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_req *req, struct page **pages,
44 struct fuse_page_desc *page_descs,
47 INIT_LIST_HEAD(&req->list);
48 INIT_LIST_HEAD(&req->intr_entry);
49 init_waitqueue_head(&req->waitq);
50 refcount_set(&req->count, 1);
52 req->page_descs = page_descs;
53 req->max_pages = npages;
54 __set_bit(FR_PENDING, &req->flags);
57 static struct page **fuse_req_pages_alloc(unsigned int npages, gfp_t flags,
58 struct fuse_page_desc **desc)
62 pages = kzalloc(npages * (sizeof(struct page *) +
63 sizeof(struct fuse_page_desc)), flags);
64 *desc = (void *) pages + npages * sizeof(struct page *);
69 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
71 struct fuse_req *req = kmem_cache_zalloc(fuse_req_cachep, flags);
73 struct page **pages = NULL;
74 struct fuse_page_desc *page_descs = NULL;
76 WARN_ON(npages > FUSE_MAX_MAX_PAGES);
77 if (npages > FUSE_REQ_INLINE_PAGES) {
78 pages = fuse_req_pages_alloc(npages, flags,
81 kmem_cache_free(fuse_req_cachep, req);
85 pages = req->inline_pages;
86 page_descs = req->inline_page_descs;
89 fuse_request_init(req, pages, page_descs, npages);
94 struct fuse_req *fuse_request_alloc(unsigned npages)
96 return __fuse_request_alloc(npages, GFP_KERNEL);
98 EXPORT_SYMBOL_GPL(fuse_request_alloc);
100 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
102 return __fuse_request_alloc(npages, GFP_NOFS);
105 static void fuse_req_pages_free(struct fuse_req *req)
107 if (req->pages != req->inline_pages)
111 bool fuse_req_realloc_pages(struct fuse_conn *fc, struct fuse_req *req,
115 struct fuse_page_desc *page_descs;
116 unsigned int npages = min_t(unsigned int,
117 max_t(unsigned int, req->max_pages * 2,
118 FUSE_DEFAULT_MAX_PAGES_PER_REQ),
120 WARN_ON(npages <= req->max_pages);
122 pages = fuse_req_pages_alloc(npages, flags, &page_descs);
126 memcpy(pages, req->pages, sizeof(struct page *) * req->max_pages);
127 memcpy(page_descs, req->page_descs,
128 sizeof(struct fuse_page_desc) * req->max_pages);
129 fuse_req_pages_free(req);
131 req->page_descs = page_descs;
132 req->max_pages = npages;
137 void fuse_request_free(struct fuse_req *req)
139 fuse_req_pages_free(req);
140 kmem_cache_free(fuse_req_cachep, req);
143 void __fuse_get_request(struct fuse_req *req)
145 refcount_inc(&req->count);
148 /* Must be called with > 1 refcount */
149 static void __fuse_put_request(struct fuse_req *req)
151 refcount_dec(&req->count);
154 void fuse_set_initialized(struct fuse_conn *fc)
156 /* Make sure stores before this are seen on another CPU */
161 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
163 return !fc->initialized || (for_background && fc->blocked);
166 static void fuse_drop_waiting(struct fuse_conn *fc)
169 * lockess check of fc->connected is okay, because atomic_dec_and_test()
170 * provides a memory barrier mached with the one in fuse_wait_aborted()
171 * to ensure no wake-up is missed.
173 if (atomic_dec_and_test(&fc->num_waiting) &&
174 !READ_ONCE(fc->connected)) {
175 /* wake up aborters */
176 wake_up_all(&fc->blocked_waitq);
180 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
183 struct fuse_req *req;
185 atomic_inc(&fc->num_waiting);
187 if (fuse_block_alloc(fc, for_background)) {
189 if (wait_event_killable_exclusive(fc->blocked_waitq,
190 !fuse_block_alloc(fc, for_background)))
193 /* Matches smp_wmb() in fuse_set_initialized() */
204 req = fuse_request_alloc(npages);
208 wake_up(&fc->blocked_waitq);
212 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid());
213 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid());
214 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
216 __set_bit(FR_WAITING, &req->flags);
218 __set_bit(FR_BACKGROUND, &req->flags);
220 if (unlikely(req->in.h.uid == ((uid_t)-1) ||
221 req->in.h.gid == ((gid_t)-1))) {
222 fuse_put_request(fc, req);
223 return ERR_PTR(-EOVERFLOW);
228 fuse_drop_waiting(fc);
232 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
234 return __fuse_get_req(fc, npages, false);
236 EXPORT_SYMBOL_GPL(fuse_get_req);
238 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
241 return __fuse_get_req(fc, npages, true);
243 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
246 * Return request in fuse_file->reserved_req. However that may
247 * currently be in use. If that is the case, wait for it to become
250 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
253 struct fuse_req *req = NULL;
254 struct fuse_file *ff = file->private_data;
257 wait_event(fc->reserved_req_waitq, ff->reserved_req);
258 spin_lock(&fc->lock);
259 if (ff->reserved_req) {
260 req = ff->reserved_req;
261 ff->reserved_req = NULL;
262 req->stolen_file = get_file(file);
264 spin_unlock(&fc->lock);
271 * Put stolen request back into fuse_file->reserved_req
273 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
275 struct file *file = req->stolen_file;
276 struct fuse_file *ff = file->private_data;
278 WARN_ON(req->max_pages);
279 spin_lock(&fc->lock);
280 memset(req, 0, sizeof(*req));
281 fuse_request_init(req, NULL, NULL, 0);
282 BUG_ON(ff->reserved_req);
283 ff->reserved_req = req;
284 wake_up_all(&fc->reserved_req_waitq);
285 spin_unlock(&fc->lock);
290 * Gets a requests for a file operation, always succeeds
292 * This is used for sending the FLUSH request, which must get to
293 * userspace, due to POSIX locks which may need to be unlocked.
295 * If allocation fails due to OOM, use the reserved request in
298 * This is very unlikely to deadlock accidentally, since the
299 * filesystem should not have it's own file open. If deadlock is
300 * intentional, it can still be broken by "aborting" the filesystem.
302 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
305 struct fuse_req *req;
307 atomic_inc(&fc->num_waiting);
308 wait_event(fc->blocked_waitq, fc->initialized);
309 /* Matches smp_wmb() in fuse_set_initialized() */
311 req = fuse_request_alloc(0);
313 req = get_reserved_req(fc, file);
315 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid());
316 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid());
317 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
319 __set_bit(FR_WAITING, &req->flags);
320 __clear_bit(FR_BACKGROUND, &req->flags);
324 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
326 if (refcount_dec_and_test(&req->count)) {
327 if (test_bit(FR_BACKGROUND, &req->flags)) {
329 * We get here in the unlikely case that a background
330 * request was allocated but not sent
332 spin_lock(&fc->bg_lock);
334 wake_up(&fc->blocked_waitq);
335 spin_unlock(&fc->bg_lock);
338 if (test_bit(FR_WAITING, &req->flags)) {
339 __clear_bit(FR_WAITING, &req->flags);
340 fuse_drop_waiting(fc);
343 if (req->stolen_file)
344 put_reserved_req(fc, req);
346 fuse_request_free(req);
349 EXPORT_SYMBOL_GPL(fuse_put_request);
351 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
356 for (i = 0; i < numargs; i++)
357 nbytes += args[i].size;
362 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
364 fiq->reqctr += FUSE_REQ_ID_STEP;
368 static unsigned int fuse_req_hash(u64 unique)
370 return hash_long(unique & ~FUSE_INT_REQ_BIT, FUSE_PQ_HASH_BITS);
373 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
375 req->in.h.len = sizeof(struct fuse_in_header) +
376 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
377 list_add_tail(&req->list, &fiq->pending);
378 wake_up_locked(&fiq->waitq);
379 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
382 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
383 u64 nodeid, u64 nlookup)
385 struct fuse_iqueue *fiq = &fc->iq;
387 forget->forget_one.nodeid = nodeid;
388 forget->forget_one.nlookup = nlookup;
390 spin_lock(&fiq->waitq.lock);
391 if (fiq->connected) {
392 fiq->forget_list_tail->next = forget;
393 fiq->forget_list_tail = forget;
394 wake_up_locked(&fiq->waitq);
395 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
399 spin_unlock(&fiq->waitq.lock);
402 static void flush_bg_queue(struct fuse_conn *fc)
404 struct fuse_iqueue *fiq = &fc->iq;
406 while (fc->active_background < fc->max_background &&
407 !list_empty(&fc->bg_queue)) {
408 struct fuse_req *req;
410 req = list_first_entry(&fc->bg_queue, struct fuse_req, list);
411 list_del(&req->list);
412 fc->active_background++;
413 spin_lock(&fiq->waitq.lock);
414 req->in.h.unique = fuse_get_unique(fiq);
415 queue_request(fiq, req);
416 spin_unlock(&fiq->waitq.lock);
421 * This function is called when a request is finished. Either a reply
422 * has arrived or it was aborted (and not yet sent) or some error
423 * occurred during communication with userspace, or the device file
424 * was closed. The requester thread is woken up (if still waiting),
425 * the 'end' callback is called if given, else the reference to the
426 * request is released
428 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
430 struct fuse_iqueue *fiq = &fc->iq;
432 if (test_and_set_bit(FR_FINISHED, &req->flags))
435 * test_and_set_bit() implies smp_mb() between bit
436 * changing and below intr_entry check. Pairs with
437 * smp_mb() from queue_interrupt().
439 if (!list_empty(&req->intr_entry)) {
440 spin_lock(&fiq->waitq.lock);
441 list_del_init(&req->intr_entry);
442 spin_unlock(&fiq->waitq.lock);
444 WARN_ON(test_bit(FR_PENDING, &req->flags));
445 WARN_ON(test_bit(FR_SENT, &req->flags));
446 if (test_bit(FR_BACKGROUND, &req->flags)) {
447 spin_lock(&fc->bg_lock);
448 clear_bit(FR_BACKGROUND, &req->flags);
449 if (fc->num_background == fc->max_background) {
451 wake_up(&fc->blocked_waitq);
452 } else if (!fc->blocked) {
454 * Wake up next waiter, if any. It's okay to use
455 * waitqueue_active(), as we've already synced up
456 * fc->blocked with waiters with the wake_up() call
459 if (waitqueue_active(&fc->blocked_waitq))
460 wake_up(&fc->blocked_waitq);
463 if (fc->num_background == fc->congestion_threshold && fc->sb) {
464 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
465 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
467 fc->num_background--;
468 fc->active_background--;
470 spin_unlock(&fc->bg_lock);
472 wake_up(&req->waitq);
476 fuse_put_request(fc, req);
479 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
481 spin_lock(&fiq->waitq.lock);
482 if (list_empty(&req->intr_entry)) {
483 list_add_tail(&req->intr_entry, &fiq->interrupts);
485 * Pairs with smp_mb() implied by test_and_set_bit()
486 * from request_end().
489 if (test_bit(FR_FINISHED, &req->flags)) {
490 list_del_init(&req->intr_entry);
491 spin_unlock(&fiq->waitq.lock);
494 wake_up_locked(&fiq->waitq);
495 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
497 spin_unlock(&fiq->waitq.lock);
500 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
502 struct fuse_iqueue *fiq = &fc->iq;
505 if (!fc->no_interrupt) {
506 /* Any signal may interrupt this */
507 err = wait_event_interruptible(req->waitq,
508 test_bit(FR_FINISHED, &req->flags));
512 set_bit(FR_INTERRUPTED, &req->flags);
513 /* matches barrier in fuse_dev_do_read() */
514 smp_mb__after_atomic();
515 if (test_bit(FR_SENT, &req->flags))
516 queue_interrupt(fiq, req);
519 if (!test_bit(FR_FORCE, &req->flags)) {
520 /* Only fatal signals may interrupt this */
521 err = wait_event_killable(req->waitq,
522 test_bit(FR_FINISHED, &req->flags));
526 spin_lock(&fiq->waitq.lock);
527 /* Request is not yet in userspace, bail out */
528 if (test_bit(FR_PENDING, &req->flags)) {
529 list_del(&req->list);
530 spin_unlock(&fiq->waitq.lock);
531 __fuse_put_request(req);
532 req->out.h.error = -EINTR;
535 spin_unlock(&fiq->waitq.lock);
539 * Either request is already in userspace, or it was forced.
542 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
545 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
547 struct fuse_iqueue *fiq = &fc->iq;
549 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
550 spin_lock(&fiq->waitq.lock);
551 if (!fiq->connected) {
552 spin_unlock(&fiq->waitq.lock);
553 req->out.h.error = -ENOTCONN;
555 req->in.h.unique = fuse_get_unique(fiq);
556 queue_request(fiq, req);
557 /* acquire extra reference, since request is still needed
558 after request_end() */
559 __fuse_get_request(req);
560 spin_unlock(&fiq->waitq.lock);
562 request_wait_answer(fc, req);
563 /* Pairs with smp_wmb() in request_end() */
568 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
570 __set_bit(FR_ISREPLY, &req->flags);
571 if (!test_bit(FR_WAITING, &req->flags)) {
572 __set_bit(FR_WAITING, &req->flags);
573 atomic_inc(&fc->num_waiting);
575 __fuse_request_send(fc, req);
577 EXPORT_SYMBOL_GPL(fuse_request_send);
579 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
581 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
582 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
585 switch (args->in.h.opcode) {
592 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
596 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
600 if (fc->minor < 12) {
601 switch (args->in.h.opcode) {
603 args->in.args[0].size = sizeof(struct fuse_open_in);
606 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
612 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
614 struct fuse_req *req;
617 req = fuse_get_req(fc, 0);
621 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
622 fuse_adjust_compat(fc, args);
624 req->in.h.opcode = args->in.h.opcode;
625 req->in.h.nodeid = args->in.h.nodeid;
626 req->in.numargs = args->in.numargs;
627 memcpy(req->in.args, args->in.args,
628 args->in.numargs * sizeof(struct fuse_in_arg));
629 req->out.argvar = args->out.argvar;
630 req->out.numargs = args->out.numargs;
631 memcpy(req->out.args, args->out.args,
632 args->out.numargs * sizeof(struct fuse_arg));
633 fuse_request_send(fc, req);
634 ret = req->out.h.error;
635 if (!ret && args->out.argvar) {
636 BUG_ON(args->out.numargs != 1);
637 ret = req->out.args[0].size;
639 fuse_put_request(fc, req);
644 bool fuse_request_queue_background(struct fuse_conn *fc, struct fuse_req *req)
648 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags));
649 if (!test_bit(FR_WAITING, &req->flags)) {
650 __set_bit(FR_WAITING, &req->flags);
651 atomic_inc(&fc->num_waiting);
653 __set_bit(FR_ISREPLY, &req->flags);
654 spin_lock(&fc->bg_lock);
655 if (likely(fc->connected)) {
656 fc->num_background++;
657 if (fc->num_background == fc->max_background)
659 if (fc->num_background == fc->congestion_threshold && fc->sb) {
660 set_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
661 set_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
663 list_add_tail(&req->list, &fc->bg_queue);
667 spin_unlock(&fc->bg_lock);
672 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
675 if (!fuse_request_queue_background(fc, req)) {
676 req->out.h.error = -ENOTCONN;
678 fuse_put_request(fc, req);
681 EXPORT_SYMBOL_GPL(fuse_request_send_background);
683 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
684 struct fuse_req *req, u64 unique)
687 struct fuse_iqueue *fiq = &fc->iq;
689 __clear_bit(FR_ISREPLY, &req->flags);
690 req->in.h.unique = unique;
691 spin_lock(&fiq->waitq.lock);
692 if (fiq->connected) {
693 queue_request(fiq, req);
696 spin_unlock(&fiq->waitq.lock);
701 void fuse_force_forget(struct file *file, u64 nodeid)
703 struct inode *inode = file_inode(file);
704 struct fuse_conn *fc = get_fuse_conn(inode);
705 struct fuse_req *req;
706 struct fuse_forget_in inarg;
708 memset(&inarg, 0, sizeof(inarg));
710 req = fuse_get_req_nofail_nopages(fc, file);
711 req->in.h.opcode = FUSE_FORGET;
712 req->in.h.nodeid = nodeid;
714 req->in.args[0].size = sizeof(inarg);
715 req->in.args[0].value = &inarg;
716 __clear_bit(FR_ISREPLY, &req->flags);
717 __fuse_request_send(fc, req);
719 fuse_put_request(fc, req);
723 * Lock the request. Up to the next unlock_request() there mustn't be
724 * anything that could cause a page-fault. If the request was already
727 static int lock_request(struct fuse_req *req)
731 spin_lock(&req->waitq.lock);
732 if (test_bit(FR_ABORTED, &req->flags))
735 set_bit(FR_LOCKED, &req->flags);
736 spin_unlock(&req->waitq.lock);
742 * Unlock request. If it was aborted while locked, caller is responsible
743 * for unlocking and ending the request.
745 static int unlock_request(struct fuse_req *req)
749 spin_lock(&req->waitq.lock);
750 if (test_bit(FR_ABORTED, &req->flags))
753 clear_bit(FR_LOCKED, &req->flags);
754 spin_unlock(&req->waitq.lock);
759 struct fuse_copy_state {
761 struct fuse_req *req;
762 struct iov_iter *iter;
763 struct pipe_buffer *pipebufs;
764 struct pipe_buffer *currbuf;
765 struct pipe_inode_info *pipe;
766 unsigned long nr_segs;
770 unsigned move_pages:1;
773 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
774 struct iov_iter *iter)
776 memset(cs, 0, sizeof(*cs));
781 /* Unmap and put previous page of userspace buffer */
782 static void fuse_copy_finish(struct fuse_copy_state *cs)
785 struct pipe_buffer *buf = cs->currbuf;
788 buf->len = PAGE_SIZE - cs->len;
792 flush_dcache_page(cs->pg);
793 set_page_dirty_lock(cs->pg);
801 * Get another pagefull of userspace buffer, and map it to kernel
802 * address space, and lock request
804 static int fuse_copy_fill(struct fuse_copy_state *cs)
809 err = unlock_request(cs->req);
813 fuse_copy_finish(cs);
815 struct pipe_buffer *buf = cs->pipebufs;
818 err = pipe_buf_confirm(cs->pipe, buf);
822 BUG_ON(!cs->nr_segs);
825 cs->offset = buf->offset;
830 if (cs->nr_segs == cs->pipe->buffers)
833 page = alloc_page(GFP_HIGHUSER);
850 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
857 iov_iter_advance(cs->iter, err);
860 return lock_request(cs->req);
863 /* Do as much copy to/from userspace buffer as we can */
864 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
866 unsigned ncpy = min(*size, cs->len);
868 void *pgaddr = kmap_atomic(cs->pg);
869 void *buf = pgaddr + cs->offset;
872 memcpy(buf, *val, ncpy);
874 memcpy(*val, buf, ncpy);
876 kunmap_atomic(pgaddr);
885 static int fuse_check_page(struct page *page)
887 if (page_mapcount(page) ||
888 page->mapping != NULL ||
889 page_count(page) != 1 ||
890 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
897 printk(KERN_WARNING "fuse: trying to steal weird page\n");
898 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);
904 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
907 struct page *oldpage = *pagep;
908 struct page *newpage;
909 struct pipe_buffer *buf = cs->pipebufs;
911 err = unlock_request(cs->req);
915 fuse_copy_finish(cs);
917 err = pipe_buf_confirm(cs->pipe, buf);
921 BUG_ON(!cs->nr_segs);
927 if (cs->len != PAGE_SIZE)
930 if (pipe_buf_steal(cs->pipe, buf) != 0)
935 if (!PageUptodate(newpage))
936 SetPageUptodate(newpage);
938 ClearPageMappedToDisk(newpage);
940 if (fuse_check_page(newpage) != 0)
941 goto out_fallback_unlock;
944 * This is a new and locked page, it shouldn't be mapped or
945 * have any special flags on it
947 if (WARN_ON(page_mapped(oldpage)))
948 goto out_fallback_unlock;
949 if (WARN_ON(page_has_private(oldpage)))
950 goto out_fallback_unlock;
951 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
952 goto out_fallback_unlock;
953 if (WARN_ON(PageMlocked(oldpage)))
954 goto out_fallback_unlock;
956 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
958 unlock_page(newpage);
964 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
965 lru_cache_add_file(newpage);
968 spin_lock(&cs->req->waitq.lock);
969 if (test_bit(FR_ABORTED, &cs->req->flags))
973 spin_unlock(&cs->req->waitq.lock);
976 unlock_page(newpage);
981 unlock_page(oldpage);
988 unlock_page(newpage);
991 cs->offset = buf->offset;
993 err = lock_request(cs->req);
1000 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
1001 unsigned offset, unsigned count)
1003 struct pipe_buffer *buf;
1006 if (cs->nr_segs == cs->pipe->buffers)
1009 err = unlock_request(cs->req);
1013 fuse_copy_finish(cs);
1018 buf->offset = offset;
1029 * Copy a page in the request to/from the userspace buffer. Must be
1032 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
1033 unsigned offset, unsigned count, int zeroing)
1036 struct page *page = *pagep;
1038 if (page && zeroing && count < PAGE_SIZE)
1039 clear_highpage(page);
1042 if (cs->write && cs->pipebufs && page) {
1043 return fuse_ref_page(cs, page, offset, count);
1044 } else if (!cs->len) {
1045 if (cs->move_pages && page &&
1046 offset == 0 && count == PAGE_SIZE) {
1047 err = fuse_try_move_page(cs, pagep);
1051 err = fuse_copy_fill(cs);
1057 void *mapaddr = kmap_atomic(page);
1058 void *buf = mapaddr + offset;
1059 offset += fuse_copy_do(cs, &buf, &count);
1060 kunmap_atomic(mapaddr);
1062 offset += fuse_copy_do(cs, NULL, &count);
1064 if (page && !cs->write)
1065 flush_dcache_page(page);
1069 /* Copy pages in the request to/from userspace buffer */
1070 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1074 struct fuse_req *req = cs->req;
1076 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1078 unsigned offset = req->page_descs[i].offset;
1079 unsigned count = min(nbytes, req->page_descs[i].length);
1081 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1091 /* Copy a single argument in the request to/from userspace buffer */
1092 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1096 int err = fuse_copy_fill(cs);
1100 fuse_copy_do(cs, &val, &size);
1105 /* Copy request arguments to/from userspace buffer */
1106 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1107 unsigned argpages, struct fuse_arg *args,
1113 for (i = 0; !err && i < numargs; i++) {
1114 struct fuse_arg *arg = &args[i];
1115 if (i == numargs - 1 && argpages)
1116 err = fuse_copy_pages(cs, arg->size, zeroing);
1118 err = fuse_copy_one(cs, arg->value, arg->size);
1123 static int forget_pending(struct fuse_iqueue *fiq)
1125 return fiq->forget_list_head.next != NULL;
1128 static int request_pending(struct fuse_iqueue *fiq)
1130 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1131 forget_pending(fiq);
1135 * Transfer an interrupt request to userspace
1137 * Unlike other requests this is assembled on demand, without a need
1138 * to allocate a separate fuse_req structure.
1140 * Called with fiq->waitq.lock held, releases it
1142 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1143 struct fuse_copy_state *cs,
1144 size_t nbytes, struct fuse_req *req)
1145 __releases(fiq->waitq.lock)
1147 struct fuse_in_header ih;
1148 struct fuse_interrupt_in arg;
1149 unsigned reqsize = sizeof(ih) + sizeof(arg);
1152 list_del_init(&req->intr_entry);
1153 memset(&ih, 0, sizeof(ih));
1154 memset(&arg, 0, sizeof(arg));
1156 ih.opcode = FUSE_INTERRUPT;
1157 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT);
1158 arg.unique = req->in.h.unique;
1160 spin_unlock(&fiq->waitq.lock);
1161 if (nbytes < reqsize)
1164 err = fuse_copy_one(cs, &ih, sizeof(ih));
1166 err = fuse_copy_one(cs, &arg, sizeof(arg));
1167 fuse_copy_finish(cs);
1169 return err ? err : reqsize;
1172 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1176 struct fuse_forget_link *head = fiq->forget_list_head.next;
1177 struct fuse_forget_link **newhead = &head;
1180 for (count = 0; *newhead != NULL && count < max; count++)
1181 newhead = &(*newhead)->next;
1183 fiq->forget_list_head.next = *newhead;
1185 if (fiq->forget_list_head.next == NULL)
1186 fiq->forget_list_tail = &fiq->forget_list_head;
1194 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1195 struct fuse_copy_state *cs,
1197 __releases(fiq->waitq.lock)
1200 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1201 struct fuse_forget_in arg = {
1202 .nlookup = forget->forget_one.nlookup,
1204 struct fuse_in_header ih = {
1205 .opcode = FUSE_FORGET,
1206 .nodeid = forget->forget_one.nodeid,
1207 .unique = fuse_get_unique(fiq),
1208 .len = sizeof(ih) + sizeof(arg),
1211 spin_unlock(&fiq->waitq.lock);
1213 if (nbytes < ih.len)
1216 err = fuse_copy_one(cs, &ih, sizeof(ih));
1218 err = fuse_copy_one(cs, &arg, sizeof(arg));
1219 fuse_copy_finish(cs);
1227 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1228 struct fuse_copy_state *cs, size_t nbytes)
1229 __releases(fiq->waitq.lock)
1232 unsigned max_forgets;
1234 struct fuse_forget_link *head;
1235 struct fuse_batch_forget_in arg = { .count = 0 };
1236 struct fuse_in_header ih = {
1237 .opcode = FUSE_BATCH_FORGET,
1238 .unique = fuse_get_unique(fiq),
1239 .len = sizeof(ih) + sizeof(arg),
1242 if (nbytes < ih.len) {
1243 spin_unlock(&fiq->waitq.lock);
1247 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1248 head = dequeue_forget(fiq, max_forgets, &count);
1249 spin_unlock(&fiq->waitq.lock);
1252 ih.len += count * sizeof(struct fuse_forget_one);
1253 err = fuse_copy_one(cs, &ih, sizeof(ih));
1255 err = fuse_copy_one(cs, &arg, sizeof(arg));
1258 struct fuse_forget_link *forget = head;
1261 err = fuse_copy_one(cs, &forget->forget_one,
1262 sizeof(forget->forget_one));
1264 head = forget->next;
1268 fuse_copy_finish(cs);
1276 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1277 struct fuse_copy_state *cs,
1279 __releases(fiq->waitq.lock)
1281 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1282 return fuse_read_single_forget(fiq, cs, nbytes);
1284 return fuse_read_batch_forget(fiq, cs, nbytes);
1288 * Read a single request into the userspace filesystem's buffer. This
1289 * function waits until a request is available, then removes it from
1290 * the pending list and copies request data to userspace buffer. If
1291 * no reply is needed (FORGET) or request has been aborted or there
1292 * was an error during the copying then it's finished by calling
1293 * request_end(). Otherwise add it to the processing list, and set
1296 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1297 struct fuse_copy_state *cs, size_t nbytes)
1300 struct fuse_conn *fc = fud->fc;
1301 struct fuse_iqueue *fiq = &fc->iq;
1302 struct fuse_pqueue *fpq = &fud->pq;
1303 struct fuse_req *req;
1309 spin_lock(&fiq->waitq.lock);
1311 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1312 !request_pending(fiq))
1315 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1316 !fiq->connected || request_pending(fiq));
1320 if (!fiq->connected) {
1321 err = (fc->aborted && fc->abort_err) ? -ECONNABORTED : -ENODEV;
1325 if (!list_empty(&fiq->interrupts)) {
1326 req = list_entry(fiq->interrupts.next, struct fuse_req,
1328 return fuse_read_interrupt(fiq, cs, nbytes, req);
1331 if (forget_pending(fiq)) {
1332 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1333 return fuse_read_forget(fc, fiq, cs, nbytes);
1335 if (fiq->forget_batch <= -8)
1336 fiq->forget_batch = 16;
1339 req = list_entry(fiq->pending.next, struct fuse_req, list);
1340 clear_bit(FR_PENDING, &req->flags);
1341 list_del_init(&req->list);
1342 spin_unlock(&fiq->waitq.lock);
1345 reqsize = in->h.len;
1347 /* If request is too large, reply with an error and restart the read */
1348 if (nbytes < reqsize) {
1349 req->out.h.error = -EIO;
1350 /* SETXATTR is special, since it may contain too large data */
1351 if (in->h.opcode == FUSE_SETXATTR)
1352 req->out.h.error = -E2BIG;
1353 request_end(fc, req);
1356 spin_lock(&fpq->lock);
1357 list_add(&req->list, &fpq->io);
1358 spin_unlock(&fpq->lock);
1360 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1362 err = fuse_copy_args(cs, in->numargs, in->argpages,
1363 (struct fuse_arg *) in->args, 0);
1364 fuse_copy_finish(cs);
1365 spin_lock(&fpq->lock);
1366 clear_bit(FR_LOCKED, &req->flags);
1367 if (!fpq->connected) {
1368 err = (fc->aborted && fc->abort_err) ? -ECONNABORTED : -ENODEV;
1372 req->out.h.error = -EIO;
1375 if (!test_bit(FR_ISREPLY, &req->flags)) {
1379 hash = fuse_req_hash(req->in.h.unique);
1380 list_move_tail(&req->list, &fpq->processing[hash]);
1381 __fuse_get_request(req);
1382 set_bit(FR_SENT, &req->flags);
1383 spin_unlock(&fpq->lock);
1384 /* matches barrier in request_wait_answer() */
1385 smp_mb__after_atomic();
1386 if (test_bit(FR_INTERRUPTED, &req->flags))
1387 queue_interrupt(fiq, req);
1388 fuse_put_request(fc, req);
1393 if (!test_bit(FR_PRIVATE, &req->flags))
1394 list_del_init(&req->list);
1395 spin_unlock(&fpq->lock);
1396 request_end(fc, req);
1400 spin_unlock(&fiq->waitq.lock);
1404 static int fuse_dev_open(struct inode *inode, struct file *file)
1407 * The fuse device's file's private_data is used to hold
1408 * the fuse_conn(ection) when it is mounted, and is used to
1409 * keep track of whether the file has been mounted already.
1411 file->private_data = NULL;
1415 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1417 struct fuse_copy_state cs;
1418 struct file *file = iocb->ki_filp;
1419 struct fuse_dev *fud = fuse_get_dev(file);
1424 if (!iter_is_iovec(to))
1427 fuse_copy_init(&cs, 1, to);
1429 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1432 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1433 struct pipe_inode_info *pipe,
1434 size_t len, unsigned int flags)
1438 struct pipe_buffer *bufs;
1439 struct fuse_copy_state cs;
1440 struct fuse_dev *fud = fuse_get_dev(in);
1445 bufs = kvmalloc_array(pipe->buffers, sizeof(struct pipe_buffer),
1450 fuse_copy_init(&cs, 1, NULL);
1453 ret = fuse_dev_do_read(fud, in, &cs, len);
1457 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1462 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1464 * Need to be careful about this. Having buf->ops in module
1465 * code can Oops if the buffer persists after module unload.
1467 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1468 bufs[page_nr].flags = 0;
1469 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1470 if (unlikely(ret < 0))
1476 for (; page_nr < cs.nr_segs; page_nr++)
1477 put_page(bufs[page_nr].page);
1483 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1484 struct fuse_copy_state *cs)
1486 struct fuse_notify_poll_wakeup_out outarg;
1489 if (size != sizeof(outarg))
1492 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1496 fuse_copy_finish(cs);
1497 return fuse_notify_poll_wakeup(fc, &outarg);
1500 fuse_copy_finish(cs);
1504 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1505 struct fuse_copy_state *cs)
1507 struct fuse_notify_inval_inode_out outarg;
1510 if (size != sizeof(outarg))
1513 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1516 fuse_copy_finish(cs);
1518 down_read(&fc->killsb);
1521 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1522 outarg.off, outarg.len);
1524 up_read(&fc->killsb);
1528 fuse_copy_finish(cs);
1532 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1533 struct fuse_copy_state *cs)
1535 struct fuse_notify_inval_entry_out outarg;
1540 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1545 if (size < sizeof(outarg))
1548 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1552 err = -ENAMETOOLONG;
1553 if (outarg.namelen > FUSE_NAME_MAX)
1557 if (size != sizeof(outarg) + outarg.namelen + 1)
1561 name.len = outarg.namelen;
1562 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1565 fuse_copy_finish(cs);
1566 buf[outarg.namelen] = 0;
1568 down_read(&fc->killsb);
1571 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1572 up_read(&fc->killsb);
1578 fuse_copy_finish(cs);
1582 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1583 struct fuse_copy_state *cs)
1585 struct fuse_notify_delete_out outarg;
1590 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1595 if (size < sizeof(outarg))
1598 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1602 err = -ENAMETOOLONG;
1603 if (outarg.namelen > FUSE_NAME_MAX)
1607 if (size != sizeof(outarg) + outarg.namelen + 1)
1611 name.len = outarg.namelen;
1612 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1615 fuse_copy_finish(cs);
1616 buf[outarg.namelen] = 0;
1618 down_read(&fc->killsb);
1621 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1622 outarg.child, &name);
1623 up_read(&fc->killsb);
1629 fuse_copy_finish(cs);
1633 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1634 struct fuse_copy_state *cs)
1636 struct fuse_notify_store_out outarg;
1637 struct inode *inode;
1638 struct address_space *mapping;
1642 unsigned int offset;
1648 if (size < sizeof(outarg))
1651 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1656 if (size - sizeof(outarg) != outarg.size)
1659 nodeid = outarg.nodeid;
1661 down_read(&fc->killsb);
1667 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1671 mapping = inode->i_mapping;
1672 index = outarg.offset >> PAGE_SHIFT;
1673 offset = outarg.offset & ~PAGE_MASK;
1674 file_size = i_size_read(inode);
1675 end = outarg.offset + outarg.size;
1676 if (end > file_size) {
1678 fuse_write_update_size(inode, file_size);
1684 unsigned int this_num;
1687 page = find_or_create_page(mapping, index,
1688 mapping_gfp_mask(mapping));
1692 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1693 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1694 if (!err && offset == 0 &&
1695 (this_num == PAGE_SIZE || file_size == end))
1696 SetPageUptodate(page);
1713 up_read(&fc->killsb);
1715 fuse_copy_finish(cs);
1719 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1721 release_pages(req->pages, req->num_pages);
1724 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1725 struct fuse_notify_retrieve_out *outarg)
1728 struct address_space *mapping = inode->i_mapping;
1729 struct fuse_req *req;
1733 unsigned int offset;
1734 size_t total_len = 0;
1735 unsigned int num_pages;
1737 offset = outarg->offset & ~PAGE_MASK;
1738 file_size = i_size_read(inode);
1741 if (outarg->offset > file_size)
1743 else if (outarg->offset + num > file_size)
1744 num = file_size - outarg->offset;
1746 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1747 num_pages = min(num_pages, fc->max_pages);
1749 req = fuse_get_req(fc, num_pages);
1751 return PTR_ERR(req);
1753 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1754 req->in.h.nodeid = outarg->nodeid;
1755 req->in.numargs = 2;
1756 req->in.argpages = 1;
1757 req->end = fuse_retrieve_end;
1759 index = outarg->offset >> PAGE_SHIFT;
1761 while (num && req->num_pages < num_pages) {
1763 unsigned int this_num;
1765 page = find_get_page(mapping, index);
1769 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1770 req->pages[req->num_pages] = page;
1771 req->page_descs[req->num_pages].offset = offset;
1772 req->page_descs[req->num_pages].length = this_num;
1777 total_len += this_num;
1780 req->misc.retrieve_in.offset = outarg->offset;
1781 req->misc.retrieve_in.size = total_len;
1782 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1783 req->in.args[0].value = &req->misc.retrieve_in;
1784 req->in.args[1].size = total_len;
1786 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1788 fuse_retrieve_end(fc, req);
1789 fuse_put_request(fc, req);
1795 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1796 struct fuse_copy_state *cs)
1798 struct fuse_notify_retrieve_out outarg;
1799 struct inode *inode;
1803 if (size != sizeof(outarg))
1806 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1810 fuse_copy_finish(cs);
1812 down_read(&fc->killsb);
1815 u64 nodeid = outarg.nodeid;
1817 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1819 err = fuse_retrieve(fc, inode, &outarg);
1823 up_read(&fc->killsb);
1828 fuse_copy_finish(cs);
1832 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1833 unsigned int size, struct fuse_copy_state *cs)
1835 /* Don't try to move pages (yet) */
1839 case FUSE_NOTIFY_POLL:
1840 return fuse_notify_poll(fc, size, cs);
1842 case FUSE_NOTIFY_INVAL_INODE:
1843 return fuse_notify_inval_inode(fc, size, cs);
1845 case FUSE_NOTIFY_INVAL_ENTRY:
1846 return fuse_notify_inval_entry(fc, size, cs);
1848 case FUSE_NOTIFY_STORE:
1849 return fuse_notify_store(fc, size, cs);
1851 case FUSE_NOTIFY_RETRIEVE:
1852 return fuse_notify_retrieve(fc, size, cs);
1854 case FUSE_NOTIFY_DELETE:
1855 return fuse_notify_delete(fc, size, cs);
1858 fuse_copy_finish(cs);
1863 /* Look up request on processing list by unique ID */
1864 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1866 unsigned int hash = fuse_req_hash(unique);
1867 struct fuse_req *req;
1869 list_for_each_entry(req, &fpq->processing[hash], list) {
1870 if (req->in.h.unique == unique)
1876 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1879 unsigned reqsize = sizeof(struct fuse_out_header);
1882 return nbytes != reqsize ? -EINVAL : 0;
1884 reqsize += len_args(out->numargs, out->args);
1886 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1888 else if (reqsize > nbytes) {
1889 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1890 unsigned diffsize = reqsize - nbytes;
1891 if (diffsize > lastarg->size)
1893 lastarg->size -= diffsize;
1895 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1900 * Write a single reply to a request. First the header is copied from
1901 * the write buffer. The request is then searched on the processing
1902 * list by the unique ID found in the header. If found, then remove
1903 * it from the list and copy the rest of the buffer to the request.
1904 * The request is finished by calling request_end()
1906 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1907 struct fuse_copy_state *cs, size_t nbytes)
1910 struct fuse_conn *fc = fud->fc;
1911 struct fuse_pqueue *fpq = &fud->pq;
1912 struct fuse_req *req;
1913 struct fuse_out_header oh;
1915 if (nbytes < sizeof(struct fuse_out_header))
1918 err = fuse_copy_one(cs, &oh, sizeof(oh));
1923 if (oh.len != nbytes)
1927 * Zero oh.unique indicates unsolicited notification message
1928 * and error contains notification code.
1931 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1932 return err ? err : nbytes;
1936 if (oh.error <= -1000 || oh.error > 0)
1939 spin_lock(&fpq->lock);
1941 if (!fpq->connected)
1944 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
1948 /* Is it an interrupt reply ID? */
1949 if (oh.unique & FUSE_INT_REQ_BIT) {
1950 __fuse_get_request(req);
1951 spin_unlock(&fpq->lock);
1954 if (nbytes != sizeof(struct fuse_out_header)) {
1955 fuse_put_request(fc, req);
1959 if (oh.error == -ENOSYS)
1960 fc->no_interrupt = 1;
1961 else if (oh.error == -EAGAIN)
1962 queue_interrupt(&fc->iq, req);
1963 fuse_put_request(fc, req);
1965 fuse_copy_finish(cs);
1969 clear_bit(FR_SENT, &req->flags);
1970 list_move(&req->list, &fpq->io);
1972 set_bit(FR_LOCKED, &req->flags);
1973 spin_unlock(&fpq->lock);
1975 if (!req->out.page_replace)
1978 err = copy_out_args(cs, &req->out, nbytes);
1979 fuse_copy_finish(cs);
1981 spin_lock(&fpq->lock);
1982 clear_bit(FR_LOCKED, &req->flags);
1983 if (!fpq->connected)
1986 req->out.h.error = -EIO;
1987 if (!test_bit(FR_PRIVATE, &req->flags))
1988 list_del_init(&req->list);
1989 spin_unlock(&fpq->lock);
1991 request_end(fc, req);
1993 return err ? err : nbytes;
1996 spin_unlock(&fpq->lock);
1998 fuse_copy_finish(cs);
2002 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
2004 struct fuse_copy_state cs;
2005 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
2010 if (!iter_is_iovec(from))
2013 fuse_copy_init(&cs, 0, from);
2015 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
2018 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
2019 struct file *out, loff_t *ppos,
2020 size_t len, unsigned int flags)
2024 struct pipe_buffer *bufs;
2025 struct fuse_copy_state cs;
2026 struct fuse_dev *fud;
2030 fud = fuse_get_dev(out);
2036 bufs = kvmalloc_array(pipe->nrbufs, sizeof(struct pipe_buffer),
2045 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
2046 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
2056 struct pipe_buffer *ibuf;
2057 struct pipe_buffer *obuf;
2059 BUG_ON(nbuf >= pipe->buffers);
2060 BUG_ON(!pipe->nrbufs);
2061 ibuf = &pipe->bufs[pipe->curbuf];
2064 if (rem >= ibuf->len) {
2067 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2070 pipe_buf_get(pipe, ibuf);
2072 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2074 ibuf->offset += obuf->len;
2075 ibuf->len -= obuf->len;
2082 fuse_copy_init(&cs, 0, NULL);
2087 if (flags & SPLICE_F_MOVE)
2090 ret = fuse_dev_do_write(fud, &cs, len);
2093 for (idx = 0; idx < nbuf; idx++)
2094 pipe_buf_release(pipe, &bufs[idx]);
2102 static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2104 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
2105 struct fuse_iqueue *fiq;
2106 struct fuse_dev *fud = fuse_get_dev(file);
2112 poll_wait(file, &fiq->waitq, wait);
2114 spin_lock(&fiq->waitq.lock);
2115 if (!fiq->connected)
2117 else if (request_pending(fiq))
2118 mask |= EPOLLIN | EPOLLRDNORM;
2119 spin_unlock(&fiq->waitq.lock);
2124 /* Abort all requests on the given list (pending or processing) */
2125 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2127 while (!list_empty(head)) {
2128 struct fuse_req *req;
2129 req = list_entry(head->next, struct fuse_req, list);
2130 req->out.h.error = -ECONNABORTED;
2131 clear_bit(FR_SENT, &req->flags);
2132 list_del_init(&req->list);
2133 request_end(fc, req);
2137 static void end_polls(struct fuse_conn *fc)
2141 p = rb_first(&fc->polled_files);
2144 struct fuse_file *ff;
2145 ff = rb_entry(p, struct fuse_file, polled_node);
2146 wake_up_interruptible_all(&ff->poll_wait);
2153 * Abort all requests.
2155 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2158 * The same effect is usually achievable through killing the filesystem daemon
2159 * and all users of the filesystem. The exception is the combination of an
2160 * asynchronous request and the tricky deadlock (see
2161 * Documentation/filesystems/fuse.txt).
2163 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2164 * requests, they should be finished off immediately. Locked requests will be
2165 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2166 * requests. It is possible that some request will finish before we can. This
2167 * is OK, the request will in that case be removed from the list before we touch
2170 void fuse_abort_conn(struct fuse_conn *fc, bool is_abort)
2172 struct fuse_iqueue *fiq = &fc->iq;
2174 spin_lock(&fc->lock);
2175 if (fc->connected) {
2176 struct fuse_dev *fud;
2177 struct fuse_req *req, *next;
2181 /* Background queuing checks fc->connected under bg_lock */
2182 spin_lock(&fc->bg_lock);
2184 spin_unlock(&fc->bg_lock);
2186 fc->aborted = is_abort;
2187 fuse_set_initialized(fc);
2188 list_for_each_entry(fud, &fc->devices, entry) {
2189 struct fuse_pqueue *fpq = &fud->pq;
2191 spin_lock(&fpq->lock);
2193 list_for_each_entry_safe(req, next, &fpq->io, list) {
2194 req->out.h.error = -ECONNABORTED;
2195 spin_lock(&req->waitq.lock);
2196 set_bit(FR_ABORTED, &req->flags);
2197 if (!test_bit(FR_LOCKED, &req->flags)) {
2198 set_bit(FR_PRIVATE, &req->flags);
2199 __fuse_get_request(req);
2200 list_move(&req->list, &to_end);
2202 spin_unlock(&req->waitq.lock);
2204 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2205 list_splice_tail_init(&fpq->processing[i],
2207 spin_unlock(&fpq->lock);
2209 spin_lock(&fc->bg_lock);
2211 fc->max_background = UINT_MAX;
2213 spin_unlock(&fc->bg_lock);
2215 spin_lock(&fiq->waitq.lock);
2217 list_for_each_entry(req, &fiq->pending, list)
2218 clear_bit(FR_PENDING, &req->flags);
2219 list_splice_tail_init(&fiq->pending, &to_end);
2220 while (forget_pending(fiq))
2221 kfree(dequeue_forget(fiq, 1, NULL));
2222 wake_up_all_locked(&fiq->waitq);
2223 spin_unlock(&fiq->waitq.lock);
2224 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2226 wake_up_all(&fc->blocked_waitq);
2227 spin_unlock(&fc->lock);
2229 end_requests(fc, &to_end);
2231 spin_unlock(&fc->lock);
2234 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2236 void fuse_wait_aborted(struct fuse_conn *fc)
2238 /* matches implicit memory barrier in fuse_drop_waiting() */
2240 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2243 int fuse_dev_release(struct inode *inode, struct file *file)
2245 struct fuse_dev *fud = fuse_get_dev(file);
2248 struct fuse_conn *fc = fud->fc;
2249 struct fuse_pqueue *fpq = &fud->pq;
2253 spin_lock(&fpq->lock);
2254 WARN_ON(!list_empty(&fpq->io));
2255 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2256 list_splice_init(&fpq->processing[i], &to_end);
2257 spin_unlock(&fpq->lock);
2259 end_requests(fc, &to_end);
2261 /* Are we the last open device? */
2262 if (atomic_dec_and_test(&fc->dev_count)) {
2263 WARN_ON(fc->iq.fasync != NULL);
2264 fuse_abort_conn(fc, false);
2270 EXPORT_SYMBOL_GPL(fuse_dev_release);
2272 static int fuse_dev_fasync(int fd, struct file *file, int on)
2274 struct fuse_dev *fud = fuse_get_dev(file);
2279 /* No locking - fasync_helper does its own locking */
2280 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2283 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2285 struct fuse_dev *fud;
2287 if (new->private_data)
2290 fud = fuse_dev_alloc(fc);
2294 new->private_data = fud;
2295 atomic_inc(&fc->dev_count);
2300 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2305 if (cmd == FUSE_DEV_IOC_CLONE) {
2309 if (!get_user(oldfd, (__u32 __user *) arg)) {
2310 struct file *old = fget(oldfd);
2314 struct fuse_dev *fud = NULL;
2317 * Check against file->f_op because CUSE
2318 * uses the same ioctl handler.
2320 if (old->f_op == file->f_op &&
2321 old->f_cred->user_ns == file->f_cred->user_ns)
2322 fud = fuse_get_dev(old);
2325 mutex_lock(&fuse_mutex);
2326 err = fuse_device_clone(fud->fc, file);
2327 mutex_unlock(&fuse_mutex);
2336 const struct file_operations fuse_dev_operations = {
2337 .owner = THIS_MODULE,
2338 .open = fuse_dev_open,
2339 .llseek = no_llseek,
2340 .read_iter = fuse_dev_read,
2341 .splice_read = fuse_dev_splice_read,
2342 .write_iter = fuse_dev_write,
2343 .splice_write = fuse_dev_splice_write,
2344 .poll = fuse_dev_poll,
2345 .release = fuse_dev_release,
2346 .fasync = fuse_dev_fasync,
2347 .unlocked_ioctl = fuse_dev_ioctl,
2348 .compat_ioctl = fuse_dev_ioctl,
2350 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2352 static struct miscdevice fuse_miscdevice = {
2353 .minor = FUSE_MINOR,
2355 .fops = &fuse_dev_operations,
2358 int __init fuse_dev_init(void)
2361 fuse_req_cachep = kmem_cache_create("fuse_request",
2362 sizeof(struct fuse_req),
2364 if (!fuse_req_cachep)
2367 err = misc_register(&fuse_miscdevice);
2369 goto out_cache_clean;
2374 kmem_cache_destroy(fuse_req_cachep);
2379 void fuse_dev_cleanup(void)
2381 misc_deregister(&fuse_miscdevice);
2382 kmem_cache_destroy(fuse_req_cachep);