]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * This file contains the procedures for the handling of select and poll | |
4 | * | |
5 | * Created for Linux based loosely upon Mathius Lattner's minix | |
6 | * patches by Peter MacDonald. Heavily edited by Linus. | |
7 | * | |
8 | * 4 February 1994 | |
9 | * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS | |
10 | * flag set in its personality we do *not* modify the given timeout | |
11 | * parameter to reflect time remaining. | |
12 | * | |
13 | * 24 January 2000 | |
14 | * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation | |
15 | * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian). | |
16 | */ | |
17 | ||
022a1692 | 18 | #include <linux/kernel.h> |
3f07c014 IM |
19 | #include <linux/sched/signal.h> |
20 | #include <linux/sched/rt.h> | |
1da177e4 | 21 | #include <linux/syscalls.h> |
630d9c47 | 22 | #include <linux/export.h> |
1da177e4 | 23 | #include <linux/slab.h> |
1da177e4 LT |
24 | #include <linux/poll.h> |
25 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ | |
26 | #include <linux/file.h> | |
9f3acc31 | 27 | #include <linux/fdtable.h> |
1da177e4 | 28 | #include <linux/fs.h> |
b835996f | 29 | #include <linux/rcupdate.h> |
8ff3e8e8 | 30 | #include <linux/hrtimer.h> |
9745cdb3 | 31 | #include <linux/freezer.h> |
076bb0c8 | 32 | #include <net/busy_poll.h> |
2d19309c | 33 | #include <linux/vmalloc.h> |
1da177e4 | 34 | |
7c0f6ba6 | 35 | #include <linux/uaccess.h> |
1da177e4 | 36 | |
90d6e24a AV |
37 | |
38 | /* | |
39 | * Estimate expected accuracy in ns from a timeval. | |
40 | * | |
41 | * After quite a bit of churning around, we've settled on | |
42 | * a simple thing of taking 0.1% of the timeout as the | |
43 | * slack, with a cap of 100 msec. | |
44 | * "nice" tasks get a 0.5% slack instead. | |
45 | * | |
46 | * Consider this comment an open invitation to come up with even | |
47 | * better solutions.. | |
48 | */ | |
49 | ||
5ae87e79 GK |
50 | #define MAX_SLACK (100 * NSEC_PER_MSEC) |
51 | ||
766b9f92 | 52 | static long __estimate_accuracy(struct timespec64 *tv) |
90d6e24a | 53 | { |
96d2ab48 | 54 | long slack; |
90d6e24a AV |
55 | int divfactor = 1000; |
56 | ||
5ae87e79 GK |
57 | if (tv->tv_sec < 0) |
58 | return 0; | |
59 | ||
4ce105d3 | 60 | if (task_nice(current) > 0) |
90d6e24a AV |
61 | divfactor = divfactor / 5; |
62 | ||
5ae87e79 GK |
63 | if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) |
64 | return MAX_SLACK; | |
65 | ||
90d6e24a AV |
66 | slack = tv->tv_nsec / divfactor; |
67 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); | |
68 | ||
5ae87e79 GK |
69 | if (slack > MAX_SLACK) |
70 | return MAX_SLACK; | |
96d2ab48 | 71 | |
90d6e24a AV |
72 | return slack; |
73 | } | |
74 | ||
766b9f92 | 75 | u64 select_estimate_accuracy(struct timespec64 *tv) |
90d6e24a | 76 | { |
da8b44d5 | 77 | u64 ret; |
766b9f92 | 78 | struct timespec64 now; |
90d6e24a AV |
79 | |
80 | /* | |
81 | * Realtime tasks get a slack of 0 for obvious reasons. | |
82 | */ | |
83 | ||
4ce105d3 | 84 | if (rt_task(current)) |
90d6e24a AV |
85 | return 0; |
86 | ||
766b9f92 DD |
87 | ktime_get_ts64(&now); |
88 | now = timespec64_sub(*tv, now); | |
90d6e24a AV |
89 | ret = __estimate_accuracy(&now); |
90 | if (ret < current->timer_slack_ns) | |
91 | return current->timer_slack_ns; | |
92 | return ret; | |
93 | } | |
94 | ||
95 | ||
96 | ||
1da177e4 LT |
97 | struct poll_table_page { |
98 | struct poll_table_page * next; | |
99 | struct poll_table_entry * entry; | |
100 | struct poll_table_entry entries[0]; | |
101 | }; | |
102 | ||
103 | #define POLL_TABLE_FULL(table) \ | |
104 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
105 | ||
106 | /* | |
107 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
108 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
109 | * follow, but it should be free of race-conditions, and it's practical. If you | |
110 | * understand what I'm doing here, then you understand how the linux | |
111 | * sleep/wakeup mechanism works. | |
112 | * | |
113 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
114 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
115 | * as all select/poll functions have to call it to add an entry to the | |
116 | * poll table. | |
117 | */ | |
75c96f85 AB |
118 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
119 | poll_table *p); | |
1da177e4 LT |
120 | |
121 | void poll_initwait(struct poll_wqueues *pwq) | |
122 | { | |
123 | init_poll_funcptr(&pwq->pt, __pollwait); | |
5f820f64 | 124 | pwq->polling_task = current; |
b2add73d | 125 | pwq->triggered = 0; |
1da177e4 LT |
126 | pwq->error = 0; |
127 | pwq->table = NULL; | |
70674f95 | 128 | pwq->inline_index = 0; |
1da177e4 | 129 | } |
1da177e4 LT |
130 | EXPORT_SYMBOL(poll_initwait); |
131 | ||
70674f95 AK |
132 | static void free_poll_entry(struct poll_table_entry *entry) |
133 | { | |
ccf6780d | 134 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
135 | fput(entry->filp); |
136 | } | |
137 | ||
1da177e4 LT |
138 | void poll_freewait(struct poll_wqueues *pwq) |
139 | { | |
140 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
141 | int i; |
142 | for (i = 0; i < pwq->inline_index; i++) | |
143 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
144 | while (p) { |
145 | struct poll_table_entry * entry; | |
146 | struct poll_table_page *old; | |
147 | ||
148 | entry = p->entry; | |
149 | do { | |
150 | entry--; | |
70674f95 | 151 | free_poll_entry(entry); |
1da177e4 LT |
152 | } while (entry > p->entries); |
153 | old = p; | |
154 | p = p->next; | |
155 | free_page((unsigned long) old); | |
156 | } | |
157 | } | |
1da177e4 LT |
158 | EXPORT_SYMBOL(poll_freewait); |
159 | ||
5f820f64 | 160 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
1da177e4 | 161 | { |
1da177e4 LT |
162 | struct poll_table_page *table = p->table; |
163 | ||
70674f95 AK |
164 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
165 | return p->inline_entries + p->inline_index++; | |
166 | ||
1da177e4 LT |
167 | if (!table || POLL_TABLE_FULL(table)) { |
168 | struct poll_table_page *new_table; | |
169 | ||
170 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
171 | if (!new_table) { | |
172 | p->error = -ENOMEM; | |
70674f95 | 173 | return NULL; |
1da177e4 LT |
174 | } |
175 | new_table->entry = new_table->entries; | |
176 | new_table->next = table; | |
177 | p->table = new_table; | |
178 | table = new_table; | |
179 | } | |
180 | ||
70674f95 AK |
181 | return table->entry++; |
182 | } | |
183 | ||
ac6424b9 | 184 | static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
5f820f64 TH |
185 | { |
186 | struct poll_wqueues *pwq = wait->private; | |
187 | DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); | |
188 | ||
189 | /* | |
190 | * Although this function is called under waitqueue lock, LOCK | |
191 | * doesn't imply write barrier and the users expect write | |
192 | * barrier semantics on wakeup functions. The following | |
193 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
b92b8b35 | 194 | * and is paired with smp_store_mb() in poll_schedule_timeout. |
5f820f64 TH |
195 | */ |
196 | smp_wmb(); | |
197 | pwq->triggered = 1; | |
198 | ||
199 | /* | |
200 | * Perform the default wake up operation using a dummy | |
201 | * waitqueue. | |
202 | * | |
203 | * TODO: This is hacky but there currently is no interface to | |
204 | * pass in @sync. @sync is scheduled to be removed and once | |
205 | * that happens, wake_up_process() can be used directly. | |
206 | */ | |
207 | return default_wake_function(&dummy_wait, mode, sync, key); | |
208 | } | |
209 | ||
ac6424b9 | 210 | static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
4938d7e0 ED |
211 | { |
212 | struct poll_table_entry *entry; | |
213 | ||
214 | entry = container_of(wait, struct poll_table_entry, wait); | |
3ad6f93e | 215 | if (key && !(key_to_poll(key) & entry->key)) |
4938d7e0 ED |
216 | return 0; |
217 | return __pollwake(wait, mode, sync, key); | |
218 | } | |
219 | ||
70674f95 AK |
220 | /* Add a new entry */ |
221 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
222 | poll_table *p) | |
223 | { | |
5f820f64 TH |
224 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
225 | struct poll_table_entry *entry = poll_get_entry(pwq); | |
70674f95 AK |
226 | if (!entry) |
227 | return; | |
cb0942b8 | 228 | entry->filp = get_file(filp); |
70674f95 | 229 | entry->wait_address = wait_address; |
626cf236 | 230 | entry->key = p->_key; |
5f820f64 TH |
231 | init_waitqueue_func_entry(&entry->wait, pollwake); |
232 | entry->wait.private = pwq; | |
ccf6780d | 233 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
234 | } |
235 | ||
8f546ae1 | 236 | static int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
5f820f64 TH |
237 | ktime_t *expires, unsigned long slack) |
238 | { | |
239 | int rc = -EINTR; | |
240 | ||
241 | set_current_state(state); | |
242 | if (!pwq->triggered) | |
59612d18 | 243 | rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); |
5f820f64 TH |
244 | __set_current_state(TASK_RUNNING); |
245 | ||
246 | /* | |
247 | * Prepare for the next iteration. | |
248 | * | |
b92b8b35 | 249 | * The following smp_store_mb() serves two purposes. First, it's |
5f820f64 TH |
250 | * the counterpart rmb of the wmb in pollwake() such that data |
251 | * written before wake up is always visible after wake up. | |
252 | * Second, the full barrier guarantees that triggered clearing | |
253 | * doesn't pass event check of the next iteration. Note that | |
254 | * this problem doesn't exist for the first iteration as | |
255 | * add_wait_queue() has full barrier semantics. | |
256 | */ | |
b92b8b35 | 257 | smp_store_mb(pwq->triggered, 0); |
5f820f64 TH |
258 | |
259 | return rc; | |
260 | } | |
5f820f64 | 261 | |
b773ad40 TG |
262 | /** |
263 | * poll_select_set_timeout - helper function to setup the timeout value | |
766b9f92 | 264 | * @to: pointer to timespec64 variable for the final timeout |
b773ad40 TG |
265 | * @sec: seconds (from user space) |
266 | * @nsec: nanoseconds (from user space) | |
267 | * | |
268 | * Note, we do not use a timespec for the user space value here, That | |
269 | * way we can use the function for timeval and compat interfaces as well. | |
270 | * | |
271 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
272 | */ | |
766b9f92 | 273 | int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec) |
b773ad40 | 274 | { |
766b9f92 | 275 | struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec}; |
b773ad40 | 276 | |
766b9f92 | 277 | if (!timespec64_valid(&ts)) |
b773ad40 TG |
278 | return -EINVAL; |
279 | ||
280 | /* Optimize for the zero timeout value here */ | |
281 | if (!sec && !nsec) { | |
282 | to->tv_sec = to->tv_nsec = 0; | |
283 | } else { | |
766b9f92 DD |
284 | ktime_get_ts64(to); |
285 | *to = timespec64_add_safe(*to, ts); | |
b773ad40 TG |
286 | } |
287 | return 0; | |
288 | } | |
289 | ||
8bd27a30 DD |
290 | enum poll_time_type { |
291 | PT_TIMEVAL = 0, | |
292 | PT_OLD_TIMEVAL = 1, | |
293 | PT_TIMESPEC = 2, | |
294 | PT_OLD_TIMESPEC = 3, | |
295 | }; | |
296 | ||
766b9f92 DD |
297 | static int poll_select_copy_remaining(struct timespec64 *end_time, |
298 | void __user *p, | |
8bd27a30 | 299 | enum poll_time_type pt_type, int ret) |
b773ad40 | 300 | { |
36819ad0 | 301 | struct timespec64 rts; |
b773ad40 TG |
302 | |
303 | if (!p) | |
304 | return ret; | |
305 | ||
306 | if (current->personality & STICKY_TIMEOUTS) | |
307 | goto sticky; | |
308 | ||
309 | /* No update for zero timeout */ | |
310 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
311 | return ret; | |
312 | ||
36819ad0 DD |
313 | ktime_get_ts64(&rts); |
314 | rts = timespec64_sub(*end_time, rts); | |
315 | if (rts.tv_sec < 0) | |
316 | rts.tv_sec = rts.tv_nsec = 0; | |
766b9f92 | 317 | |
b773ad40 | 318 | |
8bd27a30 DD |
319 | switch (pt_type) { |
320 | case PT_TIMEVAL: | |
321 | { | |
322 | struct timeval rtv; | |
b773ad40 | 323 | |
8bd27a30 DD |
324 | if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) |
325 | memset(&rtv, 0, sizeof(rtv)); | |
326 | rtv.tv_sec = rts.tv_sec; | |
327 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
328 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
329 | return ret; | |
330 | } | |
331 | break; | |
332 | case PT_OLD_TIMEVAL: | |
333 | { | |
334 | struct old_timeval32 rtv; | |
335 | ||
336 | rtv.tv_sec = rts.tv_sec; | |
337 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
338 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
339 | return ret; | |
340 | } | |
341 | break; | |
342 | case PT_TIMESPEC: | |
343 | if (!put_timespec64(&rts, p)) | |
b773ad40 | 344 | return ret; |
8bd27a30 DD |
345 | break; |
346 | case PT_OLD_TIMESPEC: | |
347 | if (!put_old_timespec32(&rts, p)) | |
348 | return ret; | |
349 | break; | |
350 | default: | |
351 | BUG(); | |
352 | } | |
b773ad40 TG |
353 | /* |
354 | * If an application puts its timeval in read-only memory, we | |
355 | * don't want the Linux-specific update to the timeval to | |
356 | * cause a fault after the select has completed | |
357 | * successfully. However, because we're not updating the | |
358 | * timeval, we can't restart the system call. | |
359 | */ | |
360 | ||
361 | sticky: | |
362 | if (ret == -ERESTARTNOHAND) | |
363 | ret = -EINTR; | |
364 | return ret; | |
365 | } | |
366 | ||
e99ca56c AV |
367 | /* |
368 | * Scalable version of the fd_set. | |
369 | */ | |
370 | ||
371 | typedef struct { | |
372 | unsigned long *in, *out, *ex; | |
373 | unsigned long *res_in, *res_out, *res_ex; | |
374 | } fd_set_bits; | |
375 | ||
376 | /* | |
377 | * How many longwords for "nr" bits? | |
378 | */ | |
379 | #define FDS_BITPERLONG (8*sizeof(long)) | |
380 | #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG) | |
381 | #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long)) | |
382 | ||
383 | /* | |
384 | * We do a VERIFY_WRITE here even though we are only reading this time: | |
385 | * we'll write to it eventually.. | |
386 | * | |
387 | * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned. | |
388 | */ | |
389 | static inline | |
390 | int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) | |
391 | { | |
392 | nr = FDS_BYTES(nr); | |
393 | if (ufdset) | |
394 | return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0; | |
395 | ||
396 | memset(fdset, 0, nr); | |
397 | return 0; | |
398 | } | |
399 | ||
400 | static inline unsigned long __must_check | |
401 | set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) | |
402 | { | |
403 | if (ufdset) | |
404 | return __copy_to_user(ufdset, fdset, FDS_BYTES(nr)); | |
405 | return 0; | |
406 | } | |
407 | ||
408 | static inline | |
409 | void zero_fd_set(unsigned long nr, unsigned long *fdset) | |
410 | { | |
411 | memset(fdset, 0, FDS_BYTES(nr)); | |
412 | } | |
413 | ||
1da177e4 LT |
414 | #define FDS_IN(fds, n) (fds->in + n) |
415 | #define FDS_OUT(fds, n) (fds->out + n) | |
416 | #define FDS_EX(fds, n) (fds->ex + n) | |
417 | ||
418 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
419 | ||
420 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
421 | { | |
422 | unsigned long *open_fds; | |
423 | unsigned long set; | |
424 | int max; | |
badf1662 | 425 | struct fdtable *fdt; |
1da177e4 LT |
426 | |
427 | /* handle last in-complete long-word first */ | |
8ded2bbc JB |
428 | set = ~(~0UL << (n & (BITS_PER_LONG-1))); |
429 | n /= BITS_PER_LONG; | |
badf1662 | 430 | fdt = files_fdtable(current->files); |
1fd36adc | 431 | open_fds = fdt->open_fds + n; |
1da177e4 LT |
432 | max = 0; |
433 | if (set) { | |
434 | set &= BITS(fds, n); | |
435 | if (set) { | |
436 | if (!(set & ~*open_fds)) | |
437 | goto get_max; | |
438 | return -EBADF; | |
439 | } | |
440 | } | |
441 | while (n) { | |
442 | open_fds--; | |
443 | n--; | |
444 | set = BITS(fds, n); | |
445 | if (!set) | |
446 | continue; | |
447 | if (set & ~*open_fds) | |
448 | return -EBADF; | |
449 | if (max) | |
450 | continue; | |
451 | get_max: | |
452 | do { | |
453 | max++; | |
454 | set >>= 1; | |
455 | } while (set); | |
8ded2bbc | 456 | max += n * BITS_PER_LONG; |
1da177e4 LT |
457 | } |
458 | ||
459 | return max; | |
460 | } | |
461 | ||
a9a08845 LT |
462 | #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR) |
463 | #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR) | |
464 | #define POLLEX_SET (EPOLLPRI) | |
1da177e4 | 465 | |
4938d7e0 | 466 | static inline void wait_key_set(poll_table *wait, unsigned long in, |
2d48d67f | 467 | unsigned long out, unsigned long bit, |
01699437 | 468 | __poll_t ll_flag) |
4938d7e0 | 469 | { |
2d48d67f | 470 | wait->_key = POLLEX_SET | ll_flag; |
626cf236 HV |
471 | if (in & bit) |
472 | wait->_key |= POLLIN_SET; | |
473 | if (out & bit) | |
474 | wait->_key |= POLLOUT_SET; | |
4938d7e0 ED |
475 | } |
476 | ||
e99ca56c | 477 | static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time) |
1da177e4 | 478 | { |
8ff3e8e8 | 479 | ktime_t expire, *to = NULL; |
1da177e4 LT |
480 | struct poll_wqueues table; |
481 | poll_table *wait; | |
8ff3e8e8 | 482 | int retval, i, timed_out = 0; |
da8b44d5 | 483 | u64 slack = 0; |
01699437 | 484 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 485 | unsigned long busy_start = 0; |
1da177e4 | 486 | |
b835996f | 487 | rcu_read_lock(); |
1da177e4 | 488 | retval = max_select_fd(n, fds); |
b835996f | 489 | rcu_read_unlock(); |
1da177e4 LT |
490 | |
491 | if (retval < 0) | |
492 | return retval; | |
493 | n = retval; | |
494 | ||
495 | poll_initwait(&table); | |
496 | wait = &table.pt; | |
8ff3e8e8 | 497 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 498 | wait->_qproc = NULL; |
8ff3e8e8 AV |
499 | timed_out = 1; |
500 | } | |
501 | ||
96d2ab48 | 502 | if (end_time && !timed_out) |
231f3d39 | 503 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 504 | |
1da177e4 LT |
505 | retval = 0; |
506 | for (;;) { | |
507 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
cbf55001 | 508 | bool can_busy_loop = false; |
1da177e4 | 509 | |
1da177e4 LT |
510 | inp = fds->in; outp = fds->out; exp = fds->ex; |
511 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
512 | ||
513 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
e6c8adca | 514 | unsigned long in, out, ex, all_bits, bit = 1, j; |
1da177e4 | 515 | unsigned long res_in = 0, res_out = 0, res_ex = 0; |
e6c8adca | 516 | __poll_t mask; |
1da177e4 LT |
517 | |
518 | in = *inp++; out = *outp++; ex = *exp++; | |
519 | all_bits = in | out | ex; | |
520 | if (all_bits == 0) { | |
8ded2bbc | 521 | i += BITS_PER_LONG; |
1da177e4 LT |
522 | continue; |
523 | } | |
524 | ||
8ded2bbc | 525 | for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { |
2903ff01 | 526 | struct fd f; |
1da177e4 LT |
527 | if (i >= n) |
528 | break; | |
529 | if (!(bit & all_bits)) | |
530 | continue; | |
2903ff01 AV |
531 | f = fdget(i); |
532 | if (f.file) { | |
9965ed17 CH |
533 | wait_key_set(wait, in, out, bit, |
534 | busy_flag); | |
535 | mask = vfs_poll(f.file, wait); | |
536 | ||
2903ff01 | 537 | fdput(f); |
1da177e4 LT |
538 | if ((mask & POLLIN_SET) && (in & bit)) { |
539 | res_in |= bit; | |
540 | retval++; | |
626cf236 | 541 | wait->_qproc = NULL; |
1da177e4 LT |
542 | } |
543 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
544 | res_out |= bit; | |
545 | retval++; | |
626cf236 | 546 | wait->_qproc = NULL; |
1da177e4 LT |
547 | } |
548 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
549 | res_ex |= bit; | |
550 | retval++; | |
626cf236 | 551 | wait->_qproc = NULL; |
1da177e4 | 552 | } |
2d48d67f | 553 | /* got something, stop busy polling */ |
cbf55001 ET |
554 | if (retval) { |
555 | can_busy_loop = false; | |
556 | busy_flag = 0; | |
557 | ||
558 | /* | |
559 | * only remember a returned | |
560 | * POLL_BUSY_LOOP if we asked for it | |
561 | */ | |
562 | } else if (busy_flag & mask) | |
563 | can_busy_loop = true; | |
564 | ||
1da177e4 | 565 | } |
1da177e4 LT |
566 | } |
567 | if (res_in) | |
568 | *rinp = res_in; | |
569 | if (res_out) | |
570 | *routp = res_out; | |
571 | if (res_ex) | |
572 | *rexp = res_ex; | |
55d85384 | 573 | cond_resched(); |
1da177e4 | 574 | } |
626cf236 | 575 | wait->_qproc = NULL; |
8ff3e8e8 | 576 | if (retval || timed_out || signal_pending(current)) |
1da177e4 | 577 | break; |
f5264481 | 578 | if (table.error) { |
1da177e4 LT |
579 | retval = table.error; |
580 | break; | |
581 | } | |
9f72949f | 582 | |
cbf55001 | 583 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 | 584 | if (can_busy_loop && !need_resched()) { |
37056719 AD |
585 | if (!busy_start) { |
586 | busy_start = busy_loop_current_time(); | |
76b1e9b9 ET |
587 | continue; |
588 | } | |
37056719 | 589 | if (!busy_loop_timeout(busy_start)) |
76b1e9b9 ET |
590 | continue; |
591 | } | |
592 | busy_flag = 0; | |
2d48d67f | 593 | |
8ff3e8e8 AV |
594 | /* |
595 | * If this is the first loop and we have a timeout | |
596 | * given, then we convert to ktime_t and set the to | |
597 | * pointer to the expiry value. | |
598 | */ | |
599 | if (end_time && !to) { | |
766b9f92 | 600 | expire = timespec64_to_ktime(*end_time); |
8ff3e8e8 | 601 | to = &expire; |
9f72949f | 602 | } |
8ff3e8e8 | 603 | |
5f820f64 TH |
604 | if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE, |
605 | to, slack)) | |
8ff3e8e8 | 606 | timed_out = 1; |
1da177e4 | 607 | } |
1da177e4 LT |
608 | |
609 | poll_freewait(&table); | |
610 | ||
1da177e4 LT |
611 | return retval; |
612 | } | |
613 | ||
1da177e4 LT |
614 | /* |
615 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
616 | * like to be certain this leads to no problems. So I return | |
617 | * EINTR just for safety. | |
618 | * | |
619 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
620 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
621 | */ | |
a2dcb44c | 622 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
766b9f92 | 623 | fd_set __user *exp, struct timespec64 *end_time) |
1da177e4 LT |
624 | { |
625 | fd_set_bits fds; | |
29ff2db5 | 626 | void *bits; |
bbea9f69 | 627 | int ret, max_fds; |
2d19309c | 628 | size_t size, alloc_size; |
badf1662 | 629 | struct fdtable *fdt; |
70674f95 | 630 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 631 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 632 | |
1da177e4 LT |
633 | ret = -EINVAL; |
634 | if (n < 0) | |
635 | goto out_nofds; | |
636 | ||
bbea9f69 | 637 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 638 | rcu_read_lock(); |
badf1662 | 639 | fdt = files_fdtable(current->files); |
bbea9f69 | 640 | max_fds = fdt->max_fds; |
b835996f | 641 | rcu_read_unlock(); |
bbea9f69 VL |
642 | if (n > max_fds) |
643 | n = max_fds; | |
1da177e4 LT |
644 | |
645 | /* | |
646 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
647 | * since we used fdset we need to allocate memory in units of | |
648 | * long-words. | |
649 | */ | |
1da177e4 | 650 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
651 | bits = stack_fds; |
652 | if (size > sizeof(stack_fds) / 6) { | |
653 | /* Not enough space in on-stack array; must use kmalloc */ | |
654 | ret = -ENOMEM; | |
2d19309c VB |
655 | if (size > (SIZE_MAX / 6)) |
656 | goto out_nofds; | |
657 | ||
658 | alloc_size = 6 * size; | |
752ade68 | 659 | bits = kvmalloc(alloc_size, GFP_KERNEL); |
b04eb6aa MBJ |
660 | if (!bits) |
661 | goto out_nofds; | |
662 | } | |
29ff2db5 AM |
663 | fds.in = bits; |
664 | fds.out = bits + size; | |
665 | fds.ex = bits + 2*size; | |
666 | fds.res_in = bits + 3*size; | |
667 | fds.res_out = bits + 4*size; | |
668 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
669 | |
670 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
671 | (ret = get_fd_set(n, outp, fds.out)) || | |
672 | (ret = get_fd_set(n, exp, fds.ex))) | |
673 | goto out; | |
674 | zero_fd_set(n, fds.res_in); | |
675 | zero_fd_set(n, fds.res_out); | |
676 | zero_fd_set(n, fds.res_ex); | |
677 | ||
8ff3e8e8 | 678 | ret = do_select(n, &fds, end_time); |
1da177e4 LT |
679 | |
680 | if (ret < 0) | |
681 | goto out; | |
682 | if (!ret) { | |
683 | ret = -ERESTARTNOHAND; | |
684 | if (signal_pending(current)) | |
685 | goto out; | |
686 | ret = 0; | |
687 | } | |
688 | ||
689 | if (set_fd_set(n, inp, fds.res_in) || | |
690 | set_fd_set(n, outp, fds.res_out) || | |
691 | set_fd_set(n, exp, fds.res_ex)) | |
692 | ret = -EFAULT; | |
693 | ||
694 | out: | |
70674f95 | 695 | if (bits != stack_fds) |
2d19309c | 696 | kvfree(bits); |
1da177e4 LT |
697 | out_nofds: |
698 | return ret; | |
699 | } | |
700 | ||
4bdb9aca DB |
701 | static int kern_select(int n, fd_set __user *inp, fd_set __user *outp, |
702 | fd_set __user *exp, struct timeval __user *tvp) | |
9f72949f | 703 | { |
766b9f92 | 704 | struct timespec64 end_time, *to = NULL; |
9f72949f DW |
705 | struct timeval tv; |
706 | int ret; | |
707 | ||
708 | if (tvp) { | |
709 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
710 | return -EFAULT; | |
711 | ||
8ff3e8e8 | 712 | to = &end_time; |
4d36a9e6 AV |
713 | if (poll_select_set_timeout(to, |
714 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
715 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
9f72949f | 716 | return -EINVAL; |
9f72949f DW |
717 | } |
718 | ||
8ff3e8e8 | 719 | ret = core_sys_select(n, inp, outp, exp, to); |
8bd27a30 | 720 | ret = poll_select_copy_remaining(&end_time, tvp, PT_TIMEVAL, ret); |
9f72949f DW |
721 | |
722 | return ret; | |
723 | } | |
724 | ||
4bdb9aca DB |
725 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
726 | fd_set __user *, exp, struct timeval __user *, tvp) | |
727 | { | |
728 | return kern_select(n, inp, outp, exp, tvp); | |
729 | } | |
730 | ||
c9da9f21 | 731 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
e024707b DD |
732 | fd_set __user *exp, void __user *tsp, |
733 | const sigset_t __user *sigmask, size_t sigsetsize, | |
734 | enum poll_time_type type) | |
9f72949f | 735 | { |
9f72949f | 736 | sigset_t ksigmask, sigsaved; |
36819ad0 | 737 | struct timespec64 ts, end_time, *to = NULL; |
9f72949f DW |
738 | int ret; |
739 | ||
740 | if (tsp) { | |
e024707b DD |
741 | switch (type) { |
742 | case PT_TIMESPEC: | |
743 | if (get_timespec64(&ts, tsp)) | |
744 | return -EFAULT; | |
745 | break; | |
746 | case PT_OLD_TIMESPEC: | |
747 | if (get_old_timespec32(&ts, tsp)) | |
748 | return -EFAULT; | |
749 | break; | |
750 | default: | |
751 | BUG(); | |
752 | } | |
9f72949f | 753 | |
8ff3e8e8 | 754 | to = &end_time; |
36819ad0 | 755 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) |
9f72949f | 756 | return -EINVAL; |
9f72949f DW |
757 | } |
758 | ||
ded653cc DD |
759 | ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); |
760 | if (ret) | |
761 | return ret; | |
9f72949f | 762 | |
62568510 | 763 | ret = core_sys_select(n, inp, outp, exp, to); |
e024707b | 764 | ret = poll_select_copy_remaining(&end_time, tsp, type, ret); |
9f72949f | 765 | |
854a6ed5 | 766 | restore_user_sigmask(sigmask, &sigsaved); |
9f72949f DW |
767 | |
768 | return ret; | |
769 | } | |
770 | ||
771 | /* | |
772 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
773 | * 6-argument version where the sixth argument is a pointer to a structure | |
774 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
775 | * the sigset size. | |
776 | */ | |
d4e82042 | 777 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
e024707b DD |
778 | fd_set __user *, exp, struct __kernel_timespec __user *, tsp, |
779 | void __user *, sig) | |
780 | { | |
781 | size_t sigsetsize = 0; | |
782 | sigset_t __user *up = NULL; | |
783 | ||
784 | if (sig) { | |
785 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
786 | || __get_user(up, (sigset_t __user * __user *)sig) | |
787 | || __get_user(sigsetsize, | |
788 | (size_t __user *)(sig+sizeof(void *)))) | |
789 | return -EFAULT; | |
790 | } | |
791 | ||
792 | return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_TIMESPEC); | |
793 | } | |
794 | ||
795 | #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) | |
796 | ||
797 | SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp, | |
798 | fd_set __user *, exp, struct old_timespec32 __user *, tsp, | |
d4e82042 | 799 | void __user *, sig) |
9f72949f DW |
800 | { |
801 | size_t sigsetsize = 0; | |
802 | sigset_t __user *up = NULL; | |
803 | ||
804 | if (sig) { | |
805 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
e110ab94 | 806 | || __get_user(up, (sigset_t __user * __user *)sig) |
9f72949f | 807 | || __get_user(sigsetsize, |
e110ab94 | 808 | (size_t __user *)(sig+sizeof(void *)))) |
9f72949f DW |
809 | return -EFAULT; |
810 | } | |
811 | ||
e024707b | 812 | return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_OLD_TIMESPEC); |
9f72949f | 813 | } |
9f72949f | 814 | |
e024707b DD |
815 | #endif |
816 | ||
5d0e5283 CH |
817 | #ifdef __ARCH_WANT_SYS_OLD_SELECT |
818 | struct sel_arg_struct { | |
819 | unsigned long n; | |
820 | fd_set __user *inp, *outp, *exp; | |
821 | struct timeval __user *tvp; | |
822 | }; | |
823 | ||
824 | SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) | |
825 | { | |
826 | struct sel_arg_struct a; | |
827 | ||
828 | if (copy_from_user(&a, arg, sizeof(a))) | |
829 | return -EFAULT; | |
4bdb9aca | 830 | return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp); |
5d0e5283 CH |
831 | } |
832 | #endif | |
833 | ||
1da177e4 LT |
834 | struct poll_list { |
835 | struct poll_list *next; | |
836 | int len; | |
837 | struct pollfd entries[0]; | |
838 | }; | |
839 | ||
840 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
841 | ||
4a4b69f7 VL |
842 | /* |
843 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
844 | * interested in events matching the pollfd->events mask, and the result | |
845 | * matching that mask is both recorded in pollfd->revents and returned. The | |
846 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
626cf236 | 847 | * if pwait->_qproc is non-NULL. |
4a4b69f7 | 848 | */ |
fb367937 | 849 | static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait, |
cbf55001 | 850 | bool *can_busy_poll, |
fb367937 | 851 | __poll_t busy_flag) |
1da177e4 | 852 | { |
a0f8dcfc CH |
853 | int fd = pollfd->fd; |
854 | __poll_t mask = 0, filter; | |
855 | struct fd f; | |
856 | ||
857 | if (fd < 0) | |
858 | goto out; | |
859 | mask = EPOLLNVAL; | |
860 | f = fdget(fd); | |
861 | if (!f.file) | |
862 | goto out; | |
863 | ||
864 | /* userland u16 ->events contains POLL... bitmap */ | |
865 | filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP; | |
9965ed17 CH |
866 | pwait->_key = filter | busy_flag; |
867 | mask = vfs_poll(f.file, pwait); | |
868 | if (mask & busy_flag) | |
869 | *can_busy_poll = true; | |
a0f8dcfc CH |
870 | mask &= filter; /* Mask out unneeded events. */ |
871 | fdput(f); | |
872 | ||
873 | out: | |
fb367937 | 874 | /* ... and so does ->revents */ |
c71d227f | 875 | pollfd->revents = mangle_poll(mask); |
4a4b69f7 | 876 | return mask; |
1da177e4 LT |
877 | } |
878 | ||
ccec5ee3 | 879 | static int do_poll(struct poll_list *list, struct poll_wqueues *wait, |
766b9f92 | 880 | struct timespec64 *end_time) |
1da177e4 | 881 | { |
1da177e4 | 882 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
883 | ktime_t expire, *to = NULL; |
884 | int timed_out = 0, count = 0; | |
da8b44d5 | 885 | u64 slack = 0; |
fb367937 | 886 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 887 | unsigned long busy_start = 0; |
1da177e4 | 888 | |
9f72949f | 889 | /* Optimise the no-wait case */ |
8ff3e8e8 | 890 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 891 | pt->_qproc = NULL; |
8ff3e8e8 AV |
892 | timed_out = 1; |
893 | } | |
9bf084f7 | 894 | |
96d2ab48 | 895 | if (end_time && !timed_out) |
231f3d39 | 896 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 897 | |
1da177e4 LT |
898 | for (;;) { |
899 | struct poll_list *walk; | |
cbf55001 | 900 | bool can_busy_loop = false; |
9f72949f | 901 | |
4a4b69f7 VL |
902 | for (walk = list; walk != NULL; walk = walk->next) { |
903 | struct pollfd * pfd, * pfd_end; | |
904 | ||
905 | pfd = walk->entries; | |
906 | pfd_end = pfd + walk->len; | |
907 | for (; pfd != pfd_end; pfd++) { | |
908 | /* | |
909 | * Fish for events. If we found one, record it | |
626cf236 | 910 | * and kill poll_table->_qproc, so we don't |
4a4b69f7 VL |
911 | * needlessly register any other waiters after |
912 | * this. They'll get immediately deregistered | |
913 | * when we break out and return. | |
914 | */ | |
cbf55001 ET |
915 | if (do_pollfd(pfd, pt, &can_busy_loop, |
916 | busy_flag)) { | |
4a4b69f7 | 917 | count++; |
626cf236 | 918 | pt->_qproc = NULL; |
cbf55001 ET |
919 | /* found something, stop busy polling */ |
920 | busy_flag = 0; | |
921 | can_busy_loop = false; | |
4a4b69f7 VL |
922 | } |
923 | } | |
1da177e4 | 924 | } |
4a4b69f7 VL |
925 | /* |
926 | * All waiters have already been registered, so don't provide | |
626cf236 | 927 | * a poll_table->_qproc to them on the next loop iteration. |
4a4b69f7 | 928 | */ |
626cf236 | 929 | pt->_qproc = NULL; |
9bf084f7 ON |
930 | if (!count) { |
931 | count = wait->error; | |
932 | if (signal_pending(current)) | |
933 | count = -EINTR; | |
934 | } | |
8ff3e8e8 | 935 | if (count || timed_out) |
1da177e4 | 936 | break; |
9f72949f | 937 | |
cbf55001 | 938 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 | 939 | if (can_busy_loop && !need_resched()) { |
37056719 AD |
940 | if (!busy_start) { |
941 | busy_start = busy_loop_current_time(); | |
76b1e9b9 ET |
942 | continue; |
943 | } | |
37056719 | 944 | if (!busy_loop_timeout(busy_start)) |
76b1e9b9 ET |
945 | continue; |
946 | } | |
947 | busy_flag = 0; | |
91e2fd33 | 948 | |
8ff3e8e8 AV |
949 | /* |
950 | * If this is the first loop and we have a timeout | |
951 | * given, then we convert to ktime_t and set the to | |
952 | * pointer to the expiry value. | |
953 | */ | |
954 | if (end_time && !to) { | |
766b9f92 | 955 | expire = timespec64_to_ktime(*end_time); |
8ff3e8e8 | 956 | to = &expire; |
9f72949f DW |
957 | } |
958 | ||
5f820f64 | 959 | if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) |
8ff3e8e8 | 960 | timed_out = 1; |
1da177e4 | 961 | } |
1da177e4 LT |
962 | return count; |
963 | } | |
964 | ||
70674f95 AK |
965 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
966 | sizeof(struct pollfd)) | |
967 | ||
e99ca56c | 968 | static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
766b9f92 | 969 | struct timespec64 *end_time) |
1da177e4 LT |
970 | { |
971 | struct poll_wqueues table; | |
252e5725 | 972 | int err = -EFAULT, fdcount, len, size; |
30c14e40 JS |
973 | /* Allocate small arguments on the stack to save memory and be |
974 | faster - use long to make sure the buffer is aligned properly | |
975 | on 64 bit archs to avoid unaligned access */ | |
976 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
977 | struct poll_list *const head = (struct poll_list *)stack_pps; |
978 | struct poll_list *walk = head; | |
979 | unsigned long todo = nfds; | |
1da177e4 | 980 | |
d554ed89 | 981 | if (nfds > rlimit(RLIMIT_NOFILE)) |
1da177e4 LT |
982 | return -EINVAL; |
983 | ||
252e5725 ON |
984 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
985 | for (;;) { | |
986 | walk->next = NULL; | |
987 | walk->len = len; | |
988 | if (!len) | |
989 | break; | |
1da177e4 | 990 | |
252e5725 ON |
991 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
992 | sizeof(struct pollfd) * walk->len)) | |
993 | goto out_fds; | |
994 | ||
995 | todo -= walk->len; | |
996 | if (!todo) | |
997 | break; | |
1da177e4 | 998 | |
252e5725 ON |
999 | len = min(todo, POLLFD_PER_PAGE); |
1000 | size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; | |
1001 | walk = walk->next = kmalloc(size, GFP_KERNEL); | |
1002 | if (!walk) { | |
1003 | err = -ENOMEM; | |
1da177e4 LT |
1004 | goto out_fds; |
1005 | } | |
1da177e4 | 1006 | } |
9f72949f | 1007 | |
252e5725 | 1008 | poll_initwait(&table); |
ccec5ee3 | 1009 | fdcount = do_poll(head, &table, end_time); |
252e5725 | 1010 | poll_freewait(&table); |
1da177e4 | 1011 | |
252e5725 | 1012 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
1013 | struct pollfd *fds = walk->entries; |
1014 | int j; | |
1015 | ||
252e5725 ON |
1016 | for (j = 0; j < walk->len; j++, ufds++) |
1017 | if (__put_user(fds[j].revents, &ufds->revents)) | |
1da177e4 | 1018 | goto out_fds; |
1da177e4 | 1019 | } |
252e5725 | 1020 | |
1da177e4 | 1021 | err = fdcount; |
1da177e4 | 1022 | out_fds: |
252e5725 ON |
1023 | walk = head->next; |
1024 | while (walk) { | |
1025 | struct poll_list *pos = walk; | |
1026 | walk = walk->next; | |
1027 | kfree(pos); | |
1da177e4 | 1028 | } |
252e5725 | 1029 | |
1da177e4 LT |
1030 | return err; |
1031 | } | |
9f72949f | 1032 | |
3075d9da CW |
1033 | static long do_restart_poll(struct restart_block *restart_block) |
1034 | { | |
8ff3e8e8 AV |
1035 | struct pollfd __user *ufds = restart_block->poll.ufds; |
1036 | int nfds = restart_block->poll.nfds; | |
766b9f92 | 1037 | struct timespec64 *to = NULL, end_time; |
3075d9da CW |
1038 | int ret; |
1039 | ||
8ff3e8e8 AV |
1040 | if (restart_block->poll.has_timeout) { |
1041 | end_time.tv_sec = restart_block->poll.tv_sec; | |
1042 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
1043 | to = &end_time; | |
1044 | } | |
1045 | ||
1046 | ret = do_sys_poll(ufds, nfds, to); | |
1047 | ||
3075d9da CW |
1048 | if (ret == -EINTR) { |
1049 | restart_block->fn = do_restart_poll; | |
3075d9da CW |
1050 | ret = -ERESTART_RESTARTBLOCK; |
1051 | } | |
1052 | return ret; | |
1053 | } | |
1054 | ||
5a8a82b1 | 1055 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
faf30900 | 1056 | int, timeout_msecs) |
9f72949f | 1057 | { |
766b9f92 | 1058 | struct timespec64 end_time, *to = NULL; |
3075d9da | 1059 | int ret; |
9f72949f | 1060 | |
8ff3e8e8 AV |
1061 | if (timeout_msecs >= 0) { |
1062 | to = &end_time; | |
1063 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
1064 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
1065 | } |
1066 | ||
8ff3e8e8 AV |
1067 | ret = do_sys_poll(ufds, nfds, to); |
1068 | ||
3075d9da CW |
1069 | if (ret == -EINTR) { |
1070 | struct restart_block *restart_block; | |
8ff3e8e8 | 1071 | |
f56141e3 | 1072 | restart_block = ¤t->restart_block; |
3075d9da | 1073 | restart_block->fn = do_restart_poll; |
8ff3e8e8 AV |
1074 | restart_block->poll.ufds = ufds; |
1075 | restart_block->poll.nfds = nfds; | |
1076 | ||
1077 | if (timeout_msecs >= 0) { | |
1078 | restart_block->poll.tv_sec = end_time.tv_sec; | |
1079 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
1080 | restart_block->poll.has_timeout = 1; | |
1081 | } else | |
1082 | restart_block->poll.has_timeout = 0; | |
1083 | ||
3075d9da CW |
1084 | ret = -ERESTART_RESTARTBLOCK; |
1085 | } | |
1086 | return ret; | |
9f72949f DW |
1087 | } |
1088 | ||
d4e82042 | 1089 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
8bd27a30 | 1090 | struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask, |
d4e82042 | 1091 | size_t, sigsetsize) |
9f72949f DW |
1092 | { |
1093 | sigset_t ksigmask, sigsaved; | |
36819ad0 | 1094 | struct timespec64 ts, end_time, *to = NULL; |
9f72949f DW |
1095 | int ret; |
1096 | ||
1097 | if (tsp) { | |
36819ad0 | 1098 | if (get_timespec64(&ts, tsp)) |
9f72949f DW |
1099 | return -EFAULT; |
1100 | ||
8ff3e8e8 AV |
1101 | to = &end_time; |
1102 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1103 | return -EINVAL; | |
9f72949f DW |
1104 | } |
1105 | ||
ded653cc DD |
1106 | ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); |
1107 | if (ret) | |
1108 | return ret; | |
9f72949f | 1109 | |
8ff3e8e8 | 1110 | ret = do_sys_poll(ufds, nfds, to); |
9f72949f | 1111 | |
854a6ed5 DD |
1112 | restore_user_sigmask(sigmask, &sigsaved); |
1113 | ||
9f72949f | 1114 | /* We can restart this syscall, usually */ |
854a6ed5 | 1115 | if (ret == -EINTR) |
9f72949f | 1116 | ret = -ERESTARTNOHAND; |
9f72949f | 1117 | |
8bd27a30 | 1118 | ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret); |
9f72949f DW |
1119 | |
1120 | return ret; | |
1121 | } | |
e99ca56c | 1122 | |
8bd27a30 | 1123 | #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) |
e99ca56c | 1124 | |
8bd27a30 DD |
1125 | SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds, |
1126 | struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask, | |
1127 | size_t, sigsetsize) | |
e99ca56c | 1128 | { |
8bd27a30 DD |
1129 | sigset_t ksigmask, sigsaved; |
1130 | struct timespec64 ts, end_time, *to = NULL; | |
1131 | int ret; | |
e99ca56c | 1132 | |
8bd27a30 DD |
1133 | if (tsp) { |
1134 | if (get_old_timespec32(&ts, tsp)) | |
1135 | return -EFAULT; | |
e99ca56c | 1136 | |
8bd27a30 DD |
1137 | to = &end_time; |
1138 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1139 | return -EINVAL; | |
1140 | } | |
e99ca56c | 1141 | |
8bd27a30 DD |
1142 | ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); |
1143 | if (ret) | |
e99ca56c AV |
1144 | return ret; |
1145 | ||
8bd27a30 | 1146 | ret = do_sys_poll(ufds, nfds, to); |
e99ca56c | 1147 | |
8bd27a30 | 1148 | restore_user_sigmask(sigmask, &sigsaved); |
e99ca56c | 1149 | |
8bd27a30 DD |
1150 | /* We can restart this syscall, usually */ |
1151 | if (ret == -EINTR) | |
1152 | ret = -ERESTARTNOHAND; | |
e99ca56c | 1153 | |
8bd27a30 | 1154 | ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret); |
e99ca56c | 1155 | |
e99ca56c AV |
1156 | return ret; |
1157 | } | |
8bd27a30 DD |
1158 | #endif |
1159 | ||
1160 | #ifdef CONFIG_COMPAT | |
1161 | #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t)) | |
e99ca56c AV |
1162 | |
1163 | /* | |
1164 | * Ooo, nasty. We need here to frob 32-bit unsigned longs to | |
1165 | * 64-bit unsigned longs. | |
1166 | */ | |
1167 | static | |
1168 | int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, | |
1169 | unsigned long *fdset) | |
1170 | { | |
e99ca56c | 1171 | if (ufdset) { |
464d6242 | 1172 | return compat_get_bitmap(fdset, ufdset, nr); |
e99ca56c | 1173 | } else { |
79de3cbe | 1174 | zero_fd_set(nr, fdset); |
464d6242 | 1175 | return 0; |
e99ca56c | 1176 | } |
e99ca56c AV |
1177 | } |
1178 | ||
1179 | static | |
1180 | int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, | |
1181 | unsigned long *fdset) | |
1182 | { | |
e99ca56c AV |
1183 | if (!ufdset) |
1184 | return 0; | |
464d6242 | 1185 | return compat_put_bitmap(ufdset, fdset, nr); |
e99ca56c AV |
1186 | } |
1187 | ||
1188 | ||
1189 | /* | |
1190 | * This is a virtual copy of sys_select from fs/select.c and probably | |
1191 | * should be compared to it from time to time | |
1192 | */ | |
1193 | ||
1194 | /* | |
1195 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
1196 | * like to be certain this leads to no problems. So I return | |
1197 | * EINTR just for safety. | |
1198 | * | |
1199 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
1200 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
1201 | */ | |
1202 | static int compat_core_sys_select(int n, compat_ulong_t __user *inp, | |
1203 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
36819ad0 | 1204 | struct timespec64 *end_time) |
e99ca56c AV |
1205 | { |
1206 | fd_set_bits fds; | |
1207 | void *bits; | |
1208 | int size, max_fds, ret = -EINVAL; | |
1209 | struct fdtable *fdt; | |
1210 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; | |
1211 | ||
1212 | if (n < 0) | |
1213 | goto out_nofds; | |
1214 | ||
1215 | /* max_fds can increase, so grab it once to avoid race */ | |
1216 | rcu_read_lock(); | |
1217 | fdt = files_fdtable(current->files); | |
1218 | max_fds = fdt->max_fds; | |
1219 | rcu_read_unlock(); | |
1220 | if (n > max_fds) | |
1221 | n = max_fds; | |
1222 | ||
1223 | /* | |
1224 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
1225 | * since we used fdset we need to allocate memory in units of | |
1226 | * long-words. | |
1227 | */ | |
1228 | size = FDS_BYTES(n); | |
1229 | bits = stack_fds; | |
1230 | if (size > sizeof(stack_fds) / 6) { | |
6da2ec56 | 1231 | bits = kmalloc_array(6, size, GFP_KERNEL); |
e99ca56c AV |
1232 | ret = -ENOMEM; |
1233 | if (!bits) | |
1234 | goto out_nofds; | |
1235 | } | |
1236 | fds.in = (unsigned long *) bits; | |
1237 | fds.out = (unsigned long *) (bits + size); | |
1238 | fds.ex = (unsigned long *) (bits + 2*size); | |
1239 | fds.res_in = (unsigned long *) (bits + 3*size); | |
1240 | fds.res_out = (unsigned long *) (bits + 4*size); | |
1241 | fds.res_ex = (unsigned long *) (bits + 5*size); | |
1242 | ||
1243 | if ((ret = compat_get_fd_set(n, inp, fds.in)) || | |
1244 | (ret = compat_get_fd_set(n, outp, fds.out)) || | |
1245 | (ret = compat_get_fd_set(n, exp, fds.ex))) | |
1246 | goto out; | |
1247 | zero_fd_set(n, fds.res_in); | |
1248 | zero_fd_set(n, fds.res_out); | |
1249 | zero_fd_set(n, fds.res_ex); | |
1250 | ||
1251 | ret = do_select(n, &fds, end_time); | |
1252 | ||
1253 | if (ret < 0) | |
1254 | goto out; | |
1255 | if (!ret) { | |
1256 | ret = -ERESTARTNOHAND; | |
1257 | if (signal_pending(current)) | |
1258 | goto out; | |
1259 | ret = 0; | |
1260 | } | |
1261 | ||
1262 | if (compat_set_fd_set(n, inp, fds.res_in) || | |
1263 | compat_set_fd_set(n, outp, fds.res_out) || | |
1264 | compat_set_fd_set(n, exp, fds.res_ex)) | |
1265 | ret = -EFAULT; | |
1266 | out: | |
1267 | if (bits != stack_fds) | |
1268 | kfree(bits); | |
1269 | out_nofds: | |
1270 | return ret; | |
1271 | } | |
1272 | ||
05585e44 DB |
1273 | static int do_compat_select(int n, compat_ulong_t __user *inp, |
1274 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
9afc5eee | 1275 | struct old_timeval32 __user *tvp) |
e99ca56c | 1276 | { |
36819ad0 | 1277 | struct timespec64 end_time, *to = NULL; |
9afc5eee | 1278 | struct old_timeval32 tv; |
e99ca56c AV |
1279 | int ret; |
1280 | ||
1281 | if (tvp) { | |
1282 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
1283 | return -EFAULT; | |
1284 | ||
1285 | to = &end_time; | |
1286 | if (poll_select_set_timeout(to, | |
1287 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
1288 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
1289 | return -EINVAL; | |
1290 | } | |
1291 | ||
1292 | ret = compat_core_sys_select(n, inp, outp, exp, to); | |
8bd27a30 | 1293 | ret = poll_select_copy_remaining(&end_time, tvp, PT_OLD_TIMEVAL, ret); |
e99ca56c AV |
1294 | |
1295 | return ret; | |
1296 | } | |
1297 | ||
05585e44 DB |
1298 | COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp, |
1299 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
9afc5eee | 1300 | struct old_timeval32 __user *, tvp) |
05585e44 DB |
1301 | { |
1302 | return do_compat_select(n, inp, outp, exp, tvp); | |
1303 | } | |
1304 | ||
e99ca56c AV |
1305 | struct compat_sel_arg_struct { |
1306 | compat_ulong_t n; | |
1307 | compat_uptr_t inp; | |
1308 | compat_uptr_t outp; | |
1309 | compat_uptr_t exp; | |
1310 | compat_uptr_t tvp; | |
1311 | }; | |
1312 | ||
1313 | COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg) | |
1314 | { | |
1315 | struct compat_sel_arg_struct a; | |
1316 | ||
1317 | if (copy_from_user(&a, arg, sizeof(a))) | |
1318 | return -EFAULT; | |
05585e44 DB |
1319 | return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp), |
1320 | compat_ptr(a.exp), compat_ptr(a.tvp)); | |
e99ca56c AV |
1321 | } |
1322 | ||
1323 | static long do_compat_pselect(int n, compat_ulong_t __user *inp, | |
1324 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
e024707b DD |
1325 | void __user *tsp, compat_sigset_t __user *sigmask, |
1326 | compat_size_t sigsetsize, enum poll_time_type type) | |
e99ca56c | 1327 | { |
e99ca56c | 1328 | sigset_t ksigmask, sigsaved; |
36819ad0 | 1329 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1330 | int ret; |
1331 | ||
1332 | if (tsp) { | |
e024707b DD |
1333 | switch (type) { |
1334 | case PT_OLD_TIMESPEC: | |
1335 | if (get_old_timespec32(&ts, tsp)) | |
1336 | return -EFAULT; | |
1337 | break; | |
1338 | case PT_TIMESPEC: | |
1339 | if (get_timespec64(&ts, tsp)) | |
1340 | return -EFAULT; | |
1341 | break; | |
1342 | default: | |
1343 | BUG(); | |
1344 | } | |
e99ca56c AV |
1345 | |
1346 | to = &end_time; | |
1347 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1348 | return -EINVAL; | |
1349 | } | |
1350 | ||
ded653cc DD |
1351 | ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); |
1352 | if (ret) | |
1353 | return ret; | |
e99ca56c AV |
1354 | |
1355 | ret = compat_core_sys_select(n, inp, outp, exp, to); | |
e024707b | 1356 | ret = poll_select_copy_remaining(&end_time, tsp, type, ret); |
e99ca56c | 1357 | |
854a6ed5 | 1358 | restore_user_sigmask(sigmask, &sigsaved); |
e99ca56c AV |
1359 | |
1360 | return ret; | |
1361 | } | |
1362 | ||
e024707b DD |
1363 | COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp, |
1364 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
1365 | struct __kernel_timespec __user *, tsp, void __user *, sig) | |
1366 | { | |
1367 | compat_size_t sigsetsize = 0; | |
1368 | compat_uptr_t up = 0; | |
1369 | ||
1370 | if (sig) { | |
1371 | if (!access_ok(VERIFY_READ, sig, | |
1372 | sizeof(compat_uptr_t)+sizeof(compat_size_t)) || | |
1373 | __get_user(up, (compat_uptr_t __user *)sig) || | |
1374 | __get_user(sigsetsize, | |
1375 | (compat_size_t __user *)(sig+sizeof(up)))) | |
1376 | return -EFAULT; | |
1377 | } | |
1378 | ||
1379 | return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up), | |
1380 | sigsetsize, PT_TIMESPEC); | |
1381 | } | |
1382 | ||
1383 | #if defined(CONFIG_COMPAT_32BIT_TIME) | |
1384 | ||
e99ca56c AV |
1385 | COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp, |
1386 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
9afc5eee | 1387 | struct old_timespec32 __user *, tsp, void __user *, sig) |
e99ca56c AV |
1388 | { |
1389 | compat_size_t sigsetsize = 0; | |
1390 | compat_uptr_t up = 0; | |
1391 | ||
1392 | if (sig) { | |
1393 | if (!access_ok(VERIFY_READ, sig, | |
1394 | sizeof(compat_uptr_t)+sizeof(compat_size_t)) || | |
1395 | __get_user(up, (compat_uptr_t __user *)sig) || | |
1396 | __get_user(sigsetsize, | |
1397 | (compat_size_t __user *)(sig+sizeof(up)))) | |
1398 | return -EFAULT; | |
1399 | } | |
e024707b | 1400 | |
e99ca56c | 1401 | return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up), |
e024707b | 1402 | sigsetsize, PT_OLD_TIMESPEC); |
e99ca56c AV |
1403 | } |
1404 | ||
e024707b DD |
1405 | #endif |
1406 | ||
8bd27a30 | 1407 | #if defined(CONFIG_COMPAT_32BIT_TIME) |
e99ca56c | 1408 | COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, |
9afc5eee | 1409 | unsigned int, nfds, struct old_timespec32 __user *, tsp, |
e99ca56c AV |
1410 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) |
1411 | { | |
e99ca56c | 1412 | sigset_t ksigmask, sigsaved; |
36819ad0 | 1413 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1414 | int ret; |
1415 | ||
1416 | if (tsp) { | |
9afc5eee | 1417 | if (get_old_timespec32(&ts, tsp)) |
e99ca56c AV |
1418 | return -EFAULT; |
1419 | ||
1420 | to = &end_time; | |
1421 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1422 | return -EINVAL; | |
1423 | } | |
1424 | ||
ded653cc DD |
1425 | ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); |
1426 | if (ret) | |
1427 | return ret; | |
e99ca56c AV |
1428 | |
1429 | ret = do_sys_poll(ufds, nfds, to); | |
1430 | ||
854a6ed5 DD |
1431 | restore_user_sigmask(sigmask, &sigsaved); |
1432 | ||
e99ca56c | 1433 | /* We can restart this syscall, usually */ |
854a6ed5 | 1434 | if (ret == -EINTR) |
e99ca56c | 1435 | ret = -ERESTARTNOHAND; |
e99ca56c | 1436 | |
8bd27a30 | 1437 | ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret); |
e99ca56c AV |
1438 | |
1439 | return ret; | |
1440 | } | |
1441 | #endif | |
8bd27a30 DD |
1442 | |
1443 | /* New compat syscall for 64 bit time_t*/ | |
1444 | COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds, | |
1445 | unsigned int, nfds, struct __kernel_timespec __user *, tsp, | |
1446 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) | |
1447 | { | |
1448 | sigset_t ksigmask, sigsaved; | |
1449 | struct timespec64 ts, end_time, *to = NULL; | |
1450 | int ret; | |
1451 | ||
1452 | if (tsp) { | |
1453 | if (get_timespec64(&ts, tsp)) | |
1454 | return -EFAULT; | |
1455 | ||
1456 | to = &end_time; | |
1457 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1458 | return -EINVAL; | |
1459 | } | |
1460 | ||
1461 | ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); | |
1462 | if (ret) | |
1463 | return ret; | |
1464 | ||
1465 | ret = do_sys_poll(ufds, nfds, to); | |
1466 | ||
1467 | restore_user_sigmask(sigmask, &sigsaved); | |
1468 | ||
1469 | /* We can restart this syscall, usually */ | |
1470 | if (ret == -EINTR) | |
1471 | ret = -ERESTARTNOHAND; | |
1472 | ||
1473 | ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret); | |
1474 | ||
1475 | return ret; | |
1476 | } | |
1477 | ||
1478 | #endif |