1/*
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>
15#include <linux/acct.h>
16#include <linux/slab.h>
17
18#define BITS_PER_PAGE		(PAGE_SIZE*8)
19
20struct pid_cache {
21	int nr_ids;
22	char name[16];
23	struct kmem_cache *cachep;
24	struct list_head list;
25};
26
27static LIST_HEAD(pid_caches_lh);
28static DEFINE_MUTEX(pid_caches_mutex);
29static struct kmem_cache *pid_ns_cachep;
30
31/*
32 * creates the kmem cache to allocate pids from.
33 * @nr_ids: the number of numerical ids this pid will have to carry
34 */
35
36static struct kmem_cache *create_pid_cachep(int nr_ids)
37{
38	struct pid_cache *pcache;
39	struct kmem_cache *cachep;
40
41	mutex_lock(&pid_caches_mutex);
42	list_for_each_entry(pcache, &pid_caches_lh, list)
43		if (pcache->nr_ids == nr_ids)
44			goto out;
45
46	pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
47	if (pcache == NULL)
48		goto err_alloc;
49
50	snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
51	cachep = kmem_cache_create(pcache->name,
52			sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
53			0, SLAB_HWCACHE_ALIGN, NULL);
54	if (cachep == NULL)
55		goto err_cachep;
56
57	pcache->nr_ids = nr_ids;
58	pcache->cachep = cachep;
59	list_add(&pcache->list, &pid_caches_lh);
60out:
61	mutex_unlock(&pid_caches_mutex);
62	return pcache->cachep;
63
64err_cachep:
65	kfree(pcache);
66err_alloc:
67	mutex_unlock(&pid_caches_mutex);
68	return NULL;
69}
70
71static struct pid_namespace *create_pid_namespace(struct pid_namespace *parent_pid_ns)
72{
73	struct pid_namespace *ns;
74	unsigned int level = parent_pid_ns->level + 1;
75	int i;
76
77	ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
78	if (ns == NULL)
79		goto out;
80
81	ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
82	if (!ns->pidmap[0].page)
83		goto out_free;
84
85	ns->pid_cachep = create_pid_cachep(level + 1);
86	if (ns->pid_cachep == NULL)
87		goto out_free_map;
88
89	kref_init(&ns->kref);
90	ns->level = level;
91	ns->parent = get_pid_ns(parent_pid_ns);
92
93	set_bit(0, ns->pidmap[0].page);
94	atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
95
96	for (i = 1; i < PIDMAP_ENTRIES; i++)
97		atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
98
99	return ns;
100
101out_free_map:
102	kfree(ns->pidmap[0].page);
103out_free:
104	kmem_cache_free(pid_ns_cachep, ns);
105out:
106	return ERR_PTR(-ENOMEM);
107}
108
109static void destroy_pid_namespace(struct pid_namespace *ns)
110{
111	int i;
112
113	for (i = 0; i < PIDMAP_ENTRIES; i++)
114		kfree(ns->pidmap[i].page);
115	kmem_cache_free(pid_ns_cachep, ns);
116}
117
118struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
119{
120	if (!(flags & CLONE_NEWPID))
121		return get_pid_ns(old_ns);
122	if (flags & (CLONE_THREAD|CLONE_PARENT))
123		return ERR_PTR(-EINVAL);
124	return create_pid_namespace(old_ns);
125}
126
127void free_pid_ns(struct kref *kref)
128{
129	struct pid_namespace *ns, *parent;
130
131	ns = container_of(kref, struct pid_namespace, kref);
132
133	parent = ns->parent;
134	destroy_pid_namespace(ns);
135
136	if (parent != NULL)
137		put_pid_ns(parent);
138}
139
140void zap_pid_ns_processes(struct pid_namespace *pid_ns)
141{
142	int nr;
143	int rc;
144	struct task_struct *task;
145
146	/*
147	 * The last thread in the cgroup-init thread group is terminating.
148	 * Find remaining pid_ts in the namespace, signal and wait for them
149	 * to exit.
150	 *
151	 * Note:  This signals each threads in the namespace - even those that
152	 * 	  belong to the same thread group, To avoid this, we would have
153	 * 	  to walk the entire tasklist looking a processes in this
154	 * 	  namespace, but that could be unnecessarily expensive if the
155	 * 	  pid namespace has just a few processes. Or we need to
156	 * 	  maintain a tasklist for each pid namespace.
157	 *
158	 */
159	read_lock(&tasklist_lock);
160	nr = next_pidmap(pid_ns, 1);
161	while (nr > 0) {
162		rcu_read_lock();
163
164		/*
165		 * Any nested-container's init processes won't ignore the
166		 * SEND_SIG_NOINFO signal, see send_signal()->si_fromuser().
167		 */
168		task = pid_task(find_vpid(nr), PIDTYPE_PID);
169		if (task)
170			send_sig_info(SIGKILL, SEND_SIG_NOINFO, task);
171
172		rcu_read_unlock();
173
174		nr = next_pidmap(pid_ns, nr);
175	}
176	read_unlock(&tasklist_lock);
177
178	do {
179		clear_thread_flag(TIF_SIGPENDING);
180		rc = sys_wait4(-1, NULL, __WALL, NULL);
181	} while (rc != -ECHILD);
182
183	acct_exit_ns(pid_ns);
184	return;
185}
186
187static __init int pid_namespaces_init(void)
188{
189	pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
190	return 0;
191}
192
193__initcall(pid_namespaces_init);
194