[PATCH] Fix up 'linux-dvb' maintainers entry
[linux-2.6/history.git] / fs / select.c
blob16fc1ef95981bb69f3c1461d28c3370361f6e506
1 /*
2 * This file contains the procedures for the handling of select and poll
4 * Created for Linux based loosely upon Mathius Lattner's minix
5 * patches by Peter MacDonald. Heavily edited by Linus.
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.
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).
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/smp_lock.h>
20 #include <linux/poll.h>
21 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
22 #include <linux/file.h>
23 #include <linux/fs.h>
25 #include <asm/uaccess.h>
27 #define ROUND_UP(x,y) (((x)+(y)-1)/(y))
28 #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)
30 struct poll_table_entry {
31 struct file * filp;
32 wait_queue_t wait;
33 wait_queue_head_t * wait_address;
36 struct poll_table_page {
37 struct poll_table_page * next;
38 struct poll_table_entry * entry;
39 struct poll_table_entry entries[0];
42 #define POLL_TABLE_FULL(table) \
43 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
46 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
47 * I have rewritten this, taking some shortcuts: This code may not be easy to
48 * follow, but it should be free of race-conditions, and it's practical. If you
49 * understand what I'm doing here, then you understand how the linux
50 * sleep/wakeup mechanism works.
52 * Two very simple procedures, poll_wait() and poll_freewait() make all the
53 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
54 * as all select/poll functions have to call it to add an entry to the
55 * poll table.
57 void __pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p);
59 void poll_initwait(struct poll_wqueues *pwq)
61 init_poll_funcptr(&pwq->pt, __pollwait);
62 pwq->error = 0;
63 pwq->table = NULL;
66 EXPORT_SYMBOL(poll_initwait);
68 void poll_freewait(struct poll_wqueues *pwq)
70 struct poll_table_page * p = pwq->table;
71 while (p) {
72 struct poll_table_entry * entry;
73 struct poll_table_page *old;
75 entry = p->entry;
76 do {
77 entry--;
78 remove_wait_queue(entry->wait_address,&entry->wait);
79 fput(entry->filp);
80 } while (entry > p->entries);
81 old = p;
82 p = p->next;
83 free_page((unsigned long) old);
87 EXPORT_SYMBOL(poll_freewait);
89 void __pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *_p)
91 struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt);
92 struct poll_table_page *table = p->table;
94 if (!table || POLL_TABLE_FULL(table)) {
95 struct poll_table_page *new_table;
97 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
98 if (!new_table) {
99 p->error = -ENOMEM;
100 __set_current_state(TASK_RUNNING);
101 return;
103 new_table->entry = new_table->entries;
104 new_table->next = table;
105 p->table = new_table;
106 table = new_table;
109 /* Add a new entry */
111 struct poll_table_entry * entry = table->entry;
112 table->entry = entry+1;
113 get_file(filp);
114 entry->filp = filp;
115 entry->wait_address = wait_address;
116 init_waitqueue_entry(&entry->wait, current);
117 add_wait_queue(wait_address,&entry->wait);
122 #define __IN(fds, n) (fds->in + n)
123 #define __OUT(fds, n) (fds->out + n)
124 #define __EX(fds, n) (fds->ex + n)
125 #define __RES_IN(fds, n) (fds->res_in + n)
126 #define __RES_OUT(fds, n) (fds->res_out + n)
127 #define __RES_EX(fds, n) (fds->res_ex + n)
129 #define BITS(fds, n) (*__IN(fds, n)|*__OUT(fds, n)|*__EX(fds, n))
131 static int max_select_fd(unsigned long n, fd_set_bits *fds)
133 unsigned long *open_fds;
134 unsigned long set;
135 int max;
137 /* handle last in-complete long-word first */
138 set = ~(~0UL << (n & (__NFDBITS-1)));
139 n /= __NFDBITS;
140 open_fds = current->files->open_fds->fds_bits+n;
141 max = 0;
142 if (set) {
143 set &= BITS(fds, n);
144 if (set) {
145 if (!(set & ~*open_fds))
146 goto get_max;
147 return -EBADF;
150 while (n) {
151 open_fds--;
152 n--;
153 set = BITS(fds, n);
154 if (!set)
155 continue;
156 if (set & ~*open_fds)
157 return -EBADF;
158 if (max)
159 continue;
160 get_max:
161 do {
162 max++;
163 set >>= 1;
164 } while (set);
165 max += n * __NFDBITS;
168 return max;
171 #define BIT(i) (1UL << ((i)&(__NFDBITS-1)))
172 #define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS)
173 #define ISSET(i,m) (((i)&*(m)) != 0)
174 #define SET(i,m) (*(m) |= (i))
176 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
177 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
178 #define POLLEX_SET (POLLPRI)
180 int do_select(int n, fd_set_bits *fds, long *timeout)
182 struct poll_wqueues table;
183 poll_table *wait;
184 int retval, i;
185 long __timeout = *timeout;
187 spin_lock(&current->files->file_lock);
188 retval = max_select_fd(n, fds);
189 spin_unlock(&current->files->file_lock);
191 if (retval < 0)
192 return retval;
193 n = retval;
195 poll_initwait(&table);
196 wait = &table.pt;
197 if (!__timeout)
198 wait = NULL;
199 retval = 0;
200 for (;;) {
201 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
203 set_current_state(TASK_INTERRUPTIBLE);
205 inp = fds->in; outp = fds->out; exp = fds->ex;
206 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
208 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
209 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
210 unsigned long res_in = 0, res_out = 0, res_ex = 0;
211 struct file_operations *f_op = NULL;
212 struct file *file = NULL;
214 in = *inp++; out = *outp++; ex = *exp++;
215 all_bits = in | out | ex;
216 if (all_bits == 0) {
217 i += __NFDBITS;
218 continue;
221 for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
222 if (i >= n)
223 break;
224 if (!(bit & all_bits))
225 continue;
226 file = fget(i);
227 if (file) {
228 f_op = file->f_op;
229 mask = DEFAULT_POLLMASK;
230 if (f_op && f_op->poll)
231 mask = (*f_op->poll)(file, retval ? NULL : wait);
232 fput(file);
233 if ((mask & POLLIN_SET) && (in & bit)) {
234 res_in |= bit;
235 retval++;
237 if ((mask & POLLOUT_SET) && (out & bit)) {
238 res_out |= bit;
239 retval++;
241 if ((mask & POLLEX_SET) && (ex & bit)) {
242 res_ex |= bit;
243 retval++;
247 if (res_in)
248 *rinp = res_in;
249 if (res_out)
250 *routp = res_out;
251 if (res_ex)
252 *rexp = res_ex;
254 wait = NULL;
255 if (retval || !__timeout || signal_pending(current))
256 break;
257 if(table.error) {
258 retval = table.error;
259 break;
261 __timeout = schedule_timeout(__timeout);
263 __set_current_state(TASK_RUNNING);
265 poll_freewait(&table);
268 * Up-to-date the caller timeout.
270 *timeout = __timeout;
271 return retval;
274 EXPORT_SYMBOL(do_select);
276 static void *select_bits_alloc(int size)
278 return kmalloc(6 * size, GFP_KERNEL);
281 static void select_bits_free(void *bits, int size)
283 kfree(bits);
287 * We can actually return ERESTARTSYS instead of EINTR, but I'd
288 * like to be certain this leads to no problems. So I return
289 * EINTR just for safety.
291 * Update: ERESTARTSYS breaks at least the xview clock binary, so
292 * I'm trying ERESTARTNOHAND which restart only when you want to.
294 #define MAX_SELECT_SECONDS \
295 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
297 asmlinkage long
298 sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, struct timeval __user *tvp)
300 fd_set_bits fds;
301 char *bits;
302 long timeout;
303 int ret, size, max_fdset;
305 timeout = MAX_SCHEDULE_TIMEOUT;
306 if (tvp) {
307 time_t sec, usec;
309 if ((ret = verify_area(VERIFY_READ, tvp, sizeof(*tvp)))
310 || (ret = __get_user(sec, &tvp->tv_sec))
311 || (ret = __get_user(usec, &tvp->tv_usec)))
312 goto out_nofds;
314 ret = -EINVAL;
315 if (sec < 0 || usec < 0)
316 goto out_nofds;
318 if ((unsigned long) sec < MAX_SELECT_SECONDS) {
319 timeout = ROUND_UP(usec, 1000000/HZ);
320 timeout += sec * (unsigned long) HZ;
324 ret = -EINVAL;
325 if (n < 0)
326 goto out_nofds;
328 /* max_fdset can increase, so grab it once to avoid race */
329 max_fdset = current->files->max_fdset;
330 if (n > max_fdset)
331 n = max_fdset;
334 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
335 * since we used fdset we need to allocate memory in units of
336 * long-words.
338 ret = -ENOMEM;
339 size = FDS_BYTES(n);
340 bits = select_bits_alloc(size);
341 if (!bits)
342 goto out_nofds;
343 fds.in = (unsigned long *) bits;
344 fds.out = (unsigned long *) (bits + size);
345 fds.ex = (unsigned long *) (bits + 2*size);
346 fds.res_in = (unsigned long *) (bits + 3*size);
347 fds.res_out = (unsigned long *) (bits + 4*size);
348 fds.res_ex = (unsigned long *) (bits + 5*size);
350 if ((ret = get_fd_set(n, inp, fds.in)) ||
351 (ret = get_fd_set(n, outp, fds.out)) ||
352 (ret = get_fd_set(n, exp, fds.ex)))
353 goto out;
354 zero_fd_set(n, fds.res_in);
355 zero_fd_set(n, fds.res_out);
356 zero_fd_set(n, fds.res_ex);
358 ret = do_select(n, &fds, &timeout);
360 if (tvp && !(current->personality & STICKY_TIMEOUTS)) {
361 time_t sec = 0, usec = 0;
362 if (timeout) {
363 sec = timeout / HZ;
364 usec = timeout % HZ;
365 usec *= (1000000/HZ);
367 put_user(sec, &tvp->tv_sec);
368 put_user(usec, &tvp->tv_usec);
371 if (ret < 0)
372 goto out;
373 if (!ret) {
374 ret = -ERESTARTNOHAND;
375 if (signal_pending(current))
376 goto out;
377 ret = 0;
380 set_fd_set(n, inp, fds.res_in);
381 set_fd_set(n, outp, fds.res_out);
382 set_fd_set(n, exp, fds.res_ex);
384 out:
385 select_bits_free(bits, size);
386 out_nofds:
387 return ret;
390 struct poll_list {
391 struct poll_list *next;
392 int len;
393 struct pollfd entries[0];
396 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
398 static void do_pollfd(unsigned int num, struct pollfd * fdpage,
399 poll_table ** pwait, int *count)
401 int i;
403 for (i = 0; i < num; i++) {
404 int fd;
405 unsigned int mask;
406 struct pollfd *fdp;
408 mask = 0;
409 fdp = fdpage+i;
410 fd = fdp->fd;
411 if (fd >= 0) {
412 struct file * file = fget(fd);
413 mask = POLLNVAL;
414 if (file != NULL) {
415 mask = DEFAULT_POLLMASK;
416 if (file->f_op && file->f_op->poll)
417 mask = file->f_op->poll(file, *pwait);
418 mask &= fdp->events | POLLERR | POLLHUP;
419 fput(file);
421 if (mask) {
422 *pwait = NULL;
423 (*count)++;
426 fdp->revents = mask;
430 static int do_poll(unsigned int nfds, struct poll_list *list,
431 struct poll_wqueues *wait, long timeout)
433 int count = 0;
434 poll_table* pt = &wait->pt;
436 if (!timeout)
437 pt = NULL;
439 for (;;) {
440 struct poll_list *walk;
441 set_current_state(TASK_INTERRUPTIBLE);
442 walk = list;
443 while(walk != NULL) {
444 do_pollfd( walk->len, walk->entries, &pt, &count);
445 walk = walk->next;
447 pt = NULL;
448 if (count || !timeout || signal_pending(current))
449 break;
450 count = wait->error;
451 if (count)
452 break;
453 timeout = schedule_timeout(timeout);
455 __set_current_state(TASK_RUNNING);
456 return count;
459 asmlinkage long sys_poll(struct pollfd __user * ufds, unsigned int nfds, long timeout)
461 struct poll_wqueues table;
462 int fdcount, err;
463 unsigned int i;
464 struct poll_list *head;
465 struct poll_list *walk;
467 /* Do a sanity check on nfds ... */
468 if (nfds > current->files->max_fdset && nfds > OPEN_MAX)
469 return -EINVAL;
471 if (timeout) {
472 /* Careful about overflow in the intermediate values */
473 if ((unsigned long) timeout < MAX_SCHEDULE_TIMEOUT / HZ)
474 timeout = (unsigned long)(timeout*HZ+999)/1000+1;
475 else /* Negative or overflow */
476 timeout = MAX_SCHEDULE_TIMEOUT;
479 poll_initwait(&table);
481 head = NULL;
482 walk = NULL;
483 i = nfds;
484 err = -ENOMEM;
485 while(i!=0) {
486 struct poll_list *pp;
487 pp = kmalloc(sizeof(struct poll_list)+
488 sizeof(struct pollfd)*
489 (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i),
490 GFP_KERNEL);
491 if(pp==NULL)
492 goto out_fds;
493 pp->next=NULL;
494 pp->len = (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i);
495 if (head == NULL)
496 head = pp;
497 else
498 walk->next = pp;
500 walk = pp;
501 if (copy_from_user(pp->entries, ufds + nfds-i,
502 sizeof(struct pollfd)*pp->len)) {
503 err = -EFAULT;
504 goto out_fds;
506 i -= pp->len;
508 fdcount = do_poll(nfds, head, &table, timeout);
510 /* OK, now copy the revents fields back to user space. */
511 walk = head;
512 err = -EFAULT;
513 while(walk != NULL) {
514 struct pollfd *fds = walk->entries;
515 int j;
517 for (j=0; j < walk->len; j++, ufds++) {
518 if(__put_user(fds[j].revents, &ufds->revents))
519 goto out_fds;
521 walk = walk->next;
523 err = fdcount;
524 if (!fdcount && signal_pending(current))
525 err = -EINTR;
526 out_fds:
527 walk = head;
528 while(walk!=NULL) {
529 struct poll_list *pp = walk->next;
530 kfree(walk);
531 walk = pp;
533 poll_freewait(&table);
534 return err;