linux/kernel/nsproxy.c
<<
>>
Prefs
   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/time_namespace.h>
  26#include <linux/proc_ns.h>
  27#include <linux/file.h>
  28#include <linux/syscalls.h>
  29#include <linux/cgroup.h>
  30#include <linux/perf_event.h>
  31
  32static struct kmem_cache *nsproxy_cachep;
  33
  34struct nsproxy init_nsproxy = {
  35        .count                  = ATOMIC_INIT(1),
  36        .uts_ns                 = &init_uts_ns,
  37#if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
  38        .ipc_ns                 = &init_ipc_ns,
  39#endif
  40        .mnt_ns                 = NULL,
  41        .pid_ns_for_children    = &init_pid_ns,
  42#ifdef CONFIG_NET
  43        .net_ns                 = &init_net,
  44#endif
  45#ifdef CONFIG_CGROUPS
  46        .cgroup_ns              = &init_cgroup_ns,
  47#endif
  48#ifdef CONFIG_TIME_NS
  49        .time_ns                = &init_time_ns,
  50        .time_ns_for_children   = &init_time_ns,
  51#endif
  52};
  53
  54static inline struct nsproxy *create_nsproxy(void)
  55{
  56        struct nsproxy *nsproxy;
  57
  58        nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
  59        if (nsproxy)
  60                atomic_set(&nsproxy->count, 1);
  61        return nsproxy;
  62}
  63
  64/*
  65 * Create new nsproxy and all of its the associated namespaces.
  66 * Return the newly created nsproxy.  Do not attach this to the task,
  67 * leave it to the caller to do proper locking and attach it to task.
  68 */
  69static struct nsproxy *create_new_namespaces(unsigned long flags,
  70        struct task_struct *tsk, struct user_namespace *user_ns,
  71        struct fs_struct *new_fs)
  72{
  73        struct nsproxy *new_nsp;
  74        int err;
  75
  76        new_nsp = create_nsproxy();
  77        if (!new_nsp)
  78                return ERR_PTR(-ENOMEM);
  79
  80        new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs);
  81        if (IS_ERR(new_nsp->mnt_ns)) {
  82                err = PTR_ERR(new_nsp->mnt_ns);
  83                goto out_ns;
  84        }
  85
  86        new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
  87        if (IS_ERR(new_nsp->uts_ns)) {
  88                err = PTR_ERR(new_nsp->uts_ns);
  89                goto out_uts;
  90        }
  91
  92        new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
  93        if (IS_ERR(new_nsp->ipc_ns)) {
  94                err = PTR_ERR(new_nsp->ipc_ns);
  95                goto out_ipc;
  96        }
  97
  98        new_nsp->pid_ns_for_children =
  99                copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
 100        if (IS_ERR(new_nsp->pid_ns_for_children)) {
 101                err = PTR_ERR(new_nsp->pid_ns_for_children);
 102                goto out_pid;
 103        }
 104
 105        new_nsp->cgroup_ns = copy_cgroup_ns(flags, user_ns,
 106                                            tsk->nsproxy->cgroup_ns);
 107        if (IS_ERR(new_nsp->cgroup_ns)) {
 108                err = PTR_ERR(new_nsp->cgroup_ns);
 109                goto out_cgroup;
 110        }
 111
 112        new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
 113        if (IS_ERR(new_nsp->net_ns)) {
 114                err = PTR_ERR(new_nsp->net_ns);
 115                goto out_net;
 116        }
 117
 118        new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns,
 119                                        tsk->nsproxy->time_ns_for_children);
 120        if (IS_ERR(new_nsp->time_ns_for_children)) {
 121                err = PTR_ERR(new_nsp->time_ns_for_children);
 122                goto out_time;
 123        }
 124        new_nsp->time_ns = get_time_ns(tsk->nsproxy->time_ns);
 125
 126        return new_nsp;
 127
 128out_time:
 129        put_net(new_nsp->net_ns);
 130out_net:
 131        put_cgroup_ns(new_nsp->cgroup_ns);
 132out_cgroup:
 133        if (new_nsp->pid_ns_for_children)
 134                put_pid_ns(new_nsp->pid_ns_for_children);
 135out_pid:
 136        if (new_nsp->ipc_ns)
 137                put_ipc_ns(new_nsp->ipc_ns);
 138out_ipc:
 139        if (new_nsp->uts_ns)
 140                put_uts_ns(new_nsp->uts_ns);
 141out_uts:
 142        if (new_nsp->mnt_ns)
 143                put_mnt_ns(new_nsp->mnt_ns);
 144out_ns:
 145        kmem_cache_free(nsproxy_cachep, new_nsp);
 146        return ERR_PTR(err);
 147}
 148
 149/*
 150 * called from clone.  This now handles copy for nsproxy and all
 151 * namespaces therein.
 152 */
 153int copy_namespaces(unsigned long flags, struct task_struct *tsk)
 154{
 155        struct nsproxy *old_ns = tsk->nsproxy;
 156        struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
 157        struct nsproxy *new_ns;
 158        int ret;
 159
 160        if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 161                              CLONE_NEWPID | CLONE_NEWNET |
 162                              CLONE_NEWCGROUP | CLONE_NEWTIME)))) {
 163                if (likely(old_ns->time_ns_for_children == old_ns->time_ns)) {
 164                        get_nsproxy(old_ns);
 165                        return 0;
 166                }
 167        } else if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 168                return -EPERM;
 169
 170        /*
 171         * CLONE_NEWIPC must detach from the undolist: after switching
 172         * to a new ipc namespace, the semaphore arrays from the old
 173         * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
 174         * means share undolist with parent, so we must forbid using
 175         * it along with CLONE_NEWIPC.
 176         */
 177        if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
 178                (CLONE_NEWIPC | CLONE_SYSVSEM)) 
 179                return -EINVAL;
 180
 181        new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
 182        if (IS_ERR(new_ns))
 183                return  PTR_ERR(new_ns);
 184
 185        ret = timens_on_fork(new_ns, tsk);
 186        if (ret) {
 187                free_nsproxy(new_ns);
 188                return ret;
 189        }
 190
 191        tsk->nsproxy = new_ns;
 192        return 0;
 193}
 194
 195void free_nsproxy(struct nsproxy *ns)
 196{
 197        if (ns->mnt_ns)
 198                put_mnt_ns(ns->mnt_ns);
 199        if (ns->uts_ns)
 200                put_uts_ns(ns->uts_ns);
 201        if (ns->ipc_ns)
 202                put_ipc_ns(ns->ipc_ns);
 203        if (ns->pid_ns_for_children)
 204                put_pid_ns(ns->pid_ns_for_children);
 205        if (ns->time_ns)
 206                put_time_ns(ns->time_ns);
 207        if (ns->time_ns_for_children)
 208                put_time_ns(ns->time_ns_for_children);
 209        put_cgroup_ns(ns->cgroup_ns);
 210        put_net(ns->net_ns);
 211        kmem_cache_free(nsproxy_cachep, ns);
 212}
 213
 214/*
 215 * Called from unshare. Unshare all the namespaces part of nsproxy.
 216 * On success, returns the new nsproxy.
 217 */
 218int unshare_nsproxy_namespaces(unsigned long unshare_flags,
 219        struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
 220{
 221        struct user_namespace *user_ns;
 222        int err = 0;
 223
 224        if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
 225                               CLONE_NEWNET | CLONE_NEWPID | CLONE_NEWCGROUP |
 226                               CLONE_NEWTIME)))
 227                return 0;
 228
 229        user_ns = new_cred ? new_cred->user_ns : current_user_ns();
 230        if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 231                return -EPERM;
 232
 233        *new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
 234                                         new_fs ? new_fs : current->fs);
 235        if (IS_ERR(*new_nsp)) {
 236                err = PTR_ERR(*new_nsp);
 237                goto out;
 238        }
 239
 240out:
 241        return err;
 242}
 243
 244void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
 245{
 246        struct nsproxy *ns;
 247
 248        might_sleep();
 249
 250        task_lock(p);
 251        ns = p->nsproxy;
 252        p->nsproxy = new;
 253        task_unlock(p);
 254
 255        if (ns && atomic_dec_and_test(&ns->count))
 256                free_nsproxy(ns);
 257}
 258
 259void exit_task_namespaces(struct task_struct *p)
 260{
 261        switch_task_namespaces(p, NULL);
 262}
 263
 264SYSCALL_DEFINE2(setns, int, fd, int, nstype)
 265{
 266        struct task_struct *tsk = current;
 267        struct nsproxy *new_nsproxy;
 268        struct file *file;
 269        struct ns_common *ns;
 270        int err;
 271
 272        file = proc_ns_fget(fd);
 273        if (IS_ERR(file))
 274                return PTR_ERR(file);
 275
 276        err = -EINVAL;
 277        ns = get_proc_ns(file_inode(file));
 278        if (nstype && (ns->ops->type != nstype))
 279                goto out;
 280
 281        new_nsproxy = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
 282        if (IS_ERR(new_nsproxy)) {
 283                err = PTR_ERR(new_nsproxy);
 284                goto out;
 285        }
 286
 287        err = ns->ops->install(new_nsproxy, ns);
 288        if (err) {
 289                free_nsproxy(new_nsproxy);
 290                goto out;
 291        }
 292        switch_task_namespaces(tsk, new_nsproxy);
 293
 294        perf_event_namespaces(tsk);
 295out:
 296        fput(file);
 297        return err;
 298}
 299
 300int __init nsproxy_cache_init(void)
 301{
 302        nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
 303        return 0;
 304}
 305