blob: 25b1ccac2f2cdcc73cbc42135837bfed752e7f9e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
25
26#include <asm/uaccess.h>
27
28#define ROUND_UP(x,y) (((x)+(y)-1)/(y))
29#define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)
30
31struct poll_table_entry {
32 struct file * filp;
33 wait_queue_t wait;
34 wait_queue_head_t * wait_address;
35};
36
37struct poll_table_page {
38 struct poll_table_page * next;
39 struct poll_table_entry * entry;
40 struct poll_table_entry entries[0];
41};
42
43#define POLL_TABLE_FULL(table) \
44 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
45
46/*
47 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
48 * I have rewritten this, taking some shortcuts: This code may not be easy to
49 * follow, but it should be free of race-conditions, and it's practical. If you
50 * understand what I'm doing here, then you understand how the linux
51 * sleep/wakeup mechanism works.
52 *
53 * Two very simple procedures, poll_wait() and poll_freewait() make all the
54 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
55 * as all select/poll functions have to call it to add an entry to the
56 * poll table.
57 */
58void __pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p);
59
60void poll_initwait(struct poll_wqueues *pwq)
61{
62 init_poll_funcptr(&pwq->pt, __pollwait);
63 pwq->error = 0;
64 pwq->table = NULL;
65}
66
67EXPORT_SYMBOL(poll_initwait);
68
69void poll_freewait(struct poll_wqueues *pwq)
70{
71 struct poll_table_page * p = pwq->table;
72 while (p) {
73 struct poll_table_entry * entry;
74 struct poll_table_page *old;
75
76 entry = p->entry;
77 do {
78 entry--;
79 remove_wait_queue(entry->wait_address,&entry->wait);
80 fput(entry->filp);
81 } while (entry > p->entries);
82 old = p;
83 p = p->next;
84 free_page((unsigned long) old);
85 }
86}
87
88EXPORT_SYMBOL(poll_freewait);
89
90void __pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *_p)
91{
92 struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt);
93 struct poll_table_page *table = p->table;
94
95 if (!table || POLL_TABLE_FULL(table)) {
96 struct poll_table_page *new_table;
97
98 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
99 if (!new_table) {
100 p->error = -ENOMEM;
101 __set_current_state(TASK_RUNNING);
102 return;
103 }
104 new_table->entry = new_table->entries;
105 new_table->next = table;
106 p->table = new_table;
107 table = new_table;
108 }
109
110 /* Add a new entry */
111 {
112 struct poll_table_entry * entry = table->entry;
113 table->entry = entry+1;
114 get_file(filp);
115 entry->filp = filp;
116 entry->wait_address = wait_address;
117 init_waitqueue_entry(&entry->wait, current);
118 add_wait_queue(wait_address,&entry->wait);
119 }
120}
121
122#define FDS_IN(fds, n) (fds->in + n)
123#define FDS_OUT(fds, n) (fds->out + n)
124#define FDS_EX(fds, n) (fds->ex + n)
125
126#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
127
128static int max_select_fd(unsigned long n, fd_set_bits *fds)
129{
130 unsigned long *open_fds;
131 unsigned long set;
132 int max;
133
134 /* handle last in-complete long-word first */
135 set = ~(~0UL << (n & (__NFDBITS-1)));
136 n /= __NFDBITS;
137 open_fds = current->files->open_fds->fds_bits+n;
138 max = 0;
139 if (set) {
140 set &= BITS(fds, n);
141 if (set) {
142 if (!(set & ~*open_fds))
143 goto get_max;
144 return -EBADF;
145 }
146 }
147 while (n) {
148 open_fds--;
149 n--;
150 set = BITS(fds, n);
151 if (!set)
152 continue;
153 if (set & ~*open_fds)
154 return -EBADF;
155 if (max)
156 continue;
157get_max:
158 do {
159 max++;
160 set >>= 1;
161 } while (set);
162 max += n * __NFDBITS;
163 }
164
165 return max;
166}
167
168#define BIT(i) (1UL << ((i)&(__NFDBITS-1)))
169#define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS)
170#define ISSET(i,m) (((i)&*(m)) != 0)
171#define SET(i,m) (*(m) |= (i))
172
173#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
174#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
175#define POLLEX_SET (POLLPRI)
176
177int do_select(int n, fd_set_bits *fds, long *timeout)
178{
179 struct poll_wqueues table;
180 poll_table *wait;
181 int retval, i;
182 long __timeout = *timeout;
183
184 spin_lock(&current->files->file_lock);
185 retval = max_select_fd(n, fds);
186 spin_unlock(&current->files->file_lock);
187
188 if (retval < 0)
189 return retval;
190 n = retval;
191
192 poll_initwait(&table);
193 wait = &table.pt;
194 if (!__timeout)
195 wait = NULL;
196 retval = 0;
197 for (;;) {
198 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
199
200 set_current_state(TASK_INTERRUPTIBLE);
201
202 inp = fds->in; outp = fds->out; exp = fds->ex;
203 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
204
205 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
206 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
207 unsigned long res_in = 0, res_out = 0, res_ex = 0;
208 struct file_operations *f_op = NULL;
209 struct file *file = NULL;
210
211 in = *inp++; out = *outp++; ex = *exp++;
212 all_bits = in | out | ex;
213 if (all_bits == 0) {
214 i += __NFDBITS;
215 continue;
216 }
217
218 for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
219 if (i >= n)
220 break;
221 if (!(bit & all_bits))
222 continue;
223 file = fget(i);
224 if (file) {
225 f_op = file->f_op;
226 mask = DEFAULT_POLLMASK;
227 if (f_op && f_op->poll)
228 mask = (*f_op->poll)(file, retval ? NULL : wait);
229 fput(file);
230 if ((mask & POLLIN_SET) && (in & bit)) {
231 res_in |= bit;
232 retval++;
233 }
234 if ((mask & POLLOUT_SET) && (out & bit)) {
235 res_out |= bit;
236 retval++;
237 }
238 if ((mask & POLLEX_SET) && (ex & bit)) {
239 res_ex |= bit;
240 retval++;
241 }
242 }
243 cond_resched();
244 }
245 if (res_in)
246 *rinp = res_in;
247 if (res_out)
248 *routp = res_out;
249 if (res_ex)
250 *rexp = res_ex;
251 }
252 wait = NULL;
253 if (retval || !__timeout || signal_pending(current))
254 break;
255 if(table.error) {
256 retval = table.error;
257 break;
258 }
259 __timeout = schedule_timeout(__timeout);
260 }
261 __set_current_state(TASK_RUNNING);
262
263 poll_freewait(&table);
264
265 /*
266 * Up-to-date the caller timeout.
267 */
268 *timeout = __timeout;
269 return retval;
270}
271
272static void *select_bits_alloc(int size)
273{
274 return kmalloc(6 * size, GFP_KERNEL);
275}
276
277static void select_bits_free(void *bits, int size)
278{
279 kfree(bits);
280}
281
282/*
283 * We can actually return ERESTARTSYS instead of EINTR, but I'd
284 * like to be certain this leads to no problems. So I return
285 * EINTR just for safety.
286 *
287 * Update: ERESTARTSYS breaks at least the xview clock binary, so
288 * I'm trying ERESTARTNOHAND which restart only when you want to.
289 */
290#define MAX_SELECT_SECONDS \
291 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
292
293asmlinkage long
294sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, struct timeval __user *tvp)
295{
296 fd_set_bits fds;
297 char *bits;
298 long timeout;
299 int ret, size, max_fdset;
300
301 timeout = MAX_SCHEDULE_TIMEOUT;
302 if (tvp) {
303 time_t sec, usec;
304
305 if (!access_ok(VERIFY_READ, tvp, sizeof(*tvp))
306 || __get_user(sec, &tvp->tv_sec)
307 || __get_user(usec, &tvp->tv_usec)) {
308 ret = -EFAULT;
309 goto out_nofds;
310 }
311
312 ret = -EINVAL;
313 if (sec < 0 || usec < 0)
314 goto out_nofds;
315
316 if ((unsigned long) sec < MAX_SELECT_SECONDS) {
317 timeout = ROUND_UP(usec, 1000000/HZ);
318 timeout += sec * (unsigned long) HZ;
319 }
320 }
321
322 ret = -EINVAL;
323 if (n < 0)
324 goto out_nofds;
325
326 /* max_fdset can increase, so grab it once to avoid race */
327 max_fdset = current->files->max_fdset;
328 if (n > max_fdset)
329 n = max_fdset;
330
331 /*
332 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
333 * since we used fdset we need to allocate memory in units of
334 * long-words.
335 */
336 ret = -ENOMEM;
337 size = FDS_BYTES(n);
338 bits = select_bits_alloc(size);
339 if (!bits)
340 goto out_nofds;
341 fds.in = (unsigned long *) bits;
342 fds.out = (unsigned long *) (bits + size);
343 fds.ex = (unsigned long *) (bits + 2*size);
344 fds.res_in = (unsigned long *) (bits + 3*size);
345 fds.res_out = (unsigned long *) (bits + 4*size);
346 fds.res_ex = (unsigned long *) (bits + 5*size);
347
348 if ((ret = get_fd_set(n, inp, fds.in)) ||
349 (ret = get_fd_set(n, outp, fds.out)) ||
350 (ret = get_fd_set(n, exp, fds.ex)))
351 goto out;
352 zero_fd_set(n, fds.res_in);
353 zero_fd_set(n, fds.res_out);
354 zero_fd_set(n, fds.res_ex);
355
356 ret = do_select(n, &fds, &timeout);
357
358 if (tvp && !(current->personality & STICKY_TIMEOUTS)) {
359 time_t sec = 0, usec = 0;
360 if (timeout) {
361 sec = timeout / HZ;
362 usec = timeout % HZ;
363 usec *= (1000000/HZ);
364 }
365 put_user(sec, &tvp->tv_sec);
366 put_user(usec, &tvp->tv_usec);
367 }
368
369 if (ret < 0)
370 goto out;
371 if (!ret) {
372 ret = -ERESTARTNOHAND;
373 if (signal_pending(current))
374 goto out;
375 ret = 0;
376 }
377
378 if (set_fd_set(n, inp, fds.res_in) ||
379 set_fd_set(n, outp, fds.res_out) ||
380 set_fd_set(n, exp, fds.res_ex))
381 ret = -EFAULT;
382
383out:
384 select_bits_free(bits, size);
385out_nofds:
386 return ret;
387}
388
389struct poll_list {
390 struct poll_list *next;
391 int len;
392 struct pollfd entries[0];
393};
394
395#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
396
397static void do_pollfd(unsigned int num, struct pollfd * fdpage,
398 poll_table ** pwait, int *count)
399{
400 int i;
401
402 for (i = 0; i < num; i++) {
403 int fd;
404 unsigned int mask;
405 struct pollfd *fdp;
406
407 mask = 0;
408 fdp = fdpage+i;
409 fd = fdp->fd;
410 if (fd >= 0) {
411 struct file * file = fget(fd);
412 mask = POLLNVAL;
413 if (file != NULL) {
414 mask = DEFAULT_POLLMASK;
415 if (file->f_op && file->f_op->poll)
416 mask = file->f_op->poll(file, *pwait);
417 mask &= fdp->events | POLLERR | POLLHUP;
418 fput(file);
419 }
420 if (mask) {
421 *pwait = NULL;
422 (*count)++;
423 }
424 }
425 fdp->revents = mask;
426 }
427}
428
429static int do_poll(unsigned int nfds, struct poll_list *list,
430 struct poll_wqueues *wait, long timeout)
431{
432 int count = 0;
433 poll_table* pt = &wait->pt;
434
435 if (!timeout)
436 pt = NULL;
437
438 for (;;) {
439 struct poll_list *walk;
440 set_current_state(TASK_INTERRUPTIBLE);
441 walk = list;
442 while(walk != NULL) {
443 do_pollfd( walk->len, walk->entries, &pt, &count);
444 walk = walk->next;
445 }
446 pt = NULL;
447 if (count || !timeout || signal_pending(current))
448 break;
449 count = wait->error;
450 if (count)
451 break;
452 timeout = schedule_timeout(timeout);
453 }
454 __set_current_state(TASK_RUNNING);
455 return count;
456}
457
458asmlinkage long sys_poll(struct pollfd __user * ufds, unsigned int nfds, long timeout)
459{
460 struct poll_wqueues table;
461 int fdcount, err;
462 unsigned int i;
463 struct poll_list *head;
464 struct poll_list *walk;
465
466 /* Do a sanity check on nfds ... */
467 if (nfds > current->files->max_fdset && nfds > OPEN_MAX)
468 return -EINVAL;
469
470 if (timeout) {
471 /* Careful about overflow in the intermediate values */
472 if ((unsigned long) timeout < MAX_SCHEDULE_TIMEOUT / HZ)
473 timeout = (unsigned long)(timeout*HZ+999)/1000+1;
474 else /* Negative or overflow */
475 timeout = MAX_SCHEDULE_TIMEOUT;
476 }
477
478 poll_initwait(&table);
479
480 head = NULL;
481 walk = NULL;
482 i = nfds;
483 err = -ENOMEM;
484 while(i!=0) {
485 struct poll_list *pp;
486 pp = kmalloc(sizeof(struct poll_list)+
487 sizeof(struct pollfd)*
488 (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i),
489 GFP_KERNEL);
490 if(pp==NULL)
491 goto out_fds;
492 pp->next=NULL;
493 pp->len = (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i);
494 if (head == NULL)
495 head = pp;
496 else
497 walk->next = pp;
498
499 walk = pp;
500 if (copy_from_user(pp->entries, ufds + nfds-i,
501 sizeof(struct pollfd)*pp->len)) {
502 err = -EFAULT;
503 goto out_fds;
504 }
505 i -= pp->len;
506 }
507 fdcount = do_poll(nfds, head, &table, timeout);
508
509 /* OK, now copy the revents fields back to user space. */
510 walk = head;
511 err = -EFAULT;
512 while(walk != NULL) {
513 struct pollfd *fds = walk->entries;
514 int j;
515
516 for (j=0; j < walk->len; j++, ufds++) {
517 if(__put_user(fds[j].revents, &ufds->revents))
518 goto out_fds;
519 }
520 walk = walk->next;
521 }
522 err = fdcount;
523 if (!fdcount && signal_pending(current))
524 err = -EINTR;
525out_fds:
526 walk = head;
527 while(walk!=NULL) {
528 struct poll_list *pp = walk->next;
529 kfree(walk);
530 walk = pp;
531 }
532 poll_freewait(&table);
533 return err;
534}