Loading...
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
22static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns)
23{
24 return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES);
25}
26
27static void dec_ipc_namespaces(struct ucounts *ucounts)
28{
29 dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES);
30}
31
32static 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_ACCOUNT);
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->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 err = -ENOMEM;
63 if (!setup_mq_sysctls(ns))
64 goto fail_put;
65
66 if (!setup_ipc_sysctls(ns))
67 goto fail_mq;
68
69 err = msg_init_ns(ns);
70 if (err)
71 goto fail_put;
72
73 sem_init_ns(ns);
74 shm_init_ns(ns);
75
76 return ns;
77
78fail_mq:
79 retire_mq_sysctls(ns);
80
81fail_put:
82 put_user_ns(ns->user_ns);
83 ns_free_inum(&ns->ns);
84fail_free:
85 kfree(ns);
86fail_dec:
87 dec_ipc_namespaces(ucounts);
88fail:
89 return ERR_PTR(err);
90}
91
92struct ipc_namespace *copy_ipcs(unsigned long flags,
93 struct user_namespace *user_ns, struct ipc_namespace *ns)
94{
95 if (!(flags & CLONE_NEWIPC))
96 return get_ipc_ns(ns);
97 return create_ipc_ns(user_ns, ns);
98}
99
100/*
101 * free_ipcs - free all ipcs of one type
102 * @ns: the namespace to remove the ipcs from
103 * @ids: the table of ipcs to free
104 * @free: the function called to free each individual ipc
105 *
106 * Called for each kind of ipc when an ipc_namespace exits.
107 */
108void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
109 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
110{
111 struct kern_ipc_perm *perm;
112 int next_id;
113 int total, in_use;
114
115 down_write(&ids->rwsem);
116
117 in_use = ids->in_use;
118
119 for (total = 0, next_id = 0; total < in_use; next_id++) {
120 perm = idr_find(&ids->ipcs_idr, next_id);
121 if (perm == NULL)
122 continue;
123 rcu_read_lock();
124 ipc_lock_object(perm);
125 free(ns, perm);
126 total++;
127 }
128 up_write(&ids->rwsem);
129}
130
131static void free_ipc_ns(struct ipc_namespace *ns)
132{
133 /* mq_put_mnt() waits for a grace period as kern_unmount()
134 * uses synchronize_rcu().
135 */
136 mq_put_mnt(ns);
137 sem_exit_ns(ns);
138 msg_exit_ns(ns);
139 shm_exit_ns(ns);
140
141 retire_mq_sysctls(ns);
142 retire_ipc_sysctls(ns);
143
144 dec_ipc_namespaces(ns->ucounts);
145 put_user_ns(ns->user_ns);
146 ns_free_inum(&ns->ns);
147 kfree(ns);
148}
149
150static LLIST_HEAD(free_ipc_list);
151static void free_ipc(struct work_struct *unused)
152{
153 struct llist_node *node = llist_del_all(&free_ipc_list);
154 struct ipc_namespace *n, *t;
155
156 llist_for_each_entry_safe(n, t, node, mnt_llist)
157 free_ipc_ns(n);
158}
159
160/*
161 * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
162 */
163static DECLARE_WORK(free_ipc_work, free_ipc);
164
165/*
166 * put_ipc_ns - drop a reference to an ipc namespace.
167 * @ns: the namespace to put
168 *
169 * If this is the last task in the namespace exiting, and
170 * it is dropping the refcount to 0, then it can race with
171 * a task in another ipc namespace but in a mounts namespace
172 * which has this ipcns's mqueuefs mounted, doing some action
173 * with one of the mqueuefs files. That can raise the refcount.
174 * So dropping the refcount, and raising the refcount when
175 * accessing it through the VFS, are protected with mq_lock.
176 *
177 * (Clearly, a task raising the refcount on its own ipc_ns
178 * needn't take mq_lock since it can't race with the last task
179 * in the ipcns exiting).
180 */
181void put_ipc_ns(struct ipc_namespace *ns)
182{
183 if (refcount_dec_and_lock(&ns->ns.count, &mq_lock)) {
184 mq_clear_sbinfo(ns);
185 spin_unlock(&mq_lock);
186
187 if (llist_add(&ns->mnt_llist, &free_ipc_list))
188 schedule_work(&free_ipc_work);
189 }
190}
191
192static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
193{
194 return container_of(ns, struct ipc_namespace, ns);
195}
196
197static struct ns_common *ipcns_get(struct task_struct *task)
198{
199 struct ipc_namespace *ns = NULL;
200 struct nsproxy *nsproxy;
201
202 task_lock(task);
203 nsproxy = task->nsproxy;
204 if (nsproxy)
205 ns = get_ipc_ns(nsproxy->ipc_ns);
206 task_unlock(task);
207
208 return ns ? &ns->ns : NULL;
209}
210
211static void ipcns_put(struct ns_common *ns)
212{
213 return put_ipc_ns(to_ipc_ns(ns));
214}
215
216static int ipcns_install(struct nsset *nsset, struct ns_common *new)
217{
218 struct nsproxy *nsproxy = nsset->nsproxy;
219 struct ipc_namespace *ns = to_ipc_ns(new);
220 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
221 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
222 return -EPERM;
223
224 put_ipc_ns(nsproxy->ipc_ns);
225 nsproxy->ipc_ns = get_ipc_ns(ns);
226 return 0;
227}
228
229static struct user_namespace *ipcns_owner(struct ns_common *ns)
230{
231 return to_ipc_ns(ns)->user_ns;
232}
233
234const struct proc_ns_operations ipcns_operations = {
235 .name = "ipc",
236 .type = CLONE_NEWIPC,
237 .get = ipcns_get,
238 .put = ipcns_put,
239 .install = ipcns_install,
240 .owner = ipcns_owner,
241};
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 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->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
100static 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 */
142void 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
152static 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
166static void ipcns_put(void *ns)
167{
168 return put_ipc_ns(ns);
169}
170
171static 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
185static unsigned int ipcns_inum(void *vp)
186{
187 struct ipc_namespace *ns = vp;
188
189 return ns->proc_inum;
190}
191
192const 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};