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 | ||
106 | struct fuse_req *fuse_get_request(struct fuse_conn *fc) | |
107 | { | |
108 | if (down_interruptible(&fc->outstanding_sem)) | |
109 | return NULL; | |
110 | return do_get_request(fc); | |
111 | } | |
112 | ||
113 | /* | |
114 | * Non-interruptible version of the above function is for operations | |
115 | * which can't legally return -ERESTART{SYS,NOINTR}. This can still | |
116 | * return NULL, but only in case the signal is SIGKILL. | |
117 | */ | |
118 | struct fuse_req *fuse_get_request_nonint(struct fuse_conn *fc) | |
119 | { | |
120 | int intr; | |
121 | sigset_t oldset; | |
122 | ||
123 | block_sigs(&oldset); | |
124 | intr = down_interruptible(&fc->outstanding_sem); | |
125 | restore_sigs(&oldset); | |
126 | return intr ? NULL : do_get_request(fc); | |
127 | } | |
128 | ||
129 | static void fuse_putback_request(struct fuse_conn *fc, struct fuse_req *req) | |
130 | { | |
131 | spin_lock(&fuse_lock); | |
132 | if (req->preallocated) | |
133 | list_add(&req->list, &fc->unused_list); | |
134 | else | |
135 | fuse_request_free(req); | |
136 | ||
137 | /* If we are in debt decrease that first */ | |
138 | if (fc->outstanding_debt) | |
139 | fc->outstanding_debt--; | |
140 | else | |
141 | up(&fc->outstanding_sem); | |
142 | spin_unlock(&fuse_lock); | |
143 | } | |
144 | ||
145 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) | |
146 | { | |
147 | if (atomic_dec_and_test(&req->count)) | |
148 | fuse_putback_request(fc, req); | |
149 | } | |
150 | ||
1e9a4ed9 MS |
151 | void fuse_release_background(struct fuse_req *req) |
152 | { | |
153 | iput(req->inode); | |
154 | iput(req->inode2); | |
155 | if (req->file) | |
156 | fput(req->file); | |
157 | spin_lock(&fuse_lock); | |
158 | list_del(&req->bg_entry); | |
159 | spin_unlock(&fuse_lock); | |
160 | } | |
161 | ||
334f485d MS |
162 | /* |
163 | * This function is called when a request is finished. Either a reply | |
164 | * has arrived or it was interrupted (and not yet sent) or some error | |
165 | * occured during communication with userspace, or the device file was | |
166 | * closed. It decreases the referece count for the request. In case | |
167 | * of a background request the referece to the stored objects are | |
168 | * released. The requester thread is woken up (if still waiting), and | |
169 | * finally the request is either freed or put on the unused_list | |
170 | * | |
171 | * Called with fuse_lock, unlocks it | |
172 | */ | |
173 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
174 | { | |
175 | int putback; | |
176 | req->finished = 1; | |
177 | putback = atomic_dec_and_test(&req->count); | |
178 | spin_unlock(&fuse_lock); | |
179 | if (req->background) { | |
1e9a4ed9 MS |
180 | down_read(&fc->sbput_sem); |
181 | if (fc->mounted) | |
182 | fuse_release_background(req); | |
183 | up_read(&fc->sbput_sem); | |
334f485d MS |
184 | } |
185 | wake_up(&req->waitq); | |
186 | if (req->in.h.opcode == FUSE_INIT) { | |
187 | int i; | |
188 | ||
189 | if (req->misc.init_in_out.major != FUSE_KERNEL_VERSION) | |
190 | fc->conn_error = 1; | |
191 | ||
192 | /* After INIT reply is received other requests can go | |
193 | out. So do (FUSE_MAX_OUTSTANDING - 1) number of | |
194 | up()s on outstanding_sem. The last up() is done in | |
195 | fuse_putback_request() */ | |
196 | for (i = 1; i < FUSE_MAX_OUTSTANDING; i++) | |
197 | up(&fc->outstanding_sem); | |
198 | } | |
199 | if (putback) | |
200 | fuse_putback_request(fc, req); | |
201 | } | |
202 | ||
1e9a4ed9 MS |
203 | /* |
204 | * Unfortunately request interruption not just solves the deadlock | |
205 | * problem, it causes problems too. These stem from the fact, that an | |
206 | * interrupted request is continued to be processed in userspace, | |
207 | * while all the locks and object references (inode and file) held | |
208 | * during the operation are released. | |
209 | * | |
210 | * To release the locks is exactly why there's a need to interrupt the | |
211 | * request, so there's not a lot that can be done about this, except | |
212 | * introduce additional locking in userspace. | |
213 | * | |
214 | * More important is to keep inode and file references until userspace | |
215 | * has replied, otherwise FORGET and RELEASE could be sent while the | |
216 | * inode/file is still used by the filesystem. | |
217 | * | |
218 | * For this reason the concept of "background" request is introduced. | |
219 | * An interrupted request is backgrounded if it has been already sent | |
220 | * to userspace. Backgrounding involves getting an extra reference to | |
221 | * inode(s) or file used in the request, and adding the request to | |
222 | * fc->background list. When a reply is received for a background | |
223 | * request, the object references are released, and the request is | |
224 | * removed from the list. If the filesystem is unmounted while there | |
225 | * are still background requests, the list is walked and references | |
226 | * are released as if a reply was received. | |
227 | * | |
228 | * There's one more use for a background request. The RELEASE message is | |
229 | * always sent as background, since it doesn't return an error or | |
230 | * data. | |
231 | */ | |
232 | static void background_request(struct fuse_conn *fc, struct fuse_req *req) | |
334f485d | 233 | { |
334f485d | 234 | req->background = 1; |
1e9a4ed9 | 235 | list_add(&req->bg_entry, &fc->background); |
334f485d MS |
236 | if (req->inode) |
237 | req->inode = igrab(req->inode); | |
238 | if (req->inode2) | |
239 | req->inode2 = igrab(req->inode2); | |
240 | if (req->file) | |
241 | get_file(req->file); | |
242 | } | |
243 | ||
244 | static int request_wait_answer_nonint(struct fuse_req *req) | |
245 | { | |
246 | int err; | |
247 | sigset_t oldset; | |
248 | block_sigs(&oldset); | |
249 | err = wait_event_interruptible(req->waitq, req->finished); | |
250 | restore_sigs(&oldset); | |
251 | return err; | |
252 | } | |
253 | ||
254 | /* Called with fuse_lock held. Releases, and then reacquires it. */ | |
1e9a4ed9 MS |
255 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req, |
256 | int interruptible) | |
334f485d MS |
257 | { |
258 | int intr; | |
259 | ||
260 | spin_unlock(&fuse_lock); | |
261 | if (interruptible) | |
262 | intr = wait_event_interruptible(req->waitq, req->finished); | |
263 | else | |
264 | intr = request_wait_answer_nonint(req); | |
265 | spin_lock(&fuse_lock); | |
266 | if (intr && interruptible && req->sent) { | |
267 | /* If request is already in userspace, only allow KILL | |
268 | signal to interrupt */ | |
269 | spin_unlock(&fuse_lock); | |
270 | intr = request_wait_answer_nonint(req); | |
271 | spin_lock(&fuse_lock); | |
272 | } | |
273 | if (!intr) | |
274 | return; | |
275 | ||
276 | if (!interruptible || req->sent) | |
277 | req->out.h.error = -EINTR; | |
278 | else | |
279 | req->out.h.error = -ERESTARTNOINTR; | |
280 | ||
281 | req->interrupted = 1; | |
282 | if (req->locked) { | |
283 | /* This is uninterruptible sleep, because data is | |
284 | being copied to/from the buffers of req. During | |
285 | locked state, there mustn't be any filesystem | |
286 | operation (e.g. page fault), since that could lead | |
287 | to deadlock */ | |
288 | spin_unlock(&fuse_lock); | |
289 | wait_event(req->waitq, !req->locked); | |
290 | spin_lock(&fuse_lock); | |
291 | } | |
292 | if (!req->sent && !list_empty(&req->list)) { | |
293 | list_del(&req->list); | |
294 | __fuse_put_request(req); | |
295 | } else if (!req->finished && req->sent) | |
1e9a4ed9 | 296 | background_request(fc, req); |
334f485d MS |
297 | } |
298 | ||
299 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) | |
300 | { | |
301 | unsigned nbytes = 0; | |
302 | unsigned i; | |
303 | ||
304 | for (i = 0; i < numargs; i++) | |
305 | nbytes += args[i].size; | |
306 | ||
307 | return nbytes; | |
308 | } | |
309 | ||
310 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) | |
311 | { | |
312 | fc->reqctr++; | |
313 | /* zero is special */ | |
314 | if (fc->reqctr == 0) | |
315 | fc->reqctr = 1; | |
316 | req->in.h.unique = fc->reqctr; | |
317 | req->in.h.len = sizeof(struct fuse_in_header) + | |
318 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
319 | if (!req->preallocated) { | |
320 | /* If request is not preallocated (either FORGET or | |
321 | RELEASE), then still decrease outstanding_sem, so | |
322 | user can't open infinite number of files while not | |
323 | processing the RELEASE requests. However for | |
324 | efficiency do it without blocking, so if down() | |
325 | would block, just increase the debt instead */ | |
326 | if (down_trylock(&fc->outstanding_sem)) | |
327 | fc->outstanding_debt++; | |
328 | } | |
329 | list_add_tail(&req->list, &fc->pending); | |
330 | wake_up(&fc->waitq); | |
331 | } | |
332 | ||
333 | static void request_send_wait(struct fuse_conn *fc, struct fuse_req *req, | |
334 | int interruptible) | |
335 | { | |
336 | req->isreply = 1; | |
337 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 338 | if (!fc->connected) |
334f485d MS |
339 | req->out.h.error = -ENOTCONN; |
340 | else if (fc->conn_error) | |
341 | req->out.h.error = -ECONNREFUSED; | |
342 | else { | |
343 | queue_request(fc, req); | |
344 | /* acquire extra reference, since request is still needed | |
345 | after request_end() */ | |
346 | __fuse_get_request(req); | |
347 | ||
1e9a4ed9 | 348 | request_wait_answer(fc, req, interruptible); |
334f485d MS |
349 | } |
350 | spin_unlock(&fuse_lock); | |
351 | } | |
352 | ||
353 | void request_send(struct fuse_conn *fc, struct fuse_req *req) | |
354 | { | |
355 | request_send_wait(fc, req, 1); | |
356 | } | |
357 | ||
358 | /* | |
359 | * Non-interruptible version of the above function is for operations | |
360 | * which can't legally return -ERESTART{SYS,NOINTR}. This can still | |
361 | * be interrupted but only with SIGKILL. | |
362 | */ | |
363 | void request_send_nonint(struct fuse_conn *fc, struct fuse_req *req) | |
364 | { | |
365 | request_send_wait(fc, req, 0); | |
366 | } | |
367 | ||
368 | static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) | |
369 | { | |
370 | spin_lock(&fuse_lock); | |
1e9a4ed9 | 371 | if (fc->connected) { |
334f485d MS |
372 | queue_request(fc, req); |
373 | spin_unlock(&fuse_lock); | |
374 | } else { | |
375 | req->out.h.error = -ENOTCONN; | |
376 | request_end(fc, req); | |
377 | } | |
378 | } | |
379 | ||
380 | void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req) | |
381 | { | |
382 | req->isreply = 0; | |
383 | request_send_nowait(fc, req); | |
384 | } | |
385 | ||
386 | void request_send_background(struct fuse_conn *fc, struct fuse_req *req) | |
387 | { | |
388 | req->isreply = 1; | |
1e9a4ed9 MS |
389 | spin_lock(&fuse_lock); |
390 | background_request(fc, req); | |
391 | spin_unlock(&fuse_lock); | |
334f485d MS |
392 | request_send_nowait(fc, req); |
393 | } | |
394 | ||
395 | void fuse_send_init(struct fuse_conn *fc) | |
396 | { | |
397 | /* This is called from fuse_read_super() so there's guaranteed | |
398 | to be a request available */ | |
399 | struct fuse_req *req = do_get_request(fc); | |
400 | struct fuse_init_in_out *arg = &req->misc.init_in_out; | |
401 | arg->major = FUSE_KERNEL_VERSION; | |
402 | arg->minor = FUSE_KERNEL_MINOR_VERSION; | |
403 | req->in.h.opcode = FUSE_INIT; | |
404 | req->in.numargs = 1; | |
405 | req->in.args[0].size = sizeof(*arg); | |
406 | req->in.args[0].value = arg; | |
407 | req->out.numargs = 1; | |
408 | req->out.args[0].size = sizeof(*arg); | |
409 | req->out.args[0].value = arg; | |
410 | request_send_background(fc, req); | |
411 | } | |
412 | ||
413 | /* | |
414 | * Lock the request. Up to the next unlock_request() there mustn't be | |
415 | * anything that could cause a page-fault. If the request was already | |
416 | * interrupted bail out. | |
417 | */ | |
418 | static inline int lock_request(struct fuse_req *req) | |
419 | { | |
420 | int err = 0; | |
421 | if (req) { | |
422 | spin_lock(&fuse_lock); | |
423 | if (req->interrupted) | |
424 | err = -ENOENT; | |
425 | else | |
426 | req->locked = 1; | |
427 | spin_unlock(&fuse_lock); | |
428 | } | |
429 | return err; | |
430 | } | |
431 | ||
432 | /* | |
433 | * Unlock request. If it was interrupted during being locked, the | |
434 | * requester thread is currently waiting for it to be unlocked, so | |
435 | * wake it up. | |
436 | */ | |
437 | static inline void unlock_request(struct fuse_req *req) | |
438 | { | |
439 | if (req) { | |
440 | spin_lock(&fuse_lock); | |
441 | req->locked = 0; | |
442 | if (req->interrupted) | |
443 | wake_up(&req->waitq); | |
444 | spin_unlock(&fuse_lock); | |
445 | } | |
446 | } | |
447 | ||
448 | struct fuse_copy_state { | |
449 | int write; | |
450 | struct fuse_req *req; | |
451 | const struct iovec *iov; | |
452 | unsigned long nr_segs; | |
453 | unsigned long seglen; | |
454 | unsigned long addr; | |
455 | struct page *pg; | |
456 | void *mapaddr; | |
457 | void *buf; | |
458 | unsigned len; | |
459 | }; | |
460 | ||
461 | static void fuse_copy_init(struct fuse_copy_state *cs, int write, | |
462 | struct fuse_req *req, const struct iovec *iov, | |
463 | unsigned long nr_segs) | |
464 | { | |
465 | memset(cs, 0, sizeof(*cs)); | |
466 | cs->write = write; | |
467 | cs->req = req; | |
468 | cs->iov = iov; | |
469 | cs->nr_segs = nr_segs; | |
470 | } | |
471 | ||
472 | /* Unmap and put previous page of userspace buffer */ | |
473 | static inline void fuse_copy_finish(struct fuse_copy_state *cs) | |
474 | { | |
475 | if (cs->mapaddr) { | |
476 | kunmap_atomic(cs->mapaddr, KM_USER0); | |
477 | if (cs->write) { | |
478 | flush_dcache_page(cs->pg); | |
479 | set_page_dirty_lock(cs->pg); | |
480 | } | |
481 | put_page(cs->pg); | |
482 | cs->mapaddr = NULL; | |
483 | } | |
484 | } | |
485 | ||
486 | /* | |
487 | * Get another pagefull of userspace buffer, and map it to kernel | |
488 | * address space, and lock request | |
489 | */ | |
490 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
491 | { | |
492 | unsigned long offset; | |
493 | int err; | |
494 | ||
495 | unlock_request(cs->req); | |
496 | fuse_copy_finish(cs); | |
497 | if (!cs->seglen) { | |
498 | BUG_ON(!cs->nr_segs); | |
499 | cs->seglen = cs->iov[0].iov_len; | |
500 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
501 | cs->iov ++; | |
502 | cs->nr_segs --; | |
503 | } | |
504 | down_read(¤t->mm->mmap_sem); | |
505 | err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0, | |
506 | &cs->pg, NULL); | |
507 | up_read(¤t->mm->mmap_sem); | |
508 | if (err < 0) | |
509 | return err; | |
510 | BUG_ON(err != 1); | |
511 | offset = cs->addr % PAGE_SIZE; | |
512 | cs->mapaddr = kmap_atomic(cs->pg, KM_USER0); | |
513 | cs->buf = cs->mapaddr + offset; | |
514 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
515 | cs->seglen -= cs->len; | |
516 | cs->addr += cs->len; | |
517 | ||
518 | return lock_request(cs->req); | |
519 | } | |
520 | ||
521 | /* Do as much copy to/from userspace buffer as we can */ | |
522 | static inline int fuse_copy_do(struct fuse_copy_state *cs, void **val, | |
523 | unsigned *size) | |
524 | { | |
525 | unsigned ncpy = min(*size, cs->len); | |
526 | if (val) { | |
527 | if (cs->write) | |
528 | memcpy(cs->buf, *val, ncpy); | |
529 | else | |
530 | memcpy(*val, cs->buf, ncpy); | |
531 | *val += ncpy; | |
532 | } | |
533 | *size -= ncpy; | |
534 | cs->len -= ncpy; | |
535 | cs->buf += ncpy; | |
536 | return ncpy; | |
537 | } | |
538 | ||
539 | /* | |
540 | * Copy a page in the request to/from the userspace buffer. Must be | |
541 | * done atomically | |
542 | */ | |
543 | static inline int fuse_copy_page(struct fuse_copy_state *cs, struct page *page, | |
544 | unsigned offset, unsigned count, int zeroing) | |
545 | { | |
546 | if (page && zeroing && count < PAGE_SIZE) { | |
547 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
548 | memset(mapaddr, 0, PAGE_SIZE); | |
549 | kunmap_atomic(mapaddr, KM_USER1); | |
550 | } | |
551 | while (count) { | |
552 | int err; | |
553 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
554 | return err; | |
555 | if (page) { | |
556 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
557 | void *buf = mapaddr + offset; | |
558 | offset += fuse_copy_do(cs, &buf, &count); | |
559 | kunmap_atomic(mapaddr, KM_USER1); | |
560 | } else | |
561 | offset += fuse_copy_do(cs, NULL, &count); | |
562 | } | |
563 | if (page && !cs->write) | |
564 | flush_dcache_page(page); | |
565 | return 0; | |
566 | } | |
567 | ||
568 | /* Copy pages in the request to/from userspace buffer */ | |
569 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
570 | int zeroing) | |
571 | { | |
572 | unsigned i; | |
573 | struct fuse_req *req = cs->req; | |
574 | unsigned offset = req->page_offset; | |
575 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
576 | ||
577 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
578 | struct page *page = req->pages[i]; | |
579 | int err = fuse_copy_page(cs, page, offset, count, zeroing); | |
580 | if (err) | |
581 | return err; | |
582 | ||
583 | nbytes -= count; | |
584 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
585 | offset = 0; | |
586 | } | |
587 | return 0; | |
588 | } | |
589 | ||
590 | /* Copy a single argument in the request to/from userspace buffer */ | |
591 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
592 | { | |
593 | while (size) { | |
594 | int err; | |
595 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
596 | return err; | |
597 | fuse_copy_do(cs, &val, &size); | |
598 | } | |
599 | return 0; | |
600 | } | |
601 | ||
602 | /* Copy request arguments to/from userspace buffer */ | |
603 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
604 | unsigned argpages, struct fuse_arg *args, | |
605 | int zeroing) | |
606 | { | |
607 | int err = 0; | |
608 | unsigned i; | |
609 | ||
610 | for (i = 0; !err && i < numargs; i++) { | |
611 | struct fuse_arg *arg = &args[i]; | |
612 | if (i == numargs - 1 && argpages) | |
613 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
614 | else | |
615 | err = fuse_copy_one(cs, arg->value, arg->size); | |
616 | } | |
617 | return err; | |
618 | } | |
619 | ||
620 | /* Wait until a request is available on the pending list */ | |
621 | static void request_wait(struct fuse_conn *fc) | |
622 | { | |
623 | DECLARE_WAITQUEUE(wait, current); | |
624 | ||
625 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
1e9a4ed9 | 626 | while (fc->mounted && list_empty(&fc->pending)) { |
334f485d MS |
627 | set_current_state(TASK_INTERRUPTIBLE); |
628 | if (signal_pending(current)) | |
629 | break; | |
630 | ||
631 | spin_unlock(&fuse_lock); | |
632 | schedule(); | |
633 | spin_lock(&fuse_lock); | |
634 | } | |
635 | set_current_state(TASK_RUNNING); | |
636 | remove_wait_queue(&fc->waitq, &wait); | |
637 | } | |
638 | ||
639 | /* | |
640 | * Read a single request into the userspace filesystem's buffer. This | |
641 | * function waits until a request is available, then removes it from | |
642 | * the pending list and copies request data to userspace buffer. If | |
643 | * no reply is needed (FORGET) or request has been interrupted or | |
644 | * there was an error during the copying then it's finished by calling | |
645 | * request_end(). Otherwise add it to the processing list, and set | |
646 | * the 'sent' flag. | |
647 | */ | |
648 | static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov, | |
649 | unsigned long nr_segs, loff_t *off) | |
650 | { | |
651 | int err; | |
652 | struct fuse_conn *fc; | |
653 | struct fuse_req *req; | |
654 | struct fuse_in *in; | |
655 | struct fuse_copy_state cs; | |
656 | unsigned reqsize; | |
657 | ||
658 | spin_lock(&fuse_lock); | |
659 | fc = file->private_data; | |
660 | err = -EPERM; | |
661 | if (!fc) | |
662 | goto err_unlock; | |
663 | request_wait(fc); | |
664 | err = -ENODEV; | |
1e9a4ed9 | 665 | if (!fc->mounted) |
334f485d MS |
666 | goto err_unlock; |
667 | err = -ERESTARTSYS; | |
668 | if (list_empty(&fc->pending)) | |
669 | goto err_unlock; | |
670 | ||
671 | req = list_entry(fc->pending.next, struct fuse_req, list); | |
672 | list_del_init(&req->list); | |
673 | spin_unlock(&fuse_lock); | |
674 | ||
675 | in = &req->in; | |
676 | reqsize = req->in.h.len; | |
677 | fuse_copy_init(&cs, 1, req, iov, nr_segs); | |
678 | err = -EINVAL; | |
679 | if (iov_length(iov, nr_segs) >= reqsize) { | |
680 | err = fuse_copy_one(&cs, &in->h, sizeof(in->h)); | |
681 | if (!err) | |
682 | err = fuse_copy_args(&cs, in->numargs, in->argpages, | |
683 | (struct fuse_arg *) in->args, 0); | |
684 | } | |
685 | fuse_copy_finish(&cs); | |
686 | ||
687 | spin_lock(&fuse_lock); | |
688 | req->locked = 0; | |
689 | if (!err && req->interrupted) | |
690 | err = -ENOENT; | |
691 | if (err) { | |
692 | if (!req->interrupted) | |
693 | req->out.h.error = -EIO; | |
694 | request_end(fc, req); | |
695 | return err; | |
696 | } | |
697 | if (!req->isreply) | |
698 | request_end(fc, req); | |
699 | else { | |
700 | req->sent = 1; | |
701 | list_add_tail(&req->list, &fc->processing); | |
702 | spin_unlock(&fuse_lock); | |
703 | } | |
704 | return reqsize; | |
705 | ||
706 | err_unlock: | |
707 | spin_unlock(&fuse_lock); | |
708 | return err; | |
709 | } | |
710 | ||
711 | static ssize_t fuse_dev_read(struct file *file, char __user *buf, | |
712 | size_t nbytes, loff_t *off) | |
713 | { | |
714 | struct iovec iov; | |
715 | iov.iov_len = nbytes; | |
716 | iov.iov_base = buf; | |
717 | return fuse_dev_readv(file, &iov, 1, off); | |
718 | } | |
719 | ||
720 | /* Look up request on processing list by unique ID */ | |
721 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
722 | { | |
723 | struct list_head *entry; | |
724 | ||
725 | list_for_each(entry, &fc->processing) { | |
726 | struct fuse_req *req; | |
727 | req = list_entry(entry, struct fuse_req, list); | |
728 | if (req->in.h.unique == unique) | |
729 | return req; | |
730 | } | |
731 | return NULL; | |
732 | } | |
733 | ||
e5e5558e MS |
734 | /* fget() needs to be done in this context */ |
735 | static void process_getdir(struct fuse_req *req) | |
736 | { | |
737 | struct fuse_getdir_out_i *arg = req->out.args[0].value; | |
738 | arg->file = fget(arg->fd); | |
739 | } | |
740 | ||
334f485d MS |
741 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, |
742 | unsigned nbytes) | |
743 | { | |
744 | unsigned reqsize = sizeof(struct fuse_out_header); | |
745 | ||
746 | if (out->h.error) | |
747 | return nbytes != reqsize ? -EINVAL : 0; | |
748 | ||
749 | reqsize += len_args(out->numargs, out->args); | |
750 | ||
751 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
752 | return -EINVAL; | |
753 | else if (reqsize > nbytes) { | |
754 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
755 | unsigned diffsize = reqsize - nbytes; | |
756 | if (diffsize > lastarg->size) | |
757 | return -EINVAL; | |
758 | lastarg->size -= diffsize; | |
759 | } | |
760 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
761 | out->page_zeroing); | |
762 | } | |
763 | ||
764 | /* | |
765 | * Write a single reply to a request. First the header is copied from | |
766 | * the write buffer. The request is then searched on the processing | |
767 | * list by the unique ID found in the header. If found, then remove | |
768 | * it from the list and copy the rest of the buffer to the request. | |
769 | * The request is finished by calling request_end() | |
770 | */ | |
771 | static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov, | |
772 | unsigned long nr_segs, loff_t *off) | |
773 | { | |
774 | int err; | |
775 | unsigned nbytes = iov_length(iov, nr_segs); | |
776 | struct fuse_req *req; | |
777 | struct fuse_out_header oh; | |
778 | struct fuse_copy_state cs; | |
779 | struct fuse_conn *fc = fuse_get_conn(file); | |
780 | if (!fc) | |
781 | return -ENODEV; | |
782 | ||
783 | fuse_copy_init(&cs, 0, NULL, iov, nr_segs); | |
784 | if (nbytes < sizeof(struct fuse_out_header)) | |
785 | return -EINVAL; | |
786 | ||
787 | err = fuse_copy_one(&cs, &oh, sizeof(oh)); | |
788 | if (err) | |
789 | goto err_finish; | |
790 | err = -EINVAL; | |
791 | if (!oh.unique || oh.error <= -1000 || oh.error > 0 || | |
792 | oh.len != nbytes) | |
793 | goto err_finish; | |
794 | ||
795 | spin_lock(&fuse_lock); | |
796 | req = request_find(fc, oh.unique); | |
797 | err = -EINVAL; | |
798 | if (!req) | |
799 | goto err_unlock; | |
800 | ||
801 | list_del_init(&req->list); | |
802 | if (req->interrupted) { | |
803 | request_end(fc, req); | |
804 | fuse_copy_finish(&cs); | |
805 | return -ENOENT; | |
806 | } | |
807 | req->out.h = oh; | |
808 | req->locked = 1; | |
809 | cs.req = req; | |
810 | spin_unlock(&fuse_lock); | |
811 | ||
812 | err = copy_out_args(&cs, &req->out, nbytes); | |
813 | fuse_copy_finish(&cs); | |
814 | ||
815 | spin_lock(&fuse_lock); | |
816 | req->locked = 0; | |
817 | if (!err) { | |
818 | if (req->interrupted) | |
819 | err = -ENOENT; | |
e5e5558e MS |
820 | else if (req->in.h.opcode == FUSE_GETDIR && !oh.error) |
821 | process_getdir(req); | |
334f485d MS |
822 | } else if (!req->interrupted) |
823 | req->out.h.error = -EIO; | |
824 | request_end(fc, req); | |
825 | ||
826 | return err ? err : nbytes; | |
827 | ||
828 | err_unlock: | |
829 | spin_unlock(&fuse_lock); | |
830 | err_finish: | |
831 | fuse_copy_finish(&cs); | |
832 | return err; | |
833 | } | |
834 | ||
835 | static ssize_t fuse_dev_write(struct file *file, const char __user *buf, | |
836 | size_t nbytes, loff_t *off) | |
837 | { | |
838 | struct iovec iov; | |
839 | iov.iov_len = nbytes; | |
840 | iov.iov_base = (char __user *) buf; | |
841 | return fuse_dev_writev(file, &iov, 1, off); | |
842 | } | |
843 | ||
844 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) | |
845 | { | |
846 | struct fuse_conn *fc = fuse_get_conn(file); | |
847 | unsigned mask = POLLOUT | POLLWRNORM; | |
848 | ||
849 | if (!fc) | |
850 | return -ENODEV; | |
851 | ||
852 | poll_wait(file, &fc->waitq, wait); | |
853 | ||
854 | spin_lock(&fuse_lock); | |
855 | if (!list_empty(&fc->pending)) | |
856 | mask |= POLLIN | POLLRDNORM; | |
857 | spin_unlock(&fuse_lock); | |
858 | ||
859 | return mask; | |
860 | } | |
861 | ||
862 | /* Abort all requests on the given list (pending or processing) */ | |
863 | static void end_requests(struct fuse_conn *fc, struct list_head *head) | |
864 | { | |
865 | while (!list_empty(head)) { | |
866 | struct fuse_req *req; | |
867 | req = list_entry(head->next, struct fuse_req, list); | |
868 | list_del_init(&req->list); | |
869 | req->out.h.error = -ECONNABORTED; | |
870 | request_end(fc, req); | |
871 | spin_lock(&fuse_lock); | |
872 | } | |
873 | } | |
874 | ||
875 | static int fuse_dev_release(struct inode *inode, struct file *file) | |
876 | { | |
877 | struct fuse_conn *fc; | |
878 | ||
879 | spin_lock(&fuse_lock); | |
880 | fc = file->private_data; | |
881 | if (fc) { | |
1e9a4ed9 | 882 | fc->connected = 0; |
334f485d MS |
883 | end_requests(fc, &fc->pending); |
884 | end_requests(fc, &fc->processing); | |
885 | fuse_release_conn(fc); | |
886 | } | |
887 | spin_unlock(&fuse_lock); | |
888 | return 0; | |
889 | } | |
890 | ||
891 | struct file_operations fuse_dev_operations = { | |
892 | .owner = THIS_MODULE, | |
893 | .llseek = no_llseek, | |
894 | .read = fuse_dev_read, | |
895 | .readv = fuse_dev_readv, | |
896 | .write = fuse_dev_write, | |
897 | .writev = fuse_dev_writev, | |
898 | .poll = fuse_dev_poll, | |
899 | .release = fuse_dev_release, | |
900 | }; | |
901 | ||
902 | static struct miscdevice fuse_miscdevice = { | |
903 | .minor = FUSE_MINOR, | |
904 | .name = "fuse", | |
905 | .fops = &fuse_dev_operations, | |
906 | }; | |
907 | ||
908 | int __init fuse_dev_init(void) | |
909 | { | |
910 | int err = -ENOMEM; | |
911 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
912 | sizeof(struct fuse_req), | |
913 | 0, 0, NULL, NULL); | |
914 | if (!fuse_req_cachep) | |
915 | goto out; | |
916 | ||
917 | err = misc_register(&fuse_miscdevice); | |
918 | if (err) | |
919 | goto out_cache_clean; | |
920 | ||
921 | return 0; | |
922 | ||
923 | out_cache_clean: | |
924 | kmem_cache_destroy(fuse_req_cachep); | |
925 | out: | |
926 | return err; | |
927 | } | |
928 | ||
929 | void fuse_dev_cleanup(void) | |
930 | { | |
931 | misc_deregister(&fuse_miscdevice); | |
932 | kmem_cache_destroy(fuse_req_cachep); | |
933 | } |