Linux Audio

Check our new training course

Loading...
v3.1
 
  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/module.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
 30#include <asm/uaccess.h>
 31
 32
 33/*
 34 * Estimate expected accuracy in ns from a timeval.
 35 *
 36 * After quite a bit of churning around, we've settled on
 37 * a simple thing of taking 0.1% of the timeout as the
 38 * slack, with a cap of 100 msec.
 39 * "nice" tasks get a 0.5% slack instead.
 40 *
 41 * Consider this comment an open invitation to come up with even
 42 * better solutions..
 43 */
 44
 45#define MAX_SLACK	(100 * NSEC_PER_MSEC)
 46
 47static long __estimate_accuracy(struct timespec *tv)
 48{
 49	long slack;
 50	int divfactor = 1000;
 51
 52	if (tv->tv_sec < 0)
 53		return 0;
 54
 55	if (task_nice(current) > 0)
 56		divfactor = divfactor / 5;
 57
 58	if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
 59		return MAX_SLACK;
 60
 61	slack = tv->tv_nsec / divfactor;
 62	slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
 63
 64	if (slack > MAX_SLACK)
 65		return MAX_SLACK;
 66
 67	return slack;
 68}
 69
 70long select_estimate_accuracy(struct timespec *tv)
 71{
 72	unsigned long ret;
 73	struct timespec now;
 74
 75	/*
 76	 * Realtime tasks get a slack of 0 for obvious reasons.
 77	 */
 78
 79	if (rt_task(current))
 80		return 0;
 81
 82	ktime_get_ts(&now);
 83	now = timespec_sub(*tv, now);
 84	ret = __estimate_accuracy(&now);
 85	if (ret < current->timer_slack_ns)
 86		return current->timer_slack_ns;
 87	return ret;
 88}
 89
 90
 91
 92struct poll_table_page {
 93	struct poll_table_page * next;
 94	struct poll_table_entry * entry;
 95	struct poll_table_entry entries[0];
 96};
 97
 98#define POLL_TABLE_FULL(table) \
 99	((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
100
101/*
102 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
103 * I have rewritten this, taking some shortcuts: This code may not be easy to
104 * follow, but it should be free of race-conditions, and it's practical. If you
105 * understand what I'm doing here, then you understand how the linux
106 * sleep/wakeup mechanism works.
107 *
108 * Two very simple procedures, poll_wait() and poll_freewait() make all the
109 * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
110 * as all select/poll functions have to call it to add an entry to the
111 * poll table.
112 */
113static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
114		       poll_table *p);
115
116void poll_initwait(struct poll_wqueues *pwq)
117{
118	init_poll_funcptr(&pwq->pt, __pollwait);
119	pwq->polling_task = current;
120	pwq->triggered = 0;
121	pwq->error = 0;
122	pwq->table = NULL;
123	pwq->inline_index = 0;
124}
125EXPORT_SYMBOL(poll_initwait);
126
127static void free_poll_entry(struct poll_table_entry *entry)
128{
129	remove_wait_queue(entry->wait_address, &entry->wait);
130	fput(entry->filp);
131}
132
133void poll_freewait(struct poll_wqueues *pwq)
134{
135	struct poll_table_page * p = pwq->table;
136	int i;
137	for (i = 0; i < pwq->inline_index; i++)
138		free_poll_entry(pwq->inline_entries + i);
139	while (p) {
140		struct poll_table_entry * entry;
141		struct poll_table_page *old;
142
143		entry = p->entry;
144		do {
145			entry--;
146			free_poll_entry(entry);
147		} while (entry > p->entries);
148		old = p;
149		p = p->next;
150		free_page((unsigned long) old);
151	}
152}
153EXPORT_SYMBOL(poll_freewait);
154
155static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
156{
157	struct poll_table_page *table = p->table;
158
159	if (p->inline_index < N_INLINE_POLL_ENTRIES)
160		return p->inline_entries + p->inline_index++;
161
162	if (!table || POLL_TABLE_FULL(table)) {
163		struct poll_table_page *new_table;
164
165		new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
166		if (!new_table) {
167			p->error = -ENOMEM;
168			return NULL;
169		}
170		new_table->entry = new_table->entries;
171		new_table->next = table;
172		p->table = new_table;
173		table = new_table;
174	}
175
176	return table->entry++;
177}
178
179static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
180{
181	struct poll_wqueues *pwq = wait->private;
182	DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
183
184	/*
185	 * Although this function is called under waitqueue lock, LOCK
186	 * doesn't imply write barrier and the users expect write
187	 * barrier semantics on wakeup functions.  The following
188	 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
189	 * and is paired with set_mb() in poll_schedule_timeout.
190	 */
191	smp_wmb();
192	pwq->triggered = 1;
193
194	/*
195	 * Perform the default wake up operation using a dummy
196	 * waitqueue.
197	 *
198	 * TODO: This is hacky but there currently is no interface to
199	 * pass in @sync.  @sync is scheduled to be removed and once
200	 * that happens, wake_up_process() can be used directly.
201	 */
202	return default_wake_function(&dummy_wait, mode, sync, key);
203}
204
205static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
206{
207	struct poll_table_entry *entry;
208
209	entry = container_of(wait, struct poll_table_entry, wait);
210	if (key && !((unsigned long)key & entry->key))
211		return 0;
212	return __pollwake(wait, mode, sync, key);
213}
214
215/* Add a new entry */
216static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
217				poll_table *p)
218{
219	struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
220	struct poll_table_entry *entry = poll_get_entry(pwq);
221	if (!entry)
222		return;
223	get_file(filp);
224	entry->filp = filp;
225	entry->wait_address = wait_address;
226	entry->key = p->key;
227	init_waitqueue_func_entry(&entry->wait, pollwake);
228	entry->wait.private = pwq;
229	add_wait_queue(wait_address, &entry->wait);
230}
231
232int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
233			  ktime_t *expires, unsigned long slack)
234{
235	int rc = -EINTR;
236
237	set_current_state(state);
238	if (!pwq->triggered)
239		rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
240	__set_current_state(TASK_RUNNING);
241
242	/*
243	 * Prepare for the next iteration.
244	 *
245	 * The following set_mb() serves two purposes.  First, it's
246	 * the counterpart rmb of the wmb in pollwake() such that data
247	 * written before wake up is always visible after wake up.
248	 * Second, the full barrier guarantees that triggered clearing
249	 * doesn't pass event check of the next iteration.  Note that
250	 * this problem doesn't exist for the first iteration as
251	 * add_wait_queue() has full barrier semantics.
252	 */
253	set_mb(pwq->triggered, 0);
254
255	return rc;
256}
257EXPORT_SYMBOL(poll_schedule_timeout);
258
259/**
260 * poll_select_set_timeout - helper function to setup the timeout value
261 * @to:		pointer to timespec variable for the final timeout
262 * @sec:	seconds (from user space)
263 * @nsec:	nanoseconds (from user space)
264 *
265 * Note, we do not use a timespec for the user space value here, That
266 * way we can use the function for timeval and compat interfaces as well.
267 *
268 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
269 */
270int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
271{
272	struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
273
274	if (!timespec_valid(&ts))
275		return -EINVAL;
276
277	/* Optimize for the zero timeout value here */
278	if (!sec && !nsec) {
279		to->tv_sec = to->tv_nsec = 0;
280	} else {
281		ktime_get_ts(to);
282		*to = timespec_add_safe(*to, ts);
283	}
284	return 0;
285}
286
287static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
 
288				      int timeval, int ret)
289{
290	struct timespec rts;
291	struct timeval rtv;
292
293	if (!p)
294		return ret;
295
296	if (current->personality & STICKY_TIMEOUTS)
297		goto sticky;
298
299	/* No update for zero timeout */
300	if (!end_time->tv_sec && !end_time->tv_nsec)
301		return ret;
302
303	ktime_get_ts(&rts);
304	rts = timespec_sub(*end_time, rts);
305	if (rts.tv_sec < 0)
306		rts.tv_sec = rts.tv_nsec = 0;
307
 
308	if (timeval) {
309		if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
310			memset(&rtv, 0, sizeof(rtv));
311		rtv.tv_sec = rts.tv_sec;
312		rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
313
314		if (!copy_to_user(p, &rtv, sizeof(rtv)))
315			return ret;
316
317	} else if (!copy_to_user(p, &rts, sizeof(rts)))
318		return ret;
319
320	/*
321	 * If an application puts its timeval in read-only memory, we
322	 * don't want the Linux-specific update to the timeval to
323	 * cause a fault after the select has completed
324	 * successfully. However, because we're not updating the
325	 * timeval, we can't restart the system call.
326	 */
327
328sticky:
329	if (ret == -ERESTARTNOHAND)
330		ret = -EINTR;
331	return ret;
332}
333
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
334#define FDS_IN(fds, n)		(fds->in + n)
335#define FDS_OUT(fds, n)		(fds->out + n)
336#define FDS_EX(fds, n)		(fds->ex + n)
337
338#define BITS(fds, n)	(*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
339
340static int max_select_fd(unsigned long n, fd_set_bits *fds)
341{
342	unsigned long *open_fds;
343	unsigned long set;
344	int max;
345	struct fdtable *fdt;
346
347	/* handle last in-complete long-word first */
348	set = ~(~0UL << (n & (__NFDBITS-1)));
349	n /= __NFDBITS;
350	fdt = files_fdtable(current->files);
351	open_fds = fdt->open_fds->fds_bits+n;
352	max = 0;
353	if (set) {
354		set &= BITS(fds, n);
355		if (set) {
356			if (!(set & ~*open_fds))
357				goto get_max;
358			return -EBADF;
359		}
360	}
361	while (n) {
362		open_fds--;
363		n--;
364		set = BITS(fds, n);
365		if (!set)
366			continue;
367		if (set & ~*open_fds)
368			return -EBADF;
369		if (max)
370			continue;
371get_max:
372		do {
373			max++;
374			set >>= 1;
375		} while (set);
376		max += n * __NFDBITS;
377	}
378
379	return max;
380}
381
382#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
383#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
384#define POLLEX_SET (POLLPRI)
385
386static inline void wait_key_set(poll_table *wait, unsigned long in,
387				unsigned long out, unsigned long bit)
 
388{
389	if (wait) {
390		wait->key = POLLEX_SET;
391		if (in & bit)
392			wait->key |= POLLIN_SET;
393		if (out & bit)
394			wait->key |= POLLOUT_SET;
395	}
396}
397
398int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
399{
400	ktime_t expire, *to = NULL;
401	struct poll_wqueues table;
402	poll_table *wait;
403	int retval, i, timed_out = 0;
404	unsigned long slack = 0;
 
 
405
406	rcu_read_lock();
407	retval = max_select_fd(n, fds);
408	rcu_read_unlock();
409
410	if (retval < 0)
411		return retval;
412	n = retval;
413
414	poll_initwait(&table);
415	wait = &table.pt;
416	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
417		wait = NULL;
418		timed_out = 1;
419	}
420
421	if (end_time && !timed_out)
422		slack = select_estimate_accuracy(end_time);
423
424	retval = 0;
425	for (;;) {
426		unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
 
427
428		inp = fds->in; outp = fds->out; exp = fds->ex;
429		rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
430
431		for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
432			unsigned long in, out, ex, all_bits, bit = 1, mask, j;
433			unsigned long res_in = 0, res_out = 0, res_ex = 0;
434			const struct file_operations *f_op = NULL;
435			struct file *file = NULL;
436
437			in = *inp++; out = *outp++; ex = *exp++;
438			all_bits = in | out | ex;
439			if (all_bits == 0) {
440				i += __NFDBITS;
441				continue;
442			}
443
444			for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
445				int fput_needed;
446				if (i >= n)
447					break;
448				if (!(bit & all_bits))
449					continue;
450				file = fget_light(i, &fput_needed);
451				if (file) {
452					f_op = file->f_op;
 
453					mask = DEFAULT_POLLMASK;
454					if (f_op && f_op->poll) {
455						wait_key_set(wait, in, out, bit);
456						mask = (*f_op->poll)(file, wait);
 
457					}
458					fput_light(file, fput_needed);
459					if ((mask & POLLIN_SET) && (in & bit)) {
460						res_in |= bit;
461						retval++;
462						wait = NULL;
463					}
464					if ((mask & POLLOUT_SET) && (out & bit)) {
465						res_out |= bit;
466						retval++;
467						wait = NULL;
468					}
469					if ((mask & POLLEX_SET) && (ex & bit)) {
470						res_ex |= bit;
471						retval++;
472						wait = NULL;
473					}
 
 
 
 
 
 
 
 
 
 
 
 
474				}
475			}
476			if (res_in)
477				*rinp = res_in;
478			if (res_out)
479				*routp = res_out;
480			if (res_ex)
481				*rexp = res_ex;
482			cond_resched();
483		}
484		wait = NULL;
485		if (retval || timed_out || signal_pending(current))
486			break;
487		if (table.error) {
488			retval = table.error;
489			break;
490		}
491
 
 
 
 
 
 
 
 
 
 
 
492		/*
493		 * If this is the first loop and we have a timeout
494		 * given, then we convert to ktime_t and set the to
495		 * pointer to the expiry value.
496		 */
497		if (end_time && !to) {
498			expire = timespec_to_ktime(*end_time);
499			to = &expire;
500		}
501
502		if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
503					   to, slack))
504			timed_out = 1;
505	}
506
507	poll_freewait(&table);
508
509	return retval;
510}
511
512/*
513 * We can actually return ERESTARTSYS instead of EINTR, but I'd
514 * like to be certain this leads to no problems. So I return
515 * EINTR just for safety.
516 *
517 * Update: ERESTARTSYS breaks at least the xview clock binary, so
518 * I'm trying ERESTARTNOHAND which restart only when you want to.
519 */
520int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
521			   fd_set __user *exp, struct timespec *end_time)
522{
523	fd_set_bits fds;
524	void *bits;
525	int ret, max_fds;
526	unsigned int size;
527	struct fdtable *fdt;
528	/* Allocate small arguments on the stack to save memory and be faster */
529	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
530
531	ret = -EINVAL;
532	if (n < 0)
533		goto out_nofds;
534
535	/* max_fds can increase, so grab it once to avoid race */
536	rcu_read_lock();
537	fdt = files_fdtable(current->files);
538	max_fds = fdt->max_fds;
539	rcu_read_unlock();
540	if (n > max_fds)
541		n = max_fds;
542
543	/*
544	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
545	 * since we used fdset we need to allocate memory in units of
546	 * long-words. 
547	 */
548	size = FDS_BYTES(n);
549	bits = stack_fds;
550	if (size > sizeof(stack_fds) / 6) {
551		/* Not enough space in on-stack array; must use kmalloc */
552		ret = -ENOMEM;
553		bits = kmalloc(6 * size, GFP_KERNEL);
 
 
 
 
554		if (!bits)
555			goto out_nofds;
556	}
557	fds.in      = bits;
558	fds.out     = bits +   size;
559	fds.ex      = bits + 2*size;
560	fds.res_in  = bits + 3*size;
561	fds.res_out = bits + 4*size;
562	fds.res_ex  = bits + 5*size;
563
564	if ((ret = get_fd_set(n, inp, fds.in)) ||
565	    (ret = get_fd_set(n, outp, fds.out)) ||
566	    (ret = get_fd_set(n, exp, fds.ex)))
567		goto out;
568	zero_fd_set(n, fds.res_in);
569	zero_fd_set(n, fds.res_out);
570	zero_fd_set(n, fds.res_ex);
571
572	ret = do_select(n, &fds, end_time);
573
574	if (ret < 0)
575		goto out;
576	if (!ret) {
577		ret = -ERESTARTNOHAND;
578		if (signal_pending(current))
579			goto out;
580		ret = 0;
581	}
582
583	if (set_fd_set(n, inp, fds.res_in) ||
584	    set_fd_set(n, outp, fds.res_out) ||
585	    set_fd_set(n, exp, fds.res_ex))
586		ret = -EFAULT;
587
588out:
589	if (bits != stack_fds)
590		kfree(bits);
591out_nofds:
592	return ret;
593}
594
595SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
596		fd_set __user *, exp, struct timeval __user *, tvp)
597{
598	struct timespec end_time, *to = NULL;
599	struct timeval tv;
600	int ret;
601
602	if (tvp) {
603		if (copy_from_user(&tv, tvp, sizeof(tv)))
604			return -EFAULT;
605
606		to = &end_time;
607		if (poll_select_set_timeout(to,
608				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
609				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
610			return -EINVAL;
611	}
612
613	ret = core_sys_select(n, inp, outp, exp, to);
614	ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
615
616	return ret;
617}
618
619#ifdef HAVE_SET_RESTORE_SIGMASK
 
 
 
 
 
620static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
621		       fd_set __user *exp, struct timespec __user *tsp,
622		       const sigset_t __user *sigmask, size_t sigsetsize)
623{
624	sigset_t ksigmask, sigsaved;
625	struct timespec ts, end_time, *to = NULL;
626	int ret;
627
628	if (tsp) {
629		if (copy_from_user(&ts, tsp, sizeof(ts)))
630			return -EFAULT;
631
632		to = &end_time;
633		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
634			return -EINVAL;
635	}
636
637	if (sigmask) {
638		/* XXX: Don't preclude handling different sized sigset_t's.  */
639		if (sigsetsize != sizeof(sigset_t))
640			return -EINVAL;
641		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
642			return -EFAULT;
643
644		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
645		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
646	}
647
648	ret = core_sys_select(n, inp, outp, exp, to);
649	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
650
651	if (ret == -ERESTARTNOHAND) {
652		/*
653		 * Don't restore the signal mask yet. Let do_signal() deliver
654		 * the signal on the way back to userspace, before the signal
655		 * mask is restored.
656		 */
657		if (sigmask) {
658			memcpy(&current->saved_sigmask, &sigsaved,
659					sizeof(sigsaved));
660			set_restore_sigmask();
661		}
662	} else if (sigmask)
663		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
664
665	return ret;
666}
667
668/*
669 * Most architectures can't handle 7-argument syscalls. So we provide a
670 * 6-argument version where the sixth argument is a pointer to a structure
671 * which has a pointer to the sigset_t itself followed by a size_t containing
672 * the sigset size.
673 */
674SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
675		fd_set __user *, exp, struct timespec __user *, tsp,
676		void __user *, sig)
677{
678	size_t sigsetsize = 0;
679	sigset_t __user *up = NULL;
680
681	if (sig) {
682		if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
683		    || __get_user(up, (sigset_t __user * __user *)sig)
684		    || __get_user(sigsetsize,
685				(size_t __user *)(sig+sizeof(void *))))
686			return -EFAULT;
687	}
688
689	return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
690}
691#endif /* HAVE_SET_RESTORE_SIGMASK */
692
693#ifdef __ARCH_WANT_SYS_OLD_SELECT
694struct sel_arg_struct {
695	unsigned long n;
696	fd_set __user *inp, *outp, *exp;
697	struct timeval __user *tvp;
698};
699
700SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
701{
702	struct sel_arg_struct a;
703
704	if (copy_from_user(&a, arg, sizeof(a)))
705		return -EFAULT;
706	return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
707}
708#endif
709
710struct poll_list {
711	struct poll_list *next;
712	int len;
713	struct pollfd entries[0];
714};
715
716#define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
717
718/*
719 * Fish for pollable events on the pollfd->fd file descriptor. We're only
720 * interested in events matching the pollfd->events mask, and the result
721 * matching that mask is both recorded in pollfd->revents and returned. The
722 * pwait poll_table will be used by the fd-provided poll handler for waiting,
723 * if non-NULL.
724 */
725static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
 
 
726{
727	unsigned int mask;
728	int fd;
729
730	mask = 0;
731	fd = pollfd->fd;
732	if (fd >= 0) {
733		int fput_needed;
734		struct file * file;
735
736		file = fget_light(fd, &fput_needed);
737		mask = POLLNVAL;
738		if (file != NULL) {
739			mask = DEFAULT_POLLMASK;
740			if (file->f_op && file->f_op->poll) {
741				if (pwait)
742					pwait->key = pollfd->events |
743							POLLERR | POLLHUP;
744				mask = file->f_op->poll(file, pwait);
 
745			}
746			/* Mask out unneeded events. */
747			mask &= pollfd->events | POLLERR | POLLHUP;
748			fput_light(file, fput_needed);
749		}
750	}
751	pollfd->revents = mask;
 
752
753	return mask;
754}
755
756static int do_poll(unsigned int nfds,  struct poll_list *list,
757		   struct poll_wqueues *wait, struct timespec *end_time)
758{
759	poll_table* pt = &wait->pt;
760	ktime_t expire, *to = NULL;
761	int timed_out = 0, count = 0;
762	unsigned long slack = 0;
 
 
763
764	/* Optimise the no-wait case */
765	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
766		pt = NULL;
767		timed_out = 1;
768	}
769
770	if (end_time && !timed_out)
771		slack = select_estimate_accuracy(end_time);
772
773	for (;;) {
774		struct poll_list *walk;
 
775
776		for (walk = list; walk != NULL; walk = walk->next) {
777			struct pollfd * pfd, * pfd_end;
778
779			pfd = walk->entries;
780			pfd_end = pfd + walk->len;
781			for (; pfd != pfd_end; pfd++) {
782				/*
783				 * Fish for events. If we found one, record it
784				 * and kill the poll_table, so we don't
785				 * needlessly register any other waiters after
786				 * this. They'll get immediately deregistered
787				 * when we break out and return.
788				 */
789				if (do_pollfd(pfd, pt)) {
 
790					count++;
791					pt = NULL;
 
 
 
792				}
793			}
794		}
795		/*
796		 * All waiters have already been registered, so don't provide
797		 * a poll_table to them on the next loop iteration.
798		 */
799		pt = NULL;
800		if (!count) {
801			count = wait->error;
802			if (signal_pending(current))
803				count = -EINTR;
804		}
805		if (count || timed_out)
806			break;
807
 
 
 
 
 
 
 
 
 
 
 
808		/*
809		 * If this is the first loop and we have a timeout
810		 * given, then we convert to ktime_t and set the to
811		 * pointer to the expiry value.
812		 */
813		if (end_time && !to) {
814			expire = timespec_to_ktime(*end_time);
815			to = &expire;
816		}
817
818		if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
819			timed_out = 1;
820	}
821	return count;
822}
823
824#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
825			sizeof(struct pollfd))
826
827int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
828		struct timespec *end_time)
829{
830	struct poll_wqueues table;
831 	int err = -EFAULT, fdcount, len, size;
832	/* Allocate small arguments on the stack to save memory and be
833	   faster - use long to make sure the buffer is aligned properly
834	   on 64 bit archs to avoid unaligned access */
835	long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
836	struct poll_list *const head = (struct poll_list *)stack_pps;
837 	struct poll_list *walk = head;
838 	unsigned long todo = nfds;
839
840	if (nfds > rlimit(RLIMIT_NOFILE))
841		return -EINVAL;
842
843	len = min_t(unsigned int, nfds, N_STACK_PPS);
844	for (;;) {
845		walk->next = NULL;
846		walk->len = len;
847		if (!len)
848			break;
849
850		if (copy_from_user(walk->entries, ufds + nfds-todo,
851					sizeof(struct pollfd) * walk->len))
852			goto out_fds;
853
854		todo -= walk->len;
855		if (!todo)
856			break;
857
858		len = min(todo, POLLFD_PER_PAGE);
859		size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
860		walk = walk->next = kmalloc(size, GFP_KERNEL);
861		if (!walk) {
862			err = -ENOMEM;
863			goto out_fds;
864		}
865	}
866
867	poll_initwait(&table);
868	fdcount = do_poll(nfds, head, &table, end_time);
869	poll_freewait(&table);
870
871	for (walk = head; walk; walk = walk->next) {
872		struct pollfd *fds = walk->entries;
873		int j;
874
875		for (j = 0; j < walk->len; j++, ufds++)
876			if (__put_user(fds[j].revents, &ufds->revents))
877				goto out_fds;
878  	}
879
880	err = fdcount;
881out_fds:
882	walk = head->next;
883	while (walk) {
884		struct poll_list *pos = walk;
885		walk = walk->next;
886		kfree(pos);
887	}
888
889	return err;
890}
891
892static long do_restart_poll(struct restart_block *restart_block)
893{
894	struct pollfd __user *ufds = restart_block->poll.ufds;
895	int nfds = restart_block->poll.nfds;
896	struct timespec *to = NULL, end_time;
897	int ret;
898
899	if (restart_block->poll.has_timeout) {
900		end_time.tv_sec = restart_block->poll.tv_sec;
901		end_time.tv_nsec = restart_block->poll.tv_nsec;
902		to = &end_time;
903	}
904
905	ret = do_sys_poll(ufds, nfds, to);
906
907	if (ret == -EINTR) {
908		restart_block->fn = do_restart_poll;
909		ret = -ERESTART_RESTARTBLOCK;
910	}
911	return ret;
912}
913
914SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
915		long, timeout_msecs)
916{
917	struct timespec end_time, *to = NULL;
918	int ret;
919
920	if (timeout_msecs >= 0) {
921		to = &end_time;
922		poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
923			NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
924	}
925
926	ret = do_sys_poll(ufds, nfds, to);
927
928	if (ret == -EINTR) {
929		struct restart_block *restart_block;
930
931		restart_block = &current_thread_info()->restart_block;
932		restart_block->fn = do_restart_poll;
933		restart_block->poll.ufds = ufds;
934		restart_block->poll.nfds = nfds;
935
936		if (timeout_msecs >= 0) {
937			restart_block->poll.tv_sec = end_time.tv_sec;
938			restart_block->poll.tv_nsec = end_time.tv_nsec;
939			restart_block->poll.has_timeout = 1;
940		} else
941			restart_block->poll.has_timeout = 0;
942
943		ret = -ERESTART_RESTARTBLOCK;
944	}
945	return ret;
946}
947
948#ifdef HAVE_SET_RESTORE_SIGMASK
949SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
950		struct timespec __user *, tsp, const sigset_t __user *, sigmask,
951		size_t, sigsetsize)
952{
953	sigset_t ksigmask, sigsaved;
954	struct timespec ts, end_time, *to = NULL;
955	int ret;
956
957	if (tsp) {
958		if (copy_from_user(&ts, tsp, sizeof(ts)))
959			return -EFAULT;
960
961		to = &end_time;
962		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
963			return -EINVAL;
964	}
965
966	if (sigmask) {
967		/* XXX: Don't preclude handling different sized sigset_t's.  */
968		if (sigsetsize != sizeof(sigset_t))
969			return -EINVAL;
970		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
971			return -EFAULT;
972
973		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
974		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
975	}
976
977	ret = do_sys_poll(ufds, nfds, to);
978
979	/* We can restart this syscall, usually */
980	if (ret == -EINTR) {
981		/*
982		 * Don't restore the signal mask yet. Let do_signal() deliver
983		 * the signal on the way back to userspace, before the signal
984		 * mask is restored.
985		 */
986		if (sigmask) {
987			memcpy(&current->saved_sigmask, &sigsaved,
988					sizeof(sigsaved));
989			set_restore_sigmask();
990		}
991		ret = -ERESTARTNOHAND;
992	} else if (sigmask)
993		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
994
995	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
996
997	return ret;
998}
999#endif /* HAVE_SET_RESTORE_SIGMASK */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * This file contains the procedures for the handling of select and poll
   4 *
   5 * Created for Linux based loosely upon Mathius Lattner's minix
   6 * patches by Peter MacDonald. Heavily edited by Linus.
   7 *
   8 *  4 February 1994
   9 *     COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
  10 *     flag set in its personality we do *not* modify the given timeout
  11 *     parameter to reflect time remaining.
  12 *
  13 *  24 January 2000
  14 *     Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation 
  15 *     of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
  16 */
  17
  18#include <linux/kernel.h>
  19#include <linux/sched/signal.h>
  20#include <linux/sched/rt.h>
  21#include <linux/syscalls.h>
  22#include <linux/export.h>
  23#include <linux/slab.h>
  24#include <linux/poll.h>
  25#include <linux/personality.h> /* for STICKY_TIMEOUTS */
  26#include <linux/file.h>
  27#include <linux/fdtable.h>
  28#include <linux/fs.h>
  29#include <linux/rcupdate.h>
  30#include <linux/hrtimer.h>
  31#include <linux/freezer.h>
  32#include <net/busy_poll.h>
  33#include <linux/vmalloc.h>
  34
  35#include <linux/uaccess.h>
  36
  37
  38/*
  39 * Estimate expected accuracy in ns from a timeval.
  40 *
  41 * After quite a bit of churning around, we've settled on
  42 * a simple thing of taking 0.1% of the timeout as the
  43 * slack, with a cap of 100 msec.
  44 * "nice" tasks get a 0.5% slack instead.
  45 *
  46 * Consider this comment an open invitation to come up with even
  47 * better solutions..
  48 */
  49
  50#define MAX_SLACK	(100 * NSEC_PER_MSEC)
  51
  52static long __estimate_accuracy(struct timespec64 *tv)
  53{
  54	long slack;
  55	int divfactor = 1000;
  56
  57	if (tv->tv_sec < 0)
  58		return 0;
  59
  60	if (task_nice(current) > 0)
  61		divfactor = divfactor / 5;
  62
  63	if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
  64		return MAX_SLACK;
  65
  66	slack = tv->tv_nsec / divfactor;
  67	slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
  68
  69	if (slack > MAX_SLACK)
  70		return MAX_SLACK;
  71
  72	return slack;
  73}
  74
  75u64 select_estimate_accuracy(struct timespec64 *tv)
  76{
  77	u64 ret;
  78	struct timespec64 now;
  79
  80	/*
  81	 * Realtime tasks get a slack of 0 for obvious reasons.
  82	 */
  83
  84	if (rt_task(current))
  85		return 0;
  86
  87	ktime_get_ts64(&now);
  88	now = timespec64_sub(*tv, now);
  89	ret = __estimate_accuracy(&now);
  90	if (ret < current->timer_slack_ns)
  91		return current->timer_slack_ns;
  92	return ret;
  93}
  94
  95
  96
  97struct poll_table_page {
  98	struct poll_table_page * next;
  99	struct poll_table_entry * entry;
 100	struct poll_table_entry entries[0];
 101};
 102
 103#define POLL_TABLE_FULL(table) \
 104	((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
 105
 106/*
 107 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
 108 * I have rewritten this, taking some shortcuts: This code may not be easy to
 109 * follow, but it should be free of race-conditions, and it's practical. If you
 110 * understand what I'm doing here, then you understand how the linux
 111 * sleep/wakeup mechanism works.
 112 *
 113 * Two very simple procedures, poll_wait() and poll_freewait() make all the
 114 * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
 115 * as all select/poll functions have to call it to add an entry to the
 116 * poll table.
 117 */
 118static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 119		       poll_table *p);
 120
 121void poll_initwait(struct poll_wqueues *pwq)
 122{
 123	init_poll_funcptr(&pwq->pt, __pollwait);
 124	pwq->polling_task = current;
 125	pwq->triggered = 0;
 126	pwq->error = 0;
 127	pwq->table = NULL;
 128	pwq->inline_index = 0;
 129}
 130EXPORT_SYMBOL(poll_initwait);
 131
 132static void free_poll_entry(struct poll_table_entry *entry)
 133{
 134	remove_wait_queue(entry->wait_address, &entry->wait);
 135	fput(entry->filp);
 136}
 137
 138void poll_freewait(struct poll_wqueues *pwq)
 139{
 140	struct poll_table_page * p = pwq->table;
 141	int i;
 142	for (i = 0; i < pwq->inline_index; i++)
 143		free_poll_entry(pwq->inline_entries + i);
 144	while (p) {
 145		struct poll_table_entry * entry;
 146		struct poll_table_page *old;
 147
 148		entry = p->entry;
 149		do {
 150			entry--;
 151			free_poll_entry(entry);
 152		} while (entry > p->entries);
 153		old = p;
 154		p = p->next;
 155		free_page((unsigned long) old);
 156	}
 157}
 158EXPORT_SYMBOL(poll_freewait);
 159
 160static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
 161{
 162	struct poll_table_page *table = p->table;
 163
 164	if (p->inline_index < N_INLINE_POLL_ENTRIES)
 165		return p->inline_entries + p->inline_index++;
 166
 167	if (!table || POLL_TABLE_FULL(table)) {
 168		struct poll_table_page *new_table;
 169
 170		new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
 171		if (!new_table) {
 172			p->error = -ENOMEM;
 173			return NULL;
 174		}
 175		new_table->entry = new_table->entries;
 176		new_table->next = table;
 177		p->table = new_table;
 178		table = new_table;
 179	}
 180
 181	return table->entry++;
 182}
 183
 184static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
 185{
 186	struct poll_wqueues *pwq = wait->private;
 187	DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
 188
 189	/*
 190	 * Although this function is called under waitqueue lock, LOCK
 191	 * doesn't imply write barrier and the users expect write
 192	 * barrier semantics on wakeup functions.  The following
 193	 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
 194	 * and is paired with smp_store_mb() in poll_schedule_timeout.
 195	 */
 196	smp_wmb();
 197	pwq->triggered = 1;
 198
 199	/*
 200	 * Perform the default wake up operation using a dummy
 201	 * waitqueue.
 202	 *
 203	 * TODO: This is hacky but there currently is no interface to
 204	 * pass in @sync.  @sync is scheduled to be removed and once
 205	 * that happens, wake_up_process() can be used directly.
 206	 */
 207	return default_wake_function(&dummy_wait, mode, sync, key);
 208}
 209
 210static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
 211{
 212	struct poll_table_entry *entry;
 213
 214	entry = container_of(wait, struct poll_table_entry, wait);
 215	if (key && !(key_to_poll(key) & entry->key))
 216		return 0;
 217	return __pollwake(wait, mode, sync, key);
 218}
 219
 220/* Add a new entry */
 221static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 222				poll_table *p)
 223{
 224	struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
 225	struct poll_table_entry *entry = poll_get_entry(pwq);
 226	if (!entry)
 227		return;
 228	entry->filp = get_file(filp);
 
 229	entry->wait_address = wait_address;
 230	entry->key = p->_key;
 231	init_waitqueue_func_entry(&entry->wait, pollwake);
 232	entry->wait.private = pwq;
 233	add_wait_queue(wait_address, &entry->wait);
 234}
 235
 236int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
 237			  ktime_t *expires, unsigned long slack)
 238{
 239	int rc = -EINTR;
 240
 241	set_current_state(state);
 242	if (!pwq->triggered)
 243		rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
 244	__set_current_state(TASK_RUNNING);
 245
 246	/*
 247	 * Prepare for the next iteration.
 248	 *
 249	 * The following smp_store_mb() serves two purposes.  First, it's
 250	 * the counterpart rmb of the wmb in pollwake() such that data
 251	 * written before wake up is always visible after wake up.
 252	 * Second, the full barrier guarantees that triggered clearing
 253	 * doesn't pass event check of the next iteration.  Note that
 254	 * this problem doesn't exist for the first iteration as
 255	 * add_wait_queue() has full barrier semantics.
 256	 */
 257	smp_store_mb(pwq->triggered, 0);
 258
 259	return rc;
 260}
 261EXPORT_SYMBOL(poll_schedule_timeout);
 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 */
 274int 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
 291static int poll_select_copy_remaining(struct timespec64 *end_time,
 292				      void __user *p,
 293				      int timeval, int ret)
 294{
 295	struct timespec64 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(&rts);
 309	rts = timespec64_sub(*end_time, rts);
 310	if (rts.tv_sec < 0)
 311		rts.tv_sec = rts.tv_nsec = 0;
 312
 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 = rts.tv_sec;
 318		rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
 319
 320		if (!copy_to_user(p, &rtv, sizeof(rtv)))
 321			return ret;
 322
 323	} else if (!put_timespec64(&rts, p))
 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
 334sticky:
 335	if (ret == -ERESTARTNOHAND)
 336		ret = -EINTR;
 337	return ret;
 338}
 339
 340/*
 341 * Scalable version of the fd_set.
 342 */
 343
 344typedef struct {
 345	unsigned long *in, *out, *ex;
 346	unsigned long *res_in, *res_out, *res_ex;
 347} fd_set_bits;
 348
 349/*
 350 * How many longwords for "nr" bits?
 351 */
 352#define FDS_BITPERLONG	(8*sizeof(long))
 353#define FDS_LONGS(nr)	(((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
 354#define FDS_BYTES(nr)	(FDS_LONGS(nr)*sizeof(long))
 355
 356/*
 357 * We do a VERIFY_WRITE here even though we are only reading this time:
 358 * we'll write to it eventually..
 359 *
 360 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
 361 */
 362static inline
 363int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
 364{
 365	nr = FDS_BYTES(nr);
 366	if (ufdset)
 367		return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
 368
 369	memset(fdset, 0, nr);
 370	return 0;
 371}
 372
 373static inline unsigned long __must_check
 374set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
 375{
 376	if (ufdset)
 377		return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
 378	return 0;
 379}
 380
 381static inline
 382void zero_fd_set(unsigned long nr, unsigned long *fdset)
 383{
 384	memset(fdset, 0, FDS_BYTES(nr));
 385}
 386
 387#define FDS_IN(fds, n)		(fds->in + n)
 388#define FDS_OUT(fds, n)		(fds->out + n)
 389#define FDS_EX(fds, n)		(fds->ex + n)
 390
 391#define BITS(fds, n)	(*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
 392
 393static int max_select_fd(unsigned long n, fd_set_bits *fds)
 394{
 395	unsigned long *open_fds;
 396	unsigned long set;
 397	int max;
 398	struct fdtable *fdt;
 399
 400	/* handle last in-complete long-word first */
 401	set = ~(~0UL << (n & (BITS_PER_LONG-1)));
 402	n /= BITS_PER_LONG;
 403	fdt = files_fdtable(current->files);
 404	open_fds = fdt->open_fds + n;
 405	max = 0;
 406	if (set) {
 407		set &= BITS(fds, n);
 408		if (set) {
 409			if (!(set & ~*open_fds))
 410				goto get_max;
 411			return -EBADF;
 412		}
 413	}
 414	while (n) {
 415		open_fds--;
 416		n--;
 417		set = BITS(fds, n);
 418		if (!set)
 419			continue;
 420		if (set & ~*open_fds)
 421			return -EBADF;
 422		if (max)
 423			continue;
 424get_max:
 425		do {
 426			max++;
 427			set >>= 1;
 428		} while (set);
 429		max += n * BITS_PER_LONG;
 430	}
 431
 432	return max;
 433}
 434
 435#define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR)
 436#define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
 437#define POLLEX_SET (EPOLLPRI)
 438
 439static inline void wait_key_set(poll_table *wait, unsigned long in,
 440				unsigned long out, unsigned long bit,
 441				__poll_t ll_flag)
 442{
 443	wait->_key = POLLEX_SET | ll_flag;
 444	if (in & bit)
 445		wait->_key |= POLLIN_SET;
 446	if (out & bit)
 447		wait->_key |= POLLOUT_SET;
 
 
 448}
 449
 450static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
 451{
 452	ktime_t expire, *to = NULL;
 453	struct poll_wqueues table;
 454	poll_table *wait;
 455	int retval, i, timed_out = 0;
 456	u64 slack = 0;
 457	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
 458	unsigned long busy_start = 0;
 459
 460	rcu_read_lock();
 461	retval = max_select_fd(n, fds);
 462	rcu_read_unlock();
 463
 464	if (retval < 0)
 465		return retval;
 466	n = retval;
 467
 468	poll_initwait(&table);
 469	wait = &table.pt;
 470	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 471		wait->_qproc = NULL;
 472		timed_out = 1;
 473	}
 474
 475	if (end_time && !timed_out)
 476		slack = select_estimate_accuracy(end_time);
 477
 478	retval = 0;
 479	for (;;) {
 480		unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
 481		bool can_busy_loop = false;
 482
 483		inp = fds->in; outp = fds->out; exp = fds->ex;
 484		rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
 485
 486		for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
 487			unsigned long in, out, ex, all_bits, bit = 1, j;
 488			unsigned long res_in = 0, res_out = 0, res_ex = 0;
 489			__poll_t mask;
 
 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 */
 599int 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 = kvmalloc(alloc_size, GFP_KERNEL);
 637		if (!bits)
 638			goto out_nofds;
 639	}
 640	fds.in      = bits;
 641	fds.out     = bits +   size;
 642	fds.ex      = bits + 2*size;
 643	fds.res_in  = bits + 3*size;
 644	fds.res_out = bits + 4*size;
 645	fds.res_ex  = bits + 5*size;
 646
 647	if ((ret = get_fd_set(n, inp, fds.in)) ||
 648	    (ret = get_fd_set(n, outp, fds.out)) ||
 649	    (ret = get_fd_set(n, exp, fds.ex)))
 650		goto out;
 651	zero_fd_set(n, fds.res_in);
 652	zero_fd_set(n, fds.res_out);
 653	zero_fd_set(n, fds.res_ex);
 654
 655	ret = do_select(n, &fds, end_time);
 656
 657	if (ret < 0)
 658		goto out;
 659	if (!ret) {
 660		ret = -ERESTARTNOHAND;
 661		if (signal_pending(current))
 662			goto out;
 663		ret = 0;
 664	}
 665
 666	if (set_fd_set(n, inp, fds.res_in) ||
 667	    set_fd_set(n, outp, fds.res_out) ||
 668	    set_fd_set(n, exp, fds.res_ex))
 669		ret = -EFAULT;
 670
 671out:
 672	if (bits != stack_fds)
 673		kvfree(bits);
 674out_nofds:
 675	return ret;
 676}
 677
 678static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
 679		       fd_set __user *exp, struct timeval __user *tvp)
 680{
 681	struct timespec64 end_time, *to = NULL;
 682	struct timeval tv;
 683	int ret;
 684
 685	if (tvp) {
 686		if (copy_from_user(&tv, tvp, sizeof(tv)))
 687			return -EFAULT;
 688
 689		to = &end_time;
 690		if (poll_select_set_timeout(to,
 691				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
 692				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
 693			return -EINVAL;
 694	}
 695
 696	ret = core_sys_select(n, inp, outp, exp, to);
 697	ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
 698
 699	return ret;
 700}
 701
 702SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
 703		fd_set __user *, exp, struct timeval __user *, tvp)
 704{
 705	return kern_select(n, inp, outp, exp, tvp);
 706}
 707
 708static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
 709		       fd_set __user *exp, struct timespec __user *tsp,
 710		       const sigset_t __user *sigmask, size_t sigsetsize)
 711{
 712	sigset_t ksigmask, sigsaved;
 713	struct timespec64 ts, end_time, *to = NULL;
 714	int ret;
 715
 716	if (tsp) {
 717		if (get_timespec64(&ts, tsp))
 718			return -EFAULT;
 719
 720		to = &end_time;
 721		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
 722			return -EINVAL;
 723	}
 724
 725	if (sigmask) {
 726		/* XXX: Don't preclude handling different sized sigset_t's.  */
 727		if (sigsetsize != sizeof(sigset_t))
 728			return -EINVAL;
 729		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
 730			return -EFAULT;
 731
 732		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
 733		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
 734	}
 735
 736	ret = core_sys_select(n, inp, outp, exp, to);
 737	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
 738
 739	if (ret == -ERESTARTNOHAND) {
 740		/*
 741		 * Don't restore the signal mask yet. Let do_signal() deliver
 742		 * the signal on the way back to userspace, before the signal
 743		 * mask is restored.
 744		 */
 745		if (sigmask) {
 746			memcpy(&current->saved_sigmask, &sigsaved,
 747					sizeof(sigsaved));
 748			set_restore_sigmask();
 749		}
 750	} else if (sigmask)
 751		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 752
 753	return ret;
 754}
 755
 756/*
 757 * Most architectures can't handle 7-argument syscalls. So we provide a
 758 * 6-argument version where the sixth argument is a pointer to a structure
 759 * which has a pointer to the sigset_t itself followed by a size_t containing
 760 * the sigset size.
 761 */
 762SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
 763		fd_set __user *, exp, struct timespec __user *, tsp,
 764		void __user *, sig)
 765{
 766	size_t sigsetsize = 0;
 767	sigset_t __user *up = NULL;
 768
 769	if (sig) {
 770		if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
 771		    || __get_user(up, (sigset_t __user * __user *)sig)
 772		    || __get_user(sigsetsize,
 773				(size_t __user *)(sig+sizeof(void *))))
 774			return -EFAULT;
 775	}
 776
 777	return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
 778}
 
 779
 780#ifdef __ARCH_WANT_SYS_OLD_SELECT
 781struct sel_arg_struct {
 782	unsigned long n;
 783	fd_set __user *inp, *outp, *exp;
 784	struct timeval __user *tvp;
 785};
 786
 787SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
 788{
 789	struct sel_arg_struct a;
 790
 791	if (copy_from_user(&a, arg, sizeof(a)))
 792		return -EFAULT;
 793	return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
 794}
 795#endif
 796
 797struct poll_list {
 798	struct poll_list *next;
 799	int len;
 800	struct pollfd entries[0];
 801};
 802
 803#define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
 804
 805/*
 806 * Fish for pollable events on the pollfd->fd file descriptor. We're only
 807 * interested in events matching the pollfd->events mask, and the result
 808 * matching that mask is both recorded in pollfd->revents and returned. The
 809 * pwait poll_table will be used by the fd-provided poll handler for waiting,
 810 * if pwait->_qproc is non-NULL.
 811 */
 812static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
 813				     bool *can_busy_poll,
 814				     __poll_t busy_flag)
 815{
 816	__poll_t mask;
 817	int fd;
 818
 819	mask = 0;
 820	fd = pollfd->fd;
 821	if (fd >= 0) {
 822		struct fd f = fdget(fd);
 823		mask = EPOLLNVAL;
 824		if (f.file) {
 825			/* userland u16 ->events contains POLL... bitmap */
 826			__poll_t filter = demangle_poll(pollfd->events) |
 827						EPOLLERR | EPOLLHUP;
 828			mask = DEFAULT_POLLMASK;
 829			if (f.file->f_op->poll) {
 830				pwait->_key = filter;
 831				pwait->_key |= busy_flag;
 832				mask = f.file->f_op->poll(f.file, pwait);
 833				if (mask & busy_flag)
 834					*can_busy_poll = true;
 835			}
 836			/* Mask out unneeded events. */
 837			mask &= filter;
 838			fdput(f);
 839		}
 840	}
 841	/* ... and so does ->revents */
 842	pollfd->revents = mangle_poll(mask);
 843
 844	return mask;
 845}
 846
 847static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
 848		   struct timespec64 *end_time)
 849{
 850	poll_table* pt = &wait->pt;
 851	ktime_t expire, *to = NULL;
 852	int timed_out = 0, count = 0;
 853	u64 slack = 0;
 854	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
 855	unsigned long busy_start = 0;
 856
 857	/* Optimise the no-wait case */
 858	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 859		pt->_qproc = NULL;
 860		timed_out = 1;
 861	}
 862
 863	if (end_time && !timed_out)
 864		slack = select_estimate_accuracy(end_time);
 865
 866	for (;;) {
 867		struct poll_list *walk;
 868		bool can_busy_loop = false;
 869
 870		for (walk = list; walk != NULL; walk = walk->next) {
 871			struct pollfd * pfd, * pfd_end;
 872
 873			pfd = walk->entries;
 874			pfd_end = pfd + walk->len;
 875			for (; pfd != pfd_end; pfd++) {
 876				/*
 877				 * Fish for events. If we found one, record it
 878				 * and kill poll_table->_qproc, so we don't
 879				 * needlessly register any other waiters after
 880				 * this. They'll get immediately deregistered
 881				 * when we break out and return.
 882				 */
 883				if (do_pollfd(pfd, pt, &can_busy_loop,
 884					      busy_flag)) {
 885					count++;
 886					pt->_qproc = NULL;
 887					/* found something, stop busy polling */
 888					busy_flag = 0;
 889					can_busy_loop = false;
 890				}
 891			}
 892		}
 893		/*
 894		 * All waiters have already been registered, so don't provide
 895		 * a poll_table->_qproc to them on the next loop iteration.
 896		 */
 897		pt->_qproc = NULL;
 898		if (!count) {
 899			count = wait->error;
 900			if (signal_pending(current))
 901				count = -EINTR;
 902		}
 903		if (count || timed_out)
 904			break;
 905
 906		/* only if found POLL_BUSY_LOOP sockets && not out of time */
 907		if (can_busy_loop && !need_resched()) {
 908			if (!busy_start) {
 909				busy_start = busy_loop_current_time();
 910				continue;
 911			}
 912			if (!busy_loop_timeout(busy_start))
 913				continue;
 914		}
 915		busy_flag = 0;
 916
 917		/*
 918		 * If this is the first loop and we have a timeout
 919		 * given, then we convert to ktime_t and set the to
 920		 * pointer to the expiry value.
 921		 */
 922		if (end_time && !to) {
 923			expire = timespec64_to_ktime(*end_time);
 924			to = &expire;
 925		}
 926
 927		if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
 928			timed_out = 1;
 929	}
 930	return count;
 931}
 932
 933#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
 934			sizeof(struct pollfd))
 935
 936static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
 937		struct timespec64 *end_time)
 938{
 939	struct poll_wqueues table;
 940 	int err = -EFAULT, fdcount, len, size;
 941	/* Allocate small arguments on the stack to save memory and be
 942	   faster - use long to make sure the buffer is aligned properly
 943	   on 64 bit archs to avoid unaligned access */
 944	long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
 945	struct poll_list *const head = (struct poll_list *)stack_pps;
 946 	struct poll_list *walk = head;
 947 	unsigned long todo = nfds;
 948
 949	if (nfds > rlimit(RLIMIT_NOFILE))
 950		return -EINVAL;
 951
 952	len = min_t(unsigned int, nfds, N_STACK_PPS);
 953	for (;;) {
 954		walk->next = NULL;
 955		walk->len = len;
 956		if (!len)
 957			break;
 958
 959		if (copy_from_user(walk->entries, ufds + nfds-todo,
 960					sizeof(struct pollfd) * walk->len))
 961			goto out_fds;
 962
 963		todo -= walk->len;
 964		if (!todo)
 965			break;
 966
 967		len = min(todo, POLLFD_PER_PAGE);
 968		size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
 969		walk = walk->next = kmalloc(size, GFP_KERNEL);
 970		if (!walk) {
 971			err = -ENOMEM;
 972			goto out_fds;
 973		}
 974	}
 975
 976	poll_initwait(&table);
 977	fdcount = do_poll(head, &table, end_time);
 978	poll_freewait(&table);
 979
 980	for (walk = head; walk; walk = walk->next) {
 981		struct pollfd *fds = walk->entries;
 982		int j;
 983
 984		for (j = 0; j < walk->len; j++, ufds++)
 985			if (__put_user(fds[j].revents, &ufds->revents))
 986				goto out_fds;
 987  	}
 988
 989	err = fdcount;
 990out_fds:
 991	walk = head->next;
 992	while (walk) {
 993		struct poll_list *pos = walk;
 994		walk = walk->next;
 995		kfree(pos);
 996	}
 997
 998	return err;
 999}
1000
1001static long do_restart_poll(struct restart_block *restart_block)
1002{
1003	struct pollfd __user *ufds = restart_block->poll.ufds;
1004	int nfds = restart_block->poll.nfds;
1005	struct timespec64 *to = NULL, end_time;
1006	int ret;
1007
1008	if (restart_block->poll.has_timeout) {
1009		end_time.tv_sec = restart_block->poll.tv_sec;
1010		end_time.tv_nsec = restart_block->poll.tv_nsec;
1011		to = &end_time;
1012	}
1013
1014	ret = do_sys_poll(ufds, nfds, to);
1015
1016	if (ret == -EINTR) {
1017		restart_block->fn = do_restart_poll;
1018		ret = -ERESTART_RESTARTBLOCK;
1019	}
1020	return ret;
1021}
1022
1023SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1024		int, timeout_msecs)
1025{
1026	struct timespec64 end_time, *to = NULL;
1027	int ret;
1028
1029	if (timeout_msecs >= 0) {
1030		to = &end_time;
1031		poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1032			NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1033	}
1034
1035	ret = do_sys_poll(ufds, nfds, to);
1036
1037	if (ret == -EINTR) {
1038		struct restart_block *restart_block;
1039
1040		restart_block = &current->restart_block;
1041		restart_block->fn = do_restart_poll;
1042		restart_block->poll.ufds = ufds;
1043		restart_block->poll.nfds = nfds;
1044
1045		if (timeout_msecs >= 0) {
1046			restart_block->poll.tv_sec = end_time.tv_sec;
1047			restart_block->poll.tv_nsec = end_time.tv_nsec;
1048			restart_block->poll.has_timeout = 1;
1049		} else
1050			restart_block->poll.has_timeout = 0;
1051
1052		ret = -ERESTART_RESTARTBLOCK;
1053	}
1054	return ret;
1055}
1056
 
1057SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1058		struct timespec __user *, tsp, const sigset_t __user *, sigmask,
1059		size_t, sigsetsize)
1060{
1061	sigset_t ksigmask, sigsaved;
1062	struct timespec64 ts, end_time, *to = NULL;
1063	int ret;
1064
1065	if (tsp) {
1066		if (get_timespec64(&ts, tsp))
1067			return -EFAULT;
1068
1069		to = &end_time;
1070		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1071			return -EINVAL;
1072	}
1073
1074	if (sigmask) {
1075		/* XXX: Don't preclude handling different sized sigset_t's.  */
1076		if (sigsetsize != sizeof(sigset_t))
1077			return -EINVAL;
1078		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1079			return -EFAULT;
1080
1081		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1082		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1083	}
1084
1085	ret = do_sys_poll(ufds, nfds, to);
1086
1087	/* We can restart this syscall, usually */
1088	if (ret == -EINTR) {
1089		/*
1090		 * Don't restore the signal mask yet. Let do_signal() deliver
1091		 * the signal on the way back to userspace, before the signal
1092		 * mask is restored.
1093		 */
1094		if (sigmask) {
1095			memcpy(&current->saved_sigmask, &sigsaved,
1096					sizeof(sigsaved));
1097			set_restore_sigmask();
1098		}
1099		ret = -ERESTARTNOHAND;
1100	} else if (sigmask)
1101		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1102
1103	ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1104
1105	return ret;
1106}
1107
1108#ifdef CONFIG_COMPAT
1109#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1110
1111static
1112int compat_poll_select_copy_remaining(struct timespec64 *end_time, void __user *p,
1113				      int timeval, int ret)
1114{
1115	struct timespec64 ts;
1116
1117	if (!p)
1118		return ret;
1119
1120	if (current->personality & STICKY_TIMEOUTS)
1121		goto sticky;
1122
1123	/* No update for zero timeout */
1124	if (!end_time->tv_sec && !end_time->tv_nsec)
1125		return ret;
1126
1127	ktime_get_ts64(&ts);
1128	ts = timespec64_sub(*end_time, ts);
1129	if (ts.tv_sec < 0)
1130		ts.tv_sec = ts.tv_nsec = 0;
1131
1132	if (timeval) {
1133		struct compat_timeval rtv;
1134
1135		rtv.tv_sec = ts.tv_sec;
1136		rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1137
1138		if (!copy_to_user(p, &rtv, sizeof(rtv)))
1139			return ret;
1140	} else {
1141		if (!compat_put_timespec64(&ts, p))
1142			return ret;
1143	}
1144	/*
1145	 * If an application puts its timeval in read-only memory, we
1146	 * don't want the Linux-specific update to the timeval to
1147	 * cause a fault after the select has completed
1148	 * successfully. However, because we're not updating the
1149	 * timeval, we can't restart the system call.
1150	 */
1151
1152sticky:
1153	if (ret == -ERESTARTNOHAND)
1154		ret = -EINTR;
1155	return ret;
1156}
1157
1158/*
1159 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1160 * 64-bit unsigned longs.
1161 */
1162static
1163int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1164			unsigned long *fdset)
1165{
1166	if (ufdset) {
1167		return compat_get_bitmap(fdset, ufdset, nr);
1168	} else {
1169		zero_fd_set(nr, fdset);
1170		return 0;
1171	}
1172}
1173
1174static
1175int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1176		      unsigned long *fdset)
1177{
1178	if (!ufdset)
1179		return 0;
1180	return compat_put_bitmap(ufdset, fdset, nr);
1181}
1182
1183
1184/*
1185 * This is a virtual copy of sys_select from fs/select.c and probably
1186 * should be compared to it from time to time
1187 */
1188
1189/*
1190 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1191 * like to be certain this leads to no problems. So I return
1192 * EINTR just for safety.
1193 *
1194 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1195 * I'm trying ERESTARTNOHAND which restart only when you want to.
1196 */
1197static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1198	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1199	struct timespec64 *end_time)
1200{
1201	fd_set_bits fds;
1202	void *bits;
1203	int size, max_fds, ret = -EINVAL;
1204	struct fdtable *fdt;
1205	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1206
1207	if (n < 0)
1208		goto out_nofds;
1209
1210	/* max_fds can increase, so grab it once to avoid race */
1211	rcu_read_lock();
1212	fdt = files_fdtable(current->files);
1213	max_fds = fdt->max_fds;
1214	rcu_read_unlock();
1215	if (n > max_fds)
1216		n = max_fds;
1217
1218	/*
1219	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1220	 * since we used fdset we need to allocate memory in units of
1221	 * long-words.
1222	 */
1223	size = FDS_BYTES(n);
1224	bits = stack_fds;
1225	if (size > sizeof(stack_fds) / 6) {
1226		bits = kmalloc(6 * size, GFP_KERNEL);
1227		ret = -ENOMEM;
1228		if (!bits)
1229			goto out_nofds;
1230	}
1231	fds.in      = (unsigned long *)  bits;
1232	fds.out     = (unsigned long *) (bits +   size);
1233	fds.ex      = (unsigned long *) (bits + 2*size);
1234	fds.res_in  = (unsigned long *) (bits + 3*size);
1235	fds.res_out = (unsigned long *) (bits + 4*size);
1236	fds.res_ex  = (unsigned long *) (bits + 5*size);
1237
1238	if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1239	    (ret = compat_get_fd_set(n, outp, fds.out)) ||
1240	    (ret = compat_get_fd_set(n, exp, fds.ex)))
1241		goto out;
1242	zero_fd_set(n, fds.res_in);
1243	zero_fd_set(n, fds.res_out);
1244	zero_fd_set(n, fds.res_ex);
1245
1246	ret = do_select(n, &fds, end_time);
1247
1248	if (ret < 0)
1249		goto out;
1250	if (!ret) {
1251		ret = -ERESTARTNOHAND;
1252		if (signal_pending(current))
1253			goto out;
1254		ret = 0;
1255	}
1256
1257	if (compat_set_fd_set(n, inp, fds.res_in) ||
1258	    compat_set_fd_set(n, outp, fds.res_out) ||
1259	    compat_set_fd_set(n, exp, fds.res_ex))
1260		ret = -EFAULT;
1261out:
1262	if (bits != stack_fds)
1263		kfree(bits);
1264out_nofds:
1265	return ret;
1266}
1267
1268static int do_compat_select(int n, compat_ulong_t __user *inp,
1269	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1270	struct compat_timeval __user *tvp)
1271{
1272	struct timespec64 end_time, *to = NULL;
1273	struct compat_timeval tv;
1274	int ret;
1275
1276	if (tvp) {
1277		if (copy_from_user(&tv, tvp, sizeof(tv)))
1278			return -EFAULT;
1279
1280		to = &end_time;
1281		if (poll_select_set_timeout(to,
1282				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1283				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1284			return -EINVAL;
1285	}
1286
1287	ret = compat_core_sys_select(n, inp, outp, exp, to);
1288	ret = compat_poll_select_copy_remaining(&end_time, tvp, 1, ret);
1289
1290	return ret;
1291}
1292
1293COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1294	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1295	struct compat_timeval __user *, tvp)
1296{
1297	return do_compat_select(n, inp, outp, exp, tvp);
1298}
1299
1300struct compat_sel_arg_struct {
1301	compat_ulong_t n;
1302	compat_uptr_t inp;
1303	compat_uptr_t outp;
1304	compat_uptr_t exp;
1305	compat_uptr_t tvp;
1306};
1307
1308COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1309{
1310	struct compat_sel_arg_struct a;
1311
1312	if (copy_from_user(&a, arg, sizeof(a)))
1313		return -EFAULT;
1314	return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1315				compat_ptr(a.exp), compat_ptr(a.tvp));
1316}
1317
1318static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1319	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1320	struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1321	compat_size_t sigsetsize)
1322{
1323	sigset_t ksigmask, sigsaved;
1324	struct timespec64 ts, end_time, *to = NULL;
1325	int ret;
1326
1327	if (tsp) {
1328		if (compat_get_timespec64(&ts, tsp))
1329			return -EFAULT;
1330
1331		to = &end_time;
1332		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1333			return -EINVAL;
1334	}
1335
1336	if (sigmask) {
1337		if (sigsetsize != sizeof(compat_sigset_t))
1338			return -EINVAL;
1339		if (get_compat_sigset(&ksigmask, sigmask))
1340			return -EFAULT;
1341
1342		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1343		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1344	}
1345
1346	ret = compat_core_sys_select(n, inp, outp, exp, to);
1347	ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret);
1348
1349	if (ret == -ERESTARTNOHAND) {
1350		/*
1351		 * Don't restore the signal mask yet. Let do_signal() deliver
1352		 * the signal on the way back to userspace, before the signal
1353		 * mask is restored.
1354		 */
1355		if (sigmask) {
1356			memcpy(&current->saved_sigmask, &sigsaved,
1357					sizeof(sigsaved));
1358			set_restore_sigmask();
1359		}
1360	} else if (sigmask)
1361		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1362
1363	return ret;
1364}
1365
1366COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp,
1367	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1368	struct compat_timespec __user *, tsp, void __user *, sig)
1369{
1370	compat_size_t sigsetsize = 0;
1371	compat_uptr_t up = 0;
1372
1373	if (sig) {
1374		if (!access_ok(VERIFY_READ, sig,
1375				sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1376		    	__get_user(up, (compat_uptr_t __user *)sig) ||
1377		    	__get_user(sigsetsize,
1378				(compat_size_t __user *)(sig+sizeof(up))))
1379			return -EFAULT;
1380	}
1381	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1382				 sigsetsize);
1383}
1384
1385COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds,
1386	unsigned int,  nfds, struct compat_timespec __user *, tsp,
1387	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1388{
1389	sigset_t ksigmask, sigsaved;
1390	struct timespec64 ts, end_time, *to = NULL;
1391	int ret;
1392
1393	if (tsp) {
1394		if (compat_get_timespec64(&ts, tsp))
1395			return -EFAULT;
1396
1397		to = &end_time;
1398		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1399			return -EINVAL;
1400	}
1401
1402	if (sigmask) {
1403		if (sigsetsize != sizeof(compat_sigset_t))
1404			return -EINVAL;
1405		if (get_compat_sigset(&ksigmask, sigmask))
1406			return -EFAULT;
1407
1408		sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1409		sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1410	}
1411
1412	ret = do_sys_poll(ufds, nfds, to);
1413
1414	/* We can restart this syscall, usually */
1415	if (ret == -EINTR) {
1416		/*
1417		 * Don't restore the signal mask yet. Let do_signal() deliver
1418		 * the signal on the way back to userspace, before the signal
1419		 * mask is restored.
1420		 */
1421		if (sigmask) {
1422			memcpy(&current->saved_sigmask, &sigsaved,
1423				sizeof(sigsaved));
1424			set_restore_sigmask();
1425		}
1426		ret = -ERESTARTNOHAND;
1427	} else if (sigmask)
1428		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1429
1430	ret = compat_poll_select_copy_remaining(&end_time, tsp, 0, ret);
1431
1432	return ret;
1433}
1434#endif