blob: a65ba137fd15b42ba840ca1e46243ef44fa6839c [file] [log] [blame]
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -08001/*
2 * Pid namespaces
3 *
4 * Authors:
5 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
6 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
7 * Many thanks to Oleg Nesterov for comments and help
8 *
9 */
10
11#include <linux/pid.h>
12#include <linux/pid_namespace.h>
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -070013#include <linux/user_namespace.h>
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080014#include <linux/syscalls.h>
15#include <linux/err.h>
Pavel Emelyanov0b6b0302008-07-25 01:48:47 -070016#include <linux/acct.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090017#include <linux/slab.h>
David Howells0bb80f22013-04-12 01:50:06 +010018#include <linux/proc_ns.h>
Daniel Lezcanocf3f8922012-03-28 14:42:51 -070019#include <linux/reboot.h>
Eric W. Biederman523a6a92012-08-03 19:11:22 -070020#include <linux/export.h>
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080021
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080022struct pid_cache {
23 int nr_ids;
24 char name[16];
25 struct kmem_cache *cachep;
26 struct list_head list;
27};
28
29static LIST_HEAD(pid_caches_lh);
30static DEFINE_MUTEX(pid_caches_mutex);
31static struct kmem_cache *pid_ns_cachep;
32
33/*
34 * creates the kmem cache to allocate pids from.
35 * @nr_ids: the number of numerical ids this pid will have to carry
36 */
37
38static struct kmem_cache *create_pid_cachep(int nr_ids)
39{
40 struct pid_cache *pcache;
41 struct kmem_cache *cachep;
42
43 mutex_lock(&pid_caches_mutex);
44 list_for_each_entry(pcache, &pid_caches_lh, list)
45 if (pcache->nr_ids == nr_ids)
46 goto out;
47
48 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
49 if (pcache == NULL)
50 goto err_alloc;
51
52 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
53 cachep = kmem_cache_create(pcache->name,
54 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
55 0, SLAB_HWCACHE_ALIGN, NULL);
56 if (cachep == NULL)
57 goto err_cachep;
58
59 pcache->nr_ids = nr_ids;
60 pcache->cachep = cachep;
61 list_add(&pcache->list, &pid_caches_lh);
62out:
63 mutex_unlock(&pid_caches_mutex);
64 return pcache->cachep;
65
66err_cachep:
67 kfree(pcache);
68err_alloc:
69 mutex_unlock(&pid_caches_mutex);
70 return NULL;
71}
72
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -070073static void proc_cleanup_work(struct work_struct *work)
74{
75 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
76 pid_ns_release_proc(ns);
77}
78
Andrew Vaginf2302502012-10-25 13:38:07 -070079/* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
80#define MAX_PID_NS_LEVEL 32
81
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -070082static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
83 struct pid_namespace *parent_pid_ns)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080084{
85 struct pid_namespace *ns;
Alexey Dobriyaned469a62009-06-17 16:27:52 -070086 unsigned int level = parent_pid_ns->level + 1;
Andrew Vaginf2302502012-10-25 13:38:07 -070087 int i;
88 int err;
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080089
Andrew Vaginf2302502012-10-25 13:38:07 -070090 if (level > MAX_PID_NS_LEVEL) {
91 err = -EINVAL;
92 goto out;
93 }
94
95 err = -ENOMEM;
Pavel Emelyanov84406c12008-07-25 01:48:42 -070096 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080097 if (ns == NULL)
98 goto out;
99
100 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
101 if (!ns->pidmap[0].page)
102 goto out_free;
103
104 ns->pid_cachep = create_pid_cachep(level + 1);
105 if (ns->pid_cachep == NULL)
106 goto out_free_map;
107
Al Viro6344c432014-11-01 00:45:45 -0400108 err = ns_alloc_inum(&ns->ns);
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700109 if (err)
110 goto out_free_map;
Al Viro33c42942014-11-01 02:32:53 -0400111 ns->ns.ops = &pidns_operations;
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700112
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800113 kref_init(&ns->kref);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800114 ns->level = level;
Alexey Dobriyaned469a62009-06-17 16:27:52 -0700115 ns->parent = get_pid_ns(parent_pid_ns);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700116 ns->user_ns = get_user_ns(user_ns);
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800117 ns->nr_hashed = PIDNS_HASH_ADDING;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700118 INIT_WORK(&ns->proc_work, proc_cleanup_work);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800119
120 set_bit(0, ns->pidmap[0].page);
121 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
122
Pavel Emelyanov84406c12008-07-25 01:48:42 -0700123 for (i = 1; i < PIDMAP_ENTRIES; i++)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800124 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800125
126 return ns;
127
128out_free_map:
129 kfree(ns->pidmap[0].page);
130out_free:
131 kmem_cache_free(pid_ns_cachep, ns);
132out:
Eric W. Biederman4308eeb2011-03-23 16:43:13 -0700133 return ERR_PTR(err);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800134}
135
Al Viro1adfcb02013-10-03 13:28:06 -0400136static void delayed_free_pidns(struct rcu_head *p)
137{
138 kmem_cache_free(pid_ns_cachep,
139 container_of(p, struct pid_namespace, rcu));
140}
141
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800142static void destroy_pid_namespace(struct pid_namespace *ns)
143{
144 int i;
145
Al Viro6344c432014-11-01 00:45:45 -0400146 ns_free_inum(&ns->ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800147 for (i = 0; i < PIDMAP_ENTRIES; i++)
148 kfree(ns->pidmap[i].page);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700149 put_user_ns(ns->user_ns);
Al Viro1adfcb02013-10-03 13:28:06 -0400150 call_rcu(&ns->rcu, delayed_free_pidns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800151}
152
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700153struct pid_namespace *copy_pid_ns(unsigned long flags,
154 struct user_namespace *user_ns, struct pid_namespace *old_ns)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800155{
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800156 if (!(flags & CLONE_NEWPID))
Alexey Dobriyandca4a972009-06-17 16:27:53 -0700157 return get_pid_ns(old_ns);
Eric W. Biederman225778d2012-08-02 08:35:35 -0700158 if (task_active_pid_ns(current) != old_ns)
159 return ERR_PTR(-EINVAL);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700160 return create_pid_namespace(user_ns, old_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800161}
162
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700163static void free_pid_ns(struct kref *kref)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800164{
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700165 struct pid_namespace *ns;
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800166
167 ns = container_of(kref, struct pid_namespace, kref);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800168 destroy_pid_namespace(ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800169}
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700170
171void put_pid_ns(struct pid_namespace *ns)
172{
173 struct pid_namespace *parent;
174
175 while (ns != &init_pid_ns) {
176 parent = ns->parent;
177 if (!kref_put(&ns->kref, free_pid_ns))
178 break;
179 ns = parent;
180 }
181}
182EXPORT_SYMBOL_GPL(put_pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800183
184void zap_pid_ns_processes(struct pid_namespace *pid_ns)
185{
186 int nr;
187 int rc;
Eric W. Biederman00c10bc2012-05-31 16:26:40 -0700188 struct task_struct *task, *me = current;
Eric W. Biederman751c6442013-03-26 02:27:11 -0700189 int init_pids = thread_group_leader(me) ? 1 : 2;
Eric W. Biederman00c10bc2012-05-31 16:26:40 -0700190
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800191 /* Don't allow any more processes into the pid namespace */
192 disable_pid_allocation(pid_ns);
193
Oleg Nesterova53b8312014-12-10 15:55:28 -0800194 /*
195 * Ignore SIGCHLD causing any terminated children to autoreap.
196 * This speeds up the namespace shutdown, plus see the comment
197 * below.
198 */
Eric W. Biederman00c10bc2012-05-31 16:26:40 -0700199 spin_lock_irq(&me->sighand->siglock);
200 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
201 spin_unlock_irq(&me->sighand->siglock);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800202
203 /*
204 * The last thread in the cgroup-init thread group is terminating.
205 * Find remaining pid_ts in the namespace, signal and wait for them
206 * to exit.
207 *
208 * Note: This signals each threads in the namespace - even those that
209 * belong to the same thread group, To avoid this, we would have
210 * to walk the entire tasklist looking a processes in this
211 * namespace, but that could be unnecessarily expensive if the
212 * pid namespace has just a few processes. Or we need to
213 * maintain a tasklist for each pid namespace.
214 *
215 */
216 read_lock(&tasklist_lock);
217 nr = next_pidmap(pid_ns, 1);
218 while (nr > 0) {
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700219 rcu_read_lock();
220
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700221 task = pid_task(find_vpid(nr), PIDTYPE_PID);
Oleg Nesterova02d6fd2012-03-23 15:02:46 -0700222 if (task && !__fatal_signal_pending(task))
223 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700224
225 rcu_read_unlock();
226
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800227 nr = next_pidmap(pid_ns, nr);
228 }
229 read_unlock(&tasklist_lock);
230
Oleg Nesterova53b8312014-12-10 15:55:28 -0800231 /*
232 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
233 * sys_wait4() will also block until our children traced from the
234 * parent namespace are detached and become EXIT_DEAD.
235 */
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800236 do {
237 clear_thread_flag(TIF_SIGPENDING);
238 rc = sys_wait4(-1, NULL, __WALL, NULL);
239 } while (rc != -ECHILD);
240
Eric W. Biederman6347e902012-06-20 12:53:03 -0700241 /*
Oleg Nesterova53b8312014-12-10 15:55:28 -0800242 * sys_wait4() above can't reap the EXIT_DEAD children but we do not
243 * really care, we could reparent them to the global init. We could
244 * exit and reap ->child_reaper even if it is not the last thread in
245 * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
246 * pid_ns can not go away until proc_kill_sb() drops the reference.
247 *
248 * But this ns can also have other tasks injected by setns()+fork().
249 * Again, ignoring the user visible semantics we do not really need
250 * to wait until they are all reaped, but they can be reparented to
251 * us and thus we need to ensure that pid->child_reaper stays valid
252 * until they all go away. See free_pid()->wake_up_process().
253 *
254 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
255 * if reparented.
Eric W. Biederman6347e902012-06-20 12:53:03 -0700256 */
257 for (;;) {
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700258 set_current_state(TASK_UNINTERRUPTIBLE);
Eric W. Biederman751c6442013-03-26 02:27:11 -0700259 if (pid_ns->nr_hashed == init_pids)
Eric W. Biederman6347e902012-06-20 12:53:03 -0700260 break;
261 schedule();
262 }
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700263 __set_current_state(TASK_RUNNING);
Eric W. Biederman6347e902012-06-20 12:53:03 -0700264
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700265 if (pid_ns->reboot)
266 current->signal->group_exit_code = pid_ns->reboot;
267
Pavel Emelyanov0b6b0302008-07-25 01:48:47 -0700268 acct_exit_ns(pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800269 return;
270}
271
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700272#ifdef CONFIG_CHECKPOINT_RESTORE
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800273static int pid_ns_ctl_handler(struct ctl_table *table, int write,
274 void __user *buffer, size_t *lenp, loff_t *ppos)
275{
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700276 struct pid_namespace *pid_ns = task_active_pid_ns(current);
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800277 struct ctl_table tmp = *table;
278
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700279 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800280 return -EPERM;
281
282 /*
283 * Writing directly to ns' last_pid field is OK, since this field
284 * is volatile in a living namespace anyway and a code writing to
285 * it should synchronize its usage with external means.
286 */
287
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700288 tmp.data = &pid_ns->last_pid;
Andrew Vagin579035d2012-09-17 14:09:12 -0700289 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800290}
291
Andrew Vagin579035d2012-09-17 14:09:12 -0700292extern int pid_max;
293static int zero = 0;
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800294static struct ctl_table pid_ns_ctl_table[] = {
295 {
296 .procname = "ns_last_pid",
297 .maxlen = sizeof(int),
298 .mode = 0666, /* permissions are checked in the handler */
299 .proc_handler = pid_ns_ctl_handler,
Andrew Vagin579035d2012-09-17 14:09:12 -0700300 .extra1 = &zero,
301 .extra2 = &pid_max,
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800302 },
303 { }
304};
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800305static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700306#endif /* CONFIG_CHECKPOINT_RESTORE */
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800307
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700308int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
309{
310 if (pid_ns == &init_pid_ns)
311 return 0;
312
313 switch (cmd) {
314 case LINUX_REBOOT_CMD_RESTART2:
315 case LINUX_REBOOT_CMD_RESTART:
316 pid_ns->reboot = SIGHUP;
317 break;
318
319 case LINUX_REBOOT_CMD_POWER_OFF:
320 case LINUX_REBOOT_CMD_HALT:
321 pid_ns->reboot = SIGINT;
322 break;
323 default:
324 return -EINVAL;
325 }
326
327 read_lock(&tasklist_lock);
328 force_sig(SIGKILL, pid_ns->child_reaper);
329 read_unlock(&tasklist_lock);
330
331 do_exit(0);
332
333 /* Not reached */
334 return 0;
335}
336
Al Viro3c041182014-11-01 00:25:30 -0400337static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
338{
339 return container_of(ns, struct pid_namespace, ns);
340}
341
Al Viro64964522014-11-01 00:37:32 -0400342static struct ns_common *pidns_get(struct task_struct *task)
Eric W. Biederman57e83912010-03-07 18:17:03 -0800343{
344 struct pid_namespace *ns;
345
346 rcu_read_lock();
Oleg Nesterovd2308222014-04-02 17:45:05 +0200347 ns = task_active_pid_ns(task);
348 if (ns)
349 get_pid_ns(ns);
Eric W. Biederman57e83912010-03-07 18:17:03 -0800350 rcu_read_unlock();
351
Al Viro3c041182014-11-01 00:25:30 -0400352 return ns ? &ns->ns : NULL;
Eric W. Biederman57e83912010-03-07 18:17:03 -0800353}
354
Al Viro64964522014-11-01 00:37:32 -0400355static void pidns_put(struct ns_common *ns)
Eric W. Biederman57e83912010-03-07 18:17:03 -0800356{
Al Viro3c041182014-11-01 00:25:30 -0400357 put_pid_ns(to_pid_ns(ns));
Eric W. Biederman57e83912010-03-07 18:17:03 -0800358}
359
Al Viro64964522014-11-01 00:37:32 -0400360static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
Eric W. Biederman57e83912010-03-07 18:17:03 -0800361{
362 struct pid_namespace *active = task_active_pid_ns(current);
Al Viro3c041182014-11-01 00:25:30 -0400363 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
Eric W. Biederman57e83912010-03-07 18:17:03 -0800364
Eric W. Biederman5e4a0842012-12-14 07:55:36 -0800365 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -0700366 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Eric W. Biederman57e83912010-03-07 18:17:03 -0800367 return -EPERM;
368
369 /*
370 * Only allow entering the current active pid namespace
371 * or a child of the current active pid namespace.
372 *
373 * This is required for fork to return a usable pid value and
374 * this maintains the property that processes and their
375 * children can not escape their current pid namespace.
376 */
377 if (new->level < active->level)
378 return -EINVAL;
379
380 ancestor = new;
381 while (ancestor->level > active->level)
382 ancestor = ancestor->parent;
383 if (ancestor != active)
384 return -EINVAL;
385
Andy Lutomirskic2b1df22013-08-22 11:39:16 -0700386 put_pid_ns(nsproxy->pid_ns_for_children);
387 nsproxy->pid_ns_for_children = get_pid_ns(new);
Eric W. Biederman57e83912010-03-07 18:17:03 -0800388 return 0;
389}
390
391const struct proc_ns_operations pidns_operations = {
392 .name = "pid",
393 .type = CLONE_NEWPID,
394 .get = pidns_get,
395 .put = pidns_put,
396 .install = pidns_install,
397};
398
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800399static __init int pid_namespaces_init(void)
400{
401 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700402
403#ifdef CONFIG_CHECKPOINT_RESTORE
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800404 register_sysctl_paths(kern_path, pid_ns_ctl_table);
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700405#endif
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800406 return 0;
407}
408
409__initcall(pid_namespaces_init);