xref: /linux/ipc/namespace.c (revision 2634682fdffd9ba6e74b76be8aa91cf8b2e05c41)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/ipc/namespace.c
4  * Copyright (C) 2006 Pavel Emelyanov <xemul@openvz.org> OpenVZ, SWsoft Inc.
5  */
6 
7 #include <linux/ipc.h>
8 #include <linux/msg.h>
9 #include <linux/ipc_namespace.h>
10 #include <linux/rcupdate.h>
11 #include <linux/nsproxy.h>
12 #include <linux/slab.h>
13 #include <linux/cred.h>
14 #include <linux/fs.h>
15 #include <linux/mount.h>
16 #include <linux/user_namespace.h>
17 #include <linux/proc_ns.h>
18 #include <linux/sched/task.h>
19 
20 #include "util.h"
21 
22 static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns)
23 {
24 	return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES);
25 }
26 
27 static void dec_ipc_namespaces(struct ucounts *ucounts)
28 {
29 	dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES);
30 }
31 
32 static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
33 					   struct ipc_namespace *old_ns)
34 {
35 	struct ipc_namespace *ns;
36 	struct ucounts *ucounts;
37 	int err;
38 
39 	err = -ENOSPC;
40 	ucounts = inc_ipc_namespaces(user_ns);
41 	if (!ucounts)
42 		goto fail;
43 
44 	err = -ENOMEM;
45 	ns = kzalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
46 	if (ns == NULL)
47 		goto fail_dec;
48 
49 	err = ns_alloc_inum(&ns->ns);
50 	if (err)
51 		goto fail_free;
52 	ns->ns.ops = &ipcns_operations;
53 
54 	refcount_set(&ns->count, 1);
55 	ns->user_ns = get_user_ns(user_ns);
56 	ns->ucounts = ucounts;
57 
58 	err = mq_init_ns(ns);
59 	if (err)
60 		goto fail_put;
61 
62 	sem_init_ns(ns);
63 	msg_init_ns(ns);
64 	shm_init_ns(ns);
65 
66 	return ns;
67 
68 fail_put:
69 	put_user_ns(ns->user_ns);
70 	ns_free_inum(&ns->ns);
71 fail_free:
72 	kfree(ns);
73 fail_dec:
74 	dec_ipc_namespaces(ucounts);
75 fail:
76 	return ERR_PTR(err);
77 }
78 
79 struct ipc_namespace *copy_ipcs(unsigned long flags,
80 	struct user_namespace *user_ns, struct ipc_namespace *ns)
81 {
82 	if (!(flags & CLONE_NEWIPC))
83 		return get_ipc_ns(ns);
84 	return create_ipc_ns(user_ns, ns);
85 }
86 
87 /*
88  * free_ipcs - free all ipcs of one type
89  * @ns:   the namespace to remove the ipcs from
90  * @ids:  the table of ipcs to free
91  * @free: the function called to free each individual ipc
92  *
93  * Called for each kind of ipc when an ipc_namespace exits.
94  */
95 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
96 	       void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
97 {
98 	struct kern_ipc_perm *perm;
99 	int next_id;
100 	int total, in_use;
101 
102 	down_write(&ids->rwsem);
103 
104 	in_use = ids->in_use;
105 
106 	for (total = 0, next_id = 0; total < in_use; next_id++) {
107 		perm = idr_find(&ids->ipcs_idr, next_id);
108 		if (perm == NULL)
109 			continue;
110 		rcu_read_lock();
111 		ipc_lock_object(perm);
112 		free(ns, perm);
113 		total++;
114 	}
115 	up_write(&ids->rwsem);
116 }
117 
118 static void free_ipc_ns(struct ipc_namespace *ns)
119 {
120 	/* mq_put_mnt() waits for a grace period as kern_unmount()
121 	 * uses synchronize_rcu().
122 	 */
123 	mq_put_mnt(ns);
124 	sem_exit_ns(ns);
125 	msg_exit_ns(ns);
126 	shm_exit_ns(ns);
127 
128 	dec_ipc_namespaces(ns->ucounts);
129 	put_user_ns(ns->user_ns);
130 	ns_free_inum(&ns->ns);
131 	kfree(ns);
132 }
133 
134 static LLIST_HEAD(free_ipc_list);
135 static void free_ipc(struct work_struct *unused)
136 {
137 	struct llist_node *node = llist_del_all(&free_ipc_list);
138 	struct ipc_namespace *n, *t;
139 
140 	llist_for_each_entry_safe(n, t, node, mnt_llist)
141 		free_ipc_ns(n);
142 }
143 
144 /*
145  * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
146  */
147 static DECLARE_WORK(free_ipc_work, free_ipc);
148 
149 /*
150  * put_ipc_ns - drop a reference to an ipc namespace.
151  * @ns: the namespace to put
152  *
153  * If this is the last task in the namespace exiting, and
154  * it is dropping the refcount to 0, then it can race with
155  * a task in another ipc namespace but in a mounts namespace
156  * which has this ipcns's mqueuefs mounted, doing some action
157  * with one of the mqueuefs files.  That can raise the refcount.
158  * So dropping the refcount, and raising the refcount when
159  * accessing it through the VFS, are protected with mq_lock.
160  *
161  * (Clearly, a task raising the refcount on its own ipc_ns
162  * needn't take mq_lock since it can't race with the last task
163  * in the ipcns exiting).
164  */
165 void put_ipc_ns(struct ipc_namespace *ns)
166 {
167 	if (refcount_dec_and_lock(&ns->count, &mq_lock)) {
168 		mq_clear_sbinfo(ns);
169 		spin_unlock(&mq_lock);
170 
171 		if (llist_add(&ns->mnt_llist, &free_ipc_list))
172 			schedule_work(&free_ipc_work);
173 	}
174 }
175 
176 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
177 {
178 	return container_of(ns, struct ipc_namespace, ns);
179 }
180 
181 static struct ns_common *ipcns_get(struct task_struct *task)
182 {
183 	struct ipc_namespace *ns = NULL;
184 	struct nsproxy *nsproxy;
185 
186 	task_lock(task);
187 	nsproxy = task->nsproxy;
188 	if (nsproxy)
189 		ns = get_ipc_ns(nsproxy->ipc_ns);
190 	task_unlock(task);
191 
192 	return ns ? &ns->ns : NULL;
193 }
194 
195 static void ipcns_put(struct ns_common *ns)
196 {
197 	return put_ipc_ns(to_ipc_ns(ns));
198 }
199 
200 static int ipcns_install(struct nsset *nsset, struct ns_common *new)
201 {
202 	struct nsproxy *nsproxy = nsset->nsproxy;
203 	struct ipc_namespace *ns = to_ipc_ns(new);
204 	if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
205 	    !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
206 		return -EPERM;
207 
208 	put_ipc_ns(nsproxy->ipc_ns);
209 	nsproxy->ipc_ns = get_ipc_ns(ns);
210 	return 0;
211 }
212 
213 static struct user_namespace *ipcns_owner(struct ns_common *ns)
214 {
215 	return to_ipc_ns(ns)->user_ns;
216 }
217 
218 const struct proc_ns_operations ipcns_operations = {
219 	.name		= "ipc",
220 	.type		= CLONE_NEWIPC,
221 	.get		= ipcns_get,
222 	.put		= ipcns_put,
223 	.install	= ipcns_install,
224 	.owner		= ipcns_owner,
225 };
226