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> |
1da177e4 LT |
27 | |
28 | #include <asm/uaccess.h> | |
29 | ||
1da177e4 LT |
30 | struct poll_table_page { |
31 | struct poll_table_page * next; | |
32 | struct poll_table_entry * entry; | |
33 | struct poll_table_entry entries[0]; | |
34 | }; | |
35 | ||
36 | #define POLL_TABLE_FULL(table) \ | |
37 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
38 | ||
39 | /* | |
40 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
41 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
42 | * follow, but it should be free of race-conditions, and it's practical. If you | |
43 | * understand what I'm doing here, then you understand how the linux | |
44 | * sleep/wakeup mechanism works. | |
45 | * | |
46 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
47 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
48 | * as all select/poll functions have to call it to add an entry to the | |
49 | * poll table. | |
50 | */ | |
75c96f85 AB |
51 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
52 | poll_table *p); | |
1da177e4 LT |
53 | |
54 | void poll_initwait(struct poll_wqueues *pwq) | |
55 | { | |
56 | init_poll_funcptr(&pwq->pt, __pollwait); | |
57 | pwq->error = 0; | |
58 | pwq->table = NULL; | |
70674f95 | 59 | pwq->inline_index = 0; |
1da177e4 LT |
60 | } |
61 | ||
62 | EXPORT_SYMBOL(poll_initwait); | |
63 | ||
70674f95 AK |
64 | static void free_poll_entry(struct poll_table_entry *entry) |
65 | { | |
ccf6780d | 66 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
67 | fput(entry->filp); |
68 | } | |
69 | ||
1da177e4 LT |
70 | void poll_freewait(struct poll_wqueues *pwq) |
71 | { | |
72 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
73 | int i; |
74 | for (i = 0; i < pwq->inline_index; i++) | |
75 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
76 | while (p) { |
77 | struct poll_table_entry * entry; | |
78 | struct poll_table_page *old; | |
79 | ||
80 | entry = p->entry; | |
81 | do { | |
82 | entry--; | |
70674f95 | 83 | free_poll_entry(entry); |
1da177e4 LT |
84 | } while (entry > p->entries); |
85 | old = p; | |
86 | p = p->next; | |
87 | free_page((unsigned long) old); | |
88 | } | |
89 | } | |
90 | ||
91 | EXPORT_SYMBOL(poll_freewait); | |
92 | ||
70674f95 | 93 | static struct poll_table_entry *poll_get_entry(poll_table *_p) |
1da177e4 LT |
94 | { |
95 | struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt); | |
96 | struct poll_table_page *table = p->table; | |
97 | ||
70674f95 AK |
98 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
99 | return p->inline_entries + p->inline_index++; | |
100 | ||
1da177e4 LT |
101 | if (!table || POLL_TABLE_FULL(table)) { |
102 | struct poll_table_page *new_table; | |
103 | ||
104 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
105 | if (!new_table) { | |
106 | p->error = -ENOMEM; | |
107 | __set_current_state(TASK_RUNNING); | |
70674f95 | 108 | return NULL; |
1da177e4 LT |
109 | } |
110 | new_table->entry = new_table->entries; | |
111 | new_table->next = table; | |
112 | p->table = new_table; | |
113 | table = new_table; | |
114 | } | |
115 | ||
70674f95 AK |
116 | return table->entry++; |
117 | } | |
118 | ||
119 | /* Add a new entry */ | |
120 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
121 | poll_table *p) | |
122 | { | |
123 | struct poll_table_entry *entry = poll_get_entry(p); | |
124 | if (!entry) | |
125 | return; | |
126 | get_file(filp); | |
127 | entry->filp = filp; | |
128 | entry->wait_address = wait_address; | |
129 | init_waitqueue_entry(&entry->wait, current); | |
ccf6780d | 130 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
131 | } |
132 | ||
b773ad40 TG |
133 | /** |
134 | * poll_select_set_timeout - helper function to setup the timeout value | |
135 | * @to: pointer to timespec variable for the final timeout | |
136 | * @sec: seconds (from user space) | |
137 | * @nsec: nanoseconds (from user space) | |
138 | * | |
139 | * Note, we do not use a timespec for the user space value here, That | |
140 | * way we can use the function for timeval and compat interfaces as well. | |
141 | * | |
142 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
143 | */ | |
144 | int poll_select_set_timeout(struct timespec *to, long sec, long nsec) | |
145 | { | |
146 | struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec}; | |
147 | ||
148 | if (!timespec_valid(&ts)) | |
149 | return -EINVAL; | |
150 | ||
151 | /* Optimize for the zero timeout value here */ | |
152 | if (!sec && !nsec) { | |
153 | to->tv_sec = to->tv_nsec = 0; | |
154 | } else { | |
155 | ktime_get_ts(to); | |
156 | *to = timespec_add_safe(*to, ts); | |
157 | } | |
158 | return 0; | |
159 | } | |
160 | ||
161 | static int poll_select_copy_remaining(struct timespec *end_time, void __user *p, | |
162 | int timeval, int ret) | |
163 | { | |
164 | struct timespec rts; | |
165 | struct timeval rtv; | |
166 | ||
167 | if (!p) | |
168 | return ret; | |
169 | ||
170 | if (current->personality & STICKY_TIMEOUTS) | |
171 | goto sticky; | |
172 | ||
173 | /* No update for zero timeout */ | |
174 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
175 | return ret; | |
176 | ||
177 | ktime_get_ts(&rts); | |
178 | rts = timespec_sub(*end_time, rts); | |
179 | if (rts.tv_sec < 0) | |
180 | rts.tv_sec = rts.tv_nsec = 0; | |
181 | ||
182 | if (timeval) { | |
183 | rtv.tv_sec = rts.tv_sec; | |
184 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
185 | ||
186 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
187 | return ret; | |
188 | ||
189 | } else if (!copy_to_user(p, &rts, sizeof(rts))) | |
190 | return ret; | |
191 | ||
192 | /* | |
193 | * If an application puts its timeval in read-only memory, we | |
194 | * don't want the Linux-specific update to the timeval to | |
195 | * cause a fault after the select has completed | |
196 | * successfully. However, because we're not updating the | |
197 | * timeval, we can't restart the system call. | |
198 | */ | |
199 | ||
200 | sticky: | |
201 | if (ret == -ERESTARTNOHAND) | |
202 | ret = -EINTR; | |
203 | return ret; | |
204 | } | |
205 | ||
206 | ||
207 | ||
1da177e4 LT |
208 | #define FDS_IN(fds, n) (fds->in + n) |
209 | #define FDS_OUT(fds, n) (fds->out + n) | |
210 | #define FDS_EX(fds, n) (fds->ex + n) | |
211 | ||
212 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
213 | ||
214 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
215 | { | |
216 | unsigned long *open_fds; | |
217 | unsigned long set; | |
218 | int max; | |
badf1662 | 219 | struct fdtable *fdt; |
1da177e4 LT |
220 | |
221 | /* handle last in-complete long-word first */ | |
222 | set = ~(~0UL << (n & (__NFDBITS-1))); | |
223 | n /= __NFDBITS; | |
badf1662 DS |
224 | fdt = files_fdtable(current->files); |
225 | open_fds = fdt->open_fds->fds_bits+n; | |
1da177e4 LT |
226 | max = 0; |
227 | if (set) { | |
228 | set &= BITS(fds, n); | |
229 | if (set) { | |
230 | if (!(set & ~*open_fds)) | |
231 | goto get_max; | |
232 | return -EBADF; | |
233 | } | |
234 | } | |
235 | while (n) { | |
236 | open_fds--; | |
237 | n--; | |
238 | set = BITS(fds, n); | |
239 | if (!set) | |
240 | continue; | |
241 | if (set & ~*open_fds) | |
242 | return -EBADF; | |
243 | if (max) | |
244 | continue; | |
245 | get_max: | |
246 | do { | |
247 | max++; | |
248 | set >>= 1; | |
249 | } while (set); | |
250 | max += n * __NFDBITS; | |
251 | } | |
252 | ||
253 | return max; | |
254 | } | |
255 | ||
1da177e4 LT |
256 | #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) |
257 | #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) | |
258 | #define POLLEX_SET (POLLPRI) | |
259 | ||
9f72949f | 260 | int do_select(int n, fd_set_bits *fds, s64 *timeout) |
1da177e4 LT |
261 | { |
262 | struct poll_wqueues table; | |
263 | poll_table *wait; | |
264 | int retval, i; | |
1da177e4 | 265 | |
b835996f | 266 | rcu_read_lock(); |
1da177e4 | 267 | retval = max_select_fd(n, fds); |
b835996f | 268 | rcu_read_unlock(); |
1da177e4 LT |
269 | |
270 | if (retval < 0) | |
271 | return retval; | |
272 | n = retval; | |
273 | ||
274 | poll_initwait(&table); | |
275 | wait = &table.pt; | |
9f72949f | 276 | if (!*timeout) |
1da177e4 LT |
277 | wait = NULL; |
278 | retval = 0; | |
279 | for (;;) { | |
280 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
9f72949f | 281 | long __timeout; |
1da177e4 LT |
282 | |
283 | set_current_state(TASK_INTERRUPTIBLE); | |
284 | ||
285 | inp = fds->in; outp = fds->out; exp = fds->ex; | |
286 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
287 | ||
288 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
289 | unsigned long in, out, ex, all_bits, bit = 1, mask, j; | |
290 | unsigned long res_in = 0, res_out = 0, res_ex = 0; | |
99ac48f5 | 291 | const struct file_operations *f_op = NULL; |
1da177e4 LT |
292 | struct file *file = NULL; |
293 | ||
294 | in = *inp++; out = *outp++; ex = *exp++; | |
295 | all_bits = in | out | ex; | |
296 | if (all_bits == 0) { | |
297 | i += __NFDBITS; | |
298 | continue; | |
299 | } | |
300 | ||
301 | for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) { | |
e4a1f129 | 302 | int fput_needed; |
1da177e4 LT |
303 | if (i >= n) |
304 | break; | |
305 | if (!(bit & all_bits)) | |
306 | continue; | |
e4a1f129 | 307 | file = fget_light(i, &fput_needed); |
1da177e4 LT |
308 | if (file) { |
309 | f_op = file->f_op; | |
310 | mask = DEFAULT_POLLMASK; | |
311 | if (f_op && f_op->poll) | |
312 | mask = (*f_op->poll)(file, retval ? NULL : wait); | |
e4a1f129 | 313 | fput_light(file, fput_needed); |
1da177e4 LT |
314 | if ((mask & POLLIN_SET) && (in & bit)) { |
315 | res_in |= bit; | |
316 | retval++; | |
317 | } | |
318 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
319 | res_out |= bit; | |
320 | retval++; | |
321 | } | |
322 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
323 | res_ex |= bit; | |
324 | retval++; | |
325 | } | |
326 | } | |
1da177e4 LT |
327 | } |
328 | if (res_in) | |
329 | *rinp = res_in; | |
330 | if (res_out) | |
331 | *routp = res_out; | |
332 | if (res_ex) | |
333 | *rexp = res_ex; | |
55d85384 | 334 | cond_resched(); |
1da177e4 LT |
335 | } |
336 | wait = NULL; | |
9f72949f | 337 | if (retval || !*timeout || signal_pending(current)) |
1da177e4 | 338 | break; |
f5264481 | 339 | if (table.error) { |
1da177e4 LT |
340 | retval = table.error; |
341 | break; | |
342 | } | |
9f72949f DW |
343 | |
344 | if (*timeout < 0) { | |
345 | /* Wait indefinitely */ | |
346 | __timeout = MAX_SCHEDULE_TIMEOUT; | |
347 | } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT - 1)) { | |
348 | /* Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in a loop */ | |
349 | __timeout = MAX_SCHEDULE_TIMEOUT - 1; | |
350 | *timeout -= __timeout; | |
351 | } else { | |
352 | __timeout = *timeout; | |
353 | *timeout = 0; | |
354 | } | |
1da177e4 | 355 | __timeout = schedule_timeout(__timeout); |
9f72949f DW |
356 | if (*timeout >= 0) |
357 | *timeout += __timeout; | |
1da177e4 LT |
358 | } |
359 | __set_current_state(TASK_RUNNING); | |
360 | ||
361 | poll_freewait(&table); | |
362 | ||
1da177e4 LT |
363 | return retval; |
364 | } | |
365 | ||
1da177e4 LT |
366 | /* |
367 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
368 | * like to be certain this leads to no problems. So I return | |
369 | * EINTR just for safety. | |
370 | * | |
371 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
372 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
373 | */ | |
374 | #define MAX_SELECT_SECONDS \ | |
375 | ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1) | |
376 | ||
a2dcb44c | 377 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
9f72949f | 378 | fd_set __user *exp, s64 *timeout) |
1da177e4 LT |
379 | { |
380 | fd_set_bits fds; | |
29ff2db5 | 381 | void *bits; |
bbea9f69 | 382 | int ret, max_fds; |
b04eb6aa | 383 | unsigned int size; |
badf1662 | 384 | struct fdtable *fdt; |
70674f95 | 385 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 386 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 387 | |
1da177e4 LT |
388 | ret = -EINVAL; |
389 | if (n < 0) | |
390 | goto out_nofds; | |
391 | ||
bbea9f69 | 392 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 393 | rcu_read_lock(); |
badf1662 | 394 | fdt = files_fdtable(current->files); |
bbea9f69 | 395 | max_fds = fdt->max_fds; |
b835996f | 396 | rcu_read_unlock(); |
bbea9f69 VL |
397 | if (n > max_fds) |
398 | n = max_fds; | |
1da177e4 LT |
399 | |
400 | /* | |
401 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
402 | * since we used fdset we need to allocate memory in units of | |
403 | * long-words. | |
404 | */ | |
1da177e4 | 405 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
406 | bits = stack_fds; |
407 | if (size > sizeof(stack_fds) / 6) { | |
408 | /* Not enough space in on-stack array; must use kmalloc */ | |
409 | ret = -ENOMEM; | |
70674f95 | 410 | bits = kmalloc(6 * size, GFP_KERNEL); |
b04eb6aa MBJ |
411 | if (!bits) |
412 | goto out_nofds; | |
413 | } | |
29ff2db5 AM |
414 | fds.in = bits; |
415 | fds.out = bits + size; | |
416 | fds.ex = bits + 2*size; | |
417 | fds.res_in = bits + 3*size; | |
418 | fds.res_out = bits + 4*size; | |
419 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
420 | |
421 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
422 | (ret = get_fd_set(n, outp, fds.out)) || | |
423 | (ret = get_fd_set(n, exp, fds.ex))) | |
424 | goto out; | |
425 | zero_fd_set(n, fds.res_in); | |
426 | zero_fd_set(n, fds.res_out); | |
427 | zero_fd_set(n, fds.res_ex); | |
428 | ||
9f72949f | 429 | ret = do_select(n, &fds, timeout); |
1da177e4 LT |
430 | |
431 | if (ret < 0) | |
432 | goto out; | |
433 | if (!ret) { | |
434 | ret = -ERESTARTNOHAND; | |
435 | if (signal_pending(current)) | |
436 | goto out; | |
437 | ret = 0; | |
438 | } | |
439 | ||
440 | if (set_fd_set(n, inp, fds.res_in) || | |
441 | set_fd_set(n, outp, fds.res_out) || | |
442 | set_fd_set(n, exp, fds.res_ex)) | |
443 | ret = -EFAULT; | |
444 | ||
445 | out: | |
70674f95 AK |
446 | if (bits != stack_fds) |
447 | kfree(bits); | |
1da177e4 LT |
448 | out_nofds: |
449 | return ret; | |
450 | } | |
451 | ||
9f72949f DW |
452 | asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
453 | fd_set __user *exp, struct timeval __user *tvp) | |
454 | { | |
455 | s64 timeout = -1; | |
456 | struct timeval tv; | |
457 | int ret; | |
458 | ||
459 | if (tvp) { | |
460 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
461 | return -EFAULT; | |
462 | ||
463 | if (tv.tv_sec < 0 || tv.tv_usec < 0) | |
464 | return -EINVAL; | |
465 | ||
466 | /* Cast to u64 to make GCC stop complaining */ | |
467 | if ((u64)tv.tv_sec >= (u64)MAX_INT64_SECONDS) | |
468 | timeout = -1; /* infinite */ | |
469 | else { | |
022a1692 | 470 | timeout = DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC/HZ); |
9f72949f DW |
471 | timeout += tv.tv_sec * HZ; |
472 | } | |
473 | } | |
474 | ||
475 | ret = core_sys_select(n, inp, outp, exp, &timeout); | |
476 | ||
477 | if (tvp) { | |
643a6545 AM |
478 | struct timeval rtv; |
479 | ||
9f72949f DW |
480 | if (current->personality & STICKY_TIMEOUTS) |
481 | goto sticky; | |
643a6545 AM |
482 | rtv.tv_usec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)); |
483 | rtv.tv_sec = timeout; | |
74910e6c | 484 | if (timeval_compare(&rtv, &tv) >= 0) |
643a6545 AM |
485 | rtv = tv; |
486 | if (copy_to_user(tvp, &rtv, sizeof(rtv))) { | |
9f72949f DW |
487 | sticky: |
488 | /* | |
489 | * If an application puts its timeval in read-only | |
490 | * memory, we don't want the Linux-specific update to | |
491 | * the timeval to cause a fault after the select has | |
492 | * completed successfully. However, because we're not | |
493 | * updating the timeval, we can't restart the system | |
494 | * call. | |
495 | */ | |
496 | if (ret == -ERESTARTNOHAND) | |
497 | ret = -EINTR; | |
498 | } | |
499 | } | |
500 | ||
501 | return ret; | |
502 | } | |
503 | ||
f3de272b | 504 | #ifdef HAVE_SET_RESTORE_SIGMASK |
9f72949f DW |
505 | asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp, |
506 | fd_set __user *exp, struct timespec __user *tsp, | |
507 | const sigset_t __user *sigmask, size_t sigsetsize) | |
508 | { | |
509 | s64 timeout = MAX_SCHEDULE_TIMEOUT; | |
510 | sigset_t ksigmask, sigsaved; | |
511 | struct timespec ts; | |
512 | int ret; | |
513 | ||
514 | if (tsp) { | |
515 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
516 | return -EFAULT; | |
517 | ||
518 | if (ts.tv_sec < 0 || ts.tv_nsec < 0) | |
519 | return -EINVAL; | |
520 | ||
521 | /* Cast to u64 to make GCC stop complaining */ | |
522 | if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS) | |
523 | timeout = -1; /* infinite */ | |
524 | else { | |
022a1692 | 525 | timeout = DIV_ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ); |
9f72949f DW |
526 | timeout += ts.tv_sec * HZ; |
527 | } | |
528 | } | |
529 | ||
530 | if (sigmask) { | |
531 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
532 | if (sigsetsize != sizeof(sigset_t)) | |
533 | return -EINVAL; | |
534 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
535 | return -EFAULT; | |
536 | ||
537 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
538 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
539 | } | |
540 | ||
541 | ret = core_sys_select(n, inp, outp, exp, &timeout); | |
542 | ||
543 | if (tsp) { | |
643a6545 AM |
544 | struct timespec rts; |
545 | ||
9f72949f DW |
546 | if (current->personality & STICKY_TIMEOUTS) |
547 | goto sticky; | |
643a6545 AM |
548 | rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) * |
549 | 1000; | |
550 | rts.tv_sec = timeout; | |
74910e6c | 551 | if (timespec_compare(&rts, &ts) >= 0) |
643a6545 AM |
552 | rts = ts; |
553 | if (copy_to_user(tsp, &rts, sizeof(rts))) { | |
9f72949f DW |
554 | sticky: |
555 | /* | |
556 | * If an application puts its timeval in read-only | |
557 | * memory, we don't want the Linux-specific update to | |
558 | * the timeval to cause a fault after the select has | |
559 | * completed successfully. However, because we're not | |
560 | * updating the timeval, we can't restart the system | |
561 | * call. | |
562 | */ | |
563 | if (ret == -ERESTARTNOHAND) | |
564 | ret = -EINTR; | |
565 | } | |
566 | } | |
567 | ||
568 | if (ret == -ERESTARTNOHAND) { | |
569 | /* | |
570 | * Don't restore the signal mask yet. Let do_signal() deliver | |
571 | * the signal on the way back to userspace, before the signal | |
572 | * mask is restored. | |
573 | */ | |
574 | if (sigmask) { | |
575 | memcpy(¤t->saved_sigmask, &sigsaved, | |
576 | sizeof(sigsaved)); | |
4e4c22c7 | 577 | set_restore_sigmask(); |
9f72949f DW |
578 | } |
579 | } else if (sigmask) | |
580 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
581 | ||
582 | return ret; | |
583 | } | |
584 | ||
585 | /* | |
586 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
587 | * 6-argument version where the sixth argument is a pointer to a structure | |
588 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
589 | * the sigset size. | |
590 | */ | |
591 | asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp, | |
592 | fd_set __user *exp, struct timespec __user *tsp, void __user *sig) | |
593 | { | |
594 | size_t sigsetsize = 0; | |
595 | sigset_t __user *up = NULL; | |
596 | ||
597 | if (sig) { | |
598 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
e110ab94 | 599 | || __get_user(up, (sigset_t __user * __user *)sig) |
9f72949f | 600 | || __get_user(sigsetsize, |
e110ab94 | 601 | (size_t __user *)(sig+sizeof(void *)))) |
9f72949f DW |
602 | return -EFAULT; |
603 | } | |
604 | ||
605 | return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize); | |
606 | } | |
f3de272b | 607 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
9f72949f | 608 | |
1da177e4 LT |
609 | struct poll_list { |
610 | struct poll_list *next; | |
611 | int len; | |
612 | struct pollfd entries[0]; | |
613 | }; | |
614 | ||
615 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
616 | ||
4a4b69f7 VL |
617 | /* |
618 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
619 | * interested in events matching the pollfd->events mask, and the result | |
620 | * matching that mask is both recorded in pollfd->revents and returned. The | |
621 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
622 | * if non-NULL. | |
623 | */ | |
624 | static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait) | |
1da177e4 | 625 | { |
4a4b69f7 VL |
626 | unsigned int mask; |
627 | int fd; | |
628 | ||
629 | mask = 0; | |
630 | fd = pollfd->fd; | |
631 | if (fd >= 0) { | |
632 | int fput_needed; | |
633 | struct file * file; | |
634 | ||
635 | file = fget_light(fd, &fput_needed); | |
636 | mask = POLLNVAL; | |
637 | if (file != NULL) { | |
638 | mask = DEFAULT_POLLMASK; | |
639 | if (file->f_op && file->f_op->poll) | |
640 | mask = file->f_op->poll(file, pwait); | |
641 | /* Mask out unneeded events. */ | |
642 | mask &= pollfd->events | POLLERR | POLLHUP; | |
643 | fput_light(file, fput_needed); | |
1da177e4 | 644 | } |
1da177e4 | 645 | } |
4a4b69f7 VL |
646 | pollfd->revents = mask; |
647 | ||
648 | return mask; | |
1da177e4 LT |
649 | } |
650 | ||
651 | static int do_poll(unsigned int nfds, struct poll_list *list, | |
9f72949f | 652 | struct poll_wqueues *wait, s64 *timeout) |
1da177e4 LT |
653 | { |
654 | int count = 0; | |
655 | poll_table* pt = &wait->pt; | |
656 | ||
9f72949f DW |
657 | /* Optimise the no-wait case */ |
658 | if (!(*timeout)) | |
1da177e4 | 659 | pt = NULL; |
9bf084f7 | 660 | |
1da177e4 LT |
661 | for (;;) { |
662 | struct poll_list *walk; | |
9f72949f DW |
663 | long __timeout; |
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 | } | |
695 | if (count || !*timeout) | |
1da177e4 | 696 | break; |
9f72949f DW |
697 | |
698 | if (*timeout < 0) { | |
699 | /* Wait indefinitely */ | |
700 | __timeout = MAX_SCHEDULE_TIMEOUT; | |
701 | } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT-1)) { | |
702 | /* | |
703 | * Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in | |
704 | * a loop | |
705 | */ | |
706 | __timeout = MAX_SCHEDULE_TIMEOUT - 1; | |
707 | *timeout -= __timeout; | |
708 | } else { | |
709 | __timeout = *timeout; | |
710 | *timeout = 0; | |
711 | } | |
712 | ||
713 | __timeout = schedule_timeout(__timeout); | |
714 | if (*timeout >= 0) | |
715 | *timeout += __timeout; | |
1da177e4 LT |
716 | } |
717 | __set_current_state(TASK_RUNNING); | |
718 | return count; | |
719 | } | |
720 | ||
70674f95 AK |
721 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
722 | sizeof(struct pollfd)) | |
723 | ||
9f72949f | 724 | int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, s64 *timeout) |
1da177e4 LT |
725 | { |
726 | struct poll_wqueues table; | |
252e5725 | 727 | int err = -EFAULT, fdcount, len, size; |
30c14e40 JS |
728 | /* Allocate small arguments on the stack to save memory and be |
729 | faster - use long to make sure the buffer is aligned properly | |
730 | on 64 bit archs to avoid unaligned access */ | |
731 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
732 | struct poll_list *const head = (struct poll_list *)stack_pps; |
733 | struct poll_list *walk = head; | |
734 | unsigned long todo = nfds; | |
1da177e4 | 735 | |
4e6fd33b | 736 | if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur) |
1da177e4 LT |
737 | return -EINVAL; |
738 | ||
252e5725 ON |
739 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
740 | for (;;) { | |
741 | walk->next = NULL; | |
742 | walk->len = len; | |
743 | if (!len) | |
744 | break; | |
1da177e4 | 745 | |
252e5725 ON |
746 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
747 | sizeof(struct pollfd) * walk->len)) | |
748 | goto out_fds; | |
749 | ||
750 | todo -= walk->len; | |
751 | if (!todo) | |
752 | break; | |
1da177e4 | 753 | |
252e5725 ON |
754 | len = min(todo, POLLFD_PER_PAGE); |
755 | size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; | |
756 | walk = walk->next = kmalloc(size, GFP_KERNEL); | |
757 | if (!walk) { | |
758 | err = -ENOMEM; | |
1da177e4 LT |
759 | goto out_fds; |
760 | } | |
1da177e4 | 761 | } |
9f72949f | 762 | |
252e5725 | 763 | poll_initwait(&table); |
1da177e4 | 764 | fdcount = do_poll(nfds, head, &table, timeout); |
252e5725 | 765 | poll_freewait(&table); |
1da177e4 | 766 | |
252e5725 | 767 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
768 | struct pollfd *fds = walk->entries; |
769 | int j; | |
770 | ||
252e5725 ON |
771 | for (j = 0; j < walk->len; j++, ufds++) |
772 | if (__put_user(fds[j].revents, &ufds->revents)) | |
1da177e4 | 773 | goto out_fds; |
1da177e4 | 774 | } |
252e5725 | 775 | |
1da177e4 | 776 | err = fdcount; |
1da177e4 | 777 | out_fds: |
252e5725 ON |
778 | walk = head->next; |
779 | while (walk) { | |
780 | struct poll_list *pos = walk; | |
781 | walk = walk->next; | |
782 | kfree(pos); | |
1da177e4 | 783 | } |
252e5725 | 784 | |
1da177e4 LT |
785 | return err; |
786 | } | |
9f72949f | 787 | |
3075d9da CW |
788 | static long do_restart_poll(struct restart_block *restart_block) |
789 | { | |
790 | struct pollfd __user *ufds = (struct pollfd __user*)restart_block->arg0; | |
791 | int nfds = restart_block->arg1; | |
792 | s64 timeout = ((s64)restart_block->arg3<<32) | (s64)restart_block->arg2; | |
793 | int ret; | |
794 | ||
795 | ret = do_sys_poll(ufds, nfds, &timeout); | |
796 | if (ret == -EINTR) { | |
797 | restart_block->fn = do_restart_poll; | |
798 | restart_block->arg2 = timeout & 0xFFFFFFFF; | |
799 | restart_block->arg3 = (u64)timeout >> 32; | |
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 | { | |
04a3446c | 808 | s64 timeout_jiffies; |
3075d9da | 809 | int ret; |
9f72949f | 810 | |
04a3446c | 811 | if (timeout_msecs > 0) { |
9f72949f DW |
812 | #if HZ > 1000 |
813 | /* We can only overflow if HZ > 1000 */ | |
814 | if (timeout_msecs / 1000 > (s64)0x7fffffffffffffffULL / (s64)HZ) | |
815 | timeout_jiffies = -1; | |
816 | else | |
817 | #endif | |
844fcc53 | 818 | timeout_jiffies = msecs_to_jiffies(timeout_msecs) + 1; |
04a3446c FI |
819 | } else { |
820 | /* Infinite (< 0) or no (0) timeout */ | |
821 | timeout_jiffies = timeout_msecs; | |
9f72949f DW |
822 | } |
823 | ||
3075d9da CW |
824 | ret = do_sys_poll(ufds, nfds, &timeout_jiffies); |
825 | if (ret == -EINTR) { | |
826 | struct restart_block *restart_block; | |
827 | restart_block = ¤t_thread_info()->restart_block; | |
828 | restart_block->fn = do_restart_poll; | |
829 | restart_block->arg0 = (unsigned long)ufds; | |
830 | restart_block->arg1 = nfds; | |
831 | restart_block->arg2 = timeout_jiffies & 0xFFFFFFFF; | |
832 | restart_block->arg3 = (u64)timeout_jiffies >> 32; | |
833 | ret = -ERESTART_RESTARTBLOCK; | |
834 | } | |
835 | return ret; | |
9f72949f DW |
836 | } |
837 | ||
f3de272b | 838 | #ifdef HAVE_SET_RESTORE_SIGMASK |
9f72949f DW |
839 | asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds, |
840 | struct timespec __user *tsp, const sigset_t __user *sigmask, | |
841 | size_t sigsetsize) | |
842 | { | |
843 | sigset_t ksigmask, sigsaved; | |
844 | struct timespec ts; | |
845 | s64 timeout = -1; | |
846 | int ret; | |
847 | ||
848 | if (tsp) { | |
849 | if (copy_from_user(&ts, tsp, sizeof(ts))) | |
850 | return -EFAULT; | |
851 | ||
852 | /* Cast to u64 to make GCC stop complaining */ | |
853 | if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS) | |
854 | timeout = -1; /* infinite */ | |
855 | else { | |
022a1692 | 856 | timeout = DIV_ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ); |
9f72949f DW |
857 | timeout += ts.tv_sec * HZ; |
858 | } | |
859 | } | |
860 | ||
861 | if (sigmask) { | |
862 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
863 | if (sigsetsize != sizeof(sigset_t)) | |
864 | return -EINVAL; | |
865 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
866 | return -EFAULT; | |
867 | ||
868 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
869 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
870 | } | |
871 | ||
872 | ret = do_sys_poll(ufds, nfds, &timeout); | |
873 | ||
874 | /* We can restart this syscall, usually */ | |
875 | if (ret == -EINTR) { | |
876 | /* | |
877 | * Don't restore the signal mask yet. Let do_signal() deliver | |
878 | * the signal on the way back to userspace, before the signal | |
879 | * mask is restored. | |
880 | */ | |
881 | if (sigmask) { | |
882 | memcpy(¤t->saved_sigmask, &sigsaved, | |
883 | sizeof(sigsaved)); | |
4e4c22c7 | 884 | set_restore_sigmask(); |
9f72949f DW |
885 | } |
886 | ret = -ERESTARTNOHAND; | |
887 | } else if (sigmask) | |
888 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
889 | ||
890 | if (tsp && timeout >= 0) { | |
643a6545 AM |
891 | struct timespec rts; |
892 | ||
9f72949f DW |
893 | if (current->personality & STICKY_TIMEOUTS) |
894 | goto sticky; | |
895 | /* Yes, we know it's actually an s64, but it's also positive. */ | |
643a6545 AM |
896 | rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) * |
897 | 1000; | |
898 | rts.tv_sec = timeout; | |
74910e6c | 899 | if (timespec_compare(&rts, &ts) >= 0) |
643a6545 AM |
900 | rts = ts; |
901 | if (copy_to_user(tsp, &rts, sizeof(rts))) { | |
9f72949f DW |
902 | sticky: |
903 | /* | |
904 | * If an application puts its timeval in read-only | |
905 | * memory, we don't want the Linux-specific update to | |
906 | * the timeval to cause a fault after the select has | |
907 | * completed successfully. However, because we're not | |
908 | * updating the timeval, we can't restart the system | |
909 | * call. | |
910 | */ | |
911 | if (ret == -ERESTARTNOHAND && timeout >= 0) | |
912 | ret = -EINTR; | |
913 | } | |
914 | } | |
915 | ||
916 | return ret; | |
917 | } | |
f3de272b | 918 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |