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_ns.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_for_children    = &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_for_children =
  89                 copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
  90         if (IS_ERR(new_nsp->pid_ns_for_children)) {
  91                 err = PTR_ERR(new_nsp->pid_ns_for_children);
  92                 goto out_pid;
  93         }
  94
  95         new_nsp->net_ns = copy_net_ns(flags, user_ns, 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_for_children)
 105                 put_pid_ns(new_nsp->pid_ns_for_children);
 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 user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
 128         struct nsproxy *new_ns;
 129
 130         if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 131                               CLONE_NEWPID | CLONE_NEWNET)))) {
 132                 get_nsproxy(old_ns);
 133                 return 0;
 134         }
 135
 136         if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 137                 return -EPERM;
 138
 139         /*
 140          * CLONE_NEWIPC must detach from the undolist: after switching
 141          * to a new ipc namespace, the semaphore arrays from the old
 142          * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
 143          * means share undolist with parent, so we must forbid using
 144          * it along with CLONE_NEWIPC.
 145          */
 146         if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
 147                 (CLONE_NEWIPC | CLONE_SYSVSEM))
 148                 return -EINVAL;
 149
 150         new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
 151         if (IS_ERR(new_ns))
 152                 return  PTR_ERR(new_ns);
 153
 154         tsk->nsproxy = new_ns;
 155         return 0;
 156 }
 157
 158 void free_nsproxy(struct nsproxy *ns)
 159 {
 160         if (ns->mnt_ns)
 161                 put_mnt_ns(ns->mnt_ns);
 162         if (ns->uts_ns)
 163                 put_uts_ns(ns->uts_ns);
 164         if (ns->ipc_ns)
 165                 put_ipc_ns(ns->ipc_ns);
 166         if (ns->pid_ns_for_children)
 167                 put_pid_ns(ns->pid_ns_for_children);
 168         put_net(ns->net_ns);
 169         kmem_cache_free(nsproxy_cachep, ns);
 170 }
 171
 172 /*
 173  * Called from unshare. Unshare all the namespaces part of nsproxy.
 174  * On success, returns the new nsproxy.
 175  */
 176 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
 177         struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
 178 {
 179         struct user_namespace *user_ns;
 180         int err = 0;
 181
 182         if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 183                                CLONE_NEWNET | CLONE_NEWPID)))
 184                 return 0;
 185
 186         user_ns = new_cred ? new_cred->user_ns : current_user_ns();
 187         if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 188                 return -EPERM;
 189
 190         *new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
 191                                          new_fs ? new_fs : current->fs);
 192         if (IS_ERR(*new_nsp)) {
 193                 err = PTR_ERR(*new_nsp);
 194                 goto out;
 195         }
 196
 197 out:
 198         return err;
 199 }
 200
 201 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
 202 {
 203         struct nsproxy *ns;
 204
 205         might_sleep();
 206
 207         ns = p->nsproxy;
 208
 209         rcu_assign_pointer(p->nsproxy, new);
 210
 211         if (ns && atomic_dec_and_test(&ns->count)) {
 212                 /*
 213                  * wait for others to get what they want from this nsproxy.
 214                  *
 215                  * cannot release this nsproxy via the call_rcu() since
 216                  * put_mnt_ns() will want to sleep
 217                  */
 218                 synchronize_rcu();
 219                 free_nsproxy(ns);
 220         }
 221 }
 222
 223 void exit_task_namespaces(struct task_struct *p)
 224 {
 225         switch_task_namespaces(p, NULL);
 226 }
 227
 228 SYSCALL_DEFINE2(setns, int, fd, int, nstype)
 229 {
 230         const struct proc_ns_operations *ops;
 231         struct task_struct *tsk = current;
 232         struct nsproxy *new_nsproxy;
 233         struct proc_ns *ei;
 234         struct file *file;
 235         int err;
 236
 237         file = proc_ns_fget(fd);
 238         if (IS_ERR(file))
 239                 return PTR_ERR(file);
 240
 241         err = -EINVAL;
 242         ei = get_proc_ns(file_inode(file));
 243         ops = ei->ns_ops;
 244         if (nstype && (ops->type != nstype))
 245                 goto out;
 246
 247         new_nsproxy = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
 248         if (IS_ERR(new_nsproxy)) {
 249                 err = PTR_ERR(new_nsproxy);
 250                 goto out;
 251         }
 252
 253         err = ops->install(new_nsproxy, ei->ns);
 254         if (err) {
 255                 free_nsproxy(new_nsproxy);
 256                 goto out;
 257         }
 258         switch_task_namespaces(tsk, new_nsproxy);
 259 out:
 260         fput(file);
 261         return err;
 262 }
 263
 264 int __init nsproxy_cache_init(void)
 265 {
 266         nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
 267         return 0;
 268 }