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);
125 inode->i_ino = attr->ino;
126 inode->i_mode = (inode->i_mode & S_IFMT) + (attr->mode & 07777);
127 inode->i_nlink = attr->nlink;
128 inode->i_uid = attr->uid;
129 inode->i_gid = attr->gid;
130 inode->i_blocks = attr->blocks;
131 inode->i_atime.tv_sec = attr->atime;
132 inode->i_atime.tv_nsec = attr->atimensec;
133 inode->i_mtime.tv_sec = attr->mtime;
134 inode->i_mtime.tv_nsec = attr->mtimensec;
135 inode->i_ctime.tv_sec = attr->ctime;
136 inode->i_ctime.tv_nsec = attr->ctimensec;
138 spin_lock(&fc->lock);
139 oldsize = inode->i_size;
140 i_size_write(inode, attr->size);
141 spin_unlock(&fc->lock);
143 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
144 if (attr->size < oldsize)
145 fuse_truncate(inode->i_mapping, attr->size);
146 invalidate_mapping_pages(inode->i_mapping, 0, -1);
150 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
152 inode->i_mode = attr->mode & S_IFMT;
153 inode->i_size = attr->size;
154 if (S_ISREG(inode->i_mode)) {
155 fuse_init_common(inode);
156 fuse_init_file_inode(inode);
157 } else if (S_ISDIR(inode->i_mode))
158 fuse_init_dir(inode);
159 else if (S_ISLNK(inode->i_mode))
160 fuse_init_symlink(inode);
161 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
162 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
163 fuse_init_common(inode);
164 init_special_inode(inode, inode->i_mode,
165 new_decode_dev(attr->rdev));
170 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
172 unsigned long nodeid = *(unsigned long *) _nodeidp;
173 if (get_node_id(inode) == nodeid)
179 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
181 unsigned long nodeid = *(unsigned long *) _nodeidp;
182 get_fuse_inode(inode)->nodeid = nodeid;
186 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
187 int generation, struct fuse_attr *attr)
190 struct fuse_inode *fi;
191 struct fuse_conn *fc = get_fuse_conn_super(sb);
194 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
198 if ((inode->i_state & I_NEW)) {
199 inode->i_flags |= S_NOATIME|S_NOCMTIME;
200 inode->i_generation = generation;
201 inode->i_data.backing_dev_info = &fc->bdi;
202 fuse_init_inode(inode, attr);
203 unlock_new_inode(inode);
204 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
205 /* Inode has changed type, any I/O on the old should fail */
206 make_bad_inode(inode);
211 fi = get_fuse_inode(inode);
212 spin_lock(&fc->lock);
214 spin_unlock(&fc->lock);
215 fuse_change_attributes(inode, attr);
219 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
221 if (flags & MNT_FORCE)
222 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
225 static void fuse_send_destroy(struct fuse_conn *fc)
227 struct fuse_req *req = fc->destroy_req;
228 if (req && fc->conn_init) {
229 fc->destroy_req = NULL;
230 req->in.h.opcode = FUSE_DESTROY;
232 request_send(fc, req);
233 fuse_put_request(fc, req);
237 static void fuse_put_super(struct super_block *sb)
239 struct fuse_conn *fc = get_fuse_conn_super(sb);
241 fuse_send_destroy(fc);
242 spin_lock(&fc->lock);
245 spin_unlock(&fc->lock);
246 /* Flush all readers on this fs */
247 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
248 wake_up_all(&fc->waitq);
249 wake_up_all(&fc->blocked_waitq);
250 wake_up_all(&fc->reserved_req_waitq);
251 mutex_lock(&fuse_mutex);
252 list_del(&fc->entry);
253 fuse_ctl_remove_conn(fc);
254 mutex_unlock(&fuse_mutex);
258 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
260 stbuf->f_type = FUSE_SUPER_MAGIC;
261 stbuf->f_bsize = attr->bsize;
262 stbuf->f_frsize = attr->frsize;
263 stbuf->f_blocks = attr->blocks;
264 stbuf->f_bfree = attr->bfree;
265 stbuf->f_bavail = attr->bavail;
266 stbuf->f_files = attr->files;
267 stbuf->f_ffree = attr->ffree;
268 stbuf->f_namelen = attr->namelen;
269 /* fsid is left zero */
272 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
274 struct super_block *sb = dentry->d_sb;
275 struct fuse_conn *fc = get_fuse_conn_super(sb);
276 struct fuse_req *req;
277 struct fuse_statfs_out outarg;
280 req = fuse_get_req(fc);
284 memset(&outarg, 0, sizeof(outarg));
286 req->in.h.opcode = FUSE_STATFS;
287 req->in.h.nodeid = get_node_id(dentry->d_inode);
288 req->out.numargs = 1;
289 req->out.args[0].size =
290 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
291 req->out.args[0].value = &outarg;
292 request_send(fc, req);
293 err = req->out.h.error;
295 convert_fuse_statfs(buf, &outarg.st);
296 fuse_put_request(fc, req);
305 OPT_DEFAULT_PERMISSIONS,
312 static match_table_t tokens = {
314 {OPT_ROOTMODE, "rootmode=%o"},
315 {OPT_USER_ID, "user_id=%u"},
316 {OPT_GROUP_ID, "group_id=%u"},
317 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
318 {OPT_ALLOW_OTHER, "allow_other"},
319 {OPT_MAX_READ, "max_read=%u"},
320 {OPT_BLKSIZE, "blksize=%u"},
324 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
327 memset(d, 0, sizeof(struct fuse_mount_data));
331 while ((p = strsep(&opt, ",")) != NULL) {
334 substring_t args[MAX_OPT_ARGS];
338 token = match_token(p, tokens, args);
341 if (match_int(&args[0], &value))
348 if (match_octal(&args[0], &value))
350 if (!fuse_valid_type(value))
353 d->rootmode_present = 1;
357 if (match_int(&args[0], &value))
360 d->user_id_present = 1;
364 if (match_int(&args[0], &value))
367 d->group_id_present = 1;
370 case OPT_DEFAULT_PERMISSIONS:
371 d->flags |= FUSE_DEFAULT_PERMISSIONS;
374 case OPT_ALLOW_OTHER:
375 d->flags |= FUSE_ALLOW_OTHER;
379 if (match_int(&args[0], &value))
385 if (!is_bdev || match_int(&args[0], &value))
395 if (!d->fd_present || !d->rootmode_present ||
396 !d->user_id_present || !d->group_id_present)
402 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
404 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
406 seq_printf(m, ",user_id=%u", fc->user_id);
407 seq_printf(m, ",group_id=%u", fc->group_id);
408 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
409 seq_puts(m, ",default_permissions");
410 if (fc->flags & FUSE_ALLOW_OTHER)
411 seq_puts(m, ",allow_other");
412 if (fc->max_read != ~0)
413 seq_printf(m, ",max_read=%u", fc->max_read);
417 static struct fuse_conn *new_conn(void)
419 struct fuse_conn *fc;
422 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
424 spin_lock_init(&fc->lock);
425 mutex_init(&fc->inst_mutex);
426 atomic_set(&fc->count, 1);
427 init_waitqueue_head(&fc->waitq);
428 init_waitqueue_head(&fc->blocked_waitq);
429 init_waitqueue_head(&fc->reserved_req_waitq);
430 INIT_LIST_HEAD(&fc->pending);
431 INIT_LIST_HEAD(&fc->processing);
432 INIT_LIST_HEAD(&fc->io);
433 INIT_LIST_HEAD(&fc->interrupts);
434 atomic_set(&fc->num_waiting, 0);
435 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
436 fc->bdi.unplug_io_fn = default_unplug_io_fn;
437 err = bdi_init(&fc->bdi);
445 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
451 void fuse_conn_put(struct fuse_conn *fc)
453 if (atomic_dec_and_test(&fc->count)) {
455 fuse_request_free(fc->destroy_req);
456 mutex_destroy(&fc->inst_mutex);
457 bdi_destroy(&fc->bdi);
462 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
464 atomic_inc(&fc->count);
468 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
470 struct fuse_attr attr;
471 memset(&attr, 0, sizeof(attr));
474 attr.ino = FUSE_ROOT_ID;
475 return fuse_iget(sb, 1, 0, &attr);
478 static const struct super_operations fuse_super_operations = {
479 .alloc_inode = fuse_alloc_inode,
480 .destroy_inode = fuse_destroy_inode,
481 .read_inode = fuse_read_inode,
482 .clear_inode = fuse_clear_inode,
483 .drop_inode = generic_delete_inode,
484 .remount_fs = fuse_remount_fs,
485 .put_super = fuse_put_super,
486 .umount_begin = fuse_umount_begin,
487 .statfs = fuse_statfs,
488 .show_options = fuse_show_options,
491 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
493 struct fuse_init_out *arg = &req->misc.init_out;
495 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
498 unsigned long ra_pages;
500 if (arg->minor >= 6) {
501 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
502 if (arg->flags & FUSE_ASYNC_READ)
504 if (!(arg->flags & FUSE_POSIX_LOCKS))
507 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
511 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
512 fc->minor = arg->minor;
513 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
516 fuse_put_request(fc, req);
518 wake_up_all(&fc->blocked_waitq);
521 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
523 struct fuse_init_in *arg = &req->misc.init_in;
525 arg->major = FUSE_KERNEL_VERSION;
526 arg->minor = FUSE_KERNEL_MINOR_VERSION;
527 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
528 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS;
529 req->in.h.opcode = FUSE_INIT;
531 req->in.args[0].size = sizeof(*arg);
532 req->in.args[0].value = arg;
533 req->out.numargs = 1;
534 /* Variable length arguement used for backward compatibility
535 with interface version < 7.5. Rest of init_out is zeroed
536 by do_get_request(), so a short reply is not a problem */
538 req->out.args[0].size = sizeof(struct fuse_init_out);
539 req->out.args[0].value = &req->misc.init_out;
540 req->end = process_init_reply;
541 request_send_background(fc, req);
544 static u64 conn_id(void)
550 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
552 struct fuse_conn *fc;
554 struct fuse_mount_data d;
556 struct dentry *root_dentry;
557 struct fuse_req *init_req;
559 int is_bdev = sb->s_bdev != NULL;
561 if (sb->s_flags & MS_MANDLOCK)
564 if (!parse_fuse_opt((char *) data, &d, is_bdev))
569 if (!sb_set_blocksize(sb, d.blksize))
573 sb->s_blocksize = PAGE_CACHE_SIZE;
574 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
576 sb->s_magic = FUSE_SUPER_MAGIC;
577 sb->s_op = &fuse_super_operations;
578 sb->s_maxbytes = MAX_LFS_FILESIZE;
584 if (file->f_op != &fuse_dev_operations)
592 fc->user_id = d.user_id;
593 fc->group_id = d.group_id;
594 fc->max_read = d.max_read;
596 /* Used by get_root_inode() */
600 root = get_root_inode(sb, d.rootmode);
604 root_dentry = d_alloc_root(root);
610 init_req = fuse_request_alloc();
615 fc->destroy_req = fuse_request_alloc();
616 if (!fc->destroy_req)
620 mutex_lock(&fuse_mutex);
622 if (file->private_data)
626 err = fuse_ctl_add_conn(fc);
630 list_add_tail(&fc->entry, &fuse_conn_list);
631 sb->s_root = root_dentry;
633 file->private_data = fuse_conn_get(fc);
634 mutex_unlock(&fuse_mutex);
636 * atomic_dec_and_test() in fput() provides the necessary
637 * memory barrier for file->private_data to be visible on all
642 fuse_send_init(fc, init_req);
647 mutex_unlock(&fuse_mutex);
648 fuse_request_free(init_req);
657 static int fuse_get_sb(struct file_system_type *fs_type,
658 int flags, const char *dev_name,
659 void *raw_data, struct vfsmount *mnt)
661 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
664 static struct file_system_type fuse_fs_type = {
665 .owner = THIS_MODULE,
667 .fs_flags = FS_HAS_SUBTYPE,
668 .get_sb = fuse_get_sb,
669 .kill_sb = kill_anon_super,
673 static int fuse_get_sb_blk(struct file_system_type *fs_type,
674 int flags, const char *dev_name,
675 void *raw_data, struct vfsmount *mnt)
677 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
681 static struct file_system_type fuseblk_fs_type = {
682 .owner = THIS_MODULE,
684 .get_sb = fuse_get_sb_blk,
685 .kill_sb = kill_block_super,
686 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
689 static inline int register_fuseblk(void)
691 return register_filesystem(&fuseblk_fs_type);
694 static inline void unregister_fuseblk(void)
696 unregister_filesystem(&fuseblk_fs_type);
699 static inline int register_fuseblk(void)
704 static inline void unregister_fuseblk(void)
709 static decl_subsys(fuse, NULL, NULL);
710 static decl_subsys(connections, NULL, NULL);
712 static void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
714 struct inode * inode = foo;
716 inode_init_once(inode);
719 static int __init fuse_fs_init(void)
723 err = register_filesystem(&fuse_fs_type);
727 err = register_fuseblk();
731 fuse_inode_cachep = kmem_cache_create("fuse_inode",
732 sizeof(struct fuse_inode),
733 0, SLAB_HWCACHE_ALIGN,
734 fuse_inode_init_once);
736 if (!fuse_inode_cachep)
742 unregister_fuseblk();
744 unregister_filesystem(&fuse_fs_type);
749 static void fuse_fs_cleanup(void)
751 unregister_filesystem(&fuse_fs_type);
752 unregister_fuseblk();
753 kmem_cache_destroy(fuse_inode_cachep);
756 static int fuse_sysfs_init(void)
760 kobj_set_kset_s(&fuse_subsys, fs_subsys);
761 err = subsystem_register(&fuse_subsys);
765 kobj_set_kset_s(&connections_subsys, fuse_subsys);
766 err = subsystem_register(&connections_subsys);
768 goto out_fuse_unregister;
773 subsystem_unregister(&fuse_subsys);
778 static void fuse_sysfs_cleanup(void)
780 subsystem_unregister(&connections_subsys);
781 subsystem_unregister(&fuse_subsys);
784 static int __init fuse_init(void)
788 printk("fuse init (API version %i.%i)\n",
789 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
791 INIT_LIST_HEAD(&fuse_conn_list);
792 res = fuse_fs_init();
796 res = fuse_dev_init();
800 res = fuse_sysfs_init();
802 goto err_dev_cleanup;
804 res = fuse_ctl_init();
806 goto err_sysfs_cleanup;
811 fuse_sysfs_cleanup();
820 static void __exit fuse_exit(void)
822 printk(KERN_DEBUG "fuse exit\n");
825 fuse_sysfs_cleanup();
830 module_init(fuse_init);
831 module_exit(fuse_exit);