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 | ||
334f485d MS |
67 | static void __fuse_get_request(struct fuse_req *req) |
68 | { | |
69 | atomic_inc(&req->count); | |
70 | } | |
71 | ||
72 | /* Must be called with > 1 refcount */ | |
73 | static void __fuse_put_request(struct fuse_req *req) | |
74 | { | |
75 | BUG_ON(atomic_read(&req->count) < 2); | |
76 | atomic_dec(&req->count); | |
77 | } | |
78 | ||
33649c91 MS |
79 | static void fuse_req_init_context(struct fuse_req *req) |
80 | { | |
81 | req->in.h.uid = current->fsuid; | |
82 | req->in.h.gid = current->fsgid; | |
83 | req->in.h.pid = current->pid; | |
84 | } | |
85 | ||
ce1d5a49 | 86 | struct fuse_req *fuse_get_req(struct fuse_conn *fc) |
334f485d | 87 | { |
08a53cdc MS |
88 | struct fuse_req *req; |
89 | sigset_t oldset; | |
9bc5ddda | 90 | int intr; |
08a53cdc MS |
91 | int err; |
92 | ||
9bc5ddda | 93 | atomic_inc(&fc->num_waiting); |
08a53cdc | 94 | block_sigs(&oldset); |
9bc5ddda | 95 | intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked); |
08a53cdc | 96 | restore_sigs(&oldset); |
9bc5ddda MS |
97 | err = -EINTR; |
98 | if (intr) | |
99 | goto out; | |
08a53cdc | 100 | |
51eb01e7 MS |
101 | err = -ENOTCONN; |
102 | if (!fc->connected) | |
103 | goto out; | |
104 | ||
08a53cdc | 105 | req = fuse_request_alloc(); |
9bc5ddda | 106 | err = -ENOMEM; |
ce1d5a49 | 107 | if (!req) |
9bc5ddda | 108 | goto out; |
334f485d | 109 | |
33649c91 | 110 | fuse_req_init_context(req); |
9bc5ddda | 111 | req->waiting = 1; |
334f485d | 112 | return req; |
9bc5ddda MS |
113 | |
114 | out: | |
115 | atomic_dec(&fc->num_waiting); | |
116 | return ERR_PTR(err); | |
334f485d MS |
117 | } |
118 | ||
33649c91 MS |
119 | /* |
120 | * Return request in fuse_file->reserved_req. However that may | |
121 | * currently be in use. If that is the case, wait for it to become | |
122 | * available. | |
123 | */ | |
124 | static struct fuse_req *get_reserved_req(struct fuse_conn *fc, | |
125 | struct file *file) | |
126 | { | |
127 | struct fuse_req *req = NULL; | |
128 | struct fuse_file *ff = file->private_data; | |
129 | ||
130 | do { | |
131 | wait_event(fc->blocked_waitq, ff->reserved_req); | |
132 | spin_lock(&fc->lock); | |
133 | if (ff->reserved_req) { | |
134 | req = ff->reserved_req; | |
135 | ff->reserved_req = NULL; | |
136 | get_file(file); | |
137 | req->stolen_file = file; | |
138 | } | |
139 | spin_unlock(&fc->lock); | |
140 | } while (!req); | |
141 | ||
142 | return req; | |
143 | } | |
144 | ||
145 | /* | |
146 | * Put stolen request back into fuse_file->reserved_req | |
147 | */ | |
148 | static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req) | |
149 | { | |
150 | struct file *file = req->stolen_file; | |
151 | struct fuse_file *ff = file->private_data; | |
152 | ||
153 | spin_lock(&fc->lock); | |
154 | fuse_request_init(req); | |
155 | BUG_ON(ff->reserved_req); | |
156 | ff->reserved_req = req; | |
157 | wake_up(&fc->blocked_waitq); | |
158 | spin_unlock(&fc->lock); | |
159 | fput(file); | |
160 | } | |
161 | ||
162 | /* | |
163 | * Gets a requests for a file operation, always succeeds | |
164 | * | |
165 | * This is used for sending the FLUSH request, which must get to | |
166 | * userspace, due to POSIX locks which may need to be unlocked. | |
167 | * | |
168 | * If allocation fails due to OOM, use the reserved request in | |
169 | * fuse_file. | |
170 | * | |
171 | * This is very unlikely to deadlock accidentally, since the | |
172 | * filesystem should not have it's own file open. If deadlock is | |
173 | * intentional, it can still be broken by "aborting" the filesystem. | |
174 | */ | |
175 | struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file) | |
176 | { | |
177 | struct fuse_req *req; | |
178 | ||
179 | atomic_inc(&fc->num_waiting); | |
180 | wait_event(fc->blocked_waitq, !fc->blocked); | |
181 | req = fuse_request_alloc(); | |
182 | if (!req) | |
183 | req = get_reserved_req(fc, file); | |
184 | ||
185 | fuse_req_init_context(req); | |
186 | req->waiting = 1; | |
187 | return req; | |
188 | } | |
189 | ||
334f485d | 190 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) |
7128ec2a MS |
191 | { |
192 | if (atomic_dec_and_test(&req->count)) { | |
9bc5ddda MS |
193 | if (req->waiting) |
194 | atomic_dec(&fc->num_waiting); | |
33649c91 MS |
195 | |
196 | if (req->stolen_file) | |
197 | put_reserved_req(fc, req); | |
198 | else | |
199 | fuse_request_free(req); | |
7128ec2a MS |
200 | } |
201 | } | |
202 | ||
334f485d MS |
203 | /* |
204 | * This function is called when a request is finished. Either a reply | |
f9a2842e | 205 | * has arrived or it was aborted (and not yet sent) or some error |
f43b155a | 206 | * occurred during communication with userspace, or the device file |
51eb01e7 MS |
207 | * was closed. The requester thread is woken up (if still waiting), |
208 | * the 'end' callback is called if given, else the reference to the | |
209 | * request is released | |
7128ec2a | 210 | * |
d7133114 | 211 | * Called with fc->lock, unlocks it |
334f485d MS |
212 | */ |
213 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
214 | { | |
51eb01e7 MS |
215 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; |
216 | req->end = NULL; | |
d77a1d5b | 217 | list_del(&req->list); |
83cfd493 | 218 | req->state = FUSE_REQ_FINISHED; |
51eb01e7 MS |
219 | if (req->background) { |
220 | if (fc->num_background == FUSE_MAX_BACKGROUND) { | |
221 | fc->blocked = 0; | |
222 | wake_up_all(&fc->blocked_waitq); | |
223 | } | |
224 | fc->num_background--; | |
334f485d | 225 | } |
51eb01e7 MS |
226 | spin_unlock(&fc->lock); |
227 | dput(req->dentry); | |
228 | mntput(req->vfsmount); | |
334f485d | 229 | if (req->file) |
51eb01e7 MS |
230 | fput(req->file); |
231 | wake_up(&req->waitq); | |
232 | if (end) | |
233 | end(fc, req); | |
234 | else | |
235 | fuse_put_request(fc, req); | |
334f485d MS |
236 | } |
237 | ||
d7133114 | 238 | /* Called with fc->lock held. Releases, and then reacquires it. */ |
7c352bdf | 239 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) |
334f485d | 240 | { |
7c352bdf | 241 | sigset_t oldset; |
334f485d | 242 | |
d7133114 | 243 | spin_unlock(&fc->lock); |
51eb01e7 MS |
244 | if (req->force) |
245 | wait_event(req->waitq, req->state == FUSE_REQ_FINISHED); | |
246 | else { | |
247 | block_sigs(&oldset); | |
248 | wait_event_interruptible(req->waitq, | |
249 | req->state == FUSE_REQ_FINISHED); | |
250 | restore_sigs(&oldset); | |
251 | } | |
d7133114 | 252 | spin_lock(&fc->lock); |
f9a2842e | 253 | if (req->state == FUSE_REQ_FINISHED && !req->aborted) |
334f485d MS |
254 | return; |
255 | ||
f9a2842e | 256 | if (!req->aborted) { |
69a53bf2 | 257 | req->out.h.error = -EINTR; |
f9a2842e | 258 | req->aborted = 1; |
69a53bf2 | 259 | } |
334f485d MS |
260 | if (req->locked) { |
261 | /* This is uninterruptible sleep, because data is | |
262 | being copied to/from the buffers of req. During | |
263 | locked state, there mustn't be any filesystem | |
264 | operation (e.g. page fault), since that could lead | |
265 | to deadlock */ | |
d7133114 | 266 | spin_unlock(&fc->lock); |
334f485d | 267 | wait_event(req->waitq, !req->locked); |
d7133114 | 268 | spin_lock(&fc->lock); |
334f485d | 269 | } |
83cfd493 | 270 | if (req->state == FUSE_REQ_PENDING) { |
334f485d MS |
271 | list_del(&req->list); |
272 | __fuse_put_request(req); | |
51eb01e7 MS |
273 | } else if (req->state == FUSE_REQ_SENT) { |
274 | spin_unlock(&fc->lock); | |
275 | wait_event(req->waitq, req->state == FUSE_REQ_FINISHED); | |
276 | spin_lock(&fc->lock); | |
277 | } | |
334f485d MS |
278 | } |
279 | ||
280 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) | |
281 | { | |
282 | unsigned nbytes = 0; | |
283 | unsigned i; | |
284 | ||
285 | for (i = 0; i < numargs; i++) | |
286 | nbytes += args[i].size; | |
287 | ||
288 | return nbytes; | |
289 | } | |
290 | ||
291 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) | |
292 | { | |
293 | fc->reqctr++; | |
294 | /* zero is special */ | |
295 | if (fc->reqctr == 0) | |
296 | fc->reqctr = 1; | |
297 | req->in.h.unique = fc->reqctr; | |
298 | req->in.h.len = sizeof(struct fuse_in_header) + | |
299 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
334f485d | 300 | list_add_tail(&req->list, &fc->pending); |
83cfd493 | 301 | req->state = FUSE_REQ_PENDING; |
9bc5ddda MS |
302 | if (!req->waiting) { |
303 | req->waiting = 1; | |
304 | atomic_inc(&fc->num_waiting); | |
305 | } | |
334f485d | 306 | wake_up(&fc->waitq); |
385a17bf | 307 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
334f485d MS |
308 | } |
309 | ||
7c352bdf MS |
310 | /* |
311 | * This can only be interrupted by a SIGKILL | |
312 | */ | |
313 | void request_send(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d MS |
314 | { |
315 | req->isreply = 1; | |
d7133114 | 316 | spin_lock(&fc->lock); |
1e9a4ed9 | 317 | if (!fc->connected) |
334f485d MS |
318 | req->out.h.error = -ENOTCONN; |
319 | else if (fc->conn_error) | |
320 | req->out.h.error = -ECONNREFUSED; | |
321 | else { | |
322 | queue_request(fc, req); | |
323 | /* acquire extra reference, since request is still needed | |
324 | after request_end() */ | |
325 | __fuse_get_request(req); | |
326 | ||
7c352bdf | 327 | request_wait_answer(fc, req); |
334f485d | 328 | } |
d7133114 | 329 | spin_unlock(&fc->lock); |
334f485d MS |
330 | } |
331 | ||
334f485d MS |
332 | static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
333 | { | |
d7133114 | 334 | spin_lock(&fc->lock); |
1e9a4ed9 | 335 | if (fc->connected) { |
51eb01e7 MS |
336 | req->background = 1; |
337 | fc->num_background++; | |
338 | if (fc->num_background == FUSE_MAX_BACKGROUND) | |
339 | fc->blocked = 1; | |
340 | ||
334f485d | 341 | queue_request(fc, req); |
d7133114 | 342 | spin_unlock(&fc->lock); |
334f485d MS |
343 | } else { |
344 | req->out.h.error = -ENOTCONN; | |
345 | request_end(fc, req); | |
346 | } | |
347 | } | |
348 | ||
349 | void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req) | |
350 | { | |
351 | req->isreply = 0; | |
352 | request_send_nowait(fc, req); | |
353 | } | |
354 | ||
355 | void request_send_background(struct fuse_conn *fc, struct fuse_req *req) | |
356 | { | |
357 | req->isreply = 1; | |
334f485d MS |
358 | request_send_nowait(fc, req); |
359 | } | |
360 | ||
334f485d MS |
361 | /* |
362 | * Lock the request. Up to the next unlock_request() there mustn't be | |
363 | * anything that could cause a page-fault. If the request was already | |
f9a2842e | 364 | * aborted bail out. |
334f485d | 365 | */ |
d7133114 | 366 | static int lock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
367 | { |
368 | int err = 0; | |
369 | if (req) { | |
d7133114 | 370 | spin_lock(&fc->lock); |
f9a2842e | 371 | if (req->aborted) |
334f485d MS |
372 | err = -ENOENT; |
373 | else | |
374 | req->locked = 1; | |
d7133114 | 375 | spin_unlock(&fc->lock); |
334f485d MS |
376 | } |
377 | return err; | |
378 | } | |
379 | ||
380 | /* | |
f9a2842e | 381 | * Unlock request. If it was aborted during being locked, the |
334f485d MS |
382 | * requester thread is currently waiting for it to be unlocked, so |
383 | * wake it up. | |
384 | */ | |
d7133114 | 385 | static void unlock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
386 | { |
387 | if (req) { | |
d7133114 | 388 | spin_lock(&fc->lock); |
334f485d | 389 | req->locked = 0; |
f9a2842e | 390 | if (req->aborted) |
334f485d | 391 | wake_up(&req->waitq); |
d7133114 | 392 | spin_unlock(&fc->lock); |
334f485d MS |
393 | } |
394 | } | |
395 | ||
396 | struct fuse_copy_state { | |
d7133114 | 397 | struct fuse_conn *fc; |
334f485d MS |
398 | int write; |
399 | struct fuse_req *req; | |
400 | const struct iovec *iov; | |
401 | unsigned long nr_segs; | |
402 | unsigned long seglen; | |
403 | unsigned long addr; | |
404 | struct page *pg; | |
405 | void *mapaddr; | |
406 | void *buf; | |
407 | unsigned len; | |
408 | }; | |
409 | ||
d7133114 MS |
410 | static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc, |
411 | int write, struct fuse_req *req, | |
412 | const struct iovec *iov, unsigned long nr_segs) | |
334f485d MS |
413 | { |
414 | memset(cs, 0, sizeof(*cs)); | |
d7133114 | 415 | cs->fc = fc; |
334f485d MS |
416 | cs->write = write; |
417 | cs->req = req; | |
418 | cs->iov = iov; | |
419 | cs->nr_segs = nr_segs; | |
420 | } | |
421 | ||
422 | /* Unmap and put previous page of userspace buffer */ | |
8bfc016d | 423 | static void fuse_copy_finish(struct fuse_copy_state *cs) |
334f485d MS |
424 | { |
425 | if (cs->mapaddr) { | |
426 | kunmap_atomic(cs->mapaddr, KM_USER0); | |
427 | if (cs->write) { | |
428 | flush_dcache_page(cs->pg); | |
429 | set_page_dirty_lock(cs->pg); | |
430 | } | |
431 | put_page(cs->pg); | |
432 | cs->mapaddr = NULL; | |
433 | } | |
434 | } | |
435 | ||
436 | /* | |
437 | * Get another pagefull of userspace buffer, and map it to kernel | |
438 | * address space, and lock request | |
439 | */ | |
440 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
441 | { | |
442 | unsigned long offset; | |
443 | int err; | |
444 | ||
d7133114 | 445 | unlock_request(cs->fc, cs->req); |
334f485d MS |
446 | fuse_copy_finish(cs); |
447 | if (!cs->seglen) { | |
448 | BUG_ON(!cs->nr_segs); | |
449 | cs->seglen = cs->iov[0].iov_len; | |
450 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
451 | cs->iov ++; | |
452 | cs->nr_segs --; | |
453 | } | |
454 | down_read(¤t->mm->mmap_sem); | |
455 | err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0, | |
456 | &cs->pg, NULL); | |
457 | up_read(¤t->mm->mmap_sem); | |
458 | if (err < 0) | |
459 | return err; | |
460 | BUG_ON(err != 1); | |
461 | offset = cs->addr % PAGE_SIZE; | |
462 | cs->mapaddr = kmap_atomic(cs->pg, KM_USER0); | |
463 | cs->buf = cs->mapaddr + offset; | |
464 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
465 | cs->seglen -= cs->len; | |
466 | cs->addr += cs->len; | |
467 | ||
d7133114 | 468 | return lock_request(cs->fc, cs->req); |
334f485d MS |
469 | } |
470 | ||
471 | /* Do as much copy to/from userspace buffer as we can */ | |
8bfc016d | 472 | static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) |
334f485d MS |
473 | { |
474 | unsigned ncpy = min(*size, cs->len); | |
475 | if (val) { | |
476 | if (cs->write) | |
477 | memcpy(cs->buf, *val, ncpy); | |
478 | else | |
479 | memcpy(*val, cs->buf, ncpy); | |
480 | *val += ncpy; | |
481 | } | |
482 | *size -= ncpy; | |
483 | cs->len -= ncpy; | |
484 | cs->buf += ncpy; | |
485 | return ncpy; | |
486 | } | |
487 | ||
488 | /* | |
489 | * Copy a page in the request to/from the userspace buffer. Must be | |
490 | * done atomically | |
491 | */ | |
8bfc016d MS |
492 | static int fuse_copy_page(struct fuse_copy_state *cs, struct page *page, |
493 | unsigned offset, unsigned count, int zeroing) | |
334f485d MS |
494 | { |
495 | if (page && zeroing && count < PAGE_SIZE) { | |
496 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
497 | memset(mapaddr, 0, PAGE_SIZE); | |
498 | kunmap_atomic(mapaddr, KM_USER1); | |
499 | } | |
500 | while (count) { | |
501 | int err; | |
502 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
503 | return err; | |
504 | if (page) { | |
505 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
506 | void *buf = mapaddr + offset; | |
507 | offset += fuse_copy_do(cs, &buf, &count); | |
508 | kunmap_atomic(mapaddr, KM_USER1); | |
509 | } else | |
510 | offset += fuse_copy_do(cs, NULL, &count); | |
511 | } | |
512 | if (page && !cs->write) | |
513 | flush_dcache_page(page); | |
514 | return 0; | |
515 | } | |
516 | ||
517 | /* Copy pages in the request to/from userspace buffer */ | |
518 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
519 | int zeroing) | |
520 | { | |
521 | unsigned i; | |
522 | struct fuse_req *req = cs->req; | |
523 | unsigned offset = req->page_offset; | |
524 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
525 | ||
526 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
527 | struct page *page = req->pages[i]; | |
528 | int err = fuse_copy_page(cs, page, offset, count, zeroing); | |
529 | if (err) | |
530 | return err; | |
531 | ||
532 | nbytes -= count; | |
533 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
534 | offset = 0; | |
535 | } | |
536 | return 0; | |
537 | } | |
538 | ||
539 | /* Copy a single argument in the request to/from userspace buffer */ | |
540 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
541 | { | |
542 | while (size) { | |
543 | int err; | |
544 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
545 | return err; | |
546 | fuse_copy_do(cs, &val, &size); | |
547 | } | |
548 | return 0; | |
549 | } | |
550 | ||
551 | /* Copy request arguments to/from userspace buffer */ | |
552 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
553 | unsigned argpages, struct fuse_arg *args, | |
554 | int zeroing) | |
555 | { | |
556 | int err = 0; | |
557 | unsigned i; | |
558 | ||
559 | for (i = 0; !err && i < numargs; i++) { | |
560 | struct fuse_arg *arg = &args[i]; | |
561 | if (i == numargs - 1 && argpages) | |
562 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
563 | else | |
564 | err = fuse_copy_one(cs, arg->value, arg->size); | |
565 | } | |
566 | return err; | |
567 | } | |
568 | ||
569 | /* Wait until a request is available on the pending list */ | |
570 | static void request_wait(struct fuse_conn *fc) | |
571 | { | |
572 | DECLARE_WAITQUEUE(wait, current); | |
573 | ||
574 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
9ba7cbba | 575 | while (fc->connected && list_empty(&fc->pending)) { |
334f485d MS |
576 | set_current_state(TASK_INTERRUPTIBLE); |
577 | if (signal_pending(current)) | |
578 | break; | |
579 | ||
d7133114 | 580 | spin_unlock(&fc->lock); |
334f485d | 581 | schedule(); |
d7133114 | 582 | spin_lock(&fc->lock); |
334f485d MS |
583 | } |
584 | set_current_state(TASK_RUNNING); | |
585 | remove_wait_queue(&fc->waitq, &wait); | |
586 | } | |
587 | ||
588 | /* | |
589 | * Read a single request into the userspace filesystem's buffer. This | |
590 | * function waits until a request is available, then removes it from | |
591 | * the pending list and copies request data to userspace buffer. If | |
f9a2842e MS |
592 | * no reply is needed (FORGET) or request has been aborted or there |
593 | * was an error during the copying then it's finished by calling | |
334f485d MS |
594 | * request_end(). Otherwise add it to the processing list, and set |
595 | * the 'sent' flag. | |
596 | */ | |
597 | static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov, | |
598 | unsigned long nr_segs, loff_t *off) | |
599 | { | |
600 | int err; | |
334f485d MS |
601 | struct fuse_req *req; |
602 | struct fuse_in *in; | |
603 | struct fuse_copy_state cs; | |
604 | unsigned reqsize; | |
0720b315 MS |
605 | struct fuse_conn *fc = fuse_get_conn(file); |
606 | if (!fc) | |
607 | return -EPERM; | |
334f485d | 608 | |
1d3d752b | 609 | restart: |
d7133114 | 610 | spin_lock(&fc->lock); |
e5ac1d1e JD |
611 | err = -EAGAIN; |
612 | if ((file->f_flags & O_NONBLOCK) && fc->connected && | |
613 | list_empty(&fc->pending)) | |
614 | goto err_unlock; | |
615 | ||
334f485d MS |
616 | request_wait(fc); |
617 | err = -ENODEV; | |
9ba7cbba | 618 | if (!fc->connected) |
334f485d MS |
619 | goto err_unlock; |
620 | err = -ERESTARTSYS; | |
621 | if (list_empty(&fc->pending)) | |
622 | goto err_unlock; | |
623 | ||
624 | req = list_entry(fc->pending.next, struct fuse_req, list); | |
83cfd493 | 625 | req->state = FUSE_REQ_READING; |
d77a1d5b | 626 | list_move(&req->list, &fc->io); |
334f485d MS |
627 | |
628 | in = &req->in; | |
1d3d752b MS |
629 | reqsize = in->h.len; |
630 | /* If request is too large, reply with an error and restart the read */ | |
631 | if (iov_length(iov, nr_segs) < reqsize) { | |
632 | req->out.h.error = -EIO; | |
633 | /* SETXATTR is special, since it may contain too large data */ | |
634 | if (in->h.opcode == FUSE_SETXATTR) | |
635 | req->out.h.error = -E2BIG; | |
636 | request_end(fc, req); | |
637 | goto restart; | |
334f485d | 638 | } |
d7133114 MS |
639 | spin_unlock(&fc->lock); |
640 | fuse_copy_init(&cs, fc, 1, req, iov, nr_segs); | |
1d3d752b MS |
641 | err = fuse_copy_one(&cs, &in->h, sizeof(in->h)); |
642 | if (!err) | |
643 | err = fuse_copy_args(&cs, in->numargs, in->argpages, | |
644 | (struct fuse_arg *) in->args, 0); | |
334f485d | 645 | fuse_copy_finish(&cs); |
d7133114 | 646 | spin_lock(&fc->lock); |
334f485d | 647 | req->locked = 0; |
f9a2842e | 648 | if (!err && req->aborted) |
334f485d MS |
649 | err = -ENOENT; |
650 | if (err) { | |
f9a2842e | 651 | if (!req->aborted) |
334f485d MS |
652 | req->out.h.error = -EIO; |
653 | request_end(fc, req); | |
654 | return err; | |
655 | } | |
656 | if (!req->isreply) | |
657 | request_end(fc, req); | |
658 | else { | |
83cfd493 | 659 | req->state = FUSE_REQ_SENT; |
d77a1d5b | 660 | list_move_tail(&req->list, &fc->processing); |
d7133114 | 661 | spin_unlock(&fc->lock); |
334f485d MS |
662 | } |
663 | return reqsize; | |
664 | ||
665 | err_unlock: | |
d7133114 | 666 | spin_unlock(&fc->lock); |
334f485d MS |
667 | return err; |
668 | } | |
669 | ||
670 | static ssize_t fuse_dev_read(struct file *file, char __user *buf, | |
671 | size_t nbytes, loff_t *off) | |
672 | { | |
673 | struct iovec iov; | |
674 | iov.iov_len = nbytes; | |
675 | iov.iov_base = buf; | |
676 | return fuse_dev_readv(file, &iov, 1, off); | |
677 | } | |
678 | ||
679 | /* Look up request on processing list by unique ID */ | |
680 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
681 | { | |
682 | struct list_head *entry; | |
683 | ||
684 | list_for_each(entry, &fc->processing) { | |
685 | struct fuse_req *req; | |
686 | req = list_entry(entry, struct fuse_req, list); | |
687 | if (req->in.h.unique == unique) | |
688 | return req; | |
689 | } | |
690 | return NULL; | |
691 | } | |
692 | ||
693 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
694 | unsigned nbytes) | |
695 | { | |
696 | unsigned reqsize = sizeof(struct fuse_out_header); | |
697 | ||
698 | if (out->h.error) | |
699 | return nbytes != reqsize ? -EINVAL : 0; | |
700 | ||
701 | reqsize += len_args(out->numargs, out->args); | |
702 | ||
703 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
704 | return -EINVAL; | |
705 | else if (reqsize > nbytes) { | |
706 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
707 | unsigned diffsize = reqsize - nbytes; | |
708 | if (diffsize > lastarg->size) | |
709 | return -EINVAL; | |
710 | lastarg->size -= diffsize; | |
711 | } | |
712 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
713 | out->page_zeroing); | |
714 | } | |
715 | ||
716 | /* | |
717 | * Write a single reply to a request. First the header is copied from | |
718 | * the write buffer. The request is then searched on the processing | |
719 | * list by the unique ID found in the header. If found, then remove | |
720 | * it from the list and copy the rest of the buffer to the request. | |
721 | * The request is finished by calling request_end() | |
722 | */ | |
723 | static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov, | |
724 | unsigned long nr_segs, loff_t *off) | |
725 | { | |
726 | int err; | |
727 | unsigned nbytes = iov_length(iov, nr_segs); | |
728 | struct fuse_req *req; | |
729 | struct fuse_out_header oh; | |
730 | struct fuse_copy_state cs; | |
731 | struct fuse_conn *fc = fuse_get_conn(file); | |
732 | if (!fc) | |
a87046d8 | 733 | return -EPERM; |
334f485d | 734 | |
d7133114 | 735 | fuse_copy_init(&cs, fc, 0, NULL, iov, nr_segs); |
334f485d MS |
736 | if (nbytes < sizeof(struct fuse_out_header)) |
737 | return -EINVAL; | |
738 | ||
739 | err = fuse_copy_one(&cs, &oh, sizeof(oh)); | |
740 | if (err) | |
741 | goto err_finish; | |
742 | err = -EINVAL; | |
743 | if (!oh.unique || oh.error <= -1000 || oh.error > 0 || | |
744 | oh.len != nbytes) | |
745 | goto err_finish; | |
746 | ||
d7133114 | 747 | spin_lock(&fc->lock); |
69a53bf2 MS |
748 | err = -ENOENT; |
749 | if (!fc->connected) | |
750 | goto err_unlock; | |
751 | ||
334f485d MS |
752 | req = request_find(fc, oh.unique); |
753 | err = -EINVAL; | |
754 | if (!req) | |
755 | goto err_unlock; | |
756 | ||
f9a2842e | 757 | if (req->aborted) { |
d7133114 | 758 | spin_unlock(&fc->lock); |
334f485d | 759 | fuse_copy_finish(&cs); |
d7133114 | 760 | spin_lock(&fc->lock); |
222f1d69 | 761 | request_end(fc, req); |
334f485d MS |
762 | return -ENOENT; |
763 | } | |
d77a1d5b | 764 | list_move(&req->list, &fc->io); |
334f485d MS |
765 | req->out.h = oh; |
766 | req->locked = 1; | |
767 | cs.req = req; | |
d7133114 | 768 | spin_unlock(&fc->lock); |
334f485d MS |
769 | |
770 | err = copy_out_args(&cs, &req->out, nbytes); | |
771 | fuse_copy_finish(&cs); | |
772 | ||
d7133114 | 773 | spin_lock(&fc->lock); |
334f485d MS |
774 | req->locked = 0; |
775 | if (!err) { | |
f9a2842e | 776 | if (req->aborted) |
334f485d | 777 | err = -ENOENT; |
f9a2842e | 778 | } else if (!req->aborted) |
334f485d MS |
779 | req->out.h.error = -EIO; |
780 | request_end(fc, req); | |
781 | ||
782 | return err ? err : nbytes; | |
783 | ||
784 | err_unlock: | |
d7133114 | 785 | spin_unlock(&fc->lock); |
334f485d MS |
786 | err_finish: |
787 | fuse_copy_finish(&cs); | |
788 | return err; | |
789 | } | |
790 | ||
791 | static ssize_t fuse_dev_write(struct file *file, const char __user *buf, | |
792 | size_t nbytes, loff_t *off) | |
793 | { | |
794 | struct iovec iov; | |
795 | iov.iov_len = nbytes; | |
796 | iov.iov_base = (char __user *) buf; | |
797 | return fuse_dev_writev(file, &iov, 1, off); | |
798 | } | |
799 | ||
800 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) | |
801 | { | |
334f485d | 802 | unsigned mask = POLLOUT | POLLWRNORM; |
7025d9ad | 803 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 804 | if (!fc) |
7025d9ad | 805 | return POLLERR; |
334f485d MS |
806 | |
807 | poll_wait(file, &fc->waitq, wait); | |
808 | ||
d7133114 | 809 | spin_lock(&fc->lock); |
7025d9ad MS |
810 | if (!fc->connected) |
811 | mask = POLLERR; | |
812 | else if (!list_empty(&fc->pending)) | |
813 | mask |= POLLIN | POLLRDNORM; | |
d7133114 | 814 | spin_unlock(&fc->lock); |
334f485d MS |
815 | |
816 | return mask; | |
817 | } | |
818 | ||
69a53bf2 MS |
819 | /* |
820 | * Abort all requests on the given list (pending or processing) | |
821 | * | |
d7133114 | 822 | * This function releases and reacquires fc->lock |
69a53bf2 | 823 | */ |
334f485d MS |
824 | static void end_requests(struct fuse_conn *fc, struct list_head *head) |
825 | { | |
826 | while (!list_empty(head)) { | |
827 | struct fuse_req *req; | |
828 | req = list_entry(head->next, struct fuse_req, list); | |
334f485d MS |
829 | req->out.h.error = -ECONNABORTED; |
830 | request_end(fc, req); | |
d7133114 | 831 | spin_lock(&fc->lock); |
334f485d MS |
832 | } |
833 | } | |
834 | ||
69a53bf2 MS |
835 | /* |
836 | * Abort requests under I/O | |
837 | * | |
f9a2842e | 838 | * The requests are set to aborted and finished, and the request |
69a53bf2 MS |
839 | * waiter is woken up. This will make request_wait_answer() wait |
840 | * until the request is unlocked and then return. | |
64c6d8ed MS |
841 | * |
842 | * If the request is asynchronous, then the end function needs to be | |
843 | * called after waiting for the request to be unlocked (if it was | |
844 | * locked). | |
69a53bf2 MS |
845 | */ |
846 | static void end_io_requests(struct fuse_conn *fc) | |
847 | { | |
848 | while (!list_empty(&fc->io)) { | |
64c6d8ed MS |
849 | struct fuse_req *req = |
850 | list_entry(fc->io.next, struct fuse_req, list); | |
851 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
852 | ||
f9a2842e | 853 | req->aborted = 1; |
69a53bf2 MS |
854 | req->out.h.error = -ECONNABORTED; |
855 | req->state = FUSE_REQ_FINISHED; | |
856 | list_del_init(&req->list); | |
857 | wake_up(&req->waitq); | |
64c6d8ed MS |
858 | if (end) { |
859 | req->end = NULL; | |
860 | /* The end function will consume this reference */ | |
861 | __fuse_get_request(req); | |
d7133114 | 862 | spin_unlock(&fc->lock); |
64c6d8ed MS |
863 | wait_event(req->waitq, !req->locked); |
864 | end(fc, req); | |
d7133114 | 865 | spin_lock(&fc->lock); |
64c6d8ed | 866 | } |
69a53bf2 MS |
867 | } |
868 | } | |
869 | ||
870 | /* | |
871 | * Abort all requests. | |
872 | * | |
873 | * Emergency exit in case of a malicious or accidental deadlock, or | |
874 | * just a hung filesystem. | |
875 | * | |
876 | * The same effect is usually achievable through killing the | |
877 | * filesystem daemon and all users of the filesystem. The exception | |
878 | * is the combination of an asynchronous request and the tricky | |
879 | * deadlock (see Documentation/filesystems/fuse.txt). | |
880 | * | |
881 | * During the aborting, progression of requests from the pending and | |
882 | * processing lists onto the io list, and progression of new requests | |
883 | * onto the pending list is prevented by req->connected being false. | |
884 | * | |
885 | * Progression of requests under I/O to the processing list is | |
f9a2842e MS |
886 | * prevented by the req->aborted flag being true for these requests. |
887 | * For this reason requests on the io list must be aborted first. | |
69a53bf2 MS |
888 | */ |
889 | void fuse_abort_conn(struct fuse_conn *fc) | |
890 | { | |
d7133114 | 891 | spin_lock(&fc->lock); |
69a53bf2 MS |
892 | if (fc->connected) { |
893 | fc->connected = 0; | |
51eb01e7 | 894 | fc->blocked = 0; |
69a53bf2 MS |
895 | end_io_requests(fc); |
896 | end_requests(fc, &fc->pending); | |
897 | end_requests(fc, &fc->processing); | |
898 | wake_up_all(&fc->waitq); | |
51eb01e7 | 899 | wake_up_all(&fc->blocked_waitq); |
385a17bf | 900 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
69a53bf2 | 901 | } |
d7133114 | 902 | spin_unlock(&fc->lock); |
69a53bf2 MS |
903 | } |
904 | ||
334f485d MS |
905 | static int fuse_dev_release(struct inode *inode, struct file *file) |
906 | { | |
0720b315 | 907 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 908 | if (fc) { |
d7133114 | 909 | spin_lock(&fc->lock); |
1e9a4ed9 | 910 | fc->connected = 0; |
334f485d MS |
911 | end_requests(fc, &fc->pending); |
912 | end_requests(fc, &fc->processing); | |
d7133114 | 913 | spin_unlock(&fc->lock); |
385a17bf | 914 | fasync_helper(-1, file, 0, &fc->fasync); |
bafa9654 | 915 | fuse_conn_put(fc); |
385a17bf | 916 | } |
f543f253 | 917 | |
334f485d MS |
918 | return 0; |
919 | } | |
920 | ||
385a17bf JD |
921 | static int fuse_dev_fasync(int fd, struct file *file, int on) |
922 | { | |
923 | struct fuse_conn *fc = fuse_get_conn(file); | |
924 | if (!fc) | |
a87046d8 | 925 | return -EPERM; |
385a17bf JD |
926 | |
927 | /* No locking - fasync_helper does its own locking */ | |
928 | return fasync_helper(fd, file, on, &fc->fasync); | |
929 | } | |
930 | ||
4b6f5d20 | 931 | const struct file_operations fuse_dev_operations = { |
334f485d MS |
932 | .owner = THIS_MODULE, |
933 | .llseek = no_llseek, | |
934 | .read = fuse_dev_read, | |
935 | .readv = fuse_dev_readv, | |
936 | .write = fuse_dev_write, | |
937 | .writev = fuse_dev_writev, | |
938 | .poll = fuse_dev_poll, | |
939 | .release = fuse_dev_release, | |
385a17bf | 940 | .fasync = fuse_dev_fasync, |
334f485d MS |
941 | }; |
942 | ||
943 | static struct miscdevice fuse_miscdevice = { | |
944 | .minor = FUSE_MINOR, | |
945 | .name = "fuse", | |
946 | .fops = &fuse_dev_operations, | |
947 | }; | |
948 | ||
949 | int __init fuse_dev_init(void) | |
950 | { | |
951 | int err = -ENOMEM; | |
952 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
953 | sizeof(struct fuse_req), | |
954 | 0, 0, NULL, NULL); | |
955 | if (!fuse_req_cachep) | |
956 | goto out; | |
957 | ||
958 | err = misc_register(&fuse_miscdevice); | |
959 | if (err) | |
960 | goto out_cache_clean; | |
961 | ||
962 | return 0; | |
963 | ||
964 | out_cache_clean: | |
965 | kmem_cache_destroy(fuse_req_cachep); | |
966 | out: | |
967 | return err; | |
968 | } | |
969 | ||
970 | void fuse_dev_cleanup(void) | |
971 | { | |
972 | misc_deregister(&fuse_miscdevice); | |
973 | kmem_cache_destroy(fuse_req_cachep); | |
974 | } |