| 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 | /* |
| 23 | * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount. |
| 24 | */ |
| 25 | static void free_ipc(struct work_struct *unused); |
| 26 | static DECLARE_WORK(free_ipc_work, free_ipc); |
| 27 | |
| 28 | static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns) |
| 29 | { |
| 30 | return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES); |
| 31 | } |
| 32 | |
| 33 | static void dec_ipc_namespaces(struct ucounts *ucounts) |
| 34 | { |
| 35 | dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES); |
| 36 | } |
| 37 | |
| 38 | static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns, |
| 39 | struct ipc_namespace *old_ns) |
| 40 | { |
| 41 | struct ipc_namespace *ns; |
| 42 | struct ucounts *ucounts; |
| 43 | int err; |
| 44 | |
| 45 | err = -ENOSPC; |
| 46 | again: |
| 47 | ucounts = inc_ipc_namespaces(user_ns); |
| 48 | if (!ucounts) { |
| 49 | /* |
| 50 | * IPC namespaces are freed asynchronously, by free_ipc_work. |
| 51 | * If frees were pending, flush_work will wait, and |
| 52 | * return true. Fail the allocation if no frees are pending. |
| 53 | */ |
| 54 | if (flush_work(&free_ipc_work)) |
| 55 | goto again; |
| 56 | goto fail; |
| 57 | } |
| 58 | |
| 59 | err = -ENOMEM; |
| 60 | ns = kzalloc(sizeof(struct ipc_namespace), GFP_KERNEL_ACCOUNT); |
| 61 | if (ns == NULL) |
| 62 | goto fail_dec; |
| 63 | |
| 64 | err = ns_alloc_inum(&ns->ns); |
| 65 | if (err) |
| 66 | goto fail_free; |
| 67 | ns->ns.ops = &ipcns_operations; |
| 68 | |
| 69 | refcount_set(&ns->ns.count, 1); |
| 70 | ns->user_ns = get_user_ns(user_ns); |
| 71 | ns->ucounts = ucounts; |
| 72 | |
| 73 | err = mq_init_ns(ns); |
| 74 | if (err) |
| 75 | goto fail_put; |
| 76 | |
| 77 | err = -ENOMEM; |
| 78 | if (!setup_mq_sysctls(ns)) |
| 79 | goto fail_put; |
| 80 | |
| 81 | if (!setup_ipc_sysctls(ns)) |
| 82 | goto fail_mq; |
| 83 | |
| 84 | err = msg_init_ns(ns); |
| 85 | if (err) |
| 86 | goto fail_put; |
| 87 | |
| 88 | sem_init_ns(ns); |
| 89 | shm_init_ns(ns); |
| 90 | |
| 91 | return ns; |
| 92 | |
| 93 | fail_mq: |
| 94 | retire_mq_sysctls(ns); |
| 95 | |
| 96 | fail_put: |
| 97 | put_user_ns(ns->user_ns); |
| 98 | ns_free_inum(&ns->ns); |
| 99 | fail_free: |
| 100 | kfree(ns); |
| 101 | fail_dec: |
| 102 | dec_ipc_namespaces(ucounts); |
| 103 | fail: |
| 104 | return ERR_PTR(err); |
| 105 | } |
| 106 | |
| 107 | struct ipc_namespace *copy_ipcs(unsigned long flags, |
| 108 | struct user_namespace *user_ns, struct ipc_namespace *ns) |
| 109 | { |
| 110 | if (!(flags & CLONE_NEWIPC)) |
| 111 | return get_ipc_ns(ns); |
| 112 | return create_ipc_ns(user_ns, ns); |
| 113 | } |
| 114 | |
| 115 | /* |
| 116 | * free_ipcs - free all ipcs of one type |
| 117 | * @ns: the namespace to remove the ipcs from |
| 118 | * @ids: the table of ipcs to free |
| 119 | * @free: the function called to free each individual ipc |
| 120 | * |
| 121 | * Called for each kind of ipc when an ipc_namespace exits. |
| 122 | */ |
| 123 | void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids, |
| 124 | void (*free)(struct ipc_namespace *, struct kern_ipc_perm *)) |
| 125 | { |
| 126 | struct kern_ipc_perm *perm; |
| 127 | int next_id; |
| 128 | int total, in_use; |
| 129 | |
| 130 | down_write(&ids->rwsem); |
| 131 | |
| 132 | in_use = ids->in_use; |
| 133 | |
| 134 | for (total = 0, next_id = 0; total < in_use; next_id++) { |
| 135 | perm = idr_find(&ids->ipcs_idr, next_id); |
| 136 | if (perm == NULL) |
| 137 | continue; |
| 138 | rcu_read_lock(); |
| 139 | ipc_lock_object(perm); |
| 140 | free(ns, perm); |
| 141 | total++; |
| 142 | } |
| 143 | up_write(&ids->rwsem); |
| 144 | } |
| 145 | |
| 146 | static void free_ipc_ns(struct ipc_namespace *ns) |
| 147 | { |
| 148 | /* |
| 149 | * Caller needs to wait for an RCU grace period to have passed |
| 150 | * after making the mount point inaccessible to new accesses. |
| 151 | */ |
| 152 | mntput(ns->mq_mnt); |
| 153 | sem_exit_ns(ns); |
| 154 | msg_exit_ns(ns); |
| 155 | shm_exit_ns(ns); |
| 156 | |
| 157 | retire_mq_sysctls(ns); |
| 158 | retire_ipc_sysctls(ns); |
| 159 | |
| 160 | dec_ipc_namespaces(ns->ucounts); |
| 161 | put_user_ns(ns->user_ns); |
| 162 | ns_free_inum(&ns->ns); |
| 163 | kfree(ns); |
| 164 | } |
| 165 | |
| 166 | static LLIST_HEAD(free_ipc_list); |
| 167 | static void free_ipc(struct work_struct *unused) |
| 168 | { |
| 169 | struct llist_node *node = llist_del_all(&free_ipc_list); |
| 170 | struct ipc_namespace *n, *t; |
| 171 | |
| 172 | llist_for_each_entry_safe(n, t, node, mnt_llist) |
| 173 | mnt_make_shortterm(n->mq_mnt); |
| 174 | |
| 175 | /* Wait for any last users to have gone away. */ |
| 176 | synchronize_rcu(); |
| 177 | |
| 178 | llist_for_each_entry_safe(n, t, node, mnt_llist) |
| 179 | free_ipc_ns(n); |
| 180 | } |
| 181 | |
| 182 | /* |
| 183 | * put_ipc_ns - drop a reference to an ipc namespace. |
| 184 | * @ns: the namespace to put |
| 185 | * |
| 186 | * If this is the last task in the namespace exiting, and |
| 187 | * it is dropping the refcount to 0, then it can race with |
| 188 | * a task in another ipc namespace but in a mounts namespace |
| 189 | * which has this ipcns's mqueuefs mounted, doing some action |
| 190 | * with one of the mqueuefs files. That can raise the refcount. |
| 191 | * So dropping the refcount, and raising the refcount when |
| 192 | * accessing it through the VFS, are protected with mq_lock. |
| 193 | * |
| 194 | * (Clearly, a task raising the refcount on its own ipc_ns |
| 195 | * needn't take mq_lock since it can't race with the last task |
| 196 | * in the ipcns exiting). |
| 197 | */ |
| 198 | void put_ipc_ns(struct ipc_namespace *ns) |
| 199 | { |
| 200 | if (refcount_dec_and_lock(&ns->ns.count, &mq_lock)) { |
| 201 | mq_clear_sbinfo(ns); |
| 202 | spin_unlock(&mq_lock); |
| 203 | |
| 204 | if (llist_add(&ns->mnt_llist, &free_ipc_list)) |
| 205 | schedule_work(&free_ipc_work); |
| 206 | } |
| 207 | } |
| 208 | |
| 209 | static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns) |
| 210 | { |
| 211 | return container_of(ns, struct ipc_namespace, ns); |
| 212 | } |
| 213 | |
| 214 | static struct ns_common *ipcns_get(struct task_struct *task) |
| 215 | { |
| 216 | struct ipc_namespace *ns = NULL; |
| 217 | struct nsproxy *nsproxy; |
| 218 | |
| 219 | task_lock(task); |
| 220 | nsproxy = task->nsproxy; |
| 221 | if (nsproxy) |
| 222 | ns = get_ipc_ns(nsproxy->ipc_ns); |
| 223 | task_unlock(task); |
| 224 | |
| 225 | return ns ? &ns->ns : NULL; |
| 226 | } |
| 227 | |
| 228 | static void ipcns_put(struct ns_common *ns) |
| 229 | { |
| 230 | return put_ipc_ns(to_ipc_ns(ns)); |
| 231 | } |
| 232 | |
| 233 | static int ipcns_install(struct nsset *nsset, struct ns_common *new) |
| 234 | { |
| 235 | struct nsproxy *nsproxy = nsset->nsproxy; |
| 236 | struct ipc_namespace *ns = to_ipc_ns(new); |
| 237 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) || |
| 238 | !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN)) |
| 239 | return -EPERM; |
| 240 | |
| 241 | put_ipc_ns(nsproxy->ipc_ns); |
| 242 | nsproxy->ipc_ns = get_ipc_ns(ns); |
| 243 | return 0; |
| 244 | } |
| 245 | |
| 246 | static struct user_namespace *ipcns_owner(struct ns_common *ns) |
| 247 | { |
| 248 | return to_ipc_ns(ns)->user_ns; |
| 249 | } |
| 250 | |
| 251 | const struct proc_ns_operations ipcns_operations = { |
| 252 | .name = "ipc", |
| 253 | .type = CLONE_NEWIPC, |
| 254 | .get = ipcns_get, |
| 255 | .put = ipcns_put, |
| 256 | .install = ipcns_install, |
| 257 | .owner = ipcns_owner, |
| 258 | }; |