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 = current_fsuid();
96 req->in.h.gid = 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;
695 struct address_space *mapping;
698 unlock_request(cs->fc, cs->req);
699 fuse_copy_finish(cs);
701 err = buf->ops->confirm(cs->pipe, buf);
705 BUG_ON(!cs->nr_segs);
711 if (cs->len != PAGE_SIZE)
714 if (buf->ops->steal(cs->pipe, buf) != 0)
719 if (!PageUptodate(newpage))
720 SetPageUptodate(newpage);
722 ClearPageMappedToDisk(newpage);
724 if (fuse_check_page(newpage) != 0)
725 goto out_fallback_unlock;
727 mapping = oldpage->mapping;
728 index = oldpage->index;
731 * This is a new and locked page, it shouldn't be mapped or
732 * have any special flags on it
734 if (WARN_ON(page_mapped(oldpage)))
735 goto out_fallback_unlock;
736 if (WARN_ON(page_has_private(oldpage)))
737 goto out_fallback_unlock;
738 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
739 goto out_fallback_unlock;
740 if (WARN_ON(PageMlocked(oldpage)))
741 goto out_fallback_unlock;
743 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
745 unlock_page(newpage);
749 page_cache_get(newpage);
751 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
752 lru_cache_add_file(newpage);
755 spin_lock(&cs->fc->lock);
756 if (cs->req->aborted)
760 spin_unlock(&cs->fc->lock);
763 unlock_page(newpage);
764 page_cache_release(newpage);
768 unlock_page(oldpage);
769 page_cache_release(oldpage);
775 unlock_page(newpage);
777 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
778 cs->buf = cs->mapaddr + buf->offset;
780 err = lock_request(cs->fc, cs->req);
787 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
788 unsigned offset, unsigned count)
790 struct pipe_buffer *buf;
792 if (cs->nr_segs == cs->pipe->buffers)
795 unlock_request(cs->fc, cs->req);
796 fuse_copy_finish(cs);
799 page_cache_get(page);
801 buf->offset = offset;
812 * Copy a page in the request to/from the userspace buffer. Must be
815 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
816 unsigned offset, unsigned count, int zeroing)
819 struct page *page = *pagep;
821 if (page && zeroing && count < PAGE_SIZE)
822 clear_highpage(page);
825 if (cs->write && cs->pipebufs && page) {
826 return fuse_ref_page(cs, page, offset, count);
827 } else if (!cs->len) {
828 if (cs->move_pages && page &&
829 offset == 0 && count == PAGE_SIZE) {
830 err = fuse_try_move_page(cs, pagep);
834 err = fuse_copy_fill(cs);
840 void *mapaddr = kmap_atomic(page, KM_USER0);
841 void *buf = mapaddr + offset;
842 offset += fuse_copy_do(cs, &buf, &count);
843 kunmap_atomic(mapaddr, KM_USER0);
845 offset += fuse_copy_do(cs, NULL, &count);
847 if (page && !cs->write)
848 flush_dcache_page(page);
852 /* Copy pages in the request to/from userspace buffer */
853 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
857 struct fuse_req *req = cs->req;
858 unsigned offset = req->page_offset;
859 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
861 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
864 err = fuse_copy_page(cs, &req->pages[i], offset, count,
870 count = min(nbytes, (unsigned) PAGE_SIZE);
876 /* Copy a single argument in the request to/from userspace buffer */
877 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
881 int err = fuse_copy_fill(cs);
885 fuse_copy_do(cs, &val, &size);
890 /* Copy request arguments to/from userspace buffer */
891 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
892 unsigned argpages, struct fuse_arg *args,
898 for (i = 0; !err && i < numargs; i++) {
899 struct fuse_arg *arg = &args[i];
900 if (i == numargs - 1 && argpages)
901 err = fuse_copy_pages(cs, arg->size, zeroing);
903 err = fuse_copy_one(cs, arg->value, arg->size);
908 static int forget_pending(struct fuse_conn *fc)
910 return fc->forget_list_head.next != NULL;
913 static int request_pending(struct fuse_conn *fc)
915 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
919 /* Wait until a request is available on the pending list */
920 static void request_wait(struct fuse_conn *fc)
924 DECLARE_WAITQUEUE(wait, current);
926 add_wait_queue_exclusive(&fc->waitq, &wait);
927 while (fc->connected && !request_pending(fc)) {
928 set_current_state(TASK_INTERRUPTIBLE);
929 if (signal_pending(current))
932 spin_unlock(&fc->lock);
934 spin_lock(&fc->lock);
936 set_current_state(TASK_RUNNING);
937 remove_wait_queue(&fc->waitq, &wait);
941 * Transfer an interrupt request to userspace
943 * Unlike other requests this is assembled on demand, without a need
944 * to allocate a separate fuse_req structure.
946 * Called with fc->lock held, releases it
948 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
949 size_t nbytes, struct fuse_req *req)
952 struct fuse_in_header ih;
953 struct fuse_interrupt_in arg;
954 unsigned reqsize = sizeof(ih) + sizeof(arg);
957 list_del_init(&req->intr_entry);
958 req->intr_unique = fuse_get_unique(fc);
959 memset(&ih, 0, sizeof(ih));
960 memset(&arg, 0, sizeof(arg));
962 ih.opcode = FUSE_INTERRUPT;
963 ih.unique = req->intr_unique;
964 arg.unique = req->in.h.unique;
966 spin_unlock(&fc->lock);
967 if (nbytes < reqsize)
970 err = fuse_copy_one(cs, &ih, sizeof(ih));
972 err = fuse_copy_one(cs, &arg, sizeof(arg));
973 fuse_copy_finish(cs);
975 return err ? err : reqsize;
978 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
982 struct fuse_forget_link *head = fc->forget_list_head.next;
983 struct fuse_forget_link **newhead = &head;
986 for (count = 0; *newhead != NULL && count < max; count++)
987 newhead = &(*newhead)->next;
989 fc->forget_list_head.next = *newhead;
991 if (fc->forget_list_head.next == NULL)
992 fc->forget_list_tail = &fc->forget_list_head;
1000 static int fuse_read_single_forget(struct fuse_conn *fc,
1001 struct fuse_copy_state *cs,
1003 __releases(fc->lock)
1006 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1007 struct fuse_forget_in arg = {
1008 .nlookup = forget->forget_one.nlookup,
1010 struct fuse_in_header ih = {
1011 .opcode = FUSE_FORGET,
1012 .nodeid = forget->forget_one.nodeid,
1013 .unique = fuse_get_unique(fc),
1014 .len = sizeof(ih) + sizeof(arg),
1017 spin_unlock(&fc->lock);
1019 if (nbytes < ih.len)
1022 err = fuse_copy_one(cs, &ih, sizeof(ih));
1024 err = fuse_copy_one(cs, &arg, sizeof(arg));
1025 fuse_copy_finish(cs);
1033 static int fuse_read_batch_forget(struct fuse_conn *fc,
1034 struct fuse_copy_state *cs, size_t nbytes)
1035 __releases(fc->lock)
1038 unsigned max_forgets;
1040 struct fuse_forget_link *head;
1041 struct fuse_batch_forget_in arg = { .count = 0 };
1042 struct fuse_in_header ih = {
1043 .opcode = FUSE_BATCH_FORGET,
1044 .unique = fuse_get_unique(fc),
1045 .len = sizeof(ih) + sizeof(arg),
1048 if (nbytes < ih.len) {
1049 spin_unlock(&fc->lock);
1053 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1054 head = dequeue_forget(fc, max_forgets, &count);
1055 spin_unlock(&fc->lock);
1058 ih.len += count * sizeof(struct fuse_forget_one);
1059 err = fuse_copy_one(cs, &ih, sizeof(ih));
1061 err = fuse_copy_one(cs, &arg, sizeof(arg));
1064 struct fuse_forget_link *forget = head;
1067 err = fuse_copy_one(cs, &forget->forget_one,
1068 sizeof(forget->forget_one));
1070 head = forget->next;
1074 fuse_copy_finish(cs);
1082 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1084 __releases(fc->lock)
1086 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1087 return fuse_read_single_forget(fc, cs, nbytes);
1089 return fuse_read_batch_forget(fc, cs, nbytes);
1093 * Read a single request into the userspace filesystem's buffer. This
1094 * function waits until a request is available, then removes it from
1095 * the pending list and copies request data to userspace buffer. If
1096 * no reply is needed (FORGET) or request has been aborted or there
1097 * was an error during the copying then it's finished by calling
1098 * request_end(). Otherwise add it to the processing list, and set
1101 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1102 struct fuse_copy_state *cs, size_t nbytes)
1105 struct fuse_req *req;
1110 spin_lock(&fc->lock);
1112 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1113 !request_pending(fc))
1121 if (!request_pending(fc))
1124 if (!list_empty(&fc->interrupts)) {
1125 req = list_entry(fc->interrupts.next, struct fuse_req,
1127 return fuse_read_interrupt(fc, cs, nbytes, req);
1130 if (forget_pending(fc)) {
1131 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1132 return fuse_read_forget(fc, cs, nbytes);
1134 if (fc->forget_batch <= -8)
1135 fc->forget_batch = 16;
1138 req = list_entry(fc->pending.next, struct fuse_req, list);
1139 req->state = FUSE_REQ_READING;
1140 list_move(&req->list, &fc->io);
1143 reqsize = in->h.len;
1144 /* If request is too large, reply with an error and restart the read */
1145 if (nbytes < reqsize) {
1146 req->out.h.error = -EIO;
1147 /* SETXATTR is special, since it may contain too large data */
1148 if (in->h.opcode == FUSE_SETXATTR)
1149 req->out.h.error = -E2BIG;
1150 request_end(fc, req);
1153 spin_unlock(&fc->lock);
1155 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1157 err = fuse_copy_args(cs, in->numargs, in->argpages,
1158 (struct fuse_arg *) in->args, 0);
1159 fuse_copy_finish(cs);
1160 spin_lock(&fc->lock);
1163 request_end(fc, req);
1167 req->out.h.error = -EIO;
1168 request_end(fc, req);
1172 request_end(fc, req);
1174 req->state = FUSE_REQ_SENT;
1175 list_move_tail(&req->list, &fc->processing);
1176 if (req->interrupted)
1177 queue_interrupt(fc, req);
1178 spin_unlock(&fc->lock);
1183 spin_unlock(&fc->lock);
1187 static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1188 unsigned long nr_segs, loff_t pos)
1190 struct fuse_copy_state cs;
1191 struct file *file = iocb->ki_filp;
1192 struct fuse_conn *fc = fuse_get_conn(file);
1196 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1198 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1201 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1202 struct pipe_inode_info *pipe,
1203 size_t len, unsigned int flags)
1208 struct pipe_buffer *bufs;
1209 struct fuse_copy_state cs;
1210 struct fuse_conn *fc = fuse_get_conn(in);
1214 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1218 fuse_copy_init(&cs, fc, 1, NULL, 0);
1221 ret = fuse_dev_do_read(fc, in, &cs, len);
1228 if (!pipe->readers) {
1229 send_sig(SIGPIPE, current, 0);
1235 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1240 while (page_nr < cs.nr_segs) {
1241 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1242 struct pipe_buffer *buf = pipe->bufs + newbuf;
1244 buf->page = bufs[page_nr].page;
1245 buf->offset = bufs[page_nr].offset;
1246 buf->len = bufs[page_nr].len;
1248 * Need to be careful about this. Having buf->ops in module
1249 * code can Oops if the buffer persists after module unload.
1251 buf->ops = &nosteal_pipe_buf_ops;
1266 if (waitqueue_active(&pipe->wait))
1267 wake_up_interruptible(&pipe->wait);
1268 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1272 for (; page_nr < cs.nr_segs; page_nr++)
1273 page_cache_release(bufs[page_nr].page);
1279 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1280 struct fuse_copy_state *cs)
1282 struct fuse_notify_poll_wakeup_out outarg;
1285 if (size != sizeof(outarg))
1288 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1292 fuse_copy_finish(cs);
1293 return fuse_notify_poll_wakeup(fc, &outarg);
1296 fuse_copy_finish(cs);
1300 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1301 struct fuse_copy_state *cs)
1303 struct fuse_notify_inval_inode_out outarg;
1306 if (size != sizeof(outarg))
1309 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1312 fuse_copy_finish(cs);
1314 down_read(&fc->killsb);
1317 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1318 outarg.off, outarg.len);
1320 up_read(&fc->killsb);
1324 fuse_copy_finish(cs);
1328 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1329 struct fuse_copy_state *cs)
1331 struct fuse_notify_inval_entry_out outarg;
1336 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1341 if (size < sizeof(outarg))
1344 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1348 err = -ENAMETOOLONG;
1349 if (outarg.namelen > FUSE_NAME_MAX)
1353 if (size != sizeof(outarg) + outarg.namelen + 1)
1357 name.len = outarg.namelen;
1358 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1361 fuse_copy_finish(cs);
1362 buf[outarg.namelen] = 0;
1363 name.hash = full_name_hash(name.name, name.len);
1365 down_read(&fc->killsb);
1368 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, &name);
1369 up_read(&fc->killsb);
1375 fuse_copy_finish(cs);
1379 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1380 struct fuse_copy_state *cs)
1382 struct fuse_notify_store_out outarg;
1383 struct inode *inode;
1384 struct address_space *mapping;
1388 unsigned int offset;
1394 if (size < sizeof(outarg))
1397 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1402 if (size - sizeof(outarg) != outarg.size)
1405 nodeid = outarg.nodeid;
1407 down_read(&fc->killsb);
1413 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1417 mapping = inode->i_mapping;
1418 index = outarg.offset >> PAGE_CACHE_SHIFT;
1419 offset = outarg.offset & ~PAGE_CACHE_MASK;
1420 file_size = i_size_read(inode);
1421 end = outarg.offset + outarg.size;
1422 if (end > file_size) {
1424 fuse_write_update_size(inode, file_size);
1430 unsigned int this_num;
1433 page = find_or_create_page(mapping, index,
1434 mapping_gfp_mask(mapping));
1438 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1439 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1440 if (!err && offset == 0 && (num != 0 || file_size == end))
1441 SetPageUptodate(page);
1443 page_cache_release(page);
1458 up_read(&fc->killsb);
1460 fuse_copy_finish(cs);
1464 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1466 release_pages(req->pages, req->num_pages, 0);
1469 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1470 struct fuse_notify_retrieve_out *outarg)
1473 struct address_space *mapping = inode->i_mapping;
1474 struct fuse_req *req;
1478 unsigned int offset;
1479 size_t total_len = 0;
1481 req = fuse_get_req(fc);
1483 return PTR_ERR(req);
1485 offset = outarg->offset & ~PAGE_CACHE_MASK;
1487 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1488 req->in.h.nodeid = outarg->nodeid;
1489 req->in.numargs = 2;
1490 req->in.argpages = 1;
1491 req->page_offset = offset;
1492 req->end = fuse_retrieve_end;
1494 index = outarg->offset >> PAGE_CACHE_SHIFT;
1495 file_size = i_size_read(inode);
1497 if (outarg->offset > file_size)
1499 else if (outarg->offset + num > file_size)
1500 num = file_size - outarg->offset;
1502 while (num && req->num_pages < FUSE_MAX_PAGES_PER_REQ) {
1504 unsigned int this_num;
1506 page = find_get_page(mapping, index);
1510 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1511 req->pages[req->num_pages] = page;
1516 total_len += this_num;
1519 req->misc.retrieve_in.offset = outarg->offset;
1520 req->misc.retrieve_in.size = total_len;
1521 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1522 req->in.args[0].value = &req->misc.retrieve_in;
1523 req->in.args[1].size = total_len;
1525 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1527 fuse_retrieve_end(fc, req);
1532 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1533 struct fuse_copy_state *cs)
1535 struct fuse_notify_retrieve_out outarg;
1536 struct inode *inode;
1540 if (size != sizeof(outarg))
1543 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1547 fuse_copy_finish(cs);
1549 down_read(&fc->killsb);
1552 u64 nodeid = outarg.nodeid;
1554 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1556 err = fuse_retrieve(fc, inode, &outarg);
1560 up_read(&fc->killsb);
1565 fuse_copy_finish(cs);
1569 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1570 unsigned int size, struct fuse_copy_state *cs)
1572 /* Don't try to move pages (yet) */
1576 case FUSE_NOTIFY_POLL:
1577 return fuse_notify_poll(fc, size, cs);
1579 case FUSE_NOTIFY_INVAL_INODE:
1580 return fuse_notify_inval_inode(fc, size, cs);
1582 case FUSE_NOTIFY_INVAL_ENTRY:
1583 return fuse_notify_inval_entry(fc, size, cs);
1585 case FUSE_NOTIFY_STORE:
1586 return fuse_notify_store(fc, size, cs);
1588 case FUSE_NOTIFY_RETRIEVE:
1589 return fuse_notify_retrieve(fc, size, cs);
1592 fuse_copy_finish(cs);
1597 /* Look up request on processing list by unique ID */
1598 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1600 struct list_head *entry;
1602 list_for_each(entry, &fc->processing) {
1603 struct fuse_req *req;
1604 req = list_entry(entry, struct fuse_req, list);
1605 if (req->in.h.unique == unique || req->intr_unique == unique)
1611 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1614 unsigned reqsize = sizeof(struct fuse_out_header);
1617 return nbytes != reqsize ? -EINVAL : 0;
1619 reqsize += len_args(out->numargs, out->args);
1621 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1623 else if (reqsize > nbytes) {
1624 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1625 unsigned diffsize = reqsize - nbytes;
1626 if (diffsize > lastarg->size)
1628 lastarg->size -= diffsize;
1630 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1635 * Write a single reply to a request. First the header is copied from
1636 * the write buffer. The request is then searched on the processing
1637 * list by the unique ID found in the header. If found, then remove
1638 * it from the list and copy the rest of the buffer to the request.
1639 * The request is finished by calling request_end()
1641 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1642 struct fuse_copy_state *cs, size_t nbytes)
1645 struct fuse_req *req;
1646 struct fuse_out_header oh;
1648 if (nbytes < sizeof(struct fuse_out_header))
1651 err = fuse_copy_one(cs, &oh, sizeof(oh));
1656 if (oh.len != nbytes)
1660 * Zero oh.unique indicates unsolicited notification message
1661 * and error contains notification code.
1664 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1665 return err ? err : nbytes;
1669 if (oh.error <= -1000 || oh.error > 0)
1672 spin_lock(&fc->lock);
1677 req = request_find(fc, oh.unique);
1682 spin_unlock(&fc->lock);
1683 fuse_copy_finish(cs);
1684 spin_lock(&fc->lock);
1685 request_end(fc, req);
1688 /* Is it an interrupt reply? */
1689 if (req->intr_unique == oh.unique) {
1691 if (nbytes != sizeof(struct fuse_out_header))
1694 if (oh.error == -ENOSYS)
1695 fc->no_interrupt = 1;
1696 else if (oh.error == -EAGAIN)
1697 queue_interrupt(fc, req);
1699 spin_unlock(&fc->lock);
1700 fuse_copy_finish(cs);
1704 req->state = FUSE_REQ_WRITING;
1705 list_move(&req->list, &fc->io);
1709 if (!req->out.page_replace)
1711 spin_unlock(&fc->lock);
1713 err = copy_out_args(cs, &req->out, nbytes);
1714 fuse_copy_finish(cs);
1716 spin_lock(&fc->lock);
1721 } else if (!req->aborted)
1722 req->out.h.error = -EIO;
1723 request_end(fc, req);
1725 return err ? err : nbytes;
1728 spin_unlock(&fc->lock);
1730 fuse_copy_finish(cs);
1734 static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1735 unsigned long nr_segs, loff_t pos)
1737 struct fuse_copy_state cs;
1738 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1742 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1744 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1747 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1748 struct file *out, loff_t *ppos,
1749 size_t len, unsigned int flags)
1753 struct pipe_buffer *bufs;
1754 struct fuse_copy_state cs;
1755 struct fuse_conn *fc;
1759 fc = fuse_get_conn(out);
1763 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1770 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1771 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1781 struct pipe_buffer *ibuf;
1782 struct pipe_buffer *obuf;
1784 BUG_ON(nbuf >= pipe->buffers);
1785 BUG_ON(!pipe->nrbufs);
1786 ibuf = &pipe->bufs[pipe->curbuf];
1789 if (rem >= ibuf->len) {
1792 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1795 ibuf->ops->get(pipe, ibuf);
1797 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1799 ibuf->offset += obuf->len;
1800 ibuf->len -= obuf->len;
1807 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1811 if (flags & SPLICE_F_MOVE)
1814 ret = fuse_dev_do_write(fc, &cs, len);
1816 for (idx = 0; idx < nbuf; idx++) {
1817 struct pipe_buffer *buf = &bufs[idx];
1818 buf->ops->release(pipe, buf);
1825 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1827 unsigned mask = POLLOUT | POLLWRNORM;
1828 struct fuse_conn *fc = fuse_get_conn(file);
1832 poll_wait(file, &fc->waitq, wait);
1834 spin_lock(&fc->lock);
1837 else if (request_pending(fc))
1838 mask |= POLLIN | POLLRDNORM;
1839 spin_unlock(&fc->lock);
1845 * Abort all requests on the given list (pending or processing)
1847 * This function releases and reacquires fc->lock
1849 static void end_requests(struct fuse_conn *fc, struct list_head *head)
1850 __releases(fc->lock)
1851 __acquires(fc->lock)
1853 while (!list_empty(head)) {
1854 struct fuse_req *req;
1855 req = list_entry(head->next, struct fuse_req, list);
1856 req->out.h.error = -ECONNABORTED;
1857 request_end(fc, req);
1858 spin_lock(&fc->lock);
1863 * Abort requests under I/O
1865 * The requests are set to aborted and finished, and the request
1866 * waiter is woken up. This will make request_wait_answer() wait
1867 * until the request is unlocked and then return.
1869 * If the request is asynchronous, then the end function needs to be
1870 * called after waiting for the request to be unlocked (if it was
1873 static void end_io_requests(struct fuse_conn *fc)
1874 __releases(fc->lock)
1875 __acquires(fc->lock)
1877 while (!list_empty(&fc->io)) {
1878 struct fuse_req *req =
1879 list_entry(fc->io.next, struct fuse_req, list);
1880 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
1883 req->out.h.error = -ECONNABORTED;
1884 req->state = FUSE_REQ_FINISHED;
1885 list_del_init(&req->list);
1886 wake_up(&req->waitq);
1889 __fuse_get_request(req);
1890 spin_unlock(&fc->lock);
1891 wait_event(req->waitq, !req->locked);
1893 fuse_put_request(fc, req);
1894 spin_lock(&fc->lock);
1899 static void end_queued_requests(struct fuse_conn *fc)
1900 __releases(fc->lock)
1901 __acquires(fc->lock)
1903 fc->max_background = UINT_MAX;
1905 end_requests(fc, &fc->pending);
1906 end_requests(fc, &fc->processing);
1907 while (forget_pending(fc))
1908 kfree(dequeue_forget(fc, 1, NULL));
1911 static void end_polls(struct fuse_conn *fc)
1915 p = rb_first(&fc->polled_files);
1918 struct fuse_file *ff;
1919 ff = rb_entry(p, struct fuse_file, polled_node);
1920 wake_up_interruptible_all(&ff->poll_wait);
1927 * Abort all requests.
1929 * Emergency exit in case of a malicious or accidental deadlock, or
1930 * just a hung filesystem.
1932 * The same effect is usually achievable through killing the
1933 * filesystem daemon and all users of the filesystem. The exception
1934 * is the combination of an asynchronous request and the tricky
1935 * deadlock (see Documentation/filesystems/fuse.txt).
1937 * During the aborting, progression of requests from the pending and
1938 * processing lists onto the io list, and progression of new requests
1939 * onto the pending list is prevented by req->connected being false.
1941 * Progression of requests under I/O to the processing list is
1942 * prevented by the req->aborted flag being true for these requests.
1943 * For this reason requests on the io list must be aborted first.
1945 void fuse_abort_conn(struct fuse_conn *fc)
1947 spin_lock(&fc->lock);
1948 if (fc->connected) {
1951 end_io_requests(fc);
1952 end_queued_requests(fc);
1954 wake_up_all(&fc->waitq);
1955 wake_up_all(&fc->blocked_waitq);
1956 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
1958 spin_unlock(&fc->lock);
1960 EXPORT_SYMBOL_GPL(fuse_abort_conn);
1962 int fuse_dev_release(struct inode *inode, struct file *file)
1964 struct fuse_conn *fc = fuse_get_conn(file);
1966 spin_lock(&fc->lock);
1969 end_queued_requests(fc);
1971 wake_up_all(&fc->blocked_waitq);
1972 spin_unlock(&fc->lock);
1978 EXPORT_SYMBOL_GPL(fuse_dev_release);
1980 static int fuse_dev_fasync(int fd, struct file *file, int on)
1982 struct fuse_conn *fc = fuse_get_conn(file);
1986 /* No locking - fasync_helper does its own locking */
1987 return fasync_helper(fd, file, on, &fc->fasync);
1990 const struct file_operations fuse_dev_operations = {
1991 .owner = THIS_MODULE,
1992 .llseek = no_llseek,
1993 .read = do_sync_read,
1994 .aio_read = fuse_dev_read,
1995 .splice_read = fuse_dev_splice_read,
1996 .write = do_sync_write,
1997 .aio_write = fuse_dev_write,
1998 .splice_write = fuse_dev_splice_write,
1999 .poll = fuse_dev_poll,
2000 .release = fuse_dev_release,
2001 .fasync = fuse_dev_fasync,
2003 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2005 static struct miscdevice fuse_miscdevice = {
2006 .minor = FUSE_MINOR,
2008 .fops = &fuse_dev_operations,
2011 int __init fuse_dev_init(void)
2014 fuse_req_cachep = kmem_cache_create("fuse_request",
2015 sizeof(struct fuse_req),
2017 if (!fuse_req_cachep)
2020 err = misc_register(&fuse_miscdevice);
2022 goto out_cache_clean;
2027 kmem_cache_destroy(fuse_req_cachep);
2032 void fuse_dev_cleanup(void)
2034 misc_deregister(&fuse_miscdevice);
2035 kmem_cache_destroy(fuse_req_cachep);