ia64/linux-2.6.18-xen.hg

view fs/fuse/dev.c @ 524:7f8b544237bf

netfront: Allow netfront in domain 0.

This is useful if your physical network device is in a utility domain.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Apr 15 15:18:58 2008 +0100 (2008-04-15)
parents 831230e53067
children
line source
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2006 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
9 #include "fuse_i.h"
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>
20 MODULE_ALIAS_MISCDEV(FUSE_MINOR);
22 static kmem_cache_t *fuse_req_cachep;
24 static struct fuse_conn *fuse_get_conn(struct file *file)
25 {
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;
31 }
33 static void fuse_request_init(struct fuse_req *req)
34 {
35 memset(req, 0, sizeof(*req));
36 INIT_LIST_HEAD(&req->list);
37 INIT_LIST_HEAD(&req->intr_entry);
38 init_waitqueue_head(&req->waitq);
39 atomic_set(&req->count, 1);
40 }
42 struct fuse_req *fuse_request_alloc(void)
43 {
44 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, SLAB_KERNEL);
45 if (req)
46 fuse_request_init(req);
47 return req;
48 }
50 void fuse_request_free(struct fuse_req *req)
51 {
52 kmem_cache_free(fuse_req_cachep, req);
53 }
55 static void block_sigs(sigset_t *oldset)
56 {
57 sigset_t mask;
59 siginitsetinv(&mask, sigmask(SIGKILL));
60 sigprocmask(SIG_BLOCK, &mask, oldset);
61 }
63 static void restore_sigs(sigset_t *oldset)
64 {
65 sigprocmask(SIG_SETMASK, oldset, NULL);
66 }
68 static void __fuse_get_request(struct fuse_req *req)
69 {
70 atomic_inc(&req->count);
71 }
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 }
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 }
87 struct fuse_req *fuse_get_req(struct fuse_conn *fc)
88 {
89 struct fuse_req *req;
90 sigset_t oldset;
91 int intr;
92 int err;
94 atomic_inc(&fc->num_waiting);
95 block_sigs(&oldset);
96 intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked);
97 restore_sigs(&oldset);
98 err = -EINTR;
99 if (intr)
100 goto out;
102 err = -ENOTCONN;
103 if (!fc->connected)
104 goto out;
106 req = fuse_request_alloc();
107 err = -ENOMEM;
108 if (!req)
109 goto out;
111 fuse_req_init_context(req);
112 req->waiting = 1;
113 return req;
115 out:
116 atomic_dec(&fc->num_waiting);
117 return ERR_PTR(err);
118 }
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;
131 do {
132 wait_event(fc->blocked_waitq, ff->reserved_req);
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);
143 return req;
144 }
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;
154 spin_lock(&fc->lock);
155 fuse_request_init(req);
156 BUG_ON(ff->reserved_req);
157 ff->reserved_req = req;
158 wake_up(&fc->blocked_waitq);
159 spin_unlock(&fc->lock);
160 fput(file);
161 }
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;
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);
186 fuse_req_init_context(req);
187 req->waiting = 1;
188 return req;
189 }
191 void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
192 {
193 if (atomic_dec_and_test(&req->count)) {
194 if (req->waiting)
195 atomic_dec(&fc->num_waiting);
197 if (req->stolen_file)
198 put_reserved_req(fc, req);
199 else
200 fuse_request_free(req);
201 }
202 }
204 /*
205 * This function is called when a request is finished. Either a reply
206 * has arrived or it was aborted (and not yet sent) or some error
207 * occurred during communication with userspace, or the device file
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
211 *
212 * Called with fc->lock, unlocks it
213 */
214 static void request_end(struct fuse_conn *fc, struct fuse_req *req)
215 {
216 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
217 req->end = NULL;
218 list_del(&req->list);
219 list_del(&req->intr_entry);
220 req->state = FUSE_REQ_FINISHED;
221 if (req->background) {
222 if (fc->num_background == FUSE_MAX_BACKGROUND) {
223 fc->blocked = 0;
224 wake_up_all(&fc->blocked_waitq);
225 }
226 fc->num_background--;
227 }
228 spin_unlock(&fc->lock);
229 dput(req->dentry);
230 mntput(req->vfsmount);
231 if (req->file)
232 fput(req->file);
233 wake_up(&req->waitq);
234 if (end)
235 end(fc, req);
236 else
237 fuse_put_request(fc, req);
238 }
240 static void wait_answer_interruptible(struct fuse_conn *fc,
241 struct fuse_req *req)
242 {
243 if (signal_pending(current))
244 return;
246 spin_unlock(&fc->lock);
247 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
248 spin_lock(&fc->lock);
249 }
251 static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
252 {
253 list_add_tail(&req->intr_entry, &fc->interrupts);
254 wake_up(&fc->waitq);
255 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
256 }
258 /* Called with fc->lock held. Releases, and then reacquires it. */
259 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
260 {
261 if (!fc->no_interrupt) {
262 /* Any signal may interrupt this */
263 wait_answer_interruptible(fc, req);
265 if (req->aborted)
266 goto aborted;
267 if (req->state == FUSE_REQ_FINISHED)
268 return;
270 req->interrupted = 1;
271 if (req->state == FUSE_REQ_SENT)
272 queue_interrupt(fc, req);
273 }
275 if (req->force) {
276 spin_unlock(&fc->lock);
277 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
278 spin_lock(&fc->lock);
279 } else {
280 sigset_t oldset;
282 /* Only fatal signals may interrupt this */
283 block_sigs(&oldset);
284 wait_answer_interruptible(fc, req);
285 restore_sigs(&oldset);
286 }
288 if (req->aborted)
289 goto aborted;
290 if (req->state == FUSE_REQ_FINISHED)
291 return;
293 req->out.h.error = -EINTR;
294 req->aborted = 1;
296 aborted:
297 if (req->locked) {
298 /* This is uninterruptible sleep, because data is
299 being copied to/from the buffers of req. During
300 locked state, there mustn't be any filesystem
301 operation (e.g. page fault), since that could lead
302 to deadlock */
303 spin_unlock(&fc->lock);
304 wait_event(req->waitq, !req->locked);
305 spin_lock(&fc->lock);
306 }
307 if (req->state == FUSE_REQ_PENDING) {
308 list_del(&req->list);
309 __fuse_put_request(req);
310 } else if (req->state == FUSE_REQ_SENT) {
311 spin_unlock(&fc->lock);
312 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
313 spin_lock(&fc->lock);
314 }
315 }
317 static unsigned len_args(unsigned numargs, struct fuse_arg *args)
318 {
319 unsigned nbytes = 0;
320 unsigned i;
322 for (i = 0; i < numargs; i++)
323 nbytes += args[i].size;
325 return nbytes;
326 }
328 static u64 fuse_get_unique(struct fuse_conn *fc)
329 {
330 fc->reqctr++;
331 /* zero is special */
332 if (fc->reqctr == 0)
333 fc->reqctr = 1;
335 return fc->reqctr;
336 }
338 static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
339 {
340 req->in.h.unique = fuse_get_unique(fc);
341 req->in.h.len = sizeof(struct fuse_in_header) +
342 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
343 list_add_tail(&req->list, &fc->pending);
344 req->state = FUSE_REQ_PENDING;
345 if (!req->waiting) {
346 req->waiting = 1;
347 atomic_inc(&fc->num_waiting);
348 }
349 wake_up(&fc->waitq);
350 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
351 }
353 void request_send(struct fuse_conn *fc, struct fuse_req *req)
354 {
355 req->isreply = 1;
356 spin_lock(&fc->lock);
357 if (!fc->connected)
358 req->out.h.error = -ENOTCONN;
359 else if (fc->conn_error)
360 req->out.h.error = -ECONNREFUSED;
361 else {
362 queue_request(fc, req);
363 /* acquire extra reference, since request is still needed
364 after request_end() */
365 __fuse_get_request(req);
367 request_wait_answer(fc, req);
368 }
369 spin_unlock(&fc->lock);
370 }
372 static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
373 {
374 spin_lock(&fc->lock);
375 if (fc->connected) {
376 req->background = 1;
377 fc->num_background++;
378 if (fc->num_background == FUSE_MAX_BACKGROUND)
379 fc->blocked = 1;
381 queue_request(fc, req);
382 spin_unlock(&fc->lock);
383 } else {
384 req->out.h.error = -ENOTCONN;
385 request_end(fc, req);
386 }
387 }
389 void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req)
390 {
391 req->isreply = 0;
392 request_send_nowait(fc, req);
393 }
395 void request_send_background(struct fuse_conn *fc, struct fuse_req *req)
396 {
397 req->isreply = 1;
398 request_send_nowait(fc, req);
399 }
401 /*
402 * Lock the request. Up to the next unlock_request() there mustn't be
403 * anything that could cause a page-fault. If the request was already
404 * aborted bail out.
405 */
406 static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
407 {
408 int err = 0;
409 if (req) {
410 spin_lock(&fc->lock);
411 if (req->aborted)
412 err = -ENOENT;
413 else
414 req->locked = 1;
415 spin_unlock(&fc->lock);
416 }
417 return err;
418 }
420 /*
421 * Unlock request. If it was aborted during being locked, the
422 * requester thread is currently waiting for it to be unlocked, so
423 * wake it up.
424 */
425 static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
426 {
427 if (req) {
428 spin_lock(&fc->lock);
429 req->locked = 0;
430 if (req->aborted)
431 wake_up(&req->waitq);
432 spin_unlock(&fc->lock);
433 }
434 }
436 struct fuse_copy_state {
437 struct fuse_conn *fc;
438 int write;
439 struct fuse_req *req;
440 const struct iovec *iov;
441 unsigned long nr_segs;
442 unsigned long seglen;
443 unsigned long addr;
444 struct page *pg;
445 void *mapaddr;
446 void *buf;
447 unsigned len;
448 };
450 static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
451 int write, struct fuse_req *req,
452 const struct iovec *iov, unsigned long nr_segs)
453 {
454 memset(cs, 0, sizeof(*cs));
455 cs->fc = fc;
456 cs->write = write;
457 cs->req = req;
458 cs->iov = iov;
459 cs->nr_segs = nr_segs;
460 }
462 /* Unmap and put previous page of userspace buffer */
463 static void fuse_copy_finish(struct fuse_copy_state *cs)
464 {
465 if (cs->mapaddr) {
466 kunmap_atomic(cs->mapaddr, KM_USER0);
467 if (cs->write) {
468 flush_dcache_page(cs->pg);
469 set_page_dirty_lock(cs->pg);
470 }
471 put_page(cs->pg);
472 cs->mapaddr = NULL;
473 }
474 }
476 /*
477 * Get another pagefull of userspace buffer, and map it to kernel
478 * address space, and lock request
479 */
480 static int fuse_copy_fill(struct fuse_copy_state *cs)
481 {
482 unsigned long offset;
483 int err;
485 unlock_request(cs->fc, cs->req);
486 fuse_copy_finish(cs);
487 if (!cs->seglen) {
488 BUG_ON(!cs->nr_segs);
489 cs->seglen = cs->iov[0].iov_len;
490 cs->addr = (unsigned long) cs->iov[0].iov_base;
491 cs->iov ++;
492 cs->nr_segs --;
493 }
494 down_read(&current->mm->mmap_sem);
495 err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0,
496 &cs->pg, NULL);
497 up_read(&current->mm->mmap_sem);
498 if (err < 0)
499 return err;
500 BUG_ON(err != 1);
501 offset = cs->addr % PAGE_SIZE;
502 cs->mapaddr = kmap_atomic(cs->pg, KM_USER0);
503 cs->buf = cs->mapaddr + offset;
504 cs->len = min(PAGE_SIZE - offset, cs->seglen);
505 cs->seglen -= cs->len;
506 cs->addr += cs->len;
508 return lock_request(cs->fc, cs->req);
509 }
511 /* Do as much copy to/from userspace buffer as we can */
512 static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
513 {
514 unsigned ncpy = min(*size, cs->len);
515 if (val) {
516 if (cs->write)
517 memcpy(cs->buf, *val, ncpy);
518 else
519 memcpy(*val, cs->buf, ncpy);
520 *val += ncpy;
521 }
522 *size -= ncpy;
523 cs->len -= ncpy;
524 cs->buf += ncpy;
525 return ncpy;
526 }
528 /*
529 * Copy a page in the request to/from the userspace buffer. Must be
530 * done atomically
531 */
532 static int fuse_copy_page(struct fuse_copy_state *cs, struct page *page,
533 unsigned offset, unsigned count, int zeroing)
534 {
535 if (page && zeroing && count < PAGE_SIZE) {
536 void *mapaddr = kmap_atomic(page, KM_USER1);
537 memset(mapaddr, 0, PAGE_SIZE);
538 kunmap_atomic(mapaddr, KM_USER1);
539 }
540 while (count) {
541 int err;
542 if (!cs->len && (err = fuse_copy_fill(cs)))
543 return err;
544 if (page) {
545 void *mapaddr = kmap_atomic(page, KM_USER1);
546 void *buf = mapaddr + offset;
547 offset += fuse_copy_do(cs, &buf, &count);
548 kunmap_atomic(mapaddr, KM_USER1);
549 } else
550 offset += fuse_copy_do(cs, NULL, &count);
551 }
552 if (page && !cs->write)
553 flush_dcache_page(page);
554 return 0;
555 }
557 /* Copy pages in the request to/from userspace buffer */
558 static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
559 int zeroing)
560 {
561 unsigned i;
562 struct fuse_req *req = cs->req;
563 unsigned offset = req->page_offset;
564 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
566 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
567 struct page *page = req->pages[i];
568 int err = fuse_copy_page(cs, page, offset, count, zeroing);
569 if (err)
570 return err;
572 nbytes -= count;
573 count = min(nbytes, (unsigned) PAGE_SIZE);
574 offset = 0;
575 }
576 return 0;
577 }
579 /* Copy a single argument in the request to/from userspace buffer */
580 static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
581 {
582 while (size) {
583 int err;
584 if (!cs->len && (err = fuse_copy_fill(cs)))
585 return err;
586 fuse_copy_do(cs, &val, &size);
587 }
588 return 0;
589 }
591 /* Copy request arguments to/from userspace buffer */
592 static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
593 unsigned argpages, struct fuse_arg *args,
594 int zeroing)
595 {
596 int err = 0;
597 unsigned i;
599 for (i = 0; !err && i < numargs; i++) {
600 struct fuse_arg *arg = &args[i];
601 if (i == numargs - 1 && argpages)
602 err = fuse_copy_pages(cs, arg->size, zeroing);
603 else
604 err = fuse_copy_one(cs, arg->value, arg->size);
605 }
606 return err;
607 }
609 static int request_pending(struct fuse_conn *fc)
610 {
611 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts);
612 }
614 /* Wait until a request is available on the pending list */
615 static void request_wait(struct fuse_conn *fc)
616 {
617 DECLARE_WAITQUEUE(wait, current);
619 add_wait_queue_exclusive(&fc->waitq, &wait);
620 while (fc->connected && !request_pending(fc)) {
621 set_current_state(TASK_INTERRUPTIBLE);
622 if (signal_pending(current))
623 break;
625 spin_unlock(&fc->lock);
626 schedule();
627 spin_lock(&fc->lock);
628 }
629 set_current_state(TASK_RUNNING);
630 remove_wait_queue(&fc->waitq, &wait);
631 }
633 /*
634 * Transfer an interrupt request to userspace
635 *
636 * Unlike other requests this is assembled on demand, without a need
637 * to allocate a separate fuse_req structure.
638 *
639 * Called with fc->lock held, releases it
640 */
641 static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_req *req,
642 const struct iovec *iov, unsigned long nr_segs)
643 {
644 struct fuse_copy_state cs;
645 struct fuse_in_header ih;
646 struct fuse_interrupt_in arg;
647 unsigned reqsize = sizeof(ih) + sizeof(arg);
648 int err;
650 list_del_init(&req->intr_entry);
651 req->intr_unique = fuse_get_unique(fc);
652 memset(&ih, 0, sizeof(ih));
653 memset(&arg, 0, sizeof(arg));
654 ih.len = reqsize;
655 ih.opcode = FUSE_INTERRUPT;
656 ih.unique = req->intr_unique;
657 arg.unique = req->in.h.unique;
659 spin_unlock(&fc->lock);
660 if (iov_length(iov, nr_segs) < reqsize)
661 return -EINVAL;
663 fuse_copy_init(&cs, fc, 1, NULL, iov, nr_segs);
664 err = fuse_copy_one(&cs, &ih, sizeof(ih));
665 if (!err)
666 err = fuse_copy_one(&cs, &arg, sizeof(arg));
667 fuse_copy_finish(&cs);
669 return err ? err : reqsize;
670 }
672 /*
673 * Read a single request into the userspace filesystem's buffer. This
674 * function waits until a request is available, then removes it from
675 * the pending list and copies request data to userspace buffer. If
676 * no reply is needed (FORGET) or request has been aborted or there
677 * was an error during the copying then it's finished by calling
678 * request_end(). Otherwise add it to the processing list, and set
679 * the 'sent' flag.
680 */
681 static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov,
682 unsigned long nr_segs, loff_t *off)
683 {
684 int err;
685 struct fuse_req *req;
686 struct fuse_in *in;
687 struct fuse_copy_state cs;
688 unsigned reqsize;
689 struct fuse_conn *fc = fuse_get_conn(file);
690 if (!fc)
691 return -EPERM;
693 restart:
694 spin_lock(&fc->lock);
695 err = -EAGAIN;
696 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
697 !request_pending(fc))
698 goto err_unlock;
700 request_wait(fc);
701 err = -ENODEV;
702 if (!fc->connected)
703 goto err_unlock;
704 err = -ERESTARTSYS;
705 if (!request_pending(fc))
706 goto err_unlock;
708 if (!list_empty(&fc->interrupts)) {
709 req = list_entry(fc->interrupts.next, struct fuse_req,
710 intr_entry);
711 return fuse_read_interrupt(fc, req, iov, nr_segs);
712 }
714 req = list_entry(fc->pending.next, struct fuse_req, list);
715 req->state = FUSE_REQ_READING;
716 list_move(&req->list, &fc->io);
718 in = &req->in;
719 reqsize = in->h.len;
720 /* If request is too large, reply with an error and restart the read */
721 if (iov_length(iov, nr_segs) < reqsize) {
722 req->out.h.error = -EIO;
723 /* SETXATTR is special, since it may contain too large data */
724 if (in->h.opcode == FUSE_SETXATTR)
725 req->out.h.error = -E2BIG;
726 request_end(fc, req);
727 goto restart;
728 }
729 spin_unlock(&fc->lock);
730 fuse_copy_init(&cs, fc, 1, req, iov, nr_segs);
731 err = fuse_copy_one(&cs, &in->h, sizeof(in->h));
732 if (!err)
733 err = fuse_copy_args(&cs, in->numargs, in->argpages,
734 (struct fuse_arg *) in->args, 0);
735 fuse_copy_finish(&cs);
736 spin_lock(&fc->lock);
737 req->locked = 0;
738 if (!err && req->aborted)
739 err = -ENOENT;
740 if (err) {
741 if (!req->aborted)
742 req->out.h.error = -EIO;
743 request_end(fc, req);
744 return err;
745 }
746 if (!req->isreply)
747 request_end(fc, req);
748 else {
749 req->state = FUSE_REQ_SENT;
750 list_move_tail(&req->list, &fc->processing);
751 if (req->interrupted)
752 queue_interrupt(fc, req);
753 spin_unlock(&fc->lock);
754 }
755 return reqsize;
757 err_unlock:
758 spin_unlock(&fc->lock);
759 return err;
760 }
762 static ssize_t fuse_dev_read(struct file *file, char __user *buf,
763 size_t nbytes, loff_t *off)
764 {
765 struct iovec iov;
766 iov.iov_len = nbytes;
767 iov.iov_base = buf;
768 return fuse_dev_readv(file, &iov, 1, off);
769 }
771 /* Look up request on processing list by unique ID */
772 static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
773 {
774 struct list_head *entry;
776 list_for_each(entry, &fc->processing) {
777 struct fuse_req *req;
778 req = list_entry(entry, struct fuse_req, list);
779 if (req->in.h.unique == unique || req->intr_unique == unique)
780 return req;
781 }
782 return NULL;
783 }
785 static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
786 unsigned nbytes)
787 {
788 unsigned reqsize = sizeof(struct fuse_out_header);
790 if (out->h.error)
791 return nbytes != reqsize ? -EINVAL : 0;
793 reqsize += len_args(out->numargs, out->args);
795 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
796 return -EINVAL;
797 else if (reqsize > nbytes) {
798 struct fuse_arg *lastarg = &out->args[out->numargs-1];
799 unsigned diffsize = reqsize - nbytes;
800 if (diffsize > lastarg->size)
801 return -EINVAL;
802 lastarg->size -= diffsize;
803 }
804 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
805 out->page_zeroing);
806 }
808 /*
809 * Write a single reply to a request. First the header is copied from
810 * the write buffer. The request is then searched on the processing
811 * list by the unique ID found in the header. If found, then remove
812 * it from the list and copy the rest of the buffer to the request.
813 * The request is finished by calling request_end()
814 */
815 static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov,
816 unsigned long nr_segs, loff_t *off)
817 {
818 int err;
819 unsigned nbytes = iov_length(iov, nr_segs);
820 struct fuse_req *req;
821 struct fuse_out_header oh;
822 struct fuse_copy_state cs;
823 struct fuse_conn *fc = fuse_get_conn(file);
824 if (!fc)
825 return -EPERM;
827 fuse_copy_init(&cs, fc, 0, NULL, iov, nr_segs);
828 if (nbytes < sizeof(struct fuse_out_header))
829 return -EINVAL;
831 err = fuse_copy_one(&cs, &oh, sizeof(oh));
832 if (err)
833 goto err_finish;
834 err = -EINVAL;
835 if (!oh.unique || oh.error <= -1000 || oh.error > 0 ||
836 oh.len != nbytes)
837 goto err_finish;
839 spin_lock(&fc->lock);
840 err = -ENOENT;
841 if (!fc->connected)
842 goto err_unlock;
844 req = request_find(fc, oh.unique);
845 if (!req)
846 goto err_unlock;
848 if (req->aborted) {
849 spin_unlock(&fc->lock);
850 fuse_copy_finish(&cs);
851 spin_lock(&fc->lock);
852 request_end(fc, req);
853 return -ENOENT;
854 }
855 /* Is it an interrupt reply? */
856 if (req->intr_unique == oh.unique) {
857 err = -EINVAL;
858 if (nbytes != sizeof(struct fuse_out_header))
859 goto err_unlock;
861 if (oh.error == -ENOSYS)
862 fc->no_interrupt = 1;
863 else if (oh.error == -EAGAIN)
864 queue_interrupt(fc, req);
866 spin_unlock(&fc->lock);
867 fuse_copy_finish(&cs);
868 return nbytes;
869 }
871 req->state = FUSE_REQ_WRITING;
872 list_move(&req->list, &fc->io);
873 req->out.h = oh;
874 req->locked = 1;
875 cs.req = req;
876 spin_unlock(&fc->lock);
878 err = copy_out_args(&cs, &req->out, nbytes);
879 fuse_copy_finish(&cs);
881 spin_lock(&fc->lock);
882 req->locked = 0;
883 if (!err) {
884 if (req->aborted)
885 err = -ENOENT;
886 } else if (!req->aborted)
887 req->out.h.error = -EIO;
888 request_end(fc, req);
890 return err ? err : nbytes;
892 err_unlock:
893 spin_unlock(&fc->lock);
894 err_finish:
895 fuse_copy_finish(&cs);
896 return err;
897 }
899 static ssize_t fuse_dev_write(struct file *file, const char __user *buf,
900 size_t nbytes, loff_t *off)
901 {
902 struct iovec iov;
903 iov.iov_len = nbytes;
904 iov.iov_base = (char __user *) buf;
905 return fuse_dev_writev(file, &iov, 1, off);
906 }
908 static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
909 {
910 unsigned mask = POLLOUT | POLLWRNORM;
911 struct fuse_conn *fc = fuse_get_conn(file);
912 if (!fc)
913 return POLLERR;
915 poll_wait(file, &fc->waitq, wait);
917 spin_lock(&fc->lock);
918 if (!fc->connected)
919 mask = POLLERR;
920 else if (request_pending(fc))
921 mask |= POLLIN | POLLRDNORM;
922 spin_unlock(&fc->lock);
924 return mask;
925 }
927 /*
928 * Abort all requests on the given list (pending or processing)
929 *
930 * This function releases and reacquires fc->lock
931 */
932 static void end_requests(struct fuse_conn *fc, struct list_head *head)
933 {
934 while (!list_empty(head)) {
935 struct fuse_req *req;
936 req = list_entry(head->next, struct fuse_req, list);
937 req->out.h.error = -ECONNABORTED;
938 request_end(fc, req);
939 spin_lock(&fc->lock);
940 }
941 }
943 /*
944 * Abort requests under I/O
945 *
946 * The requests are set to aborted and finished, and the request
947 * waiter is woken up. This will make request_wait_answer() wait
948 * until the request is unlocked and then return.
949 *
950 * If the request is asynchronous, then the end function needs to be
951 * called after waiting for the request to be unlocked (if it was
952 * locked).
953 */
954 static void end_io_requests(struct fuse_conn *fc)
955 {
956 while (!list_empty(&fc->io)) {
957 struct fuse_req *req =
958 list_entry(fc->io.next, struct fuse_req, list);
959 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
961 req->aborted = 1;
962 req->out.h.error = -ECONNABORTED;
963 req->state = FUSE_REQ_FINISHED;
964 list_del_init(&req->list);
965 wake_up(&req->waitq);
966 if (end) {
967 req->end = NULL;
968 /* The end function will consume this reference */
969 __fuse_get_request(req);
970 spin_unlock(&fc->lock);
971 wait_event(req->waitq, !req->locked);
972 end(fc, req);
973 spin_lock(&fc->lock);
974 }
975 }
976 }
978 /*
979 * Abort all requests.
980 *
981 * Emergency exit in case of a malicious or accidental deadlock, or
982 * just a hung filesystem.
983 *
984 * The same effect is usually achievable through killing the
985 * filesystem daemon and all users of the filesystem. The exception
986 * is the combination of an asynchronous request and the tricky
987 * deadlock (see Documentation/filesystems/fuse.txt).
988 *
989 * During the aborting, progression of requests from the pending and
990 * processing lists onto the io list, and progression of new requests
991 * onto the pending list is prevented by req->connected being false.
992 *
993 * Progression of requests under I/O to the processing list is
994 * prevented by the req->aborted flag being true for these requests.
995 * For this reason requests on the io list must be aborted first.
996 */
997 void fuse_abort_conn(struct fuse_conn *fc)
998 {
999 spin_lock(&fc->lock);
1000 if (fc->connected) {
1001 fc->connected = 0;
1002 fc->blocked = 0;
1003 end_io_requests(fc);
1004 end_requests(fc, &fc->pending);
1005 end_requests(fc, &fc->processing);
1006 wake_up_all(&fc->waitq);
1007 wake_up_all(&fc->blocked_waitq);
1008 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
1010 spin_unlock(&fc->lock);
1013 static int fuse_dev_release(struct inode *inode, struct file *file)
1015 struct fuse_conn *fc = fuse_get_conn(file);
1016 if (fc) {
1017 spin_lock(&fc->lock);
1018 fc->connected = 0;
1019 end_requests(fc, &fc->pending);
1020 end_requests(fc, &fc->processing);
1021 spin_unlock(&fc->lock);
1022 fasync_helper(-1, file, 0, &fc->fasync);
1023 fuse_conn_put(fc);
1026 return 0;
1029 static int fuse_dev_fasync(int fd, struct file *file, int on)
1031 struct fuse_conn *fc = fuse_get_conn(file);
1032 if (!fc)
1033 return -EPERM;
1035 /* No locking - fasync_helper does its own locking */
1036 return fasync_helper(fd, file, on, &fc->fasync);
1039 const struct file_operations fuse_dev_operations = {
1040 .owner = THIS_MODULE,
1041 .llseek = no_llseek,
1042 .read = fuse_dev_read,
1043 .readv = fuse_dev_readv,
1044 .write = fuse_dev_write,
1045 .writev = fuse_dev_writev,
1046 .poll = fuse_dev_poll,
1047 .release = fuse_dev_release,
1048 .fasync = fuse_dev_fasync,
1049 };
1051 static struct miscdevice fuse_miscdevice = {
1052 .minor = FUSE_MINOR,
1053 .name = "fuse",
1054 .fops = &fuse_dev_operations,
1055 };
1057 int __init fuse_dev_init(void)
1059 int err = -ENOMEM;
1060 fuse_req_cachep = kmem_cache_create("fuse_request",
1061 sizeof(struct fuse_req),
1062 0, 0, NULL, NULL);
1063 if (!fuse_req_cachep)
1064 goto out;
1066 err = misc_register(&fuse_miscdevice);
1067 if (err)
1068 goto out_cache_clean;
1070 return 0;
1072 out_cache_clean:
1073 kmem_cache_destroy(fuse_req_cachep);
1074 out:
1075 return err;
1078 void fuse_dev_cleanup(void)
1080 misc_deregister(&fuse_miscdevice);
1081 kmem_cache_destroy(fuse_req_cachep);