zram: fix unbalanced idr management at hot removal
[linux/fpc-iii.git] / fs / fuse / dev.c
blob70ea57c7b6bb2b48ecf47d8cc9bd211cfc430a61
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 (list_empty(&req->intr_entry)) {
403 list_add_tail(&req->intr_entry, &fiq->interrupts);
404 wake_up_locked(&fiq->waitq);
406 spin_unlock(&fiq->waitq.lock);
407 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
410 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
412 struct fuse_iqueue *fiq = &fc->iq;
413 int err;
415 if (!fc->no_interrupt) {
416 /* Any signal may interrupt this */
417 err = wait_event_interruptible(req->waitq,
418 test_bit(FR_FINISHED, &req->flags));
419 if (!err)
420 return;
422 set_bit(FR_INTERRUPTED, &req->flags);
423 /* matches barrier in fuse_dev_do_read() */
424 smp_mb__after_atomic();
425 if (test_bit(FR_SENT, &req->flags))
426 queue_interrupt(fiq, req);
429 if (!test_bit(FR_FORCE, &req->flags)) {
430 /* Only fatal signals may interrupt this */
431 err = wait_event_killable(req->waitq,
432 test_bit(FR_FINISHED, &req->flags));
433 if (!err)
434 return;
436 spin_lock(&fiq->waitq.lock);
437 /* Request is not yet in userspace, bail out */
438 if (test_bit(FR_PENDING, &req->flags)) {
439 list_del(&req->list);
440 spin_unlock(&fiq->waitq.lock);
441 __fuse_put_request(req);
442 req->out.h.error = -EINTR;
443 return;
445 spin_unlock(&fiq->waitq.lock);
449 * Either request is already in userspace, or it was forced.
450 * Wait it out.
452 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
455 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
457 struct fuse_iqueue *fiq = &fc->iq;
459 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
460 spin_lock(&fiq->waitq.lock);
461 if (!fiq->connected) {
462 spin_unlock(&fiq->waitq.lock);
463 req->out.h.error = -ENOTCONN;
464 } else {
465 req->in.h.unique = fuse_get_unique(fiq);
466 queue_request(fiq, req);
467 /* acquire extra reference, since request is still needed
468 after request_end() */
469 __fuse_get_request(req);
470 spin_unlock(&fiq->waitq.lock);
472 request_wait_answer(fc, req);
473 /* Pairs with smp_wmb() in request_end() */
474 smp_rmb();
478 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
480 __set_bit(FR_ISREPLY, &req->flags);
481 if (!test_bit(FR_WAITING, &req->flags)) {
482 __set_bit(FR_WAITING, &req->flags);
483 atomic_inc(&fc->num_waiting);
485 __fuse_request_send(fc, req);
487 EXPORT_SYMBOL_GPL(fuse_request_send);
489 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
491 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
492 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
494 if (fc->minor < 9) {
495 switch (args->in.h.opcode) {
496 case FUSE_LOOKUP:
497 case FUSE_CREATE:
498 case FUSE_MKNOD:
499 case FUSE_MKDIR:
500 case FUSE_SYMLINK:
501 case FUSE_LINK:
502 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
503 break;
504 case FUSE_GETATTR:
505 case FUSE_SETATTR:
506 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
507 break;
510 if (fc->minor < 12) {
511 switch (args->in.h.opcode) {
512 case FUSE_CREATE:
513 args->in.args[0].size = sizeof(struct fuse_open_in);
514 break;
515 case FUSE_MKNOD:
516 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
517 break;
522 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
524 struct fuse_req *req;
525 ssize_t ret;
527 req = fuse_get_req(fc, 0);
528 if (IS_ERR(req))
529 return PTR_ERR(req);
531 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
532 fuse_adjust_compat(fc, args);
534 req->in.h.opcode = args->in.h.opcode;
535 req->in.h.nodeid = args->in.h.nodeid;
536 req->in.numargs = args->in.numargs;
537 memcpy(req->in.args, args->in.args,
538 args->in.numargs * sizeof(struct fuse_in_arg));
539 req->out.argvar = args->out.argvar;
540 req->out.numargs = args->out.numargs;
541 memcpy(req->out.args, args->out.args,
542 args->out.numargs * sizeof(struct fuse_arg));
543 fuse_request_send(fc, req);
544 ret = req->out.h.error;
545 if (!ret && args->out.argvar) {
546 BUG_ON(args->out.numargs != 1);
547 ret = req->out.args[0].size;
549 fuse_put_request(fc, req);
551 return ret;
555 * Called under fc->lock
557 * fc->connected must have been checked previously
559 void fuse_request_send_background_locked(struct fuse_conn *fc,
560 struct fuse_req *req)
562 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
563 if (!test_bit(FR_WAITING, &req->flags)) {
564 __set_bit(FR_WAITING, &req->flags);
565 atomic_inc(&fc->num_waiting);
567 __set_bit(FR_ISREPLY, &req->flags);
568 fc->num_background++;
569 if (fc->num_background == fc->max_background)
570 fc->blocked = 1;
571 if (fc->num_background == fc->congestion_threshold &&
572 fc->bdi_initialized) {
573 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
574 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
576 list_add_tail(&req->list, &fc->bg_queue);
577 flush_bg_queue(fc);
580 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
582 BUG_ON(!req->end);
583 spin_lock(&fc->lock);
584 if (fc->connected) {
585 fuse_request_send_background_locked(fc, req);
586 spin_unlock(&fc->lock);
587 } else {
588 spin_unlock(&fc->lock);
589 req->out.h.error = -ENOTCONN;
590 req->end(fc, req);
591 fuse_put_request(fc, req);
594 EXPORT_SYMBOL_GPL(fuse_request_send_background);
596 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
597 struct fuse_req *req, u64 unique)
599 int err = -ENODEV;
600 struct fuse_iqueue *fiq = &fc->iq;
602 __clear_bit(FR_ISREPLY, &req->flags);
603 req->in.h.unique = unique;
604 spin_lock(&fiq->waitq.lock);
605 if (fiq->connected) {
606 queue_request(fiq, req);
607 err = 0;
609 spin_unlock(&fiq->waitq.lock);
611 return err;
614 void fuse_force_forget(struct file *file, u64 nodeid)
616 struct inode *inode = file_inode(file);
617 struct fuse_conn *fc = get_fuse_conn(inode);
618 struct fuse_req *req;
619 struct fuse_forget_in inarg;
621 memset(&inarg, 0, sizeof(inarg));
622 inarg.nlookup = 1;
623 req = fuse_get_req_nofail_nopages(fc, file);
624 req->in.h.opcode = FUSE_FORGET;
625 req->in.h.nodeid = nodeid;
626 req->in.numargs = 1;
627 req->in.args[0].size = sizeof(inarg);
628 req->in.args[0].value = &inarg;
629 __clear_bit(FR_ISREPLY, &req->flags);
630 __fuse_request_send(fc, req);
631 /* ignore errors */
632 fuse_put_request(fc, req);
636 * Lock the request. Up to the next unlock_request() there mustn't be
637 * anything that could cause a page-fault. If the request was already
638 * aborted bail out.
640 static int lock_request(struct fuse_req *req)
642 int err = 0;
643 if (req) {
644 spin_lock(&req->waitq.lock);
645 if (test_bit(FR_ABORTED, &req->flags))
646 err = -ENOENT;
647 else
648 set_bit(FR_LOCKED, &req->flags);
649 spin_unlock(&req->waitq.lock);
651 return err;
655 * Unlock request. If it was aborted while locked, caller is responsible
656 * for unlocking and ending the request.
658 static int unlock_request(struct fuse_req *req)
660 int err = 0;
661 if (req) {
662 spin_lock(&req->waitq.lock);
663 if (test_bit(FR_ABORTED, &req->flags))
664 err = -ENOENT;
665 else
666 clear_bit(FR_LOCKED, &req->flags);
667 spin_unlock(&req->waitq.lock);
669 return err;
672 struct fuse_copy_state {
673 int write;
674 struct fuse_req *req;
675 struct iov_iter *iter;
676 struct pipe_buffer *pipebufs;
677 struct pipe_buffer *currbuf;
678 struct pipe_inode_info *pipe;
679 unsigned long nr_segs;
680 struct page *pg;
681 unsigned len;
682 unsigned offset;
683 unsigned move_pages:1;
686 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
687 struct iov_iter *iter)
689 memset(cs, 0, sizeof(*cs));
690 cs->write = write;
691 cs->iter = iter;
694 /* Unmap and put previous page of userspace buffer */
695 static void fuse_copy_finish(struct fuse_copy_state *cs)
697 if (cs->currbuf) {
698 struct pipe_buffer *buf = cs->currbuf;
700 if (cs->write)
701 buf->len = PAGE_SIZE - cs->len;
702 cs->currbuf = NULL;
703 } else if (cs->pg) {
704 if (cs->write) {
705 flush_dcache_page(cs->pg);
706 set_page_dirty_lock(cs->pg);
708 put_page(cs->pg);
710 cs->pg = NULL;
714 * Get another pagefull of userspace buffer, and map it to kernel
715 * address space, and lock request
717 static int fuse_copy_fill(struct fuse_copy_state *cs)
719 struct page *page;
720 int err;
722 err = unlock_request(cs->req);
723 if (err)
724 return err;
726 fuse_copy_finish(cs);
727 if (cs->pipebufs) {
728 struct pipe_buffer *buf = cs->pipebufs;
730 if (!cs->write) {
731 err = pipe_buf_confirm(cs->pipe, buf);
732 if (err)
733 return err;
735 BUG_ON(!cs->nr_segs);
736 cs->currbuf = buf;
737 cs->pg = buf->page;
738 cs->offset = buf->offset;
739 cs->len = buf->len;
740 cs->pipebufs++;
741 cs->nr_segs--;
742 } else {
743 if (cs->nr_segs == cs->pipe->buffers)
744 return -EIO;
746 page = alloc_page(GFP_HIGHUSER);
747 if (!page)
748 return -ENOMEM;
750 buf->page = page;
751 buf->offset = 0;
752 buf->len = 0;
754 cs->currbuf = buf;
755 cs->pg = page;
756 cs->offset = 0;
757 cs->len = PAGE_SIZE;
758 cs->pipebufs++;
759 cs->nr_segs++;
761 } else {
762 size_t off;
763 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
764 if (err < 0)
765 return err;
766 BUG_ON(!err);
767 cs->len = err;
768 cs->offset = off;
769 cs->pg = page;
770 iov_iter_advance(cs->iter, err);
773 return lock_request(cs->req);
776 /* Do as much copy to/from userspace buffer as we can */
777 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
779 unsigned ncpy = min(*size, cs->len);
780 if (val) {
781 void *pgaddr = kmap_atomic(cs->pg);
782 void *buf = pgaddr + cs->offset;
784 if (cs->write)
785 memcpy(buf, *val, ncpy);
786 else
787 memcpy(*val, buf, ncpy);
789 kunmap_atomic(pgaddr);
790 *val += ncpy;
792 *size -= ncpy;
793 cs->len -= ncpy;
794 cs->offset += ncpy;
795 return ncpy;
798 static int fuse_check_page(struct page *page)
800 if (page_mapcount(page) ||
801 page->mapping != NULL ||
802 page_count(page) != 1 ||
803 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
804 ~(1 << PG_locked |
805 1 << PG_referenced |
806 1 << PG_uptodate |
807 1 << PG_lru |
808 1 << PG_active |
809 1 << PG_reclaim))) {
810 printk(KERN_WARNING "fuse: trying to steal weird page\n");
811 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);
812 return 1;
814 return 0;
817 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
819 int err;
820 struct page *oldpage = *pagep;
821 struct page *newpage;
822 struct pipe_buffer *buf = cs->pipebufs;
824 err = unlock_request(cs->req);
825 if (err)
826 return err;
828 fuse_copy_finish(cs);
830 err = pipe_buf_confirm(cs->pipe, buf);
831 if (err)
832 return err;
834 BUG_ON(!cs->nr_segs);
835 cs->currbuf = buf;
836 cs->len = buf->len;
837 cs->pipebufs++;
838 cs->nr_segs--;
840 if (cs->len != PAGE_SIZE)
841 goto out_fallback;
843 if (pipe_buf_steal(cs->pipe, buf) != 0)
844 goto out_fallback;
846 newpage = buf->page;
848 if (!PageUptodate(newpage))
849 SetPageUptodate(newpage);
851 ClearPageMappedToDisk(newpage);
853 if (fuse_check_page(newpage) != 0)
854 goto out_fallback_unlock;
857 * This is a new and locked page, it shouldn't be mapped or
858 * have any special flags on it
860 if (WARN_ON(page_mapped(oldpage)))
861 goto out_fallback_unlock;
862 if (WARN_ON(page_has_private(oldpage)))
863 goto out_fallback_unlock;
864 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
865 goto out_fallback_unlock;
866 if (WARN_ON(PageMlocked(oldpage)))
867 goto out_fallback_unlock;
869 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
870 if (err) {
871 unlock_page(newpage);
872 return err;
875 get_page(newpage);
877 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
878 lru_cache_add_file(newpage);
880 err = 0;
881 spin_lock(&cs->req->waitq.lock);
882 if (test_bit(FR_ABORTED, &cs->req->flags))
883 err = -ENOENT;
884 else
885 *pagep = newpage;
886 spin_unlock(&cs->req->waitq.lock);
888 if (err) {
889 unlock_page(newpage);
890 put_page(newpage);
891 return err;
894 unlock_page(oldpage);
895 put_page(oldpage);
896 cs->len = 0;
898 return 0;
900 out_fallback_unlock:
901 unlock_page(newpage);
902 out_fallback:
903 cs->pg = buf->page;
904 cs->offset = buf->offset;
906 err = lock_request(cs->req);
907 if (err)
908 return err;
910 return 1;
913 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
914 unsigned offset, unsigned count)
916 struct pipe_buffer *buf;
917 int err;
919 if (cs->nr_segs == cs->pipe->buffers)
920 return -EIO;
922 err = unlock_request(cs->req);
923 if (err)
924 return err;
926 fuse_copy_finish(cs);
928 buf = cs->pipebufs;
929 get_page(page);
930 buf->page = page;
931 buf->offset = offset;
932 buf->len = count;
934 cs->pipebufs++;
935 cs->nr_segs++;
936 cs->len = 0;
938 return 0;
942 * Copy a page in the request to/from the userspace buffer. Must be
943 * done atomically
945 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
946 unsigned offset, unsigned count, int zeroing)
948 int err;
949 struct page *page = *pagep;
951 if (page && zeroing && count < PAGE_SIZE)
952 clear_highpage(page);
954 while (count) {
955 if (cs->write && cs->pipebufs && page) {
956 return fuse_ref_page(cs, page, offset, count);
957 } else if (!cs->len) {
958 if (cs->move_pages && page &&
959 offset == 0 && count == PAGE_SIZE) {
960 err = fuse_try_move_page(cs, pagep);
961 if (err <= 0)
962 return err;
963 } else {
964 err = fuse_copy_fill(cs);
965 if (err)
966 return err;
969 if (page) {
970 void *mapaddr = kmap_atomic(page);
971 void *buf = mapaddr + offset;
972 offset += fuse_copy_do(cs, &buf, &count);
973 kunmap_atomic(mapaddr);
974 } else
975 offset += fuse_copy_do(cs, NULL, &count);
977 if (page && !cs->write)
978 flush_dcache_page(page);
979 return 0;
982 /* Copy pages in the request to/from userspace buffer */
983 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
984 int zeroing)
986 unsigned i;
987 struct fuse_req *req = cs->req;
989 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
990 int err;
991 unsigned offset = req->page_descs[i].offset;
992 unsigned count = min(nbytes, req->page_descs[i].length);
994 err = fuse_copy_page(cs, &req->pages[i], offset, count,
995 zeroing);
996 if (err)
997 return err;
999 nbytes -= count;
1001 return 0;
1004 /* Copy a single argument in the request to/from userspace buffer */
1005 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1007 while (size) {
1008 if (!cs->len) {
1009 int err = fuse_copy_fill(cs);
1010 if (err)
1011 return err;
1013 fuse_copy_do(cs, &val, &size);
1015 return 0;
1018 /* Copy request arguments to/from userspace buffer */
1019 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1020 unsigned argpages, struct fuse_arg *args,
1021 int zeroing)
1023 int err = 0;
1024 unsigned i;
1026 for (i = 0; !err && i < numargs; i++) {
1027 struct fuse_arg *arg = &args[i];
1028 if (i == numargs - 1 && argpages)
1029 err = fuse_copy_pages(cs, arg->size, zeroing);
1030 else
1031 err = fuse_copy_one(cs, arg->value, arg->size);
1033 return err;
1036 static int forget_pending(struct fuse_iqueue *fiq)
1038 return fiq->forget_list_head.next != NULL;
1041 static int request_pending(struct fuse_iqueue *fiq)
1043 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1044 forget_pending(fiq);
1048 * Transfer an interrupt request to userspace
1050 * Unlike other requests this is assembled on demand, without a need
1051 * to allocate a separate fuse_req structure.
1053 * Called with fiq->waitq.lock held, releases it
1055 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1056 struct fuse_copy_state *cs,
1057 size_t nbytes, struct fuse_req *req)
1058 __releases(fiq->waitq.lock)
1060 struct fuse_in_header ih;
1061 struct fuse_interrupt_in arg;
1062 unsigned reqsize = sizeof(ih) + sizeof(arg);
1063 int err;
1065 list_del_init(&req->intr_entry);
1066 req->intr_unique = fuse_get_unique(fiq);
1067 memset(&ih, 0, sizeof(ih));
1068 memset(&arg, 0, sizeof(arg));
1069 ih.len = reqsize;
1070 ih.opcode = FUSE_INTERRUPT;
1071 ih.unique = req->intr_unique;
1072 arg.unique = req->in.h.unique;
1074 spin_unlock(&fiq->waitq.lock);
1075 if (nbytes < reqsize)
1076 return -EINVAL;
1078 err = fuse_copy_one(cs, &ih, sizeof(ih));
1079 if (!err)
1080 err = fuse_copy_one(cs, &arg, sizeof(arg));
1081 fuse_copy_finish(cs);
1083 return err ? err : reqsize;
1086 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1087 unsigned max,
1088 unsigned *countp)
1090 struct fuse_forget_link *head = fiq->forget_list_head.next;
1091 struct fuse_forget_link **newhead = &head;
1092 unsigned count;
1094 for (count = 0; *newhead != NULL && count < max; count++)
1095 newhead = &(*newhead)->next;
1097 fiq->forget_list_head.next = *newhead;
1098 *newhead = NULL;
1099 if (fiq->forget_list_head.next == NULL)
1100 fiq->forget_list_tail = &fiq->forget_list_head;
1102 if (countp != NULL)
1103 *countp = count;
1105 return head;
1108 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1109 struct fuse_copy_state *cs,
1110 size_t nbytes)
1111 __releases(fiq->waitq.lock)
1113 int err;
1114 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1115 struct fuse_forget_in arg = {
1116 .nlookup = forget->forget_one.nlookup,
1118 struct fuse_in_header ih = {
1119 .opcode = FUSE_FORGET,
1120 .nodeid = forget->forget_one.nodeid,
1121 .unique = fuse_get_unique(fiq),
1122 .len = sizeof(ih) + sizeof(arg),
1125 spin_unlock(&fiq->waitq.lock);
1126 kfree(forget);
1127 if (nbytes < ih.len)
1128 return -EINVAL;
1130 err = fuse_copy_one(cs, &ih, sizeof(ih));
1131 if (!err)
1132 err = fuse_copy_one(cs, &arg, sizeof(arg));
1133 fuse_copy_finish(cs);
1135 if (err)
1136 return err;
1138 return ih.len;
1141 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1142 struct fuse_copy_state *cs, size_t nbytes)
1143 __releases(fiq->waitq.lock)
1145 int err;
1146 unsigned max_forgets;
1147 unsigned count;
1148 struct fuse_forget_link *head;
1149 struct fuse_batch_forget_in arg = { .count = 0 };
1150 struct fuse_in_header ih = {
1151 .opcode = FUSE_BATCH_FORGET,
1152 .unique = fuse_get_unique(fiq),
1153 .len = sizeof(ih) + sizeof(arg),
1156 if (nbytes < ih.len) {
1157 spin_unlock(&fiq->waitq.lock);
1158 return -EINVAL;
1161 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1162 head = dequeue_forget(fiq, max_forgets, &count);
1163 spin_unlock(&fiq->waitq.lock);
1165 arg.count = count;
1166 ih.len += count * sizeof(struct fuse_forget_one);
1167 err = fuse_copy_one(cs, &ih, sizeof(ih));
1168 if (!err)
1169 err = fuse_copy_one(cs, &arg, sizeof(arg));
1171 while (head) {
1172 struct fuse_forget_link *forget = head;
1174 if (!err) {
1175 err = fuse_copy_one(cs, &forget->forget_one,
1176 sizeof(forget->forget_one));
1178 head = forget->next;
1179 kfree(forget);
1182 fuse_copy_finish(cs);
1184 if (err)
1185 return err;
1187 return ih.len;
1190 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1191 struct fuse_copy_state *cs,
1192 size_t nbytes)
1193 __releases(fiq->waitq.lock)
1195 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1196 return fuse_read_single_forget(fiq, cs, nbytes);
1197 else
1198 return fuse_read_batch_forget(fiq, cs, nbytes);
1202 * Read a single request into the userspace filesystem's buffer. This
1203 * function waits until a request is available, then removes it from
1204 * the pending list and copies request data to userspace buffer. If
1205 * no reply is needed (FORGET) or request has been aborted or there
1206 * was an error during the copying then it's finished by calling
1207 * request_end(). Otherwise add it to the processing list, and set
1208 * the 'sent' flag.
1210 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1211 struct fuse_copy_state *cs, size_t nbytes)
1213 ssize_t err;
1214 struct fuse_conn *fc = fud->fc;
1215 struct fuse_iqueue *fiq = &fc->iq;
1216 struct fuse_pqueue *fpq = &fud->pq;
1217 struct fuse_req *req;
1218 struct fuse_in *in;
1219 unsigned reqsize;
1221 restart:
1222 spin_lock(&fiq->waitq.lock);
1223 err = -EAGAIN;
1224 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1225 !request_pending(fiq))
1226 goto err_unlock;
1228 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1229 !fiq->connected || request_pending(fiq));
1230 if (err)
1231 goto err_unlock;
1233 err = -ENODEV;
1234 if (!fiq->connected)
1235 goto err_unlock;
1237 if (!list_empty(&fiq->interrupts)) {
1238 req = list_entry(fiq->interrupts.next, struct fuse_req,
1239 intr_entry);
1240 return fuse_read_interrupt(fiq, cs, nbytes, req);
1243 if (forget_pending(fiq)) {
1244 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1245 return fuse_read_forget(fc, fiq, cs, nbytes);
1247 if (fiq->forget_batch <= -8)
1248 fiq->forget_batch = 16;
1251 req = list_entry(fiq->pending.next, struct fuse_req, list);
1252 clear_bit(FR_PENDING, &req->flags);
1253 list_del_init(&req->list);
1254 spin_unlock(&fiq->waitq.lock);
1256 in = &req->in;
1257 reqsize = in->h.len;
1258 /* If request is too large, reply with an error and restart the read */
1259 if (nbytes < reqsize) {
1260 req->out.h.error = -EIO;
1261 /* SETXATTR is special, since it may contain too large data */
1262 if (in->h.opcode == FUSE_SETXATTR)
1263 req->out.h.error = -E2BIG;
1264 request_end(fc, req);
1265 goto restart;
1267 spin_lock(&fpq->lock);
1268 list_add(&req->list, &fpq->io);
1269 spin_unlock(&fpq->lock);
1270 cs->req = req;
1271 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1272 if (!err)
1273 err = fuse_copy_args(cs, in->numargs, in->argpages,
1274 (struct fuse_arg *) in->args, 0);
1275 fuse_copy_finish(cs);
1276 spin_lock(&fpq->lock);
1277 clear_bit(FR_LOCKED, &req->flags);
1278 if (!fpq->connected) {
1279 err = -ENODEV;
1280 goto out_end;
1282 if (err) {
1283 req->out.h.error = -EIO;
1284 goto out_end;
1286 if (!test_bit(FR_ISREPLY, &req->flags)) {
1287 err = reqsize;
1288 goto out_end;
1290 list_move_tail(&req->list, &fpq->processing);
1291 spin_unlock(&fpq->lock);
1292 set_bit(FR_SENT, &req->flags);
1293 /* matches barrier in request_wait_answer() */
1294 smp_mb__after_atomic();
1295 if (test_bit(FR_INTERRUPTED, &req->flags))
1296 queue_interrupt(fiq, req);
1298 return reqsize;
1300 out_end:
1301 if (!test_bit(FR_PRIVATE, &req->flags))
1302 list_del_init(&req->list);
1303 spin_unlock(&fpq->lock);
1304 request_end(fc, req);
1305 return err;
1307 err_unlock:
1308 spin_unlock(&fiq->waitq.lock);
1309 return err;
1312 static int fuse_dev_open(struct inode *inode, struct file *file)
1315 * The fuse device's file's private_data is used to hold
1316 * the fuse_conn(ection) when it is mounted, and is used to
1317 * keep track of whether the file has been mounted already.
1319 file->private_data = NULL;
1320 return 0;
1323 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1325 struct fuse_copy_state cs;
1326 struct file *file = iocb->ki_filp;
1327 struct fuse_dev *fud = fuse_get_dev(file);
1329 if (!fud)
1330 return -EPERM;
1332 if (!iter_is_iovec(to))
1333 return -EINVAL;
1335 fuse_copy_init(&cs, 1, to);
1337 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1340 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1341 struct pipe_inode_info *pipe,
1342 size_t len, unsigned int flags)
1344 int total, ret;
1345 int page_nr = 0;
1346 struct pipe_buffer *bufs;
1347 struct fuse_copy_state cs;
1348 struct fuse_dev *fud = fuse_get_dev(in);
1350 if (!fud)
1351 return -EPERM;
1353 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1354 if (!bufs)
1355 return -ENOMEM;
1357 fuse_copy_init(&cs, 1, NULL);
1358 cs.pipebufs = bufs;
1359 cs.pipe = pipe;
1360 ret = fuse_dev_do_read(fud, in, &cs, len);
1361 if (ret < 0)
1362 goto out;
1364 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1365 ret = -EIO;
1366 goto out;
1369 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1371 * Need to be careful about this. Having buf->ops in module
1372 * code can Oops if the buffer persists after module unload.
1374 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1375 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1376 if (unlikely(ret < 0))
1377 break;
1379 if (total)
1380 ret = total;
1381 out:
1382 for (; page_nr < cs.nr_segs; page_nr++)
1383 put_page(bufs[page_nr].page);
1385 kfree(bufs);
1386 return ret;
1389 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1390 struct fuse_copy_state *cs)
1392 struct fuse_notify_poll_wakeup_out outarg;
1393 int err = -EINVAL;
1395 if (size != sizeof(outarg))
1396 goto err;
1398 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1399 if (err)
1400 goto err;
1402 fuse_copy_finish(cs);
1403 return fuse_notify_poll_wakeup(fc, &outarg);
1405 err:
1406 fuse_copy_finish(cs);
1407 return err;
1410 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1411 struct fuse_copy_state *cs)
1413 struct fuse_notify_inval_inode_out outarg;
1414 int err = -EINVAL;
1416 if (size != sizeof(outarg))
1417 goto err;
1419 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1420 if (err)
1421 goto err;
1422 fuse_copy_finish(cs);
1424 down_read(&fc->killsb);
1425 err = -ENOENT;
1426 if (fc->sb) {
1427 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1428 outarg.off, outarg.len);
1430 up_read(&fc->killsb);
1431 return err;
1433 err:
1434 fuse_copy_finish(cs);
1435 return err;
1438 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1439 struct fuse_copy_state *cs)
1441 struct fuse_notify_inval_entry_out outarg;
1442 int err = -ENOMEM;
1443 char *buf;
1444 struct qstr name;
1446 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1447 if (!buf)
1448 goto err;
1450 err = -EINVAL;
1451 if (size < sizeof(outarg))
1452 goto err;
1454 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1455 if (err)
1456 goto err;
1458 err = -ENAMETOOLONG;
1459 if (outarg.namelen > FUSE_NAME_MAX)
1460 goto err;
1462 err = -EINVAL;
1463 if (size != sizeof(outarg) + outarg.namelen + 1)
1464 goto err;
1466 name.name = buf;
1467 name.len = outarg.namelen;
1468 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1469 if (err)
1470 goto err;
1471 fuse_copy_finish(cs);
1472 buf[outarg.namelen] = 0;
1474 down_read(&fc->killsb);
1475 err = -ENOENT;
1476 if (fc->sb)
1477 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1478 up_read(&fc->killsb);
1479 kfree(buf);
1480 return err;
1482 err:
1483 kfree(buf);
1484 fuse_copy_finish(cs);
1485 return err;
1488 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1489 struct fuse_copy_state *cs)
1491 struct fuse_notify_delete_out outarg;
1492 int err = -ENOMEM;
1493 char *buf;
1494 struct qstr name;
1496 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1497 if (!buf)
1498 goto err;
1500 err = -EINVAL;
1501 if (size < sizeof(outarg))
1502 goto err;
1504 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1505 if (err)
1506 goto err;
1508 err = -ENAMETOOLONG;
1509 if (outarg.namelen > FUSE_NAME_MAX)
1510 goto err;
1512 err = -EINVAL;
1513 if (size != sizeof(outarg) + outarg.namelen + 1)
1514 goto err;
1516 name.name = buf;
1517 name.len = outarg.namelen;
1518 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1519 if (err)
1520 goto err;
1521 fuse_copy_finish(cs);
1522 buf[outarg.namelen] = 0;
1524 down_read(&fc->killsb);
1525 err = -ENOENT;
1526 if (fc->sb)
1527 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1528 outarg.child, &name);
1529 up_read(&fc->killsb);
1530 kfree(buf);
1531 return err;
1533 err:
1534 kfree(buf);
1535 fuse_copy_finish(cs);
1536 return err;
1539 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1540 struct fuse_copy_state *cs)
1542 struct fuse_notify_store_out outarg;
1543 struct inode *inode;
1544 struct address_space *mapping;
1545 u64 nodeid;
1546 int err;
1547 pgoff_t index;
1548 unsigned int offset;
1549 unsigned int num;
1550 loff_t file_size;
1551 loff_t end;
1553 err = -EINVAL;
1554 if (size < sizeof(outarg))
1555 goto out_finish;
1557 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1558 if (err)
1559 goto out_finish;
1561 err = -EINVAL;
1562 if (size - sizeof(outarg) != outarg.size)
1563 goto out_finish;
1565 nodeid = outarg.nodeid;
1567 down_read(&fc->killsb);
1569 err = -ENOENT;
1570 if (!fc->sb)
1571 goto out_up_killsb;
1573 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1574 if (!inode)
1575 goto out_up_killsb;
1577 mapping = inode->i_mapping;
1578 index = outarg.offset >> PAGE_SHIFT;
1579 offset = outarg.offset & ~PAGE_MASK;
1580 file_size = i_size_read(inode);
1581 end = outarg.offset + outarg.size;
1582 if (end > file_size) {
1583 file_size = end;
1584 fuse_write_update_size(inode, file_size);
1587 num = outarg.size;
1588 while (num) {
1589 struct page *page;
1590 unsigned int this_num;
1592 err = -ENOMEM;
1593 page = find_or_create_page(mapping, index,
1594 mapping_gfp_mask(mapping));
1595 if (!page)
1596 goto out_iput;
1598 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1599 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1600 if (!err && offset == 0 &&
1601 (this_num == PAGE_SIZE || file_size == end))
1602 SetPageUptodate(page);
1603 unlock_page(page);
1604 put_page(page);
1606 if (err)
1607 goto out_iput;
1609 num -= this_num;
1610 offset = 0;
1611 index++;
1614 err = 0;
1616 out_iput:
1617 iput(inode);
1618 out_up_killsb:
1619 up_read(&fc->killsb);
1620 out_finish:
1621 fuse_copy_finish(cs);
1622 return err;
1625 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1627 release_pages(req->pages, req->num_pages, false);
1630 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1631 struct fuse_notify_retrieve_out *outarg)
1633 int err;
1634 struct address_space *mapping = inode->i_mapping;
1635 struct fuse_req *req;
1636 pgoff_t index;
1637 loff_t file_size;
1638 unsigned int num;
1639 unsigned int offset;
1640 size_t total_len = 0;
1641 int num_pages;
1643 offset = outarg->offset & ~PAGE_MASK;
1644 file_size = i_size_read(inode);
1646 num = outarg->size;
1647 if (outarg->offset > file_size)
1648 num = 0;
1649 else if (outarg->offset + num > file_size)
1650 num = file_size - outarg->offset;
1652 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1653 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1655 req = fuse_get_req(fc, num_pages);
1656 if (IS_ERR(req))
1657 return PTR_ERR(req);
1659 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1660 req->in.h.nodeid = outarg->nodeid;
1661 req->in.numargs = 2;
1662 req->in.argpages = 1;
1663 req->page_descs[0].offset = offset;
1664 req->end = fuse_retrieve_end;
1666 index = outarg->offset >> PAGE_SHIFT;
1668 while (num && req->num_pages < num_pages) {
1669 struct page *page;
1670 unsigned int this_num;
1672 page = find_get_page(mapping, index);
1673 if (!page)
1674 break;
1676 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1677 req->pages[req->num_pages] = page;
1678 req->page_descs[req->num_pages].length = this_num;
1679 req->num_pages++;
1681 offset = 0;
1682 num -= this_num;
1683 total_len += this_num;
1684 index++;
1686 req->misc.retrieve_in.offset = outarg->offset;
1687 req->misc.retrieve_in.size = total_len;
1688 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1689 req->in.args[0].value = &req->misc.retrieve_in;
1690 req->in.args[1].size = total_len;
1692 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1693 if (err)
1694 fuse_retrieve_end(fc, req);
1696 return err;
1699 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1700 struct fuse_copy_state *cs)
1702 struct fuse_notify_retrieve_out outarg;
1703 struct inode *inode;
1704 int err;
1706 err = -EINVAL;
1707 if (size != sizeof(outarg))
1708 goto copy_finish;
1710 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1711 if (err)
1712 goto copy_finish;
1714 fuse_copy_finish(cs);
1716 down_read(&fc->killsb);
1717 err = -ENOENT;
1718 if (fc->sb) {
1719 u64 nodeid = outarg.nodeid;
1721 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1722 if (inode) {
1723 err = fuse_retrieve(fc, inode, &outarg);
1724 iput(inode);
1727 up_read(&fc->killsb);
1729 return err;
1731 copy_finish:
1732 fuse_copy_finish(cs);
1733 return err;
1736 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1737 unsigned int size, struct fuse_copy_state *cs)
1739 /* Don't try to move pages (yet) */
1740 cs->move_pages = 0;
1742 switch (code) {
1743 case FUSE_NOTIFY_POLL:
1744 return fuse_notify_poll(fc, size, cs);
1746 case FUSE_NOTIFY_INVAL_INODE:
1747 return fuse_notify_inval_inode(fc, size, cs);
1749 case FUSE_NOTIFY_INVAL_ENTRY:
1750 return fuse_notify_inval_entry(fc, size, cs);
1752 case FUSE_NOTIFY_STORE:
1753 return fuse_notify_store(fc, size, cs);
1755 case FUSE_NOTIFY_RETRIEVE:
1756 return fuse_notify_retrieve(fc, size, cs);
1758 case FUSE_NOTIFY_DELETE:
1759 return fuse_notify_delete(fc, size, cs);
1761 default:
1762 fuse_copy_finish(cs);
1763 return -EINVAL;
1767 /* Look up request on processing list by unique ID */
1768 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1770 struct fuse_req *req;
1772 list_for_each_entry(req, &fpq->processing, list) {
1773 if (req->in.h.unique == unique || req->intr_unique == unique)
1774 return req;
1776 return NULL;
1779 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1780 unsigned nbytes)
1782 unsigned reqsize = sizeof(struct fuse_out_header);
1784 if (out->h.error)
1785 return nbytes != reqsize ? -EINVAL : 0;
1787 reqsize += len_args(out->numargs, out->args);
1789 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1790 return -EINVAL;
1791 else if (reqsize > nbytes) {
1792 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1793 unsigned diffsize = reqsize - nbytes;
1794 if (diffsize > lastarg->size)
1795 return -EINVAL;
1796 lastarg->size -= diffsize;
1798 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1799 out->page_zeroing);
1803 * Write a single reply to a request. First the header is copied from
1804 * the write buffer. The request is then searched on the processing
1805 * list by the unique ID found in the header. If found, then remove
1806 * it from the list and copy the rest of the buffer to the request.
1807 * The request is finished by calling request_end()
1809 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1810 struct fuse_copy_state *cs, size_t nbytes)
1812 int err;
1813 struct fuse_conn *fc = fud->fc;
1814 struct fuse_pqueue *fpq = &fud->pq;
1815 struct fuse_req *req;
1816 struct fuse_out_header oh;
1818 if (nbytes < sizeof(struct fuse_out_header))
1819 return -EINVAL;
1821 err = fuse_copy_one(cs, &oh, sizeof(oh));
1822 if (err)
1823 goto err_finish;
1825 err = -EINVAL;
1826 if (oh.len != nbytes)
1827 goto err_finish;
1830 * Zero oh.unique indicates unsolicited notification message
1831 * and error contains notification code.
1833 if (!oh.unique) {
1834 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1835 return err ? err : nbytes;
1838 err = -EINVAL;
1839 if (oh.error <= -1000 || oh.error > 0)
1840 goto err_finish;
1842 spin_lock(&fpq->lock);
1843 err = -ENOENT;
1844 if (!fpq->connected)
1845 goto err_unlock_pq;
1847 req = request_find(fpq, oh.unique);
1848 if (!req)
1849 goto err_unlock_pq;
1851 /* Is it an interrupt reply? */
1852 if (req->intr_unique == oh.unique) {
1853 spin_unlock(&fpq->lock);
1855 err = -EINVAL;
1856 if (nbytes != sizeof(struct fuse_out_header))
1857 goto err_finish;
1859 if (oh.error == -ENOSYS)
1860 fc->no_interrupt = 1;
1861 else if (oh.error == -EAGAIN)
1862 queue_interrupt(&fc->iq, req);
1864 fuse_copy_finish(cs);
1865 return nbytes;
1868 clear_bit(FR_SENT, &req->flags);
1869 list_move(&req->list, &fpq->io);
1870 req->out.h = oh;
1871 set_bit(FR_LOCKED, &req->flags);
1872 spin_unlock(&fpq->lock);
1873 cs->req = req;
1874 if (!req->out.page_replace)
1875 cs->move_pages = 0;
1877 err = copy_out_args(cs, &req->out, nbytes);
1878 fuse_copy_finish(cs);
1880 spin_lock(&fpq->lock);
1881 clear_bit(FR_LOCKED, &req->flags);
1882 if (!fpq->connected)
1883 err = -ENOENT;
1884 else if (err)
1885 req->out.h.error = -EIO;
1886 if (!test_bit(FR_PRIVATE, &req->flags))
1887 list_del_init(&req->list);
1888 spin_unlock(&fpq->lock);
1890 request_end(fc, req);
1892 return err ? err : nbytes;
1894 err_unlock_pq:
1895 spin_unlock(&fpq->lock);
1896 err_finish:
1897 fuse_copy_finish(cs);
1898 return err;
1901 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1903 struct fuse_copy_state cs;
1904 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1906 if (!fud)
1907 return -EPERM;
1909 if (!iter_is_iovec(from))
1910 return -EINVAL;
1912 fuse_copy_init(&cs, 0, from);
1914 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1917 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1918 struct file *out, loff_t *ppos,
1919 size_t len, unsigned int flags)
1921 unsigned nbuf;
1922 unsigned idx;
1923 struct pipe_buffer *bufs;
1924 struct fuse_copy_state cs;
1925 struct fuse_dev *fud;
1926 size_t rem;
1927 ssize_t ret;
1929 fud = fuse_get_dev(out);
1930 if (!fud)
1931 return -EPERM;
1933 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1934 if (!bufs)
1935 return -ENOMEM;
1937 pipe_lock(pipe);
1938 nbuf = 0;
1939 rem = 0;
1940 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1941 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1943 ret = -EINVAL;
1944 if (rem < len) {
1945 pipe_unlock(pipe);
1946 goto out;
1949 rem = len;
1950 while (rem) {
1951 struct pipe_buffer *ibuf;
1952 struct pipe_buffer *obuf;
1954 BUG_ON(nbuf >= pipe->buffers);
1955 BUG_ON(!pipe->nrbufs);
1956 ibuf = &pipe->bufs[pipe->curbuf];
1957 obuf = &bufs[nbuf];
1959 if (rem >= ibuf->len) {
1960 *obuf = *ibuf;
1961 ibuf->ops = NULL;
1962 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1963 pipe->nrbufs--;
1964 } else {
1965 pipe_buf_get(pipe, ibuf);
1966 *obuf = *ibuf;
1967 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1968 obuf->len = rem;
1969 ibuf->offset += obuf->len;
1970 ibuf->len -= obuf->len;
1972 nbuf++;
1973 rem -= obuf->len;
1975 pipe_unlock(pipe);
1977 fuse_copy_init(&cs, 0, NULL);
1978 cs.pipebufs = bufs;
1979 cs.nr_segs = nbuf;
1980 cs.pipe = pipe;
1982 if (flags & SPLICE_F_MOVE)
1983 cs.move_pages = 1;
1985 ret = fuse_dev_do_write(fud, &cs, len);
1987 for (idx = 0; idx < nbuf; idx++)
1988 pipe_buf_release(pipe, &bufs[idx]);
1990 out:
1991 kfree(bufs);
1992 return ret;
1995 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1997 unsigned mask = POLLOUT | POLLWRNORM;
1998 struct fuse_iqueue *fiq;
1999 struct fuse_dev *fud = fuse_get_dev(file);
2001 if (!fud)
2002 return POLLERR;
2004 fiq = &fud->fc->iq;
2005 poll_wait(file, &fiq->waitq, wait);
2007 spin_lock(&fiq->waitq.lock);
2008 if (!fiq->connected)
2009 mask = POLLERR;
2010 else if (request_pending(fiq))
2011 mask |= POLLIN | POLLRDNORM;
2012 spin_unlock(&fiq->waitq.lock);
2014 return mask;
2018 * Abort all requests on the given list (pending or processing)
2020 * This function releases and reacquires fc->lock
2022 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2024 while (!list_empty(head)) {
2025 struct fuse_req *req;
2026 req = list_entry(head->next, struct fuse_req, list);
2027 req->out.h.error = -ECONNABORTED;
2028 clear_bit(FR_PENDING, &req->flags);
2029 clear_bit(FR_SENT, &req->flags);
2030 list_del_init(&req->list);
2031 request_end(fc, req);
2035 static void end_polls(struct fuse_conn *fc)
2037 struct rb_node *p;
2039 p = rb_first(&fc->polled_files);
2041 while (p) {
2042 struct fuse_file *ff;
2043 ff = rb_entry(p, struct fuse_file, polled_node);
2044 wake_up_interruptible_all(&ff->poll_wait);
2046 p = rb_next(p);
2051 * Abort all requests.
2053 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2054 * filesystem.
2056 * The same effect is usually achievable through killing the filesystem daemon
2057 * and all users of the filesystem. The exception is the combination of an
2058 * asynchronous request and the tricky deadlock (see
2059 * Documentation/filesystems/fuse.txt).
2061 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2062 * requests, they should be finished off immediately. Locked requests will be
2063 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2064 * requests. It is possible that some request will finish before we can. This
2065 * is OK, the request will in that case be removed from the list before we touch
2066 * it.
2068 void fuse_abort_conn(struct fuse_conn *fc)
2070 struct fuse_iqueue *fiq = &fc->iq;
2072 spin_lock(&fc->lock);
2073 if (fc->connected) {
2074 struct fuse_dev *fud;
2075 struct fuse_req *req, *next;
2076 LIST_HEAD(to_end1);
2077 LIST_HEAD(to_end2);
2079 fc->connected = 0;
2080 fc->blocked = 0;
2081 fuse_set_initialized(fc);
2082 list_for_each_entry(fud, &fc->devices, entry) {
2083 struct fuse_pqueue *fpq = &fud->pq;
2085 spin_lock(&fpq->lock);
2086 fpq->connected = 0;
2087 list_for_each_entry_safe(req, next, &fpq->io, list) {
2088 req->out.h.error = -ECONNABORTED;
2089 spin_lock(&req->waitq.lock);
2090 set_bit(FR_ABORTED, &req->flags);
2091 if (!test_bit(FR_LOCKED, &req->flags)) {
2092 set_bit(FR_PRIVATE, &req->flags);
2093 list_move(&req->list, &to_end1);
2095 spin_unlock(&req->waitq.lock);
2097 list_splice_init(&fpq->processing, &to_end2);
2098 spin_unlock(&fpq->lock);
2100 fc->max_background = UINT_MAX;
2101 flush_bg_queue(fc);
2103 spin_lock(&fiq->waitq.lock);
2104 fiq->connected = 0;
2105 list_splice_init(&fiq->pending, &to_end2);
2106 while (forget_pending(fiq))
2107 kfree(dequeue_forget(fiq, 1, NULL));
2108 wake_up_all_locked(&fiq->waitq);
2109 spin_unlock(&fiq->waitq.lock);
2110 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2111 end_polls(fc);
2112 wake_up_all(&fc->blocked_waitq);
2113 spin_unlock(&fc->lock);
2115 while (!list_empty(&to_end1)) {
2116 req = list_first_entry(&to_end1, struct fuse_req, list);
2117 __fuse_get_request(req);
2118 list_del_init(&req->list);
2119 request_end(fc, req);
2121 end_requests(fc, &to_end2);
2122 } else {
2123 spin_unlock(&fc->lock);
2126 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2128 int fuse_dev_release(struct inode *inode, struct file *file)
2130 struct fuse_dev *fud = fuse_get_dev(file);
2132 if (fud) {
2133 struct fuse_conn *fc = fud->fc;
2134 struct fuse_pqueue *fpq = &fud->pq;
2136 WARN_ON(!list_empty(&fpq->io));
2137 end_requests(fc, &fpq->processing);
2138 /* Are we the last open device? */
2139 if (atomic_dec_and_test(&fc->dev_count)) {
2140 WARN_ON(fc->iq.fasync != NULL);
2141 fuse_abort_conn(fc);
2143 fuse_dev_free(fud);
2145 return 0;
2147 EXPORT_SYMBOL_GPL(fuse_dev_release);
2149 static int fuse_dev_fasync(int fd, struct file *file, int on)
2151 struct fuse_dev *fud = fuse_get_dev(file);
2153 if (!fud)
2154 return -EPERM;
2156 /* No locking - fasync_helper does its own locking */
2157 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2160 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2162 struct fuse_dev *fud;
2164 if (new->private_data)
2165 return -EINVAL;
2167 fud = fuse_dev_alloc(fc);
2168 if (!fud)
2169 return -ENOMEM;
2171 new->private_data = fud;
2172 atomic_inc(&fc->dev_count);
2174 return 0;
2177 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2178 unsigned long arg)
2180 int err = -ENOTTY;
2182 if (cmd == FUSE_DEV_IOC_CLONE) {
2183 int oldfd;
2185 err = -EFAULT;
2186 if (!get_user(oldfd, (__u32 __user *) arg)) {
2187 struct file *old = fget(oldfd);
2189 err = -EINVAL;
2190 if (old) {
2191 struct fuse_dev *fud = NULL;
2194 * Check against file->f_op because CUSE
2195 * uses the same ioctl handler.
2197 if (old->f_op == file->f_op &&
2198 old->f_cred->user_ns == file->f_cred->user_ns)
2199 fud = fuse_get_dev(old);
2201 if (fud) {
2202 mutex_lock(&fuse_mutex);
2203 err = fuse_device_clone(fud->fc, file);
2204 mutex_unlock(&fuse_mutex);
2206 fput(old);
2210 return err;
2213 const struct file_operations fuse_dev_operations = {
2214 .owner = THIS_MODULE,
2215 .open = fuse_dev_open,
2216 .llseek = no_llseek,
2217 .read_iter = fuse_dev_read,
2218 .splice_read = fuse_dev_splice_read,
2219 .write_iter = fuse_dev_write,
2220 .splice_write = fuse_dev_splice_write,
2221 .poll = fuse_dev_poll,
2222 .release = fuse_dev_release,
2223 .fasync = fuse_dev_fasync,
2224 .unlocked_ioctl = fuse_dev_ioctl,
2225 .compat_ioctl = fuse_dev_ioctl,
2227 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2229 static struct miscdevice fuse_miscdevice = {
2230 .minor = FUSE_MINOR,
2231 .name = "fuse",
2232 .fops = &fuse_dev_operations,
2235 int __init fuse_dev_init(void)
2237 int err = -ENOMEM;
2238 fuse_req_cachep = kmem_cache_create("fuse_request",
2239 sizeof(struct fuse_req),
2240 0, 0, NULL);
2241 if (!fuse_req_cachep)
2242 goto out;
2244 err = misc_register(&fuse_miscdevice);
2245 if (err)
2246 goto out_cache_clean;
2248 return 0;
2250 out_cache_clean:
2251 kmem_cache_destroy(fuse_req_cachep);
2252 out:
2253 return err;
2256 void fuse_dev_cleanup(void)
2258 misc_deregister(&fuse_miscdevice);
2259 kmem_cache_destroy(fuse_req_cachep);