xref: /linux/fs/fuse/dev.c (revision 5bbfea1eacdf584d0d159e38c01ee190162706d9)
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4 
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8 
9 #include "fuse_i.h"
10 
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/sched/signal.h>
15 #include <linux/uio.h>
16 #include <linux/miscdevice.h>
17 #include <linux/pagemap.h>
18 #include <linux/file.h>
19 #include <linux/slab.h>
20 #include <linux/pipe_fs_i.h>
21 #include <linux/swap.h>
22 #include <linux/splice.h>
23 #include <linux/sched.h>
24 
25 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26 MODULE_ALIAS("devname:fuse");
27 
28 /* 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 
32 static struct kmem_cache *fuse_req_cachep;
33 
34 static 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 
43 static 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 
53 static 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 
62 static void fuse_request_free(struct fuse_req *req)
63 {
64 	kmem_cache_free(fuse_req_cachep, req);
65 }
66 
67 static void __fuse_get_request(struct fuse_req *req)
68 {
69 	refcount_inc(&req->count);
70 }
71 
72 /* Must be called with > 1 refcount */
73 static void __fuse_put_request(struct fuse_req *req)
74 {
75 	refcount_dec(&req->count);
76 }
77 
78 void 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 
85 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
86 {
87 	return !fc->initialized || (for_background && fc->blocked);
88 }
89 
90 static 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 
104 static void fuse_put_request(struct fuse_req *req);
105 
106 static 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 
158 static 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 
183 unsigned 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 }
193 EXPORT_SYMBOL_GPL(fuse_len_args);
194 
195 u64 fuse_get_unique(struct fuse_iqueue *fiq)
196 {
197 	fiq->reqctr += FUSE_REQ_ID_STEP;
198 	return fiq->reqctr;
199 }
200 EXPORT_SYMBOL_GPL(fuse_get_unique);
201 
202 static 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  */
210 static 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 
218 const 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 };
223 EXPORT_SYMBOL_GPL(fuse_dev_fiq_ops);
224 
225 static 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 
236 void 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 
255 static 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  */
280 void 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 		if (fc->num_background == fc->congestion_threshold && fm->sb) {
319 			clear_bdi_congested(fm->sb->s_bdi, BLK_RW_SYNC);
320 			clear_bdi_congested(fm->sb->s_bdi, BLK_RW_ASYNC);
321 		}
322 		fc->num_background--;
323 		fc->active_background--;
324 		flush_bg_queue(fc);
325 		spin_unlock(&fc->bg_lock);
326 	} else {
327 		/* Wake up waiter sleeping in request_wait_answer() */
328 		wake_up(&req->waitq);
329 	}
330 
331 	if (test_bit(FR_ASYNC, &req->flags))
332 		req->args->end(fm, req->args, req->out.h.error);
333 put_request:
334 	fuse_put_request(req);
335 }
336 EXPORT_SYMBOL_GPL(fuse_request_end);
337 
338 static int queue_interrupt(struct fuse_req *req)
339 {
340 	struct fuse_iqueue *fiq = &req->fm->fc->iq;
341 
342 	spin_lock(&fiq->lock);
343 	/* Check for we've sent request to interrupt this req */
344 	if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) {
345 		spin_unlock(&fiq->lock);
346 		return -EINVAL;
347 	}
348 
349 	if (list_empty(&req->intr_entry)) {
350 		list_add_tail(&req->intr_entry, &fiq->interrupts);
351 		/*
352 		 * Pairs with smp_mb() implied by test_and_set_bit()
353 		 * from fuse_request_end().
354 		 */
355 		smp_mb();
356 		if (test_bit(FR_FINISHED, &req->flags)) {
357 			list_del_init(&req->intr_entry);
358 			spin_unlock(&fiq->lock);
359 			return 0;
360 		}
361 		fiq->ops->wake_interrupt_and_unlock(fiq);
362 	} else {
363 		spin_unlock(&fiq->lock);
364 	}
365 	return 0;
366 }
367 
368 static void request_wait_answer(struct fuse_req *req)
369 {
370 	struct fuse_conn *fc = req->fm->fc;
371 	struct fuse_iqueue *fiq = &fc->iq;
372 	int err;
373 
374 	if (!fc->no_interrupt) {
375 		/* Any signal may interrupt this */
376 		err = wait_event_interruptible(req->waitq,
377 					test_bit(FR_FINISHED, &req->flags));
378 		if (!err)
379 			return;
380 
381 		set_bit(FR_INTERRUPTED, &req->flags);
382 		/* matches barrier in fuse_dev_do_read() */
383 		smp_mb__after_atomic();
384 		if (test_bit(FR_SENT, &req->flags))
385 			queue_interrupt(req);
386 	}
387 
388 	if (!test_bit(FR_FORCE, &req->flags)) {
389 		/* Only fatal signals may interrupt this */
390 		err = wait_event_killable(req->waitq,
391 					test_bit(FR_FINISHED, &req->flags));
392 		if (!err)
393 			return;
394 
395 		spin_lock(&fiq->lock);
396 		/* Request is not yet in userspace, bail out */
397 		if (test_bit(FR_PENDING, &req->flags)) {
398 			list_del(&req->list);
399 			spin_unlock(&fiq->lock);
400 			__fuse_put_request(req);
401 			req->out.h.error = -EINTR;
402 			return;
403 		}
404 		spin_unlock(&fiq->lock);
405 	}
406 
407 	/*
408 	 * Either request is already in userspace, or it was forced.
409 	 * Wait it out.
410 	 */
411 	wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
412 }
413 
414 static void __fuse_request_send(struct fuse_req *req)
415 {
416 	struct fuse_iqueue *fiq = &req->fm->fc->iq;
417 
418 	BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
419 	spin_lock(&fiq->lock);
420 	if (!fiq->connected) {
421 		spin_unlock(&fiq->lock);
422 		req->out.h.error = -ENOTCONN;
423 	} else {
424 		req->in.h.unique = fuse_get_unique(fiq);
425 		/* acquire extra reference, since request is still needed
426 		   after fuse_request_end() */
427 		__fuse_get_request(req);
428 		queue_request_and_unlock(fiq, req);
429 
430 		request_wait_answer(req);
431 		/* Pairs with smp_wmb() in fuse_request_end() */
432 		smp_rmb();
433 	}
434 }
435 
436 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
437 {
438 	if (fc->minor < 4 && args->opcode == FUSE_STATFS)
439 		args->out_args[0].size = FUSE_COMPAT_STATFS_SIZE;
440 
441 	if (fc->minor < 9) {
442 		switch (args->opcode) {
443 		case FUSE_LOOKUP:
444 		case FUSE_CREATE:
445 		case FUSE_MKNOD:
446 		case FUSE_MKDIR:
447 		case FUSE_SYMLINK:
448 		case FUSE_LINK:
449 			args->out_args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
450 			break;
451 		case FUSE_GETATTR:
452 		case FUSE_SETATTR:
453 			args->out_args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
454 			break;
455 		}
456 	}
457 	if (fc->minor < 12) {
458 		switch (args->opcode) {
459 		case FUSE_CREATE:
460 			args->in_args[0].size = sizeof(struct fuse_open_in);
461 			break;
462 		case FUSE_MKNOD:
463 			args->in_args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
464 			break;
465 		}
466 	}
467 }
468 
469 static void fuse_force_creds(struct fuse_req *req)
470 {
471 	struct fuse_conn *fc = req->fm->fc;
472 
473 	req->in.h.uid = from_kuid_munged(fc->user_ns, current_fsuid());
474 	req->in.h.gid = from_kgid_munged(fc->user_ns, current_fsgid());
475 	req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
476 }
477 
478 static void fuse_args_to_req(struct fuse_req *req, struct fuse_args *args)
479 {
480 	req->in.h.opcode = args->opcode;
481 	req->in.h.nodeid = args->nodeid;
482 	req->args = args;
483 	if (args->end)
484 		__set_bit(FR_ASYNC, &req->flags);
485 }
486 
487 ssize_t fuse_simple_request(struct fuse_mount *fm, struct fuse_args *args)
488 {
489 	struct fuse_conn *fc = fm->fc;
490 	struct fuse_req *req;
491 	ssize_t ret;
492 
493 	if (args->force) {
494 		atomic_inc(&fc->num_waiting);
495 		req = fuse_request_alloc(fm, GFP_KERNEL | __GFP_NOFAIL);
496 
497 		if (!args->nocreds)
498 			fuse_force_creds(req);
499 
500 		__set_bit(FR_WAITING, &req->flags);
501 		__set_bit(FR_FORCE, &req->flags);
502 	} else {
503 		WARN_ON(args->nocreds);
504 		req = fuse_get_req(fm, false);
505 		if (IS_ERR(req))
506 			return PTR_ERR(req);
507 	}
508 
509 	/* Needs to be done after fuse_get_req() so that fc->minor is valid */
510 	fuse_adjust_compat(fc, args);
511 	fuse_args_to_req(req, args);
512 
513 	if (!args->noreply)
514 		__set_bit(FR_ISREPLY, &req->flags);
515 	__fuse_request_send(req);
516 	ret = req->out.h.error;
517 	if (!ret && args->out_argvar) {
518 		BUG_ON(args->out_numargs == 0);
519 		ret = args->out_args[args->out_numargs - 1].size;
520 	}
521 	fuse_put_request(req);
522 
523 	return ret;
524 }
525 
526 static bool fuse_request_queue_background(struct fuse_req *req)
527 {
528 	struct fuse_mount *fm = req->fm;
529 	struct fuse_conn *fc = fm->fc;
530 	bool queued = false;
531 
532 	WARN_ON(!test_bit(FR_BACKGROUND, &req->flags));
533 	if (!test_bit(FR_WAITING, &req->flags)) {
534 		__set_bit(FR_WAITING, &req->flags);
535 		atomic_inc(&fc->num_waiting);
536 	}
537 	__set_bit(FR_ISREPLY, &req->flags);
538 	spin_lock(&fc->bg_lock);
539 	if (likely(fc->connected)) {
540 		fc->num_background++;
541 		if (fc->num_background == fc->max_background)
542 			fc->blocked = 1;
543 		if (fc->num_background == fc->congestion_threshold && fm->sb) {
544 			set_bdi_congested(fm->sb->s_bdi, BLK_RW_SYNC);
545 			set_bdi_congested(fm->sb->s_bdi, BLK_RW_ASYNC);
546 		}
547 		list_add_tail(&req->list, &fc->bg_queue);
548 		flush_bg_queue(fc);
549 		queued = true;
550 	}
551 	spin_unlock(&fc->bg_lock);
552 
553 	return queued;
554 }
555 
556 int fuse_simple_background(struct fuse_mount *fm, struct fuse_args *args,
557 			    gfp_t gfp_flags)
558 {
559 	struct fuse_req *req;
560 
561 	if (args->force) {
562 		WARN_ON(!args->nocreds);
563 		req = fuse_request_alloc(fm, gfp_flags);
564 		if (!req)
565 			return -ENOMEM;
566 		__set_bit(FR_BACKGROUND, &req->flags);
567 	} else {
568 		WARN_ON(args->nocreds);
569 		req = fuse_get_req(fm, true);
570 		if (IS_ERR(req))
571 			return PTR_ERR(req);
572 	}
573 
574 	fuse_args_to_req(req, args);
575 
576 	if (!fuse_request_queue_background(req)) {
577 		fuse_put_request(req);
578 		return -ENOTCONN;
579 	}
580 
581 	return 0;
582 }
583 EXPORT_SYMBOL_GPL(fuse_simple_background);
584 
585 static int fuse_simple_notify_reply(struct fuse_mount *fm,
586 				    struct fuse_args *args, u64 unique)
587 {
588 	struct fuse_req *req;
589 	struct fuse_iqueue *fiq = &fm->fc->iq;
590 	int err = 0;
591 
592 	req = fuse_get_req(fm, false);
593 	if (IS_ERR(req))
594 		return PTR_ERR(req);
595 
596 	__clear_bit(FR_ISREPLY, &req->flags);
597 	req->in.h.unique = unique;
598 
599 	fuse_args_to_req(req, args);
600 
601 	spin_lock(&fiq->lock);
602 	if (fiq->connected) {
603 		queue_request_and_unlock(fiq, req);
604 	} else {
605 		err = -ENODEV;
606 		spin_unlock(&fiq->lock);
607 		fuse_put_request(req);
608 	}
609 
610 	return err;
611 }
612 
613 /*
614  * Lock the request.  Up to the next unlock_request() there mustn't be
615  * anything that could cause a page-fault.  If the request was already
616  * aborted bail out.
617  */
618 static int lock_request(struct fuse_req *req)
619 {
620 	int err = 0;
621 	if (req) {
622 		spin_lock(&req->waitq.lock);
623 		if (test_bit(FR_ABORTED, &req->flags))
624 			err = -ENOENT;
625 		else
626 			set_bit(FR_LOCKED, &req->flags);
627 		spin_unlock(&req->waitq.lock);
628 	}
629 	return err;
630 }
631 
632 /*
633  * Unlock request.  If it was aborted while locked, caller is responsible
634  * for unlocking and ending the request.
635  */
636 static int unlock_request(struct fuse_req *req)
637 {
638 	int err = 0;
639 	if (req) {
640 		spin_lock(&req->waitq.lock);
641 		if (test_bit(FR_ABORTED, &req->flags))
642 			err = -ENOENT;
643 		else
644 			clear_bit(FR_LOCKED, &req->flags);
645 		spin_unlock(&req->waitq.lock);
646 	}
647 	return err;
648 }
649 
650 struct fuse_copy_state {
651 	int write;
652 	struct fuse_req *req;
653 	struct iov_iter *iter;
654 	struct pipe_buffer *pipebufs;
655 	struct pipe_buffer *currbuf;
656 	struct pipe_inode_info *pipe;
657 	unsigned long nr_segs;
658 	struct page *pg;
659 	unsigned len;
660 	unsigned offset;
661 	unsigned move_pages:1;
662 };
663 
664 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
665 			   struct iov_iter *iter)
666 {
667 	memset(cs, 0, sizeof(*cs));
668 	cs->write = write;
669 	cs->iter = iter;
670 }
671 
672 /* Unmap and put previous page of userspace buffer */
673 static void fuse_copy_finish(struct fuse_copy_state *cs)
674 {
675 	if (cs->currbuf) {
676 		struct pipe_buffer *buf = cs->currbuf;
677 
678 		if (cs->write)
679 			buf->len = PAGE_SIZE - cs->len;
680 		cs->currbuf = NULL;
681 	} else if (cs->pg) {
682 		if (cs->write) {
683 			flush_dcache_page(cs->pg);
684 			set_page_dirty_lock(cs->pg);
685 		}
686 		put_page(cs->pg);
687 	}
688 	cs->pg = NULL;
689 }
690 
691 /*
692  * Get another pagefull of userspace buffer, and map it to kernel
693  * address space, and lock request
694  */
695 static int fuse_copy_fill(struct fuse_copy_state *cs)
696 {
697 	struct page *page;
698 	int err;
699 
700 	err = unlock_request(cs->req);
701 	if (err)
702 		return err;
703 
704 	fuse_copy_finish(cs);
705 	if (cs->pipebufs) {
706 		struct pipe_buffer *buf = cs->pipebufs;
707 
708 		if (!cs->write) {
709 			err = pipe_buf_confirm(cs->pipe, buf);
710 			if (err)
711 				return err;
712 
713 			BUG_ON(!cs->nr_segs);
714 			cs->currbuf = buf;
715 			cs->pg = buf->page;
716 			cs->offset = buf->offset;
717 			cs->len = buf->len;
718 			cs->pipebufs++;
719 			cs->nr_segs--;
720 		} else {
721 			if (cs->nr_segs >= cs->pipe->max_usage)
722 				return -EIO;
723 
724 			page = alloc_page(GFP_HIGHUSER);
725 			if (!page)
726 				return -ENOMEM;
727 
728 			buf->page = page;
729 			buf->offset = 0;
730 			buf->len = 0;
731 
732 			cs->currbuf = buf;
733 			cs->pg = page;
734 			cs->offset = 0;
735 			cs->len = PAGE_SIZE;
736 			cs->pipebufs++;
737 			cs->nr_segs++;
738 		}
739 	} else {
740 		size_t off;
741 		err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
742 		if (err < 0)
743 			return err;
744 		BUG_ON(!err);
745 		cs->len = err;
746 		cs->offset = off;
747 		cs->pg = page;
748 		iov_iter_advance(cs->iter, err);
749 	}
750 
751 	return lock_request(cs->req);
752 }
753 
754 /* Do as much copy to/from userspace buffer as we can */
755 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
756 {
757 	unsigned ncpy = min(*size, cs->len);
758 	if (val) {
759 		void *pgaddr = kmap_atomic(cs->pg);
760 		void *buf = pgaddr + cs->offset;
761 
762 		if (cs->write)
763 			memcpy(buf, *val, ncpy);
764 		else
765 			memcpy(*val, buf, ncpy);
766 
767 		kunmap_atomic(pgaddr);
768 		*val += ncpy;
769 	}
770 	*size -= ncpy;
771 	cs->len -= ncpy;
772 	cs->offset += ncpy;
773 	return ncpy;
774 }
775 
776 static int fuse_check_page(struct page *page)
777 {
778 	if (page_mapcount(page) ||
779 	    page->mapping != NULL ||
780 	    (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
781 	     ~(1 << PG_locked |
782 	       1 << PG_referenced |
783 	       1 << PG_uptodate |
784 	       1 << PG_lru |
785 	       1 << PG_active |
786 	       1 << PG_workingset |
787 	       1 << PG_reclaim |
788 	       1 << PG_waiters))) {
789 		dump_page(page, "fuse: trying to steal weird page");
790 		return 1;
791 	}
792 	return 0;
793 }
794 
795 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
796 {
797 	int err;
798 	struct page *oldpage = *pagep;
799 	struct page *newpage;
800 	struct pipe_buffer *buf = cs->pipebufs;
801 
802 	get_page(oldpage);
803 	err = unlock_request(cs->req);
804 	if (err)
805 		goto out_put_old;
806 
807 	fuse_copy_finish(cs);
808 
809 	err = pipe_buf_confirm(cs->pipe, buf);
810 	if (err)
811 		goto out_put_old;
812 
813 	BUG_ON(!cs->nr_segs);
814 	cs->currbuf = buf;
815 	cs->len = buf->len;
816 	cs->pipebufs++;
817 	cs->nr_segs--;
818 
819 	if (cs->len != PAGE_SIZE)
820 		goto out_fallback;
821 
822 	if (!pipe_buf_try_steal(cs->pipe, buf))
823 		goto out_fallback;
824 
825 	newpage = buf->page;
826 
827 	if (!PageUptodate(newpage))
828 		SetPageUptodate(newpage);
829 
830 	ClearPageMappedToDisk(newpage);
831 
832 	if (fuse_check_page(newpage) != 0)
833 		goto out_fallback_unlock;
834 
835 	/*
836 	 * This is a new and locked page, it shouldn't be mapped or
837 	 * have any special flags on it
838 	 */
839 	if (WARN_ON(page_mapped(oldpage)))
840 		goto out_fallback_unlock;
841 	if (WARN_ON(page_has_private(oldpage)))
842 		goto out_fallback_unlock;
843 	if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
844 		goto out_fallback_unlock;
845 	if (WARN_ON(PageMlocked(oldpage)))
846 		goto out_fallback_unlock;
847 
848 	replace_page_cache_page(oldpage, newpage);
849 
850 	get_page(newpage);
851 
852 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
853 		lru_cache_add(newpage);
854 
855 	err = 0;
856 	spin_lock(&cs->req->waitq.lock);
857 	if (test_bit(FR_ABORTED, &cs->req->flags))
858 		err = -ENOENT;
859 	else
860 		*pagep = newpage;
861 	spin_unlock(&cs->req->waitq.lock);
862 
863 	if (err) {
864 		unlock_page(newpage);
865 		put_page(newpage);
866 		goto out_put_old;
867 	}
868 
869 	unlock_page(oldpage);
870 	/* Drop ref for ap->pages[] array */
871 	put_page(oldpage);
872 	cs->len = 0;
873 
874 	err = 0;
875 out_put_old:
876 	/* Drop ref obtained in this function */
877 	put_page(oldpage);
878 	return err;
879 
880 out_fallback_unlock:
881 	unlock_page(newpage);
882 out_fallback:
883 	cs->pg = buf->page;
884 	cs->offset = buf->offset;
885 
886 	err = lock_request(cs->req);
887 	if (!err)
888 		err = 1;
889 
890 	goto out_put_old;
891 }
892 
893 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
894 			 unsigned offset, unsigned count)
895 {
896 	struct pipe_buffer *buf;
897 	int err;
898 
899 	if (cs->nr_segs >= cs->pipe->max_usage)
900 		return -EIO;
901 
902 	get_page(page);
903 	err = unlock_request(cs->req);
904 	if (err) {
905 		put_page(page);
906 		return err;
907 	}
908 
909 	fuse_copy_finish(cs);
910 
911 	buf = cs->pipebufs;
912 	buf->page = page;
913 	buf->offset = offset;
914 	buf->len = count;
915 
916 	cs->pipebufs++;
917 	cs->nr_segs++;
918 	cs->len = 0;
919 
920 	return 0;
921 }
922 
923 /*
924  * Copy a page in the request to/from the userspace buffer.  Must be
925  * done atomically
926  */
927 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
928 			  unsigned offset, unsigned count, int zeroing)
929 {
930 	int err;
931 	struct page *page = *pagep;
932 
933 	if (page && zeroing && count < PAGE_SIZE)
934 		clear_highpage(page);
935 
936 	while (count) {
937 		if (cs->write && cs->pipebufs && page) {
938 			return fuse_ref_page(cs, page, offset, count);
939 		} else if (!cs->len) {
940 			if (cs->move_pages && page &&
941 			    offset == 0 && count == PAGE_SIZE) {
942 				err = fuse_try_move_page(cs, pagep);
943 				if (err <= 0)
944 					return err;
945 			} else {
946 				err = fuse_copy_fill(cs);
947 				if (err)
948 					return err;
949 			}
950 		}
951 		if (page) {
952 			void *mapaddr = kmap_atomic(page);
953 			void *buf = mapaddr + offset;
954 			offset += fuse_copy_do(cs, &buf, &count);
955 			kunmap_atomic(mapaddr);
956 		} else
957 			offset += fuse_copy_do(cs, NULL, &count);
958 	}
959 	if (page && !cs->write)
960 		flush_dcache_page(page);
961 	return 0;
962 }
963 
964 /* Copy pages in the request to/from userspace buffer */
965 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
966 			   int zeroing)
967 {
968 	unsigned i;
969 	struct fuse_req *req = cs->req;
970 	struct fuse_args_pages *ap = container_of(req->args, typeof(*ap), args);
971 
972 
973 	for (i = 0; i < ap->num_pages && (nbytes || zeroing); i++) {
974 		int err;
975 		unsigned int offset = ap->descs[i].offset;
976 		unsigned int count = min(nbytes, ap->descs[i].length);
977 
978 		err = fuse_copy_page(cs, &ap->pages[i], offset, count, zeroing);
979 		if (err)
980 			return err;
981 
982 		nbytes -= count;
983 	}
984 	return 0;
985 }
986 
987 /* Copy a single argument in the request to/from userspace buffer */
988 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
989 {
990 	while (size) {
991 		if (!cs->len) {
992 			int err = fuse_copy_fill(cs);
993 			if (err)
994 				return err;
995 		}
996 		fuse_copy_do(cs, &val, &size);
997 	}
998 	return 0;
999 }
1000 
1001 /* Copy request arguments to/from userspace buffer */
1002 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1003 			  unsigned argpages, struct fuse_arg *args,
1004 			  int zeroing)
1005 {
1006 	int err = 0;
1007 	unsigned i;
1008 
1009 	for (i = 0; !err && i < numargs; i++)  {
1010 		struct fuse_arg *arg = &args[i];
1011 		if (i == numargs - 1 && argpages)
1012 			err = fuse_copy_pages(cs, arg->size, zeroing);
1013 		else
1014 			err = fuse_copy_one(cs, arg->value, arg->size);
1015 	}
1016 	return err;
1017 }
1018 
1019 static int forget_pending(struct fuse_iqueue *fiq)
1020 {
1021 	return fiq->forget_list_head.next != NULL;
1022 }
1023 
1024 static int request_pending(struct fuse_iqueue *fiq)
1025 {
1026 	return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1027 		forget_pending(fiq);
1028 }
1029 
1030 /*
1031  * Transfer an interrupt request to userspace
1032  *
1033  * Unlike other requests this is assembled on demand, without a need
1034  * to allocate a separate fuse_req structure.
1035  *
1036  * Called with fiq->lock held, releases it
1037  */
1038 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1039 			       struct fuse_copy_state *cs,
1040 			       size_t nbytes, struct fuse_req *req)
1041 __releases(fiq->lock)
1042 {
1043 	struct fuse_in_header ih;
1044 	struct fuse_interrupt_in arg;
1045 	unsigned reqsize = sizeof(ih) + sizeof(arg);
1046 	int err;
1047 
1048 	list_del_init(&req->intr_entry);
1049 	memset(&ih, 0, sizeof(ih));
1050 	memset(&arg, 0, sizeof(arg));
1051 	ih.len = reqsize;
1052 	ih.opcode = FUSE_INTERRUPT;
1053 	ih.unique = (req->in.h.unique | FUSE_INT_REQ_BIT);
1054 	arg.unique = req->in.h.unique;
1055 
1056 	spin_unlock(&fiq->lock);
1057 	if (nbytes < reqsize)
1058 		return -EINVAL;
1059 
1060 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1061 	if (!err)
1062 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1063 	fuse_copy_finish(cs);
1064 
1065 	return err ? err : reqsize;
1066 }
1067 
1068 struct fuse_forget_link *fuse_dequeue_forget(struct fuse_iqueue *fiq,
1069 					     unsigned int max,
1070 					     unsigned int *countp)
1071 {
1072 	struct fuse_forget_link *head = fiq->forget_list_head.next;
1073 	struct fuse_forget_link **newhead = &head;
1074 	unsigned count;
1075 
1076 	for (count = 0; *newhead != NULL && count < max; count++)
1077 		newhead = &(*newhead)->next;
1078 
1079 	fiq->forget_list_head.next = *newhead;
1080 	*newhead = NULL;
1081 	if (fiq->forget_list_head.next == NULL)
1082 		fiq->forget_list_tail = &fiq->forget_list_head;
1083 
1084 	if (countp != NULL)
1085 		*countp = count;
1086 
1087 	return head;
1088 }
1089 EXPORT_SYMBOL(fuse_dequeue_forget);
1090 
1091 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1092 				   struct fuse_copy_state *cs,
1093 				   size_t nbytes)
1094 __releases(fiq->lock)
1095 {
1096 	int err;
1097 	struct fuse_forget_link *forget = fuse_dequeue_forget(fiq, 1, NULL);
1098 	struct fuse_forget_in arg = {
1099 		.nlookup = forget->forget_one.nlookup,
1100 	};
1101 	struct fuse_in_header ih = {
1102 		.opcode = FUSE_FORGET,
1103 		.nodeid = forget->forget_one.nodeid,
1104 		.unique = fuse_get_unique(fiq),
1105 		.len = sizeof(ih) + sizeof(arg),
1106 	};
1107 
1108 	spin_unlock(&fiq->lock);
1109 	kfree(forget);
1110 	if (nbytes < ih.len)
1111 		return -EINVAL;
1112 
1113 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1114 	if (!err)
1115 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1116 	fuse_copy_finish(cs);
1117 
1118 	if (err)
1119 		return err;
1120 
1121 	return ih.len;
1122 }
1123 
1124 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1125 				   struct fuse_copy_state *cs, size_t nbytes)
1126 __releases(fiq->lock)
1127 {
1128 	int err;
1129 	unsigned max_forgets;
1130 	unsigned count;
1131 	struct fuse_forget_link *head;
1132 	struct fuse_batch_forget_in arg = { .count = 0 };
1133 	struct fuse_in_header ih = {
1134 		.opcode = FUSE_BATCH_FORGET,
1135 		.unique = fuse_get_unique(fiq),
1136 		.len = sizeof(ih) + sizeof(arg),
1137 	};
1138 
1139 	if (nbytes < ih.len) {
1140 		spin_unlock(&fiq->lock);
1141 		return -EINVAL;
1142 	}
1143 
1144 	max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1145 	head = fuse_dequeue_forget(fiq, max_forgets, &count);
1146 	spin_unlock(&fiq->lock);
1147 
1148 	arg.count = count;
1149 	ih.len += count * sizeof(struct fuse_forget_one);
1150 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1151 	if (!err)
1152 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1153 
1154 	while (head) {
1155 		struct fuse_forget_link *forget = head;
1156 
1157 		if (!err) {
1158 			err = fuse_copy_one(cs, &forget->forget_one,
1159 					    sizeof(forget->forget_one));
1160 		}
1161 		head = forget->next;
1162 		kfree(forget);
1163 	}
1164 
1165 	fuse_copy_finish(cs);
1166 
1167 	if (err)
1168 		return err;
1169 
1170 	return ih.len;
1171 }
1172 
1173 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1174 			    struct fuse_copy_state *cs,
1175 			    size_t nbytes)
1176 __releases(fiq->lock)
1177 {
1178 	if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1179 		return fuse_read_single_forget(fiq, cs, nbytes);
1180 	else
1181 		return fuse_read_batch_forget(fiq, cs, nbytes);
1182 }
1183 
1184 /*
1185  * Read a single request into the userspace filesystem's buffer.  This
1186  * function waits until a request is available, then removes it from
1187  * the pending list and copies request data to userspace buffer.  If
1188  * no reply is needed (FORGET) or request has been aborted or there
1189  * was an error during the copying then it's finished by calling
1190  * fuse_request_end().  Otherwise add it to the processing list, and set
1191  * the 'sent' flag.
1192  */
1193 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1194 				struct fuse_copy_state *cs, size_t nbytes)
1195 {
1196 	ssize_t err;
1197 	struct fuse_conn *fc = fud->fc;
1198 	struct fuse_iqueue *fiq = &fc->iq;
1199 	struct fuse_pqueue *fpq = &fud->pq;
1200 	struct fuse_req *req;
1201 	struct fuse_args *args;
1202 	unsigned reqsize;
1203 	unsigned int hash;
1204 
1205 	/*
1206 	 * Require sane minimum read buffer - that has capacity for fixed part
1207 	 * of any request header + negotiated max_write room for data.
1208 	 *
1209 	 * Historically libfuse reserves 4K for fixed header room, but e.g.
1210 	 * GlusterFS reserves only 80 bytes
1211 	 *
1212 	 *	= `sizeof(fuse_in_header) + sizeof(fuse_write_in)`
1213 	 *
1214 	 * which is the absolute minimum any sane filesystem should be using
1215 	 * for header room.
1216 	 */
1217 	if (nbytes < max_t(size_t, FUSE_MIN_READ_BUFFER,
1218 			   sizeof(struct fuse_in_header) +
1219 			   sizeof(struct fuse_write_in) +
1220 			   fc->max_write))
1221 		return -EINVAL;
1222 
1223  restart:
1224 	for (;;) {
1225 		spin_lock(&fiq->lock);
1226 		if (!fiq->connected || request_pending(fiq))
1227 			break;
1228 		spin_unlock(&fiq->lock);
1229 
1230 		if (file->f_flags & O_NONBLOCK)
1231 			return -EAGAIN;
1232 		err = wait_event_interruptible_exclusive(fiq->waitq,
1233 				!fiq->connected || request_pending(fiq));
1234 		if (err)
1235 			return err;
1236 	}
1237 
1238 	if (!fiq->connected) {
1239 		err = fc->aborted ? -ECONNABORTED : -ENODEV;
1240 		goto err_unlock;
1241 	}
1242 
1243 	if (!list_empty(&fiq->interrupts)) {
1244 		req = list_entry(fiq->interrupts.next, struct fuse_req,
1245 				 intr_entry);
1246 		return fuse_read_interrupt(fiq, cs, nbytes, req);
1247 	}
1248 
1249 	if (forget_pending(fiq)) {
1250 		if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1251 			return fuse_read_forget(fc, fiq, cs, nbytes);
1252 
1253 		if (fiq->forget_batch <= -8)
1254 			fiq->forget_batch = 16;
1255 	}
1256 
1257 	req = list_entry(fiq->pending.next, struct fuse_req, list);
1258 	clear_bit(FR_PENDING, &req->flags);
1259 	list_del_init(&req->list);
1260 	spin_unlock(&fiq->lock);
1261 
1262 	args = req->args;
1263 	reqsize = req->in.h.len;
1264 
1265 	/* If request is too large, reply with an error and restart the read */
1266 	if (nbytes < reqsize) {
1267 		req->out.h.error = -EIO;
1268 		/* SETXATTR is special, since it may contain too large data */
1269 		if (args->opcode == FUSE_SETXATTR)
1270 			req->out.h.error = -E2BIG;
1271 		fuse_request_end(req);
1272 		goto restart;
1273 	}
1274 	spin_lock(&fpq->lock);
1275 	/*
1276 	 *  Must not put request on fpq->io queue after having been shut down by
1277 	 *  fuse_abort_conn()
1278 	 */
1279 	if (!fpq->connected) {
1280 		req->out.h.error = err = -ECONNABORTED;
1281 		goto out_end;
1282 
1283 	}
1284 	list_add(&req->list, &fpq->io);
1285 	spin_unlock(&fpq->lock);
1286 	cs->req = req;
1287 	err = fuse_copy_one(cs, &req->in.h, sizeof(req->in.h));
1288 	if (!err)
1289 		err = fuse_copy_args(cs, args->in_numargs, args->in_pages,
1290 				     (struct fuse_arg *) args->in_args, 0);
1291 	fuse_copy_finish(cs);
1292 	spin_lock(&fpq->lock);
1293 	clear_bit(FR_LOCKED, &req->flags);
1294 	if (!fpq->connected) {
1295 		err = fc->aborted ? -ECONNABORTED : -ENODEV;
1296 		goto out_end;
1297 	}
1298 	if (err) {
1299 		req->out.h.error = -EIO;
1300 		goto out_end;
1301 	}
1302 	if (!test_bit(FR_ISREPLY, &req->flags)) {
1303 		err = reqsize;
1304 		goto out_end;
1305 	}
1306 	hash = fuse_req_hash(req->in.h.unique);
1307 	list_move_tail(&req->list, &fpq->processing[hash]);
1308 	__fuse_get_request(req);
1309 	set_bit(FR_SENT, &req->flags);
1310 	spin_unlock(&fpq->lock);
1311 	/* matches barrier in request_wait_answer() */
1312 	smp_mb__after_atomic();
1313 	if (test_bit(FR_INTERRUPTED, &req->flags))
1314 		queue_interrupt(req);
1315 	fuse_put_request(req);
1316 
1317 	return reqsize;
1318 
1319 out_end:
1320 	if (!test_bit(FR_PRIVATE, &req->flags))
1321 		list_del_init(&req->list);
1322 	spin_unlock(&fpq->lock);
1323 	fuse_request_end(req);
1324 	return err;
1325 
1326  err_unlock:
1327 	spin_unlock(&fiq->lock);
1328 	return err;
1329 }
1330 
1331 static int fuse_dev_open(struct inode *inode, struct file *file)
1332 {
1333 	/*
1334 	 * The fuse device's file's private_data is used to hold
1335 	 * the fuse_conn(ection) when it is mounted, and is used to
1336 	 * keep track of whether the file has been mounted already.
1337 	 */
1338 	file->private_data = NULL;
1339 	return 0;
1340 }
1341 
1342 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1343 {
1344 	struct fuse_copy_state cs;
1345 	struct file *file = iocb->ki_filp;
1346 	struct fuse_dev *fud = fuse_get_dev(file);
1347 
1348 	if (!fud)
1349 		return -EPERM;
1350 
1351 	if (!iter_is_iovec(to))
1352 		return -EINVAL;
1353 
1354 	fuse_copy_init(&cs, 1, to);
1355 
1356 	return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1357 }
1358 
1359 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1360 				    struct pipe_inode_info *pipe,
1361 				    size_t len, unsigned int flags)
1362 {
1363 	int total, ret;
1364 	int page_nr = 0;
1365 	struct pipe_buffer *bufs;
1366 	struct fuse_copy_state cs;
1367 	struct fuse_dev *fud = fuse_get_dev(in);
1368 
1369 	if (!fud)
1370 		return -EPERM;
1371 
1372 	bufs = kvmalloc_array(pipe->max_usage, sizeof(struct pipe_buffer),
1373 			      GFP_KERNEL);
1374 	if (!bufs)
1375 		return -ENOMEM;
1376 
1377 	fuse_copy_init(&cs, 1, NULL);
1378 	cs.pipebufs = bufs;
1379 	cs.pipe = pipe;
1380 	ret = fuse_dev_do_read(fud, in, &cs, len);
1381 	if (ret < 0)
1382 		goto out;
1383 
1384 	if (pipe_occupancy(pipe->head, pipe->tail) + cs.nr_segs > pipe->max_usage) {
1385 		ret = -EIO;
1386 		goto out;
1387 	}
1388 
1389 	for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1390 		/*
1391 		 * Need to be careful about this.  Having buf->ops in module
1392 		 * code can Oops if the buffer persists after module unload.
1393 		 */
1394 		bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1395 		bufs[page_nr].flags = 0;
1396 		ret = add_to_pipe(pipe, &bufs[page_nr++]);
1397 		if (unlikely(ret < 0))
1398 			break;
1399 	}
1400 	if (total)
1401 		ret = total;
1402 out:
1403 	for (; page_nr < cs.nr_segs; page_nr++)
1404 		put_page(bufs[page_nr].page);
1405 
1406 	kvfree(bufs);
1407 	return ret;
1408 }
1409 
1410 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1411 			    struct fuse_copy_state *cs)
1412 {
1413 	struct fuse_notify_poll_wakeup_out outarg;
1414 	int err = -EINVAL;
1415 
1416 	if (size != sizeof(outarg))
1417 		goto err;
1418 
1419 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1420 	if (err)
1421 		goto err;
1422 
1423 	fuse_copy_finish(cs);
1424 	return fuse_notify_poll_wakeup(fc, &outarg);
1425 
1426 err:
1427 	fuse_copy_finish(cs);
1428 	return err;
1429 }
1430 
1431 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1432 				   struct fuse_copy_state *cs)
1433 {
1434 	struct fuse_notify_inval_inode_out outarg;
1435 	int err = -EINVAL;
1436 
1437 	if (size != sizeof(outarg))
1438 		goto err;
1439 
1440 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1441 	if (err)
1442 		goto err;
1443 	fuse_copy_finish(cs);
1444 
1445 	down_read(&fc->killsb);
1446 	err = fuse_reverse_inval_inode(fc, outarg.ino,
1447 				       outarg.off, outarg.len);
1448 	up_read(&fc->killsb);
1449 	return err;
1450 
1451 err:
1452 	fuse_copy_finish(cs);
1453 	return err;
1454 }
1455 
1456 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1457 				   struct fuse_copy_state *cs)
1458 {
1459 	struct fuse_notify_inval_entry_out outarg;
1460 	int err = -ENOMEM;
1461 	char *buf;
1462 	struct qstr name;
1463 
1464 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1465 	if (!buf)
1466 		goto err;
1467 
1468 	err = -EINVAL;
1469 	if (size < sizeof(outarg))
1470 		goto err;
1471 
1472 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1473 	if (err)
1474 		goto err;
1475 
1476 	err = -ENAMETOOLONG;
1477 	if (outarg.namelen > FUSE_NAME_MAX)
1478 		goto err;
1479 
1480 	err = -EINVAL;
1481 	if (size != sizeof(outarg) + outarg.namelen + 1)
1482 		goto err;
1483 
1484 	name.name = buf;
1485 	name.len = outarg.namelen;
1486 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1487 	if (err)
1488 		goto err;
1489 	fuse_copy_finish(cs);
1490 	buf[outarg.namelen] = 0;
1491 
1492 	down_read(&fc->killsb);
1493 	err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name);
1494 	up_read(&fc->killsb);
1495 	kfree(buf);
1496 	return err;
1497 
1498 err:
1499 	kfree(buf);
1500 	fuse_copy_finish(cs);
1501 	return err;
1502 }
1503 
1504 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1505 			      struct fuse_copy_state *cs)
1506 {
1507 	struct fuse_notify_delete_out outarg;
1508 	int err = -ENOMEM;
1509 	char *buf;
1510 	struct qstr name;
1511 
1512 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1513 	if (!buf)
1514 		goto err;
1515 
1516 	err = -EINVAL;
1517 	if (size < sizeof(outarg))
1518 		goto err;
1519 
1520 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1521 	if (err)
1522 		goto err;
1523 
1524 	err = -ENAMETOOLONG;
1525 	if (outarg.namelen > FUSE_NAME_MAX)
1526 		goto err;
1527 
1528 	err = -EINVAL;
1529 	if (size != sizeof(outarg) + outarg.namelen + 1)
1530 		goto err;
1531 
1532 	name.name = buf;
1533 	name.len = outarg.namelen;
1534 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1535 	if (err)
1536 		goto err;
1537 	fuse_copy_finish(cs);
1538 	buf[outarg.namelen] = 0;
1539 
1540 	down_read(&fc->killsb);
1541 	err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name);
1542 	up_read(&fc->killsb);
1543 	kfree(buf);
1544 	return err;
1545 
1546 err:
1547 	kfree(buf);
1548 	fuse_copy_finish(cs);
1549 	return err;
1550 }
1551 
1552 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1553 			     struct fuse_copy_state *cs)
1554 {
1555 	struct fuse_notify_store_out outarg;
1556 	struct inode *inode;
1557 	struct address_space *mapping;
1558 	u64 nodeid;
1559 	int err;
1560 	pgoff_t index;
1561 	unsigned int offset;
1562 	unsigned int num;
1563 	loff_t file_size;
1564 	loff_t end;
1565 
1566 	err = -EINVAL;
1567 	if (size < sizeof(outarg))
1568 		goto out_finish;
1569 
1570 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1571 	if (err)
1572 		goto out_finish;
1573 
1574 	err = -EINVAL;
1575 	if (size - sizeof(outarg) != outarg.size)
1576 		goto out_finish;
1577 
1578 	nodeid = outarg.nodeid;
1579 
1580 	down_read(&fc->killsb);
1581 
1582 	err = -ENOENT;
1583 	inode = fuse_ilookup(fc, nodeid,  NULL);
1584 	if (!inode)
1585 		goto out_up_killsb;
1586 
1587 	mapping = inode->i_mapping;
1588 	index = outarg.offset >> PAGE_SHIFT;
1589 	offset = outarg.offset & ~PAGE_MASK;
1590 	file_size = i_size_read(inode);
1591 	end = outarg.offset + outarg.size;
1592 	if (end > file_size) {
1593 		file_size = end;
1594 		fuse_write_update_size(inode, file_size);
1595 	}
1596 
1597 	num = outarg.size;
1598 	while (num) {
1599 		struct page *page;
1600 		unsigned int this_num;
1601 
1602 		err = -ENOMEM;
1603 		page = find_or_create_page(mapping, index,
1604 					   mapping_gfp_mask(mapping));
1605 		if (!page)
1606 			goto out_iput;
1607 
1608 		this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1609 		err = fuse_copy_page(cs, &page, offset, this_num, 0);
1610 		if (!err && offset == 0 &&
1611 		    (this_num == PAGE_SIZE || file_size == end))
1612 			SetPageUptodate(page);
1613 		unlock_page(page);
1614 		put_page(page);
1615 
1616 		if (err)
1617 			goto out_iput;
1618 
1619 		num -= this_num;
1620 		offset = 0;
1621 		index++;
1622 	}
1623 
1624 	err = 0;
1625 
1626 out_iput:
1627 	iput(inode);
1628 out_up_killsb:
1629 	up_read(&fc->killsb);
1630 out_finish:
1631 	fuse_copy_finish(cs);
1632 	return err;
1633 }
1634 
1635 struct fuse_retrieve_args {
1636 	struct fuse_args_pages ap;
1637 	struct fuse_notify_retrieve_in inarg;
1638 };
1639 
1640 static void fuse_retrieve_end(struct fuse_mount *fm, struct fuse_args *args,
1641 			      int error)
1642 {
1643 	struct fuse_retrieve_args *ra =
1644 		container_of(args, typeof(*ra), ap.args);
1645 
1646 	release_pages(ra->ap.pages, ra->ap.num_pages);
1647 	kfree(ra);
1648 }
1649 
1650 static int fuse_retrieve(struct fuse_mount *fm, struct inode *inode,
1651 			 struct fuse_notify_retrieve_out *outarg)
1652 {
1653 	int err;
1654 	struct address_space *mapping = inode->i_mapping;
1655 	pgoff_t index;
1656 	loff_t file_size;
1657 	unsigned int num;
1658 	unsigned int offset;
1659 	size_t total_len = 0;
1660 	unsigned int num_pages;
1661 	struct fuse_conn *fc = fm->fc;
1662 	struct fuse_retrieve_args *ra;
1663 	size_t args_size = sizeof(*ra);
1664 	struct fuse_args_pages *ap;
1665 	struct fuse_args *args;
1666 
1667 	offset = outarg->offset & ~PAGE_MASK;
1668 	file_size = i_size_read(inode);
1669 
1670 	num = min(outarg->size, fc->max_write);
1671 	if (outarg->offset > file_size)
1672 		num = 0;
1673 	else if (outarg->offset + num > file_size)
1674 		num = file_size - outarg->offset;
1675 
1676 	num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1677 	num_pages = min(num_pages, fc->max_pages);
1678 
1679 	args_size += num_pages * (sizeof(ap->pages[0]) + sizeof(ap->descs[0]));
1680 
1681 	ra = kzalloc(args_size, GFP_KERNEL);
1682 	if (!ra)
1683 		return -ENOMEM;
1684 
1685 	ap = &ra->ap;
1686 	ap->pages = (void *) (ra + 1);
1687 	ap->descs = (void *) (ap->pages + num_pages);
1688 
1689 	args = &ap->args;
1690 	args->nodeid = outarg->nodeid;
1691 	args->opcode = FUSE_NOTIFY_REPLY;
1692 	args->in_numargs = 2;
1693 	args->in_pages = true;
1694 	args->end = fuse_retrieve_end;
1695 
1696 	index = outarg->offset >> PAGE_SHIFT;
1697 
1698 	while (num && ap->num_pages < num_pages) {
1699 		struct page *page;
1700 		unsigned int this_num;
1701 
1702 		page = find_get_page(mapping, index);
1703 		if (!page)
1704 			break;
1705 
1706 		this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1707 		ap->pages[ap->num_pages] = page;
1708 		ap->descs[ap->num_pages].offset = offset;
1709 		ap->descs[ap->num_pages].length = this_num;
1710 		ap->num_pages++;
1711 
1712 		offset = 0;
1713 		num -= this_num;
1714 		total_len += this_num;
1715 		index++;
1716 	}
1717 	ra->inarg.offset = outarg->offset;
1718 	ra->inarg.size = total_len;
1719 	args->in_args[0].size = sizeof(ra->inarg);
1720 	args->in_args[0].value = &ra->inarg;
1721 	args->in_args[1].size = total_len;
1722 
1723 	err = fuse_simple_notify_reply(fm, args, outarg->notify_unique);
1724 	if (err)
1725 		fuse_retrieve_end(fm, args, err);
1726 
1727 	return err;
1728 }
1729 
1730 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1731 				struct fuse_copy_state *cs)
1732 {
1733 	struct fuse_notify_retrieve_out outarg;
1734 	struct fuse_mount *fm;
1735 	struct inode *inode;
1736 	u64 nodeid;
1737 	int err;
1738 
1739 	err = -EINVAL;
1740 	if (size != sizeof(outarg))
1741 		goto copy_finish;
1742 
1743 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1744 	if (err)
1745 		goto copy_finish;
1746 
1747 	fuse_copy_finish(cs);
1748 
1749 	down_read(&fc->killsb);
1750 	err = -ENOENT;
1751 	nodeid = outarg.nodeid;
1752 
1753 	inode = fuse_ilookup(fc, nodeid, &fm);
1754 	if (inode) {
1755 		err = fuse_retrieve(fm, inode, &outarg);
1756 		iput(inode);
1757 	}
1758 	up_read(&fc->killsb);
1759 
1760 	return err;
1761 
1762 copy_finish:
1763 	fuse_copy_finish(cs);
1764 	return err;
1765 }
1766 
1767 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1768 		       unsigned int size, struct fuse_copy_state *cs)
1769 {
1770 	/* Don't try to move pages (yet) */
1771 	cs->move_pages = 0;
1772 
1773 	switch (code) {
1774 	case FUSE_NOTIFY_POLL:
1775 		return fuse_notify_poll(fc, size, cs);
1776 
1777 	case FUSE_NOTIFY_INVAL_INODE:
1778 		return fuse_notify_inval_inode(fc, size, cs);
1779 
1780 	case FUSE_NOTIFY_INVAL_ENTRY:
1781 		return fuse_notify_inval_entry(fc, size, cs);
1782 
1783 	case FUSE_NOTIFY_STORE:
1784 		return fuse_notify_store(fc, size, cs);
1785 
1786 	case FUSE_NOTIFY_RETRIEVE:
1787 		return fuse_notify_retrieve(fc, size, cs);
1788 
1789 	case FUSE_NOTIFY_DELETE:
1790 		return fuse_notify_delete(fc, size, cs);
1791 
1792 	default:
1793 		fuse_copy_finish(cs);
1794 		return -EINVAL;
1795 	}
1796 }
1797 
1798 /* Look up request on processing list by unique ID */
1799 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1800 {
1801 	unsigned int hash = fuse_req_hash(unique);
1802 	struct fuse_req *req;
1803 
1804 	list_for_each_entry(req, &fpq->processing[hash], list) {
1805 		if (req->in.h.unique == unique)
1806 			return req;
1807 	}
1808 	return NULL;
1809 }
1810 
1811 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args,
1812 			 unsigned nbytes)
1813 {
1814 	unsigned reqsize = sizeof(struct fuse_out_header);
1815 
1816 	reqsize += fuse_len_args(args->out_numargs, args->out_args);
1817 
1818 	if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar))
1819 		return -EINVAL;
1820 	else if (reqsize > nbytes) {
1821 		struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1];
1822 		unsigned diffsize = reqsize - nbytes;
1823 
1824 		if (diffsize > lastarg->size)
1825 			return -EINVAL;
1826 		lastarg->size -= diffsize;
1827 	}
1828 	return fuse_copy_args(cs, args->out_numargs, args->out_pages,
1829 			      args->out_args, args->page_zeroing);
1830 }
1831 
1832 /*
1833  * Write a single reply to a request.  First the header is copied from
1834  * the write buffer.  The request is then searched on the processing
1835  * list by the unique ID found in the header.  If found, then remove
1836  * it from the list and copy the rest of the buffer to the request.
1837  * The request is finished by calling fuse_request_end().
1838  */
1839 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1840 				 struct fuse_copy_state *cs, size_t nbytes)
1841 {
1842 	int err;
1843 	struct fuse_conn *fc = fud->fc;
1844 	struct fuse_pqueue *fpq = &fud->pq;
1845 	struct fuse_req *req;
1846 	struct fuse_out_header oh;
1847 
1848 	err = -EINVAL;
1849 	if (nbytes < sizeof(struct fuse_out_header))
1850 		goto out;
1851 
1852 	err = fuse_copy_one(cs, &oh, sizeof(oh));
1853 	if (err)
1854 		goto copy_finish;
1855 
1856 	err = -EINVAL;
1857 	if (oh.len != nbytes)
1858 		goto copy_finish;
1859 
1860 	/*
1861 	 * Zero oh.unique indicates unsolicited notification message
1862 	 * and error contains notification code.
1863 	 */
1864 	if (!oh.unique) {
1865 		err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1866 		goto out;
1867 	}
1868 
1869 	err = -EINVAL;
1870 	if (oh.error <= -512 || oh.error > 0)
1871 		goto copy_finish;
1872 
1873 	spin_lock(&fpq->lock);
1874 	req = NULL;
1875 	if (fpq->connected)
1876 		req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
1877 
1878 	err = -ENOENT;
1879 	if (!req) {
1880 		spin_unlock(&fpq->lock);
1881 		goto copy_finish;
1882 	}
1883 
1884 	/* Is it an interrupt reply ID? */
1885 	if (oh.unique & FUSE_INT_REQ_BIT) {
1886 		__fuse_get_request(req);
1887 		spin_unlock(&fpq->lock);
1888 
1889 		err = 0;
1890 		if (nbytes != sizeof(struct fuse_out_header))
1891 			err = -EINVAL;
1892 		else if (oh.error == -ENOSYS)
1893 			fc->no_interrupt = 1;
1894 		else if (oh.error == -EAGAIN)
1895 			err = queue_interrupt(req);
1896 
1897 		fuse_put_request(req);
1898 
1899 		goto copy_finish;
1900 	}
1901 
1902 	clear_bit(FR_SENT, &req->flags);
1903 	list_move(&req->list, &fpq->io);
1904 	req->out.h = oh;
1905 	set_bit(FR_LOCKED, &req->flags);
1906 	spin_unlock(&fpq->lock);
1907 	cs->req = req;
1908 	if (!req->args->page_replace)
1909 		cs->move_pages = 0;
1910 
1911 	if (oh.error)
1912 		err = nbytes != sizeof(oh) ? -EINVAL : 0;
1913 	else
1914 		err = copy_out_args(cs, req->args, nbytes);
1915 	fuse_copy_finish(cs);
1916 
1917 	spin_lock(&fpq->lock);
1918 	clear_bit(FR_LOCKED, &req->flags);
1919 	if (!fpq->connected)
1920 		err = -ENOENT;
1921 	else if (err)
1922 		req->out.h.error = -EIO;
1923 	if (!test_bit(FR_PRIVATE, &req->flags))
1924 		list_del_init(&req->list);
1925 	spin_unlock(&fpq->lock);
1926 
1927 	fuse_request_end(req);
1928 out:
1929 	return err ? err : nbytes;
1930 
1931 copy_finish:
1932 	fuse_copy_finish(cs);
1933 	goto out;
1934 }
1935 
1936 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1937 {
1938 	struct fuse_copy_state cs;
1939 	struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1940 
1941 	if (!fud)
1942 		return -EPERM;
1943 
1944 	if (!iter_is_iovec(from))
1945 		return -EINVAL;
1946 
1947 	fuse_copy_init(&cs, 0, from);
1948 
1949 	return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1950 }
1951 
1952 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1953 				     struct file *out, loff_t *ppos,
1954 				     size_t len, unsigned int flags)
1955 {
1956 	unsigned int head, tail, mask, count;
1957 	unsigned nbuf;
1958 	unsigned idx;
1959 	struct pipe_buffer *bufs;
1960 	struct fuse_copy_state cs;
1961 	struct fuse_dev *fud;
1962 	size_t rem;
1963 	ssize_t ret;
1964 
1965 	fud = fuse_get_dev(out);
1966 	if (!fud)
1967 		return -EPERM;
1968 
1969 	pipe_lock(pipe);
1970 
1971 	head = pipe->head;
1972 	tail = pipe->tail;
1973 	mask = pipe->ring_size - 1;
1974 	count = head - tail;
1975 
1976 	bufs = kvmalloc_array(count, sizeof(struct pipe_buffer), GFP_KERNEL);
1977 	if (!bufs) {
1978 		pipe_unlock(pipe);
1979 		return -ENOMEM;
1980 	}
1981 
1982 	nbuf = 0;
1983 	rem = 0;
1984 	for (idx = tail; idx != head && rem < len; idx++)
1985 		rem += pipe->bufs[idx & mask].len;
1986 
1987 	ret = -EINVAL;
1988 	if (rem < len)
1989 		goto out_free;
1990 
1991 	rem = len;
1992 	while (rem) {
1993 		struct pipe_buffer *ibuf;
1994 		struct pipe_buffer *obuf;
1995 
1996 		if (WARN_ON(nbuf >= count || tail == head))
1997 			goto out_free;
1998 
1999 		ibuf = &pipe->bufs[tail & mask];
2000 		obuf = &bufs[nbuf];
2001 
2002 		if (rem >= ibuf->len) {
2003 			*obuf = *ibuf;
2004 			ibuf->ops = NULL;
2005 			tail++;
2006 			pipe->tail = tail;
2007 		} else {
2008 			if (!pipe_buf_get(pipe, ibuf))
2009 				goto out_free;
2010 
2011 			*obuf = *ibuf;
2012 			obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2013 			obuf->len = rem;
2014 			ibuf->offset += obuf->len;
2015 			ibuf->len -= obuf->len;
2016 		}
2017 		nbuf++;
2018 		rem -= obuf->len;
2019 	}
2020 	pipe_unlock(pipe);
2021 
2022 	fuse_copy_init(&cs, 0, NULL);
2023 	cs.pipebufs = bufs;
2024 	cs.nr_segs = nbuf;
2025 	cs.pipe = pipe;
2026 
2027 	if (flags & SPLICE_F_MOVE)
2028 		cs.move_pages = 1;
2029 
2030 	ret = fuse_dev_do_write(fud, &cs, len);
2031 
2032 	pipe_lock(pipe);
2033 out_free:
2034 	for (idx = 0; idx < nbuf; idx++)
2035 		pipe_buf_release(pipe, &bufs[idx]);
2036 	pipe_unlock(pipe);
2037 
2038 	kvfree(bufs);
2039 	return ret;
2040 }
2041 
2042 static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
2043 {
2044 	__poll_t mask = EPOLLOUT | EPOLLWRNORM;
2045 	struct fuse_iqueue *fiq;
2046 	struct fuse_dev *fud = fuse_get_dev(file);
2047 
2048 	if (!fud)
2049 		return EPOLLERR;
2050 
2051 	fiq = &fud->fc->iq;
2052 	poll_wait(file, &fiq->waitq, wait);
2053 
2054 	spin_lock(&fiq->lock);
2055 	if (!fiq->connected)
2056 		mask = EPOLLERR;
2057 	else if (request_pending(fiq))
2058 		mask |= EPOLLIN | EPOLLRDNORM;
2059 	spin_unlock(&fiq->lock);
2060 
2061 	return mask;
2062 }
2063 
2064 /* Abort all requests on the given list (pending or processing) */
2065 static void end_requests(struct list_head *head)
2066 {
2067 	while (!list_empty(head)) {
2068 		struct fuse_req *req;
2069 		req = list_entry(head->next, struct fuse_req, list);
2070 		req->out.h.error = -ECONNABORTED;
2071 		clear_bit(FR_SENT, &req->flags);
2072 		list_del_init(&req->list);
2073 		fuse_request_end(req);
2074 	}
2075 }
2076 
2077 static void end_polls(struct fuse_conn *fc)
2078 {
2079 	struct rb_node *p;
2080 
2081 	p = rb_first(&fc->polled_files);
2082 
2083 	while (p) {
2084 		struct fuse_file *ff;
2085 		ff = rb_entry(p, struct fuse_file, polled_node);
2086 		wake_up_interruptible_all(&ff->poll_wait);
2087 
2088 		p = rb_next(p);
2089 	}
2090 }
2091 
2092 /*
2093  * Abort all requests.
2094  *
2095  * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2096  * filesystem.
2097  *
2098  * The same effect is usually achievable through killing the filesystem daemon
2099  * and all users of the filesystem.  The exception is the combination of an
2100  * asynchronous request and the tricky deadlock (see
2101  * Documentation/filesystems/fuse.rst).
2102  *
2103  * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2104  * requests, they should be finished off immediately.  Locked requests will be
2105  * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2106  * requests.  It is possible that some request will finish before we can.  This
2107  * is OK, the request will in that case be removed from the list before we touch
2108  * it.
2109  */
2110 void fuse_abort_conn(struct fuse_conn *fc)
2111 {
2112 	struct fuse_iqueue *fiq = &fc->iq;
2113 
2114 	spin_lock(&fc->lock);
2115 	if (fc->connected) {
2116 		struct fuse_dev *fud;
2117 		struct fuse_req *req, *next;
2118 		LIST_HEAD(to_end);
2119 		unsigned int i;
2120 
2121 		/* Background queuing checks fc->connected under bg_lock */
2122 		spin_lock(&fc->bg_lock);
2123 		fc->connected = 0;
2124 		spin_unlock(&fc->bg_lock);
2125 
2126 		fuse_set_initialized(fc);
2127 		list_for_each_entry(fud, &fc->devices, entry) {
2128 			struct fuse_pqueue *fpq = &fud->pq;
2129 
2130 			spin_lock(&fpq->lock);
2131 			fpq->connected = 0;
2132 			list_for_each_entry_safe(req, next, &fpq->io, list) {
2133 				req->out.h.error = -ECONNABORTED;
2134 				spin_lock(&req->waitq.lock);
2135 				set_bit(FR_ABORTED, &req->flags);
2136 				if (!test_bit(FR_LOCKED, &req->flags)) {
2137 					set_bit(FR_PRIVATE, &req->flags);
2138 					__fuse_get_request(req);
2139 					list_move(&req->list, &to_end);
2140 				}
2141 				spin_unlock(&req->waitq.lock);
2142 			}
2143 			for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2144 				list_splice_tail_init(&fpq->processing[i],
2145 						      &to_end);
2146 			spin_unlock(&fpq->lock);
2147 		}
2148 		spin_lock(&fc->bg_lock);
2149 		fc->blocked = 0;
2150 		fc->max_background = UINT_MAX;
2151 		flush_bg_queue(fc);
2152 		spin_unlock(&fc->bg_lock);
2153 
2154 		spin_lock(&fiq->lock);
2155 		fiq->connected = 0;
2156 		list_for_each_entry(req, &fiq->pending, list)
2157 			clear_bit(FR_PENDING, &req->flags);
2158 		list_splice_tail_init(&fiq->pending, &to_end);
2159 		while (forget_pending(fiq))
2160 			kfree(fuse_dequeue_forget(fiq, 1, NULL));
2161 		wake_up_all(&fiq->waitq);
2162 		spin_unlock(&fiq->lock);
2163 		kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2164 		end_polls(fc);
2165 		wake_up_all(&fc->blocked_waitq);
2166 		spin_unlock(&fc->lock);
2167 
2168 		end_requests(&to_end);
2169 	} else {
2170 		spin_unlock(&fc->lock);
2171 	}
2172 }
2173 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2174 
2175 void fuse_wait_aborted(struct fuse_conn *fc)
2176 {
2177 	/* matches implicit memory barrier in fuse_drop_waiting() */
2178 	smp_mb();
2179 	wait_event(fc->blocked_waitq, atomic_read(&fc->num_waiting) == 0);
2180 }
2181 
2182 int fuse_dev_release(struct inode *inode, struct file *file)
2183 {
2184 	struct fuse_dev *fud = fuse_get_dev(file);
2185 
2186 	if (fud) {
2187 		struct fuse_conn *fc = fud->fc;
2188 		struct fuse_pqueue *fpq = &fud->pq;
2189 		LIST_HEAD(to_end);
2190 		unsigned int i;
2191 
2192 		spin_lock(&fpq->lock);
2193 		WARN_ON(!list_empty(&fpq->io));
2194 		for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
2195 			list_splice_init(&fpq->processing[i], &to_end);
2196 		spin_unlock(&fpq->lock);
2197 
2198 		end_requests(&to_end);
2199 
2200 		/* Are we the last open device? */
2201 		if (atomic_dec_and_test(&fc->dev_count)) {
2202 			WARN_ON(fc->iq.fasync != NULL);
2203 			fuse_abort_conn(fc);
2204 		}
2205 		fuse_dev_free(fud);
2206 	}
2207 	return 0;
2208 }
2209 EXPORT_SYMBOL_GPL(fuse_dev_release);
2210 
2211 static int fuse_dev_fasync(int fd, struct file *file, int on)
2212 {
2213 	struct fuse_dev *fud = fuse_get_dev(file);
2214 
2215 	if (!fud)
2216 		return -EPERM;
2217 
2218 	/* No locking - fasync_helper does its own locking */
2219 	return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2220 }
2221 
2222 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2223 {
2224 	struct fuse_dev *fud;
2225 
2226 	if (new->private_data)
2227 		return -EINVAL;
2228 
2229 	fud = fuse_dev_alloc_install(fc);
2230 	if (!fud)
2231 		return -ENOMEM;
2232 
2233 	new->private_data = fud;
2234 	atomic_inc(&fc->dev_count);
2235 
2236 	return 0;
2237 }
2238 
2239 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2240 			   unsigned long arg)
2241 {
2242 	int res;
2243 	int oldfd;
2244 	struct fuse_dev *fud = NULL;
2245 
2246 	switch (cmd) {
2247 	case FUSE_DEV_IOC_CLONE:
2248 		res = -EFAULT;
2249 		if (!get_user(oldfd, (__u32 __user *)arg)) {
2250 			struct file *old = fget(oldfd);
2251 
2252 			res = -EINVAL;
2253 			if (old) {
2254 				/*
2255 				 * Check against file->f_op because CUSE
2256 				 * uses the same ioctl handler.
2257 				 */
2258 				if (old->f_op == file->f_op &&
2259 				    old->f_cred->user_ns == file->f_cred->user_ns)
2260 					fud = fuse_get_dev(old);
2261 
2262 				if (fud) {
2263 					mutex_lock(&fuse_mutex);
2264 					res = fuse_device_clone(fud->fc, file);
2265 					mutex_unlock(&fuse_mutex);
2266 				}
2267 				fput(old);
2268 			}
2269 		}
2270 		break;
2271 	default:
2272 		res = -ENOTTY;
2273 		break;
2274 	}
2275 	return res;
2276 }
2277 
2278 const struct file_operations fuse_dev_operations = {
2279 	.owner		= THIS_MODULE,
2280 	.open		= fuse_dev_open,
2281 	.llseek		= no_llseek,
2282 	.read_iter	= fuse_dev_read,
2283 	.splice_read	= fuse_dev_splice_read,
2284 	.write_iter	= fuse_dev_write,
2285 	.splice_write	= fuse_dev_splice_write,
2286 	.poll		= fuse_dev_poll,
2287 	.release	= fuse_dev_release,
2288 	.fasync		= fuse_dev_fasync,
2289 	.unlocked_ioctl = fuse_dev_ioctl,
2290 	.compat_ioctl   = compat_ptr_ioctl,
2291 };
2292 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2293 
2294 static struct miscdevice fuse_miscdevice = {
2295 	.minor = FUSE_MINOR,
2296 	.name  = "fuse",
2297 	.fops = &fuse_dev_operations,
2298 };
2299 
2300 int __init fuse_dev_init(void)
2301 {
2302 	int err = -ENOMEM;
2303 	fuse_req_cachep = kmem_cache_create("fuse_request",
2304 					    sizeof(struct fuse_req),
2305 					    0, 0, NULL);
2306 	if (!fuse_req_cachep)
2307 		goto out;
2308 
2309 	err = misc_register(&fuse_miscdevice);
2310 	if (err)
2311 		goto out_cache_clean;
2312 
2313 	return 0;
2314 
2315  out_cache_clean:
2316 	kmem_cache_destroy(fuse_req_cachep);
2317  out:
2318 	return err;
2319 }
2320 
2321 void fuse_dev_cleanup(void)
2322 {
2323 	misc_deregister(&fuse_miscdevice);
2324 	kmem_cache_destroy(fuse_req_cachep);
2325 }
2326