src/lxc/attach.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   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 <errno.h>
  28 #include <fcntl.h>
  29 #include <grp.h>
  30 #include <linux/unistd.h>
  31 #include <pwd.h>
  32 #include <pthread.h>
  33 #include <signal.h>
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <string.h>
  37 #include <sys/mount.h>
  38 #include <sys/param.h>
  39 #include <sys/prctl.h>
  40 #include <sys/socket.h>
  41 #include <sys/syscall.h>
  42 #include <sys/wait.h>
  43 #include <termios.h>
  44 #include <unistd.h>
  45
  46 #include <lxc/lxccontainer.h>
  47
  48 #include "af_unix.h"
  49 #include "attach.h"
  50 #include "caps.h"
  51 #include "cgroup.h"
  52 #include "commands.h"
  53 #include "conf.h"
  54 #include "config.h"
  55 #include "confile.h"
  56 #include "log.h"
  57 #include "lsm/lsm.h"
  58 #include "lxclock.h"
  59 #include "lxcseccomp.h"
  60 #include "macro.h"
  61 #include "mainloop.h"
  62 #include "memory_utils.h"
  63 #include "namespace.h"
  64 #include "raw_syscalls.h"
  65 #include "syscall_wrappers.h"
  66 #include "terminal.h"
  67 #include "utils.h"
  68
  69 #if HAVE_SYS_PERSONALITY_H
  70 #include <sys/personality.h>
  71 #endif
  72
  73 lxc_log_define(attach, lxc);
  74
  75 /* Define default options if no options are supplied by the user. */
  76 static lxc_attach_options_t attach_static_default_options = LXC_ATTACH_OPTIONS_DEFAULT;
  77
  78 static struct lxc_proc_context_info *lxc_proc_get_context_info(pid_t pid)
  79 {
  80         __do_free char *line = NULL;
  81         __do_fclose FILE *proc_file = NULL;
  82         int ret;
  83         bool found;
  84         char proc_fn[LXC_PROC_STATUS_LEN];
  85         struct lxc_proc_context_info *info;
  86         size_t line_bufsz = 0;
  87
  88         /* Read capabilities. */
  89         ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", pid);
  90         if (ret < 0 || ret >= LXC_PROC_STATUS_LEN)
  91                 return NULL;
  92
  93         proc_file = fopen(proc_fn, "r");
  94         if (!proc_file) {
  95                 SYSERROR("Failed to open %s", proc_fn);
  96                 return NULL;
  97         }
  98
  99         info = calloc(1, sizeof(*info));
 100         if (!info)
 101                 return NULL;
 102
 103         found = false;
 104
 105         while (getline(&line, &line_bufsz, proc_file) != -1) {
 106                 ret = sscanf(line, "CapBnd: %llx", &info->capability_mask);
 107                 if (ret != EOF && ret == 1) {
 108                         found = true;
 109                         break;
 110                 }
 111         }
 112
 113         if (!found) {
 114                 ERROR("Could not read capability bounding set from %s", proc_fn);
 115                 free(info);
 116                 return NULL;
 117         }
 118
 119         info->lsm_label = lsm_process_label_get(pid);
 120         info->ns_inherited = 0;
 121         memset(info->ns_fd, -1, sizeof(int) * LXC_NS_MAX);
 122
 123         return info;
 124 }
 125
 126 static inline void lxc_proc_close_ns_fd(struct lxc_proc_context_info *ctx)
 127 {
 128         for (int i = 0; i < LXC_NS_MAX; i++) {
 129                 __do_close_prot_errno int fd = move_fd(ctx->ns_fd[i]);
 130         }
 131 }
 132
 133 static void lxc_proc_put_context_info(struct lxc_proc_context_info *ctx)
 134 {
 135         free(ctx->lsm_label);
 136         ctx->lsm_label = NULL;
 137
 138         if (ctx->container) {
 139                 lxc_container_put(ctx->container);
 140                 ctx->container = NULL;
 141         }
 142
 143         lxc_proc_close_ns_fd(ctx);
 144         free(ctx);
 145 }
 146
 147 /**
 148  * in_same_namespace - Check whether two processes are in the same namespace.
 149  * @pid1 - PID of the first process.
 150  * @pid2 - PID of the second process.
 151  * @ns   - Name of the namespace to check. Must correspond to one of the names
 152  *         for the namespaces as shown in /proc/<pid/ns/
 153  *
 154  * If the two processes are not in the same namespace returns an fd to the
 155  * namespace of the second process identified by @pid2. If the two processes are
 156  * in the same namespace returns -EINVAL, -1 if an error occurred.
 157  */
 158 static int in_same_namespace(pid_t pid1, pid_t pid2, const char *ns)
 159 {
 160         __do_close_prot_errno int ns_fd1 = -1, ns_fd2 = -1;
 161         int ret = -1;
 162         struct stat ns_st1, ns_st2;
 163
 164         ns_fd1 = lxc_preserve_ns(pid1, ns);
 165         if (ns_fd1 < 0) {
 166                 /* The kernel does not support this namespace. This is not an
 167                  * error.
 168                  */
 169                 if (errno == ENOENT)
 170                         return -EINVAL;
 171
 172                 return -1;
 173         }
 174
 175         ns_fd2 = lxc_preserve_ns(pid2, ns);
 176         if (ns_fd2 < 0)
 177                 return -1;
 178
 179         ret = fstat(ns_fd1, &ns_st1);
 180         if (ret < 0)
 181                 return -1;
 182
 183         ret = fstat(ns_fd2, &ns_st2);
 184         if (ret < 0)
 185                 return -1;
 186
 187         /* processes are in the same namespace */
 188         if ((ns_st1.st_dev == ns_st2.st_dev) && (ns_st1.st_ino == ns_st2.st_ino))
 189                 return -EINVAL;
 190
 191         /* processes are in different namespaces */
 192         return move_fd(ns_fd2);
 193 }
 194
 195 static int lxc_attach_to_ns(pid_t pid, struct lxc_proc_context_info *ctx)
 196 {
 197         int i, ret;
 198
 199         for (i = 0; i < LXC_NS_MAX; i++) {
 200                 if (ctx->ns_fd[i] < 0)
 201                         continue;
 202
 203                 ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag);
 204                 if (ret < 0) {
 205                         SYSERROR("Failed to attach to %s namespace of %d",
 206                                  ns_info[i].proc_name, pid);
 207                         return -1;
 208                 }
 209
 210                 DEBUG("Attached to %s namespace of %d", ns_info[i].proc_name, pid);
 211         }
 212
 213         return 0;
 214 }
 215
 216 static int lxc_attach_remount_sys_proc(void)
 217 {
 218         int ret;
 219
 220         ret = unshare(CLONE_NEWNS);
 221         if (ret < 0) {
 222                 SYSERROR("Failed to unshare mount namespace");
 223                 return -1;
 224         }
 225
 226         if (detect_shared_rootfs()) {
 227                 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) {
 228                         SYSERROR("Failed to make / rslave");
 229                         ERROR("Continuing...");
 230                 }
 231         }
 232
 233         /* Assume /proc is always mounted, so remount it. */
 234         ret = umount2("/proc", MNT_DETACH);
 235         if (ret < 0) {
 236                 SYSERROR("Failed to unmount /proc");
 237                 return -1;
 238         }
 239
 240         ret = mount("none", "/proc", "proc", 0, NULL);
 241         if (ret < 0) {
 242                 SYSERROR("Failed to remount /proc");
 243                 return -1;
 244         }
 245
 246         /* Try to umount /sys. If it's not a mount point, we'll get EINVAL, then
 247          * we ignore it because it may not have been mounted in the first place.
 248          */
 249         ret = umount2("/sys", MNT_DETACH);
 250         if (ret < 0 && errno != EINVAL) {
 251                 SYSERROR("Failed to unmount /sys");
 252                 return -1;
 253         } else if (ret == 0) {
 254                 /* Remount it. */
 255                 ret = mount("none", "/sys", "sysfs", 0, NULL);
 256                 if (ret < 0) {
 257                         SYSERROR("Failed to remount /sys");
 258                         return -1;
 259                 }
 260         }
 261
 262         return 0;
 263 }
 264
 265 static int lxc_attach_drop_privs(struct lxc_proc_context_info *ctx)
 266 {
 267         int cap, last_cap;
 268
 269         last_cap = lxc_caps_last_cap();
 270         for (cap = 0; cap <= last_cap; cap++) {
 271                 if (ctx->capability_mask & (1LL << cap))
 272                         continue;
 273
 274                 if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0),
 275                           prctl_arg(0), prctl_arg(0))) {
 276                         SYSERROR("Failed to drop capability %d", cap);
 277                         return -1;
 278                 }
 279
 280                 TRACE("Dropped capability %d", cap);
 281         }
 282
 283         return 0;
 284 }
 285
 286 static int lxc_attach_set_environment(struct lxc_proc_context_info *init_ctx,
 287                                       enum lxc_attach_env_policy_t policy,
 288                                       char **extra_env, char **extra_keep)
 289 {
 290         int ret;
 291         struct lxc_list *iterator;
 292
 293         if (policy == LXC_ATTACH_CLEAR_ENV) {
 294                 int path_kept = 0;
 295                 char **extra_keep_store = NULL;
 296
 297                 if (extra_keep) {
 298                         size_t count, i;
 299
 300                         for (count = 0; extra_keep[count]; count++)
 301                                 ;
 302
 303                         extra_keep_store = calloc(count, sizeof(char *));
 304                         if (!extra_keep_store)
 305                                 return -1;
 306
 307                         for (i = 0; i < count; i++) {
 308                                 char *v = getenv(extra_keep[i]);
 309                                 if (v) {
 310                                         extra_keep_store[i] = strdup(v);
 311                                         if (!extra_keep_store[i]) {
 312                                                 while (i > 0)
 313                                                         free(extra_keep_store[--i]);
 314
 315                                                 free(extra_keep_store);
 316                                                 return -1;
 317                                         }
 318
 319                                         if (strcmp(extra_keep[i], "PATH") == 0)
 320                                                 path_kept = 1;
 321                                 }
 322                         }
 323                 }
 324
 325                 if (clearenv()) {
 326                         if (extra_keep_store) {
 327                                 char **p;
 328
 329                                 for (p = extra_keep_store; *p; p++)
 330                                         free(*p);
 331
 332                                 free(extra_keep_store);
 333                         }
 334
 335                         ERROR("Failed to clear environment");
 336                         return -1;
 337                 }
 338
 339                 if (extra_keep_store) {
 340                         size_t i;
 341
 342                         for (i = 0; extra_keep[i]; i++) {
 343                                 if (extra_keep_store[i]) {
 344                                         ret = setenv(extra_keep[i], extra_keep_store[i], 1);
 345                                         if (ret < 0)
 346                                                 SYSWARN("Failed to set environment variable");
 347                                 }
 348
 349                                 free(extra_keep_store[i]);
 350                         }
 351
 352                         free(extra_keep_store);
 353                 }
 354
 355                 /* Always set a default path; shells and execlp tend to be fine
 356                  * without it, but there is a disturbing number of C programs
 357                  * out there that just assume that getenv("PATH") is never NULL
 358                  * and then die a painful segfault death.
 359                  */
 360                 if (!path_kept) {
 361                         ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1);
 362                         if (ret < 0)
 363                                 SYSWARN("Failed to set environment variable");
 364                 }
 365         }
 366
 367         ret = putenv("container=lxc");
 368         if (ret < 0) {
 369                 SYSWARN("Failed to set environment variable");
 370                 return -1;
 371         }
 372
 373         /* Set container environment variables.*/
 374         if (init_ctx && init_ctx->container && init_ctx->container->lxc_conf) {
 375                 lxc_list_for_each(iterator, &init_ctx->container->lxc_conf->environment) {
 376                         char *env_tmp;
 377
 378                         env_tmp = strdup((char *)iterator->elem);
 379                         if (!env_tmp)
 380                                 return -1;
 381
 382                         ret = putenv(env_tmp);
 383                         if (ret < 0) {
 384                                 SYSERROR("Failed to set environment variable: %s", (char *)iterator->elem);
 385                                 return -1;
 386                         }
 387                 }
 388         }
 389
 390         /* Set extra environment variables. */
 391         if (extra_env) {
 392                 for (; *extra_env; extra_env++) {
 393                         char *p;
 394
 395                         /* We just assume the user knows what they are doing, so
 396                          * we don't do any checks.
 397                          */
 398                         p = strdup(*extra_env);
 399                         if (!p)
 400                                 return -1;
 401
 402                         ret = putenv(p);
 403                         if (ret < 0)
 404                                 SYSWARN("Failed to set environment variable");
 405                 }
 406         }
 407
 408         return 0;
 409 }
 410
 411 static char *lxc_attach_getpwshell(uid_t uid)
 412 {
 413         __do_free char *line = NULL;
 414         __do_fclose FILE *pipe_f = NULL;
 415         int fd, ret;
 416         pid_t pid;
 417         int pipes[2];
 418         bool found = false;
 419         size_t line_bufsz = 0;
 420         char *result = NULL;
 421
 422         /* We need to fork off a process that runs the getent program, and we
 423          * need to capture its output, so we use a pipe for that purpose.
 424          */
 425         ret = pipe2(pipes, O_CLOEXEC);
 426         if (ret < 0)
 427                 return NULL;
 428
 429         pid = fork();
 430         if (pid < 0) {
 431                 close(pipes[0]);
 432                 close(pipes[1]);
 433                 return NULL;
 434         }
 435
 436         if (!pid) {
 437                 char uid_buf[32];
 438                 char *arguments[] = {
 439                         "getent",
 440                         "passwd",
 441                         uid_buf,
 442                         NULL
 443                 };
 444
 445                 close(pipes[0]);
 446
 447                 /* We want to capture stdout. */
 448                 ret = dup2(pipes[1], STDOUT_FILENO);
 449                 close(pipes[1]);
 450                 if (ret < 0)
 451                         _exit(EXIT_FAILURE);
 452
 453                 /* Get rid of stdin/stderr, so we try to associate it with
 454                  * /dev/null.
 455                  */
 456                 fd = open_devnull();
 457                 if (fd < 0) {
 458                         close(STDIN_FILENO);
 459                         close(STDERR_FILENO);
 460                 } else {
 461                         (void)dup3(fd, STDIN_FILENO, O_CLOEXEC);
 462                         (void)dup3(fd, STDOUT_FILENO, O_CLOEXEC);
 463                         close(fd);
 464                 }
 465
 466                 /* Finish argument list. */
 467                 ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid);
 468                 if (ret <= 0 || ret >= sizeof(uid_buf))
 469                         _exit(EXIT_FAILURE);
 470
 471                 /* Try to run getent program. */
 472                 (void)execvp("getent", arguments);
 473                 _exit(EXIT_FAILURE);
 474         }
 475
 476         close(pipes[1]);
 477
 478         pipe_f = fdopen(pipes[0], "r");
 479         while (getline(&line, &line_bufsz, pipe_f) != -1) {
 480                 int i;
 481                 long value;
 482                 char *token;
 483                 char *endptr = NULL, *saveptr = NULL;
 484
 485                 /* If we already found something, just continue to read
 486                 * until the pipe doesn't deliver any more data, but
 487                 * don't modify the existing data structure.
 488                  */
 489                 if (found)
 490                         continue;
 491
 492                 if (!line)
 493                         continue;
 494
 495                 /* Trim line on the right hand side. */
 496                 for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i)
 497                         line[i - 1] = '\0';
 498
 499                 /* Split into tokens: first: user name. */
 500                 token = strtok_r(line, ":", &saveptr);
 501                 if (!token)
 502                         continue;
 503
 504                 /* next: dummy password field */
 505                 token = strtok_r(NULL, ":", &saveptr);
 506                 if (!token)
 507                         continue;
 508
 509                 /* next: user id */
 510                 token = strtok_r(NULL, ":", &saveptr);
 511                 value = token ? strtol(token, &endptr, 10) : 0;
 512                 if (!token || !endptr || *endptr || value == LONG_MIN ||
 513                     value == LONG_MAX)
 514                         continue;
 515
 516                 /* dummy sanity check: user id matches */
 517                 if ((uid_t)value != uid)
 518                         continue;
 519
 520                 /* skip fields: gid, gecos, dir, go to next field 'shell' */
 521                 for (i = 0; i < 4; i++) {
 522                         token = strtok_r(NULL, ":", &saveptr);
 523                         if (!token)
 524                                 continue;
 525                 }
 526
 527                 if (!token)
 528                         continue;
 529
 530                 free(result);
 531                 result = strdup(token);
 532
 533                 /* Sanity check that there are no fields after that. */
 534                 token = strtok_r(NULL, ":", &saveptr);
 535                 if (token)
 536                         continue;
 537
 538                 found = true;
 539         }
 540
 541         ret = wait_for_pid(pid);
 542         if (ret < 0) {
 543                 free(result);
 544                 return NULL;
 545         }
 546
 547         if (!found) {
 548                 free(result);
 549                 return NULL;
 550         }
 551
 552         return result;
 553 }
 554
 555 static void lxc_attach_get_init_uidgid(uid_t *init_uid, gid_t *init_gid)
 556 {
 557         __do_free char *line = NULL;
 558         __do_fclose FILE *proc_file = NULL;
 559         char proc_fn[LXC_PROC_STATUS_LEN];
 560         int ret;
 561         size_t line_bufsz = 0;
 562         long value = -1;
 563         uid_t uid = (uid_t)-1;
 564         gid_t gid = (gid_t)-1;
 565
 566         ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", 1);
 567         if (ret < 0 || ret >= LXC_PROC_STATUS_LEN)
 568                 return;
 569
 570         proc_file = fopen(proc_fn, "r");
 571         if (!proc_file)
 572                 return;
 573
 574         while (getline(&line, &line_bufsz, proc_file) != -1) {
 575                 /* Format is: real, effective, saved set user, fs we only care
 576                  * about real uid.
 577                  */
 578                 ret = sscanf(line, "Uid: %ld", &value);
 579                 if (ret != EOF && ret == 1) {
 580                         uid = (uid_t)value;
 581                 } else {
 582                         ret = sscanf(line, "Gid: %ld", &value);
 583                         if (ret != EOF && ret == 1)
 584                                 gid = (gid_t)value;
 585                 }
 586
 587                 if (uid != (uid_t)-1 && gid != (gid_t)-1)
 588                         break;
 589         }
 590
 591         /* Only override arguments if we found something. */
 592         if (uid != (uid_t)-1)
 593                 *init_uid = uid;
 594
 595         if (gid != (gid_t)-1)
 596                 *init_gid = gid;
 597
 598         /* TODO: we should also parse supplementary groups and use
 599          * setgroups() to set them.
 600          */
 601 }
 602
 603 static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options)
 604 {
 605         __do_free char *path = NULL;
 606         int ret;
 607         bool bret;
 608
 609         if (!(options->namespaces & CLONE_NEWNS) ||
 610             !(options->attach_flags & LXC_ATTACH_LSM)) {
 611                 free(c->lxc_conf->seccomp);
 612                 c->lxc_conf->seccomp = NULL;
 613                 return true;
 614         }
 615
 616         /* Remove current setting. */
 617         if (!c->set_config_item(c, "lxc.seccomp.profile", "") &&
 618             !c->set_config_item(c, "lxc.seccomp", ""))
 619                 return false;
 620
 621         /* Fetch the current profile path over the cmd interface. */
 622         path = c->get_running_config_item(c, "lxc.seccomp.profile");
 623         if (!path) {
 624                 INFO("Failed to retrieve lxc.seccomp.profile");
 625
 626                 path = c->get_running_config_item(c, "lxc.seccomp");
 627                 if (!path) {
 628                         INFO("Failed to retrieve lxc.seccomp");
 629                         return true;
 630                 }
 631         }
 632
 633         /* Copy the value into the new lxc_conf. */
 634         bret = c->set_config_item(c, "lxc.seccomp.profile", path);
 635         if (!bret)
 636                 return false;
 637
 638         /* Attempt to parse the resulting config. */
 639         ret = lxc_read_seccomp_config(c->lxc_conf);
 640         if (ret < 0) {
 641                 ERROR("Failed to retrieve seccomp policy");
 642                 return false;
 643         }
 644
 645         INFO("Retrieved seccomp policy");
 646         return true;
 647 }
 648
 649 static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options)
 650 {
 651         __do_free char *val = NULL;
 652
 653         /* Remove current setting. */
 654         if (!c->set_config_item(c, "lxc.no_new_privs", "")) {
 655                 INFO("Failed to unset lxc.no_new_privs");
 656                 return false;
 657         }
 658
 659         /* Retrieve currently active setting. */
 660         val = c->get_running_config_item(c, "lxc.no_new_privs");
 661         if (!val) {
 662                 INFO("Failed to retrieve lxc.no_new_privs");
 663                 return false;
 664         }
 665
 666         /* Set currently active setting. */
 667         return c->set_config_item(c, "lxc.no_new_privs", val);
 668 }
 669
 670 static signed long get_personality(const char *name, const char *lxcpath)
 671 {
 672         __do_free char *p = NULL;
 673
 674         p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath);
 675         if (!p)
 676                 return -1;
 677
 678         return lxc_config_parse_arch(p);
 679 }
 680
 681 struct attach_clone_payload {
 682         int ipc_socket;
 683         int terminal_slave_fd;
 684         lxc_attach_options_t *options;
 685         struct lxc_proc_context_info *init_ctx;
 686         lxc_attach_exec_t exec_function;
 687         void *exec_payload;
 688 };
 689
 690 static void lxc_put_attach_clone_payload(struct attach_clone_payload *p)
 691 {
 692         __do_close_prot_errno int ipc_socket = p->ipc_socket;
 693         __do_close_prot_errno int terminal_slave_fd = p->terminal_slave_fd;
 694
 695         if (p->init_ctx) {
 696                 lxc_proc_put_context_info(p->init_ctx);
 697                 p->init_ctx = NULL;
 698         }
 699 }
 700
 701 static int attach_child_main(struct attach_clone_payload *payload)
 702 {
 703         int fd, lsm_fd, ret;
 704         uid_t new_uid;
 705         gid_t new_gid;
 706         uid_t ns_root_uid = 0;
 707         gid_t ns_root_gid = 0;
 708         lxc_attach_options_t* options = payload->options;
 709         struct lxc_proc_context_info* init_ctx = payload->init_ctx;
 710         bool needs_lsm = (options->namespaces & CLONE_NEWNS) &&
 711                          (options->attach_flags & LXC_ATTACH_LSM) &&
 712                          init_ctx->lsm_label;
 713
 714         /* A description of the purpose of this functionality is provided in the
 715          * lxc-attach(1) manual page. We have to remount here and not in the
 716          * parent process, otherwise /proc may not properly reflect the new pid
 717          * namespace.
 718          */
 719         if (!(options->namespaces & CLONE_NEWNS) &&
 720             (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) {
 721                 ret = lxc_attach_remount_sys_proc();
 722                 if (ret < 0)
 723                         goto on_error;
 724
 725                 TRACE("Remounted \"/proc\" and \"/sys\"");
 726         }
 727
 728 /* Now perform additional attachments. */
 729 #if HAVE_SYS_PERSONALITY_H
 730         if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) {
 731                 long new_personality;
 732
 733                 if (options->personality < 0)
 734                         new_personality = init_ctx->personality;
 735                 else
 736                         new_personality = options->personality;
 737
 738                 ret = personality(new_personality);
 739                 if (ret < 0)
 740                         goto on_error;
 741
 742                 TRACE("Set new personality");
 743         }
 744 #endif
 745
 746         if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) {
 747                 ret = lxc_attach_drop_privs(init_ctx);
 748                 if (ret < 0)
 749                         goto on_error;
 750
 751                 TRACE("Dropped capabilities");
 752         }
 753
 754         /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL)
 755          * if you want this to be a no-op).
 756          */
 757         ret = lxc_attach_set_environment(init_ctx,
 758                                          options->env_policy,
 759                                          options->extra_env_vars,
 760                                          options->extra_keep_env);
 761         if (ret < 0)
 762                 goto on_error;
 763
 764         TRACE("Set up environment");
 765
 766         /* This remark only affects fully unprivileged containers:
 767          * Receive fd for LSM security module before we set{g,u}id(). The reason
 768          * is that on set{g,u}id() the kernel will a) make us undumpable and b)
 769          * we will change our effective uid. This means our effective uid will
 770          * be different from the effective uid of the process that created us
 771          * which means that this processs no longer has capabilities in our
 772          * namespace including CAP_SYS_PTRACE. This means we will not be able to
 773          * read and /proc/<pid> files for the process anymore when /proc is
 774          * mounted with hidepid={1,2}. So let's get the lsm label fd before the
 775          * set{g,u}id().
 776          */
 777         if (needs_lsm) {
 778                 ret = lxc_abstract_unix_recv_fds(payload->ipc_socket, &lsm_fd, 1, NULL, 0);
 779                 if (ret <= 0) {
 780                         if (ret < 0)
 781                                 SYSERROR("Failed to receive lsm label fd");
 782
 783                         goto on_error;
 784                 }
 785
 786                 TRACE("Received LSM label file descriptor %d from parent", lsm_fd);
 787         }
 788
 789         if (options->stdin_fd > 0 && isatty(options->stdin_fd)) {
 790                 ret = lxc_make_controlling_terminal(options->stdin_fd);
 791                 if (ret < 0)
 792                         goto on_error;
 793         }
 794
 795         if (options->namespaces & CLONE_NEWUSER) {
 796                 /* Check whether nsuid 0 has a mapping. */
 797                 ns_root_uid = get_ns_uid(0);
 798
 799                 /* Check whether nsgid 0 has a mapping. */
 800                 ns_root_gid = get_ns_gid(0);
 801
 802                 /* If there's no mapping for nsuid 0 try to retrieve the nsuid
 803                  * init was started with.
 804                  */
 805                 if (ns_root_uid == LXC_INVALID_UID)
 806                         lxc_attach_get_init_uidgid(&ns_root_uid, &ns_root_gid);
 807
 808                 if (ns_root_uid == LXC_INVALID_UID)
 809                         goto on_error;
 810
 811                 if (!lxc_switch_uid_gid(ns_root_uid, ns_root_gid))
 812                         goto on_error;
 813         }
 814
 815         if (!lxc_setgroups(0, NULL) && errno != EPERM)
 816                 goto on_error;
 817
 818         /* Set {u,g}id. */
 819         if (options->uid != LXC_INVALID_UID)
 820                 new_uid = options->uid;
 821         else
 822                 new_uid = ns_root_uid;
 823
 824         if (options->gid != LXC_INVALID_GID)
 825                 new_gid = options->gid;
 826         else
 827                 new_gid = ns_root_gid;
 828
 829         if ((init_ctx->container && init_ctx->container->lxc_conf &&
 830              init_ctx->container->lxc_conf->no_new_privs) ||
 831             (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) {
 832                 ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0),
 833                             prctl_arg(0), prctl_arg(0));
 834                 if (ret < 0)
 835                         goto on_error;
 836
 837                 TRACE("Set PR_SET_NO_NEW_PRIVS");
 838         }
 839
 840         if (needs_lsm) {
 841                 bool on_exec;
 842
 843                 /* Change into our new LSM profile. */
 844                 on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false;
 845
 846                 ret = lsm_process_label_set_at(lsm_fd, init_ctx->lsm_label, on_exec);
 847                 close(lsm_fd);
 848                 if (ret < 0)
 849                         goto on_error;
 850
 851                 TRACE("Set %s LSM label to \"%s\"", lsm_name(), init_ctx->lsm_label);
 852         }
 853
 854         if (init_ctx->container && init_ctx->container->lxc_conf &&
 855             init_ctx->container->lxc_conf->seccomp) {
 856                 ret = lxc_seccomp_load(init_ctx->container->lxc_conf);
 857                 if (ret < 0)
 858                         goto on_error;
 859
 860                 TRACE("Loaded seccomp profile");
 861         }
 862
 863         close(payload->ipc_socket);
 864         payload->ipc_socket = -EBADF;
 865         lxc_proc_put_context_info(init_ctx);
 866         payload->init_ctx = NULL;
 867
 868         /* The following is done after the communication socket is shut down.
 869          * That way, all errors that might (though unlikely) occur up until this
 870          * point will have their messages printed to the original stderr (if
 871          * logging is so configured) and not the fd the user supplied, if any.
 872          */
 873
 874         /* Fd handling for stdin, stdout and stderr; ignore errors here, user
 875          * may want to make sure the fds are closed, for example.
 876          */
 877         if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO)
 878                 (void)dup2(options->stdin_fd, STDIN_FILENO);
 879
 880         if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO)
 881                 (void)dup2(options->stdout_fd, STDOUT_FILENO);
 882
 883         if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO)
 884                 (void)dup2(options->stderr_fd, STDERR_FILENO);
 885
 886         /* close the old fds */
 887         if (options->stdin_fd > STDERR_FILENO)
 888                 close(options->stdin_fd);
 889
 890         if (options->stdout_fd > STDERR_FILENO)
 891                 close(options->stdout_fd);
 892
 893         if (options->stderr_fd > STDERR_FILENO)
 894                 close(options->stderr_fd);
 895
 896         /* Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also
 897          * here, ignore errors.
 898          */
 899         for (fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) {
 900                 ret = fd_cloexec(fd, false);
 901                 if (ret < 0) {
 902                         SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd);
 903                         goto on_error;
 904                 }
 905         }
 906
 907         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
 908                 ret = lxc_terminal_prepare_login(payload->terminal_slave_fd);
 909                 if (ret < 0) {
 910                         SYSERROR("Failed to prepare terminal file descriptor %d", payload->terminal_slave_fd);
 911                         goto on_error;
 912                 }
 913
 914                 TRACE("Prepared terminal file descriptor %d", payload->terminal_slave_fd);
 915         }
 916
 917         /* Avoid unnecessary syscalls. */
 918         if (new_uid == ns_root_uid)
 919                 new_uid = LXC_INVALID_UID;
 920
 921         if (new_gid == ns_root_gid)
 922                 new_gid = LXC_INVALID_GID;
 923
 924         if (!lxc_switch_uid_gid(new_uid, new_gid))
 925                 goto on_error;
 926
 927         /* We're done, so we can now do whatever the user intended us to do. */
 928         _exit(payload->exec_function(payload->exec_payload));
 929
 930 on_error:
 931         lxc_put_attach_clone_payload(payload);
 932         _exit(EXIT_FAILURE);
 933 }
 934
 935 static int lxc_attach_terminal(struct lxc_conf *conf,
 936                                struct lxc_terminal *terminal)
 937 {
 938         int ret;
 939
 940         lxc_terminal_init(terminal);
 941
 942         ret = lxc_terminal_create(terminal);
 943         if (ret < 0) {
 944                 ERROR("Failed to create terminal");
 945                 return -1;
 946         }
 947
 948         /* Shift ttys to container. */
 949         ret = lxc_terminal_map_ids(conf, terminal);
 950         if (ret < 0) {
 951                 ERROR("Failed to chown terminal");
 952                 goto on_error;
 953         }
 954
 955         return 0;
 956
 957 on_error:
 958         lxc_terminal_delete(terminal);
 959         lxc_terminal_conf_free(terminal);
 960         return -1;
 961 }
 962
 963 static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal,
 964                                              struct lxc_epoll_descr *descr)
 965 {
 966         int ret;
 967
 968         ret = lxc_mainloop_open(descr);
 969         if (ret < 0) {
 970                 ERROR("Failed to create mainloop");
 971                 return -1;
 972         }
 973
 974         ret = lxc_terminal_mainloop_add(descr, terminal);
 975         if (ret < 0) {
 976                 ERROR("Failed to add handlers to mainloop");
 977                 lxc_mainloop_close(descr);
 978                 return -1;
 979         }
 980
 981         return 0;
 982 }
 983
 984 static inline void lxc_attach_terminal_close_master(struct lxc_terminal *terminal)
 985 {
 986         close_prot_errno_disarm(terminal->master);
 987 }
 988
 989 static inline void lxc_attach_terminal_close_slave(struct lxc_terminal *terminal)
 990 {
 991         close_prot_errno_disarm(terminal->slave);
 992 }
 993
 994 static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal)
 995 {
 996         close_prot_errno_disarm(terminal->peer);
 997 }
 998
 999 static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal)
1000 {
1001         close_prot_errno_disarm(terminal->log_fd);
1002 }
1003
1004 int lxc_attach(const char *name, const char *lxcpath,
1005                lxc_attach_exec_t exec_function, void *exec_payload,
1006                lxc_attach_options_t *options, pid_t *attached_process)
1007 {
1008         int i, ret, status;
1009         int ipc_sockets[2];
1010         char *cwd, *new_cwd;
1011         signed long personality;
1012         pid_t attached_pid, init_pid, pid;
1013         struct lxc_proc_context_info *init_ctx;
1014         struct lxc_terminal terminal;
1015         struct lxc_conf *conf;
1016         struct attach_clone_payload payload = {0};
1017
1018         ret = access("/proc/self/ns", X_OK);
1019         if (ret) {
1020                 SYSERROR("Does this kernel version support namespaces?");
1021                 return -1;
1022         }
1023
1024         if (!options)
1025                 options = &attach_static_default_options;
1026
1027         init_pid = lxc_cmd_get_init_pid(name, lxcpath);
1028         if (init_pid < 0) {
1029                 ERROR("Failed to get init pid");
1030                 return -1;
1031         }
1032
1033         init_ctx = lxc_proc_get_context_info(init_pid);
1034         if (!init_ctx) {
1035                 ERROR("Failed to get context of init process: %ld", (long)init_pid);
1036                 return -1;
1037         }
1038
1039         personality = get_personality(name, lxcpath);
1040         if (init_ctx->personality < 0) {
1041                 ERROR("Failed to get personality of the container");
1042                 lxc_proc_put_context_info(init_ctx);
1043                 return -1;
1044         }
1045         init_ctx->personality = personality;
1046
1047         init_ctx->container = lxc_container_new(name, lxcpath);
1048         if (!init_ctx->container) {
1049                 lxc_proc_put_context_info(init_ctx);
1050                 return -1;
1051         }
1052
1053         if (!init_ctx->container->lxc_conf) {
1054                 init_ctx->container->lxc_conf = lxc_conf_init();
1055                 if (!init_ctx->container->lxc_conf) {
1056                         lxc_proc_put_context_info(init_ctx);
1057                         return -1;
1058                 }
1059         }
1060         conf = init_ctx->container->lxc_conf;
1061
1062         if (!fetch_seccomp(init_ctx->container, options))
1063                 WARN("Failed to get seccomp policy");
1064
1065         if (!no_new_privs(init_ctx->container, options))
1066                 WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set");
1067
1068         cwd = getcwd(NULL, 0);
1069
1070         /* Determine which namespaces the container was created with
1071          * by asking lxc-start, if necessary.
1072          */
1073         if (options->namespaces == -1) {
1074                 options->namespaces = lxc_cmd_get_clone_flags(name, lxcpath);
1075                 /* call failed */
1076                 if (options->namespaces == -1) {
1077                         ERROR("Failed to automatically determine the "
1078                               "namespaces which the container uses");
1079                         free(cwd);
1080                         lxc_proc_put_context_info(init_ctx);
1081                         return -1;
1082                 }
1083
1084                 for (i = 0; i < LXC_NS_MAX; i++) {
1085                         if (ns_info[i].clone_flag & CLONE_NEWCGROUP)
1086                                 if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) ||
1087                                     !cgns_supported())
1088                                         continue;
1089
1090                         if (ns_info[i].clone_flag & options->namespaces)
1091                                 continue;
1092
1093                         init_ctx->ns_inherited |= ns_info[i].clone_flag;
1094                 }
1095         }
1096
1097         pid = lxc_raw_getpid();
1098
1099         for (i = 0; i < LXC_NS_MAX; i++) {
1100                 int j;
1101
1102                 if (options->namespaces & ns_info[i].clone_flag)
1103                         init_ctx->ns_fd[i] = lxc_preserve_ns(init_pid, ns_info[i].proc_name);
1104                 else if (init_ctx->ns_inherited & ns_info[i].clone_flag)
1105                         init_ctx->ns_fd[i] = in_same_namespace(pid, init_pid, ns_info[i].proc_name);
1106                 else
1107                         continue;
1108
1109                 if (init_ctx->ns_fd[i] >= 0)
1110                         continue;
1111
1112                 if (init_ctx->ns_fd[i] == -EINVAL) {
1113                         DEBUG("Inheriting %s namespace from %d",
1114                               ns_info[i].proc_name, pid);
1115                         init_ctx->ns_inherited &= ~ns_info[i].clone_flag;
1116                         continue;
1117                 }
1118
1119                 /* We failed to preserve the namespace. */
1120                 SYSERROR("Failed to attach to %s namespace of %d",
1121                          ns_info[i].proc_name, pid);
1122
1123                 /* Close all already opened file descriptors before we return an
1124                  * error, so we don't leak them.
1125                  */
1126                 for (j = 0; j < i; j++)
1127                         close(init_ctx->ns_fd[j]);
1128
1129                 free(cwd);
1130                 lxc_proc_put_context_info(init_ctx);
1131                 return -1;
1132         }
1133
1134         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1135                 ret = lxc_attach_terminal(conf, &terminal);
1136                 if (ret < 0) {
1137                         ERROR("Failed to setup new terminal");
1138                         free(cwd);
1139                         lxc_proc_put_context_info(init_ctx);
1140                         return -1;
1141                 }
1142
1143                 terminal.log_fd = options->log_fd;
1144         } else {
1145                 lxc_terminal_init(&terminal);
1146         }
1147
1148         /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order
1149          * to make sure we don't irritate other threads that want to fork+exec
1150          * away
1151          *
1152          * IMPORTANT: if the initial process is multithreaded and another call
1153          * just fork()s away without exec'ing directly after, the socket fd will
1154          * exist in the forked process from the other thread and any close() in
1155          * our own child process will not really cause the socket to close
1156          * properly, potentially causing the parent to hang.
1157          *
1158          * For this reason, while IPC is still active, we have to use shutdown()
1159          * if the child exits prematurely in order to signal that the socket is
1160          * closed and cannot assume that the child exiting will automatically do
1161          * that.
1162          *
1163          * IPC mechanism: (X is receiver)
1164          *   initial process        intermediate          attached
1165          *        X           <---  send pid of
1166          *                          attached proc,
1167          *                          then exit
1168          *    send 0 ------------------------------------>    X
1169          *                                              [do initialization]
1170          *        X  <------------------------------------  send 1
1171          *   [add to cgroup, ...]
1172          *    send 2 ------------------------------------>    X
1173          *                                              [set LXC_ATTACH_NO_NEW_PRIVS]
1174          *        X  <------------------------------------  send 3
1175          *   [open LSM label fd]
1176          *    send 4 ------------------------------------>    X
1177          *                                              [set LSM label]
1178          *   close socket                                 close socket
1179          *                                                run program
1180          */
1181         ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets);
1182         if (ret < 0) {
1183                 SYSERROR("Could not set up required IPC mechanism for attaching");
1184                 free(cwd);
1185                 lxc_proc_put_context_info(init_ctx);
1186                 return -1;
1187         }
1188
1189         /* Create intermediate subprocess, two reasons:
1190          *       1. We can't setns() in the child itself, since we want to make
1191          *          sure we are properly attached to the pidns.
1192          *       2. Also, the initial thread has to put the attached process
1193          *          into the cgroup, which we can only do if we didn't already
1194          *          setns() (otherwise, user namespaces will hate us).
1195          */
1196         pid = fork();
1197         if (pid < 0) {
1198                 SYSERROR("Failed to create first subprocess");
1199                 free(cwd);
1200                 lxc_proc_put_context_info(init_ctx);
1201                 return -1;
1202         }
1203
1204         if (pid) {
1205                 int ret_parent = -1;
1206                 pid_t to_cleanup_pid = pid;
1207                 struct lxc_epoll_descr descr = {0};
1208
1209                 /* close unneeded file descriptors */
1210                 close(ipc_sockets[1]);
1211                 free(cwd);
1212                 lxc_proc_close_ns_fd(init_ctx);
1213                 if (options->attach_flags & LXC_ATTACH_TERMINAL)
1214                         lxc_attach_terminal_close_slave(&terminal);
1215
1216                 /* Attach to cgroup, if requested. */
1217                 if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) {
1218                         struct cgroup_ops *cgroup_ops;
1219
1220                         cgroup_ops = cgroup_init(conf);
1221                         if (!cgroup_ops)
1222                                 goto on_error;
1223
1224                         if (!cgroup_ops->attach(cgroup_ops, name, lxcpath, pid))
1225                                 goto on_error;
1226
1227                         cgroup_exit(cgroup_ops);
1228                         TRACE("Moved intermediate process %d into container's cgroups", pid);
1229                 }
1230
1231                 /* Setup /proc limits */
1232                 if (!lxc_list_empty(&conf->procs)) {
1233                         ret = setup_proc_filesystem(&conf->procs, pid);
1234                         if (ret < 0)
1235                                 goto on_error;
1236                 }
1237
1238                 /* Setup resource limits */
1239                 if (!lxc_list_empty(&conf->limits)) {
1240                         ret = setup_resource_limits(&conf->limits, pid);
1241                         if (ret < 0)
1242                                 goto on_error;
1243                 }
1244
1245                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1246                         ret = lxc_attach_terminal_mainloop_init(&terminal, &descr);
1247                         if (ret < 0)
1248                                 goto on_error;
1249
1250                         TRACE("Initialized terminal mainloop");
1251                 }
1252
1253                 /* Let the child process know to go ahead. */
1254                 status = 0;
1255                 ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status));
1256                 if (ret != sizeof(status))
1257                         goto close_mainloop;
1258
1259                 TRACE("Told intermediate process to start initializing");
1260
1261                 /* Get pid of attached process from intermediate process. */
1262                 ret = lxc_read_nointr(ipc_sockets[0], &attached_pid, sizeof(attached_pid));
1263                 if (ret != sizeof(attached_pid))
1264                         goto close_mainloop;
1265
1266                 TRACE("Received pid %d of attached process in parent pid namespace", attached_pid);
1267
1268                 /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */
1269                 if (options->stdin_fd == 0) {
1270                         signal(SIGINT, SIG_IGN);
1271                         signal(SIGQUIT, SIG_IGN);
1272                 }
1273
1274                 /* Reap intermediate process. */
1275                 ret = wait_for_pid(pid);
1276                 if (ret < 0)
1277                         goto close_mainloop;
1278
1279                 TRACE("Intermediate process %d exited", pid);
1280
1281                 /* We will always have to reap the attached process now. */
1282                 to_cleanup_pid = attached_pid;
1283
1284                 /* Open LSM fd and send it to child. */
1285                 if ((options->namespaces & CLONE_NEWNS) &&
1286                     (options->attach_flags & LXC_ATTACH_LSM) &&
1287                     init_ctx->lsm_label) {
1288                         int labelfd;
1289                         bool on_exec;
1290
1291                         ret = -1;
1292                         on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false;
1293                         labelfd = lsm_process_label_fd_get(attached_pid, on_exec);
1294                         if (labelfd < 0)
1295                                 goto close_mainloop;
1296
1297                         TRACE("Opened LSM label file descriptor %d", labelfd);
1298
1299                         /* Send child fd of the LSM security module to write to. */
1300                         ret = lxc_abstract_unix_send_fds(ipc_sockets[0], &labelfd, 1, NULL, 0);
1301                         if (ret <= 0) {
1302                                 if (ret < 0)
1303                                         SYSERROR("Failed to send lsm label fd");
1304
1305                                 close(labelfd);
1306                                 goto close_mainloop;
1307                         }
1308
1309                         close(labelfd);
1310                         TRACE("Sent LSM label file descriptor %d to child", labelfd);
1311                 }
1312
1313                 /* We're done, the child process should now execute whatever it
1314                  * is that the user requested. The parent can now track it with
1315                  * waitpid() or similar.
1316                  */
1317
1318                 *attached_process = attached_pid;
1319
1320                 /* Now shut down communication with child, we're done. */
1321                 shutdown(ipc_sockets[0], SHUT_RDWR);
1322                 close(ipc_sockets[0]);
1323                 ipc_sockets[0] = -1;
1324
1325                 ret_parent = 0;
1326                 to_cleanup_pid = -1;
1327
1328                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1329                         ret = lxc_mainloop(&descr, -1);
1330                         if (ret < 0) {
1331                                 ret_parent = -1;
1332                                 to_cleanup_pid = attached_pid;
1333                         }
1334                 }
1335
1336         close_mainloop:
1337                 if (options->attach_flags & LXC_ATTACH_TERMINAL)
1338                         lxc_mainloop_close(&descr);
1339
1340         on_error:
1341                 if (ipc_sockets[0] >= 0) {
1342                         shutdown(ipc_sockets[0], SHUT_RDWR);
1343                         close(ipc_sockets[0]);
1344                 }
1345
1346                 if (to_cleanup_pid > 0)
1347                         (void)wait_for_pid(to_cleanup_pid);
1348
1349                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1350                         lxc_terminal_delete(&terminal);
1351                         lxc_terminal_conf_free(&terminal);
1352                 }
1353
1354                 lxc_proc_put_context_info(init_ctx);
1355                 return ret_parent;
1356         }
1357
1358         /* close unneeded file descriptors */
1359         close(ipc_sockets[0]);
1360         ipc_sockets[0] = -EBADF;
1361
1362         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1363                 lxc_attach_terminal_close_master(&terminal);
1364                 lxc_attach_terminal_close_peer(&terminal);
1365                 lxc_attach_terminal_close_log(&terminal);
1366         }
1367
1368         /* Wait for the parent to have setup cgroups. */
1369         ret = lxc_read_nointr(ipc_sockets[1], &status, sizeof(status));
1370         if (ret != sizeof(status)) {
1371                 shutdown(ipc_sockets[1], SHUT_RDWR);
1372                 lxc_proc_put_context_info(init_ctx);
1373                 _exit(EXIT_FAILURE);
1374         }
1375
1376         TRACE("Intermediate process starting to initialize");
1377
1378         /* Attach now, create another subprocess later, since pid namespaces
1379          * only really affect the children of the current process.
1380          */
1381         ret = lxc_attach_to_ns(init_pid, init_ctx);
1382         if (ret < 0) {
1383                 ERROR("Failed to enter namespaces");
1384                 shutdown(ipc_sockets[1], SHUT_RDWR);
1385                 lxc_proc_put_context_info(init_ctx);
1386                 _exit(EXIT_FAILURE);
1387         }
1388
1389         /* close namespace file descriptors */
1390         lxc_proc_close_ns_fd(init_ctx);
1391
1392         /* Attach succeeded, try to cwd. */
1393         if (options->initial_cwd)
1394                 new_cwd = options->initial_cwd;
1395         else
1396                 new_cwd = cwd;
1397         if (new_cwd) {
1398                 ret = chdir(new_cwd);
1399                 if (ret < 0)
1400                         WARN("Could not change directory to \"%s\"", new_cwd);
1401         }
1402         free(cwd);
1403
1404         /* Create attached process. */
1405         payload.ipc_socket = ipc_sockets[1];
1406         payload.options = options;
1407         payload.init_ctx = init_ctx;
1408         payload.terminal_slave_fd = terminal.slave;
1409         payload.exec_function = exec_function;
1410         payload.exec_payload = exec_payload;
1411
1412         pid = lxc_raw_clone(CLONE_PARENT);
1413         if (pid < 0) {
1414                 SYSERROR("Failed to clone attached process");
1415                 shutdown(ipc_sockets[1], SHUT_RDWR);
1416                 lxc_proc_put_context_info(init_ctx);
1417                 _exit(EXIT_FAILURE);
1418         }
1419
1420         if (pid == 0) {
1421                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1422                         ret = pthread_sigmask(SIG_SETMASK,
1423                                               &terminal.tty_state->oldmask, NULL);
1424                         if (ret < 0) {
1425                                 SYSERROR("Failed to reset signal mask");
1426                                 _exit(EXIT_FAILURE);
1427                         }
1428                 }
1429
1430                 ret = attach_child_main(&payload);
1431                 if (ret < 0)
1432                         ERROR("Failed to exec");
1433
1434                 _exit(EXIT_FAILURE);
1435         }
1436
1437         if (options->attach_flags & LXC_ATTACH_TERMINAL)
1438                 lxc_attach_terminal_close_slave(&terminal);
1439
1440         /* Tell grandparent the pid of the pid of the newly created child. */
1441         ret = lxc_write_nointr(ipc_sockets[1], &pid, sizeof(pid));
1442         if (ret != sizeof(pid)) {
1443                 /* If this really happens here, this is very unfortunate, since
1444                  * the parent will not know the pid of the attached process and
1445                  * will not be able to wait for it (and we won't either due to
1446                  * CLONE_PARENT) so the parent won't be able to reap it and the
1447                  * attached process will remain a zombie.
1448                  */
1449                 shutdown(ipc_sockets[1], SHUT_RDWR);
1450                 lxc_proc_put_context_info(init_ctx);
1451                 _exit(EXIT_FAILURE);
1452         }
1453
1454         TRACE("Sending pid %d of attached process", pid);
1455
1456         /* The rest is in the hands of the initial and the attached process. */
1457         lxc_proc_put_context_info(init_ctx);
1458         _exit(EXIT_SUCCESS);
1459 }
1460
1461 int lxc_attach_run_command(void *payload)
1462 {
1463         int ret = -1;
1464         lxc_attach_command_t *cmd = payload;
1465
1466         ret = execvp(cmd->program, cmd->argv);
1467         if (ret < 0) {
1468                 switch (errno) {
1469                 case ENOEXEC:
1470                         ret = 126;
1471                         break;
1472                 case ENOENT:
1473                         ret = 127;
1474                         break;
1475                 }
1476         }
1477
1478         SYSERROR("Failed to exec \"%s\"", cmd->program);
1479         return ret;
1480 }
1481
1482 int lxc_attach_run_shell(void* payload)
1483 {
1484         __do_free char *buf = NULL;
1485         uid_t uid;
1486         struct passwd pwent;
1487         struct passwd *pwentp = NULL;
1488         char *user_shell;
1489         size_t bufsize;
1490         int ret;
1491
1492         /* Ignore payload parameter. */
1493         (void)payload;
1494
1495         uid = getuid();
1496
1497         bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
1498         if (bufsize == -1)
1499                 bufsize = 1024;
1500
1501         buf = malloc(bufsize);
1502         if (buf) {
1503                 ret = getpwuid_r(uid, &pwent, buf, bufsize, &pwentp);
1504                 if (!pwentp) {
1505                         if (ret == 0)
1506                                 WARN("Could not find matched password record");
1507
1508                         WARN("Failed to get password record - %u", uid);
1509                 }
1510         }
1511
1512         /* This probably happens because of incompatible nss implementations in
1513          * host and container (remember, this code is still using the host's
1514          * glibc but our mount namespace is in the container) we may try to get
1515          * the information by spawning a [getent passwd uid] process and parsing
1516          * the result.
1517          */
1518         if (!pwentp)
1519                 user_shell = lxc_attach_getpwshell(uid);
1520         else
1521                 user_shell = pwent.pw_shell;
1522
1523         if (user_shell)
1524                 execlp(user_shell, user_shell, (char *)NULL);
1525
1526         /* Executed if either no passwd entry or execvp fails, we will fall back
1527          * on /bin/sh as a default shell.
1528          */
1529         execlp("/bin/sh", "/bin/sh", (char *)NULL);
1530
1531         SYSERROR("Failed to execute shell");
1532         if (!pwentp)
1533                 free(user_shell);
1534
1535         return -1;
1536 }