2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/uio.h>
15 #include <linux/miscdevice.h>
16 #include <linux/pagemap.h>
17 #include <linux/file.h>
18 #include <linux/slab.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/swap.h>
21 #include <linux/splice.h>
23 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24 MODULE_ALIAS("devname:fuse");
26 static struct kmem_cache *fuse_req_cachep;
28 static struct fuse_dev *fuse_get_dev(struct file *file)
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
34 return ACCESS_ONCE(file->private_data);
37 static void fuse_request_init(struct fuse_req *req, struct page **pages,
38 struct fuse_page_desc *page_descs,
41 memset(req, 0, sizeof(*req));
42 memset(pages, 0, sizeof(*pages) * npages);
43 memset(page_descs, 0, sizeof(*page_descs) * npages);
44 INIT_LIST_HEAD(&req->list);
45 INIT_LIST_HEAD(&req->intr_entry);
46 init_waitqueue_head(&req->waitq);
47 atomic_set(&req->count, 1);
49 req->page_descs = page_descs;
50 req->max_pages = npages;
51 __set_bit(FR_PENDING, &req->flags);
54 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
56 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
59 struct fuse_page_desc *page_descs;
61 if (npages <= FUSE_REQ_INLINE_PAGES) {
62 pages = req->inline_pages;
63 page_descs = req->inline_page_descs;
65 pages = kmalloc(sizeof(struct page *) * npages, flags);
66 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
70 if (!pages || !page_descs) {
73 kmem_cache_free(fuse_req_cachep, req);
77 fuse_request_init(req, pages, page_descs, npages);
82 struct fuse_req *fuse_request_alloc(unsigned npages)
84 return __fuse_request_alloc(npages, GFP_KERNEL);
86 EXPORT_SYMBOL_GPL(fuse_request_alloc);
88 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
90 return __fuse_request_alloc(npages, GFP_NOFS);
93 void fuse_request_free(struct fuse_req *req)
95 if (req->pages != req->inline_pages) {
97 kfree(req->page_descs);
99 kmem_cache_free(fuse_req_cachep, req);
102 void __fuse_get_request(struct fuse_req *req)
104 atomic_inc(&req->count);
107 /* Must be called with > 1 refcount */
108 static void __fuse_put_request(struct fuse_req *req)
110 BUG_ON(atomic_read(&req->count) < 2);
111 atomic_dec(&req->count);
114 static void fuse_req_init_context(struct fuse_req *req)
116 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
117 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
118 req->in.h.pid = current->pid;
121 void fuse_set_initialized(struct fuse_conn *fc)
123 /* Make sure stores before this are seen on another CPU */
128 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
130 return !fc->initialized || (for_background && fc->blocked);
133 static void fuse_drop_waiting(struct fuse_conn *fc)
136 * lockess check of fc->connected is okay, because atomic_dec_and_test()
137 * provides a memory barrier mached with the one in fuse_wait_aborted()
138 * to ensure no wake-up is missed.
140 if (atomic_dec_and_test(&fc->num_waiting) &&
141 !READ_ONCE(fc->connected)) {
142 /* wake up aborters */
143 wake_up_all(&fc->blocked_waitq);
147 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
150 struct fuse_req *req;
152 atomic_inc(&fc->num_waiting);
154 if (fuse_block_alloc(fc, for_background)) {
156 if (wait_event_killable_exclusive(fc->blocked_waitq,
157 !fuse_block_alloc(fc, for_background)))
160 /* Matches smp_wmb() in fuse_set_initialized() */
171 req = fuse_request_alloc(npages);
175 wake_up(&fc->blocked_waitq);
179 fuse_req_init_context(req);
180 __set_bit(FR_WAITING, &req->flags);
182 __set_bit(FR_BACKGROUND, &req->flags);
187 fuse_drop_waiting(fc);
191 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
193 return __fuse_get_req(fc, npages, false);
195 EXPORT_SYMBOL_GPL(fuse_get_req);
197 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
200 return __fuse_get_req(fc, npages, true);
202 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
205 * Return request in fuse_file->reserved_req. However that may
206 * currently be in use. If that is the case, wait for it to become
209 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
212 struct fuse_req *req = NULL;
213 struct fuse_file *ff = file->private_data;
216 wait_event(fc->reserved_req_waitq, ff->reserved_req);
217 spin_lock(&fc->lock);
218 if (ff->reserved_req) {
219 req = ff->reserved_req;
220 ff->reserved_req = NULL;
221 req->stolen_file = get_file(file);
223 spin_unlock(&fc->lock);
230 * Put stolen request back into fuse_file->reserved_req
232 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
234 struct file *file = req->stolen_file;
235 struct fuse_file *ff = file->private_data;
237 spin_lock(&fc->lock);
238 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
239 BUG_ON(ff->reserved_req);
240 ff->reserved_req = req;
241 wake_up_all(&fc->reserved_req_waitq);
242 spin_unlock(&fc->lock);
247 * Gets a requests for a file operation, always succeeds
249 * This is used for sending the FLUSH request, which must get to
250 * userspace, due to POSIX locks which may need to be unlocked.
252 * If allocation fails due to OOM, use the reserved request in
255 * This is very unlikely to deadlock accidentally, since the
256 * filesystem should not have it's own file open. If deadlock is
257 * intentional, it can still be broken by "aborting" the filesystem.
259 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
262 struct fuse_req *req;
264 atomic_inc(&fc->num_waiting);
265 wait_event(fc->blocked_waitq, fc->initialized);
266 /* Matches smp_wmb() in fuse_set_initialized() */
268 req = fuse_request_alloc(0);
270 req = get_reserved_req(fc, file);
272 fuse_req_init_context(req);
273 __set_bit(FR_WAITING, &req->flags);
274 __clear_bit(FR_BACKGROUND, &req->flags);
278 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
280 if (atomic_dec_and_test(&req->count)) {
281 if (test_bit(FR_BACKGROUND, &req->flags)) {
283 * We get here in the unlikely case that a background
284 * request was allocated but not sent
286 spin_lock(&fc->lock);
288 wake_up(&fc->blocked_waitq);
289 spin_unlock(&fc->lock);
292 if (test_bit(FR_WAITING, &req->flags)) {
293 __clear_bit(FR_WAITING, &req->flags);
294 fuse_drop_waiting(fc);
297 if (req->stolen_file)
298 put_reserved_req(fc, req);
300 fuse_request_free(req);
303 EXPORT_SYMBOL_GPL(fuse_put_request);
305 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
310 for (i = 0; i < numargs; i++)
311 nbytes += args[i].size;
316 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
318 return ++fiq->reqctr;
321 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
323 req->in.h.len = sizeof(struct fuse_in_header) +
324 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
325 list_add_tail(&req->list, &fiq->pending);
326 wake_up_locked(&fiq->waitq);
327 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
330 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
331 u64 nodeid, u64 nlookup)
333 struct fuse_iqueue *fiq = &fc->iq;
335 forget->forget_one.nodeid = nodeid;
336 forget->forget_one.nlookup = nlookup;
338 spin_lock(&fiq->waitq.lock);
339 if (fiq->connected) {
340 fiq->forget_list_tail->next = forget;
341 fiq->forget_list_tail = forget;
342 wake_up_locked(&fiq->waitq);
343 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
347 spin_unlock(&fiq->waitq.lock);
350 static void flush_bg_queue(struct fuse_conn *fc)
352 while (fc->active_background < fc->max_background &&
353 !list_empty(&fc->bg_queue)) {
354 struct fuse_req *req;
355 struct fuse_iqueue *fiq = &fc->iq;
357 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
358 list_del(&req->list);
359 fc->active_background++;
360 spin_lock(&fiq->waitq.lock);
361 req->in.h.unique = fuse_get_unique(fiq);
362 queue_request(fiq, req);
363 spin_unlock(&fiq->waitq.lock);
368 * This function is called when a request is finished. Either a reply
369 * has arrived or it was aborted (and not yet sent) or some error
370 * occurred during communication with userspace, or the device file
371 * was closed. The requester thread is woken up (if still waiting),
372 * the 'end' callback is called if given, else the reference to the
373 * request is released
375 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
377 struct fuse_iqueue *fiq = &fc->iq;
379 if (test_and_set_bit(FR_FINISHED, &req->flags))
382 spin_lock(&fiq->waitq.lock);
383 list_del_init(&req->intr_entry);
384 spin_unlock(&fiq->waitq.lock);
385 WARN_ON(test_bit(FR_PENDING, &req->flags));
386 WARN_ON(test_bit(FR_SENT, &req->flags));
387 if (test_bit(FR_BACKGROUND, &req->flags)) {
388 spin_lock(&fc->lock);
389 clear_bit(FR_BACKGROUND, &req->flags);
390 if (fc->num_background == fc->max_background) {
392 wake_up(&fc->blocked_waitq);
393 } else if (!fc->blocked) {
395 * Wake up next waiter, if any. It's okay to use
396 * waitqueue_active(), as we've already synced up
397 * fc->blocked with waiters with the wake_up() call
400 if (waitqueue_active(&fc->blocked_waitq))
401 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);
418 fuse_put_request(fc, req);
421 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
423 spin_lock(&fiq->waitq.lock);
424 if (test_bit(FR_FINISHED, &req->flags)) {
425 spin_unlock(&fiq->waitq.lock);
428 if (list_empty(&req->intr_entry)) {
429 list_add_tail(&req->intr_entry, &fiq->interrupts);
430 wake_up_locked(&fiq->waitq);
432 spin_unlock(&fiq->waitq.lock);
433 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
436 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
438 struct fuse_iqueue *fiq = &fc->iq;
441 if (!fc->no_interrupt) {
442 /* Any signal may interrupt this */
443 err = wait_event_interruptible(req->waitq,
444 test_bit(FR_FINISHED, &req->flags));
448 set_bit(FR_INTERRUPTED, &req->flags);
449 /* matches barrier in fuse_dev_do_read() */
450 smp_mb__after_atomic();
451 if (test_bit(FR_SENT, &req->flags))
452 queue_interrupt(fiq, req);
455 if (!test_bit(FR_FORCE, &req->flags)) {
456 /* Only fatal signals may interrupt this */
457 err = wait_event_killable(req->waitq,
458 test_bit(FR_FINISHED, &req->flags));
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 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
481 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
483 struct fuse_iqueue *fiq = &fc->iq;
485 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
486 spin_lock(&fiq->waitq.lock);
487 if (!fiq->connected) {
488 spin_unlock(&fiq->waitq.lock);
489 req->out.h.error = -ENOTCONN;
491 req->in.h.unique = fuse_get_unique(fiq);
492 queue_request(fiq, req);
493 /* acquire extra reference, since request is still needed
494 after request_end() */
495 __fuse_get_request(req);
496 spin_unlock(&fiq->waitq.lock);
498 request_wait_answer(fc, req);
499 /* Pairs with smp_wmb() in request_end() */
504 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
506 __set_bit(FR_ISREPLY, &req->flags);
507 if (!test_bit(FR_WAITING, &req->flags)) {
508 __set_bit(FR_WAITING, &req->flags);
509 atomic_inc(&fc->num_waiting);
511 __fuse_request_send(fc, req);
513 EXPORT_SYMBOL_GPL(fuse_request_send);
515 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
517 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
518 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
521 switch (args->in.h.opcode) {
528 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
532 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
536 if (fc->minor < 12) {
537 switch (args->in.h.opcode) {
539 args->in.args[0].size = sizeof(struct fuse_open_in);
542 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
548 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
550 struct fuse_req *req;
553 req = fuse_get_req(fc, 0);
557 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
558 fuse_adjust_compat(fc, args);
560 req->in.h.opcode = args->in.h.opcode;
561 req->in.h.nodeid = args->in.h.nodeid;
562 req->in.numargs = args->in.numargs;
563 memcpy(req->in.args, args->in.args,
564 args->in.numargs * sizeof(struct fuse_in_arg));
565 req->out.argvar = args->out.argvar;
566 req->out.numargs = args->out.numargs;
567 memcpy(req->out.args, args->out.args,
568 args->out.numargs * sizeof(struct fuse_arg));
569 fuse_request_send(fc, req);
570 ret = req->out.h.error;
571 if (!ret && args->out.argvar) {
572 BUG_ON(args->out.numargs != 1);
573 ret = req->out.args[0].size;
575 fuse_put_request(fc, req);
581 * Called under fc->lock
583 * fc->connected must have been checked previously
585 void fuse_request_send_background_locked(struct fuse_conn *fc,
586 struct fuse_req *req)
588 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
589 if (!test_bit(FR_WAITING, &req->flags)) {
590 __set_bit(FR_WAITING, &req->flags);
591 atomic_inc(&fc->num_waiting);
593 __set_bit(FR_ISREPLY, &req->flags);
594 fc->num_background++;
595 if (fc->num_background == fc->max_background)
597 if (fc->num_background == fc->congestion_threshold &&
598 fc->bdi_initialized) {
599 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
600 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
602 list_add_tail(&req->list, &fc->bg_queue);
606 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
609 spin_lock(&fc->lock);
611 fuse_request_send_background_locked(fc, req);
612 spin_unlock(&fc->lock);
614 spin_unlock(&fc->lock);
615 req->out.h.error = -ENOTCONN;
617 fuse_put_request(fc, req);
620 EXPORT_SYMBOL_GPL(fuse_request_send_background);
622 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
623 struct fuse_req *req, u64 unique)
626 struct fuse_iqueue *fiq = &fc->iq;
628 __clear_bit(FR_ISREPLY, &req->flags);
629 req->in.h.unique = unique;
630 spin_lock(&fiq->waitq.lock);
631 if (fiq->connected) {
632 queue_request(fiq, req);
635 spin_unlock(&fiq->waitq.lock);
640 void fuse_force_forget(struct file *file, u64 nodeid)
642 struct inode *inode = file_inode(file);
643 struct fuse_conn *fc = get_fuse_conn(inode);
644 struct fuse_req *req;
645 struct fuse_forget_in inarg;
647 memset(&inarg, 0, sizeof(inarg));
649 req = fuse_get_req_nofail_nopages(fc, file);
650 req->in.h.opcode = FUSE_FORGET;
651 req->in.h.nodeid = nodeid;
653 req->in.args[0].size = sizeof(inarg);
654 req->in.args[0].value = &inarg;
655 __clear_bit(FR_ISREPLY, &req->flags);
656 __fuse_request_send(fc, req);
658 fuse_put_request(fc, req);
662 * Lock the request. Up to the next unlock_request() there mustn't be
663 * anything that could cause a page-fault. If the request was already
666 static int lock_request(struct fuse_req *req)
670 spin_lock(&req->waitq.lock);
671 if (test_bit(FR_ABORTED, &req->flags))
674 set_bit(FR_LOCKED, &req->flags);
675 spin_unlock(&req->waitq.lock);
681 * Unlock request. If it was aborted while locked, caller is responsible
682 * for unlocking and ending the request.
684 static int unlock_request(struct fuse_req *req)
688 spin_lock(&req->waitq.lock);
689 if (test_bit(FR_ABORTED, &req->flags))
692 clear_bit(FR_LOCKED, &req->flags);
693 spin_unlock(&req->waitq.lock);
698 struct fuse_copy_state {
700 struct fuse_req *req;
701 struct iov_iter *iter;
702 struct pipe_buffer *pipebufs;
703 struct pipe_buffer *currbuf;
704 struct pipe_inode_info *pipe;
705 unsigned long nr_segs;
709 unsigned move_pages:1;
712 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
713 struct iov_iter *iter)
715 memset(cs, 0, sizeof(*cs));
720 /* Unmap and put previous page of userspace buffer */
721 static void fuse_copy_finish(struct fuse_copy_state *cs)
724 struct pipe_buffer *buf = cs->currbuf;
727 buf->len = PAGE_SIZE - cs->len;
731 flush_dcache_page(cs->pg);
732 set_page_dirty_lock(cs->pg);
740 * Get another pagefull of userspace buffer, and map it to kernel
741 * address space, and lock request
743 static int fuse_copy_fill(struct fuse_copy_state *cs)
748 err = unlock_request(cs->req);
752 fuse_copy_finish(cs);
754 struct pipe_buffer *buf = cs->pipebufs;
757 err = pipe_buf_confirm(cs->pipe, buf);
761 BUG_ON(!cs->nr_segs);
764 cs->offset = buf->offset;
769 if (cs->nr_segs == cs->pipe->buffers)
772 page = alloc_page(GFP_HIGHUSER);
789 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
796 iov_iter_advance(cs->iter, err);
799 return lock_request(cs->req);
802 /* Do as much copy to/from userspace buffer as we can */
803 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
805 unsigned ncpy = min(*size, cs->len);
807 void *pgaddr = kmap_atomic(cs->pg);
808 void *buf = pgaddr + cs->offset;
811 memcpy(buf, *val, ncpy);
813 memcpy(*val, buf, ncpy);
815 kunmap_atomic(pgaddr);
824 static int fuse_check_page(struct page *page)
826 if (page_mapcount(page) ||
827 page->mapping != NULL ||
828 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
835 printk(KERN_WARNING "fuse: trying to steal weird page\n");
836 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);
842 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
845 struct page *oldpage = *pagep;
846 struct page *newpage;
847 struct pipe_buffer *buf = cs->pipebufs;
850 err = unlock_request(cs->req);
854 fuse_copy_finish(cs);
856 err = pipe_buf_confirm(cs->pipe, buf);
860 BUG_ON(!cs->nr_segs);
866 if (cs->len != PAGE_SIZE)
869 if (pipe_buf_steal(cs->pipe, buf) != 0)
874 if (!PageUptodate(newpage))
875 SetPageUptodate(newpage);
877 ClearPageMappedToDisk(newpage);
879 if (fuse_check_page(newpage) != 0)
880 goto out_fallback_unlock;
883 * This is a new and locked page, it shouldn't be mapped or
884 * have any special flags on it
886 if (WARN_ON(page_mapped(oldpage)))
887 goto out_fallback_unlock;
888 if (WARN_ON(page_has_private(oldpage)))
889 goto out_fallback_unlock;
890 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
891 goto out_fallback_unlock;
892 if (WARN_ON(PageMlocked(oldpage)))
893 goto out_fallback_unlock;
895 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
897 unlock_page(newpage);
903 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
904 lru_cache_add_file(newpage);
907 * Release while we have extra ref on stolen page. Otherwise
908 * anon_pipe_buf_release() might think the page can be reused.
910 pipe_buf_release(cs->pipe, buf);
913 spin_lock(&cs->req->waitq.lock);
914 if (test_bit(FR_ABORTED, &cs->req->flags))
918 spin_unlock(&cs->req->waitq.lock);
921 unlock_page(newpage);
926 unlock_page(oldpage);
927 /* Drop ref for ap->pages[] array */
933 /* Drop ref obtained in this function */
938 unlock_page(newpage);
941 cs->offset = buf->offset;
943 err = lock_request(cs->req);
950 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
951 unsigned offset, unsigned count)
953 struct pipe_buffer *buf;
956 if (cs->nr_segs == cs->pipe->buffers)
960 err = unlock_request(cs->req);
966 fuse_copy_finish(cs);
970 buf->offset = offset;
981 * Copy a page in the request to/from the userspace buffer. Must be
984 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
985 unsigned offset, unsigned count, int zeroing)
988 struct page *page = *pagep;
990 if (page && zeroing && count < PAGE_SIZE)
991 clear_highpage(page);
994 if (cs->write && cs->pipebufs && page) {
996 * Can't control lifetime of pipe buffers, so always
999 if (cs->req->user_pages) {
1000 err = fuse_copy_fill(cs);
1004 return fuse_ref_page(cs, page, offset, count);
1006 } else if (!cs->len) {
1007 if (cs->move_pages && page &&
1008 offset == 0 && count == PAGE_SIZE) {
1009 err = fuse_try_move_page(cs, pagep);
1013 err = fuse_copy_fill(cs);
1019 void *mapaddr = kmap_atomic(page);
1020 void *buf = mapaddr + offset;
1021 offset += fuse_copy_do(cs, &buf, &count);
1022 kunmap_atomic(mapaddr);
1024 offset += fuse_copy_do(cs, NULL, &count);
1026 if (page && !cs->write)
1027 flush_dcache_page(page);
1031 /* Copy pages in the request to/from userspace buffer */
1032 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1036 struct fuse_req *req = cs->req;
1038 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1040 unsigned offset = req->page_descs[i].offset;
1041 unsigned count = min(nbytes, req->page_descs[i].length);
1043 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1053 /* Copy a single argument in the request to/from userspace buffer */
1054 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1058 int err = fuse_copy_fill(cs);
1062 fuse_copy_do(cs, &val, &size);
1067 /* Copy request arguments to/from userspace buffer */
1068 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1069 unsigned argpages, struct fuse_arg *args,
1075 for (i = 0; !err && i < numargs; i++) {
1076 struct fuse_arg *arg = &args[i];
1077 if (i == numargs - 1 && argpages)
1078 err = fuse_copy_pages(cs, arg->size, zeroing);
1080 err = fuse_copy_one(cs, arg->value, arg->size);
1085 static int forget_pending(struct fuse_iqueue *fiq)
1087 return fiq->forget_list_head.next != NULL;
1090 static int request_pending(struct fuse_iqueue *fiq)
1092 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1093 forget_pending(fiq);
1097 * Transfer an interrupt request to userspace
1099 * Unlike other requests this is assembled on demand, without a need
1100 * to allocate a separate fuse_req structure.
1102 * Called with fiq->waitq.lock held, releases it
1104 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1105 struct fuse_copy_state *cs,
1106 size_t nbytes, struct fuse_req *req)
1107 __releases(fiq->waitq.lock)
1109 struct fuse_in_header ih;
1110 struct fuse_interrupt_in arg;
1111 unsigned reqsize = sizeof(ih) + sizeof(arg);
1114 list_del_init(&req->intr_entry);
1115 req->intr_unique = fuse_get_unique(fiq);
1116 memset(&ih, 0, sizeof(ih));
1117 memset(&arg, 0, sizeof(arg));
1119 ih.opcode = FUSE_INTERRUPT;
1120 ih.unique = req->intr_unique;
1121 arg.unique = req->in.h.unique;
1123 spin_unlock(&fiq->waitq.lock);
1124 if (nbytes < reqsize)
1127 err = fuse_copy_one(cs, &ih, sizeof(ih));
1129 err = fuse_copy_one(cs, &arg, sizeof(arg));
1130 fuse_copy_finish(cs);
1132 return err ? err : reqsize;
1135 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1139 struct fuse_forget_link *head = fiq->forget_list_head.next;
1140 struct fuse_forget_link **newhead = &head;
1143 for (count = 0; *newhead != NULL && count < max; count++)
1144 newhead = &(*newhead)->next;
1146 fiq->forget_list_head.next = *newhead;
1148 if (fiq->forget_list_head.next == NULL)
1149 fiq->forget_list_tail = &fiq->forget_list_head;
1157 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1158 struct fuse_copy_state *cs,
1160 __releases(fiq->waitq.lock)
1163 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1164 struct fuse_forget_in arg = {
1165 .nlookup = forget->forget_one.nlookup,
1167 struct fuse_in_header ih = {
1168 .opcode = FUSE_FORGET,
1169 .nodeid = forget->forget_one.nodeid,
1170 .unique = fuse_get_unique(fiq),
1171 .len = sizeof(ih) + sizeof(arg),
1174 spin_unlock(&fiq->waitq.lock);
1176 if (nbytes < ih.len)
1179 err = fuse_copy_one(cs, &ih, sizeof(ih));
1181 err = fuse_copy_one(cs, &arg, sizeof(arg));
1182 fuse_copy_finish(cs);
1190 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1191 struct fuse_copy_state *cs, size_t nbytes)
1192 __releases(fiq->waitq.lock)
1195 unsigned max_forgets;
1197 struct fuse_forget_link *head;
1198 struct fuse_batch_forget_in arg = { .count = 0 };
1199 struct fuse_in_header ih = {
1200 .opcode = FUSE_BATCH_FORGET,
1201 .unique = fuse_get_unique(fiq),
1202 .len = sizeof(ih) + sizeof(arg),
1205 if (nbytes < ih.len) {
1206 spin_unlock(&fiq->waitq.lock);
1210 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1211 head = dequeue_forget(fiq, max_forgets, &count);
1212 spin_unlock(&fiq->waitq.lock);
1215 ih.len += count * sizeof(struct fuse_forget_one);
1216 err = fuse_copy_one(cs, &ih, sizeof(ih));
1218 err = fuse_copy_one(cs, &arg, sizeof(arg));
1221 struct fuse_forget_link *forget = head;
1224 err = fuse_copy_one(cs, &forget->forget_one,
1225 sizeof(forget->forget_one));
1227 head = forget->next;
1231 fuse_copy_finish(cs);
1239 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1240 struct fuse_copy_state *cs,
1242 __releases(fiq->waitq.lock)
1244 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1245 return fuse_read_single_forget(fiq, cs, nbytes);
1247 return fuse_read_batch_forget(fiq, cs, nbytes);
1251 * Read a single request into the userspace filesystem's buffer. This
1252 * function waits until a request is available, then removes it from
1253 * the pending list and copies request data to userspace buffer. If
1254 * no reply is needed (FORGET) or request has been aborted or there
1255 * was an error during the copying then it's finished by calling
1256 * request_end(). Otherwise add it to the processing list, and set
1259 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1260 struct fuse_copy_state *cs, size_t nbytes)
1263 struct fuse_conn *fc = fud->fc;
1264 struct fuse_iqueue *fiq = &fc->iq;
1265 struct fuse_pqueue *fpq = &fud->pq;
1266 struct fuse_req *req;
1271 spin_lock(&fiq->waitq.lock);
1273 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1274 !request_pending(fiq))
1277 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1278 !fiq->connected || request_pending(fiq));
1283 if (!fiq->connected)
1286 if (!list_empty(&fiq->interrupts)) {
1287 req = list_entry(fiq->interrupts.next, struct fuse_req,
1289 return fuse_read_interrupt(fiq, cs, nbytes, req);
1292 if (forget_pending(fiq)) {
1293 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1294 return fuse_read_forget(fc, fiq, cs, nbytes);
1296 if (fiq->forget_batch <= -8)
1297 fiq->forget_batch = 16;
1300 req = list_entry(fiq->pending.next, struct fuse_req, list);
1301 clear_bit(FR_PENDING, &req->flags);
1302 list_del_init(&req->list);
1303 spin_unlock(&fiq->waitq.lock);
1306 reqsize = in->h.len;
1307 /* If request is too large, reply with an error and restart the read */
1308 if (nbytes < reqsize) {
1309 req->out.h.error = -EIO;
1310 /* SETXATTR is special, since it may contain too large data */
1311 if (in->h.opcode == FUSE_SETXATTR)
1312 req->out.h.error = -E2BIG;
1313 request_end(fc, req);
1316 spin_lock(&fpq->lock);
1318 * Must not put request on fpq->io queue after having been shut down by
1321 if (!fpq->connected) {
1322 req->out.h.error = err = -ECONNABORTED;
1326 list_add(&req->list, &fpq->io);
1327 spin_unlock(&fpq->lock);
1329 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1331 err = fuse_copy_args(cs, in->numargs, in->argpages,
1332 (struct fuse_arg *) in->args, 0);
1333 fuse_copy_finish(cs);
1334 spin_lock(&fpq->lock);
1335 clear_bit(FR_LOCKED, &req->flags);
1336 if (!fpq->connected) {
1341 req->out.h.error = -EIO;
1344 if (!test_bit(FR_ISREPLY, &req->flags)) {
1348 list_move_tail(&req->list, &fpq->processing);
1349 __fuse_get_request(req);
1350 set_bit(FR_SENT, &req->flags);
1351 spin_unlock(&fpq->lock);
1352 /* matches barrier in request_wait_answer() */
1353 smp_mb__after_atomic();
1354 if (test_bit(FR_INTERRUPTED, &req->flags))
1355 queue_interrupt(fiq, req);
1356 fuse_put_request(fc, req);
1361 if (!test_bit(FR_PRIVATE, &req->flags))
1362 list_del_init(&req->list);
1363 spin_unlock(&fpq->lock);
1364 request_end(fc, req);
1368 spin_unlock(&fiq->waitq.lock);
1372 static int fuse_dev_open(struct inode *inode, struct file *file)
1375 * The fuse device's file's private_data is used to hold
1376 * the fuse_conn(ection) when it is mounted, and is used to
1377 * keep track of whether the file has been mounted already.
1379 file->private_data = NULL;
1383 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1385 struct fuse_copy_state cs;
1386 struct file *file = iocb->ki_filp;
1387 struct fuse_dev *fud = fuse_get_dev(file);
1392 if (!iter_is_iovec(to))
1395 fuse_copy_init(&cs, 1, to);
1397 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1400 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1401 struct pipe_inode_info *pipe,
1402 size_t len, unsigned int flags)
1406 struct pipe_buffer *bufs;
1407 struct fuse_copy_state cs;
1408 struct fuse_dev *fud = fuse_get_dev(in);
1413 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1417 fuse_copy_init(&cs, 1, NULL);
1420 ret = fuse_dev_do_read(fud, in, &cs, len);
1424 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1429 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1431 * Need to be careful about this. Having buf->ops in module
1432 * code can Oops if the buffer persists after module unload.
1434 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1435 bufs[page_nr].flags = 0;
1436 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1437 if (unlikely(ret < 0))
1443 for (; page_nr < cs.nr_segs; page_nr++)
1444 put_page(bufs[page_nr].page);
1450 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1451 struct fuse_copy_state *cs)
1453 struct fuse_notify_poll_wakeup_out outarg;
1456 if (size != sizeof(outarg))
1459 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1463 fuse_copy_finish(cs);
1464 return fuse_notify_poll_wakeup(fc, &outarg);
1467 fuse_copy_finish(cs);
1471 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1472 struct fuse_copy_state *cs)
1474 struct fuse_notify_inval_inode_out outarg;
1477 if (size != sizeof(outarg))
1480 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1483 fuse_copy_finish(cs);
1485 down_read(&fc->killsb);
1488 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1489 outarg.off, outarg.len);
1491 up_read(&fc->killsb);
1495 fuse_copy_finish(cs);
1499 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1500 struct fuse_copy_state *cs)
1502 struct fuse_notify_inval_entry_out outarg;
1507 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1512 if (size < sizeof(outarg))
1515 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1519 err = -ENAMETOOLONG;
1520 if (outarg.namelen > FUSE_NAME_MAX)
1524 if (size != sizeof(outarg) + outarg.namelen + 1)
1528 name.len = outarg.namelen;
1529 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1532 fuse_copy_finish(cs);
1533 buf[outarg.namelen] = 0;
1535 down_read(&fc->killsb);
1538 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1539 up_read(&fc->killsb);
1545 fuse_copy_finish(cs);
1549 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1550 struct fuse_copy_state *cs)
1552 struct fuse_notify_delete_out outarg;
1557 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1562 if (size < sizeof(outarg))
1565 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1569 err = -ENAMETOOLONG;
1570 if (outarg.namelen > FUSE_NAME_MAX)
1574 if (size != sizeof(outarg) + outarg.namelen + 1)
1578 name.len = outarg.namelen;
1579 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1582 fuse_copy_finish(cs);
1583 buf[outarg.namelen] = 0;
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_SHIFT;
1640 offset = outarg.offset & ~PAGE_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_SIZE - offset);
1660 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1661 if (!err && offset == 0 &&
1662 (this_num == PAGE_SIZE || file_size == end))
1663 SetPageUptodate(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_MASK;
1705 file_size = i_size_read(inode);
1707 num = min(outarg->size, fc->max_write);
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->end = fuse_retrieve_end;
1726 index = outarg->offset >> PAGE_SHIFT;
1728 while (num && req->num_pages < num_pages) {
1730 unsigned int this_num;
1732 page = find_get_page(mapping, index);
1736 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1737 req->pages[req->num_pages] = page;
1738 req->page_descs[req->num_pages].offset = offset;
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);
1756 fuse_put_request(fc, req);
1762 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1763 struct fuse_copy_state *cs)
1765 struct fuse_notify_retrieve_out outarg;
1766 struct inode *inode;
1770 if (size != sizeof(outarg))
1773 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1777 fuse_copy_finish(cs);
1779 down_read(&fc->killsb);
1782 u64 nodeid = outarg.nodeid;
1784 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1786 err = fuse_retrieve(fc, inode, &outarg);
1790 up_read(&fc->killsb);
1795 fuse_copy_finish(cs);
1799 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1800 unsigned int size, struct fuse_copy_state *cs)
1802 /* Don't try to move pages (yet) */
1806 case FUSE_NOTIFY_POLL:
1807 return fuse_notify_poll(fc, size, cs);
1809 case FUSE_NOTIFY_INVAL_INODE:
1810 return fuse_notify_inval_inode(fc, size, cs);
1812 case FUSE_NOTIFY_INVAL_ENTRY:
1813 return fuse_notify_inval_entry(fc, size, cs);
1815 case FUSE_NOTIFY_STORE:
1816 return fuse_notify_store(fc, size, cs);
1818 case FUSE_NOTIFY_RETRIEVE:
1819 return fuse_notify_retrieve(fc, size, cs);
1821 case FUSE_NOTIFY_DELETE:
1822 return fuse_notify_delete(fc, size, cs);
1825 fuse_copy_finish(cs);
1830 /* Look up request on processing list by unique ID */
1831 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1833 struct fuse_req *req;
1835 list_for_each_entry(req, &fpq->processing, list) {
1836 if (req->in.h.unique == unique || req->intr_unique == unique)
1842 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1845 unsigned reqsize = sizeof(struct fuse_out_header);
1848 return nbytes != reqsize ? -EINVAL : 0;
1850 reqsize += len_args(out->numargs, out->args);
1852 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1854 else if (reqsize > nbytes) {
1855 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1856 unsigned diffsize = reqsize - nbytes;
1857 if (diffsize > lastarg->size)
1859 lastarg->size -= diffsize;
1861 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1866 * Write a single reply to a request. First the header is copied from
1867 * the write buffer. The request is then searched on the processing
1868 * list by the unique ID found in the header. If found, then remove
1869 * it from the list and copy the rest of the buffer to the request.
1870 * The request is finished by calling request_end()
1872 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1873 struct fuse_copy_state *cs, size_t nbytes)
1876 struct fuse_conn *fc = fud->fc;
1877 struct fuse_pqueue *fpq = &fud->pq;
1878 struct fuse_req *req;
1879 struct fuse_out_header oh;
1881 if (nbytes < sizeof(struct fuse_out_header))
1884 err = fuse_copy_one(cs, &oh, sizeof(oh));
1889 if (oh.len != nbytes)
1893 * Zero oh.unique indicates unsolicited notification message
1894 * and error contains notification code.
1897 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1898 return err ? err : nbytes;
1902 if (oh.error <= -512 || oh.error > 0)
1905 spin_lock(&fpq->lock);
1907 if (!fpq->connected)
1910 req = request_find(fpq, oh.unique);
1914 /* Is it an interrupt reply? */
1915 if (req->intr_unique == oh.unique) {
1916 __fuse_get_request(req);
1917 spin_unlock(&fpq->lock);
1920 if (nbytes != sizeof(struct fuse_out_header)) {
1921 fuse_put_request(fc, req);
1925 if (oh.error == -ENOSYS)
1926 fc->no_interrupt = 1;
1927 else if (oh.error == -EAGAIN)
1928 queue_interrupt(&fc->iq, req);
1929 fuse_put_request(fc, req);
1931 fuse_copy_finish(cs);
1935 clear_bit(FR_SENT, &req->flags);
1936 list_move(&req->list, &fpq->io);
1938 set_bit(FR_LOCKED, &req->flags);
1939 spin_unlock(&fpq->lock);
1941 if (!req->out.page_replace)
1944 err = copy_out_args(cs, &req->out, nbytes);
1945 fuse_copy_finish(cs);
1947 spin_lock(&fpq->lock);
1948 clear_bit(FR_LOCKED, &req->flags);
1949 if (!fpq->connected)
1952 req->out.h.error = -EIO;
1953 if (!test_bit(FR_PRIVATE, &req->flags))
1954 list_del_init(&req->list);
1955 spin_unlock(&fpq->lock);
1957 request_end(fc, req);
1959 return err ? err : nbytes;
1962 spin_unlock(&fpq->lock);
1964 fuse_copy_finish(cs);
1968 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1970 struct fuse_copy_state cs;
1971 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1976 if (!iter_is_iovec(from))
1979 fuse_copy_init(&cs, 0, from);
1981 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1984 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1985 struct file *out, loff_t *ppos,
1986 size_t len, unsigned int flags)
1990 struct pipe_buffer *bufs;
1991 struct fuse_copy_state cs;
1992 struct fuse_dev *fud;
1996 fud = fuse_get_dev(out);
2002 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
2010 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
2011 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
2019 struct pipe_buffer *ibuf;
2020 struct pipe_buffer *obuf;
2022 BUG_ON(nbuf >= pipe->buffers);
2023 BUG_ON(!pipe->nrbufs);
2024 ibuf = &pipe->bufs[pipe->curbuf];
2027 if (rem >= ibuf->len) {
2030 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2033 if (!pipe_buf_get(pipe, ibuf))
2037 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2039 ibuf->offset += obuf->len;
2040 ibuf->len -= obuf->len;
2047 fuse_copy_init(&cs, 0, NULL);
2052 if (flags & SPLICE_F_MOVE)
2055 ret = fuse_dev_do_write(fud, &cs, len);
2059 for (idx = 0; idx < nbuf; idx++) {
2060 struct pipe_buffer *buf = &bufs[idx];
2063 pipe_buf_release(pipe, buf);
2071 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2073 unsigned mask = POLLOUT | POLLWRNORM;
2074 struct fuse_iqueue *fiq;
2075 struct fuse_dev *fud = fuse_get_dev(file);
2081 poll_wait(file, &fiq->waitq, wait);
2083 spin_lock(&fiq->waitq.lock);
2084 if (!fiq->connected)
2086 else if (request_pending(fiq))
2087 mask |= POLLIN | POLLRDNORM;
2088 spin_unlock(&fiq->waitq.lock);
2094 * Abort all requests on the given list (pending or processing)
2096 * This function releases and reacquires fc->lock
2098 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2100 while (!list_empty(head)) {
2101 struct fuse_req *req;
2102 req = list_entry(head->next, struct fuse_req, list);
2103 req->out.h.error = -ECONNABORTED;
2104 clear_bit(FR_SENT, &req->flags);
2105 list_del_init(&req->list);
2106 request_end(fc, req);
2110 static void end_polls(struct fuse_conn *fc)
2114 p = rb_first(&fc->polled_files);
2117 struct fuse_file *ff;
2118 ff = rb_entry(p, struct fuse_file, polled_node);
2119 wake_up_interruptible_all(&ff->poll_wait);
2126 * Abort all requests.
2128 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2131 * The same effect is usually achievable through killing the filesystem daemon
2132 * and all users of the filesystem. The exception is the combination of an
2133 * asynchronous request and the tricky deadlock (see
2134 * Documentation/filesystems/fuse.txt).
2136 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2137 * requests, they should be finished off immediately. Locked requests will be
2138 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2139 * requests. It is possible that some request will finish before we can. This
2140 * is OK, the request will in that case be removed from the list before we touch
2143 void fuse_abort_conn(struct fuse_conn *fc)
2145 struct fuse_iqueue *fiq = &fc->iq;
2147 spin_lock(&fc->lock);
2148 if (fc->connected) {
2149 struct fuse_dev *fud;
2150 struct fuse_req *req, *next;
2156 fuse_set_initialized(fc);
2157 list_for_each_entry(fud, &fc->devices, entry) {
2158 struct fuse_pqueue *fpq = &fud->pq;
2160 spin_lock(&fpq->lock);
2162 list_for_each_entry_safe(req, next, &fpq->io, list) {
2163 req->out.h.error = -ECONNABORTED;
2164 spin_lock(&req->waitq.lock);
2165 set_bit(FR_ABORTED, &req->flags);
2166 if (!test_bit(FR_LOCKED, &req->flags)) {
2167 set_bit(FR_PRIVATE, &req->flags);
2168 __fuse_get_request(req);
2169 list_move(&req->list, &to_end1);
2171 spin_unlock(&req->waitq.lock);
2173 list_splice_init(&fpq->processing, &to_end2);
2174 spin_unlock(&fpq->lock);
2176 fc->max_background = UINT_MAX;
2179 spin_lock(&fiq->waitq.lock);
2181 list_splice_init(&fiq->pending, &to_end2);
2182 list_for_each_entry(req, &to_end2, list)
2183 clear_bit(FR_PENDING, &req->flags);
2184 while (forget_pending(fiq))
2185 kfree(dequeue_forget(fiq, 1, NULL));
2186 wake_up_all_locked(&fiq->waitq);
2187 spin_unlock(&fiq->waitq.lock);
2188 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2190 wake_up_all(&fc->blocked_waitq);
2191 spin_unlock(&fc->lock);
2193 while (!list_empty(&to_end1)) {
2194 req = list_first_entry(&to_end1, struct fuse_req, list);
2195 list_del_init(&req->list);
2196 request_end(fc, req);
2198 end_requests(fc, &to_end2);
2200 spin_unlock(&fc->lock);
2203 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2205 void fuse_wait_aborted(struct fuse_conn *fc)
2207 /* matches implicit memory barrier in fuse_drop_waiting() */
2209 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2212 int fuse_dev_release(struct inode *inode, struct file *file)
2214 struct fuse_dev *fud = fuse_get_dev(file);
2217 struct fuse_conn *fc = fud->fc;
2218 struct fuse_pqueue *fpq = &fud->pq;
2221 spin_lock(&fpq->lock);
2222 WARN_ON(!list_empty(&fpq->io));
2223 list_splice_init(&fpq->processing, &to_end);
2224 spin_unlock(&fpq->lock);
2226 end_requests(fc, &to_end);
2228 /* Are we the last open device? */
2229 if (atomic_dec_and_test(&fc->dev_count)) {
2230 WARN_ON(fc->iq.fasync != NULL);
2231 fuse_abort_conn(fc);
2237 EXPORT_SYMBOL_GPL(fuse_dev_release);
2239 static int fuse_dev_fasync(int fd, struct file *file, int on)
2241 struct fuse_dev *fud = fuse_get_dev(file);
2246 /* No locking - fasync_helper does its own locking */
2247 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2250 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2252 struct fuse_dev *fud;
2254 if (new->private_data)
2257 fud = fuse_dev_alloc(fc);
2261 new->private_data = fud;
2262 atomic_inc(&fc->dev_count);
2267 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2272 if (cmd == FUSE_DEV_IOC_CLONE) {
2276 if (!get_user(oldfd, (__u32 __user *) arg)) {
2277 struct file *old = fget(oldfd);
2281 struct fuse_dev *fud = NULL;
2284 * Check against file->f_op because CUSE
2285 * uses the same ioctl handler.
2287 if (old->f_op == file->f_op &&
2288 old->f_cred->user_ns == file->f_cred->user_ns)
2289 fud = fuse_get_dev(old);
2292 mutex_lock(&fuse_mutex);
2293 err = fuse_device_clone(fud->fc, file);
2294 mutex_unlock(&fuse_mutex);
2303 const struct file_operations fuse_dev_operations = {
2304 .owner = THIS_MODULE,
2305 .open = fuse_dev_open,
2306 .llseek = no_llseek,
2307 .read_iter = fuse_dev_read,
2308 .splice_read = fuse_dev_splice_read,
2309 .write_iter = fuse_dev_write,
2310 .splice_write = fuse_dev_splice_write,
2311 .poll = fuse_dev_poll,
2312 .release = fuse_dev_release,
2313 .fasync = fuse_dev_fasync,
2314 .unlocked_ioctl = fuse_dev_ioctl,
2315 .compat_ioctl = fuse_dev_ioctl,
2317 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2319 static struct miscdevice fuse_miscdevice = {
2320 .minor = FUSE_MINOR,
2322 .fops = &fuse_dev_operations,
2325 int __init fuse_dev_init(void)
2328 fuse_req_cachep = kmem_cache_create("fuse_request",
2329 sizeof(struct fuse_req),
2331 if (!fuse_req_cachep)
2334 err = misc_register(&fuse_miscdevice);
2336 goto out_cache_clean;
2341 kmem_cache_destroy(fuse_req_cachep);
2346 void fuse_dev_cleanup(void)
2348 misc_deregister(&fuse_miscdevice);
2349 kmem_cache_destroy(fuse_req_cachep);