projects/linux·blob master · refs

362 lines · 7.9 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Author: Andrei Vagin <[email protected]>
4 * Author: Dmitry Safonov <[email protected]>
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#include <linux/cleanup.h>
22
23#include "namespace_internal.h"
24
25ktime_t do_timens_ktime_to_host(clockid_t clockid, ktime_t tim,
26 struct timens_offsets *ns_offsets)
27{
28 ktime_t offset;
29
30 switch (clockid) {
31 case CLOCK_MONOTONIC:
32 offset = timespec64_to_ktime(ns_offsets->monotonic);
33 break;
34 case CLOCK_BOOTTIME:
35 case CLOCK_BOOTTIME_ALARM:
36 offset = timespec64_to_ktime(ns_offsets->boottime);
37 break;
38 default:
39 return tim;
40 }
41
42 /*
43 * Check that @tim value is in [offset, KTIME_MAX + offset]
44 * and subtract offset.
45 */
46 if (tim < offset) {
47 /*
48 * User can specify @tim *absolute* value - if it's lesser than
49 * the time namespace's offset - it's already expired.
50 */
51 tim = 0;
52 } else {
53 tim = ktime_sub(tim, offset);
54 if (unlikely(tim > KTIME_MAX))
55 tim = KTIME_MAX;
56 }
57
58 return tim;
59}
60EXPORT_SYMBOL_GPL(do_timens_ktime_to_host);
61
62static struct ucounts *inc_time_namespaces(struct user_namespace *ns)
63{
64 return inc_ucount(ns, current_euid(), UCOUNT_TIME_NAMESPACES);
65}
66
67static void dec_time_namespaces(struct ucounts *ucounts)
68{
69 dec_ucount(ucounts, UCOUNT_TIME_NAMESPACES);
70}
71
72/**
73 * clone_time_ns - Clone a time namespace
74 * @user_ns: User namespace which owns a new namespace.
75 * @old_ns: Namespace to clone
76 *
77 * Clone @old_ns and set the clone refcount to 1
78 *
79 * Return: The new namespace or ERR_PTR.
80 */
81static struct time_namespace *clone_time_ns(struct user_namespace *user_ns,
82 struct time_namespace *old_ns)
83{
84 struct time_namespace *ns;
85 struct ucounts *ucounts;
86 int err;
87
88 err = -ENOSPC;
89 ucounts = inc_time_namespaces(user_ns);
90 if (!ucounts)
91 goto fail;
92
93 err = -ENOMEM;
94 ns = kzalloc_obj(*ns, GFP_KERNEL_ACCOUNT);
95 if (!ns)
96 goto fail_dec;
97
98 err = timens_vdso_alloc_vvar_page(ns);
99 if (err)
100 goto fail_free;
101
102 err = ns_common_init(ns);
103 if (err)
104 goto fail_free_page;
105
106 ns->ucounts = ucounts;
107 ns->user_ns = get_user_ns(user_ns);
108 ns->offsets = old_ns->offsets;
109 ns->frozen_offsets = false;
110 ns_tree_add(ns);
111 return ns;
112
113fail_free_page:
114 timens_vdso_free_vvar_page(ns);
115fail_free:
116 kfree(ns);
117fail_dec:
118 dec_time_namespaces(ucounts);
119fail:
120 return ERR_PTR(err);
121}
122
123/**
124 * copy_time_ns - Create timens_for_children from @old_ns
125 * @flags: Cloning flags
126 * @user_ns: User namespace which owns a new namespace.
127 * @old_ns: Namespace to clone
128 *
129 * If CLONE_NEWTIME specified in @flags, creates a new timens_for_children;
130 * adds a refcounter to @old_ns otherwise.
131 *
132 * Return: timens_for_children namespace or ERR_PTR.
133 */
134struct time_namespace *copy_time_ns(u64 flags,
135 struct user_namespace *user_ns, struct time_namespace *old_ns)
136{
137 if (!(flags & CLONE_NEWTIME))
138 return get_time_ns(old_ns);
139
140 return clone_time_ns(user_ns, old_ns);
141}
142
143DEFINE_MUTEX(timens_offset_lock);
144
145void free_time_ns(struct time_namespace *ns)
146{
147 ns_tree_remove(ns);
148 dec_time_namespaces(ns->ucounts);
149 put_user_ns(ns->user_ns);
150 ns_common_free(ns);
151 timens_vdso_free_vvar_page(ns);
152 /* Concurrent nstree traversal depends on a grace period. */
153 kfree_rcu(ns, ns.ns_rcu);
154}
155
156static struct ns_common *timens_get(struct task_struct *task)
157{
158 struct time_namespace *ns;
159 struct nsproxy *nsproxy;
160
161 guard(task_lock)(task);
162 nsproxy = task->nsproxy;
163 if (!nsproxy)
164 return NULL;
165
166 ns = nsproxy->time_ns;
167 get_time_ns(ns);
168 return &ns->ns;
169}
170
171static struct ns_common *timens_for_children_get(struct task_struct *task)
172{
173 struct time_namespace *ns;
174 struct nsproxy *nsproxy;
175
176 guard(task_lock)(task);
177 nsproxy = task->nsproxy;
178 if (!nsproxy)
179 return NULL;
180
181 ns = nsproxy->time_ns_for_children;
182 get_time_ns(ns);
183 return &ns->ns;
184}
185
186static void timens_put(struct ns_common *ns)
187{
188 put_time_ns(to_time_ns(ns));
189}
190
191static 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
213void 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
229static struct user_namespace *timens_owner(struct ns_common *ns)
230{
231 return to_time_ns(ns)->user_ns;
232}
233
234static 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
252void proc_timens_show_offsets(struct task_struct *p, struct seq_file *m)
253{
254 struct time_namespace *time_ns __free(time_ns) = NULL;
255 struct ns_common *ns = timens_for_children_get(p);
256
257 if (!ns)
258 return;
259
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}
265
266int proc_timens_set_offset(struct file *file, struct task_struct *p,
267 struct proc_timens_offset *offsets, int noffsets)
268{
269 struct time_namespace *time_ns __free(time_ns) = NULL;
270 struct ns_common *ns = timens_for_children_get(p);
271 struct timespec64 tp;
272 int i;
273
274 if (!ns)
275 return -ESRCH;
276
277 time_ns = to_time_ns(ns);
278
279 if (!file_ns_capable(file, time_ns->user_ns, CAP_SYS_TIME))
280 return -EPERM;
281
282 for (i = 0; i < noffsets; i++) {
283 struct proc_timens_offset *off = &offsets[i];
284
285 switch (off->clockid) {
286 case CLOCK_MONOTONIC:
287 ktime_get_ts64(&tp);
288 break;
289 case CLOCK_BOOTTIME:
290 ktime_get_boottime_ts64(&tp);
291 break;
292 default:
293 return -EINVAL;
294 }
295
296 if (off->val.tv_sec > KTIME_SEC_MAX || off->val.tv_sec < -KTIME_SEC_MAX)
297 return -ERANGE;
298
299 if (off->val.tv_nsec < 0 || off->val.tv_nsec >= NSEC_PER_SEC)
300 return -EINVAL;
301
302 tp = timespec64_add(tp, off->val);
303 /*
304 * KTIME_SEC_MAX is divided by 2 to be sure that KTIME_MAX is
305 * still unreachable.
306 */
307 if (tp.tv_sec < 0 || tp.tv_sec > KTIME_SEC_MAX / 2)
308 return -ERANGE;
309 }
310
311 guard(mutex)(&timens_offset_lock);
312 if (time_ns->frozen_offsets)
313 return -EACCES;
314
315 /* Don't report errors after this line */
316 for (i = 0; i < noffsets; i++) {
317 struct proc_timens_offset *off = &offsets[i];
318 struct timespec64 *offset = NULL;
319
320 switch (off->clockid) {
321 case CLOCK_MONOTONIC:
322 offset = &time_ns->offsets.monotonic;
323 break;
324 case CLOCK_BOOTTIME:
325 offset = &time_ns->offsets.boottime;
326 break;
327 }
328
329 *offset = off->val;
330 }
331
332 return 0;
333}
334
335const struct proc_ns_operations timens_operations = {
336 .name = "time",
337 .get = timens_get,
338 .put = timens_put,
339 .install = timens_install,
340 .owner = timens_owner,
341};
342
343const struct proc_ns_operations timens_for_children_operations = {
344 .name = "time_for_children",
345 .real_ns_name = "time",
346 .get = timens_for_children_get,
347 .put = timens_put,
348 .install = timens_install,
349 .owner = timens_owner,
350};
351
352struct time_namespace init_time_ns = {
353 .ns = NS_COMMON_INIT(init_time_ns),
354 .user_ns = &init_user_ns,
355 .frozen_offsets = true,
356};
357EXPORT_SYMBOL_GPL(init_time_ns);
358
359void __init time_ns_init(void)
360{
361 ns_tree_add(&init_time_ns);
362}