src/lxc/namespace.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2009
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #ifndef _GNU_SOURCE
  25 #define _GNU_SOURCE 1
  26 #endif
  27 #include <alloca.h>
  28 #include <errno.h>
  29 #include <fcntl.h>
  30 #include <sched.h>
  31 #include <signal.h>
  32 #include <sys/param.h>
  33 #include <sys/stat.h>
  34 #include <sys/syscall.h>
  35 #include <sys/types.h>
  36 #include <unistd.h>
  37
  38 #include "config.h"
  39 #include "log.h"
  40 #include "namespace.h"
  41 #include "utils.h"
  42
  43 lxc_log_define(namespace, lxc);
  44
  45 struct clone_arg {
  46         int (*fn)(void *);
  47         void *arg;
  48 };
  49
  50 static int do_clone(void *arg)
  51 {
  52         struct clone_arg *clone_arg = arg;
  53         return clone_arg->fn(clone_arg->arg);
  54 }
  55
  56 pid_t lxc_clone(int (*fn)(void *), void *arg, int flags)
  57 {
  58         struct clone_arg clone_arg = {
  59                 .fn = fn,
  60                 .arg = arg,
  61         };
  62
  63         size_t stack_size = lxc_getpagesize();
  64         void *stack = alloca(stack_size);
  65         pid_t ret;
  66
  67 #ifdef __ia64__
  68         ret = __clone2(do_clone, stack, stack_size, flags | SIGCHLD, &clone_arg);
  69 #else
  70         ret = clone(do_clone, stack  + stack_size, flags | SIGCHLD, &clone_arg);
  71 #endif
  72         if (ret < 0)
  73                 SYSERROR("Failed to clone (%#x)", flags);
  74
  75         return ret;
  76 }
  77
  78 /**
  79  * This is based on raw_clone in systemd but adapted to our needs. This uses
  80  * copy on write semantics and doesn't pass a stack. CLONE_VM is tricky and
  81  * doesn't really matter to us so disallow it.
  82  *
  83  * The nice thing about this is that we get fork() behavior. That is
  84  * lxc_raw_clone() returns 0 in the child and the child pid in the parent.
  85  */
  86 pid_t lxc_raw_clone(unsigned long flags)
  87 {
  88
  89         /* These flags don't interest at all so we don't jump through any hoopes
  90          * of retrieving them and passing them to the kernel.
  91          */
  92         errno = EINVAL;
  93         if ((flags & (CLONE_VM | CLONE_PARENT_SETTID | CLONE_CHILD_SETTID |
  94                       CLONE_CHILD_CLEARTID | CLONE_SETTLS)))
  95                 return -EINVAL;
  96
  97 #if defined(__s390x__) || defined(__s390__) || defined(__CRIS__)
  98         /* On s390/s390x and cris the order of the first and second arguments
  99          * of the system call is reversed.
 100          */
 101         return (int)syscall(__NR_clone, NULL, flags | SIGCHLD);
 102 #elif defined(__sparc__) && defined(__arch64__)
 103         {
 104                 /**
 105                  * sparc64 always returns the other process id in %o0, and
 106                  * a boolean flag whether this is the child or the parent in
 107                  * %o1. Inline assembly is needed to get the flag returned
 108                  * in %o1.
 109                  */
 110                 int in_child;
 111                 int child_pid;
 112                 asm volatile("mov %2, %%g1\n\t"
 113                              "mov %3, %%o0\n\t"
 114                              "mov 0 , %%o1\n\t"
 115                              "t 0x6d\n\t"
 116                              "mov %%o1, %0\n\t"
 117                              "mov %%o0, %1"
 118                              : "=r"(in_child), "=r"(child_pid)
 119                              : "i"(__NR_clone), "r"(flags | SIGCHLD)
 120                              : "%o1", "%o0", "%g1");
 121
 122                 if (in_child)
 123                         return 0;
 124                 else
 125                         return child_pid;
 126         }
 127 #elif defined(__ia64__)
 128         /* On ia64 the stack and stack size are passed as separate arguments. */
 129         return (int)syscall(__NR_clone, flags | SIGCHLD, NULL, 0);
 130 #else
 131         return (int)syscall(__NR_clone, flags | SIGCHLD, NULL);
 132 #endif
 133 }
 134
 135 pid_t lxc_raw_clone_cb(int (*fn)(void *), void *args, unsigned long flags)
 136 {
 137         pid_t pid;
 138
 139         pid = lxc_raw_clone(flags);
 140         if (pid < 0)
 141                 return -1;
 142
 143         /* exit() is not thread-safe and might mess with the parent's signal
 144          * handlers and other stuff when exec() fails.
 145          */
 146         if (pid == 0)
 147                 _exit(fn(args));
 148
 149         return pid;
 150 }
 151
 152 /* Leave the user namespace at the first position in the array of structs so
 153  * that we always attach to it first when iterating over the struct and using
 154  * setns() to switch namespaces. This especially affects lxc_attach(): Suppose
 155  * you cloned a new user namespace and mount namespace as an unprivileged user
 156  * on the host and want to setns() to the mount namespace. This requires you to
 157  * attach to the user namespace first otherwise the kernel will fail this check:
 158  *
 159  *        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
 160  *            !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
 161  *            !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 162  *            return -EPERM;
 163  *
 164  *    in
 165  *
 166  *        linux/fs/namespace.c:mntns_install().
 167  */
 168 const struct ns_info ns_info[LXC_NS_MAX] = {
 169         [LXC_NS_USER]    = { "user",   CLONE_NEWUSER,   "CLONE_NEWUSER",   "LXC_USER_NS"    },
 170         [LXC_NS_MNT]    =  { "mnt",    CLONE_NEWNS,     "CLONE_NEWNS",     "LXC_MNT_NS"     },
 171         [LXC_NS_PID]    =  { "pid",    CLONE_NEWPID,    "CLONE_NEWPID",    "LXC_PID_NS"     },
 172         [LXC_NS_UTS]    =  { "uts",    CLONE_NEWUTS,    "CLONE_NEWUTS",    "LXC_UTS_NS"     },
 173         [LXC_NS_IPC]    =  { "ipc",    CLONE_NEWIPC,    "CLONE_NEWIPC",    "LXC_IPC_NS"     },
 174         [LXC_NS_NET]    =  { "net",    CLONE_NEWNET,    "CLONE_NEWNET",    "LXC_NET_NS"     },
 175         [LXC_NS_CGROUP] =  { "cgroup", CLONE_NEWCGROUP, "CLONE_NEWCGROUP", "LXC_CGROUP_NS"  }
 176 };
 177
 178 int lxc_namespace_2_cloneflag(const char *namespace)
 179 {
 180         int i;
 181
 182         for (i = 0; i < LXC_NS_MAX; i++)
 183                 if (!strcasecmp(ns_info[i].proc_name, namespace))
 184                         return ns_info[i].clone_flag;
 185
 186         ERROR("Invalid namespace name \"%s\"", namespace);
 187         return -EINVAL;
 188 }
 189
 190 int lxc_namespace_2_ns_idx(const char *namespace)
 191 {
 192         int i;
 193
 194         for (i = 0; i < LXC_NS_MAX; i++)
 195                 if (!strcmp(ns_info[i].proc_name, namespace))
 196                         return i;
 197
 198         ERROR("Invalid namespace name \"%s\"", namespace);
 199         return -EINVAL;
 200 }
 201
 202 extern int lxc_namespace_2_std_identifiers(char *namespaces)
 203 {
 204         char **it;
 205         char *del;
 206
 207         /* The identifiers for namespaces used with lxc-attach and lxc-unshare
 208          * as given on the manpage do not align with the standard identifiers.
 209          * This affects network, mount, and uts namespaces. The standard identifiers
 210          * are: "mnt", "uts", and "net" whereas lxc-attach and lxc-unshare uses
 211          * "MOUNT", "UTSNAME", and "NETWORK". So let's use some cheap memmove()s
 212          * to replace them by their standard identifiers.
 213          * Let's illustrate this with an example:
 214          * Assume the string:
 215          *
 216          *      "IPC|MOUNT|PID"
 217          *
 218          * then we memmove()
 219          *
 220          *      dest: del + 1 == OUNT|PID
 221          *      src:  del + 3 == NT|PID
 222          */
 223         if (!namespaces)
 224                 return -1;
 225
 226         while ((del = strstr(namespaces, "MOUNT")))
 227                 memmove(del + 1, del + 3, strlen(del) - 2);
 228
 229         for (it = (char *[]){"NETWORK", "UTSNAME", NULL}; it && *it; it++)
 230                 while ((del = strstr(namespaces, *it)))
 231                         memmove(del + 3, del + 7, strlen(del) - 6);
 232
 233         return 0;
 234 }
 235
 236 int lxc_fill_namespace_flags(char *flaglist, int *flags)
 237 {
 238         char *token;
 239         int aflag;
 240
 241         if (!flaglist) {
 242                 ERROR("At least one namespace is needed.");
 243                 return -1;
 244         }
 245
 246         lxc_iterate_parts(token, flaglist, "|") {
 247                 aflag = lxc_namespace_2_cloneflag(token);
 248                 if (aflag < 0)
 249                         return -1;
 250
 251                 *flags |= aflag;
 252         }
 253
 254         return 0;
 255 }