]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/fuse/dev.c
fuse: introduce per-instance fuse_dev structure
[mirror_ubuntu-bionic-kernel.git] / fs / fuse / dev.c
CommitLineData
334f485d
MS
1/*
2 FUSE: Filesystem in Userspace
1729a16c 3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
334f485d
MS
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>
dd3bb14f 19#include <linux/pipe_fs_i.h>
ce534fb0
MS
20#include <linux/swap.h>
21#include <linux/splice.h>
334f485d
MS
22
23MODULE_ALIAS_MISCDEV(FUSE_MINOR);
578454ff 24MODULE_ALIAS("devname:fuse");
334f485d 25
e18b890b 26static struct kmem_cache *fuse_req_cachep;
334f485d 27
cc080e9e 28static struct fuse_dev *fuse_get_dev(struct file *file)
334f485d 29{
0720b315
MS
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 */
cc080e9e 34 return ACCESS_ONCE(file->private_data);
334f485d
MS
35}
36
4250c066 37static void fuse_request_init(struct fuse_req *req, struct page **pages,
b2430d75 38 struct fuse_page_desc *page_descs,
4250c066 39 unsigned npages)
334f485d
MS
40{
41 memset(req, 0, sizeof(*req));
4250c066 42 memset(pages, 0, sizeof(*pages) * npages);
b2430d75 43 memset(page_descs, 0, sizeof(*page_descs) * npages);
334f485d 44 INIT_LIST_HEAD(&req->list);
a4d27e75 45 INIT_LIST_HEAD(&req->intr_entry);
334f485d
MS
46 init_waitqueue_head(&req->waitq);
47 atomic_set(&req->count, 1);
4250c066 48 req->pages = pages;
b2430d75 49 req->page_descs = page_descs;
4250c066 50 req->max_pages = npages;
33e14b4d 51 __set_bit(FR_PENDING, &req->flags);
334f485d
MS
52}
53
4250c066 54static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
334f485d 55{
4250c066
MP
56 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
57 if (req) {
58 struct page **pages;
b2430d75 59 struct fuse_page_desc *page_descs;
4250c066 60
b2430d75 61 if (npages <= FUSE_REQ_INLINE_PAGES) {
4250c066 62 pages = req->inline_pages;
b2430d75
MP
63 page_descs = req->inline_page_descs;
64 } else {
4250c066 65 pages = kmalloc(sizeof(struct page *) * npages, flags);
b2430d75
MP
66 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
67 npages, flags);
68 }
4250c066 69
b2430d75
MP
70 if (!pages || !page_descs) {
71 kfree(pages);
72 kfree(page_descs);
4250c066
MP
73 kmem_cache_free(fuse_req_cachep, req);
74 return NULL;
75 }
76
b2430d75 77 fuse_request_init(req, pages, page_descs, npages);
4250c066 78 }
334f485d
MS
79 return req;
80}
4250c066
MP
81
82struct fuse_req *fuse_request_alloc(unsigned npages)
83{
84 return __fuse_request_alloc(npages, GFP_KERNEL);
85}
08cbf542 86EXPORT_SYMBOL_GPL(fuse_request_alloc);
334f485d 87
4250c066 88struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
3be5a52b 89{
4250c066 90 return __fuse_request_alloc(npages, GFP_NOFS);
3be5a52b
MS
91}
92
334f485d
MS
93void fuse_request_free(struct fuse_req *req)
94{
b2430d75 95 if (req->pages != req->inline_pages) {
4250c066 96 kfree(req->pages);
b2430d75
MP
97 kfree(req->page_descs);
98 }
334f485d
MS
99 kmem_cache_free(fuse_req_cachep, req);
100}
101
8bfc016d 102static void block_sigs(sigset_t *oldset)
334f485d
MS
103{
104 sigset_t mask;
105
106 siginitsetinv(&mask, sigmask(SIGKILL));
107 sigprocmask(SIG_BLOCK, &mask, oldset);
108}
109
8bfc016d 110static void restore_sigs(sigset_t *oldset)
334f485d
MS
111{
112 sigprocmask(SIG_SETMASK, oldset, NULL);
113}
114
36cf66ed 115void __fuse_get_request(struct fuse_req *req)
334f485d
MS
116{
117 atomic_inc(&req->count);
118}
119
120/* Must be called with > 1 refcount */
121static void __fuse_put_request(struct fuse_req *req)
122{
123 BUG_ON(atomic_read(&req->count) < 2);
124 atomic_dec(&req->count);
125}
126
33649c91
MS
127static void fuse_req_init_context(struct fuse_req *req)
128{
499dcf20
EB
129 req->in.h.uid = from_kuid_munged(&init_user_ns, current_fsuid());
130 req->in.h.gid = from_kgid_munged(&init_user_ns, current_fsgid());
33649c91
MS
131 req->in.h.pid = current->pid;
132}
133
9759bd51
MS
134void fuse_set_initialized(struct fuse_conn *fc)
135{
136 /* Make sure stores before this are seen on another CPU */
137 smp_wmb();
138 fc->initialized = 1;
139}
140
0aada884
MP
141static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
142{
143 return !fc->initialized || (for_background && fc->blocked);
144}
145
8b41e671
MP
146static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
147 bool for_background)
334f485d 148{
08a53cdc 149 struct fuse_req *req;
08a53cdc 150 int err;
9bc5ddda 151 atomic_inc(&fc->num_waiting);
0aada884
MP
152
153 if (fuse_block_alloc(fc, for_background)) {
154 sigset_t oldset;
155 int intr;
156
157 block_sigs(&oldset);
722d2bea 158 intr = wait_event_interruptible_exclusive(fc->blocked_waitq,
0aada884
MP
159 !fuse_block_alloc(fc, for_background));
160 restore_sigs(&oldset);
161 err = -EINTR;
162 if (intr)
163 goto out;
164 }
9759bd51
MS
165 /* Matches smp_wmb() in fuse_set_initialized() */
166 smp_rmb();
08a53cdc 167
51eb01e7
MS
168 err = -ENOTCONN;
169 if (!fc->connected)
170 goto out;
171
de155226
MS
172 err = -ECONNREFUSED;
173 if (fc->conn_error)
174 goto out;
175
b111c8c0 176 req = fuse_request_alloc(npages);
9bc5ddda 177 err = -ENOMEM;
722d2bea
MP
178 if (!req) {
179 if (for_background)
180 wake_up(&fc->blocked_waitq);
9bc5ddda 181 goto out;
722d2bea 182 }
334f485d 183
33649c91 184 fuse_req_init_context(req);
825d6d33
MS
185 __set_bit(FR_WAITING, &req->flags);
186 if (for_background)
187 __set_bit(FR_BACKGROUND, &req->flags);
188
334f485d 189 return req;
9bc5ddda
MS
190
191 out:
192 atomic_dec(&fc->num_waiting);
193 return ERR_PTR(err);
334f485d 194}
8b41e671
MP
195
196struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
197{
198 return __fuse_get_req(fc, npages, false);
199}
08cbf542 200EXPORT_SYMBOL_GPL(fuse_get_req);
334f485d 201
8b41e671
MP
202struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
203 unsigned npages)
204{
205 return __fuse_get_req(fc, npages, true);
206}
207EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
208
33649c91
MS
209/*
210 * Return request in fuse_file->reserved_req. However that may
211 * currently be in use. If that is the case, wait for it to become
212 * available.
213 */
214static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
215 struct file *file)
216{
217 struct fuse_req *req = NULL;
218 struct fuse_file *ff = file->private_data;
219
220 do {
de5e3dec 221 wait_event(fc->reserved_req_waitq, ff->reserved_req);
33649c91
MS
222 spin_lock(&fc->lock);
223 if (ff->reserved_req) {
224 req = ff->reserved_req;
225 ff->reserved_req = NULL;
cb0942b8 226 req->stolen_file = get_file(file);
33649c91
MS
227 }
228 spin_unlock(&fc->lock);
229 } while (!req);
230
231 return req;
232}
233
234/*
235 * Put stolen request back into fuse_file->reserved_req
236 */
237static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
238{
239 struct file *file = req->stolen_file;
240 struct fuse_file *ff = file->private_data;
241
242 spin_lock(&fc->lock);
b2430d75 243 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
33649c91
MS
244 BUG_ON(ff->reserved_req);
245 ff->reserved_req = req;
de5e3dec 246 wake_up_all(&fc->reserved_req_waitq);
33649c91
MS
247 spin_unlock(&fc->lock);
248 fput(file);
249}
250
251/*
252 * Gets a requests for a file operation, always succeeds
253 *
254 * This is used for sending the FLUSH request, which must get to
255 * userspace, due to POSIX locks which may need to be unlocked.
256 *
257 * If allocation fails due to OOM, use the reserved request in
258 * fuse_file.
259 *
260 * This is very unlikely to deadlock accidentally, since the
261 * filesystem should not have it's own file open. If deadlock is
262 * intentional, it can still be broken by "aborting" the filesystem.
263 */
b111c8c0
MP
264struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
265 struct file *file)
33649c91
MS
266{
267 struct fuse_req *req;
268
269 atomic_inc(&fc->num_waiting);
0aada884 270 wait_event(fc->blocked_waitq, fc->initialized);
9759bd51
MS
271 /* Matches smp_wmb() in fuse_set_initialized() */
272 smp_rmb();
b111c8c0 273 req = fuse_request_alloc(0);
33649c91
MS
274 if (!req)
275 req = get_reserved_req(fc, file);
276
277 fuse_req_init_context(req);
825d6d33
MS
278 __set_bit(FR_WAITING, &req->flags);
279 __clear_bit(FR_BACKGROUND, &req->flags);
33649c91
MS
280 return req;
281}
282
334f485d 283void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
7128ec2a
MS
284{
285 if (atomic_dec_and_test(&req->count)) {
825d6d33 286 if (test_bit(FR_BACKGROUND, &req->flags)) {
722d2bea
MP
287 /*
288 * We get here in the unlikely case that a background
289 * request was allocated but not sent
290 */
291 spin_lock(&fc->lock);
292 if (!fc->blocked)
293 wake_up(&fc->blocked_waitq);
294 spin_unlock(&fc->lock);
295 }
296
825d6d33
MS
297 if (test_bit(FR_WAITING, &req->flags)) {
298 __clear_bit(FR_WAITING, &req->flags);
9bc5ddda 299 atomic_dec(&fc->num_waiting);
73e0e738 300 }
33649c91
MS
301
302 if (req->stolen_file)
303 put_reserved_req(fc, req);
304 else
305 fuse_request_free(req);
7128ec2a
MS
306 }
307}
08cbf542 308EXPORT_SYMBOL_GPL(fuse_put_request);
7128ec2a 309
d12def1b
MS
310static unsigned len_args(unsigned numargs, struct fuse_arg *args)
311{
312 unsigned nbytes = 0;
313 unsigned i;
314
315 for (i = 0; i < numargs; i++)
316 nbytes += args[i].size;
317
318 return nbytes;
319}
320
f88996a9 321static u64 fuse_get_unique(struct fuse_iqueue *fiq)
d12def1b 322{
f88996a9 323 return ++fiq->reqctr;
d12def1b
MS
324}
325
f88996a9 326static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
d12def1b 327{
d12def1b
MS
328 req->in.h.len = sizeof(struct fuse_in_header) +
329 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
f88996a9 330 list_add_tail(&req->list, &fiq->pending);
4ce60812 331 wake_up_locked(&fiq->waitq);
f88996a9 332 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
d12def1b
MS
333}
334
07e77dca
MS
335void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
336 u64 nodeid, u64 nlookup)
337{
f88996a9
MS
338 struct fuse_iqueue *fiq = &fc->iq;
339
02c048b9
MS
340 forget->forget_one.nodeid = nodeid;
341 forget->forget_one.nlookup = nlookup;
07e77dca 342
4ce60812 343 spin_lock(&fiq->waitq.lock);
e16714d8 344 if (fiq->connected) {
f88996a9
MS
345 fiq->forget_list_tail->next = forget;
346 fiq->forget_list_tail = forget;
4ce60812 347 wake_up_locked(&fiq->waitq);
f88996a9 348 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
5dfcc87f
MS
349 } else {
350 kfree(forget);
351 }
4ce60812 352 spin_unlock(&fiq->waitq.lock);
07e77dca
MS
353}
354
d12def1b
MS
355static void flush_bg_queue(struct fuse_conn *fc)
356{
7a6d3c8b 357 while (fc->active_background < fc->max_background &&
d12def1b
MS
358 !list_empty(&fc->bg_queue)) {
359 struct fuse_req *req;
f88996a9 360 struct fuse_iqueue *fiq = &fc->iq;
d12def1b
MS
361
362 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
363 list_del(&req->list);
364 fc->active_background++;
4ce60812 365 spin_lock(&fiq->waitq.lock);
f88996a9
MS
366 req->in.h.unique = fuse_get_unique(fiq);
367 queue_request(fiq, req);
4ce60812 368 spin_unlock(&fiq->waitq.lock);
d12def1b
MS
369 }
370}
371
334f485d
MS
372/*
373 * This function is called when a request is finished. Either a reply
f9a2842e 374 * has arrived or it was aborted (and not yet sent) or some error
f43b155a 375 * occurred during communication with userspace, or the device file
51eb01e7
MS
376 * was closed. The requester thread is woken up (if still waiting),
377 * the 'end' callback is called if given, else the reference to the
378 * request is released
334f485d
MS
379 */
380static void request_end(struct fuse_conn *fc, struct fuse_req *req)
381{
4ce60812 382 struct fuse_iqueue *fiq = &fc->iq;
365ae710 383
efe2800f 384 if (test_and_set_bit(FR_FINISHED, &req->flags))
365ae710 385 return;
365ae710 386
4ce60812 387 spin_lock(&fiq->waitq.lock);
0d8e84b0 388 list_del_init(&req->intr_entry);
4ce60812 389 spin_unlock(&fiq->waitq.lock);
33e14b4d
MS
390 WARN_ON(test_bit(FR_PENDING, &req->flags));
391 WARN_ON(test_bit(FR_SENT, &req->flags));
825d6d33 392 if (test_bit(FR_BACKGROUND, &req->flags)) {
efe2800f 393 spin_lock(&fc->lock);
825d6d33 394 clear_bit(FR_BACKGROUND, &req->flags);
722d2bea 395 if (fc->num_background == fc->max_background)
51eb01e7 396 fc->blocked = 0;
722d2bea
MP
397
398 /* Wake up next waiter, if any */
3c18ef81 399 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
722d2bea
MP
400 wake_up(&fc->blocked_waitq);
401
7a6d3c8b 402 if (fc->num_background == fc->congestion_threshold &&
a325f9b9 403 fc->connected && fc->bdi_initialized) {
8aa7e847
JA
404 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
405 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
f92b99b9 406 }
51eb01e7 407 fc->num_background--;
d12def1b
MS
408 fc->active_background--;
409 flush_bg_queue(fc);
efe2800f 410 spin_unlock(&fc->lock);
334f485d 411 }
51eb01e7 412 wake_up(&req->waitq);
1e6881c3
MS
413 if (req->end)
414 req->end(fc, req);
e9bb09dd 415 fuse_put_request(fc, req);
334f485d
MS
416}
417
f88996a9 418static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
a4d27e75 419{
4ce60812 420 spin_lock(&fiq->waitq.lock);
8f7bb368
MS
421 if (list_empty(&req->intr_entry)) {
422 list_add_tail(&req->intr_entry, &fiq->interrupts);
423 wake_up_locked(&fiq->waitq);
424 }
4ce60812 425 spin_unlock(&fiq->waitq.lock);
f88996a9 426 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
a4d27e75
MS
427}
428
7c352bdf 429static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
334f485d 430{
4ce60812 431 struct fuse_iqueue *fiq = &fc->iq;
c4775267
MS
432 int err;
433
a4d27e75
MS
434 if (!fc->no_interrupt) {
435 /* Any signal may interrupt this */
c4775267 436 err = wait_event_interruptible(req->waitq,
33e14b4d 437 test_bit(FR_FINISHED, &req->flags));
c4775267 438 if (!err)
a4d27e75
MS
439 return;
440
825d6d33 441 set_bit(FR_INTERRUPTED, &req->flags);
8f7bb368
MS
442 /* matches barrier in fuse_dev_do_read() */
443 smp_mb__after_atomic();
33e14b4d 444 if (test_bit(FR_SENT, &req->flags))
4ce60812 445 queue_interrupt(fiq, req);
a4d27e75
MS
446 }
447
825d6d33 448 if (!test_bit(FR_FORCE, &req->flags)) {
a4d27e75
MS
449 sigset_t oldset;
450
451 /* Only fatal signals may interrupt this */
51eb01e7 452 block_sigs(&oldset);
c4775267 453 err = wait_event_interruptible(req->waitq,
33e14b4d 454 test_bit(FR_FINISHED, &req->flags));
51eb01e7 455 restore_sigs(&oldset);
a131de0a 456
c4775267 457 if (!err)
a131de0a
MS
458 return;
459
4ce60812 460 spin_lock(&fiq->waitq.lock);
a131de0a 461 /* Request is not yet in userspace, bail out */
33e14b4d 462 if (test_bit(FR_PENDING, &req->flags)) {
a131de0a 463 list_del(&req->list);
4ce60812 464 spin_unlock(&fiq->waitq.lock);
a131de0a
MS
465 __fuse_put_request(req);
466 req->out.h.error = -EINTR;
467 return;
468 }
4ce60812 469 spin_unlock(&fiq->waitq.lock);
51eb01e7 470 }
334f485d 471
a131de0a
MS
472 /*
473 * Either request is already in userspace, or it was forced.
474 * Wait it out.
475 */
33e14b4d 476 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
334f485d
MS
477}
478
6a4e922c 479static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
334f485d 480{
e16714d8
MS
481 struct fuse_iqueue *fiq = &fc->iq;
482
825d6d33 483 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
4ce60812 484 spin_lock(&fiq->waitq.lock);
e16714d8 485 if (!fiq->connected) {
4ce60812 486 spin_unlock(&fiq->waitq.lock);
334f485d 487 req->out.h.error = -ENOTCONN;
c4775267 488 } else {
f88996a9
MS
489 req->in.h.unique = fuse_get_unique(fiq);
490 queue_request(fiq, req);
334f485d
MS
491 /* acquire extra reference, since request is still needed
492 after request_end() */
493 __fuse_get_request(req);
4ce60812 494 spin_unlock(&fiq->waitq.lock);
334f485d 495
7c352bdf 496 request_wait_answer(fc, req);
c4775267
MS
497 /* Pairs with smp_wmb() in request_end() */
498 smp_rmb();
334f485d 499 }
334f485d 500}
6a4e922c
EW
501
502void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
503{
825d6d33
MS
504 __set_bit(FR_ISREPLY, &req->flags);
505 if (!test_bit(FR_WAITING, &req->flags)) {
506 __set_bit(FR_WAITING, &req->flags);
5437f241
MS
507 atomic_inc(&fc->num_waiting);
508 }
6a4e922c
EW
509 __fuse_request_send(fc, req);
510}
08cbf542 511EXPORT_SYMBOL_GPL(fuse_request_send);
334f485d 512
21f62174
MS
513static void fuse_adjust_compat(struct fuse_conn *fc, struct fuse_args *args)
514{
515 if (fc->minor < 4 && args->in.h.opcode == FUSE_STATFS)
516 args->out.args[0].size = FUSE_COMPAT_STATFS_SIZE;
517
518 if (fc->minor < 9) {
519 switch (args->in.h.opcode) {
520 case FUSE_LOOKUP:
521 case FUSE_CREATE:
522 case FUSE_MKNOD:
523 case FUSE_MKDIR:
524 case FUSE_SYMLINK:
525 case FUSE_LINK:
526 args->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
527 break;
528 case FUSE_GETATTR:
529 case FUSE_SETATTR:
530 args->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
531 break;
532 }
533 }
534 if (fc->minor < 12) {
535 switch (args->in.h.opcode) {
536 case FUSE_CREATE:
537 args->in.args[0].size = sizeof(struct fuse_open_in);
538 break;
539 case FUSE_MKNOD:
540 args->in.args[0].size = FUSE_COMPAT_MKNOD_IN_SIZE;
541 break;
542 }
543 }
544}
545
7078187a
MS
546ssize_t fuse_simple_request(struct fuse_conn *fc, struct fuse_args *args)
547{
548 struct fuse_req *req;
549 ssize_t ret;
550
551 req = fuse_get_req(fc, 0);
552 if (IS_ERR(req))
553 return PTR_ERR(req);
554
21f62174
MS
555 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
556 fuse_adjust_compat(fc, args);
557
7078187a
MS
558 req->in.h.opcode = args->in.h.opcode;
559 req->in.h.nodeid = args->in.h.nodeid;
560 req->in.numargs = args->in.numargs;
561 memcpy(req->in.args, args->in.args,
562 args->in.numargs * sizeof(struct fuse_in_arg));
563 req->out.argvar = args->out.argvar;
564 req->out.numargs = args->out.numargs;
565 memcpy(req->out.args, args->out.args,
566 args->out.numargs * sizeof(struct fuse_arg));
567 fuse_request_send(fc, req);
568 ret = req->out.h.error;
569 if (!ret && args->out.argvar) {
570 BUG_ON(args->out.numargs != 1);
571 ret = req->out.args[0].size;
572 }
573 fuse_put_request(fc, req);
574
575 return ret;
576}
577
f0139aa8
MS
578/*
579 * Called under fc->lock
580 *
581 * fc->connected must have been checked previously
582 */
583void fuse_request_send_background_locked(struct fuse_conn *fc,
584 struct fuse_req *req)
d12def1b 585{
825d6d33
MS
586 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
587 if (!test_bit(FR_WAITING, &req->flags)) {
588 __set_bit(FR_WAITING, &req->flags);
5437f241
MS
589 atomic_inc(&fc->num_waiting);
590 }
825d6d33 591 __set_bit(FR_ISREPLY, &req->flags);
d12def1b 592 fc->num_background++;
7a6d3c8b 593 if (fc->num_background == fc->max_background)
d12def1b 594 fc->blocked = 1;
7a6d3c8b 595 if (fc->num_background == fc->congestion_threshold &&
a325f9b9 596 fc->bdi_initialized) {
8aa7e847
JA
597 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
598 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
d12def1b
MS
599 }
600 list_add_tail(&req->list, &fc->bg_queue);
601 flush_bg_queue(fc);
602}
603
f0139aa8 604void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
334f485d 605{
42dc6211 606 BUG_ON(!req->end);
d7133114 607 spin_lock(&fc->lock);
1e9a4ed9 608 if (fc->connected) {
f0139aa8 609 fuse_request_send_background_locked(fc, req);
d7133114 610 spin_unlock(&fc->lock);
334f485d 611 } else {
42dc6211 612 spin_unlock(&fc->lock);
334f485d 613 req->out.h.error = -ENOTCONN;
42dc6211
MS
614 req->end(fc, req);
615 fuse_put_request(fc, req);
334f485d
MS
616 }
617}
08cbf542 618EXPORT_SYMBOL_GPL(fuse_request_send_background);
334f485d 619
2d45ba38
MS
620static int fuse_request_send_notify_reply(struct fuse_conn *fc,
621 struct fuse_req *req, u64 unique)
622{
623 int err = -ENODEV;
f88996a9 624 struct fuse_iqueue *fiq = &fc->iq;
2d45ba38 625
825d6d33 626 __clear_bit(FR_ISREPLY, &req->flags);
2d45ba38 627 req->in.h.unique = unique;
4ce60812 628 spin_lock(&fiq->waitq.lock);
e16714d8 629 if (fiq->connected) {
f88996a9 630 queue_request(fiq, req);
2d45ba38
MS
631 err = 0;
632 }
4ce60812 633 spin_unlock(&fiq->waitq.lock);
2d45ba38
MS
634
635 return err;
636}
637
0b05b183
AA
638void fuse_force_forget(struct file *file, u64 nodeid)
639{
6131ffaa 640 struct inode *inode = file_inode(file);
0b05b183
AA
641 struct fuse_conn *fc = get_fuse_conn(inode);
642 struct fuse_req *req;
643 struct fuse_forget_in inarg;
644
645 memset(&inarg, 0, sizeof(inarg));
646 inarg.nlookup = 1;
b111c8c0 647 req = fuse_get_req_nofail_nopages(fc, file);
0b05b183
AA
648 req->in.h.opcode = FUSE_FORGET;
649 req->in.h.nodeid = nodeid;
650 req->in.numargs = 1;
651 req->in.args[0].size = sizeof(inarg);
652 req->in.args[0].value = &inarg;
825d6d33 653 __clear_bit(FR_ISREPLY, &req->flags);
6a4e922c
EW
654 __fuse_request_send(fc, req);
655 /* ignore errors */
656 fuse_put_request(fc, req);
0b05b183
AA
657}
658
334f485d
MS
659/*
660 * Lock the request. Up to the next unlock_request() there mustn't be
661 * anything that could cause a page-fault. If the request was already
f9a2842e 662 * aborted bail out.
334f485d 663 */
dc00809a 664static int lock_request(struct fuse_req *req)
334f485d
MS
665{
666 int err = 0;
667 if (req) {
dc00809a 668 spin_lock(&req->waitq.lock);
825d6d33 669 if (test_bit(FR_ABORTED, &req->flags))
334f485d
MS
670 err = -ENOENT;
671 else
825d6d33 672 set_bit(FR_LOCKED, &req->flags);
dc00809a 673 spin_unlock(&req->waitq.lock);
334f485d
MS
674 }
675 return err;
676}
677
678/*
0d8e84b0
MS
679 * Unlock request. If it was aborted while locked, caller is responsible
680 * for unlocking and ending the request.
334f485d 681 */
dc00809a 682static int unlock_request(struct fuse_req *req)
334f485d 683{
0d8e84b0 684 int err = 0;
334f485d 685 if (req) {
dc00809a 686 spin_lock(&req->waitq.lock);
825d6d33 687 if (test_bit(FR_ABORTED, &req->flags))
0d8e84b0
MS
688 err = -ENOENT;
689 else
825d6d33 690 clear_bit(FR_LOCKED, &req->flags);
dc00809a 691 spin_unlock(&req->waitq.lock);
334f485d 692 }
0d8e84b0 693 return err;
334f485d
MS
694}
695
696struct fuse_copy_state {
697 int write;
698 struct fuse_req *req;
6c09e94a 699 struct iov_iter *iter;
dd3bb14f
MS
700 struct pipe_buffer *pipebufs;
701 struct pipe_buffer *currbuf;
702 struct pipe_inode_info *pipe;
334f485d 703 unsigned long nr_segs;
334f485d 704 struct page *pg;
334f485d 705 unsigned len;
c55a01d3 706 unsigned offset;
ce534fb0 707 unsigned move_pages:1;
334f485d
MS
708};
709
dc00809a 710static void fuse_copy_init(struct fuse_copy_state *cs, int write,
6c09e94a 711 struct iov_iter *iter)
334f485d
MS
712{
713 memset(cs, 0, sizeof(*cs));
714 cs->write = write;
6c09e94a 715 cs->iter = iter;
334f485d
MS
716}
717
718/* Unmap and put previous page of userspace buffer */
8bfc016d 719static void fuse_copy_finish(struct fuse_copy_state *cs)
334f485d 720{
dd3bb14f
MS
721 if (cs->currbuf) {
722 struct pipe_buffer *buf = cs->currbuf;
723
c55a01d3 724 if (cs->write)
c3021629 725 buf->len = PAGE_SIZE - cs->len;
dd3bb14f 726 cs->currbuf = NULL;
c55a01d3 727 } else if (cs->pg) {
334f485d
MS
728 if (cs->write) {
729 flush_dcache_page(cs->pg);
730 set_page_dirty_lock(cs->pg);
731 }
732 put_page(cs->pg);
334f485d 733 }
c55a01d3 734 cs->pg = NULL;
334f485d
MS
735}
736
737/*
738 * Get another pagefull of userspace buffer, and map it to kernel
739 * address space, and lock request
740 */
741static int fuse_copy_fill(struct fuse_copy_state *cs)
742{
c55a01d3 743 struct page *page;
334f485d
MS
744 int err;
745
dc00809a 746 err = unlock_request(cs->req);
0d8e84b0
MS
747 if (err)
748 return err;
749
334f485d 750 fuse_copy_finish(cs);
dd3bb14f
MS
751 if (cs->pipebufs) {
752 struct pipe_buffer *buf = cs->pipebufs;
753
c3021629
MS
754 if (!cs->write) {
755 err = buf->ops->confirm(cs->pipe, buf);
756 if (err)
757 return err;
758
759 BUG_ON(!cs->nr_segs);
760 cs->currbuf = buf;
c55a01d3
MS
761 cs->pg = buf->page;
762 cs->offset = buf->offset;
c3021629 763 cs->len = buf->len;
c3021629
MS
764 cs->pipebufs++;
765 cs->nr_segs--;
766 } else {
c3021629
MS
767 if (cs->nr_segs == cs->pipe->buffers)
768 return -EIO;
769
770 page = alloc_page(GFP_HIGHUSER);
771 if (!page)
772 return -ENOMEM;
773
774 buf->page = page;
775 buf->offset = 0;
776 buf->len = 0;
777
778 cs->currbuf = buf;
c55a01d3
MS
779 cs->pg = page;
780 cs->offset = 0;
c3021629
MS
781 cs->len = PAGE_SIZE;
782 cs->pipebufs++;
783 cs->nr_segs++;
784 }
dd3bb14f 785 } else {
6c09e94a
AV
786 size_t off;
787 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
dd3bb14f
MS
788 if (err < 0)
789 return err;
6c09e94a
AV
790 BUG_ON(!err);
791 cs->len = err;
792 cs->offset = off;
c55a01d3 793 cs->pg = page;
6c09e94a
AV
794 cs->offset = off;
795 iov_iter_advance(cs->iter, err);
334f485d 796 }
334f485d 797
dc00809a 798 return lock_request(cs->req);
334f485d
MS
799}
800
801/* Do as much copy to/from userspace buffer as we can */
8bfc016d 802static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
334f485d
MS
803{
804 unsigned ncpy = min(*size, cs->len);
805 if (val) {
c55a01d3
MS
806 void *pgaddr = kmap_atomic(cs->pg);
807 void *buf = pgaddr + cs->offset;
808
334f485d 809 if (cs->write)
c55a01d3 810 memcpy(buf, *val, ncpy);
334f485d 811 else
c55a01d3
MS
812 memcpy(*val, buf, ncpy);
813
814 kunmap_atomic(pgaddr);
334f485d
MS
815 *val += ncpy;
816 }
817 *size -= ncpy;
818 cs->len -= ncpy;
c55a01d3 819 cs->offset += ncpy;
334f485d
MS
820 return ncpy;
821}
822
ce534fb0
MS
823static int fuse_check_page(struct page *page)
824{
825 if (page_mapcount(page) ||
826 page->mapping != NULL ||
827 page_count(page) != 1 ||
828 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
829 ~(1 << PG_locked |
830 1 << PG_referenced |
831 1 << PG_uptodate |
832 1 << PG_lru |
833 1 << PG_active |
834 1 << PG_reclaim))) {
835 printk(KERN_WARNING "fuse: trying to steal weird page\n");
836 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);
837 return 1;
838 }
839 return 0;
840}
841
842static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
843{
844 int err;
845 struct page *oldpage = *pagep;
846 struct page *newpage;
847 struct pipe_buffer *buf = cs->pipebufs;
ce534fb0 848
dc00809a 849 err = unlock_request(cs->req);
0d8e84b0
MS
850 if (err)
851 return err;
852
ce534fb0
MS
853 fuse_copy_finish(cs);
854
855 err = buf->ops->confirm(cs->pipe, buf);
856 if (err)
857 return err;
858
859 BUG_ON(!cs->nr_segs);
860 cs->currbuf = buf;
861 cs->len = buf->len;
862 cs->pipebufs++;
863 cs->nr_segs--;
864
865 if (cs->len != PAGE_SIZE)
866 goto out_fallback;
867
868 if (buf->ops->steal(cs->pipe, buf) != 0)
869 goto out_fallback;
870
871 newpage = buf->page;
872
aa991b3b
MS
873 if (!PageUptodate(newpage))
874 SetPageUptodate(newpage);
ce534fb0
MS
875
876 ClearPageMappedToDisk(newpage);
877
878 if (fuse_check_page(newpage) != 0)
879 goto out_fallback_unlock;
880
ce534fb0
MS
881 /*
882 * This is a new and locked page, it shouldn't be mapped or
883 * have any special flags on it
884 */
885 if (WARN_ON(page_mapped(oldpage)))
886 goto out_fallback_unlock;
887 if (WARN_ON(page_has_private(oldpage)))
888 goto out_fallback_unlock;
889 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
890 goto out_fallback_unlock;
891 if (WARN_ON(PageMlocked(oldpage)))
892 goto out_fallback_unlock;
893
ef6a3c63 894 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
ce534fb0 895 if (err) {
ef6a3c63
MS
896 unlock_page(newpage);
897 return err;
ce534fb0 898 }
ef6a3c63 899
ce534fb0
MS
900 page_cache_get(newpage);
901
902 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
903 lru_cache_add_file(newpage);
904
905 err = 0;
dc00809a 906 spin_lock(&cs->req->waitq.lock);
825d6d33 907 if (test_bit(FR_ABORTED, &cs->req->flags))
ce534fb0
MS
908 err = -ENOENT;
909 else
910 *pagep = newpage;
dc00809a 911 spin_unlock(&cs->req->waitq.lock);
ce534fb0
MS
912
913 if (err) {
914 unlock_page(newpage);
915 page_cache_release(newpage);
916 return err;
917 }
918
919 unlock_page(oldpage);
920 page_cache_release(oldpage);
921 cs->len = 0;
922
923 return 0;
924
925out_fallback_unlock:
926 unlock_page(newpage);
927out_fallback:
c55a01d3
MS
928 cs->pg = buf->page;
929 cs->offset = buf->offset;
ce534fb0 930
dc00809a 931 err = lock_request(cs->req);
ce534fb0
MS
932 if (err)
933 return err;
934
935 return 1;
936}
937
c3021629
MS
938static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
939 unsigned offset, unsigned count)
940{
941 struct pipe_buffer *buf;
0d8e84b0 942 int err;
c3021629
MS
943
944 if (cs->nr_segs == cs->pipe->buffers)
945 return -EIO;
946
dc00809a 947 err = unlock_request(cs->req);
0d8e84b0
MS
948 if (err)
949 return err;
950
c3021629
MS
951 fuse_copy_finish(cs);
952
953 buf = cs->pipebufs;
954 page_cache_get(page);
955 buf->page = page;
956 buf->offset = offset;
957 buf->len = count;
958
959 cs->pipebufs++;
960 cs->nr_segs++;
961 cs->len = 0;
962
963 return 0;
964}
965
334f485d
MS
966/*
967 * Copy a page in the request to/from the userspace buffer. Must be
968 * done atomically
969 */
ce534fb0 970static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
8bfc016d 971 unsigned offset, unsigned count, int zeroing)
334f485d 972{
ce534fb0
MS
973 int err;
974 struct page *page = *pagep;
975
b6777c40
MS
976 if (page && zeroing && count < PAGE_SIZE)
977 clear_highpage(page);
978
334f485d 979 while (count) {
c3021629
MS
980 if (cs->write && cs->pipebufs && page) {
981 return fuse_ref_page(cs, page, offset, count);
982 } else if (!cs->len) {
ce534fb0
MS
983 if (cs->move_pages && page &&
984 offset == 0 && count == PAGE_SIZE) {
985 err = fuse_try_move_page(cs, pagep);
986 if (err <= 0)
987 return err;
988 } else {
989 err = fuse_copy_fill(cs);
990 if (err)
991 return err;
992 }
1729a16c 993 }
334f485d 994 if (page) {
2408f6ef 995 void *mapaddr = kmap_atomic(page);
334f485d
MS
996 void *buf = mapaddr + offset;
997 offset += fuse_copy_do(cs, &buf, &count);
2408f6ef 998 kunmap_atomic(mapaddr);
334f485d
MS
999 } else
1000 offset += fuse_copy_do(cs, NULL, &count);
1001 }
1002 if (page && !cs->write)
1003 flush_dcache_page(page);
1004 return 0;
1005}
1006
1007/* Copy pages in the request to/from userspace buffer */
1008static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
1009 int zeroing)
1010{
1011 unsigned i;
1012 struct fuse_req *req = cs->req;
334f485d
MS
1013
1014 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
ce534fb0 1015 int err;
85f40aec
MP
1016 unsigned offset = req->page_descs[i].offset;
1017 unsigned count = min(nbytes, req->page_descs[i].length);
ce534fb0
MS
1018
1019 err = fuse_copy_page(cs, &req->pages[i], offset, count,
1020 zeroing);
334f485d
MS
1021 if (err)
1022 return err;
1023
1024 nbytes -= count;
334f485d
MS
1025 }
1026 return 0;
1027}
1028
1029/* Copy a single argument in the request to/from userspace buffer */
1030static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1031{
1032 while (size) {
1729a16c
MS
1033 if (!cs->len) {
1034 int err = fuse_copy_fill(cs);
1035 if (err)
1036 return err;
1037 }
334f485d
MS
1038 fuse_copy_do(cs, &val, &size);
1039 }
1040 return 0;
1041}
1042
1043/* Copy request arguments to/from userspace buffer */
1044static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
1045 unsigned argpages, struct fuse_arg *args,
1046 int zeroing)
1047{
1048 int err = 0;
1049 unsigned i;
1050
1051 for (i = 0; !err && i < numargs; i++) {
1052 struct fuse_arg *arg = &args[i];
1053 if (i == numargs - 1 && argpages)
1054 err = fuse_copy_pages(cs, arg->size, zeroing);
1055 else
1056 err = fuse_copy_one(cs, arg->value, arg->size);
1057 }
1058 return err;
1059}
1060
f88996a9 1061static int forget_pending(struct fuse_iqueue *fiq)
07e77dca 1062{
f88996a9 1063 return fiq->forget_list_head.next != NULL;
07e77dca
MS
1064}
1065
f88996a9 1066static int request_pending(struct fuse_iqueue *fiq)
a4d27e75 1067{
f88996a9
MS
1068 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1069 forget_pending(fiq);
a4d27e75
MS
1070}
1071
a4d27e75
MS
1072/*
1073 * Transfer an interrupt request to userspace
1074 *
1075 * Unlike other requests this is assembled on demand, without a need
1076 * to allocate a separate fuse_req structure.
1077 *
fd22d62e 1078 * Called with fiq->waitq.lock held, releases it
a4d27e75 1079 */
fd22d62e
MS
1080static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1081 struct fuse_copy_state *cs,
c3021629 1082 size_t nbytes, struct fuse_req *req)
fd22d62e 1083__releases(fiq->waitq.lock)
a4d27e75 1084{
a4d27e75
MS
1085 struct fuse_in_header ih;
1086 struct fuse_interrupt_in arg;
1087 unsigned reqsize = sizeof(ih) + sizeof(arg);
1088 int err;
1089
1090 list_del_init(&req->intr_entry);
4ce60812 1091 req->intr_unique = fuse_get_unique(fiq);
a4d27e75
MS
1092 memset(&ih, 0, sizeof(ih));
1093 memset(&arg, 0, sizeof(arg));
1094 ih.len = reqsize;
1095 ih.opcode = FUSE_INTERRUPT;
1096 ih.unique = req->intr_unique;
1097 arg.unique = req->in.h.unique;
1098
4ce60812 1099 spin_unlock(&fiq->waitq.lock);
c3021629 1100 if (nbytes < reqsize)
a4d27e75
MS
1101 return -EINVAL;
1102
c3021629 1103 err = fuse_copy_one(cs, &ih, sizeof(ih));
a4d27e75 1104 if (!err)
c3021629
MS
1105 err = fuse_copy_one(cs, &arg, sizeof(arg));
1106 fuse_copy_finish(cs);
a4d27e75
MS
1107
1108 return err ? err : reqsize;
1109}
1110
f88996a9 1111static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
02c048b9
MS
1112 unsigned max,
1113 unsigned *countp)
07e77dca 1114{
f88996a9 1115 struct fuse_forget_link *head = fiq->forget_list_head.next;
02c048b9
MS
1116 struct fuse_forget_link **newhead = &head;
1117 unsigned count;
07e77dca 1118
02c048b9
MS
1119 for (count = 0; *newhead != NULL && count < max; count++)
1120 newhead = &(*newhead)->next;
1121
f88996a9 1122 fiq->forget_list_head.next = *newhead;
02c048b9 1123 *newhead = NULL;
f88996a9
MS
1124 if (fiq->forget_list_head.next == NULL)
1125 fiq->forget_list_tail = &fiq->forget_list_head;
07e77dca 1126
02c048b9
MS
1127 if (countp != NULL)
1128 *countp = count;
1129
1130 return head;
07e77dca
MS
1131}
1132
fd22d62e 1133static int fuse_read_single_forget(struct fuse_iqueue *fiq,
07e77dca
MS
1134 struct fuse_copy_state *cs,
1135 size_t nbytes)
fd22d62e 1136__releases(fiq->waitq.lock)
07e77dca
MS
1137{
1138 int err;
f88996a9 1139 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
07e77dca 1140 struct fuse_forget_in arg = {
02c048b9 1141 .nlookup = forget->forget_one.nlookup,
07e77dca
MS
1142 };
1143 struct fuse_in_header ih = {
1144 .opcode = FUSE_FORGET,
02c048b9 1145 .nodeid = forget->forget_one.nodeid,
f88996a9 1146 .unique = fuse_get_unique(fiq),
07e77dca
MS
1147 .len = sizeof(ih) + sizeof(arg),
1148 };
1149
4ce60812 1150 spin_unlock(&fiq->waitq.lock);
07e77dca
MS
1151 kfree(forget);
1152 if (nbytes < ih.len)
1153 return -EINVAL;
1154
1155 err = fuse_copy_one(cs, &ih, sizeof(ih));
1156 if (!err)
1157 err = fuse_copy_one(cs, &arg, sizeof(arg));
1158 fuse_copy_finish(cs);
1159
1160 if (err)
1161 return err;
1162
1163 return ih.len;
1164}
1165
fd22d62e 1166static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
02c048b9 1167 struct fuse_copy_state *cs, size_t nbytes)
fd22d62e 1168__releases(fiq->waitq.lock)
02c048b9
MS
1169{
1170 int err;
1171 unsigned max_forgets;
1172 unsigned count;
1173 struct fuse_forget_link *head;
1174 struct fuse_batch_forget_in arg = { .count = 0 };
1175 struct fuse_in_header ih = {
1176 .opcode = FUSE_BATCH_FORGET,
f88996a9 1177 .unique = fuse_get_unique(fiq),
02c048b9
MS
1178 .len = sizeof(ih) + sizeof(arg),
1179 };
1180
1181 if (nbytes < ih.len) {
4ce60812 1182 spin_unlock(&fiq->waitq.lock);
02c048b9
MS
1183 return -EINVAL;
1184 }
1185
1186 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
f88996a9 1187 head = dequeue_forget(fiq, max_forgets, &count);
4ce60812 1188 spin_unlock(&fiq->waitq.lock);
02c048b9
MS
1189
1190 arg.count = count;
1191 ih.len += count * sizeof(struct fuse_forget_one);
1192 err = fuse_copy_one(cs, &ih, sizeof(ih));
1193 if (!err)
1194 err = fuse_copy_one(cs, &arg, sizeof(arg));
1195
1196 while (head) {
1197 struct fuse_forget_link *forget = head;
1198
1199 if (!err) {
1200 err = fuse_copy_one(cs, &forget->forget_one,
1201 sizeof(forget->forget_one));
1202 }
1203 head = forget->next;
1204 kfree(forget);
1205 }
1206
1207 fuse_copy_finish(cs);
1208
1209 if (err)
1210 return err;
1211
1212 return ih.len;
1213}
1214
fd22d62e
MS
1215static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1216 struct fuse_copy_state *cs,
02c048b9 1217 size_t nbytes)
fd22d62e 1218__releases(fiq->waitq.lock)
02c048b9 1219{
f88996a9 1220 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
fd22d62e 1221 return fuse_read_single_forget(fiq, cs, nbytes);
02c048b9 1222 else
fd22d62e 1223 return fuse_read_batch_forget(fiq, cs, nbytes);
02c048b9
MS
1224}
1225
334f485d
MS
1226/*
1227 * Read a single request into the userspace filesystem's buffer. This
1228 * function waits until a request is available, then removes it from
1229 * the pending list and copies request data to userspace buffer. If
f9a2842e
MS
1230 * no reply is needed (FORGET) or request has been aborted or there
1231 * was an error during the copying then it's finished by calling
334f485d
MS
1232 * request_end(). Otherwise add it to the processing list, and set
1233 * the 'sent' flag.
1234 */
c3021629
MS
1235static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
1236 struct fuse_copy_state *cs, size_t nbytes)
334f485d 1237{
82cbdcd3 1238 ssize_t err;
f88996a9 1239 struct fuse_iqueue *fiq = &fc->iq;
3a2b5b9c 1240 struct fuse_pqueue *fpq = &fc->pq;
334f485d
MS
1241 struct fuse_req *req;
1242 struct fuse_in *in;
334f485d
MS
1243 unsigned reqsize;
1244
1d3d752b 1245 restart:
4ce60812 1246 spin_lock(&fiq->waitq.lock);
e5ac1d1e 1247 err = -EAGAIN;
e16714d8 1248 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
f88996a9 1249 !request_pending(fiq))
e5ac1d1e
JD
1250 goto err_unlock;
1251
5250921b
MS
1252 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1253 !fiq->connected || request_pending(fiq));
1254 if (err)
1255 goto err_unlock;
1256
334f485d 1257 err = -ENODEV;
e16714d8 1258 if (!fiq->connected)
334f485d 1259 goto err_unlock;
334f485d 1260
f88996a9
MS
1261 if (!list_empty(&fiq->interrupts)) {
1262 req = list_entry(fiq->interrupts.next, struct fuse_req,
a4d27e75 1263 intr_entry);
fd22d62e 1264 return fuse_read_interrupt(fiq, cs, nbytes, req);
a4d27e75
MS
1265 }
1266
f88996a9
MS
1267 if (forget_pending(fiq)) {
1268 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
fd22d62e 1269 return fuse_read_forget(fc, fiq, cs, nbytes);
07e77dca 1270
f88996a9
MS
1271 if (fiq->forget_batch <= -8)
1272 fiq->forget_batch = 16;
07e77dca
MS
1273 }
1274
f88996a9 1275 req = list_entry(fiq->pending.next, struct fuse_req, list);
33e14b4d 1276 clear_bit(FR_PENDING, &req->flags);
ef759258 1277 list_del_init(&req->list);
4ce60812
MS
1278 spin_unlock(&fiq->waitq.lock);
1279
334f485d 1280 in = &req->in;
1d3d752b
MS
1281 reqsize = in->h.len;
1282 /* If request is too large, reply with an error and restart the read */
c3021629 1283 if (nbytes < reqsize) {
1d3d752b
MS
1284 req->out.h.error = -EIO;
1285 /* SETXATTR is special, since it may contain too large data */
1286 if (in->h.opcode == FUSE_SETXATTR)
1287 req->out.h.error = -E2BIG;
1288 request_end(fc, req);
1289 goto restart;
334f485d 1290 }
45a91cb1 1291 spin_lock(&fpq->lock);
82cbdcd3 1292 list_add(&req->list, &fpq->io);
45a91cb1 1293 spin_unlock(&fpq->lock);
c3021629
MS
1294 cs->req = req;
1295 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1d3d752b 1296 if (!err)
c3021629 1297 err = fuse_copy_args(cs, in->numargs, in->argpages,
1d3d752b 1298 (struct fuse_arg *) in->args, 0);
c3021629 1299 fuse_copy_finish(cs);
45a91cb1 1300 spin_lock(&fpq->lock);
825d6d33 1301 clear_bit(FR_LOCKED, &req->flags);
e96edd94 1302 if (!fpq->connected) {
82cbdcd3
MS
1303 err = -ENODEV;
1304 goto out_end;
c9c9d7df 1305 }
334f485d 1306 if (err) {
c9c9d7df 1307 req->out.h.error = -EIO;
82cbdcd3 1308 goto out_end;
334f485d 1309 }
825d6d33 1310 if (!test_bit(FR_ISREPLY, &req->flags)) {
82cbdcd3
MS
1311 err = reqsize;
1312 goto out_end;
334f485d 1313 }
82cbdcd3 1314 list_move_tail(&req->list, &fpq->processing);
45a91cb1 1315 spin_unlock(&fpq->lock);
82cbdcd3
MS
1316 set_bit(FR_SENT, &req->flags);
1317 /* matches barrier in request_wait_answer() */
1318 smp_mb__after_atomic();
1319 if (test_bit(FR_INTERRUPTED, &req->flags))
1320 queue_interrupt(fiq, req);
82cbdcd3 1321
334f485d
MS
1322 return reqsize;
1323
82cbdcd3 1324out_end:
77cd9d48
MS
1325 if (!test_bit(FR_PRIVATE, &req->flags))
1326 list_del_init(&req->list);
45a91cb1 1327 spin_unlock(&fpq->lock);
82cbdcd3
MS
1328 request_end(fc, req);
1329 return err;
1330
334f485d 1331 err_unlock:
4ce60812 1332 spin_unlock(&fiq->waitq.lock);
334f485d
MS
1333 return err;
1334}
1335
94e4fe2c
TVB
1336static int fuse_dev_open(struct inode *inode, struct file *file)
1337{
1338 /*
1339 * The fuse device's file's private_data is used to hold
1340 * the fuse_conn(ection) when it is mounted, and is used to
1341 * keep track of whether the file has been mounted already.
1342 */
1343 file->private_data = NULL;
1344 return 0;
1345}
1346
fbdbacca 1347static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
c3021629
MS
1348{
1349 struct fuse_copy_state cs;
1350 struct file *file = iocb->ki_filp;
cc080e9e
MS
1351 struct fuse_dev *fud = fuse_get_dev(file);
1352
1353 if (!fud)
c3021629
MS
1354 return -EPERM;
1355
fbdbacca
AV
1356 if (!iter_is_iovec(to))
1357 return -EINVAL;
1358
dc00809a 1359 fuse_copy_init(&cs, 1, to);
c3021629 1360
cc080e9e 1361 return fuse_dev_do_read(fud->fc, file, &cs, iov_iter_count(to));
c3021629
MS
1362}
1363
c3021629
MS
1364static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1365 struct pipe_inode_info *pipe,
1366 size_t len, unsigned int flags)
1367{
1368 int ret;
1369 int page_nr = 0;
1370 int do_wakeup = 0;
1371 struct pipe_buffer *bufs;
1372 struct fuse_copy_state cs;
cc080e9e
MS
1373 struct fuse_dev *fud = fuse_get_dev(in);
1374
1375 if (!fud)
c3021629
MS
1376 return -EPERM;
1377
07e77dca 1378 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
c3021629
MS
1379 if (!bufs)
1380 return -ENOMEM;
1381
dc00809a 1382 fuse_copy_init(&cs, 1, NULL);
c3021629
MS
1383 cs.pipebufs = bufs;
1384 cs.pipe = pipe;
cc080e9e 1385 ret = fuse_dev_do_read(fud->fc, in, &cs, len);
c3021629
MS
1386 if (ret < 0)
1387 goto out;
1388
1389 ret = 0;
1390 pipe_lock(pipe);
1391
1392 if (!pipe->readers) {
1393 send_sig(SIGPIPE, current, 0);
1394 if (!ret)
1395 ret = -EPIPE;
1396 goto out_unlock;
1397 }
1398
1399 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1400 ret = -EIO;
1401 goto out_unlock;
1402 }
1403
1404 while (page_nr < cs.nr_segs) {
1405 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1406 struct pipe_buffer *buf = pipe->bufs + newbuf;
1407
1408 buf->page = bufs[page_nr].page;
1409 buf->offset = bufs[page_nr].offset;
1410 buf->len = bufs[page_nr].len;
28a625cb
MS
1411 /*
1412 * Need to be careful about this. Having buf->ops in module
1413 * code can Oops if the buffer persists after module unload.
1414 */
1415 buf->ops = &nosteal_pipe_buf_ops;
c3021629
MS
1416
1417 pipe->nrbufs++;
1418 page_nr++;
1419 ret += buf->len;
1420
6447a3cf 1421 if (pipe->files)
c3021629
MS
1422 do_wakeup = 1;
1423 }
1424
1425out_unlock:
1426 pipe_unlock(pipe);
1427
1428 if (do_wakeup) {
1429 smp_mb();
1430 if (waitqueue_active(&pipe->wait))
1431 wake_up_interruptible(&pipe->wait);
1432 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1433 }
1434
1435out:
1436 for (; page_nr < cs.nr_segs; page_nr++)
1437 page_cache_release(bufs[page_nr].page);
1438
1439 kfree(bufs);
1440 return ret;
1441}
1442
95668a69
TH
1443static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1444 struct fuse_copy_state *cs)
1445{
1446 struct fuse_notify_poll_wakeup_out outarg;
f6d47a17 1447 int err = -EINVAL;
95668a69
TH
1448
1449 if (size != sizeof(outarg))
f6d47a17 1450 goto err;
95668a69
TH
1451
1452 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1453 if (err)
f6d47a17 1454 goto err;
95668a69 1455
f6d47a17 1456 fuse_copy_finish(cs);
95668a69 1457 return fuse_notify_poll_wakeup(fc, &outarg);
f6d47a17
MS
1458
1459err:
1460 fuse_copy_finish(cs);
1461 return err;
95668a69
TH
1462}
1463
3b463ae0
JM
1464static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1465 struct fuse_copy_state *cs)
1466{
1467 struct fuse_notify_inval_inode_out outarg;
1468 int err = -EINVAL;
1469
1470 if (size != sizeof(outarg))
1471 goto err;
1472
1473 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1474 if (err)
1475 goto err;
1476 fuse_copy_finish(cs);
1477
1478 down_read(&fc->killsb);
1479 err = -ENOENT;
b21dda43
MS
1480 if (fc->sb) {
1481 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1482 outarg.off, outarg.len);
1483 }
3b463ae0
JM
1484 up_read(&fc->killsb);
1485 return err;
1486
1487err:
1488 fuse_copy_finish(cs);
1489 return err;
1490}
1491
1492static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1493 struct fuse_copy_state *cs)
1494{
1495 struct fuse_notify_inval_entry_out outarg;
b2d82ee3
FW
1496 int err = -ENOMEM;
1497 char *buf;
3b463ae0
JM
1498 struct qstr name;
1499
b2d82ee3
FW
1500 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1501 if (!buf)
1502 goto err;
1503
1504 err = -EINVAL;
3b463ae0
JM
1505 if (size < sizeof(outarg))
1506 goto err;
1507
1508 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1509 if (err)
1510 goto err;
1511
1512 err = -ENAMETOOLONG;
1513 if (outarg.namelen > FUSE_NAME_MAX)
1514 goto err;
1515
c2183d1e
MS
1516 err = -EINVAL;
1517 if (size != sizeof(outarg) + outarg.namelen + 1)
1518 goto err;
1519
3b463ae0
JM
1520 name.name = buf;
1521 name.len = outarg.namelen;
1522 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1523 if (err)
1524 goto err;
1525 fuse_copy_finish(cs);
1526 buf[outarg.namelen] = 0;
1527 name.hash = full_name_hash(name.name, name.len);
1528
1529 down_read(&fc->killsb);
1530 err = -ENOENT;
b21dda43 1531 if (fc->sb)
451d0f59
JM
1532 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1533 up_read(&fc->killsb);
1534 kfree(buf);
1535 return err;
1536
1537err:
1538 kfree(buf);
1539 fuse_copy_finish(cs);
1540 return err;
1541}
1542
1543static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1544 struct fuse_copy_state *cs)
1545{
1546 struct fuse_notify_delete_out outarg;
1547 int err = -ENOMEM;
1548 char *buf;
1549 struct qstr name;
1550
1551 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1552 if (!buf)
1553 goto err;
1554
1555 err = -EINVAL;
1556 if (size < sizeof(outarg))
1557 goto err;
1558
1559 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1560 if (err)
1561 goto err;
1562
1563 err = -ENAMETOOLONG;
1564 if (outarg.namelen > FUSE_NAME_MAX)
1565 goto err;
1566
1567 err = -EINVAL;
1568 if (size != sizeof(outarg) + outarg.namelen + 1)
1569 goto err;
1570
1571 name.name = buf;
1572 name.len = outarg.namelen;
1573 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1574 if (err)
1575 goto err;
1576 fuse_copy_finish(cs);
1577 buf[outarg.namelen] = 0;
1578 name.hash = full_name_hash(name.name, name.len);
1579
1580 down_read(&fc->killsb);
1581 err = -ENOENT;
1582 if (fc->sb)
1583 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1584 outarg.child, &name);
3b463ae0 1585 up_read(&fc->killsb);
b2d82ee3 1586 kfree(buf);
3b463ae0
JM
1587 return err;
1588
1589err:
b2d82ee3 1590 kfree(buf);
3b463ae0
JM
1591 fuse_copy_finish(cs);
1592 return err;
1593}
1594
a1d75f25
MS
1595static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1596 struct fuse_copy_state *cs)
1597{
1598 struct fuse_notify_store_out outarg;
1599 struct inode *inode;
1600 struct address_space *mapping;
1601 u64 nodeid;
1602 int err;
1603 pgoff_t index;
1604 unsigned int offset;
1605 unsigned int num;
1606 loff_t file_size;
1607 loff_t end;
1608
1609 err = -EINVAL;
1610 if (size < sizeof(outarg))
1611 goto out_finish;
1612
1613 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1614 if (err)
1615 goto out_finish;
1616
1617 err = -EINVAL;
1618 if (size - sizeof(outarg) != outarg.size)
1619 goto out_finish;
1620
1621 nodeid = outarg.nodeid;
1622
1623 down_read(&fc->killsb);
1624
1625 err = -ENOENT;
1626 if (!fc->sb)
1627 goto out_up_killsb;
1628
1629 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1630 if (!inode)
1631 goto out_up_killsb;
1632
1633 mapping = inode->i_mapping;
1634 index = outarg.offset >> PAGE_CACHE_SHIFT;
1635 offset = outarg.offset & ~PAGE_CACHE_MASK;
1636 file_size = i_size_read(inode);
1637 end = outarg.offset + outarg.size;
1638 if (end > file_size) {
1639 file_size = end;
1640 fuse_write_update_size(inode, file_size);
1641 }
1642
1643 num = outarg.size;
1644 while (num) {
1645 struct page *page;
1646 unsigned int this_num;
1647
1648 err = -ENOMEM;
1649 page = find_or_create_page(mapping, index,
1650 mapping_gfp_mask(mapping));
1651 if (!page)
1652 goto out_iput;
1653
1654 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1655 err = fuse_copy_page(cs, &page, offset, this_num, 0);
063ec1e5
MS
1656 if (!err && offset == 0 &&
1657 (this_num == PAGE_CACHE_SIZE || file_size == end))
a1d75f25
MS
1658 SetPageUptodate(page);
1659 unlock_page(page);
1660 page_cache_release(page);
1661
1662 if (err)
1663 goto out_iput;
1664
1665 num -= this_num;
1666 offset = 0;
1667 index++;
1668 }
1669
1670 err = 0;
1671
1672out_iput:
1673 iput(inode);
1674out_up_killsb:
1675 up_read(&fc->killsb);
1676out_finish:
1677 fuse_copy_finish(cs);
1678 return err;
1679}
1680
2d45ba38
MS
1681static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1682{
b745bc85 1683 release_pages(req->pages, req->num_pages, false);
2d45ba38
MS
1684}
1685
1686static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1687 struct fuse_notify_retrieve_out *outarg)
1688{
1689 int err;
1690 struct address_space *mapping = inode->i_mapping;
1691 struct fuse_req *req;
1692 pgoff_t index;
1693 loff_t file_size;
1694 unsigned int num;
1695 unsigned int offset;
0157443c 1696 size_t total_len = 0;
4d53dc99 1697 int num_pages;
2d45ba38 1698
4d53dc99
MP
1699 offset = outarg->offset & ~PAGE_CACHE_MASK;
1700 file_size = i_size_read(inode);
1701
1702 num = outarg->size;
1703 if (outarg->offset > file_size)
1704 num = 0;
1705 else if (outarg->offset + num > file_size)
1706 num = file_size - outarg->offset;
1707
1708 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1709 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1710
1711 req = fuse_get_req(fc, num_pages);
2d45ba38
MS
1712 if (IS_ERR(req))
1713 return PTR_ERR(req);
1714
2d45ba38
MS
1715 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1716 req->in.h.nodeid = outarg->nodeid;
1717 req->in.numargs = 2;
1718 req->in.argpages = 1;
b2430d75 1719 req->page_descs[0].offset = offset;
2d45ba38
MS
1720 req->end = fuse_retrieve_end;
1721
1722 index = outarg->offset >> PAGE_CACHE_SHIFT;
2d45ba38 1723
4d53dc99 1724 while (num && req->num_pages < num_pages) {
2d45ba38
MS
1725 struct page *page;
1726 unsigned int this_num;
1727
1728 page = find_get_page(mapping, index);
1729 if (!page)
1730 break;
1731
1732 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1733 req->pages[req->num_pages] = page;
85f40aec 1734 req->page_descs[req->num_pages].length = this_num;
2d45ba38
MS
1735 req->num_pages++;
1736
c9e67d48 1737 offset = 0;
2d45ba38
MS
1738 num -= this_num;
1739 total_len += this_num;
48706d0a 1740 index++;
2d45ba38
MS
1741 }
1742 req->misc.retrieve_in.offset = outarg->offset;
1743 req->misc.retrieve_in.size = total_len;
1744 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1745 req->in.args[0].value = &req->misc.retrieve_in;
1746 req->in.args[1].size = total_len;
1747
1748 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1749 if (err)
1750 fuse_retrieve_end(fc, req);
1751
1752 return err;
1753}
1754
1755static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1756 struct fuse_copy_state *cs)
1757{
1758 struct fuse_notify_retrieve_out outarg;
1759 struct inode *inode;
1760 int err;
1761
1762 err = -EINVAL;
1763 if (size != sizeof(outarg))
1764 goto copy_finish;
1765
1766 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1767 if (err)
1768 goto copy_finish;
1769
1770 fuse_copy_finish(cs);
1771
1772 down_read(&fc->killsb);
1773 err = -ENOENT;
1774 if (fc->sb) {
1775 u64 nodeid = outarg.nodeid;
1776
1777 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1778 if (inode) {
1779 err = fuse_retrieve(fc, inode, &outarg);
1780 iput(inode);
1781 }
1782 }
1783 up_read(&fc->killsb);
1784
1785 return err;
1786
1787copy_finish:
1788 fuse_copy_finish(cs);
1789 return err;
1790}
1791
8599396b
TH
1792static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1793 unsigned int size, struct fuse_copy_state *cs)
1794{
0d278362
MS
1795 /* Don't try to move pages (yet) */
1796 cs->move_pages = 0;
1797
8599396b 1798 switch (code) {
95668a69
TH
1799 case FUSE_NOTIFY_POLL:
1800 return fuse_notify_poll(fc, size, cs);
1801
3b463ae0
JM
1802 case FUSE_NOTIFY_INVAL_INODE:
1803 return fuse_notify_inval_inode(fc, size, cs);
1804
1805 case FUSE_NOTIFY_INVAL_ENTRY:
1806 return fuse_notify_inval_entry(fc, size, cs);
1807
a1d75f25
MS
1808 case FUSE_NOTIFY_STORE:
1809 return fuse_notify_store(fc, size, cs);
1810
2d45ba38
MS
1811 case FUSE_NOTIFY_RETRIEVE:
1812 return fuse_notify_retrieve(fc, size, cs);
1813
451d0f59
JM
1814 case FUSE_NOTIFY_DELETE:
1815 return fuse_notify_delete(fc, size, cs);
1816
8599396b 1817 default:
f6d47a17 1818 fuse_copy_finish(cs);
8599396b
TH
1819 return -EINVAL;
1820 }
1821}
1822
334f485d 1823/* Look up request on processing list by unique ID */
3a2b5b9c 1824static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
334f485d 1825{
05726aca 1826 struct fuse_req *req;
334f485d 1827
3a2b5b9c 1828 list_for_each_entry(req, &fpq->processing, list) {
a4d27e75 1829 if (req->in.h.unique == unique || req->intr_unique == unique)
334f485d
MS
1830 return req;
1831 }
1832 return NULL;
1833}
1834
1835static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1836 unsigned nbytes)
1837{
1838 unsigned reqsize = sizeof(struct fuse_out_header);
1839
1840 if (out->h.error)
1841 return nbytes != reqsize ? -EINVAL : 0;
1842
1843 reqsize += len_args(out->numargs, out->args);
1844
1845 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1846 return -EINVAL;
1847 else if (reqsize > nbytes) {
1848 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1849 unsigned diffsize = reqsize - nbytes;
1850 if (diffsize > lastarg->size)
1851 return -EINVAL;
1852 lastarg->size -= diffsize;
1853 }
1854 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1855 out->page_zeroing);
1856}
1857
1858/*
1859 * Write a single reply to a request. First the header is copied from
1860 * the write buffer. The request is then searched on the processing
1861 * list by the unique ID found in the header. If found, then remove
1862 * it from the list and copy the rest of the buffer to the request.
1863 * The request is finished by calling request_end()
1864 */
dd3bb14f
MS
1865static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1866 struct fuse_copy_state *cs, size_t nbytes)
334f485d
MS
1867{
1868 int err;
3a2b5b9c 1869 struct fuse_pqueue *fpq = &fc->pq;
334f485d
MS
1870 struct fuse_req *req;
1871 struct fuse_out_header oh;
334f485d 1872
334f485d
MS
1873 if (nbytes < sizeof(struct fuse_out_header))
1874 return -EINVAL;
1875
dd3bb14f 1876 err = fuse_copy_one(cs, &oh, sizeof(oh));
334f485d
MS
1877 if (err)
1878 goto err_finish;
8599396b
TH
1879
1880 err = -EINVAL;
1881 if (oh.len != nbytes)
1882 goto err_finish;
1883
1884 /*
1885 * Zero oh.unique indicates unsolicited notification message
1886 * and error contains notification code.
1887 */
1888 if (!oh.unique) {
dd3bb14f 1889 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
8599396b
TH
1890 return err ? err : nbytes;
1891 }
1892
334f485d 1893 err = -EINVAL;
8599396b 1894 if (oh.error <= -1000 || oh.error > 0)
334f485d
MS
1895 goto err_finish;
1896
45a91cb1 1897 spin_lock(&fpq->lock);
69a53bf2 1898 err = -ENOENT;
e96edd94 1899 if (!fpq->connected)
45a91cb1 1900 goto err_unlock_pq;
69a53bf2 1901
3a2b5b9c 1902 req = request_find(fpq, oh.unique);
334f485d 1903 if (!req)
45a91cb1 1904 goto err_unlock_pq;
334f485d 1905
a4d27e75
MS
1906 /* Is it an interrupt reply? */
1907 if (req->intr_unique == oh.unique) {
45a91cb1
MS
1908 spin_unlock(&fpq->lock);
1909
a4d27e75
MS
1910 err = -EINVAL;
1911 if (nbytes != sizeof(struct fuse_out_header))
46c34a34 1912 goto err_finish;
a4d27e75
MS
1913
1914 if (oh.error == -ENOSYS)
1915 fc->no_interrupt = 1;
1916 else if (oh.error == -EAGAIN)
f88996a9 1917 queue_interrupt(&fc->iq, req);
a4d27e75 1918
dd3bb14f 1919 fuse_copy_finish(cs);
a4d27e75
MS
1920 return nbytes;
1921 }
1922
33e14b4d 1923 clear_bit(FR_SENT, &req->flags);
3a2b5b9c 1924 list_move(&req->list, &fpq->io);
334f485d 1925 req->out.h = oh;
825d6d33 1926 set_bit(FR_LOCKED, &req->flags);
45a91cb1 1927 spin_unlock(&fpq->lock);
dd3bb14f 1928 cs->req = req;
ce534fb0
MS
1929 if (!req->out.page_replace)
1930 cs->move_pages = 0;
334f485d 1931
dd3bb14f
MS
1932 err = copy_out_args(cs, &req->out, nbytes);
1933 fuse_copy_finish(cs);
334f485d 1934
45a91cb1 1935 spin_lock(&fpq->lock);
825d6d33 1936 clear_bit(FR_LOCKED, &req->flags);
e96edd94 1937 if (!fpq->connected)
0d8e84b0
MS
1938 err = -ENOENT;
1939 else if (err)
334f485d 1940 req->out.h.error = -EIO;
77cd9d48
MS
1941 if (!test_bit(FR_PRIVATE, &req->flags))
1942 list_del_init(&req->list);
45a91cb1 1943 spin_unlock(&fpq->lock);
46c34a34 1944
334f485d
MS
1945 request_end(fc, req);
1946
1947 return err ? err : nbytes;
1948
45a91cb1
MS
1949 err_unlock_pq:
1950 spin_unlock(&fpq->lock);
334f485d 1951 err_finish:
dd3bb14f 1952 fuse_copy_finish(cs);
334f485d
MS
1953 return err;
1954}
1955
fbdbacca 1956static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
dd3bb14f
MS
1957{
1958 struct fuse_copy_state cs;
cc080e9e
MS
1959 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1960
1961 if (!fud)
dd3bb14f
MS
1962 return -EPERM;
1963
fbdbacca
AV
1964 if (!iter_is_iovec(from))
1965 return -EINVAL;
1966
dc00809a 1967 fuse_copy_init(&cs, 0, from);
dd3bb14f 1968
cc080e9e 1969 return fuse_dev_do_write(fud->fc, &cs, iov_iter_count(from));
dd3bb14f
MS
1970}
1971
1972static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1973 struct file *out, loff_t *ppos,
1974 size_t len, unsigned int flags)
1975{
1976 unsigned nbuf;
1977 unsigned idx;
1978 struct pipe_buffer *bufs;
1979 struct fuse_copy_state cs;
cc080e9e 1980 struct fuse_dev *fud;
dd3bb14f
MS
1981 size_t rem;
1982 ssize_t ret;
1983
cc080e9e
MS
1984 fud = fuse_get_dev(out);
1985 if (!fud)
dd3bb14f
MS
1986 return -EPERM;
1987
07e77dca 1988 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
dd3bb14f
MS
1989 if (!bufs)
1990 return -ENOMEM;
1991
1992 pipe_lock(pipe);
1993 nbuf = 0;
1994 rem = 0;
1995 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1996 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1997
1998 ret = -EINVAL;
1999 if (rem < len) {
2000 pipe_unlock(pipe);
2001 goto out;
2002 }
2003
2004 rem = len;
2005 while (rem) {
2006 struct pipe_buffer *ibuf;
2007 struct pipe_buffer *obuf;
2008
2009 BUG_ON(nbuf >= pipe->buffers);
2010 BUG_ON(!pipe->nrbufs);
2011 ibuf = &pipe->bufs[pipe->curbuf];
2012 obuf = &bufs[nbuf];
2013
2014 if (rem >= ibuf->len) {
2015 *obuf = *ibuf;
2016 ibuf->ops = NULL;
2017 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
2018 pipe->nrbufs--;
2019 } else {
2020 ibuf->ops->get(pipe, ibuf);
2021 *obuf = *ibuf;
2022 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2023 obuf->len = rem;
2024 ibuf->offset += obuf->len;
2025 ibuf->len -= obuf->len;
2026 }
2027 nbuf++;
2028 rem -= obuf->len;
2029 }
2030 pipe_unlock(pipe);
2031
dc00809a 2032 fuse_copy_init(&cs, 0, NULL);
dd3bb14f 2033 cs.pipebufs = bufs;
6c09e94a 2034 cs.nr_segs = nbuf;
dd3bb14f
MS
2035 cs.pipe = pipe;
2036
ce534fb0
MS
2037 if (flags & SPLICE_F_MOVE)
2038 cs.move_pages = 1;
2039
cc080e9e 2040 ret = fuse_dev_do_write(fud->fc, &cs, len);
dd3bb14f
MS
2041
2042 for (idx = 0; idx < nbuf; idx++) {
2043 struct pipe_buffer *buf = &bufs[idx];
2044 buf->ops->release(pipe, buf);
2045 }
2046out:
2047 kfree(bufs);
2048 return ret;
2049}
2050
334f485d
MS
2051static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2052{
334f485d 2053 unsigned mask = POLLOUT | POLLWRNORM;
f88996a9 2054 struct fuse_iqueue *fiq;
cc080e9e
MS
2055 struct fuse_dev *fud = fuse_get_dev(file);
2056
2057 if (!fud)
7025d9ad 2058 return POLLERR;
334f485d 2059
cc080e9e 2060 fiq = &fud->fc->iq;
f88996a9 2061 poll_wait(file, &fiq->waitq, wait);
334f485d 2062
4ce60812 2063 spin_lock(&fiq->waitq.lock);
e16714d8 2064 if (!fiq->connected)
7025d9ad 2065 mask = POLLERR;
f88996a9 2066 else if (request_pending(fiq))
7025d9ad 2067 mask |= POLLIN | POLLRDNORM;
4ce60812 2068 spin_unlock(&fiq->waitq.lock);
334f485d
MS
2069
2070 return mask;
2071}
2072
69a53bf2
MS
2073/*
2074 * Abort all requests on the given list (pending or processing)
2075 *
d7133114 2076 * This function releases and reacquires fc->lock
69a53bf2 2077 */
334f485d
MS
2078static void end_requests(struct fuse_conn *fc, struct list_head *head)
2079{
2080 while (!list_empty(head)) {
2081 struct fuse_req *req;
2082 req = list_entry(head->next, struct fuse_req, list);
334f485d 2083 req->out.h.error = -ECONNABORTED;
33e14b4d
MS
2084 clear_bit(FR_PENDING, &req->flags);
2085 clear_bit(FR_SENT, &req->flags);
f377cb79 2086 list_del_init(&req->list);
334f485d 2087 request_end(fc, req);
334f485d
MS
2088 }
2089}
2090
357ccf2b
BG
2091static void end_polls(struct fuse_conn *fc)
2092{
2093 struct rb_node *p;
2094
2095 p = rb_first(&fc->polled_files);
2096
2097 while (p) {
2098 struct fuse_file *ff;
2099 ff = rb_entry(p, struct fuse_file, polled_node);
2100 wake_up_interruptible_all(&ff->poll_wait);
2101
2102 p = rb_next(p);
2103 }
2104}
2105
69a53bf2
MS
2106/*
2107 * Abort all requests.
2108 *
b716d425
MS
2109 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2110 * filesystem.
2111 *
2112 * The same effect is usually achievable through killing the filesystem daemon
2113 * and all users of the filesystem. The exception is the combination of an
2114 * asynchronous request and the tricky deadlock (see
2115 * Documentation/filesystems/fuse.txt).
69a53bf2 2116 *
b716d425
MS
2117 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2118 * requests, they should be finished off immediately. Locked requests will be
2119 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2120 * requests. It is possible that some request will finish before we can. This
2121 * is OK, the request will in that case be removed from the list before we touch
2122 * it.
69a53bf2
MS
2123 */
2124void fuse_abort_conn(struct fuse_conn *fc)
2125{
f88996a9 2126 struct fuse_iqueue *fiq = &fc->iq;
3a2b5b9c 2127 struct fuse_pqueue *fpq = &fc->pq;
f88996a9 2128
d7133114 2129 spin_lock(&fc->lock);
69a53bf2 2130 if (fc->connected) {
b716d425 2131 struct fuse_req *req, *next;
41f98274
MS
2132 LIST_HEAD(to_end1);
2133 LIST_HEAD(to_end2);
b716d425 2134
69a53bf2 2135 fc->connected = 0;
51eb01e7 2136 fc->blocked = 0;
9759bd51 2137 fuse_set_initialized(fc);
45a91cb1 2138 spin_lock(&fpq->lock);
e96edd94 2139 fpq->connected = 0;
3a2b5b9c 2140 list_for_each_entry_safe(req, next, &fpq->io, list) {
b716d425
MS
2141 req->out.h.error = -ECONNABORTED;
2142 spin_lock(&req->waitq.lock);
2143 set_bit(FR_ABORTED, &req->flags);
77cd9d48
MS
2144 if (!test_bit(FR_LOCKED, &req->flags)) {
2145 set_bit(FR_PRIVATE, &req->flags);
41f98274 2146 list_move(&req->list, &to_end1);
77cd9d48 2147 }
b716d425
MS
2148 spin_unlock(&req->waitq.lock);
2149 }
24b4d33d 2150 list_splice_init(&fpq->processing, &to_end2);
45a91cb1 2151 spin_unlock(&fpq->lock);
41f98274
MS
2152 fc->max_background = UINT_MAX;
2153 flush_bg_queue(fc);
8c91189a 2154
4ce60812 2155 spin_lock(&fiq->waitq.lock);
8c91189a 2156 fiq->connected = 0;
f88996a9 2157 list_splice_init(&fiq->pending, &to_end2);
8c91189a
MS
2158 while (forget_pending(fiq))
2159 kfree(dequeue_forget(fiq, 1, NULL));
4ce60812
MS
2160 wake_up_all_locked(&fiq->waitq);
2161 spin_unlock(&fiq->waitq.lock);
8c91189a 2162 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
ee314a87
MS
2163 end_polls(fc);
2164 wake_up_all(&fc->blocked_waitq);
2165 spin_unlock(&fc->lock);
8c91189a 2166
41f98274
MS
2167 while (!list_empty(&to_end1)) {
2168 req = list_first_entry(&to_end1, struct fuse_req, list);
b716d425 2169 __fuse_get_request(req);
f377cb79 2170 list_del_init(&req->list);
b716d425 2171 request_end(fc, req);
b716d425 2172 }
41f98274 2173 end_requests(fc, &to_end2);
ee314a87
MS
2174 } else {
2175 spin_unlock(&fc->lock);
69a53bf2 2176 }
69a53bf2 2177}
08cbf542 2178EXPORT_SYMBOL_GPL(fuse_abort_conn);
69a53bf2 2179
08cbf542 2180int fuse_dev_release(struct inode *inode, struct file *file)
334f485d 2181{
cc080e9e
MS
2182 struct fuse_dev *fud = fuse_get_dev(file);
2183
2184 if (fud) {
2185 struct fuse_conn *fc = fud->fc;
2186
3a2b5b9c 2187 WARN_ON(!list_empty(&fc->pq.io));
f88996a9 2188 WARN_ON(fc->iq.fasync != NULL);
ccd0a0bd 2189 fuse_abort_conn(fc);
cc080e9e 2190 fuse_dev_free(fud);
385a17bf 2191 }
f543f253 2192
334f485d
MS
2193 return 0;
2194}
08cbf542 2195EXPORT_SYMBOL_GPL(fuse_dev_release);
334f485d 2196
385a17bf
JD
2197static int fuse_dev_fasync(int fd, struct file *file, int on)
2198{
cc080e9e
MS
2199 struct fuse_dev *fud = fuse_get_dev(file);
2200
2201 if (!fud)
a87046d8 2202 return -EPERM;
385a17bf
JD
2203
2204 /* No locking - fasync_helper does its own locking */
cc080e9e 2205 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
385a17bf
JD
2206}
2207
00c570f4
MS
2208static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2209{
cc080e9e
MS
2210 struct fuse_dev *fud;
2211
00c570f4
MS
2212 if (new->private_data)
2213 return -EINVAL;
2214
cc080e9e
MS
2215 fud = fuse_dev_alloc(fc);
2216 if (!fud)
2217 return -ENOMEM;
2218
2219 new->private_data = fud;
00c570f4
MS
2220
2221 return 0;
2222}
2223
2224static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2225 unsigned long arg)
2226{
2227 int err = -ENOTTY;
2228
2229 if (cmd == FUSE_DEV_IOC_CLONE) {
2230 int oldfd;
2231
2232 err = -EFAULT;
2233 if (!get_user(oldfd, (__u32 __user *) arg)) {
2234 struct file *old = fget(oldfd);
2235
2236 err = -EINVAL;
2237 if (old) {
cc080e9e 2238 struct fuse_dev *fud = fuse_get_dev(old);
00c570f4 2239
cc080e9e 2240 if (fud) {
00c570f4 2241 mutex_lock(&fuse_mutex);
cc080e9e 2242 err = fuse_device_clone(fud->fc, file);
00c570f4
MS
2243 mutex_unlock(&fuse_mutex);
2244 }
2245 fput(old);
2246 }
2247 }
2248 }
2249 return err;
2250}
2251
4b6f5d20 2252const struct file_operations fuse_dev_operations = {
334f485d 2253 .owner = THIS_MODULE,
94e4fe2c 2254 .open = fuse_dev_open,
334f485d 2255 .llseek = no_llseek,
fbdbacca 2256 .read_iter = fuse_dev_read,
c3021629 2257 .splice_read = fuse_dev_splice_read,
fbdbacca 2258 .write_iter = fuse_dev_write,
dd3bb14f 2259 .splice_write = fuse_dev_splice_write,
334f485d
MS
2260 .poll = fuse_dev_poll,
2261 .release = fuse_dev_release,
385a17bf 2262 .fasync = fuse_dev_fasync,
00c570f4
MS
2263 .unlocked_ioctl = fuse_dev_ioctl,
2264 .compat_ioctl = fuse_dev_ioctl,
334f485d 2265};
08cbf542 2266EXPORT_SYMBOL_GPL(fuse_dev_operations);
334f485d
MS
2267
2268static struct miscdevice fuse_miscdevice = {
2269 .minor = FUSE_MINOR,
2270 .name = "fuse",
2271 .fops = &fuse_dev_operations,
2272};
2273
2274int __init fuse_dev_init(void)
2275{
2276 int err = -ENOMEM;
2277 fuse_req_cachep = kmem_cache_create("fuse_request",
2278 sizeof(struct fuse_req),
20c2df83 2279 0, 0, NULL);
334f485d
MS
2280 if (!fuse_req_cachep)
2281 goto out;
2282
2283 err = misc_register(&fuse_miscdevice);
2284 if (err)
2285 goto out_cache_clean;
2286
2287 return 0;
2288
2289 out_cache_clean:
2290 kmem_cache_destroy(fuse_req_cachep);
2291 out:
2292 return err;
2293}
2294
2295void fuse_dev_cleanup(void)
2296{
2297 misc_deregister(&fuse_miscdevice);
2298 kmem_cache_destroy(fuse_req_cachep);
2299}