spi: davinci: fix up dma_mapping_error() incorrect patch
[linux/fpc-iii.git] / fs / fuse / dev.c
blobf11792672977d2280ec7fc301f81e9b1b1ba7741
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>
23 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24 MODULE_ALIAS("devname:fuse");
26 static struct kmem_cache *fuse_req_cachep;
28 static struct fuse_dev *fuse_get_dev(struct file *file)
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
34 return ACCESS_ONCE(file->private_data);
37 static void fuse_request_init(struct fuse_req *req, struct page **pages,
38 struct fuse_page_desc *page_descs,
39 unsigned npages)
41 memset(req, 0, sizeof(*req));
42 memset(pages, 0, sizeof(*pages) * npages);
43 memset(page_descs, 0, sizeof(*page_descs) * npages);
44 INIT_LIST_HEAD(&req->list);
45 INIT_LIST_HEAD(&req->intr_entry);
46 init_waitqueue_head(&req->waitq);
47 atomic_set(&req->count, 1);
48 req->pages = pages;
49 req->page_descs = page_descs;
50 req->max_pages = npages;
51 __set_bit(FR_PENDING, &req->flags);
54 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
56 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
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 void __fuse_get_request(struct fuse_req *req)
104 atomic_inc(&req->count);
107 /* Must be called with > 1 refcount */
108 static void __fuse_put_request(struct fuse_req *req)
110 BUG_ON(atomic_read(&req->count) < 2);
111 atomic_dec(&req->count);
114 static void fuse_req_init_context(struct fuse_req *req)
116 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
117 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
118 req->in.h.pid = current->pid;
121 void fuse_set_initialized(struct fuse_conn *fc)
123 /* Make sure stores before this are seen on another CPU */
124 smp_wmb();
125 fc->initialized = 1;
128 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
130 return !fc->initialized || (for_background && fc->blocked);
133 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
134 bool for_background)
136 struct fuse_req *req;
137 int err;
138 atomic_inc(&fc->num_waiting);
140 if (fuse_block_alloc(fc, for_background)) {
141 err = -EINTR;
142 if (wait_event_killable_exclusive(fc->blocked_waitq,
143 !fuse_block_alloc(fc, for_background)))
144 goto out;
146 /* Matches smp_wmb() in fuse_set_initialized() */
147 smp_rmb();
149 err = -ENOTCONN;
150 if (!fc->connected)
151 goto out;
153 err = -ECONNREFUSED;
154 if (fc->conn_error)
155 goto out;
157 req = fuse_request_alloc(npages);
158 err = -ENOMEM;
159 if (!req) {
160 if (for_background)
161 wake_up(&fc->blocked_waitq);
162 goto out;
165 fuse_req_init_context(req);
166 __set_bit(FR_WAITING, &req->flags);
167 if (for_background)
168 __set_bit(FR_BACKGROUND, &req->flags);
170 return req;
172 out:
173 atomic_dec(&fc->num_waiting);
174 return ERR_PTR(err);
177 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
179 return __fuse_get_req(fc, npages, false);
181 EXPORT_SYMBOL_GPL(fuse_get_req);
183 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
184 unsigned npages)
186 return __fuse_get_req(fc, npages, true);
188 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
191 * Return request in fuse_file->reserved_req. However that may
192 * currently be in use. If that is the case, wait for it to become
193 * available.
195 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
196 struct file *file)
198 struct fuse_req *req = NULL;
199 struct fuse_file *ff = file->private_data;
201 do {
202 wait_event(fc->reserved_req_waitq, ff->reserved_req);
203 spin_lock(&fc->lock);
204 if (ff->reserved_req) {
205 req = ff->reserved_req;
206 ff->reserved_req = NULL;
207 req->stolen_file = get_file(file);
209 spin_unlock(&fc->lock);
210 } while (!req);
212 return req;
216 * Put stolen request back into fuse_file->reserved_req
218 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
220 struct file *file = req->stolen_file;
221 struct fuse_file *ff = file->private_data;
223 spin_lock(&fc->lock);
224 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
225 BUG_ON(ff->reserved_req);
226 ff->reserved_req = req;
227 wake_up_all(&fc->reserved_req_waitq);
228 spin_unlock(&fc->lock);
229 fput(file);
233 * Gets a requests for a file operation, always succeeds
235 * This is used for sending the FLUSH request, which must get to
236 * userspace, due to POSIX locks which may need to be unlocked.
238 * If allocation fails due to OOM, use the reserved request in
239 * fuse_file.
241 * This is very unlikely to deadlock accidentally, since the
242 * filesystem should not have it's own file open. If deadlock is
243 * intentional, it can still be broken by "aborting" the filesystem.
245 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
246 struct file *file)
248 struct fuse_req *req;
250 atomic_inc(&fc->num_waiting);
251 wait_event(fc->blocked_waitq, fc->initialized);
252 /* Matches smp_wmb() in fuse_set_initialized() */
253 smp_rmb();
254 req = fuse_request_alloc(0);
255 if (!req)
256 req = get_reserved_req(fc, file);
258 fuse_req_init_context(req);
259 __set_bit(FR_WAITING, &req->flags);
260 __clear_bit(FR_BACKGROUND, &req->flags);
261 return req;
264 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
266 if (atomic_dec_and_test(&req->count)) {
267 if (test_bit(FR_BACKGROUND, &req->flags)) {
269 * We get here in the unlikely case that a background
270 * request was allocated but not sent
272 spin_lock(&fc->lock);
273 if (!fc->blocked)
274 wake_up(&fc->blocked_waitq);
275 spin_unlock(&fc->lock);
278 if (test_bit(FR_WAITING, &req->flags)) {
279 __clear_bit(FR_WAITING, &req->flags);
280 atomic_dec(&fc->num_waiting);
283 if (req->stolen_file)
284 put_reserved_req(fc, req);
285 else
286 fuse_request_free(req);
289 EXPORT_SYMBOL_GPL(fuse_put_request);
291 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
293 unsigned nbytes = 0;
294 unsigned i;
296 for (i = 0; i < numargs; i++)
297 nbytes += args[i].size;
299 return nbytes;
302 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
304 return ++fiq->reqctr;
307 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
309 req->in.h.len = sizeof(struct fuse_in_header) +
310 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
311 list_add_tail(&req->list, &fiq->pending);
312 wake_up_locked(&fiq->waitq);
313 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
316 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
317 u64 nodeid, u64 nlookup)
319 struct fuse_iqueue *fiq = &fc->iq;
321 forget->forget_one.nodeid = nodeid;
322 forget->forget_one.nlookup = nlookup;
324 spin_lock(&fiq->waitq.lock);
325 if (fiq->connected) {
326 fiq->forget_list_tail->next = forget;
327 fiq->forget_list_tail = forget;
328 wake_up_locked(&fiq->waitq);
329 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
330 } else {
331 kfree(forget);
333 spin_unlock(&fiq->waitq.lock);
336 static void flush_bg_queue(struct fuse_conn *fc)
338 while (fc->active_background < fc->max_background &&
339 !list_empty(&fc->bg_queue)) {
340 struct fuse_req *req;
341 struct fuse_iqueue *fiq = &fc->iq;
343 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
344 list_del(&req->list);
345 fc->active_background++;
346 spin_lock(&fiq->waitq.lock);
347 req->in.h.unique = fuse_get_unique(fiq);
348 queue_request(fiq, req);
349 spin_unlock(&fiq->waitq.lock);
354 * This function is called when a request is finished. Either a reply
355 * has arrived or it was aborted (and not yet sent) or some error
356 * occurred during communication with userspace, or the device file
357 * was closed. The requester thread is woken up (if still waiting),
358 * the 'end' callback is called if given, else the reference to the
359 * request is released
361 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
363 struct fuse_iqueue *fiq = &fc->iq;
365 if (test_and_set_bit(FR_FINISHED, &req->flags))
366 return;
368 spin_lock(&fiq->waitq.lock);
369 list_del_init(&req->intr_entry);
370 spin_unlock(&fiq->waitq.lock);
371 WARN_ON(test_bit(FR_PENDING, &req->flags));
372 WARN_ON(test_bit(FR_SENT, &req->flags));
373 if (test_bit(FR_BACKGROUND, &req->flags)) {
374 spin_lock(&fc->lock);
375 clear_bit(FR_BACKGROUND, &req->flags);
376 if (fc->num_background == fc->max_background)
377 fc->blocked = 0;
379 /* Wake up next waiter, if any */
380 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
381 wake_up(&fc->blocked_waitq);
383 if (fc->num_background == fc->congestion_threshold &&
384 fc->connected && fc->bdi_initialized) {
385 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
386 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
388 fc->num_background--;
389 fc->active_background--;
390 flush_bg_queue(fc);
391 spin_unlock(&fc->lock);
393 wake_up(&req->waitq);
394 if (req->end)
395 req->end(fc, req);
396 fuse_put_request(fc, req);
399 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
401 spin_lock(&fiq->waitq.lock);
402 if (test_bit(FR_FINISHED, &req->flags)) {
403 spin_unlock(&fiq->waitq.lock);
404 return;
406 if (list_empty(&req->intr_entry)) {
407 list_add_tail(&req->intr_entry, &fiq->interrupts);
408 wake_up_locked(&fiq->waitq);
410 spin_unlock(&fiq->waitq.lock);
411 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
414 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
416 struct fuse_iqueue *fiq = &fc->iq;
417 int err;
419 if (!fc->no_interrupt) {
420 /* Any signal may interrupt this */
421 err = wait_event_interruptible(req->waitq,
422 test_bit(FR_FINISHED, &req->flags));
423 if (!err)
424 return;
426 set_bit(FR_INTERRUPTED, &req->flags);
427 /* matches barrier in fuse_dev_do_read() */
428 smp_mb__after_atomic();
429 if (test_bit(FR_SENT, &req->flags))
430 queue_interrupt(fiq, req);
433 if (!test_bit(FR_FORCE, &req->flags)) {
434 /* Only fatal signals may interrupt this */
435 err = wait_event_killable(req->waitq,
436 test_bit(FR_FINISHED, &req->flags));
437 if (!err)
438 return;
440 spin_lock(&fiq->waitq.lock);
441 /* Request is not yet in userspace, bail out */
442 if (test_bit(FR_PENDING, &req->flags)) {
443 list_del(&req->list);
444 spin_unlock(&fiq->waitq.lock);
445 __fuse_put_request(req);
446 req->out.h.error = -EINTR;
447 return;
449 spin_unlock(&fiq->waitq.lock);
453 * Either request is already in userspace, or it was forced.
454 * Wait it out.
456 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
459 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
461 struct fuse_iqueue *fiq = &fc->iq;
463 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
464 spin_lock(&fiq->waitq.lock);
465 if (!fiq->connected) {
466 spin_unlock(&fiq->waitq.lock);
467 req->out.h.error = -ENOTCONN;
468 } else {
469 req->in.h.unique = fuse_get_unique(fiq);
470 queue_request(fiq, req);
471 /* acquire extra reference, since request is still needed
472 after request_end() */
473 __fuse_get_request(req);
474 spin_unlock(&fiq->waitq.lock);
476 request_wait_answer(fc, req);
477 /* Pairs with smp_wmb() in request_end() */
478 smp_rmb();
482 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
484 __set_bit(FR_ISREPLY, &req->flags);
485 if (!test_bit(FR_WAITING, &req->flags)) {
486 __set_bit(FR_WAITING, &req->flags);
487 atomic_inc(&fc->num_waiting);
489 __fuse_request_send(fc, req);
491 EXPORT_SYMBOL_GPL(fuse_request_send);
493 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
495 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
496 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
498 if (fc->minor < 9) {
499 switch (args->in.h.opcode) {
500 case FUSE_LOOKUP:
501 case FUSE_CREATE:
502 case FUSE_MKNOD:
503 case FUSE_MKDIR:
504 case FUSE_SYMLINK:
505 case FUSE_LINK:
506 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
507 break;
508 case FUSE_GETATTR:
509 case FUSE_SETATTR:
510 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
511 break;
514 if (fc->minor < 12) {
515 switch (args->in.h.opcode) {
516 case FUSE_CREATE:
517 args->in.args[0].size = sizeof(struct fuse_open_in);
518 break;
519 case FUSE_MKNOD:
520 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
521 break;
526 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
528 struct fuse_req *req;
529 ssize_t ret;
531 req = fuse_get_req(fc, 0);
532 if (IS_ERR(req))
533 return PTR_ERR(req);
535 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
536 fuse_adjust_compat(fc, args);
538 req->in.h.opcode = args->in.h.opcode;
539 req->in.h.nodeid = args->in.h.nodeid;
540 req->in.numargs = args->in.numargs;
541 memcpy(req->in.args, args->in.args,
542 args->in.numargs * sizeof(struct fuse_in_arg));
543 req->out.argvar = args->out.argvar;
544 req->out.numargs = args->out.numargs;
545 memcpy(req->out.args, args->out.args,
546 args->out.numargs * sizeof(struct fuse_arg));
547 fuse_request_send(fc, req);
548 ret = req->out.h.error;
549 if (!ret && args->out.argvar) {
550 BUG_ON(args->out.numargs != 1);
551 ret = req->out.args[0].size;
553 fuse_put_request(fc, req);
555 return ret;
559 * Called under fc->lock
561 * fc->connected must have been checked previously
563 void fuse_request_send_background_locked(struct fuse_conn *fc,
564 struct fuse_req *req)
566 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
567 if (!test_bit(FR_WAITING, &req->flags)) {
568 __set_bit(FR_WAITING, &req->flags);
569 atomic_inc(&fc->num_waiting);
571 __set_bit(FR_ISREPLY, &req->flags);
572 fc->num_background++;
573 if (fc->num_background == fc->max_background)
574 fc->blocked = 1;
575 if (fc->num_background == fc->congestion_threshold &&
576 fc->bdi_initialized) {
577 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
578 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
580 list_add_tail(&req->list, &fc->bg_queue);
581 flush_bg_queue(fc);
584 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
586 BUG_ON(!req->end);
587 spin_lock(&fc->lock);
588 if (fc->connected) {
589 fuse_request_send_background_locked(fc, req);
590 spin_unlock(&fc->lock);
591 } else {
592 spin_unlock(&fc->lock);
593 req->out.h.error = -ENOTCONN;
594 req->end(fc, req);
595 fuse_put_request(fc, req);
598 EXPORT_SYMBOL_GPL(fuse_request_send_background);
600 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
601 struct fuse_req *req, u64 unique)
603 int err = -ENODEV;
604 struct fuse_iqueue *fiq = &fc->iq;
606 __clear_bit(FR_ISREPLY, &req->flags);
607 req->in.h.unique = unique;
608 spin_lock(&fiq->waitq.lock);
609 if (fiq->connected) {
610 queue_request(fiq, req);
611 err = 0;
613 spin_unlock(&fiq->waitq.lock);
615 return err;
618 void fuse_force_forget(struct file *file, u64 nodeid)
620 struct inode *inode = file_inode(file);
621 struct fuse_conn *fc = get_fuse_conn(inode);
622 struct fuse_req *req;
623 struct fuse_forget_in inarg;
625 memset(&inarg, 0, sizeof(inarg));
626 inarg.nlookup = 1;
627 req = fuse_get_req_nofail_nopages(fc, file);
628 req->in.h.opcode = FUSE_FORGET;
629 req->in.h.nodeid = nodeid;
630 req->in.numargs = 1;
631 req->in.args[0].size = sizeof(inarg);
632 req->in.args[0].value = &inarg;
633 __clear_bit(FR_ISREPLY, &req->flags);
634 __fuse_request_send(fc, req);
635 /* ignore errors */
636 fuse_put_request(fc, req);
640 * Lock the request. Up to the next unlock_request() there mustn't be
641 * anything that could cause a page-fault. If the request was already
642 * aborted bail out.
644 static int lock_request(struct fuse_req *req)
646 int err = 0;
647 if (req) {
648 spin_lock(&req->waitq.lock);
649 if (test_bit(FR_ABORTED, &req->flags))
650 err = -ENOENT;
651 else
652 set_bit(FR_LOCKED, &req->flags);
653 spin_unlock(&req->waitq.lock);
655 return err;
659 * Unlock request. If it was aborted while locked, caller is responsible
660 * for unlocking and ending the request.
662 static int unlock_request(struct fuse_req *req)
664 int err = 0;
665 if (req) {
666 spin_lock(&req->waitq.lock);
667 if (test_bit(FR_ABORTED, &req->flags))
668 err = -ENOENT;
669 else
670 clear_bit(FR_LOCKED, &req->flags);
671 spin_unlock(&req->waitq.lock);
673 return err;
676 struct fuse_copy_state {
677 int write;
678 struct fuse_req *req;
679 struct iov_iter *iter;
680 struct pipe_buffer *pipebufs;
681 struct pipe_buffer *currbuf;
682 struct pipe_inode_info *pipe;
683 unsigned long nr_segs;
684 struct page *pg;
685 unsigned len;
686 unsigned offset;
687 unsigned move_pages:1;
690 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
691 struct iov_iter *iter)
693 memset(cs, 0, sizeof(*cs));
694 cs->write = write;
695 cs->iter = iter;
698 /* Unmap and put previous page of userspace buffer */
699 static void fuse_copy_finish(struct fuse_copy_state *cs)
701 if (cs->currbuf) {
702 struct pipe_buffer *buf = cs->currbuf;
704 if (cs->write)
705 buf->len = PAGE_SIZE - cs->len;
706 cs->currbuf = NULL;
707 } else if (cs->pg) {
708 if (cs->write) {
709 flush_dcache_page(cs->pg);
710 set_page_dirty_lock(cs->pg);
712 put_page(cs->pg);
714 cs->pg = NULL;
718 * Get another pagefull of userspace buffer, and map it to kernel
719 * address space, and lock request
721 static int fuse_copy_fill(struct fuse_copy_state *cs)
723 struct page *page;
724 int err;
726 err = unlock_request(cs->req);
727 if (err)
728 return err;
730 fuse_copy_finish(cs);
731 if (cs->pipebufs) {
732 struct pipe_buffer *buf = cs->pipebufs;
734 if (!cs->write) {
735 err = pipe_buf_confirm(cs->pipe, buf);
736 if (err)
737 return err;
739 BUG_ON(!cs->nr_segs);
740 cs->currbuf = buf;
741 cs->pg = buf->page;
742 cs->offset = buf->offset;
743 cs->len = buf->len;
744 cs->pipebufs++;
745 cs->nr_segs--;
746 } else {
747 if (cs->nr_segs == cs->pipe->buffers)
748 return -EIO;
750 page = alloc_page(GFP_HIGHUSER);
751 if (!page)
752 return -ENOMEM;
754 buf->page = page;
755 buf->offset = 0;
756 buf->len = 0;
758 cs->currbuf = buf;
759 cs->pg = page;
760 cs->offset = 0;
761 cs->len = PAGE_SIZE;
762 cs->pipebufs++;
763 cs->nr_segs++;
765 } else {
766 size_t off;
767 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
768 if (err < 0)
769 return err;
770 BUG_ON(!err);
771 cs->len = err;
772 cs->offset = off;
773 cs->pg = page;
774 iov_iter_advance(cs->iter, err);
777 return lock_request(cs->req);
780 /* Do as much copy to/from userspace buffer as we can */
781 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
783 unsigned ncpy = min(*size, cs->len);
784 if (val) {
785 void *pgaddr = kmap_atomic(cs->pg);
786 void *buf = pgaddr + cs->offset;
788 if (cs->write)
789 memcpy(buf, *val, ncpy);
790 else
791 memcpy(*val, buf, ncpy);
793 kunmap_atomic(pgaddr);
794 *val += ncpy;
796 *size -= ncpy;
797 cs->len -= ncpy;
798 cs->offset += ncpy;
799 return ncpy;
802 static int fuse_check_page(struct page *page)
804 if (page_mapcount(page) ||
805 page->mapping != NULL ||
806 page_count(page) != 1 ||
807 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
808 ~(1 << PG_locked |
809 1 << PG_referenced |
810 1 << PG_uptodate |
811 1 << PG_lru |
812 1 << PG_active |
813 1 << PG_reclaim))) {
814 printk(KERN_WARNING "fuse: trying to steal weird page\n");
815 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);
816 return 1;
818 return 0;
821 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
823 int err;
824 struct page *oldpage = *pagep;
825 struct page *newpage;
826 struct pipe_buffer *buf = cs->pipebufs;
828 err = unlock_request(cs->req);
829 if (err)
830 return err;
832 fuse_copy_finish(cs);
834 err = pipe_buf_confirm(cs->pipe, buf);
835 if (err)
836 return err;
838 BUG_ON(!cs->nr_segs);
839 cs->currbuf = buf;
840 cs->len = buf->len;
841 cs->pipebufs++;
842 cs->nr_segs--;
844 if (cs->len != PAGE_SIZE)
845 goto out_fallback;
847 if (pipe_buf_steal(cs->pipe, buf) != 0)
848 goto out_fallback;
850 newpage = buf->page;
852 if (!PageUptodate(newpage))
853 SetPageUptodate(newpage);
855 ClearPageMappedToDisk(newpage);
857 if (fuse_check_page(newpage) != 0)
858 goto out_fallback_unlock;
861 * This is a new and locked page, it shouldn't be mapped or
862 * have any special flags on it
864 if (WARN_ON(page_mapped(oldpage)))
865 goto out_fallback_unlock;
866 if (WARN_ON(page_has_private(oldpage)))
867 goto out_fallback_unlock;
868 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
869 goto out_fallback_unlock;
870 if (WARN_ON(PageMlocked(oldpage)))
871 goto out_fallback_unlock;
873 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
874 if (err) {
875 unlock_page(newpage);
876 return err;
879 get_page(newpage);
881 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
882 lru_cache_add_file(newpage);
884 err = 0;
885 spin_lock(&cs->req->waitq.lock);
886 if (test_bit(FR_ABORTED, &cs->req->flags))
887 err = -ENOENT;
888 else
889 *pagep = newpage;
890 spin_unlock(&cs->req->waitq.lock);
892 if (err) {
893 unlock_page(newpage);
894 put_page(newpage);
895 return err;
898 unlock_page(oldpage);
899 put_page(oldpage);
900 cs->len = 0;
902 return 0;
904 out_fallback_unlock:
905 unlock_page(newpage);
906 out_fallback:
907 cs->pg = buf->page;
908 cs->offset = buf->offset;
910 err = lock_request(cs->req);
911 if (err)
912 return err;
914 return 1;
917 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
918 unsigned offset, unsigned count)
920 struct pipe_buffer *buf;
921 int err;
923 if (cs->nr_segs == cs->pipe->buffers)
924 return -EIO;
926 err = unlock_request(cs->req);
927 if (err)
928 return err;
930 fuse_copy_finish(cs);
932 buf = cs->pipebufs;
933 get_page(page);
934 buf->page = page;
935 buf->offset = offset;
936 buf->len = count;
938 cs->pipebufs++;
939 cs->nr_segs++;
940 cs->len = 0;
942 return 0;
946 * Copy a page in the request to/from the userspace buffer. Must be
947 * done atomically
949 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
950 unsigned offset, unsigned count, int zeroing)
952 int err;
953 struct page *page = *pagep;
955 if (page && zeroing && count < PAGE_SIZE)
956 clear_highpage(page);
958 while (count) {
959 if (cs->write && cs->pipebufs && page) {
960 return fuse_ref_page(cs, page, offset, count);
961 } else if (!cs->len) {
962 if (cs->move_pages && page &&
963 offset == 0 && count == PAGE_SIZE) {
964 err = fuse_try_move_page(cs, pagep);
965 if (err <= 0)
966 return err;
967 } else {
968 err = fuse_copy_fill(cs);
969 if (err)
970 return err;
973 if (page) {
974 void *mapaddr = kmap_atomic(page);
975 void *buf = mapaddr + offset;
976 offset += fuse_copy_do(cs, &buf, &count);
977 kunmap_atomic(mapaddr);
978 } else
979 offset += fuse_copy_do(cs, NULL, &count);
981 if (page && !cs->write)
982 flush_dcache_page(page);
983 return 0;
986 /* Copy pages in the request to/from userspace buffer */
987 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
988 int zeroing)
990 unsigned i;
991 struct fuse_req *req = cs->req;
993 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
994 int err;
995 unsigned offset = req->page_descs[i].offset;
996 unsigned count = min(nbytes, req->page_descs[i].length);
998 err = fuse_copy_page(cs, &req->pages[i], offset, count,
999 zeroing);
1000 if (err)
1001 return err;
1003 nbytes -= count;
1005 return 0;
1008 /* Copy a single argument in the request to/from userspace buffer */
1009 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1011 while (size) {
1012 if (!cs->len) {
1013 int err = fuse_copy_fill(cs);
1014 if (err)
1015 return err;
1017 fuse_copy_do(cs, &val, &size);
1019 return 0;
1022 /* Copy request arguments to/from userspace buffer */
1023 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1024 unsigned argpages, struct fuse_arg *args,
1025 int zeroing)
1027 int err = 0;
1028 unsigned i;
1030 for (i = 0; !err && i < numargs; i++) {
1031 struct fuse_arg *arg = &args[i];
1032 if (i == numargs - 1 && argpages)
1033 err = fuse_copy_pages(cs, arg->size, zeroing);
1034 else
1035 err = fuse_copy_one(cs, arg->value, arg->size);
1037 return err;
1040 static int forget_pending(struct fuse_iqueue *fiq)
1042 return fiq->forget_list_head.next != NULL;
1045 static int request_pending(struct fuse_iqueue *fiq)
1047 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1048 forget_pending(fiq);
1052 * Transfer an interrupt request to userspace
1054 * Unlike other requests this is assembled on demand, without a need
1055 * to allocate a separate fuse_req structure.
1057 * Called with fiq->waitq.lock held, releases it
1059 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1060 struct fuse_copy_state *cs,
1061 size_t nbytes, struct fuse_req *req)
1062 __releases(fiq->waitq.lock)
1064 struct fuse_in_header ih;
1065 struct fuse_interrupt_in arg;
1066 unsigned reqsize = sizeof(ih) + sizeof(arg);
1067 int err;
1069 list_del_init(&req->intr_entry);
1070 req->intr_unique = fuse_get_unique(fiq);
1071 memset(&ih, 0, sizeof(ih));
1072 memset(&arg, 0, sizeof(arg));
1073 ih.len = reqsize;
1074 ih.opcode = FUSE_INTERRUPT;
1075 ih.unique = req->intr_unique;
1076 arg.unique = req->in.h.unique;
1078 spin_unlock(&fiq->waitq.lock);
1079 if (nbytes < reqsize)
1080 return -EINVAL;
1082 err = fuse_copy_one(cs, &ih, sizeof(ih));
1083 if (!err)
1084 err = fuse_copy_one(cs, &arg, sizeof(arg));
1085 fuse_copy_finish(cs);
1087 return err ? err : reqsize;
1090 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1091 unsigned max,
1092 unsigned *countp)
1094 struct fuse_forget_link *head = fiq->forget_list_head.next;
1095 struct fuse_forget_link **newhead = &head;
1096 unsigned count;
1098 for (count = 0; *newhead != NULL && count < max; count++)
1099 newhead = &(*newhead)->next;
1101 fiq->forget_list_head.next = *newhead;
1102 *newhead = NULL;
1103 if (fiq->forget_list_head.next == NULL)
1104 fiq->forget_list_tail = &fiq->forget_list_head;
1106 if (countp != NULL)
1107 *countp = count;
1109 return head;
1112 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1113 struct fuse_copy_state *cs,
1114 size_t nbytes)
1115 __releases(fiq->waitq.lock)
1117 int err;
1118 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1119 struct fuse_forget_in arg = {
1120 .nlookup = forget->forget_one.nlookup,
1122 struct fuse_in_header ih = {
1123 .opcode = FUSE_FORGET,
1124 .nodeid = forget->forget_one.nodeid,
1125 .unique = fuse_get_unique(fiq),
1126 .len = sizeof(ih) + sizeof(arg),
1129 spin_unlock(&fiq->waitq.lock);
1130 kfree(forget);
1131 if (nbytes < ih.len)
1132 return -EINVAL;
1134 err = fuse_copy_one(cs, &ih, sizeof(ih));
1135 if (!err)
1136 err = fuse_copy_one(cs, &arg, sizeof(arg));
1137 fuse_copy_finish(cs);
1139 if (err)
1140 return err;
1142 return ih.len;
1145 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1146 struct fuse_copy_state *cs, size_t nbytes)
1147 __releases(fiq->waitq.lock)
1149 int err;
1150 unsigned max_forgets;
1151 unsigned count;
1152 struct fuse_forget_link *head;
1153 struct fuse_batch_forget_in arg = { .count = 0 };
1154 struct fuse_in_header ih = {
1155 .opcode = FUSE_BATCH_FORGET,
1156 .unique = fuse_get_unique(fiq),
1157 .len = sizeof(ih) + sizeof(arg),
1160 if (nbytes < ih.len) {
1161 spin_unlock(&fiq->waitq.lock);
1162 return -EINVAL;
1165 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1166 head = dequeue_forget(fiq, max_forgets, &count);
1167 spin_unlock(&fiq->waitq.lock);
1169 arg.count = count;
1170 ih.len += count * sizeof(struct fuse_forget_one);
1171 err = fuse_copy_one(cs, &ih, sizeof(ih));
1172 if (!err)
1173 err = fuse_copy_one(cs, &arg, sizeof(arg));
1175 while (head) {
1176 struct fuse_forget_link *forget = head;
1178 if (!err) {
1179 err = fuse_copy_one(cs, &forget->forget_one,
1180 sizeof(forget->forget_one));
1182 head = forget->next;
1183 kfree(forget);
1186 fuse_copy_finish(cs);
1188 if (err)
1189 return err;
1191 return ih.len;
1194 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1195 struct fuse_copy_state *cs,
1196 size_t nbytes)
1197 __releases(fiq->waitq.lock)
1199 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1200 return fuse_read_single_forget(fiq, cs, nbytes);
1201 else
1202 return fuse_read_batch_forget(fiq, cs, nbytes);
1206 * Read a single request into the userspace filesystem's buffer. This
1207 * function waits until a request is available, then removes it from
1208 * the pending list and copies request data to userspace buffer. If
1209 * no reply is needed (FORGET) or request has been aborted or there
1210 * was an error during the copying then it's finished by calling
1211 * request_end(). Otherwise add it to the processing list, and set
1212 * the 'sent' flag.
1214 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1215 struct fuse_copy_state *cs, size_t nbytes)
1217 ssize_t err;
1218 struct fuse_conn *fc = fud->fc;
1219 struct fuse_iqueue *fiq = &fc->iq;
1220 struct fuse_pqueue *fpq = &fud->pq;
1221 struct fuse_req *req;
1222 struct fuse_in *in;
1223 unsigned reqsize;
1225 restart:
1226 spin_lock(&fiq->waitq.lock);
1227 err = -EAGAIN;
1228 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1229 !request_pending(fiq))
1230 goto err_unlock;
1232 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1233 !fiq->connected || request_pending(fiq));
1234 if (err)
1235 goto err_unlock;
1237 err = -ENODEV;
1238 if (!fiq->connected)
1239 goto err_unlock;
1241 if (!list_empty(&fiq->interrupts)) {
1242 req = list_entry(fiq->interrupts.next, struct fuse_req,
1243 intr_entry);
1244 return fuse_read_interrupt(fiq, cs, nbytes, req);
1247 if (forget_pending(fiq)) {
1248 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1249 return fuse_read_forget(fc, fiq, cs, nbytes);
1251 if (fiq->forget_batch <= -8)
1252 fiq->forget_batch = 16;
1255 req = list_entry(fiq->pending.next, struct fuse_req, list);
1256 clear_bit(FR_PENDING, &req->flags);
1257 list_del_init(&req->list);
1258 spin_unlock(&fiq->waitq.lock);
1260 in = &req->in;
1261 reqsize = in->h.len;
1262 /* If request is too large, reply with an error and restart the read */
1263 if (nbytes < reqsize) {
1264 req->out.h.error = -EIO;
1265 /* SETXATTR is special, since it may contain too large data */
1266 if (in->h.opcode == FUSE_SETXATTR)
1267 req->out.h.error = -E2BIG;
1268 request_end(fc, req);
1269 goto restart;
1271 spin_lock(&fpq->lock);
1272 list_add(&req->list, &fpq->io);
1273 spin_unlock(&fpq->lock);
1274 cs->req = req;
1275 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1276 if (!err)
1277 err = fuse_copy_args(cs, in->numargs, in->argpages,
1278 (struct fuse_arg *) in->args, 0);
1279 fuse_copy_finish(cs);
1280 spin_lock(&fpq->lock);
1281 clear_bit(FR_LOCKED, &req->flags);
1282 if (!fpq->connected) {
1283 err = -ENODEV;
1284 goto out_end;
1286 if (err) {
1287 req->out.h.error = -EIO;
1288 goto out_end;
1290 if (!test_bit(FR_ISREPLY, &req->flags)) {
1291 err = reqsize;
1292 goto out_end;
1294 list_move_tail(&req->list, &fpq->processing);
1295 spin_unlock(&fpq->lock);
1296 set_bit(FR_SENT, &req->flags);
1297 /* matches barrier in request_wait_answer() */
1298 smp_mb__after_atomic();
1299 if (test_bit(FR_INTERRUPTED, &req->flags))
1300 queue_interrupt(fiq, req);
1302 return reqsize;
1304 out_end:
1305 if (!test_bit(FR_PRIVATE, &req->flags))
1306 list_del_init(&req->list);
1307 spin_unlock(&fpq->lock);
1308 request_end(fc, req);
1309 return err;
1311 err_unlock:
1312 spin_unlock(&fiq->waitq.lock);
1313 return err;
1316 static int fuse_dev_open(struct inode *inode, struct file *file)
1319 * The fuse device's file's private_data is used to hold
1320 * the fuse_conn(ection) when it is mounted, and is used to
1321 * keep track of whether the file has been mounted already.
1323 file->private_data = NULL;
1324 return 0;
1327 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1329 struct fuse_copy_state cs;
1330 struct file *file = iocb->ki_filp;
1331 struct fuse_dev *fud = fuse_get_dev(file);
1333 if (!fud)
1334 return -EPERM;
1336 if (!iter_is_iovec(to))
1337 return -EINVAL;
1339 fuse_copy_init(&cs, 1, to);
1341 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1344 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1345 struct pipe_inode_info *pipe,
1346 size_t len, unsigned int flags)
1348 int total, ret;
1349 int page_nr = 0;
1350 struct pipe_buffer *bufs;
1351 struct fuse_copy_state cs;
1352 struct fuse_dev *fud = fuse_get_dev(in);
1354 if (!fud)
1355 return -EPERM;
1357 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1358 if (!bufs)
1359 return -ENOMEM;
1361 fuse_copy_init(&cs, 1, NULL);
1362 cs.pipebufs = bufs;
1363 cs.pipe = pipe;
1364 ret = fuse_dev_do_read(fud, in, &cs, len);
1365 if (ret < 0)
1366 goto out;
1368 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1369 ret = -EIO;
1370 goto out;
1373 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1375 * Need to be careful about this. Having buf->ops in module
1376 * code can Oops if the buffer persists after module unload.
1378 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1379 bufs[page_nr].flags = 0;
1380 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1381 if (unlikely(ret < 0))
1382 break;
1384 if (total)
1385 ret = total;
1386 out:
1387 for (; page_nr < cs.nr_segs; page_nr++)
1388 put_page(bufs[page_nr].page);
1390 kfree(bufs);
1391 return ret;
1394 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1395 struct fuse_copy_state *cs)
1397 struct fuse_notify_poll_wakeup_out outarg;
1398 int err = -EINVAL;
1400 if (size != sizeof(outarg))
1401 goto err;
1403 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1404 if (err)
1405 goto err;
1407 fuse_copy_finish(cs);
1408 return fuse_notify_poll_wakeup(fc, &outarg);
1410 err:
1411 fuse_copy_finish(cs);
1412 return err;
1415 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1416 struct fuse_copy_state *cs)
1418 struct fuse_notify_inval_inode_out outarg;
1419 int err = -EINVAL;
1421 if (size != sizeof(outarg))
1422 goto err;
1424 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1425 if (err)
1426 goto err;
1427 fuse_copy_finish(cs);
1429 down_read(&fc->killsb);
1430 err = -ENOENT;
1431 if (fc->sb) {
1432 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1433 outarg.off, outarg.len);
1435 up_read(&fc->killsb);
1436 return err;
1438 err:
1439 fuse_copy_finish(cs);
1440 return err;
1443 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1444 struct fuse_copy_state *cs)
1446 struct fuse_notify_inval_entry_out outarg;
1447 int err = -ENOMEM;
1448 char *buf;
1449 struct qstr name;
1451 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1452 if (!buf)
1453 goto err;
1455 err = -EINVAL;
1456 if (size < sizeof(outarg))
1457 goto err;
1459 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1460 if (err)
1461 goto err;
1463 err = -ENAMETOOLONG;
1464 if (outarg.namelen > FUSE_NAME_MAX)
1465 goto err;
1467 err = -EINVAL;
1468 if (size != sizeof(outarg) + outarg.namelen + 1)
1469 goto err;
1471 name.name = buf;
1472 name.len = outarg.namelen;
1473 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1474 if (err)
1475 goto err;
1476 fuse_copy_finish(cs);
1477 buf[outarg.namelen] = 0;
1479 down_read(&fc->killsb);
1480 err = -ENOENT;
1481 if (fc->sb)
1482 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1483 up_read(&fc->killsb);
1484 kfree(buf);
1485 return err;
1487 err:
1488 kfree(buf);
1489 fuse_copy_finish(cs);
1490 return err;
1493 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1494 struct fuse_copy_state *cs)
1496 struct fuse_notify_delete_out outarg;
1497 int err = -ENOMEM;
1498 char *buf;
1499 struct qstr name;
1501 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1502 if (!buf)
1503 goto err;
1505 err = -EINVAL;
1506 if (size < sizeof(outarg))
1507 goto err;
1509 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1510 if (err)
1511 goto err;
1513 err = -ENAMETOOLONG;
1514 if (outarg.namelen > FUSE_NAME_MAX)
1515 goto err;
1517 err = -EINVAL;
1518 if (size != sizeof(outarg) + outarg.namelen + 1)
1519 goto err;
1521 name.name = buf;
1522 name.len = outarg.namelen;
1523 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1524 if (err)
1525 goto err;
1526 fuse_copy_finish(cs);
1527 buf[outarg.namelen] = 0;
1529 down_read(&fc->killsb);
1530 err = -ENOENT;
1531 if (fc->sb)
1532 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1533 outarg.child, &name);
1534 up_read(&fc->killsb);
1535 kfree(buf);
1536 return err;
1538 err:
1539 kfree(buf);
1540 fuse_copy_finish(cs);
1541 return err;
1544 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1545 struct fuse_copy_state *cs)
1547 struct fuse_notify_store_out outarg;
1548 struct inode *inode;
1549 struct address_space *mapping;
1550 u64 nodeid;
1551 int err;
1552 pgoff_t index;
1553 unsigned int offset;
1554 unsigned int num;
1555 loff_t file_size;
1556 loff_t end;
1558 err = -EINVAL;
1559 if (size < sizeof(outarg))
1560 goto out_finish;
1562 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1563 if (err)
1564 goto out_finish;
1566 err = -EINVAL;
1567 if (size - sizeof(outarg) != outarg.size)
1568 goto out_finish;
1570 nodeid = outarg.nodeid;
1572 down_read(&fc->killsb);
1574 err = -ENOENT;
1575 if (!fc->sb)
1576 goto out_up_killsb;
1578 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1579 if (!inode)
1580 goto out_up_killsb;
1582 mapping = inode->i_mapping;
1583 index = outarg.offset >> PAGE_SHIFT;
1584 offset = outarg.offset & ~PAGE_MASK;
1585 file_size = i_size_read(inode);
1586 end = outarg.offset + outarg.size;
1587 if (end > file_size) {
1588 file_size = end;
1589 fuse_write_update_size(inode, file_size);
1592 num = outarg.size;
1593 while (num) {
1594 struct page *page;
1595 unsigned int this_num;
1597 err = -ENOMEM;
1598 page = find_or_create_page(mapping, index,
1599 mapping_gfp_mask(mapping));
1600 if (!page)
1601 goto out_iput;
1603 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1604 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1605 if (!err && offset == 0 &&
1606 (this_num == PAGE_SIZE || file_size == end))
1607 SetPageUptodate(page);
1608 unlock_page(page);
1609 put_page(page);
1611 if (err)
1612 goto out_iput;
1614 num -= this_num;
1615 offset = 0;
1616 index++;
1619 err = 0;
1621 out_iput:
1622 iput(inode);
1623 out_up_killsb:
1624 up_read(&fc->killsb);
1625 out_finish:
1626 fuse_copy_finish(cs);
1627 return err;
1630 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1632 release_pages(req->pages, req->num_pages, false);
1635 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1636 struct fuse_notify_retrieve_out *outarg)
1638 int err;
1639 struct address_space *mapping = inode->i_mapping;
1640 struct fuse_req *req;
1641 pgoff_t index;
1642 loff_t file_size;
1643 unsigned int num;
1644 unsigned int offset;
1645 size_t total_len = 0;
1646 int num_pages;
1648 offset = outarg->offset & ~PAGE_MASK;
1649 file_size = i_size_read(inode);
1651 num = outarg->size;
1652 if (outarg->offset > file_size)
1653 num = 0;
1654 else if (outarg->offset + num > file_size)
1655 num = file_size - outarg->offset;
1657 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1658 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1660 req = fuse_get_req(fc, num_pages);
1661 if (IS_ERR(req))
1662 return PTR_ERR(req);
1664 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1665 req->in.h.nodeid = outarg->nodeid;
1666 req->in.numargs = 2;
1667 req->in.argpages = 1;
1668 req->page_descs[0].offset = offset;
1669 req->end = fuse_retrieve_end;
1671 index = outarg->offset >> PAGE_SHIFT;
1673 while (num && req->num_pages < num_pages) {
1674 struct page *page;
1675 unsigned int this_num;
1677 page = find_get_page(mapping, index);
1678 if (!page)
1679 break;
1681 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1682 req->pages[req->num_pages] = page;
1683 req->page_descs[req->num_pages].length = this_num;
1684 req->num_pages++;
1686 offset = 0;
1687 num -= this_num;
1688 total_len += this_num;
1689 index++;
1691 req->misc.retrieve_in.offset = outarg->offset;
1692 req->misc.retrieve_in.size = total_len;
1693 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1694 req->in.args[0].value = &req->misc.retrieve_in;
1695 req->in.args[1].size = total_len;
1697 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1698 if (err)
1699 fuse_retrieve_end(fc, req);
1701 return err;
1704 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1705 struct fuse_copy_state *cs)
1707 struct fuse_notify_retrieve_out outarg;
1708 struct inode *inode;
1709 int err;
1711 err = -EINVAL;
1712 if (size != sizeof(outarg))
1713 goto copy_finish;
1715 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1716 if (err)
1717 goto copy_finish;
1719 fuse_copy_finish(cs);
1721 down_read(&fc->killsb);
1722 err = -ENOENT;
1723 if (fc->sb) {
1724 u64 nodeid = outarg.nodeid;
1726 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1727 if (inode) {
1728 err = fuse_retrieve(fc, inode, &outarg);
1729 iput(inode);
1732 up_read(&fc->killsb);
1734 return err;
1736 copy_finish:
1737 fuse_copy_finish(cs);
1738 return err;
1741 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1742 unsigned int size, struct fuse_copy_state *cs)
1744 /* Don't try to move pages (yet) */
1745 cs->move_pages = 0;
1747 switch (code) {
1748 case FUSE_NOTIFY_POLL:
1749 return fuse_notify_poll(fc, size, cs);
1751 case FUSE_NOTIFY_INVAL_INODE:
1752 return fuse_notify_inval_inode(fc, size, cs);
1754 case FUSE_NOTIFY_INVAL_ENTRY:
1755 return fuse_notify_inval_entry(fc, size, cs);
1757 case FUSE_NOTIFY_STORE:
1758 return fuse_notify_store(fc, size, cs);
1760 case FUSE_NOTIFY_RETRIEVE:
1761 return fuse_notify_retrieve(fc, size, cs);
1763 case FUSE_NOTIFY_DELETE:
1764 return fuse_notify_delete(fc, size, cs);
1766 default:
1767 fuse_copy_finish(cs);
1768 return -EINVAL;
1772 /* Look up request on processing list by unique ID */
1773 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1775 struct fuse_req *req;
1777 list_for_each_entry(req, &fpq->processing, list) {
1778 if (req->in.h.unique == unique || req->intr_unique == unique)
1779 return req;
1781 return NULL;
1784 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1785 unsigned nbytes)
1787 unsigned reqsize = sizeof(struct fuse_out_header);
1789 if (out->h.error)
1790 return nbytes != reqsize ? -EINVAL : 0;
1792 reqsize += len_args(out->numargs, out->args);
1794 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1795 return -EINVAL;
1796 else if (reqsize > nbytes) {
1797 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1798 unsigned diffsize = reqsize - nbytes;
1799 if (diffsize > lastarg->size)
1800 return -EINVAL;
1801 lastarg->size -= diffsize;
1803 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1804 out->page_zeroing);
1808 * Write a single reply to a request. First the header is copied from
1809 * the write buffer. The request is then searched on the processing
1810 * list by the unique ID found in the header. If found, then remove
1811 * it from the list and copy the rest of the buffer to the request.
1812 * The request is finished by calling request_end()
1814 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1815 struct fuse_copy_state *cs, size_t nbytes)
1817 int err;
1818 struct fuse_conn *fc = fud->fc;
1819 struct fuse_pqueue *fpq = &fud->pq;
1820 struct fuse_req *req;
1821 struct fuse_out_header oh;
1823 if (nbytes < sizeof(struct fuse_out_header))
1824 return -EINVAL;
1826 err = fuse_copy_one(cs, &oh, sizeof(oh));
1827 if (err)
1828 goto err_finish;
1830 err = -EINVAL;
1831 if (oh.len != nbytes)
1832 goto err_finish;
1835 * Zero oh.unique indicates unsolicited notification message
1836 * and error contains notification code.
1838 if (!oh.unique) {
1839 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1840 return err ? err : nbytes;
1843 err = -EINVAL;
1844 if (oh.error <= -1000 || oh.error > 0)
1845 goto err_finish;
1847 spin_lock(&fpq->lock);
1848 err = -ENOENT;
1849 if (!fpq->connected)
1850 goto err_unlock_pq;
1852 req = request_find(fpq, oh.unique);
1853 if (!req)
1854 goto err_unlock_pq;
1856 /* Is it an interrupt reply? */
1857 if (req->intr_unique == oh.unique) {
1858 spin_unlock(&fpq->lock);
1860 err = -EINVAL;
1861 if (nbytes != sizeof(struct fuse_out_header))
1862 goto err_finish;
1864 if (oh.error == -ENOSYS)
1865 fc->no_interrupt = 1;
1866 else if (oh.error == -EAGAIN)
1867 queue_interrupt(&fc->iq, req);
1869 fuse_copy_finish(cs);
1870 return nbytes;
1873 clear_bit(FR_SENT, &req->flags);
1874 list_move(&req->list, &fpq->io);
1875 req->out.h = oh;
1876 set_bit(FR_LOCKED, &req->flags);
1877 spin_unlock(&fpq->lock);
1878 cs->req = req;
1879 if (!req->out.page_replace)
1880 cs->move_pages = 0;
1882 err = copy_out_args(cs, &req->out, nbytes);
1883 fuse_copy_finish(cs);
1885 spin_lock(&fpq->lock);
1886 clear_bit(FR_LOCKED, &req->flags);
1887 if (!fpq->connected)
1888 err = -ENOENT;
1889 else if (err)
1890 req->out.h.error = -EIO;
1891 if (!test_bit(FR_PRIVATE, &req->flags))
1892 list_del_init(&req->list);
1893 spin_unlock(&fpq->lock);
1895 request_end(fc, req);
1897 return err ? err : nbytes;
1899 err_unlock_pq:
1900 spin_unlock(&fpq->lock);
1901 err_finish:
1902 fuse_copy_finish(cs);
1903 return err;
1906 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1908 struct fuse_copy_state cs;
1909 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1911 if (!fud)
1912 return -EPERM;
1914 if (!iter_is_iovec(from))
1915 return -EINVAL;
1917 fuse_copy_init(&cs, 0, from);
1919 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1922 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1923 struct file *out, loff_t *ppos,
1924 size_t len, unsigned int flags)
1926 unsigned nbuf;
1927 unsigned idx;
1928 struct pipe_buffer *bufs;
1929 struct fuse_copy_state cs;
1930 struct fuse_dev *fud;
1931 size_t rem;
1932 ssize_t ret;
1934 fud = fuse_get_dev(out);
1935 if (!fud)
1936 return -EPERM;
1938 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1939 if (!bufs)
1940 return -ENOMEM;
1942 pipe_lock(pipe);
1943 nbuf = 0;
1944 rem = 0;
1945 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1946 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1948 ret = -EINVAL;
1949 if (rem < len) {
1950 pipe_unlock(pipe);
1951 goto out;
1954 rem = len;
1955 while (rem) {
1956 struct pipe_buffer *ibuf;
1957 struct pipe_buffer *obuf;
1959 BUG_ON(nbuf >= pipe->buffers);
1960 BUG_ON(!pipe->nrbufs);
1961 ibuf = &pipe->bufs[pipe->curbuf];
1962 obuf = &bufs[nbuf];
1964 if (rem >= ibuf->len) {
1965 *obuf = *ibuf;
1966 ibuf->ops = NULL;
1967 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1968 pipe->nrbufs--;
1969 } else {
1970 pipe_buf_get(pipe, ibuf);
1971 *obuf = *ibuf;
1972 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1973 obuf->len = rem;
1974 ibuf->offset += obuf->len;
1975 ibuf->len -= obuf->len;
1977 nbuf++;
1978 rem -= obuf->len;
1980 pipe_unlock(pipe);
1982 fuse_copy_init(&cs, 0, NULL);
1983 cs.pipebufs = bufs;
1984 cs.nr_segs = nbuf;
1985 cs.pipe = pipe;
1987 if (flags & SPLICE_F_MOVE)
1988 cs.move_pages = 1;
1990 ret = fuse_dev_do_write(fud, &cs, len);
1992 for (idx = 0; idx < nbuf; idx++)
1993 pipe_buf_release(pipe, &bufs[idx]);
1995 out:
1996 kfree(bufs);
1997 return ret;
2000 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2002 unsigned mask = POLLOUT | POLLWRNORM;
2003 struct fuse_iqueue *fiq;
2004 struct fuse_dev *fud = fuse_get_dev(file);
2006 if (!fud)
2007 return POLLERR;
2009 fiq = &fud->fc->iq;
2010 poll_wait(file, &fiq->waitq, wait);
2012 spin_lock(&fiq->waitq.lock);
2013 if (!fiq->connected)
2014 mask = POLLERR;
2015 else if (request_pending(fiq))
2016 mask |= POLLIN | POLLRDNORM;
2017 spin_unlock(&fiq->waitq.lock);
2019 return mask;
2023 * Abort all requests on the given list (pending or processing)
2025 * This function releases and reacquires fc->lock
2027 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2029 while (!list_empty(head)) {
2030 struct fuse_req *req;
2031 req = list_entry(head->next, struct fuse_req, list);
2032 req->out.h.error = -ECONNABORTED;
2033 clear_bit(FR_SENT, &req->flags);
2034 list_del_init(&req->list);
2035 request_end(fc, req);
2039 static void end_polls(struct fuse_conn *fc)
2041 struct rb_node *p;
2043 p = rb_first(&fc->polled_files);
2045 while (p) {
2046 struct fuse_file *ff;
2047 ff = rb_entry(p, struct fuse_file, polled_node);
2048 wake_up_interruptible_all(&ff->poll_wait);
2050 p = rb_next(p);
2055 * Abort all requests.
2057 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2058 * filesystem.
2060 * The same effect is usually achievable through killing the filesystem daemon
2061 * and all users of the filesystem. The exception is the combination of an
2062 * asynchronous request and the tricky deadlock (see
2063 * Documentation/filesystems/fuse.txt).
2065 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2066 * requests, they should be finished off immediately. Locked requests will be
2067 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2068 * requests. It is possible that some request will finish before we can. This
2069 * is OK, the request will in that case be removed from the list before we touch
2070 * it.
2072 void fuse_abort_conn(struct fuse_conn *fc)
2074 struct fuse_iqueue *fiq = &fc->iq;
2076 spin_lock(&fc->lock);
2077 if (fc->connected) {
2078 struct fuse_dev *fud;
2079 struct fuse_req *req, *next;
2080 LIST_HEAD(to_end1);
2081 LIST_HEAD(to_end2);
2083 fc->connected = 0;
2084 fc->blocked = 0;
2085 fuse_set_initialized(fc);
2086 list_for_each_entry(fud, &fc->devices, entry) {
2087 struct fuse_pqueue *fpq = &fud->pq;
2089 spin_lock(&fpq->lock);
2090 fpq->connected = 0;
2091 list_for_each_entry_safe(req, next, &fpq->io, list) {
2092 req->out.h.error = -ECONNABORTED;
2093 spin_lock(&req->waitq.lock);
2094 set_bit(FR_ABORTED, &req->flags);
2095 if (!test_bit(FR_LOCKED, &req->flags)) {
2096 set_bit(FR_PRIVATE, &req->flags);
2097 list_move(&req->list, &to_end1);
2099 spin_unlock(&req->waitq.lock);
2101 list_splice_init(&fpq->processing, &to_end2);
2102 spin_unlock(&fpq->lock);
2104 fc->max_background = UINT_MAX;
2105 flush_bg_queue(fc);
2107 spin_lock(&fiq->waitq.lock);
2108 fiq->connected = 0;
2109 list_splice_init(&fiq->pending, &to_end2);
2110 list_for_each_entry(req, &to_end2, list)
2111 clear_bit(FR_PENDING, &req->flags);
2112 while (forget_pending(fiq))
2113 kfree(dequeue_forget(fiq, 1, NULL));
2114 wake_up_all_locked(&fiq->waitq);
2115 spin_unlock(&fiq->waitq.lock);
2116 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2117 end_polls(fc);
2118 wake_up_all(&fc->blocked_waitq);
2119 spin_unlock(&fc->lock);
2121 while (!list_empty(&to_end1)) {
2122 req = list_first_entry(&to_end1, struct fuse_req, list);
2123 __fuse_get_request(req);
2124 list_del_init(&req->list);
2125 request_end(fc, req);
2127 end_requests(fc, &to_end2);
2128 } else {
2129 spin_unlock(&fc->lock);
2132 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2134 int fuse_dev_release(struct inode *inode, struct file *file)
2136 struct fuse_dev *fud = fuse_get_dev(file);
2138 if (fud) {
2139 struct fuse_conn *fc = fud->fc;
2140 struct fuse_pqueue *fpq = &fud->pq;
2142 WARN_ON(!list_empty(&fpq->io));
2143 end_requests(fc, &fpq->processing);
2144 /* Are we the last open device? */
2145 if (atomic_dec_and_test(&fc->dev_count)) {
2146 WARN_ON(fc->iq.fasync != NULL);
2147 fuse_abort_conn(fc);
2149 fuse_dev_free(fud);
2151 return 0;
2153 EXPORT_SYMBOL_GPL(fuse_dev_release);
2155 static int fuse_dev_fasync(int fd, struct file *file, int on)
2157 struct fuse_dev *fud = fuse_get_dev(file);
2159 if (!fud)
2160 return -EPERM;
2162 /* No locking - fasync_helper does its own locking */
2163 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2166 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2168 struct fuse_dev *fud;
2170 if (new->private_data)
2171 return -EINVAL;
2173 fud = fuse_dev_alloc(fc);
2174 if (!fud)
2175 return -ENOMEM;
2177 new->private_data = fud;
2178 atomic_inc(&fc->dev_count);
2180 return 0;
2183 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2184 unsigned long arg)
2186 int err = -ENOTTY;
2188 if (cmd == FUSE_DEV_IOC_CLONE) {
2189 int oldfd;
2191 err = -EFAULT;
2192 if (!get_user(oldfd, (__u32 __user *) arg)) {
2193 struct file *old = fget(oldfd);
2195 err = -EINVAL;
2196 if (old) {
2197 struct fuse_dev *fud = NULL;
2200 * Check against file->f_op because CUSE
2201 * uses the same ioctl handler.
2203 if (old->f_op == file->f_op &&
2204 old->f_cred->user_ns == file->f_cred->user_ns)
2205 fud = fuse_get_dev(old);
2207 if (fud) {
2208 mutex_lock(&fuse_mutex);
2209 err = fuse_device_clone(fud->fc, file);
2210 mutex_unlock(&fuse_mutex);
2212 fput(old);
2216 return err;
2219 const struct file_operations fuse_dev_operations = {
2220 .owner = THIS_MODULE,
2221 .open = fuse_dev_open,
2222 .llseek = no_llseek,
2223 .read_iter = fuse_dev_read,
2224 .splice_read = fuse_dev_splice_read,
2225 .write_iter = 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,
2230 .unlocked_ioctl = fuse_dev_ioctl,
2231 .compat_ioctl = fuse_dev_ioctl,
2233 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2235 static struct miscdevice fuse_miscdevice = {
2236 .minor = FUSE_MINOR,
2237 .name = "fuse",
2238 .fops = &fuse_dev_operations,
2241 int __init fuse_dev_init(void)
2243 int err = -ENOMEM;
2244 fuse_req_cachep = kmem_cache_create("fuse_request",
2245 sizeof(struct fuse_req),
2246 0, 0, NULL);
2247 if (!fuse_req_cachep)
2248 goto out;
2250 err = misc_register(&fuse_miscdevice);
2251 if (err)
2252 goto out_cache_clean;
2254 return 0;
2256 out_cache_clean:
2257 kmem_cache_destroy(fuse_req_cachep);
2258 out:
2259 return err;
2262 void fuse_dev_cleanup(void)
2264 misc_deregister(&fuse_miscdevice);
2265 kmem_cache_destroy(fuse_req_cachep);