]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/fuse/dev.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
fuse: call pipe_buf_release() under pipe lock
[mirror_ubuntu-bionic-kernel.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/sched/signal.h>
15 #include <linux/uio.h>
16 #include <linux/miscdevice.h>
17 #include <linux/pagemap.h>
18 #include <linux/file.h>
19 #include <linux/slab.h>
20 #include <linux/pipe_fs_i.h>
21 #include <linux/swap.h>
22 #include <linux/splice.h>
23 #include <linux/sched.h>
24
25 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26 MODULE_ALIAS("devname:fuse");
27
28 static struct kmem_cache *fuse_req_cachep;
29
30 static struct fuse_dev *fuse_get_dev(struct file *file)
31 {
32 /*
33 * Lockless access is OK, because file->private data is set
34 * once during mount and is valid until the file is released.
35 */
36 return READ_ONCE(file->private_data);
37 }
38
39 static void fuse_request_init(struct fuse_req *req, struct page **pages,
40 struct fuse_page_desc *page_descs,
41 unsigned npages)
42 {
43 memset(req, 0, sizeof(*req));
44 memset(pages, 0, sizeof(*pages) * npages);
45 memset(page_descs, 0, sizeof(*page_descs) * npages);
46 INIT_LIST_HEAD(&req->list);
47 INIT_LIST_HEAD(&req->intr_entry);
48 init_waitqueue_head(&req->waitq);
49 refcount_set(&req->count, 1);
50 req->pages = pages;
51 req->page_descs = page_descs;
52 req->max_pages = npages;
53 __set_bit(FR_PENDING, &req->flags);
54 }
55
56 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
57 {
58 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
59 if (req) {
60 struct page **pages;
61 struct fuse_page_desc *page_descs;
62
63 if (npages <= FUSE_REQ_INLINE_PAGES) {
64 pages = req->inline_pages;
65 page_descs = req->inline_page_descs;
66 } else {
67 pages = kmalloc(sizeof(struct page *) * npages, flags);
68 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
69 npages, flags);
70 }
71
72 if (!pages || !page_descs) {
73 kfree(pages);
74 kfree(page_descs);
75 kmem_cache_free(fuse_req_cachep, req);
76 return NULL;
77 }
78
79 fuse_request_init(req, pages, page_descs, npages);
80 }
81 return req;
82 }
83
84 struct fuse_req *fuse_request_alloc(unsigned npages)
85 {
86 return __fuse_request_alloc(npages, GFP_KERNEL);
87 }
88 EXPORT_SYMBOL_GPL(fuse_request_alloc);
89
90 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
91 {
92 return __fuse_request_alloc(npages, GFP_NOFS);
93 }
94
95 void fuse_request_free(struct fuse_req *req)
96 {
97 if (req->pages != req->inline_pages) {
98 kfree(req->pages);
99 kfree(req->page_descs);
100 }
101 kmem_cache_free(fuse_req_cachep, req);
102 }
103
104 void __fuse_get_request(struct fuse_req *req)
105 {
106 refcount_inc(&req->count);
107 }
108
109 /* Must be called with > 1 refcount */
110 static void __fuse_put_request(struct fuse_req *req)
111 {
112 refcount_dec(&req->count);
113 }
114
115 static void fuse_req_init_context(struct fuse_conn *fc, struct fuse_req *req)
116 {
117 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid());
118 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid());
119 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
120 }
121
122 void fuse_set_initialized(struct fuse_conn *fc)
123 {
124 /* Make sure stores before this are seen on another CPU */
125 smp_wmb();
126 fc->initialized = 1;
127 }
128
129 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
130 {
131 return !fc->initialized || (for_background && fc->blocked);
132 }
133
134 static void fuse_drop_waiting(struct fuse_conn *fc)
135 {
136 /*
137 * lockess check of fc->connected is okay, because atomic_dec_and_test()
138 * provides a memory barrier mached with the one in fuse_wait_aborted()
139 * to ensure no wake-up is missed.
140 */
141 if (atomic_dec_and_test(&fc->num_waiting) &&
142 !READ_ONCE(fc->connected)) {
143 /* wake up aborters */
144 wake_up_all(&fc->blocked_waitq);
145 }
146 }
147
148 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
149 bool for_background)
150 {
151 struct fuse_req *req;
152 int err;
153 atomic_inc(&fc->num_waiting);
154
155 if (fuse_block_alloc(fc, for_background)) {
156 err = -EINTR;
157 if (wait_event_killable_exclusive(fc->blocked_waitq,
158 !fuse_block_alloc(fc, for_background)))
159 goto out;
160 }
161 /* Matches smp_wmb() in fuse_set_initialized() */
162 smp_rmb();
163
164 err = -ENOTCONN;
165 if (!fc->connected)
166 goto out;
167
168 err = -ECONNREFUSED;
169 if (fc->conn_error)
170 goto out;
171
172 req = fuse_request_alloc(npages);
173 err = -ENOMEM;
174 if (!req) {
175 if (for_background)
176 wake_up(&fc->blocked_waitq);
177 goto out;
178 }
179
180 fuse_req_init_context(fc, req);
181 __set_bit(FR_WAITING, &req->flags);
182 if (for_background)
183 __set_bit(FR_BACKGROUND, &req->flags);
184 if (req->in.h.uid == (uid_t)-1 || req->in.h.gid == (gid_t)-1) {
185 fuse_put_request(fc, req);
186 return ERR_PTR(-EOVERFLOW);
187 }
188
189 return req;
190
191 out:
192 fuse_drop_waiting(fc);
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(fc, 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 (refcount_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 fuse_drop_waiting(fc);
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_iqueue *fiq)
322 {
323 return ++fiq->reqctr;
324 }
325
326 static void queue_request(struct fuse_iqueue *fiq, 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, &fiq->pending);
331 wake_up_locked(&fiq->waitq);
332 kill_fasync(&fiq->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 struct fuse_iqueue *fiq = &fc->iq;
339
340 forget->forget_one.nodeid = nodeid;
341 forget->forget_one.nlookup = nlookup;
342
343 spin_lock(&fiq->waitq.lock);
344 if (fiq->connected) {
345 fiq->forget_list_tail->next = forget;
346 fiq->forget_list_tail = forget;
347 wake_up_locked(&fiq->waitq);
348 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
349 } else {
350 kfree(forget);
351 }
352 spin_unlock(&fiq->waitq.lock);
353 }
354
355 static void flush_bg_queue(struct fuse_conn *fc)
356 {
357 while (fc->active_background < fc->max_background &&
358 !list_empty(&fc->bg_queue)) {
359 struct fuse_req *req;
360 struct fuse_iqueue *fiq = &fc->iq;
361
362 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
363 list_del(&req->list);
364 fc->active_background++;
365 spin_lock(&fiq->waitq.lock);
366 req->in.h.unique = fuse_get_unique(fiq);
367 queue_request(fiq, req);
368 spin_unlock(&fiq->waitq.lock);
369 }
370 }
371
372 /*
373 * This function is called when a request is finished. Either a reply
374 * has arrived or it was aborted (and not yet sent) or some error
375 * occurred during communication with userspace, or the device file
376 * was closed. The requester thread is woken up (if still waiting),
377 * the 'end' callback is called if given, else the reference to the
378 * request is released
379 */
380 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
381 {
382 struct fuse_iqueue *fiq = &fc->iq;
383
384 if (test_and_set_bit(FR_FINISHED, &req->flags))
385 goto put_request;
386
387 spin_lock(&fiq->waitq.lock);
388 list_del_init(&req->intr_entry);
389 spin_unlock(&fiq->waitq.lock);
390 WARN_ON(test_bit(FR_PENDING, &req->flags));
391 WARN_ON(test_bit(FR_SENT, &req->flags));
392 if (test_bit(FR_BACKGROUND, &req->flags)) {
393 spin_lock(&fc->lock);
394 clear_bit(FR_BACKGROUND, &req->flags);
395 if (fc->num_background == fc->max_background) {
396 fc->blocked = 0;
397 wake_up(&fc->blocked_waitq);
398 } else if (!fc->blocked) {
399 /*
400 * Wake up next waiter, if any. It's okay to use
401 * waitqueue_active(), as we've already synced up
402 * fc->blocked with waiters with the wake_up() call
403 * above.
404 */
405 if (waitqueue_active(&fc->blocked_waitq))
406 wake_up(&fc->blocked_waitq);
407 }
408
409 if (fc->num_background == fc->congestion_threshold && fc->sb) {
410 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
411 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
412 }
413 fc->num_background--;
414 fc->active_background--;
415 flush_bg_queue(fc);
416 spin_unlock(&fc->lock);
417 }
418 wake_up(&req->waitq);
419 if (req->end)
420 req->end(fc, req);
421 put_request:
422 fuse_put_request(fc, req);
423 }
424
425 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
426 {
427 spin_lock(&fiq->waitq.lock);
428 if (test_bit(FR_FINISHED, &req->flags)) {
429 spin_unlock(&fiq->waitq.lock);
430 return;
431 }
432 if (list_empty(&req->intr_entry)) {
433 list_add_tail(&req->intr_entry, &fiq->interrupts);
434 wake_up_locked(&fiq->waitq);
435 }
436 spin_unlock(&fiq->waitq.lock);
437 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
438 }
439
440 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
441 {
442 struct fuse_iqueue *fiq = &fc->iq;
443 int err;
444
445 if (!fc->no_interrupt) {
446 /* Any signal may interrupt this */
447 err = wait_event_interruptible(req->waitq,
448 test_bit(FR_FINISHED, &req->flags));
449 if (!err)
450 return;
451
452 set_bit(FR_INTERRUPTED, &req->flags);
453 /* matches barrier in fuse_dev_do_read() */
454 smp_mb__after_atomic();
455 if (test_bit(FR_SENT, &req->flags))
456 queue_interrupt(fiq, req);
457 }
458
459 if (!test_bit(FR_FORCE, &req->flags)) {
460 /* Only fatal signals may interrupt this */
461 err = wait_event_killable(req->waitq,
462 test_bit(FR_FINISHED, &req->flags));
463 if (!err)
464 return;
465
466 spin_lock(&fiq->waitq.lock);
467 /* Request is not yet in userspace, bail out */
468 if (test_bit(FR_PENDING, &req->flags)) {
469 list_del(&req->list);
470 spin_unlock(&fiq->waitq.lock);
471 __fuse_put_request(req);
472 req->out.h.error = -EINTR;
473 return;
474 }
475 spin_unlock(&fiq->waitq.lock);
476 }
477
478 /*
479 * Either request is already in userspace, or it was forced.
480 * Wait it out.
481 */
482 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
483 }
484
485 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
486 {
487 struct fuse_iqueue *fiq = &fc->iq;
488
489 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
490 spin_lock(&fiq->waitq.lock);
491 if (!fiq->connected) {
492 spin_unlock(&fiq->waitq.lock);
493 req->out.h.error = -ENOTCONN;
494 } else {
495 req->in.h.unique = fuse_get_unique(fiq);
496 queue_request(fiq, req);
497 /* acquire extra reference, since request is still needed
498 after request_end() */
499 __fuse_get_request(req);
500 spin_unlock(&fiq->waitq.lock);
501
502 request_wait_answer(fc, req);
503 /* Pairs with smp_wmb() in request_end() */
504 smp_rmb();
505 }
506 }
507
508 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
509 {
510 __set_bit(FR_ISREPLY, &req->flags);
511 if (!test_bit(FR_WAITING, &req->flags)) {
512 __set_bit(FR_WAITING, &req->flags);
513 atomic_inc(&fc->num_waiting);
514 }
515 __fuse_request_send(fc, req);
516 }
517 EXPORT_SYMBOL_GPL(fuse_request_send);
518
519 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
520 {
521 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
522 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
523
524 if (fc->minor < 9) {
525 switch (args->in.h.opcode) {
526 case FUSE_LOOKUP:
527 case FUSE_CREATE:
528 case FUSE_MKNOD:
529 case FUSE_MKDIR:
530 case FUSE_SYMLINK:
531 case FUSE_LINK:
532 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
533 break;
534 case FUSE_GETATTR:
535 case FUSE_SETATTR:
536 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
537 break;
538 }
539 }
540 if (fc->minor < 12) {
541 switch (args->in.h.opcode) {
542 case FUSE_CREATE:
543 args->in.args[0].size = sizeof(struct fuse_open_in);
544 break;
545 case FUSE_MKNOD:
546 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
547 break;
548 }
549 }
550 }
551
552 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
553 {
554 struct fuse_req *req;
555 ssize_t ret;
556
557 req = fuse_get_req(fc, 0);
558 if (IS_ERR(req))
559 return PTR_ERR(req);
560
561 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
562 fuse_adjust_compat(fc, args);
563
564 req->in.h.opcode = args->in.h.opcode;
565 req->in.h.nodeid = args->in.h.nodeid;
566 req->in.numargs = args->in.numargs;
567 memcpy(req->in.args, args->in.args,
568 args->in.numargs * sizeof(struct fuse_in_arg));
569 req->out.argvar = args->out.argvar;
570 req->out.numargs = args->out.numargs;
571 memcpy(req->out.args, args->out.args,
572 args->out.numargs * sizeof(struct fuse_arg));
573 fuse_request_send(fc, req);
574 ret = req->out.h.error;
575 if (!ret && args->out.argvar) {
576 BUG_ON(args->out.numargs != 1);
577 ret = req->out.args[0].size;
578 }
579 fuse_put_request(fc, req);
580
581 return ret;
582 }
583
584 /*
585 * Called under fc->lock
586 *
587 * fc->connected must have been checked previously
588 */
589 void fuse_request_send_background_locked(struct fuse_conn *fc,
590 struct fuse_req *req)
591 {
592 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
593 if (!test_bit(FR_WAITING, &req->flags)) {
594 __set_bit(FR_WAITING, &req->flags);
595 atomic_inc(&fc->num_waiting);
596 }
597 __set_bit(FR_ISREPLY, &req->flags);
598 fc->num_background++;
599 if (fc->num_background == fc->max_background)
600 fc->blocked = 1;
601 if (fc->num_background == fc->congestion_threshold && fc->sb) {
602 set_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
603 set_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
604 }
605 list_add_tail(&req->list, &fc->bg_queue);
606 flush_bg_queue(fc);
607 }
608
609 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
610 {
611 BUG_ON(!req->end);
612 spin_lock(&fc->lock);
613 if (fc->connected) {
614 fuse_request_send_background_locked(fc, req);
615 spin_unlock(&fc->lock);
616 } else {
617 spin_unlock(&fc->lock);
618 req->out.h.error = -ENOTCONN;
619 req->end(fc, req);
620 fuse_put_request(fc, req);
621 }
622 }
623 EXPORT_SYMBOL_GPL(fuse_request_send_background);
624
625 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
626 struct fuse_req *req, u64 unique)
627 {
628 int err = -ENODEV;
629 struct fuse_iqueue *fiq = &fc->iq;
630
631 __clear_bit(FR_ISREPLY, &req->flags);
632 req->in.h.unique = unique;
633 spin_lock(&fiq->waitq.lock);
634 if (fiq->connected) {
635 queue_request(fiq, req);
636 err = 0;
637 }
638 spin_unlock(&fiq->waitq.lock);
639
640 return err;
641 }
642
643 void fuse_force_forget(struct file *file, u64 nodeid)
644 {
645 struct inode *inode = file_inode(file);
646 struct fuse_conn *fc = get_fuse_conn(inode);
647 struct fuse_req *req;
648 struct fuse_forget_in inarg;
649
650 memset(&inarg, 0, sizeof(inarg));
651 inarg.nlookup = 1;
652 req = fuse_get_req_nofail_nopages(fc, file);
653 req->in.h.opcode = FUSE_FORGET;
654 req->in.h.nodeid = nodeid;
655 req->in.numargs = 1;
656 req->in.args[0].size = sizeof(inarg);
657 req->in.args[0].value = &inarg;
658 __clear_bit(FR_ISREPLY, &req->flags);
659 __fuse_request_send(fc, req);
660 /* ignore errors */
661 fuse_put_request(fc, req);
662 }
663
664 /*
665 * Lock the request. Up to the next unlock_request() there mustn't be
666 * anything that could cause a page-fault. If the request was already
667 * aborted bail out.
668 */
669 static int lock_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 set_bit(FR_LOCKED, &req->flags);
678 spin_unlock(&req->waitq.lock);
679 }
680 return err;
681 }
682
683 /*
684 * Unlock request. If it was aborted while locked, caller is responsible
685 * for unlocking and ending the request.
686 */
687 static int unlock_request(struct fuse_req *req)
688 {
689 int err = 0;
690 if (req) {
691 spin_lock(&req->waitq.lock);
692 if (test_bit(FR_ABORTED, &req->flags))
693 err = -ENOENT;
694 else
695 clear_bit(FR_LOCKED, &req->flags);
696 spin_unlock(&req->waitq.lock);
697 }
698 return err;
699 }
700
701 struct fuse_copy_state {
702 int write;
703 struct fuse_req *req;
704 struct iov_iter *iter;
705 struct pipe_buffer *pipebufs;
706 struct pipe_buffer *currbuf;
707 struct pipe_inode_info *pipe;
708 unsigned long nr_segs;
709 struct page *pg;
710 unsigned len;
711 unsigned offset;
712 unsigned move_pages:1;
713 };
714
715 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
716 struct iov_iter *iter)
717 {
718 memset(cs, 0, sizeof(*cs));
719 cs->write = write;
720 cs->iter = iter;
721 }
722
723 /* Unmap and put previous page of userspace buffer */
724 static void fuse_copy_finish(struct fuse_copy_state *cs)
725 {
726 if (cs->currbuf) {
727 struct pipe_buffer *buf = cs->currbuf;
728
729 if (cs->write)
730 buf->len = PAGE_SIZE - cs->len;
731 cs->currbuf = NULL;
732 } else if (cs->pg) {
733 if (cs->write) {
734 flush_dcache_page(cs->pg);
735 set_page_dirty_lock(cs->pg);
736 }
737 put_page(cs->pg);
738 }
739 cs->pg = NULL;
740 }
741
742 /*
743 * Get another pagefull of userspace buffer, and map it to kernel
744 * address space, and lock request
745 */
746 static int fuse_copy_fill(struct fuse_copy_state *cs)
747 {
748 struct page *page;
749 int err;
750
751 err = unlock_request(cs->req);
752 if (err)
753 return err;
754
755 fuse_copy_finish(cs);
756 if (cs->pipebufs) {
757 struct pipe_buffer *buf = cs->pipebufs;
758
759 if (!cs->write) {
760 err = pipe_buf_confirm(cs->pipe, buf);
761 if (err)
762 return err;
763
764 BUG_ON(!cs->nr_segs);
765 cs->currbuf = buf;
766 cs->pg = buf->page;
767 cs->offset = buf->offset;
768 cs->len = buf->len;
769 cs->pipebufs++;
770 cs->nr_segs--;
771 } else {
772 if (cs->nr_segs == cs->pipe->buffers)
773 return -EIO;
774
775 page = alloc_page(GFP_HIGHUSER);
776 if (!page)
777 return -ENOMEM;
778
779 buf->page = page;
780 buf->offset = 0;
781 buf->len = 0;
782
783 cs->currbuf = buf;
784 cs->pg = page;
785 cs->offset = 0;
786 cs->len = PAGE_SIZE;
787 cs->pipebufs++;
788 cs->nr_segs++;
789 }
790 } else {
791 size_t off;
792 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
793 if (err < 0)
794 return err;
795 BUG_ON(!err);
796 cs->len = err;
797 cs->offset = off;
798 cs->pg = page;
799 iov_iter_advance(cs->iter, err);
800 }
801
802 return lock_request(cs->req);
803 }
804
805 /* Do as much copy to/from userspace buffer as we can */
806 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
807 {
808 unsigned ncpy = min(*size, cs->len);
809 if (val) {
810 void *pgaddr = kmap_atomic(cs->pg);
811 void *buf = pgaddr + cs->offset;
812
813 if (cs->write)
814 memcpy(buf, *val, ncpy);
815 else
816 memcpy(*val, buf, ncpy);
817
818 kunmap_atomic(pgaddr);
819 *val += ncpy;
820 }
821 *size -= ncpy;
822 cs->len -= ncpy;
823 cs->offset += ncpy;
824 return ncpy;
825 }
826
827 static int fuse_check_page(struct page *page)
828 {
829 if (page_mapcount(page) ||
830 page->mapping != NULL ||
831 page_count(page) != 1 ||
832 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
833 ~(1 << PG_locked |
834 1 << PG_referenced |
835 1 << PG_uptodate |
836 1 << PG_lru |
837 1 << PG_active |
838 1 << PG_reclaim))) {
839 printk(KERN_WARNING "fuse: trying to steal weird page\n");
840 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);
841 return 1;
842 }
843 return 0;
844 }
845
846 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
847 {
848 int err;
849 struct page *oldpage = *pagep;
850 struct page *newpage;
851 struct pipe_buffer *buf = cs->pipebufs;
852
853 err = unlock_request(cs->req);
854 if (err)
855 return err;
856
857 fuse_copy_finish(cs);
858
859 err = pipe_buf_confirm(cs->pipe, buf);
860 if (err)
861 return err;
862
863 BUG_ON(!cs->nr_segs);
864 cs->currbuf = buf;
865 cs->len = buf->len;
866 cs->pipebufs++;
867 cs->nr_segs--;
868
869 if (cs->len != PAGE_SIZE)
870 goto out_fallback;
871
872 if (pipe_buf_steal(cs->pipe, buf) != 0)
873 goto out_fallback;
874
875 newpage = buf->page;
876
877 if (!PageUptodate(newpage))
878 SetPageUptodate(newpage);
879
880 ClearPageMappedToDisk(newpage);
881
882 if (fuse_check_page(newpage) != 0)
883 goto out_fallback_unlock;
884
885 /*
886 * This is a new and locked page, it shouldn't be mapped or
887 * have any special flags on it
888 */
889 if (WARN_ON(page_mapped(oldpage)))
890 goto out_fallback_unlock;
891 if (WARN_ON(page_has_private(oldpage)))
892 goto out_fallback_unlock;
893 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
894 goto out_fallback_unlock;
895 if (WARN_ON(PageMlocked(oldpage)))
896 goto out_fallback_unlock;
897
898 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
899 if (err) {
900 unlock_page(newpage);
901 return err;
902 }
903
904 get_page(newpage);
905
906 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
907 lru_cache_add_file(newpage);
908
909 err = 0;
910 spin_lock(&cs->req->waitq.lock);
911 if (test_bit(FR_ABORTED, &cs->req->flags))
912 err = -ENOENT;
913 else
914 *pagep = newpage;
915 spin_unlock(&cs->req->waitq.lock);
916
917 if (err) {
918 unlock_page(newpage);
919 put_page(newpage);
920 return err;
921 }
922
923 unlock_page(oldpage);
924 put_page(oldpage);
925 cs->len = 0;
926
927 return 0;
928
929 out_fallback_unlock:
930 unlock_page(newpage);
931 out_fallback:
932 cs->pg = buf->page;
933 cs->offset = buf->offset;
934
935 err = lock_request(cs->req);
936 if (err)
937 return err;
938
939 return 1;
940 }
941
942 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
943 unsigned offset, unsigned count)
944 {
945 struct pipe_buffer *buf;
946 int err;
947
948 if (cs->nr_segs == cs->pipe->buffers)
949 return -EIO;
950
951 err = unlock_request(cs->req);
952 if (err)
953 return err;
954
955 fuse_copy_finish(cs);
956
957 buf = cs->pipebufs;
958 get_page(page);
959 buf->page = page;
960 buf->offset = offset;
961 buf->len = count;
962
963 cs->pipebufs++;
964 cs->nr_segs++;
965 cs->len = 0;
966
967 return 0;
968 }
969
970 /*
971 * Copy a page in the request to/from the userspace buffer. Must be
972 * done atomically
973 */
974 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
975 unsigned offset, unsigned count, int zeroing)
976 {
977 int err;
978 struct page *page = *pagep;
979
980 if (page && zeroing && count < PAGE_SIZE)
981 clear_highpage(page);
982
983 while (count) {
984 if (cs->write && cs->pipebufs && page) {
985 return fuse_ref_page(cs, page, offset, count);
986 } else if (!cs->len) {
987 if (cs->move_pages && page &&
988 offset == 0 && count == PAGE_SIZE) {
989 err = fuse_try_move_page(cs, pagep);
990 if (err <= 0)
991 return err;
992 } else {
993 err = fuse_copy_fill(cs);
994 if (err)
995 return err;
996 }
997 }
998 if (page) {
999 void *mapaddr = kmap_atomic(page);
1000 void *buf = mapaddr + offset;
1001 offset += fuse_copy_do(cs, &buf, &count);
1002 kunmap_atomic(mapaddr);
1003 } else
1004 offset += fuse_copy_do(cs, NULL, &count);
1005 }
1006 if (page && !cs->write)
1007 flush_dcache_page(page);
1008 return 0;
1009 }
1010
1011 /* Copy pages in the request to/from userspace buffer */
1012 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1013 int zeroing)
1014 {
1015 unsigned i;
1016 struct fuse_req *req = cs->req;
1017
1018 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
1019 int err;
1020 unsigned offset = req->page_descs[i].offset;
1021 unsigned count = min(nbytes, req->page_descs[i].length);
1022
1023 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1024 zeroing);
1025 if (err)
1026 return err;
1027
1028 nbytes -= count;
1029 }
1030 return 0;
1031 }
1032
1033 /* Copy a single argument in the request to/from userspace buffer */
1034 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1035 {
1036 while (size) {
1037 if (!cs->len) {
1038 int err = fuse_copy_fill(cs);
1039 if (err)
1040 return err;
1041 }
1042 fuse_copy_do(cs, &val, &size);
1043 }
1044 return 0;
1045 }
1046
1047 /* Copy request arguments to/from userspace buffer */
1048 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1049 unsigned argpages, struct fuse_arg *args,
1050 int zeroing)
1051 {
1052 int err = 0;
1053 unsigned i;
1054
1055 for (i = 0; !err && i < numargs; i++) {
1056 struct fuse_arg *arg = &args[i];
1057 if (i == numargs - 1 && argpages)
1058 err = fuse_copy_pages(cs, arg->size, zeroing);
1059 else
1060 err = fuse_copy_one(cs, arg->value, arg->size);
1061 }
1062 return err;
1063 }
1064
1065 static int forget_pending(struct fuse_iqueue *fiq)
1066 {
1067 return fiq->forget_list_head.next != NULL;
1068 }
1069
1070 static int request_pending(struct fuse_iqueue *fiq)
1071 {
1072 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1073 forget_pending(fiq);
1074 }
1075
1076 /*
1077 * Transfer an interrupt request to userspace
1078 *
1079 * Unlike other requests this is assembled on demand, without a need
1080 * to allocate a separate fuse_req structure.
1081 *
1082 * Called with fiq->waitq.lock held, releases it
1083 */
1084 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1085 struct fuse_copy_state *cs,
1086 size_t nbytes, struct fuse_req *req)
1087 __releases(fiq->waitq.lock)
1088 {
1089 struct fuse_in_header ih;
1090 struct fuse_interrupt_in arg;
1091 unsigned reqsize = sizeof(ih) + sizeof(arg);
1092 int err;
1093
1094 list_del_init(&req->intr_entry);
1095 req->intr_unique = fuse_get_unique(fiq);
1096 memset(&ih, 0, sizeof(ih));
1097 memset(&arg, 0, sizeof(arg));
1098 ih.len = reqsize;
1099 ih.opcode = FUSE_INTERRUPT;
1100 ih.unique = req->intr_unique;
1101 arg.unique = req->in.h.unique;
1102
1103 spin_unlock(&fiq->waitq.lock);
1104 if (nbytes < reqsize)
1105 return -EINVAL;
1106
1107 err = fuse_copy_one(cs, &ih, sizeof(ih));
1108 if (!err)
1109 err = fuse_copy_one(cs, &arg, sizeof(arg));
1110 fuse_copy_finish(cs);
1111
1112 return err ? err : reqsize;
1113 }
1114
1115 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1116 unsigned max,
1117 unsigned *countp)
1118 {
1119 struct fuse_forget_link *head = fiq->forget_list_head.next;
1120 struct fuse_forget_link **newhead = &head;
1121 unsigned count;
1122
1123 for (count = 0; *newhead != NULL && count < max; count++)
1124 newhead = &(*newhead)->next;
1125
1126 fiq->forget_list_head.next = *newhead;
1127 *newhead = NULL;
1128 if (fiq->forget_list_head.next == NULL)
1129 fiq->forget_list_tail = &fiq->forget_list_head;
1130
1131 if (countp != NULL)
1132 *countp = count;
1133
1134 return head;
1135 }
1136
1137 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1138 struct fuse_copy_state *cs,
1139 size_t nbytes)
1140 __releases(fiq->waitq.lock)
1141 {
1142 int err;
1143 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1144 struct fuse_forget_in arg = {
1145 .nlookup = forget->forget_one.nlookup,
1146 };
1147 struct fuse_in_header ih = {
1148 .opcode = FUSE_FORGET,
1149 .nodeid = forget->forget_one.nodeid,
1150 .unique = fuse_get_unique(fiq),
1151 .len = sizeof(ih) + sizeof(arg),
1152 };
1153
1154 spin_unlock(&fiq->waitq.lock);
1155 kfree(forget);
1156 if (nbytes < ih.len)
1157 return -EINVAL;
1158
1159 err = fuse_copy_one(cs, &ih, sizeof(ih));
1160 if (!err)
1161 err = fuse_copy_one(cs, &arg, sizeof(arg));
1162 fuse_copy_finish(cs);
1163
1164 if (err)
1165 return err;
1166
1167 return ih.len;
1168 }
1169
1170 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1171 struct fuse_copy_state *cs, size_t nbytes)
1172 __releases(fiq->waitq.lock)
1173 {
1174 int err;
1175 unsigned max_forgets;
1176 unsigned count;
1177 struct fuse_forget_link *head;
1178 struct fuse_batch_forget_in arg = { .count = 0 };
1179 struct fuse_in_header ih = {
1180 .opcode = FUSE_BATCH_FORGET,
1181 .unique = fuse_get_unique(fiq),
1182 .len = sizeof(ih) + sizeof(arg),
1183 };
1184
1185 if (nbytes < ih.len) {
1186 spin_unlock(&fiq->waitq.lock);
1187 return -EINVAL;
1188 }
1189
1190 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1191 head = dequeue_forget(fiq, max_forgets, &count);
1192 spin_unlock(&fiq->waitq.lock);
1193
1194 arg.count = count;
1195 ih.len += count * sizeof(struct fuse_forget_one);
1196 err = fuse_copy_one(cs, &ih, sizeof(ih));
1197 if (!err)
1198 err = fuse_copy_one(cs, &arg, sizeof(arg));
1199
1200 while (head) {
1201 struct fuse_forget_link *forget = head;
1202
1203 if (!err) {
1204 err = fuse_copy_one(cs, &forget->forget_one,
1205 sizeof(forget->forget_one));
1206 }
1207 head = forget->next;
1208 kfree(forget);
1209 }
1210
1211 fuse_copy_finish(cs);
1212
1213 if (err)
1214 return err;
1215
1216 return ih.len;
1217 }
1218
1219 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1220 struct fuse_copy_state *cs,
1221 size_t nbytes)
1222 __releases(fiq->waitq.lock)
1223 {
1224 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1225 return fuse_read_single_forget(fiq, cs, nbytes);
1226 else
1227 return fuse_read_batch_forget(fiq, cs, nbytes);
1228 }
1229
1230 /*
1231 * Read a single request into the userspace filesystem's buffer. This
1232 * function waits until a request is available, then removes it from
1233 * the pending list and copies request data to userspace buffer. If
1234 * no reply is needed (FORGET) or request has been aborted or there
1235 * was an error during the copying then it's finished by calling
1236 * request_end(). Otherwise add it to the processing list, and set
1237 * the 'sent' flag.
1238 */
1239 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1240 struct fuse_copy_state *cs, size_t nbytes)
1241 {
1242 ssize_t err;
1243 struct fuse_conn *fc = fud->fc;
1244 struct fuse_iqueue *fiq = &fc->iq;
1245 struct fuse_pqueue *fpq = &fud->pq;
1246 struct fuse_req *req;
1247 struct fuse_in *in;
1248 unsigned reqsize;
1249
1250 if (current_user_ns() != fc->user_ns)
1251 return -EIO;
1252
1253 restart:
1254 spin_lock(&fiq->waitq.lock);
1255 err = -EAGAIN;
1256 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1257 !request_pending(fiq))
1258 goto err_unlock;
1259
1260 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1261 !fiq->connected || request_pending(fiq));
1262 if (err)
1263 goto err_unlock;
1264
1265 err = -ENODEV;
1266 if (!fiq->connected)
1267 goto err_unlock;
1268
1269 if (!list_empty(&fiq->interrupts)) {
1270 req = list_entry(fiq->interrupts.next, struct fuse_req,
1271 intr_entry);
1272 return fuse_read_interrupt(fiq, cs, nbytes, req);
1273 }
1274
1275 if (forget_pending(fiq)) {
1276 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1277 return fuse_read_forget(fc, fiq, cs, nbytes);
1278
1279 if (fiq->forget_batch <= -8)
1280 fiq->forget_batch = 16;
1281 }
1282
1283 req = list_entry(fiq->pending.next, struct fuse_req, list);
1284 clear_bit(FR_PENDING, &req->flags);
1285 list_del_init(&req->list);
1286 spin_unlock(&fiq->waitq.lock);
1287
1288 in = &req->in;
1289 reqsize = in->h.len;
1290
1291 if (task_active_pid_ns(current) != fc->pid_ns) {
1292 rcu_read_lock();
1293 in->h.pid = pid_vnr(find_pid_ns(in->h.pid, fc->pid_ns));
1294 rcu_read_unlock();
1295 }
1296
1297 /* If request is too large, reply with an error and restart the read */
1298 if (nbytes < reqsize) {
1299 req->out.h.error = -EIO;
1300 /* SETXATTR is special, since it may contain too large data */
1301 if (in->h.opcode == FUSE_SETXATTR)
1302 req->out.h.error = -E2BIG;
1303 request_end(fc, req);
1304 goto restart;
1305 }
1306 spin_lock(&fpq->lock);
1307 list_add(&req->list, &fpq->io);
1308 spin_unlock(&fpq->lock);
1309 cs->req = req;
1310 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1311 if (!err)
1312 err = fuse_copy_args(cs, in->numargs, in->argpages,
1313 (struct fuse_arg *) in->args, 0);
1314 fuse_copy_finish(cs);
1315 spin_lock(&fpq->lock);
1316 clear_bit(FR_LOCKED, &req->flags);
1317 if (!fpq->connected) {
1318 err = -ENODEV;
1319 goto out_end;
1320 }
1321 if (err) {
1322 req->out.h.error = -EIO;
1323 goto out_end;
1324 }
1325 if (!test_bit(FR_ISREPLY, &req->flags)) {
1326 err = reqsize;
1327 goto out_end;
1328 }
1329 list_move_tail(&req->list, &fpq->processing);
1330 __fuse_get_request(req);
1331 set_bit(FR_SENT, &req->flags);
1332 spin_unlock(&fpq->lock);
1333 /* matches barrier in request_wait_answer() */
1334 smp_mb__after_atomic();
1335 if (test_bit(FR_INTERRUPTED, &req->flags))
1336 queue_interrupt(fiq, req);
1337 fuse_put_request(fc, req);
1338
1339 return reqsize;
1340
1341 out_end:
1342 if (!test_bit(FR_PRIVATE, &req->flags))
1343 list_del_init(&req->list);
1344 spin_unlock(&fpq->lock);
1345 request_end(fc, req);
1346 return err;
1347
1348 err_unlock:
1349 spin_unlock(&fiq->waitq.lock);
1350 return err;
1351 }
1352
1353 static int fuse_dev_open(struct inode *inode, struct file *file)
1354 {
1355 /*
1356 * The fuse device's file's private_data is used to hold
1357 * the fuse_conn(ection) when it is mounted, and is used to
1358 * keep track of whether the file has been mounted already.
1359 */
1360 file->private_data = NULL;
1361 return 0;
1362 }
1363
1364 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1365 {
1366 struct fuse_copy_state cs;
1367 struct file *file = iocb->ki_filp;
1368 struct fuse_dev *fud = fuse_get_dev(file);
1369
1370 if (!fud)
1371 return -EPERM;
1372
1373 if (!iter_is_iovec(to))
1374 return -EINVAL;
1375
1376 fuse_copy_init(&cs, 1, to);
1377
1378 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1379 }
1380
1381 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1382 struct pipe_inode_info *pipe,
1383 size_t len, unsigned int flags)
1384 {
1385 int total, ret;
1386 int page_nr = 0;
1387 struct pipe_buffer *bufs;
1388 struct fuse_copy_state cs;
1389 struct fuse_dev *fud = fuse_get_dev(in);
1390
1391 if (!fud)
1392 return -EPERM;
1393
1394 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1395 if (!bufs)
1396 return -ENOMEM;
1397
1398 fuse_copy_init(&cs, 1, NULL);
1399 cs.pipebufs = bufs;
1400 cs.pipe = pipe;
1401 ret = fuse_dev_do_read(fud, in, &cs, len);
1402 if (ret < 0)
1403 goto out;
1404
1405 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1406 ret = -EIO;
1407 goto out;
1408 }
1409
1410 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1411 /*
1412 * Need to be careful about this. Having buf->ops in module
1413 * code can Oops if the buffer persists after module unload.
1414 */
1415 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1416 bufs[page_nr].flags = 0;
1417 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1418 if (unlikely(ret < 0))
1419 break;
1420 }
1421 if (total)
1422 ret = total;
1423 out:
1424 for (; page_nr < cs.nr_segs; page_nr++)
1425 put_page(bufs[page_nr].page);
1426
1427 kfree(bufs);
1428 return ret;
1429 }
1430
1431 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1432 struct fuse_copy_state *cs)
1433 {
1434 struct fuse_notify_poll_wakeup_out outarg;
1435 int err = -EINVAL;
1436
1437 if (size != sizeof(outarg))
1438 goto err;
1439
1440 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1441 if (err)
1442 goto err;
1443
1444 fuse_copy_finish(cs);
1445 return fuse_notify_poll_wakeup(fc, &outarg);
1446
1447 err:
1448 fuse_copy_finish(cs);
1449 return err;
1450 }
1451
1452 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1453 struct fuse_copy_state *cs)
1454 {
1455 struct fuse_notify_inval_inode_out outarg;
1456 int err = -EINVAL;
1457
1458 if (size != sizeof(outarg))
1459 goto err;
1460
1461 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1462 if (err)
1463 goto err;
1464 fuse_copy_finish(cs);
1465
1466 down_read(&fc->killsb);
1467 err = -ENOENT;
1468 if (fc->sb) {
1469 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1470 outarg.off, outarg.len);
1471 }
1472 up_read(&fc->killsb);
1473 return err;
1474
1475 err:
1476 fuse_copy_finish(cs);
1477 return err;
1478 }
1479
1480 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1481 struct fuse_copy_state *cs)
1482 {
1483 struct fuse_notify_inval_entry_out outarg;
1484 int err = -ENOMEM;
1485 char *buf;
1486 struct qstr name;
1487
1488 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1489 if (!buf)
1490 goto err;
1491
1492 err = -EINVAL;
1493 if (size < sizeof(outarg))
1494 goto err;
1495
1496 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1497 if (err)
1498 goto err;
1499
1500 err = -ENAMETOOLONG;
1501 if (outarg.namelen > FUSE_NAME_MAX)
1502 goto err;
1503
1504 err = -EINVAL;
1505 if (size != sizeof(outarg) + outarg.namelen + 1)
1506 goto err;
1507
1508 name.name = buf;
1509 name.len = outarg.namelen;
1510 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1511 if (err)
1512 goto err;
1513 fuse_copy_finish(cs);
1514 buf[outarg.namelen] = 0;
1515
1516 down_read(&fc->killsb);
1517 err = -ENOENT;
1518 if (fc->sb)
1519 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1520 up_read(&fc->killsb);
1521 kfree(buf);
1522 return err;
1523
1524 err:
1525 kfree(buf);
1526 fuse_copy_finish(cs);
1527 return err;
1528 }
1529
1530 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1531 struct fuse_copy_state *cs)
1532 {
1533 struct fuse_notify_delete_out outarg;
1534 int err = -ENOMEM;
1535 char *buf;
1536 struct qstr name;
1537
1538 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1539 if (!buf)
1540 goto err;
1541
1542 err = -EINVAL;
1543 if (size < sizeof(outarg))
1544 goto err;
1545
1546 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1547 if (err)
1548 goto err;
1549
1550 err = -ENAMETOOLONG;
1551 if (outarg.namelen > FUSE_NAME_MAX)
1552 goto err;
1553
1554 err = -EINVAL;
1555 if (size != sizeof(outarg) + outarg.namelen + 1)
1556 goto err;
1557
1558 name.name = buf;
1559 name.len = outarg.namelen;
1560 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1561 if (err)
1562 goto err;
1563 fuse_copy_finish(cs);
1564 buf[outarg.namelen] = 0;
1565
1566 down_read(&fc->killsb);
1567 err = -ENOENT;
1568 if (fc->sb)
1569 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1570 outarg.child, &name);
1571 up_read(&fc->killsb);
1572 kfree(buf);
1573 return err;
1574
1575 err:
1576 kfree(buf);
1577 fuse_copy_finish(cs);
1578 return err;
1579 }
1580
1581 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1582 struct fuse_copy_state *cs)
1583 {
1584 struct fuse_notify_store_out outarg;
1585 struct inode *inode;
1586 struct address_space *mapping;
1587 u64 nodeid;
1588 int err;
1589 pgoff_t index;
1590 unsigned int offset;
1591 unsigned int num;
1592 loff_t file_size;
1593 loff_t end;
1594
1595 err = -EINVAL;
1596 if (size < sizeof(outarg))
1597 goto out_finish;
1598
1599 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1600 if (err)
1601 goto out_finish;
1602
1603 err = -EINVAL;
1604 if (size - sizeof(outarg) != outarg.size)
1605 goto out_finish;
1606
1607 nodeid = outarg.nodeid;
1608
1609 down_read(&fc->killsb);
1610
1611 err = -ENOENT;
1612 if (!fc->sb)
1613 goto out_up_killsb;
1614
1615 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1616 if (!inode)
1617 goto out_up_killsb;
1618
1619 mapping = inode->i_mapping;
1620 index = outarg.offset >> PAGE_SHIFT;
1621 offset = outarg.offset & ~PAGE_MASK;
1622 file_size = i_size_read(inode);
1623 end = outarg.offset + outarg.size;
1624 if (end > file_size) {
1625 file_size = end;
1626 fuse_write_update_size(inode, file_size);
1627 }
1628
1629 num = outarg.size;
1630 while (num) {
1631 struct page *page;
1632 unsigned int this_num;
1633
1634 err = -ENOMEM;
1635 page = find_or_create_page(mapping, index,
1636 mapping_gfp_mask(mapping));
1637 if (!page)
1638 goto out_iput;
1639
1640 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1641 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1642 if (!err && offset == 0 &&
1643 (this_num == PAGE_SIZE || file_size == end))
1644 SetPageUptodate(page);
1645 unlock_page(page);
1646 put_page(page);
1647
1648 if (err)
1649 goto out_iput;
1650
1651 num -= this_num;
1652 offset = 0;
1653 index++;
1654 }
1655
1656 err = 0;
1657
1658 out_iput:
1659 iput(inode);
1660 out_up_killsb:
1661 up_read(&fc->killsb);
1662 out_finish:
1663 fuse_copy_finish(cs);
1664 return err;
1665 }
1666
1667 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1668 {
1669 release_pages(req->pages, req->num_pages);
1670 }
1671
1672 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1673 struct fuse_notify_retrieve_out *outarg)
1674 {
1675 int err;
1676 struct address_space *mapping = inode->i_mapping;
1677 struct fuse_req *req;
1678 pgoff_t index;
1679 loff_t file_size;
1680 unsigned int num;
1681 unsigned int offset;
1682 size_t total_len = 0;
1683 int num_pages;
1684
1685 offset = outarg->offset & ~PAGE_MASK;
1686 file_size = i_size_read(inode);
1687
1688 num = outarg->size;
1689 if (outarg->offset > file_size)
1690 num = 0;
1691 else if (outarg->offset + num > file_size)
1692 num = file_size - outarg->offset;
1693
1694 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1695 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1696
1697 req = fuse_get_req(fc, num_pages);
1698 if (IS_ERR(req))
1699 return PTR_ERR(req);
1700
1701 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1702 req->in.h.nodeid = outarg->nodeid;
1703 req->in.numargs = 2;
1704 req->in.argpages = 1;
1705 req->end = fuse_retrieve_end;
1706
1707 index = outarg->offset >> PAGE_SHIFT;
1708
1709 while (num && req->num_pages < num_pages) {
1710 struct page *page;
1711 unsigned int this_num;
1712
1713 page = find_get_page(mapping, index);
1714 if (!page)
1715 break;
1716
1717 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1718 req->pages[req->num_pages] = page;
1719 req->page_descs[req->num_pages].offset = offset;
1720 req->page_descs[req->num_pages].length = this_num;
1721 req->num_pages++;
1722
1723 offset = 0;
1724 num -= this_num;
1725 total_len += this_num;
1726 index++;
1727 }
1728 req->misc.retrieve_in.offset = outarg->offset;
1729 req->misc.retrieve_in.size = total_len;
1730 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1731 req->in.args[0].value = &req->misc.retrieve_in;
1732 req->in.args[1].size = total_len;
1733
1734 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1735 if (err) {
1736 fuse_retrieve_end(fc, req);
1737 fuse_put_request(fc, req);
1738 }
1739
1740 return err;
1741 }
1742
1743 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1744 struct fuse_copy_state *cs)
1745 {
1746 struct fuse_notify_retrieve_out outarg;
1747 struct inode *inode;
1748 int err;
1749
1750 err = -EINVAL;
1751 if (size != sizeof(outarg))
1752 goto copy_finish;
1753
1754 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1755 if (err)
1756 goto copy_finish;
1757
1758 fuse_copy_finish(cs);
1759
1760 down_read(&fc->killsb);
1761 err = -ENOENT;
1762 if (fc->sb) {
1763 u64 nodeid = outarg.nodeid;
1764
1765 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1766 if (inode) {
1767 err = fuse_retrieve(fc, inode, &outarg);
1768 iput(inode);
1769 }
1770 }
1771 up_read(&fc->killsb);
1772
1773 return err;
1774
1775 copy_finish:
1776 fuse_copy_finish(cs);
1777 return err;
1778 }
1779
1780 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1781 unsigned int size, struct fuse_copy_state *cs)
1782 {
1783 /* Don't try to move pages (yet) */
1784 cs->move_pages = 0;
1785
1786 switch (code) {
1787 case FUSE_NOTIFY_POLL:
1788 return fuse_notify_poll(fc, size, cs);
1789
1790 case FUSE_NOTIFY_INVAL_INODE:
1791 return fuse_notify_inval_inode(fc, size, cs);
1792
1793 case FUSE_NOTIFY_INVAL_ENTRY:
1794 return fuse_notify_inval_entry(fc, size, cs);
1795
1796 case FUSE_NOTIFY_STORE:
1797 return fuse_notify_store(fc, size, cs);
1798
1799 case FUSE_NOTIFY_RETRIEVE:
1800 return fuse_notify_retrieve(fc, size, cs);
1801
1802 case FUSE_NOTIFY_DELETE:
1803 return fuse_notify_delete(fc, size, cs);
1804
1805 default:
1806 fuse_copy_finish(cs);
1807 return -EINVAL;
1808 }
1809 }
1810
1811 /* Look up request on processing list by unique ID */
1812 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1813 {
1814 struct fuse_req *req;
1815
1816 list_for_each_entry(req, &fpq->processing, list) {
1817 if (req->in.h.unique == unique || req->intr_unique == unique)
1818 return req;
1819 }
1820 return NULL;
1821 }
1822
1823 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1824 unsigned nbytes)
1825 {
1826 unsigned reqsize = sizeof(struct fuse_out_header);
1827
1828 if (out->h.error)
1829 return nbytes != reqsize ? -EINVAL : 0;
1830
1831 reqsize += len_args(out->numargs, out->args);
1832
1833 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1834 return -EINVAL;
1835 else if (reqsize > nbytes) {
1836 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1837 unsigned diffsize = reqsize - nbytes;
1838 if (diffsize > lastarg->size)
1839 return -EINVAL;
1840 lastarg->size -= diffsize;
1841 }
1842 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1843 out->page_zeroing);
1844 }
1845
1846 /*
1847 * Write a single reply to a request. First the header is copied from
1848 * the write buffer. The request is then searched on the processing
1849 * list by the unique ID found in the header. If found, then remove
1850 * it from the list and copy the rest of the buffer to the request.
1851 * The request is finished by calling request_end()
1852 */
1853 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1854 struct fuse_copy_state *cs, size_t nbytes)
1855 {
1856 int err;
1857 struct fuse_conn *fc = fud->fc;
1858 struct fuse_pqueue *fpq = &fud->pq;
1859 struct fuse_req *req;
1860 struct fuse_out_header oh;
1861
1862 if (current_user_ns() != fc->user_ns)
1863 return -EIO;
1864
1865 if (nbytes < sizeof(struct fuse_out_header))
1866 return -EINVAL;
1867
1868 err = fuse_copy_one(cs, &oh, sizeof(oh));
1869 if (err)
1870 goto err_finish;
1871
1872 err = -EINVAL;
1873 if (oh.len != nbytes)
1874 goto err_finish;
1875
1876 /*
1877 * Zero oh.unique indicates unsolicited notification message
1878 * and error contains notification code.
1879 */
1880 if (!oh.unique) {
1881 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1882 return err ? err : nbytes;
1883 }
1884
1885 err = -EINVAL;
1886 if (oh.error <= -1000 || oh.error > 0)
1887 goto err_finish;
1888
1889 spin_lock(&fpq->lock);
1890 err = -ENOENT;
1891 if (!fpq->connected)
1892 goto err_unlock_pq;
1893
1894 req = request_find(fpq, oh.unique);
1895 if (!req)
1896 goto err_unlock_pq;
1897
1898 /* Is it an interrupt reply? */
1899 if (req->intr_unique == oh.unique) {
1900 __fuse_get_request(req);
1901 spin_unlock(&fpq->lock);
1902
1903 err = -EINVAL;
1904 if (nbytes != sizeof(struct fuse_out_header)) {
1905 fuse_put_request(fc, req);
1906 goto err_finish;
1907 }
1908
1909 if (oh.error == -ENOSYS)
1910 fc->no_interrupt = 1;
1911 else if (oh.error == -EAGAIN)
1912 queue_interrupt(&fc->iq, req);
1913 fuse_put_request(fc, req);
1914
1915 fuse_copy_finish(cs);
1916 return nbytes;
1917 }
1918
1919 clear_bit(FR_SENT, &req->flags);
1920 list_move(&req->list, &fpq->io);
1921 req->out.h = oh;
1922 set_bit(FR_LOCKED, &req->flags);
1923 spin_unlock(&fpq->lock);
1924 cs->req = req;
1925 if (!req->out.page_replace)
1926 cs->move_pages = 0;
1927
1928 err = copy_out_args(cs, &req->out, nbytes);
1929 fuse_copy_finish(cs);
1930
1931 spin_lock(&fpq->lock);
1932 clear_bit(FR_LOCKED, &req->flags);
1933 if (!fpq->connected)
1934 err = -ENOENT;
1935 else if (err)
1936 req->out.h.error = -EIO;
1937 if (!test_bit(FR_PRIVATE, &req->flags))
1938 list_del_init(&req->list);
1939 spin_unlock(&fpq->lock);
1940
1941 request_end(fc, req);
1942
1943 return err ? err : nbytes;
1944
1945 err_unlock_pq:
1946 spin_unlock(&fpq->lock);
1947 err_finish:
1948 fuse_copy_finish(cs);
1949 return err;
1950 }
1951
1952 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1953 {
1954 struct fuse_copy_state cs;
1955 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1956
1957 if (!fud)
1958 return -EPERM;
1959
1960 if (!iter_is_iovec(from))
1961 return -EINVAL;
1962
1963 fuse_copy_init(&cs, 0, from);
1964
1965 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1966 }
1967
1968 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1969 struct file *out, loff_t *ppos,
1970 size_t len, unsigned int flags)
1971 {
1972 unsigned nbuf;
1973 unsigned idx;
1974 struct pipe_buffer *bufs;
1975 struct fuse_copy_state cs;
1976 struct fuse_dev *fud;
1977 size_t rem;
1978 ssize_t ret;
1979
1980 fud = fuse_get_dev(out);
1981 if (!fud)
1982 return -EPERM;
1983
1984 pipe_lock(pipe);
1985
1986 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1987 if (!bufs) {
1988 pipe_unlock(pipe);
1989 return -ENOMEM;
1990 }
1991
1992 nbuf = 0;
1993 rem = 0;
1994 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1995 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1996
1997 ret = -EINVAL;
1998 if (rem < len) {
1999 pipe_unlock(pipe);
2000 goto out;
2001 }
2002
2003 rem = len;
2004 while (rem) {
2005 struct pipe_buffer *ibuf;
2006 struct pipe_buffer *obuf;
2007
2008 BUG_ON(nbuf >= pipe->buffers);
2009 BUG_ON(!pipe->nrbufs);
2010 ibuf = &pipe->bufs[pipe->curbuf];
2011 obuf = &bufs[nbuf];
2012
2013 if (rem >= ibuf->len) {
2014 *obuf = *ibuf;
2015 ibuf->ops = NULL;
2016 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2017 pipe->nrbufs--;
2018 } else {
2019 pipe_buf_get(pipe, ibuf);
2020 *obuf = *ibuf;
2021 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2022 obuf->len = rem;
2023 ibuf->offset += obuf->len;
2024 ibuf->len -= obuf->len;
2025 }
2026 nbuf++;
2027 rem -= obuf->len;
2028 }
2029 pipe_unlock(pipe);
2030
2031 fuse_copy_init(&cs, 0, NULL);
2032 cs.pipebufs = bufs;
2033 cs.nr_segs = nbuf;
2034 cs.pipe = pipe;
2035
2036 if (flags & SPLICE_F_MOVE)
2037 cs.move_pages = 1;
2038
2039 ret = fuse_dev_do_write(fud, &cs, len);
2040
2041 pipe_lock(pipe);
2042 for (idx = 0; idx < nbuf; idx++)
2043 pipe_buf_release(pipe, &bufs[idx]);
2044 pipe_unlock(pipe);
2045
2046 out:
2047 kfree(bufs);
2048 return ret;
2049 }
2050
2051 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2052 {
2053 unsigned mask = POLLOUT | POLLWRNORM;
2054 struct fuse_iqueue *fiq;
2055 struct fuse_dev *fud = fuse_get_dev(file);
2056
2057 if (!fud)
2058 return POLLERR;
2059
2060 fiq = &fud->fc->iq;
2061 poll_wait(file, &fiq->waitq, wait);
2062
2063 spin_lock(&fiq->waitq.lock);
2064 if (!fiq->connected)
2065 mask = POLLERR;
2066 else if (request_pending(fiq))
2067 mask |= POLLIN | POLLRDNORM;
2068 spin_unlock(&fiq->waitq.lock);
2069
2070 return mask;
2071 }
2072
2073 /*
2074 * Abort all requests on the given list (pending or processing)
2075 *
2076 * This function releases and reacquires fc->lock
2077 */
2078 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2079 {
2080 while (!list_empty(head)) {
2081 struct fuse_req *req;
2082 req = list_entry(head->next, struct fuse_req, list);
2083 req->out.h.error = -ECONNABORTED;
2084 clear_bit(FR_SENT, &req->flags);
2085 list_del_init(&req->list);
2086 request_end(fc, req);
2087 }
2088 }
2089
2090 static void end_polls(struct fuse_conn *fc)
2091 {
2092 struct rb_node *p;
2093
2094 p = rb_first(&fc->polled_files);
2095
2096 while (p) {
2097 struct fuse_file *ff;
2098 ff = rb_entry(p, struct fuse_file, polled_node);
2099 wake_up_interruptible_all(&ff->poll_wait);
2100
2101 p = rb_next(p);
2102 }
2103 }
2104
2105 /*
2106 * Abort all requests.
2107 *
2108 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2109 * filesystem.
2110 *
2111 * The same effect is usually achievable through killing the filesystem daemon
2112 * and all users of the filesystem. The exception is the combination of an
2113 * asynchronous request and the tricky deadlock (see
2114 * Documentation/filesystems/fuse.txt).
2115 *
2116 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2117 * requests, they should be finished off immediately. Locked requests will be
2118 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2119 * requests. It is possible that some request will finish before we can. This
2120 * is OK, the request will in that case be removed from the list before we touch
2121 * it.
2122 */
2123 void fuse_abort_conn(struct fuse_conn *fc)
2124 {
2125 struct fuse_iqueue *fiq = &fc->iq;
2126
2127 spin_lock(&fc->lock);
2128 if (fc->connected) {
2129 struct fuse_dev *fud;
2130 struct fuse_req *req, *next;
2131 LIST_HEAD(to_end1);
2132 LIST_HEAD(to_end2);
2133
2134 fc->connected = 0;
2135 fc->blocked = 0;
2136 fuse_set_initialized(fc);
2137 list_for_each_entry(fud, &fc->devices, entry) {
2138 struct fuse_pqueue *fpq = &fud->pq;
2139
2140 spin_lock(&fpq->lock);
2141 fpq->connected = 0;
2142 list_for_each_entry_safe(req, next, &fpq->io, list) {
2143 req->out.h.error = -ECONNABORTED;
2144 spin_lock(&req->waitq.lock);
2145 set_bit(FR_ABORTED, &req->flags);
2146 if (!test_bit(FR_LOCKED, &req->flags)) {
2147 set_bit(FR_PRIVATE, &req->flags);
2148 __fuse_get_request(req);
2149 list_move(&req->list, &to_end1);
2150 }
2151 spin_unlock(&req->waitq.lock);
2152 }
2153 list_splice_init(&fpq->processing, &to_end2);
2154 spin_unlock(&fpq->lock);
2155 }
2156 fc->max_background = UINT_MAX;
2157 flush_bg_queue(fc);
2158
2159 spin_lock(&fiq->waitq.lock);
2160 fiq->connected = 0;
2161 list_splice_init(&fiq->pending, &to_end2);
2162 list_for_each_entry(req, &to_end2, list)
2163 clear_bit(FR_PENDING, &req->flags);
2164 while (forget_pending(fiq))
2165 kfree(dequeue_forget(fiq, 1, NULL));
2166 wake_up_all_locked(&fiq->waitq);
2167 spin_unlock(&fiq->waitq.lock);
2168 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2169 end_polls(fc);
2170 wake_up_all(&fc->blocked_waitq);
2171 spin_unlock(&fc->lock);
2172
2173 while (!list_empty(&to_end1)) {
2174 req = list_first_entry(&to_end1, struct fuse_req, list);
2175 list_del_init(&req->list);
2176 request_end(fc, req);
2177 }
2178 end_requests(fc, &to_end2);
2179 } else {
2180 spin_unlock(&fc->lock);
2181 }
2182 }
2183 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2184
2185 void fuse_wait_aborted(struct fuse_conn *fc)
2186 {
2187 /* matches implicit memory barrier in fuse_drop_waiting() */
2188 smp_mb();
2189 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2190 }
2191
2192 int fuse_dev_release(struct inode *inode, struct file *file)
2193 {
2194 struct fuse_dev *fud = fuse_get_dev(file);
2195
2196 if (fud) {
2197 struct fuse_conn *fc = fud->fc;
2198 struct fuse_pqueue *fpq = &fud->pq;
2199 LIST_HEAD(to_end);
2200
2201 spin_lock(&fpq->lock);
2202 WARN_ON(!list_empty(&fpq->io));
2203 list_splice_init(&fpq->processing, &to_end);
2204 spin_unlock(&fpq->lock);
2205
2206 end_requests(fc, &to_end);
2207
2208 /* Are we the last open device? */
2209 if (atomic_dec_and_test(&fc->dev_count)) {
2210 WARN_ON(fc->iq.fasync != NULL);
2211 fuse_abort_conn(fc);
2212 }
2213 fuse_dev_free(fud);
2214 }
2215 return 0;
2216 }
2217 EXPORT_SYMBOL_GPL(fuse_dev_release);
2218
2219 static int fuse_dev_fasync(int fd, struct file *file, int on)
2220 {
2221 struct fuse_dev *fud = fuse_get_dev(file);
2222
2223 if (!fud)
2224 return -EPERM;
2225
2226 /* No locking - fasync_helper does its own locking */
2227 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2228 }
2229
2230 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2231 {
2232 struct fuse_dev *fud;
2233
2234 if (new->private_data)
2235 return -EINVAL;
2236
2237 fud = fuse_dev_alloc(fc);
2238 if (!fud)
2239 return -ENOMEM;
2240
2241 new->private_data = fud;
2242 atomic_inc(&fc->dev_count);
2243
2244 return 0;
2245 }
2246
2247 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2248 unsigned long arg)
2249 {
2250 int err = -ENOTTY;
2251
2252 if (cmd == FUSE_DEV_IOC_CLONE) {
2253 int oldfd;
2254
2255 err = -EFAULT;
2256 if (!get_user(oldfd, (__u32 __user *) arg)) {
2257 struct file *old = fget(oldfd);
2258
2259 err = -EINVAL;
2260 if (old) {
2261 struct fuse_dev *fud = NULL;
2262
2263 /*
2264 * Check against file->f_op because CUSE
2265 * uses the same ioctl handler.
2266 */
2267 if (old->f_op == file->f_op &&
2268 old->f_cred->user_ns == file->f_cred->user_ns)
2269 fud = fuse_get_dev(old);
2270
2271 if (fud) {
2272 mutex_lock(&fuse_mutex);
2273 err = fuse_device_clone(fud->fc, file);
2274 mutex_unlock(&fuse_mutex);
2275 }
2276 fput(old);
2277 }
2278 }
2279 }
2280 return err;
2281 }
2282
2283 const struct file_operations fuse_dev_operations = {
2284 .owner = THIS_MODULE,
2285 .open = fuse_dev_open,
2286 .llseek = no_llseek,
2287 .read_iter = fuse_dev_read,
2288 .splice_read = fuse_dev_splice_read,
2289 .write_iter = fuse_dev_write,
2290 .splice_write = fuse_dev_splice_write,
2291 .poll = fuse_dev_poll,
2292 .release = fuse_dev_release,
2293 .fasync = fuse_dev_fasync,
2294 .unlocked_ioctl = fuse_dev_ioctl,
2295 .compat_ioctl = fuse_dev_ioctl,
2296 };
2297 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2298
2299 static struct miscdevice fuse_miscdevice = {
2300 .minor = FUSE_MINOR,
2301 .name = "fuse",
2302 .fops = &fuse_dev_operations,
2303 };
2304
2305 int __init fuse_dev_init(void)
2306 {
2307 int err = -ENOMEM;
2308 fuse_req_cachep = kmem_cache_create("fuse_request",
2309 sizeof(struct fuse_req),
2310 0, 0, NULL);
2311 if (!fuse_req_cachep)
2312 goto out;
2313
2314 err = misc_register(&fuse_miscdevice);
2315 if (err)
2316 goto out_cache_clean;
2317
2318 return 0;
2319
2320 out_cache_clean:
2321 kmem_cache_destroy(fuse_req_cachep);
2322 out:
2323 return err;
2324 }
2325
2326 void fuse_dev_cleanup(void)
2327 {
2328 misc_deregister(&fuse_miscdevice);
2329 kmem_cache_destroy(fuse_req_cachep);
2330 }
ubuntu-bionic-kernel mirror