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