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 | ||
44 | static unsigned long __estimate_accuracy(struct timespec *tv) | |
45 | { | |
46 | unsigned long slack; | |
47 | int divfactor = 1000; | |
48 | ||
4ce105d3 | 49 | if (task_nice(current) > 0) |
90d6e24a AV |
50 | divfactor = divfactor / 5; |
51 | ||
52 | slack = tv->tv_nsec / divfactor; | |
53 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); | |
54 | ||
55 | if (slack > 100 * NSEC_PER_MSEC) | |
56 | slack = 100 * NSEC_PER_MSEC; | |
57 | return slack; | |
58 | } | |
59 | ||
60 | static unsigned long estimate_accuracy(struct timespec *tv) | |
61 | { | |
62 | unsigned long ret; | |
63 | struct timespec now; | |
64 | ||
65 | /* | |
66 | * Realtime tasks get a slack of 0 for obvious reasons. | |
67 | */ | |
68 | ||
4ce105d3 | 69 | if (rt_task(current)) |
90d6e24a AV |
70 | return 0; |
71 | ||
72 | ktime_get_ts(&now); | |
73 | now = timespec_sub(*tv, now); | |
74 | ret = __estimate_accuracy(&now); | |
75 | if (ret < current->timer_slack_ns) | |
76 | return current->timer_slack_ns; | |
77 | return ret; | |
78 | } | |
79 | ||
80 | ||
81 | ||
1da177e4 LT |
82 | struct poll_table_page { |
83 | struct poll_table_page * next; | |
84 | struct poll_table_entry * entry; | |
85 | struct poll_table_entry entries[0]; | |
86 | }; | |
87 | ||
88 | #define POLL_TABLE_FULL(table) \ | |
89 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
90 | ||
91 | /* | |
92 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
93 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
94 | * follow, but it should be free of race-conditions, and it's practical. If you | |
95 | * understand what I'm doing here, then you understand how the linux | |
96 | * sleep/wakeup mechanism works. | |
97 | * | |
98 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
99 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
100 | * as all select/poll functions have to call it to add an entry to the | |
101 | * poll table. | |
102 | */ | |
75c96f85 AB |
103 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
104 | poll_table *p); | |
1da177e4 LT |
105 | |
106 | void poll_initwait(struct poll_wqueues *pwq) | |
107 | { | |
108 | init_poll_funcptr(&pwq->pt, __pollwait); | |
109 | pwq->error = 0; | |
110 | pwq->table = NULL; | |
70674f95 | 111 | pwq->inline_index = 0; |
1da177e4 LT |
112 | } |
113 | ||
114 | EXPORT_SYMBOL(poll_initwait); | |
115 | ||
70674f95 AK |
116 | static void free_poll_entry(struct poll_table_entry *entry) |
117 | { | |
ccf6780d | 118 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
119 | fput(entry->filp); |
120 | } | |
121 | ||
1da177e4 LT |
122 | void poll_freewait(struct poll_wqueues *pwq) |
123 | { | |
124 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
125 | int i; |
126 | for (i = 0; i < pwq->inline_index; i++) | |
127 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
128 | while (p) { |
129 | struct poll_table_entry * entry; | |
130 | struct poll_table_page *old; | |
131 | ||
132 | entry = p->entry; | |
133 | do { | |
134 | entry--; | |
70674f95 | 135 | free_poll_entry(entry); |
1da177e4 LT |
136 | } while (entry > p->entries); |
137 | old = p; | |
138 | p = p->next; | |
139 | free_page((unsigned long) old); | |
140 | } | |
141 | } | |
142 | ||
143 | EXPORT_SYMBOL(poll_freewait); | |
144 | ||
70674f95 | 145 | static struct poll_table_entry *poll_get_entry(poll_table *_p) |
1da177e4 LT |
146 | { |
147 | struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt); | |
148 | struct poll_table_page *table = p->table; | |
149 | ||
70674f95 AK |
150 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
151 | return p->inline_entries + p->inline_index++; | |
152 | ||
1da177e4 LT |
153 | if (!table || POLL_TABLE_FULL(table)) { |
154 | struct poll_table_page *new_table; | |
155 | ||
156 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
157 | if (!new_table) { | |
158 | p->error = -ENOMEM; | |
159 | __set_current_state(TASK_RUNNING); | |
70674f95 | 160 | return NULL; |
1da177e4 LT |
161 | } |
162 | new_table->entry = new_table->entries; | |
163 | new_table->next = table; | |
164 | p->table = new_table; | |
165 | table = new_table; | |
166 | } | |
167 | ||
70674f95 AK |
168 | return table->entry++; |
169 | } | |
170 | ||
171 | /* Add a new entry */ | |
172 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
173 | poll_table *p) | |
174 | { | |
175 | struct poll_table_entry *entry = poll_get_entry(p); | |
176 | if (!entry) | |
177 | return; | |
178 | get_file(filp); | |
179 | entry->filp = filp; | |
180 | entry->wait_address = wait_address; | |
181 | init_waitqueue_entry(&entry->wait, current); | |
ccf6780d | 182 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
183 | } |
184 | ||
b773ad40 TG |
185 | /** |
186 | * poll_select_set_timeout - helper function to setup the timeout value | |
187 | * @to: pointer to timespec variable for the final timeout | |
188 | * @sec: seconds (from user space) | |
189 | * @nsec: nanoseconds (from user space) | |
190 | * | |
191 | * Note, we do not use a timespec for the user space value here, That | |
192 | * way we can use the function for timeval and compat interfaces as well. | |
193 | * | |
194 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
195 | */ | |
196 | int poll_select_set_timeout(struct timespec *to, long sec, long nsec) | |
197 | { | |
198 | struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec}; | |
199 | ||
200 | if (!timespec_valid(&ts)) | |
201 | return -EINVAL; | |
202 | ||
203 | /* Optimize for the zero timeout value here */ | |
204 | if (!sec && !nsec) { | |
205 | to->tv_sec = to->tv_nsec = 0; | |
206 | } else { | |
207 | ktime_get_ts(to); | |
208 | *to = timespec_add_safe(*to, ts); | |
209 | } | |
210 | return 0; | |
211 | } | |
212 | ||
213 | static int poll_select_copy_remaining(struct timespec *end_time, void __user *p, | |
214 | int timeval, int ret) | |
215 | { | |
216 | struct timespec rts; | |
217 | struct timeval rtv; | |
218 | ||
219 | if (!p) | |
220 | return ret; | |
221 | ||
222 | if (current->personality & STICKY_TIMEOUTS) | |
223 | goto sticky; | |
224 | ||
225 | /* No update for zero timeout */ | |
226 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
227 | return ret; | |
228 | ||
229 | ktime_get_ts(&rts); | |
230 | rts = timespec_sub(*end_time, rts); | |
231 | if (rts.tv_sec < 0) | |
232 | rts.tv_sec = rts.tv_nsec = 0; | |
233 | ||
234 | if (timeval) { | |
235 | rtv.tv_sec = rts.tv_sec; | |
236 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
237 | ||
238 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
239 | return ret; | |
240 | ||
241 | } else if (!copy_to_user(p, &rts, sizeof(rts))) | |
242 | return ret; | |
243 | ||
244 | /* | |
245 | * If an application puts its timeval in read-only memory, we | |
246 | * don't want the Linux-specific update to the timeval to | |
247 | * cause a fault after the select has completed | |
248 | * successfully. However, because we're not updating the | |
249 | * timeval, we can't restart the system call. | |
250 | */ | |
251 | ||
252 | sticky: | |
253 | if (ret == -ERESTARTNOHAND) | |
254 | ret = -EINTR; | |
255 | return ret; | |
256 | } | |
257 | ||
1da177e4 LT |
258 | #define FDS_IN(fds, n) (fds->in + n) |
259 | #define FDS_OUT(fds, n) (fds->out + n) | |
260 | #define FDS_EX(fds, n) (fds->ex + n) | |
261 | ||
262 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
263 | ||
264 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
265 | { | |
266 | unsigned long *open_fds; | |
267 | unsigned long set; | |
268 | int max; | |
badf1662 | 269 | struct fdtable *fdt; |
1da177e4 LT |
270 | |
271 | /* handle last in-complete long-word first */ | |
272 | set = ~(~0UL << (n & (__NFDBITS-1))); | |
273 | n /= __NFDBITS; | |
badf1662 DS |
274 | fdt = files_fdtable(current->files); |
275 | open_fds = fdt->open_fds->fds_bits+n; | |
1da177e4 LT |
276 | max = 0; |
277 | if (set) { | |
278 | set &= BITS(fds, n); | |
279 | if (set) { | |
280 | if (!(set & ~*open_fds)) | |
281 | goto get_max; | |
282 | return -EBADF; | |
283 | } | |
284 | } | |
285 | while (n) { | |
286 | open_fds--; | |
287 | n--; | |
288 | set = BITS(fds, n); | |
289 | if (!set) | |
290 | continue; | |
291 | if (set & ~*open_fds) | |
292 | return -EBADF; | |
293 | if (max) | |
294 | continue; | |
295 | get_max: | |
296 | do { | |
297 | max++; | |
298 | set >>= 1; | |
299 | } while (set); | |
300 | max += n * __NFDBITS; | |
301 | } | |
302 | ||
303 | return max; | |
304 | } | |
305 | ||
1da177e4 LT |
306 | #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) |
307 | #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) | |
308 | #define POLLEX_SET (POLLPRI) | |
309 | ||
8ff3e8e8 | 310 | int do_select(int n, fd_set_bits *fds, struct timespec *end_time) |
1da177e4 | 311 | { |
8ff3e8e8 | 312 | ktime_t expire, *to = NULL; |
1da177e4 LT |
313 | struct poll_wqueues table; |
314 | poll_table *wait; | |
8ff3e8e8 | 315 | int retval, i, timed_out = 0; |
90d6e24a | 316 | unsigned long slack = 0; |
1da177e4 | 317 | |
b835996f | 318 | rcu_read_lock(); |
1da177e4 | 319 | retval = max_select_fd(n, fds); |
b835996f | 320 | rcu_read_unlock(); |
1da177e4 LT |
321 | |
322 | if (retval < 0) | |
323 | return retval; | |
324 | n = retval; | |
325 | ||
326 | poll_initwait(&table); | |
327 | wait = &table.pt; | |
8ff3e8e8 | 328 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
1da177e4 | 329 | wait = NULL; |
8ff3e8e8 AV |
330 | timed_out = 1; |
331 | } | |
332 | ||
90d6e24a AV |
333 | if (end_time) |
334 | slack = estimate_accuracy(end_time); | |
335 | ||
1da177e4 LT |
336 | retval = 0; |
337 | for (;;) { | |
338 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
339 | ||
340 | set_current_state(TASK_INTERRUPTIBLE); | |
341 | ||
342 | inp = fds->in; outp = fds->out; exp = fds->ex; | |
343 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
344 | ||
345 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
346 | unsigned long in, out, ex, all_bits, bit = 1, mask, j; | |
347 | unsigned long res_in = 0, res_out = 0, res_ex = 0; | |
99ac48f5 | 348 | const struct file_operations *f_op = NULL; |
1da177e4 LT |
349 | struct file *file = NULL; |
350 | ||
351 | in = *inp++; out = *outp++; ex = *exp++; | |
352 | all_bits = in | out | ex; | |
353 | if (all_bits == 0) { | |
354 | i += __NFDBITS; | |
355 | continue; | |
356 | } | |
357 | ||
358 | for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) { | |
e4a1f129 | 359 | int fput_needed; |
1da177e4 LT |
360 | if (i >= n) |
361 | break; | |
362 | if (!(bit & all_bits)) | |
363 | continue; | |
e4a1f129 | 364 | file = fget_light(i, &fput_needed); |
1da177e4 LT |
365 | if (file) { |
366 | f_op = file->f_op; | |
367 | mask = DEFAULT_POLLMASK; | |
368 | if (f_op && f_op->poll) | |
369 | mask = (*f_op->poll)(file, retval ? NULL : wait); | |
e4a1f129 | 370 | fput_light(file, fput_needed); |
1da177e4 LT |
371 | if ((mask & POLLIN_SET) && (in & bit)) { |
372 | res_in |= bit; | |
373 | retval++; | |
374 | } | |
375 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
376 | res_out |= bit; | |
377 | retval++; | |
378 | } | |
379 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
380 | res_ex |= bit; | |
381 | retval++; | |
382 | } | |
383 | } | |
1da177e4 LT |
384 | } |
385 | if (res_in) | |
386 | *rinp = res_in; | |
387 | if (res_out) | |
388 | *routp = res_out; | |
389 | if (res_ex) | |
390 | *rexp = res_ex; | |
55d85384 | 391 | cond_resched(); |
1da177e4 LT |
392 | } |
393 | wait = NULL; | |
8ff3e8e8 | 394 | if (retval || timed_out || signal_pending(current)) |
1da177e4 | 395 | break; |
f5264481 | 396 | if (table.error) { |
1da177e4 LT |
397 | retval = table.error; |
398 | break; | |
399 | } | |
9f72949f | 400 | |
8ff3e8e8 AV |
401 | /* |
402 | * If this is the first loop and we have a timeout | |
403 | * given, then we convert to ktime_t and set the to | |
404 | * pointer to the expiry value. | |
405 | */ | |
406 | if (end_time && !to) { | |
407 | expire = timespec_to_ktime(*end_time); | |
408 | to = &expire; | |
9f72949f | 409 | } |
8ff3e8e8 | 410 | |
90d6e24a | 411 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) |
8ff3e8e8 | 412 | timed_out = 1; |
1da177e4 LT |
413 | } |
414 | __set_current_state(TASK_RUNNING); | |
415 | ||
416 | poll_freewait(&table); | |
417 | ||
1da177e4 LT |
418 | return retval; |
419 | } | |
420 | ||
1da177e4 LT |
421 | /* |
422 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
423 | * like to be certain this leads to no problems. So I return | |
424 | * EINTR just for safety. | |
425 | * | |
426 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
427 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
428 | */ | |
429 | #define MAX_SELECT_SECONDS \ | |
430 | ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1) | |
431 | ||
a2dcb44c | 432 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
8ff3e8e8 | 433 | fd_set __user *exp, struct timespec *end_time) |
1da177e4 LT |
434 | { |
435 | fd_set_bits fds; | |
29ff2db5 | 436 | void *bits; |
bbea9f69 | 437 | int ret, max_fds; |
b04eb6aa | 438 | unsigned int size; |
badf1662 | 439 | struct fdtable *fdt; |
70674f95 | 440 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 441 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 442 | |
1da177e4 LT |
443 | ret = -EINVAL; |
444 | if (n < 0) | |
445 | goto out_nofds; | |
446 | ||
bbea9f69 | 447 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 448 | rcu_read_lock(); |
badf1662 | 449 | fdt = files_fdtable(current->files); |
bbea9f69 | 450 | max_fds = fdt->max_fds; |
b835996f | 451 | rcu_read_unlock(); |
bbea9f69 VL |
452 | if (n > max_fds) |
453 | n = max_fds; | |
1da177e4 LT |
454 | |
455 | /* | |
456 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
457 | * since we used fdset we need to allocate memory in units of | |
458 | * long-words. | |
459 | */ | |
1da177e4 | 460 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
461 | bits = stack_fds; |
462 | if (size > sizeof(stack_fds) / 6) { | |
463 | /* Not enough space in on-stack array; must use kmalloc */ | |
464 | ret = -ENOMEM; | |
70674f95 | 465 | bits = kmalloc(6 * size, GFP_KERNEL); |
b04eb6aa MBJ |
466 | if (!bits) |
467 | goto out_nofds; | |
468 | } | |
29ff2db5 AM |
469 | fds.in = bits; |
470 | fds.out = bits + size; | |
471 | fds.ex = bits + 2*size; | |
472 | fds.res_in = bits + 3*size; | |
473 | fds.res_out = bits + 4*size; | |
474 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
475 | |
476 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
477 | (ret = get_fd_set(n, outp, fds.out)) || | |
478 | (ret = get_fd_set(n, exp, fds.ex))) | |
479 | goto out; | |
480 | zero_fd_set(n, fds.res_in); | |
481 | zero_fd_set(n, fds.res_out); | |
482 | zero_fd_set(n, fds.res_ex); | |
483 | ||
8ff3e8e8 | 484 | ret = do_select(n, &fds, end_time); |
1da177e4 LT |
485 | |
486 | if (ret < 0) | |
487 | goto out; | |
488 | if (!ret) { | |
489 | ret = -ERESTARTNOHAND; | |
490 | if (signal_pending(current)) | |
491 | goto out; | |
492 | ret = 0; | |
493 | } | |
494 | ||
495 | if (set_fd_set(n, inp, fds.res_in) || | |
496 | set_fd_set(n, outp, fds.res_out) || | |
497 | set_fd_set(n, exp, fds.res_ex)) | |
498 | ret = -EFAULT; | |
499 | ||
500 | out: | |
70674f95 AK |
501 | if (bits != stack_fds) |
502 | kfree(bits); | |
1da177e4 LT |
503 | out_nofds: |
504 | return ret; | |
505 | } | |
506 | ||
9f72949f DW |
507 | asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
508 | fd_set __user *exp, struct timeval __user *tvp) | |
509 | { | |
8ff3e8e8 | 510 | struct timespec end_time, *to = NULL; |
9f72949f DW |
511 | struct timeval tv; |
512 | int ret; | |
513 | ||
514 | if (tvp) { | |
515 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
516 | return -EFAULT; | |
517 | ||
8ff3e8e8 AV |
518 | to = &end_time; |
519 | if (poll_select_set_timeout(to, tv.tv_sec, | |
520 | tv.tv_usec * NSEC_PER_USEC)) | |
9f72949f | 521 | return -EINVAL; |
9f72949f DW |
522 | } |
523 | ||
8ff3e8e8 AV |
524 | ret = core_sys_select(n, inp, outp, exp, to); |
525 | ret = poll_select_copy_remaining(&end_time, tvp, 1, ret); | |
9f72949f DW |
526 | |
527 | return ret; | |
528 | } | |
529 | ||
f3de272b | 530 | #ifdef HAVE_SET_RESTORE_SIGMASK |
9f72949f DW |
531 | asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp, |
532 | fd_set __user *exp, struct timespec __user *tsp, | |
533 | const sigset_t __user *sigmask, size_t sigsetsize) | |
534 | { | |
9f72949f | 535 | sigset_t ksigmask, sigsaved; |
8ff3e8e8 | 536 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
537 | int ret; |
538 | ||
539 | if (tsp) { | |
540 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
541 | return -EFAULT; | |
542 | ||
8ff3e8e8 AV |
543 | to = &end_time; |
544 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
9f72949f | 545 | return -EINVAL; |
9f72949f DW |
546 | } |
547 | ||
548 | if (sigmask) { | |
549 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
550 | if (sigsetsize != sizeof(sigset_t)) | |
551 | return -EINVAL; | |
552 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
553 | return -EFAULT; | |
554 | ||
555 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
556 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
557 | } | |
558 | ||
8ff3e8e8 AV |
559 | ret = core_sys_select(n, inp, outp, exp, &end_time); |
560 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); | |
9f72949f DW |
561 | |
562 | if (ret == -ERESTARTNOHAND) { | |
563 | /* | |
564 | * Don't restore the signal mask yet. Let do_signal() deliver | |
565 | * the signal on the way back to userspace, before the signal | |
566 | * mask is restored. | |
567 | */ | |
568 | if (sigmask) { | |
569 | memcpy(¤t->saved_sigmask, &sigsaved, | |
570 | sizeof(sigsaved)); | |
4e4c22c7 | 571 | set_restore_sigmask(); |
9f72949f DW |
572 | } |
573 | } else if (sigmask) | |
574 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
575 | ||
576 | return ret; | |
577 | } | |
578 | ||
579 | /* | |
580 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
581 | * 6-argument version where the sixth argument is a pointer to a structure | |
582 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
583 | * the sigset size. | |
584 | */ | |
585 | asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp, | |
586 | fd_set __user *exp, struct timespec __user *tsp, void __user *sig) | |
587 | { | |
588 | size_t sigsetsize = 0; | |
589 | sigset_t __user *up = NULL; | |
590 | ||
591 | if (sig) { | |
592 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
e110ab94 | 593 | || __get_user(up, (sigset_t __user * __user *)sig) |
9f72949f | 594 | || __get_user(sigsetsize, |
e110ab94 | 595 | (size_t __user *)(sig+sizeof(void *)))) |
9f72949f DW |
596 | return -EFAULT; |
597 | } | |
598 | ||
599 | return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize); | |
600 | } | |
f3de272b | 601 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
9f72949f | 602 | |
1da177e4 LT |
603 | struct poll_list { |
604 | struct poll_list *next; | |
605 | int len; | |
606 | struct pollfd entries[0]; | |
607 | }; | |
608 | ||
609 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
610 | ||
4a4b69f7 VL |
611 | /* |
612 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
613 | * interested in events matching the pollfd->events mask, and the result | |
614 | * matching that mask is both recorded in pollfd->revents and returned. The | |
615 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
616 | * if non-NULL. | |
617 | */ | |
618 | static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait) | |
1da177e4 | 619 | { |
4a4b69f7 VL |
620 | unsigned int mask; |
621 | int fd; | |
622 | ||
623 | mask = 0; | |
624 | fd = pollfd->fd; | |
625 | if (fd >= 0) { | |
626 | int fput_needed; | |
627 | struct file * file; | |
628 | ||
629 | file = fget_light(fd, &fput_needed); | |
630 | mask = POLLNVAL; | |
631 | if (file != NULL) { | |
632 | mask = DEFAULT_POLLMASK; | |
633 | if (file->f_op && file->f_op->poll) | |
634 | mask = file->f_op->poll(file, pwait); | |
635 | /* Mask out unneeded events. */ | |
636 | mask &= pollfd->events | POLLERR | POLLHUP; | |
637 | fput_light(file, fput_needed); | |
1da177e4 | 638 | } |
1da177e4 | 639 | } |
4a4b69f7 VL |
640 | pollfd->revents = mask; |
641 | ||
642 | return mask; | |
1da177e4 LT |
643 | } |
644 | ||
645 | static int do_poll(unsigned int nfds, struct poll_list *list, | |
8ff3e8e8 | 646 | struct poll_wqueues *wait, struct timespec *end_time) |
1da177e4 | 647 | { |
1da177e4 | 648 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
649 | ktime_t expire, *to = NULL; |
650 | int timed_out = 0, count = 0; | |
90d6e24a | 651 | unsigned long slack = 0; |
1da177e4 | 652 | |
9f72949f | 653 | /* Optimise the no-wait case */ |
8ff3e8e8 | 654 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
1da177e4 | 655 | pt = NULL; |
8ff3e8e8 AV |
656 | timed_out = 1; |
657 | } | |
9bf084f7 | 658 | |
90d6e24a AV |
659 | if (end_time) |
660 | slack = estimate_accuracy(end_time); | |
661 | ||
1da177e4 LT |
662 | for (;;) { |
663 | struct poll_list *walk; | |
9f72949f | 664 | |
1da177e4 | 665 | set_current_state(TASK_INTERRUPTIBLE); |
4a4b69f7 VL |
666 | for (walk = list; walk != NULL; walk = walk->next) { |
667 | struct pollfd * pfd, * pfd_end; | |
668 | ||
669 | pfd = walk->entries; | |
670 | pfd_end = pfd + walk->len; | |
671 | for (; pfd != pfd_end; pfd++) { | |
672 | /* | |
673 | * Fish for events. If we found one, record it | |
674 | * and kill the poll_table, so we don't | |
675 | * needlessly register any other waiters after | |
676 | * this. They'll get immediately deregistered | |
677 | * when we break out and return. | |
678 | */ | |
679 | if (do_pollfd(pfd, pt)) { | |
680 | count++; | |
681 | pt = NULL; | |
682 | } | |
683 | } | |
1da177e4 | 684 | } |
4a4b69f7 VL |
685 | /* |
686 | * All waiters have already been registered, so don't provide | |
687 | * a poll_table to them on the next loop iteration. | |
688 | */ | |
1da177e4 | 689 | pt = NULL; |
9bf084f7 ON |
690 | if (!count) { |
691 | count = wait->error; | |
692 | if (signal_pending(current)) | |
693 | count = -EINTR; | |
694 | } | |
8ff3e8e8 | 695 | if (count || timed_out) |
1da177e4 | 696 | break; |
9f72949f | 697 | |
8ff3e8e8 AV |
698 | /* |
699 | * If this is the first loop and we have a timeout | |
700 | * given, then we convert to ktime_t and set the to | |
701 | * pointer to the expiry value. | |
702 | */ | |
703 | if (end_time && !to) { | |
704 | expire = timespec_to_ktime(*end_time); | |
705 | to = &expire; | |
9f72949f DW |
706 | } |
707 | ||
90d6e24a | 708 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) |
8ff3e8e8 | 709 | timed_out = 1; |
1da177e4 LT |
710 | } |
711 | __set_current_state(TASK_RUNNING); | |
712 | return count; | |
713 | } | |
714 | ||
70674f95 AK |
715 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
716 | sizeof(struct pollfd)) | |
717 | ||
8ff3e8e8 AV |
718 | int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
719 | struct timespec *end_time) | |
1da177e4 LT |
720 | { |
721 | struct poll_wqueues table; | |
252e5725 | 722 | int err = -EFAULT, fdcount, len, size; |
30c14e40 JS |
723 | /* Allocate small arguments on the stack to save memory and be |
724 | faster - use long to make sure the buffer is aligned properly | |
725 | on 64 bit archs to avoid unaligned access */ | |
726 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
727 | struct poll_list *const head = (struct poll_list *)stack_pps; |
728 | struct poll_list *walk = head; | |
729 | unsigned long todo = nfds; | |
1da177e4 | 730 | |
4e6fd33b | 731 | if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur) |
1da177e4 LT |
732 | return -EINVAL; |
733 | ||
252e5725 ON |
734 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
735 | for (;;) { | |
736 | walk->next = NULL; | |
737 | walk->len = len; | |
738 | if (!len) | |
739 | break; | |
1da177e4 | 740 | |
252e5725 ON |
741 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
742 | sizeof(struct pollfd) * walk->len)) | |
743 | goto out_fds; | |
744 | ||
745 | todo -= walk->len; | |
746 | if (!todo) | |
747 | break; | |
1da177e4 | 748 | |
252e5725 ON |
749 | len = min(todo, POLLFD_PER_PAGE); |
750 | size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; | |
751 | walk = walk->next = kmalloc(size, GFP_KERNEL); | |
752 | if (!walk) { | |
753 | err = -ENOMEM; | |
1da177e4 LT |
754 | goto out_fds; |
755 | } | |
1da177e4 | 756 | } |
9f72949f | 757 | |
252e5725 | 758 | poll_initwait(&table); |
8ff3e8e8 | 759 | fdcount = do_poll(nfds, head, &table, end_time); |
252e5725 | 760 | poll_freewait(&table); |
1da177e4 | 761 | |
252e5725 | 762 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
763 | struct pollfd *fds = walk->entries; |
764 | int j; | |
765 | ||
252e5725 ON |
766 | for (j = 0; j < walk->len; j++, ufds++) |
767 | if (__put_user(fds[j].revents, &ufds->revents)) | |
1da177e4 | 768 | goto out_fds; |
1da177e4 | 769 | } |
252e5725 | 770 | |
1da177e4 | 771 | err = fdcount; |
1da177e4 | 772 | out_fds: |
252e5725 ON |
773 | walk = head->next; |
774 | while (walk) { | |
775 | struct poll_list *pos = walk; | |
776 | walk = walk->next; | |
777 | kfree(pos); | |
1da177e4 | 778 | } |
252e5725 | 779 | |
1da177e4 LT |
780 | return err; |
781 | } | |
9f72949f | 782 | |
3075d9da CW |
783 | static long do_restart_poll(struct restart_block *restart_block) |
784 | { | |
8ff3e8e8 AV |
785 | struct pollfd __user *ufds = restart_block->poll.ufds; |
786 | int nfds = restart_block->poll.nfds; | |
787 | struct timespec *to = NULL, end_time; | |
3075d9da CW |
788 | int ret; |
789 | ||
8ff3e8e8 AV |
790 | if (restart_block->poll.has_timeout) { |
791 | end_time.tv_sec = restart_block->poll.tv_sec; | |
792 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
793 | to = &end_time; | |
794 | } | |
795 | ||
796 | ret = do_sys_poll(ufds, nfds, to); | |
797 | ||
3075d9da CW |
798 | if (ret == -EINTR) { |
799 | restart_block->fn = do_restart_poll; | |
3075d9da CW |
800 | ret = -ERESTART_RESTARTBLOCK; |
801 | } | |
802 | return ret; | |
803 | } | |
804 | ||
9f72949f DW |
805 | asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
806 | long timeout_msecs) | |
807 | { | |
8ff3e8e8 | 808 | struct timespec end_time, *to = NULL; |
3075d9da | 809 | int ret; |
9f72949f | 810 | |
8ff3e8e8 AV |
811 | if (timeout_msecs >= 0) { |
812 | to = &end_time; | |
813 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
814 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
815 | } |
816 | ||
8ff3e8e8 AV |
817 | ret = do_sys_poll(ufds, nfds, to); |
818 | ||
3075d9da CW |
819 | if (ret == -EINTR) { |
820 | struct restart_block *restart_block; | |
8ff3e8e8 | 821 | |
3075d9da CW |
822 | restart_block = ¤t_thread_info()->restart_block; |
823 | restart_block->fn = do_restart_poll; | |
8ff3e8e8 AV |
824 | restart_block->poll.ufds = ufds; |
825 | restart_block->poll.nfds = nfds; | |
826 | ||
827 | if (timeout_msecs >= 0) { | |
828 | restart_block->poll.tv_sec = end_time.tv_sec; | |
829 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
830 | restart_block->poll.has_timeout = 1; | |
831 | } else | |
832 | restart_block->poll.has_timeout = 0; | |
833 | ||
3075d9da CW |
834 | ret = -ERESTART_RESTARTBLOCK; |
835 | } | |
836 | return ret; | |
9f72949f DW |
837 | } |
838 | ||
f3de272b | 839 | #ifdef HAVE_SET_RESTORE_SIGMASK |
9f72949f DW |
840 | asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds, |
841 | struct timespec __user *tsp, const sigset_t __user *sigmask, | |
842 | size_t sigsetsize) | |
843 | { | |
844 | sigset_t ksigmask, sigsaved; | |
8ff3e8e8 | 845 | struct timespec ts, end_time, *to = NULL; |
9f72949f DW |
846 | int ret; |
847 | ||
848 | if (tsp) { | |
849 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
850 | return -EFAULT; | |
851 | ||
8ff3e8e8 AV |
852 | to = &end_time; |
853 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
854 | return -EINVAL; | |
9f72949f DW |
855 | } |
856 | ||
857 | if (sigmask) { | |
858 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
859 | if (sigsetsize != sizeof(sigset_t)) | |
860 | return -EINVAL; | |
861 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
862 | return -EFAULT; | |
863 | ||
864 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
865 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
866 | } | |
867 | ||
8ff3e8e8 | 868 | ret = do_sys_poll(ufds, nfds, to); |
9f72949f DW |
869 | |
870 | /* We can restart this syscall, usually */ | |
871 | if (ret == -EINTR) { | |
872 | /* | |
873 | * Don't restore the signal mask yet. Let do_signal() deliver | |
874 | * the signal on the way back to userspace, before the signal | |
875 | * mask is restored. | |
876 | */ | |
877 | if (sigmask) { | |
878 | memcpy(¤t->saved_sigmask, &sigsaved, | |
879 | sizeof(sigsaved)); | |
4e4c22c7 | 880 | set_restore_sigmask(); |
9f72949f DW |
881 | } |
882 | ret = -ERESTARTNOHAND; | |
883 | } else if (sigmask) | |
884 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
885 | ||
8ff3e8e8 | 886 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
887 | |
888 | return ret; | |
889 | } | |
f3de272b | 890 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |