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>
19 #define BITS_PER_PAGE (PAGE_SIZE*8)
24 struct kmem_cache *cachep;
25 struct list_head list;
28 static LIST_HEAD(pid_caches_lh);
29 static DEFINE_MUTEX(pid_caches_mutex);
30 static struct kmem_cache *pid_ns_cachep;
33 * creates the kmem cache to allocate pids from.
34 * @nr_ids: the number of numerical ids this pid will have to carry
37 static struct kmem_cache *create_pid_cachep(int nr_ids)
39 struct pid_cache *pcache;
40 struct kmem_cache *cachep;
42 mutex_lock(&pid_caches_mutex);
43 list_for_each_entry(pcache, &pid_caches_lh, list)
44 if (pcache->nr_ids == nr_ids)
47 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
51 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
52 cachep = kmem_cache_create(pcache->name,
53 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
54 0, SLAB_HWCACHE_ALIGN, NULL);
58 pcache->nr_ids = nr_ids;
59 pcache->cachep = cachep;
60 list_add(&pcache->list, &pid_caches_lh);
62 mutex_unlock(&pid_caches_mutex);
63 return pcache->cachep;
68 mutex_unlock(&pid_caches_mutex);
72 static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
74 struct pid_namespace *ns;
75 unsigned int level = parent_pid_ns->level + 1;
78 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
82 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
83 if (!ns->pidmap[0].page)
86 ns->pid_cachep = create_pid_cachep(level + 1);
87 if (ns->pid_cachep == NULL)
92 ns->parent = get_pid_ns(parent_pid_ns);
94 set_bit(0, ns->pidmap[0].page);
95 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
97 for (i = 1; i < PIDMAP_ENTRIES; i++)
98 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
100 err = pid_ns_prepare_proc(ns);
102 goto out_put_parent_pid_ns;
106 out_put_parent_pid_ns:
107 put_pid_ns(parent_pid_ns);
109 kfree(ns->pidmap[0].page);
111 kmem_cache_free(pid_ns_cachep, ns);
116 static void destroy_pid_namespace(struct pid_namespace *ns)
120 for (i = 0; i < PIDMAP_ENTRIES; i++)
121 kfree(ns->pidmap[i].page);
122 kmem_cache_free(pid_ns_cachep, ns);
125 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
127 if (!(flags & CLONE_NEWPID))
128 return get_pid_ns(old_ns);
129 if (flags & (CLONE_THREAD|CLONE_PARENT))
130 return ERR_PTR(-EINVAL);
131 return create_pid_namespace(old_ns);
134 void free_pid_ns(struct kref *kref)
136 struct pid_namespace *ns, *parent;
138 ns = container_of(kref, struct pid_namespace, kref);
141 destroy_pid_namespace(ns);
147 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
151 struct task_struct *task;
154 * The last thread in the cgroup-init thread group is terminating.
155 * Find remaining pid_ts in the namespace, signal and wait for them
158 * Note: This signals each threads in the namespace - even those that
159 * belong to the same thread group, To avoid this, we would have
160 * to walk the entire tasklist looking a processes in this
161 * namespace, but that could be unnecessarily expensive if the
162 * pid namespace has just a few processes. Or we need to
163 * maintain a tasklist for each pid namespace.
166 read_lock(&tasklist_lock);
167 nr = next_pidmap(pid_ns, 1);
172 * Any nested-container's init processes won't ignore the
173 * SEND_SIG_NOINFO signal, see send_signal()->si_fromuser().
175 task = pid_task(find_vpid(nr), PIDTYPE_PID);
177 send_sig_info(SIGKILL, SEND_SIG_NOINFO, task);
181 nr = next_pidmap(pid_ns, nr);
183 read_unlock(&tasklist_lock);
186 clear_thread_flag(TIF_SIGPENDING);
187 rc = sys_wait4(-1, NULL, __WALL, NULL);
188 } while (rc != -ECHILD);
190 acct_exit_ns(pid_ns);
194 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
195 void __user *buffer, size_t *lenp, loff_t *ppos)
197 struct ctl_table tmp = *table;
199 if (write && !capable(CAP_SYS_ADMIN))
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.
208 tmp.data = ¤t->nsproxy->pid_ns->last_pid;
209 return proc_dointvec(&tmp, write, buffer, lenp, ppos);
212 static struct ctl_table pid_ns_ctl_table[] = {
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,
222 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
224 static __init int pid_namespaces_init(void)
226 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
227 register_sysctl_paths(kern_path, pid_ns_ctl_table);
231 __initcall(pid_namespaces_init);