blob: 87df51eadcf2bd170ab0271e189cf8a580aa1adb [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>
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -070027#include <linux/hrtimer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29#include <asm/uaccess.h>
30
Arjan van de Ven90d6e242008-09-01 15:55:35 -070031
32/*
33 * Estimate expected accuracy in ns from a timeval.
34 *
35 * After quite a bit of churning around, we've settled on
36 * a simple thing of taking 0.1% of the timeout as the
37 * slack, with a cap of 100 msec.
38 * "nice" tasks get a 0.5% slack instead.
39 *
40 * Consider this comment an open invitation to come up with even
41 * better solutions..
42 */
43
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070044static long __estimate_accuracy(struct timespec *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070045{
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070046 long slack;
Arjan van de Ven90d6e242008-09-01 15:55:35 -070047 int divfactor = 1000;
48
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070049 if (task_nice(current) > 0)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070050 divfactor = divfactor / 5;
51
52 slack = tv->tv_nsec / divfactor;
53 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
54
55 if (slack > 100 * NSEC_PER_MSEC)
56 slack = 100 * NSEC_PER_MSEC;
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070057
58 if (slack < 0)
59 slack = 0;
Arjan van de Ven90d6e242008-09-01 15:55:35 -070060 return slack;
61}
62
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070063static long estimate_accuracy(struct timespec *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070064{
65 unsigned long ret;
66 struct timespec now;
67
68 /*
69 * Realtime tasks get a slack of 0 for obvious reasons.
70 */
71
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070072 if (rt_task(current))
Arjan van de Ven90d6e242008-09-01 15:55:35 -070073 return 0;
74
75 ktime_get_ts(&now);
76 now = timespec_sub(*tv, now);
77 ret = __estimate_accuracy(&now);
78 if (ret < current->timer_slack_ns)
79 return current->timer_slack_ns;
80 return ret;
81}
82
83
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085struct poll_table_page {
86 struct poll_table_page * next;
87 struct poll_table_entry * entry;
88 struct poll_table_entry entries[0];
89};
90
91#define POLL_TABLE_FULL(table) \
92 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
93
94/*
95 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
96 * I have rewritten this, taking some shortcuts: This code may not be easy to
97 * follow, but it should be free of race-conditions, and it's practical. If you
98 * understand what I'm doing here, then you understand how the linux
99 * sleep/wakeup mechanism works.
100 *
101 * Two very simple procedures, poll_wait() and poll_freewait() make all the
102 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
103 * as all select/poll functions have to call it to add an entry to the
104 * poll table.
105 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700106static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
107 poll_table *p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
109void poll_initwait(struct poll_wqueues *pwq)
110{
111 init_poll_funcptr(&pwq->pt, __pollwait);
112 pwq->error = 0;
113 pwq->table = NULL;
Andi Kleen70674f92006-03-28 01:56:33 -0800114 pwq->inline_index = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
117EXPORT_SYMBOL(poll_initwait);
118
Andi Kleen70674f92006-03-28 01:56:33 -0800119static void free_poll_entry(struct poll_table_entry *entry)
120{
WANG Congccf67802007-05-09 07:10:02 +0200121 remove_wait_queue(entry->wait_address, &entry->wait);
Andi Kleen70674f92006-03-28 01:56:33 -0800122 fput(entry->filp);
123}
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125void poll_freewait(struct poll_wqueues *pwq)
126{
127 struct poll_table_page * p = pwq->table;
Andi Kleen70674f92006-03-28 01:56:33 -0800128 int i;
129 for (i = 0; i < pwq->inline_index; i++)
130 free_poll_entry(pwq->inline_entries + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 while (p) {
132 struct poll_table_entry * entry;
133 struct poll_table_page *old;
134
135 entry = p->entry;
136 do {
137 entry--;
Andi Kleen70674f92006-03-28 01:56:33 -0800138 free_poll_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 } while (entry > p->entries);
140 old = p;
141 p = p->next;
142 free_page((unsigned long) old);
143 }
144}
145
146EXPORT_SYMBOL(poll_freewait);
147
Andi Kleen70674f92006-03-28 01:56:33 -0800148static struct poll_table_entry *poll_get_entry(poll_table *_p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149{
150 struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt);
151 struct poll_table_page *table = p->table;
152
Andi Kleen70674f92006-03-28 01:56:33 -0800153 if (p->inline_index < N_INLINE_POLL_ENTRIES)
154 return p->inline_entries + p->inline_index++;
155
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 if (!table || POLL_TABLE_FULL(table)) {
157 struct poll_table_page *new_table;
158
159 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
160 if (!new_table) {
161 p->error = -ENOMEM;
162 __set_current_state(TASK_RUNNING);
Andi Kleen70674f92006-03-28 01:56:33 -0800163 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 }
165 new_table->entry = new_table->entries;
166 new_table->next = table;
167 p->table = new_table;
168 table = new_table;
169 }
170
Andi Kleen70674f92006-03-28 01:56:33 -0800171 return table->entry++;
172}
173
174/* Add a new entry */
175static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
176 poll_table *p)
177{
178 struct poll_table_entry *entry = poll_get_entry(p);
179 if (!entry)
180 return;
181 get_file(filp);
182 entry->filp = filp;
183 entry->wait_address = wait_address;
184 init_waitqueue_entry(&entry->wait, current);
WANG Congccf67802007-05-09 07:10:02 +0200185 add_wait_queue(wait_address, &entry->wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700188/**
189 * poll_select_set_timeout - helper function to setup the timeout value
190 * @to: pointer to timespec variable for the final timeout
191 * @sec: seconds (from user space)
192 * @nsec: nanoseconds (from user space)
193 *
194 * Note, we do not use a timespec for the user space value here, That
195 * way we can use the function for timeval and compat interfaces as well.
196 *
197 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
198 */
199int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
200{
201 struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
202
203 if (!timespec_valid(&ts))
204 return -EINVAL;
205
206 /* Optimize for the zero timeout value here */
207 if (!sec && !nsec) {
208 to->tv_sec = to->tv_nsec = 0;
209 } else {
210 ktime_get_ts(to);
211 *to = timespec_add_safe(*to, ts);
212 }
213 return 0;
214}
215
216static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
217 int timeval, int ret)
218{
219 struct timespec rts;
220 struct timeval rtv;
221
222 if (!p)
223 return ret;
224
225 if (current->personality & STICKY_TIMEOUTS)
226 goto sticky;
227
228 /* No update for zero timeout */
229 if (!end_time->tv_sec && !end_time->tv_nsec)
230 return ret;
231
232 ktime_get_ts(&rts);
233 rts = timespec_sub(*end_time, rts);
234 if (rts.tv_sec < 0)
235 rts.tv_sec = rts.tv_nsec = 0;
236
237 if (timeval) {
238 rtv.tv_sec = rts.tv_sec;
239 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
240
241 if (!copy_to_user(p, &rtv, sizeof(rtv)))
242 return ret;
243
244 } else if (!copy_to_user(p, &rts, sizeof(rts)))
245 return ret;
246
247 /*
248 * If an application puts its timeval in read-only memory, we
249 * don't want the Linux-specific update to the timeval to
250 * cause a fault after the select has completed
251 * successfully. However, because we're not updating the
252 * timeval, we can't restart the system call.
253 */
254
255sticky:
256 if (ret == -ERESTARTNOHAND)
257 ret = -EINTR;
258 return ret;
259}
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261#define FDS_IN(fds, n) (fds->in + n)
262#define FDS_OUT(fds, n) (fds->out + n)
263#define FDS_EX(fds, n) (fds->ex + n)
264
265#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
266
267static int max_select_fd(unsigned long n, fd_set_bits *fds)
268{
269 unsigned long *open_fds;
270 unsigned long set;
271 int max;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700272 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 /* handle last in-complete long-word first */
275 set = ~(~0UL << (n & (__NFDBITS-1)));
276 n /= __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700277 fdt = files_fdtable(current->files);
278 open_fds = fdt->open_fds->fds_bits+n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 max = 0;
280 if (set) {
281 set &= BITS(fds, n);
282 if (set) {
283 if (!(set & ~*open_fds))
284 goto get_max;
285 return -EBADF;
286 }
287 }
288 while (n) {
289 open_fds--;
290 n--;
291 set = BITS(fds, n);
292 if (!set)
293 continue;
294 if (set & ~*open_fds)
295 return -EBADF;
296 if (max)
297 continue;
298get_max:
299 do {
300 max++;
301 set >>= 1;
302 } while (set);
303 max += n * __NFDBITS;
304 }
305
306 return max;
307}
308
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
310#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
311#define POLLEX_SET (POLLPRI)
312
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700313int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700315 ktime_t expire, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 struct poll_wqueues table;
317 poll_table *wait;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700318 int retval, i, timed_out = 0;
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700319 unsigned long slack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Dipankar Sarmab8359962005-09-09 13:04:14 -0700321 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 retval = max_select_fd(n, fds);
Dipankar Sarmab8359962005-09-09 13:04:14 -0700323 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
325 if (retval < 0)
326 return retval;
327 n = retval;
328
329 poll_initwait(&table);
330 wait = &table.pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700331 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 wait = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700333 timed_out = 1;
334 }
335
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700336 if (end_time && !timed_out)
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700337 slack = estimate_accuracy(end_time);
338
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 retval = 0;
340 for (;;) {
341 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
342
343 set_current_state(TASK_INTERRUPTIBLE);
344
345 inp = fds->in; outp = fds->out; exp = fds->ex;
346 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
347
348 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
349 unsigned long in, out, ex, all_bits, bit = 1, mask, j;
350 unsigned long res_in = 0, res_out = 0, res_ex = 0;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800351 const struct file_operations *f_op = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 struct file *file = NULL;
353
354 in = *inp++; out = *outp++; ex = *exp++;
355 all_bits = in | out | ex;
356 if (all_bits == 0) {
357 i += __NFDBITS;
358 continue;
359 }
360
361 for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
Eric Dumazete4a1f122006-03-28 01:56:34 -0800362 int fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (i >= n)
364 break;
365 if (!(bit & all_bits))
366 continue;
Eric Dumazete4a1f122006-03-28 01:56:34 -0800367 file = fget_light(i, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 if (file) {
369 f_op = file->f_op;
370 mask = DEFAULT_POLLMASK;
371 if (f_op && f_op->poll)
372 mask = (*f_op->poll)(file, retval ? NULL : wait);
Eric Dumazete4a1f122006-03-28 01:56:34 -0800373 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 if ((mask & POLLIN_SET) && (in & bit)) {
375 res_in |= bit;
376 retval++;
377 }
378 if ((mask & POLLOUT_SET) && (out & bit)) {
379 res_out |= bit;
380 retval++;
381 }
382 if ((mask & POLLEX_SET) && (ex & bit)) {
383 res_ex |= bit;
384 retval++;
385 }
386 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 }
388 if (res_in)
389 *rinp = res_in;
390 if (res_out)
391 *routp = res_out;
392 if (res_ex)
393 *rexp = res_ex;
Linus Torvalds55d85382008-06-22 12:23:15 -0700394 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
396 wait = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700397 if (retval || timed_out || signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 break;
Pavel Machekf5264482008-04-21 22:15:06 +0000399 if (table.error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 retval = table.error;
401 break;
402 }
David Woodhouse9f729492006-01-18 17:44:05 -0800403
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700404 /*
405 * If this is the first loop and we have a timeout
406 * given, then we convert to ktime_t and set the to
407 * pointer to the expiry value.
408 */
409 if (end_time && !to) {
410 expire = timespec_to_ktime(*end_time);
411 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800412 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700413
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700414 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700415 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
417 __set_current_state(TASK_RUNNING);
418
419 poll_freewait(&table);
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 return retval;
422}
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424/*
425 * We can actually return ERESTARTSYS instead of EINTR, but I'd
426 * like to be certain this leads to no problems. So I return
427 * EINTR just for safety.
428 *
429 * Update: ERESTARTSYS breaks at least the xview clock binary, so
430 * I'm trying ERESTARTNOHAND which restart only when you want to.
431 */
432#define MAX_SELECT_SECONDS \
433 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
434
Al Viroa2dcb442008-04-23 14:05:15 -0400435int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700436 fd_set __user *exp, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 fd_set_bits fds;
Andrew Morton29ff2db2006-04-10 22:52:46 -0700439 void *bits;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800440 int ret, max_fds;
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700441 unsigned int size;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700442 struct fdtable *fdt;
Andi Kleen70674f92006-03-28 01:56:33 -0800443 /* Allocate small arguments on the stack to save memory and be faster */
Jes Sorensen30c14e42006-03-31 11:18:57 -0500444 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 ret = -EINVAL;
447 if (n < 0)
448 goto out_nofds;
449
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800450 /* max_fds can increase, so grab it once to avoid race */
Dipankar Sarmab8359962005-09-09 13:04:14 -0700451 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700452 fdt = files_fdtable(current->files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800453 max_fds = fdt->max_fds;
Dipankar Sarmab8359962005-09-09 13:04:14 -0700454 rcu_read_unlock();
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800455 if (n > max_fds)
456 n = max_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458 /*
459 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
460 * since we used fdset we need to allocate memory in units of
461 * long-words.
462 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 size = FDS_BYTES(n);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700464 bits = stack_fds;
465 if (size > sizeof(stack_fds) / 6) {
466 /* Not enough space in on-stack array; must use kmalloc */
467 ret = -ENOMEM;
Andi Kleen70674f92006-03-28 01:56:33 -0800468 bits = kmalloc(6 * size, GFP_KERNEL);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700469 if (!bits)
470 goto out_nofds;
471 }
Andrew Morton29ff2db2006-04-10 22:52:46 -0700472 fds.in = bits;
473 fds.out = bits + size;
474 fds.ex = bits + 2*size;
475 fds.res_in = bits + 3*size;
476 fds.res_out = bits + 4*size;
477 fds.res_ex = bits + 5*size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 if ((ret = get_fd_set(n, inp, fds.in)) ||
480 (ret = get_fd_set(n, outp, fds.out)) ||
481 (ret = get_fd_set(n, exp, fds.ex)))
482 goto out;
483 zero_fd_set(n, fds.res_in);
484 zero_fd_set(n, fds.res_out);
485 zero_fd_set(n, fds.res_ex);
486
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700487 ret = do_select(n, &fds, end_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 if (ret < 0)
490 goto out;
491 if (!ret) {
492 ret = -ERESTARTNOHAND;
493 if (signal_pending(current))
494 goto out;
495 ret = 0;
496 }
497
498 if (set_fd_set(n, inp, fds.res_in) ||
499 set_fd_set(n, outp, fds.res_out) ||
500 set_fd_set(n, exp, fds.res_ex))
501 ret = -EFAULT;
502
503out:
Andi Kleen70674f92006-03-28 01:56:33 -0800504 if (bits != stack_fds)
505 kfree(bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506out_nofds:
507 return ret;
508}
509
David Woodhouse9f729492006-01-18 17:44:05 -0800510asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
511 fd_set __user *exp, struct timeval __user *tvp)
512{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700513 struct timespec end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800514 struct timeval tv;
515 int ret;
516
517 if (tvp) {
518 if (copy_from_user(&tv, tvp, sizeof(tv)))
519 return -EFAULT;
520
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700521 to = &end_time;
Arjan van de Ven4d36a9e2008-10-25 12:41:41 -0700522 if (poll_select_set_timeout(to,
523 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
524 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
David Woodhouse9f729492006-01-18 17:44:05 -0800525 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800526 }
527
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700528 ret = core_sys_select(n, inp, outp, exp, to);
529 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800530
531 return ret;
532}
533
Roland McGrathf3de2722008-04-30 00:53:09 -0700534#ifdef HAVE_SET_RESTORE_SIGMASK
David Woodhouse9f729492006-01-18 17:44:05 -0800535asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp,
536 fd_set __user *exp, struct timespec __user *tsp,
537 const sigset_t __user *sigmask, size_t sigsetsize)
538{
David Woodhouse9f729492006-01-18 17:44:05 -0800539 sigset_t ksigmask, sigsaved;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700540 struct timespec ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800541 int ret;
542
543 if (tsp) {
544 if (copy_from_user(&ts, tsp, sizeof(ts)))
545 return -EFAULT;
546
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700547 to = &end_time;
548 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
David Woodhouse9f729492006-01-18 17:44:05 -0800549 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800550 }
551
552 if (sigmask) {
553 /* XXX: Don't preclude handling different sized sigset_t's. */
554 if (sigsetsize != sizeof(sigset_t))
555 return -EINVAL;
556 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
557 return -EFAULT;
558
559 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
560 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
561 }
562
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700563 ret = core_sys_select(n, inp, outp, exp, &end_time);
564 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800565
566 if (ret == -ERESTARTNOHAND) {
567 /*
568 * Don't restore the signal mask yet. Let do_signal() deliver
569 * the signal on the way back to userspace, before the signal
570 * mask is restored.
571 */
572 if (sigmask) {
573 memcpy(&current->saved_sigmask, &sigsaved,
574 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -0700575 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -0800576 }
577 } else if (sigmask)
578 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
579
580 return ret;
581}
582
583/*
584 * Most architectures can't handle 7-argument syscalls. So we provide a
585 * 6-argument version where the sixth argument is a pointer to a structure
586 * which has a pointer to the sigset_t itself followed by a size_t containing
587 * the sigset size.
588 */
589asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp,
590 fd_set __user *exp, struct timespec __user *tsp, void __user *sig)
591{
592 size_t sigsetsize = 0;
593 sigset_t __user *up = NULL;
594
595 if (sig) {
596 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
Al Viroe110ab92006-02-01 05:26:09 -0500597 || __get_user(up, (sigset_t __user * __user *)sig)
David Woodhouse9f729492006-01-18 17:44:05 -0800598 || __get_user(sigsetsize,
Al Viroe110ab92006-02-01 05:26:09 -0500599 (size_t __user *)(sig+sizeof(void *))))
David Woodhouse9f729492006-01-18 17:44:05 -0800600 return -EFAULT;
601 }
602
603 return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize);
604}
Roland McGrathf3de2722008-04-30 00:53:09 -0700605#endif /* HAVE_SET_RESTORE_SIGMASK */
David Woodhouse9f729492006-01-18 17:44:05 -0800606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607struct poll_list {
608 struct poll_list *next;
609 int len;
610 struct pollfd entries[0];
611};
612
613#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
614
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700615/*
616 * Fish for pollable events on the pollfd->fd file descriptor. We're only
617 * interested in events matching the pollfd->events mask, and the result
618 * matching that mask is both recorded in pollfd->revents and returned. The
619 * pwait poll_table will be used by the fd-provided poll handler for waiting,
620 * if non-NULL.
621 */
622static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700624 unsigned int mask;
625 int fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700627 mask = 0;
628 fd = pollfd->fd;
629 if (fd >= 0) {
630 int fput_needed;
631 struct file * file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700633 file = fget_light(fd, &fput_needed);
634 mask = POLLNVAL;
635 if (file != NULL) {
636 mask = DEFAULT_POLLMASK;
637 if (file->f_op && file->f_op->poll)
638 mask = file->f_op->poll(file, pwait);
639 /* Mask out unneeded events. */
640 mask &= pollfd->events | POLLERR | POLLHUP;
641 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700644 pollfd->revents = mask;
645
646 return mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649static int do_poll(unsigned int nfds, struct poll_list *list,
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700650 struct poll_wqueues *wait, struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 poll_table* pt = &wait->pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700653 ktime_t expire, *to = NULL;
654 int timed_out = 0, count = 0;
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700655 unsigned long slack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
David Woodhouse9f729492006-01-18 17:44:05 -0800657 /* Optimise the no-wait case */
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700658 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 pt = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700660 timed_out = 1;
661 }
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700662
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700663 if (end_time && !timed_out)
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700664 slack = estimate_accuracy(end_time);
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 for (;;) {
667 struct poll_list *walk;
David Woodhouse9f729492006-01-18 17:44:05 -0800668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 set_current_state(TASK_INTERRUPTIBLE);
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700670 for (walk = list; walk != NULL; walk = walk->next) {
671 struct pollfd * pfd, * pfd_end;
672
673 pfd = walk->entries;
674 pfd_end = pfd + walk->len;
675 for (; pfd != pfd_end; pfd++) {
676 /*
677 * Fish for events. If we found one, record it
678 * and kill the poll_table, so we don't
679 * needlessly register any other waiters after
680 * this. They'll get immediately deregistered
681 * when we break out and return.
682 */
683 if (do_pollfd(pfd, pt)) {
684 count++;
685 pt = NULL;
686 }
687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700689 /*
690 * All waiters have already been registered, so don't provide
691 * a poll_table to them on the next loop iteration.
692 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 pt = NULL;
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700694 if (!count) {
695 count = wait->error;
696 if (signal_pending(current))
697 count = -EINTR;
698 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700699 if (count || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 break;
David Woodhouse9f729492006-01-18 17:44:05 -0800701
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700702 /*
703 * If this is the first loop and we have a timeout
704 * given, then we convert to ktime_t and set the to
705 * pointer to the expiry value.
706 */
707 if (end_time && !to) {
708 expire = timespec_to_ktime(*end_time);
709 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800710 }
711
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700712 if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700713 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 }
715 __set_current_state(TASK_RUNNING);
716 return count;
717}
718
Andi Kleen70674f92006-03-28 01:56:33 -0800719#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
720 sizeof(struct pollfd))
721
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700722int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
723 struct timespec *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724{
725 struct poll_wqueues table;
Oleg Nesterov252e5722007-10-16 23:26:17 -0700726 int err = -EFAULT, fdcount, len, size;
Jes Sorensen30c14e42006-03-31 11:18:57 -0500727 /* Allocate small arguments on the stack to save memory and be
728 faster - use long to make sure the buffer is aligned properly
729 on 64 bit archs to avoid unaligned access */
730 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
Oleg Nesterov252e5722007-10-16 23:26:17 -0700731 struct poll_list *const head = (struct poll_list *)stack_pps;
732 struct poll_list *walk = head;
733 unsigned long todo = nfds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Chris Snook4e6fd332006-09-29 02:01:33 -0700735 if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 return -EINVAL;
737
Oleg Nesterov252e5722007-10-16 23:26:17 -0700738 len = min_t(unsigned int, nfds, N_STACK_PPS);
739 for (;;) {
740 walk->next = NULL;
741 walk->len = len;
742 if (!len)
743 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Oleg Nesterov252e5722007-10-16 23:26:17 -0700745 if (copy_from_user(walk->entries, ufds + nfds-todo,
746 sizeof(struct pollfd) * walk->len))
747 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Oleg Nesterov252e5722007-10-16 23:26:17 -0700749 todo -= walk->len;
750 if (!todo)
751 break;
752
753 len = min(todo, POLLFD_PER_PAGE);
754 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
755 walk = walk->next = kmalloc(size, GFP_KERNEL);
756 if (!walk) {
757 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 goto out_fds;
759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
David Woodhouse9f729492006-01-18 17:44:05 -0800761
Oleg Nesterov252e5722007-10-16 23:26:17 -0700762 poll_initwait(&table);
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700763 fdcount = do_poll(nfds, head, &table, end_time);
Oleg Nesterov252e5722007-10-16 23:26:17 -0700764 poll_freewait(&table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Oleg Nesterov252e5722007-10-16 23:26:17 -0700766 for (walk = head; walk; walk = walk->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 struct pollfd *fds = walk->entries;
768 int j;
769
Oleg Nesterov252e5722007-10-16 23:26:17 -0700770 for (j = 0; j < walk->len; j++, ufds++)
771 if (__put_user(fds[j].revents, &ufds->revents))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 err = fdcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776out_fds:
Oleg Nesterov252e5722007-10-16 23:26:17 -0700777 walk = head->next;
778 while (walk) {
779 struct poll_list *pos = walk;
780 walk = walk->next;
781 kfree(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
Oleg Nesterov252e5722007-10-16 23:26:17 -0700783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 return err;
785}
David Woodhouse9f729492006-01-18 17:44:05 -0800786
Chris Wright3075d9d2007-10-16 23:27:18 -0700787static long do_restart_poll(struct restart_block *restart_block)
788{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700789 struct pollfd __user *ufds = restart_block->poll.ufds;
790 int nfds = restart_block->poll.nfds;
791 struct timespec *to = NULL, end_time;
Chris Wright3075d9d2007-10-16 23:27:18 -0700792 int ret;
793
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700794 if (restart_block->poll.has_timeout) {
795 end_time.tv_sec = restart_block->poll.tv_sec;
796 end_time.tv_nsec = restart_block->poll.tv_nsec;
797 to = &end_time;
798 }
799
800 ret = do_sys_poll(ufds, nfds, to);
801
Chris Wright3075d9d2007-10-16 23:27:18 -0700802 if (ret == -EINTR) {
803 restart_block->fn = do_restart_poll;
Chris Wright3075d9d2007-10-16 23:27:18 -0700804 ret = -ERESTART_RESTARTBLOCK;
805 }
806 return ret;
807}
808
David Woodhouse9f729492006-01-18 17:44:05 -0800809asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds,
810 long timeout_msecs)
811{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700812 struct timespec end_time, *to = NULL;
Chris Wright3075d9d2007-10-16 23:27:18 -0700813 int ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800814
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700815 if (timeout_msecs >= 0) {
816 to = &end_time;
817 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
818 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
David Woodhouse9f729492006-01-18 17:44:05 -0800819 }
820
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700821 ret = do_sys_poll(ufds, nfds, to);
822
Chris Wright3075d9d2007-10-16 23:27:18 -0700823 if (ret == -EINTR) {
824 struct restart_block *restart_block;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700825
Chris Wright3075d9d2007-10-16 23:27:18 -0700826 restart_block = &current_thread_info()->restart_block;
827 restart_block->fn = do_restart_poll;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700828 restart_block->poll.ufds = ufds;
829 restart_block->poll.nfds = nfds;
830
831 if (timeout_msecs >= 0) {
832 restart_block->poll.tv_sec = end_time.tv_sec;
833 restart_block->poll.tv_nsec = end_time.tv_nsec;
834 restart_block->poll.has_timeout = 1;
835 } else
836 restart_block->poll.has_timeout = 0;
837
Chris Wright3075d9d2007-10-16 23:27:18 -0700838 ret = -ERESTART_RESTARTBLOCK;
839 }
840 return ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800841}
842
Roland McGrathf3de2722008-04-30 00:53:09 -0700843#ifdef HAVE_SET_RESTORE_SIGMASK
David Woodhouse9f729492006-01-18 17:44:05 -0800844asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds,
845 struct timespec __user *tsp, const sigset_t __user *sigmask,
846 size_t sigsetsize)
847{
848 sigset_t ksigmask, sigsaved;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700849 struct timespec ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800850 int ret;
851
852 if (tsp) {
853 if (copy_from_user(&ts, tsp, sizeof(ts)))
854 return -EFAULT;
855
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700856 to = &end_time;
857 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
858 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800859 }
860
861 if (sigmask) {
862 /* XXX: Don't preclude handling different sized sigset_t's. */
863 if (sigsetsize != sizeof(sigset_t))
864 return -EINVAL;
865 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
866 return -EFAULT;
867
868 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
869 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
870 }
871
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700872 ret = do_sys_poll(ufds, nfds, to);
David Woodhouse9f729492006-01-18 17:44:05 -0800873
874 /* We can restart this syscall, usually */
875 if (ret == -EINTR) {
876 /*
877 * Don't restore the signal mask yet. Let do_signal() deliver
878 * the signal on the way back to userspace, before the signal
879 * mask is restored.
880 */
881 if (sigmask) {
882 memcpy(&current->saved_sigmask, &sigsaved,
883 sizeof(sigsaved));
Roland McGrath4e4c22c2008-04-30 00:53:06 -0700884 set_restore_sigmask();
David Woodhouse9f729492006-01-18 17:44:05 -0800885 }
886 ret = -ERESTARTNOHAND;
887 } else if (sigmask)
888 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
889
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700890 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800891
892 return ret;
893}
Roland McGrathf3de2722008-04-30 00:53:09 -0700894#endif /* HAVE_SET_RESTORE_SIGMASK */