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_fs.h>
7 #include <linux/proc_ns.h>
8 #include <linux/magic.h>
9 #include <linux/ktime.h>
10 #include <linux/seq_file.h>
11 #include <linux/user_namespace.h>
12 #include <linux/nsfs.h>
13 #include <linux/uaccess.h>
14 
15 #include "internal.h"
16 
17 static struct vfsmount *nsfs_mnt;
18 
19 static long ns_ioctl(struct file *filp, unsigned int ioctl,
20 			unsigned long arg);
21 static const struct file_operations ns_file_operations = {
22 	.llseek		= no_llseek,
23 	.unlocked_ioctl = ns_ioctl,
24 };
25 
ns_dname(struct dentry * dentry,char * buffer,int buflen)26 static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
27 {
28 	struct inode *inode = d_inode(dentry);
29 	const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
30 
31 	return dynamic_dname(dentry, buffer, buflen, "%s:[%lu]",
32 		ns_ops->name, inode->i_ino);
33 }
34 
ns_prune_dentry(struct dentry * dentry)35 static void ns_prune_dentry(struct dentry *dentry)
36 {
37 	struct inode *inode = d_inode(dentry);
38 	if (inode) {
39 		struct ns_common *ns = inode->i_private;
40 		atomic_long_set(&ns->stashed, 0);
41 	}
42 }
43 
44 const struct dentry_operations ns_dentry_operations =
45 {
46 	.d_prune	= ns_prune_dentry,
47 	.d_delete	= always_delete_dentry,
48 	.d_dname	= ns_dname,
49 };
50 
nsfs_evict(struct inode * inode)51 static void nsfs_evict(struct inode *inode)
52 {
53 	struct ns_common *ns = inode->i_private;
54 	clear_inode(inode);
55 	ns->ops->put(ns);
56 }
57 
__ns_get_path(struct path * path,struct ns_common * ns)58 static int __ns_get_path(struct path *path, struct ns_common *ns)
59 {
60 	struct vfsmount *mnt = nsfs_mnt;
61 	struct dentry *dentry;
62 	struct inode *inode;
63 	unsigned long d;
64 
65 	rcu_read_lock();
66 	d = atomic_long_read(&ns->stashed);
67 	if (!d)
68 		goto slow;
69 	dentry = (struct dentry *)d;
70 	if (!lockref_get_not_dead(&dentry->d_lockref))
71 		goto slow;
72 	rcu_read_unlock();
73 	ns->ops->put(ns);
74 got_it:
75 	path->mnt = mntget(mnt);
76 	path->dentry = dentry;
77 	return 0;
78 slow:
79 	rcu_read_unlock();
80 	inode = new_inode_pseudo(mnt->mnt_sb);
81 	if (!inode) {
82 		ns->ops->put(ns);
83 		return -ENOMEM;
84 	}
85 	inode->i_ino = ns->inum;
86 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
87 	inode->i_flags |= S_IMMUTABLE;
88 	inode->i_mode = S_IFREG | S_IRUGO;
89 	inode->i_fop = &ns_file_operations;
90 	inode->i_private = ns;
91 
92 	dentry = d_alloc_anon(mnt->mnt_sb);
93 	if (!dentry) {
94 		iput(inode);
95 		return -ENOMEM;
96 	}
97 	d_instantiate(dentry, inode);
98 	dentry->d_fsdata = (void *)ns->ops;
99 	d = atomic_long_cmpxchg(&ns->stashed, 0, (unsigned long)dentry);
100 	if (d) {
101 		d_delete(dentry);	/* make sure ->d_prune() does nothing */
102 		dput(dentry);
103 		cpu_relax();
104 		return -EAGAIN;
105 	}
106 	goto got_it;
107 }
108 
ns_get_path_cb(struct path * path,ns_get_path_helper_t * ns_get_cb,void * private_data)109 int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
110 		     void *private_data)
111 {
112 	int ret;
113 
114 	do {
115 		struct ns_common *ns = ns_get_cb(private_data);
116 		if (!ns)
117 			return -ENOENT;
118 		ret = __ns_get_path(path, ns);
119 	} while (ret == -EAGAIN);
120 
121 	return ret;
122 }
123 
124 struct ns_get_path_task_args {
125 	const struct proc_ns_operations *ns_ops;
126 	struct task_struct *task;
127 };
128 
ns_get_path_task(void * private_data)129 static struct ns_common *ns_get_path_task(void *private_data)
130 {
131 	struct ns_get_path_task_args *args = private_data;
132 
133 	return args->ns_ops->get(args->task);
134 }
135 
ns_get_path(struct path * path,struct task_struct * task,const struct proc_ns_operations * ns_ops)136 int ns_get_path(struct path *path, struct task_struct *task,
137 		  const struct proc_ns_operations *ns_ops)
138 {
139 	struct ns_get_path_task_args args = {
140 		.ns_ops	= ns_ops,
141 		.task	= task,
142 	};
143 
144 	return ns_get_path_cb(path, ns_get_path_task, &args);
145 }
146 
open_related_ns(struct ns_common * ns,struct ns_common * (* get_ns)(struct ns_common * ns))147 int open_related_ns(struct ns_common *ns,
148 		   struct ns_common *(*get_ns)(struct ns_common *ns))
149 {
150 	struct path path = {};
151 	struct file *f;
152 	int err;
153 	int fd;
154 
155 	fd = get_unused_fd_flags(O_CLOEXEC);
156 	if (fd < 0)
157 		return fd;
158 
159 	do {
160 		struct ns_common *relative;
161 
162 		relative = get_ns(ns);
163 		if (IS_ERR(relative)) {
164 			put_unused_fd(fd);
165 			return PTR_ERR(relative);
166 		}
167 
168 		err = __ns_get_path(&path, relative);
169 	} while (err == -EAGAIN);
170 
171 	if (err) {
172 		put_unused_fd(fd);
173 		return err;
174 	}
175 
176 	f = dentry_open(&path, O_RDONLY, current_cred());
177 	path_put(&path);
178 	if (IS_ERR(f)) {
179 		put_unused_fd(fd);
180 		fd = PTR_ERR(f);
181 	} else
182 		fd_install(fd, f);
183 
184 	return fd;
185 }
186 EXPORT_SYMBOL_GPL(open_related_ns);
187 
ns_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)188 static long ns_ioctl(struct file *filp, unsigned int ioctl,
189 			unsigned long arg)
190 {
191 	struct user_namespace *user_ns;
192 	struct ns_common *ns = get_proc_ns(file_inode(filp));
193 	uid_t __user *argp;
194 	uid_t uid;
195 
196 	switch (ioctl) {
197 	case NS_GET_USERNS:
198 		return open_related_ns(ns, ns_get_owner);
199 	case NS_GET_PARENT:
200 		if (!ns->ops->get_parent)
201 			return -EINVAL;
202 		return open_related_ns(ns, ns->ops->get_parent);
203 	case NS_GET_NSTYPE:
204 		return ns->ops->type;
205 	case NS_GET_OWNER_UID:
206 		if (ns->ops->type != CLONE_NEWUSER)
207 			return -EINVAL;
208 		user_ns = container_of(ns, struct user_namespace, ns);
209 		argp = (uid_t __user *) arg;
210 		uid = from_kuid_munged(current_user_ns(), user_ns->owner);
211 		return put_user(uid, argp);
212 	default:
213 		return -ENOTTY;
214 	}
215 }
216 
ns_get_name(char * buf,size_t size,struct task_struct * task,const struct proc_ns_operations * ns_ops)217 int ns_get_name(char *buf, size_t size, struct task_struct *task,
218 			const struct proc_ns_operations *ns_ops)
219 {
220 	struct ns_common *ns;
221 	int res = -ENOENT;
222 	const char *name;
223 	ns = ns_ops->get(task);
224 	if (ns) {
225 		name = ns_ops->real_ns_name ? : ns_ops->name;
226 		res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
227 		ns_ops->put(ns);
228 	}
229 	return res;
230 }
231 
proc_ns_file(const struct file * file)232 bool proc_ns_file(const struct file *file)
233 {
234 	return file->f_op == &ns_file_operations;
235 }
236 
proc_ns_fget(int fd)237 struct file *proc_ns_fget(int fd)
238 {
239 	struct file *file;
240 
241 	file = fget(fd);
242 	if (!file)
243 		return ERR_PTR(-EBADF);
244 
245 	if (file->f_op != &ns_file_operations)
246 		goto out_invalid;
247 
248 	return file;
249 
250 out_invalid:
251 	fput(file);
252 	return ERR_PTR(-EINVAL);
253 }
254 
255 /**
256  * ns_match() - Returns true if current namespace matches dev/ino provided.
257  * @ns_common: current ns
258  * @dev: dev_t from nsfs that will be matched against current nsfs
259  * @ino: ino_t from nsfs that will be matched against current nsfs
260  *
261  * Return: true if dev and ino matches the current nsfs.
262  */
ns_match(const struct ns_common * ns,dev_t dev,ino_t ino)263 bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
264 {
265 	return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
266 }
267 
268 
nsfs_show_path(struct seq_file * seq,struct dentry * dentry)269 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
270 {
271 	struct inode *inode = d_inode(dentry);
272 	const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
273 
274 	seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
275 	return 0;
276 }
277 
278 static const struct super_operations nsfs_ops = {
279 	.statfs = simple_statfs,
280 	.evict_inode = nsfs_evict,
281 	.show_path = nsfs_show_path,
282 };
283 
nsfs_init_fs_context(struct fs_context * fc)284 static int nsfs_init_fs_context(struct fs_context *fc)
285 {
286 	struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
287 	if (!ctx)
288 		return -ENOMEM;
289 	ctx->ops = &nsfs_ops;
290 	ctx->dops = &ns_dentry_operations;
291 	return 0;
292 }
293 
294 static struct file_system_type nsfs = {
295 	.name = "nsfs",
296 	.init_fs_context = nsfs_init_fs_context,
297 	.kill_sb = kill_anon_super,
298 };
299 
nsfs_init(void)300 void __init nsfs_init(void)
301 {
302 	nsfs_mnt = kern_mount(&nsfs);
303 	if (IS_ERR(nsfs_mnt))
304 		panic("can't set nsfs up\n");
305 	nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
306 }
307