blob: db95d8eb761b4f92c4722a2667fb258e7e2b64c8 [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
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700108 err = proc_alloc_inum(&ns->proc_inum);
109 if (err)
110 goto out_free_map;
111
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800112 kref_init(&ns->kref);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800113 ns->level = level;
Alexey Dobriyaned469a62009-06-17 16:27:52 -0700114 ns->parent = get_pid_ns(parent_pid_ns);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700115 ns->user_ns = get_user_ns(user_ns);
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800116 ns->nr_hashed = PIDNS_HASH_ADDING;
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -0700117 INIT_WORK(&ns->proc_work, proc_cleanup_work);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800118
119 set_bit(0, ns->pidmap[0].page);
120 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
121
Pavel Emelyanov84406c12008-07-25 01:48:42 -0700122 for (i = 1; i < PIDMAP_ENTRIES; i++)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800123 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800124
125 return ns;
126
127out_free_map:
128 kfree(ns->pidmap[0].page);
129out_free:
130 kmem_cache_free(pid_ns_cachep, ns);
131out:
Eric W. Biederman4308eeb2011-03-23 16:43:13 -0700132 return ERR_PTR(err);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800133}
134
Al Viro1adfcb02013-10-03 13:28:06 -0400135static void delayed_free_pidns(struct rcu_head *p)
136{
137 kmem_cache_free(pid_ns_cachep,
138 container_of(p, struct pid_namespace, rcu));
139}
140
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800141static void destroy_pid_namespace(struct pid_namespace *ns)
142{
143 int i;
144
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700145 proc_free_inum(ns->proc_inum);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800146 for (i = 0; i < PIDMAP_ENTRIES; i++)
147 kfree(ns->pidmap[i].page);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700148 put_user_ns(ns->user_ns);
Al Viro1adfcb02013-10-03 13:28:06 -0400149 call_rcu(&ns->rcu, delayed_free_pidns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800150}
151
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700152struct pid_namespace *copy_pid_ns(unsigned long flags,
153 struct user_namespace *user_ns, struct pid_namespace *old_ns)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800154{
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800155 if (!(flags & CLONE_NEWPID))
Alexey Dobriyandca4a972009-06-17 16:27:53 -0700156 return get_pid_ns(old_ns);
Eric W. Biederman225778d2012-08-02 08:35:35 -0700157 if (task_active_pid_ns(current) != old_ns)
158 return ERR_PTR(-EINVAL);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700159 return create_pid_namespace(user_ns, old_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800160}
161
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700162static void free_pid_ns(struct kref *kref)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800163{
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700164 struct pid_namespace *ns;
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800165
166 ns = container_of(kref, struct pid_namespace, kref);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800167 destroy_pid_namespace(ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800168}
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700169
170void put_pid_ns(struct pid_namespace *ns)
171{
172 struct pid_namespace *parent;
173
174 while (ns != &init_pid_ns) {
175 parent = ns->parent;
176 if (!kref_put(&ns->kref, free_pid_ns))
177 break;
178 ns = parent;
179 }
180}
181EXPORT_SYMBOL_GPL(put_pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800182
183void zap_pid_ns_processes(struct pid_namespace *pid_ns)
184{
185 int nr;
186 int rc;
Eric W. Biederman00c10bc2012-05-31 16:26:40 -0700187 struct task_struct *task, *me = current;
Eric W. Biederman751c6442013-03-26 02:27:11 -0700188 int init_pids = thread_group_leader(me) ? 1 : 2;
Eric W. Biederman00c10bc2012-05-31 16:26:40 -0700189
Eric W. Biedermanc876ad762012-12-21 20:27:12 -0800190 /* Don't allow any more processes into the pid namespace */
191 disable_pid_allocation(pid_ns);
192
Eric W. Biederman00c10bc2012-05-31 16:26:40 -0700193 /* Ignore SIGCHLD causing any terminated children to autoreap */
194 spin_lock_irq(&me->sighand->siglock);
195 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
196 spin_unlock_irq(&me->sighand->siglock);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800197
198 /*
199 * The last thread in the cgroup-init thread group is terminating.
200 * Find remaining pid_ts in the namespace, signal and wait for them
201 * to exit.
202 *
203 * Note: This signals each threads in the namespace - even those that
204 * belong to the same thread group, To avoid this, we would have
205 * to walk the entire tasklist looking a processes in this
206 * namespace, but that could be unnecessarily expensive if the
207 * pid namespace has just a few processes. Or we need to
208 * maintain a tasklist for each pid namespace.
209 *
210 */
211 read_lock(&tasklist_lock);
212 nr = next_pidmap(pid_ns, 1);
213 while (nr > 0) {
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700214 rcu_read_lock();
215
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700216 task = pid_task(find_vpid(nr), PIDTYPE_PID);
Oleg Nesterova02d6fd2012-03-23 15:02:46 -0700217 if (task && !__fatal_signal_pending(task))
218 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700219
220 rcu_read_unlock();
221
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800222 nr = next_pidmap(pid_ns, nr);
223 }
224 read_unlock(&tasklist_lock);
225
Eric W. Biederman6347e902012-06-20 12:53:03 -0700226 /* Firstly reap the EXIT_ZOMBIE children we may have. */
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800227 do {
228 clear_thread_flag(TIF_SIGPENDING);
229 rc = sys_wait4(-1, NULL, __WALL, NULL);
230 } while (rc != -ECHILD);
231
Eric W. Biederman6347e902012-06-20 12:53:03 -0700232 /*
233 * sys_wait4() above can't reap the TASK_DEAD children.
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700234 * Make sure they all go away, see free_pid().
Eric W. Biederman6347e902012-06-20 12:53:03 -0700235 */
236 for (;;) {
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700237 set_current_state(TASK_UNINTERRUPTIBLE);
Eric W. Biederman751c6442013-03-26 02:27:11 -0700238 if (pid_ns->nr_hashed == init_pids)
Eric W. Biederman6347e902012-06-20 12:53:03 -0700239 break;
240 schedule();
241 }
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700242 __set_current_state(TASK_RUNNING);
Eric W. Biederman6347e902012-06-20 12:53:03 -0700243
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700244 if (pid_ns->reboot)
245 current->signal->group_exit_code = pid_ns->reboot;
246
Pavel Emelyanov0b6b0302008-07-25 01:48:47 -0700247 acct_exit_ns(pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800248 return;
249}
250
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700251#ifdef CONFIG_CHECKPOINT_RESTORE
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800252static int pid_ns_ctl_handler(struct ctl_table *table, int write,
253 void __user *buffer, size_t *lenp, loff_t *ppos)
254{
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700255 struct pid_namespace *pid_ns = task_active_pid_ns(current);
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800256 struct ctl_table tmp = *table;
257
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700258 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800259 return -EPERM;
260
261 /*
262 * Writing directly to ns' last_pid field is OK, since this field
263 * is volatile in a living namespace anyway and a code writing to
264 * it should synchronize its usage with external means.
265 */
266
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700267 tmp.data = &pid_ns->last_pid;
Andrew Vagin579035d2012-09-17 14:09:12 -0700268 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800269}
270
Andrew Vagin579035d2012-09-17 14:09:12 -0700271extern int pid_max;
272static int zero = 0;
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800273static struct ctl_table pid_ns_ctl_table[] = {
274 {
275 .procname = "ns_last_pid",
276 .maxlen = sizeof(int),
277 .mode = 0666, /* permissions are checked in the handler */
278 .proc_handler = pid_ns_ctl_handler,
Andrew Vagin579035d2012-09-17 14:09:12 -0700279 .extra1 = &zero,
280 .extra2 = &pid_max,
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800281 },
282 { }
283};
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800284static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700285#endif /* CONFIG_CHECKPOINT_RESTORE */
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800286
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700287int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
288{
289 if (pid_ns == &init_pid_ns)
290 return 0;
291
292 switch (cmd) {
293 case LINUX_REBOOT_CMD_RESTART2:
294 case LINUX_REBOOT_CMD_RESTART:
295 pid_ns->reboot = SIGHUP;
296 break;
297
298 case LINUX_REBOOT_CMD_POWER_OFF:
299 case LINUX_REBOOT_CMD_HALT:
300 pid_ns->reboot = SIGINT;
301 break;
302 default:
303 return -EINVAL;
304 }
305
306 read_lock(&tasklist_lock);
307 force_sig(SIGKILL, pid_ns->child_reaper);
308 read_unlock(&tasklist_lock);
309
310 do_exit(0);
311
312 /* Not reached */
313 return 0;
314}
315
Eric W. Biederman57e83912010-03-07 18:17:03 -0800316static void *pidns_get(struct task_struct *task)
317{
318 struct pid_namespace *ns;
319
320 rcu_read_lock();
Oleg Nesterovd2308222014-04-02 17:45:05 +0200321 ns = task_active_pid_ns(task);
322 if (ns)
323 get_pid_ns(ns);
Eric W. Biederman57e83912010-03-07 18:17:03 -0800324 rcu_read_unlock();
325
326 return ns;
327}
328
329static void pidns_put(void *ns)
330{
331 put_pid_ns(ns);
332}
333
334static int pidns_install(struct nsproxy *nsproxy, void *ns)
335{
336 struct pid_namespace *active = task_active_pid_ns(current);
337 struct pid_namespace *ancestor, *new = ns;
338
Eric W. Biederman5e4a0842012-12-14 07:55:36 -0800339 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
Eric W. Biedermanc7b96ac2013-03-20 12:49:49 -0700340 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
Eric W. Biederman57e83912010-03-07 18:17:03 -0800341 return -EPERM;
342
343 /*
344 * Only allow entering the current active pid namespace
345 * or a child of the current active pid namespace.
346 *
347 * This is required for fork to return a usable pid value and
348 * this maintains the property that processes and their
349 * children can not escape their current pid namespace.
350 */
351 if (new->level < active->level)
352 return -EINVAL;
353
354 ancestor = new;
355 while (ancestor->level > active->level)
356 ancestor = ancestor->parent;
357 if (ancestor != active)
358 return -EINVAL;
359
Andy Lutomirskic2b1df22013-08-22 11:39:16 -0700360 put_pid_ns(nsproxy->pid_ns_for_children);
361 nsproxy->pid_ns_for_children = get_pid_ns(new);
Eric W. Biederman57e83912010-03-07 18:17:03 -0800362 return 0;
363}
364
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700365static unsigned int pidns_inum(void *ns)
366{
367 struct pid_namespace *pid_ns = ns;
368 return pid_ns->proc_inum;
369}
370
Eric W. Biederman57e83912010-03-07 18:17:03 -0800371const struct proc_ns_operations pidns_operations = {
372 .name = "pid",
373 .type = CLONE_NEWPID,
374 .get = pidns_get,
375 .put = pidns_put,
376 .install = pidns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700377 .inum = pidns_inum,
Eric W. Biederman57e83912010-03-07 18:17:03 -0800378};
379
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800380static __init int pid_namespaces_init(void)
381{
382 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700383
384#ifdef CONFIG_CHECKPOINT_RESTORE
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800385 register_sysctl_paths(kern_path, pid_ns_ctl_table);
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700386#endif
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800387 return 0;
388}
389
390__initcall(pid_namespaces_init);