1
2
3
4
5
6
7
8
9#include <linux/init.h>
10#include <linux/kernel.h>
11#include <linux/sched/signal.h>
12#include <linux/fs.h>
13#include <linux/file.h>
14#include <linux/signal.h>
15#include <linux/errno.h>
16#include <linux/mm.h>
17#include <linux/slab.h>
18#include <linux/poll.h>
19#include <linux/string.h>
20#include <linux/list.h>
21#include <linux/hash.h>
22#include <linux/spinlock.h>
23#include <linux/syscalls.h>
24#include <linux/rbtree.h>
25#include <linux/wait.h>
26#include <linux/eventpoll.h>
27#include <linux/mount.h>
28#include <linux/bitops.h>
29#include <linux/mutex.h>
30#include <linux/anon_inodes.h>
31#include <linux/device.h>
32#include <linux/uaccess.h>
33#include <asm/io.h>
34#include <asm/mman.h>
35#include <linux/atomic.h>
36#include <linux/proc_fs.h>
37#include <linux/seq_file.h>
38#include <linux/compat.h>
39#include <linux/rculist.h>
40#include <net/busy_poll.h>
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91#define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE)
92
93#define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT)
94
95#define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \
96 EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE)
97
98
99#define EP_MAX_NESTS 4
100
101#define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
102
103#define EP_UNACTIVE_PTR ((void *) -1L)
104
105#define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry))
106
107struct epoll_filefd {
108 struct file *file;
109 int fd;
110} __packed;
111
112
113struct eppoll_entry {
114
115 struct eppoll_entry *next;
116
117
118 struct epitem *base;
119
120
121
122
123
124 wait_queue_entry_t wait;
125
126
127 wait_queue_head_t *whead;
128};
129
130
131
132
133
134
135
136struct epitem {
137 union {
138
139 struct rb_node rbn;
140
141 struct rcu_head rcu;
142 };
143
144
145 struct list_head rdllink;
146
147
148
149
150
151 struct epitem *next;
152
153
154 struct epoll_filefd ffd;
155
156
157 struct eppoll_entry *pwqlist;
158
159
160 struct eventpoll *ep;
161
162
163 struct hlist_node fllink;
164
165
166 struct wakeup_source __rcu *ws;
167
168
169 struct epoll_event event;
170};
171
172
173
174
175
176
177struct eventpoll {
178
179
180
181
182
183
184 struct mutex mtx;
185
186
187 wait_queue_head_t wq;
188
189
190 wait_queue_head_t poll_wait;
191
192
193 struct list_head rdllist;
194
195
196 rwlock_t lock;
197
198
199 struct rb_root_cached rbr;
200
201
202
203
204
205
206 struct epitem *ovflist;
207
208
209 struct wakeup_source *ws;
210
211
212 struct user_struct *user;
213
214 struct file *file;
215
216
217 u64 gen;
218 struct hlist_head refs;
219
220#ifdef CONFIG_NET_RX_BUSY_POLL
221
222 unsigned int napi_id;
223#endif
224
225#ifdef CONFIG_DEBUG_LOCK_ALLOC
226
227 u8 nests;
228#endif
229};
230
231
232struct ep_pqueue {
233 poll_table pt;
234 struct epitem *epi;
235};
236
237
238
239
240
241static long max_user_watches __read_mostly;
242
243
244
245
246static DEFINE_MUTEX(epmutex);
247
248static u64 loop_check_gen = 0;
249
250
251static struct eventpoll *inserting_into;
252
253
254static struct kmem_cache *epi_cache __read_mostly;
255
256
257static struct kmem_cache *pwq_cache __read_mostly;
258
259
260
261
262
263struct epitems_head {
264 struct hlist_head epitems;
265 struct epitems_head *next;
266};
267static struct epitems_head *tfile_check_list = EP_UNACTIVE_PTR;
268
269static struct kmem_cache *ephead_cache __read_mostly;
270
271static inline void free_ephead(struct epitems_head *head)
272{
273 if (head)
274 kmem_cache_free(ephead_cache, head);
275}
276
277static void list_file(struct file *file)
278{
279 struct epitems_head *head;
280
281 head = container_of(file->f_ep, struct epitems_head, epitems);
282 if (!head->next) {
283 head->next = tfile_check_list;
284 tfile_check_list = head;
285 }
286}
287
288static void unlist_file(struct epitems_head *head)
289{
290 struct epitems_head *to_free = head;
291 struct hlist_node *p = rcu_dereference(hlist_first_rcu(&head->epitems));
292 if (p) {
293 struct epitem *epi= container_of(p, struct epitem, fllink);
294 spin_lock(&epi->ffd.file->f_lock);
295 if (!hlist_empty(&head->epitems))
296 to_free = NULL;
297 head->next = NULL;
298 spin_unlock(&epi->ffd.file->f_lock);
299 }
300 free_ephead(to_free);
301}
302
303#ifdef CONFIG_SYSCTL
304
305#include <linux/sysctl.h>
306
307static long long_zero;
308static long long_max = LONG_MAX;
309
310struct ctl_table epoll_table[] = {
311 {
312 .procname = "max_user_watches",
313 .data = &max_user_watches,
314 .maxlen = sizeof(max_user_watches),
315 .mode = 0644,
316 .proc_handler = proc_doulongvec_minmax,
317 .extra1 = &long_zero,
318 .extra2 = &long_max,
319 },
320 { }
321};
322#endif
323
324static const struct file_operations eventpoll_fops;
325
326static inline int is_file_epoll(struct file *f)
327{
328 return f->f_op == &eventpoll_fops;
329}
330
331
332static inline void ep_set_ffd(struct epoll_filefd *ffd,
333 struct file *file, int fd)
334{
335 ffd->file = file;
336 ffd->fd = fd;
337}
338
339
340static inline int ep_cmp_ffd(struct epoll_filefd *p1,
341 struct epoll_filefd *p2)
342{
343 return (p1->file > p2->file ? +1:
344 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
345}
346
347
348static inline int ep_is_linked(struct epitem *epi)
349{
350 return !list_empty(&epi->rdllink);
351}
352
353static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p)
354{
355 return container_of(p, struct eppoll_entry, wait);
356}
357
358
359static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p)
360{
361 return container_of(p, struct eppoll_entry, wait)->base;
362}
363
364
365
366
367
368
369
370
371
372static inline int ep_events_available(struct eventpoll *ep)
373{
374 return !list_empty_careful(&ep->rdllist) ||
375 READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR;
376}
377
378#ifdef CONFIG_NET_RX_BUSY_POLL
379static bool ep_busy_loop_end(void *p, unsigned long start_time)
380{
381 struct eventpoll *ep = p;
382
383 return ep_events_available(ep) || busy_loop_timeout(start_time);
384}
385
386
387
388
389
390
391
392static bool ep_busy_loop(struct eventpoll *ep, int nonblock)
393{
394 unsigned int napi_id = READ_ONCE(ep->napi_id);
395
396 if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) {
397 napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep, false,
398 BUSY_POLL_BUDGET);
399 if (ep_events_available(ep))
400 return true;
401
402
403
404
405
406 ep->napi_id = 0;
407 return false;
408 }
409 return false;
410}
411
412
413
414
415static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
416{
417 struct eventpoll *ep;
418 unsigned int napi_id;
419 struct socket *sock;
420 struct sock *sk;
421
422 if (!net_busy_loop_on())
423 return;
424
425 sock = sock_from_file(epi->ffd.file);
426 if (!sock)
427 return;
428
429 sk = sock->sk;
430 if (!sk)
431 return;
432
433 napi_id = READ_ONCE(sk->sk_napi_id);
434 ep = epi->ep;
435
436
437
438
439
440 if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id)
441 return;
442
443
444 ep->napi_id = napi_id;
445}
446
447#else
448
449static inline bool ep_busy_loop(struct eventpoll *ep, int nonblock)
450{
451 return false;
452}
453
454static inline void ep_set_busy_poll_napi_id(struct epitem *epi)
455{
456}
457
458#endif
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485#ifdef CONFIG_DEBUG_LOCK_ALLOC
486
487static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi)
488{
489 struct eventpoll *ep_src;
490 unsigned long flags;
491 u8 nests = 0;
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508 if (epi) {
509 if ((is_file_epoll(epi->ffd.file))) {
510 ep_src = epi->ffd.file->private_data;
511 nests = ep_src->nests;
512 } else {
513 nests = 1;
514 }
515 }
516 spin_lock_irqsave_nested(&ep->poll_wait.lock, flags, nests);
517 ep->nests = nests + 1;
518 wake_up_locked_poll(&ep->poll_wait, EPOLLIN);
519 ep->nests = 0;
520 spin_unlock_irqrestore(&ep->poll_wait.lock, flags);
521}
522
523#else
524
525static void ep_poll_safewake(struct eventpoll *ep, struct epitem *epi)
526{
527 wake_up_poll(&ep->poll_wait, EPOLLIN);
528}
529
530#endif
531
532static void ep_remove_wait_queue(struct eppoll_entry *pwq)
533{
534 wait_queue_head_t *whead;
535
536 rcu_read_lock();
537
538
539
540
541
542
543 whead = smp_load_acquire(&pwq->whead);
544 if (whead)
545 remove_wait_queue(whead, &pwq->wait);
546 rcu_read_unlock();
547}
548
549
550
551
552
553
554static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi)
555{
556 struct eppoll_entry **p = &epi->pwqlist;
557 struct eppoll_entry *pwq;
558
559 while ((pwq = *p) != NULL) {
560 *p = pwq->next;
561 ep_remove_wait_queue(pwq);
562 kmem_cache_free(pwq_cache, pwq);
563 }
564}
565
566
567static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi)
568{
569 return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx));
570}
571
572
573static inline void ep_pm_stay_awake(struct epitem *epi)
574{
575 struct wakeup_source *ws = ep_wakeup_source(epi);
576
577 if (ws)
578 __pm_stay_awake(ws);
579}
580
581static inline bool ep_has_wakeup_source(struct epitem *epi)
582{
583 return rcu_access_pointer(epi->ws) ? true : false;
584}
585
586
587static inline void ep_pm_stay_awake_rcu(struct epitem *epi)
588{
589 struct wakeup_source *ws;
590
591 rcu_read_lock();
592 ws = rcu_dereference(epi->ws);
593 if (ws)
594 __pm_stay_awake(ws);
595 rcu_read_unlock();
596}
597
598
599
600
601
602
603static void ep_start_scan(struct eventpoll *ep, struct list_head *txlist)
604{
605
606
607
608
609
610
611
612
613 lockdep_assert_irqs_enabled();
614 write_lock_irq(&ep->lock);
615 list_splice_init(&ep->rdllist, txlist);
616 WRITE_ONCE(ep->ovflist, NULL);
617 write_unlock_irq(&ep->lock);
618}
619
620static void ep_done_scan(struct eventpoll *ep,
621 struct list_head *txlist)
622{
623 struct epitem *epi, *nepi;
624
625 write_lock_irq(&ep->lock);
626
627
628
629
630
631 for (nepi = READ_ONCE(ep->ovflist); (epi = nepi) != NULL;
632 nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
633
634
635
636
637
638
639 if (!ep_is_linked(epi)) {
640
641
642
643
644 list_add(&epi->rdllink, &ep->rdllist);
645 ep_pm_stay_awake(epi);
646 }
647 }
648
649
650
651
652
653 WRITE_ONCE(ep->ovflist, EP_UNACTIVE_PTR);
654
655
656
657
658 list_splice(txlist, &ep->rdllist);
659 __pm_relax(ep->ws);
660
661 if (!list_empty(&ep->rdllist)) {
662 if (waitqueue_active(&ep->wq))
663 wake_up(&ep->wq);
664 }
665
666 write_unlock_irq(&ep->lock);
667}
668
669static void epi_rcu_free(struct rcu_head *head)
670{
671 struct epitem *epi = container_of(head, struct epitem, rcu);
672 kmem_cache_free(epi_cache, epi);
673}
674
675
676
677
678
679static int ep_remove(struct eventpoll *ep, struct epitem *epi)
680{
681 struct file *file = epi->ffd.file;
682 struct epitems_head *to_free;
683 struct hlist_head *head;
684
685 lockdep_assert_irqs_enabled();
686
687
688
689
690 ep_unregister_pollwait(ep, epi);
691
692
693 spin_lock(&file->f_lock);
694 to_free = NULL;
695 head = file->f_ep;
696 if (head->first == &epi->fllink && !epi->fllink.next) {
697 file->f_ep = NULL;
698 if (!is_file_epoll(file)) {
699 struct epitems_head *v;
700 v = container_of(head, struct epitems_head, epitems);
701 if (!smp_load_acquire(&v->next))
702 to_free = v;
703 }
704 }
705 hlist_del_rcu(&epi->fllink);
706 spin_unlock(&file->f_lock);
707 free_ephead(to_free);
708
709 rb_erase_cached(&epi->rbn, &ep->rbr);
710
711 write_lock_irq(&ep->lock);
712 if (ep_is_linked(epi))
713 list_del_init(&epi->rdllink);
714 write_unlock_irq(&ep->lock);
715
716 wakeup_source_unregister(ep_wakeup_source(epi));
717
718
719
720
721
722
723
724 call_rcu(&epi->rcu, epi_rcu_free);
725
726 atomic_long_dec(&ep->user->epoll_watches);
727
728 return 0;
729}
730
731static void ep_free(struct eventpoll *ep)
732{
733 struct rb_node *rbp;
734 struct epitem *epi;
735
736
737 if (waitqueue_active(&ep->poll_wait))
738 ep_poll_safewake(ep, NULL);
739
740
741
742
743
744
745
746
747
748 mutex_lock(&epmutex);
749
750
751
752
753 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
754 epi = rb_entry(rbp, struct epitem, rbn);
755
756 ep_unregister_pollwait(ep, epi);
757 cond_resched();
758 }
759
760
761
762
763
764
765
766
767
768 mutex_lock(&ep->mtx);
769 while ((rbp = rb_first_cached(&ep->rbr)) != NULL) {
770 epi = rb_entry(rbp, struct epitem, rbn);
771 ep_remove(ep, epi);
772 cond_resched();
773 }
774 mutex_unlock(&ep->mtx);
775
776 mutex_unlock(&epmutex);
777 mutex_destroy(&ep->mtx);
778 free_uid(ep->user);
779 wakeup_source_unregister(ep->ws);
780 kfree(ep);
781}
782
783static int ep_eventpoll_release(struct inode *inode, struct file *file)
784{
785 struct eventpoll *ep = file->private_data;
786
787 if (ep)
788 ep_free(ep);
789
790 return 0;
791}
792
793static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, int depth);
794
795static __poll_t __ep_eventpoll_poll(struct file *file, poll_table *wait, int depth)
796{
797 struct eventpoll *ep = file->private_data;
798 LIST_HEAD(txlist);
799 struct epitem *epi, *tmp;
800 poll_table pt;
801 __poll_t res = 0;
802
803 init_poll_funcptr(&pt, NULL);
804
805
806 poll_wait(file, &ep->poll_wait, wait);
807
808
809
810
811
812 mutex_lock_nested(&ep->mtx, depth);
813 ep_start_scan(ep, &txlist);
814 list_for_each_entry_safe(epi, tmp, &txlist, rdllink) {
815 if (ep_item_poll(epi, &pt, depth + 1)) {
816 res = EPOLLIN | EPOLLRDNORM;
817 break;
818 } else {
819
820
821
822
823
824 __pm_relax(ep_wakeup_source(epi));
825 list_del_init(&epi->rdllink);
826 }
827 }
828 ep_done_scan(ep, &txlist);
829 mutex_unlock(&ep->mtx);
830 return res;
831}
832
833
834
835
836
837
838static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt,
839 int depth)
840{
841 struct file *file = epi->ffd.file;
842 __poll_t res;
843
844 pt->_key = epi->event.events;
845 if (!is_file_epoll(file))
846 res = vfs_poll(file, pt);
847 else
848 res = __ep_eventpoll_poll(file, pt, depth);
849 return res & epi->event.events;
850}
851
852static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait)
853{
854 return __ep_eventpoll_poll(file, wait, 0);
855}
856
857#ifdef CONFIG_PROC_FS
858static void ep_show_fdinfo(struct seq_file *m, struct file *f)
859{
860 struct eventpoll *ep = f->private_data;
861 struct rb_node *rbp;
862
863 mutex_lock(&ep->mtx);
864 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
865 struct epitem *epi = rb_entry(rbp, struct epitem, rbn);
866 struct inode *inode = file_inode(epi->ffd.file);
867
868 seq_printf(m, "tfd: %8d events: %8x data: %16llx "
869 " pos:%lli ino:%lx sdev:%x\n",
870 epi->ffd.fd, epi->event.events,
871 (long long)epi->event.data,
872 (long long)epi->ffd.file->f_pos,
873 inode->i_ino, inode->i_sb->s_dev);
874 if (seq_has_overflowed(m))
875 break;
876 }
877 mutex_unlock(&ep->mtx);
878}
879#endif
880
881
882static const struct file_operations eventpoll_fops = {
883#ifdef CONFIG_PROC_FS
884 .show_fdinfo = ep_show_fdinfo,
885#endif
886 .release = ep_eventpoll_release,
887 .poll = ep_eventpoll_poll,
888 .llseek = noop_llseek,
889};
890
891
892
893
894
895
896void eventpoll_release_file(struct file *file)
897{
898 struct eventpoll *ep;
899 struct epitem *epi;
900 struct hlist_node *next;
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915 mutex_lock(&epmutex);
916 if (unlikely(!file->f_ep)) {
917 mutex_unlock(&epmutex);
918 return;
919 }
920 hlist_for_each_entry_safe(epi, next, file->f_ep, fllink) {
921 ep = epi->ep;
922 mutex_lock_nested(&ep->mtx, 0);
923 ep_remove(ep, epi);
924 mutex_unlock(&ep->mtx);
925 }
926 mutex_unlock(&epmutex);
927}
928
929static int ep_alloc(struct eventpoll **pep)
930{
931 int error;
932 struct user_struct *user;
933 struct eventpoll *ep;
934
935 user = get_current_user();
936 error = -ENOMEM;
937 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
938 if (unlikely(!ep))
939 goto free_uid;
940
941 mutex_init(&ep->mtx);
942 rwlock_init(&ep->lock);
943 init_waitqueue_head(&ep->wq);
944 init_waitqueue_head(&ep->poll_wait);
945 INIT_LIST_HEAD(&ep->rdllist);
946 ep->rbr = RB_ROOT_CACHED;
947 ep->ovflist = EP_UNACTIVE_PTR;
948 ep->user = user;
949
950 *pep = ep;
951
952 return 0;
953
954free_uid:
955 free_uid(user);
956 return error;
957}
958
959
960
961
962
963
964static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd)
965{
966 int kcmp;
967 struct rb_node *rbp;
968 struct epitem *epi, *epir = NULL;
969 struct epoll_filefd ffd;
970
971 ep_set_ffd(&ffd, file, fd);
972 for (rbp = ep->rbr.rb_root.rb_node; rbp; ) {
973 epi = rb_entry(rbp, struct epitem, rbn);
974 kcmp = ep_cmp_ffd(&ffd, &epi->ffd);
975 if (kcmp > 0)
976 rbp = rbp->rb_right;
977 else if (kcmp < 0)
978 rbp = rbp->rb_left;
979 else {
980 epir = epi;
981 break;
982 }
983 }
984
985 return epir;
986}
987
988#ifdef CONFIG_KCMP
989static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff)
990{
991 struct rb_node *rbp;
992 struct epitem *epi;
993
994 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
995 epi = rb_entry(rbp, struct epitem, rbn);
996 if (epi->ffd.fd == tfd) {
997 if (toff == 0)
998 return epi;
999 else
1000 toff--;
1001 }
1002 cond_resched();
1003 }
1004
1005 return NULL;
1006}
1007
1008struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd,
1009 unsigned long toff)
1010{
1011 struct file *file_raw;
1012 struct eventpoll *ep;
1013 struct epitem *epi;
1014
1015 if (!is_file_epoll(file))
1016 return ERR_PTR(-EINVAL);
1017
1018 ep = file->private_data;
1019
1020 mutex_lock(&ep->mtx);
1021 epi = ep_find_tfd(ep, tfd, toff);
1022 if (epi)
1023 file_raw = epi->ffd.file;
1024 else
1025 file_raw = ERR_PTR(-ENOENT);
1026 mutex_unlock(&ep->mtx);
1027
1028 return file_raw;
1029}
1030#endif
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050static inline bool list_add_tail_lockless(struct list_head *new,
1051 struct list_head *head)
1052{
1053 struct list_head *prev;
1054
1055
1056
1057
1058
1059
1060
1061 if (cmpxchg(&new->next, new, head) != new)
1062 return false;
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072 prev = xchg(&head->prev, new);
1073
1074
1075
1076
1077
1078
1079 prev->next = new;
1080 new->prev = prev;
1081
1082 return true;
1083}
1084
1085
1086
1087
1088
1089
1090
1091static inline bool chain_epi_lockless(struct epitem *epi)
1092{
1093 struct eventpoll *ep = epi->ep;
1094
1095
1096 if (epi->next != EP_UNACTIVE_PTR)
1097 return false;
1098
1099
1100 if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR)
1101 return false;
1102
1103
1104 epi->next = xchg(&ep->ovflist, epi);
1105
1106 return true;
1107}
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1128{
1129 int pwake = 0;
1130 struct epitem *epi = ep_item_from_wait(wait);
1131 struct eventpoll *ep = epi->ep;
1132 __poll_t pollflags = key_to_poll(key);
1133 unsigned long flags;
1134 int ewake = 0;
1135
1136 read_lock_irqsave(&ep->lock, flags);
1137
1138 ep_set_busy_poll_napi_id(epi);
1139
1140
1141
1142
1143
1144
1145
1146 if (!(epi->event.events & ~EP_PRIVATE_BITS))
1147 goto out_unlock;
1148
1149
1150
1151
1152
1153
1154
1155 if (pollflags && !(pollflags & epi->event.events))
1156 goto out_unlock;
1157
1158
1159
1160
1161
1162
1163
1164 if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) {
1165 if (chain_epi_lockless(epi))
1166 ep_pm_stay_awake_rcu(epi);
1167 } else if (!ep_is_linked(epi)) {
1168
1169 if (list_add_tail_lockless(&epi->rdllink, &ep->rdllist))
1170 ep_pm_stay_awake_rcu(epi);
1171 }
1172
1173
1174
1175
1176
1177 if (waitqueue_active(&ep->wq)) {
1178 if ((epi->event.events & EPOLLEXCLUSIVE) &&
1179 !(pollflags & POLLFREE)) {
1180 switch (pollflags & EPOLLINOUT_BITS) {
1181 case EPOLLIN:
1182 if (epi->event.events & EPOLLIN)
1183 ewake = 1;
1184 break;
1185 case EPOLLOUT:
1186 if (epi->event.events & EPOLLOUT)
1187 ewake = 1;
1188 break;
1189 case 0:
1190 ewake = 1;
1191 break;
1192 }
1193 }
1194 wake_up(&ep->wq);
1195 }
1196 if (waitqueue_active(&ep->poll_wait))
1197 pwake++;
1198
1199out_unlock:
1200 read_unlock_irqrestore(&ep->lock, flags);
1201
1202
1203 if (pwake)
1204 ep_poll_safewake(ep, epi);
1205
1206 if (!(epi->event.events & EPOLLEXCLUSIVE))
1207 ewake = 1;
1208
1209 if (pollflags & POLLFREE) {
1210
1211
1212
1213
1214
1215 list_del_init(&wait->entry);
1216
1217
1218
1219
1220
1221
1222 smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL);
1223 }
1224
1225 return ewake;
1226}
1227
1228
1229
1230
1231
1232static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
1233 poll_table *pt)
1234{
1235 struct ep_pqueue *epq = container_of(pt, struct ep_pqueue, pt);
1236 struct epitem *epi = epq->epi;
1237 struct eppoll_entry *pwq;
1238
1239 if (unlikely(!epi))
1240 return;
1241
1242 pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL);
1243 if (unlikely(!pwq)) {
1244 epq->epi = NULL;
1245 return;
1246 }
1247
1248 init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
1249 pwq->whead = whead;
1250 pwq->base = epi;
1251 if (epi->event.events & EPOLLEXCLUSIVE)
1252 add_wait_queue_exclusive(whead, &pwq->wait);
1253 else
1254 add_wait_queue(whead, &pwq->wait);
1255 pwq->next = epi->pwqlist;
1256 epi->pwqlist = pwq;
1257}
1258
1259static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi)
1260{
1261 int kcmp;
1262 struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL;
1263 struct epitem *epic;
1264 bool leftmost = true;
1265
1266 while (*p) {
1267 parent = *p;
1268 epic = rb_entry(parent, struct epitem, rbn);
1269 kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd);
1270 if (kcmp > 0) {
1271 p = &parent->rb_right;
1272 leftmost = false;
1273 } else
1274 p = &parent->rb_left;
1275 }
1276 rb_link_node(&epi->rbn, parent, p);
1277 rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost);
1278}
1279
1280
1281
1282#define PATH_ARR_SIZE 5
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 };
1295static int path_count[PATH_ARR_SIZE];
1296
1297static int path_count_inc(int nests)
1298{
1299
1300 if (nests == 0)
1301 return 0;
1302
1303 if (++path_count[nests] > path_limits[nests])
1304 return -1;
1305 return 0;
1306}
1307
1308static void path_count_init(void)
1309{
1310 int i;
1311
1312 for (i = 0; i < PATH_ARR_SIZE; i++)
1313 path_count[i] = 0;
1314}
1315
1316static int reverse_path_check_proc(struct hlist_head *refs, int depth)
1317{
1318 int error = 0;
1319 struct epitem *epi;
1320
1321 if (depth > EP_MAX_NESTS)
1322 return -1;
1323
1324
1325 hlist_for_each_entry_rcu(epi, refs, fllink) {
1326 struct hlist_head *refs = &epi->ep->refs;
1327 if (hlist_empty(refs))
1328 error = path_count_inc(depth);
1329 else
1330 error = reverse_path_check_proc(refs, depth + 1);
1331 if (error != 0)
1332 break;
1333 }
1334 return error;
1335}
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347static int reverse_path_check(void)
1348{
1349 struct epitems_head *p;
1350
1351 for (p = tfile_check_list; p != EP_UNACTIVE_PTR; p = p->next) {
1352 int error;
1353 path_count_init();
1354 rcu_read_lock();
1355 error = reverse_path_check_proc(&p->epitems, 0);
1356 rcu_read_unlock();
1357 if (error)
1358 return error;
1359 }
1360 return 0;
1361}
1362
1363static int ep_create_wakeup_source(struct epitem *epi)
1364{
1365 struct name_snapshot n;
1366 struct wakeup_source *ws;
1367
1368 if (!epi->ep->ws) {
1369 epi->ep->ws = wakeup_source_register(NULL, "eventpoll");
1370 if (!epi->ep->ws)
1371 return -ENOMEM;
1372 }
1373
1374 take_dentry_name_snapshot(&n, epi->ffd.file->f_path.dentry);
1375 ws = wakeup_source_register(NULL, n.name.name);
1376 release_dentry_name_snapshot(&n);
1377
1378 if (!ws)
1379 return -ENOMEM;
1380 rcu_assign_pointer(epi->ws, ws);
1381
1382 return 0;
1383}
1384
1385
1386static noinline void ep_destroy_wakeup_source(struct epitem *epi)
1387{
1388 struct wakeup_source *ws = ep_wakeup_source(epi);
1389
1390 RCU_INIT_POINTER(epi->ws, NULL);
1391
1392
1393
1394
1395
1396
1397 synchronize_rcu();
1398 wakeup_source_unregister(ws);
1399}
1400
1401static int attach_epitem(struct file *file, struct epitem *epi)
1402{
1403 struct epitems_head *to_free = NULL;
1404 struct hlist_head *head = NULL;
1405 struct eventpoll *ep = NULL;
1406
1407 if (is_file_epoll(file))
1408 ep = file->private_data;
1409
1410 if (ep) {
1411 head = &ep->refs;
1412 } else if (!READ_ONCE(file->f_ep)) {
1413allocate:
1414 to_free = kmem_cache_zalloc(ephead_cache, GFP_KERNEL);
1415 if (!to_free)
1416 return -ENOMEM;
1417 head = &to_free->epitems;
1418 }
1419 spin_lock(&file->f_lock);
1420 if (!file->f_ep) {
1421 if (unlikely(!head)) {
1422 spin_unlock(&file->f_lock);
1423 goto allocate;
1424 }
1425 file->f_ep = head;
1426 to_free = NULL;
1427 }
1428 hlist_add_head_rcu(&epi->fllink, file->f_ep);
1429 spin_unlock(&file->f_lock);
1430 free_ephead(to_free);
1431 return 0;
1432}
1433
1434
1435
1436
1437static int ep_insert(struct eventpoll *ep, const struct epoll_event *event,
1438 struct file *tfile, int fd, int full_check)
1439{
1440 int error, pwake = 0;
1441 __poll_t revents;
1442 long user_watches;
1443 struct epitem *epi;
1444 struct ep_pqueue epq;
1445 struct eventpoll *tep = NULL;
1446
1447 if (is_file_epoll(tfile))
1448 tep = tfile->private_data;
1449
1450 lockdep_assert_irqs_enabled();
1451
1452 user_watches = atomic_long_read(&ep->user->epoll_watches);
1453 if (unlikely(user_watches >= max_user_watches))
1454 return -ENOSPC;
1455 if (!(epi = kmem_cache_zalloc(epi_cache, GFP_KERNEL)))
1456 return -ENOMEM;
1457
1458
1459 INIT_LIST_HEAD(&epi->rdllink);
1460 epi->ep = ep;
1461 ep_set_ffd(&epi->ffd, tfile, fd);
1462 epi->event = *event;
1463 epi->next = EP_UNACTIVE_PTR;
1464
1465 if (tep)
1466 mutex_lock_nested(&tep->mtx, 1);
1467
1468 if (unlikely(attach_epitem(tfile, epi) < 0)) {
1469 kmem_cache_free(epi_cache, epi);
1470 if (tep)
1471 mutex_unlock(&tep->mtx);
1472 return -ENOMEM;
1473 }
1474
1475 if (full_check && !tep)
1476 list_file(tfile);
1477
1478 atomic_long_inc(&ep->user->epoll_watches);
1479
1480
1481
1482
1483
1484 ep_rbtree_insert(ep, epi);
1485 if (tep)
1486 mutex_unlock(&tep->mtx);
1487
1488
1489 if (unlikely(full_check && reverse_path_check())) {
1490 ep_remove(ep, epi);
1491 return -EINVAL;
1492 }
1493
1494 if (epi->event.events & EPOLLWAKEUP) {
1495 error = ep_create_wakeup_source(epi);
1496 if (error) {
1497 ep_remove(ep, epi);
1498 return error;
1499 }
1500 }
1501
1502
1503 epq.epi = epi;
1504 init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
1505
1506
1507
1508
1509
1510
1511
1512
1513 revents = ep_item_poll(epi, &epq.pt, 1);
1514
1515
1516
1517
1518
1519
1520 if (unlikely(!epq.epi)) {
1521 ep_remove(ep, epi);
1522 return -ENOMEM;
1523 }
1524
1525
1526 write_lock_irq(&ep->lock);
1527
1528
1529 ep_set_busy_poll_napi_id(epi);
1530
1531
1532 if (revents && !ep_is_linked(epi)) {
1533 list_add_tail(&epi->rdllink, &ep->rdllist);
1534 ep_pm_stay_awake(epi);
1535
1536
1537 if (waitqueue_active(&ep->wq))
1538 wake_up(&ep->wq);
1539 if (waitqueue_active(&ep->poll_wait))
1540 pwake++;
1541 }
1542
1543 write_unlock_irq(&ep->lock);
1544
1545
1546 if (pwake)
1547 ep_poll_safewake(ep, NULL);
1548
1549 return 0;
1550}
1551
1552
1553
1554
1555
1556static int ep_modify(struct eventpoll *ep, struct epitem *epi,
1557 const struct epoll_event *event)
1558{
1559 int pwake = 0;
1560 poll_table pt;
1561
1562 lockdep_assert_irqs_enabled();
1563
1564 init_poll_funcptr(&pt, NULL);
1565
1566
1567
1568
1569
1570
1571 epi->event.events = event->events;
1572 epi->event.data = event->data;
1573 if (epi->event.events & EPOLLWAKEUP) {
1574 if (!ep_has_wakeup_source(epi))
1575 ep_create_wakeup_source(epi);
1576 } else if (ep_has_wakeup_source(epi)) {
1577 ep_destroy_wakeup_source(epi);
1578 }
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598 smp_mb();
1599
1600
1601
1602
1603
1604
1605
1606 if (ep_item_poll(epi, &pt, 1)) {
1607 write_lock_irq(&ep->lock);
1608 if (!ep_is_linked(epi)) {
1609 list_add_tail(&epi->rdllink, &ep->rdllist);
1610 ep_pm_stay_awake(epi);
1611
1612
1613 if (waitqueue_active(&ep->wq))
1614 wake_up(&ep->wq);
1615 if (waitqueue_active(&ep->poll_wait))
1616 pwake++;
1617 }
1618 write_unlock_irq(&ep->lock);
1619 }
1620
1621
1622 if (pwake)
1623 ep_poll_safewake(ep, NULL);
1624
1625 return 0;
1626}
1627
1628static int ep_send_events(struct eventpoll *ep,
1629 struct epoll_event __user *events, int maxevents)
1630{
1631 struct epitem *epi, *tmp;
1632 LIST_HEAD(txlist);
1633 poll_table pt;
1634 int res = 0;
1635
1636
1637
1638
1639
1640
1641 if (fatal_signal_pending(current))
1642 return -EINTR;
1643
1644 init_poll_funcptr(&pt, NULL);
1645
1646 mutex_lock(&ep->mtx);
1647 ep_start_scan(ep, &txlist);
1648
1649
1650
1651
1652
1653 list_for_each_entry_safe(epi, tmp, &txlist, rdllink) {
1654 struct wakeup_source *ws;
1655 __poll_t revents;
1656
1657 if (res >= maxevents)
1658 break;
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669 ws = ep_wakeup_source(epi);
1670 if (ws) {
1671 if (ws->active)
1672 __pm_stay_awake(ep->ws);
1673 __pm_relax(ws);
1674 }
1675
1676 list_del_init(&epi->rdllink);
1677
1678
1679
1680
1681
1682
1683 revents = ep_item_poll(epi, &pt, 1);
1684 if (!revents)
1685 continue;
1686
1687 if (__put_user(revents, &events->events) ||
1688 __put_user(epi->event.data, &events->data)) {
1689 list_add(&epi->rdllink, &txlist);
1690 ep_pm_stay_awake(epi);
1691 if (!res)
1692 res = -EFAULT;
1693 break;
1694 }
1695 res++;
1696 events++;
1697 if (epi->event.events & EPOLLONESHOT)
1698 epi->event.events &= EP_PRIVATE_BITS;
1699 else if (!(epi->event.events & EPOLLET)) {
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711 list_add_tail(&epi->rdllink, &ep->rdllist);
1712 ep_pm_stay_awake(epi);
1713 }
1714 }
1715 ep_done_scan(ep, &txlist);
1716 mutex_unlock(&ep->mtx);
1717
1718 return res;
1719}
1720
1721static struct timespec64 *ep_timeout_to_timespec(struct timespec64 *to, long ms)
1722{
1723 struct timespec64 now;
1724
1725 if (ms < 0)
1726 return NULL;
1727
1728 if (!ms) {
1729 to->tv_sec = 0;
1730 to->tv_nsec = 0;
1731 return to;
1732 }
1733
1734 to->tv_sec = ms / MSEC_PER_SEC;
1735 to->tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC);
1736
1737 ktime_get_ts64(&now);
1738 *to = timespec64_add_safe(now, *to);
1739 return to;
1740}
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
1760 int maxevents, struct timespec64 *timeout)
1761{
1762 int res, eavail, timed_out = 0;
1763 u64 slack = 0;
1764 wait_queue_entry_t wait;
1765 ktime_t expires, *to = NULL;
1766
1767 lockdep_assert_irqs_enabled();
1768
1769 if (timeout && (timeout->tv_sec | timeout->tv_nsec)) {
1770 slack = select_estimate_accuracy(timeout);
1771 to = &expires;
1772 *to = timespec64_to_ktime(*timeout);
1773 } else if (timeout) {
1774
1775
1776
1777
1778 timed_out = 1;
1779 }
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789 eavail = ep_events_available(ep);
1790
1791 while (1) {
1792 if (eavail) {
1793
1794
1795
1796
1797
1798 res = ep_send_events(ep, events, maxevents);
1799 if (res)
1800 return res;
1801 }
1802
1803 if (timed_out)
1804 return 0;
1805
1806 eavail = ep_busy_loop(ep, timed_out);
1807 if (eavail)
1808 continue;
1809
1810 if (signal_pending(current))
1811 return -EINTR;
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824 init_wait(&wait);
1825
1826 write_lock_irq(&ep->lock);
1827
1828
1829
1830
1831
1832 __set_current_state(TASK_INTERRUPTIBLE);
1833
1834
1835
1836
1837
1838
1839
1840
1841 eavail = ep_events_available(ep);
1842 if (!eavail)
1843 __add_wait_queue_exclusive(&ep->wq, &wait);
1844
1845 write_unlock_irq(&ep->lock);
1846
1847 if (!eavail)
1848 timed_out = !schedule_hrtimeout_range(to, slack,
1849 HRTIMER_MODE_ABS);
1850 __set_current_state(TASK_RUNNING);
1851
1852
1853
1854
1855
1856
1857 eavail = 1;
1858
1859 if (!list_empty_careful(&wait.entry)) {
1860 write_lock_irq(&ep->lock);
1861
1862
1863
1864
1865
1866
1867
1868 if (timed_out)
1869 eavail = list_empty(&wait.entry);
1870 __remove_wait_queue(&ep->wq, &wait);
1871 write_unlock_irq(&ep->lock);
1872 }
1873 }
1874}
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888static int ep_loop_check_proc(struct eventpoll *ep, int depth)
1889{
1890 int error = 0;
1891 struct rb_node *rbp;
1892 struct epitem *epi;
1893
1894 mutex_lock_nested(&ep->mtx, depth + 1);
1895 ep->gen = loop_check_gen;
1896 for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) {
1897 epi = rb_entry(rbp, struct epitem, rbn);
1898 if (unlikely(is_file_epoll(epi->ffd.file))) {
1899 struct eventpoll *ep_tovisit;
1900 ep_tovisit = epi->ffd.file->private_data;
1901 if (ep_tovisit->gen == loop_check_gen)
1902 continue;
1903 if (ep_tovisit == inserting_into || depth > EP_MAX_NESTS)
1904 error = -1;
1905 else
1906 error = ep_loop_check_proc(ep_tovisit, depth + 1);
1907 if (error != 0)
1908 break;
1909 } else {
1910
1911
1912
1913
1914
1915
1916
1917
1918 list_file(epi->ffd.file);
1919 }
1920 }
1921 mutex_unlock(&ep->mtx);
1922
1923 return error;
1924}
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937static int ep_loop_check(struct eventpoll *ep, struct eventpoll *to)
1938{
1939 inserting_into = ep;
1940 return ep_loop_check_proc(to, 0);
1941}
1942
1943static void clear_tfile_check_list(void)
1944{
1945 rcu_read_lock();
1946 while (tfile_check_list != EP_UNACTIVE_PTR) {
1947 struct epitems_head *head = tfile_check_list;
1948 tfile_check_list = head->next;
1949 unlist_file(head);
1950 }
1951 rcu_read_unlock();
1952}
1953
1954
1955
1956
1957static int do_epoll_create(int flags)
1958{
1959 int error, fd;
1960 struct eventpoll *ep = NULL;
1961 struct file *file;
1962
1963
1964 BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
1965
1966 if (flags & ~EPOLL_CLOEXEC)
1967 return -EINVAL;
1968
1969
1970
1971 error = ep_alloc(&ep);
1972 if (error < 0)
1973 return error;
1974
1975
1976
1977
1978 fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC));
1979 if (fd < 0) {
1980 error = fd;
1981 goto out_free_ep;
1982 }
1983 file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep,
1984 O_RDWR | (flags & O_CLOEXEC));
1985 if (IS_ERR(file)) {
1986 error = PTR_ERR(file);
1987 goto out_free_fd;
1988 }
1989 ep->file = file;
1990 fd_install(fd, file);
1991 return fd;
1992
1993out_free_fd:
1994 put_unused_fd(fd);
1995out_free_ep:
1996 ep_free(ep);
1997 return error;
1998}
1999
2000SYSCALL_DEFINE1(epoll_create1, int, flags)
2001{
2002 return do_epoll_create(flags);
2003}
2004
2005SYSCALL_DEFINE1(epoll_create, int, size)
2006{
2007 if (size <= 0)
2008 return -EINVAL;
2009
2010 return do_epoll_create(0);
2011}
2012
2013static inline int epoll_mutex_lock(struct mutex *mutex, int depth,
2014 bool nonblock)
2015{
2016 if (!nonblock) {
2017 mutex_lock_nested(mutex, depth);
2018 return 0;
2019 }
2020 if (mutex_trylock(mutex))
2021 return 0;
2022 return -EAGAIN;
2023}
2024
2025int do_epoll_ctl(int epfd, int op, int fd, struct epoll_event *epds,
2026 bool nonblock)
2027{
2028 int error;
2029 int full_check = 0;
2030 struct fd f, tf;
2031 struct eventpoll *ep;
2032 struct epitem *epi;
2033 struct eventpoll *tep = NULL;
2034
2035 error = -EBADF;
2036 f = fdget(epfd);
2037 if (!f.file)
2038 goto error_return;
2039
2040
2041 tf = fdget(fd);
2042 if (!tf.file)
2043 goto error_fput;
2044
2045
2046 error = -EPERM;
2047 if (!file_can_poll(tf.file))
2048 goto error_tgt_fput;
2049
2050
2051 if (ep_op_has_event(op))
2052 ep_take_care_of_epollwakeup(epds);
2053
2054
2055
2056
2057
2058
2059 error = -EINVAL;
2060 if (f.file == tf.file || !is_file_epoll(f.file))
2061 goto error_tgt_fput;
2062
2063
2064
2065
2066
2067
2068 if (ep_op_has_event(op) && (epds->events & EPOLLEXCLUSIVE)) {
2069 if (op == EPOLL_CTL_MOD)
2070 goto error_tgt_fput;
2071 if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) ||
2072 (epds->events & ~EPOLLEXCLUSIVE_OK_BITS)))
2073 goto error_tgt_fput;
2074 }
2075
2076
2077
2078
2079
2080 ep = f.file->private_data;
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097 error = epoll_mutex_lock(&ep->mtx, 0, nonblock);
2098 if (error)
2099 goto error_tgt_fput;
2100 if (op == EPOLL_CTL_ADD) {
2101 if (READ_ONCE(f.file->f_ep) || ep->gen == loop_check_gen ||
2102 is_file_epoll(tf.file)) {
2103 mutex_unlock(&ep->mtx);
2104 error = epoll_mutex_lock(&epmutex, 0, nonblock);
2105 if (error)
2106 goto error_tgt_fput;
2107 loop_check_gen++;
2108 full_check = 1;
2109 if (is_file_epoll(tf.file)) {
2110 tep = tf.file->private_data;
2111 error = -ELOOP;
2112 if (ep_loop_check(ep, tep) != 0)
2113 goto error_tgt_fput;
2114 }
2115 error = epoll_mutex_lock(&ep->mtx, 0, nonblock);
2116 if (error)
2117 goto error_tgt_fput;
2118 }
2119 }
2120
2121
2122
2123
2124
2125
2126 epi = ep_find(ep, tf.file, fd);
2127
2128 error = -EINVAL;
2129 switch (op) {
2130 case EPOLL_CTL_ADD:
2131 if (!epi) {
2132 epds->events |= EPOLLERR | EPOLLHUP;
2133 error = ep_insert(ep, epds, tf.file, fd, full_check);
2134 } else
2135 error = -EEXIST;
2136 break;
2137 case EPOLL_CTL_DEL:
2138 if (epi)
2139 error = ep_remove(ep, epi);
2140 else
2141 error = -ENOENT;
2142 break;
2143 case EPOLL_CTL_MOD:
2144 if (epi) {
2145 if (!(epi->event.events & EPOLLEXCLUSIVE)) {
2146 epds->events |= EPOLLERR | EPOLLHUP;
2147 error = ep_modify(ep, epi, epds);
2148 }
2149 } else
2150 error = -ENOENT;
2151 break;
2152 }
2153 mutex_unlock(&ep->mtx);
2154
2155error_tgt_fput:
2156 if (full_check) {
2157 clear_tfile_check_list();
2158 loop_check_gen++;
2159 mutex_unlock(&epmutex);
2160 }
2161
2162 fdput(tf);
2163error_fput:
2164 fdput(f);
2165error_return:
2166
2167 return error;
2168}
2169
2170
2171
2172
2173
2174
2175SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
2176 struct epoll_event __user *, event)
2177{
2178 struct epoll_event epds;
2179
2180 if (ep_op_has_event(op) &&
2181 copy_from_user(&epds, event, sizeof(struct epoll_event)))
2182 return -EFAULT;
2183
2184 return do_epoll_ctl(epfd, op, fd, &epds, false);
2185}
2186
2187
2188
2189
2190
2191static int do_epoll_wait(int epfd, struct epoll_event __user *events,
2192 int maxevents, struct timespec64 *to)
2193{
2194 int error;
2195 struct fd f;
2196 struct eventpoll *ep;
2197
2198
2199 if (maxevents <= 0 || maxevents > EP_MAX_EVENTS)
2200 return -EINVAL;
2201
2202
2203 if (!access_ok(events, maxevents * sizeof(struct epoll_event)))
2204 return -EFAULT;
2205
2206
2207 f = fdget(epfd);
2208 if (!f.file)
2209 return -EBADF;
2210
2211
2212
2213
2214
2215 error = -EINVAL;
2216 if (!is_file_epoll(f.file))
2217 goto error_fput;
2218
2219
2220
2221
2222
2223 ep = f.file->private_data;
2224
2225
2226 error = ep_poll(ep, events, maxevents, to);
2227
2228error_fput:
2229 fdput(f);
2230 return error;
2231}
2232
2233SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
2234 int, maxevents, int, timeout)
2235{
2236 struct timespec64 to;
2237
2238 return do_epoll_wait(epfd, events, maxevents,
2239 ep_timeout_to_timespec(&to, timeout));
2240}
2241
2242
2243
2244
2245
2246static int do_epoll_pwait(int epfd, struct epoll_event __user *events,
2247 int maxevents, struct timespec64 *to,
2248 const sigset_t __user *sigmask, size_t sigsetsize)
2249{
2250 int error;
2251
2252
2253
2254
2255
2256 error = set_user_sigmask(sigmask, sigsetsize);
2257 if (error)
2258 return error;
2259
2260 error = do_epoll_wait(epfd, events, maxevents, to);
2261
2262 restore_saved_sigmask_unless(error == -EINTR);
2263
2264 return error;
2265}
2266
2267SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
2268 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
2269 size_t, sigsetsize)
2270{
2271 struct timespec64 to;
2272
2273 return do_epoll_pwait(epfd, events, maxevents,
2274 ep_timeout_to_timespec(&to, timeout),
2275 sigmask, sigsetsize);
2276}
2277
2278SYSCALL_DEFINE6(epoll_pwait2, int, epfd, struct epoll_event __user *, events,
2279 int, maxevents, const struct __kernel_timespec __user *, timeout,
2280 const sigset_t __user *, sigmask, size_t, sigsetsize)
2281{
2282 struct timespec64 ts, *to = NULL;
2283
2284 if (timeout) {
2285 if (get_timespec64(&ts, timeout))
2286 return -EFAULT;
2287 to = &ts;
2288 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
2289 return -EINVAL;
2290 }
2291
2292 return do_epoll_pwait(epfd, events, maxevents, to,
2293 sigmask, sigsetsize);
2294}
2295
2296#ifdef CONFIG_COMPAT
2297static int do_compat_epoll_pwait(int epfd, struct epoll_event __user *events,
2298 int maxevents, struct timespec64 *timeout,
2299 const compat_sigset_t __user *sigmask,
2300 compat_size_t sigsetsize)
2301{
2302 long err;
2303
2304
2305
2306
2307
2308 err = set_compat_user_sigmask(sigmask, sigsetsize);
2309 if (err)
2310 return err;
2311
2312 err = do_epoll_wait(epfd, events, maxevents, timeout);
2313
2314 restore_saved_sigmask_unless(err == -EINTR);
2315
2316 return err;
2317}
2318
2319COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd,
2320 struct epoll_event __user *, events,
2321 int, maxevents, int, timeout,
2322 const compat_sigset_t __user *, sigmask,
2323 compat_size_t, sigsetsize)
2324{
2325 struct timespec64 to;
2326
2327 return do_compat_epoll_pwait(epfd, events, maxevents,
2328 ep_timeout_to_timespec(&to, timeout),
2329 sigmask, sigsetsize);
2330}
2331
2332COMPAT_SYSCALL_DEFINE6(epoll_pwait2, int, epfd,
2333 struct epoll_event __user *, events,
2334 int, maxevents,
2335 const struct __kernel_timespec __user *, timeout,
2336 const compat_sigset_t __user *, sigmask,
2337 compat_size_t, sigsetsize)
2338{
2339 struct timespec64 ts, *to = NULL;
2340
2341 if (timeout) {
2342 if (get_timespec64(&ts, timeout))
2343 return -EFAULT;
2344 to = &ts;
2345 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
2346 return -EINVAL;
2347 }
2348
2349 return do_compat_epoll_pwait(epfd, events, maxevents, to,
2350 sigmask, sigsetsize);
2351}
2352
2353#endif
2354
2355static int __init eventpoll_init(void)
2356{
2357 struct sysinfo si;
2358
2359 si_meminfo(&si);
2360
2361
2362
2363 max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) /
2364 EP_ITEM_COST;
2365 BUG_ON(max_user_watches < 0);
2366
2367
2368
2369
2370
2371 BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128);
2372
2373
2374 epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem),
2375 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
2376
2377
2378 pwq_cache = kmem_cache_create("eventpoll_pwq",
2379 sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL);
2380
2381 ephead_cache = kmem_cache_create("ep_head",
2382 sizeof(struct epitems_head), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL);
2383
2384 return 0;
2385}
2386fs_initcall(eventpoll_init);
2387