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))
76 retire_mq_sysctls(ns);
79 put_user_ns(ns->user_ns);
80 ns_free_inum(&ns->ns);
84 dec_ipc_namespaces(ucounts);
89 struct ipc_namespace *copy_ipcs(unsigned long flags,
90 struct user_namespace *user_ns, struct ipc_namespace *ns)
92 if (!(flags & CLONE_NEWIPC))
93 return get_ipc_ns(ns);
94 return create_ipc_ns(user_ns, ns);
98 * free_ipcs - free all ipcs of one type
99 * @ns: the namespace to remove the ipcs from
100 * @ids: the table of ipcs to free
101 * @free: the function called to free each individual ipc
103 * Called for each kind of ipc when an ipc_namespace exits.
105 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
106 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
108 struct kern_ipc_perm *perm;
112 down_write(&ids->rwsem);
114 in_use = ids->in_use;
116 for (total = 0, next_id = 0; total < in_use; next_id++) {
117 perm = idr_find(&ids->ipcs_idr, next_id);
121 ipc_lock_object(perm);
125 up_write(&ids->rwsem);
128 static void free_ipc_ns(struct ipc_namespace *ns)
130 /* mq_put_mnt() waits for a grace period as kern_unmount()
131 * uses synchronize_rcu().
138 retire_mq_sysctls(ns);
139 retire_ipc_sysctls(ns);
141 dec_ipc_namespaces(ns->ucounts);
142 put_user_ns(ns->user_ns);
143 ns_free_inum(&ns->ns);
147 static LLIST_HEAD(free_ipc_list);
148 static void free_ipc(struct work_struct *unused)
150 struct llist_node *node = llist_del_all(&free_ipc_list);
151 struct ipc_namespace *n, *t;
153 llist_for_each_entry_safe(n, t, node, mnt_llist)
158 * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
160 static DECLARE_WORK(free_ipc_work, free_ipc);
163 * put_ipc_ns - drop a reference to an ipc namespace.
164 * @ns: the namespace to put
166 * If this is the last task in the namespace exiting, and
167 * it is dropping the refcount to 0, then it can race with
168 * a task in another ipc namespace but in a mounts namespace
169 * which has this ipcns's mqueuefs mounted, doing some action
170 * with one of the mqueuefs files. That can raise the refcount.
171 * So dropping the refcount, and raising the refcount when
172 * accessing it through the VFS, are protected with mq_lock.
174 * (Clearly, a task raising the refcount on its own ipc_ns
175 * needn't take mq_lock since it can't race with the last task
176 * in the ipcns exiting).
178 void put_ipc_ns(struct ipc_namespace *ns)
180 if (refcount_dec_and_lock(&ns->ns.count, &mq_lock)) {
182 spin_unlock(&mq_lock);
184 if (llist_add(&ns->mnt_llist, &free_ipc_list))
185 schedule_work(&free_ipc_work);
189 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
191 return container_of(ns, struct ipc_namespace, ns);
194 static struct ns_common *ipcns_get(struct task_struct *task)
196 struct ipc_namespace *ns = NULL;
197 struct nsproxy *nsproxy;
200 nsproxy = task->nsproxy;
202 ns = get_ipc_ns(nsproxy->ipc_ns);
205 return ns ? &ns->ns : NULL;
208 static void ipcns_put(struct ns_common *ns)
210 return put_ipc_ns(to_ipc_ns(ns));
213 static int ipcns_install(struct nsset *nsset, struct ns_common *new)
215 struct nsproxy *nsproxy = nsset->nsproxy;
216 struct ipc_namespace *ns = to_ipc_ns(new);
217 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
218 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
221 put_ipc_ns(nsproxy->ipc_ns);
222 nsproxy->ipc_ns = get_ipc_ns(ns);
226 static struct user_namespace *ipcns_owner(struct ns_common *ns)
228 return to_ipc_ns(ns)->user_ns;
231 const struct proc_ns_operations ipcns_operations = {
233 .type = CLONE_NEWIPC,
236 .install = ipcns_install,
237 .owner = ipcns_owner,