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>
26 #include <uapi/linux/magic.h>
28 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
29 MODULE_DESCRIPTION("Filesystem in Userspace");
30 MODULE_LICENSE("GPL");
32 static struct kmem_cache *fuse_inode_cachep;
33 struct list_head fuse_conn_list;
34 DEFINE_MUTEX(fuse_mutex);
36 static int set_global_limit(const char *val, const struct kernel_param *kp);
38 unsigned max_user_bgreq;
39 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
40 &max_user_bgreq, 0644);
41 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
42 MODULE_PARM_DESC(max_user_bgreq,
43 "Global limit for the maximum number of backgrounded requests an "
44 "unprivileged user can set");
46 unsigned max_user_congthresh;
47 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
48 &max_user_congthresh, 0644);
49 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
50 MODULE_PARM_DESC(max_user_congthresh,
51 "Global limit for the maximum congestion threshold an "
52 "unprivileged user can set");
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)
63 static struct file_system_type fuseblk_fs_type;
66 struct fuse_forget_link *fuse_alloc_forget(void)
68 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
71 static struct inode *fuse_alloc_inode(struct super_block *sb)
73 struct fuse_inode *fi;
75 fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
86 mutex_init(&fi->mutex);
87 spin_lock_init(&fi->lock);
88 fi->forget = fuse_alloc_forget();
92 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
100 kmem_cache_free(fuse_inode_cachep, fi);
104 static void fuse_free_inode(struct inode *inode)
106 struct fuse_inode *fi = get_fuse_inode(inode);
108 mutex_destroy(&fi->mutex);
110 #ifdef CONFIG_FUSE_DAX
113 kmem_cache_free(fuse_inode_cachep, fi);
116 static void fuse_evict_inode(struct inode *inode)
118 struct fuse_inode *fi = get_fuse_inode(inode);
120 /* Will write inode on close/munmap and in all other dirtiers */
121 WARN_ON(inode->i_state & I_DIRTY_INODE);
123 truncate_inode_pages_final(&inode->i_data);
125 if (inode->i_sb->s_flags & SB_ACTIVE) {
126 struct fuse_conn *fc = get_fuse_conn(inode);
128 if (FUSE_IS_DAX(inode))
129 fuse_dax_inode_cleanup(inode);
131 fuse_queue_forget(fc, fi->forget, fi->nodeid,
136 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
137 WARN_ON(!list_empty(&fi->write_files));
138 WARN_ON(!list_empty(&fi->queued_writes));
142 static int fuse_reconfigure(struct fs_context *fsc)
144 struct super_block *sb = fsc->root->d_sb;
147 if (fsc->sb_flags & SB_MANDLOCK)
154 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
155 * so that it will fit.
157 static ino_t fuse_squash_ino(u64 ino64)
159 ino_t ino = (ino_t) ino64;
160 if (sizeof(ino_t) < sizeof(u64))
161 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
165 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
166 u64 attr_valid, u32 cache_mask)
168 struct fuse_conn *fc = get_fuse_conn(inode);
169 struct fuse_inode *fi = get_fuse_inode(inode);
171 lockdep_assert_held(&fi->lock);
173 fi->attr_version = atomic64_inc_return(&fc->attr_version);
174 fi->i_time = attr_valid;
175 WRITE_ONCE(fi->inval_mask, 0);
177 inode->i_ino = fuse_squash_ino(attr->ino);
178 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
179 set_nlink(inode, attr->nlink);
180 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
181 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
182 inode->i_blocks = attr->blocks;
185 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
186 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
187 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
189 inode->i_atime.tv_sec = attr->atime;
190 inode->i_atime.tv_nsec = attr->atimensec;
191 /* mtime from server may be stale due to local buffered write */
192 if (!(cache_mask & STATX_MTIME)) {
193 inode->i_mtime.tv_sec = attr->mtime;
194 inode->i_mtime.tv_nsec = attr->mtimensec;
196 if (!(cache_mask & STATX_CTIME)) {
197 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
200 if (attr->blksize != 0)
201 inode->i_blkbits = ilog2(attr->blksize);
203 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
206 * Don't set the sticky bit in i_mode, unless we want the VFS
207 * to check permissions. This prevents failures due to the
208 * check in may_delete().
210 fi->orig_i_mode = inode->i_mode;
211 if (!fc->default_permissions)
212 inode->i_mode &= ~S_ISVTX;
214 fi->orig_ino = attr->ino;
217 * We are refreshing inode data and it is possible that another
218 * client set suid/sgid or security.capability xattr. So clear
219 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
220 * was set or if security.capability xattr was set. But we don't
221 * know if security.capability has been set or not. So clear it
222 * anyway. Its less efficient but should be safe.
224 inode->i_flags &= ~S_NOSEC;
227 u32 fuse_get_cache_mask(struct inode *inode)
229 struct fuse_conn *fc = get_fuse_conn(inode);
231 if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
234 return STATX_MTIME | STATX_CTIME | STATX_SIZE;
237 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
238 u64 attr_valid, u64 attr_version)
240 struct fuse_conn *fc = get_fuse_conn(inode);
241 struct fuse_inode *fi = get_fuse_inode(inode);
244 struct timespec64 old_mtime;
246 spin_lock(&fi->lock);
248 * In case of writeback_cache enabled, writes update mtime, ctime and
249 * may update i_size. In these cases trust the cached value in the
252 cache_mask = fuse_get_cache_mask(inode);
253 if (cache_mask & STATX_SIZE)
254 attr->size = i_size_read(inode);
256 if (cache_mask & STATX_MTIME) {
257 attr->mtime = inode->i_mtime.tv_sec;
258 attr->mtimensec = inode->i_mtime.tv_nsec;
260 if (cache_mask & STATX_CTIME) {
261 attr->ctime = inode_get_ctime(inode).tv_sec;
262 attr->ctimensec = inode_get_ctime(inode).tv_nsec;
265 if ((attr_version != 0 && fi->attr_version > attr_version) ||
266 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
267 spin_unlock(&fi->lock);
271 old_mtime = inode->i_mtime;
272 fuse_change_attributes_common(inode, attr, attr_valid, cache_mask);
274 oldsize = inode->i_size;
276 * In case of writeback_cache enabled, the cached writes beyond EOF
277 * extend local i_size without keeping userspace server in sync. So,
278 * attr->size coming from server can be stale. We cannot trust it.
280 if (!(cache_mask & STATX_SIZE))
281 i_size_write(inode, attr->size);
282 spin_unlock(&fi->lock);
284 if (!cache_mask && S_ISREG(inode->i_mode)) {
287 if (oldsize != attr->size) {
288 truncate_pagecache(inode, attr->size);
289 if (!fc->explicit_inval_data)
291 } else if (fc->auto_inval_data) {
292 struct timespec64 new_mtime = {
293 .tv_sec = attr->mtime,
294 .tv_nsec = attr->mtimensec,
298 * Auto inval mode also checks and invalidates if mtime
301 if (!timespec64_equal(&old_mtime, &new_mtime))
306 invalidate_inode_pages2(inode->i_mapping);
309 if (IS_ENABLED(CONFIG_FUSE_DAX))
310 fuse_dax_dontcache(inode, attr->flags);
313 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
314 struct fuse_conn *fc)
316 inode->i_mode = attr->mode & S_IFMT;
317 inode->i_size = attr->size;
318 inode->i_mtime.tv_sec = attr->mtime;
319 inode->i_mtime.tv_nsec = attr->mtimensec;
320 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
321 if (S_ISREG(inode->i_mode)) {
322 fuse_init_common(inode);
323 fuse_init_file_inode(inode, attr->flags);
324 } else if (S_ISDIR(inode->i_mode))
325 fuse_init_dir(inode);
326 else if (S_ISLNK(inode->i_mode))
327 fuse_init_symlink(inode);
328 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
329 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
330 fuse_init_common(inode);
331 init_special_inode(inode, inode->i_mode,
332 new_decode_dev(attr->rdev));
336 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
337 * so they see the exact same behavior as before.
340 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
343 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
345 u64 nodeid = *(u64 *) _nodeidp;
346 if (get_node_id(inode) == nodeid)
352 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
354 u64 nodeid = *(u64 *) _nodeidp;
355 get_fuse_inode(inode)->nodeid = nodeid;
359 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
360 int generation, struct fuse_attr *attr,
361 u64 attr_valid, u64 attr_version)
364 struct fuse_inode *fi;
365 struct fuse_conn *fc = get_fuse_conn_super(sb);
368 * Auto mount points get their node id from the submount root, which is
369 * not a unique identifier within this filesystem.
371 * To avoid conflicts, do not place submount points into the inode hash
374 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
375 S_ISDIR(attr->mode)) {
376 inode = new_inode(sb);
380 fuse_init_inode(inode, attr, fc);
381 get_fuse_inode(inode)->nodeid = nodeid;
382 inode->i_flags |= S_AUTOMOUNT;
387 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
391 if ((inode->i_state & I_NEW)) {
392 inode->i_flags |= S_NOATIME;
393 if (!fc->writeback_cache || !S_ISREG(attr->mode))
394 inode->i_flags |= S_NOCMTIME;
395 inode->i_generation = generation;
396 fuse_init_inode(inode, attr, fc);
397 unlock_new_inode(inode);
398 } else if (fuse_stale_inode(inode, generation, attr)) {
399 /* nodeid was reused, any I/O on the old inode should fail */
400 fuse_make_bad(inode);
405 fi = get_fuse_inode(inode);
406 spin_lock(&fi->lock);
408 spin_unlock(&fi->lock);
409 fuse_change_attributes(inode, attr, attr_valid, attr_version);
414 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
415 struct fuse_mount **fm)
417 struct fuse_mount *fm_iter;
420 WARN_ON(!rwsem_is_locked(&fc->killsb));
421 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
425 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
436 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
437 loff_t offset, loff_t len)
439 struct fuse_inode *fi;
444 inode = fuse_ilookup(fc, nodeid, NULL);
448 fi = get_fuse_inode(inode);
449 spin_lock(&fi->lock);
450 fi->attr_version = atomic64_inc_return(&fc->attr_version);
451 spin_unlock(&fi->lock);
453 fuse_invalidate_attr(inode);
454 forget_all_cached_acls(inode);
456 pg_start = offset >> PAGE_SHIFT;
460 pg_end = (offset + len - 1) >> PAGE_SHIFT;
461 invalidate_inode_pages2_range(inode->i_mapping,
468 bool fuse_lock_inode(struct inode *inode)
472 if (!get_fuse_conn(inode)->parallel_dirops) {
473 mutex_lock(&get_fuse_inode(inode)->mutex);
480 void fuse_unlock_inode(struct inode *inode, bool locked)
483 mutex_unlock(&get_fuse_inode(inode)->mutex);
486 static void fuse_umount_begin(struct super_block *sb)
488 struct fuse_conn *fc = get_fuse_conn_super(sb);
490 if (fc->no_force_umount)
495 // Only retire block-device-based superblocks.
496 if (sb->s_bdev != NULL)
500 static void fuse_send_destroy(struct fuse_mount *fm)
502 if (fm->fc->conn_init) {
505 args.opcode = FUSE_DESTROY;
508 fuse_simple_request(fm, &args);
512 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
514 stbuf->f_type = FUSE_SUPER_MAGIC;
515 stbuf->f_bsize = attr->bsize;
516 stbuf->f_frsize = attr->frsize;
517 stbuf->f_blocks = attr->blocks;
518 stbuf->f_bfree = attr->bfree;
519 stbuf->f_bavail = attr->bavail;
520 stbuf->f_files = attr->files;
521 stbuf->f_ffree = attr->ffree;
522 stbuf->f_namelen = attr->namelen;
523 /* fsid is left zero */
526 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
528 struct super_block *sb = dentry->d_sb;
529 struct fuse_mount *fm = get_fuse_mount_super(sb);
531 struct fuse_statfs_out outarg;
534 if (!fuse_allow_current_process(fm->fc)) {
535 buf->f_type = FUSE_SUPER_MAGIC;
539 memset(&outarg, 0, sizeof(outarg));
541 args.opcode = FUSE_STATFS;
542 args.nodeid = get_node_id(d_inode(dentry));
543 args.out_numargs = 1;
544 args.out_args[0].size = sizeof(outarg);
545 args.out_args[0].value = &outarg;
546 err = fuse_simple_request(fm, &args);
548 convert_fuse_statfs(buf, &outarg.st);
552 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
554 struct fuse_sync_bucket *bucket;
556 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
558 init_waitqueue_head(&bucket->waitq);
559 /* Initial active count */
560 atomic_set(&bucket->count, 1);
565 static void fuse_sync_fs_writes(struct fuse_conn *fc)
567 struct fuse_sync_bucket *bucket, *new_bucket;
570 new_bucket = fuse_sync_bucket_alloc();
571 spin_lock(&fc->lock);
572 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
573 count = atomic_read(&bucket->count);
575 /* No outstanding writes? */
577 spin_unlock(&fc->lock);
583 * Completion of new bucket depends on completion of this bucket, so add
586 atomic_inc(&new_bucket->count);
587 rcu_assign_pointer(fc->curr_bucket, new_bucket);
588 spin_unlock(&fc->lock);
590 * Drop initial active count. At this point if all writes in this and
591 * ancestor buckets complete, the count will go to zero and this task
594 atomic_dec(&bucket->count);
596 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
598 /* Drop temp count on descendant bucket */
599 fuse_sync_bucket_dec(new_bucket);
600 kfree_rcu(bucket, rcu);
603 static int fuse_sync_fs(struct super_block *sb, int wait)
605 struct fuse_mount *fm = get_fuse_mount_super(sb);
606 struct fuse_conn *fc = fm->fc;
607 struct fuse_syncfs_in inarg;
612 * Userspace cannot handle the wait == 0 case. Avoid a
613 * gratuitous roundtrip.
618 /* The filesystem is being unmounted. Nothing to do. */
625 fuse_sync_fs_writes(fc);
627 memset(&inarg, 0, sizeof(inarg));
629 args.in_args[0].size = sizeof(inarg);
630 args.in_args[0].value = &inarg;
631 args.opcode = FUSE_SYNCFS;
632 args.nodeid = get_node_id(sb->s_root->d_inode);
633 args.out_numargs = 0;
635 err = fuse_simple_request(fm, &args);
636 if (err == -ENOSYS) {
651 OPT_DEFAULT_PERMISSIONS,
658 static const struct fs_parameter_spec fuse_fs_parameters[] = {
659 fsparam_string ("source", OPT_SOURCE),
660 fsparam_u32 ("fd", OPT_FD),
661 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
662 fsparam_u32 ("user_id", OPT_USER_ID),
663 fsparam_u32 ("group_id", OPT_GROUP_ID),
664 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
665 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
666 fsparam_u32 ("max_read", OPT_MAX_READ),
667 fsparam_u32 ("blksize", OPT_BLKSIZE),
668 fsparam_string ("subtype", OPT_SUBTYPE),
672 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
674 struct fs_parse_result result;
675 struct fuse_fs_context *ctx = fsc->fs_private;
678 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
680 * Ignore options coming from mount(MS_REMOUNT) for backward
686 return invalfc(fsc, "No changes allowed in reconfigure");
689 opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
696 return invalfc(fsc, "Multiple sources specified");
697 fsc->source = param->string;
698 param->string = NULL;
703 return invalfc(fsc, "Multiple subtypes specified");
704 ctx->subtype = param->string;
705 param->string = NULL;
709 ctx->fd = result.uint_32;
710 ctx->fd_present = true;
714 if (!fuse_valid_type(result.uint_32))
715 return invalfc(fsc, "Invalid rootmode");
716 ctx->rootmode = result.uint_32;
717 ctx->rootmode_present = true;
721 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
722 if (!uid_valid(ctx->user_id))
723 return invalfc(fsc, "Invalid user_id");
724 ctx->user_id_present = true;
728 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
729 if (!gid_valid(ctx->group_id))
730 return invalfc(fsc, "Invalid group_id");
731 ctx->group_id_present = true;
734 case OPT_DEFAULT_PERMISSIONS:
735 ctx->default_permissions = true;
738 case OPT_ALLOW_OTHER:
739 ctx->allow_other = true;
743 ctx->max_read = result.uint_32;
748 return invalfc(fsc, "blksize only supported for fuseblk");
749 ctx->blksize = result.uint_32;
759 static void fuse_free_fsc(struct fs_context *fsc)
761 struct fuse_fs_context *ctx = fsc->fs_private;
769 static int fuse_show_options(struct seq_file *m, struct dentry *root)
771 struct super_block *sb = root->d_sb;
772 struct fuse_conn *fc = get_fuse_conn_super(sb);
774 if (fc->legacy_opts_show) {
775 seq_printf(m, ",user_id=%u",
776 from_kuid_munged(fc->user_ns, fc->user_id));
777 seq_printf(m, ",group_id=%u",
778 from_kgid_munged(fc->user_ns, fc->group_id));
779 if (fc->default_permissions)
780 seq_puts(m, ",default_permissions");
782 seq_puts(m, ",allow_other");
783 if (fc->max_read != ~0)
784 seq_printf(m, ",max_read=%u", fc->max_read);
785 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
786 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
788 #ifdef CONFIG_FUSE_DAX
789 if (fc->dax_mode == FUSE_DAX_ALWAYS)
790 seq_puts(m, ",dax=always");
791 else if (fc->dax_mode == FUSE_DAX_NEVER)
792 seq_puts(m, ",dax=never");
793 else if (fc->dax_mode == FUSE_DAX_INODE_USER)
794 seq_puts(m, ",dax=inode");
800 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
801 const struct fuse_iqueue_ops *ops,
804 memset(fiq, 0, sizeof(struct fuse_iqueue));
805 spin_lock_init(&fiq->lock);
806 init_waitqueue_head(&fiq->waitq);
807 INIT_LIST_HEAD(&fiq->pending);
808 INIT_LIST_HEAD(&fiq->interrupts);
809 fiq->forget_list_tail = &fiq->forget_list_head;
815 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
819 spin_lock_init(&fpq->lock);
820 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
821 INIT_LIST_HEAD(&fpq->processing[i]);
822 INIT_LIST_HEAD(&fpq->io);
826 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
827 struct user_namespace *user_ns,
828 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
830 memset(fc, 0, sizeof(*fc));
831 spin_lock_init(&fc->lock);
832 spin_lock_init(&fc->bg_lock);
833 init_rwsem(&fc->killsb);
834 refcount_set(&fc->count, 1);
835 atomic_set(&fc->dev_count, 1);
836 init_waitqueue_head(&fc->blocked_waitq);
837 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
838 INIT_LIST_HEAD(&fc->bg_queue);
839 INIT_LIST_HEAD(&fc->entry);
840 INIT_LIST_HEAD(&fc->devices);
841 atomic_set(&fc->num_waiting, 0);
842 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
843 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
844 atomic64_set(&fc->khctr, 0);
845 fc->polled_files = RB_ROOT;
849 atomic64_set(&fc->attr_version, 1);
850 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
851 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
852 fc->user_ns = get_user_ns(user_ns);
853 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
854 fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
856 INIT_LIST_HEAD(&fc->mounts);
857 list_add(&fm->fc_entry, &fc->mounts);
860 EXPORT_SYMBOL_GPL(fuse_conn_init);
862 void fuse_conn_put(struct fuse_conn *fc)
864 if (refcount_dec_and_test(&fc->count)) {
865 struct fuse_iqueue *fiq = &fc->iq;
866 struct fuse_sync_bucket *bucket;
868 if (IS_ENABLED(CONFIG_FUSE_DAX))
869 fuse_dax_conn_free(fc);
870 if (fiq->ops->release)
871 fiq->ops->release(fiq);
872 put_pid_ns(fc->pid_ns);
873 put_user_ns(fc->user_ns);
874 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
876 WARN_ON(atomic_read(&bucket->count) != 1);
882 EXPORT_SYMBOL_GPL(fuse_conn_put);
884 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
886 refcount_inc(&fc->count);
889 EXPORT_SYMBOL_GPL(fuse_conn_get);
891 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
893 struct fuse_attr attr;
894 memset(&attr, 0, sizeof(attr));
897 attr.ino = FUSE_ROOT_ID;
899 return fuse_iget(sb, 1, 0, &attr, 0, 0);
902 struct fuse_inode_handle {
907 static struct dentry *fuse_get_dentry(struct super_block *sb,
908 struct fuse_inode_handle *handle)
910 struct fuse_conn *fc = get_fuse_conn_super(sb);
912 struct dentry *entry;
915 if (handle->nodeid == 0)
918 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
920 struct fuse_entry_out outarg;
921 const struct qstr name = QSTR_INIT(".", 1);
923 if (!fc->export_support)
926 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
928 if (err && err != -ENOENT)
935 if (get_node_id(inode) != handle->nodeid)
939 if (inode->i_generation != handle->generation)
942 entry = d_obtain_alias(inode);
943 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
944 fuse_invalidate_entry_cache(entry);
954 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
955 struct inode *parent)
957 int len = parent ? 6 : 3;
961 if (*max_len < len) {
963 return FILEID_INVALID;
966 nodeid = get_fuse_inode(inode)->nodeid;
967 generation = inode->i_generation;
969 fh[0] = (u32)(nodeid >> 32);
970 fh[1] = (u32)(nodeid & 0xffffffff);
974 nodeid = get_fuse_inode(parent)->nodeid;
975 generation = parent->i_generation;
977 fh[3] = (u32)(nodeid >> 32);
978 fh[4] = (u32)(nodeid & 0xffffffff);
983 return parent ? 0x82 : 0x81;
986 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
987 struct fid *fid, int fh_len, int fh_type)
989 struct fuse_inode_handle handle;
991 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
994 handle.nodeid = (u64) fid->raw[0] << 32;
995 handle.nodeid |= (u64) fid->raw[1];
996 handle.generation = fid->raw[2];
997 return fuse_get_dentry(sb, &handle);
1000 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1001 struct fid *fid, int fh_len, int fh_type)
1003 struct fuse_inode_handle parent;
1005 if (fh_type != 0x82 || fh_len < 6)
1008 parent.nodeid = (u64) fid->raw[3] << 32;
1009 parent.nodeid |= (u64) fid->raw[4];
1010 parent.generation = fid->raw[5];
1011 return fuse_get_dentry(sb, &parent);
1014 static struct dentry *fuse_get_parent(struct dentry *child)
1016 struct inode *child_inode = d_inode(child);
1017 struct fuse_conn *fc = get_fuse_conn(child_inode);
1018 struct inode *inode;
1019 struct dentry *parent;
1020 struct fuse_entry_out outarg;
1023 if (!fc->export_support)
1024 return ERR_PTR(-ESTALE);
1026 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1027 &dotdot_name, &outarg, &inode);
1030 return ERR_PTR(-ESTALE);
1031 return ERR_PTR(err);
1034 parent = d_obtain_alias(inode);
1035 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1036 fuse_invalidate_entry_cache(parent);
1041 static const struct export_operations fuse_export_operations = {
1042 .fh_to_dentry = fuse_fh_to_dentry,
1043 .fh_to_parent = fuse_fh_to_parent,
1044 .encode_fh = fuse_encode_fh,
1045 .get_parent = fuse_get_parent,
1048 static const struct super_operations fuse_super_operations = {
1049 .alloc_inode = fuse_alloc_inode,
1050 .free_inode = fuse_free_inode,
1051 .evict_inode = fuse_evict_inode,
1052 .write_inode = fuse_write_inode,
1053 .drop_inode = generic_delete_inode,
1054 .umount_begin = fuse_umount_begin,
1055 .statfs = fuse_statfs,
1056 .sync_fs = fuse_sync_fs,
1057 .show_options = fuse_show_options,
1060 static void sanitize_global_limit(unsigned *limit)
1063 * The default maximum number of async requests is calculated to consume
1064 * 1/2^13 of the total memory, assuming 392 bytes per request.
1067 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1069 if (*limit >= 1 << 16)
1070 *limit = (1 << 16) - 1;
1073 static int set_global_limit(const char *val, const struct kernel_param *kp)
1077 rv = param_set_uint(val, kp);
1081 sanitize_global_limit((unsigned *)kp->arg);
1086 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1088 int cap_sys_admin = capable(CAP_SYS_ADMIN);
1090 if (arg->minor < 13)
1093 sanitize_global_limit(&max_user_bgreq);
1094 sanitize_global_limit(&max_user_congthresh);
1096 spin_lock(&fc->bg_lock);
1097 if (arg->max_background) {
1098 fc->max_background = arg->max_background;
1100 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1101 fc->max_background = max_user_bgreq;
1103 if (arg->congestion_threshold) {
1104 fc->congestion_threshold = arg->congestion_threshold;
1106 if (!cap_sys_admin &&
1107 fc->congestion_threshold > max_user_congthresh)
1108 fc->congestion_threshold = max_user_congthresh;
1110 spin_unlock(&fc->bg_lock);
1113 struct fuse_init_args {
1114 struct fuse_args args;
1115 struct fuse_init_in in;
1116 struct fuse_init_out out;
1119 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1122 struct fuse_conn *fc = fm->fc;
1123 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1124 struct fuse_init_out *arg = &ia->out;
1127 if (error || arg->major != FUSE_KERNEL_VERSION)
1130 unsigned long ra_pages;
1132 process_init_limits(fc, arg);
1134 if (arg->minor >= 6) {
1135 u64 flags = arg->flags;
1137 if (flags & FUSE_INIT_EXT)
1138 flags |= (u64) arg->flags2 << 32;
1140 ra_pages = arg->max_readahead / PAGE_SIZE;
1141 if (flags & FUSE_ASYNC_READ)
1143 if (!(flags & FUSE_POSIX_LOCKS))
1145 if (arg->minor >= 17) {
1146 if (!(flags & FUSE_FLOCK_LOCKS))
1149 if (!(flags & FUSE_POSIX_LOCKS))
1152 if (flags & FUSE_ATOMIC_O_TRUNC)
1153 fc->atomic_o_trunc = 1;
1154 if (arg->minor >= 9) {
1155 /* LOOKUP has dependency on proto version */
1156 if (flags & FUSE_EXPORT_SUPPORT)
1157 fc->export_support = 1;
1159 if (flags & FUSE_BIG_WRITES)
1161 if (flags & FUSE_DONT_MASK)
1163 if (flags & FUSE_AUTO_INVAL_DATA)
1164 fc->auto_inval_data = 1;
1165 else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1166 fc->explicit_inval_data = 1;
1167 if (flags & FUSE_DO_READDIRPLUS) {
1168 fc->do_readdirplus = 1;
1169 if (flags & FUSE_READDIRPLUS_AUTO)
1170 fc->readdirplus_auto = 1;
1172 if (flags & FUSE_ASYNC_DIO)
1174 if (flags & FUSE_WRITEBACK_CACHE)
1175 fc->writeback_cache = 1;
1176 if (flags & FUSE_PARALLEL_DIROPS)
1177 fc->parallel_dirops = 1;
1178 if (flags & FUSE_HANDLE_KILLPRIV)
1179 fc->handle_killpriv = 1;
1180 if (arg->time_gran && arg->time_gran <= 1000000000)
1181 fm->sb->s_time_gran = arg->time_gran;
1182 if ((flags & FUSE_POSIX_ACL)) {
1183 fc->default_permissions = 1;
1186 if (flags & FUSE_CACHE_SYMLINKS)
1187 fc->cache_symlinks = 1;
1188 if (flags & FUSE_ABORT_ERROR)
1190 if (flags & FUSE_MAX_PAGES) {
1192 min_t(unsigned int, fc->max_pages_limit,
1193 max_t(unsigned int, arg->max_pages, 1));
1195 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1196 if (flags & FUSE_MAP_ALIGNMENT &&
1197 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1200 if (flags & FUSE_HAS_INODE_DAX)
1203 if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1204 fc->handle_killpriv_v2 = 1;
1205 fm->sb->s_flags |= SB_NOSEC;
1207 if (flags & FUSE_SETXATTR_EXT)
1208 fc->setxattr_ext = 1;
1209 if (flags & FUSE_SECURITY_CTX)
1210 fc->init_security = 1;
1211 if (flags & FUSE_CREATE_SUPP_GROUP)
1212 fc->create_supp_group = 1;
1214 ra_pages = fc->max_read / PAGE_SIZE;
1219 fm->sb->s_bdi->ra_pages =
1220 min(fm->sb->s_bdi->ra_pages, ra_pages);
1221 fc->minor = arg->minor;
1222 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1223 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1233 fuse_set_initialized(fc);
1234 wake_up_all(&fc->blocked_waitq);
1237 void fuse_send_init(struct fuse_mount *fm)
1239 struct fuse_init_args *ia;
1242 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1244 ia->in.major = FUSE_KERNEL_VERSION;
1245 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1246 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1248 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1249 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1250 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1251 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1252 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1253 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1254 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1255 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1256 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1257 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1258 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1259 FUSE_HAS_EXPIRE_ONLY;
1260 #ifdef CONFIG_FUSE_DAX
1262 flags |= FUSE_MAP_ALIGNMENT;
1263 if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1264 flags |= FUSE_HAS_INODE_DAX;
1266 if (fm->fc->auto_submounts)
1267 flags |= FUSE_SUBMOUNTS;
1269 ia->in.flags = flags;
1270 ia->in.flags2 = flags >> 32;
1272 ia->args.opcode = FUSE_INIT;
1273 ia->args.in_numargs = 1;
1274 ia->args.in_args[0].size = sizeof(ia->in);
1275 ia->args.in_args[0].value = &ia->in;
1276 ia->args.out_numargs = 1;
1277 /* Variable length argument used for backward compatibility
1278 with interface version < 7.5. Rest of init_out is zeroed
1279 by do_get_request(), so a short reply is not a problem */
1280 ia->args.out_argvar = true;
1281 ia->args.out_args[0].size = sizeof(ia->out);
1282 ia->args.out_args[0].value = &ia->out;
1283 ia->args.force = true;
1284 ia->args.nocreds = true;
1285 ia->args.end = process_init_reply;
1287 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1288 process_init_reply(fm, &ia->args, -ENOTCONN);
1290 EXPORT_SYMBOL_GPL(fuse_send_init);
1292 void fuse_free_conn(struct fuse_conn *fc)
1294 WARN_ON(!list_empty(&fc->devices));
1297 EXPORT_SYMBOL_GPL(fuse_free_conn);
1299 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1305 suffix = "-fuseblk";
1307 * sb->s_bdi points to blkdev's bdi however we want to redirect
1308 * it to our private bdi...
1311 sb->s_bdi = &noop_backing_dev_info;
1313 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1314 MINOR(fc->dev), suffix);
1318 /* fuse does it's own writeback accounting */
1319 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1320 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1323 * For a single fuse filesystem use max 1% of dirty +
1324 * writeback threshold.
1326 * This gives about 1M of write buffer for memory maps on a
1327 * machine with 1G and 10% dirty_ratio, which should be more
1330 * Privileged users can raise it by writing to
1332 * /sys/class/bdi/<bdi>/max_ratio
1334 bdi_set_max_ratio(sb->s_bdi, 1);
1339 struct fuse_dev *fuse_dev_alloc(void)
1341 struct fuse_dev *fud;
1342 struct list_head *pq;
1344 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1348 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1354 fud->pq.processing = pq;
1355 fuse_pqueue_init(&fud->pq);
1359 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1361 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1363 fud->fc = fuse_conn_get(fc);
1364 spin_lock(&fc->lock);
1365 list_add_tail(&fud->entry, &fc->devices);
1366 spin_unlock(&fc->lock);
1368 EXPORT_SYMBOL_GPL(fuse_dev_install);
1370 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1372 struct fuse_dev *fud;
1374 fud = fuse_dev_alloc();
1378 fuse_dev_install(fud, fc);
1381 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1383 void fuse_dev_free(struct fuse_dev *fud)
1385 struct fuse_conn *fc = fud->fc;
1388 spin_lock(&fc->lock);
1389 list_del(&fud->entry);
1390 spin_unlock(&fc->lock);
1394 kfree(fud->pq.processing);
1397 EXPORT_SYMBOL_GPL(fuse_dev_free);
1399 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1400 const struct fuse_inode *fi)
1402 struct timespec64 ctime = inode_get_ctime(&fi->inode);
1404 *attr = (struct fuse_attr){
1405 .ino = fi->inode.i_ino,
1406 .size = fi->inode.i_size,
1407 .blocks = fi->inode.i_blocks,
1408 .atime = fi->inode.i_atime.tv_sec,
1409 .mtime = fi->inode.i_mtime.tv_sec,
1410 .ctime = ctime.tv_sec,
1411 .atimensec = fi->inode.i_atime.tv_nsec,
1412 .mtimensec = fi->inode.i_mtime.tv_nsec,
1413 .ctimensec = ctime.tv_nsec,
1414 .mode = fi->inode.i_mode,
1415 .nlink = fi->inode.i_nlink,
1416 .uid = fi->inode.i_uid.val,
1417 .gid = fi->inode.i_gid.val,
1418 .rdev = fi->inode.i_rdev,
1419 .blksize = 1u << fi->inode.i_blkbits,
1423 static void fuse_sb_defaults(struct super_block *sb)
1425 sb->s_magic = FUSE_SUPER_MAGIC;
1426 sb->s_op = &fuse_super_operations;
1427 sb->s_xattr = fuse_xattr_handlers;
1428 sb->s_maxbytes = MAX_LFS_FILESIZE;
1429 sb->s_time_gran = 1;
1430 sb->s_export_op = &fuse_export_operations;
1431 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1432 if (sb->s_user_ns != &init_user_ns)
1433 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1434 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1437 static int fuse_fill_super_submount(struct super_block *sb,
1438 struct fuse_inode *parent_fi)
1440 struct fuse_mount *fm = get_fuse_mount_super(sb);
1441 struct super_block *parent_sb = parent_fi->inode.i_sb;
1442 struct fuse_attr root_attr;
1445 fuse_sb_defaults(sb);
1448 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1449 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1451 sb->s_xattr = parent_sb->s_xattr;
1452 sb->s_time_gran = parent_sb->s_time_gran;
1453 sb->s_blocksize = parent_sb->s_blocksize;
1454 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1455 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1456 if (parent_sb->s_subtype && !sb->s_subtype)
1459 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1460 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1462 * This inode is just a duplicate, so it is not looked up and
1463 * its nlookup should not be incremented. fuse_iget() does
1464 * that, though, so undo it here.
1466 get_fuse_inode(root)->nlookup--;
1467 sb->s_d_op = &fuse_dentry_operations;
1468 sb->s_root = d_make_root(root);
1475 /* Filesystem context private data holds the FUSE inode of the mount point */
1476 static int fuse_get_tree_submount(struct fs_context *fsc)
1478 struct fuse_mount *fm;
1479 struct fuse_inode *mp_fi = fsc->fs_private;
1480 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1481 struct super_block *sb;
1484 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1488 fm->fc = fuse_conn_get(fc);
1489 fsc->s_fs_info = fm;
1490 sb = sget_fc(fsc, NULL, set_anon_super_fc);
1492 fuse_mount_destroy(fm);
1496 /* Initialize superblock, making @mp_fi its root */
1497 err = fuse_fill_super_submount(sb, mp_fi);
1499 deactivate_locked_super(sb);
1503 down_write(&fc->killsb);
1504 list_add_tail(&fm->fc_entry, &fc->mounts);
1505 up_write(&fc->killsb);
1507 sb->s_flags |= SB_ACTIVE;
1508 fsc->root = dget(sb->s_root);
1513 static const struct fs_context_operations fuse_context_submount_ops = {
1514 .get_tree = fuse_get_tree_submount,
1517 int fuse_init_fs_context_submount(struct fs_context *fsc)
1519 fsc->ops = &fuse_context_submount_ops;
1522 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1524 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1526 struct fuse_dev *fud = NULL;
1527 struct fuse_mount *fm = get_fuse_mount_super(sb);
1528 struct fuse_conn *fc = fm->fc;
1530 struct dentry *root_dentry;
1534 if (sb->s_flags & SB_MANDLOCK)
1537 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1538 fuse_sb_defaults(sb);
1543 if (!sb_set_blocksize(sb, ctx->blksize))
1547 sb->s_blocksize = PAGE_SIZE;
1548 sb->s_blocksize_bits = PAGE_SHIFT;
1551 sb->s_subtype = ctx->subtype;
1552 ctx->subtype = NULL;
1553 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1554 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1561 fud = fuse_dev_alloc_install(fc);
1566 fc->dev = sb->s_dev;
1568 err = fuse_bdi_init(fc, sb);
1572 /* Handle umasking inside the fuse code */
1573 if (sb->s_flags & SB_POSIXACL)
1575 sb->s_flags |= SB_POSIXACL;
1577 fc->default_permissions = ctx->default_permissions;
1578 fc->allow_other = ctx->allow_other;
1579 fc->user_id = ctx->user_id;
1580 fc->group_id = ctx->group_id;
1581 fc->legacy_opts_show = ctx->legacy_opts_show;
1582 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1583 fc->destroy = ctx->destroy;
1584 fc->no_control = ctx->no_control;
1585 fc->no_force_umount = ctx->no_force_umount;
1588 root = fuse_get_root_inode(sb, ctx->rootmode);
1589 sb->s_d_op = &fuse_root_dentry_operations;
1590 root_dentry = d_make_root(root);
1593 /* Root dentry doesn't have .d_revalidate */
1594 sb->s_d_op = &fuse_dentry_operations;
1596 mutex_lock(&fuse_mutex);
1598 if (ctx->fudptr && *ctx->fudptr)
1601 err = fuse_ctl_add_conn(fc);
1605 list_add_tail(&fc->entry, &fuse_conn_list);
1606 sb->s_root = root_dentry;
1609 mutex_unlock(&fuse_mutex);
1613 mutex_unlock(&fuse_mutex);
1619 if (IS_ENABLED(CONFIG_FUSE_DAX))
1620 fuse_dax_conn_free(fc);
1624 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1626 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1628 struct fuse_fs_context *ctx = fsc->fs_private;
1631 if (!ctx->file || !ctx->rootmode_present ||
1632 !ctx->user_id_present || !ctx->group_id_present)
1636 * Require mount to happen from the same user namespace which
1637 * opened /dev/fuse to prevent potential attacks.
1639 if ((ctx->file->f_op != &fuse_dev_operations) ||
1640 (ctx->file->f_cred->user_ns != sb->s_user_ns))
1642 ctx->fudptr = &ctx->file->private_data;
1644 err = fuse_fill_super_common(sb, ctx);
1647 /* file->private_data shall be visible on all CPUs after this */
1649 fuse_send_init(get_fuse_mount_super(sb));
1654 * This is the path where user supplied an already initialized fuse dev. In
1655 * this case never create a new super if the old one is gone.
1657 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1662 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1665 return fsc->sget_key == get_fuse_conn_super(sb);
1668 static int fuse_get_tree(struct fs_context *fsc)
1670 struct fuse_fs_context *ctx = fsc->fs_private;
1671 struct fuse_dev *fud;
1672 struct fuse_conn *fc;
1673 struct fuse_mount *fm;
1674 struct super_block *sb;
1677 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1681 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1687 fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1688 fc->release = fuse_free_conn;
1690 fsc->s_fs_info = fm;
1692 if (ctx->fd_present)
1693 ctx->file = fget(ctx->fd);
1695 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1696 err = get_tree_bdev(fsc, fuse_fill_super);
1700 * While block dev mount can be initialized with a dummy device fd
1701 * (found by device name), normal fuse mounts can't
1708 * Allow creating a fuse mount with an already initialized fuse
1711 fud = READ_ONCE(ctx->file->private_data);
1712 if (ctx->file->f_op == &fuse_dev_operations && fud) {
1713 fsc->sget_key = fud->fc;
1714 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1715 err = PTR_ERR_OR_ZERO(sb);
1717 fsc->root = dget(sb->s_root);
1719 err = get_tree_nodev(fsc, fuse_fill_super);
1723 fuse_mount_destroy(fm);
1729 static const struct fs_context_operations fuse_context_ops = {
1730 .free = fuse_free_fsc,
1731 .parse_param = fuse_parse_param,
1732 .reconfigure = fuse_reconfigure,
1733 .get_tree = fuse_get_tree,
1737 * Set up the filesystem mount context.
1739 static int fuse_init_fs_context(struct fs_context *fsc)
1741 struct fuse_fs_context *ctx;
1743 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1748 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1749 ctx->legacy_opts_show = true;
1752 if (fsc->fs_type == &fuseblk_fs_type) {
1753 ctx->is_bdev = true;
1754 ctx->destroy = true;
1758 fsc->fs_private = ctx;
1759 fsc->ops = &fuse_context_ops;
1763 bool fuse_mount_remove(struct fuse_mount *fm)
1765 struct fuse_conn *fc = fm->fc;
1768 down_write(&fc->killsb);
1769 list_del_init(&fm->fc_entry);
1770 if (list_empty(&fc->mounts))
1772 up_write(&fc->killsb);
1776 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1778 void fuse_conn_destroy(struct fuse_mount *fm)
1780 struct fuse_conn *fc = fm->fc;
1783 fuse_send_destroy(fm);
1785 fuse_abort_conn(fc);
1786 fuse_wait_aborted(fc);
1788 if (!list_empty(&fc->entry)) {
1789 mutex_lock(&fuse_mutex);
1790 list_del(&fc->entry);
1791 fuse_ctl_remove_conn(fc);
1792 mutex_unlock(&fuse_mutex);
1795 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1797 static void fuse_sb_destroy(struct super_block *sb)
1799 struct fuse_mount *fm = get_fuse_mount_super(sb);
1803 last = fuse_mount_remove(fm);
1805 fuse_conn_destroy(fm);
1809 void fuse_mount_destroy(struct fuse_mount *fm)
1811 fuse_conn_put(fm->fc);
1814 EXPORT_SYMBOL(fuse_mount_destroy);
1816 static void fuse_kill_sb_anon(struct super_block *sb)
1818 fuse_sb_destroy(sb);
1819 kill_anon_super(sb);
1820 fuse_mount_destroy(get_fuse_mount_super(sb));
1823 static struct file_system_type fuse_fs_type = {
1824 .owner = THIS_MODULE,
1826 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1827 .init_fs_context = fuse_init_fs_context,
1828 .parameters = fuse_fs_parameters,
1829 .kill_sb = fuse_kill_sb_anon,
1831 MODULE_ALIAS_FS("fuse");
1834 static void fuse_kill_sb_blk(struct super_block *sb)
1836 fuse_sb_destroy(sb);
1837 kill_block_super(sb);
1838 fuse_mount_destroy(get_fuse_mount_super(sb));
1841 static struct file_system_type fuseblk_fs_type = {
1842 .owner = THIS_MODULE,
1844 .init_fs_context = fuse_init_fs_context,
1845 .parameters = fuse_fs_parameters,
1846 .kill_sb = fuse_kill_sb_blk,
1847 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1849 MODULE_ALIAS_FS("fuseblk");
1851 static inline int register_fuseblk(void)
1853 return register_filesystem(&fuseblk_fs_type);
1856 static inline void unregister_fuseblk(void)
1858 unregister_filesystem(&fuseblk_fs_type);
1861 static inline int register_fuseblk(void)
1866 static inline void unregister_fuseblk(void)
1871 static void fuse_inode_init_once(void *foo)
1873 struct inode *inode = foo;
1875 inode_init_once(inode);
1878 static int __init fuse_fs_init(void)
1882 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1883 sizeof(struct fuse_inode), 0,
1884 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1885 fuse_inode_init_once);
1887 if (!fuse_inode_cachep)
1890 err = register_fuseblk();
1894 err = register_filesystem(&fuse_fs_type);
1901 unregister_fuseblk();
1903 kmem_cache_destroy(fuse_inode_cachep);
1908 static void fuse_fs_cleanup(void)
1910 unregister_filesystem(&fuse_fs_type);
1911 unregister_fuseblk();
1914 * Make sure all delayed rcu free inodes are flushed before we
1918 kmem_cache_destroy(fuse_inode_cachep);
1921 static struct kobject *fuse_kobj;
1923 static int fuse_sysfs_init(void)
1927 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1933 err = sysfs_create_mount_point(fuse_kobj, "connections");
1935 goto out_fuse_unregister;
1939 out_fuse_unregister:
1940 kobject_put(fuse_kobj);
1945 static void fuse_sysfs_cleanup(void)
1947 sysfs_remove_mount_point(fuse_kobj, "connections");
1948 kobject_put(fuse_kobj);
1951 static int __init fuse_init(void)
1955 pr_info("init (API version %i.%i)\n",
1956 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1958 INIT_LIST_HEAD(&fuse_conn_list);
1959 res = fuse_fs_init();
1963 res = fuse_dev_init();
1965 goto err_fs_cleanup;
1967 res = fuse_sysfs_init();
1969 goto err_dev_cleanup;
1971 res = fuse_ctl_init();
1973 goto err_sysfs_cleanup;
1975 sanitize_global_limit(&max_user_bgreq);
1976 sanitize_global_limit(&max_user_congthresh);
1981 fuse_sysfs_cleanup();
1990 static void __exit fuse_exit(void)
1995 fuse_sysfs_cleanup();
2000 module_init(fuse_init);
2001 module_exit(fuse_exit);