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/nstree.h>
19 #include <linux/sched/task.h>
20
21 #include "util.h"
22
23 /*
24 * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
25 */
26 static void free_ipc(struct work_struct *unused);
27 static DECLARE_WORK(free_ipc_work, free_ipc);
28
inc_ipc_namespaces(struct user_namespace * ns)29 static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns)
30 {
31 return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES);
32 }
33
dec_ipc_namespaces(struct ucounts * ucounts)34 static void dec_ipc_namespaces(struct ucounts *ucounts)
35 {
36 dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES);
37 }
38
create_ipc_ns(struct user_namespace * user_ns,struct ipc_namespace * old_ns)39 static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
40 struct ipc_namespace *old_ns)
41 {
42 struct ipc_namespace *ns;
43 struct ucounts *ucounts;
44 int err;
45
46 err = -ENOSPC;
47 again:
48 ucounts = inc_ipc_namespaces(user_ns);
49 if (!ucounts) {
50 /*
51 * IPC namespaces are freed asynchronously, by free_ipc_work.
52 * If frees were pending, flush_work will wait, and
53 * return true. Fail the allocation if no frees are pending.
54 */
55 if (flush_work(&free_ipc_work))
56 goto again;
57 goto fail;
58 }
59
60 err = -ENOMEM;
61 ns = kzalloc(sizeof(struct ipc_namespace), GFP_KERNEL_ACCOUNT);
62 if (ns == NULL)
63 goto fail_dec;
64
65 err = ns_common_init(ns);
66 if (err)
67 goto fail_free;
68
69 ns->user_ns = get_user_ns(user_ns);
70 ns->ucounts = ucounts;
71
72 err = mq_init_ns(ns);
73 if (err)
74 goto fail_put;
75
76 err = -ENOMEM;
77 if (!setup_mq_sysctls(ns))
78 goto fail_put;
79
80 if (!setup_ipc_sysctls(ns))
81 goto fail_mq;
82
83 err = msg_init_ns(ns);
84 if (err)
85 goto fail_ipc;
86
87 sem_init_ns(ns);
88 shm_init_ns(ns);
89 ns_tree_add(ns);
90
91 return ns;
92
93 fail_ipc:
94 retire_ipc_sysctls(ns);
95 fail_mq:
96 retire_mq_sysctls(ns);
97
98 fail_put:
99 put_user_ns(ns->user_ns);
100 ns_common_free(ns);
101 fail_free:
102 kfree(ns);
103 fail_dec:
104 dec_ipc_namespaces(ucounts);
105 fail:
106 return ERR_PTR(err);
107 }
108
copy_ipcs(u64 flags,struct user_namespace * user_ns,struct ipc_namespace * ns)109 struct ipc_namespace *copy_ipcs(u64 flags,
110 struct user_namespace *user_ns, struct ipc_namespace *ns)
111 {
112 if (!(flags & CLONE_NEWIPC))
113 return get_ipc_ns(ns);
114 return create_ipc_ns(user_ns, ns);
115 }
116
117 /*
118 * free_ipcs - free all ipcs of one type
119 * @ns: the namespace to remove the ipcs from
120 * @ids: the table of ipcs to free
121 * @free: the function called to free each individual ipc
122 *
123 * Called for each kind of ipc when an ipc_namespace exits.
124 */
free_ipcs(struct ipc_namespace * ns,struct ipc_ids * ids,void (* free)(struct ipc_namespace *,struct kern_ipc_perm *))125 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
126 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
127 {
128 struct kern_ipc_perm *perm;
129 int next_id;
130 int total, in_use;
131
132 down_write(&ids->rwsem);
133
134 in_use = ids->in_use;
135
136 for (total = 0, next_id = 0; total < in_use; next_id++) {
137 perm = idr_find(&ids->ipcs_idr, next_id);
138 if (perm == NULL)
139 continue;
140 rcu_read_lock();
141 ipc_lock_object(perm);
142 free(ns, perm);
143 total++;
144 }
145 up_write(&ids->rwsem);
146 }
147
free_ipc_ns(struct ipc_namespace * ns)148 static void free_ipc_ns(struct ipc_namespace *ns)
149 {
150 /*
151 * Caller needs to wait for an RCU grace period to have passed
152 * after making the mount point inaccessible to new accesses.
153 */
154 mntput(ns->mq_mnt);
155 sem_exit_ns(ns);
156 msg_exit_ns(ns);
157 shm_exit_ns(ns);
158
159 retire_mq_sysctls(ns);
160 retire_ipc_sysctls(ns);
161
162 dec_ipc_namespaces(ns->ucounts);
163 put_user_ns(ns->user_ns);
164 ns_common_free(ns);
165 kfree(ns);
166 }
167
168 static LLIST_HEAD(free_ipc_list);
free_ipc(struct work_struct * unused)169 static void free_ipc(struct work_struct *unused)
170 {
171 struct llist_node *node = llist_del_all(&free_ipc_list);
172 struct ipc_namespace *n, *t;
173
174 llist_for_each_entry_safe(n, t, node, mnt_llist)
175 mnt_make_shortterm(n->mq_mnt);
176
177 /* Wait for any last users to have gone away. */
178 synchronize_rcu();
179
180 llist_for_each_entry_safe(n, t, node, mnt_llist)
181 free_ipc_ns(n);
182 }
183
184 /*
185 * put_ipc_ns - drop a reference to an ipc namespace.
186 * @ns: the namespace to put
187 *
188 * If this is the last task in the namespace exiting, and
189 * it is dropping the refcount to 0, then it can race with
190 * a task in another ipc namespace but in a mounts namespace
191 * which has this ipcns's mqueuefs mounted, doing some action
192 * with one of the mqueuefs files. That can raise the refcount.
193 * So dropping the refcount, and raising the refcount when
194 * accessing it through the VFS, are protected with mq_lock.
195 *
196 * (Clearly, a task raising the refcount on its own ipc_ns
197 * needn't take mq_lock since it can't race with the last task
198 * in the ipcns exiting).
199 */
put_ipc_ns(struct ipc_namespace * ns)200 void put_ipc_ns(struct ipc_namespace *ns)
201 {
202 if (ns_ref_put_and_lock(ns, &mq_lock)) {
203 mq_clear_sbinfo(ns);
204 spin_unlock(&mq_lock);
205
206 ns_tree_remove(ns);
207 if (llist_add(&ns->mnt_llist, &free_ipc_list))
208 schedule_work(&free_ipc_work);
209 }
210 }
211
ipcns_get(struct task_struct * task)212 static struct ns_common *ipcns_get(struct task_struct *task)
213 {
214 struct ipc_namespace *ns = NULL;
215 struct nsproxy *nsproxy;
216
217 task_lock(task);
218 nsproxy = task->nsproxy;
219 if (nsproxy)
220 ns = get_ipc_ns(nsproxy->ipc_ns);
221 task_unlock(task);
222
223 return ns ? &ns->ns : NULL;
224 }
225
ipcns_put(struct ns_common * ns)226 static void ipcns_put(struct ns_common *ns)
227 {
228 return put_ipc_ns(to_ipc_ns(ns));
229 }
230
ipcns_install(struct nsset * nsset,struct ns_common * new)231 static int ipcns_install(struct nsset *nsset, struct ns_common *new)
232 {
233 struct nsproxy *nsproxy = nsset->nsproxy;
234 struct ipc_namespace *ns = to_ipc_ns(new);
235 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
236 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
237 return -EPERM;
238
239 put_ipc_ns(nsproxy->ipc_ns);
240 nsproxy->ipc_ns = get_ipc_ns(ns);
241 return 0;
242 }
243
ipcns_owner(struct ns_common * ns)244 static struct user_namespace *ipcns_owner(struct ns_common *ns)
245 {
246 return to_ipc_ns(ns)->user_ns;
247 }
248
249 const struct proc_ns_operations ipcns_operations = {
250 .name = "ipc",
251 .get = ipcns_get,
252 .put = ipcns_put,
253 .install = ipcns_install,
254 .owner = ipcns_owner,
255 };
256