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>
21 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
22 MODULE_DESCRIPTION("Filesystem in Userspace");
23 MODULE_LICENSE("GPL");
25 static kmem_cache_t *fuse_inode_cachep;
26 struct list_head fuse_conn_list;
27 DEFINE_MUTEX(fuse_mutex);
29 #define FUSE_SUPER_MAGIC 0x65735546
31 struct fuse_mount_data {
36 unsigned fd_present : 1;
37 unsigned rootmode_present : 1;
38 unsigned user_id_present : 1;
39 unsigned group_id_present : 1;
44 static struct inode *fuse_alloc_inode(struct super_block *sb)
47 struct fuse_inode *fi;
49 inode = kmem_cache_alloc(fuse_inode_cachep, SLAB_KERNEL);
53 fi = get_fuse_inode(inode);
57 fi->forget_req = fuse_request_alloc();
58 if (!fi->forget_req) {
59 kmem_cache_free(fuse_inode_cachep, inode);
66 static void fuse_destroy_inode(struct inode *inode)
68 struct fuse_inode *fi = get_fuse_inode(inode);
70 fuse_request_free(fi->forget_req);
71 kmem_cache_free(fuse_inode_cachep, inode);
74 static void fuse_read_inode(struct inode *inode)
79 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
80 unsigned long nodeid, u64 nlookup)
82 struct fuse_forget_in *inarg = &req->misc.forget_in;
83 inarg->nlookup = nlookup;
84 req->in.h.opcode = FUSE_FORGET;
85 req->in.h.nodeid = nodeid;
87 req->in.args[0].size = sizeof(struct fuse_forget_in);
88 req->in.args[0].value = inarg;
89 request_send_noreply(fc, req);
92 static void fuse_clear_inode(struct inode *inode)
94 if (inode->i_sb->s_flags & MS_ACTIVE) {
95 struct fuse_conn *fc = get_fuse_conn(inode);
96 struct fuse_inode *fi = get_fuse_inode(inode);
97 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
98 fi->forget_req = NULL;
102 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
104 if (*flags & MS_MANDLOCK)
110 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr)
112 if (S_ISREG(inode->i_mode) && i_size_read(inode) != attr->size)
113 invalidate_inode_pages(inode->i_mapping);
115 inode->i_ino = attr->ino;
116 inode->i_mode = (inode->i_mode & S_IFMT) + (attr->mode & 07777);
117 inode->i_nlink = attr->nlink;
118 inode->i_uid = attr->uid;
119 inode->i_gid = attr->gid;
120 i_size_write(inode, attr->size);
121 inode->i_blksize = PAGE_CACHE_SIZE;
122 inode->i_blocks = attr->blocks;
123 inode->i_atime.tv_sec = attr->atime;
124 inode->i_atime.tv_nsec = attr->atimensec;
125 inode->i_mtime.tv_sec = attr->mtime;
126 inode->i_mtime.tv_nsec = attr->mtimensec;
127 inode->i_ctime.tv_sec = attr->ctime;
128 inode->i_ctime.tv_nsec = attr->ctimensec;
131 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
133 inode->i_mode = attr->mode & S_IFMT;
134 i_size_write(inode, attr->size);
135 if (S_ISREG(inode->i_mode)) {
136 fuse_init_common(inode);
137 fuse_init_file_inode(inode);
138 } else if (S_ISDIR(inode->i_mode))
139 fuse_init_dir(inode);
140 else if (S_ISLNK(inode->i_mode))
141 fuse_init_symlink(inode);
142 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
143 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
144 fuse_init_common(inode);
145 init_special_inode(inode, inode->i_mode,
146 new_decode_dev(attr->rdev));
151 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
153 unsigned long nodeid = *(unsigned long *) _nodeidp;
154 if (get_node_id(inode) == nodeid)
160 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
162 unsigned long nodeid = *(unsigned long *) _nodeidp;
163 get_fuse_inode(inode)->nodeid = nodeid;
167 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
168 int generation, struct fuse_attr *attr)
171 struct fuse_inode *fi;
172 struct fuse_conn *fc = get_fuse_conn_super(sb);
176 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
180 if ((inode->i_state & I_NEW)) {
181 inode->i_flags |= S_NOATIME|S_NOCMTIME;
182 inode->i_generation = generation;
183 inode->i_data.backing_dev_info = &fc->bdi;
184 fuse_init_inode(inode, attr);
185 unlock_new_inode(inode);
186 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
188 /* Inode has changed type, any I/O on the old should fail */
189 make_bad_inode(inode);
195 fi = get_fuse_inode(inode);
197 fuse_change_attributes(inode, attr);
201 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
203 if (flags & MNT_FORCE)
204 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
207 static void fuse_put_super(struct super_block *sb)
209 struct fuse_conn *fc = get_fuse_conn_super(sb);
211 spin_lock(&fc->lock);
214 spin_unlock(&fc->lock);
215 /* Flush all readers on this fs */
216 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
217 wake_up_all(&fc->waitq);
218 wake_up_all(&fc->blocked_waitq);
219 mutex_lock(&fuse_mutex);
220 list_del(&fc->entry);
221 fuse_ctl_remove_conn(fc);
222 mutex_unlock(&fuse_mutex);
226 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
228 stbuf->f_type = FUSE_SUPER_MAGIC;
229 stbuf->f_bsize = attr->bsize;
230 stbuf->f_frsize = attr->frsize;
231 stbuf->f_blocks = attr->blocks;
232 stbuf->f_bfree = attr->bfree;
233 stbuf->f_bavail = attr->bavail;
234 stbuf->f_files = attr->files;
235 stbuf->f_ffree = attr->ffree;
236 stbuf->f_namelen = attr->namelen;
237 /* fsid is left zero */
240 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
242 struct super_block *sb = dentry->d_sb;
243 struct fuse_conn *fc = get_fuse_conn_super(sb);
244 struct fuse_req *req;
245 struct fuse_statfs_out outarg;
248 req = fuse_get_req(fc);
252 memset(&outarg, 0, sizeof(outarg));
254 req->in.h.opcode = FUSE_STATFS;
255 req->out.numargs = 1;
256 req->out.args[0].size =
257 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
258 req->out.args[0].value = &outarg;
259 request_send(fc, req);
260 err = req->out.h.error;
262 convert_fuse_statfs(buf, &outarg.st);
263 fuse_put_request(fc, req);
272 OPT_DEFAULT_PERMISSIONS,
278 static match_table_t tokens = {
280 {OPT_ROOTMODE, "rootmode=%o"},
281 {OPT_USER_ID, "user_id=%u"},
282 {OPT_GROUP_ID, "group_id=%u"},
283 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
284 {OPT_ALLOW_OTHER, "allow_other"},
285 {OPT_MAX_READ, "max_read=%u"},
289 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d)
292 memset(d, 0, sizeof(struct fuse_mount_data));
295 while ((p = strsep(&opt, ",")) != NULL) {
298 substring_t args[MAX_OPT_ARGS];
302 token = match_token(p, tokens, args);
305 if (match_int(&args[0], &value))
312 if (match_octal(&args[0], &value))
315 d->rootmode_present = 1;
319 if (match_int(&args[0], &value))
322 d->user_id_present = 1;
326 if (match_int(&args[0], &value))
329 d->group_id_present = 1;
332 case OPT_DEFAULT_PERMISSIONS:
333 d->flags |= FUSE_DEFAULT_PERMISSIONS;
336 case OPT_ALLOW_OTHER:
337 d->flags |= FUSE_ALLOW_OTHER;
341 if (match_int(&args[0], &value))
351 if (!d->fd_present || !d->rootmode_present ||
352 !d->user_id_present || !d->group_id_present)
358 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
360 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
362 seq_printf(m, ",user_id=%u", fc->user_id);
363 seq_printf(m, ",group_id=%u", fc->group_id);
364 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
365 seq_puts(m, ",default_permissions");
366 if (fc->flags & FUSE_ALLOW_OTHER)
367 seq_puts(m, ",allow_other");
368 if (fc->max_read != ~0)
369 seq_printf(m, ",max_read=%u", fc->max_read);
373 static struct fuse_conn *new_conn(void)
375 struct fuse_conn *fc;
377 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
379 spin_lock_init(&fc->lock);
380 atomic_set(&fc->count, 1);
381 init_waitqueue_head(&fc->waitq);
382 init_waitqueue_head(&fc->blocked_waitq);
383 INIT_LIST_HEAD(&fc->pending);
384 INIT_LIST_HEAD(&fc->processing);
385 INIT_LIST_HEAD(&fc->io);
386 INIT_LIST_HEAD(&fc->interrupts);
387 atomic_set(&fc->num_waiting, 0);
388 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
389 fc->bdi.unplug_io_fn = default_unplug_io_fn;
392 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
397 void fuse_conn_put(struct fuse_conn *fc)
399 if (atomic_dec_and_test(&fc->count))
403 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
405 atomic_inc(&fc->count);
409 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
411 struct fuse_attr attr;
412 memset(&attr, 0, sizeof(attr));
415 attr.ino = FUSE_ROOT_ID;
416 return fuse_iget(sb, 1, 0, &attr);
419 static struct super_operations fuse_super_operations = {
420 .alloc_inode = fuse_alloc_inode,
421 .destroy_inode = fuse_destroy_inode,
422 .read_inode = fuse_read_inode,
423 .clear_inode = fuse_clear_inode,
424 .remount_fs = fuse_remount_fs,
425 .put_super = fuse_put_super,
426 .umount_begin = fuse_umount_begin,
427 .statfs = fuse_statfs,
428 .show_options = fuse_show_options,
431 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
433 struct fuse_init_out *arg = &req->misc.init_out;
435 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
438 unsigned long ra_pages;
440 if (arg->minor >= 6) {
441 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
442 if (arg->flags & FUSE_ASYNC_READ)
444 if (!(arg->flags & FUSE_POSIX_LOCKS))
447 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
451 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
452 fc->minor = arg->minor;
453 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
455 fuse_put_request(fc, req);
457 wake_up_all(&fc->blocked_waitq);
460 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
462 struct fuse_init_in *arg = &req->misc.init_in;
464 arg->major = FUSE_KERNEL_VERSION;
465 arg->minor = FUSE_KERNEL_MINOR_VERSION;
466 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
467 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS;
468 req->in.h.opcode = FUSE_INIT;
470 req->in.args[0].size = sizeof(*arg);
471 req->in.args[0].value = arg;
472 req->out.numargs = 1;
473 /* Variable length arguement used for backward compatibility
474 with interface version < 7.5. Rest of init_out is zeroed
475 by do_get_request(), so a short reply is not a problem */
477 req->out.args[0].size = sizeof(struct fuse_init_out);
478 req->out.args[0].value = &req->misc.init_out;
479 req->end = process_init_reply;
480 request_send_background(fc, req);
483 static u64 conn_id(void)
489 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
491 struct fuse_conn *fc;
493 struct fuse_mount_data d;
495 struct dentry *root_dentry;
496 struct fuse_req *init_req;
499 if (sb->s_flags & MS_MANDLOCK)
502 if (!parse_fuse_opt((char *) data, &d))
505 sb->s_blocksize = PAGE_CACHE_SIZE;
506 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
507 sb->s_magic = FUSE_SUPER_MAGIC;
508 sb->s_op = &fuse_super_operations;
509 sb->s_maxbytes = MAX_LFS_FILESIZE;
515 if (file->f_op != &fuse_dev_operations)
523 fc->user_id = d.user_id;
524 fc->group_id = d.group_id;
525 fc->max_read = d.max_read;
527 /* Used by get_root_inode() */
531 root = get_root_inode(sb, d.rootmode);
535 root_dentry = d_alloc_root(root);
541 init_req = fuse_request_alloc();
545 mutex_lock(&fuse_mutex);
547 if (file->private_data)
551 err = fuse_ctl_add_conn(fc);
555 list_add_tail(&fc->entry, &fuse_conn_list);
556 sb->s_root = root_dentry;
558 file->private_data = fuse_conn_get(fc);
559 mutex_unlock(&fuse_mutex);
561 * atomic_dec_and_test() in fput() provides the necessary
562 * memory barrier for file->private_data to be visible on all
567 fuse_send_init(fc, init_req);
572 mutex_unlock(&fuse_mutex);
573 fuse_request_free(init_req);
582 static int fuse_get_sb(struct file_system_type *fs_type,
583 int flags, const char *dev_name,
584 void *raw_data, struct vfsmount *mnt)
586 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
589 static struct file_system_type fuse_fs_type = {
590 .owner = THIS_MODULE,
592 .get_sb = fuse_get_sb,
593 .kill_sb = kill_anon_super,
596 static decl_subsys(fuse, NULL, NULL);
597 static decl_subsys(connections, NULL, NULL);
599 static void fuse_inode_init_once(void *foo, kmem_cache_t *cachep,
602 struct inode * inode = foo;
604 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
605 SLAB_CTOR_CONSTRUCTOR)
606 inode_init_once(inode);
609 static int __init fuse_fs_init(void)
613 err = register_filesystem(&fuse_fs_type);
615 printk("fuse: failed to register filesystem\n");
617 fuse_inode_cachep = kmem_cache_create("fuse_inode",
618 sizeof(struct fuse_inode),
619 0, SLAB_HWCACHE_ALIGN,
620 fuse_inode_init_once, NULL);
621 if (!fuse_inode_cachep) {
622 unregister_filesystem(&fuse_fs_type);
630 static void fuse_fs_cleanup(void)
632 unregister_filesystem(&fuse_fs_type);
633 kmem_cache_destroy(fuse_inode_cachep);
636 static int fuse_sysfs_init(void)
640 kset_set_kset_s(&fuse_subsys, fs_subsys);
641 err = subsystem_register(&fuse_subsys);
645 kset_set_kset_s(&connections_subsys, fuse_subsys);
646 err = subsystem_register(&connections_subsys);
648 goto out_fuse_unregister;
653 subsystem_unregister(&fuse_subsys);
658 static void fuse_sysfs_cleanup(void)
660 subsystem_unregister(&connections_subsys);
661 subsystem_unregister(&fuse_subsys);
664 static int __init fuse_init(void)
668 printk("fuse init (API version %i.%i)\n",
669 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
671 INIT_LIST_HEAD(&fuse_conn_list);
672 res = fuse_fs_init();
676 res = fuse_dev_init();
680 res = fuse_sysfs_init();
682 goto err_dev_cleanup;
684 res = fuse_ctl_init();
686 goto err_sysfs_cleanup;
691 fuse_sysfs_cleanup();
700 static void __exit fuse_exit(void)
702 printk(KERN_DEBUG "fuse exit\n");
705 fuse_sysfs_cleanup();
710 module_init(fuse_init);
711 module_exit(fuse_exit);