?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ring-buffer: Do not poll non allocated cpu buffers
[linux-2.6.git] / fs / fuse / dev.c
blob1d55f94654000dbc8e8c0de37e0cb32471e3791a
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 #include <linux/aio.h>
23
24 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
25 MODULE_ALIAS("devname:fuse");
26
27 static struct kmem_cache *fuse_req_cachep;
28
29 static struct fuse_conn *fuse_get_conn(struct file *file)
30 {
31 /*
32 * Lockless access is OK, because file->private data is set
33 * once during mount and is valid until the file is released.
34 */
35 return file->private_data;
36 }
37
38 static void fuse_request_init(struct fuse_req *req, struct page **pages,
39 struct fuse_page_desc *page_descs,
40 unsigned npages)
41 {
42 memset(req, 0, sizeof(*req));
43 memset(pages, 0, sizeof(*pages) * npages);
44 memset(page_descs, 0, sizeof(*page_descs) * npages);
45 INIT_LIST_HEAD(&req->list);
46 INIT_LIST_HEAD(&req->intr_entry);
47 init_waitqueue_head(&req->waitq);
48 atomic_set(&req->count, 1);
49 req->pages = pages;
50 req->page_descs = page_descs;
51 req->max_pages = npages;
52 }
53
54 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
55 {
56 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
57 if (req) {
58 struct page **pages;
59 struct fuse_page_desc *page_descs;
60
61 if (npages <= FUSE_REQ_INLINE_PAGES) {
62 pages = req->inline_pages;
63 page_descs = req->inline_page_descs;
64 } else {
65 pages = kmalloc(sizeof(struct page *) * npages, flags);
66 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
67 npages, flags);
68 }
69
70 if (!pages || !page_descs) {
71 kfree(pages);
72 kfree(page_descs);
73 kmem_cache_free(fuse_req_cachep, req);
74 return NULL;
75 }
76
77 fuse_request_init(req, pages, page_descs, npages);
78 }
79 return req;
80 }
81
82 struct fuse_req *fuse_request_alloc(unsigned npages)
83 {
84 return __fuse_request_alloc(npages, GFP_KERNEL);
85 }
86 EXPORT_SYMBOL_GPL(fuse_request_alloc);
87
88 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
89 {
90 return __fuse_request_alloc(npages, GFP_NOFS);
91 }
92
93 void fuse_request_free(struct fuse_req *req)
94 {
95 if (req->pages != req->inline_pages) {
96 kfree(req->pages);
97 kfree(req->page_descs);
98 }
99 kmem_cache_free(fuse_req_cachep, req);
100 }
101
102 static void block_sigs(sigset_t *oldset)
103 {
104 sigset_t mask;
105
106 siginitsetinv(&mask, sigmask(SIGKILL));
107 sigprocmask(SIG_BLOCK, &mask, oldset);
108 }
109
110 static void restore_sigs(sigset_t *oldset)
111 {
112 sigprocmask(SIG_SETMASK, oldset, NULL);
113 }
114
115 void __fuse_get_request(struct fuse_req *req)
116 {
117 atomic_inc(&req->count);
118 }
119
120 /* Must be called with > 1 refcount */
121 static void __fuse_put_request(struct fuse_req *req)
122 {
123 BUG_ON(atomic_read(&req->count) < 2);
124 atomic_dec(&req->count);
125 }
126
127 static void fuse_req_init_context(struct fuse_req *req)
128 {
129 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
130 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
131 req->in.h.pid = current->pid;
132 }
133
134 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
135 {
136 return !fc->initialized || (for_background && fc->blocked);
137 }
138
139 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
140 bool for_background)
141 {
142 struct fuse_req *req;
143 int err;
144 atomic_inc(&fc->num_waiting);
145
146 if (fuse_block_alloc(fc, for_background)) {
147 sigset_t oldset;
148 int intr;
149
150 block_sigs(&oldset);
151 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
152 !fuse_block_alloc(fc, for_background));
153 restore_sigs(&oldset);
154 err = -EINTR;
155 if (intr)
156 goto out;
157 }
158
159 err = -ENOTCONN;
160 if (!fc->connected)
161 goto out;
162
163 req = fuse_request_alloc(npages);
164 err = -ENOMEM;
165 if (!req) {
166 if (for_background)
167 wake_up(&fc->blocked_waitq);
168 goto out;
169 }
170
171 fuse_req_init_context(req);
172 req->waiting = 1;
173 req->background = for_background;
174 return req;
175
176 out:
177 atomic_dec(&fc->num_waiting);
178 return ERR_PTR(err);
179 }
180
181 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
182 {
183 return __fuse_get_req(fc, npages, false);
184 }
185 EXPORT_SYMBOL_GPL(fuse_get_req);
186
187 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
188 unsigned npages)
189 {
190 return __fuse_get_req(fc, npages, true);
191 }
192 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
193
194 /*
195 * Return request in fuse_file->reserved_req. However that may
196 * currently be in use. If that is the case, wait for it to become
197 * available.
198 */
199 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
200 struct file *file)
201 {
202 struct fuse_req *req = NULL;
203 struct fuse_file *ff = file->private_data;
204
205 do {
206 wait_event(fc->reserved_req_waitq, ff->reserved_req);
207 spin_lock(&fc->lock);
208 if (ff->reserved_req) {
209 req = ff->reserved_req;
210 ff->reserved_req = NULL;
211 req->stolen_file = get_file(file);
212 }
213 spin_unlock(&fc->lock);
214 } while (!req);
215
216 return req;
217 }
218
219 /*
220 * Put stolen request back into fuse_file->reserved_req
221 */
222 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
223 {
224 struct file *file = req->stolen_file;
225 struct fuse_file *ff = file->private_data;
226
227 spin_lock(&fc->lock);
228 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
229 BUG_ON(ff->reserved_req);
230 ff->reserved_req = req;
231 wake_up_all(&fc->reserved_req_waitq);
232 spin_unlock(&fc->lock);
233 fput(file);
234 }
235
236 /*
237 * Gets a requests for a file operation, always succeeds
238 *
239 * This is used for sending the FLUSH request, which must get to
240 * userspace, due to POSIX locks which may need to be unlocked.
241 *
242 * If allocation fails due to OOM, use the reserved request in
243 * fuse_file.
244 *
245 * This is very unlikely to deadlock accidentally, since the
246 * filesystem should not have it's own file open. If deadlock is
247 * intentional, it can still be broken by "aborting" the filesystem.
248 */
249 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
250 struct file *file)
251 {
252 struct fuse_req *req;
253
254 atomic_inc(&fc->num_waiting);
255 wait_event(fc->blocked_waitq, fc->initialized);
256 req = fuse_request_alloc(0);
257 if (!req)
258 req = get_reserved_req(fc, file);
259
260 fuse_req_init_context(req);
261 req->waiting = 1;
262 req->background = 0;
263 return req;
264 }
265
266 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
267 {
268 if (atomic_dec_and_test(&req->count)) {
269 if (unlikely(req->background)) {
270 /*
271 * We get here in the unlikely case that a background
272 * request was allocated but not sent
273 */
274 spin_lock(&fc->lock);
275 if (!fc->blocked)
276 wake_up(&fc->blocked_waitq);
277 spin_unlock(&fc->lock);
278 }
279
280 if (req->waiting)
281 atomic_dec(&fc->num_waiting);
282
283 if (req->stolen_file)
284 put_reserved_req(fc, req);
285 else
286 fuse_request_free(req);
287 }
288 }
289 EXPORT_SYMBOL_GPL(fuse_put_request);
290
291 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
292 {
293 unsigned nbytes = 0;
294 unsigned i;
295
296 for (i = 0; i < numargs; i++)
297 nbytes += args[i].size;
298
299 return nbytes;
300 }
301
302 static u64 fuse_get_unique(struct fuse_conn *fc)
303 {
304 fc->reqctr++;
305 /* zero is special */
306 if (fc->reqctr == 0)
307 fc->reqctr = 1;
308
309 return fc->reqctr;
310 }
311
312 static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
313 {
314 req->in.h.len = sizeof(struct fuse_in_header) +
315 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
316 list_add_tail(&req->list, &fc->pending);
317 req->state = FUSE_REQ_PENDING;
318 if (!req->waiting) {
319 req->waiting = 1;
320 atomic_inc(&fc->num_waiting);
321 }
322 wake_up(&fc->waitq);
323 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
324 }
325
326 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
327 u64 nodeid, u64 nlookup)
328 {
329 forget->forget_one.nodeid = nodeid;
330 forget->forget_one.nlookup = nlookup;
331
332 spin_lock(&fc->lock);
333 if (fc->connected) {
334 fc->forget_list_tail->next = forget;
335 fc->forget_list_tail = forget;
336 wake_up(&fc->waitq);
337 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
338 } else {
339 kfree(forget);
340 }
341 spin_unlock(&fc->lock);
342 }
343
344 static void flush_bg_queue(struct fuse_conn *fc)
345 {
346 while (fc->active_background < fc->max_background &&
347 !list_empty(&fc->bg_queue)) {
348 struct fuse_req *req;
349
350 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
351 list_del(&req->list);
352 fc->active_background++;
353 req->in.h.unique = fuse_get_unique(fc);
354 queue_request(fc, req);
355 }
356 }
357
358 /*
359 * This function is called when a request is finished. Either a reply
360 * has arrived or it was aborted (and not yet sent) or some error
361 * occurred during communication with userspace, or the device file
362 * was closed. The requester thread is woken up (if still waiting),
363 * the 'end' callback is called if given, else the reference to the
364 * request is released
365 *
366 * Called with fc->lock, unlocks it
367 */
368 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
369 __releases(fc->lock)
370 {
371 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
372 req->end = NULL;
373 list_del(&req->list);
374 list_del(&req->intr_entry);
375 req->state = FUSE_REQ_FINISHED;
376 if (req->background) {
377 req->background = 0;
378
379 if (fc->num_background == fc->max_background)
380 fc->blocked = 0;
381
382 /* Wake up next waiter, if any */
383 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
384 wake_up(&fc->blocked_waitq);
385
386 if (fc->num_background == fc->congestion_threshold &&
387 fc->connected && fc->bdi_initialized) {
388 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
389 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
390 }
391 fc->num_background--;
392 fc->active_background--;
393 flush_bg_queue(fc);
394 }
395 spin_unlock(&fc->lock);
396 wake_up(&req->waitq);
397 if (end)
398 end(fc, req);
399 fuse_put_request(fc, req);
400 }
401
402 static void wait_answer_interruptible(struct fuse_conn *fc,
403 struct fuse_req *req)
404 __releases(fc->lock)
405 __acquires(fc->lock)
406 {
407 if (signal_pending(current))
408 return;
409
410 spin_unlock(&fc->lock);
411 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
412 spin_lock(&fc->lock);
413 }
414
415 static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
416 {
417 list_add_tail(&req->intr_entry, &fc->interrupts);
418 wake_up(&fc->waitq);
419 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
420 }
421
422 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
423 __releases(fc->lock)
424 __acquires(fc->lock)
425 {
426 if (!fc->no_interrupt) {
427 /* Any signal may interrupt this */
428 wait_answer_interruptible(fc, req);
429
430 if (req->aborted)
431 goto aborted;
432 if (req->state == FUSE_REQ_FINISHED)
433 return;
434
435 req->interrupted = 1;
436 if (req->state == FUSE_REQ_SENT)
437 queue_interrupt(fc, req);
438 }
439
440 if (!req->force) {
441 sigset_t oldset;
442
443 /* Only fatal signals may interrupt this */
444 block_sigs(&oldset);
445 wait_answer_interruptible(fc, req);
446 restore_sigs(&oldset);
447
448 if (req->aborted)
449 goto aborted;
450 if (req->state == FUSE_REQ_FINISHED)
451 return;
452
453 /* Request is not yet in userspace, bail out */
454 if (req->state == FUSE_REQ_PENDING) {
455 list_del(&req->list);
456 __fuse_put_request(req);
457 req->out.h.error = -EINTR;
458 return;
459 }
460 }
461
462 /*
463 * Either request is already in userspace, or it was forced.
464 * Wait it out.
465 */
466 spin_unlock(&fc->lock);
467 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
468 spin_lock(&fc->lock);
469
470 if (!req->aborted)
471 return;
472
473 aborted:
474 BUG_ON(req->state != FUSE_REQ_FINISHED);
475 if (req->locked) {
476 /* This is uninterruptible sleep, because data is
477 being copied to/from the buffers of req. During
478 locked state, there mustn't be any filesystem
479 operation (e.g. page fault), since that could lead
480 to deadlock */
481 spin_unlock(&fc->lock);
482 wait_event(req->waitq, !req->locked);
483 spin_lock(&fc->lock);
484 }
485 }
486
487 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
488 {
489 BUG_ON(req->background);
490 spin_lock(&fc->lock);
491 if (!fc->connected)
492 req->out.h.error = -ENOTCONN;
493 else if (fc->conn_error)
494 req->out.h.error = -ECONNREFUSED;
495 else {
496 req->in.h.unique = fuse_get_unique(fc);
497 queue_request(fc, req);
498 /* acquire extra reference, since request is still needed
499 after request_end() */
500 __fuse_get_request(req);
501
502 request_wait_answer(fc, req);
503 }
504 spin_unlock(&fc->lock);
505 }
506
507 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
508 {
509 req->isreply = 1;
510 __fuse_request_send(fc, req);
511 }
512 EXPORT_SYMBOL_GPL(fuse_request_send);
513
514 static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
515 struct fuse_req *req)
516 {
517 BUG_ON(!req->background);
518 fc->num_background++;
519 if (fc->num_background == fc->max_background)
520 fc->blocked = 1;
521 if (fc->num_background == fc->congestion_threshold &&
522 fc->bdi_initialized) {
523 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
524 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
525 }
526 list_add_tail(&req->list, &fc->bg_queue);
527 flush_bg_queue(fc);
528 }
529
530 static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
531 {
532 spin_lock(&fc->lock);
533 if (fc->connected) {
534 fuse_request_send_nowait_locked(fc, req);
535 spin_unlock(&fc->lock);
536 } else {
537 req->out.h.error = -ENOTCONN;
538 request_end(fc, req);
539 }
540 }
541
542 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
543 {
544 req->isreply = 1;
545 fuse_request_send_nowait(fc, req);
546 }
547 EXPORT_SYMBOL_GPL(fuse_request_send_background);
548
549 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
550 struct fuse_req *req, u64 unique)
551 {
552 int err = -ENODEV;
553
554 req->isreply = 0;
555 req->in.h.unique = unique;
556 spin_lock(&fc->lock);
557 if (fc->connected) {
558 queue_request(fc, req);
559 err = 0;
560 }
561 spin_unlock(&fc->lock);
562
563 return err;
564 }
565
566 /*
567 * Called under fc->lock
568 *
569 * fc->connected must have been checked previously
570 */
571 void fuse_request_send_background_locked(struct fuse_conn *fc,
572 struct fuse_req *req)
573 {
574 req->isreply = 1;
575 fuse_request_send_nowait_locked(fc, req);
576 }
577
578 void fuse_force_forget(struct file *file, u64 nodeid)
579 {
580 struct inode *inode = file_inode(file);
581 struct fuse_conn *fc = get_fuse_conn(inode);
582 struct fuse_req *req;
583 struct fuse_forget_in inarg;
584
585 memset(&inarg, 0, sizeof(inarg));
586 inarg.nlookup = 1;
587 req = fuse_get_req_nofail_nopages(fc, file);
588 req->in.h.opcode = FUSE_FORGET;
589 req->in.h.nodeid = nodeid;
590 req->in.numargs = 1;
591 req->in.args[0].size = sizeof(inarg);
592 req->in.args[0].value = &inarg;
593 req->isreply = 0;
594 __fuse_request_send(fc, req);
595 /* ignore errors */
596 fuse_put_request(fc, req);
597 }
598
599 /*
600 * Lock the request. Up to the next unlock_request() there mustn't be
601 * anything that could cause a page-fault. If the request was already
602 * aborted bail out.
603 */
604 static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
605 {
606 int err = 0;
607 if (req) {
608 spin_lock(&fc->lock);
609 if (req->aborted)
610 err = -ENOENT;
611 else
612 req->locked = 1;
613 spin_unlock(&fc->lock);
614 }
615 return err;
616 }
617
618 /*
619 * Unlock request. If it was aborted during being locked, the
620 * requester thread is currently waiting for it to be unlocked, so
621 * wake it up.
622 */
623 static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
624 {
625 if (req) {
626 spin_lock(&fc->lock);
627 req->locked = 0;
628 if (req->aborted)
629 wake_up(&req->waitq);
630 spin_unlock(&fc->lock);
631 }
632 }
633
634 struct fuse_copy_state {
635 struct fuse_conn *fc;
636 int write;
637 struct fuse_req *req;
638 const struct iovec *iov;
639 struct pipe_buffer *pipebufs;
640 struct pipe_buffer *currbuf;
641 struct pipe_inode_info *pipe;
642 unsigned long nr_segs;
643 unsigned long seglen;
644 unsigned long addr;
645 struct page *pg;
646 void *mapaddr;
647 void *buf;
648 unsigned len;
649 unsigned move_pages:1;
650 };
651
652 static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
653 int write,
654 const struct iovec *iov, unsigned long nr_segs)
655 {
656 memset(cs, 0, sizeof(*cs));
657 cs->fc = fc;
658 cs->write = write;
659 cs->iov = iov;
660 cs->nr_segs = nr_segs;
661 }
662
663 /* Unmap and put previous page of userspace buffer */
664 static void fuse_copy_finish(struct fuse_copy_state *cs)
665 {
666 if (cs->currbuf) {
667 struct pipe_buffer *buf = cs->currbuf;
668
669 if (!cs->write) {
670 buf->ops->unmap(cs->pipe, buf, cs->mapaddr);
671 } else {
672 kunmap(buf->page);
673 buf->len = PAGE_SIZE - cs->len;
674 }
675 cs->currbuf = NULL;
676 cs->mapaddr = NULL;
677 } else if (cs->mapaddr) {
678 kunmap(cs->pg);
679 if (cs->write) {
680 flush_dcache_page(cs->pg);
681 set_page_dirty_lock(cs->pg);
682 }
683 put_page(cs->pg);
684 cs->mapaddr = NULL;
685 }
686 }
687
688 /*
689 * Get another pagefull of userspace buffer, and map it to kernel
690 * address space, and lock request
691 */
692 static int fuse_copy_fill(struct fuse_copy_state *cs)
693 {
694 unsigned long offset;
695 int err;
696
697 unlock_request(cs->fc, cs->req);
698 fuse_copy_finish(cs);
699 if (cs->pipebufs) {
700 struct pipe_buffer *buf = cs->pipebufs;
701
702 if (!cs->write) {
703 err = buf->ops->confirm(cs->pipe, buf);
704 if (err)
705 return err;
706
707 BUG_ON(!cs->nr_segs);
708 cs->currbuf = buf;
709 cs->mapaddr = buf->ops->map(cs->pipe, buf, 0);
710 cs->len = buf->len;
711 cs->buf = cs->mapaddr + buf->offset;
712 cs->pipebufs++;
713 cs->nr_segs--;
714 } else {
715 struct page *page;
716
717 if (cs->nr_segs == cs->pipe->buffers)
718 return -EIO;
719
720 page = alloc_page(GFP_HIGHUSER);
721 if (!page)
722 return -ENOMEM;
723
724 buf->page = page;
725 buf->offset = 0;
726 buf->len = 0;
727
728 cs->currbuf = buf;
729 cs->mapaddr = kmap(page);
730 cs->buf = cs->mapaddr;
731 cs->len = PAGE_SIZE;
732 cs->pipebufs++;
733 cs->nr_segs++;
734 }
735 } else {
736 if (!cs->seglen) {
737 BUG_ON(!cs->nr_segs);
738 cs->seglen = cs->iov[0].iov_len;
739 cs->addr = (unsigned long) cs->iov[0].iov_base;
740 cs->iov++;
741 cs->nr_segs--;
742 }
743 err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg);
744 if (err < 0)
745 return err;
746 BUG_ON(err != 1);
747 offset = cs->addr % PAGE_SIZE;
748 cs->mapaddr = kmap(cs->pg);
749 cs->buf = cs->mapaddr + offset;
750 cs->len = min(PAGE_SIZE - offset, cs->seglen);
751 cs->seglen -= cs->len;
752 cs->addr += cs->len;
753 }
754
755 return lock_request(cs->fc, cs->req);
756 }
757
758 /* Do as much copy to/from userspace buffer as we can */
759 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
760 {
761 unsigned ncpy = min(*size, cs->len);
762 if (val) {
763 if (cs->write)
764 memcpy(cs->buf, *val, ncpy);
765 else
766 memcpy(*val, cs->buf, ncpy);
767 *val += ncpy;
768 }
769 *size -= ncpy;
770 cs->len -= ncpy;
771 cs->buf += ncpy;
772 return ncpy;
773 }
774
775 static int fuse_check_page(struct page *page)
776 {
777 if (page_mapcount(page) ||
778 page->mapping != NULL ||
779 page_count(page) != 1 ||
780 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
781 ~(1 << PG_locked |
782 1 << PG_referenced |
783 1 << PG_uptodate |
784 1 << PG_lru |
785 1 << PG_active |
786 1 << PG_reclaim))) {
787 printk(KERN_WARNING "fuse: trying to steal weird page\n");
788 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);
789 return 1;
790 }
791 return 0;
792 }
793
794 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
795 {
796 int err;
797 struct page *oldpage = *pagep;
798 struct page *newpage;
799 struct pipe_buffer *buf = cs->pipebufs;
800
801 unlock_request(cs->fc, cs->req);
802 fuse_copy_finish(cs);
803
804 err = buf->ops->confirm(cs->pipe, buf);
805 if (err)
806 return err;
807
808 BUG_ON(!cs->nr_segs);
809 cs->currbuf = buf;
810 cs->len = buf->len;
811 cs->pipebufs++;
812 cs->nr_segs--;
813
814 if (cs->len != PAGE_SIZE)
815 goto out_fallback;
816
817 if (buf->ops->steal(cs->pipe, buf) != 0)
818 goto out_fallback;
819
820 newpage = buf->page;
821
822 if (WARN_ON(!PageUptodate(newpage)))
823 return -EIO;
824
825 ClearPageMappedToDisk(newpage);
826
827 if (fuse_check_page(newpage) != 0)
828 goto out_fallback_unlock;
829
830 /*
831 * This is a new and locked page, it shouldn't be mapped or
832 * have any special flags on it
833 */
834 if (WARN_ON(page_mapped(oldpage)))
835 goto out_fallback_unlock;
836 if (WARN_ON(page_has_private(oldpage)))
837 goto out_fallback_unlock;
838 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
839 goto out_fallback_unlock;
840 if (WARN_ON(PageMlocked(oldpage)))
841 goto out_fallback_unlock;
842
843 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
844 if (err) {
845 unlock_page(newpage);
846 return err;
847 }
848
849 page_cache_get(newpage);
850
851 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
852 lru_cache_add_file(newpage);
853
854 err = 0;
855 spin_lock(&cs->fc->lock);
856 if (cs->req->aborted)
857 err = -ENOENT;
858 else
859 *pagep = newpage;
860 spin_unlock(&cs->fc->lock);
861
862 if (err) {
863 unlock_page(newpage);
864 page_cache_release(newpage);
865 return err;
866 }
867
868 unlock_page(oldpage);
869 page_cache_release(oldpage);
870 cs->len = 0;
871
872 return 0;
873
874 out_fallback_unlock:
875 unlock_page(newpage);
876 out_fallback:
877 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
878 cs->buf = cs->mapaddr + buf->offset;
879
880 err = lock_request(cs->fc, cs->req);
881 if (err)
882 return err;
883
884 return 1;
885 }
886
887 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
888 unsigned offset, unsigned count)
889 {
890 struct pipe_buffer *buf;
891
892 if (cs->nr_segs == cs->pipe->buffers)
893 return -EIO;
894
895 unlock_request(cs->fc, cs->req);
896 fuse_copy_finish(cs);
897
898 buf = cs->pipebufs;
899 page_cache_get(page);
900 buf->page = page;
901 buf->offset = offset;
902 buf->len = count;
903
904 cs->pipebufs++;
905 cs->nr_segs++;
906 cs->len = 0;
907
908 return 0;
909 }
910
911 /*
912 * Copy a page in the request to/from the userspace buffer. Must be
913 * done atomically
914 */
915 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
916 unsigned offset, unsigned count, int zeroing)
917 {
918 int err;
919 struct page *page = *pagep;
920
921 if (page && zeroing && count < PAGE_SIZE)
922 clear_highpage(page);
923
924 while (count) {
925 if (cs->write && cs->pipebufs && page) {
926 return fuse_ref_page(cs, page, offset, count);
927 } else if (!cs->len) {
928 if (cs->move_pages && page &&
929 offset == 0 && count == PAGE_SIZE) {
930 err = fuse_try_move_page(cs, pagep);
931 if (err <= 0)
932 return err;
933 } else {
934 err = fuse_copy_fill(cs);
935 if (err)
936 return err;
937 }
938 }
939 if (page) {
940 void *mapaddr = kmap_atomic(page);
941 void *buf = mapaddr + offset;
942 offset += fuse_copy_do(cs, &buf, &count);
943 kunmap_atomic(mapaddr);
944 } else
945 offset += fuse_copy_do(cs, NULL, &count);
946 }
947 if (page && !cs->write)
948 flush_dcache_page(page);
949 return 0;
950 }
951
952 /* Copy pages in the request to/from userspace buffer */
953 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
954 int zeroing)
955 {
956 unsigned i;
957 struct fuse_req *req = cs->req;
958
959 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
960 int err;
961 unsigned offset = req->page_descs[i].offset;
962 unsigned count = min(nbytes, req->page_descs[i].length);
963
964 err = fuse_copy_page(cs, &req->pages[i], offset, count,
965 zeroing);
966 if (err)
967 return err;
968
969 nbytes -= count;
970 }
971 return 0;
972 }
973
974 /* Copy a single argument in the request to/from userspace buffer */
975 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
976 {
977 while (size) {
978 if (!cs->len) {
979 int err = fuse_copy_fill(cs);
980 if (err)
981 return err;
982 }
983 fuse_copy_do(cs, &val, &size);
984 }
985 return 0;
986 }
987
988 /* Copy request arguments to/from userspace buffer */
989 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
990 unsigned argpages, struct fuse_arg *args,
991 int zeroing)
992 {
993 int err = 0;
994 unsigned i;
995
996 for (i = 0; !err && i < numargs; i++) {
997 struct fuse_arg *arg = &args[i];
998 if (i == numargs - 1 && argpages)
999 err = fuse_copy_pages(cs, arg->size, zeroing);
1000 else
1001 err = fuse_copy_one(cs, arg->value, arg->size);
1002 }
1003 return err;
1004 }
1005
1006 static int forget_pending(struct fuse_conn *fc)
1007 {
1008 return fc->forget_list_head.next != NULL;
1009 }
1010
1011 static int request_pending(struct fuse_conn *fc)
1012 {
1013 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
1014 forget_pending(fc);
1015 }
1016
1017 /* Wait until a request is available on the pending list */
1018 static void request_wait(struct fuse_conn *fc)
1019 __releases(fc->lock)
1020 __acquires(fc->lock)
1021 {
1022 DECLARE_WAITQUEUE(wait, current);
1023
1024 add_wait_queue_exclusive(&fc->waitq, &wait);
1025 while (fc->connected && !request_pending(fc)) {
1026 set_current_state(TASK_INTERRUPTIBLE);
1027 if (signal_pending(current))
1028 break;
1029
1030 spin_unlock(&fc->lock);
1031 schedule();
1032 spin_lock(&fc->lock);
1033 }
1034 set_current_state(TASK_RUNNING);
1035 remove_wait_queue(&fc->waitq, &wait);
1036 }
1037
1038 /*
1039 * Transfer an interrupt request to userspace
1040 *
1041 * Unlike other requests this is assembled on demand, without a need
1042 * to allocate a separate fuse_req structure.
1043 *
1044 * Called with fc->lock held, releases it
1045 */
1046 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1047 size_t nbytes, struct fuse_req *req)
1048 __releases(fc->lock)
1049 {
1050 struct fuse_in_header ih;
1051 struct fuse_interrupt_in arg;
1052 unsigned reqsize = sizeof(ih) + sizeof(arg);
1053 int err;
1054
1055 list_del_init(&req->intr_entry);
1056 req->intr_unique = fuse_get_unique(fc);
1057 memset(&ih, 0, sizeof(ih));
1058 memset(&arg, 0, sizeof(arg));
1059 ih.len = reqsize;
1060 ih.opcode = FUSE_INTERRUPT;
1061 ih.unique = req->intr_unique;
1062 arg.unique = req->in.h.unique;
1063
1064 spin_unlock(&fc->lock);
1065 if (nbytes < reqsize)
1066 return -EINVAL;
1067
1068 err = fuse_copy_one(cs, &ih, sizeof(ih));
1069 if (!err)
1070 err = fuse_copy_one(cs, &arg, sizeof(arg));
1071 fuse_copy_finish(cs);
1072
1073 return err ? err : reqsize;
1074 }
1075
1076 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
1077 unsigned max,
1078 unsigned *countp)
1079 {
1080 struct fuse_forget_link *head = fc->forget_list_head.next;
1081 struct fuse_forget_link **newhead = &head;
1082 unsigned count;
1083
1084 for (count = 0; *newhead != NULL && count < max; count++)
1085 newhead = &(*newhead)->next;
1086
1087 fc->forget_list_head.next = *newhead;
1088 *newhead = NULL;
1089 if (fc->forget_list_head.next == NULL)
1090 fc->forget_list_tail = &fc->forget_list_head;
1091
1092 if (countp != NULL)
1093 *countp = count;
1094
1095 return head;
1096 }
1097
1098 static int fuse_read_single_forget(struct fuse_conn *fc,
1099 struct fuse_copy_state *cs,
1100 size_t nbytes)
1101 __releases(fc->lock)
1102 {
1103 int err;
1104 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1105 struct fuse_forget_in arg = {
1106 .nlookup = forget->forget_one.nlookup,
1107 };
1108 struct fuse_in_header ih = {
1109 .opcode = FUSE_FORGET,
1110 .nodeid = forget->forget_one.nodeid,
1111 .unique = fuse_get_unique(fc),
1112 .len = sizeof(ih) + sizeof(arg),
1113 };
1114
1115 spin_unlock(&fc->lock);
1116 kfree(forget);
1117 if (nbytes < ih.len)
1118 return -EINVAL;
1119
1120 err = fuse_copy_one(cs, &ih, sizeof(ih));
1121 if (!err)
1122 err = fuse_copy_one(cs, &arg, sizeof(arg));
1123 fuse_copy_finish(cs);
1124
1125 if (err)
1126 return err;
1127
1128 return ih.len;
1129 }
1130
1131 static int fuse_read_batch_forget(struct fuse_conn *fc,
1132 struct fuse_copy_state *cs, size_t nbytes)
1133 __releases(fc->lock)
1134 {
1135 int err;
1136 unsigned max_forgets;
1137 unsigned count;
1138 struct fuse_forget_link *head;
1139 struct fuse_batch_forget_in arg = { .count = 0 };
1140 struct fuse_in_header ih = {
1141 .opcode = FUSE_BATCH_FORGET,
1142 .unique = fuse_get_unique(fc),
1143 .len = sizeof(ih) + sizeof(arg),
1144 };
1145
1146 if (nbytes < ih.len) {
1147 spin_unlock(&fc->lock);
1148 return -EINVAL;
1149 }
1150
1151 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1152 head = dequeue_forget(fc, max_forgets, &count);
1153 spin_unlock(&fc->lock);
1154
1155 arg.count = count;
1156 ih.len += count * sizeof(struct fuse_forget_one);
1157 err = fuse_copy_one(cs, &ih, sizeof(ih));
1158 if (!err)
1159 err = fuse_copy_one(cs, &arg, sizeof(arg));
1160
1161 while (head) {
1162 struct fuse_forget_link *forget = head;
1163
1164 if (!err) {
1165 err = fuse_copy_one(cs, &forget->forget_one,
1166 sizeof(forget->forget_one));
1167 }
1168 head = forget->next;
1169 kfree(forget);
1170 }
1171
1172 fuse_copy_finish(cs);
1173
1174 if (err)
1175 return err;
1176
1177 return ih.len;
1178 }
1179
1180 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1181 size_t nbytes)
1182 __releases(fc->lock)
1183 {
1184 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1185 return fuse_read_single_forget(fc, cs, nbytes);
1186 else
1187 return fuse_read_batch_forget(fc, cs, nbytes);
1188 }
1189
1190 /*
1191 * Read a single request into the userspace filesystem's buffer. This
1192 * function waits until a request is available, then removes it from
1193 * the pending list and copies request data to userspace buffer. If
1194 * no reply is needed (FORGET) or request has been aborted or there
1195 * was an error during the copying then it's finished by calling
1196 * request_end(). Otherwise add it to the processing list, and set
1197 * the 'sent' flag.
1198 */
1199 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1200 struct fuse_copy_state *cs, size_t nbytes)
1201 {
1202 int err;
1203 struct fuse_req *req;
1204 struct fuse_in *in;
1205 unsigned reqsize;
1206
1207 restart:
1208 spin_lock(&fc->lock);
1209 err = -EAGAIN;
1210 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1211 !request_pending(fc))
1212 goto err_unlock;
1213
1214 request_wait(fc);
1215 err = -ENODEV;
1216 if (!fc->connected)
1217 goto err_unlock;
1218 err = -ERESTARTSYS;
1219 if (!request_pending(fc))
1220 goto err_unlock;
1221
1222 if (!list_empty(&fc->interrupts)) {
1223 req = list_entry(fc->interrupts.next, struct fuse_req,
1224 intr_entry);
1225 return fuse_read_interrupt(fc, cs, nbytes, req);
1226 }
1227
1228 if (forget_pending(fc)) {
1229 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1230 return fuse_read_forget(fc, cs, nbytes);
1231
1232 if (fc->forget_batch <= -8)
1233 fc->forget_batch = 16;
1234 }
1235
1236 req = list_entry(fc->pending.next, struct fuse_req, list);
1237 req->state = FUSE_REQ_READING;
1238 list_move(&req->list, &fc->io);
1239
1240 in = &req->in;
1241 reqsize = in->h.len;
1242 /* If request is too large, reply with an error and restart the read */
1243 if (nbytes < reqsize) {
1244 req->out.h.error = -EIO;
1245 /* SETXATTR is special, since it may contain too large data */
1246 if (in->h.opcode == FUSE_SETXATTR)
1247 req->out.h.error = -E2BIG;
1248 request_end(fc, req);
1249 goto restart;
1250 }
1251 spin_unlock(&fc->lock);
1252 cs->req = req;
1253 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1254 if (!err)
1255 err = fuse_copy_args(cs, in->numargs, in->argpages,
1256 (struct fuse_arg *) in->args, 0);
1257 fuse_copy_finish(cs);
1258 spin_lock(&fc->lock);
1259 req->locked = 0;
1260 if (req->aborted) {
1261 request_end(fc, req);
1262 return -ENODEV;
1263 }
1264 if (err) {
1265 req->out.h.error = -EIO;
1266 request_end(fc, req);
1267 return err;
1268 }
1269 if (!req->isreply)
1270 request_end(fc, req);
1271 else {
1272 req->state = FUSE_REQ_SENT;
1273 list_move_tail(&req->list, &fc->processing);
1274 if (req->interrupted)
1275 queue_interrupt(fc, req);
1276 spin_unlock(&fc->lock);
1277 }
1278 return reqsize;
1279
1280 err_unlock:
1281 spin_unlock(&fc->lock);
1282 return err;
1283 }
1284
1285 static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1286 unsigned long nr_segs, loff_t pos)
1287 {
1288 struct fuse_copy_state cs;
1289 struct file *file = iocb->ki_filp;
1290 struct fuse_conn *fc = fuse_get_conn(file);
1291 if (!fc)
1292 return -EPERM;
1293
1294 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1295
1296 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1297 }
1298
1299 static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe,
1300 struct pipe_buffer *buf)
1301 {
1302 return 1;
1303 }
1304
1305 static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = {
1306 .can_merge = 0,
1307 .map = generic_pipe_buf_map,
1308 .unmap = generic_pipe_buf_unmap,
1309 .confirm = generic_pipe_buf_confirm,
1310 .release = generic_pipe_buf_release,
1311 .steal = fuse_dev_pipe_buf_steal,
1312 .get = generic_pipe_buf_get,
1313 };
1314
1315 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1316 struct pipe_inode_info *pipe,
1317 size_t len, unsigned int flags)
1318 {
1319 int ret;
1320 int page_nr = 0;
1321 int do_wakeup = 0;
1322 struct pipe_buffer *bufs;
1323 struct fuse_copy_state cs;
1324 struct fuse_conn *fc = fuse_get_conn(in);
1325 if (!fc)
1326 return -EPERM;
1327
1328 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1329 if (!bufs)
1330 return -ENOMEM;
1331
1332 fuse_copy_init(&cs, fc, 1, NULL, 0);
1333 cs.pipebufs = bufs;
1334 cs.pipe = pipe;
1335 ret = fuse_dev_do_read(fc, in, &cs, len);
1336 if (ret < 0)
1337 goto out;
1338
1339 ret = 0;
1340 pipe_lock(pipe);
1341
1342 if (!pipe->readers) {
1343 send_sig(SIGPIPE, current, 0);
1344 if (!ret)
1345 ret = -EPIPE;
1346 goto out_unlock;
1347 }
1348
1349 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1350 ret = -EIO;
1351 goto out_unlock;
1352 }
1353
1354 while (page_nr < cs.nr_segs) {
1355 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1356 struct pipe_buffer *buf = pipe->bufs + newbuf;
1357
1358 buf->page = bufs[page_nr].page;
1359 buf->offset = bufs[page_nr].offset;
1360 buf->len = bufs[page_nr].len;
1361 buf->ops = &fuse_dev_pipe_buf_ops;
1362
1363 pipe->nrbufs++;
1364 page_nr++;
1365 ret += buf->len;
1366
1367 if (pipe->files)
1368 do_wakeup = 1;
1369 }
1370
1371 out_unlock:
1372 pipe_unlock(pipe);
1373
1374 if (do_wakeup) {
1375 smp_mb();
1376 if (waitqueue_active(&pipe->wait))
1377 wake_up_interruptible(&pipe->wait);
1378 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1379 }
1380
1381 out:
1382 for (; page_nr < cs.nr_segs; page_nr++)
1383 page_cache_release(bufs[page_nr].page);
1384
1385 kfree(bufs);
1386 return ret;
1387 }
1388
1389 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1390 struct fuse_copy_state *cs)
1391 {
1392 struct fuse_notify_poll_wakeup_out outarg;
1393 int err = -EINVAL;
1394
1395 if (size != sizeof(outarg))
1396 goto err;
1397
1398 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1399 if (err)
1400 goto err;
1401
1402 fuse_copy_finish(cs);
1403 return fuse_notify_poll_wakeup(fc, &outarg);
1404
1405 err:
1406 fuse_copy_finish(cs);
1407 return err;
1408 }
1409
1410 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1411 struct fuse_copy_state *cs)
1412 {
1413 struct fuse_notify_inval_inode_out outarg;
1414 int err = -EINVAL;
1415
1416 if (size != sizeof(outarg))
1417 goto err;
1418
1419 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1420 if (err)
1421 goto err;
1422 fuse_copy_finish(cs);
1423
1424 down_read(&fc->killsb);
1425 err = -ENOENT;
1426 if (fc->sb) {
1427 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1428 outarg.off, outarg.len);
1429 }
1430 up_read(&fc->killsb);
1431 return err;
1432
1433 err:
1434 fuse_copy_finish(cs);
1435 return err;
1436 }
1437
1438 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1439 struct fuse_copy_state *cs)
1440 {
1441 struct fuse_notify_inval_entry_out outarg;
1442 int err = -ENOMEM;
1443 char *buf;
1444 struct qstr name;
1445
1446 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1447 if (!buf)
1448 goto err;
1449
1450 err = -EINVAL;
1451 if (size < sizeof(outarg))
1452 goto err;
1453
1454 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1455 if (err)
1456 goto err;
1457
1458 err = -ENAMETOOLONG;
1459 if (outarg.namelen > FUSE_NAME_MAX)
1460 goto err;
1461
1462 err = -EINVAL;
1463 if (size != sizeof(outarg) + outarg.namelen + 1)
1464 goto err;
1465
1466 name.name = buf;
1467 name.len = outarg.namelen;
1468 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1469 if (err)
1470 goto err;
1471 fuse_copy_finish(cs);
1472 buf[outarg.namelen] = 0;
1473 name.hash = full_name_hash(name.name, name.len);
1474
1475 down_read(&fc->killsb);
1476 err = -ENOENT;
1477 if (fc->sb)
1478 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1479 up_read(&fc->killsb);
1480 kfree(buf);
1481 return err;
1482
1483 err:
1484 kfree(buf);
1485 fuse_copy_finish(cs);
1486 return err;
1487 }
1488
1489 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1490 struct fuse_copy_state *cs)
1491 {
1492 struct fuse_notify_delete_out outarg;
1493 int err = -ENOMEM;
1494 char *buf;
1495 struct qstr name;
1496
1497 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1498 if (!buf)
1499 goto err;
1500
1501 err = -EINVAL;
1502 if (size < sizeof(outarg))
1503 goto err;
1504
1505 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1506 if (err)
1507 goto err;
1508
1509 err = -ENAMETOOLONG;
1510 if (outarg.namelen > FUSE_NAME_MAX)
1511 goto err;
1512
1513 err = -EINVAL;
1514 if (size != sizeof(outarg) + outarg.namelen + 1)
1515 goto err;
1516
1517 name.name = buf;
1518 name.len = outarg.namelen;
1519 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1520 if (err)
1521 goto err;
1522 fuse_copy_finish(cs);
1523 buf[outarg.namelen] = 0;
1524 name.hash = full_name_hash(name.name, name.len);
1525
1526 down_read(&fc->killsb);
1527 err = -ENOENT;
1528 if (fc->sb)
1529 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1530 outarg.child, &name);
1531 up_read(&fc->killsb);
1532 kfree(buf);
1533 return err;
1534
1535 err:
1536 kfree(buf);
1537 fuse_copy_finish(cs);
1538 return err;
1539 }
1540
1541 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1542 struct fuse_copy_state *cs)
1543 {
1544 struct fuse_notify_store_out outarg;
1545 struct inode *inode;
1546 struct address_space *mapping;
1547 u64 nodeid;
1548 int err;
1549 pgoff_t index;
1550 unsigned int offset;
1551 unsigned int num;
1552 loff_t file_size;
1553 loff_t end;
1554
1555 err = -EINVAL;
1556 if (size < sizeof(outarg))
1557 goto out_finish;
1558
1559 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1560 if (err)
1561 goto out_finish;
1562
1563 err = -EINVAL;
1564 if (size - sizeof(outarg) != outarg.size)
1565 goto out_finish;
1566
1567 nodeid = outarg.nodeid;
1568
1569 down_read(&fc->killsb);
1570
1571 err = -ENOENT;
1572 if (!fc->sb)
1573 goto out_up_killsb;
1574
1575 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1576 if (!inode)
1577 goto out_up_killsb;
1578
1579 mapping = inode->i_mapping;
1580 index = outarg.offset >> PAGE_CACHE_SHIFT;
1581 offset = outarg.offset & ~PAGE_CACHE_MASK;
1582 file_size = i_size_read(inode);
1583 end = outarg.offset + outarg.size;
1584 if (end > file_size) {
1585 file_size = end;
1586 fuse_write_update_size(inode, file_size);
1587 }
1588
1589 num = outarg.size;
1590 while (num) {
1591 struct page *page;
1592 unsigned int this_num;
1593
1594 err = -ENOMEM;
1595 page = find_or_create_page(mapping, index,
1596 mapping_gfp_mask(mapping));
1597 if (!page)
1598 goto out_iput;
1599
1600 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1601 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1602 if (!err && offset == 0 && (num != 0 || file_size == end))
1603 SetPageUptodate(page);
1604 unlock_page(page);
1605 page_cache_release(page);
1606
1607 if (err)
1608 goto out_iput;
1609
1610 num -= this_num;
1611 offset = 0;
1612 index++;
1613 }
1614
1615 err = 0;
1616
1617 out_iput:
1618 iput(inode);
1619 out_up_killsb:
1620 up_read(&fc->killsb);
1621 out_finish:
1622 fuse_copy_finish(cs);
1623 return err;
1624 }
1625
1626 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1627 {
1628 release_pages(req->pages, req->num_pages, 0);
1629 }
1630
1631 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1632 struct fuse_notify_retrieve_out *outarg)
1633 {
1634 int err;
1635 struct address_space *mapping = inode->i_mapping;
1636 struct fuse_req *req;
1637 pgoff_t index;
1638 loff_t file_size;
1639 unsigned int num;
1640 unsigned int offset;
1641 size_t total_len = 0;
1642 int num_pages;
1643
1644 offset = outarg->offset & ~PAGE_CACHE_MASK;
1645 file_size = i_size_read(inode);
1646
1647 num = outarg->size;
1648 if (outarg->offset > file_size)
1649 num = 0;
1650 else if (outarg->offset + num > file_size)
1651 num = file_size - outarg->offset;
1652
1653 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1654 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1655
1656 req = fuse_get_req(fc, num_pages);
1657 if (IS_ERR(req))
1658 return PTR_ERR(req);
1659
1660 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1661 req->in.h.nodeid = outarg->nodeid;
1662 req->in.numargs = 2;
1663 req->in.argpages = 1;
1664 req->page_descs[0].offset = offset;
1665 req->end = fuse_retrieve_end;
1666
1667 index = outarg->offset >> PAGE_CACHE_SHIFT;
1668
1669 while (num && req->num_pages < num_pages) {
1670 struct page *page;
1671 unsigned int this_num;
1672
1673 page = find_get_page(mapping, index);
1674 if (!page)
1675 break;
1676
1677 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1678 req->pages[req->num_pages] = page;
1679 req->page_descs[req->num_pages].length = this_num;
1680 req->num_pages++;
1681
1682 offset = 0;
1683 num -= this_num;
1684 total_len += this_num;
1685 index++;
1686 }
1687 req->misc.retrieve_in.offset = outarg->offset;
1688 req->misc.retrieve_in.size = total_len;
1689 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1690 req->in.args[0].value = &req->misc.retrieve_in;
1691 req->in.args[1].size = total_len;
1692
1693 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1694 if (err)
1695 fuse_retrieve_end(fc, req);
1696
1697 return err;
1698 }
1699
1700 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1701 struct fuse_copy_state *cs)
1702 {
1703 struct fuse_notify_retrieve_out outarg;
1704 struct inode *inode;
1705 int err;
1706
1707 err = -EINVAL;
1708 if (size != sizeof(outarg))
1709 goto copy_finish;
1710
1711 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1712 if (err)
1713 goto copy_finish;
1714
1715 fuse_copy_finish(cs);
1716
1717 down_read(&fc->killsb);
1718 err = -ENOENT;
1719 if (fc->sb) {
1720 u64 nodeid = outarg.nodeid;
1721
1722 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1723 if (inode) {
1724 err = fuse_retrieve(fc, inode, &outarg);
1725 iput(inode);
1726 }
1727 }
1728 up_read(&fc->killsb);
1729
1730 return err;
1731
1732 copy_finish:
1733 fuse_copy_finish(cs);
1734 return err;
1735 }
1736
1737 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1738 unsigned int size, struct fuse_copy_state *cs)
1739 {
1740 switch (code) {
1741 case FUSE_NOTIFY_POLL:
1742 return fuse_notify_poll(fc, size, cs);
1743
1744 case FUSE_NOTIFY_INVAL_INODE:
1745 return fuse_notify_inval_inode(fc, size, cs);
1746
1747 case FUSE_NOTIFY_INVAL_ENTRY:
1748 return fuse_notify_inval_entry(fc, size, cs);
1749
1750 case FUSE_NOTIFY_STORE:
1751 return fuse_notify_store(fc, size, cs);
1752
1753 case FUSE_NOTIFY_RETRIEVE:
1754 return fuse_notify_retrieve(fc, size, cs);
1755
1756 case FUSE_NOTIFY_DELETE:
1757 return fuse_notify_delete(fc, size, cs);
1758
1759 default:
1760 fuse_copy_finish(cs);
1761 return -EINVAL;
1762 }
1763 }
1764
1765 /* Look up request on processing list by unique ID */
1766 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1767 {
1768 struct list_head *entry;
1769
1770 list_for_each(entry, &fc->processing) {
1771 struct fuse_req *req;
1772 req = list_entry(entry, struct fuse_req, list);
1773 if (req->in.h.unique == unique || req->intr_unique == unique)
1774 return req;
1775 }
1776 return NULL;
1777 }
1778
1779 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1780 unsigned nbytes)
1781 {
1782 unsigned reqsize = sizeof(struct fuse_out_header);
1783
1784 if (out->h.error)
1785 return nbytes != reqsize ? -EINVAL : 0;
1786
1787 reqsize += len_args(out->numargs, out->args);
1788
1789 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1790 return -EINVAL;
1791 else if (reqsize > nbytes) {
1792 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1793 unsigned diffsize = reqsize - nbytes;
1794 if (diffsize > lastarg->size)
1795 return -EINVAL;
1796 lastarg->size -= diffsize;
1797 }
1798 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1799 out->page_zeroing);
1800 }
1801
1802 /*
1803 * Write a single reply to a request. First the header is copied from
1804 * the write buffer. The request is then searched on the processing
1805 * list by the unique ID found in the header. If found, then remove
1806 * it from the list and copy the rest of the buffer to the request.
1807 * The request is finished by calling request_end()
1808 */
1809 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1810 struct fuse_copy_state *cs, size_t nbytes)
1811 {
1812 int err;
1813 struct fuse_req *req;
1814 struct fuse_out_header oh;
1815
1816 if (nbytes < sizeof(struct fuse_out_header))
1817 return -EINVAL;
1818
1819 err = fuse_copy_one(cs, &oh, sizeof(oh));
1820 if (err)
1821 goto err_finish;
1822
1823 err = -EINVAL;
1824 if (oh.len != nbytes)
1825 goto err_finish;
1826
1827 /*
1828 * Zero oh.unique indicates unsolicited notification message
1829 * and error contains notification code.
1830 */
1831 if (!oh.unique) {
1832 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1833 return err ? err : nbytes;
1834 }
1835
1836 err = -EINVAL;
1837 if (oh.error <= -1000 || oh.error > 0)
1838 goto err_finish;
1839
1840 spin_lock(&fc->lock);
1841 err = -ENOENT;
1842 if (!fc->connected)
1843 goto err_unlock;
1844
1845 req = request_find(fc, oh.unique);
1846 if (!req)
1847 goto err_unlock;
1848
1849 if (req->aborted) {
1850 spin_unlock(&fc->lock);
1851 fuse_copy_finish(cs);
1852 spin_lock(&fc->lock);
1853 request_end(fc, req);
1854 return -ENOENT;
1855 }
1856 /* Is it an interrupt reply? */
1857 if (req->intr_unique == oh.unique) {
1858 err = -EINVAL;
1859 if (nbytes != sizeof(struct fuse_out_header))
1860 goto err_unlock;
1861
1862 if (oh.error == -ENOSYS)
1863 fc->no_interrupt = 1;
1864 else if (oh.error == -EAGAIN)
1865 queue_interrupt(fc, req);
1866
1867 spin_unlock(&fc->lock);
1868 fuse_copy_finish(cs);
1869 return nbytes;
1870 }
1871
1872 req->state = FUSE_REQ_WRITING;
1873 list_move(&req->list, &fc->io);
1874 req->out.h = oh;
1875 req->locked = 1;
1876 cs->req = req;
1877 if (!req->out.page_replace)
1878 cs->move_pages = 0;
1879 spin_unlock(&fc->lock);
1880
1881 err = copy_out_args(cs, &req->out, nbytes);
1882 fuse_copy_finish(cs);
1883
1884 spin_lock(&fc->lock);
1885 req->locked = 0;
1886 if (!err) {
1887 if (req->aborted)
1888 err = -ENOENT;
1889 } else if (!req->aborted)
1890 req->out.h.error = -EIO;
1891 request_end(fc, req);
1892
1893 return err ? err : nbytes;
1894
1895 err_unlock:
1896 spin_unlock(&fc->lock);
1897 err_finish:
1898 fuse_copy_finish(cs);
1899 return err;
1900 }
1901
1902 static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1903 unsigned long nr_segs, loff_t pos)
1904 {
1905 struct fuse_copy_state cs;
1906 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1907 if (!fc)
1908 return -EPERM;
1909
1910 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1911
1912 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1913 }
1914
1915 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1916 struct file *out, loff_t *ppos,
1917 size_t len, unsigned int flags)
1918 {
1919 unsigned nbuf;
1920 unsigned idx;
1921 struct pipe_buffer *bufs;
1922 struct fuse_copy_state cs;
1923 struct fuse_conn *fc;
1924 size_t rem;
1925 ssize_t ret;
1926
1927 fc = fuse_get_conn(out);
1928 if (!fc)
1929 return -EPERM;
1930
1931 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1932 if (!bufs)
1933 return -ENOMEM;
1934
1935 pipe_lock(pipe);
1936 nbuf = 0;
1937 rem = 0;
1938 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1939 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1940
1941 ret = -EINVAL;
1942 if (rem < len) {
1943 pipe_unlock(pipe);
1944 goto out;
1945 }
1946
1947 rem = len;
1948 while (rem) {
1949 struct pipe_buffer *ibuf;
1950 struct pipe_buffer *obuf;
1951
1952 BUG_ON(nbuf >= pipe->buffers);
1953 BUG_ON(!pipe->nrbufs);
1954 ibuf = &pipe->bufs[pipe->curbuf];
1955 obuf = &bufs[nbuf];
1956
1957 if (rem >= ibuf->len) {
1958 *obuf = *ibuf;
1959 ibuf->ops = NULL;
1960 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1961 pipe->nrbufs--;
1962 } else {
1963 ibuf->ops->get(pipe, ibuf);
1964 *obuf = *ibuf;
1965 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1966 obuf->len = rem;
1967 ibuf->offset += obuf->len;
1968 ibuf->len -= obuf->len;
1969 }
1970 nbuf++;
1971 rem -= obuf->len;
1972 }
1973 pipe_unlock(pipe);
1974
1975 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1976 cs.pipebufs = bufs;
1977 cs.pipe = pipe;
1978
1979 if (flags & SPLICE_F_MOVE)
1980 cs.move_pages = 1;
1981
1982 ret = fuse_dev_do_write(fc, &cs, len);
1983
1984 for (idx = 0; idx < nbuf; idx++) {
1985 struct pipe_buffer *buf = &bufs[idx];
1986 buf->ops->release(pipe, buf);
1987 }
1988 out:
1989 kfree(bufs);
1990 return ret;
1991 }
1992
1993 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1994 {
1995 unsigned mask = POLLOUT | POLLWRNORM;
1996 struct fuse_conn *fc = fuse_get_conn(file);
1997 if (!fc)
1998 return POLLERR;
1999
2000 poll_wait(file, &fc->waitq, wait);
2001
2002 spin_lock(&fc->lock);
2003 if (!fc->connected)
2004 mask = POLLERR;
2005 else if (request_pending(fc))
2006 mask |= POLLIN | POLLRDNORM;
2007 spin_unlock(&fc->lock);
2008
2009 return mask;
2010 }
2011
2012 /*
2013 * Abort all requests on the given list (pending or processing)
2014 *
2015 * This function releases and reacquires fc->lock
2016 */
2017 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2018 __releases(fc->lock)
2019 __acquires(fc->lock)
2020 {
2021 while (!list_empty(head)) {
2022 struct fuse_req *req;
2023 req = list_entry(head->next, struct fuse_req, list);
2024 req->out.h.error = -ECONNABORTED;
2025 request_end(fc, req);
2026 spin_lock(&fc->lock);
2027 }
2028 }
2029
2030 /*
2031 * Abort requests under I/O
2032 *
2033 * The requests are set to aborted and finished, and the request
2034 * waiter is woken up. This will make request_wait_answer() wait
2035 * until the request is unlocked and then return.
2036 *
2037 * If the request is asynchronous, then the end function needs to be
2038 * called after waiting for the request to be unlocked (if it was
2039 * locked).
2040 */
2041 static void end_io_requests(struct fuse_conn *fc)
2042 __releases(fc->lock)
2043 __acquires(fc->lock)
2044 {
2045 while (!list_empty(&fc->io)) {
2046 struct fuse_req *req =
2047 list_entry(fc->io.next, struct fuse_req, list);
2048 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
2049
2050 req->aborted = 1;
2051 req->out.h.error = -ECONNABORTED;
2052 req->state = FUSE_REQ_FINISHED;
2053 list_del_init(&req->list);
2054 wake_up(&req->waitq);
2055 if (end) {
2056 req->end = NULL;
2057 __fuse_get_request(req);
2058 spin_unlock(&fc->lock);
2059 wait_event(req->waitq, !req->locked);
2060 end(fc, req);
2061 fuse_put_request(fc, req);
2062 spin_lock(&fc->lock);
2063 }
2064 }
2065 }
2066
2067 static void end_queued_requests(struct fuse_conn *fc)
2068 __releases(fc->lock)
2069 __acquires(fc->lock)
2070 {
2071 fc->max_background = UINT_MAX;
2072 flush_bg_queue(fc);
2073 end_requests(fc, &fc->pending);
2074 end_requests(fc, &fc->processing);
2075 while (forget_pending(fc))
2076 kfree(dequeue_forget(fc, 1, NULL));
2077 }
2078
2079 static void end_polls(struct fuse_conn *fc)
2080 {
2081 struct rb_node *p;
2082
2083 p = rb_first(&fc->polled_files);
2084
2085 while (p) {
2086 struct fuse_file *ff;
2087 ff = rb_entry(p, struct fuse_file, polled_node);
2088 wake_up_interruptible_all(&ff->poll_wait);
2089
2090 p = rb_next(p);
2091 }
2092 }
2093
2094 /*
2095 * Abort all requests.
2096 *
2097 * Emergency exit in case of a malicious or accidental deadlock, or
2098 * just a hung filesystem.
2099 *
2100 * The same effect is usually achievable through killing the
2101 * filesystem daemon and all users of the filesystem. The exception
2102 * is the combination of an asynchronous request and the tricky
2103 * deadlock (see Documentation/filesystems/fuse.txt).
2104 *
2105 * During the aborting, progression of requests from the pending and
2106 * processing lists onto the io list, and progression of new requests
2107 * onto the pending list is prevented by req->connected being false.
2108 *
2109 * Progression of requests under I/O to the processing list is
2110 * prevented by the req->aborted flag being true for these requests.
2111 * For this reason requests on the io list must be aborted first.
2112 */
2113 void fuse_abort_conn(struct fuse_conn *fc)
2114 {
2115 spin_lock(&fc->lock);
2116 if (fc->connected) {
2117 fc->connected = 0;
2118 fc->blocked = 0;
2119 fc->initialized = 1;
2120 end_io_requests(fc);
2121 end_queued_requests(fc);
2122 end_polls(fc);
2123 wake_up_all(&fc->waitq);
2124 wake_up_all(&fc->blocked_waitq);
2125 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2126 }
2127 spin_unlock(&fc->lock);
2128 }
2129 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2130
2131 int fuse_dev_release(struct inode *inode, struct file *file)
2132 {
2133 struct fuse_conn *fc = fuse_get_conn(file);
2134 if (fc) {
2135 spin_lock(&fc->lock);
2136 fc->connected = 0;
2137 fc->blocked = 0;
2138 fc->initialized = 1;
2139 end_queued_requests(fc);
2140 end_polls(fc);
2141 wake_up_all(&fc->blocked_waitq);
2142 spin_unlock(&fc->lock);
2143 fuse_conn_put(fc);
2144 }
2145
2146 return 0;
2147 }
2148 EXPORT_SYMBOL_GPL(fuse_dev_release);
2149
2150 static int fuse_dev_fasync(int fd, struct file *file, int on)
2151 {
2152 struct fuse_conn *fc = fuse_get_conn(file);
2153 if (!fc)
2154 return -EPERM;
2155
2156 /* No locking - fasync_helper does its own locking */
2157 return fasync_helper(fd, file, on, &fc->fasync);
2158 }
2159
2160 const struct file_operations fuse_dev_operations = {
2161 .owner = THIS_MODULE,
2162 .llseek = no_llseek,
2163 .read = do_sync_read,
2164 .aio_read = fuse_dev_read,
2165 .splice_read = fuse_dev_splice_read,
2166 .write = do_sync_write,
2167 .aio_write = fuse_dev_write,
2168 .splice_write = fuse_dev_splice_write,
2169 .poll = fuse_dev_poll,
2170 .release = fuse_dev_release,
2171 .fasync = fuse_dev_fasync,
2172 };
2173 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2174
2175 static struct miscdevice fuse_miscdevice = {
2176 .minor = FUSE_MINOR,
2177 .name = "fuse",
2178 .fops = &fuse_dev_operations,
2179 };
2180
2181 int __init fuse_dev_init(void)
2182 {
2183 int err = -ENOMEM;
2184 fuse_req_cachep = kmem_cache_create("fuse_request",
2185 sizeof(struct fuse_req),
2186 0, 0, NULL);
2187 if (!fuse_req_cachep)
2188 goto out;
2189
2190 err = misc_register(&fuse_miscdevice);
2191 if (err)
2192 goto out_cache_clean;
2193
2194 return 0;
2195
2196 out_cache_clean:
2197 kmem_cache_destroy(fuse_req_cachep);
2198 out:
2199 return err;
2200 }
2201
2202 void fuse_dev_cleanup(void)
2203 {
2204 misc_deregister(&fuse_miscdevice);
2205 kmem_cache_destroy(fuse_req_cachep);
2206 }
Linus' kernel tree