2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2006 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/parser.h>
18 #include <linux/statfs.h>
19 #include <linux/random.h>
20 #include <linux/sched.h>
22 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
23 MODULE_DESCRIPTION("Filesystem in Userspace");
24 MODULE_LICENSE("GPL");
26 static struct kmem_cache *fuse_inode_cachep;
27 struct list_head fuse_conn_list;
28 DEFINE_MUTEX(fuse_mutex);
30 #define FUSE_SUPER_MAGIC 0x65735546
32 struct fuse_mount_data {
37 unsigned fd_present : 1;
38 unsigned rootmode_present : 1;
39 unsigned user_id_present : 1;
40 unsigned group_id_present : 1;
46 static struct inode *fuse_alloc_inode(struct super_block *sb)
49 struct fuse_inode *fi;
51 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
55 fi = get_fuse_inode(inode);
59 fi->forget_req = fuse_request_alloc();
60 if (!fi->forget_req) {
61 kmem_cache_free(fuse_inode_cachep, inode);
68 static void fuse_destroy_inode(struct inode *inode)
70 struct fuse_inode *fi = get_fuse_inode(inode);
72 fuse_request_free(fi->forget_req);
73 kmem_cache_free(fuse_inode_cachep, inode);
76 static void fuse_read_inode(struct inode *inode)
81 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
82 unsigned long nodeid, u64 nlookup)
84 struct fuse_forget_in *inarg = &req->misc.forget_in;
85 inarg->nlookup = nlookup;
86 req->in.h.opcode = FUSE_FORGET;
87 req->in.h.nodeid = nodeid;
89 req->in.args[0].size = sizeof(struct fuse_forget_in);
90 req->in.args[0].value = inarg;
91 request_send_noreply(fc, req);
94 static void fuse_clear_inode(struct inode *inode)
96 if (inode->i_sb->s_flags & MS_ACTIVE) {
97 struct fuse_conn *fc = get_fuse_conn(inode);
98 struct fuse_inode *fi = get_fuse_inode(inode);
99 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
100 fi->forget_req = NULL;
104 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
106 if (*flags & MS_MANDLOCK)
112 static void fuse_truncate(struct address_space *mapping, loff_t offset)
114 /* See vmtruncate() */
115 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
116 truncate_inode_pages(mapping, offset);
117 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
121 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
122 u64 attr_valid, u64 attr_version)
124 struct fuse_conn *fc = get_fuse_conn(inode);
125 struct fuse_inode *fi = get_fuse_inode(inode);
128 spin_lock(&fc->lock);
129 if (attr_version != 0 && fi->attr_version > attr_version) {
130 spin_unlock(&fc->lock);
133 fi->attr_version = ++fc->attr_version;
134 fi->i_time = attr_valid;
136 inode->i_ino = attr->ino;
137 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
138 inode->i_nlink = attr->nlink;
139 inode->i_uid = attr->uid;
140 inode->i_gid = attr->gid;
141 inode->i_blocks = attr->blocks;
142 inode->i_atime.tv_sec = attr->atime;
143 inode->i_atime.tv_nsec = attr->atimensec;
144 inode->i_mtime.tv_sec = attr->mtime;
145 inode->i_mtime.tv_nsec = attr->mtimensec;
146 inode->i_ctime.tv_sec = attr->ctime;
147 inode->i_ctime.tv_nsec = attr->ctimensec;
150 * Don't set the sticky bit in i_mode, unless we want the VFS
151 * to check permissions. This prevents failures due to the
152 * check in may_delete().
154 fi->orig_i_mode = inode->i_mode;
155 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
156 inode->i_mode &= ~S_ISVTX;
158 oldsize = inode->i_size;
159 i_size_write(inode, attr->size);
160 spin_unlock(&fc->lock);
162 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
163 if (attr->size < oldsize)
164 fuse_truncate(inode->i_mapping, attr->size);
165 invalidate_inode_pages2(inode->i_mapping);
169 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
171 inode->i_mode = attr->mode & S_IFMT;
172 inode->i_size = attr->size;
173 if (S_ISREG(inode->i_mode)) {
174 fuse_init_common(inode);
175 fuse_init_file_inode(inode);
176 } else if (S_ISDIR(inode->i_mode))
177 fuse_init_dir(inode);
178 else if (S_ISLNK(inode->i_mode))
179 fuse_init_symlink(inode);
180 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
181 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
182 fuse_init_common(inode);
183 init_special_inode(inode, inode->i_mode,
184 new_decode_dev(attr->rdev));
189 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
191 unsigned long nodeid = *(unsigned long *) _nodeidp;
192 if (get_node_id(inode) == nodeid)
198 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
200 unsigned long nodeid = *(unsigned long *) _nodeidp;
201 get_fuse_inode(inode)->nodeid = nodeid;
205 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
206 int generation, struct fuse_attr *attr,
207 u64 attr_valid, u64 attr_version)
210 struct fuse_inode *fi;
211 struct fuse_conn *fc = get_fuse_conn_super(sb);
214 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
218 if ((inode->i_state & I_NEW)) {
219 inode->i_flags |= S_NOATIME|S_NOCMTIME;
220 inode->i_generation = generation;
221 inode->i_data.backing_dev_info = &fc->bdi;
222 fuse_init_inode(inode, attr);
223 unlock_new_inode(inode);
224 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
225 /* Inode has changed type, any I/O on the old should fail */
226 make_bad_inode(inode);
231 fi = get_fuse_inode(inode);
232 spin_lock(&fc->lock);
234 spin_unlock(&fc->lock);
235 fuse_change_attributes(inode, attr, attr_valid, attr_version);
240 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
242 if (flags & MNT_FORCE)
243 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
246 static void fuse_send_destroy(struct fuse_conn *fc)
248 struct fuse_req *req = fc->destroy_req;
249 if (req && fc->conn_init) {
250 fc->destroy_req = NULL;
251 req->in.h.opcode = FUSE_DESTROY;
253 request_send(fc, req);
254 fuse_put_request(fc, req);
258 static void fuse_put_super(struct super_block *sb)
260 struct fuse_conn *fc = get_fuse_conn_super(sb);
262 fuse_send_destroy(fc);
263 spin_lock(&fc->lock);
266 spin_unlock(&fc->lock);
267 /* Flush all readers on this fs */
268 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
269 wake_up_all(&fc->waitq);
270 wake_up_all(&fc->blocked_waitq);
271 wake_up_all(&fc->reserved_req_waitq);
272 mutex_lock(&fuse_mutex);
273 list_del(&fc->entry);
274 fuse_ctl_remove_conn(fc);
275 mutex_unlock(&fuse_mutex);
279 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
281 stbuf->f_type = FUSE_SUPER_MAGIC;
282 stbuf->f_bsize = attr->bsize;
283 stbuf->f_frsize = attr->frsize;
284 stbuf->f_blocks = attr->blocks;
285 stbuf->f_bfree = attr->bfree;
286 stbuf->f_bavail = attr->bavail;
287 stbuf->f_files = attr->files;
288 stbuf->f_ffree = attr->ffree;
289 stbuf->f_namelen = attr->namelen;
290 /* fsid is left zero */
293 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
295 struct super_block *sb = dentry->d_sb;
296 struct fuse_conn *fc = get_fuse_conn_super(sb);
297 struct fuse_req *req;
298 struct fuse_statfs_out outarg;
301 if (!fuse_allow_task(fc, current)) {
302 buf->f_type = FUSE_SUPER_MAGIC;
306 req = fuse_get_req(fc);
310 memset(&outarg, 0, sizeof(outarg));
312 req->in.h.opcode = FUSE_STATFS;
313 req->in.h.nodeid = get_node_id(dentry->d_inode);
314 req->out.numargs = 1;
315 req->out.args[0].size =
316 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
317 req->out.args[0].value = &outarg;
318 request_send(fc, req);
319 err = req->out.h.error;
321 convert_fuse_statfs(buf, &outarg.st);
322 fuse_put_request(fc, req);
331 OPT_DEFAULT_PERMISSIONS,
338 static match_table_t tokens = {
340 {OPT_ROOTMODE, "rootmode=%o"},
341 {OPT_USER_ID, "user_id=%u"},
342 {OPT_GROUP_ID, "group_id=%u"},
343 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
344 {OPT_ALLOW_OTHER, "allow_other"},
345 {OPT_MAX_READ, "max_read=%u"},
346 {OPT_BLKSIZE, "blksize=%u"},
350 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
353 memset(d, 0, sizeof(struct fuse_mount_data));
357 while ((p = strsep(&opt, ",")) != NULL) {
360 substring_t args[MAX_OPT_ARGS];
364 token = match_token(p, tokens, args);
367 if (match_int(&args[0], &value))
374 if (match_octal(&args[0], &value))
376 if (!fuse_valid_type(value))
379 d->rootmode_present = 1;
383 if (match_int(&args[0], &value))
386 d->user_id_present = 1;
390 if (match_int(&args[0], &value))
393 d->group_id_present = 1;
396 case OPT_DEFAULT_PERMISSIONS:
397 d->flags |= FUSE_DEFAULT_PERMISSIONS;
400 case OPT_ALLOW_OTHER:
401 d->flags |= FUSE_ALLOW_OTHER;
405 if (match_int(&args[0], &value))
411 if (!is_bdev || match_int(&args[0], &value))
421 if (!d->fd_present || !d->rootmode_present ||
422 !d->user_id_present || !d->group_id_present)
428 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
430 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
432 seq_printf(m, ",user_id=%u", fc->user_id);
433 seq_printf(m, ",group_id=%u", fc->group_id);
434 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
435 seq_puts(m, ",default_permissions");
436 if (fc->flags & FUSE_ALLOW_OTHER)
437 seq_puts(m, ",allow_other");
438 if (fc->max_read != ~0)
439 seq_printf(m, ",max_read=%u", fc->max_read);
443 static struct fuse_conn *new_conn(void)
445 struct fuse_conn *fc;
448 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
450 spin_lock_init(&fc->lock);
451 mutex_init(&fc->inst_mutex);
452 atomic_set(&fc->count, 1);
453 init_waitqueue_head(&fc->waitq);
454 init_waitqueue_head(&fc->blocked_waitq);
455 init_waitqueue_head(&fc->reserved_req_waitq);
456 INIT_LIST_HEAD(&fc->pending);
457 INIT_LIST_HEAD(&fc->processing);
458 INIT_LIST_HEAD(&fc->io);
459 INIT_LIST_HEAD(&fc->interrupts);
460 atomic_set(&fc->num_waiting, 0);
461 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
462 fc->bdi.unplug_io_fn = default_unplug_io_fn;
463 err = bdi_init(&fc->bdi);
471 fc->attr_version = 1;
472 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
478 void fuse_conn_put(struct fuse_conn *fc)
480 if (atomic_dec_and_test(&fc->count)) {
482 fuse_request_free(fc->destroy_req);
483 mutex_destroy(&fc->inst_mutex);
484 bdi_destroy(&fc->bdi);
489 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
491 atomic_inc(&fc->count);
495 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
497 struct fuse_attr attr;
498 memset(&attr, 0, sizeof(attr));
501 attr.ino = FUSE_ROOT_ID;
503 return fuse_iget(sb, 1, 0, &attr, 0, 0);
506 static const struct super_operations fuse_super_operations = {
507 .alloc_inode = fuse_alloc_inode,
508 .destroy_inode = fuse_destroy_inode,
509 .read_inode = fuse_read_inode,
510 .clear_inode = fuse_clear_inode,
511 .drop_inode = generic_delete_inode,
512 .remount_fs = fuse_remount_fs,
513 .put_super = fuse_put_super,
514 .umount_begin = fuse_umount_begin,
515 .statfs = fuse_statfs,
516 .show_options = fuse_show_options,
519 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
521 struct fuse_init_out *arg = &req->misc.init_out;
523 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
526 unsigned long ra_pages;
528 if (arg->minor >= 6) {
529 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
530 if (arg->flags & FUSE_ASYNC_READ)
532 if (!(arg->flags & FUSE_POSIX_LOCKS))
535 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
539 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
540 fc->minor = arg->minor;
541 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
544 fuse_put_request(fc, req);
546 wake_up_all(&fc->blocked_waitq);
549 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
551 struct fuse_init_in *arg = &req->misc.init_in;
553 arg->major = FUSE_KERNEL_VERSION;
554 arg->minor = FUSE_KERNEL_MINOR_VERSION;
555 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
556 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS;
557 req->in.h.opcode = FUSE_INIT;
559 req->in.args[0].size = sizeof(*arg);
560 req->in.args[0].value = arg;
561 req->out.numargs = 1;
562 /* Variable length arguement used for backward compatibility
563 with interface version < 7.5. Rest of init_out is zeroed
564 by do_get_request(), so a short reply is not a problem */
566 req->out.args[0].size = sizeof(struct fuse_init_out);
567 req->out.args[0].value = &req->misc.init_out;
568 req->end = process_init_reply;
569 request_send_background(fc, req);
572 static u64 conn_id(void)
578 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
580 struct fuse_conn *fc;
582 struct fuse_mount_data d;
584 struct dentry *root_dentry;
585 struct fuse_req *init_req;
587 int is_bdev = sb->s_bdev != NULL;
589 if (sb->s_flags & MS_MANDLOCK)
592 if (!parse_fuse_opt((char *) data, &d, is_bdev))
597 if (!sb_set_blocksize(sb, d.blksize))
601 sb->s_blocksize = PAGE_CACHE_SIZE;
602 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
604 sb->s_magic = FUSE_SUPER_MAGIC;
605 sb->s_op = &fuse_super_operations;
606 sb->s_maxbytes = MAX_LFS_FILESIZE;
612 if (file->f_op != &fuse_dev_operations)
620 fc->user_id = d.user_id;
621 fc->group_id = d.group_id;
622 fc->max_read = d.max_read;
624 /* Used by get_root_inode() */
628 root = get_root_inode(sb, d.rootmode);
632 root_dentry = d_alloc_root(root);
638 init_req = fuse_request_alloc();
643 fc->destroy_req = fuse_request_alloc();
644 if (!fc->destroy_req)
648 mutex_lock(&fuse_mutex);
650 if (file->private_data)
654 err = fuse_ctl_add_conn(fc);
658 list_add_tail(&fc->entry, &fuse_conn_list);
659 sb->s_root = root_dentry;
661 file->private_data = fuse_conn_get(fc);
662 mutex_unlock(&fuse_mutex);
664 * atomic_dec_and_test() in fput() provides the necessary
665 * memory barrier for file->private_data to be visible on all
670 fuse_send_init(fc, init_req);
675 mutex_unlock(&fuse_mutex);
676 fuse_request_free(init_req);
685 static int fuse_get_sb(struct file_system_type *fs_type,
686 int flags, const char *dev_name,
687 void *raw_data, struct vfsmount *mnt)
689 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
692 static struct file_system_type fuse_fs_type = {
693 .owner = THIS_MODULE,
695 .fs_flags = FS_HAS_SUBTYPE,
696 .get_sb = fuse_get_sb,
697 .kill_sb = kill_anon_super,
701 static int fuse_get_sb_blk(struct file_system_type *fs_type,
702 int flags, const char *dev_name,
703 void *raw_data, struct vfsmount *mnt)
705 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
709 static struct file_system_type fuseblk_fs_type = {
710 .owner = THIS_MODULE,
712 .get_sb = fuse_get_sb_blk,
713 .kill_sb = kill_block_super,
714 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
717 static inline int register_fuseblk(void)
719 return register_filesystem(&fuseblk_fs_type);
722 static inline void unregister_fuseblk(void)
724 unregister_filesystem(&fuseblk_fs_type);
727 static inline int register_fuseblk(void)
732 static inline void unregister_fuseblk(void)
737 static decl_subsys(fuse, NULL, NULL);
738 static decl_subsys(connections, NULL, NULL);
740 static void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
742 struct inode * inode = foo;
744 inode_init_once(inode);
747 static int __init fuse_fs_init(void)
751 err = register_filesystem(&fuse_fs_type);
755 err = register_fuseblk();
759 fuse_inode_cachep = kmem_cache_create("fuse_inode",
760 sizeof(struct fuse_inode),
761 0, SLAB_HWCACHE_ALIGN,
762 fuse_inode_init_once);
764 if (!fuse_inode_cachep)
770 unregister_fuseblk();
772 unregister_filesystem(&fuse_fs_type);
777 static void fuse_fs_cleanup(void)
779 unregister_filesystem(&fuse_fs_type);
780 unregister_fuseblk();
781 kmem_cache_destroy(fuse_inode_cachep);
784 static int fuse_sysfs_init(void)
788 kobj_set_kset_s(&fuse_subsys, fs_subsys);
789 err = subsystem_register(&fuse_subsys);
793 kobj_set_kset_s(&connections_subsys, fuse_subsys);
794 err = subsystem_register(&connections_subsys);
796 goto out_fuse_unregister;
801 subsystem_unregister(&fuse_subsys);
806 static void fuse_sysfs_cleanup(void)
808 subsystem_unregister(&connections_subsys);
809 subsystem_unregister(&fuse_subsys);
812 static int __init fuse_init(void)
816 printk("fuse init (API version %i.%i)\n",
817 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
819 INIT_LIST_HEAD(&fuse_conn_list);
820 res = fuse_fs_init();
824 res = fuse_dev_init();
828 res = fuse_sysfs_init();
830 goto err_dev_cleanup;
832 res = fuse_ctl_init();
834 goto err_sysfs_cleanup;
839 fuse_sysfs_cleanup();
848 static void __exit fuse_exit(void)
850 printk(KERN_DEBUG "fuse exit\n");
853 fuse_sysfs_cleanup();
858 module_init(fuse_init);
859 module_exit(fuse_exit);