Loading...
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
25MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26MODULE_ALIAS("devname:fuse");
27
28/* Ordinary requests have even IDs, while interrupts IDs are odd */
29#define FUSE_INT_REQ_BIT (1ULL << 0)
30#define FUSE_REQ_ID_STEP (1ULL << 1)
31
32static struct kmem_cache *fuse_req_cachep;
33
34static struct fuse_dev *fuse_get_dev(struct file *file)
35{
36 /*
37 * Lockless access is OK, because file->private data is set
38 * once during mount and is valid until the file is released.
39 */
40 return READ_ONCE(file->private_data);
41}
42
43static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)
44{
45 INIT_LIST_HEAD(&req->list);
46 INIT_LIST_HEAD(&req->intr_entry);
47 init_waitqueue_head(&req->waitq);
48 refcount_set(&req->count, 1);
49 __set_bit(FR_PENDING, &req->flags);
50 req->fm = fm;
51}
52
53static struct fuse_req *fuse_request_alloc(struct fuse_mount *fm, gfp_t flags)
54{
55 struct fuse_req *req = kmem_cache_zalloc(fuse_req_cachep, flags);
56 if (req)
57 fuse_request_init(fm, req);
58
59 return req;
60}
61
62static void fuse_request_free(struct fuse_req *req)
63{
64 kmem_cache_free(fuse_req_cachep, req);
65}
66
67static void __fuse_get_request(struct fuse_req *req)
68{
69 refcount_inc(&req->count);
70}
71
72/* Must be called with > 1 refcount */
73static void __fuse_put_request(struct fuse_req *req)
74{
75 refcount_dec(&req->count);
76}
77
78void fuse_set_initialized(struct fuse_conn *fc)
79{
80 /* Make sure stores before this are seen on another CPU */
81 smp_wmb();
82 fc->initialized = 1;
83}
84
85static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
86{
87 return !fc->initialized || (for_background && fc->blocked);
88}
89
90static void fuse_drop_waiting(struct fuse_conn *fc)
91{
92 /*
93 * lockess check of fc->connected is okay, because atomic_dec_and_test()
94 * provides a memory barrier matched with the one in fuse_wait_aborted()
95 * to ensure no wake-up is missed.
96 */
97 if (atomic_dec_and_test(&fc->num_waiting) &&
98 !READ_ONCE(fc->connected)) {
99 /* wake up aborters */
100 wake_up_all(&fc->blocked_waitq);
101 }
102}
103
104static void fuse_put_request(struct fuse_req *req);
105
106static struct fuse_req *fuse_get_req(struct fuse_mount *fm, bool for_background)
107{
108 struct fuse_conn *fc = fm->fc;
109 struct fuse_req *req;
110 int err;
111 atomic_inc(&fc->num_waiting);
112
113 if (fuse_block_alloc(fc, for_background)) {
114 err = -EINTR;
115 if (wait_event_killable_exclusive(fc->blocked_waitq,
116 !fuse_block_alloc(fc, for_background)))
117 goto out;
118 }
119 /* Matches smp_wmb() in fuse_set_initialized() */
120 smp_rmb();
121
122 err = -ENOTCONN;
123 if (!fc->connected)
124 goto out;
125
126 err = -ECONNREFUSED;
127 if (fc->conn_error)
128 goto out;
129
130 req = fuse_request_alloc(fm, GFP_KERNEL);
131 err = -ENOMEM;
132 if (!req) {
133 if (for_background)
134 wake_up(&fc->blocked_waitq);
135 goto out;
136 }
137
138 req->in.h.uid = from_kuid(fc->user_ns, current_fsuid());
139 req->in.h.gid = from_kgid(fc->user_ns, current_fsgid());
140 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
141
142 __set_bit(FR_WAITING, &req->flags);
143 if (for_background)
144 __set_bit(FR_BACKGROUND, &req->flags);
145
146 if (unlikely(req->in.h.uid == ((uid_t)-1) ||
147 req->in.h.gid == ((gid_t)-1))) {
148 fuse_put_request(req);
149 return ERR_PTR(-EOVERFLOW);
150 }
151 return req;
152
153 out:
154 fuse_drop_waiting(fc);
155 return ERR_PTR(err);
156}
157
158static void fuse_put_request(struct fuse_req *req)
159{
160 struct fuse_conn *fc = req->fm->fc;
161
162 if (refcount_dec_and_test(&req->count)) {
163 if (test_bit(FR_BACKGROUND, &req->flags)) {
164 /*
165 * We get here in the unlikely case that a background
166 * request was allocated but not sent
167 */
168 spin_lock(&fc->bg_lock);
169 if (!fc->blocked)
170 wake_up(&fc->blocked_waitq);
171 spin_unlock(&fc->bg_lock);
172 }
173
174 if (test_bit(FR_WAITING, &req->flags)) {
175 __clear_bit(FR_WAITING, &req->flags);
176 fuse_drop_waiting(fc);
177 }
178
179 fuse_request_free(req);
180 }
181}
182
183unsigned int fuse_len_args(unsigned int numargs, struct fuse_arg *args)
184{
185 unsigned nbytes = 0;
186 unsigned i;
187
188 for (i = 0; i < numargs; i++)
189 nbytes += args[i].size;
190
191 return nbytes;
192}
193EXPORT_SYMBOL_GPL(fuse_len_args);
194
195u64 fuse_get_unique(struct fuse_iqueue *fiq)
196{
197 fiq->reqctr += FUSE_REQ_ID_STEP;
198 return fiq->reqctr;
199}
200EXPORT_SYMBOL_GPL(fuse_get_unique);
201
202static unsigned int fuse_req_hash(u64 unique)
203{
204 return hash_long(unique & ~FUSE_INT_REQ_BIT, FUSE_PQ_HASH_BITS);
205}
206
207/**
208 * A new request is available, wake fiq->waitq
209 */
210static void fuse_dev_wake_and_unlock(struct fuse_iqueue *fiq)
211__releases(fiq->lock)
212{
213 wake_up(&fiq->waitq);
214 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
215 spin_unlock(&fiq->lock);
216}
217
218const struct fuse_iqueue_ops fuse_dev_fiq_ops = {
219 .wake_forget_and_unlock = fuse_dev_wake_and_unlock,
220 .wake_interrupt_and_unlock = fuse_dev_wake_and_unlock,
221 .wake_pending_and_unlock = fuse_dev_wake_and_unlock,
222};
223EXPORT_SYMBOL_GPL(fuse_dev_fiq_ops);
224
225static void queue_request_and_unlock(struct fuse_iqueue *fiq,
226 struct fuse_req *req)
227__releases(fiq->lock)
228{
229 req->in.h.len = sizeof(struct fuse_in_header) +
230 fuse_len_args(req->args->in_numargs,
231 (struct fuse_arg *) req->args->in_args);
232 list_add_tail(&req->list, &fiq->pending);
233 fiq->ops->wake_pending_and_unlock(fiq);
234}
235
236void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
237 u64 nodeid, u64 nlookup)
238{
239 struct fuse_iqueue *fiq = &fc->iq;
240
241 forget->forget_one.nodeid = nodeid;
242 forget->forget_one.nlookup = nlookup;
243
244 spin_lock(&fiq->lock);
245 if (fiq->connected) {
246 fiq->forget_list_tail->next = forget;
247 fiq->forget_list_tail = forget;
248 fiq->ops->wake_forget_and_unlock(fiq);
249 } else {
250 kfree(forget);
251 spin_unlock(&fiq->lock);
252 }
253}
254
255static void flush_bg_queue(struct fuse_conn *fc)
256{
257 struct fuse_iqueue *fiq = &fc->iq;
258
259 while (fc->active_background < fc->max_background &&
260 !list_empty(&fc->bg_queue)) {
261 struct fuse_req *req;
262
263 req = list_first_entry(&fc->bg_queue, struct fuse_req, list);
264 list_del(&req->list);
265 fc->active_background++;
266 spin_lock(&fiq->lock);
267 req->in.h.unique = fuse_get_unique(fiq);
268 queue_request_and_unlock(fiq, req);
269 }
270}
271
272/*
273 * This function is called when a request is finished. Either a reply
274 * has arrived or it was aborted (and not yet sent) or some error
275 * occurred during communication with userspace, or the device file
276 * was closed. The requester thread is woken up (if still waiting),
277 * the 'end' callback is called if given, else the reference to the
278 * request is released
279 */
280void fuse_request_end(struct fuse_req *req)
281{
282 struct fuse_mount *fm = req->fm;
283 struct fuse_conn *fc = fm->fc;
284 struct fuse_iqueue *fiq = &fc->iq;
285
286 if (test_and_set_bit(FR_FINISHED, &req->flags))
287 goto put_request;
288
289 /*
290 * test_and_set_bit() implies smp_mb() between bit
291 * changing and below FR_INTERRUPTED check. Pairs with
292 * smp_mb() from queue_interrupt().
293 */
294 if (test_bit(FR_INTERRUPTED, &req->flags)) {
295 spin_lock(&fiq->lock);
296 list_del_init(&req->intr_entry);
297 spin_unlock(&fiq->lock);
298 }
299 WARN_ON(test_bit(FR_PENDING, &req->flags));
300 WARN_ON(test_bit(FR_SENT, &req->flags));
301 if (test_bit(FR_BACKGROUND, &req->flags)) {
302 spin_lock(&fc->bg_lock);
303 clear_bit(FR_BACKGROUND, &req->flags);
304 if (fc->num_background == fc->max_background) {
305 fc->blocked = 0;
306 wake_up(&fc->blocked_waitq);
307 } else if (!fc->blocked) {
308 /*
309 * Wake up next waiter, if any. It's okay to use
310 * waitqueue_active(), as we've already synced up
311 * fc->blocked with waiters with the wake_up() call
312 * above.
313 */
314 if (waitqueue_active(&fc->blocked_waitq))
315 wake_up(&fc->blocked_waitq);
316 }
317
318 fc->num_background--;
319 fc->active_background--;
320 flush_bg_queue(fc);
321 spin_unlock(&fc->bg_lock);
322 } else {
323 /* Wake up waiter sleeping in request_wait_answer() */
324 wake_up(&req->waitq);
325 }
326
327 if (test_bit(FR_ASYNC, &req->flags))
328 req->args->end(fm, req->args, req->out.h.error);
329put_request:
330 fuse_put_request(req);
331}
332EXPORT_SYMBOL_GPL(fuse_request_end);
333
334static int queue_interrupt(struct fuse_req *req)
335{
336 struct fuse_iqueue *fiq = &req->fm->fc->iq;
337
338 spin_lock(&fiq->lock);
339 /* Check for we've sent request to interrupt this req */
340 if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) {
341 spin_unlock(&fiq->lock);
342 return -EINVAL;
343 }
344
345 if (list_empty(&req->intr_entry)) {
346 list_add_tail(&req->intr_entry, &fiq->interrupts);
347 /*
348 * Pairs with smp_mb() implied by test_and_set_bit()
349 * from fuse_request_end().
350 */
351 smp_mb();
352 if (test_bit(FR_FINISHED, &req->flags)) {
353 list_del_init(&req->intr_entry);
354 spin_unlock(&fiq->lock);
355 return 0;
356 }
357 fiq->ops->wake_interrupt_and_unlock(fiq);
358 } else {
359 spin_unlock(&fiq->lock);
360 }
361 return 0;
362}
363
364static void request_wait_answer(struct fuse_req *req)
365{
366 struct fuse_conn *fc = req->fm->fc;
367 struct fuse_iqueue *fiq = &fc->iq;
368 int err;
369
370 if (!fc->no_interrupt) {
371 /* Any signal may interrupt this */
372 err = wait_event_interruptible(req->waitq,
373 test_bit(FR_FINISHED, &req->flags));
374 if (!err)
375 return;
376
377 set_bit(FR_INTERRUPTED, &req->flags);
378 /* matches barrier in fuse_dev_do_read() */
379 smp_mb__after_atomic();
380 if (test_bit(FR_SENT, &req->flags))
381 queue_interrupt(req);
382 }
383
384 if (!test_bit(FR_FORCE, &req->flags)) {
385 /* Only fatal signals may interrupt this */
386 err = wait_event_killable(req->waitq,
387 test_bit(FR_FINISHED, &req->flags));
388 if (!err)
389 return;
390
391 spin_lock(&fiq->lock);
392 /* Request is not yet in userspace, bail out */
393 if (test_bit(FR_PENDING, &req->flags)) {
394 list_del(&req->list);
395 spin_unlock(&fiq->lock);
396 __fuse_put_request(req);
397 req->out.h.error = -EINTR;
398 return;
399 }
400 spin_unlock(&fiq->lock);
401 }
402
403 /*
404 * Either request is already in userspace, or it was forced.
405 * Wait it out.
406 */
407 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
408}
409
410static void __fuse_request_send(struct fuse_req *req)
411{
412 struct fuse_iqueue *fiq = &req->fm->fc->iq;
413
414 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
415 spin_lock(&fiq->lock);
416 if (!fiq->connected) {
417 spin_unlock(&fiq->lock);
418 req->out.h.error = -ENOTCONN;
419 } else {
420 req->in.h.unique = fuse_get_unique(fiq);
421 /* acquire extra reference, since request is still needed
422 after fuse_request_end() */
423 __fuse_get_request(req);
424 queue_request_and_unlock(fiq, req);
425
426 request_wait_answer(req);
427 /* Pairs with smp_wmb() in fuse_request_end() */
428 smp_rmb();
429 }
430}
431
432static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
433{
434 if (fc->minor < 4 && args->opcode == FUSE_STATFS)
435 args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE;
436
437 if (fc->minor < 9) {
438 switch (args->opcode) {
439 case FUSE_LOOKUP:
440 case FUSE_CREATE:
441 case FUSE_MKNOD:
442 case FUSE_MKDIR:
443 case FUSE_SYMLINK:
444 case FUSE_LINK:
445 args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
446 break;
447 case FUSE_GETATTR:
448 case FUSE_SETATTR:
449 args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
450 break;
451 }
452 }
453 if (fc->minor < 12) {
454 switch (args->opcode) {
455 case FUSE_CREATE:
456 args->in_args[0].size = sizeof(struct fuse_open_in);
457 break;
458 case FUSE_MKNOD:
459 args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
460 break;
461 }
462 }
463}
464
465static void fuse_force_creds(struct fuse_req *req)
466{
467 struct fuse_conn *fc = req->fm->fc;
468
469 req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid());
470 req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid());
471 req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
472}
473
474static void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args)
475{
476 req->in.h.opcode = args->opcode;
477 req->in.h.nodeid = args->nodeid;
478 req->args = args;
479 if (args->end)
480 __set_bit(FR_ASYNC, &req->flags);
481}
482
483ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args)
484{
485 struct fuse_conn *fc = fm->fc;
486 struct fuse_req *req;
487 ssize_t ret;
488
489 if (args->force) {
490 atomic_inc(&fc->num_waiting);
491 req = fuse_request_alloc(fm, GFP_KERNEL | __GFP_NOFAIL);
492
493 if (!args->nocreds)
494 fuse_force_creds(req);
495
496 __set_bit(FR_WAITING, &req->flags);
497 __set_bit(FR_FORCE, &req->flags);
498 } else {
499 WARN_ON(args->nocreds);
500 req = fuse_get_req(fm, false);
501 if (IS_ERR(req))
502 return PTR_ERR(req);
503 }
504
505 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
506 fuse_adjust_compat(fc, args);
507 fuse_args_to_req(req, args);
508
509 if (!args->noreply)
510 __set_bit(FR_ISREPLY, &req->flags);
511 __fuse_request_send(req);
512 ret = req->out.h.error;
513 if (!ret && args->out_argvar) {
514 BUG_ON(args->out_numargs == 0);
515 ret = args->out_args[args->out_numargs - 1].size;
516 }
517 fuse_put_request(req);
518
519 return ret;
520}
521
522static bool fuse_request_queue_background(struct fuse_req *req)
523{
524 struct fuse_mount *fm = req->fm;
525 struct fuse_conn *fc = fm->fc;
526 bool queued = false;
527
528 WARN_ON(!test_bit(FR_BACKGROUND, &req->flags));
529 if (!test_bit(FR_WAITING, &req->flags)) {
530 __set_bit(FR_WAITING, &req->flags);
531 atomic_inc(&fc->num_waiting);
532 }
533 __set_bit(FR_ISREPLY, &req->flags);
534 spin_lock(&fc->bg_lock);
535 if (likely(fc->connected)) {
536 fc->num_background++;
537 if (fc->num_background == fc->max_background)
538 fc->blocked = 1;
539 list_add_tail(&req->list, &fc->bg_queue);
540 flush_bg_queue(fc);
541 queued = true;
542 }
543 spin_unlock(&fc->bg_lock);
544
545 return queued;
546}
547
548int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args,
549 gfp_t gfp_flags)
550{
551 struct fuse_req *req;
552
553 if (args->force) {
554 WARN_ON(!args->nocreds);
555 req = fuse_request_alloc(fm, gfp_flags);
556 if (!req)
557 return -ENOMEM;
558 __set_bit(FR_BACKGROUND, &req->flags);
559 } else {
560 WARN_ON(args->nocreds);
561 req = fuse_get_req(fm, true);
562 if (IS_ERR(req))
563 return PTR_ERR(req);
564 }
565
566 fuse_args_to_req(req, args);
567
568 if (!fuse_request_queue_background(req)) {
569 fuse_put_request(req);
570 return -ENOTCONN;
571 }
572
573 return 0;
574}
575EXPORT_SYMBOL_GPL(fuse_simple_background);
576
577static int fuse_simple_notify_reply(struct fuse_mount *fm,
578 struct fuse_args *args, u64 unique)
579{
580 struct fuse_req *req;
581 struct fuse_iqueue *fiq = &fm->fc->iq;
582 int err = 0;
583
584 req = fuse_get_req(fm, false);
585 if (IS_ERR(req))
586 return PTR_ERR(req);
587
588 __clear_bit(FR_ISREPLY, &req->flags);
589 req->in.h.unique = unique;
590
591 fuse_args_to_req(req, args);
592
593 spin_lock(&fiq->lock);
594 if (fiq->connected) {
595 queue_request_and_unlock(fiq, req);
596 } else {
597 err = -ENODEV;
598 spin_unlock(&fiq->lock);
599 fuse_put_request(req);
600 }
601
602 return err;
603}
604
605/*
606 * Lock the request. Up to the next unlock_request() there mustn't be
607 * anything that could cause a page-fault. If the request was already
608 * aborted bail out.
609 */
610static int lock_request(struct fuse_req *req)
611{
612 int err = 0;
613 if (req) {
614 spin_lock(&req->waitq.lock);
615 if (test_bit(FR_ABORTED, &req->flags))
616 err = -ENOENT;
617 else
618 set_bit(FR_LOCKED, &req->flags);
619 spin_unlock(&req->waitq.lock);
620 }
621 return err;
622}
623
624/*
625 * Unlock request. If it was aborted while locked, caller is responsible
626 * for unlocking and ending the request.
627 */
628static int unlock_request(struct fuse_req *req)
629{
630 int err = 0;
631 if (req) {
632 spin_lock(&req->waitq.lock);
633 if (test_bit(FR_ABORTED, &req->flags))
634 err = -ENOENT;
635 else
636 clear_bit(FR_LOCKED, &req->flags);
637 spin_unlock(&req->waitq.lock);
638 }
639 return err;
640}
641
642struct fuse_copy_state {
643 int write;
644 struct fuse_req *req;
645 struct iov_iter *iter;
646 struct pipe_buffer *pipebufs;
647 struct pipe_buffer *currbuf;
648 struct pipe_inode_info *pipe;
649 unsigned long nr_segs;
650 struct page *pg;
651 unsigned len;
652 unsigned offset;
653 unsigned move_pages:1;
654};
655
656static void fuse_copy_init(struct fuse_copy_state *cs, int write,
657 struct iov_iter *iter)
658{
659 memset(cs, 0, sizeof(*cs));
660 cs->write = write;
661 cs->iter = iter;
662}
663
664/* Unmap and put previous page of userspace buffer */
665static void fuse_copy_finish(struct fuse_copy_state *cs)
666{
667 if (cs->currbuf) {
668 struct pipe_buffer *buf = cs->currbuf;
669
670 if (cs->write)
671 buf->len = PAGE_SIZE - cs->len;
672 cs->currbuf = NULL;
673 } else if (cs->pg) {
674 if (cs->write) {
675 flush_dcache_page(cs->pg);
676 set_page_dirty_lock(cs->pg);
677 }
678 put_page(cs->pg);
679 }
680 cs->pg = NULL;
681}
682
683/*
684 * Get another pagefull of userspace buffer, and map it to kernel
685 * address space, and lock request
686 */
687static int fuse_copy_fill(struct fuse_copy_state *cs)
688{
689 struct page *page;
690 int err;
691
692 err = unlock_request(cs->req);
693 if (err)
694 return err;
695
696 fuse_copy_finish(cs);
697 if (cs->pipebufs) {
698 struct pipe_buffer *buf = cs->pipebufs;
699
700 if (!cs->write) {
701 err = pipe_buf_confirm(cs->pipe, buf);
702 if (err)
703 return err;
704
705 BUG_ON(!cs->nr_segs);
706 cs->currbuf = buf;
707 cs->pg = buf->page;
708 cs->offset = buf->offset;
709 cs->len = buf->len;
710 cs->pipebufs++;
711 cs->nr_segs--;
712 } else {
713 if (cs->nr_segs >= cs->pipe->max_usage)
714 return -EIO;
715
716 page = alloc_page(GFP_HIGHUSER);
717 if (!page)
718 return -ENOMEM;
719
720 buf->page = page;
721 buf->offset = 0;
722 buf->len = 0;
723
724 cs->currbuf = buf;
725 cs->pg = page;
726 cs->offset = 0;
727 cs->len = PAGE_SIZE;
728 cs->pipebufs++;
729 cs->nr_segs++;
730 }
731 } else {
732 size_t off;
733 err = iov_iter_get_pages2(cs->iter, &page, PAGE_SIZE, 1, &off);
734 if (err < 0)
735 return err;
736 BUG_ON(!err);
737 cs->len = err;
738 cs->offset = off;
739 cs->pg = page;
740 }
741
742 return lock_request(cs->req);
743}
744
745/* Do as much copy to/from userspace buffer as we can */
746static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
747{
748 unsigned ncpy = min(*size, cs->len);
749 if (val) {
750 void *pgaddr = kmap_local_page(cs->pg);
751 void *buf = pgaddr + cs->offset;
752
753 if (cs->write)
754 memcpy(buf, *val, ncpy);
755 else
756 memcpy(*val, buf, ncpy);
757
758 kunmap_local(pgaddr);
759 *val += ncpy;
760 }
761 *size -= ncpy;
762 cs->len -= ncpy;
763 cs->offset += ncpy;
764 return ncpy;
765}
766
767static int fuse_check_folio(struct folio *folio)
768{
769 if (folio_mapped(folio) ||
770 folio->mapping != NULL ||
771 (folio->flags & PAGE_FLAGS_CHECK_AT_PREP &
772 ~(1 << PG_locked |
773 1 << PG_referenced |
774 1 << PG_uptodate |
775 1 << PG_lru |
776 1 << PG_active |
777 1 << PG_workingset |
778 1 << PG_reclaim |
779 1 << PG_waiters |
780 LRU_GEN_MASK | LRU_REFS_MASK))) {
781 dump_page(&folio->page, "fuse: trying to steal weird page");
782 return 1;
783 }
784 return 0;
785}
786
787static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
788{
789 int err;
790 struct folio *oldfolio = page_folio(*pagep);
791 struct folio *newfolio;
792 struct pipe_buffer *buf = cs->pipebufs;
793
794 folio_get(oldfolio);
795 err = unlock_request(cs->req);
796 if (err)
797 goto out_put_old;
798
799 fuse_copy_finish(cs);
800
801 err = pipe_buf_confirm(cs->pipe, buf);
802 if (err)
803 goto out_put_old;
804
805 BUG_ON(!cs->nr_segs);
806 cs->currbuf = buf;
807 cs->len = buf->len;
808 cs->pipebufs++;
809 cs->nr_segs--;
810
811 if (cs->len != PAGE_SIZE)
812 goto out_fallback;
813
814 if (!pipe_buf_try_steal(cs->pipe, buf))
815 goto out_fallback;
816
817 newfolio = page_folio(buf->page);
818
819 if (!folio_test_uptodate(newfolio))
820 folio_mark_uptodate(newfolio);
821
822 folio_clear_mappedtodisk(newfolio);
823
824 if (fuse_check_folio(newfolio) != 0)
825 goto out_fallback_unlock;
826
827 /*
828 * This is a new and locked page, it shouldn't be mapped or
829 * have any special flags on it
830 */
831 if (WARN_ON(folio_mapped(oldfolio)))
832 goto out_fallback_unlock;
833 if (WARN_ON(folio_has_private(oldfolio)))
834 goto out_fallback_unlock;
835 if (WARN_ON(folio_test_dirty(oldfolio) ||
836 folio_test_writeback(oldfolio)))
837 goto out_fallback_unlock;
838 if (WARN_ON(folio_test_mlocked(oldfolio)))
839 goto out_fallback_unlock;
840
841 replace_page_cache_folio(oldfolio, newfolio);
842
843 folio_get(newfolio);
844
845 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
846 folio_add_lru(newfolio);
847
848 /*
849 * Release while we have extra ref on stolen page. Otherwise
850 * anon_pipe_buf_release() might think the page can be reused.
851 */
852 pipe_buf_release(cs->pipe, buf);
853
854 err = 0;
855 spin_lock(&cs->req->waitq.lock);
856 if (test_bit(FR_ABORTED, &cs->req->flags))
857 err = -ENOENT;
858 else
859 *pagep = &newfolio->page;
860 spin_unlock(&cs->req->waitq.lock);
861
862 if (err) {
863 folio_unlock(newfolio);
864 folio_put(newfolio);
865 goto out_put_old;
866 }
867
868 folio_unlock(oldfolio);
869 /* Drop ref for ap->pages[] array */
870 folio_put(oldfolio);
871 cs->len = 0;
872
873 err = 0;
874out_put_old:
875 /* Drop ref obtained in this function */
876 folio_put(oldfolio);
877 return err;
878
879out_fallback_unlock:
880 folio_unlock(newfolio);
881out_fallback:
882 cs->pg = buf->page;
883 cs->offset = buf->offset;
884
885 err = lock_request(cs->req);
886 if (!err)
887 err = 1;
888
889 goto out_put_old;
890}
891
892static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
893 unsigned offset, unsigned count)
894{
895 struct pipe_buffer *buf;
896 int err;
897
898 if (cs->nr_segs >= cs->pipe->max_usage)
899 return -EIO;
900
901 get_page(page);
902 err = unlock_request(cs->req);
903 if (err) {
904 put_page(page);
905 return err;
906 }
907
908 fuse_copy_finish(cs);
909
910 buf = cs->pipebufs;
911 buf->page = page;
912 buf->offset = offset;
913 buf->len = count;
914
915 cs->pipebufs++;
916 cs->nr_segs++;
917 cs->len = 0;
918
919 return 0;
920}
921
922/*
923 * Copy a page in the request to/from the userspace buffer. Must be
924 * done atomically
925 */
926static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
927 unsigned offset, unsigned count, int zeroing)
928{
929 int err;
930 struct page *page = *pagep;
931
932 if (page && zeroing && count < PAGE_SIZE)
933 clear_highpage(page);
934
935 while (count) {
936 if (cs->write && cs->pipebufs && page) {
937 /*
938 * Can't control lifetime of pipe buffers, so always
939 * copy user pages.
940 */
941 if (cs->req->args->user_pages) {
942 err = fuse_copy_fill(cs);
943 if (err)
944 return err;
945 } else {
946 return fuse_ref_page(cs, page, offset, count);
947 }
948 } else if (!cs->len) {
949 if (cs->move_pages && page &&
950 offset == 0 && count == PAGE_SIZE) {
951 err = fuse_try_move_page(cs, pagep);
952 if (err <= 0)
953 return err;
954 } else {
955 err = fuse_copy_fill(cs);
956 if (err)
957 return err;
958 }
959 }
960 if (page) {
961 void *mapaddr = kmap_local_page(page);
962 void *buf = mapaddr + offset;
963 offset += fuse_copy_do(cs, &buf, &count);
964 kunmap_local(mapaddr);
965 } else
966 offset += fuse_copy_do(cs, NULL, &count);
967 }
968 if (page && !cs->write)
969 flush_dcache_page(page);
970 return 0;
971}
972
973/* Copy pages in the request to/from userspace buffer */
974static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
975 int zeroing)
976{
977 unsigned i;
978 struct fuse_req *req = cs->req;
979 struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args);
980
981
982 for (i = 0; i < ap->num_pages && (nbytes || zeroing); i++) {
983 int err;
984 unsigned int offset = ap->descs[i].offset;
985 unsigned int count = min(nbytes, ap->descs[i].length);
986
987 err = fuse_copy_page(cs, &ap->pages[i], offset, count, zeroing);
988 if (err)
989 return err;
990
991 nbytes -= count;
992 }
993 return 0;
994}
995
996/* Copy a single argument in the request to/from userspace buffer */
997static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
998{
999 while (size) {
1000 if (!cs->len) {
1001 int err = fuse_copy_fill(cs);
1002 if (err)
1003 return err;
1004 }
1005 fuse_copy_do(cs, &val, &size);
1006 }
1007 return 0;
1008}
1009
1010/* Copy request arguments to/from userspace buffer */
1011static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1012 unsigned argpages, struct fuse_arg *args,
1013 int zeroing)
1014{
1015 int err = 0;
1016 unsigned i;
1017
1018 for (i = 0; !err && i < numargs; i++) {
1019 struct fuse_arg *arg = &args[i];
1020 if (i == numargs - 1 && argpages)
1021 err = fuse_copy_pages(cs, arg->size, zeroing);
1022 else
1023 err = fuse_copy_one(cs, arg->value, arg->size);
1024 }
1025 return err;
1026}
1027
1028static int forget_pending(struct fuse_iqueue *fiq)
1029{
1030 return fiq->forget_list_head.next != NULL;
1031}
1032
1033static int request_pending(struct fuse_iqueue *fiq)
1034{
1035 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1036 forget_pending(fiq);
1037}
1038
1039/*
1040 * Transfer an interrupt request to userspace
1041 *
1042 * Unlike other requests this is assembled on demand, without a need
1043 * to allocate a separate fuse_req structure.
1044 *
1045 * Called with fiq->lock held, releases it
1046 */
1047static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1048 struct fuse_copy_state *cs,
1049 size_t nbytes, struct fuse_req *req)
1050__releases(fiq->lock)
1051{
1052 struct fuse_in_header ih;
1053 struct fuse_interrupt_in arg;
1054 unsigned reqsize = sizeof(ih) + sizeof(arg);
1055 int err;
1056
1057 list_del_init(&req->intr_entry);
1058 memset(&ih, 0, sizeof(ih));
1059 memset(&arg, 0, sizeof(arg));
1060 ih.len = reqsize;
1061 ih.opcode = FUSE_INTERRUPT;
1062 ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT);
1063 arg.unique = req->in.h.unique;
1064
1065 spin_unlock(&fiq->lock);
1066 if (nbytes < reqsize)
1067 return -EINVAL;
1068
1069 err = fuse_copy_one(cs, &ih, sizeof(ih));
1070 if (!err)
1071 err = fuse_copy_one(cs, &arg, sizeof(arg));
1072 fuse_copy_finish(cs);
1073
1074 return err ? err : reqsize;
1075}
1076
1077struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq,
1078 unsigned int max,
1079 unsigned int *countp)
1080{
1081 struct fuse_forget_link *head = fiq->forget_list_head.next;
1082 struct fuse_forget_link **newhead = &head;
1083 unsigned count;
1084
1085 for (count = 0; *newhead != NULL && count < max; count++)
1086 newhead = &(*newhead)->next;
1087
1088 fiq->forget_list_head.next = *newhead;
1089 *newhead = NULL;
1090 if (fiq->forget_list_head.next == NULL)
1091 fiq->forget_list_tail = &fiq->forget_list_head;
1092
1093 if (countp != NULL)
1094 *countp = count;
1095
1096 return head;
1097}
1098EXPORT_SYMBOL(fuse_dequeue_forget);
1099
1100static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1101 struct fuse_copy_state *cs,
1102 size_t nbytes)
1103__releases(fiq->lock)
1104{
1105 int err;
1106 struct fuse_forget_link *forget = fuse_dequeue_forget(fiq, 1, NULL);
1107 struct fuse_forget_in arg = {
1108 .nlookup = forget->forget_one.nlookup,
1109 };
1110 struct fuse_in_header ih = {
1111 .opcode = FUSE_FORGET,
1112 .nodeid = forget->forget_one.nodeid,
1113 .unique = fuse_get_unique(fiq),
1114 .len = sizeof(ih) + sizeof(arg),
1115 };
1116
1117 spin_unlock(&fiq->lock);
1118 kfree(forget);
1119 if (nbytes < ih.len)
1120 return -EINVAL;
1121
1122 err = fuse_copy_one(cs, &ih, sizeof(ih));
1123 if (!err)
1124 err = fuse_copy_one(cs, &arg, sizeof(arg));
1125 fuse_copy_finish(cs);
1126
1127 if (err)
1128 return err;
1129
1130 return ih.len;
1131}
1132
1133static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1134 struct fuse_copy_state *cs, size_t nbytes)
1135__releases(fiq->lock)
1136{
1137 int err;
1138 unsigned max_forgets;
1139 unsigned count;
1140 struct fuse_forget_link *head;
1141 struct fuse_batch_forget_in arg = { .count = 0 };
1142 struct fuse_in_header ih = {
1143 .opcode = FUSE_BATCH_FORGET,
1144 .unique = fuse_get_unique(fiq),
1145 .len = sizeof(ih) + sizeof(arg),
1146 };
1147
1148 if (nbytes < ih.len) {
1149 spin_unlock(&fiq->lock);
1150 return -EINVAL;
1151 }
1152
1153 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1154 head = fuse_dequeue_forget(fiq, max_forgets, &count);
1155 spin_unlock(&fiq->lock);
1156
1157 arg.count = count;
1158 ih.len += count * sizeof(struct fuse_forget_one);
1159 err = fuse_copy_one(cs, &ih, sizeof(ih));
1160 if (!err)
1161 err = fuse_copy_one(cs, &arg, sizeof(arg));
1162
1163 while (head) {
1164 struct fuse_forget_link *forget = head;
1165
1166 if (!err) {
1167 err = fuse_copy_one(cs, &forget->forget_one,
1168 sizeof(forget->forget_one));
1169 }
1170 head = forget->next;
1171 kfree(forget);
1172 }
1173
1174 fuse_copy_finish(cs);
1175
1176 if (err)
1177 return err;
1178
1179 return ih.len;
1180}
1181
1182static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1183 struct fuse_copy_state *cs,
1184 size_t nbytes)
1185__releases(fiq->lock)
1186{
1187 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1188 return fuse_read_single_forget(fiq, cs, nbytes);
1189 else
1190 return fuse_read_batch_forget(fiq, cs, nbytes);
1191}
1192
1193/*
1194 * Read a single request into the userspace filesystem's buffer. This
1195 * function waits until a request is available, then removes it from
1196 * the pending list and copies request data to userspace buffer. If
1197 * no reply is needed (FORGET) or request has been aborted or there
1198 * was an error during the copying then it's finished by calling
1199 * fuse_request_end(). Otherwise add it to the processing list, and set
1200 * the 'sent' flag.
1201 */
1202static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1203 struct fuse_copy_state *cs, size_t nbytes)
1204{
1205 ssize_t err;
1206 struct fuse_conn *fc = fud->fc;
1207 struct fuse_iqueue *fiq = &fc->iq;
1208 struct fuse_pqueue *fpq = &fud->pq;
1209 struct fuse_req *req;
1210 struct fuse_args *args;
1211 unsigned reqsize;
1212 unsigned int hash;
1213
1214 /*
1215 * Require sane minimum read buffer - that has capacity for fixed part
1216 * of any request header + negotiated max_write room for data.
1217 *
1218 * Historically libfuse reserves 4K for fixed header room, but e.g.
1219 * GlusterFS reserves only 80 bytes
1220 *
1221 * = `sizeof(fuse_in_header) + sizeof(fuse_write_in)`
1222 *
1223 * which is the absolute minimum any sane filesystem should be using
1224 * for header room.
1225 */
1226 if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER,
1227 sizeof(struct fuse_in_header) +
1228 sizeof(struct fuse_write_in) +
1229 fc->max_write))
1230 return -EINVAL;
1231
1232 restart:
1233 for (;;) {
1234 spin_lock(&fiq->lock);
1235 if (!fiq->connected || request_pending(fiq))
1236 break;
1237 spin_unlock(&fiq->lock);
1238
1239 if (file->f_flags & O_NONBLOCK)
1240 return -EAGAIN;
1241 err = wait_event_interruptible_exclusive(fiq->waitq,
1242 !fiq->connected || request_pending(fiq));
1243 if (err)
1244 return err;
1245 }
1246
1247 if (!fiq->connected) {
1248 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1249 goto err_unlock;
1250 }
1251
1252 if (!list_empty(&fiq->interrupts)) {
1253 req = list_entry(fiq->interrupts.next, struct fuse_req,
1254 intr_entry);
1255 return fuse_read_interrupt(fiq, cs, nbytes, req);
1256 }
1257
1258 if (forget_pending(fiq)) {
1259 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1260 return fuse_read_forget(fc, fiq, cs, nbytes);
1261
1262 if (fiq->forget_batch <= -8)
1263 fiq->forget_batch = 16;
1264 }
1265
1266 req = list_entry(fiq->pending.next, struct fuse_req, list);
1267 clear_bit(FR_PENDING, &req->flags);
1268 list_del_init(&req->list);
1269 spin_unlock(&fiq->lock);
1270
1271 args = req->args;
1272 reqsize = req->in.h.len;
1273
1274 /* If request is too large, reply with an error and restart the read */
1275 if (nbytes < reqsize) {
1276 req->out.h.error = -EIO;
1277 /* SETXATTR is special, since it may contain too large data */
1278 if (args->opcode == FUSE_SETXATTR)
1279 req->out.h.error = -E2BIG;
1280 fuse_request_end(req);
1281 goto restart;
1282 }
1283 spin_lock(&fpq->lock);
1284 /*
1285 * Must not put request on fpq->io queue after having been shut down by
1286 * fuse_abort_conn()
1287 */
1288 if (!fpq->connected) {
1289 req->out.h.error = err = -ECONNABORTED;
1290 goto out_end;
1291
1292 }
1293 list_add(&req->list, &fpq->io);
1294 spin_unlock(&fpq->lock);
1295 cs->req = req;
1296 err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h));
1297 if (!err)
1298 err = fuse_copy_args(cs, args->in_numargs, args->in_pages,
1299 (struct fuse_arg *) args->in_args, 0);
1300 fuse_copy_finish(cs);
1301 spin_lock(&fpq->lock);
1302 clear_bit(FR_LOCKED, &req->flags);
1303 if (!fpq->connected) {
1304 err = fc->aborted ? -ECONNABORTED : -ENODEV;
1305 goto out_end;
1306 }
1307 if (err) {
1308 req->out.h.error = -EIO;
1309 goto out_end;
1310 }
1311 if (!test_bit(FR_ISREPLY, &req->flags)) {
1312 err = reqsize;
1313 goto out_end;
1314 }
1315 hash = fuse_req_hash(req->in.h.unique);
1316 list_move_tail(&req->list, &fpq->processing[hash]);
1317 __fuse_get_request(req);
1318 set_bit(FR_SENT, &req->flags);
1319 spin_unlock(&fpq->lock);
1320 /* matches barrier in request_wait_answer() */
1321 smp_mb__after_atomic();
1322 if (test_bit(FR_INTERRUPTED, &req->flags))
1323 queue_interrupt(req);
1324 fuse_put_request(req);
1325
1326 return reqsize;
1327
1328out_end:
1329 if (!test_bit(FR_PRIVATE, &req->flags))
1330 list_del_init(&req->list);
1331 spin_unlock(&fpq->lock);
1332 fuse_request_end(req);
1333 return err;
1334
1335 err_unlock:
1336 spin_unlock(&fiq->lock);
1337 return err;
1338}
1339
1340static int fuse_dev_open(struct inode *inode, struct file *file)
1341{
1342 /*
1343 * The fuse device's file's private_data is used to hold
1344 * the fuse_conn(ection) when it is mounted, and is used to
1345 * keep track of whether the file has been mounted already.
1346 */
1347 file->private_data = NULL;
1348 return 0;
1349}
1350
1351static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1352{
1353 struct fuse_copy_state cs;
1354 struct file *file = iocb->ki_filp;
1355 struct fuse_dev *fud = fuse_get_dev(file);
1356
1357 if (!fud)
1358 return -EPERM;
1359
1360 if (!user_backed_iter(to))
1361 return -EINVAL;
1362
1363 fuse_copy_init(&cs, 1, to);
1364
1365 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1366}
1367
1368static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1369 struct pipe_inode_info *pipe,
1370 size_t len, unsigned int flags)
1371{
1372 int total, ret;
1373 int page_nr = 0;
1374 struct pipe_buffer *bufs;
1375 struct fuse_copy_state cs;
1376 struct fuse_dev *fud = fuse_get_dev(in);
1377
1378 if (!fud)
1379 return -EPERM;
1380
1381 bufs = kvmalloc_array(pipe->max_usage, sizeof(struct pipe_buffer),
1382 GFP_KERNEL);
1383 if (!bufs)
1384 return -ENOMEM;
1385
1386 fuse_copy_init(&cs, 1, NULL);
1387 cs.pipebufs = bufs;
1388 cs.pipe = pipe;
1389 ret = fuse_dev_do_read(fud, in, &cs, len);
1390 if (ret < 0)
1391 goto out;
1392
1393 if (pipe_occupancy(pipe->head, pipe->tail) + cs.nr_segs > pipe->max_usage) {
1394 ret = -EIO;
1395 goto out;
1396 }
1397
1398 for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1399 /*
1400 * Need to be careful about this. Having buf->ops in module
1401 * code can Oops if the buffer persists after module unload.
1402 */
1403 bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1404 bufs[page_nr].flags = 0;
1405 ret = add_to_pipe(pipe, &bufs[page_nr++]);
1406 if (unlikely(ret < 0))
1407 break;
1408 }
1409 if (total)
1410 ret = total;
1411out:
1412 for (; page_nr < cs.nr_segs; page_nr++)
1413 put_page(bufs[page_nr].page);
1414
1415 kvfree(bufs);
1416 return ret;
1417}
1418
1419static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1420 struct fuse_copy_state *cs)
1421{
1422 struct fuse_notify_poll_wakeup_out outarg;
1423 int err = -EINVAL;
1424
1425 if (size != sizeof(outarg))
1426 goto err;
1427
1428 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1429 if (err)
1430 goto err;
1431
1432 fuse_copy_finish(cs);
1433 return fuse_notify_poll_wakeup(fc, &outarg);
1434
1435err:
1436 fuse_copy_finish(cs);
1437 return err;
1438}
1439
1440static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1441 struct fuse_copy_state *cs)
1442{
1443 struct fuse_notify_inval_inode_out outarg;
1444 int err = -EINVAL;
1445
1446 if (size != sizeof(outarg))
1447 goto err;
1448
1449 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1450 if (err)
1451 goto err;
1452 fuse_copy_finish(cs);
1453
1454 down_read(&fc->killsb);
1455 err = fuse_reverse_inval_inode(fc, outarg.ino,
1456 outarg.off, outarg.len);
1457 up_read(&fc->killsb);
1458 return err;
1459
1460err:
1461 fuse_copy_finish(cs);
1462 return err;
1463}
1464
1465static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1466 struct fuse_copy_state *cs)
1467{
1468 struct fuse_notify_inval_entry_out outarg;
1469 int err = -ENOMEM;
1470 char *buf;
1471 struct qstr name;
1472
1473 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1474 if (!buf)
1475 goto err;
1476
1477 err = -EINVAL;
1478 if (size < sizeof(outarg))
1479 goto err;
1480
1481 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1482 if (err)
1483 goto err;
1484
1485 err = -ENAMETOOLONG;
1486 if (outarg.namelen > FUSE_NAME_MAX)
1487 goto err;
1488
1489 err = -EINVAL;
1490 if (size != sizeof(outarg) + outarg.namelen + 1)
1491 goto err;
1492
1493 name.name = buf;
1494 name.len = outarg.namelen;
1495 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1496 if (err)
1497 goto err;
1498 fuse_copy_finish(cs);
1499 buf[outarg.namelen] = 0;
1500
1501 down_read(&fc->killsb);
1502 err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name, outarg.flags);
1503 up_read(&fc->killsb);
1504 kfree(buf);
1505 return err;
1506
1507err:
1508 kfree(buf);
1509 fuse_copy_finish(cs);
1510 return err;
1511}
1512
1513static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1514 struct fuse_copy_state *cs)
1515{
1516 struct fuse_notify_delete_out outarg;
1517 int err = -ENOMEM;
1518 char *buf;
1519 struct qstr name;
1520
1521 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1522 if (!buf)
1523 goto err;
1524
1525 err = -EINVAL;
1526 if (size < sizeof(outarg))
1527 goto err;
1528
1529 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1530 if (err)
1531 goto err;
1532
1533 err = -ENAMETOOLONG;
1534 if (outarg.namelen > FUSE_NAME_MAX)
1535 goto err;
1536
1537 err = -EINVAL;
1538 if (size != sizeof(outarg) + outarg.namelen + 1)
1539 goto err;
1540
1541 name.name = buf;
1542 name.len = outarg.namelen;
1543 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1544 if (err)
1545 goto err;
1546 fuse_copy_finish(cs);
1547 buf[outarg.namelen] = 0;
1548
1549 down_read(&fc->killsb);
1550 err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name, 0);
1551 up_read(&fc->killsb);
1552 kfree(buf);
1553 return err;
1554
1555err:
1556 kfree(buf);
1557 fuse_copy_finish(cs);
1558 return err;
1559}
1560
1561static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1562 struct fuse_copy_state *cs)
1563{
1564 struct fuse_notify_store_out outarg;
1565 struct inode *inode;
1566 struct address_space *mapping;
1567 u64 nodeid;
1568 int err;
1569 pgoff_t index;
1570 unsigned int offset;
1571 unsigned int num;
1572 loff_t file_size;
1573 loff_t end;
1574
1575 err = -EINVAL;
1576 if (size < sizeof(outarg))
1577 goto out_finish;
1578
1579 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1580 if (err)
1581 goto out_finish;
1582
1583 err = -EINVAL;
1584 if (size - sizeof(outarg) != outarg.size)
1585 goto out_finish;
1586
1587 nodeid = outarg.nodeid;
1588
1589 down_read(&fc->killsb);
1590
1591 err = -ENOENT;
1592 inode = fuse_ilookup(fc, nodeid, NULL);
1593 if (!inode)
1594 goto out_up_killsb;
1595
1596 mapping = inode->i_mapping;
1597 index = outarg.offset >> PAGE_SHIFT;
1598 offset = outarg.offset & ~PAGE_MASK;
1599 file_size = i_size_read(inode);
1600 end = outarg.offset + outarg.size;
1601 if (end > file_size) {
1602 file_size = end;
1603 fuse_write_update_attr(inode, file_size, outarg.size);
1604 }
1605
1606 num = outarg.size;
1607 while (num) {
1608 struct page *page;
1609 unsigned int this_num;
1610
1611 err = -ENOMEM;
1612 page = find_or_create_page(mapping, index,
1613 mapping_gfp_mask(mapping));
1614 if (!page)
1615 goto out_iput;
1616
1617 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1618 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1619 if (!err && offset == 0 &&
1620 (this_num == PAGE_SIZE || file_size == end))
1621 SetPageUptodate(page);
1622 unlock_page(page);
1623 put_page(page);
1624
1625 if (err)
1626 goto out_iput;
1627
1628 num -= this_num;
1629 offset = 0;
1630 index++;
1631 }
1632
1633 err = 0;
1634
1635out_iput:
1636 iput(inode);
1637out_up_killsb:
1638 up_read(&fc->killsb);
1639out_finish:
1640 fuse_copy_finish(cs);
1641 return err;
1642}
1643
1644struct fuse_retrieve_args {
1645 struct fuse_args_pages ap;
1646 struct fuse_notify_retrieve_in inarg;
1647};
1648
1649static void fuse_retrieve_end(struct fuse_mount *fm, struct fuse_args *args,
1650 int error)
1651{
1652 struct fuse_retrieve_args *ra =
1653 container_of(args, typeof(*ra), ap.args);
1654
1655 release_pages(ra->ap.pages, ra->ap.num_pages);
1656 kfree(ra);
1657}
1658
1659static int fuse_retrieve(struct fuse_mount *fm, struct inode *inode,
1660 struct fuse_notify_retrieve_out *outarg)
1661{
1662 int err;
1663 struct address_space *mapping = inode->i_mapping;
1664 pgoff_t index;
1665 loff_t file_size;
1666 unsigned int num;
1667 unsigned int offset;
1668 size_t total_len = 0;
1669 unsigned int num_pages;
1670 struct fuse_conn *fc = fm->fc;
1671 struct fuse_retrieve_args *ra;
1672 size_t args_size = sizeof(*ra);
1673 struct fuse_args_pages *ap;
1674 struct fuse_args *args;
1675
1676 offset = outarg->offset & ~PAGE_MASK;
1677 file_size = i_size_read(inode);
1678
1679 num = min(outarg->size, fc->max_write);
1680 if (outarg->offset > file_size)
1681 num = 0;
1682 else if (outarg->offset + num > file_size)
1683 num = file_size - outarg->offset;
1684
1685 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1686 num_pages = min(num_pages, fc->max_pages);
1687
1688 args_size += num_pages * (sizeof(ap->pages[0]) + sizeof(ap->descs[0]));
1689
1690 ra = kzalloc(args_size, GFP_KERNEL);
1691 if (!ra)
1692 return -ENOMEM;
1693
1694 ap = &ra->ap;
1695 ap->pages = (void *) (ra + 1);
1696 ap->descs = (void *) (ap->pages + num_pages);
1697
1698 args = &ap->args;
1699 args->nodeid = outarg->nodeid;
1700 args->opcode = FUSE_NOTIFY_REPLY;
1701 args->in_numargs = 2;
1702 args->in_pages = true;
1703 args->end = fuse_retrieve_end;
1704
1705 index = outarg->offset >> PAGE_SHIFT;
1706
1707 while (num && ap->num_pages < num_pages) {
1708 struct page *page;
1709 unsigned int this_num;
1710
1711 page = find_get_page(mapping, index);
1712 if (!page)
1713 break;
1714
1715 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1716 ap->pages[ap->num_pages] = page;
1717 ap->descs[ap->num_pages].offset = offset;
1718 ap->descs[ap->num_pages].length = this_num;
1719 ap->num_pages++;
1720
1721 offset = 0;
1722 num -= this_num;
1723 total_len += this_num;
1724 index++;
1725 }
1726 ra->inarg.offset = outarg->offset;
1727 ra->inarg.size = total_len;
1728 args->in_args[0].size = sizeof(ra->inarg);
1729 args->in_args[0].value = &ra->inarg;
1730 args->in_args[1].size = total_len;
1731
1732 err = fuse_simple_notify_reply(fm, args, outarg->notify_unique);
1733 if (err)
1734 fuse_retrieve_end(fm, args, err);
1735
1736 return err;
1737}
1738
1739static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1740 struct fuse_copy_state *cs)
1741{
1742 struct fuse_notify_retrieve_out outarg;
1743 struct fuse_mount *fm;
1744 struct inode *inode;
1745 u64 nodeid;
1746 int err;
1747
1748 err = -EINVAL;
1749 if (size != sizeof(outarg))
1750 goto copy_finish;
1751
1752 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1753 if (err)
1754 goto copy_finish;
1755
1756 fuse_copy_finish(cs);
1757
1758 down_read(&fc->killsb);
1759 err = -ENOENT;
1760 nodeid = outarg.nodeid;
1761
1762 inode = fuse_ilookup(fc, nodeid, &fm);
1763 if (inode) {
1764 err = fuse_retrieve(fm, inode, &outarg);
1765 iput(inode);
1766 }
1767 up_read(&fc->killsb);
1768
1769 return err;
1770
1771copy_finish:
1772 fuse_copy_finish(cs);
1773 return err;
1774}
1775
1776static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1777 unsigned int size, struct fuse_copy_state *cs)
1778{
1779 /* Don't try to move pages (yet) */
1780 cs->move_pages = 0;
1781
1782 switch (code) {
1783 case FUSE_NOTIFY_POLL:
1784 return fuse_notify_poll(fc, size, cs);
1785
1786 case FUSE_NOTIFY_INVAL_INODE:
1787 return fuse_notify_inval_inode(fc, size, cs);
1788
1789 case FUSE_NOTIFY_INVAL_ENTRY:
1790 return fuse_notify_inval_entry(fc, size, cs);
1791
1792 case FUSE_NOTIFY_STORE:
1793 return fuse_notify_store(fc, size, cs);
1794
1795 case FUSE_NOTIFY_RETRIEVE:
1796 return fuse_notify_retrieve(fc, size, cs);
1797
1798 case FUSE_NOTIFY_DELETE:
1799 return fuse_notify_delete(fc, size, cs);
1800
1801 default:
1802 fuse_copy_finish(cs);
1803 return -EINVAL;
1804 }
1805}
1806
1807/* Look up request on processing list by unique ID */
1808static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1809{
1810 unsigned int hash = fuse_req_hash(unique);
1811 struct fuse_req *req;
1812
1813 list_for_each_entry(req, &fpq->processing[hash], list) {
1814 if (req->in.h.unique == unique)
1815 return req;
1816 }
1817 return NULL;
1818}
1819
1820static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args,
1821 unsigned nbytes)
1822{
1823 unsigned reqsize = sizeof(struct fuse_out_header);
1824
1825 reqsize += fuse_len_args(args->out_numargs, args->out_args);
1826
1827 if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar))
1828 return -EINVAL;
1829 else if (reqsize > nbytes) {
1830 struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1];
1831 unsigned diffsize = reqsize - nbytes;
1832
1833 if (diffsize > lastarg->size)
1834 return -EINVAL;
1835 lastarg->size -= diffsize;
1836 }
1837 return fuse_copy_args(cs, args->out_numargs, args->out_pages,
1838 args->out_args, args->page_zeroing);
1839}
1840
1841/*
1842 * Write a single reply to a request. First the header is copied from
1843 * the write buffer. The request is then searched on the processing
1844 * list by the unique ID found in the header. If found, then remove
1845 * it from the list and copy the rest of the buffer to the request.
1846 * The request is finished by calling fuse_request_end().
1847 */
1848static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1849 struct fuse_copy_state *cs, size_t nbytes)
1850{
1851 int err;
1852 struct fuse_conn *fc = fud->fc;
1853 struct fuse_pqueue *fpq = &fud->pq;
1854 struct fuse_req *req;
1855 struct fuse_out_header oh;
1856
1857 err = -EINVAL;
1858 if (nbytes < sizeof(struct fuse_out_header))
1859 goto out;
1860
1861 err = fuse_copy_one(cs, &oh, sizeof(oh));
1862 if (err)
1863 goto copy_finish;
1864
1865 err = -EINVAL;
1866 if (oh.len != nbytes)
1867 goto copy_finish;
1868
1869 /*
1870 * Zero oh.unique indicates unsolicited notification message
1871 * and error contains notification code.
1872 */
1873 if (!oh.unique) {
1874 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1875 goto out;
1876 }
1877
1878 err = -EINVAL;
1879 if (oh.error <= -512 || oh.error > 0)
1880 goto copy_finish;
1881
1882 spin_lock(&fpq->lock);
1883 req = NULL;
1884 if (fpq->connected)
1885 req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
1886
1887 err = -ENOENT;
1888 if (!req) {
1889 spin_unlock(&fpq->lock);
1890 goto copy_finish;
1891 }
1892
1893 /* Is it an interrupt reply ID? */
1894 if (oh.unique & FUSE_INT_REQ_BIT) {
1895 __fuse_get_request(req);
1896 spin_unlock(&fpq->lock);
1897
1898 err = 0;
1899 if (nbytes != sizeof(struct fuse_out_header))
1900 err = -EINVAL;
1901 else if (oh.error == -ENOSYS)
1902 fc->no_interrupt = 1;
1903 else if (oh.error == -EAGAIN)
1904 err = queue_interrupt(req);
1905
1906 fuse_put_request(req);
1907
1908 goto copy_finish;
1909 }
1910
1911 clear_bit(FR_SENT, &req->flags);
1912 list_move(&req->list, &fpq->io);
1913 req->out.h = oh;
1914 set_bit(FR_LOCKED, &req->flags);
1915 spin_unlock(&fpq->lock);
1916 cs->req = req;
1917 if (!req->args->page_replace)
1918 cs->move_pages = 0;
1919
1920 if (oh.error)
1921 err = nbytes != sizeof(oh) ? -EINVAL : 0;
1922 else
1923 err = copy_out_args(cs, req->args, nbytes);
1924 fuse_copy_finish(cs);
1925
1926 spin_lock(&fpq->lock);
1927 clear_bit(FR_LOCKED, &req->flags);
1928 if (!fpq->connected)
1929 err = -ENOENT;
1930 else if (err)
1931 req->out.h.error = -EIO;
1932 if (!test_bit(FR_PRIVATE, &req->flags))
1933 list_del_init(&req->list);
1934 spin_unlock(&fpq->lock);
1935
1936 fuse_request_end(req);
1937out:
1938 return err ? err : nbytes;
1939
1940copy_finish:
1941 fuse_copy_finish(cs);
1942 goto out;
1943}
1944
1945static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1946{
1947 struct fuse_copy_state cs;
1948 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1949
1950 if (!fud)
1951 return -EPERM;
1952
1953 if (!user_backed_iter(from))
1954 return -EINVAL;
1955
1956 fuse_copy_init(&cs, 0, from);
1957
1958 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1959}
1960
1961static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1962 struct file *out, loff_t *ppos,
1963 size_t len, unsigned int flags)
1964{
1965 unsigned int head, tail, mask, count;
1966 unsigned nbuf;
1967 unsigned idx;
1968 struct pipe_buffer *bufs;
1969 struct fuse_copy_state cs;
1970 struct fuse_dev *fud;
1971 size_t rem;
1972 ssize_t ret;
1973
1974 fud = fuse_get_dev(out);
1975 if (!fud)
1976 return -EPERM;
1977
1978 pipe_lock(pipe);
1979
1980 head = pipe->head;
1981 tail = pipe->tail;
1982 mask = pipe->ring_size - 1;
1983 count = head - tail;
1984
1985 bufs = kvmalloc_array(count, sizeof(struct pipe_buffer), GFP_KERNEL);
1986 if (!bufs) {
1987 pipe_unlock(pipe);
1988 return -ENOMEM;
1989 }
1990
1991 nbuf = 0;
1992 rem = 0;
1993 for (idx = tail; idx != head && rem < len; idx++)
1994 rem += pipe->bufs[idx & mask].len;
1995
1996 ret = -EINVAL;
1997 if (rem < len)
1998 goto out_free;
1999
2000 rem = len;
2001 while (rem) {
2002 struct pipe_buffer *ibuf;
2003 struct pipe_buffer *obuf;
2004
2005 if (WARN_ON(nbuf >= count || tail == head))
2006 goto out_free;
2007
2008 ibuf = &pipe->bufs[tail & mask];
2009 obuf = &bufs[nbuf];
2010
2011 if (rem >= ibuf->len) {
2012 *obuf = *ibuf;
2013 ibuf->ops = NULL;
2014 tail++;
2015 pipe->tail = tail;
2016 } else {
2017 if (!pipe_buf_get(pipe, ibuf))
2018 goto out_free;
2019
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);
2042out_free:
2043 for (idx = 0; idx < nbuf; idx++) {
2044 struct pipe_buffer *buf = &bufs[idx];
2045
2046 if (buf->ops)
2047 pipe_buf_release(pipe, buf);
2048 }
2049 pipe_unlock(pipe);
2050
2051 kvfree(bufs);
2052 return ret;
2053}
2054
2055static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2056{
2057 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
2058 struct fuse_iqueue *fiq;
2059 struct fuse_dev *fud = fuse_get_dev(file);
2060
2061 if (!fud)
2062 return EPOLLERR;
2063
2064 fiq = &fud->fc->iq;
2065 poll_wait(file, &fiq->waitq, wait);
2066
2067 spin_lock(&fiq->lock);
2068 if (!fiq->connected)
2069 mask = EPOLLERR;
2070 else if (request_pending(fiq))
2071 mask |= EPOLLIN | EPOLLRDNORM;
2072 spin_unlock(&fiq->lock);
2073
2074 return mask;
2075}
2076
2077/* Abort all requests on the given list (pending or processing) */
2078static void end_requests(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 fuse_request_end(req);
2087 }
2088}
2089
2090static 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.rst).
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 */
2123void 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_end);
2132 unsigned int i;
2133
2134 /* Background queuing checks fc->connected under bg_lock */
2135 spin_lock(&fc->bg_lock);
2136 fc->connected = 0;
2137 spin_unlock(&fc->bg_lock);
2138
2139 fuse_set_initialized(fc);
2140 list_for_each_entry(fud, &fc->devices, entry) {
2141 struct fuse_pqueue *fpq = &fud->pq;
2142
2143 spin_lock(&fpq->lock);
2144 fpq->connected = 0;
2145 list_for_each_entry_safe(req, next, &fpq->io, list) {
2146 req->out.h.error = -ECONNABORTED;
2147 spin_lock(&req->waitq.lock);
2148 set_bit(FR_ABORTED, &req->flags);
2149 if (!test_bit(FR_LOCKED, &req->flags)) {
2150 set_bit(FR_PRIVATE, &req->flags);
2151 __fuse_get_request(req);
2152 list_move(&req->list, &to_end);
2153 }
2154 spin_unlock(&req->waitq.lock);
2155 }
2156 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2157 list_splice_tail_init(&fpq->processing[i],
2158 &to_end);
2159 spin_unlock(&fpq->lock);
2160 }
2161 spin_lock(&fc->bg_lock);
2162 fc->blocked = 0;
2163 fc->max_background = UINT_MAX;
2164 flush_bg_queue(fc);
2165 spin_unlock(&fc->bg_lock);
2166
2167 spin_lock(&fiq->lock);
2168 fiq->connected = 0;
2169 list_for_each_entry(req, &fiq->pending, list)
2170 clear_bit(FR_PENDING, &req->flags);
2171 list_splice_tail_init(&fiq->pending, &to_end);
2172 while (forget_pending(fiq))
2173 kfree(fuse_dequeue_forget(fiq, 1, NULL));
2174 wake_up_all(&fiq->waitq);
2175 spin_unlock(&fiq->lock);
2176 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2177 end_polls(fc);
2178 wake_up_all(&fc->blocked_waitq);
2179 spin_unlock(&fc->lock);
2180
2181 end_requests(&to_end);
2182 } else {
2183 spin_unlock(&fc->lock);
2184 }
2185}
2186EXPORT_SYMBOL_GPL(fuse_abort_conn);
2187
2188void fuse_wait_aborted(struct fuse_conn *fc)
2189{
2190 /* matches implicit memory barrier in fuse_drop_waiting() */
2191 smp_mb();
2192 wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2193}
2194
2195int fuse_dev_release(struct inode *inode, struct file *file)
2196{
2197 struct fuse_dev *fud = fuse_get_dev(file);
2198
2199 if (fud) {
2200 struct fuse_conn *fc = fud->fc;
2201 struct fuse_pqueue *fpq = &fud->pq;
2202 LIST_HEAD(to_end);
2203 unsigned int i;
2204
2205 spin_lock(&fpq->lock);
2206 WARN_ON(!list_empty(&fpq->io));
2207 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2208 list_splice_init(&fpq->processing[i], &to_end);
2209 spin_unlock(&fpq->lock);
2210
2211 end_requests(&to_end);
2212
2213 /* Are we the last open device? */
2214 if (atomic_dec_and_test(&fc->dev_count)) {
2215 WARN_ON(fc->iq.fasync != NULL);
2216 fuse_abort_conn(fc);
2217 }
2218 fuse_dev_free(fud);
2219 }
2220 return 0;
2221}
2222EXPORT_SYMBOL_GPL(fuse_dev_release);
2223
2224static int fuse_dev_fasync(int fd, struct file *file, int on)
2225{
2226 struct fuse_dev *fud = fuse_get_dev(file);
2227
2228 if (!fud)
2229 return -EPERM;
2230
2231 /* No locking - fasync_helper does its own locking */
2232 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2233}
2234
2235static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2236{
2237 struct fuse_dev *fud;
2238
2239 if (new->private_data)
2240 return -EINVAL;
2241
2242 fud = fuse_dev_alloc_install(fc);
2243 if (!fud)
2244 return -ENOMEM;
2245
2246 new->private_data = fud;
2247 atomic_inc(&fc->dev_count);
2248
2249 return 0;
2250}
2251
2252static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2253 unsigned long arg)
2254{
2255 int res;
2256 int oldfd;
2257 struct fuse_dev *fud = NULL;
2258
2259 switch (cmd) {
2260 case FUSE_DEV_IOC_CLONE:
2261 res = -EFAULT;
2262 if (!get_user(oldfd, (__u32 __user *)arg)) {
2263 struct file *old = fget(oldfd);
2264
2265 res = -EINVAL;
2266 if (old) {
2267 /*
2268 * Check against file->f_op because CUSE
2269 * uses the same ioctl handler.
2270 */
2271 if (old->f_op == file->f_op)
2272 fud = fuse_get_dev(old);
2273
2274 if (fud) {
2275 mutex_lock(&fuse_mutex);
2276 res = fuse_device_clone(fud->fc, file);
2277 mutex_unlock(&fuse_mutex);
2278 }
2279 fput(old);
2280 }
2281 }
2282 break;
2283 default:
2284 res = -ENOTTY;
2285 break;
2286 }
2287 return res;
2288}
2289
2290const struct file_operations fuse_dev_operations = {
2291 .owner = THIS_MODULE,
2292 .open = fuse_dev_open,
2293 .llseek = no_llseek,
2294 .read_iter = fuse_dev_read,
2295 .splice_read = fuse_dev_splice_read,
2296 .write_iter = fuse_dev_write,
2297 .splice_write = fuse_dev_splice_write,
2298 .poll = fuse_dev_poll,
2299 .release = fuse_dev_release,
2300 .fasync = fuse_dev_fasync,
2301 .unlocked_ioctl = fuse_dev_ioctl,
2302 .compat_ioctl = compat_ptr_ioctl,
2303};
2304EXPORT_SYMBOL_GPL(fuse_dev_operations);
2305
2306static struct miscdevice fuse_miscdevice = {
2307 .minor = FUSE_MINOR,
2308 .name = "fuse",
2309 .fops = &fuse_dev_operations,
2310};
2311
2312int __init fuse_dev_init(void)
2313{
2314 int err = -ENOMEM;
2315 fuse_req_cachep = kmem_cache_create("fuse_request",
2316 sizeof(struct fuse_req),
2317 0, 0, NULL);
2318 if (!fuse_req_cachep)
2319 goto out;
2320
2321 err = misc_register(&fuse_miscdevice);
2322 if (err)
2323 goto out_cache_clean;
2324
2325 return 0;
2326
2327 out_cache_clean:
2328 kmem_cache_destroy(fuse_req_cachep);
2329 out:
2330 return err;
2331}
2332
2333void fuse_dev_cleanup(void)
2334{
2335 misc_deregister(&fuse_miscdevice);
2336 kmem_cache_destroy(fuse_req_cachep);
2337}
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/poll.h>
14#include <linux/uio.h>
15#include <linux/miscdevice.h>
16#include <linux/pagemap.h>
17#include <linux/file.h>
18#include <linux/slab.h>
19#include <linux/pipe_fs_i.h>
20#include <linux/swap.h>
21#include <linux/splice.h>
22
23MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24MODULE_ALIAS("devname:fuse");
25
26static struct kmem_cache *fuse_req_cachep;
27
28static struct fuse_conn *fuse_get_conn(struct file *file)
29{
30 /*
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
33 */
34 return file->private_data;
35}
36
37static void fuse_request_init(struct fuse_req *req)
38{
39 memset(req, 0, sizeof(*req));
40 INIT_LIST_HEAD(&req->list);
41 INIT_LIST_HEAD(&req->intr_entry);
42 init_waitqueue_head(&req->waitq);
43 atomic_set(&req->count, 1);
44}
45
46struct fuse_req *fuse_request_alloc(void)
47{
48 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL);
49 if (req)
50 fuse_request_init(req);
51 return req;
52}
53EXPORT_SYMBOL_GPL(fuse_request_alloc);
54
55struct fuse_req *fuse_request_alloc_nofs(void)
56{
57 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_NOFS);
58 if (req)
59 fuse_request_init(req);
60 return req;
61}
62
63void fuse_request_free(struct fuse_req *req)
64{
65 kmem_cache_free(fuse_req_cachep, req);
66}
67
68static void block_sigs(sigset_t *oldset)
69{
70 sigset_t mask;
71
72 siginitsetinv(&mask, sigmask(SIGKILL));
73 sigprocmask(SIG_BLOCK, &mask, oldset);
74}
75
76static void restore_sigs(sigset_t *oldset)
77{
78 sigprocmask(SIG_SETMASK, oldset, NULL);
79}
80
81static void __fuse_get_request(struct fuse_req *req)
82{
83 atomic_inc(&req->count);
84}
85
86/* Must be called with > 1 refcount */
87static void __fuse_put_request(struct fuse_req *req)
88{
89 BUG_ON(atomic_read(&req->count) < 2);
90 atomic_dec(&req->count);
91}
92
93static void fuse_req_init_context(struct fuse_req *req)
94{
95 req->in.h.uid = current_fsuid();
96 req->in.h.gid = current_fsgid();
97 req->in.h.pid = current->pid;
98}
99
100struct fuse_req *fuse_get_req(struct fuse_conn *fc)
101{
102 struct fuse_req *req;
103 sigset_t oldset;
104 int intr;
105 int err;
106
107 atomic_inc(&fc->num_waiting);
108 block_sigs(&oldset);
109 intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked);
110 restore_sigs(&oldset);
111 err = -EINTR;
112 if (intr)
113 goto out;
114
115 err = -ENOTCONN;
116 if (!fc->connected)
117 goto out;
118
119 req = fuse_request_alloc();
120 err = -ENOMEM;
121 if (!req)
122 goto out;
123
124 fuse_req_init_context(req);
125 req->waiting = 1;
126 return req;
127
128 out:
129 atomic_dec(&fc->num_waiting);
130 return ERR_PTR(err);
131}
132EXPORT_SYMBOL_GPL(fuse_get_req);
133
134/*
135 * Return request in fuse_file->reserved_req. However that may
136 * currently be in use. If that is the case, wait for it to become
137 * available.
138 */
139static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
140 struct file *file)
141{
142 struct fuse_req *req = NULL;
143 struct fuse_file *ff = file->private_data;
144
145 do {
146 wait_event(fc->reserved_req_waitq, ff->reserved_req);
147 spin_lock(&fc->lock);
148 if (ff->reserved_req) {
149 req = ff->reserved_req;
150 ff->reserved_req = NULL;
151 get_file(file);
152 req->stolen_file = file;
153 }
154 spin_unlock(&fc->lock);
155 } while (!req);
156
157 return req;
158}
159
160/*
161 * Put stolen request back into fuse_file->reserved_req
162 */
163static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
164{
165 struct file *file = req->stolen_file;
166 struct fuse_file *ff = file->private_data;
167
168 spin_lock(&fc->lock);
169 fuse_request_init(req);
170 BUG_ON(ff->reserved_req);
171 ff->reserved_req = req;
172 wake_up_all(&fc->reserved_req_waitq);
173 spin_unlock(&fc->lock);
174 fput(file);
175}
176
177/*
178 * Gets a requests for a file operation, always succeeds
179 *
180 * This is used for sending the FLUSH request, which must get to
181 * userspace, due to POSIX locks which may need to be unlocked.
182 *
183 * If allocation fails due to OOM, use the reserved request in
184 * fuse_file.
185 *
186 * This is very unlikely to deadlock accidentally, since the
187 * filesystem should not have it's own file open. If deadlock is
188 * intentional, it can still be broken by "aborting" the filesystem.
189 */
190struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file)
191{
192 struct fuse_req *req;
193
194 atomic_inc(&fc->num_waiting);
195 wait_event(fc->blocked_waitq, !fc->blocked);
196 req = fuse_request_alloc();
197 if (!req)
198 req = get_reserved_req(fc, file);
199
200 fuse_req_init_context(req);
201 req->waiting = 1;
202 return req;
203}
204
205void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
206{
207 if (atomic_dec_and_test(&req->count)) {
208 if (req->waiting)
209 atomic_dec(&fc->num_waiting);
210
211 if (req->stolen_file)
212 put_reserved_req(fc, req);
213 else
214 fuse_request_free(req);
215 }
216}
217EXPORT_SYMBOL_GPL(fuse_put_request);
218
219static unsigned len_args(unsigned numargs, struct fuse_arg *args)
220{
221 unsigned nbytes = 0;
222 unsigned i;
223
224 for (i = 0; i < numargs; i++)
225 nbytes += args[i].size;
226
227 return nbytes;
228}
229
230static u64 fuse_get_unique(struct fuse_conn *fc)
231{
232 fc->reqctr++;
233 /* zero is special */
234 if (fc->reqctr == 0)
235 fc->reqctr = 1;
236
237 return fc->reqctr;
238}
239
240static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
241{
242 req->in.h.len = sizeof(struct fuse_in_header) +
243 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
244 list_add_tail(&req->list, &fc->pending);
245 req->state = FUSE_REQ_PENDING;
246 if (!req->waiting) {
247 req->waiting = 1;
248 atomic_inc(&fc->num_waiting);
249 }
250 wake_up(&fc->waitq);
251 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
252}
253
254void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
255 u64 nodeid, u64 nlookup)
256{
257 forget->forget_one.nodeid = nodeid;
258 forget->forget_one.nlookup = nlookup;
259
260 spin_lock(&fc->lock);
261 if (fc->connected) {
262 fc->forget_list_tail->next = forget;
263 fc->forget_list_tail = forget;
264 wake_up(&fc->waitq);
265 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
266 } else {
267 kfree(forget);
268 }
269 spin_unlock(&fc->lock);
270}
271
272static void flush_bg_queue(struct fuse_conn *fc)
273{
274 while (fc->active_background < fc->max_background &&
275 !list_empty(&fc->bg_queue)) {
276 struct fuse_req *req;
277
278 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
279 list_del(&req->list);
280 fc->active_background++;
281 req->in.h.unique = fuse_get_unique(fc);
282 queue_request(fc, req);
283 }
284}
285
286/*
287 * This function is called when a request is finished. Either a reply
288 * has arrived or it was aborted (and not yet sent) or some error
289 * occurred during communication with userspace, or the device file
290 * was closed. The requester thread is woken up (if still waiting),
291 * the 'end' callback is called if given, else the reference to the
292 * request is released
293 *
294 * Called with fc->lock, unlocks it
295 */
296static void request_end(struct fuse_conn *fc, struct fuse_req *req)
297__releases(fc->lock)
298{
299 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
300 req->end = NULL;
301 list_del(&req->list);
302 list_del(&req->intr_entry);
303 req->state = FUSE_REQ_FINISHED;
304 if (req->background) {
305 if (fc->num_background == fc->max_background) {
306 fc->blocked = 0;
307 wake_up_all(&fc->blocked_waitq);
308 }
309 if (fc->num_background == fc->congestion_threshold &&
310 fc->connected && fc->bdi_initialized) {
311 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
312 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
313 }
314 fc->num_background--;
315 fc->active_background--;
316 flush_bg_queue(fc);
317 }
318 spin_unlock(&fc->lock);
319 wake_up(&req->waitq);
320 if (end)
321 end(fc, req);
322 fuse_put_request(fc, req);
323}
324
325static void wait_answer_interruptible(struct fuse_conn *fc,
326 struct fuse_req *req)
327__releases(fc->lock)
328__acquires(fc->lock)
329{
330 if (signal_pending(current))
331 return;
332
333 spin_unlock(&fc->lock);
334 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
335 spin_lock(&fc->lock);
336}
337
338static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
339{
340 list_add_tail(&req->intr_entry, &fc->interrupts);
341 wake_up(&fc->waitq);
342 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
343}
344
345static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
346__releases(fc->lock)
347__acquires(fc->lock)
348{
349 if (!fc->no_interrupt) {
350 /* Any signal may interrupt this */
351 wait_answer_interruptible(fc, req);
352
353 if (req->aborted)
354 goto aborted;
355 if (req->state == FUSE_REQ_FINISHED)
356 return;
357
358 req->interrupted = 1;
359 if (req->state == FUSE_REQ_SENT)
360 queue_interrupt(fc, req);
361 }
362
363 if (!req->force) {
364 sigset_t oldset;
365
366 /* Only fatal signals may interrupt this */
367 block_sigs(&oldset);
368 wait_answer_interruptible(fc, req);
369 restore_sigs(&oldset);
370
371 if (req->aborted)
372 goto aborted;
373 if (req->state == FUSE_REQ_FINISHED)
374 return;
375
376 /* Request is not yet in userspace, bail out */
377 if (req->state == FUSE_REQ_PENDING) {
378 list_del(&req->list);
379 __fuse_put_request(req);
380 req->out.h.error = -EINTR;
381 return;
382 }
383 }
384
385 /*
386 * Either request is already in userspace, or it was forced.
387 * Wait it out.
388 */
389 spin_unlock(&fc->lock);
390 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
391 spin_lock(&fc->lock);
392
393 if (!req->aborted)
394 return;
395
396 aborted:
397 BUG_ON(req->state != FUSE_REQ_FINISHED);
398 if (req->locked) {
399 /* This is uninterruptible sleep, because data is
400 being copied to/from the buffers of req. During
401 locked state, there mustn't be any filesystem
402 operation (e.g. page fault), since that could lead
403 to deadlock */
404 spin_unlock(&fc->lock);
405 wait_event(req->waitq, !req->locked);
406 spin_lock(&fc->lock);
407 }
408}
409
410void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
411{
412 req->isreply = 1;
413 spin_lock(&fc->lock);
414 if (!fc->connected)
415 req->out.h.error = -ENOTCONN;
416 else if (fc->conn_error)
417 req->out.h.error = -ECONNREFUSED;
418 else {
419 req->in.h.unique = fuse_get_unique(fc);
420 queue_request(fc, req);
421 /* acquire extra reference, since request is still needed
422 after request_end() */
423 __fuse_get_request(req);
424
425 request_wait_answer(fc, req);
426 }
427 spin_unlock(&fc->lock);
428}
429EXPORT_SYMBOL_GPL(fuse_request_send);
430
431static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
432 struct fuse_req *req)
433{
434 req->background = 1;
435 fc->num_background++;
436 if (fc->num_background == fc->max_background)
437 fc->blocked = 1;
438 if (fc->num_background == fc->congestion_threshold &&
439 fc->bdi_initialized) {
440 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
441 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
442 }
443 list_add_tail(&req->list, &fc->bg_queue);
444 flush_bg_queue(fc);
445}
446
447static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
448{
449 spin_lock(&fc->lock);
450 if (fc->connected) {
451 fuse_request_send_nowait_locked(fc, req);
452 spin_unlock(&fc->lock);
453 } else {
454 req->out.h.error = -ENOTCONN;
455 request_end(fc, req);
456 }
457}
458
459void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
460{
461 req->isreply = 1;
462 fuse_request_send_nowait(fc, req);
463}
464EXPORT_SYMBOL_GPL(fuse_request_send_background);
465
466static int fuse_request_send_notify_reply(struct fuse_conn *fc,
467 struct fuse_req *req, u64 unique)
468{
469 int err = -ENODEV;
470
471 req->isreply = 0;
472 req->in.h.unique = unique;
473 spin_lock(&fc->lock);
474 if (fc->connected) {
475 queue_request(fc, req);
476 err = 0;
477 }
478 spin_unlock(&fc->lock);
479
480 return err;
481}
482
483/*
484 * Called under fc->lock
485 *
486 * fc->connected must have been checked previously
487 */
488void fuse_request_send_background_locked(struct fuse_conn *fc,
489 struct fuse_req *req)
490{
491 req->isreply = 1;
492 fuse_request_send_nowait_locked(fc, req);
493}
494
495/*
496 * Lock the request. Up to the next unlock_request() there mustn't be
497 * anything that could cause a page-fault. If the request was already
498 * aborted bail out.
499 */
500static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
501{
502 int err = 0;
503 if (req) {
504 spin_lock(&fc->lock);
505 if (req->aborted)
506 err = -ENOENT;
507 else
508 req->locked = 1;
509 spin_unlock(&fc->lock);
510 }
511 return err;
512}
513
514/*
515 * Unlock request. If it was aborted during being locked, the
516 * requester thread is currently waiting for it to be unlocked, so
517 * wake it up.
518 */
519static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
520{
521 if (req) {
522 spin_lock(&fc->lock);
523 req->locked = 0;
524 if (req->aborted)
525 wake_up(&req->waitq);
526 spin_unlock(&fc->lock);
527 }
528}
529
530struct fuse_copy_state {
531 struct fuse_conn *fc;
532 int write;
533 struct fuse_req *req;
534 const struct iovec *iov;
535 struct pipe_buffer *pipebufs;
536 struct pipe_buffer *currbuf;
537 struct pipe_inode_info *pipe;
538 unsigned long nr_segs;
539 unsigned long seglen;
540 unsigned long addr;
541 struct page *pg;
542 void *mapaddr;
543 void *buf;
544 unsigned len;
545 unsigned move_pages:1;
546};
547
548static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
549 int write,
550 const struct iovec *iov, unsigned long nr_segs)
551{
552 memset(cs, 0, sizeof(*cs));
553 cs->fc = fc;
554 cs->write = write;
555 cs->iov = iov;
556 cs->nr_segs = nr_segs;
557}
558
559/* Unmap and put previous page of userspace buffer */
560static void fuse_copy_finish(struct fuse_copy_state *cs)
561{
562 if (cs->currbuf) {
563 struct pipe_buffer *buf = cs->currbuf;
564
565 if (!cs->write) {
566 buf->ops->unmap(cs->pipe, buf, cs->mapaddr);
567 } else {
568 kunmap(buf->page);
569 buf->len = PAGE_SIZE - cs->len;
570 }
571 cs->currbuf = NULL;
572 cs->mapaddr = NULL;
573 } else if (cs->mapaddr) {
574 kunmap(cs->pg);
575 if (cs->write) {
576 flush_dcache_page(cs->pg);
577 set_page_dirty_lock(cs->pg);
578 }
579 put_page(cs->pg);
580 cs->mapaddr = NULL;
581 }
582}
583
584/*
585 * Get another pagefull of userspace buffer, and map it to kernel
586 * address space, and lock request
587 */
588static int fuse_copy_fill(struct fuse_copy_state *cs)
589{
590 unsigned long offset;
591 int err;
592
593 unlock_request(cs->fc, cs->req);
594 fuse_copy_finish(cs);
595 if (cs->pipebufs) {
596 struct pipe_buffer *buf = cs->pipebufs;
597
598 if (!cs->write) {
599 err = buf->ops->confirm(cs->pipe, buf);
600 if (err)
601 return err;
602
603 BUG_ON(!cs->nr_segs);
604 cs->currbuf = buf;
605 cs->mapaddr = buf->ops->map(cs->pipe, buf, 0);
606 cs->len = buf->len;
607 cs->buf = cs->mapaddr + buf->offset;
608 cs->pipebufs++;
609 cs->nr_segs--;
610 } else {
611 struct page *page;
612
613 if (cs->nr_segs == cs->pipe->buffers)
614 return -EIO;
615
616 page = alloc_page(GFP_HIGHUSER);
617 if (!page)
618 return -ENOMEM;
619
620 buf->page = page;
621 buf->offset = 0;
622 buf->len = 0;
623
624 cs->currbuf = buf;
625 cs->mapaddr = kmap(page);
626 cs->buf = cs->mapaddr;
627 cs->len = PAGE_SIZE;
628 cs->pipebufs++;
629 cs->nr_segs++;
630 }
631 } else {
632 if (!cs->seglen) {
633 BUG_ON(!cs->nr_segs);
634 cs->seglen = cs->iov[0].iov_len;
635 cs->addr = (unsigned long) cs->iov[0].iov_base;
636 cs->iov++;
637 cs->nr_segs--;
638 }
639 err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg);
640 if (err < 0)
641 return err;
642 BUG_ON(err != 1);
643 offset = cs->addr % PAGE_SIZE;
644 cs->mapaddr = kmap(cs->pg);
645 cs->buf = cs->mapaddr + offset;
646 cs->len = min(PAGE_SIZE - offset, cs->seglen);
647 cs->seglen -= cs->len;
648 cs->addr += cs->len;
649 }
650
651 return lock_request(cs->fc, cs->req);
652}
653
654/* Do as much copy to/from userspace buffer as we can */
655static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
656{
657 unsigned ncpy = min(*size, cs->len);
658 if (val) {
659 if (cs->write)
660 memcpy(cs->buf, *val, ncpy);
661 else
662 memcpy(*val, cs->buf, ncpy);
663 *val += ncpy;
664 }
665 *size -= ncpy;
666 cs->len -= ncpy;
667 cs->buf += ncpy;
668 return ncpy;
669}
670
671static int fuse_check_page(struct page *page)
672{
673 if (page_mapcount(page) ||
674 page->mapping != NULL ||
675 page_count(page) != 1 ||
676 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
677 ~(1 << PG_locked |
678 1 << PG_referenced |
679 1 << PG_uptodate |
680 1 << PG_lru |
681 1 << PG_active |
682 1 << PG_reclaim))) {
683 printk(KERN_WARNING "fuse: trying to steal weird page\n");
684 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);
685 return 1;
686 }
687 return 0;
688}
689
690static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
691{
692 int err;
693 struct page *oldpage = *pagep;
694 struct page *newpage;
695 struct pipe_buffer *buf = cs->pipebufs;
696 struct address_space *mapping;
697 pgoff_t index;
698
699 unlock_request(cs->fc, cs->req);
700 fuse_copy_finish(cs);
701
702 err = buf->ops->confirm(cs->pipe, buf);
703 if (err)
704 return err;
705
706 BUG_ON(!cs->nr_segs);
707 cs->currbuf = buf;
708 cs->len = buf->len;
709 cs->pipebufs++;
710 cs->nr_segs--;
711
712 if (cs->len != PAGE_SIZE)
713 goto out_fallback;
714
715 if (buf->ops->steal(cs->pipe, buf) != 0)
716 goto out_fallback;
717
718 newpage = buf->page;
719
720 if (WARN_ON(!PageUptodate(newpage)))
721 return -EIO;
722
723 ClearPageMappedToDisk(newpage);
724
725 if (fuse_check_page(newpage) != 0)
726 goto out_fallback_unlock;
727
728 mapping = oldpage->mapping;
729 index = oldpage->index;
730
731 /*
732 * This is a new and locked page, it shouldn't be mapped or
733 * have any special flags on it
734 */
735 if (WARN_ON(page_mapped(oldpage)))
736 goto out_fallback_unlock;
737 if (WARN_ON(page_has_private(oldpage)))
738 goto out_fallback_unlock;
739 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
740 goto out_fallback_unlock;
741 if (WARN_ON(PageMlocked(oldpage)))
742 goto out_fallback_unlock;
743
744 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
745 if (err) {
746 unlock_page(newpage);
747 return err;
748 }
749
750 page_cache_get(newpage);
751
752 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
753 lru_cache_add_file(newpage);
754
755 err = 0;
756 spin_lock(&cs->fc->lock);
757 if (cs->req->aborted)
758 err = -ENOENT;
759 else
760 *pagep = newpage;
761 spin_unlock(&cs->fc->lock);
762
763 if (err) {
764 unlock_page(newpage);
765 page_cache_release(newpage);
766 return err;
767 }
768
769 unlock_page(oldpage);
770 page_cache_release(oldpage);
771 cs->len = 0;
772
773 return 0;
774
775out_fallback_unlock:
776 unlock_page(newpage);
777out_fallback:
778 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
779 cs->buf = cs->mapaddr + buf->offset;
780
781 err = lock_request(cs->fc, cs->req);
782 if (err)
783 return err;
784
785 return 1;
786}
787
788static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
789 unsigned offset, unsigned count)
790{
791 struct pipe_buffer *buf;
792
793 if (cs->nr_segs == cs->pipe->buffers)
794 return -EIO;
795
796 unlock_request(cs->fc, cs->req);
797 fuse_copy_finish(cs);
798
799 buf = cs->pipebufs;
800 page_cache_get(page);
801 buf->page = page;
802 buf->offset = offset;
803 buf->len = count;
804
805 cs->pipebufs++;
806 cs->nr_segs++;
807 cs->len = 0;
808
809 return 0;
810}
811
812/*
813 * Copy a page in the request to/from the userspace buffer. Must be
814 * done atomically
815 */
816static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
817 unsigned offset, unsigned count, int zeroing)
818{
819 int err;
820 struct page *page = *pagep;
821
822 if (page && zeroing && count < PAGE_SIZE)
823 clear_highpage(page);
824
825 while (count) {
826 if (cs->write && cs->pipebufs && page) {
827 return fuse_ref_page(cs, page, offset, count);
828 } else if (!cs->len) {
829 if (cs->move_pages && page &&
830 offset == 0 && count == PAGE_SIZE) {
831 err = fuse_try_move_page(cs, pagep);
832 if (err <= 0)
833 return err;
834 } else {
835 err = fuse_copy_fill(cs);
836 if (err)
837 return err;
838 }
839 }
840 if (page) {
841 void *mapaddr = kmap_atomic(page, KM_USER0);
842 void *buf = mapaddr + offset;
843 offset += fuse_copy_do(cs, &buf, &count);
844 kunmap_atomic(mapaddr, KM_USER0);
845 } else
846 offset += fuse_copy_do(cs, NULL, &count);
847 }
848 if (page && !cs->write)
849 flush_dcache_page(page);
850 return 0;
851}
852
853/* Copy pages in the request to/from userspace buffer */
854static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
855 int zeroing)
856{
857 unsigned i;
858 struct fuse_req *req = cs->req;
859 unsigned offset = req->page_offset;
860 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
861
862 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
863 int err;
864
865 err = fuse_copy_page(cs, &req->pages[i], offset, count,
866 zeroing);
867 if (err)
868 return err;
869
870 nbytes -= count;
871 count = min(nbytes, (unsigned) PAGE_SIZE);
872 offset = 0;
873 }
874 return 0;
875}
876
877/* Copy a single argument in the request to/from userspace buffer */
878static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
879{
880 while (size) {
881 if (!cs->len) {
882 int err = fuse_copy_fill(cs);
883 if (err)
884 return err;
885 }
886 fuse_copy_do(cs, &val, &size);
887 }
888 return 0;
889}
890
891/* Copy request arguments to/from userspace buffer */
892static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
893 unsigned argpages, struct fuse_arg *args,
894 int zeroing)
895{
896 int err = 0;
897 unsigned i;
898
899 for (i = 0; !err && i < numargs; i++) {
900 struct fuse_arg *arg = &args[i];
901 if (i == numargs - 1 && argpages)
902 err = fuse_copy_pages(cs, arg->size, zeroing);
903 else
904 err = fuse_copy_one(cs, arg->value, arg->size);
905 }
906 return err;
907}
908
909static int forget_pending(struct fuse_conn *fc)
910{
911 return fc->forget_list_head.next != NULL;
912}
913
914static int request_pending(struct fuse_conn *fc)
915{
916 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) ||
917 forget_pending(fc);
918}
919
920/* Wait until a request is available on the pending list */
921static void request_wait(struct fuse_conn *fc)
922__releases(fc->lock)
923__acquires(fc->lock)
924{
925 DECLARE_WAITQUEUE(wait, current);
926
927 add_wait_queue_exclusive(&fc->waitq, &wait);
928 while (fc->connected && !request_pending(fc)) {
929 set_current_state(TASK_INTERRUPTIBLE);
930 if (signal_pending(current))
931 break;
932
933 spin_unlock(&fc->lock);
934 schedule();
935 spin_lock(&fc->lock);
936 }
937 set_current_state(TASK_RUNNING);
938 remove_wait_queue(&fc->waitq, &wait);
939}
940
941/*
942 * Transfer an interrupt request to userspace
943 *
944 * Unlike other requests this is assembled on demand, without a need
945 * to allocate a separate fuse_req structure.
946 *
947 * Called with fc->lock held, releases it
948 */
949static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
950 size_t nbytes, struct fuse_req *req)
951__releases(fc->lock)
952{
953 struct fuse_in_header ih;
954 struct fuse_interrupt_in arg;
955 unsigned reqsize = sizeof(ih) + sizeof(arg);
956 int err;
957
958 list_del_init(&req->intr_entry);
959 req->intr_unique = fuse_get_unique(fc);
960 memset(&ih, 0, sizeof(ih));
961 memset(&arg, 0, sizeof(arg));
962 ih.len = reqsize;
963 ih.opcode = FUSE_INTERRUPT;
964 ih.unique = req->intr_unique;
965 arg.unique = req->in.h.unique;
966
967 spin_unlock(&fc->lock);
968 if (nbytes < reqsize)
969 return -EINVAL;
970
971 err = fuse_copy_one(cs, &ih, sizeof(ih));
972 if (!err)
973 err = fuse_copy_one(cs, &arg, sizeof(arg));
974 fuse_copy_finish(cs);
975
976 return err ? err : reqsize;
977}
978
979static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc,
980 unsigned max,
981 unsigned *countp)
982{
983 struct fuse_forget_link *head = fc->forget_list_head.next;
984 struct fuse_forget_link **newhead = &head;
985 unsigned count;
986
987 for (count = 0; *newhead != NULL && count < max; count++)
988 newhead = &(*newhead)->next;
989
990 fc->forget_list_head.next = *newhead;
991 *newhead = NULL;
992 if (fc->forget_list_head.next == NULL)
993 fc->forget_list_tail = &fc->forget_list_head;
994
995 if (countp != NULL)
996 *countp = count;
997
998 return head;
999}
1000
1001static int fuse_read_single_forget(struct fuse_conn *fc,
1002 struct fuse_copy_state *cs,
1003 size_t nbytes)
1004__releases(fc->lock)
1005{
1006 int err;
1007 struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL);
1008 struct fuse_forget_in arg = {
1009 .nlookup = forget->forget_one.nlookup,
1010 };
1011 struct fuse_in_header ih = {
1012 .opcode = FUSE_FORGET,
1013 .nodeid = forget->forget_one.nodeid,
1014 .unique = fuse_get_unique(fc),
1015 .len = sizeof(ih) + sizeof(arg),
1016 };
1017
1018 spin_unlock(&fc->lock);
1019 kfree(forget);
1020 if (nbytes < ih.len)
1021 return -EINVAL;
1022
1023 err = fuse_copy_one(cs, &ih, sizeof(ih));
1024 if (!err)
1025 err = fuse_copy_one(cs, &arg, sizeof(arg));
1026 fuse_copy_finish(cs);
1027
1028 if (err)
1029 return err;
1030
1031 return ih.len;
1032}
1033
1034static int fuse_read_batch_forget(struct fuse_conn *fc,
1035 struct fuse_copy_state *cs, size_t nbytes)
1036__releases(fc->lock)
1037{
1038 int err;
1039 unsigned max_forgets;
1040 unsigned count;
1041 struct fuse_forget_link *head;
1042 struct fuse_batch_forget_in arg = { .count = 0 };
1043 struct fuse_in_header ih = {
1044 .opcode = FUSE_BATCH_FORGET,
1045 .unique = fuse_get_unique(fc),
1046 .len = sizeof(ih) + sizeof(arg),
1047 };
1048
1049 if (nbytes < ih.len) {
1050 spin_unlock(&fc->lock);
1051 return -EINVAL;
1052 }
1053
1054 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1055 head = dequeue_forget(fc, max_forgets, &count);
1056 spin_unlock(&fc->lock);
1057
1058 arg.count = count;
1059 ih.len += count * sizeof(struct fuse_forget_one);
1060 err = fuse_copy_one(cs, &ih, sizeof(ih));
1061 if (!err)
1062 err = fuse_copy_one(cs, &arg, sizeof(arg));
1063
1064 while (head) {
1065 struct fuse_forget_link *forget = head;
1066
1067 if (!err) {
1068 err = fuse_copy_one(cs, &forget->forget_one,
1069 sizeof(forget->forget_one));
1070 }
1071 head = forget->next;
1072 kfree(forget);
1073 }
1074
1075 fuse_copy_finish(cs);
1076
1077 if (err)
1078 return err;
1079
1080 return ih.len;
1081}
1082
1083static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs,
1084 size_t nbytes)
1085__releases(fc->lock)
1086{
1087 if (fc->minor < 16 || fc->forget_list_head.next->next == NULL)
1088 return fuse_read_single_forget(fc, cs, nbytes);
1089 else
1090 return fuse_read_batch_forget(fc, cs, nbytes);
1091}
1092
1093/*
1094 * Read a single request into the userspace filesystem's buffer. This
1095 * function waits until a request is available, then removes it from
1096 * the pending list and copies request data to userspace buffer. If
1097 * no reply is needed (FORGET) or request has been aborted or there
1098 * was an error during the copying then it's finished by calling
1099 * request_end(). Otherwise add it to the processing list, and set
1100 * the 'sent' flag.
1101 */
1102static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1103 struct fuse_copy_state *cs, size_t nbytes)
1104{
1105 int err;
1106 struct fuse_req *req;
1107 struct fuse_in *in;
1108 unsigned reqsize;
1109
1110 restart:
1111 spin_lock(&fc->lock);
1112 err = -EAGAIN;
1113 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
1114 !request_pending(fc))
1115 goto err_unlock;
1116
1117 request_wait(fc);
1118 err = -ENODEV;
1119 if (!fc->connected)
1120 goto err_unlock;
1121 err = -ERESTARTSYS;
1122 if (!request_pending(fc))
1123 goto err_unlock;
1124
1125 if (!list_empty(&fc->interrupts)) {
1126 req = list_entry(fc->interrupts.next, struct fuse_req,
1127 intr_entry);
1128 return fuse_read_interrupt(fc, cs, nbytes, req);
1129 }
1130
1131 if (forget_pending(fc)) {
1132 if (list_empty(&fc->pending) || fc->forget_batch-- > 0)
1133 return fuse_read_forget(fc, cs, nbytes);
1134
1135 if (fc->forget_batch <= -8)
1136 fc->forget_batch = 16;
1137 }
1138
1139 req = list_entry(fc->pending.next, struct fuse_req, list);
1140 req->state = FUSE_REQ_READING;
1141 list_move(&req->list, &fc->io);
1142
1143 in = &req->in;
1144 reqsize = in->h.len;
1145 /* If request is too large, reply with an error and restart the read */
1146 if (nbytes < reqsize) {
1147 req->out.h.error = -EIO;
1148 /* SETXATTR is special, since it may contain too large data */
1149 if (in->h.opcode == FUSE_SETXATTR)
1150 req->out.h.error = -E2BIG;
1151 request_end(fc, req);
1152 goto restart;
1153 }
1154 spin_unlock(&fc->lock);
1155 cs->req = req;
1156 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1157 if (!err)
1158 err = fuse_copy_args(cs, in->numargs, in->argpages,
1159 (struct fuse_arg *) in->args, 0);
1160 fuse_copy_finish(cs);
1161 spin_lock(&fc->lock);
1162 req->locked = 0;
1163 if (req->aborted) {
1164 request_end(fc, req);
1165 return -ENODEV;
1166 }
1167 if (err) {
1168 req->out.h.error = -EIO;
1169 request_end(fc, req);
1170 return err;
1171 }
1172 if (!req->isreply)
1173 request_end(fc, req);
1174 else {
1175 req->state = FUSE_REQ_SENT;
1176 list_move_tail(&req->list, &fc->processing);
1177 if (req->interrupted)
1178 queue_interrupt(fc, req);
1179 spin_unlock(&fc->lock);
1180 }
1181 return reqsize;
1182
1183 err_unlock:
1184 spin_unlock(&fc->lock);
1185 return err;
1186}
1187
1188static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1189 unsigned long nr_segs, loff_t pos)
1190{
1191 struct fuse_copy_state cs;
1192 struct file *file = iocb->ki_filp;
1193 struct fuse_conn *fc = fuse_get_conn(file);
1194 if (!fc)
1195 return -EPERM;
1196
1197 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1198
1199 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1200}
1201
1202static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe,
1203 struct pipe_buffer *buf)
1204{
1205 return 1;
1206}
1207
1208static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = {
1209 .can_merge = 0,
1210 .map = generic_pipe_buf_map,
1211 .unmap = generic_pipe_buf_unmap,
1212 .confirm = generic_pipe_buf_confirm,
1213 .release = generic_pipe_buf_release,
1214 .steal = fuse_dev_pipe_buf_steal,
1215 .get = generic_pipe_buf_get,
1216};
1217
1218static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1219 struct pipe_inode_info *pipe,
1220 size_t len, unsigned int flags)
1221{
1222 int ret;
1223 int page_nr = 0;
1224 int do_wakeup = 0;
1225 struct pipe_buffer *bufs;
1226 struct fuse_copy_state cs;
1227 struct fuse_conn *fc = fuse_get_conn(in);
1228 if (!fc)
1229 return -EPERM;
1230
1231 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1232 if (!bufs)
1233 return -ENOMEM;
1234
1235 fuse_copy_init(&cs, fc, 1, NULL, 0);
1236 cs.pipebufs = bufs;
1237 cs.pipe = pipe;
1238 ret = fuse_dev_do_read(fc, in, &cs, len);
1239 if (ret < 0)
1240 goto out;
1241
1242 ret = 0;
1243 pipe_lock(pipe);
1244
1245 if (!pipe->readers) {
1246 send_sig(SIGPIPE, current, 0);
1247 if (!ret)
1248 ret = -EPIPE;
1249 goto out_unlock;
1250 }
1251
1252 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1253 ret = -EIO;
1254 goto out_unlock;
1255 }
1256
1257 while (page_nr < cs.nr_segs) {
1258 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1259 struct pipe_buffer *buf = pipe->bufs + newbuf;
1260
1261 buf->page = bufs[page_nr].page;
1262 buf->offset = bufs[page_nr].offset;
1263 buf->len = bufs[page_nr].len;
1264 buf->ops = &fuse_dev_pipe_buf_ops;
1265
1266 pipe->nrbufs++;
1267 page_nr++;
1268 ret += buf->len;
1269
1270 if (pipe->inode)
1271 do_wakeup = 1;
1272 }
1273
1274out_unlock:
1275 pipe_unlock(pipe);
1276
1277 if (do_wakeup) {
1278 smp_mb();
1279 if (waitqueue_active(&pipe->wait))
1280 wake_up_interruptible(&pipe->wait);
1281 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1282 }
1283
1284out:
1285 for (; page_nr < cs.nr_segs; page_nr++)
1286 page_cache_release(bufs[page_nr].page);
1287
1288 kfree(bufs);
1289 return ret;
1290}
1291
1292static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1293 struct fuse_copy_state *cs)
1294{
1295 struct fuse_notify_poll_wakeup_out outarg;
1296 int err = -EINVAL;
1297
1298 if (size != sizeof(outarg))
1299 goto err;
1300
1301 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1302 if (err)
1303 goto err;
1304
1305 fuse_copy_finish(cs);
1306 return fuse_notify_poll_wakeup(fc, &outarg);
1307
1308err:
1309 fuse_copy_finish(cs);
1310 return err;
1311}
1312
1313static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1314 struct fuse_copy_state *cs)
1315{
1316 struct fuse_notify_inval_inode_out outarg;
1317 int err = -EINVAL;
1318
1319 if (size != sizeof(outarg))
1320 goto err;
1321
1322 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1323 if (err)
1324 goto err;
1325 fuse_copy_finish(cs);
1326
1327 down_read(&fc->killsb);
1328 err = -ENOENT;
1329 if (fc->sb) {
1330 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1331 outarg.off, outarg.len);
1332 }
1333 up_read(&fc->killsb);
1334 return err;
1335
1336err:
1337 fuse_copy_finish(cs);
1338 return err;
1339}
1340
1341static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1342 struct fuse_copy_state *cs)
1343{
1344 struct fuse_notify_inval_entry_out outarg;
1345 int err = -ENOMEM;
1346 char *buf;
1347 struct qstr name;
1348
1349 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1350 if (!buf)
1351 goto err;
1352
1353 err = -EINVAL;
1354 if (size < sizeof(outarg))
1355 goto err;
1356
1357 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1358 if (err)
1359 goto err;
1360
1361 err = -ENAMETOOLONG;
1362 if (outarg.namelen > FUSE_NAME_MAX)
1363 goto err;
1364
1365 err = -EINVAL;
1366 if (size != sizeof(outarg) + outarg.namelen + 1)
1367 goto err;
1368
1369 name.name = buf;
1370 name.len = outarg.namelen;
1371 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1372 if (err)
1373 goto err;
1374 fuse_copy_finish(cs);
1375 buf[outarg.namelen] = 0;
1376 name.hash = full_name_hash(name.name, name.len);
1377
1378 down_read(&fc->killsb);
1379 err = -ENOENT;
1380 if (fc->sb)
1381 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, &name);
1382 up_read(&fc->killsb);
1383 kfree(buf);
1384 return err;
1385
1386err:
1387 kfree(buf);
1388 fuse_copy_finish(cs);
1389 return err;
1390}
1391
1392static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1393 struct fuse_copy_state *cs)
1394{
1395 struct fuse_notify_store_out outarg;
1396 struct inode *inode;
1397 struct address_space *mapping;
1398 u64 nodeid;
1399 int err;
1400 pgoff_t index;
1401 unsigned int offset;
1402 unsigned int num;
1403 loff_t file_size;
1404 loff_t end;
1405
1406 err = -EINVAL;
1407 if (size < sizeof(outarg))
1408 goto out_finish;
1409
1410 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1411 if (err)
1412 goto out_finish;
1413
1414 err = -EINVAL;
1415 if (size - sizeof(outarg) != outarg.size)
1416 goto out_finish;
1417
1418 nodeid = outarg.nodeid;
1419
1420 down_read(&fc->killsb);
1421
1422 err = -ENOENT;
1423 if (!fc->sb)
1424 goto out_up_killsb;
1425
1426 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1427 if (!inode)
1428 goto out_up_killsb;
1429
1430 mapping = inode->i_mapping;
1431 index = outarg.offset >> PAGE_CACHE_SHIFT;
1432 offset = outarg.offset & ~PAGE_CACHE_MASK;
1433 file_size = i_size_read(inode);
1434 end = outarg.offset + outarg.size;
1435 if (end > file_size) {
1436 file_size = end;
1437 fuse_write_update_size(inode, file_size);
1438 }
1439
1440 num = outarg.size;
1441 while (num) {
1442 struct page *page;
1443 unsigned int this_num;
1444
1445 err = -ENOMEM;
1446 page = find_or_create_page(mapping, index,
1447 mapping_gfp_mask(mapping));
1448 if (!page)
1449 goto out_iput;
1450
1451 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1452 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1453 if (!err && offset == 0 && (num != 0 || file_size == end))
1454 SetPageUptodate(page);
1455 unlock_page(page);
1456 page_cache_release(page);
1457
1458 if (err)
1459 goto out_iput;
1460
1461 num -= this_num;
1462 offset = 0;
1463 index++;
1464 }
1465
1466 err = 0;
1467
1468out_iput:
1469 iput(inode);
1470out_up_killsb:
1471 up_read(&fc->killsb);
1472out_finish:
1473 fuse_copy_finish(cs);
1474 return err;
1475}
1476
1477static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1478{
1479 release_pages(req->pages, req->num_pages, 0);
1480}
1481
1482static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1483 struct fuse_notify_retrieve_out *outarg)
1484{
1485 int err;
1486 struct address_space *mapping = inode->i_mapping;
1487 struct fuse_req *req;
1488 pgoff_t index;
1489 loff_t file_size;
1490 unsigned int num;
1491 unsigned int offset;
1492 size_t total_len = 0;
1493
1494 req = fuse_get_req(fc);
1495 if (IS_ERR(req))
1496 return PTR_ERR(req);
1497
1498 offset = outarg->offset & ~PAGE_CACHE_MASK;
1499
1500 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1501 req->in.h.nodeid = outarg->nodeid;
1502 req->in.numargs = 2;
1503 req->in.argpages = 1;
1504 req->page_offset = offset;
1505 req->end = fuse_retrieve_end;
1506
1507 index = outarg->offset >> PAGE_CACHE_SHIFT;
1508 file_size = i_size_read(inode);
1509 num = outarg->size;
1510 if (outarg->offset > file_size)
1511 num = 0;
1512 else if (outarg->offset + num > file_size)
1513 num = file_size - outarg->offset;
1514
1515 while (num) {
1516 struct page *page;
1517 unsigned int this_num;
1518
1519 page = find_get_page(mapping, index);
1520 if (!page)
1521 break;
1522
1523 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1524 req->pages[req->num_pages] = page;
1525 req->num_pages++;
1526
1527 num -= this_num;
1528 total_len += this_num;
1529 }
1530 req->misc.retrieve_in.offset = outarg->offset;
1531 req->misc.retrieve_in.size = total_len;
1532 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1533 req->in.args[0].value = &req->misc.retrieve_in;
1534 req->in.args[1].size = total_len;
1535
1536 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1537 if (err)
1538 fuse_retrieve_end(fc, req);
1539
1540 return err;
1541}
1542
1543static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1544 struct fuse_copy_state *cs)
1545{
1546 struct fuse_notify_retrieve_out outarg;
1547 struct inode *inode;
1548 int err;
1549
1550 err = -EINVAL;
1551 if (size != sizeof(outarg))
1552 goto copy_finish;
1553
1554 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1555 if (err)
1556 goto copy_finish;
1557
1558 fuse_copy_finish(cs);
1559
1560 down_read(&fc->killsb);
1561 err = -ENOENT;
1562 if (fc->sb) {
1563 u64 nodeid = outarg.nodeid;
1564
1565 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1566 if (inode) {
1567 err = fuse_retrieve(fc, inode, &outarg);
1568 iput(inode);
1569 }
1570 }
1571 up_read(&fc->killsb);
1572
1573 return err;
1574
1575copy_finish:
1576 fuse_copy_finish(cs);
1577 return err;
1578}
1579
1580static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1581 unsigned int size, struct fuse_copy_state *cs)
1582{
1583 switch (code) {
1584 case FUSE_NOTIFY_POLL:
1585 return fuse_notify_poll(fc, size, cs);
1586
1587 case FUSE_NOTIFY_INVAL_INODE:
1588 return fuse_notify_inval_inode(fc, size, cs);
1589
1590 case FUSE_NOTIFY_INVAL_ENTRY:
1591 return fuse_notify_inval_entry(fc, size, cs);
1592
1593 case FUSE_NOTIFY_STORE:
1594 return fuse_notify_store(fc, size, cs);
1595
1596 case FUSE_NOTIFY_RETRIEVE:
1597 return fuse_notify_retrieve(fc, size, cs);
1598
1599 default:
1600 fuse_copy_finish(cs);
1601 return -EINVAL;
1602 }
1603}
1604
1605/* Look up request on processing list by unique ID */
1606static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1607{
1608 struct list_head *entry;
1609
1610 list_for_each(entry, &fc->processing) {
1611 struct fuse_req *req;
1612 req = list_entry(entry, struct fuse_req, list);
1613 if (req->in.h.unique == unique || req->intr_unique == unique)
1614 return req;
1615 }
1616 return NULL;
1617}
1618
1619static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1620 unsigned nbytes)
1621{
1622 unsigned reqsize = sizeof(struct fuse_out_header);
1623
1624 if (out->h.error)
1625 return nbytes != reqsize ? -EINVAL : 0;
1626
1627 reqsize += len_args(out->numargs, out->args);
1628
1629 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1630 return -EINVAL;
1631 else if (reqsize > nbytes) {
1632 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1633 unsigned diffsize = reqsize - nbytes;
1634 if (diffsize > lastarg->size)
1635 return -EINVAL;
1636 lastarg->size -= diffsize;
1637 }
1638 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1639 out->page_zeroing);
1640}
1641
1642/*
1643 * Write a single reply to a request. First the header is copied from
1644 * the write buffer. The request is then searched on the processing
1645 * list by the unique ID found in the header. If found, then remove
1646 * it from the list and copy the rest of the buffer to the request.
1647 * The request is finished by calling request_end()
1648 */
1649static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1650 struct fuse_copy_state *cs, size_t nbytes)
1651{
1652 int err;
1653 struct fuse_req *req;
1654 struct fuse_out_header oh;
1655
1656 if (nbytes < sizeof(struct fuse_out_header))
1657 return -EINVAL;
1658
1659 err = fuse_copy_one(cs, &oh, sizeof(oh));
1660 if (err)
1661 goto err_finish;
1662
1663 err = -EINVAL;
1664 if (oh.len != nbytes)
1665 goto err_finish;
1666
1667 /*
1668 * Zero oh.unique indicates unsolicited notification message
1669 * and error contains notification code.
1670 */
1671 if (!oh.unique) {
1672 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1673 return err ? err : nbytes;
1674 }
1675
1676 err = -EINVAL;
1677 if (oh.error <= -1000 || oh.error > 0)
1678 goto err_finish;
1679
1680 spin_lock(&fc->lock);
1681 err = -ENOENT;
1682 if (!fc->connected)
1683 goto err_unlock;
1684
1685 req = request_find(fc, oh.unique);
1686 if (!req)
1687 goto err_unlock;
1688
1689 if (req->aborted) {
1690 spin_unlock(&fc->lock);
1691 fuse_copy_finish(cs);
1692 spin_lock(&fc->lock);
1693 request_end(fc, req);
1694 return -ENOENT;
1695 }
1696 /* Is it an interrupt reply? */
1697 if (req->intr_unique == oh.unique) {
1698 err = -EINVAL;
1699 if (nbytes != sizeof(struct fuse_out_header))
1700 goto err_unlock;
1701
1702 if (oh.error == -ENOSYS)
1703 fc->no_interrupt = 1;
1704 else if (oh.error == -EAGAIN)
1705 queue_interrupt(fc, req);
1706
1707 spin_unlock(&fc->lock);
1708 fuse_copy_finish(cs);
1709 return nbytes;
1710 }
1711
1712 req->state = FUSE_REQ_WRITING;
1713 list_move(&req->list, &fc->io);
1714 req->out.h = oh;
1715 req->locked = 1;
1716 cs->req = req;
1717 if (!req->out.page_replace)
1718 cs->move_pages = 0;
1719 spin_unlock(&fc->lock);
1720
1721 err = copy_out_args(cs, &req->out, nbytes);
1722 fuse_copy_finish(cs);
1723
1724 spin_lock(&fc->lock);
1725 req->locked = 0;
1726 if (!err) {
1727 if (req->aborted)
1728 err = -ENOENT;
1729 } else if (!req->aborted)
1730 req->out.h.error = -EIO;
1731 request_end(fc, req);
1732
1733 return err ? err : nbytes;
1734
1735 err_unlock:
1736 spin_unlock(&fc->lock);
1737 err_finish:
1738 fuse_copy_finish(cs);
1739 return err;
1740}
1741
1742static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1743 unsigned long nr_segs, loff_t pos)
1744{
1745 struct fuse_copy_state cs;
1746 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1747 if (!fc)
1748 return -EPERM;
1749
1750 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
1751
1752 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1753}
1754
1755static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1756 struct file *out, loff_t *ppos,
1757 size_t len, unsigned int flags)
1758{
1759 unsigned nbuf;
1760 unsigned idx;
1761 struct pipe_buffer *bufs;
1762 struct fuse_copy_state cs;
1763 struct fuse_conn *fc;
1764 size_t rem;
1765 ssize_t ret;
1766
1767 fc = fuse_get_conn(out);
1768 if (!fc)
1769 return -EPERM;
1770
1771 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1772 if (!bufs)
1773 return -ENOMEM;
1774
1775 pipe_lock(pipe);
1776 nbuf = 0;
1777 rem = 0;
1778 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1779 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1780
1781 ret = -EINVAL;
1782 if (rem < len) {
1783 pipe_unlock(pipe);
1784 goto out;
1785 }
1786
1787 rem = len;
1788 while (rem) {
1789 struct pipe_buffer *ibuf;
1790 struct pipe_buffer *obuf;
1791
1792 BUG_ON(nbuf >= pipe->buffers);
1793 BUG_ON(!pipe->nrbufs);
1794 ibuf = &pipe->bufs[pipe->curbuf];
1795 obuf = &bufs[nbuf];
1796
1797 if (rem >= ibuf->len) {
1798 *obuf = *ibuf;
1799 ibuf->ops = NULL;
1800 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1801 pipe->nrbufs--;
1802 } else {
1803 ibuf->ops->get(pipe, ibuf);
1804 *obuf = *ibuf;
1805 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1806 obuf->len = rem;
1807 ibuf->offset += obuf->len;
1808 ibuf->len -= obuf->len;
1809 }
1810 nbuf++;
1811 rem -= obuf->len;
1812 }
1813 pipe_unlock(pipe);
1814
1815 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
1816 cs.pipebufs = bufs;
1817 cs.pipe = pipe;
1818
1819 if (flags & SPLICE_F_MOVE)
1820 cs.move_pages = 1;
1821
1822 ret = fuse_dev_do_write(fc, &cs, len);
1823
1824 for (idx = 0; idx < nbuf; idx++) {
1825 struct pipe_buffer *buf = &bufs[idx];
1826 buf->ops->release(pipe, buf);
1827 }
1828out:
1829 kfree(bufs);
1830 return ret;
1831}
1832
1833static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1834{
1835 unsigned mask = POLLOUT | POLLWRNORM;
1836 struct fuse_conn *fc = fuse_get_conn(file);
1837 if (!fc)
1838 return POLLERR;
1839
1840 poll_wait(file, &fc->waitq, wait);
1841
1842 spin_lock(&fc->lock);
1843 if (!fc->connected)
1844 mask = POLLERR;
1845 else if (request_pending(fc))
1846 mask |= POLLIN | POLLRDNORM;
1847 spin_unlock(&fc->lock);
1848
1849 return mask;
1850}
1851
1852/*
1853 * Abort all requests on the given list (pending or processing)
1854 *
1855 * This function releases and reacquires fc->lock
1856 */
1857static void end_requests(struct fuse_conn *fc, struct list_head *head)
1858__releases(fc->lock)
1859__acquires(fc->lock)
1860{
1861 while (!list_empty(head)) {
1862 struct fuse_req *req;
1863 req = list_entry(head->next, struct fuse_req, list);
1864 req->out.h.error = -ECONNABORTED;
1865 request_end(fc, req);
1866 spin_lock(&fc->lock);
1867 }
1868}
1869
1870/*
1871 * Abort requests under I/O
1872 *
1873 * The requests are set to aborted and finished, and the request
1874 * waiter is woken up. This will make request_wait_answer() wait
1875 * until the request is unlocked and then return.
1876 *
1877 * If the request is asynchronous, then the end function needs to be
1878 * called after waiting for the request to be unlocked (if it was
1879 * locked).
1880 */
1881static void end_io_requests(struct fuse_conn *fc)
1882__releases(fc->lock)
1883__acquires(fc->lock)
1884{
1885 while (!list_empty(&fc->io)) {
1886 struct fuse_req *req =
1887 list_entry(fc->io.next, struct fuse_req, list);
1888 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
1889
1890 req->aborted = 1;
1891 req->out.h.error = -ECONNABORTED;
1892 req->state = FUSE_REQ_FINISHED;
1893 list_del_init(&req->list);
1894 wake_up(&req->waitq);
1895 if (end) {
1896 req->end = NULL;
1897 __fuse_get_request(req);
1898 spin_unlock(&fc->lock);
1899 wait_event(req->waitq, !req->locked);
1900 end(fc, req);
1901 fuse_put_request(fc, req);
1902 spin_lock(&fc->lock);
1903 }
1904 }
1905}
1906
1907static void end_queued_requests(struct fuse_conn *fc)
1908__releases(fc->lock)
1909__acquires(fc->lock)
1910{
1911 fc->max_background = UINT_MAX;
1912 flush_bg_queue(fc);
1913 end_requests(fc, &fc->pending);
1914 end_requests(fc, &fc->processing);
1915 while (forget_pending(fc))
1916 kfree(dequeue_forget(fc, 1, NULL));
1917}
1918
1919static void end_polls(struct fuse_conn *fc)
1920{
1921 struct rb_node *p;
1922
1923 p = rb_first(&fc->polled_files);
1924
1925 while (p) {
1926 struct fuse_file *ff;
1927 ff = rb_entry(p, struct fuse_file, polled_node);
1928 wake_up_interruptible_all(&ff->poll_wait);
1929
1930 p = rb_next(p);
1931 }
1932}
1933
1934/*
1935 * Abort all requests.
1936 *
1937 * Emergency exit in case of a malicious or accidental deadlock, or
1938 * just a hung filesystem.
1939 *
1940 * The same effect is usually achievable through killing the
1941 * filesystem daemon and all users of the filesystem. The exception
1942 * is the combination of an asynchronous request and the tricky
1943 * deadlock (see Documentation/filesystems/fuse.txt).
1944 *
1945 * During the aborting, progression of requests from the pending and
1946 * processing lists onto the io list, and progression of new requests
1947 * onto the pending list is prevented by req->connected being false.
1948 *
1949 * Progression of requests under I/O to the processing list is
1950 * prevented by the req->aborted flag being true for these requests.
1951 * For this reason requests on the io list must be aborted first.
1952 */
1953void fuse_abort_conn(struct fuse_conn *fc)
1954{
1955 spin_lock(&fc->lock);
1956 if (fc->connected) {
1957 fc->connected = 0;
1958 fc->blocked = 0;
1959 end_io_requests(fc);
1960 end_queued_requests(fc);
1961 end_polls(fc);
1962 wake_up_all(&fc->waitq);
1963 wake_up_all(&fc->blocked_waitq);
1964 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
1965 }
1966 spin_unlock(&fc->lock);
1967}
1968EXPORT_SYMBOL_GPL(fuse_abort_conn);
1969
1970int fuse_dev_release(struct inode *inode, struct file *file)
1971{
1972 struct fuse_conn *fc = fuse_get_conn(file);
1973 if (fc) {
1974 spin_lock(&fc->lock);
1975 fc->connected = 0;
1976 fc->blocked = 0;
1977 end_queued_requests(fc);
1978 end_polls(fc);
1979 wake_up_all(&fc->blocked_waitq);
1980 spin_unlock(&fc->lock);
1981 fuse_conn_put(fc);
1982 }
1983
1984 return 0;
1985}
1986EXPORT_SYMBOL_GPL(fuse_dev_release);
1987
1988static int fuse_dev_fasync(int fd, struct file *file, int on)
1989{
1990 struct fuse_conn *fc = fuse_get_conn(file);
1991 if (!fc)
1992 return -EPERM;
1993
1994 /* No locking - fasync_helper does its own locking */
1995 return fasync_helper(fd, file, on, &fc->fasync);
1996}
1997
1998const struct file_operations fuse_dev_operations = {
1999 .owner = THIS_MODULE,
2000 .llseek = no_llseek,
2001 .read = do_sync_read,
2002 .aio_read = fuse_dev_read,
2003 .splice_read = fuse_dev_splice_read,
2004 .write = do_sync_write,
2005 .aio_write = fuse_dev_write,
2006 .splice_write = fuse_dev_splice_write,
2007 .poll = fuse_dev_poll,
2008 .release = fuse_dev_release,
2009 .fasync = fuse_dev_fasync,
2010};
2011EXPORT_SYMBOL_GPL(fuse_dev_operations);
2012
2013static struct miscdevice fuse_miscdevice = {
2014 .minor = FUSE_MINOR,
2015 .name = "fuse",
2016 .fops = &fuse_dev_operations,
2017};
2018
2019int __init fuse_dev_init(void)
2020{
2021 int err = -ENOMEM;
2022 fuse_req_cachep = kmem_cache_create("fuse_request",
2023 sizeof(struct fuse_req),
2024 0, 0, NULL);
2025 if (!fuse_req_cachep)
2026 goto out;
2027
2028 err = misc_register(&fuse_miscdevice);
2029 if (err)
2030 goto out_cache_clean;
2031
2032 return 0;
2033
2034 out_cache_clean:
2035 kmem_cache_destroy(fuse_req_cachep);
2036 out:
2037 return err;
2038}
2039
2040void fuse_dev_cleanup(void)
2041{
2042 misc_deregister(&fuse_miscdevice);
2043 kmem_cache_destroy(fuse_req_cachep);
2044}