xref: /linux/fs/nsfs.c (revision 0b3bb205808195159be633a8cefb602670e856fb)
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 #include <linux/ipc_namespace.h>
17 #include <linux/time_namespace.h>
18 #include <linux/utsname.h>
19 #include <linux/exportfs.h>
20 #include <linux/nstree.h>
21 #include <net/net_namespace.h>
22 
23 #include "mount.h"
24 #include "internal.h"
25 
26 static struct vfsmount *nsfs_mnt;
27 
28 static struct path nsfs_root_path = {};
29 
nsfs_get_root(struct path * path)30 void nsfs_get_root(struct path *path)
31 {
32 	*path = nsfs_root_path;
33 	path_get(path);
34 }
35 
36 static long ns_ioctl(struct file *filp, unsigned int ioctl,
37 			unsigned long arg);
38 static const struct file_operations ns_file_operations = {
39 	.unlocked_ioctl = ns_ioctl,
40 	.compat_ioctl   = compat_ptr_ioctl,
41 };
42 
ns_dname(struct dentry * dentry,char * buffer,int buflen)43 static char *ns_dname(struct dentry *dentry, char *buffer, int buflen)
44 {
45 	struct inode *inode = d_inode(dentry);
46 	struct ns_common *ns = inode->i_private;
47 	const struct proc_ns_operations *ns_ops = ns->ops;
48 
49 	return dynamic_dname(buffer, buflen, "%s:[%lu]",
50 		ns_ops->name, inode->i_ino);
51 }
52 
53 const struct dentry_operations ns_dentry_operations = {
54 	.d_dname	= ns_dname,
55 	.d_prune	= stashed_dentry_prune,
56 };
57 
nsfs_evict(struct inode * inode)58 static void nsfs_evict(struct inode *inode)
59 {
60 	struct ns_common *ns = inode->i_private;
61 
62 	__ns_ref_active_put(ns);
63 	clear_inode(inode);
64 	ns->ops->put(ns);
65 }
66 
ns_get_path_cb(struct path * path,ns_get_path_helper_t * ns_get_cb,void * private_data)67 int ns_get_path_cb(struct path *path, ns_get_path_helper_t *ns_get_cb,
68 		     void *private_data)
69 {
70 	struct ns_common *ns;
71 
72 	ns = ns_get_cb(private_data);
73 	if (!ns)
74 		return -ENOENT;
75 
76 	return path_from_stashed(&ns->stashed, nsfs_mnt, ns, path);
77 }
78 
79 struct ns_get_path_task_args {
80 	const struct proc_ns_operations *ns_ops;
81 	struct task_struct *task;
82 };
83 
ns_get_path_task(void * private_data)84 static struct ns_common *ns_get_path_task(void *private_data)
85 {
86 	struct ns_get_path_task_args *args = private_data;
87 
88 	return args->ns_ops->get(args->task);
89 }
90 
ns_get_path(struct path * path,struct task_struct * task,const struct proc_ns_operations * ns_ops)91 int ns_get_path(struct path *path, struct task_struct *task,
92 		  const struct proc_ns_operations *ns_ops)
93 {
94 	struct ns_get_path_task_args args = {
95 		.ns_ops	= ns_ops,
96 		.task	= task,
97 	};
98 
99 	return ns_get_path_cb(path, ns_get_path_task, &args);
100 }
101 
open_namespace_file(struct ns_common * ns)102 struct file *open_namespace_file(struct ns_common *ns)
103 {
104 	struct path path __free(path_put) = {};
105 	int err;
106 
107 	/* call first to consume reference */
108 	err = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
109 	if (err < 0)
110 		return ERR_PTR(err);
111 
112 	return dentry_open(&path, O_RDONLY, current_cred());
113 }
114 
115 /**
116  * open_namespace - open a namespace
117  * @ns: the namespace to open
118  *
119  * This will consume a reference to @ns indendent of success or failure.
120  *
121  * Return: A file descriptor on success or a negative error code on failure.
122  */
open_namespace(struct ns_common * ns)123 int open_namespace(struct ns_common *ns)
124 {
125 	struct path path __free(path_put) = {};
126 	int err;
127 
128 	/* call first to consume reference */
129 	err = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
130 	if (err < 0)
131 		return err;
132 
133 	return FD_ADD(O_CLOEXEC, dentry_open(&path, O_RDONLY, current_cred()));
134 }
135 
open_related_ns(struct ns_common * ns,struct ns_common * (* get_ns)(struct ns_common * ns))136 int open_related_ns(struct ns_common *ns,
137 		   struct ns_common *(*get_ns)(struct ns_common *ns))
138 {
139 	struct ns_common *relative;
140 
141 	relative = get_ns(ns);
142 	if (IS_ERR(relative))
143 		return PTR_ERR(relative);
144 
145 	return open_namespace(relative);
146 }
147 EXPORT_SYMBOL_GPL(open_related_ns);
148 
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)149 static int copy_ns_info_to_user(const struct mnt_namespace *mnt_ns,
150 				struct mnt_ns_info __user *uinfo, size_t usize,
151 				struct mnt_ns_info *kinfo)
152 {
153 	/*
154 	 * If userspace and the kernel have the same struct size it can just
155 	 * be copied. If userspace provides an older struct, only the bits that
156 	 * userspace knows about will be copied. If userspace provides a new
157 	 * struct, only the bits that the kernel knows aobut will be copied and
158 	 * the size value will be set to the size the kernel knows about.
159 	 */
160 	kinfo->size		= min(usize, sizeof(*kinfo));
161 	kinfo->mnt_ns_id	= mnt_ns->ns.ns_id;
162 	kinfo->nr_mounts	= READ_ONCE(mnt_ns->nr_mounts);
163 	/* Subtract the root mount of the mount namespace. */
164 	if (kinfo->nr_mounts)
165 		kinfo->nr_mounts--;
166 
167 	if (copy_to_user(uinfo, kinfo, kinfo->size))
168 		return -EFAULT;
169 
170 	return 0;
171 }
172 
nsfs_ioctl_valid(unsigned int cmd)173 static bool nsfs_ioctl_valid(unsigned int cmd)
174 {
175 	switch (cmd) {
176 	case NS_GET_USERNS:
177 	case NS_GET_PARENT:
178 	case NS_GET_NSTYPE:
179 	case NS_GET_OWNER_UID:
180 	case NS_GET_MNTNS_ID:
181 	case NS_GET_PID_FROM_PIDNS:
182 	case NS_GET_TGID_FROM_PIDNS:
183 	case NS_GET_PID_IN_PIDNS:
184 	case NS_GET_TGID_IN_PIDNS:
185 	case NS_GET_ID:
186 		return true;
187 	}
188 
189 	/* Extensible ioctls require some extra handling. */
190 	switch (_IOC_NR(cmd)) {
191 	case _IOC_NR(NS_MNT_GET_INFO):
192 		return extensible_ioctl_valid(cmd, NS_MNT_GET_INFO, MNT_NS_INFO_SIZE_VER0);
193 	case _IOC_NR(NS_MNT_GET_NEXT):
194 		return extensible_ioctl_valid(cmd, NS_MNT_GET_NEXT, MNT_NS_INFO_SIZE_VER0);
195 	case _IOC_NR(NS_MNT_GET_PREV):
196 		return extensible_ioctl_valid(cmd, NS_MNT_GET_PREV, MNT_NS_INFO_SIZE_VER0);
197 	}
198 
199 	return false;
200 }
201 
may_use_nsfs_ioctl(unsigned int cmd)202 static bool may_use_nsfs_ioctl(unsigned int cmd)
203 {
204 	switch (_IOC_NR(cmd)) {
205 	case _IOC_NR(NS_MNT_GET_NEXT):
206 		fallthrough;
207 	case _IOC_NR(NS_MNT_GET_PREV):
208 		return may_see_all_namespaces();
209 	}
210 	return true;
211 }
212 
ns_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)213 static long ns_ioctl(struct file *filp, unsigned int ioctl,
214 			unsigned long arg)
215 {
216 	struct user_namespace *user_ns;
217 	struct pid_namespace *pid_ns;
218 	struct task_struct *tsk;
219 	struct ns_common *ns;
220 	struct mnt_namespace *mnt_ns;
221 	bool previous = false;
222 	uid_t __user *argp;
223 	uid_t uid;
224 	int ret;
225 
226 	if (!nsfs_ioctl_valid(ioctl))
227 		return -ENOIOCTLCMD;
228 	if (!may_use_nsfs_ioctl(ioctl))
229 		return -EPERM;
230 
231 	ns = get_proc_ns(file_inode(filp));
232 	switch (ioctl) {
233 	case NS_GET_USERNS:
234 		return open_related_ns(ns, ns_get_owner);
235 	case NS_GET_PARENT:
236 		if (!ns->ops->get_parent)
237 			return -EINVAL;
238 		return open_related_ns(ns, ns->ops->get_parent);
239 	case NS_GET_NSTYPE:
240 		return ns->ns_type;
241 	case NS_GET_OWNER_UID:
242 		if (ns->ns_type != CLONE_NEWUSER)
243 			return -EINVAL;
244 		user_ns = container_of(ns, struct user_namespace, ns);
245 		argp = (uid_t __user *) arg;
246 		uid = from_kuid_munged(current_user_ns(), user_ns->owner);
247 		return put_user(uid, argp);
248 	case NS_GET_PID_FROM_PIDNS:
249 		fallthrough;
250 	case NS_GET_TGID_FROM_PIDNS:
251 		fallthrough;
252 	case NS_GET_PID_IN_PIDNS:
253 		fallthrough;
254 	case NS_GET_TGID_IN_PIDNS: {
255 		if (ns->ns_type != CLONE_NEWPID)
256 			return -EINVAL;
257 
258 		ret = -ESRCH;
259 		pid_ns = container_of(ns, struct pid_namespace, ns);
260 
261 		guard(rcu)();
262 
263 		if (ioctl == NS_GET_PID_IN_PIDNS ||
264 		    ioctl == NS_GET_TGID_IN_PIDNS)
265 			tsk = find_task_by_vpid(arg);
266 		else
267 			tsk = find_task_by_pid_ns(arg, pid_ns);
268 		if (!tsk)
269 			break;
270 
271 		switch (ioctl) {
272 		case NS_GET_PID_FROM_PIDNS:
273 			ret = task_pid_vnr(tsk);
274 			break;
275 		case NS_GET_TGID_FROM_PIDNS:
276 			ret = task_tgid_vnr(tsk);
277 			break;
278 		case NS_GET_PID_IN_PIDNS:
279 			ret = task_pid_nr_ns(tsk, pid_ns);
280 			break;
281 		case NS_GET_TGID_IN_PIDNS:
282 			ret = task_tgid_nr_ns(tsk, pid_ns);
283 			break;
284 		default:
285 			ret = 0;
286 			break;
287 		}
288 
289 		if (!ret)
290 			ret = -ESRCH;
291 		return ret;
292 	}
293 	case NS_GET_MNTNS_ID:
294 		if (ns->ns_type != CLONE_NEWNS)
295 			return -EINVAL;
296 		fallthrough;
297 	case NS_GET_ID: {
298 		__u64 __user *idp;
299 		__u64 id;
300 
301 		idp = (__u64 __user *)arg;
302 		id = ns->ns_id;
303 		return put_user(id, idp);
304 	}
305 	}
306 
307 	/* extensible ioctls */
308 	switch (_IOC_NR(ioctl)) {
309 	case _IOC_NR(NS_MNT_GET_INFO): {
310 		struct mnt_ns_info kinfo = {};
311 		struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
312 		size_t usize = _IOC_SIZE(ioctl);
313 
314 		if (ns->ns_type != CLONE_NEWNS)
315 			return -EINVAL;
316 
317 		if (!uinfo)
318 			return -EINVAL;
319 
320 		if (usize < MNT_NS_INFO_SIZE_VER0)
321 			return -EINVAL;
322 
323 		return copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
324 	}
325 	case _IOC_NR(NS_MNT_GET_PREV):
326 		previous = true;
327 		fallthrough;
328 	case _IOC_NR(NS_MNT_GET_NEXT): {
329 		struct mnt_ns_info kinfo = {};
330 		struct mnt_ns_info __user *uinfo = (struct mnt_ns_info __user *)arg;
331 		struct path path __free(path_put) = {};
332 		size_t usize = _IOC_SIZE(ioctl);
333 
334 		if (ns->ns_type != CLONE_NEWNS)
335 			return -EINVAL;
336 
337 		if (usize < MNT_NS_INFO_SIZE_VER0)
338 			return -EINVAL;
339 
340 		mnt_ns = get_sequential_mnt_ns(to_mnt_ns(ns), previous);
341 		if (IS_ERR(mnt_ns))
342 			return PTR_ERR(mnt_ns);
343 
344 		ns = to_ns_common(mnt_ns);
345 		/* Transfer ownership of @mnt_ns reference to @path. */
346 		ret = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
347 		if (ret)
348 			return ret;
349 
350 		FD_PREPARE(fdf, O_CLOEXEC, dentry_open(&path, O_RDONLY, current_cred()));
351 		if (fdf.err)
352 			return fdf.err;
353 		/*
354 		 * If @uinfo is passed return all information about the
355 		 * mount namespace as well.
356 		 */
357 		ret = copy_ns_info_to_user(to_mnt_ns(ns), uinfo, usize, &kinfo);
358 		if (ret)
359 			return ret;
360 		ret = fd_publish(fdf);
361 		break;
362 	}
363 	default:
364 		ret = -ENOTTY;
365 	}
366 
367 	return ret;
368 }
369 
ns_get_name(char * buf,size_t size,struct task_struct * task,const struct proc_ns_operations * ns_ops)370 int ns_get_name(char *buf, size_t size, struct task_struct *task,
371 			const struct proc_ns_operations *ns_ops)
372 {
373 	struct ns_common *ns;
374 	int res = -ENOENT;
375 	const char *name;
376 	ns = ns_ops->get(task);
377 	if (ns) {
378 		name = ns_ops->real_ns_name ? : ns_ops->name;
379 		res = snprintf(buf, size, "%s:[%u]", name, ns->inum);
380 		ns_ops->put(ns);
381 	}
382 	return res;
383 }
384 
proc_ns_file(const struct file * file)385 bool proc_ns_file(const struct file *file)
386 {
387 	return file->f_op == &ns_file_operations;
388 }
389 
390 /**
391  * ns_match() - Returns true if current namespace matches dev/ino provided.
392  * @ns: current namespace
393  * @dev: dev_t from nsfs that will be matched against current nsfs
394  * @ino: ino_t from nsfs that will be matched against current nsfs
395  *
396  * Return: true if dev and ino matches the current nsfs.
397  */
ns_match(const struct ns_common * ns,dev_t dev,ino_t ino)398 bool ns_match(const struct ns_common *ns, dev_t dev, ino_t ino)
399 {
400 	return (ns->inum == ino) && (nsfs_mnt->mnt_sb->s_dev == dev);
401 }
402 
403 
nsfs_show_path(struct seq_file * seq,struct dentry * dentry)404 static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry)
405 {
406 	struct inode *inode = d_inode(dentry);
407 	const struct ns_common *ns = inode->i_private;
408 	const struct proc_ns_operations *ns_ops = ns->ops;
409 
410 	seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino);
411 	return 0;
412 }
413 
414 static const struct super_operations nsfs_ops = {
415 	.statfs = simple_statfs,
416 	.evict_inode = nsfs_evict,
417 	.show_path = nsfs_show_path,
418 	.drop_inode = inode_just_drop,
419 };
420 
nsfs_init_inode(struct inode * inode,void * data)421 static int nsfs_init_inode(struct inode *inode, void *data)
422 {
423 	struct ns_common *ns = data;
424 
425 	inode->i_private = data;
426 	inode->i_mode |= S_IRUGO;
427 	inode->i_fop = &ns_file_operations;
428 	inode->i_ino = ns->inum;
429 
430 	/*
431 	 * Bring the namespace subtree back to life if we have to. This
432 	 * can happen when e.g., all processes using a network namespace
433 	 * and all namespace files or namespace file bind-mounts have
434 	 * died but there are still sockets pinning it. The SIOCGSKNS
435 	 * ioctl on such a socket will resurrect the relevant namespace
436 	 * subtree.
437 	 */
438 	__ns_ref_active_get(ns);
439 	return 0;
440 }
441 
nsfs_put_data(void * data)442 static void nsfs_put_data(void *data)
443 {
444 	struct ns_common *ns = data;
445 	ns->ops->put(ns);
446 }
447 
448 static const struct stashed_operations nsfs_stashed_ops = {
449 	.init_inode = nsfs_init_inode,
450 	.put_data = nsfs_put_data,
451 };
452 
453 #define NSFS_FID_SIZE_U32_VER0 (NSFS_FILE_HANDLE_SIZE_VER0 / sizeof(u32))
454 #define NSFS_FID_SIZE_U32_LATEST (NSFS_FILE_HANDLE_SIZE_LATEST / sizeof(u32))
455 
nsfs_encode_fh(struct inode * inode,u32 * fh,int * max_len,struct inode * parent)456 static int nsfs_encode_fh(struct inode *inode, u32 *fh, int *max_len,
457 			  struct inode *parent)
458 {
459 	struct nsfs_file_handle *fid = (struct nsfs_file_handle *)fh;
460 	struct ns_common *ns = inode->i_private;
461 	int len = *max_len;
462 
463 	if (parent)
464 		return FILEID_INVALID;
465 
466 	if (len < NSFS_FID_SIZE_U32_VER0) {
467 		*max_len = NSFS_FID_SIZE_U32_LATEST;
468 		return FILEID_INVALID;
469 	} else if (len > NSFS_FID_SIZE_U32_LATEST) {
470 		*max_len = NSFS_FID_SIZE_U32_LATEST;
471 	}
472 
473 	fid->ns_id	= ns->ns_id;
474 	fid->ns_type	= ns->ns_type;
475 	fid->ns_inum	= inode->i_ino;
476 	return FILEID_NSFS;
477 }
478 
is_current_namespace(struct ns_common * ns)479 bool is_current_namespace(struct ns_common *ns)
480 {
481 	switch (ns->ns_type) {
482 #ifdef CONFIG_CGROUPS
483 	case CLONE_NEWCGROUP:
484 		return current_in_namespace(to_cg_ns(ns));
485 #endif
486 #ifdef CONFIG_IPC_NS
487 	case CLONE_NEWIPC:
488 		return current_in_namespace(to_ipc_ns(ns));
489 #endif
490 	case CLONE_NEWNS:
491 		return current_in_namespace(to_mnt_ns(ns));
492 #ifdef CONFIG_NET_NS
493 	case CLONE_NEWNET:
494 		return current_in_namespace(to_net_ns(ns));
495 #endif
496 #ifdef CONFIG_PID_NS
497 	case CLONE_NEWPID:
498 		return current_in_namespace(to_pid_ns(ns));
499 #endif
500 #ifdef CONFIG_TIME_NS
501 	case CLONE_NEWTIME:
502 		return current_in_namespace(to_time_ns(ns));
503 #endif
504 #ifdef CONFIG_USER_NS
505 	case CLONE_NEWUSER:
506 		return current_in_namespace(to_user_ns(ns));
507 #endif
508 #ifdef CONFIG_UTS_NS
509 	case CLONE_NEWUTS:
510 		return current_in_namespace(to_uts_ns(ns));
511 #endif
512 	default:
513 		VFS_WARN_ON_ONCE(true);
514 		return false;
515 	}
516 }
517 
nsfs_fh_to_dentry(struct super_block * sb,struct fid * fh,int fh_len,int fh_type)518 static struct dentry *nsfs_fh_to_dentry(struct super_block *sb, struct fid *fh,
519 					int fh_len, int fh_type)
520 {
521 	struct path path __free(path_put) = {};
522 	struct nsfs_file_handle *fid = (struct nsfs_file_handle *)fh;
523 	struct user_namespace *owning_ns = NULL;
524 	struct ns_common *ns;
525 	int ret;
526 
527 	if (fh_len < NSFS_FID_SIZE_U32_VER0)
528 		return NULL;
529 
530 	/* Check that any trailing bytes are zero. */
531 	if ((fh_len > NSFS_FID_SIZE_U32_LATEST) &&
532 	    memchr_inv((void *)fid + NSFS_FID_SIZE_U32_LATEST, 0,
533 		       fh_len - NSFS_FID_SIZE_U32_LATEST))
534 		return NULL;
535 
536 	switch (fh_type) {
537 	case FILEID_NSFS:
538 		break;
539 	default:
540 		return NULL;
541 	}
542 
543 	if (!fid->ns_id)
544 		return NULL;
545 	/* Either both are set or both are unset. */
546 	if (!fid->ns_inum != !fid->ns_type)
547 		return NULL;
548 
549 	scoped_guard(rcu) {
550 		ns = ns_tree_lookup_rcu(fid->ns_id, fid->ns_type);
551 		if (!ns)
552 			return NULL;
553 
554 		VFS_WARN_ON_ONCE(ns->ns_id != fid->ns_id);
555 
556 		if (fid->ns_inum && (fid->ns_inum != ns->inum))
557 			return NULL;
558 		if (fid->ns_type && (fid->ns_type != ns->ns_type))
559 			return NULL;
560 
561 		/*
562 		 * This is racy because we're not actually taking an
563 		 * active reference. IOW, it could happen that the
564 		 * namespace becomes inactive after this check.
565 		 * We don't care because nsfs_init_inode() will just
566 		 * resurrect the relevant namespace tree for us. If it
567 		 * has been active here we just allow it's resurrection.
568 		 * We could try to take an active reference here and
569 		 * then drop it again. But really, why bother.
570 		 */
571 		if (!ns_get_unless_inactive(ns))
572 			return NULL;
573 	}
574 
575 	switch (ns->ns_type) {
576 #ifdef CONFIG_CGROUPS
577 	case CLONE_NEWCGROUP:
578 		if (!current_in_namespace(to_cg_ns(ns)))
579 			owning_ns = to_cg_ns(ns)->user_ns;
580 		break;
581 #endif
582 #ifdef CONFIG_IPC_NS
583 	case CLONE_NEWIPC:
584 		if (!current_in_namespace(to_ipc_ns(ns)))
585 			owning_ns = to_ipc_ns(ns)->user_ns;
586 		break;
587 #endif
588 	case CLONE_NEWNS:
589 		if (!current_in_namespace(to_mnt_ns(ns)))
590 			owning_ns = to_mnt_ns(ns)->user_ns;
591 		break;
592 #ifdef CONFIG_NET_NS
593 	case CLONE_NEWNET:
594 		if (!current_in_namespace(to_net_ns(ns)))
595 			owning_ns = to_net_ns(ns)->user_ns;
596 		break;
597 #endif
598 #ifdef CONFIG_PID_NS
599 	case CLONE_NEWPID:
600 		if (!current_in_namespace(to_pid_ns(ns))) {
601 			owning_ns = to_pid_ns(ns)->user_ns;
602 		} else if (!READ_ONCE(to_pid_ns(ns)->child_reaper)) {
603 			ns->ops->put(ns);
604 			return ERR_PTR(-EPERM);
605 		}
606 		break;
607 #endif
608 #ifdef CONFIG_TIME_NS
609 	case CLONE_NEWTIME:
610 		if (!current_in_namespace(to_time_ns(ns)))
611 			owning_ns = to_time_ns(ns)->user_ns;
612 		break;
613 #endif
614 #ifdef CONFIG_USER_NS
615 	case CLONE_NEWUSER:
616 		if (!current_in_namespace(to_user_ns(ns)))
617 			owning_ns = to_user_ns(ns);
618 		break;
619 #endif
620 #ifdef CONFIG_UTS_NS
621 	case CLONE_NEWUTS:
622 		if (!current_in_namespace(to_uts_ns(ns)))
623 			owning_ns = to_uts_ns(ns)->user_ns;
624 		break;
625 #endif
626 	default:
627 		return ERR_PTR(-EOPNOTSUPP);
628 	}
629 
630 	if (owning_ns && !may_see_all_namespaces()) {
631 		ns->ops->put(ns);
632 		return ERR_PTR(-EPERM);
633 	}
634 
635 	/* path_from_stashed() unconditionally consumes the reference. */
636 	ret = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
637 	if (ret)
638 		return ERR_PTR(ret);
639 
640 	return no_free_ptr(path.dentry);
641 }
642 
nsfs_export_permission(struct handle_to_path_ctx * ctx,unsigned int oflags)643 static int nsfs_export_permission(struct handle_to_path_ctx *ctx,
644 				   unsigned int oflags)
645 {
646 	/* nsfs_fh_to_dentry() performs all permission checks. */
647 	return 0;
648 }
649 
nsfs_export_open(const struct path * path,unsigned int oflags)650 static struct file *nsfs_export_open(const struct path *path, unsigned int oflags)
651 {
652 	return file_open_root(path, "", oflags, 0);
653 }
654 
655 static const struct export_operations nsfs_export_operations = {
656 	.encode_fh	= nsfs_encode_fh,
657 	.fh_to_dentry	= nsfs_fh_to_dentry,
658 	.open		= nsfs_export_open,
659 	.permission	= nsfs_export_permission,
660 };
661 
nsfs_init_fs_context(struct fs_context * fc)662 static int nsfs_init_fs_context(struct fs_context *fc)
663 {
664 	struct pseudo_fs_context *ctx = init_pseudo(fc, NSFS_MAGIC);
665 	if (!ctx)
666 		return -ENOMEM;
667 	fc->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
668 	ctx->s_d_flags |= DCACHE_DONTCACHE;
669 	ctx->ops = &nsfs_ops;
670 	ctx->eops = &nsfs_export_operations;
671 	ctx->dops = &ns_dentry_operations;
672 	fc->s_fs_info = (void *)&nsfs_stashed_ops;
673 	return 0;
674 }
675 
676 static struct file_system_type nsfs = {
677 	.name = "nsfs",
678 	.init_fs_context = nsfs_init_fs_context,
679 	.kill_sb = kill_anon_super,
680 };
681 
nsfs_init(void)682 void __init nsfs_init(void)
683 {
684 	nsfs_mnt = kern_mount(&nsfs);
685 	if (IS_ERR(nsfs_mnt))
686 		panic("can't set nsfs up\n");
687 	nsfs_mnt->mnt_sb->s_flags &= ~SB_NOUSER;
688 	nsfs_root_path.mnt = nsfs_mnt;
689 	nsfs_root_path.dentry = nsfs_mnt->mnt_root;
690 }
691 
nsproxy_ns_active_get(struct nsproxy * ns)692 void nsproxy_ns_active_get(struct nsproxy *ns)
693 {
694 	ns_ref_active_get(ns->mnt_ns);
695 	ns_ref_active_get(ns->uts_ns);
696 	ns_ref_active_get(ns->ipc_ns);
697 	ns_ref_active_get(ns->pid_ns_for_children);
698 	ns_ref_active_get(ns->cgroup_ns);
699 	ns_ref_active_get(ns->net_ns);
700 	ns_ref_active_get(ns->time_ns);
701 	ns_ref_active_get(ns->time_ns_for_children);
702 }
703 
nsproxy_ns_active_put(struct nsproxy * ns)704 void nsproxy_ns_active_put(struct nsproxy *ns)
705 {
706 	ns_ref_active_put(ns->mnt_ns);
707 	ns_ref_active_put(ns->uts_ns);
708 	ns_ref_active_put(ns->ipc_ns);
709 	ns_ref_active_put(ns->pid_ns_for_children);
710 	ns_ref_active_put(ns->cgroup_ns);
711 	ns_ref_active_put(ns->net_ns);
712 	ns_ref_active_put(ns->time_ns);
713 	ns_ref_active_put(ns->time_ns_for_children);
714 }
715