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 #define FUSE_DEFAULT_BLKSIZE 512
34 struct fuse_mount_data {
39 unsigned fd_present : 1;
40 unsigned rootmode_present : 1;
41 unsigned user_id_present : 1;
42 unsigned group_id_present : 1;
48 static struct inode *fuse_alloc_inode(struct super_block *sb)
51 struct fuse_inode *fi;
53 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
57 fi = get_fuse_inode(inode);
62 INIT_LIST_HEAD(&fi->write_files);
63 fi->forget_req = fuse_request_alloc();
64 if (!fi->forget_req) {
65 kmem_cache_free(fuse_inode_cachep, inode);
72 static void fuse_destroy_inode(struct inode *inode)
74 struct fuse_inode *fi = get_fuse_inode(inode);
75 BUG_ON(!list_empty(&fi->write_files));
77 fuse_request_free(fi->forget_req);
78 kmem_cache_free(fuse_inode_cachep, 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);
121 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
122 u64 attr_valid, u64 attr_version)
124 struct fuse_conn *fc = get_fuse_conn(inode);
125 struct fuse_inode *fi = get_fuse_inode(inode);
128 spin_lock(&fc->lock);
129 if (attr_version != 0 && fi->attr_version > attr_version) {
130 spin_unlock(&fc->lock);
133 fi->attr_version = ++fc->attr_version;
134 fi->i_time = attr_valid;
136 inode->i_ino = attr->ino;
137 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
138 inode->i_nlink = attr->nlink;
139 inode->i_uid = attr->uid;
140 inode->i_gid = attr->gid;
141 inode->i_blocks = attr->blocks;
142 inode->i_atime.tv_sec = attr->atime;
143 inode->i_atime.tv_nsec = attr->atimensec;
144 inode->i_mtime.tv_sec = attr->mtime;
145 inode->i_mtime.tv_nsec = attr->mtimensec;
146 inode->i_ctime.tv_sec = attr->ctime;
147 inode->i_ctime.tv_nsec = attr->ctimensec;
149 if (attr->blksize != 0)
150 inode->i_blkbits = ilog2(attr->blksize);
152 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
155 * Don't set the sticky bit in i_mode, unless we want the VFS
156 * to check permissions. This prevents failures due to the
157 * check in may_delete().
159 fi->orig_i_mode = inode->i_mode;
160 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
161 inode->i_mode &= ~S_ISVTX;
163 oldsize = inode->i_size;
164 i_size_write(inode, attr->size);
165 spin_unlock(&fc->lock);
167 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
168 if (attr->size < oldsize)
169 fuse_truncate(inode->i_mapping, attr->size);
170 invalidate_inode_pages2(inode->i_mapping);
174 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
176 inode->i_mode = attr->mode & S_IFMT;
177 inode->i_size = attr->size;
178 if (S_ISREG(inode->i_mode)) {
179 fuse_init_common(inode);
180 fuse_init_file_inode(inode);
181 } else if (S_ISDIR(inode->i_mode))
182 fuse_init_dir(inode);
183 else if (S_ISLNK(inode->i_mode))
184 fuse_init_symlink(inode);
185 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
186 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
187 fuse_init_common(inode);
188 init_special_inode(inode, inode->i_mode,
189 new_decode_dev(attr->rdev));
194 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
196 unsigned long nodeid = *(unsigned long *) _nodeidp;
197 if (get_node_id(inode) == nodeid)
203 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
205 unsigned long nodeid = *(unsigned long *) _nodeidp;
206 get_fuse_inode(inode)->nodeid = nodeid;
210 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
211 int generation, struct fuse_attr *attr,
212 u64 attr_valid, u64 attr_version)
215 struct fuse_inode *fi;
216 struct fuse_conn *fc = get_fuse_conn_super(sb);
219 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
223 if ((inode->i_state & I_NEW)) {
224 inode->i_flags |= S_NOATIME|S_NOCMTIME;
225 inode->i_generation = generation;
226 inode->i_data.backing_dev_info = &fc->bdi;
227 fuse_init_inode(inode, attr);
228 unlock_new_inode(inode);
229 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
230 /* Inode has changed type, any I/O on the old should fail */
231 make_bad_inode(inode);
236 fi = get_fuse_inode(inode);
237 spin_lock(&fc->lock);
239 spin_unlock(&fc->lock);
240 fuse_change_attributes(inode, attr, attr_valid, attr_version);
245 static void fuse_umount_begin(struct vfsmount *vfsmnt, int flags)
247 if (flags & MNT_FORCE)
248 fuse_abort_conn(get_fuse_conn_super(vfsmnt->mnt_sb));
251 static void fuse_send_destroy(struct fuse_conn *fc)
253 struct fuse_req *req = fc->destroy_req;
254 if (req && fc->conn_init) {
255 fc->destroy_req = NULL;
256 req->in.h.opcode = FUSE_DESTROY;
258 request_send(fc, req);
259 fuse_put_request(fc, req);
263 static void fuse_put_super(struct super_block *sb)
265 struct fuse_conn *fc = get_fuse_conn_super(sb);
267 fuse_send_destroy(fc);
268 spin_lock(&fc->lock);
271 spin_unlock(&fc->lock);
272 /* Flush all readers on this fs */
273 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
274 wake_up_all(&fc->waitq);
275 wake_up_all(&fc->blocked_waitq);
276 wake_up_all(&fc->reserved_req_waitq);
277 mutex_lock(&fuse_mutex);
278 list_del(&fc->entry);
279 fuse_ctl_remove_conn(fc);
280 mutex_unlock(&fuse_mutex);
284 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
286 stbuf->f_type = FUSE_SUPER_MAGIC;
287 stbuf->f_bsize = attr->bsize;
288 stbuf->f_frsize = attr->frsize;
289 stbuf->f_blocks = attr->blocks;
290 stbuf->f_bfree = attr->bfree;
291 stbuf->f_bavail = attr->bavail;
292 stbuf->f_files = attr->files;
293 stbuf->f_ffree = attr->ffree;
294 stbuf->f_namelen = attr->namelen;
295 /* fsid is left zero */
298 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
300 struct super_block *sb = dentry->d_sb;
301 struct fuse_conn *fc = get_fuse_conn_super(sb);
302 struct fuse_req *req;
303 struct fuse_statfs_out outarg;
306 if (!fuse_allow_task(fc, current)) {
307 buf->f_type = FUSE_SUPER_MAGIC;
311 req = fuse_get_req(fc);
315 memset(&outarg, 0, sizeof(outarg));
317 req->in.h.opcode = FUSE_STATFS;
318 req->in.h.nodeid = get_node_id(dentry->d_inode);
319 req->out.numargs = 1;
320 req->out.args[0].size =
321 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
322 req->out.args[0].value = &outarg;
323 request_send(fc, req);
324 err = req->out.h.error;
326 convert_fuse_statfs(buf, &outarg.st);
327 fuse_put_request(fc, req);
336 OPT_DEFAULT_PERMISSIONS,
343 static match_table_t tokens = {
345 {OPT_ROOTMODE, "rootmode=%o"},
346 {OPT_USER_ID, "user_id=%u"},
347 {OPT_GROUP_ID, "group_id=%u"},
348 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
349 {OPT_ALLOW_OTHER, "allow_other"},
350 {OPT_MAX_READ, "max_read=%u"},
351 {OPT_BLKSIZE, "blksize=%u"},
355 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
358 memset(d, 0, sizeof(struct fuse_mount_data));
360 d->blksize = FUSE_DEFAULT_BLKSIZE;
362 while ((p = strsep(&opt, ",")) != NULL) {
365 substring_t args[MAX_OPT_ARGS];
369 token = match_token(p, tokens, args);
372 if (match_int(&args[0], &value))
379 if (match_octal(&args[0], &value))
381 if (!fuse_valid_type(value))
384 d->rootmode_present = 1;
388 if (match_int(&args[0], &value))
391 d->user_id_present = 1;
395 if (match_int(&args[0], &value))
398 d->group_id_present = 1;
401 case OPT_DEFAULT_PERMISSIONS:
402 d->flags |= FUSE_DEFAULT_PERMISSIONS;
405 case OPT_ALLOW_OTHER:
406 d->flags |= FUSE_ALLOW_OTHER;
410 if (match_int(&args[0], &value))
416 if (!is_bdev || match_int(&args[0], &value))
426 if (!d->fd_present || !d->rootmode_present ||
427 !d->user_id_present || !d->group_id_present)
433 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
435 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
437 seq_printf(m, ",user_id=%u", fc->user_id);
438 seq_printf(m, ",group_id=%u", fc->group_id);
439 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
440 seq_puts(m, ",default_permissions");
441 if (fc->flags & FUSE_ALLOW_OTHER)
442 seq_puts(m, ",allow_other");
443 if (fc->max_read != ~0)
444 seq_printf(m, ",max_read=%u", fc->max_read);
445 if (mnt->mnt_sb->s_bdev &&
446 mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
447 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
451 static struct fuse_conn *new_conn(void)
453 struct fuse_conn *fc;
456 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
458 spin_lock_init(&fc->lock);
459 mutex_init(&fc->inst_mutex);
460 atomic_set(&fc->count, 1);
461 init_waitqueue_head(&fc->waitq);
462 init_waitqueue_head(&fc->blocked_waitq);
463 init_waitqueue_head(&fc->reserved_req_waitq);
464 INIT_LIST_HEAD(&fc->pending);
465 INIT_LIST_HEAD(&fc->processing);
466 INIT_LIST_HEAD(&fc->io);
467 INIT_LIST_HEAD(&fc->interrupts);
468 INIT_LIST_HEAD(&fc->bg_queue);
469 atomic_set(&fc->num_waiting, 0);
470 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
471 fc->bdi.unplug_io_fn = default_unplug_io_fn;
472 err = bdi_init(&fc->bdi);
480 fc->attr_version = 1;
481 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
487 void fuse_conn_put(struct fuse_conn *fc)
489 if (atomic_dec_and_test(&fc->count)) {
491 fuse_request_free(fc->destroy_req);
492 mutex_destroy(&fc->inst_mutex);
493 bdi_destroy(&fc->bdi);
498 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
500 atomic_inc(&fc->count);
504 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
506 struct fuse_attr attr;
507 memset(&attr, 0, sizeof(attr));
510 attr.ino = FUSE_ROOT_ID;
512 return fuse_iget(sb, 1, 0, &attr, 0, 0);
515 static const struct super_operations fuse_super_operations = {
516 .alloc_inode = fuse_alloc_inode,
517 .destroy_inode = fuse_destroy_inode,
518 .clear_inode = fuse_clear_inode,
519 .drop_inode = generic_delete_inode,
520 .remount_fs = fuse_remount_fs,
521 .put_super = fuse_put_super,
522 .umount_begin = fuse_umount_begin,
523 .statfs = fuse_statfs,
524 .show_options = fuse_show_options,
527 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
529 struct fuse_init_out *arg = &req->misc.init_out;
531 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
534 unsigned long ra_pages;
536 if (arg->minor >= 6) {
537 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
538 if (arg->flags & FUSE_ASYNC_READ)
540 if (!(arg->flags & FUSE_POSIX_LOCKS))
542 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
543 fc->atomic_o_trunc = 1;
545 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
549 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
550 fc->minor = arg->minor;
551 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
554 fuse_put_request(fc, req);
556 wake_up_all(&fc->blocked_waitq);
559 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
561 struct fuse_init_in *arg = &req->misc.init_in;
563 arg->major = FUSE_KERNEL_VERSION;
564 arg->minor = FUSE_KERNEL_MINOR_VERSION;
565 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
566 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC;
567 req->in.h.opcode = FUSE_INIT;
569 req->in.args[0].size = sizeof(*arg);
570 req->in.args[0].value = arg;
571 req->out.numargs = 1;
572 /* Variable length arguement used for backward compatibility
573 with interface version < 7.5. Rest of init_out is zeroed
574 by do_get_request(), so a short reply is not a problem */
576 req->out.args[0].size = sizeof(struct fuse_init_out);
577 req->out.args[0].value = &req->misc.init_out;
578 req->end = process_init_reply;
579 request_send_background(fc, req);
582 static u64 conn_id(void)
588 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
590 struct fuse_conn *fc;
592 struct fuse_mount_data d;
594 struct dentry *root_dentry;
595 struct fuse_req *init_req;
597 int is_bdev = sb->s_bdev != NULL;
599 if (sb->s_flags & MS_MANDLOCK)
602 if (!parse_fuse_opt((char *) data, &d, is_bdev))
607 if (!sb_set_blocksize(sb, d.blksize))
611 sb->s_blocksize = PAGE_CACHE_SIZE;
612 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
614 sb->s_magic = FUSE_SUPER_MAGIC;
615 sb->s_op = &fuse_super_operations;
616 sb->s_maxbytes = MAX_LFS_FILESIZE;
622 if (file->f_op != &fuse_dev_operations)
630 fc->user_id = d.user_id;
631 fc->group_id = d.group_id;
632 fc->max_read = d.max_read;
634 /* Used by get_root_inode() */
638 root = get_root_inode(sb, d.rootmode);
642 root_dentry = d_alloc_root(root);
648 init_req = fuse_request_alloc();
653 fc->destroy_req = fuse_request_alloc();
654 if (!fc->destroy_req)
658 mutex_lock(&fuse_mutex);
660 if (file->private_data)
664 err = fuse_ctl_add_conn(fc);
668 list_add_tail(&fc->entry, &fuse_conn_list);
669 sb->s_root = root_dentry;
671 file->private_data = fuse_conn_get(fc);
672 mutex_unlock(&fuse_mutex);
674 * atomic_dec_and_test() in fput() provides the necessary
675 * memory barrier for file->private_data to be visible on all
680 fuse_send_init(fc, init_req);
685 mutex_unlock(&fuse_mutex);
686 fuse_request_free(init_req);
695 static int fuse_get_sb(struct file_system_type *fs_type,
696 int flags, const char *dev_name,
697 void *raw_data, struct vfsmount *mnt)
699 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
702 static struct file_system_type fuse_fs_type = {
703 .owner = THIS_MODULE,
705 .fs_flags = FS_HAS_SUBTYPE,
706 .get_sb = fuse_get_sb,
707 .kill_sb = kill_anon_super,
711 static int fuse_get_sb_blk(struct file_system_type *fs_type,
712 int flags, const char *dev_name,
713 void *raw_data, struct vfsmount *mnt)
715 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
719 static struct file_system_type fuseblk_fs_type = {
720 .owner = THIS_MODULE,
722 .get_sb = fuse_get_sb_blk,
723 .kill_sb = kill_block_super,
724 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
727 static inline int register_fuseblk(void)
729 return register_filesystem(&fuseblk_fs_type);
732 static inline void unregister_fuseblk(void)
734 unregister_filesystem(&fuseblk_fs_type);
737 static inline int register_fuseblk(void)
742 static inline void unregister_fuseblk(void)
747 static void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
749 struct inode * inode = foo;
751 inode_init_once(inode);
754 static int __init fuse_fs_init(void)
758 err = register_filesystem(&fuse_fs_type);
762 err = register_fuseblk();
766 fuse_inode_cachep = kmem_cache_create("fuse_inode",
767 sizeof(struct fuse_inode),
768 0, SLAB_HWCACHE_ALIGN,
769 fuse_inode_init_once);
771 if (!fuse_inode_cachep)
777 unregister_fuseblk();
779 unregister_filesystem(&fuse_fs_type);
784 static void fuse_fs_cleanup(void)
786 unregister_filesystem(&fuse_fs_type);
787 unregister_fuseblk();
788 kmem_cache_destroy(fuse_inode_cachep);
791 static struct kobject *fuse_kobj;
792 static struct kobject *connections_kobj;
794 static int fuse_sysfs_init(void)
798 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
804 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
805 if (!connections_kobj) {
807 goto out_fuse_unregister;
813 kobject_put(fuse_kobj);
818 static void fuse_sysfs_cleanup(void)
820 kobject_put(connections_kobj);
821 kobject_put(fuse_kobj);
824 static int __init fuse_init(void)
828 printk("fuse init (API version %i.%i)\n",
829 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
831 INIT_LIST_HEAD(&fuse_conn_list);
832 res = fuse_fs_init();
836 res = fuse_dev_init();
840 res = fuse_sysfs_init();
842 goto err_dev_cleanup;
844 res = fuse_ctl_init();
846 goto err_sysfs_cleanup;
851 fuse_sysfs_cleanup();
860 static void __exit fuse_exit(void)
862 printk(KERN_DEBUG "fuse exit\n");
865 fuse_sysfs_cleanup();
870 module_init(fuse_init);
871 module_exit(fuse_exit);