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