projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
[deliverable/linux.git] / fs / fuse / dev.c
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 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
515 {
516 struct fuse_req *req;
517 ssize_t ret;
518
519 req = fuse_get_req(fc, 0);
520 if (IS_ERR(req))
521 return PTR_ERR(req);
522
523 req->in.h.opcode = args->in.h.opcode;
524 req->in.h.nodeid = args->in.h.nodeid;
525 req->in.numargs = args->in.numargs;
526 memcpy(req->in.args, args->in.args,
527 args->in.numargs * sizeof(struct fuse_in_arg));
528 req->out.argvar = args->out.argvar;
529 req->out.numargs = args->out.numargs;
530 memcpy(req->out.args, args->out.args,
531 args->out.numargs * sizeof(struct fuse_arg));
532 fuse_request_send(fc, req);
533 ret = req->out.h.error;
534 if (!ret && args->out.argvar) {
535 BUG_ON(args->out.numargs != 1);
536 ret = req->out.args[0].size;
537 }
538 fuse_put_request(fc, req);
539
540 return ret;
541 }
542
543 static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
544 struct fuse_req *req)
545 {
546 BUG_ON(!req->background);
547 fc->num_background++;
548 if (fc->num_background == fc->max_background)
549 fc->blocked = 1;
550 if (fc->num_background == fc->congestion_threshold &&
551 fc->bdi_initialized) {
552 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
553 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
554 }
555 list_add_tail(&req->list, &fc->bg_queue);
556 flush_bg_queue(fc);
557 }
558
559 static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
560 {
561 spin_lock(&fc->lock);
562 if (fc->connected) {
563 fuse_request_send_nowait_locked(fc, req);
564 spin_unlock(&fc->lock);
565 } else {
566 req->out.h.error = -ENOTCONN;
567 request_end(fc, req);
568 }
569 }
570
571 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
572 {
573 req->isreply = 1;
574 fuse_request_send_nowait(fc, req);
575 }
576 EXPORT_SYMBOL_GPL(fuse_request_send_background);
577
578 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
579 struct fuse_req *req, u64 unique)
580 {
581 int err = -ENODEV;
582
583 req->isreply = 0;
584 req->in.h.unique = unique;
585 spin_lock(&fc->lock);
586 if (fc->connected) {
587 queue_request(fc, req);
588 err = 0;
589 }
590 spin_unlock(&fc->lock);
591
592 return err;
593 }
594
595 /*
596 * Called under fc->lock
597 *
598 * fc->connected must have been checked previously
599 */
600 void fuse_request_send_background_locked(struct fuse_conn *fc,
601 struct fuse_req *req)
602 {
603 req->isreply = 1;
604 fuse_request_send_nowait_locked(fc, req);
605 }
606
607 void fuse_force_forget(struct file *file, u64 nodeid)
608 {
609 struct inode *inode = file_inode(file);
610 struct fuse_conn *fc = get_fuse_conn(inode);
611 struct fuse_req *req;
612 struct fuse_forget_in inarg;
613
614 memset(&inarg, 0, sizeof(inarg));
615 inarg.nlookup = 1;
616 req = fuse_get_req_nofail_nopages(fc, file);
617 req->in.h.opcode = FUSE_FORGET;
618 req->in.h.nodeid = nodeid;
619 req->in.numargs = 1;
620 req->in.args[0].size = sizeof(inarg);
621 req->in.args[0].value = &inarg;
622 req->isreply = 0;
623 __fuse_request_send(fc, req);
624 /* ignore errors */
625 fuse_put_request(fc, req);
626 }
627
628 /*
629 * Lock the request. Up to the next unlock_request() there mustn't be
630 * anything that could cause a page-fault. If the request was already
631 * aborted bail out.
632 */
633 static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
634 {
635 int err = 0;
636 if (req) {
637 spin_lock(&fc->lock);
638 if (req->aborted)
639 err = -ENOENT;
640 else
641 req->locked = 1;
642 spin_unlock(&fc->lock);
643 }
644 return err;
645 }
646
647 /*
648 * Unlock request. If it was aborted during being locked, the
649 * requester thread is currently waiting for it to be unlocked, so
650 * wake it up.
651 */
652 static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
653 {
654 if (req) {
655 spin_lock(&fc->lock);
656 req->locked = 0;
657 if (req->aborted)
658 wake_up(&req->waitq);
659 spin_unlock(&fc->lock);
660 }
661 }
662
663 struct fuse_copy_state {
664 struct fuse_conn *fc;
665 int write;
666 struct fuse_req *req;
667 const struct iovec *iov;
668 struct pipe_buffer *pipebufs;
669 struct pipe_buffer *currbuf;
670 struct pipe_inode_info *pipe;
671 unsigned long nr_segs;
672 unsigned long seglen;
673 unsigned long addr;
674 struct page *pg;
675 unsigned len;
676 unsigned offset;
677 unsigned move_pages:1;
678 };
679
680 static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
681 int write,
682 const struct iovec *iov, unsigned long nr_segs)
683 {
684 memset(cs, 0, sizeof(*cs));
685 cs->fc = fc;
686 cs->write = write;
687 cs->iov = iov;
688 cs->nr_segs = nr_segs;
689 }
690
691 /* Unmap and put previous page of userspace buffer */
692 static void fuse_copy_finish(struct fuse_copy_state *cs)
693 {
694 if (cs->currbuf) {
695 struct pipe_buffer *buf = cs->currbuf;
696
697 if (cs->write)
698 buf->len = PAGE_SIZE - cs->len;
699 cs->currbuf = NULL;
700 } else if (cs->pg) {
701 if (cs->write) {
702 flush_dcache_page(cs->pg);
703 set_page_dirty_lock(cs->pg);
704 }
705 put_page(cs->pg);
706 }
707 cs->pg = NULL;
708 }
709
710 /*
711 * Get another pagefull of userspace buffer, and map it to kernel
712 * address space, and lock request
713 */
714 static int fuse_copy_fill(struct fuse_copy_state *cs)
715 {
716 struct page *page;
717 int err;
718
719 unlock_request(cs->fc, cs->req);
720 fuse_copy_finish(cs);
721 if (cs->pipebufs) {
722 struct pipe_buffer *buf = cs->pipebufs;
723
724 if (!cs->write) {
725 err = buf->ops->confirm(cs->pipe, buf);
726 if (err)
727 return err;
728
729 BUG_ON(!cs->nr_segs);
730 cs->currbuf = buf;
731 cs->pg = buf->page;
732 cs->offset = buf->offset;
733 cs->len = buf->len;
734 cs->pipebufs++;
735 cs->nr_segs--;
736 } else {
737 if (cs->nr_segs == cs->pipe->buffers)
738 return -EIO;
739
740 page = alloc_page(GFP_HIGHUSER);
741 if (!page)
742 return -ENOMEM;
743
744 buf->page = page;
745 buf->offset = 0;
746 buf->len = 0;
747
748 cs->currbuf = buf;
749 cs->pg = page;
750 cs->offset = 0;
751 cs->len = PAGE_SIZE;
752 cs->pipebufs++;
753 cs->nr_segs++;
754 }
755 } else {
756 if (!cs->seglen) {
757 BUG_ON(!cs->nr_segs);
758 cs->seglen = cs->iov[0].iov_len;
759 cs->addr = (unsigned long) cs->iov[0].iov_base;
760 cs->iov++;
761 cs->nr_segs--;
762 }
763 err = get_user_pages_fast(cs->addr, 1, cs->write, &page);
764 if (err < 0)
765 return err;
766 BUG_ON(err != 1);
767 cs->pg = page;
768 cs->offset = cs->addr % PAGE_SIZE;
769 cs->len = min(PAGE_SIZE - cs->offset, cs->seglen);
770 cs->seglen -= cs->len;
771 cs->addr += cs->len;
772 }
773
774 return lock_request(cs->fc, cs->req);
775 }
776
777 /* Do as much copy to/from userspace buffer as we can */
778 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
779 {
780 unsigned ncpy = min(*size, cs->len);
781 if (val) {
782 void *pgaddr = kmap_atomic(cs->pg);
783 void *buf = pgaddr + cs->offset;
784
785 if (cs->write)
786 memcpy(buf, *val, ncpy);
787 else
788 memcpy(*val, buf, ncpy);
789
790 kunmap_atomic(pgaddr);
791 *val += ncpy;
792 }
793 *size -= ncpy;
794 cs->len -= ncpy;
795 cs->offset += ncpy;
796 return ncpy;
797 }
798
799 static int fuse_check_page(struct page *page)
800 {
801 if (page_mapcount(page) ||
802 page->mapping != NULL ||
803 page_count(page) != 1 ||
804 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
805 ~(1 << PG_locked |
806 1 << PG_referenced |
807 1 << PG_uptodate |
808 1 << PG_lru |
809 1 << PG_active |
810 1 << PG_reclaim))) {
811 printk(KERN_WARNING "fuse: trying to steal weird page\n");
812 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);
813 return 1;
814 }
815 return 0;
816 }
817
818 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
819 {
820 int err;
821 struct page *oldpage = *pagep;
822 struct page *newpage;
823 struct pipe_buffer *buf = cs->pipebufs;
824
825 unlock_request(cs->fc, cs->req);
826 fuse_copy_finish(cs);
827
828 err = buf->ops->confirm(cs->pipe, buf);
829 if (err)
830 return err;
831
832 BUG_ON(!cs->nr_segs);
833 cs->currbuf = buf;
834 cs->len = buf->len;
835 cs->pipebufs++;
836 cs->nr_segs--;
837
838 if (cs->len != PAGE_SIZE)
839 goto out_fallback;
840
841 if (buf->ops->steal(cs->pipe, buf) != 0)
842 goto out_fallback;
843
844 newpage = buf->page;
845
846 if (WARN_ON(!PageUptodate(newpage)))
847 return -EIO;
848
849 ClearPageMappedToDisk(newpage);
850
851 if (fuse_check_page(newpage) != 0)
852 goto out_fallback_unlock;
853
854 /*
855 * This is a new and locked page, it shouldn't be mapped or
856 * have any special flags on it
857 */
858 if (WARN_ON(page_mapped(oldpage)))
859 goto out_fallback_unlock;
860 if (WARN_ON(page_has_private(oldpage)))
861 goto out_fallback_unlock;
862 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
863 goto out_fallback_unlock;
864 if (WARN_ON(PageMlocked(oldpage)))
865 goto out_fallback_unlock;
866
867 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
868 if (err) {
869 unlock_page(newpage);
870 return err;
871 }
872
873 page_cache_get(newpage);
874
875 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
876 lru_cache_add_file(newpage);
877
878 err = 0;
879 spin_lock(&cs->fc->lock);
880 if (cs->req->aborted)
881 err = -ENOENT;
882 else
883 *pagep = newpage;
884 spin_unlock(&cs->fc->lock);
885
886 if (err) {
887 unlock_page(newpage);
888 page_cache_release(newpage);
889 return err;
890 }
891
892 unlock_page(oldpage);
893 page_cache_release(oldpage);
894 cs->len = 0;
895
896 return 0;
897
898 out_fallback_unlock:
899 unlock_page(newpage);
900 out_fallback:
901 cs->pg = buf->page;
902 cs->offset = buf->offset;
903
904 err = lock_request(cs->fc, cs->req);
905 if (err)
906 return err;
907
908 return 1;
909 }
910
911 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
912 unsigned offset, unsigned count)
913 {
914 struct pipe_buffer *buf;
915
916 if (cs->nr_segs == cs->pipe->buffers)
917 return -EIO;
918
919 unlock_request(cs->fc, cs->req);
920 fuse_copy_finish(cs);
921
922 buf = cs->pipebufs;
923 page_cache_get(page);
924 buf->page = page;
925 buf->offset = offset;
926 buf->len = count;
927
928 cs->pipebufs++;
929 cs->nr_segs++;
930 cs->len = 0;
931
932 return 0;
933 }
934
935 /*
936 * Copy a page in the request to/from the userspace buffer. Must be
937 * done atomically
938 */
939 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
940 unsigned offset, unsigned count, int zeroing)
941 {
942 int err;
943 struct page *page = *pagep;
944
945 if (page && zeroing && count < PAGE_SIZE)
946 clear_highpage(page);
947
948 while (count) {
949 if (cs->write && cs->pipebufs && page) {
950 return fuse_ref_page(cs, page, offset, count);
951 } else if (!cs->len) {
952 if (cs->move_pages && page &&
953 offset == 0 && count == PAGE_SIZE) {
954 err = fuse_try_move_page(cs, pagep);
955 if (err <= 0)
956 return err;
957 } else {
958 err = fuse_copy_fill(cs);
959 if (err)
960 return err;
961 }
962 }
963 if (page) {
964 void *mapaddr = kmap_atomic(page);
965 void *buf = mapaddr + offset;
966 offset += fuse_copy_do(cs, &buf, &count);
967 kunmap_atomic(mapaddr);
968 } else
969 offset += fuse_copy_do(cs, NULL, &count);
970 }
971 if (page && !cs->write)
972 flush_dcache_page(page);
973 return 0;
974 }
975
976 /* Copy pages in the request to/from userspace buffer */
977 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
978 int zeroing)
979 {
980 unsigned i;
981 struct fuse_req *req = cs->req;
982
983 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
984 int err;
985 unsigned offset = req->page_descs[i].offset;
986 unsigned count = min(nbytes, req->page_descs[i].length);
987
988 err = fuse_copy_page(cs, &req->pages[i], offset, count,
989 zeroing);
990 if (err)
991 return err;
992
993 nbytes -= count;
994 }
995 return 0;
996 }
997
998 /* Copy a single argument in the request to/from userspace buffer */
999 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1000 {
1001 while (size) {
1002 if (!cs->len) {
1003 int err = fuse_copy_fill(cs);
1004 if (err)
1005 return err;
1006 }
1007 fuse_copy_do(cs, &val, &size);
1008 }
1009 return 0;
1010 }
1011
1012 /* Copy request arguments to/from userspace buffer */
1013 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1014 unsigned argpages, struct fuse_arg *args,
1015 int zeroing)
1016 {
1017 int err = 0;
1018 unsigned i;
1019
1020 for (i = 0; !err && i < numargs; i++) {
1021 struct fuse_arg *arg = &args[i];
1022 if (i == numargs - 1 && argpages)
1023 err = fuse_copy_pages(cs, arg->size, zeroing);
1024 else
1025 err = fuse_copy_one(cs, arg->value, arg->size);
1026 }
1027 return err;
1028 }
1029
1030 static int forget_pending(struct fuse_conn *fc)
1031 {
1032 return fc->forget_list_head.next != NULL;
1033 }
1034
1035 static int request_pending(struct fuse_conn *fc)
1036 {
1037 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
1038 forget_pending(fc);
1039 }
1040
1041 /* Wait until a request is available on the pending list */
1042 static void request_wait(struct fuse_conn *fc)
1043 __releases(fc->lock)
1044 __acquires(fc->lock)
1045 {
1046 DECLARE_WAITQUEUE(wait, current);
1047
1048 add_wait_queue_exclusive(&fc->waitq, &wait);
1049 while (fc->connected && !request_pending(fc)) {
1050 set_current_state(TASK_INTERRUPTIBLE);
1051 if (signal_pending(current))
1052 break;
1053
1054 spin_unlock(&fc->lock);
1055 schedule();
1056 spin_lock(&fc->lock);
1057 }
1058 set_current_state(TASK_RUNNING);
1059 remove_wait_queue(&fc->waitq, &wait);
1060 }
1061
1062 /*
1063 * Transfer an interrupt request to userspace
1064 *
1065 * Unlike other requests this is assembled on demand, without a need
1066 * to allocate a separate fuse_req structure.
1067 *
1068 * Called with fc->lock held, releases it
1069 */
1070 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1071 size_t nbytes, struct fuse_req *req)
1072 __releases(fc->lock)
1073 {
1074 struct fuse_in_header ih;
1075 struct fuse_interrupt_in arg;
1076 unsigned reqsize = sizeof(ih) + sizeof(arg);
1077 int err;
1078
1079 list_del_init(&req->intr_entry);
1080 req->intr_unique = fuse_get_unique(fc);
1081 memset(&ih, 0, sizeof(ih));
1082 memset(&arg, 0, sizeof(arg));
1083 ih.len = reqsize;
1084 ih.opcode = FUSE_INTERRUPT;
1085 ih.unique = req->intr_unique;
1086 arg.unique = req->in.h.unique;
1087
1088 spin_unlock(&fc->lock);
1089 if (nbytes < reqsize)
1090 return -EINVAL;
1091
1092 err = fuse_copy_one(cs, &ih, sizeof(ih));
1093 if (!err)
1094 err = fuse_copy_one(cs, &arg, sizeof(arg));
1095 fuse_copy_finish(cs);
1096
1097 return err ? err : reqsize;
1098 }
1099
1100 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
1101 unsigned max,
1102 unsigned *countp)
1103 {
1104 struct fuse_forget_link *head = fc->forget_list_head.next;
1105 struct fuse_forget_link **newhead = &head;
1106 unsigned count;
1107
1108 for (count = 0; *newhead != NULL && count < max; count++)
1109 newhead = &(*newhead)->next;
1110
1111 fc->forget_list_head.next = *newhead;
1112 *newhead = NULL;
1113 if (fc->forget_list_head.next == NULL)
1114 fc->forget_list_tail = &fc->forget_list_head;
1115
1116 if (countp != NULL)
1117 *countp = count;
1118
1119 return head;
1120 }
1121
1122 static int fuse_read_single_forget(struct fuse_conn *fc,
1123 struct fuse_copy_state *cs,
1124 size_t nbytes)
1125 __releases(fc->lock)
1126 {
1127 int err;
1128 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1129 struct fuse_forget_in arg = {
1130 .nlookup = forget->forget_one.nlookup,
1131 };
1132 struct fuse_in_header ih = {
1133 .opcode = FUSE_FORGET,
1134 .nodeid = forget->forget_one.nodeid,
1135 .unique = fuse_get_unique(fc),
1136 .len = sizeof(ih) + sizeof(arg),
1137 };
1138
1139 spin_unlock(&fc->lock);
1140 kfree(forget);
1141 if (nbytes < ih.len)
1142 return -EINVAL;
1143
1144 err = fuse_copy_one(cs, &ih, sizeof(ih));
1145 if (!err)
1146 err = fuse_copy_one(cs, &arg, sizeof(arg));
1147 fuse_copy_finish(cs);
1148
1149 if (err)
1150 return err;
1151
1152 return ih.len;
1153 }
1154
1155 static int fuse_read_batch_forget(struct fuse_conn *fc,
1156 struct fuse_copy_state *cs, size_t nbytes)
1157 __releases(fc->lock)
1158 {
1159 int err;
1160 unsigned max_forgets;
1161 unsigned count;
1162 struct fuse_forget_link *head;
1163 struct fuse_batch_forget_in arg = { .count = 0 };
1164 struct fuse_in_header ih = {
1165 .opcode = FUSE_BATCH_FORGET,
1166 .unique = fuse_get_unique(fc),
1167 .len = sizeof(ih) + sizeof(arg),
1168 };
1169
1170 if (nbytes < ih.len) {
1171 spin_unlock(&fc->lock);
1172 return -EINVAL;
1173 }
1174
1175 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1176 head = dequeue_forget(fc, max_forgets, &count);
1177 spin_unlock(&fc->lock);
1178
1179 arg.count = count;
1180 ih.len += count * sizeof(struct fuse_forget_one);
1181 err = fuse_copy_one(cs, &ih, sizeof(ih));
1182 if (!err)
1183 err = fuse_copy_one(cs, &arg, sizeof(arg));
1184
1185 while (head) {
1186 struct fuse_forget_link *forget = head;
1187
1188 if (!err) {
1189 err = fuse_copy_one(cs, &forget->forget_one,
1190 sizeof(forget->forget_one));
1191 }
1192 head = forget->next;
1193 kfree(forget);
1194 }
1195
1196 fuse_copy_finish(cs);
1197
1198 if (err)
1199 return err;
1200
1201 return ih.len;
1202 }
1203
1204 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1205 size_t nbytes)
1206 __releases(fc->lock)
1207 {
1208 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1209 return fuse_read_single_forget(fc, cs, nbytes);
1210 else
1211 return fuse_read_batch_forget(fc, cs, nbytes);
1212 }
1213
1214 /*
1215 * Read a single request into the userspace filesystem's buffer. This
1216 * function waits until a request is available, then removes it from
1217 * the pending list and copies request data to userspace buffer. If
1218 * no reply is needed (FORGET) or request has been aborted or there
1219 * was an error during the copying then it's finished by calling
1220 * request_end(). Otherwise add it to the processing list, and set
1221 * the 'sent' flag.
1222 */
1223 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1224 struct fuse_copy_state *cs, size_t nbytes)
1225 {
1226 int err;
1227 struct fuse_req *req;
1228 struct fuse_in *in;
1229 unsigned reqsize;
1230
1231 restart:
1232 spin_lock(&fc->lock);
1233 err = -EAGAIN;
1234 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1235 !request_pending(fc))
1236 goto err_unlock;
1237
1238 request_wait(fc);
1239 err = -ENODEV;
1240 if (!fc->connected)
1241 goto err_unlock;
1242 err = -ERESTARTSYS;
1243 if (!request_pending(fc))
1244 goto err_unlock;
1245
1246 if (!list_empty(&fc->interrupts)) {
1247 req = list_entry(fc->interrupts.next, struct fuse_req,
1248 intr_entry);
1249 return fuse_read_interrupt(fc, cs, nbytes, req);
1250 }
1251
1252 if (forget_pending(fc)) {
1253 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1254 return fuse_read_forget(fc, cs, nbytes);
1255
1256 if (fc->forget_batch <= -8)
1257 fc->forget_batch = 16;
1258 }
1259
1260 req = list_entry(fc->pending.next, struct fuse_req, list);
1261 req->state = FUSE_REQ_READING;
1262 list_move(&req->list, &fc->io);
1263
1264 in = &req->in;
1265 reqsize = in->h.len;
1266 /* If request is too large, reply with an error and restart the read */
1267 if (nbytes < reqsize) {
1268 req->out.h.error = -EIO;
1269 /* SETXATTR is special, since it may contain too large data */
1270 if (in->h.opcode == FUSE_SETXATTR)
1271 req->out.h.error = -E2BIG;
1272 request_end(fc, req);
1273 goto restart;
1274 }
1275 spin_unlock(&fc->lock);
1276 cs->req = req;
1277 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1278 if (!err)
1279 err = fuse_copy_args(cs, in->numargs, in->argpages,
1280 (struct fuse_arg *) in->args, 0);
1281 fuse_copy_finish(cs);
1282 spin_lock(&fc->lock);
1283 req->locked = 0;
1284 if (req->aborted) {
1285 request_end(fc, req);
1286 return -ENODEV;
1287 }
1288 if (err) {
1289 req->out.h.error = -EIO;
1290 request_end(fc, req);
1291 return err;
1292 }
1293 if (!req->isreply)
1294 request_end(fc, req);
1295 else {
1296 req->state = FUSE_REQ_SENT;
1297 list_move_tail(&req->list, &fc->processing);
1298 if (req->interrupted)
1299 queue_interrupt(fc, req);
1300 spin_unlock(&fc->lock);
1301 }
1302 return reqsize;
1303
1304 err_unlock:
1305 spin_unlock(&fc->lock);
1306 return err;
1307 }
1308
1309 static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1310 unsigned long nr_segs, loff_t pos)
1311 {
1312 struct fuse_copy_state cs;
1313 struct file *file = iocb->ki_filp;
1314 struct fuse_conn *fc = fuse_get_conn(file);
1315 if (!fc)
1316 return -EPERM;
1317
1318 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1319
1320 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1321 }
1322
1323 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1324 struct pipe_inode_info *pipe,
1325 size_t len, unsigned int flags)
1326 {
1327 int ret;
1328 int page_nr = 0;
1329 int do_wakeup = 0;
1330 struct pipe_buffer *bufs;
1331 struct fuse_copy_state cs;
1332 struct fuse_conn *fc = fuse_get_conn(in);
1333 if (!fc)
1334 return -EPERM;
1335
1336 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1337 if (!bufs)
1338 return -ENOMEM;
1339
1340 fuse_copy_init(&cs, fc, 1, NULL, 0);
1341 cs.pipebufs = bufs;
1342 cs.pipe = pipe;
1343 ret = fuse_dev_do_read(fc, in, &cs, len);
1344 if (ret < 0)
1345 goto out;
1346
1347 ret = 0;
1348 pipe_lock(pipe);
1349
1350 if (!pipe->readers) {
1351 send_sig(SIGPIPE, current, 0);
1352 if (!ret)
1353 ret = -EPIPE;
1354 goto out_unlock;
1355 }
1356
1357 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1358 ret = -EIO;
1359 goto out_unlock;
1360 }
1361
1362 while (page_nr < cs.nr_segs) {
1363 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1364 struct pipe_buffer *buf = pipe->bufs + newbuf;
1365
1366 buf->page = bufs[page_nr].page;
1367 buf->offset = bufs[page_nr].offset;
1368 buf->len = bufs[page_nr].len;
1369 /*
1370 * Need to be careful about this. Having buf->ops in module
1371 * code can Oops if the buffer persists after module unload.
1372 */
1373 buf->ops = &nosteal_pipe_buf_ops;
1374
1375 pipe->nrbufs++;
1376 page_nr++;
1377 ret += buf->len;
1378
1379 if (pipe->files)
1380 do_wakeup = 1;
1381 }
1382
1383 out_unlock:
1384 pipe_unlock(pipe);
1385
1386 if (do_wakeup) {
1387 smp_mb();
1388 if (waitqueue_active(&pipe->wait))
1389 wake_up_interruptible(&pipe->wait);
1390 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1391 }
1392
1393 out:
1394 for (; page_nr < cs.nr_segs; page_nr++)
1395 page_cache_release(bufs[page_nr].page);
1396
1397 kfree(bufs);
1398 return ret;
1399 }
1400
1401 static 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
1417 err:
1418 fuse_copy_finish(cs);
1419 return err;
1420 }
1421
1422 static 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
1445 err:
1446 fuse_copy_finish(cs);
1447 return err;
1448 }
1449
1450 static 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 name.hash = full_name_hash(name.name, name.len);
1486
1487 down_read(&fc->killsb);
1488 err = -ENOENT;
1489 if (fc->sb)
1490 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1491 up_read(&fc->killsb);
1492 kfree(buf);
1493 return err;
1494
1495 err:
1496 kfree(buf);
1497 fuse_copy_finish(cs);
1498 return err;
1499 }
1500
1501 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1502 struct fuse_copy_state *cs)
1503 {
1504 struct fuse_notify_delete_out outarg;
1505 int err = -ENOMEM;
1506 char *buf;
1507 struct qstr name;
1508
1509 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1510 if (!buf)
1511 goto err;
1512
1513 err = -EINVAL;
1514 if (size < sizeof(outarg))
1515 goto err;
1516
1517 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1518 if (err)
1519 goto err;
1520
1521 err = -ENAMETOOLONG;
1522 if (outarg.namelen > FUSE_NAME_MAX)
1523 goto err;
1524
1525 err = -EINVAL;
1526 if (size != sizeof(outarg) + outarg.namelen + 1)
1527 goto err;
1528
1529 name.name = buf;
1530 name.len = outarg.namelen;
1531 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1532 if (err)
1533 goto err;
1534 fuse_copy_finish(cs);
1535 buf[outarg.namelen] = 0;
1536 name.hash = full_name_hash(name.name, name.len);
1537
1538 down_read(&fc->killsb);
1539 err = -ENOENT;
1540 if (fc->sb)
1541 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1542 outarg.child, &name);
1543 up_read(&fc->killsb);
1544 kfree(buf);
1545 return err;
1546
1547 err:
1548 kfree(buf);
1549 fuse_copy_finish(cs);
1550 return err;
1551 }
1552
1553 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1554 struct fuse_copy_state *cs)
1555 {
1556 struct fuse_notify_store_out outarg;
1557 struct inode *inode;
1558 struct address_space *mapping;
1559 u64 nodeid;
1560 int err;
1561 pgoff_t index;
1562 unsigned int offset;
1563 unsigned int num;
1564 loff_t file_size;
1565 loff_t end;
1566
1567 err = -EINVAL;
1568 if (size < sizeof(outarg))
1569 goto out_finish;
1570
1571 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1572 if (err)
1573 goto out_finish;
1574
1575 err = -EINVAL;
1576 if (size - sizeof(outarg) != outarg.size)
1577 goto out_finish;
1578
1579 nodeid = outarg.nodeid;
1580
1581 down_read(&fc->killsb);
1582
1583 err = -ENOENT;
1584 if (!fc->sb)
1585 goto out_up_killsb;
1586
1587 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1588 if (!inode)
1589 goto out_up_killsb;
1590
1591 mapping = inode->i_mapping;
1592 index = outarg.offset >> PAGE_CACHE_SHIFT;
1593 offset = outarg.offset & ~PAGE_CACHE_MASK;
1594 file_size = i_size_read(inode);
1595 end = outarg.offset + outarg.size;
1596 if (end > file_size) {
1597 file_size = end;
1598 fuse_write_update_size(inode, file_size);
1599 }
1600
1601 num = outarg.size;
1602 while (num) {
1603 struct page *page;
1604 unsigned int this_num;
1605
1606 err = -ENOMEM;
1607 page = find_or_create_page(mapping, index,
1608 mapping_gfp_mask(mapping));
1609 if (!page)
1610 goto out_iput;
1611
1612 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1613 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1614 if (!err && offset == 0 &&
1615 (this_num == PAGE_CACHE_SIZE || file_size == end))
1616 SetPageUptodate(page);
1617 unlock_page(page);
1618 page_cache_release(page);
1619
1620 if (err)
1621 goto out_iput;
1622
1623 num -= this_num;
1624 offset = 0;
1625 index++;
1626 }
1627
1628 err = 0;
1629
1630 out_iput:
1631 iput(inode);
1632 out_up_killsb:
1633 up_read(&fc->killsb);
1634 out_finish:
1635 fuse_copy_finish(cs);
1636 return err;
1637 }
1638
1639 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1640 {
1641 release_pages(req->pages, req->num_pages, false);
1642 }
1643
1644 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1645 struct fuse_notify_retrieve_out *outarg)
1646 {
1647 int err;
1648 struct address_space *mapping = inode->i_mapping;
1649 struct fuse_req *req;
1650 pgoff_t index;
1651 loff_t file_size;
1652 unsigned int num;
1653 unsigned int offset;
1654 size_t total_len = 0;
1655 int num_pages;
1656
1657 offset = outarg->offset & ~PAGE_CACHE_MASK;
1658 file_size = i_size_read(inode);
1659
1660 num = outarg->size;
1661 if (outarg->offset > file_size)
1662 num = 0;
1663 else if (outarg->offset + num > file_size)
1664 num = file_size - outarg->offset;
1665
1666 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1667 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1668
1669 req = fuse_get_req(fc, num_pages);
1670 if (IS_ERR(req))
1671 return PTR_ERR(req);
1672
1673 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1674 req->in.h.nodeid = outarg->nodeid;
1675 req->in.numargs = 2;
1676 req->in.argpages = 1;
1677 req->page_descs[0].offset = offset;
1678 req->end = fuse_retrieve_end;
1679
1680 index = outarg->offset >> PAGE_CACHE_SHIFT;
1681
1682 while (num && req->num_pages < num_pages) {
1683 struct page *page;
1684 unsigned int this_num;
1685
1686 page = find_get_page(mapping, index);
1687 if (!page)
1688 break;
1689
1690 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1691 req->pages[req->num_pages] = page;
1692 req->page_descs[req->num_pages].length = this_num;
1693 req->num_pages++;
1694
1695 offset = 0;
1696 num -= this_num;
1697 total_len += this_num;
1698 index++;
1699 }
1700 req->misc.retrieve_in.offset = outarg->offset;
1701 req->misc.retrieve_in.size = total_len;
1702 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1703 req->in.args[0].value = &req->misc.retrieve_in;
1704 req->in.args[1].size = total_len;
1705
1706 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1707 if (err)
1708 fuse_retrieve_end(fc, req);
1709
1710 return err;
1711 }
1712
1713 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1714 struct fuse_copy_state *cs)
1715 {
1716 struct fuse_notify_retrieve_out outarg;
1717 struct inode *inode;
1718 int err;
1719
1720 err = -EINVAL;
1721 if (size != sizeof(outarg))
1722 goto copy_finish;
1723
1724 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1725 if (err)
1726 goto copy_finish;
1727
1728 fuse_copy_finish(cs);
1729
1730 down_read(&fc->killsb);
1731 err = -ENOENT;
1732 if (fc->sb) {
1733 u64 nodeid = outarg.nodeid;
1734
1735 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1736 if (inode) {
1737 err = fuse_retrieve(fc, inode, &outarg);
1738 iput(inode);
1739 }
1740 }
1741 up_read(&fc->killsb);
1742
1743 return err;
1744
1745 copy_finish:
1746 fuse_copy_finish(cs);
1747 return err;
1748 }
1749
1750 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1751 unsigned int size, struct fuse_copy_state *cs)
1752 {
1753 switch (code) {
1754 case FUSE_NOTIFY_POLL:
1755 return fuse_notify_poll(fc, size, cs);
1756
1757 case FUSE_NOTIFY_INVAL_INODE:
1758 return fuse_notify_inval_inode(fc, size, cs);
1759
1760 case FUSE_NOTIFY_INVAL_ENTRY:
1761 return fuse_notify_inval_entry(fc, size, cs);
1762
1763 case FUSE_NOTIFY_STORE:
1764 return fuse_notify_store(fc, size, cs);
1765
1766 case FUSE_NOTIFY_RETRIEVE:
1767 return fuse_notify_retrieve(fc, size, cs);
1768
1769 case FUSE_NOTIFY_DELETE:
1770 return fuse_notify_delete(fc, size, cs);
1771
1772 default:
1773 fuse_copy_finish(cs);
1774 return -EINVAL;
1775 }
1776 }
1777
1778 /* Look up request on processing list by unique ID */
1779 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1780 {
1781 struct fuse_req *req;
1782
1783 list_for_each_entry(req, &fc->processing, list) {
1784 if (req->in.h.unique == unique || req->intr_unique == unique)
1785 return req;
1786 }
1787 return NULL;
1788 }
1789
1790 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1791 unsigned nbytes)
1792 {
1793 unsigned reqsize = sizeof(struct fuse_out_header);
1794
1795 if (out->h.error)
1796 return nbytes != reqsize ? -EINVAL : 0;
1797
1798 reqsize += len_args(out->numargs, out->args);
1799
1800 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1801 return -EINVAL;
1802 else if (reqsize > nbytes) {
1803 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1804 unsigned diffsize = reqsize - nbytes;
1805 if (diffsize > lastarg->size)
1806 return -EINVAL;
1807 lastarg->size -= diffsize;
1808 }
1809 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1810 out->page_zeroing);
1811 }
1812
1813 /*
1814 * Write a single reply to a request. First the header is copied from
1815 * the write buffer. The request is then searched on the processing
1816 * list by the unique ID found in the header. If found, then remove
1817 * it from the list and copy the rest of the buffer to the request.
1818 * The request is finished by calling request_end()
1819 */
1820 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1821 struct fuse_copy_state *cs, size_t nbytes)
1822 {
1823 int err;
1824 struct fuse_req *req;
1825 struct fuse_out_header oh;
1826
1827 if (nbytes < sizeof(struct fuse_out_header))
1828 return -EINVAL;
1829
1830 err = fuse_copy_one(cs, &oh, sizeof(oh));
1831 if (err)
1832 goto err_finish;
1833
1834 err = -EINVAL;
1835 if (oh.len != nbytes)
1836 goto err_finish;
1837
1838 /*
1839 * Zero oh.unique indicates unsolicited notification message
1840 * and error contains notification code.
1841 */
1842 if (!oh.unique) {
1843 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1844 return err ? err : nbytes;
1845 }
1846
1847 err = -EINVAL;
1848 if (oh.error <= -1000 || oh.error > 0)
1849 goto err_finish;
1850
1851 spin_lock(&fc->lock);
1852 err = -ENOENT;
1853 if (!fc->connected)
1854 goto err_unlock;
1855
1856 req = request_find(fc, oh.unique);
1857 if (!req)
1858 goto err_unlock;
1859
1860 if (req->aborted) {
1861 spin_unlock(&fc->lock);
1862 fuse_copy_finish(cs);
1863 spin_lock(&fc->lock);
1864 request_end(fc, req);
1865 return -ENOENT;
1866 }
1867 /* Is it an interrupt reply? */
1868 if (req->intr_unique == oh.unique) {
1869 err = -EINVAL;
1870 if (nbytes != sizeof(struct fuse_out_header))
1871 goto err_unlock;
1872
1873 if (oh.error == -ENOSYS)
1874 fc->no_interrupt = 1;
1875 else if (oh.error == -EAGAIN)
1876 queue_interrupt(fc, req);
1877
1878 spin_unlock(&fc->lock);
1879 fuse_copy_finish(cs);
1880 return nbytes;
1881 }
1882
1883 req->state = FUSE_REQ_WRITING;
1884 list_move(&req->list, &fc->io);
1885 req->out.h = oh;
1886 req->locked = 1;
1887 cs->req = req;
1888 if (!req->out.page_replace)
1889 cs->move_pages = 0;
1890 spin_unlock(&fc->lock);
1891
1892 err = copy_out_args(cs, &req->out, nbytes);
1893 fuse_copy_finish(cs);
1894
1895 spin_lock(&fc->lock);
1896 req->locked = 0;
1897 if (!err) {
1898 if (req->aborted)
1899 err = -ENOENT;
1900 } else if (!req->aborted)
1901 req->out.h.error = -EIO;
1902 request_end(fc, req);
1903
1904 return err ? err : nbytes;
1905
1906 err_unlock:
1907 spin_unlock(&fc->lock);
1908 err_finish:
1909 fuse_copy_finish(cs);
1910 return err;
1911 }
1912
1913 static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1914 unsigned long nr_segs, loff_t pos)
1915 {
1916 struct fuse_copy_state cs;
1917 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1918 if (!fc)
1919 return -EPERM;
1920
1921 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1922
1923 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1924 }
1925
1926 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1927 struct file *out, loff_t *ppos,
1928 size_t len, unsigned int flags)
1929 {
1930 unsigned nbuf;
1931 unsigned idx;
1932 struct pipe_buffer *bufs;
1933 struct fuse_copy_state cs;
1934 struct fuse_conn *fc;
1935 size_t rem;
1936 ssize_t ret;
1937
1938 fc = fuse_get_conn(out);
1939 if (!fc)
1940 return -EPERM;
1941
1942 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1943 if (!bufs)
1944 return -ENOMEM;
1945
1946 pipe_lock(pipe);
1947 nbuf = 0;
1948 rem = 0;
1949 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1950 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1951
1952 ret = -EINVAL;
1953 if (rem < len) {
1954 pipe_unlock(pipe);
1955 goto out;
1956 }
1957
1958 rem = len;
1959 while (rem) {
1960 struct pipe_buffer *ibuf;
1961 struct pipe_buffer *obuf;
1962
1963 BUG_ON(nbuf >= pipe->buffers);
1964 BUG_ON(!pipe->nrbufs);
1965 ibuf = &pipe->bufs[pipe->curbuf];
1966 obuf = &bufs[nbuf];
1967
1968 if (rem >= ibuf->len) {
1969 *obuf = *ibuf;
1970 ibuf->ops = NULL;
1971 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1972 pipe->nrbufs--;
1973 } else {
1974 ibuf->ops->get(pipe, ibuf);
1975 *obuf = *ibuf;
1976 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1977 obuf->len = rem;
1978 ibuf->offset += obuf->len;
1979 ibuf->len -= obuf->len;
1980 }
1981 nbuf++;
1982 rem -= obuf->len;
1983 }
1984 pipe_unlock(pipe);
1985
1986 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1987 cs.pipebufs = bufs;
1988 cs.pipe = pipe;
1989
1990 if (flags & SPLICE_F_MOVE)
1991 cs.move_pages = 1;
1992
1993 ret = fuse_dev_do_write(fc, &cs, len);
1994
1995 for (idx = 0; idx < nbuf; idx++) {
1996 struct pipe_buffer *buf = &bufs[idx];
1997 buf->ops->release(pipe, buf);
1998 }
1999 out:
2000 kfree(bufs);
2001 return ret;
2002 }
2003
2004 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2005 {
2006 unsigned mask = POLLOUT | POLLWRNORM;
2007 struct fuse_conn *fc = fuse_get_conn(file);
2008 if (!fc)
2009 return POLLERR;
2010
2011 poll_wait(file, &fc->waitq, wait);
2012
2013 spin_lock(&fc->lock);
2014 if (!fc->connected)
2015 mask = POLLERR;
2016 else if (request_pending(fc))
2017 mask |= POLLIN | POLLRDNORM;
2018 spin_unlock(&fc->lock);
2019
2020 return mask;
2021 }
2022
2023 /*
2024 * Abort all requests on the given list (pending or processing)
2025 *
2026 * This function releases and reacquires fc->lock
2027 */
2028 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2029 __releases(fc->lock)
2030 __acquires(fc->lock)
2031 {
2032 while (!list_empty(head)) {
2033 struct fuse_req *req;
2034 req = list_entry(head->next, struct fuse_req, list);
2035 req->out.h.error = -ECONNABORTED;
2036 request_end(fc, req);
2037 spin_lock(&fc->lock);
2038 }
2039 }
2040
2041 /*
2042 * Abort requests under I/O
2043 *
2044 * The requests are set to aborted and finished, and the request
2045 * waiter is woken up. This will make request_wait_answer() wait
2046 * until the request is unlocked and then return.
2047 *
2048 * If the request is asynchronous, then the end function needs to be
2049 * called after waiting for the request to be unlocked (if it was
2050 * locked).
2051 */
2052 static void end_io_requests(struct fuse_conn *fc)
2053 __releases(fc->lock)
2054 __acquires(fc->lock)
2055 {
2056 while (!list_empty(&fc->io)) {
2057 struct fuse_req *req =
2058 list_entry(fc->io.next, struct fuse_req, list);
2059 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
2060
2061 req->aborted = 1;
2062 req->out.h.error = -ECONNABORTED;
2063 req->state = FUSE_REQ_FINISHED;
2064 list_del_init(&req->list);
2065 wake_up(&req->waitq);
2066 if (end) {
2067 req->end = NULL;
2068 __fuse_get_request(req);
2069 spin_unlock(&fc->lock);
2070 wait_event(req->waitq, !req->locked);
2071 end(fc, req);
2072 fuse_put_request(fc, req);
2073 spin_lock(&fc->lock);
2074 }
2075 }
2076 }
2077
2078 static void end_queued_requests(struct fuse_conn *fc)
2079 __releases(fc->lock)
2080 __acquires(fc->lock)
2081 {
2082 fc->max_background = UINT_MAX;
2083 flush_bg_queue(fc);
2084 end_requests(fc, &fc->pending);
2085 end_requests(fc, &fc->processing);
2086 while (forget_pending(fc))
2087 kfree(dequeue_forget(fc, 1, NULL));
2088 }
2089
2090 static void end_polls(struct fuse_conn *fc)
2091 {
2092 struct rb_node *p;
2093
2094 p = rb_first(&fc->polled_files);
2095
2096 while (p) {
2097 struct fuse_file *ff;
2098 ff = rb_entry(p, struct fuse_file, polled_node);
2099 wake_up_interruptible_all(&ff->poll_wait);
2100
2101 p = rb_next(p);
2102 }
2103 }
2104
2105 /*
2106 * Abort all requests.
2107 *
2108 * Emergency exit in case of a malicious or accidental deadlock, or
2109 * just a hung filesystem.
2110 *
2111 * The same effect is usually achievable through killing the
2112 * filesystem daemon and all users of the filesystem. The exception
2113 * is the combination of an asynchronous request and the tricky
2114 * deadlock (see Documentation/filesystems/fuse.txt).
2115 *
2116 * During the aborting, progression of requests from the pending and
2117 * processing lists onto the io list, and progression of new requests
2118 * onto the pending list is prevented by req->connected being false.
2119 *
2120 * Progression of requests under I/O to the processing list is
2121 * prevented by the req->aborted flag being true for these requests.
2122 * For this reason requests on the io list must be aborted first.
2123 */
2124 void fuse_abort_conn(struct fuse_conn *fc)
2125 {
2126 spin_lock(&fc->lock);
2127 if (fc->connected) {
2128 fc->connected = 0;
2129 fc->blocked = 0;
2130 fc->initialized = 1;
2131 end_io_requests(fc);
2132 end_queued_requests(fc);
2133 end_polls(fc);
2134 wake_up_all(&fc->waitq);
2135 wake_up_all(&fc->blocked_waitq);
2136 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2137 }
2138 spin_unlock(&fc->lock);
2139 }
2140 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2141
2142 int fuse_dev_release(struct inode *inode, struct file *file)
2143 {
2144 struct fuse_conn *fc = fuse_get_conn(file);
2145 if (fc) {
2146 spin_lock(&fc->lock);
2147 fc->connected = 0;
2148 fc->blocked = 0;
2149 fc->initialized = 1;
2150 end_queued_requests(fc);
2151 end_polls(fc);
2152 wake_up_all(&fc->blocked_waitq);
2153 spin_unlock(&fc->lock);
2154 fuse_conn_put(fc);
2155 }
2156
2157 return 0;
2158 }
2159 EXPORT_SYMBOL_GPL(fuse_dev_release);
2160
2161 static int fuse_dev_fasync(int fd, struct file *file, int on)
2162 {
2163 struct fuse_conn *fc = fuse_get_conn(file);
2164 if (!fc)
2165 return -EPERM;
2166
2167 /* No locking - fasync_helper does its own locking */
2168 return fasync_helper(fd, file, on, &fc->fasync);
2169 }
2170
2171 const struct file_operations fuse_dev_operations = {
2172 .owner = THIS_MODULE,
2173 .llseek = no_llseek,
2174 .read = do_sync_read,
2175 .aio_read = fuse_dev_read,
2176 .splice_read = fuse_dev_splice_read,
2177 .write = do_sync_write,
2178 .aio_write = fuse_dev_write,
2179 .splice_write = fuse_dev_splice_write,
2180 .poll = fuse_dev_poll,
2181 .release = fuse_dev_release,
2182 .fasync = fuse_dev_fasync,
2183 };
2184 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2185
2186 static struct miscdevice fuse_miscdevice = {
2187 .minor = FUSE_MINOR,
2188 .name = "fuse",
2189 .fops = &fuse_dev_operations,
2190 };
2191
2192 int __init fuse_dev_init(void)
2193 {
2194 int err = -ENOMEM;
2195 fuse_req_cachep = kmem_cache_create("fuse_request",
2196 sizeof(struct fuse_req),
2197 0, 0, NULL);
2198 if (!fuse_req_cachep)
2199 goto out;
2200
2201 err = misc_register(&fuse_miscdevice);
2202 if (err)
2203 goto out_cache_clean;
2204
2205 return 0;
2206
2207 out_cache_clean:
2208 kmem_cache_destroy(fuse_req_cachep);
2209 out:
2210 return err;
2211 }
2212
2213 void fuse_dev_cleanup(void)
2214 {
2215 misc_deregister(&fuse_miscdevice);
2216 kmem_cache_destroy(fuse_req_cachep);
2217 }
Linux kernel - Deliverables
This page took 0.081328 seconds and 5 git commands to generate.