xref: /linux/fs/select.c (revision 293d5b43948309434568f4dcbb36cce4c3c51bd5)
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.h>
19 #include <linux/syscalls.h>
20 #include <linux/export.h>
21 #include <linux/slab.h>
22 #include <linux/poll.h>
23 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/fs.h>
27 #include <linux/rcupdate.h>
28 #include <linux/hrtimer.h>
29 #include <linux/sched/rt.h>
30 #include <linux/freezer.h>
31 #include <net/busy_poll.h>
32 
33 #include <asm/uaccess.h>
34 
35 
36 /*
37  * Estimate expected accuracy in ns from a timeval.
38  *
39  * After quite a bit of churning around, we've settled on
40  * a simple thing of taking 0.1% of the timeout as the
41  * slack, with a cap of 100 msec.
42  * "nice" tasks get a 0.5% slack instead.
43  *
44  * Consider this comment an open invitation to come up with even
45  * better solutions..
46  */
47 
48 #define MAX_SLACK	(100 * NSEC_PER_MSEC)
49 
50 static long __estimate_accuracy(struct timespec64 *tv)
51 {
52 	long slack;
53 	int divfactor = 1000;
54 
55 	if (tv->tv_sec < 0)
56 		return 0;
57 
58 	if (task_nice(current) > 0)
59 		divfactor = divfactor / 5;
60 
61 	if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
62 		return MAX_SLACK;
63 
64 	slack = tv->tv_nsec / divfactor;
65 	slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
66 
67 	if (slack > MAX_SLACK)
68 		return MAX_SLACK;
69 
70 	return slack;
71 }
72 
73 u64 select_estimate_accuracy(struct timespec64 *tv)
74 {
75 	u64 ret;
76 	struct timespec64 now;
77 
78 	/*
79 	 * Realtime tasks get a slack of 0 for obvious reasons.
80 	 */
81 
82 	if (rt_task(current))
83 		return 0;
84 
85 	ktime_get_ts64(&now);
86 	now = timespec64_sub(*tv, now);
87 	ret = __estimate_accuracy(&now);
88 	if (ret < current->timer_slack_ns)
89 		return current->timer_slack_ns;
90 	return ret;
91 }
92 
93 
94 
95 struct poll_table_page {
96 	struct poll_table_page * next;
97 	struct poll_table_entry * entry;
98 	struct poll_table_entry entries[0];
99 };
100 
101 #define POLL_TABLE_FULL(table) \
102 	((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
103 
104 /*
105  * Ok, Peter made a complicated, but straightforward multiple_wait() function.
106  * I have rewritten this, taking some shortcuts: This code may not be easy to
107  * follow, but it should be free of race-conditions, and it's practical. If you
108  * understand what I'm doing here, then you understand how the linux
109  * sleep/wakeup mechanism works.
110  *
111  * Two very simple procedures, poll_wait() and poll_freewait() make all the
112  * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
113  * as all select/poll functions have to call it to add an entry to the
114  * poll table.
115  */
116 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
117 		       poll_table *p);
118 
119 void poll_initwait(struct poll_wqueues *pwq)
120 {
121 	init_poll_funcptr(&pwq->pt, __pollwait);
122 	pwq->polling_task = current;
123 	pwq->triggered = 0;
124 	pwq->error = 0;
125 	pwq->table = NULL;
126 	pwq->inline_index = 0;
127 }
128 EXPORT_SYMBOL(poll_initwait);
129 
130 static void free_poll_entry(struct poll_table_entry *entry)
131 {
132 	remove_wait_queue(entry->wait_address, &entry->wait);
133 	fput(entry->filp);
134 }
135 
136 void poll_freewait(struct poll_wqueues *pwq)
137 {
138 	struct poll_table_page * p = pwq->table;
139 	int i;
140 	for (i = 0; i < pwq->inline_index; i++)
141 		free_poll_entry(pwq->inline_entries + i);
142 	while (p) {
143 		struct poll_table_entry * entry;
144 		struct poll_table_page *old;
145 
146 		entry = p->entry;
147 		do {
148 			entry--;
149 			free_poll_entry(entry);
150 		} while (entry > p->entries);
151 		old = p;
152 		p = p->next;
153 		free_page((unsigned long) old);
154 	}
155 }
156 EXPORT_SYMBOL(poll_freewait);
157 
158 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
159 {
160 	struct poll_table_page *table = p->table;
161 
162 	if (p->inline_index < N_INLINE_POLL_ENTRIES)
163 		return p->inline_entries + p->inline_index++;
164 
165 	if (!table || POLL_TABLE_FULL(table)) {
166 		struct poll_table_page *new_table;
167 
168 		new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
169 		if (!new_table) {
170 			p->error = -ENOMEM;
171 			return NULL;
172 		}
173 		new_table->entry = new_table->entries;
174 		new_table->next = table;
175 		p->table = new_table;
176 		table = new_table;
177 	}
178 
179 	return table->entry++;
180 }
181 
182 static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
183 {
184 	struct poll_wqueues *pwq = wait->private;
185 	DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
186 
187 	/*
188 	 * Although this function is called under waitqueue lock, LOCK
189 	 * doesn't imply write barrier and the users expect write
190 	 * barrier semantics on wakeup functions.  The following
191 	 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
192 	 * and is paired with smp_store_mb() in poll_schedule_timeout.
193 	 */
194 	smp_wmb();
195 	pwq->triggered = 1;
196 
197 	/*
198 	 * Perform the default wake up operation using a dummy
199 	 * waitqueue.
200 	 *
201 	 * TODO: This is hacky but there currently is no interface to
202 	 * pass in @sync.  @sync is scheduled to be removed and once
203 	 * that happens, wake_up_process() can be used directly.
204 	 */
205 	return default_wake_function(&dummy_wait, mode, sync, key);
206 }
207 
208 static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
209 {
210 	struct poll_table_entry *entry;
211 
212 	entry = container_of(wait, struct poll_table_entry, wait);
213 	if (key && !((unsigned long)key & entry->key))
214 		return 0;
215 	return __pollwake(wait, mode, sync, key);
216 }
217 
218 /* Add a new entry */
219 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
220 				poll_table *p)
221 {
222 	struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
223 	struct poll_table_entry *entry = poll_get_entry(pwq);
224 	if (!entry)
225 		return;
226 	entry->filp = get_file(filp);
227 	entry->wait_address = wait_address;
228 	entry->key = p->_key;
229 	init_waitqueue_func_entry(&entry->wait, pollwake);
230 	entry->wait.private = pwq;
231 	add_wait_queue(wait_address, &entry->wait);
232 }
233 
234 int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
235 			  ktime_t *expires, unsigned long slack)
236 {
237 	int rc = -EINTR;
238 
239 	set_current_state(state);
240 	if (!pwq->triggered)
241 		rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
242 	__set_current_state(TASK_RUNNING);
243 
244 	/*
245 	 * Prepare for the next iteration.
246 	 *
247 	 * The following smp_store_mb() serves two purposes.  First, it's
248 	 * the counterpart rmb of the wmb in pollwake() such that data
249 	 * written before wake up is always visible after wake up.
250 	 * Second, the full barrier guarantees that triggered clearing
251 	 * doesn't pass event check of the next iteration.  Note that
252 	 * this problem doesn't exist for the first iteration as
253 	 * add_wait_queue() has full barrier semantics.
254 	 */
255 	smp_store_mb(pwq->triggered, 0);
256 
257 	return rc;
258 }
259 EXPORT_SYMBOL(poll_schedule_timeout);
260 
261 /**
262  * poll_select_set_timeout - helper function to setup the timeout value
263  * @to:		pointer to timespec64 variable for the final timeout
264  * @sec:	seconds (from user space)
265  * @nsec:	nanoseconds (from user space)
266  *
267  * Note, we do not use a timespec for the user space value here, That
268  * way we can use the function for timeval and compat interfaces as well.
269  *
270  * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
271  */
272 int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
273 {
274 	struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
275 
276 	if (!timespec64_valid(&ts))
277 		return -EINVAL;
278 
279 	/* Optimize for the zero timeout value here */
280 	if (!sec && !nsec) {
281 		to->tv_sec = to->tv_nsec = 0;
282 	} else {
283 		ktime_get_ts64(to);
284 		*to = timespec64_add_safe(*to, ts);
285 	}
286 	return 0;
287 }
288 
289 static int poll_select_copy_remaining(struct timespec64 *end_time,
290 				      void __user *p,
291 				      int timeval, int ret)
292 {
293 	struct timespec64 rts64;
294 	struct timespec rts;
295 	struct timeval rtv;
296 
297 	if (!p)
298 		return ret;
299 
300 	if (current->personality & STICKY_TIMEOUTS)
301 		goto sticky;
302 
303 	/* No update for zero timeout */
304 	if (!end_time->tv_sec && !end_time->tv_nsec)
305 		return ret;
306 
307 	ktime_get_ts64(&rts64);
308 	rts64 = timespec64_sub(*end_time, rts64);
309 	if (rts64.tv_sec < 0)
310 		rts64.tv_sec = rts64.tv_nsec = 0;
311 
312 	rts = timespec64_to_timespec(rts64);
313 
314 	if (timeval) {
315 		if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
316 			memset(&rtv, 0, sizeof(rtv));
317 		rtv.tv_sec = rts64.tv_sec;
318 		rtv.tv_usec = rts64.tv_nsec / NSEC_PER_USEC;
319 
320 		if (!copy_to_user(p, &rtv, sizeof(rtv)))
321 			return ret;
322 
323 	} else if (!copy_to_user(p, &rts, sizeof(rts)))
324 		return ret;
325 
326 	/*
327 	 * If an application puts its timeval in read-only memory, we
328 	 * don't want the Linux-specific update to the timeval to
329 	 * cause a fault after the select has completed
330 	 * successfully. However, because we're not updating the
331 	 * timeval, we can't restart the system call.
332 	 */
333 
334 sticky:
335 	if (ret == -ERESTARTNOHAND)
336 		ret = -EINTR;
337 	return ret;
338 }
339 
340 #define FDS_IN(fds, n)		(fds->in + n)
341 #define FDS_OUT(fds, n)		(fds->out + n)
342 #define FDS_EX(fds, n)		(fds->ex + n)
343 
344 #define BITS(fds, n)	(*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
345 
346 static int max_select_fd(unsigned long n, fd_set_bits *fds)
347 {
348 	unsigned long *open_fds;
349 	unsigned long set;
350 	int max;
351 	struct fdtable *fdt;
352 
353 	/* handle last in-complete long-word first */
354 	set = ~(~0UL << (n & (BITS_PER_LONG-1)));
355 	n /= BITS_PER_LONG;
356 	fdt = files_fdtable(current->files);
357 	open_fds = fdt->open_fds + n;
358 	max = 0;
359 	if (set) {
360 		set &= BITS(fds, n);
361 		if (set) {
362 			if (!(set & ~*open_fds))
363 				goto get_max;
364 			return -EBADF;
365 		}
366 	}
367 	while (n) {
368 		open_fds--;
369 		n--;
370 		set = BITS(fds, n);
371 		if (!set)
372 			continue;
373 		if (set & ~*open_fds)
374 			return -EBADF;
375 		if (max)
376 			continue;
377 get_max:
378 		do {
379 			max++;
380 			set >>= 1;
381 		} while (set);
382 		max += n * BITS_PER_LONG;
383 	}
384 
385 	return max;
386 }
387 
388 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
389 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
390 #define POLLEX_SET (POLLPRI)
391 
392 static inline void wait_key_set(poll_table *wait, unsigned long in,
393 				unsigned long out, unsigned long bit,
394 				unsigned int ll_flag)
395 {
396 	wait->_key = POLLEX_SET | ll_flag;
397 	if (in & bit)
398 		wait->_key |= POLLIN_SET;
399 	if (out & bit)
400 		wait->_key |= POLLOUT_SET;
401 }
402 
403 int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
404 {
405 	ktime_t expire, *to = NULL;
406 	struct poll_wqueues table;
407 	poll_table *wait;
408 	int retval, i, timed_out = 0;
409 	u64 slack = 0;
410 	unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
411 	unsigned long busy_end = 0;
412 
413 	rcu_read_lock();
414 	retval = max_select_fd(n, fds);
415 	rcu_read_unlock();
416 
417 	if (retval < 0)
418 		return retval;
419 	n = retval;
420 
421 	poll_initwait(&table);
422 	wait = &table.pt;
423 	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
424 		wait->_qproc = NULL;
425 		timed_out = 1;
426 	}
427 
428 	if (end_time && !timed_out)
429 		slack = select_estimate_accuracy(end_time);
430 
431 	retval = 0;
432 	for (;;) {
433 		unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
434 		bool can_busy_loop = false;
435 
436 		inp = fds->in; outp = fds->out; exp = fds->ex;
437 		rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
438 
439 		for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
440 			unsigned long in, out, ex, all_bits, bit = 1, mask, j;
441 			unsigned long res_in = 0, res_out = 0, res_ex = 0;
442 
443 			in = *inp++; out = *outp++; ex = *exp++;
444 			all_bits = in | out | ex;
445 			if (all_bits == 0) {
446 				i += BITS_PER_LONG;
447 				continue;
448 			}
449 
450 			for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
451 				struct fd f;
452 				if (i >= n)
453 					break;
454 				if (!(bit & all_bits))
455 					continue;
456 				f = fdget(i);
457 				if (f.file) {
458 					const struct file_operations *f_op;
459 					f_op = f.file->f_op;
460 					mask = DEFAULT_POLLMASK;
461 					if (f_op->poll) {
462 						wait_key_set(wait, in, out,
463 							     bit, busy_flag);
464 						mask = (*f_op->poll)(f.file, wait);
465 					}
466 					fdput(f);
467 					if ((mask & POLLIN_SET) && (in & bit)) {
468 						res_in |= bit;
469 						retval++;
470 						wait->_qproc = NULL;
471 					}
472 					if ((mask & POLLOUT_SET) && (out & bit)) {
473 						res_out |= bit;
474 						retval++;
475 						wait->_qproc = NULL;
476 					}
477 					if ((mask & POLLEX_SET) && (ex & bit)) {
478 						res_ex |= bit;
479 						retval++;
480 						wait->_qproc = NULL;
481 					}
482 					/* got something, stop busy polling */
483 					if (retval) {
484 						can_busy_loop = false;
485 						busy_flag = 0;
486 
487 					/*
488 					 * only remember a returned
489 					 * POLL_BUSY_LOOP if we asked for it
490 					 */
491 					} else if (busy_flag & mask)
492 						can_busy_loop = true;
493 
494 				}
495 			}
496 			if (res_in)
497 				*rinp = res_in;
498 			if (res_out)
499 				*routp = res_out;
500 			if (res_ex)
501 				*rexp = res_ex;
502 			cond_resched();
503 		}
504 		wait->_qproc = NULL;
505 		if (retval || timed_out || signal_pending(current))
506 			break;
507 		if (table.error) {
508 			retval = table.error;
509 			break;
510 		}
511 
512 		/* only if found POLL_BUSY_LOOP sockets && not out of time */
513 		if (can_busy_loop && !need_resched()) {
514 			if (!busy_end) {
515 				busy_end = busy_loop_end_time();
516 				continue;
517 			}
518 			if (!busy_loop_timeout(busy_end))
519 				continue;
520 		}
521 		busy_flag = 0;
522 
523 		/*
524 		 * If this is the first loop and we have a timeout
525 		 * given, then we convert to ktime_t and set the to
526 		 * pointer to the expiry value.
527 		 */
528 		if (end_time && !to) {
529 			expire = timespec64_to_ktime(*end_time);
530 			to = &expire;
531 		}
532 
533 		if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
534 					   to, slack))
535 			timed_out = 1;
536 	}
537 
538 	poll_freewait(&table);
539 
540 	return retval;
541 }
542 
543 /*
544  * We can actually return ERESTARTSYS instead of EINTR, but I'd
545  * like to be certain this leads to no problems. So I return
546  * EINTR just for safety.
547  *
548  * Update: ERESTARTSYS breaks at least the xview clock binary, so
549  * I'm trying ERESTARTNOHAND which restart only when you want to.
550  */
551 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
552 			   fd_set __user *exp, struct timespec64 *end_time)
553 {
554 	fd_set_bits fds;
555 	void *bits;
556 	int ret, max_fds;
557 	unsigned int size;
558 	struct fdtable *fdt;
559 	/* Allocate small arguments on the stack to save memory and be faster */
560 	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
561 
562 	ret = -EINVAL;
563 	if (n < 0)
564 		goto out_nofds;
565 
566 	/* max_fds can increase, so grab it once to avoid race */
567 	rcu_read_lock();
568 	fdt = files_fdtable(current->files);
569 	max_fds = fdt->max_fds;
570 	rcu_read_unlock();
571 	if (n > max_fds)
572 		n = max_fds;
573 
574 	/*
575 	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
576 	 * since we used fdset we need to allocate memory in units of
577 	 * long-words.
578 	 */
579 	size = FDS_BYTES(n);
580 	bits = stack_fds;
581 	if (size > sizeof(stack_fds) / 6) {
582 		/* Not enough space in on-stack array; must use kmalloc */
583 		ret = -ENOMEM;
584 		bits = kmalloc(6 * size, GFP_KERNEL);
585 		if (!bits)
586 			goto out_nofds;
587 	}
588 	fds.in      = bits;
589 	fds.out     = bits +   size;
590 	fds.ex      = bits + 2*size;
591 	fds.res_in  = bits + 3*size;
592 	fds.res_out = bits + 4*size;
593 	fds.res_ex  = bits + 5*size;
594 
595 	if ((ret = get_fd_set(n, inp, fds.in)) ||
596 	    (ret = get_fd_set(n, outp, fds.out)) ||
597 	    (ret = get_fd_set(n, exp, fds.ex)))
598 		goto out;
599 	zero_fd_set(n, fds.res_in);
600 	zero_fd_set(n, fds.res_out);
601 	zero_fd_set(n, fds.res_ex);
602 
603 	ret = do_select(n, &fds, end_time);
604 
605 	if (ret < 0)
606 		goto out;
607 	if (!ret) {
608 		ret = -ERESTARTNOHAND;
609 		if (signal_pending(current))
610 			goto out;
611 		ret = 0;
612 	}
613 
614 	if (set_fd_set(n, inp, fds.res_in) ||
615 	    set_fd_set(n, outp, fds.res_out) ||
616 	    set_fd_set(n, exp, fds.res_ex))
617 		ret = -EFAULT;
618 
619 out:
620 	if (bits != stack_fds)
621 		kfree(bits);
622 out_nofds:
623 	return ret;
624 }
625 
626 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
627 		fd_set __user *, exp, struct timeval __user *, tvp)
628 {
629 	struct timespec64 end_time, *to = NULL;
630 	struct timeval tv;
631 	int ret;
632 
633 	if (tvp) {
634 		if (copy_from_user(&tv, tvp, sizeof(tv)))
635 			return -EFAULT;
636 
637 		to = &end_time;
638 		if (poll_select_set_timeout(to,
639 				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
640 				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
641 			return -EINVAL;
642 	}
643 
644 	ret = core_sys_select(n, inp, outp, exp, to);
645 	ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
646 
647 	return ret;
648 }
649 
650 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
651 		       fd_set __user *exp, struct timespec __user *tsp,
652 		       const sigset_t __user *sigmask, size_t sigsetsize)
653 {
654 	sigset_t ksigmask, sigsaved;
655 	struct timespec ts;
656 	struct timespec64 ts64, end_time, *to = NULL;
657 	int ret;
658 
659 	if (tsp) {
660 		if (copy_from_user(&ts, tsp, sizeof(ts)))
661 			return -EFAULT;
662 		ts64 = timespec_to_timespec64(ts);
663 
664 		to = &end_time;
665 		if (poll_select_set_timeout(to, ts64.tv_sec, ts64.tv_nsec))
666 			return -EINVAL;
667 	}
668 
669 	if (sigmask) {
670 		/* XXX: Don't preclude handling different sized sigset_t's.  */
671 		if (sigsetsize != sizeof(sigset_t))
672 			return -EINVAL;
673 		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
674 			return -EFAULT;
675 
676 		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
677 		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
678 	}
679 
680 	ret = core_sys_select(n, inp, outp, exp, to);
681 	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
682 
683 	if (ret == -ERESTARTNOHAND) {
684 		/*
685 		 * Don't restore the signal mask yet. Let do_signal() deliver
686 		 * the signal on the way back to userspace, before the signal
687 		 * mask is restored.
688 		 */
689 		if (sigmask) {
690 			memcpy(&current->saved_sigmask, &sigsaved,
691 					sizeof(sigsaved));
692 			set_restore_sigmask();
693 		}
694 	} else if (sigmask)
695 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
696 
697 	return ret;
698 }
699 
700 /*
701  * Most architectures can't handle 7-argument syscalls. So we provide a
702  * 6-argument version where the sixth argument is a pointer to a structure
703  * which has a pointer to the sigset_t itself followed by a size_t containing
704  * the sigset size.
705  */
706 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
707 		fd_set __user *, exp, struct timespec __user *, tsp,
708 		void __user *, sig)
709 {
710 	size_t sigsetsize = 0;
711 	sigset_t __user *up = NULL;
712 
713 	if (sig) {
714 		if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
715 		    || __get_user(up, (sigset_t __user * __user *)sig)
716 		    || __get_user(sigsetsize,
717 				(size_t __user *)(sig+sizeof(void *))))
718 			return -EFAULT;
719 	}
720 
721 	return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
722 }
723 
724 #ifdef __ARCH_WANT_SYS_OLD_SELECT
725 struct sel_arg_struct {
726 	unsigned long n;
727 	fd_set __user *inp, *outp, *exp;
728 	struct timeval __user *tvp;
729 };
730 
731 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
732 {
733 	struct sel_arg_struct a;
734 
735 	if (copy_from_user(&a, arg, sizeof(a)))
736 		return -EFAULT;
737 	return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
738 }
739 #endif
740 
741 struct poll_list {
742 	struct poll_list *next;
743 	int len;
744 	struct pollfd entries[0];
745 };
746 
747 #define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
748 
749 /*
750  * Fish for pollable events on the pollfd->fd file descriptor. We're only
751  * interested in events matching the pollfd->events mask, and the result
752  * matching that mask is both recorded in pollfd->revents and returned. The
753  * pwait poll_table will be used by the fd-provided poll handler for waiting,
754  * if pwait->_qproc is non-NULL.
755  */
756 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait,
757 				     bool *can_busy_poll,
758 				     unsigned int busy_flag)
759 {
760 	unsigned int mask;
761 	int fd;
762 
763 	mask = 0;
764 	fd = pollfd->fd;
765 	if (fd >= 0) {
766 		struct fd f = fdget(fd);
767 		mask = POLLNVAL;
768 		if (f.file) {
769 			mask = DEFAULT_POLLMASK;
770 			if (f.file->f_op->poll) {
771 				pwait->_key = pollfd->events|POLLERR|POLLHUP;
772 				pwait->_key |= busy_flag;
773 				mask = f.file->f_op->poll(f.file, pwait);
774 				if (mask & busy_flag)
775 					*can_busy_poll = true;
776 			}
777 			/* Mask out unneeded events. */
778 			mask &= pollfd->events | POLLERR | POLLHUP;
779 			fdput(f);
780 		}
781 	}
782 	pollfd->revents = mask;
783 
784 	return mask;
785 }
786 
787 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
788 		   struct timespec64 *end_time)
789 {
790 	poll_table* pt = &wait->pt;
791 	ktime_t expire, *to = NULL;
792 	int timed_out = 0, count = 0;
793 	u64 slack = 0;
794 	unsigned int busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
795 	unsigned long busy_end = 0;
796 
797 	/* Optimise the no-wait case */
798 	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
799 		pt->_qproc = NULL;
800 		timed_out = 1;
801 	}
802 
803 	if (end_time && !timed_out)
804 		slack = select_estimate_accuracy(end_time);
805 
806 	for (;;) {
807 		struct poll_list *walk;
808 		bool can_busy_loop = false;
809 
810 		for (walk = list; walk != NULL; walk = walk->next) {
811 			struct pollfd * pfd, * pfd_end;
812 
813 			pfd = walk->entries;
814 			pfd_end = pfd + walk->len;
815 			for (; pfd != pfd_end; pfd++) {
816 				/*
817 				 * Fish for events. If we found one, record it
818 				 * and kill poll_table->_qproc, so we don't
819 				 * needlessly register any other waiters after
820 				 * this. They'll get immediately deregistered
821 				 * when we break out and return.
822 				 */
823 				if (do_pollfd(pfd, pt, &can_busy_loop,
824 					      busy_flag)) {
825 					count++;
826 					pt->_qproc = NULL;
827 					/* found something, stop busy polling */
828 					busy_flag = 0;
829 					can_busy_loop = false;
830 				}
831 			}
832 		}
833 		/*
834 		 * All waiters have already been registered, so don't provide
835 		 * a poll_table->_qproc to them on the next loop iteration.
836 		 */
837 		pt->_qproc = NULL;
838 		if (!count) {
839 			count = wait->error;
840 			if (signal_pending(current))
841 				count = -EINTR;
842 		}
843 		if (count || timed_out)
844 			break;
845 
846 		/* only if found POLL_BUSY_LOOP sockets && not out of time */
847 		if (can_busy_loop && !need_resched()) {
848 			if (!busy_end) {
849 				busy_end = busy_loop_end_time();
850 				continue;
851 			}
852 			if (!busy_loop_timeout(busy_end))
853 				continue;
854 		}
855 		busy_flag = 0;
856 
857 		/*
858 		 * If this is the first loop and we have a timeout
859 		 * given, then we convert to ktime_t and set the to
860 		 * pointer to the expiry value.
861 		 */
862 		if (end_time && !to) {
863 			expire = timespec64_to_ktime(*end_time);
864 			to = &expire;
865 		}
866 
867 		if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
868 			timed_out = 1;
869 	}
870 	return count;
871 }
872 
873 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
874 			sizeof(struct pollfd))
875 
876 int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
877 		struct timespec64 *end_time)
878 {
879 	struct poll_wqueues table;
880  	int err = -EFAULT, fdcount, len, size;
881 	/* Allocate small arguments on the stack to save memory and be
882 	   faster - use long to make sure the buffer is aligned properly
883 	   on 64 bit archs to avoid unaligned access */
884 	long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
885 	struct poll_list *const head = (struct poll_list *)stack_pps;
886  	struct poll_list *walk = head;
887  	unsigned long todo = nfds;
888 
889 	if (nfds > rlimit(RLIMIT_NOFILE))
890 		return -EINVAL;
891 
892 	len = min_t(unsigned int, nfds, N_STACK_PPS);
893 	for (;;) {
894 		walk->next = NULL;
895 		walk->len = len;
896 		if (!len)
897 			break;
898 
899 		if (copy_from_user(walk->entries, ufds + nfds-todo,
900 					sizeof(struct pollfd) * walk->len))
901 			goto out_fds;
902 
903 		todo -= walk->len;
904 		if (!todo)
905 			break;
906 
907 		len = min(todo, POLLFD_PER_PAGE);
908 		size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
909 		walk = walk->next = kmalloc(size, GFP_KERNEL);
910 		if (!walk) {
911 			err = -ENOMEM;
912 			goto out_fds;
913 		}
914 	}
915 
916 	poll_initwait(&table);
917 	fdcount = do_poll(head, &table, end_time);
918 	poll_freewait(&table);
919 
920 	for (walk = head; walk; walk = walk->next) {
921 		struct pollfd *fds = walk->entries;
922 		int j;
923 
924 		for (j = 0; j < walk->len; j++, ufds++)
925 			if (__put_user(fds[j].revents, &ufds->revents))
926 				goto out_fds;
927   	}
928 
929 	err = fdcount;
930 out_fds:
931 	walk = head->next;
932 	while (walk) {
933 		struct poll_list *pos = walk;
934 		walk = walk->next;
935 		kfree(pos);
936 	}
937 
938 	return err;
939 }
940 
941 static long do_restart_poll(struct restart_block *restart_block)
942 {
943 	struct pollfd __user *ufds = restart_block->poll.ufds;
944 	int nfds = restart_block->poll.nfds;
945 	struct timespec64 *to = NULL, end_time;
946 	int ret;
947 
948 	if (restart_block->poll.has_timeout) {
949 		end_time.tv_sec = restart_block->poll.tv_sec;
950 		end_time.tv_nsec = restart_block->poll.tv_nsec;
951 		to = &end_time;
952 	}
953 
954 	ret = do_sys_poll(ufds, nfds, to);
955 
956 	if (ret == -EINTR) {
957 		restart_block->fn = do_restart_poll;
958 		ret = -ERESTART_RESTARTBLOCK;
959 	}
960 	return ret;
961 }
962 
963 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
964 		int, timeout_msecs)
965 {
966 	struct timespec64 end_time, *to = NULL;
967 	int ret;
968 
969 	if (timeout_msecs >= 0) {
970 		to = &end_time;
971 		poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
972 			NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
973 	}
974 
975 	ret = do_sys_poll(ufds, nfds, to);
976 
977 	if (ret == -EINTR) {
978 		struct restart_block *restart_block;
979 
980 		restart_block = &current->restart_block;
981 		restart_block->fn = do_restart_poll;
982 		restart_block->poll.ufds = ufds;
983 		restart_block->poll.nfds = nfds;
984 
985 		if (timeout_msecs >= 0) {
986 			restart_block->poll.tv_sec = end_time.tv_sec;
987 			restart_block->poll.tv_nsec = end_time.tv_nsec;
988 			restart_block->poll.has_timeout = 1;
989 		} else
990 			restart_block->poll.has_timeout = 0;
991 
992 		ret = -ERESTART_RESTARTBLOCK;
993 	}
994 	return ret;
995 }
996 
997 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
998 		struct timespec __user *, tsp, const sigset_t __user *, sigmask,
999 		size_t, sigsetsize)
1000 {
1001 	sigset_t ksigmask, sigsaved;
1002 	struct timespec ts;
1003 	struct timespec64 end_time, *to = NULL;
1004 	int ret;
1005 
1006 	if (tsp) {
1007 		if (copy_from_user(&ts, tsp, sizeof(ts)))
1008 			return -EFAULT;
1009 
1010 		to = &end_time;
1011 		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1012 			return -EINVAL;
1013 	}
1014 
1015 	if (sigmask) {
1016 		/* XXX: Don't preclude handling different sized sigset_t's.  */
1017 		if (sigsetsize != sizeof(sigset_t))
1018 			return -EINVAL;
1019 		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1020 			return -EFAULT;
1021 
1022 		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1023 		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1024 	}
1025 
1026 	ret = do_sys_poll(ufds, nfds, to);
1027 
1028 	/* We can restart this syscall, usually */
1029 	if (ret == -EINTR) {
1030 		/*
1031 		 * Don't restore the signal mask yet. Let do_signal() deliver
1032 		 * the signal on the way back to userspace, before the signal
1033 		 * mask is restored.
1034 		 */
1035 		if (sigmask) {
1036 			memcpy(&current->saved_sigmask, &sigsaved,
1037 					sizeof(sigsaved));
1038 			set_restore_sigmask();
1039 		}
1040 		ret = -ERESTARTNOHAND;
1041 	} else if (sigmask)
1042 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1043 
1044 	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1045 
1046 	return ret;
1047 }
1048