Loading...
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
19static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns)
20{
21 return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES);
22}
23
24static void dec_ipc_namespaces(struct ucounts *ucounts)
25{
26 dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES);
27}
28
29static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
30 struct ipc_namespace *old_ns)
31{
32 struct ipc_namespace *ns;
33 struct ucounts *ucounts;
34 int err;
35
36 err = -ENOSPC;
37 ucounts = inc_ipc_namespaces(user_ns);
38 if (!ucounts)
39 goto fail;
40
41 err = -ENOMEM;
42 ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
43 if (ns == NULL)
44 goto fail_dec;
45
46 err = ns_alloc_inum(&ns->ns);
47 if (err)
48 goto fail_free;
49 ns->ns.ops = &ipcns_operations;
50
51 atomic_set(&ns->count, 1);
52 ns->user_ns = get_user_ns(user_ns);
53 ns->ucounts = ucounts;
54
55 err = mq_init_ns(ns);
56 if (err)
57 goto fail_put;
58
59 sem_init_ns(ns);
60 msg_init_ns(ns);
61 shm_init_ns(ns);
62
63 return ns;
64
65fail_put:
66 put_user_ns(ns->user_ns);
67 ns_free_inum(&ns->ns);
68fail_free:
69 kfree(ns);
70fail_dec:
71 dec_ipc_namespaces(ucounts);
72fail:
73 return ERR_PTR(err);
74}
75
76struct ipc_namespace *copy_ipcs(unsigned long flags,
77 struct user_namespace *user_ns, struct ipc_namespace *ns)
78{
79 if (!(flags & CLONE_NEWIPC))
80 return get_ipc_ns(ns);
81 return create_ipc_ns(user_ns, ns);
82}
83
84/*
85 * free_ipcs - free all ipcs of one type
86 * @ns: the namespace to remove the ipcs from
87 * @ids: the table of ipcs to free
88 * @free: the function called to free each individual ipc
89 *
90 * Called for each kind of ipc when an ipc_namespace exits.
91 */
92void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
93 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
94{
95 struct kern_ipc_perm *perm;
96 int next_id;
97 int total, in_use;
98
99 down_write(&ids->rwsem);
100
101 in_use = ids->in_use;
102
103 for (total = 0, next_id = 0; total < in_use; next_id++) {
104 perm = idr_find(&ids->ipcs_idr, next_id);
105 if (perm == NULL)
106 continue;
107 rcu_read_lock();
108 ipc_lock_object(perm);
109 free(ns, perm);
110 total++;
111 }
112 up_write(&ids->rwsem);
113}
114
115static void free_ipc_ns(struct ipc_namespace *ns)
116{
117 sem_exit_ns(ns);
118 msg_exit_ns(ns);
119 shm_exit_ns(ns);
120
121 dec_ipc_namespaces(ns->ucounts);
122 put_user_ns(ns->user_ns);
123 ns_free_inum(&ns->ns);
124 kfree(ns);
125}
126
127/*
128 * put_ipc_ns - drop a reference to an ipc namespace.
129 * @ns: the namespace to put
130 *
131 * If this is the last task in the namespace exiting, and
132 * it is dropping the refcount to 0, then it can race with
133 * a task in another ipc namespace but in a mounts namespace
134 * which has this ipcns's mqueuefs mounted, doing some action
135 * with one of the mqueuefs files. That can raise the refcount.
136 * So dropping the refcount, and raising the refcount when
137 * accessing it through the VFS, are protected with mq_lock.
138 *
139 * (Clearly, a task raising the refcount on its own ipc_ns
140 * needn't take mq_lock since it can't race with the last task
141 * in the ipcns exiting).
142 */
143void put_ipc_ns(struct ipc_namespace *ns)
144{
145 if (atomic_dec_and_lock(&ns->count, &mq_lock)) {
146 mq_clear_sbinfo(ns);
147 spin_unlock(&mq_lock);
148 mq_put_mnt(ns);
149 free_ipc_ns(ns);
150 }
151}
152
153static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
154{
155 return container_of(ns, struct ipc_namespace, ns);
156}
157
158static struct ns_common *ipcns_get(struct task_struct *task)
159{
160 struct ipc_namespace *ns = NULL;
161 struct nsproxy *nsproxy;
162
163 task_lock(task);
164 nsproxy = task->nsproxy;
165 if (nsproxy)
166 ns = get_ipc_ns(nsproxy->ipc_ns);
167 task_unlock(task);
168
169 return ns ? &ns->ns : NULL;
170}
171
172static void ipcns_put(struct ns_common *ns)
173{
174 return put_ipc_ns(to_ipc_ns(ns));
175}
176
177static int ipcns_install(struct nsproxy *nsproxy, struct ns_common *new)
178{
179 struct ipc_namespace *ns = to_ipc_ns(new);
180 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
181 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
182 return -EPERM;
183
184 /* Ditch state from the old ipc namespace */
185 exit_sem(current);
186 put_ipc_ns(nsproxy->ipc_ns);
187 nsproxy->ipc_ns = get_ipc_ns(ns);
188 return 0;
189}
190
191static struct user_namespace *ipcns_owner(struct ns_common *ns)
192{
193 return to_ipc_ns(ns)->user_ns;
194}
195
196const struct proc_ns_operations ipcns_operations = {
197 .name = "ipc",
198 .type = CLONE_NEWIPC,
199 .get = ipcns_get,
200 .put = ipcns_put,
201 .install = ipcns_install,
202 .owner = ipcns_owner,
203};
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 task_struct *tsk,
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 atomic_set(&ns->count, 1);
30 err = mq_init_ns(ns);
31 if (err) {
32 kfree(ns);
33 return ERR_PTR(err);
34 }
35 atomic_inc(&nr_ipc_ns);
36
37 sem_init_ns(ns);
38 msg_init_ns(ns);
39 shm_init_ns(ns);
40
41 /*
42 * msgmni has already been computed for the new ipc ns.
43 * Thus, do the ipcns creation notification before registering that
44 * new ipcns in the chain.
45 */
46 ipcns_notify(IPCNS_CREATED);
47 register_ipcns_notifier(ns);
48
49 ns->user_ns = get_user_ns(task_cred_xxx(tsk, user)->user_ns);
50
51 return ns;
52}
53
54struct ipc_namespace *copy_ipcs(unsigned long flags,
55 struct task_struct *tsk)
56{
57 struct ipc_namespace *ns = tsk->nsproxy->ipc_ns;
58
59 if (!(flags & CLONE_NEWIPC))
60 return get_ipc_ns(ns);
61 return create_ipc_ns(tsk, ns);
62}
63
64/*
65 * free_ipcs - free all ipcs of one type
66 * @ns: the namespace to remove the ipcs from
67 * @ids: the table of ipcs to free
68 * @free: the function called to free each individual ipc
69 *
70 * Called for each kind of ipc when an ipc_namespace exits.
71 */
72void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
73 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
74{
75 struct kern_ipc_perm *perm;
76 int next_id;
77 int total, in_use;
78
79 down_write(&ids->rw_mutex);
80
81 in_use = ids->in_use;
82
83 for (total = 0, next_id = 0; total < in_use; next_id++) {
84 perm = idr_find(&ids->ipcs_idr, next_id);
85 if (perm == NULL)
86 continue;
87 ipc_lock_by_ptr(perm);
88 free(ns, perm);
89 total++;
90 }
91 up_write(&ids->rw_mutex);
92}
93
94static void free_ipc_ns(struct ipc_namespace *ns)
95{
96 /*
97 * Unregistering the hotplug notifier at the beginning guarantees
98 * that the ipc namespace won't be freed while we are inside the
99 * callback routine. Since the blocking_notifier_chain_XXX routines
100 * hold a rw lock on the notifier list, unregister_ipcns_notifier()
101 * won't take the rw lock before blocking_notifier_call_chain() has
102 * released the rd lock.
103 */
104 unregister_ipcns_notifier(ns);
105 sem_exit_ns(ns);
106 msg_exit_ns(ns);
107 shm_exit_ns(ns);
108 atomic_dec(&nr_ipc_ns);
109
110 /*
111 * Do the ipcns removal notification after decrementing nr_ipc_ns in
112 * order to have a correct value when recomputing msgmni.
113 */
114 ipcns_notify(IPCNS_REMOVED);
115 put_user_ns(ns->user_ns);
116 kfree(ns);
117}
118
119/*
120 * put_ipc_ns - drop a reference to an ipc namespace.
121 * @ns: the namespace to put
122 *
123 * If this is the last task in the namespace exiting, and
124 * it is dropping the refcount to 0, then it can race with
125 * a task in another ipc namespace but in a mounts namespace
126 * which has this ipcns's mqueuefs mounted, doing some action
127 * with one of the mqueuefs files. That can raise the refcount.
128 * So dropping the refcount, and raising the refcount when
129 * accessing it through the VFS, are protected with mq_lock.
130 *
131 * (Clearly, a task raising the refcount on its own ipc_ns
132 * needn't take mq_lock since it can't race with the last task
133 * in the ipcns exiting).
134 */
135void put_ipc_ns(struct ipc_namespace *ns)
136{
137 if (atomic_dec_and_lock(&ns->count, &mq_lock)) {
138 mq_clear_sbinfo(ns);
139 spin_unlock(&mq_lock);
140 mq_put_mnt(ns);
141 free_ipc_ns(ns);
142 }
143}
144
145static void *ipcns_get(struct task_struct *task)
146{
147 struct ipc_namespace *ns = NULL;
148 struct nsproxy *nsproxy;
149
150 rcu_read_lock();
151 nsproxy = task_nsproxy(task);
152 if (nsproxy)
153 ns = get_ipc_ns(nsproxy->ipc_ns);
154 rcu_read_unlock();
155
156 return ns;
157}
158
159static void ipcns_put(void *ns)
160{
161 return put_ipc_ns(ns);
162}
163
164static int ipcns_install(struct nsproxy *nsproxy, void *ns)
165{
166 /* Ditch state from the old ipc namespace */
167 exit_sem(current);
168 put_ipc_ns(nsproxy->ipc_ns);
169 nsproxy->ipc_ns = get_ipc_ns(ns);
170 return 0;
171}
172
173const struct proc_ns_operations ipcns_operations = {
174 .name = "ipc",
175 .type = CLONE_NEWIPC,
176 .get = ipcns_get,
177 .put = ipcns_put,
178 .install = ipcns_install,
179};