xref: /linux/ipc/namespace.c (revision c98be0c96db00e9b6b02d31e0fa7590c54cdaaac)
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_ns.h>
16 
17 #include "util.h"
18 
19 static 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 
61 struct 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  */
77 void 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->rwsem);
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 		rcu_read_lock();
93 		ipc_lock_object(perm);
94 		free(ns, perm);
95 		total++;
96 	}
97 	up_write(&ids->rwsem);
98 }
99 
100 static void free_ipc_ns(struct ipc_namespace *ns)
101 {
102 	/*
103 	 * Unregistering the hotplug notifier at the beginning guarantees
104 	 * that the ipc namespace won't be freed while we are inside the
105 	 * callback routine. Since the blocking_notifier_chain_XXX routines
106 	 * hold a rw lock on the notifier list, unregister_ipcns_notifier()
107 	 * won't take the rw lock before blocking_notifier_call_chain() has
108 	 * released the rd lock.
109 	 */
110 	unregister_ipcns_notifier(ns);
111 	sem_exit_ns(ns);
112 	msg_exit_ns(ns);
113 	shm_exit_ns(ns);
114 	atomic_dec(&nr_ipc_ns);
115 
116 	/*
117 	 * Do the ipcns removal notification after decrementing nr_ipc_ns in
118 	 * order to have a correct value when recomputing msgmni.
119 	 */
120 	ipcns_notify(IPCNS_REMOVED);
121 	put_user_ns(ns->user_ns);
122 	proc_free_inum(ns->proc_inum);
123 	kfree(ns);
124 }
125 
126 /*
127  * put_ipc_ns - drop a reference to an ipc namespace.
128  * @ns: the namespace to put
129  *
130  * If this is the last task in the namespace exiting, and
131  * it is dropping the refcount to 0, then it can race with
132  * a task in another ipc namespace but in a mounts namespace
133  * which has this ipcns's mqueuefs mounted, doing some action
134  * with one of the mqueuefs files.  That can raise the refcount.
135  * So dropping the refcount, and raising the refcount when
136  * accessing it through the VFS, are protected with mq_lock.
137  *
138  * (Clearly, a task raising the refcount on its own ipc_ns
139  * needn't take mq_lock since it can't race with the last task
140  * in the ipcns exiting).
141  */
142 void put_ipc_ns(struct ipc_namespace *ns)
143 {
144 	if (atomic_dec_and_lock(&ns->count, &mq_lock)) {
145 		mq_clear_sbinfo(ns);
146 		spin_unlock(&mq_lock);
147 		mq_put_mnt(ns);
148 		free_ipc_ns(ns);
149 	}
150 }
151 
152 static void *ipcns_get(struct task_struct *task)
153 {
154 	struct ipc_namespace *ns = NULL;
155 	struct nsproxy *nsproxy;
156 
157 	rcu_read_lock();
158 	nsproxy = task_nsproxy(task);
159 	if (nsproxy)
160 		ns = get_ipc_ns(nsproxy->ipc_ns);
161 	rcu_read_unlock();
162 
163 	return ns;
164 }
165 
166 static void ipcns_put(void *ns)
167 {
168 	return put_ipc_ns(ns);
169 }
170 
171 static int ipcns_install(struct nsproxy *nsproxy, void *new)
172 {
173 	struct ipc_namespace *ns = new;
174 	if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
175 	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
176 		return -EPERM;
177 
178 	/* Ditch state from the old ipc namespace */
179 	exit_sem(current);
180 	put_ipc_ns(nsproxy->ipc_ns);
181 	nsproxy->ipc_ns = get_ipc_ns(ns);
182 	return 0;
183 }
184 
185 static unsigned int ipcns_inum(void *vp)
186 {
187 	struct ipc_namespace *ns = vp;
188 
189 	return ns->proc_inum;
190 }
191 
192 const struct proc_ns_operations ipcns_operations = {
193 	.name		= "ipc",
194 	.type		= CLONE_NEWIPC,
195 	.get		= ipcns_get,
196 	.put		= ipcns_put,
197 	.install	= ipcns_install,
198 	.inum		= ipcns_inum,
199 };
200