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>
22 #include <linux/freezer.h>
24 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
25 MODULE_ALIAS("devname:fuse");
27 static struct kmem_cache *fuse_req_cachep;
29 static struct fuse_dev *fuse_get_dev(struct file *file)
32 * Lockless access is OK, because file->private data is set
33 * once during mount and is valid until the file is released.
35 return ACCESS_ONCE(file->private_data);
38 static void fuse_request_init(struct fuse_req *req, struct page **pages,
39 struct fuse_page_desc *page_descs,
42 memset(req, 0, sizeof(*req));
43 memset(pages, 0, sizeof(*pages) * npages);
44 memset(page_descs, 0, sizeof(*page_descs) * npages);
45 INIT_LIST_HEAD(&req->list);
46 INIT_LIST_HEAD(&req->intr_entry);
47 init_waitqueue_head(&req->waitq);
48 atomic_set(&req->count, 1);
50 req->page_descs = page_descs;
51 req->max_pages = npages;
52 __set_bit(FR_PENDING, &req->flags);
55 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
57 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
60 struct fuse_page_desc *page_descs;
62 if (npages <= FUSE_REQ_INLINE_PAGES) {
63 pages = req->inline_pages;
64 page_descs = req->inline_page_descs;
66 pages = kmalloc(sizeof(struct page *) * npages, flags);
67 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
71 if (!pages || !page_descs) {
74 kmem_cache_free(fuse_req_cachep, req);
78 fuse_request_init(req, pages, page_descs, npages);
83 struct fuse_req *fuse_request_alloc(unsigned npages)
85 return __fuse_request_alloc(npages, GFP_KERNEL);
87 EXPORT_SYMBOL_GPL(fuse_request_alloc);
89 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
91 return __fuse_request_alloc(npages, GFP_NOFS);
94 void fuse_request_free(struct fuse_req *req)
96 if (req->pages != req->inline_pages) {
98 kfree(req->page_descs);
100 kmem_cache_free(fuse_req_cachep, req);
103 static void block_sigs(sigset_t *oldset)
107 siginitsetinv(&mask, sigmask(SIGKILL));
108 sigprocmask(SIG_BLOCK, &mask, oldset);
111 static void restore_sigs(sigset_t *oldset)
113 sigprocmask(SIG_SETMASK, oldset, NULL);
116 void __fuse_get_request(struct fuse_req *req)
118 atomic_inc(&req->count);
121 /* Must be called with > 1 refcount */
122 static void __fuse_put_request(struct fuse_req *req)
124 BUG_ON(atomic_read(&req->count) < 2);
125 atomic_dec(&req->count);
128 static void fuse_req_init_context(struct fuse_req *req)
130 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
131 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
132 req->in.h.pid = current->pid;
135 void fuse_set_initialized(struct fuse_conn *fc)
137 /* Make sure stores before this are seen on another CPU */
142 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
144 return !fc->initialized || (for_background && fc->blocked);
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)) {
159 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
160 !fuse_block_alloc(fc, for_background));
161 restore_sigs(&oldset);
166 /* Matches smp_wmb() in fuse_set_initialized() */
177 req = fuse_request_alloc(npages);
181 wake_up(&fc->blocked_waitq);
185 fuse_req_init_context(req);
186 __set_bit(FR_WAITING, &req->flags);
188 __set_bit(FR_BACKGROUND, &req->flags);
193 atomic_dec(&fc->num_waiting);
197 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
199 return __fuse_get_req(fc, npages, false);
201 EXPORT_SYMBOL_GPL(fuse_get_req);
203 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
206 return __fuse_get_req(fc, npages, true);
208 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
211 * Return request in fuse_file->reserved_req. However that may
212 * currently be in use. If that is the case, wait for it to become
215 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
218 struct fuse_req *req = NULL;
219 struct fuse_file *ff = file->private_data;
222 wait_event(fc->reserved_req_waitq, ff->reserved_req);
223 spin_lock(&fc->lock);
224 if (ff->reserved_req) {
225 req = ff->reserved_req;
226 ff->reserved_req = NULL;
227 req->stolen_file = get_file(file);
229 spin_unlock(&fc->lock);
236 * Put stolen request back into fuse_file->reserved_req
238 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
240 struct file *file = req->stolen_file;
241 struct fuse_file *ff = file->private_data;
243 spin_lock(&fc->lock);
244 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
245 BUG_ON(ff->reserved_req);
246 ff->reserved_req = req;
247 wake_up_all(&fc->reserved_req_waitq);
248 spin_unlock(&fc->lock);
253 * Gets a requests for a file operation, always succeeds
255 * This is used for sending the FLUSH request, which must get to
256 * userspace, due to POSIX locks which may need to be unlocked.
258 * If allocation fails due to OOM, use the reserved request in
261 * This is very unlikely to deadlock accidentally, since the
262 * filesystem should not have it's own file open. If deadlock is
263 * intentional, it can still be broken by "aborting" the filesystem.
265 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
268 struct fuse_req *req;
270 atomic_inc(&fc->num_waiting);
271 wait_event(fc->blocked_waitq, fc->initialized);
272 /* Matches smp_wmb() in fuse_set_initialized() */
274 req = fuse_request_alloc(0);
276 req = get_reserved_req(fc, file);
278 fuse_req_init_context(req);
279 __set_bit(FR_WAITING, &req->flags);
280 __clear_bit(FR_BACKGROUND, &req->flags);
284 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
286 if (atomic_dec_and_test(&req->count)) {
287 if (test_bit(FR_BACKGROUND, &req->flags)) {
289 * We get here in the unlikely case that a background
290 * request was allocated but not sent
292 spin_lock(&fc->lock);
294 wake_up(&fc->blocked_waitq);
295 spin_unlock(&fc->lock);
298 if (test_bit(FR_WAITING, &req->flags)) {
299 __clear_bit(FR_WAITING, &req->flags);
300 atomic_dec(&fc->num_waiting);
303 if (req->stolen_file)
304 put_reserved_req(fc, req);
306 fuse_request_free(req);
309 EXPORT_SYMBOL_GPL(fuse_put_request);
311 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
316 for (i = 0; i < numargs; i++)
317 nbytes += args[i].size;
322 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
324 return ++fiq->reqctr;
327 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
329 req->in.h.len = sizeof(struct fuse_in_header) +
330 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
331 list_add_tail(&req->list, &fiq->pending);
332 wake_up_locked(&fiq->waitq);
333 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
336 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
337 u64 nodeid, u64 nlookup)
339 struct fuse_iqueue *fiq = &fc->iq;
341 forget->forget_one.nodeid = nodeid;
342 forget->forget_one.nlookup = nlookup;
344 spin_lock(&fiq->waitq.lock);
345 if (fiq->connected) {
346 fiq->forget_list_tail->next = forget;
347 fiq->forget_list_tail = forget;
348 wake_up_locked(&fiq->waitq);
349 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
353 spin_unlock(&fiq->waitq.lock);
356 static void flush_bg_queue(struct fuse_conn *fc)
358 while (fc->active_background < fc->max_background &&
359 !list_empty(&fc->bg_queue)) {
360 struct fuse_req *req;
361 struct fuse_iqueue *fiq = &fc->iq;
363 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
364 list_del(&req->list);
365 fc->active_background++;
366 spin_lock(&fiq->waitq.lock);
367 req->in.h.unique = fuse_get_unique(fiq);
368 queue_request(fiq, req);
369 spin_unlock(&fiq->waitq.lock);
374 * This function is called when a request is finished. Either a reply
375 * has arrived or it was aborted (and not yet sent) or some error
376 * occurred during communication with userspace, or the device file
377 * was closed. The requester thread is woken up (if still waiting),
378 * the 'end' callback is called if given, else the reference to the
379 * request is released
381 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
383 struct fuse_iqueue *fiq = &fc->iq;
385 if (test_and_set_bit(FR_FINISHED, &req->flags))
388 spin_lock(&fiq->waitq.lock);
389 list_del_init(&req->intr_entry);
390 spin_unlock(&fiq->waitq.lock);
391 WARN_ON(test_bit(FR_PENDING, &req->flags));
392 WARN_ON(test_bit(FR_SENT, &req->flags));
393 if (test_bit(FR_BACKGROUND, &req->flags)) {
394 spin_lock(&fc->lock);
395 clear_bit(FR_BACKGROUND, &req->flags);
396 if (fc->num_background == fc->max_background)
399 /* Wake up next waiter, if any */
400 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
401 wake_up(&fc->blocked_waitq);
403 if (fc->num_background == fc->congestion_threshold &&
404 fc->connected && fc->bdi_initialized) {
405 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
406 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
408 fc->num_background--;
409 fc->active_background--;
411 spin_unlock(&fc->lock);
413 wake_up(&req->waitq);
416 fuse_put_request(fc, req);
419 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
421 spin_lock(&fiq->waitq.lock);
422 if (list_empty(&req->intr_entry)) {
423 list_add_tail(&req->intr_entry, &fiq->interrupts);
424 wake_up_locked(&fiq->waitq);
426 spin_unlock(&fiq->waitq.lock);
427 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
430 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
432 struct fuse_iqueue *fiq = &fc->iq;
435 if (!fc->no_interrupt) {
436 /* Any signal may interrupt this */
437 err = wait_event_interruptible(req->waitq,
438 test_bit(FR_FINISHED, &req->flags));
442 set_bit(FR_INTERRUPTED, &req->flags);
443 /* matches barrier in fuse_dev_do_read() */
444 smp_mb__after_atomic();
445 if (test_bit(FR_SENT, &req->flags))
446 queue_interrupt(fiq, req);
449 if (!test_bit(FR_FORCE, &req->flags)) {
452 /* Only fatal signals may interrupt this */
454 err = wait_event_interruptible(req->waitq,
455 test_bit(FR_FINISHED, &req->flags));
456 restore_sigs(&oldset);
461 spin_lock(&fiq->waitq.lock);
462 /* Request is not yet in userspace, bail out */
463 if (test_bit(FR_PENDING, &req->flags)) {
464 list_del(&req->list);
465 spin_unlock(&fiq->waitq.lock);
466 __fuse_put_request(req);
467 req->out.h.error = -EINTR;
470 spin_unlock(&fiq->waitq.lock);
474 * Either request is already in userspace, or it was forced.
477 while (!test_bit(FR_FINISHED, &req->flags))
478 wait_event_freezable(req->waitq,
479 test_bit(FR_FINISHED, &req->flags));
482 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
484 struct fuse_iqueue *fiq = &fc->iq;
486 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
487 spin_lock(&fiq->waitq.lock);
488 if (!fiq->connected) {
489 spin_unlock(&fiq->waitq.lock);
490 req->out.h.error = -ENOTCONN;
492 req->in.h.unique = fuse_get_unique(fiq);
493 queue_request(fiq, req);
494 /* acquire extra reference, since request is still needed
495 after request_end() */
496 __fuse_get_request(req);
497 spin_unlock(&fiq->waitq.lock);
499 request_wait_answer(fc, req);
500 /* Pairs with smp_wmb() in request_end() */
505 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
507 __set_bit(FR_ISREPLY, &req->flags);
508 if (!test_bit(FR_WAITING, &req->flags)) {
509 __set_bit(FR_WAITING, &req->flags);
510 atomic_inc(&fc->num_waiting);
512 __fuse_request_send(fc, req);
514 EXPORT_SYMBOL_GPL(fuse_request_send);
516 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
518 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
519 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
522 switch (args->in.h.opcode) {
529 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
533 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
537 if (fc->minor < 12) {
538 switch (args->in.h.opcode) {
540 args->in.args[0].size = sizeof(struct fuse_open_in);
543 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
549 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
551 struct fuse_req *req;
554 req = fuse_get_req(fc, 0);
558 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
559 fuse_adjust_compat(fc, args);
561 req->in.h.opcode = args->in.h.opcode;
562 req->in.h.nodeid = args->in.h.nodeid;
563 req->in.numargs = args->in.numargs;
564 memcpy(req->in.args, args->in.args,
565 args->in.numargs * sizeof(struct fuse_in_arg));
566 req->out.argvar = args->out.argvar;
567 req->out.numargs = args->out.numargs;
568 memcpy(req->out.args, args->out.args,
569 args->out.numargs * sizeof(struct fuse_arg));
570 fuse_request_send(fc, req);
571 ret = req->out.h.error;
572 if (!ret && args->out.argvar) {
573 BUG_ON(args->out.numargs != 1);
574 ret = req->out.args[0].size;
576 fuse_put_request(fc, req);
582 * Called under fc->lock
584 * fc->connected must have been checked previously
586 void fuse_request_send_background_locked(struct fuse_conn *fc,
587 struct fuse_req *req)
589 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
590 if (!test_bit(FR_WAITING, &req->flags)) {
591 __set_bit(FR_WAITING, &req->flags);
592 atomic_inc(&fc->num_waiting);
594 __set_bit(FR_ISREPLY, &req->flags);
595 fc->num_background++;
596 if (fc->num_background == fc->max_background)
598 if (fc->num_background == fc->congestion_threshold &&
599 fc->bdi_initialized) {
600 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
601 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
603 list_add_tail(&req->list, &fc->bg_queue);
607 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
610 spin_lock(&fc->lock);
612 fuse_request_send_background_locked(fc, req);
613 spin_unlock(&fc->lock);
615 spin_unlock(&fc->lock);
616 req->out.h.error = -ENOTCONN;
618 fuse_put_request(fc, req);
621 EXPORT_SYMBOL_GPL(fuse_request_send_background);
623 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
624 struct fuse_req *req, u64 unique)
627 struct fuse_iqueue *fiq = &fc->iq;
629 __clear_bit(FR_ISREPLY, &req->flags);
630 req->in.h.unique = unique;
631 spin_lock(&fiq->waitq.lock);
632 if (fiq->connected) {
633 queue_request(fiq, req);
636 spin_unlock(&fiq->waitq.lock);
641 void fuse_force_forget(struct file *file, u64 nodeid)
643 struct inode *inode = file_inode(file);
644 struct fuse_conn *fc = get_fuse_conn(inode);
645 struct fuse_req *req;
646 struct fuse_forget_in inarg;
648 memset(&inarg, 0, sizeof(inarg));
650 req = fuse_get_req_nofail_nopages(fc, file);
651 req->in.h.opcode = FUSE_FORGET;
652 req->in.h.nodeid = nodeid;
654 req->in.args[0].size = sizeof(inarg);
655 req->in.args[0].value = &inarg;
656 __clear_bit(FR_ISREPLY, &req->flags);
657 __fuse_request_send(fc, req);
659 fuse_put_request(fc, req);
663 * Lock the request. Up to the next unlock_request() there mustn't be
664 * anything that could cause a page-fault. If the request was already
667 static int lock_request(struct fuse_req *req)
671 spin_lock(&req->waitq.lock);
672 if (test_bit(FR_ABORTED, &req->flags))
675 set_bit(FR_LOCKED, &req->flags);
676 spin_unlock(&req->waitq.lock);
682 * Unlock request. If it was aborted while locked, caller is responsible
683 * for unlocking and ending the request.
685 static int unlock_request(struct fuse_req *req)
689 spin_lock(&req->waitq.lock);
690 if (test_bit(FR_ABORTED, &req->flags))
693 clear_bit(FR_LOCKED, &req->flags);
694 spin_unlock(&req->waitq.lock);
699 struct fuse_copy_state {
701 struct fuse_req *req;
702 struct iov_iter *iter;
703 struct pipe_buffer *pipebufs;
704 struct pipe_buffer *currbuf;
705 struct pipe_inode_info *pipe;
706 unsigned long nr_segs;
710 unsigned move_pages:1;
713 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
714 struct iov_iter *iter)
716 memset(cs, 0, sizeof(*cs));
721 /* Unmap and put previous page of userspace buffer */
722 static void fuse_copy_finish(struct fuse_copy_state *cs)
725 struct pipe_buffer *buf = cs->currbuf;
728 buf->len = PAGE_SIZE - cs->len;
732 flush_dcache_page(cs->pg);
733 set_page_dirty_lock(cs->pg);
741 * Get another pagefull of userspace buffer, and map it to kernel
742 * address space, and lock request
744 static int fuse_copy_fill(struct fuse_copy_state *cs)
749 err = unlock_request(cs->req);
753 fuse_copy_finish(cs);
755 struct pipe_buffer *buf = cs->pipebufs;
758 err = buf->ops->confirm(cs->pipe, buf);
762 BUG_ON(!cs->nr_segs);
765 cs->offset = buf->offset;
770 if (cs->nr_segs == cs->pipe->buffers)
773 page = alloc_page(GFP_HIGHUSER);
790 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
798 iov_iter_advance(cs->iter, err);
801 return lock_request(cs->req);
804 /* Do as much copy to/from userspace buffer as we can */
805 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
807 unsigned ncpy = min(*size, cs->len);
809 void *pgaddr = kmap_atomic(cs->pg);
810 void *buf = pgaddr + cs->offset;
813 memcpy(buf, *val, ncpy);
815 memcpy(*val, buf, ncpy);
817 kunmap_atomic(pgaddr);
826 static int fuse_check_page(struct page *page)
828 if (page_mapcount(page) ||
829 page->mapping != NULL ||
830 page_count(page) != 1 ||
831 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
838 printk(KERN_WARNING "fuse: trying to steal weird page\n");
839 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);
845 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
848 struct page *oldpage = *pagep;
849 struct page *newpage;
850 struct pipe_buffer *buf = cs->pipebufs;
852 err = unlock_request(cs->req);
856 fuse_copy_finish(cs);
858 err = buf->ops->confirm(cs->pipe, buf);
862 BUG_ON(!cs->nr_segs);
868 if (cs->len != PAGE_SIZE)
871 if (buf->ops->steal(cs->pipe, buf) != 0)
876 if (!PageUptodate(newpage))
877 SetPageUptodate(newpage);
879 ClearPageMappedToDisk(newpage);
881 if (fuse_check_page(newpage) != 0)
882 goto out_fallback_unlock;
885 * This is a new and locked page, it shouldn't be mapped or
886 * have any special flags on it
888 if (WARN_ON(page_mapped(oldpage)))
889 goto out_fallback_unlock;
890 if (WARN_ON(page_has_private(oldpage)))
891 goto out_fallback_unlock;
892 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
893 goto out_fallback_unlock;
894 if (WARN_ON(PageMlocked(oldpage)))
895 goto out_fallback_unlock;
897 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
899 unlock_page(newpage);
903 page_cache_get(newpage);
905 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
906 lru_cache_add_file(newpage);
909 spin_lock(&cs->req->waitq.lock);
910 if (test_bit(FR_ABORTED, &cs->req->flags))
914 spin_unlock(&cs->req->waitq.lock);
917 unlock_page(newpage);
918 page_cache_release(newpage);
922 unlock_page(oldpage);
923 page_cache_release(oldpage);
929 unlock_page(newpage);
932 cs->offset = buf->offset;
934 err = lock_request(cs->req);
941 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
942 unsigned offset, unsigned count)
944 struct pipe_buffer *buf;
947 if (cs->nr_segs == cs->pipe->buffers)
950 err = unlock_request(cs->req);
954 fuse_copy_finish(cs);
957 page_cache_get(page);
959 buf->offset = offset;
970 * Copy a page in the request to/from the userspace buffer. Must be
973 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
974 unsigned offset, unsigned count, int zeroing)
977 struct page *page = *pagep;
979 if (page && zeroing && count < PAGE_SIZE)
980 clear_highpage(page);
983 if (cs->write && cs->pipebufs && page) {
984 return fuse_ref_page(cs, page, offset, count);
985 } else if (!cs->len) {
986 if (cs->move_pages && page &&
987 offset == 0 && count == PAGE_SIZE) {
988 err = fuse_try_move_page(cs, pagep);
992 err = fuse_copy_fill(cs);
998 void *mapaddr = kmap_atomic(page);
999 void *buf = mapaddr + offset;
1000 offset += fuse_copy_do(cs, &buf, &count);
1001 kunmap_atomic(mapaddr);
1003 offset += fuse_copy_do(cs, NULL, &count);
1005 if (page && !cs->write)
1006 flush_dcache_page(page);
1010 /* Copy pages in the request to/from userspace buffer */
1011 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1015 struct fuse_req *req = cs->req;
1017 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1019 unsigned offset = req->page_descs[i].offset;
1020 unsigned count = min(nbytes, req->page_descs[i].length);
1022 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1032 /* Copy a single argument in the request to/from userspace buffer */
1033 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1037 int err = fuse_copy_fill(cs);
1041 fuse_copy_do(cs, &val, &size);
1046 /* Copy request arguments to/from userspace buffer */
1047 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1048 unsigned argpages, struct fuse_arg *args,
1054 for (i = 0; !err && i < numargs; i++) {
1055 struct fuse_arg *arg = &args[i];
1056 if (i == numargs - 1 && argpages)
1057 err = fuse_copy_pages(cs, arg->size, zeroing);
1059 err = fuse_copy_one(cs, arg->value, arg->size);
1064 static int forget_pending(struct fuse_iqueue *fiq)
1066 return fiq->forget_list_head.next != NULL;
1069 static int request_pending(struct fuse_iqueue *fiq)
1071 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1072 forget_pending(fiq);
1076 * Transfer an interrupt request to userspace
1078 * Unlike other requests this is assembled on demand, without a need
1079 * to allocate a separate fuse_req structure.
1081 * Called with fiq->waitq.lock held, releases it
1083 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1084 struct fuse_copy_state *cs,
1085 size_t nbytes, struct fuse_req *req)
1086 __releases(fiq->waitq.lock)
1088 struct fuse_in_header ih;
1089 struct fuse_interrupt_in arg;
1090 unsigned reqsize = sizeof(ih) + sizeof(arg);
1093 list_del_init(&req->intr_entry);
1094 req->intr_unique = fuse_get_unique(fiq);
1095 memset(&ih, 0, sizeof(ih));
1096 memset(&arg, 0, sizeof(arg));
1098 ih.opcode = FUSE_INTERRUPT;
1099 ih.unique = req->intr_unique;
1100 arg.unique = req->in.h.unique;
1102 spin_unlock(&fiq->waitq.lock);
1103 if (nbytes < reqsize)
1106 err = fuse_copy_one(cs, &ih, sizeof(ih));
1108 err = fuse_copy_one(cs, &arg, sizeof(arg));
1109 fuse_copy_finish(cs);
1111 return err ? err : reqsize;
1114 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1118 struct fuse_forget_link *head = fiq->forget_list_head.next;
1119 struct fuse_forget_link **newhead = &head;
1122 for (count = 0; *newhead != NULL && count < max; count++)
1123 newhead = &(*newhead)->next;
1125 fiq->forget_list_head.next = *newhead;
1127 if (fiq->forget_list_head.next == NULL)
1128 fiq->forget_list_tail = &fiq->forget_list_head;
1136 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1137 struct fuse_copy_state *cs,
1139 __releases(fiq->waitq.lock)
1142 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1143 struct fuse_forget_in arg = {
1144 .nlookup = forget->forget_one.nlookup,
1146 struct fuse_in_header ih = {
1147 .opcode = FUSE_FORGET,
1148 .nodeid = forget->forget_one.nodeid,
1149 .unique = fuse_get_unique(fiq),
1150 .len = sizeof(ih) + sizeof(arg),
1153 spin_unlock(&fiq->waitq.lock);
1155 if (nbytes < ih.len)
1158 err = fuse_copy_one(cs, &ih, sizeof(ih));
1160 err = fuse_copy_one(cs, &arg, sizeof(arg));
1161 fuse_copy_finish(cs);
1169 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1170 struct fuse_copy_state *cs, size_t nbytes)
1171 __releases(fiq->waitq.lock)
1174 unsigned max_forgets;
1176 struct fuse_forget_link *head;
1177 struct fuse_batch_forget_in arg = { .count = 0 };
1178 struct fuse_in_header ih = {
1179 .opcode = FUSE_BATCH_FORGET,
1180 .unique = fuse_get_unique(fiq),
1181 .len = sizeof(ih) + sizeof(arg),
1184 if (nbytes < ih.len) {
1185 spin_unlock(&fiq->waitq.lock);
1189 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1190 head = dequeue_forget(fiq, max_forgets, &count);
1191 spin_unlock(&fiq->waitq.lock);
1194 ih.len += count * sizeof(struct fuse_forget_one);
1195 err = fuse_copy_one(cs, &ih, sizeof(ih));
1197 err = fuse_copy_one(cs, &arg, sizeof(arg));
1200 struct fuse_forget_link *forget = head;
1203 err = fuse_copy_one(cs, &forget->forget_one,
1204 sizeof(forget->forget_one));
1206 head = forget->next;
1210 fuse_copy_finish(cs);
1218 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1219 struct fuse_copy_state *cs,
1221 __releases(fiq->waitq.lock)
1223 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1224 return fuse_read_single_forget(fiq, cs, nbytes);
1226 return fuse_read_batch_forget(fiq, cs, nbytes);
1230 * Read a single request into the userspace filesystem's buffer. This
1231 * function waits until a request is available, then removes it from
1232 * the pending list and copies request data to userspace buffer. If
1233 * no reply is needed (FORGET) or request has been aborted or there
1234 * was an error during the copying then it's finished by calling
1235 * request_end(). Otherwise add it to the processing list, and set
1238 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1239 struct fuse_copy_state *cs, size_t nbytes)
1242 struct fuse_conn *fc = fud->fc;
1243 struct fuse_iqueue *fiq = &fc->iq;
1244 struct fuse_pqueue *fpq = &fud->pq;
1245 struct fuse_req *req;
1250 spin_lock(&fiq->waitq.lock);
1252 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1253 !request_pending(fiq))
1256 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1257 !fiq->connected || request_pending(fiq));
1262 if (!fiq->connected)
1265 if (!list_empty(&fiq->interrupts)) {
1266 req = list_entry(fiq->interrupts.next, struct fuse_req,
1268 return fuse_read_interrupt(fiq, cs, nbytes, req);
1271 if (forget_pending(fiq)) {
1272 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1273 return fuse_read_forget(fc, fiq, cs, nbytes);
1275 if (fiq->forget_batch <= -8)
1276 fiq->forget_batch = 16;
1279 req = list_entry(fiq->pending.next, struct fuse_req, list);
1280 clear_bit(FR_PENDING, &req->flags);
1281 list_del_init(&req->list);
1282 spin_unlock(&fiq->waitq.lock);
1285 reqsize = in->h.len;
1286 /* If request is too large, reply with an error and restart the read */
1287 if (nbytes < reqsize) {
1288 req->out.h.error = -EIO;
1289 /* SETXATTR is special, since it may contain too large data */
1290 if (in->h.opcode == FUSE_SETXATTR)
1291 req->out.h.error = -E2BIG;
1292 request_end(fc, req);
1295 spin_lock(&fpq->lock);
1296 list_add(&req->list, &fpq->io);
1297 spin_unlock(&fpq->lock);
1299 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1301 err = fuse_copy_args(cs, in->numargs, in->argpages,
1302 (struct fuse_arg *) in->args, 0);
1303 fuse_copy_finish(cs);
1304 spin_lock(&fpq->lock);
1305 clear_bit(FR_LOCKED, &req->flags);
1306 if (!fpq->connected) {
1311 req->out.h.error = -EIO;
1314 if (!test_bit(FR_ISREPLY, &req->flags)) {
1318 list_move_tail(&req->list, &fpq->processing);
1319 spin_unlock(&fpq->lock);
1320 set_bit(FR_SENT, &req->flags);
1321 /* matches barrier in request_wait_answer() */
1322 smp_mb__after_atomic();
1323 if (test_bit(FR_INTERRUPTED, &req->flags))
1324 queue_interrupt(fiq, req);
1329 if (!test_bit(FR_PRIVATE, &req->flags))
1330 list_del_init(&req->list);
1331 spin_unlock(&fpq->lock);
1332 request_end(fc, req);
1336 spin_unlock(&fiq->waitq.lock);
1340 static int fuse_dev_open(struct inode *inode, struct file *file)
1343 * The fuse device's file's private_data is used to hold
1344 * the fuse_conn(ection) when it is mounted, and is used to
1345 * keep track of whether the file has been mounted already.
1347 file->private_data = NULL;
1351 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1353 struct fuse_copy_state cs;
1354 struct file *file = iocb->ki_filp;
1355 struct fuse_dev *fud = fuse_get_dev(file);
1360 if (!iter_is_iovec(to))
1363 fuse_copy_init(&cs, 1, to);
1365 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1368 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1369 struct pipe_inode_info *pipe,
1370 size_t len, unsigned int flags)
1375 struct pipe_buffer *bufs;
1376 struct fuse_copy_state cs;
1377 struct fuse_dev *fud = fuse_get_dev(in);
1382 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1386 fuse_copy_init(&cs, 1, NULL);
1389 ret = fuse_dev_do_read(fud, in, &cs, len);
1396 if (!pipe->readers) {
1397 send_sig(SIGPIPE, current, 0);
1403 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1408 while (page_nr < cs.nr_segs) {
1409 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1410 struct pipe_buffer *buf = pipe->bufs + newbuf;
1412 buf->page = bufs[page_nr].page;
1413 buf->offset = bufs[page_nr].offset;
1414 buf->len = bufs[page_nr].len;
1416 * Need to be careful about this. Having buf->ops in module
1417 * code can Oops if the buffer persists after module unload.
1419 buf->ops = &nosteal_pipe_buf_ops;
1434 if (waitqueue_active(&pipe->wait))
1435 wake_up_interruptible(&pipe->wait);
1436 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1440 for (; page_nr < cs.nr_segs; page_nr++)
1441 page_cache_release(bufs[page_nr].page);
1447 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1448 struct fuse_copy_state *cs)
1450 struct fuse_notify_poll_wakeup_out outarg;
1453 if (size != sizeof(outarg))
1456 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1460 fuse_copy_finish(cs);
1461 return fuse_notify_poll_wakeup(fc, &outarg);
1464 fuse_copy_finish(cs);
1468 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1469 struct fuse_copy_state *cs)
1471 struct fuse_notify_inval_inode_out outarg;
1474 if (size != sizeof(outarg))
1477 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1480 fuse_copy_finish(cs);
1482 down_read(&fc->killsb);
1485 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1486 outarg.off, outarg.len);
1488 up_read(&fc->killsb);
1492 fuse_copy_finish(cs);
1496 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1497 struct fuse_copy_state *cs)
1499 struct fuse_notify_inval_entry_out outarg;
1504 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1509 if (size < sizeof(outarg))
1512 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1516 err = -ENAMETOOLONG;
1517 if (outarg.namelen > FUSE_NAME_MAX)
1521 if (size != sizeof(outarg) + outarg.namelen + 1)
1525 name.len = outarg.namelen;
1526 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1529 fuse_copy_finish(cs);
1530 buf[outarg.namelen] = 0;
1531 name.hash = full_name_hash(name.name, name.len);
1533 down_read(&fc->killsb);
1536 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1537 up_read(&fc->killsb);
1543 fuse_copy_finish(cs);
1547 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1548 struct fuse_copy_state *cs)
1550 struct fuse_notify_delete_out outarg;
1555 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1560 if (size < sizeof(outarg))
1563 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1567 err = -ENAMETOOLONG;
1568 if (outarg.namelen > FUSE_NAME_MAX)
1572 if (size != sizeof(outarg) + outarg.namelen + 1)
1576 name.len = outarg.namelen;
1577 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1580 fuse_copy_finish(cs);
1581 buf[outarg.namelen] = 0;
1582 name.hash = full_name_hash(name.name, name.len);
1584 down_read(&fc->killsb);
1587 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1588 outarg.child, &name);
1589 up_read(&fc->killsb);
1595 fuse_copy_finish(cs);
1599 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1600 struct fuse_copy_state *cs)
1602 struct fuse_notify_store_out outarg;
1603 struct inode *inode;
1604 struct address_space *mapping;
1608 unsigned int offset;
1614 if (size < sizeof(outarg))
1617 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1622 if (size - sizeof(outarg) != outarg.size)
1625 nodeid = outarg.nodeid;
1627 down_read(&fc->killsb);
1633 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1637 mapping = inode->i_mapping;
1638 index = outarg.offset >> PAGE_CACHE_SHIFT;
1639 offset = outarg.offset & ~PAGE_CACHE_MASK;
1640 file_size = i_size_read(inode);
1641 end = outarg.offset + outarg.size;
1642 if (end > file_size) {
1644 fuse_write_update_size(inode, file_size);
1650 unsigned int this_num;
1653 page = find_or_create_page(mapping, index,
1654 mapping_gfp_mask(mapping));
1658 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1659 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1660 if (!err && offset == 0 &&
1661 (this_num == PAGE_CACHE_SIZE || file_size == end))
1662 SetPageUptodate(page);
1664 page_cache_release(page);
1679 up_read(&fc->killsb);
1681 fuse_copy_finish(cs);
1685 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1687 release_pages(req->pages, req->num_pages, false);
1690 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1691 struct fuse_notify_retrieve_out *outarg)
1694 struct address_space *mapping = inode->i_mapping;
1695 struct fuse_req *req;
1699 unsigned int offset;
1700 size_t total_len = 0;
1703 offset = outarg->offset & ~PAGE_CACHE_MASK;
1704 file_size = i_size_read(inode);
1707 if (outarg->offset > file_size)
1709 else if (outarg->offset + num > file_size)
1710 num = file_size - outarg->offset;
1712 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1713 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1715 req = fuse_get_req(fc, num_pages);
1717 return PTR_ERR(req);
1719 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1720 req->in.h.nodeid = outarg->nodeid;
1721 req->in.numargs = 2;
1722 req->in.argpages = 1;
1723 req->page_descs[0].offset = offset;
1724 req->end = fuse_retrieve_end;
1726 index = outarg->offset >> PAGE_CACHE_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_CACHE_SIZE - offset);
1737 req->pages[req->num_pages] = page;
1738 req->page_descs[req->num_pages].length = this_num;
1743 total_len += this_num;
1746 req->misc.retrieve_in.offset = outarg->offset;
1747 req->misc.retrieve_in.size = total_len;
1748 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1749 req->in.args[0].value = &req->misc.retrieve_in;
1750 req->in.args[1].size = total_len;
1752 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1754 fuse_retrieve_end(fc, req);
1759 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1760 struct fuse_copy_state *cs)
1762 struct fuse_notify_retrieve_out outarg;
1763 struct inode *inode;
1767 if (size != sizeof(outarg))
1770 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1774 fuse_copy_finish(cs);
1776 down_read(&fc->killsb);
1779 u64 nodeid = outarg.nodeid;
1781 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1783 err = fuse_retrieve(fc, inode, &outarg);
1787 up_read(&fc->killsb);
1792 fuse_copy_finish(cs);
1796 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1797 unsigned int size, struct fuse_copy_state *cs)
1799 /* Don't try to move pages (yet) */
1803 case FUSE_NOTIFY_POLL:
1804 return fuse_notify_poll(fc, size, cs);
1806 case FUSE_NOTIFY_INVAL_INODE:
1807 return fuse_notify_inval_inode(fc, size, cs);
1809 case FUSE_NOTIFY_INVAL_ENTRY:
1810 return fuse_notify_inval_entry(fc, size, cs);
1812 case FUSE_NOTIFY_STORE:
1813 return fuse_notify_store(fc, size, cs);
1815 case FUSE_NOTIFY_RETRIEVE:
1816 return fuse_notify_retrieve(fc, size, cs);
1818 case FUSE_NOTIFY_DELETE:
1819 return fuse_notify_delete(fc, size, cs);
1822 fuse_copy_finish(cs);
1827 /* Look up request on processing list by unique ID */
1828 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1830 struct fuse_req *req;
1832 list_for_each_entry(req, &fpq->processing, list) {
1833 if (req->in.h.unique == unique || req->intr_unique == unique)
1839 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1842 unsigned reqsize = sizeof(struct fuse_out_header);
1845 return nbytes != reqsize ? -EINVAL : 0;
1847 reqsize += len_args(out->numargs, out->args);
1849 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1851 else if (reqsize > nbytes) {
1852 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1853 unsigned diffsize = reqsize - nbytes;
1854 if (diffsize > lastarg->size)
1856 lastarg->size -= diffsize;
1858 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1863 * Write a single reply to a request. First the header is copied from
1864 * the write buffer. The request is then searched on the processing
1865 * list by the unique ID found in the header. If found, then remove
1866 * it from the list and copy the rest of the buffer to the request.
1867 * The request is finished by calling request_end()
1869 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1870 struct fuse_copy_state *cs, size_t nbytes)
1873 struct fuse_conn *fc = fud->fc;
1874 struct fuse_pqueue *fpq = &fud->pq;
1875 struct fuse_req *req;
1876 struct fuse_out_header oh;
1878 if (nbytes < sizeof(struct fuse_out_header))
1881 err = fuse_copy_one(cs, &oh, sizeof(oh));
1886 if (oh.len != nbytes)
1890 * Zero oh.unique indicates unsolicited notification message
1891 * and error contains notification code.
1894 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1895 return err ? err : nbytes;
1899 if (oh.error <= -1000 || oh.error > 0)
1902 spin_lock(&fpq->lock);
1904 if (!fpq->connected)
1907 req = request_find(fpq, oh.unique);
1911 /* Is it an interrupt reply? */
1912 if (req->intr_unique == oh.unique) {
1913 spin_unlock(&fpq->lock);
1916 if (nbytes != sizeof(struct fuse_out_header))
1919 if (oh.error == -ENOSYS)
1920 fc->no_interrupt = 1;
1921 else if (oh.error == -EAGAIN)
1922 queue_interrupt(&fc->iq, req);
1924 fuse_copy_finish(cs);
1928 clear_bit(FR_SENT, &req->flags);
1929 list_move(&req->list, &fpq->io);
1931 set_bit(FR_LOCKED, &req->flags);
1932 spin_unlock(&fpq->lock);
1934 if (!req->out.page_replace)
1937 err = copy_out_args(cs, &req->out, nbytes);
1938 fuse_copy_finish(cs);
1940 spin_lock(&fpq->lock);
1941 clear_bit(FR_LOCKED, &req->flags);
1942 if (!fpq->connected)
1945 req->out.h.error = -EIO;
1946 if (!test_bit(FR_PRIVATE, &req->flags))
1947 list_del_init(&req->list);
1948 spin_unlock(&fpq->lock);
1950 request_end(fc, req);
1952 return err ? err : nbytes;
1955 spin_unlock(&fpq->lock);
1957 fuse_copy_finish(cs);
1961 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1963 struct fuse_copy_state cs;
1964 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1969 if (!iter_is_iovec(from))
1972 fuse_copy_init(&cs, 0, from);
1974 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1977 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1978 struct file *out, loff_t *ppos,
1979 size_t len, unsigned int flags)
1983 struct pipe_buffer *bufs;
1984 struct fuse_copy_state cs;
1985 struct fuse_dev *fud;
1989 fud = fuse_get_dev(out);
1993 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
2000 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
2001 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
2011 struct pipe_buffer *ibuf;
2012 struct pipe_buffer *obuf;
2014 BUG_ON(nbuf >= pipe->buffers);
2015 BUG_ON(!pipe->nrbufs);
2016 ibuf = &pipe->bufs[pipe->curbuf];
2019 if (rem >= ibuf->len) {
2022 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2025 ibuf->ops->get(pipe, ibuf);
2027 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2029 ibuf->offset += obuf->len;
2030 ibuf->len -= obuf->len;
2037 fuse_copy_init(&cs, 0, NULL);
2042 if (flags & SPLICE_F_MOVE)
2045 ret = fuse_dev_do_write(fud, &cs, len);
2047 for (idx = 0; idx < nbuf; idx++) {
2048 struct pipe_buffer *buf = &bufs[idx];
2049 buf->ops->release(pipe, buf);
2056 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2058 unsigned mask = POLLOUT | POLLWRNORM;
2059 struct fuse_iqueue *fiq;
2060 struct fuse_dev *fud = fuse_get_dev(file);
2066 poll_wait(file, &fiq->waitq, wait);
2068 spin_lock(&fiq->waitq.lock);
2069 if (!fiq->connected)
2071 else if (request_pending(fiq))
2072 mask |= POLLIN | POLLRDNORM;
2073 spin_unlock(&fiq->waitq.lock);
2079 * Abort all requests on the given list (pending or processing)
2081 * This function releases and reacquires fc->lock
2083 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2085 while (!list_empty(head)) {
2086 struct fuse_req *req;
2087 req = list_entry(head->next, struct fuse_req, list);
2088 req->out.h.error = -ECONNABORTED;
2089 clear_bit(FR_PENDING, &req->flags);
2090 clear_bit(FR_SENT, &req->flags);
2091 list_del_init(&req->list);
2092 request_end(fc, req);
2096 static void end_polls(struct fuse_conn *fc)
2100 p = rb_first(&fc->polled_files);
2103 struct fuse_file *ff;
2104 ff = rb_entry(p, struct fuse_file, polled_node);
2105 wake_up_interruptible_all(&ff->poll_wait);
2112 * Abort all requests.
2114 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2117 * The same effect is usually achievable through killing the filesystem daemon
2118 * and all users of the filesystem. The exception is the combination of an
2119 * asynchronous request and the tricky deadlock (see
2120 * Documentation/filesystems/fuse.txt).
2122 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2123 * requests, they should be finished off immediately. Locked requests will be
2124 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2125 * requests. It is possible that some request will finish before we can. This
2126 * is OK, the request will in that case be removed from the list before we touch
2129 void fuse_abort_conn(struct fuse_conn *fc)
2131 struct fuse_iqueue *fiq = &fc->iq;
2133 spin_lock(&fc->lock);
2134 if (fc->connected) {
2135 struct fuse_dev *fud;
2136 struct fuse_req *req, *next;
2142 fuse_set_initialized(fc);
2143 list_for_each_entry(fud, &fc->devices, entry) {
2144 struct fuse_pqueue *fpq = &fud->pq;
2146 spin_lock(&fpq->lock);
2148 list_for_each_entry_safe(req, next, &fpq->io, list) {
2149 req->out.h.error = -ECONNABORTED;
2150 spin_lock(&req->waitq.lock);
2151 set_bit(FR_ABORTED, &req->flags);
2152 if (!test_bit(FR_LOCKED, &req->flags)) {
2153 set_bit(FR_PRIVATE, &req->flags);
2154 list_move(&req->list, &to_end1);
2156 spin_unlock(&req->waitq.lock);
2158 list_splice_init(&fpq->processing, &to_end2);
2159 spin_unlock(&fpq->lock);
2161 fc->max_background = UINT_MAX;
2164 spin_lock(&fiq->waitq.lock);
2166 list_splice_init(&fiq->pending, &to_end2);
2167 while (forget_pending(fiq))
2168 kfree(dequeue_forget(fiq, 1, NULL));
2169 wake_up_all_locked(&fiq->waitq);
2170 spin_unlock(&fiq->waitq.lock);
2171 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2173 wake_up_all(&fc->blocked_waitq);
2174 spin_unlock(&fc->lock);
2176 while (!list_empty(&to_end1)) {
2177 req = list_first_entry(&to_end1, struct fuse_req, list);
2178 __fuse_get_request(req);
2179 list_del_init(&req->list);
2180 request_end(fc, req);
2182 end_requests(fc, &to_end2);
2184 spin_unlock(&fc->lock);
2187 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2189 int fuse_dev_release(struct inode *inode, struct file *file)
2191 struct fuse_dev *fud = fuse_get_dev(file);
2194 struct fuse_conn *fc = fud->fc;
2195 struct fuse_pqueue *fpq = &fud->pq;
2197 WARN_ON(!list_empty(&fpq->io));
2198 end_requests(fc, &fpq->processing);
2199 /* Are we the last open device? */
2200 if (atomic_dec_and_test(&fc->dev_count)) {
2201 WARN_ON(fc->iq.fasync != NULL);
2202 fuse_abort_conn(fc);
2208 EXPORT_SYMBOL_GPL(fuse_dev_release);
2210 static int fuse_dev_fasync(int fd, struct file *file, int on)
2212 struct fuse_dev *fud = fuse_get_dev(file);
2217 /* No locking - fasync_helper does its own locking */
2218 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2221 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2223 struct fuse_dev *fud;
2225 if (new->private_data)
2228 fud = fuse_dev_alloc(fc);
2232 new->private_data = fud;
2233 atomic_inc(&fc->dev_count);
2238 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2243 if (cmd == FUSE_DEV_IOC_CLONE) {
2247 if (!get_user(oldfd, (__u32 __user *) arg)) {
2248 struct file *old = fget(oldfd);
2252 struct fuse_dev *fud = NULL;
2255 * Check against file->f_op because CUSE
2256 * uses the same ioctl handler.
2258 if (old->f_op == file->f_op &&
2259 old->f_cred->user_ns == file->f_cred->user_ns)
2260 fud = fuse_get_dev(old);
2263 mutex_lock(&fuse_mutex);
2264 err = fuse_device_clone(fud->fc, file);
2265 mutex_unlock(&fuse_mutex);
2274 const struct file_operations fuse_dev_operations = {
2275 .owner = THIS_MODULE,
2276 .open = fuse_dev_open,
2277 .llseek = no_llseek,
2278 .read_iter = fuse_dev_read,
2279 .splice_read = fuse_dev_splice_read,
2280 .write_iter = fuse_dev_write,
2281 .splice_write = fuse_dev_splice_write,
2282 .poll = fuse_dev_poll,
2283 .release = fuse_dev_release,
2284 .fasync = fuse_dev_fasync,
2285 .unlocked_ioctl = fuse_dev_ioctl,
2286 .compat_ioctl = fuse_dev_ioctl,
2288 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2290 static struct miscdevice fuse_miscdevice = {
2291 .minor = FUSE_MINOR,
2293 .fops = &fuse_dev_operations,
2296 int __init fuse_dev_init(void)
2299 fuse_req_cachep = kmem_cache_create("fuse_request",
2300 sizeof(struct fuse_req),
2302 if (!fuse_req_cachep)
2305 err = misc_register(&fuse_miscdevice);
2307 goto out_cache_clean;
2312 kmem_cache_destroy(fuse_req_cachep);
2317 void fuse_dev_cleanup(void)
2319 misc_deregister(&fuse_miscdevice);
2320 kmem_cache_destroy(fuse_req_cachep);