2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2006 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/kernel.h>
15 static const struct file_operations fuse_direct_io_file_operations
;
17 static int fuse_send_open(struct inode
*inode
, struct file
*file
, int isdir
,
18 struct fuse_open_out
*outargp
)
20 struct fuse_conn
*fc
= get_fuse_conn(inode
);
21 struct fuse_open_in inarg
;
25 req
= fuse_get_req(fc
);
29 memset(&inarg
, 0, sizeof(inarg
));
30 inarg
.flags
= file
->f_flags
& ~(O_CREAT
| O_EXCL
| O_NOCTTY
| O_TRUNC
);
31 req
->in
.h
.opcode
= isdir
? FUSE_OPENDIR
: FUSE_OPEN
;
32 req
->in
.h
.nodeid
= get_node_id(inode
);
34 req
->in
.args
[0].size
= sizeof(inarg
);
35 req
->in
.args
[0].value
= &inarg
;
37 req
->out
.args
[0].size
= sizeof(*outargp
);
38 req
->out
.args
[0].value
= outargp
;
39 request_send(fc
, req
);
40 err
= req
->out
.h
.error
;
41 fuse_put_request(fc
, req
);
46 struct fuse_file
*fuse_file_alloc(void)
49 ff
= kmalloc(sizeof(struct fuse_file
), GFP_KERNEL
);
51 ff
->reserved_req
= fuse_request_alloc();
52 if (!ff
->reserved_req
) {
60 void fuse_file_free(struct fuse_file
*ff
)
62 fuse_request_free(ff
->reserved_req
);
66 void fuse_finish_open(struct inode
*inode
, struct file
*file
,
67 struct fuse_file
*ff
, struct fuse_open_out
*outarg
)
69 if (outarg
->open_flags
& FOPEN_DIRECT_IO
)
70 file
->f_op
= &fuse_direct_io_file_operations
;
71 if (!(outarg
->open_flags
& FOPEN_KEEP_CACHE
))
72 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
74 file
->private_data
= ff
;
77 int fuse_open_common(struct inode
*inode
, struct file
*file
, int isdir
)
79 struct fuse_open_out outarg
;
83 /* VFS checks this, but only _after_ ->open() */
84 if (file
->f_flags
& O_DIRECT
)
87 err
= generic_file_open(inode
, file
);
91 /* If opening the root node, no lookup has been performed on
92 it, so the attributes must be refreshed */
93 if (get_node_id(inode
) == FUSE_ROOT_ID
) {
94 err
= fuse_do_getattr(inode
);
99 ff
= fuse_file_alloc();
103 err
= fuse_send_open(inode
, file
, isdir
, &outarg
);
108 outarg
.open_flags
&= ~FOPEN_DIRECT_IO
;
109 fuse_finish_open(inode
, file
, ff
, &outarg
);
115 struct fuse_req
*fuse_release_fill(struct fuse_file
*ff
, u64 nodeid
, int flags
,
118 struct fuse_req
*req
= ff
->reserved_req
;
119 struct fuse_release_in
*inarg
= &req
->misc
.release_in
;
122 inarg
->flags
= flags
;
123 req
->in
.h
.opcode
= opcode
;
124 req
->in
.h
.nodeid
= nodeid
;
126 req
->in
.args
[0].size
= sizeof(struct fuse_release_in
);
127 req
->in
.args
[0].value
= inarg
;
133 int fuse_release_common(struct inode
*inode
, struct file
*file
, int isdir
)
135 struct fuse_file
*ff
= file
->private_data
;
137 struct fuse_conn
*fc
= get_fuse_conn(inode
);
138 struct fuse_req
*req
;
140 req
= fuse_release_fill(ff
, get_node_id(inode
), file
->f_flags
,
141 isdir
? FUSE_RELEASEDIR
: FUSE_RELEASE
);
143 /* Hold vfsmount and dentry until release is finished */
144 req
->vfsmount
= mntget(file
->f_path
.mnt
);
145 req
->dentry
= dget(file
->f_path
.dentry
);
146 request_send_background(fc
, req
);
149 /* Return value is ignored by VFS */
153 static int fuse_open(struct inode
*inode
, struct file
*file
)
155 return fuse_open_common(inode
, file
, 0);
158 static int fuse_release(struct inode
*inode
, struct file
*file
)
160 return fuse_release_common(inode
, file
, 0);
164 * Scramble the ID space with XTEA, so that the value of the files_struct
165 * pointer is not exposed to userspace.
167 static u64
fuse_lock_owner_id(struct fuse_conn
*fc
, fl_owner_t id
)
169 u32
*k
= fc
->scramble_key
;
170 u64 v
= (unsigned long) id
;
176 for (i
= 0; i
< 32; i
++) {
177 v0
+= ((v1
<< 4 ^ v1
>> 5) + v1
) ^ (sum
+ k
[sum
& 3]);
179 v1
+= ((v0
<< 4 ^ v0
>> 5) + v0
) ^ (sum
+ k
[sum
>>11 & 3]);
182 return (u64
) v0
+ ((u64
) v1
<< 32);
185 static int fuse_flush(struct file
*file
, fl_owner_t id
)
187 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
188 struct fuse_conn
*fc
= get_fuse_conn(inode
);
189 struct fuse_file
*ff
= file
->private_data
;
190 struct fuse_req
*req
;
191 struct fuse_flush_in inarg
;
194 if (is_bad_inode(inode
))
200 req
= fuse_get_req_nofail(fc
, file
);
201 memset(&inarg
, 0, sizeof(inarg
));
203 inarg
.lock_owner
= fuse_lock_owner_id(fc
, id
);
204 req
->in
.h
.opcode
= FUSE_FLUSH
;
205 req
->in
.h
.nodeid
= get_node_id(inode
);
207 req
->in
.args
[0].size
= sizeof(inarg
);
208 req
->in
.args
[0].value
= &inarg
;
210 request_send(fc
, req
);
211 err
= req
->out
.h
.error
;
212 fuse_put_request(fc
, req
);
213 if (err
== -ENOSYS
) {
220 int fuse_fsync_common(struct file
*file
, struct dentry
*de
, int datasync
,
223 struct inode
*inode
= de
->d_inode
;
224 struct fuse_conn
*fc
= get_fuse_conn(inode
);
225 struct fuse_file
*ff
= file
->private_data
;
226 struct fuse_req
*req
;
227 struct fuse_fsync_in inarg
;
230 if (is_bad_inode(inode
))
233 if ((!isdir
&& fc
->no_fsync
) || (isdir
&& fc
->no_fsyncdir
))
236 req
= fuse_get_req(fc
);
240 memset(&inarg
, 0, sizeof(inarg
));
242 inarg
.fsync_flags
= datasync
? 1 : 0;
243 req
->in
.h
.opcode
= isdir
? FUSE_FSYNCDIR
: FUSE_FSYNC
;
244 req
->in
.h
.nodeid
= get_node_id(inode
);
246 req
->in
.args
[0].size
= sizeof(inarg
);
247 req
->in
.args
[0].value
= &inarg
;
248 request_send(fc
, req
);
249 err
= req
->out
.h
.error
;
250 fuse_put_request(fc
, req
);
251 if (err
== -ENOSYS
) {
261 static int fuse_fsync(struct file
*file
, struct dentry
*de
, int datasync
)
263 return fuse_fsync_common(file
, de
, datasync
, 0);
266 void fuse_read_fill(struct fuse_req
*req
, struct file
*file
,
267 struct inode
*inode
, loff_t pos
, size_t count
, int opcode
)
269 struct fuse_file
*ff
= file
->private_data
;
270 struct fuse_read_in
*inarg
= &req
->misc
.read_in
;
275 req
->in
.h
.opcode
= opcode
;
276 req
->in
.h
.nodeid
= get_node_id(inode
);
278 req
->in
.args
[0].size
= sizeof(struct fuse_read_in
);
279 req
->in
.args
[0].value
= inarg
;
280 req
->out
.argpages
= 1;
282 req
->out
.numargs
= 1;
283 req
->out
.args
[0].size
= count
;
286 static size_t fuse_send_read(struct fuse_req
*req
, struct file
*file
,
287 struct inode
*inode
, loff_t pos
, size_t count
)
289 struct fuse_conn
*fc
= get_fuse_conn(inode
);
290 fuse_read_fill(req
, file
, inode
, pos
, count
, FUSE_READ
);
291 request_send(fc
, req
);
292 return req
->out
.args
[0].size
;
295 static int fuse_readpage(struct file
*file
, struct page
*page
)
297 struct inode
*inode
= page
->mapping
->host
;
298 struct fuse_conn
*fc
= get_fuse_conn(inode
);
299 struct fuse_req
*req
;
303 if (is_bad_inode(inode
))
306 req
= fuse_get_req(fc
);
311 req
->out
.page_zeroing
= 1;
313 req
->pages
[0] = page
;
314 fuse_send_read(req
, file
, inode
, page_offset(page
), PAGE_CACHE_SIZE
);
315 err
= req
->out
.h
.error
;
316 fuse_put_request(fc
, req
);
318 SetPageUptodate(page
);
319 fuse_invalidate_attr(inode
); /* atime changed */
325 static void fuse_readpages_end(struct fuse_conn
*fc
, struct fuse_req
*req
)
329 fuse_invalidate_attr(req
->pages
[0]->mapping
->host
); /* atime changed */
331 for (i
= 0; i
< req
->num_pages
; i
++) {
332 struct page
*page
= req
->pages
[i
];
333 if (!req
->out
.h
.error
)
334 SetPageUptodate(page
);
339 fuse_put_request(fc
, req
);
342 static void fuse_send_readpages(struct fuse_req
*req
, struct file
*file
,
345 struct fuse_conn
*fc
= get_fuse_conn(inode
);
346 loff_t pos
= page_offset(req
->pages
[0]);
347 size_t count
= req
->num_pages
<< PAGE_CACHE_SHIFT
;
348 req
->out
.page_zeroing
= 1;
349 fuse_read_fill(req
, file
, inode
, pos
, count
, FUSE_READ
);
350 if (fc
->async_read
) {
353 req
->end
= fuse_readpages_end
;
354 request_send_background(fc
, req
);
356 request_send(fc
, req
);
357 fuse_readpages_end(fc
, req
);
361 struct fuse_readpages_data
{
362 struct fuse_req
*req
;
367 static int fuse_readpages_fill(void *_data
, struct page
*page
)
369 struct fuse_readpages_data
*data
= _data
;
370 struct fuse_req
*req
= data
->req
;
371 struct inode
*inode
= data
->inode
;
372 struct fuse_conn
*fc
= get_fuse_conn(inode
);
374 if (req
->num_pages
&&
375 (req
->num_pages
== FUSE_MAX_PAGES_PER_REQ
||
376 (req
->num_pages
+ 1) * PAGE_CACHE_SIZE
> fc
->max_read
||
377 req
->pages
[req
->num_pages
- 1]->index
+ 1 != page
->index
)) {
378 fuse_send_readpages(req
, data
->file
, inode
);
379 data
->req
= req
= fuse_get_req(fc
);
385 req
->pages
[req
->num_pages
] = page
;
390 static int fuse_readpages(struct file
*file
, struct address_space
*mapping
,
391 struct list_head
*pages
, unsigned nr_pages
)
393 struct inode
*inode
= mapping
->host
;
394 struct fuse_conn
*fc
= get_fuse_conn(inode
);
395 struct fuse_readpages_data data
;
399 if (is_bad_inode(inode
))
404 data
.req
= fuse_get_req(fc
);
405 err
= PTR_ERR(data
.req
);
406 if (IS_ERR(data
.req
))
409 err
= read_cache_pages(mapping
, pages
, fuse_readpages_fill
, &data
);
411 if (data
.req
->num_pages
)
412 fuse_send_readpages(data
.req
, file
, inode
);
414 fuse_put_request(fc
, data
.req
);
420 static size_t fuse_send_write(struct fuse_req
*req
, struct file
*file
,
421 struct inode
*inode
, loff_t pos
, size_t count
)
423 struct fuse_conn
*fc
= get_fuse_conn(inode
);
424 struct fuse_file
*ff
= file
->private_data
;
425 struct fuse_write_in inarg
;
426 struct fuse_write_out outarg
;
428 memset(&inarg
, 0, sizeof(struct fuse_write_in
));
432 req
->in
.h
.opcode
= FUSE_WRITE
;
433 req
->in
.h
.nodeid
= get_node_id(inode
);
434 req
->in
.argpages
= 1;
436 req
->in
.args
[0].size
= sizeof(struct fuse_write_in
);
437 req
->in
.args
[0].value
= &inarg
;
438 req
->in
.args
[1].size
= count
;
439 req
->out
.numargs
= 1;
440 req
->out
.args
[0].size
= sizeof(struct fuse_write_out
);
441 req
->out
.args
[0].value
= &outarg
;
442 request_send(fc
, req
);
446 static int fuse_prepare_write(struct file
*file
, struct page
*page
,
447 unsigned offset
, unsigned to
)
453 static int fuse_commit_write(struct file
*file
, struct page
*page
,
454 unsigned offset
, unsigned to
)
458 unsigned count
= to
- offset
;
459 struct inode
*inode
= page
->mapping
->host
;
460 struct fuse_conn
*fc
= get_fuse_conn(inode
);
461 loff_t pos
= page_offset(page
) + offset
;
462 struct fuse_req
*req
;
464 if (is_bad_inode(inode
))
467 req
= fuse_get_req(fc
);
472 req
->pages
[0] = page
;
473 req
->page_offset
= offset
;
474 nres
= fuse_send_write(req
, file
, inode
, pos
, count
);
475 err
= req
->out
.h
.error
;
476 fuse_put_request(fc
, req
);
477 if (!err
&& nres
!= count
)
481 spin_lock(&fc
->lock
);
482 if (pos
> inode
->i_size
)
483 i_size_write(inode
, pos
);
484 spin_unlock(&fc
->lock
);
486 if (offset
== 0 && to
== PAGE_CACHE_SIZE
)
487 SetPageUptodate(page
);
489 fuse_invalidate_attr(inode
);
493 static void fuse_release_user_pages(struct fuse_req
*req
, int write
)
497 for (i
= 0; i
< req
->num_pages
; i
++) {
498 struct page
*page
= req
->pages
[i
];
500 set_page_dirty_lock(page
);
505 static int fuse_get_user_pages(struct fuse_req
*req
, const char __user
*buf
,
506 unsigned nbytes
, int write
)
508 unsigned long user_addr
= (unsigned long) buf
;
509 unsigned offset
= user_addr
& ~PAGE_MASK
;
512 /* This doesn't work with nfsd */
516 nbytes
= min(nbytes
, (unsigned) FUSE_MAX_PAGES_PER_REQ
<< PAGE_SHIFT
);
517 npages
= (nbytes
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
518 npages
= min(max(npages
, 1), FUSE_MAX_PAGES_PER_REQ
);
519 down_read(¤t
->mm
->mmap_sem
);
520 npages
= get_user_pages(current
, current
->mm
, user_addr
, npages
, write
,
521 0, req
->pages
, NULL
);
522 up_read(¤t
->mm
->mmap_sem
);
526 req
->num_pages
= npages
;
527 req
->page_offset
= offset
;
531 static ssize_t
fuse_direct_io(struct file
*file
, const char __user
*buf
,
532 size_t count
, loff_t
*ppos
, int write
)
534 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
535 struct fuse_conn
*fc
= get_fuse_conn(inode
);
536 size_t nmax
= write
? fc
->max_write
: fc
->max_read
;
539 struct fuse_req
*req
;
541 if (is_bad_inode(inode
))
544 req
= fuse_get_req(fc
);
550 size_t nbytes
= min(count
, nmax
);
551 int err
= fuse_get_user_pages(req
, buf
, nbytes
, !write
);
556 nbytes
= (req
->num_pages
<< PAGE_SHIFT
) - req
->page_offset
;
557 nbytes
= min(count
, nbytes
);
559 nres
= fuse_send_write(req
, file
, inode
, pos
, nbytes
);
561 nres
= fuse_send_read(req
, file
, inode
, pos
, nbytes
);
562 fuse_release_user_pages(req
, !write
);
563 if (req
->out
.h
.error
) {
565 res
= req
->out
.h
.error
;
567 } else if (nres
> nbytes
) {
578 fuse_put_request(fc
, req
);
579 req
= fuse_get_req(fc
);
584 fuse_put_request(fc
, req
);
587 spin_lock(&fc
->lock
);
588 if (pos
> inode
->i_size
)
589 i_size_write(inode
, pos
);
590 spin_unlock(&fc
->lock
);
594 fuse_invalidate_attr(inode
);
599 static ssize_t
fuse_direct_read(struct file
*file
, char __user
*buf
,
600 size_t count
, loff_t
*ppos
)
602 return fuse_direct_io(file
, buf
, count
, ppos
, 0);
605 static ssize_t
fuse_direct_write(struct file
*file
, const char __user
*buf
,
606 size_t count
, loff_t
*ppos
)
608 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
610 /* Don't allow parallel writes to the same file */
611 mutex_lock(&inode
->i_mutex
);
612 res
= fuse_direct_io(file
, buf
, count
, ppos
, 1);
613 mutex_unlock(&inode
->i_mutex
);
617 static int fuse_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
619 if ((vma
->vm_flags
& VM_SHARED
)) {
620 if ((vma
->vm_flags
& VM_WRITE
))
623 vma
->vm_flags
&= ~VM_MAYWRITE
;
625 return generic_file_mmap(file
, vma
);
628 static int fuse_set_page_dirty(struct page
*page
)
630 printk("fuse_set_page_dirty: should not happen\n");
635 static int convert_fuse_file_lock(const struct fuse_file_lock
*ffl
,
636 struct file_lock
*fl
)
644 if (ffl
->start
> OFFSET_MAX
|| ffl
->end
> OFFSET_MAX
||
645 ffl
->end
< ffl
->start
)
648 fl
->fl_start
= ffl
->start
;
649 fl
->fl_end
= ffl
->end
;
650 fl
->fl_pid
= ffl
->pid
;
656 fl
->fl_type
= ffl
->type
;
660 static void fuse_lk_fill(struct fuse_req
*req
, struct file
*file
,
661 const struct file_lock
*fl
, int opcode
, pid_t pid
)
663 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
664 struct fuse_conn
*fc
= get_fuse_conn(inode
);
665 struct fuse_file
*ff
= file
->private_data
;
666 struct fuse_lk_in
*arg
= &req
->misc
.lk_in
;
669 arg
->owner
= fuse_lock_owner_id(fc
, fl
->fl_owner
);
670 arg
->lk
.start
= fl
->fl_start
;
671 arg
->lk
.end
= fl
->fl_end
;
672 arg
->lk
.type
= fl
->fl_type
;
674 req
->in
.h
.opcode
= opcode
;
675 req
->in
.h
.nodeid
= get_node_id(inode
);
677 req
->in
.args
[0].size
= sizeof(*arg
);
678 req
->in
.args
[0].value
= arg
;
681 static int fuse_getlk(struct file
*file
, struct file_lock
*fl
)
683 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
684 struct fuse_conn
*fc
= get_fuse_conn(inode
);
685 struct fuse_req
*req
;
686 struct fuse_lk_out outarg
;
689 req
= fuse_get_req(fc
);
693 fuse_lk_fill(req
, file
, fl
, FUSE_GETLK
, 0);
694 req
->out
.numargs
= 1;
695 req
->out
.args
[0].size
= sizeof(outarg
);
696 req
->out
.args
[0].value
= &outarg
;
697 request_send(fc
, req
);
698 err
= req
->out
.h
.error
;
699 fuse_put_request(fc
, req
);
701 err
= convert_fuse_file_lock(&outarg
.lk
, fl
);
706 static int fuse_setlk(struct file
*file
, struct file_lock
*fl
)
708 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
709 struct fuse_conn
*fc
= get_fuse_conn(inode
);
710 struct fuse_req
*req
;
711 int opcode
= (fl
->fl_flags
& FL_SLEEP
) ? FUSE_SETLKW
: FUSE_SETLK
;
712 pid_t pid
= fl
->fl_type
!= F_UNLCK
? current
->tgid
: 0;
715 /* Unlock on close is handled by the flush method */
716 if (fl
->fl_flags
& FL_CLOSE
)
719 req
= fuse_get_req(fc
);
723 fuse_lk_fill(req
, file
, fl
, opcode
, pid
);
724 request_send(fc
, req
);
725 err
= req
->out
.h
.error
;
726 /* locking is restartable */
729 fuse_put_request(fc
, req
);
733 static int fuse_file_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
735 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
736 struct fuse_conn
*fc
= get_fuse_conn(inode
);
739 if (cmd
== F_GETLK
) {
741 if (!posix_test_lock(file
, fl
, fl
))
742 fl
->fl_type
= F_UNLCK
;
745 err
= fuse_getlk(file
, fl
);
748 err
= posix_lock_file_wait(file
, fl
);
750 err
= fuse_setlk(file
, fl
);
755 static sector_t
fuse_bmap(struct address_space
*mapping
, sector_t block
)
757 struct inode
*inode
= mapping
->host
;
758 struct fuse_conn
*fc
= get_fuse_conn(inode
);
759 struct fuse_req
*req
;
760 struct fuse_bmap_in inarg
;
761 struct fuse_bmap_out outarg
;
764 if (!inode
->i_sb
->s_bdev
|| fc
->no_bmap
)
767 req
= fuse_get_req(fc
);
771 memset(&inarg
, 0, sizeof(inarg
));
773 inarg
.blocksize
= inode
->i_sb
->s_blocksize
;
774 req
->in
.h
.opcode
= FUSE_BMAP
;
775 req
->in
.h
.nodeid
= get_node_id(inode
);
777 req
->in
.args
[0].size
= sizeof(inarg
);
778 req
->in
.args
[0].value
= &inarg
;
779 req
->out
.numargs
= 1;
780 req
->out
.args
[0].size
= sizeof(outarg
);
781 req
->out
.args
[0].value
= &outarg
;
782 request_send(fc
, req
);
783 err
= req
->out
.h
.error
;
784 fuse_put_request(fc
, req
);
788 return err
? 0 : outarg
.block
;
791 static const struct file_operations fuse_file_operations
= {
792 .llseek
= generic_file_llseek
,
793 .read
= do_sync_read
,
794 .aio_read
= generic_file_aio_read
,
795 .write
= do_sync_write
,
796 .aio_write
= generic_file_aio_write
,
797 .mmap
= fuse_file_mmap
,
800 .release
= fuse_release
,
802 .lock
= fuse_file_lock
,
803 .sendfile
= generic_file_sendfile
,
806 static const struct file_operations fuse_direct_io_file_operations
= {
807 .llseek
= generic_file_llseek
,
808 .read
= fuse_direct_read
,
809 .write
= fuse_direct_write
,
812 .release
= fuse_release
,
814 .lock
= fuse_file_lock
,
815 /* no mmap and sendfile */
818 static const struct address_space_operations fuse_file_aops
= {
819 .readpage
= fuse_readpage
,
820 .prepare_write
= fuse_prepare_write
,
821 .commit_write
= fuse_commit_write
,
822 .readpages
= fuse_readpages
,
823 .set_page_dirty
= fuse_set_page_dirty
,
827 void fuse_init_file_inode(struct inode
*inode
)
829 inode
->i_fop
= &fuse_file_operations
;
830 inode
->i_data
.a_ops
= &fuse_file_aops
;