projects / GitHub / moto-9609 / android_kernel_motorola_exynos9610.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge 4.14.7 into android-4.14
[GitHub/moto-9609/android_kernel_motorola_exynos9610.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 #include <linux/freezer.h>
25
26 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
27 MODULE_ALIAS("devname:fuse");
28
29 static struct kmem_cache *fuse_req_cachep;
30
31 static struct fuse_dev *fuse_get_dev(struct file *file)
32 {
33 /*
34 * Lockless access is OK, because file->private data is set
35 * once during mount and is valid until the file is released.
36 */
37 return ACCESS_ONCE(file->private_data);
38 }
39
40 static void fuse_request_init(struct fuse_req *req, struct page **pages,
41 struct fuse_page_desc *page_descs,
42 unsigned npages)
43 {
44 memset(req, 0, sizeof(*req));
45 memset(pages, 0, sizeof(*pages) * npages);
46 memset(page_descs, 0, sizeof(*page_descs) * npages);
47 INIT_LIST_HEAD(&req->list);
48 INIT_LIST_HEAD(&req->intr_entry);
49 init_waitqueue_head(&req->waitq);
50 refcount_set(&req->count, 1);
51 req->pages = pages;
52 req->page_descs = page_descs;
53 req->max_pages = npages;
54 __set_bit(FR_PENDING, &req->flags);
55 }
56
57 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
58 {
59 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
60 if (req) {
61 struct page **pages;
62 struct fuse_page_desc *page_descs;
63
64 if (npages <= FUSE_REQ_INLINE_PAGES) {
65 pages = req->inline_pages;
66 page_descs = req->inline_page_descs;
67 } else {
68 pages = kmalloc(sizeof(struct page *) * npages, flags);
69 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
70 npages, flags);
71 }
72
73 if (!pages || !page_descs) {
74 kfree(pages);
75 kfree(page_descs);
76 kmem_cache_free(fuse_req_cachep, req);
77 return NULL;
78 }
79
80 fuse_request_init(req, pages, page_descs, npages);
81 }
82 return req;
83 }
84
85 struct fuse_req *fuse_request_alloc(unsigned npages)
86 {
87 return __fuse_request_alloc(npages, GFP_KERNEL);
88 }
89 EXPORT_SYMBOL_GPL(fuse_request_alloc);
90
91 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
92 {
93 return __fuse_request_alloc(npages, GFP_NOFS);
94 }
95
96 void fuse_request_free(struct fuse_req *req)
97 {
98 if (req->pages != req->inline_pages) {
99 kfree(req->pages);
100 kfree(req->page_descs);
101 }
102 kmem_cache_free(fuse_req_cachep, req);
103 }
104
105 void __fuse_get_request(struct fuse_req *req)
106 {
107 refcount_inc(&req->count);
108 }
109
110 /* Must be called with > 1 refcount */
111 static void __fuse_put_request(struct fuse_req *req)
112 {
113 refcount_dec(&req->count);
114 }
115
116 static void fuse_req_init_context(struct fuse_conn *fc, struct fuse_req *req)
117 {
118 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
119 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
120 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
121 }
122
123 void fuse_set_initialized(struct fuse_conn *fc)
124 {
125 /* Make sure stores before this are seen on another CPU */
126 smp_wmb();
127 fc->initialized = 1;
128 }
129
130 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
131 {
132 return !fc->initialized || (for_background && fc->blocked);
133 }
134
135 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
136 bool for_background)
137 {
138 struct fuse_req *req;
139 int err;
140 atomic_inc(&fc->num_waiting);
141
142 if (fuse_block_alloc(fc, for_background)) {
143 err = -EINTR;
144 if (wait_event_killable_exclusive(fc->blocked_waitq,
145 !fuse_block_alloc(fc, for_background)))
146 goto out;
147 }
148 /* Matches smp_wmb() in fuse_set_initialized() */
149 smp_rmb();
150
151 err = -ENOTCONN;
152 if (!fc->connected)
153 goto out;
154
155 err = -ECONNREFUSED;
156 if (fc->conn_error)
157 goto out;
158
159 req = fuse_request_alloc(npages);
160 err = -ENOMEM;
161 if (!req) {
162 if (for_background)
163 wake_up(&fc->blocked_waitq);
164 goto out;
165 }
166
167 fuse_req_init_context(fc, req);
168 __set_bit(FR_WAITING, &req->flags);
169 if (for_background)
170 __set_bit(FR_BACKGROUND, &req->flags);
171
172 return req;
173
174 out:
175 atomic_dec(&fc->num_waiting);
176 return ERR_PTR(err);
177 }
178
179 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
180 {
181 return __fuse_get_req(fc, npages, false);
182 }
183 EXPORT_SYMBOL_GPL(fuse_get_req);
184
185 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
186 unsigned npages)
187 {
188 return __fuse_get_req(fc, npages, true);
189 }
190 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
191
192 /*
193 * Return request in fuse_file->reserved_req. However that may
194 * currently be in use. If that is the case, wait for it to become
195 * available.
196 */
197 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
198 struct file *file)
199 {
200 struct fuse_req *req = NULL;
201 struct fuse_file *ff = file->private_data;
202
203 do {
204 wait_event(fc->reserved_req_waitq, ff->reserved_req);
205 spin_lock(&fc->lock);
206 if (ff->reserved_req) {
207 req = ff->reserved_req;
208 ff->reserved_req = NULL;
209 req->stolen_file = get_file(file);
210 }
211 spin_unlock(&fc->lock);
212 } while (!req);
213
214 return req;
215 }
216
217 /*
218 * Put stolen request back into fuse_file->reserved_req
219 */
220 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
221 {
222 struct file *file = req->stolen_file;
223 struct fuse_file *ff = file->private_data;
224
225 spin_lock(&fc->lock);
226 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
227 BUG_ON(ff->reserved_req);
228 ff->reserved_req = req;
229 wake_up_all(&fc->reserved_req_waitq);
230 spin_unlock(&fc->lock);
231 fput(file);
232 }
233
234 /*
235 * Gets a requests for a file operation, always succeeds
236 *
237 * This is used for sending the FLUSH request, which must get to
238 * userspace, due to POSIX locks which may need to be unlocked.
239 *
240 * If allocation fails due to OOM, use the reserved request in
241 * fuse_file.
242 *
243 * This is very unlikely to deadlock accidentally, since the
244 * filesystem should not have it's own file open. If deadlock is
245 * intentional, it can still be broken by "aborting" the filesystem.
246 */
247 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
248 struct file *file)
249 {
250 struct fuse_req *req;
251
252 atomic_inc(&fc->num_waiting);
253 wait_event(fc->blocked_waitq, fc->initialized);
254 /* Matches smp_wmb() in fuse_set_initialized() */
255 smp_rmb();
256 req = fuse_request_alloc(0);
257 if (!req)
258 req = get_reserved_req(fc, file);
259
260 fuse_req_init_context(fc, req);
261 __set_bit(FR_WAITING, &req->flags);
262 __clear_bit(FR_BACKGROUND, &req->flags);
263 return req;
264 }
265
266 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
267 {
268 if (refcount_dec_and_test(&req->count)) {
269 if (test_bit(FR_BACKGROUND, &req->flags)) {
270 /*
271 * We get here in the unlikely case that a background
272 * request was allocated but not sent
273 */
274 spin_lock(&fc->lock);
275 if (!fc->blocked)
276 wake_up(&fc->blocked_waitq);
277 spin_unlock(&fc->lock);
278 }
279
280 if (test_bit(FR_WAITING, &req->flags)) {
281 __clear_bit(FR_WAITING, &req->flags);
282 atomic_dec(&fc->num_waiting);
283 }
284
285 if (req->stolen_file)
286 put_reserved_req(fc, req);
287 else
288 fuse_request_free(req);
289 }
290 }
291 EXPORT_SYMBOL_GPL(fuse_put_request);
292
293 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
294 {
295 unsigned nbytes = 0;
296 unsigned i;
297
298 for (i = 0; i < numargs; i++)
299 nbytes += args[i].size;
300
301 return nbytes;
302 }
303
304 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
305 {
306 return ++fiq->reqctr;
307 }
308
309 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
310 {
311 req->in.h.len = sizeof(struct fuse_in_header) +
312 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
313 list_add_tail(&req->list, &fiq->pending);
314 wake_up_locked(&fiq->waitq);
315 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
316 }
317
318 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
319 u64 nodeid, u64 nlookup)
320 {
321 struct fuse_iqueue *fiq = &fc->iq;
322
323 forget->forget_one.nodeid = nodeid;
324 forget->forget_one.nlookup = nlookup;
325
326 spin_lock(&fiq->waitq.lock);
327 if (fiq->connected) {
328 fiq->forget_list_tail->next = forget;
329 fiq->forget_list_tail = forget;
330 wake_up_locked(&fiq->waitq);
331 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
332 } else {
333 kfree(forget);
334 }
335 spin_unlock(&fiq->waitq.lock);
336 }
337
338 static void flush_bg_queue(struct fuse_conn *fc)
339 {
340 while (fc->active_background < fc->max_background &&
341 !list_empty(&fc->bg_queue)) {
342 struct fuse_req *req;
343 struct fuse_iqueue *fiq = &fc->iq;
344
345 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
346 list_del(&req->list);
347 fc->active_background++;
348 spin_lock(&fiq->waitq.lock);
349 req->in.h.unique = fuse_get_unique(fiq);
350 queue_request(fiq, req);
351 spin_unlock(&fiq->waitq.lock);
352 }
353 }
354
355 /*
356 * This function is called when a request is finished. Either a reply
357 * has arrived or it was aborted (and not yet sent) or some error
358 * occurred during communication with userspace, or the device file
359 * was closed. The requester thread is woken up (if still waiting),
360 * the 'end' callback is called if given, else the reference to the
361 * request is released
362 */
363 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
364 {
365 struct fuse_iqueue *fiq = &fc->iq;
366
367 if (test_and_set_bit(FR_FINISHED, &req->flags))
368 return;
369
370 spin_lock(&fiq->waitq.lock);
371 list_del_init(&req->intr_entry);
372 spin_unlock(&fiq->waitq.lock);
373 WARN_ON(test_bit(FR_PENDING, &req->flags));
374 WARN_ON(test_bit(FR_SENT, &req->flags));
375 if (test_bit(FR_BACKGROUND, &req->flags)) {
376 spin_lock(&fc->lock);
377 clear_bit(FR_BACKGROUND, &req->flags);
378 if (fc->num_background == fc->max_background)
379 fc->blocked = 0;
380
381 /* Wake up next waiter, if any */
382 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
383 wake_up(&fc->blocked_waitq);
384
385 if (fc->num_background == fc->congestion_threshold &&
386 fc->connected && fc->sb) {
387 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
388 clear_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
389 }
390 fc->num_background--;
391 fc->active_background--;
392 flush_bg_queue(fc);
393 spin_unlock(&fc->lock);
394 }
395 wake_up(&req->waitq);
396 if (req->end)
397 req->end(fc, req);
398 fuse_put_request(fc, req);
399 }
400
401 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
402 {
403 spin_lock(&fiq->waitq.lock);
404 if (test_bit(FR_FINISHED, &req->flags)) {
405 spin_unlock(&fiq->waitq.lock);
406 return;
407 }
408 if (list_empty(&req->intr_entry)) {
409 list_add_tail(&req->intr_entry, &fiq->interrupts);
410 wake_up_locked(&fiq->waitq);
411 }
412 spin_unlock(&fiq->waitq.lock);
413 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
414 }
415
416 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
417 {
418 struct fuse_iqueue *fiq = &fc->iq;
419 int err;
420
421 if (!fc->no_interrupt) {
422 /* Any signal may interrupt this */
423 err = wait_event_interruptible(req->waitq,
424 test_bit(FR_FINISHED, &req->flags));
425 if (!err)
426 return;
427
428 set_bit(FR_INTERRUPTED, &req->flags);
429 /* matches barrier in fuse_dev_do_read() */
430 smp_mb__after_atomic();
431 if (test_bit(FR_SENT, &req->flags))
432 queue_interrupt(fiq, req);
433 }
434
435 if (!test_bit(FR_FORCE, &req->flags)) {
436 /* Only fatal signals may interrupt this */
437 err = wait_event_killable(req->waitq,
438 test_bit(FR_FINISHED, &req->flags));
439 if (!err)
440 return;
441
442 spin_lock(&fiq->waitq.lock);
443 /* Request is not yet in userspace, bail out */
444 if (test_bit(FR_PENDING, &req->flags)) {
445 list_del(&req->list);
446 spin_unlock(&fiq->waitq.lock);
447 __fuse_put_request(req);
448 req->out.h.error = -EINTR;
449 return;
450 }
451 spin_unlock(&fiq->waitq.lock);
452 }
453
454 /*
455 * Either request is already in userspace, or it was forced.
456 * Wait it out.
457 */
458 while (!test_bit(FR_FINISHED, &req->flags))
459 wait_event_freezable(req->waitq,
460 test_bit(FR_FINISHED, &req->flags));
461 }
462
463 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
464 {
465 struct fuse_iqueue *fiq = &fc->iq;
466
467 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
468 spin_lock(&fiq->waitq.lock);
469 if (!fiq->connected) {
470 spin_unlock(&fiq->waitq.lock);
471 req->out.h.error = -ENOTCONN;
472 } else {
473 req->in.h.unique = fuse_get_unique(fiq);
474 queue_request(fiq, req);
475 /* acquire extra reference, since request is still needed
476 after request_end() */
477 __fuse_get_request(req);
478 spin_unlock(&fiq->waitq.lock);
479
480 request_wait_answer(fc, req);
481 /* Pairs with smp_wmb() in request_end() */
482 smp_rmb();
483 }
484 }
485
486 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
487 {
488 __set_bit(FR_ISREPLY, &req->flags);
489 if (!test_bit(FR_WAITING, &req->flags)) {
490 __set_bit(FR_WAITING, &req->flags);
491 atomic_inc(&fc->num_waiting);
492 }
493 __fuse_request_send(fc, req);
494 }
495 EXPORT_SYMBOL_GPL(fuse_request_send);
496
497 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
498 {
499 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
500 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
501
502 if (fc->minor < 9) {
503 switch (args->in.h.opcode) {
504 case FUSE_LOOKUP:
505 case FUSE_CREATE:
506 case FUSE_MKNOD:
507 case FUSE_MKDIR:
508 case FUSE_SYMLINK:
509 case FUSE_LINK:
510 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
511 break;
512 case FUSE_GETATTR:
513 case FUSE_SETATTR:
514 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
515 break;
516 }
517 }
518 if (fc->minor < 12) {
519 switch (args->in.h.opcode) {
520 case FUSE_CREATE:
521 args->in.args[0].size = sizeof(struct fuse_open_in);
522 break;
523 case FUSE_MKNOD:
524 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
525 break;
526 }
527 }
528 }
529
530 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
531 {
532 struct fuse_req *req;
533 ssize_t ret;
534
535 req = fuse_get_req(fc, 0);
536 if (IS_ERR(req))
537 return PTR_ERR(req);
538
539 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
540 fuse_adjust_compat(fc, args);
541
542 req->in.h.opcode = args->in.h.opcode;
543 req->in.h.nodeid = args->in.h.nodeid;
544 req->in.numargs = args->in.numargs;
545 memcpy(req->in.args, args->in.args,
546 args->in.numargs * sizeof(struct fuse_in_arg));
547 req->out.argvar = args->out.argvar;
548 req->out.numargs = args->out.numargs;
549 memcpy(req->out.args, args->out.args,
550 args->out.numargs * sizeof(struct fuse_arg));
551 fuse_request_send(fc, req);
552 ret = req->out.h.error;
553 if (!ret && args->out.argvar) {
554 BUG_ON(args->out.numargs != 1);
555 ret = req->out.args[0].size;
556 }
557 fuse_put_request(fc, req);
558
559 return ret;
560 }
561
562 /*
563 * Called under fc->lock
564 *
565 * fc->connected must have been checked previously
566 */
567 void fuse_request_send_background_locked(struct fuse_conn *fc,
568 struct fuse_req *req)
569 {
570 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
571 if (!test_bit(FR_WAITING, &req->flags)) {
572 __set_bit(FR_WAITING, &req->flags);
573 atomic_inc(&fc->num_waiting);
574 }
575 __set_bit(FR_ISREPLY, &req->flags);
576 fc->num_background++;
577 if (fc->num_background == fc->max_background)
578 fc->blocked = 1;
579 if (fc->num_background == fc->congestion_threshold && fc->sb) {
580 set_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
581 set_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
582 }
583 list_add_tail(&req->list, &fc->bg_queue);
584 flush_bg_queue(fc);
585 }
586
587 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
588 {
589 BUG_ON(!req->end);
590 spin_lock(&fc->lock);
591 if (fc->connected) {
592 fuse_request_send_background_locked(fc, req);
593 spin_unlock(&fc->lock);
594 } else {
595 spin_unlock(&fc->lock);
596 req->out.h.error = -ENOTCONN;
597 req->end(fc, req);
598 fuse_put_request(fc, req);
599 }
600 }
601 EXPORT_SYMBOL_GPL(fuse_request_send_background);
602
603 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
604 struct fuse_req *req, u64 unique)
605 {
606 int err = -ENODEV;
607 struct fuse_iqueue *fiq = &fc->iq;
608
609 __clear_bit(FR_ISREPLY, &req->flags);
610 req->in.h.unique = unique;
611 spin_lock(&fiq->waitq.lock);
612 if (fiq->connected) {
613 queue_request(fiq, req);
614 err = 0;
615 }
616 spin_unlock(&fiq->waitq.lock);
617
618 return err;
619 }
620
621 void fuse_force_forget(struct file *file, u64 nodeid)
622 {
623 struct inode *inode = file_inode(file);
624 struct fuse_conn *fc = get_fuse_conn(inode);
625 struct fuse_req *req;
626 struct fuse_forget_in inarg;
627
628 memset(&inarg, 0, sizeof(inarg));
629 inarg.nlookup = 1;
630 req = fuse_get_req_nofail_nopages(fc, file);
631 req->in.h.opcode = FUSE_FORGET;
632 req->in.h.nodeid = nodeid;
633 req->in.numargs = 1;
634 req->in.args[0].size = sizeof(inarg);
635 req->in.args[0].value = &inarg;
636 __clear_bit(FR_ISREPLY, &req->flags);
637 __fuse_request_send(fc, req);
638 /* ignore errors */
639 fuse_put_request(fc, req);
640 }
641
642 /*
643 * Lock the request. Up to the next unlock_request() there mustn't be
644 * anything that could cause a page-fault. If the request was already
645 * aborted bail out.
646 */
647 static int lock_request(struct fuse_req *req)
648 {
649 int err = 0;
650 if (req) {
651 spin_lock(&req->waitq.lock);
652 if (test_bit(FR_ABORTED, &req->flags))
653 err = -ENOENT;
654 else
655 set_bit(FR_LOCKED, &req->flags);
656 spin_unlock(&req->waitq.lock);
657 }
658 return err;
659 }
660
661 /*
662 * Unlock request. If it was aborted while locked, caller is responsible
663 * for unlocking and ending the request.
664 */
665 static int unlock_request(struct fuse_req *req)
666 {
667 int err = 0;
668 if (req) {
669 spin_lock(&req->waitq.lock);
670 if (test_bit(FR_ABORTED, &req->flags))
671 err = -ENOENT;
672 else
673 clear_bit(FR_LOCKED, &req->flags);
674 spin_unlock(&req->waitq.lock);
675 }
676 return err;
677 }
678
679 struct fuse_copy_state {
680 int write;
681 struct fuse_req *req;
682 struct iov_iter *iter;
683 struct pipe_buffer *pipebufs;
684 struct pipe_buffer *currbuf;
685 struct pipe_inode_info *pipe;
686 unsigned long nr_segs;
687 struct page *pg;
688 unsigned len;
689 unsigned offset;
690 unsigned move_pages:1;
691 };
692
693 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
694 struct iov_iter *iter)
695 {
696 memset(cs, 0, sizeof(*cs));
697 cs->write = write;
698 cs->iter = iter;
699 }
700
701 /* Unmap and put previous page of userspace buffer */
702 static void fuse_copy_finish(struct fuse_copy_state *cs)
703 {
704 if (cs->currbuf) {
705 struct pipe_buffer *buf = cs->currbuf;
706
707 if (cs->write)
708 buf->len = PAGE_SIZE - cs->len;
709 cs->currbuf = NULL;
710 } else if (cs->pg) {
711 if (cs->write) {
712 flush_dcache_page(cs->pg);
713 set_page_dirty_lock(cs->pg);
714 }
715 put_page(cs->pg);
716 }
717 cs->pg = NULL;
718 }
719
720 /*
721 * Get another pagefull of userspace buffer, and map it to kernel
722 * address space, and lock request
723 */
724 static int fuse_copy_fill(struct fuse_copy_state *cs)
725 {
726 struct page *page;
727 int err;
728
729 err = unlock_request(cs->req);
730 if (err)
731 return err;
732
733 fuse_copy_finish(cs);
734 if (cs->pipebufs) {
735 struct pipe_buffer *buf = cs->pipebufs;
736
737 if (!cs->write) {
738 err = pipe_buf_confirm(cs->pipe, buf);
739 if (err)
740 return err;
741
742 BUG_ON(!cs->nr_segs);
743 cs->currbuf = buf;
744 cs->pg = buf->page;
745 cs->offset = buf->offset;
746 cs->len = buf->len;
747 cs->pipebufs++;
748 cs->nr_segs--;
749 } else {
750 if (cs->nr_segs == cs->pipe->buffers)
751 return -EIO;
752
753 page = alloc_page(GFP_HIGHUSER);
754 if (!page)
755 return -ENOMEM;
756
757 buf->page = page;
758 buf->offset = 0;
759 buf->len = 0;
760
761 cs->currbuf = buf;
762 cs->pg = page;
763 cs->offset = 0;
764 cs->len = PAGE_SIZE;
765 cs->pipebufs++;
766 cs->nr_segs++;
767 }
768 } else {
769 size_t off;
770 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
771 if (err < 0)
772 return err;
773 BUG_ON(!err);
774 cs->len = err;
775 cs->offset = off;
776 cs->pg = page;
777 iov_iter_advance(cs->iter, err);
778 }
779
780 return lock_request(cs->req);
781 }
782
783 /* Do as much copy to/from userspace buffer as we can */
784 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
785 {
786 unsigned ncpy = min(*size, cs->len);
787 if (val) {
788 void *pgaddr = kmap_atomic(cs->pg);
789 void *buf = pgaddr + cs->offset;
790
791 if (cs->write)
792 memcpy(buf, *val, ncpy);
793 else
794 memcpy(*val, buf, ncpy);
795
796 kunmap_atomic(pgaddr);
797 *val += ncpy;
798 }
799 *size -= ncpy;
800 cs->len -= ncpy;
801 cs->offset += ncpy;
802 return ncpy;
803 }
804
805 static int fuse_check_page(struct page *page)
806 {
807 if (page_mapcount(page) ||
808 page->mapping != NULL ||
809 page_count(page) != 1 ||
810 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
811 ~(1 << PG_locked |
812 1 << PG_referenced |
813 1 << PG_uptodate |
814 1 << PG_lru |
815 1 << PG_active |
816 1 << PG_reclaim))) {
817 printk(KERN_WARNING "fuse: trying to steal weird page\n");
818 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);
819 return 1;
820 }
821 return 0;
822 }
823
824 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
825 {
826 int err;
827 struct page *oldpage = *pagep;
828 struct page *newpage;
829 struct pipe_buffer *buf = cs->pipebufs;
830
831 err = unlock_request(cs->req);
832 if (err)
833 return err;
834
835 fuse_copy_finish(cs);
836
837 err = pipe_buf_confirm(cs->pipe, buf);
838 if (err)
839 return err;
840
841 BUG_ON(!cs->nr_segs);
842 cs->currbuf = buf;
843 cs->len = buf->len;
844 cs->pipebufs++;
845 cs->nr_segs--;
846
847 if (cs->len != PAGE_SIZE)
848 goto out_fallback;
849
850 if (pipe_buf_steal(cs->pipe, buf) != 0)
851 goto out_fallback;
852
853 newpage = buf->page;
854
855 if (!PageUptodate(newpage))
856 SetPageUptodate(newpage);
857
858 ClearPageMappedToDisk(newpage);
859
860 if (fuse_check_page(newpage) != 0)
861 goto out_fallback_unlock;
862
863 /*
864 * This is a new and locked page, it shouldn't be mapped or
865 * have any special flags on it
866 */
867 if (WARN_ON(page_mapped(oldpage)))
868 goto out_fallback_unlock;
869 if (WARN_ON(page_has_private(oldpage)))
870 goto out_fallback_unlock;
871 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
872 goto out_fallback_unlock;
873 if (WARN_ON(PageMlocked(oldpage)))
874 goto out_fallback_unlock;
875
876 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
877 if (err) {
878 unlock_page(newpage);
879 return err;
880 }
881
882 get_page(newpage);
883
884 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
885 lru_cache_add_file(newpage);
886
887 err = 0;
888 spin_lock(&cs->req->waitq.lock);
889 if (test_bit(FR_ABORTED, &cs->req->flags))
890 err = -ENOENT;
891 else
892 *pagep = newpage;
893 spin_unlock(&cs->req->waitq.lock);
894
895 if (err) {
896 unlock_page(newpage);
897 put_page(newpage);
898 return err;
899 }
900
901 unlock_page(oldpage);
902 put_page(oldpage);
903 cs->len = 0;
904
905 return 0;
906
907 out_fallback_unlock:
908 unlock_page(newpage);
909 out_fallback:
910 cs->pg = buf->page;
911 cs->offset = buf->offset;
912
913 err = lock_request(cs->req);
914 if (err)
915 return err;
916
917 return 1;
918 }
919
920 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
921 unsigned offset, unsigned count)
922 {
923 struct pipe_buffer *buf;
924 int err;
925
926 if (cs->nr_segs == cs->pipe->buffers)
927 return -EIO;
928
929 err = unlock_request(cs->req);
930 if (err)
931 return err;
932
933 fuse_copy_finish(cs);
934
935 buf = cs->pipebufs;
936 get_page(page);
937 buf->page = page;
938 buf->offset = offset;
939 buf->len = count;
940
941 cs->pipebufs++;
942 cs->nr_segs++;
943 cs->len = 0;
944
945 return 0;
946 }
947
948 /*
949 * Copy a page in the request to/from the userspace buffer. Must be
950 * done atomically
951 */
952 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
953 unsigned offset, unsigned count, int zeroing)
954 {
955 int err;
956 struct page *page = *pagep;
957
958 if (page && zeroing && count < PAGE_SIZE)
959 clear_highpage(page);
960
961 while (count) {
962 if (cs->write && cs->pipebufs && page) {
963 return fuse_ref_page(cs, page, offset, count);
964 } else if (!cs->len) {
965 if (cs->move_pages && page &&
966 offset == 0 && count == PAGE_SIZE) {
967 err = fuse_try_move_page(cs, pagep);
968 if (err <= 0)
969 return err;
970 } else {
971 err = fuse_copy_fill(cs);
972 if (err)
973 return err;
974 }
975 }
976 if (page) {
977 void *mapaddr = kmap_atomic(page);
978 void *buf = mapaddr + offset;
979 offset += fuse_copy_do(cs, &buf, &count);
980 kunmap_atomic(mapaddr);
981 } else
982 offset += fuse_copy_do(cs, NULL, &count);
983 }
984 if (page && !cs->write)
985 flush_dcache_page(page);
986 return 0;
987 }
988
989 /* Copy pages in the request to/from userspace buffer */
990 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
991 int zeroing)
992 {
993 unsigned i;
994 struct fuse_req *req = cs->req;
995
996 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
997 int err;
998 unsigned offset = req->page_descs[i].offset;
999 unsigned count = min(nbytes, req->page_descs[i].length);
1000
1001 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1002 zeroing);
1003 if (err)
1004 return err;
1005
1006 nbytes -= count;
1007 }
1008 return 0;
1009 }
1010
1011 /* Copy a single argument in the request to/from userspace buffer */
1012 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1013 {
1014 while (size) {
1015 if (!cs->len) {
1016 int err = fuse_copy_fill(cs);
1017 if (err)
1018 return err;
1019 }
1020 fuse_copy_do(cs, &val, &size);
1021 }
1022 return 0;
1023 }
1024
1025 /* Copy request arguments to/from userspace buffer */
1026 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1027 unsigned argpages, struct fuse_arg *args,
1028 int zeroing)
1029 {
1030 int err = 0;
1031 unsigned i;
1032
1033 for (i = 0; !err && i < numargs; i++) {
1034 struct fuse_arg *arg = &args[i];
1035 if (i == numargs - 1 && argpages)
1036 err = fuse_copy_pages(cs, arg->size, zeroing);
1037 else
1038 err = fuse_copy_one(cs, arg->value, arg->size);
1039 }
1040 return err;
1041 }
1042
1043 static int forget_pending(struct fuse_iqueue *fiq)
1044 {
1045 return fiq->forget_list_head.next != NULL;
1046 }
1047
1048 static int request_pending(struct fuse_iqueue *fiq)
1049 {
1050 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1051 forget_pending(fiq);
1052 }
1053
1054 /*
1055 * Transfer an interrupt request to userspace
1056 *
1057 * Unlike other requests this is assembled on demand, without a need
1058 * to allocate a separate fuse_req structure.
1059 *
1060 * Called with fiq->waitq.lock held, releases it
1061 */
1062 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1063 struct fuse_copy_state *cs,
1064 size_t nbytes, struct fuse_req *req)
1065 __releases(fiq->waitq.lock)
1066 {
1067 struct fuse_in_header ih;
1068 struct fuse_interrupt_in arg;
1069 unsigned reqsize = sizeof(ih) + sizeof(arg);
1070 int err;
1071
1072 list_del_init(&req->intr_entry);
1073 req->intr_unique = fuse_get_unique(fiq);
1074 memset(&ih, 0, sizeof(ih));
1075 memset(&arg, 0, sizeof(arg));
1076 ih.len = reqsize;
1077 ih.opcode = FUSE_INTERRUPT;
1078 ih.unique = req->intr_unique;
1079 arg.unique = req->in.h.unique;
1080
1081 spin_unlock(&fiq->waitq.lock);
1082 if (nbytes < reqsize)
1083 return -EINVAL;
1084
1085 err = fuse_copy_one(cs, &ih, sizeof(ih));
1086 if (!err)
1087 err = fuse_copy_one(cs, &arg, sizeof(arg));
1088 fuse_copy_finish(cs);
1089
1090 return err ? err : reqsize;
1091 }
1092
1093 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1094 unsigned max,
1095 unsigned *countp)
1096 {
1097 struct fuse_forget_link *head = fiq->forget_list_head.next;
1098 struct fuse_forget_link **newhead = &head;
1099 unsigned count;
1100
1101 for (count = 0; *newhead != NULL && count < max; count++)
1102 newhead = &(*newhead)->next;
1103
1104 fiq->forget_list_head.next = *newhead;
1105 *newhead = NULL;
1106 if (fiq->forget_list_head.next == NULL)
1107 fiq->forget_list_tail = &fiq->forget_list_head;
1108
1109 if (countp != NULL)
1110 *countp = count;
1111
1112 return head;
1113 }
1114
1115 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1116 struct fuse_copy_state *cs,
1117 size_t nbytes)
1118 __releases(fiq->waitq.lock)
1119 {
1120 int err;
1121 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1122 struct fuse_forget_in arg = {
1123 .nlookup = forget->forget_one.nlookup,
1124 };
1125 struct fuse_in_header ih = {
1126 .opcode = FUSE_FORGET,
1127 .nodeid = forget->forget_one.nodeid,
1128 .unique = fuse_get_unique(fiq),
1129 .len = sizeof(ih) + sizeof(arg),
1130 };
1131
1132 spin_unlock(&fiq->waitq.lock);
1133 kfree(forget);
1134 if (nbytes < ih.len)
1135 return -EINVAL;
1136
1137 err = fuse_copy_one(cs, &ih, sizeof(ih));
1138 if (!err)
1139 err = fuse_copy_one(cs, &arg, sizeof(arg));
1140 fuse_copy_finish(cs);
1141
1142 if (err)
1143 return err;
1144
1145 return ih.len;
1146 }
1147
1148 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1149 struct fuse_copy_state *cs, size_t nbytes)
1150 __releases(fiq->waitq.lock)
1151 {
1152 int err;
1153 unsigned max_forgets;
1154 unsigned count;
1155 struct fuse_forget_link *head;
1156 struct fuse_batch_forget_in arg = { .count = 0 };
1157 struct fuse_in_header ih = {
1158 .opcode = FUSE_BATCH_FORGET,
1159 .unique = fuse_get_unique(fiq),
1160 .len = sizeof(ih) + sizeof(arg),
1161 };
1162
1163 if (nbytes < ih.len) {
1164 spin_unlock(&fiq->waitq.lock);
1165 return -EINVAL;
1166 }
1167
1168 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1169 head = dequeue_forget(fiq, max_forgets, &count);
1170 spin_unlock(&fiq->waitq.lock);
1171
1172 arg.count = count;
1173 ih.len += count * sizeof(struct fuse_forget_one);
1174 err = fuse_copy_one(cs, &ih, sizeof(ih));
1175 if (!err)
1176 err = fuse_copy_one(cs, &arg, sizeof(arg));
1177
1178 while (head) {
1179 struct fuse_forget_link *forget = head;
1180
1181 if (!err) {
1182 err = fuse_copy_one(cs, &forget->forget_one,
1183 sizeof(forget->forget_one));
1184 }
1185 head = forget->next;
1186 kfree(forget);
1187 }
1188
1189 fuse_copy_finish(cs);
1190
1191 if (err)
1192 return err;
1193
1194 return ih.len;
1195 }
1196
1197 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1198 struct fuse_copy_state *cs,
1199 size_t nbytes)
1200 __releases(fiq->waitq.lock)
1201 {
1202 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1203 return fuse_read_single_forget(fiq, cs, nbytes);
1204 else
1205 return fuse_read_batch_forget(fiq, cs, nbytes);
1206 }
1207
1208 /*
1209 * Read a single request into the userspace filesystem's buffer. This
1210 * function waits until a request is available, then removes it from
1211 * the pending list and copies request data to userspace buffer. If
1212 * no reply is needed (FORGET) or request has been aborted or there
1213 * was an error during the copying then it's finished by calling
1214 * request_end(). Otherwise add it to the processing list, and set
1215 * the 'sent' flag.
1216 */
1217 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1218 struct fuse_copy_state *cs, size_t nbytes)
1219 {
1220 ssize_t err;
1221 struct fuse_conn *fc = fud->fc;
1222 struct fuse_iqueue *fiq = &fc->iq;
1223 struct fuse_pqueue *fpq = &fud->pq;
1224 struct fuse_req *req;
1225 struct fuse_in *in;
1226 unsigned reqsize;
1227
1228 restart:
1229 spin_lock(&fiq->waitq.lock);
1230 err = -EAGAIN;
1231 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1232 !request_pending(fiq))
1233 goto err_unlock;
1234
1235 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1236 !fiq->connected || request_pending(fiq));
1237 if (err)
1238 goto err_unlock;
1239
1240 err = -ENODEV;
1241 if (!fiq->connected)
1242 goto err_unlock;
1243
1244 if (!list_empty(&fiq->interrupts)) {
1245 req = list_entry(fiq->interrupts.next, struct fuse_req,
1246 intr_entry);
1247 return fuse_read_interrupt(fiq, cs, nbytes, req);
1248 }
1249
1250 if (forget_pending(fiq)) {
1251 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1252 return fuse_read_forget(fc, fiq, cs, nbytes);
1253
1254 if (fiq->forget_batch <= -8)
1255 fiq->forget_batch = 16;
1256 }
1257
1258 req = list_entry(fiq->pending.next, struct fuse_req, list);
1259 clear_bit(FR_PENDING, &req->flags);
1260 list_del_init(&req->list);
1261 spin_unlock(&fiq->waitq.lock);
1262
1263 in = &req->in;
1264 reqsize = in->h.len;
1265
1266 if (task_active_pid_ns(current) != fc->pid_ns) {
1267 rcu_read_lock();
1268 in->h.pid = pid_vnr(find_pid_ns(in->h.pid, fc->pid_ns));
1269 rcu_read_unlock();
1270 }
1271
1272 /* If request is too large, reply with an error and restart the read */
1273 if (nbytes < reqsize) {
1274 req->out.h.error = -EIO;
1275 /* SETXATTR is special, since it may contain too large data */
1276 if (in->h.opcode == FUSE_SETXATTR)
1277 req->out.h.error = -E2BIG;
1278 request_end(fc, req);
1279 goto restart;
1280 }
1281 spin_lock(&fpq->lock);
1282 list_add(&req->list, &fpq->io);
1283 spin_unlock(&fpq->lock);
1284 cs->req = req;
1285 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1286 if (!err)
1287 err = fuse_copy_args(cs, in->numargs, in->argpages,
1288 (struct fuse_arg *) in->args, 0);
1289 fuse_copy_finish(cs);
1290 spin_lock(&fpq->lock);
1291 clear_bit(FR_LOCKED, &req->flags);
1292 if (!fpq->connected) {
1293 err = -ENODEV;
1294 goto out_end;
1295 }
1296 if (err) {
1297 req->out.h.error = -EIO;
1298 goto out_end;
1299 }
1300 if (!test_bit(FR_ISREPLY, &req->flags)) {
1301 err = reqsize;
1302 goto out_end;
1303 }
1304 list_move_tail(&req->list, &fpq->processing);
1305 spin_unlock(&fpq->lock);
1306 set_bit(FR_SENT, &req->flags);
1307 /* matches barrier in request_wait_answer() */
1308 smp_mb__after_atomic();
1309 if (test_bit(FR_INTERRUPTED, &req->flags))
1310 queue_interrupt(fiq, req);
1311
1312 return reqsize;
1313
1314 out_end:
1315 if (!test_bit(FR_PRIVATE, &req->flags))
1316 list_del_init(&req->list);
1317 spin_unlock(&fpq->lock);
1318 request_end(fc, req);
1319 return err;
1320
1321 err_unlock:
1322 spin_unlock(&fiq->waitq.lock);
1323 return err;
1324 }
1325
1326 static int fuse_dev_open(struct inode *inode, struct file *file)
1327 {
1328 /*
1329 * The fuse device's file's private_data is used to hold
1330 * the fuse_conn(ection) when it is mounted, and is used to
1331 * keep track of whether the file has been mounted already.
1332 */
1333 file->private_data = NULL;
1334 return 0;
1335 }
1336
1337 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1338 {
1339 struct fuse_copy_state cs;
1340 struct file *file = iocb->ki_filp;
1341 struct fuse_dev *fud = fuse_get_dev(file);
1342
1343 if (!fud)
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(fud, 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 total, ret;
1359 int page_nr = 0;
1360 struct pipe_buffer *bufs;
1361 struct fuse_copy_state cs;
1362 struct fuse_dev *fud = fuse_get_dev(in);
1363
1364 if (!fud)
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(fud, in, &cs, len);
1375 if (ret < 0)
1376 goto out;
1377
1378 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1379 ret = -EIO;
1380 goto out;
1381 }
1382
1383 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1384 /*
1385 * Need to be careful about this. Having buf->ops in module
1386 * code can Oops if the buffer persists after module unload.
1387 */
1388 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1389 bufs[page_nr].flags = 0;
1390 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1391 if (unlikely(ret < 0))
1392 break;
1393 }
1394 if (total)
1395 ret = total;
1396 out:
1397 for (; page_nr < cs.nr_segs; page_nr++)
1398 put_page(bufs[page_nr].page);
1399
1400 kfree(bufs);
1401 return ret;
1402 }
1403
1404 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1405 struct fuse_copy_state *cs)
1406 {
1407 struct fuse_notify_poll_wakeup_out outarg;
1408 int err = -EINVAL;
1409
1410 if (size != sizeof(outarg))
1411 goto err;
1412
1413 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1414 if (err)
1415 goto err;
1416
1417 fuse_copy_finish(cs);
1418 return fuse_notify_poll_wakeup(fc, &outarg);
1419
1420 err:
1421 fuse_copy_finish(cs);
1422 return err;
1423 }
1424
1425 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1426 struct fuse_copy_state *cs)
1427 {
1428 struct fuse_notify_inval_inode_out outarg;
1429 int err = -EINVAL;
1430
1431 if (size != sizeof(outarg))
1432 goto err;
1433
1434 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1435 if (err)
1436 goto err;
1437 fuse_copy_finish(cs);
1438
1439 down_read(&fc->killsb);
1440 err = -ENOENT;
1441 if (fc->sb) {
1442 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1443 outarg.off, outarg.len);
1444 }
1445 up_read(&fc->killsb);
1446 return err;
1447
1448 err:
1449 fuse_copy_finish(cs);
1450 return err;
1451 }
1452
1453 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1454 struct fuse_copy_state *cs)
1455 {
1456 struct fuse_notify_inval_entry_out outarg;
1457 int err = -ENOMEM;
1458 char *buf;
1459 struct qstr name;
1460
1461 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1462 if (!buf)
1463 goto err;
1464
1465 err = -EINVAL;
1466 if (size < sizeof(outarg))
1467 goto err;
1468
1469 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1470 if (err)
1471 goto err;
1472
1473 err = -ENAMETOOLONG;
1474 if (outarg.namelen > FUSE_NAME_MAX)
1475 goto err;
1476
1477 err = -EINVAL;
1478 if (size != sizeof(outarg) + outarg.namelen + 1)
1479 goto err;
1480
1481 name.name = buf;
1482 name.len = outarg.namelen;
1483 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1484 if (err)
1485 goto err;
1486 fuse_copy_finish(cs);
1487 buf[outarg.namelen] = 0;
1488
1489 down_read(&fc->killsb);
1490 err = -ENOENT;
1491 if (fc->sb)
1492 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1493 up_read(&fc->killsb);
1494 kfree(buf);
1495 return err;
1496
1497 err:
1498 kfree(buf);
1499 fuse_copy_finish(cs);
1500 return err;
1501 }
1502
1503 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1504 struct fuse_copy_state *cs)
1505 {
1506 struct fuse_notify_delete_out outarg;
1507 int err = -ENOMEM;
1508 char *buf;
1509 struct qstr name;
1510
1511 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1512 if (!buf)
1513 goto err;
1514
1515 err = -EINVAL;
1516 if (size < sizeof(outarg))
1517 goto err;
1518
1519 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1520 if (err)
1521 goto err;
1522
1523 err = -ENAMETOOLONG;
1524 if (outarg.namelen > FUSE_NAME_MAX)
1525 goto err;
1526
1527 err = -EINVAL;
1528 if (size != sizeof(outarg) + outarg.namelen + 1)
1529 goto err;
1530
1531 name.name = buf;
1532 name.len = outarg.namelen;
1533 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1534 if (err)
1535 goto err;
1536 fuse_copy_finish(cs);
1537 buf[outarg.namelen] = 0;
1538
1539 down_read(&fc->killsb);
1540 err = -ENOENT;
1541 if (fc->sb)
1542 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1543 outarg.child, &name);
1544 up_read(&fc->killsb);
1545 kfree(buf);
1546 return err;
1547
1548 err:
1549 kfree(buf);
1550 fuse_copy_finish(cs);
1551 return err;
1552 }
1553
1554 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1555 struct fuse_copy_state *cs)
1556 {
1557 struct fuse_notify_store_out outarg;
1558 struct inode *inode;
1559 struct address_space *mapping;
1560 u64 nodeid;
1561 int err;
1562 pgoff_t index;
1563 unsigned int offset;
1564 unsigned int num;
1565 loff_t file_size;
1566 loff_t end;
1567
1568 err = -EINVAL;
1569 if (size < sizeof(outarg))
1570 goto out_finish;
1571
1572 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1573 if (err)
1574 goto out_finish;
1575
1576 err = -EINVAL;
1577 if (size - sizeof(outarg) != outarg.size)
1578 goto out_finish;
1579
1580 nodeid = outarg.nodeid;
1581
1582 down_read(&fc->killsb);
1583
1584 err = -ENOENT;
1585 if (!fc->sb)
1586 goto out_up_killsb;
1587
1588 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1589 if (!inode)
1590 goto out_up_killsb;
1591
1592 mapping = inode->i_mapping;
1593 index = outarg.offset >> PAGE_SHIFT;
1594 offset = outarg.offset & ~PAGE_MASK;
1595 file_size = i_size_read(inode);
1596 end = outarg.offset + outarg.size;
1597 if (end > file_size) {
1598 file_size = end;
1599 fuse_write_update_size(inode, file_size);
1600 }
1601
1602 num = outarg.size;
1603 while (num) {
1604 struct page *page;
1605 unsigned int this_num;
1606
1607 err = -ENOMEM;
1608 page = find_or_create_page(mapping, index,
1609 mapping_gfp_mask(mapping));
1610 if (!page)
1611 goto out_iput;
1612
1613 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1614 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1615 if (!err && offset == 0 &&
1616 (this_num == PAGE_SIZE || file_size == end))
1617 SetPageUptodate(page);
1618 unlock_page(page);
1619 put_page(page);
1620
1621 if (err)
1622 goto out_iput;
1623
1624 num -= this_num;
1625 offset = 0;
1626 index++;
1627 }
1628
1629 err = 0;
1630
1631 out_iput:
1632 iput(inode);
1633 out_up_killsb:
1634 up_read(&fc->killsb);
1635 out_finish:
1636 fuse_copy_finish(cs);
1637 return err;
1638 }
1639
1640 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1641 {
1642 release_pages(req->pages, req->num_pages, false);
1643 }
1644
1645 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1646 struct fuse_notify_retrieve_out *outarg)
1647 {
1648 int err;
1649 struct address_space *mapping = inode->i_mapping;
1650 struct fuse_req *req;
1651 pgoff_t index;
1652 loff_t file_size;
1653 unsigned int num;
1654 unsigned int offset;
1655 size_t total_len = 0;
1656 int num_pages;
1657
1658 offset = outarg->offset & ~PAGE_MASK;
1659 file_size = i_size_read(inode);
1660
1661 num = outarg->size;
1662 if (outarg->offset > file_size)
1663 num = 0;
1664 else if (outarg->offset + num > file_size)
1665 num = file_size - outarg->offset;
1666
1667 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1668 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1669
1670 req = fuse_get_req(fc, num_pages);
1671 if (IS_ERR(req))
1672 return PTR_ERR(req);
1673
1674 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1675 req->in.h.nodeid = outarg->nodeid;
1676 req->in.numargs = 2;
1677 req->in.argpages = 1;
1678 req->page_descs[0].offset = offset;
1679 req->end = fuse_retrieve_end;
1680
1681 index = outarg->offset >> PAGE_SHIFT;
1682
1683 while (num && req->num_pages < num_pages) {
1684 struct page *page;
1685 unsigned int this_num;
1686
1687 page = find_get_page(mapping, index);
1688 if (!page)
1689 break;
1690
1691 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1692 req->pages[req->num_pages] = page;
1693 req->page_descs[req->num_pages].length = this_num;
1694 req->num_pages++;
1695
1696 offset = 0;
1697 num -= this_num;
1698 total_len += this_num;
1699 index++;
1700 }
1701 req->misc.retrieve_in.offset = outarg->offset;
1702 req->misc.retrieve_in.size = total_len;
1703 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1704 req->in.args[0].value = &req->misc.retrieve_in;
1705 req->in.args[1].size = total_len;
1706
1707 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1708 if (err)
1709 fuse_retrieve_end(fc, req);
1710
1711 return err;
1712 }
1713
1714 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1715 struct fuse_copy_state *cs)
1716 {
1717 struct fuse_notify_retrieve_out outarg;
1718 struct inode *inode;
1719 int err;
1720
1721 err = -EINVAL;
1722 if (size != sizeof(outarg))
1723 goto copy_finish;
1724
1725 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1726 if (err)
1727 goto copy_finish;
1728
1729 fuse_copy_finish(cs);
1730
1731 down_read(&fc->killsb);
1732 err = -ENOENT;
1733 if (fc->sb) {
1734 u64 nodeid = outarg.nodeid;
1735
1736 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1737 if (inode) {
1738 err = fuse_retrieve(fc, inode, &outarg);
1739 iput(inode);
1740 }
1741 }
1742 up_read(&fc->killsb);
1743
1744 return err;
1745
1746 copy_finish:
1747 fuse_copy_finish(cs);
1748 return err;
1749 }
1750
1751 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1752 unsigned int size, struct fuse_copy_state *cs)
1753 {
1754 /* Don't try to move pages (yet) */
1755 cs->move_pages = 0;
1756
1757 switch (code) {
1758 case FUSE_NOTIFY_POLL:
1759 return fuse_notify_poll(fc, size, cs);
1760
1761 case FUSE_NOTIFY_INVAL_INODE:
1762 return fuse_notify_inval_inode(fc, size, cs);
1763
1764 case FUSE_NOTIFY_INVAL_ENTRY:
1765 return fuse_notify_inval_entry(fc, size, cs);
1766
1767 case FUSE_NOTIFY_STORE:
1768 return fuse_notify_store(fc, size, cs);
1769
1770 case FUSE_NOTIFY_RETRIEVE:
1771 return fuse_notify_retrieve(fc, size, cs);
1772
1773 case FUSE_NOTIFY_DELETE:
1774 return fuse_notify_delete(fc, size, cs);
1775
1776 default:
1777 fuse_copy_finish(cs);
1778 return -EINVAL;
1779 }
1780 }
1781
1782 /* Look up request on processing list by unique ID */
1783 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1784 {
1785 struct fuse_req *req;
1786
1787 list_for_each_entry(req, &fpq->processing, list) {
1788 if (req->in.h.unique == unique || req->intr_unique == unique)
1789 return req;
1790 }
1791 return NULL;
1792 }
1793
1794 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1795 unsigned nbytes)
1796 {
1797 unsigned reqsize = sizeof(struct fuse_out_header);
1798
1799 if (out->h.error)
1800 return nbytes != reqsize ? -EINVAL : 0;
1801
1802 reqsize += len_args(out->numargs, out->args);
1803
1804 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1805 return -EINVAL;
1806 else if (reqsize > nbytes) {
1807 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1808 unsigned diffsize = reqsize - nbytes;
1809 if (diffsize > lastarg->size)
1810 return -EINVAL;
1811 lastarg->size -= diffsize;
1812 }
1813 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1814 out->page_zeroing);
1815 }
1816
1817 /*
1818 * Write a single reply to a request. First the header is copied from
1819 * the write buffer. The request is then searched on the processing
1820 * list by the unique ID found in the header. If found, then remove
1821 * it from the list and copy the rest of the buffer to the request.
1822 * The request is finished by calling request_end()
1823 */
1824 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1825 struct fuse_copy_state *cs, size_t nbytes)
1826 {
1827 int err;
1828 struct fuse_conn *fc = fud->fc;
1829 struct fuse_pqueue *fpq = &fud->pq;
1830 struct fuse_req *req;
1831 struct fuse_out_header oh;
1832
1833 if (nbytes < sizeof(struct fuse_out_header))
1834 return -EINVAL;
1835
1836 err = fuse_copy_one(cs, &oh, sizeof(oh));
1837 if (err)
1838 goto err_finish;
1839
1840 err = -EINVAL;
1841 if (oh.len != nbytes)
1842 goto err_finish;
1843
1844 /*
1845 * Zero oh.unique indicates unsolicited notification message
1846 * and error contains notification code.
1847 */
1848 if (!oh.unique) {
1849 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1850 return err ? err : nbytes;
1851 }
1852
1853 err = -EINVAL;
1854 if (oh.error <= -1000 || oh.error > 0)
1855 goto err_finish;
1856
1857 spin_lock(&fpq->lock);
1858 err = -ENOENT;
1859 if (!fpq->connected)
1860 goto err_unlock_pq;
1861
1862 req = request_find(fpq, oh.unique);
1863 if (!req)
1864 goto err_unlock_pq;
1865
1866 /* Is it an interrupt reply? */
1867 if (req->intr_unique == oh.unique) {
1868 spin_unlock(&fpq->lock);
1869
1870 err = -EINVAL;
1871 if (nbytes != sizeof(struct fuse_out_header))
1872 goto err_finish;
1873
1874 if (oh.error == -ENOSYS)
1875 fc->no_interrupt = 1;
1876 else if (oh.error == -EAGAIN)
1877 queue_interrupt(&fc->iq, req);
1878
1879 fuse_copy_finish(cs);
1880 return nbytes;
1881 }
1882
1883 clear_bit(FR_SENT, &req->flags);
1884 list_move(&req->list, &fpq->io);
1885 req->out.h = oh;
1886 set_bit(FR_LOCKED, &req->flags);
1887 spin_unlock(&fpq->lock);
1888 cs->req = req;
1889 if (!req->out.page_replace)
1890 cs->move_pages = 0;
1891
1892 err = copy_out_args(cs, &req->out, nbytes);
1893 fuse_copy_finish(cs);
1894
1895 spin_lock(&fpq->lock);
1896 clear_bit(FR_LOCKED, &req->flags);
1897 if (!fpq->connected)
1898 err = -ENOENT;
1899 else if (err)
1900 req->out.h.error = -EIO;
1901 if (!test_bit(FR_PRIVATE, &req->flags))
1902 list_del_init(&req->list);
1903 spin_unlock(&fpq->lock);
1904
1905 request_end(fc, req);
1906
1907 return err ? err : nbytes;
1908
1909 err_unlock_pq:
1910 spin_unlock(&fpq->lock);
1911 err_finish:
1912 fuse_copy_finish(cs);
1913 return err;
1914 }
1915
1916 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1917 {
1918 struct fuse_copy_state cs;
1919 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1920
1921 if (!fud)
1922 return -EPERM;
1923
1924 if (!iter_is_iovec(from))
1925 return -EINVAL;
1926
1927 fuse_copy_init(&cs, 0, from);
1928
1929 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1930 }
1931
1932 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1933 struct file *out, loff_t *ppos,
1934 size_t len, unsigned int flags)
1935 {
1936 unsigned nbuf;
1937 unsigned idx;
1938 struct pipe_buffer *bufs;
1939 struct fuse_copy_state cs;
1940 struct fuse_dev *fud;
1941 size_t rem;
1942 ssize_t ret;
1943
1944 fud = fuse_get_dev(out);
1945 if (!fud)
1946 return -EPERM;
1947
1948 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1949 if (!bufs)
1950 return -ENOMEM;
1951
1952 pipe_lock(pipe);
1953 nbuf = 0;
1954 rem = 0;
1955 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1956 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1957
1958 ret = -EINVAL;
1959 if (rem < len) {
1960 pipe_unlock(pipe);
1961 goto out;
1962 }
1963
1964 rem = len;
1965 while (rem) {
1966 struct pipe_buffer *ibuf;
1967 struct pipe_buffer *obuf;
1968
1969 BUG_ON(nbuf >= pipe->buffers);
1970 BUG_ON(!pipe->nrbufs);
1971 ibuf = &pipe->bufs[pipe->curbuf];
1972 obuf = &bufs[nbuf];
1973
1974 if (rem >= ibuf->len) {
1975 *obuf = *ibuf;
1976 ibuf->ops = NULL;
1977 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1978 pipe->nrbufs--;
1979 } else {
1980 pipe_buf_get(pipe, ibuf);
1981 *obuf = *ibuf;
1982 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1983 obuf->len = rem;
1984 ibuf->offset += obuf->len;
1985 ibuf->len -= obuf->len;
1986 }
1987 nbuf++;
1988 rem -= obuf->len;
1989 }
1990 pipe_unlock(pipe);
1991
1992 fuse_copy_init(&cs, 0, NULL);
1993 cs.pipebufs = bufs;
1994 cs.nr_segs = nbuf;
1995 cs.pipe = pipe;
1996
1997 if (flags & SPLICE_F_MOVE)
1998 cs.move_pages = 1;
1999
2000 ret = fuse_dev_do_write(fud, &cs, len);
2001
2002 for (idx = 0; idx < nbuf; idx++)
2003 pipe_buf_release(pipe, &bufs[idx]);
2004
2005 out:
2006 kfree(bufs);
2007 return ret;
2008 }
2009
2010 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2011 {
2012 unsigned mask = POLLOUT | POLLWRNORM;
2013 struct fuse_iqueue *fiq;
2014 struct fuse_dev *fud = fuse_get_dev(file);
2015
2016 if (!fud)
2017 return POLLERR;
2018
2019 fiq = &fud->fc->iq;
2020 poll_wait(file, &fiq->waitq, wait);
2021
2022 spin_lock(&fiq->waitq.lock);
2023 if (!fiq->connected)
2024 mask = POLLERR;
2025 else if (request_pending(fiq))
2026 mask |= POLLIN | POLLRDNORM;
2027 spin_unlock(&fiq->waitq.lock);
2028
2029 return mask;
2030 }
2031
2032 /*
2033 * Abort all requests on the given list (pending or processing)
2034 *
2035 * This function releases and reacquires fc->lock
2036 */
2037 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2038 {
2039 while (!list_empty(head)) {
2040 struct fuse_req *req;
2041 req = list_entry(head->next, struct fuse_req, list);
2042 req->out.h.error = -ECONNABORTED;
2043 clear_bit(FR_SENT, &req->flags);
2044 list_del_init(&req->list);
2045 request_end(fc, req);
2046 }
2047 }
2048
2049 static void end_polls(struct fuse_conn *fc)
2050 {
2051 struct rb_node *p;
2052
2053 p = rb_first(&fc->polled_files);
2054
2055 while (p) {
2056 struct fuse_file *ff;
2057 ff = rb_entry(p, struct fuse_file, polled_node);
2058 wake_up_interruptible_all(&ff->poll_wait);
2059
2060 p = rb_next(p);
2061 }
2062 }
2063
2064 /*
2065 * Abort all requests.
2066 *
2067 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2068 * filesystem.
2069 *
2070 * The same effect is usually achievable through killing the filesystem daemon
2071 * and all users of the filesystem. The exception is the combination of an
2072 * asynchronous request and the tricky deadlock (see
2073 * Documentation/filesystems/fuse.txt).
2074 *
2075 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2076 * requests, they should be finished off immediately. Locked requests will be
2077 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2078 * requests. It is possible that some request will finish before we can. This
2079 * is OK, the request will in that case be removed from the list before we touch
2080 * it.
2081 */
2082 void fuse_abort_conn(struct fuse_conn *fc)
2083 {
2084 struct fuse_iqueue *fiq = &fc->iq;
2085
2086 spin_lock(&fc->lock);
2087 if (fc->connected) {
2088 struct fuse_dev *fud;
2089 struct fuse_req *req, *next;
2090 LIST_HEAD(to_end1);
2091 LIST_HEAD(to_end2);
2092
2093 fc->connected = 0;
2094 fc->blocked = 0;
2095 fuse_set_initialized(fc);
2096 list_for_each_entry(fud, &fc->devices, entry) {
2097 struct fuse_pqueue *fpq = &fud->pq;
2098
2099 spin_lock(&fpq->lock);
2100 fpq->connected = 0;
2101 list_for_each_entry_safe(req, next, &fpq->io, list) {
2102 req->out.h.error = -ECONNABORTED;
2103 spin_lock(&req->waitq.lock);
2104 set_bit(FR_ABORTED, &req->flags);
2105 if (!test_bit(FR_LOCKED, &req->flags)) {
2106 set_bit(FR_PRIVATE, &req->flags);
2107 list_move(&req->list, &to_end1);
2108 }
2109 spin_unlock(&req->waitq.lock);
2110 }
2111 list_splice_init(&fpq->processing, &to_end2);
2112 spin_unlock(&fpq->lock);
2113 }
2114 fc->max_background = UINT_MAX;
2115 flush_bg_queue(fc);
2116
2117 spin_lock(&fiq->waitq.lock);
2118 fiq->connected = 0;
2119 list_splice_init(&fiq->pending, &to_end2);
2120 list_for_each_entry(req, &to_end2, list)
2121 clear_bit(FR_PENDING, &req->flags);
2122 while (forget_pending(fiq))
2123 kfree(dequeue_forget(fiq, 1, NULL));
2124 wake_up_all_locked(&fiq->waitq);
2125 spin_unlock(&fiq->waitq.lock);
2126 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2127 end_polls(fc);
2128 wake_up_all(&fc->blocked_waitq);
2129 spin_unlock(&fc->lock);
2130
2131 while (!list_empty(&to_end1)) {
2132 req = list_first_entry(&to_end1, struct fuse_req, list);
2133 __fuse_get_request(req);
2134 list_del_init(&req->list);
2135 request_end(fc, req);
2136 }
2137 end_requests(fc, &to_end2);
2138 } else {
2139 spin_unlock(&fc->lock);
2140 }
2141 }
2142 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2143
2144 int fuse_dev_release(struct inode *inode, struct file *file)
2145 {
2146 struct fuse_dev *fud = fuse_get_dev(file);
2147
2148 if (fud) {
2149 struct fuse_conn *fc = fud->fc;
2150 struct fuse_pqueue *fpq = &fud->pq;
2151
2152 WARN_ON(!list_empty(&fpq->io));
2153 end_requests(fc, &fpq->processing);
2154 /* Are we the last open device? */
2155 if (atomic_dec_and_test(&fc->dev_count)) {
2156 WARN_ON(fc->iq.fasync != NULL);
2157 fuse_abort_conn(fc);
2158 }
2159 fuse_dev_free(fud);
2160 }
2161 return 0;
2162 }
2163 EXPORT_SYMBOL_GPL(fuse_dev_release);
2164
2165 static int fuse_dev_fasync(int fd, struct file *file, int on)
2166 {
2167 struct fuse_dev *fud = fuse_get_dev(file);
2168
2169 if (!fud)
2170 return -EPERM;
2171
2172 /* No locking - fasync_helper does its own locking */
2173 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2174 }
2175
2176 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2177 {
2178 struct fuse_dev *fud;
2179
2180 if (new->private_data)
2181 return -EINVAL;
2182
2183 fud = fuse_dev_alloc(fc);
2184 if (!fud)
2185 return -ENOMEM;
2186
2187 new->private_data = fud;
2188 atomic_inc(&fc->dev_count);
2189
2190 return 0;
2191 }
2192
2193 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2194 unsigned long arg)
2195 {
2196 int err = -ENOTTY;
2197
2198 if (cmd == FUSE_DEV_IOC_CLONE) {
2199 int oldfd;
2200
2201 err = -EFAULT;
2202 if (!get_user(oldfd, (__u32 __user *) arg)) {
2203 struct file *old = fget(oldfd);
2204
2205 err = -EINVAL;
2206 if (old) {
2207 struct fuse_dev *fud = NULL;
2208
2209 /*
2210 * Check against file->f_op because CUSE
2211 * uses the same ioctl handler.
2212 */
2213 if (old->f_op == file->f_op &&
2214 old->f_cred->user_ns == file->f_cred->user_ns)
2215 fud = fuse_get_dev(old);
2216
2217 if (fud) {
2218 mutex_lock(&fuse_mutex);
2219 err = fuse_device_clone(fud->fc, file);
2220 mutex_unlock(&fuse_mutex);
2221 }
2222 fput(old);
2223 }
2224 }
2225 }
2226 return err;
2227 }
2228
2229 const struct file_operations fuse_dev_operations = {
2230 .owner = THIS_MODULE,
2231 .open = fuse_dev_open,
2232 .llseek = no_llseek,
2233 .read_iter = fuse_dev_read,
2234 .splice_read = fuse_dev_splice_read,
2235 .write_iter = fuse_dev_write,
2236 .splice_write = fuse_dev_splice_write,
2237 .poll = fuse_dev_poll,
2238 .release = fuse_dev_release,
2239 .fasync = fuse_dev_fasync,
2240 .unlocked_ioctl = fuse_dev_ioctl,
2241 .compat_ioctl = fuse_dev_ioctl,
2242 };
2243 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2244
2245 static struct miscdevice fuse_miscdevice = {
2246 .minor = FUSE_MINOR,
2247 .name = "fuse",
2248 .fops = &fuse_dev_operations,
2249 };
2250
2251 int __init fuse_dev_init(void)
2252 {
2253 int err = -ENOMEM;
2254 fuse_req_cachep = kmem_cache_create("fuse_request",
2255 sizeof(struct fuse_req),
2256 0, 0, NULL);
2257 if (!fuse_req_cachep)
2258 goto out;
2259
2260 err = misc_register(&fuse_miscdevice);
2261 if (err)
2262 goto out_cache_clean;
2263
2264 return 0;
2265
2266 out_cache_clean:
2267 kmem_cache_destroy(fuse_req_cachep);
2268 out:
2269 return err;
2270 }
2271
2272 void fuse_dev_cleanup(void)
2273 {
2274 misc_deregister(&fuse_miscdevice);
2275 kmem_cache_destroy(fuse_req_cachep);
2276 }