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