2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 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/moduleparam.h>
18 #include <linux/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
28 static struct kmem_cache *fuse_inode_cachep;
29 struct list_head fuse_conn_list;
30 DEFINE_MUTEX(fuse_mutex);
32 static int set_global_limit(const char *val, struct kernel_param *kp);
34 unsigned max_user_bgreq;
35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36 &max_user_bgreq, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
38 MODULE_PARM_DESC(max_user_bgreq,
39 "Global limit for the maximum number of backgrounded requests an "
40 "unprivileged user can set");
42 unsigned max_user_congthresh;
43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44 &max_user_congthresh, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
46 MODULE_PARM_DESC(max_user_congthresh,
47 "Global limit for the maximum congestion threshold an "
48 "unprivileged user can set");
50 #define FUSE_SUPER_MAGIC 0x65735546
52 #define FUSE_DEFAULT_BLKSIZE 512
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
60 struct fuse_mount_data {
65 unsigned fd_present:1;
66 unsigned rootmode_present:1;
67 unsigned user_id_present:1;
68 unsigned group_id_present:1;
74 struct fuse_forget_link *fuse_alloc_forget(void)
76 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
79 static struct inode *fuse_alloc_inode(struct super_block *sb)
82 struct fuse_inode *fi;
84 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
88 fi = get_fuse_inode(inode);
96 INIT_LIST_HEAD(&fi->write_files);
97 INIT_LIST_HEAD(&fi->queued_writes);
98 INIT_LIST_HEAD(&fi->writepages);
99 init_waitqueue_head(&fi->page_waitq);
100 fi->forget = fuse_alloc_forget();
102 kmem_cache_free(fuse_inode_cachep, inode);
109 static void fuse_i_callback(struct rcu_head *head)
111 struct inode *inode = container_of(head, struct inode, i_rcu);
112 kmem_cache_free(fuse_inode_cachep, inode);
115 static void fuse_destroy_inode(struct inode *inode)
117 struct fuse_inode *fi = get_fuse_inode(inode);
118 BUG_ON(!list_empty(&fi->write_files));
119 BUG_ON(!list_empty(&fi->queued_writes));
121 call_rcu(&inode->i_rcu, fuse_i_callback);
124 static void fuse_evict_inode(struct inode *inode)
126 truncate_inode_pages(&inode->i_data, 0);
128 if (inode->i_sb->s_flags & MS_ACTIVE) {
129 struct fuse_conn *fc = get_fuse_conn(inode);
130 struct fuse_inode *fi = get_fuse_inode(inode);
131 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
136 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
139 if (*flags & MS_MANDLOCK)
146 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
147 * so that it will fit.
149 static ino_t fuse_squash_ino(u64 ino64)
151 ino_t ino = (ino_t) ino64;
152 if (sizeof(ino_t) < sizeof(u64))
153 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
157 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
160 struct fuse_conn *fc = get_fuse_conn(inode);
161 struct fuse_inode *fi = get_fuse_inode(inode);
163 fi->attr_version = ++fc->attr_version;
164 fi->i_time = attr_valid;
166 inode->i_ino = fuse_squash_ino(attr->ino);
167 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
168 set_nlink(inode, attr->nlink);
169 inode->i_uid = make_kuid(&init_user_ns, attr->uid);
170 inode->i_gid = make_kgid(&init_user_ns, attr->gid);
171 inode->i_blocks = attr->blocks;
172 inode->i_atime.tv_sec = attr->atime;
173 inode->i_atime.tv_nsec = attr->atimensec;
174 inode->i_mtime.tv_sec = attr->mtime;
175 inode->i_mtime.tv_nsec = attr->mtimensec;
176 inode->i_ctime.tv_sec = attr->ctime;
177 inode->i_ctime.tv_nsec = attr->ctimensec;
179 if (attr->blksize != 0)
180 inode->i_blkbits = ilog2(attr->blksize);
182 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
185 * Don't set the sticky bit in i_mode, unless we want the VFS
186 * to check permissions. This prevents failures due to the
187 * check in may_delete().
189 fi->orig_i_mode = inode->i_mode;
190 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
191 inode->i_mode &= ~S_ISVTX;
193 fi->orig_ino = attr->ino;
196 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
197 u64 attr_valid, u64 attr_version)
199 struct fuse_conn *fc = get_fuse_conn(inode);
200 struct fuse_inode *fi = get_fuse_inode(inode);
202 struct timespec old_mtime;
204 spin_lock(&fc->lock);
205 if ((attr_version != 0 && fi->attr_version > attr_version) ||
206 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
207 spin_unlock(&fc->lock);
211 old_mtime = inode->i_mtime;
212 fuse_change_attributes_common(inode, attr, attr_valid);
214 oldsize = inode->i_size;
215 i_size_write(inode, attr->size);
216 spin_unlock(&fc->lock);
218 if (S_ISREG(inode->i_mode)) {
221 if (oldsize != attr->size) {
222 truncate_pagecache(inode, oldsize, attr->size);
224 } else if (fc->auto_inval_data) {
225 struct timespec new_mtime = {
226 .tv_sec = attr->mtime,
227 .tv_nsec = attr->mtimensec,
231 * Auto inval mode also checks and invalidates if mtime
234 if (!timespec_equal(&old_mtime, &new_mtime))
239 invalidate_inode_pages2(inode->i_mapping);
243 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
245 inode->i_mode = attr->mode & S_IFMT;
246 inode->i_size = attr->size;
247 if (S_ISREG(inode->i_mode)) {
248 fuse_init_common(inode);
249 fuse_init_file_inode(inode);
250 } else if (S_ISDIR(inode->i_mode))
251 fuse_init_dir(inode);
252 else if (S_ISLNK(inode->i_mode))
253 fuse_init_symlink(inode);
254 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
255 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
256 fuse_init_common(inode);
257 init_special_inode(inode, inode->i_mode,
258 new_decode_dev(attr->rdev));
263 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
265 u64 nodeid = *(u64 *) _nodeidp;
266 if (get_node_id(inode) == nodeid)
272 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
274 u64 nodeid = *(u64 *) _nodeidp;
275 get_fuse_inode(inode)->nodeid = nodeid;
279 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
280 int generation, struct fuse_attr *attr,
281 u64 attr_valid, u64 attr_version)
284 struct fuse_inode *fi;
285 struct fuse_conn *fc = get_fuse_conn_super(sb);
288 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
292 if ((inode->i_state & I_NEW)) {
293 inode->i_flags |= S_NOATIME|S_NOCMTIME;
294 inode->i_generation = generation;
295 inode->i_data.backing_dev_info = &fc->bdi;
296 fuse_init_inode(inode, attr);
297 unlock_new_inode(inode);
298 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
299 /* Inode has changed type, any I/O on the old should fail */
300 make_bad_inode(inode);
305 fi = get_fuse_inode(inode);
306 spin_lock(&fc->lock);
308 spin_unlock(&fc->lock);
309 fuse_change_attributes(inode, attr, attr_valid, attr_version);
314 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
315 loff_t offset, loff_t len)
321 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
325 fuse_invalidate_attr(inode);
327 pg_start = offset >> PAGE_CACHE_SHIFT;
331 pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
332 invalidate_inode_pages2_range(inode->i_mapping,
339 static void fuse_umount_begin(struct super_block *sb)
341 fuse_abort_conn(get_fuse_conn_super(sb));
344 static void fuse_send_destroy(struct fuse_conn *fc)
346 struct fuse_req *req = fc->destroy_req;
347 if (req && fc->conn_init) {
348 fc->destroy_req = NULL;
349 req->in.h.opcode = FUSE_DESTROY;
352 fuse_request_send(fc, req);
353 fuse_put_request(fc, req);
357 static void fuse_bdi_destroy(struct fuse_conn *fc)
359 if (fc->bdi_initialized)
360 bdi_destroy(&fc->bdi);
363 void fuse_conn_kill(struct fuse_conn *fc)
365 spin_lock(&fc->lock);
369 spin_unlock(&fc->lock);
370 /* Flush all readers on this fs */
371 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
372 wake_up_all(&fc->waitq);
373 wake_up_all(&fc->blocked_waitq);
374 wake_up_all(&fc->reserved_req_waitq);
376 EXPORT_SYMBOL_GPL(fuse_conn_kill);
378 static void fuse_put_super(struct super_block *sb)
380 struct fuse_conn *fc = get_fuse_conn_super(sb);
382 fuse_send_destroy(fc);
385 mutex_lock(&fuse_mutex);
386 list_del(&fc->entry);
387 fuse_ctl_remove_conn(fc);
388 mutex_unlock(&fuse_mutex);
389 fuse_bdi_destroy(fc);
394 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
396 stbuf->f_type = FUSE_SUPER_MAGIC;
397 stbuf->f_bsize = attr->bsize;
398 stbuf->f_frsize = attr->frsize;
399 stbuf->f_blocks = attr->blocks;
400 stbuf->f_bfree = attr->bfree;
401 stbuf->f_bavail = attr->bavail;
402 stbuf->f_files = attr->files;
403 stbuf->f_ffree = attr->ffree;
404 stbuf->f_namelen = attr->namelen;
405 /* fsid is left zero */
408 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
410 struct super_block *sb = dentry->d_sb;
411 struct fuse_conn *fc = get_fuse_conn_super(sb);
412 struct fuse_req *req;
413 struct fuse_statfs_out outarg;
416 if (!fuse_allow_current_process(fc)) {
417 buf->f_type = FUSE_SUPER_MAGIC;
421 req = fuse_get_req_nopages(fc);
425 memset(&outarg, 0, sizeof(outarg));
427 req->in.h.opcode = FUSE_STATFS;
428 req->in.h.nodeid = get_node_id(dentry->d_inode);
429 req->out.numargs = 1;
430 req->out.args[0].size =
431 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
432 req->out.args[0].value = &outarg;
433 fuse_request_send(fc, req);
434 err = req->out.h.error;
436 convert_fuse_statfs(buf, &outarg.st);
437 fuse_put_request(fc, req);
446 OPT_DEFAULT_PERMISSIONS,
453 static const match_table_t tokens = {
455 {OPT_ROOTMODE, "rootmode=%o"},
456 {OPT_USER_ID, "user_id=%u"},
457 {OPT_GROUP_ID, "group_id=%u"},
458 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
459 {OPT_ALLOW_OTHER, "allow_other"},
460 {OPT_MAX_READ, "max_read=%u"},
461 {OPT_BLKSIZE, "blksize=%u"},
465 static int fuse_match_uint(substring_t *s, unsigned int *res)
468 char *buf = match_strdup(s);
470 err = kstrtouint(buf, 10, res);
476 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
479 memset(d, 0, sizeof(struct fuse_mount_data));
481 d->blksize = FUSE_DEFAULT_BLKSIZE;
483 while ((p = strsep(&opt, ",")) != NULL) {
487 substring_t args[MAX_OPT_ARGS];
491 token = match_token(p, tokens, args);
494 if (match_int(&args[0], &value))
501 if (match_octal(&args[0], &value))
503 if (!fuse_valid_type(value))
506 d->rootmode_present = 1;
510 if (fuse_match_uint(&args[0], &uv))
512 d->user_id = make_kuid(current_user_ns(), uv);
513 if (!uid_valid(d->user_id))
515 d->user_id_present = 1;
519 if (fuse_match_uint(&args[0], &uv))
521 d->group_id = make_kgid(current_user_ns(), uv);
522 if (!gid_valid(d->group_id))
524 d->group_id_present = 1;
527 case OPT_DEFAULT_PERMISSIONS:
528 d->flags |= FUSE_DEFAULT_PERMISSIONS;
531 case OPT_ALLOW_OTHER:
532 d->flags |= FUSE_ALLOW_OTHER;
536 if (match_int(&args[0], &value))
542 if (!is_bdev || match_int(&args[0], &value))
552 if (!d->fd_present || !d->rootmode_present ||
553 !d->user_id_present || !d->group_id_present)
559 static int fuse_show_options(struct seq_file *m, struct dentry *root)
561 struct super_block *sb = root->d_sb;
562 struct fuse_conn *fc = get_fuse_conn_super(sb);
564 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
565 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
566 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
567 seq_puts(m, ",default_permissions");
568 if (fc->flags & FUSE_ALLOW_OTHER)
569 seq_puts(m, ",allow_other");
570 if (fc->max_read != ~0)
571 seq_printf(m, ",max_read=%u", fc->max_read);
572 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
573 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
577 void fuse_conn_init(struct fuse_conn *fc)
579 memset(fc, 0, sizeof(*fc));
580 spin_lock_init(&fc->lock);
581 mutex_init(&fc->inst_mutex);
582 init_rwsem(&fc->killsb);
583 atomic_set(&fc->count, 1);
584 init_waitqueue_head(&fc->waitq);
585 init_waitqueue_head(&fc->blocked_waitq);
586 init_waitqueue_head(&fc->reserved_req_waitq);
587 INIT_LIST_HEAD(&fc->pending);
588 INIT_LIST_HEAD(&fc->processing);
589 INIT_LIST_HEAD(&fc->io);
590 INIT_LIST_HEAD(&fc->interrupts);
591 INIT_LIST_HEAD(&fc->bg_queue);
592 INIT_LIST_HEAD(&fc->entry);
593 fc->forget_list_tail = &fc->forget_list_head;
594 atomic_set(&fc->num_waiting, 0);
595 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
596 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
598 fc->polled_files = RB_ROOT;
602 fc->attr_version = 1;
603 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
605 EXPORT_SYMBOL_GPL(fuse_conn_init);
607 void fuse_conn_put(struct fuse_conn *fc)
609 if (atomic_dec_and_test(&fc->count)) {
611 fuse_request_free(fc->destroy_req);
612 mutex_destroy(&fc->inst_mutex);
616 EXPORT_SYMBOL_GPL(fuse_conn_put);
618 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
620 atomic_inc(&fc->count);
623 EXPORT_SYMBOL_GPL(fuse_conn_get);
625 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
627 struct fuse_attr attr;
628 memset(&attr, 0, sizeof(attr));
631 attr.ino = FUSE_ROOT_ID;
633 return fuse_iget(sb, 1, 0, &attr, 0, 0);
636 struct fuse_inode_handle {
641 static struct dentry *fuse_get_dentry(struct super_block *sb,
642 struct fuse_inode_handle *handle)
644 struct fuse_conn *fc = get_fuse_conn_super(sb);
646 struct dentry *entry;
649 if (handle->nodeid == 0)
652 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
654 struct fuse_entry_out outarg;
657 if (!fc->export_support)
662 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
664 if (err && err != -ENOENT)
671 if (get_node_id(inode) != handle->nodeid)
675 if (inode->i_generation != handle->generation)
678 entry = d_obtain_alias(inode);
679 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
680 fuse_invalidate_entry_cache(entry);
690 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
691 struct inode *parent)
693 int len = parent ? 6 : 3;
697 if (*max_len < len) {
699 return FILEID_INVALID;
702 nodeid = get_fuse_inode(inode)->nodeid;
703 generation = inode->i_generation;
705 fh[0] = (u32)(nodeid >> 32);
706 fh[1] = (u32)(nodeid & 0xffffffff);
710 nodeid = get_fuse_inode(parent)->nodeid;
711 generation = parent->i_generation;
713 fh[3] = (u32)(nodeid >> 32);
714 fh[4] = (u32)(nodeid & 0xffffffff);
719 return parent ? 0x82 : 0x81;
722 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
723 struct fid *fid, int fh_len, int fh_type)
725 struct fuse_inode_handle handle;
727 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
730 handle.nodeid = (u64) fid->raw[0] << 32;
731 handle.nodeid |= (u64) fid->raw[1];
732 handle.generation = fid->raw[2];
733 return fuse_get_dentry(sb, &handle);
736 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
737 struct fid *fid, int fh_len, int fh_type)
739 struct fuse_inode_handle parent;
741 if (fh_type != 0x82 || fh_len < 6)
744 parent.nodeid = (u64) fid->raw[3] << 32;
745 parent.nodeid |= (u64) fid->raw[4];
746 parent.generation = fid->raw[5];
747 return fuse_get_dentry(sb, &parent);
750 static struct dentry *fuse_get_parent(struct dentry *child)
752 struct inode *child_inode = child->d_inode;
753 struct fuse_conn *fc = get_fuse_conn(child_inode);
755 struct dentry *parent;
756 struct fuse_entry_out outarg;
760 if (!fc->export_support)
761 return ERR_PTR(-ESTALE);
765 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
766 &name, &outarg, &inode);
769 return ERR_PTR(-ESTALE);
773 parent = d_obtain_alias(inode);
774 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
775 fuse_invalidate_entry_cache(parent);
780 static const struct export_operations fuse_export_operations = {
781 .fh_to_dentry = fuse_fh_to_dentry,
782 .fh_to_parent = fuse_fh_to_parent,
783 .encode_fh = fuse_encode_fh,
784 .get_parent = fuse_get_parent,
787 static const struct super_operations fuse_super_operations = {
788 .alloc_inode = fuse_alloc_inode,
789 .destroy_inode = fuse_destroy_inode,
790 .evict_inode = fuse_evict_inode,
791 .drop_inode = generic_delete_inode,
792 .remount_fs = fuse_remount_fs,
793 .put_super = fuse_put_super,
794 .umount_begin = fuse_umount_begin,
795 .statfs = fuse_statfs,
796 .show_options = fuse_show_options,
799 static void sanitize_global_limit(unsigned *limit)
802 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
803 sizeof(struct fuse_req);
805 if (*limit >= 1 << 16)
806 *limit = (1 << 16) - 1;
809 static int set_global_limit(const char *val, struct kernel_param *kp)
813 rv = param_set_uint(val, kp);
817 sanitize_global_limit((unsigned *)kp->arg);
822 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
824 int cap_sys_admin = capable(CAP_SYS_ADMIN);
829 sanitize_global_limit(&max_user_bgreq);
830 sanitize_global_limit(&max_user_congthresh);
832 if (arg->max_background) {
833 fc->max_background = arg->max_background;
835 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
836 fc->max_background = max_user_bgreq;
838 if (arg->congestion_threshold) {
839 fc->congestion_threshold = arg->congestion_threshold;
841 if (!cap_sys_admin &&
842 fc->congestion_threshold > max_user_congthresh)
843 fc->congestion_threshold = max_user_congthresh;
847 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
849 struct fuse_init_out *arg = &req->misc.init_out;
851 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
854 unsigned long ra_pages;
856 process_init_limits(fc, arg);
858 if (arg->minor >= 6) {
859 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
860 if (arg->flags & FUSE_ASYNC_READ)
862 if (!(arg->flags & FUSE_POSIX_LOCKS))
864 if (arg->minor >= 17) {
865 if (!(arg->flags & FUSE_FLOCK_LOCKS))
868 if (!(arg->flags & FUSE_POSIX_LOCKS))
871 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
872 fc->atomic_o_trunc = 1;
873 if (arg->minor >= 9) {
874 /* LOOKUP has dependency on proto version */
875 if (arg->flags & FUSE_EXPORT_SUPPORT)
876 fc->export_support = 1;
878 if (arg->flags & FUSE_BIG_WRITES)
880 if (arg->flags & FUSE_DONT_MASK)
882 if (arg->flags & FUSE_AUTO_INVAL_DATA)
883 fc->auto_inval_data = 1;
884 if (arg->flags & FUSE_DO_READDIRPLUS) {
885 fc->do_readdirplus = 1;
886 if (arg->flags & FUSE_READDIRPLUS_AUTO)
887 fc->readdirplus_auto = 1;
889 if (arg->flags & FUSE_ASYNC_DIO)
892 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
897 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
898 fc->minor = arg->minor;
899 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
900 fc->max_write = max_t(unsigned, 4096, fc->max_write);
904 wake_up_all(&fc->blocked_waitq);
907 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
909 struct fuse_init_in *arg = &req->misc.init_in;
911 arg->major = FUSE_KERNEL_VERSION;
912 arg->minor = FUSE_KERNEL_MINOR_VERSION;
913 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
914 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
915 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
916 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
917 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
918 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO;
919 req->in.h.opcode = FUSE_INIT;
921 req->in.args[0].size = sizeof(*arg);
922 req->in.args[0].value = arg;
923 req->out.numargs = 1;
924 /* Variable length argument used for backward compatibility
925 with interface version < 7.5. Rest of init_out is zeroed
926 by do_get_request(), so a short reply is not a problem */
928 req->out.args[0].size = sizeof(struct fuse_init_out);
929 req->out.args[0].value = &req->misc.init_out;
930 req->end = process_init_reply;
931 fuse_request_send_background(fc, req);
934 static void fuse_free_conn(struct fuse_conn *fc)
939 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
943 fc->bdi.name = "fuse";
944 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
945 /* fuse does it's own writeback accounting */
946 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
948 err = bdi_init(&fc->bdi);
952 fc->bdi_initialized = 1;
955 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
956 MAJOR(fc->dev), MINOR(fc->dev));
958 err = bdi_register_dev(&fc->bdi, fc->dev);
965 * For a single fuse filesystem use max 1% of dirty +
966 * writeback threshold.
968 * This gives about 1M of write buffer for memory maps on a
969 * machine with 1G and 10% dirty_ratio, which should be more
972 * Privileged users can raise it by writing to
974 * /sys/class/bdi/<bdi>/max_ratio
976 bdi_set_max_ratio(&fc->bdi, 1);
981 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
983 struct fuse_conn *fc;
985 struct fuse_mount_data d;
987 struct dentry *root_dentry;
988 struct fuse_req *init_req;
990 int is_bdev = sb->s_bdev != NULL;
993 if (sb->s_flags & MS_MANDLOCK)
996 sb->s_flags &= ~MS_NOSEC;
998 if (!parse_fuse_opt((char *) data, &d, is_bdev))
1004 if (!sb_set_blocksize(sb, d.blksize))
1008 sb->s_blocksize = PAGE_CACHE_SIZE;
1009 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1011 sb->s_magic = FUSE_SUPER_MAGIC;
1012 sb->s_op = &fuse_super_operations;
1013 sb->s_maxbytes = MAX_LFS_FILESIZE;
1014 sb->s_time_gran = 1;
1015 sb->s_export_op = &fuse_export_operations;
1022 if ((file->f_op != &fuse_dev_operations) ||
1023 (file->f_cred->user_ns != &init_user_ns))
1026 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1032 fc->release = fuse_free_conn;
1034 fc->dev = sb->s_dev;
1036 err = fuse_bdi_init(fc, sb);
1040 sb->s_bdi = &fc->bdi;
1042 /* Handle umasking inside the fuse code */
1043 if (sb->s_flags & MS_POSIXACL)
1045 sb->s_flags |= MS_POSIXACL;
1047 fc->flags = d.flags;
1048 fc->user_id = d.user_id;
1049 fc->group_id = d.group_id;
1050 fc->max_read = max_t(unsigned, 4096, d.max_read);
1052 /* Used by get_root_inode() */
1056 root = fuse_get_root_inode(sb, d.rootmode);
1057 root_dentry = d_make_root(root);
1060 /* only now - we want root dentry with NULL ->d_op */
1061 sb->s_d_op = &fuse_dentry_operations;
1063 init_req = fuse_request_alloc(0);
1066 init_req->background = 1;
1069 fc->destroy_req = fuse_request_alloc(0);
1070 if (!fc->destroy_req)
1071 goto err_free_init_req;
1074 mutex_lock(&fuse_mutex);
1076 if (file->private_data)
1079 err = fuse_ctl_add_conn(fc);
1083 list_add_tail(&fc->entry, &fuse_conn_list);
1084 sb->s_root = root_dentry;
1086 file->private_data = fuse_conn_get(fc);
1087 mutex_unlock(&fuse_mutex);
1089 * atomic_dec_and_test() in fput() provides the necessary
1090 * memory barrier for file->private_data to be visible on all
1095 fuse_send_init(fc, init_req);
1100 mutex_unlock(&fuse_mutex);
1102 fuse_request_free(init_req);
1106 fuse_bdi_destroy(fc);
1114 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1115 int flags, const char *dev_name,
1118 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1121 static void fuse_kill_sb_anon(struct super_block *sb)
1123 struct fuse_conn *fc = get_fuse_conn_super(sb);
1126 down_write(&fc->killsb);
1128 up_write(&fc->killsb);
1131 kill_anon_super(sb);
1134 static struct file_system_type fuse_fs_type = {
1135 .owner = THIS_MODULE,
1137 .fs_flags = FS_HAS_SUBTYPE,
1138 .mount = fuse_mount,
1139 .kill_sb = fuse_kill_sb_anon,
1141 MODULE_ALIAS_FS("fuse");
1144 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1145 int flags, const char *dev_name,
1148 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1151 static void fuse_kill_sb_blk(struct super_block *sb)
1153 struct fuse_conn *fc = get_fuse_conn_super(sb);
1156 down_write(&fc->killsb);
1158 up_write(&fc->killsb);
1161 kill_block_super(sb);
1164 static struct file_system_type fuseblk_fs_type = {
1165 .owner = THIS_MODULE,
1167 .mount = fuse_mount_blk,
1168 .kill_sb = fuse_kill_sb_blk,
1169 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1171 MODULE_ALIAS_FS("fuseblk");
1173 static inline int register_fuseblk(void)
1175 return register_filesystem(&fuseblk_fs_type);
1178 static inline void unregister_fuseblk(void)
1180 unregister_filesystem(&fuseblk_fs_type);
1183 static inline int register_fuseblk(void)
1188 static inline void unregister_fuseblk(void)
1193 static void fuse_inode_init_once(void *foo)
1195 struct inode *inode = foo;
1197 inode_init_once(inode);
1200 static int __init fuse_fs_init(void)
1204 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1205 sizeof(struct fuse_inode),
1206 0, SLAB_HWCACHE_ALIGN,
1207 fuse_inode_init_once);
1209 if (!fuse_inode_cachep)
1212 err = register_fuseblk();
1216 err = register_filesystem(&fuse_fs_type);
1223 unregister_fuseblk();
1225 kmem_cache_destroy(fuse_inode_cachep);
1230 static void fuse_fs_cleanup(void)
1232 unregister_filesystem(&fuse_fs_type);
1233 unregister_fuseblk();
1236 * Make sure all delayed rcu free inodes are flushed before we
1240 kmem_cache_destroy(fuse_inode_cachep);
1243 static struct kobject *fuse_kobj;
1244 static struct kobject *connections_kobj;
1246 static int fuse_sysfs_init(void)
1250 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1256 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1257 if (!connections_kobj) {
1259 goto out_fuse_unregister;
1264 out_fuse_unregister:
1265 kobject_put(fuse_kobj);
1270 static void fuse_sysfs_cleanup(void)
1272 kobject_put(connections_kobj);
1273 kobject_put(fuse_kobj);
1276 static int __init fuse_init(void)
1280 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1281 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1283 INIT_LIST_HEAD(&fuse_conn_list);
1284 res = fuse_fs_init();
1288 res = fuse_dev_init();
1290 goto err_fs_cleanup;
1292 res = fuse_sysfs_init();
1294 goto err_dev_cleanup;
1296 res = fuse_ctl_init();
1298 goto err_sysfs_cleanup;
1300 sanitize_global_limit(&max_user_bgreq);
1301 sanitize_global_limit(&max_user_congthresh);
1306 fuse_sysfs_cleanup();
1315 static void __exit fuse_exit(void)
1317 printk(KERN_DEBUG "fuse exit\n");
1320 fuse_sysfs_cleanup();
1325 module_init(fuse_init);
1326 module_exit(fuse_exit);