]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/fuse/dev.c
skb_splice_bits(): get rid of callback
[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
36cf66ed 102void __fuse_get_request(struct fuse_req *req)
334f485d
MS
103{
104 atomic_inc(&req->count);
105}
106
107/* Must be called with > 1 refcount */
108static void __fuse_put_request(struct fuse_req *req)
109{
110 BUG_ON(atomic_read(&req->count) < 2);
111 atomic_dec(&req->count);
112}
113
33649c91
MS
114static void fuse_req_init_context(struct fuse_req *req)
115{
499dcf20
EB
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());
33649c91
MS
118 req->in.h.pid = current->pid;
119}
120
9759bd51
MS
121void 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
0aada884
MP
128static bool fuse_block_alloc(struct fuse_conn *fc, bool for_background)
129{
130 return !fc->initialized || (for_background && fc->blocked);
131}
132
8b41e671
MP
133static struct fuse_req *__fuse_get_req(struct fuse_conn *fc, unsigned npages,
134 bool for_background)
334f485d 135{
08a53cdc 136 struct fuse_req *req;
08a53cdc 137 int err;
9bc5ddda 138 atomic_inc(&fc->num_waiting);
0aada884
MP
139
140 if (fuse_block_alloc(fc, for_background)) {
0aada884 141 err = -EINTR;
7d3a07fc
AV
142 if (wait_event_killable_exclusive(fc->blocked_waitq,
143 !fuse_block_alloc(fc, for_background)))
0aada884
MP
144 goto out;
145 }
9759bd51
MS
146 /* Matches smp_wmb() in fuse_set_initialized() */
147 smp_rmb();
08a53cdc 148
51eb01e7
MS
149 err = -ENOTCONN;
150 if (!fc->connected)
151 goto out;
152
de155226
MS
153 err = -ECONNREFUSED;
154 if (fc->conn_error)
155 goto out;
156
b111c8c0 157 req = fuse_request_alloc(npages);
9bc5ddda 158 err = -ENOMEM;
722d2bea
MP
159 if (!req) {
160 if (for_background)
161 wake_up(&fc->blocked_waitq);
9bc5ddda 162 goto out;
722d2bea 163 }
334f485d 164
33649c91 165 fuse_req_init_context(req);
825d6d33
MS
166 __set_bit(FR_WAITING, &req->flags);
167 if (for_background)
168 __set_bit(FR_BACKGROUND, &req->flags);
169
334f485d 170 return req;
9bc5ddda
MS
171
172 out:
173 atomic_dec(&fc->num_waiting);
174 return ERR_PTR(err);
334f485d 175}
8b41e671
MP
176
177struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
178{
179 return __fuse_get_req(fc, npages, false);
180}
08cbf542 181EXPORT_SYMBOL_GPL(fuse_get_req);
334f485d 182
8b41e671
MP
183struct fuse_req *fuse_get_req_for_background(struct fuse_conn *fc,
184 unsigned npages)
185{
186 return __fuse_get_req(fc, npages, true);
187}
188EXPORT_SYMBOL_GPL(fuse_get_req_for_background);
189
33649c91
MS
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 */
195static 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 {
de5e3dec 202 wait_event(fc->reserved_req_waitq, ff->reserved_req);
33649c91
MS
203 spin_lock(&fc->lock);
204 if (ff->reserved_req) {
205 req = ff->reserved_req;
206 ff->reserved_req = NULL;
cb0942b8 207 req->stolen_file = get_file(file);
33649c91
MS
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 */
218static 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);
b2430d75 224 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
33649c91
MS
225 BUG_ON(ff->reserved_req);
226 ff->reserved_req = req;
de5e3dec 227 wake_up_all(&fc->reserved_req_waitq);
33649c91
MS
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 */
b111c8c0
MP
245struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
246 struct file *file)
33649c91
MS
247{
248 struct fuse_req *req;
249
250 atomic_inc(&fc->num_waiting);
0aada884 251 wait_event(fc->blocked_waitq, fc->initialized);
9759bd51
MS
252 /* Matches smp_wmb() in fuse_set_initialized() */
253 smp_rmb();
b111c8c0 254 req = fuse_request_alloc(0);
33649c91
MS
255 if (!req)
256 req = get_reserved_req(fc, file);
257
258 fuse_req_init_context(req);
825d6d33
MS
259 __set_bit(FR_WAITING, &req->flags);
260 __clear_bit(FR_BACKGROUND, &req->flags);
33649c91
MS
261 return req;
262}
263
334f485d 264void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
7128ec2a
MS
265{
266 if (atomic_dec_and_test(&req->count)) {
825d6d33 267 if (test_bit(FR_BACKGROUND, &req->flags)) {
722d2bea
MP
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
825d6d33
MS
278 if (test_bit(FR_WAITING, &req->flags)) {
279 __clear_bit(FR_WAITING, &req->flags);
9bc5ddda 280 atomic_dec(&fc->num_waiting);
73e0e738 281 }
33649c91
MS
282
283 if (req->stolen_file)
284 put_reserved_req(fc, req);
285 else
286 fuse_request_free(req);
7128ec2a
MS
287 }
288}
08cbf542 289EXPORT_SYMBOL_GPL(fuse_put_request);
7128ec2a 290
d12def1b
MS
291static 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
f88996a9 302static u64 fuse_get_unique(struct fuse_iqueue *fiq)
d12def1b 303{
f88996a9 304 return ++fiq->reqctr;
d12def1b
MS
305}
306
f88996a9 307static void queue_request(struct fuse_iqueue *fiq, struct fuse_req *req)
d12def1b 308{
d12def1b
MS
309 req->in.h.len = sizeof(struct fuse_in_header) +
310 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
f88996a9 311 list_add_tail(&req->list, &fiq->pending);
4ce60812 312 wake_up_locked(&fiq->waitq);
f88996a9 313 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
d12def1b
MS
314}
315
07e77dca
MS
316void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
317 u64 nodeid, u64 nlookup)
318{
f88996a9
MS
319 struct fuse_iqueue *fiq = &fc->iq;
320
02c048b9
MS
321 forget->forget_one.nodeid = nodeid;
322 forget->forget_one.nlookup = nlookup;
07e77dca 323
4ce60812 324 spin_lock(&fiq->waitq.lock);
e16714d8 325 if (fiq->connected) {
f88996a9
MS
326 fiq->forget_list_tail->next = forget;
327 fiq->forget_list_tail = forget;
4ce60812 328 wake_up_locked(&fiq->waitq);
f88996a9 329 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
5dfcc87f
MS
330 } else {
331 kfree(forget);
332 }
4ce60812 333 spin_unlock(&fiq->waitq.lock);
07e77dca
MS
334}
335
d12def1b
MS
336static void flush_bg_queue(struct fuse_conn *fc)
337{
7a6d3c8b 338 while (fc->active_background < fc->max_background &&
d12def1b
MS
339 !list_empty(&fc->bg_queue)) {
340 struct fuse_req *req;
f88996a9 341 struct fuse_iqueue *fiq = &fc->iq;
d12def1b
MS
342
343 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
344 list_del(&req->list);
345 fc->active_background++;
4ce60812 346 spin_lock(&fiq->waitq.lock);
f88996a9
MS
347 req->in.h.unique = fuse_get_unique(fiq);
348 queue_request(fiq, req);
4ce60812 349 spin_unlock(&fiq->waitq.lock);
d12def1b
MS
350 }
351}
352
334f485d
MS
353/*
354 * This function is called when a request is finished. Either a reply
f9a2842e 355 * has arrived or it was aborted (and not yet sent) or some error
f43b155a 356 * occurred during communication with userspace, or the device file
51eb01e7
MS
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
334f485d
MS
360 */
361static void request_end(struct fuse_conn *fc, struct fuse_req *req)
362{
4ce60812 363 struct fuse_iqueue *fiq = &fc->iq;
365ae710 364
efe2800f 365 if (test_and_set_bit(FR_FINISHED, &req->flags))
365ae710 366 return;
365ae710 367
4ce60812 368 spin_lock(&fiq->waitq.lock);
0d8e84b0 369 list_del_init(&req->intr_entry);
4ce60812 370 spin_unlock(&fiq->waitq.lock);
33e14b4d
MS
371 WARN_ON(test_bit(FR_PENDING, &req->flags));
372 WARN_ON(test_bit(FR_SENT, &req->flags));
825d6d33 373 if (test_bit(FR_BACKGROUND, &req->flags)) {
efe2800f 374 spin_lock(&fc->lock);
825d6d33 375 clear_bit(FR_BACKGROUND, &req->flags);
722d2bea 376 if (fc->num_background == fc->max_background)
51eb01e7 377 fc->blocked = 0;
722d2bea
MP
378
379 /* Wake up next waiter, if any */
3c18ef81 380 if (!fc->blocked && waitqueue_active(&fc->blocked_waitq))
722d2bea
MP
381 wake_up(&fc->blocked_waitq);
382
7a6d3c8b 383 if (fc->num_background == fc->congestion_threshold &&
a325f9b9 384 fc->connected && fc->bdi_initialized) {
8aa7e847
JA
385 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
386 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
f92b99b9 387 }
51eb01e7 388 fc->num_background--;
d12def1b
MS
389 fc->active_background--;
390 flush_bg_queue(fc);
efe2800f 391 spin_unlock(&fc->lock);
334f485d 392 }
51eb01e7 393 wake_up(&req->waitq);
1e6881c3
MS
394 if (req->end)
395 req->end(fc, req);
e9bb09dd 396 fuse_put_request(fc, req);
334f485d
MS
397}
398
f88996a9 399static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
a4d27e75 400{
4ce60812 401 spin_lock(&fiq->waitq.lock);
8f7bb368
MS
402 if (list_empty(&req->intr_entry)) {
403 list_add_tail(&req->intr_entry, &fiq->interrupts);
404 wake_up_locked(&fiq->waitq);
405 }
4ce60812 406 spin_unlock(&fiq->waitq.lock);
f88996a9 407 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
a4d27e75
MS
408}
409
7c352bdf 410static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
334f485d 411{
4ce60812 412 struct fuse_iqueue *fiq = &fc->iq;
c4775267
MS
413 int err;
414
a4d27e75
MS
415 if (!fc->no_interrupt) {
416 /* Any signal may interrupt this */
c4775267 417 err = wait_event_interruptible(req->waitq,
33e14b4d 418 test_bit(FR_FINISHED, &req->flags));
c4775267 419 if (!err)
a4d27e75
MS
420 return;
421
825d6d33 422 set_bit(FR_INTERRUPTED, &req->flags);
8f7bb368
MS
423 /* matches barrier in fuse_dev_do_read() */
424 smp_mb__after_atomic();
33e14b4d 425 if (test_bit(FR_SENT, &req->flags))
4ce60812 426 queue_interrupt(fiq, req);
a4d27e75
MS
427 }
428
825d6d33 429 if (!test_bit(FR_FORCE, &req->flags)) {
a4d27e75 430 /* Only fatal signals may interrupt this */
7d3a07fc 431 err = wait_event_killable(req->waitq,
33e14b4d 432 test_bit(FR_FINISHED, &req->flags));
c4775267 433 if (!err)
a131de0a
MS
434 return;
435
4ce60812 436 spin_lock(&fiq->waitq.lock);
a131de0a 437 /* Request is not yet in userspace, bail out */
33e14b4d 438 if (test_bit(FR_PENDING, &req->flags)) {
a131de0a 439 list_del(&req->list);
4ce60812 440 spin_unlock(&fiq->waitq.lock);
a131de0a
MS
441 __fuse_put_request(req);
442 req->out.h.error = -EINTR;
443 return;
444 }
4ce60812 445 spin_unlock(&fiq->waitq.lock);
51eb01e7 446 }
334f485d 447
a131de0a
MS
448 /*
449 * Either request is already in userspace, or it was forced.
450 * Wait it out.
451 */
33e14b4d 452 wait_event(req->waitq, test_bit(FR_FINISHED, &req->flags));
334f485d
MS
453}
454
6a4e922c 455static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
334f485d 456{
e16714d8
MS
457 struct fuse_iqueue *fiq = &fc->iq;
458
825d6d33 459 BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
4ce60812 460 spin_lock(&fiq->waitq.lock);
e16714d8 461 if (!fiq->connected) {
4ce60812 462 spin_unlock(&fiq->waitq.lock);
334f485d 463 req->out.h.error = -ENOTCONN;
c4775267 464 } else {
f88996a9
MS
465 req->in.h.unique = fuse_get_unique(fiq);
466 queue_request(fiq, req);
334f485d
MS
467 /* acquire extra reference, since request is still needed
468 after request_end() */
469 __fuse_get_request(req);
4ce60812 470 spin_unlock(&fiq->waitq.lock);
334f485d 471
7c352bdf 472 request_wait_answer(fc, req);
c4775267
MS
473 /* Pairs with smp_wmb() in request_end() */
474 smp_rmb();
334f485d 475 }
334f485d 476}
6a4e922c
EW
477
478void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
479{
825d6d33
MS
480 __set_bit(FR_ISREPLY, &req->flags);
481 if (!test_bit(FR_WAITING, &req->flags)) {
482 __set_bit(FR_WAITING, &req->flags);
5437f241
MS
483 atomic_inc(&fc->num_waiting);
484 }
6a4e922c
EW
485 __fuse_request_send(fc, req);
486}
08cbf542 487EXPORT_SYMBOL_GPL(fuse_request_send);
334f485d 488
21f62174
MS
489static 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
7078187a
MS
522ssize_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
21f62174
MS
531 /* Needs to be done after fuse_get_req() so that fc->minor is valid */
532 fuse_adjust_compat(fc, args);
533
7078187a
MS
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
f0139aa8
MS
554/*
555 * Called under fc->lock
556 *
557 * fc->connected must have been checked previously
558 */
559void fuse_request_send_background_locked(struct fuse_conn *fc,
560 struct fuse_req *req)
d12def1b 561{
825d6d33
MS
562 BUG_ON(!test_bit(FR_BACKGROUND, &req->flags));
563 if (!test_bit(FR_WAITING, &req->flags)) {
564 __set_bit(FR_WAITING, &req->flags);
5437f241
MS
565 atomic_inc(&fc->num_waiting);
566 }
825d6d33 567 __set_bit(FR_ISREPLY, &req->flags);
d12def1b 568 fc->num_background++;
7a6d3c8b 569 if (fc->num_background == fc->max_background)
d12def1b 570 fc->blocked = 1;
7a6d3c8b 571 if (fc->num_background == fc->congestion_threshold &&
a325f9b9 572 fc->bdi_initialized) {
8aa7e847
JA
573 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
574 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
d12def1b
MS
575 }
576 list_add_tail(&req->list, &fc->bg_queue);
577 flush_bg_queue(fc);
578}
579
f0139aa8 580void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
334f485d 581{
42dc6211 582 BUG_ON(!req->end);
d7133114 583 spin_lock(&fc->lock);
1e9a4ed9 584 if (fc->connected) {
f0139aa8 585 fuse_request_send_background_locked(fc, req);
d7133114 586 spin_unlock(&fc->lock);
334f485d 587 } else {
42dc6211 588 spin_unlock(&fc->lock);
334f485d 589 req->out.h.error = -ENOTCONN;
42dc6211
MS
590 req->end(fc, req);
591 fuse_put_request(fc, req);
334f485d
MS
592 }
593}
08cbf542 594EXPORT_SYMBOL_GPL(fuse_request_send_background);
334f485d 595
2d45ba38
MS
596static int fuse_request_send_notify_reply(struct fuse_conn *fc,
597 struct fuse_req *req, u64 unique)
598{
599 int err = -ENODEV;
f88996a9 600 struct fuse_iqueue *fiq = &fc->iq;
2d45ba38 601
825d6d33 602 __clear_bit(FR_ISREPLY, &req->flags);
2d45ba38 603 req->in.h.unique = unique;
4ce60812 604 spin_lock(&fiq->waitq.lock);
e16714d8 605 if (fiq->connected) {
f88996a9 606 queue_request(fiq, req);
2d45ba38
MS
607 err = 0;
608 }
4ce60812 609 spin_unlock(&fiq->waitq.lock);
2d45ba38
MS
610
611 return err;
612}
613
0b05b183
AA
614void fuse_force_forget(struct file *file, u64 nodeid)
615{
6131ffaa 616 struct inode *inode = file_inode(file);
0b05b183
AA
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;
b111c8c0 623 req = fuse_get_req_nofail_nopages(fc, file);
0b05b183
AA
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;
825d6d33 629 __clear_bit(FR_ISREPLY, &req->flags);
6a4e922c
EW
630 __fuse_request_send(fc, req);
631 /* ignore errors */
632 fuse_put_request(fc, req);
0b05b183
AA
633}
634
334f485d
MS
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
f9a2842e 638 * aborted bail out.
334f485d 639 */
dc00809a 640static int lock_request(struct fuse_req *req)
334f485d
MS
641{
642 int err = 0;
643 if (req) {
dc00809a 644 spin_lock(&req->waitq.lock);
825d6d33 645 if (test_bit(FR_ABORTED, &req->flags))
334f485d
MS
646 err = -ENOENT;
647 else
825d6d33 648 set_bit(FR_LOCKED, &req->flags);
dc00809a 649 spin_unlock(&req->waitq.lock);
334f485d
MS
650 }
651 return err;
652}
653
654/*
0d8e84b0
MS
655 * Unlock request. If it was aborted while locked, caller is responsible
656 * for unlocking and ending the request.
334f485d 657 */
dc00809a 658static int unlock_request(struct fuse_req *req)
334f485d 659{
0d8e84b0 660 int err = 0;
334f485d 661 if (req) {
dc00809a 662 spin_lock(&req->waitq.lock);
825d6d33 663 if (test_bit(FR_ABORTED, &req->flags))
0d8e84b0
MS
664 err = -ENOENT;
665 else
825d6d33 666 clear_bit(FR_LOCKED, &req->flags);
dc00809a 667 spin_unlock(&req->waitq.lock);
334f485d 668 }
0d8e84b0 669 return err;
334f485d
MS
670}
671
672struct fuse_copy_state {
673 int write;
674 struct fuse_req *req;
6c09e94a 675 struct iov_iter *iter;
dd3bb14f
MS
676 struct pipe_buffer *pipebufs;
677 struct pipe_buffer *currbuf;
678 struct pipe_inode_info *pipe;
334f485d 679 unsigned long nr_segs;
334f485d 680 struct page *pg;
334f485d 681 unsigned len;
c55a01d3 682 unsigned offset;
ce534fb0 683 unsigned move_pages:1;
334f485d
MS
684};
685
dc00809a 686static void fuse_copy_init(struct fuse_copy_state *cs, int write,
6c09e94a 687 struct iov_iter *iter)
334f485d
MS
688{
689 memset(cs, 0, sizeof(*cs));
690 cs->write = write;
6c09e94a 691 cs->iter = iter;
334f485d
MS
692}
693
694/* Unmap and put previous page of userspace buffer */
8bfc016d 695static void fuse_copy_finish(struct fuse_copy_state *cs)
334f485d 696{
dd3bb14f
MS
697 if (cs->currbuf) {
698 struct pipe_buffer *buf = cs->currbuf;
699
c55a01d3 700 if (cs->write)
c3021629 701 buf->len = PAGE_SIZE - cs->len;
dd3bb14f 702 cs->currbuf = NULL;
c55a01d3 703 } else if (cs->pg) {
334f485d
MS
704 if (cs->write) {
705 flush_dcache_page(cs->pg);
706 set_page_dirty_lock(cs->pg);
707 }
708 put_page(cs->pg);
334f485d 709 }
c55a01d3 710 cs->pg = NULL;
334f485d
MS
711}
712
713/*
714 * Get another pagefull of userspace buffer, and map it to kernel
715 * address space, and lock request
716 */
717static int fuse_copy_fill(struct fuse_copy_state *cs)
718{
c55a01d3 719 struct page *page;
334f485d
MS
720 int err;
721
dc00809a 722 err = unlock_request(cs->req);
0d8e84b0
MS
723 if (err)
724 return err;
725
334f485d 726 fuse_copy_finish(cs);
dd3bb14f
MS
727 if (cs->pipebufs) {
728 struct pipe_buffer *buf = cs->pipebufs;
729
c3021629
MS
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;
c55a01d3
MS
737 cs->pg = buf->page;
738 cs->offset = buf->offset;
c3021629 739 cs->len = buf->len;
c3021629
MS
740 cs->pipebufs++;
741 cs->nr_segs--;
742 } else {
c3021629
MS
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;
c55a01d3
MS
755 cs->pg = page;
756 cs->offset = 0;
c3021629
MS
757 cs->len = PAGE_SIZE;
758 cs->pipebufs++;
759 cs->nr_segs++;
760 }
dd3bb14f 761 } else {
6c09e94a
AV
762 size_t off;
763 err = iov_iter_get_pages(cs->iter, &page, PAGE_SIZE, 1, &off);
dd3bb14f
MS
764 if (err < 0)
765 return err;
6c09e94a
AV
766 BUG_ON(!err);
767 cs->len = err;
768 cs->offset = off;
c55a01d3 769 cs->pg = page;
6c09e94a
AV
770 cs->offset = off;
771 iov_iter_advance(cs->iter, err);
334f485d 772 }
334f485d 773
dc00809a 774 return lock_request(cs->req);
334f485d
MS
775}
776
777/* Do as much copy to/from userspace buffer as we can */
8bfc016d 778static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
334f485d
MS
779{
780 unsigned ncpy = min(*size, cs->len);
781 if (val) {
c55a01d3
MS
782 void *pgaddr = kmap_atomic(cs->pg);
783 void *buf = pgaddr + cs->offset;
784
334f485d 785 if (cs->write)
c55a01d3 786 memcpy(buf, *val, ncpy);
334f485d 787 else
c55a01d3
MS
788 memcpy(*val, buf, ncpy);
789
790 kunmap_atomic(pgaddr);
334f485d
MS
791 *val += ncpy;
792 }
793 *size -= ncpy;
794 cs->len -= ncpy;
c55a01d3 795 cs->offset += ncpy;
334f485d
MS
796 return ncpy;
797}
798
ce534fb0
MS
799static 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
818static 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;
ce534fb0 824
dc00809a 825 err = unlock_request(cs->req);
0d8e84b0
MS
826 if (err)
827 return err;
828
ce534fb0
MS
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
aa991b3b
MS
849 if (!PageUptodate(newpage))
850 SetPageUptodate(newpage);
ce534fb0
MS
851
852 ClearPageMappedToDisk(newpage);
853
854 if (fuse_check_page(newpage) != 0)
855 goto out_fallback_unlock;
856
ce534fb0
MS
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
ef6a3c63 870 err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
ce534fb0 871 if (err) {
ef6a3c63
MS
872 unlock_page(newpage);
873 return err;
ce534fb0 874 }
ef6a3c63 875
09cbfeaf 876 get_page(newpage);
ce534fb0
MS
877
878 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
879 lru_cache_add_file(newpage);
880
881 err = 0;
dc00809a 882 spin_lock(&cs->req->waitq.lock);
825d6d33 883 if (test_bit(FR_ABORTED, &cs->req->flags))
ce534fb0
MS
884 err = -ENOENT;
885 else
886 *pagep = newpage;
dc00809a 887 spin_unlock(&cs->req->waitq.lock);
ce534fb0
MS
888
889 if (err) {
890 unlock_page(newpage);
09cbfeaf 891 put_page(newpage);
ce534fb0
MS
892 return err;
893 }
894
895 unlock_page(oldpage);
09cbfeaf 896 put_page(oldpage);
ce534fb0
MS
897 cs->len = 0;
898
899 return 0;
900
901out_fallback_unlock:
902 unlock_page(newpage);
903out_fallback:
c55a01d3
MS
904 cs->pg = buf->page;
905 cs->offset = buf->offset;
ce534fb0 906
dc00809a 907 err = lock_request(cs->req);
ce534fb0
MS
908 if (err)
909 return err;
910
911 return 1;
912}
913
c3021629
MS
914static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
915 unsigned offset, unsigned count)
916{
917 struct pipe_buffer *buf;
0d8e84b0 918 int err;
c3021629
MS
919
920 if (cs->nr_segs == cs->pipe->buffers)
921 return -EIO;
922
dc00809a 923 err = unlock_request(cs->req);
0d8e84b0
MS
924 if (err)
925 return err;
926
c3021629
MS
927 fuse_copy_finish(cs);
928
929 buf = cs->pipebufs;
09cbfeaf 930 get_page(page);
c3021629
MS
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
334f485d
MS
942/*
943 * Copy a page in the request to/from the userspace buffer. Must be
944 * done atomically
945 */
ce534fb0 946static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
8bfc016d 947 unsigned offset, unsigned count, int zeroing)
334f485d 948{
ce534fb0
MS
949 int err;
950 struct page *page = *pagep;
951
b6777c40
MS
952 if (page && zeroing && count < PAGE_SIZE)
953 clear_highpage(page);
954
334f485d 955 while (count) {
c3021629
MS
956 if (cs->write && cs->pipebufs && page) {
957 return fuse_ref_page(cs, page, offset, count);
958 } else if (!cs->len) {
ce534fb0
MS
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 }
1729a16c 969 }
334f485d 970 if (page) {
2408f6ef 971 void *mapaddr = kmap_atomic(page);
334f485d
MS
972 void *buf = mapaddr + offset;
973 offset += fuse_copy_do(cs, &buf, &count);
2408f6ef 974 kunmap_atomic(mapaddr);
334f485d
MS
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 */
984static 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;
334f485d
MS
989
990 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
ce534fb0 991 int err;
85f40aec
MP
992 unsigned offset = req->page_descs[i].offset;
993 unsigned count = min(nbytes, req->page_descs[i].length);
ce534fb0
MS
994
995 err = fuse_copy_page(cs, &req->pages[i], offset, count,
996 zeroing);
334f485d
MS
997 if (err)
998 return err;
999
1000 nbytes -= count;
334f485d
MS
1001 }
1002 return 0;
1003}
1004
1005/* Copy a single argument in the request to/from userspace buffer */
1006static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
1007{
1008 while (size) {
1729a16c
MS
1009 if (!cs->len) {
1010 int err = fuse_copy_fill(cs);
1011 if (err)
1012 return err;
1013 }
334f485d
MS
1014 fuse_copy_do(cs, &val, &size);
1015 }
1016 return 0;
1017}
1018
1019/* Copy request arguments to/from userspace buffer */
1020static 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
f88996a9 1037static int forget_pending(struct fuse_iqueue *fiq)
07e77dca 1038{
f88996a9 1039 return fiq->forget_list_head.next != NULL;
07e77dca
MS
1040}
1041
f88996a9 1042static int request_pending(struct fuse_iqueue *fiq)
a4d27e75 1043{
f88996a9
MS
1044 return !list_empty(&fiq->pending) || !list_empty(&fiq->interrupts) ||
1045 forget_pending(fiq);
a4d27e75
MS
1046}
1047
a4d27e75
MS
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 *
fd22d62e 1054 * Called with fiq->waitq.lock held, releases it
a4d27e75 1055 */
fd22d62e
MS
1056static int fuse_read_interrupt(struct fuse_iqueue *fiq,
1057 struct fuse_copy_state *cs,
c3021629 1058 size_t nbytes, struct fuse_req *req)
fd22d62e 1059__releases(fiq->waitq.lock)
a4d27e75 1060{
a4d27e75
MS
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);
4ce60812 1067 req->intr_unique = fuse_get_unique(fiq);
a4d27e75
MS
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
4ce60812 1075 spin_unlock(&fiq->waitq.lock);
c3021629 1076 if (nbytes < reqsize)
a4d27e75
MS
1077 return -EINVAL;
1078
c3021629 1079 err = fuse_copy_one(cs, &ih, sizeof(ih));
a4d27e75 1080 if (!err)
c3021629
MS
1081 err = fuse_copy_one(cs, &arg, sizeof(arg));
1082 fuse_copy_finish(cs);
a4d27e75
MS
1083
1084 return err ? err : reqsize;
1085}
1086
f88996a9 1087static struct fuse_forget_link *dequeue_forget(struct fuse_iqueue *fiq,
02c048b9
MS
1088 unsigned max,
1089 unsigned *countp)
07e77dca 1090{
f88996a9 1091 struct fuse_forget_link *head = fiq->forget_list_head.next;
02c048b9
MS
1092 struct fuse_forget_link **newhead = &head;
1093 unsigned count;
07e77dca 1094
02c048b9
MS
1095 for (count = 0; *newhead != NULL && count < max; count++)
1096 newhead = &(*newhead)->next;
1097
f88996a9 1098 fiq->forget_list_head.next = *newhead;
02c048b9 1099 *newhead = NULL;
f88996a9
MS
1100 if (fiq->forget_list_head.next == NULL)
1101 fiq->forget_list_tail = &fiq->forget_list_head;
07e77dca 1102
02c048b9
MS
1103 if (countp != NULL)
1104 *countp = count;
1105
1106 return head;
07e77dca
MS
1107}
1108
fd22d62e 1109static int fuse_read_single_forget(struct fuse_iqueue *fiq,
07e77dca
MS
1110 struct fuse_copy_state *cs,
1111 size_t nbytes)
fd22d62e 1112__releases(fiq->waitq.lock)
07e77dca
MS
1113{
1114 int err;
f88996a9 1115 struct fuse_forget_link *forget = dequeue_forget(fiq, 1, NULL);
07e77dca 1116 struct fuse_forget_in arg = {
02c048b9 1117 .nlookup = forget->forget_one.nlookup,
07e77dca
MS
1118 };
1119 struct fuse_in_header ih = {
1120 .opcode = FUSE_FORGET,
02c048b9 1121 .nodeid = forget->forget_one.nodeid,
f88996a9 1122 .unique = fuse_get_unique(fiq),
07e77dca
MS
1123 .len = sizeof(ih) + sizeof(arg),
1124 };
1125
4ce60812 1126 spin_unlock(&fiq->waitq.lock);
07e77dca
MS
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
fd22d62e 1142static int fuse_read_batch_forget(struct fuse_iqueue *fiq,
02c048b9 1143 struct fuse_copy_state *cs, size_t nbytes)
fd22d62e 1144__releases(fiq->waitq.lock)
02c048b9
MS
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,
f88996a9 1153 .unique = fuse_get_unique(fiq),
02c048b9
MS
1154 .len = sizeof(ih) + sizeof(arg),
1155 };
1156
1157 if (nbytes < ih.len) {
4ce60812 1158 spin_unlock(&fiq->waitq.lock);
02c048b9
MS
1159 return -EINVAL;
1160 }
1161
1162 max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one);
f88996a9 1163 head = dequeue_forget(fiq, max_forgets, &count);
4ce60812 1164 spin_unlock(&fiq->waitq.lock);
02c048b9
MS
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
fd22d62e
MS
1191static int fuse_read_forget(struct fuse_conn *fc, struct fuse_iqueue *fiq,
1192 struct fuse_copy_state *cs,
02c048b9 1193 size_t nbytes)
fd22d62e 1194__releases(fiq->waitq.lock)
02c048b9 1195{
f88996a9 1196 if (fc->minor < 16 || fiq->forget_list_head.next->next == NULL)
fd22d62e 1197 return fuse_read_single_forget(fiq, cs, nbytes);
02c048b9 1198 else
fd22d62e 1199 return fuse_read_batch_forget(fiq, cs, nbytes);
02c048b9
MS
1200}
1201
334f485d
MS
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
f9a2842e
MS
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
334f485d
MS
1208 * request_end(). Otherwise add it to the processing list, and set
1209 * the 'sent' flag.
1210 */
c3696046 1211static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
c3021629 1212 struct fuse_copy_state *cs, size_t nbytes)
334f485d 1213{
82cbdcd3 1214 ssize_t err;
c3696046 1215 struct fuse_conn *fc = fud->fc;
f88996a9 1216 struct fuse_iqueue *fiq = &fc->iq;
c3696046 1217 struct fuse_pqueue *fpq = &fud->pq;
334f485d
MS
1218 struct fuse_req *req;
1219 struct fuse_in *in;
334f485d
MS
1220 unsigned reqsize;
1221
1d3d752b 1222 restart:
4ce60812 1223 spin_lock(&fiq->waitq.lock);
e5ac1d1e 1224 err = -EAGAIN;
e16714d8 1225 if ((file->f_flags & O_NONBLOCK) && fiq->connected &&
f88996a9 1226 !request_pending(fiq))
e5ac1d1e
JD
1227 goto err_unlock;
1228
5250921b
MS
1229 err = wait_event_interruptible_exclusive_locked(fiq->waitq,
1230 !fiq->connected || request_pending(fiq));
1231 if (err)
1232 goto err_unlock;
1233
334f485d 1234 err = -ENODEV;
e16714d8 1235 if (!fiq->connected)
334f485d 1236 goto err_unlock;
334f485d 1237
f88996a9
MS
1238 if (!list_empty(&fiq->interrupts)) {
1239 req = list_entry(fiq->interrupts.next, struct fuse_req,
a4d27e75 1240 intr_entry);
fd22d62e 1241 return fuse_read_interrupt(fiq, cs, nbytes, req);
a4d27e75
MS
1242 }
1243
f88996a9
MS
1244 if (forget_pending(fiq)) {
1245 if (list_empty(&fiq->pending) || fiq->forget_batch-- > 0)
fd22d62e 1246 return fuse_read_forget(fc, fiq, cs, nbytes);
07e77dca 1247
f88996a9
MS
1248 if (fiq->forget_batch <= -8)
1249 fiq->forget_batch = 16;
07e77dca
MS
1250 }
1251
f88996a9 1252 req = list_entry(fiq->pending.next, struct fuse_req, list);
33e14b4d 1253 clear_bit(FR_PENDING, &req->flags);
ef759258 1254 list_del_init(&req->list);
4ce60812
MS
1255 spin_unlock(&fiq->waitq.lock);
1256
334f485d 1257 in = &req->in;
1d3d752b
MS
1258 reqsize = in->h.len;
1259 /* If request is too large, reply with an error and restart the read */
c3021629 1260 if (nbytes < reqsize) {
1d3d752b
MS
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;
334f485d 1267 }
45a91cb1 1268 spin_lock(&fpq->lock);
82cbdcd3 1269 list_add(&req->list, &fpq->io);
45a91cb1 1270 spin_unlock(&fpq->lock);
c3021629
MS
1271 cs->req = req;
1272 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
1d3d752b 1273 if (!err)
c3021629 1274 err = fuse_copy_args(cs, in->numargs, in->argpages,
1d3d752b 1275 (struct fuse_arg *) in->args, 0);
c3021629 1276 fuse_copy_finish(cs);
45a91cb1 1277 spin_lock(&fpq->lock);
825d6d33 1278 clear_bit(FR_LOCKED, &req->flags);
e96edd94 1279 if (!fpq->connected) {
82cbdcd3
MS
1280 err = -ENODEV;
1281 goto out_end;
c9c9d7df 1282 }
334f485d 1283 if (err) {
c9c9d7df 1284 req->out.h.error = -EIO;
82cbdcd3 1285 goto out_end;
334f485d 1286 }
825d6d33 1287 if (!test_bit(FR_ISREPLY, &req->flags)) {
82cbdcd3
MS
1288 err = reqsize;
1289 goto out_end;
334f485d 1290 }
82cbdcd3 1291 list_move_tail(&req->list, &fpq->processing);
45a91cb1 1292 spin_unlock(&fpq->lock);
82cbdcd3
MS
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);
82cbdcd3 1298
334f485d
MS
1299 return reqsize;
1300
82cbdcd3 1301out_end:
77cd9d48
MS
1302 if (!test_bit(FR_PRIVATE, &req->flags))
1303 list_del_init(&req->list);
45a91cb1 1304 spin_unlock(&fpq->lock);
82cbdcd3
MS
1305 request_end(fc, req);
1306 return err;
1307
334f485d 1308 err_unlock:
4ce60812 1309 spin_unlock(&fiq->waitq.lock);
334f485d
MS
1310 return err;
1311}
1312
94e4fe2c
TVB
1313static 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
fbdbacca 1324static ssize_t fuse_dev_read(struct kiocb *iocb, struct iov_iter *to)
c3021629
MS
1325{
1326 struct fuse_copy_state cs;
1327 struct file *file = iocb->ki_filp;
cc080e9e
MS
1328 struct fuse_dev *fud = fuse_get_dev(file);
1329
1330 if (!fud)
c3021629
MS
1331 return -EPERM;
1332
fbdbacca
AV
1333 if (!iter_is_iovec(to))
1334 return -EINVAL;
1335
dc00809a 1336 fuse_copy_init(&cs, 1, to);
c3021629 1337
c3696046 1338 return fuse_dev_do_read(fud, file, &cs, iov_iter_count(to));
c3021629
MS
1339}
1340
c3021629
MS
1341static 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;
cc080e9e
MS
1350 struct fuse_dev *fud = fuse_get_dev(in);
1351
1352 if (!fud)
c3021629
MS
1353 return -EPERM;
1354
07e77dca 1355 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
c3021629
MS
1356 if (!bufs)
1357 return -ENOMEM;
1358
dc00809a 1359 fuse_copy_init(&cs, 1, NULL);
c3021629
MS
1360 cs.pipebufs = bufs;
1361 cs.pipe = pipe;
c3696046 1362 ret = fuse_dev_do_read(fud, in, &cs, len);
c3021629
MS
1363 if (ret < 0)
1364 goto out;
1365
1366 ret = 0;
c3021629
MS
1367
1368 if (!pipe->readers) {
1369 send_sig(SIGPIPE, current, 0);
1370 if (!ret)
1371 ret = -EPIPE;
1372 goto out_unlock;
1373 }
1374
1375 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1376 ret = -EIO;
1377 goto out_unlock;
1378 }
1379
1380 while (page_nr < cs.nr_segs) {
1381 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1382 struct pipe_buffer *buf = pipe->bufs + newbuf;
1383
1384 buf->page = bufs[page_nr].page;
1385 buf->offset = bufs[page_nr].offset;
1386 buf->len = bufs[page_nr].len;
28a625cb
MS
1387 /*
1388 * Need to be careful about this. Having buf->ops in module
1389 * code can Oops if the buffer persists after module unload.
1390 */
1391 buf->ops = &nosteal_pipe_buf_ops;
c3021629
MS
1392
1393 pipe->nrbufs++;
1394 page_nr++;
1395 ret += buf->len;
1396
6447a3cf 1397 if (pipe->files)
c3021629
MS
1398 do_wakeup = 1;
1399 }
1400
1401out_unlock:
c3021629
MS
1402
1403 if (do_wakeup) {
1404 smp_mb();
1405 if (waitqueue_active(&pipe->wait))
1406 wake_up_interruptible(&pipe->wait);
1407 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1408 }
1409
1410out:
1411 for (; page_nr < cs.nr_segs; page_nr++)
09cbfeaf 1412 put_page(bufs[page_nr].page);
c3021629
MS
1413
1414 kfree(bufs);
1415 return ret;
1416}
1417
95668a69
TH
1418static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1419 struct fuse_copy_state *cs)
1420{
1421 struct fuse_notify_poll_wakeup_out outarg;
f6d47a17 1422 int err = -EINVAL;
95668a69
TH
1423
1424 if (size != sizeof(outarg))
f6d47a17 1425 goto err;
95668a69
TH
1426
1427 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1428 if (err)
f6d47a17 1429 goto err;
95668a69 1430
f6d47a17 1431 fuse_copy_finish(cs);
95668a69 1432 return fuse_notify_poll_wakeup(fc, &outarg);
f6d47a17
MS
1433
1434err:
1435 fuse_copy_finish(cs);
1436 return err;
95668a69
TH
1437}
1438
3b463ae0
JM
1439static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1440 struct fuse_copy_state *cs)
1441{
1442 struct fuse_notify_inval_inode_out outarg;
1443 int err = -EINVAL;
1444
1445 if (size != sizeof(outarg))
1446 goto err;
1447
1448 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1449 if (err)
1450 goto err;
1451 fuse_copy_finish(cs);
1452
1453 down_read(&fc->killsb);
1454 err = -ENOENT;
b21dda43
MS
1455 if (fc->sb) {
1456 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1457 outarg.off, outarg.len);
1458 }
3b463ae0
JM
1459 up_read(&fc->killsb);
1460 return err;
1461
1462err:
1463 fuse_copy_finish(cs);
1464 return err;
1465}
1466
1467static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1468 struct fuse_copy_state *cs)
1469{
1470 struct fuse_notify_inval_entry_out outarg;
b2d82ee3
FW
1471 int err = -ENOMEM;
1472 char *buf;
3b463ae0
JM
1473 struct qstr name;
1474
b2d82ee3
FW
1475 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1476 if (!buf)
1477 goto err;
1478
1479 err = -EINVAL;
3b463ae0
JM
1480 if (size < sizeof(outarg))
1481 goto err;
1482
1483 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1484 if (err)
1485 goto err;
1486
1487 err = -ENAMETOOLONG;
1488 if (outarg.namelen > FUSE_NAME_MAX)
1489 goto err;
1490
c2183d1e
MS
1491 err = -EINVAL;
1492 if (size != sizeof(outarg) + outarg.namelen + 1)
1493 goto err;
1494
3b463ae0
JM
1495 name.name = buf;
1496 name.len = outarg.namelen;
1497 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1498 if (err)
1499 goto err;
1500 fuse_copy_finish(cs);
1501 buf[outarg.namelen] = 0;
3b463ae0
JM
1502
1503 down_read(&fc->killsb);
1504 err = -ENOENT;
b21dda43 1505 if (fc->sb)
451d0f59
JM
1506 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name);
1507 up_read(&fc->killsb);
1508 kfree(buf);
1509 return err;
1510
1511err:
1512 kfree(buf);
1513 fuse_copy_finish(cs);
1514 return err;
1515}
1516
1517static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
1518 struct fuse_copy_state *cs)
1519{
1520 struct fuse_notify_delete_out outarg;
1521 int err = -ENOMEM;
1522 char *buf;
1523 struct qstr name;
1524
1525 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1526 if (!buf)
1527 goto err;
1528
1529 err = -EINVAL;
1530 if (size < sizeof(outarg))
1531 goto err;
1532
1533 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1534 if (err)
1535 goto err;
1536
1537 err = -ENAMETOOLONG;
1538 if (outarg.namelen > FUSE_NAME_MAX)
1539 goto err;
1540
1541 err = -EINVAL;
1542 if (size != sizeof(outarg) + outarg.namelen + 1)
1543 goto err;
1544
1545 name.name = buf;
1546 name.len = outarg.namelen;
1547 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1548 if (err)
1549 goto err;
1550 fuse_copy_finish(cs);
1551 buf[outarg.namelen] = 0;
451d0f59
JM
1552
1553 down_read(&fc->killsb);
1554 err = -ENOENT;
1555 if (fc->sb)
1556 err = fuse_reverse_inval_entry(fc->sb, outarg.parent,
1557 outarg.child, &name);
3b463ae0 1558 up_read(&fc->killsb);
b2d82ee3 1559 kfree(buf);
3b463ae0
JM
1560 return err;
1561
1562err:
b2d82ee3 1563 kfree(buf);
3b463ae0
JM
1564 fuse_copy_finish(cs);
1565 return err;
1566}
1567
a1d75f25
MS
1568static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1569 struct fuse_copy_state *cs)
1570{
1571 struct fuse_notify_store_out outarg;
1572 struct inode *inode;
1573 struct address_space *mapping;
1574 u64 nodeid;
1575 int err;
1576 pgoff_t index;
1577 unsigned int offset;
1578 unsigned int num;
1579 loff_t file_size;
1580 loff_t end;
1581
1582 err = -EINVAL;
1583 if (size < sizeof(outarg))
1584 goto out_finish;
1585
1586 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1587 if (err)
1588 goto out_finish;
1589
1590 err = -EINVAL;
1591 if (size - sizeof(outarg) != outarg.size)
1592 goto out_finish;
1593
1594 nodeid = outarg.nodeid;
1595
1596 down_read(&fc->killsb);
1597
1598 err = -ENOENT;
1599 if (!fc->sb)
1600 goto out_up_killsb;
1601
1602 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1603 if (!inode)
1604 goto out_up_killsb;
1605
1606 mapping = inode->i_mapping;
09cbfeaf
KS
1607 index = outarg.offset >> PAGE_SHIFT;
1608 offset = outarg.offset & ~PAGE_MASK;
a1d75f25
MS
1609 file_size = i_size_read(inode);
1610 end = outarg.offset + outarg.size;
1611 if (end > file_size) {
1612 file_size = end;
1613 fuse_write_update_size(inode, file_size);
1614 }
1615
1616 num = outarg.size;
1617 while (num) {
1618 struct page *page;
1619 unsigned int this_num;
1620
1621 err = -ENOMEM;
1622 page = find_or_create_page(mapping, index,
1623 mapping_gfp_mask(mapping));
1624 if (!page)
1625 goto out_iput;
1626
09cbfeaf 1627 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
a1d75f25 1628 err = fuse_copy_page(cs, &page, offset, this_num, 0);
063ec1e5 1629 if (!err && offset == 0 &&
09cbfeaf 1630 (this_num == PAGE_SIZE || file_size == end))
a1d75f25
MS
1631 SetPageUptodate(page);
1632 unlock_page(page);
09cbfeaf 1633 put_page(page);
a1d75f25
MS
1634
1635 if (err)
1636 goto out_iput;
1637
1638 num -= this_num;
1639 offset = 0;
1640 index++;
1641 }
1642
1643 err = 0;
1644
1645out_iput:
1646 iput(inode);
1647out_up_killsb:
1648 up_read(&fc->killsb);
1649out_finish:
1650 fuse_copy_finish(cs);
1651 return err;
1652}
1653
2d45ba38
MS
1654static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1655{
b745bc85 1656 release_pages(req->pages, req->num_pages, false);
2d45ba38
MS
1657}
1658
1659static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1660 struct fuse_notify_retrieve_out *outarg)
1661{
1662 int err;
1663 struct address_space *mapping = inode->i_mapping;
1664 struct fuse_req *req;
1665 pgoff_t index;
1666 loff_t file_size;
1667 unsigned int num;
1668 unsigned int offset;
0157443c 1669 size_t total_len = 0;
4d53dc99 1670 int num_pages;
2d45ba38 1671
09cbfeaf 1672 offset = outarg->offset & ~PAGE_MASK;
4d53dc99
MP
1673 file_size = i_size_read(inode);
1674
1675 num = outarg->size;
1676 if (outarg->offset > file_size)
1677 num = 0;
1678 else if (outarg->offset + num > file_size)
1679 num = file_size - outarg->offset;
1680
1681 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1682 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1683
1684 req = fuse_get_req(fc, num_pages);
2d45ba38
MS
1685 if (IS_ERR(req))
1686 return PTR_ERR(req);
1687
2d45ba38
MS
1688 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1689 req->in.h.nodeid = outarg->nodeid;
1690 req->in.numargs = 2;
1691 req->in.argpages = 1;
b2430d75 1692 req->page_descs[0].offset = offset;
2d45ba38
MS
1693 req->end = fuse_retrieve_end;
1694
09cbfeaf 1695 index = outarg->offset >> PAGE_SHIFT;
2d45ba38 1696
4d53dc99 1697 while (num && req->num_pages < num_pages) {
2d45ba38
MS
1698 struct page *page;
1699 unsigned int this_num;
1700
1701 page = find_get_page(mapping, index);
1702 if (!page)
1703 break;
1704
09cbfeaf 1705 this_num = min_t(unsigned, num, PAGE_SIZE - offset);
2d45ba38 1706 req->pages[req->num_pages] = page;
85f40aec 1707 req->page_descs[req->num_pages].length = this_num;
2d45ba38
MS
1708 req->num_pages++;
1709
c9e67d48 1710 offset = 0;
2d45ba38
MS
1711 num -= this_num;
1712 total_len += this_num;
48706d0a 1713 index++;
2d45ba38
MS
1714 }
1715 req->misc.retrieve_in.offset = outarg->offset;
1716 req->misc.retrieve_in.size = total_len;
1717 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1718 req->in.args[0].value = &req->misc.retrieve_in;
1719 req->in.args[1].size = total_len;
1720
1721 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1722 if (err)
1723 fuse_retrieve_end(fc, req);
1724
1725 return err;
1726}
1727
1728static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1729 struct fuse_copy_state *cs)
1730{
1731 struct fuse_notify_retrieve_out outarg;
1732 struct inode *inode;
1733 int err;
1734
1735 err = -EINVAL;
1736 if (size != sizeof(outarg))
1737 goto copy_finish;
1738
1739 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1740 if (err)
1741 goto copy_finish;
1742
1743 fuse_copy_finish(cs);
1744
1745 down_read(&fc->killsb);
1746 err = -ENOENT;
1747 if (fc->sb) {
1748 u64 nodeid = outarg.nodeid;
1749
1750 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1751 if (inode) {
1752 err = fuse_retrieve(fc, inode, &outarg);
1753 iput(inode);
1754 }
1755 }
1756 up_read(&fc->killsb);
1757
1758 return err;
1759
1760copy_finish:
1761 fuse_copy_finish(cs);
1762 return err;
1763}
1764
8599396b
TH
1765static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1766 unsigned int size, struct fuse_copy_state *cs)
1767{
0d278362
MS
1768 /* Don't try to move pages (yet) */
1769 cs->move_pages = 0;
1770
8599396b 1771 switch (code) {
95668a69
TH
1772 case FUSE_NOTIFY_POLL:
1773 return fuse_notify_poll(fc, size, cs);
1774
3b463ae0
JM
1775 case FUSE_NOTIFY_INVAL_INODE:
1776 return fuse_notify_inval_inode(fc, size, cs);
1777
1778 case FUSE_NOTIFY_INVAL_ENTRY:
1779 return fuse_notify_inval_entry(fc, size, cs);
1780
a1d75f25
MS
1781 case FUSE_NOTIFY_STORE:
1782 return fuse_notify_store(fc, size, cs);
1783
2d45ba38
MS
1784 case FUSE_NOTIFY_RETRIEVE:
1785 return fuse_notify_retrieve(fc, size, cs);
1786
451d0f59
JM
1787 case FUSE_NOTIFY_DELETE:
1788 return fuse_notify_delete(fc, size, cs);
1789
8599396b 1790 default:
f6d47a17 1791 fuse_copy_finish(cs);
8599396b
TH
1792 return -EINVAL;
1793 }
1794}
1795
334f485d 1796/* Look up request on processing list by unique ID */
3a2b5b9c 1797static struct fuse_req *request_find(struct fuse_pqueue *fpq, u64 unique)
334f485d 1798{
05726aca 1799 struct fuse_req *req;
334f485d 1800
3a2b5b9c 1801 list_for_each_entry(req, &fpq->processing, list) {
a4d27e75 1802 if (req->in.h.unique == unique || req->intr_unique == unique)
334f485d
MS
1803 return req;
1804 }
1805 return NULL;
1806}
1807
1808static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1809 unsigned nbytes)
1810{
1811 unsigned reqsize = sizeof(struct fuse_out_header);
1812
1813 if (out->h.error)
1814 return nbytes != reqsize ? -EINVAL : 0;
1815
1816 reqsize += len_args(out->numargs, out->args);
1817
1818 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1819 return -EINVAL;
1820 else if (reqsize > nbytes) {
1821 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1822 unsigned diffsize = reqsize - nbytes;
1823 if (diffsize > lastarg->size)
1824 return -EINVAL;
1825 lastarg->size -= diffsize;
1826 }
1827 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1828 out->page_zeroing);
1829}
1830
1831/*
1832 * Write a single reply to a request. First the header is copied from
1833 * the write buffer. The request is then searched on the processing
1834 * list by the unique ID found in the header. If found, then remove
1835 * it from the list and copy the rest of the buffer to the request.
1836 * The request is finished by calling request_end()
1837 */
c3696046 1838static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
dd3bb14f 1839 struct fuse_copy_state *cs, size_t nbytes)
334f485d
MS
1840{
1841 int err;
c3696046
MS
1842 struct fuse_conn *fc = fud->fc;
1843 struct fuse_pqueue *fpq = &fud->pq;
334f485d
MS
1844 struct fuse_req *req;
1845 struct fuse_out_header oh;
334f485d 1846
334f485d
MS
1847 if (nbytes < sizeof(struct fuse_out_header))
1848 return -EINVAL;
1849
dd3bb14f 1850 err = fuse_copy_one(cs, &oh, sizeof(oh));
334f485d
MS
1851 if (err)
1852 goto err_finish;
8599396b
TH
1853
1854 err = -EINVAL;
1855 if (oh.len != nbytes)
1856 goto err_finish;
1857
1858 /*
1859 * Zero oh.unique indicates unsolicited notification message
1860 * and error contains notification code.
1861 */
1862 if (!oh.unique) {
dd3bb14f 1863 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
8599396b
TH
1864 return err ? err : nbytes;
1865 }
1866
334f485d 1867 err = -EINVAL;
8599396b 1868 if (oh.error <= -1000 || oh.error > 0)
334f485d
MS
1869 goto err_finish;
1870
45a91cb1 1871 spin_lock(&fpq->lock);
69a53bf2 1872 err = -ENOENT;
e96edd94 1873 if (!fpq->connected)
45a91cb1 1874 goto err_unlock_pq;
69a53bf2 1875
3a2b5b9c 1876 req = request_find(fpq, oh.unique);
334f485d 1877 if (!req)
45a91cb1 1878 goto err_unlock_pq;
334f485d 1879
a4d27e75
MS
1880 /* Is it an interrupt reply? */
1881 if (req->intr_unique == oh.unique) {
45a91cb1
MS
1882 spin_unlock(&fpq->lock);
1883
a4d27e75
MS
1884 err = -EINVAL;
1885 if (nbytes != sizeof(struct fuse_out_header))
46c34a34 1886 goto err_finish;
a4d27e75
MS
1887
1888 if (oh.error == -ENOSYS)
1889 fc->no_interrupt = 1;
1890 else if (oh.error == -EAGAIN)
f88996a9 1891 queue_interrupt(&fc->iq, req);
a4d27e75 1892
dd3bb14f 1893 fuse_copy_finish(cs);
a4d27e75
MS
1894 return nbytes;
1895 }
1896
33e14b4d 1897 clear_bit(FR_SENT, &req->flags);
3a2b5b9c 1898 list_move(&req->list, &fpq->io);
334f485d 1899 req->out.h = oh;
825d6d33 1900 set_bit(FR_LOCKED, &req->flags);
45a91cb1 1901 spin_unlock(&fpq->lock);
dd3bb14f 1902 cs->req = req;
ce534fb0
MS
1903 if (!req->out.page_replace)
1904 cs->move_pages = 0;
334f485d 1905
dd3bb14f
MS
1906 err = copy_out_args(cs, &req->out, nbytes);
1907 fuse_copy_finish(cs);
334f485d 1908
45a91cb1 1909 spin_lock(&fpq->lock);
825d6d33 1910 clear_bit(FR_LOCKED, &req->flags);
e96edd94 1911 if (!fpq->connected)
0d8e84b0
MS
1912 err = -ENOENT;
1913 else if (err)
334f485d 1914 req->out.h.error = -EIO;
77cd9d48
MS
1915 if (!test_bit(FR_PRIVATE, &req->flags))
1916 list_del_init(&req->list);
45a91cb1 1917 spin_unlock(&fpq->lock);
46c34a34 1918
334f485d
MS
1919 request_end(fc, req);
1920
1921 return err ? err : nbytes;
1922
45a91cb1
MS
1923 err_unlock_pq:
1924 spin_unlock(&fpq->lock);
334f485d 1925 err_finish:
dd3bb14f 1926 fuse_copy_finish(cs);
334f485d
MS
1927 return err;
1928}
1929
fbdbacca 1930static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
dd3bb14f
MS
1931{
1932 struct fuse_copy_state cs;
cc080e9e
MS
1933 struct fuse_dev *fud = fuse_get_dev(iocb->ki_filp);
1934
1935 if (!fud)
dd3bb14f
MS
1936 return -EPERM;
1937
fbdbacca
AV
1938 if (!iter_is_iovec(from))
1939 return -EINVAL;
1940
dc00809a 1941 fuse_copy_init(&cs, 0, from);
dd3bb14f 1942
c3696046 1943 return fuse_dev_do_write(fud, &cs, iov_iter_count(from));
dd3bb14f
MS
1944}
1945
1946static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1947 struct file *out, loff_t *ppos,
1948 size_t len, unsigned int flags)
1949{
1950 unsigned nbuf;
1951 unsigned idx;
1952 struct pipe_buffer *bufs;
1953 struct fuse_copy_state cs;
cc080e9e 1954 struct fuse_dev *fud;
dd3bb14f
MS
1955 size_t rem;
1956 ssize_t ret;
1957
cc080e9e
MS
1958 fud = fuse_get_dev(out);
1959 if (!fud)
dd3bb14f
MS
1960 return -EPERM;
1961
07e77dca 1962 bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL);
dd3bb14f
MS
1963 if (!bufs)
1964 return -ENOMEM;
1965
1966 pipe_lock(pipe);
1967 nbuf = 0;
1968 rem = 0;
1969 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1970 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1971
1972 ret = -EINVAL;
1973 if (rem < len) {
1974 pipe_unlock(pipe);
1975 goto out;
1976 }
1977
1978 rem = len;
1979 while (rem) {
1980 struct pipe_buffer *ibuf;
1981 struct pipe_buffer *obuf;
1982
1983 BUG_ON(nbuf >= pipe->buffers);
1984 BUG_ON(!pipe->nrbufs);
1985 ibuf = &pipe->bufs[pipe->curbuf];
1986 obuf = &bufs[nbuf];
1987
1988 if (rem >= ibuf->len) {
1989 *obuf = *ibuf;
1990 ibuf->ops = NULL;
1991 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1992 pipe->nrbufs--;
1993 } else {
1994 ibuf->ops->get(pipe, ibuf);
1995 *obuf = *ibuf;
1996 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1997 obuf->len = rem;
1998 ibuf->offset += obuf->len;
1999 ibuf->len -= obuf->len;
2000 }
2001 nbuf++;
2002 rem -= obuf->len;
2003 }
2004 pipe_unlock(pipe);
2005
dc00809a 2006 fuse_copy_init(&cs, 0, NULL);
dd3bb14f 2007 cs.pipebufs = bufs;
6c09e94a 2008 cs.nr_segs = nbuf;
dd3bb14f
MS
2009 cs.pipe = pipe;
2010
ce534fb0
MS
2011 if (flags & SPLICE_F_MOVE)
2012 cs.move_pages = 1;
2013
c3696046 2014 ret = fuse_dev_do_write(fud, &cs, len);
dd3bb14f
MS
2015
2016 for (idx = 0; idx < nbuf; idx++) {
2017 struct pipe_buffer *buf = &bufs[idx];
2018 buf->ops->release(pipe, buf);
2019 }
2020out:
2021 kfree(bufs);
2022 return ret;
2023}
2024
334f485d
MS
2025static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
2026{
334f485d 2027 unsigned mask = POLLOUT | POLLWRNORM;
f88996a9 2028 struct fuse_iqueue *fiq;
cc080e9e
MS
2029 struct fuse_dev *fud = fuse_get_dev(file);
2030
2031 if (!fud)
7025d9ad 2032 return POLLERR;
334f485d 2033
cc080e9e 2034 fiq = &fud->fc->iq;
f88996a9 2035 poll_wait(file, &fiq->waitq, wait);
334f485d 2036
4ce60812 2037 spin_lock(&fiq->waitq.lock);
e16714d8 2038 if (!fiq->connected)
7025d9ad 2039 mask = POLLERR;
f88996a9 2040 else if (request_pending(fiq))
7025d9ad 2041 mask |= POLLIN | POLLRDNORM;
4ce60812 2042 spin_unlock(&fiq->waitq.lock);
334f485d
MS
2043
2044 return mask;
2045}
2046
69a53bf2
MS
2047/*
2048 * Abort all requests on the given list (pending or processing)
2049 *
d7133114 2050 * This function releases and reacquires fc->lock
69a53bf2 2051 */
334f485d
MS
2052static void end_requests(struct fuse_conn *fc, struct list_head *head)
2053{
2054 while (!list_empty(head)) {
2055 struct fuse_req *req;
2056 req = list_entry(head->next, struct fuse_req, list);
334f485d 2057 req->out.h.error = -ECONNABORTED;
33e14b4d
MS
2058 clear_bit(FR_PENDING, &req->flags);
2059 clear_bit(FR_SENT, &req->flags);
f377cb79 2060 list_del_init(&req->list);
334f485d 2061 request_end(fc, req);
334f485d
MS
2062 }
2063}
2064
357ccf2b
BG
2065static void end_polls(struct fuse_conn *fc)
2066{
2067 struct rb_node *p;
2068
2069 p = rb_first(&fc->polled_files);
2070
2071 while (p) {
2072 struct fuse_file *ff;
2073 ff = rb_entry(p, struct fuse_file, polled_node);
2074 wake_up_interruptible_all(&ff->poll_wait);
2075
2076 p = rb_next(p);
2077 }
2078}
2079
69a53bf2
MS
2080/*
2081 * Abort all requests.
2082 *
b716d425
MS
2083 * Emergency exit in case of a malicious or accidental deadlock, or just a hung
2084 * filesystem.
2085 *
2086 * The same effect is usually achievable through killing the filesystem daemon
2087 * and all users of the filesystem. The exception is the combination of an
2088 * asynchronous request and the tricky deadlock (see
2089 * Documentation/filesystems/fuse.txt).
69a53bf2 2090 *
b716d425
MS
2091 * Aborting requests under I/O goes as follows: 1: Separate out unlocked
2092 * requests, they should be finished off immediately. Locked requests will be
2093 * finished after unlock; see unlock_request(). 2: Finish off the unlocked
2094 * requests. It is possible that some request will finish before we can. This
2095 * is OK, the request will in that case be removed from the list before we touch
2096 * it.
69a53bf2
MS
2097 */
2098void fuse_abort_conn(struct fuse_conn *fc)
2099{
f88996a9
MS
2100 struct fuse_iqueue *fiq = &fc->iq;
2101
d7133114 2102 spin_lock(&fc->lock);
69a53bf2 2103 if (fc->connected) {
c3696046 2104 struct fuse_dev *fud;
b716d425 2105 struct fuse_req *req, *next;
41f98274
MS
2106 LIST_HEAD(to_end1);
2107 LIST_HEAD(to_end2);
b716d425 2108
69a53bf2 2109 fc->connected = 0;
51eb01e7 2110 fc->blocked = 0;
9759bd51 2111 fuse_set_initialized(fc);
c3696046
MS
2112 list_for_each_entry(fud, &fc->devices, entry) {
2113 struct fuse_pqueue *fpq = &fud->pq;
2114
2115 spin_lock(&fpq->lock);
2116 fpq->connected = 0;
2117 list_for_each_entry_safe(req, next, &fpq->io, list) {
2118 req->out.h.error = -ECONNABORTED;
2119 spin_lock(&req->waitq.lock);
2120 set_bit(FR_ABORTED, &req->flags);
2121 if (!test_bit(FR_LOCKED, &req->flags)) {
2122 set_bit(FR_PRIVATE, &req->flags);
2123 list_move(&req->list, &to_end1);
2124 }
2125 spin_unlock(&req->waitq.lock);
77cd9d48 2126 }
c3696046
MS
2127 list_splice_init(&fpq->processing, &to_end2);
2128 spin_unlock(&fpq->lock);
b716d425 2129 }
41f98274
MS
2130 fc->max_background = UINT_MAX;
2131 flush_bg_queue(fc);
8c91189a 2132
4ce60812 2133 spin_lock(&fiq->waitq.lock);
8c91189a 2134 fiq->connected = 0;
f88996a9 2135 list_splice_init(&fiq->pending, &to_end2);
8c91189a
MS
2136 while (forget_pending(fiq))
2137 kfree(dequeue_forget(fiq, 1, NULL));
4ce60812
MS
2138 wake_up_all_locked(&fiq->waitq);
2139 spin_unlock(&fiq->waitq.lock);
8c91189a 2140 kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
ee314a87
MS
2141 end_polls(fc);
2142 wake_up_all(&fc->blocked_waitq);
2143 spin_unlock(&fc->lock);
8c91189a 2144
41f98274
MS
2145 while (!list_empty(&to_end1)) {
2146 req = list_first_entry(&to_end1, struct fuse_req, list);
b716d425 2147 __fuse_get_request(req);
f377cb79 2148 list_del_init(&req->list);
b716d425 2149 request_end(fc, req);
b716d425 2150 }
41f98274 2151 end_requests(fc, &to_end2);
ee314a87
MS
2152 } else {
2153 spin_unlock(&fc->lock);
69a53bf2 2154 }
69a53bf2 2155}
08cbf542 2156EXPORT_SYMBOL_GPL(fuse_abort_conn);
69a53bf2 2157
08cbf542 2158int fuse_dev_release(struct inode *inode, struct file *file)
334f485d 2159{
cc080e9e
MS
2160 struct fuse_dev *fud = fuse_get_dev(file);
2161
2162 if (fud) {
2163 struct fuse_conn *fc = fud->fc;
c3696046
MS
2164 struct fuse_pqueue *fpq = &fud->pq;
2165
2166 WARN_ON(!list_empty(&fpq->io));
2167 end_requests(fc, &fpq->processing);
2168 /* Are we the last open device? */
2169 if (atomic_dec_and_test(&fc->dev_count)) {
2170 WARN_ON(fc->iq.fasync != NULL);
2171 fuse_abort_conn(fc);
2172 }
cc080e9e 2173 fuse_dev_free(fud);
385a17bf 2174 }
334f485d
MS
2175 return 0;
2176}
08cbf542 2177EXPORT_SYMBOL_GPL(fuse_dev_release);
334f485d 2178
385a17bf
JD
2179static int fuse_dev_fasync(int fd, struct file *file, int on)
2180{
cc080e9e
MS
2181 struct fuse_dev *fud = fuse_get_dev(file);
2182
2183 if (!fud)
a87046d8 2184 return -EPERM;
385a17bf
JD
2185
2186 /* No locking - fasync_helper does its own locking */
cc080e9e 2187 return fasync_helper(fd, file, on, &fud->fc->iq.fasync);
385a17bf
JD
2188}
2189
00c570f4
MS
2190static int fuse_device_clone(struct fuse_conn *fc, struct file *new)
2191{
cc080e9e
MS
2192 struct fuse_dev *fud;
2193
00c570f4
MS
2194 if (new->private_data)
2195 return -EINVAL;
2196
cc080e9e
MS
2197 fud = fuse_dev_alloc(fc);
2198 if (!fud)
2199 return -ENOMEM;
2200
2201 new->private_data = fud;
c3696046 2202 atomic_inc(&fc->dev_count);
00c570f4
MS
2203
2204 return 0;
2205}
2206
2207static long fuse_dev_ioctl(struct file *file, unsigned int cmd,
2208 unsigned long arg)
2209{
2210 int err = -ENOTTY;
2211
2212 if (cmd == FUSE_DEV_IOC_CLONE) {
2213 int oldfd;
2214
2215 err = -EFAULT;
2216 if (!get_user(oldfd, (__u32 __user *) arg)) {
2217 struct file *old = fget(oldfd);
2218
2219 err = -EINVAL;
2220 if (old) {
8ed1f0e2
JH
2221 struct fuse_dev *fud = NULL;
2222
2223 /*
2224 * Check against file->f_op because CUSE
2225 * uses the same ioctl handler.
2226 */
2227 if (old->f_op == file->f_op &&
2228 old->f_cred->user_ns == file->f_cred->user_ns)
2229 fud = fuse_get_dev(old);
00c570f4 2230
cc080e9e 2231 if (fud) {
00c570f4 2232 mutex_lock(&fuse_mutex);
cc080e9e 2233 err = fuse_device_clone(fud->fc, file);
00c570f4
MS
2234 mutex_unlock(&fuse_mutex);
2235 }
2236 fput(old);
2237 }
2238 }
2239 }
2240 return err;
2241}
2242
4b6f5d20 2243const struct file_operations fuse_dev_operations = {
334f485d 2244 .owner = THIS_MODULE,
94e4fe2c 2245 .open = fuse_dev_open,
334f485d 2246 .llseek = no_llseek,
fbdbacca 2247 .read_iter = fuse_dev_read,
c3021629 2248 .splice_read = fuse_dev_splice_read,
fbdbacca 2249 .write_iter = fuse_dev_write,
dd3bb14f 2250 .splice_write = fuse_dev_splice_write,
334f485d
MS
2251 .poll = fuse_dev_poll,
2252 .release = fuse_dev_release,
385a17bf 2253 .fasync = fuse_dev_fasync,
00c570f4
MS
2254 .unlocked_ioctl = fuse_dev_ioctl,
2255 .compat_ioctl = fuse_dev_ioctl,
334f485d 2256};
08cbf542 2257EXPORT_SYMBOL_GPL(fuse_dev_operations);
334f485d
MS
2258
2259static struct miscdevice fuse_miscdevice = {
2260 .minor = FUSE_MINOR,
2261 .name = "fuse",
2262 .fops = &fuse_dev_operations,
2263};
2264
2265int __init fuse_dev_init(void)
2266{
2267 int err = -ENOMEM;
2268 fuse_req_cachep = kmem_cache_create("fuse_request",
2269 sizeof(struct fuse_req),
20c2df83 2270 0, 0, NULL);
334f485d
MS
2271 if (!fuse_req_cachep)
2272 goto out;
2273
2274 err = misc_register(&fuse_miscdevice);
2275 if (err)
2276 goto out_cache_clean;
2277
2278 return 0;
2279
2280 out_cache_clean:
2281 kmem_cache_destroy(fuse_req_cachep);
2282 out:
2283 return err;
2284}
2285
2286void fuse_dev_cleanup(void)
2287{
2288 misc_deregister(&fuse_miscdevice);
2289 kmem_cache_destroy(fuse_req_cachep);
2290}