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
13static struct vfsmount *nsfs_mnt;
14
15static long ns_ioctl(struct file *filp, unsigned int ioctl,
16 unsigned long arg);
17static const struct file_operations ns_file_operations = {
18 .llseek = no_llseek,
19 .unlocked_ioctl = ns_ioctl,
20};
21
22static 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
31static 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
40const 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
47static 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
54static 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);
70got_it:
71 path->mnt = mntget(mnt);
72 path->dentry = dentry;
73 return NULL;
74slow:
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_pseudo(mnt->mnt_sb, &empty_name);
89 if (!dentry) {
90 iput(inode);
91 return ERR_PTR(-ENOMEM);
92 }
93 d_instantiate(dentry, inode);
94 dentry->d_flags |= DCACHE_RCUACCESS;
95 dentry->d_fsdata = (void *)ns->ops;
96 d = atomic_long_cmpxchg(&ns->stashed, 0, (unsigned long)dentry);
97 if (d) {
98 d_delete(dentry); /* make sure ->d_prune() does nothing */
99 dput(dentry);
100 cpu_relax();
101 return ERR_PTR(-EAGAIN);
102 }
103 goto got_it;
104}
105
106void *ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
107 void *private_data)
108{
109 struct ns_common *ns;
110 void *ret;
111
112again:
113 ns = ns_get_cb(private_data);
114 if (!ns)
115 return ERR_PTR(-ENOENT);
116
117 ret = __ns_get_path(path, ns);
118 if (IS_ERR(ret) && PTR_ERR(ret) == -EAGAIN)
119 goto again;
120 return ret;
121}
122
123struct ns_get_path_task_args {
124 const struct proc_ns_operations *ns_ops;
125 struct task_struct *task;
126};
127
128static struct ns_common *ns_get_path_task(void *private_data)
129{
130 struct ns_get_path_task_args *args = private_data;
131
132 return args->ns_ops->get(args->task);
133}
134
135void *ns_get_path(struct path *path, struct task_struct *task,
136 const struct proc_ns_operations *ns_ops)
137{
138 struct ns_get_path_task_args args = {
139 .ns_ops = ns_ops,
140 .task = task,
141 };
142
143 return ns_get_path_cb(path, ns_get_path_task, &args);
144}
145
146int open_related_ns(struct ns_common *ns,
147 struct ns_common *(*get_ns)(struct ns_common *ns))
148{
149 struct path path = {};
150 struct file *f;
151 void *err;
152 int fd;
153
154 fd = get_unused_fd_flags(O_CLOEXEC);
155 if (fd < 0)
156 return fd;
157
158 while (1) {
159 struct ns_common *relative;
160
161 relative = get_ns(ns);
162 if (IS_ERR(relative)) {
163 put_unused_fd(fd);
164 return PTR_ERR(relative);
165 }
166
167 err = __ns_get_path(&path, relative);
168 if (IS_ERR(err) && PTR_ERR(err) == -EAGAIN)
169 continue;
170 break;
171 }
172 if (IS_ERR(err)) {
173 put_unused_fd(fd);
174 return PTR_ERR(err);
175 }
176
177 f = dentry_open(&path, O_RDONLY, current_cred());
178 path_put(&path);
179 if (IS_ERR(f)) {
180 put_unused_fd(fd);
181 fd = PTR_ERR(f);
182 } else
183 fd_install(fd, f);
184
185 return fd;
186}
187EXPORT_SYMBOL_GPL(open_related_ns);
188
189static long ns_ioctl(struct file *filp, unsigned int ioctl,
190 unsigned long arg)
191{
192 struct user_namespace *user_ns;
193 struct ns_common *ns = get_proc_ns(file_inode(filp));
194 uid_t __user *argp;
195 uid_t uid;
196
197 switch (ioctl) {
198 case NS_GET_USERNS:
199 return open_related_ns(ns, ns_get_owner);
200 case NS_GET_PARENT:
201 if (!ns->ops->get_parent)
202 return -EINVAL;
203 return open_related_ns(ns, ns->ops->get_parent);
204 case NS_GET_NSTYPE:
205 return ns->ops->type;
206 case NS_GET_OWNER_UID:
207 if (ns->ops->type != CLONE_NEWUSER)
208 return -EINVAL;
209 user_ns = container_of(ns, struct user_namespace, ns);
210 argp = (uid_t __user *) arg;
211 uid = from_kuid_munged(current_user_ns(), user_ns->owner);
212 return put_user(uid, argp);
213 default:
214 return -ENOTTY;
215 }
216}
217
218int ns_get_name(char *buf, size_t size, struct task_struct *task,
219 const struct proc_ns_operations *ns_ops)
220{
221 struct ns_common *ns;
222 int res = -ENOENT;
223 const char *name;
224 ns = ns_ops->get(task);
225 if (ns) {
226 name = ns_ops->real_ns_name ? : ns_ops->name;
227 res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
228 ns_ops->put(ns);
229 }
230 return res;
231}
232
233struct file *proc_ns_fget(int fd)
234{
235 struct file *file;
236
237 file = fget(fd);
238 if (!file)
239 return ERR_PTR(-EBADF);
240
241 if (file->f_op != &ns_file_operations)
242 goto out_invalid;
243
244 return file;
245
246out_invalid:
247 fput(file);
248 return ERR_PTR(-EINVAL);
249}
250
251static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
252{
253 struct inode *inode = d_inode(dentry);
254 const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
255
256 seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
257 return 0;
258}
259
260static const struct super_operations nsfs_ops = {
261 .statfs = simple_statfs,
262 .evict_inode = nsfs_evict,
263 .show_path = nsfs_show_path,
264};
265static struct dentry *nsfs_mount(struct file_system_type *fs_type,
266 int flags, const char *dev_name, void *data)
267{
268 return mount_pseudo(fs_type, "nsfs:", &nsfs_ops,
269 &ns_dentry_operations, NSFS_MAGIC);
270}
271static struct file_system_type nsfs = {
272 .name = "nsfs",
273 .mount = nsfs_mount,
274 .kill_sb = kill_anon_super,
275};
276
277void __init nsfs_init(void)
278{
279 nsfs_mnt = kern_mount(&nsfs);
280 if (IS_ERR(nsfs_mnt))
281 panic("can't set nsfs up\n");
282 nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
283}
284