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 #include <linux/mnt_namespace.h>
16
17 #include "mount.h"
18 #include "internal.h"
19
20 static struct vfsmount *nsfs_mnt;
21
22 static long ns_ioctl(struct file *filp, unsigned int ioctl,
23 unsigned long arg);
24 static const struct file_operations ns_file_operations = {
25 .unlocked_ioctl = ns_ioctl,
26 .compat_ioctl = compat_ptr_ioctl,
27 };
28
ns_dname(struct dentry * dentry,char * buffer,int buflen)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_dname = ns_dname,
41 .d_prune = stashed_dentry_prune,
42 };
43
nsfs_evict(struct inode * inode)44 static void nsfs_evict(struct inode *inode)
45 {
46 struct ns_common *ns = inode->i_private;
47 clear_inode(inode);
48 ns->ops->put(ns);
49 }
50
ns_get_path_cb(struct path * path,ns_get_path_helper_t * ns_get_cb,void * private_data)51 int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
52 void *private_data)
53 {
54 struct ns_common *ns;
55
56 ns = ns_get_cb(private_data);
57 if (!ns)
58 return -ENOENT;
59
60 return path_from_stashed(&ns->stashed, nsfs_mnt, ns, path);
61 }
62
63 struct ns_get_path_task_args {
64 const struct proc_ns_operations *ns_ops;
65 struct task_struct *task;
66 };
67
ns_get_path_task(void * private_data)68 static struct ns_common *ns_get_path_task(void *private_data)
69 {
70 struct ns_get_path_task_args *args = private_data;
71
72 return args->ns_ops->get(args->task);
73 }
74
ns_get_path(struct path * path,struct task_struct * task,const struct proc_ns_operations * ns_ops)75 int ns_get_path(struct path *path, struct task_struct *task,
76 const struct proc_ns_operations *ns_ops)
77 {
78 struct ns_get_path_task_args args = {
79 .ns_ops = ns_ops,
80 .task = task,
81 };
82
83 return ns_get_path_cb(path, ns_get_path_task, &args);
84 }
85
86 /**
87 * open_namespace - open a namespace
88 * @ns: the namespace to open
89 *
90 * This will consume a reference to @ns indendent of success or failure.
91 *
92 * Return: A file descriptor on success or a negative error code on failure.
93 */
open_namespace(struct ns_common * ns)94 int open_namespace(struct ns_common *ns)
95 {
96 struct path path __free(path_put) = {};
97 struct file *f;
98 int err;
99
100 /* call first to consume reference */
101 err = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
102 if (err < 0)
103 return err;
104
105 CLASS(get_unused_fd, fd)(O_CLOEXEC);
106 if (fd < 0)
107 return fd;
108
109 f = dentry_open(&path, O_RDONLY, current_cred());
110 if (IS_ERR(f))
111 return PTR_ERR(f);
112
113 fd_install(fd, f);
114 return take_fd(fd);
115 }
116
open_related_ns(struct ns_common * ns,struct ns_common * (* get_ns)(struct ns_common * ns))117 int open_related_ns(struct ns_common *ns,
118 struct ns_common *(*get_ns)(struct ns_common *ns))
119 {
120 struct ns_common *relative;
121
122 relative = get_ns(ns);
123 if (IS_ERR(relative))
124 return PTR_ERR(relative);
125
126 return open_namespace(relative);
127 }
128 EXPORT_SYMBOL_GPL(open_related_ns);
129
copy_ns_info_to_user(const struct mnt_namespace * mnt_ns,struct mnt_ns_info __user * uinfo,size_t usize,struct mnt_ns_info * kinfo)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)
133 {
134 /*
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.
140 */
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)
146 kinfo->nr_mounts--;
147
148 if (copy_to_user(uinfo, kinfo, kinfo->size))
149 return -EFAULT;
150
151 return 0;
152 }
153
ns_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)154 static long ns_ioctl(struct file *filp, unsigned int ioctl,
155 unsigned long arg)
156 {
157 struct user_namespace *user_ns;
158 struct pid_namespace *pid_ns;
159 struct task_struct *tsk;
160 struct ns_common *ns = get_proc_ns(file_inode(filp));
161 struct mnt_namespace *mnt_ns;
162 bool previous = false;
163 uid_t __user *argp;
164 uid_t uid;
165 int ret;
166
167 switch (ioctl) {
168 case NS_GET_USERNS:
169 return open_related_ns(ns, ns_get_owner);
170 case NS_GET_PARENT:
171 if (!ns->ops->get_parent)
172 return -EINVAL;
173 return open_related_ns(ns, ns->ops->get_parent);
174 case NS_GET_NSTYPE:
175 return ns->ops->type;
176 case NS_GET_OWNER_UID:
177 if (ns->ops->type != CLONE_NEWUSER)
178 return -EINVAL;
179 user_ns = container_of(ns, struct user_namespace, ns);
180 argp = (uid_t __user *) arg;
181 uid = from_kuid_munged(current_user_ns(), user_ns->owner);
182 return put_user(uid, argp);
183 case NS_GET_MNTNS_ID: {
184 __u64 __user *idp;
185 __u64 id;
186
187 if (ns->ops->type != CLONE_NEWNS)
188 return -EINVAL;
189
190 mnt_ns = container_of(ns, struct mnt_namespace, ns);
191 idp = (__u64 __user *)arg;
192 id = mnt_ns->seq;
193 return put_user(id, idp);
194 }
195 case NS_GET_PID_FROM_PIDNS:
196 fallthrough;
197 case NS_GET_TGID_FROM_PIDNS:
198 fallthrough;
199 case NS_GET_PID_IN_PIDNS:
200 fallthrough;
201 case NS_GET_TGID_IN_PIDNS: {
202 if (ns->ops->type != CLONE_NEWPID)
203 return -EINVAL;
204
205 ret = -ESRCH;
206 pid_ns = container_of(ns, struct pid_namespace, ns);
207
208 guard(rcu)();
209
210 if (ioctl == NS_GET_PID_IN_PIDNS ||
211 ioctl == NS_GET_TGID_IN_PIDNS)
212 tsk = find_task_by_vpid(arg);
213 else
214 tsk = find_task_by_pid_ns(arg, pid_ns);
215 if (!tsk)
216 break;
217
218 switch (ioctl) {
219 case NS_GET_PID_FROM_PIDNS:
220 ret = task_pid_vnr(tsk);
221 break;
222 case NS_GET_TGID_FROM_PIDNS:
223 ret = task_tgid_vnr(tsk);
224 break;
225 case NS_GET_PID_IN_PIDNS:
226 ret = task_pid_nr_ns(tsk, pid_ns);
227 break;
228 case NS_GET_TGID_IN_PIDNS:
229 ret = task_tgid_nr_ns(tsk, pid_ns);
230 break;
231 default:
232 ret = 0;
233 break;
234 }
235
236 if (!ret)
237 ret = -ESRCH;
238 return ret;
239 }
240 }
241
242 /* extensible ioctls */
243 switch (_IOC_NR(ioctl)) {
244 case _IOC_NR(NS_MNT_GET_INFO): {
245 struct mnt_ns_info kinfo = {};
246 struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
247 size_t usize = _IOC_SIZE(ioctl);
248
249 if (ns->ops->type != CLONE_NEWNS)
250 return -EINVAL;
251
252 if (!uinfo)
253 return -EINVAL;
254
255 if (usize < MNT_NS_INFO_SIZE_VER0)
256 return -EINVAL;
257
258 return copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
259 }
260 case _IOC_NR(NS_MNT_GET_PREV):
261 previous = true;
262 fallthrough;
263 case _IOC_NR(NS_MNT_GET_NEXT): {
264 struct mnt_ns_info kinfo = {};
265 struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
266 struct path path __free(path_put) = {};
267 struct file *f __free(fput) = NULL;
268 size_t usize = _IOC_SIZE(ioctl);
269
270 if (ns->ops->type != CLONE_NEWNS)
271 return -EINVAL;
272
273 if (usize < MNT_NS_INFO_SIZE_VER0)
274 return -EINVAL;
275
276 mnt_ns = get_sequential_mnt_ns(to_mnt_ns(ns), previous);
277 if (IS_ERR(mnt_ns))
278 return PTR_ERR(mnt_ns);
279
280 ns = to_ns_common(mnt_ns);
281 /* Transfer ownership of @mnt_ns reference to @path. */
282 ret = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
283 if (ret)
284 return ret;
285
286 CLASS(get_unused_fd, fd)(O_CLOEXEC);
287 if (fd < 0)
288 return fd;
289
290 f = dentry_open(&path, O_RDONLY, current_cred());
291 if (IS_ERR(f))
292 return PTR_ERR(f);
293
294 if (uinfo) {
295 /*
296 * If @uinfo is passed return all information about the
297 * mount namespace as well.
298 */
299 ret = copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
300 if (ret)
301 return ret;
302 }
303
304 /* Transfer reference of @f to caller's fdtable. */
305 fd_install(fd, no_free_ptr(f));
306 /* File descriptor is live so hand it off to the caller. */
307 return take_fd(fd);
308 }
309 default:
310 ret = -ENOTTY;
311 }
312
313 return ret;
314 }
315
ns_get_name(char * buf,size_t size,struct task_struct * task,const struct proc_ns_operations * ns_ops)316 int ns_get_name(char *buf, size_t size, struct task_struct *task,
317 const struct proc_ns_operations *ns_ops)
318 {
319 struct ns_common *ns;
320 int res = -ENOENT;
321 const char *name;
322 ns = ns_ops->get(task);
323 if (ns) {
324 name = ns_ops->real_ns_name ? : ns_ops->name;
325 res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
326 ns_ops->put(ns);
327 }
328 return res;
329 }
330
proc_ns_file(const struct file * file)331 bool proc_ns_file(const struct file *file)
332 {
333 return file->f_op == &ns_file_operations;
334 }
335
336 /**
337 * ns_match() - Returns true if current namespace matches dev/ino provided.
338 * @ns: current namespace
339 * @dev: dev_t from nsfs that will be matched against current nsfs
340 * @ino: ino_t from nsfs that will be matched against current nsfs
341 *
342 * Return: true if dev and ino matches the current nsfs.
343 */
ns_match(const struct ns_common * ns,dev_t dev,ino_t ino)344 bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
345 {
346 return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
347 }
348
349
nsfs_show_path(struct seq_file * seq,struct dentry * dentry)350 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
351 {
352 struct inode *inode = d_inode(dentry);
353 const struct ns_common *ns = inode->i_private;
354 const struct proc_ns_operations *ns_ops = ns->ops;
355
356 seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
357 return 0;
358 }
359
360 static const struct super_operations nsfs_ops = {
361 .statfs = simple_statfs,
362 .evict_inode = nsfs_evict,
363 .show_path = nsfs_show_path,
364 };
365
nsfs_init_inode(struct inode * inode,void * data)366 static int nsfs_init_inode(struct inode *inode, void *data)
367 {
368 struct ns_common *ns = data;
369
370 inode->i_private = data;
371 inode->i_mode |= S_IRUGO;
372 inode->i_fop = &ns_file_operations;
373 inode->i_ino = ns->inum;
374 return 0;
375 }
376
nsfs_put_data(void * data)377 static void nsfs_put_data(void *data)
378 {
379 struct ns_common *ns = data;
380 ns->ops->put(ns);
381 }
382
383 static const struct stashed_operations nsfs_stashed_ops = {
384 .init_inode = nsfs_init_inode,
385 .put_data = nsfs_put_data,
386 };
387
nsfs_init_fs_context(struct fs_context * fc)388 static int nsfs_init_fs_context(struct fs_context *fc)
389 {
390 struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
391 if (!ctx)
392 return -ENOMEM;
393 ctx->ops = &nsfs_ops;
394 ctx->dops = &ns_dentry_operations;
395 fc->s_fs_info = (void *)&nsfs_stashed_ops;
396 return 0;
397 }
398
399 static struct file_system_type nsfs = {
400 .name = "nsfs",
401 .init_fs_context = nsfs_init_fs_context,
402 .kill_sb = kill_anon_super,
403 };
404
nsfs_init(void)405 void __init nsfs_init(void)
406 {
407 nsfs_mnt = kern_mount(&nsfs);
408 if (IS_ERR(nsfs_mnt))
409 panic("can't set nsfs up\n");
410 nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
411 }
412