2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/uio.h>
15 #include <linux/miscdevice.h>
16 #include <linux/namei.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/freezer.h>
25 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26 MODULE_ALIAS("devname:fuse");
28 static struct kmem_cache *fuse_req_cachep;
30 static struct fuse_dev *fuse_get_dev(struct file *file)
33 * Lockless access is OK, because file->private data is set
34 * once during mount and is valid until the file is released.
36 return ACCESS_ONCE(file->private_data);
39 static void fuse_request_init(struct fuse_req *req, struct page **pages,
40 struct fuse_page_desc *page_descs,
43 memset(req, 0, sizeof(*req));
44 memset(pages, 0, sizeof(*pages) * npages);
45 memset(page_descs, 0, sizeof(*page_descs) * npages);
46 INIT_LIST_HEAD(&req->list);
47 INIT_LIST_HEAD(&req->intr_entry);
48 init_waitqueue_head(&req->waitq);
49 atomic_set(&req->count, 1);
51 req->page_descs = page_descs;
52 req->max_pages = npages;
53 __set_bit(FR_PENDING, &req->flags);
56 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
58 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
61 struct fuse_page_desc *page_descs;
63 if (npages <= FUSE_REQ_INLINE_PAGES) {
64 pages = req->inline_pages;
65 page_descs = req->inline_page_descs;
67 pages = kmalloc(sizeof(struct page *) * npages, flags);
68 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
72 if (!pages || !page_descs) {
75 kmem_cache_free(fuse_req_cachep, req);
79 fuse_request_init(req, pages, page_descs, npages);
84 struct fuse_req *fuse_request_alloc(unsigned npages)
86 return __fuse_request_alloc(npages, GFP_KERNEL);
88 EXPORT_SYMBOL_GPL(fuse_request_alloc);
90 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
92 return __fuse_request_alloc(npages, GFP_NOFS);
95 void fuse_request_free(struct fuse_req *req)
97 if (req->pages != req->inline_pages) {
99 kfree(req->page_descs);
101 kmem_cache_free(fuse_req_cachep, req);
104 static void block_sigs(sigset_t *oldset)
108 siginitsetinv(&mask, sigmask(SIGKILL));
109 sigprocmask(SIG_BLOCK, &mask, oldset);
112 static void restore_sigs(sigset_t *oldset)
114 sigprocmask(SIG_SETMASK, oldset, NULL);
117 void __fuse_get_request(struct fuse_req *req)
119 atomic_inc(&req->count);
122 /* Must be called with > 1 refcount */
123 static void __fuse_put_request(struct fuse_req *req)
125 BUG_ON(atomic_read(&req->count) < 2);
126 atomic_dec(&req->count);
129 static void fuse_req_init_context(struct fuse_req *req)
131 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
132 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
133 req->in.h.pid = current->pid;
136 void fuse_set_initialized(struct fuse_conn *fc)
138 /* Make sure stores before this are seen on another CPU */
143 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
145 return !fc->initialized || (for_background && fc->blocked);
148 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
151 struct fuse_req *req;
153 atomic_inc(&fc->num_waiting);
155 if (fuse_block_alloc(fc, for_background)) {
160 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
161 !fuse_block_alloc(fc, for_background));
162 restore_sigs(&oldset);
167 /* Matches smp_wmb() in fuse_set_initialized() */
178 req = fuse_request_alloc(npages);
182 wake_up(&fc->blocked_waitq);
186 fuse_req_init_context(req);
187 __set_bit(FR_WAITING, &req->flags);
189 __set_bit(FR_BACKGROUND, &req->flags);
194 atomic_dec(&fc->num_waiting);
198 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
200 return __fuse_get_req(fc, npages, false);
202 EXPORT_SYMBOL_GPL(fuse_get_req);
204 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
207 return __fuse_get_req(fc, npages, true);
209 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
212 * Return request in fuse_file->reserved_req. However that may
213 * currently be in use. If that is the case, wait for it to become
216 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
219 struct fuse_req *req = NULL;
220 struct fuse_file *ff = file->private_data;
223 wait_event(fc->reserved_req_waitq, ff->reserved_req);
224 spin_lock(&fc->lock);
225 if (ff->reserved_req) {
226 req = ff->reserved_req;
227 ff->reserved_req = NULL;
228 req->stolen_file = get_file(file);
230 spin_unlock(&fc->lock);
237 * Put stolen request back into fuse_file->reserved_req
239 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
241 struct file *file = req->stolen_file;
242 struct fuse_file *ff = file->private_data;
244 spin_lock(&fc->lock);
245 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
246 BUG_ON(ff->reserved_req);
247 ff->reserved_req = req;
248 wake_up_all(&fc->reserved_req_waitq);
249 spin_unlock(&fc->lock);
254 * Gets a requests for a file operation, always succeeds
256 * This is used for sending the FLUSH request, which must get to
257 * userspace, due to POSIX locks which may need to be unlocked.
259 * If allocation fails due to OOM, use the reserved request in
262 * This is very unlikely to deadlock accidentally, since the
263 * filesystem should not have it's own file open. If deadlock is
264 * intentional, it can still be broken by "aborting" the filesystem.
266 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
269 struct fuse_req *req;
271 atomic_inc(&fc->num_waiting);
272 wait_event(fc->blocked_waitq, fc->initialized);
273 /* Matches smp_wmb() in fuse_set_initialized() */
275 req = fuse_request_alloc(0);
277 req = get_reserved_req(fc, file);
279 fuse_req_init_context(req);
280 __set_bit(FR_WAITING, &req->flags);
281 __clear_bit(FR_BACKGROUND, &req->flags);
285 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
287 if (atomic_dec_and_test(&req->count)) {
288 if (test_bit(FR_BACKGROUND, &req->flags)) {
290 * We get here in the unlikely case that a background
291 * request was allocated but not sent
293 spin_lock(&fc->lock);
295 wake_up(&fc->blocked_waitq);
296 spin_unlock(&fc->lock);
299 if (test_bit(FR_WAITING, &req->flags)) {
300 __clear_bit(FR_WAITING, &req->flags);
301 atomic_dec(&fc->num_waiting);
304 if (req->stolen_file)
305 put_reserved_req(fc, req);
307 fuse_request_free(req);
310 EXPORT_SYMBOL_GPL(fuse_put_request);
312 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
317 for (i = 0; i < numargs; i++)
318 nbytes += args[i].size;
323 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
325 return ++fiq->reqctr;
328 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
330 req->in.h.len = sizeof(struct fuse_in_header) +
331 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
332 list_add_tail(&req->list, &fiq->pending);
333 wake_up_locked(&fiq->waitq);
334 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
337 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
338 u64 nodeid, u64 nlookup)
340 struct fuse_iqueue *fiq = &fc->iq;
342 forget->forget_one.nodeid = nodeid;
343 forget->forget_one.nlookup = nlookup;
345 spin_lock(&fiq->waitq.lock);
346 if (fiq->connected) {
347 fiq->forget_list_tail->next = forget;
348 fiq->forget_list_tail = forget;
349 wake_up_locked(&fiq->waitq);
350 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
354 spin_unlock(&fiq->waitq.lock);
357 static void flush_bg_queue(struct fuse_conn *fc)
359 while (fc->active_background < fc->max_background &&
360 !list_empty(&fc->bg_queue)) {
361 struct fuse_req *req;
362 struct fuse_iqueue *fiq = &fc->iq;
364 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
365 list_del(&req->list);
366 fc->active_background++;
367 spin_lock(&fiq->waitq.lock);
368 req->in.h.unique = fuse_get_unique(fiq);
369 queue_request(fiq, req);
370 spin_unlock(&fiq->waitq.lock);
375 * This function is called when a request is finished. Either a reply
376 * has arrived or it was aborted (and not yet sent) or some error
377 * occurred during communication with userspace, or the device file
378 * was closed. The requester thread is woken up (if still waiting),
379 * the 'end' callback is called if given, else the reference to the
380 * request is released
382 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
384 struct fuse_iqueue *fiq = &fc->iq;
386 if (test_and_set_bit(FR_FINISHED, &req->flags))
389 spin_lock(&fiq->waitq.lock);
390 list_del_init(&req->intr_entry);
391 spin_unlock(&fiq->waitq.lock);
392 WARN_ON(test_bit(FR_PENDING, &req->flags));
393 WARN_ON(test_bit(FR_SENT, &req->flags));
394 if (test_bit(FR_BACKGROUND, &req->flags)) {
395 spin_lock(&fc->lock);
396 clear_bit(FR_BACKGROUND, &req->flags);
397 if (fc->num_background == fc->max_background)
400 /* Wake up next waiter, if any */
401 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
402 wake_up(&fc->blocked_waitq);
404 if (fc->num_background == fc->congestion_threshold &&
405 fc->connected && fc->bdi_initialized) {
406 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
407 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
409 fc->num_background--;
410 fc->active_background--;
412 spin_unlock(&fc->lock);
414 wake_up(&req->waitq);
417 fuse_put_request(fc, req);
420 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
422 spin_lock(&fiq->waitq.lock);
423 if (list_empty(&req->intr_entry)) {
424 list_add_tail(&req->intr_entry, &fiq->interrupts);
425 wake_up_locked(&fiq->waitq);
427 spin_unlock(&fiq->waitq.lock);
428 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
431 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
433 struct fuse_iqueue *fiq = &fc->iq;
436 if (!fc->no_interrupt) {
437 /* Any signal may interrupt this */
438 err = wait_event_interruptible(req->waitq,
439 test_bit(FR_FINISHED, &req->flags));
443 set_bit(FR_INTERRUPTED, &req->flags);
444 /* matches barrier in fuse_dev_do_read() */
445 smp_mb__after_atomic();
446 if (test_bit(FR_SENT, &req->flags))
447 queue_interrupt(fiq, req);
450 if (!test_bit(FR_FORCE, &req->flags)) {
453 /* Only fatal signals may interrupt this */
455 err = wait_event_interruptible(req->waitq,
456 test_bit(FR_FINISHED, &req->flags));
457 restore_sigs(&oldset);
462 spin_lock(&fiq->waitq.lock);
463 /* Request is not yet in userspace, bail out */
464 if (test_bit(FR_PENDING, &req->flags)) {
465 list_del(&req->list);
466 spin_unlock(&fiq->waitq.lock);
467 __fuse_put_request(req);
468 req->out.h.error = -EINTR;
471 spin_unlock(&fiq->waitq.lock);
475 * Either request is already in userspace, or it was forced.
478 while (!test_bit(FR_FINISHED, &req->flags))
479 wait_event_freezable(req->waitq,
480 test_bit(FR_FINISHED, &req->flags));
483 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
485 struct fuse_iqueue *fiq = &fc->iq;
487 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
488 spin_lock(&fiq->waitq.lock);
489 if (!fiq->connected) {
490 spin_unlock(&fiq->waitq.lock);
491 req->out.h.error = -ENOTCONN;
493 req->in.h.unique = fuse_get_unique(fiq);
494 queue_request(fiq, req);
495 /* acquire extra reference, since request is still needed
496 after request_end() */
497 __fuse_get_request(req);
498 spin_unlock(&fiq->waitq.lock);
500 request_wait_answer(fc, req);
501 /* Pairs with smp_wmb() in request_end() */
506 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
508 __set_bit(FR_ISREPLY, &req->flags);
509 if (!test_bit(FR_WAITING, &req->flags)) {
510 __set_bit(FR_WAITING, &req->flags);
511 atomic_inc(&fc->num_waiting);
513 __fuse_request_send(fc, req);
515 EXPORT_SYMBOL_GPL(fuse_request_send);
517 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
519 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
520 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
523 switch (args->in.h.opcode) {
530 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
534 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
538 if (fc->minor < 12) {
539 switch (args->in.h.opcode) {
541 args->in.args[0].size = sizeof(struct fuse_open_in);
544 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
550 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
552 struct fuse_req *req;
555 req = fuse_get_req(fc, 0);
559 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
560 fuse_adjust_compat(fc, args);
562 req->in.h.opcode = args->in.h.opcode;
563 req->in.h.nodeid = args->in.h.nodeid;
564 req->in.numargs = args->in.numargs;
565 memcpy(req->in.args, args->in.args,
566 args->in.numargs * sizeof(struct fuse_in_arg));
567 req->out.argvar = args->out.argvar;
568 req->out.numargs = args->out.numargs;
569 memcpy(req->out.args, args->out.args,
570 args->out.numargs * sizeof(struct fuse_arg));
571 fuse_request_send(fc, req);
572 ret = req->out.h.error;
573 if (!ret && args->out.argvar) {
574 BUG_ON(args->out.numargs != 1);
575 ret = req->out.args[0].size;
577 fuse_put_request(fc, req);
583 * Called under fc->lock
585 * fc->connected must have been checked previously
587 void fuse_request_send_background_locked(struct fuse_conn *fc,
588 struct fuse_req *req)
590 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
591 if (!test_bit(FR_WAITING, &req->flags)) {
592 __set_bit(FR_WAITING, &req->flags);
593 atomic_inc(&fc->num_waiting);
595 __set_bit(FR_ISREPLY, &req->flags);
596 fc->num_background++;
597 if (fc->num_background == fc->max_background)
599 if (fc->num_background == fc->congestion_threshold &&
600 fc->bdi_initialized) {
601 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
602 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
604 list_add_tail(&req->list, &fc->bg_queue);
608 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
611 spin_lock(&fc->lock);
613 fuse_request_send_background_locked(fc, req);
614 spin_unlock(&fc->lock);
616 spin_unlock(&fc->lock);
617 req->out.h.error = -ENOTCONN;
619 fuse_put_request(fc, req);
622 EXPORT_SYMBOL_GPL(fuse_request_send_background);
624 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
625 struct fuse_req *req, u64 unique)
628 struct fuse_iqueue *fiq = &fc->iq;
630 __clear_bit(FR_ISREPLY, &req->flags);
631 req->in.h.unique = unique;
632 spin_lock(&fiq->waitq.lock);
633 if (fiq->connected) {
634 queue_request(fiq, req);
637 spin_unlock(&fiq->waitq.lock);
642 void fuse_force_forget(struct file *file, u64 nodeid)
644 struct inode *inode = file_inode(file);
645 struct fuse_conn *fc = get_fuse_conn(inode);
646 struct fuse_req *req;
647 struct fuse_forget_in inarg;
649 memset(&inarg, 0, sizeof(inarg));
651 req = fuse_get_req_nofail_nopages(fc, file);
652 req->in.h.opcode = FUSE_FORGET;
653 req->in.h.nodeid = nodeid;
655 req->in.args[0].size = sizeof(inarg);
656 req->in.args[0].value = &inarg;
657 __clear_bit(FR_ISREPLY, &req->flags);
658 __fuse_request_send(fc, req);
660 fuse_put_request(fc, req);
664 * Lock the request. Up to the next unlock_request() there mustn't be
665 * anything that could cause a page-fault. If the request was already
668 static int lock_request(struct fuse_req *req)
672 spin_lock(&req->waitq.lock);
673 if (test_bit(FR_ABORTED, &req->flags))
676 set_bit(FR_LOCKED, &req->flags);
677 spin_unlock(&req->waitq.lock);
683 * Unlock request. If it was aborted while locked, caller is responsible
684 * for unlocking and ending the request.
686 static int unlock_request(struct fuse_req *req)
690 spin_lock(&req->waitq.lock);
691 if (test_bit(FR_ABORTED, &req->flags))
694 clear_bit(FR_LOCKED, &req->flags);
695 spin_unlock(&req->waitq.lock);
700 struct fuse_copy_state {
702 struct fuse_req *req;
703 struct iov_iter *iter;
704 struct pipe_buffer *pipebufs;
705 struct pipe_buffer *currbuf;
706 struct pipe_inode_info *pipe;
707 unsigned long nr_segs;
711 unsigned move_pages:1;
714 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
715 struct iov_iter *iter)
717 memset(cs, 0, sizeof(*cs));
722 /* Unmap and put previous page of userspace buffer */
723 static void fuse_copy_finish(struct fuse_copy_state *cs)
726 struct pipe_buffer *buf = cs->currbuf;
729 buf->len = PAGE_SIZE - cs->len;
733 flush_dcache_page(cs->pg);
734 set_page_dirty_lock(cs->pg);
742 * Get another pagefull of userspace buffer, and map it to kernel
743 * address space, and lock request
745 static int fuse_copy_fill(struct fuse_copy_state *cs)
750 err = unlock_request(cs->req);
754 fuse_copy_finish(cs);
756 struct pipe_buffer *buf = cs->pipebufs;
759 err = buf->ops->confirm(cs->pipe, buf);
763 BUG_ON(!cs->nr_segs);
766 cs->offset = buf->offset;
771 if (cs->nr_segs == cs->pipe->buffers)
774 page = alloc_page(GFP_HIGHUSER);
791 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
799 iov_iter_advance(cs->iter, err);
802 return lock_request(cs->req);
805 /* Do as much copy to/from userspace buffer as we can */
806 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
808 unsigned ncpy = min(*size, cs->len);
810 void *pgaddr = kmap_atomic(cs->pg);
811 void *buf = pgaddr + cs->offset;
814 memcpy(buf, *val, ncpy);
816 memcpy(*val, buf, ncpy);
818 kunmap_atomic(pgaddr);
827 static int fuse_check_page(struct page *page)
829 if (page_mapcount(page) ||
830 page->mapping != NULL ||
831 page_count(page) != 1 ||
832 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
839 printk(KERN_WARNING "fuse: trying to steal weird page\n");
840 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);
846 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
849 struct page *oldpage = *pagep;
850 struct page *newpage;
851 struct pipe_buffer *buf = cs->pipebufs;
853 err = unlock_request(cs->req);
857 fuse_copy_finish(cs);
859 err = buf->ops->confirm(cs->pipe, buf);
863 BUG_ON(!cs->nr_segs);
869 if (cs->len != PAGE_SIZE)
872 if (buf->ops->steal(cs->pipe, buf) != 0)
877 if (!PageUptodate(newpage))
878 SetPageUptodate(newpage);
880 ClearPageMappedToDisk(newpage);
882 if (fuse_check_page(newpage) != 0)
883 goto out_fallback_unlock;
886 * This is a new and locked page, it shouldn't be mapped or
887 * have any special flags on it
889 if (WARN_ON(page_mapped(oldpage)))
890 goto out_fallback_unlock;
891 if (WARN_ON(page_has_private(oldpage)))
892 goto out_fallback_unlock;
893 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
894 goto out_fallback_unlock;
895 if (WARN_ON(PageMlocked(oldpage)))
896 goto out_fallback_unlock;
898 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
900 unlock_page(newpage);
904 page_cache_get(newpage);
906 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
907 lru_cache_add_file(newpage);
910 spin_lock(&cs->req->waitq.lock);
911 if (test_bit(FR_ABORTED, &cs->req->flags))
915 spin_unlock(&cs->req->waitq.lock);
918 unlock_page(newpage);
919 page_cache_release(newpage);
923 unlock_page(oldpage);
924 page_cache_release(oldpage);
930 unlock_page(newpage);
933 cs->offset = buf->offset;
935 err = lock_request(cs->req);
942 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
943 unsigned offset, unsigned count)
945 struct pipe_buffer *buf;
948 if (cs->nr_segs == cs->pipe->buffers)
951 err = unlock_request(cs->req);
955 fuse_copy_finish(cs);
958 page_cache_get(page);
960 buf->offset = offset;
971 * Copy a page in the request to/from the userspace buffer. Must be
974 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
975 unsigned offset, unsigned count, int zeroing)
978 struct page *page = *pagep;
980 if (page && zeroing && count < PAGE_SIZE)
981 clear_highpage(page);
984 if (cs->write && cs->pipebufs && page) {
985 return fuse_ref_page(cs, page, offset, count);
986 } else if (!cs->len) {
987 if (cs->move_pages && page &&
988 offset == 0 && count == PAGE_SIZE) {
989 err = fuse_try_move_page(cs, pagep);
993 err = fuse_copy_fill(cs);
999 void *mapaddr = kmap_atomic(page);
1000 void *buf = mapaddr + offset;
1001 offset += fuse_copy_do(cs, &buf, &count);
1002 kunmap_atomic(mapaddr);
1004 offset += fuse_copy_do(cs, NULL, &count);
1006 if (page && !cs->write)
1007 flush_dcache_page(page);
1011 /* Copy pages in the request to/from userspace buffer */
1012 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1016 struct fuse_req *req = cs->req;
1018 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1020 unsigned offset = req->page_descs[i].offset;
1021 unsigned count = min(nbytes, req->page_descs[i].length);
1023 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1033 /* Copy a single argument in the request to/from userspace buffer */
1034 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1038 int err = fuse_copy_fill(cs);
1042 fuse_copy_do(cs, &val, &size);
1047 /* Copy request arguments to/from userspace buffer */
1048 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1049 unsigned argpages, struct fuse_arg *args,
1055 for (i = 0; !err && i < numargs; i++) {
1056 struct fuse_arg *arg = &args[i];
1057 if (i == numargs - 1 && argpages)
1058 err = fuse_copy_pages(cs, arg->size, zeroing);
1060 err = fuse_copy_one(cs, arg->value, arg->size);
1065 static int forget_pending(struct fuse_iqueue *fiq)
1067 return fiq->forget_list_head.next != NULL;
1070 static int request_pending(struct fuse_iqueue *fiq)
1072 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1073 forget_pending(fiq);
1077 * Transfer an interrupt request to userspace
1079 * Unlike other requests this is assembled on demand, without a need
1080 * to allocate a separate fuse_req structure.
1082 * Called with fiq->waitq.lock held, releases it
1084 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1085 struct fuse_copy_state *cs,
1086 size_t nbytes, struct fuse_req *req)
1087 __releases(fiq->waitq.lock)
1089 struct fuse_in_header ih;
1090 struct fuse_interrupt_in arg;
1091 unsigned reqsize = sizeof(ih) + sizeof(arg);
1094 list_del_init(&req->intr_entry);
1095 req->intr_unique = fuse_get_unique(fiq);
1096 memset(&ih, 0, sizeof(ih));
1097 memset(&arg, 0, sizeof(arg));
1099 ih.opcode = FUSE_INTERRUPT;
1100 ih.unique = req->intr_unique;
1101 arg.unique = req->in.h.unique;
1103 spin_unlock(&fiq->waitq.lock);
1104 if (nbytes < reqsize)
1107 err = fuse_copy_one(cs, &ih, sizeof(ih));
1109 err = fuse_copy_one(cs, &arg, sizeof(arg));
1110 fuse_copy_finish(cs);
1112 return err ? err : reqsize;
1115 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1119 struct fuse_forget_link *head = fiq->forget_list_head.next;
1120 struct fuse_forget_link **newhead = &head;
1123 for (count = 0; *newhead != NULL && count < max; count++)
1124 newhead = &(*newhead)->next;
1126 fiq->forget_list_head.next = *newhead;
1128 if (fiq->forget_list_head.next == NULL)
1129 fiq->forget_list_tail = &fiq->forget_list_head;
1137 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1138 struct fuse_copy_state *cs,
1140 __releases(fiq->waitq.lock)
1143 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1144 struct fuse_forget_in arg = {
1145 .nlookup = forget->forget_one.nlookup,
1147 struct fuse_in_header ih = {
1148 .opcode = FUSE_FORGET,
1149 .nodeid = forget->forget_one.nodeid,
1150 .unique = fuse_get_unique(fiq),
1151 .len = sizeof(ih) + sizeof(arg),
1154 spin_unlock(&fiq->waitq.lock);
1156 if (nbytes < ih.len)
1159 err = fuse_copy_one(cs, &ih, sizeof(ih));
1161 err = fuse_copy_one(cs, &arg, sizeof(arg));
1162 fuse_copy_finish(cs);
1170 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1171 struct fuse_copy_state *cs, size_t nbytes)
1172 __releases(fiq->waitq.lock)
1175 unsigned max_forgets;
1177 struct fuse_forget_link *head;
1178 struct fuse_batch_forget_in arg = { .count = 0 };
1179 struct fuse_in_header ih = {
1180 .opcode = FUSE_BATCH_FORGET,
1181 .unique = fuse_get_unique(fiq),
1182 .len = sizeof(ih) + sizeof(arg),
1185 if (nbytes < ih.len) {
1186 spin_unlock(&fiq->waitq.lock);
1190 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1191 head = dequeue_forget(fiq, max_forgets, &count);
1192 spin_unlock(&fiq->waitq.lock);
1195 ih.len += count * sizeof(struct fuse_forget_one);
1196 err = fuse_copy_one(cs, &ih, sizeof(ih));
1198 err = fuse_copy_one(cs, &arg, sizeof(arg));
1201 struct fuse_forget_link *forget = head;
1204 err = fuse_copy_one(cs, &forget->forget_one,
1205 sizeof(forget->forget_one));
1207 head = forget->next;
1211 fuse_copy_finish(cs);
1219 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1220 struct fuse_copy_state *cs,
1222 __releases(fiq->waitq.lock)
1224 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1225 return fuse_read_single_forget(fiq, cs, nbytes);
1227 return fuse_read_batch_forget(fiq, cs, nbytes);
1231 * Read a single request into the userspace filesystem's buffer. This
1232 * function waits until a request is available, then removes it from
1233 * the pending list and copies request data to userspace buffer. If
1234 * no reply is needed (FORGET) or request has been aborted or there
1235 * was an error during the copying then it's finished by calling
1236 * request_end(). Otherwise add it to the processing list, and set
1239 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1240 struct fuse_copy_state *cs, size_t nbytes)
1243 struct fuse_conn *fc = fud->fc;
1244 struct fuse_iqueue *fiq = &fc->iq;
1245 struct fuse_pqueue *fpq = &fud->pq;
1246 struct fuse_req *req;
1251 spin_lock(&fiq->waitq.lock);
1253 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1254 !request_pending(fiq))
1257 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1258 !fiq->connected || request_pending(fiq));
1263 if (!fiq->connected)
1266 if (!list_empty(&fiq->interrupts)) {
1267 req = list_entry(fiq->interrupts.next, struct fuse_req,
1269 return fuse_read_interrupt(fiq, cs, nbytes, req);
1272 if (forget_pending(fiq)) {
1273 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1274 return fuse_read_forget(fc, fiq, cs, nbytes);
1276 if (fiq->forget_batch <= -8)
1277 fiq->forget_batch = 16;
1280 req = list_entry(fiq->pending.next, struct fuse_req, list);
1281 clear_bit(FR_PENDING, &req->flags);
1282 list_del_init(&req->list);
1283 spin_unlock(&fiq->waitq.lock);
1286 reqsize = in->h.len;
1287 /* If request is too large, reply with an error and restart the read */
1288 if (nbytes < reqsize) {
1289 req->out.h.error = -EIO;
1290 /* SETXATTR is special, since it may contain too large data */
1291 if (in->h.opcode == FUSE_SETXATTR)
1292 req->out.h.error = -E2BIG;
1293 request_end(fc, req);
1296 spin_lock(&fpq->lock);
1297 list_add(&req->list, &fpq->io);
1298 spin_unlock(&fpq->lock);
1300 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1302 err = fuse_copy_args(cs, in->numargs, in->argpages,
1303 (struct fuse_arg *) in->args, 0);
1304 fuse_copy_finish(cs);
1305 spin_lock(&fpq->lock);
1306 clear_bit(FR_LOCKED, &req->flags);
1307 if (!fpq->connected) {
1312 req->out.h.error = -EIO;
1315 if (!test_bit(FR_ISREPLY, &req->flags)) {
1319 list_move_tail(&req->list, &fpq->processing);
1320 spin_unlock(&fpq->lock);
1321 set_bit(FR_SENT, &req->flags);
1322 /* matches barrier in request_wait_answer() */
1323 smp_mb__after_atomic();
1324 if (test_bit(FR_INTERRUPTED, &req->flags))
1325 queue_interrupt(fiq, req);
1330 if (!test_bit(FR_PRIVATE, &req->flags))
1331 list_del_init(&req->list);
1332 spin_unlock(&fpq->lock);
1333 request_end(fc, req);
1337 spin_unlock(&fiq->waitq.lock);
1341 static int fuse_dev_open(struct inode *inode, struct file *file)
1344 * The fuse device's file's private_data is used to hold
1345 * the fuse_conn(ection) when it is mounted, and is used to
1346 * keep track of whether the file has been mounted already.
1348 file->private_data = NULL;
1352 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1354 struct fuse_copy_state cs;
1355 struct file *file = iocb->ki_filp;
1356 struct fuse_dev *fud = fuse_get_dev(file);
1361 if (!iter_is_iovec(to))
1364 fuse_copy_init(&cs, 1, to);
1366 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1369 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1370 struct pipe_inode_info *pipe,
1371 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 = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1387 fuse_copy_init(&cs, 1, NULL);
1390 ret = fuse_dev_do_read(fud, in, &cs, len);
1397 if (!pipe->readers) {
1398 send_sig(SIGPIPE, current, 0);
1404 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1409 while (page_nr < cs.nr_segs) {
1410 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1411 struct pipe_buffer *buf = pipe->bufs + newbuf;
1413 buf->page = bufs[page_nr].page;
1414 buf->offset = bufs[page_nr].offset;
1415 buf->len = bufs[page_nr].len;
1417 * Need to be careful about this. Having buf->ops in module
1418 * code can Oops if the buffer persists after module unload.
1420 buf->ops = &nosteal_pipe_buf_ops;
1435 if (waitqueue_active(&pipe->wait))
1436 wake_up_interruptible(&pipe->wait);
1437 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1441 for (; page_nr < cs.nr_segs; page_nr++)
1442 page_cache_release(bufs[page_nr].page);
1448 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1449 struct fuse_copy_state *cs)
1451 struct fuse_notify_poll_wakeup_out outarg;
1454 if (size != sizeof(outarg))
1457 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1461 fuse_copy_finish(cs);
1462 return fuse_notify_poll_wakeup(fc, &outarg);
1465 fuse_copy_finish(cs);
1469 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1470 struct fuse_copy_state *cs)
1472 struct fuse_notify_inval_inode_out outarg;
1475 if (size != sizeof(outarg))
1478 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1481 fuse_copy_finish(cs);
1483 down_read(&fc->killsb);
1486 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1487 outarg.off, outarg.len);
1489 up_read(&fc->killsb);
1493 fuse_copy_finish(cs);
1497 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1498 struct fuse_copy_state *cs)
1500 struct fuse_notify_inval_entry_out outarg;
1505 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1510 if (size < sizeof(outarg))
1513 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1517 err = -ENAMETOOLONG;
1518 if (outarg.namelen > FUSE_NAME_MAX)
1522 if (size != sizeof(outarg) + outarg.namelen + 1)
1526 name.len = outarg.namelen;
1527 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1530 fuse_copy_finish(cs);
1531 buf[outarg.namelen] = 0;
1532 name.hash = full_name_hash(name.name, name.len);
1534 down_read(&fc->killsb);
1537 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1538 up_read(&fc->killsb);
1544 fuse_copy_finish(cs);
1548 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1549 struct fuse_copy_state *cs)
1551 struct fuse_notify_delete_out outarg;
1556 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1561 if (size < sizeof(outarg))
1564 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1568 err = -ENAMETOOLONG;
1569 if (outarg.namelen > FUSE_NAME_MAX)
1573 if (size != sizeof(outarg) + outarg.namelen + 1)
1577 name.len = outarg.namelen;
1578 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1581 fuse_copy_finish(cs);
1582 buf[outarg.namelen] = 0;
1583 name.hash = full_name_hash(name.name, name.len);
1585 down_read(&fc->killsb);
1588 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1589 outarg.child, &name);
1590 up_read(&fc->killsb);
1596 fuse_copy_finish(cs);
1600 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1601 struct fuse_copy_state *cs)
1603 struct fuse_notify_store_out outarg;
1604 struct inode *inode;
1605 struct address_space *mapping;
1609 unsigned int offset;
1615 if (size < sizeof(outarg))
1618 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1623 if (size - sizeof(outarg) != outarg.size)
1626 nodeid = outarg.nodeid;
1628 down_read(&fc->killsb);
1634 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1638 mapping = inode->i_mapping;
1639 index = outarg.offset >> PAGE_CACHE_SHIFT;
1640 offset = outarg.offset & ~PAGE_CACHE_MASK;
1641 file_size = i_size_read(inode);
1642 end = outarg.offset + outarg.size;
1643 if (end > file_size) {
1645 fuse_write_update_size(inode, file_size);
1651 unsigned int this_num;
1654 page = find_or_create_page(mapping, index,
1655 mapping_gfp_mask(mapping));
1659 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1660 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1661 if (!err && offset == 0 &&
1662 (this_num == PAGE_CACHE_SIZE || file_size == end))
1663 SetPageUptodate(page);
1665 page_cache_release(page);
1680 up_read(&fc->killsb);
1682 fuse_copy_finish(cs);
1686 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1688 release_pages(req->pages, req->num_pages, false);
1691 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1692 struct fuse_notify_retrieve_out *outarg)
1695 struct address_space *mapping = inode->i_mapping;
1696 struct fuse_req *req;
1700 unsigned int offset;
1701 size_t total_len = 0;
1704 offset = outarg->offset & ~PAGE_CACHE_MASK;
1705 file_size = i_size_read(inode);
1708 if (outarg->offset > file_size)
1710 else if (outarg->offset + num > file_size)
1711 num = file_size - outarg->offset;
1713 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1714 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1716 req = fuse_get_req(fc, num_pages);
1718 return PTR_ERR(req);
1720 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1721 req->in.h.nodeid = outarg->nodeid;
1722 req->in.numargs = 2;
1723 req->in.argpages = 1;
1724 req->page_descs[0].offset = offset;
1725 req->end = fuse_retrieve_end;
1727 index = outarg->offset >> PAGE_CACHE_SHIFT;
1729 while (num && req->num_pages < num_pages) {
1731 unsigned int this_num;
1733 page = find_get_page(mapping, index);
1737 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1738 req->pages[req->num_pages] = page;
1739 req->page_descs[req->num_pages].length = this_num;
1744 total_len += this_num;
1747 req->misc.retrieve_in.offset = outarg->offset;
1748 req->misc.retrieve_in.size = total_len;
1749 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1750 req->in.args[0].value = &req->misc.retrieve_in;
1751 req->in.args[1].size = total_len;
1753 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1755 fuse_retrieve_end(fc, req);
1760 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1761 struct fuse_copy_state *cs)
1763 struct fuse_notify_retrieve_out outarg;
1764 struct inode *inode;
1768 if (size != sizeof(outarg))
1771 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1775 fuse_copy_finish(cs);
1777 down_read(&fc->killsb);
1780 u64 nodeid = outarg.nodeid;
1782 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1784 err = fuse_retrieve(fc, inode, &outarg);
1788 up_read(&fc->killsb);
1793 fuse_copy_finish(cs);
1797 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1798 unsigned int size, struct fuse_copy_state *cs)
1800 /* Don't try to move pages (yet) */
1804 case FUSE_NOTIFY_POLL:
1805 return fuse_notify_poll(fc, size, cs);
1807 case FUSE_NOTIFY_INVAL_INODE:
1808 return fuse_notify_inval_inode(fc, size, cs);
1810 case FUSE_NOTIFY_INVAL_ENTRY:
1811 return fuse_notify_inval_entry(fc, size, cs);
1813 case FUSE_NOTIFY_STORE:
1814 return fuse_notify_store(fc, size, cs);
1816 case FUSE_NOTIFY_RETRIEVE:
1817 return fuse_notify_retrieve(fc, size, cs);
1819 case FUSE_NOTIFY_DELETE:
1820 return fuse_notify_delete(fc, size, cs);
1823 fuse_copy_finish(cs);
1828 /* Look up request on processing list by unique ID */
1829 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1831 struct fuse_req *req;
1833 list_for_each_entry(req, &fpq->processing, list) {
1834 if (req->in.h.unique == unique || req->intr_unique == unique)
1840 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1843 unsigned reqsize = sizeof(struct fuse_out_header);
1846 return nbytes != reqsize ? -EINVAL : 0;
1848 reqsize += len_args(out->numargs, out->args);
1850 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1852 else if (reqsize > nbytes) {
1853 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1854 unsigned diffsize = reqsize - nbytes;
1855 if (diffsize > lastarg->size)
1857 lastarg->size -= diffsize;
1859 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1864 * Write a single reply to a request. First the header is copied from
1865 * the write buffer. The request is then searched on the processing
1866 * list by the unique ID found in the header. If found, then remove
1867 * it from the list and copy the rest of the buffer to the request.
1868 * The request is finished by calling request_end()
1870 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1871 struct fuse_copy_state *cs, size_t nbytes)
1874 struct fuse_conn *fc = fud->fc;
1875 struct fuse_pqueue *fpq = &fud->pq;
1876 struct fuse_req *req;
1877 struct fuse_out_header oh;
1879 if (nbytes < sizeof(struct fuse_out_header))
1882 err = fuse_copy_one(cs, &oh, sizeof(oh));
1887 if (oh.len != nbytes)
1891 * Zero oh.unique indicates unsolicited notification message
1892 * and error contains notification code.
1895 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1896 return err ? err : nbytes;
1900 if (oh.error <= -1000 || oh.error > 0)
1903 spin_lock(&fpq->lock);
1905 if (!fpq->connected)
1908 req = request_find(fpq, oh.unique);
1912 /* Is it an interrupt reply? */
1913 if (req->intr_unique == oh.unique) {
1914 spin_unlock(&fpq->lock);
1917 if (nbytes != sizeof(struct fuse_out_header))
1920 if (oh.error == -ENOSYS)
1921 fc->no_interrupt = 1;
1922 else if (oh.error == -EAGAIN)
1923 queue_interrupt(&fc->iq, req);
1925 fuse_copy_finish(cs);
1929 clear_bit(FR_SENT, &req->flags);
1930 list_move(&req->list, &fpq->io);
1932 set_bit(FR_LOCKED, &req->flags);
1933 spin_unlock(&fpq->lock);
1935 if (!req->out.page_replace)
1938 err = copy_out_args(cs, &req->out, nbytes);
1939 if (req->in.h.opcode == FUSE_CANONICAL_PATH) {
1940 req->out.h.error = kern_path((char *)req->out.args[0].value, 0,
1941 req->canonical_path);
1943 fuse_copy_finish(cs);
1945 spin_lock(&fpq->lock);
1946 clear_bit(FR_LOCKED, &req->flags);
1947 if (!fpq->connected)
1950 req->out.h.error = -EIO;
1951 if (!test_bit(FR_PRIVATE, &req->flags))
1952 list_del_init(&req->list);
1953 spin_unlock(&fpq->lock);
1955 request_end(fc, req);
1957 return err ? err : nbytes;
1960 spin_unlock(&fpq->lock);
1962 fuse_copy_finish(cs);
1966 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1968 struct fuse_copy_state cs;
1969 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1974 if (!iter_is_iovec(from))
1977 fuse_copy_init(&cs, 0, from);
1979 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1982 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1983 struct file *out, loff_t *ppos,
1984 size_t len, unsigned int flags)
1988 struct pipe_buffer *bufs;
1989 struct fuse_copy_state cs;
1990 struct fuse_dev *fud;
1994 fud = fuse_get_dev(out);
1998 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
2005 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
2006 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
2016 struct pipe_buffer *ibuf;
2017 struct pipe_buffer *obuf;
2019 BUG_ON(nbuf >= pipe->buffers);
2020 BUG_ON(!pipe->nrbufs);
2021 ibuf = &pipe->bufs[pipe->curbuf];
2024 if (rem >= ibuf->len) {
2027 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2030 ibuf->ops->get(pipe, ibuf);
2032 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2034 ibuf->offset += obuf->len;
2035 ibuf->len -= obuf->len;
2042 fuse_copy_init(&cs, 0, NULL);
2047 if (flags & SPLICE_F_MOVE)
2050 ret = fuse_dev_do_write(fud, &cs, len);
2052 for (idx = 0; idx < nbuf; idx++) {
2053 struct pipe_buffer *buf = &bufs[idx];
2054 buf->ops->release(pipe, buf);
2061 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2063 unsigned mask = POLLOUT | POLLWRNORM;
2064 struct fuse_iqueue *fiq;
2065 struct fuse_dev *fud = fuse_get_dev(file);
2071 poll_wait(file, &fiq->waitq, wait);
2073 spin_lock(&fiq->waitq.lock);
2074 if (!fiq->connected)
2076 else if (request_pending(fiq))
2077 mask |= POLLIN | POLLRDNORM;
2078 spin_unlock(&fiq->waitq.lock);
2084 * Abort all requests on the given list (pending or processing)
2086 * This function releases and reacquires fc->lock
2088 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2090 while (!list_empty(head)) {
2091 struct fuse_req *req;
2092 req = list_entry(head->next, struct fuse_req, list);
2093 req->out.h.error = -ECONNABORTED;
2094 clear_bit(FR_PENDING, &req->flags);
2095 clear_bit(FR_SENT, &req->flags);
2096 list_del_init(&req->list);
2097 request_end(fc, req);
2101 static void end_polls(struct fuse_conn *fc)
2105 p = rb_first(&fc->polled_files);
2108 struct fuse_file *ff;
2109 ff = rb_entry(p, struct fuse_file, polled_node);
2110 wake_up_interruptible_all(&ff->poll_wait);
2117 * Abort all requests.
2119 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2122 * The same effect is usually achievable through killing the filesystem daemon
2123 * and all users of the filesystem. The exception is the combination of an
2124 * asynchronous request and the tricky deadlock (see
2125 * Documentation/filesystems/fuse.txt).
2127 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2128 * requests, they should be finished off immediately. Locked requests will be
2129 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2130 * requests. It is possible that some request will finish before we can. This
2131 * is OK, the request will in that case be removed from the list before we touch
2134 void fuse_abort_conn(struct fuse_conn *fc)
2136 struct fuse_iqueue *fiq = &fc->iq;
2138 spin_lock(&fc->lock);
2139 if (fc->connected) {
2140 struct fuse_dev *fud;
2141 struct fuse_req *req, *next;
2147 fuse_set_initialized(fc);
2148 list_for_each_entry(fud, &fc->devices, entry) {
2149 struct fuse_pqueue *fpq = &fud->pq;
2151 spin_lock(&fpq->lock);
2153 list_for_each_entry_safe(req, next, &fpq->io, list) {
2154 req->out.h.error = -ECONNABORTED;
2155 spin_lock(&req->waitq.lock);
2156 set_bit(FR_ABORTED, &req->flags);
2157 if (!test_bit(FR_LOCKED, &req->flags)) {
2158 set_bit(FR_PRIVATE, &req->flags);
2159 list_move(&req->list, &to_end1);
2161 spin_unlock(&req->waitq.lock);
2163 list_splice_init(&fpq->processing, &to_end2);
2164 spin_unlock(&fpq->lock);
2166 fc->max_background = UINT_MAX;
2169 spin_lock(&fiq->waitq.lock);
2171 list_splice_init(&fiq->pending, &to_end2);
2172 while (forget_pending(fiq))
2173 kfree(dequeue_forget(fiq, 1, NULL));
2174 wake_up_all_locked(&fiq->waitq);
2175 spin_unlock(&fiq->waitq.lock);
2176 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2178 wake_up_all(&fc->blocked_waitq);
2179 spin_unlock(&fc->lock);
2181 while (!list_empty(&to_end1)) {
2182 req = list_first_entry(&to_end1, struct fuse_req, list);
2183 __fuse_get_request(req);
2184 list_del_init(&req->list);
2185 request_end(fc, req);
2187 end_requests(fc, &to_end2);
2189 spin_unlock(&fc->lock);
2192 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2194 int fuse_dev_release(struct inode *inode, struct file *file)
2196 struct fuse_dev *fud = fuse_get_dev(file);
2199 struct fuse_conn *fc = fud->fc;
2200 struct fuse_pqueue *fpq = &fud->pq;
2202 WARN_ON(!list_empty(&fpq->io));
2203 end_requests(fc, &fpq->processing);
2204 /* Are we the last open device? */
2205 if (atomic_dec_and_test(&fc->dev_count)) {
2206 WARN_ON(fc->iq.fasync != NULL);
2207 fuse_abort_conn(fc);
2213 EXPORT_SYMBOL_GPL(fuse_dev_release);
2215 static int fuse_dev_fasync(int fd, struct file *file, int on)
2217 struct fuse_dev *fud = fuse_get_dev(file);
2222 /* No locking - fasync_helper does its own locking */
2223 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2226 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2228 struct fuse_dev *fud;
2230 if (new->private_data)
2233 fud = fuse_dev_alloc(fc);
2237 new->private_data = fud;
2238 atomic_inc(&fc->dev_count);
2243 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2248 if (cmd == FUSE_DEV_IOC_CLONE) {
2252 if (!get_user(oldfd, (__u32 __user *) arg)) {
2253 struct file *old = fget(oldfd);
2257 struct fuse_dev *fud = NULL;
2260 * Check against file->f_op because CUSE
2261 * uses the same ioctl handler.
2263 if (old->f_op == file->f_op &&
2264 old->f_cred->user_ns == file->f_cred->user_ns)
2265 fud = fuse_get_dev(old);
2268 mutex_lock(&fuse_mutex);
2269 err = fuse_device_clone(fud->fc, file);
2270 mutex_unlock(&fuse_mutex);
2279 const struct file_operations fuse_dev_operations = {
2280 .owner = THIS_MODULE,
2281 .open = fuse_dev_open,
2282 .llseek = no_llseek,
2283 .read_iter = fuse_dev_read,
2284 .splice_read = fuse_dev_splice_read,
2285 .write_iter = fuse_dev_write,
2286 .splice_write = fuse_dev_splice_write,
2287 .poll = fuse_dev_poll,
2288 .release = fuse_dev_release,
2289 .fasync = fuse_dev_fasync,
2290 .unlocked_ioctl = fuse_dev_ioctl,
2291 .compat_ioctl = fuse_dev_ioctl,
2293 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2295 static struct miscdevice fuse_miscdevice = {
2296 .minor = FUSE_MINOR,
2298 .fops = &fuse_dev_operations,
2301 int __init fuse_dev_init(void)
2304 fuse_req_cachep = kmem_cache_create("fuse_request",
2305 sizeof(struct fuse_req),
2307 if (!fuse_req_cachep)
2310 err = misc_register(&fuse_miscdevice);
2312 goto out_cache_clean;
2317 kmem_cache_destroy(fuse_req_cachep);
2322 void fuse_dev_cleanup(void)
2324 misc_deregister(&fuse_miscdevice);
2325 kmem_cache_destroy(fuse_req_cachep);