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/export.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 user_namespace *user_ns,
  61         struct fs_struct *new_fs)
  62 {
  63         struct nsproxy *new_nsp;
  64         int err;
  65
  66         new_nsp = create_nsproxy();
  67         if (!new_nsp)
  68                 return ERR_PTR(-ENOMEM);
  69
  70         new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs);
  71         if (IS_ERR(new_nsp->mnt_ns)) {
  72                 err = PTR_ERR(new_nsp->mnt_ns);
  73                 goto out_ns;
  74         }
  75
  76         new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
  77         if (IS_ERR(new_nsp->uts_ns)) {
  78                 err = PTR_ERR(new_nsp->uts_ns);
  79                 goto out_uts;
  80         }
  81
  82         new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
  83         if (IS_ERR(new_nsp->ipc_ns)) {
  84                 err = PTR_ERR(new_nsp->ipc_ns);
  85                 goto out_ipc;
  86         }
  87
  88         new_nsp->pid_ns = copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns);
  89         if (IS_ERR(new_nsp->pid_ns)) {
  90                 err = PTR_ERR(new_nsp->pid_ns);
  91                 goto out_pid;
  92         }
  93
  94         new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
  95         if (IS_ERR(new_nsp->net_ns)) {
  96                 err = PTR_ERR(new_nsp->net_ns);
  97                 goto out_net;
  98         }
  99
 100         return new_nsp;
 101
 102 out_net:
 103         if (new_nsp->pid_ns)
 104                 put_pid_ns(new_nsp->pid_ns);
 105 out_pid:
 106         if (new_nsp->ipc_ns)
 107                 put_ipc_ns(new_nsp->ipc_ns);
 108 out_ipc:
 109         if (new_nsp->uts_ns)
 110                 put_uts_ns(new_nsp->uts_ns);
 111 out_uts:
 112         if (new_nsp->mnt_ns)
 113                 put_mnt_ns(new_nsp->mnt_ns);
 114 out_ns:
 115         kmem_cache_free(nsproxy_cachep, new_nsp);
 116         return ERR_PTR(err);
 117 }
 118
 119 /*
 120  * called from clone.  This now handles copy for nsproxy and all
 121  * namespaces therein.
 122  */
 123 int copy_namespaces(unsigned long flags, struct task_struct *tsk)
 124 {
 125         struct nsproxy *old_ns = tsk->nsproxy;
 126         struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
 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 (!ns_capable(user_ns, 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, user_ns, 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 cred *new_cred, struct fs_struct *new_fs)
 189 {
 190         struct user_namespace *user_ns;
 191         int err = 0;
 192
 193         if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 194                                CLONE_NEWNET | CLONE_NEWPID)))
 195                 return 0;
 196
 197         user_ns = new_cred ? new_cred->user_ns : current_user_ns();
 198         if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 199                 return -EPERM;
 200
 201         *new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
 202                                          new_fs ? new_fs : current->fs);
 203         if (IS_ERR(*new_nsp)) {
 204                 err = PTR_ERR(*new_nsp);
 205                 goto out;
 206         }
 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         file = proc_ns_fget(fd);
 249         if (IS_ERR(file))
 250                 return PTR_ERR(file);
 251
 252         err = -EINVAL;
 253         ei = PROC_I(file->f_dentry->d_inode);
 254         ops = ei->ns_ops;
 255         if (nstype && (ops->type != nstype))
 256                 goto out;
 257
 258         new_nsproxy = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
 259         if (IS_ERR(new_nsproxy)) {
 260                 err = PTR_ERR(new_nsproxy);
 261                 goto out;
 262         }
 263
 264         err = ops->install(new_nsproxy, ei->ns);
 265         if (err) {
 266                 free_nsproxy(new_nsproxy);
 267                 goto out;
 268         }
 269         switch_task_namespaces(tsk, new_nsproxy);
 270 out:
 271         fput(file);
 272         return err;
 273 }
 274
 275 int __init nsproxy_cache_init(void)
 276 {
 277         nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
 278         return 0;
 279 }