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)
39 memset(req, 0, sizeof(*req));
40 INIT_LIST_HEAD(&req->list);
41 INIT_LIST_HEAD(&req->intr_entry);
42 init_waitqueue_head(&req->waitq);
43 atomic_set(&req->count, 1);
46 struct fuse_req *fuse_request_alloc(void)
48 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL);
50 fuse_request_init(req);
53 EXPORT_SYMBOL_GPL(fuse_request_alloc);
55 struct fuse_req *fuse_request_alloc_nofs(void)
57 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_NOFS);
59 fuse_request_init(req);
63 void fuse_request_free(struct fuse_req *req)
65 kmem_cache_free(fuse_req_cachep, req);
68 static void block_sigs(sigset_t *oldset)
72 siginitsetinv(&mask, sigmask(SIGKILL));
73 sigprocmask(SIG_BLOCK, &mask, oldset);
76 static void restore_sigs(sigset_t *oldset)
78 sigprocmask(SIG_SETMASK, oldset, NULL);
81 static void __fuse_get_request(struct fuse_req *req)
83 atomic_inc(&req->count);
86 /* Must be called with > 1 refcount */
87 static void __fuse_put_request(struct fuse_req *req)
89 BUG_ON(atomic_read(&req->count) < 2);
90 atomic_dec(&req->count);
93 static void fuse_req_init_context(struct fuse_req *req)
95 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
96 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
97 req->in.h.pid = current->pid;
100 struct fuse_req *fuse_get_req(struct fuse_conn *fc)
102 struct fuse_req *req;
107 atomic_inc(&fc->num_waiting);
109 intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked);
110 restore_sigs(&oldset);
119 req = fuse_request_alloc();
124 fuse_req_init_context(req);
129 atomic_dec(&fc->num_waiting);
132 EXPORT_SYMBOL_GPL(fuse_get_req);
135 * Return request in fuse_file->reserved_req. However that may
136 * currently be in use. If that is the case, wait for it to become
139 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
142 struct fuse_req *req = NULL;
143 struct fuse_file *ff = file->private_data;
146 wait_event(fc->reserved_req_waitq, ff->reserved_req);
147 spin_lock(&fc->lock);
148 if (ff->reserved_req) {
149 req = ff->reserved_req;
150 ff->reserved_req = NULL;
151 req->stolen_file = get_file(file);
153 spin_unlock(&fc->lock);
160 * Put stolen request back into fuse_file->reserved_req
162 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
164 struct file *file = req->stolen_file;
165 struct fuse_file *ff = file->private_data;
167 spin_lock(&fc->lock);
168 fuse_request_init(req);
169 BUG_ON(ff->reserved_req);
170 ff->reserved_req = req;
171 wake_up_all(&fc->reserved_req_waitq);
172 spin_unlock(&fc->lock);
177 * Gets a requests for a file operation, always succeeds
179 * This is used for sending the FLUSH request, which must get to
180 * userspace, due to POSIX locks which may need to be unlocked.
182 * If allocation fails due to OOM, use the reserved request in
185 * This is very unlikely to deadlock accidentally, since the
186 * filesystem should not have it's own file open. If deadlock is
187 * intentional, it can still be broken by "aborting" the filesystem.
189 struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file)
191 struct fuse_req *req;
193 atomic_inc(&fc->num_waiting);
194 wait_event(fc->blocked_waitq, !fc->blocked);
195 req = fuse_request_alloc();
197 req = get_reserved_req(fc, file);
199 fuse_req_init_context(req);
204 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
206 if (atomic_dec_and_test(&req->count)) {
208 atomic_dec(&fc->num_waiting);
210 if (req->stolen_file)
211 put_reserved_req(fc, req);
213 fuse_request_free(req);
216 EXPORT_SYMBOL_GPL(fuse_put_request);
218 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
223 for (i = 0; i < numargs; i++)
224 nbytes += args[i].size;
229 static u64 fuse_get_unique(struct fuse_conn *fc)
232 /* zero is special */
239 static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
241 req->in.h.len = sizeof(struct fuse_in_header) +
242 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
243 list_add_tail(&req->list, &fc->pending);
244 req->state = FUSE_REQ_PENDING;
247 atomic_inc(&fc->num_waiting);
250 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
253 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
254 u64 nodeid, u64 nlookup)
256 forget->forget_one.nodeid = nodeid;
257 forget->forget_one.nlookup = nlookup;
259 spin_lock(&fc->lock);
261 fc->forget_list_tail->next = forget;
262 fc->forget_list_tail = forget;
264 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
268 spin_unlock(&fc->lock);
271 static void flush_bg_queue(struct fuse_conn *fc)
273 while (fc->active_background < fc->max_background &&
274 !list_empty(&fc->bg_queue)) {
275 struct fuse_req *req;
277 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
278 list_del(&req->list);
279 fc->active_background++;
280 req->in.h.unique = fuse_get_unique(fc);
281 queue_request(fc, req);
286 * This function is called when a request is finished. Either a reply
287 * has arrived or it was aborted (and not yet sent) or some error
288 * occurred during communication with userspace, or the device file
289 * was closed. The requester thread is woken up (if still waiting),
290 * the 'end' callback is called if given, else the reference to the
291 * request is released
293 * Called with fc->lock, unlocks it
295 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
298 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
300 list_del(&req->list);
301 list_del(&req->intr_entry);
302 req->state = FUSE_REQ_FINISHED;
303 if (req->background) {
304 if (fc->num_background == fc->max_background) {
306 wake_up_all(&fc->blocked_waitq);
308 if (fc->num_background == fc->congestion_threshold &&
309 fc->connected && fc->bdi_initialized) {
310 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
311 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
313 fc->num_background--;
314 fc->active_background--;
317 spin_unlock(&fc->lock);
318 wake_up(&req->waitq);
321 fuse_put_request(fc, req);
324 static void wait_answer_interruptible(struct fuse_conn *fc,
325 struct fuse_req *req)
329 if (signal_pending(current))
332 spin_unlock(&fc->lock);
333 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
334 spin_lock(&fc->lock);
337 static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
339 list_add_tail(&req->intr_entry, &fc->interrupts);
341 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
344 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
348 if (!fc->no_interrupt) {
349 /* Any signal may interrupt this */
350 wait_answer_interruptible(fc, req);
354 if (req->state == FUSE_REQ_FINISHED)
357 req->interrupted = 1;
358 if (req->state == FUSE_REQ_SENT)
359 queue_interrupt(fc, req);
365 /* Only fatal signals may interrupt this */
367 wait_answer_interruptible(fc, req);
368 restore_sigs(&oldset);
372 if (req->state == FUSE_REQ_FINISHED)
375 /* Request is not yet in userspace, bail out */
376 if (req->state == FUSE_REQ_PENDING) {
377 list_del(&req->list);
378 __fuse_put_request(req);
379 req->out.h.error = -EINTR;
385 * Either request is already in userspace, or it was forced.
388 spin_unlock(&fc->lock);
389 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
390 spin_lock(&fc->lock);
396 BUG_ON(req->state != FUSE_REQ_FINISHED);
398 /* This is uninterruptible sleep, because data is
399 being copied to/from the buffers of req. During
400 locked state, there mustn't be any filesystem
401 operation (e.g. page fault), since that could lead
403 spin_unlock(&fc->lock);
404 wait_event(req->waitq, !req->locked);
405 spin_lock(&fc->lock);
409 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
412 spin_lock(&fc->lock);
414 req->out.h.error = -ENOTCONN;
415 else if (fc->conn_error)
416 req->out.h.error = -ECONNREFUSED;
418 req->in.h.unique = fuse_get_unique(fc);
419 queue_request(fc, req);
420 /* acquire extra reference, since request is still needed
421 after request_end() */
422 __fuse_get_request(req);
424 request_wait_answer(fc, req);
426 spin_unlock(&fc->lock);
428 EXPORT_SYMBOL_GPL(fuse_request_send);
430 static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
431 struct fuse_req *req)
434 fc->num_background++;
435 if (fc->num_background == fc->max_background)
437 if (fc->num_background == fc->congestion_threshold &&
438 fc->bdi_initialized) {
439 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
440 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
442 list_add_tail(&req->list, &fc->bg_queue);
446 static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
448 spin_lock(&fc->lock);
450 fuse_request_send_nowait_locked(fc, req);
451 spin_unlock(&fc->lock);
453 req->out.h.error = -ENOTCONN;
454 request_end(fc, req);
458 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
461 fuse_request_send_nowait(fc, req);
463 EXPORT_SYMBOL_GPL(fuse_request_send_background);
465 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
466 struct fuse_req *req, u64 unique)
471 req->in.h.unique = unique;
472 spin_lock(&fc->lock);
474 queue_request(fc, req);
477 spin_unlock(&fc->lock);
483 * Called under fc->lock
485 * fc->connected must have been checked previously
487 void fuse_request_send_background_locked(struct fuse_conn *fc,
488 struct fuse_req *req)
491 fuse_request_send_nowait_locked(fc, req);
495 * Lock the request. Up to the next unlock_request() there mustn't be
496 * anything that could cause a page-fault. If the request was already
499 static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
503 spin_lock(&fc->lock);
508 spin_unlock(&fc->lock);
514 * Unlock request. If it was aborted during being locked, the
515 * requester thread is currently waiting for it to be unlocked, so
518 static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
521 spin_lock(&fc->lock);
524 wake_up(&req->waitq);
525 spin_unlock(&fc->lock);
529 struct fuse_copy_state {
530 struct fuse_conn *fc;
532 struct fuse_req *req;
533 const struct iovec *iov;
534 struct pipe_buffer *pipebufs;
535 struct pipe_buffer *currbuf;
536 struct pipe_inode_info *pipe;
537 unsigned long nr_segs;
538 unsigned long seglen;
544 unsigned move_pages:1;
547 static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
549 const struct iovec *iov, unsigned long nr_segs)
551 memset(cs, 0, sizeof(*cs));
555 cs->nr_segs = nr_segs;
558 /* Unmap and put previous page of userspace buffer */
559 static void fuse_copy_finish(struct fuse_copy_state *cs)
562 struct pipe_buffer *buf = cs->currbuf;
565 buf->ops->unmap(cs->pipe, buf, cs->mapaddr);
568 buf->len = PAGE_SIZE - cs->len;
572 } else if (cs->mapaddr) {
575 flush_dcache_page(cs->pg);
576 set_page_dirty_lock(cs->pg);
584 * Get another pagefull of userspace buffer, and map it to kernel
585 * address space, and lock request
587 static int fuse_copy_fill(struct fuse_copy_state *cs)
589 unsigned long offset;
592 unlock_request(cs->fc, cs->req);
593 fuse_copy_finish(cs);
595 struct pipe_buffer *buf = cs->pipebufs;
598 err = buf->ops->confirm(cs->pipe, buf);
602 BUG_ON(!cs->nr_segs);
604 cs->mapaddr = buf->ops->map(cs->pipe, buf, 0);
606 cs->buf = cs->mapaddr + buf->offset;
612 if (cs->nr_segs == cs->pipe->buffers)
615 page = alloc_page(GFP_HIGHUSER);
624 cs->mapaddr = kmap(page);
625 cs->buf = cs->mapaddr;
632 BUG_ON(!cs->nr_segs);
633 cs->seglen = cs->iov[0].iov_len;
634 cs->addr = (unsigned long) cs->iov[0].iov_base;
638 err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg);
642 offset = cs->addr % PAGE_SIZE;
643 cs->mapaddr = kmap(cs->pg);
644 cs->buf = cs->mapaddr + offset;
645 cs->len = min(PAGE_SIZE - offset, cs->seglen);
646 cs->seglen -= cs->len;
650 return lock_request(cs->fc, cs->req);
653 /* Do as much copy to/from userspace buffer as we can */
654 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
656 unsigned ncpy = min(*size, cs->len);
659 memcpy(cs->buf, *val, ncpy);
661 memcpy(*val, cs->buf, ncpy);
670 static int fuse_check_page(struct page *page)
672 if (page_mapcount(page) ||
673 page->mapping != NULL ||
674 page_count(page) != 1 ||
675 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
682 printk(KERN_WARNING "fuse: trying to steal weird page\n");
683 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);
689 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
692 struct page *oldpage = *pagep;
693 struct page *newpage;
694 struct pipe_buffer *buf = cs->pipebufs;
696 unlock_request(cs->fc, cs->req);
697 fuse_copy_finish(cs);
699 err = buf->ops->confirm(cs->pipe, buf);
703 BUG_ON(!cs->nr_segs);
709 if (cs->len != PAGE_SIZE)
712 if (buf->ops->steal(cs->pipe, buf) != 0)
717 if (WARN_ON(!PageUptodate(newpage)))
720 ClearPageMappedToDisk(newpage);
722 if (fuse_check_page(newpage) != 0)
723 goto out_fallback_unlock;
726 * This is a new and locked page, it shouldn't be mapped or
727 * have any special flags on it
729 if (WARN_ON(page_mapped(oldpage)))
730 goto out_fallback_unlock;
731 if (WARN_ON(page_has_private(oldpage)))
732 goto out_fallback_unlock;
733 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
734 goto out_fallback_unlock;
735 if (WARN_ON(PageMlocked(oldpage)))
736 goto out_fallback_unlock;
738 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
740 unlock_page(newpage);
744 page_cache_get(newpage);
746 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
747 lru_cache_add_file(newpage);
750 spin_lock(&cs->fc->lock);
751 if (cs->req->aborted)
755 spin_unlock(&cs->fc->lock);
758 unlock_page(newpage);
759 page_cache_release(newpage);
763 unlock_page(oldpage);
764 page_cache_release(oldpage);
770 unlock_page(newpage);
772 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
773 cs->buf = cs->mapaddr + buf->offset;
775 err = lock_request(cs->fc, cs->req);
782 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
783 unsigned offset, unsigned count)
785 struct pipe_buffer *buf;
787 if (cs->nr_segs == cs->pipe->buffers)
790 unlock_request(cs->fc, cs->req);
791 fuse_copy_finish(cs);
794 page_cache_get(page);
796 buf->offset = offset;
807 * Copy a page in the request to/from the userspace buffer. Must be
810 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
811 unsigned offset, unsigned count, int zeroing)
814 struct page *page = *pagep;
816 if (page && zeroing && count < PAGE_SIZE)
817 clear_highpage(page);
820 if (cs->write && cs->pipebufs && page) {
821 return fuse_ref_page(cs, page, offset, count);
822 } else if (!cs->len) {
823 if (cs->move_pages && page &&
824 offset == 0 && count == PAGE_SIZE) {
825 err = fuse_try_move_page(cs, pagep);
829 err = fuse_copy_fill(cs);
835 void *mapaddr = kmap_atomic(page);
836 void *buf = mapaddr + offset;
837 offset += fuse_copy_do(cs, &buf, &count);
838 kunmap_atomic(mapaddr);
840 offset += fuse_copy_do(cs, NULL, &count);
842 if (page && !cs->write)
843 flush_dcache_page(page);
847 /* Copy pages in the request to/from userspace buffer */
848 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
852 struct fuse_req *req = cs->req;
853 unsigned offset = req->page_offset;
854 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
856 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
859 err = fuse_copy_page(cs, &req->pages[i], offset, count,
865 count = min(nbytes, (unsigned) PAGE_SIZE);
871 /* Copy a single argument in the request to/from userspace buffer */
872 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
876 int err = fuse_copy_fill(cs);
880 fuse_copy_do(cs, &val, &size);
885 /* Copy request arguments to/from userspace buffer */
886 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
887 unsigned argpages, struct fuse_arg *args,
893 for (i = 0; !err && i < numargs; i++) {
894 struct fuse_arg *arg = &args[i];
895 if (i == numargs - 1 && argpages)
896 err = fuse_copy_pages(cs, arg->size, zeroing);
898 err = fuse_copy_one(cs, arg->value, arg->size);
903 static int forget_pending(struct fuse_conn *fc)
905 return fc->forget_list_head.next != NULL;
908 static int request_pending(struct fuse_conn *fc)
910 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
914 /* Wait until a request is available on the pending list */
915 static void request_wait(struct fuse_conn *fc)
919 DECLARE_WAITQUEUE(wait, current);
921 add_wait_queue_exclusive(&fc->waitq, &wait);
922 while (fc->connected && !request_pending(fc)) {
923 set_current_state(TASK_INTERRUPTIBLE);
924 if (signal_pending(current))
927 spin_unlock(&fc->lock);
929 spin_lock(&fc->lock);
931 set_current_state(TASK_RUNNING);
932 remove_wait_queue(&fc->waitq, &wait);
936 * Transfer an interrupt request to userspace
938 * Unlike other requests this is assembled on demand, without a need
939 * to allocate a separate fuse_req structure.
941 * Called with fc->lock held, releases it
943 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
944 size_t nbytes, struct fuse_req *req)
947 struct fuse_in_header ih;
948 struct fuse_interrupt_in arg;
949 unsigned reqsize = sizeof(ih) + sizeof(arg);
952 list_del_init(&req->intr_entry);
953 req->intr_unique = fuse_get_unique(fc);
954 memset(&ih, 0, sizeof(ih));
955 memset(&arg, 0, sizeof(arg));
957 ih.opcode = FUSE_INTERRUPT;
958 ih.unique = req->intr_unique;
959 arg.unique = req->in.h.unique;
961 spin_unlock(&fc->lock);
962 if (nbytes < reqsize)
965 err = fuse_copy_one(cs, &ih, sizeof(ih));
967 err = fuse_copy_one(cs, &arg, sizeof(arg));
968 fuse_copy_finish(cs);
970 return err ? err : reqsize;
973 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
977 struct fuse_forget_link *head = fc->forget_list_head.next;
978 struct fuse_forget_link **newhead = &head;
981 for (count = 0; *newhead != NULL && count < max; count++)
982 newhead = &(*newhead)->next;
984 fc->forget_list_head.next = *newhead;
986 if (fc->forget_list_head.next == NULL)
987 fc->forget_list_tail = &fc->forget_list_head;
995 static int fuse_read_single_forget(struct fuse_conn *fc,
996 struct fuse_copy_state *cs,
1001 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1002 struct fuse_forget_in arg = {
1003 .nlookup = forget->forget_one.nlookup,
1005 struct fuse_in_header ih = {
1006 .opcode = FUSE_FORGET,
1007 .nodeid = forget->forget_one.nodeid,
1008 .unique = fuse_get_unique(fc),
1009 .len = sizeof(ih) + sizeof(arg),
1012 spin_unlock(&fc->lock);
1014 if (nbytes < ih.len)
1017 err = fuse_copy_one(cs, &ih, sizeof(ih));
1019 err = fuse_copy_one(cs, &arg, sizeof(arg));
1020 fuse_copy_finish(cs);
1028 static int fuse_read_batch_forget(struct fuse_conn *fc,
1029 struct fuse_copy_state *cs, size_t nbytes)
1030 __releases(fc->lock)
1033 unsigned max_forgets;
1035 struct fuse_forget_link *head;
1036 struct fuse_batch_forget_in arg = { .count = 0 };
1037 struct fuse_in_header ih = {
1038 .opcode = FUSE_BATCH_FORGET,
1039 .unique = fuse_get_unique(fc),
1040 .len = sizeof(ih) + sizeof(arg),
1043 if (nbytes < ih.len) {
1044 spin_unlock(&fc->lock);
1048 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1049 head = dequeue_forget(fc, max_forgets, &count);
1050 spin_unlock(&fc->lock);
1053 ih.len += count * sizeof(struct fuse_forget_one);
1054 err = fuse_copy_one(cs, &ih, sizeof(ih));
1056 err = fuse_copy_one(cs, &arg, sizeof(arg));
1059 struct fuse_forget_link *forget = head;
1062 err = fuse_copy_one(cs, &forget->forget_one,
1063 sizeof(forget->forget_one));
1065 head = forget->next;
1069 fuse_copy_finish(cs);
1077 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1079 __releases(fc->lock)
1081 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1082 return fuse_read_single_forget(fc, cs, nbytes);
1084 return fuse_read_batch_forget(fc, cs, nbytes);
1088 * Read a single request into the userspace filesystem's buffer. This
1089 * function waits until a request is available, then removes it from
1090 * the pending list and copies request data to userspace buffer. If
1091 * no reply is needed (FORGET) or request has been aborted or there
1092 * was an error during the copying then it's finished by calling
1093 * request_end(). Otherwise add it to the processing list, and set
1096 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1097 struct fuse_copy_state *cs, size_t nbytes)
1100 struct fuse_req *req;
1105 spin_lock(&fc->lock);
1107 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1108 !request_pending(fc))
1116 if (!request_pending(fc))
1119 if (!list_empty(&fc->interrupts)) {
1120 req = list_entry(fc->interrupts.next, struct fuse_req,
1122 return fuse_read_interrupt(fc, cs, nbytes, req);
1125 if (forget_pending(fc)) {
1126 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1127 return fuse_read_forget(fc, cs, nbytes);
1129 if (fc->forget_batch <= -8)
1130 fc->forget_batch = 16;
1133 req = list_entry(fc->pending.next, struct fuse_req, list);
1134 req->state = FUSE_REQ_READING;
1135 list_move(&req->list, &fc->io);
1138 reqsize = in->h.len;
1139 /* If request is too large, reply with an error and restart the read */
1140 if (nbytes < reqsize) {
1141 req->out.h.error = -EIO;
1142 /* SETXATTR is special, since it may contain too large data */
1143 if (in->h.opcode == FUSE_SETXATTR)
1144 req->out.h.error = -E2BIG;
1145 request_end(fc, req);
1148 spin_unlock(&fc->lock);
1150 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1152 err = fuse_copy_args(cs, in->numargs, in->argpages,
1153 (struct fuse_arg *) in->args, 0);
1154 fuse_copy_finish(cs);
1155 spin_lock(&fc->lock);
1158 request_end(fc, req);
1162 req->out.h.error = -EIO;
1163 request_end(fc, req);
1167 request_end(fc, req);
1169 req->state = FUSE_REQ_SENT;
1170 list_move_tail(&req->list, &fc->processing);
1171 if (req->interrupted)
1172 queue_interrupt(fc, req);
1173 spin_unlock(&fc->lock);
1178 spin_unlock(&fc->lock);
1182 static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1183 unsigned long nr_segs, loff_t pos)
1185 struct fuse_copy_state cs;
1186 struct file *file = iocb->ki_filp;
1187 struct fuse_conn *fc = fuse_get_conn(file);
1191 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1193 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1196 static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe,
1197 struct pipe_buffer *buf)
1202 static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = {
1204 .map = generic_pipe_buf_map,
1205 .unmap = generic_pipe_buf_unmap,
1206 .confirm = generic_pipe_buf_confirm,
1207 .release = generic_pipe_buf_release,
1208 .steal = fuse_dev_pipe_buf_steal,
1209 .get = generic_pipe_buf_get,
1212 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1213 struct pipe_inode_info *pipe,
1214 size_t len, unsigned int flags)
1219 struct pipe_buffer *bufs;
1220 struct fuse_copy_state cs;
1221 struct fuse_conn *fc = fuse_get_conn(in);
1225 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1229 fuse_copy_init(&cs, fc, 1, NULL, 0);
1232 ret = fuse_dev_do_read(fc, in, &cs, len);
1239 if (!pipe->readers) {
1240 send_sig(SIGPIPE, current, 0);
1246 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1251 while (page_nr < cs.nr_segs) {
1252 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1253 struct pipe_buffer *buf = pipe->bufs + newbuf;
1255 buf->page = bufs[page_nr].page;
1256 buf->offset = bufs[page_nr].offset;
1257 buf->len = bufs[page_nr].len;
1258 buf->ops = &fuse_dev_pipe_buf_ops;
1273 if (waitqueue_active(&pipe->wait))
1274 wake_up_interruptible(&pipe->wait);
1275 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1279 for (; page_nr < cs.nr_segs; page_nr++)
1280 page_cache_release(bufs[page_nr].page);
1286 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1287 struct fuse_copy_state *cs)
1289 struct fuse_notify_poll_wakeup_out outarg;
1292 if (size != sizeof(outarg))
1295 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1299 fuse_copy_finish(cs);
1300 return fuse_notify_poll_wakeup(fc, &outarg);
1303 fuse_copy_finish(cs);
1307 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1308 struct fuse_copy_state *cs)
1310 struct fuse_notify_inval_inode_out outarg;
1313 if (size != sizeof(outarg))
1316 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1319 fuse_copy_finish(cs);
1321 down_read(&fc->killsb);
1324 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1325 outarg.off, outarg.len);
1327 up_read(&fc->killsb);
1331 fuse_copy_finish(cs);
1335 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1336 struct fuse_copy_state *cs)
1338 struct fuse_notify_inval_entry_out outarg;
1343 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1348 if (size < sizeof(outarg))
1351 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1355 err = -ENAMETOOLONG;
1356 if (outarg.namelen > FUSE_NAME_MAX)
1360 if (size != sizeof(outarg) + outarg.namelen + 1)
1364 name.len = outarg.namelen;
1365 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1368 fuse_copy_finish(cs);
1369 buf[outarg.namelen] = 0;
1370 name.hash = full_name_hash(name.name, name.len);
1372 down_read(&fc->killsb);
1375 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1376 up_read(&fc->killsb);
1382 fuse_copy_finish(cs);
1386 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1387 struct fuse_copy_state *cs)
1389 struct fuse_notify_delete_out outarg;
1394 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1399 if (size < sizeof(outarg))
1402 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1406 err = -ENAMETOOLONG;
1407 if (outarg.namelen > FUSE_NAME_MAX)
1411 if (size != sizeof(outarg) + outarg.namelen + 1)
1415 name.len = outarg.namelen;
1416 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1419 fuse_copy_finish(cs);
1420 buf[outarg.namelen] = 0;
1421 name.hash = full_name_hash(name.name, name.len);
1423 down_read(&fc->killsb);
1426 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1427 outarg.child, &name);
1428 up_read(&fc->killsb);
1434 fuse_copy_finish(cs);
1438 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1439 struct fuse_copy_state *cs)
1441 struct fuse_notify_store_out outarg;
1442 struct inode *inode;
1443 struct address_space *mapping;
1447 unsigned int offset;
1453 if (size < sizeof(outarg))
1456 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1461 if (size - sizeof(outarg) != outarg.size)
1464 nodeid = outarg.nodeid;
1466 down_read(&fc->killsb);
1472 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1476 mapping = inode->i_mapping;
1477 index = outarg.offset >> PAGE_CACHE_SHIFT;
1478 offset = outarg.offset & ~PAGE_CACHE_MASK;
1479 file_size = i_size_read(inode);
1480 end = outarg.offset + outarg.size;
1481 if (end > file_size) {
1483 fuse_write_update_size(inode, file_size);
1489 unsigned int this_num;
1492 page = find_or_create_page(mapping, index,
1493 mapping_gfp_mask(mapping));
1497 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1498 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1499 if (!err && offset == 0 && (num != 0 || file_size == end))
1500 SetPageUptodate(page);
1502 page_cache_release(page);
1517 up_read(&fc->killsb);
1519 fuse_copy_finish(cs);
1523 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1525 release_pages(req->pages, req->num_pages, 0);
1528 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1529 struct fuse_notify_retrieve_out *outarg)
1532 struct address_space *mapping = inode->i_mapping;
1533 struct fuse_req *req;
1537 unsigned int offset;
1538 size_t total_len = 0;
1540 req = fuse_get_req(fc);
1542 return PTR_ERR(req);
1544 offset = outarg->offset & ~PAGE_CACHE_MASK;
1546 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1547 req->in.h.nodeid = outarg->nodeid;
1548 req->in.numargs = 2;
1549 req->in.argpages = 1;
1550 req->page_offset = offset;
1551 req->end = fuse_retrieve_end;
1553 index = outarg->offset >> PAGE_CACHE_SHIFT;
1554 file_size = i_size_read(inode);
1556 if (outarg->offset > file_size)
1558 else if (outarg->offset + num > file_size)
1559 num = file_size - outarg->offset;
1561 while (num && req->num_pages < FUSE_MAX_PAGES_PER_REQ) {
1563 unsigned int this_num;
1565 page = find_get_page(mapping, index);
1569 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1570 req->pages[req->num_pages] = page;
1575 total_len += this_num;
1578 req->misc.retrieve_in.offset = outarg->offset;
1579 req->misc.retrieve_in.size = total_len;
1580 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1581 req->in.args[0].value = &req->misc.retrieve_in;
1582 req->in.args[1].size = total_len;
1584 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1586 fuse_retrieve_end(fc, req);
1591 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1592 struct fuse_copy_state *cs)
1594 struct fuse_notify_retrieve_out outarg;
1595 struct inode *inode;
1599 if (size != sizeof(outarg))
1602 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1606 fuse_copy_finish(cs);
1608 down_read(&fc->killsb);
1611 u64 nodeid = outarg.nodeid;
1613 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1615 err = fuse_retrieve(fc, inode, &outarg);
1619 up_read(&fc->killsb);
1624 fuse_copy_finish(cs);
1628 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1629 unsigned int size, struct fuse_copy_state *cs)
1632 case FUSE_NOTIFY_POLL:
1633 return fuse_notify_poll(fc, size, cs);
1635 case FUSE_NOTIFY_INVAL_INODE:
1636 return fuse_notify_inval_inode(fc, size, cs);
1638 case FUSE_NOTIFY_INVAL_ENTRY:
1639 return fuse_notify_inval_entry(fc, size, cs);
1641 case FUSE_NOTIFY_STORE:
1642 return fuse_notify_store(fc, size, cs);
1644 case FUSE_NOTIFY_RETRIEVE:
1645 return fuse_notify_retrieve(fc, size, cs);
1647 case FUSE_NOTIFY_DELETE:
1648 return fuse_notify_delete(fc, size, cs);
1651 fuse_copy_finish(cs);
1656 /* Look up request on processing list by unique ID */
1657 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1659 struct list_head *entry;
1661 list_for_each(entry, &fc->processing) {
1662 struct fuse_req *req;
1663 req = list_entry(entry, struct fuse_req, list);
1664 if (req->in.h.unique == unique || req->intr_unique == unique)
1670 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1673 unsigned reqsize = sizeof(struct fuse_out_header);
1676 return nbytes != reqsize ? -EINVAL : 0;
1678 reqsize += len_args(out->numargs, out->args);
1680 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1682 else if (reqsize > nbytes) {
1683 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1684 unsigned diffsize = reqsize - nbytes;
1685 if (diffsize > lastarg->size)
1687 lastarg->size -= diffsize;
1689 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1694 * Write a single reply to a request. First the header is copied from
1695 * the write buffer. The request is then searched on the processing
1696 * list by the unique ID found in the header. If found, then remove
1697 * it from the list and copy the rest of the buffer to the request.
1698 * The request is finished by calling request_end()
1700 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1701 struct fuse_copy_state *cs, size_t nbytes)
1704 struct fuse_req *req;
1705 struct fuse_out_header oh;
1707 if (nbytes < sizeof(struct fuse_out_header))
1710 err = fuse_copy_one(cs, &oh, sizeof(oh));
1715 if (oh.len != nbytes)
1719 * Zero oh.unique indicates unsolicited notification message
1720 * and error contains notification code.
1723 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1724 return err ? err : nbytes;
1728 if (oh.error <= -1000 || oh.error > 0)
1731 spin_lock(&fc->lock);
1736 req = request_find(fc, oh.unique);
1741 spin_unlock(&fc->lock);
1742 fuse_copy_finish(cs);
1743 spin_lock(&fc->lock);
1744 request_end(fc, req);
1747 /* Is it an interrupt reply? */
1748 if (req->intr_unique == oh.unique) {
1750 if (nbytes != sizeof(struct fuse_out_header))
1753 if (oh.error == -ENOSYS)
1754 fc->no_interrupt = 1;
1755 else if (oh.error == -EAGAIN)
1756 queue_interrupt(fc, req);
1758 spin_unlock(&fc->lock);
1759 fuse_copy_finish(cs);
1763 req->state = FUSE_REQ_WRITING;
1764 list_move(&req->list, &fc->io);
1768 if (!req->out.page_replace)
1770 spin_unlock(&fc->lock);
1772 err = copy_out_args(cs, &req->out, nbytes);
1773 fuse_copy_finish(cs);
1775 spin_lock(&fc->lock);
1780 } else if (!req->aborted)
1781 req->out.h.error = -EIO;
1782 request_end(fc, req);
1784 return err ? err : nbytes;
1787 spin_unlock(&fc->lock);
1789 fuse_copy_finish(cs);
1793 static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1794 unsigned long nr_segs, loff_t pos)
1796 struct fuse_copy_state cs;
1797 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1801 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1803 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1806 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1807 struct file *out, loff_t *ppos,
1808 size_t len, unsigned int flags)
1812 struct pipe_buffer *bufs;
1813 struct fuse_copy_state cs;
1814 struct fuse_conn *fc;
1818 fc = fuse_get_conn(out);
1822 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1829 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1830 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1840 struct pipe_buffer *ibuf;
1841 struct pipe_buffer *obuf;
1843 BUG_ON(nbuf >= pipe->buffers);
1844 BUG_ON(!pipe->nrbufs);
1845 ibuf = &pipe->bufs[pipe->curbuf];
1848 if (rem >= ibuf->len) {
1851 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1854 ibuf->ops->get(pipe, ibuf);
1856 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1858 ibuf->offset += obuf->len;
1859 ibuf->len -= obuf->len;
1866 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1870 if (flags & SPLICE_F_MOVE)
1873 ret = fuse_dev_do_write(fc, &cs, len);
1875 for (idx = 0; idx < nbuf; idx++) {
1876 struct pipe_buffer *buf = &bufs[idx];
1877 buf->ops->release(pipe, buf);
1884 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1886 unsigned mask = POLLOUT | POLLWRNORM;
1887 struct fuse_conn *fc = fuse_get_conn(file);
1891 poll_wait(file, &fc->waitq, wait);
1893 spin_lock(&fc->lock);
1896 else if (request_pending(fc))
1897 mask |= POLLIN | POLLRDNORM;
1898 spin_unlock(&fc->lock);
1904 * Abort all requests on the given list (pending or processing)
1906 * This function releases and reacquires fc->lock
1908 static void end_requests(struct fuse_conn *fc, struct list_head *head)
1909 __releases(fc->lock)
1910 __acquires(fc->lock)
1912 while (!list_empty(head)) {
1913 struct fuse_req *req;
1914 req = list_entry(head->next, struct fuse_req, list);
1915 req->out.h.error = -ECONNABORTED;
1916 request_end(fc, req);
1917 spin_lock(&fc->lock);
1922 * Abort requests under I/O
1924 * The requests are set to aborted and finished, and the request
1925 * waiter is woken up. This will make request_wait_answer() wait
1926 * until the request is unlocked and then return.
1928 * If the request is asynchronous, then the end function needs to be
1929 * called after waiting for the request to be unlocked (if it was
1932 static void end_io_requests(struct fuse_conn *fc)
1933 __releases(fc->lock)
1934 __acquires(fc->lock)
1936 while (!list_empty(&fc->io)) {
1937 struct fuse_req *req =
1938 list_entry(fc->io.next, struct fuse_req, list);
1939 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
1942 req->out.h.error = -ECONNABORTED;
1943 req->state = FUSE_REQ_FINISHED;
1944 list_del_init(&req->list);
1945 wake_up(&req->waitq);
1948 __fuse_get_request(req);
1949 spin_unlock(&fc->lock);
1950 wait_event(req->waitq, !req->locked);
1952 fuse_put_request(fc, req);
1953 spin_lock(&fc->lock);
1958 static void end_queued_requests(struct fuse_conn *fc)
1959 __releases(fc->lock)
1960 __acquires(fc->lock)
1962 fc->max_background = UINT_MAX;
1964 end_requests(fc, &fc->pending);
1965 end_requests(fc, &fc->processing);
1966 while (forget_pending(fc))
1967 kfree(dequeue_forget(fc, 1, NULL));
1970 static void end_polls(struct fuse_conn *fc)
1974 p = rb_first(&fc->polled_files);
1977 struct fuse_file *ff;
1978 ff = rb_entry(p, struct fuse_file, polled_node);
1979 wake_up_interruptible_all(&ff->poll_wait);
1986 * Abort all requests.
1988 * Emergency exit in case of a malicious or accidental deadlock, or
1989 * just a hung filesystem.
1991 * The same effect is usually achievable through killing the
1992 * filesystem daemon and all users of the filesystem. The exception
1993 * is the combination of an asynchronous request and the tricky
1994 * deadlock (see Documentation/filesystems/fuse.txt).
1996 * During the aborting, progression of requests from the pending and
1997 * processing lists onto the io list, and progression of new requests
1998 * onto the pending list is prevented by req->connected being false.
2000 * Progression of requests under I/O to the processing list is
2001 * prevented by the req->aborted flag being true for these requests.
2002 * For this reason requests on the io list must be aborted first.
2004 void fuse_abort_conn(struct fuse_conn *fc)
2006 spin_lock(&fc->lock);
2007 if (fc->connected) {
2010 end_io_requests(fc);
2011 end_queued_requests(fc);
2013 wake_up_all(&fc->waitq);
2014 wake_up_all(&fc->blocked_waitq);
2015 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2017 spin_unlock(&fc->lock);
2019 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2021 int fuse_dev_release(struct inode *inode, struct file *file)
2023 struct fuse_conn *fc = fuse_get_conn(file);
2025 spin_lock(&fc->lock);
2028 end_queued_requests(fc);
2030 wake_up_all(&fc->blocked_waitq);
2031 spin_unlock(&fc->lock);
2037 EXPORT_SYMBOL_GPL(fuse_dev_release);
2039 static int fuse_dev_fasync(int fd, struct file *file, int on)
2041 struct fuse_conn *fc = fuse_get_conn(file);
2045 /* No locking - fasync_helper does its own locking */
2046 return fasync_helper(fd, file, on, &fc->fasync);
2049 const struct file_operations fuse_dev_operations = {
2050 .owner = THIS_MODULE,
2051 .llseek = no_llseek,
2052 .read = do_sync_read,
2053 .aio_read = fuse_dev_read,
2054 .splice_read = fuse_dev_splice_read,
2055 .write = do_sync_write,
2056 .aio_write = fuse_dev_write,
2057 .splice_write = fuse_dev_splice_write,
2058 .poll = fuse_dev_poll,
2059 .release = fuse_dev_release,
2060 .fasync = fuse_dev_fasync,
2062 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2064 static struct miscdevice fuse_miscdevice = {
2065 .minor = FUSE_MINOR,
2067 .fops = &fuse_dev_operations,
2070 int __init fuse_dev_init(void)
2073 fuse_req_cachep = kmem_cache_create("fuse_request",
2074 sizeof(struct fuse_req),
2076 if (!fuse_req_cachep)
2079 err = misc_register(&fuse_miscdevice);
2081 goto out_cache_clean;
2086 kmem_cache_destroy(fuse_req_cachep);
2091 void fuse_dev_cleanup(void)
2093 misc_deregister(&fuse_miscdevice);
2094 kmem_cache_destroy(fuse_req_cachep);