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(&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 if (fiq->connected) {
345 fiq->forget_list_tail->next = forget;
346 fiq->forget_list_tail = forget;
347 wake_up(&fiq->waitq);
348 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
352 spin_unlock(&fc->lock);
355 static void flush_bg_queue(struct fuse_conn *fc)
357 while (fc->active_background < fc->max_background &&
358 !list_empty(&fc->bg_queue)) {
359 struct fuse_req *req;
360 struct fuse_iqueue *fiq = &fc->iq;
362 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
363 list_del(&req->list);
364 fc->active_background++;
365 req->in.h.unique = fuse_get_unique(fiq);
366 queue_request(fiq, req);
371 * This function is called when a request is finished. Either a reply
372 * has arrived or it was aborted (and not yet sent) or some error
373 * occurred during communication with userspace, or the device file
374 * was closed. The requester thread is woken up (if still waiting),
375 * the 'end' callback is called if given, else the reference to the
376 * request is released
378 * Called with fc->lock, unlocks it
380 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
383 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
385 list_del_init(&req->list);
386 list_del_init(&req->intr_entry);
387 WARN_ON(test_bit(FR_PENDING, &req->flags));
388 WARN_ON(test_bit(FR_SENT, &req->flags));
390 set_bit(FR_FINISHED, &req->flags);
391 if (test_bit(FR_BACKGROUND, &req->flags)) {
392 clear_bit(FR_BACKGROUND, &req->flags);
393 if (fc->num_background == fc->max_background)
396 /* Wake up next waiter, if any */
397 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
398 wake_up(&fc->blocked_waitq);
400 if (fc->num_background == fc->congestion_threshold &&
401 fc->connected && fc->bdi_initialized) {
402 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
403 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
405 fc->num_background--;
406 fc->active_background--;
409 spin_unlock(&fc->lock);
410 wake_up(&req->waitq);
413 fuse_put_request(fc, req);
416 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
418 list_add_tail(&req->intr_entry, &fiq->interrupts);
419 wake_up(&fiq->waitq);
420 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
423 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
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));
434 spin_lock(&fc->lock);
435 set_bit(FR_INTERRUPTED, &req->flags);
436 if (test_bit(FR_SENT, &req->flags))
437 queue_interrupt(&fc->iq, req);
438 spin_unlock(&fc->lock);
441 if (!test_bit(FR_FORCE, &req->flags)) {
444 /* Only fatal signals may interrupt this */
446 err = wait_event_interruptible(req->waitq,
447 test_bit(FR_FINISHED, &req->flags));
448 restore_sigs(&oldset);
453 spin_lock(&fc->lock);
454 /* Request is not yet in userspace, bail out */
455 if (test_bit(FR_PENDING, &req->flags)) {
456 list_del(&req->list);
457 spin_unlock(&fc->lock);
458 __fuse_put_request(req);
459 req->out.h.error = -EINTR;
462 spin_unlock(&fc->lock);
466 * Either request is already in userspace, or it was forced.
469 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
472 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
474 struct fuse_iqueue *fiq = &fc->iq;
476 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
477 spin_lock(&fc->lock);
478 if (!fiq->connected) {
479 spin_unlock(&fc->lock);
480 req->out.h.error = -ENOTCONN;
482 req->in.h.unique = fuse_get_unique(fiq);
483 queue_request(fiq, req);
484 /* acquire extra reference, since request is still needed
485 after request_end() */
486 __fuse_get_request(req);
487 spin_unlock(&fc->lock);
489 request_wait_answer(fc, req);
490 /* Pairs with smp_wmb() in request_end() */
495 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
497 __set_bit(FR_ISREPLY, &req->flags);
498 if (!test_bit(FR_WAITING, &req->flags)) {
499 __set_bit(FR_WAITING, &req->flags);
500 atomic_inc(&fc->num_waiting);
502 __fuse_request_send(fc, req);
504 EXPORT_SYMBOL_GPL(fuse_request_send);
506 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
508 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
509 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
512 switch (args->in.h.opcode) {
519 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
523 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
527 if (fc->minor < 12) {
528 switch (args->in.h.opcode) {
530 args->in.args[0].size = sizeof(struct fuse_open_in);
533 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
539 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
541 struct fuse_req *req;
544 req = fuse_get_req(fc, 0);
548 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
549 fuse_adjust_compat(fc, args);
551 req->in.h.opcode = args->in.h.opcode;
552 req->in.h.nodeid = args->in.h.nodeid;
553 req->in.numargs = args->in.numargs;
554 memcpy(req->in.args, args->in.args,
555 args->in.numargs * sizeof(struct fuse_in_arg));
556 req->out.argvar = args->out.argvar;
557 req->out.numargs = args->out.numargs;
558 memcpy(req->out.args, args->out.args,
559 args->out.numargs * sizeof(struct fuse_arg));
560 fuse_request_send(fc, req);
561 ret = req->out.h.error;
562 if (!ret && args->out.argvar) {
563 BUG_ON(args->out.numargs != 1);
564 ret = req->out.args[0].size;
566 fuse_put_request(fc, req);
572 * Called under fc->lock
574 * fc->connected must have been checked previously
576 void fuse_request_send_background_locked(struct fuse_conn *fc,
577 struct fuse_req *req)
579 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
580 if (!test_bit(FR_WAITING, &req->flags)) {
581 __set_bit(FR_WAITING, &req->flags);
582 atomic_inc(&fc->num_waiting);
584 __set_bit(FR_ISREPLY, &req->flags);
585 fc->num_background++;
586 if (fc->num_background == fc->max_background)
588 if (fc->num_background == fc->congestion_threshold &&
589 fc->bdi_initialized) {
590 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
591 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
593 list_add_tail(&req->list, &fc->bg_queue);
597 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
600 spin_lock(&fc->lock);
602 fuse_request_send_background_locked(fc, req);
603 spin_unlock(&fc->lock);
605 spin_unlock(&fc->lock);
606 req->out.h.error = -ENOTCONN;
608 fuse_put_request(fc, req);
611 EXPORT_SYMBOL_GPL(fuse_request_send_background);
613 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
614 struct fuse_req *req, u64 unique)
617 struct fuse_iqueue *fiq = &fc->iq;
619 __clear_bit(FR_ISREPLY, &req->flags);
620 req->in.h.unique = unique;
621 spin_lock(&fc->lock);
622 if (fiq->connected) {
623 queue_request(fiq, req);
626 spin_unlock(&fc->lock);
631 void fuse_force_forget(struct file *file, u64 nodeid)
633 struct inode *inode = file_inode(file);
634 struct fuse_conn *fc = get_fuse_conn(inode);
635 struct fuse_req *req;
636 struct fuse_forget_in inarg;
638 memset(&inarg, 0, sizeof(inarg));
640 req = fuse_get_req_nofail_nopages(fc, file);
641 req->in.h.opcode = FUSE_FORGET;
642 req->in.h.nodeid = nodeid;
644 req->in.args[0].size = sizeof(inarg);
645 req->in.args[0].value = &inarg;
646 __clear_bit(FR_ISREPLY, &req->flags);
647 __fuse_request_send(fc, req);
649 fuse_put_request(fc, req);
653 * Lock the request. Up to the next unlock_request() there mustn't be
654 * anything that could cause a page-fault. If the request was already
657 static int lock_request(struct fuse_req *req)
661 spin_lock(&req->waitq.lock);
662 if (test_bit(FR_ABORTED, &req->flags))
665 set_bit(FR_LOCKED, &req->flags);
666 spin_unlock(&req->waitq.lock);
672 * Unlock request. If it was aborted while locked, caller is responsible
673 * for unlocking and ending the request.
675 static int unlock_request(struct fuse_req *req)
679 spin_lock(&req->waitq.lock);
680 if (test_bit(FR_ABORTED, &req->flags))
683 clear_bit(FR_LOCKED, &req->flags);
684 spin_unlock(&req->waitq.lock);
689 struct fuse_copy_state {
691 struct fuse_req *req;
692 struct iov_iter *iter;
693 struct pipe_buffer *pipebufs;
694 struct pipe_buffer *currbuf;
695 struct pipe_inode_info *pipe;
696 unsigned long nr_segs;
700 unsigned move_pages:1;
703 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
704 struct iov_iter *iter)
706 memset(cs, 0, sizeof(*cs));
711 /* Unmap and put previous page of userspace buffer */
712 static void fuse_copy_finish(struct fuse_copy_state *cs)
715 struct pipe_buffer *buf = cs->currbuf;
718 buf->len = PAGE_SIZE - cs->len;
722 flush_dcache_page(cs->pg);
723 set_page_dirty_lock(cs->pg);
731 * Get another pagefull of userspace buffer, and map it to kernel
732 * address space, and lock request
734 static int fuse_copy_fill(struct fuse_copy_state *cs)
739 err = unlock_request(cs->req);
743 fuse_copy_finish(cs);
745 struct pipe_buffer *buf = cs->pipebufs;
748 err = buf->ops->confirm(cs->pipe, buf);
752 BUG_ON(!cs->nr_segs);
755 cs->offset = buf->offset;
760 if (cs->nr_segs == cs->pipe->buffers)
763 page = alloc_page(GFP_HIGHUSER);
780 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
788 iov_iter_advance(cs->iter, err);
791 return lock_request(cs->req);
794 /* Do as much copy to/from userspace buffer as we can */
795 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
797 unsigned ncpy = min(*size, cs->len);
799 void *pgaddr = kmap_atomic(cs->pg);
800 void *buf = pgaddr + cs->offset;
803 memcpy(buf, *val, ncpy);
805 memcpy(*val, buf, ncpy);
807 kunmap_atomic(pgaddr);
816 static int fuse_check_page(struct page *page)
818 if (page_mapcount(page) ||
819 page->mapping != NULL ||
820 page_count(page) != 1 ||
821 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
828 printk(KERN_WARNING "fuse: trying to steal weird page\n");
829 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);
835 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
838 struct page *oldpage = *pagep;
839 struct page *newpage;
840 struct pipe_buffer *buf = cs->pipebufs;
842 err = unlock_request(cs->req);
846 fuse_copy_finish(cs);
848 err = buf->ops->confirm(cs->pipe, buf);
852 BUG_ON(!cs->nr_segs);
858 if (cs->len != PAGE_SIZE)
861 if (buf->ops->steal(cs->pipe, buf) != 0)
866 if (!PageUptodate(newpage))
867 SetPageUptodate(newpage);
869 ClearPageMappedToDisk(newpage);
871 if (fuse_check_page(newpage) != 0)
872 goto out_fallback_unlock;
875 * This is a new and locked page, it shouldn't be mapped or
876 * have any special flags on it
878 if (WARN_ON(page_mapped(oldpage)))
879 goto out_fallback_unlock;
880 if (WARN_ON(page_has_private(oldpage)))
881 goto out_fallback_unlock;
882 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
883 goto out_fallback_unlock;
884 if (WARN_ON(PageMlocked(oldpage)))
885 goto out_fallback_unlock;
887 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
889 unlock_page(newpage);
893 page_cache_get(newpage);
895 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
896 lru_cache_add_file(newpage);
899 spin_lock(&cs->req->waitq.lock);
900 if (test_bit(FR_ABORTED, &cs->req->flags))
904 spin_unlock(&cs->req->waitq.lock);
907 unlock_page(newpage);
908 page_cache_release(newpage);
912 unlock_page(oldpage);
913 page_cache_release(oldpage);
919 unlock_page(newpage);
922 cs->offset = buf->offset;
924 err = lock_request(cs->req);
931 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
932 unsigned offset, unsigned count)
934 struct pipe_buffer *buf;
937 if (cs->nr_segs == cs->pipe->buffers)
940 err = unlock_request(cs->req);
944 fuse_copy_finish(cs);
947 page_cache_get(page);
949 buf->offset = offset;
960 * Copy a page in the request to/from the userspace buffer. Must be
963 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
964 unsigned offset, unsigned count, int zeroing)
967 struct page *page = *pagep;
969 if (page && zeroing && count < PAGE_SIZE)
970 clear_highpage(page);
973 if (cs->write && cs->pipebufs && page) {
974 return fuse_ref_page(cs, page, offset, count);
975 } else if (!cs->len) {
976 if (cs->move_pages && page &&
977 offset == 0 && count == PAGE_SIZE) {
978 err = fuse_try_move_page(cs, pagep);
982 err = fuse_copy_fill(cs);
988 void *mapaddr = kmap_atomic(page);
989 void *buf = mapaddr + offset;
990 offset += fuse_copy_do(cs, &buf, &count);
991 kunmap_atomic(mapaddr);
993 offset += fuse_copy_do(cs, NULL, &count);
995 if (page && !cs->write)
996 flush_dcache_page(page);
1000 /* Copy pages in the request to/from userspace buffer */
1001 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1005 struct fuse_req *req = cs->req;
1007 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1009 unsigned offset = req->page_descs[i].offset;
1010 unsigned count = min(nbytes, req->page_descs[i].length);
1012 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1022 /* Copy a single argument in the request to/from userspace buffer */
1023 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1027 int err = fuse_copy_fill(cs);
1031 fuse_copy_do(cs, &val, &size);
1036 /* Copy request arguments to/from userspace buffer */
1037 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1038 unsigned argpages, struct fuse_arg *args,
1044 for (i = 0; !err && i < numargs; i++) {
1045 struct fuse_arg *arg = &args[i];
1046 if (i == numargs - 1 && argpages)
1047 err = fuse_copy_pages(cs, arg->size, zeroing);
1049 err = fuse_copy_one(cs, arg->value, arg->size);
1054 static int forget_pending(struct fuse_iqueue *fiq)
1056 return fiq->forget_list_head.next != NULL;
1059 static int request_pending(struct fuse_iqueue *fiq)
1061 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1062 forget_pending(fiq);
1065 /* Wait until a request is available on the pending list */
1066 static void request_wait(struct fuse_conn *fc)
1067 __releases(fc->lock)
1068 __acquires(fc->lock)
1070 struct fuse_iqueue *fiq = &fc->iq;
1071 DECLARE_WAITQUEUE(wait, current);
1073 add_wait_queue_exclusive(&fiq->waitq, &wait);
1074 while (fiq->connected && !request_pending(fiq)) {
1075 set_current_state(TASK_INTERRUPTIBLE);
1076 if (signal_pending(current))
1079 spin_unlock(&fc->lock);
1081 spin_lock(&fc->lock);
1083 set_current_state(TASK_RUNNING);
1084 remove_wait_queue(&fiq->waitq, &wait);
1088 * Transfer an interrupt request to userspace
1090 * Unlike other requests this is assembled on demand, without a need
1091 * to allocate a separate fuse_req structure.
1093 * Called with fc->lock held, releases it
1095 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1096 size_t nbytes, struct fuse_req *req)
1097 __releases(fc->lock)
1099 struct fuse_in_header ih;
1100 struct fuse_interrupt_in arg;
1101 unsigned reqsize = sizeof(ih) + sizeof(arg);
1104 list_del_init(&req->intr_entry);
1105 req->intr_unique = fuse_get_unique(&fc->iq);
1106 memset(&ih, 0, sizeof(ih));
1107 memset(&arg, 0, sizeof(arg));
1109 ih.opcode = FUSE_INTERRUPT;
1110 ih.unique = req->intr_unique;
1111 arg.unique = req->in.h.unique;
1113 spin_unlock(&fc->lock);
1114 if (nbytes < reqsize)
1117 err = fuse_copy_one(cs, &ih, sizeof(ih));
1119 err = fuse_copy_one(cs, &arg, sizeof(arg));
1120 fuse_copy_finish(cs);
1122 return err ? err : reqsize;
1125 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1129 struct fuse_forget_link *head = fiq->forget_list_head.next;
1130 struct fuse_forget_link **newhead = &head;
1133 for (count = 0; *newhead != NULL && count < max; count++)
1134 newhead = &(*newhead)->next;
1136 fiq->forget_list_head.next = *newhead;
1138 if (fiq->forget_list_head.next == NULL)
1139 fiq->forget_list_tail = &fiq->forget_list_head;
1147 static int fuse_read_single_forget(struct fuse_conn *fc,
1148 struct fuse_copy_state *cs,
1150 __releases(fc->lock)
1153 struct fuse_iqueue *fiq = &fc->iq;
1154 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1155 struct fuse_forget_in arg = {
1156 .nlookup = forget->forget_one.nlookup,
1158 struct fuse_in_header ih = {
1159 .opcode = FUSE_FORGET,
1160 .nodeid = forget->forget_one.nodeid,
1161 .unique = fuse_get_unique(fiq),
1162 .len = sizeof(ih) + sizeof(arg),
1165 spin_unlock(&fc->lock);
1167 if (nbytes < ih.len)
1170 err = fuse_copy_one(cs, &ih, sizeof(ih));
1172 err = fuse_copy_one(cs, &arg, sizeof(arg));
1173 fuse_copy_finish(cs);
1181 static int fuse_read_batch_forget(struct fuse_conn *fc,
1182 struct fuse_copy_state *cs, size_t nbytes)
1183 __releases(fc->lock)
1186 unsigned max_forgets;
1188 struct fuse_forget_link *head;
1189 struct fuse_iqueue *fiq = &fc->iq;
1190 struct fuse_batch_forget_in arg = { .count = 0 };
1191 struct fuse_in_header ih = {
1192 .opcode = FUSE_BATCH_FORGET,
1193 .unique = fuse_get_unique(fiq),
1194 .len = sizeof(ih) + sizeof(arg),
1197 if (nbytes < ih.len) {
1198 spin_unlock(&fc->lock);
1202 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1203 head = dequeue_forget(fiq, max_forgets, &count);
1204 spin_unlock(&fc->lock);
1207 ih.len += count * sizeof(struct fuse_forget_one);
1208 err = fuse_copy_one(cs, &ih, sizeof(ih));
1210 err = fuse_copy_one(cs, &arg, sizeof(arg));
1213 struct fuse_forget_link *forget = head;
1216 err = fuse_copy_one(cs, &forget->forget_one,
1217 sizeof(forget->forget_one));
1219 head = forget->next;
1223 fuse_copy_finish(cs);
1231 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1233 __releases(fc->lock)
1235 struct fuse_iqueue *fiq = &fc->iq;
1237 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1238 return fuse_read_single_forget(fc, cs, nbytes);
1240 return fuse_read_batch_forget(fc, cs, nbytes);
1244 * Read a single request into the userspace filesystem's buffer. This
1245 * function waits until a request is available, then removes it from
1246 * the pending list and copies request data to userspace buffer. If
1247 * no reply is needed (FORGET) or request has been aborted or there
1248 * was an error during the copying then it's finished by calling
1249 * request_end(). Otherwise add it to the processing list, and set
1252 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1253 struct fuse_copy_state *cs, size_t nbytes)
1256 struct fuse_iqueue *fiq = &fc->iq;
1257 struct fuse_req *req;
1262 spin_lock(&fc->lock);
1264 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1265 !request_pending(fiq))
1270 if (!fiq->connected)
1273 if (!request_pending(fiq))
1276 if (!list_empty(&fiq->interrupts)) {
1277 req = list_entry(fiq->interrupts.next, struct fuse_req,
1279 return fuse_read_interrupt(fc, cs, nbytes, req);
1282 if (forget_pending(fiq)) {
1283 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1284 return fuse_read_forget(fc, cs, nbytes);
1286 if (fiq->forget_batch <= -8)
1287 fiq->forget_batch = 16;
1290 req = list_entry(fiq->pending.next, struct fuse_req, list);
1291 clear_bit(FR_PENDING, &req->flags);
1292 list_move(&req->list, &fc->io);
1295 reqsize = in->h.len;
1296 /* If request is too large, reply with an error and restart the read */
1297 if (nbytes < reqsize) {
1298 req->out.h.error = -EIO;
1299 /* SETXATTR is special, since it may contain too large data */
1300 if (in->h.opcode == FUSE_SETXATTR)
1301 req->out.h.error = -E2BIG;
1302 request_end(fc, req);
1305 spin_unlock(&fc->lock);
1307 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1309 err = fuse_copy_args(cs, in->numargs, in->argpages,
1310 (struct fuse_arg *) in->args, 0);
1311 fuse_copy_finish(cs);
1312 spin_lock(&fc->lock);
1313 clear_bit(FR_LOCKED, &req->flags);
1314 if (!fc->connected) {
1315 request_end(fc, req);
1319 req->out.h.error = -EIO;
1320 request_end(fc, req);
1323 if (!test_bit(FR_ISREPLY, &req->flags)) {
1324 request_end(fc, req);
1326 set_bit(FR_SENT, &req->flags);
1327 list_move_tail(&req->list, &fc->processing);
1328 if (test_bit(FR_INTERRUPTED, &req->flags))
1329 queue_interrupt(fiq, req);
1330 spin_unlock(&fc->lock);
1335 spin_unlock(&fc->lock);
1339 static int fuse_dev_open(struct inode *inode, struct file *file)
1342 * The fuse device's file's private_data is used to hold
1343 * the fuse_conn(ection) when it is mounted, and is used to
1344 * keep track of whether the file has been mounted already.
1346 file->private_data = NULL;
1350 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1352 struct fuse_copy_state cs;
1353 struct file *file = iocb->ki_filp;
1354 struct fuse_conn *fc = fuse_get_conn(file);
1358 if (!iter_is_iovec(to))
1361 fuse_copy_init(&cs, 1, to);
1363 return fuse_dev_do_read(fc, file, &cs, iov_iter_count(to));
1366 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1367 struct pipe_inode_info *pipe,
1368 size_t len, unsigned int flags)
1373 struct pipe_buffer *bufs;
1374 struct fuse_copy_state cs;
1375 struct fuse_conn *fc = fuse_get_conn(in);
1379 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1383 fuse_copy_init(&cs, 1, NULL);
1386 ret = fuse_dev_do_read(fc, in, &cs, len);
1393 if (!pipe->readers) {
1394 send_sig(SIGPIPE, current, 0);
1400 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1405 while (page_nr < cs.nr_segs) {
1406 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1407 struct pipe_buffer *buf = pipe->bufs + newbuf;
1409 buf->page = bufs[page_nr].page;
1410 buf->offset = bufs[page_nr].offset;
1411 buf->len = bufs[page_nr].len;
1413 * Need to be careful about this. Having buf->ops in module
1414 * code can Oops if the buffer persists after module unload.
1416 buf->ops = &nosteal_pipe_buf_ops;
1431 if (waitqueue_active(&pipe->wait))
1432 wake_up_interruptible(&pipe->wait);
1433 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1437 for (; page_nr < cs.nr_segs; page_nr++)
1438 page_cache_release(bufs[page_nr].page);
1444 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1445 struct fuse_copy_state *cs)
1447 struct fuse_notify_poll_wakeup_out outarg;
1450 if (size != sizeof(outarg))
1453 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1457 fuse_copy_finish(cs);
1458 return fuse_notify_poll_wakeup(fc, &outarg);
1461 fuse_copy_finish(cs);
1465 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1466 struct fuse_copy_state *cs)
1468 struct fuse_notify_inval_inode_out outarg;
1471 if (size != sizeof(outarg))
1474 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1477 fuse_copy_finish(cs);
1479 down_read(&fc->killsb);
1482 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1483 outarg.off, outarg.len);
1485 up_read(&fc->killsb);
1489 fuse_copy_finish(cs);
1493 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1494 struct fuse_copy_state *cs)
1496 struct fuse_notify_inval_entry_out outarg;
1501 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1506 if (size < sizeof(outarg))
1509 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1513 err = -ENAMETOOLONG;
1514 if (outarg.namelen > FUSE_NAME_MAX)
1518 if (size != sizeof(outarg) + outarg.namelen + 1)
1522 name.len = outarg.namelen;
1523 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1526 fuse_copy_finish(cs);
1527 buf[outarg.namelen] = 0;
1528 name.hash = full_name_hash(name.name, name.len);
1530 down_read(&fc->killsb);
1533 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1534 up_read(&fc->killsb);
1540 fuse_copy_finish(cs);
1544 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1545 struct fuse_copy_state *cs)
1547 struct fuse_notify_delete_out outarg;
1552 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1557 if (size < sizeof(outarg))
1560 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1564 err = -ENAMETOOLONG;
1565 if (outarg.namelen > FUSE_NAME_MAX)
1569 if (size != sizeof(outarg) + outarg.namelen + 1)
1573 name.len = outarg.namelen;
1574 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1577 fuse_copy_finish(cs);
1578 buf[outarg.namelen] = 0;
1579 name.hash = full_name_hash(name.name, name.len);
1581 down_read(&fc->killsb);
1584 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1585 outarg.child, &name);
1586 up_read(&fc->killsb);
1592 fuse_copy_finish(cs);
1596 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1597 struct fuse_copy_state *cs)
1599 struct fuse_notify_store_out outarg;
1600 struct inode *inode;
1601 struct address_space *mapping;
1605 unsigned int offset;
1611 if (size < sizeof(outarg))
1614 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1619 if (size - sizeof(outarg) != outarg.size)
1622 nodeid = outarg.nodeid;
1624 down_read(&fc->killsb);
1630 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1634 mapping = inode->i_mapping;
1635 index = outarg.offset >> PAGE_CACHE_SHIFT;
1636 offset = outarg.offset & ~PAGE_CACHE_MASK;
1637 file_size = i_size_read(inode);
1638 end = outarg.offset + outarg.size;
1639 if (end > file_size) {
1641 fuse_write_update_size(inode, file_size);
1647 unsigned int this_num;
1650 page = find_or_create_page(mapping, index,
1651 mapping_gfp_mask(mapping));
1655 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1656 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1657 if (!err && offset == 0 &&
1658 (this_num == PAGE_CACHE_SIZE || file_size == end))
1659 SetPageUptodate(page);
1661 page_cache_release(page);
1676 up_read(&fc->killsb);
1678 fuse_copy_finish(cs);
1682 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1684 release_pages(req->pages, req->num_pages, false);
1687 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1688 struct fuse_notify_retrieve_out *outarg)
1691 struct address_space *mapping = inode->i_mapping;
1692 struct fuse_req *req;
1696 unsigned int offset;
1697 size_t total_len = 0;
1700 offset = outarg->offset & ~PAGE_CACHE_MASK;
1701 file_size = i_size_read(inode);
1704 if (outarg->offset > file_size)
1706 else if (outarg->offset + num > file_size)
1707 num = file_size - outarg->offset;
1709 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1710 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1712 req = fuse_get_req(fc, num_pages);
1714 return PTR_ERR(req);
1716 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1717 req->in.h.nodeid = outarg->nodeid;
1718 req->in.numargs = 2;
1719 req->in.argpages = 1;
1720 req->page_descs[0].offset = offset;
1721 req->end = fuse_retrieve_end;
1723 index = outarg->offset >> PAGE_CACHE_SHIFT;
1725 while (num && req->num_pages < num_pages) {
1727 unsigned int this_num;
1729 page = find_get_page(mapping, index);
1733 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1734 req->pages[req->num_pages] = page;
1735 req->page_descs[req->num_pages].length = this_num;
1740 total_len += this_num;
1743 req->misc.retrieve_in.offset = outarg->offset;
1744 req->misc.retrieve_in.size = total_len;
1745 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1746 req->in.args[0].value = &req->misc.retrieve_in;
1747 req->in.args[1].size = total_len;
1749 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1751 fuse_retrieve_end(fc, req);
1756 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1757 struct fuse_copy_state *cs)
1759 struct fuse_notify_retrieve_out outarg;
1760 struct inode *inode;
1764 if (size != sizeof(outarg))
1767 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1771 fuse_copy_finish(cs);
1773 down_read(&fc->killsb);
1776 u64 nodeid = outarg.nodeid;
1778 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1780 err = fuse_retrieve(fc, inode, &outarg);
1784 up_read(&fc->killsb);
1789 fuse_copy_finish(cs);
1793 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1794 unsigned int size, struct fuse_copy_state *cs)
1796 /* Don't try to move pages (yet) */
1800 case FUSE_NOTIFY_POLL:
1801 return fuse_notify_poll(fc, size, cs);
1803 case FUSE_NOTIFY_INVAL_INODE:
1804 return fuse_notify_inval_inode(fc, size, cs);
1806 case FUSE_NOTIFY_INVAL_ENTRY:
1807 return fuse_notify_inval_entry(fc, size, cs);
1809 case FUSE_NOTIFY_STORE:
1810 return fuse_notify_store(fc, size, cs);
1812 case FUSE_NOTIFY_RETRIEVE:
1813 return fuse_notify_retrieve(fc, size, cs);
1815 case FUSE_NOTIFY_DELETE:
1816 return fuse_notify_delete(fc, size, cs);
1819 fuse_copy_finish(cs);
1824 /* Look up request on processing list by unique ID */
1825 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1827 struct fuse_req *req;
1829 list_for_each_entry(req, &fc->processing, list) {
1830 if (req->in.h.unique == unique || req->intr_unique == unique)
1836 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1839 unsigned reqsize = sizeof(struct fuse_out_header);
1842 return nbytes != reqsize ? -EINVAL : 0;
1844 reqsize += len_args(out->numargs, out->args);
1846 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1848 else if (reqsize > nbytes) {
1849 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1850 unsigned diffsize = reqsize - nbytes;
1851 if (diffsize > lastarg->size)
1853 lastarg->size -= diffsize;
1855 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1860 * Write a single reply to a request. First the header is copied from
1861 * the write buffer. The request is then searched on the processing
1862 * list by the unique ID found in the header. If found, then remove
1863 * it from the list and copy the rest of the buffer to the request.
1864 * The request is finished by calling request_end()
1866 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1867 struct fuse_copy_state *cs, size_t nbytes)
1870 struct fuse_req *req;
1871 struct fuse_out_header oh;
1873 if (nbytes < sizeof(struct fuse_out_header))
1876 err = fuse_copy_one(cs, &oh, sizeof(oh));
1881 if (oh.len != nbytes)
1885 * Zero oh.unique indicates unsolicited notification message
1886 * and error contains notification code.
1889 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1890 return err ? err : nbytes;
1894 if (oh.error <= -1000 || oh.error > 0)
1897 spin_lock(&fc->lock);
1902 req = request_find(fc, oh.unique);
1906 /* Is it an interrupt reply? */
1907 if (req->intr_unique == oh.unique) {
1909 if (nbytes != sizeof(struct fuse_out_header))
1912 if (oh.error == -ENOSYS)
1913 fc->no_interrupt = 1;
1914 else if (oh.error == -EAGAIN)
1915 queue_interrupt(&fc->iq, req);
1917 spin_unlock(&fc->lock);
1918 fuse_copy_finish(cs);
1922 clear_bit(FR_SENT, &req->flags);
1923 list_move(&req->list, &fc->io);
1925 set_bit(FR_LOCKED, &req->flags);
1927 if (!req->out.page_replace)
1929 spin_unlock(&fc->lock);
1931 err = copy_out_args(cs, &req->out, nbytes);
1932 fuse_copy_finish(cs);
1934 spin_lock(&fc->lock);
1935 clear_bit(FR_LOCKED, &req->flags);
1939 req->out.h.error = -EIO;
1940 request_end(fc, req);
1942 return err ? err : nbytes;
1945 spin_unlock(&fc->lock);
1947 fuse_copy_finish(cs);
1951 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1953 struct fuse_copy_state cs;
1954 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1958 if (!iter_is_iovec(from))
1961 fuse_copy_init(&cs, 0, from);
1963 return fuse_dev_do_write(fc, &cs, iov_iter_count(from));
1966 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1967 struct file *out, loff_t *ppos,
1968 size_t len, unsigned int flags)
1972 struct pipe_buffer *bufs;
1973 struct fuse_copy_state cs;
1974 struct fuse_conn *fc;
1978 fc = fuse_get_conn(out);
1982 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1989 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1990 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
2000 struct pipe_buffer *ibuf;
2001 struct pipe_buffer *obuf;
2003 BUG_ON(nbuf >= pipe->buffers);
2004 BUG_ON(!pipe->nrbufs);
2005 ibuf = &pipe->bufs[pipe->curbuf];
2008 if (rem >= ibuf->len) {
2011 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2014 ibuf->ops->get(pipe, ibuf);
2016 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2018 ibuf->offset += obuf->len;
2019 ibuf->len -= obuf->len;
2026 fuse_copy_init(&cs, 0, NULL);
2031 if (flags & SPLICE_F_MOVE)
2034 ret = fuse_dev_do_write(fc, &cs, len);
2036 for (idx = 0; idx < nbuf; idx++) {
2037 struct pipe_buffer *buf = &bufs[idx];
2038 buf->ops->release(pipe, buf);
2045 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2047 unsigned mask = POLLOUT | POLLWRNORM;
2048 struct fuse_iqueue *fiq;
2049 struct fuse_conn *fc = fuse_get_conn(file);
2054 poll_wait(file, &fiq->waitq, wait);
2056 spin_lock(&fc->lock);
2057 if (!fiq->connected)
2059 else if (request_pending(fiq))
2060 mask |= POLLIN | POLLRDNORM;
2061 spin_unlock(&fc->lock);
2067 * Abort all requests on the given list (pending or processing)
2069 * This function releases and reacquires fc->lock
2071 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2072 __releases(fc->lock)
2073 __acquires(fc->lock)
2075 while (!list_empty(head)) {
2076 struct fuse_req *req;
2077 req = list_entry(head->next, struct fuse_req, list);
2078 req->out.h.error = -ECONNABORTED;
2079 clear_bit(FR_PENDING, &req->flags);
2080 clear_bit(FR_SENT, &req->flags);
2081 request_end(fc, req);
2082 spin_lock(&fc->lock);
2086 static void end_polls(struct fuse_conn *fc)
2090 p = rb_first(&fc->polled_files);
2093 struct fuse_file *ff;
2094 ff = rb_entry(p, struct fuse_file, polled_node);
2095 wake_up_interruptible_all(&ff->poll_wait);
2102 * Abort all requests.
2104 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2107 * The same effect is usually achievable through killing the filesystem daemon
2108 * and all users of the filesystem. The exception is the combination of an
2109 * asynchronous request and the tricky deadlock (see
2110 * Documentation/filesystems/fuse.txt).
2112 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2113 * requests, they should be finished off immediately. Locked requests will be
2114 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2115 * requests. It is possible that some request will finish before we can. This
2116 * is OK, the request will in that case be removed from the list before we touch
2119 void fuse_abort_conn(struct fuse_conn *fc)
2121 struct fuse_iqueue *fiq = &fc->iq;
2123 spin_lock(&fc->lock);
2124 if (fc->connected) {
2125 struct fuse_req *req, *next;
2132 fuse_set_initialized(fc);
2133 list_for_each_entry_safe(req, next, &fc->io, list) {
2134 req->out.h.error = -ECONNABORTED;
2135 spin_lock(&req->waitq.lock);
2136 set_bit(FR_ABORTED, &req->flags);
2137 if (!test_bit(FR_LOCKED, &req->flags))
2138 list_move(&req->list, &to_end1);
2139 spin_unlock(&req->waitq.lock);
2141 fc->max_background = UINT_MAX;
2143 list_splice_init(&fiq->pending, &to_end2);
2144 list_splice_init(&fc->processing, &to_end2);
2145 while (!list_empty(&to_end1)) {
2146 req = list_first_entry(&to_end1, struct fuse_req, list);
2147 __fuse_get_request(req);
2148 request_end(fc, req);
2149 spin_lock(&fc->lock);
2151 end_requests(fc, &to_end2);
2152 while (forget_pending(fiq))
2153 kfree(dequeue_forget(fiq, 1, NULL));
2155 wake_up_all(&fiq->waitq);
2156 wake_up_all(&fc->blocked_waitq);
2157 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2159 spin_unlock(&fc->lock);
2161 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2163 int fuse_dev_release(struct inode *inode, struct file *file)
2165 struct fuse_conn *fc = fuse_get_conn(file);
2167 WARN_ON(!list_empty(&fc->io));
2168 WARN_ON(fc->iq.fasync != NULL);
2169 fuse_abort_conn(fc);
2175 EXPORT_SYMBOL_GPL(fuse_dev_release);
2177 static int fuse_dev_fasync(int fd, struct file *file, int on)
2179 struct fuse_conn *fc = fuse_get_conn(file);
2183 /* No locking - fasync_helper does its own locking */
2184 return fasync_helper(fd, file, on, &fc->iq.fasync);
2187 const struct file_operations fuse_dev_operations = {
2188 .owner = THIS_MODULE,
2189 .open = fuse_dev_open,
2190 .llseek = no_llseek,
2191 .read_iter = fuse_dev_read,
2192 .splice_read = fuse_dev_splice_read,
2193 .write_iter = fuse_dev_write,
2194 .splice_write = fuse_dev_splice_write,
2195 .poll = fuse_dev_poll,
2196 .release = fuse_dev_release,
2197 .fasync = fuse_dev_fasync,
2199 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2201 static struct miscdevice fuse_miscdevice = {
2202 .minor = FUSE_MINOR,
2204 .fops = &fuse_dev_operations,
2207 int __init fuse_dev_init(void)
2210 fuse_req_cachep = kmem_cache_create("fuse_request",
2211 sizeof(struct fuse_req),
2213 if (!fuse_req_cachep)
2216 err = misc_register(&fuse_miscdevice);
2218 goto out_cache_clean;
2223 kmem_cache_destroy(fuse_req_cachep);
2228 void fuse_dev_cleanup(void)
2230 misc_deregister(&fuse_miscdevice);
2231 kmem_cache_destroy(fuse_req_cachep);