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 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr)
114 struct fuse_conn *fc = get_fuse_conn(inode);
115 if (S_ISREG(inode->i_mode) && i_size_read(inode) != attr->size)
116 invalidate_mapping_pages(inode->i_mapping, 0, -1);
118 inode->i_ino = attr->ino;
119 inode->i_mode = (inode->i_mode & S_IFMT) + (attr->mode & 07777);
120 inode->i_nlink = attr->nlink;
121 inode->i_uid = attr->uid;
122 inode->i_gid = attr->gid;
123 spin_lock(&fc->lock);
124 i_size_write(inode, attr->size);
125 spin_unlock(&fc->lock);
126 inode->i_blocks = attr->blocks;
127 inode->i_atime.tv_sec = attr->atime;
128 inode->i_atime.tv_nsec = attr->atimensec;
129 inode->i_mtime.tv_sec = attr->mtime;
130 inode->i_mtime.tv_nsec = attr->mtimensec;
131 inode->i_ctime.tv_sec = attr->ctime;
132 inode->i_ctime.tv_nsec = attr->ctimensec;
135 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
137 inode->i_mode = attr->mode & S_IFMT;
138 inode->i_size = attr->size;
139 if (S_ISREG(inode->i_mode)) {
140 fuse_init_common(inode);
141 fuse_init_file_inode(inode);
142 } else if (S_ISDIR(inode->i_mode))
143 fuse_init_dir(inode);
144 else if (S_ISLNK(inode->i_mode))
145 fuse_init_symlink(inode);
146 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
147 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
148 fuse_init_common(inode);
149 init_special_inode(inode, inode->i_mode,
150 new_decode_dev(attr->rdev));
155 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
157 unsigned long nodeid = *(unsigned long *) _nodeidp;
158 if (get_node_id(inode) == nodeid)
164 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
166 unsigned long nodeid = *(unsigned long *) _nodeidp;
167 get_fuse_inode(inode)->nodeid = nodeid;
171 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
172 int generation, struct fuse_attr *attr)
175 struct fuse_inode *fi;
176 struct fuse_conn *fc = get_fuse_conn_super(sb);
179 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
183 if ((inode->i_state & I_NEW)) {
184 inode->i_flags |= S_NOATIME|S_NOCMTIME;
185 inode->i_generation = generation;
186 inode->i_data.backing_dev_info = &fc->bdi;
187 fuse_init_inode(inode, attr);
188 unlock_new_inode(inode);
189 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
190 /* Inode has changed type, any I/O on the old should fail */
191 make_bad_inode(inode);
196 fi = get_fuse_inode(inode);
197 spin_lock(&fc->lock);
199 spin_unlock(&fc->lock);
200 fuse_change_attributes(inode, attr);
204 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
206 if (flags & MNT_FORCE)
207 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
210 static void fuse_send_destroy(struct fuse_conn *fc)
212 struct fuse_req *req = fc->destroy_req;
213 if (req && fc->conn_init) {
214 fc->destroy_req = NULL;
215 req->in.h.opcode = FUSE_DESTROY;
217 request_send(fc, req);
218 fuse_put_request(fc, req);
222 static void fuse_put_super(struct super_block *sb)
224 struct fuse_conn *fc = get_fuse_conn_super(sb);
226 fuse_send_destroy(fc);
227 spin_lock(&fc->lock);
230 spin_unlock(&fc->lock);
231 /* Flush all readers on this fs */
232 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
233 wake_up_all(&fc->waitq);
234 wake_up_all(&fc->blocked_waitq);
235 wake_up_all(&fc->reserved_req_waitq);
236 mutex_lock(&fuse_mutex);
237 list_del(&fc->entry);
238 fuse_ctl_remove_conn(fc);
239 mutex_unlock(&fuse_mutex);
243 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
245 stbuf->f_type = FUSE_SUPER_MAGIC;
246 stbuf->f_bsize = attr->bsize;
247 stbuf->f_frsize = attr->frsize;
248 stbuf->f_blocks = attr->blocks;
249 stbuf->f_bfree = attr->bfree;
250 stbuf->f_bavail = attr->bavail;
251 stbuf->f_files = attr->files;
252 stbuf->f_ffree = attr->ffree;
253 stbuf->f_namelen = attr->namelen;
254 /* fsid is left zero */
257 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
259 struct super_block *sb = dentry->d_sb;
260 struct fuse_conn *fc = get_fuse_conn_super(sb);
261 struct fuse_req *req;
262 struct fuse_statfs_out outarg;
265 req = fuse_get_req(fc);
269 memset(&outarg, 0, sizeof(outarg));
271 req->in.h.opcode = FUSE_STATFS;
272 req->in.h.nodeid = get_node_id(dentry->d_inode);
273 req->out.numargs = 1;
274 req->out.args[0].size =
275 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
276 req->out.args[0].value = &outarg;
277 request_send(fc, req);
278 err = req->out.h.error;
280 convert_fuse_statfs(buf, &outarg.st);
281 fuse_put_request(fc, req);
290 OPT_DEFAULT_PERMISSIONS,
297 static match_table_t tokens = {
299 {OPT_ROOTMODE, "rootmode=%o"},
300 {OPT_USER_ID, "user_id=%u"},
301 {OPT_GROUP_ID, "group_id=%u"},
302 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
303 {OPT_ALLOW_OTHER, "allow_other"},
304 {OPT_MAX_READ, "max_read=%u"},
305 {OPT_BLKSIZE, "blksize=%u"},
309 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
312 memset(d, 0, sizeof(struct fuse_mount_data));
316 while ((p = strsep(&opt, ",")) != NULL) {
319 substring_t args[MAX_OPT_ARGS];
323 token = match_token(p, tokens, args);
326 if (match_int(&args[0], &value))
333 if (match_octal(&args[0], &value))
335 if (!fuse_valid_type(value))
338 d->rootmode_present = 1;
342 if (match_int(&args[0], &value))
345 d->user_id_present = 1;
349 if (match_int(&args[0], &value))
352 d->group_id_present = 1;
355 case OPT_DEFAULT_PERMISSIONS:
356 d->flags |= FUSE_DEFAULT_PERMISSIONS;
359 case OPT_ALLOW_OTHER:
360 d->flags |= FUSE_ALLOW_OTHER;
364 if (match_int(&args[0], &value))
370 if (!is_bdev || match_int(&args[0], &value))
380 if (!d->fd_present || !d->rootmode_present ||
381 !d->user_id_present || !d->group_id_present)
387 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
389 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
391 seq_printf(m, ",user_id=%u", fc->user_id);
392 seq_printf(m, ",group_id=%u", fc->group_id);
393 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
394 seq_puts(m, ",default_permissions");
395 if (fc->flags & FUSE_ALLOW_OTHER)
396 seq_puts(m, ",allow_other");
397 if (fc->max_read != ~0)
398 seq_printf(m, ",max_read=%u", fc->max_read);
402 static struct fuse_conn *new_conn(void)
404 struct fuse_conn *fc;
407 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
409 spin_lock_init(&fc->lock);
410 mutex_init(&fc->inst_mutex);
411 atomic_set(&fc->count, 1);
412 init_waitqueue_head(&fc->waitq);
413 init_waitqueue_head(&fc->blocked_waitq);
414 init_waitqueue_head(&fc->reserved_req_waitq);
415 INIT_LIST_HEAD(&fc->pending);
416 INIT_LIST_HEAD(&fc->processing);
417 INIT_LIST_HEAD(&fc->io);
418 INIT_LIST_HEAD(&fc->interrupts);
419 atomic_set(&fc->num_waiting, 0);
420 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
421 fc->bdi.unplug_io_fn = default_unplug_io_fn;
422 err = bdi_init(&fc->bdi);
430 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
436 void fuse_conn_put(struct fuse_conn *fc)
438 if (atomic_dec_and_test(&fc->count)) {
440 fuse_request_free(fc->destroy_req);
441 mutex_destroy(&fc->inst_mutex);
442 bdi_destroy(&fc->bdi);
447 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
449 atomic_inc(&fc->count);
453 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
455 struct fuse_attr attr;
456 memset(&attr, 0, sizeof(attr));
459 attr.ino = FUSE_ROOT_ID;
460 return fuse_iget(sb, 1, 0, &attr);
463 static const struct super_operations fuse_super_operations = {
464 .alloc_inode = fuse_alloc_inode,
465 .destroy_inode = fuse_destroy_inode,
466 .read_inode = fuse_read_inode,
467 .clear_inode = fuse_clear_inode,
468 .drop_inode = generic_delete_inode,
469 .remount_fs = fuse_remount_fs,
470 .put_super = fuse_put_super,
471 .umount_begin = fuse_umount_begin,
472 .statfs = fuse_statfs,
473 .show_options = fuse_show_options,
476 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
478 struct fuse_init_out *arg = &req->misc.init_out;
480 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
483 unsigned long ra_pages;
485 if (arg->minor >= 6) {
486 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
487 if (arg->flags & FUSE_ASYNC_READ)
489 if (!(arg->flags & FUSE_POSIX_LOCKS))
492 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
496 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
497 fc->minor = arg->minor;
498 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
501 fuse_put_request(fc, req);
503 wake_up_all(&fc->blocked_waitq);
506 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
508 struct fuse_init_in *arg = &req->misc.init_in;
510 arg->major = FUSE_KERNEL_VERSION;
511 arg->minor = FUSE_KERNEL_MINOR_VERSION;
512 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
513 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS;
514 req->in.h.opcode = FUSE_INIT;
516 req->in.args[0].size = sizeof(*arg);
517 req->in.args[0].value = arg;
518 req->out.numargs = 1;
519 /* Variable length arguement used for backward compatibility
520 with interface version < 7.5. Rest of init_out is zeroed
521 by do_get_request(), so a short reply is not a problem */
523 req->out.args[0].size = sizeof(struct fuse_init_out);
524 req->out.args[0].value = &req->misc.init_out;
525 req->end = process_init_reply;
526 request_send_background(fc, req);
529 static u64 conn_id(void)
535 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
537 struct fuse_conn *fc;
539 struct fuse_mount_data d;
541 struct dentry *root_dentry;
542 struct fuse_req *init_req;
544 int is_bdev = sb->s_bdev != NULL;
546 if (sb->s_flags & MS_MANDLOCK)
549 if (!parse_fuse_opt((char *) data, &d, is_bdev))
554 if (!sb_set_blocksize(sb, d.blksize))
558 sb->s_blocksize = PAGE_CACHE_SIZE;
559 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
561 sb->s_magic = FUSE_SUPER_MAGIC;
562 sb->s_op = &fuse_super_operations;
563 sb->s_maxbytes = MAX_LFS_FILESIZE;
569 if (file->f_op != &fuse_dev_operations)
577 fc->user_id = d.user_id;
578 fc->group_id = d.group_id;
579 fc->max_read = d.max_read;
581 /* Used by get_root_inode() */
585 root = get_root_inode(sb, d.rootmode);
589 root_dentry = d_alloc_root(root);
595 init_req = fuse_request_alloc();
600 fc->destroy_req = fuse_request_alloc();
601 if (!fc->destroy_req)
605 mutex_lock(&fuse_mutex);
607 if (file->private_data)
611 err = fuse_ctl_add_conn(fc);
615 list_add_tail(&fc->entry, &fuse_conn_list);
616 sb->s_root = root_dentry;
618 file->private_data = fuse_conn_get(fc);
619 mutex_unlock(&fuse_mutex);
621 * atomic_dec_and_test() in fput() provides the necessary
622 * memory barrier for file->private_data to be visible on all
627 fuse_send_init(fc, init_req);
632 mutex_unlock(&fuse_mutex);
633 fuse_request_free(init_req);
642 static int fuse_get_sb(struct file_system_type *fs_type,
643 int flags, const char *dev_name,
644 void *raw_data, struct vfsmount *mnt)
646 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
649 static struct file_system_type fuse_fs_type = {
650 .owner = THIS_MODULE,
652 .fs_flags = FS_HAS_SUBTYPE,
653 .get_sb = fuse_get_sb,
654 .kill_sb = kill_anon_super,
658 static int fuse_get_sb_blk(struct file_system_type *fs_type,
659 int flags, const char *dev_name,
660 void *raw_data, struct vfsmount *mnt)
662 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
666 static struct file_system_type fuseblk_fs_type = {
667 .owner = THIS_MODULE,
669 .get_sb = fuse_get_sb_blk,
670 .kill_sb = kill_block_super,
671 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
674 static inline int register_fuseblk(void)
676 return register_filesystem(&fuseblk_fs_type);
679 static inline void unregister_fuseblk(void)
681 unregister_filesystem(&fuseblk_fs_type);
684 static inline int register_fuseblk(void)
689 static inline void unregister_fuseblk(void)
694 static decl_subsys(fuse, NULL, NULL);
695 static decl_subsys(connections, NULL, NULL);
697 static void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
699 struct inode * inode = foo;
701 inode_init_once(inode);
704 static int __init fuse_fs_init(void)
708 err = register_filesystem(&fuse_fs_type);
712 err = register_fuseblk();
716 fuse_inode_cachep = kmem_cache_create("fuse_inode",
717 sizeof(struct fuse_inode),
718 0, SLAB_HWCACHE_ALIGN,
719 fuse_inode_init_once);
721 if (!fuse_inode_cachep)
727 unregister_fuseblk();
729 unregister_filesystem(&fuse_fs_type);
734 static void fuse_fs_cleanup(void)
736 unregister_filesystem(&fuse_fs_type);
737 unregister_fuseblk();
738 kmem_cache_destroy(fuse_inode_cachep);
741 static int fuse_sysfs_init(void)
745 kobj_set_kset_s(&fuse_subsys, fs_subsys);
746 err = subsystem_register(&fuse_subsys);
750 kobj_set_kset_s(&connections_subsys, fuse_subsys);
751 err = subsystem_register(&connections_subsys);
753 goto out_fuse_unregister;
758 subsystem_unregister(&fuse_subsys);
763 static void fuse_sysfs_cleanup(void)
765 subsystem_unregister(&connections_subsys);
766 subsystem_unregister(&fuse_subsys);
769 static int __init fuse_init(void)
773 printk("fuse init (API version %i.%i)\n",
774 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
776 INIT_LIST_HEAD(&fuse_conn_list);
777 res = fuse_fs_init();
781 res = fuse_dev_init();
785 res = fuse_sysfs_init();
787 goto err_dev_cleanup;
789 res = fuse_ctl_init();
791 goto err_sysfs_cleanup;
796 fuse_sysfs_cleanup();
805 static void __exit fuse_exit(void)
807 printk(KERN_DEBUG "fuse exit\n");
810 fuse_sysfs_cleanup();
815 module_init(fuse_init);
816 module_exit(fuse_exit);