Commit | Line | Data |
---|---|---|
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 | |
23 | MODULE_ALIAS_MISCDEV(FUSE_MINOR); | |
578454ff | 24 | MODULE_ALIAS("devname:fuse"); |
334f485d | 25 | |
e18b890b | 26 | static struct kmem_cache *fuse_req_cachep; |
334f485d | 27 | |
8bfc016d | 28 | static struct fuse_conn *fuse_get_conn(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 | */ | |
34 | return file->private_data; | |
334f485d MS |
35 | } |
36 | ||
8bfc016d | 37 | static void fuse_request_init(struct fuse_req *req) |
334f485d MS |
38 | { |
39 | memset(req, 0, sizeof(*req)); | |
40 | INIT_LIST_HEAD(&req->list); | |
a4d27e75 | 41 | INIT_LIST_HEAD(&req->intr_entry); |
334f485d MS |
42 | init_waitqueue_head(&req->waitq); |
43 | atomic_set(&req->count, 1); | |
44 | } | |
45 | ||
46 | struct fuse_req *fuse_request_alloc(void) | |
47 | { | |
e94b1766 | 48 | struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL); |
334f485d MS |
49 | if (req) |
50 | fuse_request_init(req); | |
51 | return req; | |
52 | } | |
08cbf542 | 53 | EXPORT_SYMBOL_GPL(fuse_request_alloc); |
334f485d | 54 | |
3be5a52b MS |
55 | struct fuse_req *fuse_request_alloc_nofs(void) |
56 | { | |
57 | struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_NOFS); | |
58 | if (req) | |
59 | fuse_request_init(req); | |
60 | return req; | |
61 | } | |
62 | ||
334f485d MS |
63 | void fuse_request_free(struct fuse_req *req) |
64 | { | |
65 | kmem_cache_free(fuse_req_cachep, req); | |
66 | } | |
67 | ||
8bfc016d | 68 | static void block_sigs(sigset_t *oldset) |
334f485d MS |
69 | { |
70 | sigset_t mask; | |
71 | ||
72 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
73 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
74 | } | |
75 | ||
8bfc016d | 76 | static void restore_sigs(sigset_t *oldset) |
334f485d MS |
77 | { |
78 | sigprocmask(SIG_SETMASK, oldset, NULL); | |
79 | } | |
80 | ||
334f485d MS |
81 | static void __fuse_get_request(struct fuse_req *req) |
82 | { | |
83 | atomic_inc(&req->count); | |
84 | } | |
85 | ||
86 | /* Must be called with > 1 refcount */ | |
87 | static void __fuse_put_request(struct fuse_req *req) | |
88 | { | |
89 | BUG_ON(atomic_read(&req->count) < 2); | |
90 | atomic_dec(&req->count); | |
91 | } | |
92 | ||
33649c91 MS |
93 | static void fuse_req_init_context(struct fuse_req *req) |
94 | { | |
2186a71c DH |
95 | req->in.h.uid = current_fsuid(); |
96 | req->in.h.gid = current_fsgid(); | |
33649c91 MS |
97 | req->in.h.pid = current->pid; |
98 | } | |
99 | ||
ce1d5a49 | 100 | struct fuse_req *fuse_get_req(struct fuse_conn *fc) |
334f485d | 101 | { |
08a53cdc MS |
102 | struct fuse_req *req; |
103 | sigset_t oldset; | |
9bc5ddda | 104 | int intr; |
08a53cdc MS |
105 | int err; |
106 | ||
9bc5ddda | 107 | atomic_inc(&fc->num_waiting); |
08a53cdc | 108 | block_sigs(&oldset); |
9bc5ddda | 109 | intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked); |
08a53cdc | 110 | restore_sigs(&oldset); |
9bc5ddda MS |
111 | err = -EINTR; |
112 | if (intr) | |
113 | goto out; | |
08a53cdc | 114 | |
51eb01e7 MS |
115 | err = -ENOTCONN; |
116 | if (!fc->connected) | |
117 | goto out; | |
118 | ||
08a53cdc | 119 | req = fuse_request_alloc(); |
9bc5ddda | 120 | err = -ENOMEM; |
ce1d5a49 | 121 | if (!req) |
9bc5ddda | 122 | goto out; |
334f485d | 123 | |
33649c91 | 124 | fuse_req_init_context(req); |
9bc5ddda | 125 | req->waiting = 1; |
334f485d | 126 | return req; |
9bc5ddda MS |
127 | |
128 | out: | |
129 | atomic_dec(&fc->num_waiting); | |
130 | return ERR_PTR(err); | |
334f485d | 131 | } |
08cbf542 | 132 | EXPORT_SYMBOL_GPL(fuse_get_req); |
334f485d | 133 | |
33649c91 MS |
134 | /* |
135 | * Return request in fuse_file->reserved_req. However that may | |
136 | * currently be in use. If that is the case, wait for it to become | |
137 | * available. | |
138 | */ | |
139 | static struct fuse_req *get_reserved_req(struct fuse_conn *fc, | |
140 | struct file *file) | |
141 | { | |
142 | struct fuse_req *req = NULL; | |
143 | struct fuse_file *ff = file->private_data; | |
144 | ||
145 | do { | |
de5e3dec | 146 | wait_event(fc->reserved_req_waitq, ff->reserved_req); |
33649c91 MS |
147 | spin_lock(&fc->lock); |
148 | if (ff->reserved_req) { | |
149 | req = ff->reserved_req; | |
150 | ff->reserved_req = NULL; | |
151 | get_file(file); | |
152 | req->stolen_file = file; | |
153 | } | |
154 | spin_unlock(&fc->lock); | |
155 | } while (!req); | |
156 | ||
157 | return req; | |
158 | } | |
159 | ||
160 | /* | |
161 | * Put stolen request back into fuse_file->reserved_req | |
162 | */ | |
163 | static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req) | |
164 | { | |
165 | struct file *file = req->stolen_file; | |
166 | struct fuse_file *ff = file->private_data; | |
167 | ||
168 | spin_lock(&fc->lock); | |
169 | fuse_request_init(req); | |
170 | BUG_ON(ff->reserved_req); | |
171 | ff->reserved_req = req; | |
de5e3dec | 172 | wake_up_all(&fc->reserved_req_waitq); |
33649c91 MS |
173 | spin_unlock(&fc->lock); |
174 | fput(file); | |
175 | } | |
176 | ||
177 | /* | |
178 | * Gets a requests for a file operation, always succeeds | |
179 | * | |
180 | * This is used for sending the FLUSH request, which must get to | |
181 | * userspace, due to POSIX locks which may need to be unlocked. | |
182 | * | |
183 | * If allocation fails due to OOM, use the reserved request in | |
184 | * fuse_file. | |
185 | * | |
186 | * This is very unlikely to deadlock accidentally, since the | |
187 | * filesystem should not have it's own file open. If deadlock is | |
188 | * intentional, it can still be broken by "aborting" the filesystem. | |
189 | */ | |
190 | struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file) | |
191 | { | |
192 | struct fuse_req *req; | |
193 | ||
194 | atomic_inc(&fc->num_waiting); | |
195 | wait_event(fc->blocked_waitq, !fc->blocked); | |
196 | req = fuse_request_alloc(); | |
197 | if (!req) | |
198 | req = get_reserved_req(fc, file); | |
199 | ||
200 | fuse_req_init_context(req); | |
201 | req->waiting = 1; | |
202 | return req; | |
203 | } | |
204 | ||
334f485d | 205 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) |
7128ec2a MS |
206 | { |
207 | if (atomic_dec_and_test(&req->count)) { | |
9bc5ddda MS |
208 | if (req->waiting) |
209 | atomic_dec(&fc->num_waiting); | |
33649c91 MS |
210 | |
211 | if (req->stolen_file) | |
212 | put_reserved_req(fc, req); | |
213 | else | |
214 | fuse_request_free(req); | |
7128ec2a MS |
215 | } |
216 | } | |
08cbf542 | 217 | EXPORT_SYMBOL_GPL(fuse_put_request); |
7128ec2a | 218 | |
d12def1b MS |
219 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) |
220 | { | |
221 | unsigned nbytes = 0; | |
222 | unsigned i; | |
223 | ||
224 | for (i = 0; i < numargs; i++) | |
225 | nbytes += args[i].size; | |
226 | ||
227 | return nbytes; | |
228 | } | |
229 | ||
230 | static u64 fuse_get_unique(struct fuse_conn *fc) | |
231 | { | |
232 | fc->reqctr++; | |
233 | /* zero is special */ | |
234 | if (fc->reqctr == 0) | |
235 | fc->reqctr = 1; | |
236 | ||
237 | return fc->reqctr; | |
238 | } | |
239 | ||
240 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) | |
241 | { | |
d12def1b MS |
242 | req->in.h.len = sizeof(struct fuse_in_header) + |
243 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
244 | list_add_tail(&req->list, &fc->pending); | |
245 | req->state = FUSE_REQ_PENDING; | |
246 | if (!req->waiting) { | |
247 | req->waiting = 1; | |
248 | atomic_inc(&fc->num_waiting); | |
249 | } | |
250 | wake_up(&fc->waitq); | |
251 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); | |
252 | } | |
253 | ||
254 | static void flush_bg_queue(struct fuse_conn *fc) | |
255 | { | |
7a6d3c8b | 256 | while (fc->active_background < fc->max_background && |
d12def1b MS |
257 | !list_empty(&fc->bg_queue)) { |
258 | struct fuse_req *req; | |
259 | ||
260 | req = list_entry(fc->bg_queue.next, struct fuse_req, list); | |
261 | list_del(&req->list); | |
262 | fc->active_background++; | |
2d45ba38 | 263 | req->in.h.unique = fuse_get_unique(fc); |
d12def1b MS |
264 | queue_request(fc, req); |
265 | } | |
266 | } | |
267 | ||
334f485d MS |
268 | /* |
269 | * This function is called when a request is finished. Either a reply | |
f9a2842e | 270 | * has arrived or it was aborted (and not yet sent) or some error |
f43b155a | 271 | * occurred during communication with userspace, or the device file |
51eb01e7 MS |
272 | * was closed. The requester thread is woken up (if still waiting), |
273 | * the 'end' callback is called if given, else the reference to the | |
274 | * request is released | |
7128ec2a | 275 | * |
d7133114 | 276 | * Called with fc->lock, unlocks it |
334f485d MS |
277 | */ |
278 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
b9ca67b2 | 279 | __releases(fc->lock) |
334f485d | 280 | { |
51eb01e7 MS |
281 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; |
282 | req->end = NULL; | |
d77a1d5b | 283 | list_del(&req->list); |
a4d27e75 | 284 | list_del(&req->intr_entry); |
83cfd493 | 285 | req->state = FUSE_REQ_FINISHED; |
51eb01e7 | 286 | if (req->background) { |
7a6d3c8b | 287 | if (fc->num_background == fc->max_background) { |
51eb01e7 MS |
288 | fc->blocked = 0; |
289 | wake_up_all(&fc->blocked_waitq); | |
290 | } | |
7a6d3c8b | 291 | if (fc->num_background == fc->congestion_threshold && |
a325f9b9 | 292 | fc->connected && fc->bdi_initialized) { |
8aa7e847 JA |
293 | clear_bdi_congested(&fc->bdi, BLK_RW_SYNC); |
294 | clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC); | |
f92b99b9 | 295 | } |
51eb01e7 | 296 | fc->num_background--; |
d12def1b MS |
297 | fc->active_background--; |
298 | flush_bg_queue(fc); | |
334f485d | 299 | } |
51eb01e7 | 300 | spin_unlock(&fc->lock); |
51eb01e7 MS |
301 | wake_up(&req->waitq); |
302 | if (end) | |
303 | end(fc, req); | |
e9bb09dd | 304 | fuse_put_request(fc, req); |
334f485d MS |
305 | } |
306 | ||
a4d27e75 MS |
307 | static void wait_answer_interruptible(struct fuse_conn *fc, |
308 | struct fuse_req *req) | |
b9ca67b2 MS |
309 | __releases(fc->lock) |
310 | __acquires(fc->lock) | |
a4d27e75 MS |
311 | { |
312 | if (signal_pending(current)) | |
313 | return; | |
314 | ||
315 | spin_unlock(&fc->lock); | |
316 | wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED); | |
317 | spin_lock(&fc->lock); | |
318 | } | |
319 | ||
320 | static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req) | |
321 | { | |
322 | list_add_tail(&req->intr_entry, &fc->interrupts); | |
323 | wake_up(&fc->waitq); | |
324 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); | |
325 | } | |
326 | ||
7c352bdf | 327 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) |
b9ca67b2 MS |
328 | __releases(fc->lock) |
329 | __acquires(fc->lock) | |
334f485d | 330 | { |
a4d27e75 MS |
331 | if (!fc->no_interrupt) { |
332 | /* Any signal may interrupt this */ | |
333 | wait_answer_interruptible(fc, req); | |
334f485d | 334 | |
a4d27e75 MS |
335 | if (req->aborted) |
336 | goto aborted; | |
337 | if (req->state == FUSE_REQ_FINISHED) | |
338 | return; | |
339 | ||
340 | req->interrupted = 1; | |
341 | if (req->state == FUSE_REQ_SENT) | |
342 | queue_interrupt(fc, req); | |
343 | } | |
344 | ||
a131de0a | 345 | if (!req->force) { |
a4d27e75 MS |
346 | sigset_t oldset; |
347 | ||
348 | /* Only fatal signals may interrupt this */ | |
51eb01e7 | 349 | block_sigs(&oldset); |
a4d27e75 | 350 | wait_answer_interruptible(fc, req); |
51eb01e7 | 351 | restore_sigs(&oldset); |
a131de0a MS |
352 | |
353 | if (req->aborted) | |
354 | goto aborted; | |
355 | if (req->state == FUSE_REQ_FINISHED) | |
356 | return; | |
357 | ||
358 | /* Request is not yet in userspace, bail out */ | |
359 | if (req->state == FUSE_REQ_PENDING) { | |
360 | list_del(&req->list); | |
361 | __fuse_put_request(req); | |
362 | req->out.h.error = -EINTR; | |
363 | return; | |
364 | } | |
51eb01e7 | 365 | } |
334f485d | 366 | |
a131de0a MS |
367 | /* |
368 | * Either request is already in userspace, or it was forced. | |
369 | * Wait it out. | |
370 | */ | |
371 | spin_unlock(&fc->lock); | |
372 | wait_event(req->waitq, req->state == FUSE_REQ_FINISHED); | |
373 | spin_lock(&fc->lock); | |
a4d27e75 | 374 | |
a131de0a MS |
375 | if (!req->aborted) |
376 | return; | |
a4d27e75 MS |
377 | |
378 | aborted: | |
a131de0a | 379 | BUG_ON(req->state != FUSE_REQ_FINISHED); |
334f485d MS |
380 | if (req->locked) { |
381 | /* This is uninterruptible sleep, because data is | |
382 | being copied to/from the buffers of req. During | |
383 | locked state, there mustn't be any filesystem | |
384 | operation (e.g. page fault), since that could lead | |
385 | to deadlock */ | |
d7133114 | 386 | spin_unlock(&fc->lock); |
334f485d | 387 | wait_event(req->waitq, !req->locked); |
d7133114 | 388 | spin_lock(&fc->lock); |
334f485d | 389 | } |
334f485d MS |
390 | } |
391 | ||
b93f858a | 392 | void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
393 | { |
394 | req->isreply = 1; | |
d7133114 | 395 | spin_lock(&fc->lock); |
1e9a4ed9 | 396 | if (!fc->connected) |
334f485d MS |
397 | req->out.h.error = -ENOTCONN; |
398 | else if (fc->conn_error) | |
399 | req->out.h.error = -ECONNREFUSED; | |
400 | else { | |
2d45ba38 | 401 | req->in.h.unique = fuse_get_unique(fc); |
334f485d MS |
402 | queue_request(fc, req); |
403 | /* acquire extra reference, since request is still needed | |
404 | after request_end() */ | |
405 | __fuse_get_request(req); | |
406 | ||
7c352bdf | 407 | request_wait_answer(fc, req); |
334f485d | 408 | } |
d7133114 | 409 | spin_unlock(&fc->lock); |
334f485d | 410 | } |
08cbf542 | 411 | EXPORT_SYMBOL_GPL(fuse_request_send); |
334f485d | 412 | |
b93f858a TH |
413 | static void fuse_request_send_nowait_locked(struct fuse_conn *fc, |
414 | struct fuse_req *req) | |
d12def1b MS |
415 | { |
416 | req->background = 1; | |
417 | fc->num_background++; | |
7a6d3c8b | 418 | if (fc->num_background == fc->max_background) |
d12def1b | 419 | fc->blocked = 1; |
7a6d3c8b | 420 | if (fc->num_background == fc->congestion_threshold && |
a325f9b9 | 421 | fc->bdi_initialized) { |
8aa7e847 JA |
422 | set_bdi_congested(&fc->bdi, BLK_RW_SYNC); |
423 | set_bdi_congested(&fc->bdi, BLK_RW_ASYNC); | |
d12def1b MS |
424 | } |
425 | list_add_tail(&req->list, &fc->bg_queue); | |
426 | flush_bg_queue(fc); | |
427 | } | |
428 | ||
b93f858a | 429 | static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
334f485d | 430 | { |
d7133114 | 431 | spin_lock(&fc->lock); |
1e9a4ed9 | 432 | if (fc->connected) { |
b93f858a | 433 | fuse_request_send_nowait_locked(fc, req); |
d7133114 | 434 | spin_unlock(&fc->lock); |
334f485d MS |
435 | } else { |
436 | req->out.h.error = -ENOTCONN; | |
437 | request_end(fc, req); | |
438 | } | |
439 | } | |
440 | ||
b93f858a | 441 | void fuse_request_send_noreply(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
442 | { |
443 | req->isreply = 0; | |
b93f858a | 444 | fuse_request_send_nowait(fc, req); |
334f485d MS |
445 | } |
446 | ||
b93f858a | 447 | void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
448 | { |
449 | req->isreply = 1; | |
b93f858a | 450 | fuse_request_send_nowait(fc, req); |
334f485d | 451 | } |
08cbf542 | 452 | EXPORT_SYMBOL_GPL(fuse_request_send_background); |
334f485d | 453 | |
2d45ba38 MS |
454 | static int fuse_request_send_notify_reply(struct fuse_conn *fc, |
455 | struct fuse_req *req, u64 unique) | |
456 | { | |
457 | int err = -ENODEV; | |
458 | ||
459 | req->isreply = 0; | |
460 | req->in.h.unique = unique; | |
461 | spin_lock(&fc->lock); | |
462 | if (fc->connected) { | |
463 | queue_request(fc, req); | |
464 | err = 0; | |
465 | } | |
466 | spin_unlock(&fc->lock); | |
467 | ||
468 | return err; | |
469 | } | |
470 | ||
3be5a52b MS |
471 | /* |
472 | * Called under fc->lock | |
473 | * | |
474 | * fc->connected must have been checked previously | |
475 | */ | |
b93f858a TH |
476 | void fuse_request_send_background_locked(struct fuse_conn *fc, |
477 | struct fuse_req *req) | |
3be5a52b MS |
478 | { |
479 | req->isreply = 1; | |
b93f858a | 480 | fuse_request_send_nowait_locked(fc, req); |
3be5a52b MS |
481 | } |
482 | ||
334f485d MS |
483 | /* |
484 | * Lock the request. Up to the next unlock_request() there mustn't be | |
485 | * anything that could cause a page-fault. If the request was already | |
f9a2842e | 486 | * aborted bail out. |
334f485d | 487 | */ |
d7133114 | 488 | static int lock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
489 | { |
490 | int err = 0; | |
491 | if (req) { | |
d7133114 | 492 | spin_lock(&fc->lock); |
f9a2842e | 493 | if (req->aborted) |
334f485d MS |
494 | err = -ENOENT; |
495 | else | |
496 | req->locked = 1; | |
d7133114 | 497 | spin_unlock(&fc->lock); |
334f485d MS |
498 | } |
499 | return err; | |
500 | } | |
501 | ||
502 | /* | |
f9a2842e | 503 | * Unlock request. If it was aborted during being locked, the |
334f485d MS |
504 | * requester thread is currently waiting for it to be unlocked, so |
505 | * wake it up. | |
506 | */ | |
d7133114 | 507 | static void unlock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
508 | { |
509 | if (req) { | |
d7133114 | 510 | spin_lock(&fc->lock); |
334f485d | 511 | req->locked = 0; |
f9a2842e | 512 | if (req->aborted) |
334f485d | 513 | wake_up(&req->waitq); |
d7133114 | 514 | spin_unlock(&fc->lock); |
334f485d MS |
515 | } |
516 | } | |
517 | ||
518 | struct fuse_copy_state { | |
d7133114 | 519 | struct fuse_conn *fc; |
334f485d MS |
520 | int write; |
521 | struct fuse_req *req; | |
522 | const struct iovec *iov; | |
dd3bb14f MS |
523 | struct pipe_buffer *pipebufs; |
524 | struct pipe_buffer *currbuf; | |
525 | struct pipe_inode_info *pipe; | |
334f485d MS |
526 | unsigned long nr_segs; |
527 | unsigned long seglen; | |
528 | unsigned long addr; | |
529 | struct page *pg; | |
530 | void *mapaddr; | |
531 | void *buf; | |
532 | unsigned len; | |
ce534fb0 | 533 | unsigned move_pages:1; |
334f485d MS |
534 | }; |
535 | ||
d7133114 | 536 | static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc, |
c3021629 | 537 | int write, |
d7133114 | 538 | const struct iovec *iov, unsigned long nr_segs) |
334f485d MS |
539 | { |
540 | memset(cs, 0, sizeof(*cs)); | |
d7133114 | 541 | cs->fc = fc; |
334f485d | 542 | cs->write = write; |
334f485d MS |
543 | cs->iov = iov; |
544 | cs->nr_segs = nr_segs; | |
545 | } | |
546 | ||
547 | /* Unmap and put previous page of userspace buffer */ | |
8bfc016d | 548 | static void fuse_copy_finish(struct fuse_copy_state *cs) |
334f485d | 549 | { |
dd3bb14f MS |
550 | if (cs->currbuf) { |
551 | struct pipe_buffer *buf = cs->currbuf; | |
552 | ||
c3021629 MS |
553 | if (!cs->write) { |
554 | buf->ops->unmap(cs->pipe, buf, cs->mapaddr); | |
555 | } else { | |
7909b1c6 | 556 | kunmap(buf->page); |
c3021629 MS |
557 | buf->len = PAGE_SIZE - cs->len; |
558 | } | |
dd3bb14f MS |
559 | cs->currbuf = NULL; |
560 | cs->mapaddr = NULL; | |
561 | } else if (cs->mapaddr) { | |
7909b1c6 | 562 | kunmap(cs->pg); |
334f485d MS |
563 | if (cs->write) { |
564 | flush_dcache_page(cs->pg); | |
565 | set_page_dirty_lock(cs->pg); | |
566 | } | |
567 | put_page(cs->pg); | |
568 | cs->mapaddr = NULL; | |
569 | } | |
570 | } | |
571 | ||
572 | /* | |
573 | * Get another pagefull of userspace buffer, and map it to kernel | |
574 | * address space, and lock request | |
575 | */ | |
576 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
577 | { | |
578 | unsigned long offset; | |
579 | int err; | |
580 | ||
d7133114 | 581 | unlock_request(cs->fc, cs->req); |
334f485d | 582 | fuse_copy_finish(cs); |
dd3bb14f MS |
583 | if (cs->pipebufs) { |
584 | struct pipe_buffer *buf = cs->pipebufs; | |
585 | ||
c3021629 MS |
586 | if (!cs->write) { |
587 | err = buf->ops->confirm(cs->pipe, buf); | |
588 | if (err) | |
589 | return err; | |
590 | ||
591 | BUG_ON(!cs->nr_segs); | |
592 | cs->currbuf = buf; | |
7909b1c6 | 593 | cs->mapaddr = buf->ops->map(cs->pipe, buf, 0); |
c3021629 MS |
594 | cs->len = buf->len; |
595 | cs->buf = cs->mapaddr + buf->offset; | |
596 | cs->pipebufs++; | |
597 | cs->nr_segs--; | |
598 | } else { | |
599 | struct page *page; | |
dd3bb14f | 600 | |
c3021629 MS |
601 | if (cs->nr_segs == cs->pipe->buffers) |
602 | return -EIO; | |
603 | ||
604 | page = alloc_page(GFP_HIGHUSER); | |
605 | if (!page) | |
606 | return -ENOMEM; | |
607 | ||
608 | buf->page = page; | |
609 | buf->offset = 0; | |
610 | buf->len = 0; | |
611 | ||
612 | cs->currbuf = buf; | |
7909b1c6 | 613 | cs->mapaddr = kmap(page); |
c3021629 MS |
614 | cs->buf = cs->mapaddr; |
615 | cs->len = PAGE_SIZE; | |
616 | cs->pipebufs++; | |
617 | cs->nr_segs++; | |
618 | } | |
dd3bb14f MS |
619 | } else { |
620 | if (!cs->seglen) { | |
621 | BUG_ON(!cs->nr_segs); | |
622 | cs->seglen = cs->iov[0].iov_len; | |
623 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
624 | cs->iov++; | |
625 | cs->nr_segs--; | |
626 | } | |
627 | err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg); | |
628 | if (err < 0) | |
629 | return err; | |
630 | BUG_ON(err != 1); | |
631 | offset = cs->addr % PAGE_SIZE; | |
7909b1c6 | 632 | cs->mapaddr = kmap(cs->pg); |
dd3bb14f MS |
633 | cs->buf = cs->mapaddr + offset; |
634 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
635 | cs->seglen -= cs->len; | |
636 | cs->addr += cs->len; | |
334f485d | 637 | } |
334f485d | 638 | |
d7133114 | 639 | return lock_request(cs->fc, cs->req); |
334f485d MS |
640 | } |
641 | ||
642 | /* Do as much copy to/from userspace buffer as we can */ | |
8bfc016d | 643 | static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) |
334f485d MS |
644 | { |
645 | unsigned ncpy = min(*size, cs->len); | |
646 | if (val) { | |
647 | if (cs->write) | |
648 | memcpy(cs->buf, *val, ncpy); | |
649 | else | |
650 | memcpy(*val, cs->buf, ncpy); | |
651 | *val += ncpy; | |
652 | } | |
653 | *size -= ncpy; | |
654 | cs->len -= ncpy; | |
655 | cs->buf += ncpy; | |
656 | return ncpy; | |
657 | } | |
658 | ||
ce534fb0 MS |
659 | static int fuse_check_page(struct page *page) |
660 | { | |
661 | if (page_mapcount(page) || | |
662 | page->mapping != NULL || | |
663 | page_count(page) != 1 || | |
664 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP & | |
665 | ~(1 << PG_locked | | |
666 | 1 << PG_referenced | | |
667 | 1 << PG_uptodate | | |
668 | 1 << PG_lru | | |
669 | 1 << PG_active | | |
670 | 1 << PG_reclaim))) { | |
671 | printk(KERN_WARNING "fuse: trying to steal weird page\n"); | |
672 | 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); | |
673 | return 1; | |
674 | } | |
675 | return 0; | |
676 | } | |
677 | ||
678 | static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep) | |
679 | { | |
680 | int err; | |
681 | struct page *oldpage = *pagep; | |
682 | struct page *newpage; | |
683 | struct pipe_buffer *buf = cs->pipebufs; | |
684 | struct address_space *mapping; | |
685 | pgoff_t index; | |
686 | ||
687 | unlock_request(cs->fc, cs->req); | |
688 | fuse_copy_finish(cs); | |
689 | ||
690 | err = buf->ops->confirm(cs->pipe, buf); | |
691 | if (err) | |
692 | return err; | |
693 | ||
694 | BUG_ON(!cs->nr_segs); | |
695 | cs->currbuf = buf; | |
696 | cs->len = buf->len; | |
697 | cs->pipebufs++; | |
698 | cs->nr_segs--; | |
699 | ||
700 | if (cs->len != PAGE_SIZE) | |
701 | goto out_fallback; | |
702 | ||
703 | if (buf->ops->steal(cs->pipe, buf) != 0) | |
704 | goto out_fallback; | |
705 | ||
706 | newpage = buf->page; | |
707 | ||
708 | if (WARN_ON(!PageUptodate(newpage))) | |
709 | return -EIO; | |
710 | ||
711 | ClearPageMappedToDisk(newpage); | |
712 | ||
713 | if (fuse_check_page(newpage) != 0) | |
714 | goto out_fallback_unlock; | |
715 | ||
716 | mapping = oldpage->mapping; | |
717 | index = oldpage->index; | |
718 | ||
719 | /* | |
720 | * This is a new and locked page, it shouldn't be mapped or | |
721 | * have any special flags on it | |
722 | */ | |
723 | if (WARN_ON(page_mapped(oldpage))) | |
724 | goto out_fallback_unlock; | |
725 | if (WARN_ON(page_has_private(oldpage))) | |
726 | goto out_fallback_unlock; | |
727 | if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage))) | |
728 | goto out_fallback_unlock; | |
729 | if (WARN_ON(PageMlocked(oldpage))) | |
730 | goto out_fallback_unlock; | |
731 | ||
732 | remove_from_page_cache(oldpage); | |
733 | page_cache_release(oldpage); | |
734 | ||
735 | err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL); | |
736 | if (err) { | |
737 | printk(KERN_WARNING "fuse_try_move_page: failed to add page"); | |
738 | goto out_fallback_unlock; | |
739 | } | |
740 | page_cache_get(newpage); | |
741 | ||
742 | if (!(buf->flags & PIPE_BUF_FLAG_LRU)) | |
743 | lru_cache_add_file(newpage); | |
744 | ||
745 | err = 0; | |
746 | spin_lock(&cs->fc->lock); | |
747 | if (cs->req->aborted) | |
748 | err = -ENOENT; | |
749 | else | |
750 | *pagep = newpage; | |
751 | spin_unlock(&cs->fc->lock); | |
752 | ||
753 | if (err) { | |
754 | unlock_page(newpage); | |
755 | page_cache_release(newpage); | |
756 | return err; | |
757 | } | |
758 | ||
759 | unlock_page(oldpage); | |
760 | page_cache_release(oldpage); | |
761 | cs->len = 0; | |
762 | ||
763 | return 0; | |
764 | ||
765 | out_fallback_unlock: | |
766 | unlock_page(newpage); | |
767 | out_fallback: | |
768 | cs->mapaddr = buf->ops->map(cs->pipe, buf, 1); | |
769 | cs->buf = cs->mapaddr + buf->offset; | |
770 | ||
771 | err = lock_request(cs->fc, cs->req); | |
772 | if (err) | |
773 | return err; | |
774 | ||
775 | return 1; | |
776 | } | |
777 | ||
c3021629 MS |
778 | static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page, |
779 | unsigned offset, unsigned count) | |
780 | { | |
781 | struct pipe_buffer *buf; | |
782 | ||
783 | if (cs->nr_segs == cs->pipe->buffers) | |
784 | return -EIO; | |
785 | ||
786 | unlock_request(cs->fc, cs->req); | |
787 | fuse_copy_finish(cs); | |
788 | ||
789 | buf = cs->pipebufs; | |
790 | page_cache_get(page); | |
791 | buf->page = page; | |
792 | buf->offset = offset; | |
793 | buf->len = count; | |
794 | ||
795 | cs->pipebufs++; | |
796 | cs->nr_segs++; | |
797 | cs->len = 0; | |
798 | ||
799 | return 0; | |
800 | } | |
801 | ||
334f485d MS |
802 | /* |
803 | * Copy a page in the request to/from the userspace buffer. Must be | |
804 | * done atomically | |
805 | */ | |
ce534fb0 | 806 | static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep, |
8bfc016d | 807 | unsigned offset, unsigned count, int zeroing) |
334f485d | 808 | { |
ce534fb0 MS |
809 | int err; |
810 | struct page *page = *pagep; | |
811 | ||
b6777c40 MS |
812 | if (page && zeroing && count < PAGE_SIZE) |
813 | clear_highpage(page); | |
814 | ||
334f485d | 815 | while (count) { |
c3021629 MS |
816 | if (cs->write && cs->pipebufs && page) { |
817 | return fuse_ref_page(cs, page, offset, count); | |
818 | } else if (!cs->len) { | |
ce534fb0 MS |
819 | if (cs->move_pages && page && |
820 | offset == 0 && count == PAGE_SIZE) { | |
821 | err = fuse_try_move_page(cs, pagep); | |
822 | if (err <= 0) | |
823 | return err; | |
824 | } else { | |
825 | err = fuse_copy_fill(cs); | |
826 | if (err) | |
827 | return err; | |
828 | } | |
1729a16c | 829 | } |
334f485d | 830 | if (page) { |
b6777c40 | 831 | void *mapaddr = kmap_atomic(page, KM_USER0); |
334f485d MS |
832 | void *buf = mapaddr + offset; |
833 | offset += fuse_copy_do(cs, &buf, &count); | |
b6777c40 | 834 | kunmap_atomic(mapaddr, KM_USER0); |
334f485d MS |
835 | } else |
836 | offset += fuse_copy_do(cs, NULL, &count); | |
837 | } | |
838 | if (page && !cs->write) | |
839 | flush_dcache_page(page); | |
840 | return 0; | |
841 | } | |
842 | ||
843 | /* Copy pages in the request to/from userspace buffer */ | |
844 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
845 | int zeroing) | |
846 | { | |
847 | unsigned i; | |
848 | struct fuse_req *req = cs->req; | |
849 | unsigned offset = req->page_offset; | |
850 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
851 | ||
852 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
ce534fb0 MS |
853 | int err; |
854 | ||
855 | err = fuse_copy_page(cs, &req->pages[i], offset, count, | |
856 | zeroing); | |
334f485d MS |
857 | if (err) |
858 | return err; | |
859 | ||
860 | nbytes -= count; | |
861 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
862 | offset = 0; | |
863 | } | |
864 | return 0; | |
865 | } | |
866 | ||
867 | /* Copy a single argument in the request to/from userspace buffer */ | |
868 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
869 | { | |
870 | while (size) { | |
1729a16c MS |
871 | if (!cs->len) { |
872 | int err = fuse_copy_fill(cs); | |
873 | if (err) | |
874 | return err; | |
875 | } | |
334f485d MS |
876 | fuse_copy_do(cs, &val, &size); |
877 | } | |
878 | return 0; | |
879 | } | |
880 | ||
881 | /* Copy request arguments to/from userspace buffer */ | |
882 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
883 | unsigned argpages, struct fuse_arg *args, | |
884 | int zeroing) | |
885 | { | |
886 | int err = 0; | |
887 | unsigned i; | |
888 | ||
889 | for (i = 0; !err && i < numargs; i++) { | |
890 | struct fuse_arg *arg = &args[i]; | |
891 | if (i == numargs - 1 && argpages) | |
892 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
893 | else | |
894 | err = fuse_copy_one(cs, arg->value, arg->size); | |
895 | } | |
896 | return err; | |
897 | } | |
898 | ||
a4d27e75 MS |
899 | static int request_pending(struct fuse_conn *fc) |
900 | { | |
901 | return !list_empty(&fc->pending) || !list_empty(&fc->interrupts); | |
902 | } | |
903 | ||
334f485d MS |
904 | /* Wait until a request is available on the pending list */ |
905 | static void request_wait(struct fuse_conn *fc) | |
b9ca67b2 MS |
906 | __releases(fc->lock) |
907 | __acquires(fc->lock) | |
334f485d MS |
908 | { |
909 | DECLARE_WAITQUEUE(wait, current); | |
910 | ||
911 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
a4d27e75 | 912 | while (fc->connected && !request_pending(fc)) { |
334f485d MS |
913 | set_current_state(TASK_INTERRUPTIBLE); |
914 | if (signal_pending(current)) | |
915 | break; | |
916 | ||
d7133114 | 917 | spin_unlock(&fc->lock); |
334f485d | 918 | schedule(); |
d7133114 | 919 | spin_lock(&fc->lock); |
334f485d MS |
920 | } |
921 | set_current_state(TASK_RUNNING); | |
922 | remove_wait_queue(&fc->waitq, &wait); | |
923 | } | |
924 | ||
a4d27e75 MS |
925 | /* |
926 | * Transfer an interrupt request to userspace | |
927 | * | |
928 | * Unlike other requests this is assembled on demand, without a need | |
929 | * to allocate a separate fuse_req structure. | |
930 | * | |
931 | * Called with fc->lock held, releases it | |
932 | */ | |
c3021629 MS |
933 | static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs, |
934 | size_t nbytes, struct fuse_req *req) | |
b9ca67b2 | 935 | __releases(fc->lock) |
a4d27e75 | 936 | { |
a4d27e75 MS |
937 | struct fuse_in_header ih; |
938 | struct fuse_interrupt_in arg; | |
939 | unsigned reqsize = sizeof(ih) + sizeof(arg); | |
940 | int err; | |
941 | ||
942 | list_del_init(&req->intr_entry); | |
943 | req->intr_unique = fuse_get_unique(fc); | |
944 | memset(&ih, 0, sizeof(ih)); | |
945 | memset(&arg, 0, sizeof(arg)); | |
946 | ih.len = reqsize; | |
947 | ih.opcode = FUSE_INTERRUPT; | |
948 | ih.unique = req->intr_unique; | |
949 | arg.unique = req->in.h.unique; | |
950 | ||
951 | spin_unlock(&fc->lock); | |
c3021629 | 952 | if (nbytes < reqsize) |
a4d27e75 MS |
953 | return -EINVAL; |
954 | ||
c3021629 | 955 | err = fuse_copy_one(cs, &ih, sizeof(ih)); |
a4d27e75 | 956 | if (!err) |
c3021629 MS |
957 | err = fuse_copy_one(cs, &arg, sizeof(arg)); |
958 | fuse_copy_finish(cs); | |
a4d27e75 MS |
959 | |
960 | return err ? err : reqsize; | |
961 | } | |
962 | ||
334f485d MS |
963 | /* |
964 | * Read a single request into the userspace filesystem's buffer. This | |
965 | * function waits until a request is available, then removes it from | |
966 | * the pending list and copies request data to userspace buffer. If | |
f9a2842e MS |
967 | * no reply is needed (FORGET) or request has been aborted or there |
968 | * was an error during the copying then it's finished by calling | |
334f485d MS |
969 | * request_end(). Otherwise add it to the processing list, and set |
970 | * the 'sent' flag. | |
971 | */ | |
c3021629 MS |
972 | static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file, |
973 | struct fuse_copy_state *cs, size_t nbytes) | |
334f485d MS |
974 | { |
975 | int err; | |
334f485d MS |
976 | struct fuse_req *req; |
977 | struct fuse_in *in; | |
334f485d MS |
978 | unsigned reqsize; |
979 | ||
1d3d752b | 980 | restart: |
d7133114 | 981 | spin_lock(&fc->lock); |
e5ac1d1e JD |
982 | err = -EAGAIN; |
983 | if ((file->f_flags & O_NONBLOCK) && fc->connected && | |
a4d27e75 | 984 | !request_pending(fc)) |
e5ac1d1e JD |
985 | goto err_unlock; |
986 | ||
334f485d MS |
987 | request_wait(fc); |
988 | err = -ENODEV; | |
9ba7cbba | 989 | if (!fc->connected) |
334f485d MS |
990 | goto err_unlock; |
991 | err = -ERESTARTSYS; | |
a4d27e75 | 992 | if (!request_pending(fc)) |
334f485d MS |
993 | goto err_unlock; |
994 | ||
a4d27e75 MS |
995 | if (!list_empty(&fc->interrupts)) { |
996 | req = list_entry(fc->interrupts.next, struct fuse_req, | |
997 | intr_entry); | |
c3021629 | 998 | return fuse_read_interrupt(fc, cs, nbytes, req); |
a4d27e75 MS |
999 | } |
1000 | ||
334f485d | 1001 | req = list_entry(fc->pending.next, struct fuse_req, list); |
83cfd493 | 1002 | req->state = FUSE_REQ_READING; |
d77a1d5b | 1003 | list_move(&req->list, &fc->io); |
334f485d MS |
1004 | |
1005 | in = &req->in; | |
1d3d752b MS |
1006 | reqsize = in->h.len; |
1007 | /* If request is too large, reply with an error and restart the read */ | |
c3021629 | 1008 | if (nbytes < reqsize) { |
1d3d752b MS |
1009 | req->out.h.error = -EIO; |
1010 | /* SETXATTR is special, since it may contain too large data */ | |
1011 | if (in->h.opcode == FUSE_SETXATTR) | |
1012 | req->out.h.error = -E2BIG; | |
1013 | request_end(fc, req); | |
1014 | goto restart; | |
334f485d | 1015 | } |
d7133114 | 1016 | spin_unlock(&fc->lock); |
c3021629 MS |
1017 | cs->req = req; |
1018 | err = fuse_copy_one(cs, &in->h, sizeof(in->h)); | |
1d3d752b | 1019 | if (!err) |
c3021629 | 1020 | err = fuse_copy_args(cs, in->numargs, in->argpages, |
1d3d752b | 1021 | (struct fuse_arg *) in->args, 0); |
c3021629 | 1022 | fuse_copy_finish(cs); |
d7133114 | 1023 | spin_lock(&fc->lock); |
334f485d | 1024 | req->locked = 0; |
c9c9d7df MS |
1025 | if (req->aborted) { |
1026 | request_end(fc, req); | |
1027 | return -ENODEV; | |
1028 | } | |
334f485d | 1029 | if (err) { |
c9c9d7df | 1030 | req->out.h.error = -EIO; |
334f485d MS |
1031 | request_end(fc, req); |
1032 | return err; | |
1033 | } | |
1034 | if (!req->isreply) | |
1035 | request_end(fc, req); | |
1036 | else { | |
83cfd493 | 1037 | req->state = FUSE_REQ_SENT; |
d77a1d5b | 1038 | list_move_tail(&req->list, &fc->processing); |
a4d27e75 MS |
1039 | if (req->interrupted) |
1040 | queue_interrupt(fc, req); | |
d7133114 | 1041 | spin_unlock(&fc->lock); |
334f485d MS |
1042 | } |
1043 | return reqsize; | |
1044 | ||
1045 | err_unlock: | |
d7133114 | 1046 | spin_unlock(&fc->lock); |
334f485d MS |
1047 | return err; |
1048 | } | |
1049 | ||
c3021629 MS |
1050 | static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov, |
1051 | unsigned long nr_segs, loff_t pos) | |
1052 | { | |
1053 | struct fuse_copy_state cs; | |
1054 | struct file *file = iocb->ki_filp; | |
1055 | struct fuse_conn *fc = fuse_get_conn(file); | |
1056 | if (!fc) | |
1057 | return -EPERM; | |
1058 | ||
1059 | fuse_copy_init(&cs, fc, 1, iov, nr_segs); | |
1060 | ||
1061 | return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs)); | |
1062 | } | |
1063 | ||
1064 | static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe, | |
1065 | struct pipe_buffer *buf) | |
1066 | { | |
1067 | return 1; | |
1068 | } | |
1069 | ||
1070 | static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = { | |
1071 | .can_merge = 0, | |
1072 | .map = generic_pipe_buf_map, | |
1073 | .unmap = generic_pipe_buf_unmap, | |
1074 | .confirm = generic_pipe_buf_confirm, | |
1075 | .release = generic_pipe_buf_release, | |
1076 | .steal = fuse_dev_pipe_buf_steal, | |
1077 | .get = generic_pipe_buf_get, | |
1078 | }; | |
1079 | ||
1080 | static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos, | |
1081 | struct pipe_inode_info *pipe, | |
1082 | size_t len, unsigned int flags) | |
1083 | { | |
1084 | int ret; | |
1085 | int page_nr = 0; | |
1086 | int do_wakeup = 0; | |
1087 | struct pipe_buffer *bufs; | |
1088 | struct fuse_copy_state cs; | |
1089 | struct fuse_conn *fc = fuse_get_conn(in); | |
1090 | if (!fc) | |
1091 | return -EPERM; | |
1092 | ||
1093 | bufs = kmalloc(pipe->buffers * sizeof (struct pipe_buffer), GFP_KERNEL); | |
1094 | if (!bufs) | |
1095 | return -ENOMEM; | |
1096 | ||
1097 | fuse_copy_init(&cs, fc, 1, NULL, 0); | |
1098 | cs.pipebufs = bufs; | |
1099 | cs.pipe = pipe; | |
1100 | ret = fuse_dev_do_read(fc, in, &cs, len); | |
1101 | if (ret < 0) | |
1102 | goto out; | |
1103 | ||
1104 | ret = 0; | |
1105 | pipe_lock(pipe); | |
1106 | ||
1107 | if (!pipe->readers) { | |
1108 | send_sig(SIGPIPE, current, 0); | |
1109 | if (!ret) | |
1110 | ret = -EPIPE; | |
1111 | goto out_unlock; | |
1112 | } | |
1113 | ||
1114 | if (pipe->nrbufs + cs.nr_segs > pipe->buffers) { | |
1115 | ret = -EIO; | |
1116 | goto out_unlock; | |
1117 | } | |
1118 | ||
1119 | while (page_nr < cs.nr_segs) { | |
1120 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
1121 | struct pipe_buffer *buf = pipe->bufs + newbuf; | |
1122 | ||
1123 | buf->page = bufs[page_nr].page; | |
1124 | buf->offset = bufs[page_nr].offset; | |
1125 | buf->len = bufs[page_nr].len; | |
1126 | buf->ops = &fuse_dev_pipe_buf_ops; | |
1127 | ||
1128 | pipe->nrbufs++; | |
1129 | page_nr++; | |
1130 | ret += buf->len; | |
1131 | ||
1132 | if (pipe->inode) | |
1133 | do_wakeup = 1; | |
1134 | } | |
1135 | ||
1136 | out_unlock: | |
1137 | pipe_unlock(pipe); | |
1138 | ||
1139 | if (do_wakeup) { | |
1140 | smp_mb(); | |
1141 | if (waitqueue_active(&pipe->wait)) | |
1142 | wake_up_interruptible(&pipe->wait); | |
1143 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
1144 | } | |
1145 | ||
1146 | out: | |
1147 | for (; page_nr < cs.nr_segs; page_nr++) | |
1148 | page_cache_release(bufs[page_nr].page); | |
1149 | ||
1150 | kfree(bufs); | |
1151 | return ret; | |
1152 | } | |
1153 | ||
95668a69 TH |
1154 | static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size, |
1155 | struct fuse_copy_state *cs) | |
1156 | { | |
1157 | struct fuse_notify_poll_wakeup_out outarg; | |
f6d47a17 | 1158 | int err = -EINVAL; |
95668a69 TH |
1159 | |
1160 | if (size != sizeof(outarg)) | |
f6d47a17 | 1161 | goto err; |
95668a69 TH |
1162 | |
1163 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1164 | if (err) | |
f6d47a17 | 1165 | goto err; |
95668a69 | 1166 | |
f6d47a17 | 1167 | fuse_copy_finish(cs); |
95668a69 | 1168 | return fuse_notify_poll_wakeup(fc, &outarg); |
f6d47a17 MS |
1169 | |
1170 | err: | |
1171 | fuse_copy_finish(cs); | |
1172 | return err; | |
95668a69 TH |
1173 | } |
1174 | ||
3b463ae0 JM |
1175 | static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size, |
1176 | struct fuse_copy_state *cs) | |
1177 | { | |
1178 | struct fuse_notify_inval_inode_out outarg; | |
1179 | int err = -EINVAL; | |
1180 | ||
1181 | if (size != sizeof(outarg)) | |
1182 | goto err; | |
1183 | ||
1184 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1185 | if (err) | |
1186 | goto err; | |
1187 | fuse_copy_finish(cs); | |
1188 | ||
1189 | down_read(&fc->killsb); | |
1190 | err = -ENOENT; | |
b21dda43 MS |
1191 | if (fc->sb) { |
1192 | err = fuse_reverse_inval_inode(fc->sb, outarg.ino, | |
1193 | outarg.off, outarg.len); | |
1194 | } | |
3b463ae0 JM |
1195 | up_read(&fc->killsb); |
1196 | return err; | |
1197 | ||
1198 | err: | |
1199 | fuse_copy_finish(cs); | |
1200 | return err; | |
1201 | } | |
1202 | ||
1203 | static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size, | |
1204 | struct fuse_copy_state *cs) | |
1205 | { | |
1206 | struct fuse_notify_inval_entry_out outarg; | |
b2d82ee3 FW |
1207 | int err = -ENOMEM; |
1208 | char *buf; | |
3b463ae0 JM |
1209 | struct qstr name; |
1210 | ||
b2d82ee3 FW |
1211 | buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); |
1212 | if (!buf) | |
1213 | goto err; | |
1214 | ||
1215 | err = -EINVAL; | |
3b463ae0 JM |
1216 | if (size < sizeof(outarg)) |
1217 | goto err; | |
1218 | ||
1219 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1220 | if (err) | |
1221 | goto err; | |
1222 | ||
1223 | err = -ENAMETOOLONG; | |
1224 | if (outarg.namelen > FUSE_NAME_MAX) | |
1225 | goto err; | |
1226 | ||
1227 | name.name = buf; | |
1228 | name.len = outarg.namelen; | |
1229 | err = fuse_copy_one(cs, buf, outarg.namelen + 1); | |
1230 | if (err) | |
1231 | goto err; | |
1232 | fuse_copy_finish(cs); | |
1233 | buf[outarg.namelen] = 0; | |
1234 | name.hash = full_name_hash(name.name, name.len); | |
1235 | ||
1236 | down_read(&fc->killsb); | |
1237 | err = -ENOENT; | |
b21dda43 MS |
1238 | if (fc->sb) |
1239 | err = fuse_reverse_inval_entry(fc->sb, outarg.parent, &name); | |
3b463ae0 | 1240 | up_read(&fc->killsb); |
b2d82ee3 | 1241 | kfree(buf); |
3b463ae0 JM |
1242 | return err; |
1243 | ||
1244 | err: | |
b2d82ee3 | 1245 | kfree(buf); |
3b463ae0 JM |
1246 | fuse_copy_finish(cs); |
1247 | return err; | |
1248 | } | |
1249 | ||
a1d75f25 MS |
1250 | static int fuse_notify_store(struct fuse_conn *fc, unsigned int size, |
1251 | struct fuse_copy_state *cs) | |
1252 | { | |
1253 | struct fuse_notify_store_out outarg; | |
1254 | struct inode *inode; | |
1255 | struct address_space *mapping; | |
1256 | u64 nodeid; | |
1257 | int err; | |
1258 | pgoff_t index; | |
1259 | unsigned int offset; | |
1260 | unsigned int num; | |
1261 | loff_t file_size; | |
1262 | loff_t end; | |
1263 | ||
1264 | err = -EINVAL; | |
1265 | if (size < sizeof(outarg)) | |
1266 | goto out_finish; | |
1267 | ||
1268 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1269 | if (err) | |
1270 | goto out_finish; | |
1271 | ||
1272 | err = -EINVAL; | |
1273 | if (size - sizeof(outarg) != outarg.size) | |
1274 | goto out_finish; | |
1275 | ||
1276 | nodeid = outarg.nodeid; | |
1277 | ||
1278 | down_read(&fc->killsb); | |
1279 | ||
1280 | err = -ENOENT; | |
1281 | if (!fc->sb) | |
1282 | goto out_up_killsb; | |
1283 | ||
1284 | inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid); | |
1285 | if (!inode) | |
1286 | goto out_up_killsb; | |
1287 | ||
1288 | mapping = inode->i_mapping; | |
1289 | index = outarg.offset >> PAGE_CACHE_SHIFT; | |
1290 | offset = outarg.offset & ~PAGE_CACHE_MASK; | |
1291 | file_size = i_size_read(inode); | |
1292 | end = outarg.offset + outarg.size; | |
1293 | if (end > file_size) { | |
1294 | file_size = end; | |
1295 | fuse_write_update_size(inode, file_size); | |
1296 | } | |
1297 | ||
1298 | num = outarg.size; | |
1299 | while (num) { | |
1300 | struct page *page; | |
1301 | unsigned int this_num; | |
1302 | ||
1303 | err = -ENOMEM; | |
1304 | page = find_or_create_page(mapping, index, | |
1305 | mapping_gfp_mask(mapping)); | |
1306 | if (!page) | |
1307 | goto out_iput; | |
1308 | ||
1309 | this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset); | |
1310 | err = fuse_copy_page(cs, &page, offset, this_num, 0); | |
1311 | if (!err && offset == 0 && (num != 0 || file_size == end)) | |
1312 | SetPageUptodate(page); | |
1313 | unlock_page(page); | |
1314 | page_cache_release(page); | |
1315 | ||
1316 | if (err) | |
1317 | goto out_iput; | |
1318 | ||
1319 | num -= this_num; | |
1320 | offset = 0; | |
1321 | index++; | |
1322 | } | |
1323 | ||
1324 | err = 0; | |
1325 | ||
1326 | out_iput: | |
1327 | iput(inode); | |
1328 | out_up_killsb: | |
1329 | up_read(&fc->killsb); | |
1330 | out_finish: | |
1331 | fuse_copy_finish(cs); | |
1332 | return err; | |
1333 | } | |
1334 | ||
2d45ba38 MS |
1335 | static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req) |
1336 | { | |
1337 | int i; | |
1338 | ||
1339 | for (i = 0; i < req->num_pages; i++) { | |
1340 | struct page *page = req->pages[i]; | |
1341 | page_cache_release(page); | |
1342 | } | |
1343 | } | |
1344 | ||
1345 | static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode, | |
1346 | struct fuse_notify_retrieve_out *outarg) | |
1347 | { | |
1348 | int err; | |
1349 | struct address_space *mapping = inode->i_mapping; | |
1350 | struct fuse_req *req; | |
1351 | pgoff_t index; | |
1352 | loff_t file_size; | |
1353 | unsigned int num; | |
1354 | unsigned int offset; | |
0157443c | 1355 | size_t total_len = 0; |
2d45ba38 MS |
1356 | |
1357 | req = fuse_get_req(fc); | |
1358 | if (IS_ERR(req)) | |
1359 | return PTR_ERR(req); | |
1360 | ||
1361 | offset = outarg->offset & ~PAGE_CACHE_MASK; | |
1362 | ||
1363 | req->in.h.opcode = FUSE_NOTIFY_REPLY; | |
1364 | req->in.h.nodeid = outarg->nodeid; | |
1365 | req->in.numargs = 2; | |
1366 | req->in.argpages = 1; | |
1367 | req->page_offset = offset; | |
1368 | req->end = fuse_retrieve_end; | |
1369 | ||
1370 | index = outarg->offset >> PAGE_CACHE_SHIFT; | |
1371 | file_size = i_size_read(inode); | |
1372 | num = outarg->size; | |
1373 | if (outarg->offset > file_size) | |
1374 | num = 0; | |
1375 | else if (outarg->offset + num > file_size) | |
1376 | num = file_size - outarg->offset; | |
1377 | ||
1378 | while (num) { | |
1379 | struct page *page; | |
1380 | unsigned int this_num; | |
1381 | ||
1382 | page = find_get_page(mapping, index); | |
1383 | if (!page) | |
1384 | break; | |
1385 | ||
1386 | this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset); | |
1387 | req->pages[req->num_pages] = page; | |
1388 | req->num_pages++; | |
1389 | ||
1390 | num -= this_num; | |
1391 | total_len += this_num; | |
1392 | } | |
1393 | req->misc.retrieve_in.offset = outarg->offset; | |
1394 | req->misc.retrieve_in.size = total_len; | |
1395 | req->in.args[0].size = sizeof(req->misc.retrieve_in); | |
1396 | req->in.args[0].value = &req->misc.retrieve_in; | |
1397 | req->in.args[1].size = total_len; | |
1398 | ||
1399 | err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique); | |
1400 | if (err) | |
1401 | fuse_retrieve_end(fc, req); | |
1402 | ||
1403 | return err; | |
1404 | } | |
1405 | ||
1406 | static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size, | |
1407 | struct fuse_copy_state *cs) | |
1408 | { | |
1409 | struct fuse_notify_retrieve_out outarg; | |
1410 | struct inode *inode; | |
1411 | int err; | |
1412 | ||
1413 | err = -EINVAL; | |
1414 | if (size != sizeof(outarg)) | |
1415 | goto copy_finish; | |
1416 | ||
1417 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1418 | if (err) | |
1419 | goto copy_finish; | |
1420 | ||
1421 | fuse_copy_finish(cs); | |
1422 | ||
1423 | down_read(&fc->killsb); | |
1424 | err = -ENOENT; | |
1425 | if (fc->sb) { | |
1426 | u64 nodeid = outarg.nodeid; | |
1427 | ||
1428 | inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid); | |
1429 | if (inode) { | |
1430 | err = fuse_retrieve(fc, inode, &outarg); | |
1431 | iput(inode); | |
1432 | } | |
1433 | } | |
1434 | up_read(&fc->killsb); | |
1435 | ||
1436 | return err; | |
1437 | ||
1438 | copy_finish: | |
1439 | fuse_copy_finish(cs); | |
1440 | return err; | |
1441 | } | |
1442 | ||
8599396b TH |
1443 | static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code, |
1444 | unsigned int size, struct fuse_copy_state *cs) | |
1445 | { | |
1446 | switch (code) { | |
95668a69 TH |
1447 | case FUSE_NOTIFY_POLL: |
1448 | return fuse_notify_poll(fc, size, cs); | |
1449 | ||
3b463ae0 JM |
1450 | case FUSE_NOTIFY_INVAL_INODE: |
1451 | return fuse_notify_inval_inode(fc, size, cs); | |
1452 | ||
1453 | case FUSE_NOTIFY_INVAL_ENTRY: | |
1454 | return fuse_notify_inval_entry(fc, size, cs); | |
1455 | ||
a1d75f25 MS |
1456 | case FUSE_NOTIFY_STORE: |
1457 | return fuse_notify_store(fc, size, cs); | |
1458 | ||
2d45ba38 MS |
1459 | case FUSE_NOTIFY_RETRIEVE: |
1460 | return fuse_notify_retrieve(fc, size, cs); | |
1461 | ||
8599396b | 1462 | default: |
f6d47a17 | 1463 | fuse_copy_finish(cs); |
8599396b TH |
1464 | return -EINVAL; |
1465 | } | |
1466 | } | |
1467 | ||
334f485d MS |
1468 | /* Look up request on processing list by unique ID */ |
1469 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
1470 | { | |
1471 | struct list_head *entry; | |
1472 | ||
1473 | list_for_each(entry, &fc->processing) { | |
1474 | struct fuse_req *req; | |
1475 | req = list_entry(entry, struct fuse_req, list); | |
a4d27e75 | 1476 | if (req->in.h.unique == unique || req->intr_unique == unique) |
334f485d MS |
1477 | return req; |
1478 | } | |
1479 | return NULL; | |
1480 | } | |
1481 | ||
1482 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
1483 | unsigned nbytes) | |
1484 | { | |
1485 | unsigned reqsize = sizeof(struct fuse_out_header); | |
1486 | ||
1487 | if (out->h.error) | |
1488 | return nbytes != reqsize ? -EINVAL : 0; | |
1489 | ||
1490 | reqsize += len_args(out->numargs, out->args); | |
1491 | ||
1492 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
1493 | return -EINVAL; | |
1494 | else if (reqsize > nbytes) { | |
1495 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
1496 | unsigned diffsize = reqsize - nbytes; | |
1497 | if (diffsize > lastarg->size) | |
1498 | return -EINVAL; | |
1499 | lastarg->size -= diffsize; | |
1500 | } | |
1501 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
1502 | out->page_zeroing); | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Write a single reply to a request. First the header is copied from | |
1507 | * the write buffer. The request is then searched on the processing | |
1508 | * list by the unique ID found in the header. If found, then remove | |
1509 | * it from the list and copy the rest of the buffer to the request. | |
1510 | * The request is finished by calling request_end() | |
1511 | */ | |
dd3bb14f MS |
1512 | static ssize_t fuse_dev_do_write(struct fuse_conn *fc, |
1513 | struct fuse_copy_state *cs, size_t nbytes) | |
334f485d MS |
1514 | { |
1515 | int err; | |
334f485d MS |
1516 | struct fuse_req *req; |
1517 | struct fuse_out_header oh; | |
334f485d | 1518 | |
334f485d MS |
1519 | if (nbytes < sizeof(struct fuse_out_header)) |
1520 | return -EINVAL; | |
1521 | ||
dd3bb14f | 1522 | err = fuse_copy_one(cs, &oh, sizeof(oh)); |
334f485d MS |
1523 | if (err) |
1524 | goto err_finish; | |
8599396b TH |
1525 | |
1526 | err = -EINVAL; | |
1527 | if (oh.len != nbytes) | |
1528 | goto err_finish; | |
1529 | ||
1530 | /* | |
1531 | * Zero oh.unique indicates unsolicited notification message | |
1532 | * and error contains notification code. | |
1533 | */ | |
1534 | if (!oh.unique) { | |
dd3bb14f | 1535 | err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs); |
8599396b TH |
1536 | return err ? err : nbytes; |
1537 | } | |
1538 | ||
334f485d | 1539 | err = -EINVAL; |
8599396b | 1540 | if (oh.error <= -1000 || oh.error > 0) |
334f485d MS |
1541 | goto err_finish; |
1542 | ||
d7133114 | 1543 | spin_lock(&fc->lock); |
69a53bf2 MS |
1544 | err = -ENOENT; |
1545 | if (!fc->connected) | |
1546 | goto err_unlock; | |
1547 | ||
334f485d | 1548 | req = request_find(fc, oh.unique); |
334f485d MS |
1549 | if (!req) |
1550 | goto err_unlock; | |
1551 | ||
f9a2842e | 1552 | if (req->aborted) { |
d7133114 | 1553 | spin_unlock(&fc->lock); |
dd3bb14f | 1554 | fuse_copy_finish(cs); |
d7133114 | 1555 | spin_lock(&fc->lock); |
222f1d69 | 1556 | request_end(fc, req); |
334f485d MS |
1557 | return -ENOENT; |
1558 | } | |
a4d27e75 MS |
1559 | /* Is it an interrupt reply? */ |
1560 | if (req->intr_unique == oh.unique) { | |
1561 | err = -EINVAL; | |
1562 | if (nbytes != sizeof(struct fuse_out_header)) | |
1563 | goto err_unlock; | |
1564 | ||
1565 | if (oh.error == -ENOSYS) | |
1566 | fc->no_interrupt = 1; | |
1567 | else if (oh.error == -EAGAIN) | |
1568 | queue_interrupt(fc, req); | |
1569 | ||
1570 | spin_unlock(&fc->lock); | |
dd3bb14f | 1571 | fuse_copy_finish(cs); |
a4d27e75 MS |
1572 | return nbytes; |
1573 | } | |
1574 | ||
1575 | req->state = FUSE_REQ_WRITING; | |
d77a1d5b | 1576 | list_move(&req->list, &fc->io); |
334f485d MS |
1577 | req->out.h = oh; |
1578 | req->locked = 1; | |
dd3bb14f | 1579 | cs->req = req; |
ce534fb0 MS |
1580 | if (!req->out.page_replace) |
1581 | cs->move_pages = 0; | |
d7133114 | 1582 | spin_unlock(&fc->lock); |
334f485d | 1583 | |
dd3bb14f MS |
1584 | err = copy_out_args(cs, &req->out, nbytes); |
1585 | fuse_copy_finish(cs); | |
334f485d | 1586 | |
d7133114 | 1587 | spin_lock(&fc->lock); |
334f485d MS |
1588 | req->locked = 0; |
1589 | if (!err) { | |
f9a2842e | 1590 | if (req->aborted) |
334f485d | 1591 | err = -ENOENT; |
f9a2842e | 1592 | } else if (!req->aborted) |
334f485d MS |
1593 | req->out.h.error = -EIO; |
1594 | request_end(fc, req); | |
1595 | ||
1596 | return err ? err : nbytes; | |
1597 | ||
1598 | err_unlock: | |
d7133114 | 1599 | spin_unlock(&fc->lock); |
334f485d | 1600 | err_finish: |
dd3bb14f | 1601 | fuse_copy_finish(cs); |
334f485d MS |
1602 | return err; |
1603 | } | |
1604 | ||
dd3bb14f MS |
1605 | static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov, |
1606 | unsigned long nr_segs, loff_t pos) | |
1607 | { | |
1608 | struct fuse_copy_state cs; | |
1609 | struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp); | |
1610 | if (!fc) | |
1611 | return -EPERM; | |
1612 | ||
c3021629 | 1613 | fuse_copy_init(&cs, fc, 0, iov, nr_segs); |
dd3bb14f MS |
1614 | |
1615 | return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs)); | |
1616 | } | |
1617 | ||
1618 | static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe, | |
1619 | struct file *out, loff_t *ppos, | |
1620 | size_t len, unsigned int flags) | |
1621 | { | |
1622 | unsigned nbuf; | |
1623 | unsigned idx; | |
1624 | struct pipe_buffer *bufs; | |
1625 | struct fuse_copy_state cs; | |
1626 | struct fuse_conn *fc; | |
1627 | size_t rem; | |
1628 | ssize_t ret; | |
1629 | ||
1630 | fc = fuse_get_conn(out); | |
1631 | if (!fc) | |
1632 | return -EPERM; | |
1633 | ||
1634 | bufs = kmalloc(pipe->buffers * sizeof (struct pipe_buffer), GFP_KERNEL); | |
1635 | if (!bufs) | |
1636 | return -ENOMEM; | |
1637 | ||
1638 | pipe_lock(pipe); | |
1639 | nbuf = 0; | |
1640 | rem = 0; | |
1641 | for (idx = 0; idx < pipe->nrbufs && rem < len; idx++) | |
1642 | rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len; | |
1643 | ||
1644 | ret = -EINVAL; | |
1645 | if (rem < len) { | |
1646 | pipe_unlock(pipe); | |
1647 | goto out; | |
1648 | } | |
1649 | ||
1650 | rem = len; | |
1651 | while (rem) { | |
1652 | struct pipe_buffer *ibuf; | |
1653 | struct pipe_buffer *obuf; | |
1654 | ||
1655 | BUG_ON(nbuf >= pipe->buffers); | |
1656 | BUG_ON(!pipe->nrbufs); | |
1657 | ibuf = &pipe->bufs[pipe->curbuf]; | |
1658 | obuf = &bufs[nbuf]; | |
1659 | ||
1660 | if (rem >= ibuf->len) { | |
1661 | *obuf = *ibuf; | |
1662 | ibuf->ops = NULL; | |
1663 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); | |
1664 | pipe->nrbufs--; | |
1665 | } else { | |
1666 | ibuf->ops->get(pipe, ibuf); | |
1667 | *obuf = *ibuf; | |
1668 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1669 | obuf->len = rem; | |
1670 | ibuf->offset += obuf->len; | |
1671 | ibuf->len -= obuf->len; | |
1672 | } | |
1673 | nbuf++; | |
1674 | rem -= obuf->len; | |
1675 | } | |
1676 | pipe_unlock(pipe); | |
1677 | ||
c3021629 | 1678 | fuse_copy_init(&cs, fc, 0, NULL, nbuf); |
dd3bb14f | 1679 | cs.pipebufs = bufs; |
dd3bb14f MS |
1680 | cs.pipe = pipe; |
1681 | ||
ce534fb0 MS |
1682 | if (flags & SPLICE_F_MOVE) |
1683 | cs.move_pages = 1; | |
1684 | ||
dd3bb14f MS |
1685 | ret = fuse_dev_do_write(fc, &cs, len); |
1686 | ||
1687 | for (idx = 0; idx < nbuf; idx++) { | |
1688 | struct pipe_buffer *buf = &bufs[idx]; | |
1689 | buf->ops->release(pipe, buf); | |
1690 | } | |
1691 | out: | |
1692 | kfree(bufs); | |
1693 | return ret; | |
1694 | } | |
1695 | ||
334f485d MS |
1696 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) |
1697 | { | |
334f485d | 1698 | unsigned mask = POLLOUT | POLLWRNORM; |
7025d9ad | 1699 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 1700 | if (!fc) |
7025d9ad | 1701 | return POLLERR; |
334f485d MS |
1702 | |
1703 | poll_wait(file, &fc->waitq, wait); | |
1704 | ||
d7133114 | 1705 | spin_lock(&fc->lock); |
7025d9ad MS |
1706 | if (!fc->connected) |
1707 | mask = POLLERR; | |
a4d27e75 | 1708 | else if (request_pending(fc)) |
7025d9ad | 1709 | mask |= POLLIN | POLLRDNORM; |
d7133114 | 1710 | spin_unlock(&fc->lock); |
334f485d MS |
1711 | |
1712 | return mask; | |
1713 | } | |
1714 | ||
69a53bf2 MS |
1715 | /* |
1716 | * Abort all requests on the given list (pending or processing) | |
1717 | * | |
d7133114 | 1718 | * This function releases and reacquires fc->lock |
69a53bf2 | 1719 | */ |
334f485d | 1720 | static void end_requests(struct fuse_conn *fc, struct list_head *head) |
b9ca67b2 MS |
1721 | __releases(fc->lock) |
1722 | __acquires(fc->lock) | |
334f485d MS |
1723 | { |
1724 | while (!list_empty(head)) { | |
1725 | struct fuse_req *req; | |
1726 | req = list_entry(head->next, struct fuse_req, list); | |
334f485d MS |
1727 | req->out.h.error = -ECONNABORTED; |
1728 | request_end(fc, req); | |
d7133114 | 1729 | spin_lock(&fc->lock); |
334f485d MS |
1730 | } |
1731 | } | |
1732 | ||
69a53bf2 MS |
1733 | /* |
1734 | * Abort requests under I/O | |
1735 | * | |
f9a2842e | 1736 | * The requests are set to aborted and finished, and the request |
69a53bf2 MS |
1737 | * waiter is woken up. This will make request_wait_answer() wait |
1738 | * until the request is unlocked and then return. | |
64c6d8ed MS |
1739 | * |
1740 | * If the request is asynchronous, then the end function needs to be | |
1741 | * called after waiting for the request to be unlocked (if it was | |
1742 | * locked). | |
69a53bf2 MS |
1743 | */ |
1744 | static void end_io_requests(struct fuse_conn *fc) | |
b9ca67b2 MS |
1745 | __releases(fc->lock) |
1746 | __acquires(fc->lock) | |
69a53bf2 MS |
1747 | { |
1748 | while (!list_empty(&fc->io)) { | |
64c6d8ed MS |
1749 | struct fuse_req *req = |
1750 | list_entry(fc->io.next, struct fuse_req, list); | |
1751 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
1752 | ||
f9a2842e | 1753 | req->aborted = 1; |
69a53bf2 MS |
1754 | req->out.h.error = -ECONNABORTED; |
1755 | req->state = FUSE_REQ_FINISHED; | |
1756 | list_del_init(&req->list); | |
1757 | wake_up(&req->waitq); | |
64c6d8ed MS |
1758 | if (end) { |
1759 | req->end = NULL; | |
64c6d8ed | 1760 | __fuse_get_request(req); |
d7133114 | 1761 | spin_unlock(&fc->lock); |
64c6d8ed MS |
1762 | wait_event(req->waitq, !req->locked); |
1763 | end(fc, req); | |
e9bb09dd | 1764 | fuse_put_request(fc, req); |
d7133114 | 1765 | spin_lock(&fc->lock); |
64c6d8ed | 1766 | } |
69a53bf2 MS |
1767 | } |
1768 | } | |
1769 | ||
595afaf9 | 1770 | static void end_queued_requests(struct fuse_conn *fc) |
b9ca67b2 MS |
1771 | __releases(fc->lock) |
1772 | __acquires(fc->lock) | |
595afaf9 MS |
1773 | { |
1774 | fc->max_background = UINT_MAX; | |
1775 | flush_bg_queue(fc); | |
1776 | end_requests(fc, &fc->pending); | |
1777 | end_requests(fc, &fc->processing); | |
1778 | } | |
1779 | ||
69a53bf2 MS |
1780 | /* |
1781 | * Abort all requests. | |
1782 | * | |
1783 | * Emergency exit in case of a malicious or accidental deadlock, or | |
1784 | * just a hung filesystem. | |
1785 | * | |
1786 | * The same effect is usually achievable through killing the | |
1787 | * filesystem daemon and all users of the filesystem. The exception | |
1788 | * is the combination of an asynchronous request and the tricky | |
1789 | * deadlock (see Documentation/filesystems/fuse.txt). | |
1790 | * | |
1791 | * During the aborting, progression of requests from the pending and | |
1792 | * processing lists onto the io list, and progression of new requests | |
1793 | * onto the pending list is prevented by req->connected being false. | |
1794 | * | |
1795 | * Progression of requests under I/O to the processing list is | |
f9a2842e MS |
1796 | * prevented by the req->aborted flag being true for these requests. |
1797 | * For this reason requests on the io list must be aborted first. | |
69a53bf2 MS |
1798 | */ |
1799 | void fuse_abort_conn(struct fuse_conn *fc) | |
1800 | { | |
d7133114 | 1801 | spin_lock(&fc->lock); |
69a53bf2 MS |
1802 | if (fc->connected) { |
1803 | fc->connected = 0; | |
51eb01e7 | 1804 | fc->blocked = 0; |
69a53bf2 | 1805 | end_io_requests(fc); |
595afaf9 | 1806 | end_queued_requests(fc); |
69a53bf2 | 1807 | wake_up_all(&fc->waitq); |
51eb01e7 | 1808 | wake_up_all(&fc->blocked_waitq); |
385a17bf | 1809 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
69a53bf2 | 1810 | } |
d7133114 | 1811 | spin_unlock(&fc->lock); |
69a53bf2 | 1812 | } |
08cbf542 | 1813 | EXPORT_SYMBOL_GPL(fuse_abort_conn); |
69a53bf2 | 1814 | |
08cbf542 | 1815 | int fuse_dev_release(struct inode *inode, struct file *file) |
334f485d | 1816 | { |
0720b315 | 1817 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 1818 | if (fc) { |
d7133114 | 1819 | spin_lock(&fc->lock); |
1e9a4ed9 | 1820 | fc->connected = 0; |
595afaf9 MS |
1821 | fc->blocked = 0; |
1822 | end_queued_requests(fc); | |
1823 | wake_up_all(&fc->blocked_waitq); | |
d7133114 | 1824 | spin_unlock(&fc->lock); |
bafa9654 | 1825 | fuse_conn_put(fc); |
385a17bf | 1826 | } |
f543f253 | 1827 | |
334f485d MS |
1828 | return 0; |
1829 | } | |
08cbf542 | 1830 | EXPORT_SYMBOL_GPL(fuse_dev_release); |
334f485d | 1831 | |
385a17bf JD |
1832 | static int fuse_dev_fasync(int fd, struct file *file, int on) |
1833 | { | |
1834 | struct fuse_conn *fc = fuse_get_conn(file); | |
1835 | if (!fc) | |
a87046d8 | 1836 | return -EPERM; |
385a17bf JD |
1837 | |
1838 | /* No locking - fasync_helper does its own locking */ | |
1839 | return fasync_helper(fd, file, on, &fc->fasync); | |
1840 | } | |
1841 | ||
4b6f5d20 | 1842 | const struct file_operations fuse_dev_operations = { |
334f485d MS |
1843 | .owner = THIS_MODULE, |
1844 | .llseek = no_llseek, | |
ee0b3e67 BP |
1845 | .read = do_sync_read, |
1846 | .aio_read = fuse_dev_read, | |
c3021629 | 1847 | .splice_read = fuse_dev_splice_read, |
ee0b3e67 BP |
1848 | .write = do_sync_write, |
1849 | .aio_write = fuse_dev_write, | |
dd3bb14f | 1850 | .splice_write = fuse_dev_splice_write, |
334f485d MS |
1851 | .poll = fuse_dev_poll, |
1852 | .release = fuse_dev_release, | |
385a17bf | 1853 | .fasync = fuse_dev_fasync, |
334f485d | 1854 | }; |
08cbf542 | 1855 | EXPORT_SYMBOL_GPL(fuse_dev_operations); |
334f485d MS |
1856 | |
1857 | static struct miscdevice fuse_miscdevice = { | |
1858 | .minor = FUSE_MINOR, | |
1859 | .name = "fuse", | |
1860 | .fops = &fuse_dev_operations, | |
1861 | }; | |
1862 | ||
1863 | int __init fuse_dev_init(void) | |
1864 | { | |
1865 | int err = -ENOMEM; | |
1866 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
1867 | sizeof(struct fuse_req), | |
20c2df83 | 1868 | 0, 0, NULL); |
334f485d MS |
1869 | if (!fuse_req_cachep) |
1870 | goto out; | |
1871 | ||
1872 | err = misc_register(&fuse_miscdevice); | |
1873 | if (err) | |
1874 | goto out_cache_clean; | |
1875 | ||
1876 | return 0; | |
1877 | ||
1878 | out_cache_clean: | |
1879 | kmem_cache_destroy(fuse_req_cachep); | |
1880 | out: | |
1881 | return err; | |
1882 | } | |
1883 | ||
1884 | void fuse_dev_cleanup(void) | |
1885 | { | |
1886 | misc_deregister(&fuse_miscdevice); | |
1887 | kmem_cache_destroy(fuse_req_cachep); | |
1888 | } |