xref: /linux/fs/fuse/dev.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4 
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8 
9 #include "fuse_i.h"
10 
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/sched/signal.h>
15 #include <linux/uio.h>
16 #include <linux/miscdevice.h>
17 #include <linux/pagemap.h>
18 #include <linux/file.h>
19 #include <linux/slab.h>
20 #include <linux/pipe_fs_i.h>
21 #include <linux/swap.h>
22 #include <linux/splice.h>
23 #include <linux/sched.h>
24 
25 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
26 MODULE_ALIAS("devname:fuse");
27 
28 static struct kmem_cache *fuse_req_cachep;
29 
30 static struct fuse_dev *fuse_get_dev(struct file *file)
31 {
32 	/*
33 	 * Lockless access is OK, because file->private data is set
34 	 * once during mount and is valid until the file is released.
35 	 */
36 	return READ_ONCE(file->private_data);
37 }
38 
39 static void fuse_request_init(struct fuse_req *req, struct page **pages,
40 			      struct fuse_page_desc *page_descs,
41 			      unsigned npages)
42 {
43 	memset(req, 0, sizeof(*req));
44 	memset(pages, 0, sizeof(*pages) * npages);
45 	memset(page_descs, 0, sizeof(*page_descs) * npages);
46 	INIT_LIST_HEAD(&req->list);
47 	INIT_LIST_HEAD(&req->intr_entry);
48 	init_waitqueue_head(&req->waitq);
49 	refcount_set(&req->count, 1);
50 	req->pages = pages;
51 	req->page_descs = page_descs;
52 	req->max_pages = npages;
53 	__set_bit(FR_PENDING, &req->flags);
54 }
55 
56 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
57 {
58 	struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
59 	if (req) {
60 		struct page **pages;
61 		struct fuse_page_desc *page_descs;
62 
63 		if (npages <= FUSE_REQ_INLINE_PAGES) {
64 			pages = req->inline_pages;
65 			page_descs = req->inline_page_descs;
66 		} else {
67 			pages = kmalloc(sizeof(struct page *) * npages, flags);
68 			page_descs = kmalloc(sizeof(struct fuse_page_desc) *
69 					     npages, flags);
70 		}
71 
72 		if (!pages || !page_descs) {
73 			kfree(pages);
74 			kfree(page_descs);
75 			kmem_cache_free(fuse_req_cachep, req);
76 			return NULL;
77 		}
78 
79 		fuse_request_init(req, pages, page_descs, npages);
80 	}
81 	return req;
82 }
83 
84 struct fuse_req *fuse_request_alloc(unsigned npages)
85 {
86 	return __fuse_request_alloc(npages, GFP_KERNEL);
87 }
88 EXPORT_SYMBOL_GPL(fuse_request_alloc);
89 
90 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
91 {
92 	return __fuse_request_alloc(npages, GFP_NOFS);
93 }
94 
95 void fuse_request_free(struct fuse_req *req)
96 {
97 	if (req->pages != req->inline_pages) {
98 		kfree(req->pages);
99 		kfree(req->page_descs);
100 	}
101 	kmem_cache_free(fuse_req_cachep, req);
102 }
103 
104 void __fuse_get_request(struct fuse_req *req)
105 {
106 	refcount_inc(&req->count);
107 }
108 
109 /* Must be called with > 1 refcount */
110 static void __fuse_put_request(struct fuse_req *req)
111 {
112 	refcount_dec(&req->count);
113 }
114 
115 static void fuse_req_init_context(struct fuse_conn *fc, struct fuse_req *req)
116 {
117 	req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
118 	req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
119 	req->in.h.pid = pid_nr_ns(task_pid(current), fc->pid_ns);
120 }
121 
122 void fuse_set_initialized(struct fuse_conn *fc)
123 {
124 	/* Make sure stores before this are seen on another CPU */
125 	smp_wmb();
126 	fc->initialized = 1;
127 }
128 
129 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
130 {
131 	return !fc->initialized || (for_background && fc->blocked);
132 }
133 
134 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
135 				       bool for_background)
136 {
137 	struct fuse_req *req;
138 	int err;
139 	atomic_inc(&fc->num_waiting);
140 
141 	if (fuse_block_alloc(fc, for_background)) {
142 		err = -EINTR;
143 		if (wait_event_killable_exclusive(fc->blocked_waitq,
144 				!fuse_block_alloc(fc, for_background)))
145 			goto out;
146 	}
147 	/* Matches smp_wmb() in fuse_set_initialized() */
148 	smp_rmb();
149 
150 	err = -ENOTCONN;
151 	if (!fc->connected)
152 		goto out;
153 
154 	err = -ECONNREFUSED;
155 	if (fc->conn_error)
156 		goto out;
157 
158 	req = fuse_request_alloc(npages);
159 	err = -ENOMEM;
160 	if (!req) {
161 		if (for_background)
162 			wake_up(&fc->blocked_waitq);
163 		goto out;
164 	}
165 
166 	fuse_req_init_context(fc, req);
167 	__set_bit(FR_WAITING, &req->flags);
168 	if (for_background)
169 		__set_bit(FR_BACKGROUND, &req->flags);
170 
171 	return req;
172 
173  out:
174 	atomic_dec(&fc->num_waiting);
175 	return ERR_PTR(err);
176 }
177 
178 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
179 {
180 	return __fuse_get_req(fc, npages, false);
181 }
182 EXPORT_SYMBOL_GPL(fuse_get_req);
183 
184 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
185 					     unsigned npages)
186 {
187 	return __fuse_get_req(fc, npages, true);
188 }
189 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
190 
191 /*
192  * Return request in fuse_file->reserved_req.  However that may
193  * currently be in use.  If that is the case, wait for it to become
194  * available.
195  */
196 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
197 					 struct file *file)
198 {
199 	struct fuse_req *req = NULL;
200 	struct fuse_file *ff = file->private_data;
201 
202 	do {
203 		wait_event(fc->reserved_req_waitq, ff->reserved_req);
204 		spin_lock(&fc->lock);
205 		if (ff->reserved_req) {
206 			req = ff->reserved_req;
207 			ff->reserved_req = NULL;
208 			req->stolen_file = get_file(file);
209 		}
210 		spin_unlock(&fc->lock);
211 	} while (!req);
212 
213 	return req;
214 }
215 
216 /*
217  * Put stolen request back into fuse_file->reserved_req
218  */
219 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
220 {
221 	struct file *file = req->stolen_file;
222 	struct fuse_file *ff = file->private_data;
223 
224 	spin_lock(&fc->lock);
225 	fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
226 	BUG_ON(ff->reserved_req);
227 	ff->reserved_req = req;
228 	wake_up_all(&fc->reserved_req_waitq);
229 	spin_unlock(&fc->lock);
230 	fput(file);
231 }
232 
233 /*
234  * Gets a requests for a file operation, always succeeds
235  *
236  * This is used for sending the FLUSH request, which must get to
237  * userspace, due to POSIX locks which may need to be unlocked.
238  *
239  * If allocation fails due to OOM, use the reserved request in
240  * fuse_file.
241  *
242  * This is very unlikely to deadlock accidentally, since the
243  * filesystem should not have it's own file open.  If deadlock is
244  * intentional, it can still be broken by "aborting" the filesystem.
245  */
246 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
247 					     struct file *file)
248 {
249 	struct fuse_req *req;
250 
251 	atomic_inc(&fc->num_waiting);
252 	wait_event(fc->blocked_waitq, fc->initialized);
253 	/* Matches smp_wmb() in fuse_set_initialized() */
254 	smp_rmb();
255 	req = fuse_request_alloc(0);
256 	if (!req)
257 		req = get_reserved_req(fc, file);
258 
259 	fuse_req_init_context(fc, req);
260 	__set_bit(FR_WAITING, &req->flags);
261 	__clear_bit(FR_BACKGROUND, &req->flags);
262 	return req;
263 }
264 
265 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
266 {
267 	if (refcount_dec_and_test(&req->count)) {
268 		if (test_bit(FR_BACKGROUND, &req->flags)) {
269 			/*
270 			 * We get here in the unlikely case that a background
271 			 * request was allocated but not sent
272 			 */
273 			spin_lock(&fc->lock);
274 			if (!fc->blocked)
275 				wake_up(&fc->blocked_waitq);
276 			spin_unlock(&fc->lock);
277 		}
278 
279 		if (test_bit(FR_WAITING, &req->flags)) {
280 			__clear_bit(FR_WAITING, &req->flags);
281 			atomic_dec(&fc->num_waiting);
282 		}
283 
284 		if (req->stolen_file)
285 			put_reserved_req(fc, req);
286 		else
287 			fuse_request_free(req);
288 	}
289 }
290 EXPORT_SYMBOL_GPL(fuse_put_request);
291 
292 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
293 {
294 	unsigned nbytes = 0;
295 	unsigned i;
296 
297 	for (i = 0; i < numargs; i++)
298 		nbytes += args[i].size;
299 
300 	return nbytes;
301 }
302 
303 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
304 {
305 	return ++fiq->reqctr;
306 }
307 
308 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
309 {
310 	req->in.h.len = sizeof(struct fuse_in_header) +
311 		len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
312 	list_add_tail(&req->list, &fiq->pending);
313 	wake_up_locked(&fiq->waitq);
314 	kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
315 }
316 
317 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
318 		       u64 nodeid, u64 nlookup)
319 {
320 	struct fuse_iqueue *fiq = &fc->iq;
321 
322 	forget->forget_one.nodeid = nodeid;
323 	forget->forget_one.nlookup = nlookup;
324 
325 	spin_lock(&fiq->waitq.lock);
326 	if (fiq->connected) {
327 		fiq->forget_list_tail->next = forget;
328 		fiq->forget_list_tail = forget;
329 		wake_up_locked(&fiq->waitq);
330 		kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
331 	} else {
332 		kfree(forget);
333 	}
334 	spin_unlock(&fiq->waitq.lock);
335 }
336 
337 static void flush_bg_queue(struct fuse_conn *fc)
338 {
339 	while (fc->active_background < fc->max_background &&
340 	       !list_empty(&fc->bg_queue)) {
341 		struct fuse_req *req;
342 		struct fuse_iqueue *fiq = &fc->iq;
343 
344 		req = list_entry(fc->bg_queue.next, struct fuse_req, list);
345 		list_del(&req->list);
346 		fc->active_background++;
347 		spin_lock(&fiq->waitq.lock);
348 		req->in.h.unique = fuse_get_unique(fiq);
349 		queue_request(fiq, req);
350 		spin_unlock(&fiq->waitq.lock);
351 	}
352 }
353 
354 /*
355  * This function is called when a request is finished.  Either a reply
356  * has arrived or it was aborted (and not yet sent) or some error
357  * occurred during communication with userspace, or the device file
358  * was closed.  The requester thread is woken up (if still waiting),
359  * the 'end' callback is called if given, else the reference to the
360  * request is released
361  */
362 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
363 {
364 	struct fuse_iqueue *fiq = &fc->iq;
365 
366 	if (test_and_set_bit(FR_FINISHED, &req->flags))
367 		return;
368 
369 	spin_lock(&fiq->waitq.lock);
370 	list_del_init(&req->intr_entry);
371 	spin_unlock(&fiq->waitq.lock);
372 	WARN_ON(test_bit(FR_PENDING, &req->flags));
373 	WARN_ON(test_bit(FR_SENT, &req->flags));
374 	if (test_bit(FR_BACKGROUND, &req->flags)) {
375 		spin_lock(&fc->lock);
376 		clear_bit(FR_BACKGROUND, &req->flags);
377 		if (fc->num_background == fc->max_background)
378 			fc->blocked = 0;
379 
380 		/* Wake up next waiter, if any */
381 		if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
382 			wake_up(&fc->blocked_waitq);
383 
384 		if (fc->num_background == fc->congestion_threshold &&
385 		    fc->connected && fc->sb) {
386 			clear_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
387 			clear_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
388 		}
389 		fc->num_background--;
390 		fc->active_background--;
391 		flush_bg_queue(fc);
392 		spin_unlock(&fc->lock);
393 	}
394 	wake_up(&req->waitq);
395 	if (req->end)
396 		req->end(fc, req);
397 	fuse_put_request(fc, req);
398 }
399 
400 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
401 {
402 	spin_lock(&fiq->waitq.lock);
403 	if (test_bit(FR_FINISHED, &req->flags)) {
404 		spin_unlock(&fiq->waitq.lock);
405 		return;
406 	}
407 	if (list_empty(&req->intr_entry)) {
408 		list_add_tail(&req->intr_entry, &fiq->interrupts);
409 		wake_up_locked(&fiq->waitq);
410 	}
411 	spin_unlock(&fiq->waitq.lock);
412 	kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
413 }
414 
415 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
416 {
417 	struct fuse_iqueue *fiq = &fc->iq;
418 	int err;
419 
420 	if (!fc->no_interrupt) {
421 		/* Any signal may interrupt this */
422 		err = wait_event_interruptible(req->waitq,
423 					test_bit(FR_FINISHED, &req->flags));
424 		if (!err)
425 			return;
426 
427 		set_bit(FR_INTERRUPTED, &req->flags);
428 		/* matches barrier in fuse_dev_do_read() */
429 		smp_mb__after_atomic();
430 		if (test_bit(FR_SENT, &req->flags))
431 			queue_interrupt(fiq, req);
432 	}
433 
434 	if (!test_bit(FR_FORCE, &req->flags)) {
435 		/* Only fatal signals may interrupt this */
436 		err = wait_event_killable(req->waitq,
437 					test_bit(FR_FINISHED, &req->flags));
438 		if (!err)
439 			return;
440 
441 		spin_lock(&fiq->waitq.lock);
442 		/* Request is not yet in userspace, bail out */
443 		if (test_bit(FR_PENDING, &req->flags)) {
444 			list_del(&req->list);
445 			spin_unlock(&fiq->waitq.lock);
446 			__fuse_put_request(req);
447 			req->out.h.error = -EINTR;
448 			return;
449 		}
450 		spin_unlock(&fiq->waitq.lock);
451 	}
452 
453 	/*
454 	 * Either request is already in userspace, or it was forced.
455 	 * Wait it out.
456 	 */
457 	wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
458 }
459 
460 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
461 {
462 	struct fuse_iqueue *fiq = &fc->iq;
463 
464 	BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
465 	spin_lock(&fiq->waitq.lock);
466 	if (!fiq->connected) {
467 		spin_unlock(&fiq->waitq.lock);
468 		req->out.h.error = -ENOTCONN;
469 	} else {
470 		req->in.h.unique = fuse_get_unique(fiq);
471 		queue_request(fiq, req);
472 		/* acquire extra reference, since request is still needed
473 		   after request_end() */
474 		__fuse_get_request(req);
475 		spin_unlock(&fiq->waitq.lock);
476 
477 		request_wait_answer(fc, req);
478 		/* Pairs with smp_wmb() in request_end() */
479 		smp_rmb();
480 	}
481 }
482 
483 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
484 {
485 	__set_bit(FR_ISREPLY, &req->flags);
486 	if (!test_bit(FR_WAITING, &req->flags)) {
487 		__set_bit(FR_WAITING, &req->flags);
488 		atomic_inc(&fc->num_waiting);
489 	}
490 	__fuse_request_send(fc, req);
491 }
492 EXPORT_SYMBOL_GPL(fuse_request_send);
493 
494 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
495 {
496 	if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
497 		args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
498 
499 	if (fc->minor < 9) {
500 		switch (args->in.h.opcode) {
501 		case FUSE_LOOKUP:
502 		case FUSE_CREATE:
503 		case FUSE_MKNOD:
504 		case FUSE_MKDIR:
505 		case FUSE_SYMLINK:
506 		case FUSE_LINK:
507 			args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
508 			break;
509 		case FUSE_GETATTR:
510 		case FUSE_SETATTR:
511 			args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
512 			break;
513 		}
514 	}
515 	if (fc->minor < 12) {
516 		switch (args->in.h.opcode) {
517 		case FUSE_CREATE:
518 			args->in.args[0].size = sizeof(struct fuse_open_in);
519 			break;
520 		case FUSE_MKNOD:
521 			args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
522 			break;
523 		}
524 	}
525 }
526 
527 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
528 {
529 	struct fuse_req *req;
530 	ssize_t ret;
531 
532 	req = fuse_get_req(fc, 0);
533 	if (IS_ERR(req))
534 		return PTR_ERR(req);
535 
536 	/* Needs to be done after fuse_get_req() so that fc->minor is valid */
537 	fuse_adjust_compat(fc, args);
538 
539 	req->in.h.opcode = args->in.h.opcode;
540 	req->in.h.nodeid = args->in.h.nodeid;
541 	req->in.numargs = args->in.numargs;
542 	memcpy(req->in.args, args->in.args,
543 	       args->in.numargs * sizeof(struct fuse_in_arg));
544 	req->out.argvar = args->out.argvar;
545 	req->out.numargs = args->out.numargs;
546 	memcpy(req->out.args, args->out.args,
547 	       args->out.numargs * sizeof(struct fuse_arg));
548 	fuse_request_send(fc, req);
549 	ret = req->out.h.error;
550 	if (!ret && args->out.argvar) {
551 		BUG_ON(args->out.numargs != 1);
552 		ret = req->out.args[0].size;
553 	}
554 	fuse_put_request(fc, req);
555 
556 	return ret;
557 }
558 
559 /*
560  * Called under fc->lock
561  *
562  * fc->connected must have been checked previously
563  */
564 void fuse_request_send_background_locked(struct fuse_conn *fc,
565 					 struct fuse_req *req)
566 {
567 	BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
568 	if (!test_bit(FR_WAITING, &req->flags)) {
569 		__set_bit(FR_WAITING, &req->flags);
570 		atomic_inc(&fc->num_waiting);
571 	}
572 	__set_bit(FR_ISREPLY, &req->flags);
573 	fc->num_background++;
574 	if (fc->num_background == fc->max_background)
575 		fc->blocked = 1;
576 	if (fc->num_background == fc->congestion_threshold && fc->sb) {
577 		set_bdi_congested(fc->sb->s_bdi, BLK_RW_SYNC);
578 		set_bdi_congested(fc->sb->s_bdi, BLK_RW_ASYNC);
579 	}
580 	list_add_tail(&req->list, &fc->bg_queue);
581 	flush_bg_queue(fc);
582 }
583 
584 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
585 {
586 	BUG_ON(!req->end);
587 	spin_lock(&fc->lock);
588 	if (fc->connected) {
589 		fuse_request_send_background_locked(fc, req);
590 		spin_unlock(&fc->lock);
591 	} else {
592 		spin_unlock(&fc->lock);
593 		req->out.h.error = -ENOTCONN;
594 		req->end(fc, req);
595 		fuse_put_request(fc, req);
596 	}
597 }
598 EXPORT_SYMBOL_GPL(fuse_request_send_background);
599 
600 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
601 					  struct fuse_req *req, u64 unique)
602 {
603 	int err = -ENODEV;
604 	struct fuse_iqueue *fiq = &fc->iq;
605 
606 	__clear_bit(FR_ISREPLY, &req->flags);
607 	req->in.h.unique = unique;
608 	spin_lock(&fiq->waitq.lock);
609 	if (fiq->connected) {
610 		queue_request(fiq, req);
611 		err = 0;
612 	}
613 	spin_unlock(&fiq->waitq.lock);
614 
615 	return err;
616 }
617 
618 void fuse_force_forget(struct file *file, u64 nodeid)
619 {
620 	struct inode *inode = file_inode(file);
621 	struct fuse_conn *fc = get_fuse_conn(inode);
622 	struct fuse_req *req;
623 	struct fuse_forget_in inarg;
624 
625 	memset(&inarg, 0, sizeof(inarg));
626 	inarg.nlookup = 1;
627 	req = fuse_get_req_nofail_nopages(fc, file);
628 	req->in.h.opcode = FUSE_FORGET;
629 	req->in.h.nodeid = nodeid;
630 	req->in.numargs = 1;
631 	req->in.args[0].size = sizeof(inarg);
632 	req->in.args[0].value = &inarg;
633 	__clear_bit(FR_ISREPLY, &req->flags);
634 	__fuse_request_send(fc, req);
635 	/* ignore errors */
636 	fuse_put_request(fc, req);
637 }
638 
639 /*
640  * Lock the request.  Up to the next unlock_request() there mustn't be
641  * anything that could cause a page-fault.  If the request was already
642  * aborted bail out.
643  */
644 static int lock_request(struct fuse_req *req)
645 {
646 	int err = 0;
647 	if (req) {
648 		spin_lock(&req->waitq.lock);
649 		if (test_bit(FR_ABORTED, &req->flags))
650 			err = -ENOENT;
651 		else
652 			set_bit(FR_LOCKED, &req->flags);
653 		spin_unlock(&req->waitq.lock);
654 	}
655 	return err;
656 }
657 
658 /*
659  * Unlock request.  If it was aborted while locked, caller is responsible
660  * for unlocking and ending the request.
661  */
662 static int unlock_request(struct fuse_req *req)
663 {
664 	int err = 0;
665 	if (req) {
666 		spin_lock(&req->waitq.lock);
667 		if (test_bit(FR_ABORTED, &req->flags))
668 			err = -ENOENT;
669 		else
670 			clear_bit(FR_LOCKED, &req->flags);
671 		spin_unlock(&req->waitq.lock);
672 	}
673 	return err;
674 }
675 
676 struct fuse_copy_state {
677 	int write;
678 	struct fuse_req *req;
679 	struct iov_iter *iter;
680 	struct pipe_buffer *pipebufs;
681 	struct pipe_buffer *currbuf;
682 	struct pipe_inode_info *pipe;
683 	unsigned long nr_segs;
684 	struct page *pg;
685 	unsigned len;
686 	unsigned offset;
687 	unsigned move_pages:1;
688 };
689 
690 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
691 			   struct iov_iter *iter)
692 {
693 	memset(cs, 0, sizeof(*cs));
694 	cs->write = write;
695 	cs->iter = iter;
696 }
697 
698 /* Unmap and put previous page of userspace buffer */
699 static void fuse_copy_finish(struct fuse_copy_state *cs)
700 {
701 	if (cs->currbuf) {
702 		struct pipe_buffer *buf = cs->currbuf;
703 
704 		if (cs->write)
705 			buf->len = PAGE_SIZE - cs->len;
706 		cs->currbuf = NULL;
707 	} else if (cs->pg) {
708 		if (cs->write) {
709 			flush_dcache_page(cs->pg);
710 			set_page_dirty_lock(cs->pg);
711 		}
712 		put_page(cs->pg);
713 	}
714 	cs->pg = NULL;
715 }
716 
717 /*
718  * Get another pagefull of userspace buffer, and map it to kernel
719  * address space, and lock request
720  */
721 static int fuse_copy_fill(struct fuse_copy_state *cs)
722 {
723 	struct page *page;
724 	int err;
725 
726 	err = unlock_request(cs->req);
727 	if (err)
728 		return err;
729 
730 	fuse_copy_finish(cs);
731 	if (cs->pipebufs) {
732 		struct pipe_buffer *buf = cs->pipebufs;
733 
734 		if (!cs->write) {
735 			err = pipe_buf_confirm(cs->pipe, buf);
736 			if (err)
737 				return err;
738 
739 			BUG_ON(!cs->nr_segs);
740 			cs->currbuf = buf;
741 			cs->pg = buf->page;
742 			cs->offset = buf->offset;
743 			cs->len = buf->len;
744 			cs->pipebufs++;
745 			cs->nr_segs--;
746 		} else {
747 			if (cs->nr_segs == cs->pipe->buffers)
748 				return -EIO;
749 
750 			page = alloc_page(GFP_HIGHUSER);
751 			if (!page)
752 				return -ENOMEM;
753 
754 			buf->page = page;
755 			buf->offset = 0;
756 			buf->len = 0;
757 
758 			cs->currbuf = buf;
759 			cs->pg = page;
760 			cs->offset = 0;
761 			cs->len = PAGE_SIZE;
762 			cs->pipebufs++;
763 			cs->nr_segs++;
764 		}
765 	} else {
766 		size_t off;
767 		err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
768 		if (err < 0)
769 			return err;
770 		BUG_ON(!err);
771 		cs->len = err;
772 		cs->offset = off;
773 		cs->pg = page;
774 		iov_iter_advance(cs->iter, err);
775 	}
776 
777 	return lock_request(cs->req);
778 }
779 
780 /* Do as much copy to/from userspace buffer as we can */
781 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
782 {
783 	unsigned ncpy = min(*size, cs->len);
784 	if (val) {
785 		void *pgaddr = kmap_atomic(cs->pg);
786 		void *buf = pgaddr + cs->offset;
787 
788 		if (cs->write)
789 			memcpy(buf, *val, ncpy);
790 		else
791 			memcpy(*val, buf, ncpy);
792 
793 		kunmap_atomic(pgaddr);
794 		*val += ncpy;
795 	}
796 	*size -= ncpy;
797 	cs->len -= ncpy;
798 	cs->offset += ncpy;
799 	return ncpy;
800 }
801 
802 static int fuse_check_page(struct page *page)
803 {
804 	if (page_mapcount(page) ||
805 	    page->mapping != NULL ||
806 	    page_count(page) != 1 ||
807 	    (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
808 	     ~(1 << PG_locked |
809 	       1 << PG_referenced |
810 	       1 << PG_uptodate |
811 	       1 << PG_lru |
812 	       1 << PG_active |
813 	       1 << PG_reclaim))) {
814 		printk(KERN_WARNING "fuse: trying to steal weird page\n");
815 		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);
816 		return 1;
817 	}
818 	return 0;
819 }
820 
821 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
822 {
823 	int err;
824 	struct page *oldpage = *pagep;
825 	struct page *newpage;
826 	struct pipe_buffer *buf = cs->pipebufs;
827 
828 	err = unlock_request(cs->req);
829 	if (err)
830 		return err;
831 
832 	fuse_copy_finish(cs);
833 
834 	err = pipe_buf_confirm(cs->pipe, buf);
835 	if (err)
836 		return err;
837 
838 	BUG_ON(!cs->nr_segs);
839 	cs->currbuf = buf;
840 	cs->len = buf->len;
841 	cs->pipebufs++;
842 	cs->nr_segs--;
843 
844 	if (cs->len != PAGE_SIZE)
845 		goto out_fallback;
846 
847 	if (pipe_buf_steal(cs->pipe, buf) != 0)
848 		goto out_fallback;
849 
850 	newpage = buf->page;
851 
852 	if (!PageUptodate(newpage))
853 		SetPageUptodate(newpage);
854 
855 	ClearPageMappedToDisk(newpage);
856 
857 	if (fuse_check_page(newpage) != 0)
858 		goto out_fallback_unlock;
859 
860 	/*
861 	 * This is a new and locked page, it shouldn't be mapped or
862 	 * have any special flags on it
863 	 */
864 	if (WARN_ON(page_mapped(oldpage)))
865 		goto out_fallback_unlock;
866 	if (WARN_ON(page_has_private(oldpage)))
867 		goto out_fallback_unlock;
868 	if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
869 		goto out_fallback_unlock;
870 	if (WARN_ON(PageMlocked(oldpage)))
871 		goto out_fallback_unlock;
872 
873 	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
874 	if (err) {
875 		unlock_page(newpage);
876 		return err;
877 	}
878 
879 	get_page(newpage);
880 
881 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
882 		lru_cache_add_file(newpage);
883 
884 	err = 0;
885 	spin_lock(&cs->req->waitq.lock);
886 	if (test_bit(FR_ABORTED, &cs->req->flags))
887 		err = -ENOENT;
888 	else
889 		*pagep = newpage;
890 	spin_unlock(&cs->req->waitq.lock);
891 
892 	if (err) {
893 		unlock_page(newpage);
894 		put_page(newpage);
895 		return err;
896 	}
897 
898 	unlock_page(oldpage);
899 	put_page(oldpage);
900 	cs->len = 0;
901 
902 	return 0;
903 
904 out_fallback_unlock:
905 	unlock_page(newpage);
906 out_fallback:
907 	cs->pg = buf->page;
908 	cs->offset = buf->offset;
909 
910 	err = lock_request(cs->req);
911 	if (err)
912 		return err;
913 
914 	return 1;
915 }
916 
917 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
918 			 unsigned offset, unsigned count)
919 {
920 	struct pipe_buffer *buf;
921 	int err;
922 
923 	if (cs->nr_segs == cs->pipe->buffers)
924 		return -EIO;
925 
926 	err = unlock_request(cs->req);
927 	if (err)
928 		return err;
929 
930 	fuse_copy_finish(cs);
931 
932 	buf = cs->pipebufs;
933 	get_page(page);
934 	buf->page = page;
935 	buf->offset = offset;
936 	buf->len = count;
937 
938 	cs->pipebufs++;
939 	cs->nr_segs++;
940 	cs->len = 0;
941 
942 	return 0;
943 }
944 
945 /*
946  * Copy a page in the request to/from the userspace buffer.  Must be
947  * done atomically
948  */
949 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
950 			  unsigned offset, unsigned count, int zeroing)
951 {
952 	int err;
953 	struct page *page = *pagep;
954 
955 	if (page && zeroing && count < PAGE_SIZE)
956 		clear_highpage(page);
957 
958 	while (count) {
959 		if (cs->write && cs->pipebufs && page) {
960 			return fuse_ref_page(cs, page, offset, count);
961 		} else if (!cs->len) {
962 			if (cs->move_pages && page &&
963 			    offset == 0 && count == PAGE_SIZE) {
964 				err = fuse_try_move_page(cs, pagep);
965 				if (err <= 0)
966 					return err;
967 			} else {
968 				err = fuse_copy_fill(cs);
969 				if (err)
970 					return err;
971 			}
972 		}
973 		if (page) {
974 			void *mapaddr = kmap_atomic(page);
975 			void *buf = mapaddr + offset;
976 			offset += fuse_copy_do(cs, &buf, &count);
977 			kunmap_atomic(mapaddr);
978 		} else
979 			offset += fuse_copy_do(cs, NULL, &count);
980 	}
981 	if (page && !cs->write)
982 		flush_dcache_page(page);
983 	return 0;
984 }
985 
986 /* Copy pages in the request to/from userspace buffer */
987 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
988 			   int zeroing)
989 {
990 	unsigned i;
991 	struct fuse_req *req = cs->req;
992 
993 	for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
994 		int err;
995 		unsigned offset = req->page_descs[i].offset;
996 		unsigned count = min(nbytes, req->page_descs[i].length);
997 
998 		err = fuse_copy_page(cs, &req->pages[i], offset, count,
999 				     zeroing);
1000 		if (err)
1001 			return err;
1002 
1003 		nbytes -= count;
1004 	}
1005 	return 0;
1006 }
1007 
1008 /* Copy a single argument in the request to/from userspace buffer */
1009 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1010 {
1011 	while (size) {
1012 		if (!cs->len) {
1013 			int err = fuse_copy_fill(cs);
1014 			if (err)
1015 				return err;
1016 		}
1017 		fuse_copy_do(cs, &val, &size);
1018 	}
1019 	return 0;
1020 }
1021 
1022 /* Copy request arguments to/from userspace buffer */
1023 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1024 			  unsigned argpages, struct fuse_arg *args,
1025 			  int zeroing)
1026 {
1027 	int err = 0;
1028 	unsigned i;
1029 
1030 	for (i = 0; !err && i < numargs; i++)  {
1031 		struct fuse_arg *arg = &args[i];
1032 		if (i == numargs - 1 && argpages)
1033 			err = fuse_copy_pages(cs, arg->size, zeroing);
1034 		else
1035 			err = fuse_copy_one(cs, arg->value, arg->size);
1036 	}
1037 	return err;
1038 }
1039 
1040 static int forget_pending(struct fuse_iqueue *fiq)
1041 {
1042 	return fiq->forget_list_head.next != NULL;
1043 }
1044 
1045 static int request_pending(struct fuse_iqueue *fiq)
1046 {
1047 	return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1048 		forget_pending(fiq);
1049 }
1050 
1051 /*
1052  * Transfer an interrupt request to userspace
1053  *
1054  * Unlike other requests this is assembled on demand, without a need
1055  * to allocate a separate fuse_req structure.
1056  *
1057  * Called with fiq->waitq.lock held, releases it
1058  */
1059 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1060 			       struct fuse_copy_state *cs,
1061 			       size_t nbytes, struct fuse_req *req)
1062 __releases(fiq->waitq.lock)
1063 {
1064 	struct fuse_in_header ih;
1065 	struct fuse_interrupt_in arg;
1066 	unsigned reqsize = sizeof(ih) + sizeof(arg);
1067 	int err;
1068 
1069 	list_del_init(&req->intr_entry);
1070 	req->intr_unique = fuse_get_unique(fiq);
1071 	memset(&ih, 0, sizeof(ih));
1072 	memset(&arg, 0, sizeof(arg));
1073 	ih.len = reqsize;
1074 	ih.opcode = FUSE_INTERRUPT;
1075 	ih.unique = req->intr_unique;
1076 	arg.unique = req->in.h.unique;
1077 
1078 	spin_unlock(&fiq->waitq.lock);
1079 	if (nbytes < reqsize)
1080 		return -EINVAL;
1081 
1082 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1083 	if (!err)
1084 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1085 	fuse_copy_finish(cs);
1086 
1087 	return err ? err : reqsize;
1088 }
1089 
1090 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1091 					       unsigned max,
1092 					       unsigned *countp)
1093 {
1094 	struct fuse_forget_link *head = fiq->forget_list_head.next;
1095 	struct fuse_forget_link **newhead = &head;
1096 	unsigned count;
1097 
1098 	for (count = 0; *newhead != NULL && count < max; count++)
1099 		newhead = &(*newhead)->next;
1100 
1101 	fiq->forget_list_head.next = *newhead;
1102 	*newhead = NULL;
1103 	if (fiq->forget_list_head.next == NULL)
1104 		fiq->forget_list_tail = &fiq->forget_list_head;
1105 
1106 	if (countp != NULL)
1107 		*countp = count;
1108 
1109 	return head;
1110 }
1111 
1112 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1113 				   struct fuse_copy_state *cs,
1114 				   size_t nbytes)
1115 __releases(fiq->waitq.lock)
1116 {
1117 	int err;
1118 	struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1119 	struct fuse_forget_in arg = {
1120 		.nlookup = forget->forget_one.nlookup,
1121 	};
1122 	struct fuse_in_header ih = {
1123 		.opcode = FUSE_FORGET,
1124 		.nodeid = forget->forget_one.nodeid,
1125 		.unique = fuse_get_unique(fiq),
1126 		.len = sizeof(ih) + sizeof(arg),
1127 	};
1128 
1129 	spin_unlock(&fiq->waitq.lock);
1130 	kfree(forget);
1131 	if (nbytes < ih.len)
1132 		return -EINVAL;
1133 
1134 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1135 	if (!err)
1136 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1137 	fuse_copy_finish(cs);
1138 
1139 	if (err)
1140 		return err;
1141 
1142 	return ih.len;
1143 }
1144 
1145 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1146 				   struct fuse_copy_state *cs, size_t nbytes)
1147 __releases(fiq->waitq.lock)
1148 {
1149 	int err;
1150 	unsigned max_forgets;
1151 	unsigned count;
1152 	struct fuse_forget_link *head;
1153 	struct fuse_batch_forget_in arg = { .count = 0 };
1154 	struct fuse_in_header ih = {
1155 		.opcode = FUSE_BATCH_FORGET,
1156 		.unique = fuse_get_unique(fiq),
1157 		.len = sizeof(ih) + sizeof(arg),
1158 	};
1159 
1160 	if (nbytes < ih.len) {
1161 		spin_unlock(&fiq->waitq.lock);
1162 		return -EINVAL;
1163 	}
1164 
1165 	max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1166 	head = dequeue_forget(fiq, max_forgets, &count);
1167 	spin_unlock(&fiq->waitq.lock);
1168 
1169 	arg.count = count;
1170 	ih.len += count * sizeof(struct fuse_forget_one);
1171 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1172 	if (!err)
1173 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1174 
1175 	while (head) {
1176 		struct fuse_forget_link *forget = head;
1177 
1178 		if (!err) {
1179 			err = fuse_copy_one(cs, &forget->forget_one,
1180 					    sizeof(forget->forget_one));
1181 		}
1182 		head = forget->next;
1183 		kfree(forget);
1184 	}
1185 
1186 	fuse_copy_finish(cs);
1187 
1188 	if (err)
1189 		return err;
1190 
1191 	return ih.len;
1192 }
1193 
1194 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1195 			    struct fuse_copy_state *cs,
1196 			    size_t nbytes)
1197 __releases(fiq->waitq.lock)
1198 {
1199 	if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1200 		return fuse_read_single_forget(fiq, cs, nbytes);
1201 	else
1202 		return fuse_read_batch_forget(fiq, cs, nbytes);
1203 }
1204 
1205 /*
1206  * Read a single request into the userspace filesystem's buffer.  This
1207  * function waits until a request is available, then removes it from
1208  * the pending list and copies request data to userspace buffer.  If
1209  * no reply is needed (FORGET) or request has been aborted or there
1210  * was an error during the copying then it's finished by calling
1211  * request_end().  Otherwise add it to the processing list, and set
1212  * the 'sent' flag.
1213  */
1214 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1215 				struct fuse_copy_state *cs, size_t nbytes)
1216 {
1217 	ssize_t err;
1218 	struct fuse_conn *fc = fud->fc;
1219 	struct fuse_iqueue *fiq = &fc->iq;
1220 	struct fuse_pqueue *fpq = &fud->pq;
1221 	struct fuse_req *req;
1222 	struct fuse_in *in;
1223 	unsigned reqsize;
1224 
1225  restart:
1226 	spin_lock(&fiq->waitq.lock);
1227 	err = -EAGAIN;
1228 	if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1229 	    !request_pending(fiq))
1230 		goto err_unlock;
1231 
1232 	err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1233 				!fiq->connected || request_pending(fiq));
1234 	if (err)
1235 		goto err_unlock;
1236 
1237 	err = -ENODEV;
1238 	if (!fiq->connected)
1239 		goto err_unlock;
1240 
1241 	if (!list_empty(&fiq->interrupts)) {
1242 		req = list_entry(fiq->interrupts.next, struct fuse_req,
1243 				 intr_entry);
1244 		return fuse_read_interrupt(fiq, cs, nbytes, req);
1245 	}
1246 
1247 	if (forget_pending(fiq)) {
1248 		if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1249 			return fuse_read_forget(fc, fiq, cs, nbytes);
1250 
1251 		if (fiq->forget_batch <= -8)
1252 			fiq->forget_batch = 16;
1253 	}
1254 
1255 	req = list_entry(fiq->pending.next, struct fuse_req, list);
1256 	clear_bit(FR_PENDING, &req->flags);
1257 	list_del_init(&req->list);
1258 	spin_unlock(&fiq->waitq.lock);
1259 
1260 	in = &req->in;
1261 	reqsize = in->h.len;
1262 
1263 	if (task_active_pid_ns(current) != fc->pid_ns) {
1264 		rcu_read_lock();
1265 		in->h.pid = pid_vnr(find_pid_ns(in->h.pid, fc->pid_ns));
1266 		rcu_read_unlock();
1267 	}
1268 
1269 	/* If request is too large, reply with an error and restart the read */
1270 	if (nbytes < reqsize) {
1271 		req->out.h.error = -EIO;
1272 		/* SETXATTR is special, since it may contain too large data */
1273 		if (in->h.opcode == FUSE_SETXATTR)
1274 			req->out.h.error = -E2BIG;
1275 		request_end(fc, req);
1276 		goto restart;
1277 	}
1278 	spin_lock(&fpq->lock);
1279 	list_add(&req->list, &fpq->io);
1280 	spin_unlock(&fpq->lock);
1281 	cs->req = req;
1282 	err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1283 	if (!err)
1284 		err = fuse_copy_args(cs, in->numargs, in->argpages,
1285 				     (struct fuse_arg *) in->args, 0);
1286 	fuse_copy_finish(cs);
1287 	spin_lock(&fpq->lock);
1288 	clear_bit(FR_LOCKED, &req->flags);
1289 	if (!fpq->connected) {
1290 		err = -ENODEV;
1291 		goto out_end;
1292 	}
1293 	if (err) {
1294 		req->out.h.error = -EIO;
1295 		goto out_end;
1296 	}
1297 	if (!test_bit(FR_ISREPLY, &req->flags)) {
1298 		err = reqsize;
1299 		goto out_end;
1300 	}
1301 	list_move_tail(&req->list, &fpq->processing);
1302 	spin_unlock(&fpq->lock);
1303 	set_bit(FR_SENT, &req->flags);
1304 	/* matches barrier in request_wait_answer() */
1305 	smp_mb__after_atomic();
1306 	if (test_bit(FR_INTERRUPTED, &req->flags))
1307 		queue_interrupt(fiq, req);
1308 
1309 	return reqsize;
1310 
1311 out_end:
1312 	if (!test_bit(FR_PRIVATE, &req->flags))
1313 		list_del_init(&req->list);
1314 	spin_unlock(&fpq->lock);
1315 	request_end(fc, req);
1316 	return err;
1317 
1318  err_unlock:
1319 	spin_unlock(&fiq->waitq.lock);
1320 	return err;
1321 }
1322 
1323 static int fuse_dev_open(struct inode *inode, struct file *file)
1324 {
1325 	/*
1326 	 * The fuse device's file's private_data is used to hold
1327 	 * the fuse_conn(ection) when it is mounted, and is used to
1328 	 * keep track of whether the file has been mounted already.
1329 	 */
1330 	file->private_data = NULL;
1331 	return 0;
1332 }
1333 
1334 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1335 {
1336 	struct fuse_copy_state cs;
1337 	struct file *file = iocb->ki_filp;
1338 	struct fuse_dev *fud = fuse_get_dev(file);
1339 
1340 	if (!fud)
1341 		return -EPERM;
1342 
1343 	if (!iter_is_iovec(to))
1344 		return -EINVAL;
1345 
1346 	fuse_copy_init(&cs, 1, to);
1347 
1348 	return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1349 }
1350 
1351 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1352 				    struct pipe_inode_info *pipe,
1353 				    size_t len, unsigned int flags)
1354 {
1355 	int total, ret;
1356 	int page_nr = 0;
1357 	struct pipe_buffer *bufs;
1358 	struct fuse_copy_state cs;
1359 	struct fuse_dev *fud = fuse_get_dev(in);
1360 
1361 	if (!fud)
1362 		return -EPERM;
1363 
1364 	bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1365 	if (!bufs)
1366 		return -ENOMEM;
1367 
1368 	fuse_copy_init(&cs, 1, NULL);
1369 	cs.pipebufs = bufs;
1370 	cs.pipe = pipe;
1371 	ret = fuse_dev_do_read(fud, in, &cs, len);
1372 	if (ret < 0)
1373 		goto out;
1374 
1375 	if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1376 		ret = -EIO;
1377 		goto out;
1378 	}
1379 
1380 	for (ret = total = 0; page_nr < cs.nr_segs; total += ret) {
1381 		/*
1382 		 * Need to be careful about this.  Having buf->ops in module
1383 		 * code can Oops if the buffer persists after module unload.
1384 		 */
1385 		bufs[page_nr].ops = &nosteal_pipe_buf_ops;
1386 		bufs[page_nr].flags = 0;
1387 		ret = add_to_pipe(pipe, &bufs[page_nr++]);
1388 		if (unlikely(ret < 0))
1389 			break;
1390 	}
1391 	if (total)
1392 		ret = total;
1393 out:
1394 	for (; page_nr < cs.nr_segs; page_nr++)
1395 		put_page(bufs[page_nr].page);
1396 
1397 	kfree(bufs);
1398 	return ret;
1399 }
1400 
1401 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1402 			    struct fuse_copy_state *cs)
1403 {
1404 	struct fuse_notify_poll_wakeup_out outarg;
1405 	int err = -EINVAL;
1406 
1407 	if (size != sizeof(outarg))
1408 		goto err;
1409 
1410 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1411 	if (err)
1412 		goto err;
1413 
1414 	fuse_copy_finish(cs);
1415 	return fuse_notify_poll_wakeup(fc, &outarg);
1416 
1417 err:
1418 	fuse_copy_finish(cs);
1419 	return err;
1420 }
1421 
1422 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1423 				   struct fuse_copy_state *cs)
1424 {
1425 	struct fuse_notify_inval_inode_out outarg;
1426 	int err = -EINVAL;
1427 
1428 	if (size != sizeof(outarg))
1429 		goto err;
1430 
1431 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1432 	if (err)
1433 		goto err;
1434 	fuse_copy_finish(cs);
1435 
1436 	down_read(&fc->killsb);
1437 	err = -ENOENT;
1438 	if (fc->sb) {
1439 		err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1440 					       outarg.off, outarg.len);
1441 	}
1442 	up_read(&fc->killsb);
1443 	return err;
1444 
1445 err:
1446 	fuse_copy_finish(cs);
1447 	return err;
1448 }
1449 
1450 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1451 				   struct fuse_copy_state *cs)
1452 {
1453 	struct fuse_notify_inval_entry_out outarg;
1454 	int err = -ENOMEM;
1455 	char *buf;
1456 	struct qstr name;
1457 
1458 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1459 	if (!buf)
1460 		goto err;
1461 
1462 	err = -EINVAL;
1463 	if (size < sizeof(outarg))
1464 		goto err;
1465 
1466 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1467 	if (err)
1468 		goto err;
1469 
1470 	err = -ENAMETOOLONG;
1471 	if (outarg.namelen > FUSE_NAME_MAX)
1472 		goto err;
1473 
1474 	err = -EINVAL;
1475 	if (size != sizeof(outarg) + outarg.namelen + 1)
1476 		goto err;
1477 
1478 	name.name = buf;
1479 	name.len = outarg.namelen;
1480 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1481 	if (err)
1482 		goto err;
1483 	fuse_copy_finish(cs);
1484 	buf[outarg.namelen] = 0;
1485 
1486 	down_read(&fc->killsb);
1487 	err = -ENOENT;
1488 	if (fc->sb)
1489 		err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1490 	up_read(&fc->killsb);
1491 	kfree(buf);
1492 	return err;
1493 
1494 err:
1495 	kfree(buf);
1496 	fuse_copy_finish(cs);
1497 	return err;
1498 }
1499 
1500 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1501 			      struct fuse_copy_state *cs)
1502 {
1503 	struct fuse_notify_delete_out outarg;
1504 	int err = -ENOMEM;
1505 	char *buf;
1506 	struct qstr name;
1507 
1508 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1509 	if (!buf)
1510 		goto err;
1511 
1512 	err = -EINVAL;
1513 	if (size < sizeof(outarg))
1514 		goto err;
1515 
1516 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1517 	if (err)
1518 		goto err;
1519 
1520 	err = -ENAMETOOLONG;
1521 	if (outarg.namelen > FUSE_NAME_MAX)
1522 		goto err;
1523 
1524 	err = -EINVAL;
1525 	if (size != sizeof(outarg) + outarg.namelen + 1)
1526 		goto err;
1527 
1528 	name.name = buf;
1529 	name.len = outarg.namelen;
1530 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1531 	if (err)
1532 		goto err;
1533 	fuse_copy_finish(cs);
1534 	buf[outarg.namelen] = 0;
1535 
1536 	down_read(&fc->killsb);
1537 	err = -ENOENT;
1538 	if (fc->sb)
1539 		err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1540 					       outarg.child, &name);
1541 	up_read(&fc->killsb);
1542 	kfree(buf);
1543 	return err;
1544 
1545 err:
1546 	kfree(buf);
1547 	fuse_copy_finish(cs);
1548 	return err;
1549 }
1550 
1551 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1552 			     struct fuse_copy_state *cs)
1553 {
1554 	struct fuse_notify_store_out outarg;
1555 	struct inode *inode;
1556 	struct address_space *mapping;
1557 	u64 nodeid;
1558 	int err;
1559 	pgoff_t index;
1560 	unsigned int offset;
1561 	unsigned int num;
1562 	loff_t file_size;
1563 	loff_t end;
1564 
1565 	err = -EINVAL;
1566 	if (size < sizeof(outarg))
1567 		goto out_finish;
1568 
1569 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1570 	if (err)
1571 		goto out_finish;
1572 
1573 	err = -EINVAL;
1574 	if (size - sizeof(outarg) != outarg.size)
1575 		goto out_finish;
1576 
1577 	nodeid = outarg.nodeid;
1578 
1579 	down_read(&fc->killsb);
1580 
1581 	err = -ENOENT;
1582 	if (!fc->sb)
1583 		goto out_up_killsb;
1584 
1585 	inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1586 	if (!inode)
1587 		goto out_up_killsb;
1588 
1589 	mapping = inode->i_mapping;
1590 	index = outarg.offset >> PAGE_SHIFT;
1591 	offset = outarg.offset & ~PAGE_MASK;
1592 	file_size = i_size_read(inode);
1593 	end = outarg.offset + outarg.size;
1594 	if (end > file_size) {
1595 		file_size = end;
1596 		fuse_write_update_size(inode, file_size);
1597 	}
1598 
1599 	num = outarg.size;
1600 	while (num) {
1601 		struct page *page;
1602 		unsigned int this_num;
1603 
1604 		err = -ENOMEM;
1605 		page = find_or_create_page(mapping, index,
1606 					   mapping_gfp_mask(mapping));
1607 		if (!page)
1608 			goto out_iput;
1609 
1610 		this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1611 		err = fuse_copy_page(cs, &page, offset, this_num, 0);
1612 		if (!err && offset == 0 &&
1613 		    (this_num == PAGE_SIZE || file_size == end))
1614 			SetPageUptodate(page);
1615 		unlock_page(page);
1616 		put_page(page);
1617 
1618 		if (err)
1619 			goto out_iput;
1620 
1621 		num -= this_num;
1622 		offset = 0;
1623 		index++;
1624 	}
1625 
1626 	err = 0;
1627 
1628 out_iput:
1629 	iput(inode);
1630 out_up_killsb:
1631 	up_read(&fc->killsb);
1632 out_finish:
1633 	fuse_copy_finish(cs);
1634 	return err;
1635 }
1636 
1637 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1638 {
1639 	release_pages(req->pages, req->num_pages);
1640 }
1641 
1642 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1643 			 struct fuse_notify_retrieve_out *outarg)
1644 {
1645 	int err;
1646 	struct address_space *mapping = inode->i_mapping;
1647 	struct fuse_req *req;
1648 	pgoff_t index;
1649 	loff_t file_size;
1650 	unsigned int num;
1651 	unsigned int offset;
1652 	size_t total_len = 0;
1653 	int num_pages;
1654 
1655 	offset = outarg->offset & ~PAGE_MASK;
1656 	file_size = i_size_read(inode);
1657 
1658 	num = outarg->size;
1659 	if (outarg->offset > file_size)
1660 		num = 0;
1661 	else if (outarg->offset + num > file_size)
1662 		num = file_size - outarg->offset;
1663 
1664 	num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1665 	num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1666 
1667 	req = fuse_get_req(fc, num_pages);
1668 	if (IS_ERR(req))
1669 		return PTR_ERR(req);
1670 
1671 	req->in.h.opcode = FUSE_NOTIFY_REPLY;
1672 	req->in.h.nodeid = outarg->nodeid;
1673 	req->in.numargs = 2;
1674 	req->in.argpages = 1;
1675 	req->page_descs[0].offset = offset;
1676 	req->end = fuse_retrieve_end;
1677 
1678 	index = outarg->offset >> PAGE_SHIFT;
1679 
1680 	while (num && req->num_pages < num_pages) {
1681 		struct page *page;
1682 		unsigned int this_num;
1683 
1684 		page = find_get_page(mapping, index);
1685 		if (!page)
1686 			break;
1687 
1688 		this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1689 		req->pages[req->num_pages] = page;
1690 		req->page_descs[req->num_pages].length = this_num;
1691 		req->num_pages++;
1692 
1693 		offset = 0;
1694 		num -= this_num;
1695 		total_len += this_num;
1696 		index++;
1697 	}
1698 	req->misc.retrieve_in.offset = outarg->offset;
1699 	req->misc.retrieve_in.size = total_len;
1700 	req->in.args[0].size = sizeof(req->misc.retrieve_in);
1701 	req->in.args[0].value = &req->misc.retrieve_in;
1702 	req->in.args[1].size = total_len;
1703 
1704 	err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1705 	if (err)
1706 		fuse_retrieve_end(fc, req);
1707 
1708 	return err;
1709 }
1710 
1711 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1712 				struct fuse_copy_state *cs)
1713 {
1714 	struct fuse_notify_retrieve_out outarg;
1715 	struct inode *inode;
1716 	int err;
1717 
1718 	err = -EINVAL;
1719 	if (size != sizeof(outarg))
1720 		goto copy_finish;
1721 
1722 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1723 	if (err)
1724 		goto copy_finish;
1725 
1726 	fuse_copy_finish(cs);
1727 
1728 	down_read(&fc->killsb);
1729 	err = -ENOENT;
1730 	if (fc->sb) {
1731 		u64 nodeid = outarg.nodeid;
1732 
1733 		inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1734 		if (inode) {
1735 			err = fuse_retrieve(fc, inode, &outarg);
1736 			iput(inode);
1737 		}
1738 	}
1739 	up_read(&fc->killsb);
1740 
1741 	return err;
1742 
1743 copy_finish:
1744 	fuse_copy_finish(cs);
1745 	return err;
1746 }
1747 
1748 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1749 		       unsigned int size, struct fuse_copy_state *cs)
1750 {
1751 	/* Don't try to move pages (yet) */
1752 	cs->move_pages = 0;
1753 
1754 	switch (code) {
1755 	case FUSE_NOTIFY_POLL:
1756 		return fuse_notify_poll(fc, size, cs);
1757 
1758 	case FUSE_NOTIFY_INVAL_INODE:
1759 		return fuse_notify_inval_inode(fc, size, cs);
1760 
1761 	case FUSE_NOTIFY_INVAL_ENTRY:
1762 		return fuse_notify_inval_entry(fc, size, cs);
1763 
1764 	case FUSE_NOTIFY_STORE:
1765 		return fuse_notify_store(fc, size, cs);
1766 
1767 	case FUSE_NOTIFY_RETRIEVE:
1768 		return fuse_notify_retrieve(fc, size, cs);
1769 
1770 	case FUSE_NOTIFY_DELETE:
1771 		return fuse_notify_delete(fc, size, cs);
1772 
1773 	default:
1774 		fuse_copy_finish(cs);
1775 		return -EINVAL;
1776 	}
1777 }
1778 
1779 /* Look up request on processing list by unique ID */
1780 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1781 {
1782 	struct fuse_req *req;
1783 
1784 	list_for_each_entry(req, &fpq->processing, list) {
1785 		if (req->in.h.unique == unique || req->intr_unique == unique)
1786 			return req;
1787 	}
1788 	return NULL;
1789 }
1790 
1791 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1792 			 unsigned nbytes)
1793 {
1794 	unsigned reqsize = sizeof(struct fuse_out_header);
1795 
1796 	if (out->h.error)
1797 		return nbytes != reqsize ? -EINVAL : 0;
1798 
1799 	reqsize += len_args(out->numargs, out->args);
1800 
1801 	if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1802 		return -EINVAL;
1803 	else if (reqsize > nbytes) {
1804 		struct fuse_arg *lastarg = &out->args[out->numargs-1];
1805 		unsigned diffsize = reqsize - nbytes;
1806 		if (diffsize > lastarg->size)
1807 			return -EINVAL;
1808 		lastarg->size -= diffsize;
1809 	}
1810 	return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1811 			      out->page_zeroing);
1812 }
1813 
1814 /*
1815  * Write a single reply to a request.  First the header is copied from
1816  * the write buffer.  The request is then searched on the processing
1817  * list by the unique ID found in the header.  If found, then remove
1818  * it from the list and copy the rest of the buffer to the request.
1819  * The request is finished by calling request_end()
1820  */
1821 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1822 				 struct fuse_copy_state *cs, size_t nbytes)
1823 {
1824 	int err;
1825 	struct fuse_conn *fc = fud->fc;
1826 	struct fuse_pqueue *fpq = &fud->pq;
1827 	struct fuse_req *req;
1828 	struct fuse_out_header oh;
1829 
1830 	if (nbytes < sizeof(struct fuse_out_header))
1831 		return -EINVAL;
1832 
1833 	err = fuse_copy_one(cs, &oh, sizeof(oh));
1834 	if (err)
1835 		goto err_finish;
1836 
1837 	err = -EINVAL;
1838 	if (oh.len != nbytes)
1839 		goto err_finish;
1840 
1841 	/*
1842 	 * Zero oh.unique indicates unsolicited notification message
1843 	 * and error contains notification code.
1844 	 */
1845 	if (!oh.unique) {
1846 		err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1847 		return err ? err : nbytes;
1848 	}
1849 
1850 	err = -EINVAL;
1851 	if (oh.error <= -1000 || oh.error > 0)
1852 		goto err_finish;
1853 
1854 	spin_lock(&fpq->lock);
1855 	err = -ENOENT;
1856 	if (!fpq->connected)
1857 		goto err_unlock_pq;
1858 
1859 	req = request_find(fpq, oh.unique);
1860 	if (!req)
1861 		goto err_unlock_pq;
1862 
1863 	/* Is it an interrupt reply? */
1864 	if (req->intr_unique == oh.unique) {
1865 		spin_unlock(&fpq->lock);
1866 
1867 		err = -EINVAL;
1868 		if (nbytes != sizeof(struct fuse_out_header))
1869 			goto err_finish;
1870 
1871 		if (oh.error == -ENOSYS)
1872 			fc->no_interrupt = 1;
1873 		else if (oh.error == -EAGAIN)
1874 			queue_interrupt(&fc->iq, req);
1875 
1876 		fuse_copy_finish(cs);
1877 		return nbytes;
1878 	}
1879 
1880 	clear_bit(FR_SENT, &req->flags);
1881 	list_move(&req->list, &fpq->io);
1882 	req->out.h = oh;
1883 	set_bit(FR_LOCKED, &req->flags);
1884 	spin_unlock(&fpq->lock);
1885 	cs->req = req;
1886 	if (!req->out.page_replace)
1887 		cs->move_pages = 0;
1888 
1889 	err = copy_out_args(cs, &req->out, nbytes);
1890 	fuse_copy_finish(cs);
1891 
1892 	spin_lock(&fpq->lock);
1893 	clear_bit(FR_LOCKED, &req->flags);
1894 	if (!fpq->connected)
1895 		err = -ENOENT;
1896 	else if (err)
1897 		req->out.h.error = -EIO;
1898 	if (!test_bit(FR_PRIVATE, &req->flags))
1899 		list_del_init(&req->list);
1900 	spin_unlock(&fpq->lock);
1901 
1902 	request_end(fc, req);
1903 
1904 	return err ? err : nbytes;
1905 
1906  err_unlock_pq:
1907 	spin_unlock(&fpq->lock);
1908  err_finish:
1909 	fuse_copy_finish(cs);
1910 	return err;
1911 }
1912 
1913 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1914 {
1915 	struct fuse_copy_state cs;
1916 	struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1917 
1918 	if (!fud)
1919 		return -EPERM;
1920 
1921 	if (!iter_is_iovec(from))
1922 		return -EINVAL;
1923 
1924 	fuse_copy_init(&cs, 0, from);
1925 
1926 	return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1927 }
1928 
1929 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1930 				     struct file *out, loff_t *ppos,
1931 				     size_t len, unsigned int flags)
1932 {
1933 	unsigned nbuf;
1934 	unsigned idx;
1935 	struct pipe_buffer *bufs;
1936 	struct fuse_copy_state cs;
1937 	struct fuse_dev *fud;
1938 	size_t rem;
1939 	ssize_t ret;
1940 
1941 	fud = fuse_get_dev(out);
1942 	if (!fud)
1943 		return -EPERM;
1944 
1945 	bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1946 	if (!bufs)
1947 		return -ENOMEM;
1948 
1949 	pipe_lock(pipe);
1950 	nbuf = 0;
1951 	rem = 0;
1952 	for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1953 		rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1954 
1955 	ret = -EINVAL;
1956 	if (rem < len) {
1957 		pipe_unlock(pipe);
1958 		goto out;
1959 	}
1960 
1961 	rem = len;
1962 	while (rem) {
1963 		struct pipe_buffer *ibuf;
1964 		struct pipe_buffer *obuf;
1965 
1966 		BUG_ON(nbuf >= pipe->buffers);
1967 		BUG_ON(!pipe->nrbufs);
1968 		ibuf = &pipe->bufs[pipe->curbuf];
1969 		obuf = &bufs[nbuf];
1970 
1971 		if (rem >= ibuf->len) {
1972 			*obuf = *ibuf;
1973 			ibuf->ops = NULL;
1974 			pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1975 			pipe->nrbufs--;
1976 		} else {
1977 			pipe_buf_get(pipe, ibuf);
1978 			*obuf = *ibuf;
1979 			obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1980 			obuf->len = rem;
1981 			ibuf->offset += obuf->len;
1982 			ibuf->len -= obuf->len;
1983 		}
1984 		nbuf++;
1985 		rem -= obuf->len;
1986 	}
1987 	pipe_unlock(pipe);
1988 
1989 	fuse_copy_init(&cs, 0, NULL);
1990 	cs.pipebufs = bufs;
1991 	cs.nr_segs = nbuf;
1992 	cs.pipe = pipe;
1993 
1994 	if (flags & SPLICE_F_MOVE)
1995 		cs.move_pages = 1;
1996 
1997 	ret = fuse_dev_do_write(fud, &cs, len);
1998 
1999 	for (idx = 0; idx < nbuf; idx++)
2000 		pipe_buf_release(pipe, &bufs[idx]);
2001 
2002 out:
2003 	kfree(bufs);
2004 	return ret;
2005 }
2006 
2007 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2008 {
2009 	unsigned mask = POLLOUT | POLLWRNORM;
2010 	struct fuse_iqueue *fiq;
2011 	struct fuse_dev *fud = fuse_get_dev(file);
2012 
2013 	if (!fud)
2014 		return POLLERR;
2015 
2016 	fiq = &fud->fc->iq;
2017 	poll_wait(file, &fiq->waitq, wait);
2018 
2019 	spin_lock(&fiq->waitq.lock);
2020 	if (!fiq->connected)
2021 		mask = POLLERR;
2022 	else if (request_pending(fiq))
2023 		mask |= POLLIN | POLLRDNORM;
2024 	spin_unlock(&fiq->waitq.lock);
2025 
2026 	return mask;
2027 }
2028 
2029 /*
2030  * Abort all requests on the given list (pending or processing)
2031  *
2032  * This function releases and reacquires fc->lock
2033  */
2034 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2035 {
2036 	while (!list_empty(head)) {
2037 		struct fuse_req *req;
2038 		req = list_entry(head->next, struct fuse_req, list);
2039 		req->out.h.error = -ECONNABORTED;
2040 		clear_bit(FR_SENT, &req->flags);
2041 		list_del_init(&req->list);
2042 		request_end(fc, req);
2043 	}
2044 }
2045 
2046 static void end_polls(struct fuse_conn *fc)
2047 {
2048 	struct rb_node *p;
2049 
2050 	p = rb_first(&fc->polled_files);
2051 
2052 	while (p) {
2053 		struct fuse_file *ff;
2054 		ff = rb_entry(p, struct fuse_file, polled_node);
2055 		wake_up_interruptible_all(&ff->poll_wait);
2056 
2057 		p = rb_next(p);
2058 	}
2059 }
2060 
2061 /*
2062  * Abort all requests.
2063  *
2064  * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2065  * filesystem.
2066  *
2067  * The same effect is usually achievable through killing the filesystem daemon
2068  * and all users of the filesystem.  The exception is the combination of an
2069  * asynchronous request and the tricky deadlock (see
2070  * Documentation/filesystems/fuse.txt).
2071  *
2072  * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2073  * requests, they should be finished off immediately.  Locked requests will be
2074  * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2075  * requests.  It is possible that some request will finish before we can.  This
2076  * is OK, the request will in that case be removed from the list before we touch
2077  * it.
2078  */
2079 void fuse_abort_conn(struct fuse_conn *fc)
2080 {
2081 	struct fuse_iqueue *fiq = &fc->iq;
2082 
2083 	spin_lock(&fc->lock);
2084 	if (fc->connected) {
2085 		struct fuse_dev *fud;
2086 		struct fuse_req *req, *next;
2087 		LIST_HEAD(to_end1);
2088 		LIST_HEAD(to_end2);
2089 
2090 		fc->connected = 0;
2091 		fc->blocked = 0;
2092 		fuse_set_initialized(fc);
2093 		list_for_each_entry(fud, &fc->devices, entry) {
2094 			struct fuse_pqueue *fpq = &fud->pq;
2095 
2096 			spin_lock(&fpq->lock);
2097 			fpq->connected = 0;
2098 			list_for_each_entry_safe(req, next, &fpq->io, list) {
2099 				req->out.h.error = -ECONNABORTED;
2100 				spin_lock(&req->waitq.lock);
2101 				set_bit(FR_ABORTED, &req->flags);
2102 				if (!test_bit(FR_LOCKED, &req->flags)) {
2103 					set_bit(FR_PRIVATE, &req->flags);
2104 					list_move(&req->list, &to_end1);
2105 				}
2106 				spin_unlock(&req->waitq.lock);
2107 			}
2108 			list_splice_init(&fpq->processing, &to_end2);
2109 			spin_unlock(&fpq->lock);
2110 		}
2111 		fc->max_background = UINT_MAX;
2112 		flush_bg_queue(fc);
2113 
2114 		spin_lock(&fiq->waitq.lock);
2115 		fiq->connected = 0;
2116 		list_splice_init(&fiq->pending, &to_end2);
2117 		list_for_each_entry(req, &to_end2, list)
2118 			clear_bit(FR_PENDING, &req->flags);
2119 		while (forget_pending(fiq))
2120 			kfree(dequeue_forget(fiq, 1, NULL));
2121 		wake_up_all_locked(&fiq->waitq);
2122 		spin_unlock(&fiq->waitq.lock);
2123 		kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2124 		end_polls(fc);
2125 		wake_up_all(&fc->blocked_waitq);
2126 		spin_unlock(&fc->lock);
2127 
2128 		while (!list_empty(&to_end1)) {
2129 			req = list_first_entry(&to_end1, struct fuse_req, list);
2130 			__fuse_get_request(req);
2131 			list_del_init(&req->list);
2132 			request_end(fc, req);
2133 		}
2134 		end_requests(fc, &to_end2);
2135 	} else {
2136 		spin_unlock(&fc->lock);
2137 	}
2138 }
2139 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2140 
2141 int fuse_dev_release(struct inode *inode, struct file *file)
2142 {
2143 	struct fuse_dev *fud = fuse_get_dev(file);
2144 
2145 	if (fud) {
2146 		struct fuse_conn *fc = fud->fc;
2147 		struct fuse_pqueue *fpq = &fud->pq;
2148 
2149 		WARN_ON(!list_empty(&fpq->io));
2150 		end_requests(fc, &fpq->processing);
2151 		/* Are we the last open device? */
2152 		if (atomic_dec_and_test(&fc->dev_count)) {
2153 			WARN_ON(fc->iq.fasync != NULL);
2154 			fuse_abort_conn(fc);
2155 		}
2156 		fuse_dev_free(fud);
2157 	}
2158 	return 0;
2159 }
2160 EXPORT_SYMBOL_GPL(fuse_dev_release);
2161 
2162 static int fuse_dev_fasync(int fd, struct file *file, int on)
2163 {
2164 	struct fuse_dev *fud = fuse_get_dev(file);
2165 
2166 	if (!fud)
2167 		return -EPERM;
2168 
2169 	/* No locking - fasync_helper does its own locking */
2170 	return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2171 }
2172 
2173 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2174 {
2175 	struct fuse_dev *fud;
2176 
2177 	if (new->private_data)
2178 		return -EINVAL;
2179 
2180 	fud = fuse_dev_alloc(fc);
2181 	if (!fud)
2182 		return -ENOMEM;
2183 
2184 	new->private_data = fud;
2185 	atomic_inc(&fc->dev_count);
2186 
2187 	return 0;
2188 }
2189 
2190 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2191 			   unsigned long arg)
2192 {
2193 	int err = -ENOTTY;
2194 
2195 	if (cmd == FUSE_DEV_IOC_CLONE) {
2196 		int oldfd;
2197 
2198 		err = -EFAULT;
2199 		if (!get_user(oldfd, (__u32 __user *) arg)) {
2200 			struct file *old = fget(oldfd);
2201 
2202 			err = -EINVAL;
2203 			if (old) {
2204 				struct fuse_dev *fud = NULL;
2205 
2206 				/*
2207 				 * Check against file->f_op because CUSE
2208 				 * uses the same ioctl handler.
2209 				 */
2210 				if (old->f_op == file->f_op &&
2211 				    old->f_cred->user_ns == file->f_cred->user_ns)
2212 					fud = fuse_get_dev(old);
2213 
2214 				if (fud) {
2215 					mutex_lock(&fuse_mutex);
2216 					err = fuse_device_clone(fud->fc, file);
2217 					mutex_unlock(&fuse_mutex);
2218 				}
2219 				fput(old);
2220 			}
2221 		}
2222 	}
2223 	return err;
2224 }
2225 
2226 const struct file_operations fuse_dev_operations = {
2227 	.owner		= THIS_MODULE,
2228 	.open		= fuse_dev_open,
2229 	.llseek		= no_llseek,
2230 	.read_iter	= fuse_dev_read,
2231 	.splice_read	= fuse_dev_splice_read,
2232 	.write_iter	= fuse_dev_write,
2233 	.splice_write	= fuse_dev_splice_write,
2234 	.poll		= fuse_dev_poll,
2235 	.release	= fuse_dev_release,
2236 	.fasync		= fuse_dev_fasync,
2237 	.unlocked_ioctl = fuse_dev_ioctl,
2238 	.compat_ioctl   = fuse_dev_ioctl,
2239 };
2240 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2241 
2242 static struct miscdevice fuse_miscdevice = {
2243 	.minor = FUSE_MINOR,
2244 	.name  = "fuse",
2245 	.fops = &fuse_dev_operations,
2246 };
2247 
2248 int __init fuse_dev_init(void)
2249 {
2250 	int err = -ENOMEM;
2251 	fuse_req_cachep = kmem_cache_create("fuse_request",
2252 					    sizeof(struct fuse_req),
2253 					    0, 0, NULL);
2254 	if (!fuse_req_cachep)
2255 		goto out;
2256 
2257 	err = misc_register(&fuse_miscdevice);
2258 	if (err)
2259 		goto out_cache_clean;
2260 
2261 	return 0;
2262 
2263  out_cache_clean:
2264 	kmem_cache_destroy(fuse_req_cachep);
2265  out:
2266 	return err;
2267 }
2268 
2269 void fuse_dev_cleanup(void)
2270 {
2271 	misc_deregister(&fuse_miscdevice);
2272 	kmem_cache_destroy(fuse_req_cachep);
2273 }
2274