1 // SPDX-License-Identifier: GPL-2.0
3 * linux/ipc/namespace.c
4 * Copyright (C) 2006 Pavel Emelyanov <xemul@openvz.org> OpenVZ, SWsoft Inc.
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>
15 #include <linux/mount.h>
16 #include <linux/user_namespace.h>
17 #include <linux/proc_ns.h>
18 #include <linux/sched/task.h>
22 static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns)
24 return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES);
27 static void dec_ipc_namespaces(struct ucounts *ucounts)
29 dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES);
32 static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
33 struct ipc_namespace *old_ns)
35 struct ipc_namespace *ns;
36 struct ucounts *ucounts;
40 ucounts = inc_ipc_namespaces(user_ns);
45 ns = kzalloc(sizeof(struct ipc_namespace), GFP_KERNEL_ACCOUNT);
49 err = ns_alloc_inum(&ns->ns);
52 ns->ns.ops = &ipcns_operations;
54 refcount_set(&ns->ns.count, 1);
55 ns->user_ns = get_user_ns(user_ns);
56 ns->ucounts = ucounts;
63 if (!setup_mq_sysctls(ns))
66 if (!setup_ipc_sysctls(ns))
69 err = msg_init_ns(ns);
79 retire_mq_sysctls(ns);
82 put_user_ns(ns->user_ns);
83 ns_free_inum(&ns->ns);
87 dec_ipc_namespaces(ucounts);
92 struct ipc_namespace *copy_ipcs(unsigned long flags,
93 struct user_namespace *user_ns, struct ipc_namespace *ns)
95 if (!(flags & CLONE_NEWIPC))
96 return get_ipc_ns(ns);
97 return create_ipc_ns(user_ns, ns);
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
106 * Called for each kind of ipc when an ipc_namespace exits.
108 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
109 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
111 struct kern_ipc_perm *perm;
115 down_write(&ids->rwsem);
117 in_use = ids->in_use;
119 for (total = 0, next_id = 0; total < in_use; next_id++) {
120 perm = idr_find(&ids->ipcs_idr, next_id);
124 ipc_lock_object(perm);
128 up_write(&ids->rwsem);
131 static void free_ipc_ns(struct ipc_namespace *ns)
133 /* mq_put_mnt() waits for a grace period as kern_unmount()
134 * uses synchronize_rcu().
141 retire_mq_sysctls(ns);
142 retire_ipc_sysctls(ns);
144 dec_ipc_namespaces(ns->ucounts);
145 put_user_ns(ns->user_ns);
146 ns_free_inum(&ns->ns);
150 static LLIST_HEAD(free_ipc_list);
151 static void free_ipc(struct work_struct *unused)
153 struct llist_node *node = llist_del_all(&free_ipc_list);
154 struct ipc_namespace *n, *t;
156 llist_for_each_entry_safe(n, t, node, mnt_llist)
161 * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
163 static DECLARE_WORK(free_ipc_work, free_ipc);
166 * put_ipc_ns - drop a reference to an ipc namespace.
167 * @ns: the namespace to put
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.
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).
181 void put_ipc_ns(struct ipc_namespace *ns)
183 if (refcount_dec_and_lock(&ns->ns.count, &mq_lock)) {
185 spin_unlock(&mq_lock);
187 if (llist_add(&ns->mnt_llist, &free_ipc_list))
188 schedule_work(&free_ipc_work);
192 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
194 return container_of(ns, struct ipc_namespace, ns);
197 static struct ns_common *ipcns_get(struct task_struct *task)
199 struct ipc_namespace *ns = NULL;
200 struct nsproxy *nsproxy;
203 nsproxy = task->nsproxy;
205 ns = get_ipc_ns(nsproxy->ipc_ns);
208 return ns ? &ns->ns : NULL;
211 static void ipcns_put(struct ns_common *ns)
213 return put_ipc_ns(to_ipc_ns(ns));
216 static int ipcns_install(struct nsset *nsset, struct ns_common *new)
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))
224 put_ipc_ns(nsproxy->ipc_ns);
225 nsproxy->ipc_ns = get_ipc_ns(ns);
229 static struct user_namespace *ipcns_owner(struct ns_common *ns)
231 return to_ipc_ns(ns)->user_ns;
234 const struct proc_ns_operations ipcns_operations = {
236 .type = CLONE_NEWIPC,
239 .install = ipcns_install,
240 .owner = ipcns_owner,