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 inode
->i_data
.backing_dev_info
= &fc
->bdi
;
312 fuse_init_inode(inode
, attr
);
313 unlock_new_inode(inode
);
314 } else if ((inode
->i_mode
^ attr
->mode
) & S_IFMT
) {
315 /* Inode has changed type, any I/O on the old should fail */
316 make_bad_inode(inode
);
321 fi
= get_fuse_inode(inode
);
322 spin_lock(&fc
->lock
);
324 spin_unlock(&fc
->lock
);
325 fuse_change_attributes(inode
, attr
, attr_valid
, attr_version
);
330 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
331 loff_t offset
, loff_t len
)
337 inode
= ilookup5(sb
, nodeid
, fuse_inode_eq
, &nodeid
);
341 fuse_invalidate_attr(inode
);
343 pg_start
= offset
>> PAGE_CACHE_SHIFT
;
347 pg_end
= (offset
+ len
- 1) >> PAGE_CACHE_SHIFT
;
348 invalidate_inode_pages2_range(inode
->i_mapping
,
355 static void fuse_umount_begin(struct super_block
*sb
)
357 fuse_abort_conn(get_fuse_conn_super(sb
));
360 static void fuse_send_destroy(struct fuse_conn
*fc
)
362 struct fuse_req
*req
= fc
->destroy_req
;
363 if (req
&& fc
->conn_init
) {
364 fc
->destroy_req
= NULL
;
365 req
->in
.h
.opcode
= FUSE_DESTROY
;
368 fuse_request_send(fc
, req
);
369 fuse_put_request(fc
, req
);
373 static void fuse_bdi_destroy(struct fuse_conn
*fc
)
375 if (fc
->bdi_initialized
)
376 bdi_destroy(&fc
->bdi
);
379 static void fuse_put_super(struct super_block
*sb
)
381 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
383 fuse_send_destroy(fc
);
386 mutex_lock(&fuse_mutex
);
387 list_del(&fc
->entry
);
388 fuse_ctl_remove_conn(fc
);
389 mutex_unlock(&fuse_mutex
);
390 fuse_bdi_destroy(fc
);
395 static void convert_fuse_statfs(struct kstatfs
*stbuf
, struct fuse_kstatfs
*attr
)
397 stbuf
->f_type
= FUSE_SUPER_MAGIC
;
398 stbuf
->f_bsize
= attr
->bsize
;
399 stbuf
->f_frsize
= attr
->frsize
;
400 stbuf
->f_blocks
= attr
->blocks
;
401 stbuf
->f_bfree
= attr
->bfree
;
402 stbuf
->f_bavail
= attr
->bavail
;
403 stbuf
->f_files
= attr
->files
;
404 stbuf
->f_ffree
= attr
->ffree
;
405 stbuf
->f_namelen
= attr
->namelen
;
406 /* fsid is left zero */
409 static int fuse_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
411 struct super_block
*sb
= dentry
->d_sb
;
412 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
414 struct fuse_statfs_out outarg
;
417 if (!fuse_allow_current_process(fc
)) {
418 buf
->f_type
= FUSE_SUPER_MAGIC
;
422 memset(&outarg
, 0, sizeof(outarg
));
424 args
.in
.h
.opcode
= FUSE_STATFS
;
425 args
.in
.h
.nodeid
= get_node_id(dentry
->d_inode
);
426 args
.out
.numargs
= 1;
427 args
.out
.args
[0].size
= sizeof(outarg
);
428 args
.out
.args
[0].value
= &outarg
;
429 err
= fuse_simple_request(fc
, &args
);
431 convert_fuse_statfs(buf
, &outarg
.st
);
440 OPT_DEFAULT_PERMISSIONS
,
447 static const match_table_t tokens
= {
449 {OPT_ROOTMODE
, "rootmode=%o"},
450 {OPT_USER_ID
, "user_id=%u"},
451 {OPT_GROUP_ID
, "group_id=%u"},
452 {OPT_DEFAULT_PERMISSIONS
, "default_permissions"},
453 {OPT_ALLOW_OTHER
, "allow_other"},
454 {OPT_MAX_READ
, "max_read=%u"},
455 {OPT_BLKSIZE
, "blksize=%u"},
459 static int fuse_match_uint(substring_t
*s
, unsigned int *res
)
462 char *buf
= match_strdup(s
);
464 err
= kstrtouint(buf
, 10, res
);
470 static int parse_fuse_opt(char *opt
, struct fuse_mount_data
*d
, int is_bdev
)
473 memset(d
, 0, sizeof(struct fuse_mount_data
));
475 d
->blksize
= FUSE_DEFAULT_BLKSIZE
;
477 while ((p
= strsep(&opt
, ",")) != NULL
) {
481 substring_t args
[MAX_OPT_ARGS
];
485 token
= match_token(p
, tokens
, args
);
488 if (match_int(&args
[0], &value
))
495 if (match_octal(&args
[0], &value
))
497 if (!fuse_valid_type(value
))
500 d
->rootmode_present
= 1;
504 if (fuse_match_uint(&args
[0], &uv
))
506 d
->user_id
= make_kuid(current_user_ns(), uv
);
507 if (!uid_valid(d
->user_id
))
509 d
->user_id_present
= 1;
513 if (fuse_match_uint(&args
[0], &uv
))
515 d
->group_id
= make_kgid(current_user_ns(), uv
);
516 if (!gid_valid(d
->group_id
))
518 d
->group_id_present
= 1;
521 case OPT_DEFAULT_PERMISSIONS
:
522 d
->flags
|= FUSE_DEFAULT_PERMISSIONS
;
525 case OPT_ALLOW_OTHER
:
526 d
->flags
|= FUSE_ALLOW_OTHER
;
530 if (match_int(&args
[0], &value
))
536 if (!is_bdev
|| match_int(&args
[0], &value
))
546 if (!d
->fd_present
|| !d
->rootmode_present
||
547 !d
->user_id_present
|| !d
->group_id_present
)
553 static int fuse_show_options(struct seq_file
*m
, struct dentry
*root
)
555 struct super_block
*sb
= root
->d_sb
;
556 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
558 seq_printf(m
, ",user_id=%u", from_kuid_munged(&init_user_ns
, fc
->user_id
));
559 seq_printf(m
, ",group_id=%u", from_kgid_munged(&init_user_ns
, fc
->group_id
));
560 if (fc
->flags
& FUSE_DEFAULT_PERMISSIONS
)
561 seq_puts(m
, ",default_permissions");
562 if (fc
->flags
& FUSE_ALLOW_OTHER
)
563 seq_puts(m
, ",allow_other");
564 if (fc
->max_read
!= ~0)
565 seq_printf(m
, ",max_read=%u", fc
->max_read
);
566 if (sb
->s_bdev
&& sb
->s_blocksize
!= FUSE_DEFAULT_BLKSIZE
)
567 seq_printf(m
, ",blksize=%lu", sb
->s_blocksize
);
571 void fuse_conn_init(struct fuse_conn
*fc
)
573 memset(fc
, 0, sizeof(*fc
));
574 spin_lock_init(&fc
->lock
);
575 init_rwsem(&fc
->killsb
);
576 atomic_set(&fc
->count
, 1);
577 init_waitqueue_head(&fc
->waitq
);
578 init_waitqueue_head(&fc
->blocked_waitq
);
579 init_waitqueue_head(&fc
->reserved_req_waitq
);
580 INIT_LIST_HEAD(&fc
->pending
);
581 INIT_LIST_HEAD(&fc
->processing
);
582 INIT_LIST_HEAD(&fc
->io
);
583 INIT_LIST_HEAD(&fc
->interrupts
);
584 INIT_LIST_HEAD(&fc
->bg_queue
);
585 INIT_LIST_HEAD(&fc
->entry
);
586 fc
->forget_list_tail
= &fc
->forget_list_head
;
587 atomic_set(&fc
->num_waiting
, 0);
588 fc
->max_background
= FUSE_DEFAULT_MAX_BACKGROUND
;
589 fc
->congestion_threshold
= FUSE_DEFAULT_CONGESTION_THRESHOLD
;
591 fc
->polled_files
= RB_ROOT
;
595 fc
->attr_version
= 1;
596 get_random_bytes(&fc
->scramble_key
, sizeof(fc
->scramble_key
));
598 EXPORT_SYMBOL_GPL(fuse_conn_init
);
600 void fuse_conn_put(struct fuse_conn
*fc
)
602 if (atomic_dec_and_test(&fc
->count
)) {
604 fuse_request_free(fc
->destroy_req
);
608 EXPORT_SYMBOL_GPL(fuse_conn_put
);
610 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
)
612 atomic_inc(&fc
->count
);
615 EXPORT_SYMBOL_GPL(fuse_conn_get
);
617 static struct inode
*fuse_get_root_inode(struct super_block
*sb
, unsigned mode
)
619 struct fuse_attr attr
;
620 memset(&attr
, 0, sizeof(attr
));
623 attr
.ino
= FUSE_ROOT_ID
;
625 return fuse_iget(sb
, 1, 0, &attr
, 0, 0);
628 struct fuse_inode_handle
{
633 static struct dentry
*fuse_get_dentry(struct super_block
*sb
,
634 struct fuse_inode_handle
*handle
)
636 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
638 struct dentry
*entry
;
641 if (handle
->nodeid
== 0)
644 inode
= ilookup5(sb
, handle
->nodeid
, fuse_inode_eq
, &handle
->nodeid
);
646 struct fuse_entry_out outarg
;
649 if (!fc
->export_support
)
654 err
= fuse_lookup_name(sb
, handle
->nodeid
, &name
, &outarg
,
656 if (err
&& err
!= -ENOENT
)
663 if (get_node_id(inode
) != handle
->nodeid
)
667 if (inode
->i_generation
!= handle
->generation
)
670 entry
= d_obtain_alias(inode
);
671 if (!IS_ERR(entry
) && get_node_id(inode
) != FUSE_ROOT_ID
)
672 fuse_invalidate_entry_cache(entry
);
682 static int fuse_encode_fh(struct inode
*inode
, u32
*fh
, int *max_len
,
683 struct inode
*parent
)
685 int len
= parent
? 6 : 3;
689 if (*max_len
< len
) {
691 return FILEID_INVALID
;
694 nodeid
= get_fuse_inode(inode
)->nodeid
;
695 generation
= inode
->i_generation
;
697 fh
[0] = (u32
)(nodeid
>> 32);
698 fh
[1] = (u32
)(nodeid
& 0xffffffff);
702 nodeid
= get_fuse_inode(parent
)->nodeid
;
703 generation
= parent
->i_generation
;
705 fh
[3] = (u32
)(nodeid
>> 32);
706 fh
[4] = (u32
)(nodeid
& 0xffffffff);
711 return parent
? 0x82 : 0x81;
714 static struct dentry
*fuse_fh_to_dentry(struct super_block
*sb
,
715 struct fid
*fid
, int fh_len
, int fh_type
)
717 struct fuse_inode_handle handle
;
719 if ((fh_type
!= 0x81 && fh_type
!= 0x82) || fh_len
< 3)
722 handle
.nodeid
= (u64
) fid
->raw
[0] << 32;
723 handle
.nodeid
|= (u64
) fid
->raw
[1];
724 handle
.generation
= fid
->raw
[2];
725 return fuse_get_dentry(sb
, &handle
);
728 static struct dentry
*fuse_fh_to_parent(struct super_block
*sb
,
729 struct fid
*fid
, int fh_len
, int fh_type
)
731 struct fuse_inode_handle parent
;
733 if (fh_type
!= 0x82 || fh_len
< 6)
736 parent
.nodeid
= (u64
) fid
->raw
[3] << 32;
737 parent
.nodeid
|= (u64
) fid
->raw
[4];
738 parent
.generation
= fid
->raw
[5];
739 return fuse_get_dentry(sb
, &parent
);
742 static struct dentry
*fuse_get_parent(struct dentry
*child
)
744 struct inode
*child_inode
= child
->d_inode
;
745 struct fuse_conn
*fc
= get_fuse_conn(child_inode
);
747 struct dentry
*parent
;
748 struct fuse_entry_out outarg
;
752 if (!fc
->export_support
)
753 return ERR_PTR(-ESTALE
);
757 err
= fuse_lookup_name(child_inode
->i_sb
, get_node_id(child_inode
),
758 &name
, &outarg
, &inode
);
761 return ERR_PTR(-ESTALE
);
765 parent
= d_obtain_alias(inode
);
766 if (!IS_ERR(parent
) && get_node_id(inode
) != FUSE_ROOT_ID
)
767 fuse_invalidate_entry_cache(parent
);
772 static const struct export_operations fuse_export_operations
= {
773 .fh_to_dentry
= fuse_fh_to_dentry
,
774 .fh_to_parent
= fuse_fh_to_parent
,
775 .encode_fh
= fuse_encode_fh
,
776 .get_parent
= fuse_get_parent
,
779 static const struct super_operations fuse_super_operations
= {
780 .alloc_inode
= fuse_alloc_inode
,
781 .destroy_inode
= fuse_destroy_inode
,
782 .evict_inode
= fuse_evict_inode
,
783 .write_inode
= fuse_write_inode
,
784 .drop_inode
= generic_delete_inode
,
785 .remount_fs
= fuse_remount_fs
,
786 .put_super
= fuse_put_super
,
787 .umount_begin
= fuse_umount_begin
,
788 .statfs
= fuse_statfs
,
789 .show_options
= fuse_show_options
,
792 static void sanitize_global_limit(unsigned *limit
)
795 *limit
= ((totalram_pages
<< PAGE_SHIFT
) >> 13) /
796 sizeof(struct fuse_req
);
798 if (*limit
>= 1 << 16)
799 *limit
= (1 << 16) - 1;
802 static int set_global_limit(const char *val
, struct kernel_param
*kp
)
806 rv
= param_set_uint(val
, kp
);
810 sanitize_global_limit((unsigned *)kp
->arg
);
815 static void process_init_limits(struct fuse_conn
*fc
, struct fuse_init_out
*arg
)
817 int cap_sys_admin
= capable(CAP_SYS_ADMIN
);
822 sanitize_global_limit(&max_user_bgreq
);
823 sanitize_global_limit(&max_user_congthresh
);
825 if (arg
->max_background
) {
826 fc
->max_background
= arg
->max_background
;
828 if (!cap_sys_admin
&& fc
->max_background
> max_user_bgreq
)
829 fc
->max_background
= max_user_bgreq
;
831 if (arg
->congestion_threshold
) {
832 fc
->congestion_threshold
= arg
->congestion_threshold
;
834 if (!cap_sys_admin
&&
835 fc
->congestion_threshold
> max_user_congthresh
)
836 fc
->congestion_threshold
= max_user_congthresh
;
840 static void process_init_reply(struct fuse_conn
*fc
, struct fuse_req
*req
)
842 struct fuse_init_out
*arg
= &req
->misc
.init_out
;
844 if (req
->out
.h
.error
|| arg
->major
!= FUSE_KERNEL_VERSION
)
847 unsigned long ra_pages
;
849 process_init_limits(fc
, arg
);
851 if (arg
->minor
>= 6) {
852 ra_pages
= arg
->max_readahead
/ PAGE_CACHE_SIZE
;
853 if (arg
->flags
& FUSE_ASYNC_READ
)
855 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
857 if (arg
->minor
>= 17) {
858 if (!(arg
->flags
& FUSE_FLOCK_LOCKS
))
861 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
864 if (arg
->flags
& FUSE_ATOMIC_O_TRUNC
)
865 fc
->atomic_o_trunc
= 1;
866 if (arg
->minor
>= 9) {
867 /* LOOKUP has dependency on proto version */
868 if (arg
->flags
& FUSE_EXPORT_SUPPORT
)
869 fc
->export_support
= 1;
871 if (arg
->flags
& FUSE_BIG_WRITES
)
873 if (arg
->flags
& FUSE_DONT_MASK
)
875 if (arg
->flags
& FUSE_AUTO_INVAL_DATA
)
876 fc
->auto_inval_data
= 1;
877 if (arg
->flags
& FUSE_DO_READDIRPLUS
) {
878 fc
->do_readdirplus
= 1;
879 if (arg
->flags
& FUSE_READDIRPLUS_AUTO
)
880 fc
->readdirplus_auto
= 1;
882 if (arg
->flags
& FUSE_ASYNC_DIO
)
884 if (arg
->flags
& FUSE_WRITEBACK_CACHE
)
885 fc
->writeback_cache
= 1;
886 if (arg
->time_gran
&& arg
->time_gran
<= 1000000000)
887 fc
->sb
->s_time_gran
= arg
->time_gran
;
889 ra_pages
= fc
->max_read
/ PAGE_CACHE_SIZE
;
894 fc
->bdi
.ra_pages
= min(fc
->bdi
.ra_pages
, ra_pages
);
895 fc
->minor
= arg
->minor
;
896 fc
->max_write
= arg
->minor
< 5 ? 4096 : arg
->max_write
;
897 fc
->max_write
= max_t(unsigned, 4096, fc
->max_write
);
900 fuse_set_initialized(fc
);
901 wake_up_all(&fc
->blocked_waitq
);
904 static void fuse_send_init(struct fuse_conn
*fc
, struct fuse_req
*req
)
906 struct fuse_init_in
*arg
= &req
->misc
.init_in
;
908 arg
->major
= FUSE_KERNEL_VERSION
;
909 arg
->minor
= FUSE_KERNEL_MINOR_VERSION
;
910 arg
->max_readahead
= fc
->bdi
.ra_pages
* PAGE_CACHE_SIZE
;
911 arg
->flags
|= FUSE_ASYNC_READ
| FUSE_POSIX_LOCKS
| FUSE_ATOMIC_O_TRUNC
|
912 FUSE_EXPORT_SUPPORT
| FUSE_BIG_WRITES
| FUSE_DONT_MASK
|
913 FUSE_SPLICE_WRITE
| FUSE_SPLICE_MOVE
| FUSE_SPLICE_READ
|
914 FUSE_FLOCK_LOCKS
| FUSE_IOCTL_DIR
| FUSE_AUTO_INVAL_DATA
|
915 FUSE_DO_READDIRPLUS
| FUSE_READDIRPLUS_AUTO
| FUSE_ASYNC_DIO
|
916 FUSE_WRITEBACK_CACHE
| FUSE_NO_OPEN_SUPPORT
;
917 req
->in
.h
.opcode
= FUSE_INIT
;
919 req
->in
.args
[0].size
= sizeof(*arg
);
920 req
->in
.args
[0].value
= arg
;
921 req
->out
.numargs
= 1;
922 /* Variable length argument used for backward compatibility
923 with interface version < 7.5. Rest of init_out is zeroed
924 by do_get_request(), so a short reply is not a problem */
926 req
->out
.args
[0].size
= sizeof(struct fuse_init_out
);
927 req
->out
.args
[0].value
= &req
->misc
.init_out
;
928 req
->end
= process_init_reply
;
929 fuse_request_send_background(fc
, req
);
932 static void fuse_free_conn(struct fuse_conn
*fc
)
937 static int fuse_bdi_init(struct fuse_conn
*fc
, struct super_block
*sb
)
941 fc
->bdi
.name
= "fuse";
942 fc
->bdi
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
943 /* fuse does it's own writeback accounting */
944 fc
->bdi
.capabilities
= BDI_CAP_NO_ACCT_WB
| BDI_CAP_STRICTLIMIT
;
946 err
= bdi_init(&fc
->bdi
);
950 fc
->bdi_initialized
= 1;
953 err
= bdi_register(&fc
->bdi
, NULL
, "%u:%u-fuseblk",
954 MAJOR(fc
->dev
), MINOR(fc
->dev
));
956 err
= bdi_register_dev(&fc
->bdi
, fc
->dev
);
963 * For a single fuse filesystem use max 1% of dirty +
964 * writeback threshold.
966 * This gives about 1M of write buffer for memory maps on a
967 * machine with 1G and 10% dirty_ratio, which should be more
970 * Privileged users can raise it by writing to
972 * /sys/class/bdi/<bdi>/max_ratio
974 bdi_set_max_ratio(&fc
->bdi
, 1);
979 static int fuse_fill_super(struct super_block
*sb
, void *data
, int silent
)
981 struct fuse_conn
*fc
;
983 struct fuse_mount_data d
;
985 struct dentry
*root_dentry
;
986 struct fuse_req
*init_req
;
988 int is_bdev
= sb
->s_bdev
!= NULL
;
991 if (sb
->s_flags
& MS_MANDLOCK
)
994 sb
->s_flags
&= ~(MS_NOSEC
| MS_I_VERSION
);
996 if (!parse_fuse_opt(data
, &d
, is_bdev
))
1002 if (!sb_set_blocksize(sb
, d
.blksize
))
1006 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1007 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1009 sb
->s_magic
= FUSE_SUPER_MAGIC
;
1010 sb
->s_op
= &fuse_super_operations
;
1011 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1012 sb
->s_time_gran
= 1;
1013 sb
->s_export_op
= &fuse_export_operations
;
1020 if ((file
->f_op
!= &fuse_dev_operations
) ||
1021 (file
->f_cred
->user_ns
!= &init_user_ns
))
1024 fc
= kmalloc(sizeof(*fc
), GFP_KERNEL
);
1031 fc
->dev
= sb
->s_dev
;
1033 err
= fuse_bdi_init(fc
, sb
);
1037 sb
->s_bdi
= &fc
->bdi
;
1039 /* Handle umasking inside the fuse code */
1040 if (sb
->s_flags
& MS_POSIXACL
)
1042 sb
->s_flags
|= MS_POSIXACL
;
1044 fc
->release
= fuse_free_conn
;
1045 fc
->flags
= d
.flags
;
1046 fc
->user_id
= d
.user_id
;
1047 fc
->group_id
= d
.group_id
;
1048 fc
->max_read
= max_t(unsigned, 4096, d
.max_read
);
1050 /* Used by get_root_inode() */
1054 root
= fuse_get_root_inode(sb
, d
.rootmode
);
1055 root_dentry
= d_make_root(root
);
1058 /* only now - we want root dentry with NULL ->d_op */
1059 sb
->s_d_op
= &fuse_dentry_operations
;
1061 init_req
= fuse_request_alloc(0);
1064 init_req
->background
= 1;
1067 fc
->destroy_req
= fuse_request_alloc(0);
1068 if (!fc
->destroy_req
)
1069 goto err_free_init_req
;
1072 mutex_lock(&fuse_mutex
);
1074 if (file
->private_data
)
1077 err
= fuse_ctl_add_conn(fc
);
1081 list_add_tail(&fc
->entry
, &fuse_conn_list
);
1082 sb
->s_root
= root_dentry
;
1084 file
->private_data
= fuse_conn_get(fc
);
1085 mutex_unlock(&fuse_mutex
);
1087 * atomic_dec_and_test() in fput() provides the necessary
1088 * memory barrier for file->private_data to be visible on all
1093 fuse_send_init(fc
, init_req
);
1098 mutex_unlock(&fuse_mutex
);
1100 fuse_request_free(init_req
);
1104 fuse_bdi_destroy(fc
);
1112 static struct dentry
*fuse_mount(struct file_system_type
*fs_type
,
1113 int flags
, const char *dev_name
,
1116 return mount_nodev(fs_type
, flags
, raw_data
, fuse_fill_super
);
1119 static void fuse_kill_sb_anon(struct super_block
*sb
)
1121 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1124 down_write(&fc
->killsb
);
1126 up_write(&fc
->killsb
);
1129 kill_anon_super(sb
);
1132 static struct file_system_type fuse_fs_type
= {
1133 .owner
= THIS_MODULE
,
1135 .fs_flags
= FS_HAS_SUBTYPE
,
1136 .mount
= fuse_mount
,
1137 .kill_sb
= fuse_kill_sb_anon
,
1139 MODULE_ALIAS_FS("fuse");
1142 static struct dentry
*fuse_mount_blk(struct file_system_type
*fs_type
,
1143 int flags
, const char *dev_name
,
1146 return mount_bdev(fs_type
, flags
, dev_name
, raw_data
, fuse_fill_super
);
1149 static void fuse_kill_sb_blk(struct super_block
*sb
)
1151 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1154 down_write(&fc
->killsb
);
1156 up_write(&fc
->killsb
);
1159 kill_block_super(sb
);
1162 static struct file_system_type fuseblk_fs_type
= {
1163 .owner
= THIS_MODULE
,
1165 .mount
= fuse_mount_blk
,
1166 .kill_sb
= fuse_kill_sb_blk
,
1167 .fs_flags
= FS_REQUIRES_DEV
| FS_HAS_SUBTYPE
,
1169 MODULE_ALIAS_FS("fuseblk");
1171 static inline int register_fuseblk(void)
1173 return register_filesystem(&fuseblk_fs_type
);
1176 static inline void unregister_fuseblk(void)
1178 unregister_filesystem(&fuseblk_fs_type
);
1181 static inline int register_fuseblk(void)
1186 static inline void unregister_fuseblk(void)
1191 static void fuse_inode_init_once(void *foo
)
1193 struct inode
*inode
= foo
;
1195 inode_init_once(inode
);
1198 static int __init
fuse_fs_init(void)
1202 fuse_inode_cachep
= kmem_cache_create("fuse_inode",
1203 sizeof(struct fuse_inode
),
1204 0, SLAB_HWCACHE_ALIGN
,
1205 fuse_inode_init_once
);
1207 if (!fuse_inode_cachep
)
1210 err
= register_fuseblk();
1214 err
= register_filesystem(&fuse_fs_type
);
1221 unregister_fuseblk();
1223 kmem_cache_destroy(fuse_inode_cachep
);
1228 static void fuse_fs_cleanup(void)
1230 unregister_filesystem(&fuse_fs_type
);
1231 unregister_fuseblk();
1234 * Make sure all delayed rcu free inodes are flushed before we
1238 kmem_cache_destroy(fuse_inode_cachep
);
1241 static struct kobject
*fuse_kobj
;
1242 static struct kobject
*connections_kobj
;
1244 static int fuse_sysfs_init(void)
1248 fuse_kobj
= kobject_create_and_add("fuse", fs_kobj
);
1254 connections_kobj
= kobject_create_and_add("connections", fuse_kobj
);
1255 if (!connections_kobj
) {
1257 goto out_fuse_unregister
;
1262 out_fuse_unregister
:
1263 kobject_put(fuse_kobj
);
1268 static void fuse_sysfs_cleanup(void)
1270 kobject_put(connections_kobj
);
1271 kobject_put(fuse_kobj
);
1274 static int __init
fuse_init(void)
1278 printk(KERN_INFO
"fuse init (API version %i.%i)\n",
1279 FUSE_KERNEL_VERSION
, FUSE_KERNEL_MINOR_VERSION
);
1281 INIT_LIST_HEAD(&fuse_conn_list
);
1282 res
= fuse_fs_init();
1286 res
= fuse_dev_init();
1288 goto err_fs_cleanup
;
1290 res
= fuse_sysfs_init();
1292 goto err_dev_cleanup
;
1294 res
= fuse_ctl_init();
1296 goto err_sysfs_cleanup
;
1298 sanitize_global_limit(&max_user_bgreq
);
1299 sanitize_global_limit(&max_user_congthresh
);
1304 fuse_sysfs_cleanup();
1313 static void __exit
fuse_exit(void)
1315 printk(KERN_DEBUG
"fuse exit\n");
1318 fuse_sysfs_cleanup();
1323 module_init(fuse_init
);
1324 module_exit(fuse_exit
);