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);
60 INIT_LIST_HEAD(&fi->write_files);
61 fi->forget_req = fuse_request_alloc();
62 if (!fi->forget_req) {
63 kmem_cache_free(fuse_inode_cachep, inode);
70 static void fuse_destroy_inode(struct inode *inode)
72 struct fuse_inode *fi = get_fuse_inode(inode);
73 BUG_ON(!list_empty(&fi->write_files));
75 fuse_request_free(fi->forget_req);
76 kmem_cache_free(fuse_inode_cachep, 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 static void fuse_truncate(struct address_space *mapping, loff_t offset)
112 /* See vmtruncate() */
113 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
114 truncate_inode_pages(mapping, offset);
115 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
119 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
120 u64 attr_valid, u64 attr_version)
122 struct fuse_conn *fc = get_fuse_conn(inode);
123 struct fuse_inode *fi = get_fuse_inode(inode);
126 spin_lock(&fc->lock);
127 if (attr_version != 0 && fi->attr_version > attr_version) {
128 spin_unlock(&fc->lock);
131 fi->attr_version = ++fc->attr_version;
132 fi->i_time = attr_valid;
134 inode->i_ino = attr->ino;
135 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
136 inode->i_nlink = attr->nlink;
137 inode->i_uid = attr->uid;
138 inode->i_gid = attr->gid;
139 inode->i_blocks = attr->blocks;
140 inode->i_atime.tv_sec = attr->atime;
141 inode->i_atime.tv_nsec = attr->atimensec;
142 inode->i_mtime.tv_sec = attr->mtime;
143 inode->i_mtime.tv_nsec = attr->mtimensec;
144 inode->i_ctime.tv_sec = attr->ctime;
145 inode->i_ctime.tv_nsec = attr->ctimensec;
147 if (attr->blksize != 0)
148 inode->i_blkbits = ilog2(attr->blksize);
150 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
153 * Don't set the sticky bit in i_mode, unless we want the VFS
154 * to check permissions. This prevents failures due to the
155 * check in may_delete().
157 fi->orig_i_mode = inode->i_mode;
158 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
159 inode->i_mode &= ~S_ISVTX;
161 oldsize = inode->i_size;
162 i_size_write(inode, attr->size);
163 spin_unlock(&fc->lock);
165 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
166 if (attr->size < oldsize)
167 fuse_truncate(inode->i_mapping, attr->size);
168 invalidate_inode_pages2(inode->i_mapping);
172 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
174 inode->i_mode = attr->mode & S_IFMT;
175 inode->i_size = attr->size;
176 if (S_ISREG(inode->i_mode)) {
177 fuse_init_common(inode);
178 fuse_init_file_inode(inode);
179 } else if (S_ISDIR(inode->i_mode))
180 fuse_init_dir(inode);
181 else if (S_ISLNK(inode->i_mode))
182 fuse_init_symlink(inode);
183 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
184 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
185 fuse_init_common(inode);
186 init_special_inode(inode, inode->i_mode,
187 new_decode_dev(attr->rdev));
192 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
194 unsigned long nodeid = *(unsigned long *) _nodeidp;
195 if (get_node_id(inode) == nodeid)
201 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
203 unsigned long nodeid = *(unsigned long *) _nodeidp;
204 get_fuse_inode(inode)->nodeid = nodeid;
208 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
209 int generation, struct fuse_attr *attr,
210 u64 attr_valid, u64 attr_version)
213 struct fuse_inode *fi;
214 struct fuse_conn *fc = get_fuse_conn_super(sb);
217 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
221 if ((inode->i_state & I_NEW)) {
222 inode->i_flags |= S_NOATIME|S_NOCMTIME;
223 inode->i_generation = generation;
224 inode->i_data.backing_dev_info = &fc->bdi;
225 fuse_init_inode(inode, attr);
226 unlock_new_inode(inode);
227 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
228 /* Inode has changed type, any I/O on the old should fail */
229 make_bad_inode(inode);
234 fi = get_fuse_inode(inode);
235 spin_lock(&fc->lock);
237 spin_unlock(&fc->lock);
238 fuse_change_attributes(inode, attr, attr_valid, attr_version);
243 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
245 if (flags & MNT_FORCE)
246 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
249 static void fuse_send_destroy(struct fuse_conn *fc)
251 struct fuse_req *req = fc->destroy_req;
252 if (req && fc->conn_init) {
253 fc->destroy_req = NULL;
254 req->in.h.opcode = FUSE_DESTROY;
256 request_send(fc, req);
257 fuse_put_request(fc, req);
261 static void fuse_put_super(struct super_block *sb)
263 struct fuse_conn *fc = get_fuse_conn_super(sb);
265 fuse_send_destroy(fc);
266 spin_lock(&fc->lock);
269 spin_unlock(&fc->lock);
270 /* Flush all readers on this fs */
271 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
272 wake_up_all(&fc->waitq);
273 wake_up_all(&fc->blocked_waitq);
274 wake_up_all(&fc->reserved_req_waitq);
275 mutex_lock(&fuse_mutex);
276 list_del(&fc->entry);
277 fuse_ctl_remove_conn(fc);
278 mutex_unlock(&fuse_mutex);
282 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
284 stbuf->f_type = FUSE_SUPER_MAGIC;
285 stbuf->f_bsize = attr->bsize;
286 stbuf->f_frsize = attr->frsize;
287 stbuf->f_blocks = attr->blocks;
288 stbuf->f_bfree = attr->bfree;
289 stbuf->f_bavail = attr->bavail;
290 stbuf->f_files = attr->files;
291 stbuf->f_ffree = attr->ffree;
292 stbuf->f_namelen = attr->namelen;
293 /* fsid is left zero */
296 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
298 struct super_block *sb = dentry->d_sb;
299 struct fuse_conn *fc = get_fuse_conn_super(sb);
300 struct fuse_req *req;
301 struct fuse_statfs_out outarg;
304 if (!fuse_allow_task(fc, current)) {
305 buf->f_type = FUSE_SUPER_MAGIC;
309 req = fuse_get_req(fc);
313 memset(&outarg, 0, sizeof(outarg));
315 req->in.h.opcode = FUSE_STATFS;
316 req->in.h.nodeid = get_node_id(dentry->d_inode);
317 req->out.numargs = 1;
318 req->out.args[0].size =
319 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
320 req->out.args[0].value = &outarg;
321 request_send(fc, req);
322 err = req->out.h.error;
324 convert_fuse_statfs(buf, &outarg.st);
325 fuse_put_request(fc, req);
334 OPT_DEFAULT_PERMISSIONS,
341 static match_table_t tokens = {
343 {OPT_ROOTMODE, "rootmode=%o"},
344 {OPT_USER_ID, "user_id=%u"},
345 {OPT_GROUP_ID, "group_id=%u"},
346 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
347 {OPT_ALLOW_OTHER, "allow_other"},
348 {OPT_MAX_READ, "max_read=%u"},
349 {OPT_BLKSIZE, "blksize=%u"},
353 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
356 memset(d, 0, sizeof(struct fuse_mount_data));
360 while ((p = strsep(&opt, ",")) != NULL) {
363 substring_t args[MAX_OPT_ARGS];
367 token = match_token(p, tokens, args);
370 if (match_int(&args[0], &value))
377 if (match_octal(&args[0], &value))
379 if (!fuse_valid_type(value))
382 d->rootmode_present = 1;
386 if (match_int(&args[0], &value))
389 d->user_id_present = 1;
393 if (match_int(&args[0], &value))
396 d->group_id_present = 1;
399 case OPT_DEFAULT_PERMISSIONS:
400 d->flags |= FUSE_DEFAULT_PERMISSIONS;
403 case OPT_ALLOW_OTHER:
404 d->flags |= FUSE_ALLOW_OTHER;
408 if (match_int(&args[0], &value))
414 if (!is_bdev || match_int(&args[0], &value))
424 if (!d->fd_present || !d->rootmode_present ||
425 !d->user_id_present || !d->group_id_present)
431 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
433 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
435 seq_printf(m, ",user_id=%u", fc->user_id);
436 seq_printf(m, ",group_id=%u", fc->group_id);
437 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
438 seq_puts(m, ",default_permissions");
439 if (fc->flags & FUSE_ALLOW_OTHER)
440 seq_puts(m, ",allow_other");
441 if (fc->max_read != ~0)
442 seq_printf(m, ",max_read=%u", fc->max_read);
446 static struct fuse_conn *new_conn(void)
448 struct fuse_conn *fc;
451 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
453 spin_lock_init(&fc->lock);
454 mutex_init(&fc->inst_mutex);
455 atomic_set(&fc->count, 1);
456 init_waitqueue_head(&fc->waitq);
457 init_waitqueue_head(&fc->blocked_waitq);
458 init_waitqueue_head(&fc->reserved_req_waitq);
459 INIT_LIST_HEAD(&fc->pending);
460 INIT_LIST_HEAD(&fc->processing);
461 INIT_LIST_HEAD(&fc->io);
462 INIT_LIST_HEAD(&fc->interrupts);
463 INIT_LIST_HEAD(&fc->bg_queue);
464 atomic_set(&fc->num_waiting, 0);
465 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
466 fc->bdi.unplug_io_fn = default_unplug_io_fn;
467 err = bdi_init(&fc->bdi);
475 fc->attr_version = 1;
476 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
482 void fuse_conn_put(struct fuse_conn *fc)
484 if (atomic_dec_and_test(&fc->count)) {
486 fuse_request_free(fc->destroy_req);
487 mutex_destroy(&fc->inst_mutex);
488 bdi_destroy(&fc->bdi);
493 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
495 atomic_inc(&fc->count);
499 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
501 struct fuse_attr attr;
502 memset(&attr, 0, sizeof(attr));
505 attr.ino = FUSE_ROOT_ID;
507 return fuse_iget(sb, 1, 0, &attr, 0, 0);
510 static const struct super_operations fuse_super_operations = {
511 .alloc_inode = fuse_alloc_inode,
512 .destroy_inode = fuse_destroy_inode,
513 .clear_inode = fuse_clear_inode,
514 .drop_inode = generic_delete_inode,
515 .remount_fs = fuse_remount_fs,
516 .put_super = fuse_put_super,
517 .umount_begin = fuse_umount_begin,
518 .statfs = fuse_statfs,
519 .show_options = fuse_show_options,
522 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
524 struct fuse_init_out *arg = &req->misc.init_out;
526 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
529 unsigned long ra_pages;
531 if (arg->minor >= 6) {
532 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
533 if (arg->flags & FUSE_ASYNC_READ)
535 if (!(arg->flags & FUSE_POSIX_LOCKS))
537 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
538 fc->atomic_o_trunc = 1;
540 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
544 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
545 fc->minor = arg->minor;
546 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
549 fuse_put_request(fc, req);
551 wake_up_all(&fc->blocked_waitq);
554 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
556 struct fuse_init_in *arg = &req->misc.init_in;
558 arg->major = FUSE_KERNEL_VERSION;
559 arg->minor = FUSE_KERNEL_MINOR_VERSION;
560 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
561 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC;
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 void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
744 struct inode * inode = foo;
746 inode_init_once(inode);
749 static int __init fuse_fs_init(void)
753 err = register_filesystem(&fuse_fs_type);
757 err = register_fuseblk();
761 fuse_inode_cachep = kmem_cache_create("fuse_inode",
762 sizeof(struct fuse_inode),
763 0, SLAB_HWCACHE_ALIGN,
764 fuse_inode_init_once);
766 if (!fuse_inode_cachep)
772 unregister_fuseblk();
774 unregister_filesystem(&fuse_fs_type);
779 static void fuse_fs_cleanup(void)
781 unregister_filesystem(&fuse_fs_type);
782 unregister_fuseblk();
783 kmem_cache_destroy(fuse_inode_cachep);
786 static struct kobject *fuse_kobj;
787 static struct kobject *connections_kobj;
789 static int fuse_sysfs_init(void)
793 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
799 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
800 if (!connections_kobj) {
802 goto out_fuse_unregister;
808 kobject_put(fuse_kobj);
813 static void fuse_sysfs_cleanup(void)
815 kobject_put(connections_kobj);
816 kobject_put(fuse_kobj);
819 static int __init fuse_init(void)
823 printk("fuse init (API version %i.%i)\n",
824 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
826 INIT_LIST_HEAD(&fuse_conn_list);
827 res = fuse_fs_init();
831 res = fuse_dev_init();
835 res = fuse_sysfs_init();
837 goto err_dev_cleanup;
839 res = fuse_ctl_init();
841 goto err_sysfs_cleanup;
846 fuse_sysfs_cleanup();
855 static void __exit fuse_exit(void)
857 printk(KERN_DEBUG "fuse exit\n");
860 fuse_sysfs_cleanup();
865 module_init(fuse_init);
866 module_exit(fuse_exit);