blob: 560da0dab230aab631dab7003597815fc4d21c6b [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>
Eric W. Biederman4308eeb2011-03-23 16:43:13 -070018#include <linux/proc_fs.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
22#define BITS_PER_PAGE (PAGE_SIZE*8)
23
24struct pid_cache {
25 int nr_ids;
26 char name[16];
27 struct kmem_cache *cachep;
28 struct list_head list;
29};
30
31static LIST_HEAD(pid_caches_lh);
32static DEFINE_MUTEX(pid_caches_mutex);
33static struct kmem_cache *pid_ns_cachep;
34
35/*
36 * creates the kmem cache to allocate pids from.
37 * @nr_ids: the number of numerical ids this pid will have to carry
38 */
39
40static struct kmem_cache *create_pid_cachep(int nr_ids)
41{
42 struct pid_cache *pcache;
43 struct kmem_cache *cachep;
44
45 mutex_lock(&pid_caches_mutex);
46 list_for_each_entry(pcache, &pid_caches_lh, list)
47 if (pcache->nr_ids == nr_ids)
48 goto out;
49
50 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
51 if (pcache == NULL)
52 goto err_alloc;
53
54 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
55 cachep = kmem_cache_create(pcache->name,
56 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
57 0, SLAB_HWCACHE_ALIGN, NULL);
58 if (cachep == NULL)
59 goto err_cachep;
60
61 pcache->nr_ids = nr_ids;
62 pcache->cachep = cachep;
63 list_add(&pcache->list, &pid_caches_lh);
64out:
65 mutex_unlock(&pid_caches_mutex);
66 return pcache->cachep;
67
68err_cachep:
69 kfree(pcache);
70err_alloc:
71 mutex_unlock(&pid_caches_mutex);
72 return NULL;
73}
74
Eric W. Biederman0a01f2c2012-08-01 10:33:47 -070075static void proc_cleanup_work(struct work_struct *work)
76{
77 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
78 pid_ns_release_proc(ns);
79}
80
Andrew Vaginf2302502012-10-25 13:38:07 -070081/* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
82#define MAX_PID_NS_LEVEL 32
83
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -070084static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
85 struct pid_namespace *parent_pid_ns)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080086{
87 struct pid_namespace *ns;
Alexey Dobriyaned469a62009-06-17 16:27:52 -070088 unsigned int level = parent_pid_ns->level + 1;
Andrew Vaginf2302502012-10-25 13:38:07 -070089 int i;
90 int err;
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080091
Andrew Vaginf2302502012-10-25 13:38:07 -070092 if (level > MAX_PID_NS_LEVEL) {
93 err = -EINVAL;
94 goto out;
95 }
96
97 err = -ENOMEM;
Pavel Emelyanov84406c12008-07-25 01:48:42 -070098 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080099 if (ns == NULL)
100 goto out;
101
102 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
103 if (!ns->pidmap[0].page)
104 goto out_free;
105
106 ns->pid_cachep = create_pid_cachep(level + 1);
107 if (ns->pid_cachep == NULL)
108 goto out_free_map;
109
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700110 err = proc_alloc_inum(&ns->proc_inum);
111 if (err)
112 goto out_free_map;
113
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800114 kref_init(&ns->kref);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800115 ns->level = level;
Alexey Dobriyaned469a62009-06-17 16:27:52 -0700116 ns->parent = get_pid_ns(parent_pid_ns);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700117 ns->user_ns = get_user_ns(user_ns);
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
136static void destroy_pid_namespace(struct pid_namespace *ns)
137{
138 int i;
139
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700140 proc_free_inum(ns->proc_inum);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800141 for (i = 0; i < PIDMAP_ENTRIES; i++)
142 kfree(ns->pidmap[i].page);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700143 put_user_ns(ns->user_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800144 kmem_cache_free(pid_ns_cachep, ns);
145}
146
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700147struct pid_namespace *copy_pid_ns(unsigned long flags,
148 struct user_namespace *user_ns, struct pid_namespace *old_ns)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800149{
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800150 if (!(flags & CLONE_NEWPID))
Alexey Dobriyandca4a972009-06-17 16:27:53 -0700151 return get_pid_ns(old_ns);
Eric W. Biederman225778d2012-08-02 08:35:35 -0700152 if (task_active_pid_ns(current) != old_ns)
153 return ERR_PTR(-EINVAL);
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700154 return create_pid_namespace(user_ns, old_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800155}
156
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700157static void free_pid_ns(struct kref *kref)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800158{
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700159 struct pid_namespace *ns;
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800160
161 ns = container_of(kref, struct pid_namespace, kref);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800162 destroy_pid_namespace(ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800163}
Cyrill Gorcunovbbc2e3e2012-10-19 13:56:53 -0700164
165void put_pid_ns(struct pid_namespace *ns)
166{
167 struct pid_namespace *parent;
168
169 while (ns != &init_pid_ns) {
170 parent = ns->parent;
171 if (!kref_put(&ns->kref, free_pid_ns))
172 break;
173 ns = parent;
174 }
175}
176EXPORT_SYMBOL_GPL(put_pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800177
178void zap_pid_ns_processes(struct pid_namespace *pid_ns)
179{
180 int nr;
181 int rc;
Eric W. Biederman00c10bc2012-05-31 16:26:40 -0700182 struct task_struct *task, *me = current;
183
184 /* Ignore SIGCHLD causing any terminated children to autoreap */
185 spin_lock_irq(&me->sighand->siglock);
186 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
187 spin_unlock_irq(&me->sighand->siglock);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800188
189 /*
190 * The last thread in the cgroup-init thread group is terminating.
191 * Find remaining pid_ts in the namespace, signal and wait for them
192 * to exit.
193 *
194 * Note: This signals each threads in the namespace - even those that
195 * belong to the same thread group, To avoid this, we would have
196 * to walk the entire tasklist looking a processes in this
197 * namespace, but that could be unnecessarily expensive if the
198 * pid namespace has just a few processes. Or we need to
199 * maintain a tasklist for each pid namespace.
200 *
201 */
202 read_lock(&tasklist_lock);
203 nr = next_pidmap(pid_ns, 1);
204 while (nr > 0) {
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700205 rcu_read_lock();
206
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700207 task = pid_task(find_vpid(nr), PIDTYPE_PID);
Oleg Nesterova02d6fd2012-03-23 15:02:46 -0700208 if (task && !__fatal_signal_pending(task))
209 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700210
211 rcu_read_unlock();
212
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800213 nr = next_pidmap(pid_ns, nr);
214 }
215 read_unlock(&tasklist_lock);
216
Eric W. Biederman6347e902012-06-20 12:53:03 -0700217 /* Firstly reap the EXIT_ZOMBIE children we may have. */
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800218 do {
219 clear_thread_flag(TIF_SIGPENDING);
220 rc = sys_wait4(-1, NULL, __WALL, NULL);
221 } while (rc != -ECHILD);
222
Eric W. Biederman6347e902012-06-20 12:53:03 -0700223 /*
224 * sys_wait4() above can't reap the TASK_DEAD children.
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700225 * Make sure they all go away, see free_pid().
Eric W. Biederman6347e902012-06-20 12:53:03 -0700226 */
227 for (;;) {
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700228 set_current_state(TASK_UNINTERRUPTIBLE);
229 if (pid_ns->nr_hashed == 1)
Eric W. Biederman6347e902012-06-20 12:53:03 -0700230 break;
231 schedule();
232 }
Eric W. Biedermanaf4b8a82012-08-01 15:03:42 -0700233 __set_current_state(TASK_RUNNING);
Eric W. Biederman6347e902012-06-20 12:53:03 -0700234
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700235 if (pid_ns->reboot)
236 current->signal->group_exit_code = pid_ns->reboot;
237
Pavel Emelyanov0b6b0302008-07-25 01:48:47 -0700238 acct_exit_ns(pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800239 return;
240}
241
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700242#ifdef CONFIG_CHECKPOINT_RESTORE
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800243static int pid_ns_ctl_handler(struct ctl_table *table, int write,
244 void __user *buffer, size_t *lenp, loff_t *ppos)
245{
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700246 struct pid_namespace *pid_ns = task_active_pid_ns(current);
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800247 struct ctl_table tmp = *table;
248
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700249 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800250 return -EPERM;
251
252 /*
253 * Writing directly to ns' last_pid field is OK, since this field
254 * is volatile in a living namespace anyway and a code writing to
255 * it should synchronize its usage with external means.
256 */
257
Eric W. Biederman49f4d8b2012-08-02 04:25:10 -0700258 tmp.data = &pid_ns->last_pid;
Andrew Vagin579035d2012-09-17 14:09:12 -0700259 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800260}
261
Andrew Vagin579035d2012-09-17 14:09:12 -0700262extern int pid_max;
263static int zero = 0;
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800264static struct ctl_table pid_ns_ctl_table[] = {
265 {
266 .procname = "ns_last_pid",
267 .maxlen = sizeof(int),
268 .mode = 0666, /* permissions are checked in the handler */
269 .proc_handler = pid_ns_ctl_handler,
Andrew Vagin579035d2012-09-17 14:09:12 -0700270 .extra1 = &zero,
271 .extra2 = &pid_max,
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800272 },
273 { }
274};
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800275static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700276#endif /* CONFIG_CHECKPOINT_RESTORE */
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800277
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700278int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
279{
280 if (pid_ns == &init_pid_ns)
281 return 0;
282
283 switch (cmd) {
284 case LINUX_REBOOT_CMD_RESTART2:
285 case LINUX_REBOOT_CMD_RESTART:
286 pid_ns->reboot = SIGHUP;
287 break;
288
289 case LINUX_REBOOT_CMD_POWER_OFF:
290 case LINUX_REBOOT_CMD_HALT:
291 pid_ns->reboot = SIGINT;
292 break;
293 default:
294 return -EINVAL;
295 }
296
297 read_lock(&tasklist_lock);
298 force_sig(SIGKILL, pid_ns->child_reaper);
299 read_unlock(&tasklist_lock);
300
301 do_exit(0);
302
303 /* Not reached */
304 return 0;
305}
306
Eric W. Biederman57e83912010-03-07 18:17:03 -0800307static void *pidns_get(struct task_struct *task)
308{
309 struct pid_namespace *ns;
310
311 rcu_read_lock();
312 ns = get_pid_ns(task_active_pid_ns(task));
313 rcu_read_unlock();
314
315 return ns;
316}
317
318static void pidns_put(void *ns)
319{
320 put_pid_ns(ns);
321}
322
323static int pidns_install(struct nsproxy *nsproxy, void *ns)
324{
325 struct pid_namespace *active = task_active_pid_ns(current);
326 struct pid_namespace *ancestor, *new = ns;
327
328 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN))
329 return -EPERM;
330
331 /*
332 * Only allow entering the current active pid namespace
333 * or a child of the current active pid namespace.
334 *
335 * This is required for fork to return a usable pid value and
336 * this maintains the property that processes and their
337 * children can not escape their current pid namespace.
338 */
339 if (new->level < active->level)
340 return -EINVAL;
341
342 ancestor = new;
343 while (ancestor->level > active->level)
344 ancestor = ancestor->parent;
345 if (ancestor != active)
346 return -EINVAL;
347
348 put_pid_ns(nsproxy->pid_ns);
349 nsproxy->pid_ns = get_pid_ns(new);
350 return 0;
351}
352
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700353static unsigned int pidns_inum(void *ns)
354{
355 struct pid_namespace *pid_ns = ns;
356 return pid_ns->proc_inum;
357}
358
Eric W. Biederman57e83912010-03-07 18:17:03 -0800359const struct proc_ns_operations pidns_operations = {
360 .name = "pid",
361 .type = CLONE_NEWPID,
362 .get = pidns_get,
363 .put = pidns_put,
364 .install = pidns_install,
Eric W. Biederman98f842e2011-06-15 10:21:48 -0700365 .inum = pidns_inum,
Eric W. Biederman57e83912010-03-07 18:17:03 -0800366};
367
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800368static __init int pid_namespaces_init(void)
369{
370 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700371
372#ifdef CONFIG_CHECKPOINT_RESTORE
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800373 register_sysctl_paths(kern_path, pid_ns_ctl_table);
Cyrill Gorcunov98ed57e2012-05-31 16:26:42 -0700374#endif
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800375 return 0;
376}
377
378__initcall(pid_namespaces_init);