xref: /openbmc/linux/fs/nsfs.c (revision ce623f89)
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 
14 static struct vfsmount *nsfs_mnt;
15 
16 static long ns_ioctl(struct file *filp, unsigned int ioctl,
17 			unsigned long arg);
18 static const struct file_operations ns_file_operations = {
19 	.llseek		= no_llseek,
20 	.unlocked_ioctl = ns_ioctl,
21 };
22 
23 static 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 
32 static 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 
41 const 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 
48 static 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 
55 static int __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);
71 got_it:
72 	path->mnt = mntget(mnt);
73 	path->dentry = dentry;
74 	return 0;
75 slow:
76 	rcu_read_unlock();
77 	inode = new_inode_pseudo(mnt->mnt_sb);
78 	if (!inode) {
79 		ns->ops->put(ns);
80 		return -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 -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 -EAGAIN;
102 	}
103 	goto got_it;
104 }
105 
106 int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
107 		     void *private_data)
108 {
109 	int ret;
110 
111 	do {
112 		struct ns_common *ns = ns_get_cb(private_data);
113 		if (!ns)
114 			return -ENOENT;
115 		ret = __ns_get_path(path, ns);
116 	} while (ret == -EAGAIN);
117 
118 	return ret;
119 }
120 
121 struct ns_get_path_task_args {
122 	const struct proc_ns_operations *ns_ops;
123 	struct task_struct *task;
124 };
125 
126 static struct ns_common *ns_get_path_task(void *private_data)
127 {
128 	struct ns_get_path_task_args *args = private_data;
129 
130 	return args->ns_ops->get(args->task);
131 }
132 
133 int ns_get_path(struct path *path, struct task_struct *task,
134 		  const struct proc_ns_operations *ns_ops)
135 {
136 	struct ns_get_path_task_args args = {
137 		.ns_ops	= ns_ops,
138 		.task	= task,
139 	};
140 
141 	return ns_get_path_cb(path, ns_get_path_task, &args);
142 }
143 
144 int open_related_ns(struct ns_common *ns,
145 		   struct ns_common *(*get_ns)(struct ns_common *ns))
146 {
147 	struct path path = {};
148 	struct file *f;
149 	int err;
150 	int fd;
151 
152 	fd = get_unused_fd_flags(O_CLOEXEC);
153 	if (fd < 0)
154 		return fd;
155 
156 	do {
157 		struct ns_common *relative;
158 
159 		relative = get_ns(ns);
160 		if (IS_ERR(relative)) {
161 			put_unused_fd(fd);
162 			return PTR_ERR(relative);
163 		}
164 
165 		err = __ns_get_path(&path, relative);
166 	} while (err == -EAGAIN);
167 
168 	if (err) {
169 		put_unused_fd(fd);
170 		return err;
171 	}
172 
173 	f = dentry_open(&path, O_RDONLY, current_cred());
174 	path_put(&path);
175 	if (IS_ERR(f)) {
176 		put_unused_fd(fd);
177 		fd = PTR_ERR(f);
178 	} else
179 		fd_install(fd, f);
180 
181 	return fd;
182 }
183 EXPORT_SYMBOL_GPL(open_related_ns);
184 
185 static long ns_ioctl(struct file *filp, unsigned int ioctl,
186 			unsigned long arg)
187 {
188 	struct user_namespace *user_ns;
189 	struct ns_common *ns = get_proc_ns(file_inode(filp));
190 	uid_t __user *argp;
191 	uid_t uid;
192 
193 	switch (ioctl) {
194 	case NS_GET_USERNS:
195 		return open_related_ns(ns, ns_get_owner);
196 	case NS_GET_PARENT:
197 		if (!ns->ops->get_parent)
198 			return -EINVAL;
199 		return open_related_ns(ns, ns->ops->get_parent);
200 	case NS_GET_NSTYPE:
201 		return ns->ops->type;
202 	case NS_GET_OWNER_UID:
203 		if (ns->ops->type != CLONE_NEWUSER)
204 			return -EINVAL;
205 		user_ns = container_of(ns, struct user_namespace, ns);
206 		argp = (uid_t __user *) arg;
207 		uid = from_kuid_munged(current_user_ns(), user_ns->owner);
208 		return put_user(uid, argp);
209 	default:
210 		return -ENOTTY;
211 	}
212 }
213 
214 int ns_get_name(char *buf, size_t size, struct task_struct *task,
215 			const struct proc_ns_operations *ns_ops)
216 {
217 	struct ns_common *ns;
218 	int res = -ENOENT;
219 	const char *name;
220 	ns = ns_ops->get(task);
221 	if (ns) {
222 		name = ns_ops->real_ns_name ? : ns_ops->name;
223 		res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
224 		ns_ops->put(ns);
225 	}
226 	return res;
227 }
228 
229 struct file *proc_ns_fget(int fd)
230 {
231 	struct file *file;
232 
233 	file = fget(fd);
234 	if (!file)
235 		return ERR_PTR(-EBADF);
236 
237 	if (file->f_op != &ns_file_operations)
238 		goto out_invalid;
239 
240 	return file;
241 
242 out_invalid:
243 	fput(file);
244 	return ERR_PTR(-EINVAL);
245 }
246 
247 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
248 {
249 	struct inode *inode = d_inode(dentry);
250 	const struct proc_ns_operations *ns_ops = dentry->d_fsdata;
251 
252 	seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
253 	return 0;
254 }
255 
256 static const struct super_operations nsfs_ops = {
257 	.statfs = simple_statfs,
258 	.evict_inode = nsfs_evict,
259 	.show_path = nsfs_show_path,
260 };
261 
262 static int nsfs_init_fs_context(struct fs_context *fc)
263 {
264 	struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
265 	if (!ctx)
266 		return -ENOMEM;
267 	ctx->ops = &nsfs_ops;
268 	ctx->dops = &ns_dentry_operations;
269 	return 0;
270 }
271 
272 static struct file_system_type nsfs = {
273 	.name = "nsfs",
274 	.init_fs_context = nsfs_init_fs_context,
275 	.kill_sb = kill_anon_super,
276 };
277 
278 void __init nsfs_init(void)
279 {
280 	nsfs_mnt = kern_mount(&nsfs);
281 	if (IS_ERR(nsfs_mnt))
282 		panic("can't set nsfs up\n");
283 	nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
284 }
285