xref: /linux/fs/nsfs.c (revision 8bc7c5e525584903ea83332e18a2118ed3b1985e)
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/pid_namespace.h>
12 #include <linux/user_namespace.h>
13 #include <linux/nsfs.h>
14 #include <linux/uaccess.h>
15 
16 #include "mount.h"
17 #include "internal.h"
18 
19 static struct vfsmount *nsfs_mnt;
20 
21 static long ns_ioctl(struct file *filp, unsigned int ioctl,
22 			unsigned long arg);
23 static const struct file_operations ns_file_operations = {
24 	.llseek		= no_llseek,
25 	.unlocked_ioctl = ns_ioctl,
26 	.compat_ioctl   = compat_ptr_ioctl,
27 };
28 
29 static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
30 {
31 	struct inode *inode = d_inode(dentry);
32 	struct ns_common *ns = inode->i_private;
33 	const struct proc_ns_operations *ns_ops = ns->ops;
34 
35 	return dynamic_dname(buffer, buflen, "%s:[%lu]",
36 		ns_ops->name, inode->i_ino);
37 }
38 
39 const struct dentry_operations ns_dentry_operations = {
40 	.d_delete	= always_delete_dentry,
41 	.d_dname	= ns_dname,
42 	.d_prune	= stashed_dentry_prune,
43 };
44 
45 static void nsfs_evict(struct inode *inode)
46 {
47 	struct ns_common *ns = inode->i_private;
48 	clear_inode(inode);
49 	ns->ops->put(ns);
50 }
51 
52 int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
53 		     void *private_data)
54 {
55 	struct ns_common *ns;
56 
57 	ns = ns_get_cb(private_data);
58 	if (!ns)
59 		return -ENOENT;
60 
61 	return path_from_stashed(&ns->stashed, nsfs_mnt, ns, path);
62 }
63 
64 struct ns_get_path_task_args {
65 	const struct proc_ns_operations *ns_ops;
66 	struct task_struct *task;
67 };
68 
69 static struct ns_common *ns_get_path_task(void *private_data)
70 {
71 	struct ns_get_path_task_args *args = private_data;
72 
73 	return args->ns_ops->get(args->task);
74 }
75 
76 int ns_get_path(struct path *path, struct task_struct *task,
77 		  const struct proc_ns_operations *ns_ops)
78 {
79 	struct ns_get_path_task_args args = {
80 		.ns_ops	= ns_ops,
81 		.task	= task,
82 	};
83 
84 	return ns_get_path_cb(path, ns_get_path_task, &args);
85 }
86 
87 /**
88  * open_namespace - open a namespace
89  * @ns: the namespace to open
90  *
91  * This will consume a reference to @ns indendent of success or failure.
92  *
93  * Return: A file descriptor on success or a negative error code on failure.
94  */
95 int open_namespace(struct ns_common *ns)
96 {
97 	struct path path __free(path_put) = {};
98 	struct file *f;
99 	int err;
100 
101 	/* call first to consume reference */
102 	err = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
103 	if (err < 0)
104 		return err;
105 
106 	CLASS(get_unused_fd, fd)(O_CLOEXEC);
107 	if (fd < 0)
108 		return fd;
109 
110 	f = dentry_open(&path, O_RDONLY, current_cred());
111 	if (IS_ERR(f))
112 		return PTR_ERR(f);
113 
114 	fd_install(fd, f);
115 	return take_fd(fd);
116 }
117 
118 int open_related_ns(struct ns_common *ns,
119 		   struct ns_common *(*get_ns)(struct ns_common *ns))
120 {
121 	struct ns_common *relative;
122 
123 	relative = get_ns(ns);
124 	if (IS_ERR(relative))
125 		return PTR_ERR(relative);
126 
127 	return open_namespace(relative);
128 }
129 EXPORT_SYMBOL_GPL(open_related_ns);
130 
131 static long ns_ioctl(struct file *filp, unsigned int ioctl,
132 			unsigned long arg)
133 {
134 	struct user_namespace *user_ns;
135 	struct pid_namespace *pid_ns;
136 	struct task_struct *tsk;
137 	struct ns_common *ns = get_proc_ns(file_inode(filp));
138 	uid_t __user *argp;
139 	uid_t uid;
140 	int ret;
141 
142 	switch (ioctl) {
143 	case NS_GET_USERNS:
144 		return open_related_ns(ns, ns_get_owner);
145 	case NS_GET_PARENT:
146 		if (!ns->ops->get_parent)
147 			return -EINVAL;
148 		return open_related_ns(ns, ns->ops->get_parent);
149 	case NS_GET_NSTYPE:
150 		return ns->ops->type;
151 	case NS_GET_OWNER_UID:
152 		if (ns->ops->type != CLONE_NEWUSER)
153 			return -EINVAL;
154 		user_ns = container_of(ns, struct user_namespace, ns);
155 		argp = (uid_t __user *) arg;
156 		uid = from_kuid_munged(current_user_ns(), user_ns->owner);
157 		return put_user(uid, argp);
158 	case NS_GET_MNTNS_ID: {
159 		struct mnt_namespace *mnt_ns;
160 		__u64 __user *idp;
161 		__u64 id;
162 
163 		if (ns->ops->type != CLONE_NEWNS)
164 			return -EINVAL;
165 
166 		mnt_ns = container_of(ns, struct mnt_namespace, ns);
167 		idp = (__u64 __user *)arg;
168 		id = mnt_ns->seq;
169 		return put_user(id, idp);
170 	}
171 	case NS_GET_PID_FROM_PIDNS:
172 		fallthrough;
173 	case NS_GET_TGID_FROM_PIDNS:
174 		fallthrough;
175 	case NS_GET_PID_IN_PIDNS:
176 		fallthrough;
177 	case NS_GET_TGID_IN_PIDNS: {
178 		if (ns->ops->type != CLONE_NEWPID)
179 			return -EINVAL;
180 
181 		ret = -ESRCH;
182 		pid_ns = container_of(ns, struct pid_namespace, ns);
183 
184 		guard(rcu)();
185 
186 		if (ioctl == NS_GET_PID_IN_PIDNS ||
187 		    ioctl == NS_GET_TGID_IN_PIDNS)
188 			tsk = find_task_by_vpid(arg);
189 		else
190 			tsk = find_task_by_pid_ns(arg, pid_ns);
191 		if (!tsk)
192 			break;
193 
194 		switch (ioctl) {
195 		case NS_GET_PID_FROM_PIDNS:
196 			ret = task_pid_vnr(tsk);
197 			break;
198 		case NS_GET_TGID_FROM_PIDNS:
199 			ret = task_tgid_vnr(tsk);
200 			break;
201 		case NS_GET_PID_IN_PIDNS:
202 			ret = task_pid_nr_ns(tsk, pid_ns);
203 			break;
204 		case NS_GET_TGID_IN_PIDNS:
205 			ret = task_tgid_nr_ns(tsk, pid_ns);
206 			break;
207 		default:
208 			ret = 0;
209 			break;
210 		}
211 
212 		if (!ret)
213 			ret = -ESRCH;
214 		break;
215 	}
216 	default:
217 		ret = -ENOTTY;
218 	}
219 
220 	return ret;
221 }
222 
223 int ns_get_name(char *buf, size_t size, struct task_struct *task,
224 			const struct proc_ns_operations *ns_ops)
225 {
226 	struct ns_common *ns;
227 	int res = -ENOENT;
228 	const char *name;
229 	ns = ns_ops->get(task);
230 	if (ns) {
231 		name = ns_ops->real_ns_name ? : ns_ops->name;
232 		res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
233 		ns_ops->put(ns);
234 	}
235 	return res;
236 }
237 
238 bool proc_ns_file(const struct file *file)
239 {
240 	return file->f_op == &ns_file_operations;
241 }
242 
243 /**
244  * ns_match() - Returns true if current namespace matches dev/ino provided.
245  * @ns: current namespace
246  * @dev: dev_t from nsfs that will be matched against current nsfs
247  * @ino: ino_t from nsfs that will be matched against current nsfs
248  *
249  * Return: true if dev and ino matches the current nsfs.
250  */
251 bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
252 {
253 	return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
254 }
255 
256 
257 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
258 {
259 	struct inode *inode = d_inode(dentry);
260 	const struct ns_common *ns = inode->i_private;
261 	const struct proc_ns_operations *ns_ops = ns->ops;
262 
263 	seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
264 	return 0;
265 }
266 
267 static const struct super_operations nsfs_ops = {
268 	.statfs = simple_statfs,
269 	.evict_inode = nsfs_evict,
270 	.show_path = nsfs_show_path,
271 };
272 
273 static int nsfs_init_inode(struct inode *inode, void *data)
274 {
275 	struct ns_common *ns = data;
276 
277 	inode->i_private = data;
278 	inode->i_mode |= S_IRUGO;
279 	inode->i_fop = &ns_file_operations;
280 	inode->i_ino = ns->inum;
281 	return 0;
282 }
283 
284 static void nsfs_put_data(void *data)
285 {
286 	struct ns_common *ns = data;
287 	ns->ops->put(ns);
288 }
289 
290 static const struct stashed_operations nsfs_stashed_ops = {
291 	.init_inode = nsfs_init_inode,
292 	.put_data = nsfs_put_data,
293 };
294 
295 static int nsfs_init_fs_context(struct fs_context *fc)
296 {
297 	struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
298 	if (!ctx)
299 		return -ENOMEM;
300 	ctx->ops = &nsfs_ops;
301 	ctx->dops = &ns_dentry_operations;
302 	fc->s_fs_info = (void *)&nsfs_stashed_ops;
303 	return 0;
304 }
305 
306 static struct file_system_type nsfs = {
307 	.name = "nsfs",
308 	.init_fs_context = nsfs_init_fs_context,
309 	.kill_sb = kill_anon_super,
310 };
311 
312 void __init nsfs_init(void)
313 {
314 	nsfs_mnt = kern_mount(&nsfs);
315 	if (IS_ERR(nsfs_mnt))
316 		panic("can't set nsfs up\n");
317 	nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
318 }
319