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 | ||
17 | #include <linux/syscalls.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/smp_lock.h> | |
21 | #include <linux/poll.h> | |
22 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ | |
23 | #include <linux/file.h> | |
24 | #include <linux/fs.h> | |
b835996f | 25 | #include <linux/rcupdate.h> |
1da177e4 LT |
26 | |
27 | #include <asm/uaccess.h> | |
28 | ||
29 | #define ROUND_UP(x,y) (((x)+(y)-1)/(y)) | |
30 | #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM) | |
31 | ||
32 | struct poll_table_entry { | |
33 | struct file * filp; | |
34 | wait_queue_t wait; | |
35 | wait_queue_head_t * wait_address; | |
36 | }; | |
37 | ||
38 | struct poll_table_page { | |
39 | struct poll_table_page * next; | |
40 | struct poll_table_entry * entry; | |
41 | struct poll_table_entry entries[0]; | |
42 | }; | |
43 | ||
44 | #define POLL_TABLE_FULL(table) \ | |
45 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
46 | ||
47 | /* | |
48 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
49 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
50 | * follow, but it should be free of race-conditions, and it's practical. If you | |
51 | * understand what I'm doing here, then you understand how the linux | |
52 | * sleep/wakeup mechanism works. | |
53 | * | |
54 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
55 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
56 | * as all select/poll functions have to call it to add an entry to the | |
57 | * poll table. | |
58 | */ | |
75c96f85 AB |
59 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
60 | poll_table *p); | |
1da177e4 LT |
61 | |
62 | void poll_initwait(struct poll_wqueues *pwq) | |
63 | { | |
64 | init_poll_funcptr(&pwq->pt, __pollwait); | |
65 | pwq->error = 0; | |
66 | pwq->table = NULL; | |
67 | } | |
68 | ||
69 | EXPORT_SYMBOL(poll_initwait); | |
70 | ||
71 | void poll_freewait(struct poll_wqueues *pwq) | |
72 | { | |
73 | struct poll_table_page * p = pwq->table; | |
74 | while (p) { | |
75 | struct poll_table_entry * entry; | |
76 | struct poll_table_page *old; | |
77 | ||
78 | entry = p->entry; | |
79 | do { | |
80 | entry--; | |
81 | remove_wait_queue(entry->wait_address,&entry->wait); | |
82 | fput(entry->filp); | |
83 | } while (entry > p->entries); | |
84 | old = p; | |
85 | p = p->next; | |
86 | free_page((unsigned long) old); | |
87 | } | |
88 | } | |
89 | ||
90 | EXPORT_SYMBOL(poll_freewait); | |
91 | ||
75c96f85 AB |
92 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
93 | 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 | ||
98 | if (!table || POLL_TABLE_FULL(table)) { | |
99 | struct poll_table_page *new_table; | |
100 | ||
101 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
102 | if (!new_table) { | |
103 | p->error = -ENOMEM; | |
104 | __set_current_state(TASK_RUNNING); | |
105 | return; | |
106 | } | |
107 | new_table->entry = new_table->entries; | |
108 | new_table->next = table; | |
109 | p->table = new_table; | |
110 | table = new_table; | |
111 | } | |
112 | ||
113 | /* Add a new entry */ | |
114 | { | |
115 | struct poll_table_entry * entry = table->entry; | |
116 | table->entry = entry+1; | |
117 | get_file(filp); | |
118 | entry->filp = filp; | |
119 | entry->wait_address = wait_address; | |
120 | init_waitqueue_entry(&entry->wait, current); | |
121 | add_wait_queue(wait_address,&entry->wait); | |
122 | } | |
123 | } | |
124 | ||
125 | #define FDS_IN(fds, n) (fds->in + n) | |
126 | #define FDS_OUT(fds, n) (fds->out + n) | |
127 | #define FDS_EX(fds, n) (fds->ex + n) | |
128 | ||
129 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
130 | ||
131 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
132 | { | |
133 | unsigned long *open_fds; | |
134 | unsigned long set; | |
135 | int max; | |
badf1662 | 136 | struct fdtable *fdt; |
1da177e4 LT |
137 | |
138 | /* handle last in-complete long-word first */ | |
139 | set = ~(~0UL << (n & (__NFDBITS-1))); | |
140 | n /= __NFDBITS; | |
badf1662 DS |
141 | fdt = files_fdtable(current->files); |
142 | open_fds = fdt->open_fds->fds_bits+n; | |
1da177e4 LT |
143 | max = 0; |
144 | if (set) { | |
145 | set &= BITS(fds, n); | |
146 | if (set) { | |
147 | if (!(set & ~*open_fds)) | |
148 | goto get_max; | |
149 | return -EBADF; | |
150 | } | |
151 | } | |
152 | while (n) { | |
153 | open_fds--; | |
154 | n--; | |
155 | set = BITS(fds, n); | |
156 | if (!set) | |
157 | continue; | |
158 | if (set & ~*open_fds) | |
159 | return -EBADF; | |
160 | if (max) | |
161 | continue; | |
162 | get_max: | |
163 | do { | |
164 | max++; | |
165 | set >>= 1; | |
166 | } while (set); | |
167 | max += n * __NFDBITS; | |
168 | } | |
169 | ||
170 | return max; | |
171 | } | |
172 | ||
173 | #define BIT(i) (1UL << ((i)&(__NFDBITS-1))) | |
174 | #define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS) | |
175 | #define ISSET(i,m) (((i)&*(m)) != 0) | |
176 | #define SET(i,m) (*(m) |= (i)) | |
177 | ||
178 | #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) | |
179 | #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) | |
180 | #define POLLEX_SET (POLLPRI) | |
181 | ||
182 | int do_select(int n, fd_set_bits *fds, long *timeout) | |
183 | { | |
184 | struct poll_wqueues table; | |
185 | poll_table *wait; | |
186 | int retval, i; | |
187 | long __timeout = *timeout; | |
188 | ||
b835996f | 189 | rcu_read_lock(); |
1da177e4 | 190 | retval = max_select_fd(n, fds); |
b835996f | 191 | rcu_read_unlock(); |
1da177e4 LT |
192 | |
193 | if (retval < 0) | |
194 | return retval; | |
195 | n = retval; | |
196 | ||
197 | poll_initwait(&table); | |
198 | wait = &table.pt; | |
199 | if (!__timeout) | |
200 | wait = NULL; | |
201 | retval = 0; | |
202 | for (;;) { | |
203 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
204 | ||
205 | set_current_state(TASK_INTERRUPTIBLE); | |
206 | ||
207 | inp = fds->in; outp = fds->out; exp = fds->ex; | |
208 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
209 | ||
210 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
211 | unsigned long in, out, ex, all_bits, bit = 1, mask, j; | |
212 | unsigned long res_in = 0, res_out = 0, res_ex = 0; | |
213 | struct file_operations *f_op = NULL; | |
214 | struct file *file = NULL; | |
215 | ||
216 | in = *inp++; out = *outp++; ex = *exp++; | |
217 | all_bits = in | out | ex; | |
218 | if (all_bits == 0) { | |
219 | i += __NFDBITS; | |
220 | continue; | |
221 | } | |
222 | ||
223 | for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) { | |
224 | if (i >= n) | |
225 | break; | |
226 | if (!(bit & all_bits)) | |
227 | continue; | |
228 | file = fget(i); | |
229 | if (file) { | |
230 | f_op = file->f_op; | |
231 | mask = DEFAULT_POLLMASK; | |
232 | if (f_op && f_op->poll) | |
233 | mask = (*f_op->poll)(file, retval ? NULL : wait); | |
234 | fput(file); | |
235 | if ((mask & POLLIN_SET) && (in & bit)) { | |
236 | res_in |= bit; | |
237 | retval++; | |
238 | } | |
239 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
240 | res_out |= bit; | |
241 | retval++; | |
242 | } | |
243 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
244 | res_ex |= bit; | |
245 | retval++; | |
246 | } | |
247 | } | |
248 | cond_resched(); | |
249 | } | |
250 | if (res_in) | |
251 | *rinp = res_in; | |
252 | if (res_out) | |
253 | *routp = res_out; | |
254 | if (res_ex) | |
255 | *rexp = res_ex; | |
256 | } | |
257 | wait = NULL; | |
258 | if (retval || !__timeout || signal_pending(current)) | |
259 | break; | |
260 | if(table.error) { | |
261 | retval = table.error; | |
262 | break; | |
263 | } | |
264 | __timeout = schedule_timeout(__timeout); | |
265 | } | |
266 | __set_current_state(TASK_RUNNING); | |
267 | ||
268 | poll_freewait(&table); | |
269 | ||
270 | /* | |
271 | * Up-to-date the caller timeout. | |
272 | */ | |
273 | *timeout = __timeout; | |
274 | return retval; | |
275 | } | |
276 | ||
277 | static void *select_bits_alloc(int size) | |
278 | { | |
279 | return kmalloc(6 * size, GFP_KERNEL); | |
280 | } | |
281 | ||
282 | static void select_bits_free(void *bits, int size) | |
283 | { | |
284 | kfree(bits); | |
285 | } | |
286 | ||
287 | /* | |
288 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
289 | * like to be certain this leads to no problems. So I return | |
290 | * EINTR just for safety. | |
291 | * | |
292 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
293 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
294 | */ | |
295 | #define MAX_SELECT_SECONDS \ | |
296 | ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1) | |
297 | ||
298 | asmlinkage long | |
299 | sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, struct timeval __user *tvp) | |
300 | { | |
301 | fd_set_bits fds; | |
302 | char *bits; | |
303 | long timeout; | |
304 | int ret, size, max_fdset; | |
badf1662 | 305 | struct fdtable *fdt; |
1da177e4 LT |
306 | |
307 | timeout = MAX_SCHEDULE_TIMEOUT; | |
308 | if (tvp) { | |
309 | time_t sec, usec; | |
310 | ||
311 | if (!access_ok(VERIFY_READ, tvp, sizeof(*tvp)) | |
312 | || __get_user(sec, &tvp->tv_sec) | |
313 | || __get_user(usec, &tvp->tv_usec)) { | |
314 | ret = -EFAULT; | |
315 | goto out_nofds; | |
316 | } | |
317 | ||
318 | ret = -EINVAL; | |
319 | if (sec < 0 || usec < 0) | |
320 | goto out_nofds; | |
321 | ||
322 | if ((unsigned long) sec < MAX_SELECT_SECONDS) { | |
323 | timeout = ROUND_UP(usec, 1000000/HZ); | |
324 | timeout += sec * (unsigned long) HZ; | |
325 | } | |
326 | } | |
327 | ||
328 | ret = -EINVAL; | |
329 | if (n < 0) | |
330 | goto out_nofds; | |
331 | ||
332 | /* max_fdset can increase, so grab it once to avoid race */ | |
b835996f | 333 | rcu_read_lock(); |
badf1662 DS |
334 | fdt = files_fdtable(current->files); |
335 | max_fdset = fdt->max_fdset; | |
b835996f | 336 | rcu_read_unlock(); |
1da177e4 LT |
337 | if (n > max_fdset) |
338 | n = max_fdset; | |
339 | ||
340 | /* | |
341 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
342 | * since we used fdset we need to allocate memory in units of | |
343 | * long-words. | |
344 | */ | |
345 | ret = -ENOMEM; | |
346 | size = FDS_BYTES(n); | |
347 | bits = select_bits_alloc(size); | |
348 | if (!bits) | |
349 | goto out_nofds; | |
350 | fds.in = (unsigned long *) bits; | |
351 | fds.out = (unsigned long *) (bits + size); | |
352 | fds.ex = (unsigned long *) (bits + 2*size); | |
353 | fds.res_in = (unsigned long *) (bits + 3*size); | |
354 | fds.res_out = (unsigned long *) (bits + 4*size); | |
355 | fds.res_ex = (unsigned long *) (bits + 5*size); | |
356 | ||
357 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
358 | (ret = get_fd_set(n, outp, fds.out)) || | |
359 | (ret = get_fd_set(n, exp, fds.ex))) | |
360 | goto out; | |
361 | zero_fd_set(n, fds.res_in); | |
362 | zero_fd_set(n, fds.res_out); | |
363 | zero_fd_set(n, fds.res_ex); | |
364 | ||
365 | ret = do_select(n, &fds, &timeout); | |
366 | ||
367 | if (tvp && !(current->personality & STICKY_TIMEOUTS)) { | |
368 | time_t sec = 0, usec = 0; | |
369 | if (timeout) { | |
370 | sec = timeout / HZ; | |
371 | usec = timeout % HZ; | |
372 | usec *= (1000000/HZ); | |
373 | } | |
374 | put_user(sec, &tvp->tv_sec); | |
375 | put_user(usec, &tvp->tv_usec); | |
376 | } | |
377 | ||
378 | if (ret < 0) | |
379 | goto out; | |
380 | if (!ret) { | |
381 | ret = -ERESTARTNOHAND; | |
382 | if (signal_pending(current)) | |
383 | goto out; | |
384 | ret = 0; | |
385 | } | |
386 | ||
387 | if (set_fd_set(n, inp, fds.res_in) || | |
388 | set_fd_set(n, outp, fds.res_out) || | |
389 | set_fd_set(n, exp, fds.res_ex)) | |
390 | ret = -EFAULT; | |
391 | ||
392 | out: | |
393 | select_bits_free(bits, size); | |
394 | out_nofds: | |
395 | return ret; | |
396 | } | |
397 | ||
398 | struct poll_list { | |
399 | struct poll_list *next; | |
400 | int len; | |
401 | struct pollfd entries[0]; | |
402 | }; | |
403 | ||
404 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
405 | ||
406 | static void do_pollfd(unsigned int num, struct pollfd * fdpage, | |
407 | poll_table ** pwait, int *count) | |
408 | { | |
409 | int i; | |
410 | ||
411 | for (i = 0; i < num; i++) { | |
412 | int fd; | |
413 | unsigned int mask; | |
414 | struct pollfd *fdp; | |
415 | ||
416 | mask = 0; | |
417 | fdp = fdpage+i; | |
418 | fd = fdp->fd; | |
419 | if (fd >= 0) { | |
420 | struct file * file = fget(fd); | |
421 | mask = POLLNVAL; | |
422 | if (file != NULL) { | |
423 | mask = DEFAULT_POLLMASK; | |
424 | if (file->f_op && file->f_op->poll) | |
425 | mask = file->f_op->poll(file, *pwait); | |
426 | mask &= fdp->events | POLLERR | POLLHUP; | |
427 | fput(file); | |
428 | } | |
429 | if (mask) { | |
430 | *pwait = NULL; | |
431 | (*count)++; | |
432 | } | |
433 | } | |
434 | fdp->revents = mask; | |
435 | } | |
436 | } | |
437 | ||
438 | static int do_poll(unsigned int nfds, struct poll_list *list, | |
439 | struct poll_wqueues *wait, long timeout) | |
440 | { | |
441 | int count = 0; | |
442 | poll_table* pt = &wait->pt; | |
443 | ||
444 | if (!timeout) | |
445 | pt = NULL; | |
446 | ||
447 | for (;;) { | |
448 | struct poll_list *walk; | |
449 | set_current_state(TASK_INTERRUPTIBLE); | |
450 | walk = list; | |
451 | while(walk != NULL) { | |
452 | do_pollfd( walk->len, walk->entries, &pt, &count); | |
453 | walk = walk->next; | |
454 | } | |
455 | pt = NULL; | |
456 | if (count || !timeout || signal_pending(current)) | |
457 | break; | |
458 | count = wait->error; | |
459 | if (count) | |
460 | break; | |
461 | timeout = schedule_timeout(timeout); | |
462 | } | |
463 | __set_current_state(TASK_RUNNING); | |
464 | return count; | |
465 | } | |
466 | ||
467 | asmlinkage long sys_poll(struct pollfd __user * ufds, unsigned int nfds, long timeout) | |
468 | { | |
469 | struct poll_wqueues table; | |
470 | int fdcount, err; | |
471 | unsigned int i; | |
472 | struct poll_list *head; | |
473 | struct poll_list *walk; | |
badf1662 | 474 | struct fdtable *fdt; |
b835996f | 475 | int max_fdset; |
1da177e4 LT |
476 | |
477 | /* Do a sanity check on nfds ... */ | |
b835996f | 478 | rcu_read_lock(); |
badf1662 | 479 | fdt = files_fdtable(current->files); |
b835996f DS |
480 | max_fdset = fdt->max_fdset; |
481 | rcu_read_unlock(); | |
482 | if (nfds > max_fdset && nfds > OPEN_MAX) | |
1da177e4 LT |
483 | return -EINVAL; |
484 | ||
485 | if (timeout) { | |
486 | /* Careful about overflow in the intermediate values */ | |
487 | if ((unsigned long) timeout < MAX_SCHEDULE_TIMEOUT / HZ) | |
488 | timeout = (unsigned long)(timeout*HZ+999)/1000+1; | |
489 | else /* Negative or overflow */ | |
490 | timeout = MAX_SCHEDULE_TIMEOUT; | |
491 | } | |
492 | ||
493 | poll_initwait(&table); | |
494 | ||
495 | head = NULL; | |
496 | walk = NULL; | |
497 | i = nfds; | |
498 | err = -ENOMEM; | |
499 | while(i!=0) { | |
500 | struct poll_list *pp; | |
501 | pp = kmalloc(sizeof(struct poll_list)+ | |
502 | sizeof(struct pollfd)* | |
503 | (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i), | |
504 | GFP_KERNEL); | |
505 | if(pp==NULL) | |
506 | goto out_fds; | |
507 | pp->next=NULL; | |
508 | pp->len = (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i); | |
509 | if (head == NULL) | |
510 | head = pp; | |
511 | else | |
512 | walk->next = pp; | |
513 | ||
514 | walk = pp; | |
515 | if (copy_from_user(pp->entries, ufds + nfds-i, | |
516 | sizeof(struct pollfd)*pp->len)) { | |
517 | err = -EFAULT; | |
518 | goto out_fds; | |
519 | } | |
520 | i -= pp->len; | |
521 | } | |
522 | fdcount = do_poll(nfds, head, &table, timeout); | |
523 | ||
524 | /* OK, now copy the revents fields back to user space. */ | |
525 | walk = head; | |
526 | err = -EFAULT; | |
527 | while(walk != NULL) { | |
528 | struct pollfd *fds = walk->entries; | |
529 | int j; | |
530 | ||
531 | for (j=0; j < walk->len; j++, ufds++) { | |
532 | if(__put_user(fds[j].revents, &ufds->revents)) | |
533 | goto out_fds; | |
534 | } | |
535 | walk = walk->next; | |
536 | } | |
537 | err = fdcount; | |
538 | if (!fdcount && signal_pending(current)) | |
539 | err = -EINTR; | |
540 | out_fds: | |
541 | walk = head; | |
542 | while(walk!=NULL) { | |
543 | struct poll_list *pp = walk->next; | |
544 | kfree(walk); | |
545 | walk = pp; | |
546 | } | |
547 | poll_freewait(&table); | |
548 | return err; | |
549 | } |