projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[GitHub/mt8127/android_kernel_alcatel_ttab.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
23 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24 MODULE_ALIAS("devname:fuse");
25
26 static struct kmem_cache *fuse_req_cachep;
27
28 static struct fuse_conn *fuse_get_conn(struct file *file)
29 {
30 /*
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
33 */
34 return file->private_data;
35 }
36
37 static void fuse_request_init(struct fuse_req *req, struct page **pages,
38 struct fuse_page_desc *page_descs,
39 unsigned npages)
40 {
41 memset(req, 0, sizeof(*req));
42 memset(pages, 0, sizeof(*pages) * npages);
43 memset(page_descs, 0, sizeof(*page_descs) * npages);
44 INIT_LIST_HEAD(&req->list);
45 INIT_LIST_HEAD(&req->intr_entry);
46 init_waitqueue_head(&req->waitq);
47 atomic_set(&req->count, 1);
48 req->pages = pages;
49 req->page_descs = page_descs;
50 req->max_pages = npages;
51 }
52
53 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
54 {
55 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
56 if (req) {
57 struct page **pages;
58 struct fuse_page_desc *page_descs;
59
60 if (npages <= FUSE_REQ_INLINE_PAGES) {
61 pages = req->inline_pages;
62 page_descs = req->inline_page_descs;
63 } else {
64 pages = kmalloc(sizeof(struct page *) * npages, flags);
65 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
66 npages, flags);
67 }
68
69 if (!pages || !page_descs) {
70 kfree(pages);
71 kfree(page_descs);
72 kmem_cache_free(fuse_req_cachep, req);
73 return NULL;
74 }
75
76 fuse_request_init(req, pages, page_descs, npages);
77 }
78 return req;
79 }
80
81 struct fuse_req *fuse_request_alloc(unsigned npages)
82 {
83 return __fuse_request_alloc(npages, GFP_KERNEL);
84 }
85 EXPORT_SYMBOL_GPL(fuse_request_alloc);
86
87 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
88 {
89 return __fuse_request_alloc(npages, GFP_NOFS);
90 }
91
92 void fuse_request_free(struct fuse_req *req)
93 {
94 if (req->pages != req->inline_pages) {
95 kfree(req->pages);
96 kfree(req->page_descs);
97 }
98 kmem_cache_free(fuse_req_cachep, req);
99 }
100
101 static void block_sigs(sigset_t *oldset)
102 {
103 sigset_t mask;
104
105 siginitsetinv(&mask, sigmask(SIGKILL));
106 sigprocmask(SIG_BLOCK, &mask, oldset);
107 }
108
109 static void restore_sigs(sigset_t *oldset)
110 {
111 sigprocmask(SIG_SETMASK, oldset, NULL);
112 }
113
114 void __fuse_get_request(struct fuse_req *req)
115 {
116 atomic_inc(&req->count);
117 }
118
119 /* Must be called with > 1 refcount */
120 static void __fuse_put_request(struct fuse_req *req)
121 {
122 BUG_ON(atomic_read(&req->count) < 2);
123 atomic_dec(&req->count);
124 }
125
126 static void fuse_req_init_context(struct fuse_req *req)
127 {
128 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
129 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
130 req->in.h.pid = current->pid;
131 }
132
133 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
134 {
135 return !fc->initialized || (for_background && fc->blocked);
136 }
137
138 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
139 bool for_background)
140 {
141 struct fuse_req *req;
142 int err;
143 atomic_inc(&fc->num_waiting);
144
145 if (fuse_block_alloc(fc, for_background)) {
146 sigset_t oldset;
147 int intr;
148
149 block_sigs(&oldset);
150 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
151 !fuse_block_alloc(fc, for_background));
152 restore_sigs(&oldset);
153 err = -EINTR;
154 if (intr)
155 goto out;
156 }
157
158 err = -ENOTCONN;
159 if (!fc->connected)
160 goto out;
161
162 req = fuse_request_alloc(npages);
163 err = -ENOMEM;
164 if (!req) {
165 if (for_background)
166 wake_up(&fc->blocked_waitq);
167 goto out;
168 }
169
170 fuse_req_init_context(req);
171 req->waiting = 1;
172 req->background = for_background;
173 return req;
174
175 out:
176 atomic_dec(&fc->num_waiting);
177 return ERR_PTR(err);
178 }
179
180 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
181 {
182 return __fuse_get_req(fc, npages, false);
183 }
184 EXPORT_SYMBOL_GPL(fuse_get_req);
185
186 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
187 unsigned npages)
188 {
189 return __fuse_get_req(fc, npages, true);
190 }
191 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
192
193 /*
194 * Return request in fuse_file->reserved_req. However that may
195 * currently be in use. If that is the case, wait for it to become
196 * available.
197 */
198 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
199 struct file *file)
200 {
201 struct fuse_req *req = NULL;
202 struct fuse_file *ff = file->private_data;
203
204 do {
205 wait_event(fc->reserved_req_waitq, ff->reserved_req);
206 spin_lock(&fc->lock);
207 if (ff->reserved_req) {
208 req = ff->reserved_req;
209 ff->reserved_req = NULL;
210 req->stolen_file = get_file(file);
211 }
212 spin_unlock(&fc->lock);
213 } while (!req);
214
215 return req;
216 }
217
218 /*
219 * Put stolen request back into fuse_file->reserved_req
220 */
221 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
222 {
223 struct file *file = req->stolen_file;
224 struct fuse_file *ff = file->private_data;
225
226 spin_lock(&fc->lock);
227 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
228 BUG_ON(ff->reserved_req);
229 ff->reserved_req = req;
230 wake_up_all(&fc->reserved_req_waitq);
231 spin_unlock(&fc->lock);
232 fput(file);
233 }
234
235 /*
236 * Gets a requests for a file operation, always succeeds
237 *
238 * This is used for sending the FLUSH request, which must get to
239 * userspace, due to POSIX locks which may need to be unlocked.
240 *
241 * If allocation fails due to OOM, use the reserved request in
242 * fuse_file.
243 *
244 * This is very unlikely to deadlock accidentally, since the
245 * filesystem should not have it's own file open. If deadlock is
246 * intentional, it can still be broken by "aborting" the filesystem.
247 */
248 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
249 struct file *file)
250 {
251 struct fuse_req *req;
252
253 atomic_inc(&fc->num_waiting);
254 wait_event(fc->blocked_waitq, fc->initialized);
255 req = fuse_request_alloc(0);
256 if (!req)
257 req = get_reserved_req(fc, file);
258
259 fuse_req_init_context(req);
260 req->waiting = 1;
261 req->background = 0;
262 return req;
263 }
264
265 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
266 {
267 if (atomic_dec_and_test(&req->count)) {
268 if (unlikely(req->background)) {
269 /*
270 * We get here in the unlikely case that a background
271 * request was allocated but not sent
272 */
273 spin_lock(&fc->lock);
274 if (!fc->blocked)
275 wake_up(&fc->blocked_waitq);
276 spin_unlock(&fc->lock);
277 }
278
279 if (req->waiting)
280 atomic_dec(&fc->num_waiting);
281
282 if (req->stolen_file)
283 put_reserved_req(fc, req);
284 else
285 fuse_request_free(req);
286 }
287 }
288 EXPORT_SYMBOL_GPL(fuse_put_request);
289
290 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
291 {
292 unsigned nbytes = 0;
293 unsigned i;
294
295 for (i = 0; i < numargs; i++)
296 nbytes += args[i].size;
297
298 return nbytes;
299 }
300
301 static u64 fuse_get_unique(struct fuse_conn *fc)
302 {
303 fc->reqctr++;
304 /* zero is special */
305 if (fc->reqctr == 0)
306 fc->reqctr = 1;
307
308 return fc->reqctr;
309 }
310
311 static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
312 {
313 req->in.h.len = sizeof(struct fuse_in_header) +
314 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
315 list_add_tail(&req->list, &fc->pending);
316 req->state = FUSE_REQ_PENDING;
317 if (!req->waiting) {
318 req->waiting = 1;
319 atomic_inc(&fc->num_waiting);
320 }
321 wake_up(&fc->waitq);
322 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
323 }
324
325 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
326 u64 nodeid, u64 nlookup)
327 {
328 forget->forget_one.nodeid = nodeid;
329 forget->forget_one.nlookup = nlookup;
330
331 spin_lock(&fc->lock);
332 if (fc->connected) {
333 fc->forget_list_tail->next = forget;
334 fc->forget_list_tail = forget;
335 wake_up(&fc->waitq);
336 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
337 } else {
338 kfree(forget);
339 }
340 spin_unlock(&fc->lock);
341 }
342
343 static void flush_bg_queue(struct fuse_conn *fc)
344 {
345 while (fc->active_background < fc->max_background &&
346 !list_empty(&fc->bg_queue)) {
347 struct fuse_req *req;
348
349 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
350 list_del(&req->list);
351 fc->active_background++;
352 req->in.h.unique = fuse_get_unique(fc);
353 queue_request(fc, req);
354 }
355 }
356
357 /*
358 * This function is called when a request is finished. Either a reply
359 * has arrived or it was aborted (and not yet sent) or some error
360 * occurred during communication with userspace, or the device file
361 * was closed. The requester thread is woken up (if still waiting),
362 * the 'end' callback is called if given, else the reference to the
363 * request is released
364 *
365 * Called with fc->lock, unlocks it
366 */
367 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
368 __releases(fc->lock)
369 {
370 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
371 req->end = NULL;
372 list_del(&req->list);
373 list_del(&req->intr_entry);
374 req->state = FUSE_REQ_FINISHED;
375 if (req->background) {
376 req->background = 0;
377
378 if (fc->num_background == fc->max_background)
379 fc->blocked = 0;
380
381 /* Wake up next waiter, if any */
382 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
383 wake_up(&fc->blocked_waitq);
384
385 if (fc->num_background == fc->congestion_threshold &&
386 fc->connected && fc->bdi_initialized) {
387 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
388 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
389 }
390 fc->num_background--;
391 fc->active_background--;
392 flush_bg_queue(fc);
393 }
394 spin_unlock(&fc->lock);
395 wake_up(&req->waitq);
396 if (end)
397 end(fc, req);
398 fuse_put_request(fc, req);
399 }
400
401 static void wait_answer_interruptible(struct fuse_conn *fc,
402 struct fuse_req *req)
403 __releases(fc->lock)
404 __acquires(fc->lock)
405 {
406 if (signal_pending(current))
407 return;
408
409 spin_unlock(&fc->lock);
410 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
411 spin_lock(&fc->lock);
412 }
413
414 static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
415 {
416 list_add_tail(&req->intr_entry, &fc->interrupts);
417 wake_up(&fc->waitq);
418 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
419 }
420
421 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
422 __releases(fc->lock)
423 __acquires(fc->lock)
424 {
425 if (!fc->no_interrupt) {
426 /* Any signal may interrupt this */
427 wait_answer_interruptible(fc, req);
428
429 if (req->aborted)
430 goto aborted;
431 if (req->state == FUSE_REQ_FINISHED)
432 return;
433
434 req->interrupted = 1;
435 if (req->state == FUSE_REQ_SENT)
436 queue_interrupt(fc, req);
437 }
438
439 if (!req->force) {
440 sigset_t oldset;
441
442 /* Only fatal signals may interrupt this */
443 block_sigs(&oldset);
444 wait_answer_interruptible(fc, req);
445 restore_sigs(&oldset);
446
447 if (req->aborted)
448 goto aborted;
449 if (req->state == FUSE_REQ_FINISHED)
450 return;
451
452 /* Request is not yet in userspace, bail out */
453 if (req->state == FUSE_REQ_PENDING) {
454 list_del(&req->list);
455 __fuse_put_request(req);
456 req->out.h.error = -EINTR;
457 return;
458 }
459 }
460
461 /*
462 * Either request is already in userspace, or it was forced.
463 * Wait it out.
464 */
465 spin_unlock(&fc->lock);
466 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
467 spin_lock(&fc->lock);
468
469 if (!req->aborted)
470 return;
471
472 aborted:
473 BUG_ON(req->state != FUSE_REQ_FINISHED);
474 if (req->locked) {
475 /* This is uninterruptible sleep, because data is
476 being copied to/from the buffers of req. During
477 locked state, there mustn't be any filesystem
478 operation (e.g. page fault), since that could lead
479 to deadlock */
480 spin_unlock(&fc->lock);
481 wait_event(req->waitq, !req->locked);
482 spin_lock(&fc->lock);
483 }
484 }
485
486 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
487 {
488 BUG_ON(req->background);
489 spin_lock(&fc->lock);
490 if (!fc->connected)
491 req->out.h.error = -ENOTCONN;
492 else if (fc->conn_error)
493 req->out.h.error = -ECONNREFUSED;
494 else {
495 req->in.h.unique = fuse_get_unique(fc);
496 queue_request(fc, req);
497 /* acquire extra reference, since request is still needed
498 after request_end() */
499 __fuse_get_request(req);
500
501 request_wait_answer(fc, req);
502 }
503 spin_unlock(&fc->lock);
504 }
505
506 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
507 {
508 req->isreply = 1;
509 __fuse_request_send(fc, req);
510 }
511 EXPORT_SYMBOL_GPL(fuse_request_send);
512
513 static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
514 struct fuse_req *req)
515 {
516 BUG_ON(!req->background);
517 fc->num_background++;
518 if (fc->num_background == fc->max_background)
519 fc->blocked = 1;
520 if (fc->num_background == fc->congestion_threshold &&
521 fc->bdi_initialized) {
522 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
523 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
524 }
525 list_add_tail(&req->list, &fc->bg_queue);
526 flush_bg_queue(fc);
527 }
528
529 static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
530 {
531 spin_lock(&fc->lock);
532 if (fc->connected) {
533 fuse_request_send_nowait_locked(fc, req);
534 spin_unlock(&fc->lock);
535 } else {
536 req->out.h.error = -ENOTCONN;
537 request_end(fc, req);
538 }
539 }
540
541 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
542 {
543 req->isreply = 1;
544 fuse_request_send_nowait(fc, req);
545 }
546 EXPORT_SYMBOL_GPL(fuse_request_send_background);
547
548 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
549 struct fuse_req *req, u64 unique)
550 {
551 int err = -ENODEV;
552
553 req->isreply = 0;
554 req->in.h.unique = unique;
555 spin_lock(&fc->lock);
556 if (fc->connected) {
557 queue_request(fc, req);
558 err = 0;
559 }
560 spin_unlock(&fc->lock);
561
562 return err;
563 }
564
565 /*
566 * Called under fc->lock
567 *
568 * fc->connected must have been checked previously
569 */
570 void fuse_request_send_background_locked(struct fuse_conn *fc,
571 struct fuse_req *req)
572 {
573 req->isreply = 1;
574 fuse_request_send_nowait_locked(fc, req);
575 }
576
577 void fuse_force_forget(struct file *file, u64 nodeid)
578 {
579 struct inode *inode = file_inode(file);
580 struct fuse_conn *fc = get_fuse_conn(inode);
581 struct fuse_req *req;
582 struct fuse_forget_in inarg;
583
584 memset(&inarg, 0, sizeof(inarg));
585 inarg.nlookup = 1;
586 req = fuse_get_req_nofail_nopages(fc, file);
587 req->in.h.opcode = FUSE_FORGET;
588 req->in.h.nodeid = nodeid;
589 req->in.numargs = 1;
590 req->in.args[0].size = sizeof(inarg);
591 req->in.args[0].value = &inarg;
592 req->isreply = 0;
593 __fuse_request_send(fc, req);
594 /* ignore errors */
595 fuse_put_request(fc, req);
596 }
597
598 /*
599 * Lock the request. Up to the next unlock_request() there mustn't be
600 * anything that could cause a page-fault. If the request was already
601 * aborted bail out.
602 */
603 static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
604 {
605 int err = 0;
606 if (req) {
607 spin_lock(&fc->lock);
608 if (req->aborted)
609 err = -ENOENT;
610 else
611 req->locked = 1;
612 spin_unlock(&fc->lock);
613 }
614 return err;
615 }
616
617 /*
618 * Unlock request. If it was aborted during being locked, the
619 * requester thread is currently waiting for it to be unlocked, so
620 * wake it up.
621 */
622 static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
623 {
624 if (req) {
625 spin_lock(&fc->lock);
626 req->locked = 0;
627 if (req->aborted)
628 wake_up(&req->waitq);
629 spin_unlock(&fc->lock);
630 }
631 }
632
633 struct fuse_copy_state {
634 struct fuse_conn *fc;
635 int write;
636 struct fuse_req *req;
637 const struct iovec *iov;
638 struct pipe_buffer *pipebufs;
639 struct pipe_buffer *currbuf;
640 struct pipe_inode_info *pipe;
641 unsigned long nr_segs;
642 unsigned long seglen;
643 unsigned long addr;
644 struct page *pg;
645 void *mapaddr;
646 void *buf;
647 unsigned len;
648 unsigned move_pages:1;
649 };
650
651 static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
652 int write,
653 const struct iovec *iov, unsigned long nr_segs)
654 {
655 memset(cs, 0, sizeof(*cs));
656 cs->fc = fc;
657 cs->write = write;
658 cs->iov = iov;
659 cs->nr_segs = nr_segs;
660 }
661
662 /* Unmap and put previous page of userspace buffer */
663 static void fuse_copy_finish(struct fuse_copy_state *cs)
664 {
665 if (cs->currbuf) {
666 struct pipe_buffer *buf = cs->currbuf;
667
668 if (!cs->write) {
669 buf->ops->unmap(cs->pipe, buf, cs->mapaddr);
670 } else {
671 kunmap(buf->page);
672 buf->len = PAGE_SIZE - cs->len;
673 }
674 cs->currbuf = NULL;
675 cs->mapaddr = NULL;
676 } else if (cs->mapaddr) {
677 kunmap(cs->pg);
678 if (cs->write) {
679 flush_dcache_page(cs->pg);
680 set_page_dirty_lock(cs->pg);
681 }
682 put_page(cs->pg);
683 cs->mapaddr = NULL;
684 }
685 }
686
687 /*
688 * Get another pagefull of userspace buffer, and map it to kernel
689 * address space, and lock request
690 */
691 static int fuse_copy_fill(struct fuse_copy_state *cs)
692 {
693 unsigned long offset;
694 int err;
695
696 unlock_request(cs->fc, cs->req);
697 fuse_copy_finish(cs);
698 if (cs->pipebufs) {
699 struct pipe_buffer *buf = cs->pipebufs;
700
701 if (!cs->write) {
702 err = buf->ops->confirm(cs->pipe, buf);
703 if (err)
704 return err;
705
706 BUG_ON(!cs->nr_segs);
707 cs->currbuf = buf;
708 cs->mapaddr = buf->ops->map(cs->pipe, buf, 0);
709 cs->len = buf->len;
710 cs->buf = cs->mapaddr + buf->offset;
711 cs->pipebufs++;
712 cs->nr_segs--;
713 } else {
714 struct page *page;
715
716 if (cs->nr_segs == cs->pipe->buffers)
717 return -EIO;
718
719 page = alloc_page(GFP_HIGHUSER);
720 if (!page)
721 return -ENOMEM;
722
723 buf->page = page;
724 buf->offset = 0;
725 buf->len = 0;
726
727 cs->currbuf = buf;
728 cs->mapaddr = kmap(page);
729 cs->buf = cs->mapaddr;
730 cs->len = PAGE_SIZE;
731 cs->pipebufs++;
732 cs->nr_segs++;
733 }
734 } else {
735 if (!cs->seglen) {
736 BUG_ON(!cs->nr_segs);
737 cs->seglen = cs->iov[0].iov_len;
738 cs->addr = (unsigned long) cs->iov[0].iov_base;
739 cs->iov++;
740 cs->nr_segs--;
741 }
742 err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg);
743 if (err < 0)
744 return err;
745 BUG_ON(err != 1);
746 offset = cs->addr % PAGE_SIZE;
747 cs->mapaddr = kmap(cs->pg);
748 cs->buf = cs->mapaddr + offset;
749 cs->len = min(PAGE_SIZE - offset, cs->seglen);
750 cs->seglen -= cs->len;
751 cs->addr += cs->len;
752 }
753
754 return lock_request(cs->fc, cs->req);
755 }
756
757 /* Do as much copy to/from userspace buffer as we can */
758 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
759 {
760 unsigned ncpy = min(*size, cs->len);
761 if (val) {
762 if (cs->write)
763 memcpy(cs->buf, *val, ncpy);
764 else
765 memcpy(*val, cs->buf, ncpy);
766 *val += ncpy;
767 }
768 *size -= ncpy;
769 cs->len -= ncpy;
770 cs->buf += ncpy;
771 return ncpy;
772 }
773
774 static int fuse_check_page(struct page *page)
775 {
776 if (page_mapcount(page) ||
777 page->mapping != NULL ||
778 page_count(page) != 1 ||
779 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
780 ~(1 << PG_locked |
781 1 << PG_referenced |
782 1 << PG_uptodate |
783 1 << PG_lru |
784 1 << PG_active |
785 1 << PG_reclaim))) {
786 printk(KERN_WARNING "fuse: trying to steal weird page\n");
787 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);
788 return 1;
789 }
790 return 0;
791 }
792
793 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
794 {
795 int err;
796 struct page *oldpage = *pagep;
797 struct page *newpage;
798 struct pipe_buffer *buf = cs->pipebufs;
799
800 unlock_request(cs->fc, cs->req);
801 fuse_copy_finish(cs);
802
803 err = buf->ops->confirm(cs->pipe, buf);
804 if (err)
805 return err;
806
807 BUG_ON(!cs->nr_segs);
808 cs->currbuf = buf;
809 cs->len = buf->len;
810 cs->pipebufs++;
811 cs->nr_segs--;
812
813 if (cs->len != PAGE_SIZE)
814 goto out_fallback;
815
816 if (buf->ops->steal(cs->pipe, buf) != 0)
817 goto out_fallback;
818
819 newpage = buf->page;
820
821 if (WARN_ON(!PageUptodate(newpage)))
822 return -EIO;
823
824 ClearPageMappedToDisk(newpage);
825
826 if (fuse_check_page(newpage) != 0)
827 goto out_fallback_unlock;
828
829 /*
830 * This is a new and locked page, it shouldn't be mapped or
831 * have any special flags on it
832 */
833 if (WARN_ON(page_mapped(oldpage)))
834 goto out_fallback_unlock;
835 if (WARN_ON(page_has_private(oldpage)))
836 goto out_fallback_unlock;
837 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
838 goto out_fallback_unlock;
839 if (WARN_ON(PageMlocked(oldpage)))
840 goto out_fallback_unlock;
841
842 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
843 if (err) {
844 unlock_page(newpage);
845 return err;
846 }
847
848 page_cache_get(newpage);
849
850 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
851 lru_cache_add_file(newpage);
852
853 err = 0;
854 spin_lock(&cs->fc->lock);
855 if (cs->req->aborted)
856 err = -ENOENT;
857 else
858 *pagep = newpage;
859 spin_unlock(&cs->fc->lock);
860
861 if (err) {
862 unlock_page(newpage);
863 page_cache_release(newpage);
864 return err;
865 }
866
867 unlock_page(oldpage);
868 page_cache_release(oldpage);
869 cs->len = 0;
870
871 return 0;
872
873 out_fallback_unlock:
874 unlock_page(newpage);
875 out_fallback:
876 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
877 cs->buf = cs->mapaddr + buf->offset;
878
879 err = lock_request(cs->fc, cs->req);
880 if (err)
881 return err;
882
883 return 1;
884 }
885
886 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
887 unsigned offset, unsigned count)
888 {
889 struct pipe_buffer *buf;
890
891 if (cs->nr_segs == cs->pipe->buffers)
892 return -EIO;
893
894 unlock_request(cs->fc, cs->req);
895 fuse_copy_finish(cs);
896
897 buf = cs->pipebufs;
898 page_cache_get(page);
899 buf->page = page;
900 buf->offset = offset;
901 buf->len = count;
902
903 cs->pipebufs++;
904 cs->nr_segs++;
905 cs->len = 0;
906
907 return 0;
908 }
909
910 /*
911 * Copy a page in the request to/from the userspace buffer. Must be
912 * done atomically
913 */
914 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
915 unsigned offset, unsigned count, int zeroing)
916 {
917 int err;
918 struct page *page = *pagep;
919
920 if (page && zeroing && count < PAGE_SIZE)
921 clear_highpage(page);
922
923 while (count) {
924 if (cs->write && cs->pipebufs && page) {
925 return fuse_ref_page(cs, page, offset, count);
926 } else if (!cs->len) {
927 if (cs->move_pages && page &&
928 offset == 0 && count == PAGE_SIZE) {
929 err = fuse_try_move_page(cs, pagep);
930 if (err <= 0)
931 return err;
932 } else {
933 err = fuse_copy_fill(cs);
934 if (err)
935 return err;
936 }
937 }
938 if (page) {
939 void *mapaddr = kmap_atomic(page);
940 void *buf = mapaddr + offset;
941 offset += fuse_copy_do(cs, &buf, &count);
942 kunmap_atomic(mapaddr);
943 } else
944 offset += fuse_copy_do(cs, NULL, &count);
945 }
946 if (page && !cs->write)
947 flush_dcache_page(page);
948 return 0;
949 }
950
951 /* Copy pages in the request to/from userspace buffer */
952 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
953 int zeroing)
954 {
955 unsigned i;
956 struct fuse_req *req = cs->req;
957
958 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
959 int err;
960 unsigned offset = req->page_descs[i].offset;
961 unsigned count = min(nbytes, req->page_descs[i].length);
962
963 err = fuse_copy_page(cs, &req->pages[i], offset, count,
964 zeroing);
965 if (err)
966 return err;
967
968 nbytes -= count;
969 }
970 return 0;
971 }
972
973 /* Copy a single argument in the request to/from userspace buffer */
974 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
975 {
976 while (size) {
977 if (!cs->len) {
978 int err = fuse_copy_fill(cs);
979 if (err)
980 return err;
981 }
982 fuse_copy_do(cs, &val, &size);
983 }
984 return 0;
985 }
986
987 /* Copy request arguments to/from userspace buffer */
988 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
989 unsigned argpages, struct fuse_arg *args,
990 int zeroing)
991 {
992 int err = 0;
993 unsigned i;
994
995 for (i = 0; !err && i < numargs; i++) {
996 struct fuse_arg *arg = &args[i];
997 if (i == numargs - 1 && argpages)
998 err = fuse_copy_pages(cs, arg->size, zeroing);
999 else
1000 err = fuse_copy_one(cs, arg->value, arg->size);
1001 }
1002 return err;
1003 }
1004
1005 static int forget_pending(struct fuse_conn *fc)
1006 {
1007 return fc->forget_list_head.next != NULL;
1008 }
1009
1010 static int request_pending(struct fuse_conn *fc)
1011 {
1012 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
1013 forget_pending(fc);
1014 }
1015
1016 /* Wait until a request is available on the pending list */
1017 static void request_wait(struct fuse_conn *fc)
1018 __releases(fc->lock)
1019 __acquires(fc->lock)
1020 {
1021 DECLARE_WAITQUEUE(wait, current);
1022
1023 add_wait_queue_exclusive(&fc->waitq, &wait);
1024 while (fc->connected && !request_pending(fc)) {
1025 set_current_state(TASK_INTERRUPTIBLE);
1026 if (signal_pending(current))
1027 break;
1028
1029 spin_unlock(&fc->lock);
1030 schedule();
1031 spin_lock(&fc->lock);
1032 }
1033 set_current_state(TASK_RUNNING);
1034 remove_wait_queue(&fc->waitq, &wait);
1035 }
1036
1037 /*
1038 * Transfer an interrupt request to userspace
1039 *
1040 * Unlike other requests this is assembled on demand, without a need
1041 * to allocate a separate fuse_req structure.
1042 *
1043 * Called with fc->lock held, releases it
1044 */
1045 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1046 size_t nbytes, struct fuse_req *req)
1047 __releases(fc->lock)
1048 {
1049 struct fuse_in_header ih;
1050 struct fuse_interrupt_in arg;
1051 unsigned reqsize = sizeof(ih) + sizeof(arg);
1052 int err;
1053
1054 list_del_init(&req->intr_entry);
1055 req->intr_unique = fuse_get_unique(fc);
1056 memset(&ih, 0, sizeof(ih));
1057 memset(&arg, 0, sizeof(arg));
1058 ih.len = reqsize;
1059 ih.opcode = FUSE_INTERRUPT;
1060 ih.unique = req->intr_unique;
1061 arg.unique = req->in.h.unique;
1062
1063 spin_unlock(&fc->lock);
1064 if (nbytes < reqsize)
1065 return -EINVAL;
1066
1067 err = fuse_copy_one(cs, &ih, sizeof(ih));
1068 if (!err)
1069 err = fuse_copy_one(cs, &arg, sizeof(arg));
1070 fuse_copy_finish(cs);
1071
1072 return err ? err : reqsize;
1073 }
1074
1075 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
1076 unsigned max,
1077 unsigned *countp)
1078 {
1079 struct fuse_forget_link *head = fc->forget_list_head.next;
1080 struct fuse_forget_link **newhead = &head;
1081 unsigned count;
1082
1083 for (count = 0; *newhead != NULL && count < max; count++)
1084 newhead = &(*newhead)->next;
1085
1086 fc->forget_list_head.next = *newhead;
1087 *newhead = NULL;
1088 if (fc->forget_list_head.next == NULL)
1089 fc->forget_list_tail = &fc->forget_list_head;
1090
1091 if (countp != NULL)
1092 *countp = count;
1093
1094 return head;
1095 }
1096
1097 static int fuse_read_single_forget(struct fuse_conn *fc,
1098 struct fuse_copy_state *cs,
1099 size_t nbytes)
1100 __releases(fc->lock)
1101 {
1102 int err;
1103 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1104 struct fuse_forget_in arg = {
1105 .nlookup = forget->forget_one.nlookup,
1106 };
1107 struct fuse_in_header ih = {
1108 .opcode = FUSE_FORGET,
1109 .nodeid = forget->forget_one.nodeid,
1110 .unique = fuse_get_unique(fc),
1111 .len = sizeof(ih) + sizeof(arg),
1112 };
1113
1114 spin_unlock(&fc->lock);
1115 kfree(forget);
1116 if (nbytes < ih.len)
1117 return -EINVAL;
1118
1119 err = fuse_copy_one(cs, &ih, sizeof(ih));
1120 if (!err)
1121 err = fuse_copy_one(cs, &arg, sizeof(arg));
1122 fuse_copy_finish(cs);
1123
1124 if (err)
1125 return err;
1126
1127 return ih.len;
1128 }
1129
1130 static int fuse_read_batch_forget(struct fuse_conn *fc,
1131 struct fuse_copy_state *cs, size_t nbytes)
1132 __releases(fc->lock)
1133 {
1134 int err;
1135 unsigned max_forgets;
1136 unsigned count;
1137 struct fuse_forget_link *head;
1138 struct fuse_batch_forget_in arg = { .count = 0 };
1139 struct fuse_in_header ih = {
1140 .opcode = FUSE_BATCH_FORGET,
1141 .unique = fuse_get_unique(fc),
1142 .len = sizeof(ih) + sizeof(arg),
1143 };
1144
1145 if (nbytes < ih.len) {
1146 spin_unlock(&fc->lock);
1147 return -EINVAL;
1148 }
1149
1150 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1151 head = dequeue_forget(fc, max_forgets, &count);
1152 spin_unlock(&fc->lock);
1153
1154 arg.count = count;
1155 ih.len += count * sizeof(struct fuse_forget_one);
1156 err = fuse_copy_one(cs, &ih, sizeof(ih));
1157 if (!err)
1158 err = fuse_copy_one(cs, &arg, sizeof(arg));
1159
1160 while (head) {
1161 struct fuse_forget_link *forget = head;
1162
1163 if (!err) {
1164 err = fuse_copy_one(cs, &forget->forget_one,
1165 sizeof(forget->forget_one));
1166 }
1167 head = forget->next;
1168 kfree(forget);
1169 }
1170
1171 fuse_copy_finish(cs);
1172
1173 if (err)
1174 return err;
1175
1176 return ih.len;
1177 }
1178
1179 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1180 size_t nbytes)
1181 __releases(fc->lock)
1182 {
1183 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1184 return fuse_read_single_forget(fc, cs, nbytes);
1185 else
1186 return fuse_read_batch_forget(fc, cs, nbytes);
1187 }
1188
1189 /*
1190 * Read a single request into the userspace filesystem's buffer. This
1191 * function waits until a request is available, then removes it from
1192 * the pending list and copies request data to userspace buffer. If
1193 * no reply is needed (FORGET) or request has been aborted or there
1194 * was an error during the copying then it's finished by calling
1195 * request_end(). Otherwise add it to the processing list, and set
1196 * the 'sent' flag.
1197 */
1198 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1199 struct fuse_copy_state *cs, size_t nbytes)
1200 {
1201 int err;
1202 struct fuse_req *req;
1203 struct fuse_in *in;
1204 unsigned reqsize;
1205
1206 restart:
1207 spin_lock(&fc->lock);
1208 err = -EAGAIN;
1209 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1210 !request_pending(fc))
1211 goto err_unlock;
1212
1213 request_wait(fc);
1214 err = -ENODEV;
1215 if (!fc->connected)
1216 goto err_unlock;
1217 err = -ERESTARTSYS;
1218 if (!request_pending(fc))
1219 goto err_unlock;
1220
1221 if (!list_empty(&fc->interrupts)) {
1222 req = list_entry(fc->interrupts.next, struct fuse_req,
1223 intr_entry);
1224 return fuse_read_interrupt(fc, cs, nbytes, req);
1225 }
1226
1227 if (forget_pending(fc)) {
1228 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1229 return fuse_read_forget(fc, cs, nbytes);
1230
1231 if (fc->forget_batch <= -8)
1232 fc->forget_batch = 16;
1233 }
1234
1235 req = list_entry(fc->pending.next, struct fuse_req, list);
1236 req->state = FUSE_REQ_READING;
1237 list_move(&req->list, &fc->io);
1238
1239 in = &req->in;
1240 reqsize = in->h.len;
1241 /* If request is too large, reply with an error and restart the read */
1242 if (nbytes < reqsize) {
1243 req->out.h.error = -EIO;
1244 /* SETXATTR is special, since it may contain too large data */
1245 if (in->h.opcode == FUSE_SETXATTR)
1246 req->out.h.error = -E2BIG;
1247 request_end(fc, req);
1248 goto restart;
1249 }
1250 spin_unlock(&fc->lock);
1251 cs->req = req;
1252 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1253 if (!err)
1254 err = fuse_copy_args(cs, in->numargs, in->argpages,
1255 (struct fuse_arg *) in->args, 0);
1256 fuse_copy_finish(cs);
1257 spin_lock(&fc->lock);
1258 req->locked = 0;
1259 if (req->aborted) {
1260 request_end(fc, req);
1261 return -ENODEV;
1262 }
1263 if (err) {
1264 req->out.h.error = -EIO;
1265 request_end(fc, req);
1266 return err;
1267 }
1268 if (!req->isreply)
1269 request_end(fc, req);
1270 else {
1271 req->state = FUSE_REQ_SENT;
1272 list_move_tail(&req->list, &fc->processing);
1273 if (req->interrupted)
1274 queue_interrupt(fc, req);
1275 spin_unlock(&fc->lock);
1276 }
1277 return reqsize;
1278
1279 err_unlock:
1280 spin_unlock(&fc->lock);
1281 return err;
1282 }
1283
1284 static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1285 unsigned long nr_segs, loff_t pos)
1286 {
1287 struct fuse_copy_state cs;
1288 struct file *file = iocb->ki_filp;
1289 struct fuse_conn *fc = fuse_get_conn(file);
1290 if (!fc)
1291 return -EPERM;
1292
1293 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1294
1295 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1296 }
1297
1298 static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe,
1299 struct pipe_buffer *buf)
1300 {
1301 return 1;
1302 }
1303
1304 static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = {
1305 .can_merge = 0,
1306 .map = generic_pipe_buf_map,
1307 .unmap = generic_pipe_buf_unmap,
1308 .confirm = generic_pipe_buf_confirm,
1309 .release = generic_pipe_buf_release,
1310 .steal = fuse_dev_pipe_buf_steal,
1311 .get = generic_pipe_buf_get,
1312 };
1313
1314 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1315 struct pipe_inode_info *pipe,
1316 size_t len, unsigned int flags)
1317 {
1318 int ret;
1319 int page_nr = 0;
1320 int do_wakeup = 0;
1321 struct pipe_buffer *bufs;
1322 struct fuse_copy_state cs;
1323 struct fuse_conn *fc = fuse_get_conn(in);
1324 if (!fc)
1325 return -EPERM;
1326
1327 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1328 if (!bufs)
1329 return -ENOMEM;
1330
1331 fuse_copy_init(&cs, fc, 1, NULL, 0);
1332 cs.pipebufs = bufs;
1333 cs.pipe = pipe;
1334 ret = fuse_dev_do_read(fc, in, &cs, len);
1335 if (ret < 0)
1336 goto out;
1337
1338 ret = 0;
1339 pipe_lock(pipe);
1340
1341 if (!pipe->readers) {
1342 send_sig(SIGPIPE, current, 0);
1343 if (!ret)
1344 ret = -EPIPE;
1345 goto out_unlock;
1346 }
1347
1348 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1349 ret = -EIO;
1350 goto out_unlock;
1351 }
1352
1353 while (page_nr < cs.nr_segs) {
1354 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1355 struct pipe_buffer *buf = pipe->bufs + newbuf;
1356
1357 buf->page = bufs[page_nr].page;
1358 buf->offset = bufs[page_nr].offset;
1359 buf->len = bufs[page_nr].len;
1360 buf->ops = &fuse_dev_pipe_buf_ops;
1361
1362 pipe->nrbufs++;
1363 page_nr++;
1364 ret += buf->len;
1365
1366 if (pipe->files)
1367 do_wakeup = 1;
1368 }
1369
1370 out_unlock:
1371 pipe_unlock(pipe);
1372
1373 if (do_wakeup) {
1374 smp_mb();
1375 if (waitqueue_active(&pipe->wait))
1376 wake_up_interruptible(&pipe->wait);
1377 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1378 }
1379
1380 out:
1381 for (; page_nr < cs.nr_segs; page_nr++)
1382 page_cache_release(bufs[page_nr].page);
1383
1384 kfree(bufs);
1385 return ret;
1386 }
1387
1388 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1389 struct fuse_copy_state *cs)
1390 {
1391 struct fuse_notify_poll_wakeup_out outarg;
1392 int err = -EINVAL;
1393
1394 if (size != sizeof(outarg))
1395 goto err;
1396
1397 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1398 if (err)
1399 goto err;
1400
1401 fuse_copy_finish(cs);
1402 return fuse_notify_poll_wakeup(fc, &outarg);
1403
1404 err:
1405 fuse_copy_finish(cs);
1406 return err;
1407 }
1408
1409 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1410 struct fuse_copy_state *cs)
1411 {
1412 struct fuse_notify_inval_inode_out outarg;
1413 int err = -EINVAL;
1414
1415 if (size != sizeof(outarg))
1416 goto err;
1417
1418 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1419 if (err)
1420 goto err;
1421 fuse_copy_finish(cs);
1422
1423 down_read(&fc->killsb);
1424 err = -ENOENT;
1425 if (fc->sb) {
1426 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1427 outarg.off, outarg.len);
1428 }
1429 up_read(&fc->killsb);
1430 return err;
1431
1432 err:
1433 fuse_copy_finish(cs);
1434 return err;
1435 }
1436
1437 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1438 struct fuse_copy_state *cs)
1439 {
1440 struct fuse_notify_inval_entry_out outarg;
1441 int err = -ENOMEM;
1442 char *buf;
1443 struct qstr name;
1444
1445 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1446 if (!buf)
1447 goto err;
1448
1449 err = -EINVAL;
1450 if (size < sizeof(outarg))
1451 goto err;
1452
1453 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1454 if (err)
1455 goto err;
1456
1457 err = -ENAMETOOLONG;
1458 if (outarg.namelen > FUSE_NAME_MAX)
1459 goto err;
1460
1461 err = -EINVAL;
1462 if (size != sizeof(outarg) + outarg.namelen + 1)
1463 goto err;
1464
1465 name.name = buf;
1466 name.len = outarg.namelen;
1467 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1468 if (err)
1469 goto err;
1470 fuse_copy_finish(cs);
1471 buf[outarg.namelen] = 0;
1472 name.hash = full_name_hash(name.name, name.len);
1473
1474 down_read(&fc->killsb);
1475 err = -ENOENT;
1476 if (fc->sb)
1477 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1478 up_read(&fc->killsb);
1479 kfree(buf);
1480 return err;
1481
1482 err:
1483 kfree(buf);
1484 fuse_copy_finish(cs);
1485 return err;
1486 }
1487
1488 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1489 struct fuse_copy_state *cs)
1490 {
1491 struct fuse_notify_delete_out outarg;
1492 int err = -ENOMEM;
1493 char *buf;
1494 struct qstr name;
1495
1496 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1497 if (!buf)
1498 goto err;
1499
1500 err = -EINVAL;
1501 if (size < sizeof(outarg))
1502 goto err;
1503
1504 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1505 if (err)
1506 goto err;
1507
1508 err = -ENAMETOOLONG;
1509 if (outarg.namelen > FUSE_NAME_MAX)
1510 goto err;
1511
1512 err = -EINVAL;
1513 if (size != sizeof(outarg) + outarg.namelen + 1)
1514 goto err;
1515
1516 name.name = buf;
1517 name.len = outarg.namelen;
1518 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1519 if (err)
1520 goto err;
1521 fuse_copy_finish(cs);
1522 buf[outarg.namelen] = 0;
1523 name.hash = full_name_hash(name.name, name.len);
1524
1525 down_read(&fc->killsb);
1526 err = -ENOENT;
1527 if (fc->sb)
1528 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1529 outarg.child, &name);
1530 up_read(&fc->killsb);
1531 kfree(buf);
1532 return err;
1533
1534 err:
1535 kfree(buf);
1536 fuse_copy_finish(cs);
1537 return err;
1538 }
1539
1540 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1541 struct fuse_copy_state *cs)
1542 {
1543 struct fuse_notify_store_out outarg;
1544 struct inode *inode;
1545 struct address_space *mapping;
1546 u64 nodeid;
1547 int err;
1548 pgoff_t index;
1549 unsigned int offset;
1550 unsigned int num;
1551 loff_t file_size;
1552 loff_t end;
1553
1554 err = -EINVAL;
1555 if (size < sizeof(outarg))
1556 goto out_finish;
1557
1558 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1559 if (err)
1560 goto out_finish;
1561
1562 err = -EINVAL;
1563 if (size - sizeof(outarg) != outarg.size)
1564 goto out_finish;
1565
1566 nodeid = outarg.nodeid;
1567
1568 down_read(&fc->killsb);
1569
1570 err = -ENOENT;
1571 if (!fc->sb)
1572 goto out_up_killsb;
1573
1574 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1575 if (!inode)
1576 goto out_up_killsb;
1577
1578 mapping = inode->i_mapping;
1579 index = outarg.offset >> PAGE_CACHE_SHIFT;
1580 offset = outarg.offset & ~PAGE_CACHE_MASK;
1581 file_size = i_size_read(inode);
1582 end = outarg.offset + outarg.size;
1583 if (end > file_size) {
1584 file_size = end;
1585 fuse_write_update_size(inode, file_size);
1586 }
1587
1588 num = outarg.size;
1589 while (num) {
1590 struct page *page;
1591 unsigned int this_num;
1592
1593 err = -ENOMEM;
1594 page = find_or_create_page(mapping, index,
1595 mapping_gfp_mask(mapping));
1596 if (!page)
1597 goto out_iput;
1598
1599 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1600 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1601 if (!err && offset == 0 && (num != 0 || file_size == end))
1602 SetPageUptodate(page);
1603 unlock_page(page);
1604 page_cache_release(page);
1605
1606 if (err)
1607 goto out_iput;
1608
1609 num -= this_num;
1610 offset = 0;
1611 index++;
1612 }
1613
1614 err = 0;
1615
1616 out_iput:
1617 iput(inode);
1618 out_up_killsb:
1619 up_read(&fc->killsb);
1620 out_finish:
1621 fuse_copy_finish(cs);
1622 return err;
1623 }
1624
1625 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1626 {
1627 release_pages(req->pages, req->num_pages, 0);
1628 }
1629
1630 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1631 struct fuse_notify_retrieve_out *outarg)
1632 {
1633 int err;
1634 struct address_space *mapping = inode->i_mapping;
1635 struct fuse_req *req;
1636 pgoff_t index;
1637 loff_t file_size;
1638 unsigned int num;
1639 unsigned int offset;
1640 size_t total_len = 0;
1641 int num_pages;
1642
1643 offset = outarg->offset & ~PAGE_CACHE_MASK;
1644 file_size = i_size_read(inode);
1645
1646 num = outarg->size;
1647 if (outarg->offset > file_size)
1648 num = 0;
1649 else if (outarg->offset + num > file_size)
1650 num = file_size - outarg->offset;
1651
1652 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1653 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1654
1655 req = fuse_get_req(fc, num_pages);
1656 if (IS_ERR(req))
1657 return PTR_ERR(req);
1658
1659 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1660 req->in.h.nodeid = outarg->nodeid;
1661 req->in.numargs = 2;
1662 req->in.argpages = 1;
1663 req->page_descs[0].offset = offset;
1664 req->end = fuse_retrieve_end;
1665
1666 index = outarg->offset >> PAGE_CACHE_SHIFT;
1667
1668 while (num && req->num_pages < num_pages) {
1669 struct page *page;
1670 unsigned int this_num;
1671
1672 page = find_get_page(mapping, index);
1673 if (!page)
1674 break;
1675
1676 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1677 req->pages[req->num_pages] = page;
1678 req->page_descs[req->num_pages].length = this_num;
1679 req->num_pages++;
1680
1681 offset = 0;
1682 num -= this_num;
1683 total_len += this_num;
1684 index++;
1685 }
1686 req->misc.retrieve_in.offset = outarg->offset;
1687 req->misc.retrieve_in.size = total_len;
1688 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1689 req->in.args[0].value = &req->misc.retrieve_in;
1690 req->in.args[1].size = total_len;
1691
1692 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1693 if (err)
1694 fuse_retrieve_end(fc, req);
1695
1696 return err;
1697 }
1698
1699 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1700 struct fuse_copy_state *cs)
1701 {
1702 struct fuse_notify_retrieve_out outarg;
1703 struct inode *inode;
1704 int err;
1705
1706 err = -EINVAL;
1707 if (size != sizeof(outarg))
1708 goto copy_finish;
1709
1710 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1711 if (err)
1712 goto copy_finish;
1713
1714 fuse_copy_finish(cs);
1715
1716 down_read(&fc->killsb);
1717 err = -ENOENT;
1718 if (fc->sb) {
1719 u64 nodeid = outarg.nodeid;
1720
1721 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1722 if (inode) {
1723 err = fuse_retrieve(fc, inode, &outarg);
1724 iput(inode);
1725 }
1726 }
1727 up_read(&fc->killsb);
1728
1729 return err;
1730
1731 copy_finish:
1732 fuse_copy_finish(cs);
1733 return err;
1734 }
1735
1736 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1737 unsigned int size, struct fuse_copy_state *cs)
1738 {
1739 switch (code) {
1740 case FUSE_NOTIFY_POLL:
1741 return fuse_notify_poll(fc, size, cs);
1742
1743 case FUSE_NOTIFY_INVAL_INODE:
1744 return fuse_notify_inval_inode(fc, size, cs);
1745
1746 case FUSE_NOTIFY_INVAL_ENTRY:
1747 return fuse_notify_inval_entry(fc, size, cs);
1748
1749 case FUSE_NOTIFY_STORE:
1750 return fuse_notify_store(fc, size, cs);
1751
1752 case FUSE_NOTIFY_RETRIEVE:
1753 return fuse_notify_retrieve(fc, size, cs);
1754
1755 case FUSE_NOTIFY_DELETE:
1756 return fuse_notify_delete(fc, size, cs);
1757
1758 default:
1759 fuse_copy_finish(cs);
1760 return -EINVAL;
1761 }
1762 }
1763
1764 /* Look up request on processing list by unique ID */
1765 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1766 {
1767 struct list_head *entry;
1768
1769 list_for_each(entry, &fc->processing) {
1770 struct fuse_req *req;
1771 req = list_entry(entry, struct fuse_req, list);
1772 if (req->in.h.unique == unique || req->intr_unique == unique)
1773 return req;
1774 }
1775 return NULL;
1776 }
1777
1778 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1779 unsigned nbytes)
1780 {
1781 unsigned reqsize = sizeof(struct fuse_out_header);
1782
1783 if (out->h.error)
1784 return nbytes != reqsize ? -EINVAL : 0;
1785
1786 reqsize += len_args(out->numargs, out->args);
1787
1788 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1789 return -EINVAL;
1790 else if (reqsize > nbytes) {
1791 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1792 unsigned diffsize = reqsize - nbytes;
1793 if (diffsize > lastarg->size)
1794 return -EINVAL;
1795 lastarg->size -= diffsize;
1796 }
1797 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1798 out->page_zeroing);
1799 }
1800
1801 /*
1802 * Write a single reply to a request. First the header is copied from
1803 * the write buffer. The request is then searched on the processing
1804 * list by the unique ID found in the header. If found, then remove
1805 * it from the list and copy the rest of the buffer to the request.
1806 * The request is finished by calling request_end()
1807 */
1808 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1809 struct fuse_copy_state *cs, size_t nbytes)
1810 {
1811 int err;
1812 struct fuse_req *req;
1813 struct fuse_out_header oh;
1814
1815 if (nbytes < sizeof(struct fuse_out_header))
1816 return -EINVAL;
1817
1818 err = fuse_copy_one(cs, &oh, sizeof(oh));
1819 if (err)
1820 goto err_finish;
1821
1822 err = -EINVAL;
1823 if (oh.len != nbytes)
1824 goto err_finish;
1825
1826 /*
1827 * Zero oh.unique indicates unsolicited notification message
1828 * and error contains notification code.
1829 */
1830 if (!oh.unique) {
1831 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1832 return err ? err : nbytes;
1833 }
1834
1835 err = -EINVAL;
1836 if (oh.error <= -1000 || oh.error > 0)
1837 goto err_finish;
1838
1839 spin_lock(&fc->lock);
1840 err = -ENOENT;
1841 if (!fc->connected)
1842 goto err_unlock;
1843
1844 req = request_find(fc, oh.unique);
1845 if (!req)
1846 goto err_unlock;
1847
1848 if (req->aborted) {
1849 spin_unlock(&fc->lock);
1850 fuse_copy_finish(cs);
1851 spin_lock(&fc->lock);
1852 request_end(fc, req);
1853 return -ENOENT;
1854 }
1855 /* Is it an interrupt reply? */
1856 if (req->intr_unique == oh.unique) {
1857 err = -EINVAL;
1858 if (nbytes != sizeof(struct fuse_out_header))
1859 goto err_unlock;
1860
1861 if (oh.error == -ENOSYS)
1862 fc->no_interrupt = 1;
1863 else if (oh.error == -EAGAIN)
1864 queue_interrupt(fc, req);
1865
1866 spin_unlock(&fc->lock);
1867 fuse_copy_finish(cs);
1868 return nbytes;
1869 }
1870
1871 req->state = FUSE_REQ_WRITING;
1872 list_move(&req->list, &fc->io);
1873 req->out.h = oh;
1874 req->locked = 1;
1875 cs->req = req;
1876 if (!req->out.page_replace)
1877 cs->move_pages = 0;
1878 spin_unlock(&fc->lock);
1879
1880 err = copy_out_args(cs, &req->out, nbytes);
1881 fuse_copy_finish(cs);
1882
1883 spin_lock(&fc->lock);
1884 req->locked = 0;
1885 if (!err) {
1886 if (req->aborted)
1887 err = -ENOENT;
1888 } else if (!req->aborted)
1889 req->out.h.error = -EIO;
1890 request_end(fc, req);
1891
1892 return err ? err : nbytes;
1893
1894 err_unlock:
1895 spin_unlock(&fc->lock);
1896 err_finish:
1897 fuse_copy_finish(cs);
1898 return err;
1899 }
1900
1901 static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1902 unsigned long nr_segs, loff_t pos)
1903 {
1904 struct fuse_copy_state cs;
1905 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1906 if (!fc)
1907 return -EPERM;
1908
1909 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1910
1911 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1912 }
1913
1914 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1915 struct file *out, loff_t *ppos,
1916 size_t len, unsigned int flags)
1917 {
1918 unsigned nbuf;
1919 unsigned idx;
1920 struct pipe_buffer *bufs;
1921 struct fuse_copy_state cs;
1922 struct fuse_conn *fc;
1923 size_t rem;
1924 ssize_t ret;
1925
1926 fc = fuse_get_conn(out);
1927 if (!fc)
1928 return -EPERM;
1929
1930 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1931 if (!bufs)
1932 return -ENOMEM;
1933
1934 pipe_lock(pipe);
1935 nbuf = 0;
1936 rem = 0;
1937 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1938 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1939
1940 ret = -EINVAL;
1941 if (rem < len) {
1942 pipe_unlock(pipe);
1943 goto out;
1944 }
1945
1946 rem = len;
1947 while (rem) {
1948 struct pipe_buffer *ibuf;
1949 struct pipe_buffer *obuf;
1950
1951 BUG_ON(nbuf >= pipe->buffers);
1952 BUG_ON(!pipe->nrbufs);
1953 ibuf = &pipe->bufs[pipe->curbuf];
1954 obuf = &bufs[nbuf];
1955
1956 if (rem >= ibuf->len) {
1957 *obuf = *ibuf;
1958 ibuf->ops = NULL;
1959 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1960 pipe->nrbufs--;
1961 } else {
1962 ibuf->ops->get(pipe, ibuf);
1963 *obuf = *ibuf;
1964 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1965 obuf->len = rem;
1966 ibuf->offset += obuf->len;
1967 ibuf->len -= obuf->len;
1968 }
1969 nbuf++;
1970 rem -= obuf->len;
1971 }
1972 pipe_unlock(pipe);
1973
1974 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1975 cs.pipebufs = bufs;
1976 cs.pipe = pipe;
1977
1978 if (flags & SPLICE_F_MOVE)
1979 cs.move_pages = 1;
1980
1981 ret = fuse_dev_do_write(fc, &cs, len);
1982
1983 for (idx = 0; idx < nbuf; idx++) {
1984 struct pipe_buffer *buf = &bufs[idx];
1985 buf->ops->release(pipe, buf);
1986 }
1987 out:
1988 kfree(bufs);
1989 return ret;
1990 }
1991
1992 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1993 {
1994 unsigned mask = POLLOUT | POLLWRNORM;
1995 struct fuse_conn *fc = fuse_get_conn(file);
1996 if (!fc)
1997 return POLLERR;
1998
1999 poll_wait(file, &fc->waitq, wait);
2000
2001 spin_lock(&fc->lock);
2002 if (!fc->connected)
2003 mask = POLLERR;
2004 else if (request_pending(fc))
2005 mask |= POLLIN | POLLRDNORM;
2006 spin_unlock(&fc->lock);
2007
2008 return mask;
2009 }
2010
2011 /*
2012 * Abort all requests on the given list (pending or processing)
2013 *
2014 * This function releases and reacquires fc->lock
2015 */
2016 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2017 __releases(fc->lock)
2018 __acquires(fc->lock)
2019 {
2020 while (!list_empty(head)) {
2021 struct fuse_req *req;
2022 req = list_entry(head->next, struct fuse_req, list);
2023 req->out.h.error = -ECONNABORTED;
2024 request_end(fc, req);
2025 spin_lock(&fc->lock);
2026 }
2027 }
2028
2029 /*
2030 * Abort requests under I/O
2031 *
2032 * The requests are set to aborted and finished, and the request
2033 * waiter is woken up. This will make request_wait_answer() wait
2034 * until the request is unlocked and then return.
2035 *
2036 * If the request is asynchronous, then the end function needs to be
2037 * called after waiting for the request to be unlocked (if it was
2038 * locked).
2039 */
2040 static void end_io_requests(struct fuse_conn *fc)
2041 __releases(fc->lock)
2042 __acquires(fc->lock)
2043 {
2044 while (!list_empty(&fc->io)) {
2045 struct fuse_req *req =
2046 list_entry(fc->io.next, struct fuse_req, list);
2047 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
2048
2049 req->aborted = 1;
2050 req->out.h.error = -ECONNABORTED;
2051 req->state = FUSE_REQ_FINISHED;
2052 list_del_init(&req->list);
2053 wake_up(&req->waitq);
2054 if (end) {
2055 req->end = NULL;
2056 __fuse_get_request(req);
2057 spin_unlock(&fc->lock);
2058 wait_event(req->waitq, !req->locked);
2059 end(fc, req);
2060 fuse_put_request(fc, req);
2061 spin_lock(&fc->lock);
2062 }
2063 }
2064 }
2065
2066 static void end_queued_requests(struct fuse_conn *fc)
2067 __releases(fc->lock)
2068 __acquires(fc->lock)
2069 {
2070 fc->max_background = UINT_MAX;
2071 flush_bg_queue(fc);
2072 end_requests(fc, &fc->pending);
2073 end_requests(fc, &fc->processing);
2074 while (forget_pending(fc))
2075 kfree(dequeue_forget(fc, 1, NULL));
2076 }
2077
2078 static void end_polls(struct fuse_conn *fc)
2079 {
2080 struct rb_node *p;
2081
2082 p = rb_first(&fc->polled_files);
2083
2084 while (p) {
2085 struct fuse_file *ff;
2086 ff = rb_entry(p, struct fuse_file, polled_node);
2087 wake_up_interruptible_all(&ff->poll_wait);
2088
2089 p = rb_next(p);
2090 }
2091 }
2092
2093 /*
2094 * Abort all requests.
2095 *
2096 * Emergency exit in case of a malicious or accidental deadlock, or
2097 * just a hung filesystem.
2098 *
2099 * The same effect is usually achievable through killing the
2100 * filesystem daemon and all users of the filesystem. The exception
2101 * is the combination of an asynchronous request and the tricky
2102 * deadlock (see Documentation/filesystems/fuse.txt).
2103 *
2104 * During the aborting, progression of requests from the pending and
2105 * processing lists onto the io list, and progression of new requests
2106 * onto the pending list is prevented by req->connected being false.
2107 *
2108 * Progression of requests under I/O to the processing list is
2109 * prevented by the req->aborted flag being true for these requests.
2110 * For this reason requests on the io list must be aborted first.
2111 */
2112 void fuse_abort_conn(struct fuse_conn *fc)
2113 {
2114 spin_lock(&fc->lock);
2115 if (fc->connected) {
2116 fc->connected = 0;
2117 fc->blocked = 0;
2118 fc->initialized = 1;
2119 end_io_requests(fc);
2120 end_queued_requests(fc);
2121 end_polls(fc);
2122 wake_up_all(&fc->waitq);
2123 wake_up_all(&fc->blocked_waitq);
2124 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2125 }
2126 spin_unlock(&fc->lock);
2127 }
2128 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2129
2130 int fuse_dev_release(struct inode *inode, struct file *file)
2131 {
2132 struct fuse_conn *fc = fuse_get_conn(file);
2133 if (fc) {
2134 spin_lock(&fc->lock);
2135 fc->connected = 0;
2136 fc->blocked = 0;
2137 fc->initialized = 1;
2138 end_queued_requests(fc);
2139 end_polls(fc);
2140 wake_up_all(&fc->blocked_waitq);
2141 spin_unlock(&fc->lock);
2142 fuse_conn_put(fc);
2143 }
2144
2145 return 0;
2146 }
2147 EXPORT_SYMBOL_GPL(fuse_dev_release);
2148
2149 static int fuse_dev_fasync(int fd, struct file *file, int on)
2150 {
2151 struct fuse_conn *fc = fuse_get_conn(file);
2152 if (!fc)
2153 return -EPERM;
2154
2155 /* No locking - fasync_helper does its own locking */
2156 return fasync_helper(fd, file, on, &fc->fasync);
2157 }
2158
2159 const struct file_operations fuse_dev_operations = {
2160 .owner = THIS_MODULE,
2161 .llseek = no_llseek,
2162 .read = do_sync_read,
2163 .aio_read = fuse_dev_read,
2164 .splice_read = fuse_dev_splice_read,
2165 .write = do_sync_write,
2166 .aio_write = fuse_dev_write,
2167 .splice_write = fuse_dev_splice_write,
2168 .poll = fuse_dev_poll,
2169 .release = fuse_dev_release,
2170 .fasync = fuse_dev_fasync,
2171 };
2172 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2173
2174 static struct miscdevice fuse_miscdevice = {
2175 .minor = FUSE_MINOR,
2176 .name = "fuse",
2177 .fops = &fuse_dev_operations,
2178 };
2179
2180 int __init fuse_dev_init(void)
2181 {
2182 int err = -ENOMEM;
2183 fuse_req_cachep = kmem_cache_create("fuse_request",
2184 sizeof(struct fuse_req),
2185 0, 0, NULL);
2186 if (!fuse_req_cachep)
2187 goto out;
2188
2189 err = misc_register(&fuse_miscdevice);
2190 if (err)
2191 goto out_cache_clean;
2192
2193 return 0;
2194
2195 out_cache_clean:
2196 kmem_cache_destroy(fuse_req_cachep);
2197 out:
2198 return err;
2199 }
2200
2201 void fuse_dev_cleanup(void)
2202 {
2203 misc_deregister(&fuse_miscdevice);
2204 kmem_cache_destroy(fuse_req_cachep);
2205 }
kernel source for telekom puls (alcatel/mediatek)