1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/mount.h>
3 #include <linux/pseudo_fs.h>
4 #include <linux/file.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 #include <linux/mnt_namespace.h>
20 static struct vfsmount *nsfs_mnt;
22 static long ns_ioctl(struct file *filp, unsigned int ioctl,
24 static const struct file_operations ns_file_operations = {
26 .unlocked_ioctl = ns_ioctl,
27 .compat_ioctl = compat_ptr_ioctl,
30 static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
32 struct inode *inode = d_inode(dentry);
33 struct ns_common *ns = inode->i_private;
34 const struct proc_ns_operations *ns_ops = ns->ops;
36 return dynamic_dname(buffer, buflen, "%s:[%lu]",
37 ns_ops->name, inode->i_ino);
40 const struct dentry_operations ns_dentry_operations = {
41 .d_delete = always_delete_dentry,
43 .d_prune = stashed_dentry_prune,
46 static void nsfs_evict(struct inode *inode)
48 struct ns_common *ns = inode->i_private;
53 int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
58 ns = ns_get_cb(private_data);
62 return path_from_stashed(&ns->stashed, nsfs_mnt, ns, path);
65 struct ns_get_path_task_args {
66 const struct proc_ns_operations *ns_ops;
67 struct task_struct *task;
70 static struct ns_common *ns_get_path_task(void *private_data)
72 struct ns_get_path_task_args *args = private_data;
74 return args->ns_ops->get(args->task);
77 int ns_get_path(struct path *path, struct task_struct *task,
78 const struct proc_ns_operations *ns_ops)
80 struct ns_get_path_task_args args = {
85 return ns_get_path_cb(path, ns_get_path_task, &args);
89 * open_namespace - open a namespace
90 * @ns: the namespace to open
92 * This will consume a reference to @ns indendent of success or failure.
94 * Return: A file descriptor on success or a negative error code on failure.
96 int open_namespace(struct ns_common *ns)
98 struct path path __free(path_put) = {};
102 /* call first to consume reference */
103 err = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
107 CLASS(get_unused_fd, fd)(O_CLOEXEC);
111 f = dentry_open(&path, O_RDONLY, current_cred());
119 int open_related_ns(struct ns_common *ns,
120 struct ns_common *(*get_ns)(struct ns_common *ns))
122 struct ns_common *relative;
124 relative = get_ns(ns);
125 if (IS_ERR(relative))
126 return PTR_ERR(relative);
128 return open_namespace(relative);
130 EXPORT_SYMBOL_GPL(open_related_ns);
132 static int copy_ns_info_to_user(const struct mnt_namespace *mnt_ns,
133 struct mnt_ns_info __user *uinfo, size_t usize,
134 struct mnt_ns_info *kinfo)
137 * If userspace and the kernel have the same struct size it can just
138 * be copied. If userspace provides an older struct, only the bits that
139 * userspace knows about will be copied. If userspace provides a new
140 * struct, only the bits that the kernel knows aobut will be copied and
141 * the size value will be set to the size the kernel knows about.
143 kinfo->size = min(usize, sizeof(*kinfo));
144 kinfo->mnt_ns_id = mnt_ns->seq;
145 kinfo->nr_mounts = READ_ONCE(mnt_ns->nr_mounts);
146 /* Subtract the root mount of the mount namespace. */
147 if (kinfo->nr_mounts)
150 if (copy_to_user(uinfo, kinfo, kinfo->size))
156 static long ns_ioctl(struct file *filp, unsigned int ioctl,
159 struct user_namespace *user_ns;
160 struct pid_namespace *pid_ns;
161 struct task_struct *tsk;
162 struct ns_common *ns = get_proc_ns(file_inode(filp));
163 struct mnt_namespace *mnt_ns;
164 bool previous = false;
171 return open_related_ns(ns, ns_get_owner);
173 if (!ns->ops->get_parent)
175 return open_related_ns(ns, ns->ops->get_parent);
177 return ns->ops->type;
178 case NS_GET_OWNER_UID:
179 if (ns->ops->type != CLONE_NEWUSER)
181 user_ns = container_of(ns, struct user_namespace, ns);
182 argp = (uid_t __user *) arg;
183 uid = from_kuid_munged(current_user_ns(), user_ns->owner);
184 return put_user(uid, argp);
185 case NS_GET_MNTNS_ID: {
189 if (ns->ops->type != CLONE_NEWNS)
192 mnt_ns = container_of(ns, struct mnt_namespace, ns);
193 idp = (__u64 __user *)arg;
195 return put_user(id, idp);
197 case NS_GET_PID_FROM_PIDNS:
199 case NS_GET_TGID_FROM_PIDNS:
201 case NS_GET_PID_IN_PIDNS:
203 case NS_GET_TGID_IN_PIDNS: {
204 if (ns->ops->type != CLONE_NEWPID)
208 pid_ns = container_of(ns, struct pid_namespace, ns);
212 if (ioctl == NS_GET_PID_IN_PIDNS ||
213 ioctl == NS_GET_TGID_IN_PIDNS)
214 tsk = find_task_by_vpid(arg);
216 tsk = find_task_by_pid_ns(arg, pid_ns);
221 case NS_GET_PID_FROM_PIDNS:
222 ret = task_pid_vnr(tsk);
224 case NS_GET_TGID_FROM_PIDNS:
225 ret = task_tgid_vnr(tsk);
227 case NS_GET_PID_IN_PIDNS:
228 ret = task_pid_nr_ns(tsk, pid_ns);
230 case NS_GET_TGID_IN_PIDNS:
231 ret = task_tgid_nr_ns(tsk, pid_ns);
244 /* extensible ioctls */
245 switch (_IOC_NR(ioctl)) {
246 case _IOC_NR(NS_MNT_GET_INFO): {
247 struct mnt_ns_info kinfo = {};
248 struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
249 size_t usize = _IOC_SIZE(ioctl);
251 if (ns->ops->type != CLONE_NEWNS)
257 if (usize < MNT_NS_INFO_SIZE_VER0)
260 return copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
262 case _IOC_NR(NS_MNT_GET_PREV):
265 case _IOC_NR(NS_MNT_GET_NEXT): {
266 struct mnt_ns_info kinfo = {};
267 struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
268 struct path path __free(path_put) = {};
269 struct file *f __free(fput) = NULL;
270 size_t usize = _IOC_SIZE(ioctl);
272 if (ns->ops->type != CLONE_NEWNS)
275 if (usize < MNT_NS_INFO_SIZE_VER0)
279 mnt_ns = lookup_prev_mnt_ns(to_mnt_ns(ns));
281 mnt_ns = lookup_next_mnt_ns(to_mnt_ns(ns));
283 return PTR_ERR(mnt_ns);
285 ns = to_ns_common(mnt_ns);
286 /* Transfer ownership of @mnt_ns reference to @path. */
287 ret = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
291 CLASS(get_unused_fd, fd)(O_CLOEXEC);
295 f = dentry_open(&path, O_RDONLY, current_cred());
301 * If @uinfo is passed return all information about the
302 * mount namespace as well.
304 ret = copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
309 /* Transfer reference of @f to caller's fdtable. */
310 fd_install(fd, no_free_ptr(f));
311 /* File descriptor is live so hand it off to the caller. */
321 int ns_get_name(char *buf, size_t size, struct task_struct *task,
322 const struct proc_ns_operations *ns_ops)
324 struct ns_common *ns;
327 ns = ns_ops->get(task);
329 name = ns_ops->real_ns_name ? : ns_ops->name;
330 res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
336 bool proc_ns_file(const struct file *file)
338 return file->f_op == &ns_file_operations;
342 * ns_match() - Returns true if current namespace matches dev/ino provided.
343 * @ns: current namespace
344 * @dev: dev_t from nsfs that will be matched against current nsfs
345 * @ino: ino_t from nsfs that will be matched against current nsfs
347 * Return: true if dev and ino matches the current nsfs.
349 bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
351 return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
355 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
357 struct inode *inode = d_inode(dentry);
358 const struct ns_common *ns = inode->i_private;
359 const struct proc_ns_operations *ns_ops = ns->ops;
361 seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
365 static const struct super_operations nsfs_ops = {
366 .statfs = simple_statfs,
367 .evict_inode = nsfs_evict,
368 .show_path = nsfs_show_path,
371 static int nsfs_init_inode(struct inode *inode, void *data)
373 struct ns_common *ns = data;
375 inode->i_private = data;
376 inode->i_mode |= S_IRUGO;
377 inode->i_fop = &ns_file_operations;
378 inode->i_ino = ns->inum;
382 static void nsfs_put_data(void *data)
384 struct ns_common *ns = data;
388 static const struct stashed_operations nsfs_stashed_ops = {
389 .init_inode = nsfs_init_inode,
390 .put_data = nsfs_put_data,
393 static int nsfs_init_fs_context(struct fs_context *fc)
395 struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
398 ctx->ops = &nsfs_ops;
399 ctx->dops = &ns_dentry_operations;
400 fc->s_fs_info = (void *)&nsfs_stashed_ops;
404 static struct file_system_type nsfs = {
406 .init_fs_context = nsfs_init_fs_context,
407 .kill_sb = kill_anon_super,
410 void __init nsfs_init(void)
412 nsfs_mnt = kern_mount(&nsfs);
413 if (IS_ERR(nsfs_mnt))
414 panic("can't set nsfs up\n");
415 nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;