blob: 57bc1fd35b3cbe6bffdbfe71af5f13fc00648b81 [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>
13#include <linux/syscalls.h>
14#include <linux/err.h>
Pavel Emelyanov0b6b0302008-07-25 01:48:47 -070015#include <linux/acct.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/slab.h>
Eric W. Biederman4308eeb2011-03-23 16:43:13 -070017#include <linux/proc_fs.h>
Daniel Lezcanocf3f8922012-03-28 14:42:51 -070018#include <linux/reboot.h>
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080019
20#define BITS_PER_PAGE (PAGE_SIZE*8)
21
22struct 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
Alexey Dobriyaned469a62009-06-17 16:27:52 -070073static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080074{
75 struct pid_namespace *ns;
Alexey Dobriyaned469a62009-06-17 16:27:52 -070076 unsigned int level = parent_pid_ns->level + 1;
Eric W. Biederman4308eeb2011-03-23 16:43:13 -070077 int i, err = -ENOMEM;
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080078
Pavel Emelyanov84406c12008-07-25 01:48:42 -070079 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080080 if (ns == NULL)
81 goto out;
82
83 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
84 if (!ns->pidmap[0].page)
85 goto out_free;
86
87 ns->pid_cachep = create_pid_cachep(level + 1);
88 if (ns->pid_cachep == NULL)
89 goto out_free_map;
90
91 kref_init(&ns->kref);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080092 ns->level = level;
Alexey Dobriyaned469a62009-06-17 16:27:52 -070093 ns->parent = get_pid_ns(parent_pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080094
95 set_bit(0, ns->pidmap[0].page);
96 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
97
Pavel Emelyanov84406c12008-07-25 01:48:42 -070098 for (i = 1; i < PIDMAP_ENTRIES; i++)
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -080099 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800100
Eric W. Biederman4308eeb2011-03-23 16:43:13 -0700101 err = pid_ns_prepare_proc(ns);
102 if (err)
103 goto out_put_parent_pid_ns;
104
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800105 return ns;
106
Eric W. Biederman4308eeb2011-03-23 16:43:13 -0700107out_put_parent_pid_ns:
108 put_pid_ns(parent_pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800109out_free_map:
110 kfree(ns->pidmap[0].page);
111out_free:
112 kmem_cache_free(pid_ns_cachep, ns);
113out:
Eric W. Biederman4308eeb2011-03-23 16:43:13 -0700114 return ERR_PTR(err);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800115}
116
117static void destroy_pid_namespace(struct pid_namespace *ns)
118{
119 int i;
120
121 for (i = 0; i < PIDMAP_ENTRIES; i++)
122 kfree(ns->pidmap[i].page);
123 kmem_cache_free(pid_ns_cachep, ns);
124}
125
126struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
127{
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800128 if (!(flags & CLONE_NEWPID))
Alexey Dobriyandca4a972009-06-17 16:27:53 -0700129 return get_pid_ns(old_ns);
Sukadev Bhattiprolue5a47382009-09-23 15:57:22 -0700130 if (flags & (CLONE_THREAD|CLONE_PARENT))
Alexey Dobriyandca4a972009-06-17 16:27:53 -0700131 return ERR_PTR(-EINVAL);
132 return create_pid_namespace(old_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800133}
134
135void free_pid_ns(struct kref *kref)
136{
137 struct pid_namespace *ns, *parent;
138
139 ns = container_of(kref, struct pid_namespace, kref);
140
141 parent = ns->parent;
142 destroy_pid_namespace(ns);
143
144 if (parent != NULL)
145 put_pid_ns(parent);
146}
147
148void zap_pid_ns_processes(struct pid_namespace *pid_ns)
149{
150 int nr;
151 int rc;
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700152 struct task_struct *task;
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800153
154 /*
155 * The last thread in the cgroup-init thread group is terminating.
156 * Find remaining pid_ts in the namespace, signal and wait for them
157 * to exit.
158 *
159 * Note: This signals each threads in the namespace - even those that
160 * belong to the same thread group, To avoid this, we would have
161 * to walk the entire tasklist looking a processes in this
162 * namespace, but that could be unnecessarily expensive if the
163 * pid namespace has just a few processes. Or we need to
164 * maintain a tasklist for each pid namespace.
165 *
166 */
167 read_lock(&tasklist_lock);
168 nr = next_pidmap(pid_ns, 1);
169 while (nr > 0) {
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700170 rcu_read_lock();
171
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700172 task = pid_task(find_vpid(nr), PIDTYPE_PID);
Oleg Nesterova02d6fd2012-03-23 15:02:46 -0700173 if (task && !__fatal_signal_pending(task))
174 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
Sukadev Bhattiprolue4da0262009-04-02 16:58:06 -0700175
176 rcu_read_unlock();
177
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800178 nr = next_pidmap(pid_ns, nr);
179 }
180 read_unlock(&tasklist_lock);
181
182 do {
183 clear_thread_flag(TIF_SIGPENDING);
184 rc = sys_wait4(-1, NULL, __WALL, NULL);
185 } while (rc != -ECHILD);
186
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700187 if (pid_ns->reboot)
188 current->signal->group_exit_code = pid_ns->reboot;
189
Pavel Emelyanov0b6b0302008-07-25 01:48:47 -0700190 acct_exit_ns(pid_ns);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800191 return;
192}
193
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800194static int pid_ns_ctl_handler(struct ctl_table *table, int write,
195 void __user *buffer, size_t *lenp, loff_t *ppos)
196{
197 struct ctl_table tmp = *table;
198
199 if (write && !capable(CAP_SYS_ADMIN))
200 return -EPERM;
201
202 /*
203 * Writing directly to ns' last_pid field is OK, since this field
204 * is volatile in a living namespace anyway and a code writing to
205 * it should synchronize its usage with external means.
206 */
207
208 tmp.data = &current->nsproxy->pid_ns->last_pid;
209 return proc_dointvec(&tmp, write, buffer, lenp, ppos);
210}
211
212static struct ctl_table pid_ns_ctl_table[] = {
213 {
214 .procname = "ns_last_pid",
215 .maxlen = sizeof(int),
216 .mode = 0666, /* permissions are checked in the handler */
217 .proc_handler = pid_ns_ctl_handler,
218 },
219 { }
220};
221
222static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
223
Daniel Lezcanocf3f8922012-03-28 14:42:51 -0700224int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
225{
226 if (pid_ns == &init_pid_ns)
227 return 0;
228
229 switch (cmd) {
230 case LINUX_REBOOT_CMD_RESTART2:
231 case LINUX_REBOOT_CMD_RESTART:
232 pid_ns->reboot = SIGHUP;
233 break;
234
235 case LINUX_REBOOT_CMD_POWER_OFF:
236 case LINUX_REBOOT_CMD_HALT:
237 pid_ns->reboot = SIGINT;
238 break;
239 default:
240 return -EINVAL;
241 }
242
243 read_lock(&tasklist_lock);
244 force_sig(SIGKILL, pid_ns->child_reaper);
245 read_unlock(&tasklist_lock);
246
247 do_exit(0);
248
249 /* Not reached */
250 return 0;
251}
252
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800253static __init int pid_namespaces_init(void)
254{
255 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
Pavel Emelyanovb8f566b2012-01-12 17:20:27 -0800256 register_sysctl_paths(kern_path, pid_ns_ctl_table);
Pavel Emelyanov74bd59b2008-02-08 04:18:24 -0800257 return 0;
258}
259
260__initcall(pid_namespaces_init);