1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/mount.h> 3 #include <linux/file.h> 4 #include <linux/fs.h> 5 #include <linux/proc_ns.h> 6 #include <linux/magic.h> 7 #include <linux/ktime.h> 8 #include <linux/seq_file.h> 9 #include <linux/user_namespace.h> 10 #include <linux/nsfs.h> 11 #include <linux/uaccess.h> 12 13 static struct vfsmount *nsfs_mnt; 14 15 static long ns_ioctl(struct file *filp, unsigned int ioctl, 16 unsigned long arg); 17 static const struct file_operations ns_file_operations = { 18 .llseek = no_llseek, 19 .unlocked_ioctl = ns_ioctl, 20 }; 21 22 static char *ns_dname(struct dentry *dentry, char *buffer, int buflen) 23 { 24 struct inode *inode = d_inode(dentry); 25 const struct proc_ns_operations *ns_ops = dentry->d_fsdata; 26 27 return dynamic_dname(dentry, buffer, buflen, "%s:[%lu]", 28 ns_ops->name, inode->i_ino); 29 } 30 31 static void ns_prune_dentry(struct dentry *dentry) 32 { 33 struct inode *inode = d_inode(dentry); 34 if (inode) { 35 struct ns_common *ns = inode->i_private; 36 atomic_long_set(&ns->stashed, 0); 37 } 38 } 39 40 const struct dentry_operations ns_dentry_operations = 41 { 42 .d_prune = ns_prune_dentry, 43 .d_delete = always_delete_dentry, 44 .d_dname = ns_dname, 45 }; 46 47 static void nsfs_evict(struct inode *inode) 48 { 49 struct ns_common *ns = inode->i_private; 50 clear_inode(inode); 51 ns->ops->put(ns); 52 } 53 54 static void *__ns_get_path(struct path *path, struct ns_common *ns) 55 { 56 struct vfsmount *mnt = nsfs_mnt; 57 struct dentry *dentry; 58 struct inode *inode; 59 unsigned long d; 60 61 rcu_read_lock(); 62 d = atomic_long_read(&ns->stashed); 63 if (!d) 64 goto slow; 65 dentry = (struct dentry *)d; 66 if (!lockref_get_not_dead(&dentry->d_lockref)) 67 goto slow; 68 rcu_read_unlock(); 69 ns->ops->put(ns); 70 got_it: 71 path->mnt = mntget(mnt); 72 path->dentry = dentry; 73 return NULL; 74 slow: 75 rcu_read_unlock(); 76 inode = new_inode_pseudo(mnt->mnt_sb); 77 if (!inode) { 78 ns->ops->put(ns); 79 return ERR_PTR(-ENOMEM); 80 } 81 inode->i_ino = ns->inum; 82 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); 83 inode->i_flags |= S_IMMUTABLE; 84 inode->i_mode = S_IFREG | S_IRUGO; 85 inode->i_fop = &ns_file_operations; 86 inode->i_private = ns; 87 88 dentry = d_alloc_anon(mnt->mnt_sb); 89 if (!dentry) { 90 iput(inode); 91 return ERR_PTR(-ENOMEM); 92 } 93 d_instantiate(dentry, inode); 94 dentry->d_fsdata = (void *)ns->ops; 95 d = atomic_long_cmpxchg(&ns->stashed, 0, (unsigned long)dentry); 96 if (d) { 97 d_delete(dentry); /* make sure ->d_prune() does nothing */ 98 dput(dentry); 99 cpu_relax(); 100 return ERR_PTR(-EAGAIN); 101 } 102 goto got_it; 103 } 104 105 void *ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb, 106 void *private_data) 107 { 108 void *ret; 109 110 do { 111 struct ns_common *ns = ns_get_cb(private_data); 112 if (!ns) 113 return ERR_PTR(-ENOENT); 114 115 ret = __ns_get_path(path, ns); 116 } while (ret == ERR_PTR(-EAGAIN)); 117 118 return ret; 119 } 120 121 struct ns_get_path_task_args { 122 const struct proc_ns_operations *ns_ops; 123 struct task_struct *task; 124 }; 125 126 static struct ns_common *ns_get_path_task(void *private_data) 127 { 128 struct ns_get_path_task_args *args = private_data; 129 130 return args->ns_ops->get(args->task); 131 } 132 133 void *ns_get_path(struct path *path, struct task_struct *task, 134 const struct proc_ns_operations *ns_ops) 135 { 136 struct ns_get_path_task_args args = { 137 .ns_ops = ns_ops, 138 .task = task, 139 }; 140 141 return ns_get_path_cb(path, ns_get_path_task, &args); 142 } 143 144 int open_related_ns(struct ns_common *ns, 145 struct ns_common *(*get_ns)(struct ns_common *ns)) 146 { 147 struct path path = {}; 148 struct file *f; 149 void *err; 150 int fd; 151 152 fd = get_unused_fd_flags(O_CLOEXEC); 153 if (fd < 0) 154 return fd; 155 156 do { 157 struct ns_common *relative; 158 159 relative = get_ns(ns); 160 if (IS_ERR(relative)) { 161 put_unused_fd(fd); 162 return PTR_ERR(relative); 163 } 164 165 err = __ns_get_path(&path, relative); 166 } while (err == ERR_PTR(-EAGAIN)); 167 168 if (IS_ERR(err)) { 169 put_unused_fd(fd); 170 return PTR_ERR(err); 171 } 172 173 f = dentry_open(&path, O_RDONLY, current_cred()); 174 path_put(&path); 175 if (IS_ERR(f)) { 176 put_unused_fd(fd); 177 fd = PTR_ERR(f); 178 } else 179 fd_install(fd, f); 180 181 return fd; 182 } 183 EXPORT_SYMBOL_GPL(open_related_ns); 184 185 static long ns_ioctl(struct file *filp, unsigned int ioctl, 186 unsigned long arg) 187 { 188 struct user_namespace *user_ns; 189 struct ns_common *ns = get_proc_ns(file_inode(filp)); 190 uid_t __user *argp; 191 uid_t uid; 192 193 switch (ioctl) { 194 case NS_GET_USERNS: 195 return open_related_ns(ns, ns_get_owner); 196 case NS_GET_PARENT: 197 if (!ns->ops->get_parent) 198 return -EINVAL; 199 return open_related_ns(ns, ns->ops->get_parent); 200 case NS_GET_NSTYPE: 201 return ns->ops->type; 202 case NS_GET_OWNER_UID: 203 if (ns->ops->type != CLONE_NEWUSER) 204 return -EINVAL; 205 user_ns = container_of(ns, struct user_namespace, ns); 206 argp = (uid_t __user *) arg; 207 uid = from_kuid_munged(current_user_ns(), user_ns->owner); 208 return put_user(uid, argp); 209 default: 210 return -ENOTTY; 211 } 212 } 213 214 int ns_get_name(char *buf, size_t size, struct task_struct *task, 215 const struct proc_ns_operations *ns_ops) 216 { 217 struct ns_common *ns; 218 int res = -ENOENT; 219 const char *name; 220 ns = ns_ops->get(task); 221 if (ns) { 222 name = ns_ops->real_ns_name ? : ns_ops->name; 223 res = snprintf(buf, size, "%s:[%u]", name, ns->inum); 224 ns_ops->put(ns); 225 } 226 return res; 227 } 228 229 struct file *proc_ns_fget(int fd) 230 { 231 struct file *file; 232 233 file = fget(fd); 234 if (!file) 235 return ERR_PTR(-EBADF); 236 237 if (file->f_op != &ns_file_operations) 238 goto out_invalid; 239 240 return file; 241 242 out_invalid: 243 fput(file); 244 return ERR_PTR(-EINVAL); 245 } 246 247 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry) 248 { 249 struct inode *inode = d_inode(dentry); 250 const struct proc_ns_operations *ns_ops = dentry->d_fsdata; 251 252 seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino); 253 return 0; 254 } 255 256 static const struct super_operations nsfs_ops = { 257 .statfs = simple_statfs, 258 .evict_inode = nsfs_evict, 259 .show_path = nsfs_show_path, 260 }; 261 static struct dentry *nsfs_mount(struct file_system_type *fs_type, 262 int flags, const char *dev_name, void *data) 263 { 264 return mount_pseudo(fs_type, "nsfs:", &nsfs_ops, 265 &ns_dentry_operations, NSFS_MAGIC); 266 } 267 static struct file_system_type nsfs = { 268 .name = "nsfs", 269 .mount = nsfs_mount, 270 .kill_sb = kill_anon_super, 271 }; 272 273 void __init nsfs_init(void) 274 { 275 nsfs_mnt = kern_mount(&nsfs); 276 if (IS_ERR(nsfs_mnt)) 277 panic("can't set nsfs up\n"); 278 nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER; 279 } 280