blob: 0ee55af1a55c29b14b118907b98e0329b8792376 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
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
Jakub Kicinskib6459412021-12-28 16:49:13 -080018#include <linux/compat.h>
Milind Arun Choudhary022a1692007-05-08 00:29:02 -070019#include <linux/kernel.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010020#include <linux/sched/signal.h>
21#include <linux/sched/rt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/syscalls.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050023#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/poll.h>
26#include <linux/personality.h> /* for STICKY_TIMEOUTS */
27#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040028#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/fs.h>
Dipankar Sarmab8359962005-09-09 13:04:14 -070030#include <linux/rcupdate.h>
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -070031#include <linux/hrtimer.h>
Colin Cross9745cdb2013-05-06 23:50:17 +000032#include <linux/freezer.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +030033#include <net/busy_poll.h>
Vlastimil Babka2d193092016-10-11 13:51:14 -070034#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080036#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Arjan van de Ven90d6e242008-09-01 15:55:35 -070038
39/*
40 * Estimate expected accuracy in ns from a timeval.
41 *
42 * After quite a bit of churning around, we've settled on
43 * a simple thing of taking 0.1% of the timeout as the
44 * slack, with a cap of 100 msec.
45 * "nice" tasks get a 0.5% slack instead.
46 *
47 * Consider this comment an open invitation to come up with even
48 * better solutions..
49 */
50
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070051#define MAX_SLACK (100 * NSEC_PER_MSEC)
52
Deepa Dinamani766b9f92016-05-19 17:09:05 -070053static long __estimate_accuracy(struct timespec64 *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070054{
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070055 long slack;
Arjan van de Ven90d6e242008-09-01 15:55:35 -070056 int divfactor = 1000;
57
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070058 if (tv->tv_sec < 0)
59 return 0;
60
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070061 if (task_nice(current) > 0)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070062 divfactor = divfactor / 5;
63
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070064 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
65 return MAX_SLACK;
66
Arjan van de Ven90d6e242008-09-01 15:55:35 -070067 slack = tv->tv_nsec / divfactor;
68 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
69
Guillaume Knispel5ae87e72009-09-22 16:43:30 -070070 if (slack > MAX_SLACK)
71 return MAX_SLACK;
Arjan van de Ven96d2ab42008-09-07 16:08:55 -070072
Arjan van de Ven90d6e242008-09-01 15:55:35 -070073 return slack;
74}
75
Deepa Dinamani766b9f92016-05-19 17:09:05 -070076u64 select_estimate_accuracy(struct timespec64 *tv)
Arjan van de Ven90d6e242008-09-01 15:55:35 -070077{
John Stultzda8b44d2016-03-17 14:20:51 -070078 u64 ret;
Deepa Dinamani766b9f92016-05-19 17:09:05 -070079 struct timespec64 now;
Arjan van de Ven90d6e242008-09-01 15:55:35 -070080
81 /*
82 * Realtime tasks get a slack of 0 for obvious reasons.
83 */
84
Arjan van de Ven4ce105d2008-09-07 15:31:39 -070085 if (rt_task(current))
Arjan van de Ven90d6e242008-09-01 15:55:35 -070086 return 0;
87
Deepa Dinamani766b9f92016-05-19 17:09:05 -070088 ktime_get_ts64(&now);
89 now = timespec64_sub(*tv, now);
Arjan van de Ven90d6e242008-09-01 15:55:35 -070090 ret = __estimate_accuracy(&now);
91 if (ret < current->timer_slack_ns)
92 return current->timer_slack_ns;
93 return ret;
94}
95
96
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098struct poll_table_page {
99 struct poll_table_page * next;
100 struct poll_table_entry * entry;
Gustavo A. R. Silva5e01fdf2020-08-31 08:25:42 -0500101 struct poll_table_entry entries[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102};
103
104#define POLL_TABLE_FULL(table) \
105 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
106
107/*
108 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
109 * I have rewritten this, taking some shortcuts: This code may not be easy to
110 * follow, but it should be free of race-conditions, and it's practical. If you
111 * understand what I'm doing here, then you understand how the linux
112 * sleep/wakeup mechanism works.
113 *
114 * Two very simple procedures, poll_wait() and poll_freewait() make all the
115 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
116 * as all select/poll functions have to call it to add an entry to the
117 * poll table.
118 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700119static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
120 poll_table *p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
122void poll_initwait(struct poll_wqueues *pwq)
123{
124 init_poll_funcptr(&pwq->pt, __pollwait);
Tejun Heo5f820f62009-01-06 14:40:59 -0800125 pwq->polling_task = current;
Guillaume Knispelb2add732009-08-15 19:30:24 +0200126 pwq->triggered = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 pwq->error = 0;
128 pwq->table = NULL;
Andi Kleen70674f92006-03-28 01:56:33 -0800129 pwq->inline_index = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131EXPORT_SYMBOL(poll_initwait);
132
Andi Kleen70674f92006-03-28 01:56:33 -0800133static void free_poll_entry(struct poll_table_entry *entry)
134{
WANG Congccf67802007-05-09 07:10:02 +0200135 remove_wait_queue(entry->wait_address, &entry->wait);
Andi Kleen70674f92006-03-28 01:56:33 -0800136 fput(entry->filp);
137}
138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139void poll_freewait(struct poll_wqueues *pwq)
140{
141 struct poll_table_page * p = pwq->table;
Andi Kleen70674f92006-03-28 01:56:33 -0800142 int i;
143 for (i = 0; i < pwq->inline_index; i++)
144 free_poll_entry(pwq->inline_entries + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 while (p) {
146 struct poll_table_entry * entry;
147 struct poll_table_page *old;
148
149 entry = p->entry;
150 do {
151 entry--;
Andi Kleen70674f92006-03-28 01:56:33 -0800152 free_poll_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 } while (entry > p->entries);
154 old = p;
155 p = p->next;
156 free_page((unsigned long) old);
157 }
158}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159EXPORT_SYMBOL(poll_freewait);
160
Tejun Heo5f820f62009-01-06 14:40:59 -0800161static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 struct poll_table_page *table = p->table;
164
Andi Kleen70674f92006-03-28 01:56:33 -0800165 if (p->inline_index < N_INLINE_POLL_ENTRIES)
166 return p->inline_entries + p->inline_index++;
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 if (!table || POLL_TABLE_FULL(table)) {
169 struct poll_table_page *new_table;
170
171 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
172 if (!new_table) {
173 p->error = -ENOMEM;
Andi Kleen70674f92006-03-28 01:56:33 -0800174 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 }
176 new_table->entry = new_table->entries;
177 new_table->next = table;
178 p->table = new_table;
179 table = new_table;
180 }
181
Andi Kleen70674f92006-03-28 01:56:33 -0800182 return table->entry++;
183}
184
Ingo Molnarac6424b2017-06-20 12:06:13 +0200185static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Tejun Heo5f820f62009-01-06 14:40:59 -0800186{
187 struct poll_wqueues *pwq = wait->private;
188 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
189
190 /*
191 * Although this function is called under waitqueue lock, LOCK
192 * doesn't imply write barrier and the users expect write
193 * barrier semantics on wakeup functions. The following
194 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
Peter Zijlstrab92b8b32015-05-12 10:51:55 +0200195 * and is paired with smp_store_mb() in poll_schedule_timeout.
Tejun Heo5f820f62009-01-06 14:40:59 -0800196 */
197 smp_wmb();
198 pwq->triggered = 1;
199
200 /*
201 * Perform the default wake up operation using a dummy
202 * waitqueue.
203 *
204 * TODO: This is hacky but there currently is no interface to
205 * pass in @sync. @sync is scheduled to be removed and once
206 * that happens, wake_up_process() can be used directly.
207 */
208 return default_wake_function(&dummy_wait, mode, sync, key);
209}
210
Ingo Molnarac6424b2017-06-20 12:06:13 +0200211static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700212{
213 struct poll_table_entry *entry;
214
215 entry = container_of(wait, struct poll_table_entry, wait);
Al Viro3ad6f932017-07-03 20:14:56 -0400216 if (key && !(key_to_poll(key) & entry->key))
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700217 return 0;
218 return __pollwake(wait, mode, sync, key);
219}
220
Andi Kleen70674f92006-03-28 01:56:33 -0800221/* Add a new entry */
222static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
223 poll_table *p)
224{
Tejun Heo5f820f62009-01-06 14:40:59 -0800225 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
226 struct poll_table_entry *entry = poll_get_entry(pwq);
Andi Kleen70674f92006-03-28 01:56:33 -0800227 if (!entry)
228 return;
Al Virocb0942b2012-08-27 14:48:26 -0400229 entry->filp = get_file(filp);
Andi Kleen70674f92006-03-28 01:56:33 -0800230 entry->wait_address = wait_address;
Hans Verkuil626cf232012-03-23 15:02:27 -0700231 entry->key = p->_key;
Tejun Heo5f820f62009-01-06 14:40:59 -0800232 init_waitqueue_func_entry(&entry->wait, pollwake);
233 entry->wait.private = pwq;
WANG Congccf67802007-05-09 07:10:02 +0200234 add_wait_queue(wait_address, &entry->wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
Christoph Hellwig8f546ae2018-01-11 12:23:05 +0100237static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
Tejun Heo5f820f62009-01-06 14:40:59 -0800238 ktime_t *expires, unsigned long slack)
239{
240 int rc = -EINTR;
241
242 set_current_state(state);
243 if (!pwq->triggered)
Rafael J. Wysocki59612d12013-10-29 23:43:08 +0100244 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
Tejun Heo5f820f62009-01-06 14:40:59 -0800245 __set_current_state(TASK_RUNNING);
246
247 /*
248 * Prepare for the next iteration.
249 *
Peter Zijlstrab92b8b32015-05-12 10:51:55 +0200250 * The following smp_store_mb() serves two purposes. First, it's
Tejun Heo5f820f62009-01-06 14:40:59 -0800251 * the counterpart rmb of the wmb in pollwake() such that data
252 * written before wake up is always visible after wake up.
253 * Second, the full barrier guarantees that triggered clearing
254 * doesn't pass event check of the next iteration. Note that
255 * this problem doesn't exist for the first iteration as
256 * add_wait_queue() has full barrier semantics.
257 */
Peter Zijlstrab92b8b32015-05-12 10:51:55 +0200258 smp_store_mb(pwq->triggered, 0);
Tejun Heo5f820f62009-01-06 14:40:59 -0800259
260 return rc;
261}
Tejun Heo5f820f62009-01-06 14:40:59 -0800262
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700263/**
264 * poll_select_set_timeout - helper function to setup the timeout value
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700265 * @to: pointer to timespec64 variable for the final timeout
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700266 * @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 */
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700274int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700275{
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700276 struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700277
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700278 if (!timespec64_valid(&ts))
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700279 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 {
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700285 ktime_get_ts64(to);
286 *to = timespec64_add_safe(*to, ts);
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700287 }
288 return 0;
289}
290
Deepa Dinamani8bd27a32018-09-19 21:41:06 -0700291enum poll_time_type {
292 PT_TIMEVAL = 0,
293 PT_OLD_TIMEVAL = 1,
294 PT_TIMESPEC = 2,
295 PT_OLD_TIMESPEC = 3,
296};
297
Oleg Nesterovac301022019-07-16 16:29:59 -0700298static int poll_select_finish(struct timespec64 *end_time,
299 void __user *p,
300 enum poll_time_type pt_type, int ret)
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700301{
Deepa Dinamani36819ad2017-08-04 21:12:31 -0700302 struct timespec64 rts;
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700303
Oleg Nesterovac301022019-07-16 16:29:59 -0700304 restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
305
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700306 if (!p)
307 return ret;
308
309 if (current->personality & STICKY_TIMEOUTS)
310 goto sticky;
311
312 /* No update for zero timeout */
313 if (!end_time->tv_sec && !end_time->tv_nsec)
314 return ret;
315
Deepa Dinamani36819ad2017-08-04 21:12:31 -0700316 ktime_get_ts64(&rts);
317 rts = timespec64_sub(*end_time, rts);
318 if (rts.tv_sec < 0)
319 rts.tv_sec = rts.tv_nsec = 0;
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700320
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700321
Deepa Dinamani8bd27a32018-09-19 21:41:06 -0700322 switch (pt_type) {
323 case PT_TIMEVAL:
324 {
Arnd Bergmann75d319c2019-10-25 22:56:17 +0200325 struct __kernel_old_timeval rtv;
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700326
Deepa Dinamani8bd27a32018-09-19 21:41:06 -0700327 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
328 memset(&rtv, 0, sizeof(rtv));
329 rtv.tv_sec = rts.tv_sec;
330 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
331 if (!copy_to_user(p, &rtv, sizeof(rtv)))
332 return ret;
333 }
334 break;
335 case PT_OLD_TIMEVAL:
336 {
337 struct old_timeval32 rtv;
338
339 rtv.tv_sec = rts.tv_sec;
340 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
341 if (!copy_to_user(p, &rtv, sizeof(rtv)))
342 return ret;
343 }
344 break;
345 case PT_TIMESPEC:
346 if (!put_timespec64(&rts, p))
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700347 return ret;
Deepa Dinamani8bd27a32018-09-19 21:41:06 -0700348 break;
349 case PT_OLD_TIMESPEC:
350 if (!put_old_timespec32(&rts, p))
351 return ret;
352 break;
353 default:
354 BUG();
355 }
Thomas Gleixnerb773ad42008-08-31 08:16:57 -0700356 /*
357 * If an application puts its timeval in read-only memory, we
358 * don't want the Linux-specific update to the timeval to
359 * cause a fault after the select has completed
360 * successfully. However, because we're not updating the
361 * timeval, we can't restart the system call.
362 */
363
364sticky:
365 if (ret == -ERESTARTNOHAND)
366 ret = -EINTR;
367 return ret;
368}
369
Al Viroe99ca562017-04-08 16:50:24 -0400370/*
371 * Scalable version of the fd_set.
372 */
373
374typedef struct {
375 unsigned long *in, *out, *ex;
376 unsigned long *res_in, *res_out, *res_ex;
377} fd_set_bits;
378
379/*
380 * How many longwords for "nr" bits?
381 */
382#define FDS_BITPERLONG (8*sizeof(long))
383#define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
384#define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
385
386/*
Al Viroe99ca562017-04-08 16:50:24 -0400387 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
388 */
389static inline
390int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
391{
392 nr = FDS_BYTES(nr);
393 if (ufdset)
394 return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
395
396 memset(fdset, 0, nr);
397 return 0;
398}
399
400static inline unsigned long __must_check
401set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
402{
403 if (ufdset)
404 return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
405 return 0;
406}
407
408static inline
409void zero_fd_set(unsigned long nr, unsigned long *fdset)
410{
411 memset(fdset, 0, FDS_BYTES(nr));
412}
413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414#define FDS_IN(fds, n) (fds->in + n)
415#define FDS_OUT(fds, n) (fds->out + n)
416#define FDS_EX(fds, n) (fds->ex + n)
417
418#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
419
420static int max_select_fd(unsigned long n, fd_set_bits *fds)
421{
422 unsigned long *open_fds;
423 unsigned long set;
424 int max;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700425 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
427 /* handle last in-complete long-word first */
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400428 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
429 n /= BITS_PER_LONG;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700430 fdt = files_fdtable(current->files);
David Howells1fd36ad2012-02-16 17:49:54 +0000431 open_fds = fdt->open_fds + n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 max = 0;
433 if (set) {
434 set &= BITS(fds, n);
435 if (set) {
436 if (!(set & ~*open_fds))
437 goto get_max;
438 return -EBADF;
439 }
440 }
441 while (n) {
442 open_fds--;
443 n--;
444 set = BITS(fds, n);
445 if (!set)
446 continue;
447 if (set & ~*open_fds)
448 return -EBADF;
449 if (max)
450 continue;
451get_max:
452 do {
453 max++;
454 set >>= 1;
455 } while (set);
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400456 max += n * BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
458
459 return max;
460}
461
Jan Kara68514da2022-01-10 19:19:23 +0100462#define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR |\
463 EPOLLNVAL)
464#define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\
465 EPOLLNVAL)
466#define POLLEX_SET (EPOLLPRI | EPOLLNVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700468static inline void wait_key_set(poll_table *wait, unsigned long in,
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300469 unsigned long out, unsigned long bit,
Al Viro01699432017-07-03 03:14:15 -0400470 __poll_t ll_flag)
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700471{
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300472 wait->_key = POLLEX_SET | ll_flag;
Hans Verkuil626cf232012-03-23 15:02:27 -0700473 if (in & bit)
474 wait->_key |= POLLIN_SET;
475 if (out & bit)
476 wait->_key |= POLLOUT_SET;
Eric Dumazet4938d7e2009-06-16 15:33:36 -0700477}
478
Al Viroe99ca562017-04-08 16:50:24 -0400479static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700481 ktime_t expire, *to = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 struct poll_wqueues table;
483 poll_table *wait;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700484 int retval, i, timed_out = 0;
John Stultzda8b44d2016-03-17 14:20:51 -0700485 u64 slack = 0;
Al Viro01699432017-07-03 03:14:15 -0400486 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
Alexander Duyck37056712017-03-24 10:08:18 -0700487 unsigned long busy_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Dipankar Sarmab8359962005-09-09 13:04:14 -0700489 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 retval = max_select_fd(n, fds);
Dipankar Sarmab8359962005-09-09 13:04:14 -0700491 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493 if (retval < 0)
494 return retval;
495 n = retval;
496
497 poll_initwait(&table);
498 wait = &table.pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700499 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700500 wait->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700501 timed_out = 1;
502 }
503
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700504 if (end_time && !timed_out)
Andrew Morton231f3d32010-10-27 15:34:53 -0700505 slack = select_estimate_accuracy(end_time);
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 retval = 0;
508 for (;;) {
509 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300510 bool can_busy_loop = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 inp = fds->in; outp = fds->out; exp = fds->ex;
513 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
514
515 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
Al Viroe6c8adc2017-07-03 22:25:56 -0400516 unsigned long in, out, ex, all_bits, bit = 1, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 unsigned long res_in = 0, res_out = 0, res_ex = 0;
Al Viroe6c8adc2017-07-03 22:25:56 -0400518 __poll_t mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
520 in = *inp++; out = *outp++; ex = *exp++;
521 all_bits = in | out | ex;
522 if (all_bits == 0) {
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400523 i += BITS_PER_LONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 continue;
525 }
526
Josh Boyer8ded2bb2012-07-25 10:40:34 -0400527 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
Al Viro2903ff02012-08-28 12:52:22 -0400528 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 if (i >= n)
530 break;
531 if (!(bit & all_bits))
532 continue;
Jan Kara68514da2022-01-10 19:19:23 +0100533 mask = EPOLLNVAL;
Al Viro2903ff02012-08-28 12:52:22 -0400534 f = fdget(i);
535 if (f.file) {
Christoph Hellwig9965ed172018-03-05 07:26:05 -0800536 wait_key_set(wait, in, out, bit,
537 busy_flag);
538 mask = vfs_poll(f.file, wait);
539
Al Viro2903ff02012-08-28 12:52:22 -0400540 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
Jan Kara68514da2022-01-10 19:19:23 +0100542 if ((mask & POLLIN_SET) && (in & bit)) {
543 res_in |= bit;
544 retval++;
545 wait->_qproc = NULL;
546 }
547 if ((mask & POLLOUT_SET) && (out & bit)) {
548 res_out |= bit;
549 retval++;
550 wait->_qproc = NULL;
551 }
552 if ((mask & POLLEX_SET) && (ex & bit)) {
553 res_ex |= bit;
554 retval++;
555 wait->_qproc = NULL;
556 }
557 /* got something, stop busy polling */
558 if (retval) {
559 can_busy_loop = false;
560 busy_flag = 0;
561
562 /*
563 * only remember a returned
564 * POLL_BUSY_LOOP if we asked for it
565 */
566 } else if (busy_flag & mask)
567 can_busy_loop = true;
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
570 if (res_in)
571 *rinp = res_in;
572 if (res_out)
573 *routp = res_out;
574 if (res_ex)
575 *rexp = res_ex;
Linus Torvalds55d85382008-06-22 12:23:15 -0700576 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 }
Hans Verkuil626cf232012-03-23 15:02:27 -0700578 wait->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700579 if (retval || timed_out || signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 break;
Pavel Machekf5264482008-04-21 22:15:06 +0000581 if (table.error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 retval = table.error;
583 break;
584 }
David Woodhouse9f729492006-01-18 17:44:05 -0800585
Eliezer Tamircbf55002013-07-08 16:20:34 +0300586 /* only if found POLL_BUSY_LOOP sockets && not out of time */
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300587 if (can_busy_loop && !need_resched()) {
Alexander Duyck37056712017-03-24 10:08:18 -0700588 if (!busy_start) {
589 busy_start = busy_loop_current_time();
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300590 continue;
591 }
Alexander Duyck37056712017-03-24 10:08:18 -0700592 if (!busy_loop_timeout(busy_start))
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300593 continue;
594 }
595 busy_flag = 0;
Eliezer Tamir2d48d672013-06-24 10:28:03 +0300596
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700597 /*
598 * If this is the first loop and we have a timeout
599 * given, then we convert to ktime_t and set the to
600 * pointer to the expiry value.
601 */
602 if (end_time && !to) {
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700603 expire = timespec64_to_ktime(*end_time);
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700604 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800605 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700606
Tejun Heo5f820f62009-01-06 14:40:59 -0800607 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
608 to, slack))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700609 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 poll_freewait(&table);
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 return retval;
615}
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617/*
618 * We can actually return ERESTARTSYS instead of EINTR, but I'd
619 * like to be certain this leads to no problems. So I return
620 * EINTR just for safety.
621 *
622 * Update: ERESTARTSYS breaks at least the xview clock binary, so
623 * I'm trying ERESTARTNOHAND which restart only when you want to.
624 */
Al Viroa2dcb442008-04-23 14:05:15 -0400625int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700626 fd_set __user *exp, struct timespec64 *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
628 fd_set_bits fds;
Andrew Morton29ff2db2006-04-10 22:52:46 -0700629 void *bits;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800630 int ret, max_fds;
Vlastimil Babka2d193092016-10-11 13:51:14 -0700631 size_t size, alloc_size;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700632 struct fdtable *fdt;
Andi Kleen70674f92006-03-28 01:56:33 -0800633 /* Allocate small arguments on the stack to save memory and be faster */
Jes Sorensen30c14e42006-03-31 11:18:57 -0500634 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 ret = -EINVAL;
637 if (n < 0)
638 goto out_nofds;
639
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800640 /* max_fds can increase, so grab it once to avoid race */
Dipankar Sarmab8359962005-09-09 13:04:14 -0700641 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700642 fdt = files_fdtable(current->files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800643 max_fds = fdt->max_fds;
Dipankar Sarmab8359962005-09-09 13:04:14 -0700644 rcu_read_unlock();
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800645 if (n > max_fds)
646 n = max_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
648 /*
649 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
650 * since we used fdset we need to allocate memory in units of
651 * long-words.
652 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 size = FDS_BYTES(n);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700654 bits = stack_fds;
655 if (size > sizeof(stack_fds) / 6) {
656 /* Not enough space in on-stack array; must use kmalloc */
657 ret = -ENOMEM;
Vlastimil Babka2d193092016-10-11 13:51:14 -0700658 if (size > (SIZE_MAX / 6))
659 goto out_nofds;
660
661 alloc_size = 6 * size;
Linus Torvalds0bcfe682021-09-07 11:26:23 -0700662 bits = kvmalloc(alloc_size, GFP_KERNEL);
Mitchell Blank Jrb04eb6a2006-04-10 22:54:08 -0700663 if (!bits)
664 goto out_nofds;
665 }
Andrew Morton29ff2db2006-04-10 22:52:46 -0700666 fds.in = bits;
667 fds.out = bits + size;
668 fds.ex = bits + 2*size;
669 fds.res_in = bits + 3*size;
670 fds.res_out = bits + 4*size;
671 fds.res_ex = bits + 5*size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673 if ((ret = get_fd_set(n, inp, fds.in)) ||
674 (ret = get_fd_set(n, outp, fds.out)) ||
675 (ret = get_fd_set(n, exp, fds.ex)))
676 goto out;
677 zero_fd_set(n, fds.res_in);
678 zero_fd_set(n, fds.res_out);
679 zero_fd_set(n, fds.res_ex);
680
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700681 ret = do_select(n, &fds, end_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
683 if (ret < 0)
684 goto out;
685 if (!ret) {
686 ret = -ERESTARTNOHAND;
687 if (signal_pending(current))
688 goto out;
689 ret = 0;
690 }
691
692 if (set_fd_set(n, inp, fds.res_in) ||
693 set_fd_set(n, outp, fds.res_out) ||
694 set_fd_set(n, exp, fds.res_ex))
695 ret = -EFAULT;
696
697out:
Andi Kleen70674f92006-03-28 01:56:33 -0800698 if (bits != stack_fds)
Vlastimil Babka2d193092016-10-11 13:51:14 -0700699 kvfree(bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700out_nofds:
701 return ret;
702}
703
Dominik Brodowski4bdb9ac2018-03-18 07:53:04 +0100704static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
Arnd Bergmann75d319c2019-10-25 22:56:17 +0200705 fd_set __user *exp, struct __kernel_old_timeval __user *tvp)
David Woodhouse9f729492006-01-18 17:44:05 -0800706{
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700707 struct timespec64 end_time, *to = NULL;
Arnd Bergmann75d319c2019-10-25 22:56:17 +0200708 struct __kernel_old_timeval tv;
David Woodhouse9f729492006-01-18 17:44:05 -0800709 int ret;
710
711 if (tvp) {
712 if (copy_from_user(&tv, tvp, sizeof(tv)))
713 return -EFAULT;
714
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700715 to = &end_time;
Arjan van de Ven4d36a9e2008-10-25 12:41:41 -0700716 if (poll_select_set_timeout(to,
717 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
718 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
David Woodhouse9f729492006-01-18 17:44:05 -0800719 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800720 }
721
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700722 ret = core_sys_select(n, inp, outp, exp, to);
Oleg Nesterovac301022019-07-16 16:29:59 -0700723 return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800724}
725
Dominik Brodowski4bdb9ac2018-03-18 07:53:04 +0100726SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
Arnd Bergmann75d319c2019-10-25 22:56:17 +0200727 fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp)
Dominik Brodowski4bdb9ac2018-03-18 07:53:04 +0100728{
729 return kern_select(n, inp, outp, exp, tvp);
730}
731
Heiko Carstensc9da9f22009-01-14 14:13:57 +0100732static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
Deepa Dinamanie0247072018-09-19 21:41:07 -0700733 fd_set __user *exp, void __user *tsp,
734 const sigset_t __user *sigmask, size_t sigsetsize,
735 enum poll_time_type type)
David Woodhouse9f729492006-01-18 17:44:05 -0800736{
Deepa Dinamani36819ad2017-08-04 21:12:31 -0700737 struct timespec64 ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -0800738 int ret;
739
740 if (tsp) {
Deepa Dinamanie0247072018-09-19 21:41:07 -0700741 switch (type) {
742 case PT_TIMESPEC:
743 if (get_timespec64(&ts, tsp))
744 return -EFAULT;
745 break;
746 case PT_OLD_TIMESPEC:
747 if (get_old_timespec32(&ts, tsp))
748 return -EFAULT;
749 break;
750 default:
751 BUG();
752 }
David Woodhouse9f729492006-01-18 17:44:05 -0800753
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700754 to = &end_time;
Deepa Dinamani36819ad2017-08-04 21:12:31 -0700755 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
David Woodhouse9f729492006-01-18 17:44:05 -0800756 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -0800757 }
758
Oleg Nesterovb7724342019-07-16 16:29:53 -0700759 ret = set_user_sigmask(sigmask, sigsetsize);
Deepa Dinamanided653c2018-09-19 21:41:04 -0700760 if (ret)
761 return ret;
David Woodhouse9f729492006-01-18 17:44:05 -0800762
Bernd Schmidt62568512009-01-13 22:14:48 +0100763 ret = core_sys_select(n, inp, outp, exp, to);
Oleg Nesterovac301022019-07-16 16:29:59 -0700764 return poll_select_finish(&end_time, tsp, type, ret);
David Woodhouse9f729492006-01-18 17:44:05 -0800765}
766
767/*
768 * Most architectures can't handle 7-argument syscalls. So we provide a
769 * 6-argument version where the sixth argument is a pointer to a structure
770 * which has a pointer to the sigset_t itself followed by a size_t containing
771 * the sigset size.
772 */
Al Viro7e716092020-02-19 09:54:24 -0500773struct sigset_argpack {
774 sigset_t __user *p;
775 size_t size;
776};
777
778static inline int get_sigset_argpack(struct sigset_argpack *to,
779 struct sigset_argpack __user *from)
780{
781 // the path is hot enough for overhead of copy_from_user() to matter
782 if (from) {
783 if (!user_read_access_begin(from, sizeof(*from)))
784 return -EFAULT;
785 unsafe_get_user(to->p, &from->p, Efault);
786 unsafe_get_user(to->size, &from->size, Efault);
787 user_read_access_end();
788 }
789 return 0;
790Efault:
791 user_access_end();
792 return -EFAULT;
793}
794
Heiko Carstensd4e82042009-01-14 14:14:34 +0100795SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
Deepa Dinamanie0247072018-09-19 21:41:07 -0700796 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
Heiko Carstensd4e82042009-01-14 14:14:34 +0100797 void __user *, sig)
David Woodhouse9f729492006-01-18 17:44:05 -0800798{
Al Viro7e716092020-02-19 09:54:24 -0500799 struct sigset_argpack x = {NULL, 0};
David Woodhouse9f729492006-01-18 17:44:05 -0800800
Al Viro7e716092020-02-19 09:54:24 -0500801 if (get_sigset_argpack(&x, sig))
802 return -EFAULT;
David Woodhouse9f729492006-01-18 17:44:05 -0800803
Al Viro7e716092020-02-19 09:54:24 -0500804 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
David Woodhouse9f729492006-01-18 17:44:05 -0800805}
David Woodhouse9f729492006-01-18 17:44:05 -0800806
Deepa Dinamanie0247072018-09-19 21:41:07 -0700807#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
808
809SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
810 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
811 void __user *, sig)
812{
Al Viro7e716092020-02-19 09:54:24 -0500813 struct sigset_argpack x = {NULL, 0};
Deepa Dinamanie0247072018-09-19 21:41:07 -0700814
Al Viro7e716092020-02-19 09:54:24 -0500815 if (get_sigset_argpack(&x, sig))
816 return -EFAULT;
Deepa Dinamanie0247072018-09-19 21:41:07 -0700817
Al Viro7e716092020-02-19 09:54:24 -0500818 return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
Deepa Dinamanie0247072018-09-19 21:41:07 -0700819}
820
821#endif
822
Christoph Hellwig5d0e5282010-03-10 15:21:13 -0800823#ifdef __ARCH_WANT_SYS_OLD_SELECT
824struct sel_arg_struct {
825 unsigned long n;
826 fd_set __user *inp, *outp, *exp;
Arnd Bergmann75d319c2019-10-25 22:56:17 +0200827 struct __kernel_old_timeval __user *tvp;
Christoph Hellwig5d0e5282010-03-10 15:21:13 -0800828};
829
830SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
831{
832 struct sel_arg_struct a;
833
834 if (copy_from_user(&a, arg, sizeof(a)))
835 return -EFAULT;
Dominik Brodowski4bdb9ac2018-03-18 07:53:04 +0100836 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
Christoph Hellwig5d0e5282010-03-10 15:21:13 -0800837}
838#endif
839
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840struct poll_list {
841 struct poll_list *next;
842 int len;
Gustavo A. R. Silva5e01fdf2020-08-31 08:25:42 -0500843 struct pollfd entries[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844};
845
846#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
847
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700848/*
849 * Fish for pollable events on the pollfd->fd file descriptor. We're only
850 * interested in events matching the pollfd->events mask, and the result
851 * matching that mask is both recorded in pollfd->revents and returned. The
852 * pwait poll_table will be used by the fd-provided poll handler for waiting,
Hans Verkuil626cf232012-03-23 15:02:27 -0700853 * if pwait->_qproc is non-NULL.
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700854 */
Al Virofb367932017-07-16 23:51:03 -0400855static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
Eliezer Tamircbf55002013-07-08 16:20:34 +0300856 bool *can_busy_poll,
Al Virofb367932017-07-16 23:51:03 -0400857 __poll_t busy_flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
Christoph Hellwiga0f8dcf2018-03-05 07:15:25 -0800859 int fd = pollfd->fd;
860 __poll_t mask = 0, filter;
861 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
Christoph Hellwiga0f8dcf2018-03-05 07:15:25 -0800863 if (fd < 0)
864 goto out;
865 mask = EPOLLNVAL;
866 f = fdget(fd);
867 if (!f.file)
868 goto out;
869
870 /* userland u16 ->events contains POLL... bitmap */
871 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
Christoph Hellwig9965ed172018-03-05 07:26:05 -0800872 pwait->_key = filter | busy_flag;
873 mask = vfs_poll(f.file, pwait);
874 if (mask & busy_flag)
875 *can_busy_poll = true;
Christoph Hellwiga0f8dcf2018-03-05 07:15:25 -0800876 mask &= filter; /* Mask out unneeded events. */
877 fdput(f);
878
879out:
Al Virofb367932017-07-16 23:51:03 -0400880 /* ... and so does ->revents */
Al Viroc71d2272017-11-29 19:00:41 -0500881 pollfd->revents = mangle_poll(mask);
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700882 return mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883}
884
Mateusz Guzikccec5ee2016-01-06 06:41:53 +0100885static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700886 struct timespec64 *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 poll_table* pt = &wait->pt;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700889 ktime_t expire, *to = NULL;
890 int timed_out = 0, count = 0;
John Stultzda8b44d2016-03-17 14:20:51 -0700891 u64 slack = 0;
Al Virofb367932017-07-16 23:51:03 -0400892 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
Alexander Duyck37056712017-03-24 10:08:18 -0700893 unsigned long busy_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
David Woodhouse9f729492006-01-18 17:44:05 -0800895 /* Optimise the no-wait case */
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700896 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
Hans Verkuil626cf232012-03-23 15:02:27 -0700897 pt->_qproc = NULL;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700898 timed_out = 1;
899 }
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700900
Arjan van de Ven96d2ab42008-09-07 16:08:55 -0700901 if (end_time && !timed_out)
Andrew Morton231f3d32010-10-27 15:34:53 -0700902 slack = select_estimate_accuracy(end_time);
Arjan van de Ven90d6e242008-09-01 15:55:35 -0700903
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 for (;;) {
905 struct poll_list *walk;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300906 bool can_busy_loop = false;
David Woodhouse9f729492006-01-18 17:44:05 -0800907
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700908 for (walk = list; walk != NULL; walk = walk->next) {
909 struct pollfd * pfd, * pfd_end;
910
911 pfd = walk->entries;
912 pfd_end = pfd + walk->len;
913 for (; pfd != pfd_end; pfd++) {
914 /*
915 * Fish for events. If we found one, record it
Hans Verkuil626cf232012-03-23 15:02:27 -0700916 * and kill poll_table->_qproc, so we don't
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700917 * needlessly register any other waiters after
918 * this. They'll get immediately deregistered
919 * when we break out and return.
920 */
Eliezer Tamircbf55002013-07-08 16:20:34 +0300921 if (do_pollfd(pfd, pt, &can_busy_loop,
922 busy_flag)) {
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700923 count++;
Hans Verkuil626cf232012-03-23 15:02:27 -0700924 pt->_qproc = NULL;
Eliezer Tamircbf55002013-07-08 16:20:34 +0300925 /* found something, stop busy polling */
926 busy_flag = 0;
927 can_busy_loop = false;
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700928 }
929 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 }
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700931 /*
932 * All waiters have already been registered, so don't provide
Hans Verkuil626cf232012-03-23 15:02:27 -0700933 * a poll_table->_qproc to them on the next loop iteration.
Vadim Lobanov4a4b69f2006-06-23 02:05:16 -0700934 */
Hans Verkuil626cf232012-03-23 15:02:27 -0700935 pt->_qproc = NULL;
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700936 if (!count) {
937 count = wait->error;
938 if (signal_pending(current))
Oleg Nesterov8cf8b552019-07-16 16:29:56 -0700939 count = -ERESTARTNOHAND;
Oleg Nesterov9bf084f2007-10-16 23:26:18 -0700940 }
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700941 if (count || timed_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 break;
David Woodhouse9f729492006-01-18 17:44:05 -0800943
Eliezer Tamircbf55002013-07-08 16:20:34 +0300944 /* only if found POLL_BUSY_LOOP sockets && not out of time */
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300945 if (can_busy_loop && !need_resched()) {
Alexander Duyck37056712017-03-24 10:08:18 -0700946 if (!busy_start) {
947 busy_start = busy_loop_current_time();
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300948 continue;
949 }
Alexander Duyck37056712017-03-24 10:08:18 -0700950 if (!busy_loop_timeout(busy_start))
Eliezer Tamir76b1e9b2013-07-09 13:09:21 +0300951 continue;
952 }
953 busy_flag = 0;
Eliezer Tamir91e2fd332013-06-28 15:59:35 +0300954
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700955 /*
956 * If this is the first loop and we have a timeout
957 * given, then we convert to ktime_t and set the to
958 * pointer to the expiry value.
959 */
960 if (end_time && !to) {
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700961 expire = timespec64_to_ktime(*end_time);
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700962 to = &expire;
David Woodhouse9f729492006-01-18 17:44:05 -0800963 }
964
Tejun Heo5f820f62009-01-06 14:40:59 -0800965 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -0700966 timed_out = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 return count;
969}
970
Andi Kleen70674f92006-03-28 01:56:33 -0800971#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
972 sizeof(struct pollfd))
973
Al Viroe99ca562017-04-08 16:50:24 -0400974static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
Deepa Dinamani766b9f92016-05-19 17:09:05 -0700975 struct timespec64 *end_time)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
977 struct poll_wqueues table;
Gustavo A. R. Silva43e11fa2019-07-16 16:30:58 -0700978 int err = -EFAULT, fdcount, len;
Jes Sorensen30c14e42006-03-31 11:18:57 -0500979 /* Allocate small arguments on the stack to save memory and be
980 faster - use long to make sure the buffer is aligned properly
981 on 64 bit archs to avoid unaligned access */
982 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
Oleg Nesterov252e5722007-10-16 23:26:17 -0700983 struct poll_list *const head = (struct poll_list *)stack_pps;
984 struct poll_list *walk = head;
985 unsigned long todo = nfds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Jiri Slabyd554ed892010-03-05 13:42:42 -0800987 if (nfds > rlimit(RLIMIT_NOFILE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 return -EINVAL;
989
Oleg Nesterov252e5722007-10-16 23:26:17 -0700990 len = min_t(unsigned int, nfds, N_STACK_PPS);
991 for (;;) {
992 walk->next = NULL;
993 walk->len = len;
994 if (!len)
995 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Oleg Nesterov252e5722007-10-16 23:26:17 -0700997 if (copy_from_user(walk->entries, ufds + nfds-todo,
998 sizeof(struct pollfd) * walk->len))
999 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Oleg Nesterov252e5722007-10-16 23:26:17 -07001001 todo -= walk->len;
1002 if (!todo)
1003 break;
1004
1005 len = min(todo, POLLFD_PER_PAGE);
Gustavo A. R. Silva43e11fa2019-07-16 16:30:58 -07001006 walk = walk->next = kmalloc(struct_size(walk, entries, len),
Linus Torvalds0bcfe682021-09-07 11:26:23 -07001007 GFP_KERNEL);
Oleg Nesterov252e5722007-10-16 23:26:17 -07001008 if (!walk) {
1009 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 goto out_fds;
1011 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 }
David Woodhouse9f729492006-01-18 17:44:05 -08001013
Oleg Nesterov252e5722007-10-16 23:26:17 -07001014 poll_initwait(&table);
Mateusz Guzikccec5ee2016-01-06 06:41:53 +01001015 fdcount = do_poll(head, &table, end_time);
Oleg Nesterov252e5722007-10-16 23:26:17 -07001016 poll_freewait(&table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Linus Torvaldsef0ba052021-01-07 09:43:54 -08001018 if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1019 goto out_fds;
1020
Oleg Nesterov252e5722007-10-16 23:26:17 -07001021 for (walk = head; walk; walk = walk->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 struct pollfd *fds = walk->entries;
1023 int j;
1024
Linus Torvaldsef0ba052021-01-07 09:43:54 -08001025 for (j = walk->len; j; fds++, ufds++, j--)
1026 unsafe_put_user(fds->revents, &ufds->revents, Efault);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 }
Linus Torvaldsef0ba052021-01-07 09:43:54 -08001028 user_write_access_end();
Oleg Nesterov252e5722007-10-16 23:26:17 -07001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 err = fdcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031out_fds:
Oleg Nesterov252e5722007-10-16 23:26:17 -07001032 walk = head->next;
1033 while (walk) {
1034 struct poll_list *pos = walk;
1035 walk = walk->next;
1036 kfree(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 }
Oleg Nesterov252e5722007-10-16 23:26:17 -07001038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 return err;
Linus Torvaldsef0ba052021-01-07 09:43:54 -08001040
1041Efault:
1042 user_write_access_end();
1043 err = -EFAULT;
1044 goto out_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045}
David Woodhouse9f729492006-01-18 17:44:05 -08001046
Chris Wright3075d9d2007-10-16 23:27:18 -07001047static long do_restart_poll(struct restart_block *restart_block)
1048{
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001049 struct pollfd __user *ufds = restart_block->poll.ufds;
1050 int nfds = restart_block->poll.nfds;
Deepa Dinamani766b9f92016-05-19 17:09:05 -07001051 struct timespec64 *to = NULL, end_time;
Chris Wright3075d9d2007-10-16 23:27:18 -07001052 int ret;
1053
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001054 if (restart_block->poll.has_timeout) {
1055 end_time.tv_sec = restart_block->poll.tv_sec;
1056 end_time.tv_nsec = restart_block->poll.tv_nsec;
1057 to = &end_time;
1058 }
1059
1060 ret = do_sys_poll(ufds, nfds, to);
1061
Oleg Nesterov5abbe512021-02-01 18:46:41 +01001062 if (ret == -ERESTARTNOHAND)
1063 ret = set_restart_fn(restart_block, do_restart_poll);
1064
Chris Wright3075d9d2007-10-16 23:27:18 -07001065 return ret;
1066}
1067
Heiko Carstens5a8a82b2009-01-14 14:14:25 +01001068SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
Linus Torvaldsfaf30902012-02-21 17:24:20 -08001069 int, timeout_msecs)
David Woodhouse9f729492006-01-18 17:44:05 -08001070{
Deepa Dinamani766b9f92016-05-19 17:09:05 -07001071 struct timespec64 end_time, *to = NULL;
Chris Wright3075d9d2007-10-16 23:27:18 -07001072 int ret;
David Woodhouse9f729492006-01-18 17:44:05 -08001073
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001074 if (timeout_msecs >= 0) {
1075 to = &end_time;
1076 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1077 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
David Woodhouse9f729492006-01-18 17:44:05 -08001078 }
1079
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001080 ret = do_sys_poll(ufds, nfds, to);
1081
Oleg Nesterov8cf8b552019-07-16 16:29:56 -07001082 if (ret == -ERESTARTNOHAND) {
Chris Wright3075d9d2007-10-16 23:27:18 -07001083 struct restart_block *restart_block;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001084
Andy Lutomirskif56141e2015-02-12 15:01:14 -08001085 restart_block = &current->restart_block;
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001086 restart_block->poll.ufds = ufds;
1087 restart_block->poll.nfds = nfds;
1088
1089 if (timeout_msecs >= 0) {
1090 restart_block->poll.tv_sec = end_time.tv_sec;
1091 restart_block->poll.tv_nsec = end_time.tv_nsec;
1092 restart_block->poll.has_timeout = 1;
1093 } else
1094 restart_block->poll.has_timeout = 0;
1095
Oleg Nesterov5abbe512021-02-01 18:46:41 +01001096 ret = set_restart_fn(restart_block, do_restart_poll);
Chris Wright3075d9d2007-10-16 23:27:18 -07001097 }
1098 return ret;
David Woodhouse9f729492006-01-18 17:44:05 -08001099}
1100
Heiko Carstensd4e82042009-01-14 14:14:34 +01001101SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001102 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
Heiko Carstensd4e82042009-01-14 14:14:34 +01001103 size_t, sigsetsize)
David Woodhouse9f729492006-01-18 17:44:05 -08001104{
Deepa Dinamani36819ad2017-08-04 21:12:31 -07001105 struct timespec64 ts, end_time, *to = NULL;
David Woodhouse9f729492006-01-18 17:44:05 -08001106 int ret;
1107
1108 if (tsp) {
Deepa Dinamani36819ad2017-08-04 21:12:31 -07001109 if (get_timespec64(&ts, tsp))
David Woodhouse9f729492006-01-18 17:44:05 -08001110 return -EFAULT;
1111
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001112 to = &end_time;
1113 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1114 return -EINVAL;
David Woodhouse9f729492006-01-18 17:44:05 -08001115 }
1116
Oleg Nesterovb7724342019-07-16 16:29:53 -07001117 ret = set_user_sigmask(sigmask, sigsetsize);
Deepa Dinamanided653c2018-09-19 21:41:04 -07001118 if (ret)
1119 return ret;
David Woodhouse9f729492006-01-18 17:44:05 -08001120
Arjan van de Ven8ff3e8e2008-08-31 08:26:40 -07001121 ret = do_sys_poll(ufds, nfds, to);
Oleg Nesterovac301022019-07-16 16:29:59 -07001122 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
David Woodhouse9f729492006-01-18 17:44:05 -08001123}
Al Viroe99ca562017-04-08 16:50:24 -04001124
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001125#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1126
1127SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1128 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1129 size_t, sigsetsize)
1130{
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001131 struct timespec64 ts, end_time, *to = NULL;
1132 int ret;
1133
1134 if (tsp) {
1135 if (get_old_timespec32(&ts, tsp))
1136 return -EFAULT;
1137
1138 to = &end_time;
1139 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1140 return -EINVAL;
1141 }
1142
Oleg Nesterovb7724342019-07-16 16:29:53 -07001143 ret = set_user_sigmask(sigmask, sigsetsize);
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001144 if (ret)
1145 return ret;
1146
1147 ret = do_sys_poll(ufds, nfds, to);
Oleg Nesterovac301022019-07-16 16:29:59 -07001148 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001149}
1150#endif
1151
Al Viroe99ca562017-04-08 16:50:24 -04001152#ifdef CONFIG_COMPAT
1153#define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1154
Al Viroe99ca562017-04-08 16:50:24 -04001155/*
1156 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1157 * 64-bit unsigned longs.
1158 */
1159static
1160int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1161 unsigned long *fdset)
1162{
Al Viroe99ca562017-04-08 16:50:24 -04001163 if (ufdset) {
Al Viro464d6242017-06-04 03:24:26 -04001164 return compat_get_bitmap(fdset, ufdset, nr);
Al Viroe99ca562017-04-08 16:50:24 -04001165 } else {
Helge Deller79de3cb2017-08-23 22:37:00 +02001166 zero_fd_set(nr, fdset);
Al Viro464d6242017-06-04 03:24:26 -04001167 return 0;
Al Viroe99ca562017-04-08 16:50:24 -04001168 }
Al Viroe99ca562017-04-08 16:50:24 -04001169}
1170
1171static
1172int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1173 unsigned long *fdset)
1174{
Al Viroe99ca562017-04-08 16:50:24 -04001175 if (!ufdset)
1176 return 0;
Al Viro464d6242017-06-04 03:24:26 -04001177 return compat_put_bitmap(ufdset, fdset, nr);
Al Viroe99ca562017-04-08 16:50:24 -04001178}
1179
1180
1181/*
1182 * This is a virtual copy of sys_select from fs/select.c and probably
1183 * should be compared to it from time to time
1184 */
1185
1186/*
1187 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1188 * like to be certain this leads to no problems. So I return
1189 * EINTR just for safety.
1190 *
1191 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1192 * I'm trying ERESTARTNOHAND which restart only when you want to.
1193 */
1194static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1195 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
Deepa Dinamani36819ad2017-08-04 21:12:31 -07001196 struct timespec64 *end_time)
Al Viroe99ca562017-04-08 16:50:24 -04001197{
1198 fd_set_bits fds;
1199 void *bits;
1200 int size, max_fds, ret = -EINVAL;
1201 struct fdtable *fdt;
1202 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1203
1204 if (n < 0)
1205 goto out_nofds;
1206
1207 /* max_fds can increase, so grab it once to avoid race */
1208 rcu_read_lock();
1209 fdt = files_fdtable(current->files);
1210 max_fds = fdt->max_fds;
1211 rcu_read_unlock();
1212 if (n > max_fds)
1213 n = max_fds;
1214
1215 /*
1216 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1217 * since we used fdset we need to allocate memory in units of
1218 * long-words.
1219 */
1220 size = FDS_BYTES(n);
1221 bits = stack_fds;
1222 if (size > sizeof(stack_fds) / 6) {
Kees Cook6da2ec52018-06-12 13:55:00 -07001223 bits = kmalloc_array(6, size, GFP_KERNEL);
Al Viroe99ca562017-04-08 16:50:24 -04001224 ret = -ENOMEM;
1225 if (!bits)
1226 goto out_nofds;
1227 }
1228 fds.in = (unsigned long *) bits;
1229 fds.out = (unsigned long *) (bits + size);
1230 fds.ex = (unsigned long *) (bits + 2*size);
1231 fds.res_in = (unsigned long *) (bits + 3*size);
1232 fds.res_out = (unsigned long *) (bits + 4*size);
1233 fds.res_ex = (unsigned long *) (bits + 5*size);
1234
1235 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1236 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1237 (ret = compat_get_fd_set(n, exp, fds.ex)))
1238 goto out;
1239 zero_fd_set(n, fds.res_in);
1240 zero_fd_set(n, fds.res_out);
1241 zero_fd_set(n, fds.res_ex);
1242
1243 ret = do_select(n, &fds, end_time);
1244
1245 if (ret < 0)
1246 goto out;
1247 if (!ret) {
1248 ret = -ERESTARTNOHAND;
1249 if (signal_pending(current))
1250 goto out;
1251 ret = 0;
1252 }
1253
1254 if (compat_set_fd_set(n, inp, fds.res_in) ||
1255 compat_set_fd_set(n, outp, fds.res_out) ||
1256 compat_set_fd_set(n, exp, fds.res_ex))
1257 ret = -EFAULT;
1258out:
1259 if (bits != stack_fds)
1260 kfree(bits);
1261out_nofds:
1262 return ret;
1263}
1264
Dominik Brodowski05585e42018-03-20 19:33:48 +01001265static int do_compat_select(int n, compat_ulong_t __user *inp,
1266 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02001267 struct old_timeval32 __user *tvp)
Al Viroe99ca562017-04-08 16:50:24 -04001268{
Deepa Dinamani36819ad2017-08-04 21:12:31 -07001269 struct timespec64 end_time, *to = NULL;
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02001270 struct old_timeval32 tv;
Al Viroe99ca562017-04-08 16:50:24 -04001271 int ret;
1272
1273 if (tvp) {
1274 if (copy_from_user(&tv, tvp, sizeof(tv)))
1275 return -EFAULT;
1276
1277 to = &end_time;
1278 if (poll_select_set_timeout(to,
1279 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1280 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1281 return -EINVAL;
1282 }
1283
1284 ret = compat_core_sys_select(n, inp, outp, exp, to);
Oleg Nesterovac301022019-07-16 16:29:59 -07001285 return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
Al Viroe99ca562017-04-08 16:50:24 -04001286}
1287
Dominik Brodowski05585e42018-03-20 19:33:48 +01001288COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1289 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02001290 struct old_timeval32 __user *, tvp)
Dominik Brodowski05585e42018-03-20 19:33:48 +01001291{
1292 return do_compat_select(n, inp, outp, exp, tvp);
1293}
1294
Al Viroe99ca562017-04-08 16:50:24 -04001295struct compat_sel_arg_struct {
1296 compat_ulong_t n;
1297 compat_uptr_t inp;
1298 compat_uptr_t outp;
1299 compat_uptr_t exp;
1300 compat_uptr_t tvp;
1301};
1302
1303COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1304{
1305 struct compat_sel_arg_struct a;
1306
1307 if (copy_from_user(&a, arg, sizeof(a)))
1308 return -EFAULT;
Dominik Brodowski05585e42018-03-20 19:33:48 +01001309 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1310 compat_ptr(a.exp), compat_ptr(a.tvp));
Al Viroe99ca562017-04-08 16:50:24 -04001311}
1312
1313static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1314 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
Deepa Dinamanie0247072018-09-19 21:41:07 -07001315 void __user *tsp, compat_sigset_t __user *sigmask,
1316 compat_size_t sigsetsize, enum poll_time_type type)
Al Viroe99ca562017-04-08 16:50:24 -04001317{
Deepa Dinamani36819ad2017-08-04 21:12:31 -07001318 struct timespec64 ts, end_time, *to = NULL;
Al Viroe99ca562017-04-08 16:50:24 -04001319 int ret;
1320
1321 if (tsp) {
Deepa Dinamanie0247072018-09-19 21:41:07 -07001322 switch (type) {
1323 case PT_OLD_TIMESPEC:
1324 if (get_old_timespec32(&ts, tsp))
1325 return -EFAULT;
1326 break;
1327 case PT_TIMESPEC:
1328 if (get_timespec64(&ts, tsp))
1329 return -EFAULT;
1330 break;
1331 default:
1332 BUG();
1333 }
Al Viroe99ca562017-04-08 16:50:24 -04001334
1335 to = &end_time;
1336 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1337 return -EINVAL;
1338 }
1339
Oleg Nesterovb7724342019-07-16 16:29:53 -07001340 ret = set_compat_user_sigmask(sigmask, sigsetsize);
Deepa Dinamanided653c2018-09-19 21:41:04 -07001341 if (ret)
1342 return ret;
Al Viroe99ca562017-04-08 16:50:24 -04001343
1344 ret = compat_core_sys_select(n, inp, outp, exp, to);
Oleg Nesterovac301022019-07-16 16:29:59 -07001345 return poll_select_finish(&end_time, tsp, type, ret);
Al Viroe99ca562017-04-08 16:50:24 -04001346}
1347
Al Viro7e716092020-02-19 09:54:24 -05001348struct compat_sigset_argpack {
1349 compat_uptr_t p;
1350 compat_size_t size;
1351};
1352static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1353 struct compat_sigset_argpack __user *from)
1354{
1355 if (from) {
1356 if (!user_read_access_begin(from, sizeof(*from)))
1357 return -EFAULT;
1358 unsafe_get_user(to->p, &from->p, Efault);
1359 unsafe_get_user(to->size, &from->size, Efault);
1360 user_read_access_end();
1361 }
1362 return 0;
1363Efault:
1364 user_access_end();
1365 return -EFAULT;
1366}
1367
Deepa Dinamanie0247072018-09-19 21:41:07 -07001368COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1369 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1370 struct __kernel_timespec __user *, tsp, void __user *, sig)
1371{
Al Viro7e716092020-02-19 09:54:24 -05001372 struct compat_sigset_argpack x = {0, 0};
Deepa Dinamanie0247072018-09-19 21:41:07 -07001373
Al Viro7e716092020-02-19 09:54:24 -05001374 if (get_compat_sigset_argpack(&x, sig))
1375 return -EFAULT;
Deepa Dinamanie0247072018-09-19 21:41:07 -07001376
Al Viro7e716092020-02-19 09:54:24 -05001377 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1378 x.size, PT_TIMESPEC);
Deepa Dinamanie0247072018-09-19 21:41:07 -07001379}
1380
1381#if defined(CONFIG_COMPAT_32BIT_TIME)
1382
Arnd Bergmann8dabe722019-01-07 00:33:08 +01001383COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
Al Viroe99ca562017-04-08 16:50:24 -04001384 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02001385 struct old_timespec32 __user *, tsp, void __user *, sig)
Al Viroe99ca562017-04-08 16:50:24 -04001386{
Al Viro7e716092020-02-19 09:54:24 -05001387 struct compat_sigset_argpack x = {0, 0};
Al Viroe99ca562017-04-08 16:50:24 -04001388
Al Viro7e716092020-02-19 09:54:24 -05001389 if (get_compat_sigset_argpack(&x, sig))
1390 return -EFAULT;
Deepa Dinamanie0247072018-09-19 21:41:07 -07001391
Al Viro7e716092020-02-19 09:54:24 -05001392 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1393 x.size, PT_OLD_TIMESPEC);
Al Viroe99ca562017-04-08 16:50:24 -04001394}
1395
Deepa Dinamanie0247072018-09-19 21:41:07 -07001396#endif
1397
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001398#if defined(CONFIG_COMPAT_32BIT_TIME)
Arnd Bergmann8dabe722019-01-07 00:33:08 +01001399COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02001400 unsigned int, nfds, struct old_timespec32 __user *, tsp,
Al Viroe99ca562017-04-08 16:50:24 -04001401 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1402{
Deepa Dinamani36819ad2017-08-04 21:12:31 -07001403 struct timespec64 ts, end_time, *to = NULL;
Al Viroe99ca562017-04-08 16:50:24 -04001404 int ret;
1405
1406 if (tsp) {
Arnd Bergmann9afc5ee2018-07-13 12:52:28 +02001407 if (get_old_timespec32(&ts, tsp))
Al Viroe99ca562017-04-08 16:50:24 -04001408 return -EFAULT;
1409
1410 to = &end_time;
1411 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1412 return -EINVAL;
1413 }
1414
Oleg Nesterovb7724342019-07-16 16:29:53 -07001415 ret = set_compat_user_sigmask(sigmask, sigsetsize);
Deepa Dinamanided653c2018-09-19 21:41:04 -07001416 if (ret)
1417 return ret;
Al Viroe99ca562017-04-08 16:50:24 -04001418
1419 ret = do_sys_poll(ufds, nfds, to);
Oleg Nesterovac301022019-07-16 16:29:59 -07001420 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
Al Viroe99ca562017-04-08 16:50:24 -04001421}
1422#endif
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001423
1424/* New compat syscall for 64 bit time_t*/
1425COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1426 unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1427 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1428{
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001429 struct timespec64 ts, end_time, *to = NULL;
1430 int ret;
1431
1432 if (tsp) {
1433 if (get_timespec64(&ts, tsp))
1434 return -EFAULT;
1435
1436 to = &end_time;
1437 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1438 return -EINVAL;
1439 }
1440
Oleg Nesterovb7724342019-07-16 16:29:53 -07001441 ret = set_compat_user_sigmask(sigmask, sigsetsize);
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001442 if (ret)
1443 return ret;
1444
1445 ret = do_sys_poll(ufds, nfds, to);
Oleg Nesterovac301022019-07-16 16:29:59 -07001446 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
Deepa Dinamani8bd27a32018-09-19 21:41:06 -07001447}
1448
1449#endif