xref: /openbmc/linux/fs/select.c (revision 8aaaf2f3af2ae212428f4db1af34214225f5cec3)
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 
81 	/*
82 	 * Realtime tasks get a slack of 0 for obvious reasons.
83 	 */
84 
85 	if (rt_task(current))
86 		return 0;
87 
88 	ktime_get_ts64(&now);
89 	now = timespec64_sub(*tv, now);
90 	ret = __estimate_accuracy(&now);
91 	if (ret < current->timer_slack_ns)
92 		return current->timer_slack_ns;
93 	return ret;
94 }
95 
96 
97 
98 struct poll_table_page {
99 	struct poll_table_page * next;
100 	struct poll_table_entry * entry;
101 	struct poll_table_entry entries[];
102 };
103 
104 #define POLL_TABLE_FULL(table) \
105 	((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
106 
107 /*
108  * Ok, Peter made a complicated, but straightforward multiple_wait() function.
109  * I have rewritten this, taking some shortcuts: This code may not be easy to
110  * follow, but it should be free of race-conditions, and it's practical. If you
111  * understand what I'm doing here, then you understand how the linux
112  * sleep/wakeup mechanism works.
113  *
114  * Two very simple procedures, poll_wait() and poll_freewait() make all the
115  * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
116  * as all select/poll functions have to call it to add an entry to the
117  * poll table.
118  */
119 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
120 		       poll_table *p);
121 
122 void poll_initwait(struct poll_wqueues *pwq)
123 {
124 	init_poll_funcptr(&pwq->pt, __pollwait);
125 	pwq->polling_task = current;
126 	pwq->triggered = 0;
127 	pwq->error = 0;
128 	pwq->table = NULL;
129 	pwq->inline_index = 0;
130 }
131 EXPORT_SYMBOL(poll_initwait);
132 
133 static void free_poll_entry(struct poll_table_entry *entry)
134 {
135 	remove_wait_queue(entry->wait_address, &entry->wait);
136 	fput(entry->filp);
137 }
138 
139 void poll_freewait(struct poll_wqueues *pwq)
140 {
141 	struct poll_table_page * p = pwq->table;
142 	int i;
143 	for (i = 0; i < pwq->inline_index; i++)
144 		free_poll_entry(pwq->inline_entries + i);
145 	while (p) {
146 		struct poll_table_entry * entry;
147 		struct poll_table_page *old;
148 
149 		entry = p->entry;
150 		do {
151 			entry--;
152 			free_poll_entry(entry);
153 		} while (entry > p->entries);
154 		old = p;
155 		p = p->next;
156 		free_page((unsigned long) old);
157 	}
158 }
159 EXPORT_SYMBOL(poll_freewait);
160 
161 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
162 {
163 	struct poll_table_page *table = p->table;
164 
165 	if (p->inline_index < N_INLINE_POLL_ENTRIES)
166 		return p->inline_entries + p->inline_index++;
167 
168 	if (!table || POLL_TABLE_FULL(table)) {
169 		struct poll_table_page *new_table;
170 
171 		new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
172 		if (!new_table) {
173 			p->error = -ENOMEM;
174 			return NULL;
175 		}
176 		new_table->entry = new_table->entries;
177 		new_table->next = table;
178 		p->table = new_table;
179 		table = new_table;
180 	}
181 
182 	return table->entry++;
183 }
184 
185 static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
186 {
187 	struct poll_wqueues *pwq = wait->private;
188 	DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
189 
190 	/*
191 	 * Although this function is called under waitqueue lock, LOCK
192 	 * doesn't imply write barrier and the users expect write
193 	 * barrier semantics on wakeup functions.  The following
194 	 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
195 	 * and is paired with smp_store_mb() in poll_schedule_timeout.
196 	 */
197 	smp_wmb();
198 	pwq->triggered = 1;
199 
200 	/*
201 	 * Perform the default wake up operation using a dummy
202 	 * waitqueue.
203 	 *
204 	 * TODO: This is hacky but there currently is no interface to
205 	 * pass in @sync.  @sync is scheduled to be removed and once
206 	 * that happens, wake_up_process() can be used directly.
207 	 */
208 	return default_wake_function(&dummy_wait, mode, sync, key);
209 }
210 
211 static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
212 {
213 	struct poll_table_entry *entry;
214 
215 	entry = container_of(wait, struct poll_table_entry, wait);
216 	if (key && !(key_to_poll(key) & entry->key))
217 		return 0;
218 	return __pollwake(wait, mode, sync, key);
219 }
220 
221 /* Add a new entry */
222 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
223 				poll_table *p)
224 {
225 	struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
226 	struct poll_table_entry *entry = poll_get_entry(pwq);
227 	if (!entry)
228 		return;
229 	entry->filp = get_file(filp);
230 	entry->wait_address = wait_address;
231 	entry->key = p->_key;
232 	init_waitqueue_func_entry(&entry->wait, pollwake);
233 	entry->wait.private = pwq;
234 	add_wait_queue(wait_address, &entry->wait);
235 }
236 
237 static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
238 			  ktime_t *expires, unsigned long slack)
239 {
240 	int rc = -EINTR;
241 
242 	set_current_state(state);
243 	if (!pwq->triggered)
244 		rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
245 	__set_current_state(TASK_RUNNING);
246 
247 	/*
248 	 * Prepare for the next iteration.
249 	 *
250 	 * The following smp_store_mb() serves two purposes.  First, it's
251 	 * the counterpart rmb of the wmb in pollwake() such that data
252 	 * written before wake up is always visible after wake up.
253 	 * Second, the full barrier guarantees that triggered clearing
254 	 * doesn't pass event check of the next iteration.  Note that
255 	 * this problem doesn't exist for the first iteration as
256 	 * add_wait_queue() has full barrier semantics.
257 	 */
258 	smp_store_mb(pwq->triggered, 0);
259 
260 	return rc;
261 }
262 
263 /**
264  * poll_select_set_timeout - helper function to setup the timeout value
265  * @to:		pointer to timespec64 variable for the final timeout
266  * @sec:	seconds (from user space)
267  * @nsec:	nanoseconds (from user space)
268  *
269  * Note, we do not use a timespec for the user space value here, That
270  * way we can use the function for timeval and compat interfaces as well.
271  *
272  * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
273  */
274 int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
275 {
276 	struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
277 
278 	if (!timespec64_valid(&ts))
279 		return -EINVAL;
280 
281 	/* Optimize for the zero timeout value here */
282 	if (!sec && !nsec) {
283 		to->tv_sec = to->tv_nsec = 0;
284 	} else {
285 		ktime_get_ts64(to);
286 		*to = timespec64_add_safe(*to, ts);
287 	}
288 	return 0;
289 }
290 
291 enum poll_time_type {
292 	PT_TIMEVAL = 0,
293 	PT_OLD_TIMEVAL = 1,
294 	PT_TIMESPEC = 2,
295 	PT_OLD_TIMESPEC = 3,
296 };
297 
298 static int poll_select_finish(struct timespec64 *end_time,
299 			      void __user *p,
300 			      enum poll_time_type pt_type, int ret)
301 {
302 	struct timespec64 rts;
303 
304 	restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
305 
306 	if (!p)
307 		return ret;
308 
309 	if (current->personality & STICKY_TIMEOUTS)
310 		goto sticky;
311 
312 	/* No update for zero timeout */
313 	if (!end_time->tv_sec && !end_time->tv_nsec)
314 		return ret;
315 
316 	ktime_get_ts64(&rts);
317 	rts = timespec64_sub(*end_time, rts);
318 	if (rts.tv_sec < 0)
319 		rts.tv_sec = rts.tv_nsec = 0;
320 
321 
322 	switch (pt_type) {
323 	case PT_TIMEVAL:
324 		{
325 			struct __kernel_old_timeval rtv;
326 
327 			if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
328 				memset(&rtv, 0, sizeof(rtv));
329 			rtv.tv_sec = rts.tv_sec;
330 			rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
331 			if (!copy_to_user(p, &rtv, sizeof(rtv)))
332 				return ret;
333 		}
334 		break;
335 	case PT_OLD_TIMEVAL:
336 		{
337 			struct old_timeval32 rtv;
338 
339 			rtv.tv_sec = rts.tv_sec;
340 			rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
341 			if (!copy_to_user(p, &rtv, sizeof(rtv)))
342 				return ret;
343 		}
344 		break;
345 	case PT_TIMESPEC:
346 		if (!put_timespec64(&rts, p))
347 			return ret;
348 		break;
349 	case PT_OLD_TIMESPEC:
350 		if (!put_old_timespec32(&rts, p))
351 			return ret;
352 		break;
353 	default:
354 		BUG();
355 	}
356 	/*
357 	 * If an application puts its timeval in read-only memory, we
358 	 * don't want the Linux-specific update to the timeval to
359 	 * cause a fault after the select has completed
360 	 * successfully. However, because we're not updating the
361 	 * timeval, we can't restart the system call.
362 	 */
363 
364 sticky:
365 	if (ret == -ERESTARTNOHAND)
366 		ret = -EINTR;
367 	return ret;
368 }
369 
370 /*
371  * Scalable version of the fd_set.
372  */
373 
374 typedef struct {
375 	unsigned long *in, *out, *ex;
376 	unsigned long *res_in, *res_out, *res_ex;
377 } fd_set_bits;
378 
379 /*
380  * How many longwords for "nr" bits?
381  */
382 #define FDS_BITPERLONG	(8*sizeof(long))
383 #define FDS_LONGS(nr)	(((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
384 #define FDS_BYTES(nr)	(FDS_LONGS(nr)*sizeof(long))
385 
386 /*
387  * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
388  */
389 static inline
390 int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
391 {
392 	nr = FDS_BYTES(nr);
393 	if (ufdset)
394 		return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
395 
396 	memset(fdset, 0, nr);
397 	return 0;
398 }
399 
400 static inline unsigned long __must_check
401 set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
402 {
403 	if (ufdset)
404 		return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
405 	return 0;
406 }
407 
408 static inline
409 void zero_fd_set(unsigned long nr, unsigned long *fdset)
410 {
411 	memset(fdset, 0, FDS_BYTES(nr));
412 }
413 
414 #define FDS_IN(fds, n)		(fds->in + n)
415 #define FDS_OUT(fds, n)		(fds->out + n)
416 #define FDS_EX(fds, n)		(fds->ex + n)
417 
418 #define BITS(fds, n)	(*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
419 
420 static int max_select_fd(unsigned long n, fd_set_bits *fds)
421 {
422 	unsigned long *open_fds;
423 	unsigned long set;
424 	int max;
425 	struct fdtable *fdt;
426 
427 	/* handle last in-complete long-word first */
428 	set = ~(~0UL << (n & (BITS_PER_LONG-1)));
429 	n /= BITS_PER_LONG;
430 	fdt = files_fdtable(current->files);
431 	open_fds = fdt->open_fds + n;
432 	max = 0;
433 	if (set) {
434 		set &= BITS(fds, n);
435 		if (set) {
436 			if (!(set & ~*open_fds))
437 				goto get_max;
438 			return -EBADF;
439 		}
440 	}
441 	while (n) {
442 		open_fds--;
443 		n--;
444 		set = BITS(fds, n);
445 		if (!set)
446 			continue;
447 		if (set & ~*open_fds)
448 			return -EBADF;
449 		if (max)
450 			continue;
451 get_max:
452 		do {
453 			max++;
454 			set >>= 1;
455 		} while (set);
456 		max += n * BITS_PER_LONG;
457 	}
458 
459 	return max;
460 }
461 
462 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR)
463 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
464 #define POLLEX_SET (EPOLLPRI)
465 
466 static inline void wait_key_set(poll_table *wait, unsigned long in,
467 				unsigned long out, unsigned long bit,
468 				__poll_t ll_flag)
469 {
470 	wait->_key = POLLEX_SET | ll_flag;
471 	if (in & bit)
472 		wait->_key |= POLLIN_SET;
473 	if (out & bit)
474 		wait->_key |= POLLOUT_SET;
475 }
476 
477 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
478 {
479 	ktime_t expire, *to = NULL;
480 	struct poll_wqueues table;
481 	poll_table *wait;
482 	int retval, i, timed_out = 0;
483 	u64 slack = 0;
484 	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
485 	unsigned long busy_start = 0;
486 
487 	rcu_read_lock();
488 	retval = max_select_fd(n, fds);
489 	rcu_read_unlock();
490 
491 	if (retval < 0)
492 		return retval;
493 	n = retval;
494 
495 	poll_initwait(&table);
496 	wait = &table.pt;
497 	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
498 		wait->_qproc = NULL;
499 		timed_out = 1;
500 	}
501 
502 	if (end_time && !timed_out)
503 		slack = select_estimate_accuracy(end_time);
504 
505 	retval = 0;
506 	for (;;) {
507 		unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
508 		bool can_busy_loop = false;
509 
510 		inp = fds->in; outp = fds->out; exp = fds->ex;
511 		rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
512 
513 		for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
514 			unsigned long in, out, ex, all_bits, bit = 1, j;
515 			unsigned long res_in = 0, res_out = 0, res_ex = 0;
516 			__poll_t mask;
517 
518 			in = *inp++; out = *outp++; ex = *exp++;
519 			all_bits = in | out | ex;
520 			if (all_bits == 0) {
521 				i += BITS_PER_LONG;
522 				continue;
523 			}
524 
525 			for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
526 				struct fd f;
527 				if (i >= n)
528 					break;
529 				if (!(bit & all_bits))
530 					continue;
531 				f = fdget(i);
532 				if (f.file) {
533 					wait_key_set(wait, in, out, bit,
534 						     busy_flag);
535 					mask = vfs_poll(f.file, wait);
536 
537 					fdput(f);
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 			}
567 			if (res_in)
568 				*rinp = res_in;
569 			if (res_out)
570 				*routp = res_out;
571 			if (res_ex)
572 				*rexp = res_ex;
573 			cond_resched();
574 		}
575 		wait->_qproc = NULL;
576 		if (retval || timed_out || signal_pending(current))
577 			break;
578 		if (table.error) {
579 			retval = table.error;
580 			break;
581 		}
582 
583 		/* only if found POLL_BUSY_LOOP sockets && not out of time */
584 		if (can_busy_loop && !need_resched()) {
585 			if (!busy_start) {
586 				busy_start = busy_loop_current_time();
587 				continue;
588 			}
589 			if (!busy_loop_timeout(busy_start))
590 				continue;
591 		}
592 		busy_flag = 0;
593 
594 		/*
595 		 * If this is the first loop and we have a timeout
596 		 * given, then we convert to ktime_t and set the to
597 		 * pointer to the expiry value.
598 		 */
599 		if (end_time && !to) {
600 			expire = timespec64_to_ktime(*end_time);
601 			to = &expire;
602 		}
603 
604 		if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
605 					   to, slack))
606 			timed_out = 1;
607 	}
608 
609 	poll_freewait(&table);
610 
611 	return retval;
612 }
613 
614 /*
615  * We can actually return ERESTARTSYS instead of EINTR, but I'd
616  * like to be certain this leads to no problems. So I return
617  * EINTR just for safety.
618  *
619  * Update: ERESTARTSYS breaks at least the xview clock binary, so
620  * I'm trying ERESTARTNOHAND which restart only when you want to.
621  */
622 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
623 			   fd_set __user *exp, struct timespec64 *end_time)
624 {
625 	fd_set_bits fds;
626 	void *bits;
627 	int ret, max_fds;
628 	size_t size, alloc_size;
629 	struct fdtable *fdt;
630 	/* Allocate small arguments on the stack to save memory and be faster */
631 	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
632 
633 	ret = -EINVAL;
634 	if (n < 0)
635 		goto out_nofds;
636 
637 	/* max_fds can increase, so grab it once to avoid race */
638 	rcu_read_lock();
639 	fdt = files_fdtable(current->files);
640 	max_fds = fdt->max_fds;
641 	rcu_read_unlock();
642 	if (n > max_fds)
643 		n = max_fds;
644 
645 	/*
646 	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
647 	 * since we used fdset we need to allocate memory in units of
648 	 * long-words.
649 	 */
650 	size = FDS_BYTES(n);
651 	bits = stack_fds;
652 	if (size > sizeof(stack_fds) / 6) {
653 		/* Not enough space in on-stack array; must use kmalloc */
654 		ret = -ENOMEM;
655 		if (size > (SIZE_MAX / 6))
656 			goto out_nofds;
657 
658 		alloc_size = 6 * size;
659 		bits = kvmalloc(alloc_size, GFP_KERNEL);
660 		if (!bits)
661 			goto out_nofds;
662 	}
663 	fds.in      = bits;
664 	fds.out     = bits +   size;
665 	fds.ex      = bits + 2*size;
666 	fds.res_in  = bits + 3*size;
667 	fds.res_out = bits + 4*size;
668 	fds.res_ex  = bits + 5*size;
669 
670 	if ((ret = get_fd_set(n, inp, fds.in)) ||
671 	    (ret = get_fd_set(n, outp, fds.out)) ||
672 	    (ret = get_fd_set(n, exp, fds.ex)))
673 		goto out;
674 	zero_fd_set(n, fds.res_in);
675 	zero_fd_set(n, fds.res_out);
676 	zero_fd_set(n, fds.res_ex);
677 
678 	ret = do_select(n, &fds, end_time);
679 
680 	if (ret < 0)
681 		goto out;
682 	if (!ret) {
683 		ret = -ERESTARTNOHAND;
684 		if (signal_pending(current))
685 			goto out;
686 		ret = 0;
687 	}
688 
689 	if (set_fd_set(n, inp, fds.res_in) ||
690 	    set_fd_set(n, outp, fds.res_out) ||
691 	    set_fd_set(n, exp, fds.res_ex))
692 		ret = -EFAULT;
693 
694 out:
695 	if (bits != stack_fds)
696 		kvfree(bits);
697 out_nofds:
698 	return ret;
699 }
700 
701 static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
702 		       fd_set __user *exp, struct __kernel_old_timeval __user *tvp)
703 {
704 	struct timespec64 end_time, *to = NULL;
705 	struct __kernel_old_timeval tv;
706 	int ret;
707 
708 	if (tvp) {
709 		if (copy_from_user(&tv, tvp, sizeof(tv)))
710 			return -EFAULT;
711 
712 		to = &end_time;
713 		if (poll_select_set_timeout(to,
714 				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
715 				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
716 			return -EINVAL;
717 	}
718 
719 	ret = core_sys_select(n, inp, outp, exp, to);
720 	return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
721 }
722 
723 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
724 		fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp)
725 {
726 	return kern_select(n, inp, outp, exp, tvp);
727 }
728 
729 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
730 		       fd_set __user *exp, void __user *tsp,
731 		       const sigset_t __user *sigmask, size_t sigsetsize,
732 		       enum poll_time_type type)
733 {
734 	struct timespec64 ts, end_time, *to = NULL;
735 	int ret;
736 
737 	if (tsp) {
738 		switch (type) {
739 		case PT_TIMESPEC:
740 			if (get_timespec64(&ts, tsp))
741 				return -EFAULT;
742 			break;
743 		case PT_OLD_TIMESPEC:
744 			if (get_old_timespec32(&ts, tsp))
745 				return -EFAULT;
746 			break;
747 		default:
748 			BUG();
749 		}
750 
751 		to = &end_time;
752 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
753 			return -EINVAL;
754 	}
755 
756 	ret = set_user_sigmask(sigmask, sigsetsize);
757 	if (ret)
758 		return ret;
759 
760 	ret = core_sys_select(n, inp, outp, exp, to);
761 	return poll_select_finish(&end_time, tsp, type, ret);
762 }
763 
764 /*
765  * Most architectures can't handle 7-argument syscalls. So we provide a
766  * 6-argument version where the sixth argument is a pointer to a structure
767  * which has a pointer to the sigset_t itself followed by a size_t containing
768  * the sigset size.
769  */
770 struct sigset_argpack {
771 	sigset_t __user *p;
772 	size_t size;
773 };
774 
775 static inline int get_sigset_argpack(struct sigset_argpack *to,
776 				     struct sigset_argpack __user *from)
777 {
778 	// the path is hot enough for overhead of copy_from_user() to matter
779 	if (from) {
780 		if (!user_read_access_begin(from, sizeof(*from)))
781 			return -EFAULT;
782 		unsafe_get_user(to->p, &from->p, Efault);
783 		unsafe_get_user(to->size, &from->size, Efault);
784 		user_read_access_end();
785 	}
786 	return 0;
787 Efault:
788 	user_access_end();
789 	return -EFAULT;
790 }
791 
792 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
793 		fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
794 		void __user *, sig)
795 {
796 	struct sigset_argpack x = {NULL, 0};
797 
798 	if (get_sigset_argpack(&x, sig))
799 		return -EFAULT;
800 
801 	return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
802 }
803 
804 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
805 
806 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
807 		fd_set __user *, exp, struct old_timespec32 __user *, tsp,
808 		void __user *, sig)
809 {
810 	struct sigset_argpack x = {NULL, 0};
811 
812 	if (get_sigset_argpack(&x, sig))
813 		return -EFAULT;
814 
815 	return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
816 }
817 
818 #endif
819 
820 #ifdef __ARCH_WANT_SYS_OLD_SELECT
821 struct sel_arg_struct {
822 	unsigned long n;
823 	fd_set __user *inp, *outp, *exp;
824 	struct __kernel_old_timeval __user *tvp;
825 };
826 
827 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
828 {
829 	struct sel_arg_struct a;
830 
831 	if (copy_from_user(&a, arg, sizeof(a)))
832 		return -EFAULT;
833 	return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
834 }
835 #endif
836 
837 struct poll_list {
838 	struct poll_list *next;
839 	int len;
840 	struct pollfd entries[];
841 };
842 
843 #define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
844 
845 /*
846  * Fish for pollable events on the pollfd->fd file descriptor. We're only
847  * interested in events matching the pollfd->events mask, and the result
848  * matching that mask is both recorded in pollfd->revents and returned. The
849  * pwait poll_table will be used by the fd-provided poll handler for waiting,
850  * if pwait->_qproc is non-NULL.
851  */
852 static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
853 				     bool *can_busy_poll,
854 				     __poll_t busy_flag)
855 {
856 	int fd = pollfd->fd;
857 	__poll_t mask = 0, filter;
858 	struct fd f;
859 
860 	if (fd < 0)
861 		goto out;
862 	mask = EPOLLNVAL;
863 	f = fdget(fd);
864 	if (!f.file)
865 		goto out;
866 
867 	/* userland u16 ->events contains POLL... bitmap */
868 	filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
869 	pwait->_key = filter | busy_flag;
870 	mask = vfs_poll(f.file, pwait);
871 	if (mask & busy_flag)
872 		*can_busy_poll = true;
873 	mask &= filter;		/* Mask out unneeded events. */
874 	fdput(f);
875 
876 out:
877 	/* ... and so does ->revents */
878 	pollfd->revents = mangle_poll(mask);
879 	return mask;
880 }
881 
882 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
883 		   struct timespec64 *end_time)
884 {
885 	poll_table* pt = &wait->pt;
886 	ktime_t expire, *to = NULL;
887 	int timed_out = 0, count = 0;
888 	u64 slack = 0;
889 	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
890 	unsigned long busy_start = 0;
891 
892 	/* Optimise the no-wait case */
893 	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
894 		pt->_qproc = NULL;
895 		timed_out = 1;
896 	}
897 
898 	if (end_time && !timed_out)
899 		slack = select_estimate_accuracy(end_time);
900 
901 	for (;;) {
902 		struct poll_list *walk;
903 		bool can_busy_loop = false;
904 
905 		for (walk = list; walk != NULL; walk = walk->next) {
906 			struct pollfd * pfd, * pfd_end;
907 
908 			pfd = walk->entries;
909 			pfd_end = pfd + walk->len;
910 			for (; pfd != pfd_end; pfd++) {
911 				/*
912 				 * Fish for events. If we found one, record it
913 				 * and kill poll_table->_qproc, so we don't
914 				 * needlessly register any other waiters after
915 				 * this. They'll get immediately deregistered
916 				 * when we break out and return.
917 				 */
918 				if (do_pollfd(pfd, pt, &can_busy_loop,
919 					      busy_flag)) {
920 					count++;
921 					pt->_qproc = NULL;
922 					/* found something, stop busy polling */
923 					busy_flag = 0;
924 					can_busy_loop = false;
925 				}
926 			}
927 		}
928 		/*
929 		 * All waiters have already been registered, so don't provide
930 		 * a poll_table->_qproc to them on the next loop iteration.
931 		 */
932 		pt->_qproc = NULL;
933 		if (!count) {
934 			count = wait->error;
935 			if (signal_pending(current))
936 				count = -ERESTARTNOHAND;
937 		}
938 		if (count || timed_out)
939 			break;
940 
941 		/* only if found POLL_BUSY_LOOP sockets && not out of time */
942 		if (can_busy_loop && !need_resched()) {
943 			if (!busy_start) {
944 				busy_start = busy_loop_current_time();
945 				continue;
946 			}
947 			if (!busy_loop_timeout(busy_start))
948 				continue;
949 		}
950 		busy_flag = 0;
951 
952 		/*
953 		 * If this is the first loop and we have a timeout
954 		 * given, then we convert to ktime_t and set the to
955 		 * pointer to the expiry value.
956 		 */
957 		if (end_time && !to) {
958 			expire = timespec64_to_ktime(*end_time);
959 			to = &expire;
960 		}
961 
962 		if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
963 			timed_out = 1;
964 	}
965 	return count;
966 }
967 
968 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
969 			sizeof(struct pollfd))
970 
971 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
972 		struct timespec64 *end_time)
973 {
974 	struct poll_wqueues table;
975 	int err = -EFAULT, fdcount, len;
976 	/* Allocate small arguments on the stack to save memory and be
977 	   faster - use long to make sure the buffer is aligned properly
978 	   on 64 bit archs to avoid unaligned access */
979 	long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
980 	struct poll_list *const head = (struct poll_list *)stack_pps;
981  	struct poll_list *walk = head;
982  	unsigned long todo = nfds;
983 
984 	if (nfds > rlimit(RLIMIT_NOFILE))
985 		return -EINVAL;
986 
987 	len = min_t(unsigned int, nfds, N_STACK_PPS);
988 	for (;;) {
989 		walk->next = NULL;
990 		walk->len = len;
991 		if (!len)
992 			break;
993 
994 		if (copy_from_user(walk->entries, ufds + nfds-todo,
995 					sizeof(struct pollfd) * walk->len))
996 			goto out_fds;
997 
998 		todo -= walk->len;
999 		if (!todo)
1000 			break;
1001 
1002 		len = min(todo, POLLFD_PER_PAGE);
1003 		walk = walk->next = kmalloc(struct_size(walk, entries, len),
1004 					    GFP_KERNEL);
1005 		if (!walk) {
1006 			err = -ENOMEM;
1007 			goto out_fds;
1008 		}
1009 	}
1010 
1011 	poll_initwait(&table);
1012 	fdcount = do_poll(head, &table, end_time);
1013 	poll_freewait(&table);
1014 
1015 	if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1016 		goto out_fds;
1017 
1018 	for (walk = head; walk; walk = walk->next) {
1019 		struct pollfd *fds = walk->entries;
1020 		int j;
1021 
1022 		for (j = walk->len; j; fds++, ufds++, j--)
1023 			unsafe_put_user(fds->revents, &ufds->revents, Efault);
1024   	}
1025 	user_write_access_end();
1026 
1027 	err = fdcount;
1028 out_fds:
1029 	walk = head->next;
1030 	while (walk) {
1031 		struct poll_list *pos = walk;
1032 		walk = walk->next;
1033 		kfree(pos);
1034 	}
1035 
1036 	return err;
1037 
1038 Efault:
1039 	user_write_access_end();
1040 	err = -EFAULT;
1041 	goto out_fds;
1042 }
1043 
1044 static long do_restart_poll(struct restart_block *restart_block)
1045 {
1046 	struct pollfd __user *ufds = restart_block->poll.ufds;
1047 	int nfds = restart_block->poll.nfds;
1048 	struct timespec64 *to = NULL, end_time;
1049 	int ret;
1050 
1051 	if (restart_block->poll.has_timeout) {
1052 		end_time.tv_sec = restart_block->poll.tv_sec;
1053 		end_time.tv_nsec = restart_block->poll.tv_nsec;
1054 		to = &end_time;
1055 	}
1056 
1057 	ret = do_sys_poll(ufds, nfds, to);
1058 
1059 	if (ret == -ERESTARTNOHAND)
1060 		ret = set_restart_fn(restart_block, do_restart_poll);
1061 
1062 	return ret;
1063 }
1064 
1065 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1066 		int, timeout_msecs)
1067 {
1068 	struct timespec64 end_time, *to = NULL;
1069 	int ret;
1070 
1071 	if (timeout_msecs >= 0) {
1072 		to = &end_time;
1073 		poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1074 			NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1075 	}
1076 
1077 	ret = do_sys_poll(ufds, nfds, to);
1078 
1079 	if (ret == -ERESTARTNOHAND) {
1080 		struct restart_block *restart_block;
1081 
1082 		restart_block = &current->restart_block;
1083 		restart_block->poll.ufds = ufds;
1084 		restart_block->poll.nfds = nfds;
1085 
1086 		if (timeout_msecs >= 0) {
1087 			restart_block->poll.tv_sec = end_time.tv_sec;
1088 			restart_block->poll.tv_nsec = end_time.tv_nsec;
1089 			restart_block->poll.has_timeout = 1;
1090 		} else
1091 			restart_block->poll.has_timeout = 0;
1092 
1093 		ret = set_restart_fn(restart_block, do_restart_poll);
1094 	}
1095 	return ret;
1096 }
1097 
1098 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1099 		struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1100 		size_t, sigsetsize)
1101 {
1102 	struct timespec64 ts, end_time, *to = NULL;
1103 	int ret;
1104 
1105 	if (tsp) {
1106 		if (get_timespec64(&ts, tsp))
1107 			return -EFAULT;
1108 
1109 		to = &end_time;
1110 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1111 			return -EINVAL;
1112 	}
1113 
1114 	ret = set_user_sigmask(sigmask, sigsetsize);
1115 	if (ret)
1116 		return ret;
1117 
1118 	ret = do_sys_poll(ufds, nfds, to);
1119 	return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1120 }
1121 
1122 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1123 
1124 SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1125 		struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1126 		size_t, sigsetsize)
1127 {
1128 	struct timespec64 ts, end_time, *to = NULL;
1129 	int ret;
1130 
1131 	if (tsp) {
1132 		if (get_old_timespec32(&ts, tsp))
1133 			return -EFAULT;
1134 
1135 		to = &end_time;
1136 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1137 			return -EINVAL;
1138 	}
1139 
1140 	ret = set_user_sigmask(sigmask, sigsetsize);
1141 	if (ret)
1142 		return ret;
1143 
1144 	ret = do_sys_poll(ufds, nfds, to);
1145 	return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1146 }
1147 #endif
1148 
1149 #ifdef CONFIG_COMPAT
1150 #define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1151 
1152 /*
1153  * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1154  * 64-bit unsigned longs.
1155  */
1156 static
1157 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1158 			unsigned long *fdset)
1159 {
1160 	if (ufdset) {
1161 		return compat_get_bitmap(fdset, ufdset, nr);
1162 	} else {
1163 		zero_fd_set(nr, fdset);
1164 		return 0;
1165 	}
1166 }
1167 
1168 static
1169 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1170 		      unsigned long *fdset)
1171 {
1172 	if (!ufdset)
1173 		return 0;
1174 	return compat_put_bitmap(ufdset, fdset, nr);
1175 }
1176 
1177 
1178 /*
1179  * This is a virtual copy of sys_select from fs/select.c and probably
1180  * should be compared to it from time to time
1181  */
1182 
1183 /*
1184  * We can actually return ERESTARTSYS instead of EINTR, but I'd
1185  * like to be certain this leads to no problems. So I return
1186  * EINTR just for safety.
1187  *
1188  * Update: ERESTARTSYS breaks at least the xview clock binary, so
1189  * I'm trying ERESTARTNOHAND which restart only when you want to.
1190  */
1191 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1192 	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1193 	struct timespec64 *end_time)
1194 {
1195 	fd_set_bits fds;
1196 	void *bits;
1197 	int size, max_fds, ret = -EINVAL;
1198 	struct fdtable *fdt;
1199 	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1200 
1201 	if (n < 0)
1202 		goto out_nofds;
1203 
1204 	/* max_fds can increase, so grab it once to avoid race */
1205 	rcu_read_lock();
1206 	fdt = files_fdtable(current->files);
1207 	max_fds = fdt->max_fds;
1208 	rcu_read_unlock();
1209 	if (n > max_fds)
1210 		n = max_fds;
1211 
1212 	/*
1213 	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1214 	 * since we used fdset we need to allocate memory in units of
1215 	 * long-words.
1216 	 */
1217 	size = FDS_BYTES(n);
1218 	bits = stack_fds;
1219 	if (size > sizeof(stack_fds) / 6) {
1220 		bits = kmalloc_array(6, size, GFP_KERNEL);
1221 		ret = -ENOMEM;
1222 		if (!bits)
1223 			goto out_nofds;
1224 	}
1225 	fds.in      = (unsigned long *)  bits;
1226 	fds.out     = (unsigned long *) (bits +   size);
1227 	fds.ex      = (unsigned long *) (bits + 2*size);
1228 	fds.res_in  = (unsigned long *) (bits + 3*size);
1229 	fds.res_out = (unsigned long *) (bits + 4*size);
1230 	fds.res_ex  = (unsigned long *) (bits + 5*size);
1231 
1232 	if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1233 	    (ret = compat_get_fd_set(n, outp, fds.out)) ||
1234 	    (ret = compat_get_fd_set(n, exp, fds.ex)))
1235 		goto out;
1236 	zero_fd_set(n, fds.res_in);
1237 	zero_fd_set(n, fds.res_out);
1238 	zero_fd_set(n, fds.res_ex);
1239 
1240 	ret = do_select(n, &fds, end_time);
1241 
1242 	if (ret < 0)
1243 		goto out;
1244 	if (!ret) {
1245 		ret = -ERESTARTNOHAND;
1246 		if (signal_pending(current))
1247 			goto out;
1248 		ret = 0;
1249 	}
1250 
1251 	if (compat_set_fd_set(n, inp, fds.res_in) ||
1252 	    compat_set_fd_set(n, outp, fds.res_out) ||
1253 	    compat_set_fd_set(n, exp, fds.res_ex))
1254 		ret = -EFAULT;
1255 out:
1256 	if (bits != stack_fds)
1257 		kfree(bits);
1258 out_nofds:
1259 	return ret;
1260 }
1261 
1262 static int do_compat_select(int n, compat_ulong_t __user *inp,
1263 	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1264 	struct old_timeval32 __user *tvp)
1265 {
1266 	struct timespec64 end_time, *to = NULL;
1267 	struct old_timeval32 tv;
1268 	int ret;
1269 
1270 	if (tvp) {
1271 		if (copy_from_user(&tv, tvp, sizeof(tv)))
1272 			return -EFAULT;
1273 
1274 		to = &end_time;
1275 		if (poll_select_set_timeout(to,
1276 				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1277 				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1278 			return -EINVAL;
1279 	}
1280 
1281 	ret = compat_core_sys_select(n, inp, outp, exp, to);
1282 	return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1283 }
1284 
1285 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1286 	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1287 	struct old_timeval32 __user *, tvp)
1288 {
1289 	return do_compat_select(n, inp, outp, exp, tvp);
1290 }
1291 
1292 struct compat_sel_arg_struct {
1293 	compat_ulong_t n;
1294 	compat_uptr_t inp;
1295 	compat_uptr_t outp;
1296 	compat_uptr_t exp;
1297 	compat_uptr_t tvp;
1298 };
1299 
1300 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1301 {
1302 	struct compat_sel_arg_struct a;
1303 
1304 	if (copy_from_user(&a, arg, sizeof(a)))
1305 		return -EFAULT;
1306 	return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1307 				compat_ptr(a.exp), compat_ptr(a.tvp));
1308 }
1309 
1310 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1311 	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1312 	void __user *tsp, compat_sigset_t __user *sigmask,
1313 	compat_size_t sigsetsize, enum poll_time_type type)
1314 {
1315 	struct timespec64 ts, end_time, *to = NULL;
1316 	int ret;
1317 
1318 	if (tsp) {
1319 		switch (type) {
1320 		case PT_OLD_TIMESPEC:
1321 			if (get_old_timespec32(&ts, tsp))
1322 				return -EFAULT;
1323 			break;
1324 		case PT_TIMESPEC:
1325 			if (get_timespec64(&ts, tsp))
1326 				return -EFAULT;
1327 			break;
1328 		default:
1329 			BUG();
1330 		}
1331 
1332 		to = &end_time;
1333 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1334 			return -EINVAL;
1335 	}
1336 
1337 	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1338 	if (ret)
1339 		return ret;
1340 
1341 	ret = compat_core_sys_select(n, inp, outp, exp, to);
1342 	return poll_select_finish(&end_time, tsp, type, ret);
1343 }
1344 
1345 struct compat_sigset_argpack {
1346 	compat_uptr_t p;
1347 	compat_size_t size;
1348 };
1349 static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1350 					    struct compat_sigset_argpack __user *from)
1351 {
1352 	if (from) {
1353 		if (!user_read_access_begin(from, sizeof(*from)))
1354 			return -EFAULT;
1355 		unsafe_get_user(to->p, &from->p, Efault);
1356 		unsafe_get_user(to->size, &from->size, Efault);
1357 		user_read_access_end();
1358 	}
1359 	return 0;
1360 Efault:
1361 	user_access_end();
1362 	return -EFAULT;
1363 }
1364 
1365 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1366 	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1367 	struct __kernel_timespec __user *, tsp, void __user *, sig)
1368 {
1369 	struct compat_sigset_argpack x = {0, 0};
1370 
1371 	if (get_compat_sigset_argpack(&x, sig))
1372 		return -EFAULT;
1373 
1374 	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1375 				 x.size, PT_TIMESPEC);
1376 }
1377 
1378 #if defined(CONFIG_COMPAT_32BIT_TIME)
1379 
1380 COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1381 	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1382 	struct old_timespec32 __user *, tsp, void __user *, sig)
1383 {
1384 	struct compat_sigset_argpack x = {0, 0};
1385 
1386 	if (get_compat_sigset_argpack(&x, sig))
1387 		return -EFAULT;
1388 
1389 	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1390 				 x.size, PT_OLD_TIMESPEC);
1391 }
1392 
1393 #endif
1394 
1395 #if defined(CONFIG_COMPAT_32BIT_TIME)
1396 COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1397 	unsigned int,  nfds, struct old_timespec32 __user *, tsp,
1398 	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1399 {
1400 	struct timespec64 ts, end_time, *to = NULL;
1401 	int ret;
1402 
1403 	if (tsp) {
1404 		if (get_old_timespec32(&ts, tsp))
1405 			return -EFAULT;
1406 
1407 		to = &end_time;
1408 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1409 			return -EINVAL;
1410 	}
1411 
1412 	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1413 	if (ret)
1414 		return ret;
1415 
1416 	ret = do_sys_poll(ufds, nfds, to);
1417 	return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1418 }
1419 #endif
1420 
1421 /* New compat syscall for 64 bit time_t*/
1422 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1423 	unsigned int,  nfds, struct __kernel_timespec __user *, tsp,
1424 	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1425 {
1426 	struct timespec64 ts, end_time, *to = NULL;
1427 	int ret;
1428 
1429 	if (tsp) {
1430 		if (get_timespec64(&ts, tsp))
1431 			return -EFAULT;
1432 
1433 		to = &end_time;
1434 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1435 			return -EINVAL;
1436 	}
1437 
1438 	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1439 	if (ret)
1440 		return ret;
1441 
1442 	ret = do_sys_poll(ufds, nfds, to);
1443 	return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1444 }
1445 
1446 #endif
1447