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) && !is_bad_inode(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 *fc)
143 struct super_block *sb = fc->root->d_sb;
146 if (fc->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 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
210 u64 attr_valid, u64 attr_version)
212 struct fuse_conn *fc = get_fuse_conn(inode);
213 struct fuse_inode *fi = get_fuse_inode(inode);
214 bool is_wb = fc->writeback_cache;
216 struct timespec64 old_mtime;
218 spin_lock(&fi->lock);
219 if ((attr_version != 0 && fi->attr_version > attr_version) ||
220 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
221 spin_unlock(&fi->lock);
225 old_mtime = inode->i_mtime;
226 fuse_change_attributes_common(inode, attr, attr_valid);
228 oldsize = inode->i_size;
230 * In case of writeback_cache enabled, the cached writes beyond EOF
231 * extend local i_size without keeping userspace server in sync. So,
232 * attr->size coming from server can be stale. We cannot trust it.
234 if (!is_wb || !S_ISREG(inode->i_mode))
235 i_size_write(inode, attr->size);
236 spin_unlock(&fi->lock);
238 if (!is_wb && S_ISREG(inode->i_mode)) {
241 if (oldsize != attr->size) {
242 truncate_pagecache(inode, attr->size);
243 if (!fc->explicit_inval_data)
245 } else if (fc->auto_inval_data) {
246 struct timespec64 new_mtime = {
247 .tv_sec = attr->mtime,
248 .tv_nsec = attr->mtimensec,
252 * Auto inval mode also checks and invalidates if mtime
255 if (!timespec64_equal(&old_mtime, &new_mtime))
260 invalidate_inode_pages2(inode->i_mapping);
264 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
266 inode->i_mode = attr->mode & S_IFMT;
267 inode->i_size = attr->size;
268 inode->i_mtime.tv_sec = attr->mtime;
269 inode->i_mtime.tv_nsec = attr->mtimensec;
270 inode->i_ctime.tv_sec = attr->ctime;
271 inode->i_ctime.tv_nsec = attr->ctimensec;
272 if (S_ISREG(inode->i_mode)) {
273 fuse_init_common(inode);
274 fuse_init_file_inode(inode);
275 } else if (S_ISDIR(inode->i_mode))
276 fuse_init_dir(inode);
277 else if (S_ISLNK(inode->i_mode))
278 fuse_init_symlink(inode);
279 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
280 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
281 fuse_init_common(inode);
282 init_special_inode(inode, inode->i_mode,
283 new_decode_dev(attr->rdev));
288 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
290 u64 nodeid = *(u64 *) _nodeidp;
291 if (get_node_id(inode) == nodeid)
297 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
299 u64 nodeid = *(u64 *) _nodeidp;
300 get_fuse_inode(inode)->nodeid = nodeid;
304 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
305 int generation, struct fuse_attr *attr,
306 u64 attr_valid, u64 attr_version)
309 struct fuse_inode *fi;
310 struct fuse_conn *fc = get_fuse_conn_super(sb);
313 * Auto mount points get their node id from the submount root, which is
314 * not a unique identifier within this filesystem.
316 * To avoid conflicts, do not place submount points into the inode hash
319 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
320 S_ISDIR(attr->mode)) {
321 inode = new_inode(sb);
325 fuse_init_inode(inode, attr);
326 get_fuse_inode(inode)->nodeid = nodeid;
327 inode->i_flags |= S_AUTOMOUNT;
332 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
336 if ((inode->i_state & I_NEW)) {
337 inode->i_flags |= S_NOATIME;
338 if (!fc->writeback_cache || !S_ISREG(attr->mode))
339 inode->i_flags |= S_NOCMTIME;
340 inode->i_generation = generation;
341 fuse_init_inode(inode, attr);
342 unlock_new_inode(inode);
343 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
344 /* Inode has changed type, any I/O on the old should fail */
345 make_bad_inode(inode);
350 fi = get_fuse_inode(inode);
351 spin_lock(&fi->lock);
353 spin_unlock(&fi->lock);
354 fuse_change_attributes(inode, attr, attr_valid, attr_version);
359 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
360 struct fuse_mount **fm)
362 struct fuse_mount *fm_iter;
365 WARN_ON(!rwsem_is_locked(&fc->killsb));
366 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
370 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
381 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
382 loff_t offset, loff_t len)
384 struct fuse_inode *fi;
389 inode = fuse_ilookup(fc, nodeid, NULL);
393 fi = get_fuse_inode(inode);
394 spin_lock(&fi->lock);
395 fi->attr_version = atomic64_inc_return(&fc->attr_version);
396 spin_unlock(&fi->lock);
398 fuse_invalidate_attr(inode);
399 forget_all_cached_acls(inode);
401 pg_start = offset >> PAGE_SHIFT;
405 pg_end = (offset + len - 1) >> PAGE_SHIFT;
406 invalidate_inode_pages2_range(inode->i_mapping,
413 bool fuse_lock_inode(struct inode *inode)
417 if (!get_fuse_conn(inode)->parallel_dirops) {
418 mutex_lock(&get_fuse_inode(inode)->mutex);
425 void fuse_unlock_inode(struct inode *inode, bool locked)
428 mutex_unlock(&get_fuse_inode(inode)->mutex);
431 static void fuse_umount_begin(struct super_block *sb)
433 struct fuse_conn *fc = get_fuse_conn_super(sb);
435 if (!fc->no_force_umount)
439 static void fuse_send_destroy(struct fuse_mount *fm)
441 if (fm->fc->conn_init) {
444 args.opcode = FUSE_DESTROY;
447 fuse_simple_request(fm, &args);
451 static void fuse_put_super(struct super_block *sb)
453 struct fuse_mount *fm = get_fuse_mount_super(sb);
455 fuse_conn_put(fm->fc);
459 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
461 stbuf->f_type = FUSE_SUPER_MAGIC;
462 stbuf->f_bsize = attr->bsize;
463 stbuf->f_frsize = attr->frsize;
464 stbuf->f_blocks = attr->blocks;
465 stbuf->f_bfree = attr->bfree;
466 stbuf->f_bavail = attr->bavail;
467 stbuf->f_files = attr->files;
468 stbuf->f_ffree = attr->ffree;
469 stbuf->f_namelen = attr->namelen;
470 /* fsid is left zero */
473 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
475 struct super_block *sb = dentry->d_sb;
476 struct fuse_mount *fm = get_fuse_mount_super(sb);
478 struct fuse_statfs_out outarg;
481 if (!fuse_allow_current_process(fm->fc)) {
482 buf->f_type = FUSE_SUPER_MAGIC;
486 memset(&outarg, 0, sizeof(outarg));
488 args.opcode = FUSE_STATFS;
489 args.nodeid = get_node_id(d_inode(dentry));
490 args.out_numargs = 1;
491 args.out_args[0].size = sizeof(outarg);
492 args.out_args[0].value = &outarg;
493 err = fuse_simple_request(fm, &args);
495 convert_fuse_statfs(buf, &outarg.st);
506 OPT_DEFAULT_PERMISSIONS,
513 static const struct fs_parameter_spec fuse_fs_parameters[] = {
514 fsparam_string ("source", OPT_SOURCE),
515 fsparam_u32 ("fd", OPT_FD),
516 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
517 fsparam_u32 ("user_id", OPT_USER_ID),
518 fsparam_u32 ("group_id", OPT_GROUP_ID),
519 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
520 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
521 fsparam_u32 ("max_read", OPT_MAX_READ),
522 fsparam_u32 ("blksize", OPT_BLKSIZE),
523 fsparam_string ("subtype", OPT_SUBTYPE),
527 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
529 struct fs_parse_result result;
530 struct fuse_fs_context *ctx = fc->fs_private;
533 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
535 * Ignore options coming from mount(MS_REMOUNT) for backward
541 return invalfc(fc, "No changes allowed in reconfigure");
544 opt = fs_parse(fc, fuse_fs_parameters, param, &result);
551 return invalfc(fc, "Multiple sources specified");
552 fc->source = param->string;
553 param->string = NULL;
558 return invalfc(fc, "Multiple subtypes specified");
559 ctx->subtype = param->string;
560 param->string = NULL;
564 ctx->fd = result.uint_32;
565 ctx->fd_present = true;
569 if (!fuse_valid_type(result.uint_32))
570 return invalfc(fc, "Invalid rootmode");
571 ctx->rootmode = result.uint_32;
572 ctx->rootmode_present = true;
576 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
577 if (!uid_valid(ctx->user_id))
578 return invalfc(fc, "Invalid user_id");
579 ctx->user_id_present = true;
583 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
584 if (!gid_valid(ctx->group_id))
585 return invalfc(fc, "Invalid group_id");
586 ctx->group_id_present = true;
589 case OPT_DEFAULT_PERMISSIONS:
590 ctx->default_permissions = true;
593 case OPT_ALLOW_OTHER:
594 ctx->allow_other = true;
598 ctx->max_read = result.uint_32;
603 return invalfc(fc, "blksize only supported for fuseblk");
604 ctx->blksize = result.uint_32;
614 static void fuse_free_fc(struct fs_context *fc)
616 struct fuse_fs_context *ctx = fc->fs_private;
624 static int fuse_show_options(struct seq_file *m, struct dentry *root)
626 struct super_block *sb = root->d_sb;
627 struct fuse_conn *fc = get_fuse_conn_super(sb);
629 if (fc->legacy_opts_show) {
630 seq_printf(m, ",user_id=%u",
631 from_kuid_munged(fc->user_ns, fc->user_id));
632 seq_printf(m, ",group_id=%u",
633 from_kgid_munged(fc->user_ns, fc->group_id));
634 if (fc->default_permissions)
635 seq_puts(m, ",default_permissions");
637 seq_puts(m, ",allow_other");
638 if (fc->max_read != ~0)
639 seq_printf(m, ",max_read=%u", fc->max_read);
640 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
641 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
643 #ifdef CONFIG_FUSE_DAX
651 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
652 const struct fuse_iqueue_ops *ops,
655 memset(fiq, 0, sizeof(struct fuse_iqueue));
656 spin_lock_init(&fiq->lock);
657 init_waitqueue_head(&fiq->waitq);
658 INIT_LIST_HEAD(&fiq->pending);
659 INIT_LIST_HEAD(&fiq->interrupts);
660 fiq->forget_list_tail = &fiq->forget_list_head;
666 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
670 spin_lock_init(&fpq->lock);
671 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
672 INIT_LIST_HEAD(&fpq->processing[i]);
673 INIT_LIST_HEAD(&fpq->io);
677 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
678 struct user_namespace *user_ns,
679 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
681 memset(fc, 0, sizeof(*fc));
682 spin_lock_init(&fc->lock);
683 spin_lock_init(&fc->bg_lock);
684 init_rwsem(&fc->killsb);
685 refcount_set(&fc->count, 1);
686 atomic_set(&fc->dev_count, 1);
687 init_waitqueue_head(&fc->blocked_waitq);
688 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
689 INIT_LIST_HEAD(&fc->bg_queue);
690 INIT_LIST_HEAD(&fc->entry);
691 INIT_LIST_HEAD(&fc->devices);
692 atomic_set(&fc->num_waiting, 0);
693 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
694 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
695 atomic64_set(&fc->khctr, 0);
696 fc->polled_files = RB_ROOT;
700 atomic64_set(&fc->attr_version, 1);
701 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
702 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
703 fc->user_ns = get_user_ns(user_ns);
704 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
706 INIT_LIST_HEAD(&fc->mounts);
707 list_add(&fm->fc_entry, &fc->mounts);
710 EXPORT_SYMBOL_GPL(fuse_conn_init);
712 void fuse_conn_put(struct fuse_conn *fc)
714 if (refcount_dec_and_test(&fc->count)) {
715 struct fuse_iqueue *fiq = &fc->iq;
717 if (IS_ENABLED(CONFIG_FUSE_DAX))
718 fuse_dax_conn_free(fc);
719 if (fiq->ops->release)
720 fiq->ops->release(fiq);
721 put_pid_ns(fc->pid_ns);
722 put_user_ns(fc->user_ns);
726 EXPORT_SYMBOL_GPL(fuse_conn_put);
728 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
730 refcount_inc(&fc->count);
733 EXPORT_SYMBOL_GPL(fuse_conn_get);
735 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
737 struct fuse_attr attr;
738 memset(&attr, 0, sizeof(attr));
741 attr.ino = FUSE_ROOT_ID;
743 return fuse_iget(sb, 1, 0, &attr, 0, 0);
746 struct fuse_inode_handle {
751 static struct dentry *fuse_get_dentry(struct super_block *sb,
752 struct fuse_inode_handle *handle)
754 struct fuse_conn *fc = get_fuse_conn_super(sb);
756 struct dentry *entry;
759 if (handle->nodeid == 0)
762 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
764 struct fuse_entry_out outarg;
765 const struct qstr name = QSTR_INIT(".", 1);
767 if (!fc->export_support)
770 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
772 if (err && err != -ENOENT)
779 if (get_node_id(inode) != handle->nodeid)
783 if (inode->i_generation != handle->generation)
786 entry = d_obtain_alias(inode);
787 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
788 fuse_invalidate_entry_cache(entry);
798 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
799 struct inode *parent)
801 int len = parent ? 6 : 3;
805 if (*max_len < len) {
807 return FILEID_INVALID;
810 nodeid = get_fuse_inode(inode)->nodeid;
811 generation = inode->i_generation;
813 fh[0] = (u32)(nodeid >> 32);
814 fh[1] = (u32)(nodeid & 0xffffffff);
818 nodeid = get_fuse_inode(parent)->nodeid;
819 generation = parent->i_generation;
821 fh[3] = (u32)(nodeid >> 32);
822 fh[4] = (u32)(nodeid & 0xffffffff);
827 return parent ? 0x82 : 0x81;
830 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
831 struct fid *fid, int fh_len, int fh_type)
833 struct fuse_inode_handle handle;
835 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
838 handle.nodeid = (u64) fid->raw[0] << 32;
839 handle.nodeid |= (u64) fid->raw[1];
840 handle.generation = fid->raw[2];
841 return fuse_get_dentry(sb, &handle);
844 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
845 struct fid *fid, int fh_len, int fh_type)
847 struct fuse_inode_handle parent;
849 if (fh_type != 0x82 || fh_len < 6)
852 parent.nodeid = (u64) fid->raw[3] << 32;
853 parent.nodeid |= (u64) fid->raw[4];
854 parent.generation = fid->raw[5];
855 return fuse_get_dentry(sb, &parent);
858 static struct dentry *fuse_get_parent(struct dentry *child)
860 struct inode *child_inode = d_inode(child);
861 struct fuse_conn *fc = get_fuse_conn(child_inode);
863 struct dentry *parent;
864 struct fuse_entry_out outarg;
865 const struct qstr name = QSTR_INIT("..", 2);
868 if (!fc->export_support)
869 return ERR_PTR(-ESTALE);
871 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
872 &name, &outarg, &inode);
875 return ERR_PTR(-ESTALE);
879 parent = d_obtain_alias(inode);
880 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
881 fuse_invalidate_entry_cache(parent);
886 static const struct export_operations fuse_export_operations = {
887 .fh_to_dentry = fuse_fh_to_dentry,
888 .fh_to_parent = fuse_fh_to_parent,
889 .encode_fh = fuse_encode_fh,
890 .get_parent = fuse_get_parent,
893 static const struct super_operations fuse_super_operations = {
894 .alloc_inode = fuse_alloc_inode,
895 .free_inode = fuse_free_inode,
896 .evict_inode = fuse_evict_inode,
897 .write_inode = fuse_write_inode,
898 .drop_inode = generic_delete_inode,
899 .put_super = fuse_put_super,
900 .umount_begin = fuse_umount_begin,
901 .statfs = fuse_statfs,
902 .show_options = fuse_show_options,
905 static void sanitize_global_limit(unsigned *limit)
908 * The default maximum number of async requests is calculated to consume
909 * 1/2^13 of the total memory, assuming 392 bytes per request.
912 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
914 if (*limit >= 1 << 16)
915 *limit = (1 << 16) - 1;
918 static int set_global_limit(const char *val, const struct kernel_param *kp)
922 rv = param_set_uint(val, kp);
926 sanitize_global_limit((unsigned *)kp->arg);
931 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
933 int cap_sys_admin = capable(CAP_SYS_ADMIN);
938 sanitize_global_limit(&max_user_bgreq);
939 sanitize_global_limit(&max_user_congthresh);
941 spin_lock(&fc->bg_lock);
942 if (arg->max_background) {
943 fc->max_background = arg->max_background;
945 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
946 fc->max_background = max_user_bgreq;
948 if (arg->congestion_threshold) {
949 fc->congestion_threshold = arg->congestion_threshold;
951 if (!cap_sys_admin &&
952 fc->congestion_threshold > max_user_congthresh)
953 fc->congestion_threshold = max_user_congthresh;
955 spin_unlock(&fc->bg_lock);
958 struct fuse_init_args {
959 struct fuse_args args;
960 struct fuse_init_in in;
961 struct fuse_init_out out;
964 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
967 struct fuse_conn *fc = fm->fc;
968 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
969 struct fuse_init_out *arg = &ia->out;
972 if (error || arg->major != FUSE_KERNEL_VERSION)
975 unsigned long ra_pages;
977 process_init_limits(fc, arg);
979 if (arg->minor >= 6) {
980 ra_pages = arg->max_readahead / PAGE_SIZE;
981 if (arg->flags & FUSE_ASYNC_READ)
983 if (!(arg->flags & FUSE_POSIX_LOCKS))
985 if (arg->minor >= 17) {
986 if (!(arg->flags & FUSE_FLOCK_LOCKS))
989 if (!(arg->flags & FUSE_POSIX_LOCKS))
992 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
993 fc->atomic_o_trunc = 1;
994 if (arg->minor >= 9) {
995 /* LOOKUP has dependency on proto version */
996 if (arg->flags & FUSE_EXPORT_SUPPORT)
997 fc->export_support = 1;
999 if (arg->flags & FUSE_BIG_WRITES)
1001 if (arg->flags & FUSE_DONT_MASK)
1003 if (arg->flags & FUSE_AUTO_INVAL_DATA)
1004 fc->auto_inval_data = 1;
1005 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1006 fc->explicit_inval_data = 1;
1007 if (arg->flags & FUSE_DO_READDIRPLUS) {
1008 fc->do_readdirplus = 1;
1009 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1010 fc->readdirplus_auto = 1;
1012 if (arg->flags & FUSE_ASYNC_DIO)
1014 if (arg->flags & FUSE_WRITEBACK_CACHE)
1015 fc->writeback_cache = 1;
1016 if (arg->flags & FUSE_PARALLEL_DIROPS)
1017 fc->parallel_dirops = 1;
1018 if (arg->flags & FUSE_HANDLE_KILLPRIV)
1019 fc->handle_killpriv = 1;
1020 if (arg->time_gran && arg->time_gran <= 1000000000)
1021 fm->sb->s_time_gran = arg->time_gran;
1022 if ((arg->flags & FUSE_POSIX_ACL)) {
1023 fc->default_permissions = 1;
1025 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1027 if (arg->flags & FUSE_CACHE_SYMLINKS)
1028 fc->cache_symlinks = 1;
1029 if (arg->flags & FUSE_ABORT_ERROR)
1031 if (arg->flags & FUSE_MAX_PAGES) {
1033 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
1034 max_t(unsigned int, arg->max_pages, 1));
1036 if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1037 arg->flags & FUSE_MAP_ALIGNMENT &&
1038 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1042 ra_pages = fc->max_read / PAGE_SIZE;
1047 fm->sb->s_bdi->ra_pages =
1048 min(fm->sb->s_bdi->ra_pages, ra_pages);
1049 fc->minor = arg->minor;
1050 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1051 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1061 fuse_set_initialized(fc);
1062 wake_up_all(&fc->blocked_waitq);
1065 void fuse_send_init(struct fuse_mount *fm)
1067 struct fuse_init_args *ia;
1069 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1071 ia->in.major = FUSE_KERNEL_VERSION;
1072 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1073 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1075 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1076 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1077 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1078 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1079 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1080 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1081 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1082 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1083 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
1084 #ifdef CONFIG_FUSE_DAX
1086 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1088 if (fm->fc->auto_submounts)
1089 ia->in.flags |= FUSE_SUBMOUNTS;
1091 ia->args.opcode = FUSE_INIT;
1092 ia->args.in_numargs = 1;
1093 ia->args.in_args[0].size = sizeof(ia->in);
1094 ia->args.in_args[0].value = &ia->in;
1095 ia->args.out_numargs = 1;
1096 /* Variable length argument used for backward compatibility
1097 with interface version < 7.5. Rest of init_out is zeroed
1098 by do_get_request(), so a short reply is not a problem */
1099 ia->args.out_argvar = true;
1100 ia->args.out_args[0].size = sizeof(ia->out);
1101 ia->args.out_args[0].value = &ia->out;
1102 ia->args.force = true;
1103 ia->args.nocreds = true;
1104 ia->args.end = process_init_reply;
1106 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1107 process_init_reply(fm, &ia->args, -ENOTCONN);
1109 EXPORT_SYMBOL_GPL(fuse_send_init);
1111 void fuse_free_conn(struct fuse_conn *fc)
1113 WARN_ON(!list_empty(&fc->devices));
1116 EXPORT_SYMBOL_GPL(fuse_free_conn);
1118 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1124 suffix = "-fuseblk";
1126 * sb->s_bdi points to blkdev's bdi however we want to redirect
1127 * it to our private bdi...
1130 sb->s_bdi = &noop_backing_dev_info;
1132 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1133 MINOR(fc->dev), suffix);
1137 /* fuse does it's own writeback accounting */
1138 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1139 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1142 * For a single fuse filesystem use max 1% of dirty +
1143 * writeback threshold.
1145 * This gives about 1M of write buffer for memory maps on a
1146 * machine with 1G and 10% dirty_ratio, which should be more
1149 * Privileged users can raise it by writing to
1151 * /sys/class/bdi/<bdi>/max_ratio
1153 bdi_set_max_ratio(sb->s_bdi, 1);
1158 struct fuse_dev *fuse_dev_alloc(void)
1160 struct fuse_dev *fud;
1161 struct list_head *pq;
1163 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1167 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1173 fud->pq.processing = pq;
1174 fuse_pqueue_init(&fud->pq);
1178 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1180 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1182 fud->fc = fuse_conn_get(fc);
1183 spin_lock(&fc->lock);
1184 list_add_tail(&fud->entry, &fc->devices);
1185 spin_unlock(&fc->lock);
1187 EXPORT_SYMBOL_GPL(fuse_dev_install);
1189 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1191 struct fuse_dev *fud;
1193 fud = fuse_dev_alloc();
1197 fuse_dev_install(fud, fc);
1200 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1202 void fuse_dev_free(struct fuse_dev *fud)
1204 struct fuse_conn *fc = fud->fc;
1207 spin_lock(&fc->lock);
1208 list_del(&fud->entry);
1209 spin_unlock(&fc->lock);
1213 kfree(fud->pq.processing);
1216 EXPORT_SYMBOL_GPL(fuse_dev_free);
1218 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1219 const struct fuse_inode *fi)
1221 *attr = (struct fuse_attr){
1222 .ino = fi->inode.i_ino,
1223 .size = fi->inode.i_size,
1224 .blocks = fi->inode.i_blocks,
1225 .atime = fi->inode.i_atime.tv_sec,
1226 .mtime = fi->inode.i_mtime.tv_sec,
1227 .ctime = fi->inode.i_ctime.tv_sec,
1228 .atimensec = fi->inode.i_atime.tv_nsec,
1229 .mtimensec = fi->inode.i_mtime.tv_nsec,
1230 .ctimensec = fi->inode.i_ctime.tv_nsec,
1231 .mode = fi->inode.i_mode,
1232 .nlink = fi->inode.i_nlink,
1233 .uid = fi->inode.i_uid.val,
1234 .gid = fi->inode.i_gid.val,
1235 .rdev = fi->inode.i_rdev,
1236 .blksize = 1u << fi->inode.i_blkbits,
1240 static void fuse_sb_defaults(struct super_block *sb)
1242 sb->s_magic = FUSE_SUPER_MAGIC;
1243 sb->s_op = &fuse_super_operations;
1244 sb->s_xattr = fuse_xattr_handlers;
1245 sb->s_maxbytes = MAX_LFS_FILESIZE;
1246 sb->s_time_gran = 1;
1247 sb->s_export_op = &fuse_export_operations;
1248 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1249 if (sb->s_user_ns != &init_user_ns)
1250 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1251 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1254 * If we are not in the initial user namespace posix
1255 * acls must be translated.
1257 if (sb->s_user_ns != &init_user_ns)
1258 sb->s_xattr = fuse_no_acl_xattr_handlers;
1261 int fuse_fill_super_submount(struct super_block *sb,
1262 struct fuse_inode *parent_fi)
1264 struct fuse_mount *fm = get_fuse_mount_super(sb);
1265 struct super_block *parent_sb = parent_fi->inode.i_sb;
1266 struct fuse_attr root_attr;
1269 fuse_sb_defaults(sb);
1272 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1273 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1275 sb->s_xattr = parent_sb->s_xattr;
1276 sb->s_time_gran = parent_sb->s_time_gran;
1277 sb->s_blocksize = parent_sb->s_blocksize;
1278 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1279 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1280 if (parent_sb->s_subtype && !sb->s_subtype)
1283 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1284 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1286 * This inode is just a duplicate, so it is not looked up and
1287 * its nlookup should not be incremented. fuse_iget() does
1288 * that, though, so undo it here.
1290 get_fuse_inode(root)->nlookup--;
1291 sb->s_d_op = &fuse_dentry_operations;
1292 sb->s_root = d_make_root(root);
1299 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1301 struct fuse_dev *fud = NULL;
1302 struct fuse_mount *fm = get_fuse_mount_super(sb);
1303 struct fuse_conn *fc = fm->fc;
1305 struct dentry *root_dentry;
1309 if (sb->s_flags & SB_MANDLOCK)
1312 fuse_sb_defaults(sb);
1317 if (!sb_set_blocksize(sb, ctx->blksize))
1321 sb->s_blocksize = PAGE_SIZE;
1322 sb->s_blocksize_bits = PAGE_SHIFT;
1325 sb->s_subtype = ctx->subtype;
1326 ctx->subtype = NULL;
1327 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1328 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1335 fud = fuse_dev_alloc_install(fc);
1340 fc->dev = sb->s_dev;
1342 err = fuse_bdi_init(fc, sb);
1346 /* Handle umasking inside the fuse code */
1347 if (sb->s_flags & SB_POSIXACL)
1349 sb->s_flags |= SB_POSIXACL;
1351 fc->default_permissions = ctx->default_permissions;
1352 fc->allow_other = ctx->allow_other;
1353 fc->user_id = ctx->user_id;
1354 fc->group_id = ctx->group_id;
1355 fc->legacy_opts_show = ctx->legacy_opts_show;
1356 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1357 fc->destroy = ctx->destroy;
1358 fc->no_control = ctx->no_control;
1359 fc->no_force_umount = ctx->no_force_umount;
1362 root = fuse_get_root_inode(sb, ctx->rootmode);
1363 sb->s_d_op = &fuse_root_dentry_operations;
1364 root_dentry = d_make_root(root);
1367 /* Root dentry doesn't have .d_revalidate */
1368 sb->s_d_op = &fuse_dentry_operations;
1370 mutex_lock(&fuse_mutex);
1372 if (ctx->fudptr && *ctx->fudptr)
1375 err = fuse_ctl_add_conn(fc);
1379 list_add_tail(&fc->entry, &fuse_conn_list);
1380 sb->s_root = root_dentry;
1383 mutex_unlock(&fuse_mutex);
1387 mutex_unlock(&fuse_mutex);
1393 if (IS_ENABLED(CONFIG_FUSE_DAX))
1394 fuse_dax_conn_free(fc);
1398 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1400 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1402 struct fuse_fs_context *ctx = fsc->fs_private;
1405 struct fuse_conn *fc;
1406 struct fuse_mount *fm;
1409 file = fget(ctx->fd);
1414 * Require mount to happen from the same user namespace which
1415 * opened /dev/fuse to prevent potential attacks.
1417 if ((file->f_op != &fuse_dev_operations) ||
1418 (file->f_cred->user_ns != sb->s_user_ns))
1420 ctx->fudptr = &file->private_data;
1422 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1427 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1433 fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1434 fc->release = fuse_free_conn;
1438 err = fuse_fill_super_common(sb, ctx);
1442 * atomic_dec_and_test() in fput() provides the necessary
1443 * memory barrier for file->private_data to be visible on all
1447 fuse_send_init(get_fuse_mount_super(sb));
1453 sb->s_fs_info = NULL;
1460 static int fuse_get_tree(struct fs_context *fc)
1462 struct fuse_fs_context *ctx = fc->fs_private;
1464 if (!ctx->fd_present || !ctx->rootmode_present ||
1465 !ctx->user_id_present || !ctx->group_id_present)
1470 return get_tree_bdev(fc, fuse_fill_super);
1473 return get_tree_nodev(fc, fuse_fill_super);
1476 static const struct fs_context_operations fuse_context_ops = {
1477 .free = fuse_free_fc,
1478 .parse_param = fuse_parse_param,
1479 .reconfigure = fuse_reconfigure,
1480 .get_tree = fuse_get_tree,
1484 * Set up the filesystem mount context.
1486 static int fuse_init_fs_context(struct fs_context *fc)
1488 struct fuse_fs_context *ctx;
1490 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1495 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1496 ctx->legacy_opts_show = true;
1499 if (fc->fs_type == &fuseblk_fs_type) {
1500 ctx->is_bdev = true;
1501 ctx->destroy = true;
1505 fc->fs_private = ctx;
1506 fc->ops = &fuse_context_ops;
1510 bool fuse_mount_remove(struct fuse_mount *fm)
1512 struct fuse_conn *fc = fm->fc;
1515 down_write(&fc->killsb);
1516 list_del_init(&fm->fc_entry);
1517 if (list_empty(&fc->mounts))
1519 up_write(&fc->killsb);
1523 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1525 void fuse_conn_destroy(struct fuse_mount *fm)
1527 struct fuse_conn *fc = fm->fc;
1530 fuse_send_destroy(fm);
1532 fuse_abort_conn(fc);
1533 fuse_wait_aborted(fc);
1535 if (!list_empty(&fc->entry)) {
1536 mutex_lock(&fuse_mutex);
1537 list_del(&fc->entry);
1538 fuse_ctl_remove_conn(fc);
1539 mutex_unlock(&fuse_mutex);
1542 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1544 static void fuse_sb_destroy(struct super_block *sb)
1546 struct fuse_mount *fm = get_fuse_mount_super(sb);
1550 last = fuse_mount_remove(fm);
1552 fuse_conn_destroy(fm);
1556 static void fuse_kill_sb_anon(struct super_block *sb)
1558 fuse_sb_destroy(sb);
1559 kill_anon_super(sb);
1562 static struct file_system_type fuse_fs_type = {
1563 .owner = THIS_MODULE,
1565 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1566 .init_fs_context = fuse_init_fs_context,
1567 .parameters = fuse_fs_parameters,
1568 .kill_sb = fuse_kill_sb_anon,
1570 MODULE_ALIAS_FS("fuse");
1573 static void fuse_kill_sb_blk(struct super_block *sb)
1575 fuse_sb_destroy(sb);
1576 kill_block_super(sb);
1579 static struct file_system_type fuseblk_fs_type = {
1580 .owner = THIS_MODULE,
1582 .init_fs_context = fuse_init_fs_context,
1583 .parameters = fuse_fs_parameters,
1584 .kill_sb = fuse_kill_sb_blk,
1585 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1587 MODULE_ALIAS_FS("fuseblk");
1589 static inline int register_fuseblk(void)
1591 return register_filesystem(&fuseblk_fs_type);
1594 static inline void unregister_fuseblk(void)
1596 unregister_filesystem(&fuseblk_fs_type);
1599 static inline int register_fuseblk(void)
1604 static inline void unregister_fuseblk(void)
1609 static void fuse_inode_init_once(void *foo)
1611 struct inode *inode = foo;
1613 inode_init_once(inode);
1616 static int __init fuse_fs_init(void)
1620 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1621 sizeof(struct fuse_inode), 0,
1622 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1623 fuse_inode_init_once);
1625 if (!fuse_inode_cachep)
1628 err = register_fuseblk();
1632 err = register_filesystem(&fuse_fs_type);
1639 unregister_fuseblk();
1641 kmem_cache_destroy(fuse_inode_cachep);
1646 static void fuse_fs_cleanup(void)
1648 unregister_filesystem(&fuse_fs_type);
1649 unregister_fuseblk();
1652 * Make sure all delayed rcu free inodes are flushed before we
1656 kmem_cache_destroy(fuse_inode_cachep);
1659 static struct kobject *fuse_kobj;
1661 static int fuse_sysfs_init(void)
1665 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1671 err = sysfs_create_mount_point(fuse_kobj, "connections");
1673 goto out_fuse_unregister;
1677 out_fuse_unregister:
1678 kobject_put(fuse_kobj);
1683 static void fuse_sysfs_cleanup(void)
1685 sysfs_remove_mount_point(fuse_kobj, "connections");
1686 kobject_put(fuse_kobj);
1689 static int __init fuse_init(void)
1693 pr_info("init (API version %i.%i)\n",
1694 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1696 INIT_LIST_HEAD(&fuse_conn_list);
1697 res = fuse_fs_init();
1701 res = fuse_dev_init();
1703 goto err_fs_cleanup;
1705 res = fuse_sysfs_init();
1707 goto err_dev_cleanup;
1709 res = fuse_ctl_init();
1711 goto err_sysfs_cleanup;
1713 sanitize_global_limit(&max_user_bgreq);
1714 sanitize_global_limit(&max_user_congthresh);
1719 fuse_sysfs_cleanup();
1728 static void __exit fuse_exit(void)
1733 fuse_sysfs_cleanup();
1738 module_init(fuse_init);
1739 module_exit(fuse_exit);