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 | ||
b6459415 | 18 | #include <linux/compat.h> |
022a1692 | 19 | #include <linux/kernel.h> |
3f07c014 IM |
20 | #include <linux/sched/signal.h> |
21 | #include <linux/sched/rt.h> | |
1da177e4 | 22 | #include <linux/syscalls.h> |
630d9c47 | 23 | #include <linux/export.h> |
1da177e4 | 24 | #include <linux/slab.h> |
1da177e4 LT |
25 | #include <linux/poll.h> |
26 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ | |
27 | #include <linux/file.h> | |
9f3acc31 | 28 | #include <linux/fdtable.h> |
1da177e4 | 29 | #include <linux/fs.h> |
b835996f | 30 | #include <linux/rcupdate.h> |
8ff3e8e8 | 31 | #include <linux/hrtimer.h> |
9745cdb3 | 32 | #include <linux/freezer.h> |
076bb0c8 | 33 | #include <net/busy_poll.h> |
2d19309c | 34 | #include <linux/vmalloc.h> |
1da177e4 | 35 | |
7c0f6ba6 | 36 | #include <linux/uaccess.h> |
1da177e4 | 37 | |
90d6e24a AV |
38 | |
39 | /* | |
40 | * Estimate expected accuracy in ns from a timeval. | |
41 | * | |
42 | * After quite a bit of churning around, we've settled on | |
43 | * a simple thing of taking 0.1% of the timeout as the | |
44 | * slack, with a cap of 100 msec. | |
45 | * "nice" tasks get a 0.5% slack instead. | |
46 | * | |
47 | * Consider this comment an open invitation to come up with even | |
48 | * better solutions.. | |
49 | */ | |
50 | ||
5ae87e79 GK |
51 | #define MAX_SLACK (100 * NSEC_PER_MSEC) |
52 | ||
766b9f92 | 53 | static long __estimate_accuracy(struct timespec64 *tv) |
90d6e24a | 54 | { |
96d2ab48 | 55 | long slack; |
90d6e24a AV |
56 | int divfactor = 1000; |
57 | ||
5ae87e79 GK |
58 | if (tv->tv_sec < 0) |
59 | return 0; | |
60 | ||
4ce105d3 | 61 | if (task_nice(current) > 0) |
90d6e24a AV |
62 | divfactor = divfactor / 5; |
63 | ||
5ae87e79 GK |
64 | if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) |
65 | return MAX_SLACK; | |
66 | ||
90d6e24a AV |
67 | slack = tv->tv_nsec / divfactor; |
68 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); | |
69 | ||
5ae87e79 GK |
70 | if (slack > MAX_SLACK) |
71 | return MAX_SLACK; | |
96d2ab48 | 72 | |
90d6e24a AV |
73 | return slack; |
74 | } | |
75 | ||
766b9f92 | 76 | u64 select_estimate_accuracy(struct timespec64 *tv) |
90d6e24a | 77 | { |
da8b44d5 | 78 | u64 ret; |
766b9f92 | 79 | struct timespec64 now; |
90d6e24a AV |
80 | |
81 | /* | |
82 | * Realtime tasks get a slack of 0 for obvious reasons. | |
83 | */ | |
84 | ||
4ce105d3 | 85 | if (rt_task(current)) |
90d6e24a AV |
86 | return 0; |
87 | ||
766b9f92 DD |
88 | ktime_get_ts64(&now); |
89 | now = timespec64_sub(*tv, now); | |
90d6e24a AV |
90 | ret = __estimate_accuracy(&now); |
91 | if (ret < current->timer_slack_ns) | |
92 | return current->timer_slack_ns; | |
93 | return ret; | |
94 | } | |
95 | ||
96 | ||
97 | ||
1da177e4 LT |
98 | struct poll_table_page { |
99 | struct poll_table_page * next; | |
100 | struct poll_table_entry * entry; | |
5e01fdff | 101 | struct poll_table_entry entries[]; |
1da177e4 LT |
102 | }; |
103 | ||
104 | #define POLL_TABLE_FULL(table) \ | |
105 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
106 | ||
107 | /* | |
108 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
109 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
110 | * follow, but it should be free of race-conditions, and it's practical. If you | |
111 | * understand what I'm doing here, then you understand how the linux | |
112 | * sleep/wakeup mechanism works. | |
113 | * | |
114 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
115 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
116 | * as all select/poll functions have to call it to add an entry to the | |
117 | * poll table. | |
118 | */ | |
75c96f85 AB |
119 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
120 | poll_table *p); | |
1da177e4 LT |
121 | |
122 | void poll_initwait(struct poll_wqueues *pwq) | |
123 | { | |
124 | init_poll_funcptr(&pwq->pt, __pollwait); | |
5f820f64 | 125 | pwq->polling_task = current; |
b2add73d | 126 | pwq->triggered = 0; |
1da177e4 LT |
127 | pwq->error = 0; |
128 | pwq->table = NULL; | |
70674f95 | 129 | pwq->inline_index = 0; |
1da177e4 | 130 | } |
1da177e4 LT |
131 | EXPORT_SYMBOL(poll_initwait); |
132 | ||
70674f95 AK |
133 | static void free_poll_entry(struct poll_table_entry *entry) |
134 | { | |
ccf6780d | 135 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
136 | fput(entry->filp); |
137 | } | |
138 | ||
1da177e4 LT |
139 | void poll_freewait(struct poll_wqueues *pwq) |
140 | { | |
141 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
142 | int i; |
143 | for (i = 0; i < pwq->inline_index; i++) | |
144 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
145 | while (p) { |
146 | struct poll_table_entry * entry; | |
147 | struct poll_table_page *old; | |
148 | ||
149 | entry = p->entry; | |
150 | do { | |
151 | entry--; | |
70674f95 | 152 | free_poll_entry(entry); |
1da177e4 LT |
153 | } while (entry > p->entries); |
154 | old = p; | |
155 | p = p->next; | |
156 | free_page((unsigned long) old); | |
157 | } | |
158 | } | |
1da177e4 LT |
159 | EXPORT_SYMBOL(poll_freewait); |
160 | ||
5f820f64 | 161 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
1da177e4 | 162 | { |
1da177e4 LT |
163 | struct poll_table_page *table = p->table; |
164 | ||
70674f95 AK |
165 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
166 | return p->inline_entries + p->inline_index++; | |
167 | ||
1da177e4 LT |
168 | if (!table || POLL_TABLE_FULL(table)) { |
169 | struct poll_table_page *new_table; | |
170 | ||
171 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
172 | if (!new_table) { | |
173 | p->error = -ENOMEM; | |
70674f95 | 174 | return NULL; |
1da177e4 LT |
175 | } |
176 | new_table->entry = new_table->entries; | |
177 | new_table->next = table; | |
178 | p->table = new_table; | |
179 | table = new_table; | |
180 | } | |
181 | ||
70674f95 AK |
182 | return table->entry++; |
183 | } | |
184 | ||
ac6424b9 | 185 | static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
5f820f64 TH |
186 | { |
187 | struct poll_wqueues *pwq = wait->private; | |
188 | DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); | |
189 | ||
190 | /* | |
191 | * Although this function is called under waitqueue lock, LOCK | |
192 | * doesn't imply write barrier and the users expect write | |
193 | * barrier semantics on wakeup functions. The following | |
194 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
b92b8b35 | 195 | * and is paired with smp_store_mb() in poll_schedule_timeout. |
5f820f64 TH |
196 | */ |
197 | smp_wmb(); | |
198 | pwq->triggered = 1; | |
199 | ||
200 | /* | |
201 | * Perform the default wake up operation using a dummy | |
202 | * waitqueue. | |
203 | * | |
204 | * TODO: This is hacky but there currently is no interface to | |
205 | * pass in @sync. @sync is scheduled to be removed and once | |
206 | * that happens, wake_up_process() can be used directly. | |
207 | */ | |
208 | return default_wake_function(&dummy_wait, mode, sync, key); | |
209 | } | |
210 | ||
ac6424b9 | 211 | static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
4938d7e0 ED |
212 | { |
213 | struct poll_table_entry *entry; | |
214 | ||
215 | entry = container_of(wait, struct poll_table_entry, wait); | |
3ad6f93e | 216 | if (key && !(key_to_poll(key) & entry->key)) |
4938d7e0 ED |
217 | return 0; |
218 | return __pollwake(wait, mode, sync, key); | |
219 | } | |
220 | ||
70674f95 AK |
221 | /* Add a new entry */ |
222 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
223 | poll_table *p) | |
224 | { | |
5f820f64 TH |
225 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
226 | struct poll_table_entry *entry = poll_get_entry(pwq); | |
70674f95 AK |
227 | if (!entry) |
228 | return; | |
cb0942b8 | 229 | entry->filp = get_file(filp); |
70674f95 | 230 | entry->wait_address = wait_address; |
626cf236 | 231 | entry->key = p->_key; |
5f820f64 TH |
232 | init_waitqueue_func_entry(&entry->wait, pollwake); |
233 | entry->wait.private = pwq; | |
ccf6780d | 234 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
235 | } |
236 | ||
8f546ae1 | 237 | static int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
5f820f64 TH |
238 | ktime_t *expires, unsigned long slack) |
239 | { | |
240 | int rc = -EINTR; | |
241 | ||
242 | set_current_state(state); | |
243 | if (!pwq->triggered) | |
59612d18 | 244 | rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); |
5f820f64 TH |
245 | __set_current_state(TASK_RUNNING); |
246 | ||
247 | /* | |
248 | * Prepare for the next iteration. | |
249 | * | |
b92b8b35 | 250 | * The following smp_store_mb() serves two purposes. First, it's |
5f820f64 TH |
251 | * the counterpart rmb of the wmb in pollwake() such that data |
252 | * written before wake up is always visible after wake up. | |
253 | * Second, the full barrier guarantees that triggered clearing | |
254 | * doesn't pass event check of the next iteration. Note that | |
255 | * this problem doesn't exist for the first iteration as | |
256 | * add_wait_queue() has full barrier semantics. | |
257 | */ | |
b92b8b35 | 258 | smp_store_mb(pwq->triggered, 0); |
5f820f64 TH |
259 | |
260 | return rc; | |
261 | } | |
5f820f64 | 262 | |
b773ad40 TG |
263 | /** |
264 | * poll_select_set_timeout - helper function to setup the timeout value | |
766b9f92 | 265 | * @to: pointer to timespec64 variable for the final timeout |
b773ad40 TG |
266 | * @sec: seconds (from user space) |
267 | * @nsec: nanoseconds (from user space) | |
268 | * | |
269 | * Note, we do not use a timespec for the user space value here, That | |
270 | * way we can use the function for timeval and compat interfaces as well. | |
271 | * | |
272 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
273 | */ | |
766b9f92 | 274 | int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec) |
b773ad40 | 275 | { |
766b9f92 | 276 | struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec}; |
b773ad40 | 277 | |
766b9f92 | 278 | if (!timespec64_valid(&ts)) |
b773ad40 TG |
279 | return -EINVAL; |
280 | ||
281 | /* Optimize for the zero timeout value here */ | |
282 | if (!sec && !nsec) { | |
283 | to->tv_sec = to->tv_nsec = 0; | |
284 | } else { | |
766b9f92 DD |
285 | ktime_get_ts64(to); |
286 | *to = timespec64_add_safe(*to, ts); | |
b773ad40 TG |
287 | } |
288 | return 0; | |
289 | } | |
290 | ||
8bd27a30 DD |
291 | enum poll_time_type { |
292 | PT_TIMEVAL = 0, | |
293 | PT_OLD_TIMEVAL = 1, | |
294 | PT_TIMESPEC = 2, | |
295 | PT_OLD_TIMESPEC = 3, | |
296 | }; | |
297 | ||
ac301020 ON |
298 | static int poll_select_finish(struct timespec64 *end_time, |
299 | void __user *p, | |
300 | enum poll_time_type pt_type, int ret) | |
b773ad40 | 301 | { |
36819ad0 | 302 | struct timespec64 rts; |
b773ad40 | 303 | |
ac301020 ON |
304 | restore_saved_sigmask_unless(ret == -ERESTARTNOHAND); |
305 | ||
b773ad40 TG |
306 | if (!p) |
307 | return ret; | |
308 | ||
309 | if (current->personality & STICKY_TIMEOUTS) | |
310 | goto sticky; | |
311 | ||
312 | /* No update for zero timeout */ | |
313 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
314 | return ret; | |
315 | ||
36819ad0 DD |
316 | ktime_get_ts64(&rts); |
317 | rts = timespec64_sub(*end_time, rts); | |
318 | if (rts.tv_sec < 0) | |
319 | rts.tv_sec = rts.tv_nsec = 0; | |
766b9f92 | 320 | |
b773ad40 | 321 | |
8bd27a30 DD |
322 | switch (pt_type) { |
323 | case PT_TIMEVAL: | |
324 | { | |
75d319c0 | 325 | struct __kernel_old_timeval rtv; |
b773ad40 | 326 | |
8bd27a30 DD |
327 | if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) |
328 | memset(&rtv, 0, sizeof(rtv)); | |
329 | rtv.tv_sec = rts.tv_sec; | |
330 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
331 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
332 | return ret; | |
333 | } | |
334 | break; | |
335 | case PT_OLD_TIMEVAL: | |
336 | { | |
337 | struct old_timeval32 rtv; | |
338 | ||
339 | rtv.tv_sec = rts.tv_sec; | |
340 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
341 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
342 | return ret; | |
343 | } | |
344 | break; | |
345 | case PT_TIMESPEC: | |
346 | if (!put_timespec64(&rts, p)) | |
b773ad40 | 347 | return ret; |
8bd27a30 DD |
348 | break; |
349 | case PT_OLD_TIMESPEC: | |
350 | if (!put_old_timespec32(&rts, p)) | |
351 | return ret; | |
352 | break; | |
353 | default: | |
354 | BUG(); | |
355 | } | |
b773ad40 TG |
356 | /* |
357 | * If an application puts its timeval in read-only memory, we | |
358 | * don't want the Linux-specific update to the timeval to | |
359 | * cause a fault after the select has completed | |
360 | * successfully. However, because we're not updating the | |
361 | * timeval, we can't restart the system call. | |
362 | */ | |
363 | ||
364 | sticky: | |
365 | if (ret == -ERESTARTNOHAND) | |
366 | ret = -EINTR; | |
367 | return ret; | |
368 | } | |
369 | ||
e99ca56c AV |
370 | /* |
371 | * Scalable version of the fd_set. | |
372 | */ | |
373 | ||
374 | typedef struct { | |
375 | unsigned long *in, *out, *ex; | |
376 | unsigned long *res_in, *res_out, *res_ex; | |
377 | } fd_set_bits; | |
378 | ||
379 | /* | |
380 | * How many longwords for "nr" bits? | |
381 | */ | |
382 | #define FDS_BITPERLONG (8*sizeof(long)) | |
383 | #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG) | |
384 | #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long)) | |
385 | ||
386 | /* | |
e99ca56c AV |
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 | ||
68514dac JK |
462 | #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR |\ |
463 | EPOLLNVAL) | |
464 | #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\ | |
465 | EPOLLNVAL) | |
466 | #define POLLEX_SET (EPOLLPRI | EPOLLNVAL) | |
1da177e4 | 467 | |
4938d7e0 | 468 | static inline void wait_key_set(poll_table *wait, unsigned long in, |
2d48d67f | 469 | unsigned long out, unsigned long bit, |
01699437 | 470 | __poll_t ll_flag) |
4938d7e0 | 471 | { |
2d48d67f | 472 | wait->_key = POLLEX_SET | ll_flag; |
626cf236 HV |
473 | if (in & bit) |
474 | wait->_key |= POLLIN_SET; | |
475 | if (out & bit) | |
476 | wait->_key |= POLLOUT_SET; | |
4938d7e0 ED |
477 | } |
478 | ||
e99ca56c | 479 | static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time) |
1da177e4 | 480 | { |
8ff3e8e8 | 481 | ktime_t expire, *to = NULL; |
1da177e4 LT |
482 | struct poll_wqueues table; |
483 | poll_table *wait; | |
8ff3e8e8 | 484 | int retval, i, timed_out = 0; |
da8b44d5 | 485 | u64 slack = 0; |
01699437 | 486 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 487 | unsigned long busy_start = 0; |
1da177e4 | 488 | |
b835996f | 489 | rcu_read_lock(); |
1da177e4 | 490 | retval = max_select_fd(n, fds); |
b835996f | 491 | rcu_read_unlock(); |
1da177e4 LT |
492 | |
493 | if (retval < 0) | |
494 | return retval; | |
495 | n = retval; | |
496 | ||
497 | poll_initwait(&table); | |
498 | wait = &table.pt; | |
8ff3e8e8 | 499 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 500 | wait->_qproc = NULL; |
8ff3e8e8 AV |
501 | timed_out = 1; |
502 | } | |
503 | ||
96d2ab48 | 504 | if (end_time && !timed_out) |
231f3d39 | 505 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 506 | |
1da177e4 LT |
507 | retval = 0; |
508 | for (;;) { | |
509 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
cbf55001 | 510 | bool can_busy_loop = false; |
1da177e4 | 511 | |
1da177e4 LT |
512 | inp = fds->in; outp = fds->out; exp = fds->ex; |
513 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
514 | ||
515 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
e6c8adca | 516 | unsigned long in, out, ex, all_bits, bit = 1, j; |
1da177e4 | 517 | unsigned long res_in = 0, res_out = 0, res_ex = 0; |
e6c8adca | 518 | __poll_t mask; |
1da177e4 LT |
519 | |
520 | in = *inp++; out = *outp++; ex = *exp++; | |
521 | all_bits = in | out | ex; | |
522 | if (all_bits == 0) { | |
8ded2bbc | 523 | i += BITS_PER_LONG; |
1da177e4 LT |
524 | continue; |
525 | } | |
526 | ||
8ded2bbc | 527 | for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { |
2903ff01 | 528 | struct fd f; |
1da177e4 LT |
529 | if (i >= n) |
530 | break; | |
531 | if (!(bit & all_bits)) | |
532 | continue; | |
68514dac | 533 | mask = EPOLLNVAL; |
2903ff01 AV |
534 | f = fdget(i); |
535 | if (f.file) { | |
9965ed17 CH |
536 | wait_key_set(wait, in, out, bit, |
537 | busy_flag); | |
538 | mask = vfs_poll(f.file, wait); | |
539 | ||
2903ff01 | 540 | fdput(f); |
1da177e4 | 541 | } |
68514dac JK |
542 | if ((mask & POLLIN_SET) && (in & bit)) { |
543 | res_in |= bit; | |
544 | retval++; | |
545 | wait->_qproc = NULL; | |
546 | } | |
547 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
548 | res_out |= bit; | |
549 | retval++; | |
550 | wait->_qproc = NULL; | |
551 | } | |
552 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
553 | res_ex |= bit; | |
554 | retval++; | |
555 | wait->_qproc = NULL; | |
556 | } | |
557 | /* got something, stop busy polling */ | |
558 | if (retval) { | |
559 | can_busy_loop = false; | |
560 | busy_flag = 0; | |
561 | ||
562 | /* | |
563 | * only remember a returned | |
564 | * POLL_BUSY_LOOP if we asked for it | |
565 | */ | |
566 | } else if (busy_flag & mask) | |
567 | can_busy_loop = true; | |
568 | ||
1da177e4 LT |
569 | } |
570 | if (res_in) | |
571 | *rinp = res_in; | |
572 | if (res_out) | |
573 | *routp = res_out; | |
574 | if (res_ex) | |
575 | *rexp = res_ex; | |
55d85384 | 576 | cond_resched(); |
1da177e4 | 577 | } |
626cf236 | 578 | wait->_qproc = NULL; |
8ff3e8e8 | 579 | if (retval || timed_out || signal_pending(current)) |
1da177e4 | 580 | break; |
f5264481 | 581 | if (table.error) { |
1da177e4 LT |
582 | retval = table.error; |
583 | break; | |
584 | } | |
9f72949f | 585 | |
cbf55001 | 586 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 | 587 | if (can_busy_loop && !need_resched()) { |
37056719 AD |
588 | if (!busy_start) { |
589 | busy_start = busy_loop_current_time(); | |
76b1e9b9 ET |
590 | continue; |
591 | } | |
37056719 | 592 | if (!busy_loop_timeout(busy_start)) |
76b1e9b9 ET |
593 | continue; |
594 | } | |
595 | busy_flag = 0; | |
2d48d67f | 596 | |
8ff3e8e8 AV |
597 | /* |
598 | * If this is the first loop and we have a timeout | |
599 | * given, then we convert to ktime_t and set the to | |
600 | * pointer to the expiry value. | |
601 | */ | |
602 | if (end_time && !to) { | |
766b9f92 | 603 | expire = timespec64_to_ktime(*end_time); |
8ff3e8e8 | 604 | to = &expire; |
9f72949f | 605 | } |
8ff3e8e8 | 606 | |
5f820f64 TH |
607 | if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE, |
608 | to, slack)) | |
8ff3e8e8 | 609 | timed_out = 1; |
1da177e4 | 610 | } |
1da177e4 LT |
611 | |
612 | poll_freewait(&table); | |
613 | ||
1da177e4 LT |
614 | return retval; |
615 | } | |
616 | ||
1da177e4 LT |
617 | /* |
618 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
619 | * like to be certain this leads to no problems. So I return | |
620 | * EINTR just for safety. | |
621 | * | |
622 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
623 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
624 | */ | |
a2dcb44c | 625 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
766b9f92 | 626 | fd_set __user *exp, struct timespec64 *end_time) |
1da177e4 LT |
627 | { |
628 | fd_set_bits fds; | |
29ff2db5 | 629 | void *bits; |
bbea9f69 | 630 | int ret, max_fds; |
2d19309c | 631 | size_t size, alloc_size; |
badf1662 | 632 | struct fdtable *fdt; |
70674f95 | 633 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 634 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 635 | |
1da177e4 LT |
636 | ret = -EINVAL; |
637 | if (n < 0) | |
638 | goto out_nofds; | |
639 | ||
bbea9f69 | 640 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 641 | rcu_read_lock(); |
badf1662 | 642 | fdt = files_fdtable(current->files); |
bbea9f69 | 643 | max_fds = fdt->max_fds; |
b835996f | 644 | rcu_read_unlock(); |
bbea9f69 VL |
645 | if (n > max_fds) |
646 | n = max_fds; | |
1da177e4 LT |
647 | |
648 | /* | |
649 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
650 | * since we used fdset we need to allocate memory in units of | |
651 | * long-words. | |
652 | */ | |
1da177e4 | 653 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
654 | bits = stack_fds; |
655 | if (size > sizeof(stack_fds) / 6) { | |
656 | /* Not enough space in on-stack array; must use kmalloc */ | |
657 | ret = -ENOMEM; | |
2d19309c VB |
658 | if (size > (SIZE_MAX / 6)) |
659 | goto out_nofds; | |
660 | ||
661 | alloc_size = 6 * size; | |
0bcfe68b | 662 | bits = kvmalloc(alloc_size, GFP_KERNEL); |
b04eb6aa MBJ |
663 | if (!bits) |
664 | goto out_nofds; | |
665 | } | |
29ff2db5 AM |
666 | fds.in = bits; |
667 | fds.out = bits + size; | |
668 | fds.ex = bits + 2*size; | |
669 | fds.res_in = bits + 3*size; | |
670 | fds.res_out = bits + 4*size; | |
671 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
672 | |
673 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
674 | (ret = get_fd_set(n, outp, fds.out)) || | |
675 | (ret = get_fd_set(n, exp, fds.ex))) | |
676 | goto out; | |
677 | zero_fd_set(n, fds.res_in); | |
678 | zero_fd_set(n, fds.res_out); | |
679 | zero_fd_set(n, fds.res_ex); | |
680 | ||
8ff3e8e8 | 681 | ret = do_select(n, &fds, end_time); |
1da177e4 LT |
682 | |
683 | if (ret < 0) | |
684 | goto out; | |
685 | if (!ret) { | |
686 | ret = -ERESTARTNOHAND; | |
687 | if (signal_pending(current)) | |
688 | goto out; | |
689 | ret = 0; | |
690 | } | |
691 | ||
692 | if (set_fd_set(n, inp, fds.res_in) || | |
693 | set_fd_set(n, outp, fds.res_out) || | |
694 | set_fd_set(n, exp, fds.res_ex)) | |
695 | ret = -EFAULT; | |
696 | ||
697 | out: | |
70674f95 | 698 | if (bits != stack_fds) |
2d19309c | 699 | kvfree(bits); |
1da177e4 LT |
700 | out_nofds: |
701 | return ret; | |
702 | } | |
703 | ||
4bdb9aca | 704 | static int kern_select(int n, fd_set __user *inp, fd_set __user *outp, |
75d319c0 | 705 | fd_set __user *exp, struct __kernel_old_timeval __user *tvp) |
9f72949f | 706 | { |
766b9f92 | 707 | struct timespec64 end_time, *to = NULL; |
75d319c0 | 708 | struct __kernel_old_timeval tv; |
9f72949f DW |
709 | int ret; |
710 | ||
711 | if (tvp) { | |
712 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
713 | return -EFAULT; | |
714 | ||
8ff3e8e8 | 715 | to = &end_time; |
4d36a9e6 AV |
716 | if (poll_select_set_timeout(to, |
717 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
718 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
9f72949f | 719 | return -EINVAL; |
9f72949f DW |
720 | } |
721 | ||
8ff3e8e8 | 722 | ret = core_sys_select(n, inp, outp, exp, to); |
ac301020 | 723 | return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret); |
9f72949f DW |
724 | } |
725 | ||
4bdb9aca | 726 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
75d319c0 | 727 | fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp) |
4bdb9aca DB |
728 | { |
729 | return kern_select(n, inp, outp, exp, tvp); | |
730 | } | |
731 | ||
c9da9f21 | 732 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
e024707b DD |
733 | fd_set __user *exp, void __user *tsp, |
734 | const sigset_t __user *sigmask, size_t sigsetsize, | |
735 | enum poll_time_type type) | |
9f72949f | 736 | { |
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 | ||
b772434b | 759 | ret = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
760 | if (ret) |
761 | return ret; | |
9f72949f | 762 | |
62568510 | 763 | ret = core_sys_select(n, inp, outp, exp, to); |
ac301020 | 764 | return poll_select_finish(&end_time, tsp, type, ret); |
9f72949f DW |
765 | } |
766 | ||
767 | /* | |
768 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
769 | * 6-argument version where the sixth argument is a pointer to a structure | |
770 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
771 | * the sigset size. | |
772 | */ | |
7e71609f AV |
773 | struct sigset_argpack { |
774 | sigset_t __user *p; | |
775 | size_t size; | |
776 | }; | |
777 | ||
778 | static inline int get_sigset_argpack(struct sigset_argpack *to, | |
779 | struct sigset_argpack __user *from) | |
780 | { | |
781 | // the path is hot enough for overhead of copy_from_user() to matter | |
782 | if (from) { | |
783 | if (!user_read_access_begin(from, sizeof(*from))) | |
784 | return -EFAULT; | |
785 | unsafe_get_user(to->p, &from->p, Efault); | |
786 | unsafe_get_user(to->size, &from->size, Efault); | |
787 | user_read_access_end(); | |
788 | } | |
789 | return 0; | |
790 | Efault: | |
791 | user_access_end(); | |
792 | return -EFAULT; | |
793 | } | |
794 | ||
d4e82042 | 795 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
e024707b DD |
796 | fd_set __user *, exp, struct __kernel_timespec __user *, tsp, |
797 | void __user *, sig) | |
798 | { | |
7e71609f AV |
799 | struct sigset_argpack x = {NULL, 0}; |
800 | ||
801 | if (get_sigset_argpack(&x, sig)) | |
802 | return -EFAULT; | |
e024707b | 803 | |
7e71609f | 804 | return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC); |
e024707b DD |
805 | } |
806 | ||
807 | #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) | |
808 | ||
809 | SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp, | |
810 | fd_set __user *, exp, struct old_timespec32 __user *, tsp, | |
d4e82042 | 811 | void __user *, sig) |
9f72949f | 812 | { |
7e71609f AV |
813 | struct sigset_argpack x = {NULL, 0}; |
814 | ||
815 | if (get_sigset_argpack(&x, sig)) | |
816 | return -EFAULT; | |
9f72949f | 817 | |
7e71609f | 818 | return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC); |
9f72949f | 819 | } |
9f72949f | 820 | |
e024707b DD |
821 | #endif |
822 | ||
5d0e5283 CH |
823 | #ifdef __ARCH_WANT_SYS_OLD_SELECT |
824 | struct sel_arg_struct { | |
825 | unsigned long n; | |
826 | fd_set __user *inp, *outp, *exp; | |
75d319c0 | 827 | struct __kernel_old_timeval __user *tvp; |
5d0e5283 CH |
828 | }; |
829 | ||
830 | SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) | |
831 | { | |
832 | struct sel_arg_struct a; | |
833 | ||
834 | if (copy_from_user(&a, arg, sizeof(a))) | |
835 | return -EFAULT; | |
4bdb9aca | 836 | return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp); |
5d0e5283 CH |
837 | } |
838 | #endif | |
839 | ||
1da177e4 LT |
840 | struct poll_list { |
841 | struct poll_list *next; | |
842 | int len; | |
5e01fdff | 843 | struct pollfd entries[]; |
1da177e4 LT |
844 | }; |
845 | ||
846 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
847 | ||
4a4b69f7 VL |
848 | /* |
849 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
850 | * interested in events matching the pollfd->events mask, and the result | |
851 | * matching that mask is both recorded in pollfd->revents and returned. The | |
852 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
626cf236 | 853 | * if pwait->_qproc is non-NULL. |
4a4b69f7 | 854 | */ |
fb367937 | 855 | static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait, |
cbf55001 | 856 | bool *can_busy_poll, |
fb367937 | 857 | __poll_t busy_flag) |
1da177e4 | 858 | { |
a0f8dcfc CH |
859 | int fd = pollfd->fd; |
860 | __poll_t mask = 0, filter; | |
861 | struct fd f; | |
862 | ||
863 | if (fd < 0) | |
864 | goto out; | |
865 | mask = EPOLLNVAL; | |
866 | f = fdget(fd); | |
867 | if (!f.file) | |
868 | goto out; | |
869 | ||
870 | /* userland u16 ->events contains POLL... bitmap */ | |
871 | filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP; | |
9965ed17 CH |
872 | pwait->_key = filter | busy_flag; |
873 | mask = vfs_poll(f.file, pwait); | |
874 | if (mask & busy_flag) | |
875 | *can_busy_poll = true; | |
a0f8dcfc CH |
876 | mask &= filter; /* Mask out unneeded events. */ |
877 | fdput(f); | |
878 | ||
879 | out: | |
fb367937 | 880 | /* ... and so does ->revents */ |
c71d227f | 881 | pollfd->revents = mangle_poll(mask); |
4a4b69f7 | 882 | return mask; |
1da177e4 LT |
883 | } |
884 | ||
ccec5ee3 | 885 | static int do_poll(struct poll_list *list, struct poll_wqueues *wait, |
766b9f92 | 886 | struct timespec64 *end_time) |
1da177e4 | 887 | { |
1da177e4 | 888 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
889 | ktime_t expire, *to = NULL; |
890 | int timed_out = 0, count = 0; | |
da8b44d5 | 891 | u64 slack = 0; |
fb367937 | 892 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 893 | unsigned long busy_start = 0; |
1da177e4 | 894 | |
9f72949f | 895 | /* Optimise the no-wait case */ |
8ff3e8e8 | 896 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 897 | pt->_qproc = NULL; |
8ff3e8e8 AV |
898 | timed_out = 1; |
899 | } | |
9bf084f7 | 900 | |
96d2ab48 | 901 | if (end_time && !timed_out) |
231f3d39 | 902 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 903 | |
1da177e4 LT |
904 | for (;;) { |
905 | struct poll_list *walk; | |
cbf55001 | 906 | bool can_busy_loop = false; |
9f72949f | 907 | |
4a4b69f7 VL |
908 | for (walk = list; walk != NULL; walk = walk->next) { |
909 | struct pollfd * pfd, * pfd_end; | |
910 | ||
911 | pfd = walk->entries; | |
912 | pfd_end = pfd + walk->len; | |
913 | for (; pfd != pfd_end; pfd++) { | |
914 | /* | |
915 | * Fish for events. If we found one, record it | |
626cf236 | 916 | * and kill poll_table->_qproc, so we don't |
4a4b69f7 VL |
917 | * needlessly register any other waiters after |
918 | * this. They'll get immediately deregistered | |
919 | * when we break out and return. | |
920 | */ | |
cbf55001 ET |
921 | if (do_pollfd(pfd, pt, &can_busy_loop, |
922 | busy_flag)) { | |
4a4b69f7 | 923 | count++; |
626cf236 | 924 | pt->_qproc = NULL; |
cbf55001 ET |
925 | /* found something, stop busy polling */ |
926 | busy_flag = 0; | |
927 | can_busy_loop = false; | |
4a4b69f7 VL |
928 | } |
929 | } | |
1da177e4 | 930 | } |
4a4b69f7 VL |
931 | /* |
932 | * All waiters have already been registered, so don't provide | |
626cf236 | 933 | * a poll_table->_qproc to them on the next loop iteration. |
4a4b69f7 | 934 | */ |
626cf236 | 935 | pt->_qproc = NULL; |
9bf084f7 ON |
936 | if (!count) { |
937 | count = wait->error; | |
938 | if (signal_pending(current)) | |
8cf8b553 | 939 | count = -ERESTARTNOHAND; |
9bf084f7 | 940 | } |
8ff3e8e8 | 941 | if (count || timed_out) |
1da177e4 | 942 | break; |
9f72949f | 943 | |
cbf55001 | 944 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 | 945 | if (can_busy_loop && !need_resched()) { |
37056719 AD |
946 | if (!busy_start) { |
947 | busy_start = busy_loop_current_time(); | |
76b1e9b9 ET |
948 | continue; |
949 | } | |
37056719 | 950 | if (!busy_loop_timeout(busy_start)) |
76b1e9b9 ET |
951 | continue; |
952 | } | |
953 | busy_flag = 0; | |
91e2fd33 | 954 | |
8ff3e8e8 AV |
955 | /* |
956 | * If this is the first loop and we have a timeout | |
957 | * given, then we convert to ktime_t and set the to | |
958 | * pointer to the expiry value. | |
959 | */ | |
960 | if (end_time && !to) { | |
766b9f92 | 961 | expire = timespec64_to_ktime(*end_time); |
8ff3e8e8 | 962 | to = &expire; |
9f72949f DW |
963 | } |
964 | ||
5f820f64 | 965 | if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) |
8ff3e8e8 | 966 | timed_out = 1; |
1da177e4 | 967 | } |
1da177e4 LT |
968 | return count; |
969 | } | |
970 | ||
70674f95 AK |
971 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
972 | sizeof(struct pollfd)) | |
973 | ||
e99ca56c | 974 | static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
766b9f92 | 975 | struct timespec64 *end_time) |
1da177e4 LT |
976 | { |
977 | struct poll_wqueues table; | |
43e11fa2 | 978 | int err = -EFAULT, fdcount, len; |
30c14e40 JS |
979 | /* Allocate small arguments on the stack to save memory and be |
980 | faster - use long to make sure the buffer is aligned properly | |
981 | on 64 bit archs to avoid unaligned access */ | |
982 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
983 | struct poll_list *const head = (struct poll_list *)stack_pps; |
984 | struct poll_list *walk = head; | |
985 | unsigned long todo = nfds; | |
1da177e4 | 986 | |
d554ed89 | 987 | if (nfds > rlimit(RLIMIT_NOFILE)) |
1da177e4 LT |
988 | return -EINVAL; |
989 | ||
252e5725 ON |
990 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
991 | for (;;) { | |
992 | walk->next = NULL; | |
993 | walk->len = len; | |
994 | if (!len) | |
995 | break; | |
1da177e4 | 996 | |
252e5725 ON |
997 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
998 | sizeof(struct pollfd) * walk->len)) | |
999 | goto out_fds; | |
1000 | ||
1001 | todo -= walk->len; | |
1002 | if (!todo) | |
1003 | break; | |
1da177e4 | 1004 | |
252e5725 | 1005 | len = min(todo, POLLFD_PER_PAGE); |
43e11fa2 | 1006 | walk = walk->next = kmalloc(struct_size(walk, entries, len), |
0bcfe68b | 1007 | GFP_KERNEL); |
252e5725 ON |
1008 | if (!walk) { |
1009 | err = -ENOMEM; | |
1da177e4 LT |
1010 | goto out_fds; |
1011 | } | |
1da177e4 | 1012 | } |
9f72949f | 1013 | |
252e5725 | 1014 | poll_initwait(&table); |
ccec5ee3 | 1015 | fdcount = do_poll(head, &table, end_time); |
252e5725 | 1016 | poll_freewait(&table); |
1da177e4 | 1017 | |
ef0ba055 LT |
1018 | if (!user_write_access_begin(ufds, nfds * sizeof(*ufds))) |
1019 | goto out_fds; | |
1020 | ||
252e5725 | 1021 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
1022 | struct pollfd *fds = walk->entries; |
1023 | int j; | |
1024 | ||
ef0ba055 LT |
1025 | for (j = walk->len; j; fds++, ufds++, j--) |
1026 | unsafe_put_user(fds->revents, &ufds->revents, Efault); | |
1da177e4 | 1027 | } |
ef0ba055 | 1028 | user_write_access_end(); |
252e5725 | 1029 | |
1da177e4 | 1030 | err = fdcount; |
1da177e4 | 1031 | out_fds: |
252e5725 ON |
1032 | walk = head->next; |
1033 | while (walk) { | |
1034 | struct poll_list *pos = walk; | |
1035 | walk = walk->next; | |
1036 | kfree(pos); | |
1da177e4 | 1037 | } |
252e5725 | 1038 | |
1da177e4 | 1039 | return err; |
ef0ba055 LT |
1040 | |
1041 | Efault: | |
1042 | user_write_access_end(); | |
1043 | err = -EFAULT; | |
1044 | goto out_fds; | |
1da177e4 | 1045 | } |
9f72949f | 1046 | |
3075d9da CW |
1047 | static long do_restart_poll(struct restart_block *restart_block) |
1048 | { | |
8ff3e8e8 AV |
1049 | struct pollfd __user *ufds = restart_block->poll.ufds; |
1050 | int nfds = restart_block->poll.nfds; | |
766b9f92 | 1051 | struct timespec64 *to = NULL, end_time; |
3075d9da CW |
1052 | int ret; |
1053 | ||
8ff3e8e8 AV |
1054 | if (restart_block->poll.has_timeout) { |
1055 | end_time.tv_sec = restart_block->poll.tv_sec; | |
1056 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
1057 | to = &end_time; | |
1058 | } | |
1059 | ||
1060 | ret = do_sys_poll(ufds, nfds, to); | |
1061 | ||
5abbe51a ON |
1062 | if (ret == -ERESTARTNOHAND) |
1063 | ret = set_restart_fn(restart_block, do_restart_poll); | |
1064 | ||
3075d9da CW |
1065 | return ret; |
1066 | } | |
1067 | ||
5a8a82b1 | 1068 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
faf30900 | 1069 | int, timeout_msecs) |
9f72949f | 1070 | { |
766b9f92 | 1071 | struct timespec64 end_time, *to = NULL; |
3075d9da | 1072 | int ret; |
9f72949f | 1073 | |
8ff3e8e8 AV |
1074 | if (timeout_msecs >= 0) { |
1075 | to = &end_time; | |
1076 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
1077 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
1078 | } |
1079 | ||
8ff3e8e8 AV |
1080 | ret = do_sys_poll(ufds, nfds, to); |
1081 | ||
8cf8b553 | 1082 | if (ret == -ERESTARTNOHAND) { |
3075d9da | 1083 | struct restart_block *restart_block; |
8ff3e8e8 | 1084 | |
f56141e3 | 1085 | restart_block = ¤t->restart_block; |
8ff3e8e8 AV |
1086 | restart_block->poll.ufds = ufds; |
1087 | restart_block->poll.nfds = nfds; | |
1088 | ||
1089 | if (timeout_msecs >= 0) { | |
1090 | restart_block->poll.tv_sec = end_time.tv_sec; | |
1091 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
1092 | restart_block->poll.has_timeout = 1; | |
1093 | } else | |
1094 | restart_block->poll.has_timeout = 0; | |
1095 | ||
5abbe51a | 1096 | ret = set_restart_fn(restart_block, do_restart_poll); |
3075d9da CW |
1097 | } |
1098 | return ret; | |
9f72949f DW |
1099 | } |
1100 | ||
d4e82042 | 1101 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
8bd27a30 | 1102 | struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask, |
d4e82042 | 1103 | size_t, sigsetsize) |
9f72949f | 1104 | { |
36819ad0 | 1105 | struct timespec64 ts, end_time, *to = NULL; |
9f72949f DW |
1106 | int ret; |
1107 | ||
1108 | if (tsp) { | |
36819ad0 | 1109 | if (get_timespec64(&ts, tsp)) |
9f72949f DW |
1110 | return -EFAULT; |
1111 | ||
8ff3e8e8 AV |
1112 | to = &end_time; |
1113 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1114 | return -EINVAL; | |
9f72949f DW |
1115 | } |
1116 | ||
b772434b | 1117 | ret = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
1118 | if (ret) |
1119 | return ret; | |
9f72949f | 1120 | |
8ff3e8e8 | 1121 | ret = do_sys_poll(ufds, nfds, to); |
ac301020 | 1122 | return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret); |
9f72949f | 1123 | } |
e99ca56c | 1124 | |
8bd27a30 | 1125 | #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) |
e99ca56c | 1126 | |
8bd27a30 DD |
1127 | SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds, |
1128 | struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask, | |
1129 | size_t, sigsetsize) | |
e99ca56c | 1130 | { |
8bd27a30 DD |
1131 | struct timespec64 ts, end_time, *to = NULL; |
1132 | int ret; | |
e99ca56c | 1133 | |
8bd27a30 DD |
1134 | if (tsp) { |
1135 | if (get_old_timespec32(&ts, tsp)) | |
1136 | return -EFAULT; | |
e99ca56c | 1137 | |
8bd27a30 DD |
1138 | to = &end_time; |
1139 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1140 | return -EINVAL; | |
1141 | } | |
e99ca56c | 1142 | |
b772434b | 1143 | ret = set_user_sigmask(sigmask, sigsetsize); |
8bd27a30 | 1144 | if (ret) |
e99ca56c AV |
1145 | return ret; |
1146 | ||
8bd27a30 | 1147 | ret = do_sys_poll(ufds, nfds, to); |
ac301020 | 1148 | return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret); |
e99ca56c | 1149 | } |
8bd27a30 DD |
1150 | #endif |
1151 | ||
1152 | #ifdef CONFIG_COMPAT | |
1153 | #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t)) | |
e99ca56c AV |
1154 | |
1155 | /* | |
1156 | * Ooo, nasty. We need here to frob 32-bit unsigned longs to | |
1157 | * 64-bit unsigned longs. | |
1158 | */ | |
1159 | static | |
1160 | int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, | |
1161 | unsigned long *fdset) | |
1162 | { | |
e99ca56c | 1163 | if (ufdset) { |
464d6242 | 1164 | return compat_get_bitmap(fdset, ufdset, nr); |
e99ca56c | 1165 | } else { |
79de3cbe | 1166 | zero_fd_set(nr, fdset); |
464d6242 | 1167 | return 0; |
e99ca56c | 1168 | } |
e99ca56c AV |
1169 | } |
1170 | ||
1171 | static | |
1172 | int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, | |
1173 | unsigned long *fdset) | |
1174 | { | |
e99ca56c AV |
1175 | if (!ufdset) |
1176 | return 0; | |
464d6242 | 1177 | return compat_put_bitmap(ufdset, fdset, nr); |
e99ca56c AV |
1178 | } |
1179 | ||
1180 | ||
1181 | /* | |
1182 | * This is a virtual copy of sys_select from fs/select.c and probably | |
1183 | * should be compared to it from time to time | |
1184 | */ | |
1185 | ||
1186 | /* | |
1187 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
1188 | * like to be certain this leads to no problems. So I return | |
1189 | * EINTR just for safety. | |
1190 | * | |
1191 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
1192 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
1193 | */ | |
1194 | static int compat_core_sys_select(int n, compat_ulong_t __user *inp, | |
1195 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
36819ad0 | 1196 | struct timespec64 *end_time) |
e99ca56c AV |
1197 | { |
1198 | fd_set_bits fds; | |
1199 | void *bits; | |
1200 | int size, max_fds, ret = -EINVAL; | |
1201 | struct fdtable *fdt; | |
1202 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; | |
1203 | ||
1204 | if (n < 0) | |
1205 | goto out_nofds; | |
1206 | ||
1207 | /* max_fds can increase, so grab it once to avoid race */ | |
1208 | rcu_read_lock(); | |
1209 | fdt = files_fdtable(current->files); | |
1210 | max_fds = fdt->max_fds; | |
1211 | rcu_read_unlock(); | |
1212 | if (n > max_fds) | |
1213 | n = max_fds; | |
1214 | ||
1215 | /* | |
1216 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
1217 | * since we used fdset we need to allocate memory in units of | |
1218 | * long-words. | |
1219 | */ | |
1220 | size = FDS_BYTES(n); | |
1221 | bits = stack_fds; | |
1222 | if (size > sizeof(stack_fds) / 6) { | |
6da2ec56 | 1223 | bits = kmalloc_array(6, size, GFP_KERNEL); |
e99ca56c AV |
1224 | ret = -ENOMEM; |
1225 | if (!bits) | |
1226 | goto out_nofds; | |
1227 | } | |
1228 | fds.in = (unsigned long *) bits; | |
1229 | fds.out = (unsigned long *) (bits + size); | |
1230 | fds.ex = (unsigned long *) (bits + 2*size); | |
1231 | fds.res_in = (unsigned long *) (bits + 3*size); | |
1232 | fds.res_out = (unsigned long *) (bits + 4*size); | |
1233 | fds.res_ex = (unsigned long *) (bits + 5*size); | |
1234 | ||
1235 | if ((ret = compat_get_fd_set(n, inp, fds.in)) || | |
1236 | (ret = compat_get_fd_set(n, outp, fds.out)) || | |
1237 | (ret = compat_get_fd_set(n, exp, fds.ex))) | |
1238 | goto out; | |
1239 | zero_fd_set(n, fds.res_in); | |
1240 | zero_fd_set(n, fds.res_out); | |
1241 | zero_fd_set(n, fds.res_ex); | |
1242 | ||
1243 | ret = do_select(n, &fds, end_time); | |
1244 | ||
1245 | if (ret < 0) | |
1246 | goto out; | |
1247 | if (!ret) { | |
1248 | ret = -ERESTARTNOHAND; | |
1249 | if (signal_pending(current)) | |
1250 | goto out; | |
1251 | ret = 0; | |
1252 | } | |
1253 | ||
1254 | if (compat_set_fd_set(n, inp, fds.res_in) || | |
1255 | compat_set_fd_set(n, outp, fds.res_out) || | |
1256 | compat_set_fd_set(n, exp, fds.res_ex)) | |
1257 | ret = -EFAULT; | |
1258 | out: | |
1259 | if (bits != stack_fds) | |
1260 | kfree(bits); | |
1261 | out_nofds: | |
1262 | return ret; | |
1263 | } | |
1264 | ||
05585e44 DB |
1265 | static int do_compat_select(int n, compat_ulong_t __user *inp, |
1266 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
9afc5eee | 1267 | struct old_timeval32 __user *tvp) |
e99ca56c | 1268 | { |
36819ad0 | 1269 | struct timespec64 end_time, *to = NULL; |
9afc5eee | 1270 | struct old_timeval32 tv; |
e99ca56c AV |
1271 | int ret; |
1272 | ||
1273 | if (tvp) { | |
1274 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
1275 | return -EFAULT; | |
1276 | ||
1277 | to = &end_time; | |
1278 | if (poll_select_set_timeout(to, | |
1279 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
1280 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
1281 | return -EINVAL; | |
1282 | } | |
1283 | ||
1284 | ret = compat_core_sys_select(n, inp, outp, exp, to); | |
ac301020 | 1285 | return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret); |
e99ca56c AV |
1286 | } |
1287 | ||
05585e44 DB |
1288 | COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp, |
1289 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
9afc5eee | 1290 | struct old_timeval32 __user *, tvp) |
05585e44 DB |
1291 | { |
1292 | return do_compat_select(n, inp, outp, exp, tvp); | |
1293 | } | |
1294 | ||
e99ca56c AV |
1295 | struct compat_sel_arg_struct { |
1296 | compat_ulong_t n; | |
1297 | compat_uptr_t inp; | |
1298 | compat_uptr_t outp; | |
1299 | compat_uptr_t exp; | |
1300 | compat_uptr_t tvp; | |
1301 | }; | |
1302 | ||
1303 | COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg) | |
1304 | { | |
1305 | struct compat_sel_arg_struct a; | |
1306 | ||
1307 | if (copy_from_user(&a, arg, sizeof(a))) | |
1308 | return -EFAULT; | |
05585e44 DB |
1309 | return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp), |
1310 | compat_ptr(a.exp), compat_ptr(a.tvp)); | |
e99ca56c AV |
1311 | } |
1312 | ||
1313 | static long do_compat_pselect(int n, compat_ulong_t __user *inp, | |
1314 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
e024707b DD |
1315 | void __user *tsp, compat_sigset_t __user *sigmask, |
1316 | compat_size_t sigsetsize, enum poll_time_type type) | |
e99ca56c | 1317 | { |
36819ad0 | 1318 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1319 | int ret; |
1320 | ||
1321 | if (tsp) { | |
e024707b DD |
1322 | switch (type) { |
1323 | case PT_OLD_TIMESPEC: | |
1324 | if (get_old_timespec32(&ts, tsp)) | |
1325 | return -EFAULT; | |
1326 | break; | |
1327 | case PT_TIMESPEC: | |
1328 | if (get_timespec64(&ts, tsp)) | |
1329 | return -EFAULT; | |
1330 | break; | |
1331 | default: | |
1332 | BUG(); | |
1333 | } | |
e99ca56c AV |
1334 | |
1335 | to = &end_time; | |
1336 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1337 | return -EINVAL; | |
1338 | } | |
1339 | ||
b772434b | 1340 | ret = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
1341 | if (ret) |
1342 | return ret; | |
e99ca56c AV |
1343 | |
1344 | ret = compat_core_sys_select(n, inp, outp, exp, to); | |
ac301020 | 1345 | return poll_select_finish(&end_time, tsp, type, ret); |
e99ca56c AV |
1346 | } |
1347 | ||
7e71609f AV |
1348 | struct compat_sigset_argpack { |
1349 | compat_uptr_t p; | |
1350 | compat_size_t size; | |
1351 | }; | |
1352 | static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to, | |
1353 | struct compat_sigset_argpack __user *from) | |
1354 | { | |
1355 | if (from) { | |
1356 | if (!user_read_access_begin(from, sizeof(*from))) | |
1357 | return -EFAULT; | |
1358 | unsafe_get_user(to->p, &from->p, Efault); | |
1359 | unsafe_get_user(to->size, &from->size, Efault); | |
1360 | user_read_access_end(); | |
1361 | } | |
1362 | return 0; | |
1363 | Efault: | |
1364 | user_access_end(); | |
1365 | return -EFAULT; | |
1366 | } | |
1367 | ||
e024707b DD |
1368 | COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp, |
1369 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
1370 | struct __kernel_timespec __user *, tsp, void __user *, sig) | |
1371 | { | |
7e71609f AV |
1372 | struct compat_sigset_argpack x = {0, 0}; |
1373 | ||
1374 | if (get_compat_sigset_argpack(&x, sig)) | |
1375 | return -EFAULT; | |
e024707b | 1376 | |
7e71609f AV |
1377 | return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p), |
1378 | x.size, PT_TIMESPEC); | |
e024707b DD |
1379 | } |
1380 | ||
1381 | #if defined(CONFIG_COMPAT_32BIT_TIME) | |
1382 | ||
8dabe724 | 1383 | COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp, |
e99ca56c | 1384 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, |
9afc5eee | 1385 | struct old_timespec32 __user *, tsp, void __user *, sig) |
e99ca56c | 1386 | { |
7e71609f AV |
1387 | struct compat_sigset_argpack x = {0, 0}; |
1388 | ||
1389 | if (get_compat_sigset_argpack(&x, sig)) | |
1390 | return -EFAULT; | |
e024707b | 1391 | |
7e71609f AV |
1392 | return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p), |
1393 | x.size, PT_OLD_TIMESPEC); | |
e99ca56c AV |
1394 | } |
1395 | ||
e024707b DD |
1396 | #endif |
1397 | ||
8bd27a30 | 1398 | #if defined(CONFIG_COMPAT_32BIT_TIME) |
8dabe724 | 1399 | COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, |
9afc5eee | 1400 | unsigned int, nfds, struct old_timespec32 __user *, tsp, |
e99ca56c AV |
1401 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) |
1402 | { | |
36819ad0 | 1403 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1404 | int ret; |
1405 | ||
1406 | if (tsp) { | |
9afc5eee | 1407 | if (get_old_timespec32(&ts, tsp)) |
e99ca56c AV |
1408 | return -EFAULT; |
1409 | ||
1410 | to = &end_time; | |
1411 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1412 | return -EINVAL; | |
1413 | } | |
1414 | ||
b772434b | 1415 | ret = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
1416 | if (ret) |
1417 | return ret; | |
e99ca56c AV |
1418 | |
1419 | ret = do_sys_poll(ufds, nfds, to); | |
ac301020 | 1420 | return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret); |
e99ca56c AV |
1421 | } |
1422 | #endif | |
8bd27a30 DD |
1423 | |
1424 | /* New compat syscall for 64 bit time_t*/ | |
1425 | COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds, | |
1426 | unsigned int, nfds, struct __kernel_timespec __user *, tsp, | |
1427 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) | |
1428 | { | |
8bd27a30 DD |
1429 | struct timespec64 ts, end_time, *to = NULL; |
1430 | int ret; | |
1431 | ||
1432 | if (tsp) { | |
1433 | if (get_timespec64(&ts, tsp)) | |
1434 | return -EFAULT; | |
1435 | ||
1436 | to = &end_time; | |
1437 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1438 | return -EINVAL; | |
1439 | } | |
1440 | ||
b772434b | 1441 | ret = set_compat_user_sigmask(sigmask, sigsetsize); |
8bd27a30 DD |
1442 | if (ret) |
1443 | return ret; | |
1444 | ||
1445 | ret = do_sys_poll(ufds, nfds, to); | |
ac301020 | 1446 | return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret); |
8bd27a30 DD |
1447 | } |
1448 | ||
1449 | #endif |