projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
fuse: req state use flags
[deliverable/linux.git] / fs / fuse / dev.c
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/uio.h>
15 #include <linux/miscdevice.h>
16 #include <linux/pagemap.h>
17 #include <linux/file.h>
18 #include <linux/slab.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/swap.h>
21 #include <linux/splice.h>
22
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 __set_bit(FR_PENDING, &req->flags);
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 void fuse_set_initialized(struct fuse_conn *fc)
135 {
136 /* Make sure stores before this are seen on another CPU */
137 smp_wmb();
138 fc->initialized = 1;
139 }
140
141 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
142 {
143 return !fc->initialized || (for_background && fc->blocked);
144 }
145
146 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
147 bool for_background)
148 {
149 struct fuse_req *req;
150 int err;
151 atomic_inc(&fc->num_waiting);
152
153 if (fuse_block_alloc(fc, for_background)) {
154 sigset_t oldset;
155 int intr;
156
157 block_sigs(&oldset);
158 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
159 !fuse_block_alloc(fc, for_background));
160 restore_sigs(&oldset);
161 err = -EINTR;
162 if (intr)
163 goto out;
164 }
165 /* Matches smp_wmb() in fuse_set_initialized() */
166 smp_rmb();
167
168 err = -ENOTCONN;
169 if (!fc->connected)
170 goto out;
171
172 err = -ECONNREFUSED;
173 if (fc->conn_error)
174 goto out;
175
176 req = fuse_request_alloc(npages);
177 err = -ENOMEM;
178 if (!req) {
179 if (for_background)
180 wake_up(&fc->blocked_waitq);
181 goto out;
182 }
183
184 fuse_req_init_context(req);
185 __set_bit(FR_WAITING, &req->flags);
186 if (for_background)
187 __set_bit(FR_BACKGROUND, &req->flags);
188
189 return req;
190
191 out:
192 atomic_dec(&fc->num_waiting);
193 return ERR_PTR(err);
194 }
195
196 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
197 {
198 return __fuse_get_req(fc, npages, false);
199 }
200 EXPORT_SYMBOL_GPL(fuse_get_req);
201
202 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
203 unsigned npages)
204 {
205 return __fuse_get_req(fc, npages, true);
206 }
207 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
208
209 /*
210 * Return request in fuse_file->reserved_req. However that may
211 * currently be in use. If that is the case, wait for it to become
212 * available.
213 */
214 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
215 struct file *file)
216 {
217 struct fuse_req *req = NULL;
218 struct fuse_file *ff = file->private_data;
219
220 do {
221 wait_event(fc->reserved_req_waitq, ff->reserved_req);
222 spin_lock(&fc->lock);
223 if (ff->reserved_req) {
224 req = ff->reserved_req;
225 ff->reserved_req = NULL;
226 req->stolen_file = get_file(file);
227 }
228 spin_unlock(&fc->lock);
229 } while (!req);
230
231 return req;
232 }
233
234 /*
235 * Put stolen request back into fuse_file->reserved_req
236 */
237 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
238 {
239 struct file *file = req->stolen_file;
240 struct fuse_file *ff = file->private_data;
241
242 spin_lock(&fc->lock);
243 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
244 BUG_ON(ff->reserved_req);
245 ff->reserved_req = req;
246 wake_up_all(&fc->reserved_req_waitq);
247 spin_unlock(&fc->lock);
248 fput(file);
249 }
250
251 /*
252 * Gets a requests for a file operation, always succeeds
253 *
254 * This is used for sending the FLUSH request, which must get to
255 * userspace, due to POSIX locks which may need to be unlocked.
256 *
257 * If allocation fails due to OOM, use the reserved request in
258 * fuse_file.
259 *
260 * This is very unlikely to deadlock accidentally, since the
261 * filesystem should not have it's own file open. If deadlock is
262 * intentional, it can still be broken by "aborting" the filesystem.
263 */
264 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
265 struct file *file)
266 {
267 struct fuse_req *req;
268
269 atomic_inc(&fc->num_waiting);
270 wait_event(fc->blocked_waitq, fc->initialized);
271 /* Matches smp_wmb() in fuse_set_initialized() */
272 smp_rmb();
273 req = fuse_request_alloc(0);
274 if (!req)
275 req = get_reserved_req(fc, file);
276
277 fuse_req_init_context(req);
278 __set_bit(FR_WAITING, &req->flags);
279 __clear_bit(FR_BACKGROUND, &req->flags);
280 return req;
281 }
282
283 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
284 {
285 if (atomic_dec_and_test(&req->count)) {
286 if (test_bit(FR_BACKGROUND, &req->flags)) {
287 /*
288 * We get here in the unlikely case that a background
289 * request was allocated but not sent
290 */
291 spin_lock(&fc->lock);
292 if (!fc->blocked)
293 wake_up(&fc->blocked_waitq);
294 spin_unlock(&fc->lock);
295 }
296
297 if (test_bit(FR_WAITING, &req->flags)) {
298 __clear_bit(FR_WAITING, &req->flags);
299 atomic_dec(&fc->num_waiting);
300 }
301
302 if (req->stolen_file)
303 put_reserved_req(fc, req);
304 else
305 fuse_request_free(req);
306 }
307 }
308 EXPORT_SYMBOL_GPL(fuse_put_request);
309
310 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
311 {
312 unsigned nbytes = 0;
313 unsigned i;
314
315 for (i = 0; i < numargs; i++)
316 nbytes += args[i].size;
317
318 return nbytes;
319 }
320
321 static u64 fuse_get_unique(struct fuse_conn *fc)
322 {
323 return ++fc->reqctr;
324 }
325
326 static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
327 {
328 req->in.h.len = sizeof(struct fuse_in_header) +
329 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
330 list_add_tail(&req->list, &fc->pending);
331 wake_up(&fc->waitq);
332 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
333 }
334
335 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
336 u64 nodeid, u64 nlookup)
337 {
338 forget->forget_one.nodeid = nodeid;
339 forget->forget_one.nlookup = nlookup;
340
341 spin_lock(&fc->lock);
342 if (fc->connected) {
343 fc->forget_list_tail->next = forget;
344 fc->forget_list_tail = forget;
345 wake_up(&fc->waitq);
346 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
347 } else {
348 kfree(forget);
349 }
350 spin_unlock(&fc->lock);
351 }
352
353 static void flush_bg_queue(struct fuse_conn *fc)
354 {
355 while (fc->active_background < fc->max_background &&
356 !list_empty(&fc->bg_queue)) {
357 struct fuse_req *req;
358
359 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
360 list_del(&req->list);
361 fc->active_background++;
362 req->in.h.unique = fuse_get_unique(fc);
363 queue_request(fc, req);
364 }
365 }
366
367 /*
368 * This function is called when a request is finished. Either a reply
369 * has arrived or it was aborted (and not yet sent) or some error
370 * occurred during communication with userspace, or the device file
371 * was closed. The requester thread is woken up (if still waiting),
372 * the 'end' callback is called if given, else the reference to the
373 * request is released
374 *
375 * Called with fc->lock, unlocks it
376 */
377 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
378 __releases(fc->lock)
379 {
380 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
381 req->end = NULL;
382 list_del_init(&req->list);
383 list_del_init(&req->intr_entry);
384 WARN_ON(test_bit(FR_PENDING, &req->flags));
385 WARN_ON(test_bit(FR_SENT, &req->flags));
386 smp_wmb();
387 set_bit(FR_FINISHED, &req->flags);
388 if (test_bit(FR_BACKGROUND, &req->flags)) {
389 clear_bit(FR_BACKGROUND, &req->flags);
390 if (fc->num_background == fc->max_background)
391 fc->blocked = 0;
392
393 /* Wake up next waiter, if any */
394 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
395 wake_up(&fc->blocked_waitq);
396
397 if (fc->num_background == fc->congestion_threshold &&
398 fc->connected && fc->bdi_initialized) {
399 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
400 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
401 }
402 fc->num_background--;
403 fc->active_background--;
404 flush_bg_queue(fc);
405 }
406 spin_unlock(&fc->lock);
407 wake_up(&req->waitq);
408 if (end)
409 end(fc, req);
410 fuse_put_request(fc, req);
411 }
412
413 static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
414 {
415 list_add_tail(&req->intr_entry, &fc->interrupts);
416 wake_up(&fc->waitq);
417 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
418 }
419
420 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
421 {
422 int err;
423
424 if (!fc->no_interrupt) {
425 /* Any signal may interrupt this */
426 err = wait_event_interruptible(req->waitq,
427 test_bit(FR_FINISHED, &req->flags));
428 if (!err)
429 return;
430
431 spin_lock(&fc->lock);
432 set_bit(FR_INTERRUPTED, &req->flags);
433 if (test_bit(FR_SENT, &req->flags))
434 queue_interrupt(fc, req);
435 spin_unlock(&fc->lock);
436 }
437
438 if (!test_bit(FR_FORCE, &req->flags)) {
439 sigset_t oldset;
440
441 /* Only fatal signals may interrupt this */
442 block_sigs(&oldset);
443 err = wait_event_interruptible(req->waitq,
444 test_bit(FR_FINISHED, &req->flags));
445 restore_sigs(&oldset);
446
447 if (!err)
448 return;
449
450 spin_lock(&fc->lock);
451 /* Request is not yet in userspace, bail out */
452 if (test_bit(FR_PENDING, &req->flags)) {
453 list_del(&req->list);
454 spin_unlock(&fc->lock);
455 __fuse_put_request(req);
456 req->out.h.error = -EINTR;
457 return;
458 }
459 spin_unlock(&fc->lock);
460 }
461
462 /*
463 * Either request is already in userspace, or it was forced.
464 * Wait it out.
465 */
466 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
467 }
468
469 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
470 {
471 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
472 spin_lock(&fc->lock);
473 if (!fc->connected) {
474 spin_unlock(&fc->lock);
475 req->out.h.error = -ENOTCONN;
476 } else {
477 req->in.h.unique = fuse_get_unique(fc);
478 queue_request(fc, req);
479 /* acquire extra reference, since request is still needed
480 after request_end() */
481 __fuse_get_request(req);
482 spin_unlock(&fc->lock);
483
484 request_wait_answer(fc, req);
485 /* Pairs with smp_wmb() in request_end() */
486 smp_rmb();
487 }
488 }
489
490 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
491 {
492 __set_bit(FR_ISREPLY, &req->flags);
493 if (!test_bit(FR_WAITING, &req->flags)) {
494 __set_bit(FR_WAITING, &req->flags);
495 atomic_inc(&fc->num_waiting);
496 }
497 __fuse_request_send(fc, req);
498 }
499 EXPORT_SYMBOL_GPL(fuse_request_send);
500
501 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
502 {
503 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
504 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
505
506 if (fc->minor < 9) {
507 switch (args->in.h.opcode) {
508 case FUSE_LOOKUP:
509 case FUSE_CREATE:
510 case FUSE_MKNOD:
511 case FUSE_MKDIR:
512 case FUSE_SYMLINK:
513 case FUSE_LINK:
514 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
515 break;
516 case FUSE_GETATTR:
517 case FUSE_SETATTR:
518 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
519 break;
520 }
521 }
522 if (fc->minor < 12) {
523 switch (args->in.h.opcode) {
524 case FUSE_CREATE:
525 args->in.args[0].size = sizeof(struct fuse_open_in);
526 break;
527 case FUSE_MKNOD:
528 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
529 break;
530 }
531 }
532 }
533
534 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
535 {
536 struct fuse_req *req;
537 ssize_t ret;
538
539 req = fuse_get_req(fc, 0);
540 if (IS_ERR(req))
541 return PTR_ERR(req);
542
543 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
544 fuse_adjust_compat(fc, args);
545
546 req->in.h.opcode = args->in.h.opcode;
547 req->in.h.nodeid = args->in.h.nodeid;
548 req->in.numargs = args->in.numargs;
549 memcpy(req->in.args, args->in.args,
550 args->in.numargs * sizeof(struct fuse_in_arg));
551 req->out.argvar = args->out.argvar;
552 req->out.numargs = args->out.numargs;
553 memcpy(req->out.args, args->out.args,
554 args->out.numargs * sizeof(struct fuse_arg));
555 fuse_request_send(fc, req);
556 ret = req->out.h.error;
557 if (!ret && args->out.argvar) {
558 BUG_ON(args->out.numargs != 1);
559 ret = req->out.args[0].size;
560 }
561 fuse_put_request(fc, req);
562
563 return ret;
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 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
575 if (!test_bit(FR_WAITING, &req->flags)) {
576 __set_bit(FR_WAITING, &req->flags);
577 atomic_inc(&fc->num_waiting);
578 }
579 __set_bit(FR_ISREPLY, &req->flags);
580 fc->num_background++;
581 if (fc->num_background == fc->max_background)
582 fc->blocked = 1;
583 if (fc->num_background == fc->congestion_threshold &&
584 fc->bdi_initialized) {
585 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
586 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
587 }
588 list_add_tail(&req->list, &fc->bg_queue);
589 flush_bg_queue(fc);
590 }
591
592 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
593 {
594 BUG_ON(!req->end);
595 spin_lock(&fc->lock);
596 if (fc->connected) {
597 fuse_request_send_background_locked(fc, req);
598 spin_unlock(&fc->lock);
599 } else {
600 spin_unlock(&fc->lock);
601 req->out.h.error = -ENOTCONN;
602 req->end(fc, req);
603 fuse_put_request(fc, req);
604 }
605 }
606 EXPORT_SYMBOL_GPL(fuse_request_send_background);
607
608 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
609 struct fuse_req *req, u64 unique)
610 {
611 int err = -ENODEV;
612
613 __clear_bit(FR_ISREPLY, &req->flags);
614 req->in.h.unique = unique;
615 spin_lock(&fc->lock);
616 if (fc->connected) {
617 queue_request(fc, req);
618 err = 0;
619 }
620 spin_unlock(&fc->lock);
621
622 return err;
623 }
624
625 void fuse_force_forget(struct file *file, u64 nodeid)
626 {
627 struct inode *inode = file_inode(file);
628 struct fuse_conn *fc = get_fuse_conn(inode);
629 struct fuse_req *req;
630 struct fuse_forget_in inarg;
631
632 memset(&inarg, 0, sizeof(inarg));
633 inarg.nlookup = 1;
634 req = fuse_get_req_nofail_nopages(fc, file);
635 req->in.h.opcode = FUSE_FORGET;
636 req->in.h.nodeid = nodeid;
637 req->in.numargs = 1;
638 req->in.args[0].size = sizeof(inarg);
639 req->in.args[0].value = &inarg;
640 __clear_bit(FR_ISREPLY, &req->flags);
641 __fuse_request_send(fc, req);
642 /* ignore errors */
643 fuse_put_request(fc, req);
644 }
645
646 /*
647 * Lock the request. Up to the next unlock_request() there mustn't be
648 * anything that could cause a page-fault. If the request was already
649 * aborted bail out.
650 */
651 static int lock_request(struct fuse_req *req)
652 {
653 int err = 0;
654 if (req) {
655 spin_lock(&req->waitq.lock);
656 if (test_bit(FR_ABORTED, &req->flags))
657 err = -ENOENT;
658 else
659 set_bit(FR_LOCKED, &req->flags);
660 spin_unlock(&req->waitq.lock);
661 }
662 return err;
663 }
664
665 /*
666 * Unlock request. If it was aborted while locked, caller is responsible
667 * for unlocking and ending the request.
668 */
669 static int unlock_request(struct fuse_req *req)
670 {
671 int err = 0;
672 if (req) {
673 spin_lock(&req->waitq.lock);
674 if (test_bit(FR_ABORTED, &req->flags))
675 err = -ENOENT;
676 else
677 clear_bit(FR_LOCKED, &req->flags);
678 spin_unlock(&req->waitq.lock);
679 }
680 return err;
681 }
682
683 struct fuse_copy_state {
684 int write;
685 struct fuse_req *req;
686 struct iov_iter *iter;
687 struct pipe_buffer *pipebufs;
688 struct pipe_buffer *currbuf;
689 struct pipe_inode_info *pipe;
690 unsigned long nr_segs;
691 struct page *pg;
692 unsigned len;
693 unsigned offset;
694 unsigned move_pages:1;
695 };
696
697 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
698 struct iov_iter *iter)
699 {
700 memset(cs, 0, sizeof(*cs));
701 cs->write = write;
702 cs->iter = iter;
703 }
704
705 /* Unmap and put previous page of userspace buffer */
706 static void fuse_copy_finish(struct fuse_copy_state *cs)
707 {
708 if (cs->currbuf) {
709 struct pipe_buffer *buf = cs->currbuf;
710
711 if (cs->write)
712 buf->len = PAGE_SIZE - cs->len;
713 cs->currbuf = NULL;
714 } else if (cs->pg) {
715 if (cs->write) {
716 flush_dcache_page(cs->pg);
717 set_page_dirty_lock(cs->pg);
718 }
719 put_page(cs->pg);
720 }
721 cs->pg = NULL;
722 }
723
724 /*
725 * Get another pagefull of userspace buffer, and map it to kernel
726 * address space, and lock request
727 */
728 static int fuse_copy_fill(struct fuse_copy_state *cs)
729 {
730 struct page *page;
731 int err;
732
733 err = unlock_request(cs->req);
734 if (err)
735 return err;
736
737 fuse_copy_finish(cs);
738 if (cs->pipebufs) {
739 struct pipe_buffer *buf = cs->pipebufs;
740
741 if (!cs->write) {
742 err = buf->ops->confirm(cs->pipe, buf);
743 if (err)
744 return err;
745
746 BUG_ON(!cs->nr_segs);
747 cs->currbuf = buf;
748 cs->pg = buf->page;
749 cs->offset = buf->offset;
750 cs->len = buf->len;
751 cs->pipebufs++;
752 cs->nr_segs--;
753 } else {
754 if (cs->nr_segs == cs->pipe->buffers)
755 return -EIO;
756
757 page = alloc_page(GFP_HIGHUSER);
758 if (!page)
759 return -ENOMEM;
760
761 buf->page = page;
762 buf->offset = 0;
763 buf->len = 0;
764
765 cs->currbuf = buf;
766 cs->pg = page;
767 cs->offset = 0;
768 cs->len = PAGE_SIZE;
769 cs->pipebufs++;
770 cs->nr_segs++;
771 }
772 } else {
773 size_t off;
774 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
775 if (err < 0)
776 return err;
777 BUG_ON(!err);
778 cs->len = err;
779 cs->offset = off;
780 cs->pg = page;
781 cs->offset = off;
782 iov_iter_advance(cs->iter, err);
783 }
784
785 return lock_request(cs->req);
786 }
787
788 /* Do as much copy to/from userspace buffer as we can */
789 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
790 {
791 unsigned ncpy = min(*size, cs->len);
792 if (val) {
793 void *pgaddr = kmap_atomic(cs->pg);
794 void *buf = pgaddr + cs->offset;
795
796 if (cs->write)
797 memcpy(buf, *val, ncpy);
798 else
799 memcpy(*val, buf, ncpy);
800
801 kunmap_atomic(pgaddr);
802 *val += ncpy;
803 }
804 *size -= ncpy;
805 cs->len -= ncpy;
806 cs->offset += ncpy;
807 return ncpy;
808 }
809
810 static int fuse_check_page(struct page *page)
811 {
812 if (page_mapcount(page) ||
813 page->mapping != NULL ||
814 page_count(page) != 1 ||
815 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
816 ~(1 << PG_locked |
817 1 << PG_referenced |
818 1 << PG_uptodate |
819 1 << PG_lru |
820 1 << PG_active |
821 1 << PG_reclaim))) {
822 printk(KERN_WARNING "fuse: trying to steal weird page\n");
823 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);
824 return 1;
825 }
826 return 0;
827 }
828
829 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
830 {
831 int err;
832 struct page *oldpage = *pagep;
833 struct page *newpage;
834 struct pipe_buffer *buf = cs->pipebufs;
835
836 err = unlock_request(cs->req);
837 if (err)
838 return err;
839
840 fuse_copy_finish(cs);
841
842 err = buf->ops->confirm(cs->pipe, buf);
843 if (err)
844 return err;
845
846 BUG_ON(!cs->nr_segs);
847 cs->currbuf = buf;
848 cs->len = buf->len;
849 cs->pipebufs++;
850 cs->nr_segs--;
851
852 if (cs->len != PAGE_SIZE)
853 goto out_fallback;
854
855 if (buf->ops->steal(cs->pipe, buf) != 0)
856 goto out_fallback;
857
858 newpage = buf->page;
859
860 if (!PageUptodate(newpage))
861 SetPageUptodate(newpage);
862
863 ClearPageMappedToDisk(newpage);
864
865 if (fuse_check_page(newpage) != 0)
866 goto out_fallback_unlock;
867
868 /*
869 * This is a new and locked page, it shouldn't be mapped or
870 * have any special flags on it
871 */
872 if (WARN_ON(page_mapped(oldpage)))
873 goto out_fallback_unlock;
874 if (WARN_ON(page_has_private(oldpage)))
875 goto out_fallback_unlock;
876 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
877 goto out_fallback_unlock;
878 if (WARN_ON(PageMlocked(oldpage)))
879 goto out_fallback_unlock;
880
881 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
882 if (err) {
883 unlock_page(newpage);
884 return err;
885 }
886
887 page_cache_get(newpage);
888
889 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
890 lru_cache_add_file(newpage);
891
892 err = 0;
893 spin_lock(&cs->req->waitq.lock);
894 if (test_bit(FR_ABORTED, &cs->req->flags))
895 err = -ENOENT;
896 else
897 *pagep = newpage;
898 spin_unlock(&cs->req->waitq.lock);
899
900 if (err) {
901 unlock_page(newpage);
902 page_cache_release(newpage);
903 return err;
904 }
905
906 unlock_page(oldpage);
907 page_cache_release(oldpage);
908 cs->len = 0;
909
910 return 0;
911
912 out_fallback_unlock:
913 unlock_page(newpage);
914 out_fallback:
915 cs->pg = buf->page;
916 cs->offset = buf->offset;
917
918 err = lock_request(cs->req);
919 if (err)
920 return err;
921
922 return 1;
923 }
924
925 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
926 unsigned offset, unsigned count)
927 {
928 struct pipe_buffer *buf;
929 int err;
930
931 if (cs->nr_segs == cs->pipe->buffers)
932 return -EIO;
933
934 err = unlock_request(cs->req);
935 if (err)
936 return err;
937
938 fuse_copy_finish(cs);
939
940 buf = cs->pipebufs;
941 page_cache_get(page);
942 buf->page = page;
943 buf->offset = offset;
944 buf->len = count;
945
946 cs->pipebufs++;
947 cs->nr_segs++;
948 cs->len = 0;
949
950 return 0;
951 }
952
953 /*
954 * Copy a page in the request to/from the userspace buffer. Must be
955 * done atomically
956 */
957 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
958 unsigned offset, unsigned count, int zeroing)
959 {
960 int err;
961 struct page *page = *pagep;
962
963 if (page && zeroing && count < PAGE_SIZE)
964 clear_highpage(page);
965
966 while (count) {
967 if (cs->write && cs->pipebufs && page) {
968 return fuse_ref_page(cs, page, offset, count);
969 } else if (!cs->len) {
970 if (cs->move_pages && page &&
971 offset == 0 && count == PAGE_SIZE) {
972 err = fuse_try_move_page(cs, pagep);
973 if (err <= 0)
974 return err;
975 } else {
976 err = fuse_copy_fill(cs);
977 if (err)
978 return err;
979 }
980 }
981 if (page) {
982 void *mapaddr = kmap_atomic(page);
983 void *buf = mapaddr + offset;
984 offset += fuse_copy_do(cs, &buf, &count);
985 kunmap_atomic(mapaddr);
986 } else
987 offset += fuse_copy_do(cs, NULL, &count);
988 }
989 if (page && !cs->write)
990 flush_dcache_page(page);
991 return 0;
992 }
993
994 /* Copy pages in the request to/from userspace buffer */
995 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
996 int zeroing)
997 {
998 unsigned i;
999 struct fuse_req *req = cs->req;
1000
1001 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1002 int err;
1003 unsigned offset = req->page_descs[i].offset;
1004 unsigned count = min(nbytes, req->page_descs[i].length);
1005
1006 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1007 zeroing);
1008 if (err)
1009 return err;
1010
1011 nbytes -= count;
1012 }
1013 return 0;
1014 }
1015
1016 /* Copy a single argument in the request to/from userspace buffer */
1017 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1018 {
1019 while (size) {
1020 if (!cs->len) {
1021 int err = fuse_copy_fill(cs);
1022 if (err)
1023 return err;
1024 }
1025 fuse_copy_do(cs, &val, &size);
1026 }
1027 return 0;
1028 }
1029
1030 /* Copy request arguments to/from userspace buffer */
1031 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1032 unsigned argpages, struct fuse_arg *args,
1033 int zeroing)
1034 {
1035 int err = 0;
1036 unsigned i;
1037
1038 for (i = 0; !err && i < numargs; i++) {
1039 struct fuse_arg *arg = &args[i];
1040 if (i == numargs - 1 && argpages)
1041 err = fuse_copy_pages(cs, arg->size, zeroing);
1042 else
1043 err = fuse_copy_one(cs, arg->value, arg->size);
1044 }
1045 return err;
1046 }
1047
1048 static int forget_pending(struct fuse_conn *fc)
1049 {
1050 return fc->forget_list_head.next != NULL;
1051 }
1052
1053 static int request_pending(struct fuse_conn *fc)
1054 {
1055 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
1056 forget_pending(fc);
1057 }
1058
1059 /* Wait until a request is available on the pending list */
1060 static void request_wait(struct fuse_conn *fc)
1061 __releases(fc->lock)
1062 __acquires(fc->lock)
1063 {
1064 DECLARE_WAITQUEUE(wait, current);
1065
1066 add_wait_queue_exclusive(&fc->waitq, &wait);
1067 while (fc->connected && !request_pending(fc)) {
1068 set_current_state(TASK_INTERRUPTIBLE);
1069 if (signal_pending(current))
1070 break;
1071
1072 spin_unlock(&fc->lock);
1073 schedule();
1074 spin_lock(&fc->lock);
1075 }
1076 set_current_state(TASK_RUNNING);
1077 remove_wait_queue(&fc->waitq, &wait);
1078 }
1079
1080 /*
1081 * Transfer an interrupt request to userspace
1082 *
1083 * Unlike other requests this is assembled on demand, without a need
1084 * to allocate a separate fuse_req structure.
1085 *
1086 * Called with fc->lock held, releases it
1087 */
1088 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
1089 size_t nbytes, struct fuse_req *req)
1090 __releases(fc->lock)
1091 {
1092 struct fuse_in_header ih;
1093 struct fuse_interrupt_in arg;
1094 unsigned reqsize = sizeof(ih) + sizeof(arg);
1095 int err;
1096
1097 list_del_init(&req->intr_entry);
1098 req->intr_unique = fuse_get_unique(fc);
1099 memset(&ih, 0, sizeof(ih));
1100 memset(&arg, 0, sizeof(arg));
1101 ih.len = reqsize;
1102 ih.opcode = FUSE_INTERRUPT;
1103 ih.unique = req->intr_unique;
1104 arg.unique = req->in.h.unique;
1105
1106 spin_unlock(&fc->lock);
1107 if (nbytes < reqsize)
1108 return -EINVAL;
1109
1110 err = fuse_copy_one(cs, &ih, sizeof(ih));
1111 if (!err)
1112 err = fuse_copy_one(cs, &arg, sizeof(arg));
1113 fuse_copy_finish(cs);
1114
1115 return err ? err : reqsize;
1116 }
1117
1118 static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
1119 unsigned max,
1120 unsigned *countp)
1121 {
1122 struct fuse_forget_link *head = fc->forget_list_head.next;
1123 struct fuse_forget_link **newhead = &head;
1124 unsigned count;
1125
1126 for (count = 0; *newhead != NULL && count < max; count++)
1127 newhead = &(*newhead)->next;
1128
1129 fc->forget_list_head.next = *newhead;
1130 *newhead = NULL;
1131 if (fc->forget_list_head.next == NULL)
1132 fc->forget_list_tail = &fc->forget_list_head;
1133
1134 if (countp != NULL)
1135 *countp = count;
1136
1137 return head;
1138 }
1139
1140 static int fuse_read_single_forget(struct fuse_conn *fc,
1141 struct fuse_copy_state *cs,
1142 size_t nbytes)
1143 __releases(fc->lock)
1144 {
1145 int err;
1146 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1147 struct fuse_forget_in arg = {
1148 .nlookup = forget->forget_one.nlookup,
1149 };
1150 struct fuse_in_header ih = {
1151 .opcode = FUSE_FORGET,
1152 .nodeid = forget->forget_one.nodeid,
1153 .unique = fuse_get_unique(fc),
1154 .len = sizeof(ih) + sizeof(arg),
1155 };
1156
1157 spin_unlock(&fc->lock);
1158 kfree(forget);
1159 if (nbytes < ih.len)
1160 return -EINVAL;
1161
1162 err = fuse_copy_one(cs, &ih, sizeof(ih));
1163 if (!err)
1164 err = fuse_copy_one(cs, &arg, sizeof(arg));
1165 fuse_copy_finish(cs);
1166
1167 if (err)
1168 return err;
1169
1170 return ih.len;
1171 }
1172
1173 static int fuse_read_batch_forget(struct fuse_conn *fc,
1174 struct fuse_copy_state *cs, size_t nbytes)
1175 __releases(fc->lock)
1176 {
1177 int err;
1178 unsigned max_forgets;
1179 unsigned count;
1180 struct fuse_forget_link *head;
1181 struct fuse_batch_forget_in arg = { .count = 0 };
1182 struct fuse_in_header ih = {
1183 .opcode = FUSE_BATCH_FORGET,
1184 .unique = fuse_get_unique(fc),
1185 .len = sizeof(ih) + sizeof(arg),
1186 };
1187
1188 if (nbytes < ih.len) {
1189 spin_unlock(&fc->lock);
1190 return -EINVAL;
1191 }
1192
1193 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1194 head = dequeue_forget(fc, max_forgets, &count);
1195 spin_unlock(&fc->lock);
1196
1197 arg.count = count;
1198 ih.len += count * sizeof(struct fuse_forget_one);
1199 err = fuse_copy_one(cs, &ih, sizeof(ih));
1200 if (!err)
1201 err = fuse_copy_one(cs, &arg, sizeof(arg));
1202
1203 while (head) {
1204 struct fuse_forget_link *forget = head;
1205
1206 if (!err) {
1207 err = fuse_copy_one(cs, &forget->forget_one,
1208 sizeof(forget->forget_one));
1209 }
1210 head = forget->next;
1211 kfree(forget);
1212 }
1213
1214 fuse_copy_finish(cs);
1215
1216 if (err)
1217 return err;
1218
1219 return ih.len;
1220 }
1221
1222 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1223 size_t nbytes)
1224 __releases(fc->lock)
1225 {
1226 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1227 return fuse_read_single_forget(fc, cs, nbytes);
1228 else
1229 return fuse_read_batch_forget(fc, cs, nbytes);
1230 }
1231
1232 /*
1233 * Read a single request into the userspace filesystem's buffer. This
1234 * function waits until a request is available, then removes it from
1235 * the pending list and copies request data to userspace buffer. If
1236 * no reply is needed (FORGET) or request has been aborted or there
1237 * was an error during the copying then it's finished by calling
1238 * request_end(). Otherwise add it to the processing list, and set
1239 * the 'sent' flag.
1240 */
1241 static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1242 struct fuse_copy_state *cs, size_t nbytes)
1243 {
1244 int err;
1245 struct fuse_req *req;
1246 struct fuse_in *in;
1247 unsigned reqsize;
1248
1249 restart:
1250 spin_lock(&fc->lock);
1251 err = -EAGAIN;
1252 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1253 !request_pending(fc))
1254 goto err_unlock;
1255
1256 request_wait(fc);
1257 err = -ENODEV;
1258 if (!fc->connected)
1259 goto err_unlock;
1260 err = -ERESTARTSYS;
1261 if (!request_pending(fc))
1262 goto err_unlock;
1263
1264 if (!list_empty(&fc->interrupts)) {
1265 req = list_entry(fc->interrupts.next, struct fuse_req,
1266 intr_entry);
1267 return fuse_read_interrupt(fc, cs, nbytes, req);
1268 }
1269
1270 if (forget_pending(fc)) {
1271 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1272 return fuse_read_forget(fc, cs, nbytes);
1273
1274 if (fc->forget_batch <= -8)
1275 fc->forget_batch = 16;
1276 }
1277
1278 req = list_entry(fc->pending.next, struct fuse_req, list);
1279 clear_bit(FR_PENDING, &req->flags);
1280 list_move(&req->list, &fc->io);
1281
1282 in = &req->in;
1283 reqsize = in->h.len;
1284 /* If request is too large, reply with an error and restart the read */
1285 if (nbytes < reqsize) {
1286 req->out.h.error = -EIO;
1287 /* SETXATTR is special, since it may contain too large data */
1288 if (in->h.opcode == FUSE_SETXATTR)
1289 req->out.h.error = -E2BIG;
1290 request_end(fc, req);
1291 goto restart;
1292 }
1293 spin_unlock(&fc->lock);
1294 cs->req = req;
1295 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1296 if (!err)
1297 err = fuse_copy_args(cs, in->numargs, in->argpages,
1298 (struct fuse_arg *) in->args, 0);
1299 fuse_copy_finish(cs);
1300 spin_lock(&fc->lock);
1301 clear_bit(FR_LOCKED, &req->flags);
1302 if (!fc->connected) {
1303 request_end(fc, req);
1304 return -ENODEV;
1305 }
1306 if (err) {
1307 req->out.h.error = -EIO;
1308 request_end(fc, req);
1309 return err;
1310 }
1311 if (!test_bit(FR_ISREPLY, &req->flags)) {
1312 request_end(fc, req);
1313 } else {
1314 set_bit(FR_SENT, &req->flags);
1315 list_move_tail(&req->list, &fc->processing);
1316 if (test_bit(FR_INTERRUPTED, &req->flags))
1317 queue_interrupt(fc, req);
1318 spin_unlock(&fc->lock);
1319 }
1320 return reqsize;
1321
1322 err_unlock:
1323 spin_unlock(&fc->lock);
1324 return err;
1325 }
1326
1327 static int fuse_dev_open(struct inode *inode, struct file *file)
1328 {
1329 /*
1330 * The fuse device's file's private_data is used to hold
1331 * the fuse_conn(ection) when it is mounted, and is used to
1332 * keep track of whether the file has been mounted already.
1333 */
1334 file->private_data = NULL;
1335 return 0;
1336 }
1337
1338 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1339 {
1340 struct fuse_copy_state cs;
1341 struct file *file = iocb->ki_filp;
1342 struct fuse_conn *fc = fuse_get_conn(file);
1343 if (!fc)
1344 return -EPERM;
1345
1346 if (!iter_is_iovec(to))
1347 return -EINVAL;
1348
1349 fuse_copy_init(&cs, 1, to);
1350
1351 return fuse_dev_do_read(fc, file, &cs, iov_iter_count(to));
1352 }
1353
1354 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1355 struct pipe_inode_info *pipe,
1356 size_t len, unsigned int flags)
1357 {
1358 int ret;
1359 int page_nr = 0;
1360 int do_wakeup = 0;
1361 struct pipe_buffer *bufs;
1362 struct fuse_copy_state cs;
1363 struct fuse_conn *fc = fuse_get_conn(in);
1364 if (!fc)
1365 return -EPERM;
1366
1367 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1368 if (!bufs)
1369 return -ENOMEM;
1370
1371 fuse_copy_init(&cs, 1, NULL);
1372 cs.pipebufs = bufs;
1373 cs.pipe = pipe;
1374 ret = fuse_dev_do_read(fc, in, &cs, len);
1375 if (ret < 0)
1376 goto out;
1377
1378 ret = 0;
1379 pipe_lock(pipe);
1380
1381 if (!pipe->readers) {
1382 send_sig(SIGPIPE, current, 0);
1383 if (!ret)
1384 ret = -EPIPE;
1385 goto out_unlock;
1386 }
1387
1388 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1389 ret = -EIO;
1390 goto out_unlock;
1391 }
1392
1393 while (page_nr < cs.nr_segs) {
1394 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1395 struct pipe_buffer *buf = pipe->bufs + newbuf;
1396
1397 buf->page = bufs[page_nr].page;
1398 buf->offset = bufs[page_nr].offset;
1399 buf->len = bufs[page_nr].len;
1400 /*
1401 * Need to be careful about this. Having buf->ops in module
1402 * code can Oops if the buffer persists after module unload.
1403 */
1404 buf->ops = &nosteal_pipe_buf_ops;
1405
1406 pipe->nrbufs++;
1407 page_nr++;
1408 ret += buf->len;
1409
1410 if (pipe->files)
1411 do_wakeup = 1;
1412 }
1413
1414 out_unlock:
1415 pipe_unlock(pipe);
1416
1417 if (do_wakeup) {
1418 smp_mb();
1419 if (waitqueue_active(&pipe->wait))
1420 wake_up_interruptible(&pipe->wait);
1421 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1422 }
1423
1424 out:
1425 for (; page_nr < cs.nr_segs; page_nr++)
1426 page_cache_release(bufs[page_nr].page);
1427
1428 kfree(bufs);
1429 return ret;
1430 }
1431
1432 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1433 struct fuse_copy_state *cs)
1434 {
1435 struct fuse_notify_poll_wakeup_out outarg;
1436 int err = -EINVAL;
1437
1438 if (size != sizeof(outarg))
1439 goto err;
1440
1441 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1442 if (err)
1443 goto err;
1444
1445 fuse_copy_finish(cs);
1446 return fuse_notify_poll_wakeup(fc, &outarg);
1447
1448 err:
1449 fuse_copy_finish(cs);
1450 return err;
1451 }
1452
1453 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1454 struct fuse_copy_state *cs)
1455 {
1456 struct fuse_notify_inval_inode_out outarg;
1457 int err = -EINVAL;
1458
1459 if (size != sizeof(outarg))
1460 goto err;
1461
1462 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1463 if (err)
1464 goto err;
1465 fuse_copy_finish(cs);
1466
1467 down_read(&fc->killsb);
1468 err = -ENOENT;
1469 if (fc->sb) {
1470 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1471 outarg.off, outarg.len);
1472 }
1473 up_read(&fc->killsb);
1474 return err;
1475
1476 err:
1477 fuse_copy_finish(cs);
1478 return err;
1479 }
1480
1481 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1482 struct fuse_copy_state *cs)
1483 {
1484 struct fuse_notify_inval_entry_out outarg;
1485 int err = -ENOMEM;
1486 char *buf;
1487 struct qstr name;
1488
1489 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1490 if (!buf)
1491 goto err;
1492
1493 err = -EINVAL;
1494 if (size < sizeof(outarg))
1495 goto err;
1496
1497 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1498 if (err)
1499 goto err;
1500
1501 err = -ENAMETOOLONG;
1502 if (outarg.namelen > FUSE_NAME_MAX)
1503 goto err;
1504
1505 err = -EINVAL;
1506 if (size != sizeof(outarg) + outarg.namelen + 1)
1507 goto err;
1508
1509 name.name = buf;
1510 name.len = outarg.namelen;
1511 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1512 if (err)
1513 goto err;
1514 fuse_copy_finish(cs);
1515 buf[outarg.namelen] = 0;
1516 name.hash = full_name_hash(name.name, name.len);
1517
1518 down_read(&fc->killsb);
1519 err = -ENOENT;
1520 if (fc->sb)
1521 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1522 up_read(&fc->killsb);
1523 kfree(buf);
1524 return err;
1525
1526 err:
1527 kfree(buf);
1528 fuse_copy_finish(cs);
1529 return err;
1530 }
1531
1532 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1533 struct fuse_copy_state *cs)
1534 {
1535 struct fuse_notify_delete_out outarg;
1536 int err = -ENOMEM;
1537 char *buf;
1538 struct qstr name;
1539
1540 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1541 if (!buf)
1542 goto err;
1543
1544 err = -EINVAL;
1545 if (size < sizeof(outarg))
1546 goto err;
1547
1548 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1549 if (err)
1550 goto err;
1551
1552 err = -ENAMETOOLONG;
1553 if (outarg.namelen > FUSE_NAME_MAX)
1554 goto err;
1555
1556 err = -EINVAL;
1557 if (size != sizeof(outarg) + outarg.namelen + 1)
1558 goto err;
1559
1560 name.name = buf;
1561 name.len = outarg.namelen;
1562 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1563 if (err)
1564 goto err;
1565 fuse_copy_finish(cs);
1566 buf[outarg.namelen] = 0;
1567 name.hash = full_name_hash(name.name, name.len);
1568
1569 down_read(&fc->killsb);
1570 err = -ENOENT;
1571 if (fc->sb)
1572 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1573 outarg.child, &name);
1574 up_read(&fc->killsb);
1575 kfree(buf);
1576 return err;
1577
1578 err:
1579 kfree(buf);
1580 fuse_copy_finish(cs);
1581 return err;
1582 }
1583
1584 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1585 struct fuse_copy_state *cs)
1586 {
1587 struct fuse_notify_store_out outarg;
1588 struct inode *inode;
1589 struct address_space *mapping;
1590 u64 nodeid;
1591 int err;
1592 pgoff_t index;
1593 unsigned int offset;
1594 unsigned int num;
1595 loff_t file_size;
1596 loff_t end;
1597
1598 err = -EINVAL;
1599 if (size < sizeof(outarg))
1600 goto out_finish;
1601
1602 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1603 if (err)
1604 goto out_finish;
1605
1606 err = -EINVAL;
1607 if (size - sizeof(outarg) != outarg.size)
1608 goto out_finish;
1609
1610 nodeid = outarg.nodeid;
1611
1612 down_read(&fc->killsb);
1613
1614 err = -ENOENT;
1615 if (!fc->sb)
1616 goto out_up_killsb;
1617
1618 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1619 if (!inode)
1620 goto out_up_killsb;
1621
1622 mapping = inode->i_mapping;
1623 index = outarg.offset >> PAGE_CACHE_SHIFT;
1624 offset = outarg.offset & ~PAGE_CACHE_MASK;
1625 file_size = i_size_read(inode);
1626 end = outarg.offset + outarg.size;
1627 if (end > file_size) {
1628 file_size = end;
1629 fuse_write_update_size(inode, file_size);
1630 }
1631
1632 num = outarg.size;
1633 while (num) {
1634 struct page *page;
1635 unsigned int this_num;
1636
1637 err = -ENOMEM;
1638 page = find_or_create_page(mapping, index,
1639 mapping_gfp_mask(mapping));
1640 if (!page)
1641 goto out_iput;
1642
1643 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1644 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1645 if (!err && offset == 0 &&
1646 (this_num == PAGE_CACHE_SIZE || file_size == end))
1647 SetPageUptodate(page);
1648 unlock_page(page);
1649 page_cache_release(page);
1650
1651 if (err)
1652 goto out_iput;
1653
1654 num -= this_num;
1655 offset = 0;
1656 index++;
1657 }
1658
1659 err = 0;
1660
1661 out_iput:
1662 iput(inode);
1663 out_up_killsb:
1664 up_read(&fc->killsb);
1665 out_finish:
1666 fuse_copy_finish(cs);
1667 return err;
1668 }
1669
1670 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1671 {
1672 release_pages(req->pages, req->num_pages, false);
1673 }
1674
1675 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1676 struct fuse_notify_retrieve_out *outarg)
1677 {
1678 int err;
1679 struct address_space *mapping = inode->i_mapping;
1680 struct fuse_req *req;
1681 pgoff_t index;
1682 loff_t file_size;
1683 unsigned int num;
1684 unsigned int offset;
1685 size_t total_len = 0;
1686 int num_pages;
1687
1688 offset = outarg->offset & ~PAGE_CACHE_MASK;
1689 file_size = i_size_read(inode);
1690
1691 num = outarg->size;
1692 if (outarg->offset > file_size)
1693 num = 0;
1694 else if (outarg->offset + num > file_size)
1695 num = file_size - outarg->offset;
1696
1697 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1698 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1699
1700 req = fuse_get_req(fc, num_pages);
1701 if (IS_ERR(req))
1702 return PTR_ERR(req);
1703
1704 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1705 req->in.h.nodeid = outarg->nodeid;
1706 req->in.numargs = 2;
1707 req->in.argpages = 1;
1708 req->page_descs[0].offset = offset;
1709 req->end = fuse_retrieve_end;
1710
1711 index = outarg->offset >> PAGE_CACHE_SHIFT;
1712
1713 while (num && req->num_pages < num_pages) {
1714 struct page *page;
1715 unsigned int this_num;
1716
1717 page = find_get_page(mapping, index);
1718 if (!page)
1719 break;
1720
1721 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1722 req->pages[req->num_pages] = page;
1723 req->page_descs[req->num_pages].length = this_num;
1724 req->num_pages++;
1725
1726 offset = 0;
1727 num -= this_num;
1728 total_len += this_num;
1729 index++;
1730 }
1731 req->misc.retrieve_in.offset = outarg->offset;
1732 req->misc.retrieve_in.size = total_len;
1733 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1734 req->in.args[0].value = &req->misc.retrieve_in;
1735 req->in.args[1].size = total_len;
1736
1737 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1738 if (err)
1739 fuse_retrieve_end(fc, req);
1740
1741 return err;
1742 }
1743
1744 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1745 struct fuse_copy_state *cs)
1746 {
1747 struct fuse_notify_retrieve_out outarg;
1748 struct inode *inode;
1749 int err;
1750
1751 err = -EINVAL;
1752 if (size != sizeof(outarg))
1753 goto copy_finish;
1754
1755 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1756 if (err)
1757 goto copy_finish;
1758
1759 fuse_copy_finish(cs);
1760
1761 down_read(&fc->killsb);
1762 err = -ENOENT;
1763 if (fc->sb) {
1764 u64 nodeid = outarg.nodeid;
1765
1766 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1767 if (inode) {
1768 err = fuse_retrieve(fc, inode, &outarg);
1769 iput(inode);
1770 }
1771 }
1772 up_read(&fc->killsb);
1773
1774 return err;
1775
1776 copy_finish:
1777 fuse_copy_finish(cs);
1778 return err;
1779 }
1780
1781 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1782 unsigned int size, struct fuse_copy_state *cs)
1783 {
1784 /* Don't try to move pages (yet) */
1785 cs->move_pages = 0;
1786
1787 switch (code) {
1788 case FUSE_NOTIFY_POLL:
1789 return fuse_notify_poll(fc, size, cs);
1790
1791 case FUSE_NOTIFY_INVAL_INODE:
1792 return fuse_notify_inval_inode(fc, size, cs);
1793
1794 case FUSE_NOTIFY_INVAL_ENTRY:
1795 return fuse_notify_inval_entry(fc, size, cs);
1796
1797 case FUSE_NOTIFY_STORE:
1798 return fuse_notify_store(fc, size, cs);
1799
1800 case FUSE_NOTIFY_RETRIEVE:
1801 return fuse_notify_retrieve(fc, size, cs);
1802
1803 case FUSE_NOTIFY_DELETE:
1804 return fuse_notify_delete(fc, size, cs);
1805
1806 default:
1807 fuse_copy_finish(cs);
1808 return -EINVAL;
1809 }
1810 }
1811
1812 /* Look up request on processing list by unique ID */
1813 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1814 {
1815 struct fuse_req *req;
1816
1817 list_for_each_entry(req, &fc->processing, list) {
1818 if (req->in.h.unique == unique || req->intr_unique == unique)
1819 return req;
1820 }
1821 return NULL;
1822 }
1823
1824 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1825 unsigned nbytes)
1826 {
1827 unsigned reqsize = sizeof(struct fuse_out_header);
1828
1829 if (out->h.error)
1830 return nbytes != reqsize ? -EINVAL : 0;
1831
1832 reqsize += len_args(out->numargs, out->args);
1833
1834 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1835 return -EINVAL;
1836 else if (reqsize > nbytes) {
1837 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1838 unsigned diffsize = reqsize - nbytes;
1839 if (diffsize > lastarg->size)
1840 return -EINVAL;
1841 lastarg->size -= diffsize;
1842 }
1843 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1844 out->page_zeroing);
1845 }
1846
1847 /*
1848 * Write a single reply to a request. First the header is copied from
1849 * the write buffer. The request is then searched on the processing
1850 * list by the unique ID found in the header. If found, then remove
1851 * it from the list and copy the rest of the buffer to the request.
1852 * The request is finished by calling request_end()
1853 */
1854 static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1855 struct fuse_copy_state *cs, size_t nbytes)
1856 {
1857 int err;
1858 struct fuse_req *req;
1859 struct fuse_out_header oh;
1860
1861 if (nbytes < sizeof(struct fuse_out_header))
1862 return -EINVAL;
1863
1864 err = fuse_copy_one(cs, &oh, sizeof(oh));
1865 if (err)
1866 goto err_finish;
1867
1868 err = -EINVAL;
1869 if (oh.len != nbytes)
1870 goto err_finish;
1871
1872 /*
1873 * Zero oh.unique indicates unsolicited notification message
1874 * and error contains notification code.
1875 */
1876 if (!oh.unique) {
1877 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1878 return err ? err : nbytes;
1879 }
1880
1881 err = -EINVAL;
1882 if (oh.error <= -1000 || oh.error > 0)
1883 goto err_finish;
1884
1885 spin_lock(&fc->lock);
1886 err = -ENOENT;
1887 if (!fc->connected)
1888 goto err_unlock;
1889
1890 req = request_find(fc, oh.unique);
1891 if (!req)
1892 goto err_unlock;
1893
1894 /* Is it an interrupt reply? */
1895 if (req->intr_unique == oh.unique) {
1896 err = -EINVAL;
1897 if (nbytes != sizeof(struct fuse_out_header))
1898 goto err_unlock;
1899
1900 if (oh.error == -ENOSYS)
1901 fc->no_interrupt = 1;
1902 else if (oh.error == -EAGAIN)
1903 queue_interrupt(fc, req);
1904
1905 spin_unlock(&fc->lock);
1906 fuse_copy_finish(cs);
1907 return nbytes;
1908 }
1909
1910 clear_bit(FR_SENT, &req->flags);
1911 list_move(&req->list, &fc->io);
1912 req->out.h = oh;
1913 set_bit(FR_LOCKED, &req->flags);
1914 cs->req = req;
1915 if (!req->out.page_replace)
1916 cs->move_pages = 0;
1917 spin_unlock(&fc->lock);
1918
1919 err = copy_out_args(cs, &req->out, nbytes);
1920 fuse_copy_finish(cs);
1921
1922 spin_lock(&fc->lock);
1923 clear_bit(FR_LOCKED, &req->flags);
1924 if (!fc->connected)
1925 err = -ENOENT;
1926 else if (err)
1927 req->out.h.error = -EIO;
1928 request_end(fc, req);
1929
1930 return err ? err : nbytes;
1931
1932 err_unlock:
1933 spin_unlock(&fc->lock);
1934 err_finish:
1935 fuse_copy_finish(cs);
1936 return err;
1937 }
1938
1939 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1940 {
1941 struct fuse_copy_state cs;
1942 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1943 if (!fc)
1944 return -EPERM;
1945
1946 if (!iter_is_iovec(from))
1947 return -EINVAL;
1948
1949 fuse_copy_init(&cs, 0, from);
1950
1951 return fuse_dev_do_write(fc, &cs, iov_iter_count(from));
1952 }
1953
1954 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1955 struct file *out, loff_t *ppos,
1956 size_t len, unsigned int flags)
1957 {
1958 unsigned nbuf;
1959 unsigned idx;
1960 struct pipe_buffer *bufs;
1961 struct fuse_copy_state cs;
1962 struct fuse_conn *fc;
1963 size_t rem;
1964 ssize_t ret;
1965
1966 fc = fuse_get_conn(out);
1967 if (!fc)
1968 return -EPERM;
1969
1970 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1971 if (!bufs)
1972 return -ENOMEM;
1973
1974 pipe_lock(pipe);
1975 nbuf = 0;
1976 rem = 0;
1977 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1978 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1979
1980 ret = -EINVAL;
1981 if (rem < len) {
1982 pipe_unlock(pipe);
1983 goto out;
1984 }
1985
1986 rem = len;
1987 while (rem) {
1988 struct pipe_buffer *ibuf;
1989 struct pipe_buffer *obuf;
1990
1991 BUG_ON(nbuf >= pipe->buffers);
1992 BUG_ON(!pipe->nrbufs);
1993 ibuf = &pipe->bufs[pipe->curbuf];
1994 obuf = &bufs[nbuf];
1995
1996 if (rem >= ibuf->len) {
1997 *obuf = *ibuf;
1998 ibuf->ops = NULL;
1999 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2000 pipe->nrbufs--;
2001 } else {
2002 ibuf->ops->get(pipe, ibuf);
2003 *obuf = *ibuf;
2004 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2005 obuf->len = rem;
2006 ibuf->offset += obuf->len;
2007 ibuf->len -= obuf->len;
2008 }
2009 nbuf++;
2010 rem -= obuf->len;
2011 }
2012 pipe_unlock(pipe);
2013
2014 fuse_copy_init(&cs, 0, NULL);
2015 cs.pipebufs = bufs;
2016 cs.nr_segs = nbuf;
2017 cs.pipe = pipe;
2018
2019 if (flags & SPLICE_F_MOVE)
2020 cs.move_pages = 1;
2021
2022 ret = fuse_dev_do_write(fc, &cs, len);
2023
2024 for (idx = 0; idx < nbuf; idx++) {
2025 struct pipe_buffer *buf = &bufs[idx];
2026 buf->ops->release(pipe, buf);
2027 }
2028 out:
2029 kfree(bufs);
2030 return ret;
2031 }
2032
2033 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2034 {
2035 unsigned mask = POLLOUT | POLLWRNORM;
2036 struct fuse_conn *fc = fuse_get_conn(file);
2037 if (!fc)
2038 return POLLERR;
2039
2040 poll_wait(file, &fc->waitq, wait);
2041
2042 spin_lock(&fc->lock);
2043 if (!fc->connected)
2044 mask = POLLERR;
2045 else if (request_pending(fc))
2046 mask |= POLLIN | POLLRDNORM;
2047 spin_unlock(&fc->lock);
2048
2049 return mask;
2050 }
2051
2052 /*
2053 * Abort all requests on the given list (pending or processing)
2054 *
2055 * This function releases and reacquires fc->lock
2056 */
2057 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2058 __releases(fc->lock)
2059 __acquires(fc->lock)
2060 {
2061 while (!list_empty(head)) {
2062 struct fuse_req *req;
2063 req = list_entry(head->next, struct fuse_req, list);
2064 req->out.h.error = -ECONNABORTED;
2065 clear_bit(FR_PENDING, &req->flags);
2066 clear_bit(FR_SENT, &req->flags);
2067 request_end(fc, req);
2068 spin_lock(&fc->lock);
2069 }
2070 }
2071
2072 static void end_polls(struct fuse_conn *fc)
2073 {
2074 struct rb_node *p;
2075
2076 p = rb_first(&fc->polled_files);
2077
2078 while (p) {
2079 struct fuse_file *ff;
2080 ff = rb_entry(p, struct fuse_file, polled_node);
2081 wake_up_interruptible_all(&ff->poll_wait);
2082
2083 p = rb_next(p);
2084 }
2085 }
2086
2087 /*
2088 * Abort all requests.
2089 *
2090 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2091 * filesystem.
2092 *
2093 * The same effect is usually achievable through killing the filesystem daemon
2094 * and all users of the filesystem. The exception is the combination of an
2095 * asynchronous request and the tricky deadlock (see
2096 * Documentation/filesystems/fuse.txt).
2097 *
2098 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2099 * requests, they should be finished off immediately. Locked requests will be
2100 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2101 * requests. It is possible that some request will finish before we can. This
2102 * is OK, the request will in that case be removed from the list before we touch
2103 * it.
2104 */
2105 void fuse_abort_conn(struct fuse_conn *fc)
2106 {
2107 spin_lock(&fc->lock);
2108 if (fc->connected) {
2109 struct fuse_req *req, *next;
2110 LIST_HEAD(to_end1);
2111 LIST_HEAD(to_end2);
2112
2113 fc->connected = 0;
2114 fc->blocked = 0;
2115 fuse_set_initialized(fc);
2116 list_for_each_entry_safe(req, next, &fc->io, list) {
2117 req->out.h.error = -ECONNABORTED;
2118 spin_lock(&req->waitq.lock);
2119 set_bit(FR_ABORTED, &req->flags);
2120 if (!test_bit(FR_LOCKED, &req->flags))
2121 list_move(&req->list, &to_end1);
2122 spin_unlock(&req->waitq.lock);
2123 }
2124 fc->max_background = UINT_MAX;
2125 flush_bg_queue(fc);
2126 list_splice_init(&fc->pending, &to_end2);
2127 list_splice_init(&fc->processing, &to_end2);
2128 while (!list_empty(&to_end1)) {
2129 req = list_first_entry(&to_end1, struct fuse_req, list);
2130 __fuse_get_request(req);
2131 request_end(fc, req);
2132 spin_lock(&fc->lock);
2133 }
2134 end_requests(fc, &to_end2);
2135 while (forget_pending(fc))
2136 kfree(dequeue_forget(fc, 1, NULL));
2137 end_polls(fc);
2138 wake_up_all(&fc->waitq);
2139 wake_up_all(&fc->blocked_waitq);
2140 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
2141 }
2142 spin_unlock(&fc->lock);
2143 }
2144 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2145
2146 int fuse_dev_release(struct inode *inode, struct file *file)
2147 {
2148 struct fuse_conn *fc = fuse_get_conn(file);
2149 if (fc) {
2150 WARN_ON(!list_empty(&fc->io));
2151 WARN_ON(fc->fasync != NULL);
2152 fuse_abort_conn(fc);
2153 fuse_conn_put(fc);
2154 }
2155
2156 return 0;
2157 }
2158 EXPORT_SYMBOL_GPL(fuse_dev_release);
2159
2160 static int fuse_dev_fasync(int fd, struct file *file, int on)
2161 {
2162 struct fuse_conn *fc = fuse_get_conn(file);
2163 if (!fc)
2164 return -EPERM;
2165
2166 /* No locking - fasync_helper does its own locking */
2167 return fasync_helper(fd, file, on, &fc->fasync);
2168 }
2169
2170 const struct file_operations fuse_dev_operations = {
2171 .owner = THIS_MODULE,
2172 .open = fuse_dev_open,
2173 .llseek = no_llseek,
2174 .read_iter = fuse_dev_read,
2175 .splice_read = fuse_dev_splice_read,
2176 .write_iter = fuse_dev_write,
2177 .splice_write = fuse_dev_splice_write,
2178 .poll = fuse_dev_poll,
2179 .release = fuse_dev_release,
2180 .fasync = fuse_dev_fasync,
2181 };
2182 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2183
2184 static struct miscdevice fuse_miscdevice = {
2185 .minor = FUSE_MINOR,
2186 .name = "fuse",
2187 .fops = &fuse_dev_operations,
2188 };
2189
2190 int __init fuse_dev_init(void)
2191 {
2192 int err = -ENOMEM;
2193 fuse_req_cachep = kmem_cache_create("fuse_request",
2194 sizeof(struct fuse_req),
2195 0, 0, NULL);
2196 if (!fuse_req_cachep)
2197 goto out;
2198
2199 err = misc_register(&fuse_miscdevice);
2200 if (err)
2201 goto out_cache_clean;
2202
2203 return 0;
2204
2205 out_cache_clean:
2206 kmem_cache_destroy(fuse_req_cachep);
2207 out:
2208 return err;
2209 }
2210
2211 void fuse_dev_cleanup(void)
2212 {
2213 misc_deregister(&fuse_miscdevice);
2214 kmem_cache_destroy(fuse_req_cachep);
2215 }
Linux kernel - Deliverables
This page took 0.084843 seconds and 6 git commands to generate.