1// SPDX-License-Identifier: GPL-2.0
  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);	/* 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	void *ret;
110
111	do {
112		struct ns_common *ns = ns_get_cb(private_data);
113		if (!ns)
114			return ERR_PTR(-ENOENT);
115
116		ret = __ns_get_path(path, ns);
117	} while (ret == ERR_PTR(-EAGAIN));
118
119	return ret;
120}
121
122struct ns_get_path_task_args {
123	const struct proc_ns_operations *ns_ops;
124	struct task_struct *task;
125};
126
127static struct ns_common *ns_get_path_task(void *private_data)
128{
129	struct ns_get_path_task_args *args = private_data;
130
131	return args->ns_ops->get(args->task);
132}
133
134void *ns_get_path(struct path *path, struct task_struct *task,
135		  const struct proc_ns_operations *ns_ops)
136{
137	struct ns_get_path_task_args args = {
138		.ns_ops	= ns_ops,
139		.task	= task,
140	};
141
142	return ns_get_path_cb(path, ns_get_path_task, &args);
143}
144
145int open_related_ns(struct ns_common *ns,
146		   struct ns_common *(*get_ns)(struct ns_common *ns))
147{
148	struct path path = {};
149	struct file *f;
150	void *err;
151	int fd;
152
153	fd = get_unused_fd_flags(O_CLOEXEC);
154	if (fd < 0)
155		return fd;
156
157	do {
158		struct ns_common *relative;
159
160		relative = get_ns(ns);
161		if (IS_ERR(relative)) {
162			put_unused_fd(fd);
163			return PTR_ERR(relative);
164		}
165
166		err = __ns_get_path(&path, relative);
167	} while (err == ERR_PTR(-EAGAIN));
168
169	if (IS_ERR(err)) {
170		put_unused_fd(fd);
171		return PTR_ERR(err);
172	}
173
174	f = dentry_open(&path, O_RDONLY, current_cred());
175	path_put(&path);
176	if (IS_ERR(f)) {
177		put_unused_fd(fd);
178		fd = PTR_ERR(f);
179	} else
180		fd_install(fd, f);
181
182	return fd;
183}
184EXPORT_SYMBOL_GPL(open_related_ns);
185
186static long ns_ioctl(struct file *filp, unsigned int ioctl,
187			unsigned long arg)
188{
189	struct user_namespace *user_ns;
190	struct ns_common *ns = get_proc_ns(file_inode(filp));
191	uid_t __user *argp;
192	uid_t uid;
193
194	switch (ioctl) {
195	case NS_GET_USERNS:
196		return open_related_ns(ns, ns_get_owner);
197	case NS_GET_PARENT:
198		if (!ns->ops->get_parent)
199			return -EINVAL;
200		return open_related_ns(ns, ns->ops->get_parent);
201	case NS_GET_NSTYPE:
202		return ns->ops->type;
203	case NS_GET_OWNER_UID:
204		if (ns->ops->type != CLONE_NEWUSER)
205			return -EINVAL;
206		user_ns = container_of(ns, struct user_namespace, ns);
207		argp = (uid_t __user *) arg;
208		uid = from_kuid_munged(current_user_ns(), user_ns->owner);
209		return put_user(uid, argp);
210	default:
211		return -ENOTTY;
212	}
213}
214
215int ns_get_name(char *buf, size_t size, struct task_struct *task,
216			const struct proc_ns_operations *ns_ops)
217{
218	struct ns_common *ns;
219	int res = -ENOENT;
220	const char *name;
221	ns = ns_ops->get(task);
222	if (ns) {
223		name = ns_ops->real_ns_name ? : ns_ops->name;
224		res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
225		ns_ops->put(ns);
226	}
227	return res;
228}
229
230struct file *proc_ns_fget(int fd)
231{
232	struct file *file;
233
234	file = fget(fd);
235	if (!file)
236		return ERR_PTR(-EBADF);
237
238	if (file->f_op != &ns_file_operations)
239		goto out_invalid;
240
241	return file;
242
243out_invalid:
244	fput(file);
245	return ERR_PTR(-EINVAL);
246}
247
248static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
249{
250	struct inode *inode = d_inode(dentry);
251	const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
252
253	seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
254	return 0;
255}
256
257static const struct super_operations nsfs_ops = {
258	.statfs = simple_statfs,
259	.evict_inode = nsfs_evict,
260	.show_path = nsfs_show_path,
261};
262
263static int nsfs_init_fs_context(struct fs_context *fc)
264{
265	struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
266	if (!ctx)
267		return -ENOMEM;
268	ctx->ops = &nsfs_ops;
269	ctx->dops = &ns_dentry_operations;
270	return 0;
271}
272
273static struct file_system_type nsfs = {
274	.name = "nsfs",
275	.init_fs_context = nsfs_init_fs_context,
276	.kill_sb = kill_anon_super,
277};
278
279void __init nsfs_init(void)
280{
281	nsfs_mnt = kern_mount(&nsfs);
282	if (IS_ERR(nsfs_mnt))
283		panic("can't set nsfs up\n");
284	nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
285}