kernel/nsproxy.c

   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/slab.h>
  17 #include <linux/module.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 = {
  29         .count  = ATOMIC_INIT(1),
  30         .uts_ns = &init_uts_ns,
  31 #if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
  32         .ipc_ns = &init_ipc_ns,
  33 #endif
  34         .mnt_ns = NULL,
  35         .pid_ns = &init_pid_ns,
  36 #ifdef CONFIG_NET
  37         .net_ns = &init_net,
  38 #endif
  39 };
  40
  41 static inline struct nsproxy *create_nsproxy(void)
  42 {
  43         struct nsproxy *nsproxy;
  44
  45         nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
  46         if (nsproxy)
  47                 atomic_set(&nsproxy->count, 1);
  48         return nsproxy;
  49 }
  50
  51 /*
  52  * Create new nsproxy and all of its the associated namespaces.
  53  * Return the newly created nsproxy.  Do not attach this to the task,
  54  * leave it to the caller to do proper locking and attach it to task.
  55  */
  56 static struct nsproxy *create_new_namespaces(unsigned long flags,
  57                         struct task_struct *tsk, struct fs_struct *new_fs)
  58 {
  59         struct nsproxy *new_nsp;
  60         int err;
  61
  62         new_nsp = create_nsproxy();
  63         if (!new_nsp)
  64                 return ERR_PTR(-ENOMEM);
  65
  66         new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, new_fs);
  67         if (IS_ERR(new_nsp->mnt_ns)) {
  68                 err = PTR_ERR(new_nsp->mnt_ns);
  69                 goto out_ns;
  70         }
  71
  72         new_nsp->uts_ns = copy_utsname(flags, tsk->nsproxy->uts_ns);
  73         if (IS_ERR(new_nsp->uts_ns)) {
  74                 err = PTR_ERR(new_nsp->uts_ns);
  75                 goto out_uts;
  76         }
  77         if (new_nsp->uts_ns != tsk->nsproxy->uts_ns) {
  78                 put_user_ns(new_nsp->uts_ns->user_ns);
  79                 new_nsp->uts_ns->user_ns = task_cred_xxx(tsk, user)->user_ns;
  80                 get_user_ns(new_nsp->uts_ns->user_ns);
  81         }
  82
  83         new_nsp->ipc_ns = copy_ipcs(flags, tsk->nsproxy->ipc_ns);
  84         if (IS_ERR(new_nsp->ipc_ns)) {
  85                 err = PTR_ERR(new_nsp->ipc_ns);
  86                 goto out_ipc;
  87         }
  88
  89         new_nsp->pid_ns = copy_pid_ns(flags, task_active_pid_ns(tsk));
  90         if (IS_ERR(new_nsp->pid_ns)) {
  91                 err = PTR_ERR(new_nsp->pid_ns);
  92                 goto out_pid;
  93         }
  94
  95         new_nsp->net_ns = copy_net_ns(flags, tsk->nsproxy->net_ns);
  96         if (IS_ERR(new_nsp->net_ns)) {
  97                 err = PTR_ERR(new_nsp->net_ns);
  98                 goto out_net;
  99         }
 100
 101         return new_nsp;
 102
 103 out_net:
 104         if (new_nsp->pid_ns)
 105                 put_pid_ns(new_nsp->pid_ns);
 106 out_pid:
 107         if (new_nsp->ipc_ns)
 108                 put_ipc_ns(new_nsp->ipc_ns);
 109 out_ipc:
 110         if (new_nsp->uts_ns)
 111                 put_uts_ns(new_nsp->uts_ns);
 112 out_uts:
 113         if (new_nsp->mnt_ns)
 114                 put_mnt_ns(new_nsp->mnt_ns);
 115 out_ns:
 116         kmem_cache_free(nsproxy_cachep, new_nsp);
 117         return ERR_PTR(err);
 118 }
 119
 120 /*
 121  * called from clone.  This now handles copy for nsproxy and all
 122  * namespaces therein.
 123  */
 124 int copy_namespaces(unsigned long flags, struct task_struct *tsk)
 125 {
 126         struct nsproxy *old_ns = tsk->nsproxy;
 127         struct nsproxy *new_ns;
 128         int err = 0;
 129
 130         if (!old_ns)
 131                 return 0;
 132
 133         get_nsproxy(old_ns);
 134
 135         if (!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 136                                 CLONE_NEWPID | CLONE_NEWNET)))
 137                 return 0;
 138
 139         if (!capable(CAP_SYS_ADMIN)) {
 140                 err = -EPERM;
 141                 goto out;
 142         }
 143
 144         /*
 145          * CLONE_NEWIPC must detach from the undolist: after switching
 146          * to a new ipc namespace, the semaphore arrays from the old
 147          * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
 148          * means share undolist with parent, so we must forbid using
 149          * it along with CLONE_NEWIPC.
 150          */
 151         if ((flags & CLONE_NEWIPC) && (flags & CLONE_SYSVSEM)) {
 152                 err = -EINVAL;
 153                 goto out;
 154         }
 155
 156         new_ns = create_new_namespaces(flags, tsk, tsk->fs);
 157         if (IS_ERR(new_ns)) {
 158                 err = PTR_ERR(new_ns);
 159                 goto out;
 160         }
 161
 162         tsk->nsproxy = new_ns;
 163
 164 out:
 165         put_nsproxy(old_ns);
 166         return err;
 167 }
 168
 169 void free_nsproxy(struct nsproxy *ns)
 170 {
 171         if (ns->mnt_ns)
 172                 put_mnt_ns(ns->mnt_ns);
 173         if (ns->uts_ns)
 174                 put_uts_ns(ns->uts_ns);
 175         if (ns->ipc_ns)
 176                 put_ipc_ns(ns->ipc_ns);
 177         if (ns->pid_ns)
 178                 put_pid_ns(ns->pid_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_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);