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 #include <linux/proc_fs.h>
  26 #include <linux/file.h>
  27 #include <linux/syscalls.h>
  28
  29 static struct kmem_cache *nsproxy_cachep;
  30
  31 struct nsproxy init_nsproxy = {
  32         .count  = ATOMIC_INIT(1),
  33         .uts_ns = &init_uts_ns,
  34 #if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
  35         .ipc_ns = &init_ipc_ns,
  36 #endif
  37         .mnt_ns = NULL,
  38         .pid_ns = &init_pid_ns,
  39 #ifdef CONFIG_NET
  40         .net_ns = &init_net,
  41 #endif
  42 };
  43
  44 static inline struct nsproxy *create_nsproxy(void)
  45 {
  46         struct nsproxy *nsproxy;
  47
  48         nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
  49         if (nsproxy)
  50                 atomic_set(&nsproxy->count, 1);
  51         return nsproxy;
  52 }
  53
  54 /*
  55  * Create new nsproxy and all of its the associated namespaces.
  56  * Return the newly created nsproxy.  Do not attach this to the task,
  57  * leave it to the caller to do proper locking and attach it to task.
  58  */
  59 static struct nsproxy *create_new_namespaces(unsigned long flags,
  60                         struct task_struct *tsk, struct fs_struct *new_fs)
  61 {
  62         struct nsproxy *new_nsp;
  63         int err;
  64
  65         new_nsp = create_nsproxy();
  66         if (!new_nsp)
  67                 return ERR_PTR(-ENOMEM);
  68
  69         new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, new_fs);
  70         if (IS_ERR(new_nsp->mnt_ns)) {
  71                 err = PTR_ERR(new_nsp->mnt_ns);
  72                 goto out_ns;
  73         }
  74
  75         new_nsp->uts_ns = copy_utsname(flags, tsk);
  76         if (IS_ERR(new_nsp->uts_ns)) {
  77                 err = PTR_ERR(new_nsp->uts_ns);
  78                 goto out_uts;
  79         }
  80
  81         new_nsp->ipc_ns = copy_ipcs(flags, tsk);
  82         if (IS_ERR(new_nsp->ipc_ns)) {
  83                 err = PTR_ERR(new_nsp->ipc_ns);
  84                 goto out_ipc;
  85         }
  86
  87         new_nsp->pid_ns = copy_pid_ns(flags, task_active_pid_ns(tsk));
  88         if (IS_ERR(new_nsp->pid_ns)) {
  89                 err = PTR_ERR(new_nsp->pid_ns);
  90                 goto out_pid;
  91         }
  92
  93         new_nsp->net_ns = copy_net_ns(flags, tsk->nsproxy->net_ns);
  94         if (IS_ERR(new_nsp->net_ns)) {
  95                 err = PTR_ERR(new_nsp->net_ns);
  96                 goto out_net;
  97         }
  98
  99         return new_nsp;
 100
 101 out_net:
 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_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         put_net(ns->net_ns);
 178         kmem_cache_free(nsproxy_cachep, ns);
 179 }
 180
 181 /*
 182  * Called from unshare. Unshare all the namespaces part of nsproxy.
 183  * On success, returns the new nsproxy.
 184  */
 185 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
 186                 struct nsproxy **new_nsp, struct fs_struct *new_fs)
 187 {
 188         int err = 0;
 189
 190         if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 191                                CLONE_NEWNET)))
 192                 return 0;
 193
 194         if (!capable(CAP_SYS_ADMIN))
 195                 return -EPERM;
 196
 197         *new_nsp = create_new_namespaces(unshare_flags, current,
 198                                 new_fs ? new_fs : current->fs);
 199         if (IS_ERR(*new_nsp)) {
 200                 err = PTR_ERR(*new_nsp);
 201                 goto out;
 202         }
 203
 204         err = ns_cgroup_clone(current, task_pid(current));
 205         if (err)
 206                 put_nsproxy(*new_nsp);
 207
 208 out:
 209         return err;
 210 }
 211
 212 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
 213 {
 214         struct nsproxy *ns;
 215
 216         might_sleep();
 217
 218         ns = p->nsproxy;
 219
 220         rcu_assign_pointer(p->nsproxy, new);
 221
 222         if (ns && atomic_dec_and_test(&ns->count)) {
 223                 /*
 224                  * wait for others to get what they want from this nsproxy.
 225                  *
 226                  * cannot release this nsproxy via the call_rcu() since
 227                  * put_mnt_ns() will want to sleep
 228                  */
 229                 synchronize_rcu();
 230                 free_nsproxy(ns);
 231         }
 232 }
 233
 234 void exit_task_namespaces(struct task_struct *p)
 235 {
 236         switch_task_namespaces(p, NULL);
 237 }
 238
 239 SYSCALL_DEFINE2(setns, int, fd, int, nstype)
 240 {
 241         const struct proc_ns_operations *ops;
 242         struct task_struct *tsk = current;
 243         struct nsproxy *new_nsproxy;
 244         struct proc_inode *ei;
 245         struct file *file;
 246         int err;
 247
 248         if (!capable(CAP_SYS_ADMIN))
 249                 return -EPERM;
 250
 251         file = proc_ns_fget(fd);
 252         if (IS_ERR(file))
 253                 return PTR_ERR(file);
 254
 255         err = -EINVAL;
 256         ei = PROC_I(file->f_dentry->d_inode);
 257         ops = ei->ns_ops;
 258         if (nstype && (ops->type != nstype))
 259                 goto out;
 260
 261         new_nsproxy = create_new_namespaces(0, tsk, tsk->fs);
 262         if (IS_ERR(new_nsproxy)) {
 263                 err = PTR_ERR(new_nsproxy);
 264                 goto out;
 265         }
 266
 267         err = ops->install(new_nsproxy, ei->ns);
 268         if (err) {
 269                 free_nsproxy(new_nsproxy);
 270                 goto out;
 271         }
 272         switch_task_namespaces(tsk, new_nsproxy);
 273 out:
 274         fput(file);
 275         return err;
 276 }
 277
 278 static int __init nsproxy_cache_init(void)
 279 {
 280         nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
 281         return 0;
 282 }
 283
 284 module_init(nsproxy_cache_init);