blob: 8dda969614a9163f602fa62dca53a411c3400136 [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
Milind Arun Choudhary022a1692007-05-08 00:29:02 -070017#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/syscalls.h>
19#include <linux/module.h>
20#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/poll.h>
22#include <linux/personality.h> /* for STICKY_TIMEOUTS */
23#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040024#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/fs.h>
Dipankar Sarmab8359962005-09-09 13:04:14 -070026#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28#include <asm/uaccess.h>
29
Linus Torvalds1da177e2005-04-16 15:20:36 -070030struct 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 */
Adrian Bunk75c96f82005-05-05 16:16:09 -070051static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
52 poll_table *p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54void poll_initwait(struct poll_wqueues *pwq)
55{
56 init_poll_funcptr(&pwq->pt, __pollwait);
57 pwq->error = 0;
58 pwq->table = NULL;
Andi Kleen70674f92006-03-28 01:56:33 -080059 pwq->inline_index = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060}
61
62EXPORT_SYMBOL(poll_initwait);
63
Andi Kleen70674f92006-03-28 01:56:33 -080064static void free_poll_entry(struct poll_table_entry *entry)
65{
WANG Congccf67802007-05-09 07:10:02 +020066 remove_wait_queue(entry->wait_address, &entry->wait);
Andi Kleen70674f92006-03-28 01:56:33 -080067 fput(entry->filp);
68}
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070void poll_freewait(struct poll_wqueues *pwq)
71{
72 struct poll_table_page * p = pwq->table;
Andi Kleen70674f92006-03-28 01:56:33 -080073 int i;
74 for (i = 0; i < pwq->inline_index; i++)
75 free_poll_entry(pwq->inline_entries + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 while (p) {
77 struct poll_table_entry * entry;
78 struct poll_table_page *old;
79
80 entry = p->entry;
81 do {
82 entry--;
Andi Kleen70674f92006-03-28 01:56:33 -080083 free_poll_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 } while (entry > p->entries);
85 old = p;
86 p = p->next;
87 free_page((unsigned long) old);
88 }
89}
90
91EXPORT_SYMBOL(poll_freewait);
92
Andi Kleen70674f92006-03-28 01:56:33 -080093static struct poll_table_entry *poll_get_entry(poll_table *_p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
95 struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt);
96 struct poll_table_page *table = p->table;
97
Andi Kleen70674f92006-03-28 01:56:33 -080098 if (p->inline_index < N_INLINE_POLL_ENTRIES)
99 return p->inline_entries + p->inline_index++;
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 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);
Andi Kleen70674f92006-03-28 01:56:33 -0800108 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 }
110 new_table->entry = new_table->entries;
111 new_table->next = table;
112 p->table = new_table;
113 table = new_table;
114 }
115
Andi Kleen70674f92006-03-28 01:56:33 -0800116 return table->entry++;
117}
118
119/* Add a new entry */
120static 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);
WANG Congccf67802007-05-09 07:10:02 +0200130 add_wait_queue(wait_address, &entry->wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
133#define FDS_IN(fds, n) (fds->in + n)
134#define FDS_OUT(fds, n) (fds->out + n)
135#define FDS_EX(fds, n) (fds->ex + n)
136
137#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
138
139static int max_select_fd(unsigned long n, fd_set_bits *fds)
140{
141 unsigned long *open_fds;
142 unsigned long set;
143 int max;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700144 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146 /* handle last in-complete long-word first */
147 set = ~(~0UL << (n & (__NFDBITS-1)));
148 n /= __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700149 fdt = files_fdtable(current->files);
150 open_fds = fdt->open_fds->fds_bits+n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 max = 0;
152 if (set) {
153 set &= BITS(fds, n);
154 if (set) {
155 if (!(set & ~*open_fds))
156 goto get_max;
157 return -EBADF;
158 }
159 }
160 while (n) {
161 open_fds--;
162 n--;
163 set = BITS(fds, n);
164 if (!set)
165 continue;
166 if (set & ~*open_fds)
167 return -EBADF;
168 if (max)
169 continue;
170get_max:
171 do {
172 max++;
173 set >>= 1;
174 } while (set);
175 max += n * __NFDBITS;
176 }
177
178 return max;
179}
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
182#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
183#define POLLEX_SET (POLLPRI)
184
David Woodhouse9f729492006-01-18 17:44:05 -0800185int do_select(int n, fd_set_bits *fds, s64 *timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186{
187 struct poll_wqueues table;
188 poll_table *wait;
189 int retval, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Dipankar Sarmab8359962005-09-09 13:04:14 -0700191 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 retval = max_select_fd(n, fds);
Dipankar Sarmab8359962005-09-09 13:04:14 -0700193 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 if (retval < 0)
196 return retval;
197 n = retval;
198
199 poll_initwait(&table);
200 wait = &table.pt;
David Woodhouse9f729492006-01-18 17:44:05 -0800201 if (!*timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 wait = NULL;
203 retval = 0;
204 for (;;) {
205 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
David Woodhouse9f729492006-01-18 17:44:05 -0800206 long __timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
208 set_current_state(TASK_INTERRUPTIBLE);
209
210 inp = fds->in; outp = fds->out; exp = fds->ex;
211 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
212
213 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
214 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
215 unsigned long res_in = 0, res_out = 0, res_ex = 0;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800216 const struct file_operations *f_op = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 struct file *file = NULL;
218
219 in = *inp++; out = *outp++; ex = *exp++;
220 all_bits = in | out | ex;
221 if (all_bits == 0) {
222 i += __NFDBITS;
223 continue;
224 }
225
226 for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
Eric Dumazete4a1f122006-03-28 01:56:34 -0800227 int fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 if (i >= n)
229 break;
230 if (!(bit & all_bits))
231 continue;
Eric Dumazete4a1f122006-03-28 01:56:34 -0800232 file = fget_light(i, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (file) {
234 f_op = file->f_op;
235 mask = DEFAULT_POLLMASK;
236 if (f_op && f_op->poll)
237 mask = (*f_op->poll)(file, retval ? NULL : wait);
Eric Dumazete4a1f122006-03-28 01:56:34 -0800238 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 if ((mask & POLLIN_SET) && (in & bit)) {
240 res_in |= bit;
241 retval++;
242 }
243 if ((mask & POLLOUT_SET) && (out & bit)) {
244 res_out |= bit;
245 retval++;
246 }
247 if ((mask & POLLEX_SET) && (ex & bit)) {
248 res_ex |= bit;
249 retval++;
250 }
251 }
252 cond_resched();
253 }
254 if (res_in)
255 *rinp = res_in;
256 if (res_out)
257 *routp = res_out;
258 if (res_ex)
259 *rexp = res_ex;
260 }
261 wait = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800262 if (retval || !*timeout || signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 break;
Pavel Machekf5264482008-04-21 22:15:06 +0000264 if (table.error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 retval = table.error;
266 break;
267 }
David Woodhouse9f729492006-01-18 17:44:05 -0800268
269 if (*timeout < 0) {
270 /* Wait indefinitely */
271 __timeout = MAX_SCHEDULE_TIMEOUT;
272 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT - 1)) {
273 /* Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in a loop */
274 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
275 *timeout -= __timeout;
276 } else {
277 __timeout = *timeout;
278 *timeout = 0;
279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 __timeout = schedule_timeout(__timeout);
David Woodhouse9f729492006-01-18 17:44:05 -0800281 if (*timeout >= 0)
282 *timeout += __timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 }
284 __set_current_state(TASK_RUNNING);
285
286 poll_freewait(&table);
287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return retval;
289}
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291/*
292 * We can actually return ERESTARTSYS instead of EINTR, but I'd
293 * like to be certain this leads to no problems. So I return
294 * EINTR just for safety.
295 *
296 * Update: ERESTARTSYS breaks at least the xview clock binary, so
297 * I'm trying ERESTARTNOHAND which restart only when you want to.
298 */
299#define MAX_SELECT_SECONDS \
300 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
301
Al Viroa2dcb442008-04-23 14:05:15 -0400302int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
David Woodhouse9f729492006-01-18 17:44:05 -0800303 fd_set __user *exp, s64 *timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
305 fd_set_bits fds;
Andrew Morton29ff2db2006-04-10 22:52:46 -0700306 void *bits;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800307 int ret, max_fds;
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700308 unsigned int size;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700309 struct fdtable *fdt;
Andi Kleen70674f92006-03-28 01:56:33 -0800310 /* Allocate small arguments on the stack to save memory and be faster */
Jes Sorensen30c14e42006-03-31 11:18:57 -0500311 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 ret = -EINVAL;
314 if (n < 0)
315 goto out_nofds;
316
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800317 /* max_fds can increase, so grab it once to avoid race */
Dipankar Sarmab8359962005-09-09 13:04:14 -0700318 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700319 fdt = files_fdtable(current->files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800320 max_fds = fdt->max_fds;
Dipankar Sarmab8359962005-09-09 13:04:14 -0700321 rcu_read_unlock();
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800322 if (n > max_fds)
323 n = max_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
325 /*
326 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
327 * since we used fdset we need to allocate memory in units of
328 * long-words.
329 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 size = FDS_BYTES(n);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700331 bits = stack_fds;
332 if (size > sizeof(stack_fds) / 6) {
333 /* Not enough space in on-stack array; must use kmalloc */
334 ret = -ENOMEM;
Andi Kleen70674f92006-03-28 01:56:33 -0800335 bits = kmalloc(6 * size, GFP_KERNEL);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700336 if (!bits)
337 goto out_nofds;
338 }
Andrew Morton29ff2db2006-04-10 22:52:46 -0700339 fds.in = bits;
340 fds.out = bits + size;
341 fds.ex = bits + 2*size;
342 fds.res_in = bits + 3*size;
343 fds.res_out = bits + 4*size;
344 fds.res_ex = bits + 5*size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 if ((ret = get_fd_set(n, inp, fds.in)) ||
347 (ret = get_fd_set(n, outp, fds.out)) ||
348 (ret = get_fd_set(n, exp, fds.ex)))
349 goto out;
350 zero_fd_set(n, fds.res_in);
351 zero_fd_set(n, fds.res_out);
352 zero_fd_set(n, fds.res_ex);
353
David Woodhouse9f729492006-01-18 17:44:05 -0800354 ret = do_select(n, &fds, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356 if (ret < 0)
357 goto out;
358 if (!ret) {
359 ret = -ERESTARTNOHAND;
360 if (signal_pending(current))
361 goto out;
362 ret = 0;
363 }
364
365 if (set_fd_set(n, inp, fds.res_in) ||
366 set_fd_set(n, outp, fds.res_out) ||
367 set_fd_set(n, exp, fds.res_ex))
368 ret = -EFAULT;
369
370out:
Andi Kleen70674f92006-03-28 01:56:33 -0800371 if (bits != stack_fds)
372 kfree(bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373out_nofds:
374 return ret;
375}
376
David Woodhouse9f729492006-01-18 17:44:05 -0800377asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
378 fd_set __user *exp, struct timeval __user *tvp)
379{
380 s64 timeout = -1;
381 struct timeval tv;
382 int ret;
383
384 if (tvp) {
385 if (copy_from_user(&tv, tvp, sizeof(tv)))
386 return -EFAULT;
387
388 if (tv.tv_sec < 0 || tv.tv_usec < 0)
389 return -EINVAL;
390
391 /* Cast to u64 to make GCC stop complaining */
392 if ((u64)tv.tv_sec >= (u64)MAX_INT64_SECONDS)
393 timeout = -1; /* infinite */
394 else {
Milind Arun Choudhary022a1692007-05-08 00:29:02 -0700395 timeout = DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC/HZ);
David Woodhouse9f729492006-01-18 17:44:05 -0800396 timeout += tv.tv_sec * HZ;
397 }
398 }
399
400 ret = core_sys_select(n, inp, outp, exp, &timeout);
401
402 if (tvp) {
Andrew Morton643a6542006-02-11 17:55:52 -0800403 struct timeval rtv;
404
David Woodhouse9f729492006-01-18 17:44:05 -0800405 if (current->personality & STICKY_TIMEOUTS)
406 goto sticky;
Andrew Morton643a6542006-02-11 17:55:52 -0800407 rtv.tv_usec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ));
408 rtv.tv_sec = timeout;
Andrew Morton74910e62006-02-17 13:52:58 -0800409 if (timeval_compare(&rtv, &tv) >= 0)
Andrew Morton643a6542006-02-11 17:55:52 -0800410 rtv = tv;
411 if (copy_to_user(tvp, &rtv, sizeof(rtv))) {
David Woodhouse9f729492006-01-18 17:44:05 -0800412sticky:
413 /*
414 * If an application puts its timeval in read-only
415 * memory, we don't want the Linux-specific update to
416 * the timeval to cause a fault after the select has
417 * completed successfully. However, because we're not
418 * updating the timeval, we can't restart the system
419 * call.
420 */
421 if (ret == -ERESTARTNOHAND)
422 ret = -EINTR;
423 }
424 }
425
426 return ret;
427}
428
Roland McGrathf3de2722008-04-30 00:53:09 -0700429#ifdef HAVE_SET_RESTORE_SIGMASK
David Woodhouse9f729492006-01-18 17:44:05 -0800430asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp,
431 fd_set __user *exp, struct timespec __user *tsp,
432 const sigset_t __user *sigmask, size_t sigsetsize)
433{
434 s64 timeout = MAX_SCHEDULE_TIMEOUT;
435 sigset_t ksigmask, sigsaved;
436 struct timespec ts;
437 int ret;
438
439 if (tsp) {
440 if (copy_from_user(&ts, tsp, sizeof(ts)))
441 return -EFAULT;
442
443 if (ts.tv_sec < 0 || ts.tv_nsec < 0)
444 return -EINVAL;
445
446 /* Cast to u64 to make GCC stop complaining */
447 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
448 timeout = -1; /* infinite */
449 else {
Milind Arun Choudhary022a1692007-05-08 00:29:02 -0700450 timeout = DIV_ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
David Woodhouse9f729492006-01-18 17:44:05 -0800451 timeout += ts.tv_sec * HZ;
452 }
453 }
454
455 if (sigmask) {
456 /* XXX: Don't preclude handling different sized sigset_t's. */
457 if (sigsetsize != sizeof(sigset_t))
458 return -EINVAL;
459 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
460 return -EFAULT;
461
462 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
463 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
464 }
465
466 ret = core_sys_select(n, inp, outp, exp, &timeout);
467
468 if (tsp) {
Andrew Morton643a6542006-02-11 17:55:52 -0800469 struct timespec rts;
470
David Woodhouse9f729492006-01-18 17:44:05 -0800471 if (current->personality & STICKY_TIMEOUTS)
472 goto sticky;
Andrew Morton643a6542006-02-11 17:55:52 -0800473 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
474 1000;
475 rts.tv_sec = timeout;
Andrew Morton74910e62006-02-17 13:52:58 -0800476 if (timespec_compare(&rts, &ts) >= 0)
Andrew Morton643a6542006-02-11 17:55:52 -0800477 rts = ts;
478 if (copy_to_user(tsp, &rts, sizeof(rts))) {
David Woodhouse9f729492006-01-18 17:44:05 -0800479sticky:
480 /*
481 * If an application puts its timeval in read-only
482 * memory, we don't want the Linux-specific update to
483 * the timeval to cause a fault after the select has
484 * completed successfully. However, because we're not
485 * updating the timeval, we can't restart the system
486 * call.
487 */
488 if (ret == -ERESTARTNOHAND)
489 ret = -EINTR;
490 }
491 }
492
493 if (ret == -ERESTARTNOHAND) {
494 /*
495 * Don't restore the signal mask yet. Let do_signal() deliver
496 * the signal on the way back to userspace, before the signal
497 * mask is restored.
498 */
499 if (sigmask) {
500 memcpy(&current->saved_sigmask, &sigsaved,
501 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -0700502 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -0800503 }
504 } else if (sigmask)
505 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
506
507 return ret;
508}
509
510/*
511 * Most architectures can't handle 7-argument syscalls. So we provide a
512 * 6-argument version where the sixth argument is a pointer to a structure
513 * which has a pointer to the sigset_t itself followed by a size_t containing
514 * the sigset size.
515 */
516asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp,
517 fd_set __user *exp, struct timespec __user *tsp, void __user *sig)
518{
519 size_t sigsetsize = 0;
520 sigset_t __user *up = NULL;
521
522 if (sig) {
523 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
Al Viroe110ab92006-02-01 05:26:09 -0500524 || __get_user(up, (sigset_t __user * __user *)sig)
David Woodhouse9f729492006-01-18 17:44:05 -0800525 || __get_user(sigsetsize,
Al Viroe110ab92006-02-01 05:26:09 -0500526 (size_t __user *)(sig+sizeof(void *))))
David Woodhouse9f729492006-01-18 17:44:05 -0800527 return -EFAULT;
528 }
529
530 return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize);
531}
Roland McGrathf3de2722008-04-30 00:53:09 -0700532#endif /* HAVE_SET_RESTORE_SIGMASK */
David Woodhouse9f729492006-01-18 17:44:05 -0800533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534struct poll_list {
535 struct poll_list *next;
536 int len;
537 struct pollfd entries[0];
538};
539
540#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
541
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700542/*
543 * Fish for pollable events on the pollfd->fd file descriptor. We're only
544 * interested in events matching the pollfd->events mask, and the result
545 * matching that mask is both recorded in pollfd->revents and returned. The
546 * pwait poll_table will be used by the fd-provided poll handler for waiting,
547 * if non-NULL.
548 */
549static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700551 unsigned int mask;
552 int fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700554 mask = 0;
555 fd = pollfd->fd;
556 if (fd >= 0) {
557 int fput_needed;
558 struct file * file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700560 file = fget_light(fd, &fput_needed);
561 mask = POLLNVAL;
562 if (file != NULL) {
563 mask = DEFAULT_POLLMASK;
564 if (file->f_op && file->f_op->poll)
565 mask = file->f_op->poll(file, pwait);
566 /* Mask out unneeded events. */
567 mask &= pollfd->events | POLLERR | POLLHUP;
568 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700571 pollfd->revents = mask;
572
573 return mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
576static int do_poll(unsigned int nfds, struct poll_list *list,
David Woodhouse9f729492006-01-18 17:44:05 -0800577 struct poll_wqueues *wait, s64 *timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
579 int count = 0;
580 poll_table* pt = &wait->pt;
581
David Woodhouse9f729492006-01-18 17:44:05 -0800582 /* Optimise the no-wait case */
583 if (!(*timeout))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 pt = NULL;
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 for (;;) {
587 struct poll_list *walk;
David Woodhouse9f729492006-01-18 17:44:05 -0800588 long __timeout;
589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 set_current_state(TASK_INTERRUPTIBLE);
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700591 for (walk = list; walk != NULL; walk = walk->next) {
592 struct pollfd * pfd, * pfd_end;
593
594 pfd = walk->entries;
595 pfd_end = pfd + walk->len;
596 for (; pfd != pfd_end; pfd++) {
597 /*
598 * Fish for events. If we found one, record it
599 * and kill the poll_table, so we don't
600 * needlessly register any other waiters after
601 * this. They'll get immediately deregistered
602 * when we break out and return.
603 */
604 if (do_pollfd(pfd, pt)) {
605 count++;
606 pt = NULL;
607 }
608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700610 /*
611 * All waiters have already been registered, so don't provide
612 * a poll_table to them on the next loop iteration.
613 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 pt = NULL;
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700615 if (!count) {
616 count = wait->error;
617 if (signal_pending(current))
618 count = -EINTR;
619 }
620 if (count || !*timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 break;
David Woodhouse9f729492006-01-18 17:44:05 -0800622
623 if (*timeout < 0) {
624 /* Wait indefinitely */
625 __timeout = MAX_SCHEDULE_TIMEOUT;
626 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT-1)) {
627 /*
628 * Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in
629 * a loop
630 */
631 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
632 *timeout -= __timeout;
633 } else {
634 __timeout = *timeout;
635 *timeout = 0;
636 }
637
638 __timeout = schedule_timeout(__timeout);
639 if (*timeout >= 0)
640 *timeout += __timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
642 __set_current_state(TASK_RUNNING);
643 return count;
644}
645
Andi Kleen70674f92006-03-28 01:56:33 -0800646#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
647 sizeof(struct pollfd))
648
David Woodhouse9f729492006-01-18 17:44:05 -0800649int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, s64 *timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 struct poll_wqueues table;
Oleg Nesterov252e5722007-10-16 23:26:17 -0700652 int err = -EFAULT, fdcount, len, size;
Jes Sorensen30c14e42006-03-31 11:18:57 -0500653 /* Allocate small arguments on the stack to save memory and be
654 faster - use long to make sure the buffer is aligned properly
655 on 64 bit archs to avoid unaligned access */
656 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
Oleg Nesterov252e5722007-10-16 23:26:17 -0700657 struct poll_list *const head = (struct poll_list *)stack_pps;
658 struct poll_list *walk = head;
659 unsigned long todo = nfds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Chris Snook4e6fd332006-09-29 02:01:33 -0700661 if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return -EINVAL;
663
Oleg Nesterov252e5722007-10-16 23:26:17 -0700664 len = min_t(unsigned int, nfds, N_STACK_PPS);
665 for (;;) {
666 walk->next = NULL;
667 walk->len = len;
668 if (!len)
669 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Oleg Nesterov252e5722007-10-16 23:26:17 -0700671 if (copy_from_user(walk->entries, ufds + nfds-todo,
672 sizeof(struct pollfd) * walk->len))
673 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Oleg Nesterov252e5722007-10-16 23:26:17 -0700675 todo -= walk->len;
676 if (!todo)
677 break;
678
679 len = min(todo, POLLFD_PER_PAGE);
680 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
681 walk = walk->next = kmalloc(size, GFP_KERNEL);
682 if (!walk) {
683 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 goto out_fds;
685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 }
David Woodhouse9f729492006-01-18 17:44:05 -0800687
Oleg Nesterov252e5722007-10-16 23:26:17 -0700688 poll_initwait(&table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 fdcount = do_poll(nfds, head, &table, timeout);
Oleg Nesterov252e5722007-10-16 23:26:17 -0700690 poll_freewait(&table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Oleg Nesterov252e5722007-10-16 23:26:17 -0700692 for (walk = head; walk; walk = walk->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 struct pollfd *fds = walk->entries;
694 int j;
695
Oleg Nesterov252e5722007-10-16 23:26:17 -0700696 for (j = 0; j < walk->len; j++, ufds++)
697 if (__put_user(fds[j].revents, &ufds->revents))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 err = fdcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702out_fds:
Oleg Nesterov252e5722007-10-16 23:26:17 -0700703 walk = head->next;
704 while (walk) {
705 struct poll_list *pos = walk;
706 walk = walk->next;
707 kfree(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 return err;
711}
David Woodhouse9f729492006-01-18 17:44:05 -0800712
Chris Wright3075d9d2007-10-16 23:27:18 -0700713static long do_restart_poll(struct restart_block *restart_block)
714{
715 struct pollfd __user *ufds = (struct pollfd __user*)restart_block->arg0;
716 int nfds = restart_block->arg1;
717 s64 timeout = ((s64)restart_block->arg3<<32) | (s64)restart_block->arg2;
718 int ret;
719
720 ret = do_sys_poll(ufds, nfds, &timeout);
721 if (ret == -EINTR) {
722 restart_block->fn = do_restart_poll;
723 restart_block->arg2 = timeout & 0xFFFFFFFF;
724 restart_block->arg3 = (u64)timeout >> 32;
725 ret = -ERESTART_RESTARTBLOCK;
726 }
727 return ret;
728}
729
David Woodhouse9f729492006-01-18 17:44:05 -0800730asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds,
731 long timeout_msecs)
732{
Frode Isaksen04a34462006-06-25 05:49:09 -0700733 s64 timeout_jiffies;
Chris Wright3075d9d2007-10-16 23:27:18 -0700734 int ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800735
Frode Isaksen04a34462006-06-25 05:49:09 -0700736 if (timeout_msecs > 0) {
David Woodhouse9f729492006-01-18 17:44:05 -0800737#if HZ > 1000
738 /* We can only overflow if HZ > 1000 */
739 if (timeout_msecs / 1000 > (s64)0x7fffffffffffffffULL / (s64)HZ)
740 timeout_jiffies = -1;
741 else
742#endif
Karsten Wiese844fcc52008-02-06 01:37:59 -0800743 timeout_jiffies = msecs_to_jiffies(timeout_msecs) + 1;
Frode Isaksen04a34462006-06-25 05:49:09 -0700744 } else {
745 /* Infinite (< 0) or no (0) timeout */
746 timeout_jiffies = timeout_msecs;
David Woodhouse9f729492006-01-18 17:44:05 -0800747 }
748
Chris Wright3075d9d2007-10-16 23:27:18 -0700749 ret = do_sys_poll(ufds, nfds, &timeout_jiffies);
750 if (ret == -EINTR) {
751 struct restart_block *restart_block;
752 restart_block = &current_thread_info()->restart_block;
753 restart_block->fn = do_restart_poll;
754 restart_block->arg0 = (unsigned long)ufds;
755 restart_block->arg1 = nfds;
756 restart_block->arg2 = timeout_jiffies & 0xFFFFFFFF;
757 restart_block->arg3 = (u64)timeout_jiffies >> 32;
758 ret = -ERESTART_RESTARTBLOCK;
759 }
760 return ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800761}
762
Roland McGrathf3de2722008-04-30 00:53:09 -0700763#ifdef HAVE_SET_RESTORE_SIGMASK
David Woodhouse9f729492006-01-18 17:44:05 -0800764asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds,
765 struct timespec __user *tsp, const sigset_t __user *sigmask,
766 size_t sigsetsize)
767{
768 sigset_t ksigmask, sigsaved;
769 struct timespec ts;
770 s64 timeout = -1;
771 int ret;
772
773 if (tsp) {
774 if (copy_from_user(&ts, tsp, sizeof(ts)))
775 return -EFAULT;
776
777 /* Cast to u64 to make GCC stop complaining */
778 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
779 timeout = -1; /* infinite */
780 else {
Milind Arun Choudhary022a1692007-05-08 00:29:02 -0700781 timeout = DIV_ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
David Woodhouse9f729492006-01-18 17:44:05 -0800782 timeout += ts.tv_sec * HZ;
783 }
784 }
785
786 if (sigmask) {
787 /* XXX: Don't preclude handling different sized sigset_t's. */
788 if (sigsetsize != sizeof(sigset_t))
789 return -EINVAL;
790 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
791 return -EFAULT;
792
793 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
794 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
795 }
796
797 ret = do_sys_poll(ufds, nfds, &timeout);
798
799 /* We can restart this syscall, usually */
800 if (ret == -EINTR) {
801 /*
802 * Don't restore the signal mask yet. Let do_signal() deliver
803 * the signal on the way back to userspace, before the signal
804 * mask is restored.
805 */
806 if (sigmask) {
807 memcpy(&current->saved_sigmask, &sigsaved,
808 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -0700809 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -0800810 }
811 ret = -ERESTARTNOHAND;
812 } else if (sigmask)
813 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
814
815 if (tsp && timeout >= 0) {
Andrew Morton643a6542006-02-11 17:55:52 -0800816 struct timespec rts;
817
David Woodhouse9f729492006-01-18 17:44:05 -0800818 if (current->personality & STICKY_TIMEOUTS)
819 goto sticky;
820 /* Yes, we know it's actually an s64, but it's also positive. */
Andrew Morton643a6542006-02-11 17:55:52 -0800821 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
822 1000;
823 rts.tv_sec = timeout;
Andrew Morton74910e62006-02-17 13:52:58 -0800824 if (timespec_compare(&rts, &ts) >= 0)
Andrew Morton643a6542006-02-11 17:55:52 -0800825 rts = ts;
826 if (copy_to_user(tsp, &rts, sizeof(rts))) {
David Woodhouse9f729492006-01-18 17:44:05 -0800827 sticky:
828 /*
829 * If an application puts its timeval in read-only
830 * memory, we don't want the Linux-specific update to
831 * the timeval to cause a fault after the select has
832 * completed successfully. However, because we're not
833 * updating the timeval, we can't restart the system
834 * call.
835 */
836 if (ret == -ERESTARTNOHAND && timeout >= 0)
837 ret = -EINTR;
838 }
839 }
840
841 return ret;
842}
Roland McGrathf3de2722008-04-30 00:53:09 -0700843#endif /* HAVE_SET_RESTORE_SIGMASK */