1// SPDX-License-Identifier: GPL-2.0
2/* Watch queue and general notification mechanism, built on pipes
3 *
4 * Copyright (C) 2020 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 *
7 * See Documentation/core-api/watch_queue.rst
8 */
9
10#define pr_fmt(fmt) "watchq: " fmt
11#include <linux/module.h>
12#include <linux/init.h>
13#include <linux/sched.h>
14#include <linux/slab.h>
15#include <linux/printk.h>
16#include <linux/miscdevice.h>
17#include <linux/fs.h>
18#include <linux/mm.h>
19#include <linux/pagemap.h>
20#include <linux/poll.h>
21#include <linux/uaccess.h>
22#include <linux/vmalloc.h>
23#include <linux/file.h>
24#include <linux/security.h>
25#include <linux/cred.h>
26#include <linux/sched/signal.h>
27#include <linux/watch_queue.h>
28#include <linux/pipe_fs_i.h>
29
30MODULE_DESCRIPTION("Watch queue");
31MODULE_AUTHOR("Red Hat, Inc.");
32
33#define WATCH_QUEUE_NOTE_SIZE 128
34#define WATCH_QUEUE_NOTES_PER_PAGE (PAGE_SIZE / WATCH_QUEUE_NOTE_SIZE)
35
36/*
37 * This must be called under the RCU read-lock, which makes
38 * sure that the wqueue still exists. It can then take the lock,
39 * and check that the wqueue hasn't been destroyed, which in
40 * turn makes sure that the notification pipe still exists.
41 */
42static inline bool lock_wqueue(struct watch_queue *wqueue)
43{
44 spin_lock_bh(lock: &wqueue->lock);
45 if (unlikely(!wqueue->pipe)) {
46 spin_unlock_bh(lock: &wqueue->lock);
47 return false;
48 }
49 return true;
50}
51
52static inline void unlock_wqueue(struct watch_queue *wqueue)
53{
54 spin_unlock_bh(lock: &wqueue->lock);
55}
56
57static void watch_queue_pipe_buf_release(struct pipe_inode_info *pipe,
58 struct pipe_buffer *buf)
59{
60 struct watch_queue *wqueue = (struct watch_queue *)buf->private;
61 struct page *page;
62 unsigned int bit;
63
64 /* We need to work out which note within the page this refers to, but
65 * the note might have been maximum size, so merely ANDing the offset
66 * off doesn't work. OTOH, the note must've been more than zero size.
67 */
68 bit = buf->offset + buf->len;
69 if ((bit & (WATCH_QUEUE_NOTE_SIZE - 1)) == 0)
70 bit -= WATCH_QUEUE_NOTE_SIZE;
71 bit /= WATCH_QUEUE_NOTE_SIZE;
72
73 page = buf->page;
74 bit += page->index;
75
76 set_bit(nr: bit, addr: wqueue->notes_bitmap);
77 generic_pipe_buf_release(pipe, buf);
78}
79
80// No try_steal function => no stealing
81#define watch_queue_pipe_buf_try_steal NULL
82
83/* New data written to a pipe may be appended to a buffer with this type. */
84static const struct pipe_buf_operations watch_queue_pipe_buf_ops = {
85 .release = watch_queue_pipe_buf_release,
86 .try_steal = watch_queue_pipe_buf_try_steal,
87 .get = generic_pipe_buf_get,
88};
89
90/*
91 * Post a notification to a watch queue.
92 *
93 * Must be called with the RCU lock for reading, and the
94 * watch_queue lock held, which guarantees that the pipe
95 * hasn't been released.
96 */
97static bool post_one_notification(struct watch_queue *wqueue,
98 struct watch_notification *n)
99{
100 void *p;
101 struct pipe_inode_info *pipe = wqueue->pipe;
102 struct pipe_buffer *buf;
103 struct page *page;
104 unsigned int head, tail, mask, note, offset, len;
105 bool done = false;
106
107 spin_lock_irq(lock: &pipe->rd_wait.lock);
108
109 mask = pipe->ring_size - 1;
110 head = pipe->head;
111 tail = pipe->tail;
112 if (pipe_full(head, tail, limit: pipe->ring_size))
113 goto lost;
114
115 note = find_first_bit(addr: wqueue->notes_bitmap, size: wqueue->nr_notes);
116 if (note >= wqueue->nr_notes)
117 goto lost;
118
119 page = wqueue->notes[note / WATCH_QUEUE_NOTES_PER_PAGE];
120 offset = note % WATCH_QUEUE_NOTES_PER_PAGE * WATCH_QUEUE_NOTE_SIZE;
121 get_page(page);
122 len = n->info & WATCH_INFO_LENGTH;
123 p = kmap_atomic(page);
124 memcpy(p + offset, n, len);
125 kunmap_atomic(p);
126
127 buf = &pipe->bufs[head & mask];
128 buf->page = page;
129 buf->private = (unsigned long)wqueue;
130 buf->ops = &watch_queue_pipe_buf_ops;
131 buf->offset = offset;
132 buf->len = len;
133 buf->flags = PIPE_BUF_FLAG_WHOLE;
134 smp_store_release(&pipe->head, head + 1); /* vs pipe_read() */
135
136 if (!test_and_clear_bit(nr: note, addr: wqueue->notes_bitmap)) {
137 spin_unlock_irq(lock: &pipe->rd_wait.lock);
138 BUG();
139 }
140 wake_up_interruptible_sync_poll_locked(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM);
141 done = true;
142
143out:
144 spin_unlock_irq(lock: &pipe->rd_wait.lock);
145 if (done)
146 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
147 return done;
148
149lost:
150 buf = &pipe->bufs[(head - 1) & mask];
151 buf->flags |= PIPE_BUF_FLAG_LOSS;
152 goto out;
153}
154
155/*
156 * Apply filter rules to a notification.
157 */
158static bool filter_watch_notification(const struct watch_filter *wf,
159 const struct watch_notification *n)
160{
161 const struct watch_type_filter *wt;
162 unsigned int st_bits = sizeof(wt->subtype_filter[0]) * 8;
163 unsigned int st_index = n->subtype / st_bits;
164 unsigned int st_bit = 1U << (n->subtype % st_bits);
165 int i;
166
167 if (!test_bit(n->type, wf->type_filter))
168 return false;
169
170 for (i = 0; i < wf->nr_filters; i++) {
171 wt = &wf->filters[i];
172 if (n->type == wt->type &&
173 (wt->subtype_filter[st_index] & st_bit) &&
174 (n->info & wt->info_mask) == wt->info_filter)
175 return true;
176 }
177
178 return false; /* If there is a filter, the default is to reject. */
179}
180
181/**
182 * __post_watch_notification - Post an event notification
183 * @wlist: The watch list to post the event to.
184 * @n: The notification record to post.
185 * @cred: The creds of the process that triggered the notification.
186 * @id: The ID to match on the watch.
187 *
188 * Post a notification of an event into a set of watch queues and let the users
189 * know.
190 *
191 * The size of the notification should be set in n->info & WATCH_INFO_LENGTH and
192 * should be in units of sizeof(*n).
193 */
194void __post_watch_notification(struct watch_list *wlist,
195 struct watch_notification *n,
196 const struct cred *cred,
197 u64 id)
198{
199 const struct watch_filter *wf;
200 struct watch_queue *wqueue;
201 struct watch *watch;
202
203 if (((n->info & WATCH_INFO_LENGTH) >> WATCH_INFO_LENGTH__SHIFT) == 0) {
204 WARN_ON(1);
205 return;
206 }
207
208 rcu_read_lock();
209
210 hlist_for_each_entry_rcu(watch, &wlist->watchers, list_node) {
211 if (watch->id != id)
212 continue;
213 n->info &= ~WATCH_INFO_ID;
214 n->info |= watch->info_id;
215
216 wqueue = rcu_dereference(watch->queue);
217 wf = rcu_dereference(wqueue->filter);
218 if (wf && !filter_watch_notification(wf, n))
219 continue;
220
221 if (security_post_notification(w_cred: watch->cred, cred, n) < 0)
222 continue;
223
224 if (lock_wqueue(wqueue)) {
225 post_one_notification(wqueue, n);
226 unlock_wqueue(wqueue);
227 }
228 }
229
230 rcu_read_unlock();
231}
232EXPORT_SYMBOL(__post_watch_notification);
233
234/*
235 * Allocate sufficient pages to preallocation for the requested number of
236 * notifications.
237 */
238long watch_queue_set_size(struct pipe_inode_info *pipe, unsigned int nr_notes)
239{
240 struct watch_queue *wqueue = pipe->watch_queue;
241 struct page **pages;
242 unsigned long *bitmap;
243 unsigned long user_bufs;
244 int ret, i, nr_pages;
245
246 if (!wqueue)
247 return -ENODEV;
248 if (wqueue->notes)
249 return -EBUSY;
250
251 if (nr_notes < 1 ||
252 nr_notes > 512) /* TODO: choose a better hard limit */
253 return -EINVAL;
254
255 nr_pages = (nr_notes + WATCH_QUEUE_NOTES_PER_PAGE - 1);
256 nr_pages /= WATCH_QUEUE_NOTES_PER_PAGE;
257 user_bufs = account_pipe_buffers(user: pipe->user, old: pipe->nr_accounted, new: nr_pages);
258
259 if (nr_pages > pipe->max_usage &&
260 (too_many_pipe_buffers_hard(user_bufs) ||
261 too_many_pipe_buffers_soft(user_bufs)) &&
262 pipe_is_unprivileged_user()) {
263 ret = -EPERM;
264 goto error;
265 }
266
267 nr_notes = nr_pages * WATCH_QUEUE_NOTES_PER_PAGE;
268 ret = pipe_resize_ring(pipe, roundup_pow_of_two(nr_notes));
269 if (ret < 0)
270 goto error;
271
272 ret = -ENOMEM;
273 pages = kcalloc(n: sizeof(struct page *), size: nr_pages, GFP_KERNEL);
274 if (!pages)
275 goto error;
276
277 for (i = 0; i < nr_pages; i++) {
278 pages[i] = alloc_page(GFP_KERNEL);
279 if (!pages[i])
280 goto error_p;
281 pages[i]->index = i * WATCH_QUEUE_NOTES_PER_PAGE;
282 }
283
284 bitmap = bitmap_alloc(nbits: nr_notes, GFP_KERNEL);
285 if (!bitmap)
286 goto error_p;
287
288 bitmap_fill(dst: bitmap, nbits: nr_notes);
289 wqueue->notes = pages;
290 wqueue->notes_bitmap = bitmap;
291 wqueue->nr_pages = nr_pages;
292 wqueue->nr_notes = nr_notes;
293 return 0;
294
295error_p:
296 while (--i >= 0)
297 __free_page(pages[i]);
298 kfree(objp: pages);
299error:
300 (void) account_pipe_buffers(user: pipe->user, old: nr_pages, new: pipe->nr_accounted);
301 return ret;
302}
303
304/*
305 * Set the filter on a watch queue.
306 */
307long watch_queue_set_filter(struct pipe_inode_info *pipe,
308 struct watch_notification_filter __user *_filter)
309{
310 struct watch_notification_type_filter *tf;
311 struct watch_notification_filter filter;
312 struct watch_type_filter *q;
313 struct watch_filter *wfilter;
314 struct watch_queue *wqueue = pipe->watch_queue;
315 int ret, nr_filter = 0, i;
316
317 if (!wqueue)
318 return -ENODEV;
319
320 if (!_filter) {
321 /* Remove the old filter */
322 wfilter = NULL;
323 goto set;
324 }
325
326 /* Grab the user's filter specification */
327 if (copy_from_user(to: &filter, from: _filter, n: sizeof(filter)) != 0)
328 return -EFAULT;
329 if (filter.nr_filters == 0 ||
330 filter.nr_filters > 16 ||
331 filter.__reserved != 0)
332 return -EINVAL;
333
334 tf = memdup_array_user(src: _filter->filters, n: filter.nr_filters, size: sizeof(*tf));
335 if (IS_ERR(ptr: tf))
336 return PTR_ERR(ptr: tf);
337
338 ret = -EINVAL;
339 for (i = 0; i < filter.nr_filters; i++) {
340 if ((tf[i].info_filter & ~tf[i].info_mask) ||
341 tf[i].info_mask & WATCH_INFO_LENGTH)
342 goto err_filter;
343 /* Ignore any unknown types */
344 if (tf[i].type >= WATCH_TYPE__NR)
345 continue;
346 nr_filter++;
347 }
348
349 /* Now we need to build the internal filter from only the relevant
350 * user-specified filters.
351 */
352 ret = -ENOMEM;
353 wfilter = kzalloc(struct_size(wfilter, filters, nr_filter), GFP_KERNEL);
354 if (!wfilter)
355 goto err_filter;
356 wfilter->nr_filters = nr_filter;
357
358 q = wfilter->filters;
359 for (i = 0; i < filter.nr_filters; i++) {
360 if (tf[i].type >= WATCH_TYPE__NR)
361 continue;
362
363 q->type = tf[i].type;
364 q->info_filter = tf[i].info_filter;
365 q->info_mask = tf[i].info_mask;
366 q->subtype_filter[0] = tf[i].subtype_filter[0];
367 __set_bit(q->type, wfilter->type_filter);
368 q++;
369 }
370
371 kfree(objp: tf);
372set:
373 pipe_lock(pipe);
374 wfilter = rcu_replace_pointer(wqueue->filter, wfilter,
375 lockdep_is_held(&pipe->mutex));
376 pipe_unlock(pipe);
377 if (wfilter)
378 kfree_rcu(wfilter, rcu);
379 return 0;
380
381err_filter:
382 kfree(objp: tf);
383 return ret;
384}
385
386static void __put_watch_queue(struct kref *kref)
387{
388 struct watch_queue *wqueue =
389 container_of(kref, struct watch_queue, usage);
390 struct watch_filter *wfilter;
391 int i;
392
393 for (i = 0; i < wqueue->nr_pages; i++)
394 __free_page(wqueue->notes[i]);
395 kfree(objp: wqueue->notes);
396 bitmap_free(bitmap: wqueue->notes_bitmap);
397
398 wfilter = rcu_access_pointer(wqueue->filter);
399 if (wfilter)
400 kfree_rcu(wfilter, rcu);
401 kfree_rcu(wqueue, rcu);
402}
403
404/**
405 * put_watch_queue - Dispose of a ref on a watchqueue.
406 * @wqueue: The watch queue to unref.
407 */
408void put_watch_queue(struct watch_queue *wqueue)
409{
410 kref_put(kref: &wqueue->usage, release: __put_watch_queue);
411}
412EXPORT_SYMBOL(put_watch_queue);
413
414static void free_watch(struct rcu_head *rcu)
415{
416 struct watch *watch = container_of(rcu, struct watch, rcu);
417
418 put_watch_queue(rcu_access_pointer(watch->queue));
419 atomic_dec(v: &watch->cred->user->nr_watches);
420 put_cred(cred: watch->cred);
421 kfree(objp: watch);
422}
423
424static void __put_watch(struct kref *kref)
425{
426 struct watch *watch = container_of(kref, struct watch, usage);
427
428 call_rcu(head: &watch->rcu, func: free_watch);
429}
430
431/*
432 * Discard a watch.
433 */
434static void put_watch(struct watch *watch)
435{
436 kref_put(kref: &watch->usage, release: __put_watch);
437}
438
439/**
440 * init_watch - Initialise a watch
441 * @watch: The watch to initialise.
442 * @wqueue: The queue to assign.
443 *
444 * Initialise a watch and set the watch queue.
445 */
446void init_watch(struct watch *watch, struct watch_queue *wqueue)
447{
448 kref_init(kref: &watch->usage);
449 INIT_HLIST_NODE(h: &watch->list_node);
450 INIT_HLIST_NODE(h: &watch->queue_node);
451 rcu_assign_pointer(watch->queue, wqueue);
452}
453
454static int add_one_watch(struct watch *watch, struct watch_list *wlist, struct watch_queue *wqueue)
455{
456 const struct cred *cred;
457 struct watch *w;
458
459 hlist_for_each_entry(w, &wlist->watchers, list_node) {
460 struct watch_queue *wq = rcu_access_pointer(w->queue);
461 if (wqueue == wq && watch->id == w->id)
462 return -EBUSY;
463 }
464
465 cred = current_cred();
466 if (atomic_inc_return(v: &cred->user->nr_watches) > task_rlimit(current, RLIMIT_NOFILE)) {
467 atomic_dec(v: &cred->user->nr_watches);
468 return -EAGAIN;
469 }
470
471 watch->cred = get_cred(cred);
472 rcu_assign_pointer(watch->watch_list, wlist);
473
474 kref_get(kref: &wqueue->usage);
475 kref_get(kref: &watch->usage);
476 hlist_add_head(n: &watch->queue_node, h: &wqueue->watches);
477 hlist_add_head_rcu(n: &watch->list_node, h: &wlist->watchers);
478 return 0;
479}
480
481/**
482 * add_watch_to_object - Add a watch on an object to a watch list
483 * @watch: The watch to add
484 * @wlist: The watch list to add to
485 *
486 * @watch->queue must have been set to point to the queue to post notifications
487 * to and the watch list of the object to be watched. @watch->cred must also
488 * have been set to the appropriate credentials and a ref taken on them.
489 *
490 * The caller must pin the queue and the list both and must hold the list
491 * locked against racing watch additions/removals.
492 */
493int add_watch_to_object(struct watch *watch, struct watch_list *wlist)
494{
495 struct watch_queue *wqueue;
496 int ret = -ENOENT;
497
498 rcu_read_lock();
499
500 wqueue = rcu_access_pointer(watch->queue);
501 if (lock_wqueue(wqueue)) {
502 spin_lock(lock: &wlist->lock);
503 ret = add_one_watch(watch, wlist, wqueue);
504 spin_unlock(lock: &wlist->lock);
505 unlock_wqueue(wqueue);
506 }
507
508 rcu_read_unlock();
509 return ret;
510}
511EXPORT_SYMBOL(add_watch_to_object);
512
513/**
514 * remove_watch_from_object - Remove a watch or all watches from an object.
515 * @wlist: The watch list to remove from
516 * @wq: The watch queue of interest (ignored if @all is true)
517 * @id: The ID of the watch to remove (ignored if @all is true)
518 * @all: True to remove all objects
519 *
520 * Remove a specific watch or all watches from an object. A notification is
521 * sent to the watcher to tell them that this happened.
522 */
523int remove_watch_from_object(struct watch_list *wlist, struct watch_queue *wq,
524 u64 id, bool all)
525{
526 struct watch_notification_removal n;
527 struct watch_queue *wqueue;
528 struct watch *watch;
529 int ret = -EBADSLT;
530
531 rcu_read_lock();
532
533again:
534 spin_lock(lock: &wlist->lock);
535 hlist_for_each_entry(watch, &wlist->watchers, list_node) {
536 if (all ||
537 (watch->id == id && rcu_access_pointer(watch->queue) == wq))
538 goto found;
539 }
540 spin_unlock(lock: &wlist->lock);
541 goto out;
542
543found:
544 ret = 0;
545 hlist_del_init_rcu(n: &watch->list_node);
546 rcu_assign_pointer(watch->watch_list, NULL);
547 spin_unlock(lock: &wlist->lock);
548
549 /* We now own the reference on watch that used to belong to wlist. */
550
551 n.watch.type = WATCH_TYPE_META;
552 n.watch.subtype = WATCH_META_REMOVAL_NOTIFICATION;
553 n.watch.info = watch->info_id | watch_sizeof(n.watch);
554 n.id = id;
555 if (id != 0)
556 n.watch.info = watch->info_id | watch_sizeof(n);
557
558 wqueue = rcu_dereference(watch->queue);
559
560 if (lock_wqueue(wqueue)) {
561 post_one_notification(wqueue, n: &n.watch);
562
563 if (!hlist_unhashed(h: &watch->queue_node)) {
564 hlist_del_init_rcu(n: &watch->queue_node);
565 put_watch(watch);
566 }
567
568 unlock_wqueue(wqueue);
569 }
570
571 if (wlist->release_watch) {
572 void (*release_watch)(struct watch *);
573
574 release_watch = wlist->release_watch;
575 rcu_read_unlock();
576 (*release_watch)(watch);
577 rcu_read_lock();
578 }
579 put_watch(watch);
580
581 if (all && !hlist_empty(h: &wlist->watchers))
582 goto again;
583out:
584 rcu_read_unlock();
585 return ret;
586}
587EXPORT_SYMBOL(remove_watch_from_object);
588
589/*
590 * Remove all the watches that are contributory to a queue. This has the
591 * potential to race with removal of the watches by the destruction of the
592 * objects being watched or with the distribution of notifications.
593 */
594void watch_queue_clear(struct watch_queue *wqueue)
595{
596 struct watch_list *wlist;
597 struct watch *watch;
598 bool release;
599
600 rcu_read_lock();
601 spin_lock_bh(lock: &wqueue->lock);
602
603 /*
604 * This pipe can be freed by callers like free_pipe_info().
605 * Removing this reference also prevents new notifications.
606 */
607 wqueue->pipe = NULL;
608
609 while (!hlist_empty(h: &wqueue->watches)) {
610 watch = hlist_entry(wqueue->watches.first, struct watch, queue_node);
611 hlist_del_init_rcu(n: &watch->queue_node);
612 /* We now own a ref on the watch. */
613 spin_unlock_bh(lock: &wqueue->lock);
614
615 /* We can't do the next bit under the queue lock as we need to
616 * get the list lock - which would cause a deadlock if someone
617 * was removing from the opposite direction at the same time or
618 * posting a notification.
619 */
620 wlist = rcu_dereference(watch->watch_list);
621 if (wlist) {
622 void (*release_watch)(struct watch *);
623
624 spin_lock(lock: &wlist->lock);
625
626 release = !hlist_unhashed(h: &watch->list_node);
627 if (release) {
628 hlist_del_init_rcu(n: &watch->list_node);
629 rcu_assign_pointer(watch->watch_list, NULL);
630
631 /* We now own a second ref on the watch. */
632 }
633
634 release_watch = wlist->release_watch;
635 spin_unlock(lock: &wlist->lock);
636
637 if (release) {
638 if (release_watch) {
639 rcu_read_unlock();
640 /* This might need to call dput(), so
641 * we have to drop all the locks.
642 */
643 (*release_watch)(watch);
644 rcu_read_lock();
645 }
646 put_watch(watch);
647 }
648 }
649
650 put_watch(watch);
651 spin_lock_bh(lock: &wqueue->lock);
652 }
653
654 spin_unlock_bh(lock: &wqueue->lock);
655 rcu_read_unlock();
656}
657
658/**
659 * get_watch_queue - Get a watch queue from its file descriptor.
660 * @fd: The fd to query.
661 */
662struct watch_queue *get_watch_queue(int fd)
663{
664 struct pipe_inode_info *pipe;
665 struct watch_queue *wqueue = ERR_PTR(error: -EINVAL);
666 struct fd f;
667
668 f = fdget(fd);
669 if (f.file) {
670 pipe = get_pipe_info(file: f.file, for_splice: false);
671 if (pipe && pipe->watch_queue) {
672 wqueue = pipe->watch_queue;
673 kref_get(kref: &wqueue->usage);
674 }
675 fdput(fd: f);
676 }
677
678 return wqueue;
679}
680EXPORT_SYMBOL(get_watch_queue);
681
682/*
683 * Initialise a watch queue
684 */
685int watch_queue_init(struct pipe_inode_info *pipe)
686{
687 struct watch_queue *wqueue;
688
689 wqueue = kzalloc(size: sizeof(*wqueue), GFP_KERNEL);
690 if (!wqueue)
691 return -ENOMEM;
692
693 wqueue->pipe = pipe;
694 kref_init(kref: &wqueue->usage);
695 spin_lock_init(&wqueue->lock);
696 INIT_HLIST_HEAD(&wqueue->watches);
697
698 pipe->watch_queue = wqueue;
699 return 0;
700}
701

source code of linux/kernel/watch_queue.c