 6) { | |
548 | /* Not enough space in on-stack array; must use kmalloc */ | |
549 | ret = -ENOMEM; | |
70674f95 | 550 | bits = kmalloc(6 * size, GFP_KERNEL); |
b04eb6aa MBJ |
551 | if (!bits) |
552 | goto out_nofds; | |
553 | } | |
29ff2db5 AM |
554 | fds.in = bits; |
555 | fds.out = bits + size; | |
556 | fds.ex = bits + 2*size; | |
557 | fds.res_in = bits + 3*size; | |
558 | fds.res_out = bits + 4*size; | |
559 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
560 | |
561 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
562 | (ret = get_fd_set(n, outp, fds.out)) || | |
563 | (ret = get_fd_set(n, exp, fds.ex))) | |
564 | goto out; | |
565 | zero_fd_set(n, fds.res_in); | |
566 | zero_fd_set(n, fds.res_out); | |
567 | zero_fd_set(n, fds.res_ex); | |
568 | ||
8ff3e8e8 | 569 | ret = do_select(n, &fds, end_time); |
1da177e4 LT |
570 | |
571 | if (ret < 0) | |
572 | goto out; | |
573 | if (!ret) { | |
574 | ret = -ERESTARTNOHAND; | |
575 | if (signal_pending(current)) | |
576 | goto out; | |
577 | ret = 0; | |
578 | } | |
579 | ||
580 | if (set_fd_set(n, inp, fds.res_in) || | |
581 | set_fd_set(n, outp, fds.res_out) || | |
582 | set_fd_set(n, exp, fds.res_ex)) | |
583 | ret = -EFAULT; | |
584 | ||
585 | out: | |
70674f95 AK |
586 | if (bits != stack_fds) |
587 | kfree(bits); | |
1da177e4 LT |
588 | out_nofds: |
589 | return ret; | |
590 | } | |
591 | ||
5a8a82b1 HC |
592 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
593 | fd_set __user *, exp, struct timeval __user *, tvp) | |
9f72949f | 594 | { |
8ff3e8e8 | 595 | struct timespec end_time, *to = NULL; |
9f72949f DW |
596 | struct timeval tv; |
597 | int ret; | |
598 | ||
599 | if (tvp) { | |
600 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
601 | return -EFAULT; | |
602 | ||
8ff3e8e8 | 603 | to = &end_time; |
4d36a9e6 AV |
604 | if (poll_select_set_timeout(to, |
605 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
606 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
9f72949f | 607 | return -EINVAL; |
9f72949f DW |
608 | } |
609 | ||
8ff3e8e8 AV |
610 | ret = core_sys_select(n, inp, outp, exp, to); |
611 | ret = poll_select_copy_remaining(&end_time, tvp, 1, ret); | |
9f72949f DW |
612 | |
613 | return ret; | |
614 | } | |
615 | ||
c9da9f21 HC |
616 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
617 | fd_set __user *exp, struct timespec __user *tsp, | |
618 | const sigset_t __user *sigmask, size_t sigsetsize) | |
9f72949f | 619 | { |
9f72949f | 620 | sigset_t ksigmask, sigsaved; |
8ff3e8e8 | 621 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
622 | int ret; |
623 | ||
624 | if (tsp) { | |
625 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
626 | return -EFAULT; | |
627 | ||
8ff3e8e8 AV |
628 | to = &end_time; |
629 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
9f72949f | 630 | return -EINVAL; |
9f72949f DW |
631 | } |
632 | ||
633 | if (sigmask) { | |
634 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
635 | if (sigsetsize != sizeof(sigset_t)) | |
636 | return -EINVAL; | |
637 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
638 | return -EFAULT; | |
639 | ||
640 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
641 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
642 | } | |
643 | ||
62568510 | 644 | ret = core_sys_select(n, inp, outp, exp, to); |
8ff3e8e8 | 645 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
646 | |
647 | if (ret == -ERESTARTNOHAND) { | |
648 | /* | |
649 | * Don't restore the signal mask yet. Let do_signal() deliver | |
650 | * the signal on the way back to userspace, before the signal | |
651 | * mask is restored. | |
652 | */ | |
653 | if (sigmask) { | |
654 | memcpy(¤t->saved_sigmask, &sigsaved, | |
655 | sizeof(sigsaved)); | |
4e4c22c7 | 656 | set_restore_sigmask(); |
9f72949f DW |
657 | } |
658 | } else if (sigmask) | |
659 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
660 | ||
661 | return ret; | |
662 | } | |
663 | ||
664 | /* | |
665 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
666 | * 6-argument version where the sixth argument is a pointer to a structure | |
667 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
668 | * the sigset size. | |
669 | */ | |
d4e82042 HC |
670 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
671 | fd_set __user *, exp, struct timespec __user *, tsp, | |
672 | void __user *, sig) | |
9f72949f DW |
673 | { |
674 | size_t sigsetsize = 0; | |
675 | sigset_t __user *up = NULL; | |
676 | ||
677 | if (sig) { | |
678 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
e110ab94 | 679 | || __get_user(up, (sigset_t __user * __user *)sig) |
9f72949f | 680 | || __get_user(sigsetsize, |
e110ab94 | 681 | (size_t __user *)(sig+sizeof(void *)))) |
9f72949f DW |
682 | return -EFAULT; |
683 | } | |
684 | ||
c9da9f21 | 685 | return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize); |
9f72949f | 686 | } |
9f72949f | 687 | |
5d0e5283 CH |
688 | #ifdef __ARCH_WANT_SYS_OLD_SELECT |
689 | struct sel_arg_struct { | |
690 | unsigned long n; | |
691 | fd_set __user *inp, *outp, *exp; | |
692 | struct timeval __user *tvp; | |
693 | }; | |
694 | ||
695 | SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) | |
696 | { | |
697 | struct sel_arg_struct a; | |
698 | ||
699 | if (copy_from_user(&a, arg, sizeof(a))) | |
700 | return -EFAULT; | |
701 | return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp); | |
702 | } | |
703 | #endif | |
704 | ||
1da177e4 LT |
705 | struct poll_list { |
706 | struct poll_list *next; | |
707 | int len; | |
708 | struct pollfd entries[0]; | |
709 | }; | |
710 | ||
711 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
712 | ||
4a4b69f7 VL |
713 | /* |
714 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
715 | * interested in events matching the pollfd->events mask, and the result | |
716 | * matching that mask is both recorded in pollfd->revents and returned. The | |
717 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
626cf236 | 718 | * if pwait->_qproc is non-NULL. |
4a4b69f7 VL |
719 | */ |
720 | static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait) | |
1da177e4 | 721 | { |
4a4b69f7 VL |
722 | unsigned int mask; |
723 | int fd; | |
724 | ||
725 | mask = 0; | |
726 | fd = pollfd->fd; | |
727 | if (fd >= 0) { | |
2903ff01 | 728 | struct fd f = fdget(fd); |
4a4b69f7 | 729 | mask = POLLNVAL; |
2903ff01 | 730 | if (f.file) { |
4a4b69f7 | 731 | mask = DEFAULT_POLLMASK; |
2903ff01 | 732 | if (f.file->f_op && f.file->f_op->poll) { |
626cf236 | 733 | pwait->_key = pollfd->events|POLLERR|POLLHUP; |
2903ff01 | 734 | mask = f.file->f_op->poll(f.file, pwait); |
4938d7e0 | 735 | } |
4a4b69f7 VL |
736 | /* Mask out unneeded events. */ |
737 | mask &= pollfd->events | POLLERR | POLLHUP; | |
2903ff01 | 738 | fdput(f); |
1da177e4 | 739 | } |
1da177e4 | 740 | } |
4a4b69f7 VL |
741 | pollfd->revents = mask; |
742 | ||
743 | return mask; | |
1da177e4 LT |
744 | } |
745 | ||
746 | static int do_poll(unsigned int nfds, struct poll_list *list, | |
8ff3e8e8 | 747 | struct poll_wqueues *wait, struct timespec *end_time) |
1da177e4 | 748 | { |
1da177e4 | 749 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
750 | ktime_t expire, *to = NULL; |
751 | int timed_out = 0, count = 0; | |
90d6e24a | 752 | unsigned long slack = 0; |
1da177e4 | 753 | |
9f72949f | 754 | /* Optimise the no-wait case */ |
8ff3e8e8 | 755 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 756 | pt->_qproc = NULL; |
8ff3e8e8 AV |
757 | timed_out = 1; |
758 | } | |
9bf084f7 | 759 | |
96d2ab48 | 760 | if (end_time && !timed_out) |
231f3d39 | 761 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 762 | |
1da177e4 LT |
763 | for (;;) { |
764 | struct poll_list *walk; | |
9f72949f | 765 | |
4a4b69f7 VL |
766 | for (walk = list; walk != NULL; walk = walk->next) { |
767 | struct pollfd * pfd, * pfd_end; | |
768 | ||
769 | pfd = walk->entries; | |
770 | pfd_end = pfd + walk->len; | |
771 | for (; pfd != pfd_end; pfd++) { | |
772 | /* | |
773 | * Fish for events. If we found one, record it | |
626cf236 | 774 | * and kill poll_table->_qproc, so we don't |
4a4b69f7 VL |
775 | * needlessly register any other waiters after |
776 | * this. They'll get immediately deregistered | |
777 | * when we break out and return. | |
778 | */ | |
779 | if (do_pollfd(pfd, pt)) { | |
780 | count++; | |
626cf236 | 781 | pt->_qproc = NULL; |
4a4b69f7 VL |
782 | } |
783 | } | |
1da177e4 | 784 | } |
4a4b69f7 VL |
785 | /* |
786 | * All waiters have already been registered, so don't provide | |
626cf236 | 787 | * a poll_table->_qproc to them on the next loop iteration. |
4a4b69f7 | 788 | */ |
626cf236 | 789 | pt->_qproc = NULL; |
9bf084f7 ON |
790 | if (!count) { |
791 | count = wait->error; | |
792 | if (signal_pending(current)) | |
793 | count = -EINTR; | |
794 | } | |
8ff3e8e8 | 795 | if (count || timed_out) |
1da177e4 | 796 | break; |
9f72949f | 797 | |
8ff3e8e8 AV |
798 | /* |
799 | * If this is the first loop and we have a timeout | |
800 | * given, then we convert to ktime_t and set the to | |
801 | * pointer to the expiry value. | |
802 | */ | |
803 | if (end_time && !to) { | |
804 | expire = timespec_to_ktime(*end_time); | |
805 | to = &expire; | |
9f72949f DW |
806 | } |
807 | ||
5f820f64 | 808 | if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) |
8ff3e8e8 | 809 | timed_out = 1; |
1da177e4 | 810 | } |
1da177e4 LT |
811 | return count; |
812 | } | |
813 | ||
70674f95 AK |
814 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
815 | sizeof(struct pollfd)) | |
816 | ||
8ff3e8e8 AV |
817 | int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
818 | struct timespec *end_time) | |
1da177e4 LT |
819 | { |
820 | struct poll_wqueues table; | |
252e5725 | 821 | int err = -EFAULT, fdcount, len, size; |
30c14e40 JS |
822 | /* Allocate small arguments on the stack to save memory and be |
823 | faster - use long to make sure the buffer is aligned properly | |
824 | on 64 bit archs to avoid unaligned access */ | |
825 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
826 | struct poll_list *const head = (struct poll_list *)stack_pps; |
827 | struct poll_list *walk = head; | |
828 | unsigned long todo = nfds; | |
1da177e4 | 829 | |
d554ed89 | 830 | if (nfds > rlimit(RLIMIT_NOFILE)) |
1da177e4 LT |
831 | return -EINVAL; |
832 | ||
252e5725 ON |
833 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
834 | for (;;) { | |
835 | walk->next = NULL; | |
836 | walk->len = len; | |
837 | if (!len) | |
838 | break; | |
1da177e4 | 839 | |
252e5725 ON |
840 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
841 | sizeof(struct pollfd) * walk->len)) | |
842 | goto out_fds; | |
843 | ||
844 | todo -= walk->len; | |
845 | if (!todo) | |
846 | break; | |
1da177e4 | 847 | |
252e5725 ON |
848 | len = min(todo, POLLFD_PER_PAGE); |
849 | size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; | |
850 | walk = walk->next = kmalloc(size, GFP_KERNEL); | |
851 | if (!walk) { | |
852 | err = -ENOMEM; | |
1da177e4 LT |
853 | goto out_fds; |
854 | } | |
1da177e4 | 855 | } |
9f72949f | 856 | |
252e5725 | 857 | poll_initwait(&table); |
8ff3e8e8 | 858 | fdcount = do_poll(nfds, head, &table, end_time); |
252e5725 | 859 | poll_freewait(&table); |
1da177e4 | 860 | |
252e5725 | 861 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
862 | struct pollfd *fds = walk->entries; |
863 | int j; | |
864 | ||
252e5725 ON |
865 | for (j = 0; j < walk->len; j++, ufds++) |
866 | if (__put_user(fds[j].revents, &ufds->revents)) | |
1da177e4 | 867 | goto out_fds; |
1da177e4 | 868 | } |
252e5725 | 869 | |
1da177e4 | 870 | err = fdcount; |
1da177e4 | 871 | out_fds: |
252e5725 ON |
872 | walk = head->next; |
873 | while (walk) { | |
874 | struct poll_list *pos = walk; | |
875 | walk = walk->next; | |
876 | kfree(pos); | |
1da177e4 | 877 | } |
252e5725 | 878 | |
1da177e4 LT |
879 | return err; |
880 | } | |
9f72949f | 881 | |
3075d9da CW |
882 | static long do_restart_poll(struct restart_block *restart_block) |
883 | { | |
8ff3e8e8 AV |
884 | struct pollfd __user *ufds = restart_block->poll.ufds; |
885 | int nfds = restart_block->poll.nfds; | |
886 | struct timespec *to = NULL, end_time; | |
3075d9da CW |
887 | int ret; |
888 | ||
8ff3e8e8 AV |
889 | if (restart_block->poll.has_timeout) { |
890 | end_time.tv_sec = restart_block->poll.tv_sec; | |
891 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
892 | to = &end_time; | |
893 | } | |
894 | ||
895 | ret = do_sys_poll(ufds, nfds, to); | |
896 | ||
3075d9da CW |
897 | if (ret == -EINTR) { |
898 | restart_block->fn = do_restart_poll; | |
3075d9da CW |
899 | ret = -ERESTART_RESTARTBLOCK; |
900 | } | |
901 | return ret; | |
902 | } | |
903 | ||
5a8a82b1 | 904 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
faf30900 | 905 | int, timeout_msecs) |
9f72949f | 906 | { |
8ff3e8e8 | 907 | struct timespec end_time, *to = NULL; |
3075d9da | 908 | int ret; |
9f72949f | 909 | |
8ff3e8e8 AV |
910 | if (timeout_msecs >= 0) { |
911 | to = &end_time; | |
912 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
913 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
914 | } |
915 | ||
8ff3e8e8 AV |
916 | ret = do_sys_poll(ufds, nfds, to); |
917 | ||
3075d9da CW |
918 | if (ret == -EINTR) { |
919 | struct restart_block *restart_block; | |
8ff3e8e8 | 920 | |
3075d9da CW |
921 | restart_block = ¤t_thread_info()->restart_block; |
922 | restart_block->fn = do_restart_poll; | |
8ff3e8e8 AV |
923 | restart_block->poll.ufds = ufds; |
924 | restart_block->poll.nfds = nfds; | |
925 | ||
926 | if (timeout_msecs >= 0) { | |
927 | restart_block->poll.tv_sec = end_time.tv_sec; | |
928 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
929 | restart_block->poll.has_timeout = 1; | |
930 | } else | |
931 | restart_block->poll.has_timeout = 0; | |
932 | ||
3075d9da CW |
933 | ret = -ERESTART_RESTARTBLOCK; |
934 | } | |
935 | return ret; | |
9f72949f DW |
936 | } |
937 | ||
d4e82042 HC |
938 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
939 | struct timespec __user *, tsp, const sigset_t __user *, sigmask, | |
940 | size_t, sigsetsize) | |
9f72949f DW |
941 | { |
942 | sigset_t ksigmask, sigsaved; | |
8ff3e8e8 | 943 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
944 | int ret; |
945 | ||
946 | if (tsp) { | |
947 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
948 | return -EFAULT; | |
949 | ||
8ff3e8e8 AV |
950 | to = &end_time; |
951 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
952 | return -EINVAL; | |
9f72949f DW |
953 | } |
954 | ||
955 | if (sigmask) { | |
956 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
957 | if (sigsetsize != sizeof(sigset_t)) | |
958 | return -EINVAL; | |
959 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
960 | return -EFAULT; | |
961 | ||
962 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
963 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
964 | } | |
965 | ||
8ff3e8e8 | 966 | ret = do_sys_poll(ufds, nfds, to); |
9f72949f DW |
967 | |
968 | /* We can restart this syscall, usually */ | |
969 | if (ret == -EINTR) { | |
970 | /* | |
971 | * Don't restore the signal mask yet. Let do_signal() deliver | |
972 | * the signal on the way back to userspace, before the signal | |
973 | * mask is restored. | |
974 | */ | |
975 | if (sigmask) { | |
976 | memcpy(¤t->saved_sigmask, &sigsaved, | |
977 | sizeof(sigsaved)); | |
4e4c22c7 | 978 | set_restore_sigmask(); |
9f72949f DW |
979 | } |
980 | ret = -ERESTARTNOHAND; | |
981 | } else if (sigmask) | |
982 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
983 | ||
8ff3e8e8 | 984 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
985 | |
986 | return ret; | |
987 | } |