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 | ||
e18b890b | 22 | static struct kmem_cache *fuse_req_cachep; |
334f485d | 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); | |
a4d27e75 | 37 | INIT_LIST_HEAD(&req->intr_entry); |
334f485d MS |
38 | init_waitqueue_head(&req->waitq); |
39 | atomic_set(&req->count, 1); | |
40 | } | |
41 | ||
42 | struct fuse_req *fuse_request_alloc(void) | |
43 | { | |
e94b1766 | 44 | struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL); |
334f485d MS |
45 | if (req) |
46 | fuse_request_init(req); | |
47 | return req; | |
48 | } | |
49 | ||
50 | void fuse_request_free(struct fuse_req *req) | |
51 | { | |
52 | kmem_cache_free(fuse_req_cachep, req); | |
53 | } | |
54 | ||
8bfc016d | 55 | static void block_sigs(sigset_t *oldset) |
334f485d MS |
56 | { |
57 | sigset_t mask; | |
58 | ||
59 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
60 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
61 | } | |
62 | ||
8bfc016d | 63 | static void restore_sigs(sigset_t *oldset) |
334f485d MS |
64 | { |
65 | sigprocmask(SIG_SETMASK, oldset, NULL); | |
66 | } | |
67 | ||
334f485d MS |
68 | static void __fuse_get_request(struct fuse_req *req) |
69 | { | |
70 | atomic_inc(&req->count); | |
71 | } | |
72 | ||
73 | /* Must be called with > 1 refcount */ | |
74 | static void __fuse_put_request(struct fuse_req *req) | |
75 | { | |
76 | BUG_ON(atomic_read(&req->count) < 2); | |
77 | atomic_dec(&req->count); | |
78 | } | |
79 | ||
33649c91 MS |
80 | static void fuse_req_init_context(struct fuse_req *req) |
81 | { | |
82 | req->in.h.uid = current->fsuid; | |
83 | req->in.h.gid = current->fsgid; | |
84 | req->in.h.pid = current->pid; | |
85 | } | |
86 | ||
ce1d5a49 | 87 | struct fuse_req *fuse_get_req(struct fuse_conn *fc) |
334f485d | 88 | { |
08a53cdc MS |
89 | struct fuse_req *req; |
90 | sigset_t oldset; | |
9bc5ddda | 91 | int intr; |
08a53cdc MS |
92 | int err; |
93 | ||
9bc5ddda | 94 | atomic_inc(&fc->num_waiting); |
08a53cdc | 95 | block_sigs(&oldset); |
9bc5ddda | 96 | intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked); |
08a53cdc | 97 | restore_sigs(&oldset); |
9bc5ddda MS |
98 | err = -EINTR; |
99 | if (intr) | |
100 | goto out; | |
08a53cdc | 101 | |
51eb01e7 MS |
102 | err = -ENOTCONN; |
103 | if (!fc->connected) | |
104 | goto out; | |
105 | ||
08a53cdc | 106 | req = fuse_request_alloc(); |
9bc5ddda | 107 | err = -ENOMEM; |
ce1d5a49 | 108 | if (!req) |
9bc5ddda | 109 | goto out; |
334f485d | 110 | |
33649c91 | 111 | fuse_req_init_context(req); |
9bc5ddda | 112 | req->waiting = 1; |
334f485d | 113 | return req; |
9bc5ddda MS |
114 | |
115 | out: | |
116 | atomic_dec(&fc->num_waiting); | |
117 | return ERR_PTR(err); | |
334f485d MS |
118 | } |
119 | ||
33649c91 MS |
120 | /* |
121 | * Return request in fuse_file->reserved_req. However that may | |
122 | * currently be in use. If that is the case, wait for it to become | |
123 | * available. | |
124 | */ | |
125 | static struct fuse_req *get_reserved_req(struct fuse_conn *fc, | |
126 | struct file *file) | |
127 | { | |
128 | struct fuse_req *req = NULL; | |
129 | struct fuse_file *ff = file->private_data; | |
130 | ||
131 | do { | |
de5e3dec | 132 | wait_event(fc->reserved_req_waitq, ff->reserved_req); |
33649c91 MS |
133 | spin_lock(&fc->lock); |
134 | if (ff->reserved_req) { | |
135 | req = ff->reserved_req; | |
136 | ff->reserved_req = NULL; | |
137 | get_file(file); | |
138 | req->stolen_file = file; | |
139 | } | |
140 | spin_unlock(&fc->lock); | |
141 | } while (!req); | |
142 | ||
143 | return req; | |
144 | } | |
145 | ||
146 | /* | |
147 | * Put stolen request back into fuse_file->reserved_req | |
148 | */ | |
149 | static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req) | |
150 | { | |
151 | struct file *file = req->stolen_file; | |
152 | struct fuse_file *ff = file->private_data; | |
153 | ||
154 | spin_lock(&fc->lock); | |
155 | fuse_request_init(req); | |
156 | BUG_ON(ff->reserved_req); | |
157 | ff->reserved_req = req; | |
de5e3dec | 158 | wake_up_all(&fc->reserved_req_waitq); |
33649c91 MS |
159 | spin_unlock(&fc->lock); |
160 | fput(file); | |
161 | } | |
162 | ||
163 | /* | |
164 | * Gets a requests for a file operation, always succeeds | |
165 | * | |
166 | * This is used for sending the FLUSH request, which must get to | |
167 | * userspace, due to POSIX locks which may need to be unlocked. | |
168 | * | |
169 | * If allocation fails due to OOM, use the reserved request in | |
170 | * fuse_file. | |
171 | * | |
172 | * This is very unlikely to deadlock accidentally, since the | |
173 | * filesystem should not have it's own file open. If deadlock is | |
174 | * intentional, it can still be broken by "aborting" the filesystem. | |
175 | */ | |
176 | struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file) | |
177 | { | |
178 | struct fuse_req *req; | |
179 | ||
180 | atomic_inc(&fc->num_waiting); | |
181 | wait_event(fc->blocked_waitq, !fc->blocked); | |
182 | req = fuse_request_alloc(); | |
183 | if (!req) | |
184 | req = get_reserved_req(fc, file); | |
185 | ||
186 | fuse_req_init_context(req); | |
187 | req->waiting = 1; | |
188 | return req; | |
189 | } | |
190 | ||
334f485d | 191 | void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req) |
7128ec2a MS |
192 | { |
193 | if (atomic_dec_and_test(&req->count)) { | |
9bc5ddda MS |
194 | if (req->waiting) |
195 | atomic_dec(&fc->num_waiting); | |
33649c91 MS |
196 | |
197 | if (req->stolen_file) | |
198 | put_reserved_req(fc, req); | |
199 | else | |
200 | fuse_request_free(req); | |
7128ec2a MS |
201 | } |
202 | } | |
203 | ||
334f485d MS |
204 | /* |
205 | * This function is called when a request is finished. Either a reply | |
f9a2842e | 206 | * has arrived or it was aborted (and not yet sent) or some error |
f43b155a | 207 | * occurred during communication with userspace, or the device file |
51eb01e7 MS |
208 | * was closed. The requester thread is woken up (if still waiting), |
209 | * the 'end' callback is called if given, else the reference to the | |
210 | * request is released | |
7128ec2a | 211 | * |
d7133114 | 212 | * Called with fc->lock, unlocks it |
334f485d MS |
213 | */ |
214 | static void request_end(struct fuse_conn *fc, struct fuse_req *req) | |
105f4d7a | 215 | __releases(fc->lock) |
334f485d | 216 | { |
51eb01e7 MS |
217 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; |
218 | req->end = NULL; | |
d77a1d5b | 219 | list_del(&req->list); |
a4d27e75 | 220 | list_del(&req->intr_entry); |
83cfd493 | 221 | req->state = FUSE_REQ_FINISHED; |
51eb01e7 MS |
222 | if (req->background) { |
223 | if (fc->num_background == FUSE_MAX_BACKGROUND) { | |
224 | fc->blocked = 0; | |
225 | wake_up_all(&fc->blocked_waitq); | |
226 | } | |
f92b99b9 MS |
227 | if (fc->num_background == FUSE_CONGESTION_THRESHOLD) { |
228 | clear_bdi_congested(&fc->bdi, READ); | |
229 | clear_bdi_congested(&fc->bdi, WRITE); | |
230 | } | |
51eb01e7 | 231 | fc->num_background--; |
334f485d | 232 | } |
51eb01e7 MS |
233 | spin_unlock(&fc->lock); |
234 | dput(req->dentry); | |
235 | mntput(req->vfsmount); | |
334f485d | 236 | if (req->file) |
51eb01e7 MS |
237 | fput(req->file); |
238 | wake_up(&req->waitq); | |
239 | if (end) | |
240 | end(fc, req); | |
241 | else | |
242 | fuse_put_request(fc, req); | |
334f485d MS |
243 | } |
244 | ||
a4d27e75 MS |
245 | static void wait_answer_interruptible(struct fuse_conn *fc, |
246 | struct fuse_req *req) | |
247 | { | |
248 | if (signal_pending(current)) | |
249 | return; | |
250 | ||
251 | spin_unlock(&fc->lock); | |
252 | wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED); | |
253 | spin_lock(&fc->lock); | |
254 | } | |
255 | ||
256 | static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req) | |
257 | { | |
258 | list_add_tail(&req->intr_entry, &fc->interrupts); | |
259 | wake_up(&fc->waitq); | |
260 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); | |
261 | } | |
262 | ||
d7133114 | 263 | /* Called with fc->lock held. Releases, and then reacquires it. */ |
7c352bdf | 264 | static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req) |
334f485d | 265 | { |
a4d27e75 MS |
266 | if (!fc->no_interrupt) { |
267 | /* Any signal may interrupt this */ | |
268 | wait_answer_interruptible(fc, req); | |
334f485d | 269 | |
a4d27e75 MS |
270 | if (req->aborted) |
271 | goto aborted; | |
272 | if (req->state == FUSE_REQ_FINISHED) | |
273 | return; | |
274 | ||
275 | req->interrupted = 1; | |
276 | if (req->state == FUSE_REQ_SENT) | |
277 | queue_interrupt(fc, req); | |
278 | } | |
279 | ||
280 | if (req->force) { | |
281 | spin_unlock(&fc->lock); | |
51eb01e7 | 282 | wait_event(req->waitq, req->state == FUSE_REQ_FINISHED); |
a4d27e75 MS |
283 | spin_lock(&fc->lock); |
284 | } else { | |
285 | sigset_t oldset; | |
286 | ||
287 | /* Only fatal signals may interrupt this */ | |
51eb01e7 | 288 | block_sigs(&oldset); |
a4d27e75 | 289 | wait_answer_interruptible(fc, req); |
51eb01e7 MS |
290 | restore_sigs(&oldset); |
291 | } | |
334f485d | 292 | |
a4d27e75 MS |
293 | if (req->aborted) |
294 | goto aborted; | |
295 | if (req->state == FUSE_REQ_FINISHED) | |
296 | return; | |
297 | ||
298 | req->out.h.error = -EINTR; | |
299 | req->aborted = 1; | |
300 | ||
301 | aborted: | |
334f485d MS |
302 | if (req->locked) { |
303 | /* This is uninterruptible sleep, because data is | |
304 | being copied to/from the buffers of req. During | |
305 | locked state, there mustn't be any filesystem | |
306 | operation (e.g. page fault), since that could lead | |
307 | to deadlock */ | |
d7133114 | 308 | spin_unlock(&fc->lock); |
334f485d | 309 | wait_event(req->waitq, !req->locked); |
d7133114 | 310 | spin_lock(&fc->lock); |
334f485d | 311 | } |
83cfd493 | 312 | if (req->state == FUSE_REQ_PENDING) { |
334f485d MS |
313 | list_del(&req->list); |
314 | __fuse_put_request(req); | |
51eb01e7 MS |
315 | } else if (req->state == FUSE_REQ_SENT) { |
316 | spin_unlock(&fc->lock); | |
317 | wait_event(req->waitq, req->state == FUSE_REQ_FINISHED); | |
318 | spin_lock(&fc->lock); | |
319 | } | |
334f485d MS |
320 | } |
321 | ||
322 | static unsigned len_args(unsigned numargs, struct fuse_arg *args) | |
323 | { | |
324 | unsigned nbytes = 0; | |
325 | unsigned i; | |
326 | ||
327 | for (i = 0; i < numargs; i++) | |
328 | nbytes += args[i].size; | |
329 | ||
330 | return nbytes; | |
331 | } | |
332 | ||
a4d27e75 MS |
333 | static u64 fuse_get_unique(struct fuse_conn *fc) |
334 | { | |
335 | fc->reqctr++; | |
336 | /* zero is special */ | |
337 | if (fc->reqctr == 0) | |
338 | fc->reqctr = 1; | |
339 | ||
340 | return fc->reqctr; | |
341 | } | |
342 | ||
334f485d MS |
343 | static void queue_request(struct fuse_conn *fc, struct fuse_req *req) |
344 | { | |
a4d27e75 | 345 | req->in.h.unique = fuse_get_unique(fc); |
334f485d MS |
346 | req->in.h.len = sizeof(struct fuse_in_header) + |
347 | len_args(req->in.numargs, (struct fuse_arg *) req->in.args); | |
334f485d | 348 | list_add_tail(&req->list, &fc->pending); |
83cfd493 | 349 | req->state = FUSE_REQ_PENDING; |
9bc5ddda MS |
350 | if (!req->waiting) { |
351 | req->waiting = 1; | |
352 | atomic_inc(&fc->num_waiting); | |
353 | } | |
334f485d | 354 | wake_up(&fc->waitq); |
385a17bf | 355 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
334f485d MS |
356 | } |
357 | ||
7c352bdf | 358 | void request_send(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
359 | { |
360 | req->isreply = 1; | |
d7133114 | 361 | spin_lock(&fc->lock); |
1e9a4ed9 | 362 | if (!fc->connected) |
334f485d MS |
363 | req->out.h.error = -ENOTCONN; |
364 | else if (fc->conn_error) | |
365 | req->out.h.error = -ECONNREFUSED; | |
366 | else { | |
367 | queue_request(fc, req); | |
368 | /* acquire extra reference, since request is still needed | |
369 | after request_end() */ | |
370 | __fuse_get_request(req); | |
371 | ||
7c352bdf | 372 | request_wait_answer(fc, req); |
334f485d | 373 | } |
d7133114 | 374 | spin_unlock(&fc->lock); |
334f485d MS |
375 | } |
376 | ||
334f485d MS |
377 | static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) |
378 | { | |
d7133114 | 379 | spin_lock(&fc->lock); |
1e9a4ed9 | 380 | if (fc->connected) { |
51eb01e7 MS |
381 | req->background = 1; |
382 | fc->num_background++; | |
383 | if (fc->num_background == FUSE_MAX_BACKGROUND) | |
384 | fc->blocked = 1; | |
f92b99b9 MS |
385 | if (fc->num_background == FUSE_CONGESTION_THRESHOLD) { |
386 | set_bdi_congested(&fc->bdi, READ); | |
387 | set_bdi_congested(&fc->bdi, WRITE); | |
388 | } | |
51eb01e7 | 389 | |
334f485d | 390 | queue_request(fc, req); |
d7133114 | 391 | spin_unlock(&fc->lock); |
334f485d MS |
392 | } else { |
393 | req->out.h.error = -ENOTCONN; | |
394 | request_end(fc, req); | |
395 | } | |
396 | } | |
397 | ||
398 | void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req) | |
399 | { | |
400 | req->isreply = 0; | |
401 | request_send_nowait(fc, req); | |
402 | } | |
403 | ||
404 | void request_send_background(struct fuse_conn *fc, struct fuse_req *req) | |
405 | { | |
406 | req->isreply = 1; | |
334f485d MS |
407 | request_send_nowait(fc, req); |
408 | } | |
409 | ||
334f485d MS |
410 | /* |
411 | * Lock the request. Up to the next unlock_request() there mustn't be | |
412 | * anything that could cause a page-fault. If the request was already | |
f9a2842e | 413 | * aborted bail out. |
334f485d | 414 | */ |
d7133114 | 415 | static int lock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
416 | { |
417 | int err = 0; | |
418 | if (req) { | |
d7133114 | 419 | spin_lock(&fc->lock); |
f9a2842e | 420 | if (req->aborted) |
334f485d MS |
421 | err = -ENOENT; |
422 | else | |
423 | req->locked = 1; | |
d7133114 | 424 | spin_unlock(&fc->lock); |
334f485d MS |
425 | } |
426 | return err; | |
427 | } | |
428 | ||
429 | /* | |
f9a2842e | 430 | * Unlock request. If it was aborted during being locked, the |
334f485d MS |
431 | * requester thread is currently waiting for it to be unlocked, so |
432 | * wake it up. | |
433 | */ | |
d7133114 | 434 | static void unlock_request(struct fuse_conn *fc, struct fuse_req *req) |
334f485d MS |
435 | { |
436 | if (req) { | |
d7133114 | 437 | spin_lock(&fc->lock); |
334f485d | 438 | req->locked = 0; |
f9a2842e | 439 | if (req->aborted) |
334f485d | 440 | wake_up(&req->waitq); |
d7133114 | 441 | spin_unlock(&fc->lock); |
334f485d MS |
442 | } |
443 | } | |
444 | ||
445 | struct fuse_copy_state { | |
d7133114 | 446 | struct fuse_conn *fc; |
334f485d MS |
447 | int write; |
448 | struct fuse_req *req; | |
449 | const struct iovec *iov; | |
450 | unsigned long nr_segs; | |
451 | unsigned long seglen; | |
452 | unsigned long addr; | |
453 | struct page *pg; | |
454 | void *mapaddr; | |
455 | void *buf; | |
456 | unsigned len; | |
457 | }; | |
458 | ||
d7133114 MS |
459 | static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc, |
460 | int write, struct fuse_req *req, | |
461 | const struct iovec *iov, unsigned long nr_segs) | |
334f485d MS |
462 | { |
463 | memset(cs, 0, sizeof(*cs)); | |
d7133114 | 464 | cs->fc = fc; |
334f485d MS |
465 | cs->write = write; |
466 | cs->req = req; | |
467 | cs->iov = iov; | |
468 | cs->nr_segs = nr_segs; | |
469 | } | |
470 | ||
471 | /* Unmap and put previous page of userspace buffer */ | |
8bfc016d | 472 | static void fuse_copy_finish(struct fuse_copy_state *cs) |
334f485d MS |
473 | { |
474 | if (cs->mapaddr) { | |
475 | kunmap_atomic(cs->mapaddr, KM_USER0); | |
476 | if (cs->write) { | |
477 | flush_dcache_page(cs->pg); | |
478 | set_page_dirty_lock(cs->pg); | |
479 | } | |
480 | put_page(cs->pg); | |
481 | cs->mapaddr = NULL; | |
482 | } | |
483 | } | |
484 | ||
485 | /* | |
486 | * Get another pagefull of userspace buffer, and map it to kernel | |
487 | * address space, and lock request | |
488 | */ | |
489 | static int fuse_copy_fill(struct fuse_copy_state *cs) | |
490 | { | |
491 | unsigned long offset; | |
492 | int err; | |
493 | ||
d7133114 | 494 | unlock_request(cs->fc, cs->req); |
334f485d MS |
495 | fuse_copy_finish(cs); |
496 | if (!cs->seglen) { | |
497 | BUG_ON(!cs->nr_segs); | |
498 | cs->seglen = cs->iov[0].iov_len; | |
499 | cs->addr = (unsigned long) cs->iov[0].iov_base; | |
500 | cs->iov ++; | |
501 | cs->nr_segs --; | |
502 | } | |
503 | down_read(¤t->mm->mmap_sem); | |
504 | err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0, | |
505 | &cs->pg, NULL); | |
506 | up_read(¤t->mm->mmap_sem); | |
507 | if (err < 0) | |
508 | return err; | |
509 | BUG_ON(err != 1); | |
510 | offset = cs->addr % PAGE_SIZE; | |
511 | cs->mapaddr = kmap_atomic(cs->pg, KM_USER0); | |
512 | cs->buf = cs->mapaddr + offset; | |
513 | cs->len = min(PAGE_SIZE - offset, cs->seglen); | |
514 | cs->seglen -= cs->len; | |
515 | cs->addr += cs->len; | |
516 | ||
d7133114 | 517 | return lock_request(cs->fc, cs->req); |
334f485d MS |
518 | } |
519 | ||
520 | /* Do as much copy to/from userspace buffer as we can */ | |
8bfc016d | 521 | static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size) |
334f485d MS |
522 | { |
523 | unsigned ncpy = min(*size, cs->len); | |
524 | if (val) { | |
525 | if (cs->write) | |
526 | memcpy(cs->buf, *val, ncpy); | |
527 | else | |
528 | memcpy(*val, cs->buf, ncpy); | |
529 | *val += ncpy; | |
530 | } | |
531 | *size -= ncpy; | |
532 | cs->len -= ncpy; | |
533 | cs->buf += ncpy; | |
534 | return ncpy; | |
535 | } | |
536 | ||
537 | /* | |
538 | * Copy a page in the request to/from the userspace buffer. Must be | |
539 | * done atomically | |
540 | */ | |
8bfc016d MS |
541 | static int fuse_copy_page(struct fuse_copy_state *cs, struct page *page, |
542 | unsigned offset, unsigned count, int zeroing) | |
334f485d MS |
543 | { |
544 | if (page && zeroing && count < PAGE_SIZE) { | |
545 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
546 | memset(mapaddr, 0, PAGE_SIZE); | |
547 | kunmap_atomic(mapaddr, KM_USER1); | |
548 | } | |
549 | while (count) { | |
550 | int err; | |
551 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
552 | return err; | |
553 | if (page) { | |
554 | void *mapaddr = kmap_atomic(page, KM_USER1); | |
555 | void *buf = mapaddr + offset; | |
556 | offset += fuse_copy_do(cs, &buf, &count); | |
557 | kunmap_atomic(mapaddr, KM_USER1); | |
558 | } else | |
559 | offset += fuse_copy_do(cs, NULL, &count); | |
560 | } | |
561 | if (page && !cs->write) | |
562 | flush_dcache_page(page); | |
563 | return 0; | |
564 | } | |
565 | ||
566 | /* Copy pages in the request to/from userspace buffer */ | |
567 | static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes, | |
568 | int zeroing) | |
569 | { | |
570 | unsigned i; | |
571 | struct fuse_req *req = cs->req; | |
572 | unsigned offset = req->page_offset; | |
573 | unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset); | |
574 | ||
575 | for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { | |
576 | struct page *page = req->pages[i]; | |
577 | int err = fuse_copy_page(cs, page, offset, count, zeroing); | |
578 | if (err) | |
579 | return err; | |
580 | ||
581 | nbytes -= count; | |
582 | count = min(nbytes, (unsigned) PAGE_SIZE); | |
583 | offset = 0; | |
584 | } | |
585 | return 0; | |
586 | } | |
587 | ||
588 | /* Copy a single argument in the request to/from userspace buffer */ | |
589 | static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size) | |
590 | { | |
591 | while (size) { | |
592 | int err; | |
593 | if (!cs->len && (err = fuse_copy_fill(cs))) | |
594 | return err; | |
595 | fuse_copy_do(cs, &val, &size); | |
596 | } | |
597 | return 0; | |
598 | } | |
599 | ||
600 | /* Copy request arguments to/from userspace buffer */ | |
601 | static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs, | |
602 | unsigned argpages, struct fuse_arg *args, | |
603 | int zeroing) | |
604 | { | |
605 | int err = 0; | |
606 | unsigned i; | |
607 | ||
608 | for (i = 0; !err && i < numargs; i++) { | |
609 | struct fuse_arg *arg = &args[i]; | |
610 | if (i == numargs - 1 && argpages) | |
611 | err = fuse_copy_pages(cs, arg->size, zeroing); | |
612 | else | |
613 | err = fuse_copy_one(cs, arg->value, arg->size); | |
614 | } | |
615 | return err; | |
616 | } | |
617 | ||
a4d27e75 MS |
618 | static int request_pending(struct fuse_conn *fc) |
619 | { | |
620 | return !list_empty(&fc->pending) || !list_empty(&fc->interrupts); | |
621 | } | |
622 | ||
334f485d MS |
623 | /* Wait until a request is available on the pending list */ |
624 | static void request_wait(struct fuse_conn *fc) | |
625 | { | |
626 | DECLARE_WAITQUEUE(wait, current); | |
627 | ||
628 | add_wait_queue_exclusive(&fc->waitq, &wait); | |
a4d27e75 | 629 | while (fc->connected && !request_pending(fc)) { |
334f485d MS |
630 | set_current_state(TASK_INTERRUPTIBLE); |
631 | if (signal_pending(current)) | |
632 | break; | |
633 | ||
d7133114 | 634 | spin_unlock(&fc->lock); |
334f485d | 635 | schedule(); |
d7133114 | 636 | spin_lock(&fc->lock); |
334f485d MS |
637 | } |
638 | set_current_state(TASK_RUNNING); | |
639 | remove_wait_queue(&fc->waitq, &wait); | |
640 | } | |
641 | ||
a4d27e75 MS |
642 | /* |
643 | * Transfer an interrupt request to userspace | |
644 | * | |
645 | * Unlike other requests this is assembled on demand, without a need | |
646 | * to allocate a separate fuse_req structure. | |
647 | * | |
648 | * Called with fc->lock held, releases it | |
649 | */ | |
650 | static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_req *req, | |
651 | const struct iovec *iov, unsigned long nr_segs) | |
105f4d7a | 652 | __releases(fc->lock) |
a4d27e75 MS |
653 | { |
654 | struct fuse_copy_state cs; | |
655 | struct fuse_in_header ih; | |
656 | struct fuse_interrupt_in arg; | |
657 | unsigned reqsize = sizeof(ih) + sizeof(arg); | |
658 | int err; | |
659 | ||
660 | list_del_init(&req->intr_entry); | |
661 | req->intr_unique = fuse_get_unique(fc); | |
662 | memset(&ih, 0, sizeof(ih)); | |
663 | memset(&arg, 0, sizeof(arg)); | |
664 | ih.len = reqsize; | |
665 | ih.opcode = FUSE_INTERRUPT; | |
666 | ih.unique = req->intr_unique; | |
667 | arg.unique = req->in.h.unique; | |
668 | ||
669 | spin_unlock(&fc->lock); | |
670 | if (iov_length(iov, nr_segs) < reqsize) | |
671 | return -EINVAL; | |
672 | ||
673 | fuse_copy_init(&cs, fc, 1, NULL, iov, nr_segs); | |
674 | err = fuse_copy_one(&cs, &ih, sizeof(ih)); | |
675 | if (!err) | |
676 | err = fuse_copy_one(&cs, &arg, sizeof(arg)); | |
677 | fuse_copy_finish(&cs); | |
678 | ||
679 | return err ? err : reqsize; | |
680 | } | |
681 | ||
334f485d MS |
682 | /* |
683 | * Read a single request into the userspace filesystem's buffer. This | |
684 | * function waits until a request is available, then removes it from | |
685 | * the pending list and copies request data to userspace buffer. If | |
f9a2842e MS |
686 | * no reply is needed (FORGET) or request has been aborted or there |
687 | * was an error during the copying then it's finished by calling | |
334f485d MS |
688 | * request_end(). Otherwise add it to the processing list, and set |
689 | * the 'sent' flag. | |
690 | */ | |
ee0b3e67 BP |
691 | static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov, |
692 | unsigned long nr_segs, loff_t pos) | |
334f485d MS |
693 | { |
694 | int err; | |
334f485d MS |
695 | struct fuse_req *req; |
696 | struct fuse_in *in; | |
697 | struct fuse_copy_state cs; | |
698 | unsigned reqsize; | |
ee0b3e67 | 699 | struct file *file = iocb->ki_filp; |
0720b315 MS |
700 | struct fuse_conn *fc = fuse_get_conn(file); |
701 | if (!fc) | |
702 | return -EPERM; | |
334f485d | 703 | |
1d3d752b | 704 | restart: |
d7133114 | 705 | spin_lock(&fc->lock); |
e5ac1d1e JD |
706 | err = -EAGAIN; |
707 | if ((file->f_flags & O_NONBLOCK) && fc->connected && | |
a4d27e75 | 708 | !request_pending(fc)) |
e5ac1d1e JD |
709 | goto err_unlock; |
710 | ||
334f485d MS |
711 | request_wait(fc); |
712 | err = -ENODEV; | |
9ba7cbba | 713 | if (!fc->connected) |
334f485d MS |
714 | goto err_unlock; |
715 | err = -ERESTARTSYS; | |
a4d27e75 | 716 | if (!request_pending(fc)) |
334f485d MS |
717 | goto err_unlock; |
718 | ||
a4d27e75 MS |
719 | if (!list_empty(&fc->interrupts)) { |
720 | req = list_entry(fc->interrupts.next, struct fuse_req, | |
721 | intr_entry); | |
722 | return fuse_read_interrupt(fc, req, iov, nr_segs); | |
723 | } | |
724 | ||
334f485d | 725 | req = list_entry(fc->pending.next, struct fuse_req, list); |
83cfd493 | 726 | req->state = FUSE_REQ_READING; |
d77a1d5b | 727 | list_move(&req->list, &fc->io); |
334f485d MS |
728 | |
729 | in = &req->in; | |
1d3d752b MS |
730 | reqsize = in->h.len; |
731 | /* If request is too large, reply with an error and restart the read */ | |
732 | if (iov_length(iov, nr_segs) < reqsize) { | |
733 | req->out.h.error = -EIO; | |
734 | /* SETXATTR is special, since it may contain too large data */ | |
735 | if (in->h.opcode == FUSE_SETXATTR) | |
736 | req->out.h.error = -E2BIG; | |
737 | request_end(fc, req); | |
738 | goto restart; | |
334f485d | 739 | } |
d7133114 MS |
740 | spin_unlock(&fc->lock); |
741 | fuse_copy_init(&cs, fc, 1, req, iov, nr_segs); | |
1d3d752b MS |
742 | err = fuse_copy_one(&cs, &in->h, sizeof(in->h)); |
743 | if (!err) | |
744 | err = fuse_copy_args(&cs, in->numargs, in->argpages, | |
745 | (struct fuse_arg *) in->args, 0); | |
334f485d | 746 | fuse_copy_finish(&cs); |
d7133114 | 747 | spin_lock(&fc->lock); |
334f485d | 748 | req->locked = 0; |
f9a2842e | 749 | if (!err && req->aborted) |
334f485d MS |
750 | err = -ENOENT; |
751 | if (err) { | |
f9a2842e | 752 | if (!req->aborted) |
334f485d MS |
753 | req->out.h.error = -EIO; |
754 | request_end(fc, req); | |
755 | return err; | |
756 | } | |
757 | if (!req->isreply) | |
758 | request_end(fc, req); | |
759 | else { | |
83cfd493 | 760 | req->state = FUSE_REQ_SENT; |
d77a1d5b | 761 | list_move_tail(&req->list, &fc->processing); |
a4d27e75 MS |
762 | if (req->interrupted) |
763 | queue_interrupt(fc, req); | |
d7133114 | 764 | spin_unlock(&fc->lock); |
334f485d MS |
765 | } |
766 | return reqsize; | |
767 | ||
768 | err_unlock: | |
d7133114 | 769 | spin_unlock(&fc->lock); |
334f485d MS |
770 | return err; |
771 | } | |
772 | ||
334f485d MS |
773 | /* Look up request on processing list by unique ID */ |
774 | static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique) | |
775 | { | |
776 | struct list_head *entry; | |
777 | ||
778 | list_for_each(entry, &fc->processing) { | |
779 | struct fuse_req *req; | |
780 | req = list_entry(entry, struct fuse_req, list); | |
a4d27e75 | 781 | if (req->in.h.unique == unique || req->intr_unique == unique) |
334f485d MS |
782 | return req; |
783 | } | |
784 | return NULL; | |
785 | } | |
786 | ||
787 | static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out, | |
788 | unsigned nbytes) | |
789 | { | |
790 | unsigned reqsize = sizeof(struct fuse_out_header); | |
791 | ||
792 | if (out->h.error) | |
793 | return nbytes != reqsize ? -EINVAL : 0; | |
794 | ||
795 | reqsize += len_args(out->numargs, out->args); | |
796 | ||
797 | if (reqsize < nbytes || (reqsize > nbytes && !out->argvar)) | |
798 | return -EINVAL; | |
799 | else if (reqsize > nbytes) { | |
800 | struct fuse_arg *lastarg = &out->args[out->numargs-1]; | |
801 | unsigned diffsize = reqsize - nbytes; | |
802 | if (diffsize > lastarg->size) | |
803 | return -EINVAL; | |
804 | lastarg->size -= diffsize; | |
805 | } | |
806 | return fuse_copy_args(cs, out->numargs, out->argpages, out->args, | |
807 | out->page_zeroing); | |
808 | } | |
809 | ||
810 | /* | |
811 | * Write a single reply to a request. First the header is copied from | |
812 | * the write buffer. The request is then searched on the processing | |
813 | * list by the unique ID found in the header. If found, then remove | |
814 | * it from the list and copy the rest of the buffer to the request. | |
815 | * The request is finished by calling request_end() | |
816 | */ | |
ee0b3e67 BP |
817 | static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov, |
818 | unsigned long nr_segs, loff_t pos) | |
334f485d MS |
819 | { |
820 | int err; | |
821 | unsigned nbytes = iov_length(iov, nr_segs); | |
822 | struct fuse_req *req; | |
823 | struct fuse_out_header oh; | |
824 | struct fuse_copy_state cs; | |
ee0b3e67 | 825 | struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp); |
334f485d | 826 | if (!fc) |
a87046d8 | 827 | return -EPERM; |
334f485d | 828 | |
d7133114 | 829 | fuse_copy_init(&cs, fc, 0, NULL, iov, nr_segs); |
334f485d MS |
830 | if (nbytes < sizeof(struct fuse_out_header)) |
831 | return -EINVAL; | |
832 | ||
833 | err = fuse_copy_one(&cs, &oh, sizeof(oh)); | |
834 | if (err) | |
835 | goto err_finish; | |
836 | err = -EINVAL; | |
837 | if (!oh.unique || oh.error <= -1000 || oh.error > 0 || | |
838 | oh.len != nbytes) | |
839 | goto err_finish; | |
840 | ||
d7133114 | 841 | spin_lock(&fc->lock); |
69a53bf2 MS |
842 | err = -ENOENT; |
843 | if (!fc->connected) | |
844 | goto err_unlock; | |
845 | ||
334f485d | 846 | req = request_find(fc, oh.unique); |
334f485d MS |
847 | if (!req) |
848 | goto err_unlock; | |
849 | ||
f9a2842e | 850 | if (req->aborted) { |
d7133114 | 851 | spin_unlock(&fc->lock); |
334f485d | 852 | fuse_copy_finish(&cs); |
d7133114 | 853 | spin_lock(&fc->lock); |
222f1d69 | 854 | request_end(fc, req); |
334f485d MS |
855 | return -ENOENT; |
856 | } | |
a4d27e75 MS |
857 | /* Is it an interrupt reply? */ |
858 | if (req->intr_unique == oh.unique) { | |
859 | err = -EINVAL; | |
860 | if (nbytes != sizeof(struct fuse_out_header)) | |
861 | goto err_unlock; | |
862 | ||
863 | if (oh.error == -ENOSYS) | |
864 | fc->no_interrupt = 1; | |
865 | else if (oh.error == -EAGAIN) | |
866 | queue_interrupt(fc, req); | |
867 | ||
868 | spin_unlock(&fc->lock); | |
869 | fuse_copy_finish(&cs); | |
870 | return nbytes; | |
871 | } | |
872 | ||
873 | req->state = FUSE_REQ_WRITING; | |
d77a1d5b | 874 | list_move(&req->list, &fc->io); |
334f485d MS |
875 | req->out.h = oh; |
876 | req->locked = 1; | |
877 | cs.req = req; | |
d7133114 | 878 | spin_unlock(&fc->lock); |
334f485d MS |
879 | |
880 | err = copy_out_args(&cs, &req->out, nbytes); | |
881 | fuse_copy_finish(&cs); | |
882 | ||
d7133114 | 883 | spin_lock(&fc->lock); |
334f485d MS |
884 | req->locked = 0; |
885 | if (!err) { | |
f9a2842e | 886 | if (req->aborted) |
334f485d | 887 | err = -ENOENT; |
f9a2842e | 888 | } else if (!req->aborted) |
334f485d MS |
889 | req->out.h.error = -EIO; |
890 | request_end(fc, req); | |
891 | ||
892 | return err ? err : nbytes; | |
893 | ||
894 | err_unlock: | |
d7133114 | 895 | spin_unlock(&fc->lock); |
334f485d MS |
896 | err_finish: |
897 | fuse_copy_finish(&cs); | |
898 | return err; | |
899 | } | |
900 | ||
334f485d MS |
901 | static unsigned fuse_dev_poll(struct file *file, poll_table *wait) |
902 | { | |
334f485d | 903 | unsigned mask = POLLOUT | POLLWRNORM; |
7025d9ad | 904 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 905 | if (!fc) |
7025d9ad | 906 | return POLLERR; |
334f485d MS |
907 | |
908 | poll_wait(file, &fc->waitq, wait); | |
909 | ||
d7133114 | 910 | spin_lock(&fc->lock); |
7025d9ad MS |
911 | if (!fc->connected) |
912 | mask = POLLERR; | |
a4d27e75 | 913 | else if (request_pending(fc)) |
7025d9ad | 914 | mask |= POLLIN | POLLRDNORM; |
d7133114 | 915 | spin_unlock(&fc->lock); |
334f485d MS |
916 | |
917 | return mask; | |
918 | } | |
919 | ||
69a53bf2 MS |
920 | /* |
921 | * Abort all requests on the given list (pending or processing) | |
922 | * | |
d7133114 | 923 | * This function releases and reacquires fc->lock |
69a53bf2 | 924 | */ |
334f485d MS |
925 | static void end_requests(struct fuse_conn *fc, struct list_head *head) |
926 | { | |
927 | while (!list_empty(head)) { | |
928 | struct fuse_req *req; | |
929 | req = list_entry(head->next, struct fuse_req, list); | |
334f485d MS |
930 | req->out.h.error = -ECONNABORTED; |
931 | request_end(fc, req); | |
d7133114 | 932 | spin_lock(&fc->lock); |
334f485d MS |
933 | } |
934 | } | |
935 | ||
69a53bf2 MS |
936 | /* |
937 | * Abort requests under I/O | |
938 | * | |
f9a2842e | 939 | * The requests are set to aborted and finished, and the request |
69a53bf2 MS |
940 | * waiter is woken up. This will make request_wait_answer() wait |
941 | * until the request is unlocked and then return. | |
64c6d8ed MS |
942 | * |
943 | * If the request is asynchronous, then the end function needs to be | |
944 | * called after waiting for the request to be unlocked (if it was | |
945 | * locked). | |
69a53bf2 MS |
946 | */ |
947 | static void end_io_requests(struct fuse_conn *fc) | |
948 | { | |
949 | while (!list_empty(&fc->io)) { | |
64c6d8ed MS |
950 | struct fuse_req *req = |
951 | list_entry(fc->io.next, struct fuse_req, list); | |
952 | void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; | |
953 | ||
f9a2842e | 954 | req->aborted = 1; |
69a53bf2 MS |
955 | req->out.h.error = -ECONNABORTED; |
956 | req->state = FUSE_REQ_FINISHED; | |
957 | list_del_init(&req->list); | |
958 | wake_up(&req->waitq); | |
64c6d8ed MS |
959 | if (end) { |
960 | req->end = NULL; | |
961 | /* The end function will consume this reference */ | |
962 | __fuse_get_request(req); | |
d7133114 | 963 | spin_unlock(&fc->lock); |
64c6d8ed MS |
964 | wait_event(req->waitq, !req->locked); |
965 | end(fc, req); | |
d7133114 | 966 | spin_lock(&fc->lock); |
64c6d8ed | 967 | } |
69a53bf2 MS |
968 | } |
969 | } | |
970 | ||
971 | /* | |
972 | * Abort all requests. | |
973 | * | |
974 | * Emergency exit in case of a malicious or accidental deadlock, or | |
975 | * just a hung filesystem. | |
976 | * | |
977 | * The same effect is usually achievable through killing the | |
978 | * filesystem daemon and all users of the filesystem. The exception | |
979 | * is the combination of an asynchronous request and the tricky | |
980 | * deadlock (see Documentation/filesystems/fuse.txt). | |
981 | * | |
982 | * During the aborting, progression of requests from the pending and | |
983 | * processing lists onto the io list, and progression of new requests | |
984 | * onto the pending list is prevented by req->connected being false. | |
985 | * | |
986 | * Progression of requests under I/O to the processing list is | |
f9a2842e MS |
987 | * prevented by the req->aborted flag being true for these requests. |
988 | * For this reason requests on the io list must be aborted first. | |
69a53bf2 MS |
989 | */ |
990 | void fuse_abort_conn(struct fuse_conn *fc) | |
991 | { | |
d7133114 | 992 | spin_lock(&fc->lock); |
69a53bf2 MS |
993 | if (fc->connected) { |
994 | fc->connected = 0; | |
51eb01e7 | 995 | fc->blocked = 0; |
69a53bf2 MS |
996 | end_io_requests(fc); |
997 | end_requests(fc, &fc->pending); | |
998 | end_requests(fc, &fc->processing); | |
999 | wake_up_all(&fc->waitq); | |
51eb01e7 | 1000 | wake_up_all(&fc->blocked_waitq); |
385a17bf | 1001 | kill_fasync(&fc->fasync, SIGIO, POLL_IN); |
69a53bf2 | 1002 | } |
d7133114 | 1003 | spin_unlock(&fc->lock); |
69a53bf2 MS |
1004 | } |
1005 | ||
334f485d MS |
1006 | static int fuse_dev_release(struct inode *inode, struct file *file) |
1007 | { | |
0720b315 | 1008 | struct fuse_conn *fc = fuse_get_conn(file); |
334f485d | 1009 | if (fc) { |
d7133114 | 1010 | spin_lock(&fc->lock); |
1e9a4ed9 | 1011 | fc->connected = 0; |
334f485d MS |
1012 | end_requests(fc, &fc->pending); |
1013 | end_requests(fc, &fc->processing); | |
d7133114 | 1014 | spin_unlock(&fc->lock); |
385a17bf | 1015 | fasync_helper(-1, file, 0, &fc->fasync); |
bafa9654 | 1016 | fuse_conn_put(fc); |
385a17bf | 1017 | } |
f543f253 | 1018 | |
334f485d MS |
1019 | return 0; |
1020 | } | |
1021 | ||
385a17bf JD |
1022 | static int fuse_dev_fasync(int fd, struct file *file, int on) |
1023 | { | |
1024 | struct fuse_conn *fc = fuse_get_conn(file); | |
1025 | if (!fc) | |
a87046d8 | 1026 | return -EPERM; |
385a17bf JD |
1027 | |
1028 | /* No locking - fasync_helper does its own locking */ | |
1029 | return fasync_helper(fd, file, on, &fc->fasync); | |
1030 | } | |
1031 | ||
4b6f5d20 | 1032 | const struct file_operations fuse_dev_operations = { |
334f485d MS |
1033 | .owner = THIS_MODULE, |
1034 | .llseek = no_llseek, | |
ee0b3e67 BP |
1035 | .read = do_sync_read, |
1036 | .aio_read = fuse_dev_read, | |
1037 | .write = do_sync_write, | |
1038 | .aio_write = fuse_dev_write, | |
334f485d MS |
1039 | .poll = fuse_dev_poll, |
1040 | .release = fuse_dev_release, | |
385a17bf | 1041 | .fasync = fuse_dev_fasync, |
334f485d MS |
1042 | }; |
1043 | ||
1044 | static struct miscdevice fuse_miscdevice = { | |
1045 | .minor = FUSE_MINOR, | |
1046 | .name = "fuse", | |
1047 | .fops = &fuse_dev_operations, | |
1048 | }; | |
1049 | ||
1050 | int __init fuse_dev_init(void) | |
1051 | { | |
1052 | int err = -ENOMEM; | |
1053 | fuse_req_cachep = kmem_cache_create("fuse_request", | |
1054 | sizeof(struct fuse_req), | |
20c2df83 | 1055 | 0, 0, NULL); |
334f485d MS |
1056 | if (!fuse_req_cachep) |
1057 | goto out; | |
1058 | ||
1059 | err = misc_register(&fuse_miscdevice); | |
1060 | if (err) | |
1061 | goto out_cache_clean; | |
1062 | ||
1063 | return 0; | |
1064 | ||
1065 | out_cache_clean: | |
1066 | kmem_cache_destroy(fuse_req_cachep); | |
1067 | out: | |
1068 | return err; | |
1069 | } | |
1070 | ||
1071 | void fuse_dev_cleanup(void) | |
1072 | { | |
1073 | misc_deregister(&fuse_miscdevice); | |
1074 | kmem_cache_destroy(fuse_req_cachep); | |
1075 | } |