]>
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 | ||
24 | static inline struct fuse_conn *fuse_get_conn(struct file *file) | |
25 | { | |
26 | struct fuse_conn *fc; | |
27 | spin_lock(&fuse_lock); | |
28 | fc = file->private_data; | |
1e9a4ed9 | 29 | if (fc && !fc->mounted) |
334f485d MS |
30 | fc = NULL; |
31 | spin_unlock(&fuse_lock); | |
32 | return fc; | |
33 | } | |
34 | ||
35 | static inline void fuse_request_init(struct fuse_req *req) | |
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 | ||
56 | static inline void block_sigs(sigset_t *oldset) | |
57 | { | |
58 | sigset_t mask; | |
59 | ||
60 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
61 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
62 | } | |
63 | ||
64 | static inline void restore_sigs(sigset_t *oldset) | |
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 | ||
112 | block_sigs(&oldset); | |
113 | intr = down_interruptible(&fc->outstanding_sem); | |
114 | restore_sigs(&oldset); | |
115 | return intr ? NULL : do_get_request(fc); | |
116 | } | |
117 | ||
118 | static void fuse_putback_request(struct fuse_conn *fc, struct fuse_req *req) | |
119 | { | |
120 | spin_lock(&fuse_lock); | |
121 | if (req->preallocated) | |
122 | list_add(&req->list, &fc->unused_list); | |
123 | else | |
124 | fuse_request_free(req); | |
125 | ||
126 | /* If we are in debt decrease that first */ | |
127 | if (fc->outstanding_debt) | |
128 | fc->outstanding_debt--; | |
129 | else | |
130 | up(&fc->outstanding_sem); | |
131 | spin_unlock(&fuse_lock); | |
132 | } | |
133 | ||
134 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) | |
135 | { | |
136 | if (atomic_dec_and_test(&req->count)) | |
137 | fuse_putback_request(fc, req); | |
138 | } | |
139 | ||
1e9a4ed9 MS |
140 | void fuse_release_background(struct fuse_req *req) |
141 | { | |
142 | iput(req->inode); | |
143 | iput(req->inode2); | |
144 | if (req->file) | |
145 | fput(req->file); | |
146 | spin_lock(&fuse_lock); | |
147 | list_del(&req->bg_entry); | |
148 | spin_unlock(&fuse_lock); | |
149 | } | |
150 | ||
334f485d MS |
151 | /* |
152 | * This function is called when a request is finished. Either a reply | |
153 | * has arrived or it was interrupted (and not yet sent) or some error | |
1779381d MS |
154 | * occurred during communication with userspace, or the device file was |
155 | * closed. It decreases the reference count for the request. In case | |
156 | * of a background request the reference to the stored objects are | |
334f485d MS |
157 | * released. The requester thread is woken up (if still waiting), and |
158 | * finally the request is either freed or put on the unused_list | |
159 | * | |
160 | * Called with fuse_lock, unlocks it | |
161 | */ | |
162 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
163 | { | |
164 | int putback; | |
165 | req->finished = 1; | |
166 | putback = atomic_dec_and_test(&req->count); | |
167 | spin_unlock(&fuse_lock); | |
168 | if (req->background) { | |
1e9a4ed9 MS |
169 | down_read(&fc->sbput_sem); |
170 | if (fc->mounted) | |
171 | fuse_release_background(req); | |
172 | up_read(&fc->sbput_sem); | |
334f485d MS |
173 | } |
174 | wake_up(&req->waitq); | |
175 | if (req->in.h.opcode == FUSE_INIT) { | |
176 | int i; | |
177 | ||
178 | if (req->misc.init_in_out.major != FUSE_KERNEL_VERSION) | |
179 | fc->conn_error = 1; | |
180 | ||
45714d65 MS |
181 | fc->minor = req->misc.init_in_out.minor; |
182 | ||
334f485d MS |
183 | /* After INIT reply is received other requests can go |
184 | out. So do (FUSE_MAX_OUTSTANDING - 1) number of | |
185 | up()s on outstanding_sem. The last up() is done in | |
186 | fuse_putback_request() */ | |
187 | for (i = 1; i < FUSE_MAX_OUTSTANDING; i++) | |
188 | up(&fc->outstanding_sem); | |
fd72faac MS |
189 | } else if (req->in.h.opcode == FUSE_RELEASE && req->inode == NULL) { |
190 | /* Special case for failed iget in CREATE */ | |
191 | u64 nodeid = req->in.h.nodeid; | |
192 | __fuse_get_request(req); | |
193 | fuse_reset_request(req); | |
194 | fuse_send_forget(fc, req, nodeid, 1); | |
195 | putback = 0; | |
334f485d MS |
196 | } |
197 | if (putback) | |
198 | fuse_putback_request(fc, req); | |
199 | } | |
200 | ||
1e9a4ed9 MS |
201 | /* |
202 | * Unfortunately request interruption not just solves the deadlock | |
203 | * problem, it causes problems too. These stem from the fact, that an | |
204 | * interrupted request is continued to be processed in userspace, | |
205 | * while all the locks and object references (inode and file) held | |
206 | * during the operation are released. | |
207 | * | |
208 | * To release the locks is exactly why there's a need to interrupt the | |
209 | * request, so there's not a lot that can be done about this, except | |
210 | * introduce additional locking in userspace. | |
211 | * | |
212 | * More important is to keep inode and file references until userspace | |
213 | * has replied, otherwise FORGET and RELEASE could be sent while the | |
214 | * inode/file is still used by the filesystem. | |
215 | * | |
216 | * For this reason the concept of "background" request is introduced. | |
217 | * An interrupted request is backgrounded if it has been already sent | |
218 | * to userspace. Backgrounding involves getting an extra reference to | |
219 | * inode(s) or file used in the request, and adding the request to | |
220 | * fc->background list. When a reply is received for a background | |
221 | * request, the object references are released, and the request is | |
222 | * removed from the list. If the filesystem is unmounted while there | |
223 | * are still background requests, the list is walked and references | |
224 | * are released as if a reply was received. | |
225 | * | |
226 | * There's one more use for a background request. The RELEASE message is | |
227 | * always sent as background, since it doesn't return an error or | |
228 | * data. | |
229 | */ | |
230 | static void background_request(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d | 231 | { |
334f485d | 232 | req->background = 1; |
1e9a4ed9 | 233 | list_add(&req->bg_entry, &fc->background); |
334f485d MS |
234 | if (req->inode) |
235 | req->inode = igrab(req->inode); | |
236 | if (req->inode2) | |
237 | req->inode2 = igrab(req->inode2); | |
238 | if (req->file) | |
239 | get_file(req->file); | |
240 | } | |
241 | ||
334f485d | 242 | /* Called with fuse_lock held. Releases, and then reacquires it. */ |
7c352bdf | 243 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) |
334f485d | 244 | { |
7c352bdf | 245 | sigset_t oldset; |
334f485d MS |
246 | |
247 | spin_unlock(&fuse_lock); | |
7c352bdf MS |
248 | block_sigs(&oldset); |
249 | wait_event_interruptible(req->waitq, req->finished); | |
250 | restore_sigs(&oldset); | |
334f485d | 251 | spin_lock(&fuse_lock); |
7c352bdf | 252 | if (req->finished) |
334f485d MS |
253 | return; |
254 | ||
7c352bdf | 255 | req->out.h.error = -EINTR; |
334f485d MS |
256 | req->interrupted = 1; |
257 | if (req->locked) { | |
258 | /* This is uninterruptible sleep, because data is | |
259 | being copied to/from the buffers of req. During | |
260 | locked state, there mustn't be any filesystem | |
261 | operation (e.g. page fault), since that could lead | |
262 | to deadlock */ | |
263 | spin_unlock(&fuse_lock); | |
264 | wait_event(req->waitq, !req->locked); | |
265 | spin_lock(&fuse_lock); | |
266 | } | |
267 | if (!req->sent && !list_empty(&req->list)) { | |
268 | list_del(&req->list); | |
269 | __fuse_put_request(req); | |
270 | } else if (!req->finished && req->sent) | |
1e9a4ed9 | 271 | background_request(fc, req); |
334f485d MS |
272 | } |
273 | ||
274 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) | |
275 | { | |
276 | unsigned nbytes = 0; | |
277 | unsigned i; | |
278 | ||
279 | for (i = 0; i < numargs; i++) | |
280 | nbytes += args[i].size; | |
281 | ||
282 | return nbytes; | |
283 | } | |
284 | ||
285 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) | |
286 | { | |
287 | fc->reqctr++; | |
288 | /* zero is special */ | |
289 | if (fc->reqctr == 0) | |
290 | fc->reqctr = 1; | |
291 | req->in.h.unique = fc->reqctr; | |
292 | req->in.h.len = sizeof(struct fuse_in_header) + | |
293 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
294 | if (!req->preallocated) { | |
295 | /* If request is not preallocated (either FORGET or | |
296 | RELEASE), then still decrease outstanding_sem, so | |
297 | user can't open infinite number of files while not | |
298 | processing the RELEASE requests. However for | |
299 | efficiency do it without blocking, so if down() | |
300 | would block, just increase the debt instead */ | |
301 | if (down_trylock(&fc->outstanding_sem)) | |
302 | fc->outstanding_debt++; | |
303 | } | |
304 | list_add_tail(&req->list, &fc->pending); | |
305 | wake_up(&fc->waitq); | |
306 | } | |
307 | ||
7c352bdf MS |
308 | /* |
309 | * This can only be interrupted by a SIGKILL | |
310 | */ | |
311 | void request_send(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d MS |
312 | { |
313 | req->isreply = 1; | |
314 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 315 | if (!fc->connected) |
334f485d MS |
316 | req->out.h.error = -ENOTCONN; |
317 | else if (fc->conn_error) | |
318 | req->out.h.error = -ECONNREFUSED; | |
319 | else { | |
320 | queue_request(fc, req); | |
321 | /* acquire extra reference, since request is still needed | |
322 | after request_end() */ | |
323 | __fuse_get_request(req); | |
324 | ||
7c352bdf | 325 | request_wait_answer(fc, req); |
334f485d MS |
326 | } |
327 | spin_unlock(&fuse_lock); | |
328 | } | |
329 | ||
334f485d MS |
330 | static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
331 | { | |
332 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 333 | if (fc->connected) { |
334f485d MS |
334 | queue_request(fc, req); |
335 | spin_unlock(&fuse_lock); | |
336 | } else { | |
337 | req->out.h.error = -ENOTCONN; | |
338 | request_end(fc, req); | |
339 | } | |
340 | } | |
341 | ||
342 | void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req) | |
343 | { | |
344 | req->isreply = 0; | |
345 | request_send_nowait(fc, req); | |
346 | } | |
347 | ||
348 | void request_send_background(struct fuse_conn *fc, struct fuse_req *req) | |
349 | { | |
350 | req->isreply = 1; | |
1e9a4ed9 MS |
351 | spin_lock(&fuse_lock); |
352 | background_request(fc, req); | |
353 | spin_unlock(&fuse_lock); | |
334f485d MS |
354 | request_send_nowait(fc, req); |
355 | } | |
356 | ||
357 | void fuse_send_init(struct fuse_conn *fc) | |
358 | { | |
359 | /* This is called from fuse_read_super() so there's guaranteed | |
360 | to be a request available */ | |
361 | struct fuse_req *req = do_get_request(fc); | |
362 | struct fuse_init_in_out *arg = &req->misc.init_in_out; | |
363 | arg->major = FUSE_KERNEL_VERSION; | |
364 | arg->minor = FUSE_KERNEL_MINOR_VERSION; | |
365 | req->in.h.opcode = FUSE_INIT; | |
366 | req->in.numargs = 1; | |
367 | req->in.args[0].size = sizeof(*arg); | |
368 | req->in.args[0].value = arg; | |
369 | req->out.numargs = 1; | |
370 | req->out.args[0].size = sizeof(*arg); | |
371 | req->out.args[0].value = arg; | |
372 | request_send_background(fc, req); | |
373 | } | |
374 | ||
375 | /* | |
376 | * Lock the request. Up to the next unlock_request() there mustn't be | |
377 | * anything that could cause a page-fault. If the request was already | |
378 | * interrupted bail out. | |
379 | */ | |
380 | static inline int lock_request(struct fuse_req *req) | |
381 | { | |
382 | int err = 0; | |
383 | if (req) { | |
384 | spin_lock(&fuse_lock); | |
385 | if (req->interrupted) | |
386 | err = -ENOENT; | |
387 | else | |
388 | req->locked = 1; | |
389 | spin_unlock(&fuse_lock); | |
390 | } | |
391 | return err; | |
392 | } | |
393 | ||
394 | /* | |
395 | * Unlock request. If it was interrupted during being locked, the | |
396 | * requester thread is currently waiting for it to be unlocked, so | |
397 | * wake it up. | |
398 | */ | |
399 | static inline void unlock_request(struct fuse_req *req) | |
400 | { | |
401 | if (req) { | |
402 | spin_lock(&fuse_lock); | |
403 | req->locked = 0; | |
404 | if (req->interrupted) | |
405 | wake_up(&req->waitq); | |
406 | spin_unlock(&fuse_lock); | |
407 | } | |
408 | } | |
409 | ||
410 | struct fuse_copy_state { | |
411 | int write; | |
412 | struct fuse_req *req; | |
413 | const struct iovec *iov; | |
414 | unsigned long nr_segs; | |
415 | unsigned long seglen; | |
416 | unsigned long addr; | |
417 | struct page *pg; | |
418 | void *mapaddr; | |
419 | void *buf; | |
420 | unsigned len; | |
421 | }; | |
422 | ||
423 | static void fuse_copy_init(struct fuse_copy_state *cs, int write, | |
424 | struct fuse_req *req, const struct iovec *iov, | |
425 | unsigned long nr_segs) | |
426 | { | |
427 | memset(cs, 0, sizeof(*cs)); | |
428 | cs->write = write; | |
429 | cs->req = req; | |
430 | cs->iov = iov; | |
431 | cs->nr_segs = nr_segs; | |
432 | } | |
433 | ||
434 | /* Unmap and put previous page of userspace buffer */ | |
435 | static inline void fuse_copy_finish(struct fuse_copy_state *cs) | |
436 | { | |
437 | if (cs->mapaddr) { | |
438 | kunmap_atomic(cs->mapaddr, KM_USER0); | |
439 | if (cs->write) { | |
440 | flush_dcache_page(cs->pg); | |
441 | set_page_dirty_lock(cs->pg); | |
442 | } | |
443 | put_page(cs->pg); | |
444 | cs->mapaddr = NULL; | |
445 | } | |
446 | } | |
447 | ||
448 | /* | |
449 | * Get another pagefull of userspace buffer, and map it to kernel | |
450 | * address space, and lock request | |
451 | */ | |
452 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
453 | { | |
454 | unsigned long offset; | |
455 | int err; | |
456 | ||
457 | unlock_request(cs->req); | |
458 | fuse_copy_finish(cs); | |
459 | if (!cs->seglen) { | |
460 | BUG_ON(!cs->nr_segs); | |
461 | cs->seglen = cs->iov[0].iov_len; | |
462 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
463 | cs->iov ++; | |
464 | cs->nr_segs --; | |
465 | } | |
466 | down_read(¤t->mm->mmap_sem); | |
467 | err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0, | |
468 | &cs->pg, NULL); | |
469 | up_read(¤t->mm->mmap_sem); | |
470 | if (err < 0) | |
471 | return err; | |
472 | BUG_ON(err != 1); | |
473 | offset = cs->addr % PAGE_SIZE; | |
474 | cs->mapaddr = kmap_atomic(cs->pg, KM_USER0); | |
475 | cs->buf = cs->mapaddr + offset; | |
476 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
477 | cs->seglen -= cs->len; | |
478 | cs->addr += cs->len; | |
479 | ||
480 | return lock_request(cs->req); | |
481 | } | |
482 | ||
483 | /* Do as much copy to/from userspace buffer as we can */ | |
484 | static inline int fuse_copy_do(struct fuse_copy_state *cs, void **val, | |
485 | unsigned *size) | |
486 | { | |
487 | unsigned ncpy = min(*size, cs->len); | |
488 | if (val) { | |
489 | if (cs->write) | |
490 | memcpy(cs->buf, *val, ncpy); | |
491 | else | |
492 | memcpy(*val, cs->buf, ncpy); | |
493 | *val += ncpy; | |
494 | } | |
495 | *size -= ncpy; | |
496 | cs->len -= ncpy; | |
497 | cs->buf += ncpy; | |
498 | return ncpy; | |
499 | } | |
500 | ||
501 | /* | |
502 | * Copy a page in the request to/from the userspace buffer. Must be | |
503 | * done atomically | |
504 | */ | |
505 | static inline int fuse_copy_page(struct fuse_copy_state *cs, struct page *page, | |
506 | unsigned offset, unsigned count, int zeroing) | |
507 | { | |
508 | if (page && zeroing && count < PAGE_SIZE) { | |
509 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
510 | memset(mapaddr, 0, PAGE_SIZE); | |
511 | kunmap_atomic(mapaddr, KM_USER1); | |
512 | } | |
513 | while (count) { | |
514 | int err; | |
515 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
516 | return err; | |
517 | if (page) { | |
518 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
519 | void *buf = mapaddr + offset; | |
520 | offset += fuse_copy_do(cs, &buf, &count); | |
521 | kunmap_atomic(mapaddr, KM_USER1); | |
522 | } else | |
523 | offset += fuse_copy_do(cs, NULL, &count); | |
524 | } | |
525 | if (page && !cs->write) | |
526 | flush_dcache_page(page); | |
527 | return 0; | |
528 | } | |
529 | ||
530 | /* Copy pages in the request to/from userspace buffer */ | |
531 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
532 | int zeroing) | |
533 | { | |
534 | unsigned i; | |
535 | struct fuse_req *req = cs->req; | |
536 | unsigned offset = req->page_offset; | |
537 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
538 | ||
539 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
540 | struct page *page = req->pages[i]; | |
541 | int err = fuse_copy_page(cs, page, offset, count, zeroing); | |
542 | if (err) | |
543 | return err; | |
544 | ||
545 | nbytes -= count; | |
546 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
547 | offset = 0; | |
548 | } | |
549 | return 0; | |
550 | } | |
551 | ||
552 | /* Copy a single argument in the request to/from userspace buffer */ | |
553 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
554 | { | |
555 | while (size) { | |
556 | int err; | |
557 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
558 | return err; | |
559 | fuse_copy_do(cs, &val, &size); | |
560 | } | |
561 | return 0; | |
562 | } | |
563 | ||
564 | /* Copy request arguments to/from userspace buffer */ | |
565 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
566 | unsigned argpages, struct fuse_arg *args, | |
567 | int zeroing) | |
568 | { | |
569 | int err = 0; | |
570 | unsigned i; | |
571 | ||
572 | for (i = 0; !err && i < numargs; i++) { | |
573 | struct fuse_arg *arg = &args[i]; | |
574 | if (i == numargs - 1 && argpages) | |
575 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
576 | else | |
577 | err = fuse_copy_one(cs, arg->value, arg->size); | |
578 | } | |
579 | return err; | |
580 | } | |
581 | ||
582 | /* Wait until a request is available on the pending list */ | |
583 | static void request_wait(struct fuse_conn *fc) | |
584 | { | |
585 | DECLARE_WAITQUEUE(wait, current); | |
586 | ||
587 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
1e9a4ed9 | 588 | while (fc->mounted && list_empty(&fc->pending)) { |
334f485d MS |
589 | set_current_state(TASK_INTERRUPTIBLE); |
590 | if (signal_pending(current)) | |
591 | break; | |
592 | ||
593 | spin_unlock(&fuse_lock); | |
594 | schedule(); | |
595 | spin_lock(&fuse_lock); | |
596 | } | |
597 | set_current_state(TASK_RUNNING); | |
598 | remove_wait_queue(&fc->waitq, &wait); | |
599 | } | |
600 | ||
601 | /* | |
602 | * Read a single request into the userspace filesystem's buffer. This | |
603 | * function waits until a request is available, then removes it from | |
604 | * the pending list and copies request data to userspace buffer. If | |
605 | * no reply is needed (FORGET) or request has been interrupted or | |
606 | * there was an error during the copying then it's finished by calling | |
607 | * request_end(). Otherwise add it to the processing list, and set | |
608 | * the 'sent' flag. | |
609 | */ | |
610 | static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov, | |
611 | unsigned long nr_segs, loff_t *off) | |
612 | { | |
613 | int err; | |
614 | struct fuse_conn *fc; | |
615 | struct fuse_req *req; | |
616 | struct fuse_in *in; | |
617 | struct fuse_copy_state cs; | |
618 | unsigned reqsize; | |
619 | ||
620 | spin_lock(&fuse_lock); | |
621 | fc = file->private_data; | |
622 | err = -EPERM; | |
623 | if (!fc) | |
624 | goto err_unlock; | |
625 | request_wait(fc); | |
626 | err = -ENODEV; | |
1e9a4ed9 | 627 | if (!fc->mounted) |
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); | |
634 | list_del_init(&req->list); | |
635 | spin_unlock(&fuse_lock); | |
636 | ||
637 | in = &req->in; | |
638 | reqsize = req->in.h.len; | |
639 | fuse_copy_init(&cs, 1, req, iov, nr_segs); | |
640 | err = -EINVAL; | |
641 | if (iov_length(iov, nr_segs) >= reqsize) { | |
642 | err = fuse_copy_one(&cs, &in->h, sizeof(in->h)); | |
643 | if (!err) | |
644 | err = fuse_copy_args(&cs, in->numargs, in->argpages, | |
645 | (struct fuse_arg *) in->args, 0); | |
646 | } | |
647 | fuse_copy_finish(&cs); | |
648 | ||
649 | spin_lock(&fuse_lock); | |
650 | req->locked = 0; | |
651 | if (!err && req->interrupted) | |
652 | err = -ENOENT; | |
653 | if (err) { | |
654 | if (!req->interrupted) | |
655 | req->out.h.error = -EIO; | |
656 | request_end(fc, req); | |
657 | return err; | |
658 | } | |
659 | if (!req->isreply) | |
660 | request_end(fc, req); | |
661 | else { | |
662 | req->sent = 1; | |
663 | list_add_tail(&req->list, &fc->processing); | |
664 | spin_unlock(&fuse_lock); | |
665 | } | |
666 | return reqsize; | |
667 | ||
668 | err_unlock: | |
669 | spin_unlock(&fuse_lock); | |
670 | return err; | |
671 | } | |
672 | ||
673 | static ssize_t fuse_dev_read(struct file *file, char __user *buf, | |
674 | size_t nbytes, loff_t *off) | |
675 | { | |
676 | struct iovec iov; | |
677 | iov.iov_len = nbytes; | |
678 | iov.iov_base = buf; | |
679 | return fuse_dev_readv(file, &iov, 1, off); | |
680 | } | |
681 | ||
682 | /* Look up request on processing list by unique ID */ | |
683 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
684 | { | |
685 | struct list_head *entry; | |
686 | ||
687 | list_for_each(entry, &fc->processing) { | |
688 | struct fuse_req *req; | |
689 | req = list_entry(entry, struct fuse_req, list); | |
690 | if (req->in.h.unique == unique) | |
691 | return req; | |
692 | } | |
693 | return NULL; | |
694 | } | |
695 | ||
696 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
697 | unsigned nbytes) | |
698 | { | |
699 | unsigned reqsize = sizeof(struct fuse_out_header); | |
700 | ||
701 | if (out->h.error) | |
702 | return nbytes != reqsize ? -EINVAL : 0; | |
703 | ||
704 | reqsize += len_args(out->numargs, out->args); | |
705 | ||
706 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
707 | return -EINVAL; | |
708 | else if (reqsize > nbytes) { | |
709 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
710 | unsigned diffsize = reqsize - nbytes; | |
711 | if (diffsize > lastarg->size) | |
712 | return -EINVAL; | |
713 | lastarg->size -= diffsize; | |
714 | } | |
715 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
716 | out->page_zeroing); | |
717 | } | |
718 | ||
719 | /* | |
720 | * Write a single reply to a request. First the header is copied from | |
721 | * the write buffer. The request is then searched on the processing | |
722 | * list by the unique ID found in the header. If found, then remove | |
723 | * it from the list and copy the rest of the buffer to the request. | |
724 | * The request is finished by calling request_end() | |
725 | */ | |
726 | static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov, | |
727 | unsigned long nr_segs, loff_t *off) | |
728 | { | |
729 | int err; | |
730 | unsigned nbytes = iov_length(iov, nr_segs); | |
731 | struct fuse_req *req; | |
732 | struct fuse_out_header oh; | |
733 | struct fuse_copy_state cs; | |
734 | struct fuse_conn *fc = fuse_get_conn(file); | |
735 | if (!fc) | |
736 | return -ENODEV; | |
737 | ||
738 | fuse_copy_init(&cs, 0, NULL, iov, nr_segs); | |
739 | if (nbytes < sizeof(struct fuse_out_header)) | |
740 | return -EINVAL; | |
741 | ||
742 | err = fuse_copy_one(&cs, &oh, sizeof(oh)); | |
743 | if (err) | |
744 | goto err_finish; | |
745 | err = -EINVAL; | |
746 | if (!oh.unique || oh.error <= -1000 || oh.error > 0 || | |
747 | oh.len != nbytes) | |
748 | goto err_finish; | |
749 | ||
750 | spin_lock(&fuse_lock); | |
751 | req = request_find(fc, oh.unique); | |
752 | err = -EINVAL; | |
753 | if (!req) | |
754 | goto err_unlock; | |
755 | ||
756 | list_del_init(&req->list); | |
757 | if (req->interrupted) { | |
758 | request_end(fc, req); | |
759 | fuse_copy_finish(&cs); | |
760 | return -ENOENT; | |
761 | } | |
762 | req->out.h = oh; | |
763 | req->locked = 1; | |
764 | cs.req = req; | |
765 | spin_unlock(&fuse_lock); | |
766 | ||
767 | err = copy_out_args(&cs, &req->out, nbytes); | |
768 | fuse_copy_finish(&cs); | |
769 | ||
770 | spin_lock(&fuse_lock); | |
771 | req->locked = 0; | |
772 | if (!err) { | |
773 | if (req->interrupted) | |
774 | err = -ENOENT; | |
775 | } else if (!req->interrupted) | |
776 | req->out.h.error = -EIO; | |
777 | request_end(fc, req); | |
778 | ||
779 | return err ? err : nbytes; | |
780 | ||
781 | err_unlock: | |
782 | spin_unlock(&fuse_lock); | |
783 | err_finish: | |
784 | fuse_copy_finish(&cs); | |
785 | return err; | |
786 | } | |
787 | ||
788 | static ssize_t fuse_dev_write(struct file *file, const char __user *buf, | |
789 | size_t nbytes, loff_t *off) | |
790 | { | |
791 | struct iovec iov; | |
792 | iov.iov_len = nbytes; | |
793 | iov.iov_base = (char __user *) buf; | |
794 | return fuse_dev_writev(file, &iov, 1, off); | |
795 | } | |
796 | ||
797 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) | |
798 | { | |
799 | struct fuse_conn *fc = fuse_get_conn(file); | |
800 | unsigned mask = POLLOUT | POLLWRNORM; | |
801 | ||
802 | if (!fc) | |
803 | return -ENODEV; | |
804 | ||
805 | poll_wait(file, &fc->waitq, wait); | |
806 | ||
807 | spin_lock(&fuse_lock); | |
808 | if (!list_empty(&fc->pending)) | |
809 | mask |= POLLIN | POLLRDNORM; | |
810 | spin_unlock(&fuse_lock); | |
811 | ||
812 | return mask; | |
813 | } | |
814 | ||
815 | /* Abort all requests on the given list (pending or processing) */ | |
816 | static void end_requests(struct fuse_conn *fc, struct list_head *head) | |
817 | { | |
818 | while (!list_empty(head)) { | |
819 | struct fuse_req *req; | |
820 | req = list_entry(head->next, struct fuse_req, list); | |
821 | list_del_init(&req->list); | |
822 | req->out.h.error = -ECONNABORTED; | |
823 | request_end(fc, req); | |
824 | spin_lock(&fuse_lock); | |
825 | } | |
826 | } | |
827 | ||
828 | static int fuse_dev_release(struct inode *inode, struct file *file) | |
829 | { | |
830 | struct fuse_conn *fc; | |
831 | ||
832 | spin_lock(&fuse_lock); | |
833 | fc = file->private_data; | |
834 | if (fc) { | |
1e9a4ed9 | 835 | fc->connected = 0; |
334f485d MS |
836 | end_requests(fc, &fc->pending); |
837 | end_requests(fc, &fc->processing); | |
838 | fuse_release_conn(fc); | |
839 | } | |
840 | spin_unlock(&fuse_lock); | |
841 | return 0; | |
842 | } | |
843 | ||
844 | struct file_operations fuse_dev_operations = { | |
845 | .owner = THIS_MODULE, | |
846 | .llseek = no_llseek, | |
847 | .read = fuse_dev_read, | |
848 | .readv = fuse_dev_readv, | |
849 | .write = fuse_dev_write, | |
850 | .writev = fuse_dev_writev, | |
851 | .poll = fuse_dev_poll, | |
852 | .release = fuse_dev_release, | |
853 | }; | |
854 | ||
855 | static struct miscdevice fuse_miscdevice = { | |
856 | .minor = FUSE_MINOR, | |
857 | .name = "fuse", | |
858 | .fops = &fuse_dev_operations, | |
859 | }; | |
860 | ||
861 | int __init fuse_dev_init(void) | |
862 | { | |
863 | int err = -ENOMEM; | |
864 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
865 | sizeof(struct fuse_req), | |
866 | 0, 0, NULL, NULL); | |
867 | if (!fuse_req_cachep) | |
868 | goto out; | |
869 | ||
870 | err = misc_register(&fuse_miscdevice); | |
871 | if (err) | |
872 | goto out_cache_clean; | |
873 | ||
874 | return 0; | |
875 | ||
876 | out_cache_clean: | |
877 | kmem_cache_destroy(fuse_req_cachep); | |
878 | out: | |
879 | return err; | |
880 | } | |
881 | ||
882 | void fuse_dev_cleanup(void) | |
883 | { | |
884 | misc_deregister(&fuse_miscdevice); | |
885 | kmem_cache_destroy(fuse_req_cachep); | |
886 | } |