xref: /linux/fs/fuse/dev.c (revision 59024954a1e7e26b62680e1f2b5725249a6c09f7)
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4 
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8 
9 #include "fuse_i.h"
10 
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/poll.h>
14 #include <linux/uio.h>
15 #include <linux/miscdevice.h>
16 #include <linux/pagemap.h>
17 #include <linux/file.h>
18 #include <linux/slab.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/swap.h>
21 #include <linux/splice.h>
22 
23 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
24 MODULE_ALIAS("devname:fuse");
25 
26 static struct kmem_cache *fuse_req_cachep;
27 
28 static struct fuse_dev *fuse_get_dev(struct file *file)
29 {
30 	/*
31 	 * Lockless access is OK, because file->private data is set
32 	 * once during mount and is valid until the file is released.
33 	 */
34 	return ACCESS_ONCE(file->private_data);
35 }
36 
37 static void fuse_request_init(struct fuse_req *req, struct page **pages,
38 			      struct fuse_page_desc *page_descs,
39 			      unsigned npages)
40 {
41 	memset(req, 0, sizeof(*req));
42 	memset(pages, 0, sizeof(*pages) * npages);
43 	memset(page_descs, 0, sizeof(*page_descs) * npages);
44 	INIT_LIST_HEAD(&req->list);
45 	INIT_LIST_HEAD(&req->intr_entry);
46 	init_waitqueue_head(&req->waitq);
47 	atomic_set(&req->count, 1);
48 	req->pages = pages;
49 	req->page_descs = page_descs;
50 	req->max_pages = npages;
51 	__set_bit(FR_PENDING, &req->flags);
52 }
53 
54 static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
55 {
56 	struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
57 	if (req) {
58 		struct page **pages;
59 		struct fuse_page_desc *page_descs;
60 
61 		if (npages <= FUSE_REQ_INLINE_PAGES) {
62 			pages = req->inline_pages;
63 			page_descs = req->inline_page_descs;
64 		} else {
65 			pages = kmalloc(sizeof(struct page *) * npages, flags);
66 			page_descs = kmalloc(sizeof(struct fuse_page_desc) *
67 					     npages, flags);
68 		}
69 
70 		if (!pages || !page_descs) {
71 			kfree(pages);
72 			kfree(page_descs);
73 			kmem_cache_free(fuse_req_cachep, req);
74 			return NULL;
75 		}
76 
77 		fuse_request_init(req, pages, page_descs, npages);
78 	}
79 	return req;
80 }
81 
82 struct fuse_req *fuse_request_alloc(unsigned npages)
83 {
84 	return __fuse_request_alloc(npages, GFP_KERNEL);
85 }
86 EXPORT_SYMBOL_GPL(fuse_request_alloc);
87 
88 struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
89 {
90 	return __fuse_request_alloc(npages, GFP_NOFS);
91 }
92 
93 void fuse_request_free(struct fuse_req *req)
94 {
95 	if (req->pages != req->inline_pages) {
96 		kfree(req->pages);
97 		kfree(req->page_descs);
98 	}
99 	kmem_cache_free(fuse_req_cachep, req);
100 }
101 
102 void __fuse_get_request(struct fuse_req *req)
103 {
104 	atomic_inc(&req->count);
105 }
106 
107 /* Must be called with > 1 refcount */
108 static void __fuse_put_request(struct fuse_req *req)
109 {
110 	BUG_ON(atomic_read(&req->count) < 2);
111 	atomic_dec(&req->count);
112 }
113 
114 static void fuse_req_init_context(struct fuse_req *req)
115 {
116 	req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
117 	req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
118 	req->in.h.pid = current->pid;
119 }
120 
121 void fuse_set_initialized(struct fuse_conn *fc)
122 {
123 	/* Make sure stores before this are seen on another CPU */
124 	smp_wmb();
125 	fc->initialized = 1;
126 }
127 
128 static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
129 {
130 	return !fc->initialized || (for_background && fc->blocked);
131 }
132 
133 static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
134 				       bool for_background)
135 {
136 	struct fuse_req *req;
137 	int err;
138 	atomic_inc(&fc->num_waiting);
139 
140 	if (fuse_block_alloc(fc, for_background)) {
141 		err = -EINTR;
142 		if (wait_event_killable_exclusive(fc->blocked_waitq,
143 				!fuse_block_alloc(fc, for_background)))
144 			goto out;
145 	}
146 	/* Matches smp_wmb() in fuse_set_initialized() */
147 	smp_rmb();
148 
149 	err = -ENOTCONN;
150 	if (!fc->connected)
151 		goto out;
152 
153 	err = -ECONNREFUSED;
154 	if (fc->conn_error)
155 		goto out;
156 
157 	req = fuse_request_alloc(npages);
158 	err = -ENOMEM;
159 	if (!req) {
160 		if (for_background)
161 			wake_up(&fc->blocked_waitq);
162 		goto out;
163 	}
164 
165 	fuse_req_init_context(req);
166 	__set_bit(FR_WAITING, &req->flags);
167 	if (for_background)
168 		__set_bit(FR_BACKGROUND, &req->flags);
169 
170 	return req;
171 
172  out:
173 	atomic_dec(&fc->num_waiting);
174 	return ERR_PTR(err);
175 }
176 
177 struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
178 {
179 	return __fuse_get_req(fc, npages, false);
180 }
181 EXPORT_SYMBOL_GPL(fuse_get_req);
182 
183 struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
184 					     unsigned npages)
185 {
186 	return __fuse_get_req(fc, npages, true);
187 }
188 EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
189 
190 /*
191  * Return request in fuse_file->reserved_req.  However that may
192  * currently be in use.  If that is the case, wait for it to become
193  * available.
194  */
195 static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
196 					 struct file *file)
197 {
198 	struct fuse_req *req = NULL;
199 	struct fuse_file *ff = file->private_data;
200 
201 	do {
202 		wait_event(fc->reserved_req_waitq, ff->reserved_req);
203 		spin_lock(&fc->lock);
204 		if (ff->reserved_req) {
205 			req = ff->reserved_req;
206 			ff->reserved_req = NULL;
207 			req->stolen_file = get_file(file);
208 		}
209 		spin_unlock(&fc->lock);
210 	} while (!req);
211 
212 	return req;
213 }
214 
215 /*
216  * Put stolen request back into fuse_file->reserved_req
217  */
218 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
219 {
220 	struct file *file = req->stolen_file;
221 	struct fuse_file *ff = file->private_data;
222 
223 	spin_lock(&fc->lock);
224 	fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
225 	BUG_ON(ff->reserved_req);
226 	ff->reserved_req = req;
227 	wake_up_all(&fc->reserved_req_waitq);
228 	spin_unlock(&fc->lock);
229 	fput(file);
230 }
231 
232 /*
233  * Gets a requests for a file operation, always succeeds
234  *
235  * This is used for sending the FLUSH request, which must get to
236  * userspace, due to POSIX locks which may need to be unlocked.
237  *
238  * If allocation fails due to OOM, use the reserved request in
239  * fuse_file.
240  *
241  * This is very unlikely to deadlock accidentally, since the
242  * filesystem should not have it's own file open.  If deadlock is
243  * intentional, it can still be broken by "aborting" the filesystem.
244  */
245 struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
246 					     struct file *file)
247 {
248 	struct fuse_req *req;
249 
250 	atomic_inc(&fc->num_waiting);
251 	wait_event(fc->blocked_waitq, fc->initialized);
252 	/* Matches smp_wmb() in fuse_set_initialized() */
253 	smp_rmb();
254 	req = fuse_request_alloc(0);
255 	if (!req)
256 		req = get_reserved_req(fc, file);
257 
258 	fuse_req_init_context(req);
259 	__set_bit(FR_WAITING, &req->flags);
260 	__clear_bit(FR_BACKGROUND, &req->flags);
261 	return req;
262 }
263 
264 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
265 {
266 	if (atomic_dec_and_test(&req->count)) {
267 		if (test_bit(FR_BACKGROUND, &req->flags)) {
268 			/*
269 			 * We get here in the unlikely case that a background
270 			 * request was allocated but not sent
271 			 */
272 			spin_lock(&fc->lock);
273 			if (!fc->blocked)
274 				wake_up(&fc->blocked_waitq);
275 			spin_unlock(&fc->lock);
276 		}
277 
278 		if (test_bit(FR_WAITING, &req->flags)) {
279 			__clear_bit(FR_WAITING, &req->flags);
280 			atomic_dec(&fc->num_waiting);
281 		}
282 
283 		if (req->stolen_file)
284 			put_reserved_req(fc, req);
285 		else
286 			fuse_request_free(req);
287 	}
288 }
289 EXPORT_SYMBOL_GPL(fuse_put_request);
290 
291 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
292 {
293 	unsigned nbytes = 0;
294 	unsigned i;
295 
296 	for (i = 0; i < numargs; i++)
297 		nbytes += args[i].size;
298 
299 	return nbytes;
300 }
301 
302 static u64 fuse_get_unique(struct fuse_iqueue *fiq)
303 {
304 	return ++fiq->reqctr;
305 }
306 
307 static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
308 {
309 	req->in.h.len = sizeof(struct fuse_in_header) +
310 		len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
311 	list_add_tail(&req->list, &fiq->pending);
312 	wake_up_locked(&fiq->waitq);
313 	kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
314 }
315 
316 void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
317 		       u64 nodeid, u64 nlookup)
318 {
319 	struct fuse_iqueue *fiq = &fc->iq;
320 
321 	forget->forget_one.nodeid = nodeid;
322 	forget->forget_one.nlookup = nlookup;
323 
324 	spin_lock(&fiq->waitq.lock);
325 	if (fiq->connected) {
326 		fiq->forget_list_tail->next = forget;
327 		fiq->forget_list_tail = forget;
328 		wake_up_locked(&fiq->waitq);
329 		kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
330 	} else {
331 		kfree(forget);
332 	}
333 	spin_unlock(&fiq->waitq.lock);
334 }
335 
336 static void flush_bg_queue(struct fuse_conn *fc)
337 {
338 	while (fc->active_background < fc->max_background &&
339 	       !list_empty(&fc->bg_queue)) {
340 		struct fuse_req *req;
341 		struct fuse_iqueue *fiq = &fc->iq;
342 
343 		req = list_entry(fc->bg_queue.next, struct fuse_req, list);
344 		list_del(&req->list);
345 		fc->active_background++;
346 		spin_lock(&fiq->waitq.lock);
347 		req->in.h.unique = fuse_get_unique(fiq);
348 		queue_request(fiq, req);
349 		spin_unlock(&fiq->waitq.lock);
350 	}
351 }
352 
353 /*
354  * This function is called when a request is finished.  Either a reply
355  * has arrived or it was aborted (and not yet sent) or some error
356  * occurred during communication with userspace, or the device file
357  * was closed.  The requester thread is woken up (if still waiting),
358  * the 'end' callback is called if given, else the reference to the
359  * request is released
360  */
361 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
362 {
363 	struct fuse_iqueue *fiq = &fc->iq;
364 
365 	if (test_and_set_bit(FR_FINISHED, &req->flags))
366 		return;
367 
368 	spin_lock(&fiq->waitq.lock);
369 	list_del_init(&req->intr_entry);
370 	spin_unlock(&fiq->waitq.lock);
371 	WARN_ON(test_bit(FR_PENDING, &req->flags));
372 	WARN_ON(test_bit(FR_SENT, &req->flags));
373 	if (test_bit(FR_BACKGROUND, &req->flags)) {
374 		spin_lock(&fc->lock);
375 		clear_bit(FR_BACKGROUND, &req->flags);
376 		if (fc->num_background == fc->max_background)
377 			fc->blocked = 0;
378 
379 		/* Wake up next waiter, if any */
380 		if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
381 			wake_up(&fc->blocked_waitq);
382 
383 		if (fc->num_background == fc->congestion_threshold &&
384 		    fc->connected && fc->bdi_initialized) {
385 			clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
386 			clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
387 		}
388 		fc->num_background--;
389 		fc->active_background--;
390 		flush_bg_queue(fc);
391 		spin_unlock(&fc->lock);
392 	}
393 	wake_up(&req->waitq);
394 	if (req->end)
395 		req->end(fc, req);
396 	fuse_put_request(fc, req);
397 }
398 
399 static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
400 {
401 	spin_lock(&fiq->waitq.lock);
402 	if (list_empty(&req->intr_entry)) {
403 		list_add_tail(&req->intr_entry, &fiq->interrupts);
404 		wake_up_locked(&fiq->waitq);
405 	}
406 	spin_unlock(&fiq->waitq.lock);
407 	kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
408 }
409 
410 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
411 {
412 	struct fuse_iqueue *fiq = &fc->iq;
413 	int err;
414 
415 	if (!fc->no_interrupt) {
416 		/* Any signal may interrupt this */
417 		err = wait_event_interruptible(req->waitq,
418 					test_bit(FR_FINISHED, &req->flags));
419 		if (!err)
420 			return;
421 
422 		set_bit(FR_INTERRUPTED, &req->flags);
423 		/* matches barrier in fuse_dev_do_read() */
424 		smp_mb__after_atomic();
425 		if (test_bit(FR_SENT, &req->flags))
426 			queue_interrupt(fiq, req);
427 	}
428 
429 	if (!test_bit(FR_FORCE, &req->flags)) {
430 		/* Only fatal signals may interrupt this */
431 		err = wait_event_killable(req->waitq,
432 					test_bit(FR_FINISHED, &req->flags));
433 		if (!err)
434 			return;
435 
436 		spin_lock(&fiq->waitq.lock);
437 		/* Request is not yet in userspace, bail out */
438 		if (test_bit(FR_PENDING, &req->flags)) {
439 			list_del(&req->list);
440 			spin_unlock(&fiq->waitq.lock);
441 			__fuse_put_request(req);
442 			req->out.h.error = -EINTR;
443 			return;
444 		}
445 		spin_unlock(&fiq->waitq.lock);
446 	}
447 
448 	/*
449 	 * Either request is already in userspace, or it was forced.
450 	 * Wait it out.
451 	 */
452 	wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
453 }
454 
455 static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
456 {
457 	struct fuse_iqueue *fiq = &fc->iq;
458 
459 	BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
460 	spin_lock(&fiq->waitq.lock);
461 	if (!fiq->connected) {
462 		spin_unlock(&fiq->waitq.lock);
463 		req->out.h.error = -ENOTCONN;
464 	} else {
465 		req->in.h.unique = fuse_get_unique(fiq);
466 		queue_request(fiq, req);
467 		/* acquire extra reference, since request is still needed
468 		   after request_end() */
469 		__fuse_get_request(req);
470 		spin_unlock(&fiq->waitq.lock);
471 
472 		request_wait_answer(fc, req);
473 		/* Pairs with smp_wmb() in request_end() */
474 		smp_rmb();
475 	}
476 }
477 
478 void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
479 {
480 	__set_bit(FR_ISREPLY, &req->flags);
481 	if (!test_bit(FR_WAITING, &req->flags)) {
482 		__set_bit(FR_WAITING, &req->flags);
483 		atomic_inc(&fc->num_waiting);
484 	}
485 	__fuse_request_send(fc, req);
486 }
487 EXPORT_SYMBOL_GPL(fuse_request_send);
488 
489 static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
490 {
491 	if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
492 		args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
493 
494 	if (fc->minor < 9) {
495 		switch (args->in.h.opcode) {
496 		case FUSE_LOOKUP:
497 		case FUSE_CREATE:
498 		case FUSE_MKNOD:
499 		case FUSE_MKDIR:
500 		case FUSE_SYMLINK:
501 		case FUSE_LINK:
502 			args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
503 			break;
504 		case FUSE_GETATTR:
505 		case FUSE_SETATTR:
506 			args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
507 			break;
508 		}
509 	}
510 	if (fc->minor < 12) {
511 		switch (args->in.h.opcode) {
512 		case FUSE_CREATE:
513 			args->in.args[0].size = sizeof(struct fuse_open_in);
514 			break;
515 		case FUSE_MKNOD:
516 			args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
517 			break;
518 		}
519 	}
520 }
521 
522 ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
523 {
524 	struct fuse_req *req;
525 	ssize_t ret;
526 
527 	req = fuse_get_req(fc, 0);
528 	if (IS_ERR(req))
529 		return PTR_ERR(req);
530 
531 	/* Needs to be done after fuse_get_req() so that fc->minor is valid */
532 	fuse_adjust_compat(fc, args);
533 
534 	req->in.h.opcode = args->in.h.opcode;
535 	req->in.h.nodeid = args->in.h.nodeid;
536 	req->in.numargs = args->in.numargs;
537 	memcpy(req->in.args, args->in.args,
538 	       args->in.numargs * sizeof(struct fuse_in_arg));
539 	req->out.argvar = args->out.argvar;
540 	req->out.numargs = args->out.numargs;
541 	memcpy(req->out.args, args->out.args,
542 	       args->out.numargs * sizeof(struct fuse_arg));
543 	fuse_request_send(fc, req);
544 	ret = req->out.h.error;
545 	if (!ret && args->out.argvar) {
546 		BUG_ON(args->out.numargs != 1);
547 		ret = req->out.args[0].size;
548 	}
549 	fuse_put_request(fc, req);
550 
551 	return ret;
552 }
553 
554 /*
555  * Called under fc->lock
556  *
557  * fc->connected must have been checked previously
558  */
559 void fuse_request_send_background_locked(struct fuse_conn *fc,
560 					 struct fuse_req *req)
561 {
562 	BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
563 	if (!test_bit(FR_WAITING, &req->flags)) {
564 		__set_bit(FR_WAITING, &req->flags);
565 		atomic_inc(&fc->num_waiting);
566 	}
567 	__set_bit(FR_ISREPLY, &req->flags);
568 	fc->num_background++;
569 	if (fc->num_background == fc->max_background)
570 		fc->blocked = 1;
571 	if (fc->num_background == fc->congestion_threshold &&
572 	    fc->bdi_initialized) {
573 		set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
574 		set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
575 	}
576 	list_add_tail(&req->list, &fc->bg_queue);
577 	flush_bg_queue(fc);
578 }
579 
580 void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
581 {
582 	BUG_ON(!req->end);
583 	spin_lock(&fc->lock);
584 	if (fc->connected) {
585 		fuse_request_send_background_locked(fc, req);
586 		spin_unlock(&fc->lock);
587 	} else {
588 		spin_unlock(&fc->lock);
589 		req->out.h.error = -ENOTCONN;
590 		req->end(fc, req);
591 		fuse_put_request(fc, req);
592 	}
593 }
594 EXPORT_SYMBOL_GPL(fuse_request_send_background);
595 
596 static int fuse_request_send_notify_reply(struct fuse_conn *fc,
597 					  struct fuse_req *req, u64 unique)
598 {
599 	int err = -ENODEV;
600 	struct fuse_iqueue *fiq = &fc->iq;
601 
602 	__clear_bit(FR_ISREPLY, &req->flags);
603 	req->in.h.unique = unique;
604 	spin_lock(&fiq->waitq.lock);
605 	if (fiq->connected) {
606 		queue_request(fiq, req);
607 		err = 0;
608 	}
609 	spin_unlock(&fiq->waitq.lock);
610 
611 	return err;
612 }
613 
614 void fuse_force_forget(struct file *file, u64 nodeid)
615 {
616 	struct inode *inode = file_inode(file);
617 	struct fuse_conn *fc = get_fuse_conn(inode);
618 	struct fuse_req *req;
619 	struct fuse_forget_in inarg;
620 
621 	memset(&inarg, 0, sizeof(inarg));
622 	inarg.nlookup = 1;
623 	req = fuse_get_req_nofail_nopages(fc, file);
624 	req->in.h.opcode = FUSE_FORGET;
625 	req->in.h.nodeid = nodeid;
626 	req->in.numargs = 1;
627 	req->in.args[0].size = sizeof(inarg);
628 	req->in.args[0].value = &inarg;
629 	__clear_bit(FR_ISREPLY, &req->flags);
630 	__fuse_request_send(fc, req);
631 	/* ignore errors */
632 	fuse_put_request(fc, req);
633 }
634 
635 /*
636  * Lock the request.  Up to the next unlock_request() there mustn't be
637  * anything that could cause a page-fault.  If the request was already
638  * aborted bail out.
639  */
640 static int lock_request(struct fuse_req *req)
641 {
642 	int err = 0;
643 	if (req) {
644 		spin_lock(&req->waitq.lock);
645 		if (test_bit(FR_ABORTED, &req->flags))
646 			err = -ENOENT;
647 		else
648 			set_bit(FR_LOCKED, &req->flags);
649 		spin_unlock(&req->waitq.lock);
650 	}
651 	return err;
652 }
653 
654 /*
655  * Unlock request.  If it was aborted while locked, caller is responsible
656  * for unlocking and ending the request.
657  */
658 static int unlock_request(struct fuse_req *req)
659 {
660 	int err = 0;
661 	if (req) {
662 		spin_lock(&req->waitq.lock);
663 		if (test_bit(FR_ABORTED, &req->flags))
664 			err = -ENOENT;
665 		else
666 			clear_bit(FR_LOCKED, &req->flags);
667 		spin_unlock(&req->waitq.lock);
668 	}
669 	return err;
670 }
671 
672 struct fuse_copy_state {
673 	int write;
674 	struct fuse_req *req;
675 	struct iov_iter *iter;
676 	struct pipe_buffer *pipebufs;
677 	struct pipe_buffer *currbuf;
678 	struct pipe_inode_info *pipe;
679 	unsigned long nr_segs;
680 	struct page *pg;
681 	unsigned len;
682 	unsigned offset;
683 	unsigned move_pages:1;
684 };
685 
686 static void fuse_copy_init(struct fuse_copy_state *cs, int write,
687 			   struct iov_iter *iter)
688 {
689 	memset(cs, 0, sizeof(*cs));
690 	cs->write = write;
691 	cs->iter = iter;
692 }
693 
694 /* Unmap and put previous page of userspace buffer */
695 static void fuse_copy_finish(struct fuse_copy_state *cs)
696 {
697 	if (cs->currbuf) {
698 		struct pipe_buffer *buf = cs->currbuf;
699 
700 		if (cs->write)
701 			buf->len = PAGE_SIZE - cs->len;
702 		cs->currbuf = NULL;
703 	} else if (cs->pg) {
704 		if (cs->write) {
705 			flush_dcache_page(cs->pg);
706 			set_page_dirty_lock(cs->pg);
707 		}
708 		put_page(cs->pg);
709 	}
710 	cs->pg = NULL;
711 }
712 
713 /*
714  * Get another pagefull of userspace buffer, and map it to kernel
715  * address space, and lock request
716  */
717 static int fuse_copy_fill(struct fuse_copy_state *cs)
718 {
719 	struct page *page;
720 	int err;
721 
722 	err = unlock_request(cs->req);
723 	if (err)
724 		return err;
725 
726 	fuse_copy_finish(cs);
727 	if (cs->pipebufs) {
728 		struct pipe_buffer *buf = cs->pipebufs;
729 
730 		if (!cs->write) {
731 			err = buf->ops->confirm(cs->pipe, buf);
732 			if (err)
733 				return err;
734 
735 			BUG_ON(!cs->nr_segs);
736 			cs->currbuf = buf;
737 			cs->pg = buf->page;
738 			cs->offset = buf->offset;
739 			cs->len = buf->len;
740 			cs->pipebufs++;
741 			cs->nr_segs--;
742 		} else {
743 			if (cs->nr_segs == cs->pipe->buffers)
744 				return -EIO;
745 
746 			page = alloc_page(GFP_HIGHUSER);
747 			if (!page)
748 				return -ENOMEM;
749 
750 			buf->page = page;
751 			buf->offset = 0;
752 			buf->len = 0;
753 
754 			cs->currbuf = buf;
755 			cs->pg = page;
756 			cs->offset = 0;
757 			cs->len = PAGE_SIZE;
758 			cs->pipebufs++;
759 			cs->nr_segs++;
760 		}
761 	} else {
762 		size_t off;
763 		err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
764 		if (err < 0)
765 			return err;
766 		BUG_ON(!err);
767 		cs->len = err;
768 		cs->offset = off;
769 		cs->pg = page;
770 		iov_iter_advance(cs->iter, err);
771 	}
772 
773 	return lock_request(cs->req);
774 }
775 
776 /* Do as much copy to/from userspace buffer as we can */
777 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
778 {
779 	unsigned ncpy = min(*size, cs->len);
780 	if (val) {
781 		void *pgaddr = kmap_atomic(cs->pg);
782 		void *buf = pgaddr + cs->offset;
783 
784 		if (cs->write)
785 			memcpy(buf, *val, ncpy);
786 		else
787 			memcpy(*val, buf, ncpy);
788 
789 		kunmap_atomic(pgaddr);
790 		*val += ncpy;
791 	}
792 	*size -= ncpy;
793 	cs->len -= ncpy;
794 	cs->offset += ncpy;
795 	return ncpy;
796 }
797 
798 static int fuse_check_page(struct page *page)
799 {
800 	if (page_mapcount(page) ||
801 	    page->mapping != NULL ||
802 	    page_count(page) != 1 ||
803 	    (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
804 	     ~(1 << PG_locked |
805 	       1 << PG_referenced |
806 	       1 << PG_uptodate |
807 	       1 << PG_lru |
808 	       1 << PG_active |
809 	       1 << PG_reclaim))) {
810 		printk(KERN_WARNING "fuse: trying to steal weird page\n");
811 		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);
812 		return 1;
813 	}
814 	return 0;
815 }
816 
817 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
818 {
819 	int err;
820 	struct page *oldpage = *pagep;
821 	struct page *newpage;
822 	struct pipe_buffer *buf = cs->pipebufs;
823 
824 	err = unlock_request(cs->req);
825 	if (err)
826 		return err;
827 
828 	fuse_copy_finish(cs);
829 
830 	err = buf->ops->confirm(cs->pipe, buf);
831 	if (err)
832 		return err;
833 
834 	BUG_ON(!cs->nr_segs);
835 	cs->currbuf = buf;
836 	cs->len = buf->len;
837 	cs->pipebufs++;
838 	cs->nr_segs--;
839 
840 	if (cs->len != PAGE_SIZE)
841 		goto out_fallback;
842 
843 	if (buf->ops->steal(cs->pipe, buf) != 0)
844 		goto out_fallback;
845 
846 	newpage = buf->page;
847 
848 	if (!PageUptodate(newpage))
849 		SetPageUptodate(newpage);
850 
851 	ClearPageMappedToDisk(newpage);
852 
853 	if (fuse_check_page(newpage) != 0)
854 		goto out_fallback_unlock;
855 
856 	/*
857 	 * This is a new and locked page, it shouldn't be mapped or
858 	 * have any special flags on it
859 	 */
860 	if (WARN_ON(page_mapped(oldpage)))
861 		goto out_fallback_unlock;
862 	if (WARN_ON(page_has_private(oldpage)))
863 		goto out_fallback_unlock;
864 	if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
865 		goto out_fallback_unlock;
866 	if (WARN_ON(PageMlocked(oldpage)))
867 		goto out_fallback_unlock;
868 
869 	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
870 	if (err) {
871 		unlock_page(newpage);
872 		return err;
873 	}
874 
875 	get_page(newpage);
876 
877 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
878 		lru_cache_add_file(newpage);
879 
880 	err = 0;
881 	spin_lock(&cs->req->waitq.lock);
882 	if (test_bit(FR_ABORTED, &cs->req->flags))
883 		err = -ENOENT;
884 	else
885 		*pagep = newpage;
886 	spin_unlock(&cs->req->waitq.lock);
887 
888 	if (err) {
889 		unlock_page(newpage);
890 		put_page(newpage);
891 		return err;
892 	}
893 
894 	unlock_page(oldpage);
895 	put_page(oldpage);
896 	cs->len = 0;
897 
898 	return 0;
899 
900 out_fallback_unlock:
901 	unlock_page(newpage);
902 out_fallback:
903 	cs->pg = buf->page;
904 	cs->offset = buf->offset;
905 
906 	err = lock_request(cs->req);
907 	if (err)
908 		return err;
909 
910 	return 1;
911 }
912 
913 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
914 			 unsigned offset, unsigned count)
915 {
916 	struct pipe_buffer *buf;
917 	int err;
918 
919 	if (cs->nr_segs == cs->pipe->buffers)
920 		return -EIO;
921 
922 	err = unlock_request(cs->req);
923 	if (err)
924 		return err;
925 
926 	fuse_copy_finish(cs);
927 
928 	buf = cs->pipebufs;
929 	get_page(page);
930 	buf->page = page;
931 	buf->offset = offset;
932 	buf->len = count;
933 
934 	cs->pipebufs++;
935 	cs->nr_segs++;
936 	cs->len = 0;
937 
938 	return 0;
939 }
940 
941 /*
942  * Copy a page in the request to/from the userspace buffer.  Must be
943  * done atomically
944  */
945 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
946 			  unsigned offset, unsigned count, int zeroing)
947 {
948 	int err;
949 	struct page *page = *pagep;
950 
951 	if (page && zeroing && count < PAGE_SIZE)
952 		clear_highpage(page);
953 
954 	while (count) {
955 		if (cs->write && cs->pipebufs && page) {
956 			return fuse_ref_page(cs, page, offset, count);
957 		} else if (!cs->len) {
958 			if (cs->move_pages && page &&
959 			    offset == 0 && count == PAGE_SIZE) {
960 				err = fuse_try_move_page(cs, pagep);
961 				if (err <= 0)
962 					return err;
963 			} else {
964 				err = fuse_copy_fill(cs);
965 				if (err)
966 					return err;
967 			}
968 		}
969 		if (page) {
970 			void *mapaddr = kmap_atomic(page);
971 			void *buf = mapaddr + offset;
972 			offset += fuse_copy_do(cs, &buf, &count);
973 			kunmap_atomic(mapaddr);
974 		} else
975 			offset += fuse_copy_do(cs, NULL, &count);
976 	}
977 	if (page && !cs->write)
978 		flush_dcache_page(page);
979 	return 0;
980 }
981 
982 /* Copy pages in the request to/from userspace buffer */
983 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
984 			   int zeroing)
985 {
986 	unsigned i;
987 	struct fuse_req *req = cs->req;
988 
989 	for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
990 		int err;
991 		unsigned offset = req->page_descs[i].offset;
992 		unsigned count = min(nbytes, req->page_descs[i].length);
993 
994 		err = fuse_copy_page(cs, &req->pages[i], offset, count,
995 				     zeroing);
996 		if (err)
997 			return err;
998 
999 		nbytes -= count;
1000 	}
1001 	return 0;
1002 }
1003 
1004 /* Copy a single argument in the request to/from userspace buffer */
1005 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1006 {
1007 	while (size) {
1008 		if (!cs->len) {
1009 			int err = fuse_copy_fill(cs);
1010 			if (err)
1011 				return err;
1012 		}
1013 		fuse_copy_do(cs, &val, &size);
1014 	}
1015 	return 0;
1016 }
1017 
1018 /* Copy request arguments to/from userspace buffer */
1019 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1020 			  unsigned argpages, struct fuse_arg *args,
1021 			  int zeroing)
1022 {
1023 	int err = 0;
1024 	unsigned i;
1025 
1026 	for (i = 0; !err && i < numargs; i++)  {
1027 		struct fuse_arg *arg = &args[i];
1028 		if (i == numargs - 1 && argpages)
1029 			err = fuse_copy_pages(cs, arg->size, zeroing);
1030 		else
1031 			err = fuse_copy_one(cs, arg->value, arg->size);
1032 	}
1033 	return err;
1034 }
1035 
1036 static int forget_pending(struct fuse_iqueue *fiq)
1037 {
1038 	return fiq->forget_list_head.next != NULL;
1039 }
1040 
1041 static int request_pending(struct fuse_iqueue *fiq)
1042 {
1043 	return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1044 		forget_pending(fiq);
1045 }
1046 
1047 /*
1048  * Transfer an interrupt request to userspace
1049  *
1050  * Unlike other requests this is assembled on demand, without a need
1051  * to allocate a separate fuse_req structure.
1052  *
1053  * Called with fiq->waitq.lock held, releases it
1054  */
1055 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1056 			       struct fuse_copy_state *cs,
1057 			       size_t nbytes, struct fuse_req *req)
1058 __releases(fiq->waitq.lock)
1059 {
1060 	struct fuse_in_header ih;
1061 	struct fuse_interrupt_in arg;
1062 	unsigned reqsize = sizeof(ih) + sizeof(arg);
1063 	int err;
1064 
1065 	list_del_init(&req->intr_entry);
1066 	req->intr_unique = fuse_get_unique(fiq);
1067 	memset(&ih, 0, sizeof(ih));
1068 	memset(&arg, 0, sizeof(arg));
1069 	ih.len = reqsize;
1070 	ih.opcode = FUSE_INTERRUPT;
1071 	ih.unique = req->intr_unique;
1072 	arg.unique = req->in.h.unique;
1073 
1074 	spin_unlock(&fiq->waitq.lock);
1075 	if (nbytes < reqsize)
1076 		return -EINVAL;
1077 
1078 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1079 	if (!err)
1080 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1081 	fuse_copy_finish(cs);
1082 
1083 	return err ? err : reqsize;
1084 }
1085 
1086 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1087 					       unsigned max,
1088 					       unsigned *countp)
1089 {
1090 	struct fuse_forget_link *head = fiq->forget_list_head.next;
1091 	struct fuse_forget_link **newhead = &head;
1092 	unsigned count;
1093 
1094 	for (count = 0; *newhead != NULL && count < max; count++)
1095 		newhead = &(*newhead)->next;
1096 
1097 	fiq->forget_list_head.next = *newhead;
1098 	*newhead = NULL;
1099 	if (fiq->forget_list_head.next == NULL)
1100 		fiq->forget_list_tail = &fiq->forget_list_head;
1101 
1102 	if (countp != NULL)
1103 		*countp = count;
1104 
1105 	return head;
1106 }
1107 
1108 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1109 				   struct fuse_copy_state *cs,
1110 				   size_t nbytes)
1111 __releases(fiq->waitq.lock)
1112 {
1113 	int err;
1114 	struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1115 	struct fuse_forget_in arg = {
1116 		.nlookup = forget->forget_one.nlookup,
1117 	};
1118 	struct fuse_in_header ih = {
1119 		.opcode = FUSE_FORGET,
1120 		.nodeid = forget->forget_one.nodeid,
1121 		.unique = fuse_get_unique(fiq),
1122 		.len = sizeof(ih) + sizeof(arg),
1123 	};
1124 
1125 	spin_unlock(&fiq->waitq.lock);
1126 	kfree(forget);
1127 	if (nbytes < ih.len)
1128 		return -EINVAL;
1129 
1130 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1131 	if (!err)
1132 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1133 	fuse_copy_finish(cs);
1134 
1135 	if (err)
1136 		return err;
1137 
1138 	return ih.len;
1139 }
1140 
1141 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1142 				   struct fuse_copy_state *cs, size_t nbytes)
1143 __releases(fiq->waitq.lock)
1144 {
1145 	int err;
1146 	unsigned max_forgets;
1147 	unsigned count;
1148 	struct fuse_forget_link *head;
1149 	struct fuse_batch_forget_in arg = { .count = 0 };
1150 	struct fuse_in_header ih = {
1151 		.opcode = FUSE_BATCH_FORGET,
1152 		.unique = fuse_get_unique(fiq),
1153 		.len = sizeof(ih) + sizeof(arg),
1154 	};
1155 
1156 	if (nbytes < ih.len) {
1157 		spin_unlock(&fiq->waitq.lock);
1158 		return -EINVAL;
1159 	}
1160 
1161 	max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1162 	head = dequeue_forget(fiq, max_forgets, &count);
1163 	spin_unlock(&fiq->waitq.lock);
1164 
1165 	arg.count = count;
1166 	ih.len += count * sizeof(struct fuse_forget_one);
1167 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1168 	if (!err)
1169 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1170 
1171 	while (head) {
1172 		struct fuse_forget_link *forget = head;
1173 
1174 		if (!err) {
1175 			err = fuse_copy_one(cs, &forget->forget_one,
1176 					    sizeof(forget->forget_one));
1177 		}
1178 		head = forget->next;
1179 		kfree(forget);
1180 	}
1181 
1182 	fuse_copy_finish(cs);
1183 
1184 	if (err)
1185 		return err;
1186 
1187 	return ih.len;
1188 }
1189 
1190 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1191 			    struct fuse_copy_state *cs,
1192 			    size_t nbytes)
1193 __releases(fiq->waitq.lock)
1194 {
1195 	if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1196 		return fuse_read_single_forget(fiq, cs, nbytes);
1197 	else
1198 		return fuse_read_batch_forget(fiq, cs, nbytes);
1199 }
1200 
1201 /*
1202  * Read a single request into the userspace filesystem's buffer.  This
1203  * function waits until a request is available, then removes it from
1204  * the pending list and copies request data to userspace buffer.  If
1205  * no reply is needed (FORGET) or request has been aborted or there
1206  * was an error during the copying then it's finished by calling
1207  * request_end().  Otherwise add it to the processing list, and set
1208  * the 'sent' flag.
1209  */
1210 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1211 				struct fuse_copy_state *cs, size_t nbytes)
1212 {
1213 	ssize_t err;
1214 	struct fuse_conn *fc = fud->fc;
1215 	struct fuse_iqueue *fiq = &fc->iq;
1216 	struct fuse_pqueue *fpq = &fud->pq;
1217 	struct fuse_req *req;
1218 	struct fuse_in *in;
1219 	unsigned reqsize;
1220 
1221  restart:
1222 	spin_lock(&fiq->waitq.lock);
1223 	err = -EAGAIN;
1224 	if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1225 	    !request_pending(fiq))
1226 		goto err_unlock;
1227 
1228 	err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1229 				!fiq->connected || request_pending(fiq));
1230 	if (err)
1231 		goto err_unlock;
1232 
1233 	err = -ENODEV;
1234 	if (!fiq->connected)
1235 		goto err_unlock;
1236 
1237 	if (!list_empty(&fiq->interrupts)) {
1238 		req = list_entry(fiq->interrupts.next, struct fuse_req,
1239 				 intr_entry);
1240 		return fuse_read_interrupt(fiq, cs, nbytes, req);
1241 	}
1242 
1243 	if (forget_pending(fiq)) {
1244 		if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1245 			return fuse_read_forget(fc, fiq, cs, nbytes);
1246 
1247 		if (fiq->forget_batch <= -8)
1248 			fiq->forget_batch = 16;
1249 	}
1250 
1251 	req = list_entry(fiq->pending.next, struct fuse_req, list);
1252 	clear_bit(FR_PENDING, &req->flags);
1253 	list_del_init(&req->list);
1254 	spin_unlock(&fiq->waitq.lock);
1255 
1256 	in = &req->in;
1257 	reqsize = in->h.len;
1258 	/* If request is too large, reply with an error and restart the read */
1259 	if (nbytes < reqsize) {
1260 		req->out.h.error = -EIO;
1261 		/* SETXATTR is special, since it may contain too large data */
1262 		if (in->h.opcode == FUSE_SETXATTR)
1263 			req->out.h.error = -E2BIG;
1264 		request_end(fc, req);
1265 		goto restart;
1266 	}
1267 	spin_lock(&fpq->lock);
1268 	list_add(&req->list, &fpq->io);
1269 	spin_unlock(&fpq->lock);
1270 	cs->req = req;
1271 	err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1272 	if (!err)
1273 		err = fuse_copy_args(cs, in->numargs, in->argpages,
1274 				     (struct fuse_arg *) in->args, 0);
1275 	fuse_copy_finish(cs);
1276 	spin_lock(&fpq->lock);
1277 	clear_bit(FR_LOCKED, &req->flags);
1278 	if (!fpq->connected) {
1279 		err = -ENODEV;
1280 		goto out_end;
1281 	}
1282 	if (err) {
1283 		req->out.h.error = -EIO;
1284 		goto out_end;
1285 	}
1286 	if (!test_bit(FR_ISREPLY, &req->flags)) {
1287 		err = reqsize;
1288 		goto out_end;
1289 	}
1290 	list_move_tail(&req->list, &fpq->processing);
1291 	spin_unlock(&fpq->lock);
1292 	set_bit(FR_SENT, &req->flags);
1293 	/* matches barrier in request_wait_answer() */
1294 	smp_mb__after_atomic();
1295 	if (test_bit(FR_INTERRUPTED, &req->flags))
1296 		queue_interrupt(fiq, req);
1297 
1298 	return reqsize;
1299 
1300 out_end:
1301 	if (!test_bit(FR_PRIVATE, &req->flags))
1302 		list_del_init(&req->list);
1303 	spin_unlock(&fpq->lock);
1304 	request_end(fc, req);
1305 	return err;
1306 
1307  err_unlock:
1308 	spin_unlock(&fiq->waitq.lock);
1309 	return err;
1310 }
1311 
1312 static int fuse_dev_open(struct inode *inode, struct file *file)
1313 {
1314 	/*
1315 	 * The fuse device's file's private_data is used to hold
1316 	 * the fuse_conn(ection) when it is mounted, and is used to
1317 	 * keep track of whether the file has been mounted already.
1318 	 */
1319 	file->private_data = NULL;
1320 	return 0;
1321 }
1322 
1323 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1324 {
1325 	struct fuse_copy_state cs;
1326 	struct file *file = iocb->ki_filp;
1327 	struct fuse_dev *fud = fuse_get_dev(file);
1328 
1329 	if (!fud)
1330 		return -EPERM;
1331 
1332 	if (!iter_is_iovec(to))
1333 		return -EINVAL;
1334 
1335 	fuse_copy_init(&cs, 1, to);
1336 
1337 	return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1338 }
1339 
1340 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1341 				    struct pipe_inode_info *pipe,
1342 				    size_t len, unsigned int flags)
1343 {
1344 	int ret;
1345 	int page_nr = 0;
1346 	int do_wakeup = 0;
1347 	struct pipe_buffer *bufs;
1348 	struct fuse_copy_state cs;
1349 	struct fuse_dev *fud = fuse_get_dev(in);
1350 
1351 	if (!fud)
1352 		return -EPERM;
1353 
1354 	bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1355 	if (!bufs)
1356 		return -ENOMEM;
1357 
1358 	fuse_copy_init(&cs, 1, NULL);
1359 	cs.pipebufs = bufs;
1360 	cs.pipe = pipe;
1361 	ret = fuse_dev_do_read(fud, in, &cs, len);
1362 	if (ret < 0)
1363 		goto out;
1364 
1365 	ret = 0;
1366 	pipe_lock(pipe);
1367 
1368 	if (!pipe->readers) {
1369 		send_sig(SIGPIPE, current, 0);
1370 		if (!ret)
1371 			ret = -EPIPE;
1372 		goto out_unlock;
1373 	}
1374 
1375 	if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1376 		ret = -EIO;
1377 		goto out_unlock;
1378 	}
1379 
1380 	while (page_nr < cs.nr_segs) {
1381 		int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1382 		struct pipe_buffer *buf = pipe->bufs + newbuf;
1383 
1384 		buf->page = bufs[page_nr].page;
1385 		buf->offset = bufs[page_nr].offset;
1386 		buf->len = bufs[page_nr].len;
1387 		/*
1388 		 * Need to be careful about this.  Having buf->ops in module
1389 		 * code can Oops if the buffer persists after module unload.
1390 		 */
1391 		buf->ops = &nosteal_pipe_buf_ops;
1392 
1393 		pipe->nrbufs++;
1394 		page_nr++;
1395 		ret += buf->len;
1396 
1397 		if (pipe->files)
1398 			do_wakeup = 1;
1399 	}
1400 
1401 out_unlock:
1402 	pipe_unlock(pipe);
1403 
1404 	if (do_wakeup) {
1405 		smp_mb();
1406 		if (waitqueue_active(&pipe->wait))
1407 			wake_up_interruptible(&pipe->wait);
1408 		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1409 	}
1410 
1411 out:
1412 	for (; page_nr < cs.nr_segs; page_nr++)
1413 		put_page(bufs[page_nr].page);
1414 
1415 	kfree(bufs);
1416 	return ret;
1417 }
1418 
1419 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1420 			    struct fuse_copy_state *cs)
1421 {
1422 	struct fuse_notify_poll_wakeup_out outarg;
1423 	int err = -EINVAL;
1424 
1425 	if (size != sizeof(outarg))
1426 		goto err;
1427 
1428 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1429 	if (err)
1430 		goto err;
1431 
1432 	fuse_copy_finish(cs);
1433 	return fuse_notify_poll_wakeup(fc, &outarg);
1434 
1435 err:
1436 	fuse_copy_finish(cs);
1437 	return err;
1438 }
1439 
1440 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1441 				   struct fuse_copy_state *cs)
1442 {
1443 	struct fuse_notify_inval_inode_out outarg;
1444 	int err = -EINVAL;
1445 
1446 	if (size != sizeof(outarg))
1447 		goto err;
1448 
1449 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1450 	if (err)
1451 		goto err;
1452 	fuse_copy_finish(cs);
1453 
1454 	down_read(&fc->killsb);
1455 	err = -ENOENT;
1456 	if (fc->sb) {
1457 		err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1458 					       outarg.off, outarg.len);
1459 	}
1460 	up_read(&fc->killsb);
1461 	return err;
1462 
1463 err:
1464 	fuse_copy_finish(cs);
1465 	return err;
1466 }
1467 
1468 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1469 				   struct fuse_copy_state *cs)
1470 {
1471 	struct fuse_notify_inval_entry_out outarg;
1472 	int err = -ENOMEM;
1473 	char *buf;
1474 	struct qstr name;
1475 
1476 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1477 	if (!buf)
1478 		goto err;
1479 
1480 	err = -EINVAL;
1481 	if (size < sizeof(outarg))
1482 		goto err;
1483 
1484 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1485 	if (err)
1486 		goto err;
1487 
1488 	err = -ENAMETOOLONG;
1489 	if (outarg.namelen > FUSE_NAME_MAX)
1490 		goto err;
1491 
1492 	err = -EINVAL;
1493 	if (size != sizeof(outarg) + outarg.namelen + 1)
1494 		goto err;
1495 
1496 	name.name = buf;
1497 	name.len = outarg.namelen;
1498 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1499 	if (err)
1500 		goto err;
1501 	fuse_copy_finish(cs);
1502 	buf[outarg.namelen] = 0;
1503 
1504 	down_read(&fc->killsb);
1505 	err = -ENOENT;
1506 	if (fc->sb)
1507 		err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1508 	up_read(&fc->killsb);
1509 	kfree(buf);
1510 	return err;
1511 
1512 err:
1513 	kfree(buf);
1514 	fuse_copy_finish(cs);
1515 	return err;
1516 }
1517 
1518 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1519 			      struct fuse_copy_state *cs)
1520 {
1521 	struct fuse_notify_delete_out outarg;
1522 	int err = -ENOMEM;
1523 	char *buf;
1524 	struct qstr name;
1525 
1526 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1527 	if (!buf)
1528 		goto err;
1529 
1530 	err = -EINVAL;
1531 	if (size < sizeof(outarg))
1532 		goto err;
1533 
1534 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1535 	if (err)
1536 		goto err;
1537 
1538 	err = -ENAMETOOLONG;
1539 	if (outarg.namelen > FUSE_NAME_MAX)
1540 		goto err;
1541 
1542 	err = -EINVAL;
1543 	if (size != sizeof(outarg) + outarg.namelen + 1)
1544 		goto err;
1545 
1546 	name.name = buf;
1547 	name.len = outarg.namelen;
1548 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1549 	if (err)
1550 		goto err;
1551 	fuse_copy_finish(cs);
1552 	buf[outarg.namelen] = 0;
1553 
1554 	down_read(&fc->killsb);
1555 	err = -ENOENT;
1556 	if (fc->sb)
1557 		err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1558 					       outarg.child, &name);
1559 	up_read(&fc->killsb);
1560 	kfree(buf);
1561 	return err;
1562 
1563 err:
1564 	kfree(buf);
1565 	fuse_copy_finish(cs);
1566 	return err;
1567 }
1568 
1569 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1570 			     struct fuse_copy_state *cs)
1571 {
1572 	struct fuse_notify_store_out outarg;
1573 	struct inode *inode;
1574 	struct address_space *mapping;
1575 	u64 nodeid;
1576 	int err;
1577 	pgoff_t index;
1578 	unsigned int offset;
1579 	unsigned int num;
1580 	loff_t file_size;
1581 	loff_t end;
1582 
1583 	err = -EINVAL;
1584 	if (size < sizeof(outarg))
1585 		goto out_finish;
1586 
1587 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1588 	if (err)
1589 		goto out_finish;
1590 
1591 	err = -EINVAL;
1592 	if (size - sizeof(outarg) != outarg.size)
1593 		goto out_finish;
1594 
1595 	nodeid = outarg.nodeid;
1596 
1597 	down_read(&fc->killsb);
1598 
1599 	err = -ENOENT;
1600 	if (!fc->sb)
1601 		goto out_up_killsb;
1602 
1603 	inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1604 	if (!inode)
1605 		goto out_up_killsb;
1606 
1607 	mapping = inode->i_mapping;
1608 	index = outarg.offset >> PAGE_SHIFT;
1609 	offset = outarg.offset & ~PAGE_MASK;
1610 	file_size = i_size_read(inode);
1611 	end = outarg.offset + outarg.size;
1612 	if (end > file_size) {
1613 		file_size = end;
1614 		fuse_write_update_size(inode, file_size);
1615 	}
1616 
1617 	num = outarg.size;
1618 	while (num) {
1619 		struct page *page;
1620 		unsigned int this_num;
1621 
1622 		err = -ENOMEM;
1623 		page = find_or_create_page(mapping, index,
1624 					   mapping_gfp_mask(mapping));
1625 		if (!page)
1626 			goto out_iput;
1627 
1628 		this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1629 		err = fuse_copy_page(cs, &page, offset, this_num, 0);
1630 		if (!err && offset == 0 &&
1631 		    (this_num == PAGE_SIZE || file_size == end))
1632 			SetPageUptodate(page);
1633 		unlock_page(page);
1634 		put_page(page);
1635 
1636 		if (err)
1637 			goto out_iput;
1638 
1639 		num -= this_num;
1640 		offset = 0;
1641 		index++;
1642 	}
1643 
1644 	err = 0;
1645 
1646 out_iput:
1647 	iput(inode);
1648 out_up_killsb:
1649 	up_read(&fc->killsb);
1650 out_finish:
1651 	fuse_copy_finish(cs);
1652 	return err;
1653 }
1654 
1655 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1656 {
1657 	release_pages(req->pages, req->num_pages, false);
1658 }
1659 
1660 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1661 			 struct fuse_notify_retrieve_out *outarg)
1662 {
1663 	int err;
1664 	struct address_space *mapping = inode->i_mapping;
1665 	struct fuse_req *req;
1666 	pgoff_t index;
1667 	loff_t file_size;
1668 	unsigned int num;
1669 	unsigned int offset;
1670 	size_t total_len = 0;
1671 	int num_pages;
1672 
1673 	offset = outarg->offset & ~PAGE_MASK;
1674 	file_size = i_size_read(inode);
1675 
1676 	num = outarg->size;
1677 	if (outarg->offset > file_size)
1678 		num = 0;
1679 	else if (outarg->offset + num > file_size)
1680 		num = file_size - outarg->offset;
1681 
1682 	num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1683 	num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1684 
1685 	req = fuse_get_req(fc, num_pages);
1686 	if (IS_ERR(req))
1687 		return PTR_ERR(req);
1688 
1689 	req->in.h.opcode = FUSE_NOTIFY_REPLY;
1690 	req->in.h.nodeid = outarg->nodeid;
1691 	req->in.numargs = 2;
1692 	req->in.argpages = 1;
1693 	req->page_descs[0].offset = offset;
1694 	req->end = fuse_retrieve_end;
1695 
1696 	index = outarg->offset >> PAGE_SHIFT;
1697 
1698 	while (num && req->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 		req->pages[req->num_pages] = page;
1708 		req->page_descs[req->num_pages].length = this_num;
1709 		req->num_pages++;
1710 
1711 		offset = 0;
1712 		num -= this_num;
1713 		total_len += this_num;
1714 		index++;
1715 	}
1716 	req->misc.retrieve_in.offset = outarg->offset;
1717 	req->misc.retrieve_in.size = total_len;
1718 	req->in.args[0].size = sizeof(req->misc.retrieve_in);
1719 	req->in.args[0].value = &req->misc.retrieve_in;
1720 	req->in.args[1].size = total_len;
1721 
1722 	err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1723 	if (err)
1724 		fuse_retrieve_end(fc, req);
1725 
1726 	return err;
1727 }
1728 
1729 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1730 				struct fuse_copy_state *cs)
1731 {
1732 	struct fuse_notify_retrieve_out outarg;
1733 	struct inode *inode;
1734 	int err;
1735 
1736 	err = -EINVAL;
1737 	if (size != sizeof(outarg))
1738 		goto copy_finish;
1739 
1740 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1741 	if (err)
1742 		goto copy_finish;
1743 
1744 	fuse_copy_finish(cs);
1745 
1746 	down_read(&fc->killsb);
1747 	err = -ENOENT;
1748 	if (fc->sb) {
1749 		u64 nodeid = outarg.nodeid;
1750 
1751 		inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1752 		if (inode) {
1753 			err = fuse_retrieve(fc, inode, &outarg);
1754 			iput(inode);
1755 		}
1756 	}
1757 	up_read(&fc->killsb);
1758 
1759 	return err;
1760 
1761 copy_finish:
1762 	fuse_copy_finish(cs);
1763 	return err;
1764 }
1765 
1766 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1767 		       unsigned int size, struct fuse_copy_state *cs)
1768 {
1769 	/* Don't try to move pages (yet) */
1770 	cs->move_pages = 0;
1771 
1772 	switch (code) {
1773 	case FUSE_NOTIFY_POLL:
1774 		return fuse_notify_poll(fc, size, cs);
1775 
1776 	case FUSE_NOTIFY_INVAL_INODE:
1777 		return fuse_notify_inval_inode(fc, size, cs);
1778 
1779 	case FUSE_NOTIFY_INVAL_ENTRY:
1780 		return fuse_notify_inval_entry(fc, size, cs);
1781 
1782 	case FUSE_NOTIFY_STORE:
1783 		return fuse_notify_store(fc, size, cs);
1784 
1785 	case FUSE_NOTIFY_RETRIEVE:
1786 		return fuse_notify_retrieve(fc, size, cs);
1787 
1788 	case FUSE_NOTIFY_DELETE:
1789 		return fuse_notify_delete(fc, size, cs);
1790 
1791 	default:
1792 		fuse_copy_finish(cs);
1793 		return -EINVAL;
1794 	}
1795 }
1796 
1797 /* Look up request on processing list by unique ID */
1798 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1799 {
1800 	struct fuse_req *req;
1801 
1802 	list_for_each_entry(req, &fpq->processing, list) {
1803 		if (req->in.h.unique == unique || req->intr_unique == unique)
1804 			return req;
1805 	}
1806 	return NULL;
1807 }
1808 
1809 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1810 			 unsigned nbytes)
1811 {
1812 	unsigned reqsize = sizeof(struct fuse_out_header);
1813 
1814 	if (out->h.error)
1815 		return nbytes != reqsize ? -EINVAL : 0;
1816 
1817 	reqsize += len_args(out->numargs, out->args);
1818 
1819 	if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1820 		return -EINVAL;
1821 	else if (reqsize > nbytes) {
1822 		struct fuse_arg *lastarg = &out->args[out->numargs-1];
1823 		unsigned diffsize = reqsize - nbytes;
1824 		if (diffsize > lastarg->size)
1825 			return -EINVAL;
1826 		lastarg->size -= diffsize;
1827 	}
1828 	return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1829 			      out->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 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 	if (nbytes < sizeof(struct fuse_out_header))
1849 		return -EINVAL;
1850 
1851 	err = fuse_copy_one(cs, &oh, sizeof(oh));
1852 	if (err)
1853 		goto err_finish;
1854 
1855 	err = -EINVAL;
1856 	if (oh.len != nbytes)
1857 		goto err_finish;
1858 
1859 	/*
1860 	 * Zero oh.unique indicates unsolicited notification message
1861 	 * and error contains notification code.
1862 	 */
1863 	if (!oh.unique) {
1864 		err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1865 		return err ? err : nbytes;
1866 	}
1867 
1868 	err = -EINVAL;
1869 	if (oh.error <= -1000 || oh.error > 0)
1870 		goto err_finish;
1871 
1872 	spin_lock(&fpq->lock);
1873 	err = -ENOENT;
1874 	if (!fpq->connected)
1875 		goto err_unlock_pq;
1876 
1877 	req = request_find(fpq, oh.unique);
1878 	if (!req)
1879 		goto err_unlock_pq;
1880 
1881 	/* Is it an interrupt reply? */
1882 	if (req->intr_unique == oh.unique) {
1883 		spin_unlock(&fpq->lock);
1884 
1885 		err = -EINVAL;
1886 		if (nbytes != sizeof(struct fuse_out_header))
1887 			goto err_finish;
1888 
1889 		if (oh.error == -ENOSYS)
1890 			fc->no_interrupt = 1;
1891 		else if (oh.error == -EAGAIN)
1892 			queue_interrupt(&fc->iq, req);
1893 
1894 		fuse_copy_finish(cs);
1895 		return nbytes;
1896 	}
1897 
1898 	clear_bit(FR_SENT, &req->flags);
1899 	list_move(&req->list, &fpq->io);
1900 	req->out.h = oh;
1901 	set_bit(FR_LOCKED, &req->flags);
1902 	spin_unlock(&fpq->lock);
1903 	cs->req = req;
1904 	if (!req->out.page_replace)
1905 		cs->move_pages = 0;
1906 
1907 	err = copy_out_args(cs, &req->out, nbytes);
1908 	fuse_copy_finish(cs);
1909 
1910 	spin_lock(&fpq->lock);
1911 	clear_bit(FR_LOCKED, &req->flags);
1912 	if (!fpq->connected)
1913 		err = -ENOENT;
1914 	else if (err)
1915 		req->out.h.error = -EIO;
1916 	if (!test_bit(FR_PRIVATE, &req->flags))
1917 		list_del_init(&req->list);
1918 	spin_unlock(&fpq->lock);
1919 
1920 	request_end(fc, req);
1921 
1922 	return err ? err : nbytes;
1923 
1924  err_unlock_pq:
1925 	spin_unlock(&fpq->lock);
1926  err_finish:
1927 	fuse_copy_finish(cs);
1928 	return err;
1929 }
1930 
1931 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1932 {
1933 	struct fuse_copy_state cs;
1934 	struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1935 
1936 	if (!fud)
1937 		return -EPERM;
1938 
1939 	if (!iter_is_iovec(from))
1940 		return -EINVAL;
1941 
1942 	fuse_copy_init(&cs, 0, from);
1943 
1944 	return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1945 }
1946 
1947 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1948 				     struct file *out, loff_t *ppos,
1949 				     size_t len, unsigned int flags)
1950 {
1951 	unsigned nbuf;
1952 	unsigned idx;
1953 	struct pipe_buffer *bufs;
1954 	struct fuse_copy_state cs;
1955 	struct fuse_dev *fud;
1956 	size_t rem;
1957 	ssize_t ret;
1958 
1959 	fud = fuse_get_dev(out);
1960 	if (!fud)
1961 		return -EPERM;
1962 
1963 	bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1964 	if (!bufs)
1965 		return -ENOMEM;
1966 
1967 	pipe_lock(pipe);
1968 	nbuf = 0;
1969 	rem = 0;
1970 	for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1971 		rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1972 
1973 	ret = -EINVAL;
1974 	if (rem < len) {
1975 		pipe_unlock(pipe);
1976 		goto out;
1977 	}
1978 
1979 	rem = len;
1980 	while (rem) {
1981 		struct pipe_buffer *ibuf;
1982 		struct pipe_buffer *obuf;
1983 
1984 		BUG_ON(nbuf >= pipe->buffers);
1985 		BUG_ON(!pipe->nrbufs);
1986 		ibuf = &pipe->bufs[pipe->curbuf];
1987 		obuf = &bufs[nbuf];
1988 
1989 		if (rem >= ibuf->len) {
1990 			*obuf = *ibuf;
1991 			ibuf->ops = NULL;
1992 			pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1993 			pipe->nrbufs--;
1994 		} else {
1995 			ibuf->ops->get(pipe, ibuf);
1996 			*obuf = *ibuf;
1997 			obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1998 			obuf->len = rem;
1999 			ibuf->offset += obuf->len;
2000 			ibuf->len -= obuf->len;
2001 		}
2002 		nbuf++;
2003 		rem -= obuf->len;
2004 	}
2005 	pipe_unlock(pipe);
2006 
2007 	fuse_copy_init(&cs, 0, NULL);
2008 	cs.pipebufs = bufs;
2009 	cs.nr_segs = nbuf;
2010 	cs.pipe = pipe;
2011 
2012 	if (flags & SPLICE_F_MOVE)
2013 		cs.move_pages = 1;
2014 
2015 	ret = fuse_dev_do_write(fud, &cs, len);
2016 
2017 	for (idx = 0; idx < nbuf; idx++) {
2018 		struct pipe_buffer *buf = &bufs[idx];
2019 		buf->ops->release(pipe, buf);
2020 	}
2021 out:
2022 	kfree(bufs);
2023 	return ret;
2024 }
2025 
2026 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2027 {
2028 	unsigned mask = POLLOUT | POLLWRNORM;
2029 	struct fuse_iqueue *fiq;
2030 	struct fuse_dev *fud = fuse_get_dev(file);
2031 
2032 	if (!fud)
2033 		return POLLERR;
2034 
2035 	fiq = &fud->fc->iq;
2036 	poll_wait(file, &fiq->waitq, wait);
2037 
2038 	spin_lock(&fiq->waitq.lock);
2039 	if (!fiq->connected)
2040 		mask = POLLERR;
2041 	else if (request_pending(fiq))
2042 		mask |= POLLIN | POLLRDNORM;
2043 	spin_unlock(&fiq->waitq.lock);
2044 
2045 	return mask;
2046 }
2047 
2048 /*
2049  * Abort all requests on the given list (pending or processing)
2050  *
2051  * This function releases and reacquires fc->lock
2052  */
2053 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2054 {
2055 	while (!list_empty(head)) {
2056 		struct fuse_req *req;
2057 		req = list_entry(head->next, struct fuse_req, list);
2058 		req->out.h.error = -ECONNABORTED;
2059 		clear_bit(FR_PENDING, &req->flags);
2060 		clear_bit(FR_SENT, &req->flags);
2061 		list_del_init(&req->list);
2062 		request_end(fc, req);
2063 	}
2064 }
2065 
2066 static void end_polls(struct fuse_conn *fc)
2067 {
2068 	struct rb_node *p;
2069 
2070 	p = rb_first(&fc->polled_files);
2071 
2072 	while (p) {
2073 		struct fuse_file *ff;
2074 		ff = rb_entry(p, struct fuse_file, polled_node);
2075 		wake_up_interruptible_all(&ff->poll_wait);
2076 
2077 		p = rb_next(p);
2078 	}
2079 }
2080 
2081 /*
2082  * Abort all requests.
2083  *
2084  * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2085  * filesystem.
2086  *
2087  * The same effect is usually achievable through killing the filesystem daemon
2088  * and all users of the filesystem.  The exception is the combination of an
2089  * asynchronous request and the tricky deadlock (see
2090  * Documentation/filesystems/fuse.txt).
2091  *
2092  * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2093  * requests, they should be finished off immediately.  Locked requests will be
2094  * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2095  * requests.  It is possible that some request will finish before we can.  This
2096  * is OK, the request will in that case be removed from the list before we touch
2097  * it.
2098  */
2099 void fuse_abort_conn(struct fuse_conn *fc)
2100 {
2101 	struct fuse_iqueue *fiq = &fc->iq;
2102 
2103 	spin_lock(&fc->lock);
2104 	if (fc->connected) {
2105 		struct fuse_dev *fud;
2106 		struct fuse_req *req, *next;
2107 		LIST_HEAD(to_end1);
2108 		LIST_HEAD(to_end2);
2109 
2110 		fc->connected = 0;
2111 		fc->blocked = 0;
2112 		fuse_set_initialized(fc);
2113 		list_for_each_entry(fud, &fc->devices, entry) {
2114 			struct fuse_pqueue *fpq = &fud->pq;
2115 
2116 			spin_lock(&fpq->lock);
2117 			fpq->connected = 0;
2118 			list_for_each_entry_safe(req, next, &fpq->io, list) {
2119 				req->out.h.error = -ECONNABORTED;
2120 				spin_lock(&req->waitq.lock);
2121 				set_bit(FR_ABORTED, &req->flags);
2122 				if (!test_bit(FR_LOCKED, &req->flags)) {
2123 					set_bit(FR_PRIVATE, &req->flags);
2124 					list_move(&req->list, &to_end1);
2125 				}
2126 				spin_unlock(&req->waitq.lock);
2127 			}
2128 			list_splice_init(&fpq->processing, &to_end2);
2129 			spin_unlock(&fpq->lock);
2130 		}
2131 		fc->max_background = UINT_MAX;
2132 		flush_bg_queue(fc);
2133 
2134 		spin_lock(&fiq->waitq.lock);
2135 		fiq->connected = 0;
2136 		list_splice_init(&fiq->pending, &to_end2);
2137 		while (forget_pending(fiq))
2138 			kfree(dequeue_forget(fiq, 1, NULL));
2139 		wake_up_all_locked(&fiq->waitq);
2140 		spin_unlock(&fiq->waitq.lock);
2141 		kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2142 		end_polls(fc);
2143 		wake_up_all(&fc->blocked_waitq);
2144 		spin_unlock(&fc->lock);
2145 
2146 		while (!list_empty(&to_end1)) {
2147 			req = list_first_entry(&to_end1, struct fuse_req, list);
2148 			__fuse_get_request(req);
2149 			list_del_init(&req->list);
2150 			request_end(fc, req);
2151 		}
2152 		end_requests(fc, &to_end2);
2153 	} else {
2154 		spin_unlock(&fc->lock);
2155 	}
2156 }
2157 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2158 
2159 int fuse_dev_release(struct inode *inode, struct file *file)
2160 {
2161 	struct fuse_dev *fud = fuse_get_dev(file);
2162 
2163 	if (fud) {
2164 		struct fuse_conn *fc = fud->fc;
2165 		struct fuse_pqueue *fpq = &fud->pq;
2166 
2167 		WARN_ON(!list_empty(&fpq->io));
2168 		end_requests(fc, &fpq->processing);
2169 		/* Are we the last open device? */
2170 		if (atomic_dec_and_test(&fc->dev_count)) {
2171 			WARN_ON(fc->iq.fasync != NULL);
2172 			fuse_abort_conn(fc);
2173 		}
2174 		fuse_dev_free(fud);
2175 	}
2176 	return 0;
2177 }
2178 EXPORT_SYMBOL_GPL(fuse_dev_release);
2179 
2180 static int fuse_dev_fasync(int fd, struct file *file, int on)
2181 {
2182 	struct fuse_dev *fud = fuse_get_dev(file);
2183 
2184 	if (!fud)
2185 		return -EPERM;
2186 
2187 	/* No locking - fasync_helper does its own locking */
2188 	return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2189 }
2190 
2191 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2192 {
2193 	struct fuse_dev *fud;
2194 
2195 	if (new->private_data)
2196 		return -EINVAL;
2197 
2198 	fud = fuse_dev_alloc(fc);
2199 	if (!fud)
2200 		return -ENOMEM;
2201 
2202 	new->private_data = fud;
2203 	atomic_inc(&fc->dev_count);
2204 
2205 	return 0;
2206 }
2207 
2208 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2209 			   unsigned long arg)
2210 {
2211 	int err = -ENOTTY;
2212 
2213 	if (cmd == FUSE_DEV_IOC_CLONE) {
2214 		int oldfd;
2215 
2216 		err = -EFAULT;
2217 		if (!get_user(oldfd, (__u32 __user *) arg)) {
2218 			struct file *old = fget(oldfd);
2219 
2220 			err = -EINVAL;
2221 			if (old) {
2222 				struct fuse_dev *fud = NULL;
2223 
2224 				/*
2225 				 * Check against file->f_op because CUSE
2226 				 * uses the same ioctl handler.
2227 				 */
2228 				if (old->f_op == file->f_op &&
2229 				    old->f_cred->user_ns == file->f_cred->user_ns)
2230 					fud = fuse_get_dev(old);
2231 
2232 				if (fud) {
2233 					mutex_lock(&fuse_mutex);
2234 					err = fuse_device_clone(fud->fc, file);
2235 					mutex_unlock(&fuse_mutex);
2236 				}
2237 				fput(old);
2238 			}
2239 		}
2240 	}
2241 	return err;
2242 }
2243 
2244 const struct file_operations fuse_dev_operations = {
2245 	.owner		= THIS_MODULE,
2246 	.open		= fuse_dev_open,
2247 	.llseek		= no_llseek,
2248 	.read_iter	= fuse_dev_read,
2249 	.splice_read	= fuse_dev_splice_read,
2250 	.write_iter	= fuse_dev_write,
2251 	.splice_write	= fuse_dev_splice_write,
2252 	.poll		= fuse_dev_poll,
2253 	.release	= fuse_dev_release,
2254 	.fasync		= fuse_dev_fasync,
2255 	.unlocked_ioctl = fuse_dev_ioctl,
2256 	.compat_ioctl   = fuse_dev_ioctl,
2257 };
2258 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2259 
2260 static struct miscdevice fuse_miscdevice = {
2261 	.minor = FUSE_MINOR,
2262 	.name  = "fuse",
2263 	.fops = &fuse_dev_operations,
2264 };
2265 
2266 int __init fuse_dev_init(void)
2267 {
2268 	int err = -ENOMEM;
2269 	fuse_req_cachep = kmem_cache_create("fuse_request",
2270 					    sizeof(struct fuse_req),
2271 					    0, 0, NULL);
2272 	if (!fuse_req_cachep)
2273 		goto out;
2274 
2275 	err = misc_register(&fuse_miscdevice);
2276 	if (err)
2277 		goto out_cache_clean;
2278 
2279 	return 0;
2280 
2281  out_cache_clean:
2282 	kmem_cache_destroy(fuse_req_cachep);
2283  out:
2284 	return err;
2285 }
2286 
2287 void fuse_dev_cleanup(void)
2288 {
2289 	misc_deregister(&fuse_miscdevice);
2290 	kmem_cache_destroy(fuse_req_cachep);
2291 }
2292