xref: /openbmc/linux/fs/fuse/dev.c (revision 80483c3a)
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 		cs->offset = off;
771 		iov_iter_advance(cs->iter, err);
772 	}
773 
774 	return lock_request(cs->req);
775 }
776 
777 /* Do as much copy to/from userspace buffer as we can */
778 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
779 {
780 	unsigned ncpy = min(*size, cs->len);
781 	if (val) {
782 		void *pgaddr = kmap_atomic(cs->pg);
783 		void *buf = pgaddr + cs->offset;
784 
785 		if (cs->write)
786 			memcpy(buf, *val, ncpy);
787 		else
788 			memcpy(*val, buf, ncpy);
789 
790 		kunmap_atomic(pgaddr);
791 		*val += ncpy;
792 	}
793 	*size -= ncpy;
794 	cs->len -= ncpy;
795 	cs->offset += ncpy;
796 	return ncpy;
797 }
798 
799 static int fuse_check_page(struct page *page)
800 {
801 	if (page_mapcount(page) ||
802 	    page->mapping != NULL ||
803 	    page_count(page) != 1 ||
804 	    (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
805 	     ~(1 << PG_locked |
806 	       1 << PG_referenced |
807 	       1 << PG_uptodate |
808 	       1 << PG_lru |
809 	       1 << PG_active |
810 	       1 << PG_reclaim))) {
811 		printk(KERN_WARNING "fuse: trying to steal weird page\n");
812 		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);
813 		return 1;
814 	}
815 	return 0;
816 }
817 
818 static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
819 {
820 	int err;
821 	struct page *oldpage = *pagep;
822 	struct page *newpage;
823 	struct pipe_buffer *buf = cs->pipebufs;
824 
825 	err = unlock_request(cs->req);
826 	if (err)
827 		return err;
828 
829 	fuse_copy_finish(cs);
830 
831 	err = buf->ops->confirm(cs->pipe, buf);
832 	if (err)
833 		return err;
834 
835 	BUG_ON(!cs->nr_segs);
836 	cs->currbuf = buf;
837 	cs->len = buf->len;
838 	cs->pipebufs++;
839 	cs->nr_segs--;
840 
841 	if (cs->len != PAGE_SIZE)
842 		goto out_fallback;
843 
844 	if (buf->ops->steal(cs->pipe, buf) != 0)
845 		goto out_fallback;
846 
847 	newpage = buf->page;
848 
849 	if (!PageUptodate(newpage))
850 		SetPageUptodate(newpage);
851 
852 	ClearPageMappedToDisk(newpage);
853 
854 	if (fuse_check_page(newpage) != 0)
855 		goto out_fallback_unlock;
856 
857 	/*
858 	 * This is a new and locked page, it shouldn't be mapped or
859 	 * have any special flags on it
860 	 */
861 	if (WARN_ON(page_mapped(oldpage)))
862 		goto out_fallback_unlock;
863 	if (WARN_ON(page_has_private(oldpage)))
864 		goto out_fallback_unlock;
865 	if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
866 		goto out_fallback_unlock;
867 	if (WARN_ON(PageMlocked(oldpage)))
868 		goto out_fallback_unlock;
869 
870 	err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
871 	if (err) {
872 		unlock_page(newpage);
873 		return err;
874 	}
875 
876 	get_page(newpage);
877 
878 	if (!(buf->flags & PIPE_BUF_FLAG_LRU))
879 		lru_cache_add_file(newpage);
880 
881 	err = 0;
882 	spin_lock(&cs->req->waitq.lock);
883 	if (test_bit(FR_ABORTED, &cs->req->flags))
884 		err = -ENOENT;
885 	else
886 		*pagep = newpage;
887 	spin_unlock(&cs->req->waitq.lock);
888 
889 	if (err) {
890 		unlock_page(newpage);
891 		put_page(newpage);
892 		return err;
893 	}
894 
895 	unlock_page(oldpage);
896 	put_page(oldpage);
897 	cs->len = 0;
898 
899 	return 0;
900 
901 out_fallback_unlock:
902 	unlock_page(newpage);
903 out_fallback:
904 	cs->pg = buf->page;
905 	cs->offset = buf->offset;
906 
907 	err = lock_request(cs->req);
908 	if (err)
909 		return err;
910 
911 	return 1;
912 }
913 
914 static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
915 			 unsigned offset, unsigned count)
916 {
917 	struct pipe_buffer *buf;
918 	int err;
919 
920 	if (cs->nr_segs == cs->pipe->buffers)
921 		return -EIO;
922 
923 	err = unlock_request(cs->req);
924 	if (err)
925 		return err;
926 
927 	fuse_copy_finish(cs);
928 
929 	buf = cs->pipebufs;
930 	get_page(page);
931 	buf->page = page;
932 	buf->offset = offset;
933 	buf->len = count;
934 
935 	cs->pipebufs++;
936 	cs->nr_segs++;
937 	cs->len = 0;
938 
939 	return 0;
940 }
941 
942 /*
943  * Copy a page in the request to/from the userspace buffer.  Must be
944  * done atomically
945  */
946 static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
947 			  unsigned offset, unsigned count, int zeroing)
948 {
949 	int err;
950 	struct page *page = *pagep;
951 
952 	if (page && zeroing && count < PAGE_SIZE)
953 		clear_highpage(page);
954 
955 	while (count) {
956 		if (cs->write && cs->pipebufs && page) {
957 			return fuse_ref_page(cs, page, offset, count);
958 		} else if (!cs->len) {
959 			if (cs->move_pages && page &&
960 			    offset == 0 && count == PAGE_SIZE) {
961 				err = fuse_try_move_page(cs, pagep);
962 				if (err <= 0)
963 					return err;
964 			} else {
965 				err = fuse_copy_fill(cs);
966 				if (err)
967 					return err;
968 			}
969 		}
970 		if (page) {
971 			void *mapaddr = kmap_atomic(page);
972 			void *buf = mapaddr + offset;
973 			offset += fuse_copy_do(cs, &buf, &count);
974 			kunmap_atomic(mapaddr);
975 		} else
976 			offset += fuse_copy_do(cs, NULL, &count);
977 	}
978 	if (page && !cs->write)
979 		flush_dcache_page(page);
980 	return 0;
981 }
982 
983 /* Copy pages in the request to/from userspace buffer */
984 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
985 			   int zeroing)
986 {
987 	unsigned i;
988 	struct fuse_req *req = cs->req;
989 
990 	for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
991 		int err;
992 		unsigned offset = req->page_descs[i].offset;
993 		unsigned count = min(nbytes, req->page_descs[i].length);
994 
995 		err = fuse_copy_page(cs, &req->pages[i], offset, count,
996 				     zeroing);
997 		if (err)
998 			return err;
999 
1000 		nbytes -= count;
1001 	}
1002 	return 0;
1003 }
1004 
1005 /* Copy a single argument in the request to/from userspace buffer */
1006 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1007 {
1008 	while (size) {
1009 		if (!cs->len) {
1010 			int err = fuse_copy_fill(cs);
1011 			if (err)
1012 				return err;
1013 		}
1014 		fuse_copy_do(cs, &val, &size);
1015 	}
1016 	return 0;
1017 }
1018 
1019 /* Copy request arguments to/from userspace buffer */
1020 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1021 			  unsigned argpages, struct fuse_arg *args,
1022 			  int zeroing)
1023 {
1024 	int err = 0;
1025 	unsigned i;
1026 
1027 	for (i = 0; !err && i < numargs; i++)  {
1028 		struct fuse_arg *arg = &args[i];
1029 		if (i == numargs - 1 && argpages)
1030 			err = fuse_copy_pages(cs, arg->size, zeroing);
1031 		else
1032 			err = fuse_copy_one(cs, arg->value, arg->size);
1033 	}
1034 	return err;
1035 }
1036 
1037 static int forget_pending(struct fuse_iqueue *fiq)
1038 {
1039 	return fiq->forget_list_head.next != NULL;
1040 }
1041 
1042 static int request_pending(struct fuse_iqueue *fiq)
1043 {
1044 	return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1045 		forget_pending(fiq);
1046 }
1047 
1048 /*
1049  * Transfer an interrupt request to userspace
1050  *
1051  * Unlike other requests this is assembled on demand, without a need
1052  * to allocate a separate fuse_req structure.
1053  *
1054  * Called with fiq->waitq.lock held, releases it
1055  */
1056 static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1057 			       struct fuse_copy_state *cs,
1058 			       size_t nbytes, struct fuse_req *req)
1059 __releases(fiq->waitq.lock)
1060 {
1061 	struct fuse_in_header ih;
1062 	struct fuse_interrupt_in arg;
1063 	unsigned reqsize = sizeof(ih) + sizeof(arg);
1064 	int err;
1065 
1066 	list_del_init(&req->intr_entry);
1067 	req->intr_unique = fuse_get_unique(fiq);
1068 	memset(&ih, 0, sizeof(ih));
1069 	memset(&arg, 0, sizeof(arg));
1070 	ih.len = reqsize;
1071 	ih.opcode = FUSE_INTERRUPT;
1072 	ih.unique = req->intr_unique;
1073 	arg.unique = req->in.h.unique;
1074 
1075 	spin_unlock(&fiq->waitq.lock);
1076 	if (nbytes < reqsize)
1077 		return -EINVAL;
1078 
1079 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1080 	if (!err)
1081 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1082 	fuse_copy_finish(cs);
1083 
1084 	return err ? err : reqsize;
1085 }
1086 
1087 static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
1088 					       unsigned max,
1089 					       unsigned *countp)
1090 {
1091 	struct fuse_forget_link *head = fiq->forget_list_head.next;
1092 	struct fuse_forget_link **newhead = &head;
1093 	unsigned count;
1094 
1095 	for (count = 0; *newhead != NULL && count < max; count++)
1096 		newhead = &(*newhead)->next;
1097 
1098 	fiq->forget_list_head.next = *newhead;
1099 	*newhead = NULL;
1100 	if (fiq->forget_list_head.next == NULL)
1101 		fiq->forget_list_tail = &fiq->forget_list_head;
1102 
1103 	if (countp != NULL)
1104 		*countp = count;
1105 
1106 	return head;
1107 }
1108 
1109 static int fuse_read_single_forget(struct fuse_iqueue *fiq,
1110 				   struct fuse_copy_state *cs,
1111 				   size_t nbytes)
1112 __releases(fiq->waitq.lock)
1113 {
1114 	int err;
1115 	struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
1116 	struct fuse_forget_in arg = {
1117 		.nlookup = forget->forget_one.nlookup,
1118 	};
1119 	struct fuse_in_header ih = {
1120 		.opcode = FUSE_FORGET,
1121 		.nodeid = forget->forget_one.nodeid,
1122 		.unique = fuse_get_unique(fiq),
1123 		.len = sizeof(ih) + sizeof(arg),
1124 	};
1125 
1126 	spin_unlock(&fiq->waitq.lock);
1127 	kfree(forget);
1128 	if (nbytes < ih.len)
1129 		return -EINVAL;
1130 
1131 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1132 	if (!err)
1133 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1134 	fuse_copy_finish(cs);
1135 
1136 	if (err)
1137 		return err;
1138 
1139 	return ih.len;
1140 }
1141 
1142 static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
1143 				   struct fuse_copy_state *cs, size_t nbytes)
1144 __releases(fiq->waitq.lock)
1145 {
1146 	int err;
1147 	unsigned max_forgets;
1148 	unsigned count;
1149 	struct fuse_forget_link *head;
1150 	struct fuse_batch_forget_in arg = { .count = 0 };
1151 	struct fuse_in_header ih = {
1152 		.opcode = FUSE_BATCH_FORGET,
1153 		.unique = fuse_get_unique(fiq),
1154 		.len = sizeof(ih) + sizeof(arg),
1155 	};
1156 
1157 	if (nbytes < ih.len) {
1158 		spin_unlock(&fiq->waitq.lock);
1159 		return -EINVAL;
1160 	}
1161 
1162 	max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
1163 	head = dequeue_forget(fiq, max_forgets, &count);
1164 	spin_unlock(&fiq->waitq.lock);
1165 
1166 	arg.count = count;
1167 	ih.len += count * sizeof(struct fuse_forget_one);
1168 	err = fuse_copy_one(cs, &ih, sizeof(ih));
1169 	if (!err)
1170 		err = fuse_copy_one(cs, &arg, sizeof(arg));
1171 
1172 	while (head) {
1173 		struct fuse_forget_link *forget = head;
1174 
1175 		if (!err) {
1176 			err = fuse_copy_one(cs, &forget->forget_one,
1177 					    sizeof(forget->forget_one));
1178 		}
1179 		head = forget->next;
1180 		kfree(forget);
1181 	}
1182 
1183 	fuse_copy_finish(cs);
1184 
1185 	if (err)
1186 		return err;
1187 
1188 	return ih.len;
1189 }
1190 
1191 static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1192 			    struct fuse_copy_state *cs,
1193 			    size_t nbytes)
1194 __releases(fiq->waitq.lock)
1195 {
1196 	if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
1197 		return fuse_read_single_forget(fiq, cs, nbytes);
1198 	else
1199 		return fuse_read_batch_forget(fiq, cs, nbytes);
1200 }
1201 
1202 /*
1203  * Read a single request into the userspace filesystem's buffer.  This
1204  * function waits until a request is available, then removes it from
1205  * the pending list and copies request data to userspace buffer.  If
1206  * no reply is needed (FORGET) or request has been aborted or there
1207  * was an error during the copying then it's finished by calling
1208  * request_end().  Otherwise add it to the processing list, and set
1209  * the 'sent' flag.
1210  */
1211 static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
1212 				struct fuse_copy_state *cs, size_t nbytes)
1213 {
1214 	ssize_t err;
1215 	struct fuse_conn *fc = fud->fc;
1216 	struct fuse_iqueue *fiq = &fc->iq;
1217 	struct fuse_pqueue *fpq = &fud->pq;
1218 	struct fuse_req *req;
1219 	struct fuse_in *in;
1220 	unsigned reqsize;
1221 
1222  restart:
1223 	spin_lock(&fiq->waitq.lock);
1224 	err = -EAGAIN;
1225 	if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
1226 	    !request_pending(fiq))
1227 		goto err_unlock;
1228 
1229 	err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1230 				!fiq->connected || request_pending(fiq));
1231 	if (err)
1232 		goto err_unlock;
1233 
1234 	err = -ENODEV;
1235 	if (!fiq->connected)
1236 		goto err_unlock;
1237 
1238 	if (!list_empty(&fiq->interrupts)) {
1239 		req = list_entry(fiq->interrupts.next, struct fuse_req,
1240 				 intr_entry);
1241 		return fuse_read_interrupt(fiq, cs, nbytes, req);
1242 	}
1243 
1244 	if (forget_pending(fiq)) {
1245 		if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
1246 			return fuse_read_forget(fc, fiq, cs, nbytes);
1247 
1248 		if (fiq->forget_batch <= -8)
1249 			fiq->forget_batch = 16;
1250 	}
1251 
1252 	req = list_entry(fiq->pending.next, struct fuse_req, list);
1253 	clear_bit(FR_PENDING, &req->flags);
1254 	list_del_init(&req->list);
1255 	spin_unlock(&fiq->waitq.lock);
1256 
1257 	in = &req->in;
1258 	reqsize = in->h.len;
1259 	/* If request is too large, reply with an error and restart the read */
1260 	if (nbytes < reqsize) {
1261 		req->out.h.error = -EIO;
1262 		/* SETXATTR is special, since it may contain too large data */
1263 		if (in->h.opcode == FUSE_SETXATTR)
1264 			req->out.h.error = -E2BIG;
1265 		request_end(fc, req);
1266 		goto restart;
1267 	}
1268 	spin_lock(&fpq->lock);
1269 	list_add(&req->list, &fpq->io);
1270 	spin_unlock(&fpq->lock);
1271 	cs->req = req;
1272 	err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1273 	if (!err)
1274 		err = fuse_copy_args(cs, in->numargs, in->argpages,
1275 				     (struct fuse_arg *) in->args, 0);
1276 	fuse_copy_finish(cs);
1277 	spin_lock(&fpq->lock);
1278 	clear_bit(FR_LOCKED, &req->flags);
1279 	if (!fpq->connected) {
1280 		err = -ENODEV;
1281 		goto out_end;
1282 	}
1283 	if (err) {
1284 		req->out.h.error = -EIO;
1285 		goto out_end;
1286 	}
1287 	if (!test_bit(FR_ISREPLY, &req->flags)) {
1288 		err = reqsize;
1289 		goto out_end;
1290 	}
1291 	list_move_tail(&req->list, &fpq->processing);
1292 	spin_unlock(&fpq->lock);
1293 	set_bit(FR_SENT, &req->flags);
1294 	/* matches barrier in request_wait_answer() */
1295 	smp_mb__after_atomic();
1296 	if (test_bit(FR_INTERRUPTED, &req->flags))
1297 		queue_interrupt(fiq, req);
1298 
1299 	return reqsize;
1300 
1301 out_end:
1302 	if (!test_bit(FR_PRIVATE, &req->flags))
1303 		list_del_init(&req->list);
1304 	spin_unlock(&fpq->lock);
1305 	request_end(fc, req);
1306 	return err;
1307 
1308  err_unlock:
1309 	spin_unlock(&fiq->waitq.lock);
1310 	return err;
1311 }
1312 
1313 static int fuse_dev_open(struct inode *inode, struct file *file)
1314 {
1315 	/*
1316 	 * The fuse device's file's private_data is used to hold
1317 	 * the fuse_conn(ection) when it is mounted, and is used to
1318 	 * keep track of whether the file has been mounted already.
1319 	 */
1320 	file->private_data = NULL;
1321 	return 0;
1322 }
1323 
1324 static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
1325 {
1326 	struct fuse_copy_state cs;
1327 	struct file *file = iocb->ki_filp;
1328 	struct fuse_dev *fud = fuse_get_dev(file);
1329 
1330 	if (!fud)
1331 		return -EPERM;
1332 
1333 	if (!iter_is_iovec(to))
1334 		return -EINVAL;
1335 
1336 	fuse_copy_init(&cs, 1, to);
1337 
1338 	return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
1339 }
1340 
1341 static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1342 				    struct pipe_inode_info *pipe,
1343 				    size_t len, unsigned int flags)
1344 {
1345 	int ret;
1346 	int page_nr = 0;
1347 	int do_wakeup = 0;
1348 	struct pipe_buffer *bufs;
1349 	struct fuse_copy_state cs;
1350 	struct fuse_dev *fud = fuse_get_dev(in);
1351 
1352 	if (!fud)
1353 		return -EPERM;
1354 
1355 	bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1356 	if (!bufs)
1357 		return -ENOMEM;
1358 
1359 	fuse_copy_init(&cs, 1, NULL);
1360 	cs.pipebufs = bufs;
1361 	cs.pipe = pipe;
1362 	ret = fuse_dev_do_read(fud, in, &cs, len);
1363 	if (ret < 0)
1364 		goto out;
1365 
1366 	ret = 0;
1367 	pipe_lock(pipe);
1368 
1369 	if (!pipe->readers) {
1370 		send_sig(SIGPIPE, current, 0);
1371 		if (!ret)
1372 			ret = -EPIPE;
1373 		goto out_unlock;
1374 	}
1375 
1376 	if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1377 		ret = -EIO;
1378 		goto out_unlock;
1379 	}
1380 
1381 	while (page_nr < cs.nr_segs) {
1382 		int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1383 		struct pipe_buffer *buf = pipe->bufs + newbuf;
1384 
1385 		buf->page = bufs[page_nr].page;
1386 		buf->offset = bufs[page_nr].offset;
1387 		buf->len = bufs[page_nr].len;
1388 		/*
1389 		 * Need to be careful about this.  Having buf->ops in module
1390 		 * code can Oops if the buffer persists after module unload.
1391 		 */
1392 		buf->ops = &nosteal_pipe_buf_ops;
1393 
1394 		pipe->nrbufs++;
1395 		page_nr++;
1396 		ret += buf->len;
1397 
1398 		if (pipe->files)
1399 			do_wakeup = 1;
1400 	}
1401 
1402 out_unlock:
1403 	pipe_unlock(pipe);
1404 
1405 	if (do_wakeup) {
1406 		smp_mb();
1407 		if (waitqueue_active(&pipe->wait))
1408 			wake_up_interruptible(&pipe->wait);
1409 		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1410 	}
1411 
1412 out:
1413 	for (; page_nr < cs.nr_segs; page_nr++)
1414 		put_page(bufs[page_nr].page);
1415 
1416 	kfree(bufs);
1417 	return ret;
1418 }
1419 
1420 static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1421 			    struct fuse_copy_state *cs)
1422 {
1423 	struct fuse_notify_poll_wakeup_out outarg;
1424 	int err = -EINVAL;
1425 
1426 	if (size != sizeof(outarg))
1427 		goto err;
1428 
1429 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1430 	if (err)
1431 		goto err;
1432 
1433 	fuse_copy_finish(cs);
1434 	return fuse_notify_poll_wakeup(fc, &outarg);
1435 
1436 err:
1437 	fuse_copy_finish(cs);
1438 	return err;
1439 }
1440 
1441 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1442 				   struct fuse_copy_state *cs)
1443 {
1444 	struct fuse_notify_inval_inode_out outarg;
1445 	int err = -EINVAL;
1446 
1447 	if (size != sizeof(outarg))
1448 		goto err;
1449 
1450 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1451 	if (err)
1452 		goto err;
1453 	fuse_copy_finish(cs);
1454 
1455 	down_read(&fc->killsb);
1456 	err = -ENOENT;
1457 	if (fc->sb) {
1458 		err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1459 					       outarg.off, outarg.len);
1460 	}
1461 	up_read(&fc->killsb);
1462 	return err;
1463 
1464 err:
1465 	fuse_copy_finish(cs);
1466 	return err;
1467 }
1468 
1469 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1470 				   struct fuse_copy_state *cs)
1471 {
1472 	struct fuse_notify_inval_entry_out outarg;
1473 	int err = -ENOMEM;
1474 	char *buf;
1475 	struct qstr name;
1476 
1477 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1478 	if (!buf)
1479 		goto err;
1480 
1481 	err = -EINVAL;
1482 	if (size < sizeof(outarg))
1483 		goto err;
1484 
1485 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1486 	if (err)
1487 		goto err;
1488 
1489 	err = -ENAMETOOLONG;
1490 	if (outarg.namelen > FUSE_NAME_MAX)
1491 		goto err;
1492 
1493 	err = -EINVAL;
1494 	if (size != sizeof(outarg) + outarg.namelen + 1)
1495 		goto err;
1496 
1497 	name.name = buf;
1498 	name.len = outarg.namelen;
1499 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1500 	if (err)
1501 		goto err;
1502 	fuse_copy_finish(cs);
1503 	buf[outarg.namelen] = 0;
1504 
1505 	down_read(&fc->killsb);
1506 	err = -ENOENT;
1507 	if (fc->sb)
1508 		err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1509 	up_read(&fc->killsb);
1510 	kfree(buf);
1511 	return err;
1512 
1513 err:
1514 	kfree(buf);
1515 	fuse_copy_finish(cs);
1516 	return err;
1517 }
1518 
1519 static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1520 			      struct fuse_copy_state *cs)
1521 {
1522 	struct fuse_notify_delete_out outarg;
1523 	int err = -ENOMEM;
1524 	char *buf;
1525 	struct qstr name;
1526 
1527 	buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1528 	if (!buf)
1529 		goto err;
1530 
1531 	err = -EINVAL;
1532 	if (size < sizeof(outarg))
1533 		goto err;
1534 
1535 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1536 	if (err)
1537 		goto err;
1538 
1539 	err = -ENAMETOOLONG;
1540 	if (outarg.namelen > FUSE_NAME_MAX)
1541 		goto err;
1542 
1543 	err = -EINVAL;
1544 	if (size != sizeof(outarg) + outarg.namelen + 1)
1545 		goto err;
1546 
1547 	name.name = buf;
1548 	name.len = outarg.namelen;
1549 	err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1550 	if (err)
1551 		goto err;
1552 	fuse_copy_finish(cs);
1553 	buf[outarg.namelen] = 0;
1554 
1555 	down_read(&fc->killsb);
1556 	err = -ENOENT;
1557 	if (fc->sb)
1558 		err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1559 					       outarg.child, &name);
1560 	up_read(&fc->killsb);
1561 	kfree(buf);
1562 	return err;
1563 
1564 err:
1565 	kfree(buf);
1566 	fuse_copy_finish(cs);
1567 	return err;
1568 }
1569 
1570 static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1571 			     struct fuse_copy_state *cs)
1572 {
1573 	struct fuse_notify_store_out outarg;
1574 	struct inode *inode;
1575 	struct address_space *mapping;
1576 	u64 nodeid;
1577 	int err;
1578 	pgoff_t index;
1579 	unsigned int offset;
1580 	unsigned int num;
1581 	loff_t file_size;
1582 	loff_t end;
1583 
1584 	err = -EINVAL;
1585 	if (size < sizeof(outarg))
1586 		goto out_finish;
1587 
1588 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1589 	if (err)
1590 		goto out_finish;
1591 
1592 	err = -EINVAL;
1593 	if (size - sizeof(outarg) != outarg.size)
1594 		goto out_finish;
1595 
1596 	nodeid = outarg.nodeid;
1597 
1598 	down_read(&fc->killsb);
1599 
1600 	err = -ENOENT;
1601 	if (!fc->sb)
1602 		goto out_up_killsb;
1603 
1604 	inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1605 	if (!inode)
1606 		goto out_up_killsb;
1607 
1608 	mapping = inode->i_mapping;
1609 	index = outarg.offset >> PAGE_SHIFT;
1610 	offset = outarg.offset & ~PAGE_MASK;
1611 	file_size = i_size_read(inode);
1612 	end = outarg.offset + outarg.size;
1613 	if (end > file_size) {
1614 		file_size = end;
1615 		fuse_write_update_size(inode, file_size);
1616 	}
1617 
1618 	num = outarg.size;
1619 	while (num) {
1620 		struct page *page;
1621 		unsigned int this_num;
1622 
1623 		err = -ENOMEM;
1624 		page = find_or_create_page(mapping, index,
1625 					   mapping_gfp_mask(mapping));
1626 		if (!page)
1627 			goto out_iput;
1628 
1629 		this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1630 		err = fuse_copy_page(cs, &page, offset, this_num, 0);
1631 		if (!err && offset == 0 &&
1632 		    (this_num == PAGE_SIZE || file_size == end))
1633 			SetPageUptodate(page);
1634 		unlock_page(page);
1635 		put_page(page);
1636 
1637 		if (err)
1638 			goto out_iput;
1639 
1640 		num -= this_num;
1641 		offset = 0;
1642 		index++;
1643 	}
1644 
1645 	err = 0;
1646 
1647 out_iput:
1648 	iput(inode);
1649 out_up_killsb:
1650 	up_read(&fc->killsb);
1651 out_finish:
1652 	fuse_copy_finish(cs);
1653 	return err;
1654 }
1655 
1656 static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1657 {
1658 	release_pages(req->pages, req->num_pages, false);
1659 }
1660 
1661 static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1662 			 struct fuse_notify_retrieve_out *outarg)
1663 {
1664 	int err;
1665 	struct address_space *mapping = inode->i_mapping;
1666 	struct fuse_req *req;
1667 	pgoff_t index;
1668 	loff_t file_size;
1669 	unsigned int num;
1670 	unsigned int offset;
1671 	size_t total_len = 0;
1672 	int num_pages;
1673 
1674 	offset = outarg->offset & ~PAGE_MASK;
1675 	file_size = i_size_read(inode);
1676 
1677 	num = outarg->size;
1678 	if (outarg->offset > file_size)
1679 		num = 0;
1680 	else if (outarg->offset + num > file_size)
1681 		num = file_size - outarg->offset;
1682 
1683 	num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1684 	num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1685 
1686 	req = fuse_get_req(fc, num_pages);
1687 	if (IS_ERR(req))
1688 		return PTR_ERR(req);
1689 
1690 	req->in.h.opcode = FUSE_NOTIFY_REPLY;
1691 	req->in.h.nodeid = outarg->nodeid;
1692 	req->in.numargs = 2;
1693 	req->in.argpages = 1;
1694 	req->page_descs[0].offset = offset;
1695 	req->end = fuse_retrieve_end;
1696 
1697 	index = outarg->offset >> PAGE_SHIFT;
1698 
1699 	while (num && req->num_pages < num_pages) {
1700 		struct page *page;
1701 		unsigned int this_num;
1702 
1703 		page = find_get_page(mapping, index);
1704 		if (!page)
1705 			break;
1706 
1707 		this_num = min_t(unsigned, num, PAGE_SIZE - offset);
1708 		req->pages[req->num_pages] = page;
1709 		req->page_descs[req->num_pages].length = this_num;
1710 		req->num_pages++;
1711 
1712 		offset = 0;
1713 		num -= this_num;
1714 		total_len += this_num;
1715 		index++;
1716 	}
1717 	req->misc.retrieve_in.offset = outarg->offset;
1718 	req->misc.retrieve_in.size = total_len;
1719 	req->in.args[0].size = sizeof(req->misc.retrieve_in);
1720 	req->in.args[0].value = &req->misc.retrieve_in;
1721 	req->in.args[1].size = total_len;
1722 
1723 	err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1724 	if (err)
1725 		fuse_retrieve_end(fc, req);
1726 
1727 	return err;
1728 }
1729 
1730 static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1731 				struct fuse_copy_state *cs)
1732 {
1733 	struct fuse_notify_retrieve_out outarg;
1734 	struct inode *inode;
1735 	int err;
1736 
1737 	err = -EINVAL;
1738 	if (size != sizeof(outarg))
1739 		goto copy_finish;
1740 
1741 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1742 	if (err)
1743 		goto copy_finish;
1744 
1745 	fuse_copy_finish(cs);
1746 
1747 	down_read(&fc->killsb);
1748 	err = -ENOENT;
1749 	if (fc->sb) {
1750 		u64 nodeid = outarg.nodeid;
1751 
1752 		inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1753 		if (inode) {
1754 			err = fuse_retrieve(fc, inode, &outarg);
1755 			iput(inode);
1756 		}
1757 	}
1758 	up_read(&fc->killsb);
1759 
1760 	return err;
1761 
1762 copy_finish:
1763 	fuse_copy_finish(cs);
1764 	return err;
1765 }
1766 
1767 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1768 		       unsigned int size, struct fuse_copy_state *cs)
1769 {
1770 	/* Don't try to move pages (yet) */
1771 	cs->move_pages = 0;
1772 
1773 	switch (code) {
1774 	case FUSE_NOTIFY_POLL:
1775 		return fuse_notify_poll(fc, size, cs);
1776 
1777 	case FUSE_NOTIFY_INVAL_INODE:
1778 		return fuse_notify_inval_inode(fc, size, cs);
1779 
1780 	case FUSE_NOTIFY_INVAL_ENTRY:
1781 		return fuse_notify_inval_entry(fc, size, cs);
1782 
1783 	case FUSE_NOTIFY_STORE:
1784 		return fuse_notify_store(fc, size, cs);
1785 
1786 	case FUSE_NOTIFY_RETRIEVE:
1787 		return fuse_notify_retrieve(fc, size, cs);
1788 
1789 	case FUSE_NOTIFY_DELETE:
1790 		return fuse_notify_delete(fc, size, cs);
1791 
1792 	default:
1793 		fuse_copy_finish(cs);
1794 		return -EINVAL;
1795 	}
1796 }
1797 
1798 /* Look up request on processing list by unique ID */
1799 static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
1800 {
1801 	struct fuse_req *req;
1802 
1803 	list_for_each_entry(req, &fpq->processing, list) {
1804 		if (req->in.h.unique == unique || req->intr_unique == unique)
1805 			return req;
1806 	}
1807 	return NULL;
1808 }
1809 
1810 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1811 			 unsigned nbytes)
1812 {
1813 	unsigned reqsize = sizeof(struct fuse_out_header);
1814 
1815 	if (out->h.error)
1816 		return nbytes != reqsize ? -EINVAL : 0;
1817 
1818 	reqsize += len_args(out->numargs, out->args);
1819 
1820 	if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1821 		return -EINVAL;
1822 	else if (reqsize > nbytes) {
1823 		struct fuse_arg *lastarg = &out->args[out->numargs-1];
1824 		unsigned diffsize = reqsize - nbytes;
1825 		if (diffsize > lastarg->size)
1826 			return -EINVAL;
1827 		lastarg->size -= diffsize;
1828 	}
1829 	return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1830 			      out->page_zeroing);
1831 }
1832 
1833 /*
1834  * Write a single reply to a request.  First the header is copied from
1835  * the write buffer.  The request is then searched on the processing
1836  * list by the unique ID found in the header.  If found, then remove
1837  * it from the list and copy the rest of the buffer to the request.
1838  * The request is finished by calling request_end()
1839  */
1840 static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
1841 				 struct fuse_copy_state *cs, size_t nbytes)
1842 {
1843 	int err;
1844 	struct fuse_conn *fc = fud->fc;
1845 	struct fuse_pqueue *fpq = &fud->pq;
1846 	struct fuse_req *req;
1847 	struct fuse_out_header oh;
1848 
1849 	if (nbytes < sizeof(struct fuse_out_header))
1850 		return -EINVAL;
1851 
1852 	err = fuse_copy_one(cs, &oh, sizeof(oh));
1853 	if (err)
1854 		goto err_finish;
1855 
1856 	err = -EINVAL;
1857 	if (oh.len != nbytes)
1858 		goto err_finish;
1859 
1860 	/*
1861 	 * Zero oh.unique indicates unsolicited notification message
1862 	 * and error contains notification code.
1863 	 */
1864 	if (!oh.unique) {
1865 		err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
1866 		return err ? err : nbytes;
1867 	}
1868 
1869 	err = -EINVAL;
1870 	if (oh.error <= -1000 || oh.error > 0)
1871 		goto err_finish;
1872 
1873 	spin_lock(&fpq->lock);
1874 	err = -ENOENT;
1875 	if (!fpq->connected)
1876 		goto err_unlock_pq;
1877 
1878 	req = request_find(fpq, oh.unique);
1879 	if (!req)
1880 		goto err_unlock_pq;
1881 
1882 	/* Is it an interrupt reply? */
1883 	if (req->intr_unique == oh.unique) {
1884 		spin_unlock(&fpq->lock);
1885 
1886 		err = -EINVAL;
1887 		if (nbytes != sizeof(struct fuse_out_header))
1888 			goto err_finish;
1889 
1890 		if (oh.error == -ENOSYS)
1891 			fc->no_interrupt = 1;
1892 		else if (oh.error == -EAGAIN)
1893 			queue_interrupt(&fc->iq, req);
1894 
1895 		fuse_copy_finish(cs);
1896 		return nbytes;
1897 	}
1898 
1899 	clear_bit(FR_SENT, &req->flags);
1900 	list_move(&req->list, &fpq->io);
1901 	req->out.h = oh;
1902 	set_bit(FR_LOCKED, &req->flags);
1903 	spin_unlock(&fpq->lock);
1904 	cs->req = req;
1905 	if (!req->out.page_replace)
1906 		cs->move_pages = 0;
1907 
1908 	err = copy_out_args(cs, &req->out, nbytes);
1909 	fuse_copy_finish(cs);
1910 
1911 	spin_lock(&fpq->lock);
1912 	clear_bit(FR_LOCKED, &req->flags);
1913 	if (!fpq->connected)
1914 		err = -ENOENT;
1915 	else if (err)
1916 		req->out.h.error = -EIO;
1917 	if (!test_bit(FR_PRIVATE, &req->flags))
1918 		list_del_init(&req->list);
1919 	spin_unlock(&fpq->lock);
1920 
1921 	request_end(fc, req);
1922 
1923 	return err ? err : nbytes;
1924 
1925  err_unlock_pq:
1926 	spin_unlock(&fpq->lock);
1927  err_finish:
1928 	fuse_copy_finish(cs);
1929 	return err;
1930 }
1931 
1932 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
1933 {
1934 	struct fuse_copy_state cs;
1935 	struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1936 
1937 	if (!fud)
1938 		return -EPERM;
1939 
1940 	if (!iter_is_iovec(from))
1941 		return -EINVAL;
1942 
1943 	fuse_copy_init(&cs, 0, from);
1944 
1945 	return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
1946 }
1947 
1948 static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1949 				     struct file *out, loff_t *ppos,
1950 				     size_t len, unsigned int flags)
1951 {
1952 	unsigned nbuf;
1953 	unsigned idx;
1954 	struct pipe_buffer *bufs;
1955 	struct fuse_copy_state cs;
1956 	struct fuse_dev *fud;
1957 	size_t rem;
1958 	ssize_t ret;
1959 
1960 	fud = fuse_get_dev(out);
1961 	if (!fud)
1962 		return -EPERM;
1963 
1964 	bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
1965 	if (!bufs)
1966 		return -ENOMEM;
1967 
1968 	pipe_lock(pipe);
1969 	nbuf = 0;
1970 	rem = 0;
1971 	for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1972 		rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1973 
1974 	ret = -EINVAL;
1975 	if (rem < len) {
1976 		pipe_unlock(pipe);
1977 		goto out;
1978 	}
1979 
1980 	rem = len;
1981 	while (rem) {
1982 		struct pipe_buffer *ibuf;
1983 		struct pipe_buffer *obuf;
1984 
1985 		BUG_ON(nbuf >= pipe->buffers);
1986 		BUG_ON(!pipe->nrbufs);
1987 		ibuf = &pipe->bufs[pipe->curbuf];
1988 		obuf = &bufs[nbuf];
1989 
1990 		if (rem >= ibuf->len) {
1991 			*obuf = *ibuf;
1992 			ibuf->ops = NULL;
1993 			pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1994 			pipe->nrbufs--;
1995 		} else {
1996 			ibuf->ops->get(pipe, ibuf);
1997 			*obuf = *ibuf;
1998 			obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1999 			obuf->len = rem;
2000 			ibuf->offset += obuf->len;
2001 			ibuf->len -= obuf->len;
2002 		}
2003 		nbuf++;
2004 		rem -= obuf->len;
2005 	}
2006 	pipe_unlock(pipe);
2007 
2008 	fuse_copy_init(&cs, 0, NULL);
2009 	cs.pipebufs = bufs;
2010 	cs.nr_segs = nbuf;
2011 	cs.pipe = pipe;
2012 
2013 	if (flags & SPLICE_F_MOVE)
2014 		cs.move_pages = 1;
2015 
2016 	ret = fuse_dev_do_write(fud, &cs, len);
2017 
2018 	for (idx = 0; idx < nbuf; idx++) {
2019 		struct pipe_buffer *buf = &bufs[idx];
2020 		buf->ops->release(pipe, buf);
2021 	}
2022 out:
2023 	kfree(bufs);
2024 	return ret;
2025 }
2026 
2027 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2028 {
2029 	unsigned mask = POLLOUT | POLLWRNORM;
2030 	struct fuse_iqueue *fiq;
2031 	struct fuse_dev *fud = fuse_get_dev(file);
2032 
2033 	if (!fud)
2034 		return POLLERR;
2035 
2036 	fiq = &fud->fc->iq;
2037 	poll_wait(file, &fiq->waitq, wait);
2038 
2039 	spin_lock(&fiq->waitq.lock);
2040 	if (!fiq->connected)
2041 		mask = POLLERR;
2042 	else if (request_pending(fiq))
2043 		mask |= POLLIN | POLLRDNORM;
2044 	spin_unlock(&fiq->waitq.lock);
2045 
2046 	return mask;
2047 }
2048 
2049 /*
2050  * Abort all requests on the given list (pending or processing)
2051  *
2052  * This function releases and reacquires fc->lock
2053  */
2054 static void end_requests(struct fuse_conn *fc, struct list_head *head)
2055 {
2056 	while (!list_empty(head)) {
2057 		struct fuse_req *req;
2058 		req = list_entry(head->next, struct fuse_req, list);
2059 		req->out.h.error = -ECONNABORTED;
2060 		clear_bit(FR_PENDING, &req->flags);
2061 		clear_bit(FR_SENT, &req->flags);
2062 		list_del_init(&req->list);
2063 		request_end(fc, req);
2064 	}
2065 }
2066 
2067 static void end_polls(struct fuse_conn *fc)
2068 {
2069 	struct rb_node *p;
2070 
2071 	p = rb_first(&fc->polled_files);
2072 
2073 	while (p) {
2074 		struct fuse_file *ff;
2075 		ff = rb_entry(p, struct fuse_file, polled_node);
2076 		wake_up_interruptible_all(&ff->poll_wait);
2077 
2078 		p = rb_next(p);
2079 	}
2080 }
2081 
2082 /*
2083  * Abort all requests.
2084  *
2085  * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2086  * filesystem.
2087  *
2088  * The same effect is usually achievable through killing the filesystem daemon
2089  * and all users of the filesystem.  The exception is the combination of an
2090  * asynchronous request and the tricky deadlock (see
2091  * Documentation/filesystems/fuse.txt).
2092  *
2093  * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2094  * requests, they should be finished off immediately.  Locked requests will be
2095  * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2096  * requests.  It is possible that some request will finish before we can.  This
2097  * is OK, the request will in that case be removed from the list before we touch
2098  * it.
2099  */
2100 void fuse_abort_conn(struct fuse_conn *fc)
2101 {
2102 	struct fuse_iqueue *fiq = &fc->iq;
2103 
2104 	spin_lock(&fc->lock);
2105 	if (fc->connected) {
2106 		struct fuse_dev *fud;
2107 		struct fuse_req *req, *next;
2108 		LIST_HEAD(to_end1);
2109 		LIST_HEAD(to_end2);
2110 
2111 		fc->connected = 0;
2112 		fc->blocked = 0;
2113 		fuse_set_initialized(fc);
2114 		list_for_each_entry(fud, &fc->devices, entry) {
2115 			struct fuse_pqueue *fpq = &fud->pq;
2116 
2117 			spin_lock(&fpq->lock);
2118 			fpq->connected = 0;
2119 			list_for_each_entry_safe(req, next, &fpq->io, list) {
2120 				req->out.h.error = -ECONNABORTED;
2121 				spin_lock(&req->waitq.lock);
2122 				set_bit(FR_ABORTED, &req->flags);
2123 				if (!test_bit(FR_LOCKED, &req->flags)) {
2124 					set_bit(FR_PRIVATE, &req->flags);
2125 					list_move(&req->list, &to_end1);
2126 				}
2127 				spin_unlock(&req->waitq.lock);
2128 			}
2129 			list_splice_init(&fpq->processing, &to_end2);
2130 			spin_unlock(&fpq->lock);
2131 		}
2132 		fc->max_background = UINT_MAX;
2133 		flush_bg_queue(fc);
2134 
2135 		spin_lock(&fiq->waitq.lock);
2136 		fiq->connected = 0;
2137 		list_splice_init(&fiq->pending, &to_end2);
2138 		while (forget_pending(fiq))
2139 			kfree(dequeue_forget(fiq, 1, NULL));
2140 		wake_up_all_locked(&fiq->waitq);
2141 		spin_unlock(&fiq->waitq.lock);
2142 		kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
2143 		end_polls(fc);
2144 		wake_up_all(&fc->blocked_waitq);
2145 		spin_unlock(&fc->lock);
2146 
2147 		while (!list_empty(&to_end1)) {
2148 			req = list_first_entry(&to_end1, struct fuse_req, list);
2149 			__fuse_get_request(req);
2150 			list_del_init(&req->list);
2151 			request_end(fc, req);
2152 		}
2153 		end_requests(fc, &to_end2);
2154 	} else {
2155 		spin_unlock(&fc->lock);
2156 	}
2157 }
2158 EXPORT_SYMBOL_GPL(fuse_abort_conn);
2159 
2160 int fuse_dev_release(struct inode *inode, struct file *file)
2161 {
2162 	struct fuse_dev *fud = fuse_get_dev(file);
2163 
2164 	if (fud) {
2165 		struct fuse_conn *fc = fud->fc;
2166 		struct fuse_pqueue *fpq = &fud->pq;
2167 
2168 		WARN_ON(!list_empty(&fpq->io));
2169 		end_requests(fc, &fpq->processing);
2170 		/* Are we the last open device? */
2171 		if (atomic_dec_and_test(&fc->dev_count)) {
2172 			WARN_ON(fc->iq.fasync != NULL);
2173 			fuse_abort_conn(fc);
2174 		}
2175 		fuse_dev_free(fud);
2176 	}
2177 	return 0;
2178 }
2179 EXPORT_SYMBOL_GPL(fuse_dev_release);
2180 
2181 static int fuse_dev_fasync(int fd, struct file *file, int on)
2182 {
2183 	struct fuse_dev *fud = fuse_get_dev(file);
2184 
2185 	if (!fud)
2186 		return -EPERM;
2187 
2188 	/* No locking - fasync_helper does its own locking */
2189 	return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
2190 }
2191 
2192 static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2193 {
2194 	struct fuse_dev *fud;
2195 
2196 	if (new->private_data)
2197 		return -EINVAL;
2198 
2199 	fud = fuse_dev_alloc(fc);
2200 	if (!fud)
2201 		return -ENOMEM;
2202 
2203 	new->private_data = fud;
2204 	atomic_inc(&fc->dev_count);
2205 
2206 	return 0;
2207 }
2208 
2209 static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2210 			   unsigned long arg)
2211 {
2212 	int err = -ENOTTY;
2213 
2214 	if (cmd == FUSE_DEV_IOC_CLONE) {
2215 		int oldfd;
2216 
2217 		err = -EFAULT;
2218 		if (!get_user(oldfd, (__u32 __user *) arg)) {
2219 			struct file *old = fget(oldfd);
2220 
2221 			err = -EINVAL;
2222 			if (old) {
2223 				struct fuse_dev *fud = NULL;
2224 
2225 				/*
2226 				 * Check against file->f_op because CUSE
2227 				 * uses the same ioctl handler.
2228 				 */
2229 				if (old->f_op == file->f_op &&
2230 				    old->f_cred->user_ns == file->f_cred->user_ns)
2231 					fud = fuse_get_dev(old);
2232 
2233 				if (fud) {
2234 					mutex_lock(&fuse_mutex);
2235 					err = fuse_device_clone(fud->fc, file);
2236 					mutex_unlock(&fuse_mutex);
2237 				}
2238 				fput(old);
2239 			}
2240 		}
2241 	}
2242 	return err;
2243 }
2244 
2245 const struct file_operations fuse_dev_operations = {
2246 	.owner		= THIS_MODULE,
2247 	.open		= fuse_dev_open,
2248 	.llseek		= no_llseek,
2249 	.read_iter	= fuse_dev_read,
2250 	.splice_read	= fuse_dev_splice_read,
2251 	.write_iter	= fuse_dev_write,
2252 	.splice_write	= fuse_dev_splice_write,
2253 	.poll		= fuse_dev_poll,
2254 	.release	= fuse_dev_release,
2255 	.fasync		= fuse_dev_fasync,
2256 	.unlocked_ioctl = fuse_dev_ioctl,
2257 	.compat_ioctl   = fuse_dev_ioctl,
2258 };
2259 EXPORT_SYMBOL_GPL(fuse_dev_operations);
2260 
2261 static struct miscdevice fuse_miscdevice = {
2262 	.minor = FUSE_MINOR,
2263 	.name  = "fuse",
2264 	.fops = &fuse_dev_operations,
2265 };
2266 
2267 int __init fuse_dev_init(void)
2268 {
2269 	int err = -ENOMEM;
2270 	fuse_req_cachep = kmem_cache_create("fuse_request",
2271 					    sizeof(struct fuse_req),
2272 					    0, 0, NULL);
2273 	if (!fuse_req_cachep)
2274 		goto out;
2275 
2276 	err = misc_register(&fuse_miscdevice);
2277 	if (err)
2278 		goto out_cache_clean;
2279 
2280 	return 0;
2281 
2282  out_cache_clean:
2283 	kmem_cache_destroy(fuse_req_cachep);
2284  out:
2285 	return err;
2286 }
2287 
2288 void fuse_dev_cleanup(void)
2289 {
2290 	misc_deregister(&fuse_miscdevice);
2291 	kmem_cache_destroy(fuse_req_cachep);
2292 }
2293