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 | ||
4250c066 MP |
37 | static void fuse_request_init(struct fuse_req *req, struct page **pages, |
38 | unsigned npages) | |
334f485d MS |
39 | { |
40 | memset(req, 0, sizeof(*req)); | |
4250c066 | 41 | memset(pages, 0, sizeof(*pages) * npages); |
334f485d | 42 | INIT_LIST_HEAD(&req->list); |
a4d27e75 | 43 | INIT_LIST_HEAD(&req->intr_entry); |
334f485d MS |
44 | init_waitqueue_head(&req->waitq); |
45 | atomic_set(&req->count, 1); | |
4250c066 MP |
46 | req->pages = pages; |
47 | req->max_pages = npages; | |
334f485d MS |
48 | } |
49 | ||
4250c066 | 50 | static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags) |
334f485d | 51 | { |
4250c066 MP |
52 | struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags); |
53 | if (req) { | |
54 | struct page **pages; | |
55 | ||
56 | if (npages <= FUSE_REQ_INLINE_PAGES) | |
57 | pages = req->inline_pages; | |
58 | else | |
59 | pages = kmalloc(sizeof(struct page *) * npages, flags); | |
60 | ||
61 | if (!pages) { | |
62 | kmem_cache_free(fuse_req_cachep, req); | |
63 | return NULL; | |
64 | } | |
65 | ||
66 | fuse_request_init(req, pages, npages); | |
67 | } | |
334f485d MS |
68 | return req; |
69 | } | |
4250c066 MP |
70 | |
71 | struct fuse_req *fuse_request_alloc(unsigned npages) | |
72 | { | |
73 | return __fuse_request_alloc(npages, GFP_KERNEL); | |
74 | } | |
08cbf542 | 75 | EXPORT_SYMBOL_GPL(fuse_request_alloc); |
334f485d | 76 | |
4250c066 | 77 | struct fuse_req *fuse_request_alloc_nofs(unsigned npages) |
3be5a52b | 78 | { |
4250c066 | 79 | return __fuse_request_alloc(npages, GFP_NOFS); |
3be5a52b MS |
80 | } |
81 | ||
334f485d MS |
82 | void fuse_request_free(struct fuse_req *req) |
83 | { | |
4250c066 MP |
84 | if (req->pages != req->inline_pages) |
85 | kfree(req->pages); | |
334f485d MS |
86 | kmem_cache_free(fuse_req_cachep, req); |
87 | } | |
88 | ||
8bfc016d | 89 | static void block_sigs(sigset_t *oldset) |
334f485d MS |
90 | { |
91 | sigset_t mask; | |
92 | ||
93 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
94 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
95 | } | |
96 | ||
8bfc016d | 97 | static void restore_sigs(sigset_t *oldset) |
334f485d MS |
98 | { |
99 | sigprocmask(SIG_SETMASK, oldset, NULL); | |
100 | } | |
101 | ||
334f485d MS |
102 | static void __fuse_get_request(struct fuse_req *req) |
103 | { | |
104 | atomic_inc(&req->count); | |
105 | } | |
106 | ||
107 | /* Must be called with > 1 refcount */ | |
108 | static void __fuse_put_request(struct fuse_req *req) | |
109 | { | |
110 | BUG_ON(atomic_read(&req->count) < 2); | |
111 | atomic_dec(&req->count); | |
112 | } | |
113 | ||
33649c91 MS |
114 | static 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 | ||
ce1d5a49 | 121 | struct fuse_req *fuse_get_req(struct fuse_conn *fc) |
334f485d | 122 | { |
08a53cdc MS |
123 | struct fuse_req *req; |
124 | sigset_t oldset; | |
9bc5ddda | 125 | int intr; |
08a53cdc MS |
126 | int err; |
127 | ||
9bc5ddda | 128 | atomic_inc(&fc->num_waiting); |
08a53cdc | 129 | block_sigs(&oldset); |
9bc5ddda | 130 | intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked); |
08a53cdc | 131 | restore_sigs(&oldset); |
9bc5ddda MS |
132 | err = -EINTR; |
133 | if (intr) | |
134 | goto out; | |
08a53cdc | 135 | |
51eb01e7 MS |
136 | err = -ENOTCONN; |
137 | if (!fc->connected) | |
138 | goto out; | |
139 | ||
4250c066 | 140 | req = fuse_request_alloc(FUSE_MAX_PAGES_PER_REQ); |
9bc5ddda | 141 | err = -ENOMEM; |
ce1d5a49 | 142 | if (!req) |
9bc5ddda | 143 | goto out; |
334f485d | 144 | |
33649c91 | 145 | fuse_req_init_context(req); |
9bc5ddda | 146 | req->waiting = 1; |
334f485d | 147 | return req; |
9bc5ddda MS |
148 | |
149 | out: | |
150 | atomic_dec(&fc->num_waiting); | |
151 | return ERR_PTR(err); | |
334f485d | 152 | } |
08cbf542 | 153 | EXPORT_SYMBOL_GPL(fuse_get_req); |
334f485d | 154 | |
33649c91 MS |
155 | /* |
156 | * Return request in fuse_file->reserved_req. However that may | |
157 | * currently be in use. If that is the case, wait for it to become | |
158 | * available. | |
159 | */ | |
160 | static struct fuse_req *get_reserved_req(struct fuse_conn *fc, | |
161 | struct file *file) | |
162 | { | |
163 | struct fuse_req *req = NULL; | |
164 | struct fuse_file *ff = file->private_data; | |
165 | ||
166 | do { | |
de5e3dec | 167 | wait_event(fc->reserved_req_waitq, ff->reserved_req); |
33649c91 MS |
168 | spin_lock(&fc->lock); |
169 | if (ff->reserved_req) { | |
170 | req = ff->reserved_req; | |
171 | ff->reserved_req = NULL; | |
cb0942b8 | 172 | req->stolen_file = get_file(file); |
33649c91 MS |
173 | } |
174 | spin_unlock(&fc->lock); | |
175 | } while (!req); | |
176 | ||
177 | return req; | |
178 | } | |
179 | ||
180 | /* | |
181 | * Put stolen request back into fuse_file->reserved_req | |
182 | */ | |
183 | static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req) | |
184 | { | |
185 | struct file *file = req->stolen_file; | |
186 | struct fuse_file *ff = file->private_data; | |
187 | ||
188 | spin_lock(&fc->lock); | |
4250c066 | 189 | fuse_request_init(req, req->pages, req->max_pages); |
33649c91 MS |
190 | BUG_ON(ff->reserved_req); |
191 | ff->reserved_req = req; | |
de5e3dec | 192 | wake_up_all(&fc->reserved_req_waitq); |
33649c91 MS |
193 | spin_unlock(&fc->lock); |
194 | fput(file); | |
195 | } | |
196 | ||
197 | /* | |
198 | * Gets a requests for a file operation, always succeeds | |
199 | * | |
200 | * This is used for sending the FLUSH request, which must get to | |
201 | * userspace, due to POSIX locks which may need to be unlocked. | |
202 | * | |
203 | * If allocation fails due to OOM, use the reserved request in | |
204 | * fuse_file. | |
205 | * | |
206 | * This is very unlikely to deadlock accidentally, since the | |
207 | * filesystem should not have it's own file open. If deadlock is | |
208 | * intentional, it can still be broken by "aborting" the filesystem. | |
209 | */ | |
210 | struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file) | |
211 | { | |
212 | struct fuse_req *req; | |
213 | ||
214 | atomic_inc(&fc->num_waiting); | |
215 | wait_event(fc->blocked_waitq, !fc->blocked); | |
4250c066 | 216 | req = fuse_request_alloc(FUSE_MAX_PAGES_PER_REQ); |
33649c91 MS |
217 | if (!req) |
218 | req = get_reserved_req(fc, file); | |
219 | ||
220 | fuse_req_init_context(req); | |
221 | req->waiting = 1; | |
222 | return req; | |
223 | } | |
224 | ||
334f485d | 225 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) |
7128ec2a MS |
226 | { |
227 | if (atomic_dec_and_test(&req->count)) { | |
9bc5ddda MS |
228 | if (req->waiting) |
229 | atomic_dec(&fc->num_waiting); | |
33649c91 MS |
230 | |
231 | if (req->stolen_file) | |
232 | put_reserved_req(fc, req); | |
233 | else | |
234 | fuse_request_free(req); | |
7128ec2a MS |
235 | } |
236 | } | |
08cbf542 | 237 | EXPORT_SYMBOL_GPL(fuse_put_request); |
7128ec2a | 238 | |
d12def1b MS |
239 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) |
240 | { | |
241 | unsigned nbytes = 0; | |
242 | unsigned i; | |
243 | ||
244 | for (i = 0; i < numargs; i++) | |
245 | nbytes += args[i].size; | |
246 | ||
247 | return nbytes; | |
248 | } | |
249 | ||
250 | static u64 fuse_get_unique(struct fuse_conn *fc) | |
251 | { | |
252 | fc->reqctr++; | |
253 | /* zero is special */ | |
254 | if (fc->reqctr == 0) | |
255 | fc->reqctr = 1; | |
256 | ||
257 | return fc->reqctr; | |
258 | } | |
259 | ||
260 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) | |
261 | { | |
d12def1b MS |
262 | req->in.h.len = sizeof(struct fuse_in_header) + |
263 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
264 | list_add_tail(&req->list, &fc->pending); | |
265 | req->state = FUSE_REQ_PENDING; | |
266 | if (!req->waiting) { | |
267 | req->waiting = 1; | |
268 | atomic_inc(&fc->num_waiting); | |
269 | } | |
270 | wake_up(&fc->waitq); | |
271 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); | |
272 | } | |
273 | ||
07e77dca MS |
274 | void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget, |
275 | u64 nodeid, u64 nlookup) | |
276 | { | |
02c048b9 MS |
277 | forget->forget_one.nodeid = nodeid; |
278 | forget->forget_one.nlookup = nlookup; | |
07e77dca MS |
279 | |
280 | spin_lock(&fc->lock); | |
5dfcc87f MS |
281 | if (fc->connected) { |
282 | fc->forget_list_tail->next = forget; | |
283 | fc->forget_list_tail = forget; | |
284 | wake_up(&fc->waitq); | |
285 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); | |
286 | } else { | |
287 | kfree(forget); | |
288 | } | |
07e77dca MS |
289 | spin_unlock(&fc->lock); |
290 | } | |
291 | ||
d12def1b MS |
292 | static void flush_bg_queue(struct fuse_conn *fc) |
293 | { | |
7a6d3c8b | 294 | while (fc->active_background < fc->max_background && |
d12def1b MS |
295 | !list_empty(&fc->bg_queue)) { |
296 | struct fuse_req *req; | |
297 | ||
298 | req = list_entry(fc->bg_queue.next, struct fuse_req, list); | |
299 | list_del(&req->list); | |
300 | fc->active_background++; | |
2d45ba38 | 301 | req->in.h.unique = fuse_get_unique(fc); |
d12def1b MS |
302 | queue_request(fc, req); |
303 | } | |
304 | } | |
305 | ||
334f485d MS |
306 | /* |
307 | * This function is called when a request is finished. Either a reply | |
f9a2842e | 308 | * has arrived or it was aborted (and not yet sent) or some error |
f43b155a | 309 | * occurred during communication with userspace, or the device file |
51eb01e7 MS |
310 | * was closed. The requester thread is woken up (if still waiting), |
311 | * the 'end' callback is called if given, else the reference to the | |
312 | * request is released | |
7128ec2a | 313 | * |
d7133114 | 314 | * Called with fc->lock, unlocks it |
334f485d MS |
315 | */ |
316 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
b9ca67b2 | 317 | __releases(fc->lock) |
334f485d | 318 | { |
51eb01e7 MS |
319 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; |
320 | req->end = NULL; | |
d77a1d5b | 321 | list_del(&req->list); |
a4d27e75 | 322 | list_del(&req->intr_entry); |
83cfd493 | 323 | req->state = FUSE_REQ_FINISHED; |
51eb01e7 | 324 | if (req->background) { |
7a6d3c8b | 325 | if (fc->num_background == fc->max_background) { |
51eb01e7 MS |
326 | fc->blocked = 0; |
327 | wake_up_all(&fc->blocked_waitq); | |
328 | } | |
7a6d3c8b | 329 | if (fc->num_background == fc->congestion_threshold && |
a325f9b9 | 330 | fc->connected && fc->bdi_initialized) { |
8aa7e847 JA |
331 | clear_bdi_congested(&fc->bdi, BLK_RW_SYNC); |
332 | clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC); | |
f92b99b9 | 333 | } |
51eb01e7 | 334 | fc->num_background--; |
d12def1b MS |
335 | fc->active_background--; |
336 | flush_bg_queue(fc); | |
334f485d | 337 | } |
51eb01e7 | 338 | spin_unlock(&fc->lock); |
51eb01e7 MS |
339 | wake_up(&req->waitq); |
340 | if (end) | |
341 | end(fc, req); | |
e9bb09dd | 342 | fuse_put_request(fc, req); |
334f485d MS |
343 | } |
344 | ||
a4d27e75 MS |
345 | static void wait_answer_interruptible(struct fuse_conn *fc, |
346 | struct fuse_req *req) | |
b9ca67b2 MS |
347 | __releases(fc->lock) |
348 | __acquires(fc->lock) | |
a4d27e75 MS |
349 | { |
350 | if (signal_pending(current)) | |
351 | return; | |
352 | ||
353 | spin_unlock(&fc->lock); | |
354 | wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED); | |
355 | spin_lock(&fc->lock); | |
356 | } | |
357 | ||
358 | static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req) | |
359 | { | |
360 | list_add_tail(&req->intr_entry, &fc->interrupts); | |
361 | wake_up(&fc->waitq); | |
362 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); | |
363 | } | |
364 | ||
7c352bdf | 365 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) |
b9ca67b2 MS |
366 | __releases(fc->lock) |
367 | __acquires(fc->lock) | |
334f485d | 368 | { |
a4d27e75 MS |
369 | if (!fc->no_interrupt) { |
370 | /* Any signal may interrupt this */ | |
371 | wait_answer_interruptible(fc, req); | |
334f485d | 372 | |
a4d27e75 MS |
373 | if (req->aborted) |
374 | goto aborted; | |
375 | if (req->state == FUSE_REQ_FINISHED) | |
376 | return; | |
377 | ||
378 | req->interrupted = 1; | |
379 | if (req->state == FUSE_REQ_SENT) | |
380 | queue_interrupt(fc, req); | |
381 | } | |
382 | ||
a131de0a | 383 | if (!req->force) { |
a4d27e75 MS |
384 | sigset_t oldset; |
385 | ||
386 | /* Only fatal signals may interrupt this */ | |
51eb01e7 | 387 | block_sigs(&oldset); |
a4d27e75 | 388 | wait_answer_interruptible(fc, req); |
51eb01e7 | 389 | restore_sigs(&oldset); |
a131de0a MS |
390 | |
391 | if (req->aborted) | |
392 | goto aborted; | |
393 | if (req->state == FUSE_REQ_FINISHED) | |
394 | return; | |
395 | ||
396 | /* Request is not yet in userspace, bail out */ | |
397 | if (req->state == FUSE_REQ_PENDING) { | |
398 | list_del(&req->list); | |
399 | __fuse_put_request(req); | |
400 | req->out.h.error = -EINTR; | |
401 | return; | |
402 | } | |
51eb01e7 | 403 | } |
334f485d | 404 | |
a131de0a MS |
405 | /* |
406 | * Either request is already in userspace, or it was forced. | |
407 | * Wait it out. | |
408 | */ | |
409 | spin_unlock(&fc->lock); | |
410 | wait_event(req->waitq, req->state == FUSE_REQ_FINISHED); | |
411 | spin_lock(&fc->lock); | |
a4d27e75 | 412 | |
a131de0a MS |
413 | if (!req->aborted) |
414 | return; | |
a4d27e75 MS |
415 | |
416 | aborted: | |
a131de0a | 417 | BUG_ON(req->state != FUSE_REQ_FINISHED); |
334f485d MS |
418 | if (req->locked) { |
419 | /* This is uninterruptible sleep, because data is | |
420 | being copied to/from the buffers of req. During | |
421 | locked state, there mustn't be any filesystem | |
422 | operation (e.g. page fault), since that could lead | |
423 | to deadlock */ | |
d7133114 | 424 | spin_unlock(&fc->lock); |
334f485d | 425 | wait_event(req->waitq, !req->locked); |
d7133114 | 426 | spin_lock(&fc->lock); |
334f485d | 427 | } |
334f485d MS |
428 | } |
429 | ||
b93f858a | 430 | void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
431 | { |
432 | req->isreply = 1; | |
d7133114 | 433 | spin_lock(&fc->lock); |
1e9a4ed9 | 434 | if (!fc->connected) |
334f485d MS |
435 | req->out.h.error = -ENOTCONN; |
436 | else if (fc->conn_error) | |
437 | req->out.h.error = -ECONNREFUSED; | |
438 | else { | |
2d45ba38 | 439 | req->in.h.unique = fuse_get_unique(fc); |
334f485d MS |
440 | queue_request(fc, req); |
441 | /* acquire extra reference, since request is still needed | |
442 | after request_end() */ | |
443 | __fuse_get_request(req); | |
444 | ||
7c352bdf | 445 | request_wait_answer(fc, req); |
334f485d | 446 | } |
d7133114 | 447 | spin_unlock(&fc->lock); |
334f485d | 448 | } |
08cbf542 | 449 | EXPORT_SYMBOL_GPL(fuse_request_send); |
334f485d | 450 | |
b93f858a TH |
451 | static void fuse_request_send_nowait_locked(struct fuse_conn *fc, |
452 | struct fuse_req *req) | |
d12def1b MS |
453 | { |
454 | req->background = 1; | |
455 | fc->num_background++; | |
7a6d3c8b | 456 | if (fc->num_background == fc->max_background) |
d12def1b | 457 | fc->blocked = 1; |
7a6d3c8b | 458 | if (fc->num_background == fc->congestion_threshold && |
a325f9b9 | 459 | fc->bdi_initialized) { |
8aa7e847 JA |
460 | set_bdi_congested(&fc->bdi, BLK_RW_SYNC); |
461 | set_bdi_congested(&fc->bdi, BLK_RW_ASYNC); | |
d12def1b MS |
462 | } |
463 | list_add_tail(&req->list, &fc->bg_queue); | |
464 | flush_bg_queue(fc); | |
465 | } | |
466 | ||
b93f858a | 467 | static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
334f485d | 468 | { |
d7133114 | 469 | spin_lock(&fc->lock); |
1e9a4ed9 | 470 | if (fc->connected) { |
b93f858a | 471 | fuse_request_send_nowait_locked(fc, req); |
d7133114 | 472 | spin_unlock(&fc->lock); |
334f485d MS |
473 | } else { |
474 | req->out.h.error = -ENOTCONN; | |
475 | request_end(fc, req); | |
476 | } | |
477 | } | |
478 | ||
b93f858a | 479 | void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
480 | { |
481 | req->isreply = 1; | |
b93f858a | 482 | fuse_request_send_nowait(fc, req); |
334f485d | 483 | } |
08cbf542 | 484 | EXPORT_SYMBOL_GPL(fuse_request_send_background); |
334f485d | 485 | |
2d45ba38 MS |
486 | static int fuse_request_send_notify_reply(struct fuse_conn *fc, |
487 | struct fuse_req *req, u64 unique) | |
488 | { | |
489 | int err = -ENODEV; | |
490 | ||
491 | req->isreply = 0; | |
492 | req->in.h.unique = unique; | |
493 | spin_lock(&fc->lock); | |
494 | if (fc->connected) { | |
495 | queue_request(fc, req); | |
496 | err = 0; | |
497 | } | |
498 | spin_unlock(&fc->lock); | |
499 | ||
500 | return err; | |
501 | } | |
502 | ||
3be5a52b MS |
503 | /* |
504 | * Called under fc->lock | |
505 | * | |
506 | * fc->connected must have been checked previously | |
507 | */ | |
b93f858a TH |
508 | void fuse_request_send_background_locked(struct fuse_conn *fc, |
509 | struct fuse_req *req) | |
3be5a52b MS |
510 | { |
511 | req->isreply = 1; | |
b93f858a | 512 | fuse_request_send_nowait_locked(fc, req); |
3be5a52b MS |
513 | } |
514 | ||
0b05b183 AA |
515 | void fuse_force_forget(struct file *file, u64 nodeid) |
516 | { | |
517 | struct inode *inode = file->f_path.dentry->d_inode; | |
518 | struct fuse_conn *fc = get_fuse_conn(inode); | |
519 | struct fuse_req *req; | |
520 | struct fuse_forget_in inarg; | |
521 | ||
522 | memset(&inarg, 0, sizeof(inarg)); | |
523 | inarg.nlookup = 1; | |
524 | req = fuse_get_req_nofail(fc, file); | |
525 | req->in.h.opcode = FUSE_FORGET; | |
526 | req->in.h.nodeid = nodeid; | |
527 | req->in.numargs = 1; | |
528 | req->in.args[0].size = sizeof(inarg); | |
529 | req->in.args[0].value = &inarg; | |
530 | req->isreply = 0; | |
531 | fuse_request_send_nowait(fc, req); | |
532 | } | |
533 | ||
334f485d MS |
534 | /* |
535 | * Lock the request. Up to the next unlock_request() there mustn't be | |
536 | * anything that could cause a page-fault. If the request was already | |
f9a2842e | 537 | * aborted bail out. |
334f485d | 538 | */ |
d7133114 | 539 | static int lock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
540 | { |
541 | int err = 0; | |
542 | if (req) { | |
d7133114 | 543 | spin_lock(&fc->lock); |
f9a2842e | 544 | if (req->aborted) |
334f485d MS |
545 | err = -ENOENT; |
546 | else | |
547 | req->locked = 1; | |
d7133114 | 548 | spin_unlock(&fc->lock); |
334f485d MS |
549 | } |
550 | return err; | |
551 | } | |
552 | ||
553 | /* | |
f9a2842e | 554 | * Unlock request. If it was aborted during being locked, the |
334f485d MS |
555 | * requester thread is currently waiting for it to be unlocked, so |
556 | * wake it up. | |
557 | */ | |
d7133114 | 558 | static void unlock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
559 | { |
560 | if (req) { | |
d7133114 | 561 | spin_lock(&fc->lock); |
334f485d | 562 | req->locked = 0; |
f9a2842e | 563 | if (req->aborted) |
334f485d | 564 | wake_up(&req->waitq); |
d7133114 | 565 | spin_unlock(&fc->lock); |
334f485d MS |
566 | } |
567 | } | |
568 | ||
569 | struct fuse_copy_state { | |
d7133114 | 570 | struct fuse_conn *fc; |
334f485d MS |
571 | int write; |
572 | struct fuse_req *req; | |
573 | const struct iovec *iov; | |
dd3bb14f MS |
574 | struct pipe_buffer *pipebufs; |
575 | struct pipe_buffer *currbuf; | |
576 | struct pipe_inode_info *pipe; | |
334f485d MS |
577 | unsigned long nr_segs; |
578 | unsigned long seglen; | |
579 | unsigned long addr; | |
580 | struct page *pg; | |
581 | void *mapaddr; | |
582 | void *buf; | |
583 | unsigned len; | |
ce534fb0 | 584 | unsigned move_pages:1; |
334f485d MS |
585 | }; |
586 | ||
d7133114 | 587 | static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc, |
c3021629 | 588 | int write, |
d7133114 | 589 | const struct iovec *iov, unsigned long nr_segs) |
334f485d MS |
590 | { |
591 | memset(cs, 0, sizeof(*cs)); | |
d7133114 | 592 | cs->fc = fc; |
334f485d | 593 | cs->write = write; |
334f485d MS |
594 | cs->iov = iov; |
595 | cs->nr_segs = nr_segs; | |
596 | } | |
597 | ||
598 | /* Unmap and put previous page of userspace buffer */ | |
8bfc016d | 599 | static void fuse_copy_finish(struct fuse_copy_state *cs) |
334f485d | 600 | { |
dd3bb14f MS |
601 | if (cs->currbuf) { |
602 | struct pipe_buffer *buf = cs->currbuf; | |
603 | ||
c3021629 MS |
604 | if (!cs->write) { |
605 | buf->ops->unmap(cs->pipe, buf, cs->mapaddr); | |
606 | } else { | |
7909b1c6 | 607 | kunmap(buf->page); |
c3021629 MS |
608 | buf->len = PAGE_SIZE - cs->len; |
609 | } | |
dd3bb14f MS |
610 | cs->currbuf = NULL; |
611 | cs->mapaddr = NULL; | |
612 | } else if (cs->mapaddr) { | |
7909b1c6 | 613 | kunmap(cs->pg); |
334f485d MS |
614 | if (cs->write) { |
615 | flush_dcache_page(cs->pg); | |
616 | set_page_dirty_lock(cs->pg); | |
617 | } | |
618 | put_page(cs->pg); | |
619 | cs->mapaddr = NULL; | |
620 | } | |
621 | } | |
622 | ||
623 | /* | |
624 | * Get another pagefull of userspace buffer, and map it to kernel | |
625 | * address space, and lock request | |
626 | */ | |
627 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
628 | { | |
629 | unsigned long offset; | |
630 | int err; | |
631 | ||
d7133114 | 632 | unlock_request(cs->fc, cs->req); |
334f485d | 633 | fuse_copy_finish(cs); |
dd3bb14f MS |
634 | if (cs->pipebufs) { |
635 | struct pipe_buffer *buf = cs->pipebufs; | |
636 | ||
c3021629 MS |
637 | if (!cs->write) { |
638 | err = buf->ops->confirm(cs->pipe, buf); | |
639 | if (err) | |
640 | return err; | |
641 | ||
642 | BUG_ON(!cs->nr_segs); | |
643 | cs->currbuf = buf; | |
7909b1c6 | 644 | cs->mapaddr = buf->ops->map(cs->pipe, buf, 0); |
c3021629 MS |
645 | cs->len = buf->len; |
646 | cs->buf = cs->mapaddr + buf->offset; | |
647 | cs->pipebufs++; | |
648 | cs->nr_segs--; | |
649 | } else { | |
650 | struct page *page; | |
dd3bb14f | 651 | |
c3021629 MS |
652 | if (cs->nr_segs == cs->pipe->buffers) |
653 | return -EIO; | |
654 | ||
655 | page = alloc_page(GFP_HIGHUSER); | |
656 | if (!page) | |
657 | return -ENOMEM; | |
658 | ||
659 | buf->page = page; | |
660 | buf->offset = 0; | |
661 | buf->len = 0; | |
662 | ||
663 | cs->currbuf = buf; | |
7909b1c6 | 664 | cs->mapaddr = kmap(page); |
c3021629 MS |
665 | cs->buf = cs->mapaddr; |
666 | cs->len = PAGE_SIZE; | |
667 | cs->pipebufs++; | |
668 | cs->nr_segs++; | |
669 | } | |
dd3bb14f MS |
670 | } else { |
671 | if (!cs->seglen) { | |
672 | BUG_ON(!cs->nr_segs); | |
673 | cs->seglen = cs->iov[0].iov_len; | |
674 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
675 | cs->iov++; | |
676 | cs->nr_segs--; | |
677 | } | |
678 | err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg); | |
679 | if (err < 0) | |
680 | return err; | |
681 | BUG_ON(err != 1); | |
682 | offset = cs->addr % PAGE_SIZE; | |
7909b1c6 | 683 | cs->mapaddr = kmap(cs->pg); |
dd3bb14f MS |
684 | cs->buf = cs->mapaddr + offset; |
685 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
686 | cs->seglen -= cs->len; | |
687 | cs->addr += cs->len; | |
334f485d | 688 | } |
334f485d | 689 | |
d7133114 | 690 | return lock_request(cs->fc, cs->req); |
334f485d MS |
691 | } |
692 | ||
693 | /* Do as much copy to/from userspace buffer as we can */ | |
8bfc016d | 694 | static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) |
334f485d MS |
695 | { |
696 | unsigned ncpy = min(*size, cs->len); | |
697 | if (val) { | |
698 | if (cs->write) | |
699 | memcpy(cs->buf, *val, ncpy); | |
700 | else | |
701 | memcpy(*val, cs->buf, ncpy); | |
702 | *val += ncpy; | |
703 | } | |
704 | *size -= ncpy; | |
705 | cs->len -= ncpy; | |
706 | cs->buf += ncpy; | |
707 | return ncpy; | |
708 | } | |
709 | ||
ce534fb0 MS |
710 | static int fuse_check_page(struct page *page) |
711 | { | |
712 | if (page_mapcount(page) || | |
713 | page->mapping != NULL || | |
714 | page_count(page) != 1 || | |
715 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP & | |
716 | ~(1 << PG_locked | | |
717 | 1 << PG_referenced | | |
718 | 1 << PG_uptodate | | |
719 | 1 << PG_lru | | |
720 | 1 << PG_active | | |
721 | 1 << PG_reclaim))) { | |
722 | printk(KERN_WARNING "fuse: trying to steal weird page\n"); | |
723 | 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); | |
724 | return 1; | |
725 | } | |
726 | return 0; | |
727 | } | |
728 | ||
729 | static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep) | |
730 | { | |
731 | int err; | |
732 | struct page *oldpage = *pagep; | |
733 | struct page *newpage; | |
734 | struct pipe_buffer *buf = cs->pipebufs; | |
ce534fb0 MS |
735 | |
736 | unlock_request(cs->fc, cs->req); | |
737 | fuse_copy_finish(cs); | |
738 | ||
739 | err = buf->ops->confirm(cs->pipe, buf); | |
740 | if (err) | |
741 | return err; | |
742 | ||
743 | BUG_ON(!cs->nr_segs); | |
744 | cs->currbuf = buf; | |
745 | cs->len = buf->len; | |
746 | cs->pipebufs++; | |
747 | cs->nr_segs--; | |
748 | ||
749 | if (cs->len != PAGE_SIZE) | |
750 | goto out_fallback; | |
751 | ||
752 | if (buf->ops->steal(cs->pipe, buf) != 0) | |
753 | goto out_fallback; | |
754 | ||
755 | newpage = buf->page; | |
756 | ||
757 | if (WARN_ON(!PageUptodate(newpage))) | |
758 | return -EIO; | |
759 | ||
760 | ClearPageMappedToDisk(newpage); | |
761 | ||
762 | if (fuse_check_page(newpage) != 0) | |
763 | goto out_fallback_unlock; | |
764 | ||
ce534fb0 MS |
765 | /* |
766 | * This is a new and locked page, it shouldn't be mapped or | |
767 | * have any special flags on it | |
768 | */ | |
769 | if (WARN_ON(page_mapped(oldpage))) | |
770 | goto out_fallback_unlock; | |
771 | if (WARN_ON(page_has_private(oldpage))) | |
772 | goto out_fallback_unlock; | |
773 | if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage))) | |
774 | goto out_fallback_unlock; | |
775 | if (WARN_ON(PageMlocked(oldpage))) | |
776 | goto out_fallback_unlock; | |
777 | ||
ef6a3c63 | 778 | err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL); |
ce534fb0 | 779 | if (err) { |
ef6a3c63 MS |
780 | unlock_page(newpage); |
781 | return err; | |
ce534fb0 | 782 | } |
ef6a3c63 | 783 | |
ce534fb0 MS |
784 | page_cache_get(newpage); |
785 | ||
786 | if (!(buf->flags & PIPE_BUF_FLAG_LRU)) | |
787 | lru_cache_add_file(newpage); | |
788 | ||
789 | err = 0; | |
790 | spin_lock(&cs->fc->lock); | |
791 | if (cs->req->aborted) | |
792 | err = -ENOENT; | |
793 | else | |
794 | *pagep = newpage; | |
795 | spin_unlock(&cs->fc->lock); | |
796 | ||
797 | if (err) { | |
798 | unlock_page(newpage); | |
799 | page_cache_release(newpage); | |
800 | return err; | |
801 | } | |
802 | ||
803 | unlock_page(oldpage); | |
804 | page_cache_release(oldpage); | |
805 | cs->len = 0; | |
806 | ||
807 | return 0; | |
808 | ||
809 | out_fallback_unlock: | |
810 | unlock_page(newpage); | |
811 | out_fallback: | |
812 | cs->mapaddr = buf->ops->map(cs->pipe, buf, 1); | |
813 | cs->buf = cs->mapaddr + buf->offset; | |
814 | ||
815 | err = lock_request(cs->fc, cs->req); | |
816 | if (err) | |
817 | return err; | |
818 | ||
819 | return 1; | |
820 | } | |
821 | ||
c3021629 MS |
822 | static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page, |
823 | unsigned offset, unsigned count) | |
824 | { | |
825 | struct pipe_buffer *buf; | |
826 | ||
827 | if (cs->nr_segs == cs->pipe->buffers) | |
828 | return -EIO; | |
829 | ||
830 | unlock_request(cs->fc, cs->req); | |
831 | fuse_copy_finish(cs); | |
832 | ||
833 | buf = cs->pipebufs; | |
834 | page_cache_get(page); | |
835 | buf->page = page; | |
836 | buf->offset = offset; | |
837 | buf->len = count; | |
838 | ||
839 | cs->pipebufs++; | |
840 | cs->nr_segs++; | |
841 | cs->len = 0; | |
842 | ||
843 | return 0; | |
844 | } | |
845 | ||
334f485d MS |
846 | /* |
847 | * Copy a page in the request to/from the userspace buffer. Must be | |
848 | * done atomically | |
849 | */ | |
ce534fb0 | 850 | static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep, |
8bfc016d | 851 | unsigned offset, unsigned count, int zeroing) |
334f485d | 852 | { |
ce534fb0 MS |
853 | int err; |
854 | struct page *page = *pagep; | |
855 | ||
b6777c40 MS |
856 | if (page && zeroing && count < PAGE_SIZE) |
857 | clear_highpage(page); | |
858 | ||
334f485d | 859 | while (count) { |
c3021629 MS |
860 | if (cs->write && cs->pipebufs && page) { |
861 | return fuse_ref_page(cs, page, offset, count); | |
862 | } else if (!cs->len) { | |
ce534fb0 MS |
863 | if (cs->move_pages && page && |
864 | offset == 0 && count == PAGE_SIZE) { | |
865 | err = fuse_try_move_page(cs, pagep); | |
866 | if (err <= 0) | |
867 | return err; | |
868 | } else { | |
869 | err = fuse_copy_fill(cs); | |
870 | if (err) | |
871 | return err; | |
872 | } | |
1729a16c | 873 | } |
334f485d | 874 | if (page) { |
2408f6ef | 875 | void *mapaddr = kmap_atomic(page); |
334f485d MS |
876 | void *buf = mapaddr + offset; |
877 | offset += fuse_copy_do(cs, &buf, &count); | |
2408f6ef | 878 | kunmap_atomic(mapaddr); |
334f485d MS |
879 | } else |
880 | offset += fuse_copy_do(cs, NULL, &count); | |
881 | } | |
882 | if (page && !cs->write) | |
883 | flush_dcache_page(page); | |
884 | return 0; | |
885 | } | |
886 | ||
887 | /* Copy pages in the request to/from userspace buffer */ | |
888 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
889 | int zeroing) | |
890 | { | |
891 | unsigned i; | |
892 | struct fuse_req *req = cs->req; | |
893 | unsigned offset = req->page_offset; | |
894 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
895 | ||
896 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
ce534fb0 MS |
897 | int err; |
898 | ||
899 | err = fuse_copy_page(cs, &req->pages[i], offset, count, | |
900 | zeroing); | |
334f485d MS |
901 | if (err) |
902 | return err; | |
903 | ||
904 | nbytes -= count; | |
905 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
906 | offset = 0; | |
907 | } | |
908 | return 0; | |
909 | } | |
910 | ||
911 | /* Copy a single argument in the request to/from userspace buffer */ | |
912 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
913 | { | |
914 | while (size) { | |
1729a16c MS |
915 | if (!cs->len) { |
916 | int err = fuse_copy_fill(cs); | |
917 | if (err) | |
918 | return err; | |
919 | } | |
334f485d MS |
920 | fuse_copy_do(cs, &val, &size); |
921 | } | |
922 | return 0; | |
923 | } | |
924 | ||
925 | /* Copy request arguments to/from userspace buffer */ | |
926 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
927 | unsigned argpages, struct fuse_arg *args, | |
928 | int zeroing) | |
929 | { | |
930 | int err = 0; | |
931 | unsigned i; | |
932 | ||
933 | for (i = 0; !err && i < numargs; i++) { | |
934 | struct fuse_arg *arg = &args[i]; | |
935 | if (i == numargs - 1 && argpages) | |
936 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
937 | else | |
938 | err = fuse_copy_one(cs, arg->value, arg->size); | |
939 | } | |
940 | return err; | |
941 | } | |
942 | ||
07e77dca MS |
943 | static int forget_pending(struct fuse_conn *fc) |
944 | { | |
945 | return fc->forget_list_head.next != NULL; | |
946 | } | |
947 | ||
a4d27e75 MS |
948 | static int request_pending(struct fuse_conn *fc) |
949 | { | |
07e77dca MS |
950 | return !list_empty(&fc->pending) || !list_empty(&fc->interrupts) || |
951 | forget_pending(fc); | |
a4d27e75 MS |
952 | } |
953 | ||
334f485d MS |
954 | /* Wait until a request is available on the pending list */ |
955 | static void request_wait(struct fuse_conn *fc) | |
b9ca67b2 MS |
956 | __releases(fc->lock) |
957 | __acquires(fc->lock) | |
334f485d MS |
958 | { |
959 | DECLARE_WAITQUEUE(wait, current); | |
960 | ||
961 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
a4d27e75 | 962 | while (fc->connected && !request_pending(fc)) { |
334f485d MS |
963 | set_current_state(TASK_INTERRUPTIBLE); |
964 | if (signal_pending(current)) | |
965 | break; | |
966 | ||
d7133114 | 967 | spin_unlock(&fc->lock); |
334f485d | 968 | schedule(); |
d7133114 | 969 | spin_lock(&fc->lock); |
334f485d MS |
970 | } |
971 | set_current_state(TASK_RUNNING); | |
972 | remove_wait_queue(&fc->waitq, &wait); | |
973 | } | |
974 | ||
a4d27e75 MS |
975 | /* |
976 | * Transfer an interrupt request to userspace | |
977 | * | |
978 | * Unlike other requests this is assembled on demand, without a need | |
979 | * to allocate a separate fuse_req structure. | |
980 | * | |
981 | * Called with fc->lock held, releases it | |
982 | */ | |
c3021629 MS |
983 | static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs, |
984 | size_t nbytes, struct fuse_req *req) | |
b9ca67b2 | 985 | __releases(fc->lock) |
a4d27e75 | 986 | { |
a4d27e75 MS |
987 | struct fuse_in_header ih; |
988 | struct fuse_interrupt_in arg; | |
989 | unsigned reqsize = sizeof(ih) + sizeof(arg); | |
990 | int err; | |
991 | ||
992 | list_del_init(&req->intr_entry); | |
993 | req->intr_unique = fuse_get_unique(fc); | |
994 | memset(&ih, 0, sizeof(ih)); | |
995 | memset(&arg, 0, sizeof(arg)); | |
996 | ih.len = reqsize; | |
997 | ih.opcode = FUSE_INTERRUPT; | |
998 | ih.unique = req->intr_unique; | |
999 | arg.unique = req->in.h.unique; | |
1000 | ||
1001 | spin_unlock(&fc->lock); | |
c3021629 | 1002 | if (nbytes < reqsize) |
a4d27e75 MS |
1003 | return -EINVAL; |
1004 | ||
c3021629 | 1005 | err = fuse_copy_one(cs, &ih, sizeof(ih)); |
a4d27e75 | 1006 | if (!err) |
c3021629 MS |
1007 | err = fuse_copy_one(cs, &arg, sizeof(arg)); |
1008 | fuse_copy_finish(cs); | |
a4d27e75 MS |
1009 | |
1010 | return err ? err : reqsize; | |
1011 | } | |
1012 | ||
02c048b9 MS |
1013 | static struct fuse_forget_link *dequeue_forget(struct fuse_conn *fc, |
1014 | unsigned max, | |
1015 | unsigned *countp) | |
07e77dca | 1016 | { |
02c048b9 MS |
1017 | struct fuse_forget_link *head = fc->forget_list_head.next; |
1018 | struct fuse_forget_link **newhead = &head; | |
1019 | unsigned count; | |
07e77dca | 1020 | |
02c048b9 MS |
1021 | for (count = 0; *newhead != NULL && count < max; count++) |
1022 | newhead = &(*newhead)->next; | |
1023 | ||
1024 | fc->forget_list_head.next = *newhead; | |
1025 | *newhead = NULL; | |
07e77dca MS |
1026 | if (fc->forget_list_head.next == NULL) |
1027 | fc->forget_list_tail = &fc->forget_list_head; | |
1028 | ||
02c048b9 MS |
1029 | if (countp != NULL) |
1030 | *countp = count; | |
1031 | ||
1032 | return head; | |
07e77dca MS |
1033 | } |
1034 | ||
1035 | static int fuse_read_single_forget(struct fuse_conn *fc, | |
1036 | struct fuse_copy_state *cs, | |
1037 | size_t nbytes) | |
1038 | __releases(fc->lock) | |
1039 | { | |
1040 | int err; | |
02c048b9 | 1041 | struct fuse_forget_link *forget = dequeue_forget(fc, 1, NULL); |
07e77dca | 1042 | struct fuse_forget_in arg = { |
02c048b9 | 1043 | .nlookup = forget->forget_one.nlookup, |
07e77dca MS |
1044 | }; |
1045 | struct fuse_in_header ih = { | |
1046 | .opcode = FUSE_FORGET, | |
02c048b9 | 1047 | .nodeid = forget->forget_one.nodeid, |
07e77dca MS |
1048 | .unique = fuse_get_unique(fc), |
1049 | .len = sizeof(ih) + sizeof(arg), | |
1050 | }; | |
1051 | ||
1052 | spin_unlock(&fc->lock); | |
1053 | kfree(forget); | |
1054 | if (nbytes < ih.len) | |
1055 | return -EINVAL; | |
1056 | ||
1057 | err = fuse_copy_one(cs, &ih, sizeof(ih)); | |
1058 | if (!err) | |
1059 | err = fuse_copy_one(cs, &arg, sizeof(arg)); | |
1060 | fuse_copy_finish(cs); | |
1061 | ||
1062 | if (err) | |
1063 | return err; | |
1064 | ||
1065 | return ih.len; | |
1066 | } | |
1067 | ||
02c048b9 MS |
1068 | static int fuse_read_batch_forget(struct fuse_conn *fc, |
1069 | struct fuse_copy_state *cs, size_t nbytes) | |
1070 | __releases(fc->lock) | |
1071 | { | |
1072 | int err; | |
1073 | unsigned max_forgets; | |
1074 | unsigned count; | |
1075 | struct fuse_forget_link *head; | |
1076 | struct fuse_batch_forget_in arg = { .count = 0 }; | |
1077 | struct fuse_in_header ih = { | |
1078 | .opcode = FUSE_BATCH_FORGET, | |
1079 | .unique = fuse_get_unique(fc), | |
1080 | .len = sizeof(ih) + sizeof(arg), | |
1081 | }; | |
1082 | ||
1083 | if (nbytes < ih.len) { | |
1084 | spin_unlock(&fc->lock); | |
1085 | return -EINVAL; | |
1086 | } | |
1087 | ||
1088 | max_forgets = (nbytes - ih.len) / sizeof(struct fuse_forget_one); | |
1089 | head = dequeue_forget(fc, max_forgets, &count); | |
1090 | spin_unlock(&fc->lock); | |
1091 | ||
1092 | arg.count = count; | |
1093 | ih.len += count * sizeof(struct fuse_forget_one); | |
1094 | err = fuse_copy_one(cs, &ih, sizeof(ih)); | |
1095 | if (!err) | |
1096 | err = fuse_copy_one(cs, &arg, sizeof(arg)); | |
1097 | ||
1098 | while (head) { | |
1099 | struct fuse_forget_link *forget = head; | |
1100 | ||
1101 | if (!err) { | |
1102 | err = fuse_copy_one(cs, &forget->forget_one, | |
1103 | sizeof(forget->forget_one)); | |
1104 | } | |
1105 | head = forget->next; | |
1106 | kfree(forget); | |
1107 | } | |
1108 | ||
1109 | fuse_copy_finish(cs); | |
1110 | ||
1111 | if (err) | |
1112 | return err; | |
1113 | ||
1114 | return ih.len; | |
1115 | } | |
1116 | ||
1117 | static int fuse_read_forget(struct fuse_conn *fc, struct fuse_copy_state *cs, | |
1118 | size_t nbytes) | |
1119 | __releases(fc->lock) | |
1120 | { | |
1121 | if (fc->minor < 16 || fc->forget_list_head.next->next == NULL) | |
1122 | return fuse_read_single_forget(fc, cs, nbytes); | |
1123 | else | |
1124 | return fuse_read_batch_forget(fc, cs, nbytes); | |
1125 | } | |
1126 | ||
334f485d MS |
1127 | /* |
1128 | * Read a single request into the userspace filesystem's buffer. This | |
1129 | * function waits until a request is available, then removes it from | |
1130 | * the pending list and copies request data to userspace buffer. If | |
f9a2842e MS |
1131 | * no reply is needed (FORGET) or request has been aborted or there |
1132 | * was an error during the copying then it's finished by calling | |
334f485d MS |
1133 | * request_end(). Otherwise add it to the processing list, and set |
1134 | * the 'sent' flag. | |
1135 | */ | |
c3021629 MS |
1136 | static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file, |
1137 | struct fuse_copy_state *cs, size_t nbytes) | |
334f485d MS |
1138 | { |
1139 | int err; | |
334f485d MS |
1140 | struct fuse_req *req; |
1141 | struct fuse_in *in; | |
334f485d MS |
1142 | unsigned reqsize; |
1143 | ||
1d3d752b | 1144 | restart: |
d7133114 | 1145 | spin_lock(&fc->lock); |
e5ac1d1e JD |
1146 | err = -EAGAIN; |
1147 | if ((file->f_flags & O_NONBLOCK) && fc->connected && | |
a4d27e75 | 1148 | !request_pending(fc)) |
e5ac1d1e JD |
1149 | goto err_unlock; |
1150 | ||
334f485d MS |
1151 | request_wait(fc); |
1152 | err = -ENODEV; | |
9ba7cbba | 1153 | if (!fc->connected) |
334f485d MS |
1154 | goto err_unlock; |
1155 | err = -ERESTARTSYS; | |
a4d27e75 | 1156 | if (!request_pending(fc)) |
334f485d MS |
1157 | goto err_unlock; |
1158 | ||
a4d27e75 MS |
1159 | if (!list_empty(&fc->interrupts)) { |
1160 | req = list_entry(fc->interrupts.next, struct fuse_req, | |
1161 | intr_entry); | |
c3021629 | 1162 | return fuse_read_interrupt(fc, cs, nbytes, req); |
a4d27e75 MS |
1163 | } |
1164 | ||
07e77dca MS |
1165 | if (forget_pending(fc)) { |
1166 | if (list_empty(&fc->pending) || fc->forget_batch-- > 0) | |
02c048b9 | 1167 | return fuse_read_forget(fc, cs, nbytes); |
07e77dca MS |
1168 | |
1169 | if (fc->forget_batch <= -8) | |
1170 | fc->forget_batch = 16; | |
1171 | } | |
1172 | ||
334f485d | 1173 | req = list_entry(fc->pending.next, struct fuse_req, list); |
83cfd493 | 1174 | req->state = FUSE_REQ_READING; |
d77a1d5b | 1175 | list_move(&req->list, &fc->io); |
334f485d MS |
1176 | |
1177 | in = &req->in; | |
1d3d752b MS |
1178 | reqsize = in->h.len; |
1179 | /* If request is too large, reply with an error and restart the read */ | |
c3021629 | 1180 | if (nbytes < reqsize) { |
1d3d752b MS |
1181 | req->out.h.error = -EIO; |
1182 | /* SETXATTR is special, since it may contain too large data */ | |
1183 | if (in->h.opcode == FUSE_SETXATTR) | |
1184 | req->out.h.error = -E2BIG; | |
1185 | request_end(fc, req); | |
1186 | goto restart; | |
334f485d | 1187 | } |
d7133114 | 1188 | spin_unlock(&fc->lock); |
c3021629 MS |
1189 | cs->req = req; |
1190 | err = fuse_copy_one(cs, &in->h, sizeof(in->h)); | |
1d3d752b | 1191 | if (!err) |
c3021629 | 1192 | err = fuse_copy_args(cs, in->numargs, in->argpages, |
1d3d752b | 1193 | (struct fuse_arg *) in->args, 0); |
c3021629 | 1194 | fuse_copy_finish(cs); |
d7133114 | 1195 | spin_lock(&fc->lock); |
334f485d | 1196 | req->locked = 0; |
c9c9d7df MS |
1197 | if (req->aborted) { |
1198 | request_end(fc, req); | |
1199 | return -ENODEV; | |
1200 | } | |
334f485d | 1201 | if (err) { |
c9c9d7df | 1202 | req->out.h.error = -EIO; |
334f485d MS |
1203 | request_end(fc, req); |
1204 | return err; | |
1205 | } | |
1206 | if (!req->isreply) | |
1207 | request_end(fc, req); | |
1208 | else { | |
83cfd493 | 1209 | req->state = FUSE_REQ_SENT; |
d77a1d5b | 1210 | list_move_tail(&req->list, &fc->processing); |
a4d27e75 MS |
1211 | if (req->interrupted) |
1212 | queue_interrupt(fc, req); | |
d7133114 | 1213 | spin_unlock(&fc->lock); |
334f485d MS |
1214 | } |
1215 | return reqsize; | |
1216 | ||
1217 | err_unlock: | |
d7133114 | 1218 | spin_unlock(&fc->lock); |
334f485d MS |
1219 | return err; |
1220 | } | |
1221 | ||
c3021629 MS |
1222 | static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov, |
1223 | unsigned long nr_segs, loff_t pos) | |
1224 | { | |
1225 | struct fuse_copy_state cs; | |
1226 | struct file *file = iocb->ki_filp; | |
1227 | struct fuse_conn *fc = fuse_get_conn(file); | |
1228 | if (!fc) | |
1229 | return -EPERM; | |
1230 | ||
1231 | fuse_copy_init(&cs, fc, 1, iov, nr_segs); | |
1232 | ||
1233 | return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs)); | |
1234 | } | |
1235 | ||
1236 | static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe, | |
1237 | struct pipe_buffer *buf) | |
1238 | { | |
1239 | return 1; | |
1240 | } | |
1241 | ||
1242 | static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = { | |
1243 | .can_merge = 0, | |
1244 | .map = generic_pipe_buf_map, | |
1245 | .unmap = generic_pipe_buf_unmap, | |
1246 | .confirm = generic_pipe_buf_confirm, | |
1247 | .release = generic_pipe_buf_release, | |
1248 | .steal = fuse_dev_pipe_buf_steal, | |
1249 | .get = generic_pipe_buf_get, | |
1250 | }; | |
1251 | ||
1252 | static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos, | |
1253 | struct pipe_inode_info *pipe, | |
1254 | size_t len, unsigned int flags) | |
1255 | { | |
1256 | int ret; | |
1257 | int page_nr = 0; | |
1258 | int do_wakeup = 0; | |
1259 | struct pipe_buffer *bufs; | |
1260 | struct fuse_copy_state cs; | |
1261 | struct fuse_conn *fc = fuse_get_conn(in); | |
1262 | if (!fc) | |
1263 | return -EPERM; | |
1264 | ||
07e77dca | 1265 | bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL); |
c3021629 MS |
1266 | if (!bufs) |
1267 | return -ENOMEM; | |
1268 | ||
1269 | fuse_copy_init(&cs, fc, 1, NULL, 0); | |
1270 | cs.pipebufs = bufs; | |
1271 | cs.pipe = pipe; | |
1272 | ret = fuse_dev_do_read(fc, in, &cs, len); | |
1273 | if (ret < 0) | |
1274 | goto out; | |
1275 | ||
1276 | ret = 0; | |
1277 | pipe_lock(pipe); | |
1278 | ||
1279 | if (!pipe->readers) { | |
1280 | send_sig(SIGPIPE, current, 0); | |
1281 | if (!ret) | |
1282 | ret = -EPIPE; | |
1283 | goto out_unlock; | |
1284 | } | |
1285 | ||
1286 | if (pipe->nrbufs + cs.nr_segs > pipe->buffers) { | |
1287 | ret = -EIO; | |
1288 | goto out_unlock; | |
1289 | } | |
1290 | ||
1291 | while (page_nr < cs.nr_segs) { | |
1292 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
1293 | struct pipe_buffer *buf = pipe->bufs + newbuf; | |
1294 | ||
1295 | buf->page = bufs[page_nr].page; | |
1296 | buf->offset = bufs[page_nr].offset; | |
1297 | buf->len = bufs[page_nr].len; | |
1298 | buf->ops = &fuse_dev_pipe_buf_ops; | |
1299 | ||
1300 | pipe->nrbufs++; | |
1301 | page_nr++; | |
1302 | ret += buf->len; | |
1303 | ||
1304 | if (pipe->inode) | |
1305 | do_wakeup = 1; | |
1306 | } | |
1307 | ||
1308 | out_unlock: | |
1309 | pipe_unlock(pipe); | |
1310 | ||
1311 | if (do_wakeup) { | |
1312 | smp_mb(); | |
1313 | if (waitqueue_active(&pipe->wait)) | |
1314 | wake_up_interruptible(&pipe->wait); | |
1315 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
1316 | } | |
1317 | ||
1318 | out: | |
1319 | for (; page_nr < cs.nr_segs; page_nr++) | |
1320 | page_cache_release(bufs[page_nr].page); | |
1321 | ||
1322 | kfree(bufs); | |
1323 | return ret; | |
1324 | } | |
1325 | ||
95668a69 TH |
1326 | static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size, |
1327 | struct fuse_copy_state *cs) | |
1328 | { | |
1329 | struct fuse_notify_poll_wakeup_out outarg; | |
f6d47a17 | 1330 | int err = -EINVAL; |
95668a69 TH |
1331 | |
1332 | if (size != sizeof(outarg)) | |
f6d47a17 | 1333 | goto err; |
95668a69 TH |
1334 | |
1335 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1336 | if (err) | |
f6d47a17 | 1337 | goto err; |
95668a69 | 1338 | |
f6d47a17 | 1339 | fuse_copy_finish(cs); |
95668a69 | 1340 | return fuse_notify_poll_wakeup(fc, &outarg); |
f6d47a17 MS |
1341 | |
1342 | err: | |
1343 | fuse_copy_finish(cs); | |
1344 | return err; | |
95668a69 TH |
1345 | } |
1346 | ||
3b463ae0 JM |
1347 | static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size, |
1348 | struct fuse_copy_state *cs) | |
1349 | { | |
1350 | struct fuse_notify_inval_inode_out outarg; | |
1351 | int err = -EINVAL; | |
1352 | ||
1353 | if (size != sizeof(outarg)) | |
1354 | goto err; | |
1355 | ||
1356 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1357 | if (err) | |
1358 | goto err; | |
1359 | fuse_copy_finish(cs); | |
1360 | ||
1361 | down_read(&fc->killsb); | |
1362 | err = -ENOENT; | |
b21dda43 MS |
1363 | if (fc->sb) { |
1364 | err = fuse_reverse_inval_inode(fc->sb, outarg.ino, | |
1365 | outarg.off, outarg.len); | |
1366 | } | |
3b463ae0 JM |
1367 | up_read(&fc->killsb); |
1368 | return err; | |
1369 | ||
1370 | err: | |
1371 | fuse_copy_finish(cs); | |
1372 | return err; | |
1373 | } | |
1374 | ||
1375 | static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size, | |
1376 | struct fuse_copy_state *cs) | |
1377 | { | |
1378 | struct fuse_notify_inval_entry_out outarg; | |
b2d82ee3 FW |
1379 | int err = -ENOMEM; |
1380 | char *buf; | |
3b463ae0 JM |
1381 | struct qstr name; |
1382 | ||
b2d82ee3 FW |
1383 | buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); |
1384 | if (!buf) | |
1385 | goto err; | |
1386 | ||
1387 | err = -EINVAL; | |
3b463ae0 JM |
1388 | if (size < sizeof(outarg)) |
1389 | goto err; | |
1390 | ||
1391 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1392 | if (err) | |
1393 | goto err; | |
1394 | ||
1395 | err = -ENAMETOOLONG; | |
1396 | if (outarg.namelen > FUSE_NAME_MAX) | |
1397 | goto err; | |
1398 | ||
c2183d1e MS |
1399 | err = -EINVAL; |
1400 | if (size != sizeof(outarg) + outarg.namelen + 1) | |
1401 | goto err; | |
1402 | ||
3b463ae0 JM |
1403 | name.name = buf; |
1404 | name.len = outarg.namelen; | |
1405 | err = fuse_copy_one(cs, buf, outarg.namelen + 1); | |
1406 | if (err) | |
1407 | goto err; | |
1408 | fuse_copy_finish(cs); | |
1409 | buf[outarg.namelen] = 0; | |
1410 | name.hash = full_name_hash(name.name, name.len); | |
1411 | ||
1412 | down_read(&fc->killsb); | |
1413 | err = -ENOENT; | |
b21dda43 | 1414 | if (fc->sb) |
451d0f59 JM |
1415 | err = fuse_reverse_inval_entry(fc->sb, outarg.parent, 0, &name); |
1416 | up_read(&fc->killsb); | |
1417 | kfree(buf); | |
1418 | return err; | |
1419 | ||
1420 | err: | |
1421 | kfree(buf); | |
1422 | fuse_copy_finish(cs); | |
1423 | return err; | |
1424 | } | |
1425 | ||
1426 | static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size, | |
1427 | struct fuse_copy_state *cs) | |
1428 | { | |
1429 | struct fuse_notify_delete_out outarg; | |
1430 | int err = -ENOMEM; | |
1431 | char *buf; | |
1432 | struct qstr name; | |
1433 | ||
1434 | buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL); | |
1435 | if (!buf) | |
1436 | goto err; | |
1437 | ||
1438 | err = -EINVAL; | |
1439 | if (size < sizeof(outarg)) | |
1440 | goto err; | |
1441 | ||
1442 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1443 | if (err) | |
1444 | goto err; | |
1445 | ||
1446 | err = -ENAMETOOLONG; | |
1447 | if (outarg.namelen > FUSE_NAME_MAX) | |
1448 | goto err; | |
1449 | ||
1450 | err = -EINVAL; | |
1451 | if (size != sizeof(outarg) + outarg.namelen + 1) | |
1452 | goto err; | |
1453 | ||
1454 | name.name = buf; | |
1455 | name.len = outarg.namelen; | |
1456 | err = fuse_copy_one(cs, buf, outarg.namelen + 1); | |
1457 | if (err) | |
1458 | goto err; | |
1459 | fuse_copy_finish(cs); | |
1460 | buf[outarg.namelen] = 0; | |
1461 | name.hash = full_name_hash(name.name, name.len); | |
1462 | ||
1463 | down_read(&fc->killsb); | |
1464 | err = -ENOENT; | |
1465 | if (fc->sb) | |
1466 | err = fuse_reverse_inval_entry(fc->sb, outarg.parent, | |
1467 | outarg.child, &name); | |
3b463ae0 | 1468 | up_read(&fc->killsb); |
b2d82ee3 | 1469 | kfree(buf); |
3b463ae0 JM |
1470 | return err; |
1471 | ||
1472 | err: | |
b2d82ee3 | 1473 | kfree(buf); |
3b463ae0 JM |
1474 | fuse_copy_finish(cs); |
1475 | return err; | |
1476 | } | |
1477 | ||
a1d75f25 MS |
1478 | static int fuse_notify_store(struct fuse_conn *fc, unsigned int size, |
1479 | struct fuse_copy_state *cs) | |
1480 | { | |
1481 | struct fuse_notify_store_out outarg; | |
1482 | struct inode *inode; | |
1483 | struct address_space *mapping; | |
1484 | u64 nodeid; | |
1485 | int err; | |
1486 | pgoff_t index; | |
1487 | unsigned int offset; | |
1488 | unsigned int num; | |
1489 | loff_t file_size; | |
1490 | loff_t end; | |
1491 | ||
1492 | err = -EINVAL; | |
1493 | if (size < sizeof(outarg)) | |
1494 | goto out_finish; | |
1495 | ||
1496 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1497 | if (err) | |
1498 | goto out_finish; | |
1499 | ||
1500 | err = -EINVAL; | |
1501 | if (size - sizeof(outarg) != outarg.size) | |
1502 | goto out_finish; | |
1503 | ||
1504 | nodeid = outarg.nodeid; | |
1505 | ||
1506 | down_read(&fc->killsb); | |
1507 | ||
1508 | err = -ENOENT; | |
1509 | if (!fc->sb) | |
1510 | goto out_up_killsb; | |
1511 | ||
1512 | inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid); | |
1513 | if (!inode) | |
1514 | goto out_up_killsb; | |
1515 | ||
1516 | mapping = inode->i_mapping; | |
1517 | index = outarg.offset >> PAGE_CACHE_SHIFT; | |
1518 | offset = outarg.offset & ~PAGE_CACHE_MASK; | |
1519 | file_size = i_size_read(inode); | |
1520 | end = outarg.offset + outarg.size; | |
1521 | if (end > file_size) { | |
1522 | file_size = end; | |
1523 | fuse_write_update_size(inode, file_size); | |
1524 | } | |
1525 | ||
1526 | num = outarg.size; | |
1527 | while (num) { | |
1528 | struct page *page; | |
1529 | unsigned int this_num; | |
1530 | ||
1531 | err = -ENOMEM; | |
1532 | page = find_or_create_page(mapping, index, | |
1533 | mapping_gfp_mask(mapping)); | |
1534 | if (!page) | |
1535 | goto out_iput; | |
1536 | ||
1537 | this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset); | |
1538 | err = fuse_copy_page(cs, &page, offset, this_num, 0); | |
1539 | if (!err && offset == 0 && (num != 0 || file_size == end)) | |
1540 | SetPageUptodate(page); | |
1541 | unlock_page(page); | |
1542 | page_cache_release(page); | |
1543 | ||
1544 | if (err) | |
1545 | goto out_iput; | |
1546 | ||
1547 | num -= this_num; | |
1548 | offset = 0; | |
1549 | index++; | |
1550 | } | |
1551 | ||
1552 | err = 0; | |
1553 | ||
1554 | out_iput: | |
1555 | iput(inode); | |
1556 | out_up_killsb: | |
1557 | up_read(&fc->killsb); | |
1558 | out_finish: | |
1559 | fuse_copy_finish(cs); | |
1560 | return err; | |
1561 | } | |
1562 | ||
2d45ba38 MS |
1563 | static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req) |
1564 | { | |
0be8557b | 1565 | release_pages(req->pages, req->num_pages, 0); |
2d45ba38 MS |
1566 | } |
1567 | ||
1568 | static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode, | |
1569 | struct fuse_notify_retrieve_out *outarg) | |
1570 | { | |
1571 | int err; | |
1572 | struct address_space *mapping = inode->i_mapping; | |
1573 | struct fuse_req *req; | |
1574 | pgoff_t index; | |
1575 | loff_t file_size; | |
1576 | unsigned int num; | |
1577 | unsigned int offset; | |
0157443c | 1578 | size_t total_len = 0; |
2d45ba38 MS |
1579 | |
1580 | req = fuse_get_req(fc); | |
1581 | if (IS_ERR(req)) | |
1582 | return PTR_ERR(req); | |
1583 | ||
1584 | offset = outarg->offset & ~PAGE_CACHE_MASK; | |
1585 | ||
1586 | req->in.h.opcode = FUSE_NOTIFY_REPLY; | |
1587 | req->in.h.nodeid = outarg->nodeid; | |
1588 | req->in.numargs = 2; | |
1589 | req->in.argpages = 1; | |
1590 | req->page_offset = offset; | |
1591 | req->end = fuse_retrieve_end; | |
1592 | ||
1593 | index = outarg->offset >> PAGE_CACHE_SHIFT; | |
1594 | file_size = i_size_read(inode); | |
1595 | num = outarg->size; | |
1596 | if (outarg->offset > file_size) | |
1597 | num = 0; | |
1598 | else if (outarg->offset + num > file_size) | |
1599 | num = file_size - outarg->offset; | |
1600 | ||
48706d0a | 1601 | while (num && req->num_pages < FUSE_MAX_PAGES_PER_REQ) { |
2d45ba38 MS |
1602 | struct page *page; |
1603 | unsigned int this_num; | |
1604 | ||
1605 | page = find_get_page(mapping, index); | |
1606 | if (!page) | |
1607 | break; | |
1608 | ||
1609 | this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset); | |
1610 | req->pages[req->num_pages] = page; | |
1611 | req->num_pages++; | |
1612 | ||
c9e67d48 | 1613 | offset = 0; |
2d45ba38 MS |
1614 | num -= this_num; |
1615 | total_len += this_num; | |
48706d0a | 1616 | index++; |
2d45ba38 MS |
1617 | } |
1618 | req->misc.retrieve_in.offset = outarg->offset; | |
1619 | req->misc.retrieve_in.size = total_len; | |
1620 | req->in.args[0].size = sizeof(req->misc.retrieve_in); | |
1621 | req->in.args[0].value = &req->misc.retrieve_in; | |
1622 | req->in.args[1].size = total_len; | |
1623 | ||
1624 | err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique); | |
1625 | if (err) | |
1626 | fuse_retrieve_end(fc, req); | |
1627 | ||
1628 | return err; | |
1629 | } | |
1630 | ||
1631 | static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size, | |
1632 | struct fuse_copy_state *cs) | |
1633 | { | |
1634 | struct fuse_notify_retrieve_out outarg; | |
1635 | struct inode *inode; | |
1636 | int err; | |
1637 | ||
1638 | err = -EINVAL; | |
1639 | if (size != sizeof(outarg)) | |
1640 | goto copy_finish; | |
1641 | ||
1642 | err = fuse_copy_one(cs, &outarg, sizeof(outarg)); | |
1643 | if (err) | |
1644 | goto copy_finish; | |
1645 | ||
1646 | fuse_copy_finish(cs); | |
1647 | ||
1648 | down_read(&fc->killsb); | |
1649 | err = -ENOENT; | |
1650 | if (fc->sb) { | |
1651 | u64 nodeid = outarg.nodeid; | |
1652 | ||
1653 | inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid); | |
1654 | if (inode) { | |
1655 | err = fuse_retrieve(fc, inode, &outarg); | |
1656 | iput(inode); | |
1657 | } | |
1658 | } | |
1659 | up_read(&fc->killsb); | |
1660 | ||
1661 | return err; | |
1662 | ||
1663 | copy_finish: | |
1664 | fuse_copy_finish(cs); | |
1665 | return err; | |
1666 | } | |
1667 | ||
8599396b TH |
1668 | static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code, |
1669 | unsigned int size, struct fuse_copy_state *cs) | |
1670 | { | |
1671 | switch (code) { | |
95668a69 TH |
1672 | case FUSE_NOTIFY_POLL: |
1673 | return fuse_notify_poll(fc, size, cs); | |
1674 | ||
3b463ae0 JM |
1675 | case FUSE_NOTIFY_INVAL_INODE: |
1676 | return fuse_notify_inval_inode(fc, size, cs); | |
1677 | ||
1678 | case FUSE_NOTIFY_INVAL_ENTRY: | |
1679 | return fuse_notify_inval_entry(fc, size, cs); | |
1680 | ||
a1d75f25 MS |
1681 | case FUSE_NOTIFY_STORE: |
1682 | return fuse_notify_store(fc, size, cs); | |
1683 | ||
2d45ba38 MS |
1684 | case FUSE_NOTIFY_RETRIEVE: |
1685 | return fuse_notify_retrieve(fc, size, cs); | |
1686 | ||
451d0f59 JM |
1687 | case FUSE_NOTIFY_DELETE: |
1688 | return fuse_notify_delete(fc, size, cs); | |
1689 | ||
8599396b | 1690 | default: |
f6d47a17 | 1691 | fuse_copy_finish(cs); |
8599396b TH |
1692 | return -EINVAL; |
1693 | } | |
1694 | } | |
1695 | ||
334f485d MS |
1696 | /* Look up request on processing list by unique ID */ |
1697 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
1698 | { | |
1699 | struct list_head *entry; | |
1700 | ||
1701 | list_for_each(entry, &fc->processing) { | |
1702 | struct fuse_req *req; | |
1703 | req = list_entry(entry, struct fuse_req, list); | |
a4d27e75 | 1704 | if (req->in.h.unique == unique || req->intr_unique == unique) |
334f485d MS |
1705 | return req; |
1706 | } | |
1707 | return NULL; | |
1708 | } | |
1709 | ||
1710 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
1711 | unsigned nbytes) | |
1712 | { | |
1713 | unsigned reqsize = sizeof(struct fuse_out_header); | |
1714 | ||
1715 | if (out->h.error) | |
1716 | return nbytes != reqsize ? -EINVAL : 0; | |
1717 | ||
1718 | reqsize += len_args(out->numargs, out->args); | |
1719 | ||
1720 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
1721 | return -EINVAL; | |
1722 | else if (reqsize > nbytes) { | |
1723 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
1724 | unsigned diffsize = reqsize - nbytes; | |
1725 | if (diffsize > lastarg->size) | |
1726 | return -EINVAL; | |
1727 | lastarg->size -= diffsize; | |
1728 | } | |
1729 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
1730 | out->page_zeroing); | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * Write a single reply to a request. First the header is copied from | |
1735 | * the write buffer. The request is then searched on the processing | |
1736 | * list by the unique ID found in the header. If found, then remove | |
1737 | * it from the list and copy the rest of the buffer to the request. | |
1738 | * The request is finished by calling request_end() | |
1739 | */ | |
dd3bb14f MS |
1740 | static ssize_t fuse_dev_do_write(struct fuse_conn *fc, |
1741 | struct fuse_copy_state *cs, size_t nbytes) | |
334f485d MS |
1742 | { |
1743 | int err; | |
334f485d MS |
1744 | struct fuse_req *req; |
1745 | struct fuse_out_header oh; | |
334f485d | 1746 | |
334f485d MS |
1747 | if (nbytes < sizeof(struct fuse_out_header)) |
1748 | return -EINVAL; | |
1749 | ||
dd3bb14f | 1750 | err = fuse_copy_one(cs, &oh, sizeof(oh)); |
334f485d MS |
1751 | if (err) |
1752 | goto err_finish; | |
8599396b TH |
1753 | |
1754 | err = -EINVAL; | |
1755 | if (oh.len != nbytes) | |
1756 | goto err_finish; | |
1757 | ||
1758 | /* | |
1759 | * Zero oh.unique indicates unsolicited notification message | |
1760 | * and error contains notification code. | |
1761 | */ | |
1762 | if (!oh.unique) { | |
dd3bb14f | 1763 | err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs); |
8599396b TH |
1764 | return err ? err : nbytes; |
1765 | } | |
1766 | ||
334f485d | 1767 | err = -EINVAL; |
8599396b | 1768 | if (oh.error <= -1000 || oh.error > 0) |
334f485d MS |
1769 | goto err_finish; |
1770 | ||
d7133114 | 1771 | spin_lock(&fc->lock); |
69a53bf2 MS |
1772 | err = -ENOENT; |
1773 | if (!fc->connected) | |
1774 | goto err_unlock; | |
1775 | ||
334f485d | 1776 | req = request_find(fc, oh.unique); |
334f485d MS |
1777 | if (!req) |
1778 | goto err_unlock; | |
1779 | ||
f9a2842e | 1780 | if (req->aborted) { |
d7133114 | 1781 | spin_unlock(&fc->lock); |
dd3bb14f | 1782 | fuse_copy_finish(cs); |
d7133114 | 1783 | spin_lock(&fc->lock); |
222f1d69 | 1784 | request_end(fc, req); |
334f485d MS |
1785 | return -ENOENT; |
1786 | } | |
a4d27e75 MS |
1787 | /* Is it an interrupt reply? */ |
1788 | if (req->intr_unique == oh.unique) { | |
1789 | err = -EINVAL; | |
1790 | if (nbytes != sizeof(struct fuse_out_header)) | |
1791 | goto err_unlock; | |
1792 | ||
1793 | if (oh.error == -ENOSYS) | |
1794 | fc->no_interrupt = 1; | |
1795 | else if (oh.error == -EAGAIN) | |
1796 | queue_interrupt(fc, req); | |
1797 | ||
1798 | spin_unlock(&fc->lock); | |
dd3bb14f | 1799 | fuse_copy_finish(cs); |
a4d27e75 MS |
1800 | return nbytes; |
1801 | } | |
1802 | ||
1803 | req->state = FUSE_REQ_WRITING; | |
d77a1d5b | 1804 | list_move(&req->list, &fc->io); |
334f485d MS |
1805 | req->out.h = oh; |
1806 | req->locked = 1; | |
dd3bb14f | 1807 | cs->req = req; |
ce534fb0 MS |
1808 | if (!req->out.page_replace) |
1809 | cs->move_pages = 0; | |
d7133114 | 1810 | spin_unlock(&fc->lock); |
334f485d | 1811 | |
dd3bb14f MS |
1812 | err = copy_out_args(cs, &req->out, nbytes); |
1813 | fuse_copy_finish(cs); | |
334f485d | 1814 | |
d7133114 | 1815 | spin_lock(&fc->lock); |
334f485d MS |
1816 | req->locked = 0; |
1817 | if (!err) { | |
f9a2842e | 1818 | if (req->aborted) |
334f485d | 1819 | err = -ENOENT; |
f9a2842e | 1820 | } else if (!req->aborted) |
334f485d MS |
1821 | req->out.h.error = -EIO; |
1822 | request_end(fc, req); | |
1823 | ||
1824 | return err ? err : nbytes; | |
1825 | ||
1826 | err_unlock: | |
d7133114 | 1827 | spin_unlock(&fc->lock); |
334f485d | 1828 | err_finish: |
dd3bb14f | 1829 | fuse_copy_finish(cs); |
334f485d MS |
1830 | return err; |
1831 | } | |
1832 | ||
dd3bb14f MS |
1833 | static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov, |
1834 | unsigned long nr_segs, loff_t pos) | |
1835 | { | |
1836 | struct fuse_copy_state cs; | |
1837 | struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp); | |
1838 | if (!fc) | |
1839 | return -EPERM; | |
1840 | ||
c3021629 | 1841 | fuse_copy_init(&cs, fc, 0, iov, nr_segs); |
dd3bb14f MS |
1842 | |
1843 | return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs)); | |
1844 | } | |
1845 | ||
1846 | static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe, | |
1847 | struct file *out, loff_t *ppos, | |
1848 | size_t len, unsigned int flags) | |
1849 | { | |
1850 | unsigned nbuf; | |
1851 | unsigned idx; | |
1852 | struct pipe_buffer *bufs; | |
1853 | struct fuse_copy_state cs; | |
1854 | struct fuse_conn *fc; | |
1855 | size_t rem; | |
1856 | ssize_t ret; | |
1857 | ||
1858 | fc = fuse_get_conn(out); | |
1859 | if (!fc) | |
1860 | return -EPERM; | |
1861 | ||
07e77dca | 1862 | bufs = kmalloc(pipe->buffers * sizeof(struct pipe_buffer), GFP_KERNEL); |
dd3bb14f MS |
1863 | if (!bufs) |
1864 | return -ENOMEM; | |
1865 | ||
1866 | pipe_lock(pipe); | |
1867 | nbuf = 0; | |
1868 | rem = 0; | |
1869 | for (idx = 0; idx < pipe->nrbufs && rem < len; idx++) | |
1870 | rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len; | |
1871 | ||
1872 | ret = -EINVAL; | |
1873 | if (rem < len) { | |
1874 | pipe_unlock(pipe); | |
1875 | goto out; | |
1876 | } | |
1877 | ||
1878 | rem = len; | |
1879 | while (rem) { | |
1880 | struct pipe_buffer *ibuf; | |
1881 | struct pipe_buffer *obuf; | |
1882 | ||
1883 | BUG_ON(nbuf >= pipe->buffers); | |
1884 | BUG_ON(!pipe->nrbufs); | |
1885 | ibuf = &pipe->bufs[pipe->curbuf]; | |
1886 | obuf = &bufs[nbuf]; | |
1887 | ||
1888 | if (rem >= ibuf->len) { | |
1889 | *obuf = *ibuf; | |
1890 | ibuf->ops = NULL; | |
1891 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); | |
1892 | pipe->nrbufs--; | |
1893 | } else { | |
1894 | ibuf->ops->get(pipe, ibuf); | |
1895 | *obuf = *ibuf; | |
1896 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1897 | obuf->len = rem; | |
1898 | ibuf->offset += obuf->len; | |
1899 | ibuf->len -= obuf->len; | |
1900 | } | |
1901 | nbuf++; | |
1902 | rem -= obuf->len; | |
1903 | } | |
1904 | pipe_unlock(pipe); | |
1905 | ||
c3021629 | 1906 | fuse_copy_init(&cs, fc, 0, NULL, nbuf); |
dd3bb14f | 1907 | cs.pipebufs = bufs; |
dd3bb14f MS |
1908 | cs.pipe = pipe; |
1909 | ||
ce534fb0 MS |
1910 | if (flags & SPLICE_F_MOVE) |
1911 | cs.move_pages = 1; | |
1912 | ||
dd3bb14f MS |
1913 | ret = fuse_dev_do_write(fc, &cs, len); |
1914 | ||
1915 | for (idx = 0; idx < nbuf; idx++) { | |
1916 | struct pipe_buffer *buf = &bufs[idx]; | |
1917 | buf->ops->release(pipe, buf); | |
1918 | } | |
1919 | out: | |
1920 | kfree(bufs); | |
1921 | return ret; | |
1922 | } | |
1923 | ||
334f485d MS |
1924 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) |
1925 | { | |
334f485d | 1926 | unsigned mask = POLLOUT | POLLWRNORM; |
7025d9ad | 1927 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 1928 | if (!fc) |
7025d9ad | 1929 | return POLLERR; |
334f485d MS |
1930 | |
1931 | poll_wait(file, &fc->waitq, wait); | |
1932 | ||
d7133114 | 1933 | spin_lock(&fc->lock); |
7025d9ad MS |
1934 | if (!fc->connected) |
1935 | mask = POLLERR; | |
a4d27e75 | 1936 | else if (request_pending(fc)) |
7025d9ad | 1937 | mask |= POLLIN | POLLRDNORM; |
d7133114 | 1938 | spin_unlock(&fc->lock); |
334f485d MS |
1939 | |
1940 | return mask; | |
1941 | } | |
1942 | ||
69a53bf2 MS |
1943 | /* |
1944 | * Abort all requests on the given list (pending or processing) | |
1945 | * | |
d7133114 | 1946 | * This function releases and reacquires fc->lock |
69a53bf2 | 1947 | */ |
334f485d | 1948 | static void end_requests(struct fuse_conn *fc, struct list_head *head) |
b9ca67b2 MS |
1949 | __releases(fc->lock) |
1950 | __acquires(fc->lock) | |
334f485d MS |
1951 | { |
1952 | while (!list_empty(head)) { | |
1953 | struct fuse_req *req; | |
1954 | req = list_entry(head->next, struct fuse_req, list); | |
334f485d MS |
1955 | req->out.h.error = -ECONNABORTED; |
1956 | request_end(fc, req); | |
d7133114 | 1957 | spin_lock(&fc->lock); |
334f485d MS |
1958 | } |
1959 | } | |
1960 | ||
69a53bf2 MS |
1961 | /* |
1962 | * Abort requests under I/O | |
1963 | * | |
f9a2842e | 1964 | * The requests are set to aborted and finished, and the request |
69a53bf2 MS |
1965 | * waiter is woken up. This will make request_wait_answer() wait |
1966 | * until the request is unlocked and then return. | |
64c6d8ed MS |
1967 | * |
1968 | * If the request is asynchronous, then the end function needs to be | |
1969 | * called after waiting for the request to be unlocked (if it was | |
1970 | * locked). | |
69a53bf2 MS |
1971 | */ |
1972 | static void end_io_requests(struct fuse_conn *fc) | |
b9ca67b2 MS |
1973 | __releases(fc->lock) |
1974 | __acquires(fc->lock) | |
69a53bf2 MS |
1975 | { |
1976 | while (!list_empty(&fc->io)) { | |
64c6d8ed MS |
1977 | struct fuse_req *req = |
1978 | list_entry(fc->io.next, struct fuse_req, list); | |
1979 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
1980 | ||
f9a2842e | 1981 | req->aborted = 1; |
69a53bf2 MS |
1982 | req->out.h.error = -ECONNABORTED; |
1983 | req->state = FUSE_REQ_FINISHED; | |
1984 | list_del_init(&req->list); | |
1985 | wake_up(&req->waitq); | |
64c6d8ed MS |
1986 | if (end) { |
1987 | req->end = NULL; | |
64c6d8ed | 1988 | __fuse_get_request(req); |
d7133114 | 1989 | spin_unlock(&fc->lock); |
64c6d8ed MS |
1990 | wait_event(req->waitq, !req->locked); |
1991 | end(fc, req); | |
e9bb09dd | 1992 | fuse_put_request(fc, req); |
d7133114 | 1993 | spin_lock(&fc->lock); |
64c6d8ed | 1994 | } |
69a53bf2 MS |
1995 | } |
1996 | } | |
1997 | ||
595afaf9 | 1998 | static void end_queued_requests(struct fuse_conn *fc) |
b9ca67b2 MS |
1999 | __releases(fc->lock) |
2000 | __acquires(fc->lock) | |
595afaf9 MS |
2001 | { |
2002 | fc->max_background = UINT_MAX; | |
2003 | flush_bg_queue(fc); | |
2004 | end_requests(fc, &fc->pending); | |
2005 | end_requests(fc, &fc->processing); | |
07e77dca | 2006 | while (forget_pending(fc)) |
02c048b9 | 2007 | kfree(dequeue_forget(fc, 1, NULL)); |
595afaf9 MS |
2008 | } |
2009 | ||
357ccf2b BG |
2010 | static void end_polls(struct fuse_conn *fc) |
2011 | { | |
2012 | struct rb_node *p; | |
2013 | ||
2014 | p = rb_first(&fc->polled_files); | |
2015 | ||
2016 | while (p) { | |
2017 | struct fuse_file *ff; | |
2018 | ff = rb_entry(p, struct fuse_file, polled_node); | |
2019 | wake_up_interruptible_all(&ff->poll_wait); | |
2020 | ||
2021 | p = rb_next(p); | |
2022 | } | |
2023 | } | |
2024 | ||
69a53bf2 MS |
2025 | /* |
2026 | * Abort all requests. | |
2027 | * | |
2028 | * Emergency exit in case of a malicious or accidental deadlock, or | |
2029 | * just a hung filesystem. | |
2030 | * | |
2031 | * The same effect is usually achievable through killing the | |
2032 | * filesystem daemon and all users of the filesystem. The exception | |
2033 | * is the combination of an asynchronous request and the tricky | |
2034 | * deadlock (see Documentation/filesystems/fuse.txt). | |
2035 | * | |
2036 | * During the aborting, progression of requests from the pending and | |
2037 | * processing lists onto the io list, and progression of new requests | |
2038 | * onto the pending list is prevented by req->connected being false. | |
2039 | * | |
2040 | * Progression of requests under I/O to the processing list is | |
f9a2842e MS |
2041 | * prevented by the req->aborted flag being true for these requests. |
2042 | * For this reason requests on the io list must be aborted first. | |
69a53bf2 MS |
2043 | */ |
2044 | void fuse_abort_conn(struct fuse_conn *fc) | |
2045 | { | |
d7133114 | 2046 | spin_lock(&fc->lock); |
69a53bf2 MS |
2047 | if (fc->connected) { |
2048 | fc->connected = 0; | |
51eb01e7 | 2049 | fc->blocked = 0; |
69a53bf2 | 2050 | end_io_requests(fc); |
595afaf9 | 2051 | end_queued_requests(fc); |
357ccf2b | 2052 | end_polls(fc); |
69a53bf2 | 2053 | wake_up_all(&fc->waitq); |
51eb01e7 | 2054 | wake_up_all(&fc->blocked_waitq); |
385a17bf | 2055 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
69a53bf2 | 2056 | } |
d7133114 | 2057 | spin_unlock(&fc->lock); |
69a53bf2 | 2058 | } |
08cbf542 | 2059 | EXPORT_SYMBOL_GPL(fuse_abort_conn); |
69a53bf2 | 2060 | |
08cbf542 | 2061 | int fuse_dev_release(struct inode *inode, struct file *file) |
334f485d | 2062 | { |
0720b315 | 2063 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 2064 | if (fc) { |
d7133114 | 2065 | spin_lock(&fc->lock); |
1e9a4ed9 | 2066 | fc->connected = 0; |
595afaf9 MS |
2067 | fc->blocked = 0; |
2068 | end_queued_requests(fc); | |
357ccf2b | 2069 | end_polls(fc); |
595afaf9 | 2070 | wake_up_all(&fc->blocked_waitq); |
d7133114 | 2071 | spin_unlock(&fc->lock); |
bafa9654 | 2072 | fuse_conn_put(fc); |
385a17bf | 2073 | } |
f543f253 | 2074 | |
334f485d MS |
2075 | return 0; |
2076 | } | |
08cbf542 | 2077 | EXPORT_SYMBOL_GPL(fuse_dev_release); |
334f485d | 2078 | |
385a17bf JD |
2079 | static int fuse_dev_fasync(int fd, struct file *file, int on) |
2080 | { | |
2081 | struct fuse_conn *fc = fuse_get_conn(file); | |
2082 | if (!fc) | |
a87046d8 | 2083 | return -EPERM; |
385a17bf JD |
2084 | |
2085 | /* No locking - fasync_helper does its own locking */ | |
2086 | return fasync_helper(fd, file, on, &fc->fasync); | |
2087 | } | |
2088 | ||
4b6f5d20 | 2089 | const struct file_operations fuse_dev_operations = { |
334f485d MS |
2090 | .owner = THIS_MODULE, |
2091 | .llseek = no_llseek, | |
ee0b3e67 BP |
2092 | .read = do_sync_read, |
2093 | .aio_read = fuse_dev_read, | |
c3021629 | 2094 | .splice_read = fuse_dev_splice_read, |
ee0b3e67 BP |
2095 | .write = do_sync_write, |
2096 | .aio_write = fuse_dev_write, | |
dd3bb14f | 2097 | .splice_write = fuse_dev_splice_write, |
334f485d MS |
2098 | .poll = fuse_dev_poll, |
2099 | .release = fuse_dev_release, | |
385a17bf | 2100 | .fasync = fuse_dev_fasync, |
334f485d | 2101 | }; |
08cbf542 | 2102 | EXPORT_SYMBOL_GPL(fuse_dev_operations); |
334f485d MS |
2103 | |
2104 | static struct miscdevice fuse_miscdevice = { | |
2105 | .minor = FUSE_MINOR, | |
2106 | .name = "fuse", | |
2107 | .fops = &fuse_dev_operations, | |
2108 | }; | |
2109 | ||
2110 | int __init fuse_dev_init(void) | |
2111 | { | |
2112 | int err = -ENOMEM; | |
2113 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
2114 | sizeof(struct fuse_req), | |
20c2df83 | 2115 | 0, 0, NULL); |
334f485d MS |
2116 | if (!fuse_req_cachep) |
2117 | goto out; | |
2118 | ||
2119 | err = misc_register(&fuse_miscdevice); | |
2120 | if (err) | |
2121 | goto out_cache_clean; | |
2122 | ||
2123 | return 0; | |
2124 | ||
2125 | out_cache_clean: | |
2126 | kmem_cache_destroy(fuse_req_cachep); | |
2127 | out: | |
2128 | return err; | |
2129 | } | |
2130 | ||
2131 | void fuse_dev_cleanup(void) | |
2132 | { | |
2133 | misc_deregister(&fuse_miscdevice); | |
2134 | kmem_cache_destroy(fuse_req_cachep); | |
2135 | } |