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);
120 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr)
122 struct fuse_conn *fc = get_fuse_conn(inode);
123 struct fuse_inode *fi = get_fuse_inode(inode);
126 inode->i_ino = attr->ino;
127 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
128 inode->i_nlink = attr->nlink;
129 inode->i_uid = attr->uid;
130 inode->i_gid = attr->gid;
131 inode->i_blocks = attr->blocks;
132 inode->i_atime.tv_sec = attr->atime;
133 inode->i_atime.tv_nsec = attr->atimensec;
134 inode->i_mtime.tv_sec = attr->mtime;
135 inode->i_mtime.tv_nsec = attr->mtimensec;
136 inode->i_ctime.tv_sec = attr->ctime;
137 inode->i_ctime.tv_nsec = attr->ctimensec;
140 * Don't set the sticky bit in i_mode, unless we want the VFS
141 * to check permissions. This prevents failures due to the
142 * check in may_delete().
144 fi->orig_i_mode = inode->i_mode;
145 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
146 inode->i_mode &= ~S_ISVTX;
148 spin_lock(&fc->lock);
149 oldsize = inode->i_size;
150 i_size_write(inode, attr->size);
151 spin_unlock(&fc->lock);
153 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
154 if (attr->size < oldsize)
155 fuse_truncate(inode->i_mapping, attr->size);
156 invalidate_inode_pages2(inode->i_mapping);
160 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
162 inode->i_mode = attr->mode & S_IFMT;
163 inode->i_size = attr->size;
164 if (S_ISREG(inode->i_mode)) {
165 fuse_init_common(inode);
166 fuse_init_file_inode(inode);
167 } else if (S_ISDIR(inode->i_mode))
168 fuse_init_dir(inode);
169 else if (S_ISLNK(inode->i_mode))
170 fuse_init_symlink(inode);
171 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
172 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
173 fuse_init_common(inode);
174 init_special_inode(inode, inode->i_mode,
175 new_decode_dev(attr->rdev));
180 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
182 unsigned long nodeid = *(unsigned long *) _nodeidp;
183 if (get_node_id(inode) == nodeid)
189 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
191 unsigned long nodeid = *(unsigned long *) _nodeidp;
192 get_fuse_inode(inode)->nodeid = nodeid;
196 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
197 int generation, struct fuse_attr *attr)
200 struct fuse_inode *fi;
201 struct fuse_conn *fc = get_fuse_conn_super(sb);
204 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
208 if ((inode->i_state & I_NEW)) {
209 inode->i_flags |= S_NOATIME|S_NOCMTIME;
210 inode->i_generation = generation;
211 inode->i_data.backing_dev_info = &fc->bdi;
212 fuse_init_inode(inode, attr);
213 unlock_new_inode(inode);
214 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
215 /* Inode has changed type, any I/O on the old should fail */
216 make_bad_inode(inode);
221 fi = get_fuse_inode(inode);
222 spin_lock(&fc->lock);
224 spin_unlock(&fc->lock);
225 fuse_change_attributes(inode, attr);
229 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
231 if (flags & MNT_FORCE)
232 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
235 static void fuse_send_destroy(struct fuse_conn *fc)
237 struct fuse_req *req = fc->destroy_req;
238 if (req && fc->conn_init) {
239 fc->destroy_req = NULL;
240 req->in.h.opcode = FUSE_DESTROY;
242 request_send(fc, req);
243 fuse_put_request(fc, req);
247 static void fuse_put_super(struct super_block *sb)
249 struct fuse_conn *fc = get_fuse_conn_super(sb);
251 fuse_send_destroy(fc);
252 spin_lock(&fc->lock);
255 spin_unlock(&fc->lock);
256 /* Flush all readers on this fs */
257 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
258 wake_up_all(&fc->waitq);
259 wake_up_all(&fc->blocked_waitq);
260 wake_up_all(&fc->reserved_req_waitq);
261 mutex_lock(&fuse_mutex);
262 list_del(&fc->entry);
263 fuse_ctl_remove_conn(fc);
264 mutex_unlock(&fuse_mutex);
268 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
270 stbuf->f_type = FUSE_SUPER_MAGIC;
271 stbuf->f_bsize = attr->bsize;
272 stbuf->f_frsize = attr->frsize;
273 stbuf->f_blocks = attr->blocks;
274 stbuf->f_bfree = attr->bfree;
275 stbuf->f_bavail = attr->bavail;
276 stbuf->f_files = attr->files;
277 stbuf->f_ffree = attr->ffree;
278 stbuf->f_namelen = attr->namelen;
279 /* fsid is left zero */
282 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
284 struct super_block *sb = dentry->d_sb;
285 struct fuse_conn *fc = get_fuse_conn_super(sb);
286 struct fuse_req *req;
287 struct fuse_statfs_out outarg;
290 req = fuse_get_req(fc);
294 memset(&outarg, 0, sizeof(outarg));
296 req->in.h.opcode = FUSE_STATFS;
297 req->in.h.nodeid = get_node_id(dentry->d_inode);
298 req->out.numargs = 1;
299 req->out.args[0].size =
300 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
301 req->out.args[0].value = &outarg;
302 request_send(fc, req);
303 err = req->out.h.error;
305 convert_fuse_statfs(buf, &outarg.st);
306 fuse_put_request(fc, req);
315 OPT_DEFAULT_PERMISSIONS,
322 static match_table_t tokens = {
324 {OPT_ROOTMODE, "rootmode=%o"},
325 {OPT_USER_ID, "user_id=%u"},
326 {OPT_GROUP_ID, "group_id=%u"},
327 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
328 {OPT_ALLOW_OTHER, "allow_other"},
329 {OPT_MAX_READ, "max_read=%u"},
330 {OPT_BLKSIZE, "blksize=%u"},
334 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
337 memset(d, 0, sizeof(struct fuse_mount_data));
341 while ((p = strsep(&opt, ",")) != NULL) {
344 substring_t args[MAX_OPT_ARGS];
348 token = match_token(p, tokens, args);
351 if (match_int(&args[0], &value))
358 if (match_octal(&args[0], &value))
360 if (!fuse_valid_type(value))
363 d->rootmode_present = 1;
367 if (match_int(&args[0], &value))
370 d->user_id_present = 1;
374 if (match_int(&args[0], &value))
377 d->group_id_present = 1;
380 case OPT_DEFAULT_PERMISSIONS:
381 d->flags |= FUSE_DEFAULT_PERMISSIONS;
384 case OPT_ALLOW_OTHER:
385 d->flags |= FUSE_ALLOW_OTHER;
389 if (match_int(&args[0], &value))
395 if (!is_bdev || match_int(&args[0], &value))
405 if (!d->fd_present || !d->rootmode_present ||
406 !d->user_id_present || !d->group_id_present)
412 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
414 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
416 seq_printf(m, ",user_id=%u", fc->user_id);
417 seq_printf(m, ",group_id=%u", fc->group_id);
418 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
419 seq_puts(m, ",default_permissions");
420 if (fc->flags & FUSE_ALLOW_OTHER)
421 seq_puts(m, ",allow_other");
422 if (fc->max_read != ~0)
423 seq_printf(m, ",max_read=%u", fc->max_read);
427 static struct fuse_conn *new_conn(void)
429 struct fuse_conn *fc;
432 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
434 spin_lock_init(&fc->lock);
435 mutex_init(&fc->inst_mutex);
436 atomic_set(&fc->count, 1);
437 init_waitqueue_head(&fc->waitq);
438 init_waitqueue_head(&fc->blocked_waitq);
439 init_waitqueue_head(&fc->reserved_req_waitq);
440 INIT_LIST_HEAD(&fc->pending);
441 INIT_LIST_HEAD(&fc->processing);
442 INIT_LIST_HEAD(&fc->io);
443 INIT_LIST_HEAD(&fc->interrupts);
444 atomic_set(&fc->num_waiting, 0);
445 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
446 fc->bdi.unplug_io_fn = default_unplug_io_fn;
447 err = bdi_init(&fc->bdi);
455 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
461 void fuse_conn_put(struct fuse_conn *fc)
463 if (atomic_dec_and_test(&fc->count)) {
465 fuse_request_free(fc->destroy_req);
466 mutex_destroy(&fc->inst_mutex);
467 bdi_destroy(&fc->bdi);
472 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
474 atomic_inc(&fc->count);
478 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
480 struct fuse_attr attr;
481 memset(&attr, 0, sizeof(attr));
484 attr.ino = FUSE_ROOT_ID;
486 return fuse_iget(sb, 1, 0, &attr);
489 static const struct super_operations fuse_super_operations = {
490 .alloc_inode = fuse_alloc_inode,
491 .destroy_inode = fuse_destroy_inode,
492 .read_inode = fuse_read_inode,
493 .clear_inode = fuse_clear_inode,
494 .drop_inode = generic_delete_inode,
495 .remount_fs = fuse_remount_fs,
496 .put_super = fuse_put_super,
497 .umount_begin = fuse_umount_begin,
498 .statfs = fuse_statfs,
499 .show_options = fuse_show_options,
502 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
504 struct fuse_init_out *arg = &req->misc.init_out;
506 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
509 unsigned long ra_pages;
511 if (arg->minor >= 6) {
512 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
513 if (arg->flags & FUSE_ASYNC_READ)
515 if (!(arg->flags & FUSE_POSIX_LOCKS))
518 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
522 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
523 fc->minor = arg->minor;
524 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
527 fuse_put_request(fc, req);
529 wake_up_all(&fc->blocked_waitq);
532 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
534 struct fuse_init_in *arg = &req->misc.init_in;
536 arg->major = FUSE_KERNEL_VERSION;
537 arg->minor = FUSE_KERNEL_MINOR_VERSION;
538 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
539 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS;
540 req->in.h.opcode = FUSE_INIT;
542 req->in.args[0].size = sizeof(*arg);
543 req->in.args[0].value = arg;
544 req->out.numargs = 1;
545 /* Variable length arguement used for backward compatibility
546 with interface version < 7.5. Rest of init_out is zeroed
547 by do_get_request(), so a short reply is not a problem */
549 req->out.args[0].size = sizeof(struct fuse_init_out);
550 req->out.args[0].value = &req->misc.init_out;
551 req->end = process_init_reply;
552 request_send_background(fc, req);
555 static u64 conn_id(void)
561 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
563 struct fuse_conn *fc;
565 struct fuse_mount_data d;
567 struct dentry *root_dentry;
568 struct fuse_req *init_req;
570 int is_bdev = sb->s_bdev != NULL;
572 if (sb->s_flags & MS_MANDLOCK)
575 if (!parse_fuse_opt((char *) data, &d, is_bdev))
580 if (!sb_set_blocksize(sb, d.blksize))
584 sb->s_blocksize = PAGE_CACHE_SIZE;
585 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
587 sb->s_magic = FUSE_SUPER_MAGIC;
588 sb->s_op = &fuse_super_operations;
589 sb->s_maxbytes = MAX_LFS_FILESIZE;
595 if (file->f_op != &fuse_dev_operations)
603 fc->user_id = d.user_id;
604 fc->group_id = d.group_id;
605 fc->max_read = d.max_read;
607 /* Used by get_root_inode() */
611 root = get_root_inode(sb, d.rootmode);
615 root_dentry = d_alloc_root(root);
621 init_req = fuse_request_alloc();
626 fc->destroy_req = fuse_request_alloc();
627 if (!fc->destroy_req)
631 mutex_lock(&fuse_mutex);
633 if (file->private_data)
637 err = fuse_ctl_add_conn(fc);
641 list_add_tail(&fc->entry, &fuse_conn_list);
642 sb->s_root = root_dentry;
644 file->private_data = fuse_conn_get(fc);
645 mutex_unlock(&fuse_mutex);
647 * atomic_dec_and_test() in fput() provides the necessary
648 * memory barrier for file->private_data to be visible on all
653 fuse_send_init(fc, init_req);
658 mutex_unlock(&fuse_mutex);
659 fuse_request_free(init_req);
668 static int fuse_get_sb(struct file_system_type *fs_type,
669 int flags, const char *dev_name,
670 void *raw_data, struct vfsmount *mnt)
672 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
675 static struct file_system_type fuse_fs_type = {
676 .owner = THIS_MODULE,
678 .fs_flags = FS_HAS_SUBTYPE,
679 .get_sb = fuse_get_sb,
680 .kill_sb = kill_anon_super,
684 static int fuse_get_sb_blk(struct file_system_type *fs_type,
685 int flags, const char *dev_name,
686 void *raw_data, struct vfsmount *mnt)
688 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
692 static struct file_system_type fuseblk_fs_type = {
693 .owner = THIS_MODULE,
695 .get_sb = fuse_get_sb_blk,
696 .kill_sb = kill_block_super,
697 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
700 static inline int register_fuseblk(void)
702 return register_filesystem(&fuseblk_fs_type);
705 static inline void unregister_fuseblk(void)
707 unregister_filesystem(&fuseblk_fs_type);
710 static inline int register_fuseblk(void)
715 static inline void unregister_fuseblk(void)
720 static decl_subsys(fuse, NULL, NULL);
721 static decl_subsys(connections, NULL, NULL);
723 static void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
725 struct inode * inode = foo;
727 inode_init_once(inode);
730 static int __init fuse_fs_init(void)
734 err = register_filesystem(&fuse_fs_type);
738 err = register_fuseblk();
742 fuse_inode_cachep = kmem_cache_create("fuse_inode",
743 sizeof(struct fuse_inode),
744 0, SLAB_HWCACHE_ALIGN,
745 fuse_inode_init_once);
747 if (!fuse_inode_cachep)
753 unregister_fuseblk();
755 unregister_filesystem(&fuse_fs_type);
760 static void fuse_fs_cleanup(void)
762 unregister_filesystem(&fuse_fs_type);
763 unregister_fuseblk();
764 kmem_cache_destroy(fuse_inode_cachep);
767 static int fuse_sysfs_init(void)
771 kobj_set_kset_s(&fuse_subsys, fs_subsys);
772 err = subsystem_register(&fuse_subsys);
776 kobj_set_kset_s(&connections_subsys, fuse_subsys);
777 err = subsystem_register(&connections_subsys);
779 goto out_fuse_unregister;
784 subsystem_unregister(&fuse_subsys);
789 static void fuse_sysfs_cleanup(void)
791 subsystem_unregister(&connections_subsys);
792 subsystem_unregister(&fuse_subsys);
795 static int __init fuse_init(void)
799 printk("fuse init (API version %i.%i)\n",
800 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
802 INIT_LIST_HEAD(&fuse_conn_list);
803 res = fuse_fs_init();
807 res = fuse_dev_init();
811 res = fuse_sysfs_init();
813 goto err_dev_cleanup;
815 res = fuse_ctl_init();
817 goto err_sysfs_cleanup;
822 fuse_sysfs_cleanup();
831 static void __exit fuse_exit(void)
833 printk(KERN_DEBUG "fuse exit\n");
836 fuse_sysfs_cleanup();
841 module_init(fuse_init);
842 module_exit(fuse_exit);