xref: /openbmc/linux/kernel/pid_namespace.c (revision 545e4006)
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 
16 #define BITS_PER_PAGE		(PAGE_SIZE*8)
17 
18 struct pid_cache {
19 	int nr_ids;
20 	char name[16];
21 	struct kmem_cache *cachep;
22 	struct list_head list;
23 };
24 
25 static LIST_HEAD(pid_caches_lh);
26 static DEFINE_MUTEX(pid_caches_mutex);
27 static struct kmem_cache *pid_ns_cachep;
28 
29 /*
30  * creates the kmem cache to allocate pids from.
31  * @nr_ids: the number of numerical ids this pid will have to carry
32  */
33 
34 static struct kmem_cache *create_pid_cachep(int nr_ids)
35 {
36 	struct pid_cache *pcache;
37 	struct kmem_cache *cachep;
38 
39 	mutex_lock(&pid_caches_mutex);
40 	list_for_each_entry(pcache, &pid_caches_lh, list)
41 		if (pcache->nr_ids == nr_ids)
42 			goto out;
43 
44 	pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
45 	if (pcache == NULL)
46 		goto err_alloc;
47 
48 	snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
49 	cachep = kmem_cache_create(pcache->name,
50 			sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
51 			0, SLAB_HWCACHE_ALIGN, NULL);
52 	if (cachep == NULL)
53 		goto err_cachep;
54 
55 	pcache->nr_ids = nr_ids;
56 	pcache->cachep = cachep;
57 	list_add(&pcache->list, &pid_caches_lh);
58 out:
59 	mutex_unlock(&pid_caches_mutex);
60 	return pcache->cachep;
61 
62 err_cachep:
63 	kfree(pcache);
64 err_alloc:
65 	mutex_unlock(&pid_caches_mutex);
66 	return NULL;
67 }
68 
69 static struct pid_namespace *create_pid_namespace(unsigned int level)
70 {
71 	struct pid_namespace *ns;
72 	int i;
73 
74 	ns = kmem_cache_alloc(pid_ns_cachep, GFP_KERNEL);
75 	if (ns == NULL)
76 		goto out;
77 
78 	ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
79 	if (!ns->pidmap[0].page)
80 		goto out_free;
81 
82 	ns->pid_cachep = create_pid_cachep(level + 1);
83 	if (ns->pid_cachep == NULL)
84 		goto out_free_map;
85 
86 	kref_init(&ns->kref);
87 	ns->last_pid = 0;
88 	ns->child_reaper = NULL;
89 	ns->level = level;
90 
91 	set_bit(0, ns->pidmap[0].page);
92 	atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
93 
94 	for (i = 1; i < PIDMAP_ENTRIES; i++) {
95 		ns->pidmap[i].page = NULL;
96 		atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
97 	}
98 
99 	return ns;
100 
101 out_free_map:
102 	kfree(ns->pidmap[0].page);
103 out_free:
104 	kmem_cache_free(pid_ns_cachep, ns);
105 out:
106 	return ERR_PTR(-ENOMEM);
107 }
108 
109 static 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 
118 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
119 {
120 	struct pid_namespace *new_ns;
121 
122 	BUG_ON(!old_ns);
123 	new_ns = get_pid_ns(old_ns);
124 	if (!(flags & CLONE_NEWPID))
125 		goto out;
126 
127 	new_ns = ERR_PTR(-EINVAL);
128 	if (flags & CLONE_THREAD)
129 		goto out_put;
130 
131 	new_ns = create_pid_namespace(old_ns->level + 1);
132 	if (!IS_ERR(new_ns))
133 		new_ns->parent = get_pid_ns(old_ns);
134 
135 out_put:
136 	put_pid_ns(old_ns);
137 out:
138 	return new_ns;
139 }
140 
141 void free_pid_ns(struct kref *kref)
142 {
143 	struct pid_namespace *ns, *parent;
144 
145 	ns = container_of(kref, struct pid_namespace, kref);
146 
147 	parent = ns->parent;
148 	destroy_pid_namespace(ns);
149 
150 	if (parent != NULL)
151 		put_pid_ns(parent);
152 }
153 
154 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
155 {
156 	int nr;
157 	int rc;
158 
159 	/*
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
162 	 * to exit.
163 	 *
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.
170 	 *
171 	 */
172 	read_lock(&tasklist_lock);
173 	nr = next_pidmap(pid_ns, 1);
174 	while (nr > 0) {
175 		kill_proc_info(SIGKILL, SEND_SIG_PRIV, nr);
176 		nr = next_pidmap(pid_ns, nr);
177 	}
178 	read_unlock(&tasklist_lock);
179 
180 	do {
181 		clear_thread_flag(TIF_SIGPENDING);
182 		rc = sys_wait4(-1, NULL, __WALL, NULL);
183 	} while (rc != -ECHILD);
184 
185 
186 	/* Child reaper for the pid namespace is going away */
187 	pid_ns->child_reaper = NULL;
188 	return;
189 }
190 
191 static __init int pid_namespaces_init(void)
192 {
193 	pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
194 	return 0;
195 }
196 
197 __initcall(pid_namespaces_init);
198