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)
138 if (*flags & MS_MANDLOCK)
145 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
146 * so that it will fit.
148 static ino_t fuse_squash_ino(u64 ino64)
150 ino_t ino = (ino_t) ino64;
151 if (sizeof(ino_t) < sizeof(u64))
152 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
156 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
159 struct fuse_conn *fc = get_fuse_conn(inode);
160 struct fuse_inode *fi = get_fuse_inode(inode);
162 fi->attr_version = ++fc->attr_version;
163 fi->i_time = attr_valid;
165 inode->i_ino = fuse_squash_ino(attr->ino);
166 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
167 set_nlink(inode, attr->nlink);
168 inode->i_uid = make_kuid(&init_user_ns, attr->uid);
169 inode->i_gid = make_kgid(&init_user_ns, attr->gid);
170 inode->i_blocks = attr->blocks;
171 inode->i_atime.tv_sec = attr->atime;
172 inode->i_atime.tv_nsec = attr->atimensec;
173 inode->i_mtime.tv_sec = attr->mtime;
174 inode->i_mtime.tv_nsec = attr->mtimensec;
175 inode->i_ctime.tv_sec = attr->ctime;
176 inode->i_ctime.tv_nsec = attr->ctimensec;
178 if (attr->blksize != 0)
179 inode->i_blkbits = ilog2(attr->blksize);
181 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
184 * Don't set the sticky bit in i_mode, unless we want the VFS
185 * to check permissions. This prevents failures due to the
186 * check in may_delete().
188 fi->orig_i_mode = inode->i_mode;
189 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
190 inode->i_mode &= ~S_ISVTX;
192 fi->orig_ino = attr->ino;
195 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
196 u64 attr_valid, u64 attr_version)
198 struct fuse_conn *fc = get_fuse_conn(inode);
199 struct fuse_inode *fi = get_fuse_inode(inode);
201 struct timespec old_mtime;
203 spin_lock(&fc->lock);
204 if ((attr_version != 0 && fi->attr_version > attr_version) ||
205 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
206 spin_unlock(&fc->lock);
210 old_mtime = inode->i_mtime;
211 fuse_change_attributes_common(inode, attr, attr_valid);
213 oldsize = inode->i_size;
214 i_size_write(inode, attr->size);
215 spin_unlock(&fc->lock);
217 if (S_ISREG(inode->i_mode)) {
220 if (oldsize != attr->size) {
221 truncate_pagecache(inode, oldsize, attr->size);
223 } else if (fc->auto_inval_data) {
224 struct timespec new_mtime = {
225 .tv_sec = attr->mtime,
226 .tv_nsec = attr->mtimensec,
230 * Auto inval mode also checks and invalidates if mtime
233 if (!timespec_equal(&old_mtime, &new_mtime))
238 invalidate_inode_pages2(inode->i_mapping);
242 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
244 inode->i_mode = attr->mode & S_IFMT;
245 inode->i_size = attr->size;
246 if (S_ISREG(inode->i_mode)) {
247 fuse_init_common(inode);
248 fuse_init_file_inode(inode);
249 } else if (S_ISDIR(inode->i_mode))
250 fuse_init_dir(inode);
251 else if (S_ISLNK(inode->i_mode))
252 fuse_init_symlink(inode);
253 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
254 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
255 fuse_init_common(inode);
256 init_special_inode(inode, inode->i_mode,
257 new_decode_dev(attr->rdev));
262 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
264 u64 nodeid = *(u64 *) _nodeidp;
265 if (get_node_id(inode) == nodeid)
271 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
273 u64 nodeid = *(u64 *) _nodeidp;
274 get_fuse_inode(inode)->nodeid = nodeid;
278 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
279 int generation, struct fuse_attr *attr,
280 u64 attr_valid, u64 attr_version)
283 struct fuse_inode *fi;
284 struct fuse_conn *fc = get_fuse_conn_super(sb);
287 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
291 if ((inode->i_state & I_NEW)) {
292 inode->i_flags |= S_NOATIME|S_NOCMTIME;
293 inode->i_generation = generation;
294 inode->i_data.backing_dev_info = &fc->bdi;
295 fuse_init_inode(inode, attr);
296 unlock_new_inode(inode);
297 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
298 /* Inode has changed type, any I/O on the old should fail */
299 make_bad_inode(inode);
304 fi = get_fuse_inode(inode);
305 spin_lock(&fc->lock);
307 spin_unlock(&fc->lock);
308 fuse_change_attributes(inode, attr, attr_valid, attr_version);
313 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
314 loff_t offset, loff_t len)
320 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
324 fuse_invalidate_attr(inode);
326 pg_start = offset >> PAGE_CACHE_SHIFT;
330 pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
331 invalidate_inode_pages2_range(inode->i_mapping,
338 static void fuse_umount_begin(struct super_block *sb)
340 fuse_abort_conn(get_fuse_conn_super(sb));
343 static void fuse_send_destroy(struct fuse_conn *fc)
345 struct fuse_req *req = fc->destroy_req;
346 if (req && fc->conn_init) {
347 fc->destroy_req = NULL;
348 req->in.h.opcode = FUSE_DESTROY;
351 fuse_request_send(fc, req);
352 fuse_put_request(fc, req);
356 static void fuse_bdi_destroy(struct fuse_conn *fc)
358 if (fc->bdi_initialized)
359 bdi_destroy(&fc->bdi);
362 void fuse_conn_kill(struct fuse_conn *fc)
364 spin_lock(&fc->lock);
368 spin_unlock(&fc->lock);
369 /* Flush all readers on this fs */
370 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
371 wake_up_all(&fc->waitq);
372 wake_up_all(&fc->blocked_waitq);
373 wake_up_all(&fc->reserved_req_waitq);
375 EXPORT_SYMBOL_GPL(fuse_conn_kill);
377 static void fuse_put_super(struct super_block *sb)
379 struct fuse_conn *fc = get_fuse_conn_super(sb);
381 fuse_send_destroy(fc);
384 mutex_lock(&fuse_mutex);
385 list_del(&fc->entry);
386 fuse_ctl_remove_conn(fc);
387 mutex_unlock(&fuse_mutex);
388 fuse_bdi_destroy(fc);
393 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
395 stbuf->f_type = FUSE_SUPER_MAGIC;
396 stbuf->f_bsize = attr->bsize;
397 stbuf->f_frsize = attr->frsize;
398 stbuf->f_blocks = attr->blocks;
399 stbuf->f_bfree = attr->bfree;
400 stbuf->f_bavail = attr->bavail;
401 stbuf->f_files = attr->files;
402 stbuf->f_ffree = attr->ffree;
403 stbuf->f_namelen = attr->namelen;
404 /* fsid is left zero */
407 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
409 struct super_block *sb = dentry->d_sb;
410 struct fuse_conn *fc = get_fuse_conn_super(sb);
411 struct fuse_req *req;
412 struct fuse_statfs_out outarg;
415 if (!fuse_allow_current_process(fc)) {
416 buf->f_type = FUSE_SUPER_MAGIC;
420 req = fuse_get_req_nopages(fc);
424 memset(&outarg, 0, sizeof(outarg));
426 req->in.h.opcode = FUSE_STATFS;
427 req->in.h.nodeid = get_node_id(dentry->d_inode);
428 req->out.numargs = 1;
429 req->out.args[0].size =
430 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
431 req->out.args[0].value = &outarg;
432 fuse_request_send(fc, req);
433 err = req->out.h.error;
435 convert_fuse_statfs(buf, &outarg.st);
436 fuse_put_request(fc, req);
445 OPT_DEFAULT_PERMISSIONS,
452 static const match_table_t tokens = {
454 {OPT_ROOTMODE, "rootmode=%o"},
455 {OPT_USER_ID, "user_id=%u"},
456 {OPT_GROUP_ID, "group_id=%u"},
457 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
458 {OPT_ALLOW_OTHER, "allow_other"},
459 {OPT_MAX_READ, "max_read=%u"},
460 {OPT_BLKSIZE, "blksize=%u"},
464 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
467 memset(d, 0, sizeof(struct fuse_mount_data));
469 d->blksize = FUSE_DEFAULT_BLKSIZE;
471 while ((p = strsep(&opt, ",")) != NULL) {
474 substring_t args[MAX_OPT_ARGS];
478 token = match_token(p, tokens, args);
481 if (match_int(&args[0], &value))
488 if (match_octal(&args[0], &value))
490 if (!fuse_valid_type(value))
493 d->rootmode_present = 1;
497 if (match_int(&args[0], &value))
499 d->user_id = make_kuid(current_user_ns(), value);
500 if (!uid_valid(d->user_id))
502 d->user_id_present = 1;
506 if (match_int(&args[0], &value))
508 d->group_id = make_kgid(current_user_ns(), value);
509 if (!gid_valid(d->group_id))
511 d->group_id_present = 1;
514 case OPT_DEFAULT_PERMISSIONS:
515 d->flags |= FUSE_DEFAULT_PERMISSIONS;
518 case OPT_ALLOW_OTHER:
519 d->flags |= FUSE_ALLOW_OTHER;
523 if (match_int(&args[0], &value))
529 if (!is_bdev || match_int(&args[0], &value))
539 if (!d->fd_present || !d->rootmode_present ||
540 !d->user_id_present || !d->group_id_present)
546 static int fuse_show_options(struct seq_file *m, struct dentry *root)
548 struct super_block *sb = root->d_sb;
549 struct fuse_conn *fc = get_fuse_conn_super(sb);
551 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
552 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
553 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
554 seq_puts(m, ",default_permissions");
555 if (fc->flags & FUSE_ALLOW_OTHER)
556 seq_puts(m, ",allow_other");
557 if (fc->max_read != ~0)
558 seq_printf(m, ",max_read=%u", fc->max_read);
559 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
560 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
564 void fuse_conn_init(struct fuse_conn *fc)
566 memset(fc, 0, sizeof(*fc));
567 spin_lock_init(&fc->lock);
568 mutex_init(&fc->inst_mutex);
569 init_rwsem(&fc->killsb);
570 atomic_set(&fc->count, 1);
571 init_waitqueue_head(&fc->waitq);
572 init_waitqueue_head(&fc->blocked_waitq);
573 init_waitqueue_head(&fc->reserved_req_waitq);
574 INIT_LIST_HEAD(&fc->pending);
575 INIT_LIST_HEAD(&fc->processing);
576 INIT_LIST_HEAD(&fc->io);
577 INIT_LIST_HEAD(&fc->interrupts);
578 INIT_LIST_HEAD(&fc->bg_queue);
579 INIT_LIST_HEAD(&fc->entry);
580 fc->forget_list_tail = &fc->forget_list_head;
581 atomic_set(&fc->num_waiting, 0);
582 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
583 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
585 fc->polled_files = RB_ROOT;
589 fc->attr_version = 1;
590 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
592 EXPORT_SYMBOL_GPL(fuse_conn_init);
594 void fuse_conn_put(struct fuse_conn *fc)
596 if (atomic_dec_and_test(&fc->count)) {
598 fuse_request_free(fc->destroy_req);
599 mutex_destroy(&fc->inst_mutex);
603 EXPORT_SYMBOL_GPL(fuse_conn_put);
605 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
607 atomic_inc(&fc->count);
610 EXPORT_SYMBOL_GPL(fuse_conn_get);
612 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
614 struct fuse_attr attr;
615 memset(&attr, 0, sizeof(attr));
618 attr.ino = FUSE_ROOT_ID;
620 return fuse_iget(sb, 1, 0, &attr, 0, 0);
623 struct fuse_inode_handle {
628 static struct dentry *fuse_get_dentry(struct super_block *sb,
629 struct fuse_inode_handle *handle)
631 struct fuse_conn *fc = get_fuse_conn_super(sb);
633 struct dentry *entry;
636 if (handle->nodeid == 0)
639 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
641 struct fuse_entry_out outarg;
644 if (!fc->export_support)
649 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
651 if (err && err != -ENOENT)
658 if (get_node_id(inode) != handle->nodeid)
662 if (inode->i_generation != handle->generation)
665 entry = d_obtain_alias(inode);
666 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
667 fuse_invalidate_entry_cache(entry);
677 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
678 struct inode *parent)
680 int len = parent ? 6 : 3;
684 if (*max_len < len) {
686 return FILEID_INVALID;
689 nodeid = get_fuse_inode(inode)->nodeid;
690 generation = inode->i_generation;
692 fh[0] = (u32)(nodeid >> 32);
693 fh[1] = (u32)(nodeid & 0xffffffff);
697 nodeid = get_fuse_inode(parent)->nodeid;
698 generation = parent->i_generation;
700 fh[3] = (u32)(nodeid >> 32);
701 fh[4] = (u32)(nodeid & 0xffffffff);
706 return parent ? 0x82 : 0x81;
709 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
710 struct fid *fid, int fh_len, int fh_type)
712 struct fuse_inode_handle handle;
714 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
717 handle.nodeid = (u64) fid->raw[0] << 32;
718 handle.nodeid |= (u64) fid->raw[1];
719 handle.generation = fid->raw[2];
720 return fuse_get_dentry(sb, &handle);
723 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
724 struct fid *fid, int fh_len, int fh_type)
726 struct fuse_inode_handle parent;
728 if (fh_type != 0x82 || fh_len < 6)
731 parent.nodeid = (u64) fid->raw[3] << 32;
732 parent.nodeid |= (u64) fid->raw[4];
733 parent.generation = fid->raw[5];
734 return fuse_get_dentry(sb, &parent);
737 static struct dentry *fuse_get_parent(struct dentry *child)
739 struct inode *child_inode = child->d_inode;
740 struct fuse_conn *fc = get_fuse_conn(child_inode);
742 struct dentry *parent;
743 struct fuse_entry_out outarg;
747 if (!fc->export_support)
748 return ERR_PTR(-ESTALE);
752 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
753 &name, &outarg, &inode);
756 return ERR_PTR(-ESTALE);
760 parent = d_obtain_alias(inode);
761 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
762 fuse_invalidate_entry_cache(parent);
767 static const struct export_operations fuse_export_operations = {
768 .fh_to_dentry = fuse_fh_to_dentry,
769 .fh_to_parent = fuse_fh_to_parent,
770 .encode_fh = fuse_encode_fh,
771 .get_parent = fuse_get_parent,
774 static const struct super_operations fuse_super_operations = {
775 .alloc_inode = fuse_alloc_inode,
776 .destroy_inode = fuse_destroy_inode,
777 .evict_inode = fuse_evict_inode,
778 .drop_inode = generic_delete_inode,
779 .remount_fs = fuse_remount_fs,
780 .put_super = fuse_put_super,
781 .umount_begin = fuse_umount_begin,
782 .statfs = fuse_statfs,
783 .show_options = fuse_show_options,
786 static void sanitize_global_limit(unsigned *limit)
789 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
790 sizeof(struct fuse_req);
792 if (*limit >= 1 << 16)
793 *limit = (1 << 16) - 1;
796 static int set_global_limit(const char *val, struct kernel_param *kp)
800 rv = param_set_uint(val, kp);
804 sanitize_global_limit((unsigned *)kp->arg);
809 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
811 int cap_sys_admin = capable(CAP_SYS_ADMIN);
816 sanitize_global_limit(&max_user_bgreq);
817 sanitize_global_limit(&max_user_congthresh);
819 if (arg->max_background) {
820 fc->max_background = arg->max_background;
822 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
823 fc->max_background = max_user_bgreq;
825 if (arg->congestion_threshold) {
826 fc->congestion_threshold = arg->congestion_threshold;
828 if (!cap_sys_admin &&
829 fc->congestion_threshold > max_user_congthresh)
830 fc->congestion_threshold = max_user_congthresh;
834 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
836 struct fuse_init_out *arg = &req->misc.init_out;
838 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
841 unsigned long ra_pages;
843 process_init_limits(fc, arg);
845 if (arg->minor >= 6) {
846 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
847 if (arg->flags & FUSE_ASYNC_READ)
849 if (!(arg->flags & FUSE_POSIX_LOCKS))
851 if (arg->minor >= 17) {
852 if (!(arg->flags & FUSE_FLOCK_LOCKS))
855 if (!(arg->flags & FUSE_POSIX_LOCKS))
858 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
859 fc->atomic_o_trunc = 1;
860 if (arg->minor >= 9) {
861 /* LOOKUP has dependency on proto version */
862 if (arg->flags & FUSE_EXPORT_SUPPORT)
863 fc->export_support = 1;
865 if (arg->flags & FUSE_BIG_WRITES)
867 if (arg->flags & FUSE_DONT_MASK)
869 if (arg->flags & FUSE_AUTO_INVAL_DATA)
870 fc->auto_inval_data = 1;
871 if (arg->flags & FUSE_DO_READDIRPLUS) {
872 fc->do_readdirplus = 1;
873 if (arg->flags & FUSE_READDIRPLUS_AUTO)
874 fc->readdirplus_auto = 1;
876 if (arg->flags & FUSE_ASYNC_DIO)
879 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
884 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
885 fc->minor = arg->minor;
886 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
887 fc->max_write = max_t(unsigned, 4096, fc->max_write);
891 wake_up_all(&fc->blocked_waitq);
894 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
896 struct fuse_init_in *arg = &req->misc.init_in;
898 arg->major = FUSE_KERNEL_VERSION;
899 arg->minor = FUSE_KERNEL_MINOR_VERSION;
900 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
901 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
902 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
903 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
904 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
905 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO;
906 req->in.h.opcode = FUSE_INIT;
908 req->in.args[0].size = sizeof(*arg);
909 req->in.args[0].value = arg;
910 req->out.numargs = 1;
911 /* Variable length argument used for backward compatibility
912 with interface version < 7.5. Rest of init_out is zeroed
913 by do_get_request(), so a short reply is not a problem */
915 req->out.args[0].size = sizeof(struct fuse_init_out);
916 req->out.args[0].value = &req->misc.init_out;
917 req->end = process_init_reply;
918 fuse_request_send_background(fc, req);
921 static void fuse_free_conn(struct fuse_conn *fc)
926 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
930 fc->bdi.name = "fuse";
931 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
932 /* fuse does it's own writeback accounting */
933 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
935 err = bdi_init(&fc->bdi);
939 fc->bdi_initialized = 1;
942 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
943 MAJOR(fc->dev), MINOR(fc->dev));
945 err = bdi_register_dev(&fc->bdi, fc->dev);
952 * For a single fuse filesystem use max 1% of dirty +
953 * writeback threshold.
955 * This gives about 1M of write buffer for memory maps on a
956 * machine with 1G and 10% dirty_ratio, which should be more
959 * Privileged users can raise it by writing to
961 * /sys/class/bdi/<bdi>/max_ratio
963 bdi_set_max_ratio(&fc->bdi, 1);
968 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
970 struct fuse_conn *fc;
972 struct fuse_mount_data d;
974 struct dentry *root_dentry;
975 struct fuse_req *init_req;
977 int is_bdev = sb->s_bdev != NULL;
980 if (sb->s_flags & MS_MANDLOCK)
983 sb->s_flags &= ~MS_NOSEC;
985 if (!parse_fuse_opt((char *) data, &d, is_bdev))
991 if (!sb_set_blocksize(sb, d.blksize))
995 sb->s_blocksize = PAGE_CACHE_SIZE;
996 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
998 sb->s_magic = FUSE_SUPER_MAGIC;
999 sb->s_op = &fuse_super_operations;
1000 sb->s_maxbytes = MAX_LFS_FILESIZE;
1001 sb->s_time_gran = 1;
1002 sb->s_export_op = &fuse_export_operations;
1009 if ((file->f_op != &fuse_dev_operations) ||
1010 (file->f_cred->user_ns != &init_user_ns))
1013 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1020 fc->dev = sb->s_dev;
1022 err = fuse_bdi_init(fc, sb);
1026 sb->s_bdi = &fc->bdi;
1028 /* Handle umasking inside the fuse code */
1029 if (sb->s_flags & MS_POSIXACL)
1031 sb->s_flags |= MS_POSIXACL;
1033 fc->release = fuse_free_conn;
1034 fc->flags = d.flags;
1035 fc->user_id = d.user_id;
1036 fc->group_id = d.group_id;
1037 fc->max_read = max_t(unsigned, 4096, d.max_read);
1039 /* Used by get_root_inode() */
1043 root = fuse_get_root_inode(sb, d.rootmode);
1044 root_dentry = d_make_root(root);
1047 /* only now - we want root dentry with NULL ->d_op */
1048 sb->s_d_op = &fuse_dentry_operations;
1050 init_req = fuse_request_alloc(0);
1053 init_req->background = 1;
1056 fc->destroy_req = fuse_request_alloc(0);
1057 if (!fc->destroy_req)
1058 goto err_free_init_req;
1061 mutex_lock(&fuse_mutex);
1063 if (file->private_data)
1066 err = fuse_ctl_add_conn(fc);
1070 list_add_tail(&fc->entry, &fuse_conn_list);
1071 sb->s_root = root_dentry;
1073 file->private_data = fuse_conn_get(fc);
1074 mutex_unlock(&fuse_mutex);
1076 * atomic_dec_and_test() in fput() provides the necessary
1077 * memory barrier for file->private_data to be visible on all
1082 fuse_send_init(fc, init_req);
1087 mutex_unlock(&fuse_mutex);
1089 fuse_request_free(init_req);
1093 fuse_bdi_destroy(fc);
1101 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1102 int flags, const char *dev_name,
1105 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1108 static void fuse_kill_sb_anon(struct super_block *sb)
1110 struct fuse_conn *fc = get_fuse_conn_super(sb);
1113 down_write(&fc->killsb);
1115 up_write(&fc->killsb);
1118 kill_anon_super(sb);
1121 static struct file_system_type fuse_fs_type = {
1122 .owner = THIS_MODULE,
1124 .fs_flags = FS_HAS_SUBTYPE,
1125 .mount = fuse_mount,
1126 .kill_sb = fuse_kill_sb_anon,
1128 MODULE_ALIAS_FS("fuse");
1131 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1132 int flags, const char *dev_name,
1135 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1138 static void fuse_kill_sb_blk(struct super_block *sb)
1140 struct fuse_conn *fc = get_fuse_conn_super(sb);
1143 down_write(&fc->killsb);
1145 up_write(&fc->killsb);
1148 kill_block_super(sb);
1151 static struct file_system_type fuseblk_fs_type = {
1152 .owner = THIS_MODULE,
1154 .mount = fuse_mount_blk,
1155 .kill_sb = fuse_kill_sb_blk,
1156 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1158 MODULE_ALIAS_FS("fuseblk");
1160 static inline int register_fuseblk(void)
1162 return register_filesystem(&fuseblk_fs_type);
1165 static inline void unregister_fuseblk(void)
1167 unregister_filesystem(&fuseblk_fs_type);
1170 static inline int register_fuseblk(void)
1175 static inline void unregister_fuseblk(void)
1180 static void fuse_inode_init_once(void *foo)
1182 struct inode *inode = foo;
1184 inode_init_once(inode);
1187 static int __init fuse_fs_init(void)
1191 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1192 sizeof(struct fuse_inode),
1193 0, SLAB_HWCACHE_ALIGN,
1194 fuse_inode_init_once);
1196 if (!fuse_inode_cachep)
1199 err = register_fuseblk();
1203 err = register_filesystem(&fuse_fs_type);
1210 unregister_fuseblk();
1212 kmem_cache_destroy(fuse_inode_cachep);
1217 static void fuse_fs_cleanup(void)
1219 unregister_filesystem(&fuse_fs_type);
1220 unregister_fuseblk();
1223 * Make sure all delayed rcu free inodes are flushed before we
1227 kmem_cache_destroy(fuse_inode_cachep);
1230 static struct kobject *fuse_kobj;
1231 static struct kobject *connections_kobj;
1233 static int fuse_sysfs_init(void)
1237 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1243 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1244 if (!connections_kobj) {
1246 goto out_fuse_unregister;
1251 out_fuse_unregister:
1252 kobject_put(fuse_kobj);
1257 static void fuse_sysfs_cleanup(void)
1259 kobject_put(connections_kobj);
1260 kobject_put(fuse_kobj);
1263 static int __init fuse_init(void)
1267 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1268 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1270 INIT_LIST_HEAD(&fuse_conn_list);
1271 res = fuse_fs_init();
1275 res = fuse_dev_init();
1277 goto err_fs_cleanup;
1279 res = fuse_sysfs_init();
1281 goto err_dev_cleanup;
1283 res = fuse_ctl_init();
1285 goto err_sysfs_cleanup;
1287 sanitize_global_limit(&max_user_bgreq);
1288 sanitize_global_limit(&max_user_congthresh);
1293 fuse_sysfs_cleanup();
1302 static void __exit fuse_exit(void)
1304 printk(KERN_DEBUG "fuse exit\n");
1307 fuse_sysfs_cleanup();
1312 module_init(fuse_init);
1313 module_exit(fuse_exit);