?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
hyperv: Add processing of MTU reduced by the host
[linux/fpc-iii.git] / fs / fuse / dev.c
blobca887314aba9deb6d59811f0d19b8c9b9a0381c3
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/uio.h>
15 #include <linux/miscdevice.h>
16 #include <linux/pagemap.h>
17 #include <linux/file.h>
18 #include <linux/slab.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/swap.h>
21 #include <linux/splice.h>
22 #include <linux/aio.h>
23
24 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
25 MODULE_ALIAS("devname:fuse");
26
27 static struct kmem_cache *fuse_req_cachep;
28
29 static struct fuse_conn *fuse_get_conn(struct file *file)
30 {
31 /*
32 * Lockless access is OK, because file->private data is set
33 * once during mount and is valid until the file is released.
34 */
35 return file->private_data;
36 }
37
38 static void fuse_request_init(struct fuse_req *req, struct page **pages,
39 struct fuse_page_desc *page_descs,
40 unsigned npages)
41 {
42 memset(req, 0, sizeof(*req));
43 memset(pages, 0, sizeof(*pages) * npages);
44 memset(page_descs, 0, sizeof(*page_descs) * npages);
45 INIT_LIST_HEAD(&req->list);
46 INIT_LIST_HEAD(&req->intr_entry);
47 init_waitqueue_head(&req->waitq);
48 atomic_set(&req->count, 1);
49 req->pages = pages;
50 req->page_descs = page_descs;
51 req->max_pages = npages;
52 }
53
54 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
55 {
56 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
57 if (req) {
58 struct page **pages;
59 struct fuse_page_desc *page_descs;
60
61 if (npages <= FUSE_REQ_INLINE_PAGES) {
62 pages = req->inline_pages;
63 page_descs = req->inline_page_descs;
64 } else {
65 pages = kmalloc(sizeof(struct page *) * npages, flags);
66 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
67 npages, flags);
68 }
69
70 if (!pages || !page_descs) {
71 kfree(pages);
72 kfree(page_descs);
73 kmem_cache_free(fuse_req_cachep, req);
74 return NULL;
75 }
76
77 fuse_request_init(req, pages, page_descs, npages);
78 }
79 return req;
80 }
81
82 struct fuse_req *fuse_request_alloc(unsigned npages)
83 {
84 return __fuse_request_alloc(npages, GFP_KERNEL);
85 }
86 EXPORT_SYMBOL_GPL(fuse_request_alloc);
87
88 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
89 {
90 return __fuse_request_alloc(npages, GFP_NOFS);
91 }
92
93 void fuse_request_free(struct fuse_req *req)
94 {
95 if (req->pages != req->inline_pages) {
96 kfree(req->pages);
97 kfree(req->page_descs);
98 }
99 kmem_cache_free(fuse_req_cachep, req);
100 }
101
102 static void block_sigs(sigset_t *oldset)
103 {
104 sigset_t mask;
105
106 siginitsetinv(&mask, sigmask(SIGKILL));
107 sigprocmask(SIG_BLOCK, &mask, oldset);
108 }
109
110 static void restore_sigs(sigset_t *oldset)
111 {
112 sigprocmask(SIG_SETMASK, oldset, NULL);
113 }
114
115 void __fuse_get_request(struct fuse_req *req)
116 {
117 atomic_inc(&req->count);
118 }
119
120 /* Must be called with > 1 refcount */
121 static void __fuse_put_request(struct fuse_req *req)
122 {
123 BUG_ON(atomic_read(&req->count) < 2);
124 atomic_dec(&req->count);
125 }
126
127 static void fuse_req_init_context(struct fuse_req *req)
128 {
129 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
130 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
131 req->in.h.pid = current->pid;
132 }
133
134 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
135 {
136 return !fc->initialized || (for_background && fc->blocked);
137 }
138
139 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
140 bool for_background)
141 {
142 struct fuse_req *req;
143 int err;
144 atomic_inc(&fc->num_waiting);
145
146 if (fuse_block_alloc(fc, for_background)) {
147 sigset_t oldset;
148 int intr;
149
150 block_sigs(&oldset);
151 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
152 !fuse_block_alloc(fc, for_background));
153 restore_sigs(&oldset);
154 err = -EINTR;
155 if (intr)
156 goto out;
157 }
158
159 err = -ENOTCONN;
160 if (!fc->connected)
161 goto out;
162
163 req = fuse_request_alloc(npages);
164 err = -ENOMEM;
165 if (!req) {
166 if (for_background)
167 wake_up(&fc->blocked_waitq);
168 goto out;
169 }
170
171 fuse_req_init_context(req);
172 req->waiting = 1;
173 req->background = for_background;
174 return req;
175
176 out:
177 atomic_dec(&fc->num_waiting);
178 return ERR_PTR(err);
179 }
180
181 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
182 {
183 return __fuse_get_req(fc, npages, false);
184 }
185 EXPORT_SYMBOL_GPL(fuse_get_req);
186
187 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
188 unsigned npages)
189 {
190 return __fuse_get_req(fc, npages, true);
191 }
192 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
193
194 /*
195 * Return request in fuse_file->reserved_req. However that may
196 * currently be in use. If that is the case, wait for it to become
197 * available.
198 */
199 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
200 struct file *file)
201 {
202 struct fuse_req *req = NULL;
203 struct fuse_file *ff = file->private_data;
204
205 do {
206 wait_event(fc->reserved_req_waitq, ff->reserved_req);
207 spin_lock(&fc->lock);
208 if (ff->reserved_req) {
209 req = ff->reserved_req;
210 ff->reserved_req = NULL;
211 req->stolen_file = get_file(file);
212 }
213 spin_unlock(&fc->lock);
214 } while (!req);
215
216 return req;
217 }
218
219 /*
220 * Put stolen request back into fuse_file->reserved_req
221 */
222 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
223 {
224 struct file *file = req->stolen_file;
225 struct fuse_file *ff = file->private_data;
226
227 spin_lock(&fc->lock);
228 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
229 BUG_ON(ff->reserved_req);
230 ff->reserved_req = req;
231 wake_up_all(&fc->reserved_req_waitq);
232 spin_unlock(&fc->lock);
233 fput(file);
234 }
235
236 /*
237 * Gets a requests for a file operation, always succeeds
238 *
239 * This is used for sending the FLUSH request, which must get to
240 * userspace, due to POSIX locks which may need to be unlocked.
241 *
242 * If allocation fails due to OOM, use the reserved request in
243 * fuse_file.
244 *
245 * This is very unlikely to deadlock accidentally, since the
246 * filesystem should not have it's own file open. If deadlock is
247 * intentional, it can still be broken by "aborting" the filesystem.
248 */
249 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
250 struct file *file)
251 {
252 struct fuse_req *req;
253
254 atomic_inc(&fc->num_waiting);
255 wait_event(fc->blocked_waitq, fc->initialized);
256 req = fuse_request_alloc(0);
257 if (!req)
258 req = get_reserved_req(fc, file);
259
260 fuse_req_init_context(req);
261 req->waiting = 1;
262 req->background = 0;
263 return req;
264 }
265
266 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
267 {
268 if (atomic_dec_and_test(&req->count)) {
269 if (unlikely(req->background)) {
270 /*
271 * We get here in the unlikely case that a background
272 * request was allocated but not sent
273 */
274 spin_lock(&fc->lock);
275 if (!fc->blocked)
276 wake_up(&fc->blocked_waitq);
277 spin_unlock(&fc->lock);
278 }
279
280 if (req->waiting)
281 atomic_dec(&fc->num_waiting);
282
283 if (req->stolen_file)
284 put_reserved_req(fc, req);
285 else
286 fuse_request_free(req);
287 }
288 }
289 EXPORT_SYMBOL_GPL(fuse_put_request);
290
291 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
292 {
293 unsigned nbytes = 0;
294 unsigned i;
295
296 for (i = 0; i < numargs; i++)
297 nbytes += args[i].size;
298
299 return nbytes;
300 }
301
302 static u64 fuse_get_unique(struct fuse_conn *fc)
303 {
304 fc->reqctr++;
305 /* zero is special */
306 if (fc->reqctr == 0)
307 fc->reqctr = 1;
308
309 return fc->reqctr;
310 }
311
312 static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
313 {
314 req->in.h.len = sizeof(struct fuse_in_header) +
315 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
316 list_add_tail(&req->list, &fc->pending);
317 req->state = FUSE_REQ_PENDING;
318 if (!req->waiting) {
319 req->waiting = 1;
320 atomic_inc(&fc->num_waiting);
321 }
322 wake_up(&fc->waitq);
323 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
324 }
325
326 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
327 u64 nodeid, u64 nlookup)
328 {
329 forget->forget_one.nodeid = nodeid;
330 forget->forget_one.nlookup = nlookup;
331
332 spin_lock(&fc->lock);
333 if (fc->connected) {
334 fc->forget_list_tail->next = forget;
335 fc->forget_list_tail = forget;
336 wake_up(&fc->waitq);
337 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
338 } else {
339 kfree(forget);
340 }
341 spin_unlock(&fc->lock);
342 }
343
344 static void flush_bg_queue(struct fuse_conn *fc)
345 {
346 while (fc->active_background < fc->max_background &&
347 !list_empty(&fc->bg_queue)) {
348 struct fuse_req *req;
349
350 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
351 list_del(&req->list);
352 fc->active_background++;
353 req->in.h.unique = fuse_get_unique(fc);
354 queue_request(fc, req);
355 }
356 }
357
358 /*
359 * This function is called when a request is finished. Either a reply
360 * has arrived or it was aborted (and not yet sent) or some error
361 * occurred during communication with userspace, or the device file
362 * was closed. The requester thread is woken up (if still waiting),
363 * the 'end' callback is called if given, else the reference to the
364 * request is released
365 *
366 * Called with fc->lock, unlocks it
367 */
368 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
369 __releases(fc->lock)
370 {
371 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
372 req->end = NULL;
373 list_del(&req->list);
374 list_del(&req->intr_entry);
375 req->state = FUSE_REQ_FINISHED;
376 if (req->background) {
377 req->background = 0;
378
379 if (fc->num_background == fc->max_background)
380 fc->blocked = 0;
381
382 /* Wake up next waiter, if any */
383 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
384 wake_up(&fc->blocked_waitq);
385
386 if (fc->num_background == fc->congestion_threshold &&
387 fc->connected && fc->bdi_initialized) {
388 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
389 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
390 }
391 fc->num_background--;
392 fc->active_background--;
393 flush_bg_queue(fc);
394 }
395 spin_unlock(&fc->lock);
396 wake_up(&req->waitq);
397 if (end)
398 end(fc, req);
399 fuse_put_request(fc, req);
400 }
401
402 static void wait_answer_interruptible(struct fuse_conn *fc,
403 struct fuse_req *req)
404 __releases(fc->lock)
405 __acquires(fc->lock)
406 {
407 if (signal_pending(current))
408 return;
409
410 spin_unlock(&fc->lock);
411 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
412 spin_lock(&fc->lock);
413 }
414
415 static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
416 {
417 list_add_tail(&req->intr_entry, &fc->interrupts);
418 wake_up(&fc->waitq);
419 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
420 }
421
422 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
423 __releases(fc->lock)
424 __acquires(fc->lock)
425 {
426 if (!fc->no_interrupt) {
427 /* Any signal may interrupt this */
428 wait_answer_interruptible(fc, req);
429
430 if (req->aborted)
431 goto aborted;
432 if (req->state == FUSE_REQ_FINISHED)
433 return;
434
435 req->interrupted = 1;
436 if (req->state == FUSE_REQ_SENT)
437 queue_interrupt(fc, req);
438 }
439
440 if (!req->force) {
441 sigset_t oldset;
442
443 /* Only fatal signals may interrupt this */
444 block_sigs(&oldset);
445 wait_answer_interruptible(fc, req);
446 restore_sigs(&oldset);
447
448 if (req->aborted)
449 goto aborted;
450 if (req->state == FUSE_REQ_FINISHED)
451 return;
452
453 /* Request is not yet in userspace, bail out */
454 if (req->state == FUSE_REQ_PENDING) {
455 list_del(&req->list);
456 __fuse_put_request(req);
457 req->out.h.error = -EINTR;
458 return;
459 }
460 }
461
462 /*
463 * Either request is already in userspace, or it was forced.
464 * Wait it out.
465 */
466 spin_unlock(&fc->lock);
467 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
468 spin_lock(&fc->lock);
469
470 if (!req->aborted)
471 return;
472
473 aborted:
474 BUG_ON(req->state != FUSE_REQ_FINISHED);
475 if (req->locked) {
476 /* This is uninterruptible sleep, because data is
477 being copied to/from the buffers of req. During
478 locked state, there mustn't be any filesystem
479 operation (e.g. page fault), since that could lead
480 to deadlock */
481 spin_unlock(&fc->lock);
482 wait_event(req->waitq, !req->locked);
483 spin_lock(&fc->lock);
484 }
485 }
486
487 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
488 {
489 BUG_ON(req->background);
490 spin_lock(&fc->lock);
491 if (!fc->connected)
492 req->out.h.error = -ENOTCONN;
493 else if (fc->conn_error)
494 req->out.h.error = -ECONNREFUSED;
495 else {
496 req->in.h.unique = fuse_get_unique(fc);
497 queue_request(fc, req);
498 /* acquire extra reference, since request is still needed
499 after request_end() */
500 __fuse_get_request(req);
501
502 request_wait_answer(fc, req);
503 }
504 spin_unlock(&fc->lock);
505 }
506
507 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
508 {
509 req->isreply = 1;
510 __fuse_request_send(fc, req);
511 }
512 EXPORT_SYMBOL_GPL(fuse_request_send);
513
514 static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
515 struct fuse_req *req)
516 {
517 BUG_ON(!req->background);
518 fc->num_background++;
519 if (fc->num_background == fc->max_background)
520 fc->blocked = 1;
521 if (fc->num_background == fc->congestion_threshold &&
522 fc->bdi_initialized) {
523 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
524 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
525 }
526 list_add_tail(&req->list, &fc->bg_queue);
527 flush_bg_queue(fc);
528 }
529
530 static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
531 {
532 spin_lock(&fc->lock);
533 if (fc->connected) {
534 fuse_request_send_nowait_locked(fc, req);
535 spin_unlock(&fc->lock);
536 } else {
537 req->out.h.error = -ENOTCONN;
538 request_end(fc, req);
539 }
540 }
541
542 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
543 {
544 req->isreply = 1;
545 fuse_request_send_nowait(fc, req);
546 }
547 EXPORT_SYMBOL_GPL(fuse_request_send_background);
548
549 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
550 struct fuse_req *req, u64 unique)
551 {
552 int err = -ENODEV;
553
554 req->isreply = 0;
555 req->in.h.unique = unique;
556 spin_lock(&fc->lock);
557 if (fc->connected) {
558 queue_request(fc, req);
559 err = 0;
560 }
561 spin_unlock(&fc->lock);
562
563 return err;
564 }
565
566 /*
567 * Called under fc->lock
568 *
569 * fc->connected must have been checked previously
570 */
571 void fuse_request_send_background_locked(struct fuse_conn *fc,
572 struct fuse_req *req)
573 {
574 req->isreply = 1;
575 fuse_request_send_nowait_locked(fc, req);
576 }
577
578 void fuse_force_forget(struct file *file, u64 nodeid)
579 {
580 struct inode *inode = file_inode(file);
581 struct fuse_conn *fc = get_fuse_conn(inode);
582 struct fuse_req *req;
583 struct fuse_forget_in inarg;
584
585 memset(&inarg, 0, sizeof(inarg));
586 inarg.nlookup = 1;
587 req = fuse_get_req_nofail_nopages(fc, file);
588 req->in.h.opcode = FUSE_FORGET;
589 req->in.h.nodeid = nodeid;
590 req->in.numargs = 1;
591 req->in.args[0].size = sizeof(inarg);
592 req->in.args[0].value = &inarg;
593 req->isreply = 0;
594 __fuse_request_send(fc, req);
595 /* ignore errors */
596 fuse_put_request(fc, req);
597 }
598
599 /*
600 * Lock the request. Up to the next unlock_request() there mustn't be
601 * anything that could cause a page-fault. If the request was already
602 * aborted bail out.
603 */
604 static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
605 {
606 int err = 0;
607 if (req) {
608 spin_lock(&fc->lock);
609 if (req->aborted)
610 err = -ENOENT;
611 else
612 req->locked = 1;
613 spin_unlock(&fc->lock);
614 }
615 return err;
616 }
617
618 /*
619 * Unlock request. If it was aborted during being locked, the
620 * requester thread is currently waiting for it to be unlocked, so
621 * wake it up.
622 */
623 static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
624 {
625 if (req) {
626 spin_lock(&fc->lock);
627 req->locked = 0;
628 if (req->aborted)
629 wake_up(&req->waitq);
630 spin_unlock(&fc->lock);
631 }
632 }
633
634 struct fuse_copy_state {
635 struct fuse_conn *fc;
636 int write;
637 struct fuse_req *req;
638 const struct iovec *iov;
639 struct pipe_buffer *pipebufs;
640 struct pipe_buffer *currbuf;
641 struct pipe_inode_info *pipe;
642 unsigned long nr_segs;
643 unsigned long seglen;
644 unsigned long addr;
645 struct page *pg;
646 unsigned len;
647 unsigned offset;
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->len = PAGE_SIZE - cs->len;
670 cs->currbuf = NULL;
671 } else if (cs->pg) {
672 if (cs->write) {
673 flush_dcache_page(cs->pg);
674 set_page_dirty_lock(cs->pg);
675 }
676 put_page(cs->pg);
677 }
678 cs->pg = NULL;
679 }
680
681 /*
682 * Get another pagefull of userspace buffer, and map it to kernel
683 * address space, and lock request
684 */
685 static int fuse_copy_fill(struct fuse_copy_state *cs)
686 {
687 struct page *page;
688 int err;
689
690 unlock_request(cs->fc, cs->req);
691 fuse_copy_finish(cs);
692 if (cs->pipebufs) {
693 struct pipe_buffer *buf = cs->pipebufs;
694
695 if (!cs->write) {
696 err = buf->ops->confirm(cs->pipe, buf);
697 if (err)
698 return err;
699
700 BUG_ON(!cs->nr_segs);
701 cs->currbuf = buf;
702 cs->pg = buf->page;
703 cs->offset = buf->offset;
704 cs->len = buf->len;
705 cs->pipebufs++;
706 cs->nr_segs--;
707 } else {
708 if (cs->nr_segs == cs->pipe->buffers)
709 return -EIO;
710
711 page = alloc_page(GFP_HIGHUSER);
712 if (!page)
713 return -ENOMEM;
714
715 buf->page = page;
716 buf->offset = 0;
717 buf->len = 0;
718
719 cs->currbuf = buf;
720 cs->pg = page;
721 cs->offset = 0;
722 cs->len = PAGE_SIZE;
723 cs->pipebufs++;
724 cs->nr_segs++;
725 }
726 } else {
727 if (!cs->seglen) {
728 BUG_ON(!cs->nr_segs);
729 cs->seglen = cs->iov[0].iov_len;
730 cs->addr = (unsigned long) cs->iov[0].iov_base;
731 cs->iov++;
732 cs->nr_segs--;
733 }
734 err = get_user_pages_fast(cs->addr, 1, cs->write, &page);
735 if (err < 0)
736 return err;
737 BUG_ON(err != 1);
738 cs->pg = page;
739 cs->offset = cs->addr % PAGE_SIZE;
740 cs->len = min(PAGE_SIZE - cs->offset, cs->seglen);
741 cs->seglen -= cs->len;
742 cs->addr += cs->len;
743 }
744
745 return lock_request(cs->fc, cs->req);
746 }
747
748 /* Do as much copy to/from userspace buffer as we can */
749 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
750 {
751 unsigned ncpy = min(*size, cs->len);
752 if (val) {
753 void *pgaddr = kmap_atomic(cs->pg);
754 void *buf = pgaddr + cs->offset;
755
756 if (cs->write)
757 memcpy(buf, *val, ncpy);
758 else
759 memcpy(*val, buf, ncpy);
760
761 kunmap_atomic(pgaddr);
762 *val += ncpy;
763 }
764 *size -= ncpy;
765 cs->len -= ncpy;
766 cs->offset += ncpy;
767 return ncpy;
768 }
769
770 static int fuse_check_page(struct page *page)
771 {
772 if (page_mapcount(page) ||
773 page->mapping != NULL ||
774 page_count(page) != 1 ||
775 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
776 ~(1 << PG_locked |
777 1 << PG_referenced |
778 1 << PG_uptodate |
779 1 << PG_lru |
780 1 << PG_active |
781 1 << PG_reclaim))) {
782 printk(KERN_WARNING "fuse: trying to steal weird page\n");
783 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);
784 return 1;
785 }
786 return 0;
787 }
788
789 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
790 {
791 int err;
792 struct page *oldpage = *pagep;
793 struct page *newpage;
794 struct pipe_buffer *buf = cs->pipebufs;
795
796 unlock_request(cs->fc, cs->req);
797 fuse_copy_finish(cs);
798
799 err = buf->ops->confirm(cs->pipe, buf);
800 if (err)
801 return err;
802
803 BUG_ON(!cs->nr_segs);
804 cs->currbuf = buf;
805 cs->len = buf->len;
806 cs->pipebufs++;
807 cs->nr_segs--;
808
809 if (cs->len != PAGE_SIZE)
810 goto out_fallback;
811
812 if (buf->ops->steal(cs->pipe, buf) != 0)
813 goto out_fallback;
814
815 newpage = buf->page;
816
817 if (WARN_ON(!PageUptodate(newpage)))
818 return -EIO;
819
820 ClearPageMappedToDisk(newpage);
821
822 if (fuse_check_page(newpage) != 0)
823 goto out_fallback_unlock;
824
825 /*
826 * This is a new and locked page, it shouldn't be mapped or
827 * have any special flags on it
828 */
829 if (WARN_ON(page_mapped(oldpage)))
830 goto out_fallback_unlock;
831 if (WARN_ON(page_has_private(oldpage)))
832 goto out_fallback_unlock;
833 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
834 goto out_fallback_unlock;
835 if (WARN_ON(PageMlocked(oldpage)))
836 goto out_fallback_unlock;
837
838 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
839 if (err) {
840 unlock_page(newpage);
841 return err;
842 }
843
844 page_cache_get(newpage);
845
846 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
847 lru_cache_add_file(newpage);
848
849 err = 0;
850 spin_lock(&cs->fc->lock);
851 if (cs->req->aborted)
852 err = -ENOENT;
853 else
854 *pagep = newpage;
855 spin_unlock(&cs->fc->lock);
856
857 if (err) {
858 unlock_page(newpage);
859 page_cache_release(newpage);
860 return err;
861 }
862
863 unlock_page(oldpage);
864 page_cache_release(oldpage);
865 cs->len = 0;
866
867 return 0;
868
869 out_fallback_unlock:
870 unlock_page(newpage);
871 out_fallback:
872 cs->pg = buf->page;
873 cs->offset = buf->offset;
874
875 err = lock_request(cs->fc, cs->req);
876 if (err)
877 return err;
878
879 return 1;
880 }
881
882 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
883 unsigned offset, unsigned count)
884 {
885 struct pipe_buffer *buf;
886
887 if (cs->nr_segs == cs->pipe->buffers)
888 return -EIO;
889
890 unlock_request(cs->fc, cs->req);
891 fuse_copy_finish(cs);
892
893 buf = cs->pipebufs;
894 page_cache_get(page);
895 buf->page = page;
896 buf->offset = offset;
897 buf->len = count;
898
899 cs->pipebufs++;
900 cs->nr_segs++;
901 cs->len = 0;
902
903 return 0;
904 }
905
906 /*
907 * Copy a page in the request to/from the userspace buffer. Must be
908 * done atomically
909 */
910 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
911 unsigned offset, unsigned count, int zeroing)
912 {
913 int err;
914 struct page *page = *pagep;
915
916 if (page && zeroing && count < PAGE_SIZE)
917 clear_highpage(page);
918
919 while (count) {
920 if (cs->write && cs->pipebufs && page) {
921 return fuse_ref_page(cs, page, offset, count);
922 } else if (!cs->len) {
923 if (cs->move_pages && page &&
924 offset == 0 && count == PAGE_SIZE) {
925 err = fuse_try_move_page(cs, pagep);
926 if (err <= 0)
927 return err;
928 } else {
929 err = fuse_copy_fill(cs);
930 if (err)
931 return err;
932 }
933 }
934 if (page) {
935 void *mapaddr = kmap_atomic(page);
936 void *buf = mapaddr + offset;
937 offset += fuse_copy_do(cs, &buf, &count);
938 kunmap_atomic(mapaddr);
939 } else
940 offset += fuse_copy_do(cs, NULL, &count);
941 }
942 if (page && !cs->write)
943 flush_dcache_page(page);
944 return 0;
945 }
946
947 /* Copy pages in the request to/from userspace buffer */
948 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
949 int zeroing)
950 {
951 unsigned i;
952 struct fuse_req *req = cs->req;
953
954 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
955 int err;
956 unsigned offset = req->page_descs[i].offset;
957 unsigned count = min(nbytes, req->page_descs[i].length);
958
959 err = fuse_copy_page(cs, &req->pages[i], offset, count,
960 zeroing);
961 if (err)
962 return err;
963
964 nbytes -= count;
965 }
966 return 0;
967 }
968
969 /* Copy a single argument in the request to/from userspace buffer */
970 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
971 {
972 while (size) {
973 if (!cs->len) {
974 int err = fuse_copy_fill(cs);
975 if (err)
976 return err;
977 }
978 fuse_copy_do(cs, &val, &size);
979 }
980 return 0;
981 }
982
983 /* Copy request arguments to/from userspace buffer */
984 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
985 unsigned argpages, struct fuse_arg *args,
986 int zeroing)
987 {
988 int err = 0;
989 unsigned i;
990
991 for (i = 0; !err && i < numargs; i++) {
992 struct fuse_arg *arg = &args[i];
993 if (i == numargs - 1 && argpages)
994 err = fuse_copy_pages(cs, arg->size, zeroing);
995 else
996 err = fuse_copy_one(cs, arg->value, arg->size);
997 }
998 return err;
999 }
1000
1001 static int forget_pending(struct fuse_conn *fc)
1002 {
1003 return fc->forget_list_head.next != NULL;
1004 }
1005
1006 static int request_pending(struct fuse_conn *fc)
1007 {
1008 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
1009 forget_pending(fc);
1010 }
1011
1012 /* Wait until a request is available on the pending list */
1013 static void request_wait(struct fuse_conn *fc)
1014 __releases(fc->lock)
1015 __acquires(fc->lock)
1016 {
1017 DECLARE_WAITQUEUE(wait, current);
1018
1019 add_wait_queue_exclusive(&fc->waitq, &wait);
1020 while (fc->connected && !request_pending(fc)) {
1021 set_current_state(TASK_INTERRUPTIBLE);
1022 if (signal_pending(current))
1023 break;
1024
1025 spin_unlock(&fc->lock);
1026 schedule();
1027 spin_lock(&fc->lock);
1028 }
1029 set_current_state(TASK_RUNNING);
1030 remove_wait_queue(&fc->waitq, &wait);
1031 }
1032
1033 /*
1034 * Transfer an interrupt request to userspace
1035 *
1036 * Unlike other requests this is assembled on demand, without a need
1037 * to allocate a separate fuse_req structure.
1038 *
1039 * Called with fc->lock held, releases it
1040 */
1041 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1042 size_t nbytes, struct fuse_req *req)
1043 __releases(fc->lock)
1044 {
1045 struct fuse_in_header ih;
1046 struct fuse_interrupt_in arg;
1047 unsigned reqsize = sizeof(ih) + sizeof(arg);
1048 int err;
1049
1050 list_del_init(&req->intr_entry);
1051 req->intr_unique = fuse_get_unique(fc);
1052 memset(&ih, 0, sizeof(ih));
1053 memset(&arg, 0, sizeof(arg));
1054 ih.len = reqsize;
1055 ih.opcode = FUSE_INTERRUPT;
1056 ih.unique = req->intr_unique;
1057 arg.unique = req->in.h.unique;
1058
1059 spin_unlock(&fc->lock);
1060 if (nbytes < reqsize)
1061 return -EINVAL;
1062
1063 err = fuse_copy_one(cs, &ih, sizeof(ih));
1064 if (!err)
1065 err = fuse_copy_one(cs, &arg, sizeof(arg));
1066 fuse_copy_finish(cs);
1067
1068 return err ? err : reqsize;
1069 }
1070
1071 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
1072 unsigned max,
1073 unsigned *countp)
1074 {
1075 struct fuse_forget_link *head = fc->forget_list_head.next;
1076 struct fuse_forget_link **newhead = &head;
1077 unsigned count;
1078
1079 for (count = 0; *newhead != NULL && count < max; count++)
1080 newhead = &(*newhead)->next;
1081
1082 fc->forget_list_head.next = *newhead;
1083 *newhead = NULL;
1084 if (fc->forget_list_head.next == NULL)
1085 fc->forget_list_tail = &fc->forget_list_head;
1086
1087 if (countp != NULL)
1088 *countp = count;
1089
1090 return head;
1091 }
1092
1093 static int fuse_read_single_forget(struct fuse_conn *fc,
1094 struct fuse_copy_state *cs,
1095 size_t nbytes)
1096 __releases(fc->lock)
1097 {
1098 int err;
1099 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1100 struct fuse_forget_in arg = {
1101 .nlookup = forget->forget_one.nlookup,
1102 };
1103 struct fuse_in_header ih = {
1104 .opcode = FUSE_FORGET,
1105 .nodeid = forget->forget_one.nodeid,
1106 .unique = fuse_get_unique(fc),
1107 .len = sizeof(ih) + sizeof(arg),
1108 };
1109
1110 spin_unlock(&fc->lock);
1111 kfree(forget);
1112 if (nbytes < ih.len)
1113 return -EINVAL;
1114
1115 err = fuse_copy_one(cs, &ih, sizeof(ih));
1116 if (!err)
1117 err = fuse_copy_one(cs, &arg, sizeof(arg));
1118 fuse_copy_finish(cs);
1119
1120 if (err)
1121 return err;
1122
1123 return ih.len;
1124 }
1125
1126 static int fuse_read_batch_forget(struct fuse_conn *fc,
1127 struct fuse_copy_state *cs, size_t nbytes)
1128 __releases(fc->lock)
1129 {
1130 int err;
1131 unsigned max_forgets;
1132 unsigned count;
1133 struct fuse_forget_link *head;
1134 struct fuse_batch_forget_in arg = { .count = 0 };
1135 struct fuse_in_header ih = {
1136 .opcode = FUSE_BATCH_FORGET,
1137 .unique = fuse_get_unique(fc),
1138 .len = sizeof(ih) + sizeof(arg),
1139 };
1140
1141 if (nbytes < ih.len) {
1142 spin_unlock(&fc->lock);
1143 return -EINVAL;
1144 }
1145
1146 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1147 head = dequeue_forget(fc, max_forgets, &count);
1148 spin_unlock(&fc->lock);
1149
1150 arg.count = count;
1151 ih.len += count * sizeof(struct fuse_forget_one);
1152 err = fuse_copy_one(cs, &ih, sizeof(ih));
1153 if (!err)
1154 err = fuse_copy_one(cs, &arg, sizeof(arg));
1155
1156 while (head) {
1157 struct fuse_forget_link *forget = head;
1158
1159 if (!err) {
1160 err = fuse_copy_one(cs, &forget->forget_one,
1161 sizeof(forget->forget_one));
1162 }
1163 head = forget->next;
1164 kfree(forget);
1165 }
1166
1167 fuse_copy_finish(cs);
1168
1169 if (err)
1170 return err;
1171
1172 return ih.len;
1173 }
1174
1175 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1176 size_t nbytes)
1177 __releases(fc->lock)
1178 {
1179 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1180 return fuse_read_single_forget(fc, cs, nbytes);
1181 else
1182 return fuse_read_batch_forget(fc, cs, nbytes);
1183 }
1184
1185 /*
1186 * Read a single request into the userspace filesystem's buffer. This
1187 * function waits until a request is available, then removes it from
1188 * the pending list and copies request data to userspace buffer. If
1189 * no reply is needed (FORGET) or request has been aborted or there
1190 * was an error during the copying then it's finished by calling
1191 * request_end(). Otherwise add it to the processing list, and set
1192 * the 'sent' flag.
1193 */
1194 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1195 struct fuse_copy_state *cs, size_t nbytes)
1196 {
1197 int err;
1198 struct fuse_req *req;
1199 struct fuse_in *in;
1200 unsigned reqsize;
1201
1202 restart:
1203 spin_lock(&fc->lock);
1204 err = -EAGAIN;
1205 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1206 !request_pending(fc))
1207 goto err_unlock;
1208
1209 request_wait(fc);
1210 err = -ENODEV;
1211 if (!fc->connected)
1212 goto err_unlock;
1213 err = -ERESTARTSYS;
1214 if (!request_pending(fc))
1215 goto err_unlock;
1216
1217 if (!list_empty(&fc->interrupts)) {
1218 req = list_entry(fc->interrupts.next, struct fuse_req,
1219 intr_entry);
1220 return fuse_read_interrupt(fc, cs, nbytes, req);
1221 }
1222
1223 if (forget_pending(fc)) {
1224 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1225 return fuse_read_forget(fc, cs, nbytes);
1226
1227 if (fc->forget_batch <= -8)
1228 fc->forget_batch = 16;
1229 }
1230
1231 req = list_entry(fc->pending.next, struct fuse_req, list);
1232 req->state = FUSE_REQ_READING;
1233 list_move(&req->list, &fc->io);
1234
1235 in = &req->in;
1236 reqsize = in->h.len;
1237 /* If request is too large, reply with an error and restart the read */
1238 if (nbytes < reqsize) {
1239 req->out.h.error = -EIO;
1240 /* SETXATTR is special, since it may contain too large data */
1241 if (in->h.opcode == FUSE_SETXATTR)
1242 req->out.h.error = -E2BIG;
1243 request_end(fc, req);
1244 goto restart;
1245 }
1246 spin_unlock(&fc->lock);
1247 cs->req = req;
1248 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1249 if (!err)
1250 err = fuse_copy_args(cs, in->numargs, in->argpages,
1251 (struct fuse_arg *) in->args, 0);
1252 fuse_copy_finish(cs);
1253 spin_lock(&fc->lock);
1254 req->locked = 0;
1255 if (req->aborted) {
1256 request_end(fc, req);
1257 return -ENODEV;
1258 }
1259 if (err) {
1260 req->out.h.error = -EIO;
1261 request_end(fc, req);
1262 return err;
1263 }
1264 if (!req->isreply)
1265 request_end(fc, req);
1266 else {
1267 req->state = FUSE_REQ_SENT;
1268 list_move_tail(&req->list, &fc->processing);
1269 if (req->interrupted)
1270 queue_interrupt(fc, req);
1271 spin_unlock(&fc->lock);
1272 }
1273 return reqsize;
1274
1275 err_unlock:
1276 spin_unlock(&fc->lock);
1277 return err;
1278 }
1279
1280 static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1281 unsigned long nr_segs, loff_t pos)
1282 {
1283 struct fuse_copy_state cs;
1284 struct file *file = iocb->ki_filp;
1285 struct fuse_conn *fc = fuse_get_conn(file);
1286 if (!fc)
1287 return -EPERM;
1288
1289 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1290
1291 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1292 }
1293
1294 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1295 struct pipe_inode_info *pipe,
1296 size_t len, unsigned int flags)
1297 {
1298 int ret;
1299 int page_nr = 0;
1300 int do_wakeup = 0;
1301 struct pipe_buffer *bufs;
1302 struct fuse_copy_state cs;
1303 struct fuse_conn *fc = fuse_get_conn(in);
1304 if (!fc)
1305 return -EPERM;
1306
1307 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1308 if (!bufs)
1309 return -ENOMEM;
1310
1311 fuse_copy_init(&cs, fc, 1, NULL, 0);
1312 cs.pipebufs = bufs;
1313 cs.pipe = pipe;
1314 ret = fuse_dev_do_read(fc, in, &cs, len);
1315 if (ret < 0)
1316 goto out;
1317
1318 ret = 0;
1319 pipe_lock(pipe);
1320
1321 if (!pipe->readers) {
1322 send_sig(SIGPIPE, current, 0);
1323 if (!ret)
1324 ret = -EPIPE;
1325 goto out_unlock;
1326 }
1327
1328 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1329 ret = -EIO;
1330 goto out_unlock;
1331 }
1332
1333 while (page_nr < cs.nr_segs) {
1334 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1335 struct pipe_buffer *buf = pipe->bufs + newbuf;
1336
1337 buf->page = bufs[page_nr].page;
1338 buf->offset = bufs[page_nr].offset;
1339 buf->len = bufs[page_nr].len;
1340 /*
1341 * Need to be careful about this. Having buf->ops in module
1342 * code can Oops if the buffer persists after module unload.
1343 */
1344 buf->ops = &nosteal_pipe_buf_ops;
1345
1346 pipe->nrbufs++;
1347 page_nr++;
1348 ret += buf->len;
1349
1350 if (pipe->files)
1351 do_wakeup = 1;
1352 }
1353
1354 out_unlock:
1355 pipe_unlock(pipe);
1356
1357 if (do_wakeup) {
1358 smp_mb();
1359 if (waitqueue_active(&pipe->wait))
1360 wake_up_interruptible(&pipe->wait);
1361 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1362 }
1363
1364 out:
1365 for (; page_nr < cs.nr_segs; page_nr++)
1366 page_cache_release(bufs[page_nr].page);
1367
1368 kfree(bufs);
1369 return ret;
1370 }
1371
1372 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1373 struct fuse_copy_state *cs)
1374 {
1375 struct fuse_notify_poll_wakeup_out outarg;
1376 int err = -EINVAL;
1377
1378 if (size != sizeof(outarg))
1379 goto err;
1380
1381 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1382 if (err)
1383 goto err;
1384
1385 fuse_copy_finish(cs);
1386 return fuse_notify_poll_wakeup(fc, &outarg);
1387
1388 err:
1389 fuse_copy_finish(cs);
1390 return err;
1391 }
1392
1393 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1394 struct fuse_copy_state *cs)
1395 {
1396 struct fuse_notify_inval_inode_out outarg;
1397 int err = -EINVAL;
1398
1399 if (size != sizeof(outarg))
1400 goto err;
1401
1402 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1403 if (err)
1404 goto err;
1405 fuse_copy_finish(cs);
1406
1407 down_read(&fc->killsb);
1408 err = -ENOENT;
1409 if (fc->sb) {
1410 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1411 outarg.off, outarg.len);
1412 }
1413 up_read(&fc->killsb);
1414 return err;
1415
1416 err:
1417 fuse_copy_finish(cs);
1418 return err;
1419 }
1420
1421 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1422 struct fuse_copy_state *cs)
1423 {
1424 struct fuse_notify_inval_entry_out outarg;
1425 int err = -ENOMEM;
1426 char *buf;
1427 struct qstr name;
1428
1429 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1430 if (!buf)
1431 goto err;
1432
1433 err = -EINVAL;
1434 if (size < sizeof(outarg))
1435 goto err;
1436
1437 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1438 if (err)
1439 goto err;
1440
1441 err = -ENAMETOOLONG;
1442 if (outarg.namelen > FUSE_NAME_MAX)
1443 goto err;
1444
1445 err = -EINVAL;
1446 if (size != sizeof(outarg) + outarg.namelen + 1)
1447 goto err;
1448
1449 name.name = buf;
1450 name.len = outarg.namelen;
1451 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1452 if (err)
1453 goto err;
1454 fuse_copy_finish(cs);
1455 buf[outarg.namelen] = 0;
1456 name.hash = full_name_hash(name.name, name.len);
1457
1458 down_read(&fc->killsb);
1459 err = -ENOENT;
1460 if (fc->sb)
1461 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1462 up_read(&fc->killsb);
1463 kfree(buf);
1464 return err;
1465
1466 err:
1467 kfree(buf);
1468 fuse_copy_finish(cs);
1469 return err;
1470 }
1471
1472 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1473 struct fuse_copy_state *cs)
1474 {
1475 struct fuse_notify_delete_out outarg;
1476 int err = -ENOMEM;
1477 char *buf;
1478 struct qstr name;
1479
1480 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1481 if (!buf)
1482 goto err;
1483
1484 err = -EINVAL;
1485 if (size < sizeof(outarg))
1486 goto err;
1487
1488 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1489 if (err)
1490 goto err;
1491
1492 err = -ENAMETOOLONG;
1493 if (outarg.namelen > FUSE_NAME_MAX)
1494 goto err;
1495
1496 err = -EINVAL;
1497 if (size != sizeof(outarg) + outarg.namelen + 1)
1498 goto err;
1499
1500 name.name = buf;
1501 name.len = outarg.namelen;
1502 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1503 if (err)
1504 goto err;
1505 fuse_copy_finish(cs);
1506 buf[outarg.namelen] = 0;
1507 name.hash = full_name_hash(name.name, name.len);
1508
1509 down_read(&fc->killsb);
1510 err = -ENOENT;
1511 if (fc->sb)
1512 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1513 outarg.child, &name);
1514 up_read(&fc->killsb);
1515 kfree(buf);
1516 return err;
1517
1518 err:
1519 kfree(buf);
1520 fuse_copy_finish(cs);
1521 return err;
1522 }
1523
1524 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1525 struct fuse_copy_state *cs)
1526 {
1527 struct fuse_notify_store_out outarg;
1528 struct inode *inode;
1529 struct address_space *mapping;
1530 u64 nodeid;
1531 int err;
1532 pgoff_t index;
1533 unsigned int offset;
1534 unsigned int num;
1535 loff_t file_size;
1536 loff_t end;
1537
1538 err = -EINVAL;
1539 if (size < sizeof(outarg))
1540 goto out_finish;
1541
1542 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1543 if (err)
1544 goto out_finish;
1545
1546 err = -EINVAL;
1547 if (size - sizeof(outarg) != outarg.size)
1548 goto out_finish;
1549
1550 nodeid = outarg.nodeid;
1551
1552 down_read(&fc->killsb);
1553
1554 err = -ENOENT;
1555 if (!fc->sb)
1556 goto out_up_killsb;
1557
1558 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1559 if (!inode)
1560 goto out_up_killsb;
1561
1562 mapping = inode->i_mapping;
1563 index = outarg.offset >> PAGE_CACHE_SHIFT;
1564 offset = outarg.offset & ~PAGE_CACHE_MASK;
1565 file_size = i_size_read(inode);
1566 end = outarg.offset + outarg.size;
1567 if (end > file_size) {
1568 file_size = end;
1569 fuse_write_update_size(inode, file_size);
1570 }
1571
1572 num = outarg.size;
1573 while (num) {
1574 struct page *page;
1575 unsigned int this_num;
1576
1577 err = -ENOMEM;
1578 page = find_or_create_page(mapping, index,
1579 mapping_gfp_mask(mapping));
1580 if (!page)
1581 goto out_iput;
1582
1583 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1584 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1585 if (!err && offset == 0 &&
1586 (this_num == PAGE_CACHE_SIZE || file_size == end))
1587 SetPageUptodate(page);
1588 unlock_page(page);
1589 page_cache_release(page);
1590
1591 if (err)
1592 goto out_iput;
1593
1594 num -= this_num;
1595 offset = 0;
1596 index++;
1597 }
1598
1599 err = 0;
1600
1601 out_iput:
1602 iput(inode);
1603 out_up_killsb:
1604 up_read(&fc->killsb);
1605 out_finish:
1606 fuse_copy_finish(cs);
1607 return err;
1608 }
1609
1610 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1611 {
1612 release_pages(req->pages, req->num_pages, false);
1613 }
1614
1615 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1616 struct fuse_notify_retrieve_out *outarg)
1617 {
1618 int err;
1619 struct address_space *mapping = inode->i_mapping;
1620 struct fuse_req *req;
1621 pgoff_t index;
1622 loff_t file_size;
1623 unsigned int num;
1624 unsigned int offset;
1625 size_t total_len = 0;
1626 int num_pages;
1627
1628 offset = outarg->offset & ~PAGE_CACHE_MASK;
1629 file_size = i_size_read(inode);
1630
1631 num = outarg->size;
1632 if (outarg->offset > file_size)
1633 num = 0;
1634 else if (outarg->offset + num > file_size)
1635 num = file_size - outarg->offset;
1636
1637 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1638 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1639
1640 req = fuse_get_req(fc, num_pages);
1641 if (IS_ERR(req))
1642 return PTR_ERR(req);
1643
1644 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1645 req->in.h.nodeid = outarg->nodeid;
1646 req->in.numargs = 2;
1647 req->in.argpages = 1;
1648 req->page_descs[0].offset = offset;
1649 req->end = fuse_retrieve_end;
1650
1651 index = outarg->offset >> PAGE_CACHE_SHIFT;
1652
1653 while (num && req->num_pages < num_pages) {
1654 struct page *page;
1655 unsigned int this_num;
1656
1657 page = find_get_page(mapping, index);
1658 if (!page)
1659 break;
1660
1661 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1662 req->pages[req->num_pages] = page;
1663 req->page_descs[req->num_pages].length = this_num;
1664 req->num_pages++;
1665
1666 offset = 0;
1667 num -= this_num;
1668 total_len += this_num;
1669 index++;
1670 }
1671 req->misc.retrieve_in.offset = outarg->offset;
1672 req->misc.retrieve_in.size = total_len;
1673 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1674 req->in.args[0].value = &req->misc.retrieve_in;
1675 req->in.args[1].size = total_len;
1676
1677 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1678 if (err)
1679 fuse_retrieve_end(fc, req);
1680
1681 return err;
1682 }
1683
1684 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1685 struct fuse_copy_state *cs)
1686 {
1687 struct fuse_notify_retrieve_out outarg;
1688 struct inode *inode;
1689 int err;
1690
1691 err = -EINVAL;
1692 if (size != sizeof(outarg))
1693 goto copy_finish;
1694
1695 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1696 if (err)
1697 goto copy_finish;
1698
1699 fuse_copy_finish(cs);
1700
1701 down_read(&fc->killsb);
1702 err = -ENOENT;
1703 if (fc->sb) {
1704 u64 nodeid = outarg.nodeid;
1705
1706 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1707 if (inode) {
1708 err = fuse_retrieve(fc, inode, &outarg);
1709 iput(inode);
1710 }
1711 }
1712 up_read(&fc->killsb);
1713
1714 return err;
1715
1716 copy_finish:
1717 fuse_copy_finish(cs);
1718 return err;
1719 }
1720
1721 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1722 unsigned int size, struct fuse_copy_state *cs)
1723 {
1724 switch (code) {
1725 case FUSE_NOTIFY_POLL:
1726 return fuse_notify_poll(fc, size, cs);
1727
1728 case FUSE_NOTIFY_INVAL_INODE:
1729 return fuse_notify_inval_inode(fc, size, cs);
1730
1731 case FUSE_NOTIFY_INVAL_ENTRY:
1732 return fuse_notify_inval_entry(fc, size, cs);
1733
1734 case FUSE_NOTIFY_STORE:
1735 return fuse_notify_store(fc, size, cs);
1736
1737 case FUSE_NOTIFY_RETRIEVE:
1738 return fuse_notify_retrieve(fc, size, cs);
1739
1740 case FUSE_NOTIFY_DELETE:
1741 return fuse_notify_delete(fc, size, cs);
1742
1743 default:
1744 fuse_copy_finish(cs);
1745 return -EINVAL;
1746 }
1747 }
1748
1749 /* Look up request on processing list by unique ID */
1750 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1751 {
1752 struct fuse_req *req;
1753
1754 list_for_each_entry(req, &fc->processing, list) {
1755 if (req->in.h.unique == unique || req->intr_unique == unique)
1756 return req;
1757 }
1758 return NULL;
1759 }
1760
1761 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1762 unsigned nbytes)
1763 {
1764 unsigned reqsize = sizeof(struct fuse_out_header);
1765
1766 if (out->h.error)
1767 return nbytes != reqsize ? -EINVAL : 0;
1768
1769 reqsize += len_args(out->numargs, out->args);
1770
1771 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1772 return -EINVAL;
1773 else if (reqsize > nbytes) {
1774 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1775 unsigned diffsize = reqsize - nbytes;
1776 if (diffsize > lastarg->size)
1777 return -EINVAL;
1778 lastarg->size -= diffsize;
1779 }
1780 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1781 out->page_zeroing);
1782 }
1783
1784 /*
1785 * Write a single reply to a request. First the header is copied from
1786 * the write buffer. The request is then searched on the processing
1787 * list by the unique ID found in the header. If found, then remove
1788 * it from the list and copy the rest of the buffer to the request.
1789 * The request is finished by calling request_end()
1790 */
1791 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1792 struct fuse_copy_state *cs, size_t nbytes)
1793 {
1794 int err;
1795 struct fuse_req *req;
1796 struct fuse_out_header oh;
1797
1798 if (nbytes < sizeof(struct fuse_out_header))
1799 return -EINVAL;
1800
1801 err = fuse_copy_one(cs, &oh, sizeof(oh));
1802 if (err)
1803 goto err_finish;
1804
1805 err = -EINVAL;
1806 if (oh.len != nbytes)
1807 goto err_finish;
1808
1809 /*
1810 * Zero oh.unique indicates unsolicited notification message
1811 * and error contains notification code.
1812 */
1813 if (!oh.unique) {
1814 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1815 return err ? err : nbytes;
1816 }
1817
1818 err = -EINVAL;
1819 if (oh.error <= -1000 || oh.error > 0)
1820 goto err_finish;
1821
1822 spin_lock(&fc->lock);
1823 err = -ENOENT;
1824 if (!fc->connected)
1825 goto err_unlock;
1826
1827 req = request_find(fc, oh.unique);
1828 if (!req)
1829 goto err_unlock;
1830
1831 if (req->aborted) {
1832 spin_unlock(&fc->lock);
1833 fuse_copy_finish(cs);
1834 spin_lock(&fc->lock);
1835 request_end(fc, req);
1836 return -ENOENT;
1837 }
1838 /* Is it an interrupt reply? */
1839 if (req->intr_unique == oh.unique) {
1840 err = -EINVAL;
1841 if (nbytes != sizeof(struct fuse_out_header))
1842 goto err_unlock;
1843
1844 if (oh.error == -ENOSYS)
1845 fc->no_interrupt = 1;
1846 else if (oh.error == -EAGAIN)
1847 queue_interrupt(fc, req);
1848
1849 spin_unlock(&fc->lock);
1850 fuse_copy_finish(cs);
1851 return nbytes;
1852 }
1853
1854 req->state = FUSE_REQ_WRITING;
1855 list_move(&req->list, &fc->io);
1856 req->out.h = oh;
1857 req->locked = 1;
1858 cs->req = req;
1859 if (!req->out.page_replace)
1860 cs->move_pages = 0;
1861 spin_unlock(&fc->lock);
1862
1863 err = copy_out_args(cs, &req->out, nbytes);
1864 fuse_copy_finish(cs);
1865
1866 spin_lock(&fc->lock);
1867 req->locked = 0;
1868 if (!err) {
1869 if (req->aborted)
1870 err = -ENOENT;
1871 } else if (!req->aborted)
1872 req->out.h.error = -EIO;
1873 request_end(fc, req);
1874
1875 return err ? err : nbytes;
1876
1877 err_unlock:
1878 spin_unlock(&fc->lock);
1879 err_finish:
1880 fuse_copy_finish(cs);
1881 return err;
1882 }
1883
1884 static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1885 unsigned long nr_segs, loff_t pos)
1886 {
1887 struct fuse_copy_state cs;
1888 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1889 if (!fc)
1890 return -EPERM;
1891
1892 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1893
1894 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1895 }
1896
1897 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1898 struct file *out, loff_t *ppos,
1899 size_t len, unsigned int flags)
1900 {
1901 unsigned nbuf;
1902 unsigned idx;
1903 struct pipe_buffer *bufs;
1904 struct fuse_copy_state cs;
1905 struct fuse_conn *fc;
1906 size_t rem;
1907 ssize_t ret;
1908
1909 fc = fuse_get_conn(out);
1910 if (!fc)
1911 return -EPERM;
1912
1913 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1914 if (!bufs)
1915 return -ENOMEM;
1916
1917 pipe_lock(pipe);
1918 nbuf = 0;
1919 rem = 0;
1920 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1921 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1922
1923 ret = -EINVAL;
1924 if (rem < len) {
1925 pipe_unlock(pipe);
1926 goto out;
1927 }
1928
1929 rem = len;
1930 while (rem) {
1931 struct pipe_buffer *ibuf;
1932 struct pipe_buffer *obuf;
1933
1934 BUG_ON(nbuf >= pipe->buffers);
1935 BUG_ON(!pipe->nrbufs);
1936 ibuf = &pipe->bufs[pipe->curbuf];
1937 obuf = &bufs[nbuf];
1938
1939 if (rem >= ibuf->len) {
1940 *obuf = *ibuf;
1941 ibuf->ops = NULL;
1942 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1943 pipe->nrbufs--;
1944 } else {
1945 ibuf->ops->get(pipe, ibuf);
1946 *obuf = *ibuf;
1947 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1948 obuf->len = rem;
1949 ibuf->offset += obuf->len;
1950 ibuf->len -= obuf->len;
1951 }
1952 nbuf++;
1953 rem -= obuf->len;
1954 }
1955 pipe_unlock(pipe);
1956
1957 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1958 cs.pipebufs = bufs;
1959 cs.pipe = pipe;
1960
1961 if (flags & SPLICE_F_MOVE)
1962 cs.move_pages = 1;
1963
1964 ret = fuse_dev_do_write(fc, &cs, len);
1965
1966 for (idx = 0; idx < nbuf; idx++) {
1967 struct pipe_buffer *buf = &bufs[idx];
1968 buf->ops->release(pipe, buf);
1969 }
1970 out:
1971 kfree(bufs);
1972 return ret;
1973 }
1974
1975 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1976 {
1977 unsigned mask = POLLOUT | POLLWRNORM;
1978 struct fuse_conn *fc = fuse_get_conn(file);
1979 if (!fc)
1980 return POLLERR;
1981
1982 poll_wait(file, &fc->waitq, wait);
1983
1984 spin_lock(&fc->lock);
1985 if (!fc->connected)
1986 mask = POLLERR;
1987 else if (request_pending(fc))
1988 mask |= POLLIN | POLLRDNORM;
1989 spin_unlock(&fc->lock);
1990
1991 return mask;
1992 }
1993
1994 /*
1995 * Abort all requests on the given list (pending or processing)
1996 *
1997 * This function releases and reacquires fc->lock
1998 */
1999 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2000 __releases(fc->lock)
2001 __acquires(fc->lock)
2002 {
2003 while (!list_empty(head)) {
2004 struct fuse_req *req;
2005 req = list_entry(head->next, struct fuse_req, list);
2006 req->out.h.error = -ECONNABORTED;
2007 request_end(fc, req);
2008 spin_lock(&fc->lock);
2009 }
2010 }
2011
2012 /*
2013 * Abort requests under I/O
2014 *
2015 * The requests are set to aborted and finished, and the request
2016 * waiter is woken up. This will make request_wait_answer() wait
2017 * until the request is unlocked and then return.
2018 *
2019 * If the request is asynchronous, then the end function needs to be
2020 * called after waiting for the request to be unlocked (if it was
2021 * locked).
2022 */
2023 static void end_io_requests(struct fuse_conn *fc)
2024 __releases(fc->lock)
2025 __acquires(fc->lock)
2026 {
2027 while (!list_empty(&fc->io)) {
2028 struct fuse_req *req =
2029 list_entry(fc->io.next, struct fuse_req, list);
2030 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
2031
2032 req->aborted = 1;
2033 req->out.h.error = -ECONNABORTED;
2034 req->state = FUSE_REQ_FINISHED;
2035 list_del_init(&req->list);
2036 wake_up(&req->waitq);
2037 if (end) {
2038 req->end = NULL;
2039 __fuse_get_request(req);
2040 spin_unlock(&fc->lock);
2041 wait_event(req->waitq, !req->locked);
2042 end(fc, req);
2043 fuse_put_request(fc, req);
2044 spin_lock(&fc->lock);
2045 }
2046 }
2047 }
2048
2049 static void end_queued_requests(struct fuse_conn *fc)
2050 __releases(fc->lock)
2051 __acquires(fc->lock)
2052 {
2053 fc->max_background = UINT_MAX;
2054 flush_bg_queue(fc);
2055 end_requests(fc, &fc->pending);
2056 end_requests(fc, &fc->processing);
2057 while (forget_pending(fc))
2058 kfree(dequeue_forget(fc, 1, NULL));
2059 }
2060
2061 static void end_polls(struct fuse_conn *fc)
2062 {
2063 struct rb_node *p;
2064
2065 p = rb_first(&fc->polled_files);
2066
2067 while (p) {
2068 struct fuse_file *ff;
2069 ff = rb_entry(p, struct fuse_file, polled_node);
2070 wake_up_interruptible_all(&ff->poll_wait);
2071
2072 p = rb_next(p);
2073 }
2074 }
2075
2076 /*
2077 * Abort all requests.
2078 *
2079 * Emergency exit in case of a malicious or accidental deadlock, or
2080 * just a hung filesystem.
2081 *
2082 * The same effect is usually achievable through killing the
2083 * filesystem daemon and all users of the filesystem. The exception
2084 * is the combination of an asynchronous request and the tricky
2085 * deadlock (see Documentation/filesystems/fuse.txt).
2086 *
2087 * During the aborting, progression of requests from the pending and
2088 * processing lists onto the io list, and progression of new requests
2089 * onto the pending list is prevented by req->connected being false.
2090 *
2091 * Progression of requests under I/O to the processing list is
2092 * prevented by the req->aborted flag being true for these requests.
2093 * For this reason requests on the io list must be aborted first.
2094 */
2095 void fuse_abort_conn(struct fuse_conn *fc)
2096 {
2097 spin_lock(&fc->lock);
2098 if (fc->connected) {
2099 fc->connected = 0;
2100 fc->blocked = 0;
2101 fc->initialized = 1;
2102 end_io_requests(fc);
2103 end_queued_requests(fc);
2104 end_polls(fc);
2105 wake_up_all(&fc->waitq);
2106 wake_up_all(&fc->blocked_waitq);
2107 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2108 }
2109 spin_unlock(&fc->lock);
2110 }
2111 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2112
2113 int fuse_dev_release(struct inode *inode, struct file *file)
2114 {
2115 struct fuse_conn *fc = fuse_get_conn(file);
2116 if (fc) {
2117 spin_lock(&fc->lock);
2118 fc->connected = 0;
2119 fc->blocked = 0;
2120 fc->initialized = 1;
2121 end_queued_requests(fc);
2122 end_polls(fc);
2123 wake_up_all(&fc->blocked_waitq);
2124 spin_unlock(&fc->lock);
2125 fuse_conn_put(fc);
2126 }
2127
2128 return 0;
2129 }
2130 EXPORT_SYMBOL_GPL(fuse_dev_release);
2131
2132 static int fuse_dev_fasync(int fd, struct file *file, int on)
2133 {
2134 struct fuse_conn *fc = fuse_get_conn(file);
2135 if (!fc)
2136 return -EPERM;
2137
2138 /* No locking - fasync_helper does its own locking */
2139 return fasync_helper(fd, file, on, &fc->fasync);
2140 }
2141
2142 const struct file_operations fuse_dev_operations = {
2143 .owner = THIS_MODULE,
2144 .llseek = no_llseek,
2145 .read = do_sync_read,
2146 .aio_read = fuse_dev_read,
2147 .splice_read = fuse_dev_splice_read,
2148 .write = do_sync_write,
2149 .aio_write = fuse_dev_write,
2150 .splice_write = fuse_dev_splice_write,
2151 .poll = fuse_dev_poll,
2152 .release = fuse_dev_release,
2153 .fasync = fuse_dev_fasync,
2154 };
2155 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2156
2157 static struct miscdevice fuse_miscdevice = {
2158 .minor = FUSE_MINOR,
2159 .name = "fuse",
2160 .fops = &fuse_dev_operations,
2161 };
2162
2163 int __init fuse_dev_init(void)
2164 {
2165 int err = -ENOMEM;
2166 fuse_req_cachep = kmem_cache_create("fuse_request",
2167 sizeof(struct fuse_req),
2168 0, 0, NULL);
2169 if (!fuse_req_cachep)
2170 goto out;
2171
2172 err = misc_register(&fuse_miscdevice);
2173 if (err)
2174 goto out_cache_clean;
2175
2176 return 0;
2177
2178 out_cache_clean:
2179 kmem_cache_destroy(fuse_req_cachep);
2180 out:
2181 return err;
2182 }
2183
2184 void fuse_dev_cleanup(void)
2185 {
2186 misc_deregister(&fuse_miscdevice);
2187 kmem_cache_destroy(fuse_req_cachep);
2188 }
Linux with added FPC-III support