Merge tag 'firewire-updates' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee139...
[linux/fpc-iii.git] / fs / fuse / dev.c
blobed19a7d622fa35decaa08b10e83b8bdee8712419
1 /*
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.
6 See the file COPYING.
7 */
9 #include "fuse_i.h"
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/aio.h>
24 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
25 MODULE_ALIAS("devname:fuse");
27 static struct kmem_cache *fuse_req_cachep;
29 static struct fuse_conn *fuse_get_conn(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 file->private_data;
38 static void fuse_request_init(struct fuse_req *req, struct page **pages,
39 struct fuse_page_desc *page_descs,
40 unsigned npages)
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);
49 req->pages = pages;
50 req->page_descs = page_descs;
51 req->max_pages = npages;
54 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
56 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
57 if (req) {
58 struct page **pages;
59 struct fuse_page_desc *page_descs;
61 if (npages <= FUSE_REQ_INLINE_PAGES) {
62 pages = req->inline_pages;
63 page_descs = req->inline_page_descs;
64 } else {
65 pages = kmalloc(sizeof(struct page *) * npages, flags);
66 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
67 npages, flags);
70 if (!pages || !page_descs) {
71 kfree(pages);
72 kfree(page_descs);
73 kmem_cache_free(fuse_req_cachep, req);
74 return NULL;
77 fuse_request_init(req, pages, page_descs, npages);
79 return req;
82 struct fuse_req *fuse_request_alloc(unsigned npages)
84 return __fuse_request_alloc(npages, GFP_KERNEL);
86 EXPORT_SYMBOL_GPL(fuse_request_alloc);
88 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
90 return __fuse_request_alloc(npages, GFP_NOFS);
93 void fuse_request_free(struct fuse_req *req)
95 if (req->pages != req->inline_pages) {
96 kfree(req->pages);
97 kfree(req->page_descs);
99 kmem_cache_free(fuse_req_cachep, req);
102 static void block_sigs(sigset_t *oldset)
104 sigset_t mask;
106 siginitsetinv(&mask, sigmask(SIGKILL));
107 sigprocmask(SIG_BLOCK, &mask, oldset);
110 static void restore_sigs(sigset_t *oldset)
112 sigprocmask(SIG_SETMASK, oldset, NULL);
115 void __fuse_get_request(struct fuse_req *req)
117 atomic_inc(&req->count);
120 /* Must be called with > 1 refcount */
121 static void __fuse_put_request(struct fuse_req *req)
123 BUG_ON(atomic_read(&req->count) < 2);
124 atomic_dec(&req->count);
127 static void fuse_req_init_context(struct fuse_req *req)
129 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
130 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
131 req->in.h.pid = current->pid;
134 void fuse_set_initialized(struct fuse_conn *fc)
136 /* Make sure stores before this are seen on another CPU */
137 smp_wmb();
138 fc->initialized = 1;
141 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
143 return !fc->initialized || (for_background && fc->blocked);
146 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
147 bool for_background)
149 struct fuse_req *req;
150 int err;
151 atomic_inc(&fc->num_waiting);
153 if (fuse_block_alloc(fc, for_background)) {
154 sigset_t oldset;
155 int intr;
157 block_sigs(&oldset);
158 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
159 !fuse_block_alloc(fc, for_background));
160 restore_sigs(&oldset);
161 err = -EINTR;
162 if (intr)
163 goto out;
165 /* Matches smp_wmb() in fuse_set_initialized() */
166 smp_rmb();
168 err = -ENOTCONN;
169 if (!fc->connected)
170 goto out;
172 req = fuse_request_alloc(npages);
173 err = -ENOMEM;
174 if (!req) {
175 if (for_background)
176 wake_up(&fc->blocked_waitq);
177 goto out;
180 fuse_req_init_context(req);
181 req->waiting = 1;
182 req->background = for_background;
183 return req;
185 out:
186 atomic_dec(&fc->num_waiting);
187 return ERR_PTR(err);
190 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
192 return __fuse_get_req(fc, npages, false);
194 EXPORT_SYMBOL_GPL(fuse_get_req);
196 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
197 unsigned npages)
199 return __fuse_get_req(fc, npages, true);
201 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
204 * Return request in fuse_file->reserved_req. However that may
205 * currently be in use. If that is the case, wait for it to become
206 * available.
208 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
209 struct file *file)
211 struct fuse_req *req = NULL;
212 struct fuse_file *ff = file->private_data;
214 do {
215 wait_event(fc->reserved_req_waitq, ff->reserved_req);
216 spin_lock(&fc->lock);
217 if (ff->reserved_req) {
218 req = ff->reserved_req;
219 ff->reserved_req = NULL;
220 req->stolen_file = get_file(file);
222 spin_unlock(&fc->lock);
223 } while (!req);
225 return req;
229 * Put stolen request back into fuse_file->reserved_req
231 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
233 struct file *file = req->stolen_file;
234 struct fuse_file *ff = file->private_data;
236 spin_lock(&fc->lock);
237 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
238 BUG_ON(ff->reserved_req);
239 ff->reserved_req = req;
240 wake_up_all(&fc->reserved_req_waitq);
241 spin_unlock(&fc->lock);
242 fput(file);
246 * Gets a requests for a file operation, always succeeds
248 * This is used for sending the FLUSH request, which must get to
249 * userspace, due to POSIX locks which may need to be unlocked.
251 * If allocation fails due to OOM, use the reserved request in
252 * fuse_file.
254 * This is very unlikely to deadlock accidentally, since the
255 * filesystem should not have it's own file open. If deadlock is
256 * intentional, it can still be broken by "aborting" the filesystem.
258 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
259 struct file *file)
261 struct fuse_req *req;
263 atomic_inc(&fc->num_waiting);
264 wait_event(fc->blocked_waitq, fc->initialized);
265 /* Matches smp_wmb() in fuse_set_initialized() */
266 smp_rmb();
267 req = fuse_request_alloc(0);
268 if (!req)
269 req = get_reserved_req(fc, file);
271 fuse_req_init_context(req);
272 req->waiting = 1;
273 req->background = 0;
274 return req;
277 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
279 if (atomic_dec_and_test(&req->count)) {
280 if (unlikely(req->background)) {
282 * We get here in the unlikely case that a background
283 * request was allocated but not sent
285 spin_lock(&fc->lock);
286 if (!fc->blocked)
287 wake_up(&fc->blocked_waitq);
288 spin_unlock(&fc->lock);
291 if (req->waiting)
292 atomic_dec(&fc->num_waiting);
294 if (req->stolen_file)
295 put_reserved_req(fc, req);
296 else
297 fuse_request_free(req);
300 EXPORT_SYMBOL_GPL(fuse_put_request);
302 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
304 unsigned nbytes = 0;
305 unsigned i;
307 for (i = 0; i < numargs; i++)
308 nbytes += args[i].size;
310 return nbytes;
313 static u64 fuse_get_unique(struct fuse_conn *fc)
315 fc->reqctr++;
316 /* zero is special */
317 if (fc->reqctr == 0)
318 fc->reqctr = 1;
320 return fc->reqctr;
323 static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
325 req->in.h.len = sizeof(struct fuse_in_header) +
326 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
327 list_add_tail(&req->list, &fc->pending);
328 req->state = FUSE_REQ_PENDING;
329 if (!req->waiting) {
330 req->waiting = 1;
331 atomic_inc(&fc->num_waiting);
333 wake_up(&fc->waitq);
334 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
337 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
338 u64 nodeid, u64 nlookup)
340 forget->forget_one.nodeid = nodeid;
341 forget->forget_one.nlookup = nlookup;
343 spin_lock(&fc->lock);
344 if (fc->connected) {
345 fc->forget_list_tail->next = forget;
346 fc->forget_list_tail = forget;
347 wake_up(&fc->waitq);
348 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
349 } else {
350 kfree(forget);
352 spin_unlock(&fc->lock);
355 static void flush_bg_queue(struct fuse_conn *fc)
357 while (fc->active_background < fc->max_background &&
358 !list_empty(&fc->bg_queue)) {
359 struct fuse_req *req;
361 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
362 list_del(&req->list);
363 fc->active_background++;
364 req->in.h.unique = fuse_get_unique(fc);
365 queue_request(fc, req);
370 * This function is called when a request is finished. Either a reply
371 * has arrived or it was aborted (and not yet sent) or some error
372 * occurred during communication with userspace, or the device file
373 * was closed. The requester thread is woken up (if still waiting),
374 * the 'end' callback is called if given, else the reference to the
375 * request is released
377 * Called with fc->lock, unlocks it
379 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
380 __releases(fc->lock)
382 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
383 req->end = NULL;
384 list_del(&req->list);
385 list_del(&req->intr_entry);
386 req->state = FUSE_REQ_FINISHED;
387 if (req->background) {
388 req->background = 0;
390 if (fc->num_background == fc->max_background)
391 fc->blocked = 0;
393 /* Wake up next waiter, if any */
394 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
395 wake_up(&fc->blocked_waitq);
397 if (fc->num_background == fc->congestion_threshold &&
398 fc->connected && fc->bdi_initialized) {
399 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
400 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
402 fc->num_background--;
403 fc->active_background--;
404 flush_bg_queue(fc);
406 spin_unlock(&fc->lock);
407 wake_up(&req->waitq);
408 if (end)
409 end(fc, req);
410 fuse_put_request(fc, req);
413 static void wait_answer_interruptible(struct fuse_conn *fc,
414 struct fuse_req *req)
415 __releases(fc->lock)
416 __acquires(fc->lock)
418 if (signal_pending(current))
419 return;
421 spin_unlock(&fc->lock);
422 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
423 spin_lock(&fc->lock);
426 static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
428 list_add_tail(&req->intr_entry, &fc->interrupts);
429 wake_up(&fc->waitq);
430 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
433 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
434 __releases(fc->lock)
435 __acquires(fc->lock)
437 if (!fc->no_interrupt) {
438 /* Any signal may interrupt this */
439 wait_answer_interruptible(fc, req);
441 if (req->aborted)
442 goto aborted;
443 if (req->state == FUSE_REQ_FINISHED)
444 return;
446 req->interrupted = 1;
447 if (req->state == FUSE_REQ_SENT)
448 queue_interrupt(fc, req);
451 if (!req->force) {
452 sigset_t oldset;
454 /* Only fatal signals may interrupt this */
455 block_sigs(&oldset);
456 wait_answer_interruptible(fc, req);
457 restore_sigs(&oldset);
459 if (req->aborted)
460 goto aborted;
461 if (req->state == FUSE_REQ_FINISHED)
462 return;
464 /* Request is not yet in userspace, bail out */
465 if (req->state == FUSE_REQ_PENDING) {
466 list_del(&req->list);
467 __fuse_put_request(req);
468 req->out.h.error = -EINTR;
469 return;
474 * Either request is already in userspace, or it was forced.
475 * Wait it out.
477 spin_unlock(&fc->lock);
478 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
479 spin_lock(&fc->lock);
481 if (!req->aborted)
482 return;
484 aborted:
485 BUG_ON(req->state != FUSE_REQ_FINISHED);
486 if (req->locked) {
487 /* This is uninterruptible sleep, because data is
488 being copied to/from the buffers of req. During
489 locked state, there mustn't be any filesystem
490 operation (e.g. page fault), since that could lead
491 to deadlock */
492 spin_unlock(&fc->lock);
493 wait_event(req->waitq, !req->locked);
494 spin_lock(&fc->lock);
498 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
500 BUG_ON(req->background);
501 spin_lock(&fc->lock);
502 if (!fc->connected)
503 req->out.h.error = -ENOTCONN;
504 else if (fc->conn_error)
505 req->out.h.error = -ECONNREFUSED;
506 else {
507 req->in.h.unique = fuse_get_unique(fc);
508 queue_request(fc, req);
509 /* acquire extra reference, since request is still needed
510 after request_end() */
511 __fuse_get_request(req);
513 request_wait_answer(fc, req);
515 spin_unlock(&fc->lock);
518 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
520 req->isreply = 1;
521 __fuse_request_send(fc, req);
523 EXPORT_SYMBOL_GPL(fuse_request_send);
525 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
527 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
528 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
530 if (fc->minor < 9) {
531 switch (args->in.h.opcode) {
532 case FUSE_LOOKUP:
533 case FUSE_CREATE:
534 case FUSE_MKNOD:
535 case FUSE_MKDIR:
536 case FUSE_SYMLINK:
537 case FUSE_LINK:
538 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
539 break;
540 case FUSE_GETATTR:
541 case FUSE_SETATTR:
542 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
543 break;
546 if (fc->minor < 12) {
547 switch (args->in.h.opcode) {
548 case FUSE_CREATE:
549 args->in.args[0].size = sizeof(struct fuse_open_in);
550 break;
551 case FUSE_MKNOD:
552 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
553 break;
558 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
560 struct fuse_req *req;
561 ssize_t ret;
563 req = fuse_get_req(fc, 0);
564 if (IS_ERR(req))
565 return PTR_ERR(req);
567 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
568 fuse_adjust_compat(fc, args);
570 req->in.h.opcode = args->in.h.opcode;
571 req->in.h.nodeid = args->in.h.nodeid;
572 req->in.numargs = args->in.numargs;
573 memcpy(req->in.args, args->in.args,
574 args->in.numargs * sizeof(struct fuse_in_arg));
575 req->out.argvar = args->out.argvar;
576 req->out.numargs = args->out.numargs;
577 memcpy(req->out.args, args->out.args,
578 args->out.numargs * sizeof(struct fuse_arg));
579 fuse_request_send(fc, req);
580 ret = req->out.h.error;
581 if (!ret && args->out.argvar) {
582 BUG_ON(args->out.numargs != 1);
583 ret = req->out.args[0].size;
585 fuse_put_request(fc, req);
587 return ret;
590 static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
591 struct fuse_req *req)
593 BUG_ON(!req->background);
594 fc->num_background++;
595 if (fc->num_background == fc->max_background)
596 fc->blocked = 1;
597 if (fc->num_background == fc->congestion_threshold &&
598 fc->bdi_initialized) {
599 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
600 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
602 list_add_tail(&req->list, &fc->bg_queue);
603 flush_bg_queue(fc);
606 static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
608 spin_lock(&fc->lock);
609 if (fc->connected) {
610 fuse_request_send_nowait_locked(fc, req);
611 spin_unlock(&fc->lock);
612 } else {
613 req->out.h.error = -ENOTCONN;
614 request_end(fc, req);
618 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
620 req->isreply = 1;
621 fuse_request_send_nowait(fc, req);
623 EXPORT_SYMBOL_GPL(fuse_request_send_background);
625 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
626 struct fuse_req *req, u64 unique)
628 int err = -ENODEV;
630 req->isreply = 0;
631 req->in.h.unique = unique;
632 spin_lock(&fc->lock);
633 if (fc->connected) {
634 queue_request(fc, req);
635 err = 0;
637 spin_unlock(&fc->lock);
639 return err;
643 * Called under fc->lock
645 * fc->connected must have been checked previously
647 void fuse_request_send_background_locked(struct fuse_conn *fc,
648 struct fuse_req *req)
650 req->isreply = 1;
651 fuse_request_send_nowait_locked(fc, req);
654 void fuse_force_forget(struct file *file, u64 nodeid)
656 struct inode *inode = file_inode(file);
657 struct fuse_conn *fc = get_fuse_conn(inode);
658 struct fuse_req *req;
659 struct fuse_forget_in inarg;
661 memset(&inarg, 0, sizeof(inarg));
662 inarg.nlookup = 1;
663 req = fuse_get_req_nofail_nopages(fc, file);
664 req->in.h.opcode = FUSE_FORGET;
665 req->in.h.nodeid = nodeid;
666 req->in.numargs = 1;
667 req->in.args[0].size = sizeof(inarg);
668 req->in.args[0].value = &inarg;
669 req->isreply = 0;
670 __fuse_request_send(fc, req);
671 /* ignore errors */
672 fuse_put_request(fc, req);
676 * Lock the request. Up to the next unlock_request() there mustn't be
677 * anything that could cause a page-fault. If the request was already
678 * aborted bail out.
680 static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
682 int err = 0;
683 if (req) {
684 spin_lock(&fc->lock);
685 if (req->aborted)
686 err = -ENOENT;
687 else
688 req->locked = 1;
689 spin_unlock(&fc->lock);
691 return err;
695 * Unlock request. If it was aborted during being locked, the
696 * requester thread is currently waiting for it to be unlocked, so
697 * wake it up.
699 static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
701 if (req) {
702 spin_lock(&fc->lock);
703 req->locked = 0;
704 if (req->aborted)
705 wake_up(&req->waitq);
706 spin_unlock(&fc->lock);
710 struct fuse_copy_state {
711 struct fuse_conn *fc;
712 int write;
713 struct fuse_req *req;
714 const struct iovec *iov;
715 struct pipe_buffer *pipebufs;
716 struct pipe_buffer *currbuf;
717 struct pipe_inode_info *pipe;
718 unsigned long nr_segs;
719 unsigned long seglen;
720 unsigned long addr;
721 struct page *pg;
722 unsigned len;
723 unsigned offset;
724 unsigned move_pages:1;
727 static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
728 int write,
729 const struct iovec *iov, unsigned long nr_segs)
731 memset(cs, 0, sizeof(*cs));
732 cs->fc = fc;
733 cs->write = write;
734 cs->iov = iov;
735 cs->nr_segs = nr_segs;
738 /* Unmap and put previous page of userspace buffer */
739 static void fuse_copy_finish(struct fuse_copy_state *cs)
741 if (cs->currbuf) {
742 struct pipe_buffer *buf = cs->currbuf;
744 if (cs->write)
745 buf->len = PAGE_SIZE - cs->len;
746 cs->currbuf = NULL;
747 } else if (cs->pg) {
748 if (cs->write) {
749 flush_dcache_page(cs->pg);
750 set_page_dirty_lock(cs->pg);
752 put_page(cs->pg);
754 cs->pg = NULL;
758 * Get another pagefull of userspace buffer, and map it to kernel
759 * address space, and lock request
761 static int fuse_copy_fill(struct fuse_copy_state *cs)
763 struct page *page;
764 int err;
766 unlock_request(cs->fc, cs->req);
767 fuse_copy_finish(cs);
768 if (cs->pipebufs) {
769 struct pipe_buffer *buf = cs->pipebufs;
771 if (!cs->write) {
772 err = buf->ops->confirm(cs->pipe, buf);
773 if (err)
774 return err;
776 BUG_ON(!cs->nr_segs);
777 cs->currbuf = buf;
778 cs->pg = buf->page;
779 cs->offset = buf->offset;
780 cs->len = buf->len;
781 cs->pipebufs++;
782 cs->nr_segs--;
783 } else {
784 if (cs->nr_segs == cs->pipe->buffers)
785 return -EIO;
787 page = alloc_page(GFP_HIGHUSER);
788 if (!page)
789 return -ENOMEM;
791 buf->page = page;
792 buf->offset = 0;
793 buf->len = 0;
795 cs->currbuf = buf;
796 cs->pg = page;
797 cs->offset = 0;
798 cs->len = PAGE_SIZE;
799 cs->pipebufs++;
800 cs->nr_segs++;
802 } else {
803 if (!cs->seglen) {
804 BUG_ON(!cs->nr_segs);
805 cs->seglen = cs->iov[0].iov_len;
806 cs->addr = (unsigned long) cs->iov[0].iov_base;
807 cs->iov++;
808 cs->nr_segs--;
810 err = get_user_pages_fast(cs->addr, 1, cs->write, &page);
811 if (err < 0)
812 return err;
813 BUG_ON(err != 1);
814 cs->pg = page;
815 cs->offset = cs->addr % PAGE_SIZE;
816 cs->len = min(PAGE_SIZE - cs->offset, cs->seglen);
817 cs->seglen -= cs->len;
818 cs->addr += cs->len;
821 return lock_request(cs->fc, cs->req);
824 /* Do as much copy to/from userspace buffer as we can */
825 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
827 unsigned ncpy = min(*size, cs->len);
828 if (val) {
829 void *pgaddr = kmap_atomic(cs->pg);
830 void *buf = pgaddr + cs->offset;
832 if (cs->write)
833 memcpy(buf, *val, ncpy);
834 else
835 memcpy(*val, buf, ncpy);
837 kunmap_atomic(pgaddr);
838 *val += ncpy;
840 *size -= ncpy;
841 cs->len -= ncpy;
842 cs->offset += ncpy;
843 return ncpy;
846 static int fuse_check_page(struct page *page)
848 if (page_mapcount(page) ||
849 page->mapping != NULL ||
850 page_count(page) != 1 ||
851 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
852 ~(1 << PG_locked |
853 1 << PG_referenced |
854 1 << PG_uptodate |
855 1 << PG_lru |
856 1 << PG_active |
857 1 << PG_reclaim))) {
858 printk(KERN_WARNING "fuse: trying to steal weird page\n");
859 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);
860 return 1;
862 return 0;
865 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
867 int err;
868 struct page *oldpage = *pagep;
869 struct page *newpage;
870 struct pipe_buffer *buf = cs->pipebufs;
872 unlock_request(cs->fc, cs->req);
873 fuse_copy_finish(cs);
875 err = buf->ops->confirm(cs->pipe, buf);
876 if (err)
877 return err;
879 BUG_ON(!cs->nr_segs);
880 cs->currbuf = buf;
881 cs->len = buf->len;
882 cs->pipebufs++;
883 cs->nr_segs--;
885 if (cs->len != PAGE_SIZE)
886 goto out_fallback;
888 if (buf->ops->steal(cs->pipe, buf) != 0)
889 goto out_fallback;
891 newpage = buf->page;
893 if (WARN_ON(!PageUptodate(newpage)))
894 return -EIO;
896 ClearPageMappedToDisk(newpage);
898 if (fuse_check_page(newpage) != 0)
899 goto out_fallback_unlock;
902 * This is a new and locked page, it shouldn't be mapped or
903 * have any special flags on it
905 if (WARN_ON(page_mapped(oldpage)))
906 goto out_fallback_unlock;
907 if (WARN_ON(page_has_private(oldpage)))
908 goto out_fallback_unlock;
909 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
910 goto out_fallback_unlock;
911 if (WARN_ON(PageMlocked(oldpage)))
912 goto out_fallback_unlock;
914 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
915 if (err) {
916 unlock_page(newpage);
917 return err;
920 page_cache_get(newpage);
922 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
923 lru_cache_add_file(newpage);
925 err = 0;
926 spin_lock(&cs->fc->lock);
927 if (cs->req->aborted)
928 err = -ENOENT;
929 else
930 *pagep = newpage;
931 spin_unlock(&cs->fc->lock);
933 if (err) {
934 unlock_page(newpage);
935 page_cache_release(newpage);
936 return err;
939 unlock_page(oldpage);
940 page_cache_release(oldpage);
941 cs->len = 0;
943 return 0;
945 out_fallback_unlock:
946 unlock_page(newpage);
947 out_fallback:
948 cs->pg = buf->page;
949 cs->offset = buf->offset;
951 err = lock_request(cs->fc, cs->req);
952 if (err)
953 return err;
955 return 1;
958 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
959 unsigned offset, unsigned count)
961 struct pipe_buffer *buf;
963 if (cs->nr_segs == cs->pipe->buffers)
964 return -EIO;
966 unlock_request(cs->fc, cs->req);
967 fuse_copy_finish(cs);
969 buf = cs->pipebufs;
970 page_cache_get(page);
971 buf->page = page;
972 buf->offset = offset;
973 buf->len = count;
975 cs->pipebufs++;
976 cs->nr_segs++;
977 cs->len = 0;
979 return 0;
983 * Copy a page in the request to/from the userspace buffer. Must be
984 * done atomically
986 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
987 unsigned offset, unsigned count, int zeroing)
989 int err;
990 struct page *page = *pagep;
992 if (page && zeroing && count < PAGE_SIZE)
993 clear_highpage(page);
995 while (count) {
996 if (cs->write && cs->pipebufs && page) {
997 return fuse_ref_page(cs, page, offset, count);
998 } else if (!cs->len) {
999 if (cs->move_pages && page &&
1000 offset == 0 && count == PAGE_SIZE) {
1001 err = fuse_try_move_page(cs, pagep);
1002 if (err <= 0)
1003 return err;
1004 } else {
1005 err = fuse_copy_fill(cs);
1006 if (err)
1007 return err;
1010 if (page) {
1011 void *mapaddr = kmap_atomic(page);
1012 void *buf = mapaddr + offset;
1013 offset += fuse_copy_do(cs, &buf, &count);
1014 kunmap_atomic(mapaddr);
1015 } else
1016 offset += fuse_copy_do(cs, NULL, &count);
1018 if (page && !cs->write)
1019 flush_dcache_page(page);
1020 return 0;
1023 /* Copy pages in the request to/from userspace buffer */
1024 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1025 int zeroing)
1027 unsigned i;
1028 struct fuse_req *req = cs->req;
1030 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1031 int err;
1032 unsigned offset = req->page_descs[i].offset;
1033 unsigned count = min(nbytes, req->page_descs[i].length);
1035 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1036 zeroing);
1037 if (err)
1038 return err;
1040 nbytes -= count;
1042 return 0;
1045 /* Copy a single argument in the request to/from userspace buffer */
1046 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1048 while (size) {
1049 if (!cs->len) {
1050 int err = fuse_copy_fill(cs);
1051 if (err)
1052 return err;
1054 fuse_copy_do(cs, &val, &size);
1056 return 0;
1059 /* Copy request arguments to/from userspace buffer */
1060 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1061 unsigned argpages, struct fuse_arg *args,
1062 int zeroing)
1064 int err = 0;
1065 unsigned i;
1067 for (i = 0; !err && i < numargs; i++) {
1068 struct fuse_arg *arg = &args[i];
1069 if (i == numargs - 1 && argpages)
1070 err = fuse_copy_pages(cs, arg->size, zeroing);
1071 else
1072 err = fuse_copy_one(cs, arg->value, arg->size);
1074 return err;
1077 static int forget_pending(struct fuse_conn *fc)
1079 return fc->forget_list_head.next != NULL;
1082 static int request_pending(struct fuse_conn *fc)
1084 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
1085 forget_pending(fc);
1088 /* Wait until a request is available on the pending list */
1089 static void request_wait(struct fuse_conn *fc)
1090 __releases(fc->lock)
1091 __acquires(fc->lock)
1093 DECLARE_WAITQUEUE(wait, current);
1095 add_wait_queue_exclusive(&fc->waitq, &wait);
1096 while (fc->connected && !request_pending(fc)) {
1097 set_current_state(TASK_INTERRUPTIBLE);
1098 if (signal_pending(current))
1099 break;
1101 spin_unlock(&fc->lock);
1102 schedule();
1103 spin_lock(&fc->lock);
1105 set_current_state(TASK_RUNNING);
1106 remove_wait_queue(&fc->waitq, &wait);
1110 * Transfer an interrupt request to userspace
1112 * Unlike other requests this is assembled on demand, without a need
1113 * to allocate a separate fuse_req structure.
1115 * Called with fc->lock held, releases it
1117 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1118 size_t nbytes, struct fuse_req *req)
1119 __releases(fc->lock)
1121 struct fuse_in_header ih;
1122 struct fuse_interrupt_in arg;
1123 unsigned reqsize = sizeof(ih) + sizeof(arg);
1124 int err;
1126 list_del_init(&req->intr_entry);
1127 req->intr_unique = fuse_get_unique(fc);
1128 memset(&ih, 0, sizeof(ih));
1129 memset(&arg, 0, sizeof(arg));
1130 ih.len = reqsize;
1131 ih.opcode = FUSE_INTERRUPT;
1132 ih.unique = req->intr_unique;
1133 arg.unique = req->in.h.unique;
1135 spin_unlock(&fc->lock);
1136 if (nbytes < reqsize)
1137 return -EINVAL;
1139 err = fuse_copy_one(cs, &ih, sizeof(ih));
1140 if (!err)
1141 err = fuse_copy_one(cs, &arg, sizeof(arg));
1142 fuse_copy_finish(cs);
1144 return err ? err : reqsize;
1147 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
1148 unsigned max,
1149 unsigned *countp)
1151 struct fuse_forget_link *head = fc->forget_list_head.next;
1152 struct fuse_forget_link **newhead = &head;
1153 unsigned count;
1155 for (count = 0; *newhead != NULL && count < max; count++)
1156 newhead = &(*newhead)->next;
1158 fc->forget_list_head.next = *newhead;
1159 *newhead = NULL;
1160 if (fc->forget_list_head.next == NULL)
1161 fc->forget_list_tail = &fc->forget_list_head;
1163 if (countp != NULL)
1164 *countp = count;
1166 return head;
1169 static int fuse_read_single_forget(struct fuse_conn *fc,
1170 struct fuse_copy_state *cs,
1171 size_t nbytes)
1172 __releases(fc->lock)
1174 int err;
1175 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1176 struct fuse_forget_in arg = {
1177 .nlookup = forget->forget_one.nlookup,
1179 struct fuse_in_header ih = {
1180 .opcode = FUSE_FORGET,
1181 .nodeid = forget->forget_one.nodeid,
1182 .unique = fuse_get_unique(fc),
1183 .len = sizeof(ih) + sizeof(arg),
1186 spin_unlock(&fc->lock);
1187 kfree(forget);
1188 if (nbytes < ih.len)
1189 return -EINVAL;
1191 err = fuse_copy_one(cs, &ih, sizeof(ih));
1192 if (!err)
1193 err = fuse_copy_one(cs, &arg, sizeof(arg));
1194 fuse_copy_finish(cs);
1196 if (err)
1197 return err;
1199 return ih.len;
1202 static int fuse_read_batch_forget(struct fuse_conn *fc,
1203 struct fuse_copy_state *cs, size_t nbytes)
1204 __releases(fc->lock)
1206 int err;
1207 unsigned max_forgets;
1208 unsigned count;
1209 struct fuse_forget_link *head;
1210 struct fuse_batch_forget_in arg = { .count = 0 };
1211 struct fuse_in_header ih = {
1212 .opcode = FUSE_BATCH_FORGET,
1213 .unique = fuse_get_unique(fc),
1214 .len = sizeof(ih) + sizeof(arg),
1217 if (nbytes < ih.len) {
1218 spin_unlock(&fc->lock);
1219 return -EINVAL;
1222 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1223 head = dequeue_forget(fc, max_forgets, &count);
1224 spin_unlock(&fc->lock);
1226 arg.count = count;
1227 ih.len += count * sizeof(struct fuse_forget_one);
1228 err = fuse_copy_one(cs, &ih, sizeof(ih));
1229 if (!err)
1230 err = fuse_copy_one(cs, &arg, sizeof(arg));
1232 while (head) {
1233 struct fuse_forget_link *forget = head;
1235 if (!err) {
1236 err = fuse_copy_one(cs, &forget->forget_one,
1237 sizeof(forget->forget_one));
1239 head = forget->next;
1240 kfree(forget);
1243 fuse_copy_finish(cs);
1245 if (err)
1246 return err;
1248 return ih.len;
1251 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1252 size_t nbytes)
1253 __releases(fc->lock)
1255 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1256 return fuse_read_single_forget(fc, cs, nbytes);
1257 else
1258 return fuse_read_batch_forget(fc, cs, nbytes);
1262 * Read a single request into the userspace filesystem's buffer. This
1263 * function waits until a request is available, then removes it from
1264 * the pending list and copies request data to userspace buffer. If
1265 * no reply is needed (FORGET) or request has been aborted or there
1266 * was an error during the copying then it's finished by calling
1267 * request_end(). Otherwise add it to the processing list, and set
1268 * the 'sent' flag.
1270 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1271 struct fuse_copy_state *cs, size_t nbytes)
1273 int err;
1274 struct fuse_req *req;
1275 struct fuse_in *in;
1276 unsigned reqsize;
1278 restart:
1279 spin_lock(&fc->lock);
1280 err = -EAGAIN;
1281 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1282 !request_pending(fc))
1283 goto err_unlock;
1285 request_wait(fc);
1286 err = -ENODEV;
1287 if (!fc->connected)
1288 goto err_unlock;
1289 err = -ERESTARTSYS;
1290 if (!request_pending(fc))
1291 goto err_unlock;
1293 if (!list_empty(&fc->interrupts)) {
1294 req = list_entry(fc->interrupts.next, struct fuse_req,
1295 intr_entry);
1296 return fuse_read_interrupt(fc, cs, nbytes, req);
1299 if (forget_pending(fc)) {
1300 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1301 return fuse_read_forget(fc, cs, nbytes);
1303 if (fc->forget_batch <= -8)
1304 fc->forget_batch = 16;
1307 req = list_entry(fc->pending.next, struct fuse_req, list);
1308 req->state = FUSE_REQ_READING;
1309 list_move(&req->list, &fc->io);
1311 in = &req->in;
1312 reqsize = in->h.len;
1313 /* If request is too large, reply with an error and restart the read */
1314 if (nbytes < reqsize) {
1315 req->out.h.error = -EIO;
1316 /* SETXATTR is special, since it may contain too large data */
1317 if (in->h.opcode == FUSE_SETXATTR)
1318 req->out.h.error = -E2BIG;
1319 request_end(fc, req);
1320 goto restart;
1322 spin_unlock(&fc->lock);
1323 cs->req = req;
1324 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1325 if (!err)
1326 err = fuse_copy_args(cs, in->numargs, in->argpages,
1327 (struct fuse_arg *) in->args, 0);
1328 fuse_copy_finish(cs);
1329 spin_lock(&fc->lock);
1330 req->locked = 0;
1331 if (req->aborted) {
1332 request_end(fc, req);
1333 return -ENODEV;
1335 if (err) {
1336 req->out.h.error = -EIO;
1337 request_end(fc, req);
1338 return err;
1340 if (!req->isreply)
1341 request_end(fc, req);
1342 else {
1343 req->state = FUSE_REQ_SENT;
1344 list_move_tail(&req->list, &fc->processing);
1345 if (req->interrupted)
1346 queue_interrupt(fc, req);
1347 spin_unlock(&fc->lock);
1349 return reqsize;
1351 err_unlock:
1352 spin_unlock(&fc->lock);
1353 return err;
1356 static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1357 unsigned long nr_segs, loff_t pos)
1359 struct fuse_copy_state cs;
1360 struct file *file = iocb->ki_filp;
1361 struct fuse_conn *fc = fuse_get_conn(file);
1362 if (!fc)
1363 return -EPERM;
1365 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1367 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1370 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1371 struct pipe_inode_info *pipe,
1372 size_t len, unsigned int flags)
1374 int ret;
1375 int page_nr = 0;
1376 int do_wakeup = 0;
1377 struct pipe_buffer *bufs;
1378 struct fuse_copy_state cs;
1379 struct fuse_conn *fc = fuse_get_conn(in);
1380 if (!fc)
1381 return -EPERM;
1383 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1384 if (!bufs)
1385 return -ENOMEM;
1387 fuse_copy_init(&cs, fc, 1, NULL, 0);
1388 cs.pipebufs = bufs;
1389 cs.pipe = pipe;
1390 ret = fuse_dev_do_read(fc, in, &cs, len);
1391 if (ret < 0)
1392 goto out;
1394 ret = 0;
1395 pipe_lock(pipe);
1397 if (!pipe->readers) {
1398 send_sig(SIGPIPE, current, 0);
1399 if (!ret)
1400 ret = -EPIPE;
1401 goto out_unlock;
1404 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1405 ret = -EIO;
1406 goto out_unlock;
1409 while (page_nr < cs.nr_segs) {
1410 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1411 struct pipe_buffer *buf = pipe->bufs + newbuf;
1413 buf->page = bufs[page_nr].page;
1414 buf->offset = bufs[page_nr].offset;
1415 buf->len = bufs[page_nr].len;
1417 * Need to be careful about this. Having buf->ops in module
1418 * code can Oops if the buffer persists after module unload.
1420 buf->ops = &nosteal_pipe_buf_ops;
1422 pipe->nrbufs++;
1423 page_nr++;
1424 ret += buf->len;
1426 if (pipe->files)
1427 do_wakeup = 1;
1430 out_unlock:
1431 pipe_unlock(pipe);
1433 if (do_wakeup) {
1434 smp_mb();
1435 if (waitqueue_active(&pipe->wait))
1436 wake_up_interruptible(&pipe->wait);
1437 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1440 out:
1441 for (; page_nr < cs.nr_segs; page_nr++)
1442 page_cache_release(bufs[page_nr].page);
1444 kfree(bufs);
1445 return ret;
1448 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1449 struct fuse_copy_state *cs)
1451 struct fuse_notify_poll_wakeup_out outarg;
1452 int err = -EINVAL;
1454 if (size != sizeof(outarg))
1455 goto err;
1457 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1458 if (err)
1459 goto err;
1461 fuse_copy_finish(cs);
1462 return fuse_notify_poll_wakeup(fc, &outarg);
1464 err:
1465 fuse_copy_finish(cs);
1466 return err;
1469 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1470 struct fuse_copy_state *cs)
1472 struct fuse_notify_inval_inode_out outarg;
1473 int err = -EINVAL;
1475 if (size != sizeof(outarg))
1476 goto err;
1478 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1479 if (err)
1480 goto err;
1481 fuse_copy_finish(cs);
1483 down_read(&fc->killsb);
1484 err = -ENOENT;
1485 if (fc->sb) {
1486 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1487 outarg.off, outarg.len);
1489 up_read(&fc->killsb);
1490 return err;
1492 err:
1493 fuse_copy_finish(cs);
1494 return err;
1497 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1498 struct fuse_copy_state *cs)
1500 struct fuse_notify_inval_entry_out outarg;
1501 int err = -ENOMEM;
1502 char *buf;
1503 struct qstr name;
1505 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1506 if (!buf)
1507 goto err;
1509 err = -EINVAL;
1510 if (size < sizeof(outarg))
1511 goto err;
1513 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1514 if (err)
1515 goto err;
1517 err = -ENAMETOOLONG;
1518 if (outarg.namelen > FUSE_NAME_MAX)
1519 goto err;
1521 err = -EINVAL;
1522 if (size != sizeof(outarg) + outarg.namelen + 1)
1523 goto err;
1525 name.name = buf;
1526 name.len = outarg.namelen;
1527 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1528 if (err)
1529 goto err;
1530 fuse_copy_finish(cs);
1531 buf[outarg.namelen] = 0;
1532 name.hash = full_name_hash(name.name, name.len);
1534 down_read(&fc->killsb);
1535 err = -ENOENT;
1536 if (fc->sb)
1537 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1538 up_read(&fc->killsb);
1539 kfree(buf);
1540 return err;
1542 err:
1543 kfree(buf);
1544 fuse_copy_finish(cs);
1545 return err;
1548 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1549 struct fuse_copy_state *cs)
1551 struct fuse_notify_delete_out outarg;
1552 int err = -ENOMEM;
1553 char *buf;
1554 struct qstr name;
1556 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1557 if (!buf)
1558 goto err;
1560 err = -EINVAL;
1561 if (size < sizeof(outarg))
1562 goto err;
1564 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1565 if (err)
1566 goto err;
1568 err = -ENAMETOOLONG;
1569 if (outarg.namelen > FUSE_NAME_MAX)
1570 goto err;
1572 err = -EINVAL;
1573 if (size != sizeof(outarg) + outarg.namelen + 1)
1574 goto err;
1576 name.name = buf;
1577 name.len = outarg.namelen;
1578 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1579 if (err)
1580 goto err;
1581 fuse_copy_finish(cs);
1582 buf[outarg.namelen] = 0;
1583 name.hash = full_name_hash(name.name, name.len);
1585 down_read(&fc->killsb);
1586 err = -ENOENT;
1587 if (fc->sb)
1588 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1589 outarg.child, &name);
1590 up_read(&fc->killsb);
1591 kfree(buf);
1592 return err;
1594 err:
1595 kfree(buf);
1596 fuse_copy_finish(cs);
1597 return err;
1600 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1601 struct fuse_copy_state *cs)
1603 struct fuse_notify_store_out outarg;
1604 struct inode *inode;
1605 struct address_space *mapping;
1606 u64 nodeid;
1607 int err;
1608 pgoff_t index;
1609 unsigned int offset;
1610 unsigned int num;
1611 loff_t file_size;
1612 loff_t end;
1614 err = -EINVAL;
1615 if (size < sizeof(outarg))
1616 goto out_finish;
1618 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1619 if (err)
1620 goto out_finish;
1622 err = -EINVAL;
1623 if (size - sizeof(outarg) != outarg.size)
1624 goto out_finish;
1626 nodeid = outarg.nodeid;
1628 down_read(&fc->killsb);
1630 err = -ENOENT;
1631 if (!fc->sb)
1632 goto out_up_killsb;
1634 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1635 if (!inode)
1636 goto out_up_killsb;
1638 mapping = inode->i_mapping;
1639 index = outarg.offset >> PAGE_CACHE_SHIFT;
1640 offset = outarg.offset & ~PAGE_CACHE_MASK;
1641 file_size = i_size_read(inode);
1642 end = outarg.offset + outarg.size;
1643 if (end > file_size) {
1644 file_size = end;
1645 fuse_write_update_size(inode, file_size);
1648 num = outarg.size;
1649 while (num) {
1650 struct page *page;
1651 unsigned int this_num;
1653 err = -ENOMEM;
1654 page = find_or_create_page(mapping, index,
1655 mapping_gfp_mask(mapping));
1656 if (!page)
1657 goto out_iput;
1659 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1660 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1661 if (!err && offset == 0 &&
1662 (this_num == PAGE_CACHE_SIZE || file_size == end))
1663 SetPageUptodate(page);
1664 unlock_page(page);
1665 page_cache_release(page);
1667 if (err)
1668 goto out_iput;
1670 num -= this_num;
1671 offset = 0;
1672 index++;
1675 err = 0;
1677 out_iput:
1678 iput(inode);
1679 out_up_killsb:
1680 up_read(&fc->killsb);
1681 out_finish:
1682 fuse_copy_finish(cs);
1683 return err;
1686 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1688 release_pages(req->pages, req->num_pages, false);
1691 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1692 struct fuse_notify_retrieve_out *outarg)
1694 int err;
1695 struct address_space *mapping = inode->i_mapping;
1696 struct fuse_req *req;
1697 pgoff_t index;
1698 loff_t file_size;
1699 unsigned int num;
1700 unsigned int offset;
1701 size_t total_len = 0;
1702 int num_pages;
1704 offset = outarg->offset & ~PAGE_CACHE_MASK;
1705 file_size = i_size_read(inode);
1707 num = outarg->size;
1708 if (outarg->offset > file_size)
1709 num = 0;
1710 else if (outarg->offset + num > file_size)
1711 num = file_size - outarg->offset;
1713 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1714 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1716 req = fuse_get_req(fc, num_pages);
1717 if (IS_ERR(req))
1718 return PTR_ERR(req);
1720 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1721 req->in.h.nodeid = outarg->nodeid;
1722 req->in.numargs = 2;
1723 req->in.argpages = 1;
1724 req->page_descs[0].offset = offset;
1725 req->end = fuse_retrieve_end;
1727 index = outarg->offset >> PAGE_CACHE_SHIFT;
1729 while (num && req->num_pages < num_pages) {
1730 struct page *page;
1731 unsigned int this_num;
1733 page = find_get_page(mapping, index);
1734 if (!page)
1735 break;
1737 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1738 req->pages[req->num_pages] = page;
1739 req->page_descs[req->num_pages].length = this_num;
1740 req->num_pages++;
1742 offset = 0;
1743 num -= this_num;
1744 total_len += this_num;
1745 index++;
1747 req->misc.retrieve_in.offset = outarg->offset;
1748 req->misc.retrieve_in.size = total_len;
1749 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1750 req->in.args[0].value = &req->misc.retrieve_in;
1751 req->in.args[1].size = total_len;
1753 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1754 if (err)
1755 fuse_retrieve_end(fc, req);
1757 return err;
1760 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1761 struct fuse_copy_state *cs)
1763 struct fuse_notify_retrieve_out outarg;
1764 struct inode *inode;
1765 int err;
1767 err = -EINVAL;
1768 if (size != sizeof(outarg))
1769 goto copy_finish;
1771 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1772 if (err)
1773 goto copy_finish;
1775 fuse_copy_finish(cs);
1777 down_read(&fc->killsb);
1778 err = -ENOENT;
1779 if (fc->sb) {
1780 u64 nodeid = outarg.nodeid;
1782 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1783 if (inode) {
1784 err = fuse_retrieve(fc, inode, &outarg);
1785 iput(inode);
1788 up_read(&fc->killsb);
1790 return err;
1792 copy_finish:
1793 fuse_copy_finish(cs);
1794 return err;
1797 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1798 unsigned int size, struct fuse_copy_state *cs)
1800 switch (code) {
1801 case FUSE_NOTIFY_POLL:
1802 return fuse_notify_poll(fc, size, cs);
1804 case FUSE_NOTIFY_INVAL_INODE:
1805 return fuse_notify_inval_inode(fc, size, cs);
1807 case FUSE_NOTIFY_INVAL_ENTRY:
1808 return fuse_notify_inval_entry(fc, size, cs);
1810 case FUSE_NOTIFY_STORE:
1811 return fuse_notify_store(fc, size, cs);
1813 case FUSE_NOTIFY_RETRIEVE:
1814 return fuse_notify_retrieve(fc, size, cs);
1816 case FUSE_NOTIFY_DELETE:
1817 return fuse_notify_delete(fc, size, cs);
1819 default:
1820 fuse_copy_finish(cs);
1821 return -EINVAL;
1825 /* Look up request on processing list by unique ID */
1826 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1828 struct fuse_req *req;
1830 list_for_each_entry(req, &fc->processing, list) {
1831 if (req->in.h.unique == unique || req->intr_unique == unique)
1832 return req;
1834 return NULL;
1837 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1838 unsigned nbytes)
1840 unsigned reqsize = sizeof(struct fuse_out_header);
1842 if (out->h.error)
1843 return nbytes != reqsize ? -EINVAL : 0;
1845 reqsize += len_args(out->numargs, out->args);
1847 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1848 return -EINVAL;
1849 else if (reqsize > nbytes) {
1850 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1851 unsigned diffsize = reqsize - nbytes;
1852 if (diffsize > lastarg->size)
1853 return -EINVAL;
1854 lastarg->size -= diffsize;
1856 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1857 out->page_zeroing);
1861 * Write a single reply to a request. First the header is copied from
1862 * the write buffer. The request is then searched on the processing
1863 * list by the unique ID found in the header. If found, then remove
1864 * it from the list and copy the rest of the buffer to the request.
1865 * The request is finished by calling request_end()
1867 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1868 struct fuse_copy_state *cs, size_t nbytes)
1870 int err;
1871 struct fuse_req *req;
1872 struct fuse_out_header oh;
1874 if (nbytes < sizeof(struct fuse_out_header))
1875 return -EINVAL;
1877 err = fuse_copy_one(cs, &oh, sizeof(oh));
1878 if (err)
1879 goto err_finish;
1881 err = -EINVAL;
1882 if (oh.len != nbytes)
1883 goto err_finish;
1886 * Zero oh.unique indicates unsolicited notification message
1887 * and error contains notification code.
1889 if (!oh.unique) {
1890 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1891 return err ? err : nbytes;
1894 err = -EINVAL;
1895 if (oh.error <= -1000 || oh.error > 0)
1896 goto err_finish;
1898 spin_lock(&fc->lock);
1899 err = -ENOENT;
1900 if (!fc->connected)
1901 goto err_unlock;
1903 req = request_find(fc, oh.unique);
1904 if (!req)
1905 goto err_unlock;
1907 if (req->aborted) {
1908 spin_unlock(&fc->lock);
1909 fuse_copy_finish(cs);
1910 spin_lock(&fc->lock);
1911 request_end(fc, req);
1912 return -ENOENT;
1914 /* Is it an interrupt reply? */
1915 if (req->intr_unique == oh.unique) {
1916 err = -EINVAL;
1917 if (nbytes != sizeof(struct fuse_out_header))
1918 goto err_unlock;
1920 if (oh.error == -ENOSYS)
1921 fc->no_interrupt = 1;
1922 else if (oh.error == -EAGAIN)
1923 queue_interrupt(fc, req);
1925 spin_unlock(&fc->lock);
1926 fuse_copy_finish(cs);
1927 return nbytes;
1930 req->state = FUSE_REQ_WRITING;
1931 list_move(&req->list, &fc->io);
1932 req->out.h = oh;
1933 req->locked = 1;
1934 cs->req = req;
1935 if (!req->out.page_replace)
1936 cs->move_pages = 0;
1937 spin_unlock(&fc->lock);
1939 err = copy_out_args(cs, &req->out, nbytes);
1940 fuse_copy_finish(cs);
1942 spin_lock(&fc->lock);
1943 req->locked = 0;
1944 if (!err) {
1945 if (req->aborted)
1946 err = -ENOENT;
1947 } else if (!req->aborted)
1948 req->out.h.error = -EIO;
1949 request_end(fc, req);
1951 return err ? err : nbytes;
1953 err_unlock:
1954 spin_unlock(&fc->lock);
1955 err_finish:
1956 fuse_copy_finish(cs);
1957 return err;
1960 static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1961 unsigned long nr_segs, loff_t pos)
1963 struct fuse_copy_state cs;
1964 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1965 if (!fc)
1966 return -EPERM;
1968 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1970 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1973 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1974 struct file *out, loff_t *ppos,
1975 size_t len, unsigned int flags)
1977 unsigned nbuf;
1978 unsigned idx;
1979 struct pipe_buffer *bufs;
1980 struct fuse_copy_state cs;
1981 struct fuse_conn *fc;
1982 size_t rem;
1983 ssize_t ret;
1985 fc = fuse_get_conn(out);
1986 if (!fc)
1987 return -EPERM;
1989 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1990 if (!bufs)
1991 return -ENOMEM;
1993 pipe_lock(pipe);
1994 nbuf = 0;
1995 rem = 0;
1996 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1997 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1999 ret = -EINVAL;
2000 if (rem < len) {
2001 pipe_unlock(pipe);
2002 goto out;
2005 rem = len;
2006 while (rem) {
2007 struct pipe_buffer *ibuf;
2008 struct pipe_buffer *obuf;
2010 BUG_ON(nbuf >= pipe->buffers);
2011 BUG_ON(!pipe->nrbufs);
2012 ibuf = &pipe->bufs[pipe->curbuf];
2013 obuf = &bufs[nbuf];
2015 if (rem >= ibuf->len) {
2016 *obuf = *ibuf;
2017 ibuf->ops = NULL;
2018 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2019 pipe->nrbufs--;
2020 } else {
2021 ibuf->ops->get(pipe, ibuf);
2022 *obuf = *ibuf;
2023 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2024 obuf->len = rem;
2025 ibuf->offset += obuf->len;
2026 ibuf->len -= obuf->len;
2028 nbuf++;
2029 rem -= obuf->len;
2031 pipe_unlock(pipe);
2033 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
2034 cs.pipebufs = bufs;
2035 cs.pipe = pipe;
2037 if (flags & SPLICE_F_MOVE)
2038 cs.move_pages = 1;
2040 ret = fuse_dev_do_write(fc, &cs, len);
2042 for (idx = 0; idx < nbuf; idx++) {
2043 struct pipe_buffer *buf = &bufs[idx];
2044 buf->ops->release(pipe, buf);
2046 out:
2047 kfree(bufs);
2048 return ret;
2051 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2053 unsigned mask = POLLOUT | POLLWRNORM;
2054 struct fuse_conn *fc = fuse_get_conn(file);
2055 if (!fc)
2056 return POLLERR;
2058 poll_wait(file, &fc->waitq, wait);
2060 spin_lock(&fc->lock);
2061 if (!fc->connected)
2062 mask = POLLERR;
2063 else if (request_pending(fc))
2064 mask |= POLLIN | POLLRDNORM;
2065 spin_unlock(&fc->lock);
2067 return mask;
2071 * Abort all requests on the given list (pending or processing)
2073 * This function releases and reacquires fc->lock
2075 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2076 __releases(fc->lock)
2077 __acquires(fc->lock)
2079 while (!list_empty(head)) {
2080 struct fuse_req *req;
2081 req = list_entry(head->next, struct fuse_req, list);
2082 req->out.h.error = -ECONNABORTED;
2083 request_end(fc, req);
2084 spin_lock(&fc->lock);
2089 * Abort requests under I/O
2091 * The requests are set to aborted and finished, and the request
2092 * waiter is woken up. This will make request_wait_answer() wait
2093 * until the request is unlocked and then return.
2095 * If the request is asynchronous, then the end function needs to be
2096 * called after waiting for the request to be unlocked (if it was
2097 * locked).
2099 static void end_io_requests(struct fuse_conn *fc)
2100 __releases(fc->lock)
2101 __acquires(fc->lock)
2103 while (!list_empty(&fc->io)) {
2104 struct fuse_req *req =
2105 list_entry(fc->io.next, struct fuse_req, list);
2106 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
2108 req->aborted = 1;
2109 req->out.h.error = -ECONNABORTED;
2110 req->state = FUSE_REQ_FINISHED;
2111 list_del_init(&req->list);
2112 wake_up(&req->waitq);
2113 if (end) {
2114 req->end = NULL;
2115 __fuse_get_request(req);
2116 spin_unlock(&fc->lock);
2117 wait_event(req->waitq, !req->locked);
2118 end(fc, req);
2119 fuse_put_request(fc, req);
2120 spin_lock(&fc->lock);
2125 static void end_queued_requests(struct fuse_conn *fc)
2126 __releases(fc->lock)
2127 __acquires(fc->lock)
2129 fc->max_background = UINT_MAX;
2130 flush_bg_queue(fc);
2131 end_requests(fc, &fc->pending);
2132 end_requests(fc, &fc->processing);
2133 while (forget_pending(fc))
2134 kfree(dequeue_forget(fc, 1, NULL));
2137 static void end_polls(struct fuse_conn *fc)
2139 struct rb_node *p;
2141 p = rb_first(&fc->polled_files);
2143 while (p) {
2144 struct fuse_file *ff;
2145 ff = rb_entry(p, struct fuse_file, polled_node);
2146 wake_up_interruptible_all(&ff->poll_wait);
2148 p = rb_next(p);
2153 * Abort all requests.
2155 * Emergency exit in case of a malicious or accidental deadlock, or
2156 * just a hung filesystem.
2158 * The same effect is usually achievable through killing the
2159 * filesystem daemon and all users of the filesystem. The exception
2160 * is the combination of an asynchronous request and the tricky
2161 * deadlock (see Documentation/filesystems/fuse.txt).
2163 * During the aborting, progression of requests from the pending and
2164 * processing lists onto the io list, and progression of new requests
2165 * onto the pending list is prevented by req->connected being false.
2167 * Progression of requests under I/O to the processing list is
2168 * prevented by the req->aborted flag being true for these requests.
2169 * For this reason requests on the io list must be aborted first.
2171 void fuse_abort_conn(struct fuse_conn *fc)
2173 spin_lock(&fc->lock);
2174 if (fc->connected) {
2175 fc->connected = 0;
2176 fc->blocked = 0;
2177 fuse_set_initialized(fc);
2178 end_io_requests(fc);
2179 end_queued_requests(fc);
2180 end_polls(fc);
2181 wake_up_all(&fc->waitq);
2182 wake_up_all(&fc->blocked_waitq);
2183 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2185 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_conn *fc = fuse_get_conn(file);
2192 if (fc) {
2193 spin_lock(&fc->lock);
2194 fc->connected = 0;
2195 fc->blocked = 0;
2196 fuse_set_initialized(fc);
2197 end_queued_requests(fc);
2198 end_polls(fc);
2199 wake_up_all(&fc->blocked_waitq);
2200 spin_unlock(&fc->lock);
2201 fuse_conn_put(fc);
2204 return 0;
2206 EXPORT_SYMBOL_GPL(fuse_dev_release);
2208 static int fuse_dev_fasync(int fd, struct file *file, int on)
2210 struct fuse_conn *fc = fuse_get_conn(file);
2211 if (!fc)
2212 return -EPERM;
2214 /* No locking - fasync_helper does its own locking */
2215 return fasync_helper(fd, file, on, &fc->fasync);
2218 const struct file_operations fuse_dev_operations = {
2219 .owner = THIS_MODULE,
2220 .llseek = no_llseek,
2221 .read = do_sync_read,
2222 .aio_read = fuse_dev_read,
2223 .splice_read = fuse_dev_splice_read,
2224 .write = do_sync_write,
2225 .aio_write = fuse_dev_write,
2226 .splice_write = fuse_dev_splice_write,
2227 .poll = fuse_dev_poll,
2228 .release = fuse_dev_release,
2229 .fasync = fuse_dev_fasync,
2231 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2233 static struct miscdevice fuse_miscdevice = {
2234 .minor = FUSE_MINOR,
2235 .name = "fuse",
2236 .fops = &fuse_dev_operations,
2239 int __init fuse_dev_init(void)
2241 int err = -ENOMEM;
2242 fuse_req_cachep = kmem_cache_create("fuse_request",
2243 sizeof(struct fuse_req),
2244 0, 0, NULL);
2245 if (!fuse_req_cachep)
2246 goto out;
2248 err = misc_register(&fuse_miscdevice);
2249 if (err)
2250 goto out_cache_clean;
2252 return 0;
2254 out_cache_clean:
2255 kmem_cache_destroy(fuse_req_cachep);
2256 out:
2257 return err;
2260 void fuse_dev_cleanup(void)
2262 misc_deregister(&fuse_miscdevice);
2263 kmem_cache_destroy(fuse_req_cachep);