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>
23 #include <linux/posix_acl.h>
24 #include <linux/pid_namespace.h>
26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27 MODULE_DESCRIPTION("Filesystem in Userspace");
28 MODULE_LICENSE("GPL");
30 static struct kmem_cache *fuse_inode_cachep;
31 struct list_head fuse_conn_list;
32 DEFINE_MUTEX(fuse_mutex);
34 static int set_global_limit(const char *val, const struct kernel_param *kp);
36 unsigned max_user_bgreq;
37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
38 &max_user_bgreq, 0644);
39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
40 MODULE_PARM_DESC(max_user_bgreq,
41 "Global limit for the maximum number of backgrounded requests an "
42 "unprivileged user can set");
44 unsigned max_user_congthresh;
45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
46 &max_user_congthresh, 0644);
47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
48 MODULE_PARM_DESC(max_user_congthresh,
49 "Global limit for the maximum congestion threshold an "
50 "unprivileged user can set");
52 #define FUSE_SUPER_MAGIC 0x65735546
54 #define FUSE_DEFAULT_BLKSIZE 512
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62 struct fuse_mount_data {
67 unsigned fd_present:1;
68 unsigned rootmode_present:1;
69 unsigned user_id_present:1;
70 unsigned group_id_present:1;
71 unsigned default_permissions:1;
72 unsigned allow_other:1;
77 struct fuse_forget_link *fuse_alloc_forget(void)
79 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
82 static struct inode *fuse_alloc_inode(struct super_block *sb)
85 struct fuse_inode *fi;
87 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
91 fi = get_fuse_inode(inode);
99 INIT_LIST_HEAD(&fi->write_files);
100 INIT_LIST_HEAD(&fi->queued_writes);
101 INIT_LIST_HEAD(&fi->writepages);
102 init_waitqueue_head(&fi->page_waitq);
103 mutex_init(&fi->mutex);
104 fi->forget = fuse_alloc_forget();
106 kmem_cache_free(fuse_inode_cachep, inode);
113 static void fuse_i_callback(struct rcu_head *head)
115 struct inode *inode = container_of(head, struct inode, i_rcu);
116 kmem_cache_free(fuse_inode_cachep, inode);
119 static void fuse_destroy_inode(struct inode *inode)
121 struct fuse_inode *fi = get_fuse_inode(inode);
122 BUG_ON(!list_empty(&fi->write_files));
123 BUG_ON(!list_empty(&fi->queued_writes));
124 mutex_destroy(&fi->mutex);
126 call_rcu(&inode->i_rcu, fuse_i_callback);
129 static void fuse_evict_inode(struct inode *inode)
131 truncate_inode_pages_final(&inode->i_data);
133 if (inode->i_sb->s_flags & SB_ACTIVE) {
134 struct fuse_conn *fc = get_fuse_conn(inode);
135 struct fuse_inode *fi = get_fuse_inode(inode);
136 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
141 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
144 if (*flags & SB_MANDLOCK)
151 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
152 * so that it will fit.
154 static ino_t fuse_squash_ino(u64 ino64)
156 ino_t ino = (ino_t) ino64;
157 if (sizeof(ino_t) < sizeof(u64))
158 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
162 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
165 struct fuse_conn *fc = get_fuse_conn(inode);
166 struct fuse_inode *fi = get_fuse_inode(inode);
168 fi->attr_version = ++fc->attr_version;
169 fi->i_time = attr_valid;
171 inode->i_ino = fuse_squash_ino(attr->ino);
172 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
173 set_nlink(inode, attr->nlink);
174 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
175 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
176 inode->i_blocks = attr->blocks;
179 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
180 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
181 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
183 inode->i_atime.tv_sec = attr->atime;
184 inode->i_atime.tv_nsec = attr->atimensec;
185 /* mtime from server may be stale due to local buffered write */
186 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
187 inode->i_mtime.tv_sec = attr->mtime;
188 inode->i_mtime.tv_nsec = attr->mtimensec;
189 inode->i_ctime.tv_sec = attr->ctime;
190 inode->i_ctime.tv_nsec = attr->ctimensec;
193 if (attr->blksize != 0)
194 inode->i_blkbits = ilog2(attr->blksize);
196 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
199 * Don't set the sticky bit in i_mode, unless we want the VFS
200 * to check permissions. This prevents failures due to the
201 * check in may_delete().
203 fi->orig_i_mode = inode->i_mode;
204 if (!fc->default_permissions)
205 inode->i_mode &= ~S_ISVTX;
207 fi->orig_ino = attr->ino;
210 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
211 u64 attr_valid, u64 attr_version)
213 struct fuse_conn *fc = get_fuse_conn(inode);
214 struct fuse_inode *fi = get_fuse_inode(inode);
215 bool is_wb = fc->writeback_cache;
217 struct timespec64 old_mtime;
219 spin_lock(&fc->lock);
220 if ((attr_version != 0 && fi->attr_version > attr_version) ||
221 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
222 spin_unlock(&fc->lock);
226 old_mtime = inode->i_mtime;
227 fuse_change_attributes_common(inode, attr, attr_valid);
229 oldsize = inode->i_size;
231 * In case of writeback_cache enabled, the cached writes beyond EOF
232 * extend local i_size without keeping userspace server in sync. So,
233 * attr->size coming from server can be stale. We cannot trust it.
235 if (!is_wb || !S_ISREG(inode->i_mode))
236 i_size_write(inode, attr->size);
237 spin_unlock(&fc->lock);
239 if (!is_wb && S_ISREG(inode->i_mode)) {
242 if (oldsize != attr->size) {
243 truncate_pagecache(inode, attr->size);
245 } else if (fc->auto_inval_data) {
246 struct timespec64 new_mtime = {
247 .tv_sec = attr->mtime,
248 .tv_nsec = attr->mtimensec,
252 * Auto inval mode also checks and invalidates if mtime
255 if (!timespec64_equal(&old_mtime, &new_mtime))
260 invalidate_inode_pages2(inode->i_mapping);
264 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
266 inode->i_mode = attr->mode & S_IFMT;
267 inode->i_size = attr->size;
268 inode->i_mtime.tv_sec = attr->mtime;
269 inode->i_mtime.tv_nsec = attr->mtimensec;
270 inode->i_ctime.tv_sec = attr->ctime;
271 inode->i_ctime.tv_nsec = attr->ctimensec;
272 if (S_ISREG(inode->i_mode)) {
273 fuse_init_common(inode);
274 fuse_init_file_inode(inode);
275 } else if (S_ISDIR(inode->i_mode))
276 fuse_init_dir(inode);
277 else if (S_ISLNK(inode->i_mode))
278 fuse_init_symlink(inode);
279 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
280 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
281 fuse_init_common(inode);
282 init_special_inode(inode, inode->i_mode,
283 new_decode_dev(attr->rdev));
288 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
290 u64 nodeid = *(u64 *) _nodeidp;
291 if (get_node_id(inode) == nodeid)
297 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
299 u64 nodeid = *(u64 *) _nodeidp;
300 get_fuse_inode(inode)->nodeid = nodeid;
304 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
305 int generation, struct fuse_attr *attr,
306 u64 attr_valid, u64 attr_version)
309 struct fuse_inode *fi;
310 struct fuse_conn *fc = get_fuse_conn_super(sb);
313 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
317 if ((inode->i_state & I_NEW)) {
318 inode->i_flags |= S_NOATIME;
319 if (!fc->writeback_cache || !S_ISREG(attr->mode))
320 inode->i_flags |= S_NOCMTIME;
321 inode->i_generation = generation;
322 fuse_init_inode(inode, attr);
323 unlock_new_inode(inode);
324 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
325 /* Inode has changed type, any I/O on the old should fail */
326 fuse_make_bad(inode);
331 fi = get_fuse_inode(inode);
332 spin_lock(&fc->lock);
334 spin_unlock(&fc->lock);
335 fuse_change_attributes(inode, attr, attr_valid, attr_version);
340 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
341 loff_t offset, loff_t len)
347 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
351 fuse_invalidate_attr(inode);
352 forget_all_cached_acls(inode);
354 pg_start = offset >> PAGE_SHIFT;
358 pg_end = (offset + len - 1) >> PAGE_SHIFT;
359 invalidate_inode_pages2_range(inode->i_mapping,
366 bool fuse_lock_inode(struct inode *inode)
370 if (!get_fuse_conn(inode)->parallel_dirops) {
371 mutex_lock(&get_fuse_inode(inode)->mutex);
378 void fuse_unlock_inode(struct inode *inode, bool locked)
381 mutex_unlock(&get_fuse_inode(inode)->mutex);
384 static void fuse_umount_begin(struct super_block *sb)
386 fuse_abort_conn(get_fuse_conn_super(sb), false);
389 static void fuse_send_destroy(struct fuse_conn *fc)
391 struct fuse_req *req = fc->destroy_req;
392 if (req && fc->conn_init) {
393 fc->destroy_req = NULL;
394 req->in.h.opcode = FUSE_DESTROY;
395 __set_bit(FR_FORCE, &req->flags);
396 __clear_bit(FR_BACKGROUND, &req->flags);
397 fuse_request_send(fc, req);
398 fuse_put_request(fc, req);
402 static void fuse_put_super(struct super_block *sb)
404 struct fuse_conn *fc = get_fuse_conn_super(sb);
406 mutex_lock(&fuse_mutex);
407 list_del(&fc->entry);
408 fuse_ctl_remove_conn(fc);
409 mutex_unlock(&fuse_mutex);
414 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
416 stbuf->f_type = FUSE_SUPER_MAGIC;
417 stbuf->f_bsize = attr->bsize;
418 stbuf->f_frsize = attr->frsize;
419 stbuf->f_blocks = attr->blocks;
420 stbuf->f_bfree = attr->bfree;
421 stbuf->f_bavail = attr->bavail;
422 stbuf->f_files = attr->files;
423 stbuf->f_ffree = attr->ffree;
424 stbuf->f_namelen = attr->namelen;
425 /* fsid is left zero */
428 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
430 struct super_block *sb = dentry->d_sb;
431 struct fuse_conn *fc = get_fuse_conn_super(sb);
433 struct fuse_statfs_out outarg;
436 if (!fuse_allow_current_process(fc)) {
437 buf->f_type = FUSE_SUPER_MAGIC;
441 memset(&outarg, 0, sizeof(outarg));
443 args.in.h.opcode = FUSE_STATFS;
444 args.in.h.nodeid = get_node_id(d_inode(dentry));
445 args.out.numargs = 1;
446 args.out.args[0].size = sizeof(outarg);
447 args.out.args[0].value = &outarg;
448 err = fuse_simple_request(fc, &args);
450 convert_fuse_statfs(buf, &outarg.st);
459 OPT_DEFAULT_PERMISSIONS,
466 static const match_table_t tokens = {
468 {OPT_ROOTMODE, "rootmode=%o"},
469 {OPT_USER_ID, "user_id=%u"},
470 {OPT_GROUP_ID, "group_id=%u"},
471 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
472 {OPT_ALLOW_OTHER, "allow_other"},
473 {OPT_MAX_READ, "max_read=%u"},
474 {OPT_BLKSIZE, "blksize=%u"},
478 static int fuse_match_uint(substring_t *s, unsigned int *res)
481 char *buf = match_strdup(s);
483 err = kstrtouint(buf, 10, res);
489 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
490 struct user_namespace *user_ns)
493 memset(d, 0, sizeof(struct fuse_mount_data));
495 d->blksize = FUSE_DEFAULT_BLKSIZE;
497 while ((p = strsep(&opt, ",")) != NULL) {
501 substring_t args[MAX_OPT_ARGS];
505 token = match_token(p, tokens, args);
508 if (match_int(&args[0], &value))
515 if (match_octal(&args[0], &value))
517 if (!fuse_valid_type(value))
520 d->rootmode_present = 1;
524 if (fuse_match_uint(&args[0], &uv))
526 d->user_id = make_kuid(user_ns, uv);
527 if (!uid_valid(d->user_id))
529 d->user_id_present = 1;
533 if (fuse_match_uint(&args[0], &uv))
535 d->group_id = make_kgid(user_ns, uv);
536 if (!gid_valid(d->group_id))
538 d->group_id_present = 1;
541 case OPT_DEFAULT_PERMISSIONS:
542 d->default_permissions = 1;
545 case OPT_ALLOW_OTHER:
550 if (match_int(&args[0], &value))
556 if (!is_bdev || match_int(&args[0], &value))
566 if (!d->fd_present || !d->rootmode_present ||
567 !d->user_id_present || !d->group_id_present)
573 static int fuse_show_options(struct seq_file *m, struct dentry *root)
575 struct super_block *sb = root->d_sb;
576 struct fuse_conn *fc = get_fuse_conn_super(sb);
578 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
579 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
580 if (fc->default_permissions)
581 seq_puts(m, ",default_permissions");
583 seq_puts(m, ",allow_other");
584 if (fc->max_read != ~0)
585 seq_printf(m, ",max_read=%u", fc->max_read);
586 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
587 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
591 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
593 memset(fiq, 0, sizeof(struct fuse_iqueue));
594 spin_lock_init(&fiq->lock);
595 init_waitqueue_head(&fiq->waitq);
596 INIT_LIST_HEAD(&fiq->pending);
597 INIT_LIST_HEAD(&fiq->interrupts);
598 fiq->forget_list_tail = &fiq->forget_list_head;
602 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
604 memset(fpq, 0, sizeof(struct fuse_pqueue));
605 spin_lock_init(&fpq->lock);
606 INIT_LIST_HEAD(&fpq->processing);
607 INIT_LIST_HEAD(&fpq->io);
611 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
613 memset(fc, 0, sizeof(*fc));
614 spin_lock_init(&fc->lock);
615 init_rwsem(&fc->killsb);
616 refcount_set(&fc->count, 1);
617 atomic_set(&fc->dev_count, 1);
618 init_waitqueue_head(&fc->blocked_waitq);
619 init_waitqueue_head(&fc->reserved_req_waitq);
620 fuse_iqueue_init(&fc->iq);
621 INIT_LIST_HEAD(&fc->bg_queue);
622 INIT_LIST_HEAD(&fc->entry);
623 INIT_LIST_HEAD(&fc->devices);
624 atomic_set(&fc->num_waiting, 0);
625 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
626 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
628 fc->polled_files = RB_ROOT;
632 fc->attr_version = 1;
633 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
634 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
635 fc->user_ns = get_user_ns(user_ns);
637 EXPORT_SYMBOL_GPL(fuse_conn_init);
639 void fuse_conn_put(struct fuse_conn *fc)
641 if (refcount_dec_and_test(&fc->count)) {
643 fuse_request_free(fc->destroy_req);
644 put_pid_ns(fc->pid_ns);
645 put_user_ns(fc->user_ns);
649 EXPORT_SYMBOL_GPL(fuse_conn_put);
651 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
653 refcount_inc(&fc->count);
656 EXPORT_SYMBOL_GPL(fuse_conn_get);
658 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
660 struct fuse_attr attr;
661 memset(&attr, 0, sizeof(attr));
664 attr.ino = FUSE_ROOT_ID;
666 return fuse_iget(sb, 1, 0, &attr, 0, 0);
669 struct fuse_inode_handle {
674 static struct dentry *fuse_get_dentry(struct super_block *sb,
675 struct fuse_inode_handle *handle)
677 struct fuse_conn *fc = get_fuse_conn_super(sb);
679 struct dentry *entry;
682 if (handle->nodeid == 0)
685 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
687 struct fuse_entry_out outarg;
688 const struct qstr name = QSTR_INIT(".", 1);
690 if (!fc->export_support)
693 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
695 if (err && err != -ENOENT)
702 if (get_node_id(inode) != handle->nodeid)
706 if (inode->i_generation != handle->generation)
709 entry = d_obtain_alias(inode);
710 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
711 fuse_invalidate_entry_cache(entry);
721 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
722 struct inode *parent)
724 int len = parent ? 6 : 3;
728 if (*max_len < len) {
730 return FILEID_INVALID;
733 nodeid = get_fuse_inode(inode)->nodeid;
734 generation = inode->i_generation;
736 fh[0] = (u32)(nodeid >> 32);
737 fh[1] = (u32)(nodeid & 0xffffffff);
741 nodeid = get_fuse_inode(parent)->nodeid;
742 generation = parent->i_generation;
744 fh[3] = (u32)(nodeid >> 32);
745 fh[4] = (u32)(nodeid & 0xffffffff);
750 return parent ? 0x82 : 0x81;
753 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
754 struct fid *fid, int fh_len, int fh_type)
756 struct fuse_inode_handle handle;
758 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
761 handle.nodeid = (u64) fid->raw[0] << 32;
762 handle.nodeid |= (u64) fid->raw[1];
763 handle.generation = fid->raw[2];
764 return fuse_get_dentry(sb, &handle);
767 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
768 struct fid *fid, int fh_len, int fh_type)
770 struct fuse_inode_handle parent;
772 if (fh_type != 0x82 || fh_len < 6)
775 parent.nodeid = (u64) fid->raw[3] << 32;
776 parent.nodeid |= (u64) fid->raw[4];
777 parent.generation = fid->raw[5];
778 return fuse_get_dentry(sb, &parent);
781 static struct dentry *fuse_get_parent(struct dentry *child)
783 struct inode *child_inode = d_inode(child);
784 struct fuse_conn *fc = get_fuse_conn(child_inode);
786 struct dentry *parent;
787 struct fuse_entry_out outarg;
788 const struct qstr name = QSTR_INIT("..", 2);
791 if (!fc->export_support)
792 return ERR_PTR(-ESTALE);
794 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
795 &name, &outarg, &inode);
798 return ERR_PTR(-ESTALE);
802 parent = d_obtain_alias(inode);
803 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
804 fuse_invalidate_entry_cache(parent);
809 static const struct export_operations fuse_export_operations = {
810 .fh_to_dentry = fuse_fh_to_dentry,
811 .fh_to_parent = fuse_fh_to_parent,
812 .encode_fh = fuse_encode_fh,
813 .get_parent = fuse_get_parent,
816 static const struct super_operations fuse_super_operations = {
817 .alloc_inode = fuse_alloc_inode,
818 .destroy_inode = fuse_destroy_inode,
819 .evict_inode = fuse_evict_inode,
820 .write_inode = fuse_write_inode,
821 .drop_inode = generic_delete_inode,
822 .remount_fs = fuse_remount_fs,
823 .put_super = fuse_put_super,
824 .umount_begin = fuse_umount_begin,
825 .statfs = fuse_statfs,
826 .show_options = fuse_show_options,
829 static void sanitize_global_limit(unsigned *limit)
832 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
833 sizeof(struct fuse_req);
835 if (*limit >= 1 << 16)
836 *limit = (1 << 16) - 1;
839 static int set_global_limit(const char *val, const struct kernel_param *kp)
843 rv = param_set_uint(val, kp);
847 sanitize_global_limit((unsigned *)kp->arg);
852 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
854 int cap_sys_admin = capable(CAP_SYS_ADMIN);
859 sanitize_global_limit(&max_user_bgreq);
860 sanitize_global_limit(&max_user_congthresh);
862 if (arg->max_background) {
863 fc->max_background = arg->max_background;
865 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
866 fc->max_background = max_user_bgreq;
868 if (arg->congestion_threshold) {
869 fc->congestion_threshold = arg->congestion_threshold;
871 if (!cap_sys_admin &&
872 fc->congestion_threshold > max_user_congthresh)
873 fc->congestion_threshold = max_user_congthresh;
877 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
879 struct fuse_init_out *arg = &req->misc.init_out;
881 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
884 unsigned long ra_pages;
886 process_init_limits(fc, arg);
888 if (arg->minor >= 6) {
889 ra_pages = arg->max_readahead / PAGE_SIZE;
890 if (arg->flags & FUSE_ASYNC_READ)
892 if (!(arg->flags & FUSE_POSIX_LOCKS))
894 if (arg->minor >= 17) {
895 if (!(arg->flags & FUSE_FLOCK_LOCKS))
898 if (!(arg->flags & FUSE_POSIX_LOCKS))
901 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
902 fc->atomic_o_trunc = 1;
903 if (arg->minor >= 9) {
904 /* LOOKUP has dependency on proto version */
905 if (arg->flags & FUSE_EXPORT_SUPPORT)
906 fc->export_support = 1;
908 if (arg->flags & FUSE_BIG_WRITES)
910 if (arg->flags & FUSE_DONT_MASK)
912 if (arg->flags & FUSE_AUTO_INVAL_DATA)
913 fc->auto_inval_data = 1;
914 if (arg->flags & FUSE_DO_READDIRPLUS) {
915 fc->do_readdirplus = 1;
916 if (arg->flags & FUSE_READDIRPLUS_AUTO)
917 fc->readdirplus_auto = 1;
919 if (arg->flags & FUSE_ASYNC_DIO)
921 if (arg->flags & FUSE_WRITEBACK_CACHE)
922 fc->writeback_cache = 1;
923 if (arg->flags & FUSE_PARALLEL_DIROPS)
924 fc->parallel_dirops = 1;
925 if (arg->flags & FUSE_HANDLE_KILLPRIV)
926 fc->handle_killpriv = 1;
927 if (arg->time_gran && arg->time_gran <= 1000000000)
928 fc->sb->s_time_gran = arg->time_gran;
929 if ((arg->flags & FUSE_POSIX_ACL)) {
930 fc->default_permissions = 1;
932 fc->sb->s_xattr = fuse_acl_xattr_handlers;
934 if (arg->flags & FUSE_ABORT_ERROR)
937 ra_pages = fc->max_read / PAGE_SIZE;
942 fc->sb->s_bdi->ra_pages =
943 min(fc->sb->s_bdi->ra_pages, ra_pages);
944 fc->minor = arg->minor;
945 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
946 fc->max_write = max_t(unsigned, 4096, fc->max_write);
949 fuse_set_initialized(fc);
950 wake_up_all(&fc->blocked_waitq);
953 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
955 struct fuse_init_in *arg = &req->misc.init_in;
957 arg->major = FUSE_KERNEL_VERSION;
958 arg->minor = FUSE_KERNEL_MINOR_VERSION;
959 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
960 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
961 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
962 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
963 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
964 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
965 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
966 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
968 req->in.h.opcode = FUSE_INIT;
970 req->in.args[0].size = sizeof(*arg);
971 req->in.args[0].value = arg;
972 req->out.numargs = 1;
973 /* Variable length argument used for backward compatibility
974 with interface version < 7.5. Rest of init_out is zeroed
975 by do_get_request(), so a short reply is not a problem */
977 req->out.args[0].size = sizeof(struct fuse_init_out);
978 req->out.args[0].value = &req->misc.init_out;
979 req->end = process_init_reply;
980 fuse_request_send_background(fc, req);
983 static void fuse_free_conn(struct fuse_conn *fc)
985 WARN_ON(!list_empty(&fc->devices));
989 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
997 * sb->s_bdi points to blkdev's bdi however we want to redirect
998 * it to our private bdi...
1001 sb->s_bdi = &noop_backing_dev_info;
1003 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1004 MINOR(fc->dev), suffix);
1008 sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1009 /* fuse does it's own writeback accounting */
1010 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1013 * For a single fuse filesystem use max 1% of dirty +
1014 * writeback threshold.
1016 * This gives about 1M of write buffer for memory maps on a
1017 * machine with 1G and 10% dirty_ratio, which should be more
1020 * Privileged users can raise it by writing to
1022 * /sys/class/bdi/<bdi>/max_ratio
1024 bdi_set_max_ratio(sb->s_bdi, 1);
1029 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1031 struct fuse_dev *fud;
1033 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1035 fud->fc = fuse_conn_get(fc);
1036 fuse_pqueue_init(&fud->pq);
1038 spin_lock(&fc->lock);
1039 list_add_tail(&fud->entry, &fc->devices);
1040 spin_unlock(&fc->lock);
1045 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1047 void fuse_dev_free(struct fuse_dev *fud)
1049 struct fuse_conn *fc = fud->fc;
1052 spin_lock(&fc->lock);
1053 list_del(&fud->entry);
1054 spin_unlock(&fc->lock);
1060 EXPORT_SYMBOL_GPL(fuse_dev_free);
1062 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1064 struct fuse_dev *fud;
1065 struct fuse_conn *fc;
1067 struct fuse_mount_data d;
1069 struct dentry *root_dentry;
1070 struct fuse_req *init_req;
1072 int is_bdev = sb->s_bdev != NULL;
1075 if (sb->s_flags & SB_MANDLOCK)
1078 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1080 if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1086 if (!sb_set_blocksize(sb, d.blksize))
1090 sb->s_blocksize = PAGE_SIZE;
1091 sb->s_blocksize_bits = PAGE_SHIFT;
1093 sb->s_magic = FUSE_SUPER_MAGIC;
1094 sb->s_op = &fuse_super_operations;
1095 sb->s_xattr = fuse_xattr_handlers;
1096 sb->s_maxbytes = MAX_LFS_FILESIZE;
1097 sb->s_time_gran = 1;
1098 sb->s_export_op = &fuse_export_operations;
1099 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1100 if (sb->s_user_ns != &init_user_ns)
1101 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1109 * Require mount to happen from the same user namespace which
1110 * opened /dev/fuse to prevent potential attacks.
1112 if (file->f_op != &fuse_dev_operations ||
1113 file->f_cred->user_ns != sb->s_user_ns)
1117 * If we are not in the initial user namespace posix
1118 * acls must be translated.
1120 if (sb->s_user_ns != &init_user_ns)
1121 sb->s_xattr = fuse_no_acl_xattr_handlers;
1123 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1128 fuse_conn_init(fc, sb->s_user_ns);
1129 fc->release = fuse_free_conn;
1131 fud = fuse_dev_alloc(fc);
1135 fc->dev = sb->s_dev;
1137 err = fuse_bdi_init(fc, sb);
1141 /* Handle umasking inside the fuse code */
1142 if (sb->s_flags & SB_POSIXACL)
1144 sb->s_flags |= SB_POSIXACL;
1146 fc->default_permissions = d.default_permissions;
1147 fc->allow_other = d.allow_other;
1148 fc->user_id = d.user_id;
1149 fc->group_id = d.group_id;
1150 fc->max_read = max_t(unsigned, 4096, d.max_read);
1152 /* Used by get_root_inode() */
1156 root = fuse_get_root_inode(sb, d.rootmode);
1157 sb->s_d_op = &fuse_root_dentry_operations;
1158 root_dentry = d_make_root(root);
1161 /* Root dentry doesn't have .d_revalidate */
1162 sb->s_d_op = &fuse_dentry_operations;
1164 init_req = fuse_request_alloc(0);
1167 __set_bit(FR_BACKGROUND, &init_req->flags);
1170 fc->destroy_req = fuse_request_alloc(0);
1171 if (!fc->destroy_req)
1172 goto err_free_init_req;
1175 mutex_lock(&fuse_mutex);
1177 if (file->private_data)
1180 err = fuse_ctl_add_conn(fc);
1184 list_add_tail(&fc->entry, &fuse_conn_list);
1185 sb->s_root = root_dentry;
1186 file->private_data = fud;
1187 mutex_unlock(&fuse_mutex);
1189 * atomic_dec_and_test() in fput() provides the necessary
1190 * memory barrier for file->private_data to be visible on all
1195 fuse_send_init(fc, init_req);
1200 mutex_unlock(&fuse_mutex);
1202 fuse_request_free(init_req);
1209 sb->s_fs_info = NULL;
1216 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1217 int flags, const char *dev_name,
1220 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1223 static void fuse_sb_destroy(struct super_block *sb)
1225 struct fuse_conn *fc = get_fuse_conn_super(sb);
1228 fuse_send_destroy(fc);
1230 fuse_abort_conn(fc, false);
1231 fuse_wait_aborted(fc);
1233 down_write(&fc->killsb);
1235 up_write(&fc->killsb);
1239 static void fuse_kill_sb_anon(struct super_block *sb)
1241 fuse_sb_destroy(sb);
1242 kill_anon_super(sb);
1245 static struct file_system_type fuse_fs_type = {
1246 .owner = THIS_MODULE,
1248 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1249 .mount = fuse_mount,
1250 .kill_sb = fuse_kill_sb_anon,
1252 MODULE_ALIAS_FS("fuse");
1255 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1256 int flags, const char *dev_name,
1259 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1262 static void fuse_kill_sb_blk(struct super_block *sb)
1264 fuse_sb_destroy(sb);
1265 kill_block_super(sb);
1268 static struct file_system_type fuseblk_fs_type = {
1269 .owner = THIS_MODULE,
1271 .mount = fuse_mount_blk,
1272 .kill_sb = fuse_kill_sb_blk,
1273 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1275 MODULE_ALIAS_FS("fuseblk");
1277 static inline int register_fuseblk(void)
1279 return register_filesystem(&fuseblk_fs_type);
1282 static inline void unregister_fuseblk(void)
1284 unregister_filesystem(&fuseblk_fs_type);
1287 static inline int register_fuseblk(void)
1292 static inline void unregister_fuseblk(void)
1297 static void fuse_inode_init_once(void *foo)
1299 struct inode *inode = foo;
1301 inode_init_once(inode);
1304 static int __init fuse_fs_init(void)
1308 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1309 sizeof(struct fuse_inode), 0,
1310 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1311 fuse_inode_init_once);
1313 if (!fuse_inode_cachep)
1316 err = register_fuseblk();
1320 err = register_filesystem(&fuse_fs_type);
1327 unregister_fuseblk();
1329 kmem_cache_destroy(fuse_inode_cachep);
1334 static void fuse_fs_cleanup(void)
1336 unregister_filesystem(&fuse_fs_type);
1337 unregister_fuseblk();
1340 * Make sure all delayed rcu free inodes are flushed before we
1344 kmem_cache_destroy(fuse_inode_cachep);
1347 static struct kobject *fuse_kobj;
1349 static int fuse_sysfs_init(void)
1353 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1359 err = sysfs_create_mount_point(fuse_kobj, "connections");
1361 goto out_fuse_unregister;
1365 out_fuse_unregister:
1366 kobject_put(fuse_kobj);
1371 static void fuse_sysfs_cleanup(void)
1373 sysfs_remove_mount_point(fuse_kobj, "connections");
1374 kobject_put(fuse_kobj);
1377 static int __init fuse_init(void)
1381 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1382 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1384 INIT_LIST_HEAD(&fuse_conn_list);
1385 res = fuse_fs_init();
1389 res = fuse_dev_init();
1391 goto err_fs_cleanup;
1393 res = fuse_sysfs_init();
1395 goto err_dev_cleanup;
1397 res = fuse_ctl_init();
1399 goto err_sysfs_cleanup;
1401 sanitize_global_limit(&max_user_bgreq);
1402 sanitize_global_limit(&max_user_congthresh);
1407 fuse_sysfs_cleanup();
1416 static void __exit fuse_exit(void)
1418 printk(KERN_DEBUG "fuse exit\n");
1421 fuse_sysfs_cleanup();
1426 module_init(fuse_init);
1427 module_exit(fuse_exit);