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 #include <linux/cgroup.h>
  29 #include <linux/perf_event.h>
  30
  31 static struct kmem_cache *nsproxy_cachep;
  32
  33 struct nsproxy init_nsproxy = {
  34         .count                  = ATOMIC_INIT(1),
  35         .uts_ns                 = &init_uts_ns,
  36 #if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
  37         .ipc_ns                 = &init_ipc_ns,
  38 #endif
  39         .mnt_ns                 = NULL,
  40         .pid_ns_for_children    = &init_pid_ns,
  41 #ifdef CONFIG_NET
  42         .net_ns                 = &init_net,
  43 #endif
  44 #ifdef CONFIG_CGROUPS
  45         .cgroup_ns              = &init_cgroup_ns,
  46 #endif
  47 };
  48
  49 static inline struct nsproxy *create_nsproxy(void)
  50 {
  51         struct nsproxy *nsproxy;
  52
  53         nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
  54         if (nsproxy)
  55                 atomic_set(&nsproxy->count, 1);
  56         return nsproxy;
  57 }
  58
  59 /*
  60  * Create new nsproxy and all of its the associated namespaces.
  61  * Return the newly created nsproxy.  Do not attach this to the task,
  62  * leave it to the caller to do proper locking and attach it to task.
  63  */
  64 static struct nsproxy *create_new_namespaces(unsigned long flags,
  65         struct task_struct *tsk, struct user_namespace *user_ns,
  66         struct fs_struct *new_fs)
  67 {
  68         struct nsproxy *new_nsp;
  69         int err;
  70
  71         new_nsp = create_nsproxy();
  72         if (!new_nsp)
  73                 return ERR_PTR(-ENOMEM);
  74
  75         new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs);
  76         if (IS_ERR(new_nsp->mnt_ns)) {
  77                 err = PTR_ERR(new_nsp->mnt_ns);
  78                 goto out_ns;
  79         }
  80
  81         new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
  82         if (IS_ERR(new_nsp->uts_ns)) {
  83                 err = PTR_ERR(new_nsp->uts_ns);
  84                 goto out_uts;
  85         }
  86
  87         new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
  88         if (IS_ERR(new_nsp->ipc_ns)) {
  89                 err = PTR_ERR(new_nsp->ipc_ns);
  90                 goto out_ipc;
  91         }
  92
  93         new_nsp->pid_ns_for_children =
  94                 copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
  95         if (IS_ERR(new_nsp->pid_ns_for_children)) {
  96                 err = PTR_ERR(new_nsp->pid_ns_for_children);
  97                 goto out_pid;
  98         }
  99
 100         new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns,
 101                                             tsk->nsproxy->cgroup_ns);
 102         if (IS_ERR(new_nsp->cgroup_ns)) {
 103                 err = PTR_ERR(new_nsp->cgroup_ns);
 104                 goto out_cgroup;
 105         }
 106
 107         new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
 108         if (IS_ERR(new_nsp->net_ns)) {
 109                 err = PTR_ERR(new_nsp->net_ns);
 110                 goto out_net;
 111         }
 112
 113         return new_nsp;
 114
 115 out_net:
 116         put_cgroup_ns(new_nsp->cgroup_ns);
 117 out_cgroup:
 118         if (new_nsp->pid_ns_for_children)
 119                 put_pid_ns(new_nsp->pid_ns_for_children);
 120 out_pid:
 121         if (new_nsp->ipc_ns)
 122                 put_ipc_ns(new_nsp->ipc_ns);
 123 out_ipc:
 124         if (new_nsp->uts_ns)
 125                 put_uts_ns(new_nsp->uts_ns);
 126 out_uts:
 127         if (new_nsp->mnt_ns)
 128                 put_mnt_ns(new_nsp->mnt_ns);
 129 out_ns:
 130         kmem_cache_free(nsproxy_cachep, new_nsp);
 131         return ERR_PTR(err);
 132 }
 133
 134 /*
 135  * called from clone.  This now handles copy for nsproxy and all
 136  * namespaces therein.
 137  */
 138 int copy_namespaces(unsigned long flags, struct task_struct *tsk)
 139 {
 140         struct nsproxy *old_ns = tsk->nsproxy;
 141         struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
 142         struct nsproxy *new_ns;
 143
 144         if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 145                               CLONE_NEWPID | CLONE_NEWNET |
 146                               CLONE_NEWCGROUP)))) {
 147                 get_nsproxy(old_ns);
 148                 return 0;
 149         }
 150
 151         if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 152                 return -EPERM;
 153
 154         /*
 155          * CLONE_NEWIPC must detach from the undolist: after switching
 156          * to a new ipc namespace, the semaphore arrays from the old
 157          * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
 158          * means share undolist with parent, so we must forbid using
 159          * it along with CLONE_NEWIPC.
 160          */
 161         if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
 162                 (CLONE_NEWIPC | CLONE_SYSVSEM))
 163                 return -EINVAL;
 164
 165         new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
 166         if (IS_ERR(new_ns))
 167                 return  PTR_ERR(new_ns);
 168
 169         tsk->nsproxy = new_ns;
 170         return 0;
 171 }
 172
 173 void free_nsproxy(struct nsproxy *ns)
 174 {
 175         if (ns->mnt_ns)
 176                 put_mnt_ns(ns->mnt_ns);
 177         if (ns->uts_ns)
 178                 put_uts_ns(ns->uts_ns);
 179         if (ns->ipc_ns)
 180                 put_ipc_ns(ns->ipc_ns);
 181         if (ns->pid_ns_for_children)
 182                 put_pid_ns(ns->pid_ns_for_children);
 183         put_cgroup_ns(ns->cgroup_ns);
 184         put_net(ns->net_ns);
 185         kmem_cache_free(nsproxy_cachep, ns);
 186 }
 187
 188 /*
 189  * Called from unshare. Unshare all the namespaces part of nsproxy.
 190  * On success, returns the new nsproxy.
 191  */
 192 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
 193         struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
 194 {
 195         struct user_namespace *user_ns;
 196         int err = 0;
 197
 198         if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 199                                CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP)))
 200                 return 0;
 201
 202         user_ns = new_cred ? new_cred->user_ns : current_user_ns();
 203         if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 204                 return -EPERM;
 205
 206         *new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
 207                                          new_fs ? new_fs : current->fs);
 208         if (IS_ERR(*new_nsp)) {
 209                 err = PTR_ERR(*new_nsp);
 210                 goto out;
 211         }
 212
 213 out:
 214         return err;
 215 }
 216
 217 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
 218 {
 219         struct nsproxy *ns;
 220
 221         might_sleep();
 222
 223         task_lock(p);
 224         ns = p->nsproxy;
 225         p->nsproxy = new;
 226         task_unlock(p);
 227
 228         if (ns && atomic_dec_and_test(&ns->count))
 229                 free_nsproxy(ns);
 230 }
 231
 232 void exit_task_namespaces(struct task_struct *p)
 233 {
 234         switch_task_namespaces(p, NULL);
 235 }
 236
 237 SYSCALL_DEFINE2(setns, int, fd, int, nstype)
 238 {
 239         struct task_struct *tsk = current;
 240         struct nsproxy *new_nsproxy;
 241         struct file *file;
 242         struct ns_common *ns;
 243         int err;
 244
 245         file = proc_ns_fget(fd);
 246         if (IS_ERR(file))
 247                 return PTR_ERR(file);
 248
 249         err = -EINVAL;
 250         ns = get_proc_ns(file_inode(file));
 251         if (nstype && (ns->ops->type != nstype))
 252                 goto out;
 253
 254         new_nsproxy = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
 255         if (IS_ERR(new_nsproxy)) {
 256                 err = PTR_ERR(new_nsproxy);
 257                 goto out;
 258         }
 259
 260         err = ns->ops->install(new_nsproxy, ns);
 261         if (err) {
 262                 free_nsproxy(new_nsproxy);
 263                 goto out;
 264         }
 265         switch_task_namespaces(tsk, new_nsproxy);
 266
 267         perf_event_namespaces(tsk);
 268 out:
 269         fput(file);
 270         return err;
 271 }
 272
 273 int __init nsproxy_cache_init(void)
 274 {
 275         nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
 276         return 0;
 277 }