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
, const 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
& SB_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
& SB_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(fc
->user_ns
, attr
->uid
);
175 inode
->i_gid
= make_kgid(fc
->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
= timespec64_to_timespec(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
), false);
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_put_super(struct super_block
*sb
)
392 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
394 fuse_send_destroy(fc
);
396 fuse_abort_conn(fc
, false);
397 mutex_lock(&fuse_mutex
);
398 list_del(&fc
->entry
);
399 fuse_ctl_remove_conn(fc
);
400 mutex_unlock(&fuse_mutex
);
405 static void convert_fuse_statfs(struct kstatfs
*stbuf
, struct fuse_kstatfs
*attr
)
407 stbuf
->f_type
= FUSE_SUPER_MAGIC
;
408 stbuf
->f_bsize
= attr
->bsize
;
409 stbuf
->f_frsize
= attr
->frsize
;
410 stbuf
->f_blocks
= attr
->blocks
;
411 stbuf
->f_bfree
= attr
->bfree
;
412 stbuf
->f_bavail
= attr
->bavail
;
413 stbuf
->f_files
= attr
->files
;
414 stbuf
->f_ffree
= attr
->ffree
;
415 stbuf
->f_namelen
= attr
->namelen
;
416 /* fsid is left zero */
419 static int fuse_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
421 struct super_block
*sb
= dentry
->d_sb
;
422 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
424 struct fuse_statfs_out outarg
;
427 if (!fuse_allow_current_process(fc
)) {
428 buf
->f_type
= FUSE_SUPER_MAGIC
;
432 memset(&outarg
, 0, sizeof(outarg
));
434 args
.in
.h
.opcode
= FUSE_STATFS
;
435 args
.in
.h
.nodeid
= get_node_id(d_inode(dentry
));
436 args
.out
.numargs
= 1;
437 args
.out
.args
[0].size
= sizeof(outarg
);
438 args
.out
.args
[0].value
= &outarg
;
439 err
= fuse_simple_request(fc
, &args
);
441 convert_fuse_statfs(buf
, &outarg
.st
);
450 OPT_DEFAULT_PERMISSIONS
,
457 static const match_table_t tokens
= {
459 {OPT_ROOTMODE
, "rootmode=%o"},
460 {OPT_USER_ID
, "user_id=%u"},
461 {OPT_GROUP_ID
, "group_id=%u"},
462 {OPT_DEFAULT_PERMISSIONS
, "default_permissions"},
463 {OPT_ALLOW_OTHER
, "allow_other"},
464 {OPT_MAX_READ
, "max_read=%u"},
465 {OPT_BLKSIZE
, "blksize=%u"},
469 static int fuse_match_uint(substring_t
*s
, unsigned int *res
)
472 char *buf
= match_strdup(s
);
474 err
= kstrtouint(buf
, 10, res
);
480 static int parse_fuse_opt(char *opt
, struct fuse_mount_data
*d
, int is_bdev
,
481 struct user_namespace
*user_ns
)
484 memset(d
, 0, sizeof(struct fuse_mount_data
));
486 d
->blksize
= FUSE_DEFAULT_BLKSIZE
;
488 while ((p
= strsep(&opt
, ",")) != NULL
) {
492 substring_t args
[MAX_OPT_ARGS
];
496 token
= match_token(p
, tokens
, args
);
499 if (match_int(&args
[0], &value
))
506 if (match_octal(&args
[0], &value
))
508 if (!fuse_valid_type(value
))
511 d
->rootmode_present
= 1;
515 if (fuse_match_uint(&args
[0], &uv
))
517 d
->user_id
= make_kuid(user_ns
, uv
);
518 if (!uid_valid(d
->user_id
))
520 d
->user_id_present
= 1;
524 if (fuse_match_uint(&args
[0], &uv
))
526 d
->group_id
= make_kgid(user_ns
, uv
);
527 if (!gid_valid(d
->group_id
))
529 d
->group_id_present
= 1;
532 case OPT_DEFAULT_PERMISSIONS
:
533 d
->default_permissions
= 1;
536 case OPT_ALLOW_OTHER
:
541 if (match_int(&args
[0], &value
))
547 if (!is_bdev
|| match_int(&args
[0], &value
))
557 if (!d
->fd_present
|| !d
->rootmode_present
||
558 !d
->user_id_present
|| !d
->group_id_present
)
564 static int fuse_show_options(struct seq_file
*m
, struct dentry
*root
)
566 struct super_block
*sb
= root
->d_sb
;
567 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
569 seq_printf(m
, ",user_id=%u", from_kuid_munged(fc
->user_ns
, fc
->user_id
));
570 seq_printf(m
, ",group_id=%u", from_kgid_munged(fc
->user_ns
, fc
->group_id
));
571 if (fc
->default_permissions
)
572 seq_puts(m
, ",default_permissions");
574 seq_puts(m
, ",allow_other");
575 if (fc
->max_read
!= ~0)
576 seq_printf(m
, ",max_read=%u", fc
->max_read
);
577 if (sb
->s_bdev
&& sb
->s_blocksize
!= FUSE_DEFAULT_BLKSIZE
)
578 seq_printf(m
, ",blksize=%lu", sb
->s_blocksize
);
582 static void fuse_iqueue_init(struct fuse_iqueue
*fiq
)
584 memset(fiq
, 0, sizeof(struct fuse_iqueue
));
585 init_waitqueue_head(&fiq
->waitq
);
586 INIT_LIST_HEAD(&fiq
->pending
);
587 INIT_LIST_HEAD(&fiq
->interrupts
);
588 fiq
->forget_list_tail
= &fiq
->forget_list_head
;
592 static void fuse_pqueue_init(struct fuse_pqueue
*fpq
)
594 memset(fpq
, 0, sizeof(struct fuse_pqueue
));
595 spin_lock_init(&fpq
->lock
);
596 INIT_LIST_HEAD(&fpq
->processing
);
597 INIT_LIST_HEAD(&fpq
->io
);
601 void fuse_conn_init(struct fuse_conn
*fc
, struct user_namespace
*user_ns
)
603 memset(fc
, 0, sizeof(*fc
));
604 spin_lock_init(&fc
->lock
);
605 init_rwsem(&fc
->killsb
);
606 refcount_set(&fc
->count
, 1);
607 atomic_set(&fc
->dev_count
, 1);
608 init_waitqueue_head(&fc
->blocked_waitq
);
609 init_waitqueue_head(&fc
->reserved_req_waitq
);
610 fuse_iqueue_init(&fc
->iq
);
611 INIT_LIST_HEAD(&fc
->bg_queue
);
612 INIT_LIST_HEAD(&fc
->entry
);
613 INIT_LIST_HEAD(&fc
->devices
);
614 atomic_set(&fc
->num_waiting
, 0);
615 fc
->max_background
= FUSE_DEFAULT_MAX_BACKGROUND
;
616 fc
->congestion_threshold
= FUSE_DEFAULT_CONGESTION_THRESHOLD
;
618 fc
->polled_files
= RB_ROOT
;
622 fc
->attr_version
= 1;
623 get_random_bytes(&fc
->scramble_key
, sizeof(fc
->scramble_key
));
624 fc
->pid_ns
= get_pid_ns(task_active_pid_ns(current
));
625 fc
->user_ns
= get_user_ns(user_ns
);
627 EXPORT_SYMBOL_GPL(fuse_conn_init
);
629 void fuse_conn_put(struct fuse_conn
*fc
)
631 if (refcount_dec_and_test(&fc
->count
)) {
633 fuse_request_free(fc
->destroy_req
);
634 put_pid_ns(fc
->pid_ns
);
635 put_user_ns(fc
->user_ns
);
639 EXPORT_SYMBOL_GPL(fuse_conn_put
);
641 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
)
643 refcount_inc(&fc
->count
);
646 EXPORT_SYMBOL_GPL(fuse_conn_get
);
648 static struct inode
*fuse_get_root_inode(struct super_block
*sb
, unsigned mode
)
650 struct fuse_attr attr
;
651 memset(&attr
, 0, sizeof(attr
));
654 attr
.ino
= FUSE_ROOT_ID
;
656 return fuse_iget(sb
, 1, 0, &attr
, 0, 0);
659 struct fuse_inode_handle
{
664 static struct dentry
*fuse_get_dentry(struct super_block
*sb
,
665 struct fuse_inode_handle
*handle
)
667 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
669 struct dentry
*entry
;
672 if (handle
->nodeid
== 0)
675 inode
= ilookup5(sb
, handle
->nodeid
, fuse_inode_eq
, &handle
->nodeid
);
677 struct fuse_entry_out outarg
;
678 const struct qstr name
= QSTR_INIT(".", 1);
680 if (!fc
->export_support
)
683 err
= fuse_lookup_name(sb
, handle
->nodeid
, &name
, &outarg
,
685 if (err
&& err
!= -ENOENT
)
692 if (get_node_id(inode
) != handle
->nodeid
)
696 if (inode
->i_generation
!= handle
->generation
)
699 entry
= d_obtain_alias(inode
);
700 if (!IS_ERR(entry
) && get_node_id(inode
) != FUSE_ROOT_ID
)
701 fuse_invalidate_entry_cache(entry
);
711 static int fuse_encode_fh(struct inode
*inode
, u32
*fh
, int *max_len
,
712 struct inode
*parent
)
714 int len
= parent
? 6 : 3;
718 if (*max_len
< len
) {
720 return FILEID_INVALID
;
723 nodeid
= get_fuse_inode(inode
)->nodeid
;
724 generation
= inode
->i_generation
;
726 fh
[0] = (u32
)(nodeid
>> 32);
727 fh
[1] = (u32
)(nodeid
& 0xffffffff);
731 nodeid
= get_fuse_inode(parent
)->nodeid
;
732 generation
= parent
->i_generation
;
734 fh
[3] = (u32
)(nodeid
>> 32);
735 fh
[4] = (u32
)(nodeid
& 0xffffffff);
740 return parent
? 0x82 : 0x81;
743 static struct dentry
*fuse_fh_to_dentry(struct super_block
*sb
,
744 struct fid
*fid
, int fh_len
, int fh_type
)
746 struct fuse_inode_handle handle
;
748 if ((fh_type
!= 0x81 && fh_type
!= 0x82) || fh_len
< 3)
751 handle
.nodeid
= (u64
) fid
->raw
[0] << 32;
752 handle
.nodeid
|= (u64
) fid
->raw
[1];
753 handle
.generation
= fid
->raw
[2];
754 return fuse_get_dentry(sb
, &handle
);
757 static struct dentry
*fuse_fh_to_parent(struct super_block
*sb
,
758 struct fid
*fid
, int fh_len
, int fh_type
)
760 struct fuse_inode_handle parent
;
762 if (fh_type
!= 0x82 || fh_len
< 6)
765 parent
.nodeid
= (u64
) fid
->raw
[3] << 32;
766 parent
.nodeid
|= (u64
) fid
->raw
[4];
767 parent
.generation
= fid
->raw
[5];
768 return fuse_get_dentry(sb
, &parent
);
771 static struct dentry
*fuse_get_parent(struct dentry
*child
)
773 struct inode
*child_inode
= d_inode(child
);
774 struct fuse_conn
*fc
= get_fuse_conn(child_inode
);
776 struct dentry
*parent
;
777 struct fuse_entry_out outarg
;
778 const struct qstr name
= QSTR_INIT("..", 2);
781 if (!fc
->export_support
)
782 return ERR_PTR(-ESTALE
);
784 err
= fuse_lookup_name(child_inode
->i_sb
, get_node_id(child_inode
),
785 &name
, &outarg
, &inode
);
788 return ERR_PTR(-ESTALE
);
792 parent
= d_obtain_alias(inode
);
793 if (!IS_ERR(parent
) && get_node_id(inode
) != FUSE_ROOT_ID
)
794 fuse_invalidate_entry_cache(parent
);
799 static const struct export_operations fuse_export_operations
= {
800 .fh_to_dentry
= fuse_fh_to_dentry
,
801 .fh_to_parent
= fuse_fh_to_parent
,
802 .encode_fh
= fuse_encode_fh
,
803 .get_parent
= fuse_get_parent
,
806 static const struct super_operations fuse_super_operations
= {
807 .alloc_inode
= fuse_alloc_inode
,
808 .destroy_inode
= fuse_destroy_inode
,
809 .evict_inode
= fuse_evict_inode
,
810 .write_inode
= fuse_write_inode
,
811 .drop_inode
= generic_delete_inode
,
812 .remount_fs
= fuse_remount_fs
,
813 .put_super
= fuse_put_super
,
814 .umount_begin
= fuse_umount_begin
,
815 .statfs
= fuse_statfs
,
816 .show_options
= fuse_show_options
,
819 static void sanitize_global_limit(unsigned *limit
)
822 *limit
= ((totalram_pages
<< PAGE_SHIFT
) >> 13) /
823 sizeof(struct fuse_req
);
825 if (*limit
>= 1 << 16)
826 *limit
= (1 << 16) - 1;
829 static int set_global_limit(const char *val
, const struct kernel_param
*kp
)
833 rv
= param_set_uint(val
, kp
);
837 sanitize_global_limit((unsigned *)kp
->arg
);
842 static void process_init_limits(struct fuse_conn
*fc
, struct fuse_init_out
*arg
)
844 int cap_sys_admin
= capable(CAP_SYS_ADMIN
);
849 sanitize_global_limit(&max_user_bgreq
);
850 sanitize_global_limit(&max_user_congthresh
);
852 if (arg
->max_background
) {
853 fc
->max_background
= arg
->max_background
;
855 if (!cap_sys_admin
&& fc
->max_background
> max_user_bgreq
)
856 fc
->max_background
= max_user_bgreq
;
858 if (arg
->congestion_threshold
) {
859 fc
->congestion_threshold
= arg
->congestion_threshold
;
861 if (!cap_sys_admin
&&
862 fc
->congestion_threshold
> max_user_congthresh
)
863 fc
->congestion_threshold
= max_user_congthresh
;
867 static void process_init_reply(struct fuse_conn
*fc
, struct fuse_req
*req
)
869 struct fuse_init_out
*arg
= &req
->misc
.init_out
;
871 if (req
->out
.h
.error
|| arg
->major
!= FUSE_KERNEL_VERSION
)
874 unsigned long ra_pages
;
876 process_init_limits(fc
, arg
);
878 if (arg
->minor
>= 6) {
879 ra_pages
= arg
->max_readahead
/ PAGE_SIZE
;
880 if (arg
->flags
& FUSE_ASYNC_READ
)
882 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
884 if (arg
->minor
>= 17) {
885 if (!(arg
->flags
& FUSE_FLOCK_LOCKS
))
888 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
891 if (arg
->flags
& FUSE_ATOMIC_O_TRUNC
)
892 fc
->atomic_o_trunc
= 1;
893 if (arg
->minor
>= 9) {
894 /* LOOKUP has dependency on proto version */
895 if (arg
->flags
& FUSE_EXPORT_SUPPORT
)
896 fc
->export_support
= 1;
898 if (arg
->flags
& FUSE_BIG_WRITES
)
900 if (arg
->flags
& FUSE_DONT_MASK
)
902 if (arg
->flags
& FUSE_AUTO_INVAL_DATA
)
903 fc
->auto_inval_data
= 1;
904 if (arg
->flags
& FUSE_DO_READDIRPLUS
) {
905 fc
->do_readdirplus
= 1;
906 if (arg
->flags
& FUSE_READDIRPLUS_AUTO
)
907 fc
->readdirplus_auto
= 1;
909 if (arg
->flags
& FUSE_ASYNC_DIO
)
911 if (arg
->flags
& FUSE_WRITEBACK_CACHE
)
912 fc
->writeback_cache
= 1;
913 if (arg
->flags
& FUSE_PARALLEL_DIROPS
)
914 fc
->parallel_dirops
= 1;
915 if (arg
->flags
& FUSE_HANDLE_KILLPRIV
)
916 fc
->handle_killpriv
= 1;
917 if (arg
->time_gran
&& arg
->time_gran
<= 1000000000)
918 fc
->sb
->s_time_gran
= arg
->time_gran
;
919 if ((arg
->flags
& FUSE_POSIX_ACL
)) {
920 fc
->default_permissions
= 1;
922 fc
->sb
->s_xattr
= fuse_acl_xattr_handlers
;
924 if (arg
->flags
& FUSE_ABORT_ERROR
)
927 ra_pages
= fc
->max_read
/ PAGE_SIZE
;
932 fc
->sb
->s_bdi
->ra_pages
=
933 min(fc
->sb
->s_bdi
->ra_pages
, ra_pages
);
934 fc
->minor
= arg
->minor
;
935 fc
->max_write
= arg
->minor
< 5 ? 4096 : arg
->max_write
;
936 fc
->max_write
= max_t(unsigned, 4096, fc
->max_write
);
939 fuse_set_initialized(fc
);
940 wake_up_all(&fc
->blocked_waitq
);
943 static void fuse_send_init(struct fuse_conn
*fc
, struct fuse_req
*req
)
945 struct fuse_init_in
*arg
= &req
->misc
.init_in
;
947 arg
->major
= FUSE_KERNEL_VERSION
;
948 arg
->minor
= FUSE_KERNEL_MINOR_VERSION
;
949 arg
->max_readahead
= fc
->sb
->s_bdi
->ra_pages
* PAGE_SIZE
;
950 arg
->flags
|= FUSE_ASYNC_READ
| FUSE_POSIX_LOCKS
| FUSE_ATOMIC_O_TRUNC
|
951 FUSE_EXPORT_SUPPORT
| FUSE_BIG_WRITES
| FUSE_DONT_MASK
|
952 FUSE_SPLICE_WRITE
| FUSE_SPLICE_MOVE
| FUSE_SPLICE_READ
|
953 FUSE_FLOCK_LOCKS
| FUSE_HAS_IOCTL_DIR
| FUSE_AUTO_INVAL_DATA
|
954 FUSE_DO_READDIRPLUS
| FUSE_READDIRPLUS_AUTO
| FUSE_ASYNC_DIO
|
955 FUSE_WRITEBACK_CACHE
| FUSE_NO_OPEN_SUPPORT
|
956 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
)
987 * sb->s_bdi points to blkdev's bdi however we want to redirect
988 * it to our private bdi...
991 sb
->s_bdi
= &noop_backing_dev_info
;
993 err
= super_setup_bdi_name(sb
, "%u:%u%s", MAJOR(fc
->dev
),
994 MINOR(fc
->dev
), suffix
);
998 sb
->s_bdi
->ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
999 /* fuse does it's own writeback accounting */
1000 sb
->s_bdi
->capabilities
= BDI_CAP_NO_ACCT_WB
| BDI_CAP_STRICTLIMIT
;
1003 * For a single fuse filesystem use max 1% of dirty +
1004 * writeback threshold.
1006 * This gives about 1M of write buffer for memory maps on a
1007 * machine with 1G and 10% dirty_ratio, which should be more
1010 * Privileged users can raise it by writing to
1012 * /sys/class/bdi/<bdi>/max_ratio
1014 bdi_set_max_ratio(sb
->s_bdi
, 1);
1019 struct fuse_dev
*fuse_dev_alloc(struct fuse_conn
*fc
)
1021 struct fuse_dev
*fud
;
1023 fud
= kzalloc(sizeof(struct fuse_dev
), GFP_KERNEL
);
1025 fud
->fc
= fuse_conn_get(fc
);
1026 fuse_pqueue_init(&fud
->pq
);
1028 spin_lock(&fc
->lock
);
1029 list_add_tail(&fud
->entry
, &fc
->devices
);
1030 spin_unlock(&fc
->lock
);
1035 EXPORT_SYMBOL_GPL(fuse_dev_alloc
);
1037 void fuse_dev_free(struct fuse_dev
*fud
)
1039 struct fuse_conn
*fc
= fud
->fc
;
1042 spin_lock(&fc
->lock
);
1043 list_del(&fud
->entry
);
1044 spin_unlock(&fc
->lock
);
1050 EXPORT_SYMBOL_GPL(fuse_dev_free
);
1052 static int fuse_fill_super(struct super_block
*sb
, void *data
, int silent
)
1054 struct fuse_dev
*fud
;
1055 struct fuse_conn
*fc
;
1057 struct fuse_mount_data d
;
1059 struct dentry
*root_dentry
;
1060 struct fuse_req
*init_req
;
1062 int is_bdev
= sb
->s_bdev
!= NULL
;
1065 if (sb
->s_flags
& SB_MANDLOCK
)
1068 sb
->s_flags
&= ~(SB_NOSEC
| SB_I_VERSION
);
1070 if (!parse_fuse_opt(data
, &d
, is_bdev
, sb
->s_user_ns
))
1076 if (!sb_set_blocksize(sb
, d
.blksize
))
1080 sb
->s_blocksize
= PAGE_SIZE
;
1081 sb
->s_blocksize_bits
= PAGE_SHIFT
;
1083 sb
->s_magic
= FUSE_SUPER_MAGIC
;
1084 sb
->s_op
= &fuse_super_operations
;
1085 sb
->s_xattr
= fuse_xattr_handlers
;
1086 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1087 sb
->s_time_gran
= 1;
1088 sb
->s_export_op
= &fuse_export_operations
;
1089 sb
->s_iflags
|= SB_I_IMA_UNVERIFIABLE_SIGNATURE
;
1090 if (sb
->s_user_ns
!= &init_user_ns
)
1091 sb
->s_iflags
|= SB_I_UNTRUSTED_MOUNTER
;
1099 * Require mount to happen from the same user namespace which
1100 * opened /dev/fuse to prevent potential attacks.
1102 if (file
->f_op
!= &fuse_dev_operations
||
1103 file
->f_cred
->user_ns
!= sb
->s_user_ns
)
1107 * If we are not in the initial user namespace posix
1108 * acls must be translated.
1110 if (sb
->s_user_ns
!= &init_user_ns
)
1111 sb
->s_xattr
= fuse_no_acl_xattr_handlers
;
1113 fc
= kmalloc(sizeof(*fc
), GFP_KERNEL
);
1118 fuse_conn_init(fc
, sb
->s_user_ns
);
1119 fc
->release
= fuse_free_conn
;
1121 fud
= fuse_dev_alloc(fc
);
1125 fc
->dev
= sb
->s_dev
;
1127 err
= fuse_bdi_init(fc
, sb
);
1131 /* Handle umasking inside the fuse code */
1132 if (sb
->s_flags
& SB_POSIXACL
)
1134 sb
->s_flags
|= SB_POSIXACL
;
1136 fc
->default_permissions
= d
.default_permissions
;
1137 fc
->allow_other
= d
.allow_other
;
1138 fc
->user_id
= d
.user_id
;
1139 fc
->group_id
= d
.group_id
;
1140 fc
->max_read
= max_t(unsigned, 4096, d
.max_read
);
1142 /* Used by get_root_inode() */
1146 root
= fuse_get_root_inode(sb
, d
.rootmode
);
1147 sb
->s_d_op
= &fuse_root_dentry_operations
;
1148 root_dentry
= d_make_root(root
);
1151 /* Root dentry doesn't have .d_revalidate */
1152 sb
->s_d_op
= &fuse_dentry_operations
;
1154 init_req
= fuse_request_alloc(0);
1157 __set_bit(FR_BACKGROUND
, &init_req
->flags
);
1160 fc
->destroy_req
= fuse_request_alloc(0);
1161 if (!fc
->destroy_req
)
1162 goto err_free_init_req
;
1165 mutex_lock(&fuse_mutex
);
1167 if (file
->private_data
)
1170 err
= fuse_ctl_add_conn(fc
);
1174 list_add_tail(&fc
->entry
, &fuse_conn_list
);
1175 sb
->s_root
= root_dentry
;
1176 file
->private_data
= fud
;
1177 mutex_unlock(&fuse_mutex
);
1179 * atomic_dec_and_test() in fput() provides the necessary
1180 * memory barrier for file->private_data to be visible on all
1185 fuse_send_init(fc
, init_req
);
1190 mutex_unlock(&fuse_mutex
);
1192 fuse_request_free(init_req
);
1199 sb
->s_fs_info
= NULL
;
1206 static struct dentry
*fuse_mount(struct file_system_type
*fs_type
,
1207 int flags
, const char *dev_name
,
1210 return mount_nodev(fs_type
, flags
, raw_data
, fuse_fill_super
);
1213 static void fuse_kill_sb_anon(struct super_block
*sb
)
1215 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1218 down_write(&fc
->killsb
);
1220 up_write(&fc
->killsb
);
1223 kill_anon_super(sb
);
1226 static struct file_system_type fuse_fs_type
= {
1227 .owner
= THIS_MODULE
,
1229 .fs_flags
= FS_HAS_SUBTYPE
| FS_USERNS_MOUNT
,
1230 .mount
= fuse_mount
,
1231 .kill_sb
= fuse_kill_sb_anon
,
1233 MODULE_ALIAS_FS("fuse");
1236 static struct dentry
*fuse_mount_blk(struct file_system_type
*fs_type
,
1237 int flags
, const char *dev_name
,
1240 return mount_bdev(fs_type
, flags
, dev_name
, raw_data
, fuse_fill_super
);
1243 static void fuse_kill_sb_blk(struct super_block
*sb
)
1245 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1248 down_write(&fc
->killsb
);
1250 up_write(&fc
->killsb
);
1253 kill_block_super(sb
);
1256 static struct file_system_type fuseblk_fs_type
= {
1257 .owner
= THIS_MODULE
,
1259 .mount
= fuse_mount_blk
,
1260 .kill_sb
= fuse_kill_sb_blk
,
1261 .fs_flags
= FS_REQUIRES_DEV
| FS_HAS_SUBTYPE
,
1263 MODULE_ALIAS_FS("fuseblk");
1265 static inline int register_fuseblk(void)
1267 return register_filesystem(&fuseblk_fs_type
);
1270 static inline void unregister_fuseblk(void)
1272 unregister_filesystem(&fuseblk_fs_type
);
1275 static inline int register_fuseblk(void)
1280 static inline void unregister_fuseblk(void)
1285 static void fuse_inode_init_once(void *foo
)
1287 struct inode
*inode
= foo
;
1289 inode_init_once(inode
);
1292 static int __init
fuse_fs_init(void)
1296 fuse_inode_cachep
= kmem_cache_create("fuse_inode",
1297 sizeof(struct fuse_inode
), 0,
1298 SLAB_HWCACHE_ALIGN
|SLAB_ACCOUNT
|SLAB_RECLAIM_ACCOUNT
,
1299 fuse_inode_init_once
);
1301 if (!fuse_inode_cachep
)
1304 err
= register_fuseblk();
1308 err
= register_filesystem(&fuse_fs_type
);
1315 unregister_fuseblk();
1317 kmem_cache_destroy(fuse_inode_cachep
);
1322 static void fuse_fs_cleanup(void)
1324 unregister_filesystem(&fuse_fs_type
);
1325 unregister_fuseblk();
1328 * Make sure all delayed rcu free inodes are flushed before we
1332 kmem_cache_destroy(fuse_inode_cachep
);
1335 static struct kobject
*fuse_kobj
;
1337 static int fuse_sysfs_init(void)
1341 fuse_kobj
= kobject_create_and_add("fuse", fs_kobj
);
1347 err
= sysfs_create_mount_point(fuse_kobj
, "connections");
1349 goto out_fuse_unregister
;
1353 out_fuse_unregister
:
1354 kobject_put(fuse_kobj
);
1359 static void fuse_sysfs_cleanup(void)
1361 sysfs_remove_mount_point(fuse_kobj
, "connections");
1362 kobject_put(fuse_kobj
);
1365 static int __init
fuse_init(void)
1369 printk(KERN_INFO
"fuse init (API version %i.%i)\n",
1370 FUSE_KERNEL_VERSION
, FUSE_KERNEL_MINOR_VERSION
);
1372 INIT_LIST_HEAD(&fuse_conn_list
);
1373 res
= fuse_fs_init();
1377 res
= fuse_dev_init();
1379 goto err_fs_cleanup
;
1381 res
= fuse_sysfs_init();
1383 goto err_dev_cleanup
;
1385 res
= fuse_ctl_init();
1387 goto err_sysfs_cleanup
;
1389 sanitize_global_limit(&max_user_bgreq
);
1390 sanitize_global_limit(&max_user_congthresh
);
1395 fuse_sysfs_cleanup();
1404 static void __exit
fuse_exit(void)
1406 printk(KERN_DEBUG
"fuse exit\n");
1409 fuse_sysfs_cleanup();
1414 module_init(fuse_init
);
1415 module_exit(fuse_exit
);