xref: /linux/kernel/time/namespace.c (revision 5dc9cf835aba73c882348aa4f99be83b6e45ad9b) 
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Author: Andrei Vagin <avagin@openvz.org>
4  * Author: Dmitry Safonov <dima@arista.com>
5  */
6 
7 #include <linux/time_namespace.h>
8 #include <linux/user_namespace.h>
9 #include <linux/sched/signal.h>
10 #include <linux/sched/task.h>
11 #include <linux/clocksource.h>
12 #include <linux/seq_file.h>
13 #include <linux/proc_ns.h>
14 #include <linux/export.h>
15 #include <linux/nstree.h>
16 #include <linux/time.h>
17 #include <linux/slab.h>
18 #include <linux/cred.h>
19 #include <linux/err.h>
20 #include <linux/mm.h>
21 
22 #include "namespace_internal.h"
23 
24 ktime_t do_timens_ktime_to_host(clockid_t clockid, ktime_t tim,
25 				struct timens_offsets *ns_offsets)
26 {
27 	ktime_t offset;
28 
29 	switch (clockid) {
30 	case CLOCK_MONOTONIC:
31 		offset = timespec64_to_ktime(ns_offsets->monotonic);
32 		break;
33 	case CLOCK_BOOTTIME:
34 	case CLOCK_BOOTTIME_ALARM:
35 		offset = timespec64_to_ktime(ns_offsets->boottime);
36 		break;
37 	default:
38 		return tim;
39 	}
40 
41 	/*
42 	 * Check that @tim value is in [offset, KTIME_MAX + offset]
43 	 * and subtract offset.
44 	 */
45 	if (tim < offset) {
46 		/*
47 		 * User can specify @tim *absolute* value - if it's lesser than
48 		 * the time namespace's offset - it's already expired.
49 		 */
50 		tim = 0;
51 	} else {
52 		tim = ktime_sub(tim, offset);
53 		if (unlikely(tim > KTIME_MAX))
54 			tim = KTIME_MAX;
55 	}
56 
57 	return tim;
58 }
59 
60 static struct ucounts *inc_time_namespaces(struct user_namespace *ns)
61 {
62 	return inc_ucount(ns, current_euid(), UCOUNT_TIME_NAMESPACES);
63 }
64 
65 static void dec_time_namespaces(struct ucounts *ucounts)
66 {
67 	dec_ucount(ucounts, UCOUNT_TIME_NAMESPACES);
68 }
69 
70 /**
71  * clone_time_ns - Clone a time namespace
72  * @user_ns:	User namespace which owns a new namespace.
73  * @old_ns:	Namespace to clone
74  *
75  * Clone @old_ns and set the clone refcount to 1
76  *
77  * Return: The new namespace or ERR_PTR.
78  */
79 static struct time_namespace *clone_time_ns(struct user_namespace *user_ns,
80 					  struct time_namespace *old_ns)
81 {
82 	struct time_namespace *ns;
83 	struct ucounts *ucounts;
84 	int err;
85 
86 	err = -ENOSPC;
87 	ucounts = inc_time_namespaces(user_ns);
88 	if (!ucounts)
89 		goto fail;
90 
91 	err = -ENOMEM;
92 	ns = kzalloc_obj(*ns, GFP_KERNEL_ACCOUNT);
93 	if (!ns)
94 		goto fail_dec;
95 
96 	ns->vvar_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
97 	if (!ns->vvar_page)
98 		goto fail_free;
99 
100 	err = ns_common_init(ns);
101 	if (err)
102 		goto fail_free_page;
103 
104 	ns->ucounts = ucounts;
105 	ns->user_ns = get_user_ns(user_ns);
106 	ns->offsets = old_ns->offsets;
107 	ns->frozen_offsets = false;
108 	ns_tree_add(ns);
109 	return ns;
110 
111 fail_free_page:
112 	__free_page(ns->vvar_page);
113 fail_free:
114 	kfree(ns);
115 fail_dec:
116 	dec_time_namespaces(ucounts);
117 fail:
118 	return ERR_PTR(err);
119 }
120 
121 /**
122  * copy_time_ns - Create timens_for_children from @old_ns
123  * @flags:	Cloning flags
124  * @user_ns:	User namespace which owns a new namespace.
125  * @old_ns:	Namespace to clone
126  *
127  * If CLONE_NEWTIME specified in @flags, creates a new timens_for_children;
128  * adds a refcounter to @old_ns otherwise.
129  *
130  * Return: timens_for_children namespace or ERR_PTR.
131  */
132 struct time_namespace *copy_time_ns(u64 flags,
133 	struct user_namespace *user_ns, struct time_namespace *old_ns)
134 {
135 	if (!(flags & CLONE_NEWTIME))
136 		return get_time_ns(old_ns);
137 
138 	return clone_time_ns(user_ns, old_ns);
139 }
140 
141 DEFINE_MUTEX(timens_offset_lock);
142 
143 void free_time_ns(struct time_namespace *ns)
144 {
145 	ns_tree_remove(ns);
146 	dec_time_namespaces(ns->ucounts);
147 	put_user_ns(ns->user_ns);
148 	ns_common_free(ns);
149 	__free_page(ns->vvar_page);
150 	/* Concurrent nstree traversal depends on a grace period. */
151 	kfree_rcu(ns, ns.ns_rcu);
152 }
153 
154 static struct ns_common *timens_get(struct task_struct *task)
155 {
156 	struct time_namespace *ns = NULL;
157 	struct nsproxy *nsproxy;
158 
159 	task_lock(task);
160 	nsproxy = task->nsproxy;
161 	if (nsproxy) {
162 		ns = nsproxy->time_ns;
163 		get_time_ns(ns);
164 	}
165 	task_unlock(task);
166 
167 	return ns ? &ns->ns : NULL;
168 }
169 
170 static struct ns_common *timens_for_children_get(struct task_struct *task)
171 {
172 	struct time_namespace *ns = NULL;
173 	struct nsproxy *nsproxy;
174 
175 	task_lock(task);
176 	nsproxy = task->nsproxy;
177 	if (nsproxy) {
178 		ns = nsproxy->time_ns_for_children;
179 		get_time_ns(ns);
180 	}
181 	task_unlock(task);
182 
183 	return ns ? &ns->ns : NULL;
184 }
185 
186 static void timens_put(struct ns_common *ns)
187 {
188 	put_time_ns(to_time_ns(ns));
189 }
190 
191 static int timens_install(struct nsset *nsset, struct ns_common *new)
192 {
193 	struct nsproxy *nsproxy = nsset->nsproxy;
194 	struct time_namespace *ns = to_time_ns(new);
195 
196 	if (!current_is_single_threaded())
197 		return -EUSERS;
198 
199 	if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
200 	    !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
201 		return -EPERM;
202 
203 	get_time_ns(ns);
204 	put_time_ns(nsproxy->time_ns);
205 	nsproxy->time_ns = ns;
206 
207 	get_time_ns(ns);
208 	put_time_ns(nsproxy->time_ns_for_children);
209 	nsproxy->time_ns_for_children = ns;
210 	return 0;
211 }
212 
213 void timens_on_fork(struct nsproxy *nsproxy, struct task_struct *tsk)
214 {
215 	struct ns_common *nsc = &nsproxy->time_ns_for_children->ns;
216 	struct time_namespace *ns = to_time_ns(nsc);
217 
218 	/* create_new_namespaces() already incremented the ref counter */
219 	if (nsproxy->time_ns == nsproxy->time_ns_for_children)
220 		return;
221 
222 	get_time_ns(ns);
223 	put_time_ns(nsproxy->time_ns);
224 	nsproxy->time_ns = ns;
225 
226 	timens_commit(tsk, ns);
227 }
228 
229 static struct user_namespace *timens_owner(struct ns_common *ns)
230 {
231 	return to_time_ns(ns)->user_ns;
232 }
233 
234 static void show_offset(struct seq_file *m, int clockid, struct timespec64 *ts)
235 {
236 	char *clock;
237 
238 	switch (clockid) {
239 	case CLOCK_BOOTTIME:
240 		clock = "boottime";
241 		break;
242 	case CLOCK_MONOTONIC:
243 		clock = "monotonic";
244 		break;
245 	default:
246 		clock = "unknown";
247 		break;
248 	}
249 	seq_printf(m, "%-10s %10lld %9ld\n", clock, ts->tv_sec, ts->tv_nsec);
250 }
251 
252 void proc_timens_show_offsets(struct task_struct *p, struct seq_file *m)
253 {
254 	struct ns_common *ns;
255 	struct time_namespace *time_ns;
256 
257 	ns = timens_for_children_get(p);
258 	if (!ns)
259 		return;
260 	time_ns = to_time_ns(ns);
261 
262 	show_offset(m, CLOCK_MONOTONIC, &time_ns->offsets.monotonic);
263 	show_offset(m, CLOCK_BOOTTIME, &time_ns->offsets.boottime);
264 	put_time_ns(time_ns);
265 }
266 
267 int proc_timens_set_offset(struct file *file, struct task_struct *p,
268 			   struct proc_timens_offset *offsets, int noffsets)
269 {
270 	struct ns_common *ns;
271 	struct time_namespace *time_ns;
272 	struct timespec64 tp;
273 	int i, err;
274 
275 	ns = timens_for_children_get(p);
276 	if (!ns)
277 		return -ESRCH;
278 	time_ns = to_time_ns(ns);
279 
280 	if (!file_ns_capable(file, time_ns->user_ns, CAP_SYS_TIME)) {
281 		put_time_ns(time_ns);
282 		return -EPERM;
283 	}
284 
285 	for (i = 0; i < noffsets; i++) {
286 		struct proc_timens_offset *off = &offsets[i];
287 
288 		switch (off->clockid) {
289 		case CLOCK_MONOTONIC:
290 			ktime_get_ts64(&tp);
291 			break;
292 		case CLOCK_BOOTTIME:
293 			ktime_get_boottime_ts64(&tp);
294 			break;
295 		default:
296 			err = -EINVAL;
297 			goto out;
298 		}
299 
300 		err = -ERANGE;
301 
302 		if (off->val.tv_sec > KTIME_SEC_MAX ||
303 		    off->val.tv_sec < -KTIME_SEC_MAX)
304 			goto out;
305 
306 		tp = timespec64_add(tp, off->val);
307 		/*
308 		 * KTIME_SEC_MAX is divided by 2 to be sure that KTIME_MAX is
309 		 * still unreachable.
310 		 */
311 		if (tp.tv_sec < 0 || tp.tv_sec > KTIME_SEC_MAX / 2)
312 			goto out;
313 	}
314 
315 	mutex_lock(&timens_offset_lock);
316 	if (time_ns->frozen_offsets) {
317 		err = -EACCES;
318 		goto out_unlock;
319 	}
320 
321 	err = 0;
322 	/* Don't report errors after this line */
323 	for (i = 0; i < noffsets; i++) {
324 		struct proc_timens_offset *off = &offsets[i];
325 		struct timespec64 *offset = NULL;
326 
327 		switch (off->clockid) {
328 		case CLOCK_MONOTONIC:
329 			offset = &time_ns->offsets.monotonic;
330 			break;
331 		case CLOCK_BOOTTIME:
332 			offset = &time_ns->offsets.boottime;
333 			break;
334 		}
335 
336 		*offset = off->val;
337 	}
338 
339 out_unlock:
340 	mutex_unlock(&timens_offset_lock);
341 out:
342 	put_time_ns(time_ns);
343 
344 	return err;
345 }
346 
347 const struct proc_ns_operations timens_operations = {
348 	.name		= "time",
349 	.get		= timens_get,
350 	.put		= timens_put,
351 	.install	= timens_install,
352 	.owner		= timens_owner,
353 };
354 
355 const struct proc_ns_operations timens_for_children_operations = {
356 	.name		= "time_for_children",
357 	.real_ns_name	= "time",
358 	.get		= timens_for_children_get,
359 	.put		= timens_put,
360 	.install	= timens_install,
361 	.owner		= timens_owner,
362 };
363 
364 struct time_namespace init_time_ns = {
365 	.ns		= NS_COMMON_INIT(init_time_ns),
366 	.user_ns	= &init_user_ns,
367 	.frozen_offsets	= true,
368 };
369 
370 void __init time_ns_init(void)
371 {
372 	ns_tree_add(&init_time_ns);
373 }
374