aboutsummaryrefslogtreecommitdiffstats
path: root/fs/fuse
diff options
context:
space:
mode:
authorJeff Garzik <jgarzik@pobox.com>2005-09-14 08:19:08 -0400
committerJeff Garzik <jgarzik@pobox.com>2005-09-14 08:19:08 -0400
commit905ec87e93bc9e01b15c60035cd6a50c636cbaef (patch)
tree46fd7618d6511611ffc19eb0dd4d7bc6b90a41c2 /fs/fuse
parent1d6ae775d7a948c9575658eb41184fd2e506c0df (diff)
parent2f4ba45a75d6383b4a1201169a808ffea416ffa0 (diff)
Merge /spare/repo/linux-2.6/
Diffstat (limited to 'fs/fuse')
-rw-r--r--fs/fuse/Makefile7
-rw-r--r--fs/fuse/dev.c877
-rw-r--r--fs/fuse/dir.c982
-rw-r--r--fs/fuse/file.c555
-rw-r--r--fs/fuse/fuse_i.h451
-rw-r--r--fs/fuse/inode.c591
6 files changed, 3463 insertions, 0 deletions
diff --git a/fs/fuse/Makefile b/fs/fuse/Makefile
new file mode 100644
index 000000000000..c3e1f760cac9
--- /dev/null
+++ b/fs/fuse/Makefile
@@ -0,0 +1,7 @@
1#
2# Makefile for the FUSE filesystem.
3#
4
5obj-$(CONFIG_FUSE_FS) += fuse.o
6
7fuse-objs := dev.o dir.o file.o inode.o
diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
new file mode 100644
index 000000000000..d4c869c6d01b
--- /dev/null
+++ b/fs/fuse/dev.c
@@ -0,0 +1,877 @@
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2005 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/poll.h>
14#include <linux/uio.h>
15#include <linux/miscdevice.h>
16#include <linux/pagemap.h>
17#include <linux/file.h>
18#include <linux/slab.h>
19
20MODULE_ALIAS_MISCDEV(FUSE_MINOR);
21
22static kmem_cache_t *fuse_req_cachep;
23
24static inline struct fuse_conn *fuse_get_conn(struct file *file)
25{
26 struct fuse_conn *fc;
27 spin_lock(&fuse_lock);
28 fc = file->private_data;
29 if (fc && !fc->mounted)
30 fc = NULL;
31 spin_unlock(&fuse_lock);
32 return fc;
33}
34
35static inline void fuse_request_init(struct fuse_req *req)
36{
37 memset(req, 0, sizeof(*req));
38 INIT_LIST_HEAD(&req->list);
39 init_waitqueue_head(&req->waitq);
40 atomic_set(&req->count, 1);
41}
42
43struct fuse_req *fuse_request_alloc(void)
44{
45 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, SLAB_KERNEL);
46 if (req)
47 fuse_request_init(req);
48 return req;
49}
50
51void fuse_request_free(struct fuse_req *req)
52{
53 kmem_cache_free(fuse_req_cachep, req);
54}
55
56static inline void block_sigs(sigset_t *oldset)
57{
58 sigset_t mask;
59
60 siginitsetinv(&mask, sigmask(SIGKILL));
61 sigprocmask(SIG_BLOCK, &mask, oldset);
62}
63
64static inline void restore_sigs(sigset_t *oldset)
65{
66 sigprocmask(SIG_SETMASK, oldset, NULL);
67}
68
69void fuse_reset_request(struct fuse_req *req)
70{
71 int preallocated = req->preallocated;
72 BUG_ON(atomic_read(&req->count) != 1);
73 fuse_request_init(req);
74 req->preallocated = preallocated;
75}
76
77static void __fuse_get_request(struct fuse_req *req)
78{
79 atomic_inc(&req->count);
80}
81
82/* Must be called with > 1 refcount */
83static void __fuse_put_request(struct fuse_req *req)
84{
85 BUG_ON(atomic_read(&req->count) < 2);
86 atomic_dec(&req->count);
87}
88
89static struct fuse_req *do_get_request(struct fuse_conn *fc)
90{
91 struct fuse_req *req;
92
93 spin_lock(&fuse_lock);
94 BUG_ON(list_empty(&fc->unused_list));
95 req = list_entry(fc->unused_list.next, struct fuse_req, list);
96 list_del_init(&req->list);
97 spin_unlock(&fuse_lock);
98 fuse_request_init(req);
99 req->preallocated = 1;
100 req->in.h.uid = current->fsuid;
101 req->in.h.gid = current->fsgid;
102 req->in.h.pid = current->pid;
103 return req;
104}
105
106/* This can return NULL, but only in case it's interrupted by a SIGKILL */
107struct fuse_req *fuse_get_request(struct fuse_conn *fc)
108{
109 int intr;
110 sigset_t oldset;
111
112 block_sigs(&oldset);
113 intr = down_interruptible(&fc->outstanding_sem);
114 restore_sigs(&oldset);
115 return intr ? NULL : do_get_request(fc);
116}
117
118static void fuse_putback_request(struct fuse_conn *fc, struct fuse_req *req)
119{
120 spin_lock(&fuse_lock);
121 if (req->preallocated)
122 list_add(&req->list, &fc->unused_list);
123 else
124 fuse_request_free(req);
125
126 /* If we are in debt decrease that first */
127 if (fc->outstanding_debt)
128 fc->outstanding_debt--;
129 else
130 up(&fc->outstanding_sem);
131 spin_unlock(&fuse_lock);
132}
133
134void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
135{
136 if (atomic_dec_and_test(&req->count))
137 fuse_putback_request(fc, req);
138}
139
140void fuse_release_background(struct fuse_req *req)
141{
142 iput(req->inode);
143 iput(req->inode2);
144 if (req->file)
145 fput(req->file);
146 spin_lock(&fuse_lock);
147 list_del(&req->bg_entry);
148 spin_unlock(&fuse_lock);
149}
150
151/*
152 * This function is called when a request is finished. Either a reply
153 * has arrived or it was interrupted (and not yet sent) or some error
154 * occured during communication with userspace, or the device file was
155 * closed. It decreases the referece count for the request. In case
156 * of a background request the referece to the stored objects are
157 * released. The requester thread is woken up (if still waiting), and
158 * finally the request is either freed or put on the unused_list
159 *
160 * Called with fuse_lock, unlocks it
161 */
162static void request_end(struct fuse_conn *fc, struct fuse_req *req)
163{
164 int putback;
165 req->finished = 1;
166 putback = atomic_dec_and_test(&req->count);
167 spin_unlock(&fuse_lock);
168 if (req->background) {
169 down_read(&fc->sbput_sem);
170 if (fc->mounted)
171 fuse_release_background(req);
172 up_read(&fc->sbput_sem);
173 }
174 wake_up(&req->waitq);
175 if (req->in.h.opcode == FUSE_INIT) {
176 int i;
177
178 if (req->misc.init_in_out.major != FUSE_KERNEL_VERSION)
179 fc->conn_error = 1;
180
181 /* After INIT reply is received other requests can go
182 out. So do (FUSE_MAX_OUTSTANDING - 1) number of
183 up()s on outstanding_sem. The last up() is done in
184 fuse_putback_request() */
185 for (i = 1; i < FUSE_MAX_OUTSTANDING; i++)
186 up(&fc->outstanding_sem);
187 }
188 if (putback)
189 fuse_putback_request(fc, req);
190}
191
192/*
193 * Unfortunately request interruption not just solves the deadlock
194 * problem, it causes problems too. These stem from the fact, that an
195 * interrupted request is continued to be processed in userspace,
196 * while all the locks and object references (inode and file) held
197 * during the operation are released.
198 *
199 * To release the locks is exactly why there's a need to interrupt the
200 * request, so there's not a lot that can be done about this, except
201 * introduce additional locking in userspace.
202 *
203 * More important is to keep inode and file references until userspace
204 * has replied, otherwise FORGET and RELEASE could be sent while the
205 * inode/file is still used by the filesystem.
206 *
207 * For this reason the concept of "background" request is introduced.
208 * An interrupted request is backgrounded if it has been already sent
209 * to userspace. Backgrounding involves getting an extra reference to
210 * inode(s) or file used in the request, and adding the request to
211 * fc->background list. When a reply is received for a background
212 * request, the object references are released, and the request is
213 * removed from the list. If the filesystem is unmounted while there
214 * are still background requests, the list is walked and references
215 * are released as if a reply was received.
216 *
217 * There's one more use for a background request. The RELEASE message is
218 * always sent as background, since it doesn't return an error or
219 * data.
220 */
221static void background_request(struct fuse_conn *fc, struct fuse_req *req)
222{
223 req->background = 1;
224 list_add(&req->bg_entry, &fc->background);
225 if (req->inode)
226 req->inode = igrab(req->inode);
227 if (req->inode2)
228 req->inode2 = igrab(req->inode2);
229 if (req->file)
230 get_file(req->file);
231}
232
233/* Called with fuse_lock held. Releases, and then reacquires it. */
234static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
235{
236 sigset_t oldset;
237
238 spin_unlock(&fuse_lock);
239 block_sigs(&oldset);
240 wait_event_interruptible(req->waitq, req->finished);
241 restore_sigs(&oldset);
242 spin_lock(&fuse_lock);
243 if (req->finished)
244 return;
245
246 req->out.h.error = -EINTR;
247 req->interrupted = 1;
248 if (req->locked) {
249 /* This is uninterruptible sleep, because data is
250 being copied to/from the buffers of req. During
251 locked state, there mustn't be any filesystem
252 operation (e.g. page fault), since that could lead
253 to deadlock */
254 spin_unlock(&fuse_lock);
255 wait_event(req->waitq, !req->locked);
256 spin_lock(&fuse_lock);
257 }
258 if (!req->sent && !list_empty(&req->list)) {
259 list_del(&req->list);
260 __fuse_put_request(req);
261 } else if (!req->finished && req->sent)
262 background_request(fc, req);
263}
264
265static unsigned len_args(unsigned numargs, struct fuse_arg *args)
266{
267 unsigned nbytes = 0;
268 unsigned i;
269
270 for (i = 0; i < numargs; i++)
271 nbytes += args[i].size;
272
273 return nbytes;
274}
275
276static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
277{
278 fc->reqctr++;
279 /* zero is special */
280 if (fc->reqctr == 0)
281 fc->reqctr = 1;
282 req->in.h.unique = fc->reqctr;
283 req->in.h.len = sizeof(struct fuse_in_header) +
284 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
285 if (!req->preallocated) {
286 /* If request is not preallocated (either FORGET or
287 RELEASE), then still decrease outstanding_sem, so
288 user can't open infinite number of files while not
289 processing the RELEASE requests. However for
290 efficiency do it without blocking, so if down()
291 would block, just increase the debt instead */
292 if (down_trylock(&fc->outstanding_sem))
293 fc->outstanding_debt++;
294 }
295 list_add_tail(&req->list, &fc->pending);
296 wake_up(&fc->waitq);
297}
298
299/*
300 * This can only be interrupted by a SIGKILL
301 */
302void request_send(struct fuse_conn *fc, struct fuse_req *req)
303{
304 req->isreply = 1;
305 spin_lock(&fuse_lock);
306 if (!fc->connected)
307 req->out.h.error = -ENOTCONN;
308 else if (fc->conn_error)
309 req->out.h.error = -ECONNREFUSED;
310 else {
311 queue_request(fc, req);
312 /* acquire extra reference, since request is still needed
313 after request_end() */
314 __fuse_get_request(req);
315
316 request_wait_answer(fc, req);
317 }
318 spin_unlock(&fuse_lock);
319}
320
321static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
322{
323 spin_lock(&fuse_lock);
324 if (fc->connected) {
325 queue_request(fc, req);
326 spin_unlock(&fuse_lock);
327 } else {
328 req->out.h.error = -ENOTCONN;
329 request_end(fc, req);
330 }
331}
332
333void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req)
334{
335 req->isreply = 0;
336 request_send_nowait(fc, req);
337}
338
339void request_send_background(struct fuse_conn *fc, struct fuse_req *req)
340{
341 req->isreply = 1;
342 spin_lock(&fuse_lock);
343 background_request(fc, req);
344 spin_unlock(&fuse_lock);
345 request_send_nowait(fc, req);
346}
347
348void fuse_send_init(struct fuse_conn *fc)
349{
350 /* This is called from fuse_read_super() so there's guaranteed
351 to be a request available */
352 struct fuse_req *req = do_get_request(fc);
353 struct fuse_init_in_out *arg = &req->misc.init_in_out;
354 arg->major = FUSE_KERNEL_VERSION;
355 arg->minor = FUSE_KERNEL_MINOR_VERSION;
356 req->in.h.opcode = FUSE_INIT;
357 req->in.numargs = 1;
358 req->in.args[0].size = sizeof(*arg);
359 req->in.args[0].value = arg;
360 req->out.numargs = 1;
361 req->out.args[0].size = sizeof(*arg);
362 req->out.args[0].value = arg;
363 request_send_background(fc, req);
364}
365
366/*
367 * Lock the request. Up to the next unlock_request() there mustn't be
368 * anything that could cause a page-fault. If the request was already
369 * interrupted bail out.
370 */
371static inline int lock_request(struct fuse_req *req)
372{
373 int err = 0;
374 if (req) {
375 spin_lock(&fuse_lock);
376 if (req->interrupted)
377 err = -ENOENT;
378 else
379 req->locked = 1;
380 spin_unlock(&fuse_lock);
381 }
382 return err;
383}
384
385/*
386 * Unlock request. If it was interrupted during being locked, the
387 * requester thread is currently waiting for it to be unlocked, so
388 * wake it up.
389 */
390static inline void unlock_request(struct fuse_req *req)
391{
392 if (req) {
393 spin_lock(&fuse_lock);
394 req->locked = 0;
395 if (req->interrupted)
396 wake_up(&req->waitq);
397 spin_unlock(&fuse_lock);
398 }
399}
400
401struct fuse_copy_state {
402 int write;
403 struct fuse_req *req;
404 const struct iovec *iov;
405 unsigned long nr_segs;
406 unsigned long seglen;
407 unsigned long addr;
408 struct page *pg;
409 void *mapaddr;
410 void *buf;
411 unsigned len;
412};
413
414static void fuse_copy_init(struct fuse_copy_state *cs, int write,
415 struct fuse_req *req, const struct iovec *iov,
416 unsigned long nr_segs)
417{
418 memset(cs, 0, sizeof(*cs));
419 cs->write = write;
420 cs->req = req;
421 cs->iov = iov;
422 cs->nr_segs = nr_segs;
423}
424
425/* Unmap and put previous page of userspace buffer */
426static inline void fuse_copy_finish(struct fuse_copy_state *cs)
427{
428 if (cs->mapaddr) {
429 kunmap_atomic(cs->mapaddr, KM_USER0);
430 if (cs->write) {
431 flush_dcache_page(cs->pg);
432 set_page_dirty_lock(cs->pg);
433 }
434 put_page(cs->pg);
435 cs->mapaddr = NULL;
436 }
437}
438
439/*
440 * Get another pagefull of userspace buffer, and map it to kernel
441 * address space, and lock request
442 */
443static int fuse_copy_fill(struct fuse_copy_state *cs)
444{
445 unsigned long offset;
446 int err;
447
448 unlock_request(cs->req);
449 fuse_copy_finish(cs);
450 if (!cs->seglen) {
451 BUG_ON(!cs->nr_segs);
452 cs->seglen = cs->iov[0].iov_len;
453 cs->addr = (unsigned long) cs->iov[0].iov_base;
454 cs->iov ++;
455 cs->nr_segs --;
456 }
457 down_read(&current->mm->mmap_sem);
458 err = get_user_pages(current, current->mm, cs->addr, 1, cs->write, 0,
459 &cs->pg, NULL);
460 up_read(&current->mm->mmap_sem);
461 if (err < 0)
462 return err;
463 BUG_ON(err != 1);
464 offset = cs->addr % PAGE_SIZE;
465 cs->mapaddr = kmap_atomic(cs->pg, KM_USER0);
466 cs->buf = cs->mapaddr + offset;
467 cs->len = min(PAGE_SIZE - offset, cs->seglen);
468 cs->seglen -= cs->len;
469 cs->addr += cs->len;
470
471 return lock_request(cs->req);
472}
473
474/* Do as much copy to/from userspace buffer as we can */
475static inline int fuse_copy_do(struct fuse_copy_state *cs, void **val,
476 unsigned *size)
477{
478 unsigned ncpy = min(*size, cs->len);
479 if (val) {
480 if (cs->write)
481 memcpy(cs->buf, *val, ncpy);
482 else
483 memcpy(*val, cs->buf, ncpy);
484 *val += ncpy;
485 }
486 *size -= ncpy;
487 cs->len -= ncpy;
488 cs->buf += ncpy;
489 return ncpy;
490}
491
492/*
493 * Copy a page in the request to/from the userspace buffer. Must be
494 * done atomically
495 */
496static inline int fuse_copy_page(struct fuse_copy_state *cs, struct page *page,
497 unsigned offset, unsigned count, int zeroing)
498{
499 if (page && zeroing && count < PAGE_SIZE) {
500 void *mapaddr = kmap_atomic(page, KM_USER1);
501 memset(mapaddr, 0, PAGE_SIZE);
502 kunmap_atomic(mapaddr, KM_USER1);
503 }
504 while (count) {
505 int err;
506 if (!cs->len && (err = fuse_copy_fill(cs)))
507 return err;
508 if (page) {
509 void *mapaddr = kmap_atomic(page, KM_USER1);
510 void *buf = mapaddr + offset;
511 offset += fuse_copy_do(cs, &buf, &count);
512 kunmap_atomic(mapaddr, KM_USER1);
513 } else
514 offset += fuse_copy_do(cs, NULL, &count);
515 }
516 if (page && !cs->write)
517 flush_dcache_page(page);
518 return 0;
519}
520
521/* Copy pages in the request to/from userspace buffer */
522static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
523 int zeroing)
524{
525 unsigned i;
526 struct fuse_req *req = cs->req;
527 unsigned offset = req->page_offset;
528 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
529
530 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
531 struct page *page = req->pages[i];
532 int err = fuse_copy_page(cs, page, offset, count, zeroing);
533 if (err)
534 return err;
535
536 nbytes -= count;
537 count = min(nbytes, (unsigned) PAGE_SIZE);
538 offset = 0;
539 }
540 return 0;
541}
542
543/* Copy a single argument in the request to/from userspace buffer */
544static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
545{
546 while (size) {
547 int err;
548 if (!cs->len && (err = fuse_copy_fill(cs)))
549 return err;
550 fuse_copy_do(cs, &val, &size);
551 }
552 return 0;
553}
554
555/* Copy request arguments to/from userspace buffer */
556static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
557 unsigned argpages, struct fuse_arg *args,
558 int zeroing)
559{
560 int err = 0;
561 unsigned i;
562
563 for (i = 0; !err && i < numargs; i++) {
564 struct fuse_arg *arg = &args[i];
565 if (i == numargs - 1 && argpages)
566 err = fuse_copy_pages(cs, arg->size, zeroing);
567 else
568 err = fuse_copy_one(cs, arg->value, arg->size);
569 }
570 return err;
571}
572
573/* Wait until a request is available on the pending list */
574static void request_wait(struct fuse_conn *fc)
575{
576 DECLARE_WAITQUEUE(wait, current);
577
578 add_wait_queue_exclusive(&fc->waitq, &wait);
579 while (fc->mounted && list_empty(&fc->pending)) {
580 set_current_state(TASK_INTERRUPTIBLE);
581 if (signal_pending(current))
582 break;
583
584 spin_unlock(&fuse_lock);
585 schedule();
586 spin_lock(&fuse_lock);
587 }
588 set_current_state(TASK_RUNNING);
589 remove_wait_queue(&fc->waitq, &wait);
590}
591
592/*
593 * Read a single request into the userspace filesystem's buffer. This
594 * function waits until a request is available, then removes it from
595 * the pending list and copies request data to userspace buffer. If
596 * no reply is needed (FORGET) or request has been interrupted or
597 * there was an error during the copying then it's finished by calling
598 * request_end(). Otherwise add it to the processing list, and set
599 * the 'sent' flag.
600 */
601static ssize_t fuse_dev_readv(struct file *file, const struct iovec *iov,
602 unsigned long nr_segs, loff_t *off)
603{
604 int err;
605 struct fuse_conn *fc;
606 struct fuse_req *req;
607 struct fuse_in *in;
608 struct fuse_copy_state cs;
609 unsigned reqsize;
610
611 spin_lock(&fuse_lock);
612 fc = file->private_data;
613 err = -EPERM;
614 if (!fc)
615 goto err_unlock;
616 request_wait(fc);
617 err = -ENODEV;
618 if (!fc->mounted)
619 goto err_unlock;
620 err = -ERESTARTSYS;
621 if (list_empty(&fc->pending))
622 goto err_unlock;
623
624 req = list_entry(fc->pending.next, struct fuse_req, list);
625 list_del_init(&req->list);
626 spin_unlock(&fuse_lock);
627
628 in = &req->in;
629 reqsize = req->in.h.len;
630 fuse_copy_init(&cs, 1, req, iov, nr_segs);
631 err = -EINVAL;
632 if (iov_length(iov, nr_segs) >= reqsize) {
633 err = fuse_copy_one(&cs, &in->h, sizeof(in->h));
634 if (!err)
635 err = fuse_copy_args(&cs, in->numargs, in->argpages,
636 (struct fuse_arg *) in->args, 0);
637 }
638 fuse_copy_finish(&cs);
639
640 spin_lock(&fuse_lock);
641 req->locked = 0;
642 if (!err && req->interrupted)
643 err = -ENOENT;
644 if (err) {
645 if (!req->interrupted)
646 req->out.h.error = -EIO;
647 request_end(fc, req);
648 return err;
649 }
650 if (!req->isreply)
651 request_end(fc, req);
652 else {
653 req->sent = 1;
654 list_add_tail(&req->list, &fc->processing);
655 spin_unlock(&fuse_lock);
656 }
657 return reqsize;
658
659 err_unlock:
660 spin_unlock(&fuse_lock);
661 return err;
662}
663
664static ssize_t fuse_dev_read(struct file *file, char __user *buf,
665 size_t nbytes, loff_t *off)
666{
667 struct iovec iov;
668 iov.iov_len = nbytes;
669 iov.iov_base = buf;
670 return fuse_dev_readv(file, &iov, 1, off);
671}
672
673/* Look up request on processing list by unique ID */
674static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
675{
676 struct list_head *entry;
677
678 list_for_each(entry, &fc->processing) {
679 struct fuse_req *req;
680 req = list_entry(entry, struct fuse_req, list);
681 if (req->in.h.unique == unique)
682 return req;
683 }
684 return NULL;
685}
686
687static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
688 unsigned nbytes)
689{
690 unsigned reqsize = sizeof(struct fuse_out_header);
691
692 if (out->h.error)
693 return nbytes != reqsize ? -EINVAL : 0;
694
695 reqsize += len_args(out->numargs, out->args);
696
697 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
698 return -EINVAL;
699 else if (reqsize > nbytes) {
700 struct fuse_arg *lastarg = &out->args[out->numargs-1];
701 unsigned diffsize = reqsize - nbytes;
702 if (diffsize > lastarg->size)
703 return -EINVAL;
704 lastarg->size -= diffsize;
705 }
706 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
707 out->page_zeroing);
708}
709
710/*
711 * Write a single reply to a request. First the header is copied from
712 * the write buffer. The request is then searched on the processing
713 * list by the unique ID found in the header. If found, then remove
714 * it from the list and copy the rest of the buffer to the request.
715 * The request is finished by calling request_end()
716 */
717static ssize_t fuse_dev_writev(struct file *file, const struct iovec *iov,
718 unsigned long nr_segs, loff_t *off)
719{
720 int err;
721 unsigned nbytes = iov_length(iov, nr_segs);
722 struct fuse_req *req;
723 struct fuse_out_header oh;
724 struct fuse_copy_state cs;
725 struct fuse_conn *fc = fuse_get_conn(file);
726 if (!fc)
727 return -ENODEV;
728
729 fuse_copy_init(&cs, 0, NULL, iov, nr_segs);
730 if (nbytes < sizeof(struct fuse_out_header))
731 return -EINVAL;
732
733 err = fuse_copy_one(&cs, &oh, sizeof(oh));
734 if (err)
735 goto err_finish;
736 err = -EINVAL;
737 if (!oh.unique || oh.error <= -1000 || oh.error > 0 ||
738 oh.len != nbytes)
739 goto err_finish;
740
741 spin_lock(&fuse_lock);
742 req = request_find(fc, oh.unique);
743 err = -EINVAL;
744 if (!req)
745 goto err_unlock;
746
747 list_del_init(&req->list);
748 if (req->interrupted) {
749 request_end(fc, req);
750 fuse_copy_finish(&cs);
751 return -ENOENT;
752 }
753 req->out.h = oh;
754 req->locked = 1;
755 cs.req = req;
756 spin_unlock(&fuse_lock);
757
758 err = copy_out_args(&cs, &req->out, nbytes);
759 fuse_copy_finish(&cs);
760
761 spin_lock(&fuse_lock);
762 req->locked = 0;
763 if (!err) {
764 if (req->interrupted)
765 err = -ENOENT;
766 } else if (!req->interrupted)
767 req->out.h.error = -EIO;
768 request_end(fc, req);
769
770 return err ? err : nbytes;
771
772 err_unlock:
773 spin_unlock(&fuse_lock);
774 err_finish:
775 fuse_copy_finish(&cs);
776 return err;
777}
778
779static ssize_t fuse_dev_write(struct file *file, const char __user *buf,
780 size_t nbytes, loff_t *off)
781{
782 struct iovec iov;
783 iov.iov_len = nbytes;
784 iov.iov_base = (char __user *) buf;
785 return fuse_dev_writev(file, &iov, 1, off);
786}
787
788static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
789{
790 struct fuse_conn *fc = fuse_get_conn(file);
791 unsigned mask = POLLOUT | POLLWRNORM;
792
793 if (!fc)
794 return -ENODEV;
795
796 poll_wait(file, &fc->waitq, wait);
797
798 spin_lock(&fuse_lock);
799 if (!list_empty(&fc->pending))
800 mask |= POLLIN | POLLRDNORM;
801 spin_unlock(&fuse_lock);
802
803 return mask;
804}
805
806/* Abort all requests on the given list (pending or processing) */
807static void end_requests(struct fuse_conn *fc, struct list_head *head)
808{
809 while (!list_empty(head)) {
810 struct fuse_req *req;
811 req = list_entry(head->next, struct fuse_req, list);
812 list_del_init(&req->list);
813 req->out.h.error = -ECONNABORTED;
814 request_end(fc, req);
815 spin_lock(&fuse_lock);
816 }
817}
818
819static int fuse_dev_release(struct inode *inode, struct file *file)
820{
821 struct fuse_conn *fc;
822
823 spin_lock(&fuse_lock);
824 fc = file->private_data;
825 if (fc) {
826 fc->connected = 0;
827 end_requests(fc, &fc->pending);
828 end_requests(fc, &fc->processing);
829 fuse_release_conn(fc);
830 }
831 spin_unlock(&fuse_lock);
832 return 0;
833}
834
835struct file_operations fuse_dev_operations = {
836 .owner = THIS_MODULE,
837 .llseek = no_llseek,
838 .read = fuse_dev_read,
839 .readv = fuse_dev_readv,
840 .write = fuse_dev_write,
841 .writev = fuse_dev_writev,
842 .poll = fuse_dev_poll,
843 .release = fuse_dev_release,
844};
845
846static struct miscdevice fuse_miscdevice = {
847 .minor = FUSE_MINOR,
848 .name = "fuse",
849 .fops = &fuse_dev_operations,
850};
851
852int __init fuse_dev_init(void)
853{
854 int err = -ENOMEM;
855 fuse_req_cachep = kmem_cache_create("fuse_request",
856 sizeof(struct fuse_req),
857 0, 0, NULL, NULL);
858 if (!fuse_req_cachep)
859 goto out;
860
861 err = misc_register(&fuse_miscdevice);
862 if (err)
863 goto out_cache_clean;
864
865 return 0;
866
867 out_cache_clean:
868 kmem_cache_destroy(fuse_req_cachep);
869 out:
870 return err;
871}
872
873void fuse_dev_cleanup(void)
874{
875 misc_deregister(&fuse_miscdevice);
876 kmem_cache_destroy(fuse_req_cachep);
877}
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
new file mode 100644
index 000000000000..e79e49b3eec7
--- /dev/null
+++ b/fs/fuse/dir.c
@@ -0,0 +1,982 @@
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2005 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/pagemap.h>
12#include <linux/file.h>
13#include <linux/gfp.h>
14#include <linux/sched.h>
15#include <linux/namei.h>
16
17static inline unsigned long time_to_jiffies(unsigned long sec,
18 unsigned long nsec)
19{
20 struct timespec ts = {sec, nsec};
21 return jiffies + timespec_to_jiffies(&ts);
22}
23
24static void fuse_lookup_init(struct fuse_req *req, struct inode *dir,
25 struct dentry *entry,
26 struct fuse_entry_out *outarg)
27{
28 req->in.h.opcode = FUSE_LOOKUP;
29 req->in.h.nodeid = get_node_id(dir);
30 req->inode = dir;
31 req->in.numargs = 1;
32 req->in.args[0].size = entry->d_name.len + 1;
33 req->in.args[0].value = entry->d_name.name;
34 req->out.numargs = 1;
35 req->out.args[0].size = sizeof(struct fuse_entry_out);
36 req->out.args[0].value = outarg;
37}
38
39static int fuse_dentry_revalidate(struct dentry *entry, struct nameidata *nd)
40{
41 if (!entry->d_inode || is_bad_inode(entry->d_inode))
42 return 0;
43 else if (time_after(jiffies, entry->d_time)) {
44 int err;
45 struct fuse_entry_out outarg;
46 struct inode *inode = entry->d_inode;
47 struct fuse_inode *fi = get_fuse_inode(inode);
48 struct fuse_conn *fc = get_fuse_conn(inode);
49 struct fuse_req *req = fuse_get_request(fc);
50 if (!req)
51 return 0;
52
53 fuse_lookup_init(req, entry->d_parent->d_inode, entry, &outarg);
54 request_send(fc, req);
55 err = req->out.h.error;
56 if (!err) {
57 if (outarg.nodeid != get_node_id(inode)) {
58 fuse_send_forget(fc, req, outarg.nodeid, 1);
59 return 0;
60 }
61 fi->nlookup ++;
62 }
63 fuse_put_request(fc, req);
64 if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
65 return 0;
66
67 fuse_change_attributes(inode, &outarg.attr);
68 entry->d_time = time_to_jiffies(outarg.entry_valid,
69 outarg.entry_valid_nsec);
70 fi->i_time = time_to_jiffies(outarg.attr_valid,
71 outarg.attr_valid_nsec);
72 }
73 return 1;
74}
75
76static struct dentry_operations fuse_dentry_operations = {
77 .d_revalidate = fuse_dentry_revalidate,
78};
79
80static int fuse_lookup_iget(struct inode *dir, struct dentry *entry,
81 struct inode **inodep)
82{
83 int err;
84 struct fuse_entry_out outarg;
85 struct inode *inode = NULL;
86 struct fuse_conn *fc = get_fuse_conn(dir);
87 struct fuse_req *req;
88
89 if (entry->d_name.len > FUSE_NAME_MAX)
90 return -ENAMETOOLONG;
91
92 req = fuse_get_request(fc);
93 if (!req)
94 return -EINTR;
95
96 fuse_lookup_init(req, dir, entry, &outarg);
97 request_send(fc, req);
98 err = req->out.h.error;
99 if (!err) {
100 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
101 &outarg.attr);
102 if (!inode) {
103 fuse_send_forget(fc, req, outarg.nodeid, 1);
104 return -ENOMEM;
105 }
106 }
107 fuse_put_request(fc, req);
108 if (err && err != -ENOENT)
109 return err;
110
111 if (inode) {
112 struct fuse_inode *fi = get_fuse_inode(inode);
113 entry->d_time = time_to_jiffies(outarg.entry_valid,
114 outarg.entry_valid_nsec);
115 fi->i_time = time_to_jiffies(outarg.attr_valid,
116 outarg.attr_valid_nsec);
117 }
118
119 entry->d_op = &fuse_dentry_operations;
120 *inodep = inode;
121 return 0;
122}
123
124void fuse_invalidate_attr(struct inode *inode)
125{
126 get_fuse_inode(inode)->i_time = jiffies - 1;
127}
128
129static void fuse_invalidate_entry(struct dentry *entry)
130{
131 d_invalidate(entry);
132 entry->d_time = jiffies - 1;
133}
134
135static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
136 struct inode *dir, struct dentry *entry,
137 int mode)
138{
139 struct fuse_entry_out outarg;
140 struct inode *inode;
141 struct fuse_inode *fi;
142 int err;
143
144 req->in.h.nodeid = get_node_id(dir);
145 req->inode = dir;
146 req->out.numargs = 1;
147 req->out.args[0].size = sizeof(outarg);
148 req->out.args[0].value = &outarg;
149 request_send(fc, req);
150 err = req->out.h.error;
151 if (err) {
152 fuse_put_request(fc, req);
153 return err;
154 }
155 inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
156 &outarg.attr);
157 if (!inode) {
158 fuse_send_forget(fc, req, outarg.nodeid, 1);
159 return -ENOMEM;
160 }
161 fuse_put_request(fc, req);
162
163 /* Don't allow userspace to do really stupid things... */
164 if ((inode->i_mode ^ mode) & S_IFMT) {
165 iput(inode);
166 return -EIO;
167 }
168
169 entry->d_time = time_to_jiffies(outarg.entry_valid,
170 outarg.entry_valid_nsec);
171
172 fi = get_fuse_inode(inode);
173 fi->i_time = time_to_jiffies(outarg.attr_valid,
174 outarg.attr_valid_nsec);
175
176 d_instantiate(entry, inode);
177 fuse_invalidate_attr(dir);
178 return 0;
179}
180
181static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
182 dev_t rdev)
183{
184 struct fuse_mknod_in inarg;
185 struct fuse_conn *fc = get_fuse_conn(dir);
186 struct fuse_req *req = fuse_get_request(fc);
187 if (!req)
188 return -EINTR;
189
190 memset(&inarg, 0, sizeof(inarg));
191 inarg.mode = mode;
192 inarg.rdev = new_encode_dev(rdev);
193 req->in.h.opcode = FUSE_MKNOD;
194 req->in.numargs = 2;
195 req->in.args[0].size = sizeof(inarg);
196 req->in.args[0].value = &inarg;
197 req->in.args[1].size = entry->d_name.len + 1;
198 req->in.args[1].value = entry->d_name.name;
199 return create_new_entry(fc, req, dir, entry, mode);
200}
201
202static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
203 struct nameidata *nd)
204{
205 return fuse_mknod(dir, entry, mode, 0);
206}
207
208static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
209{
210 struct fuse_mkdir_in inarg;
211 struct fuse_conn *fc = get_fuse_conn(dir);
212 struct fuse_req *req = fuse_get_request(fc);
213 if (!req)
214 return -EINTR;
215
216 memset(&inarg, 0, sizeof(inarg));
217 inarg.mode = mode;
218 req->in.h.opcode = FUSE_MKDIR;
219 req->in.numargs = 2;
220 req->in.args[0].size = sizeof(inarg);
221 req->in.args[0].value = &inarg;
222 req->in.args[1].size = entry->d_name.len + 1;
223 req->in.args[1].value = entry->d_name.name;
224 return create_new_entry(fc, req, dir, entry, S_IFDIR);
225}
226
227static int fuse_symlink(struct inode *dir, struct dentry *entry,
228 const char *link)
229{
230 struct fuse_conn *fc = get_fuse_conn(dir);
231 unsigned len = strlen(link) + 1;
232 struct fuse_req *req;
233
234 if (len > FUSE_SYMLINK_MAX)
235 return -ENAMETOOLONG;
236
237 req = fuse_get_request(fc);
238 if (!req)
239 return -EINTR;
240
241 req->in.h.opcode = FUSE_SYMLINK;
242 req->in.numargs = 2;
243 req->in.args[0].size = entry->d_name.len + 1;
244 req->in.args[0].value = entry->d_name.name;
245 req->in.args[1].size = len;
246 req->in.args[1].value = link;
247 return create_new_entry(fc, req, dir, entry, S_IFLNK);
248}
249
250static int fuse_unlink(struct inode *dir, struct dentry *entry)
251{
252 int err;
253 struct fuse_conn *fc = get_fuse_conn(dir);
254 struct fuse_req *req = fuse_get_request(fc);
255 if (!req)
256 return -EINTR;
257
258 req->in.h.opcode = FUSE_UNLINK;
259 req->in.h.nodeid = get_node_id(dir);
260 req->inode = dir;
261 req->in.numargs = 1;
262 req->in.args[0].size = entry->d_name.len + 1;
263 req->in.args[0].value = entry->d_name.name;
264 request_send(fc, req);
265 err = req->out.h.error;
266 fuse_put_request(fc, req);
267 if (!err) {
268 struct inode *inode = entry->d_inode;
269
270 /* Set nlink to zero so the inode can be cleared, if
271 the inode does have more links this will be
272 discovered at the next lookup/getattr */
273 inode->i_nlink = 0;
274 fuse_invalidate_attr(inode);
275 fuse_invalidate_attr(dir);
276 } else if (err == -EINTR)
277 fuse_invalidate_entry(entry);
278 return err;
279}
280
281static int fuse_rmdir(struct inode *dir, struct dentry *entry)
282{
283 int err;
284 struct fuse_conn *fc = get_fuse_conn(dir);
285 struct fuse_req *req = fuse_get_request(fc);
286 if (!req)
287 return -EINTR;
288
289 req->in.h.opcode = FUSE_RMDIR;
290 req->in.h.nodeid = get_node_id(dir);
291 req->inode = dir;
292 req->in.numargs = 1;
293 req->in.args[0].size = entry->d_name.len + 1;
294 req->in.args[0].value = entry->d_name.name;
295 request_send(fc, req);
296 err = req->out.h.error;
297 fuse_put_request(fc, req);
298 if (!err) {
299 entry->d_inode->i_nlink = 0;
300 fuse_invalidate_attr(dir);
301 } else if (err == -EINTR)
302 fuse_invalidate_entry(entry);
303 return err;
304}
305
306static int fuse_rename(struct inode *olddir, struct dentry *oldent,
307 struct inode *newdir, struct dentry *newent)
308{
309 int err;
310 struct fuse_rename_in inarg;
311 struct fuse_conn *fc = get_fuse_conn(olddir);
312 struct fuse_req *req = fuse_get_request(fc);
313 if (!req)
314 return -EINTR;
315
316 memset(&inarg, 0, sizeof(inarg));
317 inarg.newdir = get_node_id(newdir);
318 req->in.h.opcode = FUSE_RENAME;
319 req->in.h.nodeid = get_node_id(olddir);
320 req->inode = olddir;
321 req->inode2 = newdir;
322 req->in.numargs = 3;
323 req->in.args[0].size = sizeof(inarg);
324 req->in.args[0].value = &inarg;
325 req->in.args[1].size = oldent->d_name.len + 1;
326 req->in.args[1].value = oldent->d_name.name;
327 req->in.args[2].size = newent->d_name.len + 1;
328 req->in.args[2].value = newent->d_name.name;
329 request_send(fc, req);
330 err = req->out.h.error;
331 fuse_put_request(fc, req);
332 if (!err) {
333 fuse_invalidate_attr(olddir);
334 if (olddir != newdir)
335 fuse_invalidate_attr(newdir);
336 } else if (err == -EINTR) {
337 /* If request was interrupted, DEITY only knows if the
338 rename actually took place. If the invalidation
339 fails (e.g. some process has CWD under the renamed
340 directory), then there can be inconsistency between
341 the dcache and the real filesystem. Tough luck. */
342 fuse_invalidate_entry(oldent);
343 if (newent->d_inode)
344 fuse_invalidate_entry(newent);
345 }
346
347 return err;
348}
349
350static int fuse_link(struct dentry *entry, struct inode *newdir,
351 struct dentry *newent)
352{
353 int err;
354 struct fuse_link_in inarg;
355 struct inode *inode = entry->d_inode;
356 struct fuse_conn *fc = get_fuse_conn(inode);
357 struct fuse_req *req = fuse_get_request(fc);
358 if (!req)
359 return -EINTR;
360
361 memset(&inarg, 0, sizeof(inarg));
362 inarg.oldnodeid = get_node_id(inode);
363 req->in.h.opcode = FUSE_LINK;
364 req->inode2 = inode;
365 req->in.numargs = 2;
366 req->in.args[0].size = sizeof(inarg);
367 req->in.args[0].value = &inarg;
368 req->in.args[1].size = newent->d_name.len + 1;
369 req->in.args[1].value = newent->d_name.name;
370 err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
371 /* Contrary to "normal" filesystems it can happen that link
372 makes two "logical" inodes point to the same "physical"
373 inode. We invalidate the attributes of the old one, so it
374 will reflect changes in the backing inode (link count,
375 etc.)
376 */
377 if (!err || err == -EINTR)
378 fuse_invalidate_attr(inode);
379 return err;
380}
381
382int fuse_do_getattr(struct inode *inode)
383{
384 int err;
385 struct fuse_attr_out arg;
386 struct fuse_conn *fc = get_fuse_conn(inode);
387 struct fuse_req *req = fuse_get_request(fc);
388 if (!req)
389 return -EINTR;
390
391 req->in.h.opcode = FUSE_GETATTR;
392 req->in.h.nodeid = get_node_id(inode);
393 req->inode = inode;
394 req->out.numargs = 1;
395 req->out.args[0].size = sizeof(arg);
396 req->out.args[0].value = &arg;
397 request_send(fc, req);
398 err = req->out.h.error;
399 fuse_put_request(fc, req);
400 if (!err) {
401 if ((inode->i_mode ^ arg.attr.mode) & S_IFMT) {
402 make_bad_inode(inode);
403 err = -EIO;
404 } else {
405 struct fuse_inode *fi = get_fuse_inode(inode);
406 fuse_change_attributes(inode, &arg.attr);
407 fi->i_time = time_to_jiffies(arg.attr_valid,
408 arg.attr_valid_nsec);
409 }
410 }
411 return err;
412}
413
414/*
415 * Calling into a user-controlled filesystem gives the filesystem
416 * daemon ptrace-like capabilities over the requester process. This
417 * means, that the filesystem daemon is able to record the exact
418 * filesystem operations performed, and can also control the behavior
419 * of the requester process in otherwise impossible ways. For example
420 * it can delay the operation for arbitrary length of time allowing
421 * DoS against the requester.
422 *
423 * For this reason only those processes can call into the filesystem,
424 * for which the owner of the mount has ptrace privilege. This
425 * excludes processes started by other users, suid or sgid processes.
426 */
427static int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
428{
429 if (fc->flags & FUSE_ALLOW_OTHER)
430 return 1;
431
432 if (task->euid == fc->user_id &&
433 task->suid == fc->user_id &&
434 task->uid == fc->user_id &&
435 task->egid == fc->group_id &&
436 task->sgid == fc->group_id &&
437 task->gid == fc->group_id)
438 return 1;
439
440 return 0;
441}
442
443static int fuse_revalidate(struct dentry *entry)
444{
445 struct inode *inode = entry->d_inode;
446 struct fuse_inode *fi = get_fuse_inode(inode);
447 struct fuse_conn *fc = get_fuse_conn(inode);
448
449 if (!fuse_allow_task(fc, current))
450 return -EACCES;
451 if (get_node_id(inode) != FUSE_ROOT_ID &&
452 time_before_eq(jiffies, fi->i_time))
453 return 0;
454
455 return fuse_do_getattr(inode);
456}
457
458static int fuse_permission(struct inode *inode, int mask, struct nameidata *nd)
459{
460 struct fuse_conn *fc = get_fuse_conn(inode);
461
462 if (!fuse_allow_task(fc, current))
463 return -EACCES;
464 else if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
465 int err = generic_permission(inode, mask, NULL);
466
467 /* If permission is denied, try to refresh file
468 attributes. This is also needed, because the root
469 node will at first have no permissions */
470 if (err == -EACCES) {
471 err = fuse_do_getattr(inode);
472 if (!err)
473 err = generic_permission(inode, mask, NULL);
474 }
475
476 /* FIXME: Need some mechanism to revoke permissions:
477 currently if the filesystem suddenly changes the
478 file mode, we will not be informed about it, and
479 continue to allow access to the file/directory.
480
481 This is actually not so grave, since the user can
482 simply keep access to the file/directory anyway by
483 keeping it open... */
484
485 return err;
486 } else {
487 int mode = inode->i_mode;
488 if ((mask & MAY_WRITE) && IS_RDONLY(inode) &&
489 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
490 return -EROFS;
491 if ((mask & MAY_EXEC) && !S_ISDIR(mode) && !(mode & S_IXUGO))
492 return -EACCES;
493 return 0;
494 }
495}
496
497static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
498 void *dstbuf, filldir_t filldir)
499{
500 while (nbytes >= FUSE_NAME_OFFSET) {
501 struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
502 size_t reclen = FUSE_DIRENT_SIZE(dirent);
503 int over;
504 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
505 return -EIO;
506 if (reclen > nbytes)
507 break;
508
509 over = filldir(dstbuf, dirent->name, dirent->namelen,
510 file->f_pos, dirent->ino, dirent->type);
511 if (over)
512 break;
513
514 buf += reclen;
515 nbytes -= reclen;
516 file->f_pos = dirent->off;
517 }
518
519 return 0;
520}
521
522static inline size_t fuse_send_readdir(struct fuse_req *req, struct file *file,
523 struct inode *inode, loff_t pos,
524 size_t count)
525{
526 return fuse_send_read_common(req, file, inode, pos, count, 1);
527}
528
529static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
530{
531 int err;
532 size_t nbytes;
533 struct page *page;
534 struct inode *inode = file->f_dentry->d_inode;
535 struct fuse_conn *fc = get_fuse_conn(inode);
536 struct fuse_req *req = fuse_get_request(fc);
537 if (!req)
538 return -EINTR;
539
540 page = alloc_page(GFP_KERNEL);
541 if (!page) {
542 fuse_put_request(fc, req);
543 return -ENOMEM;
544 }
545 req->num_pages = 1;
546 req->pages[0] = page;
547 nbytes = fuse_send_readdir(req, file, inode, file->f_pos, PAGE_SIZE);
548 err = req->out.h.error;
549 fuse_put_request(fc, req);
550 if (!err)
551 err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
552 filldir);
553
554 __free_page(page);
555 fuse_invalidate_attr(inode); /* atime changed */
556 return err;
557}
558
559static char *read_link(struct dentry *dentry)
560{
561 struct inode *inode = dentry->d_inode;
562 struct fuse_conn *fc = get_fuse_conn(inode);
563 struct fuse_req *req = fuse_get_request(fc);
564 char *link;
565
566 if (!req)
567 return ERR_PTR(-EINTR);
568
569 link = (char *) __get_free_page(GFP_KERNEL);
570 if (!link) {
571 link = ERR_PTR(-ENOMEM);
572 goto out;
573 }
574 req->in.h.opcode = FUSE_READLINK;
575 req->in.h.nodeid = get_node_id(inode);
576 req->inode = inode;
577 req->out.argvar = 1;
578 req->out.numargs = 1;
579 req->out.args[0].size = PAGE_SIZE - 1;
580 req->out.args[0].value = link;
581 request_send(fc, req);
582 if (req->out.h.error) {
583 free_page((unsigned long) link);
584 link = ERR_PTR(req->out.h.error);
585 } else
586 link[req->out.args[0].size] = '\0';
587 out:
588 fuse_put_request(fc, req);
589 fuse_invalidate_attr(inode); /* atime changed */
590 return link;
591}
592
593static void free_link(char *link)
594{
595 if (!IS_ERR(link))
596 free_page((unsigned long) link);
597}
598
599static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
600{
601 nd_set_link(nd, read_link(dentry));
602 return NULL;
603}
604
605static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
606{
607 free_link(nd_get_link(nd));
608}
609
610static int fuse_dir_open(struct inode *inode, struct file *file)
611{
612 return fuse_open_common(inode, file, 1);
613}
614
615static int fuse_dir_release(struct inode *inode, struct file *file)
616{
617 return fuse_release_common(inode, file, 1);
618}
619
620static int fuse_dir_fsync(struct file *file, struct dentry *de, int datasync)
621{
622 /* nfsd can call this with no file */
623 return file ? fuse_fsync_common(file, de, datasync, 1) : 0;
624}
625
626static unsigned iattr_to_fattr(struct iattr *iattr, struct fuse_attr *fattr)
627{
628 unsigned ivalid = iattr->ia_valid;
629 unsigned fvalid = 0;
630
631 memset(fattr, 0, sizeof(*fattr));
632
633 if (ivalid & ATTR_MODE)
634 fvalid |= FATTR_MODE, fattr->mode = iattr->ia_mode;
635 if (ivalid & ATTR_UID)
636 fvalid |= FATTR_UID, fattr->uid = iattr->ia_uid;
637 if (ivalid & ATTR_GID)
638 fvalid |= FATTR_GID, fattr->gid = iattr->ia_gid;
639 if (ivalid & ATTR_SIZE)
640 fvalid |= FATTR_SIZE, fattr->size = iattr->ia_size;
641 /* You can only _set_ these together (they may change by themselves) */
642 if ((ivalid & (ATTR_ATIME | ATTR_MTIME)) == (ATTR_ATIME | ATTR_MTIME)) {
643 fvalid |= FATTR_ATIME | FATTR_MTIME;
644 fattr->atime = iattr->ia_atime.tv_sec;
645 fattr->mtime = iattr->ia_mtime.tv_sec;
646 }
647
648 return fvalid;
649}
650
651static int fuse_setattr(struct dentry *entry, struct iattr *attr)
652{
653 struct inode *inode = entry->d_inode;
654 struct fuse_conn *fc = get_fuse_conn(inode);
655 struct fuse_inode *fi = get_fuse_inode(inode);
656 struct fuse_req *req;
657 struct fuse_setattr_in inarg;
658 struct fuse_attr_out outarg;
659 int err;
660 int is_truncate = 0;
661
662 if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
663 err = inode_change_ok(inode, attr);
664 if (err)
665 return err;
666 }
667
668 if (attr->ia_valid & ATTR_SIZE) {
669 unsigned long limit;
670 is_truncate = 1;
671 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
672 if (limit != RLIM_INFINITY && attr->ia_size > (loff_t) limit) {
673 send_sig(SIGXFSZ, current, 0);
674 return -EFBIG;
675 }
676 }
677
678 req = fuse_get_request(fc);
679 if (!req)
680 return -EINTR;
681
682 memset(&inarg, 0, sizeof(inarg));
683 inarg.valid = iattr_to_fattr(attr, &inarg.attr);
684 req->in.h.opcode = FUSE_SETATTR;
685 req->in.h.nodeid = get_node_id(inode);
686 req->inode = inode;
687 req->in.numargs = 1;
688 req->in.args[0].size = sizeof(inarg);
689 req->in.args[0].value = &inarg;
690 req->out.numargs = 1;
691 req->out.args[0].size = sizeof(outarg);
692 req->out.args[0].value = &outarg;
693 request_send(fc, req);
694 err = req->out.h.error;
695 fuse_put_request(fc, req);
696 if (!err) {
697 if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
698 make_bad_inode(inode);
699 err = -EIO;
700 } else {
701 if (is_truncate) {
702 loff_t origsize = i_size_read(inode);
703 i_size_write(inode, outarg.attr.size);
704 if (origsize > outarg.attr.size)
705 vmtruncate(inode, outarg.attr.size);
706 }
707 fuse_change_attributes(inode, &outarg.attr);
708 fi->i_time = time_to_jiffies(outarg.attr_valid,
709 outarg.attr_valid_nsec);
710 }
711 } else if (err == -EINTR)
712 fuse_invalidate_attr(inode);
713
714 return err;
715}
716
717static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
718 struct kstat *stat)
719{
720 struct inode *inode = entry->d_inode;
721 int err = fuse_revalidate(entry);
722 if (!err)
723 generic_fillattr(inode, stat);
724
725 return err;
726}
727
728static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
729 struct nameidata *nd)
730{
731 struct inode *inode;
732 int err = fuse_lookup_iget(dir, entry, &inode);
733 if (err)
734 return ERR_PTR(err);
735 if (inode && S_ISDIR(inode->i_mode)) {
736 /* Don't allow creating an alias to a directory */
737 struct dentry *alias = d_find_alias(inode);
738 if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
739 dput(alias);
740 iput(inode);
741 return ERR_PTR(-EIO);
742 }
743 }
744 return d_splice_alias(inode, entry);
745}
746
747static int fuse_setxattr(struct dentry *entry, const char *name,
748 const void *value, size_t size, int flags)
749{
750 struct inode *inode = entry->d_inode;
751 struct fuse_conn *fc = get_fuse_conn(inode);
752 struct fuse_req *req;
753 struct fuse_setxattr_in inarg;
754 int err;
755
756 if (size > FUSE_XATTR_SIZE_MAX)
757 return -E2BIG;
758
759 if (fc->no_setxattr)
760 return -EOPNOTSUPP;
761
762 req = fuse_get_request(fc);
763 if (!req)
764 return -EINTR;
765
766 memset(&inarg, 0, sizeof(inarg));
767 inarg.size = size;
768 inarg.flags = flags;
769 req->in.h.opcode = FUSE_SETXATTR;
770 req->in.h.nodeid = get_node_id(inode);
771 req->inode = inode;
772 req->in.numargs = 3;
773 req->in.args[0].size = sizeof(inarg);
774 req->in.args[0].value = &inarg;
775 req->in.args[1].size = strlen(name) + 1;
776 req->in.args[1].value = name;
777 req->in.args[2].size = size;
778 req->in.args[2].value = value;
779 request_send(fc, req);
780 err = req->out.h.error;
781 fuse_put_request(fc, req);
782 if (err == -ENOSYS) {
783 fc->no_setxattr = 1;
784 err = -EOPNOTSUPP;
785 }
786 return err;
787}
788
789static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
790 void *value, size_t size)
791{
792 struct inode *inode = entry->d_inode;
793 struct fuse_conn *fc = get_fuse_conn(inode);
794 struct fuse_req *req;
795 struct fuse_getxattr_in inarg;
796 struct fuse_getxattr_out outarg;
797 ssize_t ret;
798
799 if (fc->no_getxattr)
800 return -EOPNOTSUPP;
801
802 req = fuse_get_request(fc);
803 if (!req)
804 return -EINTR;
805
806 memset(&inarg, 0, sizeof(inarg));
807 inarg.size = size;
808 req->in.h.opcode = FUSE_GETXATTR;
809 req->in.h.nodeid = get_node_id(inode);
810 req->inode = inode;
811 req->in.numargs = 2;
812 req->in.args[0].size = sizeof(inarg);
813 req->in.args[0].value = &inarg;
814 req->in.args[1].size = strlen(name) + 1;
815 req->in.args[1].value = name;
816 /* This is really two different operations rolled into one */
817 req->out.numargs = 1;
818 if (size) {
819 req->out.argvar = 1;
820 req->out.args[0].size = size;
821 req->out.args[0].value = value;
822 } else {
823 req->out.args[0].size = sizeof(outarg);
824 req->out.args[0].value = &outarg;
825 }
826 request_send(fc, req);
827 ret = req->out.h.error;
828 if (!ret)
829 ret = size ? req->out.args[0].size : outarg.size;
830 else {
831 if (ret == -ENOSYS) {
832 fc->no_getxattr = 1;
833 ret = -EOPNOTSUPP;
834 }
835 }
836 fuse_put_request(fc, req);
837 return ret;
838}
839
840static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
841{
842 struct inode *inode = entry->d_inode;
843 struct fuse_conn *fc = get_fuse_conn(inode);
844 struct fuse_req *req;
845 struct fuse_getxattr_in inarg;
846 struct fuse_getxattr_out outarg;
847 ssize_t ret;
848
849 if (fc->no_listxattr)
850 return -EOPNOTSUPP;
851
852 req = fuse_get_request(fc);
853 if (!req)
854 return -EINTR;
855
856 memset(&inarg, 0, sizeof(inarg));
857 inarg.size = size;
858 req->in.h.opcode = FUSE_LISTXATTR;
859 req->in.h.nodeid = get_node_id(inode);
860 req->inode = inode;
861 req->in.numargs = 1;
862 req->in.args[0].size = sizeof(inarg);
863 req->in.args[0].value = &inarg;
864 /* This is really two different operations rolled into one */
865 req->out.numargs = 1;
866 if (size) {
867 req->out.argvar = 1;
868 req->out.args[0].size = size;
869 req->out.args[0].value = list;
870 } else {
871 req->out.args[0].size = sizeof(outarg);
872 req->out.args[0].value = &outarg;
873 }
874 request_send(fc, req);
875 ret = req->out.h.error;
876 if (!ret)
877 ret = size ? req->out.args[0].size : outarg.size;
878 else {
879 if (ret == -ENOSYS) {
880 fc->no_listxattr = 1;
881 ret = -EOPNOTSUPP;
882 }
883 }
884 fuse_put_request(fc, req);
885 return ret;
886}
887
888static int fuse_removexattr(struct dentry *entry, const char *name)
889{
890 struct inode *inode = entry->d_inode;
891 struct fuse_conn *fc = get_fuse_conn(inode);
892 struct fuse_req *req;
893 int err;
894
895 if (fc->no_removexattr)
896 return -EOPNOTSUPP;
897
898 req = fuse_get_request(fc);
899 if (!req)
900 return -EINTR;
901
902 req->in.h.opcode = FUSE_REMOVEXATTR;
903 req->in.h.nodeid = get_node_id(inode);
904 req->inode = inode;
905 req->in.numargs = 1;
906 req->in.args[0].size = strlen(name) + 1;
907 req->in.args[0].value = name;
908 request_send(fc, req);
909 err = req->out.h.error;
910 fuse_put_request(fc, req);
911 if (err == -ENOSYS) {
912 fc->no_removexattr = 1;
913 err = -EOPNOTSUPP;
914 }
915 return err;
916}
917
918static struct inode_operations fuse_dir_inode_operations = {
919 .lookup = fuse_lookup,
920 .mkdir = fuse_mkdir,
921 .symlink = fuse_symlink,
922 .unlink = fuse_unlink,
923 .rmdir = fuse_rmdir,
924 .rename = fuse_rename,
925 .link = fuse_link,
926 .setattr = fuse_setattr,
927 .create = fuse_create,
928 .mknod = fuse_mknod,
929 .permission = fuse_permission,
930 .getattr = fuse_getattr,
931 .setxattr = fuse_setxattr,
932 .getxattr = fuse_getxattr,
933 .listxattr = fuse_listxattr,
934 .removexattr = fuse_removexattr,
935};
936
937static struct file_operations fuse_dir_operations = {
938 .llseek = generic_file_llseek,
939 .read = generic_read_dir,
940 .readdir = fuse_readdir,
941 .open = fuse_dir_open,
942 .release = fuse_dir_release,
943 .fsync = fuse_dir_fsync,
944};
945
946static struct inode_operations fuse_common_inode_operations = {
947 .setattr = fuse_setattr,
948 .permission = fuse_permission,
949 .getattr = fuse_getattr,
950 .setxattr = fuse_setxattr,
951 .getxattr = fuse_getxattr,
952 .listxattr = fuse_listxattr,
953 .removexattr = fuse_removexattr,
954};
955
956static struct inode_operations fuse_symlink_inode_operations = {
957 .setattr = fuse_setattr,
958 .follow_link = fuse_follow_link,
959 .put_link = fuse_put_link,
960 .readlink = generic_readlink,
961 .getattr = fuse_getattr,
962 .setxattr = fuse_setxattr,
963 .getxattr = fuse_getxattr,
964 .listxattr = fuse_listxattr,
965 .removexattr = fuse_removexattr,
966};
967
968void fuse_init_common(struct inode *inode)
969{
970 inode->i_op = &fuse_common_inode_operations;
971}
972
973void fuse_init_dir(struct inode *inode)
974{
975 inode->i_op = &fuse_dir_inode_operations;
976 inode->i_fop = &fuse_dir_operations;
977}
978
979void fuse_init_symlink(struct inode *inode)
980{
981 inode->i_op = &fuse_symlink_inode_operations;
982}
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
new file mode 100644
index 000000000000..6454022b0536
--- /dev/null
+++ b/fs/fuse/file.c
@@ -0,0 +1,555 @@
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2005 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/pagemap.h>
12#include <linux/slab.h>
13#include <linux/kernel.h>
14
15static struct file_operations fuse_direct_io_file_operations;
16
17int fuse_open_common(struct inode *inode, struct file *file, int isdir)
18{
19 struct fuse_conn *fc = get_fuse_conn(inode);
20 struct fuse_req *req;
21 struct fuse_open_in inarg;
22 struct fuse_open_out outarg;
23 struct fuse_file *ff;
24 int err;
25
26 err = generic_file_open(inode, file);
27 if (err)
28 return err;
29
30 /* If opening the root node, no lookup has been performed on
31 it, so the attributes must be refreshed */
32 if (get_node_id(inode) == FUSE_ROOT_ID) {
33 int err = fuse_do_getattr(inode);
34 if (err)
35 return err;
36 }
37
38 req = fuse_get_request(fc);
39 if (!req)
40 return -EINTR;
41
42 err = -ENOMEM;
43 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
44 if (!ff)
45 goto out_put_request;
46
47 ff->release_req = fuse_request_alloc();
48 if (!ff->release_req) {
49 kfree(ff);
50 goto out_put_request;
51 }
52
53 memset(&inarg, 0, sizeof(inarg));
54 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
55 req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
56 req->in.h.nodeid = get_node_id(inode);
57 req->inode = inode;
58 req->in.numargs = 1;
59 req->in.args[0].size = sizeof(inarg);
60 req->in.args[0].value = &inarg;
61 req->out.numargs = 1;
62 req->out.args[0].size = sizeof(outarg);
63 req->out.args[0].value = &outarg;
64 request_send(fc, req);
65 err = req->out.h.error;
66 if (err) {
67 fuse_request_free(ff->release_req);
68 kfree(ff);
69 } else {
70 if (!isdir && (outarg.open_flags & FOPEN_DIRECT_IO))
71 file->f_op = &fuse_direct_io_file_operations;
72 if (!(outarg.open_flags & FOPEN_KEEP_CACHE))
73 invalidate_inode_pages(inode->i_mapping);
74 ff->fh = outarg.fh;
75 file->private_data = ff;
76 }
77
78 out_put_request:
79 fuse_put_request(fc, req);
80 return err;
81}
82
83int fuse_release_common(struct inode *inode, struct file *file, int isdir)
84{
85 struct fuse_conn *fc = get_fuse_conn(inode);
86 struct fuse_file *ff = file->private_data;
87 struct fuse_req *req = ff->release_req;
88 struct fuse_release_in *inarg = &req->misc.release_in;
89
90 inarg->fh = ff->fh;
91 inarg->flags = file->f_flags & ~O_EXCL;
92 req->in.h.opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE;
93 req->in.h.nodeid = get_node_id(inode);
94 req->inode = inode;
95 req->in.numargs = 1;
96 req->in.args[0].size = sizeof(struct fuse_release_in);
97 req->in.args[0].value = inarg;
98 request_send_background(fc, req);
99 kfree(ff);
100
101 /* Return value is ignored by VFS */
102 return 0;
103}
104
105static int fuse_open(struct inode *inode, struct file *file)
106{
107 return fuse_open_common(inode, file, 0);
108}
109
110static int fuse_release(struct inode *inode, struct file *file)
111{
112 return fuse_release_common(inode, file, 0);
113}
114
115static int fuse_flush(struct file *file)
116{
117 struct inode *inode = file->f_dentry->d_inode;
118 struct fuse_conn *fc = get_fuse_conn(inode);
119 struct fuse_file *ff = file->private_data;
120 struct fuse_req *req;
121 struct fuse_flush_in inarg;
122 int err;
123
124 if (fc->no_flush)
125 return 0;
126
127 req = fuse_get_request(fc);
128 if (!req)
129 return -EINTR;
130
131 memset(&inarg, 0, sizeof(inarg));
132 inarg.fh = ff->fh;
133 req->in.h.opcode = FUSE_FLUSH;
134 req->in.h.nodeid = get_node_id(inode);
135 req->inode = inode;
136 req->file = file;
137 req->in.numargs = 1;
138 req->in.args[0].size = sizeof(inarg);
139 req->in.args[0].value = &inarg;
140 request_send(fc, req);
141 err = req->out.h.error;
142 fuse_put_request(fc, req);
143 if (err == -ENOSYS) {
144 fc->no_flush = 1;
145 err = 0;
146 }
147 return err;
148}
149
150int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
151 int isdir)
152{
153 struct inode *inode = de->d_inode;
154 struct fuse_conn *fc = get_fuse_conn(inode);
155 struct fuse_file *ff = file->private_data;
156 struct fuse_req *req;
157 struct fuse_fsync_in inarg;
158 int err;
159
160 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
161 return 0;
162
163 req = fuse_get_request(fc);
164 if (!req)
165 return -EINTR;
166
167 memset(&inarg, 0, sizeof(inarg));
168 inarg.fh = ff->fh;
169 inarg.fsync_flags = datasync ? 1 : 0;
170 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
171 req->in.h.nodeid = get_node_id(inode);
172 req->inode = inode;
173 req->file = file;
174 req->in.numargs = 1;
175 req->in.args[0].size = sizeof(inarg);
176 req->in.args[0].value = &inarg;
177 request_send(fc, req);
178 err = req->out.h.error;
179 fuse_put_request(fc, req);
180 if (err == -ENOSYS) {
181 if (isdir)
182 fc->no_fsyncdir = 1;
183 else
184 fc->no_fsync = 1;
185 err = 0;
186 }
187 return err;
188}
189
190static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
191{
192 return fuse_fsync_common(file, de, datasync, 0);
193}
194
195size_t fuse_send_read_common(struct fuse_req *req, struct file *file,
196 struct inode *inode, loff_t pos, size_t count,
197 int isdir)
198{
199 struct fuse_conn *fc = get_fuse_conn(inode);
200 struct fuse_file *ff = file->private_data;
201 struct fuse_read_in inarg;
202
203 memset(&inarg, 0, sizeof(struct fuse_read_in));
204 inarg.fh = ff->fh;
205 inarg.offset = pos;
206 inarg.size = count;
207 req->in.h.opcode = isdir ? FUSE_READDIR : FUSE_READ;
208 req->in.h.nodeid = get_node_id(inode);
209 req->inode = inode;
210 req->file = file;
211 req->in.numargs = 1;
212 req->in.args[0].size = sizeof(struct fuse_read_in);
213 req->in.args[0].value = &inarg;
214 req->out.argpages = 1;
215 req->out.argvar = 1;
216 req->out.numargs = 1;
217 req->out.args[0].size = count;
218 request_send(fc, req);
219 return req->out.args[0].size;
220}
221
222static inline size_t fuse_send_read(struct fuse_req *req, struct file *file,
223 struct inode *inode, loff_t pos,
224 size_t count)
225{
226 return fuse_send_read_common(req, file, inode, pos, count, 0);
227}
228
229static int fuse_readpage(struct file *file, struct page *page)
230{
231 struct inode *inode = page->mapping->host;
232 struct fuse_conn *fc = get_fuse_conn(inode);
233 loff_t pos = (loff_t) page->index << PAGE_CACHE_SHIFT;
234 struct fuse_req *req = fuse_get_request(fc);
235 int err = -EINTR;
236 if (!req)
237 goto out;
238
239 req->out.page_zeroing = 1;
240 req->num_pages = 1;
241 req->pages[0] = page;
242 fuse_send_read(req, file, inode, pos, PAGE_CACHE_SIZE);
243 err = req->out.h.error;
244 fuse_put_request(fc, req);
245 if (!err)
246 SetPageUptodate(page);
247 fuse_invalidate_attr(inode); /* atime changed */
248 out:
249 unlock_page(page);
250 return err;
251}
252
253static int fuse_send_readpages(struct fuse_req *req, struct file *file,
254 struct inode *inode)
255{
256 loff_t pos = (loff_t) req->pages[0]->index << PAGE_CACHE_SHIFT;
257 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
258 unsigned i;
259 req->out.page_zeroing = 1;
260 fuse_send_read(req, file, inode, pos, count);
261 for (i = 0; i < req->num_pages; i++) {
262 struct page *page = req->pages[i];
263 if (!req->out.h.error)
264 SetPageUptodate(page);
265 unlock_page(page);
266 }
267 return req->out.h.error;
268}
269
270struct fuse_readpages_data {
271 struct fuse_req *req;
272 struct file *file;
273 struct inode *inode;
274};
275
276static int fuse_readpages_fill(void *_data, struct page *page)
277{
278 struct fuse_readpages_data *data = _data;
279 struct fuse_req *req = data->req;
280 struct inode *inode = data->inode;
281 struct fuse_conn *fc = get_fuse_conn(inode);
282
283 if (req->num_pages &&
284 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
285 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
286 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
287 int err = fuse_send_readpages(req, data->file, inode);
288 if (err) {
289 unlock_page(page);
290 return err;
291 }
292 fuse_reset_request(req);
293 }
294 req->pages[req->num_pages] = page;
295 req->num_pages ++;
296 return 0;
297}
298
299static int fuse_readpages(struct file *file, struct address_space *mapping,
300 struct list_head *pages, unsigned nr_pages)
301{
302 struct inode *inode = mapping->host;
303 struct fuse_conn *fc = get_fuse_conn(inode);
304 struct fuse_readpages_data data;
305 int err;
306 data.file = file;
307 data.inode = inode;
308 data.req = fuse_get_request(fc);
309 if (!data.req)
310 return -EINTR;
311
312 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
313 if (!err && data.req->num_pages)
314 err = fuse_send_readpages(data.req, file, inode);
315 fuse_put_request(fc, data.req);
316 fuse_invalidate_attr(inode); /* atime changed */
317 return err;
318}
319
320static size_t fuse_send_write(struct fuse_req *req, struct file *file,
321 struct inode *inode, loff_t pos, size_t count)
322{
323 struct fuse_conn *fc = get_fuse_conn(inode);
324 struct fuse_file *ff = file->private_data;
325 struct fuse_write_in inarg;
326 struct fuse_write_out outarg;
327
328 memset(&inarg, 0, sizeof(struct fuse_write_in));
329 inarg.fh = ff->fh;
330 inarg.offset = pos;
331 inarg.size = count;
332 req->in.h.opcode = FUSE_WRITE;
333 req->in.h.nodeid = get_node_id(inode);
334 req->inode = inode;
335 req->file = file;
336 req->in.argpages = 1;
337 req->in.numargs = 2;
338 req->in.args[0].size = sizeof(struct fuse_write_in);
339 req->in.args[0].value = &inarg;
340 req->in.args[1].size = count;
341 req->out.numargs = 1;
342 req->out.args[0].size = sizeof(struct fuse_write_out);
343 req->out.args[0].value = &outarg;
344 request_send(fc, req);
345 return outarg.size;
346}
347
348static int fuse_prepare_write(struct file *file, struct page *page,
349 unsigned offset, unsigned to)
350{
351 /* No op */
352 return 0;
353}
354
355static int fuse_commit_write(struct file *file, struct page *page,
356 unsigned offset, unsigned to)
357{
358 int err;
359 size_t nres;
360 unsigned count = to - offset;
361 struct inode *inode = page->mapping->host;
362 struct fuse_conn *fc = get_fuse_conn(inode);
363 loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + offset;
364 struct fuse_req *req = fuse_get_request(fc);
365 if (!req)
366 return -EINTR;
367
368 req->num_pages = 1;
369 req->pages[0] = page;
370 req->page_offset = offset;
371 nres = fuse_send_write(req, file, inode, pos, count);
372 err = req->out.h.error;
373 fuse_put_request(fc, req);
374 if (!err && nres != count)
375 err = -EIO;
376 if (!err) {
377 pos += count;
378 if (pos > i_size_read(inode))
379 i_size_write(inode, pos);
380
381 if (offset == 0 && to == PAGE_CACHE_SIZE) {
382 clear_page_dirty(page);
383 SetPageUptodate(page);
384 }
385 }
386 fuse_invalidate_attr(inode);
387 return err;
388}
389
390static void fuse_release_user_pages(struct fuse_req *req, int write)
391{
392 unsigned i;
393
394 for (i = 0; i < req->num_pages; i++) {
395 struct page *page = req->pages[i];
396 if (write)
397 set_page_dirty_lock(page);
398 put_page(page);
399 }
400}
401
402static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
403 unsigned nbytes, int write)
404{
405 unsigned long user_addr = (unsigned long) buf;
406 unsigned offset = user_addr & ~PAGE_MASK;
407 int npages;
408
409 /* This doesn't work with nfsd */
410 if (!current->mm)
411 return -EPERM;
412
413 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
414 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
415 npages = min(npages, FUSE_MAX_PAGES_PER_REQ);
416 down_read(&current->mm->mmap_sem);
417 npages = get_user_pages(current, current->mm, user_addr, npages, write,
418 0, req->pages, NULL);
419 up_read(&current->mm->mmap_sem);
420 if (npages < 0)
421 return npages;
422
423 req->num_pages = npages;
424 req->page_offset = offset;
425 return 0;
426}
427
428static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
429 size_t count, loff_t *ppos, int write)
430{
431 struct inode *inode = file->f_dentry->d_inode;
432 struct fuse_conn *fc = get_fuse_conn(inode);
433 size_t nmax = write ? fc->max_write : fc->max_read;
434 loff_t pos = *ppos;
435 ssize_t res = 0;
436 struct fuse_req *req = fuse_get_request(fc);
437 if (!req)
438 return -EINTR;
439
440 while (count) {
441 size_t tmp;
442 size_t nres;
443 size_t nbytes = min(count, nmax);
444 int err = fuse_get_user_pages(req, buf, nbytes, !write);
445 if (err) {
446 res = err;
447 break;
448 }
449 tmp = (req->num_pages << PAGE_SHIFT) - req->page_offset;
450 nbytes = min(nbytes, tmp);
451 if (write)
452 nres = fuse_send_write(req, file, inode, pos, nbytes);
453 else
454 nres = fuse_send_read(req, file, inode, pos, nbytes);
455 fuse_release_user_pages(req, !write);
456 if (req->out.h.error) {
457 if (!res)
458 res = req->out.h.error;
459 break;
460 } else if (nres > nbytes) {
461 res = -EIO;
462 break;
463 }
464 count -= nres;
465 res += nres;
466 pos += nres;
467 buf += nres;
468 if (nres != nbytes)
469 break;
470 if (count)
471 fuse_reset_request(req);
472 }
473 fuse_put_request(fc, req);
474 if (res > 0) {
475 if (write && pos > i_size_read(inode))
476 i_size_write(inode, pos);
477 *ppos = pos;
478 }
479 fuse_invalidate_attr(inode);
480
481 return res;
482}
483
484static ssize_t fuse_direct_read(struct file *file, char __user *buf,
485 size_t count, loff_t *ppos)
486{
487 return fuse_direct_io(file, buf, count, ppos, 0);
488}
489
490static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
491 size_t count, loff_t *ppos)
492{
493 struct inode *inode = file->f_dentry->d_inode;
494 ssize_t res;
495 /* Don't allow parallel writes to the same file */
496 down(&inode->i_sem);
497 res = fuse_direct_io(file, buf, count, ppos, 1);
498 up(&inode->i_sem);
499 return res;
500}
501
502static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
503{
504 if ((vma->vm_flags & VM_SHARED)) {
505 if ((vma->vm_flags & VM_WRITE))
506 return -ENODEV;
507 else
508 vma->vm_flags &= ~VM_MAYWRITE;
509 }
510 return generic_file_mmap(file, vma);
511}
512
513static int fuse_set_page_dirty(struct page *page)
514{
515 printk("fuse_set_page_dirty: should not happen\n");
516 dump_stack();
517 return 0;
518}
519
520static struct file_operations fuse_file_operations = {
521 .llseek = generic_file_llseek,
522 .read = generic_file_read,
523 .write = generic_file_write,
524 .mmap = fuse_file_mmap,
525 .open = fuse_open,
526 .flush = fuse_flush,
527 .release = fuse_release,
528 .fsync = fuse_fsync,
529 .sendfile = generic_file_sendfile,
530};
531
532static struct file_operations fuse_direct_io_file_operations = {
533 .llseek = generic_file_llseek,
534 .read = fuse_direct_read,
535 .write = fuse_direct_write,
536 .open = fuse_open,
537 .flush = fuse_flush,
538 .release = fuse_release,
539 .fsync = fuse_fsync,
540 /* no mmap and sendfile */
541};
542
543static struct address_space_operations fuse_file_aops = {
544 .readpage = fuse_readpage,
545 .prepare_write = fuse_prepare_write,
546 .commit_write = fuse_commit_write,
547 .readpages = fuse_readpages,
548 .set_page_dirty = fuse_set_page_dirty,
549};
550
551void fuse_init_file_inode(struct inode *inode)
552{
553 inode->i_fop = &fuse_file_operations;
554 inode->i_data.a_ops = &fuse_file_aops;
555}
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
new file mode 100644
index 000000000000..24d761518d86
--- /dev/null
+++ b/fs/fuse/fuse_i.h
@@ -0,0 +1,451 @@
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2005 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include <linux/fuse.h>
10#include <linux/fs.h>
11#include <linux/wait.h>
12#include <linux/list.h>
13#include <linux/spinlock.h>
14#include <linux/mm.h>
15#include <linux/backing-dev.h>
16#include <asm/semaphore.h>
17
18/** Max number of pages that can be used in a single read request */
19#define FUSE_MAX_PAGES_PER_REQ 32
20
21/** If more requests are outstanding, then the operation will block */
22#define FUSE_MAX_OUTSTANDING 10
23
24/** If the FUSE_DEFAULT_PERMISSIONS flag is given, the filesystem
25 module will check permissions based on the file mode. Otherwise no
26 permission checking is done in the kernel */
27#define FUSE_DEFAULT_PERMISSIONS (1 << 0)
28
29/** If the FUSE_ALLOW_OTHER flag is given, then not only the user
30 doing the mount will be allowed to access the filesystem */
31#define FUSE_ALLOW_OTHER (1 << 1)
32
33
34/** FUSE inode */
35struct fuse_inode {
36 /** Inode data */
37 struct inode inode;
38
39 /** Unique ID, which identifies the inode between userspace
40 * and kernel */
41 u64 nodeid;
42
43 /** Number of lookups on this inode */
44 u64 nlookup;
45
46 /** The request used for sending the FORGET message */
47 struct fuse_req *forget_req;
48
49 /** Time in jiffies until the file attributes are valid */
50 unsigned long i_time;
51};
52
53/** FUSE specific file data */
54struct fuse_file {
55 /** Request reserved for flush and release */
56 struct fuse_req *release_req;
57
58 /** File handle used by userspace */
59 u64 fh;
60};
61
62/** One input argument of a request */
63struct fuse_in_arg {
64 unsigned size;
65 const void *value;
66};
67
68/** The request input */
69struct fuse_in {
70 /** The request header */
71 struct fuse_in_header h;
72
73 /** True if the data for the last argument is in req->pages */
74 unsigned argpages:1;
75
76 /** Number of arguments */
77 unsigned numargs;
78
79 /** Array of arguments */
80 struct fuse_in_arg args[3];
81};
82
83/** One output argument of a request */
84struct fuse_arg {
85 unsigned size;
86 void *value;
87};
88
89/** The request output */
90struct fuse_out {
91 /** Header returned from userspace */
92 struct fuse_out_header h;
93
94 /** Last argument is variable length (can be shorter than
95 arg->size) */
96 unsigned argvar:1;
97
98 /** Last argument is a list of pages to copy data to */
99 unsigned argpages:1;
100
101 /** Zero partially or not copied pages */
102 unsigned page_zeroing:1;
103
104 /** Number or arguments */
105 unsigned numargs;
106
107 /** Array of arguments */
108 struct fuse_arg args[3];
109};
110
111struct fuse_req;
112struct fuse_conn;
113
114/**
115 * A request to the client
116 */
117struct fuse_req {
118 /** This can be on either unused_list, pending or processing
119 lists in fuse_conn */
120 struct list_head list;
121
122 /** Entry on the background list */
123 struct list_head bg_entry;
124
125 /** refcount */
126 atomic_t count;
127
128 /** True if the request has reply */
129 unsigned isreply:1;
130
131 /** The request is preallocated */
132 unsigned preallocated:1;
133
134 /** The request was interrupted */
135 unsigned interrupted:1;
136
137 /** Request is sent in the background */
138 unsigned background:1;
139
140 /** Data is being copied to/from the request */
141 unsigned locked:1;
142
143 /** Request has been sent to userspace */
144 unsigned sent:1;
145
146 /** The request is finished */
147 unsigned finished:1;
148
149 /** The request input */
150 struct fuse_in in;
151
152 /** The request output */
153 struct fuse_out out;
154
155 /** Used to wake up the task waiting for completion of request*/
156 wait_queue_head_t waitq;
157
158 /** Data for asynchronous requests */
159 union {
160 struct fuse_forget_in forget_in;
161 struct fuse_release_in release_in;
162 struct fuse_init_in_out init_in_out;
163 } misc;
164
165 /** page vector */
166 struct page *pages[FUSE_MAX_PAGES_PER_REQ];
167
168 /** number of pages in vector */
169 unsigned num_pages;
170
171 /** offset of data on first page */
172 unsigned page_offset;
173
174 /** Inode used in the request */
175 struct inode *inode;
176
177 /** Second inode used in the request (or NULL) */
178 struct inode *inode2;
179
180 /** File used in the request (or NULL) */
181 struct file *file;
182};
183
184/**
185 * A Fuse connection.
186 *
187 * This structure is created, when the filesystem is mounted, and is
188 * destroyed, when the client device is closed and the filesystem is
189 * unmounted.
190 */
191struct fuse_conn {
192 /** Reference count */
193 int count;
194
195 /** The user id for this mount */
196 uid_t user_id;
197
198 /** The group id for this mount */
199 gid_t group_id;
200
201 /** The fuse mount flags for this mount */
202 unsigned flags;
203
204 /** Maximum read size */
205 unsigned max_read;
206
207 /** Maximum write size */
208 unsigned max_write;
209
210 /** Readers of the connection are waiting on this */
211 wait_queue_head_t waitq;
212
213 /** The list of pending requests */
214 struct list_head pending;
215
216 /** The list of requests being processed */
217 struct list_head processing;
218
219 /** Requests put in the background (RELEASE or any other
220 interrupted request) */
221 struct list_head background;
222
223 /** Controls the maximum number of outstanding requests */
224 struct semaphore outstanding_sem;
225
226 /** This counts the number of outstanding requests if
227 outstanding_sem would go negative */
228 unsigned outstanding_debt;
229
230 /** RW semaphore for exclusion with fuse_put_super() */
231 struct rw_semaphore sbput_sem;
232
233 /** The list of unused requests */
234 struct list_head unused_list;
235
236 /** The next unique request id */
237 u64 reqctr;
238
239 /** Mount is active */
240 unsigned mounted : 1;
241
242 /** Connection established */
243 unsigned connected : 1;
244
245 /** Connection failed (version mismatch) */
246 unsigned conn_error : 1;
247
248 /** Is fsync not implemented by fs? */
249 unsigned no_fsync : 1;
250
251 /** Is fsyncdir not implemented by fs? */
252 unsigned no_fsyncdir : 1;
253
254 /** Is flush not implemented by fs? */
255 unsigned no_flush : 1;
256
257 /** Is setxattr not implemented by fs? */
258 unsigned no_setxattr : 1;
259
260 /** Is getxattr not implemented by fs? */
261 unsigned no_getxattr : 1;
262
263 /** Is listxattr not implemented by fs? */
264 unsigned no_listxattr : 1;
265
266 /** Is removexattr not implemented by fs? */
267 unsigned no_removexattr : 1;
268
269 /** Backing dev info */
270 struct backing_dev_info bdi;
271};
272
273static inline struct fuse_conn **get_fuse_conn_super_p(struct super_block *sb)
274{
275 return (struct fuse_conn **) &sb->s_fs_info;
276}
277
278static inline struct fuse_conn *get_fuse_conn_super(struct super_block *sb)
279{
280 return *get_fuse_conn_super_p(sb);
281}
282
283static inline struct fuse_conn *get_fuse_conn(struct inode *inode)
284{
285 return get_fuse_conn_super(inode->i_sb);
286}
287
288static inline struct fuse_inode *get_fuse_inode(struct inode *inode)
289{
290 return container_of(inode, struct fuse_inode, inode);
291}
292
293static inline u64 get_node_id(struct inode *inode)
294{
295 return get_fuse_inode(inode)->nodeid;
296}
297
298/** Device operations */
299extern struct file_operations fuse_dev_operations;
300
301/**
302 * This is the single global spinlock which protects FUSE's structures
303 *
304 * The following data is protected by this lock:
305 *
306 * - the private_data field of the device file
307 * - the s_fs_info field of the super block
308 * - unused_list, pending, processing lists in fuse_conn
309 * - background list in fuse_conn
310 * - the unique request ID counter reqctr in fuse_conn
311 * - the sb (super_block) field in fuse_conn
312 * - the file (device file) field in fuse_conn
313 */
314extern spinlock_t fuse_lock;
315
316/**
317 * Get a filled in inode
318 */
319struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
320 int generation, struct fuse_attr *attr);
321
322/**
323 * Send FORGET command
324 */
325void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
326 unsigned long nodeid, u64 nlookup);
327
328/**
329 * Send READ or READDIR request
330 */
331size_t fuse_send_read_common(struct fuse_req *req, struct file *file,
332 struct inode *inode, loff_t pos, size_t count,
333 int isdir);
334
335/**
336 * Send OPEN or OPENDIR request
337 */
338int fuse_open_common(struct inode *inode, struct file *file, int isdir);
339
340/**
341 * Send RELEASE or RELEASEDIR request
342 */
343int fuse_release_common(struct inode *inode, struct file *file, int isdir);
344
345/**
346 * Send FSYNC or FSYNCDIR request
347 */
348int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
349 int isdir);
350
351/**
352 * Initialise file operations on a regular file
353 */
354void fuse_init_file_inode(struct inode *inode);
355
356/**
357 * Initialise inode operations on regular files and special files
358 */
359void fuse_init_common(struct inode *inode);
360
361/**
362 * Initialise inode and file operations on a directory
363 */
364void fuse_init_dir(struct inode *inode);
365
366/**
367 * Initialise inode operations on a symlink
368 */
369void fuse_init_symlink(struct inode *inode);
370
371/**
372 * Change attributes of an inode
373 */
374void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr);
375
376/**
377 * Check if the connection can be released, and if yes, then free the
378 * connection structure
379 */
380void fuse_release_conn(struct fuse_conn *fc);
381
382/**
383 * Initialize the client device
384 */
385int fuse_dev_init(void);
386
387/**
388 * Cleanup the client device
389 */
390void fuse_dev_cleanup(void);
391
392/**
393 * Allocate a request
394 */
395struct fuse_req *fuse_request_alloc(void);
396
397/**
398 * Free a request
399 */
400void fuse_request_free(struct fuse_req *req);
401
402/**
403 * Reinitialize a request, the preallocated flag is left unmodified
404 */
405void fuse_reset_request(struct fuse_req *req);
406
407/**
408 * Reserve a preallocated request
409 */
410struct fuse_req *fuse_get_request(struct fuse_conn *fc);
411
412/**
413 * Decrement reference count of a request. If count goes to zero put
414 * on unused list (preallocated) or free reqest (not preallocated).
415 */
416void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req);
417
418/**
419 * Send a request (synchronous)
420 */
421void request_send(struct fuse_conn *fc, struct fuse_req *req);
422
423/**
424 * Send a request with no reply
425 */
426void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req);
427
428/**
429 * Send a request in the background
430 */
431void request_send_background(struct fuse_conn *fc, struct fuse_req *req);
432
433/**
434 * Release inodes and file assiciated with background request
435 */
436void fuse_release_background(struct fuse_req *req);
437
438/**
439 * Get the attributes of a file
440 */
441int fuse_do_getattr(struct inode *inode);
442
443/**
444 * Invalidate inode attributes
445 */
446void fuse_invalidate_attr(struct inode *inode);
447
448/**
449 * Send the INIT message
450 */
451void fuse_send_init(struct fuse_conn *fc);
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
new file mode 100644
index 000000000000..e69a546844d0
--- /dev/null
+++ b/fs/fuse/inode.c
@@ -0,0 +1,591 @@
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2005 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/pagemap.h>
12#include <linux/slab.h>
13#include <linux/file.h>
14#include <linux/mount.h>
15#include <linux/seq_file.h>
16#include <linux/init.h>
17#include <linux/module.h>
18#include <linux/parser.h>
19#include <linux/statfs.h>
20
21MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
22MODULE_DESCRIPTION("Filesystem in Userspace");
23MODULE_LICENSE("GPL");
24
25spinlock_t fuse_lock;
26static kmem_cache_t *fuse_inode_cachep;
27
28#define FUSE_SUPER_MAGIC 0x65735546
29
30struct fuse_mount_data {
31 int fd;
32 unsigned rootmode;
33 unsigned user_id;
34 unsigned group_id;
35 unsigned fd_present : 1;
36 unsigned rootmode_present : 1;
37 unsigned user_id_present : 1;
38 unsigned group_id_present : 1;
39 unsigned flags;
40 unsigned max_read;
41};
42
43static struct inode *fuse_alloc_inode(struct super_block *sb)
44{
45 struct inode *inode;
46 struct fuse_inode *fi;
47
48 inode = kmem_cache_alloc(fuse_inode_cachep, SLAB_KERNEL);
49 if (!inode)
50 return NULL;
51
52 fi = get_fuse_inode(inode);
53 fi->i_time = jiffies - 1;
54 fi->nodeid = 0;
55 fi->nlookup = 0;
56 fi->forget_req = fuse_request_alloc();
57 if (!fi->forget_req) {
58 kmem_cache_free(fuse_inode_cachep, inode);
59 return NULL;
60 }
61
62 return inode;
63}
64
65static void fuse_destroy_inode(struct inode *inode)
66{
67 struct fuse_inode *fi = get_fuse_inode(inode);
68 if (fi->forget_req)
69 fuse_request_free(fi->forget_req);
70 kmem_cache_free(fuse_inode_cachep, inode);
71}
72
73static void fuse_read_inode(struct inode *inode)
74{
75 /* No op */
76}
77
78void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
79 unsigned long nodeid, u64 nlookup)
80{
81 struct fuse_forget_in *inarg = &req->misc.forget_in;
82 inarg->nlookup = nlookup;
83 req->in.h.opcode = FUSE_FORGET;
84 req->in.h.nodeid = nodeid;
85 req->in.numargs = 1;
86 req->in.args[0].size = sizeof(struct fuse_forget_in);
87 req->in.args[0].value = inarg;
88 request_send_noreply(fc, req);
89}
90
91static void fuse_clear_inode(struct inode *inode)
92{
93 if (inode->i_sb->s_flags & MS_ACTIVE) {
94 struct fuse_conn *fc = get_fuse_conn(inode);
95 struct fuse_inode *fi = get_fuse_inode(inode);
96 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
97 fi->forget_req = NULL;
98 }
99}
100
101void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr)
102{
103 if (S_ISREG(inode->i_mode) && i_size_read(inode) != attr->size)
104 invalidate_inode_pages(inode->i_mapping);
105
106 inode->i_ino = attr->ino;
107 inode->i_mode = (inode->i_mode & S_IFMT) + (attr->mode & 07777);
108 inode->i_nlink = attr->nlink;
109 inode->i_uid = attr->uid;
110 inode->i_gid = attr->gid;
111 i_size_write(inode, attr->size);
112 inode->i_blksize = PAGE_CACHE_SIZE;
113 inode->i_blocks = attr->blocks;
114 inode->i_atime.tv_sec = attr->atime;
115 inode->i_atime.tv_nsec = attr->atimensec;
116 inode->i_mtime.tv_sec = attr->mtime;
117 inode->i_mtime.tv_nsec = attr->mtimensec;
118 inode->i_ctime.tv_sec = attr->ctime;
119 inode->i_ctime.tv_nsec = attr->ctimensec;
120}
121
122static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
123{
124 inode->i_mode = attr->mode & S_IFMT;
125 i_size_write(inode, attr->size);
126 if (S_ISREG(inode->i_mode)) {
127 fuse_init_common(inode);
128 fuse_init_file_inode(inode);
129 } else if (S_ISDIR(inode->i_mode))
130 fuse_init_dir(inode);
131 else if (S_ISLNK(inode->i_mode))
132 fuse_init_symlink(inode);
133 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
134 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
135 fuse_init_common(inode);
136 init_special_inode(inode, inode->i_mode,
137 new_decode_dev(attr->rdev));
138 } else {
139 /* Don't let user create weird files */
140 inode->i_mode = S_IFREG;
141 fuse_init_common(inode);
142 fuse_init_file_inode(inode);
143 }
144}
145
146static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
147{
148 unsigned long nodeid = *(unsigned long *) _nodeidp;
149 if (get_node_id(inode) == nodeid)
150 return 1;
151 else
152 return 0;
153}
154
155static int fuse_inode_set(struct inode *inode, void *_nodeidp)
156{
157 unsigned long nodeid = *(unsigned long *) _nodeidp;
158 get_fuse_inode(inode)->nodeid = nodeid;
159 return 0;
160}
161
162struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
163 int generation, struct fuse_attr *attr)
164{
165 struct inode *inode;
166 struct fuse_inode *fi;
167 struct fuse_conn *fc = get_fuse_conn_super(sb);
168 int retried = 0;
169
170 retry:
171 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
172 if (!inode)
173 return NULL;
174
175 if ((inode->i_state & I_NEW)) {
176 inode->i_flags |= S_NOATIME|S_NOCMTIME;
177 inode->i_generation = generation;
178 inode->i_data.backing_dev_info = &fc->bdi;
179 fuse_init_inode(inode, attr);
180 unlock_new_inode(inode);
181 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
182 BUG_ON(retried);
183 /* Inode has changed type, any I/O on the old should fail */
184 make_bad_inode(inode);
185 iput(inode);
186 retried = 1;
187 goto retry;
188 }
189
190 fi = get_fuse_inode(inode);
191 fi->nlookup ++;
192 fuse_change_attributes(inode, attr);
193 return inode;
194}
195
196static void fuse_put_super(struct super_block *sb)
197{
198 struct fuse_conn *fc = get_fuse_conn_super(sb);
199
200 down_write(&fc->sbput_sem);
201 while (!list_empty(&fc->background))
202 fuse_release_background(list_entry(fc->background.next,
203 struct fuse_req, bg_entry));
204
205 spin_lock(&fuse_lock);
206 fc->mounted = 0;
207 fc->user_id = 0;
208 fc->group_id = 0;
209 fc->flags = 0;
210 /* Flush all readers on this fs */
211 wake_up_all(&fc->waitq);
212 up_write(&fc->sbput_sem);
213 fuse_release_conn(fc);
214 spin_unlock(&fuse_lock);
215}
216
217static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
218{
219 stbuf->f_type = FUSE_SUPER_MAGIC;
220 stbuf->f_bsize = attr->bsize;
221 stbuf->f_blocks = attr->blocks;
222 stbuf->f_bfree = attr->bfree;
223 stbuf->f_bavail = attr->bavail;
224 stbuf->f_files = attr->files;
225 stbuf->f_ffree = attr->ffree;
226 stbuf->f_namelen = attr->namelen;
227 /* fsid is left zero */
228}
229
230static int fuse_statfs(struct super_block *sb, struct kstatfs *buf)
231{
232 struct fuse_conn *fc = get_fuse_conn_super(sb);
233 struct fuse_req *req;
234 struct fuse_statfs_out outarg;
235 int err;
236
237 req = fuse_get_request(fc);
238 if (!req)
239 return -EINTR;
240
241 req->in.numargs = 0;
242 req->in.h.opcode = FUSE_STATFS;
243 req->out.numargs = 1;
244 req->out.args[0].size = sizeof(outarg);
245 req->out.args[0].value = &outarg;
246 request_send(fc, req);
247 err = req->out.h.error;
248 if (!err)
249 convert_fuse_statfs(buf, &outarg.st);
250 fuse_put_request(fc, req);
251 return err;
252}
253
254enum {
255 OPT_FD,
256 OPT_ROOTMODE,
257 OPT_USER_ID,
258 OPT_GROUP_ID,
259 OPT_DEFAULT_PERMISSIONS,
260 OPT_ALLOW_OTHER,
261 OPT_MAX_READ,
262 OPT_ERR
263};
264
265static match_table_t tokens = {
266 {OPT_FD, "fd=%u"},
267 {OPT_ROOTMODE, "rootmode=%o"},
268 {OPT_USER_ID, "user_id=%u"},
269 {OPT_GROUP_ID, "group_id=%u"},
270 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
271 {OPT_ALLOW_OTHER, "allow_other"},
272 {OPT_MAX_READ, "max_read=%u"},
273 {OPT_ERR, NULL}
274};
275
276static int parse_fuse_opt(char *opt, struct fuse_mount_data *d)
277{
278 char *p;
279 memset(d, 0, sizeof(struct fuse_mount_data));
280 d->max_read = ~0;
281
282 while ((p = strsep(&opt, ",")) != NULL) {
283 int token;
284 int value;
285 substring_t args[MAX_OPT_ARGS];
286 if (!*p)
287 continue;
288
289 token = match_token(p, tokens, args);
290 switch (token) {
291 case OPT_FD:
292 if (match_int(&args[0], &value))
293 return 0;
294 d->fd = value;
295 d->fd_present = 1;
296 break;
297
298 case OPT_ROOTMODE:
299 if (match_octal(&args[0], &value))
300 return 0;
301 d->rootmode = value;
302 d->rootmode_present = 1;
303 break;
304
305 case OPT_USER_ID:
306 if (match_int(&args[0], &value))
307 return 0;
308 d->user_id = value;
309 d->user_id_present = 1;
310 break;
311
312 case OPT_GROUP_ID:
313 if (match_int(&args[0], &value))
314 return 0;
315 d->group_id = value;
316 d->group_id_present = 1;
317 break;
318
319 case OPT_DEFAULT_PERMISSIONS:
320 d->flags |= FUSE_DEFAULT_PERMISSIONS;
321 break;
322
323 case OPT_ALLOW_OTHER:
324 d->flags |= FUSE_ALLOW_OTHER;
325 break;
326
327 case OPT_MAX_READ:
328 if (match_int(&args[0], &value))
329 return 0;
330 d->max_read = value;
331 break;
332
333 default:
334 return 0;
335 }
336 }
337
338 if (!d->fd_present || !d->rootmode_present ||
339 !d->user_id_present || !d->group_id_present)
340 return 0;
341
342 return 1;
343}
344
345static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
346{
347 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
348
349 seq_printf(m, ",user_id=%u", fc->user_id);
350 seq_printf(m, ",group_id=%u", fc->group_id);
351 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
352 seq_puts(m, ",default_permissions");
353 if (fc->flags & FUSE_ALLOW_OTHER)
354 seq_puts(m, ",allow_other");
355 if (fc->max_read != ~0)
356 seq_printf(m, ",max_read=%u", fc->max_read);
357 return 0;
358}
359
360static void free_conn(struct fuse_conn *fc)
361{
362 while (!list_empty(&fc->unused_list)) {
363 struct fuse_req *req;
364 req = list_entry(fc->unused_list.next, struct fuse_req, list);
365 list_del(&req->list);
366 fuse_request_free(req);
367 }
368 kfree(fc);
369}
370
371/* Must be called with the fuse lock held */
372void fuse_release_conn(struct fuse_conn *fc)
373{
374 fc->count--;
375 if (!fc->count)
376 free_conn(fc);
377}
378
379static struct fuse_conn *new_conn(void)
380{
381 struct fuse_conn *fc;
382
383 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
384 if (fc != NULL) {
385 int i;
386 memset(fc, 0, sizeof(*fc));
387 init_waitqueue_head(&fc->waitq);
388 INIT_LIST_HEAD(&fc->pending);
389 INIT_LIST_HEAD(&fc->processing);
390 INIT_LIST_HEAD(&fc->unused_list);
391 INIT_LIST_HEAD(&fc->background);
392 sema_init(&fc->outstanding_sem, 0);
393 init_rwsem(&fc->sbput_sem);
394 for (i = 0; i < FUSE_MAX_OUTSTANDING; i++) {
395 struct fuse_req *req = fuse_request_alloc();
396 if (!req) {
397 free_conn(fc);
398 return NULL;
399 }
400 list_add(&req->list, &fc->unused_list);
401 }
402 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
403 fc->bdi.unplug_io_fn = default_unplug_io_fn;
404 fc->reqctr = 0;
405 }
406 return fc;
407}
408
409static struct fuse_conn *get_conn(struct file *file, struct super_block *sb)
410{
411 struct fuse_conn *fc;
412
413 if (file->f_op != &fuse_dev_operations)
414 return ERR_PTR(-EINVAL);
415 fc = new_conn();
416 if (fc == NULL)
417 return ERR_PTR(-ENOMEM);
418 spin_lock(&fuse_lock);
419 if (file->private_data) {
420 free_conn(fc);
421 fc = ERR_PTR(-EINVAL);
422 } else {
423 file->private_data = fc;
424 *get_fuse_conn_super_p(sb) = fc;
425 fc->mounted = 1;
426 fc->connected = 1;
427 fc->count = 2;
428 }
429 spin_unlock(&fuse_lock);
430 return fc;
431}
432
433static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
434{
435 struct fuse_attr attr;
436 memset(&attr, 0, sizeof(attr));
437
438 attr.mode = mode;
439 attr.ino = FUSE_ROOT_ID;
440 return fuse_iget(sb, 1, 0, &attr);
441}
442
443static struct super_operations fuse_super_operations = {
444 .alloc_inode = fuse_alloc_inode,
445 .destroy_inode = fuse_destroy_inode,
446 .read_inode = fuse_read_inode,
447 .clear_inode = fuse_clear_inode,
448 .put_super = fuse_put_super,
449 .statfs = fuse_statfs,
450 .show_options = fuse_show_options,
451};
452
453static int fuse_fill_super(struct super_block *sb, void *data, int silent)
454{
455 struct fuse_conn *fc;
456 struct inode *root;
457 struct fuse_mount_data d;
458 struct file *file;
459 int err;
460
461 if (!parse_fuse_opt((char *) data, &d))
462 return -EINVAL;
463
464 sb->s_blocksize = PAGE_CACHE_SIZE;
465 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
466 sb->s_magic = FUSE_SUPER_MAGIC;
467 sb->s_op = &fuse_super_operations;
468 sb->s_maxbytes = MAX_LFS_FILESIZE;
469
470 file = fget(d.fd);
471 if (!file)
472 return -EINVAL;
473
474 fc = get_conn(file, sb);
475 fput(file);
476 if (IS_ERR(fc))
477 return PTR_ERR(fc);
478
479 fc->flags = d.flags;
480 fc->user_id = d.user_id;
481 fc->group_id = d.group_id;
482 fc->max_read = d.max_read;
483 if (fc->max_read / PAGE_CACHE_SIZE < fc->bdi.ra_pages)
484 fc->bdi.ra_pages = fc->max_read / PAGE_CACHE_SIZE;
485 fc->max_write = FUSE_MAX_IN / 2;
486
487 err = -ENOMEM;
488 root = get_root_inode(sb, d.rootmode);
489 if (root == NULL)
490 goto err;
491
492 sb->s_root = d_alloc_root(root);
493 if (!sb->s_root) {
494 iput(root);
495 goto err;
496 }
497 fuse_send_init(fc);
498 return 0;
499
500 err:
501 spin_lock(&fuse_lock);
502 fuse_release_conn(fc);
503 spin_unlock(&fuse_lock);
504 return err;
505}
506
507static struct super_block *fuse_get_sb(struct file_system_type *fs_type,
508 int flags, const char *dev_name,
509 void *raw_data)
510{
511 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super);
512}
513
514static struct file_system_type fuse_fs_type = {
515 .owner = THIS_MODULE,
516 .name = "fuse",
517 .get_sb = fuse_get_sb,
518 .kill_sb = kill_anon_super,
519};
520
521static void fuse_inode_init_once(void *foo, kmem_cache_t *cachep,
522 unsigned long flags)
523{
524 struct inode * inode = foo;
525
526 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
527 SLAB_CTOR_CONSTRUCTOR)
528 inode_init_once(inode);
529}
530
531static int __init fuse_fs_init(void)
532{
533 int err;
534
535 err = register_filesystem(&fuse_fs_type);
536 if (err)
537 printk("fuse: failed to register filesystem\n");
538 else {
539 fuse_inode_cachep = kmem_cache_create("fuse_inode",
540 sizeof(struct fuse_inode),
541 0, SLAB_HWCACHE_ALIGN,
542 fuse_inode_init_once, NULL);
543 if (!fuse_inode_cachep) {
544 unregister_filesystem(&fuse_fs_type);
545 err = -ENOMEM;
546 }
547 }
548
549 return err;
550}
551
552static void fuse_fs_cleanup(void)
553{
554 unregister_filesystem(&fuse_fs_type);
555 kmem_cache_destroy(fuse_inode_cachep);
556}
557
558static int __init fuse_init(void)
559{
560 int res;
561
562 printk("fuse init (API version %i.%i)\n",
563 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
564
565 spin_lock_init(&fuse_lock);
566 res = fuse_fs_init();
567 if (res)
568 goto err;
569
570 res = fuse_dev_init();
571 if (res)
572 goto err_fs_cleanup;
573
574 return 0;
575
576 err_fs_cleanup:
577 fuse_fs_cleanup();
578 err:
579 return res;
580}
581
582static void __exit fuse_exit(void)
583{
584 printk(KERN_DEBUG "fuse exit\n");
585
586 fuse_fs_cleanup();
587 fuse_dev_cleanup();
588}
589
590module_init(fuse_init);
591module_exit(fuse_exit);