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