2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
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>
23 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24 MODULE_ALIAS("devname:fuse");
26 static struct kmem_cache *fuse_req_cachep;
28 static struct fuse_conn *fuse_get_conn(struct file *file)
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
34 return file->private_data;
37 static void fuse_request_init(struct fuse_req *req, struct page **pages,
38 struct fuse_page_desc *page_descs,
41 memset(req, 0, sizeof(*req));
42 memset(pages, 0, sizeof(*pages) * npages);
43 memset(page_descs, 0, sizeof(*page_descs) * npages);
44 INIT_LIST_HEAD(&req->list);
45 INIT_LIST_HEAD(&req->intr_entry);
46 init_waitqueue_head(&req->waitq);
47 atomic_set(&req->count, 1);
49 req->page_descs = page_descs;
50 req->max_pages = npages;
51 __set_bit(FR_PENDING, &req->flags);
54 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
56 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
59 struct fuse_page_desc *page_descs;
61 if (npages <= FUSE_REQ_INLINE_PAGES) {
62 pages = req->inline_pages;
63 page_descs = req->inline_page_descs;
65 pages = kmalloc(sizeof(struct page *) * npages, flags);
66 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
70 if (!pages || !page_descs) {
73 kmem_cache_free(fuse_req_cachep, req);
77 fuse_request_init(req, pages, page_descs, npages);
82 struct fuse_req *fuse_request_alloc(unsigned npages)
84 return __fuse_request_alloc(npages, GFP_KERNEL);
86 EXPORT_SYMBOL_GPL(fuse_request_alloc);
88 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
90 return __fuse_request_alloc(npages, GFP_NOFS);
93 void fuse_request_free(struct fuse_req *req)
95 if (req->pages != req->inline_pages) {
97 kfree(req->page_descs);
99 kmem_cache_free(fuse_req_cachep, req);
102 static void block_sigs(sigset_t *oldset)
106 siginitsetinv(&mask, sigmask(SIGKILL));
107 sigprocmask(SIG_BLOCK, &mask, oldset);
110 static void restore_sigs(sigset_t *oldset)
112 sigprocmask(SIG_SETMASK, oldset, NULL);
115 void __fuse_get_request(struct fuse_req *req)
117 atomic_inc(&req->count);
120 /* Must be called with > 1 refcount */
121 static void __fuse_put_request(struct fuse_req *req)
123 BUG_ON(atomic_read(&req->count) < 2);
124 atomic_dec(&req->count);
127 static void fuse_req_init_context(struct fuse_req *req)
129 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
130 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
131 req->in.h.pid = current->pid;
134 void fuse_set_initialized(struct fuse_conn *fc)
136 /* Make sure stores before this are seen on another CPU */
141 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
143 return !fc->initialized || (for_background && fc->blocked);
146 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
149 struct fuse_req *req;
151 atomic_inc(&fc->num_waiting);
153 if (fuse_block_alloc(fc, for_background)) {
158 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
159 !fuse_block_alloc(fc, for_background));
160 restore_sigs(&oldset);
165 /* Matches smp_wmb() in fuse_set_initialized() */
176 req = fuse_request_alloc(npages);
180 wake_up(&fc->blocked_waitq);
184 fuse_req_init_context(req);
185 __set_bit(FR_WAITING, &req->flags);
187 __set_bit(FR_BACKGROUND, &req->flags);
192 atomic_dec(&fc->num_waiting);
196 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
198 return __fuse_get_req(fc, npages, false);
200 EXPORT_SYMBOL_GPL(fuse_get_req);
202 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
205 return __fuse_get_req(fc, npages, true);
207 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
210 * Return request in fuse_file->reserved_req. However that may
211 * currently be in use. If that is the case, wait for it to become
214 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
217 struct fuse_req *req = NULL;
218 struct fuse_file *ff = file->private_data;
221 wait_event(fc->reserved_req_waitq, ff->reserved_req);
222 spin_lock(&fc->lock);
223 if (ff->reserved_req) {
224 req = ff->reserved_req;
225 ff->reserved_req = NULL;
226 req->stolen_file = get_file(file);
228 spin_unlock(&fc->lock);
235 * Put stolen request back into fuse_file->reserved_req
237 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
239 struct file *file = req->stolen_file;
240 struct fuse_file *ff = file->private_data;
242 spin_lock(&fc->lock);
243 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
244 BUG_ON(ff->reserved_req);
245 ff->reserved_req = req;
246 wake_up_all(&fc->reserved_req_waitq);
247 spin_unlock(&fc->lock);
252 * Gets a requests for a file operation, always succeeds
254 * This is used for sending the FLUSH request, which must get to
255 * userspace, due to POSIX locks which may need to be unlocked.
257 * If allocation fails due to OOM, use the reserved request in
260 * This is very unlikely to deadlock accidentally, since the
261 * filesystem should not have it's own file open. If deadlock is
262 * intentional, it can still be broken by "aborting" the filesystem.
264 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
267 struct fuse_req *req;
269 atomic_inc(&fc->num_waiting);
270 wait_event(fc->blocked_waitq, fc->initialized);
271 /* Matches smp_wmb() in fuse_set_initialized() */
273 req = fuse_request_alloc(0);
275 req = get_reserved_req(fc, file);
277 fuse_req_init_context(req);
278 __set_bit(FR_WAITING, &req->flags);
279 __clear_bit(FR_BACKGROUND, &req->flags);
283 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
285 if (atomic_dec_and_test(&req->count)) {
286 if (test_bit(FR_BACKGROUND, &req->flags)) {
288 * We get here in the unlikely case that a background
289 * request was allocated but not sent
291 spin_lock(&fc->lock);
293 wake_up(&fc->blocked_waitq);
294 spin_unlock(&fc->lock);
297 if (test_bit(FR_WAITING, &req->flags)) {
298 __clear_bit(FR_WAITING, &req->flags);
299 atomic_dec(&fc->num_waiting);
302 if (req->stolen_file)
303 put_reserved_req(fc, req);
305 fuse_request_free(req);
308 EXPORT_SYMBOL_GPL(fuse_put_request);
310 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
315 for (i = 0; i < numargs; i++)
316 nbytes += args[i].size;
321 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
323 return ++fiq->reqctr;
326 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
328 req->in.h.len = sizeof(struct fuse_in_header) +
329 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
330 list_add_tail(&req->list, &fiq->pending);
331 wake_up_locked(&fiq->waitq);
332 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
335 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
336 u64 nodeid, u64 nlookup)
338 struct fuse_iqueue *fiq = &fc->iq;
340 forget->forget_one.nodeid = nodeid;
341 forget->forget_one.nlookup = nlookup;
343 spin_lock(&fc->lock);
344 spin_lock(&fiq->waitq.lock);
345 if (fiq->connected) {
346 fiq->forget_list_tail->next = forget;
347 fiq->forget_list_tail = forget;
348 wake_up_locked(&fiq->waitq);
349 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
353 spin_unlock(&fiq->waitq.lock);
354 spin_unlock(&fc->lock);
357 static void flush_bg_queue(struct fuse_conn *fc)
359 while (fc->active_background < fc->max_background &&
360 !list_empty(&fc->bg_queue)) {
361 struct fuse_req *req;
362 struct fuse_iqueue *fiq = &fc->iq;
364 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
365 list_del(&req->list);
366 fc->active_background++;
367 spin_lock(&fiq->waitq.lock);
368 req->in.h.unique = fuse_get_unique(fiq);
369 queue_request(fiq, req);
370 spin_unlock(&fiq->waitq.lock);
375 * This function is called when a request is finished. Either a reply
376 * has arrived or it was aborted (and not yet sent) or some error
377 * occurred during communication with userspace, or the device file
378 * was closed. The requester thread is woken up (if still waiting),
379 * the 'end' callback is called if given, else the reference to the
380 * request is released
382 * Called with fc->lock, unlocks it
384 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
387 struct fuse_iqueue *fiq = &fc->iq;
388 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
390 list_del_init(&req->list);
391 spin_lock(&fiq->waitq.lock);
392 list_del_init(&req->intr_entry);
393 spin_unlock(&fiq->waitq.lock);
394 WARN_ON(test_bit(FR_PENDING, &req->flags));
395 WARN_ON(test_bit(FR_SENT, &req->flags));
397 set_bit(FR_FINISHED, &req->flags);
398 if (test_bit(FR_BACKGROUND, &req->flags)) {
399 clear_bit(FR_BACKGROUND, &req->flags);
400 if (fc->num_background == fc->max_background)
403 /* Wake up next waiter, if any */
404 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
405 wake_up(&fc->blocked_waitq);
407 if (fc->num_background == fc->congestion_threshold &&
408 fc->connected && fc->bdi_initialized) {
409 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
410 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
412 fc->num_background--;
413 fc->active_background--;
416 spin_unlock(&fc->lock);
417 wake_up(&req->waitq);
420 fuse_put_request(fc, req);
423 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
425 spin_lock(&fiq->waitq.lock);
426 list_add_tail(&req->intr_entry, &fiq->interrupts);
427 wake_up_locked(&fiq->waitq);
428 spin_unlock(&fiq->waitq.lock);
429 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
432 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
434 struct fuse_iqueue *fiq = &fc->iq;
437 if (!fc->no_interrupt) {
438 /* Any signal may interrupt this */
439 err = wait_event_interruptible(req->waitq,
440 test_bit(FR_FINISHED, &req->flags));
444 spin_lock(&fc->lock);
445 set_bit(FR_INTERRUPTED, &req->flags);
446 if (test_bit(FR_SENT, &req->flags))
447 queue_interrupt(fiq, req);
448 spin_unlock(&fc->lock);
451 if (!test_bit(FR_FORCE, &req->flags)) {
454 /* Only fatal signals may interrupt this */
456 err = wait_event_interruptible(req->waitq,
457 test_bit(FR_FINISHED, &req->flags));
458 restore_sigs(&oldset);
463 spin_lock(&fc->lock);
464 spin_lock(&fiq->waitq.lock);
465 /* Request is not yet in userspace, bail out */
466 if (test_bit(FR_PENDING, &req->flags)) {
467 list_del(&req->list);
468 spin_unlock(&fiq->waitq.lock);
469 spin_unlock(&fc->lock);
470 __fuse_put_request(req);
471 req->out.h.error = -EINTR;
474 spin_unlock(&fiq->waitq.lock);
475 spin_unlock(&fc->lock);
479 * Either request is already in userspace, or it was forced.
482 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
485 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
487 struct fuse_iqueue *fiq = &fc->iq;
489 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
490 spin_lock(&fc->lock);
491 spin_lock(&fiq->waitq.lock);
492 if (!fiq->connected) {
493 spin_unlock(&fc->lock);
494 spin_unlock(&fiq->waitq.lock);
495 req->out.h.error = -ENOTCONN;
497 req->in.h.unique = fuse_get_unique(fiq);
498 queue_request(fiq, req);
499 /* acquire extra reference, since request is still needed
500 after request_end() */
501 __fuse_get_request(req);
502 spin_unlock(&fiq->waitq.lock);
503 spin_unlock(&fc->lock);
505 request_wait_answer(fc, req);
506 /* Pairs with smp_wmb() in request_end() */
511 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
513 __set_bit(FR_ISREPLY, &req->flags);
514 if (!test_bit(FR_WAITING, &req->flags)) {
515 __set_bit(FR_WAITING, &req->flags);
516 atomic_inc(&fc->num_waiting);
518 __fuse_request_send(fc, req);
520 EXPORT_SYMBOL_GPL(fuse_request_send);
522 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
524 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
525 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
528 switch (args->in.h.opcode) {
535 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
539 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
543 if (fc->minor < 12) {
544 switch (args->in.h.opcode) {
546 args->in.args[0].size = sizeof(struct fuse_open_in);
549 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
555 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
557 struct fuse_req *req;
560 req = fuse_get_req(fc, 0);
564 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
565 fuse_adjust_compat(fc, args);
567 req->in.h.opcode = args->in.h.opcode;
568 req->in.h.nodeid = args->in.h.nodeid;
569 req->in.numargs = args->in.numargs;
570 memcpy(req->in.args, args->in.args,
571 args->in.numargs * sizeof(struct fuse_in_arg));
572 req->out.argvar = args->out.argvar;
573 req->out.numargs = args->out.numargs;
574 memcpy(req->out.args, args->out.args,
575 args->out.numargs * sizeof(struct fuse_arg));
576 fuse_request_send(fc, req);
577 ret = req->out.h.error;
578 if (!ret && args->out.argvar) {
579 BUG_ON(args->out.numargs != 1);
580 ret = req->out.args[0].size;
582 fuse_put_request(fc, req);
588 * Called under fc->lock
590 * fc->connected must have been checked previously
592 void fuse_request_send_background_locked(struct fuse_conn *fc,
593 struct fuse_req *req)
595 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
596 if (!test_bit(FR_WAITING, &req->flags)) {
597 __set_bit(FR_WAITING, &req->flags);
598 atomic_inc(&fc->num_waiting);
600 __set_bit(FR_ISREPLY, &req->flags);
601 fc->num_background++;
602 if (fc->num_background == fc->max_background)
604 if (fc->num_background == fc->congestion_threshold &&
605 fc->bdi_initialized) {
606 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
607 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
609 list_add_tail(&req->list, &fc->bg_queue);
613 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
616 spin_lock(&fc->lock);
618 fuse_request_send_background_locked(fc, req);
619 spin_unlock(&fc->lock);
621 spin_unlock(&fc->lock);
622 req->out.h.error = -ENOTCONN;
624 fuse_put_request(fc, req);
627 EXPORT_SYMBOL_GPL(fuse_request_send_background);
629 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
630 struct fuse_req *req, u64 unique)
633 struct fuse_iqueue *fiq = &fc->iq;
635 __clear_bit(FR_ISREPLY, &req->flags);
636 req->in.h.unique = unique;
637 spin_lock(&fc->lock);
638 spin_lock(&fiq->waitq.lock);
639 if (fiq->connected) {
640 queue_request(fiq, req);
643 spin_unlock(&fiq->waitq.lock);
644 spin_unlock(&fc->lock);
649 void fuse_force_forget(struct file *file, u64 nodeid)
651 struct inode *inode = file_inode(file);
652 struct fuse_conn *fc = get_fuse_conn(inode);
653 struct fuse_req *req;
654 struct fuse_forget_in inarg;
656 memset(&inarg, 0, sizeof(inarg));
658 req = fuse_get_req_nofail_nopages(fc, file);
659 req->in.h.opcode = FUSE_FORGET;
660 req->in.h.nodeid = nodeid;
662 req->in.args[0].size = sizeof(inarg);
663 req->in.args[0].value = &inarg;
664 __clear_bit(FR_ISREPLY, &req->flags);
665 __fuse_request_send(fc, req);
667 fuse_put_request(fc, req);
671 * Lock the request. Up to the next unlock_request() there mustn't be
672 * anything that could cause a page-fault. If the request was already
675 static int lock_request(struct fuse_req *req)
679 spin_lock(&req->waitq.lock);
680 if (test_bit(FR_ABORTED, &req->flags))
683 set_bit(FR_LOCKED, &req->flags);
684 spin_unlock(&req->waitq.lock);
690 * Unlock request. If it was aborted while locked, caller is responsible
691 * for unlocking and ending the request.
693 static int unlock_request(struct fuse_req *req)
697 spin_lock(&req->waitq.lock);
698 if (test_bit(FR_ABORTED, &req->flags))
701 clear_bit(FR_LOCKED, &req->flags);
702 spin_unlock(&req->waitq.lock);
707 struct fuse_copy_state {
709 struct fuse_req *req;
710 struct iov_iter *iter;
711 struct pipe_buffer *pipebufs;
712 struct pipe_buffer *currbuf;
713 struct pipe_inode_info *pipe;
714 unsigned long nr_segs;
718 unsigned move_pages:1;
721 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
722 struct iov_iter *iter)
724 memset(cs, 0, sizeof(*cs));
729 /* Unmap and put previous page of userspace buffer */
730 static void fuse_copy_finish(struct fuse_copy_state *cs)
733 struct pipe_buffer *buf = cs->currbuf;
736 buf->len = PAGE_SIZE - cs->len;
740 flush_dcache_page(cs->pg);
741 set_page_dirty_lock(cs->pg);
749 * Get another pagefull of userspace buffer, and map it to kernel
750 * address space, and lock request
752 static int fuse_copy_fill(struct fuse_copy_state *cs)
757 err = unlock_request(cs->req);
761 fuse_copy_finish(cs);
763 struct pipe_buffer *buf = cs->pipebufs;
766 err = buf->ops->confirm(cs->pipe, buf);
770 BUG_ON(!cs->nr_segs);
773 cs->offset = buf->offset;
778 if (cs->nr_segs == cs->pipe->buffers)
781 page = alloc_page(GFP_HIGHUSER);
798 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
806 iov_iter_advance(cs->iter, err);
809 return lock_request(cs->req);
812 /* Do as much copy to/from userspace buffer as we can */
813 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
815 unsigned ncpy = min(*size, cs->len);
817 void *pgaddr = kmap_atomic(cs->pg);
818 void *buf = pgaddr + cs->offset;
821 memcpy(buf, *val, ncpy);
823 memcpy(*val, buf, ncpy);
825 kunmap_atomic(pgaddr);
834 static int fuse_check_page(struct page *page)
836 if (page_mapcount(page) ||
837 page->mapping != NULL ||
838 page_count(page) != 1 ||
839 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
846 printk(KERN_WARNING "fuse: trying to steal weird page\n");
847 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);
853 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
856 struct page *oldpage = *pagep;
857 struct page *newpage;
858 struct pipe_buffer *buf = cs->pipebufs;
860 err = unlock_request(cs->req);
864 fuse_copy_finish(cs);
866 err = buf->ops->confirm(cs->pipe, buf);
870 BUG_ON(!cs->nr_segs);
876 if (cs->len != PAGE_SIZE)
879 if (buf->ops->steal(cs->pipe, buf) != 0)
884 if (!PageUptodate(newpage))
885 SetPageUptodate(newpage);
887 ClearPageMappedToDisk(newpage);
889 if (fuse_check_page(newpage) != 0)
890 goto out_fallback_unlock;
893 * This is a new and locked page, it shouldn't be mapped or
894 * have any special flags on it
896 if (WARN_ON(page_mapped(oldpage)))
897 goto out_fallback_unlock;
898 if (WARN_ON(page_has_private(oldpage)))
899 goto out_fallback_unlock;
900 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
901 goto out_fallback_unlock;
902 if (WARN_ON(PageMlocked(oldpage)))
903 goto out_fallback_unlock;
905 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
907 unlock_page(newpage);
911 page_cache_get(newpage);
913 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
914 lru_cache_add_file(newpage);
917 spin_lock(&cs->req->waitq.lock);
918 if (test_bit(FR_ABORTED, &cs->req->flags))
922 spin_unlock(&cs->req->waitq.lock);
925 unlock_page(newpage);
926 page_cache_release(newpage);
930 unlock_page(oldpage);
931 page_cache_release(oldpage);
937 unlock_page(newpage);
940 cs->offset = buf->offset;
942 err = lock_request(cs->req);
949 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
950 unsigned offset, unsigned count)
952 struct pipe_buffer *buf;
955 if (cs->nr_segs == cs->pipe->buffers)
958 err = unlock_request(cs->req);
962 fuse_copy_finish(cs);
965 page_cache_get(page);
967 buf->offset = offset;
978 * Copy a page in the request to/from the userspace buffer. Must be
981 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
982 unsigned offset, unsigned count, int zeroing)
985 struct page *page = *pagep;
987 if (page && zeroing && count < PAGE_SIZE)
988 clear_highpage(page);
991 if (cs->write && cs->pipebufs && page) {
992 return fuse_ref_page(cs, page, offset, count);
993 } else if (!cs->len) {
994 if (cs->move_pages && page &&
995 offset == 0 && count == PAGE_SIZE) {
996 err = fuse_try_move_page(cs, pagep);
1000 err = fuse_copy_fill(cs);
1006 void *mapaddr = kmap_atomic(page);
1007 void *buf = mapaddr + offset;
1008 offset += fuse_copy_do(cs, &buf, &count);
1009 kunmap_atomic(mapaddr);
1011 offset += fuse_copy_do(cs, NULL, &count);
1013 if (page && !cs->write)
1014 flush_dcache_page(page);
1018 /* Copy pages in the request to/from userspace buffer */
1019 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1023 struct fuse_req *req = cs->req;
1025 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1027 unsigned offset = req->page_descs[i].offset;
1028 unsigned count = min(nbytes, req->page_descs[i].length);
1030 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1040 /* Copy a single argument in the request to/from userspace buffer */
1041 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1045 int err = fuse_copy_fill(cs);
1049 fuse_copy_do(cs, &val, &size);
1054 /* Copy request arguments to/from userspace buffer */
1055 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1056 unsigned argpages, struct fuse_arg *args,
1062 for (i = 0; !err && i < numargs; i++) {
1063 struct fuse_arg *arg = &args[i];
1064 if (i == numargs - 1 && argpages)
1065 err = fuse_copy_pages(cs, arg->size, zeroing);
1067 err = fuse_copy_one(cs, arg->value, arg->size);
1072 static int forget_pending(struct fuse_iqueue *fiq)
1074 return fiq->forget_list_head.next != NULL;
1077 static int request_pending(struct fuse_iqueue *fiq)
1079 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1080 forget_pending(fiq);
1083 /* Wait until a request is available on the pending list */
1084 static void request_wait(struct fuse_conn *fc)
1085 __releases(fc->iq.waitq.lock)
1086 __releases(fc->lock)
1087 __acquires(fc->lock)
1088 __acquires(fc->iq.waitq.lock)
1090 struct fuse_iqueue *fiq = &fc->iq;
1091 DECLARE_WAITQUEUE(wait, current);
1093 add_wait_queue_exclusive(&fiq->waitq, &wait);
1094 while (fiq->connected && !request_pending(fiq)) {
1095 set_current_state(TASK_INTERRUPTIBLE);
1096 if (signal_pending(current))
1099 spin_unlock(&fiq->waitq.lock);
1100 spin_unlock(&fc->lock);
1102 spin_lock(&fc->lock);
1103 spin_lock(&fiq->waitq.lock);
1105 set_current_state(TASK_RUNNING);
1106 remove_wait_queue(&fiq->waitq, &wait);
1110 * Transfer an interrupt request to userspace
1112 * Unlike other requests this is assembled on demand, without a need
1113 * to allocate a separate fuse_req structure.
1115 * Called with fc->lock held, releases it
1117 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1118 size_t nbytes, struct fuse_req *req)
1119 __releases(fc->iq.waitq.lock)
1120 __releases(fc->lock)
1122 struct fuse_iqueue *fiq = &fc->iq;
1123 struct fuse_in_header ih;
1124 struct fuse_interrupt_in arg;
1125 unsigned reqsize = sizeof(ih) + sizeof(arg);
1128 list_del_init(&req->intr_entry);
1129 req->intr_unique = fuse_get_unique(fiq);
1130 memset(&ih, 0, sizeof(ih));
1131 memset(&arg, 0, sizeof(arg));
1133 ih.opcode = FUSE_INTERRUPT;
1134 ih.unique = req->intr_unique;
1135 arg.unique = req->in.h.unique;
1137 spin_unlock(&fiq->waitq.lock);
1138 spin_unlock(&fc->lock);
1139 if (nbytes < reqsize)
1142 err = fuse_copy_one(cs, &ih, sizeof(ih));
1144 err = fuse_copy_one(cs, &arg, sizeof(arg));
1145 fuse_copy_finish(cs);
1147 return err ? err : reqsize;
1150 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1154 struct fuse_forget_link *head = fiq->forget_list_head.next;
1155 struct fuse_forget_link **newhead = &head;
1158 for (count = 0; *newhead != NULL && count < max; count++)
1159 newhead = &(*newhead)->next;
1161 fiq->forget_list_head.next = *newhead;
1163 if (fiq->forget_list_head.next == NULL)
1164 fiq->forget_list_tail = &fiq->forget_list_head;
1172 static int fuse_read_single_forget(struct fuse_conn *fc,
1173 struct fuse_copy_state *cs,
1175 __releases(fc->iq.waitq.lock)
1176 __releases(fc->lock)
1179 struct fuse_iqueue *fiq = &fc->iq;
1180 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1181 struct fuse_forget_in arg = {
1182 .nlookup = forget->forget_one.nlookup,
1184 struct fuse_in_header ih = {
1185 .opcode = FUSE_FORGET,
1186 .nodeid = forget->forget_one.nodeid,
1187 .unique = fuse_get_unique(fiq),
1188 .len = sizeof(ih) + sizeof(arg),
1191 spin_unlock(&fiq->waitq.lock);
1192 spin_unlock(&fc->lock);
1194 if (nbytes < ih.len)
1197 err = fuse_copy_one(cs, &ih, sizeof(ih));
1199 err = fuse_copy_one(cs, &arg, sizeof(arg));
1200 fuse_copy_finish(cs);
1208 static int fuse_read_batch_forget(struct fuse_conn *fc,
1209 struct fuse_copy_state *cs, size_t nbytes)
1210 __releases(fc->iq.waitq.lock)
1211 __releases(fc->lock)
1214 unsigned max_forgets;
1216 struct fuse_forget_link *head;
1217 struct fuse_iqueue *fiq = &fc->iq;
1218 struct fuse_batch_forget_in arg = { .count = 0 };
1219 struct fuse_in_header ih = {
1220 .opcode = FUSE_BATCH_FORGET,
1221 .unique = fuse_get_unique(fiq),
1222 .len = sizeof(ih) + sizeof(arg),
1225 if (nbytes < ih.len) {
1226 spin_unlock(&fiq->waitq.lock);
1227 spin_unlock(&fc->lock);
1231 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1232 head = dequeue_forget(fiq, max_forgets, &count);
1233 spin_unlock(&fiq->waitq.lock);
1234 spin_unlock(&fc->lock);
1237 ih.len += count * sizeof(struct fuse_forget_one);
1238 err = fuse_copy_one(cs, &ih, sizeof(ih));
1240 err = fuse_copy_one(cs, &arg, sizeof(arg));
1243 struct fuse_forget_link *forget = head;
1246 err = fuse_copy_one(cs, &forget->forget_one,
1247 sizeof(forget->forget_one));
1249 head = forget->next;
1253 fuse_copy_finish(cs);
1261 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1263 __releases(fc->iq.waitq.lock)
1264 __releases(fc->lock)
1266 struct fuse_iqueue *fiq = &fc->iq;
1268 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1269 return fuse_read_single_forget(fc, cs, nbytes);
1271 return fuse_read_batch_forget(fc, cs, nbytes);
1275 * Read a single request into the userspace filesystem's buffer. This
1276 * function waits until a request is available, then removes it from
1277 * the pending list and copies request data to userspace buffer. If
1278 * no reply is needed (FORGET) or request has been aborted or there
1279 * was an error during the copying then it's finished by calling
1280 * request_end(). Otherwise add it to the processing list, and set
1283 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1284 struct fuse_copy_state *cs, size_t nbytes)
1287 struct fuse_iqueue *fiq = &fc->iq;
1288 struct fuse_req *req;
1293 spin_lock(&fc->lock);
1294 spin_lock(&fiq->waitq.lock);
1296 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1297 !request_pending(fiq))
1302 if (!fiq->connected)
1305 if (!request_pending(fiq))
1308 if (!list_empty(&fiq->interrupts)) {
1309 req = list_entry(fiq->interrupts.next, struct fuse_req,
1311 return fuse_read_interrupt(fc, cs, nbytes, req);
1314 if (forget_pending(fiq)) {
1315 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1316 return fuse_read_forget(fc, cs, nbytes);
1318 if (fiq->forget_batch <= -8)
1319 fiq->forget_batch = 16;
1322 req = list_entry(fiq->pending.next, struct fuse_req, list);
1323 clear_bit(FR_PENDING, &req->flags);
1324 list_del_init(&req->list);
1325 spin_unlock(&fiq->waitq.lock);
1327 list_add(&req->list, &fc->io);
1330 reqsize = in->h.len;
1331 /* If request is too large, reply with an error and restart the read */
1332 if (nbytes < reqsize) {
1333 req->out.h.error = -EIO;
1334 /* SETXATTR is special, since it may contain too large data */
1335 if (in->h.opcode == FUSE_SETXATTR)
1336 req->out.h.error = -E2BIG;
1337 request_end(fc, req);
1340 spin_unlock(&fc->lock);
1342 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1344 err = fuse_copy_args(cs, in->numargs, in->argpages,
1345 (struct fuse_arg *) in->args, 0);
1346 fuse_copy_finish(cs);
1347 spin_lock(&fc->lock);
1348 clear_bit(FR_LOCKED, &req->flags);
1349 if (!fc->connected) {
1350 request_end(fc, req);
1354 req->out.h.error = -EIO;
1355 request_end(fc, req);
1358 if (!test_bit(FR_ISREPLY, &req->flags)) {
1359 request_end(fc, req);
1361 set_bit(FR_SENT, &req->flags);
1362 list_move_tail(&req->list, &fc->processing);
1363 if (test_bit(FR_INTERRUPTED, &req->flags))
1364 queue_interrupt(fiq, req);
1365 spin_unlock(&fc->lock);
1370 spin_unlock(&fiq->waitq.lock);
1371 spin_unlock(&fc->lock);
1375 static int fuse_dev_open(struct inode *inode, struct file *file)
1378 * The fuse device's file's private_data is used to hold
1379 * the fuse_conn(ection) when it is mounted, and is used to
1380 * keep track of whether the file has been mounted already.
1382 file->private_data = NULL;
1386 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1388 struct fuse_copy_state cs;
1389 struct file *file = iocb->ki_filp;
1390 struct fuse_conn *fc = fuse_get_conn(file);
1394 if (!iter_is_iovec(to))
1397 fuse_copy_init(&cs, 1, to);
1399 return fuse_dev_do_read(fc, file, &cs, iov_iter_count(to));
1402 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1403 struct pipe_inode_info *pipe,
1404 size_t len, unsigned int flags)
1409 struct pipe_buffer *bufs;
1410 struct fuse_copy_state cs;
1411 struct fuse_conn *fc = fuse_get_conn(in);
1415 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1419 fuse_copy_init(&cs, 1, NULL);
1422 ret = fuse_dev_do_read(fc, in, &cs, len);
1429 if (!pipe->readers) {
1430 send_sig(SIGPIPE, current, 0);
1436 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1441 while (page_nr < cs.nr_segs) {
1442 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1443 struct pipe_buffer *buf = pipe->bufs + newbuf;
1445 buf->page = bufs[page_nr].page;
1446 buf->offset = bufs[page_nr].offset;
1447 buf->len = bufs[page_nr].len;
1449 * Need to be careful about this. Having buf->ops in module
1450 * code can Oops if the buffer persists after module unload.
1452 buf->ops = &nosteal_pipe_buf_ops;
1467 if (waitqueue_active(&pipe->wait))
1468 wake_up_interruptible(&pipe->wait);
1469 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1473 for (; page_nr < cs.nr_segs; page_nr++)
1474 page_cache_release(bufs[page_nr].page);
1480 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1481 struct fuse_copy_state *cs)
1483 struct fuse_notify_poll_wakeup_out outarg;
1486 if (size != sizeof(outarg))
1489 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1493 fuse_copy_finish(cs);
1494 return fuse_notify_poll_wakeup(fc, &outarg);
1497 fuse_copy_finish(cs);
1501 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1502 struct fuse_copy_state *cs)
1504 struct fuse_notify_inval_inode_out outarg;
1507 if (size != sizeof(outarg))
1510 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1513 fuse_copy_finish(cs);
1515 down_read(&fc->killsb);
1518 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1519 outarg.off, outarg.len);
1521 up_read(&fc->killsb);
1525 fuse_copy_finish(cs);
1529 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1530 struct fuse_copy_state *cs)
1532 struct fuse_notify_inval_entry_out outarg;
1537 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1542 if (size < sizeof(outarg))
1545 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1549 err = -ENAMETOOLONG;
1550 if (outarg.namelen > FUSE_NAME_MAX)
1554 if (size != sizeof(outarg) + outarg.namelen + 1)
1558 name.len = outarg.namelen;
1559 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1562 fuse_copy_finish(cs);
1563 buf[outarg.namelen] = 0;
1564 name.hash = full_name_hash(name.name, name.len);
1566 down_read(&fc->killsb);
1569 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1570 up_read(&fc->killsb);
1576 fuse_copy_finish(cs);
1580 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1581 struct fuse_copy_state *cs)
1583 struct fuse_notify_delete_out outarg;
1588 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1593 if (size < sizeof(outarg))
1596 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1600 err = -ENAMETOOLONG;
1601 if (outarg.namelen > FUSE_NAME_MAX)
1605 if (size != sizeof(outarg) + outarg.namelen + 1)
1609 name.len = outarg.namelen;
1610 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1613 fuse_copy_finish(cs);
1614 buf[outarg.namelen] = 0;
1615 name.hash = full_name_hash(name.name, name.len);
1617 down_read(&fc->killsb);
1620 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1621 outarg.child, &name);
1622 up_read(&fc->killsb);
1628 fuse_copy_finish(cs);
1632 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1633 struct fuse_copy_state *cs)
1635 struct fuse_notify_store_out outarg;
1636 struct inode *inode;
1637 struct address_space *mapping;
1641 unsigned int offset;
1647 if (size < sizeof(outarg))
1650 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1655 if (size - sizeof(outarg) != outarg.size)
1658 nodeid = outarg.nodeid;
1660 down_read(&fc->killsb);
1666 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1670 mapping = inode->i_mapping;
1671 index = outarg.offset >> PAGE_CACHE_SHIFT;
1672 offset = outarg.offset & ~PAGE_CACHE_MASK;
1673 file_size = i_size_read(inode);
1674 end = outarg.offset + outarg.size;
1675 if (end > file_size) {
1677 fuse_write_update_size(inode, file_size);
1683 unsigned int this_num;
1686 page = find_or_create_page(mapping, index,
1687 mapping_gfp_mask(mapping));
1691 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1692 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1693 if (!err && offset == 0 &&
1694 (this_num == PAGE_CACHE_SIZE || file_size == end))
1695 SetPageUptodate(page);
1697 page_cache_release(page);
1712 up_read(&fc->killsb);
1714 fuse_copy_finish(cs);
1718 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1720 release_pages(req->pages, req->num_pages, false);
1723 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1724 struct fuse_notify_retrieve_out *outarg)
1727 struct address_space *mapping = inode->i_mapping;
1728 struct fuse_req *req;
1732 unsigned int offset;
1733 size_t total_len = 0;
1736 offset = outarg->offset & ~PAGE_CACHE_MASK;
1737 file_size = i_size_read(inode);
1740 if (outarg->offset > file_size)
1742 else if (outarg->offset + num > file_size)
1743 num = file_size - outarg->offset;
1745 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1746 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1748 req = fuse_get_req(fc, num_pages);
1750 return PTR_ERR(req);
1752 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1753 req->in.h.nodeid = outarg->nodeid;
1754 req->in.numargs = 2;
1755 req->in.argpages = 1;
1756 req->page_descs[0].offset = offset;
1757 req->end = fuse_retrieve_end;
1759 index = outarg->offset >> PAGE_CACHE_SHIFT;
1761 while (num && req->num_pages < num_pages) {
1763 unsigned int this_num;
1765 page = find_get_page(mapping, index);
1769 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1770 req->pages[req->num_pages] = page;
1771 req->page_descs[req->num_pages].length = this_num;
1776 total_len += this_num;
1779 req->misc.retrieve_in.offset = outarg->offset;
1780 req->misc.retrieve_in.size = total_len;
1781 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1782 req->in.args[0].value = &req->misc.retrieve_in;
1783 req->in.args[1].size = total_len;
1785 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1787 fuse_retrieve_end(fc, req);
1792 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1793 struct fuse_copy_state *cs)
1795 struct fuse_notify_retrieve_out outarg;
1796 struct inode *inode;
1800 if (size != sizeof(outarg))
1803 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1807 fuse_copy_finish(cs);
1809 down_read(&fc->killsb);
1812 u64 nodeid = outarg.nodeid;
1814 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1816 err = fuse_retrieve(fc, inode, &outarg);
1820 up_read(&fc->killsb);
1825 fuse_copy_finish(cs);
1829 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1830 unsigned int size, struct fuse_copy_state *cs)
1832 /* Don't try to move pages (yet) */
1836 case FUSE_NOTIFY_POLL:
1837 return fuse_notify_poll(fc, size, cs);
1839 case FUSE_NOTIFY_INVAL_INODE:
1840 return fuse_notify_inval_inode(fc, size, cs);
1842 case FUSE_NOTIFY_INVAL_ENTRY:
1843 return fuse_notify_inval_entry(fc, size, cs);
1845 case FUSE_NOTIFY_STORE:
1846 return fuse_notify_store(fc, size, cs);
1848 case FUSE_NOTIFY_RETRIEVE:
1849 return fuse_notify_retrieve(fc, size, cs);
1851 case FUSE_NOTIFY_DELETE:
1852 return fuse_notify_delete(fc, size, cs);
1855 fuse_copy_finish(cs);
1860 /* Look up request on processing list by unique ID */
1861 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1863 struct fuse_req *req;
1865 list_for_each_entry(req, &fc->processing, list) {
1866 if (req->in.h.unique == unique || req->intr_unique == unique)
1872 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1875 unsigned reqsize = sizeof(struct fuse_out_header);
1878 return nbytes != reqsize ? -EINVAL : 0;
1880 reqsize += len_args(out->numargs, out->args);
1882 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1884 else if (reqsize > nbytes) {
1885 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1886 unsigned diffsize = reqsize - nbytes;
1887 if (diffsize > lastarg->size)
1889 lastarg->size -= diffsize;
1891 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1896 * Write a single reply to a request. First the header is copied from
1897 * the write buffer. The request is then searched on the processing
1898 * list by the unique ID found in the header. If found, then remove
1899 * it from the list and copy the rest of the buffer to the request.
1900 * The request is finished by calling request_end()
1902 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1903 struct fuse_copy_state *cs, size_t nbytes)
1906 struct fuse_req *req;
1907 struct fuse_out_header oh;
1909 if (nbytes < sizeof(struct fuse_out_header))
1912 err = fuse_copy_one(cs, &oh, sizeof(oh));
1917 if (oh.len != nbytes)
1921 * Zero oh.unique indicates unsolicited notification message
1922 * and error contains notification code.
1925 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1926 return err ? err : nbytes;
1930 if (oh.error <= -1000 || oh.error > 0)
1933 spin_lock(&fc->lock);
1938 req = request_find(fc, oh.unique);
1942 /* Is it an interrupt reply? */
1943 if (req->intr_unique == oh.unique) {
1945 if (nbytes != sizeof(struct fuse_out_header))
1948 if (oh.error == -ENOSYS)
1949 fc->no_interrupt = 1;
1950 else if (oh.error == -EAGAIN)
1951 queue_interrupt(&fc->iq, req);
1953 spin_unlock(&fc->lock);
1954 fuse_copy_finish(cs);
1958 clear_bit(FR_SENT, &req->flags);
1959 list_move(&req->list, &fc->io);
1961 set_bit(FR_LOCKED, &req->flags);
1963 if (!req->out.page_replace)
1965 spin_unlock(&fc->lock);
1967 err = copy_out_args(cs, &req->out, nbytes);
1968 fuse_copy_finish(cs);
1970 spin_lock(&fc->lock);
1971 clear_bit(FR_LOCKED, &req->flags);
1975 req->out.h.error = -EIO;
1976 request_end(fc, req);
1978 return err ? err : nbytes;
1981 spin_unlock(&fc->lock);
1983 fuse_copy_finish(cs);
1987 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1989 struct fuse_copy_state cs;
1990 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1994 if (!iter_is_iovec(from))
1997 fuse_copy_init(&cs, 0, from);
1999 return fuse_dev_do_write(fc, &cs, iov_iter_count(from));
2002 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
2003 struct file *out, loff_t *ppos,
2004 size_t len, unsigned int flags)
2008 struct pipe_buffer *bufs;
2009 struct fuse_copy_state cs;
2010 struct fuse_conn *fc;
2014 fc = fuse_get_conn(out);
2018 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
2025 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
2026 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
2036 struct pipe_buffer *ibuf;
2037 struct pipe_buffer *obuf;
2039 BUG_ON(nbuf >= pipe->buffers);
2040 BUG_ON(!pipe->nrbufs);
2041 ibuf = &pipe->bufs[pipe->curbuf];
2044 if (rem >= ibuf->len) {
2047 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2050 ibuf->ops->get(pipe, ibuf);
2052 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2054 ibuf->offset += obuf->len;
2055 ibuf->len -= obuf->len;
2062 fuse_copy_init(&cs, 0, NULL);
2067 if (flags & SPLICE_F_MOVE)
2070 ret = fuse_dev_do_write(fc, &cs, len);
2072 for (idx = 0; idx < nbuf; idx++) {
2073 struct pipe_buffer *buf = &bufs[idx];
2074 buf->ops->release(pipe, buf);
2081 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2083 unsigned mask = POLLOUT | POLLWRNORM;
2084 struct fuse_iqueue *fiq;
2085 struct fuse_conn *fc = fuse_get_conn(file);
2090 poll_wait(file, &fiq->waitq, wait);
2092 spin_lock(&fc->lock);
2093 spin_lock(&fiq->waitq.lock);
2094 if (!fiq->connected)
2096 else if (request_pending(fiq))
2097 mask |= POLLIN | POLLRDNORM;
2098 spin_unlock(&fiq->waitq.lock);
2099 spin_unlock(&fc->lock);
2105 * Abort all requests on the given list (pending or processing)
2107 * This function releases and reacquires fc->lock
2109 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2110 __releases(fc->lock)
2111 __acquires(fc->lock)
2113 while (!list_empty(head)) {
2114 struct fuse_req *req;
2115 req = list_entry(head->next, struct fuse_req, list);
2116 req->out.h.error = -ECONNABORTED;
2117 clear_bit(FR_PENDING, &req->flags);
2118 clear_bit(FR_SENT, &req->flags);
2119 request_end(fc, req);
2120 spin_lock(&fc->lock);
2124 static void end_polls(struct fuse_conn *fc)
2128 p = rb_first(&fc->polled_files);
2131 struct fuse_file *ff;
2132 ff = rb_entry(p, struct fuse_file, polled_node);
2133 wake_up_interruptible_all(&ff->poll_wait);
2140 * Abort all requests.
2142 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2145 * The same effect is usually achievable through killing the filesystem daemon
2146 * and all users of the filesystem. The exception is the combination of an
2147 * asynchronous request and the tricky deadlock (see
2148 * Documentation/filesystems/fuse.txt).
2150 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2151 * requests, they should be finished off immediately. Locked requests will be
2152 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2153 * requests. It is possible that some request will finish before we can. This
2154 * is OK, the request will in that case be removed from the list before we touch
2157 void fuse_abort_conn(struct fuse_conn *fc)
2159 struct fuse_iqueue *fiq = &fc->iq;
2161 spin_lock(&fc->lock);
2162 if (fc->connected) {
2163 struct fuse_req *req, *next;
2169 fuse_set_initialized(fc);
2170 list_for_each_entry_safe(req, next, &fc->io, list) {
2171 req->out.h.error = -ECONNABORTED;
2172 spin_lock(&req->waitq.lock);
2173 set_bit(FR_ABORTED, &req->flags);
2174 if (!test_bit(FR_LOCKED, &req->flags))
2175 list_move(&req->list, &to_end1);
2176 spin_unlock(&req->waitq.lock);
2178 fc->max_background = UINT_MAX;
2181 spin_lock(&fiq->waitq.lock);
2183 list_splice_init(&fiq->pending, &to_end2);
2184 while (forget_pending(fiq))
2185 kfree(dequeue_forget(fiq, 1, NULL));
2186 wake_up_all_locked(&fiq->waitq);
2187 spin_unlock(&fiq->waitq.lock);
2188 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2190 list_splice_init(&fc->processing, &to_end2);
2191 while (!list_empty(&to_end1)) {
2192 req = list_first_entry(&to_end1, struct fuse_req, list);
2193 __fuse_get_request(req);
2194 request_end(fc, req);
2195 spin_lock(&fc->lock);
2197 end_requests(fc, &to_end2);
2199 wake_up_all(&fc->blocked_waitq);
2201 spin_unlock(&fc->lock);
2203 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2205 int fuse_dev_release(struct inode *inode, struct file *file)
2207 struct fuse_conn *fc = fuse_get_conn(file);
2209 WARN_ON(!list_empty(&fc->io));
2210 WARN_ON(fc->iq.fasync != NULL);
2211 fuse_abort_conn(fc);
2217 EXPORT_SYMBOL_GPL(fuse_dev_release);
2219 static int fuse_dev_fasync(int fd, struct file *file, int on)
2221 struct fuse_conn *fc = fuse_get_conn(file);
2225 /* No locking - fasync_helper does its own locking */
2226 return fasync_helper(fd, file, on, &fc->iq.fasync);
2229 const struct file_operations fuse_dev_operations = {
2230 .owner = THIS_MODULE,
2231 .open = fuse_dev_open,
2232 .llseek = no_llseek,
2233 .read_iter = fuse_dev_read,
2234 .splice_read = fuse_dev_splice_read,
2235 .write_iter = fuse_dev_write,
2236 .splice_write = fuse_dev_splice_write,
2237 .poll = fuse_dev_poll,
2238 .release = fuse_dev_release,
2239 .fasync = fuse_dev_fasync,
2241 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2243 static struct miscdevice fuse_miscdevice = {
2244 .minor = FUSE_MINOR,
2246 .fops = &fuse_dev_operations,
2249 int __init fuse_dev_init(void)
2252 fuse_req_cachep = kmem_cache_create("fuse_request",
2253 sizeof(struct fuse_req),
2255 if (!fuse_req_cachep)
2258 err = misc_register(&fuse_miscdevice);
2260 goto out_cache_clean;
2265 kmem_cache_destroy(fuse_req_cachep);
2270 void fuse_dev_cleanup(void)
2272 misc_deregister(&fuse_miscdevice);
2273 kmem_cache_destroy(fuse_req_cachep);