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/uio.h>
15#include <linux/miscdevice.h>
16#include <linux/pagemap.h>
17#include <linux/file.h>
18#include <linux/slab.h>
19#include <linux/pipe_fs_i.h>
20#include <linux/swap.h>
21#include <linux/splice.h>
22
23MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24MODULE_ALIAS("devname:fuse");
25
26static struct kmem_cache *fuse_req_cachep;
27
28static struct fuse_conn *fuse_get_conn(struct file *file)
29{
30 /*
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
33 */
34 return file->private_data;
35}
36
37static void fuse_request_init(struct fuse_req *req)
38{
39 memset(req, 0, sizeof(*req));
40 INIT_LIST_HEAD(&req->list);
41 INIT_LIST_HEAD(&req->intr_entry);
42 init_waitqueue_head(&req->waitq);
43 atomic_set(&req->count, 1);
44}
45
46struct fuse_req *fuse_request_alloc(void)
47{
48 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL);
49 if (req)
50 fuse_request_init(req);
51 return req;
52}
53EXPORT_SYMBOL_GPL(fuse_request_alloc);
54
55struct fuse_req *fuse_request_alloc_nofs(void)
56{
57 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_NOFS);
58 if (req)
59 fuse_request_init(req);
60 return req;
61}
62
63void fuse_request_free(struct fuse_req *req)
64{
65 kmem_cache_free(fuse_req_cachep, req);
66}
67
68static void block_sigs(sigset_t *oldset)
69{
70 sigset_t mask;
71
72 siginitsetinv(&mask, sigmask(SIGKILL));
73 sigprocmask(SIG_BLOCK, &mask, oldset);
74}
75
76static void restore_sigs(sigset_t *oldset)
77{
78 sigprocmask(SIG_SETMASK, oldset, NULL);
79}
80
81static void __fuse_get_request(struct fuse_req *req)
82{
83 atomic_inc(&req->count);
84}
85
86/* Must be called with > 1 refcount */
87static void __fuse_put_request(struct fuse_req *req)
88{
89 BUG_ON(atomic_read(&req->count) < 2);
90 atomic_dec(&req->count);
91}
92
93static void fuse_req_init_context(struct fuse_req *req)
94{
95 req->in.h.uid = current_fsuid();
96 req->in.h.gid = current_fsgid();
97 req->in.h.pid = current->pid;
98}
99
100struct fuse_req *fuse_get_req(struct fuse_conn *fc)
101{
102 struct fuse_req *req;
103 sigset_t oldset;
104 int intr;
105 int err;
106
107 atomic_inc(&fc->num_waiting);
108 block_sigs(&oldset);
109 intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked);
110 restore_sigs(&oldset);
111 err = -EINTR;
112 if (intr)
113 goto out;
114
115 err = -ENOTCONN;
116 if (!fc->connected)
117 goto out;
118
119 req = fuse_request_alloc();
120 err = -ENOMEM;
121 if (!req)
122 goto out;
123
124 fuse_req_init_context(req);
125 req->waiting = 1;
126 return req;
127
128 out:
129 atomic_dec(&fc->num_waiting);
130 return ERR_PTR(err);
131}
132EXPORT_SYMBOL_GPL(fuse_get_req);
133
134/*
135 * Return request in fuse_file->reserved_req. However that may
136 * currently be in use. If that is the case, wait for it to become
137 * available.
138 */
139static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
140 struct file *file)
141{
142 struct fuse_req *req = NULL;
143 struct fuse_file *ff = file->private_data;
144
145 do {
146 wait_event(fc->reserved_req_waitq, ff->reserved_req);
147 spin_lock(&fc->lock);
148 if (ff->reserved_req) {
149 req = ff->reserved_req;
150 ff->reserved_req = NULL;
151 get_file(file);
152 req->stolen_file = file;
153 }
154 spin_unlock(&fc->lock);
155 } while (!req);
156
157 return req;
158}
159
160/*
161 * Put stolen request back into fuse_file->reserved_req
162 */
163static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
164{
165 struct file *file = req->stolen_file;
166 struct fuse_file *ff = file->private_data;
167
168 spin_lock(&fc->lock);
169 fuse_request_init(req);
170 BUG_ON(ff->reserved_req);
171 ff->reserved_req = req;
172 wake_up_all(&fc->reserved_req_waitq);
173 spin_unlock(&fc->lock);
174 fput(file);
175}
176
177/*
178 * Gets a requests for a file operation, always succeeds
179 *
180 * This is used for sending the FLUSH request, which must get to
181 * userspace, due to POSIX locks which may need to be unlocked.
182 *
183 * If allocation fails due to OOM, use the reserved request in
184 * fuse_file.
185 *
186 * This is very unlikely to deadlock accidentally, since the
187 * filesystem should not have it's own file open. If deadlock is
188 * intentional, it can still be broken by "aborting" the filesystem.
189 */
190struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file)
191{
192 struct fuse_req *req;
193
194 atomic_inc(&fc->num_waiting);
195 wait_event(fc->blocked_waitq, !fc->blocked);
196 req = fuse_request_alloc();
197 if (!req)
198 req = get_reserved_req(fc, file);
199
200 fuse_req_init_context(req);
201 req->waiting = 1;
202 return req;
203}
204
205void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
206{
207 if (atomic_dec_and_test(&req->count)) {
208 if (req->waiting)
209 atomic_dec(&fc->num_waiting);
210
211 if (req->stolen_file)
212 put_reserved_req(fc, req);
213 else
214 fuse_request_free(req);
215 }
216}
217EXPORT_SYMBOL_GPL(fuse_put_request);
218
219static unsigned len_args(unsigned numargs, struct fuse_arg *args)
220{
221 unsigned nbytes = 0;
222 unsigned i;
223
224 for (i = 0; i < numargs; i++)
225 nbytes += args[i].size;
226
227 return nbytes;
228}
229
230static u64 fuse_get_unique(struct fuse_conn *fc)
231{
232 fc->reqctr++;
233 /* zero is special */
234 if (fc->reqctr == 0)
235 fc->reqctr = 1;
236
237 return fc->reqctr;
238}
239
240static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
241{
242 req->in.h.len = sizeof(struct fuse_in_header) +
243 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
244 list_add_tail(&req->list, &fc->pending);
245 req->state = FUSE_REQ_PENDING;
246 if (!req->waiting) {
247 req->waiting = 1;
248 atomic_inc(&fc->num_waiting);
249 }
250 wake_up(&fc->waitq);
251 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
252}
253
254void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
255 u64 nodeid, u64 nlookup)
256{
257 forget->forget_one.nodeid = nodeid;
258 forget->forget_one.nlookup = nlookup;
259
260 spin_lock(&fc->lock);
261 if (fc->connected) {
262 fc->forget_list_tail->next = forget;
263 fc->forget_list_tail = forget;
264 wake_up(&fc->waitq);
265 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
266 } else {
267 kfree(forget);
268 }
269 spin_unlock(&fc->lock);
270}
271
272static void flush_bg_queue(struct fuse_conn *fc)
273{
274 while (fc->active_background < fc->max_background &&
275 !list_empty(&fc->bg_queue)) {
276 struct fuse_req *req;
277
278 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
279 list_del(&req->list);
280 fc->active_background++;
281 req->in.h.unique = fuse_get_unique(fc);
282 queue_request(fc, req);
283 }
284}
285
286/*
287 * This function is called when a request is finished. Either a reply
288 * has arrived or it was aborted (and not yet sent) or some error
289 * occurred during communication with userspace, or the device file
290 * was closed. The requester thread is woken up (if still waiting),
291 * the 'end' callback is called if given, else the reference to the
292 * request is released
293 *
294 * Called with fc->lock, unlocks it
295 */
296static void request_end(struct fuse_conn *fc, struct fuse_req *req)
297__releases(fc->lock)
298{
299 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
300 req->end = NULL;
301 list_del(&req->list);
302 list_del(&req->intr_entry);
303 req->state = FUSE_REQ_FINISHED;
304 if (req->background) {
305 if (fc->num_background == fc->max_background) {
306 fc->blocked = 0;
307 wake_up_all(&fc->blocked_waitq);
308 }
309 if (fc->num_background == fc->congestion_threshold &&
310 fc->connected && fc->bdi_initialized) {
311 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
312 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
313 }
314 fc->num_background--;
315 fc->active_background--;
316 flush_bg_queue(fc);
317 }
318 spin_unlock(&fc->lock);
319 wake_up(&req->waitq);
320 if (end)
321 end(fc, req);
322 fuse_put_request(fc, req);
323}
324
325static void wait_answer_interruptible(struct fuse_conn *fc,
326 struct fuse_req *req)
327__releases(fc->lock)
328__acquires(fc->lock)
329{
330 if (signal_pending(current))
331 return;
332
333 spin_unlock(&fc->lock);
334 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
335 spin_lock(&fc->lock);
336}
337
338static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
339{
340 list_add_tail(&req->intr_entry, &fc->interrupts);
341 wake_up(&fc->waitq);
342 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
343}
344
345static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
346__releases(fc->lock)
347__acquires(fc->lock)
348{
349 if (!fc->no_interrupt) {
350 /* Any signal may interrupt this */
351 wait_answer_interruptible(fc, req);
352
353 if (req->aborted)
354 goto aborted;
355 if (req->state == FUSE_REQ_FINISHED)
356 return;
357
358 req->interrupted = 1;
359 if (req->state == FUSE_REQ_SENT)
360 queue_interrupt(fc, req);
361 }
362
363 if (!req->force) {
364 sigset_t oldset;
365
366 /* Only fatal signals may interrupt this */
367 block_sigs(&oldset);
368 wait_answer_interruptible(fc, req);
369 restore_sigs(&oldset);
370
371 if (req->aborted)
372 goto aborted;
373 if (req->state == FUSE_REQ_FINISHED)
374 return;
375
376 /* Request is not yet in userspace, bail out */
377 if (req->state == FUSE_REQ_PENDING) {
378 list_del(&req->list);
379 __fuse_put_request(req);
380 req->out.h.error = -EINTR;
381 return;
382 }
383 }
384
385 /*
386 * Either request is already in userspace, or it was forced.
387 * Wait it out.
388 */
389 spin_unlock(&fc->lock);
390 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
391 spin_lock(&fc->lock);
392
393 if (!req->aborted)
394 return;
395
396 aborted:
397 BUG_ON(req->state != FUSE_REQ_FINISHED);
398 if (req->locked) {
399 /* This is uninterruptible sleep, because data is
400 being copied to/from the buffers of req. During
401 locked state, there mustn't be any filesystem
402 operation (e.g. page fault), since that could lead
403 to deadlock */
404 spin_unlock(&fc->lock);
405 wait_event(req->waitq, !req->locked);
406 spin_lock(&fc->lock);
407 }
408}
409
410void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
411{
412 req->isreply = 1;
413 spin_lock(&fc->lock);
414 if (!fc->connected)
415 req->out.h.error = -ENOTCONN;
416 else if (fc->conn_error)
417 req->out.h.error = -ECONNREFUSED;
418 else {
419 req->in.h.unique = fuse_get_unique(fc);
420 queue_request(fc, req);
421 /* acquire extra reference, since request is still needed
422 after request_end() */
423 __fuse_get_request(req);
424
425 request_wait_answer(fc, req);
426 }
427 spin_unlock(&fc->lock);
428}
429EXPORT_SYMBOL_GPL(fuse_request_send);
430
431static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
432 struct fuse_req *req)
433{
434 req->background = 1;
435 fc->num_background++;
436 if (fc->num_background == fc->max_background)
437 fc->blocked = 1;
438 if (fc->num_background == fc->congestion_threshold &&
439 fc->bdi_initialized) {
440 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
441 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
442 }
443 list_add_tail(&req->list, &fc->bg_queue);
444 flush_bg_queue(fc);
445}
446
447static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
448{
449 spin_lock(&fc->lock);
450 if (fc->connected) {
451 fuse_request_send_nowait_locked(fc, req);
452 spin_unlock(&fc->lock);
453 } else {
454 req->out.h.error = -ENOTCONN;
455 request_end(fc, req);
456 }
457}
458
459void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
460{
461 req->isreply = 1;
462 fuse_request_send_nowait(fc, req);
463}
464EXPORT_SYMBOL_GPL(fuse_request_send_background);
465
466static int fuse_request_send_notify_reply(struct fuse_conn *fc,
467 struct fuse_req *req, u64 unique)
468{
469 int err = -ENODEV;
470
471 req->isreply = 0;
472 req->in.h.unique = unique;
473 spin_lock(&fc->lock);
474 if (fc->connected) {
475 queue_request(fc, req);
476 err = 0;
477 }
478 spin_unlock(&fc->lock);
479
480 return err;
481}
482
483/*
484 * Called under fc->lock
485 *
486 * fc->connected must have been checked previously
487 */
488void fuse_request_send_background_locked(struct fuse_conn *fc,
489 struct fuse_req *req)
490{
491 req->isreply = 1;
492 fuse_request_send_nowait_locked(fc, req);
493}
494
495/*
496 * Lock the request. Up to the next unlock_request() there mustn't be
497 * anything that could cause a page-fault. If the request was already
498 * aborted bail out.
499 */
500static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
501{
502 int err = 0;
503 if (req) {
504 spin_lock(&fc->lock);
505 if (req->aborted)
506 err = -ENOENT;
507 else
508 req->locked = 1;
509 spin_unlock(&fc->lock);
510 }
511 return err;
512}
513
514/*
515 * Unlock request. If it was aborted during being locked, the
516 * requester thread is currently waiting for it to be unlocked, so
517 * wake it up.
518 */
519static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
520{
521 if (req) {
522 spin_lock(&fc->lock);
523 req->locked = 0;
524 if (req->aborted)
525 wake_up(&req->waitq);
526 spin_unlock(&fc->lock);
527 }
528}
529
530struct fuse_copy_state {
531 struct fuse_conn *fc;
532 int write;
533 struct fuse_req *req;
534 const struct iovec *iov;
535 struct pipe_buffer *pipebufs;
536 struct pipe_buffer *currbuf;
537 struct pipe_inode_info *pipe;
538 unsigned long nr_segs;
539 unsigned long seglen;
540 unsigned long addr;
541 struct page *pg;
542 void *mapaddr;
543 void *buf;
544 unsigned len;
545 unsigned move_pages:1;
546};
547
548static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
549 int write,
550 const struct iovec *iov, unsigned long nr_segs)
551{
552 memset(cs, 0, sizeof(*cs));
553 cs->fc = fc;
554 cs->write = write;
555 cs->iov = iov;
556 cs->nr_segs = nr_segs;
557}
558
559/* Unmap and put previous page of userspace buffer */
560static void fuse_copy_finish(struct fuse_copy_state *cs)
561{
562 if (cs->currbuf) {
563 struct pipe_buffer *buf = cs->currbuf;
564
565 if (!cs->write) {
566 buf->ops->unmap(cs->pipe, buf, cs->mapaddr);
567 } else {
568 kunmap(buf->page);
569 buf->len = PAGE_SIZE - cs->len;
570 }
571 cs->currbuf = NULL;
572 cs->mapaddr = NULL;
573 } else if (cs->mapaddr) {
574 kunmap(cs->pg);
575 if (cs->write) {
576 flush_dcache_page(cs->pg);
577 set_page_dirty_lock(cs->pg);
578 }
579 put_page(cs->pg);
580 cs->mapaddr = NULL;
581 }
582}
583
584/*
585 * Get another pagefull of userspace buffer, and map it to kernel
586 * address space, and lock request
587 */
588static int fuse_copy_fill(struct fuse_copy_state *cs)
589{
590 unsigned long offset;
591 int err;
592
593 unlock_request(cs->fc, cs->req);
594 fuse_copy_finish(cs);
595 if (cs->pipebufs) {
596 struct pipe_buffer *buf = cs->pipebufs;
597
598 if (!cs->write) {
599 err = buf->ops->confirm(cs->pipe, buf);
600 if (err)
601 return err;
602
603 BUG_ON(!cs->nr_segs);
604 cs->currbuf = buf;
605 cs->mapaddr = buf->ops->map(cs->pipe, buf, 0);
606 cs->len = buf->len;
607 cs->buf = cs->mapaddr + buf->offset;
608 cs->pipebufs++;
609 cs->nr_segs--;
610 } else {
611 struct page *page;
612
613 if (cs->nr_segs == cs->pipe->buffers)
614 return -EIO;
615
616 page = alloc_page(GFP_HIGHUSER);
617 if (!page)
618 return -ENOMEM;
619
620 buf->page = page;
621 buf->offset = 0;
622 buf->len = 0;
623
624 cs->currbuf = buf;
625 cs->mapaddr = kmap(page);
626 cs->buf = cs->mapaddr;
627 cs->len = PAGE_SIZE;
628 cs->pipebufs++;
629 cs->nr_segs++;
630 }
631 } else {
632 if (!cs->seglen) {
633 BUG_ON(!cs->nr_segs);
634 cs->seglen = cs->iov[0].iov_len;
635 cs->addr = (unsigned long) cs->iov[0].iov_base;
636 cs->iov++;
637 cs->nr_segs--;
638 }
639 err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg);
640 if (err < 0)
641 return err;
642 BUG_ON(err != 1);
643 offset = cs->addr % PAGE_SIZE;
644 cs->mapaddr = kmap(cs->pg);
645 cs->buf = cs->mapaddr + offset;
646 cs->len = min(PAGE_SIZE - offset, cs->seglen);
647 cs->seglen -= cs->len;
648 cs->addr += cs->len;
649 }
650
651 return lock_request(cs->fc, cs->req);
652}
653
654/* Do as much copy to/from userspace buffer as we can */
655static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
656{
657 unsigned ncpy = min(*size, cs->len);
658 if (val) {
659 if (cs->write)
660 memcpy(cs->buf, *val, ncpy);
661 else
662 memcpy(*val, cs->buf, ncpy);
663 *val += ncpy;
664 }
665 *size -= ncpy;
666 cs->len -= ncpy;
667 cs->buf += ncpy;
668 return ncpy;
669}
670
671static int fuse_check_page(struct page *page)
672{
673 if (page_mapcount(page) ||
674 page->mapping != NULL ||
675 page_count(page) != 1 ||
676 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
677 ~(1 << PG_locked |
678 1 << PG_referenced |
679 1 << PG_uptodate |
680 1 << PG_lru |
681 1 << PG_active |
682 1 << PG_reclaim))) {
683 printk(KERN_WARNING "fuse: trying to steal weird page\n");
684 printk(KERN_WARNING " page=%p index=%li flags=%08lx, count=%i, mapcount=%i, mapping=%p\n", page, page->index, page->flags, page_count(page), page_mapcount(page), page->mapping);
685 return 1;
686 }
687 return 0;
688}
689
690static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
691{
692 int err;
693 struct page *oldpage = *pagep;
694 struct page *newpage;
695 struct pipe_buffer *buf = cs->pipebufs;
696 struct address_space *mapping;
697 pgoff_t index;
698
699 unlock_request(cs->fc, cs->req);
700 fuse_copy_finish(cs);
701
702 err = buf->ops->confirm(cs->pipe, buf);
703 if (err)
704 return err;
705
706 BUG_ON(!cs->nr_segs);
707 cs->currbuf = buf;
708 cs->len = buf->len;
709 cs->pipebufs++;
710 cs->nr_segs--;
711
712 if (cs->len != PAGE_SIZE)
713 goto out_fallback;
714
715 if (buf->ops->steal(cs->pipe, buf) != 0)
716 goto out_fallback;
717
718 newpage = buf->page;
719
720 if (WARN_ON(!PageUptodate(newpage)))
721 return -EIO;
722
723 ClearPageMappedToDisk(newpage);
724
725 if (fuse_check_page(newpage) != 0)
726 goto out_fallback_unlock;
727
728 mapping = oldpage->mapping;
729 index = oldpage->index;
730
731 /*
732 * This is a new and locked page, it shouldn't be mapped or
733 * have any special flags on it
734 */
735 if (WARN_ON(page_mapped(oldpage)))
736 goto out_fallback_unlock;
737 if (WARN_ON(page_has_private(oldpage)))
738 goto out_fallback_unlock;
739 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
740 goto out_fallback_unlock;
741 if (WARN_ON(PageMlocked(oldpage)))
742 goto out_fallback_unlock;
743
744 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
745 if (err) {
746 unlock_page(newpage);
747 return err;
748 }
749
750 page_cache_get(newpage);
751
752 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
753 lru_cache_add_file(newpage);
754
755 err = 0;
756 spin_lock(&cs->fc->lock);
757 if (cs->req->aborted)
758 err = -ENOENT;
759 else
760 *pagep = newpage;
761 spin_unlock(&cs->fc->lock);
762
763 if (err) {
764 unlock_page(newpage);
765 page_cache_release(newpage);
766 return err;
767 }
768
769 unlock_page(oldpage);
770 page_cache_release(oldpage);
771 cs->len = 0;
772
773 return 0;
774
775out_fallback_unlock:
776 unlock_page(newpage);
777out_fallback:
778 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
779 cs->buf = cs->mapaddr + buf->offset;
780
781 err = lock_request(cs->fc, cs->req);
782 if (err)
783 return err;
784
785 return 1;
786}
787
788static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
789 unsigned offset, unsigned count)
790{
791 struct pipe_buffer *buf;
792
793 if (cs->nr_segs == cs->pipe->buffers)
794 return -EIO;
795
796 unlock_request(cs->fc, cs->req);
797 fuse_copy_finish(cs);
798
799 buf = cs->pipebufs;
800 page_cache_get(page);
801 buf->page = page;
802 buf->offset = offset;
803 buf->len = count;
804
805 cs->pipebufs++;
806 cs->nr_segs++;
807 cs->len = 0;
808
809 return 0;
810}
811
812/*
813 * Copy a page in the request to/from the userspace buffer. Must be
814 * done atomically
815 */
816static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
817 unsigned offset, unsigned count, int zeroing)
818{
819 int err;
820 struct page *page = *pagep;
821
822 if (page && zeroing && count < PAGE_SIZE)
823 clear_highpage(page);
824
825 while (count) {
826 if (cs->write && cs->pipebufs && page) {
827 return fuse_ref_page(cs, page, offset, count);
828 } else if (!cs->len) {
829 if (cs->move_pages && page &&
830 offset == 0 && count == PAGE_SIZE) {
831 err = fuse_try_move_page(cs, pagep);
832 if (err <= 0)
833 return err;
834 } else {
835 err = fuse_copy_fill(cs);
836 if (err)
837 return err;
838 }
839 }
840 if (page) {
841 void *mapaddr = kmap_atomic(page);
842 void *buf = mapaddr + offset;
843 offset += fuse_copy_do(cs, &buf, &count);
844 kunmap_atomic(mapaddr);
845 } else
846 offset += fuse_copy_do(cs, NULL, &count);
847 }
848 if (page && !cs->write)
849 flush_dcache_page(page);
850 return 0;
851}
852
853/* Copy pages in the request to/from userspace buffer */
854static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
855 int zeroing)
856{
857 unsigned i;
858 struct fuse_req *req = cs->req;
859 unsigned offset = req->page_offset;
860 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
861
862 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
863 int err;
864
865 err = fuse_copy_page(cs, &req->pages[i], offset, count,
866 zeroing);
867 if (err)
868 return err;
869
870 nbytes -= count;
871 count = min(nbytes, (unsigned) PAGE_SIZE);
872 offset = 0;
873 }
874 return 0;
875}
876
877/* Copy a single argument in the request to/from userspace buffer */
878static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
879{
880 while (size) {
881 if (!cs->len) {
882 int err = fuse_copy_fill(cs);
883 if (err)
884 return err;
885 }
886 fuse_copy_do(cs, &val, &size);
887 }
888 return 0;
889}
890
891/* Copy request arguments to/from userspace buffer */
892static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
893 unsigned argpages, struct fuse_arg *args,
894 int zeroing)
895{
896 int err = 0;
897 unsigned i;
898
899 for (i = 0; !err && i < numargs; i++) {
900 struct fuse_arg *arg = &args[i];
901 if (i == numargs - 1 && argpages)
902 err = fuse_copy_pages(cs, arg->size, zeroing);
903 else
904 err = fuse_copy_one(cs, arg->value, arg->size);
905 }
906 return err;
907}
908
909static int forget_pending(struct fuse_conn *fc)
910{
911 return fc->forget_list_head.next != NULL;
912}
913
914static int request_pending(struct fuse_conn *fc)
915{
916 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
917 forget_pending(fc);
918}
919
920/* Wait until a request is available on the pending list */
921static void request_wait(struct fuse_conn *fc)
922__releases(fc->lock)
923__acquires(fc->lock)
924{
925 DECLARE_WAITQUEUE(wait, current);
926
927 add_wait_queue_exclusive(&fc->waitq, &wait);
928 while (fc->connected && !request_pending(fc)) {
929 set_current_state(TASK_INTERRUPTIBLE);
930 if (signal_pending(current))
931 break;
932
933 spin_unlock(&fc->lock);
934 schedule();
935 spin_lock(&fc->lock);
936 }
937 set_current_state(TASK_RUNNING);
938 remove_wait_queue(&fc->waitq, &wait);
939}
940
941/*
942 * Transfer an interrupt request to userspace
943 *
944 * Unlike other requests this is assembled on demand, without a need
945 * to allocate a separate fuse_req structure.
946 *
947 * Called with fc->lock held, releases it
948 */
949static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
950 size_t nbytes, struct fuse_req *req)
951__releases(fc->lock)
952{
953 struct fuse_in_header ih;
954 struct fuse_interrupt_in arg;
955 unsigned reqsize = sizeof(ih) + sizeof(arg);
956 int err;
957
958 list_del_init(&req->intr_entry);
959 req->intr_unique = fuse_get_unique(fc);
960 memset(&ih, 0, sizeof(ih));
961 memset(&arg, 0, sizeof(arg));
962 ih.len = reqsize;
963 ih.opcode = FUSE_INTERRUPT;
964 ih.unique = req->intr_unique;
965 arg.unique = req->in.h.unique;
966
967 spin_unlock(&fc->lock);
968 if (nbytes < reqsize)
969 return -EINVAL;
970
971 err = fuse_copy_one(cs, &ih, sizeof(ih));
972 if (!err)
973 err = fuse_copy_one(cs, &arg, sizeof(arg));
974 fuse_copy_finish(cs);
975
976 return err ? err : reqsize;
977}
978
979static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
980 unsigned max,
981 unsigned *countp)
982{
983 struct fuse_forget_link *head = fc->forget_list_head.next;
984 struct fuse_forget_link **newhead = &head;
985 unsigned count;
986
987 for (count = 0; *newhead != NULL && count < max; count++)
988 newhead = &(*newhead)->next;
989
990 fc->forget_list_head.next = *newhead;
991 *newhead = NULL;
992 if (fc->forget_list_head.next == NULL)
993 fc->forget_list_tail = &fc->forget_list_head;
994
995 if (countp != NULL)
996 *countp = count;
997
998 return head;
999}
1000
1001static int fuse_read_single_forget(struct fuse_conn *fc,
1002 struct fuse_copy_state *cs,
1003 size_t nbytes)
1004__releases(fc->lock)
1005{
1006 int err;
1007 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1008 struct fuse_forget_in arg = {
1009 .nlookup = forget->forget_one.nlookup,
1010 };
1011 struct fuse_in_header ih = {
1012 .opcode = FUSE_FORGET,
1013 .nodeid = forget->forget_one.nodeid,
1014 .unique = fuse_get_unique(fc),
1015 .len = sizeof(ih) + sizeof(arg),
1016 };
1017
1018 spin_unlock(&fc->lock);
1019 kfree(forget);
1020 if (nbytes < ih.len)
1021 return -EINVAL;
1022
1023 err = fuse_copy_one(cs, &ih, sizeof(ih));
1024 if (!err)
1025 err = fuse_copy_one(cs, &arg, sizeof(arg));
1026 fuse_copy_finish(cs);
1027
1028 if (err)
1029 return err;
1030
1031 return ih.len;
1032}
1033
1034static int fuse_read_batch_forget(struct fuse_conn *fc,
1035 struct fuse_copy_state *cs, size_t nbytes)
1036__releases(fc->lock)
1037{
1038 int err;
1039 unsigned max_forgets;
1040 unsigned count;
1041 struct fuse_forget_link *head;
1042 struct fuse_batch_forget_in arg = { .count = 0 };
1043 struct fuse_in_header ih = {
1044 .opcode = FUSE_BATCH_FORGET,
1045 .unique = fuse_get_unique(fc),
1046 .len = sizeof(ih) + sizeof(arg),
1047 };
1048
1049 if (nbytes < ih.len) {
1050 spin_unlock(&fc->lock);
1051 return -EINVAL;
1052 }
1053
1054 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1055 head = dequeue_forget(fc, max_forgets, &count);
1056 spin_unlock(&fc->lock);
1057
1058 arg.count = count;
1059 ih.len += count * sizeof(struct fuse_forget_one);
1060 err = fuse_copy_one(cs, &ih, sizeof(ih));
1061 if (!err)
1062 err = fuse_copy_one(cs, &arg, sizeof(arg));
1063
1064 while (head) {
1065 struct fuse_forget_link *forget = head;
1066
1067 if (!err) {
1068 err = fuse_copy_one(cs, &forget->forget_one,
1069 sizeof(forget->forget_one));
1070 }
1071 head = forget->next;
1072 kfree(forget);
1073 }
1074
1075 fuse_copy_finish(cs);
1076
1077 if (err)
1078 return err;
1079
1080 return ih.len;
1081}
1082
1083static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1084 size_t nbytes)
1085__releases(fc->lock)
1086{
1087 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1088 return fuse_read_single_forget(fc, cs, nbytes);
1089 else
1090 return fuse_read_batch_forget(fc, cs, nbytes);
1091}
1092
1093/*
1094 * Read a single request into the userspace filesystem's buffer. This
1095 * function waits until a request is available, then removes it from
1096 * the pending list and copies request data to userspace buffer. If
1097 * no reply is needed (FORGET) or request has been aborted or there
1098 * was an error during the copying then it's finished by calling
1099 * request_end(). Otherwise add it to the processing list, and set
1100 * the 'sent' flag.
1101 */
1102static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1103 struct fuse_copy_state *cs, size_t nbytes)
1104{
1105 int err;
1106 struct fuse_req *req;
1107 struct fuse_in *in;
1108 unsigned reqsize;
1109
1110 restart:
1111 spin_lock(&fc->lock);
1112 err = -EAGAIN;
1113 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1114 !request_pending(fc))
1115 goto err_unlock;
1116
1117 request_wait(fc);
1118 err = -ENODEV;
1119 if (!fc->connected)
1120 goto err_unlock;
1121 err = -ERESTARTSYS;
1122 if (!request_pending(fc))
1123 goto err_unlock;
1124
1125 if (!list_empty(&fc->interrupts)) {
1126 req = list_entry(fc->interrupts.next, struct fuse_req,
1127 intr_entry);
1128 return fuse_read_interrupt(fc, cs, nbytes, req);
1129 }
1130
1131 if (forget_pending(fc)) {
1132 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1133 return fuse_read_forget(fc, cs, nbytes);
1134
1135 if (fc->forget_batch <= -8)
1136 fc->forget_batch = 16;
1137 }
1138
1139 req = list_entry(fc->pending.next, struct fuse_req, list);
1140 req->state = FUSE_REQ_READING;
1141 list_move(&req->list, &fc->io);
1142
1143 in = &req->in;
1144 reqsize = in->h.len;
1145 /* If request is too large, reply with an error and restart the read */
1146 if (nbytes < reqsize) {
1147 req->out.h.error = -EIO;
1148 /* SETXATTR is special, since it may contain too large data */
1149 if (in->h.opcode == FUSE_SETXATTR)
1150 req->out.h.error = -E2BIG;
1151 request_end(fc, req);
1152 goto restart;
1153 }
1154 spin_unlock(&fc->lock);
1155 cs->req = req;
1156 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1157 if (!err)
1158 err = fuse_copy_args(cs, in->numargs, in->argpages,
1159 (struct fuse_arg *) in->args, 0);
1160 fuse_copy_finish(cs);
1161 spin_lock(&fc->lock);
1162 req->locked = 0;
1163 if (req->aborted) {
1164 request_end(fc, req);
1165 return -ENODEV;
1166 }
1167 if (err) {
1168 req->out.h.error = -EIO;
1169 request_end(fc, req);
1170 return err;
1171 }
1172 if (!req->isreply)
1173 request_end(fc, req);
1174 else {
1175 req->state = FUSE_REQ_SENT;
1176 list_move_tail(&req->list, &fc->processing);
1177 if (req->interrupted)
1178 queue_interrupt(fc, req);
1179 spin_unlock(&fc->lock);
1180 }
1181 return reqsize;
1182
1183 err_unlock:
1184 spin_unlock(&fc->lock);
1185 return err;
1186}
1187
1188static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1189 unsigned long nr_segs, loff_t pos)
1190{
1191 struct fuse_copy_state cs;
1192 struct file *file = iocb->ki_filp;
1193 struct fuse_conn *fc = fuse_get_conn(file);
1194 if (!fc)
1195 return -EPERM;
1196
1197 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1198
1199 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1200}
1201
1202static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe,
1203 struct pipe_buffer *buf)
1204{
1205 return 1;
1206}
1207
1208static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = {
1209 .can_merge = 0,
1210 .map = generic_pipe_buf_map,
1211 .unmap = generic_pipe_buf_unmap,
1212 .confirm = generic_pipe_buf_confirm,
1213 .release = generic_pipe_buf_release,
1214 .steal = fuse_dev_pipe_buf_steal,
1215 .get = generic_pipe_buf_get,
1216};
1217
1218static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1219 struct pipe_inode_info *pipe,
1220 size_t len, unsigned int flags)
1221{
1222 int ret;
1223 int page_nr = 0;
1224 int do_wakeup = 0;
1225 struct pipe_buffer *bufs;
1226 struct fuse_copy_state cs;
1227 struct fuse_conn *fc = fuse_get_conn(in);
1228 if (!fc)
1229 return -EPERM;
1230
1231 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1232 if (!bufs)
1233 return -ENOMEM;
1234
1235 fuse_copy_init(&cs, fc, 1, NULL, 0);
1236 cs.pipebufs = bufs;
1237 cs.pipe = pipe;
1238 ret = fuse_dev_do_read(fc, in, &cs, len);
1239 if (ret < 0)
1240 goto out;
1241
1242 ret = 0;
1243 pipe_lock(pipe);
1244
1245 if (!pipe->readers) {
1246 send_sig(SIGPIPE, current, 0);
1247 if (!ret)
1248 ret = -EPIPE;
1249 goto out_unlock;
1250 }
1251
1252 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1253 ret = -EIO;
1254 goto out_unlock;
1255 }
1256
1257 while (page_nr < cs.nr_segs) {
1258 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1259 struct pipe_buffer *buf = pipe->bufs + newbuf;
1260
1261 buf->page = bufs[page_nr].page;
1262 buf->offset = bufs[page_nr].offset;
1263 buf->len = bufs[page_nr].len;
1264 buf->ops = &fuse_dev_pipe_buf_ops;
1265
1266 pipe->nrbufs++;
1267 page_nr++;
1268 ret += buf->len;
1269
1270 if (pipe->inode)
1271 do_wakeup = 1;
1272 }
1273
1274out_unlock:
1275 pipe_unlock(pipe);
1276
1277 if (do_wakeup) {
1278 smp_mb();
1279 if (waitqueue_active(&pipe->wait))
1280 wake_up_interruptible(&pipe->wait);
1281 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1282 }
1283
1284out:
1285 for (; page_nr < cs.nr_segs; page_nr++)
1286 page_cache_release(bufs[page_nr].page);
1287
1288 kfree(bufs);
1289 return ret;
1290}
1291
1292static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1293 struct fuse_copy_state *cs)
1294{
1295 struct fuse_notify_poll_wakeup_out outarg;
1296 int err = -EINVAL;
1297
1298 if (size != sizeof(outarg))
1299 goto err;
1300
1301 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1302 if (err)
1303 goto err;
1304
1305 fuse_copy_finish(cs);
1306 return fuse_notify_poll_wakeup(fc, &outarg);
1307
1308err:
1309 fuse_copy_finish(cs);
1310 return err;
1311}
1312
1313static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1314 struct fuse_copy_state *cs)
1315{
1316 struct fuse_notify_inval_inode_out outarg;
1317 int err = -EINVAL;
1318
1319 if (size != sizeof(outarg))
1320 goto err;
1321
1322 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1323 if (err)
1324 goto err;
1325 fuse_copy_finish(cs);
1326
1327 down_read(&fc->killsb);
1328 err = -ENOENT;
1329 if (fc->sb) {
1330 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1331 outarg.off, outarg.len);
1332 }
1333 up_read(&fc->killsb);
1334 return err;
1335
1336err:
1337 fuse_copy_finish(cs);
1338 return err;
1339}
1340
1341static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1342 struct fuse_copy_state *cs)
1343{
1344 struct fuse_notify_inval_entry_out outarg;
1345 int err = -ENOMEM;
1346 char *buf;
1347 struct qstr name;
1348
1349 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1350 if (!buf)
1351 goto err;
1352
1353 err = -EINVAL;
1354 if (size < sizeof(outarg))
1355 goto err;
1356
1357 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1358 if (err)
1359 goto err;
1360
1361 err = -ENAMETOOLONG;
1362 if (outarg.namelen > FUSE_NAME_MAX)
1363 goto err;
1364
1365 err = -EINVAL;
1366 if (size != sizeof(outarg) + outarg.namelen + 1)
1367 goto err;
1368
1369 name.name = buf;
1370 name.len = outarg.namelen;
1371 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1372 if (err)
1373 goto err;
1374 fuse_copy_finish(cs);
1375 buf[outarg.namelen] = 0;
1376 name.hash = full_name_hash(name.name, name.len);
1377
1378 down_read(&fc->killsb);
1379 err = -ENOENT;
1380 if (fc->sb)
1381 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1382 up_read(&fc->killsb);
1383 kfree(buf);
1384 return err;
1385
1386err:
1387 kfree(buf);
1388 fuse_copy_finish(cs);
1389 return err;
1390}
1391
1392static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1393 struct fuse_copy_state *cs)
1394{
1395 struct fuse_notify_delete_out outarg;
1396 int err = -ENOMEM;
1397 char *buf;
1398 struct qstr name;
1399
1400 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1401 if (!buf)
1402 goto err;
1403
1404 err = -EINVAL;
1405 if (size < sizeof(outarg))
1406 goto err;
1407
1408 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1409 if (err)
1410 goto err;
1411
1412 err = -ENAMETOOLONG;
1413 if (outarg.namelen > FUSE_NAME_MAX)
1414 goto err;
1415
1416 err = -EINVAL;
1417 if (size != sizeof(outarg) + outarg.namelen + 1)
1418 goto err;
1419
1420 name.name = buf;
1421 name.len = outarg.namelen;
1422 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1423 if (err)
1424 goto err;
1425 fuse_copy_finish(cs);
1426 buf[outarg.namelen] = 0;
1427 name.hash = full_name_hash(name.name, name.len);
1428
1429 down_read(&fc->killsb);
1430 err = -ENOENT;
1431 if (fc->sb)
1432 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1433 outarg.child, &name);
1434 up_read(&fc->killsb);
1435 kfree(buf);
1436 return err;
1437
1438err:
1439 kfree(buf);
1440 fuse_copy_finish(cs);
1441 return err;
1442}
1443
1444static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1445 struct fuse_copy_state *cs)
1446{
1447 struct fuse_notify_store_out outarg;
1448 struct inode *inode;
1449 struct address_space *mapping;
1450 u64 nodeid;
1451 int err;
1452 pgoff_t index;
1453 unsigned int offset;
1454 unsigned int num;
1455 loff_t file_size;
1456 loff_t end;
1457
1458 err = -EINVAL;
1459 if (size < sizeof(outarg))
1460 goto out_finish;
1461
1462 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1463 if (err)
1464 goto out_finish;
1465
1466 err = -EINVAL;
1467 if (size - sizeof(outarg) != outarg.size)
1468 goto out_finish;
1469
1470 nodeid = outarg.nodeid;
1471
1472 down_read(&fc->killsb);
1473
1474 err = -ENOENT;
1475 if (!fc->sb)
1476 goto out_up_killsb;
1477
1478 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1479 if (!inode)
1480 goto out_up_killsb;
1481
1482 mapping = inode->i_mapping;
1483 index = outarg.offset >> PAGE_CACHE_SHIFT;
1484 offset = outarg.offset & ~PAGE_CACHE_MASK;
1485 file_size = i_size_read(inode);
1486 end = outarg.offset + outarg.size;
1487 if (end > file_size) {
1488 file_size = end;
1489 fuse_write_update_size(inode, file_size);
1490 }
1491
1492 num = outarg.size;
1493 while (num) {
1494 struct page *page;
1495 unsigned int this_num;
1496
1497 err = -ENOMEM;
1498 page = find_or_create_page(mapping, index,
1499 mapping_gfp_mask(mapping));
1500 if (!page)
1501 goto out_iput;
1502
1503 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1504 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1505 if (!err && offset == 0 && (num != 0 || file_size == end))
1506 SetPageUptodate(page);
1507 unlock_page(page);
1508 page_cache_release(page);
1509
1510 if (err)
1511 goto out_iput;
1512
1513 num -= this_num;
1514 offset = 0;
1515 index++;
1516 }
1517
1518 err = 0;
1519
1520out_iput:
1521 iput(inode);
1522out_up_killsb:
1523 up_read(&fc->killsb);
1524out_finish:
1525 fuse_copy_finish(cs);
1526 return err;
1527}
1528
1529static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1530{
1531 release_pages(req->pages, req->num_pages, 0);
1532}
1533
1534static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1535 struct fuse_notify_retrieve_out *outarg)
1536{
1537 int err;
1538 struct address_space *mapping = inode->i_mapping;
1539 struct fuse_req *req;
1540 pgoff_t index;
1541 loff_t file_size;
1542 unsigned int num;
1543 unsigned int offset;
1544 size_t total_len = 0;
1545
1546 req = fuse_get_req(fc);
1547 if (IS_ERR(req))
1548 return PTR_ERR(req);
1549
1550 offset = outarg->offset & ~PAGE_CACHE_MASK;
1551
1552 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1553 req->in.h.nodeid = outarg->nodeid;
1554 req->in.numargs = 2;
1555 req->in.argpages = 1;
1556 req->page_offset = offset;
1557 req->end = fuse_retrieve_end;
1558
1559 index = outarg->offset >> PAGE_CACHE_SHIFT;
1560 file_size = i_size_read(inode);
1561 num = outarg->size;
1562 if (outarg->offset > file_size)
1563 num = 0;
1564 else if (outarg->offset + num > file_size)
1565 num = file_size - outarg->offset;
1566
1567 while (num && req->num_pages < FUSE_MAX_PAGES_PER_REQ) {
1568 struct page *page;
1569 unsigned int this_num;
1570
1571 page = find_get_page(mapping, index);
1572 if (!page)
1573 break;
1574
1575 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1576 req->pages[req->num_pages] = page;
1577 req->num_pages++;
1578
1579 offset = 0;
1580 num -= this_num;
1581 total_len += this_num;
1582 index++;
1583 }
1584 req->misc.retrieve_in.offset = outarg->offset;
1585 req->misc.retrieve_in.size = total_len;
1586 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1587 req->in.args[0].value = &req->misc.retrieve_in;
1588 req->in.args[1].size = total_len;
1589
1590 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1591 if (err)
1592 fuse_retrieve_end(fc, req);
1593
1594 return err;
1595}
1596
1597static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1598 struct fuse_copy_state *cs)
1599{
1600 struct fuse_notify_retrieve_out outarg;
1601 struct inode *inode;
1602 int err;
1603
1604 err = -EINVAL;
1605 if (size != sizeof(outarg))
1606 goto copy_finish;
1607
1608 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1609 if (err)
1610 goto copy_finish;
1611
1612 fuse_copy_finish(cs);
1613
1614 down_read(&fc->killsb);
1615 err = -ENOENT;
1616 if (fc->sb) {
1617 u64 nodeid = outarg.nodeid;
1618
1619 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1620 if (inode) {
1621 err = fuse_retrieve(fc, inode, &outarg);
1622 iput(inode);
1623 }
1624 }
1625 up_read(&fc->killsb);
1626
1627 return err;
1628
1629copy_finish:
1630 fuse_copy_finish(cs);
1631 return err;
1632}
1633
1634static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1635 unsigned int size, struct fuse_copy_state *cs)
1636{
1637 switch (code) {
1638 case FUSE_NOTIFY_POLL:
1639 return fuse_notify_poll(fc, size, cs);
1640
1641 case FUSE_NOTIFY_INVAL_INODE:
1642 return fuse_notify_inval_inode(fc, size, cs);
1643
1644 case FUSE_NOTIFY_INVAL_ENTRY:
1645 return fuse_notify_inval_entry(fc, size, cs);
1646
1647 case FUSE_NOTIFY_STORE:
1648 return fuse_notify_store(fc, size, cs);
1649
1650 case FUSE_NOTIFY_RETRIEVE:
1651 return fuse_notify_retrieve(fc, size, cs);
1652
1653 case FUSE_NOTIFY_DELETE:
1654 return fuse_notify_delete(fc, size, cs);
1655
1656 default:
1657 fuse_copy_finish(cs);
1658 return -EINVAL;
1659 }
1660}
1661
1662/* Look up request on processing list by unique ID */
1663static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1664{
1665 struct list_head *entry;
1666
1667 list_for_each(entry, &fc->processing) {
1668 struct fuse_req *req;
1669 req = list_entry(entry, struct fuse_req, list);
1670 if (req->in.h.unique == unique || req->intr_unique == unique)
1671 return req;
1672 }
1673 return NULL;
1674}
1675
1676static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1677 unsigned nbytes)
1678{
1679 unsigned reqsize = sizeof(struct fuse_out_header);
1680
1681 if (out->h.error)
1682 return nbytes != reqsize ? -EINVAL : 0;
1683
1684 reqsize += len_args(out->numargs, out->args);
1685
1686 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1687 return -EINVAL;
1688 else if (reqsize > nbytes) {
1689 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1690 unsigned diffsize = reqsize - nbytes;
1691 if (diffsize > lastarg->size)
1692 return -EINVAL;
1693 lastarg->size -= diffsize;
1694 }
1695 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1696 out->page_zeroing);
1697}
1698
1699/*
1700 * Write a single reply to a request. First the header is copied from
1701 * the write buffer. The request is then searched on the processing
1702 * list by the unique ID found in the header. If found, then remove
1703 * it from the list and copy the rest of the buffer to the request.
1704 * The request is finished by calling request_end()
1705 */
1706static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1707 struct fuse_copy_state *cs, size_t nbytes)
1708{
1709 int err;
1710 struct fuse_req *req;
1711 struct fuse_out_header oh;
1712
1713 if (nbytes < sizeof(struct fuse_out_header))
1714 return -EINVAL;
1715
1716 err = fuse_copy_one(cs, &oh, sizeof(oh));
1717 if (err)
1718 goto err_finish;
1719
1720 err = -EINVAL;
1721 if (oh.len != nbytes)
1722 goto err_finish;
1723
1724 /*
1725 * Zero oh.unique indicates unsolicited notification message
1726 * and error contains notification code.
1727 */
1728 if (!oh.unique) {
1729 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1730 return err ? err : nbytes;
1731 }
1732
1733 err = -EINVAL;
1734 if (oh.error <= -1000 || oh.error > 0)
1735 goto err_finish;
1736
1737 spin_lock(&fc->lock);
1738 err = -ENOENT;
1739 if (!fc->connected)
1740 goto err_unlock;
1741
1742 req = request_find(fc, oh.unique);
1743 if (!req)
1744 goto err_unlock;
1745
1746 if (req->aborted) {
1747 spin_unlock(&fc->lock);
1748 fuse_copy_finish(cs);
1749 spin_lock(&fc->lock);
1750 request_end(fc, req);
1751 return -ENOENT;
1752 }
1753 /* Is it an interrupt reply? */
1754 if (req->intr_unique == oh.unique) {
1755 err = -EINVAL;
1756 if (nbytes != sizeof(struct fuse_out_header))
1757 goto err_unlock;
1758
1759 if (oh.error == -ENOSYS)
1760 fc->no_interrupt = 1;
1761 else if (oh.error == -EAGAIN)
1762 queue_interrupt(fc, req);
1763
1764 spin_unlock(&fc->lock);
1765 fuse_copy_finish(cs);
1766 return nbytes;
1767 }
1768
1769 req->state = FUSE_REQ_WRITING;
1770 list_move(&req->list, &fc->io);
1771 req->out.h = oh;
1772 req->locked = 1;
1773 cs->req = req;
1774 if (!req->out.page_replace)
1775 cs->move_pages = 0;
1776 spin_unlock(&fc->lock);
1777
1778 err = copy_out_args(cs, &req->out, nbytes);
1779 fuse_copy_finish(cs);
1780
1781 spin_lock(&fc->lock);
1782 req->locked = 0;
1783 if (!err) {
1784 if (req->aborted)
1785 err = -ENOENT;
1786 } else if (!req->aborted)
1787 req->out.h.error = -EIO;
1788 request_end(fc, req);
1789
1790 return err ? err : nbytes;
1791
1792 err_unlock:
1793 spin_unlock(&fc->lock);
1794 err_finish:
1795 fuse_copy_finish(cs);
1796 return err;
1797}
1798
1799static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1800 unsigned long nr_segs, loff_t pos)
1801{
1802 struct fuse_copy_state cs;
1803 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1804 if (!fc)
1805 return -EPERM;
1806
1807 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1808
1809 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1810}
1811
1812static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1813 struct file *out, loff_t *ppos,
1814 size_t len, unsigned int flags)
1815{
1816 unsigned nbuf;
1817 unsigned idx;
1818 struct pipe_buffer *bufs;
1819 struct fuse_copy_state cs;
1820 struct fuse_conn *fc;
1821 size_t rem;
1822 ssize_t ret;
1823
1824 fc = fuse_get_conn(out);
1825 if (!fc)
1826 return -EPERM;
1827
1828 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1829 if (!bufs)
1830 return -ENOMEM;
1831
1832 pipe_lock(pipe);
1833 nbuf = 0;
1834 rem = 0;
1835 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1836 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1837
1838 ret = -EINVAL;
1839 if (rem < len) {
1840 pipe_unlock(pipe);
1841 goto out;
1842 }
1843
1844 rem = len;
1845 while (rem) {
1846 struct pipe_buffer *ibuf;
1847 struct pipe_buffer *obuf;
1848
1849 BUG_ON(nbuf >= pipe->buffers);
1850 BUG_ON(!pipe->nrbufs);
1851 ibuf = &pipe->bufs[pipe->curbuf];
1852 obuf = &bufs[nbuf];
1853
1854 if (rem >= ibuf->len) {
1855 *obuf = *ibuf;
1856 ibuf->ops = NULL;
1857 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1858 pipe->nrbufs--;
1859 } else {
1860 ibuf->ops->get(pipe, ibuf);
1861 *obuf = *ibuf;
1862 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1863 obuf->len = rem;
1864 ibuf->offset += obuf->len;
1865 ibuf->len -= obuf->len;
1866 }
1867 nbuf++;
1868 rem -= obuf->len;
1869 }
1870 pipe_unlock(pipe);
1871
1872 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1873 cs.pipebufs = bufs;
1874 cs.pipe = pipe;
1875
1876 if (flags & SPLICE_F_MOVE)
1877 cs.move_pages = 1;
1878
1879 ret = fuse_dev_do_write(fc, &cs, len);
1880
1881 for (idx = 0; idx < nbuf; idx++) {
1882 struct pipe_buffer *buf = &bufs[idx];
1883 buf->ops->release(pipe, buf);
1884 }
1885out:
1886 kfree(bufs);
1887 return ret;
1888}
1889
1890static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1891{
1892 unsigned mask = POLLOUT | POLLWRNORM;
1893 struct fuse_conn *fc = fuse_get_conn(file);
1894 if (!fc)
1895 return POLLERR;
1896
1897 poll_wait(file, &fc->waitq, wait);
1898
1899 spin_lock(&fc->lock);
1900 if (!fc->connected)
1901 mask = POLLERR;
1902 else if (request_pending(fc))
1903 mask |= POLLIN | POLLRDNORM;
1904 spin_unlock(&fc->lock);
1905
1906 return mask;
1907}
1908
1909/*
1910 * Abort all requests on the given list (pending or processing)
1911 *
1912 * This function releases and reacquires fc->lock
1913 */
1914static void end_requests(struct fuse_conn *fc, struct list_head *head)
1915__releases(fc->lock)
1916__acquires(fc->lock)
1917{
1918 while (!list_empty(head)) {
1919 struct fuse_req *req;
1920 req = list_entry(head->next, struct fuse_req, list);
1921 req->out.h.error = -ECONNABORTED;
1922 request_end(fc, req);
1923 spin_lock(&fc->lock);
1924 }
1925}
1926
1927/*
1928 * Abort requests under I/O
1929 *
1930 * The requests are set to aborted and finished, and the request
1931 * waiter is woken up. This will make request_wait_answer() wait
1932 * until the request is unlocked and then return.
1933 *
1934 * If the request is asynchronous, then the end function needs to be
1935 * called after waiting for the request to be unlocked (if it was
1936 * locked).
1937 */
1938static void end_io_requests(struct fuse_conn *fc)
1939__releases(fc->lock)
1940__acquires(fc->lock)
1941{
1942 while (!list_empty(&fc->io)) {
1943 struct fuse_req *req =
1944 list_entry(fc->io.next, struct fuse_req, list);
1945 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
1946
1947 req->aborted = 1;
1948 req->out.h.error = -ECONNABORTED;
1949 req->state = FUSE_REQ_FINISHED;
1950 list_del_init(&req->list);
1951 wake_up(&req->waitq);
1952 if (end) {
1953 req->end = NULL;
1954 __fuse_get_request(req);
1955 spin_unlock(&fc->lock);
1956 wait_event(req->waitq, !req->locked);
1957 end(fc, req);
1958 fuse_put_request(fc, req);
1959 spin_lock(&fc->lock);
1960 }
1961 }
1962}
1963
1964static void end_queued_requests(struct fuse_conn *fc)
1965__releases(fc->lock)
1966__acquires(fc->lock)
1967{
1968 fc->max_background = UINT_MAX;
1969 flush_bg_queue(fc);
1970 end_requests(fc, &fc->pending);
1971 end_requests(fc, &fc->processing);
1972 while (forget_pending(fc))
1973 kfree(dequeue_forget(fc, 1, NULL));
1974}
1975
1976static void end_polls(struct fuse_conn *fc)
1977{
1978 struct rb_node *p;
1979
1980 p = rb_first(&fc->polled_files);
1981
1982 while (p) {
1983 struct fuse_file *ff;
1984 ff = rb_entry(p, struct fuse_file, polled_node);
1985 wake_up_interruptible_all(&ff->poll_wait);
1986
1987 p = rb_next(p);
1988 }
1989}
1990
1991/*
1992 * Abort all requests.
1993 *
1994 * Emergency exit in case of a malicious or accidental deadlock, or
1995 * just a hung filesystem.
1996 *
1997 * The same effect is usually achievable through killing the
1998 * filesystem daemon and all users of the filesystem. The exception
1999 * is the combination of an asynchronous request and the tricky
2000 * deadlock (see Documentation/filesystems/fuse.txt).
2001 *
2002 * During the aborting, progression of requests from the pending and
2003 * processing lists onto the io list, and progression of new requests
2004 * onto the pending list is prevented by req->connected being false.
2005 *
2006 * Progression of requests under I/O to the processing list is
2007 * prevented by the req->aborted flag being true for these requests.
2008 * For this reason requests on the io list must be aborted first.
2009 */
2010void fuse_abort_conn(struct fuse_conn *fc)
2011{
2012 spin_lock(&fc->lock);
2013 if (fc->connected) {
2014 fc->connected = 0;
2015 fc->blocked = 0;
2016 end_io_requests(fc);
2017 end_queued_requests(fc);
2018 end_polls(fc);
2019 wake_up_all(&fc->waitq);
2020 wake_up_all(&fc->blocked_waitq);
2021 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2022 }
2023 spin_unlock(&fc->lock);
2024}
2025EXPORT_SYMBOL_GPL(fuse_abort_conn);
2026
2027int fuse_dev_release(struct inode *inode, struct file *file)
2028{
2029 struct fuse_conn *fc = fuse_get_conn(file);
2030 if (fc) {
2031 spin_lock(&fc->lock);
2032 fc->connected = 0;
2033 fc->blocked = 0;
2034 end_queued_requests(fc);
2035 end_polls(fc);
2036 wake_up_all(&fc->blocked_waitq);
2037 spin_unlock(&fc->lock);
2038 fuse_conn_put(fc);
2039 }
2040
2041 return 0;
2042}
2043EXPORT_SYMBOL_GPL(fuse_dev_release);
2044
2045static int fuse_dev_fasync(int fd, struct file *file, int on)
2046{
2047 struct fuse_conn *fc = fuse_get_conn(file);
2048 if (!fc)
2049 return -EPERM;
2050
2051 /* No locking - fasync_helper does its own locking */
2052 return fasync_helper(fd, file, on, &fc->fasync);
2053}
2054
2055const struct file_operations fuse_dev_operations = {
2056 .owner = THIS_MODULE,
2057 .llseek = no_llseek,
2058 .read = do_sync_read,
2059 .aio_read = fuse_dev_read,
2060 .splice_read = fuse_dev_splice_read,
2061 .write = do_sync_write,
2062 .aio_write = fuse_dev_write,
2063 .splice_write = fuse_dev_splice_write,
2064 .poll = fuse_dev_poll,
2065 .release = fuse_dev_release,
2066 .fasync = fuse_dev_fasync,
2067};
2068EXPORT_SYMBOL_GPL(fuse_dev_operations);
2069
2070static struct miscdevice fuse_miscdevice = {
2071 .minor = FUSE_MINOR,
2072 .name = "fuse",
2073 .fops = &fuse_dev_operations,
2074};
2075
2076int __init fuse_dev_init(void)
2077{
2078 int err = -ENOMEM;
2079 fuse_req_cachep = kmem_cache_create("fuse_request",
2080 sizeof(struct fuse_req),
2081 0, 0, NULL);
2082 if (!fuse_req_cachep)
2083 goto out;
2084
2085 err = misc_register(&fuse_miscdevice);
2086 if (err)
2087 goto out_cache_clean;
2088
2089 return 0;
2090
2091 out_cache_clean:
2092 kmem_cache_destroy(fuse_req_cachep);
2093 out:
2094 return err;
2095}
2096
2097void fuse_dev_cleanup(void)
2098{
2099 misc_deregister(&fuse_miscdevice);
2100 kmem_cache_destroy(fuse_req_cachep);
2101}
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}