Loading...
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/poll.h>
14#include <linux/sched/signal.h>
15#include <linux/uio.h>
16#include <linux/miscdevice.h>
17#include <linux/pagemap.h>
18#include <linux/file.h>
19#include <linux/slab.h>
20#include <linux/pipe_fs_i.h>
21#include <linux/swap.h>
22#include <linux/splice.h>
23#include <linux/sched.h>
24
25MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26MODULE_ALIAS("devname:fuse");
27
28/* Ordinary requests have even IDs, while interrupts IDs are odd */
29#define FUSE_INT_REQ_BIT (1ULL << 0)
30#define FUSE_REQ_ID_STEP (1ULL << 1)
31
32static struct kmem_cache *fuse_req_cachep;
33
34static struct fuse_dev *fuse_get_dev(struct file *file)
35{
36 /*
37 * Lockless access is OK, because file->private data is set
38 * once during mount and is valid until the file is released.
39 */
40 return READ_ONCE(file->private_data);
41}
42
43static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)
44{
45 INIT_LIST_HEAD(&req->list);
46 INIT_LIST_HEAD(&req->intr_entry);
47 init_waitqueue_head(&req->waitq);
48 refcount_set(&req->count, 1);
49 __set_bit(FR_PENDING, &req->flags);
50 req->fm = fm;
51}
52
53static struct fuse_req *fuse_request_alloc(struct fuse_mount *fm, gfp_t flags)
54{
55 struct fuse_req *req = kmem_cache_zalloc(fuse_req_cachep, flags);
56 if (req)
57 fuse_request_init(fm, req);
58
59 return req;
60}
61
62static void fuse_request_free(struct fuse_req *req)
63{
64 kmem_cache_free(fuse_req_cachep, req);
65}
66
67static void __fuse_get_request(struct fuse_req *req)
68{
69 refcount_inc(&req->count);
70}
71
72/* Must be called with > 1 refcount */
73static void __fuse_put_request(struct fuse_req *req)
74{
75 refcount_dec(&req->count);
76}
77
78void fuse_set_initialized(struct fuse_conn *fc)
79{
80 /* Make sure stores before this are seen on another CPU */
81 smp_wmb();
82 fc->initialized = 1;
83}
84
85static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
86{
87 return !fc->initialized || (for_background && fc->blocked);
88}
89
90static void fuse_drop_waiting(struct fuse_conn *fc)
91{
92 /*
93 * lockess check of fc->connected is okay, because atomic_dec_and_test()
94 * provides a memory barrier matched with the one in fuse_wait_aborted()
95 * to ensure no wake-up is missed.
96 */
97 if (atomic_dec_and_test(&fc->num_waiting) &&
98 !READ_ONCE(fc->connected)) {
99 /* wake up aborters */
100 wake_up_all(&fc->blocked_waitq);
101 }
102}
103
104static void fuse_put_request(struct fuse_req *req);
105
106static struct fuse_req *fuse_get_req(struct fuse_mount *fm, bool for_background)
107{
108 struct fuse_conn *fc = fm->fc;
109 struct fuse_req *req;
110 int err;
111 atomic_inc(&fc->num_waiting);
112
113 if (fuse_block_alloc(fc, for_background)) {
114 err = -EINTR;
115 if (wait_event_killable_exclusive(fc->blocked_waitq,
116 !fuse_block_alloc(fc, for_background)))
117 goto out;
118 }
119 /* Matches smp_wmb() in fuse_set_initialized() */
120 smp_rmb();
121
122 err = -ENOTCONN;
123 if (!fc->connected)
124 goto out;
125
126 err = -ECONNREFUSED;
127 if (fc->conn_error)
128 goto out;
129
130 req = fuse_request_alloc(fm, GFP_KERNEL);
131 err = -ENOMEM;
132 if (!req) {
133 if (for_background)
134 wake_up(&fc->blocked_waitq);
135 goto out;
136 }
137
138 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid());
139 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid());
140 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
141
142 __set_bit(FR_WAITING, &req->flags);
143 if (for_background)
144 __set_bit(FR_BACKGROUND, &req->flags);
145
146 if (unlikely(req->in.h.uid == ((uid_t)-1) ||
147 req->in.h.gid == ((gid_t)-1))) {
148 fuse_put_request(req);
149 return ERR_PTR(-EOVERFLOW);
150 }
151 return req;
152
153 out:
154 fuse_drop_waiting(fc);
155 return ERR_PTR(err);
156}
157
158static void fuse_put_request(struct fuse_req *req)
159{
160 struct fuse_conn *fc = req->fm->fc;
161
162 if (refcount_dec_and_test(&req->count)) {
163 if (test_bit(FR_BACKGROUND, &req->flags)) {
164 /*
165 * We get here in the unlikely case that a background
166 * request was allocated but not sent
167 */
168 spin_lock(&fc->bg_lock);
169 if (!fc->blocked)
170 wake_up(&fc->blocked_waitq);
171 spin_unlock(&fc->bg_lock);
172 }
173
174 if (test_bit(FR_WAITING, &req->flags)) {
175 __clear_bit(FR_WAITING, &req->flags);
176 fuse_drop_waiting(fc);
177 }
178
179 fuse_request_free(req);
180 }
181}
182
183unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args)
184{
185 unsigned nbytes = 0;
186 unsigned i;
187
188 for (i = 0; i < numargs; i++)
189 nbytes += args[i].size;
190
191 return nbytes;
192}
193EXPORT_SYMBOL_GPL(fuse_len_args);
194
195u64 fuse_get_unique(struct fuse_iqueue *fiq)
196{
197 fiq->reqctr += FUSE_REQ_ID_STEP;
198 return fiq->reqctr;
199}
200EXPORT_SYMBOL_GPL(fuse_get_unique);
201
202static unsigned int fuse_req_hash(u64 unique)
203{
204 return hash_long(unique & ~FUSE_INT_REQ_BIT, FUSE_PQ_HASH_BITS);
205}
206
207/**
208 * A new request is available, wake fiq->waitq
209 */
210static void fuse_dev_wake_and_unlock(struct fuse_iqueue *fiq)
211__releases(fiq->lock)
212{
213 wake_up(&fiq->waitq);
214 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
215 spin_unlock(&fiq->lock);
216}
217
218const struct fuse_iqueue_ops fuse_dev_fiq_ops = {
219 .wake_forget_and_unlock = fuse_dev_wake_and_unlock,
220 .wake_interrupt_and_unlock = fuse_dev_wake_and_unlock,
221 .wake_pending_and_unlock = fuse_dev_wake_and_unlock,
222};
223EXPORT_SYMBOL_GPL(fuse_dev_fiq_ops);
224
225static void queue_request_and_unlock(struct fuse_iqueue *fiq,
226 struct fuse_req *req)
227__releases(fiq->lock)
228{
229 req->in.h.len = sizeof(struct fuse_in_header) +
230 fuse_len_args(req->args->in_numargs,
231 (struct fuse_arg *) req->args->in_args);
232 list_add_tail(&req->list, &fiq->pending);
233 fiq->ops->wake_pending_and_unlock(fiq);
234}
235
236void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
237 u64 nodeid, u64 nlookup)
238{
239 struct fuse_iqueue *fiq = &fc->iq;
240
241 forget->forget_one.nodeid = nodeid;
242 forget->forget_one.nlookup = nlookup;
243
244 spin_lock(&fiq->lock);
245 if (fiq->connected) {
246 fiq->forget_list_tail->next = forget;
247 fiq->forget_list_tail = forget;
248 fiq->ops->wake_forget_and_unlock(fiq);
249 } else {
250 kfree(forget);
251 spin_unlock(&fiq->lock);
252 }
253}
254
255static void flush_bg_queue(struct fuse_conn *fc)
256{
257 struct fuse_iqueue *fiq = &fc->iq;
258
259 while (fc->active_background < fc->max_background &&
260 !list_empty(&fc->bg_queue)) {
261 struct fuse_req *req;
262
263 req = list_first_entry(&fc->bg_queue, struct fuse_req, list);
264 list_del(&req->list);
265 fc->active_background++;
266 spin_lock(&fiq->lock);
267 req->in.h.unique = fuse_get_unique(fiq);
268 queue_request_and_unlock(fiq, req);
269 }
270}
271
272/*
273 * This function is called when a request is finished. Either a reply
274 * has arrived or it was aborted (and not yet sent) or some error
275 * occurred during communication with userspace, or the device file
276 * was closed. The requester thread is woken up (if still waiting),
277 * the 'end' callback is called if given, else the reference to the
278 * request is released
279 */
280void fuse_request_end(struct fuse_req *req)
281{
282 struct fuse_mount *fm = req->fm;
283 struct fuse_conn *fc = fm->fc;
284 struct fuse_iqueue *fiq = &fc->iq;
285
286 if (test_and_set_bit(FR_FINISHED, &req->flags))
287 goto put_request;
288
289 /*
290 * test_and_set_bit() implies smp_mb() between bit
291 * changing and below FR_INTERRUPTED check. Pairs with
292 * smp_mb() from queue_interrupt().
293 */
294 if (test_bit(FR_INTERRUPTED, &req->flags)) {
295 spin_lock(&fiq->lock);
296 list_del_init(&req->intr_entry);
297 spin_unlock(&fiq->lock);
298 }
299 WARN_ON(test_bit(FR_PENDING, &req->flags));
300 WARN_ON(test_bit(FR_SENT, &req->flags));
301 if (test_bit(FR_BACKGROUND, &req->flags)) {
302 spin_lock(&fc->bg_lock);
303 clear_bit(FR_BACKGROUND, &req->flags);
304 if (fc->num_background == fc->max_background) {
305 fc->blocked = 0;
306 wake_up(&fc->blocked_waitq);
307 } else if (!fc->blocked) {
308 /*
309 * Wake up next waiter, if any. It's okay to use
310 * waitqueue_active(), as we've already synced up
311 * fc->blocked with waiters with the wake_up() call
312 * above.
313 */
314 if (waitqueue_active(&fc->blocked_waitq))
315 wake_up(&fc->blocked_waitq);
316 }
317
318 fc->num_background--;
319 fc->active_background--;
320 flush_bg_queue(fc);
321 spin_unlock(&fc->bg_lock);
322 } else {
323 /* Wake up waiter sleeping in request_wait_answer() */
324 wake_up(&req->waitq);
325 }
326
327 if (test_bit(FR_ASYNC, &req->flags))
328 req->args->end(fm, req->args, req->out.h.error);
329put_request:
330 fuse_put_request(req);
331}
332EXPORT_SYMBOL_GPL(fuse_request_end);
333
334static int queue_interrupt(struct fuse_req *req)
335{
336 struct fuse_iqueue *fiq = &req->fm->fc->iq;
337
338 spin_lock(&fiq->lock);
339 /* Check for we've sent request to interrupt this req */
340 if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) {
341 spin_unlock(&fiq->lock);
342 return -EINVAL;
343 }
344
345 if (list_empty(&req->intr_entry)) {
346 list_add_tail(&req->intr_entry, &fiq->interrupts);
347 /*
348 * Pairs with smp_mb() implied by test_and_set_bit()
349 * from fuse_request_end().
350 */
351 smp_mb();
352 if (test_bit(FR_FINISHED, &req->flags)) {
353 list_del_init(&req->intr_entry);
354 spin_unlock(&fiq->lock);
355 return 0;
356 }
357 fiq->ops->wake_interrupt_and_unlock(fiq);
358 } else {
359 spin_unlock(&fiq->lock);
360 }
361 return 0;
362}
363
364static void request_wait_answer(struct fuse_req *req)
365{
366 struct fuse_conn *fc = req->fm->fc;
367 struct fuse_iqueue *fiq = &fc->iq;
368 int err;
369
370 if (!fc->no_interrupt) {
371 /* Any signal may interrupt this */
372 err = wait_event_interruptible(req->waitq,
373 test_bit(FR_FINISHED, &req->flags));
374 if (!err)
375 return;
376
377 set_bit(FR_INTERRUPTED, &req->flags);
378 /* matches barrier in fuse_dev_do_read() */
379 smp_mb__after_atomic();
380 if (test_bit(FR_SENT, &req->flags))
381 queue_interrupt(req);
382 }
383
384 if (!test_bit(FR_FORCE, &req->flags)) {
385 /* Only fatal signals may interrupt this */
386 err = wait_event_killable(req->waitq,
387 test_bit(FR_FINISHED, &req->flags));
388 if (!err)
389 return;
390
391 spin_lock(&fiq->lock);
392 /* Request is not yet in userspace, bail out */
393 if (test_bit(FR_PENDING, &req->flags)) {
394 list_del(&req->list);
395 spin_unlock(&fiq->lock);
396 __fuse_put_request(req);
397 req->out.h.error = -EINTR;
398 return;
399 }
400 spin_unlock(&fiq->lock);
401 }
402
403 /*
404 * Either request is already in userspace, or it was forced.
405 * Wait it out.
406 */
407 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
408}
409
410static void __fuse_request_send(struct fuse_req *req)
411{
412 struct fuse_iqueue *fiq = &req->fm->fc->iq;
413
414 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
415 spin_lock(&fiq->lock);
416 if (!fiq->connected) {
417 spin_unlock(&fiq->lock);
418 req->out.h.error = -ENOTCONN;
419 } else {
420 req->in.h.unique = fuse_get_unique(fiq);
421 /* acquire extra reference, since request is still needed
422 after fuse_request_end() */
423 __fuse_get_request(req);
424 queue_request_and_unlock(fiq, req);
425
426 request_wait_answer(req);
427 /* Pairs with smp_wmb() in fuse_request_end() */
428 smp_rmb();
429 }
430}
431
432static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
433{
434 if (fc->minor < 4 && args->opcode == FUSE_STATFS)
435 args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE;
436
437 if (fc->minor < 9) {
438 switch (args->opcode) {
439 case FUSE_LOOKUP:
440 case FUSE_CREATE:
441 case FUSE_MKNOD:
442 case FUSE_MKDIR:
443 case FUSE_SYMLINK:
444 case FUSE_LINK:
445 args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
446 break;
447 case FUSE_GETATTR:
448 case FUSE_SETATTR:
449 args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
450 break;
451 }
452 }
453 if (fc->minor < 12) {
454 switch (args->opcode) {
455 case FUSE_CREATE:
456 args->in_args[0].size = sizeof(struct fuse_open_in);
457 break;
458 case FUSE_MKNOD:
459 args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
460 break;
461 }
462 }
463}
464
465static void fuse_force_creds(struct fuse_req *req)
466{
467 struct fuse_conn *fc = req->fm->fc;
468
469 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid());
470 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid());
471 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
472}
473
474static void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args)
475{
476 req->in.h.opcode = args->opcode;
477 req->in.h.nodeid = args->nodeid;
478 req->args = args;
479 if (args->end)
480 __set_bit(FR_ASYNC, &req->flags);
481}
482
483ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args)
484{
485 struct fuse_conn *fc = fm->fc;
486 struct fuse_req *req;
487 ssize_t ret;
488
489 if (args->force) {
490 atomic_inc(&fc->num_waiting);
491 req = fuse_request_alloc(fm, GFP_KERNEL | __GFP_NOFAIL);
492
493 if (!args->nocreds)
494 fuse_force_creds(req);
495
496 __set_bit(FR_WAITING, &req->flags);
497 __set_bit(FR_FORCE, &req->flags);
498 } else {
499 WARN_ON(args->nocreds);
500 req = fuse_get_req(fm, false);
501 if (IS_ERR(req))
502 return PTR_ERR(req);
503 }
504
505 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
506 fuse_adjust_compat(fc, args);
507 fuse_args_to_req(req, args);
508
509 if (!args->noreply)
510 __set_bit(FR_ISREPLY, &req->flags);
511 __fuse_request_send(req);
512 ret = req->out.h.error;
513 if (!ret && args->out_argvar) {
514 BUG_ON(args->out_numargs == 0);
515 ret = args->out_args[args->out_numargs - 1].size;
516 }
517 fuse_put_request(req);
518
519 return ret;
520}
521
522static bool fuse_request_queue_background(struct fuse_req *req)
523{
524 struct fuse_mount *fm = req->fm;
525 struct fuse_conn *fc = fm->fc;
526 bool queued = false;
527
528 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags));
529 if (!test_bit(FR_WAITING, &req->flags)) {
530 __set_bit(FR_WAITING, &req->flags);
531 atomic_inc(&fc->num_waiting);
532 }
533 __set_bit(FR_ISREPLY, &req->flags);
534 spin_lock(&fc->bg_lock);
535 if (likely(fc->connected)) {
536 fc->num_background++;
537 if (fc->num_background == fc->max_background)
538 fc->blocked = 1;
539 list_add_tail(&req->list, &fc->bg_queue);
540 flush_bg_queue(fc);
541 queued = true;
542 }
543 spin_unlock(&fc->bg_lock);
544
545 return queued;
546}
547
548int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args,
549 gfp_t gfp_flags)
550{
551 struct fuse_req *req;
552
553 if (args->force) {
554 WARN_ON(!args->nocreds);
555 req = fuse_request_alloc(fm, gfp_flags);
556 if (!req)
557 return -ENOMEM;
558 __set_bit(FR_BACKGROUND, &req->flags);
559 } else {
560 WARN_ON(args->nocreds);
561 req = fuse_get_req(fm, true);
562 if (IS_ERR(req))
563 return PTR_ERR(req);
564 }
565
566 fuse_args_to_req(req, args);
567
568 if (!fuse_request_queue_background(req)) {
569 fuse_put_request(req);
570 return -ENOTCONN;
571 }
572
573 return 0;
574}
575EXPORT_SYMBOL_GPL(fuse_simple_background);
576
577static int fuse_simple_notify_reply(struct fuse_mount *fm,
578 struct fuse_args *args, u64 unique)
579{
580 struct fuse_req *req;
581 struct fuse_iqueue *fiq = &fm->fc->iq;
582 int err = 0;
583
584 req = fuse_get_req(fm, false);
585 if (IS_ERR(req))
586 return PTR_ERR(req);
587
588 __clear_bit(FR_ISREPLY, &req->flags);
589 req->in.h.unique = unique;
590
591 fuse_args_to_req(req, args);
592
593 spin_lock(&fiq->lock);
594 if (fiq->connected) {
595 queue_request_and_unlock(fiq, req);
596 } else {
597 err = -ENODEV;
598 spin_unlock(&fiq->lock);
599 fuse_put_request(req);
600 }
601
602 return err;
603}
604
605/*
606 * Lock the request. Up to the next unlock_request() there mustn't be
607 * anything that could cause a page-fault. If the request was already
608 * aborted bail out.
609 */
610static int lock_request(struct fuse_req *req)
611{
612 int err = 0;
613 if (req) {
614 spin_lock(&req->waitq.lock);
615 if (test_bit(FR_ABORTED, &req->flags))
616 err = -ENOENT;
617 else
618 set_bit(FR_LOCKED, &req->flags);
619 spin_unlock(&req->waitq.lock);
620 }
621 return err;
622}
623
624/*
625 * Unlock request. If it was aborted while locked, caller is responsible
626 * for unlocking and ending the request.
627 */
628static int unlock_request(struct fuse_req *req)
629{
630 int err = 0;
631 if (req) {
632 spin_lock(&req->waitq.lock);
633 if (test_bit(FR_ABORTED, &req->flags))
634 err = -ENOENT;
635 else
636 clear_bit(FR_LOCKED, &req->flags);
637 spin_unlock(&req->waitq.lock);
638 }
639 return err;
640}
641
642struct fuse_copy_state {
643 int write;
644 struct fuse_req *req;
645 struct iov_iter *iter;
646 struct pipe_buffer *pipebufs;
647 struct pipe_buffer *currbuf;
648 struct pipe_inode_info *pipe;
649 unsigned long nr_segs;
650 struct page *pg;
651 unsigned len;
652 unsigned offset;
653 unsigned move_pages:1;
654};
655
656static void fuse_copy_init(struct fuse_copy_state *cs, int write,
657 struct iov_iter *iter)
658{
659 memset(cs, 0, sizeof(*cs));
660 cs->write = write;
661 cs->iter = iter;
662}
663
664/* Unmap and put previous page of userspace buffer */
665static void fuse_copy_finish(struct fuse_copy_state *cs)
666{
667 if (cs->currbuf) {
668 struct pipe_buffer *buf = cs->currbuf;
669
670 if (cs->write)
671 buf->len = PAGE_SIZE - cs->len;
672 cs->currbuf = NULL;
673 } else if (cs->pg) {
674 if (cs->write) {
675 flush_dcache_page(cs->pg);
676 set_page_dirty_lock(cs->pg);
677 }
678 put_page(cs->pg);
679 }
680 cs->pg = NULL;
681}
682
683/*
684 * Get another pagefull of userspace buffer, and map it to kernel
685 * address space, and lock request
686 */
687static int fuse_copy_fill(struct fuse_copy_state *cs)
688{
689 struct page *page;
690 int err;
691
692 err = unlock_request(cs->req);
693 if (err)
694 return err;
695
696 fuse_copy_finish(cs);
697 if (cs->pipebufs) {
698 struct pipe_buffer *buf = cs->pipebufs;
699
700 if (!cs->write) {
701 err = pipe_buf_confirm(cs->pipe, buf);
702 if (err)
703 return err;
704
705 BUG_ON(!cs->nr_segs);
706 cs->currbuf = buf;
707 cs->pg = buf->page;
708 cs->offset = buf->offset;
709 cs->len = buf->len;
710 cs->pipebufs++;
711 cs->nr_segs--;
712 } else {
713 if (cs->nr_segs >= cs->pipe->max_usage)
714 return -EIO;
715
716 page = alloc_page(GFP_HIGHUSER);
717 if (!page)
718 return -ENOMEM;
719
720 buf->page = page;
721 buf->offset = 0;
722 buf->len = 0;
723
724 cs->currbuf = buf;
725 cs->pg = page;
726 cs->offset = 0;
727 cs->len = PAGE_SIZE;
728 cs->pipebufs++;
729 cs->nr_segs++;
730 }
731 } else {
732 size_t off;
733 err = iov_iter_get_pages2(cs->iter, &page, PAGE_SIZE, 1, &off);
734 if (err < 0)
735 return err;
736 BUG_ON(!err);
737 cs->len = err;
738 cs->offset = off;
739 cs->pg = page;
740 }
741
742 return lock_request(cs->req);
743}
744
745/* Do as much copy to/from userspace buffer as we can */
746static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
747{
748 unsigned ncpy = min(*size, cs->len);
749 if (val) {
750 void *pgaddr = kmap_local_page(cs->pg);
751 void *buf = pgaddr + cs->offset;
752
753 if (cs->write)
754 memcpy(buf, *val, ncpy);
755 else
756 memcpy(*val, buf, ncpy);
757
758 kunmap_local(pgaddr);
759 *val += ncpy;
760 }
761 *size -= ncpy;
762 cs->len -= ncpy;
763 cs->offset += ncpy;
764 return ncpy;
765}
766
767static int fuse_check_folio(struct folio *folio)
768{
769 if (folio_mapped(folio) ||
770 folio->mapping != NULL ||
771 (folio->flags & PAGE_FLAGS_CHECK_AT_PREP &
772 ~(1 << PG_locked |
773 1 << PG_referenced |
774 1 << PG_uptodate |
775 1 << PG_lru |
776 1 << PG_active |
777 1 << PG_workingset |
778 1 << PG_reclaim |
779 1 << PG_waiters |
780 LRU_GEN_MASK | LRU_REFS_MASK))) {
781 dump_page(&folio->page, "fuse: trying to steal weird page");
782 return 1;
783 }
784 return 0;
785}
786
787static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
788{
789 int err;
790 struct folio *oldfolio = page_folio(*pagep);
791 struct folio *newfolio;
792 struct pipe_buffer *buf = cs->pipebufs;
793
794 folio_get(oldfolio);
795 err = unlock_request(cs->req);
796 if (err)
797 goto out_put_old;
798
799 fuse_copy_finish(cs);
800
801 err = pipe_buf_confirm(cs->pipe, buf);
802 if (err)
803 goto out_put_old;
804
805 BUG_ON(!cs->nr_segs);
806 cs->currbuf = buf;
807 cs->len = buf->len;
808 cs->pipebufs++;
809 cs->nr_segs--;
810
811 if (cs->len != PAGE_SIZE)
812 goto out_fallback;
813
814 if (!pipe_buf_try_steal(cs->pipe, buf))
815 goto out_fallback;
816
817 newfolio = page_folio(buf->page);
818
819 if (!folio_test_uptodate(newfolio))
820 folio_mark_uptodate(newfolio);
821
822 folio_clear_mappedtodisk(newfolio);
823
824 if (fuse_check_folio(newfolio) != 0)
825 goto out_fallback_unlock;
826
827 /*
828 * This is a new and locked page, it shouldn't be mapped or
829 * have any special flags on it
830 */
831 if (WARN_ON(folio_mapped(oldfolio)))
832 goto out_fallback_unlock;
833 if (WARN_ON(folio_has_private(oldfolio)))
834 goto out_fallback_unlock;
835 if (WARN_ON(folio_test_dirty(oldfolio) ||
836 folio_test_writeback(oldfolio)))
837 goto out_fallback_unlock;
838 if (WARN_ON(folio_test_mlocked(oldfolio)))
839 goto out_fallback_unlock;
840
841 replace_page_cache_folio(oldfolio, newfolio);
842
843 folio_get(newfolio);
844
845 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
846 folio_add_lru(newfolio);
847
848 /*
849 * Release while we have extra ref on stolen page. Otherwise
850 * anon_pipe_buf_release() might think the page can be reused.
851 */
852 pipe_buf_release(cs->pipe, buf);
853
854 err = 0;
855 spin_lock(&cs->req->waitq.lock);
856 if (test_bit(FR_ABORTED, &cs->req->flags))
857 err = -ENOENT;
858 else
859 *pagep = &newfolio->page;
860 spin_unlock(&cs->req->waitq.lock);
861
862 if (err) {
863 folio_unlock(newfolio);
864 folio_put(newfolio);
865 goto out_put_old;
866 }
867
868 folio_unlock(oldfolio);
869 /* Drop ref for ap->pages[] array */
870 folio_put(oldfolio);
871 cs->len = 0;
872
873 err = 0;
874out_put_old:
875 /* Drop ref obtained in this function */
876 folio_put(oldfolio);
877 return err;
878
879out_fallback_unlock:
880 folio_unlock(newfolio);
881out_fallback:
882 cs->pg = buf->page;
883 cs->offset = buf->offset;
884
885 err = lock_request(cs->req);
886 if (!err)
887 err = 1;
888
889 goto out_put_old;
890}
891
892static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
893 unsigned offset, unsigned count)
894{
895 struct pipe_buffer *buf;
896 int err;
897
898 if (cs->nr_segs >= cs->pipe->max_usage)
899 return -EIO;
900
901 get_page(page);
902 err = unlock_request(cs->req);
903 if (err) {
904 put_page(page);
905 return err;
906 }
907
908 fuse_copy_finish(cs);
909
910 buf = cs->pipebufs;
911 buf->page = page;
912 buf->offset = offset;
913 buf->len = count;
914
915 cs->pipebufs++;
916 cs->nr_segs++;
917 cs->len = 0;
918
919 return 0;
920}
921
922/*
923 * Copy a page in the request to/from the userspace buffer. Must be
924 * done atomically
925 */
926static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
927 unsigned offset, unsigned count, int zeroing)
928{
929 int err;
930 struct page *page = *pagep;
931
932 if (page && zeroing && count < PAGE_SIZE)
933 clear_highpage(page);
934
935 while (count) {
936 if (cs->write && cs->pipebufs && page) {
937 /*
938 * Can't control lifetime of pipe buffers, so always
939 * copy user pages.
940 */
941 if (cs->req->args->user_pages) {
942 err = fuse_copy_fill(cs);
943 if (err)
944 return err;
945 } else {
946 return fuse_ref_page(cs, page, offset, count);
947 }
948 } else if (!cs->len) {
949 if (cs->move_pages && page &&
950 offset == 0 && count == PAGE_SIZE) {
951 err = fuse_try_move_page(cs, pagep);
952 if (err <= 0)
953 return err;
954 } else {
955 err = fuse_copy_fill(cs);
956 if (err)
957 return err;
958 }
959 }
960 if (page) {
961 void *mapaddr = kmap_local_page(page);
962 void *buf = mapaddr + offset;
963 offset += fuse_copy_do(cs, &buf, &count);
964 kunmap_local(mapaddr);
965 } else
966 offset += fuse_copy_do(cs, NULL, &count);
967 }
968 if (page && !cs->write)
969 flush_dcache_page(page);
970 return 0;
971}
972
973/* Copy pages in the request to/from userspace buffer */
974static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
975 int zeroing)
976{
977 unsigned i;
978 struct fuse_req *req = cs->req;
979 struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args);
980
981
982 for (i = 0; i < ap->num_pages && (nbytes || zeroing); i++) {
983 int err;
984 unsigned int offset = ap->descs[i].offset;
985 unsigned int count = min(nbytes, ap->descs[i].length);
986
987 err = fuse_copy_page(cs, &ap->pages[i], offset, count, zeroing);
988 if (err)
989 return err;
990
991 nbytes -= count;
992 }
993 return 0;
994}
995
996/* Copy a single argument in the request to/from userspace buffer */
997static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
998{
999 while (size) {
1000 if (!cs->len) {
1001 int err = fuse_copy_fill(cs);
1002 if (err)
1003 return err;
1004 }
1005 fuse_copy_do(cs, &val, &size);
1006 }
1007 return 0;
1008}
1009
1010/* Copy request arguments to/from userspace buffer */
1011static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1012 unsigned argpages, struct fuse_arg *args,
1013 int zeroing)
1014{
1015 int err = 0;
1016 unsigned i;
1017
1018 for (i = 0; !err && i < numargs; i++) {
1019 struct fuse_arg *arg = &args[i];
1020 if (i == numargs - 1 && argpages)
1021 err = fuse_copy_pages(cs, arg->size, zeroing);
1022 else
1023 err = fuse_copy_one(cs, arg->value, arg->size);
1024 }
1025 return err;
1026}
1027
1028static int forget_pending(struct fuse_iqueue *fiq)
1029{
1030 return fiq->forget_list_head.next != NULL;
1031}
1032
1033static int request_pending(struct fuse_iqueue *fiq)
1034{
1035 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1036 forget_pending(fiq);
1037}
1038
1039/*
1040 * Transfer an interrupt request to userspace
1041 *
1042 * Unlike other requests this is assembled on demand, without a need
1043 * to allocate a separate fuse_req structure.
1044 *
1045 * Called with fiq->lock held, releases it
1046 */
1047static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1048 struct fuse_copy_state *cs,
1049 size_t nbytes, struct fuse_req *req)
1050__releases(fiq->lock)
1051{
1052 struct fuse_in_header ih;
1053 struct fuse_interrupt_in arg;
1054 unsigned reqsize = sizeof(ih) + sizeof(arg);
1055 int err;
1056
1057 list_del_init(&req->intr_entry);
1058 memset(&ih, 0, sizeof(ih));
1059 memset(&arg, 0, sizeof(arg));
1060 ih.len = reqsize;
1061 ih.opcode = FUSE_INTERRUPT;
1062 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT);
1063 arg.unique = req->in.h.unique;
1064
1065 spin_unlock(&fiq->lock);
1066 if (nbytes < reqsize)
1067 return -EINVAL;
1068
1069 err = fuse_copy_one(cs, &ih, sizeof(ih));
1070 if (!err)
1071 err = fuse_copy_one(cs, &arg, sizeof(arg));
1072 fuse_copy_finish(cs);
1073
1074 return err ? err : reqsize;
1075}
1076
1077struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq,
1078 unsigned int max,
1079 unsigned int *countp)
1080{
1081 struct fuse_forget_link *head = fiq->forget_list_head.next;
1082 struct fuse_forget_link **newhead = &head;
1083 unsigned count;
1084
1085 for (count = 0; *newhead != NULL && count < max; count++)
1086 newhead = &(*newhead)->next;
1087
1088 fiq->forget_list_head.next = *newhead;
1089 *newhead = NULL;
1090 if (fiq->forget_list_head.next == NULL)
1091 fiq->forget_list_tail = &fiq->forget_list_head;
1092
1093 if (countp != NULL)
1094 *countp = count;
1095
1096 return head;
1097}
1098EXPORT_SYMBOL(fuse_dequeue_forget);
1099
1100static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1101 struct fuse_copy_state *cs,
1102 size_t nbytes)
1103__releases(fiq->lock)
1104{
1105 int err;
1106 struct fuse_forget_link *forget = fuse_dequeue_forget(fiq, 1, NULL);
1107 struct fuse_forget_in arg = {
1108 .nlookup = forget->forget_one.nlookup,
1109 };
1110 struct fuse_in_header ih = {
1111 .opcode = FUSE_FORGET,
1112 .nodeid = forget->forget_one.nodeid,
1113 .unique = fuse_get_unique(fiq),
1114 .len = sizeof(ih) + sizeof(arg),
1115 };
1116
1117 spin_unlock(&fiq->lock);
1118 kfree(forget);
1119 if (nbytes < ih.len)
1120 return -EINVAL;
1121
1122 err = fuse_copy_one(cs, &ih, sizeof(ih));
1123 if (!err)
1124 err = fuse_copy_one(cs, &arg, sizeof(arg));
1125 fuse_copy_finish(cs);
1126
1127 if (err)
1128 return err;
1129
1130 return ih.len;
1131}
1132
1133static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1134 struct fuse_copy_state *cs, size_t nbytes)
1135__releases(fiq->lock)
1136{
1137 int err;
1138 unsigned max_forgets;
1139 unsigned count;
1140 struct fuse_forget_link *head;
1141 struct fuse_batch_forget_in arg = { .count = 0 };
1142 struct fuse_in_header ih = {
1143 .opcode = FUSE_BATCH_FORGET,
1144 .unique = fuse_get_unique(fiq),
1145 .len = sizeof(ih) + sizeof(arg),
1146 };
1147
1148 if (nbytes < ih.len) {
1149 spin_unlock(&fiq->lock);
1150 return -EINVAL;
1151 }
1152
1153 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1154 head = fuse_dequeue_forget(fiq, max_forgets, &count);
1155 spin_unlock(&fiq->lock);
1156
1157 arg.count = count;
1158 ih.len += count * sizeof(struct fuse_forget_one);
1159 err = fuse_copy_one(cs, &ih, sizeof(ih));
1160 if (!err)
1161 err = fuse_copy_one(cs, &arg, sizeof(arg));
1162
1163 while (head) {
1164 struct fuse_forget_link *forget = head;
1165
1166 if (!err) {
1167 err = fuse_copy_one(cs, &forget->forget_one,
1168 sizeof(forget->forget_one));
1169 }
1170 head = forget->next;
1171 kfree(forget);
1172 }
1173
1174 fuse_copy_finish(cs);
1175
1176 if (err)
1177 return err;
1178
1179 return ih.len;
1180}
1181
1182static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1183 struct fuse_copy_state *cs,
1184 size_t nbytes)
1185__releases(fiq->lock)
1186{
1187 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1188 return fuse_read_single_forget(fiq, cs, nbytes);
1189 else
1190 return fuse_read_batch_forget(fiq, cs, nbytes);
1191}
1192
1193/*
1194 * Read a single request into the userspace filesystem's buffer. This
1195 * function waits until a request is available, then removes it from
1196 * the pending list and copies request data to userspace buffer. If
1197 * no reply is needed (FORGET) or request has been aborted or there
1198 * was an error during the copying then it's finished by calling
1199 * fuse_request_end(). Otherwise add it to the processing list, and set
1200 * the 'sent' flag.
1201 */
1202static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1203 struct fuse_copy_state *cs, size_t nbytes)
1204{
1205 ssize_t err;
1206 struct fuse_conn *fc = fud->fc;
1207 struct fuse_iqueue *fiq = &fc->iq;
1208 struct fuse_pqueue *fpq = &fud->pq;
1209 struct fuse_req *req;
1210 struct fuse_args *args;
1211 unsigned reqsize;
1212 unsigned int hash;
1213
1214 /*
1215 * Require sane minimum read buffer - that has capacity for fixed part
1216 * of any request header + negotiated max_write room for data.
1217 *
1218 * Historically libfuse reserves 4K for fixed header room, but e.g.
1219 * GlusterFS reserves only 80 bytes
1220 *
1221 * = `sizeof(fuse_in_header) + sizeof(fuse_write_in)`
1222 *
1223 * which is the absolute minimum any sane filesystem should be using
1224 * for header room.
1225 */
1226 if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER,
1227 sizeof(struct fuse_in_header) +
1228 sizeof(struct fuse_write_in) +
1229 fc->max_write))
1230 return -EINVAL;
1231
1232 restart:
1233 for (;;) {
1234 spin_lock(&fiq->lock);
1235 if (!fiq->connected || request_pending(fiq))
1236 break;
1237 spin_unlock(&fiq->lock);
1238
1239 if (file->f_flags & O_NONBLOCK)
1240 return -EAGAIN;
1241 err = wait_event_interruptible_exclusive(fiq->waitq,
1242 !fiq->connected || request_pending(fiq));
1243 if (err)
1244 return err;
1245 }
1246
1247 if (!fiq->connected) {
1248 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1249 goto err_unlock;
1250 }
1251
1252 if (!list_empty(&fiq->interrupts)) {
1253 req = list_entry(fiq->interrupts.next, struct fuse_req,
1254 intr_entry);
1255 return fuse_read_interrupt(fiq, cs, nbytes, req);
1256 }
1257
1258 if (forget_pending(fiq)) {
1259 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1260 return fuse_read_forget(fc, fiq, cs, nbytes);
1261
1262 if (fiq->forget_batch <= -8)
1263 fiq->forget_batch = 16;
1264 }
1265
1266 req = list_entry(fiq->pending.next, struct fuse_req, list);
1267 clear_bit(FR_PENDING, &req->flags);
1268 list_del_init(&req->list);
1269 spin_unlock(&fiq->lock);
1270
1271 args = req->args;
1272 reqsize = req->in.h.len;
1273
1274 /* If request is too large, reply with an error and restart the read */
1275 if (nbytes < reqsize) {
1276 req->out.h.error = -EIO;
1277 /* SETXATTR is special, since it may contain too large data */
1278 if (args->opcode == FUSE_SETXATTR)
1279 req->out.h.error = -E2BIG;
1280 fuse_request_end(req);
1281 goto restart;
1282 }
1283 spin_lock(&fpq->lock);
1284 /*
1285 * Must not put request on fpq->io queue after having been shut down by
1286 * fuse_abort_conn()
1287 */
1288 if (!fpq->connected) {
1289 req->out.h.error = err = -ECONNABORTED;
1290 goto out_end;
1291
1292 }
1293 list_add(&req->list, &fpq->io);
1294 spin_unlock(&fpq->lock);
1295 cs->req = req;
1296 err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h));
1297 if (!err)
1298 err = fuse_copy_args(cs, args->in_numargs, args->in_pages,
1299 (struct fuse_arg *) args->in_args, 0);
1300 fuse_copy_finish(cs);
1301 spin_lock(&fpq->lock);
1302 clear_bit(FR_LOCKED, &req->flags);
1303 if (!fpq->connected) {
1304 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1305 goto out_end;
1306 }
1307 if (err) {
1308 req->out.h.error = -EIO;
1309 goto out_end;
1310 }
1311 if (!test_bit(FR_ISREPLY, &req->flags)) {
1312 err = reqsize;
1313 goto out_end;
1314 }
1315 hash = fuse_req_hash(req->in.h.unique);
1316 list_move_tail(&req->list, &fpq->processing[hash]);
1317 __fuse_get_request(req);
1318 set_bit(FR_SENT, &req->flags);
1319 spin_unlock(&fpq->lock);
1320 /* matches barrier in request_wait_answer() */
1321 smp_mb__after_atomic();
1322 if (test_bit(FR_INTERRUPTED, &req->flags))
1323 queue_interrupt(req);
1324 fuse_put_request(req);
1325
1326 return reqsize;
1327
1328out_end:
1329 if (!test_bit(FR_PRIVATE, &req->flags))
1330 list_del_init(&req->list);
1331 spin_unlock(&fpq->lock);
1332 fuse_request_end(req);
1333 return err;
1334
1335 err_unlock:
1336 spin_unlock(&fiq->lock);
1337 return err;
1338}
1339
1340static int fuse_dev_open(struct inode *inode, struct file *file)
1341{
1342 /*
1343 * The fuse device's file's private_data is used to hold
1344 * the fuse_conn(ection) when it is mounted, and is used to
1345 * keep track of whether the file has been mounted already.
1346 */
1347 file->private_data = NULL;
1348 return 0;
1349}
1350
1351static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1352{
1353 struct fuse_copy_state cs;
1354 struct file *file = iocb->ki_filp;
1355 struct fuse_dev *fud = fuse_get_dev(file);
1356
1357 if (!fud)
1358 return -EPERM;
1359
1360 if (!user_backed_iter(to))
1361 return -EINVAL;
1362
1363 fuse_copy_init(&cs, 1, to);
1364
1365 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1366}
1367
1368static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1369 struct pipe_inode_info *pipe,
1370 size_t len, unsigned int flags)
1371{
1372 int total, ret;
1373 int page_nr = 0;
1374 struct pipe_buffer *bufs;
1375 struct fuse_copy_state cs;
1376 struct fuse_dev *fud = fuse_get_dev(in);
1377
1378 if (!fud)
1379 return -EPERM;
1380
1381 bufs = kvmalloc_array(pipe->max_usage, sizeof(struct pipe_buffer),
1382 GFP_KERNEL);
1383 if (!bufs)
1384 return -ENOMEM;
1385
1386 fuse_copy_init(&cs, 1, NULL);
1387 cs.pipebufs = bufs;
1388 cs.pipe = pipe;
1389 ret = fuse_dev_do_read(fud, in, &cs, len);
1390 if (ret < 0)
1391 goto out;
1392
1393 if (pipe_occupancy(pipe->head, pipe->tail) + cs.nr_segs > pipe->max_usage) {
1394 ret = -EIO;
1395 goto out;
1396 }
1397
1398 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1399 /*
1400 * Need to be careful about this. Having buf->ops in module
1401 * code can Oops if the buffer persists after module unload.
1402 */
1403 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1404 bufs[page_nr].flags = 0;
1405 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1406 if (unlikely(ret < 0))
1407 break;
1408 }
1409 if (total)
1410 ret = total;
1411out:
1412 for (; page_nr < cs.nr_segs; page_nr++)
1413 put_page(bufs[page_nr].page);
1414
1415 kvfree(bufs);
1416 return ret;
1417}
1418
1419static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1420 struct fuse_copy_state *cs)
1421{
1422 struct fuse_notify_poll_wakeup_out outarg;
1423 int err = -EINVAL;
1424
1425 if (size != sizeof(outarg))
1426 goto err;
1427
1428 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1429 if (err)
1430 goto err;
1431
1432 fuse_copy_finish(cs);
1433 return fuse_notify_poll_wakeup(fc, &outarg);
1434
1435err:
1436 fuse_copy_finish(cs);
1437 return err;
1438}
1439
1440static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1441 struct fuse_copy_state *cs)
1442{
1443 struct fuse_notify_inval_inode_out outarg;
1444 int err = -EINVAL;
1445
1446 if (size != sizeof(outarg))
1447 goto err;
1448
1449 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1450 if (err)
1451 goto err;
1452 fuse_copy_finish(cs);
1453
1454 down_read(&fc->killsb);
1455 err = fuse_reverse_inval_inode(fc, outarg.ino,
1456 outarg.off, outarg.len);
1457 up_read(&fc->killsb);
1458 return err;
1459
1460err:
1461 fuse_copy_finish(cs);
1462 return err;
1463}
1464
1465static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1466 struct fuse_copy_state *cs)
1467{
1468 struct fuse_notify_inval_entry_out outarg;
1469 int err = -ENOMEM;
1470 char *buf;
1471 struct qstr name;
1472
1473 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1474 if (!buf)
1475 goto err;
1476
1477 err = -EINVAL;
1478 if (size < sizeof(outarg))
1479 goto err;
1480
1481 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1482 if (err)
1483 goto err;
1484
1485 err = -ENAMETOOLONG;
1486 if (outarg.namelen > FUSE_NAME_MAX)
1487 goto err;
1488
1489 err = -EINVAL;
1490 if (size != sizeof(outarg) + outarg.namelen + 1)
1491 goto err;
1492
1493 name.name = buf;
1494 name.len = outarg.namelen;
1495 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1496 if (err)
1497 goto err;
1498 fuse_copy_finish(cs);
1499 buf[outarg.namelen] = 0;
1500
1501 down_read(&fc->killsb);
1502 err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name, outarg.flags);
1503 up_read(&fc->killsb);
1504 kfree(buf);
1505 return err;
1506
1507err:
1508 kfree(buf);
1509 fuse_copy_finish(cs);
1510 return err;
1511}
1512
1513static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1514 struct fuse_copy_state *cs)
1515{
1516 struct fuse_notify_delete_out outarg;
1517 int err = -ENOMEM;
1518 char *buf;
1519 struct qstr name;
1520
1521 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1522 if (!buf)
1523 goto err;
1524
1525 err = -EINVAL;
1526 if (size < sizeof(outarg))
1527 goto err;
1528
1529 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1530 if (err)
1531 goto err;
1532
1533 err = -ENAMETOOLONG;
1534 if (outarg.namelen > FUSE_NAME_MAX)
1535 goto err;
1536
1537 err = -EINVAL;
1538 if (size != sizeof(outarg) + outarg.namelen + 1)
1539 goto err;
1540
1541 name.name = buf;
1542 name.len = outarg.namelen;
1543 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1544 if (err)
1545 goto err;
1546 fuse_copy_finish(cs);
1547 buf[outarg.namelen] = 0;
1548
1549 down_read(&fc->killsb);
1550 err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name, 0);
1551 up_read(&fc->killsb);
1552 kfree(buf);
1553 return err;
1554
1555err:
1556 kfree(buf);
1557 fuse_copy_finish(cs);
1558 return err;
1559}
1560
1561static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1562 struct fuse_copy_state *cs)
1563{
1564 struct fuse_notify_store_out outarg;
1565 struct inode *inode;
1566 struct address_space *mapping;
1567 u64 nodeid;
1568 int err;
1569 pgoff_t index;
1570 unsigned int offset;
1571 unsigned int num;
1572 loff_t file_size;
1573 loff_t end;
1574
1575 err = -EINVAL;
1576 if (size < sizeof(outarg))
1577 goto out_finish;
1578
1579 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1580 if (err)
1581 goto out_finish;
1582
1583 err = -EINVAL;
1584 if (size - sizeof(outarg) != outarg.size)
1585 goto out_finish;
1586
1587 nodeid = outarg.nodeid;
1588
1589 down_read(&fc->killsb);
1590
1591 err = -ENOENT;
1592 inode = fuse_ilookup(fc, nodeid, NULL);
1593 if (!inode)
1594 goto out_up_killsb;
1595
1596 mapping = inode->i_mapping;
1597 index = outarg.offset >> PAGE_SHIFT;
1598 offset = outarg.offset & ~PAGE_MASK;
1599 file_size = i_size_read(inode);
1600 end = outarg.offset + outarg.size;
1601 if (end > file_size) {
1602 file_size = end;
1603 fuse_write_update_attr(inode, file_size, outarg.size);
1604 }
1605
1606 num = outarg.size;
1607 while (num) {
1608 struct page *page;
1609 unsigned int this_num;
1610
1611 err = -ENOMEM;
1612 page = find_or_create_page(mapping, index,
1613 mapping_gfp_mask(mapping));
1614 if (!page)
1615 goto out_iput;
1616
1617 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1618 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1619 if (!err && offset == 0 &&
1620 (this_num == PAGE_SIZE || file_size == end))
1621 SetPageUptodate(page);
1622 unlock_page(page);
1623 put_page(page);
1624
1625 if (err)
1626 goto out_iput;
1627
1628 num -= this_num;
1629 offset = 0;
1630 index++;
1631 }
1632
1633 err = 0;
1634
1635out_iput:
1636 iput(inode);
1637out_up_killsb:
1638 up_read(&fc->killsb);
1639out_finish:
1640 fuse_copy_finish(cs);
1641 return err;
1642}
1643
1644struct fuse_retrieve_args {
1645 struct fuse_args_pages ap;
1646 struct fuse_notify_retrieve_in inarg;
1647};
1648
1649static void fuse_retrieve_end(struct fuse_mount *fm, struct fuse_args *args,
1650 int error)
1651{
1652 struct fuse_retrieve_args *ra =
1653 container_of(args, typeof(*ra), ap.args);
1654
1655 release_pages(ra->ap.pages, ra->ap.num_pages);
1656 kfree(ra);
1657}
1658
1659static int fuse_retrieve(struct fuse_mount *fm, struct inode *inode,
1660 struct fuse_notify_retrieve_out *outarg)
1661{
1662 int err;
1663 struct address_space *mapping = inode->i_mapping;
1664 pgoff_t index;
1665 loff_t file_size;
1666 unsigned int num;
1667 unsigned int offset;
1668 size_t total_len = 0;
1669 unsigned int num_pages;
1670 struct fuse_conn *fc = fm->fc;
1671 struct fuse_retrieve_args *ra;
1672 size_t args_size = sizeof(*ra);
1673 struct fuse_args_pages *ap;
1674 struct fuse_args *args;
1675
1676 offset = outarg->offset & ~PAGE_MASK;
1677 file_size = i_size_read(inode);
1678
1679 num = min(outarg->size, fc->max_write);
1680 if (outarg->offset > file_size)
1681 num = 0;
1682 else if (outarg->offset + num > file_size)
1683 num = file_size - outarg->offset;
1684
1685 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1686 num_pages = min(num_pages, fc->max_pages);
1687
1688 args_size += num_pages * (sizeof(ap->pages[0]) + sizeof(ap->descs[0]));
1689
1690 ra = kzalloc(args_size, GFP_KERNEL);
1691 if (!ra)
1692 return -ENOMEM;
1693
1694 ap = &ra->ap;
1695 ap->pages = (void *) (ra + 1);
1696 ap->descs = (void *) (ap->pages + num_pages);
1697
1698 args = &ap->args;
1699 args->nodeid = outarg->nodeid;
1700 args->opcode = FUSE_NOTIFY_REPLY;
1701 args->in_numargs = 2;
1702 args->in_pages = true;
1703 args->end = fuse_retrieve_end;
1704
1705 index = outarg->offset >> PAGE_SHIFT;
1706
1707 while (num && ap->num_pages < num_pages) {
1708 struct page *page;
1709 unsigned int this_num;
1710
1711 page = find_get_page(mapping, index);
1712 if (!page)
1713 break;
1714
1715 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1716 ap->pages[ap->num_pages] = page;
1717 ap->descs[ap->num_pages].offset = offset;
1718 ap->descs[ap->num_pages].length = this_num;
1719 ap->num_pages++;
1720
1721 offset = 0;
1722 num -= this_num;
1723 total_len += this_num;
1724 index++;
1725 }
1726 ra->inarg.offset = outarg->offset;
1727 ra->inarg.size = total_len;
1728 args->in_args[0].size = sizeof(ra->inarg);
1729 args->in_args[0].value = &ra->inarg;
1730 args->in_args[1].size = total_len;
1731
1732 err = fuse_simple_notify_reply(fm, args, outarg->notify_unique);
1733 if (err)
1734 fuse_retrieve_end(fm, args, err);
1735
1736 return err;
1737}
1738
1739static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1740 struct fuse_copy_state *cs)
1741{
1742 struct fuse_notify_retrieve_out outarg;
1743 struct fuse_mount *fm;
1744 struct inode *inode;
1745 u64 nodeid;
1746 int err;
1747
1748 err = -EINVAL;
1749 if (size != sizeof(outarg))
1750 goto copy_finish;
1751
1752 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1753 if (err)
1754 goto copy_finish;
1755
1756 fuse_copy_finish(cs);
1757
1758 down_read(&fc->killsb);
1759 err = -ENOENT;
1760 nodeid = outarg.nodeid;
1761
1762 inode = fuse_ilookup(fc, nodeid, &fm);
1763 if (inode) {
1764 err = fuse_retrieve(fm, inode, &outarg);
1765 iput(inode);
1766 }
1767 up_read(&fc->killsb);
1768
1769 return err;
1770
1771copy_finish:
1772 fuse_copy_finish(cs);
1773 return err;
1774}
1775
1776static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1777 unsigned int size, struct fuse_copy_state *cs)
1778{
1779 /* Don't try to move pages (yet) */
1780 cs->move_pages = 0;
1781
1782 switch (code) {
1783 case FUSE_NOTIFY_POLL:
1784 return fuse_notify_poll(fc, size, cs);
1785
1786 case FUSE_NOTIFY_INVAL_INODE:
1787 return fuse_notify_inval_inode(fc, size, cs);
1788
1789 case FUSE_NOTIFY_INVAL_ENTRY:
1790 return fuse_notify_inval_entry(fc, size, cs);
1791
1792 case FUSE_NOTIFY_STORE:
1793 return fuse_notify_store(fc, size, cs);
1794
1795 case FUSE_NOTIFY_RETRIEVE:
1796 return fuse_notify_retrieve(fc, size, cs);
1797
1798 case FUSE_NOTIFY_DELETE:
1799 return fuse_notify_delete(fc, size, cs);
1800
1801 default:
1802 fuse_copy_finish(cs);
1803 return -EINVAL;
1804 }
1805}
1806
1807/* Look up request on processing list by unique ID */
1808static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1809{
1810 unsigned int hash = fuse_req_hash(unique);
1811 struct fuse_req *req;
1812
1813 list_for_each_entry(req, &fpq->processing[hash], list) {
1814 if (req->in.h.unique == unique)
1815 return req;
1816 }
1817 return NULL;
1818}
1819
1820static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args,
1821 unsigned nbytes)
1822{
1823 unsigned reqsize = sizeof(struct fuse_out_header);
1824
1825 reqsize += fuse_len_args(args->out_numargs, args->out_args);
1826
1827 if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar))
1828 return -EINVAL;
1829 else if (reqsize > nbytes) {
1830 struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1];
1831 unsigned diffsize = reqsize - nbytes;
1832
1833 if (diffsize > lastarg->size)
1834 return -EINVAL;
1835 lastarg->size -= diffsize;
1836 }
1837 return fuse_copy_args(cs, args->out_numargs, args->out_pages,
1838 args->out_args, args->page_zeroing);
1839}
1840
1841/*
1842 * Write a single reply to a request. First the header is copied from
1843 * the write buffer. The request is then searched on the processing
1844 * list by the unique ID found in the header. If found, then remove
1845 * it from the list and copy the rest of the buffer to the request.
1846 * The request is finished by calling fuse_request_end().
1847 */
1848static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1849 struct fuse_copy_state *cs, size_t nbytes)
1850{
1851 int err;
1852 struct fuse_conn *fc = fud->fc;
1853 struct fuse_pqueue *fpq = &fud->pq;
1854 struct fuse_req *req;
1855 struct fuse_out_header oh;
1856
1857 err = -EINVAL;
1858 if (nbytes < sizeof(struct fuse_out_header))
1859 goto out;
1860
1861 err = fuse_copy_one(cs, &oh, sizeof(oh));
1862 if (err)
1863 goto copy_finish;
1864
1865 err = -EINVAL;
1866 if (oh.len != nbytes)
1867 goto copy_finish;
1868
1869 /*
1870 * Zero oh.unique indicates unsolicited notification message
1871 * and error contains notification code.
1872 */
1873 if (!oh.unique) {
1874 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1875 goto out;
1876 }
1877
1878 err = -EINVAL;
1879 if (oh.error <= -512 || oh.error > 0)
1880 goto copy_finish;
1881
1882 spin_lock(&fpq->lock);
1883 req = NULL;
1884 if (fpq->connected)
1885 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
1886
1887 err = -ENOENT;
1888 if (!req) {
1889 spin_unlock(&fpq->lock);
1890 goto copy_finish;
1891 }
1892
1893 /* Is it an interrupt reply ID? */
1894 if (oh.unique & FUSE_INT_REQ_BIT) {
1895 __fuse_get_request(req);
1896 spin_unlock(&fpq->lock);
1897
1898 err = 0;
1899 if (nbytes != sizeof(struct fuse_out_header))
1900 err = -EINVAL;
1901 else if (oh.error == -ENOSYS)
1902 fc->no_interrupt = 1;
1903 else if (oh.error == -EAGAIN)
1904 err = queue_interrupt(req);
1905
1906 fuse_put_request(req);
1907
1908 goto copy_finish;
1909 }
1910
1911 clear_bit(FR_SENT, &req->flags);
1912 list_move(&req->list, &fpq->io);
1913 req->out.h = oh;
1914 set_bit(FR_LOCKED, &req->flags);
1915 spin_unlock(&fpq->lock);
1916 cs->req = req;
1917 if (!req->args->page_replace)
1918 cs->move_pages = 0;
1919
1920 if (oh.error)
1921 err = nbytes != sizeof(oh) ? -EINVAL : 0;
1922 else
1923 err = copy_out_args(cs, req->args, nbytes);
1924 fuse_copy_finish(cs);
1925
1926 spin_lock(&fpq->lock);
1927 clear_bit(FR_LOCKED, &req->flags);
1928 if (!fpq->connected)
1929 err = -ENOENT;
1930 else if (err)
1931 req->out.h.error = -EIO;
1932 if (!test_bit(FR_PRIVATE, &req->flags))
1933 list_del_init(&req->list);
1934 spin_unlock(&fpq->lock);
1935
1936 fuse_request_end(req);
1937out:
1938 return err ? err : nbytes;
1939
1940copy_finish:
1941 fuse_copy_finish(cs);
1942 goto out;
1943}
1944
1945static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1946{
1947 struct fuse_copy_state cs;
1948 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1949
1950 if (!fud)
1951 return -EPERM;
1952
1953 if (!user_backed_iter(from))
1954 return -EINVAL;
1955
1956 fuse_copy_init(&cs, 0, from);
1957
1958 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1959}
1960
1961static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1962 struct file *out, loff_t *ppos,
1963 size_t len, unsigned int flags)
1964{
1965 unsigned int head, tail, mask, count;
1966 unsigned nbuf;
1967 unsigned idx;
1968 struct pipe_buffer *bufs;
1969 struct fuse_copy_state cs;
1970 struct fuse_dev *fud;
1971 size_t rem;
1972 ssize_t ret;
1973
1974 fud = fuse_get_dev(out);
1975 if (!fud)
1976 return -EPERM;
1977
1978 pipe_lock(pipe);
1979
1980 head = pipe->head;
1981 tail = pipe->tail;
1982 mask = pipe->ring_size - 1;
1983 count = head - tail;
1984
1985 bufs = kvmalloc_array(count, sizeof(struct pipe_buffer), GFP_KERNEL);
1986 if (!bufs) {
1987 pipe_unlock(pipe);
1988 return -ENOMEM;
1989 }
1990
1991 nbuf = 0;
1992 rem = 0;
1993 for (idx = tail; idx != head && rem < len; idx++)
1994 rem += pipe->bufs[idx & mask].len;
1995
1996 ret = -EINVAL;
1997 if (rem < len)
1998 goto out_free;
1999
2000 rem = len;
2001 while (rem) {
2002 struct pipe_buffer *ibuf;
2003 struct pipe_buffer *obuf;
2004
2005 if (WARN_ON(nbuf >= count || tail == head))
2006 goto out_free;
2007
2008 ibuf = &pipe->bufs[tail & mask];
2009 obuf = &bufs[nbuf];
2010
2011 if (rem >= ibuf->len) {
2012 *obuf = *ibuf;
2013 ibuf->ops = NULL;
2014 tail++;
2015 pipe->tail = tail;
2016 } else {
2017 if (!pipe_buf_get(pipe, ibuf))
2018 goto out_free;
2019
2020 *obuf = *ibuf;
2021 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2022 obuf->len = rem;
2023 ibuf->offset += obuf->len;
2024 ibuf->len -= obuf->len;
2025 }
2026 nbuf++;
2027 rem -= obuf->len;
2028 }
2029 pipe_unlock(pipe);
2030
2031 fuse_copy_init(&cs, 0, NULL);
2032 cs.pipebufs = bufs;
2033 cs.nr_segs = nbuf;
2034 cs.pipe = pipe;
2035
2036 if (flags & SPLICE_F_MOVE)
2037 cs.move_pages = 1;
2038
2039 ret = fuse_dev_do_write(fud, &cs, len);
2040
2041 pipe_lock(pipe);
2042out_free:
2043 for (idx = 0; idx < nbuf; idx++) {
2044 struct pipe_buffer *buf = &bufs[idx];
2045
2046 if (buf->ops)
2047 pipe_buf_release(pipe, buf);
2048 }
2049 pipe_unlock(pipe);
2050
2051 kvfree(bufs);
2052 return ret;
2053}
2054
2055static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2056{
2057 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
2058 struct fuse_iqueue *fiq;
2059 struct fuse_dev *fud = fuse_get_dev(file);
2060
2061 if (!fud)
2062 return EPOLLERR;
2063
2064 fiq = &fud->fc->iq;
2065 poll_wait(file, &fiq->waitq, wait);
2066
2067 spin_lock(&fiq->lock);
2068 if (!fiq->connected)
2069 mask = EPOLLERR;
2070 else if (request_pending(fiq))
2071 mask |= EPOLLIN | EPOLLRDNORM;
2072 spin_unlock(&fiq->lock);
2073
2074 return mask;
2075}
2076
2077/* Abort all requests on the given list (pending or processing) */
2078static void end_requests(struct list_head *head)
2079{
2080 while (!list_empty(head)) {
2081 struct fuse_req *req;
2082 req = list_entry(head->next, struct fuse_req, list);
2083 req->out.h.error = -ECONNABORTED;
2084 clear_bit(FR_SENT, &req->flags);
2085 list_del_init(&req->list);
2086 fuse_request_end(req);
2087 }
2088}
2089
2090static void end_polls(struct fuse_conn *fc)
2091{
2092 struct rb_node *p;
2093
2094 p = rb_first(&fc->polled_files);
2095
2096 while (p) {
2097 struct fuse_file *ff;
2098 ff = rb_entry(p, struct fuse_file, polled_node);
2099 wake_up_interruptible_all(&ff->poll_wait);
2100
2101 p = rb_next(p);
2102 }
2103}
2104
2105/*
2106 * Abort all requests.
2107 *
2108 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2109 * filesystem.
2110 *
2111 * The same effect is usually achievable through killing the filesystem daemon
2112 * and all users of the filesystem. The exception is the combination of an
2113 * asynchronous request and the tricky deadlock (see
2114 * Documentation/filesystems/fuse.rst).
2115 *
2116 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2117 * requests, they should be finished off immediately. Locked requests will be
2118 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2119 * requests. It is possible that some request will finish before we can. This
2120 * is OK, the request will in that case be removed from the list before we touch
2121 * it.
2122 */
2123void fuse_abort_conn(struct fuse_conn *fc)
2124{
2125 struct fuse_iqueue *fiq = &fc->iq;
2126
2127 spin_lock(&fc->lock);
2128 if (fc->connected) {
2129 struct fuse_dev *fud;
2130 struct fuse_req *req, *next;
2131 LIST_HEAD(to_end);
2132 unsigned int i;
2133
2134 /* Background queuing checks fc->connected under bg_lock */
2135 spin_lock(&fc->bg_lock);
2136 fc->connected = 0;
2137 spin_unlock(&fc->bg_lock);
2138
2139 fuse_set_initialized(fc);
2140 list_for_each_entry(fud, &fc->devices, entry) {
2141 struct fuse_pqueue *fpq = &fud->pq;
2142
2143 spin_lock(&fpq->lock);
2144 fpq->connected = 0;
2145 list_for_each_entry_safe(req, next, &fpq->io, list) {
2146 req->out.h.error = -ECONNABORTED;
2147 spin_lock(&req->waitq.lock);
2148 set_bit(FR_ABORTED, &req->flags);
2149 if (!test_bit(FR_LOCKED, &req->flags)) {
2150 set_bit(FR_PRIVATE, &req->flags);
2151 __fuse_get_request(req);
2152 list_move(&req->list, &to_end);
2153 }
2154 spin_unlock(&req->waitq.lock);
2155 }
2156 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2157 list_splice_tail_init(&fpq->processing[i],
2158 &to_end);
2159 spin_unlock(&fpq->lock);
2160 }
2161 spin_lock(&fc->bg_lock);
2162 fc->blocked = 0;
2163 fc->max_background = UINT_MAX;
2164 flush_bg_queue(fc);
2165 spin_unlock(&fc->bg_lock);
2166
2167 spin_lock(&fiq->lock);
2168 fiq->connected = 0;
2169 list_for_each_entry(req, &fiq->pending, list)
2170 clear_bit(FR_PENDING, &req->flags);
2171 list_splice_tail_init(&fiq->pending, &to_end);
2172 while (forget_pending(fiq))
2173 kfree(fuse_dequeue_forget(fiq, 1, NULL));
2174 wake_up_all(&fiq->waitq);
2175 spin_unlock(&fiq->lock);
2176 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2177 end_polls(fc);
2178 wake_up_all(&fc->blocked_waitq);
2179 spin_unlock(&fc->lock);
2180
2181 end_requests(&to_end);
2182 } else {
2183 spin_unlock(&fc->lock);
2184 }
2185}
2186EXPORT_SYMBOL_GPL(fuse_abort_conn);
2187
2188void fuse_wait_aborted(struct fuse_conn *fc)
2189{
2190 /* matches implicit memory barrier in fuse_drop_waiting() */
2191 smp_mb();
2192 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2193}
2194
2195int fuse_dev_release(struct inode *inode, struct file *file)
2196{
2197 struct fuse_dev *fud = fuse_get_dev(file);
2198
2199 if (fud) {
2200 struct fuse_conn *fc = fud->fc;
2201 struct fuse_pqueue *fpq = &fud->pq;
2202 LIST_HEAD(to_end);
2203 unsigned int i;
2204
2205 spin_lock(&fpq->lock);
2206 WARN_ON(!list_empty(&fpq->io));
2207 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2208 list_splice_init(&fpq->processing[i], &to_end);
2209 spin_unlock(&fpq->lock);
2210
2211 end_requests(&to_end);
2212
2213 /* Are we the last open device? */
2214 if (atomic_dec_and_test(&fc->dev_count)) {
2215 WARN_ON(fc->iq.fasync != NULL);
2216 fuse_abort_conn(fc);
2217 }
2218 fuse_dev_free(fud);
2219 }
2220 return 0;
2221}
2222EXPORT_SYMBOL_GPL(fuse_dev_release);
2223
2224static int fuse_dev_fasync(int fd, struct file *file, int on)
2225{
2226 struct fuse_dev *fud = fuse_get_dev(file);
2227
2228 if (!fud)
2229 return -EPERM;
2230
2231 /* No locking - fasync_helper does its own locking */
2232 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2233}
2234
2235static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2236{
2237 struct fuse_dev *fud;
2238
2239 if (new->private_data)
2240 return -EINVAL;
2241
2242 fud = fuse_dev_alloc_install(fc);
2243 if (!fud)
2244 return -ENOMEM;
2245
2246 new->private_data = fud;
2247 atomic_inc(&fc->dev_count);
2248
2249 return 0;
2250}
2251
2252static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2253 unsigned long arg)
2254{
2255 int res;
2256 int oldfd;
2257 struct fuse_dev *fud = NULL;
2258
2259 switch (cmd) {
2260 case FUSE_DEV_IOC_CLONE:
2261 res = -EFAULT;
2262 if (!get_user(oldfd, (__u32 __user *)arg)) {
2263 struct file *old = fget(oldfd);
2264
2265 res = -EINVAL;
2266 if (old) {
2267 /*
2268 * Check against file->f_op because CUSE
2269 * uses the same ioctl handler.
2270 */
2271 if (old->f_op == file->f_op)
2272 fud = fuse_get_dev(old);
2273
2274 if (fud) {
2275 mutex_lock(&fuse_mutex);
2276 res = fuse_device_clone(fud->fc, file);
2277 mutex_unlock(&fuse_mutex);
2278 }
2279 fput(old);
2280 }
2281 }
2282 break;
2283 default:
2284 res = -ENOTTY;
2285 break;
2286 }
2287 return res;
2288}
2289
2290const struct file_operations fuse_dev_operations = {
2291 .owner = THIS_MODULE,
2292 .open = fuse_dev_open,
2293 .llseek = no_llseek,
2294 .read_iter = fuse_dev_read,
2295 .splice_read = fuse_dev_splice_read,
2296 .write_iter = fuse_dev_write,
2297 .splice_write = fuse_dev_splice_write,
2298 .poll = fuse_dev_poll,
2299 .release = fuse_dev_release,
2300 .fasync = fuse_dev_fasync,
2301 .unlocked_ioctl = fuse_dev_ioctl,
2302 .compat_ioctl = compat_ptr_ioctl,
2303};
2304EXPORT_SYMBOL_GPL(fuse_dev_operations);
2305
2306static struct miscdevice fuse_miscdevice = {
2307 .minor = FUSE_MINOR,
2308 .name = "fuse",
2309 .fops = &fuse_dev_operations,
2310};
2311
2312int __init fuse_dev_init(void)
2313{
2314 int err = -ENOMEM;
2315 fuse_req_cachep = kmem_cache_create("fuse_request",
2316 sizeof(struct fuse_req),
2317 0, 0, NULL);
2318 if (!fuse_req_cachep)
2319 goto out;
2320
2321 err = misc_register(&fuse_miscdevice);
2322 if (err)
2323 goto out_cache_clean;
2324
2325 return 0;
2326
2327 out_cache_clean:
2328 kmem_cache_destroy(fuse_req_cachep);
2329 out:
2330 return err;
2331}
2332
2333void fuse_dev_cleanup(void)
2334{
2335 misc_deregister(&fuse_miscdevice);
2336 kmem_cache_destroy(fuse_req_cachep);
2337}
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/poll.h>
14#include <linux/sched/signal.h>
15#include <linux/uio.h>
16#include <linux/miscdevice.h>
17#include <linux/pagemap.h>
18#include <linux/file.h>
19#include <linux/slab.h>
20#include <linux/pipe_fs_i.h>
21#include <linux/swap.h>
22#include <linux/splice.h>
23#include <linux/sched.h>
24
25#define CREATE_TRACE_POINTS
26#include "fuse_trace.h"
27
28MODULE_ALIAS_MISCDEV(FUSE_MINOR);
29MODULE_ALIAS("devname:fuse");
30
31/* Ordinary requests have even IDs, while interrupts IDs are odd */
32#define FUSE_INT_REQ_BIT (1ULL << 0)
33#define FUSE_REQ_ID_STEP (1ULL << 1)
34
35static struct kmem_cache *fuse_req_cachep;
36
37static void end_requests(struct list_head *head);
38
39static struct fuse_dev *fuse_get_dev(struct file *file)
40{
41 /*
42 * Lockless access is OK, because file->private data is set
43 * once during mount and is valid until the file is released.
44 */
45 return READ_ONCE(file->private_data);
46}
47
48static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)
49{
50 INIT_LIST_HEAD(&req->list);
51 INIT_LIST_HEAD(&req->intr_entry);
52 init_waitqueue_head(&req->waitq);
53 refcount_set(&req->count, 1);
54 __set_bit(FR_PENDING, &req->flags);
55 req->fm = fm;
56}
57
58static struct fuse_req *fuse_request_alloc(struct fuse_mount *fm, gfp_t flags)
59{
60 struct fuse_req *req = kmem_cache_zalloc(fuse_req_cachep, flags);
61 if (req)
62 fuse_request_init(fm, req);
63
64 return req;
65}
66
67static void fuse_request_free(struct fuse_req *req)
68{
69 kmem_cache_free(fuse_req_cachep, req);
70}
71
72static void __fuse_get_request(struct fuse_req *req)
73{
74 refcount_inc(&req->count);
75}
76
77/* Must be called with > 1 refcount */
78static void __fuse_put_request(struct fuse_req *req)
79{
80 refcount_dec(&req->count);
81}
82
83void fuse_set_initialized(struct fuse_conn *fc)
84{
85 /* Make sure stores before this are seen on another CPU */
86 smp_wmb();
87 fc->initialized = 1;
88}
89
90static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
91{
92 return !fc->initialized || (for_background && fc->blocked);
93}
94
95static void fuse_drop_waiting(struct fuse_conn *fc)
96{
97 /*
98 * lockess check of fc->connected is okay, because atomic_dec_and_test()
99 * provides a memory barrier matched with the one in fuse_wait_aborted()
100 * to ensure no wake-up is missed.
101 */
102 if (atomic_dec_and_test(&fc->num_waiting) &&
103 !READ_ONCE(fc->connected)) {
104 /* wake up aborters */
105 wake_up_all(&fc->blocked_waitq);
106 }
107}
108
109static void fuse_put_request(struct fuse_req *req);
110
111static struct fuse_req *fuse_get_req(struct mnt_idmap *idmap,
112 struct fuse_mount *fm,
113 bool for_background)
114{
115 struct fuse_conn *fc = fm->fc;
116 struct fuse_req *req;
117 bool no_idmap = !fm->sb || (fm->sb->s_iflags & SB_I_NOIDMAP);
118 kuid_t fsuid;
119 kgid_t fsgid;
120 int err;
121
122 atomic_inc(&fc->num_waiting);
123
124 if (fuse_block_alloc(fc, for_background)) {
125 err = -EINTR;
126 if (wait_event_killable_exclusive(fc->blocked_waitq,
127 !fuse_block_alloc(fc, for_background)))
128 goto out;
129 }
130 /* Matches smp_wmb() in fuse_set_initialized() */
131 smp_rmb();
132
133 err = -ENOTCONN;
134 if (!fc->connected)
135 goto out;
136
137 err = -ECONNREFUSED;
138 if (fc->conn_error)
139 goto out;
140
141 req = fuse_request_alloc(fm, GFP_KERNEL);
142 err = -ENOMEM;
143 if (!req) {
144 if (for_background)
145 wake_up(&fc->blocked_waitq);
146 goto out;
147 }
148
149 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
150
151 __set_bit(FR_WAITING, &req->flags);
152 if (for_background)
153 __set_bit(FR_BACKGROUND, &req->flags);
154
155 /*
156 * Keep the old behavior when idmappings support was not
157 * declared by a FUSE server.
158 *
159 * For those FUSE servers who support idmapped mounts,
160 * we send UID/GID only along with "inode creation"
161 * fuse requests, otherwise idmap == &invalid_mnt_idmap and
162 * req->in.h.{u,g}id will be equal to FUSE_INVALID_UIDGID.
163 */
164 fsuid = no_idmap ? current_fsuid() : mapped_fsuid(idmap, fc->user_ns);
165 fsgid = no_idmap ? current_fsgid() : mapped_fsgid(idmap, fc->user_ns);
166 req->in.h.uid = from_kuid(fc->user_ns, fsuid);
167 req->in.h.gid = from_kgid(fc->user_ns, fsgid);
168
169 if (no_idmap && unlikely(req->in.h.uid == ((uid_t)-1) ||
170 req->in.h.gid == ((gid_t)-1))) {
171 fuse_put_request(req);
172 return ERR_PTR(-EOVERFLOW);
173 }
174
175 return req;
176
177 out:
178 fuse_drop_waiting(fc);
179 return ERR_PTR(err);
180}
181
182static void fuse_put_request(struct fuse_req *req)
183{
184 struct fuse_conn *fc = req->fm->fc;
185
186 if (refcount_dec_and_test(&req->count)) {
187 if (test_bit(FR_BACKGROUND, &req->flags)) {
188 /*
189 * We get here in the unlikely case that a background
190 * request was allocated but not sent
191 */
192 spin_lock(&fc->bg_lock);
193 if (!fc->blocked)
194 wake_up(&fc->blocked_waitq);
195 spin_unlock(&fc->bg_lock);
196 }
197
198 if (test_bit(FR_WAITING, &req->flags)) {
199 __clear_bit(FR_WAITING, &req->flags);
200 fuse_drop_waiting(fc);
201 }
202
203 fuse_request_free(req);
204 }
205}
206
207unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args)
208{
209 unsigned nbytes = 0;
210 unsigned i;
211
212 for (i = 0; i < numargs; i++)
213 nbytes += args[i].size;
214
215 return nbytes;
216}
217EXPORT_SYMBOL_GPL(fuse_len_args);
218
219static u64 fuse_get_unique_locked(struct fuse_iqueue *fiq)
220{
221 fiq->reqctr += FUSE_REQ_ID_STEP;
222 return fiq->reqctr;
223}
224
225u64 fuse_get_unique(struct fuse_iqueue *fiq)
226{
227 u64 ret;
228
229 spin_lock(&fiq->lock);
230 ret = fuse_get_unique_locked(fiq);
231 spin_unlock(&fiq->lock);
232
233 return ret;
234}
235EXPORT_SYMBOL_GPL(fuse_get_unique);
236
237static unsigned int fuse_req_hash(u64 unique)
238{
239 return hash_long(unique & ~FUSE_INT_REQ_BIT, FUSE_PQ_HASH_BITS);
240}
241
242/*
243 * A new request is available, wake fiq->waitq
244 */
245static void fuse_dev_wake_and_unlock(struct fuse_iqueue *fiq)
246__releases(fiq->lock)
247{
248 wake_up(&fiq->waitq);
249 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
250 spin_unlock(&fiq->lock);
251}
252
253static void fuse_dev_queue_forget(struct fuse_iqueue *fiq, struct fuse_forget_link *forget)
254{
255 spin_lock(&fiq->lock);
256 if (fiq->connected) {
257 fiq->forget_list_tail->next = forget;
258 fiq->forget_list_tail = forget;
259 fuse_dev_wake_and_unlock(fiq);
260 } else {
261 kfree(forget);
262 spin_unlock(&fiq->lock);
263 }
264}
265
266static void fuse_dev_queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
267{
268 spin_lock(&fiq->lock);
269 if (list_empty(&req->intr_entry)) {
270 list_add_tail(&req->intr_entry, &fiq->interrupts);
271 /*
272 * Pairs with smp_mb() implied by test_and_set_bit()
273 * from fuse_request_end().
274 */
275 smp_mb();
276 if (test_bit(FR_FINISHED, &req->flags)) {
277 list_del_init(&req->intr_entry);
278 spin_unlock(&fiq->lock);
279 } else {
280 fuse_dev_wake_and_unlock(fiq);
281 }
282 } else {
283 spin_unlock(&fiq->lock);
284 }
285}
286
287static void fuse_dev_queue_req(struct fuse_iqueue *fiq, struct fuse_req *req)
288{
289 spin_lock(&fiq->lock);
290 if (fiq->connected) {
291 if (req->in.h.opcode != FUSE_NOTIFY_REPLY)
292 req->in.h.unique = fuse_get_unique_locked(fiq);
293 list_add_tail(&req->list, &fiq->pending);
294 fuse_dev_wake_and_unlock(fiq);
295 } else {
296 spin_unlock(&fiq->lock);
297 req->out.h.error = -ENOTCONN;
298 clear_bit(FR_PENDING, &req->flags);
299 fuse_request_end(req);
300 }
301}
302
303const struct fuse_iqueue_ops fuse_dev_fiq_ops = {
304 .send_forget = fuse_dev_queue_forget,
305 .send_interrupt = fuse_dev_queue_interrupt,
306 .send_req = fuse_dev_queue_req,
307};
308EXPORT_SYMBOL_GPL(fuse_dev_fiq_ops);
309
310static void fuse_send_one(struct fuse_iqueue *fiq, struct fuse_req *req)
311{
312 req->in.h.len = sizeof(struct fuse_in_header) +
313 fuse_len_args(req->args->in_numargs,
314 (struct fuse_arg *) req->args->in_args);
315 trace_fuse_request_send(req);
316 fiq->ops->send_req(fiq, req);
317}
318
319void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
320 u64 nodeid, u64 nlookup)
321{
322 struct fuse_iqueue *fiq = &fc->iq;
323
324 forget->forget_one.nodeid = nodeid;
325 forget->forget_one.nlookup = nlookup;
326
327 fiq->ops->send_forget(fiq, forget);
328}
329
330static void flush_bg_queue(struct fuse_conn *fc)
331{
332 struct fuse_iqueue *fiq = &fc->iq;
333
334 while (fc->active_background < fc->max_background &&
335 !list_empty(&fc->bg_queue)) {
336 struct fuse_req *req;
337
338 req = list_first_entry(&fc->bg_queue, struct fuse_req, list);
339 list_del(&req->list);
340 fc->active_background++;
341 fuse_send_one(fiq, req);
342 }
343}
344
345/*
346 * This function is called when a request is finished. Either a reply
347 * has arrived or it was aborted (and not yet sent) or some error
348 * occurred during communication with userspace, or the device file
349 * was closed. The requester thread is woken up (if still waiting),
350 * the 'end' callback is called if given, else the reference to the
351 * request is released
352 */
353void fuse_request_end(struct fuse_req *req)
354{
355 struct fuse_mount *fm = req->fm;
356 struct fuse_conn *fc = fm->fc;
357 struct fuse_iqueue *fiq = &fc->iq;
358
359 if (test_and_set_bit(FR_FINISHED, &req->flags))
360 goto put_request;
361
362 trace_fuse_request_end(req);
363 /*
364 * test_and_set_bit() implies smp_mb() between bit
365 * changing and below FR_INTERRUPTED check. Pairs with
366 * smp_mb() from queue_interrupt().
367 */
368 if (test_bit(FR_INTERRUPTED, &req->flags)) {
369 spin_lock(&fiq->lock);
370 list_del_init(&req->intr_entry);
371 spin_unlock(&fiq->lock);
372 }
373 WARN_ON(test_bit(FR_PENDING, &req->flags));
374 WARN_ON(test_bit(FR_SENT, &req->flags));
375 if (test_bit(FR_BACKGROUND, &req->flags)) {
376 spin_lock(&fc->bg_lock);
377 clear_bit(FR_BACKGROUND, &req->flags);
378 if (fc->num_background == fc->max_background) {
379 fc->blocked = 0;
380 wake_up(&fc->blocked_waitq);
381 } else if (!fc->blocked) {
382 /*
383 * Wake up next waiter, if any. It's okay to use
384 * waitqueue_active(), as we've already synced up
385 * fc->blocked with waiters with the wake_up() call
386 * above.
387 */
388 if (waitqueue_active(&fc->blocked_waitq))
389 wake_up(&fc->blocked_waitq);
390 }
391
392 fc->num_background--;
393 fc->active_background--;
394 flush_bg_queue(fc);
395 spin_unlock(&fc->bg_lock);
396 } else {
397 /* Wake up waiter sleeping in request_wait_answer() */
398 wake_up(&req->waitq);
399 }
400
401 if (test_bit(FR_ASYNC, &req->flags))
402 req->args->end(fm, req->args, req->out.h.error);
403put_request:
404 fuse_put_request(req);
405}
406EXPORT_SYMBOL_GPL(fuse_request_end);
407
408static int queue_interrupt(struct fuse_req *req)
409{
410 struct fuse_iqueue *fiq = &req->fm->fc->iq;
411
412 /* Check for we've sent request to interrupt this req */
413 if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags)))
414 return -EINVAL;
415
416 fiq->ops->send_interrupt(fiq, req);
417
418 return 0;
419}
420
421static void request_wait_answer(struct fuse_req *req)
422{
423 struct fuse_conn *fc = req->fm->fc;
424 struct fuse_iqueue *fiq = &fc->iq;
425 int err;
426
427 if (!fc->no_interrupt) {
428 /* Any signal may interrupt this */
429 err = wait_event_interruptible(req->waitq,
430 test_bit(FR_FINISHED, &req->flags));
431 if (!err)
432 return;
433
434 set_bit(FR_INTERRUPTED, &req->flags);
435 /* matches barrier in fuse_dev_do_read() */
436 smp_mb__after_atomic();
437 if (test_bit(FR_SENT, &req->flags))
438 queue_interrupt(req);
439 }
440
441 if (!test_bit(FR_FORCE, &req->flags)) {
442 /* Only fatal signals may interrupt this */
443 err = wait_event_killable(req->waitq,
444 test_bit(FR_FINISHED, &req->flags));
445 if (!err)
446 return;
447
448 spin_lock(&fiq->lock);
449 /* Request is not yet in userspace, bail out */
450 if (test_bit(FR_PENDING, &req->flags)) {
451 list_del(&req->list);
452 spin_unlock(&fiq->lock);
453 __fuse_put_request(req);
454 req->out.h.error = -EINTR;
455 return;
456 }
457 spin_unlock(&fiq->lock);
458 }
459
460 /*
461 * Either request is already in userspace, or it was forced.
462 * Wait it out.
463 */
464 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
465}
466
467static void __fuse_request_send(struct fuse_req *req)
468{
469 struct fuse_iqueue *fiq = &req->fm->fc->iq;
470
471 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
472
473 /* acquire extra reference, since request is still needed after
474 fuse_request_end() */
475 __fuse_get_request(req);
476 fuse_send_one(fiq, req);
477
478 request_wait_answer(req);
479 /* Pairs with smp_wmb() in fuse_request_end() */
480 smp_rmb();
481}
482
483static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
484{
485 if (fc->minor < 4 && args->opcode == FUSE_STATFS)
486 args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE;
487
488 if (fc->minor < 9) {
489 switch (args->opcode) {
490 case FUSE_LOOKUP:
491 case FUSE_CREATE:
492 case FUSE_MKNOD:
493 case FUSE_MKDIR:
494 case FUSE_SYMLINK:
495 case FUSE_LINK:
496 args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
497 break;
498 case FUSE_GETATTR:
499 case FUSE_SETATTR:
500 args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
501 break;
502 }
503 }
504 if (fc->minor < 12) {
505 switch (args->opcode) {
506 case FUSE_CREATE:
507 args->in_args[0].size = sizeof(struct fuse_open_in);
508 break;
509 case FUSE_MKNOD:
510 args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
511 break;
512 }
513 }
514}
515
516static void fuse_force_creds(struct fuse_req *req)
517{
518 struct fuse_conn *fc = req->fm->fc;
519
520 if (!req->fm->sb || req->fm->sb->s_iflags & SB_I_NOIDMAP) {
521 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid());
522 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid());
523 } else {
524 req->in.h.uid = FUSE_INVALID_UIDGID;
525 req->in.h.gid = FUSE_INVALID_UIDGID;
526 }
527
528 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
529}
530
531static void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args)
532{
533 req->in.h.opcode = args->opcode;
534 req->in.h.nodeid = args->nodeid;
535 req->args = args;
536 if (args->is_ext)
537 req->in.h.total_extlen = args->in_args[args->ext_idx].size / 8;
538 if (args->end)
539 __set_bit(FR_ASYNC, &req->flags);
540}
541
542ssize_t __fuse_simple_request(struct mnt_idmap *idmap,
543 struct fuse_mount *fm,
544 struct fuse_args *args)
545{
546 struct fuse_conn *fc = fm->fc;
547 struct fuse_req *req;
548 ssize_t ret;
549
550 if (args->force) {
551 atomic_inc(&fc->num_waiting);
552 req = fuse_request_alloc(fm, GFP_KERNEL | __GFP_NOFAIL);
553
554 if (!args->nocreds)
555 fuse_force_creds(req);
556
557 __set_bit(FR_WAITING, &req->flags);
558 __set_bit(FR_FORCE, &req->flags);
559 } else {
560 WARN_ON(args->nocreds);
561 req = fuse_get_req(idmap, fm, false);
562 if (IS_ERR(req))
563 return PTR_ERR(req);
564 }
565
566 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
567 fuse_adjust_compat(fc, args);
568 fuse_args_to_req(req, args);
569
570 if (!args->noreply)
571 __set_bit(FR_ISREPLY, &req->flags);
572 __fuse_request_send(req);
573 ret = req->out.h.error;
574 if (!ret && args->out_argvar) {
575 BUG_ON(args->out_numargs == 0);
576 ret = args->out_args[args->out_numargs - 1].size;
577 }
578 fuse_put_request(req);
579
580 return ret;
581}
582
583static bool fuse_request_queue_background(struct fuse_req *req)
584{
585 struct fuse_mount *fm = req->fm;
586 struct fuse_conn *fc = fm->fc;
587 bool queued = false;
588
589 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags));
590 if (!test_bit(FR_WAITING, &req->flags)) {
591 __set_bit(FR_WAITING, &req->flags);
592 atomic_inc(&fc->num_waiting);
593 }
594 __set_bit(FR_ISREPLY, &req->flags);
595 spin_lock(&fc->bg_lock);
596 if (likely(fc->connected)) {
597 fc->num_background++;
598 if (fc->num_background == fc->max_background)
599 fc->blocked = 1;
600 list_add_tail(&req->list, &fc->bg_queue);
601 flush_bg_queue(fc);
602 queued = true;
603 }
604 spin_unlock(&fc->bg_lock);
605
606 return queued;
607}
608
609int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args,
610 gfp_t gfp_flags)
611{
612 struct fuse_req *req;
613
614 if (args->force) {
615 WARN_ON(!args->nocreds);
616 req = fuse_request_alloc(fm, gfp_flags);
617 if (!req)
618 return -ENOMEM;
619 __set_bit(FR_BACKGROUND, &req->flags);
620 } else {
621 WARN_ON(args->nocreds);
622 req = fuse_get_req(&invalid_mnt_idmap, fm, true);
623 if (IS_ERR(req))
624 return PTR_ERR(req);
625 }
626
627 fuse_args_to_req(req, args);
628
629 if (!fuse_request_queue_background(req)) {
630 fuse_put_request(req);
631 return -ENOTCONN;
632 }
633
634 return 0;
635}
636EXPORT_SYMBOL_GPL(fuse_simple_background);
637
638static int fuse_simple_notify_reply(struct fuse_mount *fm,
639 struct fuse_args *args, u64 unique)
640{
641 struct fuse_req *req;
642 struct fuse_iqueue *fiq = &fm->fc->iq;
643
644 req = fuse_get_req(&invalid_mnt_idmap, fm, false);
645 if (IS_ERR(req))
646 return PTR_ERR(req);
647
648 __clear_bit(FR_ISREPLY, &req->flags);
649 req->in.h.unique = unique;
650
651 fuse_args_to_req(req, args);
652
653 fuse_send_one(fiq, req);
654
655 return 0;
656}
657
658/*
659 * Lock the request. Up to the next unlock_request() there mustn't be
660 * anything that could cause a page-fault. If the request was already
661 * aborted bail out.
662 */
663static int lock_request(struct fuse_req *req)
664{
665 int err = 0;
666 if (req) {
667 spin_lock(&req->waitq.lock);
668 if (test_bit(FR_ABORTED, &req->flags))
669 err = -ENOENT;
670 else
671 set_bit(FR_LOCKED, &req->flags);
672 spin_unlock(&req->waitq.lock);
673 }
674 return err;
675}
676
677/*
678 * Unlock request. If it was aborted while locked, caller is responsible
679 * for unlocking and ending the request.
680 */
681static int unlock_request(struct fuse_req *req)
682{
683 int err = 0;
684 if (req) {
685 spin_lock(&req->waitq.lock);
686 if (test_bit(FR_ABORTED, &req->flags))
687 err = -ENOENT;
688 else
689 clear_bit(FR_LOCKED, &req->flags);
690 spin_unlock(&req->waitq.lock);
691 }
692 return err;
693}
694
695struct fuse_copy_state {
696 int write;
697 struct fuse_req *req;
698 struct iov_iter *iter;
699 struct pipe_buffer *pipebufs;
700 struct pipe_buffer *currbuf;
701 struct pipe_inode_info *pipe;
702 unsigned long nr_segs;
703 struct page *pg;
704 unsigned len;
705 unsigned offset;
706 unsigned move_pages:1;
707};
708
709static void fuse_copy_init(struct fuse_copy_state *cs, int write,
710 struct iov_iter *iter)
711{
712 memset(cs, 0, sizeof(*cs));
713 cs->write = write;
714 cs->iter = iter;
715}
716
717/* Unmap and put previous page of userspace buffer */
718static void fuse_copy_finish(struct fuse_copy_state *cs)
719{
720 if (cs->currbuf) {
721 struct pipe_buffer *buf = cs->currbuf;
722
723 if (cs->write)
724 buf->len = PAGE_SIZE - cs->len;
725 cs->currbuf = NULL;
726 } else if (cs->pg) {
727 if (cs->write) {
728 flush_dcache_page(cs->pg);
729 set_page_dirty_lock(cs->pg);
730 }
731 put_page(cs->pg);
732 }
733 cs->pg = NULL;
734}
735
736/*
737 * Get another pagefull of userspace buffer, and map it to kernel
738 * address space, and lock request
739 */
740static int fuse_copy_fill(struct fuse_copy_state *cs)
741{
742 struct page *page;
743 int err;
744
745 err = unlock_request(cs->req);
746 if (err)
747 return err;
748
749 fuse_copy_finish(cs);
750 if (cs->pipebufs) {
751 struct pipe_buffer *buf = cs->pipebufs;
752
753 if (!cs->write) {
754 err = pipe_buf_confirm(cs->pipe, buf);
755 if (err)
756 return err;
757
758 BUG_ON(!cs->nr_segs);
759 cs->currbuf = buf;
760 cs->pg = buf->page;
761 cs->offset = buf->offset;
762 cs->len = buf->len;
763 cs->pipebufs++;
764 cs->nr_segs--;
765 } else {
766 if (cs->nr_segs >= cs->pipe->max_usage)
767 return -EIO;
768
769 page = alloc_page(GFP_HIGHUSER);
770 if (!page)
771 return -ENOMEM;
772
773 buf->page = page;
774 buf->offset = 0;
775 buf->len = 0;
776
777 cs->currbuf = buf;
778 cs->pg = page;
779 cs->offset = 0;
780 cs->len = PAGE_SIZE;
781 cs->pipebufs++;
782 cs->nr_segs++;
783 }
784 } else {
785 size_t off;
786 err = iov_iter_get_pages2(cs->iter, &page, PAGE_SIZE, 1, &off);
787 if (err < 0)
788 return err;
789 BUG_ON(!err);
790 cs->len = err;
791 cs->offset = off;
792 cs->pg = page;
793 }
794
795 return lock_request(cs->req);
796}
797
798/* Do as much copy to/from userspace buffer as we can */
799static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
800{
801 unsigned ncpy = min(*size, cs->len);
802 if (val) {
803 void *pgaddr = kmap_local_page(cs->pg);
804 void *buf = pgaddr + cs->offset;
805
806 if (cs->write)
807 memcpy(buf, *val, ncpy);
808 else
809 memcpy(*val, buf, ncpy);
810
811 kunmap_local(pgaddr);
812 *val += ncpy;
813 }
814 *size -= ncpy;
815 cs->len -= ncpy;
816 cs->offset += ncpy;
817 return ncpy;
818}
819
820static int fuse_check_folio(struct folio *folio)
821{
822 if (folio_mapped(folio) ||
823 folio->mapping != NULL ||
824 (folio->flags & PAGE_FLAGS_CHECK_AT_PREP &
825 ~(1 << PG_locked |
826 1 << PG_referenced |
827 1 << PG_lru |
828 1 << PG_active |
829 1 << PG_workingset |
830 1 << PG_reclaim |
831 1 << PG_waiters |
832 LRU_GEN_MASK | LRU_REFS_MASK))) {
833 dump_page(&folio->page, "fuse: trying to steal weird page");
834 return 1;
835 }
836 return 0;
837}
838
839/*
840 * Attempt to steal a page from the splice() pipe and move it into the
841 * pagecache. If successful, the pointer in @pagep will be updated. The
842 * folio that was originally in @pagep will lose a reference and the new
843 * folio returned in @pagep will carry a reference.
844 */
845static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
846{
847 int err;
848 struct folio *oldfolio = page_folio(*pagep);
849 struct folio *newfolio;
850 struct pipe_buffer *buf = cs->pipebufs;
851
852 folio_get(oldfolio);
853 err = unlock_request(cs->req);
854 if (err)
855 goto out_put_old;
856
857 fuse_copy_finish(cs);
858
859 err = pipe_buf_confirm(cs->pipe, buf);
860 if (err)
861 goto out_put_old;
862
863 BUG_ON(!cs->nr_segs);
864 cs->currbuf = buf;
865 cs->len = buf->len;
866 cs->pipebufs++;
867 cs->nr_segs--;
868
869 if (cs->len != PAGE_SIZE)
870 goto out_fallback;
871
872 if (!pipe_buf_try_steal(cs->pipe, buf))
873 goto out_fallback;
874
875 newfolio = page_folio(buf->page);
876
877 folio_clear_uptodate(newfolio);
878 folio_clear_mappedtodisk(newfolio);
879
880 if (fuse_check_folio(newfolio) != 0)
881 goto out_fallback_unlock;
882
883 /*
884 * This is a new and locked page, it shouldn't be mapped or
885 * have any special flags on it
886 */
887 if (WARN_ON(folio_mapped(oldfolio)))
888 goto out_fallback_unlock;
889 if (WARN_ON(folio_has_private(oldfolio)))
890 goto out_fallback_unlock;
891 if (WARN_ON(folio_test_dirty(oldfolio) ||
892 folio_test_writeback(oldfolio)))
893 goto out_fallback_unlock;
894 if (WARN_ON(folio_test_mlocked(oldfolio)))
895 goto out_fallback_unlock;
896
897 replace_page_cache_folio(oldfolio, newfolio);
898
899 folio_get(newfolio);
900
901 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
902 folio_add_lru(newfolio);
903
904 /*
905 * Release while we have extra ref on stolen page. Otherwise
906 * anon_pipe_buf_release() might think the page can be reused.
907 */
908 pipe_buf_release(cs->pipe, buf);
909
910 err = 0;
911 spin_lock(&cs->req->waitq.lock);
912 if (test_bit(FR_ABORTED, &cs->req->flags))
913 err = -ENOENT;
914 else
915 *pagep = &newfolio->page;
916 spin_unlock(&cs->req->waitq.lock);
917
918 if (err) {
919 folio_unlock(newfolio);
920 folio_put(newfolio);
921 goto out_put_old;
922 }
923
924 folio_unlock(oldfolio);
925 /* Drop ref for ap->pages[] array */
926 folio_put(oldfolio);
927 cs->len = 0;
928
929 err = 0;
930out_put_old:
931 /* Drop ref obtained in this function */
932 folio_put(oldfolio);
933 return err;
934
935out_fallback_unlock:
936 folio_unlock(newfolio);
937out_fallback:
938 cs->pg = buf->page;
939 cs->offset = buf->offset;
940
941 err = lock_request(cs->req);
942 if (!err)
943 err = 1;
944
945 goto out_put_old;
946}
947
948static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
949 unsigned offset, unsigned count)
950{
951 struct pipe_buffer *buf;
952 int err;
953
954 if (cs->nr_segs >= cs->pipe->max_usage)
955 return -EIO;
956
957 get_page(page);
958 err = unlock_request(cs->req);
959 if (err) {
960 put_page(page);
961 return err;
962 }
963
964 fuse_copy_finish(cs);
965
966 buf = cs->pipebufs;
967 buf->page = page;
968 buf->offset = offset;
969 buf->len = count;
970
971 cs->pipebufs++;
972 cs->nr_segs++;
973 cs->len = 0;
974
975 return 0;
976}
977
978/*
979 * Copy a page in the request to/from the userspace buffer. Must be
980 * done atomically
981 */
982static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
983 unsigned offset, unsigned count, int zeroing)
984{
985 int err;
986 struct page *page = *pagep;
987
988 if (page && zeroing && count < PAGE_SIZE)
989 clear_highpage(page);
990
991 while (count) {
992 if (cs->write && cs->pipebufs && page) {
993 /*
994 * Can't control lifetime of pipe buffers, so always
995 * copy user pages.
996 */
997 if (cs->req->args->user_pages) {
998 err = fuse_copy_fill(cs);
999 if (err)
1000 return err;
1001 } else {
1002 return fuse_ref_page(cs, page, offset, count);
1003 }
1004 } else if (!cs->len) {
1005 if (cs->move_pages && page &&
1006 offset == 0 && count == PAGE_SIZE) {
1007 err = fuse_try_move_page(cs, pagep);
1008 if (err <= 0)
1009 return err;
1010 } else {
1011 err = fuse_copy_fill(cs);
1012 if (err)
1013 return err;
1014 }
1015 }
1016 if (page) {
1017 void *mapaddr = kmap_local_page(page);
1018 void *buf = mapaddr + offset;
1019 offset += fuse_copy_do(cs, &buf, &count);
1020 kunmap_local(mapaddr);
1021 } else
1022 offset += fuse_copy_do(cs, NULL, &count);
1023 }
1024 if (page && !cs->write)
1025 flush_dcache_page(page);
1026 return 0;
1027}
1028
1029/* Copy pages in the request to/from userspace buffer */
1030static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1031 int zeroing)
1032{
1033 unsigned i;
1034 struct fuse_req *req = cs->req;
1035 struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args);
1036
1037 for (i = 0; i < ap->num_folios && (nbytes || zeroing); i++) {
1038 int err;
1039 unsigned int offset = ap->descs[i].offset;
1040 unsigned int count = min(nbytes, ap->descs[i].length);
1041 struct page *orig, *pagep;
1042
1043 orig = pagep = &ap->folios[i]->page;
1044
1045 err = fuse_copy_page(cs, &pagep, offset, count, zeroing);
1046 if (err)
1047 return err;
1048
1049 nbytes -= count;
1050
1051 /*
1052 * fuse_copy_page may have moved a page from a pipe instead of
1053 * copying into our given page, so update the folios if it was
1054 * replaced.
1055 */
1056 if (pagep != orig)
1057 ap->folios[i] = page_folio(pagep);
1058 }
1059 return 0;
1060}
1061
1062/* Copy a single argument in the request to/from userspace buffer */
1063static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1064{
1065 while (size) {
1066 if (!cs->len) {
1067 int err = fuse_copy_fill(cs);
1068 if (err)
1069 return err;
1070 }
1071 fuse_copy_do(cs, &val, &size);
1072 }
1073 return 0;
1074}
1075
1076/* Copy request arguments to/from userspace buffer */
1077static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1078 unsigned argpages, struct fuse_arg *args,
1079 int zeroing)
1080{
1081 int err = 0;
1082 unsigned i;
1083
1084 for (i = 0; !err && i < numargs; i++) {
1085 struct fuse_arg *arg = &args[i];
1086 if (i == numargs - 1 && argpages)
1087 err = fuse_copy_pages(cs, arg->size, zeroing);
1088 else
1089 err = fuse_copy_one(cs, arg->value, arg->size);
1090 }
1091 return err;
1092}
1093
1094static int forget_pending(struct fuse_iqueue *fiq)
1095{
1096 return fiq->forget_list_head.next != NULL;
1097}
1098
1099static int request_pending(struct fuse_iqueue *fiq)
1100{
1101 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1102 forget_pending(fiq);
1103}
1104
1105/*
1106 * Transfer an interrupt request to userspace
1107 *
1108 * Unlike other requests this is assembled on demand, without a need
1109 * to allocate a separate fuse_req structure.
1110 *
1111 * Called with fiq->lock held, releases it
1112 */
1113static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1114 struct fuse_copy_state *cs,
1115 size_t nbytes, struct fuse_req *req)
1116__releases(fiq->lock)
1117{
1118 struct fuse_in_header ih;
1119 struct fuse_interrupt_in arg;
1120 unsigned reqsize = sizeof(ih) + sizeof(arg);
1121 int err;
1122
1123 list_del_init(&req->intr_entry);
1124 memset(&ih, 0, sizeof(ih));
1125 memset(&arg, 0, sizeof(arg));
1126 ih.len = reqsize;
1127 ih.opcode = FUSE_INTERRUPT;
1128 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT);
1129 arg.unique = req->in.h.unique;
1130
1131 spin_unlock(&fiq->lock);
1132 if (nbytes < reqsize)
1133 return -EINVAL;
1134
1135 err = fuse_copy_one(cs, &ih, sizeof(ih));
1136 if (!err)
1137 err = fuse_copy_one(cs, &arg, sizeof(arg));
1138 fuse_copy_finish(cs);
1139
1140 return err ? err : reqsize;
1141}
1142
1143static struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq,
1144 unsigned int max,
1145 unsigned int *countp)
1146{
1147 struct fuse_forget_link *head = fiq->forget_list_head.next;
1148 struct fuse_forget_link **newhead = &head;
1149 unsigned count;
1150
1151 for (count = 0; *newhead != NULL && count < max; count++)
1152 newhead = &(*newhead)->next;
1153
1154 fiq->forget_list_head.next = *newhead;
1155 *newhead = NULL;
1156 if (fiq->forget_list_head.next == NULL)
1157 fiq->forget_list_tail = &fiq->forget_list_head;
1158
1159 if (countp != NULL)
1160 *countp = count;
1161
1162 return head;
1163}
1164
1165static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1166 struct fuse_copy_state *cs,
1167 size_t nbytes)
1168__releases(fiq->lock)
1169{
1170 int err;
1171 struct fuse_forget_link *forget = fuse_dequeue_forget(fiq, 1, NULL);
1172 struct fuse_forget_in arg = {
1173 .nlookup = forget->forget_one.nlookup,
1174 };
1175 struct fuse_in_header ih = {
1176 .opcode = FUSE_FORGET,
1177 .nodeid = forget->forget_one.nodeid,
1178 .unique = fuse_get_unique_locked(fiq),
1179 .len = sizeof(ih) + sizeof(arg),
1180 };
1181
1182 spin_unlock(&fiq->lock);
1183 kfree(forget);
1184 if (nbytes < ih.len)
1185 return -EINVAL;
1186
1187 err = fuse_copy_one(cs, &ih, sizeof(ih));
1188 if (!err)
1189 err = fuse_copy_one(cs, &arg, sizeof(arg));
1190 fuse_copy_finish(cs);
1191
1192 if (err)
1193 return err;
1194
1195 return ih.len;
1196}
1197
1198static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1199 struct fuse_copy_state *cs, size_t nbytes)
1200__releases(fiq->lock)
1201{
1202 int err;
1203 unsigned max_forgets;
1204 unsigned count;
1205 struct fuse_forget_link *head;
1206 struct fuse_batch_forget_in arg = { .count = 0 };
1207 struct fuse_in_header ih = {
1208 .opcode = FUSE_BATCH_FORGET,
1209 .unique = fuse_get_unique_locked(fiq),
1210 .len = sizeof(ih) + sizeof(arg),
1211 };
1212
1213 if (nbytes < ih.len) {
1214 spin_unlock(&fiq->lock);
1215 return -EINVAL;
1216 }
1217
1218 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1219 head = fuse_dequeue_forget(fiq, max_forgets, &count);
1220 spin_unlock(&fiq->lock);
1221
1222 arg.count = count;
1223 ih.len += count * sizeof(struct fuse_forget_one);
1224 err = fuse_copy_one(cs, &ih, sizeof(ih));
1225 if (!err)
1226 err = fuse_copy_one(cs, &arg, sizeof(arg));
1227
1228 while (head) {
1229 struct fuse_forget_link *forget = head;
1230
1231 if (!err) {
1232 err = fuse_copy_one(cs, &forget->forget_one,
1233 sizeof(forget->forget_one));
1234 }
1235 head = forget->next;
1236 kfree(forget);
1237 }
1238
1239 fuse_copy_finish(cs);
1240
1241 if (err)
1242 return err;
1243
1244 return ih.len;
1245}
1246
1247static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1248 struct fuse_copy_state *cs,
1249 size_t nbytes)
1250__releases(fiq->lock)
1251{
1252 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1253 return fuse_read_single_forget(fiq, cs, nbytes);
1254 else
1255 return fuse_read_batch_forget(fiq, cs, nbytes);
1256}
1257
1258/*
1259 * Read a single request into the userspace filesystem's buffer. This
1260 * function waits until a request is available, then removes it from
1261 * the pending list and copies request data to userspace buffer. If
1262 * no reply is needed (FORGET) or request has been aborted or there
1263 * was an error during the copying then it's finished by calling
1264 * fuse_request_end(). Otherwise add it to the processing list, and set
1265 * the 'sent' flag.
1266 */
1267static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1268 struct fuse_copy_state *cs, size_t nbytes)
1269{
1270 ssize_t err;
1271 struct fuse_conn *fc = fud->fc;
1272 struct fuse_iqueue *fiq = &fc->iq;
1273 struct fuse_pqueue *fpq = &fud->pq;
1274 struct fuse_req *req;
1275 struct fuse_args *args;
1276 unsigned reqsize;
1277 unsigned int hash;
1278
1279 /*
1280 * Require sane minimum read buffer - that has capacity for fixed part
1281 * of any request header + negotiated max_write room for data.
1282 *
1283 * Historically libfuse reserves 4K for fixed header room, but e.g.
1284 * GlusterFS reserves only 80 bytes
1285 *
1286 * = `sizeof(fuse_in_header) + sizeof(fuse_write_in)`
1287 *
1288 * which is the absolute minimum any sane filesystem should be using
1289 * for header room.
1290 */
1291 if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER,
1292 sizeof(struct fuse_in_header) +
1293 sizeof(struct fuse_write_in) +
1294 fc->max_write))
1295 return -EINVAL;
1296
1297 restart:
1298 for (;;) {
1299 spin_lock(&fiq->lock);
1300 if (!fiq->connected || request_pending(fiq))
1301 break;
1302 spin_unlock(&fiq->lock);
1303
1304 if (file->f_flags & O_NONBLOCK)
1305 return -EAGAIN;
1306 err = wait_event_interruptible_exclusive(fiq->waitq,
1307 !fiq->connected || request_pending(fiq));
1308 if (err)
1309 return err;
1310 }
1311
1312 if (!fiq->connected) {
1313 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1314 goto err_unlock;
1315 }
1316
1317 if (!list_empty(&fiq->interrupts)) {
1318 req = list_entry(fiq->interrupts.next, struct fuse_req,
1319 intr_entry);
1320 return fuse_read_interrupt(fiq, cs, nbytes, req);
1321 }
1322
1323 if (forget_pending(fiq)) {
1324 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1325 return fuse_read_forget(fc, fiq, cs, nbytes);
1326
1327 if (fiq->forget_batch <= -8)
1328 fiq->forget_batch = 16;
1329 }
1330
1331 req = list_entry(fiq->pending.next, struct fuse_req, list);
1332 clear_bit(FR_PENDING, &req->flags);
1333 list_del_init(&req->list);
1334 spin_unlock(&fiq->lock);
1335
1336 args = req->args;
1337 reqsize = req->in.h.len;
1338
1339 /* If request is too large, reply with an error and restart the read */
1340 if (nbytes < reqsize) {
1341 req->out.h.error = -EIO;
1342 /* SETXATTR is special, since it may contain too large data */
1343 if (args->opcode == FUSE_SETXATTR)
1344 req->out.h.error = -E2BIG;
1345 fuse_request_end(req);
1346 goto restart;
1347 }
1348 spin_lock(&fpq->lock);
1349 /*
1350 * Must not put request on fpq->io queue after having been shut down by
1351 * fuse_abort_conn()
1352 */
1353 if (!fpq->connected) {
1354 req->out.h.error = err = -ECONNABORTED;
1355 goto out_end;
1356
1357 }
1358 list_add(&req->list, &fpq->io);
1359 spin_unlock(&fpq->lock);
1360 cs->req = req;
1361 err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h));
1362 if (!err)
1363 err = fuse_copy_args(cs, args->in_numargs, args->in_pages,
1364 (struct fuse_arg *) args->in_args, 0);
1365 fuse_copy_finish(cs);
1366 spin_lock(&fpq->lock);
1367 clear_bit(FR_LOCKED, &req->flags);
1368 if (!fpq->connected) {
1369 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1370 goto out_end;
1371 }
1372 if (err) {
1373 req->out.h.error = -EIO;
1374 goto out_end;
1375 }
1376 if (!test_bit(FR_ISREPLY, &req->flags)) {
1377 err = reqsize;
1378 goto out_end;
1379 }
1380 hash = fuse_req_hash(req->in.h.unique);
1381 list_move_tail(&req->list, &fpq->processing[hash]);
1382 __fuse_get_request(req);
1383 set_bit(FR_SENT, &req->flags);
1384 spin_unlock(&fpq->lock);
1385 /* matches barrier in request_wait_answer() */
1386 smp_mb__after_atomic();
1387 if (test_bit(FR_INTERRUPTED, &req->flags))
1388 queue_interrupt(req);
1389 fuse_put_request(req);
1390
1391 return reqsize;
1392
1393out_end:
1394 if (!test_bit(FR_PRIVATE, &req->flags))
1395 list_del_init(&req->list);
1396 spin_unlock(&fpq->lock);
1397 fuse_request_end(req);
1398 return err;
1399
1400 err_unlock:
1401 spin_unlock(&fiq->lock);
1402 return err;
1403}
1404
1405static int fuse_dev_open(struct inode *inode, struct file *file)
1406{
1407 /*
1408 * The fuse device's file's private_data is used to hold
1409 * the fuse_conn(ection) when it is mounted, and is used to
1410 * keep track of whether the file has been mounted already.
1411 */
1412 file->private_data = NULL;
1413 return 0;
1414}
1415
1416static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1417{
1418 struct fuse_copy_state cs;
1419 struct file *file = iocb->ki_filp;
1420 struct fuse_dev *fud = fuse_get_dev(file);
1421
1422 if (!fud)
1423 return -EPERM;
1424
1425 if (!user_backed_iter(to))
1426 return -EINVAL;
1427
1428 fuse_copy_init(&cs, 1, to);
1429
1430 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1431}
1432
1433static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1434 struct pipe_inode_info *pipe,
1435 size_t len, unsigned int flags)
1436{
1437 int total, ret;
1438 int page_nr = 0;
1439 struct pipe_buffer *bufs;
1440 struct fuse_copy_state cs;
1441 struct fuse_dev *fud = fuse_get_dev(in);
1442
1443 if (!fud)
1444 return -EPERM;
1445
1446 bufs = kvmalloc_array(pipe->max_usage, sizeof(struct pipe_buffer),
1447 GFP_KERNEL);
1448 if (!bufs)
1449 return -ENOMEM;
1450
1451 fuse_copy_init(&cs, 1, NULL);
1452 cs.pipebufs = bufs;
1453 cs.pipe = pipe;
1454 ret = fuse_dev_do_read(fud, in, &cs, len);
1455 if (ret < 0)
1456 goto out;
1457
1458 if (pipe_occupancy(pipe->head, pipe->tail) + cs.nr_segs > pipe->max_usage) {
1459 ret = -EIO;
1460 goto out;
1461 }
1462
1463 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1464 /*
1465 * Need to be careful about this. Having buf->ops in module
1466 * code can Oops if the buffer persists after module unload.
1467 */
1468 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1469 bufs[page_nr].flags = 0;
1470 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1471 if (unlikely(ret < 0))
1472 break;
1473 }
1474 if (total)
1475 ret = total;
1476out:
1477 for (; page_nr < cs.nr_segs; page_nr++)
1478 put_page(bufs[page_nr].page);
1479
1480 kvfree(bufs);
1481 return ret;
1482}
1483
1484static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1485 struct fuse_copy_state *cs)
1486{
1487 struct fuse_notify_poll_wakeup_out outarg;
1488 int err = -EINVAL;
1489
1490 if (size != sizeof(outarg))
1491 goto err;
1492
1493 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1494 if (err)
1495 goto err;
1496
1497 fuse_copy_finish(cs);
1498 return fuse_notify_poll_wakeup(fc, &outarg);
1499
1500err:
1501 fuse_copy_finish(cs);
1502 return err;
1503}
1504
1505static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1506 struct fuse_copy_state *cs)
1507{
1508 struct fuse_notify_inval_inode_out outarg;
1509 int err = -EINVAL;
1510
1511 if (size != sizeof(outarg))
1512 goto err;
1513
1514 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1515 if (err)
1516 goto err;
1517 fuse_copy_finish(cs);
1518
1519 down_read(&fc->killsb);
1520 err = fuse_reverse_inval_inode(fc, outarg.ino,
1521 outarg.off, outarg.len);
1522 up_read(&fc->killsb);
1523 return err;
1524
1525err:
1526 fuse_copy_finish(cs);
1527 return err;
1528}
1529
1530static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1531 struct fuse_copy_state *cs)
1532{
1533 struct fuse_notify_inval_entry_out outarg;
1534 int err = -ENOMEM;
1535 char *buf;
1536 struct qstr name;
1537
1538 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1539 if (!buf)
1540 goto err;
1541
1542 err = -EINVAL;
1543 if (size < sizeof(outarg))
1544 goto err;
1545
1546 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1547 if (err)
1548 goto err;
1549
1550 err = -ENAMETOOLONG;
1551 if (outarg.namelen > FUSE_NAME_MAX)
1552 goto err;
1553
1554 err = -EINVAL;
1555 if (size != sizeof(outarg) + outarg.namelen + 1)
1556 goto err;
1557
1558 name.name = buf;
1559 name.len = outarg.namelen;
1560 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1561 if (err)
1562 goto err;
1563 fuse_copy_finish(cs);
1564 buf[outarg.namelen] = 0;
1565
1566 down_read(&fc->killsb);
1567 err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name, outarg.flags);
1568 up_read(&fc->killsb);
1569 kfree(buf);
1570 return err;
1571
1572err:
1573 kfree(buf);
1574 fuse_copy_finish(cs);
1575 return err;
1576}
1577
1578static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1579 struct fuse_copy_state *cs)
1580{
1581 struct fuse_notify_delete_out outarg;
1582 int err = -ENOMEM;
1583 char *buf;
1584 struct qstr name;
1585
1586 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1587 if (!buf)
1588 goto err;
1589
1590 err = -EINVAL;
1591 if (size < sizeof(outarg))
1592 goto err;
1593
1594 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1595 if (err)
1596 goto err;
1597
1598 err = -ENAMETOOLONG;
1599 if (outarg.namelen > FUSE_NAME_MAX)
1600 goto err;
1601
1602 err = -EINVAL;
1603 if (size != sizeof(outarg) + outarg.namelen + 1)
1604 goto err;
1605
1606 name.name = buf;
1607 name.len = outarg.namelen;
1608 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1609 if (err)
1610 goto err;
1611 fuse_copy_finish(cs);
1612 buf[outarg.namelen] = 0;
1613
1614 down_read(&fc->killsb);
1615 err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name, 0);
1616 up_read(&fc->killsb);
1617 kfree(buf);
1618 return err;
1619
1620err:
1621 kfree(buf);
1622 fuse_copy_finish(cs);
1623 return err;
1624}
1625
1626static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1627 struct fuse_copy_state *cs)
1628{
1629 struct fuse_notify_store_out outarg;
1630 struct inode *inode;
1631 struct address_space *mapping;
1632 u64 nodeid;
1633 int err;
1634 pgoff_t index;
1635 unsigned int offset;
1636 unsigned int num;
1637 loff_t file_size;
1638 loff_t end;
1639
1640 err = -EINVAL;
1641 if (size < sizeof(outarg))
1642 goto out_finish;
1643
1644 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1645 if (err)
1646 goto out_finish;
1647
1648 err = -EINVAL;
1649 if (size - sizeof(outarg) != outarg.size)
1650 goto out_finish;
1651
1652 nodeid = outarg.nodeid;
1653
1654 down_read(&fc->killsb);
1655
1656 err = -ENOENT;
1657 inode = fuse_ilookup(fc, nodeid, NULL);
1658 if (!inode)
1659 goto out_up_killsb;
1660
1661 mapping = inode->i_mapping;
1662 index = outarg.offset >> PAGE_SHIFT;
1663 offset = outarg.offset & ~PAGE_MASK;
1664 file_size = i_size_read(inode);
1665 end = outarg.offset + outarg.size;
1666 if (end > file_size) {
1667 file_size = end;
1668 fuse_write_update_attr(inode, file_size, outarg.size);
1669 }
1670
1671 num = outarg.size;
1672 while (num) {
1673 struct folio *folio;
1674 struct page *page;
1675 unsigned int this_num;
1676
1677 folio = filemap_grab_folio(mapping, index);
1678 err = PTR_ERR(folio);
1679 if (IS_ERR(folio))
1680 goto out_iput;
1681
1682 page = &folio->page;
1683 this_num = min_t(unsigned, num, folio_size(folio) - offset);
1684 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1685 if (!folio_test_uptodate(folio) && !err && offset == 0 &&
1686 (this_num == folio_size(folio) || file_size == end)) {
1687 folio_zero_segment(folio, this_num, folio_size(folio));
1688 folio_mark_uptodate(folio);
1689 }
1690 folio_unlock(folio);
1691 folio_put(folio);
1692
1693 if (err)
1694 goto out_iput;
1695
1696 num -= this_num;
1697 offset = 0;
1698 index++;
1699 }
1700
1701 err = 0;
1702
1703out_iput:
1704 iput(inode);
1705out_up_killsb:
1706 up_read(&fc->killsb);
1707out_finish:
1708 fuse_copy_finish(cs);
1709 return err;
1710}
1711
1712struct fuse_retrieve_args {
1713 struct fuse_args_pages ap;
1714 struct fuse_notify_retrieve_in inarg;
1715};
1716
1717static void fuse_retrieve_end(struct fuse_mount *fm, struct fuse_args *args,
1718 int error)
1719{
1720 struct fuse_retrieve_args *ra =
1721 container_of(args, typeof(*ra), ap.args);
1722
1723 release_pages(ra->ap.folios, ra->ap.num_folios);
1724 kfree(ra);
1725}
1726
1727static int fuse_retrieve(struct fuse_mount *fm, struct inode *inode,
1728 struct fuse_notify_retrieve_out *outarg)
1729{
1730 int err;
1731 struct address_space *mapping = inode->i_mapping;
1732 pgoff_t index;
1733 loff_t file_size;
1734 unsigned int num;
1735 unsigned int offset;
1736 size_t total_len = 0;
1737 unsigned int num_pages, cur_pages = 0;
1738 struct fuse_conn *fc = fm->fc;
1739 struct fuse_retrieve_args *ra;
1740 size_t args_size = sizeof(*ra);
1741 struct fuse_args_pages *ap;
1742 struct fuse_args *args;
1743
1744 offset = outarg->offset & ~PAGE_MASK;
1745 file_size = i_size_read(inode);
1746
1747 num = min(outarg->size, fc->max_write);
1748 if (outarg->offset > file_size)
1749 num = 0;
1750 else if (outarg->offset + num > file_size)
1751 num = file_size - outarg->offset;
1752
1753 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1754 num_pages = min(num_pages, fc->max_pages);
1755
1756 args_size += num_pages * (sizeof(ap->folios[0]) + sizeof(ap->descs[0]));
1757
1758 ra = kzalloc(args_size, GFP_KERNEL);
1759 if (!ra)
1760 return -ENOMEM;
1761
1762 ap = &ra->ap;
1763 ap->folios = (void *) (ra + 1);
1764 ap->descs = (void *) (ap->folios + num_pages);
1765
1766 args = &ap->args;
1767 args->nodeid = outarg->nodeid;
1768 args->opcode = FUSE_NOTIFY_REPLY;
1769 args->in_numargs = 2;
1770 args->in_pages = true;
1771 args->end = fuse_retrieve_end;
1772
1773 index = outarg->offset >> PAGE_SHIFT;
1774
1775 while (num && cur_pages < num_pages) {
1776 struct folio *folio;
1777 unsigned int this_num;
1778
1779 folio = filemap_get_folio(mapping, index);
1780 if (IS_ERR(folio))
1781 break;
1782
1783 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1784 ap->folios[ap->num_folios] = folio;
1785 ap->descs[ap->num_folios].offset = offset;
1786 ap->descs[ap->num_folios].length = this_num;
1787 ap->num_folios++;
1788 cur_pages++;
1789
1790 offset = 0;
1791 num -= this_num;
1792 total_len += this_num;
1793 index++;
1794 }
1795 ra->inarg.offset = outarg->offset;
1796 ra->inarg.size = total_len;
1797 args->in_args[0].size = sizeof(ra->inarg);
1798 args->in_args[0].value = &ra->inarg;
1799 args->in_args[1].size = total_len;
1800
1801 err = fuse_simple_notify_reply(fm, args, outarg->notify_unique);
1802 if (err)
1803 fuse_retrieve_end(fm, args, err);
1804
1805 return err;
1806}
1807
1808static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1809 struct fuse_copy_state *cs)
1810{
1811 struct fuse_notify_retrieve_out outarg;
1812 struct fuse_mount *fm;
1813 struct inode *inode;
1814 u64 nodeid;
1815 int err;
1816
1817 err = -EINVAL;
1818 if (size != sizeof(outarg))
1819 goto copy_finish;
1820
1821 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1822 if (err)
1823 goto copy_finish;
1824
1825 fuse_copy_finish(cs);
1826
1827 down_read(&fc->killsb);
1828 err = -ENOENT;
1829 nodeid = outarg.nodeid;
1830
1831 inode = fuse_ilookup(fc, nodeid, &fm);
1832 if (inode) {
1833 err = fuse_retrieve(fm, inode, &outarg);
1834 iput(inode);
1835 }
1836 up_read(&fc->killsb);
1837
1838 return err;
1839
1840copy_finish:
1841 fuse_copy_finish(cs);
1842 return err;
1843}
1844
1845/*
1846 * Resending all processing queue requests.
1847 *
1848 * During a FUSE daemon panics and failover, it is possible for some inflight
1849 * requests to be lost and never returned. As a result, applications awaiting
1850 * replies would become stuck forever. To address this, we can use notification
1851 * to trigger resending of these pending requests to the FUSE daemon, ensuring
1852 * they are properly processed again.
1853 *
1854 * Please note that this strategy is applicable only to idempotent requests or
1855 * if the FUSE daemon takes careful measures to avoid processing duplicated
1856 * non-idempotent requests.
1857 */
1858static void fuse_resend(struct fuse_conn *fc)
1859{
1860 struct fuse_dev *fud;
1861 struct fuse_req *req, *next;
1862 struct fuse_iqueue *fiq = &fc->iq;
1863 LIST_HEAD(to_queue);
1864 unsigned int i;
1865
1866 spin_lock(&fc->lock);
1867 if (!fc->connected) {
1868 spin_unlock(&fc->lock);
1869 return;
1870 }
1871
1872 list_for_each_entry(fud, &fc->devices, entry) {
1873 struct fuse_pqueue *fpq = &fud->pq;
1874
1875 spin_lock(&fpq->lock);
1876 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
1877 list_splice_tail_init(&fpq->processing[i], &to_queue);
1878 spin_unlock(&fpq->lock);
1879 }
1880 spin_unlock(&fc->lock);
1881
1882 list_for_each_entry_safe(req, next, &to_queue, list) {
1883 set_bit(FR_PENDING, &req->flags);
1884 clear_bit(FR_SENT, &req->flags);
1885 /* mark the request as resend request */
1886 req->in.h.unique |= FUSE_UNIQUE_RESEND;
1887 }
1888
1889 spin_lock(&fiq->lock);
1890 if (!fiq->connected) {
1891 spin_unlock(&fiq->lock);
1892 list_for_each_entry(req, &to_queue, list)
1893 clear_bit(FR_PENDING, &req->flags);
1894 end_requests(&to_queue);
1895 return;
1896 }
1897 /* iq and pq requests are both oldest to newest */
1898 list_splice(&to_queue, &fiq->pending);
1899 fuse_dev_wake_and_unlock(fiq);
1900}
1901
1902static int fuse_notify_resend(struct fuse_conn *fc)
1903{
1904 fuse_resend(fc);
1905 return 0;
1906}
1907
1908static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1909 unsigned int size, struct fuse_copy_state *cs)
1910{
1911 /* Don't try to move pages (yet) */
1912 cs->move_pages = 0;
1913
1914 switch (code) {
1915 case FUSE_NOTIFY_POLL:
1916 return fuse_notify_poll(fc, size, cs);
1917
1918 case FUSE_NOTIFY_INVAL_INODE:
1919 return fuse_notify_inval_inode(fc, size, cs);
1920
1921 case FUSE_NOTIFY_INVAL_ENTRY:
1922 return fuse_notify_inval_entry(fc, size, cs);
1923
1924 case FUSE_NOTIFY_STORE:
1925 return fuse_notify_store(fc, size, cs);
1926
1927 case FUSE_NOTIFY_RETRIEVE:
1928 return fuse_notify_retrieve(fc, size, cs);
1929
1930 case FUSE_NOTIFY_DELETE:
1931 return fuse_notify_delete(fc, size, cs);
1932
1933 case FUSE_NOTIFY_RESEND:
1934 return fuse_notify_resend(fc);
1935
1936 default:
1937 fuse_copy_finish(cs);
1938 return -EINVAL;
1939 }
1940}
1941
1942/* Look up request on processing list by unique ID */
1943static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1944{
1945 unsigned int hash = fuse_req_hash(unique);
1946 struct fuse_req *req;
1947
1948 list_for_each_entry(req, &fpq->processing[hash], list) {
1949 if (req->in.h.unique == unique)
1950 return req;
1951 }
1952 return NULL;
1953}
1954
1955static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args,
1956 unsigned nbytes)
1957{
1958 unsigned reqsize = sizeof(struct fuse_out_header);
1959
1960 reqsize += fuse_len_args(args->out_numargs, args->out_args);
1961
1962 if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar))
1963 return -EINVAL;
1964 else if (reqsize > nbytes) {
1965 struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1];
1966 unsigned diffsize = reqsize - nbytes;
1967
1968 if (diffsize > lastarg->size)
1969 return -EINVAL;
1970 lastarg->size -= diffsize;
1971 }
1972 return fuse_copy_args(cs, args->out_numargs, args->out_pages,
1973 args->out_args, args->page_zeroing);
1974}
1975
1976/*
1977 * Write a single reply to a request. First the header is copied from
1978 * the write buffer. The request is then searched on the processing
1979 * list by the unique ID found in the header. If found, then remove
1980 * it from the list and copy the rest of the buffer to the request.
1981 * The request is finished by calling fuse_request_end().
1982 */
1983static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1984 struct fuse_copy_state *cs, size_t nbytes)
1985{
1986 int err;
1987 struct fuse_conn *fc = fud->fc;
1988 struct fuse_pqueue *fpq = &fud->pq;
1989 struct fuse_req *req;
1990 struct fuse_out_header oh;
1991
1992 err = -EINVAL;
1993 if (nbytes < sizeof(struct fuse_out_header))
1994 goto out;
1995
1996 err = fuse_copy_one(cs, &oh, sizeof(oh));
1997 if (err)
1998 goto copy_finish;
1999
2000 err = -EINVAL;
2001 if (oh.len != nbytes)
2002 goto copy_finish;
2003
2004 /*
2005 * Zero oh.unique indicates unsolicited notification message
2006 * and error contains notification code.
2007 */
2008 if (!oh.unique) {
2009 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
2010 goto out;
2011 }
2012
2013 err = -EINVAL;
2014 if (oh.error <= -512 || oh.error > 0)
2015 goto copy_finish;
2016
2017 spin_lock(&fpq->lock);
2018 req = NULL;
2019 if (fpq->connected)
2020 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
2021
2022 err = -ENOENT;
2023 if (!req) {
2024 spin_unlock(&fpq->lock);
2025 goto copy_finish;
2026 }
2027
2028 /* Is it an interrupt reply ID? */
2029 if (oh.unique & FUSE_INT_REQ_BIT) {
2030 __fuse_get_request(req);
2031 spin_unlock(&fpq->lock);
2032
2033 err = 0;
2034 if (nbytes != sizeof(struct fuse_out_header))
2035 err = -EINVAL;
2036 else if (oh.error == -ENOSYS)
2037 fc->no_interrupt = 1;
2038 else if (oh.error == -EAGAIN)
2039 err = queue_interrupt(req);
2040
2041 fuse_put_request(req);
2042
2043 goto copy_finish;
2044 }
2045
2046 clear_bit(FR_SENT, &req->flags);
2047 list_move(&req->list, &fpq->io);
2048 req->out.h = oh;
2049 set_bit(FR_LOCKED, &req->flags);
2050 spin_unlock(&fpq->lock);
2051 cs->req = req;
2052 if (!req->args->page_replace)
2053 cs->move_pages = 0;
2054
2055 if (oh.error)
2056 err = nbytes != sizeof(oh) ? -EINVAL : 0;
2057 else
2058 err = copy_out_args(cs, req->args, nbytes);
2059 fuse_copy_finish(cs);
2060
2061 spin_lock(&fpq->lock);
2062 clear_bit(FR_LOCKED, &req->flags);
2063 if (!fpq->connected)
2064 err = -ENOENT;
2065 else if (err)
2066 req->out.h.error = -EIO;
2067 if (!test_bit(FR_PRIVATE, &req->flags))
2068 list_del_init(&req->list);
2069 spin_unlock(&fpq->lock);
2070
2071 fuse_request_end(req);
2072out:
2073 return err ? err : nbytes;
2074
2075copy_finish:
2076 fuse_copy_finish(cs);
2077 goto out;
2078}
2079
2080static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
2081{
2082 struct fuse_copy_state cs;
2083 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
2084
2085 if (!fud)
2086 return -EPERM;
2087
2088 if (!user_backed_iter(from))
2089 return -EINVAL;
2090
2091 fuse_copy_init(&cs, 0, from);
2092
2093 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
2094}
2095
2096static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
2097 struct file *out, loff_t *ppos,
2098 size_t len, unsigned int flags)
2099{
2100 unsigned int head, tail, mask, count;
2101 unsigned nbuf;
2102 unsigned idx;
2103 struct pipe_buffer *bufs;
2104 struct fuse_copy_state cs;
2105 struct fuse_dev *fud;
2106 size_t rem;
2107 ssize_t ret;
2108
2109 fud = fuse_get_dev(out);
2110 if (!fud)
2111 return -EPERM;
2112
2113 pipe_lock(pipe);
2114
2115 head = pipe->head;
2116 tail = pipe->tail;
2117 mask = pipe->ring_size - 1;
2118 count = head - tail;
2119
2120 bufs = kvmalloc_array(count, sizeof(struct pipe_buffer), GFP_KERNEL);
2121 if (!bufs) {
2122 pipe_unlock(pipe);
2123 return -ENOMEM;
2124 }
2125
2126 nbuf = 0;
2127 rem = 0;
2128 for (idx = tail; idx != head && rem < len; idx++)
2129 rem += pipe->bufs[idx & mask].len;
2130
2131 ret = -EINVAL;
2132 if (rem < len)
2133 goto out_free;
2134
2135 rem = len;
2136 while (rem) {
2137 struct pipe_buffer *ibuf;
2138 struct pipe_buffer *obuf;
2139
2140 if (WARN_ON(nbuf >= count || tail == head))
2141 goto out_free;
2142
2143 ibuf = &pipe->bufs[tail & mask];
2144 obuf = &bufs[nbuf];
2145
2146 if (rem >= ibuf->len) {
2147 *obuf = *ibuf;
2148 ibuf->ops = NULL;
2149 tail++;
2150 pipe->tail = tail;
2151 } else {
2152 if (!pipe_buf_get(pipe, ibuf))
2153 goto out_free;
2154
2155 *obuf = *ibuf;
2156 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2157 obuf->len = rem;
2158 ibuf->offset += obuf->len;
2159 ibuf->len -= obuf->len;
2160 }
2161 nbuf++;
2162 rem -= obuf->len;
2163 }
2164 pipe_unlock(pipe);
2165
2166 fuse_copy_init(&cs, 0, NULL);
2167 cs.pipebufs = bufs;
2168 cs.nr_segs = nbuf;
2169 cs.pipe = pipe;
2170
2171 if (flags & SPLICE_F_MOVE)
2172 cs.move_pages = 1;
2173
2174 ret = fuse_dev_do_write(fud, &cs, len);
2175
2176 pipe_lock(pipe);
2177out_free:
2178 for (idx = 0; idx < nbuf; idx++) {
2179 struct pipe_buffer *buf = &bufs[idx];
2180
2181 if (buf->ops)
2182 pipe_buf_release(pipe, buf);
2183 }
2184 pipe_unlock(pipe);
2185
2186 kvfree(bufs);
2187 return ret;
2188}
2189
2190static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2191{
2192 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
2193 struct fuse_iqueue *fiq;
2194 struct fuse_dev *fud = fuse_get_dev(file);
2195
2196 if (!fud)
2197 return EPOLLERR;
2198
2199 fiq = &fud->fc->iq;
2200 poll_wait(file, &fiq->waitq, wait);
2201
2202 spin_lock(&fiq->lock);
2203 if (!fiq->connected)
2204 mask = EPOLLERR;
2205 else if (request_pending(fiq))
2206 mask |= EPOLLIN | EPOLLRDNORM;
2207 spin_unlock(&fiq->lock);
2208
2209 return mask;
2210}
2211
2212/* Abort all requests on the given list (pending or processing) */
2213static void end_requests(struct list_head *head)
2214{
2215 while (!list_empty(head)) {
2216 struct fuse_req *req;
2217 req = list_entry(head->next, struct fuse_req, list);
2218 req->out.h.error = -ECONNABORTED;
2219 clear_bit(FR_SENT, &req->flags);
2220 list_del_init(&req->list);
2221 fuse_request_end(req);
2222 }
2223}
2224
2225static void end_polls(struct fuse_conn *fc)
2226{
2227 struct rb_node *p;
2228
2229 p = rb_first(&fc->polled_files);
2230
2231 while (p) {
2232 struct fuse_file *ff;
2233 ff = rb_entry(p, struct fuse_file, polled_node);
2234 wake_up_interruptible_all(&ff->poll_wait);
2235
2236 p = rb_next(p);
2237 }
2238}
2239
2240/*
2241 * Abort all requests.
2242 *
2243 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2244 * filesystem.
2245 *
2246 * The same effect is usually achievable through killing the filesystem daemon
2247 * and all users of the filesystem. The exception is the combination of an
2248 * asynchronous request and the tricky deadlock (see
2249 * Documentation/filesystems/fuse.rst).
2250 *
2251 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2252 * requests, they should be finished off immediately. Locked requests will be
2253 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2254 * requests. It is possible that some request will finish before we can. This
2255 * is OK, the request will in that case be removed from the list before we touch
2256 * it.
2257 */
2258void fuse_abort_conn(struct fuse_conn *fc)
2259{
2260 struct fuse_iqueue *fiq = &fc->iq;
2261
2262 spin_lock(&fc->lock);
2263 if (fc->connected) {
2264 struct fuse_dev *fud;
2265 struct fuse_req *req, *next;
2266 LIST_HEAD(to_end);
2267 unsigned int i;
2268
2269 /* Background queuing checks fc->connected under bg_lock */
2270 spin_lock(&fc->bg_lock);
2271 fc->connected = 0;
2272 spin_unlock(&fc->bg_lock);
2273
2274 fuse_set_initialized(fc);
2275 list_for_each_entry(fud, &fc->devices, entry) {
2276 struct fuse_pqueue *fpq = &fud->pq;
2277
2278 spin_lock(&fpq->lock);
2279 fpq->connected = 0;
2280 list_for_each_entry_safe(req, next, &fpq->io, list) {
2281 req->out.h.error = -ECONNABORTED;
2282 spin_lock(&req->waitq.lock);
2283 set_bit(FR_ABORTED, &req->flags);
2284 if (!test_bit(FR_LOCKED, &req->flags)) {
2285 set_bit(FR_PRIVATE, &req->flags);
2286 __fuse_get_request(req);
2287 list_move(&req->list, &to_end);
2288 }
2289 spin_unlock(&req->waitq.lock);
2290 }
2291 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2292 list_splice_tail_init(&fpq->processing[i],
2293 &to_end);
2294 spin_unlock(&fpq->lock);
2295 }
2296 spin_lock(&fc->bg_lock);
2297 fc->blocked = 0;
2298 fc->max_background = UINT_MAX;
2299 flush_bg_queue(fc);
2300 spin_unlock(&fc->bg_lock);
2301
2302 spin_lock(&fiq->lock);
2303 fiq->connected = 0;
2304 list_for_each_entry(req, &fiq->pending, list)
2305 clear_bit(FR_PENDING, &req->flags);
2306 list_splice_tail_init(&fiq->pending, &to_end);
2307 while (forget_pending(fiq))
2308 kfree(fuse_dequeue_forget(fiq, 1, NULL));
2309 wake_up_all(&fiq->waitq);
2310 spin_unlock(&fiq->lock);
2311 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2312 end_polls(fc);
2313 wake_up_all(&fc->blocked_waitq);
2314 spin_unlock(&fc->lock);
2315
2316 end_requests(&to_end);
2317 } else {
2318 spin_unlock(&fc->lock);
2319 }
2320}
2321EXPORT_SYMBOL_GPL(fuse_abort_conn);
2322
2323void fuse_wait_aborted(struct fuse_conn *fc)
2324{
2325 /* matches implicit memory barrier in fuse_drop_waiting() */
2326 smp_mb();
2327 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2328}
2329
2330int fuse_dev_release(struct inode *inode, struct file *file)
2331{
2332 struct fuse_dev *fud = fuse_get_dev(file);
2333
2334 if (fud) {
2335 struct fuse_conn *fc = fud->fc;
2336 struct fuse_pqueue *fpq = &fud->pq;
2337 LIST_HEAD(to_end);
2338 unsigned int i;
2339
2340 spin_lock(&fpq->lock);
2341 WARN_ON(!list_empty(&fpq->io));
2342 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2343 list_splice_init(&fpq->processing[i], &to_end);
2344 spin_unlock(&fpq->lock);
2345
2346 end_requests(&to_end);
2347
2348 /* Are we the last open device? */
2349 if (atomic_dec_and_test(&fc->dev_count)) {
2350 WARN_ON(fc->iq.fasync != NULL);
2351 fuse_abort_conn(fc);
2352 }
2353 fuse_dev_free(fud);
2354 }
2355 return 0;
2356}
2357EXPORT_SYMBOL_GPL(fuse_dev_release);
2358
2359static int fuse_dev_fasync(int fd, struct file *file, int on)
2360{
2361 struct fuse_dev *fud = fuse_get_dev(file);
2362
2363 if (!fud)
2364 return -EPERM;
2365
2366 /* No locking - fasync_helper does its own locking */
2367 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2368}
2369
2370static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2371{
2372 struct fuse_dev *fud;
2373
2374 if (new->private_data)
2375 return -EINVAL;
2376
2377 fud = fuse_dev_alloc_install(fc);
2378 if (!fud)
2379 return -ENOMEM;
2380
2381 new->private_data = fud;
2382 atomic_inc(&fc->dev_count);
2383
2384 return 0;
2385}
2386
2387static long fuse_dev_ioctl_clone(struct file *file, __u32 __user *argp)
2388{
2389 int res;
2390 int oldfd;
2391 struct fuse_dev *fud = NULL;
2392
2393 if (get_user(oldfd, argp))
2394 return -EFAULT;
2395
2396 CLASS(fd, f)(oldfd);
2397 if (fd_empty(f))
2398 return -EINVAL;
2399
2400 /*
2401 * Check against file->f_op because CUSE
2402 * uses the same ioctl handler.
2403 */
2404 if (fd_file(f)->f_op == file->f_op)
2405 fud = fuse_get_dev(fd_file(f));
2406
2407 res = -EINVAL;
2408 if (fud) {
2409 mutex_lock(&fuse_mutex);
2410 res = fuse_device_clone(fud->fc, file);
2411 mutex_unlock(&fuse_mutex);
2412 }
2413
2414 return res;
2415}
2416
2417static long fuse_dev_ioctl_backing_open(struct file *file,
2418 struct fuse_backing_map __user *argp)
2419{
2420 struct fuse_dev *fud = fuse_get_dev(file);
2421 struct fuse_backing_map map;
2422
2423 if (!fud)
2424 return -EPERM;
2425
2426 if (!IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
2427 return -EOPNOTSUPP;
2428
2429 if (copy_from_user(&map, argp, sizeof(map)))
2430 return -EFAULT;
2431
2432 return fuse_backing_open(fud->fc, &map);
2433}
2434
2435static long fuse_dev_ioctl_backing_close(struct file *file, __u32 __user *argp)
2436{
2437 struct fuse_dev *fud = fuse_get_dev(file);
2438 int backing_id;
2439
2440 if (!fud)
2441 return -EPERM;
2442
2443 if (!IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
2444 return -EOPNOTSUPP;
2445
2446 if (get_user(backing_id, argp))
2447 return -EFAULT;
2448
2449 return fuse_backing_close(fud->fc, backing_id);
2450}
2451
2452static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2453 unsigned long arg)
2454{
2455 void __user *argp = (void __user *)arg;
2456
2457 switch (cmd) {
2458 case FUSE_DEV_IOC_CLONE:
2459 return fuse_dev_ioctl_clone(file, argp);
2460
2461 case FUSE_DEV_IOC_BACKING_OPEN:
2462 return fuse_dev_ioctl_backing_open(file, argp);
2463
2464 case FUSE_DEV_IOC_BACKING_CLOSE:
2465 return fuse_dev_ioctl_backing_close(file, argp);
2466
2467 default:
2468 return -ENOTTY;
2469 }
2470}
2471
2472const struct file_operations fuse_dev_operations = {
2473 .owner = THIS_MODULE,
2474 .open = fuse_dev_open,
2475 .read_iter = fuse_dev_read,
2476 .splice_read = fuse_dev_splice_read,
2477 .write_iter = fuse_dev_write,
2478 .splice_write = fuse_dev_splice_write,
2479 .poll = fuse_dev_poll,
2480 .release = fuse_dev_release,
2481 .fasync = fuse_dev_fasync,
2482 .unlocked_ioctl = fuse_dev_ioctl,
2483 .compat_ioctl = compat_ptr_ioctl,
2484};
2485EXPORT_SYMBOL_GPL(fuse_dev_operations);
2486
2487static struct miscdevice fuse_miscdevice = {
2488 .minor = FUSE_MINOR,
2489 .name = "fuse",
2490 .fops = &fuse_dev_operations,
2491};
2492
2493int __init fuse_dev_init(void)
2494{
2495 int err = -ENOMEM;
2496 fuse_req_cachep = kmem_cache_create("fuse_request",
2497 sizeof(struct fuse_req),
2498 0, 0, NULL);
2499 if (!fuse_req_cachep)
2500 goto out;
2501
2502 err = misc_register(&fuse_miscdevice);
2503 if (err)
2504 goto out_cache_clean;
2505
2506 return 0;
2507
2508 out_cache_clean:
2509 kmem_cache_destroy(fuse_req_cachep);
2510 out:
2511 return err;
2512}
2513
2514void fuse_dev_cleanup(void)
2515{
2516 misc_deregister(&fuse_miscdevice);
2517 kmem_cache_destroy(fuse_req_cachep);
2518}