1
2#include <linux/mount.h>
3#include <linux/pseudo_fs.h>
4#include <linux/file.h>
5#include <linux/fs.h>
6#include <linux/proc_ns.h>
7#include <linux/magic.h>
8#include <linux/ktime.h>
9#include <linux/seq_file.h>
10#include <linux/user_namespace.h>
11#include <linux/nsfs.h>
12#include <linux/uaccess.h>
13
14static struct vfsmount *nsfs_mnt;
15
16static long ns_ioctl(struct file *filp, unsigned int ioctl,
17 unsigned long arg);
18static const struct file_operations ns_file_operations = {
19 .llseek = no_llseek,
20 .unlocked_ioctl = ns_ioctl,
21};
22
23static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
24{
25 struct inode *inode = d_inode(dentry);
26 const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
27
28 return dynamic_dname(dentry, buffer, buflen, "%s:[%lu]",
29 ns_ops->name, inode->i_ino);
30}
31
32static void ns_prune_dentry(struct dentry *dentry)
33{
34 struct inode *inode = d_inode(dentry);
35 if (inode) {
36 struct ns_common *ns = inode->i_private;
37 atomic_long_set(&ns->stashed, 0);
38 }
39}
40
41const struct dentry_operations ns_dentry_operations =
42{
43 .d_prune = ns_prune_dentry,
44 .d_delete = always_delete_dentry,
45 .d_dname = ns_dname,
46};
47
48static void nsfs_evict(struct inode *inode)
49{
50 struct ns_common *ns = inode->i_private;
51 clear_inode(inode);
52 ns->ops->put(ns);
53}
54
55static void *__ns_get_path(struct path *path, struct ns_common *ns)
56{
57 struct vfsmount *mnt = nsfs_mnt;
58 struct dentry *dentry;
59 struct inode *inode;
60 unsigned long d;
61
62 rcu_read_lock();
63 d = atomic_long_read(&ns->stashed);
64 if (!d)
65 goto slow;
66 dentry = (struct dentry *)d;
67 if (!lockref_get_not_dead(&dentry->d_lockref))
68 goto slow;
69 rcu_read_unlock();
70 ns->ops->put(ns);
71got_it:
72 path->mnt = mntget(mnt);
73 path->dentry = dentry;
74 return NULL;
75slow:
76 rcu_read_unlock();
77 inode = new_inode_pseudo(mnt->mnt_sb);
78 if (!inode) {
79 ns->ops->put(ns);
80 return ERR_PTR(-ENOMEM);
81 }
82 inode->i_ino = ns->inum;
83 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
84 inode->i_flags |= S_IMMUTABLE;
85 inode->i_mode = S_IFREG | S_IRUGO;
86 inode->i_fop = &ns_file_operations;
87 inode->i_private = ns;
88
89 dentry = d_alloc_anon(mnt->mnt_sb);
90 if (!dentry) {
91 iput(inode);
92 return ERR_PTR(-ENOMEM);
93 }
94 d_instantiate(dentry, inode);
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);
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
251
252
253
254
255
256
257
258
259bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
260{
261 return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
262}
263
264
265static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
266{
267 struct inode *inode = d_inode(dentry);
268 const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
269
270 seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
271 return 0;
272}
273
274static const struct super_operations nsfs_ops = {
275 .statfs = simple_statfs,
276 .evict_inode = nsfs_evict,
277 .show_path = nsfs_show_path,
278};
279
280static int nsfs_init_fs_context(struct fs_context *fc)
281{
282 struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
283 if (!ctx)
284 return -ENOMEM;
285 ctx->ops = &nsfs_ops;
286 ctx->dops = &ns_dentry_operations;
287 return 0;
288}
289
290static struct file_system_type nsfs = {
291 .name = "nsfs",
292 .init_fs_context = nsfs_init_fs_context,
293 .kill_sb = kill_anon_super,
294};
295
296void __init nsfs_init(void)
297{
298 nsfs_mnt = kern_mount(&nsfs);
299 if (IS_ERR(nsfs_mnt))
300 panic("can't set nsfs up\n");
301 nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
302}
303
