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>
19 #include <linux/export.h>
21 #define BITS_PER_PAGE (PAGE_SIZE*8)
26 struct kmem_cache *cachep;
27 struct list_head list;
30 static LIST_HEAD(pid_caches_lh);
31 static DEFINE_MUTEX(pid_caches_mutex);
32 static struct kmem_cache *pid_ns_cachep;
35 * creates the kmem cache to allocate pids from.
36 * @nr_ids: the number of numerical ids this pid will have to carry
39 static struct kmem_cache *create_pid_cachep(int nr_ids)
41 struct pid_cache *pcache;
42 struct kmem_cache *cachep;
44 mutex_lock(&pid_caches_mutex);
45 list_for_each_entry(pcache, &pid_caches_lh, list)
46 if (pcache->nr_ids == nr_ids)
49 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
53 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
54 cachep = kmem_cache_create(pcache->name,
55 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
56 0, SLAB_HWCACHE_ALIGN, NULL);
60 pcache->nr_ids = nr_ids;
61 pcache->cachep = cachep;
62 list_add(&pcache->list, &pid_caches_lh);
64 mutex_unlock(&pid_caches_mutex);
65 return pcache->cachep;
70 mutex_unlock(&pid_caches_mutex);
74 static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
76 struct pid_namespace *ns;
77 unsigned int level = parent_pid_ns->level + 1;
80 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
84 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
85 if (!ns->pidmap[0].page)
88 ns->pid_cachep = create_pid_cachep(level + 1);
89 if (ns->pid_cachep == NULL)
94 ns->parent = get_pid_ns(parent_pid_ns);
96 set_bit(0, ns->pidmap[0].page);
97 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
99 for (i = 1; i < PIDMAP_ENTRIES; i++)
100 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
102 err = pid_ns_prepare_proc(ns);
104 goto out_put_parent_pid_ns;
108 out_put_parent_pid_ns:
109 put_pid_ns(parent_pid_ns);
111 kfree(ns->pidmap[0].page);
113 kmem_cache_free(pid_ns_cachep, ns);
118 static void destroy_pid_namespace(struct pid_namespace *ns)
122 for (i = 0; i < PIDMAP_ENTRIES; i++)
123 kfree(ns->pidmap[i].page);
124 kmem_cache_free(pid_ns_cachep, ns);
127 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
129 if (!(flags & CLONE_NEWPID))
130 return get_pid_ns(old_ns);
131 if (flags & (CLONE_THREAD|CLONE_PARENT))
132 return ERR_PTR(-EINVAL);
133 return create_pid_namespace(old_ns);
136 void free_pid_ns(struct kref *kref)
138 struct pid_namespace *ns, *parent;
140 ns = container_of(kref, struct pid_namespace, kref);
143 destroy_pid_namespace(ns);
148 EXPORT_SYMBOL_GPL(free_pid_ns);
150 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
154 struct task_struct *task, *me = current;
156 /* Ignore SIGCHLD causing any terminated children to autoreap */
157 spin_lock_irq(&me->sighand->siglock);
158 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
159 spin_unlock_irq(&me->sighand->siglock);
162 * The last thread in the cgroup-init thread group is terminating.
163 * Find remaining pid_ts in the namespace, signal and wait for them
166 * Note: This signals each threads in the namespace - even those that
167 * belong to the same thread group, To avoid this, we would have
168 * to walk the entire tasklist looking a processes in this
169 * namespace, but that could be unnecessarily expensive if the
170 * pid namespace has just a few processes. Or we need to
171 * maintain a tasklist for each pid namespace.
174 read_lock(&tasklist_lock);
175 nr = next_pidmap(pid_ns, 1);
179 task = pid_task(find_vpid(nr), PIDTYPE_PID);
180 if (task && !__fatal_signal_pending(task))
181 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
185 nr = next_pidmap(pid_ns, nr);
187 read_unlock(&tasklist_lock);
189 /* Firstly reap the EXIT_ZOMBIE children we may have. */
191 clear_thread_flag(TIF_SIGPENDING);
192 rc = sys_wait4(-1, NULL, __WALL, NULL);
193 } while (rc != -ECHILD);
196 * sys_wait4() above can't reap the TASK_DEAD children.
197 * Make sure they all go away, see __unhash_process().
200 bool need_wait = false;
202 read_lock(&tasklist_lock);
203 if (!list_empty(¤t->children)) {
204 __set_current_state(TASK_UNINTERRUPTIBLE);
207 read_unlock(&tasklist_lock);
215 current->signal->group_exit_code = pid_ns->reboot;
217 acct_exit_ns(pid_ns);
221 #ifdef CONFIG_CHECKPOINT_RESTORE
222 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
223 void __user *buffer, size_t *lenp, loff_t *ppos)
225 struct ctl_table tmp = *table;
227 if (write && !capable(CAP_SYS_ADMIN))
231 * Writing directly to ns' last_pid field is OK, since this field
232 * is volatile in a living namespace anyway and a code writing to
233 * it should synchronize its usage with external means.
236 tmp.data = ¤t->nsproxy->pid_ns->last_pid;
237 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
242 static struct ctl_table pid_ns_ctl_table[] = {
244 .procname = "ns_last_pid",
245 .maxlen = sizeof(int),
246 .mode = 0666, /* permissions are checked in the handler */
247 .proc_handler = pid_ns_ctl_handler,
253 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
254 #endif /* CONFIG_CHECKPOINT_RESTORE */
256 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
258 if (pid_ns == &init_pid_ns)
262 case LINUX_REBOOT_CMD_RESTART2:
263 case LINUX_REBOOT_CMD_RESTART:
264 pid_ns->reboot = SIGHUP;
267 case LINUX_REBOOT_CMD_POWER_OFF:
268 case LINUX_REBOOT_CMD_HALT:
269 pid_ns->reboot = SIGINT;
275 read_lock(&tasklist_lock);
276 force_sig(SIGKILL, pid_ns->child_reaper);
277 read_unlock(&tasklist_lock);
285 static __init int pid_namespaces_init(void)
287 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
289 #ifdef CONFIG_CHECKPOINT_RESTORE
290 register_sysctl_paths(kern_path, pid_ns_ctl_table);
295 __initcall(pid_namespaces_init);