namespace.c revision 5e4a0847
1/*
2 * linux/ipc/namespace.c
3 * Copyright (C) 2006 Pavel Emelyanov <xemul@openvz.org> OpenVZ, SWsoft Inc.
4 */
5
6#include <linux/ipc.h>
7#include <linux/msg.h>
8#include <linux/ipc_namespace.h>
9#include <linux/rcupdate.h>
10#include <linux/nsproxy.h>
11#include <linux/slab.h>
12#include <linux/fs.h>
13#include <linux/mount.h>
14#include <linux/user_namespace.h>
15#include <linux/proc_fs.h>
16
17#include "util.h"
18
19static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
20					   struct ipc_namespace *old_ns)
21{
22	struct ipc_namespace *ns;
23	int err;
24
25	ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
26	if (ns == NULL)
27		return ERR_PTR(-ENOMEM);
28
29	err = proc_alloc_inum(&ns->proc_inum);
30	if (err) {
31		kfree(ns);
32		return ERR_PTR(err);
33	}
34
35	atomic_set(&ns->count, 1);
36	err = mq_init_ns(ns);
37	if (err) {
38		proc_free_inum(ns->proc_inum);
39		kfree(ns);
40		return ERR_PTR(err);
41	}
42	atomic_inc(&nr_ipc_ns);
43
44	sem_init_ns(ns);
45	msg_init_ns(ns);
46	shm_init_ns(ns);
47
48	/*
49	 * msgmni has already been computed for the new ipc ns.
50	 * Thus, do the ipcns creation notification before registering that
51	 * new ipcns in the chain.
52	 */
53	ipcns_notify(IPCNS_CREATED);
54	register_ipcns_notifier(ns);
55
56	ns->user_ns = get_user_ns(user_ns);
57
58	return ns;
59}
60
61struct ipc_namespace *copy_ipcs(unsigned long flags,
62	struct user_namespace *user_ns, struct ipc_namespace *ns)
63{
64	if (!(flags & CLONE_NEWIPC))
65		return get_ipc_ns(ns);
66	return create_ipc_ns(user_ns, ns);
67}
68
69/*
70 * free_ipcs - free all ipcs of one type
71 * @ns:   the namespace to remove the ipcs from
72 * @ids:  the table of ipcs to free
73 * @free: the function called to free each individual ipc
74 *
75 * Called for each kind of ipc when an ipc_namespace exits.
76 */
77void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
78	       void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
79{
80	struct kern_ipc_perm *perm;
81	int next_id;
82	int total, in_use;
83
84	down_write(&ids->rw_mutex);
85
86	in_use = ids->in_use;
87
88	for (total = 0, next_id = 0; total < in_use; next_id++) {
89		perm = idr_find(&ids->ipcs_idr, next_id);
90		if (perm == NULL)
91			continue;
92		ipc_lock_by_ptr(perm);
93		free(ns, perm);
94		total++;
95	}
96	up_write(&ids->rw_mutex);
97}
98
99static void free_ipc_ns(struct ipc_namespace *ns)
100{
101	/*
102	 * Unregistering the hotplug notifier at the beginning guarantees
103	 * that the ipc namespace won't be freed while we are inside the
104	 * callback routine. Since the blocking_notifier_chain_XXX routines
105	 * hold a rw lock on the notifier list, unregister_ipcns_notifier()
106	 * won't take the rw lock before blocking_notifier_call_chain() has
107	 * released the rd lock.
108	 */
109	unregister_ipcns_notifier(ns);
110	sem_exit_ns(ns);
111	msg_exit_ns(ns);
112	shm_exit_ns(ns);
113	atomic_dec(&nr_ipc_ns);
114
115	/*
116	 * Do the ipcns removal notification after decrementing nr_ipc_ns in
117	 * order to have a correct value when recomputing msgmni.
118	 */
119	ipcns_notify(IPCNS_REMOVED);
120	put_user_ns(ns->user_ns);
121	proc_free_inum(ns->proc_inum);
122	kfree(ns);
123}
124
125/*
126 * put_ipc_ns - drop a reference to an ipc namespace.
127 * @ns: the namespace to put
128 *
129 * If this is the last task in the namespace exiting, and
130 * it is dropping the refcount to 0, then it can race with
131 * a task in another ipc namespace but in a mounts namespace
132 * which has this ipcns's mqueuefs mounted, doing some action
133 * with one of the mqueuefs files.  That can raise the refcount.
134 * So dropping the refcount, and raising the refcount when
135 * accessing it through the VFS, are protected with mq_lock.
136 *
137 * (Clearly, a task raising the refcount on its own ipc_ns
138 * needn't take mq_lock since it can't race with the last task
139 * in the ipcns exiting).
140 */
141void put_ipc_ns(struct ipc_namespace *ns)
142{
143	if (atomic_dec_and_lock(&ns->count, &mq_lock)) {
144		mq_clear_sbinfo(ns);
145		spin_unlock(&mq_lock);
146		mq_put_mnt(ns);
147		free_ipc_ns(ns);
148	}
149}
150
151static void *ipcns_get(struct task_struct *task)
152{
153	struct ipc_namespace *ns = NULL;
154	struct nsproxy *nsproxy;
155
156	rcu_read_lock();
157	nsproxy = task_nsproxy(task);
158	if (nsproxy)
159		ns = get_ipc_ns(nsproxy->ipc_ns);
160	rcu_read_unlock();
161
162	return ns;
163}
164
165static void ipcns_put(void *ns)
166{
167	return put_ipc_ns(ns);
168}
169
170static int ipcns_install(struct nsproxy *nsproxy, void *new)
171{
172	struct ipc_namespace *ns = new;
173	if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
174	    !nsown_capable(CAP_SYS_ADMIN))
175		return -EPERM;
176
177	/* Ditch state from the old ipc namespace */
178	exit_sem(current);
179	put_ipc_ns(nsproxy->ipc_ns);
180	nsproxy->ipc_ns = get_ipc_ns(ns);
181	return 0;
182}
183
184static unsigned int ipcns_inum(void *vp)
185{
186	struct ipc_namespace *ns = vp;
187
188	return ns->proc_inum;
189}
190
191const struct proc_ns_operations ipcns_operations = {
192	.name		= "ipc",
193	.type		= CLONE_NEWIPC,
194	.get		= ipcns_get,
195	.put		= ipcns_put,
196	.install	= ipcns_install,
197	.inum		= ipcns_inum,
198};
199