Ok. I didn't make 2.4.0 in 2000. Tough. I tried, but we had some
[davej-history.git] / fs / select.c
blob975013e7f8097dfe3cfc8c99320b6721306935be
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/malloc.h>
18 #include <linux/smp_lock.h>
19 #include <linux/poll.h>
20 #include <linux/file.h>
22 #include <asm/uaccess.h>
24 #define ROUND_UP(x,y) (((x)+(y)-1)/(y))
25 #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)
27 struct poll_table_entry {
28 struct file * filp;
29 wait_queue_t wait;
30 wait_queue_head_t * wait_address;
33 struct poll_table_page {
34 struct poll_table_page * next;
35 struct poll_table_entry * entry;
36 struct poll_table_entry entries[0];
39 #define POLL_TABLE_FULL(table) \
40 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
43 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
44 * I have rewritten this, taking some shortcuts: This code may not be easy to
45 * follow, but it should be free of race-conditions, and it's practical. If you
46 * understand what I'm doing here, then you understand how the linux
47 * sleep/wakeup mechanism works.
49 * Two very simple procedures, poll_wait() and poll_freewait() make all the
50 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
51 * as all select/poll functions have to call it to add an entry to the
52 * poll table.
55 void poll_freewait(poll_table* pt)
57 struct poll_table_page * p = pt->table;
58 while (p) {
59 struct poll_table_entry * entry;
60 struct poll_table_page *old;
62 entry = p->entry;
63 do {
64 entry--;
65 remove_wait_queue(entry->wait_address,&entry->wait);
66 fput(entry->filp);
67 } while (entry > p->entries);
68 old = p;
69 p = p->next;
70 free_page((unsigned long) old);
74 void __pollwait(struct file * filp, wait_queue_head_t * wait_address, poll_table *p)
76 struct poll_table_page *table = p->table;
78 if (!table || POLL_TABLE_FULL(table)) {
79 struct poll_table_page *new_table;
81 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
82 if (!new_table) {
83 p->error = -ENOMEM;
84 __set_current_state(TASK_RUNNING);
85 return;
87 new_table->entry = new_table->entries;
88 new_table->next = table;
89 p->table = new_table;
90 table = new_table;
93 /* Add a new entry */
95 struct poll_table_entry * entry = table->entry;
96 table->entry = entry+1;
97 get_file(filp);
98 entry->filp = filp;
99 entry->wait_address = wait_address;
100 init_waitqueue_entry(&entry->wait, current);
101 add_wait_queue(wait_address,&entry->wait);
105 #define __IN(fds, n) (fds->in + n)
106 #define __OUT(fds, n) (fds->out + n)
107 #define __EX(fds, n) (fds->ex + n)
108 #define __RES_IN(fds, n) (fds->res_in + n)
109 #define __RES_OUT(fds, n) (fds->res_out + n)
110 #define __RES_EX(fds, n) (fds->res_ex + n)
112 #define BITS(fds, n) (*__IN(fds, n)|*__OUT(fds, n)|*__EX(fds, n))
114 static int max_select_fd(unsigned long n, fd_set_bits *fds)
116 unsigned long *open_fds;
117 unsigned long set;
118 int max;
120 /* handle last in-complete long-word first */
121 set = ~(~0UL << (n & (__NFDBITS-1)));
122 n /= __NFDBITS;
123 open_fds = current->files->open_fds->fds_bits+n;
124 max = 0;
125 if (set) {
126 set &= BITS(fds, n);
127 if (set) {
128 if (!(set & ~*open_fds))
129 goto get_max;
130 return -EBADF;
133 while (n) {
134 open_fds--;
135 n--;
136 set = BITS(fds, n);
137 if (!set)
138 continue;
139 if (set & ~*open_fds)
140 return -EBADF;
141 if (max)
142 continue;
143 get_max:
144 do {
145 max++;
146 set >>= 1;
147 } while (set);
148 max += n * __NFDBITS;
151 return max;
154 #define BIT(i) (1UL << ((i)&(__NFDBITS-1)))
155 #define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS)
156 #define ISSET(i,m) (((i)&*(m)) != 0)
157 #define SET(i,m) (*(m) |= (i))
159 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
160 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
161 #define POLLEX_SET (POLLPRI)
163 int do_select(int n, fd_set_bits *fds, long *timeout)
165 poll_table table, *wait;
166 int retval, i, off;
167 long __timeout = *timeout;
169 read_lock(&current->files->file_lock);
170 retval = max_select_fd(n, fds);
171 read_unlock(&current->files->file_lock);
173 if (retval < 0)
174 return retval;
175 n = retval;
177 poll_initwait(&table);
178 wait = &table;
179 if (!__timeout)
180 wait = NULL;
181 retval = 0;
182 for (;;) {
183 set_current_state(TASK_INTERRUPTIBLE);
184 for (i = 0 ; i < n; i++) {
185 unsigned long bit = BIT(i);
186 unsigned long mask;
187 struct file *file;
189 off = i / __NFDBITS;
190 if (!(bit & BITS(fds, off)))
191 continue;
192 file = fget(i);
193 mask = POLLNVAL;
194 if (file) {
195 mask = DEFAULT_POLLMASK;
196 if (file->f_op && file->f_op->poll)
197 mask = file->f_op->poll(file, wait);
198 fput(file);
200 if ((mask & POLLIN_SET) && ISSET(bit, __IN(fds,off))) {
201 SET(bit, __RES_IN(fds,off));
202 retval++;
203 wait = NULL;
205 if ((mask & POLLOUT_SET) && ISSET(bit, __OUT(fds,off))) {
206 SET(bit, __RES_OUT(fds,off));
207 retval++;
208 wait = NULL;
210 if ((mask & POLLEX_SET) && ISSET(bit, __EX(fds,off))) {
211 SET(bit, __RES_EX(fds,off));
212 retval++;
213 wait = NULL;
216 wait = NULL;
217 if (retval || !__timeout || signal_pending(current))
218 break;
219 if(table.error) {
220 retval = table.error;
221 break;
223 __timeout = schedule_timeout(__timeout);
225 current->state = TASK_RUNNING;
227 poll_freewait(&table);
230 * Up-to-date the caller timeout.
232 *timeout = __timeout;
233 return retval;
236 static void *select_bits_alloc(int size)
238 return kmalloc(6 * size, GFP_KERNEL);
241 static void select_bits_free(void *bits, int size)
243 kfree(bits);
247 * We can actually return ERESTARTSYS instead of EINTR, but I'd
248 * like to be certain this leads to no problems. So I return
249 * EINTR just for safety.
251 * Update: ERESTARTSYS breaks at least the xview clock binary, so
252 * I'm trying ERESTARTNOHAND which restart only when you want to.
254 #define MAX_SELECT_SECONDS \
255 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
257 asmlinkage long
258 sys_select(int n, fd_set *inp, fd_set *outp, fd_set *exp, struct timeval *tvp)
260 fd_set_bits fds;
261 char *bits;
262 long timeout;
263 int ret, size;
265 timeout = MAX_SCHEDULE_TIMEOUT;
266 if (tvp) {
267 time_t sec, usec;
269 if ((ret = verify_area(VERIFY_READ, tvp, sizeof(*tvp)))
270 || (ret = __get_user(sec, &tvp->tv_sec))
271 || (ret = __get_user(usec, &tvp->tv_usec)))
272 goto out_nofds;
274 ret = -EINVAL;
275 if (sec < 0 || usec < 0)
276 goto out_nofds;
278 if ((unsigned long) sec < MAX_SELECT_SECONDS) {
279 timeout = ROUND_UP(usec, 1000000/HZ);
280 timeout += sec * (unsigned long) HZ;
284 ret = -EINVAL;
285 if (n < 0)
286 goto out_nofds;
288 if (n > current->files->max_fdset)
289 n = current->files->max_fdset;
292 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
293 * since we used fdset we need to allocate memory in units of
294 * long-words.
296 ret = -ENOMEM;
297 size = FDS_BYTES(n);
298 bits = select_bits_alloc(size);
299 if (!bits)
300 goto out_nofds;
301 fds.in = (unsigned long *) bits;
302 fds.out = (unsigned long *) (bits + size);
303 fds.ex = (unsigned long *) (bits + 2*size);
304 fds.res_in = (unsigned long *) (bits + 3*size);
305 fds.res_out = (unsigned long *) (bits + 4*size);
306 fds.res_ex = (unsigned long *) (bits + 5*size);
308 if ((ret = get_fd_set(n, inp, fds.in)) ||
309 (ret = get_fd_set(n, outp, fds.out)) ||
310 (ret = get_fd_set(n, exp, fds.ex)))
311 goto out;
312 zero_fd_set(n, fds.res_in);
313 zero_fd_set(n, fds.res_out);
314 zero_fd_set(n, fds.res_ex);
316 ret = do_select(n, &fds, &timeout);
318 if (tvp && !(current->personality & STICKY_TIMEOUTS)) {
319 time_t sec = 0, usec = 0;
320 if (timeout) {
321 sec = timeout / HZ;
322 usec = timeout % HZ;
323 usec *= (1000000/HZ);
325 put_user(sec, &tvp->tv_sec);
326 put_user(usec, &tvp->tv_usec);
329 if (ret < 0)
330 goto out;
331 if (!ret) {
332 ret = -ERESTARTNOHAND;
333 if (signal_pending(current))
334 goto out;
335 ret = 0;
338 set_fd_set(n, inp, fds.res_in);
339 set_fd_set(n, outp, fds.res_out);
340 set_fd_set(n, exp, fds.res_ex);
342 out:
343 select_bits_free(bits, size);
344 out_nofds:
345 return ret;
348 #define POLLFD_PER_PAGE ((PAGE_SIZE) / sizeof(struct pollfd))
350 static void do_pollfd(unsigned int num, struct pollfd * fdpage,
351 poll_table ** pwait, int *count)
353 int i;
355 for (i = 0; i < num; i++) {
356 int fd;
357 unsigned int mask;
358 struct pollfd *fdp;
360 mask = 0;
361 fdp = fdpage+i;
362 fd = fdp->fd;
363 if (fd >= 0) {
364 struct file * file = fget(fd);
365 mask = POLLNVAL;
366 if (file != NULL) {
367 mask = DEFAULT_POLLMASK;
368 if (file->f_op && file->f_op->poll)
369 mask = file->f_op->poll(file, *pwait);
370 mask &= fdp->events | POLLERR | POLLHUP;
371 fput(file);
373 if (mask) {
374 *pwait = NULL;
375 (*count)++;
378 fdp->revents = mask;
382 static int do_poll(unsigned int nfds, unsigned int nchunks, unsigned int nleft,
383 struct pollfd *fds[], poll_table *wait, long timeout)
385 int count;
386 poll_table* pt = wait;
388 for (;;) {
389 unsigned int i;
391 set_current_state(TASK_INTERRUPTIBLE);
392 count = 0;
393 for (i=0; i < nchunks; i++)
394 do_pollfd(POLLFD_PER_PAGE, fds[i], &pt, &count);
395 if (nleft)
396 do_pollfd(nleft, fds[nchunks], &pt, &count);
397 pt = NULL;
398 if (count || !timeout || signal_pending(current))
399 break;
400 count = wait->error;
401 if (count)
402 break;
403 timeout = schedule_timeout(timeout);
405 current->state = TASK_RUNNING;
406 return count;
409 asmlinkage long sys_poll(struct pollfd * ufds, unsigned int nfds, long timeout)
411 int i, j, fdcount, err;
412 struct pollfd **fds;
413 poll_table table, *wait;
414 int nchunks, nleft;
416 /* Do a sanity check on nfds ... */
417 if (nfds > current->files->max_fds)
418 return -EINVAL;
420 if (timeout) {
421 /* Careful about overflow in the intermediate values */
422 if ((unsigned long) timeout < MAX_SCHEDULE_TIMEOUT / HZ)
423 timeout = (unsigned long)(timeout*HZ+999)/1000+1;
424 else /* Negative or overflow */
425 timeout = MAX_SCHEDULE_TIMEOUT;
428 poll_initwait(&table);
429 wait = &table;
430 if (!timeout)
431 wait = NULL;
433 err = -ENOMEM;
434 fds = NULL;
435 if (nfds != 0) {
436 fds = (struct pollfd **)kmalloc(
437 (1 + (nfds - 1) / POLLFD_PER_PAGE) * sizeof(struct pollfd *),
438 GFP_KERNEL);
439 if (fds == NULL)
440 goto out;
443 nchunks = 0;
444 nleft = nfds;
445 while (nleft > POLLFD_PER_PAGE) { /* allocate complete PAGE_SIZE chunks */
446 fds[nchunks] = (struct pollfd *)__get_free_page(GFP_KERNEL);
447 if (fds[nchunks] == NULL)
448 goto out_fds;
449 nchunks++;
450 nleft -= POLLFD_PER_PAGE;
452 if (nleft) { /* allocate last PAGE_SIZE chunk, only nleft elements used */
453 fds[nchunks] = (struct pollfd *)__get_free_page(GFP_KERNEL);
454 if (fds[nchunks] == NULL)
455 goto out_fds;
458 err = -EFAULT;
459 for (i=0; i < nchunks; i++)
460 if (copy_from_user(fds[i], ufds + i*POLLFD_PER_PAGE, PAGE_SIZE))
461 goto out_fds1;
462 if (nleft) {
463 if (copy_from_user(fds[nchunks], ufds + nchunks*POLLFD_PER_PAGE,
464 nleft * sizeof(struct pollfd)))
465 goto out_fds1;
468 fdcount = do_poll(nfds, nchunks, nleft, fds, wait, timeout);
470 /* OK, now copy the revents fields back to user space. */
471 for(i=0; i < nchunks; i++)
472 for (j=0; j < POLLFD_PER_PAGE; j++, ufds++)
473 __put_user((fds[i] + j)->revents, &ufds->revents);
474 if (nleft)
475 for (j=0; j < nleft; j++, ufds++)
476 __put_user((fds[nchunks] + j)->revents, &ufds->revents);
478 err = fdcount;
479 if (!fdcount && signal_pending(current))
480 err = -EINTR;
482 out_fds1:
483 if (nleft)
484 free_page((unsigned long)(fds[nchunks]));
485 out_fds:
486 for (i=0; i < nchunks; i++)
487 free_page((unsigned long)(fds[i]));
488 if (nfds != 0)
489 kfree(fds);
490 out:
491 poll_freewait(&table);
492 return err;