Commit | Line | Data |
---|---|---|
334f485d MS |
1 | /* |
2 | FUSE: Filesystem in Userspace | |
3 | Copyright (C) 2001-2005 Miklos Szeredi <miklos@szeredi.hu> | |
4 | ||
5 | This program can be distributed under the terms of the GNU GPL. | |
6 | See the file COPYING. | |
7 | */ | |
8 | ||
9 | #include "fuse_i.h" | |
10 | ||
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/poll.h> | |
14 | #include <linux/uio.h> | |
15 | #include <linux/miscdevice.h> | |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/file.h> | |
18 | #include <linux/slab.h> | |
19 | ||
20 | MODULE_ALIAS_MISCDEV(FUSE_MINOR); | |
21 | ||
22 | static kmem_cache_t *fuse_req_cachep; | |
23 | ||
8bfc016d | 24 | static struct fuse_conn *fuse_get_conn(struct file *file) |
334f485d | 25 | { |
0720b315 MS |
26 | /* |
27 | * Lockless access is OK, because file->private data is set | |
28 | * once during mount and is valid until the file is released. | |
29 | */ | |
30 | return file->private_data; | |
334f485d MS |
31 | } |
32 | ||
8bfc016d | 33 | static void fuse_request_init(struct fuse_req *req) |
334f485d MS |
34 | { |
35 | memset(req, 0, sizeof(*req)); | |
36 | INIT_LIST_HEAD(&req->list); | |
37 | init_waitqueue_head(&req->waitq); | |
38 | atomic_set(&req->count, 1); | |
39 | } | |
40 | ||
41 | struct fuse_req *fuse_request_alloc(void) | |
42 | { | |
43 | struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, SLAB_KERNEL); | |
44 | if (req) | |
45 | fuse_request_init(req); | |
46 | return req; | |
47 | } | |
48 | ||
49 | void fuse_request_free(struct fuse_req *req) | |
50 | { | |
51 | kmem_cache_free(fuse_req_cachep, req); | |
52 | } | |
53 | ||
8bfc016d | 54 | static void block_sigs(sigset_t *oldset) |
334f485d MS |
55 | { |
56 | sigset_t mask; | |
57 | ||
58 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
59 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
60 | } | |
61 | ||
8bfc016d | 62 | static void restore_sigs(sigset_t *oldset) |
334f485d MS |
63 | { |
64 | sigprocmask(SIG_SETMASK, oldset, NULL); | |
65 | } | |
66 | ||
77e7f250 MS |
67 | /* |
68 | * Reset request, so that it can be reused | |
69 | * | |
70 | * The caller must be _very_ careful to make sure, that it is holding | |
71 | * the only reference to req | |
72 | */ | |
334f485d MS |
73 | void fuse_reset_request(struct fuse_req *req) |
74 | { | |
75 | int preallocated = req->preallocated; | |
76 | BUG_ON(atomic_read(&req->count) != 1); | |
77 | fuse_request_init(req); | |
78 | req->preallocated = preallocated; | |
79 | } | |
80 | ||
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 | ||
93 | static struct fuse_req *do_get_request(struct fuse_conn *fc) | |
94 | { | |
95 | struct fuse_req *req; | |
96 | ||
97 | spin_lock(&fuse_lock); | |
98 | BUG_ON(list_empty(&fc->unused_list)); | |
99 | req = list_entry(fc->unused_list.next, struct fuse_req, list); | |
100 | list_del_init(&req->list); | |
101 | spin_unlock(&fuse_lock); | |
102 | fuse_request_init(req); | |
103 | req->preallocated = 1; | |
104 | req->in.h.uid = current->fsuid; | |
105 | req->in.h.gid = current->fsgid; | |
106 | req->in.h.pid = current->pid; | |
107 | return req; | |
108 | } | |
109 | ||
7c352bdf | 110 | /* This can return NULL, but only in case it's interrupted by a SIGKILL */ |
334f485d | 111 | struct fuse_req *fuse_get_request(struct fuse_conn *fc) |
334f485d MS |
112 | { |
113 | int intr; | |
114 | sigset_t oldset; | |
115 | ||
0cd5b885 | 116 | atomic_inc(&fc->num_waiting); |
334f485d MS |
117 | block_sigs(&oldset); |
118 | intr = down_interruptible(&fc->outstanding_sem); | |
119 | restore_sigs(&oldset); | |
0cd5b885 MS |
120 | if (intr) { |
121 | atomic_dec(&fc->num_waiting); | |
122 | return NULL; | |
123 | } | |
124 | return do_get_request(fc); | |
334f485d MS |
125 | } |
126 | ||
7128ec2a | 127 | /* Must be called with fuse_lock held */ |
334f485d MS |
128 | static void fuse_putback_request(struct fuse_conn *fc, struct fuse_req *req) |
129 | { | |
0cd5b885 MS |
130 | if (req->preallocated) { |
131 | atomic_dec(&fc->num_waiting); | |
334f485d | 132 | list_add(&req->list, &fc->unused_list); |
0cd5b885 | 133 | } else |
334f485d MS |
134 | fuse_request_free(req); |
135 | ||
136 | /* If we are in debt decrease that first */ | |
137 | if (fc->outstanding_debt) | |
138 | fc->outstanding_debt--; | |
139 | else | |
140 | up(&fc->outstanding_sem); | |
334f485d MS |
141 | } |
142 | ||
143 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) | |
7128ec2a MS |
144 | { |
145 | if (atomic_dec_and_test(&req->count)) { | |
146 | spin_lock(&fuse_lock); | |
147 | fuse_putback_request(fc, req); | |
148 | spin_unlock(&fuse_lock); | |
149 | } | |
150 | } | |
151 | ||
152 | static void fuse_put_request_locked(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d MS |
153 | { |
154 | if (atomic_dec_and_test(&req->count)) | |
155 | fuse_putback_request(fc, req); | |
156 | } | |
157 | ||
1e9a4ed9 MS |
158 | void fuse_release_background(struct fuse_req *req) |
159 | { | |
160 | iput(req->inode); | |
161 | iput(req->inode2); | |
162 | if (req->file) | |
163 | fput(req->file); | |
164 | spin_lock(&fuse_lock); | |
165 | list_del(&req->bg_entry); | |
166 | spin_unlock(&fuse_lock); | |
167 | } | |
168 | ||
334f485d MS |
169 | /* |
170 | * This function is called when a request is finished. Either a reply | |
171 | * has arrived or it was interrupted (and not yet sent) or some error | |
f43b155a MS |
172 | * occurred during communication with userspace, or the device file |
173 | * was closed. In case of a background request the reference to the | |
174 | * stored objects are released. The requester thread is woken up (if | |
64c6d8ed MS |
175 | * still waiting), the 'end' callback is called if given, else the |
176 | * reference to the request is released | |
334f485d | 177 | * |
7128ec2a MS |
178 | * Releasing extra reference for foreground requests must be done |
179 | * within the same locked region as setting state to finished. This | |
180 | * is because fuse_reset_request() may be called after request is | |
181 | * finished and it must be the sole possessor. If request is | |
182 | * interrupted and put in the background, it will return with an error | |
183 | * and hence never be reset and reused. | |
184 | * | |
334f485d MS |
185 | * Called with fuse_lock, unlocks it |
186 | */ | |
187 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
188 | { | |
d77a1d5b | 189 | list_del(&req->list); |
83cfd493 | 190 | req->state = FUSE_REQ_FINISHED; |
7128ec2a MS |
191 | if (!req->background) { |
192 | wake_up(&req->waitq); | |
193 | fuse_put_request_locked(fc, req); | |
194 | spin_unlock(&fuse_lock); | |
195 | } else { | |
196 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
197 | req->end = NULL; | |
198 | spin_unlock(&fuse_lock); | |
1e9a4ed9 MS |
199 | down_read(&fc->sbput_sem); |
200 | if (fc->mounted) | |
201 | fuse_release_background(req); | |
202 | up_read(&fc->sbput_sem); | |
7128ec2a MS |
203 | if (end) |
204 | end(fc, req); | |
205 | else | |
206 | fuse_put_request(fc, req); | |
334f485d | 207 | } |
334f485d MS |
208 | } |
209 | ||
1e9a4ed9 MS |
210 | /* |
211 | * Unfortunately request interruption not just solves the deadlock | |
212 | * problem, it causes problems too. These stem from the fact, that an | |
213 | * interrupted request is continued to be processed in userspace, | |
214 | * while all the locks and object references (inode and file) held | |
215 | * during the operation are released. | |
216 | * | |
217 | * To release the locks is exactly why there's a need to interrupt the | |
218 | * request, so there's not a lot that can be done about this, except | |
219 | * introduce additional locking in userspace. | |
220 | * | |
221 | * More important is to keep inode and file references until userspace | |
222 | * has replied, otherwise FORGET and RELEASE could be sent while the | |
223 | * inode/file is still used by the filesystem. | |
224 | * | |
225 | * For this reason the concept of "background" request is introduced. | |
226 | * An interrupted request is backgrounded if it has been already sent | |
227 | * to userspace. Backgrounding involves getting an extra reference to | |
228 | * inode(s) or file used in the request, and adding the request to | |
229 | * fc->background list. When a reply is received for a background | |
230 | * request, the object references are released, and the request is | |
231 | * removed from the list. If the filesystem is unmounted while there | |
232 | * are still background requests, the list is walked and references | |
233 | * are released as if a reply was received. | |
234 | * | |
235 | * There's one more use for a background request. The RELEASE message is | |
236 | * always sent as background, since it doesn't return an error or | |
237 | * data. | |
238 | */ | |
239 | static void background_request(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d | 240 | { |
334f485d | 241 | req->background = 1; |
1e9a4ed9 | 242 | list_add(&req->bg_entry, &fc->background); |
334f485d MS |
243 | if (req->inode) |
244 | req->inode = igrab(req->inode); | |
245 | if (req->inode2) | |
246 | req->inode2 = igrab(req->inode2); | |
247 | if (req->file) | |
248 | get_file(req->file); | |
249 | } | |
250 | ||
334f485d | 251 | /* Called with fuse_lock held. Releases, and then reacquires it. */ |
7c352bdf | 252 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) |
334f485d | 253 | { |
7c352bdf | 254 | sigset_t oldset; |
334f485d MS |
255 | |
256 | spin_unlock(&fuse_lock); | |
7c352bdf | 257 | block_sigs(&oldset); |
83cfd493 | 258 | wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED); |
7c352bdf | 259 | restore_sigs(&oldset); |
334f485d | 260 | spin_lock(&fuse_lock); |
69a53bf2 | 261 | if (req->state == FUSE_REQ_FINISHED && !req->interrupted) |
334f485d MS |
262 | return; |
263 | ||
69a53bf2 MS |
264 | if (!req->interrupted) { |
265 | req->out.h.error = -EINTR; | |
266 | req->interrupted = 1; | |
267 | } | |
334f485d MS |
268 | if (req->locked) { |
269 | /* This is uninterruptible sleep, because data is | |
270 | being copied to/from the buffers of req. During | |
271 | locked state, there mustn't be any filesystem | |
272 | operation (e.g. page fault), since that could lead | |
273 | to deadlock */ | |
274 | spin_unlock(&fuse_lock); | |
275 | wait_event(req->waitq, !req->locked); | |
276 | spin_lock(&fuse_lock); | |
277 | } | |
83cfd493 | 278 | if (req->state == FUSE_REQ_PENDING) { |
334f485d MS |
279 | list_del(&req->list); |
280 | __fuse_put_request(req); | |
83cfd493 | 281 | } else if (req->state == FUSE_REQ_SENT) |
1e9a4ed9 | 282 | background_request(fc, req); |
334f485d MS |
283 | } |
284 | ||
285 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) | |
286 | { | |
287 | unsigned nbytes = 0; | |
288 | unsigned i; | |
289 | ||
290 | for (i = 0; i < numargs; i++) | |
291 | nbytes += args[i].size; | |
292 | ||
293 | return nbytes; | |
294 | } | |
295 | ||
296 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) | |
297 | { | |
298 | fc->reqctr++; | |
299 | /* zero is special */ | |
300 | if (fc->reqctr == 0) | |
301 | fc->reqctr = 1; | |
302 | req->in.h.unique = fc->reqctr; | |
303 | req->in.h.len = sizeof(struct fuse_in_header) + | |
304 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
305 | if (!req->preallocated) { | |
306 | /* If request is not preallocated (either FORGET or | |
307 | RELEASE), then still decrease outstanding_sem, so | |
308 | user can't open infinite number of files while not | |
309 | processing the RELEASE requests. However for | |
310 | efficiency do it without blocking, so if down() | |
311 | would block, just increase the debt instead */ | |
312 | if (down_trylock(&fc->outstanding_sem)) | |
313 | fc->outstanding_debt++; | |
314 | } | |
315 | list_add_tail(&req->list, &fc->pending); | |
83cfd493 | 316 | req->state = FUSE_REQ_PENDING; |
334f485d | 317 | wake_up(&fc->waitq); |
385a17bf | 318 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
334f485d MS |
319 | } |
320 | ||
7c352bdf MS |
321 | /* |
322 | * This can only be interrupted by a SIGKILL | |
323 | */ | |
324 | void request_send(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d MS |
325 | { |
326 | req->isreply = 1; | |
327 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 328 | if (!fc->connected) |
334f485d MS |
329 | req->out.h.error = -ENOTCONN; |
330 | else if (fc->conn_error) | |
331 | req->out.h.error = -ECONNREFUSED; | |
332 | else { | |
333 | queue_request(fc, req); | |
334 | /* acquire extra reference, since request is still needed | |
335 | after request_end() */ | |
336 | __fuse_get_request(req); | |
337 | ||
7c352bdf | 338 | request_wait_answer(fc, req); |
334f485d MS |
339 | } |
340 | spin_unlock(&fuse_lock); | |
341 | } | |
342 | ||
334f485d MS |
343 | static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
344 | { | |
345 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 346 | if (fc->connected) { |
334f485d MS |
347 | queue_request(fc, req); |
348 | spin_unlock(&fuse_lock); | |
349 | } else { | |
350 | req->out.h.error = -ENOTCONN; | |
351 | request_end(fc, req); | |
352 | } | |
353 | } | |
354 | ||
355 | void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req) | |
356 | { | |
357 | req->isreply = 0; | |
358 | request_send_nowait(fc, req); | |
359 | } | |
360 | ||
361 | void request_send_background(struct fuse_conn *fc, struct fuse_req *req) | |
362 | { | |
363 | req->isreply = 1; | |
1e9a4ed9 MS |
364 | spin_lock(&fuse_lock); |
365 | background_request(fc, req); | |
366 | spin_unlock(&fuse_lock); | |
334f485d MS |
367 | request_send_nowait(fc, req); |
368 | } | |
369 | ||
334f485d MS |
370 | /* |
371 | * Lock the request. Up to the next unlock_request() there mustn't be | |
372 | * anything that could cause a page-fault. If the request was already | |
373 | * interrupted bail out. | |
374 | */ | |
8bfc016d | 375 | static int lock_request(struct fuse_req *req) |
334f485d MS |
376 | { |
377 | int err = 0; | |
378 | if (req) { | |
379 | spin_lock(&fuse_lock); | |
380 | if (req->interrupted) | |
381 | err = -ENOENT; | |
382 | else | |
383 | req->locked = 1; | |
384 | spin_unlock(&fuse_lock); | |
385 | } | |
386 | return err; | |
387 | } | |
388 | ||
389 | /* | |
390 | * Unlock request. If it was interrupted during being locked, the | |
391 | * requester thread is currently waiting for it to be unlocked, so | |
392 | * wake it up. | |
393 | */ | |
8bfc016d | 394 | static void unlock_request(struct fuse_req *req) |
334f485d MS |
395 | { |
396 | if (req) { | |
397 | spin_lock(&fuse_lock); | |
398 | req->locked = 0; | |
399 | if (req->interrupted) | |
400 | wake_up(&req->waitq); | |
401 | spin_unlock(&fuse_lock); | |
402 | } | |
403 | } | |
404 | ||
405 | struct fuse_copy_state { | |
406 | int write; | |
407 | struct fuse_req *req; | |
408 | const struct iovec *iov; | |
409 | unsigned long nr_segs; | |
410 | unsigned long seglen; | |
411 | unsigned long addr; | |
412 | struct page *pg; | |
413 | void *mapaddr; | |
414 | void *buf; | |
415 | unsigned len; | |
416 | }; | |
417 | ||
418 | static void fuse_copy_init(struct fuse_copy_state *cs, int write, | |
419 | struct fuse_req *req, const struct iovec *iov, | |
420 | unsigned long nr_segs) | |
421 | { | |
422 | memset(cs, 0, sizeof(*cs)); | |
423 | cs->write = write; | |
424 | cs->req = req; | |
425 | cs->iov = iov; | |
426 | cs->nr_segs = nr_segs; | |
427 | } | |
428 | ||
429 | /* Unmap and put previous page of userspace buffer */ | |
8bfc016d | 430 | static void fuse_copy_finish(struct fuse_copy_state *cs) |
334f485d MS |
431 | { |
432 | if (cs->mapaddr) { | |
433 | kunmap_atomic(cs->mapaddr, KM_USER0); | |
434 | if (cs->write) { | |
435 | flush_dcache_page(cs->pg); | |
436 | set_page_dirty_lock(cs->pg); | |
437 | } | |
438 | put_page(cs->pg); | |
439 | cs->mapaddr = NULL; | |
440 | } | |
441 | } | |
442 | ||
443 | /* | |
444 | * Get another pagefull of userspace buffer, and map it to kernel | |
445 | * address space, and lock request | |
446 | */ | |
447 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
448 | { | |
449 | unsigned long offset; | |
450 | int err; | |
451 | ||
452 | unlock_request(cs->req); | |
453 | fuse_copy_finish(cs); | |
454 | if (!cs->seglen) { | |
455 | BUG_ON(!cs->nr_segs); | |
456 | cs->seglen = cs->iov[0].iov_len; | |
457 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
458 | cs->iov ++; | |
459 | cs->nr_segs --; | |
460 | } | |
461 | down_read(¤t->mm->mmap_sem); | |
462 | err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0, | |
463 | &cs->pg, NULL); | |
464 | up_read(¤t->mm->mmap_sem); | |
465 | if (err < 0) | |
466 | return err; | |
467 | BUG_ON(err != 1); | |
468 | offset = cs->addr % PAGE_SIZE; | |
469 | cs->mapaddr = kmap_atomic(cs->pg, KM_USER0); | |
470 | cs->buf = cs->mapaddr + offset; | |
471 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
472 | cs->seglen -= cs->len; | |
473 | cs->addr += cs->len; | |
474 | ||
475 | return lock_request(cs->req); | |
476 | } | |
477 | ||
478 | /* Do as much copy to/from userspace buffer as we can */ | |
8bfc016d | 479 | static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) |
334f485d MS |
480 | { |
481 | unsigned ncpy = min(*size, cs->len); | |
482 | if (val) { | |
483 | if (cs->write) | |
484 | memcpy(cs->buf, *val, ncpy); | |
485 | else | |
486 | memcpy(*val, cs->buf, ncpy); | |
487 | *val += ncpy; | |
488 | } | |
489 | *size -= ncpy; | |
490 | cs->len -= ncpy; | |
491 | cs->buf += ncpy; | |
492 | return ncpy; | |
493 | } | |
494 | ||
495 | /* | |
496 | * Copy a page in the request to/from the userspace buffer. Must be | |
497 | * done atomically | |
498 | */ | |
8bfc016d MS |
499 | static int fuse_copy_page(struct fuse_copy_state *cs, struct page *page, |
500 | unsigned offset, unsigned count, int zeroing) | |
334f485d MS |
501 | { |
502 | if (page && zeroing && count < PAGE_SIZE) { | |
503 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
504 | memset(mapaddr, 0, PAGE_SIZE); | |
505 | kunmap_atomic(mapaddr, KM_USER1); | |
506 | } | |
507 | while (count) { | |
508 | int err; | |
509 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
510 | return err; | |
511 | if (page) { | |
512 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
513 | void *buf = mapaddr + offset; | |
514 | offset += fuse_copy_do(cs, &buf, &count); | |
515 | kunmap_atomic(mapaddr, KM_USER1); | |
516 | } else | |
517 | offset += fuse_copy_do(cs, NULL, &count); | |
518 | } | |
519 | if (page && !cs->write) | |
520 | flush_dcache_page(page); | |
521 | return 0; | |
522 | } | |
523 | ||
524 | /* Copy pages in the request to/from userspace buffer */ | |
525 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
526 | int zeroing) | |
527 | { | |
528 | unsigned i; | |
529 | struct fuse_req *req = cs->req; | |
530 | unsigned offset = req->page_offset; | |
531 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
532 | ||
533 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
534 | struct page *page = req->pages[i]; | |
535 | int err = fuse_copy_page(cs, page, offset, count, zeroing); | |
536 | if (err) | |
537 | return err; | |
538 | ||
539 | nbytes -= count; | |
540 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
541 | offset = 0; | |
542 | } | |
543 | return 0; | |
544 | } | |
545 | ||
546 | /* Copy a single argument in the request to/from userspace buffer */ | |
547 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
548 | { | |
549 | while (size) { | |
550 | int err; | |
551 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
552 | return err; | |
553 | fuse_copy_do(cs, &val, &size); | |
554 | } | |
555 | return 0; | |
556 | } | |
557 | ||
558 | /* Copy request arguments to/from userspace buffer */ | |
559 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
560 | unsigned argpages, struct fuse_arg *args, | |
561 | int zeroing) | |
562 | { | |
563 | int err = 0; | |
564 | unsigned i; | |
565 | ||
566 | for (i = 0; !err && i < numargs; i++) { | |
567 | struct fuse_arg *arg = &args[i]; | |
568 | if (i == numargs - 1 && argpages) | |
569 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
570 | else | |
571 | err = fuse_copy_one(cs, arg->value, arg->size); | |
572 | } | |
573 | return err; | |
574 | } | |
575 | ||
576 | /* Wait until a request is available on the pending list */ | |
577 | static void request_wait(struct fuse_conn *fc) | |
578 | { | |
579 | DECLARE_WAITQUEUE(wait, current); | |
580 | ||
581 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
9ba7cbba | 582 | while (fc->connected && list_empty(&fc->pending)) { |
334f485d MS |
583 | set_current_state(TASK_INTERRUPTIBLE); |
584 | if (signal_pending(current)) | |
585 | break; | |
586 | ||
587 | spin_unlock(&fuse_lock); | |
588 | schedule(); | |
589 | spin_lock(&fuse_lock); | |
590 | } | |
591 | set_current_state(TASK_RUNNING); | |
592 | remove_wait_queue(&fc->waitq, &wait); | |
593 | } | |
594 | ||
595 | /* | |
596 | * Read a single request into the userspace filesystem's buffer. This | |
597 | * function waits until a request is available, then removes it from | |
598 | * the pending list and copies request data to userspace buffer. If | |
599 | * no reply is needed (FORGET) or request has been interrupted or | |
600 | * there was an error during the copying then it's finished by calling | |
601 | * request_end(). Otherwise add it to the processing list, and set | |
602 | * the 'sent' flag. | |
603 | */ | |
604 | static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov, | |
605 | unsigned long nr_segs, loff_t *off) | |
606 | { | |
607 | int err; | |
334f485d MS |
608 | struct fuse_req *req; |
609 | struct fuse_in *in; | |
610 | struct fuse_copy_state cs; | |
611 | unsigned reqsize; | |
0720b315 MS |
612 | struct fuse_conn *fc = fuse_get_conn(file); |
613 | if (!fc) | |
614 | return -EPERM; | |
334f485d | 615 | |
1d3d752b | 616 | restart: |
334f485d | 617 | spin_lock(&fuse_lock); |
e5ac1d1e JD |
618 | err = -EAGAIN; |
619 | if ((file->f_flags & O_NONBLOCK) && fc->connected && | |
620 | list_empty(&fc->pending)) | |
621 | goto err_unlock; | |
622 | ||
334f485d MS |
623 | request_wait(fc); |
624 | err = -ENODEV; | |
9ba7cbba | 625 | if (!fc->connected) |
334f485d MS |
626 | goto err_unlock; |
627 | err = -ERESTARTSYS; | |
628 | if (list_empty(&fc->pending)) | |
629 | goto err_unlock; | |
630 | ||
631 | req = list_entry(fc->pending.next, struct fuse_req, list); | |
83cfd493 | 632 | req->state = FUSE_REQ_READING; |
d77a1d5b | 633 | list_move(&req->list, &fc->io); |
334f485d MS |
634 | |
635 | in = &req->in; | |
1d3d752b MS |
636 | reqsize = in->h.len; |
637 | /* If request is too large, reply with an error and restart the read */ | |
638 | if (iov_length(iov, nr_segs) < reqsize) { | |
639 | req->out.h.error = -EIO; | |
640 | /* SETXATTR is special, since it may contain too large data */ | |
641 | if (in->h.opcode == FUSE_SETXATTR) | |
642 | req->out.h.error = -E2BIG; | |
643 | request_end(fc, req); | |
644 | goto restart; | |
334f485d | 645 | } |
1d3d752b MS |
646 | spin_unlock(&fuse_lock); |
647 | fuse_copy_init(&cs, 1, req, iov, nr_segs); | |
648 | err = fuse_copy_one(&cs, &in->h, sizeof(in->h)); | |
649 | if (!err) | |
650 | err = fuse_copy_args(&cs, in->numargs, in->argpages, | |
651 | (struct fuse_arg *) in->args, 0); | |
334f485d | 652 | fuse_copy_finish(&cs); |
334f485d MS |
653 | spin_lock(&fuse_lock); |
654 | req->locked = 0; | |
655 | if (!err && req->interrupted) | |
656 | err = -ENOENT; | |
657 | if (err) { | |
658 | if (!req->interrupted) | |
659 | req->out.h.error = -EIO; | |
660 | request_end(fc, req); | |
661 | return err; | |
662 | } | |
663 | if (!req->isreply) | |
664 | request_end(fc, req); | |
665 | else { | |
83cfd493 | 666 | req->state = FUSE_REQ_SENT; |
d77a1d5b | 667 | list_move_tail(&req->list, &fc->processing); |
334f485d MS |
668 | spin_unlock(&fuse_lock); |
669 | } | |
670 | return reqsize; | |
671 | ||
672 | err_unlock: | |
673 | spin_unlock(&fuse_lock); | |
674 | return err; | |
675 | } | |
676 | ||
677 | static ssize_t fuse_dev_read(struct file *file, char __user *buf, | |
678 | size_t nbytes, loff_t *off) | |
679 | { | |
680 | struct iovec iov; | |
681 | iov.iov_len = nbytes; | |
682 | iov.iov_base = buf; | |
683 | return fuse_dev_readv(file, &iov, 1, off); | |
684 | } | |
685 | ||
686 | /* Look up request on processing list by unique ID */ | |
687 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
688 | { | |
689 | struct list_head *entry; | |
690 | ||
691 | list_for_each(entry, &fc->processing) { | |
692 | struct fuse_req *req; | |
693 | req = list_entry(entry, struct fuse_req, list); | |
694 | if (req->in.h.unique == unique) | |
695 | return req; | |
696 | } | |
697 | return NULL; | |
698 | } | |
699 | ||
700 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
701 | unsigned nbytes) | |
702 | { | |
703 | unsigned reqsize = sizeof(struct fuse_out_header); | |
704 | ||
705 | if (out->h.error) | |
706 | return nbytes != reqsize ? -EINVAL : 0; | |
707 | ||
708 | reqsize += len_args(out->numargs, out->args); | |
709 | ||
710 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
711 | return -EINVAL; | |
712 | else if (reqsize > nbytes) { | |
713 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
714 | unsigned diffsize = reqsize - nbytes; | |
715 | if (diffsize > lastarg->size) | |
716 | return -EINVAL; | |
717 | lastarg->size -= diffsize; | |
718 | } | |
719 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
720 | out->page_zeroing); | |
721 | } | |
722 | ||
723 | /* | |
724 | * Write a single reply to a request. First the header is copied from | |
725 | * the write buffer. The request is then searched on the processing | |
726 | * list by the unique ID found in the header. If found, then remove | |
727 | * it from the list and copy the rest of the buffer to the request. | |
728 | * The request is finished by calling request_end() | |
729 | */ | |
730 | static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov, | |
731 | unsigned long nr_segs, loff_t *off) | |
732 | { | |
733 | int err; | |
734 | unsigned nbytes = iov_length(iov, nr_segs); | |
735 | struct fuse_req *req; | |
736 | struct fuse_out_header oh; | |
737 | struct fuse_copy_state cs; | |
738 | struct fuse_conn *fc = fuse_get_conn(file); | |
739 | if (!fc) | |
740 | return -ENODEV; | |
741 | ||
742 | fuse_copy_init(&cs, 0, NULL, iov, nr_segs); | |
743 | if (nbytes < sizeof(struct fuse_out_header)) | |
744 | return -EINVAL; | |
745 | ||
746 | err = fuse_copy_one(&cs, &oh, sizeof(oh)); | |
747 | if (err) | |
748 | goto err_finish; | |
749 | err = -EINVAL; | |
750 | if (!oh.unique || oh.error <= -1000 || oh.error > 0 || | |
751 | oh.len != nbytes) | |
752 | goto err_finish; | |
753 | ||
754 | spin_lock(&fuse_lock); | |
69a53bf2 MS |
755 | err = -ENOENT; |
756 | if (!fc->connected) | |
757 | goto err_unlock; | |
758 | ||
334f485d MS |
759 | req = request_find(fc, oh.unique); |
760 | err = -EINVAL; | |
761 | if (!req) | |
762 | goto err_unlock; | |
763 | ||
334f485d | 764 | if (req->interrupted) { |
222f1d69 | 765 | spin_unlock(&fuse_lock); |
334f485d | 766 | fuse_copy_finish(&cs); |
222f1d69 MS |
767 | spin_lock(&fuse_lock); |
768 | request_end(fc, req); | |
334f485d MS |
769 | return -ENOENT; |
770 | } | |
d77a1d5b | 771 | list_move(&req->list, &fc->io); |
334f485d MS |
772 | req->out.h = oh; |
773 | req->locked = 1; | |
774 | cs.req = req; | |
775 | spin_unlock(&fuse_lock); | |
776 | ||
777 | err = copy_out_args(&cs, &req->out, nbytes); | |
778 | fuse_copy_finish(&cs); | |
779 | ||
780 | spin_lock(&fuse_lock); | |
781 | req->locked = 0; | |
782 | if (!err) { | |
783 | if (req->interrupted) | |
784 | err = -ENOENT; | |
785 | } else if (!req->interrupted) | |
786 | req->out.h.error = -EIO; | |
787 | request_end(fc, req); | |
788 | ||
789 | return err ? err : nbytes; | |
790 | ||
791 | err_unlock: | |
792 | spin_unlock(&fuse_lock); | |
793 | err_finish: | |
794 | fuse_copy_finish(&cs); | |
795 | return err; | |
796 | } | |
797 | ||
798 | static ssize_t fuse_dev_write(struct file *file, const char __user *buf, | |
799 | size_t nbytes, loff_t *off) | |
800 | { | |
801 | struct iovec iov; | |
802 | iov.iov_len = nbytes; | |
803 | iov.iov_base = (char __user *) buf; | |
804 | return fuse_dev_writev(file, &iov, 1, off); | |
805 | } | |
806 | ||
807 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) | |
808 | { | |
334f485d | 809 | unsigned mask = POLLOUT | POLLWRNORM; |
7025d9ad | 810 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 811 | if (!fc) |
7025d9ad | 812 | return POLLERR; |
334f485d MS |
813 | |
814 | poll_wait(file, &fc->waitq, wait); | |
815 | ||
816 | spin_lock(&fuse_lock); | |
7025d9ad MS |
817 | if (!fc->connected) |
818 | mask = POLLERR; | |
819 | else if (!list_empty(&fc->pending)) | |
820 | mask |= POLLIN | POLLRDNORM; | |
334f485d MS |
821 | spin_unlock(&fuse_lock); |
822 | ||
823 | return mask; | |
824 | } | |
825 | ||
69a53bf2 MS |
826 | /* |
827 | * Abort all requests on the given list (pending or processing) | |
828 | * | |
829 | * This function releases and reacquires fuse_lock | |
830 | */ | |
334f485d MS |
831 | static void end_requests(struct fuse_conn *fc, struct list_head *head) |
832 | { | |
833 | while (!list_empty(head)) { | |
834 | struct fuse_req *req; | |
835 | req = list_entry(head->next, struct fuse_req, list); | |
334f485d MS |
836 | req->out.h.error = -ECONNABORTED; |
837 | request_end(fc, req); | |
838 | spin_lock(&fuse_lock); | |
839 | } | |
840 | } | |
841 | ||
69a53bf2 MS |
842 | /* |
843 | * Abort requests under I/O | |
844 | * | |
845 | * The requests are set to interrupted and finished, and the request | |
846 | * waiter is woken up. This will make request_wait_answer() wait | |
847 | * until the request is unlocked and then return. | |
64c6d8ed MS |
848 | * |
849 | * If the request is asynchronous, then the end function needs to be | |
850 | * called after waiting for the request to be unlocked (if it was | |
851 | * locked). | |
69a53bf2 MS |
852 | */ |
853 | static void end_io_requests(struct fuse_conn *fc) | |
854 | { | |
855 | while (!list_empty(&fc->io)) { | |
64c6d8ed MS |
856 | struct fuse_req *req = |
857 | list_entry(fc->io.next, struct fuse_req, list); | |
858 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
859 | ||
69a53bf2 MS |
860 | req->interrupted = 1; |
861 | req->out.h.error = -ECONNABORTED; | |
862 | req->state = FUSE_REQ_FINISHED; | |
863 | list_del_init(&req->list); | |
864 | wake_up(&req->waitq); | |
64c6d8ed MS |
865 | if (end) { |
866 | req->end = NULL; | |
867 | /* The end function will consume this reference */ | |
868 | __fuse_get_request(req); | |
869 | spin_unlock(&fuse_lock); | |
870 | wait_event(req->waitq, !req->locked); | |
871 | end(fc, req); | |
872 | spin_lock(&fuse_lock); | |
873 | } | |
69a53bf2 MS |
874 | } |
875 | } | |
876 | ||
877 | /* | |
878 | * Abort all requests. | |
879 | * | |
880 | * Emergency exit in case of a malicious or accidental deadlock, or | |
881 | * just a hung filesystem. | |
882 | * | |
883 | * The same effect is usually achievable through killing the | |
884 | * filesystem daemon and all users of the filesystem. The exception | |
885 | * is the combination of an asynchronous request and the tricky | |
886 | * deadlock (see Documentation/filesystems/fuse.txt). | |
887 | * | |
888 | * During the aborting, progression of requests from the pending and | |
889 | * processing lists onto the io list, and progression of new requests | |
890 | * onto the pending list is prevented by req->connected being false. | |
891 | * | |
892 | * Progression of requests under I/O to the processing list is | |
893 | * prevented by the req->interrupted flag being true for these | |
894 | * requests. For this reason requests on the io list must be aborted | |
895 | * first. | |
896 | */ | |
897 | void fuse_abort_conn(struct fuse_conn *fc) | |
898 | { | |
899 | spin_lock(&fuse_lock); | |
900 | if (fc->connected) { | |
901 | fc->connected = 0; | |
902 | end_io_requests(fc); | |
903 | end_requests(fc, &fc->pending); | |
904 | end_requests(fc, &fc->processing); | |
905 | wake_up_all(&fc->waitq); | |
385a17bf | 906 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
69a53bf2 MS |
907 | } |
908 | spin_unlock(&fuse_lock); | |
909 | } | |
910 | ||
334f485d MS |
911 | static int fuse_dev_release(struct inode *inode, struct file *file) |
912 | { | |
0720b315 | 913 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 914 | if (fc) { |
0720b315 | 915 | spin_lock(&fuse_lock); |
1e9a4ed9 | 916 | fc->connected = 0; |
334f485d MS |
917 | end_requests(fc, &fc->pending); |
918 | end_requests(fc, &fc->processing); | |
0720b315 | 919 | spin_unlock(&fuse_lock); |
385a17bf | 920 | fasync_helper(-1, file, 0, &fc->fasync); |
f543f253 | 921 | kobject_put(&fc->kobj); |
385a17bf | 922 | } |
f543f253 | 923 | |
334f485d MS |
924 | return 0; |
925 | } | |
926 | ||
385a17bf JD |
927 | static int fuse_dev_fasync(int fd, struct file *file, int on) |
928 | { | |
929 | struct fuse_conn *fc = fuse_get_conn(file); | |
930 | if (!fc) | |
931 | return -ENODEV; | |
932 | ||
933 | /* No locking - fasync_helper does its own locking */ | |
934 | return fasync_helper(fd, file, on, &fc->fasync); | |
935 | } | |
936 | ||
4b6f5d20 | 937 | const struct file_operations fuse_dev_operations = { |
334f485d MS |
938 | .owner = THIS_MODULE, |
939 | .llseek = no_llseek, | |
940 | .read = fuse_dev_read, | |
941 | .readv = fuse_dev_readv, | |
942 | .write = fuse_dev_write, | |
943 | .writev = fuse_dev_writev, | |
944 | .poll = fuse_dev_poll, | |
945 | .release = fuse_dev_release, | |
385a17bf | 946 | .fasync = fuse_dev_fasync, |
334f485d MS |
947 | }; |
948 | ||
949 | static struct miscdevice fuse_miscdevice = { | |
950 | .minor = FUSE_MINOR, | |
951 | .name = "fuse", | |
952 | .fops = &fuse_dev_operations, | |
953 | }; | |
954 | ||
955 | int __init fuse_dev_init(void) | |
956 | { | |
957 | int err = -ENOMEM; | |
958 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
959 | sizeof(struct fuse_req), | |
960 | 0, 0, NULL, NULL); | |
961 | if (!fuse_req_cachep) | |
962 | goto out; | |
963 | ||
964 | err = misc_register(&fuse_miscdevice); | |
965 | if (err) | |
966 | goto out_cache_clean; | |
967 | ||
968 | return 0; | |
969 | ||
970 | out_cache_clean: | |
971 | kmem_cache_destroy(fuse_req_cachep); | |
972 | out: | |
973 | return err; | |
974 | } | |
975 | ||
976 | void fuse_dev_cleanup(void) | |
977 | { | |
978 | misc_deregister(&fuse_miscdevice); | |
979 | kmem_cache_destroy(fuse_req_cachep); | |
980 | } |