Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * This file contains the procedures for the handling of select and poll | |
4 | * | |
5 | * Created for Linux based loosely upon Mathius Lattner's minix | |
6 | * patches by Peter MacDonald. Heavily edited by Linus. | |
7 | * | |
8 | * 4 February 1994 | |
9 | * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS | |
10 | * flag set in its personality we do *not* modify the given timeout | |
11 | * parameter to reflect time remaining. | |
12 | * | |
13 | * 24 January 2000 | |
14 | * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation | |
15 | * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian). | |
16 | */ | |
17 | ||
022a1692 | 18 | #include <linux/kernel.h> |
3f07c014 IM |
19 | #include <linux/sched/signal.h> |
20 | #include <linux/sched/rt.h> | |
1da177e4 | 21 | #include <linux/syscalls.h> |
630d9c47 | 22 | #include <linux/export.h> |
1da177e4 | 23 | #include <linux/slab.h> |
1da177e4 LT |
24 | #include <linux/poll.h> |
25 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ | |
26 | #include <linux/file.h> | |
9f3acc31 | 27 | #include <linux/fdtable.h> |
1da177e4 | 28 | #include <linux/fs.h> |
b835996f | 29 | #include <linux/rcupdate.h> |
8ff3e8e8 | 30 | #include <linux/hrtimer.h> |
9745cdb3 | 31 | #include <linux/freezer.h> |
076bb0c8 | 32 | #include <net/busy_poll.h> |
2d19309c | 33 | #include <linux/vmalloc.h> |
1da177e4 | 34 | |
7c0f6ba6 | 35 | #include <linux/uaccess.h> |
1da177e4 | 36 | |
90d6e24a AV |
37 | |
38 | /* | |
39 | * Estimate expected accuracy in ns from a timeval. | |
40 | * | |
41 | * After quite a bit of churning around, we've settled on | |
42 | * a simple thing of taking 0.1% of the timeout as the | |
43 | * slack, with a cap of 100 msec. | |
44 | * "nice" tasks get a 0.5% slack instead. | |
45 | * | |
46 | * Consider this comment an open invitation to come up with even | |
47 | * better solutions.. | |
48 | */ | |
49 | ||
5ae87e79 GK |
50 | #define MAX_SLACK (100 * NSEC_PER_MSEC) |
51 | ||
766b9f92 | 52 | static long __estimate_accuracy(struct timespec64 *tv) |
90d6e24a | 53 | { |
96d2ab48 | 54 | long slack; |
90d6e24a AV |
55 | int divfactor = 1000; |
56 | ||
5ae87e79 GK |
57 | if (tv->tv_sec < 0) |
58 | return 0; | |
59 | ||
4ce105d3 | 60 | if (task_nice(current) > 0) |
90d6e24a AV |
61 | divfactor = divfactor / 5; |
62 | ||
5ae87e79 GK |
63 | if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) |
64 | return MAX_SLACK; | |
65 | ||
90d6e24a AV |
66 | slack = tv->tv_nsec / divfactor; |
67 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); | |
68 | ||
5ae87e79 GK |
69 | if (slack > MAX_SLACK) |
70 | return MAX_SLACK; | |
96d2ab48 | 71 | |
90d6e24a AV |
72 | return slack; |
73 | } | |
74 | ||
766b9f92 | 75 | u64 select_estimate_accuracy(struct timespec64 *tv) |
90d6e24a | 76 | { |
da8b44d5 | 77 | u64 ret; |
766b9f92 | 78 | struct timespec64 now; |
90d6e24a AV |
79 | |
80 | /* | |
81 | * Realtime tasks get a slack of 0 for obvious reasons. | |
82 | */ | |
83 | ||
4ce105d3 | 84 | if (rt_task(current)) |
90d6e24a AV |
85 | return 0; |
86 | ||
766b9f92 DD |
87 | ktime_get_ts64(&now); |
88 | now = timespec64_sub(*tv, now); | |
90d6e24a AV |
89 | ret = __estimate_accuracy(&now); |
90 | if (ret < current->timer_slack_ns) | |
91 | return current->timer_slack_ns; | |
92 | return ret; | |
93 | } | |
94 | ||
95 | ||
96 | ||
1da177e4 LT |
97 | struct poll_table_page { |
98 | struct poll_table_page * next; | |
99 | struct poll_table_entry * entry; | |
100 | struct poll_table_entry entries[0]; | |
101 | }; | |
102 | ||
103 | #define POLL_TABLE_FULL(table) \ | |
104 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) | |
105 | ||
106 | /* | |
107 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. | |
108 | * I have rewritten this, taking some shortcuts: This code may not be easy to | |
109 | * follow, but it should be free of race-conditions, and it's practical. If you | |
110 | * understand what I'm doing here, then you understand how the linux | |
111 | * sleep/wakeup mechanism works. | |
112 | * | |
113 | * Two very simple procedures, poll_wait() and poll_freewait() make all the | |
114 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, | |
115 | * as all select/poll functions have to call it to add an entry to the | |
116 | * poll table. | |
117 | */ | |
75c96f85 AB |
118 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
119 | poll_table *p); | |
1da177e4 LT |
120 | |
121 | void poll_initwait(struct poll_wqueues *pwq) | |
122 | { | |
123 | init_poll_funcptr(&pwq->pt, __pollwait); | |
5f820f64 | 124 | pwq->polling_task = current; |
b2add73d | 125 | pwq->triggered = 0; |
1da177e4 LT |
126 | pwq->error = 0; |
127 | pwq->table = NULL; | |
70674f95 | 128 | pwq->inline_index = 0; |
1da177e4 | 129 | } |
1da177e4 LT |
130 | EXPORT_SYMBOL(poll_initwait); |
131 | ||
70674f95 AK |
132 | static void free_poll_entry(struct poll_table_entry *entry) |
133 | { | |
ccf6780d | 134 | remove_wait_queue(entry->wait_address, &entry->wait); |
70674f95 AK |
135 | fput(entry->filp); |
136 | } | |
137 | ||
1da177e4 LT |
138 | void poll_freewait(struct poll_wqueues *pwq) |
139 | { | |
140 | struct poll_table_page * p = pwq->table; | |
70674f95 AK |
141 | int i; |
142 | for (i = 0; i < pwq->inline_index; i++) | |
143 | free_poll_entry(pwq->inline_entries + i); | |
1da177e4 LT |
144 | while (p) { |
145 | struct poll_table_entry * entry; | |
146 | struct poll_table_page *old; | |
147 | ||
148 | entry = p->entry; | |
149 | do { | |
150 | entry--; | |
70674f95 | 151 | free_poll_entry(entry); |
1da177e4 LT |
152 | } while (entry > p->entries); |
153 | old = p; | |
154 | p = p->next; | |
155 | free_page((unsigned long) old); | |
156 | } | |
157 | } | |
1da177e4 LT |
158 | EXPORT_SYMBOL(poll_freewait); |
159 | ||
5f820f64 | 160 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
1da177e4 | 161 | { |
1da177e4 LT |
162 | struct poll_table_page *table = p->table; |
163 | ||
70674f95 AK |
164 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
165 | return p->inline_entries + p->inline_index++; | |
166 | ||
1da177e4 LT |
167 | if (!table || POLL_TABLE_FULL(table)) { |
168 | struct poll_table_page *new_table; | |
169 | ||
170 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); | |
171 | if (!new_table) { | |
172 | p->error = -ENOMEM; | |
70674f95 | 173 | return NULL; |
1da177e4 LT |
174 | } |
175 | new_table->entry = new_table->entries; | |
176 | new_table->next = table; | |
177 | p->table = new_table; | |
178 | table = new_table; | |
179 | } | |
180 | ||
70674f95 AK |
181 | return table->entry++; |
182 | } | |
183 | ||
ac6424b9 | 184 | static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
5f820f64 TH |
185 | { |
186 | struct poll_wqueues *pwq = wait->private; | |
187 | DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); | |
188 | ||
189 | /* | |
190 | * Although this function is called under waitqueue lock, LOCK | |
191 | * doesn't imply write barrier and the users expect write | |
192 | * barrier semantics on wakeup functions. The following | |
193 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
b92b8b35 | 194 | * and is paired with smp_store_mb() in poll_schedule_timeout. |
5f820f64 TH |
195 | */ |
196 | smp_wmb(); | |
197 | pwq->triggered = 1; | |
198 | ||
199 | /* | |
200 | * Perform the default wake up operation using a dummy | |
201 | * waitqueue. | |
202 | * | |
203 | * TODO: This is hacky but there currently is no interface to | |
204 | * pass in @sync. @sync is scheduled to be removed and once | |
205 | * that happens, wake_up_process() can be used directly. | |
206 | */ | |
207 | return default_wake_function(&dummy_wait, mode, sync, key); | |
208 | } | |
209 | ||
ac6424b9 | 210 | static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
4938d7e0 ED |
211 | { |
212 | struct poll_table_entry *entry; | |
213 | ||
214 | entry = container_of(wait, struct poll_table_entry, wait); | |
3ad6f93e | 215 | if (key && !(key_to_poll(key) & entry->key)) |
4938d7e0 ED |
216 | return 0; |
217 | return __pollwake(wait, mode, sync, key); | |
218 | } | |
219 | ||
70674f95 AK |
220 | /* Add a new entry */ |
221 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, | |
222 | poll_table *p) | |
223 | { | |
5f820f64 TH |
224 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
225 | struct poll_table_entry *entry = poll_get_entry(pwq); | |
70674f95 AK |
226 | if (!entry) |
227 | return; | |
cb0942b8 | 228 | entry->filp = get_file(filp); |
70674f95 | 229 | entry->wait_address = wait_address; |
626cf236 | 230 | entry->key = p->_key; |
5f820f64 TH |
231 | init_waitqueue_func_entry(&entry->wait, pollwake); |
232 | entry->wait.private = pwq; | |
ccf6780d | 233 | add_wait_queue(wait_address, &entry->wait); |
1da177e4 LT |
234 | } |
235 | ||
8f546ae1 | 236 | static int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
5f820f64 TH |
237 | ktime_t *expires, unsigned long slack) |
238 | { | |
239 | int rc = -EINTR; | |
240 | ||
241 | set_current_state(state); | |
242 | if (!pwq->triggered) | |
59612d18 | 243 | rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); |
5f820f64 TH |
244 | __set_current_state(TASK_RUNNING); |
245 | ||
246 | /* | |
247 | * Prepare for the next iteration. | |
248 | * | |
b92b8b35 | 249 | * The following smp_store_mb() serves two purposes. First, it's |
5f820f64 TH |
250 | * the counterpart rmb of the wmb in pollwake() such that data |
251 | * written before wake up is always visible after wake up. | |
252 | * Second, the full barrier guarantees that triggered clearing | |
253 | * doesn't pass event check of the next iteration. Note that | |
254 | * this problem doesn't exist for the first iteration as | |
255 | * add_wait_queue() has full barrier semantics. | |
256 | */ | |
b92b8b35 | 257 | smp_store_mb(pwq->triggered, 0); |
5f820f64 TH |
258 | |
259 | return rc; | |
260 | } | |
5f820f64 | 261 | |
b773ad40 TG |
262 | /** |
263 | * poll_select_set_timeout - helper function to setup the timeout value | |
766b9f92 | 264 | * @to: pointer to timespec64 variable for the final timeout |
b773ad40 TG |
265 | * @sec: seconds (from user space) |
266 | * @nsec: nanoseconds (from user space) | |
267 | * | |
268 | * Note, we do not use a timespec for the user space value here, That | |
269 | * way we can use the function for timeval and compat interfaces as well. | |
270 | * | |
271 | * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. | |
272 | */ | |
766b9f92 | 273 | int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec) |
b773ad40 | 274 | { |
766b9f92 | 275 | struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec}; |
b773ad40 | 276 | |
766b9f92 | 277 | if (!timespec64_valid(&ts)) |
b773ad40 TG |
278 | return -EINVAL; |
279 | ||
280 | /* Optimize for the zero timeout value here */ | |
281 | if (!sec && !nsec) { | |
282 | to->tv_sec = to->tv_nsec = 0; | |
283 | } else { | |
766b9f92 DD |
284 | ktime_get_ts64(to); |
285 | *to = timespec64_add_safe(*to, ts); | |
b773ad40 TG |
286 | } |
287 | return 0; | |
288 | } | |
289 | ||
766b9f92 DD |
290 | static int poll_select_copy_remaining(struct timespec64 *end_time, |
291 | void __user *p, | |
b773ad40 TG |
292 | int timeval, int ret) |
293 | { | |
36819ad0 | 294 | struct timespec64 rts; |
b773ad40 TG |
295 | struct timeval rtv; |
296 | ||
297 | if (!p) | |
298 | return ret; | |
299 | ||
300 | if (current->personality & STICKY_TIMEOUTS) | |
301 | goto sticky; | |
302 | ||
303 | /* No update for zero timeout */ | |
304 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
305 | return ret; | |
306 | ||
36819ad0 DD |
307 | ktime_get_ts64(&rts); |
308 | rts = timespec64_sub(*end_time, rts); | |
309 | if (rts.tv_sec < 0) | |
310 | rts.tv_sec = rts.tv_nsec = 0; | |
766b9f92 | 311 | |
b773ad40 TG |
312 | |
313 | if (timeval) { | |
65329bf4 VK |
314 | if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) |
315 | memset(&rtv, 0, sizeof(rtv)); | |
36819ad0 DD |
316 | rtv.tv_sec = rts.tv_sec; |
317 | rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; | |
b773ad40 TG |
318 | |
319 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
320 | return ret; | |
321 | ||
36819ad0 | 322 | } else if (!put_timespec64(&rts, p)) |
b773ad40 TG |
323 | return ret; |
324 | ||
325 | /* | |
326 | * If an application puts its timeval in read-only memory, we | |
327 | * don't want the Linux-specific update to the timeval to | |
328 | * cause a fault after the select has completed | |
329 | * successfully. However, because we're not updating the | |
330 | * timeval, we can't restart the system call. | |
331 | */ | |
332 | ||
333 | sticky: | |
334 | if (ret == -ERESTARTNOHAND) | |
335 | ret = -EINTR; | |
336 | return ret; | |
337 | } | |
338 | ||
e99ca56c AV |
339 | /* |
340 | * Scalable version of the fd_set. | |
341 | */ | |
342 | ||
343 | typedef struct { | |
344 | unsigned long *in, *out, *ex; | |
345 | unsigned long *res_in, *res_out, *res_ex; | |
346 | } fd_set_bits; | |
347 | ||
348 | /* | |
349 | * How many longwords for "nr" bits? | |
350 | */ | |
351 | #define FDS_BITPERLONG (8*sizeof(long)) | |
352 | #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG) | |
353 | #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long)) | |
354 | ||
355 | /* | |
356 | * We do a VERIFY_WRITE here even though we are only reading this time: | |
357 | * we'll write to it eventually.. | |
358 | * | |
359 | * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned. | |
360 | */ | |
361 | static inline | |
362 | int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) | |
363 | { | |
364 | nr = FDS_BYTES(nr); | |
365 | if (ufdset) | |
366 | return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0; | |
367 | ||
368 | memset(fdset, 0, nr); | |
369 | return 0; | |
370 | } | |
371 | ||
372 | static inline unsigned long __must_check | |
373 | set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) | |
374 | { | |
375 | if (ufdset) | |
376 | return __copy_to_user(ufdset, fdset, FDS_BYTES(nr)); | |
377 | return 0; | |
378 | } | |
379 | ||
380 | static inline | |
381 | void zero_fd_set(unsigned long nr, unsigned long *fdset) | |
382 | { | |
383 | memset(fdset, 0, FDS_BYTES(nr)); | |
384 | } | |
385 | ||
1da177e4 LT |
386 | #define FDS_IN(fds, n) (fds->in + n) |
387 | #define FDS_OUT(fds, n) (fds->out + n) | |
388 | #define FDS_EX(fds, n) (fds->ex + n) | |
389 | ||
390 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) | |
391 | ||
392 | static int max_select_fd(unsigned long n, fd_set_bits *fds) | |
393 | { | |
394 | unsigned long *open_fds; | |
395 | unsigned long set; | |
396 | int max; | |
badf1662 | 397 | struct fdtable *fdt; |
1da177e4 LT |
398 | |
399 | /* handle last in-complete long-word first */ | |
8ded2bbc JB |
400 | set = ~(~0UL << (n & (BITS_PER_LONG-1))); |
401 | n /= BITS_PER_LONG; | |
badf1662 | 402 | fdt = files_fdtable(current->files); |
1fd36adc | 403 | open_fds = fdt->open_fds + n; |
1da177e4 LT |
404 | max = 0; |
405 | if (set) { | |
406 | set &= BITS(fds, n); | |
407 | if (set) { | |
408 | if (!(set & ~*open_fds)) | |
409 | goto get_max; | |
410 | return -EBADF; | |
411 | } | |
412 | } | |
413 | while (n) { | |
414 | open_fds--; | |
415 | n--; | |
416 | set = BITS(fds, n); | |
417 | if (!set) | |
418 | continue; | |
419 | if (set & ~*open_fds) | |
420 | return -EBADF; | |
421 | if (max) | |
422 | continue; | |
423 | get_max: | |
424 | do { | |
425 | max++; | |
426 | set >>= 1; | |
427 | } while (set); | |
8ded2bbc | 428 | max += n * BITS_PER_LONG; |
1da177e4 LT |
429 | } |
430 | ||
431 | return max; | |
432 | } | |
433 | ||
a9a08845 LT |
434 | #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR) |
435 | #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR) | |
436 | #define POLLEX_SET (EPOLLPRI) | |
1da177e4 | 437 | |
4938d7e0 | 438 | static inline void wait_key_set(poll_table *wait, unsigned long in, |
2d48d67f | 439 | unsigned long out, unsigned long bit, |
01699437 | 440 | __poll_t ll_flag) |
4938d7e0 | 441 | { |
2d48d67f | 442 | wait->_key = POLLEX_SET | ll_flag; |
626cf236 HV |
443 | if (in & bit) |
444 | wait->_key |= POLLIN_SET; | |
445 | if (out & bit) | |
446 | wait->_key |= POLLOUT_SET; | |
4938d7e0 ED |
447 | } |
448 | ||
e99ca56c | 449 | static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time) |
1da177e4 | 450 | { |
8ff3e8e8 | 451 | ktime_t expire, *to = NULL; |
1da177e4 LT |
452 | struct poll_wqueues table; |
453 | poll_table *wait; | |
8ff3e8e8 | 454 | int retval, i, timed_out = 0; |
da8b44d5 | 455 | u64 slack = 0; |
01699437 | 456 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 457 | unsigned long busy_start = 0; |
1da177e4 | 458 | |
b835996f | 459 | rcu_read_lock(); |
1da177e4 | 460 | retval = max_select_fd(n, fds); |
b835996f | 461 | rcu_read_unlock(); |
1da177e4 LT |
462 | |
463 | if (retval < 0) | |
464 | return retval; | |
465 | n = retval; | |
466 | ||
467 | poll_initwait(&table); | |
468 | wait = &table.pt; | |
8ff3e8e8 | 469 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 470 | wait->_qproc = NULL; |
8ff3e8e8 AV |
471 | timed_out = 1; |
472 | } | |
473 | ||
96d2ab48 | 474 | if (end_time && !timed_out) |
231f3d39 | 475 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 476 | |
1da177e4 LT |
477 | retval = 0; |
478 | for (;;) { | |
479 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; | |
cbf55001 | 480 | bool can_busy_loop = false; |
1da177e4 | 481 | |
1da177e4 LT |
482 | inp = fds->in; outp = fds->out; exp = fds->ex; |
483 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; | |
484 | ||
485 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { | |
e6c8adca | 486 | unsigned long in, out, ex, all_bits, bit = 1, j; |
1da177e4 | 487 | unsigned long res_in = 0, res_out = 0, res_ex = 0; |
e6c8adca | 488 | __poll_t mask; |
1da177e4 LT |
489 | |
490 | in = *inp++; out = *outp++; ex = *exp++; | |
491 | all_bits = in | out | ex; | |
492 | if (all_bits == 0) { | |
8ded2bbc | 493 | i += BITS_PER_LONG; |
1da177e4 LT |
494 | continue; |
495 | } | |
496 | ||
8ded2bbc | 497 | for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { |
2903ff01 | 498 | struct fd f; |
1da177e4 LT |
499 | if (i >= n) |
500 | break; | |
501 | if (!(bit & all_bits)) | |
502 | continue; | |
2903ff01 AV |
503 | f = fdget(i); |
504 | if (f.file) { | |
505 | const struct file_operations *f_op; | |
506 | f_op = f.file->f_op; | |
1da177e4 | 507 | mask = DEFAULT_POLLMASK; |
72c2d531 | 508 | if (f_op->poll) { |
2d48d67f | 509 | wait_key_set(wait, in, out, |
cbf55001 | 510 | bit, busy_flag); |
2903ff01 | 511 | mask = (*f_op->poll)(f.file, wait); |
4938d7e0 | 512 | } |
2903ff01 | 513 | fdput(f); |
1da177e4 LT |
514 | if ((mask & POLLIN_SET) && (in & bit)) { |
515 | res_in |= bit; | |
516 | retval++; | |
626cf236 | 517 | wait->_qproc = NULL; |
1da177e4 LT |
518 | } |
519 | if ((mask & POLLOUT_SET) && (out & bit)) { | |
520 | res_out |= bit; | |
521 | retval++; | |
626cf236 | 522 | wait->_qproc = NULL; |
1da177e4 LT |
523 | } |
524 | if ((mask & POLLEX_SET) && (ex & bit)) { | |
525 | res_ex |= bit; | |
526 | retval++; | |
626cf236 | 527 | wait->_qproc = NULL; |
1da177e4 | 528 | } |
2d48d67f | 529 | /* got something, stop busy polling */ |
cbf55001 ET |
530 | if (retval) { |
531 | can_busy_loop = false; | |
532 | busy_flag = 0; | |
533 | ||
534 | /* | |
535 | * only remember a returned | |
536 | * POLL_BUSY_LOOP if we asked for it | |
537 | */ | |
538 | } else if (busy_flag & mask) | |
539 | can_busy_loop = true; | |
540 | ||
1da177e4 | 541 | } |
1da177e4 LT |
542 | } |
543 | if (res_in) | |
544 | *rinp = res_in; | |
545 | if (res_out) | |
546 | *routp = res_out; | |
547 | if (res_ex) | |
548 | *rexp = res_ex; | |
55d85384 | 549 | cond_resched(); |
1da177e4 | 550 | } |
626cf236 | 551 | wait->_qproc = NULL; |
8ff3e8e8 | 552 | if (retval || timed_out || signal_pending(current)) |
1da177e4 | 553 | break; |
f5264481 | 554 | if (table.error) { |
1da177e4 LT |
555 | retval = table.error; |
556 | break; | |
557 | } | |
9f72949f | 558 | |
cbf55001 | 559 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 | 560 | if (can_busy_loop && !need_resched()) { |
37056719 AD |
561 | if (!busy_start) { |
562 | busy_start = busy_loop_current_time(); | |
76b1e9b9 ET |
563 | continue; |
564 | } | |
37056719 | 565 | if (!busy_loop_timeout(busy_start)) |
76b1e9b9 ET |
566 | continue; |
567 | } | |
568 | busy_flag = 0; | |
2d48d67f | 569 | |
8ff3e8e8 AV |
570 | /* |
571 | * If this is the first loop and we have a timeout | |
572 | * given, then we convert to ktime_t and set the to | |
573 | * pointer to the expiry value. | |
574 | */ | |
575 | if (end_time && !to) { | |
766b9f92 | 576 | expire = timespec64_to_ktime(*end_time); |
8ff3e8e8 | 577 | to = &expire; |
9f72949f | 578 | } |
8ff3e8e8 | 579 | |
5f820f64 TH |
580 | if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE, |
581 | to, slack)) | |
8ff3e8e8 | 582 | timed_out = 1; |
1da177e4 | 583 | } |
1da177e4 LT |
584 | |
585 | poll_freewait(&table); | |
586 | ||
1da177e4 LT |
587 | return retval; |
588 | } | |
589 | ||
1da177e4 LT |
590 | /* |
591 | * We can actually return ERESTARTSYS instead of EINTR, but I'd | |
592 | * like to be certain this leads to no problems. So I return | |
593 | * EINTR just for safety. | |
594 | * | |
595 | * Update: ERESTARTSYS breaks at least the xview clock binary, so | |
596 | * I'm trying ERESTARTNOHAND which restart only when you want to. | |
597 | */ | |
a2dcb44c | 598 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
766b9f92 | 599 | fd_set __user *exp, struct timespec64 *end_time) |
1da177e4 LT |
600 | { |
601 | fd_set_bits fds; | |
29ff2db5 | 602 | void *bits; |
bbea9f69 | 603 | int ret, max_fds; |
2d19309c | 604 | size_t size, alloc_size; |
badf1662 | 605 | struct fdtable *fdt; |
70674f95 | 606 | /* Allocate small arguments on the stack to save memory and be faster */ |
30c14e40 | 607 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1da177e4 | 608 | |
1da177e4 LT |
609 | ret = -EINVAL; |
610 | if (n < 0) | |
611 | goto out_nofds; | |
612 | ||
bbea9f69 | 613 | /* max_fds can increase, so grab it once to avoid race */ |
b835996f | 614 | rcu_read_lock(); |
badf1662 | 615 | fdt = files_fdtable(current->files); |
bbea9f69 | 616 | max_fds = fdt->max_fds; |
b835996f | 617 | rcu_read_unlock(); |
bbea9f69 VL |
618 | if (n > max_fds) |
619 | n = max_fds; | |
1da177e4 LT |
620 | |
621 | /* | |
622 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), | |
623 | * since we used fdset we need to allocate memory in units of | |
624 | * long-words. | |
625 | */ | |
1da177e4 | 626 | size = FDS_BYTES(n); |
b04eb6aa MBJ |
627 | bits = stack_fds; |
628 | if (size > sizeof(stack_fds) / 6) { | |
629 | /* Not enough space in on-stack array; must use kmalloc */ | |
630 | ret = -ENOMEM; | |
2d19309c VB |
631 | if (size > (SIZE_MAX / 6)) |
632 | goto out_nofds; | |
633 | ||
634 | alloc_size = 6 * size; | |
752ade68 | 635 | bits = kvmalloc(alloc_size, GFP_KERNEL); |
b04eb6aa MBJ |
636 | if (!bits) |
637 | goto out_nofds; | |
638 | } | |
29ff2db5 AM |
639 | fds.in = bits; |
640 | fds.out = bits + size; | |
641 | fds.ex = bits + 2*size; | |
642 | fds.res_in = bits + 3*size; | |
643 | fds.res_out = bits + 4*size; | |
644 | fds.res_ex = bits + 5*size; | |
1da177e4 LT |
645 | |
646 | if ((ret = get_fd_set(n, inp, fds.in)) || | |
647 | (ret = get_fd_set(n, outp, fds.out)) || | |
648 | (ret = get_fd_set(n, exp, fds.ex))) | |
649 | goto out; | |
650 | zero_fd_set(n, fds.res_in); | |
651 | zero_fd_set(n, fds.res_out); | |
652 | zero_fd_set(n, fds.res_ex); | |
653 | ||
8ff3e8e8 | 654 | ret = do_select(n, &fds, end_time); |
1da177e4 LT |
655 | |
656 | if (ret < 0) | |
657 | goto out; | |
658 | if (!ret) { | |
659 | ret = -ERESTARTNOHAND; | |
660 | if (signal_pending(current)) | |
661 | goto out; | |
662 | ret = 0; | |
663 | } | |
664 | ||
665 | if (set_fd_set(n, inp, fds.res_in) || | |
666 | set_fd_set(n, outp, fds.res_out) || | |
667 | set_fd_set(n, exp, fds.res_ex)) | |
668 | ret = -EFAULT; | |
669 | ||
670 | out: | |
70674f95 | 671 | if (bits != stack_fds) |
2d19309c | 672 | kvfree(bits); |
1da177e4 LT |
673 | out_nofds: |
674 | return ret; | |
675 | } | |
676 | ||
4bdb9aca DB |
677 | static int kern_select(int n, fd_set __user *inp, fd_set __user *outp, |
678 | fd_set __user *exp, struct timeval __user *tvp) | |
9f72949f | 679 | { |
766b9f92 | 680 | struct timespec64 end_time, *to = NULL; |
9f72949f DW |
681 | struct timeval tv; |
682 | int ret; | |
683 | ||
684 | if (tvp) { | |
685 | if (copy_from_user(&tv, tvp, sizeof(tv))) | |
686 | return -EFAULT; | |
687 | ||
8ff3e8e8 | 688 | to = &end_time; |
4d36a9e6 AV |
689 | if (poll_select_set_timeout(to, |
690 | tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), | |
691 | (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) | |
9f72949f | 692 | return -EINVAL; |
9f72949f DW |
693 | } |
694 | ||
8ff3e8e8 AV |
695 | ret = core_sys_select(n, inp, outp, exp, to); |
696 | ret = poll_select_copy_remaining(&end_time, tvp, 1, ret); | |
9f72949f DW |
697 | |
698 | return ret; | |
699 | } | |
700 | ||
4bdb9aca DB |
701 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
702 | fd_set __user *, exp, struct timeval __user *, tvp) | |
703 | { | |
704 | return kern_select(n, inp, outp, exp, tvp); | |
705 | } | |
706 | ||
c9da9f21 HC |
707 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
708 | fd_set __user *exp, struct timespec __user *tsp, | |
709 | const sigset_t __user *sigmask, size_t sigsetsize) | |
9f72949f | 710 | { |
9f72949f | 711 | sigset_t ksigmask, sigsaved; |
36819ad0 | 712 | struct timespec64 ts, end_time, *to = NULL; |
9f72949f DW |
713 | int ret; |
714 | ||
715 | if (tsp) { | |
36819ad0 | 716 | if (get_timespec64(&ts, tsp)) |
9f72949f DW |
717 | return -EFAULT; |
718 | ||
8ff3e8e8 | 719 | to = &end_time; |
36819ad0 | 720 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) |
9f72949f | 721 | return -EINVAL; |
9f72949f DW |
722 | } |
723 | ||
724 | if (sigmask) { | |
725 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
726 | if (sigsetsize != sizeof(sigset_t)) | |
727 | return -EINVAL; | |
728 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
729 | return -EFAULT; | |
730 | ||
731 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
732 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
733 | } | |
734 | ||
62568510 | 735 | ret = core_sys_select(n, inp, outp, exp, to); |
8ff3e8e8 | 736 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
737 | |
738 | if (ret == -ERESTARTNOHAND) { | |
739 | /* | |
740 | * Don't restore the signal mask yet. Let do_signal() deliver | |
741 | * the signal on the way back to userspace, before the signal | |
742 | * mask is restored. | |
743 | */ | |
744 | if (sigmask) { | |
745 | memcpy(¤t->saved_sigmask, &sigsaved, | |
746 | sizeof(sigsaved)); | |
4e4c22c7 | 747 | set_restore_sigmask(); |
9f72949f DW |
748 | } |
749 | } else if (sigmask) | |
750 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
751 | ||
752 | return ret; | |
753 | } | |
754 | ||
755 | /* | |
756 | * Most architectures can't handle 7-argument syscalls. So we provide a | |
757 | * 6-argument version where the sixth argument is a pointer to a structure | |
758 | * which has a pointer to the sigset_t itself followed by a size_t containing | |
759 | * the sigset size. | |
760 | */ | |
d4e82042 HC |
761 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
762 | fd_set __user *, exp, struct timespec __user *, tsp, | |
763 | void __user *, sig) | |
9f72949f DW |
764 | { |
765 | size_t sigsetsize = 0; | |
766 | sigset_t __user *up = NULL; | |
767 | ||
768 | if (sig) { | |
769 | if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) | |
e110ab94 | 770 | || __get_user(up, (sigset_t __user * __user *)sig) |
9f72949f | 771 | || __get_user(sigsetsize, |
e110ab94 | 772 | (size_t __user *)(sig+sizeof(void *)))) |
9f72949f DW |
773 | return -EFAULT; |
774 | } | |
775 | ||
c9da9f21 | 776 | return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize); |
9f72949f | 777 | } |
9f72949f | 778 | |
5d0e5283 CH |
779 | #ifdef __ARCH_WANT_SYS_OLD_SELECT |
780 | struct sel_arg_struct { | |
781 | unsigned long n; | |
782 | fd_set __user *inp, *outp, *exp; | |
783 | struct timeval __user *tvp; | |
784 | }; | |
785 | ||
786 | SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) | |
787 | { | |
788 | struct sel_arg_struct a; | |
789 | ||
790 | if (copy_from_user(&a, arg, sizeof(a))) | |
791 | return -EFAULT; | |
4bdb9aca | 792 | return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp); |
5d0e5283 CH |
793 | } |
794 | #endif | |
795 | ||
1da177e4 LT |
796 | struct poll_list { |
797 | struct poll_list *next; | |
798 | int len; | |
799 | struct pollfd entries[0]; | |
800 | }; | |
801 | ||
802 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) | |
803 | ||
4a4b69f7 VL |
804 | /* |
805 | * Fish for pollable events on the pollfd->fd file descriptor. We're only | |
806 | * interested in events matching the pollfd->events mask, and the result | |
807 | * matching that mask is both recorded in pollfd->revents and returned. The | |
808 | * pwait poll_table will be used by the fd-provided poll handler for waiting, | |
626cf236 | 809 | * if pwait->_qproc is non-NULL. |
4a4b69f7 | 810 | */ |
fb367937 | 811 | static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait, |
cbf55001 | 812 | bool *can_busy_poll, |
fb367937 | 813 | __poll_t busy_flag) |
1da177e4 | 814 | { |
a0f8dcfc CH |
815 | int fd = pollfd->fd; |
816 | __poll_t mask = 0, filter; | |
817 | struct fd f; | |
818 | ||
819 | if (fd < 0) | |
820 | goto out; | |
821 | mask = EPOLLNVAL; | |
822 | f = fdget(fd); | |
823 | if (!f.file) | |
824 | goto out; | |
825 | ||
826 | /* userland u16 ->events contains POLL... bitmap */ | |
827 | filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP; | |
828 | mask = DEFAULT_POLLMASK; | |
829 | if (f.file->f_op->poll) { | |
830 | pwait->_key = filter | busy_flag; | |
831 | mask = f.file->f_op->poll(f.file, pwait); | |
832 | if (mask & busy_flag) | |
833 | *can_busy_poll = true; | |
1da177e4 | 834 | } |
a0f8dcfc CH |
835 | mask &= filter; /* Mask out unneeded events. */ |
836 | fdput(f); | |
837 | ||
838 | out: | |
fb367937 | 839 | /* ... and so does ->revents */ |
c71d227f | 840 | pollfd->revents = mangle_poll(mask); |
4a4b69f7 | 841 | return mask; |
1da177e4 LT |
842 | } |
843 | ||
ccec5ee3 | 844 | static int do_poll(struct poll_list *list, struct poll_wqueues *wait, |
766b9f92 | 845 | struct timespec64 *end_time) |
1da177e4 | 846 | { |
1da177e4 | 847 | poll_table* pt = &wait->pt; |
8ff3e8e8 AV |
848 | ktime_t expire, *to = NULL; |
849 | int timed_out = 0, count = 0; | |
da8b44d5 | 850 | u64 slack = 0; |
fb367937 | 851 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
37056719 | 852 | unsigned long busy_start = 0; |
1da177e4 | 853 | |
9f72949f | 854 | /* Optimise the no-wait case */ |
8ff3e8e8 | 855 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
626cf236 | 856 | pt->_qproc = NULL; |
8ff3e8e8 AV |
857 | timed_out = 1; |
858 | } | |
9bf084f7 | 859 | |
96d2ab48 | 860 | if (end_time && !timed_out) |
231f3d39 | 861 | slack = select_estimate_accuracy(end_time); |
90d6e24a | 862 | |
1da177e4 LT |
863 | for (;;) { |
864 | struct poll_list *walk; | |
cbf55001 | 865 | bool can_busy_loop = false; |
9f72949f | 866 | |
4a4b69f7 VL |
867 | for (walk = list; walk != NULL; walk = walk->next) { |
868 | struct pollfd * pfd, * pfd_end; | |
869 | ||
870 | pfd = walk->entries; | |
871 | pfd_end = pfd + walk->len; | |
872 | for (; pfd != pfd_end; pfd++) { | |
873 | /* | |
874 | * Fish for events. If we found one, record it | |
626cf236 | 875 | * and kill poll_table->_qproc, so we don't |
4a4b69f7 VL |
876 | * needlessly register any other waiters after |
877 | * this. They'll get immediately deregistered | |
878 | * when we break out and return. | |
879 | */ | |
cbf55001 ET |
880 | if (do_pollfd(pfd, pt, &can_busy_loop, |
881 | busy_flag)) { | |
4a4b69f7 | 882 | count++; |
626cf236 | 883 | pt->_qproc = NULL; |
cbf55001 ET |
884 | /* found something, stop busy polling */ |
885 | busy_flag = 0; | |
886 | can_busy_loop = false; | |
4a4b69f7 VL |
887 | } |
888 | } | |
1da177e4 | 889 | } |
4a4b69f7 VL |
890 | /* |
891 | * All waiters have already been registered, so don't provide | |
626cf236 | 892 | * a poll_table->_qproc to them on the next loop iteration. |
4a4b69f7 | 893 | */ |
626cf236 | 894 | pt->_qproc = NULL; |
9bf084f7 ON |
895 | if (!count) { |
896 | count = wait->error; | |
897 | if (signal_pending(current)) | |
898 | count = -EINTR; | |
899 | } | |
8ff3e8e8 | 900 | if (count || timed_out) |
1da177e4 | 901 | break; |
9f72949f | 902 | |
cbf55001 | 903 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
76b1e9b9 | 904 | if (can_busy_loop && !need_resched()) { |
37056719 AD |
905 | if (!busy_start) { |
906 | busy_start = busy_loop_current_time(); | |
76b1e9b9 ET |
907 | continue; |
908 | } | |
37056719 | 909 | if (!busy_loop_timeout(busy_start)) |
76b1e9b9 ET |
910 | continue; |
911 | } | |
912 | busy_flag = 0; | |
91e2fd33 | 913 | |
8ff3e8e8 AV |
914 | /* |
915 | * If this is the first loop and we have a timeout | |
916 | * given, then we convert to ktime_t and set the to | |
917 | * pointer to the expiry value. | |
918 | */ | |
919 | if (end_time && !to) { | |
766b9f92 | 920 | expire = timespec64_to_ktime(*end_time); |
8ff3e8e8 | 921 | to = &expire; |
9f72949f DW |
922 | } |
923 | ||
5f820f64 | 924 | if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) |
8ff3e8e8 | 925 | timed_out = 1; |
1da177e4 | 926 | } |
1da177e4 LT |
927 | return count; |
928 | } | |
929 | ||
70674f95 AK |
930 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
931 | sizeof(struct pollfd)) | |
932 | ||
e99ca56c | 933 | static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
766b9f92 | 934 | struct timespec64 *end_time) |
1da177e4 LT |
935 | { |
936 | struct poll_wqueues table; | |
252e5725 | 937 | int err = -EFAULT, fdcount, len, size; |
30c14e40 JS |
938 | /* Allocate small arguments on the stack to save memory and be |
939 | faster - use long to make sure the buffer is aligned properly | |
940 | on 64 bit archs to avoid unaligned access */ | |
941 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; | |
252e5725 ON |
942 | struct poll_list *const head = (struct poll_list *)stack_pps; |
943 | struct poll_list *walk = head; | |
944 | unsigned long todo = nfds; | |
1da177e4 | 945 | |
d554ed89 | 946 | if (nfds > rlimit(RLIMIT_NOFILE)) |
1da177e4 LT |
947 | return -EINVAL; |
948 | ||
252e5725 ON |
949 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
950 | for (;;) { | |
951 | walk->next = NULL; | |
952 | walk->len = len; | |
953 | if (!len) | |
954 | break; | |
1da177e4 | 955 | |
252e5725 ON |
956 | if (copy_from_user(walk->entries, ufds + nfds-todo, |
957 | sizeof(struct pollfd) * walk->len)) | |
958 | goto out_fds; | |
959 | ||
960 | todo -= walk->len; | |
961 | if (!todo) | |
962 | break; | |
1da177e4 | 963 | |
252e5725 ON |
964 | len = min(todo, POLLFD_PER_PAGE); |
965 | size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; | |
966 | walk = walk->next = kmalloc(size, GFP_KERNEL); | |
967 | if (!walk) { | |
968 | err = -ENOMEM; | |
1da177e4 LT |
969 | goto out_fds; |
970 | } | |
1da177e4 | 971 | } |
9f72949f | 972 | |
252e5725 | 973 | poll_initwait(&table); |
ccec5ee3 | 974 | fdcount = do_poll(head, &table, end_time); |
252e5725 | 975 | poll_freewait(&table); |
1da177e4 | 976 | |
252e5725 | 977 | for (walk = head; walk; walk = walk->next) { |
1da177e4 LT |
978 | struct pollfd *fds = walk->entries; |
979 | int j; | |
980 | ||
252e5725 ON |
981 | for (j = 0; j < walk->len; j++, ufds++) |
982 | if (__put_user(fds[j].revents, &ufds->revents)) | |
1da177e4 | 983 | goto out_fds; |
1da177e4 | 984 | } |
252e5725 | 985 | |
1da177e4 | 986 | err = fdcount; |
1da177e4 | 987 | out_fds: |
252e5725 ON |
988 | walk = head->next; |
989 | while (walk) { | |
990 | struct poll_list *pos = walk; | |
991 | walk = walk->next; | |
992 | kfree(pos); | |
1da177e4 | 993 | } |
252e5725 | 994 | |
1da177e4 LT |
995 | return err; |
996 | } | |
9f72949f | 997 | |
3075d9da CW |
998 | static long do_restart_poll(struct restart_block *restart_block) |
999 | { | |
8ff3e8e8 AV |
1000 | struct pollfd __user *ufds = restart_block->poll.ufds; |
1001 | int nfds = restart_block->poll.nfds; | |
766b9f92 | 1002 | struct timespec64 *to = NULL, end_time; |
3075d9da CW |
1003 | int ret; |
1004 | ||
8ff3e8e8 AV |
1005 | if (restart_block->poll.has_timeout) { |
1006 | end_time.tv_sec = restart_block->poll.tv_sec; | |
1007 | end_time.tv_nsec = restart_block->poll.tv_nsec; | |
1008 | to = &end_time; | |
1009 | } | |
1010 | ||
1011 | ret = do_sys_poll(ufds, nfds, to); | |
1012 | ||
3075d9da CW |
1013 | if (ret == -EINTR) { |
1014 | restart_block->fn = do_restart_poll; | |
3075d9da CW |
1015 | ret = -ERESTART_RESTARTBLOCK; |
1016 | } | |
1017 | return ret; | |
1018 | } | |
1019 | ||
5a8a82b1 | 1020 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
faf30900 | 1021 | int, timeout_msecs) |
9f72949f | 1022 | { |
766b9f92 | 1023 | struct timespec64 end_time, *to = NULL; |
3075d9da | 1024 | int ret; |
9f72949f | 1025 | |
8ff3e8e8 AV |
1026 | if (timeout_msecs >= 0) { |
1027 | to = &end_time; | |
1028 | poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, | |
1029 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); | |
9f72949f DW |
1030 | } |
1031 | ||
8ff3e8e8 AV |
1032 | ret = do_sys_poll(ufds, nfds, to); |
1033 | ||
3075d9da CW |
1034 | if (ret == -EINTR) { |
1035 | struct restart_block *restart_block; | |
8ff3e8e8 | 1036 | |
f56141e3 | 1037 | restart_block = ¤t->restart_block; |
3075d9da | 1038 | restart_block->fn = do_restart_poll; |
8ff3e8e8 AV |
1039 | restart_block->poll.ufds = ufds; |
1040 | restart_block->poll.nfds = nfds; | |
1041 | ||
1042 | if (timeout_msecs >= 0) { | |
1043 | restart_block->poll.tv_sec = end_time.tv_sec; | |
1044 | restart_block->poll.tv_nsec = end_time.tv_nsec; | |
1045 | restart_block->poll.has_timeout = 1; | |
1046 | } else | |
1047 | restart_block->poll.has_timeout = 0; | |
1048 | ||
3075d9da CW |
1049 | ret = -ERESTART_RESTARTBLOCK; |
1050 | } | |
1051 | return ret; | |
9f72949f DW |
1052 | } |
1053 | ||
d4e82042 HC |
1054 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
1055 | struct timespec __user *, tsp, const sigset_t __user *, sigmask, | |
1056 | size_t, sigsetsize) | |
9f72949f DW |
1057 | { |
1058 | sigset_t ksigmask, sigsaved; | |
36819ad0 | 1059 | struct timespec64 ts, end_time, *to = NULL; |
9f72949f DW |
1060 | int ret; |
1061 | ||
1062 | if (tsp) { | |
36819ad0 | 1063 | if (get_timespec64(&ts, tsp)) |
9f72949f DW |
1064 | return -EFAULT; |
1065 | ||
8ff3e8e8 AV |
1066 | to = &end_time; |
1067 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1068 | return -EINVAL; | |
9f72949f DW |
1069 | } |
1070 | ||
1071 | if (sigmask) { | |
1072 | /* XXX: Don't preclude handling different sized sigset_t's. */ | |
1073 | if (sigsetsize != sizeof(sigset_t)) | |
1074 | return -EINVAL; | |
1075 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) | |
1076 | return -EFAULT; | |
1077 | ||
1078 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
1079 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1080 | } | |
1081 | ||
8ff3e8e8 | 1082 | ret = do_sys_poll(ufds, nfds, to); |
9f72949f DW |
1083 | |
1084 | /* We can restart this syscall, usually */ | |
1085 | if (ret == -EINTR) { | |
1086 | /* | |
1087 | * Don't restore the signal mask yet. Let do_signal() deliver | |
1088 | * the signal on the way back to userspace, before the signal | |
1089 | * mask is restored. | |
1090 | */ | |
1091 | if (sigmask) { | |
1092 | memcpy(¤t->saved_sigmask, &sigsaved, | |
1093 | sizeof(sigsaved)); | |
4e4c22c7 | 1094 | set_restore_sigmask(); |
9f72949f DW |
1095 | } |
1096 | ret = -ERESTARTNOHAND; | |
1097 | } else if (sigmask) | |
1098 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1099 | ||
8ff3e8e8 | 1100 | ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); |
9f72949f DW |
1101 | |
1102 | return ret; | |
1103 | } | |
e99ca56c AV |
1104 | |
1105 | #ifdef CONFIG_COMPAT | |
1106 | #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t)) | |
1107 | ||
1108 | static | |
36819ad0 | 1109 | int compat_poll_select_copy_remaining(struct timespec64 *end_time, void __user *p, |
e99ca56c AV |
1110 | int timeval, int ret) |
1111 | { | |
36819ad0 | 1112 | struct timespec64 ts; |
e99ca56c AV |
1113 | |
1114 | if (!p) | |
1115 | return ret; | |
1116 | ||
1117 | if (current->personality & STICKY_TIMEOUTS) | |
1118 | goto sticky; | |
1119 | ||
1120 | /* No update for zero timeout */ | |
1121 | if (!end_time->tv_sec && !end_time->tv_nsec) | |
1122 | return ret; | |
1123 | ||
36819ad0 DD |
1124 | ktime_get_ts64(&ts); |
1125 | ts = timespec64_sub(*end_time, ts); | |
e99ca56c AV |
1126 | if (ts.tv_sec < 0) |
1127 | ts.tv_sec = ts.tv_nsec = 0; | |
1128 | ||
1129 | if (timeval) { | |
1130 | struct compat_timeval rtv; | |
1131 | ||
1132 | rtv.tv_sec = ts.tv_sec; | |
1133 | rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC; | |
1134 | ||
1135 | if (!copy_to_user(p, &rtv, sizeof(rtv))) | |
1136 | return ret; | |
1137 | } else { | |
36819ad0 | 1138 | if (!compat_put_timespec64(&ts, p)) |
e99ca56c AV |
1139 | return ret; |
1140 | } | |
1141 | /* | |
1142 | * If an application puts its timeval in read-only memory, we | |
1143 | * don't want the Linux-specific update to the timeval to | |
1144 | * cause a fault after the select has completed | |
1145 | * successfully. However, because we're not updating the | |
1146 | * timeval, we can't restart the system call. | |
1147 | */ | |
1148 | ||
1149 | sticky: | |
1150 | if (ret == -ERESTARTNOHAND) | |
1151 | ret = -EINTR; | |
1152 | return ret; | |
1153 | } | |
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) { | |
1223 | bits = kmalloc(6 * size, GFP_KERNEL); | |
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, | |
1267 | struct compat_timeval __user *tvp) | |
e99ca56c | 1268 | { |
36819ad0 | 1269 | struct timespec64 end_time, *to = NULL; |
e99ca56c AV |
1270 | struct compat_timeval tv; |
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); | |
1285 | ret = compat_poll_select_copy_remaining(&end_time, tvp, 1, ret); | |
1286 | ||
1287 | return ret; | |
1288 | } | |
1289 | ||
05585e44 DB |
1290 | COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp, |
1291 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
1292 | struct compat_timeval __user *, tvp) | |
1293 | { | |
1294 | return do_compat_select(n, inp, outp, exp, tvp); | |
1295 | } | |
1296 | ||
e99ca56c AV |
1297 | struct compat_sel_arg_struct { |
1298 | compat_ulong_t n; | |
1299 | compat_uptr_t inp; | |
1300 | compat_uptr_t outp; | |
1301 | compat_uptr_t exp; | |
1302 | compat_uptr_t tvp; | |
1303 | }; | |
1304 | ||
1305 | COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg) | |
1306 | { | |
1307 | struct compat_sel_arg_struct a; | |
1308 | ||
1309 | if (copy_from_user(&a, arg, sizeof(a))) | |
1310 | return -EFAULT; | |
05585e44 DB |
1311 | return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp), |
1312 | compat_ptr(a.exp), compat_ptr(a.tvp)); | |
e99ca56c AV |
1313 | } |
1314 | ||
1315 | static long do_compat_pselect(int n, compat_ulong_t __user *inp, | |
1316 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, | |
1317 | struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask, | |
1318 | compat_size_t sigsetsize) | |
1319 | { | |
e99ca56c | 1320 | sigset_t ksigmask, sigsaved; |
36819ad0 | 1321 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1322 | int ret; |
1323 | ||
1324 | if (tsp) { | |
36819ad0 | 1325 | if (compat_get_timespec64(&ts, tsp)) |
e99ca56c AV |
1326 | return -EFAULT; |
1327 | ||
1328 | to = &end_time; | |
1329 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1330 | return -EINVAL; | |
1331 | } | |
1332 | ||
1333 | if (sigmask) { | |
1334 | if (sigsetsize != sizeof(compat_sigset_t)) | |
1335 | return -EINVAL; | |
3968cf62 | 1336 | if (get_compat_sigset(&ksigmask, sigmask)) |
e99ca56c | 1337 | return -EFAULT; |
e99ca56c AV |
1338 | |
1339 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
1340 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1341 | } | |
1342 | ||
1343 | ret = compat_core_sys_select(n, inp, outp, exp, to); | |
1344 | ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret); | |
1345 | ||
1346 | if (ret == -ERESTARTNOHAND) { | |
1347 | /* | |
1348 | * Don't restore the signal mask yet. Let do_signal() deliver | |
1349 | * the signal on the way back to userspace, before the signal | |
1350 | * mask is restored. | |
1351 | */ | |
1352 | if (sigmask) { | |
1353 | memcpy(¤t->saved_sigmask, &sigsaved, | |
1354 | sizeof(sigsaved)); | |
1355 | set_restore_sigmask(); | |
1356 | } | |
1357 | } else if (sigmask) | |
1358 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1359 | ||
1360 | return ret; | |
1361 | } | |
1362 | ||
1363 | COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp, | |
1364 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, | |
1365 | struct compat_timespec __user *, tsp, void __user *, sig) | |
1366 | { | |
1367 | compat_size_t sigsetsize = 0; | |
1368 | compat_uptr_t up = 0; | |
1369 | ||
1370 | if (sig) { | |
1371 | if (!access_ok(VERIFY_READ, sig, | |
1372 | sizeof(compat_uptr_t)+sizeof(compat_size_t)) || | |
1373 | __get_user(up, (compat_uptr_t __user *)sig) || | |
1374 | __get_user(sigsetsize, | |
1375 | (compat_size_t __user *)(sig+sizeof(up)))) | |
1376 | return -EFAULT; | |
1377 | } | |
1378 | return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up), | |
1379 | sigsetsize); | |
1380 | } | |
1381 | ||
1382 | COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, | |
1383 | unsigned int, nfds, struct compat_timespec __user *, tsp, | |
1384 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) | |
1385 | { | |
e99ca56c | 1386 | sigset_t ksigmask, sigsaved; |
36819ad0 | 1387 | struct timespec64 ts, end_time, *to = NULL; |
e99ca56c AV |
1388 | int ret; |
1389 | ||
1390 | if (tsp) { | |
36819ad0 | 1391 | if (compat_get_timespec64(&ts, tsp)) |
e99ca56c AV |
1392 | return -EFAULT; |
1393 | ||
1394 | to = &end_time; | |
1395 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) | |
1396 | return -EINVAL; | |
1397 | } | |
1398 | ||
1399 | if (sigmask) { | |
1400 | if (sigsetsize != sizeof(compat_sigset_t)) | |
1401 | return -EINVAL; | |
3968cf62 | 1402 | if (get_compat_sigset(&ksigmask, sigmask)) |
e99ca56c | 1403 | return -EFAULT; |
e99ca56c AV |
1404 | |
1405 | sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
1406 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); | |
1407 | } | |
1408 | ||
1409 | ret = do_sys_poll(ufds, nfds, to); | |
1410 | ||
1411 | /* We can restart this syscall, usually */ | |
1412 | if (ret == -EINTR) { | |
1413 | /* | |
1414 | * Don't restore the signal mask yet. Let do_signal() deliver | |
1415 | * the signal on the way back to userspace, before the signal | |
1416 | * mask is restored. | |
1417 | */ | |
1418 | if (sigmask) { | |
1419 | memcpy(¤t->saved_sigmask, &sigsaved, | |
1420 | sizeof(sigsaved)); | |
1421 | set_restore_sigmask(); | |
1422 | } | |
1423 | ret = -ERESTARTNOHAND; | |
1424 | } else if (sigmask) | |
1425 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1426 | ||
1427 | ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret); | |
1428 | ||
1429 | return ret; | |
1430 | } | |
1431 | #endif |