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
43 #define BINDERFS_MAX_MINOR (1U << MINORBITS)
44 /* Ensure that the initial ipc namespace always has devices available. */
45 #define BINDERFS_MAX_MINOR_CAPPED (BINDERFS_MAX_MINOR - 4)
47 static dev_t binderfs_dev;
48 static DEFINE_MUTEX(binderfs_minors_mutex);
49 static DEFINE_IDA(binderfs_minors);
56 enum binderfs_stats_mode {
57 binderfs_stats_mode_unset,
58 binderfs_stats_mode_global,
61 static const struct constant_table binderfs_param_stats[] = {
62 { "global", binderfs_stats_mode_global },
66 static const struct fs_parameter_spec binderfs_fs_parameters[] = {
67 fsparam_u32("max", Opt_max),
68 fsparam_enum("stats", Opt_stats_mode, binderfs_param_stats),
72 static inline struct binderfs_info *BINDERFS_SB(const struct super_block *sb)
77 bool is_binderfs_device(const struct inode *inode)
79 if (inode->i_sb->s_magic == BINDERFS_SUPER_MAGIC)
86 * binderfs_binder_device_create - allocate inode from super block of a
88 * @ref_inode: inode from wich the super block will be taken
89 * @userp: buffer to copy information about new device for userspace to
90 * @req: struct binderfs_device as copied from userspace
92 * This function allocates a new binder_device and reserves a new minor
94 * Minor numbers are limited and tracked globally in binderfs_minors. The
95 * function will stash a struct binder_device for the specific binder
96 * device in i_private of the inode.
97 * It will go on to allocate a new inode from the super block of the
98 * filesystem mount, stash a struct binder_device in its i_private field
99 * and attach a dentry to that inode.
101 * Return: 0 on success, negative errno on failure
103 static int binderfs_binder_device_create(struct inode *ref_inode,
104 struct binderfs_device __user *userp,
105 struct binderfs_device *req)
108 struct dentry *dentry, *root;
109 struct binder_device *device;
112 struct inode *inode = NULL;
113 struct super_block *sb = ref_inode->i_sb;
114 struct binderfs_info *info = sb->s_fs_info;
115 #if defined(CONFIG_IPC_NS)
116 bool use_reserve = (info->ipc_ns == &init_ipc_ns);
118 bool use_reserve = true;
121 /* Reserve new minor number for the new device. */
122 mutex_lock(&binderfs_minors_mutex);
123 if (++info->device_count <= info->mount_opts.max)
124 minor = ida_alloc_max(&binderfs_minors,
125 use_reserve ? BINDERFS_MAX_MINOR :
126 BINDERFS_MAX_MINOR_CAPPED,
131 --info->device_count;
132 mutex_unlock(&binderfs_minors_mutex);
135 mutex_unlock(&binderfs_minors_mutex);
138 device = kzalloc(sizeof(*device), GFP_KERNEL);
142 inode = new_inode(sb);
146 inode->i_ino = minor + INODE_OFFSET;
147 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
148 init_special_inode(inode, S_IFCHR | 0600,
149 MKDEV(MAJOR(binderfs_dev), minor));
150 inode->i_fop = &binder_fops;
151 inode->i_uid = info->root_uid;
152 inode->i_gid = info->root_gid;
154 req->name[BINDERFS_MAX_NAME] = '\0'; /* NUL-terminate */
155 name_len = strlen(req->name);
156 /* Make sure to include terminating NUL byte */
157 name = kmemdup(req->name, name_len + 1, GFP_KERNEL);
161 refcount_set(&device->ref, 1);
162 device->binderfs_inode = inode;
163 device->context.binder_context_mgr_uid = INVALID_UID;
164 device->context.name = name;
165 device->miscdev.name = name;
166 device->miscdev.minor = minor;
167 mutex_init(&device->context.context_mgr_node_lock);
169 req->major = MAJOR(binderfs_dev);
172 if (userp && copy_to_user(userp, req, sizeof(*req))) {
178 inode_lock(d_inode(root));
181 dentry = lookup_one_len(name, root, name_len);
182 if (IS_ERR(dentry)) {
183 inode_unlock(d_inode(root));
184 ret = PTR_ERR(dentry);
188 if (d_really_is_positive(dentry)) {
191 inode_unlock(d_inode(root));
196 inode->i_private = device;
197 d_instantiate(dentry, inode);
198 fsnotify_create(root->d_inode, dentry);
199 inode_unlock(d_inode(root));
206 mutex_lock(&binderfs_minors_mutex);
207 --info->device_count;
208 ida_free(&binderfs_minors, minor);
209 mutex_unlock(&binderfs_minors_mutex);
216 * binderfs_ctl_ioctl - handle binder device node allocation requests
218 * The request handler for the binder-control device. All requests operate on
219 * the binderfs mount the binder-control device resides in:
221 * Allocate a new binder device.
223 * Return: 0 on success, negative errno on failure
225 static long binder_ctl_ioctl(struct file *file, unsigned int cmd,
229 struct inode *inode = file_inode(file);
230 struct binderfs_device __user *device = (struct binderfs_device __user *)arg;
231 struct binderfs_device device_req;
235 ret = copy_from_user(&device_req, device, sizeof(device_req));
241 ret = binderfs_binder_device_create(inode, device, &device_req);
250 static void binderfs_evict_inode(struct inode *inode)
252 struct binder_device *device = inode->i_private;
253 struct binderfs_info *info = BINDERFS_SB(inode->i_sb);
257 if (!S_ISCHR(inode->i_mode) || !device)
260 mutex_lock(&binderfs_minors_mutex);
261 --info->device_count;
262 ida_free(&binderfs_minors, device->miscdev.minor);
263 mutex_unlock(&binderfs_minors_mutex);
265 if (refcount_dec_and_test(&device->ref)) {
266 kfree(device->context.name);
271 static int binderfs_fs_context_parse_param(struct fs_context *fc,
272 struct fs_parameter *param)
275 struct binderfs_mount_opts *ctx = fc->fs_private;
276 struct fs_parse_result result;
278 opt = fs_parse(fc, binderfs_fs_parameters, param, &result);
284 if (result.uint_32 > BINDERFS_MAX_MINOR)
285 return invalfc(fc, "Bad value for '%s'", param->key);
287 ctx->max = result.uint_32;
290 if (!capable(CAP_SYS_ADMIN))
293 ctx->stats_mode = result.uint_32;
296 return invalfc(fc, "Unsupported parameter '%s'", param->key);
302 static int binderfs_fs_context_reconfigure(struct fs_context *fc)
304 struct binderfs_mount_opts *ctx = fc->fs_private;
305 struct binderfs_info *info = BINDERFS_SB(fc->root->d_sb);
307 if (info->mount_opts.stats_mode != ctx->stats_mode)
308 return invalfc(fc, "Binderfs stats mode cannot be changed during a remount");
310 info->mount_opts.stats_mode = ctx->stats_mode;
311 info->mount_opts.max = ctx->max;
315 static int binderfs_show_options(struct seq_file *seq, struct dentry *root)
317 struct binderfs_info *info = BINDERFS_SB(root->d_sb);
319 if (info->mount_opts.max <= BINDERFS_MAX_MINOR)
320 seq_printf(seq, ",max=%d", info->mount_opts.max);
322 switch (info->mount_opts.stats_mode) {
323 case binderfs_stats_mode_unset:
325 case binderfs_stats_mode_global:
326 seq_printf(seq, ",stats=global");
333 static void binderfs_put_super(struct super_block *sb)
335 struct binderfs_info *info = sb->s_fs_info;
337 if (info && info->ipc_ns)
338 put_ipc_ns(info->ipc_ns);
341 sb->s_fs_info = NULL;
344 static const struct super_operations binderfs_super_ops = {
345 .evict_inode = binderfs_evict_inode,
346 .show_options = binderfs_show_options,
347 .statfs = simple_statfs,
348 .put_super = binderfs_put_super,
351 static inline bool is_binderfs_control_device(const struct dentry *dentry)
353 struct binderfs_info *info = dentry->d_sb->s_fs_info;
355 return info->control_dentry == dentry;
358 static int binderfs_rename(struct user_namespace *mnt_userns,
359 struct inode *old_dir, struct dentry *old_dentry,
360 struct inode *new_dir, struct dentry *new_dentry,
363 if (is_binderfs_control_device(old_dentry) ||
364 is_binderfs_control_device(new_dentry))
367 return simple_rename(&init_user_ns, old_dir, old_dentry, new_dir,
371 static int binderfs_unlink(struct inode *dir, struct dentry *dentry)
373 if (is_binderfs_control_device(dentry))
376 return simple_unlink(dir, dentry);
379 static const struct file_operations binder_ctl_fops = {
380 .owner = THIS_MODULE,
381 .open = nonseekable_open,
382 .unlocked_ioctl = binder_ctl_ioctl,
383 .compat_ioctl = binder_ctl_ioctl,
384 .llseek = noop_llseek,
388 * binderfs_binder_ctl_create - create a new binder-control device
389 * @sb: super block of the binderfs mount
391 * This function creates a new binder-control device node in the binderfs mount
392 * referred to by @sb.
394 * Return: 0 on success, negative errno on failure
396 static int binderfs_binder_ctl_create(struct super_block *sb)
399 struct dentry *dentry;
400 struct binder_device *device;
401 struct inode *inode = NULL;
402 struct dentry *root = sb->s_root;
403 struct binderfs_info *info = sb->s_fs_info;
404 #if defined(CONFIG_IPC_NS)
405 bool use_reserve = (info->ipc_ns == &init_ipc_ns);
407 bool use_reserve = true;
410 device = kzalloc(sizeof(*device), GFP_KERNEL);
414 /* If we have already created a binder-control node, return. */
415 if (info->control_dentry) {
421 inode = new_inode(sb);
425 /* Reserve a new minor number for the new device. */
426 mutex_lock(&binderfs_minors_mutex);
427 minor = ida_alloc_max(&binderfs_minors,
428 use_reserve ? BINDERFS_MAX_MINOR :
429 BINDERFS_MAX_MINOR_CAPPED,
431 mutex_unlock(&binderfs_minors_mutex);
437 inode->i_ino = SECOND_INODE;
438 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
439 init_special_inode(inode, S_IFCHR | 0600,
440 MKDEV(MAJOR(binderfs_dev), minor));
441 inode->i_fop = &binder_ctl_fops;
442 inode->i_uid = info->root_uid;
443 inode->i_gid = info->root_gid;
445 refcount_set(&device->ref, 1);
446 device->binderfs_inode = inode;
447 device->miscdev.minor = minor;
449 dentry = d_alloc_name(root, "binder-control");
453 inode->i_private = device;
454 info->control_dentry = dentry;
455 d_add(dentry, inode);
466 static const struct inode_operations binderfs_dir_inode_operations = {
467 .lookup = simple_lookup,
468 .rename = binderfs_rename,
469 .unlink = binderfs_unlink,
472 static struct inode *binderfs_make_inode(struct super_block *sb, int mode)
478 ret->i_ino = iunique(sb, BINDERFS_MAX_MINOR + INODE_OFFSET);
480 ret->i_atime = ret->i_mtime = ret->i_ctime = current_time(ret);
485 static struct dentry *binderfs_create_dentry(struct dentry *parent,
488 struct dentry *dentry;
490 dentry = lookup_one_len(name, parent, strlen(name));
494 /* Return error if the file/dir already exists. */
495 if (d_really_is_positive(dentry)) {
497 return ERR_PTR(-EEXIST);
503 void binderfs_remove_file(struct dentry *dentry)
505 struct inode *parent_inode;
507 parent_inode = d_inode(dentry->d_parent);
508 inode_lock(parent_inode);
509 if (simple_positive(dentry)) {
511 simple_unlink(parent_inode, dentry);
515 inode_unlock(parent_inode);
518 struct dentry *binderfs_create_file(struct dentry *parent, const char *name,
519 const struct file_operations *fops,
522 struct dentry *dentry;
523 struct inode *new_inode, *parent_inode;
524 struct super_block *sb;
526 parent_inode = d_inode(parent);
527 inode_lock(parent_inode);
529 dentry = binderfs_create_dentry(parent, name);
533 sb = parent_inode->i_sb;
534 new_inode = binderfs_make_inode(sb, S_IFREG | 0444);
537 dentry = ERR_PTR(-ENOMEM);
541 new_inode->i_fop = fops;
542 new_inode->i_private = data;
543 d_instantiate(dentry, new_inode);
544 fsnotify_create(parent_inode, dentry);
547 inode_unlock(parent_inode);
551 static struct dentry *binderfs_create_dir(struct dentry *parent,
554 struct dentry *dentry;
555 struct inode *new_inode, *parent_inode;
556 struct super_block *sb;
558 parent_inode = d_inode(parent);
559 inode_lock(parent_inode);
561 dentry = binderfs_create_dentry(parent, name);
565 sb = parent_inode->i_sb;
566 new_inode = binderfs_make_inode(sb, S_IFDIR | 0755);
569 dentry = ERR_PTR(-ENOMEM);
573 new_inode->i_fop = &simple_dir_operations;
574 new_inode->i_op = &simple_dir_inode_operations;
576 set_nlink(new_inode, 2);
577 d_instantiate(dentry, new_inode);
578 inc_nlink(parent_inode);
579 fsnotify_mkdir(parent_inode, dentry);
582 inode_unlock(parent_inode);
586 static int init_binder_logs(struct super_block *sb)
588 struct dentry *binder_logs_root_dir, *dentry, *proc_log_dir;
589 struct binderfs_info *info;
592 binder_logs_root_dir = binderfs_create_dir(sb->s_root,
594 if (IS_ERR(binder_logs_root_dir)) {
595 ret = PTR_ERR(binder_logs_root_dir);
599 dentry = binderfs_create_file(binder_logs_root_dir, "stats",
600 &binder_stats_fops, NULL);
601 if (IS_ERR(dentry)) {
602 ret = PTR_ERR(dentry);
606 dentry = binderfs_create_file(binder_logs_root_dir, "state",
607 &binder_state_fops, NULL);
608 if (IS_ERR(dentry)) {
609 ret = PTR_ERR(dentry);
613 dentry = binderfs_create_file(binder_logs_root_dir, "transactions",
614 &binder_transactions_fops, NULL);
615 if (IS_ERR(dentry)) {
616 ret = PTR_ERR(dentry);
620 dentry = binderfs_create_file(binder_logs_root_dir,
622 &binder_transaction_log_fops,
623 &binder_transaction_log);
624 if (IS_ERR(dentry)) {
625 ret = PTR_ERR(dentry);
629 dentry = binderfs_create_file(binder_logs_root_dir,
630 "failed_transaction_log",
631 &binder_transaction_log_fops,
632 &binder_transaction_log_failed);
633 if (IS_ERR(dentry)) {
634 ret = PTR_ERR(dentry);
638 proc_log_dir = binderfs_create_dir(binder_logs_root_dir, "proc");
639 if (IS_ERR(proc_log_dir)) {
640 ret = PTR_ERR(proc_log_dir);
643 info = sb->s_fs_info;
644 info->proc_log_dir = proc_log_dir;
650 static int binderfs_fill_super(struct super_block *sb, struct fs_context *fc)
653 struct binderfs_info *info;
654 struct binderfs_mount_opts *ctx = fc->fs_private;
655 struct inode *inode = NULL;
656 struct binderfs_device device_info = {};
660 sb->s_blocksize = PAGE_SIZE;
661 sb->s_blocksize_bits = PAGE_SHIFT;
664 * The binderfs filesystem can be mounted by userns root in a
665 * non-initial userns. By default such mounts have the SB_I_NODEV flag
666 * set in s_iflags to prevent security issues where userns root can
667 * just create random device nodes via mknod() since it owns the
668 * filesystem mount. But binderfs does not allow to create any files
669 * including devices nodes. The only way to create binder devices nodes
670 * is through the binder-control device which userns root is explicitly
671 * allowed to do. So removing the SB_I_NODEV flag from s_iflags is both
672 * necessary and safe.
674 sb->s_iflags &= ~SB_I_NODEV;
675 sb->s_iflags |= SB_I_NOEXEC;
676 sb->s_magic = BINDERFS_SUPER_MAGIC;
677 sb->s_op = &binderfs_super_ops;
680 sb->s_fs_info = kzalloc(sizeof(struct binderfs_info), GFP_KERNEL);
683 info = sb->s_fs_info;
685 info->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
687 info->root_gid = make_kgid(sb->s_user_ns, 0);
688 if (!gid_valid(info->root_gid))
689 info->root_gid = GLOBAL_ROOT_GID;
690 info->root_uid = make_kuid(sb->s_user_ns, 0);
691 if (!uid_valid(info->root_uid))
692 info->root_uid = GLOBAL_ROOT_UID;
693 info->mount_opts.max = ctx->max;
694 info->mount_opts.stats_mode = ctx->stats_mode;
696 inode = new_inode(sb);
700 inode->i_ino = FIRST_INODE;
701 inode->i_fop = &simple_dir_operations;
702 inode->i_mode = S_IFDIR | 0755;
703 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
704 inode->i_op = &binderfs_dir_inode_operations;
707 sb->s_root = d_make_root(inode);
711 ret = binderfs_binder_ctl_create(sb);
715 name = binder_devices_param;
716 for (len = strcspn(name, ","); len > 0; len = strcspn(name, ",")) {
717 strscpy(device_info.name, name, len + 1);
718 ret = binderfs_binder_device_create(inode, NULL, &device_info);
726 if (info->mount_opts.stats_mode == binderfs_stats_mode_global)
727 return init_binder_logs(sb);
732 static int binderfs_fs_context_get_tree(struct fs_context *fc)
734 return get_tree_nodev(fc, binderfs_fill_super);
737 static void binderfs_fs_context_free(struct fs_context *fc)
739 struct binderfs_mount_opts *ctx = fc->fs_private;
744 static const struct fs_context_operations binderfs_fs_context_ops = {
745 .free = binderfs_fs_context_free,
746 .get_tree = binderfs_fs_context_get_tree,
747 .parse_param = binderfs_fs_context_parse_param,
748 .reconfigure = binderfs_fs_context_reconfigure,
751 static int binderfs_init_fs_context(struct fs_context *fc)
753 struct binderfs_mount_opts *ctx;
755 ctx = kzalloc(sizeof(struct binderfs_mount_opts), GFP_KERNEL);
759 ctx->max = BINDERFS_MAX_MINOR;
760 ctx->stats_mode = binderfs_stats_mode_unset;
762 fc->fs_private = ctx;
763 fc->ops = &binderfs_fs_context_ops;
768 static struct file_system_type binder_fs_type = {
770 .init_fs_context = binderfs_init_fs_context,
771 .parameters = binderfs_fs_parameters,
772 .kill_sb = kill_litter_super,
773 .fs_flags = FS_USERNS_MOUNT,
776 int __init init_binderfs(void)
782 /* Verify that the default binderfs device names are valid. */
783 name = binder_devices_param;
784 for (len = strcspn(name, ","); len > 0; len = strcspn(name, ",")) {
785 if (len > BINDERFS_MAX_NAME)
792 /* Allocate new major number for binderfs. */
793 ret = alloc_chrdev_region(&binderfs_dev, 0, BINDERFS_MAX_MINOR,
798 ret = register_filesystem(&binder_fs_type);
800 unregister_chrdev_region(binderfs_dev, BINDERFS_MAX_MINOR);