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 mutex_lock(&fuse_mutex);
236 list_del(&fc->entry);
237 fuse_ctl_remove_conn(fc);
238 mutex_unlock(&fuse_mutex);
242 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
244 stbuf->f_type = FUSE_SUPER_MAGIC;
245 stbuf->f_bsize = attr->bsize;
246 stbuf->f_frsize = attr->frsize;
247 stbuf->f_blocks = attr->blocks;
248 stbuf->f_bfree = attr->bfree;
249 stbuf->f_bavail = attr->bavail;
250 stbuf->f_files = attr->files;
251 stbuf->f_ffree = attr->ffree;
252 stbuf->f_namelen = attr->namelen;
253 /* fsid is left zero */
256 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
258 struct super_block *sb = dentry->d_sb;
259 struct fuse_conn *fc = get_fuse_conn_super(sb);
260 struct fuse_req *req;
261 struct fuse_statfs_out outarg;
264 req = fuse_get_req(fc);
268 memset(&outarg, 0, sizeof(outarg));
270 req->in.h.opcode = FUSE_STATFS;
271 req->in.h.nodeid = get_node_id(dentry->d_inode);
272 req->out.numargs = 1;
273 req->out.args[0].size =
274 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
275 req->out.args[0].value = &outarg;
276 request_send(fc, req);
277 err = req->out.h.error;
279 convert_fuse_statfs(buf, &outarg.st);
280 fuse_put_request(fc, req);
289 OPT_DEFAULT_PERMISSIONS,
296 static match_table_t tokens = {
298 {OPT_ROOTMODE, "rootmode=%o"},
299 {OPT_USER_ID, "user_id=%u"},
300 {OPT_GROUP_ID, "group_id=%u"},
301 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
302 {OPT_ALLOW_OTHER, "allow_other"},
303 {OPT_MAX_READ, "max_read=%u"},
304 {OPT_BLKSIZE, "blksize=%u"},
308 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
311 memset(d, 0, sizeof(struct fuse_mount_data));
315 while ((p = strsep(&opt, ",")) != NULL) {
318 substring_t args[MAX_OPT_ARGS];
322 token = match_token(p, tokens, args);
325 if (match_int(&args[0], &value))
332 if (match_octal(&args[0], &value))
334 if (!fuse_valid_type(value))
337 d->rootmode_present = 1;
341 if (match_int(&args[0], &value))
344 d->user_id_present = 1;
348 if (match_int(&args[0], &value))
351 d->group_id_present = 1;
354 case OPT_DEFAULT_PERMISSIONS:
355 d->flags |= FUSE_DEFAULT_PERMISSIONS;
358 case OPT_ALLOW_OTHER:
359 d->flags |= FUSE_ALLOW_OTHER;
363 if (match_int(&args[0], &value))
369 if (!is_bdev || match_int(&args[0], &value))
379 if (!d->fd_present || !d->rootmode_present ||
380 !d->user_id_present || !d->group_id_present)
386 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
388 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
390 seq_printf(m, ",user_id=%u", fc->user_id);
391 seq_printf(m, ",group_id=%u", fc->group_id);
392 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
393 seq_puts(m, ",default_permissions");
394 if (fc->flags & FUSE_ALLOW_OTHER)
395 seq_puts(m, ",allow_other");
396 if (fc->max_read != ~0)
397 seq_printf(m, ",max_read=%u", fc->max_read);
401 static struct fuse_conn *new_conn(void)
403 struct fuse_conn *fc;
405 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
407 spin_lock_init(&fc->lock);
408 mutex_init(&fc->inst_mutex);
409 atomic_set(&fc->count, 1);
410 init_waitqueue_head(&fc->waitq);
411 init_waitqueue_head(&fc->blocked_waitq);
412 INIT_LIST_HEAD(&fc->pending);
413 INIT_LIST_HEAD(&fc->processing);
414 INIT_LIST_HEAD(&fc->io);
415 INIT_LIST_HEAD(&fc->interrupts);
416 atomic_set(&fc->num_waiting, 0);
417 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
418 fc->bdi.unplug_io_fn = default_unplug_io_fn;
421 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
426 void fuse_conn_put(struct fuse_conn *fc)
428 if (atomic_dec_and_test(&fc->count)) {
430 fuse_request_free(fc->destroy_req);
431 mutex_destroy(&fc->inst_mutex);
436 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
438 atomic_inc(&fc->count);
442 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
444 struct fuse_attr attr;
445 memset(&attr, 0, sizeof(attr));
448 attr.ino = FUSE_ROOT_ID;
449 return fuse_iget(sb, 1, 0, &attr);
452 static const struct super_operations fuse_super_operations = {
453 .alloc_inode = fuse_alloc_inode,
454 .destroy_inode = fuse_destroy_inode,
455 .read_inode = fuse_read_inode,
456 .clear_inode = fuse_clear_inode,
457 .remount_fs = fuse_remount_fs,
458 .put_super = fuse_put_super,
459 .umount_begin = fuse_umount_begin,
460 .statfs = fuse_statfs,
461 .show_options = fuse_show_options,
464 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
466 struct fuse_init_out *arg = &req->misc.init_out;
468 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
471 unsigned long ra_pages;
473 if (arg->minor >= 6) {
474 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
475 if (arg->flags & FUSE_ASYNC_READ)
477 if (!(arg->flags & FUSE_POSIX_LOCKS))
480 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
484 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
485 fc->minor = arg->minor;
486 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
489 fuse_put_request(fc, req);
491 wake_up_all(&fc->blocked_waitq);
494 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
496 struct fuse_init_in *arg = &req->misc.init_in;
498 arg->major = FUSE_KERNEL_VERSION;
499 arg->minor = FUSE_KERNEL_MINOR_VERSION;
500 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
501 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS;
502 req->in.h.opcode = FUSE_INIT;
504 req->in.args[0].size = sizeof(*arg);
505 req->in.args[0].value = arg;
506 req->out.numargs = 1;
507 /* Variable length arguement used for backward compatibility
508 with interface version < 7.5. Rest of init_out is zeroed
509 by do_get_request(), so a short reply is not a problem */
511 req->out.args[0].size = sizeof(struct fuse_init_out);
512 req->out.args[0].value = &req->misc.init_out;
513 req->end = process_init_reply;
514 request_send_background(fc, req);
517 static u64 conn_id(void)
523 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
525 struct fuse_conn *fc;
527 struct fuse_mount_data d;
529 struct dentry *root_dentry;
530 struct fuse_req *init_req;
532 int is_bdev = sb->s_bdev != NULL;
534 if (sb->s_flags & MS_MANDLOCK)
537 if (!parse_fuse_opt((char *) data, &d, is_bdev))
542 if (!sb_set_blocksize(sb, d.blksize))
546 sb->s_blocksize = PAGE_CACHE_SIZE;
547 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
549 sb->s_magic = FUSE_SUPER_MAGIC;
550 sb->s_op = &fuse_super_operations;
551 sb->s_maxbytes = MAX_LFS_FILESIZE;
557 if (file->f_op != &fuse_dev_operations)
565 fc->user_id = d.user_id;
566 fc->group_id = d.group_id;
567 fc->max_read = d.max_read;
569 /* Used by get_root_inode() */
573 root = get_root_inode(sb, d.rootmode);
577 root_dentry = d_alloc_root(root);
583 init_req = fuse_request_alloc();
588 fc->destroy_req = fuse_request_alloc();
589 if (!fc->destroy_req)
593 mutex_lock(&fuse_mutex);
595 if (file->private_data)
599 err = fuse_ctl_add_conn(fc);
603 list_add_tail(&fc->entry, &fuse_conn_list);
604 sb->s_root = root_dentry;
606 file->private_data = fuse_conn_get(fc);
607 mutex_unlock(&fuse_mutex);
609 * atomic_dec_and_test() in fput() provides the necessary
610 * memory barrier for file->private_data to be visible on all
615 fuse_send_init(fc, init_req);
620 mutex_unlock(&fuse_mutex);
621 fuse_request_free(init_req);
630 static int fuse_get_sb(struct file_system_type *fs_type,
631 int flags, const char *dev_name,
632 void *raw_data, struct vfsmount *mnt)
634 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
637 static struct file_system_type fuse_fs_type = {
638 .owner = THIS_MODULE,
640 .fs_flags = FS_HAS_SUBTYPE,
641 .get_sb = fuse_get_sb,
642 .kill_sb = kill_anon_super,
646 static int fuse_get_sb_blk(struct file_system_type *fs_type,
647 int flags, const char *dev_name,
648 void *raw_data, struct vfsmount *mnt)
650 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
654 static struct file_system_type fuseblk_fs_type = {
655 .owner = THIS_MODULE,
657 .fs_flags = FS_HAS_SUBTYPE,
658 .get_sb = fuse_get_sb_blk,
659 .kill_sb = kill_block_super,
660 .fs_flags = FS_REQUIRES_DEV,
663 static inline int register_fuseblk(void)
665 return register_filesystem(&fuseblk_fs_type);
668 static inline void unregister_fuseblk(void)
670 unregister_filesystem(&fuseblk_fs_type);
673 static inline int register_fuseblk(void)
678 static inline void unregister_fuseblk(void)
683 static decl_subsys(fuse, NULL, NULL);
684 static decl_subsys(connections, NULL, NULL);
686 static void fuse_inode_init_once(void *foo, struct kmem_cache *cachep,
689 struct inode * inode = foo;
691 inode_init_once(inode);
694 static int __init fuse_fs_init(void)
698 err = register_filesystem(&fuse_fs_type);
702 err = register_fuseblk();
706 fuse_inode_cachep = kmem_cache_create("fuse_inode",
707 sizeof(struct fuse_inode),
708 0, SLAB_HWCACHE_ALIGN,
709 fuse_inode_init_once, NULL);
711 if (!fuse_inode_cachep)
717 unregister_fuseblk();
719 unregister_filesystem(&fuse_fs_type);
724 static void fuse_fs_cleanup(void)
726 unregister_filesystem(&fuse_fs_type);
727 unregister_fuseblk();
728 kmem_cache_destroy(fuse_inode_cachep);
731 static int fuse_sysfs_init(void)
735 kobj_set_kset_s(&fuse_subsys, fs_subsys);
736 err = subsystem_register(&fuse_subsys);
740 kobj_set_kset_s(&connections_subsys, fuse_subsys);
741 err = subsystem_register(&connections_subsys);
743 goto out_fuse_unregister;
748 subsystem_unregister(&fuse_subsys);
753 static void fuse_sysfs_cleanup(void)
755 subsystem_unregister(&connections_subsys);
756 subsystem_unregister(&fuse_subsys);
759 static int __init fuse_init(void)
763 printk("fuse init (API version %i.%i)\n",
764 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
766 INIT_LIST_HEAD(&fuse_conn_list);
767 res = fuse_fs_init();
771 res = fuse_dev_init();
775 res = fuse_sysfs_init();
777 goto err_dev_cleanup;
779 res = fuse_ctl_init();
781 goto err_sysfs_cleanup;
786 fuse_sysfs_cleanup();
795 static void __exit fuse_exit(void)
797 printk(KERN_DEBUG "fuse exit\n");
800 fuse_sysfs_cleanup();
805 module_init(fuse_init);
806 module_exit(fuse_exit);