Loading...
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/poll.h>
14#include <linux/uio.h>
15#include <linux/miscdevice.h>
16#include <linux/pagemap.h>
17#include <linux/file.h>
18#include <linux/slab.h>
19#include <linux/pipe_fs_i.h>
20#include <linux/swap.h>
21#include <linux/splice.h>
22
23MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24MODULE_ALIAS("devname:fuse");
25
26static struct kmem_cache *fuse_req_cachep;
27
28static struct fuse_conn *fuse_get_conn(struct file *file)
29{
30 /*
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
33 */
34 return file->private_data;
35}
36
37static void fuse_request_init(struct fuse_req *req)
38{
39 memset(req, 0, sizeof(*req));
40 INIT_LIST_HEAD(&req->list);
41 INIT_LIST_HEAD(&req->intr_entry);
42 init_waitqueue_head(&req->waitq);
43 atomic_set(&req->count, 1);
44}
45
46struct fuse_req *fuse_request_alloc(void)
47{
48 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL);
49 if (req)
50 fuse_request_init(req);
51 return req;
52}
53EXPORT_SYMBOL_GPL(fuse_request_alloc);
54
55struct fuse_req *fuse_request_alloc_nofs(void)
56{
57 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_NOFS);
58 if (req)
59 fuse_request_init(req);
60 return req;
61}
62
63void fuse_request_free(struct fuse_req *req)
64{
65 kmem_cache_free(fuse_req_cachep, req);
66}
67
68static void block_sigs(sigset_t *oldset)
69{
70 sigset_t mask;
71
72 siginitsetinv(&mask, sigmask(SIGKILL));
73 sigprocmask(SIG_BLOCK, &mask, oldset);
74}
75
76static void restore_sigs(sigset_t *oldset)
77{
78 sigprocmask(SIG_SETMASK, oldset, NULL);
79}
80
81static void __fuse_get_request(struct fuse_req *req)
82{
83 atomic_inc(&req->count);
84}
85
86/* Must be called with > 1 refcount */
87static void __fuse_put_request(struct fuse_req *req)
88{
89 BUG_ON(atomic_read(&req->count) < 2);
90 atomic_dec(&req->count);
91}
92
93static void fuse_req_init_context(struct fuse_req *req)
94{
95 req->in.h.uid = current_fsuid();
96 req->in.h.gid = current_fsgid();
97 req->in.h.pid = current->pid;
98}
99
100struct fuse_req *fuse_get_req(struct fuse_conn *fc)
101{
102 struct fuse_req *req;
103 sigset_t oldset;
104 int intr;
105 int err;
106
107 atomic_inc(&fc->num_waiting);
108 block_sigs(&oldset);
109 intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked);
110 restore_sigs(&oldset);
111 err = -EINTR;
112 if (intr)
113 goto out;
114
115 err = -ENOTCONN;
116 if (!fc->connected)
117 goto out;
118
119 req = fuse_request_alloc();
120 err = -ENOMEM;
121 if (!req)
122 goto out;
123
124 fuse_req_init_context(req);
125 req->waiting = 1;
126 return req;
127
128 out:
129 atomic_dec(&fc->num_waiting);
130 return ERR_PTR(err);
131}
132EXPORT_SYMBOL_GPL(fuse_get_req);
133
134/*
135 * Return request in fuse_file->reserved_req. However that may
136 * currently be in use. If that is the case, wait for it to become
137 * available.
138 */
139static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
140 struct file *file)
141{
142 struct fuse_req *req = NULL;
143 struct fuse_file *ff = file->private_data;
144
145 do {
146 wait_event(fc->reserved_req_waitq, ff->reserved_req);
147 spin_lock(&fc->lock);
148 if (ff->reserved_req) {
149 req = ff->reserved_req;
150 ff->reserved_req = NULL;
151 get_file(file);
152 req->stolen_file = file;
153 }
154 spin_unlock(&fc->lock);
155 } while (!req);
156
157 return req;
158}
159
160/*
161 * Put stolen request back into fuse_file->reserved_req
162 */
163static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
164{
165 struct file *file = req->stolen_file;
166 struct fuse_file *ff = file->private_data;
167
168 spin_lock(&fc->lock);
169 fuse_request_init(req);
170 BUG_ON(ff->reserved_req);
171 ff->reserved_req = req;
172 wake_up_all(&fc->reserved_req_waitq);
173 spin_unlock(&fc->lock);
174 fput(file);
175}
176
177/*
178 * Gets a requests for a file operation, always succeeds
179 *
180 * This is used for sending the FLUSH request, which must get to
181 * userspace, due to POSIX locks which may need to be unlocked.
182 *
183 * If allocation fails due to OOM, use the reserved request in
184 * fuse_file.
185 *
186 * This is very unlikely to deadlock accidentally, since the
187 * filesystem should not have it's own file open. If deadlock is
188 * intentional, it can still be broken by "aborting" the filesystem.
189 */
190struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file)
191{
192 struct fuse_req *req;
193
194 atomic_inc(&fc->num_waiting);
195 wait_event(fc->blocked_waitq, !fc->blocked);
196 req = fuse_request_alloc();
197 if (!req)
198 req = get_reserved_req(fc, file);
199
200 fuse_req_init_context(req);
201 req->waiting = 1;
202 return req;
203}
204
205void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
206{
207 if (atomic_dec_and_test(&req->count)) {
208 if (req->waiting)
209 atomic_dec(&fc->num_waiting);
210
211 if (req->stolen_file)
212 put_reserved_req(fc, req);
213 else
214 fuse_request_free(req);
215 }
216}
217EXPORT_SYMBOL_GPL(fuse_put_request);
218
219static unsigned len_args(unsigned numargs, struct fuse_arg *args)
220{
221 unsigned nbytes = 0;
222 unsigned i;
223
224 for (i = 0; i < numargs; i++)
225 nbytes += args[i].size;
226
227 return nbytes;
228}
229
230static u64 fuse_get_unique(struct fuse_conn *fc)
231{
232 fc->reqctr++;
233 /* zero is special */
234 if (fc->reqctr == 0)
235 fc->reqctr = 1;
236
237 return fc->reqctr;
238}
239
240static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
241{
242 req->in.h.len = sizeof(struct fuse_in_header) +
243 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
244 list_add_tail(&req->list, &fc->pending);
245 req->state = FUSE_REQ_PENDING;
246 if (!req->waiting) {
247 req->waiting = 1;
248 atomic_inc(&fc->num_waiting);
249 }
250 wake_up(&fc->waitq);
251 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
252}
253
254void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
255 u64 nodeid, u64 nlookup)
256{
257 forget->forget_one.nodeid = nodeid;
258 forget->forget_one.nlookup = nlookup;
259
260 spin_lock(&fc->lock);
261 if (fc->connected) {
262 fc->forget_list_tail->next = forget;
263 fc->forget_list_tail = forget;
264 wake_up(&fc->waitq);
265 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
266 } else {
267 kfree(forget);
268 }
269 spin_unlock(&fc->lock);
270}
271
272static void flush_bg_queue(struct fuse_conn *fc)
273{
274 while (fc->active_background < fc->max_background &&
275 !list_empty(&fc->bg_queue)) {
276 struct fuse_req *req;
277
278 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
279 list_del(&req->list);
280 fc->active_background++;
281 req->in.h.unique = fuse_get_unique(fc);
282 queue_request(fc, req);
283 }
284}
285
286/*
287 * This function is called when a request is finished. Either a reply
288 * has arrived or it was aborted (and not yet sent) or some error
289 * occurred during communication with userspace, or the device file
290 * was closed. The requester thread is woken up (if still waiting),
291 * the 'end' callback is called if given, else the reference to the
292 * request is released
293 *
294 * Called with fc->lock, unlocks it
295 */
296static void request_end(struct fuse_conn *fc, struct fuse_req *req)
297__releases(fc->lock)
298{
299 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
300 req->end = NULL;
301 list_del(&req->list);
302 list_del(&req->intr_entry);
303 req->state = FUSE_REQ_FINISHED;
304 if (req->background) {
305 if (fc->num_background == fc->max_background) {
306 fc->blocked = 0;
307 wake_up_all(&fc->blocked_waitq);
308 }
309 if (fc->num_background == fc->congestion_threshold &&
310 fc->connected && fc->bdi_initialized) {
311 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
312 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
313 }
314 fc->num_background--;
315 fc->active_background--;
316 flush_bg_queue(fc);
317 }
318 spin_unlock(&fc->lock);
319 wake_up(&req->waitq);
320 if (end)
321 end(fc, req);
322 fuse_put_request(fc, req);
323}
324
325static void wait_answer_interruptible(struct fuse_conn *fc,
326 struct fuse_req *req)
327__releases(fc->lock)
328__acquires(fc->lock)
329{
330 if (signal_pending(current))
331 return;
332
333 spin_unlock(&fc->lock);
334 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
335 spin_lock(&fc->lock);
336}
337
338static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
339{
340 list_add_tail(&req->intr_entry, &fc->interrupts);
341 wake_up(&fc->waitq);
342 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
343}
344
345static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
346__releases(fc->lock)
347__acquires(fc->lock)
348{
349 if (!fc->no_interrupt) {
350 /* Any signal may interrupt this */
351 wait_answer_interruptible(fc, req);
352
353 if (req->aborted)
354 goto aborted;
355 if (req->state == FUSE_REQ_FINISHED)
356 return;
357
358 req->interrupted = 1;
359 if (req->state == FUSE_REQ_SENT)
360 queue_interrupt(fc, req);
361 }
362
363 if (!req->force) {
364 sigset_t oldset;
365
366 /* Only fatal signals may interrupt this */
367 block_sigs(&oldset);
368 wait_answer_interruptible(fc, req);
369 restore_sigs(&oldset);
370
371 if (req->aborted)
372 goto aborted;
373 if (req->state == FUSE_REQ_FINISHED)
374 return;
375
376 /* Request is not yet in userspace, bail out */
377 if (req->state == FUSE_REQ_PENDING) {
378 list_del(&req->list);
379 __fuse_put_request(req);
380 req->out.h.error = -EINTR;
381 return;
382 }
383 }
384
385 /*
386 * Either request is already in userspace, or it was forced.
387 * Wait it out.
388 */
389 spin_unlock(&fc->lock);
390 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
391 spin_lock(&fc->lock);
392
393 if (!req->aborted)
394 return;
395
396 aborted:
397 BUG_ON(req->state != FUSE_REQ_FINISHED);
398 if (req->locked) {
399 /* This is uninterruptible sleep, because data is
400 being copied to/from the buffers of req. During
401 locked state, there mustn't be any filesystem
402 operation (e.g. page fault), since that could lead
403 to deadlock */
404 spin_unlock(&fc->lock);
405 wait_event(req->waitq, !req->locked);
406 spin_lock(&fc->lock);
407 }
408}
409
410void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
411{
412 req->isreply = 1;
413 spin_lock(&fc->lock);
414 if (!fc->connected)
415 req->out.h.error = -ENOTCONN;
416 else if (fc->conn_error)
417 req->out.h.error = -ECONNREFUSED;
418 else {
419 req->in.h.unique = fuse_get_unique(fc);
420 queue_request(fc, req);
421 /* acquire extra reference, since request is still needed
422 after request_end() */
423 __fuse_get_request(req);
424
425 request_wait_answer(fc, req);
426 }
427 spin_unlock(&fc->lock);
428}
429EXPORT_SYMBOL_GPL(fuse_request_send);
430
431static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
432 struct fuse_req *req)
433{
434 req->background = 1;
435 fc->num_background++;
436 if (fc->num_background == fc->max_background)
437 fc->blocked = 1;
438 if (fc->num_background == fc->congestion_threshold &&
439 fc->bdi_initialized) {
440 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
441 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
442 }
443 list_add_tail(&req->list, &fc->bg_queue);
444 flush_bg_queue(fc);
445}
446
447static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
448{
449 spin_lock(&fc->lock);
450 if (fc->connected) {
451 fuse_request_send_nowait_locked(fc, req);
452 spin_unlock(&fc->lock);
453 } else {
454 req->out.h.error = -ENOTCONN;
455 request_end(fc, req);
456 }
457}
458
459void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
460{
461 req->isreply = 1;
462 fuse_request_send_nowait(fc, req);
463}
464EXPORT_SYMBOL_GPL(fuse_request_send_background);
465
466static int fuse_request_send_notify_reply(struct fuse_conn *fc,
467 struct fuse_req *req, u64 unique)
468{
469 int err = -ENODEV;
470
471 req->isreply = 0;
472 req->in.h.unique = unique;
473 spin_lock(&fc->lock);
474 if (fc->connected) {
475 queue_request(fc, req);
476 err = 0;
477 }
478 spin_unlock(&fc->lock);
479
480 return err;
481}
482
483/*
484 * Called under fc->lock
485 *
486 * fc->connected must have been checked previously
487 */
488void fuse_request_send_background_locked(struct fuse_conn *fc,
489 struct fuse_req *req)
490{
491 req->isreply = 1;
492 fuse_request_send_nowait_locked(fc, req);
493}
494
495/*
496 * Lock the request. Up to the next unlock_request() there mustn't be
497 * anything that could cause a page-fault. If the request was already
498 * aborted bail out.
499 */
500static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
501{
502 int err = 0;
503 if (req) {
504 spin_lock(&fc->lock);
505 if (req->aborted)
506 err = -ENOENT;
507 else
508 req->locked = 1;
509 spin_unlock(&fc->lock);
510 }
511 return err;
512}
513
514/*
515 * Unlock request. If it was aborted during being locked, the
516 * requester thread is currently waiting for it to be unlocked, so
517 * wake it up.
518 */
519static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
520{
521 if (req) {
522 spin_lock(&fc->lock);
523 req->locked = 0;
524 if (req->aborted)
525 wake_up(&req->waitq);
526 spin_unlock(&fc->lock);
527 }
528}
529
530struct fuse_copy_state {
531 struct fuse_conn *fc;
532 int write;
533 struct fuse_req *req;
534 const struct iovec *iov;
535 struct pipe_buffer *pipebufs;
536 struct pipe_buffer *currbuf;
537 struct pipe_inode_info *pipe;
538 unsigned long nr_segs;
539 unsigned long seglen;
540 unsigned long addr;
541 struct page *pg;
542 void *mapaddr;
543 void *buf;
544 unsigned len;
545 unsigned move_pages:1;
546};
547
548static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
549 int write,
550 const struct iovec *iov, unsigned long nr_segs)
551{
552 memset(cs, 0, sizeof(*cs));
553 cs->fc = fc;
554 cs->write = write;
555 cs->iov = iov;
556 cs->nr_segs = nr_segs;
557}
558
559/* Unmap and put previous page of userspace buffer */
560static void fuse_copy_finish(struct fuse_copy_state *cs)
561{
562 if (cs->currbuf) {
563 struct pipe_buffer *buf = cs->currbuf;
564
565 if (!cs->write) {
566 buf->ops->unmap(cs->pipe, buf, cs->mapaddr);
567 } else {
568 kunmap(buf->page);
569 buf->len = PAGE_SIZE - cs->len;
570 }
571 cs->currbuf = NULL;
572 cs->mapaddr = NULL;
573 } else if (cs->mapaddr) {
574 kunmap(cs->pg);
575 if (cs->write) {
576 flush_dcache_page(cs->pg);
577 set_page_dirty_lock(cs->pg);
578 }
579 put_page(cs->pg);
580 cs->mapaddr = NULL;
581 }
582}
583
584/*
585 * Get another pagefull of userspace buffer, and map it to kernel
586 * address space, and lock request
587 */
588static int fuse_copy_fill(struct fuse_copy_state *cs)
589{
590 unsigned long offset;
591 int err;
592
593 unlock_request(cs->fc, cs->req);
594 fuse_copy_finish(cs);
595 if (cs->pipebufs) {
596 struct pipe_buffer *buf = cs->pipebufs;
597
598 if (!cs->write) {
599 err = buf->ops->confirm(cs->pipe, buf);
600 if (err)
601 return err;
602
603 BUG_ON(!cs->nr_segs);
604 cs->currbuf = buf;
605 cs->mapaddr = buf->ops->map(cs->pipe, buf, 0);
606 cs->len = buf->len;
607 cs->buf = cs->mapaddr + buf->offset;
608 cs->pipebufs++;
609 cs->nr_segs--;
610 } else {
611 struct page *page;
612
613 if (cs->nr_segs == cs->pipe->buffers)
614 return -EIO;
615
616 page = alloc_page(GFP_HIGHUSER);
617 if (!page)
618 return -ENOMEM;
619
620 buf->page = page;
621 buf->offset = 0;
622 buf->len = 0;
623
624 cs->currbuf = buf;
625 cs->mapaddr = kmap(page);
626 cs->buf = cs->mapaddr;
627 cs->len = PAGE_SIZE;
628 cs->pipebufs++;
629 cs->nr_segs++;
630 }
631 } else {
632 if (!cs->seglen) {
633 BUG_ON(!cs->nr_segs);
634 cs->seglen = cs->iov[0].iov_len;
635 cs->addr = (unsigned long) cs->iov[0].iov_base;
636 cs->iov++;
637 cs->nr_segs--;
638 }
639 err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg);
640 if (err < 0)
641 return err;
642 BUG_ON(err != 1);
643 offset = cs->addr % PAGE_SIZE;
644 cs->mapaddr = kmap(cs->pg);
645 cs->buf = cs->mapaddr + offset;
646 cs->len = min(PAGE_SIZE - offset, cs->seglen);
647 cs->seglen -= cs->len;
648 cs->addr += cs->len;
649 }
650
651 return lock_request(cs->fc, cs->req);
652}
653
654/* Do as much copy to/from userspace buffer as we can */
655static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
656{
657 unsigned ncpy = min(*size, cs->len);
658 if (val) {
659 if (cs->write)
660 memcpy(cs->buf, *val, ncpy);
661 else
662 memcpy(*val, cs->buf, ncpy);
663 *val += ncpy;
664 }
665 *size -= ncpy;
666 cs->len -= ncpy;
667 cs->buf += ncpy;
668 return ncpy;
669}
670
671static int fuse_check_page(struct page *page)
672{
673 if (page_mapcount(page) ||
674 page->mapping != NULL ||
675 page_count(page) != 1 ||
676 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
677 ~(1 << PG_locked |
678 1 << PG_referenced |
679 1 << PG_uptodate |
680 1 << PG_lru |
681 1 << PG_active |
682 1 << PG_reclaim))) {
683 printk(KERN_WARNING "fuse: trying to steal weird page\n");
684 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);
685 return 1;
686 }
687 return 0;
688}
689
690static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
691{
692 int err;
693 struct page *oldpage = *pagep;
694 struct page *newpage;
695 struct pipe_buffer *buf = cs->pipebufs;
696 struct address_space *mapping;
697 pgoff_t index;
698
699 unlock_request(cs->fc, cs->req);
700 fuse_copy_finish(cs);
701
702 err = buf->ops->confirm(cs->pipe, buf);
703 if (err)
704 return err;
705
706 BUG_ON(!cs->nr_segs);
707 cs->currbuf = buf;
708 cs->len = buf->len;
709 cs->pipebufs++;
710 cs->nr_segs--;
711
712 if (cs->len != PAGE_SIZE)
713 goto out_fallback;
714
715 if (buf->ops->steal(cs->pipe, buf) != 0)
716 goto out_fallback;
717
718 newpage = buf->page;
719
720 if (WARN_ON(!PageUptodate(newpage)))
721 return -EIO;
722
723 ClearPageMappedToDisk(newpage);
724
725 if (fuse_check_page(newpage) != 0)
726 goto out_fallback_unlock;
727
728 mapping = oldpage->mapping;
729 index = oldpage->index;
730
731 /*
732 * This is a new and locked page, it shouldn't be mapped or
733 * have any special flags on it
734 */
735 if (WARN_ON(page_mapped(oldpage)))
736 goto out_fallback_unlock;
737 if (WARN_ON(page_has_private(oldpage)))
738 goto out_fallback_unlock;
739 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
740 goto out_fallback_unlock;
741 if (WARN_ON(PageMlocked(oldpage)))
742 goto out_fallback_unlock;
743
744 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
745 if (err) {
746 unlock_page(newpage);
747 return err;
748 }
749
750 page_cache_get(newpage);
751
752 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
753 lru_cache_add_file(newpage);
754
755 err = 0;
756 spin_lock(&cs->fc->lock);
757 if (cs->req->aborted)
758 err = -ENOENT;
759 else
760 *pagep = newpage;
761 spin_unlock(&cs->fc->lock);
762
763 if (err) {
764 unlock_page(newpage);
765 page_cache_release(newpage);
766 return err;
767 }
768
769 unlock_page(oldpage);
770 page_cache_release(oldpage);
771 cs->len = 0;
772
773 return 0;
774
775out_fallback_unlock:
776 unlock_page(newpage);
777out_fallback:
778 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
779 cs->buf = cs->mapaddr + buf->offset;
780
781 err = lock_request(cs->fc, cs->req);
782 if (err)
783 return err;
784
785 return 1;
786}
787
788static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
789 unsigned offset, unsigned count)
790{
791 struct pipe_buffer *buf;
792
793 if (cs->nr_segs == cs->pipe->buffers)
794 return -EIO;
795
796 unlock_request(cs->fc, cs->req);
797 fuse_copy_finish(cs);
798
799 buf = cs->pipebufs;
800 page_cache_get(page);
801 buf->page = page;
802 buf->offset = offset;
803 buf->len = count;
804
805 cs->pipebufs++;
806 cs->nr_segs++;
807 cs->len = 0;
808
809 return 0;
810}
811
812/*
813 * Copy a page in the request to/from the userspace buffer. Must be
814 * done atomically
815 */
816static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
817 unsigned offset, unsigned count, int zeroing)
818{
819 int err;
820 struct page *page = *pagep;
821
822 if (page && zeroing && count < PAGE_SIZE)
823 clear_highpage(page);
824
825 while (count) {
826 if (cs->write && cs->pipebufs && page) {
827 return fuse_ref_page(cs, page, offset, count);
828 } else if (!cs->len) {
829 if (cs->move_pages && page &&
830 offset == 0 && count == PAGE_SIZE) {
831 err = fuse_try_move_page(cs, pagep);
832 if (err <= 0)
833 return err;
834 } else {
835 err = fuse_copy_fill(cs);
836 if (err)
837 return err;
838 }
839 }
840 if (page) {
841 void *mapaddr = kmap_atomic(page);
842 void *buf = mapaddr + offset;
843 offset += fuse_copy_do(cs, &buf, &count);
844 kunmap_atomic(mapaddr);
845 } else
846 offset += fuse_copy_do(cs, NULL, &count);
847 }
848 if (page && !cs->write)
849 flush_dcache_page(page);
850 return 0;
851}
852
853/* Copy pages in the request to/from userspace buffer */
854static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
855 int zeroing)
856{
857 unsigned i;
858 struct fuse_req *req = cs->req;
859 unsigned offset = req->page_offset;
860 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
861
862 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
863 int err;
864
865 err = fuse_copy_page(cs, &req->pages[i], offset, count,
866 zeroing);
867 if (err)
868 return err;
869
870 nbytes -= count;
871 count = min(nbytes, (unsigned) PAGE_SIZE);
872 offset = 0;
873 }
874 return 0;
875}
876
877/* Copy a single argument in the request to/from userspace buffer */
878static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
879{
880 while (size) {
881 if (!cs->len) {
882 int err = fuse_copy_fill(cs);
883 if (err)
884 return err;
885 }
886 fuse_copy_do(cs, &val, &size);
887 }
888 return 0;
889}
890
891/* Copy request arguments to/from userspace buffer */
892static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
893 unsigned argpages, struct fuse_arg *args,
894 int zeroing)
895{
896 int err = 0;
897 unsigned i;
898
899 for (i = 0; !err && i < numargs; i++) {
900 struct fuse_arg *arg = &args[i];
901 if (i == numargs - 1 && argpages)
902 err = fuse_copy_pages(cs, arg->size, zeroing);
903 else
904 err = fuse_copy_one(cs, arg->value, arg->size);
905 }
906 return err;
907}
908
909static int forget_pending(struct fuse_conn *fc)
910{
911 return fc->forget_list_head.next != NULL;
912}
913
914static int request_pending(struct fuse_conn *fc)
915{
916 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
917 forget_pending(fc);
918}
919
920/* Wait until a request is available on the pending list */
921static void request_wait(struct fuse_conn *fc)
922__releases(fc->lock)
923__acquires(fc->lock)
924{
925 DECLARE_WAITQUEUE(wait, current);
926
927 add_wait_queue_exclusive(&fc->waitq, &wait);
928 while (fc->connected && !request_pending(fc)) {
929 set_current_state(TASK_INTERRUPTIBLE);
930 if (signal_pending(current))
931 break;
932
933 spin_unlock(&fc->lock);
934 schedule();
935 spin_lock(&fc->lock);
936 }
937 set_current_state(TASK_RUNNING);
938 remove_wait_queue(&fc->waitq, &wait);
939}
940
941/*
942 * Transfer an interrupt request to userspace
943 *
944 * Unlike other requests this is assembled on demand, without a need
945 * to allocate a separate fuse_req structure.
946 *
947 * Called with fc->lock held, releases it
948 */
949static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
950 size_t nbytes, struct fuse_req *req)
951__releases(fc->lock)
952{
953 struct fuse_in_header ih;
954 struct fuse_interrupt_in arg;
955 unsigned reqsize = sizeof(ih) + sizeof(arg);
956 int err;
957
958 list_del_init(&req->intr_entry);
959 req->intr_unique = fuse_get_unique(fc);
960 memset(&ih, 0, sizeof(ih));
961 memset(&arg, 0, sizeof(arg));
962 ih.len = reqsize;
963 ih.opcode = FUSE_INTERRUPT;
964 ih.unique = req->intr_unique;
965 arg.unique = req->in.h.unique;
966
967 spin_unlock(&fc->lock);
968 if (nbytes < reqsize)
969 return -EINVAL;
970
971 err = fuse_copy_one(cs, &ih, sizeof(ih));
972 if (!err)
973 err = fuse_copy_one(cs, &arg, sizeof(arg));
974 fuse_copy_finish(cs);
975
976 return err ? err : reqsize;
977}
978
979static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
980 unsigned max,
981 unsigned *countp)
982{
983 struct fuse_forget_link *head = fc->forget_list_head.next;
984 struct fuse_forget_link **newhead = &head;
985 unsigned count;
986
987 for (count = 0; *newhead != NULL && count < max; count++)
988 newhead = &(*newhead)->next;
989
990 fc->forget_list_head.next = *newhead;
991 *newhead = NULL;
992 if (fc->forget_list_head.next == NULL)
993 fc->forget_list_tail = &fc->forget_list_head;
994
995 if (countp != NULL)
996 *countp = count;
997
998 return head;
999}
1000
1001static int fuse_read_single_forget(struct fuse_conn *fc,
1002 struct fuse_copy_state *cs,
1003 size_t nbytes)
1004__releases(fc->lock)
1005{
1006 int err;
1007 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1008 struct fuse_forget_in arg = {
1009 .nlookup = forget->forget_one.nlookup,
1010 };
1011 struct fuse_in_header ih = {
1012 .opcode = FUSE_FORGET,
1013 .nodeid = forget->forget_one.nodeid,
1014 .unique = fuse_get_unique(fc),
1015 .len = sizeof(ih) + sizeof(arg),
1016 };
1017
1018 spin_unlock(&fc->lock);
1019 kfree(forget);
1020 if (nbytes < ih.len)
1021 return -EINVAL;
1022
1023 err = fuse_copy_one(cs, &ih, sizeof(ih));
1024 if (!err)
1025 err = fuse_copy_one(cs, &arg, sizeof(arg));
1026 fuse_copy_finish(cs);
1027
1028 if (err)
1029 return err;
1030
1031 return ih.len;
1032}
1033
1034static int fuse_read_batch_forget(struct fuse_conn *fc,
1035 struct fuse_copy_state *cs, size_t nbytes)
1036__releases(fc->lock)
1037{
1038 int err;
1039 unsigned max_forgets;
1040 unsigned count;
1041 struct fuse_forget_link *head;
1042 struct fuse_batch_forget_in arg = { .count = 0 };
1043 struct fuse_in_header ih = {
1044 .opcode = FUSE_BATCH_FORGET,
1045 .unique = fuse_get_unique(fc),
1046 .len = sizeof(ih) + sizeof(arg),
1047 };
1048
1049 if (nbytes < ih.len) {
1050 spin_unlock(&fc->lock);
1051 return -EINVAL;
1052 }
1053
1054 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1055 head = dequeue_forget(fc, max_forgets, &count);
1056 spin_unlock(&fc->lock);
1057
1058 arg.count = count;
1059 ih.len += count * sizeof(struct fuse_forget_one);
1060 err = fuse_copy_one(cs, &ih, sizeof(ih));
1061 if (!err)
1062 err = fuse_copy_one(cs, &arg, sizeof(arg));
1063
1064 while (head) {
1065 struct fuse_forget_link *forget = head;
1066
1067 if (!err) {
1068 err = fuse_copy_one(cs, &forget->forget_one,
1069 sizeof(forget->forget_one));
1070 }
1071 head = forget->next;
1072 kfree(forget);
1073 }
1074
1075 fuse_copy_finish(cs);
1076
1077 if (err)
1078 return err;
1079
1080 return ih.len;
1081}
1082
1083static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1084 size_t nbytes)
1085__releases(fc->lock)
1086{
1087 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1088 return fuse_read_single_forget(fc, cs, nbytes);
1089 else
1090 return fuse_read_batch_forget(fc, cs, nbytes);
1091}
1092
1093/*
1094 * Read a single request into the userspace filesystem's buffer. This
1095 * function waits until a request is available, then removes it from
1096 * the pending list and copies request data to userspace buffer. If
1097 * no reply is needed (FORGET) or request has been aborted or there
1098 * was an error during the copying then it's finished by calling
1099 * request_end(). Otherwise add it to the processing list, and set
1100 * the 'sent' flag.
1101 */
1102static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1103 struct fuse_copy_state *cs, size_t nbytes)
1104{
1105 int err;
1106 struct fuse_req *req;
1107 struct fuse_in *in;
1108 unsigned reqsize;
1109
1110 restart:
1111 spin_lock(&fc->lock);
1112 err = -EAGAIN;
1113 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1114 !request_pending(fc))
1115 goto err_unlock;
1116
1117 request_wait(fc);
1118 err = -ENODEV;
1119 if (!fc->connected)
1120 goto err_unlock;
1121 err = -ERESTARTSYS;
1122 if (!request_pending(fc))
1123 goto err_unlock;
1124
1125 if (!list_empty(&fc->interrupts)) {
1126 req = list_entry(fc->interrupts.next, struct fuse_req,
1127 intr_entry);
1128 return fuse_read_interrupt(fc, cs, nbytes, req);
1129 }
1130
1131 if (forget_pending(fc)) {
1132 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1133 return fuse_read_forget(fc, cs, nbytes);
1134
1135 if (fc->forget_batch <= -8)
1136 fc->forget_batch = 16;
1137 }
1138
1139 req = list_entry(fc->pending.next, struct fuse_req, list);
1140 req->state = FUSE_REQ_READING;
1141 list_move(&req->list, &fc->io);
1142
1143 in = &req->in;
1144 reqsize = in->h.len;
1145 /* If request is too large, reply with an error and restart the read */
1146 if (nbytes < reqsize) {
1147 req->out.h.error = -EIO;
1148 /* SETXATTR is special, since it may contain too large data */
1149 if (in->h.opcode == FUSE_SETXATTR)
1150 req->out.h.error = -E2BIG;
1151 request_end(fc, req);
1152 goto restart;
1153 }
1154 spin_unlock(&fc->lock);
1155 cs->req = req;
1156 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1157 if (!err)
1158 err = fuse_copy_args(cs, in->numargs, in->argpages,
1159 (struct fuse_arg *) in->args, 0);
1160 fuse_copy_finish(cs);
1161 spin_lock(&fc->lock);
1162 req->locked = 0;
1163 if (req->aborted) {
1164 request_end(fc, req);
1165 return -ENODEV;
1166 }
1167 if (err) {
1168 req->out.h.error = -EIO;
1169 request_end(fc, req);
1170 return err;
1171 }
1172 if (!req->isreply)
1173 request_end(fc, req);
1174 else {
1175 req->state = FUSE_REQ_SENT;
1176 list_move_tail(&req->list, &fc->processing);
1177 if (req->interrupted)
1178 queue_interrupt(fc, req);
1179 spin_unlock(&fc->lock);
1180 }
1181 return reqsize;
1182
1183 err_unlock:
1184 spin_unlock(&fc->lock);
1185 return err;
1186}
1187
1188static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1189 unsigned long nr_segs, loff_t pos)
1190{
1191 struct fuse_copy_state cs;
1192 struct file *file = iocb->ki_filp;
1193 struct fuse_conn *fc = fuse_get_conn(file);
1194 if (!fc)
1195 return -EPERM;
1196
1197 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1198
1199 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1200}
1201
1202static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe,
1203 struct pipe_buffer *buf)
1204{
1205 return 1;
1206}
1207
1208static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = {
1209 .can_merge = 0,
1210 .map = generic_pipe_buf_map,
1211 .unmap = generic_pipe_buf_unmap,
1212 .confirm = generic_pipe_buf_confirm,
1213 .release = generic_pipe_buf_release,
1214 .steal = fuse_dev_pipe_buf_steal,
1215 .get = generic_pipe_buf_get,
1216};
1217
1218static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1219 struct pipe_inode_info *pipe,
1220 size_t len, unsigned int flags)
1221{
1222 int ret;
1223 int page_nr = 0;
1224 int do_wakeup = 0;
1225 struct pipe_buffer *bufs;
1226 struct fuse_copy_state cs;
1227 struct fuse_conn *fc = fuse_get_conn(in);
1228 if (!fc)
1229 return -EPERM;
1230
1231 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1232 if (!bufs)
1233 return -ENOMEM;
1234
1235 fuse_copy_init(&cs, fc, 1, NULL, 0);
1236 cs.pipebufs = bufs;
1237 cs.pipe = pipe;
1238 ret = fuse_dev_do_read(fc, in, &cs, len);
1239 if (ret < 0)
1240 goto out;
1241
1242 ret = 0;
1243 pipe_lock(pipe);
1244
1245 if (!pipe->readers) {
1246 send_sig(SIGPIPE, current, 0);
1247 if (!ret)
1248 ret = -EPIPE;
1249 goto out_unlock;
1250 }
1251
1252 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1253 ret = -EIO;
1254 goto out_unlock;
1255 }
1256
1257 while (page_nr < cs.nr_segs) {
1258 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1259 struct pipe_buffer *buf = pipe->bufs + newbuf;
1260
1261 buf->page = bufs[page_nr].page;
1262 buf->offset = bufs[page_nr].offset;
1263 buf->len = bufs[page_nr].len;
1264 buf->ops = &fuse_dev_pipe_buf_ops;
1265
1266 pipe->nrbufs++;
1267 page_nr++;
1268 ret += buf->len;
1269
1270 if (pipe->inode)
1271 do_wakeup = 1;
1272 }
1273
1274out_unlock:
1275 pipe_unlock(pipe);
1276
1277 if (do_wakeup) {
1278 smp_mb();
1279 if (waitqueue_active(&pipe->wait))
1280 wake_up_interruptible(&pipe->wait);
1281 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1282 }
1283
1284out:
1285 for (; page_nr < cs.nr_segs; page_nr++)
1286 page_cache_release(bufs[page_nr].page);
1287
1288 kfree(bufs);
1289 return ret;
1290}
1291
1292static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1293 struct fuse_copy_state *cs)
1294{
1295 struct fuse_notify_poll_wakeup_out outarg;
1296 int err = -EINVAL;
1297
1298 if (size != sizeof(outarg))
1299 goto err;
1300
1301 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1302 if (err)
1303 goto err;
1304
1305 fuse_copy_finish(cs);
1306 return fuse_notify_poll_wakeup(fc, &outarg);
1307
1308err:
1309 fuse_copy_finish(cs);
1310 return err;
1311}
1312
1313static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1314 struct fuse_copy_state *cs)
1315{
1316 struct fuse_notify_inval_inode_out outarg;
1317 int err = -EINVAL;
1318
1319 if (size != sizeof(outarg))
1320 goto err;
1321
1322 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1323 if (err)
1324 goto err;
1325 fuse_copy_finish(cs);
1326
1327 down_read(&fc->killsb);
1328 err = -ENOENT;
1329 if (fc->sb) {
1330 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1331 outarg.off, outarg.len);
1332 }
1333 up_read(&fc->killsb);
1334 return err;
1335
1336err:
1337 fuse_copy_finish(cs);
1338 return err;
1339}
1340
1341static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1342 struct fuse_copy_state *cs)
1343{
1344 struct fuse_notify_inval_entry_out outarg;
1345 int err = -ENOMEM;
1346 char *buf;
1347 struct qstr name;
1348
1349 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1350 if (!buf)
1351 goto err;
1352
1353 err = -EINVAL;
1354 if (size < sizeof(outarg))
1355 goto err;
1356
1357 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1358 if (err)
1359 goto err;
1360
1361 err = -ENAMETOOLONG;
1362 if (outarg.namelen > FUSE_NAME_MAX)
1363 goto err;
1364
1365 err = -EINVAL;
1366 if (size != sizeof(outarg) + outarg.namelen + 1)
1367 goto err;
1368
1369 name.name = buf;
1370 name.len = outarg.namelen;
1371 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1372 if (err)
1373 goto err;
1374 fuse_copy_finish(cs);
1375 buf[outarg.namelen] = 0;
1376 name.hash = full_name_hash(name.name, name.len);
1377
1378 down_read(&fc->killsb);
1379 err = -ENOENT;
1380 if (fc->sb)
1381 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1382 up_read(&fc->killsb);
1383 kfree(buf);
1384 return err;
1385
1386err:
1387 kfree(buf);
1388 fuse_copy_finish(cs);
1389 return err;
1390}
1391
1392static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1393 struct fuse_copy_state *cs)
1394{
1395 struct fuse_notify_delete_out outarg;
1396 int err = -ENOMEM;
1397 char *buf;
1398 struct qstr name;
1399
1400 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1401 if (!buf)
1402 goto err;
1403
1404 err = -EINVAL;
1405 if (size < sizeof(outarg))
1406 goto err;
1407
1408 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1409 if (err)
1410 goto err;
1411
1412 err = -ENAMETOOLONG;
1413 if (outarg.namelen > FUSE_NAME_MAX)
1414 goto err;
1415
1416 err = -EINVAL;
1417 if (size != sizeof(outarg) + outarg.namelen + 1)
1418 goto err;
1419
1420 name.name = buf;
1421 name.len = outarg.namelen;
1422 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1423 if (err)
1424 goto err;
1425 fuse_copy_finish(cs);
1426 buf[outarg.namelen] = 0;
1427 name.hash = full_name_hash(name.name, name.len);
1428
1429 down_read(&fc->killsb);
1430 err = -ENOENT;
1431 if (fc->sb)
1432 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1433 outarg.child, &name);
1434 up_read(&fc->killsb);
1435 kfree(buf);
1436 return err;
1437
1438err:
1439 kfree(buf);
1440 fuse_copy_finish(cs);
1441 return err;
1442}
1443
1444static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1445 struct fuse_copy_state *cs)
1446{
1447 struct fuse_notify_store_out outarg;
1448 struct inode *inode;
1449 struct address_space *mapping;
1450 u64 nodeid;
1451 int err;
1452 pgoff_t index;
1453 unsigned int offset;
1454 unsigned int num;
1455 loff_t file_size;
1456 loff_t end;
1457
1458 err = -EINVAL;
1459 if (size < sizeof(outarg))
1460 goto out_finish;
1461
1462 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1463 if (err)
1464 goto out_finish;
1465
1466 err = -EINVAL;
1467 if (size - sizeof(outarg) != outarg.size)
1468 goto out_finish;
1469
1470 nodeid = outarg.nodeid;
1471
1472 down_read(&fc->killsb);
1473
1474 err = -ENOENT;
1475 if (!fc->sb)
1476 goto out_up_killsb;
1477
1478 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1479 if (!inode)
1480 goto out_up_killsb;
1481
1482 mapping = inode->i_mapping;
1483 index = outarg.offset >> PAGE_CACHE_SHIFT;
1484 offset = outarg.offset & ~PAGE_CACHE_MASK;
1485 file_size = i_size_read(inode);
1486 end = outarg.offset + outarg.size;
1487 if (end > file_size) {
1488 file_size = end;
1489 fuse_write_update_size(inode, file_size);
1490 }
1491
1492 num = outarg.size;
1493 while (num) {
1494 struct page *page;
1495 unsigned int this_num;
1496
1497 err = -ENOMEM;
1498 page = find_or_create_page(mapping, index,
1499 mapping_gfp_mask(mapping));
1500 if (!page)
1501 goto out_iput;
1502
1503 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1504 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1505 if (!err && offset == 0 && (num != 0 || file_size == end))
1506 SetPageUptodate(page);
1507 unlock_page(page);
1508 page_cache_release(page);
1509
1510 if (err)
1511 goto out_iput;
1512
1513 num -= this_num;
1514 offset = 0;
1515 index++;
1516 }
1517
1518 err = 0;
1519
1520out_iput:
1521 iput(inode);
1522out_up_killsb:
1523 up_read(&fc->killsb);
1524out_finish:
1525 fuse_copy_finish(cs);
1526 return err;
1527}
1528
1529static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1530{
1531 release_pages(req->pages, req->num_pages, 0);
1532}
1533
1534static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1535 struct fuse_notify_retrieve_out *outarg)
1536{
1537 int err;
1538 struct address_space *mapping = inode->i_mapping;
1539 struct fuse_req *req;
1540 pgoff_t index;
1541 loff_t file_size;
1542 unsigned int num;
1543 unsigned int offset;
1544 size_t total_len = 0;
1545
1546 req = fuse_get_req(fc);
1547 if (IS_ERR(req))
1548 return PTR_ERR(req);
1549
1550 offset = outarg->offset & ~PAGE_CACHE_MASK;
1551
1552 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1553 req->in.h.nodeid = outarg->nodeid;
1554 req->in.numargs = 2;
1555 req->in.argpages = 1;
1556 req->page_offset = offset;
1557 req->end = fuse_retrieve_end;
1558
1559 index = outarg->offset >> PAGE_CACHE_SHIFT;
1560 file_size = i_size_read(inode);
1561 num = outarg->size;
1562 if (outarg->offset > file_size)
1563 num = 0;
1564 else if (outarg->offset + num > file_size)
1565 num = file_size - outarg->offset;
1566
1567 while (num && req->num_pages < FUSE_MAX_PAGES_PER_REQ) {
1568 struct page *page;
1569 unsigned int this_num;
1570
1571 page = find_get_page(mapping, index);
1572 if (!page)
1573 break;
1574
1575 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1576 req->pages[req->num_pages] = page;
1577 req->num_pages++;
1578
1579 offset = 0;
1580 num -= this_num;
1581 total_len += this_num;
1582 index++;
1583 }
1584 req->misc.retrieve_in.offset = outarg->offset;
1585 req->misc.retrieve_in.size = total_len;
1586 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1587 req->in.args[0].value = &req->misc.retrieve_in;
1588 req->in.args[1].size = total_len;
1589
1590 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1591 if (err)
1592 fuse_retrieve_end(fc, req);
1593
1594 return err;
1595}
1596
1597static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1598 struct fuse_copy_state *cs)
1599{
1600 struct fuse_notify_retrieve_out outarg;
1601 struct inode *inode;
1602 int err;
1603
1604 err = -EINVAL;
1605 if (size != sizeof(outarg))
1606 goto copy_finish;
1607
1608 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1609 if (err)
1610 goto copy_finish;
1611
1612 fuse_copy_finish(cs);
1613
1614 down_read(&fc->killsb);
1615 err = -ENOENT;
1616 if (fc->sb) {
1617 u64 nodeid = outarg.nodeid;
1618
1619 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1620 if (inode) {
1621 err = fuse_retrieve(fc, inode, &outarg);
1622 iput(inode);
1623 }
1624 }
1625 up_read(&fc->killsb);
1626
1627 return err;
1628
1629copy_finish:
1630 fuse_copy_finish(cs);
1631 return err;
1632}
1633
1634static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1635 unsigned int size, struct fuse_copy_state *cs)
1636{
1637 switch (code) {
1638 case FUSE_NOTIFY_POLL:
1639 return fuse_notify_poll(fc, size, cs);
1640
1641 case FUSE_NOTIFY_INVAL_INODE:
1642 return fuse_notify_inval_inode(fc, size, cs);
1643
1644 case FUSE_NOTIFY_INVAL_ENTRY:
1645 return fuse_notify_inval_entry(fc, size, cs);
1646
1647 case FUSE_NOTIFY_STORE:
1648 return fuse_notify_store(fc, size, cs);
1649
1650 case FUSE_NOTIFY_RETRIEVE:
1651 return fuse_notify_retrieve(fc, size, cs);
1652
1653 case FUSE_NOTIFY_DELETE:
1654 return fuse_notify_delete(fc, size, cs);
1655
1656 default:
1657 fuse_copy_finish(cs);
1658 return -EINVAL;
1659 }
1660}
1661
1662/* Look up request on processing list by unique ID */
1663static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1664{
1665 struct list_head *entry;
1666
1667 list_for_each(entry, &fc->processing) {
1668 struct fuse_req *req;
1669 req = list_entry(entry, struct fuse_req, list);
1670 if (req->in.h.unique == unique || req->intr_unique == unique)
1671 return req;
1672 }
1673 return NULL;
1674}
1675
1676static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1677 unsigned nbytes)
1678{
1679 unsigned reqsize = sizeof(struct fuse_out_header);
1680
1681 if (out->h.error)
1682 return nbytes != reqsize ? -EINVAL : 0;
1683
1684 reqsize += len_args(out->numargs, out->args);
1685
1686 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1687 return -EINVAL;
1688 else if (reqsize > nbytes) {
1689 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1690 unsigned diffsize = reqsize - nbytes;
1691 if (diffsize > lastarg->size)
1692 return -EINVAL;
1693 lastarg->size -= diffsize;
1694 }
1695 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1696 out->page_zeroing);
1697}
1698
1699/*
1700 * Write a single reply to a request. First the header is copied from
1701 * the write buffer. The request is then searched on the processing
1702 * list by the unique ID found in the header. If found, then remove
1703 * it from the list and copy the rest of the buffer to the request.
1704 * The request is finished by calling request_end()
1705 */
1706static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1707 struct fuse_copy_state *cs, size_t nbytes)
1708{
1709 int err;
1710 struct fuse_req *req;
1711 struct fuse_out_header oh;
1712
1713 if (nbytes < sizeof(struct fuse_out_header))
1714 return -EINVAL;
1715
1716 err = fuse_copy_one(cs, &oh, sizeof(oh));
1717 if (err)
1718 goto err_finish;
1719
1720 err = -EINVAL;
1721 if (oh.len != nbytes)
1722 goto err_finish;
1723
1724 /*
1725 * Zero oh.unique indicates unsolicited notification message
1726 * and error contains notification code.
1727 */
1728 if (!oh.unique) {
1729 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1730 return err ? err : nbytes;
1731 }
1732
1733 err = -EINVAL;
1734 if (oh.error <= -1000 || oh.error > 0)
1735 goto err_finish;
1736
1737 spin_lock(&fc->lock);
1738 err = -ENOENT;
1739 if (!fc->connected)
1740 goto err_unlock;
1741
1742 req = request_find(fc, oh.unique);
1743 if (!req)
1744 goto err_unlock;
1745
1746 if (req->aborted) {
1747 spin_unlock(&fc->lock);
1748 fuse_copy_finish(cs);
1749 spin_lock(&fc->lock);
1750 request_end(fc, req);
1751 return -ENOENT;
1752 }
1753 /* Is it an interrupt reply? */
1754 if (req->intr_unique == oh.unique) {
1755 err = -EINVAL;
1756 if (nbytes != sizeof(struct fuse_out_header))
1757 goto err_unlock;
1758
1759 if (oh.error == -ENOSYS)
1760 fc->no_interrupt = 1;
1761 else if (oh.error == -EAGAIN)
1762 queue_interrupt(fc, req);
1763
1764 spin_unlock(&fc->lock);
1765 fuse_copy_finish(cs);
1766 return nbytes;
1767 }
1768
1769 req->state = FUSE_REQ_WRITING;
1770 list_move(&req->list, &fc->io);
1771 req->out.h = oh;
1772 req->locked = 1;
1773 cs->req = req;
1774 if (!req->out.page_replace)
1775 cs->move_pages = 0;
1776 spin_unlock(&fc->lock);
1777
1778 err = copy_out_args(cs, &req->out, nbytes);
1779 fuse_copy_finish(cs);
1780
1781 spin_lock(&fc->lock);
1782 req->locked = 0;
1783 if (!err) {
1784 if (req->aborted)
1785 err = -ENOENT;
1786 } else if (!req->aborted)
1787 req->out.h.error = -EIO;
1788 request_end(fc, req);
1789
1790 return err ? err : nbytes;
1791
1792 err_unlock:
1793 spin_unlock(&fc->lock);
1794 err_finish:
1795 fuse_copy_finish(cs);
1796 return err;
1797}
1798
1799static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1800 unsigned long nr_segs, loff_t pos)
1801{
1802 struct fuse_copy_state cs;
1803 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1804 if (!fc)
1805 return -EPERM;
1806
1807 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1808
1809 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1810}
1811
1812static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1813 struct file *out, loff_t *ppos,
1814 size_t len, unsigned int flags)
1815{
1816 unsigned nbuf;
1817 unsigned idx;
1818 struct pipe_buffer *bufs;
1819 struct fuse_copy_state cs;
1820 struct fuse_conn *fc;
1821 size_t rem;
1822 ssize_t ret;
1823
1824 fc = fuse_get_conn(out);
1825 if (!fc)
1826 return -EPERM;
1827
1828 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1829 if (!bufs)
1830 return -ENOMEM;
1831
1832 pipe_lock(pipe);
1833 nbuf = 0;
1834 rem = 0;
1835 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1836 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1837
1838 ret = -EINVAL;
1839 if (rem < len) {
1840 pipe_unlock(pipe);
1841 goto out;
1842 }
1843
1844 rem = len;
1845 while (rem) {
1846 struct pipe_buffer *ibuf;
1847 struct pipe_buffer *obuf;
1848
1849 BUG_ON(nbuf >= pipe->buffers);
1850 BUG_ON(!pipe->nrbufs);
1851 ibuf = &pipe->bufs[pipe->curbuf];
1852 obuf = &bufs[nbuf];
1853
1854 if (rem >= ibuf->len) {
1855 *obuf = *ibuf;
1856 ibuf->ops = NULL;
1857 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1858 pipe->nrbufs--;
1859 } else {
1860 ibuf->ops->get(pipe, ibuf);
1861 *obuf = *ibuf;
1862 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1863 obuf->len = rem;
1864 ibuf->offset += obuf->len;
1865 ibuf->len -= obuf->len;
1866 }
1867 nbuf++;
1868 rem -= obuf->len;
1869 }
1870 pipe_unlock(pipe);
1871
1872 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1873 cs.pipebufs = bufs;
1874 cs.pipe = pipe;
1875
1876 if (flags & SPLICE_F_MOVE)
1877 cs.move_pages = 1;
1878
1879 ret = fuse_dev_do_write(fc, &cs, len);
1880
1881 for (idx = 0; idx < nbuf; idx++) {
1882 struct pipe_buffer *buf = &bufs[idx];
1883 buf->ops->release(pipe, buf);
1884 }
1885out:
1886 kfree(bufs);
1887 return ret;
1888}
1889
1890static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1891{
1892 unsigned mask = POLLOUT | POLLWRNORM;
1893 struct fuse_conn *fc = fuse_get_conn(file);
1894 if (!fc)
1895 return POLLERR;
1896
1897 poll_wait(file, &fc->waitq, wait);
1898
1899 spin_lock(&fc->lock);
1900 if (!fc->connected)
1901 mask = POLLERR;
1902 else if (request_pending(fc))
1903 mask |= POLLIN | POLLRDNORM;
1904 spin_unlock(&fc->lock);
1905
1906 return mask;
1907}
1908
1909/*
1910 * Abort all requests on the given list (pending or processing)
1911 *
1912 * This function releases and reacquires fc->lock
1913 */
1914static void end_requests(struct fuse_conn *fc, struct list_head *head)
1915__releases(fc->lock)
1916__acquires(fc->lock)
1917{
1918 while (!list_empty(head)) {
1919 struct fuse_req *req;
1920 req = list_entry(head->next, struct fuse_req, list);
1921 req->out.h.error = -ECONNABORTED;
1922 request_end(fc, req);
1923 spin_lock(&fc->lock);
1924 }
1925}
1926
1927/*
1928 * Abort requests under I/O
1929 *
1930 * The requests are set to aborted and finished, and the request
1931 * waiter is woken up. This will make request_wait_answer() wait
1932 * until the request is unlocked and then return.
1933 *
1934 * If the request is asynchronous, then the end function needs to be
1935 * called after waiting for the request to be unlocked (if it was
1936 * locked).
1937 */
1938static void end_io_requests(struct fuse_conn *fc)
1939__releases(fc->lock)
1940__acquires(fc->lock)
1941{
1942 while (!list_empty(&fc->io)) {
1943 struct fuse_req *req =
1944 list_entry(fc->io.next, struct fuse_req, list);
1945 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
1946
1947 req->aborted = 1;
1948 req->out.h.error = -ECONNABORTED;
1949 req->state = FUSE_REQ_FINISHED;
1950 list_del_init(&req->list);
1951 wake_up(&req->waitq);
1952 if (end) {
1953 req->end = NULL;
1954 __fuse_get_request(req);
1955 spin_unlock(&fc->lock);
1956 wait_event(req->waitq, !req->locked);
1957 end(fc, req);
1958 fuse_put_request(fc, req);
1959 spin_lock(&fc->lock);
1960 }
1961 }
1962}
1963
1964static void end_queued_requests(struct fuse_conn *fc)
1965__releases(fc->lock)
1966__acquires(fc->lock)
1967{
1968 fc->max_background = UINT_MAX;
1969 flush_bg_queue(fc);
1970 end_requests(fc, &fc->pending);
1971 end_requests(fc, &fc->processing);
1972 while (forget_pending(fc))
1973 kfree(dequeue_forget(fc, 1, NULL));
1974}
1975
1976static void end_polls(struct fuse_conn *fc)
1977{
1978 struct rb_node *p;
1979
1980 p = rb_first(&fc->polled_files);
1981
1982 while (p) {
1983 struct fuse_file *ff;
1984 ff = rb_entry(p, struct fuse_file, polled_node);
1985 wake_up_interruptible_all(&ff->poll_wait);
1986
1987 p = rb_next(p);
1988 }
1989}
1990
1991/*
1992 * Abort all requests.
1993 *
1994 * Emergency exit in case of a malicious or accidental deadlock, or
1995 * just a hung filesystem.
1996 *
1997 * The same effect is usually achievable through killing the
1998 * filesystem daemon and all users of the filesystem. The exception
1999 * is the combination of an asynchronous request and the tricky
2000 * deadlock (see Documentation/filesystems/fuse.txt).
2001 *
2002 * During the aborting, progression of requests from the pending and
2003 * processing lists onto the io list, and progression of new requests
2004 * onto the pending list is prevented by req->connected being false.
2005 *
2006 * Progression of requests under I/O to the processing list is
2007 * prevented by the req->aborted flag being true for these requests.
2008 * For this reason requests on the io list must be aborted first.
2009 */
2010void fuse_abort_conn(struct fuse_conn *fc)
2011{
2012 spin_lock(&fc->lock);
2013 if (fc->connected) {
2014 fc->connected = 0;
2015 fc->blocked = 0;
2016 end_io_requests(fc);
2017 end_queued_requests(fc);
2018 end_polls(fc);
2019 wake_up_all(&fc->waitq);
2020 wake_up_all(&fc->blocked_waitq);
2021 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2022 }
2023 spin_unlock(&fc->lock);
2024}
2025EXPORT_SYMBOL_GPL(fuse_abort_conn);
2026
2027int fuse_dev_release(struct inode *inode, struct file *file)
2028{
2029 struct fuse_conn *fc = fuse_get_conn(file);
2030 if (fc) {
2031 spin_lock(&fc->lock);
2032 fc->connected = 0;
2033 fc->blocked = 0;
2034 end_queued_requests(fc);
2035 end_polls(fc);
2036 wake_up_all(&fc->blocked_waitq);
2037 spin_unlock(&fc->lock);
2038 fuse_conn_put(fc);
2039 }
2040
2041 return 0;
2042}
2043EXPORT_SYMBOL_GPL(fuse_dev_release);
2044
2045static int fuse_dev_fasync(int fd, struct file *file, int on)
2046{
2047 struct fuse_conn *fc = fuse_get_conn(file);
2048 if (!fc)
2049 return -EPERM;
2050
2051 /* No locking - fasync_helper does its own locking */
2052 return fasync_helper(fd, file, on, &fc->fasync);
2053}
2054
2055const struct file_operations fuse_dev_operations = {
2056 .owner = THIS_MODULE,
2057 .llseek = no_llseek,
2058 .read = do_sync_read,
2059 .aio_read = fuse_dev_read,
2060 .splice_read = fuse_dev_splice_read,
2061 .write = do_sync_write,
2062 .aio_write = fuse_dev_write,
2063 .splice_write = fuse_dev_splice_write,
2064 .poll = fuse_dev_poll,
2065 .release = fuse_dev_release,
2066 .fasync = fuse_dev_fasync,
2067};
2068EXPORT_SYMBOL_GPL(fuse_dev_operations);
2069
2070static struct miscdevice fuse_miscdevice = {
2071 .minor = FUSE_MINOR,
2072 .name = "fuse",
2073 .fops = &fuse_dev_operations,
2074};
2075
2076int __init fuse_dev_init(void)
2077{
2078 int err = -ENOMEM;
2079 fuse_req_cachep = kmem_cache_create("fuse_request",
2080 sizeof(struct fuse_req),
2081 0, 0, NULL);
2082 if (!fuse_req_cachep)
2083 goto out;
2084
2085 err = misc_register(&fuse_miscdevice);
2086 if (err)
2087 goto out_cache_clean;
2088
2089 return 0;
2090
2091 out_cache_clean:
2092 kmem_cache_destroy(fuse_req_cachep);
2093 out:
2094 return err;
2095}
2096
2097void fuse_dev_cleanup(void)
2098{
2099 misc_deregister(&fuse_miscdevice);
2100 kmem_cache_destroy(fuse_req_cachep);
2101}
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/poll.h>
14#include <linux/sched/signal.h>
15#include <linux/uio.h>
16#include <linux/miscdevice.h>
17#include <linux/pagemap.h>
18#include <linux/file.h>
19#include <linux/slab.h>
20#include <linux/pipe_fs_i.h>
21#include <linux/swap.h>
22#include <linux/splice.h>
23#include <linux/sched.h>
24
25MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26MODULE_ALIAS("devname:fuse");
27
28static struct kmem_cache *fuse_req_cachep;
29
30static struct fuse_dev *fuse_get_dev(struct file *file)
31{
32 /*
33 * Lockless access is OK, because file->private data is set
34 * once during mount and is valid until the file is released.
35 */
36 return READ_ONCE(file->private_data);
37}
38
39static void fuse_request_init(struct fuse_req *req, struct page **pages,
40 struct fuse_page_desc *page_descs,
41 unsigned npages)
42{
43 memset(req, 0, sizeof(*req));
44 memset(pages, 0, sizeof(*pages) * npages);
45 memset(page_descs, 0, sizeof(*page_descs) * npages);
46 INIT_LIST_HEAD(&req->list);
47 INIT_LIST_HEAD(&req->intr_entry);
48 init_waitqueue_head(&req->waitq);
49 refcount_set(&req->count, 1);
50 req->pages = pages;
51 req->page_descs = page_descs;
52 req->max_pages = npages;
53 __set_bit(FR_PENDING, &req->flags);
54}
55
56static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
57{
58 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
59 if (req) {
60 struct page **pages;
61 struct fuse_page_desc *page_descs;
62
63 if (npages <= FUSE_REQ_INLINE_PAGES) {
64 pages = req->inline_pages;
65 page_descs = req->inline_page_descs;
66 } else {
67 pages = kmalloc(sizeof(struct page *) * npages, flags);
68 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
69 npages, flags);
70 }
71
72 if (!pages || !page_descs) {
73 kfree(pages);
74 kfree(page_descs);
75 kmem_cache_free(fuse_req_cachep, req);
76 return NULL;
77 }
78
79 fuse_request_init(req, pages, page_descs, npages);
80 }
81 return req;
82}
83
84struct fuse_req *fuse_request_alloc(unsigned npages)
85{
86 return __fuse_request_alloc(npages, GFP_KERNEL);
87}
88EXPORT_SYMBOL_GPL(fuse_request_alloc);
89
90struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
91{
92 return __fuse_request_alloc(npages, GFP_NOFS);
93}
94
95void fuse_request_free(struct fuse_req *req)
96{
97 if (req->pages != req->inline_pages) {
98 kfree(req->pages);
99 kfree(req->page_descs);
100 }
101 kmem_cache_free(fuse_req_cachep, req);
102}
103
104void __fuse_get_request(struct fuse_req *req)
105{
106 refcount_inc(&req->count);
107}
108
109/* Must be called with > 1 refcount */
110static void __fuse_put_request(struct fuse_req *req)
111{
112 refcount_dec(&req->count);
113}
114
115static void fuse_req_init_context(struct fuse_conn *fc, struct fuse_req *req)
116{
117 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
118 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
119 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
120}
121
122void fuse_set_initialized(struct fuse_conn *fc)
123{
124 /* Make sure stores before this are seen on another CPU */
125 smp_wmb();
126 fc->initialized = 1;
127}
128
129static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
130{
131 return !fc->initialized || (for_background && fc->blocked);
132}
133
134static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
135 bool for_background)
136{
137 struct fuse_req *req;
138 int err;
139 atomic_inc(&fc->num_waiting);
140
141 if (fuse_block_alloc(fc, for_background)) {
142 err = -EINTR;
143 if (wait_event_killable_exclusive(fc->blocked_waitq,
144 !fuse_block_alloc(fc, for_background)))
145 goto out;
146 }
147 /* Matches smp_wmb() in fuse_set_initialized() */
148 smp_rmb();
149
150 err = -ENOTCONN;
151 if (!fc->connected)
152 goto out;
153
154 err = -ECONNREFUSED;
155 if (fc->conn_error)
156 goto out;
157
158 req = fuse_request_alloc(npages);
159 err = -ENOMEM;
160 if (!req) {
161 if (for_background)
162 wake_up(&fc->blocked_waitq);
163 goto out;
164 }
165
166 fuse_req_init_context(fc, req);
167 __set_bit(FR_WAITING, &req->flags);
168 if (for_background)
169 __set_bit(FR_BACKGROUND, &req->flags);
170
171 return req;
172
173 out:
174 atomic_dec(&fc->num_waiting);
175 return ERR_PTR(err);
176}
177
178struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
179{
180 return __fuse_get_req(fc, npages, false);
181}
182EXPORT_SYMBOL_GPL(fuse_get_req);
183
184struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
185 unsigned npages)
186{
187 return __fuse_get_req(fc, npages, true);
188}
189EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
190
191/*
192 * Return request in fuse_file->reserved_req. However that may
193 * currently be in use. If that is the case, wait for it to become
194 * available.
195 */
196static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
197 struct file *file)
198{
199 struct fuse_req *req = NULL;
200 struct fuse_file *ff = file->private_data;
201
202 do {
203 wait_event(fc->reserved_req_waitq, ff->reserved_req);
204 spin_lock(&fc->lock);
205 if (ff->reserved_req) {
206 req = ff->reserved_req;
207 ff->reserved_req = NULL;
208 req->stolen_file = get_file(file);
209 }
210 spin_unlock(&fc->lock);
211 } while (!req);
212
213 return req;
214}
215
216/*
217 * Put stolen request back into fuse_file->reserved_req
218 */
219static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
220{
221 struct file *file = req->stolen_file;
222 struct fuse_file *ff = file->private_data;
223
224 spin_lock(&fc->lock);
225 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
226 BUG_ON(ff->reserved_req);
227 ff->reserved_req = req;
228 wake_up_all(&fc->reserved_req_waitq);
229 spin_unlock(&fc->lock);
230 fput(file);
231}
232
233/*
234 * Gets a requests for a file operation, always succeeds
235 *
236 * This is used for sending the FLUSH request, which must get to
237 * userspace, due to POSIX locks which may need to be unlocked.
238 *
239 * If allocation fails due to OOM, use the reserved request in
240 * fuse_file.
241 *
242 * This is very unlikely to deadlock accidentally, since the
243 * filesystem should not have it's own file open. If deadlock is
244 * intentional, it can still be broken by "aborting" the filesystem.
245 */
246struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
247 struct file *file)
248{
249 struct fuse_req *req;
250
251 atomic_inc(&fc->num_waiting);
252 wait_event(fc->blocked_waitq, fc->initialized);
253 /* Matches smp_wmb() in fuse_set_initialized() */
254 smp_rmb();
255 req = fuse_request_alloc(0);
256 if (!req)
257 req = get_reserved_req(fc, file);
258
259 fuse_req_init_context(fc, req);
260 __set_bit(FR_WAITING, &req->flags);
261 __clear_bit(FR_BACKGROUND, &req->flags);
262 return req;
263}
264
265void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
266{
267 if (refcount_dec_and_test(&req->count)) {
268 if (test_bit(FR_BACKGROUND, &req->flags)) {
269 /*
270 * We get here in the unlikely case that a background
271 * request was allocated but not sent
272 */
273 spin_lock(&fc->lock);
274 if (!fc->blocked)
275 wake_up(&fc->blocked_waitq);
276 spin_unlock(&fc->lock);
277 }
278
279 if (test_bit(FR_WAITING, &req->flags)) {
280 __clear_bit(FR_WAITING, &req->flags);
281 atomic_dec(&fc->num_waiting);
282 }
283
284 if (req->stolen_file)
285 put_reserved_req(fc, req);
286 else
287 fuse_request_free(req);
288 }
289}
290EXPORT_SYMBOL_GPL(fuse_put_request);
291
292static unsigned len_args(unsigned numargs, struct fuse_arg *args)
293{
294 unsigned nbytes = 0;
295 unsigned i;
296
297 for (i = 0; i < numargs; i++)
298 nbytes += args[i].size;
299
300 return nbytes;
301}
302
303static u64 fuse_get_unique(struct fuse_iqueue *fiq)
304{
305 return ++fiq->reqctr;
306}
307
308static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
309{
310 req->in.h.len = sizeof(struct fuse_in_header) +
311 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
312 list_add_tail(&req->list, &fiq->pending);
313 wake_up_locked(&fiq->waitq);
314 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
315}
316
317void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
318 u64 nodeid, u64 nlookup)
319{
320 struct fuse_iqueue *fiq = &fc->iq;
321
322 forget->forget_one.nodeid = nodeid;
323 forget->forget_one.nlookup = nlookup;
324
325 spin_lock(&fiq->waitq.lock);
326 if (fiq->connected) {
327 fiq->forget_list_tail->next = forget;
328 fiq->forget_list_tail = forget;
329 wake_up_locked(&fiq->waitq);
330 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
331 } else {
332 kfree(forget);
333 }
334 spin_unlock(&fiq->waitq.lock);
335}
336
337static void flush_bg_queue(struct fuse_conn *fc)
338{
339 while (fc->active_background < fc->max_background &&
340 !list_empty(&fc->bg_queue)) {
341 struct fuse_req *req;
342 struct fuse_iqueue *fiq = &fc->iq;
343
344 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
345 list_del(&req->list);
346 fc->active_background++;
347 spin_lock(&fiq->waitq.lock);
348 req->in.h.unique = fuse_get_unique(fiq);
349 queue_request(fiq, req);
350 spin_unlock(&fiq->waitq.lock);
351 }
352}
353
354/*
355 * This function is called when a request is finished. Either a reply
356 * has arrived or it was aborted (and not yet sent) or some error
357 * occurred during communication with userspace, or the device file
358 * was closed. The requester thread is woken up (if still waiting),
359 * the 'end' callback is called if given, else the reference to the
360 * request is released
361 */
362static void request_end(struct fuse_conn *fc, struct fuse_req *req)
363{
364 struct fuse_iqueue *fiq = &fc->iq;
365
366 if (test_and_set_bit(FR_FINISHED, &req->flags))
367 return;
368
369 spin_lock(&fiq->waitq.lock);
370 list_del_init(&req->intr_entry);
371 spin_unlock(&fiq->waitq.lock);
372 WARN_ON(test_bit(FR_PENDING, &req->flags));
373 WARN_ON(test_bit(FR_SENT, &req->flags));
374 if (test_bit(FR_BACKGROUND, &req->flags)) {
375 spin_lock(&fc->lock);
376 clear_bit(FR_BACKGROUND, &req->flags);
377 if (fc->num_background == fc->max_background)
378 fc->blocked = 0;
379
380 /* Wake up next waiter, if any */
381 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
382 wake_up(&fc->blocked_waitq);
383
384 if (fc->num_background == fc->congestion_threshold &&
385 fc->connected && fc->sb) {
386 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
387 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
388 }
389 fc->num_background--;
390 fc->active_background--;
391 flush_bg_queue(fc);
392 spin_unlock(&fc->lock);
393 }
394 wake_up(&req->waitq);
395 if (req->end)
396 req->end(fc, req);
397 fuse_put_request(fc, req);
398}
399
400static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
401{
402 spin_lock(&fiq->waitq.lock);
403 if (test_bit(FR_FINISHED, &req->flags)) {
404 spin_unlock(&fiq->waitq.lock);
405 return;
406 }
407 if (list_empty(&req->intr_entry)) {
408 list_add_tail(&req->intr_entry, &fiq->interrupts);
409 wake_up_locked(&fiq->waitq);
410 }
411 spin_unlock(&fiq->waitq.lock);
412 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
413}
414
415static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
416{
417 struct fuse_iqueue *fiq = &fc->iq;
418 int err;
419
420 if (!fc->no_interrupt) {
421 /* Any signal may interrupt this */
422 err = wait_event_interruptible(req->waitq,
423 test_bit(FR_FINISHED, &req->flags));
424 if (!err)
425 return;
426
427 set_bit(FR_INTERRUPTED, &req->flags);
428 /* matches barrier in fuse_dev_do_read() */
429 smp_mb__after_atomic();
430 if (test_bit(FR_SENT, &req->flags))
431 queue_interrupt(fiq, req);
432 }
433
434 if (!test_bit(FR_FORCE, &req->flags)) {
435 /* Only fatal signals may interrupt this */
436 err = wait_event_killable(req->waitq,
437 test_bit(FR_FINISHED, &req->flags));
438 if (!err)
439 return;
440
441 spin_lock(&fiq->waitq.lock);
442 /* Request is not yet in userspace, bail out */
443 if (test_bit(FR_PENDING, &req->flags)) {
444 list_del(&req->list);
445 spin_unlock(&fiq->waitq.lock);
446 __fuse_put_request(req);
447 req->out.h.error = -EINTR;
448 return;
449 }
450 spin_unlock(&fiq->waitq.lock);
451 }
452
453 /*
454 * Either request is already in userspace, or it was forced.
455 * Wait it out.
456 */
457 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
458}
459
460static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
461{
462 struct fuse_iqueue *fiq = &fc->iq;
463
464 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
465 spin_lock(&fiq->waitq.lock);
466 if (!fiq->connected) {
467 spin_unlock(&fiq->waitq.lock);
468 req->out.h.error = -ENOTCONN;
469 } else {
470 req->in.h.unique = fuse_get_unique(fiq);
471 queue_request(fiq, req);
472 /* acquire extra reference, since request is still needed
473 after request_end() */
474 __fuse_get_request(req);
475 spin_unlock(&fiq->waitq.lock);
476
477 request_wait_answer(fc, req);
478 /* Pairs with smp_wmb() in request_end() */
479 smp_rmb();
480 }
481}
482
483void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
484{
485 __set_bit(FR_ISREPLY, &req->flags);
486 if (!test_bit(FR_WAITING, &req->flags)) {
487 __set_bit(FR_WAITING, &req->flags);
488 atomic_inc(&fc->num_waiting);
489 }
490 __fuse_request_send(fc, req);
491}
492EXPORT_SYMBOL_GPL(fuse_request_send);
493
494static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
495{
496 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
497 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
498
499 if (fc->minor < 9) {
500 switch (args->in.h.opcode) {
501 case FUSE_LOOKUP:
502 case FUSE_CREATE:
503 case FUSE_MKNOD:
504 case FUSE_MKDIR:
505 case FUSE_SYMLINK:
506 case FUSE_LINK:
507 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
508 break;
509 case FUSE_GETATTR:
510 case FUSE_SETATTR:
511 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
512 break;
513 }
514 }
515 if (fc->minor < 12) {
516 switch (args->in.h.opcode) {
517 case FUSE_CREATE:
518 args->in.args[0].size = sizeof(struct fuse_open_in);
519 break;
520 case FUSE_MKNOD:
521 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
522 break;
523 }
524 }
525}
526
527ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
528{
529 struct fuse_req *req;
530 ssize_t ret;
531
532 req = fuse_get_req(fc, 0);
533 if (IS_ERR(req))
534 return PTR_ERR(req);
535
536 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
537 fuse_adjust_compat(fc, args);
538
539 req->in.h.opcode = args->in.h.opcode;
540 req->in.h.nodeid = args->in.h.nodeid;
541 req->in.numargs = args->in.numargs;
542 memcpy(req->in.args, args->in.args,
543 args->in.numargs * sizeof(struct fuse_in_arg));
544 req->out.argvar = args->out.argvar;
545 req->out.numargs = args->out.numargs;
546 memcpy(req->out.args, args->out.args,
547 args->out.numargs * sizeof(struct fuse_arg));
548 fuse_request_send(fc, req);
549 ret = req->out.h.error;
550 if (!ret && args->out.argvar) {
551 BUG_ON(args->out.numargs != 1);
552 ret = req->out.args[0].size;
553 }
554 fuse_put_request(fc, req);
555
556 return ret;
557}
558
559/*
560 * Called under fc->lock
561 *
562 * fc->connected must have been checked previously
563 */
564void fuse_request_send_background_locked(struct fuse_conn *fc,
565 struct fuse_req *req)
566{
567 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
568 if (!test_bit(FR_WAITING, &req->flags)) {
569 __set_bit(FR_WAITING, &req->flags);
570 atomic_inc(&fc->num_waiting);
571 }
572 __set_bit(FR_ISREPLY, &req->flags);
573 fc->num_background++;
574 if (fc->num_background == fc->max_background)
575 fc->blocked = 1;
576 if (fc->num_background == fc->congestion_threshold && fc->sb) {
577 set_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
578 set_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
579 }
580 list_add_tail(&req->list, &fc->bg_queue);
581 flush_bg_queue(fc);
582}
583
584void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
585{
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);
596 }
597}
598EXPORT_SYMBOL_GPL(fuse_request_send_background);
599
600static int fuse_request_send_notify_reply(struct fuse_conn *fc,
601 struct fuse_req *req, u64 unique)
602{
603 int err = -ENODEV;
604 struct fuse_iqueue *fiq = &fc->iq;
605
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;
612 }
613 spin_unlock(&fiq->waitq.lock);
614
615 return err;
616}
617
618void fuse_force_forget(struct file *file, u64 nodeid)
619{
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;
624
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);
637}
638
639/*
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.
643 */
644static int lock_request(struct fuse_req *req)
645{
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);
654 }
655 return err;
656}
657
658/*
659 * Unlock request. If it was aborted while locked, caller is responsible
660 * for unlocking and ending the request.
661 */
662static int unlock_request(struct fuse_req *req)
663{
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);
672 }
673 return err;
674}
675
676struct 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;
688};
689
690static void fuse_copy_init(struct fuse_copy_state *cs, int write,
691 struct iov_iter *iter)
692{
693 memset(cs, 0, sizeof(*cs));
694 cs->write = write;
695 cs->iter = iter;
696}
697
698/* Unmap and put previous page of userspace buffer */
699static void fuse_copy_finish(struct fuse_copy_state *cs)
700{
701 if (cs->currbuf) {
702 struct pipe_buffer *buf = cs->currbuf;
703
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);
711 }
712 put_page(cs->pg);
713 }
714 cs->pg = NULL;
715}
716
717/*
718 * Get another pagefull of userspace buffer, and map it to kernel
719 * address space, and lock request
720 */
721static int fuse_copy_fill(struct fuse_copy_state *cs)
722{
723 struct page *page;
724 int err;
725
726 err = unlock_request(cs->req);
727 if (err)
728 return err;
729
730 fuse_copy_finish(cs);
731 if (cs->pipebufs) {
732 struct pipe_buffer *buf = cs->pipebufs;
733
734 if (!cs->write) {
735 err = pipe_buf_confirm(cs->pipe, buf);
736 if (err)
737 return err;
738
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;
749
750 page = alloc_page(GFP_HIGHUSER);
751 if (!page)
752 return -ENOMEM;
753
754 buf->page = page;
755 buf->offset = 0;
756 buf->len = 0;
757
758 cs->currbuf = buf;
759 cs->pg = page;
760 cs->offset = 0;
761 cs->len = PAGE_SIZE;
762 cs->pipebufs++;
763 cs->nr_segs++;
764 }
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);
775 }
776
777 return lock_request(cs->req);
778}
779
780/* Do as much copy to/from userspace buffer as we can */
781static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
782{
783 unsigned ncpy = min(*size, cs->len);
784 if (val) {
785 void *pgaddr = kmap_atomic(cs->pg);
786 void *buf = pgaddr + cs->offset;
787
788 if (cs->write)
789 memcpy(buf, *val, ncpy);
790 else
791 memcpy(*val, buf, ncpy);
792
793 kunmap_atomic(pgaddr);
794 *val += ncpy;
795 }
796 *size -= ncpy;
797 cs->len -= ncpy;
798 cs->offset += ncpy;
799 return ncpy;
800}
801
802static int fuse_check_page(struct page *page)
803{
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;
817 }
818 return 0;
819}
820
821static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
822{
823 int err;
824 struct page *oldpage = *pagep;
825 struct page *newpage;
826 struct pipe_buffer *buf = cs->pipebufs;
827
828 err = unlock_request(cs->req);
829 if (err)
830 return err;
831
832 fuse_copy_finish(cs);
833
834 err = pipe_buf_confirm(cs->pipe, buf);
835 if (err)
836 return err;
837
838 BUG_ON(!cs->nr_segs);
839 cs->currbuf = buf;
840 cs->len = buf->len;
841 cs->pipebufs++;
842 cs->nr_segs--;
843
844 if (cs->len != PAGE_SIZE)
845 goto out_fallback;
846
847 if (pipe_buf_steal(cs->pipe, buf) != 0)
848 goto out_fallback;
849
850 newpage = buf->page;
851
852 if (!PageUptodate(newpage))
853 SetPageUptodate(newpage);
854
855 ClearPageMappedToDisk(newpage);
856
857 if (fuse_check_page(newpage) != 0)
858 goto out_fallback_unlock;
859
860 /*
861 * This is a new and locked page, it shouldn't be mapped or
862 * have any special flags on it
863 */
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;
872
873 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
874 if (err) {
875 unlock_page(newpage);
876 return err;
877 }
878
879 get_page(newpage);
880
881 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
882 lru_cache_add_file(newpage);
883
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);
891
892 if (err) {
893 unlock_page(newpage);
894 put_page(newpage);
895 return err;
896 }
897
898 unlock_page(oldpage);
899 put_page(oldpage);
900 cs->len = 0;
901
902 return 0;
903
904out_fallback_unlock:
905 unlock_page(newpage);
906out_fallback:
907 cs->pg = buf->page;
908 cs->offset = buf->offset;
909
910 err = lock_request(cs->req);
911 if (err)
912 return err;
913
914 return 1;
915}
916
917static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
918 unsigned offset, unsigned count)
919{
920 struct pipe_buffer *buf;
921 int err;
922
923 if (cs->nr_segs == cs->pipe->buffers)
924 return -EIO;
925
926 err = unlock_request(cs->req);
927 if (err)
928 return err;
929
930 fuse_copy_finish(cs);
931
932 buf = cs->pipebufs;
933 get_page(page);
934 buf->page = page;
935 buf->offset = offset;
936 buf->len = count;
937
938 cs->pipebufs++;
939 cs->nr_segs++;
940 cs->len = 0;
941
942 return 0;
943}
944
945/*
946 * Copy a page in the request to/from the userspace buffer. Must be
947 * done atomically
948 */
949static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
950 unsigned offset, unsigned count, int zeroing)
951{
952 int err;
953 struct page *page = *pagep;
954
955 if (page && zeroing && count < PAGE_SIZE)
956 clear_highpage(page);
957
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;
971 }
972 }
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);
980 }
981 if (page && !cs->write)
982 flush_dcache_page(page);
983 return 0;
984}
985
986/* Copy pages in the request to/from userspace buffer */
987static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
988 int zeroing)
989{
990 unsigned i;
991 struct fuse_req *req = cs->req;
992
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);
997
998 err = fuse_copy_page(cs, &req->pages[i], offset, count,
999 zeroing);
1000 if (err)
1001 return err;
1002
1003 nbytes -= count;
1004 }
1005 return 0;
1006}
1007
1008/* Copy a single argument in the request to/from userspace buffer */
1009static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1010{
1011 while (size) {
1012 if (!cs->len) {
1013 int err = fuse_copy_fill(cs);
1014 if (err)
1015 return err;
1016 }
1017 fuse_copy_do(cs, &val, &size);
1018 }
1019 return 0;
1020}
1021
1022/* Copy request arguments to/from userspace buffer */
1023static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1024 unsigned argpages, struct fuse_arg *args,
1025 int zeroing)
1026{
1027 int err = 0;
1028 unsigned i;
1029
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);
1036 }
1037 return err;
1038}
1039
1040static int forget_pending(struct fuse_iqueue *fiq)
1041{
1042 return fiq->forget_list_head.next != NULL;
1043}
1044
1045static int request_pending(struct fuse_iqueue *fiq)
1046{
1047 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1048 forget_pending(fiq);
1049}
1050
1051/*
1052 * Transfer an interrupt request to userspace
1053 *
1054 * Unlike other requests this is assembled on demand, without a need
1055 * to allocate a separate fuse_req structure.
1056 *
1057 * Called with fiq->waitq.lock held, releases it
1058 */
1059static 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)
1063{
1064 struct fuse_in_header ih;
1065 struct fuse_interrupt_in arg;
1066 unsigned reqsize = sizeof(ih) + sizeof(arg);
1067 int err;
1068
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;
1077
1078 spin_unlock(&fiq->waitq.lock);
1079 if (nbytes < reqsize)
1080 return -EINVAL;
1081
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);
1086
1087 return err ? err : reqsize;
1088}
1089
1090static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1091 unsigned max,
1092 unsigned *countp)
1093{
1094 struct fuse_forget_link *head = fiq->forget_list_head.next;
1095 struct fuse_forget_link **newhead = &head;
1096 unsigned count;
1097
1098 for (count = 0; *newhead != NULL && count < max; count++)
1099 newhead = &(*newhead)->next;
1100
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;
1105
1106 if (countp != NULL)
1107 *countp = count;
1108
1109 return head;
1110}
1111
1112static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1113 struct fuse_copy_state *cs,
1114 size_t nbytes)
1115__releases(fiq->waitq.lock)
1116{
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,
1121 };
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),
1127 };
1128
1129 spin_unlock(&fiq->waitq.lock);
1130 kfree(forget);
1131 if (nbytes < ih.len)
1132 return -EINVAL;
1133
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);
1138
1139 if (err)
1140 return err;
1141
1142 return ih.len;
1143}
1144
1145static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1146 struct fuse_copy_state *cs, size_t nbytes)
1147__releases(fiq->waitq.lock)
1148{
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),
1158 };
1159
1160 if (nbytes < ih.len) {
1161 spin_unlock(&fiq->waitq.lock);
1162 return -EINVAL;
1163 }
1164
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);
1168
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));
1174
1175 while (head) {
1176 struct fuse_forget_link *forget = head;
1177
1178 if (!err) {
1179 err = fuse_copy_one(cs, &forget->forget_one,
1180 sizeof(forget->forget_one));
1181 }
1182 head = forget->next;
1183 kfree(forget);
1184 }
1185
1186 fuse_copy_finish(cs);
1187
1188 if (err)
1189 return err;
1190
1191 return ih.len;
1192}
1193
1194static 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)
1198{
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);
1203}
1204
1205/*
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.
1213 */
1214static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1215 struct fuse_copy_state *cs, size_t nbytes)
1216{
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;
1224
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;
1231
1232 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1233 !fiq->connected || request_pending(fiq));
1234 if (err)
1235 goto err_unlock;
1236
1237 err = -ENODEV;
1238 if (!fiq->connected)
1239 goto err_unlock;
1240
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);
1245 }
1246
1247 if (forget_pending(fiq)) {
1248 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1249 return fuse_read_forget(fc, fiq, cs, nbytes);
1250
1251 if (fiq->forget_batch <= -8)
1252 fiq->forget_batch = 16;
1253 }
1254
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);
1259
1260 in = &req->in;
1261 reqsize = in->h.len;
1262
1263 if (task_active_pid_ns(current) != fc->pid_ns) {
1264 rcu_read_lock();
1265 in->h.pid = pid_vnr(find_pid_ns(in->h.pid, fc->pid_ns));
1266 rcu_read_unlock();
1267 }
1268
1269 /* If request is too large, reply with an error and restart the read */
1270 if (nbytes < reqsize) {
1271 req->out.h.error = -EIO;
1272 /* SETXATTR is special, since it may contain too large data */
1273 if (in->h.opcode == FUSE_SETXATTR)
1274 req->out.h.error = -E2BIG;
1275 request_end(fc, req);
1276 goto restart;
1277 }
1278 spin_lock(&fpq->lock);
1279 list_add(&req->list, &fpq->io);
1280 spin_unlock(&fpq->lock);
1281 cs->req = req;
1282 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1283 if (!err)
1284 err = fuse_copy_args(cs, in->numargs, in->argpages,
1285 (struct fuse_arg *) in->args, 0);
1286 fuse_copy_finish(cs);
1287 spin_lock(&fpq->lock);
1288 clear_bit(FR_LOCKED, &req->flags);
1289 if (!fpq->connected) {
1290 err = -ENODEV;
1291 goto out_end;
1292 }
1293 if (err) {
1294 req->out.h.error = -EIO;
1295 goto out_end;
1296 }
1297 if (!test_bit(FR_ISREPLY, &req->flags)) {
1298 err = reqsize;
1299 goto out_end;
1300 }
1301 list_move_tail(&req->list, &fpq->processing);
1302 spin_unlock(&fpq->lock);
1303 set_bit(FR_SENT, &req->flags);
1304 /* matches barrier in request_wait_answer() */
1305 smp_mb__after_atomic();
1306 if (test_bit(FR_INTERRUPTED, &req->flags))
1307 queue_interrupt(fiq, req);
1308
1309 return reqsize;
1310
1311out_end:
1312 if (!test_bit(FR_PRIVATE, &req->flags))
1313 list_del_init(&req->list);
1314 spin_unlock(&fpq->lock);
1315 request_end(fc, req);
1316 return err;
1317
1318 err_unlock:
1319 spin_unlock(&fiq->waitq.lock);
1320 return err;
1321}
1322
1323static int fuse_dev_open(struct inode *inode, struct file *file)
1324{
1325 /*
1326 * The fuse device's file's private_data is used to hold
1327 * the fuse_conn(ection) when it is mounted, and is used to
1328 * keep track of whether the file has been mounted already.
1329 */
1330 file->private_data = NULL;
1331 return 0;
1332}
1333
1334static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1335{
1336 struct fuse_copy_state cs;
1337 struct file *file = iocb->ki_filp;
1338 struct fuse_dev *fud = fuse_get_dev(file);
1339
1340 if (!fud)
1341 return -EPERM;
1342
1343 if (!iter_is_iovec(to))
1344 return -EINVAL;
1345
1346 fuse_copy_init(&cs, 1, to);
1347
1348 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1349}
1350
1351static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1352 struct pipe_inode_info *pipe,
1353 size_t len, unsigned int flags)
1354{
1355 int total, ret;
1356 int page_nr = 0;
1357 struct pipe_buffer *bufs;
1358 struct fuse_copy_state cs;
1359 struct fuse_dev *fud = fuse_get_dev(in);
1360
1361 if (!fud)
1362 return -EPERM;
1363
1364 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1365 if (!bufs)
1366 return -ENOMEM;
1367
1368 fuse_copy_init(&cs, 1, NULL);
1369 cs.pipebufs = bufs;
1370 cs.pipe = pipe;
1371 ret = fuse_dev_do_read(fud, in, &cs, len);
1372 if (ret < 0)
1373 goto out;
1374
1375 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1376 ret = -EIO;
1377 goto out;
1378 }
1379
1380 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1381 /*
1382 * Need to be careful about this. Having buf->ops in module
1383 * code can Oops if the buffer persists after module unload.
1384 */
1385 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1386 bufs[page_nr].flags = 0;
1387 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1388 if (unlikely(ret < 0))
1389 break;
1390 }
1391 if (total)
1392 ret = total;
1393out:
1394 for (; page_nr < cs.nr_segs; page_nr++)
1395 put_page(bufs[page_nr].page);
1396
1397 kfree(bufs);
1398 return ret;
1399}
1400
1401static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1402 struct fuse_copy_state *cs)
1403{
1404 struct fuse_notify_poll_wakeup_out outarg;
1405 int err = -EINVAL;
1406
1407 if (size != sizeof(outarg))
1408 goto err;
1409
1410 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1411 if (err)
1412 goto err;
1413
1414 fuse_copy_finish(cs);
1415 return fuse_notify_poll_wakeup(fc, &outarg);
1416
1417err:
1418 fuse_copy_finish(cs);
1419 return err;
1420}
1421
1422static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1423 struct fuse_copy_state *cs)
1424{
1425 struct fuse_notify_inval_inode_out outarg;
1426 int err = -EINVAL;
1427
1428 if (size != sizeof(outarg))
1429 goto err;
1430
1431 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1432 if (err)
1433 goto err;
1434 fuse_copy_finish(cs);
1435
1436 down_read(&fc->killsb);
1437 err = -ENOENT;
1438 if (fc->sb) {
1439 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1440 outarg.off, outarg.len);
1441 }
1442 up_read(&fc->killsb);
1443 return err;
1444
1445err:
1446 fuse_copy_finish(cs);
1447 return err;
1448}
1449
1450static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1451 struct fuse_copy_state *cs)
1452{
1453 struct fuse_notify_inval_entry_out outarg;
1454 int err = -ENOMEM;
1455 char *buf;
1456 struct qstr name;
1457
1458 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1459 if (!buf)
1460 goto err;
1461
1462 err = -EINVAL;
1463 if (size < sizeof(outarg))
1464 goto err;
1465
1466 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1467 if (err)
1468 goto err;
1469
1470 err = -ENAMETOOLONG;
1471 if (outarg.namelen > FUSE_NAME_MAX)
1472 goto err;
1473
1474 err = -EINVAL;
1475 if (size != sizeof(outarg) + outarg.namelen + 1)
1476 goto err;
1477
1478 name.name = buf;
1479 name.len = outarg.namelen;
1480 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1481 if (err)
1482 goto err;
1483 fuse_copy_finish(cs);
1484 buf[outarg.namelen] = 0;
1485
1486 down_read(&fc->killsb);
1487 err = -ENOENT;
1488 if (fc->sb)
1489 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1490 up_read(&fc->killsb);
1491 kfree(buf);
1492 return err;
1493
1494err:
1495 kfree(buf);
1496 fuse_copy_finish(cs);
1497 return err;
1498}
1499
1500static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1501 struct fuse_copy_state *cs)
1502{
1503 struct fuse_notify_delete_out outarg;
1504 int err = -ENOMEM;
1505 char *buf;
1506 struct qstr name;
1507
1508 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1509 if (!buf)
1510 goto err;
1511
1512 err = -EINVAL;
1513 if (size < sizeof(outarg))
1514 goto err;
1515
1516 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1517 if (err)
1518 goto err;
1519
1520 err = -ENAMETOOLONG;
1521 if (outarg.namelen > FUSE_NAME_MAX)
1522 goto err;
1523
1524 err = -EINVAL;
1525 if (size != sizeof(outarg) + outarg.namelen + 1)
1526 goto err;
1527
1528 name.name = buf;
1529 name.len = outarg.namelen;
1530 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1531 if (err)
1532 goto err;
1533 fuse_copy_finish(cs);
1534 buf[outarg.namelen] = 0;
1535
1536 down_read(&fc->killsb);
1537 err = -ENOENT;
1538 if (fc->sb)
1539 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1540 outarg.child, &name);
1541 up_read(&fc->killsb);
1542 kfree(buf);
1543 return err;
1544
1545err:
1546 kfree(buf);
1547 fuse_copy_finish(cs);
1548 return err;
1549}
1550
1551static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1552 struct fuse_copy_state *cs)
1553{
1554 struct fuse_notify_store_out outarg;
1555 struct inode *inode;
1556 struct address_space *mapping;
1557 u64 nodeid;
1558 int err;
1559 pgoff_t index;
1560 unsigned int offset;
1561 unsigned int num;
1562 loff_t file_size;
1563 loff_t end;
1564
1565 err = -EINVAL;
1566 if (size < sizeof(outarg))
1567 goto out_finish;
1568
1569 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1570 if (err)
1571 goto out_finish;
1572
1573 err = -EINVAL;
1574 if (size - sizeof(outarg) != outarg.size)
1575 goto out_finish;
1576
1577 nodeid = outarg.nodeid;
1578
1579 down_read(&fc->killsb);
1580
1581 err = -ENOENT;
1582 if (!fc->sb)
1583 goto out_up_killsb;
1584
1585 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1586 if (!inode)
1587 goto out_up_killsb;
1588
1589 mapping = inode->i_mapping;
1590 index = outarg.offset >> PAGE_SHIFT;
1591 offset = outarg.offset & ~PAGE_MASK;
1592 file_size = i_size_read(inode);
1593 end = outarg.offset + outarg.size;
1594 if (end > file_size) {
1595 file_size = end;
1596 fuse_write_update_size(inode, file_size);
1597 }
1598
1599 num = outarg.size;
1600 while (num) {
1601 struct page *page;
1602 unsigned int this_num;
1603
1604 err = -ENOMEM;
1605 page = find_or_create_page(mapping, index,
1606 mapping_gfp_mask(mapping));
1607 if (!page)
1608 goto out_iput;
1609
1610 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1611 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1612 if (!err && offset == 0 &&
1613 (this_num == PAGE_SIZE || file_size == end))
1614 SetPageUptodate(page);
1615 unlock_page(page);
1616 put_page(page);
1617
1618 if (err)
1619 goto out_iput;
1620
1621 num -= this_num;
1622 offset = 0;
1623 index++;
1624 }
1625
1626 err = 0;
1627
1628out_iput:
1629 iput(inode);
1630out_up_killsb:
1631 up_read(&fc->killsb);
1632out_finish:
1633 fuse_copy_finish(cs);
1634 return err;
1635}
1636
1637static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1638{
1639 release_pages(req->pages, req->num_pages);
1640}
1641
1642static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1643 struct fuse_notify_retrieve_out *outarg)
1644{
1645 int err;
1646 struct address_space *mapping = inode->i_mapping;
1647 struct fuse_req *req;
1648 pgoff_t index;
1649 loff_t file_size;
1650 unsigned int num;
1651 unsigned int offset;
1652 size_t total_len = 0;
1653 int num_pages;
1654
1655 offset = outarg->offset & ~PAGE_MASK;
1656 file_size = i_size_read(inode);
1657
1658 num = outarg->size;
1659 if (outarg->offset > file_size)
1660 num = 0;
1661 else if (outarg->offset + num > file_size)
1662 num = file_size - outarg->offset;
1663
1664 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1665 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1666
1667 req = fuse_get_req(fc, num_pages);
1668 if (IS_ERR(req))
1669 return PTR_ERR(req);
1670
1671 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1672 req->in.h.nodeid = outarg->nodeid;
1673 req->in.numargs = 2;
1674 req->in.argpages = 1;
1675 req->page_descs[0].offset = offset;
1676 req->end = fuse_retrieve_end;
1677
1678 index = outarg->offset >> PAGE_SHIFT;
1679
1680 while (num && req->num_pages < num_pages) {
1681 struct page *page;
1682 unsigned int this_num;
1683
1684 page = find_get_page(mapping, index);
1685 if (!page)
1686 break;
1687
1688 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1689 req->pages[req->num_pages] = page;
1690 req->page_descs[req->num_pages].length = this_num;
1691 req->num_pages++;
1692
1693 offset = 0;
1694 num -= this_num;
1695 total_len += this_num;
1696 index++;
1697 }
1698 req->misc.retrieve_in.offset = outarg->offset;
1699 req->misc.retrieve_in.size = total_len;
1700 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1701 req->in.args[0].value = &req->misc.retrieve_in;
1702 req->in.args[1].size = total_len;
1703
1704 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1705 if (err)
1706 fuse_retrieve_end(fc, req);
1707
1708 return err;
1709}
1710
1711static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1712 struct fuse_copy_state *cs)
1713{
1714 struct fuse_notify_retrieve_out outarg;
1715 struct inode *inode;
1716 int err;
1717
1718 err = -EINVAL;
1719 if (size != sizeof(outarg))
1720 goto copy_finish;
1721
1722 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1723 if (err)
1724 goto copy_finish;
1725
1726 fuse_copy_finish(cs);
1727
1728 down_read(&fc->killsb);
1729 err = -ENOENT;
1730 if (fc->sb) {
1731 u64 nodeid = outarg.nodeid;
1732
1733 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1734 if (inode) {
1735 err = fuse_retrieve(fc, inode, &outarg);
1736 iput(inode);
1737 }
1738 }
1739 up_read(&fc->killsb);
1740
1741 return err;
1742
1743copy_finish:
1744 fuse_copy_finish(cs);
1745 return err;
1746}
1747
1748static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1749 unsigned int size, struct fuse_copy_state *cs)
1750{
1751 /* Don't try to move pages (yet) */
1752 cs->move_pages = 0;
1753
1754 switch (code) {
1755 case FUSE_NOTIFY_POLL:
1756 return fuse_notify_poll(fc, size, cs);
1757
1758 case FUSE_NOTIFY_INVAL_INODE:
1759 return fuse_notify_inval_inode(fc, size, cs);
1760
1761 case FUSE_NOTIFY_INVAL_ENTRY:
1762 return fuse_notify_inval_entry(fc, size, cs);
1763
1764 case FUSE_NOTIFY_STORE:
1765 return fuse_notify_store(fc, size, cs);
1766
1767 case FUSE_NOTIFY_RETRIEVE:
1768 return fuse_notify_retrieve(fc, size, cs);
1769
1770 case FUSE_NOTIFY_DELETE:
1771 return fuse_notify_delete(fc, size, cs);
1772
1773 default:
1774 fuse_copy_finish(cs);
1775 return -EINVAL;
1776 }
1777}
1778
1779/* Look up request on processing list by unique ID */
1780static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1781{
1782 struct fuse_req *req;
1783
1784 list_for_each_entry(req, &fpq->processing, list) {
1785 if (req->in.h.unique == unique || req->intr_unique == unique)
1786 return req;
1787 }
1788 return NULL;
1789}
1790
1791static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1792 unsigned nbytes)
1793{
1794 unsigned reqsize = sizeof(struct fuse_out_header);
1795
1796 if (out->h.error)
1797 return nbytes != reqsize ? -EINVAL : 0;
1798
1799 reqsize += len_args(out->numargs, out->args);
1800
1801 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1802 return -EINVAL;
1803 else if (reqsize > nbytes) {
1804 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1805 unsigned diffsize = reqsize - nbytes;
1806 if (diffsize > lastarg->size)
1807 return -EINVAL;
1808 lastarg->size -= diffsize;
1809 }
1810 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1811 out->page_zeroing);
1812}
1813
1814/*
1815 * Write a single reply to a request. First the header is copied from
1816 * the write buffer. The request is then searched on the processing
1817 * list by the unique ID found in the header. If found, then remove
1818 * it from the list and copy the rest of the buffer to the request.
1819 * The request is finished by calling request_end()
1820 */
1821static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1822 struct fuse_copy_state *cs, size_t nbytes)
1823{
1824 int err;
1825 struct fuse_conn *fc = fud->fc;
1826 struct fuse_pqueue *fpq = &fud->pq;
1827 struct fuse_req *req;
1828 struct fuse_out_header oh;
1829
1830 if (nbytes < sizeof(struct fuse_out_header))
1831 return -EINVAL;
1832
1833 err = fuse_copy_one(cs, &oh, sizeof(oh));
1834 if (err)
1835 goto err_finish;
1836
1837 err = -EINVAL;
1838 if (oh.len != nbytes)
1839 goto err_finish;
1840
1841 /*
1842 * Zero oh.unique indicates unsolicited notification message
1843 * and error contains notification code.
1844 */
1845 if (!oh.unique) {
1846 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1847 return err ? err : nbytes;
1848 }
1849
1850 err = -EINVAL;
1851 if (oh.error <= -1000 || oh.error > 0)
1852 goto err_finish;
1853
1854 spin_lock(&fpq->lock);
1855 err = -ENOENT;
1856 if (!fpq->connected)
1857 goto err_unlock_pq;
1858
1859 req = request_find(fpq, oh.unique);
1860 if (!req)
1861 goto err_unlock_pq;
1862
1863 /* Is it an interrupt reply? */
1864 if (req->intr_unique == oh.unique) {
1865 spin_unlock(&fpq->lock);
1866
1867 err = -EINVAL;
1868 if (nbytes != sizeof(struct fuse_out_header))
1869 goto err_finish;
1870
1871 if (oh.error == -ENOSYS)
1872 fc->no_interrupt = 1;
1873 else if (oh.error == -EAGAIN)
1874 queue_interrupt(&fc->iq, req);
1875
1876 fuse_copy_finish(cs);
1877 return nbytes;
1878 }
1879
1880 clear_bit(FR_SENT, &req->flags);
1881 list_move(&req->list, &fpq->io);
1882 req->out.h = oh;
1883 set_bit(FR_LOCKED, &req->flags);
1884 spin_unlock(&fpq->lock);
1885 cs->req = req;
1886 if (!req->out.page_replace)
1887 cs->move_pages = 0;
1888
1889 err = copy_out_args(cs, &req->out, nbytes);
1890 fuse_copy_finish(cs);
1891
1892 spin_lock(&fpq->lock);
1893 clear_bit(FR_LOCKED, &req->flags);
1894 if (!fpq->connected)
1895 err = -ENOENT;
1896 else if (err)
1897 req->out.h.error = -EIO;
1898 if (!test_bit(FR_PRIVATE, &req->flags))
1899 list_del_init(&req->list);
1900 spin_unlock(&fpq->lock);
1901
1902 request_end(fc, req);
1903
1904 return err ? err : nbytes;
1905
1906 err_unlock_pq:
1907 spin_unlock(&fpq->lock);
1908 err_finish:
1909 fuse_copy_finish(cs);
1910 return err;
1911}
1912
1913static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1914{
1915 struct fuse_copy_state cs;
1916 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1917
1918 if (!fud)
1919 return -EPERM;
1920
1921 if (!iter_is_iovec(from))
1922 return -EINVAL;
1923
1924 fuse_copy_init(&cs, 0, from);
1925
1926 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1927}
1928
1929static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1930 struct file *out, loff_t *ppos,
1931 size_t len, unsigned int flags)
1932{
1933 unsigned nbuf;
1934 unsigned idx;
1935 struct pipe_buffer *bufs;
1936 struct fuse_copy_state cs;
1937 struct fuse_dev *fud;
1938 size_t rem;
1939 ssize_t ret;
1940
1941 fud = fuse_get_dev(out);
1942 if (!fud)
1943 return -EPERM;
1944
1945 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1946 if (!bufs)
1947 return -ENOMEM;
1948
1949 pipe_lock(pipe);
1950 nbuf = 0;
1951 rem = 0;
1952 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1953 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1954
1955 ret = -EINVAL;
1956 if (rem < len) {
1957 pipe_unlock(pipe);
1958 goto out;
1959 }
1960
1961 rem = len;
1962 while (rem) {
1963 struct pipe_buffer *ibuf;
1964 struct pipe_buffer *obuf;
1965
1966 BUG_ON(nbuf >= pipe->buffers);
1967 BUG_ON(!pipe->nrbufs);
1968 ibuf = &pipe->bufs[pipe->curbuf];
1969 obuf = &bufs[nbuf];
1970
1971 if (rem >= ibuf->len) {
1972 *obuf = *ibuf;
1973 ibuf->ops = NULL;
1974 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1975 pipe->nrbufs--;
1976 } else {
1977 pipe_buf_get(pipe, ibuf);
1978 *obuf = *ibuf;
1979 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1980 obuf->len = rem;
1981 ibuf->offset += obuf->len;
1982 ibuf->len -= obuf->len;
1983 }
1984 nbuf++;
1985 rem -= obuf->len;
1986 }
1987 pipe_unlock(pipe);
1988
1989 fuse_copy_init(&cs, 0, NULL);
1990 cs.pipebufs = bufs;
1991 cs.nr_segs = nbuf;
1992 cs.pipe = pipe;
1993
1994 if (flags & SPLICE_F_MOVE)
1995 cs.move_pages = 1;
1996
1997 ret = fuse_dev_do_write(fud, &cs, len);
1998
1999 for (idx = 0; idx < nbuf; idx++)
2000 pipe_buf_release(pipe, &bufs[idx]);
2001
2002out:
2003 kfree(bufs);
2004 return ret;
2005}
2006
2007static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2008{
2009 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
2010 struct fuse_iqueue *fiq;
2011 struct fuse_dev *fud = fuse_get_dev(file);
2012
2013 if (!fud)
2014 return EPOLLERR;
2015
2016 fiq = &fud->fc->iq;
2017 poll_wait(file, &fiq->waitq, wait);
2018
2019 spin_lock(&fiq->waitq.lock);
2020 if (!fiq->connected)
2021 mask = EPOLLERR;
2022 else if (request_pending(fiq))
2023 mask |= EPOLLIN | EPOLLRDNORM;
2024 spin_unlock(&fiq->waitq.lock);
2025
2026 return mask;
2027}
2028
2029/*
2030 * Abort all requests on the given list (pending or processing)
2031 *
2032 * This function releases and reacquires fc->lock
2033 */
2034static void end_requests(struct fuse_conn *fc, struct list_head *head)
2035{
2036 while (!list_empty(head)) {
2037 struct fuse_req *req;
2038 req = list_entry(head->next, struct fuse_req, list);
2039 req->out.h.error = -ECONNABORTED;
2040 clear_bit(FR_SENT, &req->flags);
2041 list_del_init(&req->list);
2042 request_end(fc, req);
2043 }
2044}
2045
2046static void end_polls(struct fuse_conn *fc)
2047{
2048 struct rb_node *p;
2049
2050 p = rb_first(&fc->polled_files);
2051
2052 while (p) {
2053 struct fuse_file *ff;
2054 ff = rb_entry(p, struct fuse_file, polled_node);
2055 wake_up_interruptible_all(&ff->poll_wait);
2056
2057 p = rb_next(p);
2058 }
2059}
2060
2061/*
2062 * Abort all requests.
2063 *
2064 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2065 * filesystem.
2066 *
2067 * The same effect is usually achievable through killing the filesystem daemon
2068 * and all users of the filesystem. The exception is the combination of an
2069 * asynchronous request and the tricky deadlock (see
2070 * Documentation/filesystems/fuse.txt).
2071 *
2072 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2073 * requests, they should be finished off immediately. Locked requests will be
2074 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2075 * requests. It is possible that some request will finish before we can. This
2076 * is OK, the request will in that case be removed from the list before we touch
2077 * it.
2078 */
2079void fuse_abort_conn(struct fuse_conn *fc)
2080{
2081 struct fuse_iqueue *fiq = &fc->iq;
2082
2083 spin_lock(&fc->lock);
2084 if (fc->connected) {
2085 struct fuse_dev *fud;
2086 struct fuse_req *req, *next;
2087 LIST_HEAD(to_end1);
2088 LIST_HEAD(to_end2);
2089
2090 fc->connected = 0;
2091 fc->blocked = 0;
2092 fuse_set_initialized(fc);
2093 list_for_each_entry(fud, &fc->devices, entry) {
2094 struct fuse_pqueue *fpq = &fud->pq;
2095
2096 spin_lock(&fpq->lock);
2097 fpq->connected = 0;
2098 list_for_each_entry_safe(req, next, &fpq->io, list) {
2099 req->out.h.error = -ECONNABORTED;
2100 spin_lock(&req->waitq.lock);
2101 set_bit(FR_ABORTED, &req->flags);
2102 if (!test_bit(FR_LOCKED, &req->flags)) {
2103 set_bit(FR_PRIVATE, &req->flags);
2104 list_move(&req->list, &to_end1);
2105 }
2106 spin_unlock(&req->waitq.lock);
2107 }
2108 list_splice_init(&fpq->processing, &to_end2);
2109 spin_unlock(&fpq->lock);
2110 }
2111 fc->max_background = UINT_MAX;
2112 flush_bg_queue(fc);
2113
2114 spin_lock(&fiq->waitq.lock);
2115 fiq->connected = 0;
2116 list_splice_init(&fiq->pending, &to_end2);
2117 list_for_each_entry(req, &to_end2, list)
2118 clear_bit(FR_PENDING, &req->flags);
2119 while (forget_pending(fiq))
2120 kfree(dequeue_forget(fiq, 1, NULL));
2121 wake_up_all_locked(&fiq->waitq);
2122 spin_unlock(&fiq->waitq.lock);
2123 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2124 end_polls(fc);
2125 wake_up_all(&fc->blocked_waitq);
2126 spin_unlock(&fc->lock);
2127
2128 while (!list_empty(&to_end1)) {
2129 req = list_first_entry(&to_end1, struct fuse_req, list);
2130 __fuse_get_request(req);
2131 list_del_init(&req->list);
2132 request_end(fc, req);
2133 }
2134 end_requests(fc, &to_end2);
2135 } else {
2136 spin_unlock(&fc->lock);
2137 }
2138}
2139EXPORT_SYMBOL_GPL(fuse_abort_conn);
2140
2141int fuse_dev_release(struct inode *inode, struct file *file)
2142{
2143 struct fuse_dev *fud = fuse_get_dev(file);
2144
2145 if (fud) {
2146 struct fuse_conn *fc = fud->fc;
2147 struct fuse_pqueue *fpq = &fud->pq;
2148
2149 WARN_ON(!list_empty(&fpq->io));
2150 end_requests(fc, &fpq->processing);
2151 /* Are we the last open device? */
2152 if (atomic_dec_and_test(&fc->dev_count)) {
2153 WARN_ON(fc->iq.fasync != NULL);
2154 fuse_abort_conn(fc);
2155 }
2156 fuse_dev_free(fud);
2157 }
2158 return 0;
2159}
2160EXPORT_SYMBOL_GPL(fuse_dev_release);
2161
2162static int fuse_dev_fasync(int fd, struct file *file, int on)
2163{
2164 struct fuse_dev *fud = fuse_get_dev(file);
2165
2166 if (!fud)
2167 return -EPERM;
2168
2169 /* No locking - fasync_helper does its own locking */
2170 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2171}
2172
2173static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2174{
2175 struct fuse_dev *fud;
2176
2177 if (new->private_data)
2178 return -EINVAL;
2179
2180 fud = fuse_dev_alloc(fc);
2181 if (!fud)
2182 return -ENOMEM;
2183
2184 new->private_data = fud;
2185 atomic_inc(&fc->dev_count);
2186
2187 return 0;
2188}
2189
2190static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2191 unsigned long arg)
2192{
2193 int err = -ENOTTY;
2194
2195 if (cmd == FUSE_DEV_IOC_CLONE) {
2196 int oldfd;
2197
2198 err = -EFAULT;
2199 if (!get_user(oldfd, (__u32 __user *) arg)) {
2200 struct file *old = fget(oldfd);
2201
2202 err = -EINVAL;
2203 if (old) {
2204 struct fuse_dev *fud = NULL;
2205
2206 /*
2207 * Check against file->f_op because CUSE
2208 * uses the same ioctl handler.
2209 */
2210 if (old->f_op == file->f_op &&
2211 old->f_cred->user_ns == file->f_cred->user_ns)
2212 fud = fuse_get_dev(old);
2213
2214 if (fud) {
2215 mutex_lock(&fuse_mutex);
2216 err = fuse_device_clone(fud->fc, file);
2217 mutex_unlock(&fuse_mutex);
2218 }
2219 fput(old);
2220 }
2221 }
2222 }
2223 return err;
2224}
2225
2226const struct file_operations fuse_dev_operations = {
2227 .owner = THIS_MODULE,
2228 .open = fuse_dev_open,
2229 .llseek = no_llseek,
2230 .read_iter = fuse_dev_read,
2231 .splice_read = fuse_dev_splice_read,
2232 .write_iter = fuse_dev_write,
2233 .splice_write = fuse_dev_splice_write,
2234 .poll = fuse_dev_poll,
2235 .release = fuse_dev_release,
2236 .fasync = fuse_dev_fasync,
2237 .unlocked_ioctl = fuse_dev_ioctl,
2238 .compat_ioctl = fuse_dev_ioctl,
2239};
2240EXPORT_SYMBOL_GPL(fuse_dev_operations);
2241
2242static struct miscdevice fuse_miscdevice = {
2243 .minor = FUSE_MINOR,
2244 .name = "fuse",
2245 .fops = &fuse_dev_operations,
2246};
2247
2248int __init fuse_dev_init(void)
2249{
2250 int err = -ENOMEM;
2251 fuse_req_cachep = kmem_cache_create("fuse_request",
2252 sizeof(struct fuse_req),
2253 0, 0, NULL);
2254 if (!fuse_req_cachep)
2255 goto out;
2256
2257 err = misc_register(&fuse_miscdevice);
2258 if (err)
2259 goto out_cache_clean;
2260
2261 return 0;
2262
2263 out_cache_clean:
2264 kmem_cache_destroy(fuse_req_cachep);
2265 out:
2266 return err;
2267}
2268
2269void fuse_dev_cleanup(void)
2270{
2271 misc_deregister(&fuse_miscdevice);
2272 kmem_cache_destroy(fuse_req_cachep);
2273}