1 | // SPDX-License-Identifier: GPL-2.0 |
---|---|
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 | |
18 | #include <linux/compat.h> |
19 | #include <linux/kernel.h> |
20 | #include <linux/sched/signal.h> |
21 | #include <linux/sched/rt.h> |
22 | #include <linux/syscalls.h> |
23 | #include <linux/export.h> |
24 | #include <linux/slab.h> |
25 | #include <linux/poll.h> |
26 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ |
27 | #include <linux/file.h> |
28 | #include <linux/fdtable.h> |
29 | #include <linux/fs.h> |
30 | #include <linux/rcupdate.h> |
31 | #include <linux/hrtimer.h> |
32 | #include <linux/freezer.h> |
33 | #include <net/busy_poll.h> |
34 | #include <linux/vmalloc.h> |
35 | |
36 | #include <linux/uaccess.h> |
37 | |
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 | |
51 | #define MAX_SLACK (100 * NSEC_PER_MSEC) |
52 | |
53 | static long __estimate_accuracy(struct timespec64 *tv) |
54 | { |
55 | long slack; |
56 | int divfactor = 1000; |
57 | |
58 | if (tv->tv_sec < 0) |
59 | return 0; |
60 | |
61 | if (task_nice(current) > 0) |
62 | divfactor = divfactor / 5; |
63 | |
64 | if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) |
65 | return MAX_SLACK; |
66 | |
67 | slack = tv->tv_nsec / divfactor; |
68 | slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); |
69 | |
70 | if (slack > MAX_SLACK) |
71 | return MAX_SLACK; |
72 | |
73 | return slack; |
74 | } |
75 | |
76 | u64 select_estimate_accuracy(struct timespec64 *tv) |
77 | { |
78 | u64 ret; |
79 | struct timespec64 now; |
80 | u64 slack = current->timer_slack_ns; |
81 | |
82 | if (slack == 0) |
83 | return 0; |
84 | |
85 | ktime_get_ts64(ts: &now); |
86 | now = timespec64_sub(lhs: *tv, rhs: now); |
87 | ret = __estimate_accuracy(tv: &now); |
88 | if (ret < slack) |
89 | return slack; |
90 | return ret; |
91 | } |
92 | |
93 | |
94 | |
95 | struct poll_table_page { |
96 | struct poll_table_page * next; |
97 | struct poll_table_entry * entry; |
98 | struct poll_table_entry entries[]; |
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 | */ |
116 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
117 | poll_table *p); |
118 | |
119 | void poll_initwait(struct poll_wqueues *pwq) |
120 | { |
121 | init_poll_funcptr(pt: &pwq->pt, qproc: __pollwait); |
122 | pwq->polling_task = current; |
123 | pwq->triggered = 0; |
124 | pwq->error = 0; |
125 | pwq->table = NULL; |
126 | pwq->inline_index = 0; |
127 | } |
128 | EXPORT_SYMBOL(poll_initwait); |
129 | |
130 | static void free_poll_entry(struct poll_table_entry *entry) |
131 | { |
132 | remove_wait_queue(wq_head: entry->wait_address, wq_entry: &entry->wait); |
133 | fput(entry->filp); |
134 | } |
135 | |
136 | void poll_freewait(struct poll_wqueues *pwq) |
137 | { |
138 | struct poll_table_page * p = pwq->table; |
139 | int i; |
140 | for (i = 0; i < pwq->inline_index; i++) |
141 | free_poll_entry(entry: pwq->inline_entries + i); |
142 | while (p) { |
143 | struct poll_table_entry * entry; |
144 | struct poll_table_page *old; |
145 | |
146 | entry = p->entry; |
147 | do { |
148 | entry--; |
149 | free_poll_entry(entry); |
150 | } while (entry > p->entries); |
151 | old = p; |
152 | p = p->next; |
153 | free_page((unsigned long) old); |
154 | } |
155 | } |
156 | EXPORT_SYMBOL(poll_freewait); |
157 | |
158 | static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) |
159 | { |
160 | struct poll_table_page *table = p->table; |
161 | |
162 | if (p->inline_index < N_INLINE_POLL_ENTRIES) |
163 | return p->inline_entries + p->inline_index++; |
164 | |
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; |
171 | return NULL; |
172 | } |
173 | new_table->entry = new_table->entries; |
174 | new_table->next = table; |
175 | p->table = new_table; |
176 | table = new_table; |
177 | } |
178 | |
179 | return table->entry++; |
180 | } |
181 | |
182 | static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
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() |
192 | * and is paired with smp_store_mb() in poll_schedule_timeout. |
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(wq_entry: &dummy_wait, mode, flags: sync, key); |
206 | } |
207 | |
208 | static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
209 | { |
210 | struct poll_table_entry *entry; |
211 | |
212 | entry = container_of(wait, struct poll_table_entry, wait); |
213 | if (key && !(key_to_poll(key) & entry->key)) |
214 | return 0; |
215 | return __pollwake(wait, mode, sync, key); |
216 | } |
217 | |
218 | /* Add a new entry */ |
219 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
220 | poll_table *p) |
221 | { |
222 | struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); |
223 | struct poll_table_entry *entry = poll_get_entry(p: pwq); |
224 | if (!entry) |
225 | return; |
226 | entry->filp = get_file(f: filp); |
227 | entry->wait_address = wait_address; |
228 | entry->key = p->_key; |
229 | init_waitqueue_func_entry(wq_entry: &entry->wait, func: pollwake); |
230 | entry->wait.private = pwq; |
231 | add_wait_queue(wq_head: wait_address, wq_entry: &entry->wait); |
232 | } |
233 | |
234 | static int poll_schedule_timeout(struct poll_wqueues *pwq, int state, |
235 | ktime_t *expires, unsigned long slack) |
236 | { |
237 | int rc = -EINTR; |
238 | |
239 | set_current_state(state); |
240 | if (!pwq->triggered) |
241 | rc = schedule_hrtimeout_range(expires, delta: slack, mode: HRTIMER_MODE_ABS); |
242 | __set_current_state(TASK_RUNNING); |
243 | |
244 | /* |
245 | * Prepare for the next iteration. |
246 | * |
247 | * The following smp_store_mb() serves two purposes. First, it's |
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 | */ |
255 | smp_store_mb(pwq->triggered, 0); |
256 | |
257 | return rc; |
258 | } |
259 | |
260 | /** |
261 | * poll_select_set_timeout - helper function to setup the timeout value |
262 | * @to: pointer to timespec64 variable for the final timeout |
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 | */ |
271 | int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec) |
272 | { |
273 | struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec}; |
274 | |
275 | if (!timespec64_valid(ts: &ts)) |
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 { |
282 | ktime_get_ts64(ts: to); |
283 | *to = timespec64_add_safe(lhs: *to, rhs: ts); |
284 | } |
285 | return 0; |
286 | } |
287 | |
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 | |
295 | static int poll_select_finish(struct timespec64 *end_time, |
296 | void __user *p, |
297 | enum poll_time_type pt_type, int ret) |
298 | { |
299 | struct timespec64 rts; |
300 | |
301 | restore_saved_sigmask_unless(interrupted: ret == -ERESTARTNOHAND); |
302 | |
303 | if (!p) |
304 | return ret; |
305 | |
306 | if (current->personality & STICKY_TIMEOUTS) |
307 | goto sticky; |
308 | |
309 | /* No update for zero timeout */ |
310 | if (!end_time->tv_sec && !end_time->tv_nsec) |
311 | return ret; |
312 | |
313 | ktime_get_ts64(ts: &rts); |
314 | rts = timespec64_sub(lhs: *end_time, rhs: rts); |
315 | if (rts.tv_sec < 0) |
316 | rts.tv_sec = rts.tv_nsec = 0; |
317 | |
318 | |
319 | switch (pt_type) { |
320 | case PT_TIMEVAL: |
321 | { |
322 | struct __kernel_old_timeval rtv; |
323 | |
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(to: p, from: &rtv, n: 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(to: p, from: &rtv, n: sizeof(rtv))) |
339 | return ret; |
340 | } |
341 | break; |
342 | case PT_TIMESPEC: |
343 | if (!put_timespec64(ts: &rts, uts: p)) |
344 | return ret; |
345 | break; |
346 | case PT_OLD_TIMESPEC: |
347 | if (!put_old_timespec32(&rts, p)) |
348 | return ret; |
349 | break; |
350 | default: |
351 | BUG(); |
352 | } |
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 | |
367 | /* |
368 | * Scalable version of the fd_set. |
369 | */ |
370 | |
371 | typedef struct { |
372 | unsigned long *in, *out, *ex; |
373 | unsigned long *res_in, *res_out, *res_ex; |
374 | } fd_set_bits; |
375 | |
376 | /* |
377 | * How many longwords for "nr" bits? |
378 | */ |
379 | #define FDS_BITPERLONG (8*sizeof(long)) |
380 | #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG) |
381 | #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long)) |
382 | |
383 | /* |
384 | * 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(to: fdset, from: ufdset, n: 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(to: ufdset, from: 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 | |
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; |
422 | struct fdtable *fdt; |
423 | |
424 | /* handle last in-complete long-word first */ |
425 | set = ~(~0UL << (n & (BITS_PER_LONG-1))); |
426 | n /= BITS_PER_LONG; |
427 | fdt = files_fdtable(current->files); |
428 | open_fds = fdt->open_fds + n; |
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); |
453 | max += n * BITS_PER_LONG; |
454 | } |
455 | |
456 | return max; |
457 | } |
458 | |
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) |
464 | |
465 | static inline __poll_t select_poll_one(int fd, poll_table *wait, unsigned long in, |
466 | unsigned long out, unsigned long bit, |
467 | __poll_t ll_flag) |
468 | { |
469 | CLASS(fd, f)(fd); |
470 | |
471 | if (fd_empty(f)) |
472 | return EPOLLNVAL; |
473 | |
474 | wait->_key = POLLEX_SET | ll_flag; |
475 | if (in & bit) |
476 | wait->_key |= POLLIN_SET; |
477 | if (out & bit) |
478 | wait->_key |= POLLOUT_SET; |
479 | |
480 | return vfs_poll(fd_file(f), pt: wait); |
481 | } |
482 | |
483 | static noinline_for_stack int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time) |
484 | { |
485 | ktime_t expire, *to = NULL; |
486 | struct poll_wqueues table; |
487 | poll_table *wait; |
488 | int retval, i, timed_out = 0; |
489 | u64 slack = 0; |
490 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
491 | unsigned long busy_start = 0; |
492 | |
493 | rcu_read_lock(); |
494 | retval = max_select_fd(n, fds); |
495 | rcu_read_unlock(); |
496 | |
497 | if (retval < 0) |
498 | return retval; |
499 | n = retval; |
500 | |
501 | poll_initwait(&table); |
502 | wait = &table.pt; |
503 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
504 | wait->_qproc = NULL; |
505 | timed_out = 1; |
506 | } |
507 | |
508 | if (end_time && !timed_out) |
509 | slack = select_estimate_accuracy(tv: end_time); |
510 | |
511 | retval = 0; |
512 | for (;;) { |
513 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; |
514 | bool can_busy_loop = false; |
515 | |
516 | inp = fds->in; outp = fds->out; exp = fds->ex; |
517 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; |
518 | |
519 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { |
520 | unsigned long in, out, ex, all_bits, bit = 1, j; |
521 | unsigned long res_in = 0, res_out = 0, res_ex = 0; |
522 | __poll_t mask; |
523 | |
524 | in = *inp++; out = *outp++; ex = *exp++; |
525 | all_bits = in | out | ex; |
526 | if (all_bits == 0) { |
527 | i += BITS_PER_LONG; |
528 | continue; |
529 | } |
530 | |
531 | for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { |
532 | if (i >= n) |
533 | break; |
534 | if (!(bit & all_bits)) |
535 | continue; |
536 | mask = select_poll_one(fd: i, wait, in, out, bit, |
537 | ll_flag: busy_flag); |
538 | if ((mask & POLLIN_SET) && (in & bit)) { |
539 | res_in |= bit; |
540 | retval++; |
541 | wait->_qproc = NULL; |
542 | } |
543 | if ((mask & POLLOUT_SET) && (out & bit)) { |
544 | res_out |= bit; |
545 | retval++; |
546 | wait->_qproc = NULL; |
547 | } |
548 | if ((mask & POLLEX_SET) && (ex & bit)) { |
549 | res_ex |= bit; |
550 | retval++; |
551 | wait->_qproc = NULL; |
552 | } |
553 | /* got something, stop busy polling */ |
554 | if (retval) { |
555 | can_busy_loop = false; |
556 | busy_flag = 0; |
557 | |
558 | /* |
559 | * only remember a returned |
560 | * POLL_BUSY_LOOP if we asked for it |
561 | */ |
562 | } else if (busy_flag & mask) |
563 | can_busy_loop = true; |
564 | |
565 | } |
566 | if (res_in) |
567 | *rinp = res_in; |
568 | if (res_out) |
569 | *routp = res_out; |
570 | if (res_ex) |
571 | *rexp = res_ex; |
572 | cond_resched(); |
573 | } |
574 | wait->_qproc = NULL; |
575 | if (retval || timed_out || signal_pending(current)) |
576 | break; |
577 | if (table.error) { |
578 | retval = table.error; |
579 | break; |
580 | } |
581 | |
582 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
583 | if (can_busy_loop && !need_resched()) { |
584 | if (!busy_start) { |
585 | busy_start = busy_loop_current_time(); |
586 | continue; |
587 | } |
588 | if (!busy_loop_timeout(start_time: busy_start)) |
589 | continue; |
590 | } |
591 | busy_flag = 0; |
592 | |
593 | /* |
594 | * If this is the first loop and we have a timeout |
595 | * given, then we convert to ktime_t and set the to |
596 | * pointer to the expiry value. |
597 | */ |
598 | if (end_time && !to) { |
599 | expire = timespec64_to_ktime(ts: *end_time); |
600 | to = &expire; |
601 | } |
602 | |
603 | if (!poll_schedule_timeout(pwq: &table, TASK_INTERRUPTIBLE, |
604 | expires: to, slack)) |
605 | timed_out = 1; |
606 | } |
607 | |
608 | poll_freewait(&table); |
609 | |
610 | return retval; |
611 | } |
612 | |
613 | /* |
614 | * We can actually return ERESTARTSYS instead of EINTR, but I'd |
615 | * like to be certain this leads to no problems. So I return |
616 | * EINTR just for safety. |
617 | * |
618 | * Update: ERESTARTSYS breaks at least the xview clock binary, so |
619 | * I'm trying ERESTARTNOHAND which restart only when you want to. |
620 | */ |
621 | int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, |
622 | fd_set __user *exp, struct timespec64 *end_time) |
623 | { |
624 | fd_set_bits fds; |
625 | void *bits; |
626 | int ret, max_fds; |
627 | size_t size, alloc_size; |
628 | struct fdtable *fdt; |
629 | /* Allocate small arguments on the stack to save memory and be faster */ |
630 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
631 | |
632 | ret = -EINVAL; |
633 | if (unlikely(n < 0)) |
634 | goto out_nofds; |
635 | |
636 | /* max_fds can increase, so grab it once to avoid race */ |
637 | rcu_read_lock(); |
638 | fdt = files_fdtable(current->files); |
639 | max_fds = fdt->max_fds; |
640 | rcu_read_unlock(); |
641 | if (n > max_fds) |
642 | n = max_fds; |
643 | |
644 | /* |
645 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), |
646 | * since we used fdset we need to allocate memory in units of |
647 | * long-words. |
648 | */ |
649 | size = FDS_BYTES(n); |
650 | bits = stack_fds; |
651 | if (size > sizeof(stack_fds) / 6) { |
652 | /* Not enough space in on-stack array; must use kmalloc */ |
653 | ret = -ENOMEM; |
654 | if (size > (SIZE_MAX / 6)) |
655 | goto out_nofds; |
656 | |
657 | alloc_size = 6 * size; |
658 | bits = kvmalloc(alloc_size, GFP_KERNEL); |
659 | if (!bits) |
660 | goto out_nofds; |
661 | } |
662 | fds.in = bits; |
663 | fds.out = bits + size; |
664 | fds.ex = bits + 2*size; |
665 | fds.res_in = bits + 3*size; |
666 | fds.res_out = bits + 4*size; |
667 | fds.res_ex = bits + 5*size; |
668 | |
669 | if ((ret = get_fd_set(nr: n, ufdset: inp, fdset: fds.in)) || |
670 | (ret = get_fd_set(nr: n, ufdset: outp, fdset: fds.out)) || |
671 | (ret = get_fd_set(nr: n, ufdset: exp, fdset: fds.ex))) |
672 | goto out; |
673 | zero_fd_set(nr: n, fdset: fds.res_in); |
674 | zero_fd_set(nr: n, fdset: fds.res_out); |
675 | zero_fd_set(nr: n, fdset: fds.res_ex); |
676 | |
677 | ret = do_select(n, fds: &fds, end_time); |
678 | |
679 | if (ret < 0) |
680 | goto out; |
681 | if (!ret) { |
682 | ret = -ERESTARTNOHAND; |
683 | if (signal_pending(current)) |
684 | goto out; |
685 | ret = 0; |
686 | } |
687 | |
688 | if (set_fd_set(nr: n, ufdset: inp, fdset: fds.res_in) || |
689 | set_fd_set(nr: n, ufdset: outp, fdset: fds.res_out) || |
690 | set_fd_set(nr: n, ufdset: exp, fdset: fds.res_ex)) |
691 | ret = -EFAULT; |
692 | |
693 | out: |
694 | if (bits != stack_fds) |
695 | kvfree(addr: bits); |
696 | out_nofds: |
697 | return ret; |
698 | } |
699 | |
700 | static int kern_select(int n, fd_set __user *inp, fd_set __user *outp, |
701 | fd_set __user *exp, struct __kernel_old_timeval __user *tvp) |
702 | { |
703 | struct timespec64 end_time, *to = NULL; |
704 | struct __kernel_old_timeval tv; |
705 | int ret; |
706 | |
707 | if (tvp) { |
708 | if (copy_from_user(to: &tv, from: tvp, n: sizeof(tv))) |
709 | return -EFAULT; |
710 | |
711 | to = &end_time; |
712 | if (poll_select_set_timeout(to, |
713 | sec: tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), |
714 | nsec: (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) |
715 | return -EINVAL; |
716 | } |
717 | |
718 | ret = core_sys_select(n, inp, outp, exp, end_time: to); |
719 | return poll_select_finish(end_time: &end_time, p: tvp, pt_type: PT_TIMEVAL, ret); |
720 | } |
721 | |
722 | SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, |
723 | fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp) |
724 | { |
725 | return kern_select(n, inp, outp, exp, tvp); |
726 | } |
727 | |
728 | static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, |
729 | fd_set __user *exp, void __user *tsp, |
730 | const sigset_t __user *sigmask, size_t sigsetsize, |
731 | enum poll_time_type type) |
732 | { |
733 | struct timespec64 ts, end_time, *to = NULL; |
734 | int ret; |
735 | |
736 | if (tsp) { |
737 | switch (type) { |
738 | case PT_TIMESPEC: |
739 | if (get_timespec64(ts: &ts, uts: tsp)) |
740 | return -EFAULT; |
741 | break; |
742 | case PT_OLD_TIMESPEC: |
743 | if (get_old_timespec32(&ts, tsp)) |
744 | return -EFAULT; |
745 | break; |
746 | default: |
747 | BUG(); |
748 | } |
749 | |
750 | to = &end_time; |
751 | if (poll_select_set_timeout(to, sec: ts.tv_sec, nsec: ts.tv_nsec)) |
752 | return -EINVAL; |
753 | } |
754 | |
755 | ret = set_user_sigmask(umask: sigmask, sigsetsize); |
756 | if (ret) |
757 | return ret; |
758 | |
759 | ret = core_sys_select(n, inp, outp, exp, end_time: to); |
760 | return poll_select_finish(end_time: &end_time, p: tsp, pt_type: type, ret); |
761 | } |
762 | |
763 | /* |
764 | * Most architectures can't handle 7-argument syscalls. So we provide a |
765 | * 6-argument version where the sixth argument is a pointer to a structure |
766 | * which has a pointer to the sigset_t itself followed by a size_t containing |
767 | * the sigset size. |
768 | */ |
769 | struct sigset_argpack { |
770 | sigset_t __user *p; |
771 | size_t size; |
772 | }; |
773 | |
774 | static inline int get_sigset_argpack(struct sigset_argpack *to, |
775 | struct sigset_argpack __user *from) |
776 | { |
777 | // the path is hot enough for overhead of copy_from_user() to matter |
778 | if (from) { |
779 | if (can_do_masked_user_access()) |
780 | from = masked_user_access_begin(from); |
781 | else if (!user_read_access_begin(from, sizeof(*from))) |
782 | return -EFAULT; |
783 | unsafe_get_user(to->p, &from->p, Efault); |
784 | unsafe_get_user(to->size, &from->size, Efault); |
785 | user_read_access_end(); |
786 | } |
787 | return 0; |
788 | Efault: |
789 | user_read_access_end(); |
790 | return -EFAULT; |
791 | } |
792 | |
793 | SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, |
794 | fd_set __user *, exp, struct __kernel_timespec __user *, tsp, |
795 | void __user *, sig) |
796 | { |
797 | struct sigset_argpack x = {NULL, 0}; |
798 | |
799 | if (get_sigset_argpack(to: &x, from: sig)) |
800 | return -EFAULT; |
801 | |
802 | return do_pselect(n, inp, outp, exp, tsp, sigmask: x.p, sigsetsize: x.size, type: PT_TIMESPEC); |
803 | } |
804 | |
805 | #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) |
806 | |
807 | SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp, |
808 | fd_set __user *, exp, struct old_timespec32 __user *, tsp, |
809 | void __user *, sig) |
810 | { |
811 | struct sigset_argpack x = {NULL, 0}; |
812 | |
813 | if (get_sigset_argpack(&x, sig)) |
814 | return -EFAULT; |
815 | |
816 | return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC); |
817 | } |
818 | |
819 | #endif |
820 | |
821 | #ifdef __ARCH_WANT_SYS_OLD_SELECT |
822 | struct sel_arg_struct { |
823 | unsigned long n; |
824 | fd_set __user *inp, *outp, *exp; |
825 | struct __kernel_old_timeval __user *tvp; |
826 | }; |
827 | |
828 | SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) |
829 | { |
830 | struct sel_arg_struct a; |
831 | |
832 | if (copy_from_user(&a, arg, sizeof(a))) |
833 | return -EFAULT; |
834 | return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp); |
835 | } |
836 | #endif |
837 | |
838 | struct poll_list { |
839 | struct poll_list *next; |
840 | unsigned int len; |
841 | struct pollfd entries[] __counted_by(len); |
842 | }; |
843 | |
844 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) |
845 | |
846 | /* |
847 | * Fish for pollable events on the pollfd->fd file descriptor. We're only |
848 | * interested in events matching the pollfd->events mask, and the result |
849 | * matching that mask is both recorded in pollfd->revents and returned. The |
850 | * pwait poll_table will be used by the fd-provided poll handler for waiting, |
851 | * if pwait->_qproc is non-NULL. |
852 | */ |
853 | static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait, |
854 | bool *can_busy_poll, |
855 | __poll_t busy_flag) |
856 | { |
857 | int fd = pollfd->fd; |
858 | __poll_t mask, filter; |
859 | |
860 | if (unlikely(fd < 0)) |
861 | return 0; |
862 | |
863 | CLASS(fd, f)(fd); |
864 | if (fd_empty(f)) |
865 | return EPOLLNVAL; |
866 | |
867 | /* userland u16 ->events contains POLL... bitmap */ |
868 | filter = demangle_poll(val: pollfd->events) | EPOLLERR | EPOLLHUP; |
869 | pwait->_key = filter | busy_flag; |
870 | mask = vfs_poll(fd_file(f), pt: pwait); |
871 | if (mask & busy_flag) |
872 | *can_busy_poll = true; |
873 | return mask & filter; /* Mask out unneeded events. */ |
874 | } |
875 | |
876 | static int do_poll(struct poll_list *list, struct poll_wqueues *wait, |
877 | struct timespec64 *end_time) |
878 | { |
879 | poll_table* pt = &wait->pt; |
880 | ktime_t expire, *to = NULL; |
881 | int timed_out = 0, count = 0; |
882 | u64 slack = 0; |
883 | __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; |
884 | unsigned long busy_start = 0; |
885 | |
886 | /* Optimise the no-wait case */ |
887 | if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { |
888 | pt->_qproc = NULL; |
889 | timed_out = 1; |
890 | } |
891 | |
892 | if (end_time && !timed_out) |
893 | slack = select_estimate_accuracy(tv: end_time); |
894 | |
895 | for (;;) { |
896 | struct poll_list *walk; |
897 | bool can_busy_loop = false; |
898 | |
899 | for (walk = list; walk != NULL; walk = walk->next) { |
900 | struct pollfd * pfd, * pfd_end; |
901 | |
902 | pfd = walk->entries; |
903 | pfd_end = pfd + walk->len; |
904 | for (; pfd != pfd_end; pfd++) { |
905 | __poll_t mask; |
906 | /* |
907 | * Fish for events. If we found one, record it |
908 | * and kill poll_table->_qproc, so we don't |
909 | * needlessly register any other waiters after |
910 | * this. They'll get immediately deregistered |
911 | * when we break out and return. |
912 | */ |
913 | mask = do_pollfd(pollfd: pfd, pwait: pt, can_busy_poll: &can_busy_loop, busy_flag); |
914 | pfd->revents = mangle_poll(val: mask); |
915 | if (mask) { |
916 | count++; |
917 | pt->_qproc = NULL; |
918 | /* found something, stop busy polling */ |
919 | busy_flag = 0; |
920 | can_busy_loop = false; |
921 | } |
922 | } |
923 | } |
924 | /* |
925 | * All waiters have already been registered, so don't provide |
926 | * a poll_table->_qproc to them on the next loop iteration. |
927 | */ |
928 | pt->_qproc = NULL; |
929 | if (!count) { |
930 | count = wait->error; |
931 | if (signal_pending(current)) |
932 | count = -ERESTARTNOHAND; |
933 | } |
934 | if (count || timed_out) |
935 | break; |
936 | |
937 | /* only if found POLL_BUSY_LOOP sockets && not out of time */ |
938 | if (can_busy_loop && !need_resched()) { |
939 | if (!busy_start) { |
940 | busy_start = busy_loop_current_time(); |
941 | continue; |
942 | } |
943 | if (!busy_loop_timeout(start_time: busy_start)) |
944 | continue; |
945 | } |
946 | busy_flag = 0; |
947 | |
948 | /* |
949 | * If this is the first loop and we have a timeout |
950 | * given, then we convert to ktime_t and set the to |
951 | * pointer to the expiry value. |
952 | */ |
953 | if (end_time && !to) { |
954 | expire = timespec64_to_ktime(ts: *end_time); |
955 | to = &expire; |
956 | } |
957 | |
958 | if (!poll_schedule_timeout(pwq: wait, TASK_INTERRUPTIBLE, expires: to, slack)) |
959 | timed_out = 1; |
960 | } |
961 | return count; |
962 | } |
963 | |
964 | #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ |
965 | sizeof(struct pollfd)) |
966 | |
967 | static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, |
968 | struct timespec64 *end_time) |
969 | { |
970 | struct poll_wqueues table; |
971 | int err = -EFAULT, fdcount; |
972 | /* Allocate small arguments on the stack to save memory and be |
973 | faster - use long to make sure the buffer is aligned properly |
974 | on 64 bit archs to avoid unaligned access */ |
975 | long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; |
976 | struct poll_list *const head = (struct poll_list *)stack_pps; |
977 | struct poll_list *walk = head; |
978 | unsigned int todo = nfds; |
979 | unsigned int len; |
980 | |
981 | if (nfds > rlimit(RLIMIT_NOFILE)) |
982 | return -EINVAL; |
983 | |
984 | len = min_t(unsigned int, nfds, N_STACK_PPS); |
985 | for (;;) { |
986 | walk->next = NULL; |
987 | walk->len = len; |
988 | if (!len) |
989 | break; |
990 | |
991 | if (copy_from_user(to: walk->entries, from: ufds + nfds-todo, |
992 | n: sizeof(struct pollfd) * walk->len)) |
993 | goto out_fds; |
994 | |
995 | if (walk->len >= todo) |
996 | break; |
997 | todo -= walk->len; |
998 | |
999 | len = min(todo, POLLFD_PER_PAGE); |
1000 | walk = walk->next = kmalloc(struct_size(walk, entries, len), |
1001 | GFP_KERNEL); |
1002 | if (!walk) { |
1003 | err = -ENOMEM; |
1004 | goto out_fds; |
1005 | } |
1006 | } |
1007 | |
1008 | poll_initwait(&table); |
1009 | fdcount = do_poll(list: head, wait: &table, end_time); |
1010 | poll_freewait(&table); |
1011 | |
1012 | if (!user_write_access_begin(ufds, nfds * sizeof(*ufds))) |
1013 | goto out_fds; |
1014 | |
1015 | for (walk = head; walk; walk = walk->next) { |
1016 | struct pollfd *fds = walk->entries; |
1017 | unsigned int j; |
1018 | |
1019 | for (j = walk->len; j; fds++, ufds++, j--) |
1020 | unsafe_put_user(fds->revents, &ufds->revents, Efault); |
1021 | } |
1022 | user_write_access_end(); |
1023 | |
1024 | err = fdcount; |
1025 | out_fds: |
1026 | walk = head->next; |
1027 | while (walk) { |
1028 | struct poll_list *pos = walk; |
1029 | walk = walk->next; |
1030 | kfree(objp: pos); |
1031 | } |
1032 | |
1033 | return err; |
1034 | |
1035 | Efault: |
1036 | user_write_access_end(); |
1037 | err = -EFAULT; |
1038 | goto out_fds; |
1039 | } |
1040 | |
1041 | static long do_restart_poll(struct restart_block *restart_block) |
1042 | { |
1043 | struct pollfd __user *ufds = restart_block->poll.ufds; |
1044 | int nfds = restart_block->poll.nfds; |
1045 | struct timespec64 *to = NULL, end_time; |
1046 | int ret; |
1047 | |
1048 | if (restart_block->poll.has_timeout) { |
1049 | end_time.tv_sec = restart_block->poll.tv_sec; |
1050 | end_time.tv_nsec = restart_block->poll.tv_nsec; |
1051 | to = &end_time; |
1052 | } |
1053 | |
1054 | ret = do_sys_poll(ufds, nfds, end_time: to); |
1055 | |
1056 | if (ret == -ERESTARTNOHAND) |
1057 | ret = set_restart_fn(restart: restart_block, fn: do_restart_poll); |
1058 | |
1059 | return ret; |
1060 | } |
1061 | |
1062 | SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, |
1063 | int, timeout_msecs) |
1064 | { |
1065 | struct timespec64 end_time, *to = NULL; |
1066 | int ret; |
1067 | |
1068 | if (timeout_msecs >= 0) { |
1069 | to = &end_time; |
1070 | poll_select_set_timeout(to, sec: timeout_msecs / MSEC_PER_SEC, |
1071 | NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); |
1072 | } |
1073 | |
1074 | ret = do_sys_poll(ufds, nfds, end_time: to); |
1075 | |
1076 | if (ret == -ERESTARTNOHAND) { |
1077 | struct restart_block *restart_block; |
1078 | |
1079 | restart_block = ¤t->restart_block; |
1080 | restart_block->poll.ufds = ufds; |
1081 | restart_block->poll.nfds = nfds; |
1082 | |
1083 | if (timeout_msecs >= 0) { |
1084 | restart_block->poll.tv_sec = end_time.tv_sec; |
1085 | restart_block->poll.tv_nsec = end_time.tv_nsec; |
1086 | restart_block->poll.has_timeout = 1; |
1087 | } else |
1088 | restart_block->poll.has_timeout = 0; |
1089 | |
1090 | ret = set_restart_fn(restart: restart_block, fn: do_restart_poll); |
1091 | } |
1092 | return ret; |
1093 | } |
1094 | |
1095 | SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, |
1096 | struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask, |
1097 | size_t, sigsetsize) |
1098 | { |
1099 | struct timespec64 ts, end_time, *to = NULL; |
1100 | int ret; |
1101 | |
1102 | if (tsp) { |
1103 | if (get_timespec64(ts: &ts, uts: tsp)) |
1104 | return -EFAULT; |
1105 | |
1106 | to = &end_time; |
1107 | if (poll_select_set_timeout(to, sec: ts.tv_sec, nsec: ts.tv_nsec)) |
1108 | return -EINVAL; |
1109 | } |
1110 | |
1111 | ret = set_user_sigmask(umask: sigmask, sigsetsize); |
1112 | if (ret) |
1113 | return ret; |
1114 | |
1115 | ret = do_sys_poll(ufds, nfds, end_time: to); |
1116 | return poll_select_finish(end_time: &end_time, p: tsp, pt_type: PT_TIMESPEC, ret); |
1117 | } |
1118 | |
1119 | #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) |
1120 | |
1121 | SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds, |
1122 | struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask, |
1123 | size_t, sigsetsize) |
1124 | { |
1125 | struct timespec64 ts, end_time, *to = NULL; |
1126 | int ret; |
1127 | |
1128 | if (tsp) { |
1129 | if (get_old_timespec32(&ts, tsp)) |
1130 | return -EFAULT; |
1131 | |
1132 | to = &end_time; |
1133 | if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) |
1134 | return -EINVAL; |
1135 | } |
1136 | |
1137 | ret = set_user_sigmask(sigmask, sigsetsize); |
1138 | if (ret) |
1139 | return ret; |
1140 | |
1141 | ret = do_sys_poll(ufds, nfds, to); |
1142 | return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret); |
1143 | } |
1144 | #endif |
1145 | |
1146 | #ifdef CONFIG_COMPAT |
1147 | #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t)) |
1148 | |
1149 | /* |
1150 | * Ooo, nasty. We need here to frob 32-bit unsigned longs to |
1151 | * 64-bit unsigned longs. |
1152 | */ |
1153 | static |
1154 | int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, |
1155 | unsigned long *fdset) |
1156 | { |
1157 | if (ufdset) { |
1158 | return compat_get_bitmap(mask: fdset, umask: ufdset, bitmap_size: nr); |
1159 | } else { |
1160 | zero_fd_set(nr, fdset); |
1161 | return 0; |
1162 | } |
1163 | } |
1164 | |
1165 | static |
1166 | int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, |
1167 | unsigned long *fdset) |
1168 | { |
1169 | if (!ufdset) |
1170 | return 0; |
1171 | return compat_put_bitmap(umask: ufdset, mask: fdset, bitmap_size: nr); |
1172 | } |
1173 | |
1174 | |
1175 | /* |
1176 | * This is a virtual copy of sys_select from fs/select.c and probably |
1177 | * should be compared to it from time to time |
1178 | */ |
1179 | |
1180 | /* |
1181 | * We can actually return ERESTARTSYS instead of EINTR, but I'd |
1182 | * like to be certain this leads to no problems. So I return |
1183 | * EINTR just for safety. |
1184 | * |
1185 | * Update: ERESTARTSYS breaks at least the xview clock binary, so |
1186 | * I'm trying ERESTARTNOHAND which restart only when you want to. |
1187 | */ |
1188 | static int compat_core_sys_select(int n, compat_ulong_t __user *inp, |
1189 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, |
1190 | struct timespec64 *end_time) |
1191 | { |
1192 | fd_set_bits fds; |
1193 | void *bits; |
1194 | int size, max_fds, ret = -EINVAL; |
1195 | struct fdtable *fdt; |
1196 | long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; |
1197 | |
1198 | if (n < 0) |
1199 | goto out_nofds; |
1200 | |
1201 | /* max_fds can increase, so grab it once to avoid race */ |
1202 | rcu_read_lock(); |
1203 | fdt = files_fdtable(current->files); |
1204 | max_fds = fdt->max_fds; |
1205 | rcu_read_unlock(); |
1206 | if (n > max_fds) |
1207 | n = max_fds; |
1208 | |
1209 | /* |
1210 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), |
1211 | * since we used fdset we need to allocate memory in units of |
1212 | * long-words. |
1213 | */ |
1214 | size = FDS_BYTES(n); |
1215 | bits = stack_fds; |
1216 | if (size > sizeof(stack_fds) / 6) { |
1217 | bits = kmalloc_array(6, size, GFP_KERNEL); |
1218 | ret = -ENOMEM; |
1219 | if (!bits) |
1220 | goto out_nofds; |
1221 | } |
1222 | fds.in = (unsigned long *) bits; |
1223 | fds.out = (unsigned long *) (bits + size); |
1224 | fds.ex = (unsigned long *) (bits + 2*size); |
1225 | fds.res_in = (unsigned long *) (bits + 3*size); |
1226 | fds.res_out = (unsigned long *) (bits + 4*size); |
1227 | fds.res_ex = (unsigned long *) (bits + 5*size); |
1228 | |
1229 | if ((ret = compat_get_fd_set(nr: n, ufdset: inp, fdset: fds.in)) || |
1230 | (ret = compat_get_fd_set(nr: n, ufdset: outp, fdset: fds.out)) || |
1231 | (ret = compat_get_fd_set(nr: n, ufdset: exp, fdset: fds.ex))) |
1232 | goto out; |
1233 | zero_fd_set(nr: n, fdset: fds.res_in); |
1234 | zero_fd_set(nr: n, fdset: fds.res_out); |
1235 | zero_fd_set(nr: n, fdset: fds.res_ex); |
1236 | |
1237 | ret = do_select(n, fds: &fds, end_time); |
1238 | |
1239 | if (ret < 0) |
1240 | goto out; |
1241 | if (!ret) { |
1242 | ret = -ERESTARTNOHAND; |
1243 | if (signal_pending(current)) |
1244 | goto out; |
1245 | ret = 0; |
1246 | } |
1247 | |
1248 | if (compat_set_fd_set(nr: n, ufdset: inp, fdset: fds.res_in) || |
1249 | compat_set_fd_set(nr: n, ufdset: outp, fdset: fds.res_out) || |
1250 | compat_set_fd_set(nr: n, ufdset: exp, fdset: fds.res_ex)) |
1251 | ret = -EFAULT; |
1252 | out: |
1253 | if (bits != stack_fds) |
1254 | kfree(objp: bits); |
1255 | out_nofds: |
1256 | return ret; |
1257 | } |
1258 | |
1259 | static int do_compat_select(int n, compat_ulong_t __user *inp, |
1260 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, |
1261 | struct old_timeval32 __user *tvp) |
1262 | { |
1263 | struct timespec64 end_time, *to = NULL; |
1264 | struct old_timeval32 tv; |
1265 | int ret; |
1266 | |
1267 | if (tvp) { |
1268 | if (copy_from_user(to: &tv, from: tvp, n: sizeof(tv))) |
1269 | return -EFAULT; |
1270 | |
1271 | to = &end_time; |
1272 | if (poll_select_set_timeout(to, |
1273 | sec: tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), |
1274 | nsec: (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) |
1275 | return -EINVAL; |
1276 | } |
1277 | |
1278 | ret = compat_core_sys_select(n, inp, outp, exp, end_time: to); |
1279 | return poll_select_finish(end_time: &end_time, p: tvp, pt_type: PT_OLD_TIMEVAL, ret); |
1280 | } |
1281 | |
1282 | COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp, |
1283 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, |
1284 | struct old_timeval32 __user *, tvp) |
1285 | { |
1286 | return do_compat_select(n, inp, outp, exp, tvp); |
1287 | } |
1288 | |
1289 | struct compat_sel_arg_struct { |
1290 | compat_ulong_t n; |
1291 | compat_uptr_t inp; |
1292 | compat_uptr_t outp; |
1293 | compat_uptr_t exp; |
1294 | compat_uptr_t tvp; |
1295 | }; |
1296 | |
1297 | COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg) |
1298 | { |
1299 | struct compat_sel_arg_struct a; |
1300 | |
1301 | if (copy_from_user(to: &a, from: arg, n: sizeof(a))) |
1302 | return -EFAULT; |
1303 | return do_compat_select(n: a.n, inp: compat_ptr(uptr: a.inp), outp: compat_ptr(uptr: a.outp), |
1304 | exp: compat_ptr(uptr: a.exp), tvp: compat_ptr(uptr: a.tvp)); |
1305 | } |
1306 | |
1307 | static long do_compat_pselect(int n, compat_ulong_t __user *inp, |
1308 | compat_ulong_t __user *outp, compat_ulong_t __user *exp, |
1309 | void __user *tsp, compat_sigset_t __user *sigmask, |
1310 | compat_size_t sigsetsize, enum poll_time_type type) |
1311 | { |
1312 | struct timespec64 ts, end_time, *to = NULL; |
1313 | int ret; |
1314 | |
1315 | if (tsp) { |
1316 | switch (type) { |
1317 | case PT_OLD_TIMESPEC: |
1318 | if (get_old_timespec32(&ts, tsp)) |
1319 | return -EFAULT; |
1320 | break; |
1321 | case PT_TIMESPEC: |
1322 | if (get_timespec64(ts: &ts, uts: tsp)) |
1323 | return -EFAULT; |
1324 | break; |
1325 | default: |
1326 | BUG(); |
1327 | } |
1328 | |
1329 | to = &end_time; |
1330 | if (poll_select_set_timeout(to, sec: ts.tv_sec, nsec: ts.tv_nsec)) |
1331 | return -EINVAL; |
1332 | } |
1333 | |
1334 | ret = set_compat_user_sigmask(umask: sigmask, sigsetsize); |
1335 | if (ret) |
1336 | return ret; |
1337 | |
1338 | ret = compat_core_sys_select(n, inp, outp, exp, end_time: to); |
1339 | return poll_select_finish(end_time: &end_time, p: tsp, pt_type: type, ret); |
1340 | } |
1341 | |
1342 | struct compat_sigset_argpack { |
1343 | compat_uptr_t p; |
1344 | compat_size_t size; |
1345 | }; |
1346 | static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to, |
1347 | struct compat_sigset_argpack __user *from) |
1348 | { |
1349 | if (from) { |
1350 | if (!user_read_access_begin(from, sizeof(*from))) |
1351 | return -EFAULT; |
1352 | unsafe_get_user(to->p, &from->p, Efault); |
1353 | unsafe_get_user(to->size, &from->size, Efault); |
1354 | user_read_access_end(); |
1355 | } |
1356 | return 0; |
1357 | Efault: |
1358 | user_read_access_end(); |
1359 | return -EFAULT; |
1360 | } |
1361 | |
1362 | COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp, |
1363 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, |
1364 | struct __kernel_timespec __user *, tsp, void __user *, sig) |
1365 | { |
1366 | struct compat_sigset_argpack x = {0, 0}; |
1367 | |
1368 | if (get_compat_sigset_argpack(to: &x, from: sig)) |
1369 | return -EFAULT; |
1370 | |
1371 | return do_compat_pselect(n, inp, outp, exp, tsp, sigmask: compat_ptr(uptr: x.p), |
1372 | sigsetsize: x.size, type: PT_TIMESPEC); |
1373 | } |
1374 | |
1375 | #if defined(CONFIG_COMPAT_32BIT_TIME) |
1376 | |
1377 | COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp, |
1378 | compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, |
1379 | struct old_timespec32 __user *, tsp, void __user *, sig) |
1380 | { |
1381 | struct compat_sigset_argpack x = {0, 0}; |
1382 | |
1383 | if (get_compat_sigset_argpack(to: &x, from: sig)) |
1384 | return -EFAULT; |
1385 | |
1386 | return do_compat_pselect(n, inp, outp, exp, tsp, sigmask: compat_ptr(uptr: x.p), |
1387 | sigsetsize: x.size, type: PT_OLD_TIMESPEC); |
1388 | } |
1389 | |
1390 | #endif |
1391 | |
1392 | #if defined(CONFIG_COMPAT_32BIT_TIME) |
1393 | COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, |
1394 | unsigned int, nfds, struct old_timespec32 __user *, tsp, |
1395 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) |
1396 | { |
1397 | struct timespec64 ts, end_time, *to = NULL; |
1398 | int ret; |
1399 | |
1400 | if (tsp) { |
1401 | if (get_old_timespec32(&ts, tsp)) |
1402 | return -EFAULT; |
1403 | |
1404 | to = &end_time; |
1405 | if (poll_select_set_timeout(to, sec: ts.tv_sec, nsec: ts.tv_nsec)) |
1406 | return -EINVAL; |
1407 | } |
1408 | |
1409 | ret = set_compat_user_sigmask(umask: sigmask, sigsetsize); |
1410 | if (ret) |
1411 | return ret; |
1412 | |
1413 | ret = do_sys_poll(ufds, nfds, end_time: to); |
1414 | return poll_select_finish(end_time: &end_time, p: tsp, pt_type: PT_OLD_TIMESPEC, ret); |
1415 | } |
1416 | #endif |
1417 | |
1418 | /* New compat syscall for 64 bit time_t*/ |
1419 | COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds, |
1420 | unsigned int, nfds, struct __kernel_timespec __user *, tsp, |
1421 | const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) |
1422 | { |
1423 | struct timespec64 ts, end_time, *to = NULL; |
1424 | int ret; |
1425 | |
1426 | if (tsp) { |
1427 | if (get_timespec64(ts: &ts, uts: tsp)) |
1428 | return -EFAULT; |
1429 | |
1430 | to = &end_time; |
1431 | if (poll_select_set_timeout(to, sec: ts.tv_sec, nsec: ts.tv_nsec)) |
1432 | return -EINVAL; |
1433 | } |
1434 | |
1435 | ret = set_compat_user_sigmask(umask: sigmask, sigsetsize); |
1436 | if (ret) |
1437 | return ret; |
1438 | |
1439 | ret = do_sys_poll(ufds, nfds, end_time: to); |
1440 | return poll_select_finish(end_time: &end_time, p: tsp, pt_type: PT_TIMESPEC, ret); |
1441 | } |
1442 | |
1443 | #endif |
1444 |
Definitions
- __estimate_accuracy
- select_estimate_accuracy
- poll_table_page
- poll_initwait
- free_poll_entry
- poll_freewait
- poll_get_entry
- __pollwake
- pollwake
- __pollwait
- poll_schedule_timeout
- poll_select_set_timeout
- poll_time_type
- poll_select_finish
- get_fd_set
- set_fd_set
- zero_fd_set
- max_select_fd
- select_poll_one
- do_select
- core_sys_select
- kern_select
- do_pselect
- sigset_argpack
- get_sigset_argpack
- poll_list
- do_pollfd
- do_poll
- do_sys_poll
- do_restart_poll
- compat_get_fd_set
- compat_set_fd_set
- compat_core_sys_select
- do_compat_select
- compat_sel_arg_struct
- do_compat_pselect
- compat_sigset_argpack
Improve your Profiling and Debugging skills
Find out more