xref: /openbmc/linux/fs/select.c (revision 1c2dd16a)
1 /*
2  * This file contains the procedures for the handling of select and poll
3  *
4  * Created for Linux based loosely upon Mathius Lattner's minix
5  * patches by Peter MacDonald. Heavily edited by Linus.
6  *
7  *  4 February 1994
8  *     COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
9  *     flag set in its personality we do *not* modify the given timeout
10  *     parameter to reflect time remaining.
11  *
12  *  24 January 2000
13  *     Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
14  *     of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
15  */
16 
17 #include <linux/kernel.h>
18 #include <linux/sched/signal.h>
19 #include <linux/sched/rt.h>
20 #include <linux/syscalls.h>
21 #include <linux/export.h>
22 #include <linux/slab.h>
23 #include <linux/poll.h>
24 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
25 #include <linux/file.h>
26 #include <linux/fdtable.h>
27 #include <linux/fs.h>
28 #include <linux/rcupdate.h>
29 #include <linux/hrtimer.h>
30 #include <linux/freezer.h>
31 #include <net/busy_poll.h>
32 #include <linux/vmalloc.h>
33 
34 #include <linux/uaccess.h>
35 
36 
37 /*
38  * Estimate expected accuracy in ns from a timeval.
39  *
40  * After quite a bit of churning around, we've settled on
41  * a simple thing of taking 0.1% of the timeout as the
42  * slack, with a cap of 100 msec.
43  * "nice" tasks get a 0.5% slack instead.
44  *
45  * Consider this comment an open invitation to come up with even
46  * better solutions..
47  */
48 
49 #define MAX_SLACK	(100 * NSEC_PER_MSEC)
50 
51 static long __estimate_accuracy(struct timespec64 *tv)
52 {
53 	long slack;
54 	int divfactor = 1000;
55 
56 	if (tv->tv_sec < 0)
57 		return 0;
58 
59 	if (task_nice(current) > 0)
60 		divfactor = divfactor / 5;
61 
62 	if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
63 		return MAX_SLACK;
64 
65 	slack = tv->tv_nsec / divfactor;
66 	slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
67 
68 	if (slack > MAX_SLACK)
69 		return MAX_SLACK;
70 
71 	return slack;
72 }
73 
74 u64 select_estimate_accuracy(struct timespec64 *tv)
75 {
76 	u64 ret;
77 	struct timespec64 now;
78 
79 	/*
80 	 * Realtime tasks get a slack of 0 for obvious reasons.
81 	 */
82 
83 	if (rt_task(current))
84 		return 0;
85 
86 	ktime_get_ts64(&now);
87 	now = timespec64_sub(*tv, now);
88 	ret = __estimate_accuracy(&now);
89 	if (ret < current->timer_slack_ns)
90 		return current->timer_slack_ns;
91 	return ret;
92 }
93 
94 
95 
96 struct poll_table_page {
97 	struct poll_table_page * next;
98 	struct poll_table_entry * entry;
99 	struct poll_table_entry entries[0];
100 };
101 
102 #define POLL_TABLE_FULL(table) \
103 	((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
104 
105 /*
106  * Ok, Peter made a complicated, but straightforward multiple_wait() function.
107  * I have rewritten this, taking some shortcuts: This code may not be easy to
108  * follow, but it should be free of race-conditions, and it's practical. If you
109  * understand what I'm doing here, then you understand how the linux
110  * sleep/wakeup mechanism works.
111  *
112  * Two very simple procedures, poll_wait() and poll_freewait() make all the
113  * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
114  * as all select/poll functions have to call it to add an entry to the
115  * poll table.
116  */
117 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
118 		       poll_table *p);
119 
120 void poll_initwait(struct poll_wqueues *pwq)
121 {
122 	init_poll_funcptr(&pwq->pt, __pollwait);
123 	pwq->polling_task = current;
124 	pwq->triggered = 0;
125 	pwq->error = 0;
126 	pwq->table = NULL;
127 	pwq->inline_index = 0;
128 }
129 EXPORT_SYMBOL(poll_initwait);
130 
131 static void free_poll_entry(struct poll_table_entry *entry)
132 {
133 	remove_wait_queue(entry->wait_address, &entry->wait);
134 	fput(entry->filp);
135 }
136 
137 void poll_freewait(struct poll_wqueues *pwq)
138 {
139 	struct poll_table_page * p = pwq->table;
140 	int i;
141 	for (i = 0; i < pwq->inline_index; i++)
142 		free_poll_entry(pwq->inline_entries + i);
143 	while (p) {
144 		struct poll_table_entry * entry;
145 		struct poll_table_page *old;
146 
147 		entry = p->entry;
148 		do {
149 			entry--;
150 			free_poll_entry(entry);
151 		} while (entry > p->entries);
152 		old = p;
153 		p = p->next;
154 		free_page((unsigned long) old);
155 	}
156 }
157 EXPORT_SYMBOL(poll_freewait);
158 
159 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
160 {
161 	struct poll_table_page *table = p->table;
162 
163 	if (p->inline_index < N_INLINE_POLL_ENTRIES)
164 		return p->inline_entries + p->inline_index++;
165 
166 	if (!table || POLL_TABLE_FULL(table)) {
167 		struct poll_table_page *new_table;
168 
169 		new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
170 		if (!new_table) {
171 			p->error = -ENOMEM;
172 			return NULL;
173 		}
174 		new_table->entry = new_table->entries;
175 		new_table->next = table;
176 		p->table = new_table;
177 		table = new_table;
178 	}
179 
180 	return table->entry++;
181 }
182 
183 static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
184 {
185 	struct poll_wqueues *pwq = wait->private;
186 	DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
187 
188 	/*
189 	 * Although this function is called under waitqueue lock, LOCK
190 	 * doesn't imply write barrier and the users expect write
191 	 * barrier semantics on wakeup functions.  The following
192 	 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
193 	 * and is paired with smp_store_mb() in poll_schedule_timeout.
194 	 */
195 	smp_wmb();
196 	pwq->triggered = 1;
197 
198 	/*
199 	 * Perform the default wake up operation using a dummy
200 	 * waitqueue.
201 	 *
202 	 * TODO: This is hacky but there currently is no interface to
203 	 * pass in @sync.  @sync is scheduled to be removed and once
204 	 * that happens, wake_up_process() can be used directly.
205 	 */
206 	return default_wake_function(&dummy_wait, mode, sync, key);
207 }
208 
209 static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
210 {
211 	struct poll_table_entry *entry;
212 
213 	entry = container_of(wait, struct poll_table_entry, wait);
214 	if (key && !((unsigned long)key & entry->key))
215 		return 0;
216 	return __pollwake(wait, mode, sync, key);
217 }
218 
219 /* Add a new entry */
220 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
221 				poll_table *p)
222 {
223 	struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
224 	struct poll_table_entry *entry = poll_get_entry(pwq);
225 	if (!entry)
226 		return;
227 	entry->filp = get_file(filp);
228 	entry->wait_address = wait_address;
229 	entry->key = p->_key;
230 	init_waitqueue_func_entry(&entry->wait, pollwake);
231 	entry->wait.private = pwq;
232 	add_wait_queue(wait_address, &entry->wait);
233 }
234 
235 int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
236 			  ktime_t *expires, unsigned long slack)
237 {
238 	int rc = -EINTR;
239 
240 	set_current_state(state);
241 	if (!pwq->triggered)
242 		rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
243 	__set_current_state(TASK_RUNNING);
244 
245 	/*
246 	 * Prepare for the next iteration.
247 	 *
248 	 * The following smp_store_mb() serves two purposes.  First, it's
249 	 * the counterpart rmb of the wmb in pollwake() such that data
250 	 * written before wake up is always visible after wake up.
251 	 * Second, the full barrier guarantees that triggered clearing
252 	 * doesn't pass event check of the next iteration.  Note that
253 	 * this problem doesn't exist for the first iteration as
254 	 * add_wait_queue() has full barrier semantics.
255 	 */
256 	smp_store_mb(pwq->triggered, 0);
257 
258 	return rc;
259 }
260 EXPORT_SYMBOL(poll_schedule_timeout);
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 int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
274 {
275 	struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
276 
277 	if (!timespec64_valid(&ts))
278 		return -EINVAL;
279 
280 	/* Optimize for the zero timeout value here */
281 	if (!sec && !nsec) {
282 		to->tv_sec = to->tv_nsec = 0;
283 	} else {
284 		ktime_get_ts64(to);
285 		*to = timespec64_add_safe(*to, ts);
286 	}
287 	return 0;
288 }
289 
290 static int poll_select_copy_remaining(struct timespec64 *end_time,
291 				      void __user *p,
292 				      int timeval, int ret)
293 {
294 	struct timespec64 rts64;
295 	struct timespec rts;
296 	struct timeval rtv;
297 
298 	if (!p)
299 		return ret;
300 
301 	if (current->personality & STICKY_TIMEOUTS)
302 		goto sticky;
303 
304 	/* No update for zero timeout */
305 	if (!end_time->tv_sec && !end_time->tv_nsec)
306 		return ret;
307 
308 	ktime_get_ts64(&rts64);
309 	rts64 = timespec64_sub(*end_time, rts64);
310 	if (rts64.tv_sec < 0)
311 		rts64.tv_sec = rts64.tv_nsec = 0;
312 
313 	rts = timespec64_to_timespec(rts64);
314 
315 	if (timeval) {
316 		if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
317 			memset(&rtv, 0, sizeof(rtv));
318 		rtv.tv_sec = rts64.tv_sec;
319 		rtv.tv_usec = rts64.tv_nsec / NSEC_PER_USEC;
320 
321 		if (!copy_to_user(p, &rtv, sizeof(rtv)))
322 			return ret;
323 
324 	} else if (!copy_to_user(p, &rts, sizeof(rts)))
325 		return ret;
326 
327 	/*
328 	 * If an application puts its timeval in read-only memory, we
329 	 * don't want the Linux-specific update to the timeval to
330 	 * cause a fault after the select has completed
331 	 * successfully. However, because we're not updating the
332 	 * timeval, we can't restart the system call.
333 	 */
334 
335 sticky:
336 	if (ret == -ERESTARTNOHAND)
337 		ret = -EINTR;
338 	return ret;
339 }
340 
341 /*
342  * Scalable version of the fd_set.
343  */
344 
345 typedef struct {
346 	unsigned long *in, *out, *ex;
347 	unsigned long *res_in, *res_out, *res_ex;
348 } fd_set_bits;
349 
350 /*
351  * How many longwords for "nr" bits?
352  */
353 #define FDS_BITPERLONG	(8*sizeof(long))
354 #define FDS_LONGS(nr)	(((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
355 #define FDS_BYTES(nr)	(FDS_LONGS(nr)*sizeof(long))
356 
357 /*
358  * We do a VERIFY_WRITE here even though we are only reading this time:
359  * we'll write to it eventually..
360  *
361  * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
362  */
363 static inline
364 int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
365 {
366 	nr = FDS_BYTES(nr);
367 	if (ufdset)
368 		return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
369 
370 	memset(fdset, 0, nr);
371 	return 0;
372 }
373 
374 static inline unsigned long __must_check
375 set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
376 {
377 	if (ufdset)
378 		return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
379 	return 0;
380 }
381 
382 static inline
383 void zero_fd_set(unsigned long nr, unsigned long *fdset)
384 {
385 	memset(fdset, 0, FDS_BYTES(nr));
386 }
387 
388 #define FDS_IN(fds, n)		(fds->in + n)
389 #define FDS_OUT(fds, n)		(fds->out + n)
390 #define FDS_EX(fds, n)		(fds->ex + n)
391 
392 #define BITS(fds, n)	(*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
393 
394 static int max_select_fd(unsigned long n, fd_set_bits *fds)
395 {
396 	unsigned long *open_fds;
397 	unsigned long set;
398 	int max;
399 	struct fdtable *fdt;
400 
401 	/* handle last in-complete long-word first */
402 	set = ~(~0UL << (n & (BITS_PER_LONG-1)));
403 	n /= BITS_PER_LONG;
404 	fdt = files_fdtable(current->files);
405 	open_fds = fdt->open_fds + n;
406 	max = 0;
407 	if (set) {
408 		set &= BITS(fds, n);
409 		if (set) {
410 			if (!(set & ~*open_fds))
411 				goto get_max;
412 			return -EBADF;
413 		}
414 	}
415 	while (n) {
416 		open_fds--;
417 		n--;
418 		set = BITS(fds, n);
419 		if (!set)
420 			continue;
421 		if (set & ~*open_fds)
422 			return -EBADF;
423 		if (max)
424 			continue;
425 get_max:
426 		do {
427 			max++;
428 			set >>= 1;
429 		} while (set);
430 		max += n * BITS_PER_LONG;
431 	}
432 
433 	return max;
434 }
435 
436 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
437 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
438 #define POLLEX_SET (POLLPRI)
439 
440 static inline void wait_key_set(poll_table *wait, unsigned long in,
441 				unsigned long out, unsigned long bit,
442 				unsigned int ll_flag)
443 {
444 	wait->_key = POLLEX_SET | ll_flag;
445 	if (in & bit)
446 		wait->_key |= POLLIN_SET;
447 	if (out & bit)
448 		wait->_key |= POLLOUT_SET;
449 }
450 
451 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
452 {
453 	ktime_t expire, *to = NULL;
454 	struct poll_wqueues table;
455 	poll_table *wait;
456 	int retval, i, timed_out = 0;
457 	u64 slack = 0;
458 	unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
459 	unsigned long busy_start = 0;
460 
461 	rcu_read_lock();
462 	retval = max_select_fd(n, fds);
463 	rcu_read_unlock();
464 
465 	if (retval < 0)
466 		return retval;
467 	n = retval;
468 
469 	poll_initwait(&table);
470 	wait = &table.pt;
471 	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
472 		wait->_qproc = NULL;
473 		timed_out = 1;
474 	}
475 
476 	if (end_time && !timed_out)
477 		slack = select_estimate_accuracy(end_time);
478 
479 	retval = 0;
480 	for (;;) {
481 		unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
482 		bool can_busy_loop = false;
483 
484 		inp = fds->in; outp = fds->out; exp = fds->ex;
485 		rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
486 
487 		for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
488 			unsigned long in, out, ex, all_bits, bit = 1, mask, j;
489 			unsigned long res_in = 0, res_out = 0, res_ex = 0;
490 
491 			in = *inp++; out = *outp++; ex = *exp++;
492 			all_bits = in | out | ex;
493 			if (all_bits == 0) {
494 				i += BITS_PER_LONG;
495 				continue;
496 			}
497 
498 			for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
499 				struct fd f;
500 				if (i >= n)
501 					break;
502 				if (!(bit & all_bits))
503 					continue;
504 				f = fdget(i);
505 				if (f.file) {
506 					const struct file_operations *f_op;
507 					f_op = f.file->f_op;
508 					mask = DEFAULT_POLLMASK;
509 					if (f_op->poll) {
510 						wait_key_set(wait, in, out,
511 							     bit, busy_flag);
512 						mask = (*f_op->poll)(f.file, wait);
513 					}
514 					fdput(f);
515 					if ((mask & POLLIN_SET) && (in & bit)) {
516 						res_in |= bit;
517 						retval++;
518 						wait->_qproc = NULL;
519 					}
520 					if ((mask & POLLOUT_SET) && (out & bit)) {
521 						res_out |= bit;
522 						retval++;
523 						wait->_qproc = NULL;
524 					}
525 					if ((mask & POLLEX_SET) && (ex & bit)) {
526 						res_ex |= bit;
527 						retval++;
528 						wait->_qproc = NULL;
529 					}
530 					/* got something, stop busy polling */
531 					if (retval) {
532 						can_busy_loop = false;
533 						busy_flag = 0;
534 
535 					/*
536 					 * only remember a returned
537 					 * POLL_BUSY_LOOP if we asked for it
538 					 */
539 					} else if (busy_flag & mask)
540 						can_busy_loop = true;
541 
542 				}
543 			}
544 			if (res_in)
545 				*rinp = res_in;
546 			if (res_out)
547 				*routp = res_out;
548 			if (res_ex)
549 				*rexp = res_ex;
550 			cond_resched();
551 		}
552 		wait->_qproc = NULL;
553 		if (retval || timed_out || signal_pending(current))
554 			break;
555 		if (table.error) {
556 			retval = table.error;
557 			break;
558 		}
559 
560 		/* only if found POLL_BUSY_LOOP sockets && not out of time */
561 		if (can_busy_loop && !need_resched()) {
562 			if (!busy_start) {
563 				busy_start = busy_loop_current_time();
564 				continue;
565 			}
566 			if (!busy_loop_timeout(busy_start))
567 				continue;
568 		}
569 		busy_flag = 0;
570 
571 		/*
572 		 * If this is the first loop and we have a timeout
573 		 * given, then we convert to ktime_t and set the to
574 		 * pointer to the expiry value.
575 		 */
576 		if (end_time && !to) {
577 			expire = timespec64_to_ktime(*end_time);
578 			to = &expire;
579 		}
580 
581 		if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
582 					   to, slack))
583 			timed_out = 1;
584 	}
585 
586 	poll_freewait(&table);
587 
588 	return retval;
589 }
590 
591 /*
592  * We can actually return ERESTARTSYS instead of EINTR, but I'd
593  * like to be certain this leads to no problems. So I return
594  * EINTR just for safety.
595  *
596  * Update: ERESTARTSYS breaks at least the xview clock binary, so
597  * I'm trying ERESTARTNOHAND which restart only when you want to.
598  */
599 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
600 			   fd_set __user *exp, struct timespec64 *end_time)
601 {
602 	fd_set_bits fds;
603 	void *bits;
604 	int ret, max_fds;
605 	size_t size, alloc_size;
606 	struct fdtable *fdt;
607 	/* Allocate small arguments on the stack to save memory and be faster */
608 	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
609 
610 	ret = -EINVAL;
611 	if (n < 0)
612 		goto out_nofds;
613 
614 	/* max_fds can increase, so grab it once to avoid race */
615 	rcu_read_lock();
616 	fdt = files_fdtable(current->files);
617 	max_fds = fdt->max_fds;
618 	rcu_read_unlock();
619 	if (n > max_fds)
620 		n = max_fds;
621 
622 	/*
623 	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
624 	 * since we used fdset we need to allocate memory in units of
625 	 * long-words.
626 	 */
627 	size = FDS_BYTES(n);
628 	bits = stack_fds;
629 	if (size > sizeof(stack_fds) / 6) {
630 		/* Not enough space in on-stack array; must use kmalloc */
631 		ret = -ENOMEM;
632 		if (size > (SIZE_MAX / 6))
633 			goto out_nofds;
634 
635 		alloc_size = 6 * size;
636 		bits = kmalloc(alloc_size, GFP_KERNEL|__GFP_NOWARN);
637 		if (!bits && alloc_size > PAGE_SIZE)
638 			bits = vmalloc(alloc_size);
639 
640 		if (!bits)
641 			goto out_nofds;
642 	}
643 	fds.in      = bits;
644 	fds.out     = bits +   size;
645 	fds.ex      = bits + 2*size;
646 	fds.res_in  = bits + 3*size;
647 	fds.res_out = bits + 4*size;
648 	fds.res_ex  = bits + 5*size;
649 
650 	if ((ret = get_fd_set(n, inp, fds.in)) ||
651 	    (ret = get_fd_set(n, outp, fds.out)) ||
652 	    (ret = get_fd_set(n, exp, fds.ex)))
653 		goto out;
654 	zero_fd_set(n, fds.res_in);
655 	zero_fd_set(n, fds.res_out);
656 	zero_fd_set(n, fds.res_ex);
657 
658 	ret = do_select(n, &fds, end_time);
659 
660 	if (ret < 0)
661 		goto out;
662 	if (!ret) {
663 		ret = -ERESTARTNOHAND;
664 		if (signal_pending(current))
665 			goto out;
666 		ret = 0;
667 	}
668 
669 	if (set_fd_set(n, inp, fds.res_in) ||
670 	    set_fd_set(n, outp, fds.res_out) ||
671 	    set_fd_set(n, exp, fds.res_ex))
672 		ret = -EFAULT;
673 
674 out:
675 	if (bits != stack_fds)
676 		kvfree(bits);
677 out_nofds:
678 	return ret;
679 }
680 
681 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
682 		fd_set __user *, exp, struct timeval __user *, tvp)
683 {
684 	struct timespec64 end_time, *to = NULL;
685 	struct timeval tv;
686 	int ret;
687 
688 	if (tvp) {
689 		if (copy_from_user(&tv, tvp, sizeof(tv)))
690 			return -EFAULT;
691 
692 		to = &end_time;
693 		if (poll_select_set_timeout(to,
694 				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
695 				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
696 			return -EINVAL;
697 	}
698 
699 	ret = core_sys_select(n, inp, outp, exp, to);
700 	ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
701 
702 	return ret;
703 }
704 
705 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
706 		       fd_set __user *exp, struct timespec __user *tsp,
707 		       const sigset_t __user *sigmask, size_t sigsetsize)
708 {
709 	sigset_t ksigmask, sigsaved;
710 	struct timespec ts;
711 	struct timespec64 ts64, end_time, *to = NULL;
712 	int ret;
713 
714 	if (tsp) {
715 		if (copy_from_user(&ts, tsp, sizeof(ts)))
716 			return -EFAULT;
717 		ts64 = timespec_to_timespec64(ts);
718 
719 		to = &end_time;
720 		if (poll_select_set_timeout(to, ts64.tv_sec, ts64.tv_nsec))
721 			return -EINVAL;
722 	}
723 
724 	if (sigmask) {
725 		/* XXX: Don't preclude handling different sized sigset_t's.  */
726 		if (sigsetsize != sizeof(sigset_t))
727 			return -EINVAL;
728 		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
729 			return -EFAULT;
730 
731 		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
732 		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
733 	}
734 
735 	ret = core_sys_select(n, inp, outp, exp, to);
736 	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
737 
738 	if (ret == -ERESTARTNOHAND) {
739 		/*
740 		 * Don't restore the signal mask yet. Let do_signal() deliver
741 		 * the signal on the way back to userspace, before the signal
742 		 * mask is restored.
743 		 */
744 		if (sigmask) {
745 			memcpy(&current->saved_sigmask, &sigsaved,
746 					sizeof(sigsaved));
747 			set_restore_sigmask();
748 		}
749 	} else if (sigmask)
750 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
751 
752 	return ret;
753 }
754 
755 /*
756  * Most architectures can't handle 7-argument syscalls. So we provide a
757  * 6-argument version where the sixth argument is a pointer to a structure
758  * which has a pointer to the sigset_t itself followed by a size_t containing
759  * the sigset size.
760  */
761 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
762 		fd_set __user *, exp, struct timespec __user *, tsp,
763 		void __user *, sig)
764 {
765 	size_t sigsetsize = 0;
766 	sigset_t __user *up = NULL;
767 
768 	if (sig) {
769 		if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
770 		    || __get_user(up, (sigset_t __user * __user *)sig)
771 		    || __get_user(sigsetsize,
772 				(size_t __user *)(sig+sizeof(void *))))
773 			return -EFAULT;
774 	}
775 
776 	return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
777 }
778 
779 #ifdef __ARCH_WANT_SYS_OLD_SELECT
780 struct sel_arg_struct {
781 	unsigned long n;
782 	fd_set __user *inp, *outp, *exp;
783 	struct timeval __user *tvp;
784 };
785 
786 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
787 {
788 	struct sel_arg_struct a;
789 
790 	if (copy_from_user(&a, arg, sizeof(a)))
791 		return -EFAULT;
792 	return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
793 }
794 #endif
795 
796 struct poll_list {
797 	struct poll_list *next;
798 	int len;
799 	struct pollfd entries[0];
800 };
801 
802 #define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
803 
804 /*
805  * Fish for pollable events on the pollfd->fd file descriptor. We're only
806  * interested in events matching the pollfd->events mask, and the result
807  * matching that mask is both recorded in pollfd->revents and returned. The
808  * pwait poll_table will be used by the fd-provided poll handler for waiting,
809  * if pwait->_qproc is non-NULL.
810  */
811 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait,
812 				     bool *can_busy_poll,
813 				     unsigned int busy_flag)
814 {
815 	unsigned int mask;
816 	int fd;
817 
818 	mask = 0;
819 	fd = pollfd->fd;
820 	if (fd >= 0) {
821 		struct fd f = fdget(fd);
822 		mask = POLLNVAL;
823 		if (f.file) {
824 			mask = DEFAULT_POLLMASK;
825 			if (f.file->f_op->poll) {
826 				pwait->_key = pollfd->events|POLLERR|POLLHUP;
827 				pwait->_key |= busy_flag;
828 				mask = f.file->f_op->poll(f.file, pwait);
829 				if (mask & busy_flag)
830 					*can_busy_poll = true;
831 			}
832 			/* Mask out unneeded events. */
833 			mask &= pollfd->events | POLLERR | POLLHUP;
834 			fdput(f);
835 		}
836 	}
837 	pollfd->revents = mask;
838 
839 	return mask;
840 }
841 
842 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
843 		   struct timespec64 *end_time)
844 {
845 	poll_table* pt = &wait->pt;
846 	ktime_t expire, *to = NULL;
847 	int timed_out = 0, count = 0;
848 	u64 slack = 0;
849 	unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
850 	unsigned long busy_start = 0;
851 
852 	/* Optimise the no-wait case */
853 	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
854 		pt->_qproc = NULL;
855 		timed_out = 1;
856 	}
857 
858 	if (end_time && !timed_out)
859 		slack = select_estimate_accuracy(end_time);
860 
861 	for (;;) {
862 		struct poll_list *walk;
863 		bool can_busy_loop = false;
864 
865 		for (walk = list; walk != NULL; walk = walk->next) {
866 			struct pollfd * pfd, * pfd_end;
867 
868 			pfd = walk->entries;
869 			pfd_end = pfd + walk->len;
870 			for (; pfd != pfd_end; pfd++) {
871 				/*
872 				 * Fish for events. If we found one, record it
873 				 * and kill poll_table->_qproc, so we don't
874 				 * needlessly register any other waiters after
875 				 * this. They'll get immediately deregistered
876 				 * when we break out and return.
877 				 */
878 				if (do_pollfd(pfd, pt, &can_busy_loop,
879 					      busy_flag)) {
880 					count++;
881 					pt->_qproc = NULL;
882 					/* found something, stop busy polling */
883 					busy_flag = 0;
884 					can_busy_loop = false;
885 				}
886 			}
887 		}
888 		/*
889 		 * All waiters have already been registered, so don't provide
890 		 * a poll_table->_qproc to them on the next loop iteration.
891 		 */
892 		pt->_qproc = NULL;
893 		if (!count) {
894 			count = wait->error;
895 			if (signal_pending(current))
896 				count = -EINTR;
897 		}
898 		if (count || timed_out)
899 			break;
900 
901 		/* only if found POLL_BUSY_LOOP sockets && not out of time */
902 		if (can_busy_loop && !need_resched()) {
903 			if (!busy_start) {
904 				busy_start = busy_loop_current_time();
905 				continue;
906 			}
907 			if (!busy_loop_timeout(busy_start))
908 				continue;
909 		}
910 		busy_flag = 0;
911 
912 		/*
913 		 * If this is the first loop and we have a timeout
914 		 * given, then we convert to ktime_t and set the to
915 		 * pointer to the expiry value.
916 		 */
917 		if (end_time && !to) {
918 			expire = timespec64_to_ktime(*end_time);
919 			to = &expire;
920 		}
921 
922 		if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
923 			timed_out = 1;
924 	}
925 	return count;
926 }
927 
928 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
929 			sizeof(struct pollfd))
930 
931 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
932 		struct timespec64 *end_time)
933 {
934 	struct poll_wqueues table;
935  	int err = -EFAULT, fdcount, len, size;
936 	/* Allocate small arguments on the stack to save memory and be
937 	   faster - use long to make sure the buffer is aligned properly
938 	   on 64 bit archs to avoid unaligned access */
939 	long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
940 	struct poll_list *const head = (struct poll_list *)stack_pps;
941  	struct poll_list *walk = head;
942  	unsigned long todo = nfds;
943 
944 	if (nfds > rlimit(RLIMIT_NOFILE))
945 		return -EINVAL;
946 
947 	len = min_t(unsigned int, nfds, N_STACK_PPS);
948 	for (;;) {
949 		walk->next = NULL;
950 		walk->len = len;
951 		if (!len)
952 			break;
953 
954 		if (copy_from_user(walk->entries, ufds + nfds-todo,
955 					sizeof(struct pollfd) * walk->len))
956 			goto out_fds;
957 
958 		todo -= walk->len;
959 		if (!todo)
960 			break;
961 
962 		len = min(todo, POLLFD_PER_PAGE);
963 		size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
964 		walk = walk->next = kmalloc(size, GFP_KERNEL);
965 		if (!walk) {
966 			err = -ENOMEM;
967 			goto out_fds;
968 		}
969 	}
970 
971 	poll_initwait(&table);
972 	fdcount = do_poll(head, &table, end_time);
973 	poll_freewait(&table);
974 
975 	for (walk = head; walk; walk = walk->next) {
976 		struct pollfd *fds = walk->entries;
977 		int j;
978 
979 		for (j = 0; j < walk->len; j++, ufds++)
980 			if (__put_user(fds[j].revents, &ufds->revents))
981 				goto out_fds;
982   	}
983 
984 	err = fdcount;
985 out_fds:
986 	walk = head->next;
987 	while (walk) {
988 		struct poll_list *pos = walk;
989 		walk = walk->next;
990 		kfree(pos);
991 	}
992 
993 	return err;
994 }
995 
996 static long do_restart_poll(struct restart_block *restart_block)
997 {
998 	struct pollfd __user *ufds = restart_block->poll.ufds;
999 	int nfds = restart_block->poll.nfds;
1000 	struct timespec64 *to = NULL, end_time;
1001 	int ret;
1002 
1003 	if (restart_block->poll.has_timeout) {
1004 		end_time.tv_sec = restart_block->poll.tv_sec;
1005 		end_time.tv_nsec = restart_block->poll.tv_nsec;
1006 		to = &end_time;
1007 	}
1008 
1009 	ret = do_sys_poll(ufds, nfds, to);
1010 
1011 	if (ret == -EINTR) {
1012 		restart_block->fn = do_restart_poll;
1013 		ret = -ERESTART_RESTARTBLOCK;
1014 	}
1015 	return ret;
1016 }
1017 
1018 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1019 		int, timeout_msecs)
1020 {
1021 	struct timespec64 end_time, *to = NULL;
1022 	int ret;
1023 
1024 	if (timeout_msecs >= 0) {
1025 		to = &end_time;
1026 		poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1027 			NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1028 	}
1029 
1030 	ret = do_sys_poll(ufds, nfds, to);
1031 
1032 	if (ret == -EINTR) {
1033 		struct restart_block *restart_block;
1034 
1035 		restart_block = &current->restart_block;
1036 		restart_block->fn = do_restart_poll;
1037 		restart_block->poll.ufds = ufds;
1038 		restart_block->poll.nfds = nfds;
1039 
1040 		if (timeout_msecs >= 0) {
1041 			restart_block->poll.tv_sec = end_time.tv_sec;
1042 			restart_block->poll.tv_nsec = end_time.tv_nsec;
1043 			restart_block->poll.has_timeout = 1;
1044 		} else
1045 			restart_block->poll.has_timeout = 0;
1046 
1047 		ret = -ERESTART_RESTARTBLOCK;
1048 	}
1049 	return ret;
1050 }
1051 
1052 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1053 		struct timespec __user *, tsp, const sigset_t __user *, sigmask,
1054 		size_t, sigsetsize)
1055 {
1056 	sigset_t ksigmask, sigsaved;
1057 	struct timespec ts;
1058 	struct timespec64 end_time, *to = NULL;
1059 	int ret;
1060 
1061 	if (tsp) {
1062 		if (copy_from_user(&ts, tsp, sizeof(ts)))
1063 			return -EFAULT;
1064 
1065 		to = &end_time;
1066 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1067 			return -EINVAL;
1068 	}
1069 
1070 	if (sigmask) {
1071 		/* XXX: Don't preclude handling different sized sigset_t's.  */
1072 		if (sigsetsize != sizeof(sigset_t))
1073 			return -EINVAL;
1074 		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1075 			return -EFAULT;
1076 
1077 		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1078 		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1079 	}
1080 
1081 	ret = do_sys_poll(ufds, nfds, to);
1082 
1083 	/* We can restart this syscall, usually */
1084 	if (ret == -EINTR) {
1085 		/*
1086 		 * Don't restore the signal mask yet. Let do_signal() deliver
1087 		 * the signal on the way back to userspace, before the signal
1088 		 * mask is restored.
1089 		 */
1090 		if (sigmask) {
1091 			memcpy(&current->saved_sigmask, &sigsaved,
1092 					sizeof(sigsaved));
1093 			set_restore_sigmask();
1094 		}
1095 		ret = -ERESTARTNOHAND;
1096 	} else if (sigmask)
1097 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1098 
1099 	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1100 
1101 	return ret;
1102 }
1103 
1104 #ifdef CONFIG_COMPAT
1105 #define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1106 
1107 static
1108 int compat_poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1109 				      int timeval, int ret)
1110 {
1111 	struct timespec ts;
1112 
1113 	if (!p)
1114 		return ret;
1115 
1116 	if (current->personality & STICKY_TIMEOUTS)
1117 		goto sticky;
1118 
1119 	/* No update for zero timeout */
1120 	if (!end_time->tv_sec && !end_time->tv_nsec)
1121 		return ret;
1122 
1123 	ktime_get_ts(&ts);
1124 	ts = timespec_sub(*end_time, ts);
1125 	if (ts.tv_sec < 0)
1126 		ts.tv_sec = ts.tv_nsec = 0;
1127 
1128 	if (timeval) {
1129 		struct compat_timeval rtv;
1130 
1131 		rtv.tv_sec = ts.tv_sec;
1132 		rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1133 
1134 		if (!copy_to_user(p, &rtv, sizeof(rtv)))
1135 			return ret;
1136 	} else {
1137 		struct compat_timespec rts;
1138 
1139 		rts.tv_sec = ts.tv_sec;
1140 		rts.tv_nsec = ts.tv_nsec;
1141 
1142 		if (!copy_to_user(p, &rts, sizeof(rts)))
1143 			return ret;
1144 	}
1145 	/*
1146 	 * If an application puts its timeval in read-only memory, we
1147 	 * don't want the Linux-specific update to the timeval to
1148 	 * cause a fault after the select has completed
1149 	 * successfully. However, because we're not updating the
1150 	 * timeval, we can't restart the system call.
1151 	 */
1152 
1153 sticky:
1154 	if (ret == -ERESTARTNOHAND)
1155 		ret = -EINTR;
1156 	return ret;
1157 }
1158 
1159 /*
1160  * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1161  * 64-bit unsigned longs.
1162  */
1163 static
1164 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1165 			unsigned long *fdset)
1166 {
1167 	nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1168 	if (ufdset) {
1169 		unsigned long odd;
1170 
1171 		if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1172 			return -EFAULT;
1173 
1174 		odd = nr & 1UL;
1175 		nr &= ~1UL;
1176 		while (nr) {
1177 			unsigned long h, l;
1178 			if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1179 				return -EFAULT;
1180 			ufdset += 2;
1181 			*fdset++ = h << 32 | l;
1182 			nr -= 2;
1183 		}
1184 		if (odd && __get_user(*fdset, ufdset))
1185 			return -EFAULT;
1186 	} else {
1187 		/* Tricky, must clear full unsigned long in the
1188 		 * kernel fdset at the end, this makes sure that
1189 		 * actually happens.
1190 		 */
1191 		memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1192 	}
1193 	return 0;
1194 }
1195 
1196 static
1197 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1198 		      unsigned long *fdset)
1199 {
1200 	unsigned long odd;
1201 	nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1202 
1203 	if (!ufdset)
1204 		return 0;
1205 
1206 	odd = nr & 1UL;
1207 	nr &= ~1UL;
1208 	while (nr) {
1209 		unsigned long h, l;
1210 		l = *fdset++;
1211 		h = l >> 32;
1212 		if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1213 			return -EFAULT;
1214 		ufdset += 2;
1215 		nr -= 2;
1216 	}
1217 	if (odd && __put_user(*fdset, ufdset))
1218 		return -EFAULT;
1219 	return 0;
1220 }
1221 
1222 
1223 /*
1224  * This is a virtual copy of sys_select from fs/select.c and probably
1225  * should be compared to it from time to time
1226  */
1227 
1228 /*
1229  * We can actually return ERESTARTSYS instead of EINTR, but I'd
1230  * like to be certain this leads to no problems. So I return
1231  * EINTR just for safety.
1232  *
1233  * Update: ERESTARTSYS breaks at least the xview clock binary, so
1234  * I'm trying ERESTARTNOHAND which restart only when you want to.
1235  */
1236 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1237 	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1238 	struct timespec *end_time)
1239 {
1240 	fd_set_bits fds;
1241 	void *bits;
1242 	int size, max_fds, ret = -EINVAL;
1243 	struct fdtable *fdt;
1244 	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1245 
1246 	if (n < 0)
1247 		goto out_nofds;
1248 
1249 	/* max_fds can increase, so grab it once to avoid race */
1250 	rcu_read_lock();
1251 	fdt = files_fdtable(current->files);
1252 	max_fds = fdt->max_fds;
1253 	rcu_read_unlock();
1254 	if (n > max_fds)
1255 		n = max_fds;
1256 
1257 	/*
1258 	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1259 	 * since we used fdset we need to allocate memory in units of
1260 	 * long-words.
1261 	 */
1262 	size = FDS_BYTES(n);
1263 	bits = stack_fds;
1264 	if (size > sizeof(stack_fds) / 6) {
1265 		bits = kmalloc(6 * size, GFP_KERNEL);
1266 		ret = -ENOMEM;
1267 		if (!bits)
1268 			goto out_nofds;
1269 	}
1270 	fds.in      = (unsigned long *)  bits;
1271 	fds.out     = (unsigned long *) (bits +   size);
1272 	fds.ex      = (unsigned long *) (bits + 2*size);
1273 	fds.res_in  = (unsigned long *) (bits + 3*size);
1274 	fds.res_out = (unsigned long *) (bits + 4*size);
1275 	fds.res_ex  = (unsigned long *) (bits + 5*size);
1276 
1277 	if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1278 	    (ret = compat_get_fd_set(n, outp, fds.out)) ||
1279 	    (ret = compat_get_fd_set(n, exp, fds.ex)))
1280 		goto out;
1281 	zero_fd_set(n, fds.res_in);
1282 	zero_fd_set(n, fds.res_out);
1283 	zero_fd_set(n, fds.res_ex);
1284 
1285 	ret = do_select(n, &fds, end_time);
1286 
1287 	if (ret < 0)
1288 		goto out;
1289 	if (!ret) {
1290 		ret = -ERESTARTNOHAND;
1291 		if (signal_pending(current))
1292 			goto out;
1293 		ret = 0;
1294 	}
1295 
1296 	if (compat_set_fd_set(n, inp, fds.res_in) ||
1297 	    compat_set_fd_set(n, outp, fds.res_out) ||
1298 	    compat_set_fd_set(n, exp, fds.res_ex))
1299 		ret = -EFAULT;
1300 out:
1301 	if (bits != stack_fds)
1302 		kfree(bits);
1303 out_nofds:
1304 	return ret;
1305 }
1306 
1307 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1308 	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1309 	struct compat_timeval __user *, tvp)
1310 {
1311 	struct timespec end_time, *to = NULL;
1312 	struct compat_timeval tv;
1313 	int ret;
1314 
1315 	if (tvp) {
1316 		if (copy_from_user(&tv, tvp, sizeof(tv)))
1317 			return -EFAULT;
1318 
1319 		to = &end_time;
1320 		if (poll_select_set_timeout(to,
1321 				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1322 				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1323 			return -EINVAL;
1324 	}
1325 
1326 	ret = compat_core_sys_select(n, inp, outp, exp, to);
1327 	ret = compat_poll_select_copy_remaining(&end_time, tvp, 1, ret);
1328 
1329 	return ret;
1330 }
1331 
1332 struct compat_sel_arg_struct {
1333 	compat_ulong_t n;
1334 	compat_uptr_t inp;
1335 	compat_uptr_t outp;
1336 	compat_uptr_t exp;
1337 	compat_uptr_t tvp;
1338 };
1339 
1340 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1341 {
1342 	struct compat_sel_arg_struct a;
1343 
1344 	if (copy_from_user(&a, arg, sizeof(a)))
1345 		return -EFAULT;
1346 	return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1347 				 compat_ptr(a.exp), compat_ptr(a.tvp));
1348 }
1349 
1350 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1351 	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1352 	struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1353 	compat_size_t sigsetsize)
1354 {
1355 	compat_sigset_t ss32;
1356 	sigset_t ksigmask, sigsaved;
1357 	struct compat_timespec ts;
1358 	struct timespec end_time, *to = NULL;
1359 	int ret;
1360 
1361 	if (tsp) {
1362 		if (copy_from_user(&ts, tsp, sizeof(ts)))
1363 			return -EFAULT;
1364 
1365 		to = &end_time;
1366 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1367 			return -EINVAL;
1368 	}
1369 
1370 	if (sigmask) {
1371 		if (sigsetsize != sizeof(compat_sigset_t))
1372 			return -EINVAL;
1373 		if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1374 			return -EFAULT;
1375 		sigset_from_compat(&ksigmask, &ss32);
1376 
1377 		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1378 		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1379 	}
1380 
1381 	ret = compat_core_sys_select(n, inp, outp, exp, to);
1382 	ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret);
1383 
1384 	if (ret == -ERESTARTNOHAND) {
1385 		/*
1386 		 * Don't restore the signal mask yet. Let do_signal() deliver
1387 		 * the signal on the way back to userspace, before the signal
1388 		 * mask is restored.
1389 		 */
1390 		if (sigmask) {
1391 			memcpy(&current->saved_sigmask, &sigsaved,
1392 					sizeof(sigsaved));
1393 			set_restore_sigmask();
1394 		}
1395 	} else if (sigmask)
1396 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1397 
1398 	return ret;
1399 }
1400 
1401 COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp,
1402 	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1403 	struct compat_timespec __user *, tsp, void __user *, sig)
1404 {
1405 	compat_size_t sigsetsize = 0;
1406 	compat_uptr_t up = 0;
1407 
1408 	if (sig) {
1409 		if (!access_ok(VERIFY_READ, sig,
1410 				sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1411 		    	__get_user(up, (compat_uptr_t __user *)sig) ||
1412 		    	__get_user(sigsetsize,
1413 				(compat_size_t __user *)(sig+sizeof(up))))
1414 			return -EFAULT;
1415 	}
1416 	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1417 				 sigsetsize);
1418 }
1419 
1420 COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds,
1421 	unsigned int,  nfds, struct compat_timespec __user *, tsp,
1422 	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1423 {
1424 	compat_sigset_t ss32;
1425 	sigset_t ksigmask, sigsaved;
1426 	struct compat_timespec ts;
1427 	struct timespec end_time, *to = NULL;
1428 	int ret;
1429 
1430 	if (tsp) {
1431 		if (copy_from_user(&ts, tsp, sizeof(ts)))
1432 			return -EFAULT;
1433 
1434 		to = &end_time;
1435 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1436 			return -EINVAL;
1437 	}
1438 
1439 	if (sigmask) {
1440 		if (sigsetsize != sizeof(compat_sigset_t))
1441 			return -EINVAL;
1442 		if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1443 			return -EFAULT;
1444 		sigset_from_compat(&ksigmask, &ss32);
1445 
1446 		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1447 		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1448 	}
1449 
1450 	ret = do_sys_poll(ufds, nfds, to);
1451 
1452 	/* We can restart this syscall, usually */
1453 	if (ret == -EINTR) {
1454 		/*
1455 		 * Don't restore the signal mask yet. Let do_signal() deliver
1456 		 * the signal on the way back to userspace, before the signal
1457 		 * mask is restored.
1458 		 */
1459 		if (sigmask) {
1460 			memcpy(&current->saved_sigmask, &sigsaved,
1461 				sizeof(sigsaved));
1462 			set_restore_sigmask();
1463 		}
1464 		ret = -ERESTARTNOHAND;
1465 	} else if (sigmask)
1466 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1467 
1468 	ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret);
1469 
1470 	return ret;
1471 }
1472 #endif
1473