Commit | Line | Data |
---|---|---|
334f485d MS |
1 | /* |
2 | FUSE: Filesystem in Userspace | |
d7133114 | 3 | Copyright (C) 2001-2006 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> | |
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 | ||
d7133114 | 97 | spin_lock(&fc->lock); |
334f485d MS |
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); | |
d7133114 | 101 | spin_unlock(&fc->lock); |
334f485d MS |
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 | ||
d7133114 | 127 | /* Must be called with fc->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)) { | |
d7133114 | 146 | spin_lock(&fc->lock); |
7128ec2a | 147 | fuse_putback_request(fc, req); |
d7133114 | 148 | spin_unlock(&fc->lock); |
7128ec2a MS |
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 | ||
d7133114 | 158 | void fuse_release_background(struct fuse_conn *fc, struct fuse_req *req) |
1e9a4ed9 MS |
159 | { |
160 | iput(req->inode); | |
161 | iput(req->inode2); | |
162 | if (req->file) | |
163 | fput(req->file); | |
d7133114 | 164 | spin_lock(&fc->lock); |
1e9a4ed9 | 165 | list_del(&req->bg_entry); |
d7133114 | 166 | spin_unlock(&fc->lock); |
1e9a4ed9 MS |
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 | * | |
d7133114 | 185 | * Called with fc->lock, unlocks it |
334f485d MS |
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); | |
d7133114 | 194 | spin_unlock(&fc->lock); |
7128ec2a MS |
195 | } else { |
196 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
197 | req->end = NULL; | |
d7133114 | 198 | spin_unlock(&fc->lock); |
1e9a4ed9 MS |
199 | down_read(&fc->sbput_sem); |
200 | if (fc->mounted) | |
d7133114 | 201 | fuse_release_background(fc, req); |
1e9a4ed9 | 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 | ||
d7133114 | 251 | /* Called with fc->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 | 255 | |
d7133114 | 256 | spin_unlock(&fc->lock); |
7c352bdf | 257 | block_sigs(&oldset); |
83cfd493 | 258 | wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED); |
7c352bdf | 259 | restore_sigs(&oldset); |
d7133114 | 260 | spin_lock(&fc->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 */ | |
d7133114 | 274 | spin_unlock(&fc->lock); |
334f485d | 275 | wait_event(req->waitq, !req->locked); |
d7133114 | 276 | spin_lock(&fc->lock); |
334f485d | 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; | |
d7133114 | 327 | spin_lock(&fc->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 | 339 | } |
d7133114 | 340 | spin_unlock(&fc->lock); |
334f485d MS |
341 | } |
342 | ||
334f485d MS |
343 | static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
344 | { | |
d7133114 | 345 | spin_lock(&fc->lock); |
1e9a4ed9 | 346 | if (fc->connected) { |
334f485d | 347 | queue_request(fc, req); |
d7133114 | 348 | spin_unlock(&fc->lock); |
334f485d MS |
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; | |
d7133114 | 364 | spin_lock(&fc->lock); |
1e9a4ed9 | 365 | background_request(fc, req); |
d7133114 | 366 | spin_unlock(&fc->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 | */ | |
d7133114 | 375 | static int lock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
376 | { |
377 | int err = 0; | |
378 | if (req) { | |
d7133114 | 379 | spin_lock(&fc->lock); |
334f485d MS |
380 | if (req->interrupted) |
381 | err = -ENOENT; | |
382 | else | |
383 | req->locked = 1; | |
d7133114 | 384 | spin_unlock(&fc->lock); |
334f485d MS |
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 | */ | |
d7133114 | 394 | static void unlock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
395 | { |
396 | if (req) { | |
d7133114 | 397 | spin_lock(&fc->lock); |
334f485d MS |
398 | req->locked = 0; |
399 | if (req->interrupted) | |
400 | wake_up(&req->waitq); | |
d7133114 | 401 | spin_unlock(&fc->lock); |
334f485d MS |
402 | } |
403 | } | |
404 | ||
405 | struct fuse_copy_state { | |
d7133114 | 406 | struct fuse_conn *fc; |
334f485d MS |
407 | int write; |
408 | struct fuse_req *req; | |
409 | const struct iovec *iov; | |
410 | unsigned long nr_segs; | |
411 | unsigned long seglen; | |
412 | unsigned long addr; | |
413 | struct page *pg; | |
414 | void *mapaddr; | |
415 | void *buf; | |
416 | unsigned len; | |
417 | }; | |
418 | ||
d7133114 MS |
419 | static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc, |
420 | int write, struct fuse_req *req, | |
421 | const struct iovec *iov, unsigned long nr_segs) | |
334f485d MS |
422 | { |
423 | memset(cs, 0, sizeof(*cs)); | |
d7133114 | 424 | cs->fc = fc; |
334f485d MS |
425 | cs->write = write; |
426 | cs->req = req; | |
427 | cs->iov = iov; | |
428 | cs->nr_segs = nr_segs; | |
429 | } | |
430 | ||
431 | /* Unmap and put previous page of userspace buffer */ | |
8bfc016d | 432 | static void fuse_copy_finish(struct fuse_copy_state *cs) |
334f485d MS |
433 | { |
434 | if (cs->mapaddr) { | |
435 | kunmap_atomic(cs->mapaddr, KM_USER0); | |
436 | if (cs->write) { | |
437 | flush_dcache_page(cs->pg); | |
438 | set_page_dirty_lock(cs->pg); | |
439 | } | |
440 | put_page(cs->pg); | |
441 | cs->mapaddr = NULL; | |
442 | } | |
443 | } | |
444 | ||
445 | /* | |
446 | * Get another pagefull of userspace buffer, and map it to kernel | |
447 | * address space, and lock request | |
448 | */ | |
449 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
450 | { | |
451 | unsigned long offset; | |
452 | int err; | |
453 | ||
d7133114 | 454 | unlock_request(cs->fc, cs->req); |
334f485d MS |
455 | fuse_copy_finish(cs); |
456 | if (!cs->seglen) { | |
457 | BUG_ON(!cs->nr_segs); | |
458 | cs->seglen = cs->iov[0].iov_len; | |
459 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
460 | cs->iov ++; | |
461 | cs->nr_segs --; | |
462 | } | |
463 | down_read(¤t->mm->mmap_sem); | |
464 | err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0, | |
465 | &cs->pg, NULL); | |
466 | up_read(¤t->mm->mmap_sem); | |
467 | if (err < 0) | |
468 | return err; | |
469 | BUG_ON(err != 1); | |
470 | offset = cs->addr % PAGE_SIZE; | |
471 | cs->mapaddr = kmap_atomic(cs->pg, KM_USER0); | |
472 | cs->buf = cs->mapaddr + offset; | |
473 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
474 | cs->seglen -= cs->len; | |
475 | cs->addr += cs->len; | |
476 | ||
d7133114 | 477 | return lock_request(cs->fc, cs->req); |
334f485d MS |
478 | } |
479 | ||
480 | /* Do as much copy to/from userspace buffer as we can */ | |
8bfc016d | 481 | static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) |
334f485d MS |
482 | { |
483 | unsigned ncpy = min(*size, cs->len); | |
484 | if (val) { | |
485 | if (cs->write) | |
486 | memcpy(cs->buf, *val, ncpy); | |
487 | else | |
488 | memcpy(*val, cs->buf, ncpy); | |
489 | *val += ncpy; | |
490 | } | |
491 | *size -= ncpy; | |
492 | cs->len -= ncpy; | |
493 | cs->buf += ncpy; | |
494 | return ncpy; | |
495 | } | |
496 | ||
497 | /* | |
498 | * Copy a page in the request to/from the userspace buffer. Must be | |
499 | * done atomically | |
500 | */ | |
8bfc016d MS |
501 | static int fuse_copy_page(struct fuse_copy_state *cs, struct page *page, |
502 | unsigned offset, unsigned count, int zeroing) | |
334f485d MS |
503 | { |
504 | if (page && zeroing && count < PAGE_SIZE) { | |
505 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
506 | memset(mapaddr, 0, PAGE_SIZE); | |
507 | kunmap_atomic(mapaddr, KM_USER1); | |
508 | } | |
509 | while (count) { | |
510 | int err; | |
511 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
512 | return err; | |
513 | if (page) { | |
514 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
515 | void *buf = mapaddr + offset; | |
516 | offset += fuse_copy_do(cs, &buf, &count); | |
517 | kunmap_atomic(mapaddr, KM_USER1); | |
518 | } else | |
519 | offset += fuse_copy_do(cs, NULL, &count); | |
520 | } | |
521 | if (page && !cs->write) | |
522 | flush_dcache_page(page); | |
523 | return 0; | |
524 | } | |
525 | ||
526 | /* Copy pages in the request to/from userspace buffer */ | |
527 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
528 | int zeroing) | |
529 | { | |
530 | unsigned i; | |
531 | struct fuse_req *req = cs->req; | |
532 | unsigned offset = req->page_offset; | |
533 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
534 | ||
535 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
536 | struct page *page = req->pages[i]; | |
537 | int err = fuse_copy_page(cs, page, offset, count, zeroing); | |
538 | if (err) | |
539 | return err; | |
540 | ||
541 | nbytes -= count; | |
542 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
543 | offset = 0; | |
544 | } | |
545 | return 0; | |
546 | } | |
547 | ||
548 | /* Copy a single argument in the request to/from userspace buffer */ | |
549 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
550 | { | |
551 | while (size) { | |
552 | int err; | |
553 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
554 | return err; | |
555 | fuse_copy_do(cs, &val, &size); | |
556 | } | |
557 | return 0; | |
558 | } | |
559 | ||
560 | /* Copy request arguments to/from userspace buffer */ | |
561 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
562 | unsigned argpages, struct fuse_arg *args, | |
563 | int zeroing) | |
564 | { | |
565 | int err = 0; | |
566 | unsigned i; | |
567 | ||
568 | for (i = 0; !err && i < numargs; i++) { | |
569 | struct fuse_arg *arg = &args[i]; | |
570 | if (i == numargs - 1 && argpages) | |
571 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
572 | else | |
573 | err = fuse_copy_one(cs, arg->value, arg->size); | |
574 | } | |
575 | return err; | |
576 | } | |
577 | ||
578 | /* Wait until a request is available on the pending list */ | |
579 | static void request_wait(struct fuse_conn *fc) | |
580 | { | |
581 | DECLARE_WAITQUEUE(wait, current); | |
582 | ||
583 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
9ba7cbba | 584 | while (fc->connected && list_empty(&fc->pending)) { |
334f485d MS |
585 | set_current_state(TASK_INTERRUPTIBLE); |
586 | if (signal_pending(current)) | |
587 | break; | |
588 | ||
d7133114 | 589 | spin_unlock(&fc->lock); |
334f485d | 590 | schedule(); |
d7133114 | 591 | spin_lock(&fc->lock); |
334f485d MS |
592 | } |
593 | set_current_state(TASK_RUNNING); | |
594 | remove_wait_queue(&fc->waitq, &wait); | |
595 | } | |
596 | ||
597 | /* | |
598 | * Read a single request into the userspace filesystem's buffer. This | |
599 | * function waits until a request is available, then removes it from | |
600 | * the pending list and copies request data to userspace buffer. If | |
601 | * no reply is needed (FORGET) or request has been interrupted or | |
602 | * there was an error during the copying then it's finished by calling | |
603 | * request_end(). Otherwise add it to the processing list, and set | |
604 | * the 'sent' flag. | |
605 | */ | |
606 | static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov, | |
607 | unsigned long nr_segs, loff_t *off) | |
608 | { | |
609 | int err; | |
334f485d MS |
610 | struct fuse_req *req; |
611 | struct fuse_in *in; | |
612 | struct fuse_copy_state cs; | |
613 | unsigned reqsize; | |
0720b315 MS |
614 | struct fuse_conn *fc = fuse_get_conn(file); |
615 | if (!fc) | |
616 | return -EPERM; | |
334f485d | 617 | |
1d3d752b | 618 | restart: |
d7133114 | 619 | spin_lock(&fc->lock); |
e5ac1d1e JD |
620 | err = -EAGAIN; |
621 | if ((file->f_flags & O_NONBLOCK) && fc->connected && | |
622 | list_empty(&fc->pending)) | |
623 | goto err_unlock; | |
624 | ||
334f485d MS |
625 | request_wait(fc); |
626 | err = -ENODEV; | |
9ba7cbba | 627 | if (!fc->connected) |
334f485d MS |
628 | goto err_unlock; |
629 | err = -ERESTARTSYS; | |
630 | if (list_empty(&fc->pending)) | |
631 | goto err_unlock; | |
632 | ||
633 | req = list_entry(fc->pending.next, struct fuse_req, list); | |
83cfd493 | 634 | req->state = FUSE_REQ_READING; |
d77a1d5b | 635 | list_move(&req->list, &fc->io); |
334f485d MS |
636 | |
637 | in = &req->in; | |
1d3d752b MS |
638 | reqsize = in->h.len; |
639 | /* If request is too large, reply with an error and restart the read */ | |
640 | if (iov_length(iov, nr_segs) < reqsize) { | |
641 | req->out.h.error = -EIO; | |
642 | /* SETXATTR is special, since it may contain too large data */ | |
643 | if (in->h.opcode == FUSE_SETXATTR) | |
644 | req->out.h.error = -E2BIG; | |
645 | request_end(fc, req); | |
646 | goto restart; | |
334f485d | 647 | } |
d7133114 MS |
648 | spin_unlock(&fc->lock); |
649 | fuse_copy_init(&cs, fc, 1, req, iov, nr_segs); | |
1d3d752b MS |
650 | err = fuse_copy_one(&cs, &in->h, sizeof(in->h)); |
651 | if (!err) | |
652 | err = fuse_copy_args(&cs, in->numargs, in->argpages, | |
653 | (struct fuse_arg *) in->args, 0); | |
334f485d | 654 | fuse_copy_finish(&cs); |
d7133114 | 655 | spin_lock(&fc->lock); |
334f485d MS |
656 | req->locked = 0; |
657 | if (!err && req->interrupted) | |
658 | err = -ENOENT; | |
659 | if (err) { | |
660 | if (!req->interrupted) | |
661 | req->out.h.error = -EIO; | |
662 | request_end(fc, req); | |
663 | return err; | |
664 | } | |
665 | if (!req->isreply) | |
666 | request_end(fc, req); | |
667 | else { | |
83cfd493 | 668 | req->state = FUSE_REQ_SENT; |
d77a1d5b | 669 | list_move_tail(&req->list, &fc->processing); |
d7133114 | 670 | spin_unlock(&fc->lock); |
334f485d MS |
671 | } |
672 | return reqsize; | |
673 | ||
674 | err_unlock: | |
d7133114 | 675 | spin_unlock(&fc->lock); |
334f485d MS |
676 | return err; |
677 | } | |
678 | ||
679 | static ssize_t fuse_dev_read(struct file *file, char __user *buf, | |
680 | size_t nbytes, loff_t *off) | |
681 | { | |
682 | struct iovec iov; | |
683 | iov.iov_len = nbytes; | |
684 | iov.iov_base = buf; | |
685 | return fuse_dev_readv(file, &iov, 1, off); | |
686 | } | |
687 | ||
688 | /* Look up request on processing list by unique ID */ | |
689 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
690 | { | |
691 | struct list_head *entry; | |
692 | ||
693 | list_for_each(entry, &fc->processing) { | |
694 | struct fuse_req *req; | |
695 | req = list_entry(entry, struct fuse_req, list); | |
696 | if (req->in.h.unique == unique) | |
697 | return req; | |
698 | } | |
699 | return NULL; | |
700 | } | |
701 | ||
702 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
703 | unsigned nbytes) | |
704 | { | |
705 | unsigned reqsize = sizeof(struct fuse_out_header); | |
706 | ||
707 | if (out->h.error) | |
708 | return nbytes != reqsize ? -EINVAL : 0; | |
709 | ||
710 | reqsize += len_args(out->numargs, out->args); | |
711 | ||
712 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
713 | return -EINVAL; | |
714 | else if (reqsize > nbytes) { | |
715 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
716 | unsigned diffsize = reqsize - nbytes; | |
717 | if (diffsize > lastarg->size) | |
718 | return -EINVAL; | |
719 | lastarg->size -= diffsize; | |
720 | } | |
721 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
722 | out->page_zeroing); | |
723 | } | |
724 | ||
725 | /* | |
726 | * Write a single reply to a request. First the header is copied from | |
727 | * the write buffer. The request is then searched on the processing | |
728 | * list by the unique ID found in the header. If found, then remove | |
729 | * it from the list and copy the rest of the buffer to the request. | |
730 | * The request is finished by calling request_end() | |
731 | */ | |
732 | static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov, | |
733 | unsigned long nr_segs, loff_t *off) | |
734 | { | |
735 | int err; | |
736 | unsigned nbytes = iov_length(iov, nr_segs); | |
737 | struct fuse_req *req; | |
738 | struct fuse_out_header oh; | |
739 | struct fuse_copy_state cs; | |
740 | struct fuse_conn *fc = fuse_get_conn(file); | |
741 | if (!fc) | |
a87046d8 | 742 | return -EPERM; |
334f485d | 743 | |
d7133114 | 744 | fuse_copy_init(&cs, fc, 0, NULL, iov, nr_segs); |
334f485d MS |
745 | if (nbytes < sizeof(struct fuse_out_header)) |
746 | return -EINVAL; | |
747 | ||
748 | err = fuse_copy_one(&cs, &oh, sizeof(oh)); | |
749 | if (err) | |
750 | goto err_finish; | |
751 | err = -EINVAL; | |
752 | if (!oh.unique || oh.error <= -1000 || oh.error > 0 || | |
753 | oh.len != nbytes) | |
754 | goto err_finish; | |
755 | ||
d7133114 | 756 | spin_lock(&fc->lock); |
69a53bf2 MS |
757 | err = -ENOENT; |
758 | if (!fc->connected) | |
759 | goto err_unlock; | |
760 | ||
334f485d MS |
761 | req = request_find(fc, oh.unique); |
762 | err = -EINVAL; | |
763 | if (!req) | |
764 | goto err_unlock; | |
765 | ||
334f485d | 766 | if (req->interrupted) { |
d7133114 | 767 | spin_unlock(&fc->lock); |
334f485d | 768 | fuse_copy_finish(&cs); |
d7133114 | 769 | spin_lock(&fc->lock); |
222f1d69 | 770 | request_end(fc, req); |
334f485d MS |
771 | return -ENOENT; |
772 | } | |
d77a1d5b | 773 | list_move(&req->list, &fc->io); |
334f485d MS |
774 | req->out.h = oh; |
775 | req->locked = 1; | |
776 | cs.req = req; | |
d7133114 | 777 | spin_unlock(&fc->lock); |
334f485d MS |
778 | |
779 | err = copy_out_args(&cs, &req->out, nbytes); | |
780 | fuse_copy_finish(&cs); | |
781 | ||
d7133114 | 782 | spin_lock(&fc->lock); |
334f485d MS |
783 | req->locked = 0; |
784 | if (!err) { | |
785 | if (req->interrupted) | |
786 | err = -ENOENT; | |
787 | } else if (!req->interrupted) | |
788 | req->out.h.error = -EIO; | |
789 | request_end(fc, req); | |
790 | ||
791 | return err ? err : nbytes; | |
792 | ||
793 | err_unlock: | |
d7133114 | 794 | spin_unlock(&fc->lock); |
334f485d MS |
795 | err_finish: |
796 | fuse_copy_finish(&cs); | |
797 | return err; | |
798 | } | |
799 | ||
800 | static ssize_t fuse_dev_write(struct file *file, const char __user *buf, | |
801 | size_t nbytes, loff_t *off) | |
802 | { | |
803 | struct iovec iov; | |
804 | iov.iov_len = nbytes; | |
805 | iov.iov_base = (char __user *) buf; | |
806 | return fuse_dev_writev(file, &iov, 1, off); | |
807 | } | |
808 | ||
809 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) | |
810 | { | |
334f485d | 811 | unsigned mask = POLLOUT | POLLWRNORM; |
7025d9ad | 812 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 813 | if (!fc) |
7025d9ad | 814 | return POLLERR; |
334f485d MS |
815 | |
816 | poll_wait(file, &fc->waitq, wait); | |
817 | ||
d7133114 | 818 | spin_lock(&fc->lock); |
7025d9ad MS |
819 | if (!fc->connected) |
820 | mask = POLLERR; | |
821 | else if (!list_empty(&fc->pending)) | |
822 | mask |= POLLIN | POLLRDNORM; | |
d7133114 | 823 | spin_unlock(&fc->lock); |
334f485d MS |
824 | |
825 | return mask; | |
826 | } | |
827 | ||
69a53bf2 MS |
828 | /* |
829 | * Abort all requests on the given list (pending or processing) | |
830 | * | |
d7133114 | 831 | * This function releases and reacquires fc->lock |
69a53bf2 | 832 | */ |
334f485d MS |
833 | static void end_requests(struct fuse_conn *fc, struct list_head *head) |
834 | { | |
835 | while (!list_empty(head)) { | |
836 | struct fuse_req *req; | |
837 | req = list_entry(head->next, struct fuse_req, list); | |
334f485d MS |
838 | req->out.h.error = -ECONNABORTED; |
839 | request_end(fc, req); | |
d7133114 | 840 | spin_lock(&fc->lock); |
334f485d MS |
841 | } |
842 | } | |
843 | ||
69a53bf2 MS |
844 | /* |
845 | * Abort requests under I/O | |
846 | * | |
847 | * The requests are set to interrupted and finished, and the request | |
848 | * waiter is woken up. This will make request_wait_answer() wait | |
849 | * until the request is unlocked and then return. | |
64c6d8ed MS |
850 | * |
851 | * If the request is asynchronous, then the end function needs to be | |
852 | * called after waiting for the request to be unlocked (if it was | |
853 | * locked). | |
69a53bf2 MS |
854 | */ |
855 | static void end_io_requests(struct fuse_conn *fc) | |
856 | { | |
857 | while (!list_empty(&fc->io)) { | |
64c6d8ed MS |
858 | struct fuse_req *req = |
859 | list_entry(fc->io.next, struct fuse_req, list); | |
860 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
861 | ||
69a53bf2 MS |
862 | req->interrupted = 1; |
863 | req->out.h.error = -ECONNABORTED; | |
864 | req->state = FUSE_REQ_FINISHED; | |
865 | list_del_init(&req->list); | |
866 | wake_up(&req->waitq); | |
64c6d8ed MS |
867 | if (end) { |
868 | req->end = NULL; | |
869 | /* The end function will consume this reference */ | |
870 | __fuse_get_request(req); | |
d7133114 | 871 | spin_unlock(&fc->lock); |
64c6d8ed MS |
872 | wait_event(req->waitq, !req->locked); |
873 | end(fc, req); | |
d7133114 | 874 | spin_lock(&fc->lock); |
64c6d8ed | 875 | } |
69a53bf2 MS |
876 | } |
877 | } | |
878 | ||
879 | /* | |
880 | * Abort all requests. | |
881 | * | |
882 | * Emergency exit in case of a malicious or accidental deadlock, or | |
883 | * just a hung filesystem. | |
884 | * | |
885 | * The same effect is usually achievable through killing the | |
886 | * filesystem daemon and all users of the filesystem. The exception | |
887 | * is the combination of an asynchronous request and the tricky | |
888 | * deadlock (see Documentation/filesystems/fuse.txt). | |
889 | * | |
890 | * During the aborting, progression of requests from the pending and | |
891 | * processing lists onto the io list, and progression of new requests | |
892 | * onto the pending list is prevented by req->connected being false. | |
893 | * | |
894 | * Progression of requests under I/O to the processing list is | |
895 | * prevented by the req->interrupted flag being true for these | |
896 | * requests. For this reason requests on the io list must be aborted | |
897 | * first. | |
898 | */ | |
899 | void fuse_abort_conn(struct fuse_conn *fc) | |
900 | { | |
d7133114 | 901 | spin_lock(&fc->lock); |
69a53bf2 MS |
902 | if (fc->connected) { |
903 | fc->connected = 0; | |
904 | end_io_requests(fc); | |
905 | end_requests(fc, &fc->pending); | |
906 | end_requests(fc, &fc->processing); | |
907 | wake_up_all(&fc->waitq); | |
385a17bf | 908 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
69a53bf2 | 909 | } |
d7133114 | 910 | spin_unlock(&fc->lock); |
69a53bf2 MS |
911 | } |
912 | ||
334f485d MS |
913 | static int fuse_dev_release(struct inode *inode, struct file *file) |
914 | { | |
0720b315 | 915 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 916 | if (fc) { |
d7133114 | 917 | spin_lock(&fc->lock); |
1e9a4ed9 | 918 | fc->connected = 0; |
334f485d MS |
919 | end_requests(fc, &fc->pending); |
920 | end_requests(fc, &fc->processing); | |
d7133114 | 921 | spin_unlock(&fc->lock); |
385a17bf | 922 | fasync_helper(-1, file, 0, &fc->fasync); |
f543f253 | 923 | kobject_put(&fc->kobj); |
385a17bf | 924 | } |
f543f253 | 925 | |
334f485d MS |
926 | return 0; |
927 | } | |
928 | ||
385a17bf JD |
929 | static int fuse_dev_fasync(int fd, struct file *file, int on) |
930 | { | |
931 | struct fuse_conn *fc = fuse_get_conn(file); | |
932 | if (!fc) | |
a87046d8 | 933 | return -EPERM; |
385a17bf JD |
934 | |
935 | /* No locking - fasync_helper does its own locking */ | |
936 | return fasync_helper(fd, file, on, &fc->fasync); | |
937 | } | |
938 | ||
4b6f5d20 | 939 | const struct file_operations fuse_dev_operations = { |
334f485d MS |
940 | .owner = THIS_MODULE, |
941 | .llseek = no_llseek, | |
942 | .read = fuse_dev_read, | |
943 | .readv = fuse_dev_readv, | |
944 | .write = fuse_dev_write, | |
945 | .writev = fuse_dev_writev, | |
946 | .poll = fuse_dev_poll, | |
947 | .release = fuse_dev_release, | |
385a17bf | 948 | .fasync = fuse_dev_fasync, |
334f485d MS |
949 | }; |
950 | ||
951 | static struct miscdevice fuse_miscdevice = { | |
952 | .minor = FUSE_MINOR, | |
953 | .name = "fuse", | |
954 | .fops = &fuse_dev_operations, | |
955 | }; | |
956 | ||
957 | int __init fuse_dev_init(void) | |
958 | { | |
959 | int err = -ENOMEM; | |
960 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
961 | sizeof(struct fuse_req), | |
962 | 0, 0, NULL, NULL); | |
963 | if (!fuse_req_cachep) | |
964 | goto out; | |
965 | ||
966 | err = misc_register(&fuse_miscdevice); | |
967 | if (err) | |
968 | goto out_cache_clean; | |
969 | ||
970 | return 0; | |
971 | ||
972 | out_cache_clean: | |
973 | kmem_cache_destroy(fuse_req_cachep); | |
974 | out: | |
975 | return err; | |
976 | } | |
977 | ||
978 | void fuse_dev_cleanup(void) | |
979 | { | |
980 | misc_deregister(&fuse_miscdevice); | |
981 | kmem_cache_destroy(fuse_req_cachep); | |
982 | } |