src/lxc/attach.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #ifndef _GNU_SOURCE
   4 #define _GNU_SOURCE 1
   5 #endif
   6 #include <errno.h>
   7 #include <fcntl.h>
   8 #include <grp.h>
   9 #include <linux/unistd.h>
  10 #include <pwd.h>
  11 #include <pthread.h>
  12 #include <signal.h>
  13 #include <stdio.h>
  14 #include <stdlib.h>
  15 #include <string.h>
  16 #include <sys/mount.h>
  17 #include <sys/param.h>
  18 #include <sys/prctl.h>
  19 #include <sys/socket.h>
  20 #include <sys/syscall.h>
  21 #include <sys/wait.h>
  22 #include <termios.h>
  23 #include <unistd.h>
  24
  25 #include <lxc/lxccontainer.h>
  26
  27 #include "af_unix.h"
  28 #include "attach.h"
  29 #include "caps.h"
  30 #include "cgroup.h"
  31 #include "commands.h"
  32 #include "conf.h"
  33 #include "config.h"
  34 #include "confile.h"
  35 #include "log.h"
  36 #include "lsm/lsm.h"
  37 #include "lxclock.h"
  38 #include "lxcseccomp.h"
  39 #include "macro.h"
  40 #include "mainloop.h"
  41 #include "memory_utils.h"
  42 #include "namespace.h"
  43 #include "raw_syscalls.h"
  44 #include "syscall_wrappers.h"
  45 #include "terminal.h"
  46 #include "utils.h"
  47
  48 #if HAVE_SYS_PERSONALITY_H
  49 #include <sys/personality.h>
  50 #endif
  51
  52 lxc_log_define(attach, lxc);
  53
  54 /* Define default options if no options are supplied by the user. */
  55 static lxc_attach_options_t attach_static_default_options = LXC_ATTACH_OPTIONS_DEFAULT;
  56
  57 static struct lxc_proc_context_info *lxc_proc_get_context_info(pid_t pid)
  58 {
  59         __do_free char *line = NULL;
  60         __do_fclose FILE *proc_file = NULL;
  61         __do_free struct lxc_proc_context_info *info = NULL;
  62         int ret;
  63         bool found;
  64         char proc_fn[LXC_PROC_STATUS_LEN];
  65         size_t line_bufsz = 0;
  66
  67         /* Read capabilities. */
  68         ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", pid);
  69         if (ret < 0 || ret >= LXC_PROC_STATUS_LEN)
  70                 return NULL;
  71
  72         proc_file = fopen(proc_fn, "re");
  73         if (!proc_file)
  74                 return log_error_errno(NULL, errno, "Failed to open %s", proc_fn);
  75
  76         info = calloc(1, sizeof(*info));
  77         if (!info)
  78                 return NULL;
  79
  80         found = false;
  81
  82         while (getline(&line, &line_bufsz, proc_file) != -1) {
  83                 ret = sscanf(line, "CapBnd: %llx", &info->capability_mask);
  84                 if (ret != EOF && ret == 1) {
  85                         found = true;
  86                         break;
  87                 }
  88         }
  89
  90         if (!found)
  91                 return log_error_errno(NULL, ENOENT, "Failed to read capability bounding set from %s", proc_fn);
  92
  93         info->lsm_label = lsm_process_label_get(pid);
  94         info->ns_inherited = 0;
  95         memset(info->ns_fd, -1, sizeof(int) * LXC_NS_MAX);
  96
  97         return move_ptr(info);
  98 }
  99
 100 static inline void lxc_proc_close_ns_fd(struct lxc_proc_context_info *ctx)
 101 {
 102         for (int i = 0; i < LXC_NS_MAX; i++)
 103                 close_prot_errno_disarm(ctx->ns_fd[i]);
 104 }
 105
 106 static void lxc_proc_put_context_info(struct lxc_proc_context_info *ctx)
 107 {
 108         free(ctx->lsm_label);
 109         ctx->lsm_label = NULL;
 110
 111         if (ctx->container) {
 112                 lxc_container_put(ctx->container);
 113                 ctx->container = NULL;
 114         }
 115
 116         lxc_proc_close_ns_fd(ctx);
 117         free(ctx);
 118 }
 119
 120 /**
 121  * in_same_namespace - Check whether two processes are in the same namespace.
 122  * @pid1 - PID of the first process.
 123  * @pid2 - PID of the second process.
 124  * @ns   - Name of the namespace to check. Must correspond to one of the names
 125  *         for the namespaces as shown in /proc/<pid/ns/
 126  *
 127  * If the two processes are not in the same namespace returns an fd to the
 128  * namespace of the second process identified by @pid2. If the two processes are
 129  * in the same namespace returns -EINVAL, -1 if an error occurred.
 130  */
 131 static int in_same_namespace(pid_t pid1, pid_t pid2, const char *ns)
 132 {
 133         __do_close_prot_errno int ns_fd1 = -1, ns_fd2 = -1;
 134         int ret = -1;
 135         struct stat ns_st1, ns_st2;
 136
 137         ns_fd1 = lxc_preserve_ns(pid1, ns);
 138         if (ns_fd1 < 0) {
 139                 /* The kernel does not support this namespace. This is not an
 140                  * error.
 141                  */
 142                 if (errno == ENOENT)
 143                         return -EINVAL;
 144
 145                 return -1;
 146         }
 147
 148         ns_fd2 = lxc_preserve_ns(pid2, ns);
 149         if (ns_fd2 < 0)
 150                 return -1;
 151
 152         ret = fstat(ns_fd1, &ns_st1);
 153         if (ret < 0)
 154                 return -1;
 155
 156         ret = fstat(ns_fd2, &ns_st2);
 157         if (ret < 0)
 158                 return -1;
 159
 160         /* processes are in the same namespace */
 161         if ((ns_st1.st_dev == ns_st2.st_dev) && (ns_st1.st_ino == ns_st2.st_ino))
 162                 return -EINVAL;
 163
 164         /* processes are in different namespaces */
 165         return move_fd(ns_fd2);
 166 }
 167
 168 static int lxc_attach_to_ns(pid_t pid, struct lxc_proc_context_info *ctx)
 169 {
 170         for (int i = 0; i < LXC_NS_MAX; i++) {
 171                 int ret;
 172
 173                 if (ctx->ns_fd[i] < 0)
 174                         continue;
 175
 176                 ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag);
 177                 if (ret < 0)
 178                         return log_error_errno(-1,
 179                                                errno, "Failed to attach to %s namespace of %d",
 180                                                ns_info[i].proc_name, pid);
 181
 182                 DEBUG("Attached to %s namespace of %d", ns_info[i].proc_name, pid);
 183         }
 184
 185         return 0;
 186 }
 187
 188 int lxc_attach_remount_sys_proc(void)
 189 {
 190         int ret;
 191
 192         ret = unshare(CLONE_NEWNS);
 193         if (ret < 0)
 194                 return log_error_errno(-1, errno, "Failed to unshare mount namespace");
 195
 196         if (detect_shared_rootfs()) {
 197                 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) {
 198                         SYSERROR("Failed to make / rslave");
 199                         ERROR("Continuing...");
 200                 }
 201         }
 202
 203         /* Assume /proc is always mounted, so remount it. */
 204         ret = umount2("/proc", MNT_DETACH);
 205         if (ret < 0)
 206                 return log_error_errno(-1, errno, "Failed to unmount /proc");
 207
 208         ret = mount("none", "/proc", "proc", 0, NULL);
 209         if (ret < 0)
 210                 return log_error_errno(-1, errno, "Failed to remount /proc");
 211
 212         /*
 213          * Try to umount /sys. If it's not a mount point, we'll get EINVAL, then
 214          * we ignore it because it may not have been mounted in the first place.
 215          */
 216         ret = umount2("/sys", MNT_DETACH);
 217         if (ret < 0 && errno != EINVAL)
 218                 return log_error_errno(-1, errno, "Failed to unmount /sys");
 219
 220         /* Remount it. */
 221         if (ret == 0 && mount("none", "/sys", "sysfs", 0, NULL))
 222                 return log_error_errno(-1, errno, "Failed to remount /sys");
 223
 224         return 0;
 225 }
 226
 227 static int lxc_attach_drop_privs(struct lxc_proc_context_info *ctx)
 228 {
 229         int last_cap;
 230
 231         last_cap = lxc_caps_last_cap();
 232         for (int cap = 0; cap <= last_cap; cap++) {
 233                 if (ctx->capability_mask & (1LL << cap))
 234                         continue;
 235
 236                 if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0),
 237                           prctl_arg(0), prctl_arg(0)))
 238                         return log_error_errno(-1, errno, "Failed to drop capability %d", cap);
 239
 240                 TRACE("Dropped capability %d", cap);
 241         }
 242
 243         return 0;
 244 }
 245
 246 static int lxc_attach_set_environment(struct lxc_proc_context_info *init_ctx,
 247                                       enum lxc_attach_env_policy_t policy,
 248                                       char **extra_env, char **extra_keep)
 249 {
 250         int ret;
 251         struct lxc_list *iterator;
 252
 253         if (policy == LXC_ATTACH_CLEAR_ENV) {
 254                 int path_kept = 0;
 255                 char **extra_keep_store = NULL;
 256
 257                 if (extra_keep) {
 258                         size_t count, i;
 259
 260                         for (count = 0; extra_keep[count]; count++)
 261                                 ;
 262
 263                         extra_keep_store = calloc(count, sizeof(char *));
 264                         if (!extra_keep_store)
 265                                 return -1;
 266
 267                         for (i = 0; i < count; i++) {
 268                                 char *v = getenv(extra_keep[i]);
 269                                 if (v) {
 270                                         extra_keep_store[i] = strdup(v);
 271                                         if (!extra_keep_store[i]) {
 272                                                 while (i > 0)
 273                                                         free(extra_keep_store[--i]);
 274
 275                                                 free(extra_keep_store);
 276                                                 return -1;
 277                                         }
 278
 279                                         if (strcmp(extra_keep[i], "PATH") == 0)
 280                                                 path_kept = 1;
 281                                 }
 282                         }
 283                 }
 284
 285                 if (clearenv()) {
 286                         if (extra_keep_store) {
 287                                 char **p;
 288
 289                                 for (p = extra_keep_store; *p; p++)
 290                                         free(*p);
 291
 292                                 free(extra_keep_store);
 293                         }
 294
 295                         return log_error(-1, "Failed to clear environment");
 296                 }
 297
 298                 if (extra_keep_store) {
 299                         size_t i;
 300
 301                         for (i = 0; extra_keep[i]; i++) {
 302                                 if (extra_keep_store[i]) {
 303                                         ret = setenv(extra_keep[i], extra_keep_store[i], 1);
 304                                         if (ret < 0)
 305                                                 SYSWARN("Failed to set environment variable");
 306                                 }
 307
 308                                 free(extra_keep_store[i]);
 309                         }
 310
 311                         free(extra_keep_store);
 312                 }
 313
 314                 /* Always set a default path; shells and execlp tend to be fine
 315                  * without it, but there is a disturbing number of C programs
 316                  * out there that just assume that getenv("PATH") is never NULL
 317                  * and then die a painful segfault death.
 318                  */
 319                 if (!path_kept) {
 320                         ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1);
 321                         if (ret < 0)
 322                                 SYSWARN("Failed to set environment variable");
 323                 }
 324         }
 325
 326         ret = putenv("container=lxc");
 327         if (ret < 0)
 328                 return log_warn(-1, errno, "Failed to set environment variable");
 329
 330         /* Set container environment variables.*/
 331         if (init_ctx && init_ctx->container && init_ctx->container->lxc_conf) {
 332                 lxc_list_for_each(iterator, &init_ctx->container->lxc_conf->environment) {
 333                         char *env_tmp;
 334
 335                         env_tmp = strdup((char *)iterator->elem);
 336                         if (!env_tmp)
 337                                 return -1;
 338
 339                         ret = putenv(env_tmp);
 340                         if (ret < 0)
 341                                 return log_error_errno(-1, errno, "Failed to set environment variable: %s", (char *)iterator->elem);
 342                 }
 343         }
 344
 345         /* Set extra environment variables. */
 346         if (extra_env) {
 347                 for (; *extra_env; extra_env++) {
 348                         char *p;
 349
 350                         /* We just assume the user knows what they are doing, so
 351                          * we don't do any checks.
 352                          */
 353                         p = strdup(*extra_env);
 354                         if (!p)
 355                                 return -1;
 356
 357                         ret = putenv(p);
 358                         if (ret < 0)
 359                                 SYSWARN("Failed to set environment variable");
 360                 }
 361         }
 362
 363         return 0;
 364 }
 365
 366 static char *lxc_attach_getpwshell(uid_t uid)
 367 {
 368         __do_free char *line = NULL, *result = NULL;
 369         __do_fclose FILE *pipe_f = NULL;
 370         int fd, ret;
 371         pid_t pid;
 372         int pipes[2];
 373         bool found = false;
 374         size_t line_bufsz = 0;
 375
 376         /* We need to fork off a process that runs the getent program, and we
 377          * need to capture its output, so we use a pipe for that purpose.
 378          */
 379         ret = pipe2(pipes, O_CLOEXEC);
 380         if (ret < 0)
 381                 return NULL;
 382
 383         pid = fork();
 384         if (pid < 0) {
 385                 close(pipes[0]);
 386                 close(pipes[1]);
 387                 return NULL;
 388         }
 389
 390         if (!pid) {
 391                 char uid_buf[32];
 392                 char *arguments[] = {
 393                         "getent",
 394                         "passwd",
 395                         uid_buf,
 396                         NULL
 397                 };
 398
 399                 close(pipes[0]);
 400
 401                 /* We want to capture stdout. */
 402                 ret = dup2(pipes[1], STDOUT_FILENO);
 403                 close(pipes[1]);
 404                 if (ret < 0)
 405                         _exit(EXIT_FAILURE);
 406
 407                 /* Get rid of stdin/stderr, so we try to associate it with
 408                  * /dev/null.
 409                  */
 410                 fd = open_devnull();
 411                 if (fd < 0) {
 412                         close(STDIN_FILENO);
 413                         close(STDERR_FILENO);
 414                 } else {
 415                         (void)dup3(fd, STDIN_FILENO, O_CLOEXEC);
 416                         (void)dup3(fd, STDERR_FILENO, O_CLOEXEC);
 417                         close(fd);
 418                 }
 419
 420                 /* Finish argument list. */
 421                 ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid);
 422                 if (ret <= 0 || ret >= sizeof(uid_buf))
 423                         _exit(EXIT_FAILURE);
 424
 425                 /* Try to run getent program. */
 426                 (void)execvp("getent", arguments);
 427                 _exit(EXIT_FAILURE);
 428         }
 429
 430         close(pipes[1]);
 431
 432         pipe_f = fdopen(pipes[0], "r");
 433         if (!pipe_f) {
 434                 close(pipes[0]);
 435                 goto reap_child;
 436         }
 437         /* Transfer ownership of pipes[0] to pipe_f. */
 438         move_fd(pipes[0]);
 439
 440         while (getline(&line, &line_bufsz, pipe_f) != -1) {
 441                 int i;
 442                 long value;
 443                 char *token;
 444                 char *endptr = NULL, *saveptr = NULL;
 445
 446                 /* If we already found something, just continue to read
 447                 * until the pipe doesn't deliver any more data, but
 448                 * don't modify the existing data structure.
 449                  */
 450                 if (found)
 451                         continue;
 452
 453                 if (!line)
 454                         continue;
 455
 456                 /* Trim line on the right hand side. */
 457                 for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i)
 458                         line[i - 1] = '\0';
 459
 460                 /* Split into tokens: first: user name. */
 461                 token = strtok_r(line, ":", &saveptr);
 462                 if (!token)
 463                         continue;
 464
 465                 /* next: dummy password field */
 466                 token = strtok_r(NULL, ":", &saveptr);
 467                 if (!token)
 468                         continue;
 469
 470                 /* next: user id */
 471                 token = strtok_r(NULL, ":", &saveptr);
 472                 value = token ? strtol(token, &endptr, 10) : 0;
 473                 if (!token || !endptr || *endptr || value == LONG_MIN ||
 474                     value == LONG_MAX)
 475                         continue;
 476
 477                 /* dummy sanity check: user id matches */
 478                 if ((uid_t)value != uid)
 479                         continue;
 480
 481                 /* skip fields: gid, gecos, dir, go to next field 'shell' */
 482                 for (i = 0; i < 4; i++) {
 483                         token = strtok_r(NULL, ":", &saveptr);
 484                         if (!token)
 485                                 continue;
 486                 }
 487
 488                 if (!token)
 489                         continue;
 490
 491                 free_disarm(result);
 492                 result = strdup(token);
 493
 494                 /* Sanity check that there are no fields after that. */
 495                 token = strtok_r(NULL, ":", &saveptr);
 496                 if (token)
 497                         continue;
 498
 499                 found = true;
 500         }
 501
 502 reap_child:
 503         ret = wait_for_pid(pid);
 504         if (ret < 0)
 505                 return NULL;
 506
 507         if (!found)
 508                 return NULL;
 509
 510         return move_ptr(result);
 511 }
 512
 513 static void lxc_attach_get_init_uidgid(uid_t *init_uid, gid_t *init_gid)
 514 {
 515         __do_free char *line = NULL;
 516         __do_fclose FILE *proc_file = NULL;
 517         char proc_fn[LXC_PROC_STATUS_LEN];
 518         int ret;
 519         size_t line_bufsz = 0;
 520         long value = -1;
 521         uid_t uid = LXC_INVALID_UID;
 522         gid_t gid = LXC_INVALID_GID;
 523
 524         ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", 1);
 525         if (ret < 0 || ret >= LXC_PROC_STATUS_LEN)
 526                 return;
 527
 528         proc_file = fopen(proc_fn, "r");
 529         if (!proc_file)
 530                 return;
 531
 532         while (getline(&line, &line_bufsz, proc_file) != -1) {
 533                 /* Format is: real, effective, saved set user, fs we only care
 534                  * about real uid.
 535                  */
 536                 ret = sscanf(line, "Uid: %ld", &value);
 537                 if (ret != EOF && ret == 1) {
 538                         uid = (uid_t)value;
 539                 } else {
 540                         ret = sscanf(line, "Gid: %ld", &value);
 541                         if (ret != EOF && ret == 1)
 542                                 gid = (gid_t)value;
 543                 }
 544
 545                 if (uid != LXC_INVALID_UID && gid != LXC_INVALID_GID)
 546                         break;
 547         }
 548
 549         /* Only override arguments if we found something. */
 550         if (uid != LXC_INVALID_UID)
 551                 *init_uid = uid;
 552
 553         if (gid != LXC_INVALID_GID)
 554                 *init_gid = gid;
 555
 556         /* TODO: we should also parse supplementary groups and use
 557          * setgroups() to set them.
 558          */
 559 }
 560
 561 static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options)
 562 {
 563         __do_free char *path = NULL;
 564         int ret;
 565         bool bret;
 566
 567         if (!(options->namespaces & CLONE_NEWNS) ||
 568             !(options->attach_flags & LXC_ATTACH_LSM)) {
 569                 free_disarm(c->lxc_conf->seccomp.seccomp);
 570                 return true;
 571         }
 572
 573         /* Remove current setting. */
 574         if (!c->set_config_item(c, "lxc.seccomp.profile", "") &&
 575             !c->set_config_item(c, "lxc.seccomp", ""))
 576                 return false;
 577
 578         /* Fetch the current profile path over the cmd interface. */
 579         path = c->get_running_config_item(c, "lxc.seccomp.profile");
 580         if (!path) {
 581                 INFO("Failed to retrieve lxc.seccomp.profile");
 582
 583                 path = c->get_running_config_item(c, "lxc.seccomp");
 584                 if (!path)
 585                         return log_info(true, "Failed to retrieve lxc.seccomp");
 586         }
 587
 588         /* Copy the value into the new lxc_conf. */
 589         bret = c->set_config_item(c, "lxc.seccomp.profile", path);
 590         if (!bret)
 591                 return false;
 592
 593         /* Attempt to parse the resulting config. */
 594         ret = lxc_read_seccomp_config(c->lxc_conf);
 595         if (ret < 0)
 596                 return log_error(false, "Failed to retrieve seccomp policy");
 597
 598         return log_info(true, "Retrieved seccomp policy");
 599 }
 600
 601 static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options)
 602 {
 603         __do_free char *val = NULL;
 604
 605         /* Remove current setting. */
 606         if (!c->set_config_item(c, "lxc.no_new_privs", ""))
 607                 return log_info(false, "Failed to unset lxc.no_new_privs");
 608
 609         /* Retrieve currently active setting. */
 610         val = c->get_running_config_item(c, "lxc.no_new_privs");
 611         if (!val)
 612                 return log_info(false, "Failed to retrieve lxc.no_new_privs");
 613
 614         /* Set currently active setting. */
 615         return c->set_config_item(c, "lxc.no_new_privs", val);
 616 }
 617
 618 static signed long get_personality(const char *name, const char *lxcpath)
 619 {
 620         __do_free char *p = NULL;
 621
 622         p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath);
 623         if (!p)
 624                 return -1;
 625
 626         return lxc_config_parse_arch(p);
 627 }
 628
 629 struct attach_clone_payload {
 630         int ipc_socket;
 631         int terminal_slave_fd;
 632         lxc_attach_options_t *options;
 633         struct lxc_proc_context_info *init_ctx;
 634         lxc_attach_exec_t exec_function;
 635         void *exec_payload;
 636 };
 637
 638 static void lxc_put_attach_clone_payload(struct attach_clone_payload *p)
 639 {
 640         close_prot_errno_disarm(p->ipc_socket);
 641         close_prot_errno_disarm(p->terminal_slave_fd);
 642         if (p->init_ctx) {
 643                 lxc_proc_put_context_info(p->init_ctx);
 644                 p->init_ctx = NULL;
 645         }
 646 }
 647
 648 static int attach_child_main(struct attach_clone_payload *payload)
 649 {
 650         int lsm_fd, ret;
 651         uid_t new_uid;
 652         gid_t new_gid;
 653         uid_t ns_root_uid = 0;
 654         gid_t ns_root_gid = 0;
 655         lxc_attach_options_t* options = payload->options;
 656         struct lxc_proc_context_info* init_ctx = payload->init_ctx;
 657         bool needs_lsm = (options->namespaces & CLONE_NEWNS) &&
 658                          (options->attach_flags & LXC_ATTACH_LSM) &&
 659                          init_ctx->lsm_label;
 660
 661         /* A description of the purpose of this functionality is provided in the
 662          * lxc-attach(1) manual page. We have to remount here and not in the
 663          * parent process, otherwise /proc may not properly reflect the new pid
 664          * namespace.
 665          */
 666         if (!(options->namespaces & CLONE_NEWNS) &&
 667             (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) {
 668                 ret = lxc_attach_remount_sys_proc();
 669                 if (ret < 0)
 670                         goto on_error;
 671
 672                 TRACE("Remounted \"/proc\" and \"/sys\"");
 673         }
 674
 675 /* Now perform additional attachments. */
 676 #if HAVE_SYS_PERSONALITY_H
 677         if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) {
 678                 long new_personality;
 679
 680                 if (options->personality < 0)
 681                         new_personality = init_ctx->personality;
 682                 else
 683                         new_personality = options->personality;
 684
 685                 ret = personality(new_personality);
 686                 if (ret < 0)
 687                         goto on_error;
 688
 689                 TRACE("Set new personality");
 690         }
 691 #endif
 692
 693         if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) {
 694                 ret = lxc_attach_drop_privs(init_ctx);
 695                 if (ret < 0)
 696                         goto on_error;
 697
 698                 TRACE("Dropped capabilities");
 699         }
 700
 701         /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL)
 702          * if you want this to be a no-op).
 703          */
 704         ret = lxc_attach_set_environment(init_ctx,
 705                                          options->env_policy,
 706                                          options->extra_env_vars,
 707                                          options->extra_keep_env);
 708         if (ret < 0)
 709                 goto on_error;
 710
 711         TRACE("Set up environment");
 712
 713         /* This remark only affects fully unprivileged containers:
 714          * Receive fd for LSM security module before we set{g,u}id(). The reason
 715          * is that on set{g,u}id() the kernel will a) make us undumpable and b)
 716          * we will change our effective uid. This means our effective uid will
 717          * be different from the effective uid of the process that created us
 718          * which means that this processs no longer has capabilities in our
 719          * namespace including CAP_SYS_PTRACE. This means we will not be able to
 720          * read and /proc/<pid> files for the process anymore when /proc is
 721          * mounted with hidepid={1,2}. So let's get the lsm label fd before the
 722          * set{g,u}id().
 723          */
 724         if (needs_lsm) {
 725                 ret = lxc_abstract_unix_recv_fds(payload->ipc_socket, &lsm_fd, 1, NULL, 0);
 726                 if (ret <= 0) {
 727                         if (ret < 0)
 728                                 SYSERROR("Failed to receive lsm label fd");
 729
 730                         goto on_error;
 731                 }
 732
 733                 TRACE("Received LSM label file descriptor %d from parent", lsm_fd);
 734         }
 735
 736         if (options->stdin_fd > 0 && isatty(options->stdin_fd)) {
 737                 ret = lxc_make_controlling_terminal(options->stdin_fd);
 738                 if (ret < 0)
 739                         goto on_error;
 740         }
 741
 742         if (!lxc_setgroups(0, NULL) && errno != EPERM)
 743                 goto on_error;
 744
 745         if (options->namespaces & CLONE_NEWUSER) {
 746                 /* Check whether nsuid 0 has a mapping. */
 747                 ns_root_uid = get_ns_uid(0);
 748
 749                 /* Check whether nsgid 0 has a mapping. */
 750                 ns_root_gid = get_ns_gid(0);
 751
 752                 /* If there's no mapping for nsuid 0 try to retrieve the nsuid
 753                  * init was started with.
 754                  */
 755                 if (ns_root_uid == LXC_INVALID_UID)
 756                         lxc_attach_get_init_uidgid(&ns_root_uid, &ns_root_gid);
 757
 758                 if (ns_root_uid == LXC_INVALID_UID)
 759                         goto on_error;
 760
 761                 if (!lxc_switch_uid_gid(ns_root_uid, ns_root_gid))
 762                         goto on_error;
 763         }
 764
 765         /* Set {u,g}id. */
 766         if (options->uid != LXC_INVALID_UID)
 767                 new_uid = options->uid;
 768         else
 769                 new_uid = ns_root_uid;
 770
 771         if (options->gid != LXC_INVALID_GID)
 772                 new_gid = options->gid;
 773         else
 774                 new_gid = ns_root_gid;
 775
 776         if ((init_ctx->container && init_ctx->container->lxc_conf &&
 777              init_ctx->container->lxc_conf->no_new_privs) ||
 778             (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) {
 779                 ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0),
 780                             prctl_arg(0), prctl_arg(0));
 781                 if (ret < 0)
 782                         goto on_error;
 783
 784                 TRACE("Set PR_SET_NO_NEW_PRIVS");
 785         }
 786
 787         if (needs_lsm) {
 788                 bool on_exec;
 789
 790                 /* Change into our new LSM profile. */
 791                 on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false;
 792
 793                 ret = lsm_process_label_set_at(lsm_fd, init_ctx->lsm_label, on_exec);
 794                 close(lsm_fd);
 795                 if (ret < 0)
 796                         goto on_error;
 797
 798                 TRACE("Set %s LSM label to \"%s\"", lsm_name(), init_ctx->lsm_label);
 799         }
 800
 801         if (init_ctx->container && init_ctx->container->lxc_conf &&
 802             init_ctx->container->lxc_conf->seccomp.seccomp) {
 803                 struct lxc_conf *conf = init_ctx->container->lxc_conf;
 804
 805                 ret = lxc_seccomp_load(conf);
 806                 if (ret < 0)
 807                         goto on_error;
 808
 809                 TRACE("Loaded seccomp profile");
 810
 811                 ret = lxc_seccomp_send_notifier_fd(&conf->seccomp, payload->ipc_socket);
 812                 if (ret < 0)
 813                         goto on_error;
 814         }
 815
 816         close(payload->ipc_socket);
 817         payload->ipc_socket = -EBADF;
 818         lxc_proc_put_context_info(init_ctx);
 819         payload->init_ctx = NULL;
 820
 821         /* The following is done after the communication socket is shut down.
 822          * That way, all errors that might (though unlikely) occur up until this
 823          * point will have their messages printed to the original stderr (if
 824          * logging is so configured) and not the fd the user supplied, if any.
 825          */
 826
 827         /* Fd handling for stdin, stdout and stderr; ignore errors here, user
 828          * may want to make sure the fds are closed, for example.
 829          */
 830         if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO)
 831                 (void)dup2(options->stdin_fd, STDIN_FILENO);
 832
 833         if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO)
 834                 (void)dup2(options->stdout_fd, STDOUT_FILENO);
 835
 836         if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO)
 837                 (void)dup2(options->stderr_fd, STDERR_FILENO);
 838
 839         /* close the old fds */
 840         if (options->stdin_fd > STDERR_FILENO)
 841                 close(options->stdin_fd);
 842
 843         if (options->stdout_fd > STDERR_FILENO)
 844                 close(options->stdout_fd);
 845
 846         if (options->stderr_fd > STDERR_FILENO)
 847                 close(options->stderr_fd);
 848
 849         /*
 850          * Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also
 851          * here, ignore errors.
 852          */
 853         for (int fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) {
 854                 ret = fd_cloexec(fd, false);
 855                 if (ret < 0) {
 856                         SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd);
 857                         goto on_error;
 858                 }
 859         }
 860
 861         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
 862                 ret = lxc_terminal_prepare_login(payload->terminal_slave_fd);
 863                 if (ret < 0) {
 864                         SYSERROR("Failed to prepare terminal file descriptor %d", payload->terminal_slave_fd);
 865                         goto on_error;
 866                 }
 867
 868                 TRACE("Prepared terminal file descriptor %d", payload->terminal_slave_fd);
 869         }
 870
 871         /* Avoid unnecessary syscalls. */
 872         if (new_uid == ns_root_uid)
 873                 new_uid = LXC_INVALID_UID;
 874
 875         if (new_gid == ns_root_gid)
 876                 new_gid = LXC_INVALID_GID;
 877
 878         if (!lxc_switch_uid_gid(new_uid, new_gid))
 879                 goto on_error;
 880
 881         /* We're done, so we can now do whatever the user intended us to do. */
 882         _exit(payload->exec_function(payload->exec_payload));
 883
 884 on_error:
 885         lxc_put_attach_clone_payload(payload);
 886         _exit(EXIT_FAILURE);
 887 }
 888
 889 static int lxc_attach_terminal(struct lxc_conf *conf,
 890                                struct lxc_terminal *terminal)
 891 {
 892         int ret;
 893
 894         lxc_terminal_init(terminal);
 895
 896         ret = lxc_terminal_create(terminal);
 897         if (ret < 0)
 898                 return log_error(-1, "Failed to create terminal");
 899
 900         /* Shift ttys to container. */
 901         ret = lxc_terminal_map_ids(conf, terminal);
 902         if (ret < 0) {
 903                 ERROR("Failed to chown terminal");
 904                 goto on_error;
 905         }
 906
 907         return 0;
 908
 909 on_error:
 910         lxc_terminal_delete(terminal);
 911         lxc_terminal_conf_free(terminal);
 912         return -1;
 913 }
 914
 915 static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal,
 916                                              struct lxc_epoll_descr *descr)
 917 {
 918         int ret;
 919
 920         ret = lxc_mainloop_open(descr);
 921         if (ret < 0)
 922                 return log_error(-1, "Failed to create mainloop");
 923
 924         ret = lxc_terminal_mainloop_add(descr, terminal);
 925         if (ret < 0) {
 926                 lxc_mainloop_close(descr);
 927                 return log_error(-1, "Failed to add handlers to mainloop");
 928         }
 929
 930         return 0;
 931 }
 932
 933 static inline void lxc_attach_terminal_close_master(struct lxc_terminal *terminal)
 934 {
 935         close_prot_errno_disarm(terminal->master);
 936 }
 937
 938 static inline void lxc_attach_terminal_close_slave(struct lxc_terminal *terminal)
 939 {
 940         close_prot_errno_disarm(terminal->slave);
 941 }
 942
 943 static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal)
 944 {
 945         close_prot_errno_disarm(terminal->peer);
 946 }
 947
 948 static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal)
 949 {
 950         close_prot_errno_disarm(terminal->log_fd);
 951 }
 952
 953 int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function,
 954                void *exec_payload, lxc_attach_options_t *options,
 955                pid_t *attached_process)
 956 {
 957         int i, ret, status;
 958         int ipc_sockets[2];
 959         char *cwd, *new_cwd;
 960         signed long personality;
 961         pid_t attached_pid, init_pid, pid;
 962         struct lxc_proc_context_info *init_ctx;
 963         struct lxc_terminal terminal;
 964         struct lxc_conf *conf;
 965         char *name, *lxcpath;
 966         struct attach_clone_payload payload = {0};
 967
 968         ret = access("/proc/self/ns", X_OK);
 969         if (ret)
 970                 return log_error_errno(-1, errno, "Does this kernel version support namespaces?");
 971
 972         if (!container)
 973                 return ret_set_errno(-1, EINVAL);
 974
 975         if (!lxc_container_get(container))
 976                 return ret_set_errno(-1, EINVAL);
 977
 978         name = container->name;
 979         lxcpath = container->config_path;
 980
 981         if (!options)
 982                 options = &attach_static_default_options;
 983
 984         init_pid = lxc_cmd_get_init_pid(name, lxcpath);
 985         if (init_pid < 0) {
 986                 lxc_container_put(container);
 987                 return log_error(-1, "Failed to get init pid");
 988         }
 989
 990         init_ctx = lxc_proc_get_context_info(init_pid);
 991         if (!init_ctx) {
 992                 ERROR("Failed to get context of init process: %ld", (long)init_pid);
 993                 lxc_container_put(container);
 994                 return -1;
 995         }
 996
 997         init_ctx->container = container;
 998
 999         personality = get_personality(name, lxcpath);
1000         if (init_ctx->personality < 0) {
1001                 ERROR("Failed to get personality of the container");
1002                 lxc_proc_put_context_info(init_ctx);
1003                 return -1;
1004         }
1005         init_ctx->personality = personality;
1006
1007         if (!init_ctx->container->lxc_conf) {
1008                 init_ctx->container->lxc_conf = lxc_conf_init();
1009                 if (!init_ctx->container->lxc_conf) {
1010                         lxc_proc_put_context_info(init_ctx);
1011                         return -1;
1012                 }
1013         }
1014         conf = init_ctx->container->lxc_conf;
1015
1016         if (!fetch_seccomp(init_ctx->container, options))
1017                 WARN("Failed to get seccomp policy");
1018
1019         if (!no_new_privs(init_ctx->container, options))
1020                 WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set");
1021
1022         cwd = getcwd(NULL, 0);
1023
1024         /* Determine which namespaces the container was created with
1025          * by asking lxc-start, if necessary.
1026          */
1027         if (options->namespaces == -1) {
1028                 options->namespaces = lxc_cmd_get_clone_flags(name, lxcpath);
1029                 /* call failed */
1030                 if (options->namespaces == -1) {
1031                         ERROR("Failed to automatically determine the "
1032                               "namespaces which the container uses");
1033                         free(cwd);
1034                         lxc_proc_put_context_info(init_ctx);
1035                         return -1;
1036                 }
1037
1038                 for (i = 0; i < LXC_NS_MAX; i++) {
1039                         if (ns_info[i].clone_flag & CLONE_NEWCGROUP)
1040                                 if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) ||
1041                                     !cgns_supported())
1042                                         continue;
1043
1044                         if (ns_info[i].clone_flag & options->namespaces)
1045                                 continue;
1046
1047                         init_ctx->ns_inherited |= ns_info[i].clone_flag;
1048                 }
1049         }
1050
1051         pid = lxc_raw_getpid();
1052
1053         for (i = 0; i < LXC_NS_MAX; i++) {
1054                 int j;
1055
1056                 if (options->namespaces & ns_info[i].clone_flag)
1057                         init_ctx->ns_fd[i] = lxc_preserve_ns(init_pid, ns_info[i].proc_name);
1058                 else if (init_ctx->ns_inherited & ns_info[i].clone_flag)
1059                         init_ctx->ns_fd[i] = in_same_namespace(pid, init_pid, ns_info[i].proc_name);
1060                 else
1061                         continue;
1062
1063                 if (init_ctx->ns_fd[i] >= 0)
1064                         continue;
1065
1066                 if (init_ctx->ns_fd[i] == -EINVAL) {
1067                         DEBUG("Inheriting %s namespace from %d",
1068                               ns_info[i].proc_name, pid);
1069                         init_ctx->ns_inherited &= ~ns_info[i].clone_flag;
1070                         continue;
1071                 }
1072
1073                 /* We failed to preserve the namespace. */
1074                 SYSERROR("Failed to attach to %s namespace of %d",
1075                          ns_info[i].proc_name, pid);
1076
1077                 /* Close all already opened file descriptors before we return an
1078                  * error, so we don't leak them.
1079                  */
1080                 for (j = 0; j < i; j++)
1081                         close(init_ctx->ns_fd[j]);
1082
1083                 free(cwd);
1084                 lxc_proc_put_context_info(init_ctx);
1085                 return -1;
1086         }
1087
1088         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1089                 ret = lxc_attach_terminal(conf, &terminal);
1090                 if (ret < 0) {
1091                         ERROR("Failed to setup new terminal");
1092                         free(cwd);
1093                         lxc_proc_put_context_info(init_ctx);
1094                         return -1;
1095                 }
1096
1097                 terminal.log_fd = options->log_fd;
1098         } else {
1099                 lxc_terminal_init(&terminal);
1100         }
1101
1102         /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order
1103          * to make sure we don't irritate other threads that want to fork+exec
1104          * away
1105          *
1106          * IMPORTANT: if the initial process is multithreaded and another call
1107          * just fork()s away without exec'ing directly after, the socket fd will
1108          * exist in the forked process from the other thread and any close() in
1109          * our own child process will not really cause the socket to close
1110          * properly, potentially causing the parent to hang.
1111          *
1112          * For this reason, while IPC is still active, we have to use shutdown()
1113          * if the child exits prematurely in order to signal that the socket is
1114          * closed and cannot assume that the child exiting will automatically do
1115          * that.
1116          *
1117          * IPC mechanism: (X is receiver)
1118          *   initial process        intermediate          attached
1119          *        X           <---  send pid of
1120          *                          attached proc,
1121          *                          then exit
1122          *    send 0 ------------------------------------>    X
1123          *                                              [do initialization]
1124          *        X  <------------------------------------  send 1
1125          *   [add to cgroup, ...]
1126          *    send 2 ------------------------------------>    X
1127          *                                              [set LXC_ATTACH_NO_NEW_PRIVS]
1128          *        X  <------------------------------------  send 3
1129          *   [open LSM label fd]
1130          *    send 4 ------------------------------------>    X
1131          *                                              [set LSM label]
1132          *   close socket                                 close socket
1133          *                                                run program
1134          */
1135         ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets);
1136         if (ret < 0) {
1137                 SYSERROR("Could not set up required IPC mechanism for attaching");
1138                 free(cwd);
1139                 lxc_proc_put_context_info(init_ctx);
1140                 return -1;
1141         }
1142
1143         /* Create intermediate subprocess, two reasons:
1144          *       1. We can't setns() in the child itself, since we want to make
1145          *          sure we are properly attached to the pidns.
1146          *       2. Also, the initial thread has to put the attached process
1147          *          into the cgroup, which we can only do if we didn't already
1148          *          setns() (otherwise, user namespaces will hate us).
1149          */
1150         pid = fork();
1151         if (pid < 0) {
1152                 SYSERROR("Failed to create first subprocess");
1153                 free(cwd);
1154                 lxc_proc_put_context_info(init_ctx);
1155                 return -1;
1156         }
1157
1158         if (pid) {
1159                 int ret_parent = -1;
1160                 pid_t to_cleanup_pid = pid;
1161                 struct lxc_epoll_descr descr = {0};
1162
1163                 /* close unneeded file descriptors */
1164                 close(ipc_sockets[1]);
1165                 free(cwd);
1166                 lxc_proc_close_ns_fd(init_ctx);
1167                 if (options->attach_flags & LXC_ATTACH_TERMINAL)
1168                         lxc_attach_terminal_close_slave(&terminal);
1169
1170                 /* Attach to cgroup, if requested. */
1171                 if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) {
1172                         /*
1173                          * If this is the unified hierarchy cgroup_attach() is
1174                          * enough.
1175                          */
1176                         ret = cgroup_attach(name, lxcpath, pid);
1177                         if (ret) {
1178                                 __do_cgroup_exit struct cgroup_ops *cgroup_ops = NULL;
1179
1180                                 cgroup_ops = cgroup_init(conf);
1181                                 if (!cgroup_ops)
1182                                         goto on_error;
1183
1184                                 if (!cgroup_ops->attach(cgroup_ops, name, lxcpath, pid))
1185                                         goto on_error;
1186                         }
1187                         TRACE("Moved intermediate process %d into container's cgroups", pid);
1188                 }
1189
1190                 /* Setup /proc limits */
1191                 if (!lxc_list_empty(&conf->procs)) {
1192                         ret = setup_proc_filesystem(&conf->procs, pid);
1193                         if (ret < 0)
1194                                 goto on_error;
1195                 }
1196
1197                 /* Setup resource limits */
1198                 if (!lxc_list_empty(&conf->limits)) {
1199                         ret = setup_resource_limits(&conf->limits, pid);
1200                         if (ret < 0)
1201                                 goto on_error;
1202                 }
1203
1204                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1205                         ret = lxc_attach_terminal_mainloop_init(&terminal, &descr);
1206                         if (ret < 0)
1207                                 goto on_error;
1208
1209                         TRACE("Initialized terminal mainloop");
1210                 }
1211
1212                 /* Let the child process know to go ahead. */
1213                 status = 0;
1214                 ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status));
1215                 if (ret != sizeof(status))
1216                         goto close_mainloop;
1217
1218                 TRACE("Told intermediate process to start initializing");
1219
1220                 /* Get pid of attached process from intermediate process. */
1221                 ret = lxc_read_nointr(ipc_sockets[0], &attached_pid, sizeof(attached_pid));
1222                 if (ret != sizeof(attached_pid))
1223                         goto close_mainloop;
1224
1225                 TRACE("Received pid %d of attached process in parent pid namespace", attached_pid);
1226
1227                 /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */
1228                 if (options->stdin_fd == 0) {
1229                         signal(SIGINT, SIG_IGN);
1230                         signal(SIGQUIT, SIG_IGN);
1231                 }
1232
1233                 /* Reap intermediate process. */
1234                 ret = wait_for_pid(pid);
1235                 if (ret < 0)
1236                         goto close_mainloop;
1237
1238                 TRACE("Intermediate process %d exited", pid);
1239
1240                 /* We will always have to reap the attached process now. */
1241                 to_cleanup_pid = attached_pid;
1242
1243                 /* Open LSM fd and send it to child. */
1244                 if ((options->namespaces & CLONE_NEWNS) &&
1245                     (options->attach_flags & LXC_ATTACH_LSM) &&
1246                     init_ctx->lsm_label) {
1247                         int labelfd;
1248                         bool on_exec;
1249
1250                         ret = -1;
1251                         on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false;
1252                         labelfd = lsm_process_label_fd_get(attached_pid, on_exec);
1253                         if (labelfd < 0)
1254                                 goto close_mainloop;
1255
1256                         TRACE("Opened LSM label file descriptor %d", labelfd);
1257
1258                         /* Send child fd of the LSM security module to write to. */
1259                         ret = lxc_abstract_unix_send_fds(ipc_sockets[0], &labelfd, 1, NULL, 0);
1260                         if (ret <= 0) {
1261                                 if (ret < 0)
1262                                         SYSERROR("Failed to send lsm label fd");
1263
1264                                 close(labelfd);
1265                                 goto close_mainloop;
1266                         }
1267
1268                         close(labelfd);
1269                         TRACE("Sent LSM label file descriptor %d to child", labelfd);
1270                 }
1271
1272                 if (conf && conf->seccomp.seccomp) {
1273                         ret = lxc_seccomp_recv_notifier_fd(&conf->seccomp, ipc_sockets[0]);
1274                         if (ret < 0)
1275                                 goto close_mainloop;
1276
1277                         ret = lxc_seccomp_add_notifier(name, lxcpath, &conf->seccomp);
1278                         if (ret < 0)
1279                                 goto close_mainloop;
1280                 }
1281
1282                 /* We're done, the child process should now execute whatever it
1283                  * is that the user requested. The parent can now track it with
1284                  * waitpid() or similar.
1285                  */
1286
1287                 *attached_process = attached_pid;
1288
1289                 /* Now shut down communication with child, we're done. */
1290                 shutdown(ipc_sockets[0], SHUT_RDWR);
1291                 close(ipc_sockets[0]);
1292                 ipc_sockets[0] = -1;
1293
1294                 ret_parent = 0;
1295                 to_cleanup_pid = -1;
1296
1297                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1298                         ret = lxc_mainloop(&descr, -1);
1299                         if (ret < 0) {
1300                                 ret_parent = -1;
1301                                 to_cleanup_pid = attached_pid;
1302                         }
1303                 }
1304
1305         close_mainloop:
1306                 if (options->attach_flags & LXC_ATTACH_TERMINAL)
1307                         lxc_mainloop_close(&descr);
1308
1309         on_error:
1310                 if (ipc_sockets[0] >= 0) {
1311                         shutdown(ipc_sockets[0], SHUT_RDWR);
1312                         close(ipc_sockets[0]);
1313                 }
1314
1315                 if (to_cleanup_pid > 0)
1316                         (void)wait_for_pid(to_cleanup_pid);
1317
1318                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1319                         lxc_terminal_delete(&terminal);
1320                         lxc_terminal_conf_free(&terminal);
1321                 }
1322
1323                 lxc_proc_put_context_info(init_ctx);
1324                 return ret_parent;
1325         }
1326
1327         /* close unneeded file descriptors */
1328         close(ipc_sockets[0]);
1329         ipc_sockets[0] = -EBADF;
1330
1331         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1332                 lxc_attach_terminal_close_master(&terminal);
1333                 lxc_attach_terminal_close_peer(&terminal);
1334                 lxc_attach_terminal_close_log(&terminal);
1335         }
1336
1337         /* Wait for the parent to have setup cgroups. */
1338         ret = lxc_read_nointr(ipc_sockets[1], &status, sizeof(status));
1339         if (ret != sizeof(status)) {
1340                 shutdown(ipc_sockets[1], SHUT_RDWR);
1341                 lxc_proc_put_context_info(init_ctx);
1342                 _exit(EXIT_FAILURE);
1343         }
1344
1345         TRACE("Intermediate process starting to initialize");
1346
1347         /* Attach now, create another subprocess later, since pid namespaces
1348          * only really affect the children of the current process.
1349          */
1350         ret = lxc_attach_to_ns(init_pid, init_ctx);
1351         if (ret < 0) {
1352                 ERROR("Failed to enter namespaces");
1353                 shutdown(ipc_sockets[1], SHUT_RDWR);
1354                 lxc_proc_put_context_info(init_ctx);
1355                 _exit(EXIT_FAILURE);
1356         }
1357
1358         /* close namespace file descriptors */
1359         lxc_proc_close_ns_fd(init_ctx);
1360
1361         /* Attach succeeded, try to cwd. */
1362         if (options->initial_cwd)
1363                 new_cwd = options->initial_cwd;
1364         else
1365                 new_cwd = cwd;
1366         if (new_cwd) {
1367                 ret = chdir(new_cwd);
1368                 if (ret < 0)
1369                         WARN("Could not change directory to \"%s\"", new_cwd);
1370         }
1371         free(cwd);
1372
1373         /* Create attached process. */
1374         payload.ipc_socket = ipc_sockets[1];
1375         payload.options = options;
1376         payload.init_ctx = init_ctx;
1377         payload.terminal_slave_fd = terminal.slave;
1378         payload.exec_function = exec_function;
1379         payload.exec_payload = exec_payload;
1380
1381         pid = lxc_raw_clone(CLONE_PARENT, NULL);
1382         if (pid < 0) {
1383                 SYSERROR("Failed to clone attached process");
1384                 shutdown(ipc_sockets[1], SHUT_RDWR);
1385                 lxc_proc_put_context_info(init_ctx);
1386                 _exit(EXIT_FAILURE);
1387         }
1388
1389         if (pid == 0) {
1390                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1391                         ret = pthread_sigmask(SIG_SETMASK,
1392                                               &terminal.tty_state->oldmask, NULL);
1393                         if (ret < 0) {
1394                                 SYSERROR("Failed to reset signal mask");
1395                                 _exit(EXIT_FAILURE);
1396                         }
1397                 }
1398
1399                 ret = attach_child_main(&payload);
1400                 if (ret < 0)
1401                         ERROR("Failed to exec");
1402
1403                 _exit(EXIT_FAILURE);
1404         }
1405
1406         if (options->attach_flags & LXC_ATTACH_TERMINAL)
1407                 lxc_attach_terminal_close_slave(&terminal);
1408
1409         /* Tell grandparent the pid of the pid of the newly created child. */
1410         ret = lxc_write_nointr(ipc_sockets[1], &pid, sizeof(pid));
1411         if (ret != sizeof(pid)) {
1412                 /* If this really happens here, this is very unfortunate, since
1413                  * the parent will not know the pid of the attached process and
1414                  * will not be able to wait for it (and we won't either due to
1415                  * CLONE_PARENT) so the parent won't be able to reap it and the
1416                  * attached process will remain a zombie.
1417                  */
1418                 shutdown(ipc_sockets[1], SHUT_RDWR);
1419                 lxc_proc_put_context_info(init_ctx);
1420                 _exit(EXIT_FAILURE);
1421         }
1422
1423         TRACE("Sending pid %d of attached process", pid);
1424
1425         /* The rest is in the hands of the initial and the attached process. */
1426         lxc_proc_put_context_info(init_ctx);
1427         _exit(EXIT_SUCCESS);
1428 }
1429
1430 int lxc_attach_run_command(void *payload)
1431 {
1432         int ret = -1;
1433         lxc_attach_command_t *cmd = payload;
1434
1435         ret = execvp(cmd->program, cmd->argv);
1436         if (ret < 0) {
1437                 switch (errno) {
1438                 case ENOEXEC:
1439                         ret = 126;
1440                         break;
1441                 case ENOENT:
1442                         ret = 127;
1443                         break;
1444                 }
1445         }
1446
1447         return log_error_errno(ret, errno, "Failed to exec \"%s\"", cmd->program);
1448 }
1449
1450 int lxc_attach_run_shell(void* payload)
1451 {
1452         __do_free char *buf = NULL;
1453         uid_t uid;
1454         struct passwd pwent;
1455         struct passwd *pwentp = NULL;
1456         char *user_shell;
1457         size_t bufsize;
1458         int ret;
1459
1460         /* Ignore payload parameter. */
1461         (void)payload;
1462
1463         uid = getuid();
1464
1465         bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
1466         if (bufsize == -1)
1467                 bufsize = 1024;
1468
1469         buf = malloc(bufsize);
1470         if (buf) {
1471                 ret = getpwuid_r(uid, &pwent, buf, bufsize, &pwentp);
1472                 if (!pwentp) {
1473                         if (ret == 0)
1474                                 WARN("Could not find matched password record");
1475
1476                         WARN("Failed to get password record - %u", uid);
1477                 }
1478         }
1479
1480         /* This probably happens because of incompatible nss implementations in
1481          * host and container (remember, this code is still using the host's
1482          * glibc but our mount namespace is in the container) we may try to get
1483          * the information by spawning a [getent passwd uid] process and parsing
1484          * the result.
1485          */
1486         if (!pwentp)
1487                 user_shell = lxc_attach_getpwshell(uid);
1488         else
1489                 user_shell = pwent.pw_shell;
1490
1491         if (user_shell)
1492                 execlp(user_shell, user_shell, (char *)NULL);
1493
1494         /* Executed if either no passwd entry or execvp fails, we will fall back
1495          * on /bin/sh as a default shell.
1496          */
1497         execlp("/bin/sh", "/bin/sh", (char *)NULL);
1498
1499         SYSERROR("Failed to execute shell");
1500         if (!pwentp)
1501                 free(user_shell);
1502
1503         return -1;
1504 }