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_mq; 68 69 sem_init_ns(ns); 70 msg_init_ns(ns); 71 shm_init_ns(ns); 72 73 return ns; 74 75 fail_mq: 76 retire_mq_sysctls(ns); 77 78 fail_put: 79 put_user_ns(ns->user_ns); 80 ns_free_inum(&ns->ns); 81 fail_free: 82 kfree(ns); 83 fail_dec: 84 dec_ipc_namespaces(ucounts); 85 fail: 86 return ERR_PTR(err); 87 } 88 89 struct ipc_namespace *copy_ipcs(unsigned long flags, 90 struct user_namespace *user_ns, struct ipc_namespace *ns) 91 { 92 if (!(flags & CLONE_NEWIPC)) 93 return get_ipc_ns(ns); 94 return create_ipc_ns(user_ns, ns); 95 } 96 97 /* 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 102 * 103 * Called for each kind of ipc when an ipc_namespace exits. 104 */ 105 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids, 106 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *)) 107 { 108 struct kern_ipc_perm *perm; 109 int next_id; 110 int total, in_use; 111 112 down_write(&ids->rwsem); 113 114 in_use = ids->in_use; 115 116 for (total = 0, next_id = 0; total < in_use; next_id++) { 117 perm = idr_find(&ids->ipcs_idr, next_id); 118 if (perm == NULL) 119 continue; 120 rcu_read_lock(); 121 ipc_lock_object(perm); 122 free(ns, perm); 123 total++; 124 } 125 up_write(&ids->rwsem); 126 } 127 128 static void free_ipc_ns(struct ipc_namespace *ns) 129 { 130 /* mq_put_mnt() waits for a grace period as kern_unmount() 131 * uses synchronize_rcu(). 132 */ 133 mq_put_mnt(ns); 134 sem_exit_ns(ns); 135 msg_exit_ns(ns); 136 shm_exit_ns(ns); 137 138 retire_mq_sysctls(ns); 139 retire_ipc_sysctls(ns); 140 141 dec_ipc_namespaces(ns->ucounts); 142 put_user_ns(ns->user_ns); 143 ns_free_inum(&ns->ns); 144 kfree(ns); 145 } 146 147 static LLIST_HEAD(free_ipc_list); 148 static void free_ipc(struct work_struct *unused) 149 { 150 struct llist_node *node = llist_del_all(&free_ipc_list); 151 struct ipc_namespace *n, *t; 152 153 llist_for_each_entry_safe(n, t, node, mnt_llist) 154 free_ipc_ns(n); 155 } 156 157 /* 158 * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount. 159 */ 160 static DECLARE_WORK(free_ipc_work, free_ipc); 161 162 /* 163 * put_ipc_ns - drop a reference to an ipc namespace. 164 * @ns: the namespace to put 165 * 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. 173 * 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). 177 */ 178 void put_ipc_ns(struct ipc_namespace *ns) 179 { 180 if (refcount_dec_and_lock(&ns->ns.count, &mq_lock)) { 181 mq_clear_sbinfo(ns); 182 spin_unlock(&mq_lock); 183 184 if (llist_add(&ns->mnt_llist, &free_ipc_list)) 185 schedule_work(&free_ipc_work); 186 } 187 } 188 189 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns) 190 { 191 return container_of(ns, struct ipc_namespace, ns); 192 } 193 194 static struct ns_common *ipcns_get(struct task_struct *task) 195 { 196 struct ipc_namespace *ns = NULL; 197 struct nsproxy *nsproxy; 198 199 task_lock(task); 200 nsproxy = task->nsproxy; 201 if (nsproxy) 202 ns = get_ipc_ns(nsproxy->ipc_ns); 203 task_unlock(task); 204 205 return ns ? &ns->ns : NULL; 206 } 207 208 static void ipcns_put(struct ns_common *ns) 209 { 210 return put_ipc_ns(to_ipc_ns(ns)); 211 } 212 213 static int ipcns_install(struct nsset *nsset, struct ns_common *new) 214 { 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)) 219 return -EPERM; 220 221 put_ipc_ns(nsproxy->ipc_ns); 222 nsproxy->ipc_ns = get_ipc_ns(ns); 223 return 0; 224 } 225 226 static struct user_namespace *ipcns_owner(struct ns_common *ns) 227 { 228 return to_ipc_ns(ns)->user_ns; 229 } 230 231 const struct proc_ns_operations ipcns_operations = { 232 .name = "ipc", 233 .type = CLONE_NEWIPC, 234 .get = ipcns_get, 235 .put = ipcns_put, 236 .install = ipcns_install, 237 .owner = ipcns_owner, 238 }; 239