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/sched/signal.h>
15 #include <linux/uio.h>
16 #include <linux/miscdevice.h>
17 #include <linux/pagemap.h>
18 #include <linux/file.h>
19 #include <linux/slab.h>
20 #include <linux/pipe_fs_i.h>
21 #include <linux/swap.h>
22 #include <linux/splice.h>
23 #include <linux/sched.h>
25 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26 MODULE_ALIAS("devname:fuse");
28 /* Ordinary requests have even IDs, while interrupts IDs are odd */
29 #define FUSE_INT_REQ_BIT (1ULL << 0)
30 #define FUSE_REQ_ID_STEP (1ULL << 1)
32 static struct kmem_cache *fuse_req_cachep;
34 static struct fuse_dev *fuse_get_dev(struct file *file)
37 * Lockless access is OK, because file->private data is set
38 * once during mount and is valid until the file is released.
40 return READ_ONCE(file->private_data);
43 static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)
45 INIT_LIST_HEAD(&req->list);
46 INIT_LIST_HEAD(&req->intr_entry);
47 init_waitqueue_head(&req->waitq);
48 refcount_set(&req->count, 1);
49 __set_bit(FR_PENDING, &req->flags);
53 static struct fuse_req *fuse_request_alloc(struct fuse_mount *fm, gfp_t flags)
55 struct fuse_req *req = kmem_cache_zalloc(fuse_req_cachep, flags);
57 fuse_request_init(fm, req);
62 static void fuse_request_free(struct fuse_req *req)
64 kmem_cache_free(fuse_req_cachep, req);
67 static void __fuse_get_request(struct fuse_req *req)
69 refcount_inc(&req->count);
72 /* Must be called with > 1 refcount */
73 static void __fuse_put_request(struct fuse_req *req)
75 refcount_dec(&req->count);
78 void fuse_set_initialized(struct fuse_conn *fc)
80 /* Make sure stores before this are seen on another CPU */
85 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
87 return !fc->initialized || (for_background && fc->blocked);
90 static void fuse_drop_waiting(struct fuse_conn *fc)
93 * lockess check of fc->connected is okay, because atomic_dec_and_test()
94 * provides a memory barrier matched with the one in fuse_wait_aborted()
95 * to ensure no wake-up is missed.
97 if (atomic_dec_and_test(&fc->num_waiting) &&
98 !READ_ONCE(fc->connected)) {
99 /* wake up aborters */
100 wake_up_all(&fc->blocked_waitq);
104 static void fuse_put_request(struct fuse_req *req);
106 static struct fuse_req *fuse_get_req(struct fuse_mount *fm, bool for_background)
108 struct fuse_conn *fc = fm->fc;
109 struct fuse_req *req;
111 atomic_inc(&fc->num_waiting);
113 if (fuse_block_alloc(fc, for_background)) {
115 if (wait_event_killable_exclusive(fc->blocked_waitq,
116 !fuse_block_alloc(fc, for_background)))
119 /* Matches smp_wmb() in fuse_set_initialized() */
130 req = fuse_request_alloc(fm, GFP_KERNEL);
134 wake_up(&fc->blocked_waitq);
138 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid());
139 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid());
140 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
142 __set_bit(FR_WAITING, &req->flags);
144 __set_bit(FR_BACKGROUND, &req->flags);
146 if (unlikely(req->in.h.uid == ((uid_t)-1) ||
147 req->in.h.gid == ((gid_t)-1))) {
148 fuse_put_request(req);
149 return ERR_PTR(-EOVERFLOW);
154 fuse_drop_waiting(fc);
158 static void fuse_put_request(struct fuse_req *req)
160 struct fuse_conn *fc = req->fm->fc;
162 if (refcount_dec_and_test(&req->count)) {
163 if (test_bit(FR_BACKGROUND, &req->flags)) {
165 * We get here in the unlikely case that a background
166 * request was allocated but not sent
168 spin_lock(&fc->bg_lock);
170 wake_up(&fc->blocked_waitq);
171 spin_unlock(&fc->bg_lock);
174 if (test_bit(FR_WAITING, &req->flags)) {
175 __clear_bit(FR_WAITING, &req->flags);
176 fuse_drop_waiting(fc);
179 fuse_request_free(req);
183 unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args)
188 for (i = 0; i < numargs; i++)
189 nbytes += args[i].size;
193 EXPORT_SYMBOL_GPL(fuse_len_args);
195 u64 fuse_get_unique(struct fuse_iqueue *fiq)
197 fiq->reqctr += FUSE_REQ_ID_STEP;
200 EXPORT_SYMBOL_GPL(fuse_get_unique);
202 static unsigned int fuse_req_hash(u64 unique)
204 return hash_long(unique & ~FUSE_INT_REQ_BIT, FUSE_PQ_HASH_BITS);
208 * A new request is available, wake fiq->waitq
210 static void fuse_dev_wake_and_unlock(struct fuse_iqueue *fiq)
211 __releases(fiq->lock)
213 wake_up(&fiq->waitq);
214 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
215 spin_unlock(&fiq->lock);
218 const struct fuse_iqueue_ops fuse_dev_fiq_ops = {
219 .wake_forget_and_unlock = fuse_dev_wake_and_unlock,
220 .wake_interrupt_and_unlock = fuse_dev_wake_and_unlock,
221 .wake_pending_and_unlock = fuse_dev_wake_and_unlock,
223 EXPORT_SYMBOL_GPL(fuse_dev_fiq_ops);
225 static void queue_request_and_unlock(struct fuse_iqueue *fiq,
226 struct fuse_req *req)
227 __releases(fiq->lock)
229 req->in.h.len = sizeof(struct fuse_in_header) +
230 fuse_len_args(req->args->in_numargs,
231 (struct fuse_arg *) req->args->in_args);
232 list_add_tail(&req->list, &fiq->pending);
233 fiq->ops->wake_pending_and_unlock(fiq);
236 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
237 u64 nodeid, u64 nlookup)
239 struct fuse_iqueue *fiq = &fc->iq;
241 forget->forget_one.nodeid = nodeid;
242 forget->forget_one.nlookup = nlookup;
244 spin_lock(&fiq->lock);
245 if (fiq->connected) {
246 fiq->forget_list_tail->next = forget;
247 fiq->forget_list_tail = forget;
248 fiq->ops->wake_forget_and_unlock(fiq);
251 spin_unlock(&fiq->lock);
255 static void flush_bg_queue(struct fuse_conn *fc)
257 struct fuse_iqueue *fiq = &fc->iq;
259 while (fc->active_background < fc->max_background &&
260 !list_empty(&fc->bg_queue)) {
261 struct fuse_req *req;
263 req = list_first_entry(&fc->bg_queue, struct fuse_req, list);
264 list_del(&req->list);
265 fc->active_background++;
266 spin_lock(&fiq->lock);
267 req->in.h.unique = fuse_get_unique(fiq);
268 queue_request_and_unlock(fiq, req);
273 * This function is called when a request is finished. Either a reply
274 * has arrived or it was aborted (and not yet sent) or some error
275 * occurred during communication with userspace, or the device file
276 * was closed. The requester thread is woken up (if still waiting),
277 * the 'end' callback is called if given, else the reference to the
278 * request is released
280 void fuse_request_end(struct fuse_req *req)
282 struct fuse_mount *fm = req->fm;
283 struct fuse_conn *fc = fm->fc;
284 struct fuse_iqueue *fiq = &fc->iq;
286 if (test_and_set_bit(FR_FINISHED, &req->flags))
290 * test_and_set_bit() implies smp_mb() between bit
291 * changing and below FR_INTERRUPTED check. Pairs with
292 * smp_mb() from queue_interrupt().
294 if (test_bit(FR_INTERRUPTED, &req->flags)) {
295 spin_lock(&fiq->lock);
296 list_del_init(&req->intr_entry);
297 spin_unlock(&fiq->lock);
299 WARN_ON(test_bit(FR_PENDING, &req->flags));
300 WARN_ON(test_bit(FR_SENT, &req->flags));
301 if (test_bit(FR_BACKGROUND, &req->flags)) {
302 spin_lock(&fc->bg_lock);
303 clear_bit(FR_BACKGROUND, &req->flags);
304 if (fc->num_background == fc->max_background) {
306 wake_up(&fc->blocked_waitq);
307 } else if (!fc->blocked) {
309 * Wake up next waiter, if any. It's okay to use
310 * waitqueue_active(), as we've already synced up
311 * fc->blocked with waiters with the wake_up() call
314 if (waitqueue_active(&fc->blocked_waitq))
315 wake_up(&fc->blocked_waitq);
318 fc->num_background--;
319 fc->active_background--;
321 spin_unlock(&fc->bg_lock);
323 /* Wake up waiter sleeping in request_wait_answer() */
324 wake_up(&req->waitq);
327 if (test_bit(FR_ASYNC, &req->flags))
328 req->args->end(fm, req->args, req->out.h.error);
330 fuse_put_request(req);
332 EXPORT_SYMBOL_GPL(fuse_request_end);
334 static int queue_interrupt(struct fuse_req *req)
336 struct fuse_iqueue *fiq = &req->fm->fc->iq;
338 spin_lock(&fiq->lock);
339 /* Check for we've sent request to interrupt this req */
340 if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) {
341 spin_unlock(&fiq->lock);
345 if (list_empty(&req->intr_entry)) {
346 list_add_tail(&req->intr_entry, &fiq->interrupts);
348 * Pairs with smp_mb() implied by test_and_set_bit()
349 * from fuse_request_end().
352 if (test_bit(FR_FINISHED, &req->flags)) {
353 list_del_init(&req->intr_entry);
354 spin_unlock(&fiq->lock);
357 fiq->ops->wake_interrupt_and_unlock(fiq);
359 spin_unlock(&fiq->lock);
364 static void request_wait_answer(struct fuse_req *req)
366 struct fuse_conn *fc = req->fm->fc;
367 struct fuse_iqueue *fiq = &fc->iq;
370 if (!fc->no_interrupt) {
371 /* Any signal may interrupt this */
372 err = wait_event_interruptible(req->waitq,
373 test_bit(FR_FINISHED, &req->flags));
377 set_bit(FR_INTERRUPTED, &req->flags);
378 /* matches barrier in fuse_dev_do_read() */
379 smp_mb__after_atomic();
380 if (test_bit(FR_SENT, &req->flags))
381 queue_interrupt(req);
384 if (!test_bit(FR_FORCE, &req->flags)) {
385 /* Only fatal signals may interrupt this */
386 err = wait_event_killable(req->waitq,
387 test_bit(FR_FINISHED, &req->flags));
391 spin_lock(&fiq->lock);
392 /* Request is not yet in userspace, bail out */
393 if (test_bit(FR_PENDING, &req->flags)) {
394 list_del(&req->list);
395 spin_unlock(&fiq->lock);
396 __fuse_put_request(req);
397 req->out.h.error = -EINTR;
400 spin_unlock(&fiq->lock);
404 * Either request is already in userspace, or it was forced.
407 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
410 static void __fuse_request_send(struct fuse_req *req)
412 struct fuse_iqueue *fiq = &req->fm->fc->iq;
414 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
415 spin_lock(&fiq->lock);
416 if (!fiq->connected) {
417 spin_unlock(&fiq->lock);
418 req->out.h.error = -ENOTCONN;
420 req->in.h.unique = fuse_get_unique(fiq);
421 /* acquire extra reference, since request is still needed
422 after fuse_request_end() */
423 __fuse_get_request(req);
424 queue_request_and_unlock(fiq, req);
426 request_wait_answer(req);
427 /* Pairs with smp_wmb() in fuse_request_end() */
432 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
434 if (fc->minor < 4 && args->opcode == FUSE_STATFS)
435 args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE;
438 switch (args->opcode) {
445 args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
449 args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
453 if (fc->minor < 12) {
454 switch (args->opcode) {
456 args->in_args[0].size = sizeof(struct fuse_open_in);
459 args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
465 static void fuse_force_creds(struct fuse_req *req)
467 struct fuse_conn *fc = req->fm->fc;
469 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid());
470 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid());
471 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
474 static void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args)
476 req->in.h.opcode = args->opcode;
477 req->in.h.nodeid = args->nodeid;
480 req->in.h.total_extlen = args->in_args[args->ext_idx].size / 8;
482 __set_bit(FR_ASYNC, &req->flags);
485 ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args)
487 struct fuse_conn *fc = fm->fc;
488 struct fuse_req *req;
492 atomic_inc(&fc->num_waiting);
493 req = fuse_request_alloc(fm, GFP_KERNEL | __GFP_NOFAIL);
496 fuse_force_creds(req);
498 __set_bit(FR_WAITING, &req->flags);
499 __set_bit(FR_FORCE, &req->flags);
501 WARN_ON(args->nocreds);
502 req = fuse_get_req(fm, false);
507 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
508 fuse_adjust_compat(fc, args);
509 fuse_args_to_req(req, args);
512 __set_bit(FR_ISREPLY, &req->flags);
513 __fuse_request_send(req);
514 ret = req->out.h.error;
515 if (!ret && args->out_argvar) {
516 BUG_ON(args->out_numargs == 0);
517 ret = args->out_args[args->out_numargs - 1].size;
519 fuse_put_request(req);
524 static bool fuse_request_queue_background(struct fuse_req *req)
526 struct fuse_mount *fm = req->fm;
527 struct fuse_conn *fc = fm->fc;
530 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags));
531 if (!test_bit(FR_WAITING, &req->flags)) {
532 __set_bit(FR_WAITING, &req->flags);
533 atomic_inc(&fc->num_waiting);
535 __set_bit(FR_ISREPLY, &req->flags);
536 spin_lock(&fc->bg_lock);
537 if (likely(fc->connected)) {
538 fc->num_background++;
539 if (fc->num_background == fc->max_background)
541 list_add_tail(&req->list, &fc->bg_queue);
545 spin_unlock(&fc->bg_lock);
550 int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args,
553 struct fuse_req *req;
556 WARN_ON(!args->nocreds);
557 req = fuse_request_alloc(fm, gfp_flags);
560 __set_bit(FR_BACKGROUND, &req->flags);
562 WARN_ON(args->nocreds);
563 req = fuse_get_req(fm, true);
568 fuse_args_to_req(req, args);
570 if (!fuse_request_queue_background(req)) {
571 fuse_put_request(req);
577 EXPORT_SYMBOL_GPL(fuse_simple_background);
579 static int fuse_simple_notify_reply(struct fuse_mount *fm,
580 struct fuse_args *args, u64 unique)
582 struct fuse_req *req;
583 struct fuse_iqueue *fiq = &fm->fc->iq;
586 req = fuse_get_req(fm, false);
590 __clear_bit(FR_ISREPLY, &req->flags);
591 req->in.h.unique = unique;
593 fuse_args_to_req(req, args);
595 spin_lock(&fiq->lock);
596 if (fiq->connected) {
597 queue_request_and_unlock(fiq, req);
600 spin_unlock(&fiq->lock);
601 fuse_put_request(req);
608 * Lock the request. Up to the next unlock_request() there mustn't be
609 * anything that could cause a page-fault. If the request was already
612 static int lock_request(struct fuse_req *req)
616 spin_lock(&req->waitq.lock);
617 if (test_bit(FR_ABORTED, &req->flags))
620 set_bit(FR_LOCKED, &req->flags);
621 spin_unlock(&req->waitq.lock);
627 * Unlock request. If it was aborted while locked, caller is responsible
628 * for unlocking and ending the request.
630 static int unlock_request(struct fuse_req *req)
634 spin_lock(&req->waitq.lock);
635 if (test_bit(FR_ABORTED, &req->flags))
638 clear_bit(FR_LOCKED, &req->flags);
639 spin_unlock(&req->waitq.lock);
644 struct fuse_copy_state {
646 struct fuse_req *req;
647 struct iov_iter *iter;
648 struct pipe_buffer *pipebufs;
649 struct pipe_buffer *currbuf;
650 struct pipe_inode_info *pipe;
651 unsigned long nr_segs;
655 unsigned move_pages:1;
658 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
659 struct iov_iter *iter)
661 memset(cs, 0, sizeof(*cs));
666 /* Unmap and put previous page of userspace buffer */
667 static void fuse_copy_finish(struct fuse_copy_state *cs)
670 struct pipe_buffer *buf = cs->currbuf;
673 buf->len = PAGE_SIZE - cs->len;
677 flush_dcache_page(cs->pg);
678 set_page_dirty_lock(cs->pg);
686 * Get another pagefull of userspace buffer, and map it to kernel
687 * address space, and lock request
689 static int fuse_copy_fill(struct fuse_copy_state *cs)
694 err = unlock_request(cs->req);
698 fuse_copy_finish(cs);
700 struct pipe_buffer *buf = cs->pipebufs;
703 err = pipe_buf_confirm(cs->pipe, buf);
707 BUG_ON(!cs->nr_segs);
710 cs->offset = buf->offset;
715 if (cs->nr_segs >= cs->pipe->max_usage)
718 page = alloc_page(GFP_HIGHUSER);
735 err = iov_iter_get_pages2(cs->iter, &page, PAGE_SIZE, 1, &off);
744 return lock_request(cs->req);
747 /* Do as much copy to/from userspace buffer as we can */
748 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
750 unsigned ncpy = min(*size, cs->len);
752 void *pgaddr = kmap_local_page(cs->pg);
753 void *buf = pgaddr + cs->offset;
756 memcpy(buf, *val, ncpy);
758 memcpy(*val, buf, ncpy);
760 kunmap_local(pgaddr);
769 static int fuse_check_folio(struct folio *folio)
771 if (folio_mapped(folio) ||
772 folio->mapping != NULL ||
773 (folio->flags & PAGE_FLAGS_CHECK_AT_PREP &
782 LRU_GEN_MASK | LRU_REFS_MASK))) {
783 dump_page(&folio->page, "fuse: trying to steal weird page");
789 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
792 struct folio *oldfolio = page_folio(*pagep);
793 struct folio *newfolio;
794 struct pipe_buffer *buf = cs->pipebufs;
797 err = unlock_request(cs->req);
801 fuse_copy_finish(cs);
803 err = pipe_buf_confirm(cs->pipe, buf);
807 BUG_ON(!cs->nr_segs);
813 if (cs->len != PAGE_SIZE)
816 if (!pipe_buf_try_steal(cs->pipe, buf))
819 newfolio = page_folio(buf->page);
821 if (!folio_test_uptodate(newfolio))
822 folio_mark_uptodate(newfolio);
824 folio_clear_mappedtodisk(newfolio);
826 if (fuse_check_folio(newfolio) != 0)
827 goto out_fallback_unlock;
830 * This is a new and locked page, it shouldn't be mapped or
831 * have any special flags on it
833 if (WARN_ON(folio_mapped(oldfolio)))
834 goto out_fallback_unlock;
835 if (WARN_ON(folio_has_private(oldfolio)))
836 goto out_fallback_unlock;
837 if (WARN_ON(folio_test_dirty(oldfolio) ||
838 folio_test_writeback(oldfolio)))
839 goto out_fallback_unlock;
840 if (WARN_ON(folio_test_mlocked(oldfolio)))
841 goto out_fallback_unlock;
843 replace_page_cache_folio(oldfolio, newfolio);
847 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
848 folio_add_lru(newfolio);
851 * Release while we have extra ref on stolen page. Otherwise
852 * anon_pipe_buf_release() might think the page can be reused.
854 pipe_buf_release(cs->pipe, buf);
857 spin_lock(&cs->req->waitq.lock);
858 if (test_bit(FR_ABORTED, &cs->req->flags))
861 *pagep = &newfolio->page;
862 spin_unlock(&cs->req->waitq.lock);
865 folio_unlock(newfolio);
870 folio_unlock(oldfolio);
871 /* Drop ref for ap->pages[] array */
877 /* Drop ref obtained in this function */
882 folio_unlock(newfolio);
885 cs->offset = buf->offset;
887 err = lock_request(cs->req);
894 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
895 unsigned offset, unsigned count)
897 struct pipe_buffer *buf;
900 if (cs->nr_segs >= cs->pipe->max_usage)
904 err = unlock_request(cs->req);
910 fuse_copy_finish(cs);
914 buf->offset = offset;
925 * Copy a page in the request to/from the userspace buffer. Must be
928 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
929 unsigned offset, unsigned count, int zeroing)
932 struct page *page = *pagep;
934 if (page && zeroing && count < PAGE_SIZE)
935 clear_highpage(page);
938 if (cs->write && cs->pipebufs && page) {
940 * Can't control lifetime of pipe buffers, so always
943 if (cs->req->args->user_pages) {
944 err = fuse_copy_fill(cs);
948 return fuse_ref_page(cs, page, offset, count);
950 } else if (!cs->len) {
951 if (cs->move_pages && page &&
952 offset == 0 && count == PAGE_SIZE) {
953 err = fuse_try_move_page(cs, pagep);
957 err = fuse_copy_fill(cs);
963 void *mapaddr = kmap_local_page(page);
964 void *buf = mapaddr + offset;
965 offset += fuse_copy_do(cs, &buf, &count);
966 kunmap_local(mapaddr);
968 offset += fuse_copy_do(cs, NULL, &count);
970 if (page && !cs->write)
971 flush_dcache_page(page);
975 /* Copy pages in the request to/from userspace buffer */
976 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
980 struct fuse_req *req = cs->req;
981 struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args);
984 for (i = 0; i < ap->num_pages && (nbytes || zeroing); i++) {
986 unsigned int offset = ap->descs[i].offset;
987 unsigned int count = min(nbytes, ap->descs[i].length);
989 err = fuse_copy_page(cs, &ap->pages[i], offset, count, zeroing);
998 /* Copy a single argument in the request to/from userspace buffer */
999 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1003 int err = fuse_copy_fill(cs);
1007 fuse_copy_do(cs, &val, &size);
1012 /* Copy request arguments to/from userspace buffer */
1013 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1014 unsigned argpages, struct fuse_arg *args,
1020 for (i = 0; !err && i < numargs; i++) {
1021 struct fuse_arg *arg = &args[i];
1022 if (i == numargs - 1 && argpages)
1023 err = fuse_copy_pages(cs, arg->size, zeroing);
1025 err = fuse_copy_one(cs, arg->value, arg->size);
1030 static int forget_pending(struct fuse_iqueue *fiq)
1032 return fiq->forget_list_head.next != NULL;
1035 static int request_pending(struct fuse_iqueue *fiq)
1037 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1038 forget_pending(fiq);
1042 * Transfer an interrupt request to userspace
1044 * Unlike other requests this is assembled on demand, without a need
1045 * to allocate a separate fuse_req structure.
1047 * Called with fiq->lock held, releases it
1049 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1050 struct fuse_copy_state *cs,
1051 size_t nbytes, struct fuse_req *req)
1052 __releases(fiq->lock)
1054 struct fuse_in_header ih;
1055 struct fuse_interrupt_in arg;
1056 unsigned reqsize = sizeof(ih) + sizeof(arg);
1059 list_del_init(&req->intr_entry);
1060 memset(&ih, 0, sizeof(ih));
1061 memset(&arg, 0, sizeof(arg));
1063 ih.opcode = FUSE_INTERRUPT;
1064 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT);
1065 arg.unique = req->in.h.unique;
1067 spin_unlock(&fiq->lock);
1068 if (nbytes < reqsize)
1071 err = fuse_copy_one(cs, &ih, sizeof(ih));
1073 err = fuse_copy_one(cs, &arg, sizeof(arg));
1074 fuse_copy_finish(cs);
1076 return err ? err : reqsize;
1079 struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq,
1081 unsigned int *countp)
1083 struct fuse_forget_link *head = fiq->forget_list_head.next;
1084 struct fuse_forget_link **newhead = &head;
1087 for (count = 0; *newhead != NULL && count < max; count++)
1088 newhead = &(*newhead)->next;
1090 fiq->forget_list_head.next = *newhead;
1092 if (fiq->forget_list_head.next == NULL)
1093 fiq->forget_list_tail = &fiq->forget_list_head;
1100 EXPORT_SYMBOL(fuse_dequeue_forget);
1102 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1103 struct fuse_copy_state *cs,
1105 __releases(fiq->lock)
1108 struct fuse_forget_link *forget = fuse_dequeue_forget(fiq, 1, NULL);
1109 struct fuse_forget_in arg = {
1110 .nlookup = forget->forget_one.nlookup,
1112 struct fuse_in_header ih = {
1113 .opcode = FUSE_FORGET,
1114 .nodeid = forget->forget_one.nodeid,
1115 .unique = fuse_get_unique(fiq),
1116 .len = sizeof(ih) + sizeof(arg),
1119 spin_unlock(&fiq->lock);
1121 if (nbytes < ih.len)
1124 err = fuse_copy_one(cs, &ih, sizeof(ih));
1126 err = fuse_copy_one(cs, &arg, sizeof(arg));
1127 fuse_copy_finish(cs);
1135 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1136 struct fuse_copy_state *cs, size_t nbytes)
1137 __releases(fiq->lock)
1140 unsigned max_forgets;
1142 struct fuse_forget_link *head;
1143 struct fuse_batch_forget_in arg = { .count = 0 };
1144 struct fuse_in_header ih = {
1145 .opcode = FUSE_BATCH_FORGET,
1146 .unique = fuse_get_unique(fiq),
1147 .len = sizeof(ih) + sizeof(arg),
1150 if (nbytes < ih.len) {
1151 spin_unlock(&fiq->lock);
1155 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1156 head = fuse_dequeue_forget(fiq, max_forgets, &count);
1157 spin_unlock(&fiq->lock);
1160 ih.len += count * sizeof(struct fuse_forget_one);
1161 err = fuse_copy_one(cs, &ih, sizeof(ih));
1163 err = fuse_copy_one(cs, &arg, sizeof(arg));
1166 struct fuse_forget_link *forget = head;
1169 err = fuse_copy_one(cs, &forget->forget_one,
1170 sizeof(forget->forget_one));
1172 head = forget->next;
1176 fuse_copy_finish(cs);
1184 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1185 struct fuse_copy_state *cs,
1187 __releases(fiq->lock)
1189 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1190 return fuse_read_single_forget(fiq, cs, nbytes);
1192 return fuse_read_batch_forget(fiq, cs, nbytes);
1196 * Read a single request into the userspace filesystem's buffer. This
1197 * function waits until a request is available, then removes it from
1198 * the pending list and copies request data to userspace buffer. If
1199 * no reply is needed (FORGET) or request has been aborted or there
1200 * was an error during the copying then it's finished by calling
1201 * fuse_request_end(). Otherwise add it to the processing list, and set
1204 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1205 struct fuse_copy_state *cs, size_t nbytes)
1208 struct fuse_conn *fc = fud->fc;
1209 struct fuse_iqueue *fiq = &fc->iq;
1210 struct fuse_pqueue *fpq = &fud->pq;
1211 struct fuse_req *req;
1212 struct fuse_args *args;
1217 * Require sane minimum read buffer - that has capacity for fixed part
1218 * of any request header + negotiated max_write room for data.
1220 * Historically libfuse reserves 4K for fixed header room, but e.g.
1221 * GlusterFS reserves only 80 bytes
1223 * = `sizeof(fuse_in_header) + sizeof(fuse_write_in)`
1225 * which is the absolute minimum any sane filesystem should be using
1228 if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER,
1229 sizeof(struct fuse_in_header) +
1230 sizeof(struct fuse_write_in) +
1236 spin_lock(&fiq->lock);
1237 if (!fiq->connected || request_pending(fiq))
1239 spin_unlock(&fiq->lock);
1241 if (file->f_flags & O_NONBLOCK)
1243 err = wait_event_interruptible_exclusive(fiq->waitq,
1244 !fiq->connected || request_pending(fiq));
1249 if (!fiq->connected) {
1250 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1254 if (!list_empty(&fiq->interrupts)) {
1255 req = list_entry(fiq->interrupts.next, struct fuse_req,
1257 return fuse_read_interrupt(fiq, cs, nbytes, req);
1260 if (forget_pending(fiq)) {
1261 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1262 return fuse_read_forget(fc, fiq, cs, nbytes);
1264 if (fiq->forget_batch <= -8)
1265 fiq->forget_batch = 16;
1268 req = list_entry(fiq->pending.next, struct fuse_req, list);
1269 clear_bit(FR_PENDING, &req->flags);
1270 list_del_init(&req->list);
1271 spin_unlock(&fiq->lock);
1274 reqsize = req->in.h.len;
1276 /* If request is too large, reply with an error and restart the read */
1277 if (nbytes < reqsize) {
1278 req->out.h.error = -EIO;
1279 /* SETXATTR is special, since it may contain too large data */
1280 if (args->opcode == FUSE_SETXATTR)
1281 req->out.h.error = -E2BIG;
1282 fuse_request_end(req);
1285 spin_lock(&fpq->lock);
1287 * Must not put request on fpq->io queue after having been shut down by
1290 if (!fpq->connected) {
1291 req->out.h.error = err = -ECONNABORTED;
1295 list_add(&req->list, &fpq->io);
1296 spin_unlock(&fpq->lock);
1298 err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h));
1300 err = fuse_copy_args(cs, args->in_numargs, args->in_pages,
1301 (struct fuse_arg *) args->in_args, 0);
1302 fuse_copy_finish(cs);
1303 spin_lock(&fpq->lock);
1304 clear_bit(FR_LOCKED, &req->flags);
1305 if (!fpq->connected) {
1306 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1310 req->out.h.error = -EIO;
1313 if (!test_bit(FR_ISREPLY, &req->flags)) {
1317 hash = fuse_req_hash(req->in.h.unique);
1318 list_move_tail(&req->list, &fpq->processing[hash]);
1319 __fuse_get_request(req);
1320 set_bit(FR_SENT, &req->flags);
1321 spin_unlock(&fpq->lock);
1322 /* matches barrier in request_wait_answer() */
1323 smp_mb__after_atomic();
1324 if (test_bit(FR_INTERRUPTED, &req->flags))
1325 queue_interrupt(req);
1326 fuse_put_request(req);
1331 if (!test_bit(FR_PRIVATE, &req->flags))
1332 list_del_init(&req->list);
1333 spin_unlock(&fpq->lock);
1334 fuse_request_end(req);
1338 spin_unlock(&fiq->lock);
1342 static int fuse_dev_open(struct inode *inode, struct file *file)
1345 * The fuse device's file's private_data is used to hold
1346 * the fuse_conn(ection) when it is mounted, and is used to
1347 * keep track of whether the file has been mounted already.
1349 file->private_data = NULL;
1353 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1355 struct fuse_copy_state cs;
1356 struct file *file = iocb->ki_filp;
1357 struct fuse_dev *fud = fuse_get_dev(file);
1362 if (!user_backed_iter(to))
1365 fuse_copy_init(&cs, 1, to);
1367 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1370 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1371 struct pipe_inode_info *pipe,
1372 size_t len, unsigned int flags)
1376 struct pipe_buffer *bufs;
1377 struct fuse_copy_state cs;
1378 struct fuse_dev *fud = fuse_get_dev(in);
1383 bufs = kvmalloc_array(pipe->max_usage, sizeof(struct pipe_buffer),
1388 fuse_copy_init(&cs, 1, NULL);
1391 ret = fuse_dev_do_read(fud, in, &cs, len);
1395 if (pipe_occupancy(pipe->head, pipe->tail) + cs.nr_segs > pipe->max_usage) {
1400 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1402 * Need to be careful about this. Having buf->ops in module
1403 * code can Oops if the buffer persists after module unload.
1405 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1406 bufs[page_nr].flags = 0;
1407 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1408 if (unlikely(ret < 0))
1414 for (; page_nr < cs.nr_segs; page_nr++)
1415 put_page(bufs[page_nr].page);
1421 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1422 struct fuse_copy_state *cs)
1424 struct fuse_notify_poll_wakeup_out outarg;
1427 if (size != sizeof(outarg))
1430 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1434 fuse_copy_finish(cs);
1435 return fuse_notify_poll_wakeup(fc, &outarg);
1438 fuse_copy_finish(cs);
1442 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1443 struct fuse_copy_state *cs)
1445 struct fuse_notify_inval_inode_out outarg;
1448 if (size != sizeof(outarg))
1451 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1454 fuse_copy_finish(cs);
1456 down_read(&fc->killsb);
1457 err = fuse_reverse_inval_inode(fc, outarg.ino,
1458 outarg.off, outarg.len);
1459 up_read(&fc->killsb);
1463 fuse_copy_finish(cs);
1467 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1468 struct fuse_copy_state *cs)
1470 struct fuse_notify_inval_entry_out outarg;
1475 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1480 if (size < sizeof(outarg))
1483 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1487 err = -ENAMETOOLONG;
1488 if (outarg.namelen > FUSE_NAME_MAX)
1492 if (size != sizeof(outarg) + outarg.namelen + 1)
1496 name.len = outarg.namelen;
1497 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1500 fuse_copy_finish(cs);
1501 buf[outarg.namelen] = 0;
1503 down_read(&fc->killsb);
1504 err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name, outarg.flags);
1505 up_read(&fc->killsb);
1511 fuse_copy_finish(cs);
1515 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1516 struct fuse_copy_state *cs)
1518 struct fuse_notify_delete_out outarg;
1523 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1528 if (size < sizeof(outarg))
1531 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1535 err = -ENAMETOOLONG;
1536 if (outarg.namelen > FUSE_NAME_MAX)
1540 if (size != sizeof(outarg) + outarg.namelen + 1)
1544 name.len = outarg.namelen;
1545 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1548 fuse_copy_finish(cs);
1549 buf[outarg.namelen] = 0;
1551 down_read(&fc->killsb);
1552 err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name, 0);
1553 up_read(&fc->killsb);
1559 fuse_copy_finish(cs);
1563 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1564 struct fuse_copy_state *cs)
1566 struct fuse_notify_store_out outarg;
1567 struct inode *inode;
1568 struct address_space *mapping;
1572 unsigned int offset;
1578 if (size < sizeof(outarg))
1581 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1586 if (size - sizeof(outarg) != outarg.size)
1589 nodeid = outarg.nodeid;
1591 down_read(&fc->killsb);
1594 inode = fuse_ilookup(fc, nodeid, NULL);
1598 mapping = inode->i_mapping;
1599 index = outarg.offset >> PAGE_SHIFT;
1600 offset = outarg.offset & ~PAGE_MASK;
1601 file_size = i_size_read(inode);
1602 end = outarg.offset + outarg.size;
1603 if (end > file_size) {
1605 fuse_write_update_attr(inode, file_size, outarg.size);
1611 unsigned int this_num;
1614 page = find_or_create_page(mapping, index,
1615 mapping_gfp_mask(mapping));
1619 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1620 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1621 if (!err && offset == 0 &&
1622 (this_num == PAGE_SIZE || file_size == end))
1623 SetPageUptodate(page);
1640 up_read(&fc->killsb);
1642 fuse_copy_finish(cs);
1646 struct fuse_retrieve_args {
1647 struct fuse_args_pages ap;
1648 struct fuse_notify_retrieve_in inarg;
1651 static void fuse_retrieve_end(struct fuse_mount *fm, struct fuse_args *args,
1654 struct fuse_retrieve_args *ra =
1655 container_of(args, typeof(*ra), ap.args);
1657 release_pages(ra->ap.pages, ra->ap.num_pages);
1661 static int fuse_retrieve(struct fuse_mount *fm, struct inode *inode,
1662 struct fuse_notify_retrieve_out *outarg)
1665 struct address_space *mapping = inode->i_mapping;
1669 unsigned int offset;
1670 size_t total_len = 0;
1671 unsigned int num_pages;
1672 struct fuse_conn *fc = fm->fc;
1673 struct fuse_retrieve_args *ra;
1674 size_t args_size = sizeof(*ra);
1675 struct fuse_args_pages *ap;
1676 struct fuse_args *args;
1678 offset = outarg->offset & ~PAGE_MASK;
1679 file_size = i_size_read(inode);
1681 num = min(outarg->size, fc->max_write);
1682 if (outarg->offset > file_size)
1684 else if (outarg->offset + num > file_size)
1685 num = file_size - outarg->offset;
1687 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1688 num_pages = min(num_pages, fc->max_pages);
1690 args_size += num_pages * (sizeof(ap->pages[0]) + sizeof(ap->descs[0]));
1692 ra = kzalloc(args_size, GFP_KERNEL);
1697 ap->pages = (void *) (ra + 1);
1698 ap->descs = (void *) (ap->pages + num_pages);
1701 args->nodeid = outarg->nodeid;
1702 args->opcode = FUSE_NOTIFY_REPLY;
1703 args->in_numargs = 2;
1704 args->in_pages = true;
1705 args->end = fuse_retrieve_end;
1707 index = outarg->offset >> PAGE_SHIFT;
1709 while (num && ap->num_pages < num_pages) {
1711 unsigned int this_num;
1713 page = find_get_page(mapping, index);
1717 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1718 ap->pages[ap->num_pages] = page;
1719 ap->descs[ap->num_pages].offset = offset;
1720 ap->descs[ap->num_pages].length = this_num;
1725 total_len += this_num;
1728 ra->inarg.offset = outarg->offset;
1729 ra->inarg.size = total_len;
1730 args->in_args[0].size = sizeof(ra->inarg);
1731 args->in_args[0].value = &ra->inarg;
1732 args->in_args[1].size = total_len;
1734 err = fuse_simple_notify_reply(fm, args, outarg->notify_unique);
1736 fuse_retrieve_end(fm, args, err);
1741 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1742 struct fuse_copy_state *cs)
1744 struct fuse_notify_retrieve_out outarg;
1745 struct fuse_mount *fm;
1746 struct inode *inode;
1751 if (size != sizeof(outarg))
1754 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1758 fuse_copy_finish(cs);
1760 down_read(&fc->killsb);
1762 nodeid = outarg.nodeid;
1764 inode = fuse_ilookup(fc, nodeid, &fm);
1766 err = fuse_retrieve(fm, inode, &outarg);
1769 up_read(&fc->killsb);
1774 fuse_copy_finish(cs);
1778 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1779 unsigned int size, struct fuse_copy_state *cs)
1781 /* Don't try to move pages (yet) */
1785 case FUSE_NOTIFY_POLL:
1786 return fuse_notify_poll(fc, size, cs);
1788 case FUSE_NOTIFY_INVAL_INODE:
1789 return fuse_notify_inval_inode(fc, size, cs);
1791 case FUSE_NOTIFY_INVAL_ENTRY:
1792 return fuse_notify_inval_entry(fc, size, cs);
1794 case FUSE_NOTIFY_STORE:
1795 return fuse_notify_store(fc, size, cs);
1797 case FUSE_NOTIFY_RETRIEVE:
1798 return fuse_notify_retrieve(fc, size, cs);
1800 case FUSE_NOTIFY_DELETE:
1801 return fuse_notify_delete(fc, size, cs);
1804 fuse_copy_finish(cs);
1809 /* Look up request on processing list by unique ID */
1810 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1812 unsigned int hash = fuse_req_hash(unique);
1813 struct fuse_req *req;
1815 list_for_each_entry(req, &fpq->processing[hash], list) {
1816 if (req->in.h.unique == unique)
1822 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args,
1825 unsigned reqsize = sizeof(struct fuse_out_header);
1827 reqsize += fuse_len_args(args->out_numargs, args->out_args);
1829 if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar))
1831 else if (reqsize > nbytes) {
1832 struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1];
1833 unsigned diffsize = reqsize - nbytes;
1835 if (diffsize > lastarg->size)
1837 lastarg->size -= diffsize;
1839 return fuse_copy_args(cs, args->out_numargs, args->out_pages,
1840 args->out_args, args->page_zeroing);
1844 * Write a single reply to a request. First the header is copied from
1845 * the write buffer. The request is then searched on the processing
1846 * list by the unique ID found in the header. If found, then remove
1847 * it from the list and copy the rest of the buffer to the request.
1848 * The request is finished by calling fuse_request_end().
1850 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1851 struct fuse_copy_state *cs, size_t nbytes)
1854 struct fuse_conn *fc = fud->fc;
1855 struct fuse_pqueue *fpq = &fud->pq;
1856 struct fuse_req *req;
1857 struct fuse_out_header oh;
1860 if (nbytes < sizeof(struct fuse_out_header))
1863 err = fuse_copy_one(cs, &oh, sizeof(oh));
1868 if (oh.len != nbytes)
1872 * Zero oh.unique indicates unsolicited notification message
1873 * and error contains notification code.
1876 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1881 if (oh.error <= -512 || oh.error > 0)
1884 spin_lock(&fpq->lock);
1887 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
1891 spin_unlock(&fpq->lock);
1895 /* Is it an interrupt reply ID? */
1896 if (oh.unique & FUSE_INT_REQ_BIT) {
1897 __fuse_get_request(req);
1898 spin_unlock(&fpq->lock);
1901 if (nbytes != sizeof(struct fuse_out_header))
1903 else if (oh.error == -ENOSYS)
1904 fc->no_interrupt = 1;
1905 else if (oh.error == -EAGAIN)
1906 err = queue_interrupt(req);
1908 fuse_put_request(req);
1913 clear_bit(FR_SENT, &req->flags);
1914 list_move(&req->list, &fpq->io);
1916 set_bit(FR_LOCKED, &req->flags);
1917 spin_unlock(&fpq->lock);
1919 if (!req->args->page_replace)
1923 err = nbytes != sizeof(oh) ? -EINVAL : 0;
1925 err = copy_out_args(cs, req->args, nbytes);
1926 fuse_copy_finish(cs);
1928 spin_lock(&fpq->lock);
1929 clear_bit(FR_LOCKED, &req->flags);
1930 if (!fpq->connected)
1933 req->out.h.error = -EIO;
1934 if (!test_bit(FR_PRIVATE, &req->flags))
1935 list_del_init(&req->list);
1936 spin_unlock(&fpq->lock);
1938 fuse_request_end(req);
1940 return err ? err : nbytes;
1943 fuse_copy_finish(cs);
1947 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1949 struct fuse_copy_state cs;
1950 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1955 if (!user_backed_iter(from))
1958 fuse_copy_init(&cs, 0, from);
1960 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1963 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1964 struct file *out, loff_t *ppos,
1965 size_t len, unsigned int flags)
1967 unsigned int head, tail, mask, count;
1970 struct pipe_buffer *bufs;
1971 struct fuse_copy_state cs;
1972 struct fuse_dev *fud;
1976 fud = fuse_get_dev(out);
1984 mask = pipe->ring_size - 1;
1985 count = head - tail;
1987 bufs = kvmalloc_array(count, sizeof(struct pipe_buffer), GFP_KERNEL);
1995 for (idx = tail; idx != head && rem < len; idx++)
1996 rem += pipe->bufs[idx & mask].len;
2004 struct pipe_buffer *ibuf;
2005 struct pipe_buffer *obuf;
2007 if (WARN_ON(nbuf >= count || tail == head))
2010 ibuf = &pipe->bufs[tail & mask];
2013 if (rem >= ibuf->len) {
2019 if (!pipe_buf_get(pipe, ibuf))
2023 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2025 ibuf->offset += obuf->len;
2026 ibuf->len -= obuf->len;
2033 fuse_copy_init(&cs, 0, NULL);
2038 if (flags & SPLICE_F_MOVE)
2041 ret = fuse_dev_do_write(fud, &cs, len);
2045 for (idx = 0; idx < nbuf; idx++) {
2046 struct pipe_buffer *buf = &bufs[idx];
2049 pipe_buf_release(pipe, buf);
2057 static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2059 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
2060 struct fuse_iqueue *fiq;
2061 struct fuse_dev *fud = fuse_get_dev(file);
2067 poll_wait(file, &fiq->waitq, wait);
2069 spin_lock(&fiq->lock);
2070 if (!fiq->connected)
2072 else if (request_pending(fiq))
2073 mask |= EPOLLIN | EPOLLRDNORM;
2074 spin_unlock(&fiq->lock);
2079 /* Abort all requests on the given list (pending or processing) */
2080 static void end_requests(struct list_head *head)
2082 while (!list_empty(head)) {
2083 struct fuse_req *req;
2084 req = list_entry(head->next, struct fuse_req, list);
2085 req->out.h.error = -ECONNABORTED;
2086 clear_bit(FR_SENT, &req->flags);
2087 list_del_init(&req->list);
2088 fuse_request_end(req);
2092 static void end_polls(struct fuse_conn *fc)
2096 p = rb_first(&fc->polled_files);
2099 struct fuse_file *ff;
2100 ff = rb_entry(p, struct fuse_file, polled_node);
2101 wake_up_interruptible_all(&ff->poll_wait);
2108 * Abort all requests.
2110 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2113 * The same effect is usually achievable through killing the filesystem daemon
2114 * and all users of the filesystem. The exception is the combination of an
2115 * asynchronous request and the tricky deadlock (see
2116 * Documentation/filesystems/fuse.rst).
2118 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2119 * requests, they should be finished off immediately. Locked requests will be
2120 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2121 * requests. It is possible that some request will finish before we can. This
2122 * is OK, the request will in that case be removed from the list before we touch
2125 void fuse_abort_conn(struct fuse_conn *fc)
2127 struct fuse_iqueue *fiq = &fc->iq;
2129 spin_lock(&fc->lock);
2130 if (fc->connected) {
2131 struct fuse_dev *fud;
2132 struct fuse_req *req, *next;
2136 /* Background queuing checks fc->connected under bg_lock */
2137 spin_lock(&fc->bg_lock);
2139 spin_unlock(&fc->bg_lock);
2141 fuse_set_initialized(fc);
2142 list_for_each_entry(fud, &fc->devices, entry) {
2143 struct fuse_pqueue *fpq = &fud->pq;
2145 spin_lock(&fpq->lock);
2147 list_for_each_entry_safe(req, next, &fpq->io, list) {
2148 req->out.h.error = -ECONNABORTED;
2149 spin_lock(&req->waitq.lock);
2150 set_bit(FR_ABORTED, &req->flags);
2151 if (!test_bit(FR_LOCKED, &req->flags)) {
2152 set_bit(FR_PRIVATE, &req->flags);
2153 __fuse_get_request(req);
2154 list_move(&req->list, &to_end);
2156 spin_unlock(&req->waitq.lock);
2158 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2159 list_splice_tail_init(&fpq->processing[i],
2161 spin_unlock(&fpq->lock);
2163 spin_lock(&fc->bg_lock);
2165 fc->max_background = UINT_MAX;
2167 spin_unlock(&fc->bg_lock);
2169 spin_lock(&fiq->lock);
2171 list_for_each_entry(req, &fiq->pending, list)
2172 clear_bit(FR_PENDING, &req->flags);
2173 list_splice_tail_init(&fiq->pending, &to_end);
2174 while (forget_pending(fiq))
2175 kfree(fuse_dequeue_forget(fiq, 1, NULL));
2176 wake_up_all(&fiq->waitq);
2177 spin_unlock(&fiq->lock);
2178 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2180 wake_up_all(&fc->blocked_waitq);
2181 spin_unlock(&fc->lock);
2183 end_requests(&to_end);
2185 spin_unlock(&fc->lock);
2188 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2190 void fuse_wait_aborted(struct fuse_conn *fc)
2192 /* matches implicit memory barrier in fuse_drop_waiting() */
2194 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2197 int fuse_dev_release(struct inode *inode, struct file *file)
2199 struct fuse_dev *fud = fuse_get_dev(file);
2202 struct fuse_conn *fc = fud->fc;
2203 struct fuse_pqueue *fpq = &fud->pq;
2207 spin_lock(&fpq->lock);
2208 WARN_ON(!list_empty(&fpq->io));
2209 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2210 list_splice_init(&fpq->processing[i], &to_end);
2211 spin_unlock(&fpq->lock);
2213 end_requests(&to_end);
2215 /* Are we the last open device? */
2216 if (atomic_dec_and_test(&fc->dev_count)) {
2217 WARN_ON(fc->iq.fasync != NULL);
2218 fuse_abort_conn(fc);
2224 EXPORT_SYMBOL_GPL(fuse_dev_release);
2226 static int fuse_dev_fasync(int fd, struct file *file, int on)
2228 struct fuse_dev *fud = fuse_get_dev(file);
2233 /* No locking - fasync_helper does its own locking */
2234 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2237 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2239 struct fuse_dev *fud;
2241 if (new->private_data)
2244 fud = fuse_dev_alloc_install(fc);
2248 new->private_data = fud;
2249 atomic_inc(&fc->dev_count);
2254 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2259 struct fuse_dev *fud = NULL;
2262 case FUSE_DEV_IOC_CLONE:
2264 if (!get_user(oldfd, (__u32 __user *)arg)) {
2265 struct file *old = fget(oldfd);
2270 * Check against file->f_op because CUSE
2271 * uses the same ioctl handler.
2273 if (old->f_op == file->f_op)
2274 fud = fuse_get_dev(old);
2277 mutex_lock(&fuse_mutex);
2278 res = fuse_device_clone(fud->fc, file);
2279 mutex_unlock(&fuse_mutex);
2292 const struct file_operations fuse_dev_operations = {
2293 .owner = THIS_MODULE,
2294 .open = fuse_dev_open,
2295 .llseek = no_llseek,
2296 .read_iter = fuse_dev_read,
2297 .splice_read = fuse_dev_splice_read,
2298 .write_iter = fuse_dev_write,
2299 .splice_write = fuse_dev_splice_write,
2300 .poll = fuse_dev_poll,
2301 .release = fuse_dev_release,
2302 .fasync = fuse_dev_fasync,
2303 .unlocked_ioctl = fuse_dev_ioctl,
2304 .compat_ioctl = compat_ptr_ioctl,
2306 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2308 static struct miscdevice fuse_miscdevice = {
2309 .minor = FUSE_MINOR,
2311 .fops = &fuse_dev_operations,
2314 int __init fuse_dev_init(void)
2317 fuse_req_cachep = kmem_cache_create("fuse_request",
2318 sizeof(struct fuse_req),
2320 if (!fuse_req_cachep)
2323 err = misc_register(&fuse_miscdevice);
2325 goto out_cache_clean;
2330 kmem_cache_destroy(fuse_req_cachep);
2335 void fuse_dev_cleanup(void)
2337 misc_deregister(&fuse_miscdevice);
2338 kmem_cache_destroy(fuse_req_cachep);