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 INIT_LIST_HEAD(&fi->write_files);
60 fi->forget_req = fuse_request_alloc();
61 if (!fi->forget_req) {
62 kmem_cache_free(fuse_inode_cachep, inode);
69 static void fuse_destroy_inode(struct inode *inode)
71 struct fuse_inode *fi = get_fuse_inode(inode);
72 BUG_ON(!list_empty(&fi->write_files));
74 fuse_request_free(fi->forget_req);
75 kmem_cache_free(fuse_inode_cachep, inode);
78 static void fuse_read_inode(struct inode *inode)
83 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
84 unsigned long nodeid, u64 nlookup)
86 struct fuse_forget_in *inarg = &req->misc.forget_in;
87 inarg->nlookup = nlookup;
88 req->in.h.opcode = FUSE_FORGET;
89 req->in.h.nodeid = nodeid;
91 req->in.args[0].size = sizeof(struct fuse_forget_in);
92 req->in.args[0].value = inarg;
93 request_send_noreply(fc, req);
96 static void fuse_clear_inode(struct inode *inode)
98 if (inode->i_sb->s_flags & MS_ACTIVE) {
99 struct fuse_conn *fc = get_fuse_conn(inode);
100 struct fuse_inode *fi = get_fuse_inode(inode);
101 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
102 fi->forget_req = NULL;
106 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
108 if (*flags & MS_MANDLOCK)
114 static void fuse_truncate(struct address_space *mapping, loff_t offset)
116 /* See vmtruncate() */
117 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
118 truncate_inode_pages(mapping, offset);
119 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
123 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
124 u64 attr_valid, u64 attr_version)
126 struct fuse_conn *fc = get_fuse_conn(inode);
127 struct fuse_inode *fi = get_fuse_inode(inode);
130 spin_lock(&fc->lock);
131 if (attr_version != 0 && fi->attr_version > attr_version) {
132 spin_unlock(&fc->lock);
135 fi->attr_version = ++fc->attr_version;
136 fi->i_time = attr_valid;
138 inode->i_ino = attr->ino;
139 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
140 inode->i_nlink = attr->nlink;
141 inode->i_uid = attr->uid;
142 inode->i_gid = attr->gid;
143 inode->i_blocks = attr->blocks;
144 inode->i_atime.tv_sec = attr->atime;
145 inode->i_atime.tv_nsec = attr->atimensec;
146 inode->i_mtime.tv_sec = attr->mtime;
147 inode->i_mtime.tv_nsec = attr->mtimensec;
148 inode->i_ctime.tv_sec = attr->ctime;
149 inode->i_ctime.tv_nsec = attr->ctimensec;
152 * Don't set the sticky bit in i_mode, unless we want the VFS
153 * to check permissions. This prevents failures due to the
154 * check in may_delete().
156 fi->orig_i_mode = inode->i_mode;
157 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
158 inode->i_mode &= ~S_ISVTX;
160 oldsize = inode->i_size;
161 i_size_write(inode, attr->size);
162 spin_unlock(&fc->lock);
164 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
165 if (attr->size < oldsize)
166 fuse_truncate(inode->i_mapping, attr->size);
167 invalidate_inode_pages2(inode->i_mapping);
171 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
173 inode->i_mode = attr->mode & S_IFMT;
174 inode->i_size = attr->size;
175 if (S_ISREG(inode->i_mode)) {
176 fuse_init_common(inode);
177 fuse_init_file_inode(inode);
178 } else if (S_ISDIR(inode->i_mode))
179 fuse_init_dir(inode);
180 else if (S_ISLNK(inode->i_mode))
181 fuse_init_symlink(inode);
182 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
183 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
184 fuse_init_common(inode);
185 init_special_inode(inode, inode->i_mode,
186 new_decode_dev(attr->rdev));
191 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
193 unsigned long nodeid = *(unsigned long *) _nodeidp;
194 if (get_node_id(inode) == nodeid)
200 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
202 unsigned long nodeid = *(unsigned long *) _nodeidp;
203 get_fuse_inode(inode)->nodeid = nodeid;
207 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
208 int generation, struct fuse_attr *attr,
209 u64 attr_valid, u64 attr_version)
212 struct fuse_inode *fi;
213 struct fuse_conn *fc = get_fuse_conn_super(sb);
216 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
220 if ((inode->i_state & I_NEW)) {
221 inode->i_flags |= S_NOATIME|S_NOCMTIME;
222 inode->i_generation = generation;
223 inode->i_data.backing_dev_info = &fc->bdi;
224 fuse_init_inode(inode, attr);
225 unlock_new_inode(inode);
226 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
227 /* Inode has changed type, any I/O on the old should fail */
228 make_bad_inode(inode);
233 fi = get_fuse_inode(inode);
234 spin_lock(&fc->lock);
236 spin_unlock(&fc->lock);
237 fuse_change_attributes(inode, attr, attr_valid, attr_version);
242 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
244 if (flags & MNT_FORCE)
245 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
248 static void fuse_send_destroy(struct fuse_conn *fc)
250 struct fuse_req *req = fc->destroy_req;
251 if (req && fc->conn_init) {
252 fc->destroy_req = NULL;
253 req->in.h.opcode = FUSE_DESTROY;
255 request_send(fc, req);
256 fuse_put_request(fc, req);
260 static void fuse_put_super(struct super_block *sb)
262 struct fuse_conn *fc = get_fuse_conn_super(sb);
264 fuse_send_destroy(fc);
265 spin_lock(&fc->lock);
268 spin_unlock(&fc->lock);
269 /* Flush all readers on this fs */
270 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
271 wake_up_all(&fc->waitq);
272 wake_up_all(&fc->blocked_waitq);
273 wake_up_all(&fc->reserved_req_waitq);
274 mutex_lock(&fuse_mutex);
275 list_del(&fc->entry);
276 fuse_ctl_remove_conn(fc);
277 mutex_unlock(&fuse_mutex);
281 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
283 stbuf->f_type = FUSE_SUPER_MAGIC;
284 stbuf->f_bsize = attr->bsize;
285 stbuf->f_frsize = attr->frsize;
286 stbuf->f_blocks = attr->blocks;
287 stbuf->f_bfree = attr->bfree;
288 stbuf->f_bavail = attr->bavail;
289 stbuf->f_files = attr->files;
290 stbuf->f_ffree = attr->ffree;
291 stbuf->f_namelen = attr->namelen;
292 /* fsid is left zero */
295 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
297 struct super_block *sb = dentry->d_sb;
298 struct fuse_conn *fc = get_fuse_conn_super(sb);
299 struct fuse_req *req;
300 struct fuse_statfs_out outarg;
303 if (!fuse_allow_task(fc, current)) {
304 buf->f_type = FUSE_SUPER_MAGIC;
308 req = fuse_get_req(fc);
312 memset(&outarg, 0, sizeof(outarg));
314 req->in.h.opcode = FUSE_STATFS;
315 req->in.h.nodeid = get_node_id(dentry->d_inode);
316 req->out.numargs = 1;
317 req->out.args[0].size =
318 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
319 req->out.args[0].value = &outarg;
320 request_send(fc, req);
321 err = req->out.h.error;
323 convert_fuse_statfs(buf, &outarg.st);
324 fuse_put_request(fc, req);
333 OPT_DEFAULT_PERMISSIONS,
340 static match_table_t tokens = {
342 {OPT_ROOTMODE, "rootmode=%o"},
343 {OPT_USER_ID, "user_id=%u"},
344 {OPT_GROUP_ID, "group_id=%u"},
345 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
346 {OPT_ALLOW_OTHER, "allow_other"},
347 {OPT_MAX_READ, "max_read=%u"},
348 {OPT_BLKSIZE, "blksize=%u"},
352 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
355 memset(d, 0, sizeof(struct fuse_mount_data));
359 while ((p = strsep(&opt, ",")) != NULL) {
362 substring_t args[MAX_OPT_ARGS];
366 token = match_token(p, tokens, args);
369 if (match_int(&args[0], &value))
376 if (match_octal(&args[0], &value))
378 if (!fuse_valid_type(value))
381 d->rootmode_present = 1;
385 if (match_int(&args[0], &value))
388 d->user_id_present = 1;
392 if (match_int(&args[0], &value))
395 d->group_id_present = 1;
398 case OPT_DEFAULT_PERMISSIONS:
399 d->flags |= FUSE_DEFAULT_PERMISSIONS;
402 case OPT_ALLOW_OTHER:
403 d->flags |= FUSE_ALLOW_OTHER;
407 if (match_int(&args[0], &value))
413 if (!is_bdev || match_int(&args[0], &value))
423 if (!d->fd_present || !d->rootmode_present ||
424 !d->user_id_present || !d->group_id_present)
430 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
432 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
434 seq_printf(m, ",user_id=%u", fc->user_id);
435 seq_printf(m, ",group_id=%u", fc->group_id);
436 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
437 seq_puts(m, ",default_permissions");
438 if (fc->flags & FUSE_ALLOW_OTHER)
439 seq_puts(m, ",allow_other");
440 if (fc->max_read != ~0)
441 seq_printf(m, ",max_read=%u", fc->max_read);
445 static struct fuse_conn *new_conn(void)
447 struct fuse_conn *fc;
450 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
452 spin_lock_init(&fc->lock);
453 mutex_init(&fc->inst_mutex);
454 atomic_set(&fc->count, 1);
455 init_waitqueue_head(&fc->waitq);
456 init_waitqueue_head(&fc->blocked_waitq);
457 init_waitqueue_head(&fc->reserved_req_waitq);
458 INIT_LIST_HEAD(&fc->pending);
459 INIT_LIST_HEAD(&fc->processing);
460 INIT_LIST_HEAD(&fc->io);
461 INIT_LIST_HEAD(&fc->interrupts);
462 atomic_set(&fc->num_waiting, 0);
463 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
464 fc->bdi.unplug_io_fn = default_unplug_io_fn;
465 err = bdi_init(&fc->bdi);
473 fc->attr_version = 1;
474 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
480 void fuse_conn_put(struct fuse_conn *fc)
482 if (atomic_dec_and_test(&fc->count)) {
484 fuse_request_free(fc->destroy_req);
485 mutex_destroy(&fc->inst_mutex);
486 bdi_destroy(&fc->bdi);
491 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
493 atomic_inc(&fc->count);
497 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
499 struct fuse_attr attr;
500 memset(&attr, 0, sizeof(attr));
503 attr.ino = FUSE_ROOT_ID;
505 return fuse_iget(sb, 1, 0, &attr, 0, 0);
508 static const struct super_operations fuse_super_operations = {
509 .alloc_inode = fuse_alloc_inode,
510 .destroy_inode = fuse_destroy_inode,
511 .read_inode = fuse_read_inode,
512 .clear_inode = fuse_clear_inode,
513 .drop_inode = generic_delete_inode,
514 .remount_fs = fuse_remount_fs,
515 .put_super = fuse_put_super,
516 .umount_begin = fuse_umount_begin,
517 .statfs = fuse_statfs,
518 .show_options = fuse_show_options,
521 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
523 struct fuse_init_out *arg = &req->misc.init_out;
525 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
528 unsigned long ra_pages;
530 if (arg->minor >= 6) {
531 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
532 if (arg->flags & FUSE_ASYNC_READ)
534 if (!(arg->flags & FUSE_POSIX_LOCKS))
536 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
537 fc->atomic_o_trunc = 1;
539 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
543 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
544 fc->minor = arg->minor;
545 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
548 fuse_put_request(fc, req);
550 wake_up_all(&fc->blocked_waitq);
553 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
555 struct fuse_init_in *arg = &req->misc.init_in;
557 arg->major = FUSE_KERNEL_VERSION;
558 arg->minor = FUSE_KERNEL_MINOR_VERSION;
559 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
560 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_FILE_OPS |
562 req->in.h.opcode = FUSE_INIT;
564 req->in.args[0].size = sizeof(*arg);
565 req->in.args[0].value = arg;
566 req->out.numargs = 1;
567 /* Variable length arguement used for backward compatibility
568 with interface version < 7.5. Rest of init_out is zeroed
569 by do_get_request(), so a short reply is not a problem */
571 req->out.args[0].size = sizeof(struct fuse_init_out);
572 req->out.args[0].value = &req->misc.init_out;
573 req->end = process_init_reply;
574 request_send_background(fc, req);
577 static u64 conn_id(void)
583 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
585 struct fuse_conn *fc;
587 struct fuse_mount_data d;
589 struct dentry *root_dentry;
590 struct fuse_req *init_req;
592 int is_bdev = sb->s_bdev != NULL;
594 if (sb->s_flags & MS_MANDLOCK)
597 if (!parse_fuse_opt((char *) data, &d, is_bdev))
602 if (!sb_set_blocksize(sb, d.blksize))
606 sb->s_blocksize = PAGE_CACHE_SIZE;
607 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
609 sb->s_magic = FUSE_SUPER_MAGIC;
610 sb->s_op = &fuse_super_operations;
611 sb->s_maxbytes = MAX_LFS_FILESIZE;
617 if (file->f_op != &fuse_dev_operations)
625 fc->user_id = d.user_id;
626 fc->group_id = d.group_id;
627 fc->max_read = d.max_read;
629 /* Used by get_root_inode() */
633 root = get_root_inode(sb, d.rootmode);
637 root_dentry = d_alloc_root(root);
643 init_req = fuse_request_alloc();
648 fc->destroy_req = fuse_request_alloc();
649 if (!fc->destroy_req)
653 mutex_lock(&fuse_mutex);
655 if (file->private_data)
659 err = fuse_ctl_add_conn(fc);
663 list_add_tail(&fc->entry, &fuse_conn_list);
664 sb->s_root = root_dentry;
666 file->private_data = fuse_conn_get(fc);
667 mutex_unlock(&fuse_mutex);
669 * atomic_dec_and_test() in fput() provides the necessary
670 * memory barrier for file->private_data to be visible on all
675 fuse_send_init(fc, init_req);
680 mutex_unlock(&fuse_mutex);
681 fuse_request_free(init_req);
690 static int fuse_get_sb(struct file_system_type *fs_type,
691 int flags, const char *dev_name,
692 void *raw_data, struct vfsmount *mnt)
694 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
697 static struct file_system_type fuse_fs_type = {
698 .owner = THIS_MODULE,
700 .fs_flags = FS_HAS_SUBTYPE,
701 .get_sb = fuse_get_sb,
702 .kill_sb = kill_anon_super,
706 static int fuse_get_sb_blk(struct file_system_type *fs_type,
707 int flags, const char *dev_name,
708 void *raw_data, struct vfsmount *mnt)
710 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
714 static struct file_system_type fuseblk_fs_type = {
715 .owner = THIS_MODULE,
717 .get_sb = fuse_get_sb_blk,
718 .kill_sb = kill_block_super,
719 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
722 static inline int register_fuseblk(void)
724 return register_filesystem(&fuseblk_fs_type);
727 static inline void unregister_fuseblk(void)
729 unregister_filesystem(&fuseblk_fs_type);
732 static inline int register_fuseblk(void)
737 static inline void unregister_fuseblk(void)
742 static decl_subsys(fuse, NULL, NULL);
743 static decl_subsys(connections, NULL, NULL);
745 static void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
747 struct inode * inode = foo;
749 inode_init_once(inode);
752 static int __init fuse_fs_init(void)
756 err = register_filesystem(&fuse_fs_type);
760 err = register_fuseblk();
764 fuse_inode_cachep = kmem_cache_create("fuse_inode",
765 sizeof(struct fuse_inode),
766 0, SLAB_HWCACHE_ALIGN,
767 fuse_inode_init_once);
769 if (!fuse_inode_cachep)
775 unregister_fuseblk();
777 unregister_filesystem(&fuse_fs_type);
782 static void fuse_fs_cleanup(void)
784 unregister_filesystem(&fuse_fs_type);
785 unregister_fuseblk();
786 kmem_cache_destroy(fuse_inode_cachep);
789 static int fuse_sysfs_init(void)
793 kobj_set_kset_s(&fuse_subsys, fs_subsys);
794 err = subsystem_register(&fuse_subsys);
798 kobj_set_kset_s(&connections_subsys, fuse_subsys);
799 err = subsystem_register(&connections_subsys);
801 goto out_fuse_unregister;
806 subsystem_unregister(&fuse_subsys);
811 static void fuse_sysfs_cleanup(void)
813 subsystem_unregister(&connections_subsys);
814 subsystem_unregister(&fuse_subsys);
817 static int __init fuse_init(void)
821 printk("fuse init (API version %i.%i)\n",
822 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
824 INIT_LIST_HEAD(&fuse_conn_list);
825 res = fuse_fs_init();
829 res = fuse_dev_init();
833 res = fuse_sysfs_init();
835 goto err_dev_cleanup;
837 res = fuse_ctl_init();
839 goto err_sysfs_cleanup;
844 fuse_sysfs_cleanup();
853 static void __exit fuse_exit(void)
855 printk(KERN_DEBUG "fuse exit\n");
858 fuse_sysfs_cleanup();
863 module_init(fuse_init);
864 module_exit(fuse_exit);