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_dev *fuse_get_dev(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 ACCESS_ONCE(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 void __fuse_get_request(struct fuse_req *req)
104 atomic_inc(&req->count);
107 /* Must be called with > 1 refcount */
108 static void __fuse_put_request(struct fuse_req *req)
110 BUG_ON(atomic_read(&req->count) < 2);
111 atomic_dec(&req->count);
114 static void fuse_req_init_context(struct fuse_req *req)
116 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
117 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
118 req->in.h.pid = current->pid;
121 void fuse_set_initialized(struct fuse_conn *fc)
123 /* Make sure stores before this are seen on another CPU */
128 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
130 return !fc->initialized || (for_background && fc->blocked);
133 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
136 struct fuse_req *req;
138 atomic_inc(&fc->num_waiting);
140 if (fuse_block_alloc(fc, for_background)) {
142 if (wait_event_killable_exclusive(fc->blocked_waitq,
143 !fuse_block_alloc(fc, for_background)))
146 /* Matches smp_wmb() in fuse_set_initialized() */
157 req = fuse_request_alloc(npages);
161 wake_up(&fc->blocked_waitq);
165 fuse_req_init_context(req);
166 __set_bit(FR_WAITING, &req->flags);
168 __set_bit(FR_BACKGROUND, &req->flags);
173 atomic_dec(&fc->num_waiting);
177 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
179 return __fuse_get_req(fc, npages, false);
181 EXPORT_SYMBOL_GPL(fuse_get_req);
183 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
186 return __fuse_get_req(fc, npages, true);
188 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
191 * Return request in fuse_file->reserved_req. However that may
192 * currently be in use. If that is the case, wait for it to become
195 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
198 struct fuse_req *req = NULL;
199 struct fuse_file *ff = file->private_data;
202 wait_event(fc->reserved_req_waitq, ff->reserved_req);
203 spin_lock(&fc->lock);
204 if (ff->reserved_req) {
205 req = ff->reserved_req;
206 ff->reserved_req = NULL;
207 req->stolen_file = get_file(file);
209 spin_unlock(&fc->lock);
216 * Put stolen request back into fuse_file->reserved_req
218 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
220 struct file *file = req->stolen_file;
221 struct fuse_file *ff = file->private_data;
223 spin_lock(&fc->lock);
224 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
225 BUG_ON(ff->reserved_req);
226 ff->reserved_req = req;
227 wake_up_all(&fc->reserved_req_waitq);
228 spin_unlock(&fc->lock);
233 * Gets a requests for a file operation, always succeeds
235 * This is used for sending the FLUSH request, which must get to
236 * userspace, due to POSIX locks which may need to be unlocked.
238 * If allocation fails due to OOM, use the reserved request in
241 * This is very unlikely to deadlock accidentally, since the
242 * filesystem should not have it's own file open. If deadlock is
243 * intentional, it can still be broken by "aborting" the filesystem.
245 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
248 struct fuse_req *req;
250 atomic_inc(&fc->num_waiting);
251 wait_event(fc->blocked_waitq, fc->initialized);
252 /* Matches smp_wmb() in fuse_set_initialized() */
254 req = fuse_request_alloc(0);
256 req = get_reserved_req(fc, file);
258 fuse_req_init_context(req);
259 __set_bit(FR_WAITING, &req->flags);
260 __clear_bit(FR_BACKGROUND, &req->flags);
264 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
266 if (atomic_dec_and_test(&req->count)) {
267 if (test_bit(FR_BACKGROUND, &req->flags)) {
269 * We get here in the unlikely case that a background
270 * request was allocated but not sent
272 spin_lock(&fc->lock);
274 wake_up(&fc->blocked_waitq);
275 spin_unlock(&fc->lock);
278 if (test_bit(FR_WAITING, &req->flags)) {
279 __clear_bit(FR_WAITING, &req->flags);
280 atomic_dec(&fc->num_waiting);
283 if (req->stolen_file)
284 put_reserved_req(fc, req);
286 fuse_request_free(req);
289 EXPORT_SYMBOL_GPL(fuse_put_request);
291 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
296 for (i = 0; i < numargs; i++)
297 nbytes += args[i].size;
302 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
304 return ++fiq->reqctr;
307 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
309 req->in.h.len = sizeof(struct fuse_in_header) +
310 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
311 list_add_tail(&req->list, &fiq->pending);
312 wake_up_locked(&fiq->waitq);
313 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
316 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
317 u64 nodeid, u64 nlookup)
319 struct fuse_iqueue *fiq = &fc->iq;
321 forget->forget_one.nodeid = nodeid;
322 forget->forget_one.nlookup = nlookup;
324 spin_lock(&fiq->waitq.lock);
325 if (fiq->connected) {
326 fiq->forget_list_tail->next = forget;
327 fiq->forget_list_tail = forget;
328 wake_up_locked(&fiq->waitq);
329 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
333 spin_unlock(&fiq->waitq.lock);
336 static void flush_bg_queue(struct fuse_conn *fc)
338 while (fc->active_background < fc->max_background &&
339 !list_empty(&fc->bg_queue)) {
340 struct fuse_req *req;
341 struct fuse_iqueue *fiq = &fc->iq;
343 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
344 list_del(&req->list);
345 fc->active_background++;
346 spin_lock(&fiq->waitq.lock);
347 req->in.h.unique = fuse_get_unique(fiq);
348 queue_request(fiq, req);
349 spin_unlock(&fiq->waitq.lock);
354 * This function is called when a request is finished. Either a reply
355 * has arrived or it was aborted (and not yet sent) or some error
356 * occurred during communication with userspace, or the device file
357 * was closed. The requester thread is woken up (if still waiting),
358 * the 'end' callback is called if given, else the reference to the
359 * request is released
361 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
363 struct fuse_iqueue *fiq = &fc->iq;
365 if (test_and_set_bit(FR_FINISHED, &req->flags))
368 spin_lock(&fiq->waitq.lock);
369 list_del_init(&req->intr_entry);
370 spin_unlock(&fiq->waitq.lock);
371 WARN_ON(test_bit(FR_PENDING, &req->flags));
372 WARN_ON(test_bit(FR_SENT, &req->flags));
373 if (test_bit(FR_BACKGROUND, &req->flags)) {
374 spin_lock(&fc->lock);
375 clear_bit(FR_BACKGROUND, &req->flags);
376 if (fc->num_background == fc->max_background)
379 /* Wake up next waiter, if any */
380 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
381 wake_up(&fc->blocked_waitq);
383 if (fc->num_background == fc->congestion_threshold &&
384 fc->connected && fc->bdi_initialized) {
385 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
386 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
388 fc->num_background--;
389 fc->active_background--;
391 spin_unlock(&fc->lock);
393 wake_up(&req->waitq);
396 fuse_put_request(fc, req);
399 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
401 spin_lock(&fiq->waitq.lock);
402 if (list_empty(&req->intr_entry)) {
403 list_add_tail(&req->intr_entry, &fiq->interrupts);
404 wake_up_locked(&fiq->waitq);
406 spin_unlock(&fiq->waitq.lock);
407 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
410 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
412 struct fuse_iqueue *fiq = &fc->iq;
415 if (!fc->no_interrupt) {
416 /* Any signal may interrupt this */
417 err = wait_event_interruptible(req->waitq,
418 test_bit(FR_FINISHED, &req->flags));
422 set_bit(FR_INTERRUPTED, &req->flags);
423 /* matches barrier in fuse_dev_do_read() */
424 smp_mb__after_atomic();
425 if (test_bit(FR_SENT, &req->flags))
426 queue_interrupt(fiq, req);
429 if (!test_bit(FR_FORCE, &req->flags)) {
430 /* Only fatal signals may interrupt this */
431 err = wait_event_killable(req->waitq,
432 test_bit(FR_FINISHED, &req->flags));
436 spin_lock(&fiq->waitq.lock);
437 /* Request is not yet in userspace, bail out */
438 if (test_bit(FR_PENDING, &req->flags)) {
439 list_del(&req->list);
440 spin_unlock(&fiq->waitq.lock);
441 __fuse_put_request(req);
442 req->out.h.error = -EINTR;
445 spin_unlock(&fiq->waitq.lock);
449 * Either request is already in userspace, or it was forced.
452 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
455 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
457 struct fuse_iqueue *fiq = &fc->iq;
459 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
460 spin_lock(&fiq->waitq.lock);
461 if (!fiq->connected) {
462 spin_unlock(&fiq->waitq.lock);
463 req->out.h.error = -ENOTCONN;
465 req->in.h.unique = fuse_get_unique(fiq);
466 queue_request(fiq, req);
467 /* acquire extra reference, since request is still needed
468 after request_end() */
469 __fuse_get_request(req);
470 spin_unlock(&fiq->waitq.lock);
472 request_wait_answer(fc, req);
473 /* Pairs with smp_wmb() in request_end() */
478 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
480 __set_bit(FR_ISREPLY, &req->flags);
481 if (!test_bit(FR_WAITING, &req->flags)) {
482 __set_bit(FR_WAITING, &req->flags);
483 atomic_inc(&fc->num_waiting);
485 __fuse_request_send(fc, req);
487 EXPORT_SYMBOL_GPL(fuse_request_send);
489 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
491 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
492 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
495 switch (args->in.h.opcode) {
502 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
506 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
510 if (fc->minor < 12) {
511 switch (args->in.h.opcode) {
513 args->in.args[0].size = sizeof(struct fuse_open_in);
516 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
522 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
524 struct fuse_req *req;
527 req = fuse_get_req(fc, 0);
531 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
532 fuse_adjust_compat(fc, args);
534 req->in.h.opcode = args->in.h.opcode;
535 req->in.h.nodeid = args->in.h.nodeid;
536 req->in.numargs = args->in.numargs;
537 memcpy(req->in.args, args->in.args,
538 args->in.numargs * sizeof(struct fuse_in_arg));
539 req->out.argvar = args->out.argvar;
540 req->out.numargs = args->out.numargs;
541 memcpy(req->out.args, args->out.args,
542 args->out.numargs * sizeof(struct fuse_arg));
543 fuse_request_send(fc, req);
544 ret = req->out.h.error;
545 if (!ret && args->out.argvar) {
546 BUG_ON(args->out.numargs != 1);
547 ret = req->out.args[0].size;
549 fuse_put_request(fc, req);
555 * Called under fc->lock
557 * fc->connected must have been checked previously
559 void fuse_request_send_background_locked(struct fuse_conn *fc,
560 struct fuse_req *req)
562 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
563 if (!test_bit(FR_WAITING, &req->flags)) {
564 __set_bit(FR_WAITING, &req->flags);
565 atomic_inc(&fc->num_waiting);
567 __set_bit(FR_ISREPLY, &req->flags);
568 fc->num_background++;
569 if (fc->num_background == fc->max_background)
571 if (fc->num_background == fc->congestion_threshold &&
572 fc->bdi_initialized) {
573 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
574 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
576 list_add_tail(&req->list, &fc->bg_queue);
580 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
583 spin_lock(&fc->lock);
585 fuse_request_send_background_locked(fc, req);
586 spin_unlock(&fc->lock);
588 spin_unlock(&fc->lock);
589 req->out.h.error = -ENOTCONN;
591 fuse_put_request(fc, req);
594 EXPORT_SYMBOL_GPL(fuse_request_send_background);
596 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
597 struct fuse_req *req, u64 unique)
600 struct fuse_iqueue *fiq = &fc->iq;
602 __clear_bit(FR_ISREPLY, &req->flags);
603 req->in.h.unique = unique;
604 spin_lock(&fiq->waitq.lock);
605 if (fiq->connected) {
606 queue_request(fiq, req);
609 spin_unlock(&fiq->waitq.lock);
614 void fuse_force_forget(struct file *file, u64 nodeid)
616 struct inode *inode = file_inode(file);
617 struct fuse_conn *fc = get_fuse_conn(inode);
618 struct fuse_req *req;
619 struct fuse_forget_in inarg;
621 memset(&inarg, 0, sizeof(inarg));
623 req = fuse_get_req_nofail_nopages(fc, file);
624 req->in.h.opcode = FUSE_FORGET;
625 req->in.h.nodeid = nodeid;
627 req->in.args[0].size = sizeof(inarg);
628 req->in.args[0].value = &inarg;
629 __clear_bit(FR_ISREPLY, &req->flags);
630 __fuse_request_send(fc, req);
632 fuse_put_request(fc, req);
636 * Lock the request. Up to the next unlock_request() there mustn't be
637 * anything that could cause a page-fault. If the request was already
640 static int lock_request(struct fuse_req *req)
644 spin_lock(&req->waitq.lock);
645 if (test_bit(FR_ABORTED, &req->flags))
648 set_bit(FR_LOCKED, &req->flags);
649 spin_unlock(&req->waitq.lock);
655 * Unlock request. If it was aborted while locked, caller is responsible
656 * for unlocking and ending the request.
658 static int unlock_request(struct fuse_req *req)
662 spin_lock(&req->waitq.lock);
663 if (test_bit(FR_ABORTED, &req->flags))
666 clear_bit(FR_LOCKED, &req->flags);
667 spin_unlock(&req->waitq.lock);
672 struct fuse_copy_state {
674 struct fuse_req *req;
675 struct iov_iter *iter;
676 struct pipe_buffer *pipebufs;
677 struct pipe_buffer *currbuf;
678 struct pipe_inode_info *pipe;
679 unsigned long nr_segs;
683 unsigned move_pages:1;
686 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
687 struct iov_iter *iter)
689 memset(cs, 0, sizeof(*cs));
694 /* Unmap and put previous page of userspace buffer */
695 static void fuse_copy_finish(struct fuse_copy_state *cs)
698 struct pipe_buffer *buf = cs->currbuf;
701 buf->len = PAGE_SIZE - cs->len;
705 flush_dcache_page(cs->pg);
706 set_page_dirty_lock(cs->pg);
714 * Get another pagefull of userspace buffer, and map it to kernel
715 * address space, and lock request
717 static int fuse_copy_fill(struct fuse_copy_state *cs)
722 err = unlock_request(cs->req);
726 fuse_copy_finish(cs);
728 struct pipe_buffer *buf = cs->pipebufs;
731 err = pipe_buf_confirm(cs->pipe, buf);
735 BUG_ON(!cs->nr_segs);
738 cs->offset = buf->offset;
743 if (cs->nr_segs == cs->pipe->buffers)
746 page = alloc_page(GFP_HIGHUSER);
763 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
771 iov_iter_advance(cs->iter, err);
774 return lock_request(cs->req);
777 /* Do as much copy to/from userspace buffer as we can */
778 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
780 unsigned ncpy = min(*size, cs->len);
782 void *pgaddr = kmap_atomic(cs->pg);
783 void *buf = pgaddr + cs->offset;
786 memcpy(buf, *val, ncpy);
788 memcpy(*val, buf, ncpy);
790 kunmap_atomic(pgaddr);
799 static int fuse_check_page(struct page *page)
801 if (page_mapcount(page) ||
802 page->mapping != NULL ||
803 page_count(page) != 1 ||
804 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
811 printk(KERN_WARNING "fuse: trying to steal weird page\n");
812 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);
818 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
821 struct page *oldpage = *pagep;
822 struct page *newpage;
823 struct pipe_buffer *buf = cs->pipebufs;
825 err = unlock_request(cs->req);
829 fuse_copy_finish(cs);
831 err = pipe_buf_confirm(cs->pipe, buf);
835 BUG_ON(!cs->nr_segs);
841 if (cs->len != PAGE_SIZE)
844 if (buf->ops->steal(cs->pipe, buf) != 0)
849 if (!PageUptodate(newpage))
850 SetPageUptodate(newpage);
852 ClearPageMappedToDisk(newpage);
854 if (fuse_check_page(newpage) != 0)
855 goto out_fallback_unlock;
858 * This is a new and locked page, it shouldn't be mapped or
859 * have any special flags on it
861 if (WARN_ON(page_mapped(oldpage)))
862 goto out_fallback_unlock;
863 if (WARN_ON(page_has_private(oldpage)))
864 goto out_fallback_unlock;
865 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
866 goto out_fallback_unlock;
867 if (WARN_ON(PageMlocked(oldpage)))
868 goto out_fallback_unlock;
870 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
872 unlock_page(newpage);
878 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
879 lru_cache_add_file(newpage);
882 spin_lock(&cs->req->waitq.lock);
883 if (test_bit(FR_ABORTED, &cs->req->flags))
887 spin_unlock(&cs->req->waitq.lock);
890 unlock_page(newpage);
895 unlock_page(oldpage);
902 unlock_page(newpage);
905 cs->offset = buf->offset;
907 err = lock_request(cs->req);
914 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
915 unsigned offset, unsigned count)
917 struct pipe_buffer *buf;
920 if (cs->nr_segs == cs->pipe->buffers)
923 err = unlock_request(cs->req);
927 fuse_copy_finish(cs);
932 buf->offset = offset;
943 * Copy a page in the request to/from the userspace buffer. Must be
946 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
947 unsigned offset, unsigned count, int zeroing)
950 struct page *page = *pagep;
952 if (page && zeroing && count < PAGE_SIZE)
953 clear_highpage(page);
956 if (cs->write && cs->pipebufs && page) {
957 return fuse_ref_page(cs, page, offset, count);
958 } else if (!cs->len) {
959 if (cs->move_pages && page &&
960 offset == 0 && count == PAGE_SIZE) {
961 err = fuse_try_move_page(cs, pagep);
965 err = fuse_copy_fill(cs);
971 void *mapaddr = kmap_atomic(page);
972 void *buf = mapaddr + offset;
973 offset += fuse_copy_do(cs, &buf, &count);
974 kunmap_atomic(mapaddr);
976 offset += fuse_copy_do(cs, NULL, &count);
978 if (page && !cs->write)
979 flush_dcache_page(page);
983 /* Copy pages in the request to/from userspace buffer */
984 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
988 struct fuse_req *req = cs->req;
990 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
992 unsigned offset = req->page_descs[i].offset;
993 unsigned count = min(nbytes, req->page_descs[i].length);
995 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1005 /* Copy a single argument in the request to/from userspace buffer */
1006 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1010 int err = fuse_copy_fill(cs);
1014 fuse_copy_do(cs, &val, &size);
1019 /* Copy request arguments to/from userspace buffer */
1020 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1021 unsigned argpages, struct fuse_arg *args,
1027 for (i = 0; !err && i < numargs; i++) {
1028 struct fuse_arg *arg = &args[i];
1029 if (i == numargs - 1 && argpages)
1030 err = fuse_copy_pages(cs, arg->size, zeroing);
1032 err = fuse_copy_one(cs, arg->value, arg->size);
1037 static int forget_pending(struct fuse_iqueue *fiq)
1039 return fiq->forget_list_head.next != NULL;
1042 static int request_pending(struct fuse_iqueue *fiq)
1044 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1045 forget_pending(fiq);
1049 * Transfer an interrupt request to userspace
1051 * Unlike other requests this is assembled on demand, without a need
1052 * to allocate a separate fuse_req structure.
1054 * Called with fiq->waitq.lock held, releases it
1056 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1057 struct fuse_copy_state *cs,
1058 size_t nbytes, struct fuse_req *req)
1059 __releases(fiq->waitq.lock)
1061 struct fuse_in_header ih;
1062 struct fuse_interrupt_in arg;
1063 unsigned reqsize = sizeof(ih) + sizeof(arg);
1066 list_del_init(&req->intr_entry);
1067 req->intr_unique = fuse_get_unique(fiq);
1068 memset(&ih, 0, sizeof(ih));
1069 memset(&arg, 0, sizeof(arg));
1071 ih.opcode = FUSE_INTERRUPT;
1072 ih.unique = req->intr_unique;
1073 arg.unique = req->in.h.unique;
1075 spin_unlock(&fiq->waitq.lock);
1076 if (nbytes < reqsize)
1079 err = fuse_copy_one(cs, &ih, sizeof(ih));
1081 err = fuse_copy_one(cs, &arg, sizeof(arg));
1082 fuse_copy_finish(cs);
1084 return err ? err : reqsize;
1087 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1091 struct fuse_forget_link *head = fiq->forget_list_head.next;
1092 struct fuse_forget_link **newhead = &head;
1095 for (count = 0; *newhead != NULL && count < max; count++)
1096 newhead = &(*newhead)->next;
1098 fiq->forget_list_head.next = *newhead;
1100 if (fiq->forget_list_head.next == NULL)
1101 fiq->forget_list_tail = &fiq->forget_list_head;
1109 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1110 struct fuse_copy_state *cs,
1112 __releases(fiq->waitq.lock)
1115 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1116 struct fuse_forget_in arg = {
1117 .nlookup = forget->forget_one.nlookup,
1119 struct fuse_in_header ih = {
1120 .opcode = FUSE_FORGET,
1121 .nodeid = forget->forget_one.nodeid,
1122 .unique = fuse_get_unique(fiq),
1123 .len = sizeof(ih) + sizeof(arg),
1126 spin_unlock(&fiq->waitq.lock);
1128 if (nbytes < ih.len)
1131 err = fuse_copy_one(cs, &ih, sizeof(ih));
1133 err = fuse_copy_one(cs, &arg, sizeof(arg));
1134 fuse_copy_finish(cs);
1142 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1143 struct fuse_copy_state *cs, size_t nbytes)
1144 __releases(fiq->waitq.lock)
1147 unsigned max_forgets;
1149 struct fuse_forget_link *head;
1150 struct fuse_batch_forget_in arg = { .count = 0 };
1151 struct fuse_in_header ih = {
1152 .opcode = FUSE_BATCH_FORGET,
1153 .unique = fuse_get_unique(fiq),
1154 .len = sizeof(ih) + sizeof(arg),
1157 if (nbytes < ih.len) {
1158 spin_unlock(&fiq->waitq.lock);
1162 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1163 head = dequeue_forget(fiq, max_forgets, &count);
1164 spin_unlock(&fiq->waitq.lock);
1167 ih.len += count * sizeof(struct fuse_forget_one);
1168 err = fuse_copy_one(cs, &ih, sizeof(ih));
1170 err = fuse_copy_one(cs, &arg, sizeof(arg));
1173 struct fuse_forget_link *forget = head;
1176 err = fuse_copy_one(cs, &forget->forget_one,
1177 sizeof(forget->forget_one));
1179 head = forget->next;
1183 fuse_copy_finish(cs);
1191 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1192 struct fuse_copy_state *cs,
1194 __releases(fiq->waitq.lock)
1196 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1197 return fuse_read_single_forget(fiq, cs, nbytes);
1199 return fuse_read_batch_forget(fiq, cs, nbytes);
1203 * Read a single request into the userspace filesystem's buffer. This
1204 * function waits until a request is available, then removes it from
1205 * the pending list and copies request data to userspace buffer. If
1206 * no reply is needed (FORGET) or request has been aborted or there
1207 * was an error during the copying then it's finished by calling
1208 * request_end(). Otherwise add it to the processing list, and set
1211 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1212 struct fuse_copy_state *cs, size_t nbytes)
1215 struct fuse_conn *fc = fud->fc;
1216 struct fuse_iqueue *fiq = &fc->iq;
1217 struct fuse_pqueue *fpq = &fud->pq;
1218 struct fuse_req *req;
1223 spin_lock(&fiq->waitq.lock);
1225 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1226 !request_pending(fiq))
1229 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1230 !fiq->connected || request_pending(fiq));
1235 if (!fiq->connected)
1238 if (!list_empty(&fiq->interrupts)) {
1239 req = list_entry(fiq->interrupts.next, struct fuse_req,
1241 return fuse_read_interrupt(fiq, cs, nbytes, req);
1244 if (forget_pending(fiq)) {
1245 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1246 return fuse_read_forget(fc, fiq, cs, nbytes);
1248 if (fiq->forget_batch <= -8)
1249 fiq->forget_batch = 16;
1252 req = list_entry(fiq->pending.next, struct fuse_req, list);
1253 clear_bit(FR_PENDING, &req->flags);
1254 list_del_init(&req->list);
1255 spin_unlock(&fiq->waitq.lock);
1258 reqsize = in->h.len;
1259 /* If request is too large, reply with an error and restart the read */
1260 if (nbytes < reqsize) {
1261 req->out.h.error = -EIO;
1262 /* SETXATTR is special, since it may contain too large data */
1263 if (in->h.opcode == FUSE_SETXATTR)
1264 req->out.h.error = -E2BIG;
1265 request_end(fc, req);
1268 spin_lock(&fpq->lock);
1269 list_add(&req->list, &fpq->io);
1270 spin_unlock(&fpq->lock);
1272 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1274 err = fuse_copy_args(cs, in->numargs, in->argpages,
1275 (struct fuse_arg *) in->args, 0);
1276 fuse_copy_finish(cs);
1277 spin_lock(&fpq->lock);
1278 clear_bit(FR_LOCKED, &req->flags);
1279 if (!fpq->connected) {
1284 req->out.h.error = -EIO;
1287 if (!test_bit(FR_ISREPLY, &req->flags)) {
1291 list_move_tail(&req->list, &fpq->processing);
1292 spin_unlock(&fpq->lock);
1293 set_bit(FR_SENT, &req->flags);
1294 /* matches barrier in request_wait_answer() */
1295 smp_mb__after_atomic();
1296 if (test_bit(FR_INTERRUPTED, &req->flags))
1297 queue_interrupt(fiq, req);
1302 if (!test_bit(FR_PRIVATE, &req->flags))
1303 list_del_init(&req->list);
1304 spin_unlock(&fpq->lock);
1305 request_end(fc, req);
1309 spin_unlock(&fiq->waitq.lock);
1313 static int fuse_dev_open(struct inode *inode, struct file *file)
1316 * The fuse device's file's private_data is used to hold
1317 * the fuse_conn(ection) when it is mounted, and is used to
1318 * keep track of whether the file has been mounted already.
1320 file->private_data = NULL;
1324 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1326 struct fuse_copy_state cs;
1327 struct file *file = iocb->ki_filp;
1328 struct fuse_dev *fud = fuse_get_dev(file);
1333 if (!iter_is_iovec(to))
1336 fuse_copy_init(&cs, 1, to);
1338 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1341 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1342 struct pipe_inode_info *pipe,
1343 size_t len, unsigned int flags)
1347 struct pipe_buffer *bufs;
1348 struct fuse_copy_state cs;
1349 struct fuse_dev *fud = fuse_get_dev(in);
1354 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1358 fuse_copy_init(&cs, 1, NULL);
1361 ret = fuse_dev_do_read(fud, in, &cs, len);
1365 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1370 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1372 * Need to be careful about this. Having buf->ops in module
1373 * code can Oops if the buffer persists after module unload.
1375 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1376 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1377 if (unlikely(ret < 0))
1383 for (; page_nr < cs.nr_segs; page_nr++)
1384 put_page(bufs[page_nr].page);
1390 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1391 struct fuse_copy_state *cs)
1393 struct fuse_notify_poll_wakeup_out outarg;
1396 if (size != sizeof(outarg))
1399 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1403 fuse_copy_finish(cs);
1404 return fuse_notify_poll_wakeup(fc, &outarg);
1407 fuse_copy_finish(cs);
1411 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1412 struct fuse_copy_state *cs)
1414 struct fuse_notify_inval_inode_out outarg;
1417 if (size != sizeof(outarg))
1420 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1423 fuse_copy_finish(cs);
1425 down_read(&fc->killsb);
1428 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1429 outarg.off, outarg.len);
1431 up_read(&fc->killsb);
1435 fuse_copy_finish(cs);
1439 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1440 struct fuse_copy_state *cs)
1442 struct fuse_notify_inval_entry_out outarg;
1447 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1452 if (size < sizeof(outarg))
1455 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1459 err = -ENAMETOOLONG;
1460 if (outarg.namelen > FUSE_NAME_MAX)
1464 if (size != sizeof(outarg) + outarg.namelen + 1)
1468 name.len = outarg.namelen;
1469 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1472 fuse_copy_finish(cs);
1473 buf[outarg.namelen] = 0;
1475 down_read(&fc->killsb);
1478 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1479 up_read(&fc->killsb);
1485 fuse_copy_finish(cs);
1489 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1490 struct fuse_copy_state *cs)
1492 struct fuse_notify_delete_out outarg;
1497 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1502 if (size < sizeof(outarg))
1505 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1509 err = -ENAMETOOLONG;
1510 if (outarg.namelen > FUSE_NAME_MAX)
1514 if (size != sizeof(outarg) + outarg.namelen + 1)
1518 name.len = outarg.namelen;
1519 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1522 fuse_copy_finish(cs);
1523 buf[outarg.namelen] = 0;
1525 down_read(&fc->killsb);
1528 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1529 outarg.child, &name);
1530 up_read(&fc->killsb);
1536 fuse_copy_finish(cs);
1540 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1541 struct fuse_copy_state *cs)
1543 struct fuse_notify_store_out outarg;
1544 struct inode *inode;
1545 struct address_space *mapping;
1549 unsigned int offset;
1555 if (size < sizeof(outarg))
1558 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1563 if (size - sizeof(outarg) != outarg.size)
1566 nodeid = outarg.nodeid;
1568 down_read(&fc->killsb);
1574 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1578 mapping = inode->i_mapping;
1579 index = outarg.offset >> PAGE_SHIFT;
1580 offset = outarg.offset & ~PAGE_MASK;
1581 file_size = i_size_read(inode);
1582 end = outarg.offset + outarg.size;
1583 if (end > file_size) {
1585 fuse_write_update_size(inode, file_size);
1591 unsigned int this_num;
1594 page = find_or_create_page(mapping, index,
1595 mapping_gfp_mask(mapping));
1599 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1600 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1601 if (!err && offset == 0 &&
1602 (this_num == PAGE_SIZE || file_size == end))
1603 SetPageUptodate(page);
1620 up_read(&fc->killsb);
1622 fuse_copy_finish(cs);
1626 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1628 release_pages(req->pages, req->num_pages, false);
1631 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1632 struct fuse_notify_retrieve_out *outarg)
1635 struct address_space *mapping = inode->i_mapping;
1636 struct fuse_req *req;
1640 unsigned int offset;
1641 size_t total_len = 0;
1644 offset = outarg->offset & ~PAGE_MASK;
1645 file_size = i_size_read(inode);
1648 if (outarg->offset > file_size)
1650 else if (outarg->offset + num > file_size)
1651 num = file_size - outarg->offset;
1653 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1654 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1656 req = fuse_get_req(fc, num_pages);
1658 return PTR_ERR(req);
1660 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1661 req->in.h.nodeid = outarg->nodeid;
1662 req->in.numargs = 2;
1663 req->in.argpages = 1;
1664 req->page_descs[0].offset = offset;
1665 req->end = fuse_retrieve_end;
1667 index = outarg->offset >> PAGE_SHIFT;
1669 while (num && req->num_pages < num_pages) {
1671 unsigned int this_num;
1673 page = find_get_page(mapping, index);
1677 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1678 req->pages[req->num_pages] = page;
1679 req->page_descs[req->num_pages].length = this_num;
1684 total_len += this_num;
1687 req->misc.retrieve_in.offset = outarg->offset;
1688 req->misc.retrieve_in.size = total_len;
1689 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1690 req->in.args[0].value = &req->misc.retrieve_in;
1691 req->in.args[1].size = total_len;
1693 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1695 fuse_retrieve_end(fc, req);
1700 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1701 struct fuse_copy_state *cs)
1703 struct fuse_notify_retrieve_out outarg;
1704 struct inode *inode;
1708 if (size != sizeof(outarg))
1711 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1715 fuse_copy_finish(cs);
1717 down_read(&fc->killsb);
1720 u64 nodeid = outarg.nodeid;
1722 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1724 err = fuse_retrieve(fc, inode, &outarg);
1728 up_read(&fc->killsb);
1733 fuse_copy_finish(cs);
1737 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1738 unsigned int size, struct fuse_copy_state *cs)
1740 /* Don't try to move pages (yet) */
1744 case FUSE_NOTIFY_POLL:
1745 return fuse_notify_poll(fc, size, cs);
1747 case FUSE_NOTIFY_INVAL_INODE:
1748 return fuse_notify_inval_inode(fc, size, cs);
1750 case FUSE_NOTIFY_INVAL_ENTRY:
1751 return fuse_notify_inval_entry(fc, size, cs);
1753 case FUSE_NOTIFY_STORE:
1754 return fuse_notify_store(fc, size, cs);
1756 case FUSE_NOTIFY_RETRIEVE:
1757 return fuse_notify_retrieve(fc, size, cs);
1759 case FUSE_NOTIFY_DELETE:
1760 return fuse_notify_delete(fc, size, cs);
1763 fuse_copy_finish(cs);
1768 /* Look up request on processing list by unique ID */
1769 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1771 struct fuse_req *req;
1773 list_for_each_entry(req, &fpq->processing, list) {
1774 if (req->in.h.unique == unique || req->intr_unique == unique)
1780 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1783 unsigned reqsize = sizeof(struct fuse_out_header);
1786 return nbytes != reqsize ? -EINVAL : 0;
1788 reqsize += len_args(out->numargs, out->args);
1790 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1792 else if (reqsize > nbytes) {
1793 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1794 unsigned diffsize = reqsize - nbytes;
1795 if (diffsize > lastarg->size)
1797 lastarg->size -= diffsize;
1799 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1804 * Write a single reply to a request. First the header is copied from
1805 * the write buffer. The request is then searched on the processing
1806 * list by the unique ID found in the header. If found, then remove
1807 * it from the list and copy the rest of the buffer to the request.
1808 * The request is finished by calling request_end()
1810 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1811 struct fuse_copy_state *cs, size_t nbytes)
1814 struct fuse_conn *fc = fud->fc;
1815 struct fuse_pqueue *fpq = &fud->pq;
1816 struct fuse_req *req;
1817 struct fuse_out_header oh;
1819 if (nbytes < sizeof(struct fuse_out_header))
1822 err = fuse_copy_one(cs, &oh, sizeof(oh));
1827 if (oh.len != nbytes)
1831 * Zero oh.unique indicates unsolicited notification message
1832 * and error contains notification code.
1835 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1836 return err ? err : nbytes;
1840 if (oh.error <= -1000 || oh.error > 0)
1843 spin_lock(&fpq->lock);
1845 if (!fpq->connected)
1848 req = request_find(fpq, oh.unique);
1852 /* Is it an interrupt reply? */
1853 if (req->intr_unique == oh.unique) {
1854 spin_unlock(&fpq->lock);
1857 if (nbytes != sizeof(struct fuse_out_header))
1860 if (oh.error == -ENOSYS)
1861 fc->no_interrupt = 1;
1862 else if (oh.error == -EAGAIN)
1863 queue_interrupt(&fc->iq, req);
1865 fuse_copy_finish(cs);
1869 clear_bit(FR_SENT, &req->flags);
1870 list_move(&req->list, &fpq->io);
1872 set_bit(FR_LOCKED, &req->flags);
1873 spin_unlock(&fpq->lock);
1875 if (!req->out.page_replace)
1878 err = copy_out_args(cs, &req->out, nbytes);
1879 fuse_copy_finish(cs);
1881 spin_lock(&fpq->lock);
1882 clear_bit(FR_LOCKED, &req->flags);
1883 if (!fpq->connected)
1886 req->out.h.error = -EIO;
1887 if (!test_bit(FR_PRIVATE, &req->flags))
1888 list_del_init(&req->list);
1889 spin_unlock(&fpq->lock);
1891 request_end(fc, req);
1893 return err ? err : nbytes;
1896 spin_unlock(&fpq->lock);
1898 fuse_copy_finish(cs);
1902 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1904 struct fuse_copy_state cs;
1905 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1910 if (!iter_is_iovec(from))
1913 fuse_copy_init(&cs, 0, from);
1915 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1918 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1919 struct file *out, loff_t *ppos,
1920 size_t len, unsigned int flags)
1924 struct pipe_buffer *bufs;
1925 struct fuse_copy_state cs;
1926 struct fuse_dev *fud;
1930 fud = fuse_get_dev(out);
1934 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1941 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1942 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1952 struct pipe_buffer *ibuf;
1953 struct pipe_buffer *obuf;
1955 BUG_ON(nbuf >= pipe->buffers);
1956 BUG_ON(!pipe->nrbufs);
1957 ibuf = &pipe->bufs[pipe->curbuf];
1960 if (rem >= ibuf->len) {
1963 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1966 pipe_buf_get(pipe, ibuf);
1968 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1970 ibuf->offset += obuf->len;
1971 ibuf->len -= obuf->len;
1978 fuse_copy_init(&cs, 0, NULL);
1983 if (flags & SPLICE_F_MOVE)
1986 ret = fuse_dev_do_write(fud, &cs, len);
1988 for (idx = 0; idx < nbuf; idx++)
1989 pipe_buf_release(pipe, &bufs[idx]);
1996 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1998 unsigned mask = POLLOUT | POLLWRNORM;
1999 struct fuse_iqueue *fiq;
2000 struct fuse_dev *fud = fuse_get_dev(file);
2006 poll_wait(file, &fiq->waitq, wait);
2008 spin_lock(&fiq->waitq.lock);
2009 if (!fiq->connected)
2011 else if (request_pending(fiq))
2012 mask |= POLLIN | POLLRDNORM;
2013 spin_unlock(&fiq->waitq.lock);
2019 * Abort all requests on the given list (pending or processing)
2021 * This function releases and reacquires fc->lock
2023 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2025 while (!list_empty(head)) {
2026 struct fuse_req *req;
2027 req = list_entry(head->next, struct fuse_req, list);
2028 req->out.h.error = -ECONNABORTED;
2029 clear_bit(FR_PENDING, &req->flags);
2030 clear_bit(FR_SENT, &req->flags);
2031 list_del_init(&req->list);
2032 request_end(fc, req);
2036 static void end_polls(struct fuse_conn *fc)
2040 p = rb_first(&fc->polled_files);
2043 struct fuse_file *ff;
2044 ff = rb_entry(p, struct fuse_file, polled_node);
2045 wake_up_interruptible_all(&ff->poll_wait);
2052 * Abort all requests.
2054 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2057 * The same effect is usually achievable through killing the filesystem daemon
2058 * and all users of the filesystem. The exception is the combination of an
2059 * asynchronous request and the tricky deadlock (see
2060 * Documentation/filesystems/fuse.txt).
2062 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2063 * requests, they should be finished off immediately. Locked requests will be
2064 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2065 * requests. It is possible that some request will finish before we can. This
2066 * is OK, the request will in that case be removed from the list before we touch
2069 void fuse_abort_conn(struct fuse_conn *fc)
2071 struct fuse_iqueue *fiq = &fc->iq;
2073 spin_lock(&fc->lock);
2074 if (fc->connected) {
2075 struct fuse_dev *fud;
2076 struct fuse_req *req, *next;
2082 fuse_set_initialized(fc);
2083 list_for_each_entry(fud, &fc->devices, entry) {
2084 struct fuse_pqueue *fpq = &fud->pq;
2086 spin_lock(&fpq->lock);
2088 list_for_each_entry_safe(req, next, &fpq->io, list) {
2089 req->out.h.error = -ECONNABORTED;
2090 spin_lock(&req->waitq.lock);
2091 set_bit(FR_ABORTED, &req->flags);
2092 if (!test_bit(FR_LOCKED, &req->flags)) {
2093 set_bit(FR_PRIVATE, &req->flags);
2094 list_move(&req->list, &to_end1);
2096 spin_unlock(&req->waitq.lock);
2098 list_splice_init(&fpq->processing, &to_end2);
2099 spin_unlock(&fpq->lock);
2101 fc->max_background = UINT_MAX;
2104 spin_lock(&fiq->waitq.lock);
2106 list_splice_init(&fiq->pending, &to_end2);
2107 while (forget_pending(fiq))
2108 kfree(dequeue_forget(fiq, 1, NULL));
2109 wake_up_all_locked(&fiq->waitq);
2110 spin_unlock(&fiq->waitq.lock);
2111 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2113 wake_up_all(&fc->blocked_waitq);
2114 spin_unlock(&fc->lock);
2116 while (!list_empty(&to_end1)) {
2117 req = list_first_entry(&to_end1, struct fuse_req, list);
2118 __fuse_get_request(req);
2119 list_del_init(&req->list);
2120 request_end(fc, req);
2122 end_requests(fc, &to_end2);
2124 spin_unlock(&fc->lock);
2127 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2129 int fuse_dev_release(struct inode *inode, struct file *file)
2131 struct fuse_dev *fud = fuse_get_dev(file);
2134 struct fuse_conn *fc = fud->fc;
2135 struct fuse_pqueue *fpq = &fud->pq;
2137 WARN_ON(!list_empty(&fpq->io));
2138 end_requests(fc, &fpq->processing);
2139 /* Are we the last open device? */
2140 if (atomic_dec_and_test(&fc->dev_count)) {
2141 WARN_ON(fc->iq.fasync != NULL);
2142 fuse_abort_conn(fc);
2148 EXPORT_SYMBOL_GPL(fuse_dev_release);
2150 static int fuse_dev_fasync(int fd, struct file *file, int on)
2152 struct fuse_dev *fud = fuse_get_dev(file);
2157 /* No locking - fasync_helper does its own locking */
2158 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2161 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2163 struct fuse_dev *fud;
2165 if (new->private_data)
2168 fud = fuse_dev_alloc(fc);
2172 new->private_data = fud;
2173 atomic_inc(&fc->dev_count);
2178 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2183 if (cmd == FUSE_DEV_IOC_CLONE) {
2187 if (!get_user(oldfd, (__u32 __user *) arg)) {
2188 struct file *old = fget(oldfd);
2192 struct fuse_dev *fud = NULL;
2195 * Check against file->f_op because CUSE
2196 * uses the same ioctl handler.
2198 if (old->f_op == file->f_op &&
2199 old->f_cred->user_ns == file->f_cred->user_ns)
2200 fud = fuse_get_dev(old);
2203 mutex_lock(&fuse_mutex);
2204 err = fuse_device_clone(fud->fc, file);
2205 mutex_unlock(&fuse_mutex);
2214 const struct file_operations fuse_dev_operations = {
2215 .owner = THIS_MODULE,
2216 .open = fuse_dev_open,
2217 .llseek = no_llseek,
2218 .read_iter = fuse_dev_read,
2219 .splice_read = fuse_dev_splice_read,
2220 .write_iter = fuse_dev_write,
2221 .splice_write = fuse_dev_splice_write,
2222 .poll = fuse_dev_poll,
2223 .release = fuse_dev_release,
2224 .fasync = fuse_dev_fasync,
2225 .unlocked_ioctl = fuse_dev_ioctl,
2226 .compat_ioctl = fuse_dev_ioctl,
2228 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2230 static struct miscdevice fuse_miscdevice = {
2231 .minor = FUSE_MINOR,
2233 .fops = &fuse_dev_operations,
2236 int __init fuse_dev_init(void)
2239 fuse_req_cachep = kmem_cache_create("fuse_request",
2240 sizeof(struct fuse_req),
2242 if (!fuse_req_cachep)
2245 err = misc_register(&fuse_miscdevice);
2247 goto out_cache_clean;
2252 kmem_cache_destroy(fuse_req_cachep);
2257 void fuse_dev_cleanup(void)
2259 misc_deregister(&fuse_miscdevice);
2260 kmem_cache_destroy(fuse_req_cachep);