2 FUSE: Filesystem in Userspace
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>
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 spin_lock_init(&fi->lock);
89 fi->forget = fuse_alloc_forget();
93 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
101 kmem_cache_free(fuse_inode_cachep, fi);
105 static void fuse_free_inode(struct inode *inode)
107 struct fuse_inode *fi = get_fuse_inode(inode);
109 mutex_destroy(&fi->mutex);
111 #ifdef CONFIG_FUSE_DAX
114 kmem_cache_free(fuse_inode_cachep, fi);
117 static void fuse_evict_inode(struct inode *inode)
119 struct fuse_inode *fi = get_fuse_inode(inode);
121 truncate_inode_pages_final(&inode->i_data);
123 if (inode->i_sb->s_flags & SB_ACTIVE) {
124 struct fuse_conn *fc = get_fuse_conn(inode);
126 if (FUSE_IS_DAX(inode))
127 fuse_dax_inode_cleanup(inode);
129 fuse_queue_forget(fc, fi->forget, fi->nodeid,
134 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
135 WARN_ON(!list_empty(&fi->write_files));
136 WARN_ON(!list_empty(&fi->queued_writes));
140 static int fuse_reconfigure(struct fs_context *fsc)
142 struct super_block *sb = fsc->root->d_sb;
145 if (fsc->sb_flags & SB_MANDLOCK)
152 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
153 * so that it will fit.
155 static ino_t fuse_squash_ino(u64 ino64)
157 ino_t ino = (ino_t) ino64;
158 if (sizeof(ino_t) < sizeof(u64))
159 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
163 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
166 struct fuse_conn *fc = get_fuse_conn(inode);
167 struct fuse_inode *fi = get_fuse_inode(inode);
169 lockdep_assert_held(&fi->lock);
171 fi->attr_version = atomic64_inc_return(&fc->attr_version);
172 fi->i_time = attr_valid;
173 WRITE_ONCE(fi->inval_mask, 0);
175 inode->i_ino = fuse_squash_ino(attr->ino);
176 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
177 set_nlink(inode, attr->nlink);
178 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
179 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
180 inode->i_blocks = attr->blocks;
181 inode->i_atime.tv_sec = attr->atime;
182 inode->i_atime.tv_nsec = attr->atimensec;
183 /* mtime from server may be stale due to local buffered write */
184 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
185 inode->i_mtime.tv_sec = attr->mtime;
186 inode->i_mtime.tv_nsec = attr->mtimensec;
187 inode->i_ctime.tv_sec = attr->ctime;
188 inode->i_ctime.tv_nsec = attr->ctimensec;
191 if (attr->blksize != 0)
192 inode->i_blkbits = ilog2(attr->blksize);
194 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
197 * Don't set the sticky bit in i_mode, unless we want the VFS
198 * to check permissions. This prevents failures due to the
199 * check in may_delete().
201 fi->orig_i_mode = inode->i_mode;
202 if (!fc->default_permissions)
203 inode->i_mode &= ~S_ISVTX;
205 fi->orig_ino = attr->ino;
208 * We are refreshing inode data and it is possible that another
209 * client set suid/sgid or security.capability xattr. So clear
210 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
211 * was set or if security.capability xattr was set. But we don't
212 * know if security.capability has been set or not. So clear it
213 * anyway. Its less efficient but should be safe.
215 inode->i_flags &= ~S_NOSEC;
218 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
219 u64 attr_valid, u64 attr_version)
221 struct fuse_conn *fc = get_fuse_conn(inode);
222 struct fuse_inode *fi = get_fuse_inode(inode);
223 bool is_wb = fc->writeback_cache;
225 struct timespec64 old_mtime;
227 spin_lock(&fi->lock);
228 if ((attr_version != 0 && fi->attr_version > attr_version) ||
229 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
230 spin_unlock(&fi->lock);
234 old_mtime = inode->i_mtime;
235 fuse_change_attributes_common(inode, attr, attr_valid);
237 oldsize = inode->i_size;
239 * In case of writeback_cache enabled, the cached writes beyond EOF
240 * extend local i_size without keeping userspace server in sync. So,
241 * attr->size coming from server can be stale. We cannot trust it.
243 if (!is_wb || !S_ISREG(inode->i_mode))
244 i_size_write(inode, attr->size);
245 spin_unlock(&fi->lock);
247 if (!is_wb && S_ISREG(inode->i_mode)) {
250 if (oldsize != attr->size) {
251 truncate_pagecache(inode, attr->size);
252 if (!fc->explicit_inval_data)
254 } else if (fc->auto_inval_data) {
255 struct timespec64 new_mtime = {
256 .tv_sec = attr->mtime,
257 .tv_nsec = attr->mtimensec,
261 * Auto inval mode also checks and invalidates if mtime
264 if (!timespec64_equal(&old_mtime, &new_mtime))
269 invalidate_inode_pages2(inode->i_mapping);
273 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
275 inode->i_mode = attr->mode & S_IFMT;
276 inode->i_size = attr->size;
277 inode->i_mtime.tv_sec = attr->mtime;
278 inode->i_mtime.tv_nsec = attr->mtimensec;
279 inode->i_ctime.tv_sec = attr->ctime;
280 inode->i_ctime.tv_nsec = attr->ctimensec;
281 if (S_ISREG(inode->i_mode)) {
282 fuse_init_common(inode);
283 fuse_init_file_inode(inode);
284 } else if (S_ISDIR(inode->i_mode))
285 fuse_init_dir(inode);
286 else if (S_ISLNK(inode->i_mode))
287 fuse_init_symlink(inode);
288 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
289 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
290 fuse_init_common(inode);
291 init_special_inode(inode, inode->i_mode,
292 new_decode_dev(attr->rdev));
297 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
299 u64 nodeid = *(u64 *) _nodeidp;
300 if (get_node_id(inode) == nodeid)
306 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
308 u64 nodeid = *(u64 *) _nodeidp;
309 get_fuse_inode(inode)->nodeid = nodeid;
313 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
314 int generation, struct fuse_attr *attr,
315 u64 attr_valid, u64 attr_version)
318 struct fuse_inode *fi;
319 struct fuse_conn *fc = get_fuse_conn_super(sb);
322 * Auto mount points get their node id from the submount root, which is
323 * not a unique identifier within this filesystem.
325 * To avoid conflicts, do not place submount points into the inode hash
328 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
329 S_ISDIR(attr->mode)) {
330 inode = new_inode(sb);
334 fuse_init_inode(inode, attr);
335 get_fuse_inode(inode)->nodeid = nodeid;
336 inode->i_flags |= S_AUTOMOUNT;
341 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
345 if ((inode->i_state & I_NEW)) {
346 inode->i_flags |= S_NOATIME;
347 if (!fc->writeback_cache || !S_ISREG(attr->mode))
348 inode->i_flags |= S_NOCMTIME;
349 inode->i_generation = generation;
350 fuse_init_inode(inode, attr);
351 unlock_new_inode(inode);
352 } else if (fuse_stale_inode(inode, generation, attr)) {
353 /* nodeid was reused, any I/O on the old inode should fail */
354 fuse_make_bad(inode);
359 fi = get_fuse_inode(inode);
360 spin_lock(&fi->lock);
362 spin_unlock(&fi->lock);
363 fuse_change_attributes(inode, attr, attr_valid, attr_version);
368 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
369 struct fuse_mount **fm)
371 struct fuse_mount *fm_iter;
374 WARN_ON(!rwsem_is_locked(&fc->killsb));
375 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
379 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
390 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
391 loff_t offset, loff_t len)
393 struct fuse_inode *fi;
398 inode = fuse_ilookup(fc, nodeid, NULL);
402 fi = get_fuse_inode(inode);
403 spin_lock(&fi->lock);
404 fi->attr_version = atomic64_inc_return(&fc->attr_version);
405 spin_unlock(&fi->lock);
407 fuse_invalidate_attr(inode);
408 forget_all_cached_acls(inode);
410 pg_start = offset >> PAGE_SHIFT;
414 pg_end = (offset + len - 1) >> PAGE_SHIFT;
415 invalidate_inode_pages2_range(inode->i_mapping,
422 bool fuse_lock_inode(struct inode *inode)
426 if (!get_fuse_conn(inode)->parallel_dirops) {
427 mutex_lock(&get_fuse_inode(inode)->mutex);
434 void fuse_unlock_inode(struct inode *inode, bool locked)
437 mutex_unlock(&get_fuse_inode(inode)->mutex);
440 static void fuse_umount_begin(struct super_block *sb)
442 struct fuse_conn *fc = get_fuse_conn_super(sb);
444 if (!fc->no_force_umount)
448 static void fuse_send_destroy(struct fuse_mount *fm)
450 if (fm->fc->conn_init) {
453 args.opcode = FUSE_DESTROY;
456 fuse_simple_request(fm, &args);
460 static void fuse_put_super(struct super_block *sb)
462 struct fuse_mount *fm = get_fuse_mount_super(sb);
464 fuse_conn_put(fm->fc);
468 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
470 stbuf->f_type = FUSE_SUPER_MAGIC;
471 stbuf->f_bsize = attr->bsize;
472 stbuf->f_frsize = attr->frsize;
473 stbuf->f_blocks = attr->blocks;
474 stbuf->f_bfree = attr->bfree;
475 stbuf->f_bavail = attr->bavail;
476 stbuf->f_files = attr->files;
477 stbuf->f_ffree = attr->ffree;
478 stbuf->f_namelen = attr->namelen;
479 /* fsid is left zero */
482 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
484 struct super_block *sb = dentry->d_sb;
485 struct fuse_mount *fm = get_fuse_mount_super(sb);
487 struct fuse_statfs_out outarg;
490 if (!fuse_allow_current_process(fm->fc)) {
491 buf->f_type = FUSE_SUPER_MAGIC;
495 memset(&outarg, 0, sizeof(outarg));
497 args.opcode = FUSE_STATFS;
498 args.nodeid = get_node_id(d_inode(dentry));
499 args.out_numargs = 1;
500 args.out_args[0].size = sizeof(outarg);
501 args.out_args[0].value = &outarg;
502 err = fuse_simple_request(fm, &args);
504 convert_fuse_statfs(buf, &outarg.st);
508 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
510 struct fuse_sync_bucket *bucket;
512 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
514 init_waitqueue_head(&bucket->waitq);
515 /* Initial active count */
516 atomic_set(&bucket->count, 1);
521 static void fuse_sync_fs_writes(struct fuse_conn *fc)
523 struct fuse_sync_bucket *bucket, *new_bucket;
526 new_bucket = fuse_sync_bucket_alloc();
527 spin_lock(&fc->lock);
528 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
529 count = atomic_read(&bucket->count);
531 /* No outstanding writes? */
533 spin_unlock(&fc->lock);
539 * Completion of new bucket depends on completion of this bucket, so add
542 atomic_inc(&new_bucket->count);
543 rcu_assign_pointer(fc->curr_bucket, new_bucket);
544 spin_unlock(&fc->lock);
546 * Drop initial active count. At this point if all writes in this and
547 * ancestor buckets complete, the count will go to zero and this task
550 atomic_dec(&bucket->count);
552 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
554 /* Drop temp count on descendant bucket */
555 fuse_sync_bucket_dec(new_bucket);
556 kfree_rcu(bucket, rcu);
559 static int fuse_sync_fs(struct super_block *sb, int wait)
561 struct fuse_mount *fm = get_fuse_mount_super(sb);
562 struct fuse_conn *fc = fm->fc;
563 struct fuse_syncfs_in inarg;
568 * Userspace cannot handle the wait == 0 case. Avoid a
569 * gratuitous roundtrip.
574 /* The filesystem is being unmounted. Nothing to do. */
581 fuse_sync_fs_writes(fc);
583 memset(&inarg, 0, sizeof(inarg));
585 args.in_args[0].size = sizeof(inarg);
586 args.in_args[0].value = &inarg;
587 args.opcode = FUSE_SYNCFS;
588 args.nodeid = get_node_id(sb->s_root->d_inode);
589 args.out_numargs = 0;
591 err = fuse_simple_request(fm, &args);
592 if (err == -ENOSYS) {
607 OPT_DEFAULT_PERMISSIONS,
614 static const struct fs_parameter_spec fuse_fs_parameters[] = {
615 fsparam_string ("source", OPT_SOURCE),
616 fsparam_u32 ("fd", OPT_FD),
617 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
618 fsparam_u32 ("user_id", OPT_USER_ID),
619 fsparam_u32 ("group_id", OPT_GROUP_ID),
620 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
621 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
622 fsparam_u32 ("max_read", OPT_MAX_READ),
623 fsparam_u32 ("blksize", OPT_BLKSIZE),
624 fsparam_string ("subtype", OPT_SUBTYPE),
628 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
630 struct fs_parse_result result;
631 struct fuse_fs_context *ctx = fsc->fs_private;
634 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
636 * Ignore options coming from mount(MS_REMOUNT) for backward
642 return invalfc(fsc, "No changes allowed in reconfigure");
645 opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
652 return invalfc(fsc, "Multiple sources specified");
653 fsc->source = param->string;
654 param->string = NULL;
659 return invalfc(fsc, "Multiple subtypes specified");
660 ctx->subtype = param->string;
661 param->string = NULL;
665 ctx->fd = result.uint_32;
666 ctx->fd_present = true;
670 if (!fuse_valid_type(result.uint_32))
671 return invalfc(fsc, "Invalid rootmode");
672 ctx->rootmode = result.uint_32;
673 ctx->rootmode_present = true;
677 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
678 if (!uid_valid(ctx->user_id))
679 return invalfc(fsc, "Invalid user_id");
680 ctx->user_id_present = true;
684 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
685 if (!gid_valid(ctx->group_id))
686 return invalfc(fsc, "Invalid group_id");
687 ctx->group_id_present = true;
690 case OPT_DEFAULT_PERMISSIONS:
691 ctx->default_permissions = true;
694 case OPT_ALLOW_OTHER:
695 ctx->allow_other = true;
699 ctx->max_read = result.uint_32;
704 return invalfc(fsc, "blksize only supported for fuseblk");
705 ctx->blksize = result.uint_32;
715 static void fuse_free_fsc(struct fs_context *fsc)
717 struct fuse_fs_context *ctx = fsc->fs_private;
725 static int fuse_show_options(struct seq_file *m, struct dentry *root)
727 struct super_block *sb = root->d_sb;
728 struct fuse_conn *fc = get_fuse_conn_super(sb);
730 if (fc->legacy_opts_show) {
731 seq_printf(m, ",user_id=%u",
732 from_kuid_munged(fc->user_ns, fc->user_id));
733 seq_printf(m, ",group_id=%u",
734 from_kgid_munged(fc->user_ns, fc->group_id));
735 if (fc->default_permissions)
736 seq_puts(m, ",default_permissions");
738 seq_puts(m, ",allow_other");
739 if (fc->max_read != ~0)
740 seq_printf(m, ",max_read=%u", fc->max_read);
741 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
742 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
744 #ifdef CONFIG_FUSE_DAX
752 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
753 const struct fuse_iqueue_ops *ops,
756 memset(fiq, 0, sizeof(struct fuse_iqueue));
757 spin_lock_init(&fiq->lock);
758 init_waitqueue_head(&fiq->waitq);
759 INIT_LIST_HEAD(&fiq->pending);
760 INIT_LIST_HEAD(&fiq->interrupts);
761 fiq->forget_list_tail = &fiq->forget_list_head;
767 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
771 spin_lock_init(&fpq->lock);
772 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
773 INIT_LIST_HEAD(&fpq->processing[i]);
774 INIT_LIST_HEAD(&fpq->io);
778 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
779 struct user_namespace *user_ns,
780 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
782 memset(fc, 0, sizeof(*fc));
783 spin_lock_init(&fc->lock);
784 spin_lock_init(&fc->bg_lock);
785 init_rwsem(&fc->killsb);
786 refcount_set(&fc->count, 1);
787 atomic_set(&fc->dev_count, 1);
788 init_waitqueue_head(&fc->blocked_waitq);
789 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
790 INIT_LIST_HEAD(&fc->bg_queue);
791 INIT_LIST_HEAD(&fc->entry);
792 INIT_LIST_HEAD(&fc->devices);
793 atomic_set(&fc->num_waiting, 0);
794 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
795 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
796 atomic64_set(&fc->khctr, 0);
797 fc->polled_files = RB_ROOT;
801 atomic64_set(&fc->attr_version, 1);
802 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
803 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
804 fc->user_ns = get_user_ns(user_ns);
805 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
806 fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
808 INIT_LIST_HEAD(&fc->mounts);
809 list_add(&fm->fc_entry, &fc->mounts);
812 EXPORT_SYMBOL_GPL(fuse_conn_init);
814 void fuse_conn_put(struct fuse_conn *fc)
816 if (refcount_dec_and_test(&fc->count)) {
817 struct fuse_iqueue *fiq = &fc->iq;
818 struct fuse_sync_bucket *bucket;
820 if (IS_ENABLED(CONFIG_FUSE_DAX))
821 fuse_dax_conn_free(fc);
822 if (fiq->ops->release)
823 fiq->ops->release(fiq);
824 put_pid_ns(fc->pid_ns);
825 put_user_ns(fc->user_ns);
826 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
828 WARN_ON(atomic_read(&bucket->count) != 1);
834 EXPORT_SYMBOL_GPL(fuse_conn_put);
836 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
838 refcount_inc(&fc->count);
841 EXPORT_SYMBOL_GPL(fuse_conn_get);
843 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
845 struct fuse_attr attr;
846 memset(&attr, 0, sizeof(attr));
849 attr.ino = FUSE_ROOT_ID;
851 return fuse_iget(sb, 1, 0, &attr, 0, 0);
854 struct fuse_inode_handle {
859 static struct dentry *fuse_get_dentry(struct super_block *sb,
860 struct fuse_inode_handle *handle)
862 struct fuse_conn *fc = get_fuse_conn_super(sb);
864 struct dentry *entry;
867 if (handle->nodeid == 0)
870 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
872 struct fuse_entry_out outarg;
873 const struct qstr name = QSTR_INIT(".", 1);
875 if (!fc->export_support)
878 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
880 if (err && err != -ENOENT)
887 if (get_node_id(inode) != handle->nodeid)
891 if (inode->i_generation != handle->generation)
894 entry = d_obtain_alias(inode);
895 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
896 fuse_invalidate_entry_cache(entry);
906 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
907 struct inode *parent)
909 int len = parent ? 6 : 3;
913 if (*max_len < len) {
915 return FILEID_INVALID;
918 nodeid = get_fuse_inode(inode)->nodeid;
919 generation = inode->i_generation;
921 fh[0] = (u32)(nodeid >> 32);
922 fh[1] = (u32)(nodeid & 0xffffffff);
926 nodeid = get_fuse_inode(parent)->nodeid;
927 generation = parent->i_generation;
929 fh[3] = (u32)(nodeid >> 32);
930 fh[4] = (u32)(nodeid & 0xffffffff);
935 return parent ? 0x82 : 0x81;
938 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
939 struct fid *fid, int fh_len, int fh_type)
941 struct fuse_inode_handle handle;
943 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
946 handle.nodeid = (u64) fid->raw[0] << 32;
947 handle.nodeid |= (u64) fid->raw[1];
948 handle.generation = fid->raw[2];
949 return fuse_get_dentry(sb, &handle);
952 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
953 struct fid *fid, int fh_len, int fh_type)
955 struct fuse_inode_handle parent;
957 if (fh_type != 0x82 || fh_len < 6)
960 parent.nodeid = (u64) fid->raw[3] << 32;
961 parent.nodeid |= (u64) fid->raw[4];
962 parent.generation = fid->raw[5];
963 return fuse_get_dentry(sb, &parent);
966 static struct dentry *fuse_get_parent(struct dentry *child)
968 struct inode *child_inode = d_inode(child);
969 struct fuse_conn *fc = get_fuse_conn(child_inode);
971 struct dentry *parent;
972 struct fuse_entry_out outarg;
975 if (!fc->export_support)
976 return ERR_PTR(-ESTALE);
978 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
979 &dotdot_name, &outarg, &inode);
982 return ERR_PTR(-ESTALE);
986 parent = d_obtain_alias(inode);
987 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
988 fuse_invalidate_entry_cache(parent);
993 static const struct export_operations fuse_export_operations = {
994 .fh_to_dentry = fuse_fh_to_dentry,
995 .fh_to_parent = fuse_fh_to_parent,
996 .encode_fh = fuse_encode_fh,
997 .get_parent = fuse_get_parent,
1000 static const struct super_operations fuse_super_operations = {
1001 .alloc_inode = fuse_alloc_inode,
1002 .free_inode = fuse_free_inode,
1003 .evict_inode = fuse_evict_inode,
1004 .write_inode = fuse_write_inode,
1005 .drop_inode = generic_delete_inode,
1006 .put_super = fuse_put_super,
1007 .umount_begin = fuse_umount_begin,
1008 .statfs = fuse_statfs,
1009 .sync_fs = fuse_sync_fs,
1010 .show_options = fuse_show_options,
1013 static void sanitize_global_limit(unsigned *limit)
1016 * The default maximum number of async requests is calculated to consume
1017 * 1/2^13 of the total memory, assuming 392 bytes per request.
1020 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1022 if (*limit >= 1 << 16)
1023 *limit = (1 << 16) - 1;
1026 static int set_global_limit(const char *val, const struct kernel_param *kp)
1030 rv = param_set_uint(val, kp);
1034 sanitize_global_limit((unsigned *)kp->arg);
1039 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1041 int cap_sys_admin = capable(CAP_SYS_ADMIN);
1043 if (arg->minor < 13)
1046 sanitize_global_limit(&max_user_bgreq);
1047 sanitize_global_limit(&max_user_congthresh);
1049 spin_lock(&fc->bg_lock);
1050 if (arg->max_background) {
1051 fc->max_background = arg->max_background;
1053 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1054 fc->max_background = max_user_bgreq;
1056 if (arg->congestion_threshold) {
1057 fc->congestion_threshold = arg->congestion_threshold;
1059 if (!cap_sys_admin &&
1060 fc->congestion_threshold > max_user_congthresh)
1061 fc->congestion_threshold = max_user_congthresh;
1063 spin_unlock(&fc->bg_lock);
1066 struct fuse_init_args {
1067 struct fuse_args args;
1068 struct fuse_init_in in;
1069 struct fuse_init_out out;
1072 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1075 struct fuse_conn *fc = fm->fc;
1076 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1077 struct fuse_init_out *arg = &ia->out;
1080 if (error || arg->major != FUSE_KERNEL_VERSION)
1083 unsigned long ra_pages;
1085 process_init_limits(fc, arg);
1087 if (arg->minor >= 6) {
1088 ra_pages = arg->max_readahead / PAGE_SIZE;
1089 if (arg->flags & FUSE_ASYNC_READ)
1091 if (!(arg->flags & FUSE_POSIX_LOCKS))
1093 if (arg->minor >= 17) {
1094 if (!(arg->flags & FUSE_FLOCK_LOCKS))
1097 if (!(arg->flags & FUSE_POSIX_LOCKS))
1100 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1101 fc->atomic_o_trunc = 1;
1102 if (arg->minor >= 9) {
1103 /* LOOKUP has dependency on proto version */
1104 if (arg->flags & FUSE_EXPORT_SUPPORT)
1105 fc->export_support = 1;
1107 if (arg->flags & FUSE_BIG_WRITES)
1109 if (arg->flags & FUSE_DONT_MASK)
1111 if (arg->flags & FUSE_AUTO_INVAL_DATA)
1112 fc->auto_inval_data = 1;
1113 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1114 fc->explicit_inval_data = 1;
1115 if (arg->flags & FUSE_DO_READDIRPLUS) {
1116 fc->do_readdirplus = 1;
1117 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1118 fc->readdirplus_auto = 1;
1120 if (arg->flags & FUSE_ASYNC_DIO)
1122 if (arg->flags & FUSE_WRITEBACK_CACHE)
1123 fc->writeback_cache = 1;
1124 if (arg->flags & FUSE_PARALLEL_DIROPS)
1125 fc->parallel_dirops = 1;
1126 if (arg->flags & FUSE_HANDLE_KILLPRIV)
1127 fc->handle_killpriv = 1;
1128 if (arg->time_gran && arg->time_gran <= 1000000000)
1129 fm->sb->s_time_gran = arg->time_gran;
1130 if ((arg->flags & FUSE_POSIX_ACL)) {
1131 fc->default_permissions = 1;
1133 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1135 if (arg->flags & FUSE_CACHE_SYMLINKS)
1136 fc->cache_symlinks = 1;
1137 if (arg->flags & FUSE_ABORT_ERROR)
1139 if (arg->flags & FUSE_MAX_PAGES) {
1141 min_t(unsigned int, fc->max_pages_limit,
1142 max_t(unsigned int, arg->max_pages, 1));
1144 if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1145 arg->flags & FUSE_MAP_ALIGNMENT &&
1146 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1149 if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) {
1150 fc->handle_killpriv_v2 = 1;
1151 fm->sb->s_flags |= SB_NOSEC;
1153 if (arg->flags & FUSE_SETXATTR_EXT)
1154 fc->setxattr_ext = 1;
1156 ra_pages = fc->max_read / PAGE_SIZE;
1161 fm->sb->s_bdi->ra_pages =
1162 min(fm->sb->s_bdi->ra_pages, ra_pages);
1163 fc->minor = arg->minor;
1164 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1165 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1175 fuse_set_initialized(fc);
1176 wake_up_all(&fc->blocked_waitq);
1179 void fuse_send_init(struct fuse_mount *fm)
1181 struct fuse_init_args *ia;
1183 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1185 ia->in.major = FUSE_KERNEL_VERSION;
1186 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1187 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1189 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1190 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1191 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1192 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1193 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1194 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1195 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1196 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1197 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1198 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT;
1199 #ifdef CONFIG_FUSE_DAX
1201 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1203 if (fm->fc->auto_submounts)
1204 ia->in.flags |= FUSE_SUBMOUNTS;
1206 ia->args.opcode = FUSE_INIT;
1207 ia->args.in_numargs = 1;
1208 ia->args.in_args[0].size = sizeof(ia->in);
1209 ia->args.in_args[0].value = &ia->in;
1210 ia->args.out_numargs = 1;
1211 /* Variable length argument used for backward compatibility
1212 with interface version < 7.5. Rest of init_out is zeroed
1213 by do_get_request(), so a short reply is not a problem */
1214 ia->args.out_argvar = true;
1215 ia->args.out_args[0].size = sizeof(ia->out);
1216 ia->args.out_args[0].value = &ia->out;
1217 ia->args.force = true;
1218 ia->args.nocreds = true;
1219 ia->args.end = process_init_reply;
1221 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1222 process_init_reply(fm, &ia->args, -ENOTCONN);
1224 EXPORT_SYMBOL_GPL(fuse_send_init);
1226 void fuse_free_conn(struct fuse_conn *fc)
1228 WARN_ON(!list_empty(&fc->devices));
1231 EXPORT_SYMBOL_GPL(fuse_free_conn);
1233 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1239 suffix = "-fuseblk";
1241 * sb->s_bdi points to blkdev's bdi however we want to redirect
1242 * it to our private bdi...
1245 sb->s_bdi = &noop_backing_dev_info;
1247 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1248 MINOR(fc->dev), suffix);
1252 /* fuse does it's own writeback accounting */
1253 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1254 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1257 * For a single fuse filesystem use max 1% of dirty +
1258 * writeback threshold.
1260 * This gives about 1M of write buffer for memory maps on a
1261 * machine with 1G and 10% dirty_ratio, which should be more
1264 * Privileged users can raise it by writing to
1266 * /sys/class/bdi/<bdi>/max_ratio
1268 bdi_set_max_ratio(sb->s_bdi, 1);
1273 struct fuse_dev *fuse_dev_alloc(void)
1275 struct fuse_dev *fud;
1276 struct list_head *pq;
1278 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1282 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1288 fud->pq.processing = pq;
1289 fuse_pqueue_init(&fud->pq);
1293 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1295 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1297 fud->fc = fuse_conn_get(fc);
1298 spin_lock(&fc->lock);
1299 list_add_tail(&fud->entry, &fc->devices);
1300 spin_unlock(&fc->lock);
1302 EXPORT_SYMBOL_GPL(fuse_dev_install);
1304 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1306 struct fuse_dev *fud;
1308 fud = fuse_dev_alloc();
1312 fuse_dev_install(fud, fc);
1315 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1317 void fuse_dev_free(struct fuse_dev *fud)
1319 struct fuse_conn *fc = fud->fc;
1322 spin_lock(&fc->lock);
1323 list_del(&fud->entry);
1324 spin_unlock(&fc->lock);
1328 kfree(fud->pq.processing);
1331 EXPORT_SYMBOL_GPL(fuse_dev_free);
1333 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1334 const struct fuse_inode *fi)
1336 *attr = (struct fuse_attr){
1337 .ino = fi->inode.i_ino,
1338 .size = fi->inode.i_size,
1339 .blocks = fi->inode.i_blocks,
1340 .atime = fi->inode.i_atime.tv_sec,
1341 .mtime = fi->inode.i_mtime.tv_sec,
1342 .ctime = fi->inode.i_ctime.tv_sec,
1343 .atimensec = fi->inode.i_atime.tv_nsec,
1344 .mtimensec = fi->inode.i_mtime.tv_nsec,
1345 .ctimensec = fi->inode.i_ctime.tv_nsec,
1346 .mode = fi->inode.i_mode,
1347 .nlink = fi->inode.i_nlink,
1348 .uid = fi->inode.i_uid.val,
1349 .gid = fi->inode.i_gid.val,
1350 .rdev = fi->inode.i_rdev,
1351 .blksize = 1u << fi->inode.i_blkbits,
1355 static void fuse_sb_defaults(struct super_block *sb)
1357 sb->s_magic = FUSE_SUPER_MAGIC;
1358 sb->s_op = &fuse_super_operations;
1359 sb->s_xattr = fuse_xattr_handlers;
1360 sb->s_maxbytes = MAX_LFS_FILESIZE;
1361 sb->s_time_gran = 1;
1362 sb->s_export_op = &fuse_export_operations;
1363 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1364 if (sb->s_user_ns != &init_user_ns)
1365 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1366 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1369 * If we are not in the initial user namespace posix
1370 * acls must be translated.
1372 if (sb->s_user_ns != &init_user_ns)
1373 sb->s_xattr = fuse_no_acl_xattr_handlers;
1376 static int fuse_fill_super_submount(struct super_block *sb,
1377 struct fuse_inode *parent_fi)
1379 struct fuse_mount *fm = get_fuse_mount_super(sb);
1380 struct super_block *parent_sb = parent_fi->inode.i_sb;
1381 struct fuse_attr root_attr;
1384 fuse_sb_defaults(sb);
1387 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1388 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1390 sb->s_xattr = parent_sb->s_xattr;
1391 sb->s_time_gran = parent_sb->s_time_gran;
1392 sb->s_blocksize = parent_sb->s_blocksize;
1393 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1394 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1395 if (parent_sb->s_subtype && !sb->s_subtype)
1398 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1399 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1401 * This inode is just a duplicate, so it is not looked up and
1402 * its nlookup should not be incremented. fuse_iget() does
1403 * that, though, so undo it here.
1405 get_fuse_inode(root)->nlookup--;
1406 sb->s_d_op = &fuse_dentry_operations;
1407 sb->s_root = d_make_root(root);
1414 /* Filesystem context private data holds the FUSE inode of the mount point */
1415 static int fuse_get_tree_submount(struct fs_context *fsc)
1417 struct fuse_mount *fm;
1418 struct fuse_inode *mp_fi = fsc->fs_private;
1419 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1420 struct super_block *sb;
1423 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1427 fsc->s_fs_info = fm;
1428 sb = sget_fc(fsc, NULL, set_anon_super_fc);
1433 fm->fc = fuse_conn_get(fc);
1435 /* Initialize superblock, making @mp_fi its root */
1436 err = fuse_fill_super_submount(sb, mp_fi);
1440 sb->s_fs_info = NULL;
1441 deactivate_locked_super(sb);
1445 down_write(&fc->killsb);
1446 list_add_tail(&fm->fc_entry, &fc->mounts);
1447 up_write(&fc->killsb);
1449 sb->s_flags |= SB_ACTIVE;
1450 fsc->root = dget(sb->s_root);
1455 static const struct fs_context_operations fuse_context_submount_ops = {
1456 .get_tree = fuse_get_tree_submount,
1459 int fuse_init_fs_context_submount(struct fs_context *fsc)
1461 fsc->ops = &fuse_context_submount_ops;
1464 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1466 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1468 struct fuse_dev *fud = NULL;
1469 struct fuse_mount *fm = get_fuse_mount_super(sb);
1470 struct fuse_conn *fc = fm->fc;
1472 struct dentry *root_dentry;
1476 if (sb->s_flags & SB_MANDLOCK)
1479 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1480 fuse_sb_defaults(sb);
1485 if (!sb_set_blocksize(sb, ctx->blksize))
1489 sb->s_blocksize = PAGE_SIZE;
1490 sb->s_blocksize_bits = PAGE_SHIFT;
1493 sb->s_subtype = ctx->subtype;
1494 ctx->subtype = NULL;
1495 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1496 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1503 fud = fuse_dev_alloc_install(fc);
1508 fc->dev = sb->s_dev;
1510 err = fuse_bdi_init(fc, sb);
1514 /* Handle umasking inside the fuse code */
1515 if (sb->s_flags & SB_POSIXACL)
1517 sb->s_flags |= SB_POSIXACL;
1519 fc->default_permissions = ctx->default_permissions;
1520 fc->allow_other = ctx->allow_other;
1521 fc->user_id = ctx->user_id;
1522 fc->group_id = ctx->group_id;
1523 fc->legacy_opts_show = ctx->legacy_opts_show;
1524 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1525 fc->destroy = ctx->destroy;
1526 fc->no_control = ctx->no_control;
1527 fc->no_force_umount = ctx->no_force_umount;
1530 root = fuse_get_root_inode(sb, ctx->rootmode);
1531 sb->s_d_op = &fuse_root_dentry_operations;
1532 root_dentry = d_make_root(root);
1535 /* Root dentry doesn't have .d_revalidate */
1536 sb->s_d_op = &fuse_dentry_operations;
1538 mutex_lock(&fuse_mutex);
1540 if (ctx->fudptr && *ctx->fudptr)
1543 err = fuse_ctl_add_conn(fc);
1547 list_add_tail(&fc->entry, &fuse_conn_list);
1548 sb->s_root = root_dentry;
1551 mutex_unlock(&fuse_mutex);
1555 mutex_unlock(&fuse_mutex);
1561 if (IS_ENABLED(CONFIG_FUSE_DAX))
1562 fuse_dax_conn_free(fc);
1566 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1568 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1570 struct fuse_fs_context *ctx = fsc->fs_private;
1572 struct fuse_conn *fc;
1573 struct fuse_mount *fm;
1575 if (!ctx->file || !ctx->rootmode_present ||
1576 !ctx->user_id_present || !ctx->group_id_present)
1580 * Require mount to happen from the same user namespace which
1581 * opened /dev/fuse to prevent potential attacks.
1584 if ((ctx->file->f_op != &fuse_dev_operations) ||
1585 (ctx->file->f_cred->user_ns != sb->s_user_ns))
1587 ctx->fudptr = &ctx->file->private_data;
1589 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1594 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1600 fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1601 fc->release = fuse_free_conn;
1605 err = fuse_fill_super_common(sb, ctx);
1608 /* file->private_data shall be visible on all CPUs after this */
1610 fuse_send_init(get_fuse_mount_super(sb));
1616 sb->s_fs_info = NULL;
1622 * This is the path where user supplied an already initialized fuse dev. In
1623 * this case never create a new super if the old one is gone.
1625 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1630 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1633 return fsc->sget_key == get_fuse_conn_super(sb);
1636 static int fuse_get_tree(struct fs_context *fsc)
1638 struct fuse_fs_context *ctx = fsc->fs_private;
1639 struct fuse_dev *fud;
1640 struct super_block *sb;
1643 if (ctx->fd_present)
1644 ctx->file = fget(ctx->fd);
1646 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1647 err = get_tree_bdev(fsc, fuse_fill_super);
1651 * While block dev mount can be initialized with a dummy device fd
1652 * (found by device name), normal fuse mounts can't
1658 * Allow creating a fuse mount with an already initialized fuse
1661 fud = READ_ONCE(ctx->file->private_data);
1662 if (ctx->file->f_op == &fuse_dev_operations && fud) {
1663 fsc->sget_key = fud->fc;
1664 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1665 err = PTR_ERR_OR_ZERO(sb);
1667 fsc->root = dget(sb->s_root);
1669 err = get_tree_nodev(fsc, fuse_fill_super);
1677 static const struct fs_context_operations fuse_context_ops = {
1678 .free = fuse_free_fsc,
1679 .parse_param = fuse_parse_param,
1680 .reconfigure = fuse_reconfigure,
1681 .get_tree = fuse_get_tree,
1685 * Set up the filesystem mount context.
1687 static int fuse_init_fs_context(struct fs_context *fsc)
1689 struct fuse_fs_context *ctx;
1691 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1696 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1697 ctx->legacy_opts_show = true;
1700 if (fsc->fs_type == &fuseblk_fs_type) {
1701 ctx->is_bdev = true;
1702 ctx->destroy = true;
1706 fsc->fs_private = ctx;
1707 fsc->ops = &fuse_context_ops;
1711 bool fuse_mount_remove(struct fuse_mount *fm)
1713 struct fuse_conn *fc = fm->fc;
1716 down_write(&fc->killsb);
1717 list_del_init(&fm->fc_entry);
1718 if (list_empty(&fc->mounts))
1720 up_write(&fc->killsb);
1724 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1726 void fuse_conn_destroy(struct fuse_mount *fm)
1728 struct fuse_conn *fc = fm->fc;
1731 fuse_send_destroy(fm);
1733 fuse_abort_conn(fc);
1734 fuse_wait_aborted(fc);
1736 if (!list_empty(&fc->entry)) {
1737 mutex_lock(&fuse_mutex);
1738 list_del(&fc->entry);
1739 fuse_ctl_remove_conn(fc);
1740 mutex_unlock(&fuse_mutex);
1743 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1745 static void fuse_sb_destroy(struct super_block *sb)
1747 struct fuse_mount *fm = get_fuse_mount_super(sb);
1751 last = fuse_mount_remove(fm);
1753 fuse_conn_destroy(fm);
1757 static void fuse_kill_sb_anon(struct super_block *sb)
1759 fuse_sb_destroy(sb);
1760 kill_anon_super(sb);
1763 static struct file_system_type fuse_fs_type = {
1764 .owner = THIS_MODULE,
1766 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1767 .init_fs_context = fuse_init_fs_context,
1768 .parameters = fuse_fs_parameters,
1769 .kill_sb = fuse_kill_sb_anon,
1771 MODULE_ALIAS_FS("fuse");
1774 static void fuse_kill_sb_blk(struct super_block *sb)
1776 fuse_sb_destroy(sb);
1777 kill_block_super(sb);
1780 static struct file_system_type fuseblk_fs_type = {
1781 .owner = THIS_MODULE,
1783 .init_fs_context = fuse_init_fs_context,
1784 .parameters = fuse_fs_parameters,
1785 .kill_sb = fuse_kill_sb_blk,
1786 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1788 MODULE_ALIAS_FS("fuseblk");
1790 static inline int register_fuseblk(void)
1792 return register_filesystem(&fuseblk_fs_type);
1795 static inline void unregister_fuseblk(void)
1797 unregister_filesystem(&fuseblk_fs_type);
1800 static inline int register_fuseblk(void)
1805 static inline void unregister_fuseblk(void)
1810 static void fuse_inode_init_once(void *foo)
1812 struct inode *inode = foo;
1814 inode_init_once(inode);
1817 static int __init fuse_fs_init(void)
1821 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1822 sizeof(struct fuse_inode), 0,
1823 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1824 fuse_inode_init_once);
1826 if (!fuse_inode_cachep)
1829 err = register_fuseblk();
1833 err = register_filesystem(&fuse_fs_type);
1840 unregister_fuseblk();
1842 kmem_cache_destroy(fuse_inode_cachep);
1847 static void fuse_fs_cleanup(void)
1849 unregister_filesystem(&fuse_fs_type);
1850 unregister_fuseblk();
1853 * Make sure all delayed rcu free inodes are flushed before we
1857 kmem_cache_destroy(fuse_inode_cachep);
1860 static struct kobject *fuse_kobj;
1862 static int fuse_sysfs_init(void)
1866 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1872 err = sysfs_create_mount_point(fuse_kobj, "connections");
1874 goto out_fuse_unregister;
1878 out_fuse_unregister:
1879 kobject_put(fuse_kobj);
1884 static void fuse_sysfs_cleanup(void)
1886 sysfs_remove_mount_point(fuse_kobj, "connections");
1887 kobject_put(fuse_kobj);
1890 static int __init fuse_init(void)
1894 pr_info("init (API version %i.%i)\n",
1895 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1897 INIT_LIST_HEAD(&fuse_conn_list);
1898 res = fuse_fs_init();
1902 res = fuse_dev_init();
1904 goto err_fs_cleanup;
1906 res = fuse_sysfs_init();
1908 goto err_dev_cleanup;
1910 res = fuse_ctl_init();
1912 goto err_sysfs_cleanup;
1914 sanitize_global_limit(&max_user_bgreq);
1915 sanitize_global_limit(&max_user_congthresh);
1920 fuse_sysfs_cleanup();
1929 static void __exit fuse_exit(void)
1934 fuse_sysfs_cleanup();
1939 module_init(fuse_init);
1940 module_exit(fuse_exit);