2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 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/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
28 static struct kmem_cache
*fuse_inode_cachep
;
29 struct list_head fuse_conn_list
;
30 DEFINE_MUTEX(fuse_mutex
);
32 static int set_global_limit(const char *val
, struct kernel_param
*kp
);
34 unsigned max_user_bgreq
;
35 module_param_call(max_user_bgreq
, set_global_limit
, param_get_uint
,
36 &max_user_bgreq
, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq
, "uint");
38 MODULE_PARM_DESC(max_user_bgreq
,
39 "Global limit for the maximum number of backgrounded requests an "
40 "unprivileged user can set");
42 unsigned max_user_congthresh
;
43 module_param_call(max_user_congthresh
, set_global_limit
, param_get_uint
,
44 &max_user_congthresh
, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh
, "uint");
46 MODULE_PARM_DESC(max_user_congthresh
,
47 "Global limit for the maximum congestion threshold an "
48 "unprivileged user can set");
50 #define FUSE_SUPER_MAGIC 0x65735546
52 #define FUSE_DEFAULT_BLKSIZE 512
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
60 struct fuse_mount_data
{
65 unsigned fd_present
:1;
66 unsigned rootmode_present
:1;
67 unsigned user_id_present
:1;
68 unsigned group_id_present
:1;
74 struct fuse_forget_link
*fuse_alloc_forget(void)
76 return kzalloc(sizeof(struct fuse_forget_link
), GFP_KERNEL
);
79 static struct inode
*fuse_alloc_inode(struct super_block
*sb
)
82 struct fuse_inode
*fi
;
84 inode
= kmem_cache_alloc(fuse_inode_cachep
, GFP_KERNEL
);
88 fi
= get_fuse_inode(inode
);
96 INIT_LIST_HEAD(&fi
->write_files
);
97 INIT_LIST_HEAD(&fi
->queued_writes
);
98 INIT_LIST_HEAD(&fi
->writepages
);
99 init_waitqueue_head(&fi
->page_waitq
);
100 fi
->forget
= fuse_alloc_forget();
102 kmem_cache_free(fuse_inode_cachep
, inode
);
109 static void fuse_i_callback(struct rcu_head
*head
)
111 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
112 kmem_cache_free(fuse_inode_cachep
, inode
);
115 static void fuse_destroy_inode(struct inode
*inode
)
117 struct fuse_inode
*fi
= get_fuse_inode(inode
);
118 BUG_ON(!list_empty(&fi
->write_files
));
119 BUG_ON(!list_empty(&fi
->queued_writes
));
121 call_rcu(&inode
->i_rcu
, fuse_i_callback
);
124 static void fuse_evict_inode(struct inode
*inode
)
126 truncate_inode_pages_final(&inode
->i_data
);
128 if (inode
->i_sb
->s_flags
& MS_ACTIVE
) {
129 struct fuse_conn
*fc
= get_fuse_conn(inode
);
130 struct fuse_inode
*fi
= get_fuse_inode(inode
);
131 fuse_queue_forget(fc
, fi
->forget
, fi
->nodeid
, fi
->nlookup
);
136 static int fuse_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
139 if (*flags
& MS_MANDLOCK
)
146 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
147 * so that it will fit.
149 static ino_t
fuse_squash_ino(u64 ino64
)
151 ino_t ino
= (ino_t
) ino64
;
152 if (sizeof(ino_t
) < sizeof(u64
))
153 ino
^= ino64
>> (sizeof(u64
) - sizeof(ino_t
)) * 8;
157 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
160 struct fuse_conn
*fc
= get_fuse_conn(inode
);
161 struct fuse_inode
*fi
= get_fuse_inode(inode
);
163 fi
->attr_version
= ++fc
->attr_version
;
164 fi
->i_time
= attr_valid
;
166 inode
->i_ino
= fuse_squash_ino(attr
->ino
);
167 inode
->i_mode
= (inode
->i_mode
& S_IFMT
) | (attr
->mode
& 07777);
168 set_nlink(inode
, attr
->nlink
);
169 inode
->i_uid
= make_kuid(&init_user_ns
, attr
->uid
);
170 inode
->i_gid
= make_kgid(&init_user_ns
, attr
->gid
);
171 inode
->i_blocks
= attr
->blocks
;
172 inode
->i_atime
.tv_sec
= attr
->atime
;
173 inode
->i_atime
.tv_nsec
= attr
->atimensec
;
174 /* mtime from server may be stale due to local buffered write */
175 if (!fc
->writeback_cache
|| !S_ISREG(inode
->i_mode
)) {
176 inode
->i_mtime
.tv_sec
= attr
->mtime
;
177 inode
->i_mtime
.tv_nsec
= attr
->mtimensec
;
178 inode
->i_ctime
.tv_sec
= attr
->ctime
;
179 inode
->i_ctime
.tv_nsec
= attr
->ctimensec
;
182 if (attr
->blksize
!= 0)
183 inode
->i_blkbits
= ilog2(attr
->blksize
);
185 inode
->i_blkbits
= inode
->i_sb
->s_blocksize_bits
;
188 * Don't set the sticky bit in i_mode, unless we want the VFS
189 * to check permissions. This prevents failures due to the
190 * check in may_delete().
192 fi
->orig_i_mode
= inode
->i_mode
;
193 if (!(fc
->flags
& FUSE_DEFAULT_PERMISSIONS
))
194 inode
->i_mode
&= ~S_ISVTX
;
196 fi
->orig_ino
= attr
->ino
;
199 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
200 u64 attr_valid
, u64 attr_version
)
202 struct fuse_conn
*fc
= get_fuse_conn(inode
);
203 struct fuse_inode
*fi
= get_fuse_inode(inode
);
204 bool is_wb
= fc
->writeback_cache
;
206 struct timespec old_mtime
;
208 spin_lock(&fc
->lock
);
209 if ((attr_version
!= 0 && fi
->attr_version
> attr_version
) ||
210 test_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
)) {
211 spin_unlock(&fc
->lock
);
215 old_mtime
= inode
->i_mtime
;
216 fuse_change_attributes_common(inode
, attr
, attr_valid
);
218 oldsize
= inode
->i_size
;
220 * In case of writeback_cache enabled, the cached writes beyond EOF
221 * extend local i_size without keeping userspace server in sync. So,
222 * attr->size coming from server can be stale. We cannot trust it.
224 if (!is_wb
|| !S_ISREG(inode
->i_mode
))
225 i_size_write(inode
, attr
->size
);
226 spin_unlock(&fc
->lock
);
228 if (!is_wb
&& S_ISREG(inode
->i_mode
)) {
231 if (oldsize
!= attr
->size
) {
232 truncate_pagecache(inode
, attr
->size
);
234 } else if (fc
->auto_inval_data
) {
235 struct timespec new_mtime
= {
236 .tv_sec
= attr
->mtime
,
237 .tv_nsec
= attr
->mtimensec
,
241 * Auto inval mode also checks and invalidates if mtime
244 if (!timespec_equal(&old_mtime
, &new_mtime
))
249 invalidate_inode_pages2(inode
->i_mapping
);
253 static void fuse_init_inode(struct inode
*inode
, struct fuse_attr
*attr
)
255 inode
->i_mode
= attr
->mode
& S_IFMT
;
256 inode
->i_size
= attr
->size
;
257 inode
->i_mtime
.tv_sec
= attr
->mtime
;
258 inode
->i_mtime
.tv_nsec
= attr
->mtimensec
;
259 inode
->i_ctime
.tv_sec
= attr
->ctime
;
260 inode
->i_ctime
.tv_nsec
= attr
->ctimensec
;
261 if (S_ISREG(inode
->i_mode
)) {
262 fuse_init_common(inode
);
263 fuse_init_file_inode(inode
);
264 } else if (S_ISDIR(inode
->i_mode
))
265 fuse_init_dir(inode
);
266 else if (S_ISLNK(inode
->i_mode
))
267 fuse_init_symlink(inode
);
268 else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
269 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
270 fuse_init_common(inode
);
271 init_special_inode(inode
, inode
->i_mode
,
272 new_decode_dev(attr
->rdev
));
277 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
)
279 u64 nodeid
= *(u64
*) _nodeidp
;
280 if (get_node_id(inode
) == nodeid
)
286 static int fuse_inode_set(struct inode
*inode
, void *_nodeidp
)
288 u64 nodeid
= *(u64
*) _nodeidp
;
289 get_fuse_inode(inode
)->nodeid
= nodeid
;
293 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
294 int generation
, struct fuse_attr
*attr
,
295 u64 attr_valid
, u64 attr_version
)
298 struct fuse_inode
*fi
;
299 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
302 inode
= iget5_locked(sb
, nodeid
, fuse_inode_eq
, fuse_inode_set
, &nodeid
);
306 if ((inode
->i_state
& I_NEW
)) {
307 inode
->i_flags
|= S_NOATIME
;
308 if (!fc
->writeback_cache
|| !S_ISREG(attr
->mode
))
309 inode
->i_flags
|= S_NOCMTIME
;
310 inode
->i_generation
= generation
;
311 fuse_init_inode(inode
, attr
);
312 unlock_new_inode(inode
);
313 } else if ((inode
->i_mode
^ attr
->mode
) & S_IFMT
) {
314 /* Inode has changed type, any I/O on the old should fail */
315 make_bad_inode(inode
);
320 fi
= get_fuse_inode(inode
);
321 spin_lock(&fc
->lock
);
323 spin_unlock(&fc
->lock
);
324 fuse_change_attributes(inode
, attr
, attr_valid
, attr_version
);
329 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
330 loff_t offset
, loff_t len
)
336 inode
= ilookup5(sb
, nodeid
, fuse_inode_eq
, &nodeid
);
340 fuse_invalidate_attr(inode
);
342 pg_start
= offset
>> PAGE_CACHE_SHIFT
;
346 pg_end
= (offset
+ len
- 1) >> PAGE_CACHE_SHIFT
;
347 invalidate_inode_pages2_range(inode
->i_mapping
,
354 static void fuse_umount_begin(struct super_block
*sb
)
356 fuse_abort_conn(get_fuse_conn_super(sb
));
359 static void fuse_send_destroy(struct fuse_conn
*fc
)
361 struct fuse_req
*req
= fc
->destroy_req
;
362 if (req
&& fc
->conn_init
) {
363 fc
->destroy_req
= NULL
;
364 req
->in
.h
.opcode
= FUSE_DESTROY
;
367 fuse_request_send(fc
, req
);
368 fuse_put_request(fc
, req
);
372 static void fuse_bdi_destroy(struct fuse_conn
*fc
)
374 if (fc
->bdi_initialized
)
375 bdi_destroy(&fc
->bdi
);
378 static void fuse_put_super(struct super_block
*sb
)
380 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
382 fuse_send_destroy(fc
);
385 mutex_lock(&fuse_mutex
);
386 list_del(&fc
->entry
);
387 fuse_ctl_remove_conn(fc
);
388 mutex_unlock(&fuse_mutex
);
389 fuse_bdi_destroy(fc
);
394 static void convert_fuse_statfs(struct kstatfs
*stbuf
, struct fuse_kstatfs
*attr
)
396 stbuf
->f_type
= FUSE_SUPER_MAGIC
;
397 stbuf
->f_bsize
= attr
->bsize
;
398 stbuf
->f_frsize
= attr
->frsize
;
399 stbuf
->f_blocks
= attr
->blocks
;
400 stbuf
->f_bfree
= attr
->bfree
;
401 stbuf
->f_bavail
= attr
->bavail
;
402 stbuf
->f_files
= attr
->files
;
403 stbuf
->f_ffree
= attr
->ffree
;
404 stbuf
->f_namelen
= attr
->namelen
;
405 /* fsid is left zero */
408 static int fuse_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
410 struct super_block
*sb
= dentry
->d_sb
;
411 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
413 struct fuse_statfs_out outarg
;
416 if (!fuse_allow_current_process(fc
)) {
417 buf
->f_type
= FUSE_SUPER_MAGIC
;
421 memset(&outarg
, 0, sizeof(outarg
));
423 args
.in
.h
.opcode
= FUSE_STATFS
;
424 args
.in
.h
.nodeid
= get_node_id(dentry
->d_inode
);
425 args
.out
.numargs
= 1;
426 args
.out
.args
[0].size
= sizeof(outarg
);
427 args
.out
.args
[0].value
= &outarg
;
428 err
= fuse_simple_request(fc
, &args
);
430 convert_fuse_statfs(buf
, &outarg
.st
);
439 OPT_DEFAULT_PERMISSIONS
,
446 static const match_table_t tokens
= {
448 {OPT_ROOTMODE
, "rootmode=%o"},
449 {OPT_USER_ID
, "user_id=%u"},
450 {OPT_GROUP_ID
, "group_id=%u"},
451 {OPT_DEFAULT_PERMISSIONS
, "default_permissions"},
452 {OPT_ALLOW_OTHER
, "allow_other"},
453 {OPT_MAX_READ
, "max_read=%u"},
454 {OPT_BLKSIZE
, "blksize=%u"},
458 static int fuse_match_uint(substring_t
*s
, unsigned int *res
)
461 char *buf
= match_strdup(s
);
463 err
= kstrtouint(buf
, 10, res
);
469 static int parse_fuse_opt(char *opt
, struct fuse_mount_data
*d
, int is_bdev
)
472 memset(d
, 0, sizeof(struct fuse_mount_data
));
474 d
->blksize
= FUSE_DEFAULT_BLKSIZE
;
476 while ((p
= strsep(&opt
, ",")) != NULL
) {
480 substring_t args
[MAX_OPT_ARGS
];
484 token
= match_token(p
, tokens
, args
);
487 if (match_int(&args
[0], &value
))
494 if (match_octal(&args
[0], &value
))
496 if (!fuse_valid_type(value
))
499 d
->rootmode_present
= 1;
503 if (fuse_match_uint(&args
[0], &uv
))
505 d
->user_id
= make_kuid(current_user_ns(), uv
);
506 if (!uid_valid(d
->user_id
))
508 d
->user_id_present
= 1;
512 if (fuse_match_uint(&args
[0], &uv
))
514 d
->group_id
= make_kgid(current_user_ns(), uv
);
515 if (!gid_valid(d
->group_id
))
517 d
->group_id_present
= 1;
520 case OPT_DEFAULT_PERMISSIONS
:
521 d
->flags
|= FUSE_DEFAULT_PERMISSIONS
;
524 case OPT_ALLOW_OTHER
:
525 d
->flags
|= FUSE_ALLOW_OTHER
;
529 if (match_int(&args
[0], &value
))
535 if (!is_bdev
|| match_int(&args
[0], &value
))
545 if (!d
->fd_present
|| !d
->rootmode_present
||
546 !d
->user_id_present
|| !d
->group_id_present
)
552 static int fuse_show_options(struct seq_file
*m
, struct dentry
*root
)
554 struct super_block
*sb
= root
->d_sb
;
555 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
557 seq_printf(m
, ",user_id=%u", from_kuid_munged(&init_user_ns
, fc
->user_id
));
558 seq_printf(m
, ",group_id=%u", from_kgid_munged(&init_user_ns
, fc
->group_id
));
559 if (fc
->flags
& FUSE_DEFAULT_PERMISSIONS
)
560 seq_puts(m
, ",default_permissions");
561 if (fc
->flags
& FUSE_ALLOW_OTHER
)
562 seq_puts(m
, ",allow_other");
563 if (fc
->max_read
!= ~0)
564 seq_printf(m
, ",max_read=%u", fc
->max_read
);
565 if (sb
->s_bdev
&& sb
->s_blocksize
!= FUSE_DEFAULT_BLKSIZE
)
566 seq_printf(m
, ",blksize=%lu", sb
->s_blocksize
);
570 void fuse_conn_init(struct fuse_conn
*fc
)
572 memset(fc
, 0, sizeof(*fc
));
573 spin_lock_init(&fc
->lock
);
574 init_rwsem(&fc
->killsb
);
575 atomic_set(&fc
->count
, 1);
576 init_waitqueue_head(&fc
->waitq
);
577 init_waitqueue_head(&fc
->blocked_waitq
);
578 init_waitqueue_head(&fc
->reserved_req_waitq
);
579 INIT_LIST_HEAD(&fc
->pending
);
580 INIT_LIST_HEAD(&fc
->processing
);
581 INIT_LIST_HEAD(&fc
->io
);
582 INIT_LIST_HEAD(&fc
->interrupts
);
583 INIT_LIST_HEAD(&fc
->bg_queue
);
584 INIT_LIST_HEAD(&fc
->entry
);
585 fc
->forget_list_tail
= &fc
->forget_list_head
;
586 atomic_set(&fc
->num_waiting
, 0);
587 fc
->max_background
= FUSE_DEFAULT_MAX_BACKGROUND
;
588 fc
->congestion_threshold
= FUSE_DEFAULT_CONGESTION_THRESHOLD
;
590 fc
->polled_files
= RB_ROOT
;
594 fc
->attr_version
= 1;
595 get_random_bytes(&fc
->scramble_key
, sizeof(fc
->scramble_key
));
597 EXPORT_SYMBOL_GPL(fuse_conn_init
);
599 void fuse_conn_put(struct fuse_conn
*fc
)
601 if (atomic_dec_and_test(&fc
->count
)) {
603 fuse_request_free(fc
->destroy_req
);
607 EXPORT_SYMBOL_GPL(fuse_conn_put
);
609 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
)
611 atomic_inc(&fc
->count
);
614 EXPORT_SYMBOL_GPL(fuse_conn_get
);
616 static struct inode
*fuse_get_root_inode(struct super_block
*sb
, unsigned mode
)
618 struct fuse_attr attr
;
619 memset(&attr
, 0, sizeof(attr
));
622 attr
.ino
= FUSE_ROOT_ID
;
624 return fuse_iget(sb
, 1, 0, &attr
, 0, 0);
627 struct fuse_inode_handle
{
632 static struct dentry
*fuse_get_dentry(struct super_block
*sb
,
633 struct fuse_inode_handle
*handle
)
635 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
637 struct dentry
*entry
;
640 if (handle
->nodeid
== 0)
643 inode
= ilookup5(sb
, handle
->nodeid
, fuse_inode_eq
, &handle
->nodeid
);
645 struct fuse_entry_out outarg
;
648 if (!fc
->export_support
)
653 err
= fuse_lookup_name(sb
, handle
->nodeid
, &name
, &outarg
,
655 if (err
&& err
!= -ENOENT
)
662 if (get_node_id(inode
) != handle
->nodeid
)
666 if (inode
->i_generation
!= handle
->generation
)
669 entry
= d_obtain_alias(inode
);
670 if (!IS_ERR(entry
) && get_node_id(inode
) != FUSE_ROOT_ID
)
671 fuse_invalidate_entry_cache(entry
);
681 static int fuse_encode_fh(struct inode
*inode
, u32
*fh
, int *max_len
,
682 struct inode
*parent
)
684 int len
= parent
? 6 : 3;
688 if (*max_len
< len
) {
690 return FILEID_INVALID
;
693 nodeid
= get_fuse_inode(inode
)->nodeid
;
694 generation
= inode
->i_generation
;
696 fh
[0] = (u32
)(nodeid
>> 32);
697 fh
[1] = (u32
)(nodeid
& 0xffffffff);
701 nodeid
= get_fuse_inode(parent
)->nodeid
;
702 generation
= parent
->i_generation
;
704 fh
[3] = (u32
)(nodeid
>> 32);
705 fh
[4] = (u32
)(nodeid
& 0xffffffff);
710 return parent
? 0x82 : 0x81;
713 static struct dentry
*fuse_fh_to_dentry(struct super_block
*sb
,
714 struct fid
*fid
, int fh_len
, int fh_type
)
716 struct fuse_inode_handle handle
;
718 if ((fh_type
!= 0x81 && fh_type
!= 0x82) || fh_len
< 3)
721 handle
.nodeid
= (u64
) fid
->raw
[0] << 32;
722 handle
.nodeid
|= (u64
) fid
->raw
[1];
723 handle
.generation
= fid
->raw
[2];
724 return fuse_get_dentry(sb
, &handle
);
727 static struct dentry
*fuse_fh_to_parent(struct super_block
*sb
,
728 struct fid
*fid
, int fh_len
, int fh_type
)
730 struct fuse_inode_handle parent
;
732 if (fh_type
!= 0x82 || fh_len
< 6)
735 parent
.nodeid
= (u64
) fid
->raw
[3] << 32;
736 parent
.nodeid
|= (u64
) fid
->raw
[4];
737 parent
.generation
= fid
->raw
[5];
738 return fuse_get_dentry(sb
, &parent
);
741 static struct dentry
*fuse_get_parent(struct dentry
*child
)
743 struct inode
*child_inode
= child
->d_inode
;
744 struct fuse_conn
*fc
= get_fuse_conn(child_inode
);
746 struct dentry
*parent
;
747 struct fuse_entry_out outarg
;
751 if (!fc
->export_support
)
752 return ERR_PTR(-ESTALE
);
756 err
= fuse_lookup_name(child_inode
->i_sb
, get_node_id(child_inode
),
757 &name
, &outarg
, &inode
);
760 return ERR_PTR(-ESTALE
);
764 parent
= d_obtain_alias(inode
);
765 if (!IS_ERR(parent
) && get_node_id(inode
) != FUSE_ROOT_ID
)
766 fuse_invalidate_entry_cache(parent
);
771 static const struct export_operations fuse_export_operations
= {
772 .fh_to_dentry
= fuse_fh_to_dentry
,
773 .fh_to_parent
= fuse_fh_to_parent
,
774 .encode_fh
= fuse_encode_fh
,
775 .get_parent
= fuse_get_parent
,
778 static const struct super_operations fuse_super_operations
= {
779 .alloc_inode
= fuse_alloc_inode
,
780 .destroy_inode
= fuse_destroy_inode
,
781 .evict_inode
= fuse_evict_inode
,
782 .write_inode
= fuse_write_inode
,
783 .drop_inode
= generic_delete_inode
,
784 .remount_fs
= fuse_remount_fs
,
785 .put_super
= fuse_put_super
,
786 .umount_begin
= fuse_umount_begin
,
787 .statfs
= fuse_statfs
,
788 .show_options
= fuse_show_options
,
791 static void sanitize_global_limit(unsigned *limit
)
794 *limit
= ((totalram_pages
<< PAGE_SHIFT
) >> 13) /
795 sizeof(struct fuse_req
);
797 if (*limit
>= 1 << 16)
798 *limit
= (1 << 16) - 1;
801 static int set_global_limit(const char *val
, struct kernel_param
*kp
)
805 rv
= param_set_uint(val
, kp
);
809 sanitize_global_limit((unsigned *)kp
->arg
);
814 static void process_init_limits(struct fuse_conn
*fc
, struct fuse_init_out
*arg
)
816 int cap_sys_admin
= capable(CAP_SYS_ADMIN
);
821 sanitize_global_limit(&max_user_bgreq
);
822 sanitize_global_limit(&max_user_congthresh
);
824 if (arg
->max_background
) {
825 fc
->max_background
= arg
->max_background
;
827 if (!cap_sys_admin
&& fc
->max_background
> max_user_bgreq
)
828 fc
->max_background
= max_user_bgreq
;
830 if (arg
->congestion_threshold
) {
831 fc
->congestion_threshold
= arg
->congestion_threshold
;
833 if (!cap_sys_admin
&&
834 fc
->congestion_threshold
> max_user_congthresh
)
835 fc
->congestion_threshold
= max_user_congthresh
;
839 static void process_init_reply(struct fuse_conn
*fc
, struct fuse_req
*req
)
841 struct fuse_init_out
*arg
= &req
->misc
.init_out
;
843 if (req
->out
.h
.error
|| arg
->major
!= FUSE_KERNEL_VERSION
)
846 unsigned long ra_pages
;
848 process_init_limits(fc
, arg
);
850 if (arg
->minor
>= 6) {
851 ra_pages
= arg
->max_readahead
/ PAGE_CACHE_SIZE
;
852 if (arg
->flags
& FUSE_ASYNC_READ
)
854 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
856 if (arg
->minor
>= 17) {
857 if (!(arg
->flags
& FUSE_FLOCK_LOCKS
))
860 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
863 if (arg
->flags
& FUSE_ATOMIC_O_TRUNC
)
864 fc
->atomic_o_trunc
= 1;
865 if (arg
->minor
>= 9) {
866 /* LOOKUP has dependency on proto version */
867 if (arg
->flags
& FUSE_EXPORT_SUPPORT
)
868 fc
->export_support
= 1;
870 if (arg
->flags
& FUSE_BIG_WRITES
)
872 if (arg
->flags
& FUSE_DONT_MASK
)
874 if (arg
->flags
& FUSE_AUTO_INVAL_DATA
)
875 fc
->auto_inval_data
= 1;
876 if (arg
->flags
& FUSE_DO_READDIRPLUS
) {
877 fc
->do_readdirplus
= 1;
878 if (arg
->flags
& FUSE_READDIRPLUS_AUTO
)
879 fc
->readdirplus_auto
= 1;
881 if (arg
->flags
& FUSE_ASYNC_DIO
)
883 if (arg
->flags
& FUSE_WRITEBACK_CACHE
)
884 fc
->writeback_cache
= 1;
885 if (arg
->time_gran
&& arg
->time_gran
<= 1000000000)
886 fc
->sb
->s_time_gran
= arg
->time_gran
;
888 ra_pages
= fc
->max_read
/ PAGE_CACHE_SIZE
;
893 fc
->bdi
.ra_pages
= min(fc
->bdi
.ra_pages
, ra_pages
);
894 fc
->minor
= arg
->minor
;
895 fc
->max_write
= arg
->minor
< 5 ? 4096 : arg
->max_write
;
896 fc
->max_write
= max_t(unsigned, 4096, fc
->max_write
);
899 fuse_set_initialized(fc
);
900 wake_up_all(&fc
->blocked_waitq
);
903 static void fuse_send_init(struct fuse_conn
*fc
, struct fuse_req
*req
)
905 struct fuse_init_in
*arg
= &req
->misc
.init_in
;
907 arg
->major
= FUSE_KERNEL_VERSION
;
908 arg
->minor
= FUSE_KERNEL_MINOR_VERSION
;
909 arg
->max_readahead
= fc
->bdi
.ra_pages
* PAGE_CACHE_SIZE
;
910 arg
->flags
|= FUSE_ASYNC_READ
| FUSE_POSIX_LOCKS
| FUSE_ATOMIC_O_TRUNC
|
911 FUSE_EXPORT_SUPPORT
| FUSE_BIG_WRITES
| FUSE_DONT_MASK
|
912 FUSE_SPLICE_WRITE
| FUSE_SPLICE_MOVE
| FUSE_SPLICE_READ
|
913 FUSE_FLOCK_LOCKS
| FUSE_IOCTL_DIR
| FUSE_AUTO_INVAL_DATA
|
914 FUSE_DO_READDIRPLUS
| FUSE_READDIRPLUS_AUTO
| FUSE_ASYNC_DIO
|
915 FUSE_WRITEBACK_CACHE
| FUSE_NO_OPEN_SUPPORT
;
916 req
->in
.h
.opcode
= FUSE_INIT
;
918 req
->in
.args
[0].size
= sizeof(*arg
);
919 req
->in
.args
[0].value
= arg
;
920 req
->out
.numargs
= 1;
921 /* Variable length argument used for backward compatibility
922 with interface version < 7.5. Rest of init_out is zeroed
923 by do_get_request(), so a short reply is not a problem */
925 req
->out
.args
[0].size
= sizeof(struct fuse_init_out
);
926 req
->out
.args
[0].value
= &req
->misc
.init_out
;
927 req
->end
= process_init_reply
;
928 fuse_request_send_background(fc
, req
);
931 static void fuse_free_conn(struct fuse_conn
*fc
)
936 static int fuse_bdi_init(struct fuse_conn
*fc
, struct super_block
*sb
)
940 fc
->bdi
.name
= "fuse";
941 fc
->bdi
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
942 /* fuse does it's own writeback accounting */
943 fc
->bdi
.capabilities
= BDI_CAP_NO_ACCT_WB
| BDI_CAP_STRICTLIMIT
;
945 err
= bdi_init(&fc
->bdi
);
949 fc
->bdi_initialized
= 1;
952 err
= bdi_register(&fc
->bdi
, NULL
, "%u:%u-fuseblk",
953 MAJOR(fc
->dev
), MINOR(fc
->dev
));
955 err
= bdi_register_dev(&fc
->bdi
, fc
->dev
);
962 * For a single fuse filesystem use max 1% of dirty +
963 * writeback threshold.
965 * This gives about 1M of write buffer for memory maps on a
966 * machine with 1G and 10% dirty_ratio, which should be more
969 * Privileged users can raise it by writing to
971 * /sys/class/bdi/<bdi>/max_ratio
973 bdi_set_max_ratio(&fc
->bdi
, 1);
978 static int fuse_fill_super(struct super_block
*sb
, void *data
, int silent
)
980 struct fuse_conn
*fc
;
982 struct fuse_mount_data d
;
984 struct dentry
*root_dentry
;
985 struct fuse_req
*init_req
;
987 int is_bdev
= sb
->s_bdev
!= NULL
;
990 if (sb
->s_flags
& MS_MANDLOCK
)
993 sb
->s_flags
&= ~(MS_NOSEC
| MS_I_VERSION
);
995 if (!parse_fuse_opt(data
, &d
, is_bdev
))
1001 if (!sb_set_blocksize(sb
, d
.blksize
))
1005 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1006 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1008 sb
->s_magic
= FUSE_SUPER_MAGIC
;
1009 sb
->s_op
= &fuse_super_operations
;
1010 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1011 sb
->s_time_gran
= 1;
1012 sb
->s_export_op
= &fuse_export_operations
;
1019 if ((file
->f_op
!= &fuse_dev_operations
) ||
1020 (file
->f_cred
->user_ns
!= &init_user_ns
))
1023 fc
= kmalloc(sizeof(*fc
), GFP_KERNEL
);
1030 fc
->dev
= sb
->s_dev
;
1032 err
= fuse_bdi_init(fc
, sb
);
1036 sb
->s_bdi
= &fc
->bdi
;
1038 /* Handle umasking inside the fuse code */
1039 if (sb
->s_flags
& MS_POSIXACL
)
1041 sb
->s_flags
|= MS_POSIXACL
;
1043 fc
->release
= fuse_free_conn
;
1044 fc
->flags
= d
.flags
;
1045 fc
->user_id
= d
.user_id
;
1046 fc
->group_id
= d
.group_id
;
1047 fc
->max_read
= max_t(unsigned, 4096, d
.max_read
);
1049 /* Used by get_root_inode() */
1053 root
= fuse_get_root_inode(sb
, d
.rootmode
);
1054 root_dentry
= d_make_root(root
);
1057 /* only now - we want root dentry with NULL ->d_op */
1058 sb
->s_d_op
= &fuse_dentry_operations
;
1060 init_req
= fuse_request_alloc(0);
1063 init_req
->background
= 1;
1066 fc
->destroy_req
= fuse_request_alloc(0);
1067 if (!fc
->destroy_req
)
1068 goto err_free_init_req
;
1071 mutex_lock(&fuse_mutex
);
1073 if (file
->private_data
)
1076 err
= fuse_ctl_add_conn(fc
);
1080 list_add_tail(&fc
->entry
, &fuse_conn_list
);
1081 sb
->s_root
= root_dentry
;
1083 file
->private_data
= fuse_conn_get(fc
);
1084 mutex_unlock(&fuse_mutex
);
1086 * atomic_dec_and_test() in fput() provides the necessary
1087 * memory barrier for file->private_data to be visible on all
1092 fuse_send_init(fc
, init_req
);
1097 mutex_unlock(&fuse_mutex
);
1099 fuse_request_free(init_req
);
1103 fuse_bdi_destroy(fc
);
1111 static struct dentry
*fuse_mount(struct file_system_type
*fs_type
,
1112 int flags
, const char *dev_name
,
1115 return mount_nodev(fs_type
, flags
, raw_data
, fuse_fill_super
);
1118 static void fuse_kill_sb_anon(struct super_block
*sb
)
1120 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1123 down_write(&fc
->killsb
);
1125 up_write(&fc
->killsb
);
1128 kill_anon_super(sb
);
1131 static struct file_system_type fuse_fs_type
= {
1132 .owner
= THIS_MODULE
,
1134 .fs_flags
= FS_HAS_SUBTYPE
,
1135 .mount
= fuse_mount
,
1136 .kill_sb
= fuse_kill_sb_anon
,
1138 MODULE_ALIAS_FS("fuse");
1141 static struct dentry
*fuse_mount_blk(struct file_system_type
*fs_type
,
1142 int flags
, const char *dev_name
,
1145 return mount_bdev(fs_type
, flags
, dev_name
, raw_data
, fuse_fill_super
);
1148 static void fuse_kill_sb_blk(struct super_block
*sb
)
1150 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1153 down_write(&fc
->killsb
);
1155 up_write(&fc
->killsb
);
1158 kill_block_super(sb
);
1161 static struct file_system_type fuseblk_fs_type
= {
1162 .owner
= THIS_MODULE
,
1164 .mount
= fuse_mount_blk
,
1165 .kill_sb
= fuse_kill_sb_blk
,
1166 .fs_flags
= FS_REQUIRES_DEV
| FS_HAS_SUBTYPE
,
1168 MODULE_ALIAS_FS("fuseblk");
1170 static inline int register_fuseblk(void)
1172 return register_filesystem(&fuseblk_fs_type
);
1175 static inline void unregister_fuseblk(void)
1177 unregister_filesystem(&fuseblk_fs_type
);
1180 static inline int register_fuseblk(void)
1185 static inline void unregister_fuseblk(void)
1190 static void fuse_inode_init_once(void *foo
)
1192 struct inode
*inode
= foo
;
1194 inode_init_once(inode
);
1197 static int __init
fuse_fs_init(void)
1201 fuse_inode_cachep
= kmem_cache_create("fuse_inode",
1202 sizeof(struct fuse_inode
),
1203 0, SLAB_HWCACHE_ALIGN
,
1204 fuse_inode_init_once
);
1206 if (!fuse_inode_cachep
)
1209 err
= register_fuseblk();
1213 err
= register_filesystem(&fuse_fs_type
);
1220 unregister_fuseblk();
1222 kmem_cache_destroy(fuse_inode_cachep
);
1227 static void fuse_fs_cleanup(void)
1229 unregister_filesystem(&fuse_fs_type
);
1230 unregister_fuseblk();
1233 * Make sure all delayed rcu free inodes are flushed before we
1237 kmem_cache_destroy(fuse_inode_cachep
);
1240 static struct kobject
*fuse_kobj
;
1241 static struct kobject
*connections_kobj
;
1243 static int fuse_sysfs_init(void)
1247 fuse_kobj
= kobject_create_and_add("fuse", fs_kobj
);
1253 connections_kobj
= kobject_create_and_add("connections", fuse_kobj
);
1254 if (!connections_kobj
) {
1256 goto out_fuse_unregister
;
1261 out_fuse_unregister
:
1262 kobject_put(fuse_kobj
);
1267 static void fuse_sysfs_cleanup(void)
1269 kobject_put(connections_kobj
);
1270 kobject_put(fuse_kobj
);
1273 static int __init
fuse_init(void)
1277 printk(KERN_INFO
"fuse init (API version %i.%i)\n",
1278 FUSE_KERNEL_VERSION
, FUSE_KERNEL_MINOR_VERSION
);
1280 INIT_LIST_HEAD(&fuse_conn_list
);
1281 res
= fuse_fs_init();
1285 res
= fuse_dev_init();
1287 goto err_fs_cleanup
;
1289 res
= fuse_sysfs_init();
1291 goto err_dev_cleanup
;
1293 res
= fuse_ctl_init();
1295 goto err_sysfs_cleanup
;
1297 sanitize_global_limit(&max_user_bgreq
);
1298 sanitize_global_limit(&max_user_congthresh
);
1303 fuse_sysfs_cleanup();
1312 static void __exit
fuse_exit(void)
1314 printk(KERN_DEBUG
"fuse exit\n");
1317 fuse_sysfs_cleanup();
1322 module_init(fuse_init
);
1323 module_exit(fuse_exit
);