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 MS |
25 | { |
26 | struct fuse_conn *fc; | |
27 | spin_lock(&fuse_lock); | |
28 | fc = file->private_data; | |
9ba7cbba | 29 | if (fc && !fc->connected) |
334f485d MS |
30 | fc = NULL; |
31 | spin_unlock(&fuse_lock); | |
32 | return fc; | |
33 | } | |
34 | ||
8bfc016d | 35 | static void fuse_request_init(struct fuse_req *req) |
334f485d MS |
36 | { |
37 | memset(req, 0, sizeof(*req)); | |
38 | INIT_LIST_HEAD(&req->list); | |
39 | init_waitqueue_head(&req->waitq); | |
40 | atomic_set(&req->count, 1); | |
41 | } | |
42 | ||
43 | struct fuse_req *fuse_request_alloc(void) | |
44 | { | |
45 | struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, SLAB_KERNEL); | |
46 | if (req) | |
47 | fuse_request_init(req); | |
48 | return req; | |
49 | } | |
50 | ||
51 | void fuse_request_free(struct fuse_req *req) | |
52 | { | |
53 | kmem_cache_free(fuse_req_cachep, req); | |
54 | } | |
55 | ||
8bfc016d | 56 | static void block_sigs(sigset_t *oldset) |
334f485d MS |
57 | { |
58 | sigset_t mask; | |
59 | ||
60 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
61 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
62 | } | |
63 | ||
8bfc016d | 64 | static void restore_sigs(sigset_t *oldset) |
334f485d MS |
65 | { |
66 | sigprocmask(SIG_SETMASK, oldset, NULL); | |
67 | } | |
68 | ||
69 | void fuse_reset_request(struct fuse_req *req) | |
70 | { | |
71 | int preallocated = req->preallocated; | |
72 | BUG_ON(atomic_read(&req->count) != 1); | |
73 | fuse_request_init(req); | |
74 | req->preallocated = preallocated; | |
75 | } | |
76 | ||
77 | static void __fuse_get_request(struct fuse_req *req) | |
78 | { | |
79 | atomic_inc(&req->count); | |
80 | } | |
81 | ||
82 | /* Must be called with > 1 refcount */ | |
83 | static void __fuse_put_request(struct fuse_req *req) | |
84 | { | |
85 | BUG_ON(atomic_read(&req->count) < 2); | |
86 | atomic_dec(&req->count); | |
87 | } | |
88 | ||
89 | static struct fuse_req *do_get_request(struct fuse_conn *fc) | |
90 | { | |
91 | struct fuse_req *req; | |
92 | ||
93 | spin_lock(&fuse_lock); | |
94 | BUG_ON(list_empty(&fc->unused_list)); | |
95 | req = list_entry(fc->unused_list.next, struct fuse_req, list); | |
96 | list_del_init(&req->list); | |
97 | spin_unlock(&fuse_lock); | |
98 | fuse_request_init(req); | |
99 | req->preallocated = 1; | |
100 | req->in.h.uid = current->fsuid; | |
101 | req->in.h.gid = current->fsgid; | |
102 | req->in.h.pid = current->pid; | |
103 | return req; | |
104 | } | |
105 | ||
7c352bdf | 106 | /* This can return NULL, but only in case it's interrupted by a SIGKILL */ |
334f485d | 107 | struct fuse_req *fuse_get_request(struct fuse_conn *fc) |
334f485d MS |
108 | { |
109 | int intr; | |
110 | sigset_t oldset; | |
111 | ||
0cd5b885 | 112 | atomic_inc(&fc->num_waiting); |
334f485d MS |
113 | block_sigs(&oldset); |
114 | intr = down_interruptible(&fc->outstanding_sem); | |
115 | restore_sigs(&oldset); | |
0cd5b885 MS |
116 | if (intr) { |
117 | atomic_dec(&fc->num_waiting); | |
118 | return NULL; | |
119 | } | |
120 | return do_get_request(fc); | |
334f485d MS |
121 | } |
122 | ||
123 | static void fuse_putback_request(struct fuse_conn *fc, struct fuse_req *req) | |
124 | { | |
125 | spin_lock(&fuse_lock); | |
0cd5b885 MS |
126 | if (req->preallocated) { |
127 | atomic_dec(&fc->num_waiting); | |
334f485d | 128 | list_add(&req->list, &fc->unused_list); |
0cd5b885 | 129 | } else |
334f485d MS |
130 | fuse_request_free(req); |
131 | ||
132 | /* If we are in debt decrease that first */ | |
133 | if (fc->outstanding_debt) | |
134 | fc->outstanding_debt--; | |
135 | else | |
136 | up(&fc->outstanding_sem); | |
137 | spin_unlock(&fuse_lock); | |
138 | } | |
139 | ||
140 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) | |
141 | { | |
142 | if (atomic_dec_and_test(&req->count)) | |
143 | fuse_putback_request(fc, req); | |
144 | } | |
145 | ||
1e9a4ed9 MS |
146 | void fuse_release_background(struct fuse_req *req) |
147 | { | |
148 | iput(req->inode); | |
149 | iput(req->inode2); | |
150 | if (req->file) | |
151 | fput(req->file); | |
152 | spin_lock(&fuse_lock); | |
153 | list_del(&req->bg_entry); | |
154 | spin_unlock(&fuse_lock); | |
155 | } | |
156 | ||
334f485d MS |
157 | /* |
158 | * This function is called when a request is finished. Either a reply | |
159 | * has arrived or it was interrupted (and not yet sent) or some error | |
f43b155a MS |
160 | * occurred during communication with userspace, or the device file |
161 | * was closed. In case of a background request the reference to the | |
162 | * stored objects are released. The requester thread is woken up (if | |
64c6d8ed MS |
163 | * still waiting), the 'end' callback is called if given, else the |
164 | * reference to the request is released | |
334f485d MS |
165 | * |
166 | * Called with fuse_lock, unlocks it | |
167 | */ | |
168 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
169 | { | |
64c6d8ed MS |
170 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; |
171 | req->end = NULL; | |
d77a1d5b | 172 | list_del(&req->list); |
83cfd493 | 173 | req->state = FUSE_REQ_FINISHED; |
334f485d MS |
174 | spin_unlock(&fuse_lock); |
175 | if (req->background) { | |
1e9a4ed9 MS |
176 | down_read(&fc->sbput_sem); |
177 | if (fc->mounted) | |
178 | fuse_release_background(req); | |
179 | up_read(&fc->sbput_sem); | |
334f485d MS |
180 | } |
181 | wake_up(&req->waitq); | |
64c6d8ed MS |
182 | if (end) |
183 | end(fc, req); | |
184 | else | |
185 | fuse_put_request(fc, req); | |
334f485d MS |
186 | } |
187 | ||
1e9a4ed9 MS |
188 | /* |
189 | * Unfortunately request interruption not just solves the deadlock | |
190 | * problem, it causes problems too. These stem from the fact, that an | |
191 | * interrupted request is continued to be processed in userspace, | |
192 | * while all the locks and object references (inode and file) held | |
193 | * during the operation are released. | |
194 | * | |
195 | * To release the locks is exactly why there's a need to interrupt the | |
196 | * request, so there's not a lot that can be done about this, except | |
197 | * introduce additional locking in userspace. | |
198 | * | |
199 | * More important is to keep inode and file references until userspace | |
200 | * has replied, otherwise FORGET and RELEASE could be sent while the | |
201 | * inode/file is still used by the filesystem. | |
202 | * | |
203 | * For this reason the concept of "background" request is introduced. | |
204 | * An interrupted request is backgrounded if it has been already sent | |
205 | * to userspace. Backgrounding involves getting an extra reference to | |
206 | * inode(s) or file used in the request, and adding the request to | |
207 | * fc->background list. When a reply is received for a background | |
208 | * request, the object references are released, and the request is | |
209 | * removed from the list. If the filesystem is unmounted while there | |
210 | * are still background requests, the list is walked and references | |
211 | * are released as if a reply was received. | |
212 | * | |
213 | * There's one more use for a background request. The RELEASE message is | |
214 | * always sent as background, since it doesn't return an error or | |
215 | * data. | |
216 | */ | |
217 | static void background_request(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d | 218 | { |
334f485d | 219 | req->background = 1; |
1e9a4ed9 | 220 | list_add(&req->bg_entry, &fc->background); |
334f485d MS |
221 | if (req->inode) |
222 | req->inode = igrab(req->inode); | |
223 | if (req->inode2) | |
224 | req->inode2 = igrab(req->inode2); | |
225 | if (req->file) | |
226 | get_file(req->file); | |
227 | } | |
228 | ||
334f485d | 229 | /* Called with fuse_lock held. Releases, and then reacquires it. */ |
7c352bdf | 230 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) |
334f485d | 231 | { |
7c352bdf | 232 | sigset_t oldset; |
334f485d MS |
233 | |
234 | spin_unlock(&fuse_lock); | |
7c352bdf | 235 | block_sigs(&oldset); |
83cfd493 | 236 | wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED); |
7c352bdf | 237 | restore_sigs(&oldset); |
334f485d | 238 | spin_lock(&fuse_lock); |
69a53bf2 | 239 | if (req->state == FUSE_REQ_FINISHED && !req->interrupted) |
334f485d MS |
240 | return; |
241 | ||
69a53bf2 MS |
242 | if (!req->interrupted) { |
243 | req->out.h.error = -EINTR; | |
244 | req->interrupted = 1; | |
245 | } | |
334f485d MS |
246 | if (req->locked) { |
247 | /* This is uninterruptible sleep, because data is | |
248 | being copied to/from the buffers of req. During | |
249 | locked state, there mustn't be any filesystem | |
250 | operation (e.g. page fault), since that could lead | |
251 | to deadlock */ | |
252 | spin_unlock(&fuse_lock); | |
253 | wait_event(req->waitq, !req->locked); | |
254 | spin_lock(&fuse_lock); | |
255 | } | |
83cfd493 | 256 | if (req->state == FUSE_REQ_PENDING) { |
334f485d MS |
257 | list_del(&req->list); |
258 | __fuse_put_request(req); | |
83cfd493 | 259 | } else if (req->state == FUSE_REQ_SENT) |
1e9a4ed9 | 260 | background_request(fc, req); |
334f485d MS |
261 | } |
262 | ||
263 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) | |
264 | { | |
265 | unsigned nbytes = 0; | |
266 | unsigned i; | |
267 | ||
268 | for (i = 0; i < numargs; i++) | |
269 | nbytes += args[i].size; | |
270 | ||
271 | return nbytes; | |
272 | } | |
273 | ||
274 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) | |
275 | { | |
276 | fc->reqctr++; | |
277 | /* zero is special */ | |
278 | if (fc->reqctr == 0) | |
279 | fc->reqctr = 1; | |
280 | req->in.h.unique = fc->reqctr; | |
281 | req->in.h.len = sizeof(struct fuse_in_header) + | |
282 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
283 | if (!req->preallocated) { | |
284 | /* If request is not preallocated (either FORGET or | |
285 | RELEASE), then still decrease outstanding_sem, so | |
286 | user can't open infinite number of files while not | |
287 | processing the RELEASE requests. However for | |
288 | efficiency do it without blocking, so if down() | |
289 | would block, just increase the debt instead */ | |
290 | if (down_trylock(&fc->outstanding_sem)) | |
291 | fc->outstanding_debt++; | |
292 | } | |
293 | list_add_tail(&req->list, &fc->pending); | |
83cfd493 | 294 | req->state = FUSE_REQ_PENDING; |
334f485d MS |
295 | wake_up(&fc->waitq); |
296 | } | |
297 | ||
7c352bdf MS |
298 | /* |
299 | * This can only be interrupted by a SIGKILL | |
300 | */ | |
301 | void request_send(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d MS |
302 | { |
303 | req->isreply = 1; | |
304 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 305 | if (!fc->connected) |
334f485d MS |
306 | req->out.h.error = -ENOTCONN; |
307 | else if (fc->conn_error) | |
308 | req->out.h.error = -ECONNREFUSED; | |
309 | else { | |
310 | queue_request(fc, req); | |
311 | /* acquire extra reference, since request is still needed | |
312 | after request_end() */ | |
313 | __fuse_get_request(req); | |
314 | ||
7c352bdf | 315 | request_wait_answer(fc, req); |
334f485d MS |
316 | } |
317 | spin_unlock(&fuse_lock); | |
318 | } | |
319 | ||
334f485d MS |
320 | static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
321 | { | |
322 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 323 | if (fc->connected) { |
334f485d MS |
324 | queue_request(fc, req); |
325 | spin_unlock(&fuse_lock); | |
326 | } else { | |
327 | req->out.h.error = -ENOTCONN; | |
328 | request_end(fc, req); | |
329 | } | |
330 | } | |
331 | ||
332 | void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req) | |
333 | { | |
334 | req->isreply = 0; | |
335 | request_send_nowait(fc, req); | |
336 | } | |
337 | ||
338 | void request_send_background(struct fuse_conn *fc, struct fuse_req *req) | |
339 | { | |
340 | req->isreply = 1; | |
1e9a4ed9 MS |
341 | spin_lock(&fuse_lock); |
342 | background_request(fc, req); | |
343 | spin_unlock(&fuse_lock); | |
334f485d MS |
344 | request_send_nowait(fc, req); |
345 | } | |
346 | ||
334f485d MS |
347 | /* |
348 | * Lock the request. Up to the next unlock_request() there mustn't be | |
349 | * anything that could cause a page-fault. If the request was already | |
350 | * interrupted bail out. | |
351 | */ | |
8bfc016d | 352 | static int lock_request(struct fuse_req *req) |
334f485d MS |
353 | { |
354 | int err = 0; | |
355 | if (req) { | |
356 | spin_lock(&fuse_lock); | |
357 | if (req->interrupted) | |
358 | err = -ENOENT; | |
359 | else | |
360 | req->locked = 1; | |
361 | spin_unlock(&fuse_lock); | |
362 | } | |
363 | return err; | |
364 | } | |
365 | ||
366 | /* | |
367 | * Unlock request. If it was interrupted during being locked, the | |
368 | * requester thread is currently waiting for it to be unlocked, so | |
369 | * wake it up. | |
370 | */ | |
8bfc016d | 371 | static void unlock_request(struct fuse_req *req) |
334f485d MS |
372 | { |
373 | if (req) { | |
374 | spin_lock(&fuse_lock); | |
375 | req->locked = 0; | |
376 | if (req->interrupted) | |
377 | wake_up(&req->waitq); | |
378 | spin_unlock(&fuse_lock); | |
379 | } | |
380 | } | |
381 | ||
382 | struct fuse_copy_state { | |
383 | int write; | |
384 | struct fuse_req *req; | |
385 | const struct iovec *iov; | |
386 | unsigned long nr_segs; | |
387 | unsigned long seglen; | |
388 | unsigned long addr; | |
389 | struct page *pg; | |
390 | void *mapaddr; | |
391 | void *buf; | |
392 | unsigned len; | |
393 | }; | |
394 | ||
395 | static void fuse_copy_init(struct fuse_copy_state *cs, int write, | |
396 | struct fuse_req *req, const struct iovec *iov, | |
397 | unsigned long nr_segs) | |
398 | { | |
399 | memset(cs, 0, sizeof(*cs)); | |
400 | cs->write = write; | |
401 | cs->req = req; | |
402 | cs->iov = iov; | |
403 | cs->nr_segs = nr_segs; | |
404 | } | |
405 | ||
406 | /* Unmap and put previous page of userspace buffer */ | |
8bfc016d | 407 | static void fuse_copy_finish(struct fuse_copy_state *cs) |
334f485d MS |
408 | { |
409 | if (cs->mapaddr) { | |
410 | kunmap_atomic(cs->mapaddr, KM_USER0); | |
411 | if (cs->write) { | |
412 | flush_dcache_page(cs->pg); | |
413 | set_page_dirty_lock(cs->pg); | |
414 | } | |
415 | put_page(cs->pg); | |
416 | cs->mapaddr = NULL; | |
417 | } | |
418 | } | |
419 | ||
420 | /* | |
421 | * Get another pagefull of userspace buffer, and map it to kernel | |
422 | * address space, and lock request | |
423 | */ | |
424 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
425 | { | |
426 | unsigned long offset; | |
427 | int err; | |
428 | ||
429 | unlock_request(cs->req); | |
430 | fuse_copy_finish(cs); | |
431 | if (!cs->seglen) { | |
432 | BUG_ON(!cs->nr_segs); | |
433 | cs->seglen = cs->iov[0].iov_len; | |
434 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
435 | cs->iov ++; | |
436 | cs->nr_segs --; | |
437 | } | |
438 | down_read(¤t->mm->mmap_sem); | |
439 | err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0, | |
440 | &cs->pg, NULL); | |
441 | up_read(¤t->mm->mmap_sem); | |
442 | if (err < 0) | |
443 | return err; | |
444 | BUG_ON(err != 1); | |
445 | offset = cs->addr % PAGE_SIZE; | |
446 | cs->mapaddr = kmap_atomic(cs->pg, KM_USER0); | |
447 | cs->buf = cs->mapaddr + offset; | |
448 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
449 | cs->seglen -= cs->len; | |
450 | cs->addr += cs->len; | |
451 | ||
452 | return lock_request(cs->req); | |
453 | } | |
454 | ||
455 | /* Do as much copy to/from userspace buffer as we can */ | |
8bfc016d | 456 | static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) |
334f485d MS |
457 | { |
458 | unsigned ncpy = min(*size, cs->len); | |
459 | if (val) { | |
460 | if (cs->write) | |
461 | memcpy(cs->buf, *val, ncpy); | |
462 | else | |
463 | memcpy(*val, cs->buf, ncpy); | |
464 | *val += ncpy; | |
465 | } | |
466 | *size -= ncpy; | |
467 | cs->len -= ncpy; | |
468 | cs->buf += ncpy; | |
469 | return ncpy; | |
470 | } | |
471 | ||
472 | /* | |
473 | * Copy a page in the request to/from the userspace buffer. Must be | |
474 | * done atomically | |
475 | */ | |
8bfc016d MS |
476 | static int fuse_copy_page(struct fuse_copy_state *cs, struct page *page, |
477 | unsigned offset, unsigned count, int zeroing) | |
334f485d MS |
478 | { |
479 | if (page && zeroing && count < PAGE_SIZE) { | |
480 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
481 | memset(mapaddr, 0, PAGE_SIZE); | |
482 | kunmap_atomic(mapaddr, KM_USER1); | |
483 | } | |
484 | while (count) { | |
485 | int err; | |
486 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
487 | return err; | |
488 | if (page) { | |
489 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
490 | void *buf = mapaddr + offset; | |
491 | offset += fuse_copy_do(cs, &buf, &count); | |
492 | kunmap_atomic(mapaddr, KM_USER1); | |
493 | } else | |
494 | offset += fuse_copy_do(cs, NULL, &count); | |
495 | } | |
496 | if (page && !cs->write) | |
497 | flush_dcache_page(page); | |
498 | return 0; | |
499 | } | |
500 | ||
501 | /* Copy pages in the request to/from userspace buffer */ | |
502 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
503 | int zeroing) | |
504 | { | |
505 | unsigned i; | |
506 | struct fuse_req *req = cs->req; | |
507 | unsigned offset = req->page_offset; | |
508 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
509 | ||
510 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
511 | struct page *page = req->pages[i]; | |
512 | int err = fuse_copy_page(cs, page, offset, count, zeroing); | |
513 | if (err) | |
514 | return err; | |
515 | ||
516 | nbytes -= count; | |
517 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
518 | offset = 0; | |
519 | } | |
520 | return 0; | |
521 | } | |
522 | ||
523 | /* Copy a single argument in the request to/from userspace buffer */ | |
524 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
525 | { | |
526 | while (size) { | |
527 | int err; | |
528 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
529 | return err; | |
530 | fuse_copy_do(cs, &val, &size); | |
531 | } | |
532 | return 0; | |
533 | } | |
534 | ||
535 | /* Copy request arguments to/from userspace buffer */ | |
536 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
537 | unsigned argpages, struct fuse_arg *args, | |
538 | int zeroing) | |
539 | { | |
540 | int err = 0; | |
541 | unsigned i; | |
542 | ||
543 | for (i = 0; !err && i < numargs; i++) { | |
544 | struct fuse_arg *arg = &args[i]; | |
545 | if (i == numargs - 1 && argpages) | |
546 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
547 | else | |
548 | err = fuse_copy_one(cs, arg->value, arg->size); | |
549 | } | |
550 | return err; | |
551 | } | |
552 | ||
553 | /* Wait until a request is available on the pending list */ | |
554 | static void request_wait(struct fuse_conn *fc) | |
555 | { | |
556 | DECLARE_WAITQUEUE(wait, current); | |
557 | ||
558 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
9ba7cbba | 559 | while (fc->connected && list_empty(&fc->pending)) { |
334f485d MS |
560 | set_current_state(TASK_INTERRUPTIBLE); |
561 | if (signal_pending(current)) | |
562 | break; | |
563 | ||
564 | spin_unlock(&fuse_lock); | |
565 | schedule(); | |
566 | spin_lock(&fuse_lock); | |
567 | } | |
568 | set_current_state(TASK_RUNNING); | |
569 | remove_wait_queue(&fc->waitq, &wait); | |
570 | } | |
571 | ||
572 | /* | |
573 | * Read a single request into the userspace filesystem's buffer. This | |
574 | * function waits until a request is available, then removes it from | |
575 | * the pending list and copies request data to userspace buffer. If | |
576 | * no reply is needed (FORGET) or request has been interrupted or | |
577 | * there was an error during the copying then it's finished by calling | |
578 | * request_end(). Otherwise add it to the processing list, and set | |
579 | * the 'sent' flag. | |
580 | */ | |
581 | static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov, | |
582 | unsigned long nr_segs, loff_t *off) | |
583 | { | |
584 | int err; | |
585 | struct fuse_conn *fc; | |
586 | struct fuse_req *req; | |
587 | struct fuse_in *in; | |
588 | struct fuse_copy_state cs; | |
589 | unsigned reqsize; | |
590 | ||
1d3d752b | 591 | restart: |
334f485d MS |
592 | spin_lock(&fuse_lock); |
593 | fc = file->private_data; | |
594 | err = -EPERM; | |
595 | if (!fc) | |
596 | goto err_unlock; | |
597 | request_wait(fc); | |
598 | err = -ENODEV; | |
9ba7cbba | 599 | if (!fc->connected) |
334f485d MS |
600 | goto err_unlock; |
601 | err = -ERESTARTSYS; | |
602 | if (list_empty(&fc->pending)) | |
603 | goto err_unlock; | |
604 | ||
605 | req = list_entry(fc->pending.next, struct fuse_req, list); | |
83cfd493 | 606 | req->state = FUSE_REQ_READING; |
d77a1d5b | 607 | list_move(&req->list, &fc->io); |
334f485d MS |
608 | |
609 | in = &req->in; | |
1d3d752b MS |
610 | reqsize = in->h.len; |
611 | /* If request is too large, reply with an error and restart the read */ | |
612 | if (iov_length(iov, nr_segs) < reqsize) { | |
613 | req->out.h.error = -EIO; | |
614 | /* SETXATTR is special, since it may contain too large data */ | |
615 | if (in->h.opcode == FUSE_SETXATTR) | |
616 | req->out.h.error = -E2BIG; | |
617 | request_end(fc, req); | |
618 | goto restart; | |
334f485d | 619 | } |
1d3d752b MS |
620 | spin_unlock(&fuse_lock); |
621 | fuse_copy_init(&cs, 1, req, iov, nr_segs); | |
622 | err = fuse_copy_one(&cs, &in->h, sizeof(in->h)); | |
623 | if (!err) | |
624 | err = fuse_copy_args(&cs, in->numargs, in->argpages, | |
625 | (struct fuse_arg *) in->args, 0); | |
334f485d | 626 | fuse_copy_finish(&cs); |
334f485d MS |
627 | spin_lock(&fuse_lock); |
628 | req->locked = 0; | |
629 | if (!err && req->interrupted) | |
630 | err = -ENOENT; | |
631 | if (err) { | |
632 | if (!req->interrupted) | |
633 | req->out.h.error = -EIO; | |
634 | request_end(fc, req); | |
635 | return err; | |
636 | } | |
637 | if (!req->isreply) | |
638 | request_end(fc, req); | |
639 | else { | |
83cfd493 | 640 | req->state = FUSE_REQ_SENT; |
d77a1d5b | 641 | list_move_tail(&req->list, &fc->processing); |
334f485d MS |
642 | spin_unlock(&fuse_lock); |
643 | } | |
644 | return reqsize; | |
645 | ||
646 | err_unlock: | |
647 | spin_unlock(&fuse_lock); | |
648 | return err; | |
649 | } | |
650 | ||
651 | static ssize_t fuse_dev_read(struct file *file, char __user *buf, | |
652 | size_t nbytes, loff_t *off) | |
653 | { | |
654 | struct iovec iov; | |
655 | iov.iov_len = nbytes; | |
656 | iov.iov_base = buf; | |
657 | return fuse_dev_readv(file, &iov, 1, off); | |
658 | } | |
659 | ||
660 | /* Look up request on processing list by unique ID */ | |
661 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
662 | { | |
663 | struct list_head *entry; | |
664 | ||
665 | list_for_each(entry, &fc->processing) { | |
666 | struct fuse_req *req; | |
667 | req = list_entry(entry, struct fuse_req, list); | |
668 | if (req->in.h.unique == unique) | |
669 | return req; | |
670 | } | |
671 | return NULL; | |
672 | } | |
673 | ||
674 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
675 | unsigned nbytes) | |
676 | { | |
677 | unsigned reqsize = sizeof(struct fuse_out_header); | |
678 | ||
679 | if (out->h.error) | |
680 | return nbytes != reqsize ? -EINVAL : 0; | |
681 | ||
682 | reqsize += len_args(out->numargs, out->args); | |
683 | ||
684 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
685 | return -EINVAL; | |
686 | else if (reqsize > nbytes) { | |
687 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
688 | unsigned diffsize = reqsize - nbytes; | |
689 | if (diffsize > lastarg->size) | |
690 | return -EINVAL; | |
691 | lastarg->size -= diffsize; | |
692 | } | |
693 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
694 | out->page_zeroing); | |
695 | } | |
696 | ||
697 | /* | |
698 | * Write a single reply to a request. First the header is copied from | |
699 | * the write buffer. The request is then searched on the processing | |
700 | * list by the unique ID found in the header. If found, then remove | |
701 | * it from the list and copy the rest of the buffer to the request. | |
702 | * The request is finished by calling request_end() | |
703 | */ | |
704 | static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov, | |
705 | unsigned long nr_segs, loff_t *off) | |
706 | { | |
707 | int err; | |
708 | unsigned nbytes = iov_length(iov, nr_segs); | |
709 | struct fuse_req *req; | |
710 | struct fuse_out_header oh; | |
711 | struct fuse_copy_state cs; | |
712 | struct fuse_conn *fc = fuse_get_conn(file); | |
713 | if (!fc) | |
714 | return -ENODEV; | |
715 | ||
716 | fuse_copy_init(&cs, 0, NULL, iov, nr_segs); | |
717 | if (nbytes < sizeof(struct fuse_out_header)) | |
718 | return -EINVAL; | |
719 | ||
720 | err = fuse_copy_one(&cs, &oh, sizeof(oh)); | |
721 | if (err) | |
722 | goto err_finish; | |
723 | err = -EINVAL; | |
724 | if (!oh.unique || oh.error <= -1000 || oh.error > 0 || | |
725 | oh.len != nbytes) | |
726 | goto err_finish; | |
727 | ||
728 | spin_lock(&fuse_lock); | |
69a53bf2 MS |
729 | err = -ENOENT; |
730 | if (!fc->connected) | |
731 | goto err_unlock; | |
732 | ||
334f485d MS |
733 | req = request_find(fc, oh.unique); |
734 | err = -EINVAL; | |
735 | if (!req) | |
736 | goto err_unlock; | |
737 | ||
334f485d | 738 | if (req->interrupted) { |
222f1d69 | 739 | spin_unlock(&fuse_lock); |
334f485d | 740 | fuse_copy_finish(&cs); |
222f1d69 MS |
741 | spin_lock(&fuse_lock); |
742 | request_end(fc, req); | |
334f485d MS |
743 | return -ENOENT; |
744 | } | |
d77a1d5b | 745 | list_move(&req->list, &fc->io); |
334f485d MS |
746 | req->out.h = oh; |
747 | req->locked = 1; | |
748 | cs.req = req; | |
749 | spin_unlock(&fuse_lock); | |
750 | ||
751 | err = copy_out_args(&cs, &req->out, nbytes); | |
752 | fuse_copy_finish(&cs); | |
753 | ||
754 | spin_lock(&fuse_lock); | |
755 | req->locked = 0; | |
756 | if (!err) { | |
757 | if (req->interrupted) | |
758 | err = -ENOENT; | |
759 | } else if (!req->interrupted) | |
760 | req->out.h.error = -EIO; | |
761 | request_end(fc, req); | |
762 | ||
763 | return err ? err : nbytes; | |
764 | ||
765 | err_unlock: | |
766 | spin_unlock(&fuse_lock); | |
767 | err_finish: | |
768 | fuse_copy_finish(&cs); | |
769 | return err; | |
770 | } | |
771 | ||
772 | static ssize_t fuse_dev_write(struct file *file, const char __user *buf, | |
773 | size_t nbytes, loff_t *off) | |
774 | { | |
775 | struct iovec iov; | |
776 | iov.iov_len = nbytes; | |
777 | iov.iov_base = (char __user *) buf; | |
778 | return fuse_dev_writev(file, &iov, 1, off); | |
779 | } | |
780 | ||
781 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) | |
782 | { | |
783 | struct fuse_conn *fc = fuse_get_conn(file); | |
784 | unsigned mask = POLLOUT | POLLWRNORM; | |
785 | ||
786 | if (!fc) | |
787 | return -ENODEV; | |
788 | ||
789 | poll_wait(file, &fc->waitq, wait); | |
790 | ||
791 | spin_lock(&fuse_lock); | |
792 | if (!list_empty(&fc->pending)) | |
793 | mask |= POLLIN | POLLRDNORM; | |
794 | spin_unlock(&fuse_lock); | |
795 | ||
796 | return mask; | |
797 | } | |
798 | ||
69a53bf2 MS |
799 | /* |
800 | * Abort all requests on the given list (pending or processing) | |
801 | * | |
802 | * This function releases and reacquires fuse_lock | |
803 | */ | |
334f485d MS |
804 | static void end_requests(struct fuse_conn *fc, struct list_head *head) |
805 | { | |
806 | while (!list_empty(head)) { | |
807 | struct fuse_req *req; | |
808 | req = list_entry(head->next, struct fuse_req, list); | |
334f485d MS |
809 | req->out.h.error = -ECONNABORTED; |
810 | request_end(fc, req); | |
811 | spin_lock(&fuse_lock); | |
812 | } | |
813 | } | |
814 | ||
69a53bf2 MS |
815 | /* |
816 | * Abort requests under I/O | |
817 | * | |
818 | * The requests are set to interrupted and finished, and the request | |
819 | * waiter is woken up. This will make request_wait_answer() wait | |
820 | * until the request is unlocked and then return. | |
64c6d8ed MS |
821 | * |
822 | * If the request is asynchronous, then the end function needs to be | |
823 | * called after waiting for the request to be unlocked (if it was | |
824 | * locked). | |
69a53bf2 MS |
825 | */ |
826 | static void end_io_requests(struct fuse_conn *fc) | |
827 | { | |
828 | while (!list_empty(&fc->io)) { | |
64c6d8ed MS |
829 | struct fuse_req *req = |
830 | list_entry(fc->io.next, struct fuse_req, list); | |
831 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
832 | ||
69a53bf2 MS |
833 | req->interrupted = 1; |
834 | req->out.h.error = -ECONNABORTED; | |
835 | req->state = FUSE_REQ_FINISHED; | |
836 | list_del_init(&req->list); | |
837 | wake_up(&req->waitq); | |
64c6d8ed MS |
838 | if (end) { |
839 | req->end = NULL; | |
840 | /* The end function will consume this reference */ | |
841 | __fuse_get_request(req); | |
842 | spin_unlock(&fuse_lock); | |
843 | wait_event(req->waitq, !req->locked); | |
844 | end(fc, req); | |
845 | spin_lock(&fuse_lock); | |
846 | } | |
69a53bf2 MS |
847 | } |
848 | } | |
849 | ||
850 | /* | |
851 | * Abort all requests. | |
852 | * | |
853 | * Emergency exit in case of a malicious or accidental deadlock, or | |
854 | * just a hung filesystem. | |
855 | * | |
856 | * The same effect is usually achievable through killing the | |
857 | * filesystem daemon and all users of the filesystem. The exception | |
858 | * is the combination of an asynchronous request and the tricky | |
859 | * deadlock (see Documentation/filesystems/fuse.txt). | |
860 | * | |
861 | * During the aborting, progression of requests from the pending and | |
862 | * processing lists onto the io list, and progression of new requests | |
863 | * onto the pending list is prevented by req->connected being false. | |
864 | * | |
865 | * Progression of requests under I/O to the processing list is | |
866 | * prevented by the req->interrupted flag being true for these | |
867 | * requests. For this reason requests on the io list must be aborted | |
868 | * first. | |
869 | */ | |
870 | void fuse_abort_conn(struct fuse_conn *fc) | |
871 | { | |
872 | spin_lock(&fuse_lock); | |
873 | if (fc->connected) { | |
874 | fc->connected = 0; | |
875 | end_io_requests(fc); | |
876 | end_requests(fc, &fc->pending); | |
877 | end_requests(fc, &fc->processing); | |
878 | wake_up_all(&fc->waitq); | |
879 | } | |
880 | spin_unlock(&fuse_lock); | |
881 | } | |
882 | ||
334f485d MS |
883 | static int fuse_dev_release(struct inode *inode, struct file *file) |
884 | { | |
885 | struct fuse_conn *fc; | |
886 | ||
887 | spin_lock(&fuse_lock); | |
888 | fc = file->private_data; | |
889 | if (fc) { | |
1e9a4ed9 | 890 | fc->connected = 0; |
334f485d MS |
891 | end_requests(fc, &fc->pending); |
892 | end_requests(fc, &fc->processing); | |
334f485d MS |
893 | } |
894 | spin_unlock(&fuse_lock); | |
f543f253 MS |
895 | if (fc) |
896 | kobject_put(&fc->kobj); | |
897 | ||
334f485d MS |
898 | return 0; |
899 | } | |
900 | ||
901 | struct file_operations fuse_dev_operations = { | |
902 | .owner = THIS_MODULE, | |
903 | .llseek = no_llseek, | |
904 | .read = fuse_dev_read, | |
905 | .readv = fuse_dev_readv, | |
906 | .write = fuse_dev_write, | |
907 | .writev = fuse_dev_writev, | |
908 | .poll = fuse_dev_poll, | |
909 | .release = fuse_dev_release, | |
910 | }; | |
911 | ||
912 | static struct miscdevice fuse_miscdevice = { | |
913 | .minor = FUSE_MINOR, | |
914 | .name = "fuse", | |
915 | .fops = &fuse_dev_operations, | |
916 | }; | |
917 | ||
918 | int __init fuse_dev_init(void) | |
919 | { | |
920 | int err = -ENOMEM; | |
921 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
922 | sizeof(struct fuse_req), | |
923 | 0, 0, NULL, NULL); | |
924 | if (!fuse_req_cachep) | |
925 | goto out; | |
926 | ||
927 | err = misc_register(&fuse_miscdevice); | |
928 | if (err) | |
929 | goto out_cache_clean; | |
930 | ||
931 | return 0; | |
932 | ||
933 | out_cache_clean: | |
934 | kmem_cache_destroy(fuse_req_cachep); | |
935 | out: | |
936 | return err; | |
937 | } | |
938 | ||
939 | void fuse_dev_cleanup(void) | |
940 | { | |
941 | misc_deregister(&fuse_miscdevice); | |
942 | kmem_cache_destroy(fuse_req_cachep); | |
943 | } |