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; |
ed4fb6d7 | 80 | u64 slack = current->timer_slack_ns; |
90d6e24a | 81 | |
ed4fb6d7 | 82 | if (slack == 0) |
90d6e24a AV |
83 | return 0; |
84 | ||
766b9f92 DD |
85 | ktime_get_ts64(&now); |
86 | now = timespec64_sub(*tv, now); | |
90d6e24a | 87 | ret = __estimate_accuracy(&now); |
ed4fb6d7 FM |
88 | if (ret < slack) |
89 | return slack; | |
90d6e24a AV |
90 | return ret; |
91 | } | |
92 | ||
93 | ||
94 | ||
1da177e4 LT |
95 | struct poll_table_page { |
96 | struct poll_table_page * next; | |
97 | struct poll_table_entry * entry; | |
5e01fdff | 98 | struct poll_table_entry entries[]; |
1da177e4 LT |
99 | }; |
100 | ||
101 | #define POLL_TABLE_FULL(table) \ | |
102 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
103 | ||
104 | /* | |
105 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
106 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
107 | * follow, but it should be free of race-conditions, and it's practical. If you | |
108 | * understand what I'm doing here, then you understand how the linux | |
109 | * sleep/wakeup mechanism works. | |
110 | * | |
111 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
112 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
113 | * as all select/poll functions have to call it to add an entry to the | |
114 | * poll table. | |
115 | */ | |
75c96f85 AB |
116 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
117 | poll_table *p); | |
1da177e4 LT |
118 | |
119 | void poll_initwait(struct poll_wqueues *pwq) | |
120 | { | |
121 | init_poll_funcptr(&pwq->pt, __pollwait); | |
5f820f64 | 122 | pwq->polling_task = current; |
b2add73d | 123 | pwq->triggered = 0; |
1da177e4 LT |
124 | pwq->error = 0; |
125 | pwq->table = NULL; | |
70674f95 | 126 | pwq->inline_index = 0; |
1da177e4 | 127 | } |
1da177e4 LT |
128 | EXPORT_SYMBOL(poll_initwait); |
129 | ||
70674f95 AK |
130 | static void free_poll_entry(struct poll_table_entry *entry) |
131 | { | |
ccf6780d | 132 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
133 | fput(entry->filp); |
134 | } | |
135 | ||
1da177e4 LT |
136 | void poll_freewait(struct poll_wqueues *pwq) |
137 | { | |
138 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
139 | int i; |
140 | for (i = 0; i < pwq->inline_index; i++) | |
141 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
142 | while (p) { |
143 | struct poll_table_entry * entry; | |
144 | struct poll_table_page *old; | |
145 | ||
146 | entry = p->entry; | |
147 | do { | |
148 | entry--; | |
70674f95 | 149 | free_poll_entry(entry); |
1da177e4 LT |
150 | } while (entry > p->entries); |
151 | old = p; | |
152 | p = p->next; | |
153 | free_page((unsigned long) old); | |
154 | } | |
155 | } | |
1da177e4 LT |
156 | EXPORT_SYMBOL(poll_freewait); |
157 | ||
5f820f64 | 158 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
1da177e4 | 159 | { |
1da177e4 LT |
160 | struct poll_table_page *table = p->table; |
161 | ||
70674f95 AK |
162 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
163 | return p->inline_entries + p->inline_index++; | |
164 | ||
1da177e4 LT |
165 | if (!table || POLL_TABLE_FULL(table)) { |
166 | struct poll_table_page *new_table; | |
167 | ||
168 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
169 | if (!new_table) { | |
170 | p->error = -ENOMEM; | |
70674f95 | 171 | return NULL; |
1da177e4 LT |
172 | } |
173 | new_table->entry = new_table->entries; | |
174 | new_table->next = table; | |
175 | p->table = new_table; | |
176 | table = new_table; | |
177 | } | |
178 | ||
70674f95 AK |
179 | return table->entry++; |
180 | } | |
181 | ||
ac6424b9 | 182 | static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
5f820f64 TH |
183 | { |
184 | struct poll_wqueues *pwq = wait->private; | |
185 | DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); | |
186 | ||
187 | /* | |
188 | * Although this function is called under waitqueue lock, LOCK | |
189 | * doesn't imply write barrier and the users expect write | |
190 | * barrier semantics on wakeup functions. The following | |
191 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
b92b8b35 | 192 | * and is paired with smp_store_mb() in poll_schedule_timeout. |
5f820f64 TH |
193 | */ |
194 | smp_wmb(); | |
195 | pwq->triggered = 1; | |
196 | ||
197 | /* | |
198 | * Perform the default wake up operation using a dummy | |
199 | * waitqueue. | |
200 | * | |
201 | * TODO: This is hacky but there currently is no interface to | |
202 | * pass in @sync. @sync is scheduled to be removed and once | |
203 | * that happens, wake_up_process() can be used directly. | |
204 | */ | |
205 | return default_wake_function(&dummy_wait, mode, sync, key); | |
206 | } | |
207 | ||
ac6424b9 | 208 | static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
4938d7e0 ED |
209 | { |
210 | struct poll_table_entry *entry; | |
211 | ||
212 | entry = container_of(wait, struct poll_table_entry, wait); | |
3ad6f93e | 213 | if (key && !(key_to_poll(key) & entry->key)) |
4938d7e0 ED |
214 | return 0; |
215 | return __pollwake(wait, mode, sync, key); | |
216 | } | |
217 | ||
70674f95 AK |
218 | /* Add a new entry */ |
219 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
220 | poll_table *p) | |
221 | { | |
5f820f64 TH |
222 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
223 | struct poll_table_entry *entry = poll_get_entry(pwq); | |
70674f95 AK |
224 | if (!entry) |
225 | return; | |
cb0942b8 | 226 | entry->filp = get_file(filp); |
70674f95 | 227 | entry->wait_address = wait_address; |
626cf236 | 228 | entry->key = p->_key; |
5f820f64 TH |
229 | init_waitqueue_func_entry(&entry->wait, pollwake); |
230 | entry->wait.private = pwq; | |
ccf6780d | 231 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
232 | } |
233 | ||
8f546ae1 | 234 | static int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
5f820f64 TH |
235 | ktime_t *expires, unsigned long slack) |
236 | { | |
237 | int rc = -EINTR; | |
238 | ||
239 | set_current_state(state); | |
240 | if (!pwq->triggered) | |
59612d18 | 241 | rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); |
5f820f64 TH |
242 | __set_current_state(TASK_RUNNING); |
243 | ||
244 | /* | |
245 | * Prepare for the next iteration. | |
246 | * | |
b92b8b35 | 247 | * The following smp_store_mb() serves two purposes. First, it's |
5f820f64 TH |
248 | * the counterpart rmb of the wmb in pollwake() such that data |
249 | * written before wake up is always visible after wake up. | |
250 | * Second, the full barrier guarantees that triggered clearing | |
251 | * doesn't pass event check of the next iteration. Note that | |
252 | * this problem doesn't exist for the first iteration as | |
253 | * add_wait_queue() has full barrier semantics. | |
254 | */ | |
b92b8b35 | 255 | smp_store_mb(pwq->triggered, 0); |
5f820f64 TH |
256 | |
257 | return rc; | |
258 | } | |
5f820f64 | 259 | |
b773ad40 TG |
260 | /** |
261 | * poll_select_set_timeout - helper function to setup the timeout value | |
766b9f92 | 262 | * @to: pointer to timespec64 variable for the final timeout |
b773ad40 TG |
263 | * @sec: seconds (from user space) |
264 | * @nsec: nanoseconds (from user space) | |
265 | * | |
266 | * Note, we do not use a timespec for the user space value here, That | |
267 | * way we can use the function for timeval and compat interfaces as well. | |
268 | * | |
269 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
270 | */ | |
766b9f92 | 271 | int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec) |
b773ad40 | 272 | { |
766b9f92 | 273 | struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec}; |
b773ad40 | 274 | |
766b9f92 | 275 | if (!timespec64_valid(&ts)) |
b773ad40 TG |
276 | return -EINVAL; |
277 | ||
278 | /* Optimize for the zero timeout value here */ | |
279 | if (!sec && !nsec) { | |
280 | to->tv_sec = to->tv_nsec = 0; | |
281 | } else { | |
766b9f92 DD |
282 | ktime_get_ts64(to); |
283 | *to = timespec64_add_safe(*to, ts); | |
b773ad40 TG |
284 | } |
285 | return 0; | |
286 | } | |
287 | ||
8bd27a30 DD |
288 | enum poll_time_type { |
289 | PT_TIMEVAL = 0, | |
290 | PT_OLD_TIMEVAL = 1, | |
291 | PT_TIMESPEC = 2, | |
292 | PT_OLD_TIMESPEC = 3, | |
293 | }; | |
294 | ||
ac301020 ON |
295 | static int poll_select_finish(struct timespec64 *end_time, |
296 | void __user *p, | |
297 | enum poll_time_type pt_type, int ret) | |
b773ad40 | 298 | { |
36819ad0 | 299 | struct timespec64 rts; |
b773ad40 | 300 | |
ac301020 ON |
301 | restore_saved_sigmask_unless(ret == -ERESTARTNOHAND); |
302 | ||
b773ad40 TG |
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 | { | |
75d319c0 | 322 | struct __kernel_old_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 | /* | |
e99ca56c AV |
384 | * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned. |
385 | */ | |
386 | static inline | |
387 | int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) | |
388 | { | |
389 | nr = FDS_BYTES(nr); | |
390 | if (ufdset) | |
391 | return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0; | |
392 | ||
393 | memset(fdset, 0, nr); | |
394 | return 0; | |
395 | } | |
396 | ||
397 | static inline unsigned long __must_check | |
398 | set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) | |
399 | { | |
400 | if (ufdset) | |
401 | return __copy_to_user(ufdset, fdset, FDS_BYTES(nr)); | |
402 | return 0; | |
403 | } | |
404 | ||
405 | static inline | |
406 | void zero_fd_set(unsigned long nr, unsigned long *fdset) | |
407 | { | |
408 | memset(fdset, 0, FDS_BYTES(nr)); | |
409 | } | |
410 | ||
1da177e4 LT |
411 | #define FDS_IN(fds, n) (fds->in + n) |
412 | #define FDS_OUT(fds, n) (fds->out + n) | |
413 | #define FDS_EX(fds, n) (fds->ex + n) | |
414 | ||
415 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
416 | ||
417 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
418 | { | |
419 | unsigned long *open_fds; | |
420 | unsigned long set; | |
421 | int max; | |
badf1662 | 422 | struct fdtable *fdt; |
1da177e4 LT |
423 | |
424 | /* handle last in-complete long-word first */ | |
8ded2bbc JB |
425 | set = ~(~0UL << (n & (BITS_PER_LONG-1))); |
426 | n /= BITS_PER_LONG; | |
badf1662 | 427 | fdt = files_fdtable(current->files); |
1fd36adc | 428 | open_fds = fdt->open_fds + n; |
1da177e4 LT |
429 | max = 0; |
430 | if (set) { | |
431 | set &= BITS(fds, n); | |
432 | if (set) { | |
433 | if (!(set & ~*open_fds)) | |
434 | goto get_max; | |
435 | return -EBADF; | |
436 | } | |
437 | } | |
438 | while (n) { | |
439 | open_fds--; | |
440 | n--; | |
441 | set = BITS(fds, n); | |
442 | if (!set) | |
443 | continue; | |
444 | if (set & ~*open_fds) | |
445 | return -EBADF; | |
446 | if (max) | |
447 | continue; | |
448 | get_max: | |
449 | do { | |
450 | max++; | |
451 | set >>= 1; | |
452 | } while (set); | |
8ded2bbc | 453 | max += n * BITS_PER_LONG; |
1da177e4 LT |
454 | } |
455 | ||
456 | return max; | |
457 | } | |
458 | ||
68514dac JK |
459 | #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR |\ |
460 | EPOLLNVAL) | |
461 | #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\ | |
462 | EPOLLNVAL) | |
463 | #define POLLEX_SET (EPOLLPRI | EPOLLNVAL) | |
1da177e4 | 464 | |
4938d7e0 | 465 | static inline void wait_key_set(poll_table *wait, unsigned long in, |
2d48d67f | 466 | unsigned long out, unsigned long bit, |
01699437 | 467 | __poll_t ll_flag) |
4938d7e0 | 468 | { |
2d48d67f | 469 | wait->_key = POLLEX_SET | ll_flag; |
626cf236 HV |
470 | if (in & bit) |
471 | wait->_key |= POLLIN_SET; | |
472 | if (out & bit) | |
473 | wait->_key |= POLLOUT_SET; | |
4938d7e0 ED |
474 | } |
475 | ||
ddb9fd7a | 476 | static noinline_for_stack int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time) |
1da177e4 | 477 | { |
8ff3e8e8 | 478 | ktime_t expire, *to = NULL; |
1da177e4 LT |
479 | struct poll_wqueues table; |
480 | poll_table *wait; | |
8ff3e8e8 | 481 | int retval, i, timed_out = 0; |
da8b44d5 | 482 | u64 slack = 0; |
01699437 | 483 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 484 | unsigned long busy_start = 0; |
1da177e4 | 485 | |
b835996f | 486 | rcu_read_lock(); |
1da177e4 | 487 | retval = max_select_fd(n, fds); |
b835996f | 488 | rcu_read_unlock(); |
1da177e4 LT |
489 | |
490 | if (retval < 0) | |
491 | return retval; | |
492 | n = retval; | |
493 | ||
494 | poll_initwait(&table); | |
495 | wait = &table.pt; | |
8ff3e8e8 | 496 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 497 | wait->_qproc = NULL; |
8ff3e8e8 AV |
498 | timed_out = 1; |
499 | } | |
500 | ||
96d2ab48 | 501 | if (end_time && !timed_out) |
231f3d39 | 502 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 503 | |
1da177e4 LT |
504 | retval = 0; |
505 | for (;;) { | |
506 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
cbf55001 | 507 | bool can_busy_loop = false; |
1da177e4 | 508 | |
1da177e4 LT |
509 | inp = fds->in; outp = fds->out; exp = fds->ex; |
510 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
511 | ||
512 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
e6c8adca | 513 | unsigned long in, out, ex, all_bits, bit = 1, j; |
1da177e4 | 514 | unsigned long res_in = 0, res_out = 0, res_ex = 0; |
e6c8adca | 515 | __poll_t mask; |
1da177e4 LT |
516 | |
517 | in = *inp++; out = *outp++; ex = *exp++; | |
518 | all_bits = in | out | ex; | |
519 | if (all_bits == 0) { | |
8ded2bbc | 520 | i += BITS_PER_LONG; |
1da177e4 LT |
521 | continue; |
522 | } | |
523 | ||
8ded2bbc | 524 | for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { |
2903ff01 | 525 | struct fd f; |
1da177e4 LT |
526 | if (i >= n) |
527 | break; | |
528 | if (!(bit & all_bits)) | |
529 | continue; | |
68514dac | 530 | mask = EPOLLNVAL; |
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 | 538 | } |
68514dac JK |
539 | if ((mask & POLLIN_SET) && (in & bit)) { |
540 | res_in |= bit; | |
541 | retval++; | |
542 | wait->_qproc = NULL; | |
543 | } | |
544 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
545 | res_out |= bit; | |
546 | retval++; | |
547 | wait->_qproc = NULL; | |
548 | } | |
549 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
550 | res_ex |= bit; | |
551 | retval++; | |
552 | wait->_qproc = NULL; | |
553 | } | |
554 | /* got something, stop busy polling */ | |
555 | if (retval) { | |
556 | can_busy_loop = false; | |
557 | busy_flag = 0; | |
558 | ||
559 | /* | |
560 | * only remember a returned | |
561 | * POLL_BUSY_LOOP if we asked for it | |
562 | */ | |
563 | } else if (busy_flag & mask) | |
564 | can_busy_loop = true; | |
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; | |
0bcfe68b | 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 | 701 | static int kern_select(int n, fd_set __user *inp, fd_set __user *outp, |
75d319c0 | 702 | fd_set __user *exp, struct __kernel_old_timeval __user *tvp) |
9f72949f | 703 | { |
766b9f92 | 704 | struct timespec64 end_time, *to = NULL; |
75d319c0 | 705 | struct __kernel_old_timeval tv; |
9f72949f DW |
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); |
ac301020 | 720 | return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret); |
9f72949f DW |
721 | } |
722 | ||
4bdb9aca | 723 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
75d319c0 | 724 | fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp) |
4bdb9aca DB |
725 | { |
726 | return kern_select(n, inp, outp, exp, tvp); | |
727 | } | |
728 | ||
c9da9f21 | 729 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
e024707b DD |
730 | fd_set __user *exp, void __user *tsp, |
731 | const sigset_t __user *sigmask, size_t sigsetsize, | |
732 | enum poll_time_type type) | |
9f72949f | 733 | { |
36819ad0 | 734 | struct timespec64 ts, end_time, *to = NULL; |
9f72949f DW |
735 | int ret; |
736 | ||
737 | if (tsp) { | |
e024707b DD |
738 | switch (type) { |
739 | case PT_TIMESPEC: | |
740 | if (get_timespec64(&ts, tsp)) | |
741 | return -EFAULT; | |
742 | break; | |
743 | case PT_OLD_TIMESPEC: | |
744 | if (get_old_timespec32(&ts, tsp)) | |
745 | return -EFAULT; | |
746 | break; | |
747 | default: | |
748 | BUG(); | |
749 | } | |
9f72949f | 750 | |
8ff3e8e8 | 751 | to = &end_time; |
36819ad0 | 752 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) |
9f72949f | 753 | return -EINVAL; |
9f72949f DW |
754 | } |
755 | ||
b772434b | 756 | ret = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
757 | if (ret) |
758 | return ret; | |
9f72949f | 759 | |
62568510 | 760 | ret = core_sys_select(n, inp, outp, exp, to); |
ac301020 | 761 | return poll_select_finish(&end_time, tsp, type, ret); |
9f72949f DW |
762 | } |
763 | ||
764 | /* | |
765 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
766 | * 6-argument version where the sixth argument is a pointer to a structure | |
767 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
768 | * the sigset size. | |
769 | */ | |
7e71609f AV |
770 | struct sigset_argpack { |
771 | sigset_t __user *p; | |
772 | size_t size; | |
773 | }; | |
774 | ||
775 | static inline int get_sigset_argpack(struct sigset_argpack *to, | |
776 | struct sigset_argpack __user *from) | |
777 | { | |
778 | // the path is hot enough for overhead of copy_from_user() to matter | |
779 | if (from) { | |
780 | if (!user_read_access_begin(from, sizeof(*from))) | |
781 | return -EFAULT; | |
782 | unsafe_get_user(to->p, &from->p, Efault); | |
783 | unsafe_get_user(to->size, &from->size, Efault); | |
784 | user_read_access_end(); | |
785 | } | |
786 | return 0; | |
787 | Efault: | |
788 | user_access_end(); | |
789 | return -EFAULT; | |
790 | } | |
791 | ||
d4e82042 | 792 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
e024707b DD |
793 | fd_set __user *, exp, struct __kernel_timespec __user *, tsp, |
794 | void __user *, sig) | |
795 | { | |
7e71609f AV |
796 | struct sigset_argpack x = {NULL, 0}; |
797 | ||
798 | if (get_sigset_argpack(&x, sig)) | |
799 | return -EFAULT; | |
e024707b | 800 | |
7e71609f | 801 | return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC); |
e024707b DD |
802 | } |
803 | ||
804 | #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) | |
805 | ||
806 | SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp, | |
807 | fd_set __user *, exp, struct old_timespec32 __user *, tsp, | |
d4e82042 | 808 | void __user *, sig) |
9f72949f | 809 | { |
7e71609f AV |
810 | struct sigset_argpack x = {NULL, 0}; |
811 | ||
812 | if (get_sigset_argpack(&x, sig)) | |
813 | return -EFAULT; | |
9f72949f | 814 | |
7e71609f | 815 | return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC); |
9f72949f | 816 | } |
9f72949f | 817 | |
e024707b DD |
818 | #endif |
819 | ||
5d0e5283 CH |
820 | #ifdef __ARCH_WANT_SYS_OLD_SELECT |
821 | struct sel_arg_struct { | |
822 | unsigned long n; | |
823 | fd_set __user *inp, *outp, *exp; | |
75d319c0 | 824 | struct __kernel_old_timeval __user *tvp; |
5d0e5283 CH |
825 | }; |
826 | ||
827 | SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) | |
828 | { | |
829 | struct sel_arg_struct a; | |
830 | ||
831 | if (copy_from_user(&a, arg, sizeof(a))) | |
832 | return -EFAULT; | |
4bdb9aca | 833 | return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp); |
5d0e5283 CH |
834 | } |
835 | #endif | |
836 | ||
1da177e4 LT |
837 | struct poll_list { |
838 | struct poll_list *next; | |
c67ef897 | 839 | unsigned int len; |
193b7279 | 840 | struct pollfd entries[] __counted_by(len); |
1da177e4 LT |
841 | }; |
842 | ||
843 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
844 | ||
4a4b69f7 VL |
845 | /* |
846 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
847 | * interested in events matching the pollfd->events mask, and the result | |
848 | * matching that mask is both recorded in pollfd->revents and returned. The | |
849 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
626cf236 | 850 | * if pwait->_qproc is non-NULL. |
4a4b69f7 | 851 | */ |
fb367937 | 852 | static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait, |
cbf55001 | 853 | bool *can_busy_poll, |
fb367937 | 854 | __poll_t busy_flag) |
1da177e4 | 855 | { |
a0f8dcfc CH |
856 | int fd = pollfd->fd; |
857 | __poll_t mask = 0, filter; | |
858 | struct fd f; | |
859 | ||
860 | if (fd < 0) | |
861 | goto out; | |
862 | mask = EPOLLNVAL; | |
863 | f = fdget(fd); | |
864 | if (!f.file) | |
865 | goto out; | |
866 | ||
867 | /* userland u16 ->events contains POLL... bitmap */ | |
868 | filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP; | |
9965ed17 CH |
869 | pwait->_key = filter | busy_flag; |
870 | mask = vfs_poll(f.file, pwait); | |
871 | if (mask & busy_flag) | |
872 | *can_busy_poll = true; | |
a0f8dcfc CH |
873 | mask &= filter; /* Mask out unneeded events. */ |
874 | fdput(f); | |
875 | ||
876 | out: | |
fb367937 | 877 | /* ... and so does ->revents */ |
c71d227f | 878 | pollfd->revents = mangle_poll(mask); |
4a4b69f7 | 879 | return mask; |
1da177e4 LT |
880 | } |
881 | ||
ccec5ee3 | 882 | static int do_poll(struct poll_list *list, struct poll_wqueues *wait, |
766b9f92 | 883 | struct timespec64 *end_time) |
1da177e4 | 884 | { |
1da177e4 | 885 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
886 | ktime_t expire, *to = NULL; |
887 | int timed_out = 0, count = 0; | |
da8b44d5 | 888 | u64 slack = 0; |
fb367937 | 889 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 890 | unsigned long busy_start = 0; |
1da177e4 | 891 | |
9f72949f | 892 | /* Optimise the no-wait case */ |
8ff3e8e8 | 893 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 894 | pt->_qproc = NULL; |
8ff3e8e8 AV |
895 | timed_out = 1; |
896 | } | |
9bf084f7 | 897 | |
96d2ab48 | 898 | if (end_time && !timed_out) |
231f3d39 | 899 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 900 | |
1da177e4 LT |
901 | for (;;) { |
902 | struct poll_list *walk; | |
cbf55001 | 903 | bool can_busy_loop = false; |
9f72949f | 904 | |
4a4b69f7 VL |
905 | for (walk = list; walk != NULL; walk = walk->next) { |
906 | struct pollfd * pfd, * pfd_end; | |
907 | ||
908 | pfd = walk->entries; | |
909 | pfd_end = pfd + walk->len; | |
910 | for (; pfd != pfd_end; pfd++) { | |
911 | /* | |
912 | * Fish for events. If we found one, record it | |
626cf236 | 913 | * and kill poll_table->_qproc, so we don't |
4a4b69f7 VL |
914 | * needlessly register any other waiters after |
915 | * this. They'll get immediately deregistered | |
916 | * when we break out and return. | |
917 | */ | |
cbf55001 ET |
918 | if (do_pollfd(pfd, pt, &can_busy_loop, |
919 | busy_flag)) { | |
4a4b69f7 | 920 | count++; |
626cf236 | 921 | pt->_qproc = NULL; |
cbf55001 ET |
922 | /* found something, stop busy polling */ |
923 | busy_flag = 0; | |
924 | can_busy_loop = false; | |
4a4b69f7 VL |
925 | } |
926 | } | |
1da177e4 | 927 | } |
4a4b69f7 VL |
928 | /* |
929 | * All waiters have already been registered, so don't provide | |
626cf236 | 930 | * a poll_table->_qproc to them on the next loop iteration. |
4a4b69f7 | 931 | */ |
626cf236 | 932 | pt->_qproc = NULL; |
9bf084f7 ON |
933 | if (!count) { |
934 | count = wait->error; | |
935 | if (signal_pending(current)) | |
8cf8b553 | 936 | count = -ERESTARTNOHAND; |
9bf084f7 | 937 | } |
8ff3e8e8 | 938 | if (count || timed_out) |
1da177e4 | 939 | break; |
9f72949f | 940 | |
cbf55001 | 941 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 | 942 | if (can_busy_loop && !need_resched()) { |
37056719 AD |
943 | if (!busy_start) { |
944 | busy_start = busy_loop_current_time(); | |
76b1e9b9 ET |
945 | continue; |
946 | } | |
37056719 | 947 | if (!busy_loop_timeout(busy_start)) |
76b1e9b9 ET |
948 | continue; |
949 | } | |
950 | busy_flag = 0; | |
91e2fd33 | 951 | |
8ff3e8e8 AV |
952 | /* |
953 | * If this is the first loop and we have a timeout | |
954 | * given, then we convert to ktime_t and set the to | |
955 | * pointer to the expiry value. | |
956 | */ | |
957 | if (end_time && !to) { | |
766b9f92 | 958 | expire = timespec64_to_ktime(*end_time); |
8ff3e8e8 | 959 | to = &expire; |
9f72949f DW |
960 | } |
961 | ||
5f820f64 | 962 | if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) |
8ff3e8e8 | 963 | timed_out = 1; |
1da177e4 | 964 | } |
1da177e4 LT |
965 | return count; |
966 | } | |
967 | ||
70674f95 AK |
968 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
969 | sizeof(struct pollfd)) | |
970 | ||
e99ca56c | 971 | static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
766b9f92 | 972 | struct timespec64 *end_time) |
1da177e4 LT |
973 | { |
974 | struct poll_wqueues table; | |
c67ef897 | 975 | int err = -EFAULT, fdcount; |
30c14e40 JS |
976 | /* Allocate small arguments on the stack to save memory and be |
977 | faster - use long to make sure the buffer is aligned properly | |
978 | on 64 bit archs to avoid unaligned access */ | |
979 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
980 | struct poll_list *const head = (struct poll_list *)stack_pps; |
981 | struct poll_list *walk = head; | |
c67ef897 KC |
982 | unsigned int todo = nfds; |
983 | unsigned int len; | |
1da177e4 | 984 | |
d554ed89 | 985 | if (nfds > rlimit(RLIMIT_NOFILE)) |
1da177e4 LT |
986 | return -EINVAL; |
987 | ||
252e5725 ON |
988 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
989 | for (;;) { | |
990 | walk->next = NULL; | |
991 | walk->len = len; | |
992 | if (!len) | |
993 | break; | |
1da177e4 | 994 | |
252e5725 ON |
995 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
996 | sizeof(struct pollfd) * walk->len)) | |
997 | goto out_fds; | |
998 | ||
c67ef897 | 999 | if (walk->len >= todo) |
252e5725 | 1000 | break; |
c67ef897 | 1001 | todo -= walk->len; |
1da177e4 | 1002 | |
252e5725 | 1003 | len = min(todo, POLLFD_PER_PAGE); |
43e11fa2 | 1004 | walk = walk->next = kmalloc(struct_size(walk, entries, len), |
0bcfe68b | 1005 | GFP_KERNEL); |
252e5725 ON |
1006 | if (!walk) { |
1007 | err = -ENOMEM; | |
1da177e4 LT |
1008 | goto out_fds; |
1009 | } | |
1da177e4 | 1010 | } |
9f72949f | 1011 | |
252e5725 | 1012 | poll_initwait(&table); |
ccec5ee3 | 1013 | fdcount = do_poll(head, &table, end_time); |
252e5725 | 1014 | poll_freewait(&table); |
1da177e4 | 1015 | |
ef0ba055 LT |
1016 | if (!user_write_access_begin(ufds, nfds * sizeof(*ufds))) |
1017 | goto out_fds; | |
1018 | ||
252e5725 | 1019 | for (walk = head; walk; walk = walk->next) { |
1da177e4 | 1020 | struct pollfd *fds = walk->entries; |
c67ef897 | 1021 | unsigned int j; |
1da177e4 | 1022 | |
ef0ba055 LT |
1023 | for (j = walk->len; j; fds++, ufds++, j--) |
1024 | unsafe_put_user(fds->revents, &ufds->revents, Efault); | |
1da177e4 | 1025 | } |
ef0ba055 | 1026 | user_write_access_end(); |
252e5725 | 1027 | |
1da177e4 | 1028 | err = fdcount; |
1da177e4 | 1029 | out_fds: |
252e5725 ON |
1030 | walk = head->next; |
1031 | while (walk) { | |
1032 | struct poll_list *pos = walk; | |
1033 | walk = walk->next; | |
1034 | kfree(pos); | |
1da177e4 | 1035 | } |
252e5725 | 1036 | |
1da177e4 | 1037 | return err; |
ef0ba055 LT |
1038 | |
1039 | Efault: | |
1040 | user_write_access_end(); | |
1041 | err = -EFAULT; | |
1042 | goto out_fds; | |
1da177e4 | 1043 | } |
9f72949f | 1044 | |
3075d9da CW |
1045 | static long do_restart_poll(struct restart_block *restart_block) |
1046 | { | |
8ff3e8e8 AV |
1047 | struct pollfd __user *ufds = restart_block->poll.ufds; |
1048 | int nfds = restart_block->poll.nfds; | |
766b9f92 | 1049 | struct timespec64 *to = NULL, end_time; |
3075d9da CW |
1050 | int ret; |
1051 | ||
8ff3e8e8 AV |
1052 | if (restart_block->poll.has_timeout) { |
1053 | end_time.tv_sec = restart_block->poll.tv_sec; | |
1054 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
1055 | to = &end_time; | |
1056 | } | |
1057 | ||
1058 | ret = do_sys_poll(ufds, nfds, to); | |
1059 | ||
5abbe51a ON |
1060 | if (ret == -ERESTARTNOHAND) |
1061 | ret = set_restart_fn(restart_block, do_restart_poll); | |
1062 | ||
3075d9da CW |
1063 | return ret; |
1064 | } | |
1065 | ||
5a8a82b1 | 1066 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
faf30900 | 1067 | int, timeout_msecs) |
9f72949f | 1068 | { |
766b9f92 | 1069 | struct timespec64 end_time, *to = NULL; |
3075d9da | 1070 | int ret; |
9f72949f | 1071 | |
8ff3e8e8 AV |
1072 | if (timeout_msecs >= 0) { |
1073 | to = &end_time; | |
1074 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
1075 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
1076 | } |
1077 | ||
8ff3e8e8 AV |
1078 | ret = do_sys_poll(ufds, nfds, to); |
1079 | ||
8cf8b553 | 1080 | if (ret == -ERESTARTNOHAND) { |
3075d9da | 1081 | struct restart_block *restart_block; |
8ff3e8e8 | 1082 | |
f56141e3 | 1083 | restart_block = ¤t->restart_block; |
8ff3e8e8 AV |
1084 | restart_block->poll.ufds = ufds; |
1085 | restart_block->poll.nfds = nfds; | |
1086 | ||
1087 | if (timeout_msecs >= 0) { | |
1088 | restart_block->poll.tv_sec = end_time.tv_sec; | |
1089 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
1090 | restart_block->poll.has_timeout = 1; | |
1091 | } else | |
1092 | restart_block->poll.has_timeout = 0; | |
1093 | ||
5abbe51a | 1094 | ret = set_restart_fn(restart_block, do_restart_poll); |
3075d9da CW |
1095 | } |
1096 | return ret; | |
9f72949f DW |
1097 | } |
1098 | ||
d4e82042 | 1099 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
8bd27a30 | 1100 | struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask, |
d4e82042 | 1101 | size_t, sigsetsize) |
9f72949f | 1102 | { |
36819ad0 | 1103 | struct timespec64 ts, end_time, *to = NULL; |
9f72949f DW |
1104 | int ret; |
1105 | ||
1106 | if (tsp) { | |
36819ad0 | 1107 | if (get_timespec64(&ts, tsp)) |
9f72949f DW |
1108 | return -EFAULT; |
1109 | ||
8ff3e8e8 AV |
1110 | to = &end_time; |
1111 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1112 | return -EINVAL; | |
9f72949f DW |
1113 | } |
1114 | ||
b772434b | 1115 | ret = set_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
1116 | if (ret) |
1117 | return ret; | |
9f72949f | 1118 | |
8ff3e8e8 | 1119 | ret = do_sys_poll(ufds, nfds, to); |
ac301020 | 1120 | return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret); |
9f72949f | 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 | struct timespec64 ts, end_time, *to = NULL; |
1130 | int ret; | |
e99ca56c | 1131 | |
8bd27a30 DD |
1132 | if (tsp) { |
1133 | if (get_old_timespec32(&ts, tsp)) | |
1134 | return -EFAULT; | |
e99ca56c | 1135 | |
8bd27a30 DD |
1136 | to = &end_time; |
1137 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1138 | return -EINVAL; | |
1139 | } | |
e99ca56c | 1140 | |
b772434b | 1141 | ret = set_user_sigmask(sigmask, sigsetsize); |
8bd27a30 | 1142 | if (ret) |
e99ca56c AV |
1143 | return ret; |
1144 | ||
8bd27a30 | 1145 | ret = do_sys_poll(ufds, nfds, to); |
ac301020 | 1146 | return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret); |
e99ca56c | 1147 | } |
8bd27a30 DD |
1148 | #endif |
1149 | ||
1150 | #ifdef CONFIG_COMPAT | |
1151 | #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t)) | |
e99ca56c AV |
1152 | |
1153 | /* | |
1154 | * Ooo, nasty. We need here to frob 32-bit unsigned longs to | |
1155 | * 64-bit unsigned longs. | |
1156 | */ | |
1157 | static | |
1158 | int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, | |
1159 | unsigned long *fdset) | |
1160 | { | |
e99ca56c | 1161 | if (ufdset) { |
464d6242 | 1162 | return compat_get_bitmap(fdset, ufdset, nr); |
e99ca56c | 1163 | } else { |
79de3cbe | 1164 | zero_fd_set(nr, fdset); |
464d6242 | 1165 | return 0; |
e99ca56c | 1166 | } |
e99ca56c AV |
1167 | } |
1168 | ||
1169 | static | |
1170 | int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, | |
1171 | unsigned long *fdset) | |
1172 | { | |
e99ca56c AV |
1173 | if (!ufdset) |
1174 | return 0; | |
464d6242 | 1175 | return compat_put_bitmap(ufdset, fdset, nr); |
e99ca56c AV |
1176 | } |
1177 | ||
1178 | ||
1179 | /* | |
1180 | * This is a virtual copy of sys_select from fs/select.c and probably | |
1181 | * should be compared to it from time to time | |
1182 | */ | |
1183 | ||
1184 | /* | |
1185 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
1186 | * like to be certain this leads to no problems. So I return | |
1187 | * EINTR just for safety. | |
1188 | * | |
1189 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
1190 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
1191 | */ | |
1192 | static int compat_core_sys_select(int n, compat_ulong_t __user *inp, | |
1193 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
36819ad0 | 1194 | struct timespec64 *end_time) |
e99ca56c AV |
1195 | { |
1196 | fd_set_bits fds; | |
1197 | void *bits; | |
1198 | int size, max_fds, ret = -EINVAL; | |
1199 | struct fdtable *fdt; | |
1200 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; | |
1201 | ||
1202 | if (n < 0) | |
1203 | goto out_nofds; | |
1204 | ||
1205 | /* max_fds can increase, so grab it once to avoid race */ | |
1206 | rcu_read_lock(); | |
1207 | fdt = files_fdtable(current->files); | |
1208 | max_fds = fdt->max_fds; | |
1209 | rcu_read_unlock(); | |
1210 | if (n > max_fds) | |
1211 | n = max_fds; | |
1212 | ||
1213 | /* | |
1214 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
1215 | * since we used fdset we need to allocate memory in units of | |
1216 | * long-words. | |
1217 | */ | |
1218 | size = FDS_BYTES(n); | |
1219 | bits = stack_fds; | |
1220 | if (size > sizeof(stack_fds) / 6) { | |
6da2ec56 | 1221 | bits = kmalloc_array(6, size, GFP_KERNEL); |
e99ca56c AV |
1222 | ret = -ENOMEM; |
1223 | if (!bits) | |
1224 | goto out_nofds; | |
1225 | } | |
1226 | fds.in = (unsigned long *) bits; | |
1227 | fds.out = (unsigned long *) (bits + size); | |
1228 | fds.ex = (unsigned long *) (bits + 2*size); | |
1229 | fds.res_in = (unsigned long *) (bits + 3*size); | |
1230 | fds.res_out = (unsigned long *) (bits + 4*size); | |
1231 | fds.res_ex = (unsigned long *) (bits + 5*size); | |
1232 | ||
1233 | if ((ret = compat_get_fd_set(n, inp, fds.in)) || | |
1234 | (ret = compat_get_fd_set(n, outp, fds.out)) || | |
1235 | (ret = compat_get_fd_set(n, exp, fds.ex))) | |
1236 | goto out; | |
1237 | zero_fd_set(n, fds.res_in); | |
1238 | zero_fd_set(n, fds.res_out); | |
1239 | zero_fd_set(n, fds.res_ex); | |
1240 | ||
1241 | ret = do_select(n, &fds, end_time); | |
1242 | ||
1243 | if (ret < 0) | |
1244 | goto out; | |
1245 | if (!ret) { | |
1246 | ret = -ERESTARTNOHAND; | |
1247 | if (signal_pending(current)) | |
1248 | goto out; | |
1249 | ret = 0; | |
1250 | } | |
1251 | ||
1252 | if (compat_set_fd_set(n, inp, fds.res_in) || | |
1253 | compat_set_fd_set(n, outp, fds.res_out) || | |
1254 | compat_set_fd_set(n, exp, fds.res_ex)) | |
1255 | ret = -EFAULT; | |
1256 | out: | |
1257 | if (bits != stack_fds) | |
1258 | kfree(bits); | |
1259 | out_nofds: | |
1260 | return ret; | |
1261 | } | |
1262 | ||
05585e44 DB |
1263 | static int do_compat_select(int n, compat_ulong_t __user *inp, |
1264 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
9afc5eee | 1265 | struct old_timeval32 __user *tvp) |
e99ca56c | 1266 | { |
36819ad0 | 1267 | struct timespec64 end_time, *to = NULL; |
9afc5eee | 1268 | struct old_timeval32 tv; |
e99ca56c AV |
1269 | int ret; |
1270 | ||
1271 | if (tvp) { | |
1272 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
1273 | return -EFAULT; | |
1274 | ||
1275 | to = &end_time; | |
1276 | if (poll_select_set_timeout(to, | |
1277 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
1278 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
1279 | return -EINVAL; | |
1280 | } | |
1281 | ||
1282 | ret = compat_core_sys_select(n, inp, outp, exp, to); | |
ac301020 | 1283 | return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret); |
e99ca56c AV |
1284 | } |
1285 | ||
05585e44 DB |
1286 | COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp, |
1287 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
9afc5eee | 1288 | struct old_timeval32 __user *, tvp) |
05585e44 DB |
1289 | { |
1290 | return do_compat_select(n, inp, outp, exp, tvp); | |
1291 | } | |
1292 | ||
e99ca56c AV |
1293 | struct compat_sel_arg_struct { |
1294 | compat_ulong_t n; | |
1295 | compat_uptr_t inp; | |
1296 | compat_uptr_t outp; | |
1297 | compat_uptr_t exp; | |
1298 | compat_uptr_t tvp; | |
1299 | }; | |
1300 | ||
1301 | COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg) | |
1302 | { | |
1303 | struct compat_sel_arg_struct a; | |
1304 | ||
1305 | if (copy_from_user(&a, arg, sizeof(a))) | |
1306 | return -EFAULT; | |
05585e44 DB |
1307 | return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp), |
1308 | compat_ptr(a.exp), compat_ptr(a.tvp)); | |
e99ca56c AV |
1309 | } |
1310 | ||
1311 | static long do_compat_pselect(int n, compat_ulong_t __user *inp, | |
1312 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
e024707b DD |
1313 | void __user *tsp, compat_sigset_t __user *sigmask, |
1314 | compat_size_t sigsetsize, enum poll_time_type type) | |
e99ca56c | 1315 | { |
36819ad0 | 1316 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1317 | int ret; |
1318 | ||
1319 | if (tsp) { | |
e024707b DD |
1320 | switch (type) { |
1321 | case PT_OLD_TIMESPEC: | |
1322 | if (get_old_timespec32(&ts, tsp)) | |
1323 | return -EFAULT; | |
1324 | break; | |
1325 | case PT_TIMESPEC: | |
1326 | if (get_timespec64(&ts, tsp)) | |
1327 | return -EFAULT; | |
1328 | break; | |
1329 | default: | |
1330 | BUG(); | |
1331 | } | |
e99ca56c AV |
1332 | |
1333 | to = &end_time; | |
1334 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1335 | return -EINVAL; | |
1336 | } | |
1337 | ||
b772434b | 1338 | ret = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
1339 | if (ret) |
1340 | return ret; | |
e99ca56c AV |
1341 | |
1342 | ret = compat_core_sys_select(n, inp, outp, exp, to); | |
ac301020 | 1343 | return poll_select_finish(&end_time, tsp, type, ret); |
e99ca56c AV |
1344 | } |
1345 | ||
7e71609f AV |
1346 | struct compat_sigset_argpack { |
1347 | compat_uptr_t p; | |
1348 | compat_size_t size; | |
1349 | }; | |
1350 | static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to, | |
1351 | struct compat_sigset_argpack __user *from) | |
1352 | { | |
1353 | if (from) { | |
1354 | if (!user_read_access_begin(from, sizeof(*from))) | |
1355 | return -EFAULT; | |
1356 | unsafe_get_user(to->p, &from->p, Efault); | |
1357 | unsafe_get_user(to->size, &from->size, Efault); | |
1358 | user_read_access_end(); | |
1359 | } | |
1360 | return 0; | |
1361 | Efault: | |
1362 | user_access_end(); | |
1363 | return -EFAULT; | |
1364 | } | |
1365 | ||
e024707b DD |
1366 | COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp, |
1367 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
1368 | struct __kernel_timespec __user *, tsp, void __user *, sig) | |
1369 | { | |
7e71609f AV |
1370 | struct compat_sigset_argpack x = {0, 0}; |
1371 | ||
1372 | if (get_compat_sigset_argpack(&x, sig)) | |
1373 | return -EFAULT; | |
e024707b | 1374 | |
7e71609f AV |
1375 | return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p), |
1376 | x.size, PT_TIMESPEC); | |
e024707b DD |
1377 | } |
1378 | ||
1379 | #if defined(CONFIG_COMPAT_32BIT_TIME) | |
1380 | ||
8dabe724 | 1381 | COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp, |
e99ca56c | 1382 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, |
9afc5eee | 1383 | struct old_timespec32 __user *, tsp, void __user *, sig) |
e99ca56c | 1384 | { |
7e71609f AV |
1385 | struct compat_sigset_argpack x = {0, 0}; |
1386 | ||
1387 | if (get_compat_sigset_argpack(&x, sig)) | |
1388 | return -EFAULT; | |
e024707b | 1389 | |
7e71609f AV |
1390 | return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p), |
1391 | x.size, PT_OLD_TIMESPEC); | |
e99ca56c AV |
1392 | } |
1393 | ||
e024707b DD |
1394 | #endif |
1395 | ||
8bd27a30 | 1396 | #if defined(CONFIG_COMPAT_32BIT_TIME) |
8dabe724 | 1397 | COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, |
9afc5eee | 1398 | unsigned int, nfds, struct old_timespec32 __user *, tsp, |
e99ca56c AV |
1399 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) |
1400 | { | |
36819ad0 | 1401 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1402 | int ret; |
1403 | ||
1404 | if (tsp) { | |
9afc5eee | 1405 | if (get_old_timespec32(&ts, tsp)) |
e99ca56c AV |
1406 | return -EFAULT; |
1407 | ||
1408 | to = &end_time; | |
1409 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1410 | return -EINVAL; | |
1411 | } | |
1412 | ||
b772434b | 1413 | ret = set_compat_user_sigmask(sigmask, sigsetsize); |
ded653cc DD |
1414 | if (ret) |
1415 | return ret; | |
e99ca56c AV |
1416 | |
1417 | ret = do_sys_poll(ufds, nfds, to); | |
ac301020 | 1418 | return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret); |
e99ca56c AV |
1419 | } |
1420 | #endif | |
8bd27a30 DD |
1421 | |
1422 | /* New compat syscall for 64 bit time_t*/ | |
1423 | COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds, | |
1424 | unsigned int, nfds, struct __kernel_timespec __user *, tsp, | |
1425 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) | |
1426 | { | |
8bd27a30 DD |
1427 | struct timespec64 ts, end_time, *to = NULL; |
1428 | int ret; | |
1429 | ||
1430 | if (tsp) { | |
1431 | if (get_timespec64(&ts, tsp)) | |
1432 | return -EFAULT; | |
1433 | ||
1434 | to = &end_time; | |
1435 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1436 | return -EINVAL; | |
1437 | } | |
1438 | ||
b772434b | 1439 | ret = set_compat_user_sigmask(sigmask, sigsetsize); |
8bd27a30 DD |
1440 | if (ret) |
1441 | return ret; | |
1442 | ||
1443 | ret = do_sys_poll(ufds, nfds, to); | |
ac301020 | 1444 | return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret); |
8bd27a30 DD |
1445 | } |
1446 | ||
1447 | #endif |