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
11 #include <linux/pid.h>
12 #include <linux/pid_namespace.h>
13 #include <linux/syscalls.h>
14 #include <linux/err.h>
15 #include <linux/acct.h>
16 #include <linux/slab.h>
17 #include <linux/proc_fs.h>
18 #include <linux/reboot.h>
20 #define BITS_PER_PAGE (PAGE_SIZE*8)
25 struct kmem_cache *cachep;
26 struct list_head list;
29 static LIST_HEAD(pid_caches_lh);
30 static DEFINE_MUTEX(pid_caches_mutex);
31 static struct kmem_cache *pid_ns_cachep;
34 * creates the kmem cache to allocate pids from.
35 * @nr_ids: the number of numerical ids this pid will have to carry
38 static struct kmem_cache *create_pid_cachep(int nr_ids)
40 struct pid_cache *pcache;
41 struct kmem_cache *cachep;
43 mutex_lock(&pid_caches_mutex);
44 list_for_each_entry(pcache, &pid_caches_lh, list)
45 if (pcache->nr_ids == nr_ids)
48 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
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);
59 pcache->nr_ids = nr_ids;
60 pcache->cachep = cachep;
61 list_add(&pcache->list, &pid_caches_lh);
63 mutex_unlock(&pid_caches_mutex);
64 return pcache->cachep;
69 mutex_unlock(&pid_caches_mutex);
73 static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
75 struct pid_namespace *ns;
76 unsigned int level = parent_pid_ns->level + 1;
79 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
83 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
84 if (!ns->pidmap[0].page)
87 ns->pid_cachep = create_pid_cachep(level + 1);
88 if (ns->pid_cachep == NULL)
93 ns->parent = get_pid_ns(parent_pid_ns);
95 set_bit(0, ns->pidmap[0].page);
96 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
98 for (i = 1; i < PIDMAP_ENTRIES; i++)
99 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
101 err = pid_ns_prepare_proc(ns);
103 goto out_put_parent_pid_ns;
107 out_put_parent_pid_ns:
108 put_pid_ns(parent_pid_ns);
110 kfree(ns->pidmap[0].page);
112 kmem_cache_free(pid_ns_cachep, ns);
117 static void destroy_pid_namespace(struct pid_namespace *ns)
121 for (i = 0; i < PIDMAP_ENTRIES; i++)
122 kfree(ns->pidmap[i].page);
123 kmem_cache_free(pid_ns_cachep, ns);
126 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
128 if (!(flags & CLONE_NEWPID))
129 return get_pid_ns(old_ns);
130 if (flags & (CLONE_THREAD|CLONE_PARENT))
131 return ERR_PTR(-EINVAL);
132 return create_pid_namespace(old_ns);
135 void free_pid_ns(struct kref *kref)
137 struct pid_namespace *ns, *parent;
139 ns = container_of(kref, struct pid_namespace, kref);
142 destroy_pid_namespace(ns);
148 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
152 struct task_struct *task, *me = current;
154 /* Ignore SIGCHLD causing any terminated children to autoreap */
155 spin_lock_irq(&me->sighand->siglock);
156 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
157 spin_unlock_irq(&me->sighand->siglock);
160 * The last thread in the cgroup-init thread group is terminating.
161 * Find remaining pid_ts in the namespace, signal and wait for them
164 * Note: This signals each threads in the namespace - even those that
165 * belong to the same thread group, To avoid this, we would have
166 * to walk the entire tasklist looking a processes in this
167 * namespace, but that could be unnecessarily expensive if the
168 * pid namespace has just a few processes. Or we need to
169 * maintain a tasklist for each pid namespace.
172 read_lock(&tasklist_lock);
173 nr = next_pidmap(pid_ns, 1);
177 task = pid_task(find_vpid(nr), PIDTYPE_PID);
178 if (task && !__fatal_signal_pending(task))
179 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
183 nr = next_pidmap(pid_ns, nr);
185 read_unlock(&tasklist_lock);
187 /* Firstly reap the EXIT_ZOMBIE children we may have. */
189 clear_thread_flag(TIF_SIGPENDING);
190 rc = sys_wait4(-1, NULL, __WALL, NULL);
191 } while (rc != -ECHILD);
194 * sys_wait4() above can't reap the TASK_DEAD children.
195 * Make sure they all go away, see __unhash_process().
198 bool need_wait = false;
200 read_lock(&tasklist_lock);
201 if (!list_empty(¤t->children)) {
202 __set_current_state(TASK_UNINTERRUPTIBLE);
205 read_unlock(&tasklist_lock);
213 current->signal->group_exit_code = pid_ns->reboot;
215 acct_exit_ns(pid_ns);
219 #ifdef CONFIG_CHECKPOINT_RESTORE
220 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
221 void __user *buffer, size_t *lenp, loff_t *ppos)
223 struct ctl_table tmp = *table;
225 if (write && !capable(CAP_SYS_ADMIN))
229 * Writing directly to ns' last_pid field is OK, since this field
230 * is volatile in a living namespace anyway and a code writing to
231 * it should synchronize its usage with external means.
234 tmp.data = ¤t->nsproxy->pid_ns->last_pid;
235 return proc_dointvec(&tmp, write, buffer, lenp, ppos);
238 static struct ctl_table pid_ns_ctl_table[] = {
240 .procname = "ns_last_pid",
241 .maxlen = sizeof(int),
242 .mode = 0666, /* permissions are checked in the handler */
243 .proc_handler = pid_ns_ctl_handler,
247 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
248 #endif /* CONFIG_CHECKPOINT_RESTORE */
250 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
252 if (pid_ns == &init_pid_ns)
256 case LINUX_REBOOT_CMD_RESTART2:
257 case LINUX_REBOOT_CMD_RESTART:
258 pid_ns->reboot = SIGHUP;
261 case LINUX_REBOOT_CMD_POWER_OFF:
262 case LINUX_REBOOT_CMD_HALT:
263 pid_ns->reboot = SIGINT;
269 read_lock(&tasklist_lock);
270 force_sig(SIGKILL, pid_ns->child_reaper);
271 read_unlock(&tasklist_lock);
279 static __init int pid_namespaces_init(void)
281 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
283 #ifdef CONFIG_CHECKPOINT_RESTORE
284 register_sysctl_paths(kern_path, pid_ns_ctl_table);
289 __initcall(pid_namespaces_init);