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 = {
25 .unlocked_ioctl = ns_ioctl,
26 .compat_ioctl = compat_ptr_ioctl,
29 static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
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;
35 return dynamic_dname(buffer, buflen, "%s:[%lu]",
36 ns_ops->name, inode->i_ino);
39 const struct dentry_operations ns_dentry_operations = {
41 .d_prune = stashed_dentry_prune,
44 static void nsfs_evict(struct inode *inode)
46 struct ns_common *ns = inode->i_private;
51 int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
56 ns = ns_get_cb(private_data);
60 return path_from_stashed(&ns->stashed, nsfs_mnt, ns, path);
63 struct ns_get_path_task_args {
64 const struct proc_ns_operations *ns_ops;
65 struct task_struct *task;
68 static struct ns_common *ns_get_path_task(void *private_data)
70 struct ns_get_path_task_args *args = private_data;
72 return args->ns_ops->get(args->task);
75 int ns_get_path(struct path *path, struct task_struct *task,
76 const struct proc_ns_operations *ns_ops)
78 struct ns_get_path_task_args args = {
83 return ns_get_path_cb(path, ns_get_path_task, &args);
87 * open_namespace - open a namespace
88 * @ns: the namespace to open
90 * This will consume a reference to @ns indendent of success or failure.
92 * Return: A file descriptor on success or a negative error code on failure.
94 int open_namespace(struct ns_common *ns)
96 struct path path __free(path_put) = {};
100 /* call first to consume reference */
101 err = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
105 CLASS(get_unused_fd, fd)(O_CLOEXEC);
109 f = dentry_open(&path, O_RDONLY, current_cred());
117 int open_related_ns(struct ns_common *ns,
118 struct ns_common *(*get_ns)(struct ns_common *ns))
120 struct ns_common *relative;
122 relative = get_ns(ns);
123 if (IS_ERR(relative))
124 return PTR_ERR(relative);
126 return open_namespace(relative);
128 EXPORT_SYMBOL_GPL(open_related_ns);
130 static int copy_ns_info_to_user(const struct mnt_namespace *mnt_ns,
131 struct mnt_ns_info __user *uinfo, size_t usize,
132 struct mnt_ns_info *kinfo)
135 * If userspace and the kernel have the same struct size it can just
136 * be copied. If userspace provides an older struct, only the bits that
137 * userspace knows about will be copied. If userspace provides a new
138 * struct, only the bits that the kernel knows aobut will be copied and
139 * the size value will be set to the size the kernel knows about.
141 kinfo->size = min(usize, sizeof(*kinfo));
142 kinfo->mnt_ns_id = mnt_ns->seq;
143 kinfo->nr_mounts = READ_ONCE(mnt_ns->nr_mounts);
144 /* Subtract the root mount of the mount namespace. */
145 if (kinfo->nr_mounts)
148 if (copy_to_user(uinfo, kinfo, kinfo->size))
154 static bool nsfs_ioctl_valid(unsigned int cmd)
160 case NS_GET_OWNER_UID:
161 case NS_GET_MNTNS_ID:
162 case NS_GET_PID_FROM_PIDNS:
163 case NS_GET_TGID_FROM_PIDNS:
164 case NS_GET_PID_IN_PIDNS:
165 case NS_GET_TGID_IN_PIDNS:
166 return (_IOC_TYPE(cmd) == _IOC_TYPE(cmd));
169 /* Extensible ioctls require some extra handling. */
170 switch (_IOC_NR(cmd)) {
171 case _IOC_NR(NS_MNT_GET_INFO):
172 case _IOC_NR(NS_MNT_GET_NEXT):
173 case _IOC_NR(NS_MNT_GET_PREV):
174 return (_IOC_TYPE(cmd) == _IOC_TYPE(cmd));
180 static long ns_ioctl(struct file *filp, unsigned int ioctl,
183 struct user_namespace *user_ns;
184 struct pid_namespace *pid_ns;
185 struct task_struct *tsk;
186 struct ns_common *ns;
187 struct mnt_namespace *mnt_ns;
188 bool previous = false;
193 if (!nsfs_ioctl_valid(ioctl))
196 ns = get_proc_ns(file_inode(filp));
199 return open_related_ns(ns, ns_get_owner);
201 if (!ns->ops->get_parent)
203 return open_related_ns(ns, ns->ops->get_parent);
205 return ns->ops->type;
206 case NS_GET_OWNER_UID:
207 if (ns->ops->type != CLONE_NEWUSER)
209 user_ns = container_of(ns, struct user_namespace, ns);
210 argp = (uid_t __user *) arg;
211 uid = from_kuid_munged(current_user_ns(), user_ns->owner);
212 return put_user(uid, argp);
213 case NS_GET_MNTNS_ID: {
217 if (ns->ops->type != CLONE_NEWNS)
220 mnt_ns = container_of(ns, struct mnt_namespace, ns);
221 idp = (__u64 __user *)arg;
223 return put_user(id, idp);
225 case NS_GET_PID_FROM_PIDNS:
227 case NS_GET_TGID_FROM_PIDNS:
229 case NS_GET_PID_IN_PIDNS:
231 case NS_GET_TGID_IN_PIDNS: {
232 if (ns->ops->type != CLONE_NEWPID)
236 pid_ns = container_of(ns, struct pid_namespace, ns);
240 if (ioctl == NS_GET_PID_IN_PIDNS ||
241 ioctl == NS_GET_TGID_IN_PIDNS)
242 tsk = find_task_by_vpid(arg);
244 tsk = find_task_by_pid_ns(arg, pid_ns);
249 case NS_GET_PID_FROM_PIDNS:
250 ret = task_pid_vnr(tsk);
252 case NS_GET_TGID_FROM_PIDNS:
253 ret = task_tgid_vnr(tsk);
255 case NS_GET_PID_IN_PIDNS:
256 ret = task_pid_nr_ns(tsk, pid_ns);
258 case NS_GET_TGID_IN_PIDNS:
259 ret = task_tgid_nr_ns(tsk, pid_ns);
272 /* extensible ioctls */
273 switch (_IOC_NR(ioctl)) {
274 case _IOC_NR(NS_MNT_GET_INFO): {
275 struct mnt_ns_info kinfo = {};
276 struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
277 size_t usize = _IOC_SIZE(ioctl);
279 if (ns->ops->type != CLONE_NEWNS)
285 if (usize < MNT_NS_INFO_SIZE_VER0)
288 return copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
290 case _IOC_NR(NS_MNT_GET_PREV):
293 case _IOC_NR(NS_MNT_GET_NEXT): {
294 struct mnt_ns_info kinfo = {};
295 struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
296 struct path path __free(path_put) = {};
297 struct file *f __free(fput) = NULL;
298 size_t usize = _IOC_SIZE(ioctl);
300 if (ns->ops->type != CLONE_NEWNS)
303 if (usize < MNT_NS_INFO_SIZE_VER0)
306 mnt_ns = get_sequential_mnt_ns(to_mnt_ns(ns), previous);
308 return PTR_ERR(mnt_ns);
310 ns = to_ns_common(mnt_ns);
311 /* Transfer ownership of @mnt_ns reference to @path. */
312 ret = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
316 CLASS(get_unused_fd, fd)(O_CLOEXEC);
320 f = dentry_open(&path, O_RDONLY, current_cred());
326 * If @uinfo is passed return all information about the
327 * mount namespace as well.
329 ret = copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
334 /* Transfer reference of @f to caller's fdtable. */
335 fd_install(fd, no_free_ptr(f));
336 /* File descriptor is live so hand it off to the caller. */
346 int ns_get_name(char *buf, size_t size, struct task_struct *task,
347 const struct proc_ns_operations *ns_ops)
349 struct ns_common *ns;
352 ns = ns_ops->get(task);
354 name = ns_ops->real_ns_name ? : ns_ops->name;
355 res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
361 bool proc_ns_file(const struct file *file)
363 return file->f_op == &ns_file_operations;
367 * ns_match() - Returns true if current namespace matches dev/ino provided.
368 * @ns: current namespace
369 * @dev: dev_t from nsfs that will be matched against current nsfs
370 * @ino: ino_t from nsfs that will be matched against current nsfs
372 * Return: true if dev and ino matches the current nsfs.
374 bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
376 return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
380 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
382 struct inode *inode = d_inode(dentry);
383 const struct ns_common *ns = inode->i_private;
384 const struct proc_ns_operations *ns_ops = ns->ops;
386 seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
390 static const struct super_operations nsfs_ops = {
391 .statfs = simple_statfs,
392 .evict_inode = nsfs_evict,
393 .show_path = nsfs_show_path,
396 static int nsfs_init_inode(struct inode *inode, void *data)
398 struct ns_common *ns = data;
400 inode->i_private = data;
401 inode->i_mode |= S_IRUGO;
402 inode->i_fop = &ns_file_operations;
403 inode->i_ino = ns->inum;
407 static void nsfs_put_data(void *data)
409 struct ns_common *ns = data;
413 static const struct stashed_operations nsfs_stashed_ops = {
414 .init_inode = nsfs_init_inode,
415 .put_data = nsfs_put_data,
418 static int nsfs_init_fs_context(struct fs_context *fc)
420 struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
423 ctx->ops = &nsfs_ops;
424 ctx->dops = &ns_dentry_operations;
425 fc->s_fs_info = (void *)&nsfs_stashed_ops;
429 static struct file_system_type nsfs = {
431 .init_fs_context = nsfs_init_fs_context,
432 .kill_sb = kill_anon_super,
435 void __init nsfs_init(void)
437 nsfs_mnt = kern_mount(&nsfs);
438 if (IS_ERR(nsfs_mnt))
439 panic("can't set nsfs up\n");
440 nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;