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>
23 #include <linux/posix_acl.h>
24 #include <linux/pid_namespace.h>
26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27 MODULE_DESCRIPTION("Filesystem in Userspace");
28 MODULE_LICENSE("GPL");
30 static struct kmem_cache
*fuse_inode_cachep
;
31 struct list_head fuse_conn_list
;
32 DEFINE_MUTEX(fuse_mutex
);
34 static int set_global_limit(const char *val
, struct kernel_param
*kp
);
36 unsigned max_user_bgreq
;
37 module_param_call(max_user_bgreq
, set_global_limit
, param_get_uint
,
38 &max_user_bgreq
, 0644);
39 __MODULE_PARM_TYPE(max_user_bgreq
, "uint");
40 MODULE_PARM_DESC(max_user_bgreq
,
41 "Global limit for the maximum number of backgrounded requests an "
42 "unprivileged user can set");
44 unsigned max_user_congthresh
;
45 module_param_call(max_user_congthresh
, set_global_limit
, param_get_uint
,
46 &max_user_congthresh
, 0644);
47 __MODULE_PARM_TYPE(max_user_congthresh
, "uint");
48 MODULE_PARM_DESC(max_user_congthresh
,
49 "Global limit for the maximum congestion threshold an "
50 "unprivileged user can set");
52 #define FUSE_SUPER_MAGIC 0x65735546
54 #define FUSE_DEFAULT_BLKSIZE 512
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62 struct fuse_mount_data
{
67 unsigned fd_present
:1;
68 unsigned rootmode_present
:1;
69 unsigned user_id_present
:1;
70 unsigned group_id_present
:1;
71 unsigned default_permissions
:1;
72 unsigned allow_other
:1;
77 struct fuse_forget_link
*fuse_alloc_forget(void)
79 return kzalloc(sizeof(struct fuse_forget_link
), GFP_KERNEL
);
82 static struct inode
*fuse_alloc_inode(struct super_block
*sb
)
85 struct fuse_inode
*fi
;
87 inode
= kmem_cache_alloc(fuse_inode_cachep
, GFP_KERNEL
);
91 fi
= get_fuse_inode(inode
);
99 INIT_LIST_HEAD(&fi
->write_files
);
100 INIT_LIST_HEAD(&fi
->queued_writes
);
101 INIT_LIST_HEAD(&fi
->writepages
);
102 init_waitqueue_head(&fi
->page_waitq
);
103 mutex_init(&fi
->mutex
);
104 fi
->forget
= fuse_alloc_forget();
106 kmem_cache_free(fuse_inode_cachep
, inode
);
113 static void fuse_i_callback(struct rcu_head
*head
)
115 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
116 kmem_cache_free(fuse_inode_cachep
, inode
);
119 static void fuse_destroy_inode(struct inode
*inode
)
121 struct fuse_inode
*fi
= get_fuse_inode(inode
);
122 BUG_ON(!list_empty(&fi
->write_files
));
123 BUG_ON(!list_empty(&fi
->queued_writes
));
124 mutex_destroy(&fi
->mutex
);
126 call_rcu(&inode
->i_rcu
, fuse_i_callback
);
129 static void fuse_evict_inode(struct inode
*inode
)
131 truncate_inode_pages_final(&inode
->i_data
);
133 if (inode
->i_sb
->s_flags
& MS_ACTIVE
) {
134 struct fuse_conn
*fc
= get_fuse_conn(inode
);
135 struct fuse_inode
*fi
= get_fuse_inode(inode
);
136 fuse_queue_forget(fc
, fi
->forget
, fi
->nodeid
, fi
->nlookup
);
141 static int fuse_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
144 if (*flags
& MS_MANDLOCK
)
151 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
152 * so that it will fit.
154 static ino_t
fuse_squash_ino(u64 ino64
)
156 ino_t ino
= (ino_t
) ino64
;
157 if (sizeof(ino_t
) < sizeof(u64
))
158 ino
^= ino64
>> (sizeof(u64
) - sizeof(ino_t
)) * 8;
162 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
165 struct fuse_conn
*fc
= get_fuse_conn(inode
);
166 struct fuse_inode
*fi
= get_fuse_inode(inode
);
168 fi
->attr_version
= ++fc
->attr_version
;
169 fi
->i_time
= attr_valid
;
171 inode
->i_ino
= fuse_squash_ino(attr
->ino
);
172 inode
->i_mode
= (inode
->i_mode
& S_IFMT
) | (attr
->mode
& 07777);
173 set_nlink(inode
, attr
->nlink
);
174 inode
->i_uid
= make_kuid(&init_user_ns
, attr
->uid
);
175 inode
->i_gid
= make_kgid(&init_user_ns
, attr
->gid
);
176 inode
->i_blocks
= attr
->blocks
;
177 inode
->i_atime
.tv_sec
= attr
->atime
;
178 inode
->i_atime
.tv_nsec
= attr
->atimensec
;
179 /* mtime from server may be stale due to local buffered write */
180 if (!fc
->writeback_cache
|| !S_ISREG(inode
->i_mode
)) {
181 inode
->i_mtime
.tv_sec
= attr
->mtime
;
182 inode
->i_mtime
.tv_nsec
= attr
->mtimensec
;
183 inode
->i_ctime
.tv_sec
= attr
->ctime
;
184 inode
->i_ctime
.tv_nsec
= attr
->ctimensec
;
187 if (attr
->blksize
!= 0)
188 inode
->i_blkbits
= ilog2(attr
->blksize
);
190 inode
->i_blkbits
= inode
->i_sb
->s_blocksize_bits
;
193 * Don't set the sticky bit in i_mode, unless we want the VFS
194 * to check permissions. This prevents failures due to the
195 * check in may_delete().
197 fi
->orig_i_mode
= inode
->i_mode
;
198 if (!fc
->default_permissions
)
199 inode
->i_mode
&= ~S_ISVTX
;
201 fi
->orig_ino
= attr
->ino
;
204 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
205 u64 attr_valid
, u64 attr_version
)
207 struct fuse_conn
*fc
= get_fuse_conn(inode
);
208 struct fuse_inode
*fi
= get_fuse_inode(inode
);
209 bool is_wb
= fc
->writeback_cache
;
211 struct timespec old_mtime
;
213 spin_lock(&fc
->lock
);
214 if ((attr_version
!= 0 && fi
->attr_version
> attr_version
) ||
215 test_bit(FUSE_I_SIZE_UNSTABLE
, &fi
->state
)) {
216 spin_unlock(&fc
->lock
);
220 old_mtime
= inode
->i_mtime
;
221 fuse_change_attributes_common(inode
, attr
, attr_valid
);
223 oldsize
= inode
->i_size
;
225 * In case of writeback_cache enabled, the cached writes beyond EOF
226 * extend local i_size without keeping userspace server in sync. So,
227 * attr->size coming from server can be stale. We cannot trust it.
229 if (!is_wb
|| !S_ISREG(inode
->i_mode
))
230 i_size_write(inode
, attr
->size
);
231 spin_unlock(&fc
->lock
);
233 if (!is_wb
&& S_ISREG(inode
->i_mode
)) {
236 if (oldsize
!= attr
->size
) {
237 truncate_pagecache(inode
, attr
->size
);
239 } else if (fc
->auto_inval_data
) {
240 struct timespec new_mtime
= {
241 .tv_sec
= attr
->mtime
,
242 .tv_nsec
= attr
->mtimensec
,
246 * Auto inval mode also checks and invalidates if mtime
249 if (!timespec_equal(&old_mtime
, &new_mtime
))
254 invalidate_inode_pages2(inode
->i_mapping
);
258 static void fuse_init_inode(struct inode
*inode
, struct fuse_attr
*attr
)
260 inode
->i_mode
= attr
->mode
& S_IFMT
;
261 inode
->i_size
= attr
->size
;
262 inode
->i_mtime
.tv_sec
= attr
->mtime
;
263 inode
->i_mtime
.tv_nsec
= attr
->mtimensec
;
264 inode
->i_ctime
.tv_sec
= attr
->ctime
;
265 inode
->i_ctime
.tv_nsec
= attr
->ctimensec
;
266 if (S_ISREG(inode
->i_mode
)) {
267 fuse_init_common(inode
);
268 fuse_init_file_inode(inode
);
269 } else if (S_ISDIR(inode
->i_mode
))
270 fuse_init_dir(inode
);
271 else if (S_ISLNK(inode
->i_mode
))
272 fuse_init_symlink(inode
);
273 else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
274 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
275 fuse_init_common(inode
);
276 init_special_inode(inode
, inode
->i_mode
,
277 new_decode_dev(attr
->rdev
));
282 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
)
284 u64 nodeid
= *(u64
*) _nodeidp
;
285 if (get_node_id(inode
) == nodeid
)
291 static int fuse_inode_set(struct inode
*inode
, void *_nodeidp
)
293 u64 nodeid
= *(u64
*) _nodeidp
;
294 get_fuse_inode(inode
)->nodeid
= nodeid
;
298 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
299 int generation
, struct fuse_attr
*attr
,
300 u64 attr_valid
, u64 attr_version
)
303 struct fuse_inode
*fi
;
304 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
307 inode
= iget5_locked(sb
, nodeid
, fuse_inode_eq
, fuse_inode_set
, &nodeid
);
311 if ((inode
->i_state
& I_NEW
)) {
312 inode
->i_flags
|= S_NOATIME
;
313 if (!fc
->writeback_cache
|| !S_ISREG(attr
->mode
))
314 inode
->i_flags
|= S_NOCMTIME
;
315 inode
->i_generation
= generation
;
316 fuse_init_inode(inode
, attr
);
317 unlock_new_inode(inode
);
318 } else if ((inode
->i_mode
^ attr
->mode
) & S_IFMT
) {
319 /* Inode has changed type, any I/O on the old should fail */
320 make_bad_inode(inode
);
325 fi
= get_fuse_inode(inode
);
326 spin_lock(&fc
->lock
);
328 spin_unlock(&fc
->lock
);
329 fuse_change_attributes(inode
, attr
, attr_valid
, attr_version
);
334 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
335 loff_t offset
, loff_t len
)
341 inode
= ilookup5(sb
, nodeid
, fuse_inode_eq
, &nodeid
);
345 fuse_invalidate_attr(inode
);
346 forget_all_cached_acls(inode
);
348 pg_start
= offset
>> PAGE_SHIFT
;
352 pg_end
= (offset
+ len
- 1) >> PAGE_SHIFT
;
353 invalidate_inode_pages2_range(inode
->i_mapping
,
360 void fuse_lock_inode(struct inode
*inode
)
362 if (!get_fuse_conn(inode
)->parallel_dirops
)
363 mutex_lock(&get_fuse_inode(inode
)->mutex
);
366 void fuse_unlock_inode(struct inode
*inode
)
368 if (!get_fuse_conn(inode
)->parallel_dirops
)
369 mutex_unlock(&get_fuse_inode(inode
)->mutex
);
372 static void fuse_umount_begin(struct super_block
*sb
)
374 fuse_abort_conn(get_fuse_conn_super(sb
));
377 static void fuse_send_destroy(struct fuse_conn
*fc
)
379 struct fuse_req
*req
= fc
->destroy_req
;
380 if (req
&& fc
->conn_init
) {
381 fc
->destroy_req
= NULL
;
382 req
->in
.h
.opcode
= FUSE_DESTROY
;
383 __set_bit(FR_FORCE
, &req
->flags
);
384 __clear_bit(FR_BACKGROUND
, &req
->flags
);
385 fuse_request_send(fc
, req
);
386 fuse_put_request(fc
, req
);
390 static void fuse_bdi_destroy(struct fuse_conn
*fc
)
392 if (fc
->bdi_initialized
)
393 bdi_destroy(&fc
->bdi
);
396 static void fuse_put_super(struct super_block
*sb
)
398 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
400 fuse_send_destroy(fc
);
403 mutex_lock(&fuse_mutex
);
404 list_del(&fc
->entry
);
405 fuse_ctl_remove_conn(fc
);
406 mutex_unlock(&fuse_mutex
);
407 fuse_bdi_destroy(fc
);
412 static void convert_fuse_statfs(struct kstatfs
*stbuf
, struct fuse_kstatfs
*attr
)
414 stbuf
->f_type
= FUSE_SUPER_MAGIC
;
415 stbuf
->f_bsize
= attr
->bsize
;
416 stbuf
->f_frsize
= attr
->frsize
;
417 stbuf
->f_blocks
= attr
->blocks
;
418 stbuf
->f_bfree
= attr
->bfree
;
419 stbuf
->f_bavail
= attr
->bavail
;
420 stbuf
->f_files
= attr
->files
;
421 stbuf
->f_ffree
= attr
->ffree
;
422 stbuf
->f_namelen
= attr
->namelen
;
423 /* fsid is left zero */
426 static int fuse_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
428 struct super_block
*sb
= dentry
->d_sb
;
429 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
431 struct fuse_statfs_out outarg
;
434 if (!fuse_allow_current_process(fc
)) {
435 buf
->f_type
= FUSE_SUPER_MAGIC
;
439 memset(&outarg
, 0, sizeof(outarg
));
441 args
.in
.h
.opcode
= FUSE_STATFS
;
442 args
.in
.h
.nodeid
= get_node_id(d_inode(dentry
));
443 args
.out
.numargs
= 1;
444 args
.out
.args
[0].size
= sizeof(outarg
);
445 args
.out
.args
[0].value
= &outarg
;
446 err
= fuse_simple_request(fc
, &args
);
448 convert_fuse_statfs(buf
, &outarg
.st
);
457 OPT_DEFAULT_PERMISSIONS
,
464 static const match_table_t tokens
= {
466 {OPT_ROOTMODE
, "rootmode=%o"},
467 {OPT_USER_ID
, "user_id=%u"},
468 {OPT_GROUP_ID
, "group_id=%u"},
469 {OPT_DEFAULT_PERMISSIONS
, "default_permissions"},
470 {OPT_ALLOW_OTHER
, "allow_other"},
471 {OPT_MAX_READ
, "max_read=%u"},
472 {OPT_BLKSIZE
, "blksize=%u"},
476 static int fuse_match_uint(substring_t
*s
, unsigned int *res
)
479 char *buf
= match_strdup(s
);
481 err
= kstrtouint(buf
, 10, res
);
487 static int parse_fuse_opt(char *opt
, struct fuse_mount_data
*d
, int is_bdev
)
490 memset(d
, 0, sizeof(struct fuse_mount_data
));
492 d
->blksize
= FUSE_DEFAULT_BLKSIZE
;
494 while ((p
= strsep(&opt
, ",")) != NULL
) {
498 substring_t args
[MAX_OPT_ARGS
];
502 token
= match_token(p
, tokens
, args
);
505 if (match_int(&args
[0], &value
))
512 if (match_octal(&args
[0], &value
))
514 if (!fuse_valid_type(value
))
517 d
->rootmode_present
= 1;
521 if (fuse_match_uint(&args
[0], &uv
))
523 d
->user_id
= make_kuid(current_user_ns(), uv
);
524 if (!uid_valid(d
->user_id
))
526 d
->user_id_present
= 1;
530 if (fuse_match_uint(&args
[0], &uv
))
532 d
->group_id
= make_kgid(current_user_ns(), uv
);
533 if (!gid_valid(d
->group_id
))
535 d
->group_id_present
= 1;
538 case OPT_DEFAULT_PERMISSIONS
:
539 d
->default_permissions
= 1;
542 case OPT_ALLOW_OTHER
:
547 if (match_int(&args
[0], &value
))
553 if (!is_bdev
|| match_int(&args
[0], &value
))
563 if (!d
->fd_present
|| !d
->rootmode_present
||
564 !d
->user_id_present
|| !d
->group_id_present
)
570 static int fuse_show_options(struct seq_file
*m
, struct dentry
*root
)
572 struct super_block
*sb
= root
->d_sb
;
573 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
575 seq_printf(m
, ",user_id=%u", from_kuid_munged(&init_user_ns
, fc
->user_id
));
576 seq_printf(m
, ",group_id=%u", from_kgid_munged(&init_user_ns
, fc
->group_id
));
577 if (fc
->default_permissions
)
578 seq_puts(m
, ",default_permissions");
580 seq_puts(m
, ",allow_other");
581 if (fc
->max_read
!= ~0)
582 seq_printf(m
, ",max_read=%u", fc
->max_read
);
583 if (sb
->s_bdev
&& sb
->s_blocksize
!= FUSE_DEFAULT_BLKSIZE
)
584 seq_printf(m
, ",blksize=%lu", sb
->s_blocksize
);
588 static void fuse_iqueue_init(struct fuse_iqueue
*fiq
)
590 memset(fiq
, 0, sizeof(struct fuse_iqueue
));
591 init_waitqueue_head(&fiq
->waitq
);
592 INIT_LIST_HEAD(&fiq
->pending
);
593 INIT_LIST_HEAD(&fiq
->interrupts
);
594 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
598 static void fuse_pqueue_init(struct fuse_pqueue
*fpq
)
600 memset(fpq
, 0, sizeof(struct fuse_pqueue
));
601 spin_lock_init(&fpq
->lock
);
602 INIT_LIST_HEAD(&fpq
->processing
);
603 INIT_LIST_HEAD(&fpq
->io
);
607 void fuse_conn_init(struct fuse_conn
*fc
)
609 memset(fc
, 0, sizeof(*fc
));
610 spin_lock_init(&fc
->lock
);
611 init_rwsem(&fc
->killsb
);
612 atomic_set(&fc
->count
, 1);
613 atomic_set(&fc
->dev_count
, 1);
614 init_waitqueue_head(&fc
->blocked_waitq
);
615 init_waitqueue_head(&fc
->reserved_req_waitq
);
616 fuse_iqueue_init(&fc
->iq
);
617 INIT_LIST_HEAD(&fc
->bg_queue
);
618 INIT_LIST_HEAD(&fc
->entry
);
619 INIT_LIST_HEAD(&fc
->devices
);
620 atomic_set(&fc
->num_waiting
, 0);
621 fc
->max_background
= FUSE_DEFAULT_MAX_BACKGROUND
;
622 fc
->congestion_threshold
= FUSE_DEFAULT_CONGESTION_THRESHOLD
;
624 fc
->polled_files
= RB_ROOT
;
628 fc
->attr_version
= 1;
629 get_random_bytes(&fc
->scramble_key
, sizeof(fc
->scramble_key
));
630 fc
->pid_ns
= get_pid_ns(task_active_pid_ns(current
));
632 EXPORT_SYMBOL_GPL(fuse_conn_init
);
634 void fuse_conn_put(struct fuse_conn
*fc
)
636 if (atomic_dec_and_test(&fc
->count
)) {
638 fuse_request_free(fc
->destroy_req
);
639 put_pid_ns(fc
->pid_ns
);
643 EXPORT_SYMBOL_GPL(fuse_conn_put
);
645 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
)
647 atomic_inc(&fc
->count
);
650 EXPORT_SYMBOL_GPL(fuse_conn_get
);
652 static struct inode
*fuse_get_root_inode(struct super_block
*sb
, unsigned mode
)
654 struct fuse_attr attr
;
655 memset(&attr
, 0, sizeof(attr
));
658 attr
.ino
= FUSE_ROOT_ID
;
660 return fuse_iget(sb
, 1, 0, &attr
, 0, 0);
663 struct fuse_inode_handle
{
668 static struct dentry
*fuse_get_dentry(struct super_block
*sb
,
669 struct fuse_inode_handle
*handle
)
671 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
673 struct dentry
*entry
;
676 if (handle
->nodeid
== 0)
679 inode
= ilookup5(sb
, handle
->nodeid
, fuse_inode_eq
, &handle
->nodeid
);
681 struct fuse_entry_out outarg
;
682 const struct qstr name
= QSTR_INIT(".", 1);
684 if (!fc
->export_support
)
687 err
= fuse_lookup_name(sb
, handle
->nodeid
, &name
, &outarg
,
689 if (err
&& err
!= -ENOENT
)
696 if (get_node_id(inode
) != handle
->nodeid
)
700 if (inode
->i_generation
!= handle
->generation
)
703 entry
= d_obtain_alias(inode
);
704 if (!IS_ERR(entry
) && get_node_id(inode
) != FUSE_ROOT_ID
)
705 fuse_invalidate_entry_cache(entry
);
715 static int fuse_encode_fh(struct inode
*inode
, u32
*fh
, int *max_len
,
716 struct inode
*parent
)
718 int len
= parent
? 6 : 3;
722 if (*max_len
< len
) {
724 return FILEID_INVALID
;
727 nodeid
= get_fuse_inode(inode
)->nodeid
;
728 generation
= inode
->i_generation
;
730 fh
[0] = (u32
)(nodeid
>> 32);
731 fh
[1] = (u32
)(nodeid
& 0xffffffff);
735 nodeid
= get_fuse_inode(parent
)->nodeid
;
736 generation
= parent
->i_generation
;
738 fh
[3] = (u32
)(nodeid
>> 32);
739 fh
[4] = (u32
)(nodeid
& 0xffffffff);
744 return parent
? 0x82 : 0x81;
747 static struct dentry
*fuse_fh_to_dentry(struct super_block
*sb
,
748 struct fid
*fid
, int fh_len
, int fh_type
)
750 struct fuse_inode_handle handle
;
752 if ((fh_type
!= 0x81 && fh_type
!= 0x82) || fh_len
< 3)
755 handle
.nodeid
= (u64
) fid
->raw
[0] << 32;
756 handle
.nodeid
|= (u64
) fid
->raw
[1];
757 handle
.generation
= fid
->raw
[2];
758 return fuse_get_dentry(sb
, &handle
);
761 static struct dentry
*fuse_fh_to_parent(struct super_block
*sb
,
762 struct fid
*fid
, int fh_len
, int fh_type
)
764 struct fuse_inode_handle parent
;
766 if (fh_type
!= 0x82 || fh_len
< 6)
769 parent
.nodeid
= (u64
) fid
->raw
[3] << 32;
770 parent
.nodeid
|= (u64
) fid
->raw
[4];
771 parent
.generation
= fid
->raw
[5];
772 return fuse_get_dentry(sb
, &parent
);
775 static struct dentry
*fuse_get_parent(struct dentry
*child
)
777 struct inode
*child_inode
= d_inode(child
);
778 struct fuse_conn
*fc
= get_fuse_conn(child_inode
);
780 struct dentry
*parent
;
781 struct fuse_entry_out outarg
;
782 const struct qstr name
= QSTR_INIT("..", 2);
785 if (!fc
->export_support
)
786 return ERR_PTR(-ESTALE
);
788 err
= fuse_lookup_name(child_inode
->i_sb
, get_node_id(child_inode
),
789 &name
, &outarg
, &inode
);
792 return ERR_PTR(-ESTALE
);
796 parent
= d_obtain_alias(inode
);
797 if (!IS_ERR(parent
) && get_node_id(inode
) != FUSE_ROOT_ID
)
798 fuse_invalidate_entry_cache(parent
);
803 static const struct export_operations fuse_export_operations
= {
804 .fh_to_dentry
= fuse_fh_to_dentry
,
805 .fh_to_parent
= fuse_fh_to_parent
,
806 .encode_fh
= fuse_encode_fh
,
807 .get_parent
= fuse_get_parent
,
810 static const struct super_operations fuse_super_operations
= {
811 .alloc_inode
= fuse_alloc_inode
,
812 .destroy_inode
= fuse_destroy_inode
,
813 .evict_inode
= fuse_evict_inode
,
814 .write_inode
= fuse_write_inode
,
815 .drop_inode
= generic_delete_inode
,
816 .remount_fs
= fuse_remount_fs
,
817 .put_super
= fuse_put_super
,
818 .umount_begin
= fuse_umount_begin
,
819 .statfs
= fuse_statfs
,
820 .show_options
= fuse_show_options
,
823 static void sanitize_global_limit(unsigned *limit
)
826 *limit
= ((totalram_pages
<< PAGE_SHIFT
) >> 13) /
827 sizeof(struct fuse_req
);
829 if (*limit
>= 1 << 16)
830 *limit
= (1 << 16) - 1;
833 static int set_global_limit(const char *val
, struct kernel_param
*kp
)
837 rv
= param_set_uint(val
, kp
);
841 sanitize_global_limit((unsigned *)kp
->arg
);
846 static void process_init_limits(struct fuse_conn
*fc
, struct fuse_init_out
*arg
)
848 int cap_sys_admin
= capable(CAP_SYS_ADMIN
);
853 sanitize_global_limit(&max_user_bgreq
);
854 sanitize_global_limit(&max_user_congthresh
);
856 if (arg
->max_background
) {
857 fc
->max_background
= arg
->max_background
;
859 if (!cap_sys_admin
&& fc
->max_background
> max_user_bgreq
)
860 fc
->max_background
= max_user_bgreq
;
862 if (arg
->congestion_threshold
) {
863 fc
->congestion_threshold
= arg
->congestion_threshold
;
865 if (!cap_sys_admin
&&
866 fc
->congestion_threshold
> max_user_congthresh
)
867 fc
->congestion_threshold
= max_user_congthresh
;
871 static void process_init_reply(struct fuse_conn
*fc
, struct fuse_req
*req
)
873 struct fuse_init_out
*arg
= &req
->misc
.init_out
;
875 if (req
->out
.h
.error
|| arg
->major
!= FUSE_KERNEL_VERSION
)
878 unsigned long ra_pages
;
880 process_init_limits(fc
, arg
);
882 if (arg
->minor
>= 6) {
883 ra_pages
= arg
->max_readahead
/ PAGE_SIZE
;
884 if (arg
->flags
& FUSE_ASYNC_READ
)
886 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
888 if (arg
->minor
>= 17) {
889 if (!(arg
->flags
& FUSE_FLOCK_LOCKS
))
892 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
895 if (arg
->flags
& FUSE_ATOMIC_O_TRUNC
)
896 fc
->atomic_o_trunc
= 1;
897 if (arg
->minor
>= 9) {
898 /* LOOKUP has dependency on proto version */
899 if (arg
->flags
& FUSE_EXPORT_SUPPORT
)
900 fc
->export_support
= 1;
902 if (arg
->flags
& FUSE_BIG_WRITES
)
904 if (arg
->flags
& FUSE_DONT_MASK
)
906 if (arg
->flags
& FUSE_AUTO_INVAL_DATA
)
907 fc
->auto_inval_data
= 1;
908 if (arg
->flags
& FUSE_DO_READDIRPLUS
) {
909 fc
->do_readdirplus
= 1;
910 if (arg
->flags
& FUSE_READDIRPLUS_AUTO
)
911 fc
->readdirplus_auto
= 1;
913 if (arg
->flags
& FUSE_ASYNC_DIO
)
915 if (arg
->flags
& FUSE_WRITEBACK_CACHE
)
916 fc
->writeback_cache
= 1;
917 if (arg
->flags
& FUSE_PARALLEL_DIROPS
)
918 fc
->parallel_dirops
= 1;
919 if (arg
->flags
& FUSE_HANDLE_KILLPRIV
)
920 fc
->handle_killpriv
= 1;
921 if (arg
->time_gran
&& arg
->time_gran
<= 1000000000)
922 fc
->sb
->s_time_gran
= arg
->time_gran
;
923 if ((arg
->flags
& FUSE_POSIX_ACL
)) {
924 fc
->default_permissions
= 1;
926 fc
->sb
->s_xattr
= fuse_acl_xattr_handlers
;
929 ra_pages
= fc
->max_read
/ PAGE_SIZE
;
934 fc
->bdi
.ra_pages
= min(fc
->bdi
.ra_pages
, ra_pages
);
935 fc
->minor
= arg
->minor
;
936 fc
->max_write
= arg
->minor
< 5 ? 4096 : arg
->max_write
;
937 fc
->max_write
= max_t(unsigned, 4096, fc
->max_write
);
940 fuse_set_initialized(fc
);
941 wake_up_all(&fc
->blocked_waitq
);
944 static void fuse_send_init(struct fuse_conn
*fc
, struct fuse_req
*req
)
946 struct fuse_init_in
*arg
= &req
->misc
.init_in
;
948 arg
->major
= FUSE_KERNEL_VERSION
;
949 arg
->minor
= FUSE_KERNEL_MINOR_VERSION
;
950 arg
->max_readahead
= fc
->bdi
.ra_pages
* PAGE_SIZE
;
951 arg
->flags
|= FUSE_ASYNC_READ
| FUSE_POSIX_LOCKS
| FUSE_ATOMIC_O_TRUNC
|
952 FUSE_EXPORT_SUPPORT
| FUSE_BIG_WRITES
| FUSE_DONT_MASK
|
953 FUSE_SPLICE_WRITE
| FUSE_SPLICE_MOVE
| FUSE_SPLICE_READ
|
954 FUSE_FLOCK_LOCKS
| FUSE_HAS_IOCTL_DIR
| FUSE_AUTO_INVAL_DATA
|
955 FUSE_DO_READDIRPLUS
| FUSE_READDIRPLUS_AUTO
| FUSE_ASYNC_DIO
|
956 FUSE_WRITEBACK_CACHE
| FUSE_NO_OPEN_SUPPORT
|
957 FUSE_PARALLEL_DIROPS
| FUSE_HANDLE_KILLPRIV
| FUSE_POSIX_ACL
;
958 req
->in
.h
.opcode
= FUSE_INIT
;
960 req
->in
.args
[0].size
= sizeof(*arg
);
961 req
->in
.args
[0].value
= arg
;
962 req
->out
.numargs
= 1;
963 /* Variable length argument used for backward compatibility
964 with interface version < 7.5. Rest of init_out is zeroed
965 by do_get_request(), so a short reply is not a problem */
967 req
->out
.args
[0].size
= sizeof(struct fuse_init_out
);
968 req
->out
.args
[0].value
= &req
->misc
.init_out
;
969 req
->end
= process_init_reply
;
970 fuse_request_send_background(fc
, req
);
973 static void fuse_free_conn(struct fuse_conn
*fc
)
975 WARN_ON(!list_empty(&fc
->devices
));
979 static int fuse_bdi_init(struct fuse_conn
*fc
, struct super_block
*sb
)
983 fc
->bdi
.name
= "fuse";
984 fc
->bdi
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
985 /* fuse does it's own writeback accounting */
986 fc
->bdi
.capabilities
= BDI_CAP_NO_ACCT_WB
| BDI_CAP_STRICTLIMIT
;
988 err
= bdi_init(&fc
->bdi
);
992 fc
->bdi_initialized
= 1;
995 err
= bdi_register(&fc
->bdi
, NULL
, "%u:%u-fuseblk",
996 MAJOR(fc
->dev
), MINOR(fc
->dev
));
998 err
= bdi_register_dev(&fc
->bdi
, fc
->dev
);
1005 * For a single fuse filesystem use max 1% of dirty +
1006 * writeback threshold.
1008 * This gives about 1M of write buffer for memory maps on a
1009 * machine with 1G and 10% dirty_ratio, which should be more
1012 * Privileged users can raise it by writing to
1014 * /sys/class/bdi/<bdi>/max_ratio
1016 bdi_set_max_ratio(&fc
->bdi
, 1);
1021 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
)
1023 struct fuse_dev
*fud
;
1025 fud
= kzalloc(sizeof(struct fuse_dev
), GFP_KERNEL
);
1027 fud
->fc
= fuse_conn_get(fc
);
1028 fuse_pqueue_init(&fud
->pq
);
1030 spin_lock(&fc
->lock
);
1031 list_add_tail(&fud
->entry
, &fc
->devices
);
1032 spin_unlock(&fc
->lock
);
1037 EXPORT_SYMBOL_GPL(fuse_dev_alloc
);
1039 void fuse_dev_free(struct fuse_dev
*fud
)
1041 struct fuse_conn
*fc
= fud
->fc
;
1044 spin_lock(&fc
->lock
);
1045 list_del(&fud
->entry
);
1046 spin_unlock(&fc
->lock
);
1052 EXPORT_SYMBOL_GPL(fuse_dev_free
);
1054 static int fuse_fill_super(struct super_block
*sb
, void *data
, int silent
)
1056 struct fuse_dev
*fud
;
1057 struct fuse_conn
*fc
;
1059 struct fuse_mount_data d
;
1061 struct dentry
*root_dentry
;
1062 struct fuse_req
*init_req
;
1064 int is_bdev
= sb
->s_bdev
!= NULL
;
1067 if (sb
->s_flags
& MS_MANDLOCK
)
1070 sb
->s_flags
&= ~(MS_NOSEC
| MS_I_VERSION
);
1072 if (!parse_fuse_opt(data
, &d
, is_bdev
))
1078 if (!sb_set_blocksize(sb
, d
.blksize
))
1082 sb
->s_blocksize
= PAGE_SIZE
;
1083 sb
->s_blocksize_bits
= PAGE_SHIFT
;
1085 sb
->s_magic
= FUSE_SUPER_MAGIC
;
1086 sb
->s_op
= &fuse_super_operations
;
1087 sb
->s_xattr
= fuse_xattr_handlers
;
1088 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1089 sb
->s_time_gran
= 1;
1090 sb
->s_export_op
= &fuse_export_operations
;
1097 if ((file
->f_op
!= &fuse_dev_operations
) ||
1098 (file
->f_cred
->user_ns
!= &init_user_ns
))
1101 fc
= kmalloc(sizeof(*fc
), GFP_KERNEL
);
1107 fc
->release
= fuse_free_conn
;
1109 fud
= fuse_dev_alloc(fc
);
1113 fc
->dev
= sb
->s_dev
;
1115 err
= fuse_bdi_init(fc
, sb
);
1119 sb
->s_bdi
= &fc
->bdi
;
1121 /* Handle umasking inside the fuse code */
1122 if (sb
->s_flags
& MS_POSIXACL
)
1124 sb
->s_flags
|= MS_POSIXACL
;
1126 fc
->default_permissions
= d
.default_permissions
;
1127 fc
->allow_other
= d
.allow_other
;
1128 fc
->user_id
= d
.user_id
;
1129 fc
->group_id
= d
.group_id
;
1130 fc
->max_read
= max_t(unsigned, 4096, d
.max_read
);
1132 /* Used by get_root_inode() */
1136 root
= fuse_get_root_inode(sb
, d
.rootmode
);
1137 sb
->s_d_op
= &fuse_root_dentry_operations
;
1138 root_dentry
= d_make_root(root
);
1141 /* Root dentry doesn't have .d_revalidate */
1142 sb
->s_d_op
= &fuse_dentry_operations
;
1144 init_req
= fuse_request_alloc(0);
1147 __set_bit(FR_BACKGROUND
, &init_req
->flags
);
1150 fc
->destroy_req
= fuse_request_alloc(0);
1151 if (!fc
->destroy_req
)
1152 goto err_free_init_req
;
1155 mutex_lock(&fuse_mutex
);
1157 if (file
->private_data
)
1160 err
= fuse_ctl_add_conn(fc
);
1164 list_add_tail(&fc
->entry
, &fuse_conn_list
);
1165 sb
->s_root
= root_dentry
;
1166 file
->private_data
= fud
;
1167 mutex_unlock(&fuse_mutex
);
1169 * atomic_dec_and_test() in fput() provides the necessary
1170 * memory barrier for file->private_data to be visible on all
1175 fuse_send_init(fc
, init_req
);
1180 mutex_unlock(&fuse_mutex
);
1182 fuse_request_free(init_req
);
1188 fuse_bdi_destroy(fc
);
1196 static struct dentry
*fuse_mount(struct file_system_type
*fs_type
,
1197 int flags
, const char *dev_name
,
1200 return mount_nodev(fs_type
, flags
, raw_data
, fuse_fill_super
);
1203 static void fuse_kill_sb_anon(struct super_block
*sb
)
1205 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1208 down_write(&fc
->killsb
);
1210 up_write(&fc
->killsb
);
1213 kill_anon_super(sb
);
1216 static struct file_system_type fuse_fs_type
= {
1217 .owner
= THIS_MODULE
,
1219 .fs_flags
= FS_HAS_SUBTYPE
,
1220 .mount
= fuse_mount
,
1221 .kill_sb
= fuse_kill_sb_anon
,
1223 MODULE_ALIAS_FS("fuse");
1226 static struct dentry
*fuse_mount_blk(struct file_system_type
*fs_type
,
1227 int flags
, const char *dev_name
,
1230 return mount_bdev(fs_type
, flags
, dev_name
, raw_data
, fuse_fill_super
);
1233 static void fuse_kill_sb_blk(struct super_block
*sb
)
1235 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1238 down_write(&fc
->killsb
);
1240 up_write(&fc
->killsb
);
1243 kill_block_super(sb
);
1246 static struct file_system_type fuseblk_fs_type
= {
1247 .owner
= THIS_MODULE
,
1249 .mount
= fuse_mount_blk
,
1250 .kill_sb
= fuse_kill_sb_blk
,
1251 .fs_flags
= FS_REQUIRES_DEV
| FS_HAS_SUBTYPE
,
1253 MODULE_ALIAS_FS("fuseblk");
1255 static inline int register_fuseblk(void)
1257 return register_filesystem(&fuseblk_fs_type
);
1260 static inline void unregister_fuseblk(void)
1262 unregister_filesystem(&fuseblk_fs_type
);
1265 static inline int register_fuseblk(void)
1270 static inline void unregister_fuseblk(void)
1275 static void fuse_inode_init_once(void *foo
)
1277 struct inode
*inode
= foo
;
1279 inode_init_once(inode
);
1282 static int __init
fuse_fs_init(void)
1286 fuse_inode_cachep
= kmem_cache_create("fuse_inode",
1287 sizeof(struct fuse_inode
), 0,
1288 SLAB_HWCACHE_ALIGN
|SLAB_ACCOUNT
,
1289 fuse_inode_init_once
);
1291 if (!fuse_inode_cachep
)
1294 err
= register_fuseblk();
1298 err
= register_filesystem(&fuse_fs_type
);
1305 unregister_fuseblk();
1307 kmem_cache_destroy(fuse_inode_cachep
);
1312 static void fuse_fs_cleanup(void)
1314 unregister_filesystem(&fuse_fs_type
);
1315 unregister_fuseblk();
1318 * Make sure all delayed rcu free inodes are flushed before we
1322 kmem_cache_destroy(fuse_inode_cachep
);
1325 static struct kobject
*fuse_kobj
;
1327 static int fuse_sysfs_init(void)
1331 fuse_kobj
= kobject_create_and_add("fuse", fs_kobj
);
1337 err
= sysfs_create_mount_point(fuse_kobj
, "connections");
1339 goto out_fuse_unregister
;
1343 out_fuse_unregister
:
1344 kobject_put(fuse_kobj
);
1349 static void fuse_sysfs_cleanup(void)
1351 sysfs_remove_mount_point(fuse_kobj
, "connections");
1352 kobject_put(fuse_kobj
);
1355 static int __init
fuse_init(void)
1359 printk(KERN_INFO
"fuse init (API version %i.%i)\n",
1360 FUSE_KERNEL_VERSION
, FUSE_KERNEL_MINOR_VERSION
);
1362 INIT_LIST_HEAD(&fuse_conn_list
);
1363 res
= fuse_fs_init();
1367 res
= fuse_dev_init();
1369 goto err_fs_cleanup
;
1371 res
= fuse_sysfs_init();
1373 goto err_dev_cleanup
;
1375 res
= fuse_ctl_init();
1377 goto err_sysfs_cleanup
;
1379 sanitize_global_limit(&max_user_bgreq
);
1380 sanitize_global_limit(&max_user_congthresh
);
1385 fuse_sysfs_cleanup();
1394 static void __exit
fuse_exit(void)
1396 printk(KERN_DEBUG
"fuse exit\n");
1399 fuse_sysfs_cleanup();
1404 module_init(fuse_init
);
1405 module_exit(fuse_exit
);