2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/sched/signal.h>
15 #include <linux/uio.h>
16 #include <linux/miscdevice.h>
17 #include <linux/pagemap.h>
18 #include <linux/file.h>
19 #include <linux/slab.h>
20 #include <linux/pipe_fs_i.h>
21 #include <linux/swap.h>
22 #include <linux/splice.h>
23 #include <linux/sched.h>
25 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26 MODULE_ALIAS("devname:fuse");
28 /* Ordinary requests have even IDs, while interrupts IDs are odd */
29 #define FUSE_INT_REQ_BIT (1ULL << 0)
30 #define FUSE_REQ_ID_STEP (1ULL << 1)
32 static struct kmem_cache *fuse_req_cachep;
34 static struct fuse_dev *fuse_get_dev(struct file *file)
37 * Lockless access is OK, because file->private data is set
38 * once during mount and is valid until the file is released.
40 return READ_ONCE(file->private_data);
43 static void fuse_request_init(struct fuse_req *req, struct page **pages,
44 struct fuse_page_desc *page_descs,
47 memset(req, 0, sizeof(*req));
48 memset(pages, 0, sizeof(*pages) * npages);
49 memset(page_descs, 0, sizeof(*page_descs) * npages);
50 INIT_LIST_HEAD(&req->list);
51 INIT_LIST_HEAD(&req->intr_entry);
52 init_waitqueue_head(&req->waitq);
53 refcount_set(&req->count, 1);
55 req->page_descs = page_descs;
56 req->max_pages = npages;
57 __set_bit(FR_PENDING, &req->flags);
60 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
62 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
65 struct fuse_page_desc *page_descs;
67 if (npages <= FUSE_REQ_INLINE_PAGES) {
68 pages = req->inline_pages;
69 page_descs = req->inline_page_descs;
71 pages = kmalloc_array(npages, sizeof(struct page *),
75 sizeof(struct fuse_page_desc),
79 if (!pages || !page_descs) {
82 kmem_cache_free(fuse_req_cachep, req);
86 fuse_request_init(req, pages, page_descs, npages);
91 struct fuse_req *fuse_request_alloc(unsigned npages)
93 return __fuse_request_alloc(npages, GFP_KERNEL);
95 EXPORT_SYMBOL_GPL(fuse_request_alloc);
97 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
99 return __fuse_request_alloc(npages, GFP_NOFS);
102 void fuse_request_free(struct fuse_req *req)
104 if (req->pages != req->inline_pages) {
106 kfree(req->page_descs);
108 kmem_cache_free(fuse_req_cachep, req);
111 void __fuse_get_request(struct fuse_req *req)
113 refcount_inc(&req->count);
116 /* Must be called with > 1 refcount */
117 static void __fuse_put_request(struct fuse_req *req)
119 refcount_dec(&req->count);
122 void fuse_set_initialized(struct fuse_conn *fc)
124 /* Make sure stores before this are seen on another CPU */
129 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
131 return !fc->initialized || (for_background && fc->blocked);
134 static void fuse_drop_waiting(struct fuse_conn *fc)
137 atomic_dec(&fc->num_waiting);
138 } else if (atomic_dec_and_test(&fc->num_waiting)) {
139 /* wake up aborters */
140 wake_up_all(&fc->blocked_waitq);
144 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
147 struct fuse_req *req;
149 atomic_inc(&fc->num_waiting);
151 if (fuse_block_alloc(fc, for_background)) {
153 if (wait_event_killable_exclusive(fc->blocked_waitq,
154 !fuse_block_alloc(fc, for_background)))
157 /* Matches smp_wmb() in fuse_set_initialized() */
168 req = fuse_request_alloc(npages);
172 wake_up(&fc->blocked_waitq);
176 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid());
177 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid());
178 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
180 __set_bit(FR_WAITING, &req->flags);
182 __set_bit(FR_BACKGROUND, &req->flags);
184 if (unlikely(req->in.h.uid == ((uid_t)-1) ||
185 req->in.h.gid == ((gid_t)-1))) {
186 fuse_put_request(fc, req);
187 return ERR_PTR(-EOVERFLOW);
192 fuse_drop_waiting(fc);
196 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
198 return __fuse_get_req(fc, npages, false);
200 EXPORT_SYMBOL_GPL(fuse_get_req);
202 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
205 return __fuse_get_req(fc, npages, true);
207 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
210 * Return request in fuse_file->reserved_req. However that may
211 * currently be in use. If that is the case, wait for it to become
214 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
217 struct fuse_req *req = NULL;
218 struct fuse_file *ff = file->private_data;
221 wait_event(fc->reserved_req_waitq, ff->reserved_req);
222 spin_lock(&fc->lock);
223 if (ff->reserved_req) {
224 req = ff->reserved_req;
225 ff->reserved_req = NULL;
226 req->stolen_file = get_file(file);
228 spin_unlock(&fc->lock);
235 * Put stolen request back into fuse_file->reserved_req
237 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
239 struct file *file = req->stolen_file;
240 struct fuse_file *ff = file->private_data;
242 spin_lock(&fc->lock);
243 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
244 BUG_ON(ff->reserved_req);
245 ff->reserved_req = req;
246 wake_up_all(&fc->reserved_req_waitq);
247 spin_unlock(&fc->lock);
252 * Gets a requests for a file operation, always succeeds
254 * This is used for sending the FLUSH request, which must get to
255 * userspace, due to POSIX locks which may need to be unlocked.
257 * If allocation fails due to OOM, use the reserved request in
260 * This is very unlikely to deadlock accidentally, since the
261 * filesystem should not have it's own file open. If deadlock is
262 * intentional, it can still be broken by "aborting" the filesystem.
264 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
267 struct fuse_req *req;
269 atomic_inc(&fc->num_waiting);
270 wait_event(fc->blocked_waitq, fc->initialized);
271 /* Matches smp_wmb() in fuse_set_initialized() */
273 req = fuse_request_alloc(0);
275 req = get_reserved_req(fc, file);
277 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid());
278 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid());
279 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
281 __set_bit(FR_WAITING, &req->flags);
282 __clear_bit(FR_BACKGROUND, &req->flags);
286 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
288 if (refcount_dec_and_test(&req->count)) {
289 if (test_bit(FR_BACKGROUND, &req->flags)) {
291 * We get here in the unlikely case that a background
292 * request was allocated but not sent
294 spin_lock(&fc->bg_lock);
296 wake_up(&fc->blocked_waitq);
297 spin_unlock(&fc->bg_lock);
300 if (test_bit(FR_WAITING, &req->flags)) {
301 __clear_bit(FR_WAITING, &req->flags);
302 fuse_drop_waiting(fc);
305 if (req->stolen_file)
306 put_reserved_req(fc, req);
308 fuse_request_free(req);
311 EXPORT_SYMBOL_GPL(fuse_put_request);
313 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
318 for (i = 0; i < numargs; i++)
319 nbytes += args[i].size;
324 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
326 fiq->reqctr += FUSE_REQ_ID_STEP;
330 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
332 req->in.h.len = sizeof(struct fuse_in_header) +
333 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
334 list_add_tail(&req->list, &fiq->pending);
335 wake_up_locked(&fiq->waitq);
336 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
339 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
340 u64 nodeid, u64 nlookup)
342 struct fuse_iqueue *fiq = &fc->iq;
344 forget->forget_one.nodeid = nodeid;
345 forget->forget_one.nlookup = nlookup;
347 spin_lock(&fiq->waitq.lock);
348 if (fiq->connected) {
349 fiq->forget_list_tail->next = forget;
350 fiq->forget_list_tail = forget;
351 wake_up_locked(&fiq->waitq);
352 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
356 spin_unlock(&fiq->waitq.lock);
359 static void flush_bg_queue(struct fuse_conn *fc)
361 struct fuse_iqueue *fiq = &fc->iq;
363 while (fc->active_background < fc->max_background &&
364 !list_empty(&fc->bg_queue)) {
365 struct fuse_req *req;
367 req = list_first_entry(&fc->bg_queue, struct fuse_req, list);
368 list_del(&req->list);
369 fc->active_background++;
370 spin_lock(&fiq->waitq.lock);
371 req->in.h.unique = fuse_get_unique(fiq);
372 queue_request(fiq, req);
373 spin_unlock(&fiq->waitq.lock);
378 * This function is called when a request is finished. Either a reply
379 * has arrived or it was aborted (and not yet sent) or some error
380 * occurred during communication with userspace, or the device file
381 * was closed. The requester thread is woken up (if still waiting),
382 * the 'end' callback is called if given, else the reference to the
383 * request is released
385 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
387 struct fuse_iqueue *fiq = &fc->iq;
389 if (test_and_set_bit(FR_FINISHED, &req->flags))
392 spin_lock(&fiq->waitq.lock);
393 list_del_init(&req->intr_entry);
394 spin_unlock(&fiq->waitq.lock);
395 WARN_ON(test_bit(FR_PENDING, &req->flags));
396 WARN_ON(test_bit(FR_SENT, &req->flags));
397 if (test_bit(FR_BACKGROUND, &req->flags)) {
398 spin_lock(&fc->bg_lock);
399 clear_bit(FR_BACKGROUND, &req->flags);
400 if (fc->num_background == fc->max_background) {
402 wake_up(&fc->blocked_waitq);
403 } else if (!fc->blocked) {
405 * Wake up next waiter, if any. It's okay to use
406 * waitqueue_active(), as we've already synced up
407 * fc->blocked with waiters with the wake_up() call
410 if (waitqueue_active(&fc->blocked_waitq))
411 wake_up(&fc->blocked_waitq);
414 if (fc->num_background == fc->congestion_threshold && fc->sb) {
415 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
416 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
418 fc->num_background--;
419 fc->active_background--;
421 spin_unlock(&fc->bg_lock);
423 wake_up(&req->waitq);
427 fuse_put_request(fc, req);
430 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
432 spin_lock(&fiq->waitq.lock);
433 if (test_bit(FR_FINISHED, &req->flags)) {
434 spin_unlock(&fiq->waitq.lock);
437 if (list_empty(&req->intr_entry)) {
438 list_add_tail(&req->intr_entry, &fiq->interrupts);
439 wake_up_locked(&fiq->waitq);
441 spin_unlock(&fiq->waitq.lock);
442 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
445 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
447 struct fuse_iqueue *fiq = &fc->iq;
450 if (!fc->no_interrupt) {
451 /* Any signal may interrupt this */
452 err = wait_event_interruptible(req->waitq,
453 test_bit(FR_FINISHED, &req->flags));
457 set_bit(FR_INTERRUPTED, &req->flags);
458 /* matches barrier in fuse_dev_do_read() */
459 smp_mb__after_atomic();
460 if (test_bit(FR_SENT, &req->flags))
461 queue_interrupt(fiq, req);
464 if (!test_bit(FR_FORCE, &req->flags)) {
465 /* Only fatal signals may interrupt this */
466 err = wait_event_killable(req->waitq,
467 test_bit(FR_FINISHED, &req->flags));
471 spin_lock(&fiq->waitq.lock);
472 /* Request is not yet in userspace, bail out */
473 if (test_bit(FR_PENDING, &req->flags)) {
474 list_del(&req->list);
475 spin_unlock(&fiq->waitq.lock);
476 __fuse_put_request(req);
477 req->out.h.error = -EINTR;
480 spin_unlock(&fiq->waitq.lock);
484 * Either request is already in userspace, or it was forced.
487 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
490 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
492 struct fuse_iqueue *fiq = &fc->iq;
494 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
495 spin_lock(&fiq->waitq.lock);
496 if (!fiq->connected) {
497 spin_unlock(&fiq->waitq.lock);
498 req->out.h.error = -ENOTCONN;
500 req->in.h.unique = fuse_get_unique(fiq);
501 queue_request(fiq, req);
502 /* acquire extra reference, since request is still needed
503 after request_end() */
504 __fuse_get_request(req);
505 spin_unlock(&fiq->waitq.lock);
507 request_wait_answer(fc, req);
508 /* Pairs with smp_wmb() in request_end() */
513 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
515 __set_bit(FR_ISREPLY, &req->flags);
516 if (!test_bit(FR_WAITING, &req->flags)) {
517 __set_bit(FR_WAITING, &req->flags);
518 atomic_inc(&fc->num_waiting);
520 __fuse_request_send(fc, req);
522 EXPORT_SYMBOL_GPL(fuse_request_send);
524 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
526 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
527 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
530 switch (args->in.h.opcode) {
537 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
541 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
545 if (fc->minor < 12) {
546 switch (args->in.h.opcode) {
548 args->in.args[0].size = sizeof(struct fuse_open_in);
551 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
557 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
559 struct fuse_req *req;
562 req = fuse_get_req(fc, 0);
566 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
567 fuse_adjust_compat(fc, args);
569 req->in.h.opcode = args->in.h.opcode;
570 req->in.h.nodeid = args->in.h.nodeid;
571 req->in.numargs = args->in.numargs;
572 memcpy(req->in.args, args->in.args,
573 args->in.numargs * sizeof(struct fuse_in_arg));
574 req->out.argvar = args->out.argvar;
575 req->out.numargs = args->out.numargs;
576 memcpy(req->out.args, args->out.args,
577 args->out.numargs * sizeof(struct fuse_arg));
578 fuse_request_send(fc, req);
579 ret = req->out.h.error;
580 if (!ret && args->out.argvar) {
581 BUG_ON(args->out.numargs != 1);
582 ret = req->out.args[0].size;
584 fuse_put_request(fc, req);
589 bool fuse_request_queue_background(struct fuse_conn *fc, struct fuse_req *req)
593 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags));
594 if (!test_bit(FR_WAITING, &req->flags)) {
595 __set_bit(FR_WAITING, &req->flags);
596 atomic_inc(&fc->num_waiting);
598 __set_bit(FR_ISREPLY, &req->flags);
599 spin_lock(&fc->bg_lock);
600 if (likely(fc->connected)) {
601 fc->num_background++;
602 if (fc->num_background == fc->max_background)
604 if (fc->num_background == fc->congestion_threshold && fc->sb) {
605 set_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
606 set_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
608 list_add_tail(&req->list, &fc->bg_queue);
612 spin_unlock(&fc->bg_lock);
617 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
620 if (!fuse_request_queue_background(fc, req)) {
621 req->out.h.error = -ENOTCONN;
623 fuse_put_request(fc, req);
626 EXPORT_SYMBOL_GPL(fuse_request_send_background);
628 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
629 struct fuse_req *req, u64 unique)
632 struct fuse_iqueue *fiq = &fc->iq;
634 __clear_bit(FR_ISREPLY, &req->flags);
635 req->in.h.unique = unique;
636 spin_lock(&fiq->waitq.lock);
637 if (fiq->connected) {
638 queue_request(fiq, req);
641 spin_unlock(&fiq->waitq.lock);
646 void fuse_force_forget(struct file *file, u64 nodeid)
648 struct inode *inode = file_inode(file);
649 struct fuse_conn *fc = get_fuse_conn(inode);
650 struct fuse_req *req;
651 struct fuse_forget_in inarg;
653 memset(&inarg, 0, sizeof(inarg));
655 req = fuse_get_req_nofail_nopages(fc, file);
656 req->in.h.opcode = FUSE_FORGET;
657 req->in.h.nodeid = nodeid;
659 req->in.args[0].size = sizeof(inarg);
660 req->in.args[0].value = &inarg;
661 __clear_bit(FR_ISREPLY, &req->flags);
662 __fuse_request_send(fc, req);
664 fuse_put_request(fc, req);
668 * Lock the request. Up to the next unlock_request() there mustn't be
669 * anything that could cause a page-fault. If the request was already
672 static int lock_request(struct fuse_req *req)
676 spin_lock(&req->waitq.lock);
677 if (test_bit(FR_ABORTED, &req->flags))
680 set_bit(FR_LOCKED, &req->flags);
681 spin_unlock(&req->waitq.lock);
687 * Unlock request. If it was aborted while locked, caller is responsible
688 * for unlocking and ending the request.
690 static int unlock_request(struct fuse_req *req)
694 spin_lock(&req->waitq.lock);
695 if (test_bit(FR_ABORTED, &req->flags))
698 clear_bit(FR_LOCKED, &req->flags);
699 spin_unlock(&req->waitq.lock);
704 struct fuse_copy_state {
706 struct fuse_req *req;
707 struct iov_iter *iter;
708 struct pipe_buffer *pipebufs;
709 struct pipe_buffer *currbuf;
710 struct pipe_inode_info *pipe;
711 unsigned long nr_segs;
715 unsigned move_pages:1;
718 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
719 struct iov_iter *iter)
721 memset(cs, 0, sizeof(*cs));
726 /* Unmap and put previous page of userspace buffer */
727 static void fuse_copy_finish(struct fuse_copy_state *cs)
730 struct pipe_buffer *buf = cs->currbuf;
733 buf->len = PAGE_SIZE - cs->len;
737 flush_dcache_page(cs->pg);
738 set_page_dirty_lock(cs->pg);
746 * Get another pagefull of userspace buffer, and map it to kernel
747 * address space, and lock request
749 static int fuse_copy_fill(struct fuse_copy_state *cs)
754 err = unlock_request(cs->req);
758 fuse_copy_finish(cs);
760 struct pipe_buffer *buf = cs->pipebufs;
763 err = pipe_buf_confirm(cs->pipe, buf);
767 BUG_ON(!cs->nr_segs);
770 cs->offset = buf->offset;
775 if (cs->nr_segs == cs->pipe->buffers)
778 page = alloc_page(GFP_HIGHUSER);
795 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
802 iov_iter_advance(cs->iter, err);
805 return lock_request(cs->req);
808 /* Do as much copy to/from userspace buffer as we can */
809 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
811 unsigned ncpy = min(*size, cs->len);
813 void *pgaddr = kmap_atomic(cs->pg);
814 void *buf = pgaddr + cs->offset;
817 memcpy(buf, *val, ncpy);
819 memcpy(*val, buf, ncpy);
821 kunmap_atomic(pgaddr);
830 static int fuse_check_page(struct page *page)
832 if (page_mapcount(page) ||
833 page->mapping != NULL ||
834 page_count(page) != 1 ||
835 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
842 printk(KERN_WARNING "fuse: trying to steal weird page\n");
843 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);
849 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
852 struct page *oldpage = *pagep;
853 struct page *newpage;
854 struct pipe_buffer *buf = cs->pipebufs;
856 err = unlock_request(cs->req);
860 fuse_copy_finish(cs);
862 err = pipe_buf_confirm(cs->pipe, buf);
866 BUG_ON(!cs->nr_segs);
872 if (cs->len != PAGE_SIZE)
875 if (pipe_buf_steal(cs->pipe, buf) != 0)
880 if (!PageUptodate(newpage))
881 SetPageUptodate(newpage);
883 ClearPageMappedToDisk(newpage);
885 if (fuse_check_page(newpage) != 0)
886 goto out_fallback_unlock;
889 * This is a new and locked page, it shouldn't be mapped or
890 * have any special flags on it
892 if (WARN_ON(page_mapped(oldpage)))
893 goto out_fallback_unlock;
894 if (WARN_ON(page_has_private(oldpage)))
895 goto out_fallback_unlock;
896 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
897 goto out_fallback_unlock;
898 if (WARN_ON(PageMlocked(oldpage)))
899 goto out_fallback_unlock;
901 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
903 unlock_page(newpage);
909 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
910 lru_cache_add_file(newpage);
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);
933 unlock_page(newpage);
936 cs->offset = buf->offset;
938 err = lock_request(cs->req);
945 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
946 unsigned offset, unsigned count)
948 struct pipe_buffer *buf;
951 if (cs->nr_segs == cs->pipe->buffers)
954 err = unlock_request(cs->req);
958 fuse_copy_finish(cs);
963 buf->offset = offset;
974 * Copy a page in the request to/from the userspace buffer. Must be
977 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
978 unsigned offset, unsigned count, int zeroing)
981 struct page *page = *pagep;
983 if (page && zeroing && count < PAGE_SIZE)
984 clear_highpage(page);
987 if (cs->write && cs->pipebufs && page) {
988 return fuse_ref_page(cs, page, offset, count);
989 } else if (!cs->len) {
990 if (cs->move_pages && page &&
991 offset == 0 && count == PAGE_SIZE) {
992 err = fuse_try_move_page(cs, pagep);
996 err = fuse_copy_fill(cs);
1002 void *mapaddr = kmap_atomic(page);
1003 void *buf = mapaddr + offset;
1004 offset += fuse_copy_do(cs, &buf, &count);
1005 kunmap_atomic(mapaddr);
1007 offset += fuse_copy_do(cs, NULL, &count);
1009 if (page && !cs->write)
1010 flush_dcache_page(page);
1014 /* Copy pages in the request to/from userspace buffer */
1015 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1019 struct fuse_req *req = cs->req;
1021 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1023 unsigned offset = req->page_descs[i].offset;
1024 unsigned count = min(nbytes, req->page_descs[i].length);
1026 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1036 /* Copy a single argument in the request to/from userspace buffer */
1037 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1041 int err = fuse_copy_fill(cs);
1045 fuse_copy_do(cs, &val, &size);
1050 /* Copy request arguments to/from userspace buffer */
1051 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1052 unsigned argpages, struct fuse_arg *args,
1058 for (i = 0; !err && i < numargs; i++) {
1059 struct fuse_arg *arg = &args[i];
1060 if (i == numargs - 1 && argpages)
1061 err = fuse_copy_pages(cs, arg->size, zeroing);
1063 err = fuse_copy_one(cs, arg->value, arg->size);
1068 static int forget_pending(struct fuse_iqueue *fiq)
1070 return fiq->forget_list_head.next != NULL;
1073 static int request_pending(struct fuse_iqueue *fiq)
1075 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1076 forget_pending(fiq);
1080 * Transfer an interrupt request to userspace
1082 * Unlike other requests this is assembled on demand, without a need
1083 * to allocate a separate fuse_req structure.
1085 * Called with fiq->waitq.lock held, releases it
1087 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1088 struct fuse_copy_state *cs,
1089 size_t nbytes, struct fuse_req *req)
1090 __releases(fiq->waitq.lock)
1092 struct fuse_in_header ih;
1093 struct fuse_interrupt_in arg;
1094 unsigned reqsize = sizeof(ih) + sizeof(arg);
1097 list_del_init(&req->intr_entry);
1098 req->intr_unique = (req->in.h.unique | FUSE_INT_REQ_BIT);
1099 memset(&ih, 0, sizeof(ih));
1100 memset(&arg, 0, sizeof(arg));
1102 ih.opcode = FUSE_INTERRUPT;
1103 ih.unique = req->intr_unique;
1104 arg.unique = req->in.h.unique;
1106 spin_unlock(&fiq->waitq.lock);
1107 if (nbytes < reqsize)
1110 err = fuse_copy_one(cs, &ih, sizeof(ih));
1112 err = fuse_copy_one(cs, &arg, sizeof(arg));
1113 fuse_copy_finish(cs);
1115 return err ? err : reqsize;
1118 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1122 struct fuse_forget_link *head = fiq->forget_list_head.next;
1123 struct fuse_forget_link **newhead = &head;
1126 for (count = 0; *newhead != NULL && count < max; count++)
1127 newhead = &(*newhead)->next;
1129 fiq->forget_list_head.next = *newhead;
1131 if (fiq->forget_list_head.next == NULL)
1132 fiq->forget_list_tail = &fiq->forget_list_head;
1140 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1141 struct fuse_copy_state *cs,
1143 __releases(fiq->waitq.lock)
1146 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1147 struct fuse_forget_in arg = {
1148 .nlookup = forget->forget_one.nlookup,
1150 struct fuse_in_header ih = {
1151 .opcode = FUSE_FORGET,
1152 .nodeid = forget->forget_one.nodeid,
1153 .unique = fuse_get_unique(fiq),
1154 .len = sizeof(ih) + sizeof(arg),
1157 spin_unlock(&fiq->waitq.lock);
1159 if (nbytes < ih.len)
1162 err = fuse_copy_one(cs, &ih, sizeof(ih));
1164 err = fuse_copy_one(cs, &arg, sizeof(arg));
1165 fuse_copy_finish(cs);
1173 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1174 struct fuse_copy_state *cs, size_t nbytes)
1175 __releases(fiq->waitq.lock)
1178 unsigned max_forgets;
1180 struct fuse_forget_link *head;
1181 struct fuse_batch_forget_in arg = { .count = 0 };
1182 struct fuse_in_header ih = {
1183 .opcode = FUSE_BATCH_FORGET,
1184 .unique = fuse_get_unique(fiq),
1185 .len = sizeof(ih) + sizeof(arg),
1188 if (nbytes < ih.len) {
1189 spin_unlock(&fiq->waitq.lock);
1193 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1194 head = dequeue_forget(fiq, max_forgets, &count);
1195 spin_unlock(&fiq->waitq.lock);
1198 ih.len += count * sizeof(struct fuse_forget_one);
1199 err = fuse_copy_one(cs, &ih, sizeof(ih));
1201 err = fuse_copy_one(cs, &arg, sizeof(arg));
1204 struct fuse_forget_link *forget = head;
1207 err = fuse_copy_one(cs, &forget->forget_one,
1208 sizeof(forget->forget_one));
1210 head = forget->next;
1214 fuse_copy_finish(cs);
1222 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1223 struct fuse_copy_state *cs,
1225 __releases(fiq->waitq.lock)
1227 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1228 return fuse_read_single_forget(fiq, cs, nbytes);
1230 return fuse_read_batch_forget(fiq, cs, nbytes);
1234 * Read a single request into the userspace filesystem's buffer. This
1235 * function waits until a request is available, then removes it from
1236 * the pending list and copies request data to userspace buffer. If
1237 * no reply is needed (FORGET) or request has been aborted or there
1238 * was an error during the copying then it's finished by calling
1239 * request_end(). Otherwise add it to the processing list, and set
1242 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1243 struct fuse_copy_state *cs, size_t nbytes)
1246 struct fuse_conn *fc = fud->fc;
1247 struct fuse_iqueue *fiq = &fc->iq;
1248 struct fuse_pqueue *fpq = &fud->pq;
1249 struct fuse_req *req;
1254 spin_lock(&fiq->waitq.lock);
1256 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1257 !request_pending(fiq))
1260 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1261 !fiq->connected || request_pending(fiq));
1265 if (!fiq->connected) {
1266 err = (fc->aborted && fc->abort_err) ? -ECONNABORTED : -ENODEV;
1270 if (!list_empty(&fiq->interrupts)) {
1271 req = list_entry(fiq->interrupts.next, struct fuse_req,
1273 return fuse_read_interrupt(fiq, cs, nbytes, req);
1276 if (forget_pending(fiq)) {
1277 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1278 return fuse_read_forget(fc, fiq, cs, nbytes);
1280 if (fiq->forget_batch <= -8)
1281 fiq->forget_batch = 16;
1284 req = list_entry(fiq->pending.next, struct fuse_req, list);
1285 clear_bit(FR_PENDING, &req->flags);
1286 list_del_init(&req->list);
1287 spin_unlock(&fiq->waitq.lock);
1290 reqsize = in->h.len;
1292 /* If request is too large, reply with an error and restart the read */
1293 if (nbytes < reqsize) {
1294 req->out.h.error = -EIO;
1295 /* SETXATTR is special, since it may contain too large data */
1296 if (in->h.opcode == FUSE_SETXATTR)
1297 req->out.h.error = -E2BIG;
1298 request_end(fc, req);
1301 spin_lock(&fpq->lock);
1302 list_add(&req->list, &fpq->io);
1303 spin_unlock(&fpq->lock);
1305 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1307 err = fuse_copy_args(cs, in->numargs, in->argpages,
1308 (struct fuse_arg *) in->args, 0);
1309 fuse_copy_finish(cs);
1310 spin_lock(&fpq->lock);
1311 clear_bit(FR_LOCKED, &req->flags);
1312 if (!fpq->connected) {
1313 err = (fc->aborted && fc->abort_err) ? -ECONNABORTED : -ENODEV;
1317 req->out.h.error = -EIO;
1320 if (!test_bit(FR_ISREPLY, &req->flags)) {
1324 list_move_tail(&req->list, &fpq->processing);
1325 __fuse_get_request(req);
1326 set_bit(FR_SENT, &req->flags);
1327 spin_unlock(&fpq->lock);
1328 /* matches barrier in request_wait_answer() */
1329 smp_mb__after_atomic();
1330 if (test_bit(FR_INTERRUPTED, &req->flags))
1331 queue_interrupt(fiq, req);
1332 fuse_put_request(fc, req);
1337 if (!test_bit(FR_PRIVATE, &req->flags))
1338 list_del_init(&req->list);
1339 spin_unlock(&fpq->lock);
1340 request_end(fc, req);
1344 spin_unlock(&fiq->waitq.lock);
1348 static int fuse_dev_open(struct inode *inode, struct file *file)
1351 * The fuse device's file's private_data is used to hold
1352 * the fuse_conn(ection) when it is mounted, and is used to
1353 * keep track of whether the file has been mounted already.
1355 file->private_data = NULL;
1359 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1361 struct fuse_copy_state cs;
1362 struct file *file = iocb->ki_filp;
1363 struct fuse_dev *fud = fuse_get_dev(file);
1368 if (!iter_is_iovec(to))
1371 fuse_copy_init(&cs, 1, to);
1373 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1376 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1377 struct pipe_inode_info *pipe,
1378 size_t len, unsigned int flags)
1382 struct pipe_buffer *bufs;
1383 struct fuse_copy_state cs;
1384 struct fuse_dev *fud = fuse_get_dev(in);
1389 bufs = kvmalloc_array(pipe->buffers, sizeof(struct pipe_buffer),
1394 fuse_copy_init(&cs, 1, NULL);
1397 ret = fuse_dev_do_read(fud, in, &cs, len);
1401 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1406 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1408 * Need to be careful about this. Having buf->ops in module
1409 * code can Oops if the buffer persists after module unload.
1411 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1412 bufs[page_nr].flags = 0;
1413 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1414 if (unlikely(ret < 0))
1420 for (; page_nr < cs.nr_segs; page_nr++)
1421 put_page(bufs[page_nr].page);
1427 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1428 struct fuse_copy_state *cs)
1430 struct fuse_notify_poll_wakeup_out outarg;
1433 if (size != sizeof(outarg))
1436 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1440 fuse_copy_finish(cs);
1441 return fuse_notify_poll_wakeup(fc, &outarg);
1444 fuse_copy_finish(cs);
1448 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1449 struct fuse_copy_state *cs)
1451 struct fuse_notify_inval_inode_out outarg;
1454 if (size != sizeof(outarg))
1457 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1460 fuse_copy_finish(cs);
1462 down_read(&fc->killsb);
1465 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1466 outarg.off, outarg.len);
1468 up_read(&fc->killsb);
1472 fuse_copy_finish(cs);
1476 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1477 struct fuse_copy_state *cs)
1479 struct fuse_notify_inval_entry_out outarg;
1484 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1489 if (size < sizeof(outarg))
1492 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1496 err = -ENAMETOOLONG;
1497 if (outarg.namelen > FUSE_NAME_MAX)
1501 if (size != sizeof(outarg) + outarg.namelen + 1)
1505 name.len = outarg.namelen;
1506 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1509 fuse_copy_finish(cs);
1510 buf[outarg.namelen] = 0;
1512 down_read(&fc->killsb);
1515 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1516 up_read(&fc->killsb);
1522 fuse_copy_finish(cs);
1526 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1527 struct fuse_copy_state *cs)
1529 struct fuse_notify_delete_out outarg;
1534 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1539 if (size < sizeof(outarg))
1542 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1546 err = -ENAMETOOLONG;
1547 if (outarg.namelen > FUSE_NAME_MAX)
1551 if (size != sizeof(outarg) + outarg.namelen + 1)
1555 name.len = outarg.namelen;
1556 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1559 fuse_copy_finish(cs);
1560 buf[outarg.namelen] = 0;
1562 down_read(&fc->killsb);
1565 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1566 outarg.child, &name);
1567 up_read(&fc->killsb);
1573 fuse_copy_finish(cs);
1577 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1578 struct fuse_copy_state *cs)
1580 struct fuse_notify_store_out outarg;
1581 struct inode *inode;
1582 struct address_space *mapping;
1586 unsigned int offset;
1592 if (size < sizeof(outarg))
1595 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1600 if (size - sizeof(outarg) != outarg.size)
1603 nodeid = outarg.nodeid;
1605 down_read(&fc->killsb);
1611 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1615 mapping = inode->i_mapping;
1616 index = outarg.offset >> PAGE_SHIFT;
1617 offset = outarg.offset & ~PAGE_MASK;
1618 file_size = i_size_read(inode);
1619 end = outarg.offset + outarg.size;
1620 if (end > file_size) {
1622 fuse_write_update_size(inode, file_size);
1628 unsigned int this_num;
1631 page = find_or_create_page(mapping, index,
1632 mapping_gfp_mask(mapping));
1636 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1637 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1638 if (!err && offset == 0 &&
1639 (this_num == PAGE_SIZE || file_size == end))
1640 SetPageUptodate(page);
1657 up_read(&fc->killsb);
1659 fuse_copy_finish(cs);
1663 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1665 release_pages(req->pages, req->num_pages);
1668 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1669 struct fuse_notify_retrieve_out *outarg)
1672 struct address_space *mapping = inode->i_mapping;
1673 struct fuse_req *req;
1677 unsigned int offset;
1678 size_t total_len = 0;
1681 offset = outarg->offset & ~PAGE_MASK;
1682 file_size = i_size_read(inode);
1685 if (outarg->offset > file_size)
1687 else if (outarg->offset + num > file_size)
1688 num = file_size - outarg->offset;
1690 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1691 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1693 req = fuse_get_req(fc, num_pages);
1695 return PTR_ERR(req);
1697 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1698 req->in.h.nodeid = outarg->nodeid;
1699 req->in.numargs = 2;
1700 req->in.argpages = 1;
1701 req->page_descs[0].offset = offset;
1702 req->end = fuse_retrieve_end;
1704 index = outarg->offset >> PAGE_SHIFT;
1706 while (num && req->num_pages < num_pages) {
1708 unsigned int this_num;
1710 page = find_get_page(mapping, index);
1714 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1715 req->pages[req->num_pages] = page;
1716 req->page_descs[req->num_pages].length = this_num;
1721 total_len += this_num;
1724 req->misc.retrieve_in.offset = outarg->offset;
1725 req->misc.retrieve_in.size = total_len;
1726 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1727 req->in.args[0].value = &req->misc.retrieve_in;
1728 req->in.args[1].size = total_len;
1730 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1732 fuse_retrieve_end(fc, req);
1737 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1738 struct fuse_copy_state *cs)
1740 struct fuse_notify_retrieve_out outarg;
1741 struct inode *inode;
1745 if (size != sizeof(outarg))
1748 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1752 fuse_copy_finish(cs);
1754 down_read(&fc->killsb);
1757 u64 nodeid = outarg.nodeid;
1759 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1761 err = fuse_retrieve(fc, inode, &outarg);
1765 up_read(&fc->killsb);
1770 fuse_copy_finish(cs);
1774 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1775 unsigned int size, struct fuse_copy_state *cs)
1777 /* Don't try to move pages (yet) */
1781 case FUSE_NOTIFY_POLL:
1782 return fuse_notify_poll(fc, size, cs);
1784 case FUSE_NOTIFY_INVAL_INODE:
1785 return fuse_notify_inval_inode(fc, size, cs);
1787 case FUSE_NOTIFY_INVAL_ENTRY:
1788 return fuse_notify_inval_entry(fc, size, cs);
1790 case FUSE_NOTIFY_STORE:
1791 return fuse_notify_store(fc, size, cs);
1793 case FUSE_NOTIFY_RETRIEVE:
1794 return fuse_notify_retrieve(fc, size, cs);
1796 case FUSE_NOTIFY_DELETE:
1797 return fuse_notify_delete(fc, size, cs);
1800 fuse_copy_finish(cs);
1805 /* Look up request on processing list by unique ID */
1806 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1808 struct fuse_req *req;
1810 list_for_each_entry(req, &fpq->processing, list) {
1811 if (req->in.h.unique == unique || req->intr_unique == unique)
1817 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1820 unsigned reqsize = sizeof(struct fuse_out_header);
1823 return nbytes != reqsize ? -EINVAL : 0;
1825 reqsize += len_args(out->numargs, out->args);
1827 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1829 else if (reqsize > nbytes) {
1830 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1831 unsigned diffsize = reqsize - nbytes;
1832 if (diffsize > lastarg->size)
1834 lastarg->size -= diffsize;
1836 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1841 * Write a single reply to a request. First the header is copied from
1842 * the write buffer. The request is then searched on the processing
1843 * list by the unique ID found in the header. If found, then remove
1844 * it from the list and copy the rest of the buffer to the request.
1845 * The request is finished by calling request_end()
1847 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1848 struct fuse_copy_state *cs, size_t nbytes)
1851 struct fuse_conn *fc = fud->fc;
1852 struct fuse_pqueue *fpq = &fud->pq;
1853 struct fuse_req *req;
1854 struct fuse_out_header oh;
1856 if (nbytes < sizeof(struct fuse_out_header))
1859 err = fuse_copy_one(cs, &oh, sizeof(oh));
1864 if (oh.len != nbytes)
1868 * Zero oh.unique indicates unsolicited notification message
1869 * and error contains notification code.
1872 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1873 return err ? err : nbytes;
1877 if (oh.error <= -1000 || oh.error > 0)
1880 spin_lock(&fpq->lock);
1882 if (!fpq->connected)
1885 req = request_find(fpq, oh.unique);
1889 /* Is it an interrupt reply? */
1890 if (req->intr_unique == oh.unique) {
1891 __fuse_get_request(req);
1892 spin_unlock(&fpq->lock);
1895 if (nbytes != sizeof(struct fuse_out_header)) {
1896 fuse_put_request(fc, req);
1900 if (oh.error == -ENOSYS)
1901 fc->no_interrupt = 1;
1902 else if (oh.error == -EAGAIN)
1903 queue_interrupt(&fc->iq, req);
1904 fuse_put_request(fc, req);
1906 fuse_copy_finish(cs);
1910 clear_bit(FR_SENT, &req->flags);
1911 list_move(&req->list, &fpq->io);
1913 set_bit(FR_LOCKED, &req->flags);
1914 spin_unlock(&fpq->lock);
1916 if (!req->out.page_replace)
1919 err = copy_out_args(cs, &req->out, nbytes);
1920 fuse_copy_finish(cs);
1922 spin_lock(&fpq->lock);
1923 clear_bit(FR_LOCKED, &req->flags);
1924 if (!fpq->connected)
1927 req->out.h.error = -EIO;
1928 if (!test_bit(FR_PRIVATE, &req->flags))
1929 list_del_init(&req->list);
1930 spin_unlock(&fpq->lock);
1932 request_end(fc, req);
1934 return err ? err : nbytes;
1937 spin_unlock(&fpq->lock);
1939 fuse_copy_finish(cs);
1943 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1945 struct fuse_copy_state cs;
1946 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1951 if (!iter_is_iovec(from))
1954 fuse_copy_init(&cs, 0, from);
1956 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1959 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1960 struct file *out, loff_t *ppos,
1961 size_t len, unsigned int flags)
1965 struct pipe_buffer *bufs;
1966 struct fuse_copy_state cs;
1967 struct fuse_dev *fud;
1971 fud = fuse_get_dev(out);
1977 bufs = kvmalloc_array(pipe->nrbufs, sizeof(struct pipe_buffer),
1986 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1987 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1997 struct pipe_buffer *ibuf;
1998 struct pipe_buffer *obuf;
2000 BUG_ON(nbuf >= pipe->buffers);
2001 BUG_ON(!pipe->nrbufs);
2002 ibuf = &pipe->bufs[pipe->curbuf];
2005 if (rem >= ibuf->len) {
2008 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2011 pipe_buf_get(pipe, ibuf);
2013 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2015 ibuf->offset += obuf->len;
2016 ibuf->len -= obuf->len;
2023 fuse_copy_init(&cs, 0, NULL);
2028 if (flags & SPLICE_F_MOVE)
2031 ret = fuse_dev_do_write(fud, &cs, len);
2033 for (idx = 0; idx < nbuf; idx++)
2034 pipe_buf_release(pipe, &bufs[idx]);
2041 static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2043 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
2044 struct fuse_iqueue *fiq;
2045 struct fuse_dev *fud = fuse_get_dev(file);
2051 poll_wait(file, &fiq->waitq, wait);
2053 spin_lock(&fiq->waitq.lock);
2054 if (!fiq->connected)
2056 else if (request_pending(fiq))
2057 mask |= EPOLLIN | EPOLLRDNORM;
2058 spin_unlock(&fiq->waitq.lock);
2064 * Abort all requests on the given list (pending or processing)
2066 * This function releases and reacquires fc->lock
2068 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2070 while (!list_empty(head)) {
2071 struct fuse_req *req;
2072 req = list_entry(head->next, struct fuse_req, list);
2073 req->out.h.error = -ECONNABORTED;
2074 clear_bit(FR_SENT, &req->flags);
2075 list_del_init(&req->list);
2076 request_end(fc, req);
2080 static void end_polls(struct fuse_conn *fc)
2084 p = rb_first(&fc->polled_files);
2087 struct fuse_file *ff;
2088 ff = rb_entry(p, struct fuse_file, polled_node);
2089 wake_up_interruptible_all(&ff->poll_wait);
2096 * Abort all requests.
2098 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2101 * The same effect is usually achievable through killing the filesystem daemon
2102 * and all users of the filesystem. The exception is the combination of an
2103 * asynchronous request and the tricky deadlock (see
2104 * Documentation/filesystems/fuse.txt).
2106 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2107 * requests, they should be finished off immediately. Locked requests will be
2108 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2109 * requests. It is possible that some request will finish before we can. This
2110 * is OK, the request will in that case be removed from the list before we touch
2113 void fuse_abort_conn(struct fuse_conn *fc, bool is_abort)
2115 struct fuse_iqueue *fiq = &fc->iq;
2117 spin_lock(&fc->lock);
2118 if (fc->connected) {
2119 struct fuse_dev *fud;
2120 struct fuse_req *req, *next;
2123 /* Background queuing checks fc->connected under bg_lock */
2124 spin_lock(&fc->bg_lock);
2126 spin_unlock(&fc->bg_lock);
2128 fc->aborted = is_abort;
2129 fuse_set_initialized(fc);
2130 list_for_each_entry(fud, &fc->devices, entry) {
2131 struct fuse_pqueue *fpq = &fud->pq;
2133 spin_lock(&fpq->lock);
2135 list_for_each_entry_safe(req, next, &fpq->io, list) {
2136 req->out.h.error = -ECONNABORTED;
2137 spin_lock(&req->waitq.lock);
2138 set_bit(FR_ABORTED, &req->flags);
2139 if (!test_bit(FR_LOCKED, &req->flags)) {
2140 set_bit(FR_PRIVATE, &req->flags);
2141 __fuse_get_request(req);
2142 list_move(&req->list, &to_end);
2144 spin_unlock(&req->waitq.lock);
2146 list_splice_tail_init(&fpq->processing, &to_end);
2147 spin_unlock(&fpq->lock);
2149 spin_lock(&fc->bg_lock);
2151 fc->max_background = UINT_MAX;
2153 spin_unlock(&fc->bg_lock);
2155 spin_lock(&fiq->waitq.lock);
2157 list_for_each_entry(req, &fiq->pending, list)
2158 clear_bit(FR_PENDING, &req->flags);
2159 list_splice_tail_init(&fiq->pending, &to_end);
2160 while (forget_pending(fiq))
2161 kfree(dequeue_forget(fiq, 1, NULL));
2162 wake_up_all_locked(&fiq->waitq);
2163 spin_unlock(&fiq->waitq.lock);
2164 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2166 wake_up_all(&fc->blocked_waitq);
2167 spin_unlock(&fc->lock);
2169 end_requests(fc, &to_end);
2171 spin_unlock(&fc->lock);
2174 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2176 void fuse_wait_aborted(struct fuse_conn *fc)
2178 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2181 int fuse_dev_release(struct inode *inode, struct file *file)
2183 struct fuse_dev *fud = fuse_get_dev(file);
2186 struct fuse_conn *fc = fud->fc;
2187 struct fuse_pqueue *fpq = &fud->pq;
2190 spin_lock(&fpq->lock);
2191 WARN_ON(!list_empty(&fpq->io));
2192 list_splice_init(&fpq->processing, &to_end);
2193 spin_unlock(&fpq->lock);
2195 end_requests(fc, &to_end);
2197 /* Are we the last open device? */
2198 if (atomic_dec_and_test(&fc->dev_count)) {
2199 WARN_ON(fc->iq.fasync != NULL);
2200 fuse_abort_conn(fc, false);
2206 EXPORT_SYMBOL_GPL(fuse_dev_release);
2208 static int fuse_dev_fasync(int fd, struct file *file, int on)
2210 struct fuse_dev *fud = fuse_get_dev(file);
2215 /* No locking - fasync_helper does its own locking */
2216 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2219 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2221 struct fuse_dev *fud;
2223 if (new->private_data)
2226 fud = fuse_dev_alloc(fc);
2230 new->private_data = fud;
2231 atomic_inc(&fc->dev_count);
2236 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2241 if (cmd == FUSE_DEV_IOC_CLONE) {
2245 if (!get_user(oldfd, (__u32 __user *) arg)) {
2246 struct file *old = fget(oldfd);
2250 struct fuse_dev *fud = NULL;
2253 * Check against file->f_op because CUSE
2254 * uses the same ioctl handler.
2256 if (old->f_op == file->f_op &&
2257 old->f_cred->user_ns == file->f_cred->user_ns)
2258 fud = fuse_get_dev(old);
2261 mutex_lock(&fuse_mutex);
2262 err = fuse_device_clone(fud->fc, file);
2263 mutex_unlock(&fuse_mutex);
2272 const struct file_operations fuse_dev_operations = {
2273 .owner = THIS_MODULE,
2274 .open = fuse_dev_open,
2275 .llseek = no_llseek,
2276 .read_iter = fuse_dev_read,
2277 .splice_read = fuse_dev_splice_read,
2278 .write_iter = fuse_dev_write,
2279 .splice_write = fuse_dev_splice_write,
2280 .poll = fuse_dev_poll,
2281 .release = fuse_dev_release,
2282 .fasync = fuse_dev_fasync,
2283 .unlocked_ioctl = fuse_dev_ioctl,
2284 .compat_ioctl = fuse_dev_ioctl,
2286 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2288 static struct miscdevice fuse_miscdevice = {
2289 .minor = FUSE_MINOR,
2291 .fops = &fuse_dev_operations,
2294 int __init fuse_dev_init(void)
2297 fuse_req_cachep = kmem_cache_create("fuse_request",
2298 sizeof(struct fuse_req),
2300 if (!fuse_req_cachep)
2303 err = misc_register(&fuse_miscdevice);
2305 goto out_cache_clean;
2310 kmem_cache_destroy(fuse_req_cachep);
2315 void fuse_dev_cleanup(void)
2317 misc_deregister(&fuse_miscdevice);
2318 kmem_cache_destroy(fuse_req_cachep);