1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/compiler_types.h>
4 #include <linux/errno.h>
6 #include <linux/fsnotify.h>
9 #include <linux/init.h>
10 #include <linux/ipc_namespace.h>
11 #include <linux/kdev_t.h>
12 #include <linux/kernel.h>
13 #include <linux/list.h>
14 #include <linux/namei.h>
15 #include <linux/magic.h>
16 #include <linux/major.h>
17 #include <linux/miscdevice.h>
18 #include <linux/module.h>
19 #include <linux/mutex.h>
20 #include <linux/mount.h>
21 #include <linux/fs_parser.h>
22 #include <linux/radix-tree.h>
23 #include <linux/sched.h>
24 #include <linux/seq_file.h>
25 #include <linux/slab.h>
26 #include <linux/spinlock_types.h>
27 #include <linux/stddef.h>
28 #include <linux/string.h>
29 #include <linux/types.h>
30 #include <linux/uaccess.h>
31 #include <linux/user_namespace.h>
32 #include <linux/xarray.h>
33 #include <uapi/asm-generic/errno-base.h>
34 #include <uapi/linux/android/binder.h>
35 #include <uapi/linux/android/binderfs.h>
37 #include "binder_internal.h"
40 #define SECOND_INODE 2
41 #define INODE_OFFSET 3
42 #define BINDERFS_MAX_MINOR (1U << MINORBITS)
43 /* Ensure that the initial ipc namespace always has devices available. */
44 #define BINDERFS_MAX_MINOR_CAPPED (BINDERFS_MAX_MINOR - 4)
46 static dev_t binderfs_dev;
47 static DEFINE_MUTEX(binderfs_minors_mutex);
48 static DEFINE_IDA(binderfs_minors);
55 enum binderfs_stats_mode {
56 binderfs_stats_mode_unset,
57 binderfs_stats_mode_global,
60 struct binder_features {
61 bool oneway_spam_detection;
65 static const struct constant_table binderfs_param_stats[] = {
66 { "global", binderfs_stats_mode_global },
70 static const struct fs_parameter_spec binderfs_fs_parameters[] = {
71 fsparam_u32("max", Opt_max),
72 fsparam_enum("stats", Opt_stats_mode, binderfs_param_stats),
76 static struct binder_features binder_features = {
77 .oneway_spam_detection = true,
78 .extended_error = true,
81 static inline struct binderfs_info *BINDERFS_SB(const struct super_block *sb)
86 bool is_binderfs_device(const struct inode *inode)
88 if (inode->i_sb->s_magic == BINDERFS_SUPER_MAGIC)
95 * binderfs_binder_device_create - allocate inode from super block of a
97 * @ref_inode: inode from wich the super block will be taken
98 * @userp: buffer to copy information about new device for userspace to
99 * @req: struct binderfs_device as copied from userspace
101 * This function allocates a new binder_device and reserves a new minor
103 * Minor numbers are limited and tracked globally in binderfs_minors. The
104 * function will stash a struct binder_device for the specific binder
105 * device in i_private of the inode.
106 * It will go on to allocate a new inode from the super block of the
107 * filesystem mount, stash a struct binder_device in its i_private field
108 * and attach a dentry to that inode.
110 * Return: 0 on success, negative errno on failure
112 static int binderfs_binder_device_create(struct inode *ref_inode,
113 struct binderfs_device __user *userp,
114 struct binderfs_device *req)
117 struct dentry *dentry, *root;
118 struct binder_device *device;
121 struct inode *inode = NULL;
122 struct super_block *sb = ref_inode->i_sb;
123 struct binderfs_info *info = sb->s_fs_info;
124 #if defined(CONFIG_IPC_NS)
125 bool use_reserve = (info->ipc_ns == &init_ipc_ns);
127 bool use_reserve = true;
130 /* Reserve new minor number for the new device. */
131 mutex_lock(&binderfs_minors_mutex);
132 if (++info->device_count <= info->mount_opts.max)
133 minor = ida_alloc_max(&binderfs_minors,
134 use_reserve ? BINDERFS_MAX_MINOR :
135 BINDERFS_MAX_MINOR_CAPPED,
140 --info->device_count;
141 mutex_unlock(&binderfs_minors_mutex);
144 mutex_unlock(&binderfs_minors_mutex);
147 device = kzalloc(sizeof(*device), GFP_KERNEL);
151 inode = new_inode(sb);
155 inode->i_ino = minor + INODE_OFFSET;
156 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
157 init_special_inode(inode, S_IFCHR | 0600,
158 MKDEV(MAJOR(binderfs_dev), minor));
159 inode->i_fop = &binder_fops;
160 inode->i_uid = info->root_uid;
161 inode->i_gid = info->root_gid;
163 req->name[BINDERFS_MAX_NAME] = '\0'; /* NUL-terminate */
164 name_len = strlen(req->name);
165 /* Make sure to include terminating NUL byte */
166 name = kmemdup(req->name, name_len + 1, GFP_KERNEL);
170 refcount_set(&device->ref, 1);
171 device->binderfs_inode = inode;
172 device->context.binder_context_mgr_uid = INVALID_UID;
173 device->context.name = name;
174 device->miscdev.name = name;
175 device->miscdev.minor = minor;
176 mutex_init(&device->context.context_mgr_node_lock);
178 req->major = MAJOR(binderfs_dev);
181 if (userp && copy_to_user(userp, req, sizeof(*req))) {
187 inode_lock(d_inode(root));
190 dentry = lookup_one_len(name, root, name_len);
191 if (IS_ERR(dentry)) {
192 inode_unlock(d_inode(root));
193 ret = PTR_ERR(dentry);
197 if (d_really_is_positive(dentry)) {
200 inode_unlock(d_inode(root));
205 inode->i_private = device;
206 d_instantiate(dentry, inode);
207 fsnotify_create(root->d_inode, dentry);
208 inode_unlock(d_inode(root));
215 mutex_lock(&binderfs_minors_mutex);
216 --info->device_count;
217 ida_free(&binderfs_minors, minor);
218 mutex_unlock(&binderfs_minors_mutex);
225 * binderfs_ctl_ioctl - handle binder device node allocation requests
227 * The request handler for the binder-control device. All requests operate on
228 * the binderfs mount the binder-control device resides in:
230 * Allocate a new binder device.
232 * Return: 0 on success, negative errno on failure
234 static long binder_ctl_ioctl(struct file *file, unsigned int cmd,
238 struct inode *inode = file_inode(file);
239 struct binderfs_device __user *device = (struct binderfs_device __user *)arg;
240 struct binderfs_device device_req;
244 ret = copy_from_user(&device_req, device, sizeof(device_req));
250 ret = binderfs_binder_device_create(inode, device, &device_req);
259 static void binderfs_evict_inode(struct inode *inode)
261 struct binder_device *device = inode->i_private;
262 struct binderfs_info *info = BINDERFS_SB(inode->i_sb);
266 if (!S_ISCHR(inode->i_mode) || !device)
269 mutex_lock(&binderfs_minors_mutex);
270 --info->device_count;
271 ida_free(&binderfs_minors, device->miscdev.minor);
272 mutex_unlock(&binderfs_minors_mutex);
274 if (refcount_dec_and_test(&device->ref)) {
275 kfree(device->context.name);
280 static int binderfs_fs_context_parse_param(struct fs_context *fc,
281 struct fs_parameter *param)
284 struct binderfs_mount_opts *ctx = fc->fs_private;
285 struct fs_parse_result result;
287 opt = fs_parse(fc, binderfs_fs_parameters, param, &result);
293 if (result.uint_32 > BINDERFS_MAX_MINOR)
294 return invalfc(fc, "Bad value for '%s'", param->key);
296 ctx->max = result.uint_32;
299 if (!capable(CAP_SYS_ADMIN))
302 ctx->stats_mode = result.uint_32;
305 return invalfc(fc, "Unsupported parameter '%s'", param->key);
311 static int binderfs_fs_context_reconfigure(struct fs_context *fc)
313 struct binderfs_mount_opts *ctx = fc->fs_private;
314 struct binderfs_info *info = BINDERFS_SB(fc->root->d_sb);
316 if (info->mount_opts.stats_mode != ctx->stats_mode)
317 return invalfc(fc, "Binderfs stats mode cannot be changed during a remount");
319 info->mount_opts.stats_mode = ctx->stats_mode;
320 info->mount_opts.max = ctx->max;
324 static int binderfs_show_options(struct seq_file *seq, struct dentry *root)
326 struct binderfs_info *info = BINDERFS_SB(root->d_sb);
328 if (info->mount_opts.max <= BINDERFS_MAX_MINOR)
329 seq_printf(seq, ",max=%d", info->mount_opts.max);
331 switch (info->mount_opts.stats_mode) {
332 case binderfs_stats_mode_unset:
334 case binderfs_stats_mode_global:
335 seq_printf(seq, ",stats=global");
342 static const struct super_operations binderfs_super_ops = {
343 .evict_inode = binderfs_evict_inode,
344 .show_options = binderfs_show_options,
345 .statfs = simple_statfs,
348 static inline bool is_binderfs_control_device(const struct dentry *dentry)
350 struct binderfs_info *info = dentry->d_sb->s_fs_info;
352 return info->control_dentry == dentry;
355 static int binderfs_rename(struct user_namespace *mnt_userns,
356 struct inode *old_dir, struct dentry *old_dentry,
357 struct inode *new_dir, struct dentry *new_dentry,
360 if (is_binderfs_control_device(old_dentry) ||
361 is_binderfs_control_device(new_dentry))
364 return simple_rename(&init_user_ns, old_dir, old_dentry, new_dir,
368 static int binderfs_unlink(struct inode *dir, struct dentry *dentry)
370 if (is_binderfs_control_device(dentry))
373 return simple_unlink(dir, dentry);
376 static const struct file_operations binder_ctl_fops = {
377 .owner = THIS_MODULE,
378 .open = nonseekable_open,
379 .unlocked_ioctl = binder_ctl_ioctl,
380 .compat_ioctl = binder_ctl_ioctl,
381 .llseek = noop_llseek,
385 * binderfs_binder_ctl_create - create a new binder-control device
386 * @sb: super block of the binderfs mount
388 * This function creates a new binder-control device node in the binderfs mount
389 * referred to by @sb.
391 * Return: 0 on success, negative errno on failure
393 static int binderfs_binder_ctl_create(struct super_block *sb)
396 struct dentry *dentry;
397 struct binder_device *device;
398 struct inode *inode = NULL;
399 struct dentry *root = sb->s_root;
400 struct binderfs_info *info = sb->s_fs_info;
401 #if defined(CONFIG_IPC_NS)
402 bool use_reserve = (info->ipc_ns == &init_ipc_ns);
404 bool use_reserve = true;
407 device = kzalloc(sizeof(*device), GFP_KERNEL);
411 /* If we have already created a binder-control node, return. */
412 if (info->control_dentry) {
418 inode = new_inode(sb);
422 /* Reserve a new minor number for the new device. */
423 mutex_lock(&binderfs_minors_mutex);
424 minor = ida_alloc_max(&binderfs_minors,
425 use_reserve ? BINDERFS_MAX_MINOR :
426 BINDERFS_MAX_MINOR_CAPPED,
428 mutex_unlock(&binderfs_minors_mutex);
434 inode->i_ino = SECOND_INODE;
435 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
436 init_special_inode(inode, S_IFCHR | 0600,
437 MKDEV(MAJOR(binderfs_dev), minor));
438 inode->i_fop = &binder_ctl_fops;
439 inode->i_uid = info->root_uid;
440 inode->i_gid = info->root_gid;
442 refcount_set(&device->ref, 1);
443 device->binderfs_inode = inode;
444 device->miscdev.minor = minor;
446 dentry = d_alloc_name(root, "binder-control");
450 inode->i_private = device;
451 info->control_dentry = dentry;
452 d_add(dentry, inode);
463 static const struct inode_operations binderfs_dir_inode_operations = {
464 .lookup = simple_lookup,
465 .rename = binderfs_rename,
466 .unlink = binderfs_unlink,
469 static struct inode *binderfs_make_inode(struct super_block *sb, int mode)
475 ret->i_ino = iunique(sb, BINDERFS_MAX_MINOR + INODE_OFFSET);
477 ret->i_atime = ret->i_mtime = ret->i_ctime = current_time(ret);
482 static struct dentry *binderfs_create_dentry(struct dentry *parent,
485 struct dentry *dentry;
487 dentry = lookup_one_len(name, parent, strlen(name));
491 /* Return error if the file/dir already exists. */
492 if (d_really_is_positive(dentry)) {
494 return ERR_PTR(-EEXIST);
500 void binderfs_remove_file(struct dentry *dentry)
502 struct inode *parent_inode;
504 parent_inode = d_inode(dentry->d_parent);
505 inode_lock(parent_inode);
506 if (simple_positive(dentry)) {
508 simple_unlink(parent_inode, dentry);
512 inode_unlock(parent_inode);
515 struct dentry *binderfs_create_file(struct dentry *parent, const char *name,
516 const struct file_operations *fops,
519 struct dentry *dentry;
520 struct inode *new_inode, *parent_inode;
521 struct super_block *sb;
523 parent_inode = d_inode(parent);
524 inode_lock(parent_inode);
526 dentry = binderfs_create_dentry(parent, name);
530 sb = parent_inode->i_sb;
531 new_inode = binderfs_make_inode(sb, S_IFREG | 0444);
534 dentry = ERR_PTR(-ENOMEM);
538 new_inode->i_fop = fops;
539 new_inode->i_private = data;
540 d_instantiate(dentry, new_inode);
541 fsnotify_create(parent_inode, dentry);
544 inode_unlock(parent_inode);
548 static struct dentry *binderfs_create_dir(struct dentry *parent,
551 struct dentry *dentry;
552 struct inode *new_inode, *parent_inode;
553 struct super_block *sb;
555 parent_inode = d_inode(parent);
556 inode_lock(parent_inode);
558 dentry = binderfs_create_dentry(parent, name);
562 sb = parent_inode->i_sb;
563 new_inode = binderfs_make_inode(sb, S_IFDIR | 0755);
566 dentry = ERR_PTR(-ENOMEM);
570 new_inode->i_fop = &simple_dir_operations;
571 new_inode->i_op = &simple_dir_inode_operations;
573 set_nlink(new_inode, 2);
574 d_instantiate(dentry, new_inode);
575 inc_nlink(parent_inode);
576 fsnotify_mkdir(parent_inode, dentry);
579 inode_unlock(parent_inode);
583 static int binder_features_show(struct seq_file *m, void *unused)
585 bool *feature = m->private;
587 seq_printf(m, "%d\n", *feature);
591 DEFINE_SHOW_ATTRIBUTE(binder_features);
593 static int init_binder_features(struct super_block *sb)
595 struct dentry *dentry, *dir;
597 dir = binderfs_create_dir(sb->s_root, "features");
601 dentry = binderfs_create_file(dir, "oneway_spam_detection",
602 &binder_features_fops,
603 &binder_features.oneway_spam_detection);
605 return PTR_ERR(dentry);
607 dentry = binderfs_create_file(dir, "extended_error",
608 &binder_features_fops,
609 &binder_features.extended_error);
611 return PTR_ERR(dentry);
616 static int init_binder_logs(struct super_block *sb)
618 struct dentry *binder_logs_root_dir, *dentry, *proc_log_dir;
619 const struct binder_debugfs_entry *db_entry;
620 struct binderfs_info *info;
623 binder_logs_root_dir = binderfs_create_dir(sb->s_root,
625 if (IS_ERR(binder_logs_root_dir)) {
626 ret = PTR_ERR(binder_logs_root_dir);
630 binder_for_each_debugfs_entry(db_entry) {
631 dentry = binderfs_create_file(binder_logs_root_dir,
635 if (IS_ERR(dentry)) {
636 ret = PTR_ERR(dentry);
641 proc_log_dir = binderfs_create_dir(binder_logs_root_dir, "proc");
642 if (IS_ERR(proc_log_dir)) {
643 ret = PTR_ERR(proc_log_dir);
646 info = sb->s_fs_info;
647 info->proc_log_dir = proc_log_dir;
653 static int binderfs_fill_super(struct super_block *sb, struct fs_context *fc)
656 struct binderfs_info *info;
657 struct binderfs_mount_opts *ctx = fc->fs_private;
658 struct inode *inode = NULL;
659 struct binderfs_device device_info = {};
663 sb->s_blocksize = PAGE_SIZE;
664 sb->s_blocksize_bits = PAGE_SHIFT;
667 * The binderfs filesystem can be mounted by userns root in a
668 * non-initial userns. By default such mounts have the SB_I_NODEV flag
669 * set in s_iflags to prevent security issues where userns root can
670 * just create random device nodes via mknod() since it owns the
671 * filesystem mount. But binderfs does not allow to create any files
672 * including devices nodes. The only way to create binder devices nodes
673 * is through the binder-control device which userns root is explicitly
674 * allowed to do. So removing the SB_I_NODEV flag from s_iflags is both
675 * necessary and safe.
677 sb->s_iflags &= ~SB_I_NODEV;
678 sb->s_iflags |= SB_I_NOEXEC;
679 sb->s_magic = BINDERFS_SUPER_MAGIC;
680 sb->s_op = &binderfs_super_ops;
683 sb->s_fs_info = kzalloc(sizeof(struct binderfs_info), GFP_KERNEL);
686 info = sb->s_fs_info;
688 info->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
690 info->root_gid = make_kgid(sb->s_user_ns, 0);
691 if (!gid_valid(info->root_gid))
692 info->root_gid = GLOBAL_ROOT_GID;
693 info->root_uid = make_kuid(sb->s_user_ns, 0);
694 if (!uid_valid(info->root_uid))
695 info->root_uid = GLOBAL_ROOT_UID;
696 info->mount_opts.max = ctx->max;
697 info->mount_opts.stats_mode = ctx->stats_mode;
699 inode = new_inode(sb);
703 inode->i_ino = FIRST_INODE;
704 inode->i_fop = &simple_dir_operations;
705 inode->i_mode = S_IFDIR | 0755;
706 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
707 inode->i_op = &binderfs_dir_inode_operations;
710 sb->s_root = d_make_root(inode);
714 ret = binderfs_binder_ctl_create(sb);
718 name = binder_devices_param;
719 for (len = strcspn(name, ","); len > 0; len = strcspn(name, ",")) {
720 strscpy(device_info.name, name, len + 1);
721 ret = binderfs_binder_device_create(inode, NULL, &device_info);
729 ret = init_binder_features(sb);
733 if (info->mount_opts.stats_mode == binderfs_stats_mode_global)
734 return init_binder_logs(sb);
739 static int binderfs_fs_context_get_tree(struct fs_context *fc)
741 return get_tree_nodev(fc, binderfs_fill_super);
744 static void binderfs_fs_context_free(struct fs_context *fc)
746 struct binderfs_mount_opts *ctx = fc->fs_private;
751 static const struct fs_context_operations binderfs_fs_context_ops = {
752 .free = binderfs_fs_context_free,
753 .get_tree = binderfs_fs_context_get_tree,
754 .parse_param = binderfs_fs_context_parse_param,
755 .reconfigure = binderfs_fs_context_reconfigure,
758 static int binderfs_init_fs_context(struct fs_context *fc)
760 struct binderfs_mount_opts *ctx;
762 ctx = kzalloc(sizeof(struct binderfs_mount_opts), GFP_KERNEL);
766 ctx->max = BINDERFS_MAX_MINOR;
767 ctx->stats_mode = binderfs_stats_mode_unset;
769 fc->fs_private = ctx;
770 fc->ops = &binderfs_fs_context_ops;
775 static void binderfs_kill_super(struct super_block *sb)
777 struct binderfs_info *info = sb->s_fs_info;
780 * During inode eviction struct binderfs_info is needed.
781 * So first wipe the super_block then free struct binderfs_info.
783 kill_litter_super(sb);
785 if (info && info->ipc_ns)
786 put_ipc_ns(info->ipc_ns);
791 static struct file_system_type binder_fs_type = {
793 .init_fs_context = binderfs_init_fs_context,
794 .parameters = binderfs_fs_parameters,
795 .kill_sb = binderfs_kill_super,
796 .fs_flags = FS_USERNS_MOUNT,
799 int __init init_binderfs(void)
805 /* Verify that the default binderfs device names are valid. */
806 name = binder_devices_param;
807 for (len = strcspn(name, ","); len > 0; len = strcspn(name, ",")) {
808 if (len > BINDERFS_MAX_NAME)
815 /* Allocate new major number for binderfs. */
816 ret = alloc_chrdev_region(&binderfs_dev, 0, BINDERFS_MAX_MINOR,
821 ret = register_filesystem(&binder_fs_type);
823 unregister_chrdev_region(binderfs_dev, BINDERFS_MAX_MINOR);