| 1 | /* |
| 2 | * Copyright (C) 2006 IBM Corporation |
| 3 | * |
| 4 | * Author: Serge Hallyn <serue@us.ibm.com> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License as |
| 8 | * published by the Free Software Foundation, version 2 of the |
| 9 | * License. |
| 10 | * |
| 11 | * Jun 2006 - namespaces support |
| 12 | * OpenVZ, SWsoft Inc. |
| 13 | * Pavel Emelianov <xemul@openvz.org> |
| 14 | */ |
| 15 | |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/version.h> |
| 18 | #include <linux/nsproxy.h> |
| 19 | #include <linux/init_task.h> |
| 20 | #include <linux/mnt_namespace.h> |
| 21 | #include <linux/utsname.h> |
| 22 | #include <linux/pid_namespace.h> |
| 23 | #include <net/net_namespace.h> |
| 24 | #include <linux/ipc_namespace.h> |
| 25 | |
| 26 | static struct kmem_cache *nsproxy_cachep; |
| 27 | |
| 28 | struct nsproxy init_nsproxy = INIT_NSPROXY(init_nsproxy); |
| 29 | |
| 30 | /* |
| 31 | * creates a copy of "orig" with refcount 1. |
| 32 | */ |
| 33 | static inline struct nsproxy *clone_nsproxy(struct nsproxy *orig) |
| 34 | { |
| 35 | struct nsproxy *ns; |
| 36 | |
| 37 | ns = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL); |
| 38 | if (ns) { |
| 39 | memcpy(ns, orig, sizeof(struct nsproxy)); |
| 40 | atomic_set(&ns->count, 1); |
| 41 | } |
| 42 | return ns; |
| 43 | } |
| 44 | |
| 45 | /* |
| 46 | * Create new nsproxy and all of its the associated namespaces. |
| 47 | * Return the newly created nsproxy. Do not attach this to the task, |
| 48 | * leave it to the caller to do proper locking and attach it to task. |
| 49 | */ |
| 50 | static struct nsproxy *create_new_namespaces(unsigned long flags, |
| 51 | struct task_struct *tsk, struct fs_struct *new_fs) |
| 52 | { |
| 53 | struct nsproxy *new_nsp; |
| 54 | int err; |
| 55 | |
| 56 | new_nsp = clone_nsproxy(tsk->nsproxy); |
| 57 | if (!new_nsp) |
| 58 | return ERR_PTR(-ENOMEM); |
| 59 | |
| 60 | new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, new_fs); |
| 61 | if (IS_ERR(new_nsp->mnt_ns)) { |
| 62 | err = PTR_ERR(new_nsp->mnt_ns); |
| 63 | goto out_ns; |
| 64 | } |
| 65 | |
| 66 | new_nsp->uts_ns = copy_utsname(flags, tsk->nsproxy->uts_ns); |
| 67 | if (IS_ERR(new_nsp->uts_ns)) { |
| 68 | err = PTR_ERR(new_nsp->uts_ns); |
| 69 | goto out_uts; |
| 70 | } |
| 71 | |
| 72 | new_nsp->ipc_ns = copy_ipcs(flags, tsk->nsproxy->ipc_ns); |
| 73 | if (IS_ERR(new_nsp->ipc_ns)) { |
| 74 | err = PTR_ERR(new_nsp->ipc_ns); |
| 75 | goto out_ipc; |
| 76 | } |
| 77 | |
| 78 | new_nsp->pid_ns = copy_pid_ns(flags, task_active_pid_ns(tsk)); |
| 79 | if (IS_ERR(new_nsp->pid_ns)) { |
| 80 | err = PTR_ERR(new_nsp->pid_ns); |
| 81 | goto out_pid; |
| 82 | } |
| 83 | |
| 84 | new_nsp->user_ns = copy_user_ns(flags, tsk->nsproxy->user_ns); |
| 85 | if (IS_ERR(new_nsp->user_ns)) { |
| 86 | err = PTR_ERR(new_nsp->user_ns); |
| 87 | goto out_user; |
| 88 | } |
| 89 | |
| 90 | new_nsp->net_ns = copy_net_ns(flags, tsk->nsproxy->net_ns); |
| 91 | if (IS_ERR(new_nsp->net_ns)) { |
| 92 | err = PTR_ERR(new_nsp->net_ns); |
| 93 | goto out_net; |
| 94 | } |
| 95 | |
| 96 | return new_nsp; |
| 97 | |
| 98 | out_net: |
| 99 | if (new_nsp->user_ns) |
| 100 | put_user_ns(new_nsp->user_ns); |
| 101 | out_user: |
| 102 | if (new_nsp->pid_ns) |
| 103 | put_pid_ns(new_nsp->pid_ns); |
| 104 | out_pid: |
| 105 | if (new_nsp->ipc_ns) |
| 106 | put_ipc_ns(new_nsp->ipc_ns); |
| 107 | out_ipc: |
| 108 | if (new_nsp->uts_ns) |
| 109 | put_uts_ns(new_nsp->uts_ns); |
| 110 | out_uts: |
| 111 | if (new_nsp->mnt_ns) |
| 112 | put_mnt_ns(new_nsp->mnt_ns); |
| 113 | out_ns: |
| 114 | kmem_cache_free(nsproxy_cachep, new_nsp); |
| 115 | return ERR_PTR(err); |
| 116 | } |
| 117 | |
| 118 | /* |
| 119 | * called from clone. This now handles copy for nsproxy and all |
| 120 | * namespaces therein. |
| 121 | */ |
| 122 | int copy_namespaces(unsigned long flags, struct task_struct *tsk) |
| 123 | { |
| 124 | struct nsproxy *old_ns = tsk->nsproxy; |
| 125 | struct nsproxy *new_ns; |
| 126 | int err = 0; |
| 127 | |
| 128 | if (!old_ns) |
| 129 | return 0; |
| 130 | |
| 131 | get_nsproxy(old_ns); |
| 132 | |
| 133 | if (!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC | |
| 134 | CLONE_NEWUSER | CLONE_NEWPID | CLONE_NEWNET))) |
| 135 | return 0; |
| 136 | |
| 137 | if (!capable(CAP_SYS_ADMIN)) { |
| 138 | err = -EPERM; |
| 139 | goto out; |
| 140 | } |
| 141 | |
| 142 | /* |
| 143 | * CLONE_NEWIPC must detach from the undolist: after switching |
| 144 | * to a new ipc namespace, the semaphore arrays from the old |
| 145 | * namespace are unreachable. In clone parlance, CLONE_SYSVSEM |
| 146 | * means share undolist with parent, so we must forbid using |
| 147 | * it along with CLONE_NEWIPC. |
| 148 | */ |
| 149 | if ((flags & CLONE_NEWIPC) && (flags & CLONE_SYSVSEM)) { |
| 150 | err = -EINVAL; |
| 151 | goto out; |
| 152 | } |
| 153 | |
| 154 | new_ns = create_new_namespaces(flags, tsk, tsk->fs); |
| 155 | if (IS_ERR(new_ns)) { |
| 156 | err = PTR_ERR(new_ns); |
| 157 | goto out; |
| 158 | } |
| 159 | |
| 160 | tsk->nsproxy = new_ns; |
| 161 | |
| 162 | out: |
| 163 | put_nsproxy(old_ns); |
| 164 | return err; |
| 165 | } |
| 166 | |
| 167 | void free_nsproxy(struct nsproxy *ns) |
| 168 | { |
| 169 | if (ns->mnt_ns) |
| 170 | put_mnt_ns(ns->mnt_ns); |
| 171 | if (ns->uts_ns) |
| 172 | put_uts_ns(ns->uts_ns); |
| 173 | if (ns->ipc_ns) |
| 174 | put_ipc_ns(ns->ipc_ns); |
| 175 | if (ns->pid_ns) |
| 176 | put_pid_ns(ns->pid_ns); |
| 177 | if (ns->user_ns) |
| 178 | put_user_ns(ns->user_ns); |
| 179 | put_net(ns->net_ns); |
| 180 | kmem_cache_free(nsproxy_cachep, ns); |
| 181 | } |
| 182 | |
| 183 | /* |
| 184 | * Called from unshare. Unshare all the namespaces part of nsproxy. |
| 185 | * On success, returns the new nsproxy. |
| 186 | */ |
| 187 | int unshare_nsproxy_namespaces(unsigned long unshare_flags, |
| 188 | struct nsproxy **new_nsp, struct fs_struct *new_fs) |
| 189 | { |
| 190 | int err = 0; |
| 191 | |
| 192 | if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC | |
| 193 | CLONE_NEWUSER | CLONE_NEWNET))) |
| 194 | return 0; |
| 195 | |
| 196 | if (!capable(CAP_SYS_ADMIN)) |
| 197 | return -EPERM; |
| 198 | |
| 199 | *new_nsp = create_new_namespaces(unshare_flags, current, |
| 200 | new_fs ? new_fs : current->fs); |
| 201 | if (IS_ERR(*new_nsp)) { |
| 202 | err = PTR_ERR(*new_nsp); |
| 203 | goto out; |
| 204 | } |
| 205 | |
| 206 | err = ns_cgroup_clone(current, task_pid(current)); |
| 207 | if (err) |
| 208 | put_nsproxy(*new_nsp); |
| 209 | |
| 210 | out: |
| 211 | return err; |
| 212 | } |
| 213 | |
| 214 | void switch_task_namespaces(struct task_struct *p, struct nsproxy *new) |
| 215 | { |
| 216 | struct nsproxy *ns; |
| 217 | |
| 218 | might_sleep(); |
| 219 | |
| 220 | ns = p->nsproxy; |
| 221 | |
| 222 | rcu_assign_pointer(p->nsproxy, new); |
| 223 | |
| 224 | if (ns && atomic_dec_and_test(&ns->count)) { |
| 225 | /* |
| 226 | * wait for others to get what they want from this nsproxy. |
| 227 | * |
| 228 | * cannot release this nsproxy via the call_rcu() since |
| 229 | * put_mnt_ns() will want to sleep |
| 230 | */ |
| 231 | synchronize_rcu(); |
| 232 | free_nsproxy(ns); |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | void exit_task_namespaces(struct task_struct *p) |
| 237 | { |
| 238 | switch_task_namespaces(p, NULL); |
| 239 | } |
| 240 | |
| 241 | static int __init nsproxy_cache_init(void) |
| 242 | { |
| 243 | nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC); |
| 244 | return 0; |
| 245 | } |
| 246 | |
| 247 | module_init(nsproxy_cache_init); |