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/fs_context.h>
19 #include <linux/fs_parser.h>
20 #include <linux/statfs.h>
21 #include <linux/random.h>
22 #include <linux/sched.h>
23 #include <linux/exportfs.h>
24 #include <linux/posix_acl.h>
25 #include <linux/pid_namespace.h>
27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28 MODULE_DESCRIPTION("Filesystem in Userspace");
29 MODULE_LICENSE("GPL");
31 static struct kmem_cache *fuse_inode_cachep;
32 struct list_head fuse_conn_list;
33 DEFINE_MUTEX(fuse_mutex);
35 static int set_global_limit(const char *val, const struct kernel_param *kp);
37 unsigned max_user_bgreq;
38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
39 &max_user_bgreq, 0644);
40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
41 MODULE_PARM_DESC(max_user_bgreq,
42 "Global limit for the maximum number of backgrounded requests an "
43 "unprivileged user can set");
45 unsigned max_user_congthresh;
46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
47 &max_user_congthresh, 0644);
48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
49 MODULE_PARM_DESC(max_user_congthresh,
50 "Global limit for the maximum congestion threshold an "
51 "unprivileged user can set");
53 #define FUSE_SUPER_MAGIC 0x65735546
55 #define FUSE_DEFAULT_BLKSIZE 512
57 /** Maximum number of outstanding background requests */
58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
60 /** Congestion starts at 75% of maximum */
61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
64 static struct file_system_type fuseblk_fs_type;
67 struct fuse_forget_link *fuse_alloc_forget(void)
69 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
72 static struct inode *fuse_alloc_inode(struct super_block *sb)
74 struct fuse_inode *fi;
76 fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
87 mutex_init(&fi->mutex);
88 init_rwsem(&fi->i_mmap_sem);
89 spin_lock_init(&fi->lock);
90 fi->forget = fuse_alloc_forget();
94 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
102 kmem_cache_free(fuse_inode_cachep, fi);
106 static void fuse_free_inode(struct inode *inode)
108 struct fuse_inode *fi = get_fuse_inode(inode);
110 mutex_destroy(&fi->mutex);
112 #ifdef CONFIG_FUSE_DAX
115 kmem_cache_free(fuse_inode_cachep, fi);
118 static void fuse_evict_inode(struct inode *inode)
120 struct fuse_inode *fi = get_fuse_inode(inode);
122 truncate_inode_pages_final(&inode->i_data);
124 if (inode->i_sb->s_flags & SB_ACTIVE) {
125 struct fuse_conn *fc = get_fuse_conn(inode);
127 if (FUSE_IS_DAX(inode))
128 fuse_dax_inode_cleanup(inode);
130 fuse_queue_forget(fc, fi->forget, fi->nodeid,
135 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
136 WARN_ON(!list_empty(&fi->write_files));
137 WARN_ON(!list_empty(&fi->queued_writes));
141 static int fuse_reconfigure(struct fs_context *fsc)
143 struct super_block *sb = fsc->root->d_sb;
146 if (fsc->sb_flags & SB_MANDLOCK)
153 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
154 * so that it will fit.
156 static ino_t fuse_squash_ino(u64 ino64)
158 ino_t ino = (ino_t) ino64;
159 if (sizeof(ino_t) < sizeof(u64))
160 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
164 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
167 struct fuse_conn *fc = get_fuse_conn(inode);
168 struct fuse_inode *fi = get_fuse_inode(inode);
170 lockdep_assert_held(&fi->lock);
172 fi->attr_version = atomic64_inc_return(&fc->attr_version);
173 fi->i_time = attr_valid;
174 WRITE_ONCE(fi->inval_mask, 0);
176 inode->i_ino = fuse_squash_ino(attr->ino);
177 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
178 set_nlink(inode, attr->nlink);
179 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
180 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
181 inode->i_blocks = attr->blocks;
182 inode->i_atime.tv_sec = attr->atime;
183 inode->i_atime.tv_nsec = attr->atimensec;
184 /* mtime from server may be stale due to local buffered write */
185 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
186 inode->i_mtime.tv_sec = attr->mtime;
187 inode->i_mtime.tv_nsec = attr->mtimensec;
188 inode->i_ctime.tv_sec = attr->ctime;
189 inode->i_ctime.tv_nsec = attr->ctimensec;
192 if (attr->blksize != 0)
193 inode->i_blkbits = ilog2(attr->blksize);
195 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
198 * Don't set the sticky bit in i_mode, unless we want the VFS
199 * to check permissions. This prevents failures due to the
200 * check in may_delete().
202 fi->orig_i_mode = inode->i_mode;
203 if (!fc->default_permissions)
204 inode->i_mode &= ~S_ISVTX;
206 fi->orig_ino = attr->ino;
209 * We are refreshing inode data and it is possible that another
210 * client set suid/sgid or security.capability xattr. So clear
211 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
212 * was set or if security.capability xattr was set. But we don't
213 * know if security.capability has been set or not. So clear it
214 * anyway. Its less efficient but should be safe.
216 inode->i_flags &= ~S_NOSEC;
219 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
220 u64 attr_valid, u64 attr_version)
222 struct fuse_conn *fc = get_fuse_conn(inode);
223 struct fuse_inode *fi = get_fuse_inode(inode);
224 bool is_wb = fc->writeback_cache;
226 struct timespec64 old_mtime;
228 spin_lock(&fi->lock);
229 if ((attr_version != 0 && fi->attr_version > attr_version) ||
230 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
231 spin_unlock(&fi->lock);
235 old_mtime = inode->i_mtime;
236 fuse_change_attributes_common(inode, attr, attr_valid);
238 oldsize = inode->i_size;
240 * In case of writeback_cache enabled, the cached writes beyond EOF
241 * extend local i_size without keeping userspace server in sync. So,
242 * attr->size coming from server can be stale. We cannot trust it.
244 if (!is_wb || !S_ISREG(inode->i_mode))
245 i_size_write(inode, attr->size);
246 spin_unlock(&fi->lock);
248 if (!is_wb && S_ISREG(inode->i_mode)) {
251 if (oldsize != attr->size) {
252 truncate_pagecache(inode, attr->size);
253 if (!fc->explicit_inval_data)
255 } else if (fc->auto_inval_data) {
256 struct timespec64 new_mtime = {
257 .tv_sec = attr->mtime,
258 .tv_nsec = attr->mtimensec,
262 * Auto inval mode also checks and invalidates if mtime
265 if (!timespec64_equal(&old_mtime, &new_mtime))
270 invalidate_inode_pages2(inode->i_mapping);
274 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
276 inode->i_mode = attr->mode & S_IFMT;
277 inode->i_size = attr->size;
278 inode->i_mtime.tv_sec = attr->mtime;
279 inode->i_mtime.tv_nsec = attr->mtimensec;
280 inode->i_ctime.tv_sec = attr->ctime;
281 inode->i_ctime.tv_nsec = attr->ctimensec;
282 if (S_ISREG(inode->i_mode)) {
283 fuse_init_common(inode);
284 fuse_init_file_inode(inode);
285 } else if (S_ISDIR(inode->i_mode))
286 fuse_init_dir(inode);
287 else if (S_ISLNK(inode->i_mode))
288 fuse_init_symlink(inode);
289 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
290 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
291 fuse_init_common(inode);
292 init_special_inode(inode, inode->i_mode,
293 new_decode_dev(attr->rdev));
298 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
300 u64 nodeid = *(u64 *) _nodeidp;
301 if (get_node_id(inode) == nodeid)
307 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
309 u64 nodeid = *(u64 *) _nodeidp;
310 get_fuse_inode(inode)->nodeid = nodeid;
314 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
315 int generation, struct fuse_attr *attr,
316 u64 attr_valid, u64 attr_version)
319 struct fuse_inode *fi;
320 struct fuse_conn *fc = get_fuse_conn_super(sb);
323 * Auto mount points get their node id from the submount root, which is
324 * not a unique identifier within this filesystem.
326 * To avoid conflicts, do not place submount points into the inode hash
329 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
330 S_ISDIR(attr->mode)) {
331 inode = new_inode(sb);
335 fuse_init_inode(inode, attr);
336 get_fuse_inode(inode)->nodeid = nodeid;
337 inode->i_flags |= S_AUTOMOUNT;
342 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
346 if ((inode->i_state & I_NEW)) {
347 inode->i_flags |= S_NOATIME;
348 if (!fc->writeback_cache || !S_ISREG(attr->mode))
349 inode->i_flags |= S_NOCMTIME;
350 inode->i_generation = generation;
351 fuse_init_inode(inode, attr);
352 unlock_new_inode(inode);
353 } else if (fuse_stale_inode(inode, generation, attr)) {
354 /* nodeid was reused, any I/O on the old inode should fail */
355 fuse_make_bad(inode);
360 fi = get_fuse_inode(inode);
361 spin_lock(&fi->lock);
363 spin_unlock(&fi->lock);
364 fuse_change_attributes(inode, attr, attr_valid, attr_version);
369 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
370 struct fuse_mount **fm)
372 struct fuse_mount *fm_iter;
375 WARN_ON(!rwsem_is_locked(&fc->killsb));
376 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
380 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
391 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
392 loff_t offset, loff_t len)
394 struct fuse_inode *fi;
399 inode = fuse_ilookup(fc, nodeid, NULL);
403 fi = get_fuse_inode(inode);
404 spin_lock(&fi->lock);
405 fi->attr_version = atomic64_inc_return(&fc->attr_version);
406 spin_unlock(&fi->lock);
408 fuse_invalidate_attr(inode);
409 forget_all_cached_acls(inode);
411 pg_start = offset >> PAGE_SHIFT;
415 pg_end = (offset + len - 1) >> PAGE_SHIFT;
416 invalidate_inode_pages2_range(inode->i_mapping,
423 bool fuse_lock_inode(struct inode *inode)
427 if (!get_fuse_conn(inode)->parallel_dirops) {
428 mutex_lock(&get_fuse_inode(inode)->mutex);
435 void fuse_unlock_inode(struct inode *inode, bool locked)
438 mutex_unlock(&get_fuse_inode(inode)->mutex);
441 static void fuse_umount_begin(struct super_block *sb)
443 struct fuse_conn *fc = get_fuse_conn_super(sb);
445 if (!fc->no_force_umount)
449 static void fuse_send_destroy(struct fuse_mount *fm)
451 if (fm->fc->conn_init) {
454 args.opcode = FUSE_DESTROY;
457 fuse_simple_request(fm, &args);
461 static void fuse_put_super(struct super_block *sb)
463 struct fuse_mount *fm = get_fuse_mount_super(sb);
465 fuse_conn_put(fm->fc);
469 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
471 stbuf->f_type = FUSE_SUPER_MAGIC;
472 stbuf->f_bsize = attr->bsize;
473 stbuf->f_frsize = attr->frsize;
474 stbuf->f_blocks = attr->blocks;
475 stbuf->f_bfree = attr->bfree;
476 stbuf->f_bavail = attr->bavail;
477 stbuf->f_files = attr->files;
478 stbuf->f_ffree = attr->ffree;
479 stbuf->f_namelen = attr->namelen;
480 /* fsid is left zero */
483 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
485 struct super_block *sb = dentry->d_sb;
486 struct fuse_mount *fm = get_fuse_mount_super(sb);
488 struct fuse_statfs_out outarg;
491 if (!fuse_allow_current_process(fm->fc)) {
492 buf->f_type = FUSE_SUPER_MAGIC;
496 memset(&outarg, 0, sizeof(outarg));
498 args.opcode = FUSE_STATFS;
499 args.nodeid = get_node_id(d_inode(dentry));
500 args.out_numargs = 1;
501 args.out_args[0].size = sizeof(outarg);
502 args.out_args[0].value = &outarg;
503 err = fuse_simple_request(fm, &args);
505 convert_fuse_statfs(buf, &outarg.st);
509 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
511 struct fuse_sync_bucket *bucket;
513 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
515 init_waitqueue_head(&bucket->waitq);
516 /* Initial active count */
517 atomic_set(&bucket->count, 1);
522 static void fuse_sync_fs_writes(struct fuse_conn *fc)
524 struct fuse_sync_bucket *bucket, *new_bucket;
527 new_bucket = fuse_sync_bucket_alloc();
528 spin_lock(&fc->lock);
529 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
530 count = atomic_read(&bucket->count);
532 /* No outstanding writes? */
534 spin_unlock(&fc->lock);
540 * Completion of new bucket depends on completion of this bucket, so add
543 atomic_inc(&new_bucket->count);
544 rcu_assign_pointer(fc->curr_bucket, new_bucket);
545 spin_unlock(&fc->lock);
547 * Drop initial active count. At this point if all writes in this and
548 * ancestor buckets complete, the count will go to zero and this task
551 atomic_dec(&bucket->count);
553 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
555 /* Drop temp count on descendant bucket */
556 fuse_sync_bucket_dec(new_bucket);
557 kfree_rcu(bucket, rcu);
560 static int fuse_sync_fs(struct super_block *sb, int wait)
562 struct fuse_mount *fm = get_fuse_mount_super(sb);
563 struct fuse_conn *fc = fm->fc;
564 struct fuse_syncfs_in inarg;
569 * Userspace cannot handle the wait == 0 case. Avoid a
570 * gratuitous roundtrip.
575 /* The filesystem is being unmounted. Nothing to do. */
582 fuse_sync_fs_writes(fc);
584 memset(&inarg, 0, sizeof(inarg));
586 args.in_args[0].size = sizeof(inarg);
587 args.in_args[0].value = &inarg;
588 args.opcode = FUSE_SYNCFS;
589 args.nodeid = get_node_id(sb->s_root->d_inode);
590 args.out_numargs = 0;
592 err = fuse_simple_request(fm, &args);
593 if (err == -ENOSYS) {
608 OPT_DEFAULT_PERMISSIONS,
615 static const struct fs_parameter_spec fuse_fs_parameters[] = {
616 fsparam_string ("source", OPT_SOURCE),
617 fsparam_u32 ("fd", OPT_FD),
618 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
619 fsparam_u32 ("user_id", OPT_USER_ID),
620 fsparam_u32 ("group_id", OPT_GROUP_ID),
621 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
622 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
623 fsparam_u32 ("max_read", OPT_MAX_READ),
624 fsparam_u32 ("blksize", OPT_BLKSIZE),
625 fsparam_string ("subtype", OPT_SUBTYPE),
629 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
631 struct fs_parse_result result;
632 struct fuse_fs_context *ctx = fsc->fs_private;
635 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
637 * Ignore options coming from mount(MS_REMOUNT) for backward
643 return invalfc(fsc, "No changes allowed in reconfigure");
646 opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
653 return invalfc(fsc, "Multiple sources specified");
654 fsc->source = param->string;
655 param->string = NULL;
660 return invalfc(fsc, "Multiple subtypes specified");
661 ctx->subtype = param->string;
662 param->string = NULL;
666 ctx->fd = result.uint_32;
667 ctx->fd_present = true;
671 if (!fuse_valid_type(result.uint_32))
672 return invalfc(fsc, "Invalid rootmode");
673 ctx->rootmode = result.uint_32;
674 ctx->rootmode_present = true;
678 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
679 if (!uid_valid(ctx->user_id))
680 return invalfc(fsc, "Invalid user_id");
681 ctx->user_id_present = true;
685 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
686 if (!gid_valid(ctx->group_id))
687 return invalfc(fsc, "Invalid group_id");
688 ctx->group_id_present = true;
691 case OPT_DEFAULT_PERMISSIONS:
692 ctx->default_permissions = true;
695 case OPT_ALLOW_OTHER:
696 ctx->allow_other = true;
700 ctx->max_read = result.uint_32;
705 return invalfc(fsc, "blksize only supported for fuseblk");
706 ctx->blksize = result.uint_32;
716 static void fuse_free_fsc(struct fs_context *fsc)
718 struct fuse_fs_context *ctx = fsc->fs_private;
726 static int fuse_show_options(struct seq_file *m, struct dentry *root)
728 struct super_block *sb = root->d_sb;
729 struct fuse_conn *fc = get_fuse_conn_super(sb);
731 if (fc->legacy_opts_show) {
732 seq_printf(m, ",user_id=%u",
733 from_kuid_munged(fc->user_ns, fc->user_id));
734 seq_printf(m, ",group_id=%u",
735 from_kgid_munged(fc->user_ns, fc->group_id));
736 if (fc->default_permissions)
737 seq_puts(m, ",default_permissions");
739 seq_puts(m, ",allow_other");
740 if (fc->max_read != ~0)
741 seq_printf(m, ",max_read=%u", fc->max_read);
742 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
743 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
745 #ifdef CONFIG_FUSE_DAX
753 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
754 const struct fuse_iqueue_ops *ops,
757 memset(fiq, 0, sizeof(struct fuse_iqueue));
758 spin_lock_init(&fiq->lock);
759 init_waitqueue_head(&fiq->waitq);
760 INIT_LIST_HEAD(&fiq->pending);
761 INIT_LIST_HEAD(&fiq->interrupts);
762 fiq->forget_list_tail = &fiq->forget_list_head;
768 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
772 spin_lock_init(&fpq->lock);
773 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
774 INIT_LIST_HEAD(&fpq->processing[i]);
775 INIT_LIST_HEAD(&fpq->io);
779 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
780 struct user_namespace *user_ns,
781 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
783 memset(fc, 0, sizeof(*fc));
784 spin_lock_init(&fc->lock);
785 spin_lock_init(&fc->bg_lock);
786 init_rwsem(&fc->killsb);
787 refcount_set(&fc->count, 1);
788 atomic_set(&fc->dev_count, 1);
789 init_waitqueue_head(&fc->blocked_waitq);
790 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
791 INIT_LIST_HEAD(&fc->bg_queue);
792 INIT_LIST_HEAD(&fc->entry);
793 INIT_LIST_HEAD(&fc->devices);
794 atomic_set(&fc->num_waiting, 0);
795 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
796 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
797 atomic64_set(&fc->khctr, 0);
798 fc->polled_files = RB_ROOT;
802 atomic64_set(&fc->attr_version, 1);
803 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
804 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
805 fc->user_ns = get_user_ns(user_ns);
806 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
807 fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
809 INIT_LIST_HEAD(&fc->mounts);
810 list_add(&fm->fc_entry, &fc->mounts);
813 EXPORT_SYMBOL_GPL(fuse_conn_init);
815 void fuse_conn_put(struct fuse_conn *fc)
817 if (refcount_dec_and_test(&fc->count)) {
818 struct fuse_iqueue *fiq = &fc->iq;
819 struct fuse_sync_bucket *bucket;
821 if (IS_ENABLED(CONFIG_FUSE_DAX))
822 fuse_dax_conn_free(fc);
823 if (fiq->ops->release)
824 fiq->ops->release(fiq);
825 put_pid_ns(fc->pid_ns);
826 put_user_ns(fc->user_ns);
827 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
829 WARN_ON(atomic_read(&bucket->count) != 1);
835 EXPORT_SYMBOL_GPL(fuse_conn_put);
837 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
839 refcount_inc(&fc->count);
842 EXPORT_SYMBOL_GPL(fuse_conn_get);
844 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
846 struct fuse_attr attr;
847 memset(&attr, 0, sizeof(attr));
850 attr.ino = FUSE_ROOT_ID;
852 return fuse_iget(sb, 1, 0, &attr, 0, 0);
855 struct fuse_inode_handle {
860 static struct dentry *fuse_get_dentry(struct super_block *sb,
861 struct fuse_inode_handle *handle)
863 struct fuse_conn *fc = get_fuse_conn_super(sb);
865 struct dentry *entry;
868 if (handle->nodeid == 0)
871 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
873 struct fuse_entry_out outarg;
874 const struct qstr name = QSTR_INIT(".", 1);
876 if (!fc->export_support)
879 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
881 if (err && err != -ENOENT)
888 if (get_node_id(inode) != handle->nodeid)
892 if (inode->i_generation != handle->generation)
895 entry = d_obtain_alias(inode);
896 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
897 fuse_invalidate_entry_cache(entry);
907 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
908 struct inode *parent)
910 int len = parent ? 6 : 3;
914 if (*max_len < len) {
916 return FILEID_INVALID;
919 nodeid = get_fuse_inode(inode)->nodeid;
920 generation = inode->i_generation;
922 fh[0] = (u32)(nodeid >> 32);
923 fh[1] = (u32)(nodeid & 0xffffffff);
927 nodeid = get_fuse_inode(parent)->nodeid;
928 generation = parent->i_generation;
930 fh[3] = (u32)(nodeid >> 32);
931 fh[4] = (u32)(nodeid & 0xffffffff);
936 return parent ? 0x82 : 0x81;
939 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
940 struct fid *fid, int fh_len, int fh_type)
942 struct fuse_inode_handle handle;
944 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
947 handle.nodeid = (u64) fid->raw[0] << 32;
948 handle.nodeid |= (u64) fid->raw[1];
949 handle.generation = fid->raw[2];
950 return fuse_get_dentry(sb, &handle);
953 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
954 struct fid *fid, int fh_len, int fh_type)
956 struct fuse_inode_handle parent;
958 if (fh_type != 0x82 || fh_len < 6)
961 parent.nodeid = (u64) fid->raw[3] << 32;
962 parent.nodeid |= (u64) fid->raw[4];
963 parent.generation = fid->raw[5];
964 return fuse_get_dentry(sb, &parent);
967 static struct dentry *fuse_get_parent(struct dentry *child)
969 struct inode *child_inode = d_inode(child);
970 struct fuse_conn *fc = get_fuse_conn(child_inode);
972 struct dentry *parent;
973 struct fuse_entry_out outarg;
976 if (!fc->export_support)
977 return ERR_PTR(-ESTALE);
979 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
980 &dotdot_name, &outarg, &inode);
983 return ERR_PTR(-ESTALE);
987 parent = d_obtain_alias(inode);
988 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
989 fuse_invalidate_entry_cache(parent);
994 static const struct export_operations fuse_export_operations = {
995 .fh_to_dentry = fuse_fh_to_dentry,
996 .fh_to_parent = fuse_fh_to_parent,
997 .encode_fh = fuse_encode_fh,
998 .get_parent = fuse_get_parent,
1001 static const struct super_operations fuse_super_operations = {
1002 .alloc_inode = fuse_alloc_inode,
1003 .free_inode = fuse_free_inode,
1004 .evict_inode = fuse_evict_inode,
1005 .write_inode = fuse_write_inode,
1006 .drop_inode = generic_delete_inode,
1007 .put_super = fuse_put_super,
1008 .umount_begin = fuse_umount_begin,
1009 .statfs = fuse_statfs,
1010 .sync_fs = fuse_sync_fs,
1011 .show_options = fuse_show_options,
1014 static void sanitize_global_limit(unsigned *limit)
1017 * The default maximum number of async requests is calculated to consume
1018 * 1/2^13 of the total memory, assuming 392 bytes per request.
1021 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1023 if (*limit >= 1 << 16)
1024 *limit = (1 << 16) - 1;
1027 static int set_global_limit(const char *val, const struct kernel_param *kp)
1031 rv = param_set_uint(val, kp);
1035 sanitize_global_limit((unsigned *)kp->arg);
1040 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1042 int cap_sys_admin = capable(CAP_SYS_ADMIN);
1044 if (arg->minor < 13)
1047 sanitize_global_limit(&max_user_bgreq);
1048 sanitize_global_limit(&max_user_congthresh);
1050 spin_lock(&fc->bg_lock);
1051 if (arg->max_background) {
1052 fc->max_background = arg->max_background;
1054 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1055 fc->max_background = max_user_bgreq;
1057 if (arg->congestion_threshold) {
1058 fc->congestion_threshold = arg->congestion_threshold;
1060 if (!cap_sys_admin &&
1061 fc->congestion_threshold > max_user_congthresh)
1062 fc->congestion_threshold = max_user_congthresh;
1064 spin_unlock(&fc->bg_lock);
1067 struct fuse_init_args {
1068 struct fuse_args args;
1069 struct fuse_init_in in;
1070 struct fuse_init_out out;
1073 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1076 struct fuse_conn *fc = fm->fc;
1077 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1078 struct fuse_init_out *arg = &ia->out;
1081 if (error || arg->major != FUSE_KERNEL_VERSION)
1084 unsigned long ra_pages;
1086 process_init_limits(fc, arg);
1088 if (arg->minor >= 6) {
1089 ra_pages = arg->max_readahead / PAGE_SIZE;
1090 if (arg->flags & FUSE_ASYNC_READ)
1092 if (!(arg->flags & FUSE_POSIX_LOCKS))
1094 if (arg->minor >= 17) {
1095 if (!(arg->flags & FUSE_FLOCK_LOCKS))
1098 if (!(arg->flags & FUSE_POSIX_LOCKS))
1101 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1102 fc->atomic_o_trunc = 1;
1103 if (arg->minor >= 9) {
1104 /* LOOKUP has dependency on proto version */
1105 if (arg->flags & FUSE_EXPORT_SUPPORT)
1106 fc->export_support = 1;
1108 if (arg->flags & FUSE_BIG_WRITES)
1110 if (arg->flags & FUSE_DONT_MASK)
1112 if (arg->flags & FUSE_AUTO_INVAL_DATA)
1113 fc->auto_inval_data = 1;
1114 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1115 fc->explicit_inval_data = 1;
1116 if (arg->flags & FUSE_DO_READDIRPLUS) {
1117 fc->do_readdirplus = 1;
1118 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1119 fc->readdirplus_auto = 1;
1121 if (arg->flags & FUSE_ASYNC_DIO)
1123 if (arg->flags & FUSE_WRITEBACK_CACHE)
1124 fc->writeback_cache = 1;
1125 if (arg->flags & FUSE_PARALLEL_DIROPS)
1126 fc->parallel_dirops = 1;
1127 if (arg->flags & FUSE_HANDLE_KILLPRIV)
1128 fc->handle_killpriv = 1;
1129 if (arg->time_gran && arg->time_gran <= 1000000000)
1130 fm->sb->s_time_gran = arg->time_gran;
1131 if ((arg->flags & FUSE_POSIX_ACL)) {
1132 fc->default_permissions = 1;
1134 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1136 if (arg->flags & FUSE_CACHE_SYMLINKS)
1137 fc->cache_symlinks = 1;
1138 if (arg->flags & FUSE_ABORT_ERROR)
1140 if (arg->flags & FUSE_MAX_PAGES) {
1142 min_t(unsigned int, fc->max_pages_limit,
1143 max_t(unsigned int, arg->max_pages, 1));
1145 if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1146 arg->flags & FUSE_MAP_ALIGNMENT &&
1147 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1150 if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) {
1151 fc->handle_killpriv_v2 = 1;
1152 fm->sb->s_flags |= SB_NOSEC;
1154 if (arg->flags & FUSE_SETXATTR_EXT)
1155 fc->setxattr_ext = 1;
1157 ra_pages = fc->max_read / PAGE_SIZE;
1162 fm->sb->s_bdi->ra_pages =
1163 min(fm->sb->s_bdi->ra_pages, ra_pages);
1164 fc->minor = arg->minor;
1165 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1166 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1176 fuse_set_initialized(fc);
1177 wake_up_all(&fc->blocked_waitq);
1180 void fuse_send_init(struct fuse_mount *fm)
1182 struct fuse_init_args *ia;
1184 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1186 ia->in.major = FUSE_KERNEL_VERSION;
1187 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1188 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1190 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1191 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1192 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1193 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1194 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1195 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1196 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1197 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1198 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1199 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT;
1200 #ifdef CONFIG_FUSE_DAX
1202 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1204 if (fm->fc->auto_submounts)
1205 ia->in.flags |= FUSE_SUBMOUNTS;
1207 ia->args.opcode = FUSE_INIT;
1208 ia->args.in_numargs = 1;
1209 ia->args.in_args[0].size = sizeof(ia->in);
1210 ia->args.in_args[0].value = &ia->in;
1211 ia->args.out_numargs = 1;
1212 /* Variable length argument used for backward compatibility
1213 with interface version < 7.5. Rest of init_out is zeroed
1214 by do_get_request(), so a short reply is not a problem */
1215 ia->args.out_argvar = true;
1216 ia->args.out_args[0].size = sizeof(ia->out);
1217 ia->args.out_args[0].value = &ia->out;
1218 ia->args.force = true;
1219 ia->args.nocreds = true;
1220 ia->args.end = process_init_reply;
1222 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1223 process_init_reply(fm, &ia->args, -ENOTCONN);
1225 EXPORT_SYMBOL_GPL(fuse_send_init);
1227 void fuse_free_conn(struct fuse_conn *fc)
1229 WARN_ON(!list_empty(&fc->devices));
1232 EXPORT_SYMBOL_GPL(fuse_free_conn);
1234 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1240 suffix = "-fuseblk";
1242 * sb->s_bdi points to blkdev's bdi however we want to redirect
1243 * it to our private bdi...
1246 sb->s_bdi = &noop_backing_dev_info;
1248 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1249 MINOR(fc->dev), suffix);
1253 /* fuse does it's own writeback accounting */
1254 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1255 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1258 * For a single fuse filesystem use max 1% of dirty +
1259 * writeback threshold.
1261 * This gives about 1M of write buffer for memory maps on a
1262 * machine with 1G and 10% dirty_ratio, which should be more
1265 * Privileged users can raise it by writing to
1267 * /sys/class/bdi/<bdi>/max_ratio
1269 bdi_set_max_ratio(sb->s_bdi, 1);
1274 struct fuse_dev *fuse_dev_alloc(void)
1276 struct fuse_dev *fud;
1277 struct list_head *pq;
1279 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1283 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1289 fud->pq.processing = pq;
1290 fuse_pqueue_init(&fud->pq);
1294 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1296 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1298 fud->fc = fuse_conn_get(fc);
1299 spin_lock(&fc->lock);
1300 list_add_tail(&fud->entry, &fc->devices);
1301 spin_unlock(&fc->lock);
1303 EXPORT_SYMBOL_GPL(fuse_dev_install);
1305 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1307 struct fuse_dev *fud;
1309 fud = fuse_dev_alloc();
1313 fuse_dev_install(fud, fc);
1316 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1318 void fuse_dev_free(struct fuse_dev *fud)
1320 struct fuse_conn *fc = fud->fc;
1323 spin_lock(&fc->lock);
1324 list_del(&fud->entry);
1325 spin_unlock(&fc->lock);
1329 kfree(fud->pq.processing);
1332 EXPORT_SYMBOL_GPL(fuse_dev_free);
1334 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1335 const struct fuse_inode *fi)
1337 *attr = (struct fuse_attr){
1338 .ino = fi->inode.i_ino,
1339 .size = fi->inode.i_size,
1340 .blocks = fi->inode.i_blocks,
1341 .atime = fi->inode.i_atime.tv_sec,
1342 .mtime = fi->inode.i_mtime.tv_sec,
1343 .ctime = fi->inode.i_ctime.tv_sec,
1344 .atimensec = fi->inode.i_atime.tv_nsec,
1345 .mtimensec = fi->inode.i_mtime.tv_nsec,
1346 .ctimensec = fi->inode.i_ctime.tv_nsec,
1347 .mode = fi->inode.i_mode,
1348 .nlink = fi->inode.i_nlink,
1349 .uid = fi->inode.i_uid.val,
1350 .gid = fi->inode.i_gid.val,
1351 .rdev = fi->inode.i_rdev,
1352 .blksize = 1u << fi->inode.i_blkbits,
1356 static void fuse_sb_defaults(struct super_block *sb)
1358 sb->s_magic = FUSE_SUPER_MAGIC;
1359 sb->s_op = &fuse_super_operations;
1360 sb->s_xattr = fuse_xattr_handlers;
1361 sb->s_maxbytes = MAX_LFS_FILESIZE;
1362 sb->s_time_gran = 1;
1363 sb->s_export_op = &fuse_export_operations;
1364 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1365 if (sb->s_user_ns != &init_user_ns)
1366 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1367 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1370 * If we are not in the initial user namespace posix
1371 * acls must be translated.
1373 if (sb->s_user_ns != &init_user_ns)
1374 sb->s_xattr = fuse_no_acl_xattr_handlers;
1377 static int fuse_fill_super_submount(struct super_block *sb,
1378 struct fuse_inode *parent_fi)
1380 struct fuse_mount *fm = get_fuse_mount_super(sb);
1381 struct super_block *parent_sb = parent_fi->inode.i_sb;
1382 struct fuse_attr root_attr;
1385 fuse_sb_defaults(sb);
1388 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1389 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1391 sb->s_xattr = parent_sb->s_xattr;
1392 sb->s_time_gran = parent_sb->s_time_gran;
1393 sb->s_blocksize = parent_sb->s_blocksize;
1394 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1395 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1396 if (parent_sb->s_subtype && !sb->s_subtype)
1399 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1400 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1402 * This inode is just a duplicate, so it is not looked up and
1403 * its nlookup should not be incremented. fuse_iget() does
1404 * that, though, so undo it here.
1406 get_fuse_inode(root)->nlookup--;
1407 sb->s_d_op = &fuse_dentry_operations;
1408 sb->s_root = d_make_root(root);
1415 /* Filesystem context private data holds the FUSE inode of the mount point */
1416 static int fuse_get_tree_submount(struct fs_context *fsc)
1418 struct fuse_mount *fm;
1419 struct fuse_inode *mp_fi = fsc->fs_private;
1420 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1421 struct super_block *sb;
1424 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1428 fsc->s_fs_info = fm;
1429 sb = sget_fc(fsc, NULL, set_anon_super_fc);
1434 fm->fc = fuse_conn_get(fc);
1436 /* Initialize superblock, making @mp_fi its root */
1437 err = fuse_fill_super_submount(sb, mp_fi);
1441 sb->s_fs_info = NULL;
1442 deactivate_locked_super(sb);
1446 down_write(&fc->killsb);
1447 list_add_tail(&fm->fc_entry, &fc->mounts);
1448 up_write(&fc->killsb);
1450 sb->s_flags |= SB_ACTIVE;
1451 fsc->root = dget(sb->s_root);
1456 static const struct fs_context_operations fuse_context_submount_ops = {
1457 .get_tree = fuse_get_tree_submount,
1460 int fuse_init_fs_context_submount(struct fs_context *fsc)
1462 fsc->ops = &fuse_context_submount_ops;
1465 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1467 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1469 struct fuse_dev *fud = NULL;
1470 struct fuse_mount *fm = get_fuse_mount_super(sb);
1471 struct fuse_conn *fc = fm->fc;
1473 struct dentry *root_dentry;
1477 if (sb->s_flags & SB_MANDLOCK)
1480 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1481 fuse_sb_defaults(sb);
1486 if (!sb_set_blocksize(sb, ctx->blksize))
1490 sb->s_blocksize = PAGE_SIZE;
1491 sb->s_blocksize_bits = PAGE_SHIFT;
1494 sb->s_subtype = ctx->subtype;
1495 ctx->subtype = NULL;
1496 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1497 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1504 fud = fuse_dev_alloc_install(fc);
1509 fc->dev = sb->s_dev;
1511 err = fuse_bdi_init(fc, sb);
1515 /* Handle umasking inside the fuse code */
1516 if (sb->s_flags & SB_POSIXACL)
1518 sb->s_flags |= SB_POSIXACL;
1520 fc->default_permissions = ctx->default_permissions;
1521 fc->allow_other = ctx->allow_other;
1522 fc->user_id = ctx->user_id;
1523 fc->group_id = ctx->group_id;
1524 fc->legacy_opts_show = ctx->legacy_opts_show;
1525 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1526 fc->destroy = ctx->destroy;
1527 fc->no_control = ctx->no_control;
1528 fc->no_force_umount = ctx->no_force_umount;
1531 root = fuse_get_root_inode(sb, ctx->rootmode);
1532 sb->s_d_op = &fuse_root_dentry_operations;
1533 root_dentry = d_make_root(root);
1536 /* Root dentry doesn't have .d_revalidate */
1537 sb->s_d_op = &fuse_dentry_operations;
1539 mutex_lock(&fuse_mutex);
1541 if (ctx->fudptr && *ctx->fudptr)
1544 err = fuse_ctl_add_conn(fc);
1548 list_add_tail(&fc->entry, &fuse_conn_list);
1549 sb->s_root = root_dentry;
1552 mutex_unlock(&fuse_mutex);
1556 mutex_unlock(&fuse_mutex);
1562 if (IS_ENABLED(CONFIG_FUSE_DAX))
1563 fuse_dax_conn_free(fc);
1567 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1569 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1571 struct fuse_fs_context *ctx = fsc->fs_private;
1573 struct fuse_conn *fc;
1574 struct fuse_mount *fm;
1576 if (!ctx->file || !ctx->rootmode_present ||
1577 !ctx->user_id_present || !ctx->group_id_present)
1581 * Require mount to happen from the same user namespace which
1582 * opened /dev/fuse to prevent potential attacks.
1585 if ((ctx->file->f_op != &fuse_dev_operations) ||
1586 (ctx->file->f_cred->user_ns != sb->s_user_ns))
1588 ctx->fudptr = &ctx->file->private_data;
1590 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1595 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1601 fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1602 fc->release = fuse_free_conn;
1606 err = fuse_fill_super_common(sb, ctx);
1609 /* file->private_data shall be visible on all CPUs after this */
1611 fuse_send_init(get_fuse_mount_super(sb));
1617 sb->s_fs_info = NULL;
1623 * This is the path where user supplied an already initialized fuse dev. In
1624 * this case never create a new super if the old one is gone.
1626 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1631 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1634 return fsc->sget_key == get_fuse_conn_super(sb);
1637 static int fuse_get_tree(struct fs_context *fsc)
1639 struct fuse_fs_context *ctx = fsc->fs_private;
1640 struct fuse_dev *fud;
1641 struct super_block *sb;
1644 if (ctx->fd_present)
1645 ctx->file = fget(ctx->fd);
1647 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1648 err = get_tree_bdev(fsc, fuse_fill_super);
1652 * While block dev mount can be initialized with a dummy device fd
1653 * (found by device name), normal fuse mounts can't
1659 * Allow creating a fuse mount with an already initialized fuse
1662 fud = READ_ONCE(ctx->file->private_data);
1663 if (ctx->file->f_op == &fuse_dev_operations && fud) {
1664 fsc->sget_key = fud->fc;
1665 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1666 err = PTR_ERR_OR_ZERO(sb);
1668 fsc->root = dget(sb->s_root);
1670 err = get_tree_nodev(fsc, fuse_fill_super);
1678 static const struct fs_context_operations fuse_context_ops = {
1679 .free = fuse_free_fsc,
1680 .parse_param = fuse_parse_param,
1681 .reconfigure = fuse_reconfigure,
1682 .get_tree = fuse_get_tree,
1686 * Set up the filesystem mount context.
1688 static int fuse_init_fs_context(struct fs_context *fsc)
1690 struct fuse_fs_context *ctx;
1692 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1697 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1698 ctx->legacy_opts_show = true;
1701 if (fsc->fs_type == &fuseblk_fs_type) {
1702 ctx->is_bdev = true;
1703 ctx->destroy = true;
1707 fsc->fs_private = ctx;
1708 fsc->ops = &fuse_context_ops;
1712 bool fuse_mount_remove(struct fuse_mount *fm)
1714 struct fuse_conn *fc = fm->fc;
1717 down_write(&fc->killsb);
1718 list_del_init(&fm->fc_entry);
1719 if (list_empty(&fc->mounts))
1721 up_write(&fc->killsb);
1725 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1727 void fuse_conn_destroy(struct fuse_mount *fm)
1729 struct fuse_conn *fc = fm->fc;
1732 fuse_send_destroy(fm);
1734 fuse_abort_conn(fc);
1735 fuse_wait_aborted(fc);
1737 if (!list_empty(&fc->entry)) {
1738 mutex_lock(&fuse_mutex);
1739 list_del(&fc->entry);
1740 fuse_ctl_remove_conn(fc);
1741 mutex_unlock(&fuse_mutex);
1744 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1746 static void fuse_sb_destroy(struct super_block *sb)
1748 struct fuse_mount *fm = get_fuse_mount_super(sb);
1752 last = fuse_mount_remove(fm);
1754 fuse_conn_destroy(fm);
1758 static void fuse_kill_sb_anon(struct super_block *sb)
1760 fuse_sb_destroy(sb);
1761 kill_anon_super(sb);
1764 static struct file_system_type fuse_fs_type = {
1765 .owner = THIS_MODULE,
1767 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1768 .init_fs_context = fuse_init_fs_context,
1769 .parameters = fuse_fs_parameters,
1770 .kill_sb = fuse_kill_sb_anon,
1772 MODULE_ALIAS_FS("fuse");
1775 static void fuse_kill_sb_blk(struct super_block *sb)
1777 fuse_sb_destroy(sb);
1778 kill_block_super(sb);
1781 static struct file_system_type fuseblk_fs_type = {
1782 .owner = THIS_MODULE,
1784 .init_fs_context = fuse_init_fs_context,
1785 .parameters = fuse_fs_parameters,
1786 .kill_sb = fuse_kill_sb_blk,
1787 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1789 MODULE_ALIAS_FS("fuseblk");
1791 static inline int register_fuseblk(void)
1793 return register_filesystem(&fuseblk_fs_type);
1796 static inline void unregister_fuseblk(void)
1798 unregister_filesystem(&fuseblk_fs_type);
1801 static inline int register_fuseblk(void)
1806 static inline void unregister_fuseblk(void)
1811 static void fuse_inode_init_once(void *foo)
1813 struct inode *inode = foo;
1815 inode_init_once(inode);
1818 static int __init fuse_fs_init(void)
1822 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1823 sizeof(struct fuse_inode), 0,
1824 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1825 fuse_inode_init_once);
1827 if (!fuse_inode_cachep)
1830 err = register_fuseblk();
1834 err = register_filesystem(&fuse_fs_type);
1841 unregister_fuseblk();
1843 kmem_cache_destroy(fuse_inode_cachep);
1848 static void fuse_fs_cleanup(void)
1850 unregister_filesystem(&fuse_fs_type);
1851 unregister_fuseblk();
1854 * Make sure all delayed rcu free inodes are flushed before we
1858 kmem_cache_destroy(fuse_inode_cachep);
1861 static struct kobject *fuse_kobj;
1863 static int fuse_sysfs_init(void)
1867 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1873 err = sysfs_create_mount_point(fuse_kobj, "connections");
1875 goto out_fuse_unregister;
1879 out_fuse_unregister:
1880 kobject_put(fuse_kobj);
1885 static void fuse_sysfs_cleanup(void)
1887 sysfs_remove_mount_point(fuse_kobj, "connections");
1888 kobject_put(fuse_kobj);
1891 static int __init fuse_init(void)
1895 pr_info("init (API version %i.%i)\n",
1896 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1898 INIT_LIST_HEAD(&fuse_conn_list);
1899 res = fuse_fs_init();
1903 res = fuse_dev_init();
1905 goto err_fs_cleanup;
1907 res = fuse_sysfs_init();
1909 goto err_dev_cleanup;
1911 res = fuse_ctl_init();
1913 goto err_sysfs_cleanup;
1915 sanitize_global_limit(&max_user_bgreq);
1916 sanitize_global_limit(&max_user_congthresh);
1921 fuse_sysfs_cleanup();
1930 static void __exit fuse_exit(void)
1935 fuse_sysfs_cleanup();
1940 module_init(fuse_init);
1941 module_exit(fuse_exit);