Line data Source code
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/kernel.h>
19 : #include <linux/sched/signal.h>
20 : #include <linux/sched/rt.h>
21 : #include <linux/syscalls.h>
22 : #include <linux/export.h>
23 : #include <linux/slab.h>
24 : #include <linux/poll.h>
25 : #include <linux/personality.h> /* for STICKY_TIMEOUTS */
26 : #include <linux/file.h>
27 : #include <linux/fdtable.h>
28 : #include <linux/fs.h>
29 : #include <linux/rcupdate.h>
30 : #include <linux/hrtimer.h>
31 : #include <linux/freezer.h>
32 : #include <net/busy_poll.h>
33 : #include <linux/vmalloc.h>
34 :
35 : #include <linux/uaccess.h>
36 :
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 :
50 : #define MAX_SLACK (100 * NSEC_PER_MSEC)
51 :
52 303 : static long __estimate_accuracy(struct timespec64 *tv)
53 : {
54 303 : long slack;
55 303 : int divfactor = 1000;
56 :
57 303 : if (tv->tv_sec < 0)
58 : return 0;
59 :
60 303 : if (task_nice(current) > 0)
61 2 : divfactor = divfactor / 5;
62 :
63 303 : if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
64 : return MAX_SLACK;
65 :
66 296 : slack = tv->tv_nsec / divfactor;
67 296 : slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
68 :
69 296 : if (slack > MAX_SLACK)
70 : return MAX_SLACK;
71 :
72 : return slack;
73 : }
74 :
75 303 : u64 select_estimate_accuracy(struct timespec64 *tv)
76 : {
77 303 : u64 ret;
78 303 : struct timespec64 now;
79 :
80 : /*
81 : * Realtime tasks get a slack of 0 for obvious reasons.
82 : */
83 :
84 303 : if (rt_task(current))
85 : return 0;
86 :
87 303 : ktime_get_ts64(&now);
88 303 : now = timespec64_sub(*tv, now);
89 303 : ret = __estimate_accuracy(&now);
90 303 : if (ret < current->timer_slack_ns)
91 0 : return current->timer_slack_ns;
92 : return ret;
93 : }
94 :
95 :
96 :
97 : struct poll_table_page {
98 : struct poll_table_page * next;
99 : struct poll_table_entry * entry;
100 : struct poll_table_entry entries[];
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 : */
118 : static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
119 : poll_table *p);
120 :
121 1150 : void poll_initwait(struct poll_wqueues *pwq)
122 : {
123 1150 : init_poll_funcptr(&pwq->pt, __pollwait);
124 1150 : pwq->polling_task = current;
125 1150 : pwq->triggered = 0;
126 1150 : pwq->error = 0;
127 1150 : pwq->table = NULL;
128 1150 : pwq->inline_index = 0;
129 0 : }
130 : EXPORT_SYMBOL(poll_initwait);
131 :
132 1473 : static void free_poll_entry(struct poll_table_entry *entry)
133 : {
134 1473 : remove_wait_queue(entry->wait_address, &entry->wait);
135 1473 : fput(entry->filp);
136 1473 : }
137 :
138 1142 : void poll_freewait(struct poll_wqueues *pwq)
139 : {
140 1142 : struct poll_table_page * p = pwq->table;
141 1142 : int i;
142 2615 : for (i = 0; i < pwq->inline_index; i++)
143 1473 : free_poll_entry(pwq->inline_entries + i);
144 1142 : while (p) {
145 0 : struct poll_table_entry * entry;
146 0 : struct poll_table_page *old;
147 :
148 0 : entry = p->entry;
149 0 : do {
150 0 : entry--;
151 0 : free_poll_entry(entry);
152 0 : } while (entry > p->entries);
153 0 : old = p;
154 0 : p = p->next;
155 0 : free_page((unsigned long) old);
156 : }
157 1142 : }
158 : EXPORT_SYMBOL(poll_freewait);
159 :
160 1488 : static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
161 : {
162 1488 : struct poll_table_page *table = p->table;
163 :
164 1488 : if (p->inline_index < N_INLINE_POLL_ENTRIES)
165 1488 : return p->inline_entries + p->inline_index++;
166 :
167 0 : if (!table || POLL_TABLE_FULL(table)) {
168 0 : struct poll_table_page *new_table;
169 :
170 0 : new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
171 0 : if (!new_table) {
172 0 : p->error = -ENOMEM;
173 0 : return NULL;
174 : }
175 0 : new_table->entry = new_table->entries;
176 0 : new_table->next = table;
177 0 : p->table = new_table;
178 0 : table = new_table;
179 : }
180 :
181 0 : return table->entry++;
182 : }
183 :
184 608 : static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
185 : {
186 608 : struct poll_wqueues *pwq = wait->private;
187 608 : 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()
194 : * and is paired with smp_store_mb() in poll_schedule_timeout.
195 : */
196 608 : smp_wmb();
197 608 : 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 : */
207 608 : return default_wake_function(&dummy_wait, mode, sync, key);
208 : }
209 :
210 800 : static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
211 : {
212 800 : struct poll_table_entry *entry;
213 :
214 800 : entry = container_of(wait, struct poll_table_entry, wait);
215 800 : if (key && !(key_to_poll(key) & entry->key))
216 : return 0;
217 608 : return __pollwake(wait, mode, sync, key);
218 : }
219 :
220 : /* Add a new entry */
221 1488 : static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
222 : poll_table *p)
223 : {
224 1488 : struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
225 1488 : struct poll_table_entry *entry = poll_get_entry(pwq);
226 1488 : if (!entry)
227 : return;
228 1488 : entry->filp = get_file(filp);
229 1488 : entry->wait_address = wait_address;
230 1488 : entry->key = p->_key;
231 1488 : init_waitqueue_func_entry(&entry->wait, pollwake);
232 1488 : entry->wait.private = pwq;
233 1488 : add_wait_queue(wait_address, &entry->wait);
234 : }
235 :
236 574 : static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
237 : ktime_t *expires, unsigned long slack)
238 : {
239 574 : int rc = -EINTR;
240 :
241 574 : set_current_state(state);
242 574 : if (!pwq->triggered)
243 574 : rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
244 566 : __set_current_state(TASK_RUNNING);
245 :
246 : /*
247 : * Prepare for the next iteration.
248 : *
249 : * The following smp_store_mb() serves two purposes. First, it's
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 : */
257 566 : smp_store_mb(pwq->triggered, 0);
258 :
259 566 : return rc;
260 : }
261 :
262 : /**
263 : * poll_select_set_timeout - helper function to setup the timeout value
264 : * @to: pointer to timespec64 variable for the final timeout
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 : */
273 268 : int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
274 : {
275 268 : struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
276 :
277 268 : if (!timespec64_valid(&ts))
278 : return -EINVAL;
279 :
280 : /* Optimize for the zero timeout value here */
281 268 : if (!sec && !nsec) {
282 43 : to->tv_sec = to->tv_nsec = 0;
283 : } else {
284 225 : ktime_get_ts64(to);
285 225 : *to = timespec64_add_safe(*to, ts);
286 : }
287 : return 0;
288 : }
289 :
290 : enum poll_time_type {
291 : PT_TIMEVAL = 0,
292 : PT_OLD_TIMEVAL = 1,
293 : PT_TIMESPEC = 2,
294 : PT_OLD_TIMESPEC = 3,
295 : };
296 :
297 777 : static int poll_select_finish(struct timespec64 *end_time,
298 : void __user *p,
299 : enum poll_time_type pt_type, int ret)
300 : {
301 777 : struct timespec64 rts;
302 :
303 777 : restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
304 :
305 777 : if (!p)
306 : return ret;
307 :
308 211 : if (current->personality & STICKY_TIMEOUTS)
309 0 : goto sticky;
310 :
311 : /* No update for zero timeout */
312 211 : if (!end_time->tv_sec && !end_time->tv_nsec)
313 : return ret;
314 :
315 210 : ktime_get_ts64(&rts);
316 210 : rts = timespec64_sub(*end_time, rts);
317 210 : if (rts.tv_sec < 0)
318 2 : rts.tv_sec = rts.tv_nsec = 0;
319 :
320 :
321 210 : switch (pt_type) {
322 : case PT_TIMEVAL:
323 : {
324 7 : struct __kernel_old_timeval rtv;
325 :
326 7 : if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
327 : memset(&rtv, 0, sizeof(rtv));
328 7 : rtv.tv_sec = rts.tv_sec;
329 7 : rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
330 7 : if (!copy_to_user(p, &rtv, sizeof(rtv)))
331 7 : return ret;
332 : }
333 0 : break;
334 0 : case PT_OLD_TIMEVAL:
335 : {
336 0 : struct old_timeval32 rtv;
337 :
338 0 : rtv.tv_sec = rts.tv_sec;
339 0 : rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
340 0 : if (!copy_to_user(p, &rtv, sizeof(rtv)))
341 0 : return ret;
342 : }
343 0 : break;
344 203 : case PT_TIMESPEC:
345 203 : if (!put_timespec64(&rts, p))
346 : return ret;
347 : break;
348 0 : case PT_OLD_TIMESPEC:
349 0 : if (!put_old_timespec32(&rts, p))
350 : return ret;
351 : break;
352 0 : default:
353 0 : BUG();
354 : }
355 : /*
356 : * If an application puts its timeval in read-only memory, we
357 : * don't want the Linux-specific update to the timeval to
358 : * cause a fault after the select has completed
359 : * successfully. However, because we're not updating the
360 : * timeval, we can't restart the system call.
361 : */
362 :
363 0 : sticky:
364 0 : if (ret == -ERESTARTNOHAND)
365 0 : ret = -EINTR;
366 : return ret;
367 : }
368 :
369 : /*
370 : * Scalable version of the fd_set.
371 : */
372 :
373 : typedef struct {
374 : unsigned long *in, *out, *ex;
375 : unsigned long *res_in, *res_out, *res_ex;
376 : } fd_set_bits;
377 :
378 : /*
379 : * How many longwords for "nr" bits?
380 : */
381 : #define FDS_BITPERLONG (8*sizeof(long))
382 : #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
383 : #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
384 :
385 : /*
386 : * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
387 : */
388 : static inline
389 1734 : int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
390 : {
391 1734 : nr = FDS_BYTES(nr);
392 1734 : if (ufdset)
393 2218 : return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
394 :
395 625 : memset(fdset, 0, nr);
396 625 : return 0;
397 : }
398 :
399 : static inline unsigned long __must_check
400 1704 : set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
401 : {
402 1704 : if (ufdset)
403 2200 : return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
404 : return 0;
405 : }
406 :
407 : static inline
408 1734 : void zero_fd_set(unsigned long nr, unsigned long *fdset)
409 : {
410 1734 : memset(fdset, 0, FDS_BYTES(nr));
411 1734 : }
412 :
413 : #define FDS_IN(fds, n) (fds->in + n)
414 : #define FDS_OUT(fds, n) (fds->out + n)
415 : #define FDS_EX(fds, n) (fds->ex + n)
416 :
417 : #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
418 :
419 578 : static int max_select_fd(unsigned long n, fd_set_bits *fds)
420 : {
421 578 : unsigned long *open_fds;
422 578 : unsigned long set;
423 578 : int max;
424 578 : struct fdtable *fdt;
425 :
426 : /* handle last in-complete long-word first */
427 578 : set = ~(~0UL << (n & (BITS_PER_LONG-1)));
428 578 : n /= BITS_PER_LONG;
429 578 : fdt = files_fdtable(current->files);
430 578 : open_fds = fdt->open_fds + n;
431 578 : max = 0;
432 578 : if (set) {
433 578 : set &= BITS(fds, n);
434 578 : if (set) {
435 575 : if (!(set & ~*open_fds))
436 575 : goto get_max;
437 : return -EBADF;
438 : }
439 : }
440 578 : while (n) {
441 0 : open_fds--;
442 0 : n--;
443 0 : set = BITS(fds, n);
444 0 : if (!set)
445 0 : continue;
446 0 : if (set & ~*open_fds)
447 : return -EBADF;
448 0 : if (max)
449 0 : continue;
450 0 : get_max:
451 7457 : do {
452 7457 : max++;
453 7457 : set >>= 1;
454 7457 : } while (set);
455 575 : max += n * BITS_PER_LONG;
456 : }
457 :
458 : return max;
459 : }
460 :
461 : #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR)
462 : #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
463 : #define POLLEX_SET (EPOLLPRI)
464 :
465 2618 : static inline void wait_key_set(poll_table *wait, unsigned long in,
466 : unsigned long out, unsigned long bit,
467 : __poll_t ll_flag)
468 : {
469 2618 : wait->_key = POLLEX_SET | ll_flag;
470 2618 : if (in & bit)
471 2590 : wait->_key |= POLLIN_SET;
472 2618 : if (out & bit)
473 403 : wait->_key |= POLLOUT_SET;
474 : }
475 :
476 578 : static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
477 : {
478 578 : ktime_t expire, *to = NULL;
479 578 : struct poll_wqueues table;
480 578 : poll_table *wait;
481 578 : int retval, i, timed_out = 0;
482 578 : u64 slack = 0;
483 578 : __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
484 578 : unsigned long busy_start = 0;
485 :
486 578 : rcu_read_lock();
487 578 : retval = max_select_fd(n, fds);
488 578 : rcu_read_unlock();
489 :
490 578 : if (retval < 0)
491 : return retval;
492 578 : n = retval;
493 :
494 578 : poll_initwait(&table);
495 578 : wait = &table.pt;
496 578 : if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
497 1 : wait->_qproc = NULL;
498 1 : timed_out = 1;
499 : }
500 :
501 578 : if (end_time && !timed_out)
502 9 : slack = select_estimate_accuracy(end_time);
503 :
504 578 : retval = 0;
505 722 : for (;;) {
506 722 : unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
507 722 : bool can_busy_loop = false;
508 :
509 722 : inp = fds->in; outp = fds->out; exp = fds->ex;
510 722 : rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
511 :
512 1439 : for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
513 717 : unsigned long in, out, ex, all_bits, bit = 1, j;
514 717 : unsigned long res_in = 0, res_out = 0, res_ex = 0;
515 717 : __poll_t mask;
516 :
517 717 : in = *inp++; out = *outp++; ex = *exp++;
518 717 : all_bits = in | out | ex;
519 717 : if (all_bits == 0) {
520 0 : i += BITS_PER_LONG;
521 0 : continue;
522 : }
523 :
524 9774 : for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
525 9774 : struct fd f;
526 9774 : if (i >= n)
527 : break;
528 9057 : if (!(bit & all_bits))
529 6439 : continue;
530 2618 : f = fdget(i);
531 2618 : if (f.file) {
532 2618 : wait_key_set(wait, in, out, bit,
533 : busy_flag);
534 2618 : mask = vfs_poll(f.file, wait);
535 :
536 2618 : fdput(f);
537 2618 : if ((mask & POLLIN_SET) && (in & bit)) {
538 454 : res_in |= bit;
539 454 : retval++;
540 454 : wait->_qproc = NULL;
541 : }
542 2618 : if ((mask & POLLOUT_SET) && (out & bit)) {
543 403 : res_out |= bit;
544 403 : retval++;
545 403 : wait->_qproc = NULL;
546 : }
547 2618 : if ((mask & POLLEX_SET) && (ex & bit)) {
548 0 : res_ex |= bit;
549 0 : retval++;
550 0 : wait->_qproc = NULL;
551 : }
552 : /* got something, stop busy polling */
553 2618 : if (retval) {
554 : can_busy_loop = false;
555 : busy_flag = 0;
556 :
557 : /*
558 : * only remember a returned
559 : * POLL_BUSY_LOOP if we asked for it
560 : */
561 704 : } else if (busy_flag & mask)
562 0 : can_busy_loop = true;
563 :
564 : }
565 : }
566 717 : if (res_in)
567 449 : *rinp = res_in;
568 717 : if (res_out)
569 400 : *routp = res_out;
570 717 : if (res_ex)
571 0 : *rexp = res_ex;
572 717 : cond_resched();
573 : }
574 722 : wait->_qproc = NULL;
575 722 : if (retval || timed_out || signal_pending(current))
576 : break;
577 149 : if (table.error) {
578 : retval = table.error;
579 : break;
580 : }
581 :
582 : /* only if found POLL_BUSY_LOOP sockets && not out of time */
583 149 : if (can_busy_loop && !need_resched()) {
584 0 : if (!busy_start) {
585 0 : busy_start = busy_loop_current_time();
586 0 : continue;
587 : }
588 0 : if (!busy_loop_timeout(busy_start))
589 0 : continue;
590 : }
591 149 : 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 : */
598 149 : if (end_time && !to) {
599 7 : expire = timespec64_to_ktime(*end_time);
600 7 : to = &expire;
601 : }
602 :
603 149 : if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
604 : to, slack))
605 1 : timed_out = 1;
606 : }
607 :
608 573 : poll_freewait(&table);
609 :
610 573 : 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 : */
621 578 : int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
622 : fd_set __user *exp, struct timespec64 *end_time)
623 : {
624 578 : fd_set_bits fds;
625 578 : void *bits;
626 578 : int ret, max_fds;
627 578 : size_t size, alloc_size;
628 578 : struct fdtable *fdt;
629 : /* Allocate small arguments on the stack to save memory and be faster */
630 578 : long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
631 :
632 578 : ret = -EINVAL;
633 578 : if (n < 0)
634 0 : goto out_nofds;
635 :
636 : /* max_fds can increase, so grab it once to avoid race */
637 578 : rcu_read_lock();
638 578 : fdt = files_fdtable(current->files);
639 578 : max_fds = fdt->max_fds;
640 578 : rcu_read_unlock();
641 578 : 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 : */
649 578 : size = FDS_BYTES(n);
650 578 : bits = stack_fds;
651 578 : if (size > sizeof(stack_fds) / 6) {
652 : /* Not enough space in on-stack array; must use kmalloc */
653 0 : ret = -ENOMEM;
654 0 : if (size > (SIZE_MAX / 6))
655 : goto out_nofds;
656 :
657 0 : alloc_size = 6 * size;
658 0 : bits = kvmalloc(alloc_size, GFP_KERNEL);
659 0 : if (!bits)
660 0 : goto out_nofds;
661 : }
662 578 : fds.in = bits;
663 578 : fds.out = bits + size;
664 578 : fds.ex = bits + 2*size;
665 578 : fds.res_in = bits + 3*size;
666 578 : fds.res_out = bits + 4*size;
667 578 : fds.res_ex = bits + 5*size;
668 :
669 578 : if ((ret = get_fd_set(n, inp, fds.in)) ||
670 578 : (ret = get_fd_set(n, outp, fds.out)) ||
671 578 : (ret = get_fd_set(n, exp, fds.ex)))
672 0 : goto out;
673 578 : zero_fd_set(n, fds.res_in);
674 578 : zero_fd_set(n, fds.res_out);
675 578 : zero_fd_set(n, fds.res_ex);
676 :
677 578 : ret = do_select(n, &fds, end_time);
678 :
679 573 : if (ret < 0)
680 0 : goto out;
681 573 : if (!ret) {
682 7 : ret = -ERESTARTNOHAND;
683 7 : if (signal_pending(current))
684 5 : goto out;
685 : ret = 0;
686 : }
687 :
688 1136 : if (set_fd_set(n, inp, fds.res_in) ||
689 1136 : set_fd_set(n, outp, fds.res_out) ||
690 568 : set_fd_set(n, exp, fds.res_ex))
691 : ret = -EFAULT;
692 :
693 568 : out:
694 573 : if (bits != stack_fds)
695 0 : kvfree(bits);
696 573 : out_nofds:
697 573 : return ret;
698 : }
699 :
700 578 : 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 : {
703 578 : struct timespec64 end_time, *to = NULL;
704 578 : struct __kernel_old_timeval tv;
705 578 : int ret;
706 :
707 578 : if (tvp) {
708 10 : if (copy_from_user(&tv, tvp, sizeof(tv)))
709 : return -EFAULT;
710 :
711 10 : to = &end_time;
712 10 : if (poll_select_set_timeout(to,
713 10 : tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
714 10 : (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
715 : return -EINVAL;
716 : }
717 :
718 578 : ret = core_sys_select(n, inp, outp, exp, to);
719 573 : return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
720 : }
721 :
722 1151 : 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 : {
725 578 : return kern_select(n, inp, outp, exp, tvp);
726 : }
727 :
728 0 : 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 0 : struct timespec64 ts, end_time, *to = NULL;
734 0 : int ret;
735 :
736 0 : if (tsp) {
737 0 : switch (type) {
738 0 : case PT_TIMESPEC:
739 0 : if (get_timespec64(&ts, tsp))
740 : return -EFAULT;
741 : break;
742 0 : case PT_OLD_TIMESPEC:
743 0 : if (get_old_timespec32(&ts, tsp))
744 : return -EFAULT;
745 : break;
746 0 : default:
747 0 : BUG();
748 : }
749 :
750 0 : to = &end_time;
751 0 : if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
752 : return -EINVAL;
753 : }
754 :
755 0 : ret = set_user_sigmask(sigmask, sigsetsize);
756 0 : if (ret)
757 0 : return ret;
758 :
759 0 : ret = core_sys_select(n, inp, outp, exp, to);
760 0 : return poll_select_finish(&end_time, tsp, 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 0 : 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 0 : if (from) {
779 0 : if (!user_read_access_begin(from, sizeof(*from)))
780 : return -EFAULT;
781 0 : unsafe_get_user(to->p, &from->p, Efault);
782 0 : unsafe_get_user(to->size, &from->size, Efault);
783 : user_read_access_end();
784 : }
785 : return 0;
786 : Efault:
787 : user_access_end();
788 : return -EFAULT;
789 : }
790 :
791 0 : SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
792 : fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
793 : void __user *, sig)
794 : {
795 0 : struct sigset_argpack x = {NULL, 0};
796 :
797 0 : if (get_sigset_argpack(&x, sig))
798 : return -EFAULT;
799 :
800 0 : return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
801 : }
802 :
803 : #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
804 :
805 : SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
806 : fd_set __user *, exp, struct old_timespec32 __user *, tsp,
807 : void __user *, sig)
808 : {
809 : struct sigset_argpack x = {NULL, 0};
810 :
811 : if (get_sigset_argpack(&x, sig))
812 : return -EFAULT;
813 :
814 : return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
815 : }
816 :
817 : #endif
818 :
819 : #ifdef __ARCH_WANT_SYS_OLD_SELECT
820 : struct sel_arg_struct {
821 : unsigned long n;
822 : fd_set __user *inp, *outp, *exp;
823 : struct __kernel_old_timeval __user *tvp;
824 : };
825 :
826 : SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
827 : {
828 : struct sel_arg_struct a;
829 :
830 : if (copy_from_user(&a, arg, sizeof(a)))
831 : return -EFAULT;
832 : return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
833 : }
834 : #endif
835 :
836 : struct poll_list {
837 : struct poll_list *next;
838 : int len;
839 : struct pollfd entries[];
840 : };
841 :
842 : #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
843 :
844 : /*
845 : * Fish for pollable events on the pollfd->fd file descriptor. We're only
846 : * interested in events matching the pollfd->events mask, and the result
847 : * matching that mask is both recorded in pollfd->revents and returned. The
848 : * pwait poll_table will be used by the fd-provided poll handler for waiting,
849 : * if pwait->_qproc is non-NULL.
850 : */
851 1051 : static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
852 : bool *can_busy_poll,
853 : __poll_t busy_flag)
854 : {
855 1051 : int fd = pollfd->fd;
856 1051 : __poll_t mask = 0, filter;
857 1051 : struct fd f;
858 :
859 1051 : if (fd < 0)
860 0 : goto out;
861 1051 : mask = EPOLLNVAL;
862 1051 : f = fdget(fd);
863 1051 : if (!f.file)
864 0 : goto out;
865 :
866 : /* userland u16 ->events contains POLL... bitmap */
867 1051 : filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
868 1051 : pwait->_key = filter | busy_flag;
869 1051 : mask = vfs_poll(f.file, pwait);
870 1051 : if (mask & busy_flag)
871 0 : *can_busy_poll = true;
872 1051 : mask &= filter; /* Mask out unneeded events. */
873 1506 : fdput(f);
874 :
875 1051 : out:
876 : /* ... and so does ->revents */
877 1051 : pollfd->revents = mangle_poll(mask);
878 1051 : return mask;
879 : }
880 :
881 572 : static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
882 : struct timespec64 *end_time)
883 : {
884 572 : poll_table* pt = &wait->pt;
885 572 : ktime_t expire, *to = NULL;
886 572 : int timed_out = 0, count = 0;
887 572 : u64 slack = 0;
888 572 : __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
889 572 : unsigned long busy_start = 0;
890 :
891 : /* Optimise the no-wait case */
892 572 : if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
893 42 : pt->_qproc = NULL;
894 42 : timed_out = 1;
895 : }
896 :
897 572 : if (end_time && !timed_out)
898 216 : slack = select_estimate_accuracy(end_time);
899 :
900 994 : for (;;) {
901 994 : struct poll_list *walk;
902 994 : bool can_busy_loop = false;
903 :
904 1988 : for (walk = list; walk != NULL; walk = walk->next) {
905 994 : struct pollfd * pfd, * pfd_end;
906 :
907 994 : pfd = walk->entries;
908 994 : pfd_end = pfd + walk->len;
909 2045 : for (; pfd != pfd_end; pfd++) {
910 : /*
911 : * Fish for events. If we found one, record it
912 : * and kill poll_table->_qproc, so we don't
913 : * needlessly register any other waiters after
914 : * this. They'll get immediately deregistered
915 : * when we break out and return.
916 : */
917 1051 : if (do_pollfd(pfd, pt, &can_busy_loop,
918 : busy_flag)) {
919 530 : count++;
920 530 : pt->_qproc = NULL;
921 : /* found something, stop busy polling */
922 530 : busy_flag = 0;
923 530 : can_busy_loop = false;
924 : }
925 : }
926 : }
927 : /*
928 : * All waiters have already been registered, so don't provide
929 : * a poll_table->_qproc to them on the next loop iteration.
930 : */
931 994 : pt->_qproc = NULL;
932 994 : if (!count) {
933 464 : count = wait->error;
934 464 : if (signal_pending(current))
935 0 : count = -ERESTARTNOHAND;
936 : }
937 994 : if (count || timed_out)
938 : break;
939 :
940 : /* only if found POLL_BUSY_LOOP sockets && not out of time */
941 425 : if (can_busy_loop && !need_resched()) {
942 0 : if (!busy_start) {
943 0 : busy_start = busy_loop_current_time();
944 0 : continue;
945 : }
946 0 : if (!busy_loop_timeout(busy_start))
947 0 : continue;
948 : }
949 425 : busy_flag = 0;
950 :
951 : /*
952 : * If this is the first loop and we have a timeout
953 : * given, then we convert to ktime_t and set the to
954 : * pointer to the expiry value.
955 : */
956 425 : if (end_time && !to) {
957 210 : expire = timespec64_to_ktime(*end_time);
958 210 : to = &expire;
959 : }
960 :
961 425 : if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
962 9 : timed_out = 1;
963 : }
964 569 : return count;
965 : }
966 :
967 : #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
968 : sizeof(struct pollfd))
969 :
970 572 : static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
971 : struct timespec64 *end_time)
972 : {
973 572 : struct poll_wqueues table;
974 572 : int err = -EFAULT, fdcount, len;
975 : /* Allocate small arguments on the stack to save memory and be
976 : faster - use long to make sure the buffer is aligned properly
977 : on 64 bit archs to avoid unaligned access */
978 572 : long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
979 572 : struct poll_list *const head = (struct poll_list *)stack_pps;
980 572 : struct poll_list *walk = head;
981 572 : unsigned long todo = nfds;
982 :
983 572 : if (nfds > rlimit(RLIMIT_NOFILE))
984 : return -EINVAL;
985 :
986 572 : len = min_t(unsigned int, nfds, N_STACK_PPS);
987 572 : for (;;) {
988 572 : walk->next = NULL;
989 572 : walk->len = len;
990 572 : if (!len)
991 : break;
992 :
993 572 : if (copy_from_user(walk->entries, ufds + nfds-todo,
994 572 : sizeof(struct pollfd) * walk->len))
995 0 : goto out_fds;
996 :
997 572 : todo -= walk->len;
998 572 : if (!todo)
999 : break;
1000 :
1001 0 : len = min(todo, POLLFD_PER_PAGE);
1002 0 : walk = walk->next = kmalloc(struct_size(walk, entries, len),
1003 : GFP_KERNEL);
1004 0 : if (!walk) {
1005 0 : err = -ENOMEM;
1006 0 : goto out_fds;
1007 : }
1008 : }
1009 :
1010 572 : poll_initwait(&table);
1011 572 : fdcount = do_poll(head, &table, end_time);
1012 569 : poll_freewait(&table);
1013 :
1014 1138 : if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1015 0 : goto out_fds;
1016 :
1017 1138 : for (walk = head; walk; walk = walk->next) {
1018 569 : struct pollfd *fds = walk->entries;
1019 569 : int j;
1020 :
1021 1169 : for (j = walk->len; j; fds++, ufds++, j--)
1022 600 : unsafe_put_user(fds->revents, &ufds->revents, Efault);
1023 : }
1024 : user_write_access_end();
1025 :
1026 : err = fdcount;
1027 569 : out_fds:
1028 569 : walk = head->next;
1029 569 : while (walk) {
1030 0 : struct poll_list *pos = walk;
1031 0 : walk = walk->next;
1032 0 : kfree(pos);
1033 : }
1034 :
1035 : return err;
1036 :
1037 0 : Efault:
1038 0 : user_write_access_end();
1039 0 : err = -EFAULT;
1040 0 : goto out_fds;
1041 : }
1042 :
1043 0 : static long do_restart_poll(struct restart_block *restart_block)
1044 : {
1045 0 : struct pollfd __user *ufds = restart_block->poll.ufds;
1046 0 : int nfds = restart_block->poll.nfds;
1047 0 : struct timespec64 *to = NULL, end_time;
1048 0 : int ret;
1049 :
1050 0 : if (restart_block->poll.has_timeout) {
1051 0 : end_time.tv_sec = restart_block->poll.tv_sec;
1052 0 : end_time.tv_nsec = restart_block->poll.tv_nsec;
1053 0 : to = &end_time;
1054 : }
1055 :
1056 0 : ret = do_sys_poll(ufds, nfds, to);
1057 :
1058 0 : if (ret == -ERESTARTNOHAND) {
1059 0 : restart_block->fn = do_restart_poll;
1060 0 : ret = -ERESTART_RESTARTBLOCK;
1061 : }
1062 0 : return ret;
1063 : }
1064 :
1065 736 : SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1066 : int, timeout_msecs)
1067 : {
1068 368 : struct timespec64 end_time, *to = NULL;
1069 368 : int ret;
1070 :
1071 368 : if (timeout_msecs >= 0) {
1072 55 : to = &end_time;
1073 55 : poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1074 55 : NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1075 : }
1076 :
1077 368 : ret = do_sys_poll(ufds, nfds, to);
1078 :
1079 365 : if (ret == -ERESTARTNOHAND) {
1080 0 : struct restart_block *restart_block;
1081 :
1082 0 : restart_block = ¤t->restart_block;
1083 0 : restart_block->fn = do_restart_poll;
1084 0 : restart_block->poll.ufds = ufds;
1085 0 : restart_block->poll.nfds = nfds;
1086 :
1087 0 : if (timeout_msecs >= 0) {
1088 0 : restart_block->poll.tv_sec = end_time.tv_sec;
1089 0 : restart_block->poll.tv_nsec = end_time.tv_nsec;
1090 0 : restart_block->poll.has_timeout = 1;
1091 : } else
1092 0 : restart_block->poll.has_timeout = 0;
1093 :
1094 : ret = -ERESTART_RESTARTBLOCK;
1095 : }
1096 365 : return ret;
1097 : }
1098 :
1099 408 : SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1100 : struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1101 : size_t, sigsetsize)
1102 : {
1103 204 : struct timespec64 ts, end_time, *to = NULL;
1104 204 : int ret;
1105 :
1106 204 : if (tsp) {
1107 203 : if (get_timespec64(&ts, tsp))
1108 : return -EFAULT;
1109 :
1110 203 : to = &end_time;
1111 203 : if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1112 : return -EINVAL;
1113 : }
1114 :
1115 204 : ret = set_user_sigmask(sigmask, sigsetsize);
1116 204 : if (ret)
1117 0 : return ret;
1118 :
1119 204 : ret = do_sys_poll(ufds, nfds, to);
1120 204 : return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1121 : }
1122 :
1123 : #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1124 :
1125 : SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1126 : struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1127 : size_t, sigsetsize)
1128 : {
1129 : struct timespec64 ts, end_time, *to = NULL;
1130 : int ret;
1131 :
1132 : if (tsp) {
1133 : if (get_old_timespec32(&ts, tsp))
1134 : return -EFAULT;
1135 :
1136 : to = &end_time;
1137 : if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1138 : return -EINVAL;
1139 : }
1140 :
1141 : ret = set_user_sigmask(sigmask, sigsetsize);
1142 : if (ret)
1143 : return ret;
1144 :
1145 : ret = do_sys_poll(ufds, nfds, to);
1146 : return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1147 : }
1148 : #endif
1149 :
1150 : #ifdef CONFIG_COMPAT
1151 : #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1152 :
1153 : /*
1154 : * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1155 : * 64-bit unsigned longs.
1156 : */
1157 : static
1158 0 : int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1159 : unsigned long *fdset)
1160 : {
1161 0 : if (ufdset) {
1162 0 : return compat_get_bitmap(fdset, ufdset, nr);
1163 : } else {
1164 0 : zero_fd_set(nr, fdset);
1165 0 : return 0;
1166 : }
1167 : }
1168 :
1169 : static
1170 0 : int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1171 : unsigned long *fdset)
1172 : {
1173 0 : if (!ufdset)
1174 : return 0;
1175 0 : return compat_put_bitmap(ufdset, fdset, nr);
1176 : }
1177 :
1178 :
1179 : /*
1180 : * This is a virtual copy of sys_select from fs/select.c and probably
1181 : * should be compared to it from time to time
1182 : */
1183 :
1184 : /*
1185 : * We can actually return ERESTARTSYS instead of EINTR, but I'd
1186 : * like to be certain this leads to no problems. So I return
1187 : * EINTR just for safety.
1188 : *
1189 : * Update: ERESTARTSYS breaks at least the xview clock binary, so
1190 : * I'm trying ERESTARTNOHAND which restart only when you want to.
1191 : */
1192 0 : static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1193 : compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1194 : struct timespec64 *end_time)
1195 : {
1196 0 : fd_set_bits fds;
1197 0 : void *bits;
1198 0 : int size, max_fds, ret = -EINVAL;
1199 0 : struct fdtable *fdt;
1200 0 : long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1201 :
1202 0 : if (n < 0)
1203 0 : goto out_nofds;
1204 :
1205 : /* max_fds can increase, so grab it once to avoid race */
1206 0 : rcu_read_lock();
1207 0 : fdt = files_fdtable(current->files);
1208 0 : max_fds = fdt->max_fds;
1209 0 : rcu_read_unlock();
1210 0 : if (n > max_fds)
1211 : n = max_fds;
1212 :
1213 : /*
1214 : * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1215 : * since we used fdset we need to allocate memory in units of
1216 : * long-words.
1217 : */
1218 0 : size = FDS_BYTES(n);
1219 0 : bits = stack_fds;
1220 0 : if (size > sizeof(stack_fds) / 6) {
1221 0 : bits = kmalloc_array(6, size, GFP_KERNEL);
1222 0 : ret = -ENOMEM;
1223 0 : if (!bits)
1224 0 : goto out_nofds;
1225 : }
1226 0 : fds.in = (unsigned long *) bits;
1227 0 : fds.out = (unsigned long *) (bits + size);
1228 0 : fds.ex = (unsigned long *) (bits + 2*size);
1229 0 : fds.res_in = (unsigned long *) (bits + 3*size);
1230 0 : fds.res_out = (unsigned long *) (bits + 4*size);
1231 0 : fds.res_ex = (unsigned long *) (bits + 5*size);
1232 :
1233 0 : if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1234 0 : (ret = compat_get_fd_set(n, outp, fds.out)) ||
1235 0 : (ret = compat_get_fd_set(n, exp, fds.ex)))
1236 0 : goto out;
1237 0 : zero_fd_set(n, fds.res_in);
1238 0 : zero_fd_set(n, fds.res_out);
1239 0 : zero_fd_set(n, fds.res_ex);
1240 :
1241 0 : ret = do_select(n, &fds, end_time);
1242 :
1243 0 : if (ret < 0)
1244 0 : goto out;
1245 0 : if (!ret) {
1246 0 : ret = -ERESTARTNOHAND;
1247 0 : if (signal_pending(current))
1248 0 : goto out;
1249 : ret = 0;
1250 : }
1251 :
1252 0 : if (compat_set_fd_set(n, inp, fds.res_in) ||
1253 0 : compat_set_fd_set(n, outp, fds.res_out) ||
1254 0 : compat_set_fd_set(n, exp, fds.res_ex))
1255 : ret = -EFAULT;
1256 0 : out:
1257 0 : if (bits != stack_fds)
1258 0 : kfree(bits);
1259 0 : out_nofds:
1260 0 : return ret;
1261 : }
1262 :
1263 0 : static int do_compat_select(int n, compat_ulong_t __user *inp,
1264 : compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1265 : struct old_timeval32 __user *tvp)
1266 : {
1267 0 : struct timespec64 end_time, *to = NULL;
1268 0 : struct old_timeval32 tv;
1269 0 : int ret;
1270 :
1271 0 : if (tvp) {
1272 0 : if (copy_from_user(&tv, tvp, sizeof(tv)))
1273 : return -EFAULT;
1274 :
1275 0 : to = &end_time;
1276 0 : if (poll_select_set_timeout(to,
1277 0 : tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1278 0 : (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1279 : return -EINVAL;
1280 : }
1281 :
1282 0 : ret = compat_core_sys_select(n, inp, outp, exp, to);
1283 0 : return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1284 : }
1285 :
1286 0 : COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1287 : compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1288 : struct old_timeval32 __user *, tvp)
1289 : {
1290 0 : return do_compat_select(n, inp, outp, exp, tvp);
1291 : }
1292 :
1293 : struct compat_sel_arg_struct {
1294 : compat_ulong_t n;
1295 : compat_uptr_t inp;
1296 : compat_uptr_t outp;
1297 : compat_uptr_t exp;
1298 : compat_uptr_t tvp;
1299 : };
1300 :
1301 0 : COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1302 : {
1303 0 : struct compat_sel_arg_struct a;
1304 :
1305 0 : if (copy_from_user(&a, arg, sizeof(a)))
1306 : return -EFAULT;
1307 0 : return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1308 0 : compat_ptr(a.exp), compat_ptr(a.tvp));
1309 : }
1310 :
1311 0 : static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1312 : compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1313 : void __user *tsp, compat_sigset_t __user *sigmask,
1314 : compat_size_t sigsetsize, enum poll_time_type type)
1315 : {
1316 0 : struct timespec64 ts, end_time, *to = NULL;
1317 0 : int ret;
1318 :
1319 0 : if (tsp) {
1320 0 : switch (type) {
1321 0 : case PT_OLD_TIMESPEC:
1322 0 : if (get_old_timespec32(&ts, tsp))
1323 : return -EFAULT;
1324 : break;
1325 0 : case PT_TIMESPEC:
1326 0 : if (get_timespec64(&ts, tsp))
1327 : return -EFAULT;
1328 : break;
1329 0 : default:
1330 0 : BUG();
1331 : }
1332 :
1333 0 : to = &end_time;
1334 0 : if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1335 : return -EINVAL;
1336 : }
1337 :
1338 0 : ret = set_compat_user_sigmask(sigmask, sigsetsize);
1339 0 : if (ret)
1340 0 : return ret;
1341 :
1342 0 : ret = compat_core_sys_select(n, inp, outp, exp, to);
1343 0 : return poll_select_finish(&end_time, tsp, type, ret);
1344 : }
1345 :
1346 : struct compat_sigset_argpack {
1347 : compat_uptr_t p;
1348 : compat_size_t size;
1349 : };
1350 0 : static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1351 : struct compat_sigset_argpack __user *from)
1352 : {
1353 0 : if (from) {
1354 0 : if (!user_read_access_begin(from, sizeof(*from)))
1355 : return -EFAULT;
1356 0 : unsafe_get_user(to->p, &from->p, Efault);
1357 0 : unsafe_get_user(to->size, &from->size, Efault);
1358 : user_read_access_end();
1359 : }
1360 : return 0;
1361 : Efault:
1362 : user_access_end();
1363 : return -EFAULT;
1364 : }
1365 :
1366 0 : COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1367 : compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1368 : struct __kernel_timespec __user *, tsp, void __user *, sig)
1369 : {
1370 0 : struct compat_sigset_argpack x = {0, 0};
1371 :
1372 0 : if (get_compat_sigset_argpack(&x, sig))
1373 : return -EFAULT;
1374 :
1375 0 : return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1376 : x.size, PT_TIMESPEC);
1377 : }
1378 :
1379 : #if defined(CONFIG_COMPAT_32BIT_TIME)
1380 :
1381 : COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1382 : compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1383 : struct old_timespec32 __user *, tsp, void __user *, sig)
1384 : {
1385 : struct compat_sigset_argpack x = {0, 0};
1386 :
1387 : if (get_compat_sigset_argpack(&x, sig))
1388 : return -EFAULT;
1389 :
1390 : return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1391 : x.size, PT_OLD_TIMESPEC);
1392 : }
1393 :
1394 : #endif
1395 :
1396 : #if defined(CONFIG_COMPAT_32BIT_TIME)
1397 : COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1398 : unsigned int, nfds, struct old_timespec32 __user *, tsp,
1399 : const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1400 : {
1401 : struct timespec64 ts, end_time, *to = NULL;
1402 : int ret;
1403 :
1404 : if (tsp) {
1405 : if (get_old_timespec32(&ts, tsp))
1406 : return -EFAULT;
1407 :
1408 : to = &end_time;
1409 : if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1410 : return -EINVAL;
1411 : }
1412 :
1413 : ret = set_compat_user_sigmask(sigmask, sigsetsize);
1414 : if (ret)
1415 : return ret;
1416 :
1417 : ret = do_sys_poll(ufds, nfds, to);
1418 : return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1419 : }
1420 : #endif
1421 :
1422 : /* New compat syscall for 64 bit time_t*/
1423 0 : COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1424 : unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1425 : const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1426 : {
1427 0 : struct timespec64 ts, end_time, *to = NULL;
1428 0 : int ret;
1429 :
1430 0 : if (tsp) {
1431 0 : if (get_timespec64(&ts, tsp))
1432 : return -EFAULT;
1433 :
1434 0 : to = &end_time;
1435 0 : if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1436 : return -EINVAL;
1437 : }
1438 :
1439 0 : ret = set_compat_user_sigmask(sigmask, sigsetsize);
1440 0 : if (ret)
1441 0 : return ret;
1442 :
1443 0 : ret = do_sys_poll(ufds, nfds, to);
1444 0 : return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1445 : }
1446 :
1447 : #endif
|
