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 #define _GNU_SOURCE
  25 #include <errno.h>
  26 #include <fcntl.h>
  27 #include <grp.h>
  28 #include <pwd.h>
  29 #include <signal.h>
  30 #include <stdio.h>
  31 #include <stdlib.h>
  32 #include <string.h>
  33 #include <unistd.h>
  34 #include <linux/unistd.h>
  35 #include <sys/mount.h>
  36 #include <sys/param.h>
  37 #include <sys/prctl.h>
  38 #include <sys/socket.h>
  39 #include <sys/syscall.h>
  40 #include <sys/wait.h>
  41
  42 #include <lxc/lxccontainer.h>
  43
  44 #ifndef HAVE_DECL_PR_CAPBSET_DROP
  45 #define PR_CAPBSET_DROP 24
  46 #endif
  47
  48 #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS
  49 #define PR_SET_NO_NEW_PRIVS 38
  50 #endif
  51
  52 #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS
  53 #define PR_GET_NO_NEW_PRIVS 39
  54 #endif
  55
  56 #include "af_unix.h"
  57 #include "attach.h"
  58 #include "caps.h"
  59 #include "cgroup.h"
  60 #include "commands.h"
  61 #include "conf.h"
  62 #include "config.h"
  63 #include "confile.h"
  64 #include "log.h"
  65 #include "lsm/lsm.h"
  66 #include "lxclock.h"
  67 #include "lxcseccomp.h"
  68 #include "mainloop.h"
  69 #include "namespace.h"
  70 #include "terminal.h"
  71 #include "utils.h"
  72
  73 #if HAVE_SYS_PERSONALITY_H
  74 #include <sys/personality.h>
  75 #endif
  76
  77 #ifndef SOCK_CLOEXEC
  78 #define SOCK_CLOEXEC 02000000
  79 #endif
  80
  81 #ifndef MS_REC
  82 #define MS_REC 16384
  83 #endif
  84
  85 #ifndef MS_SLAVE
  86 #define MS_SLAVE (1 << 19)
  87 #endif
  88
  89 lxc_log_define(lxc_attach, lxc);
  90
  91 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
  92 #define __PROC_STATUS_LEN (5 + (LXC_NUMSTRLEN64) + 7 + 1)
  93 static struct lxc_proc_context_info *lxc_proc_get_context_info(pid_t pid)
  94 {
  95         int ret;
  96         bool found;
  97         FILE *proc_file;
  98         char proc_fn[__PROC_STATUS_LEN];
  99         size_t line_bufsz = 0;
 100         char *line = NULL;
 101         struct lxc_proc_context_info *info = NULL;
 102
 103         /* Read capabilities. */
 104         ret = snprintf(proc_fn, __PROC_STATUS_LEN, "/proc/%d/status", pid);
 105         if (ret < 0 || ret >= __PROC_STATUS_LEN)
 106                 goto on_error;
 107
 108         proc_file = fopen(proc_fn, "r");
 109         if (!proc_file) {
 110                 SYSERROR("Could not open %s.", proc_fn);
 111                 goto on_error;
 112         }
 113
 114         info = calloc(1, sizeof(*info));
 115         if (!info) {
 116                 SYSERROR("Could not allocate memory.");
 117                 fclose(proc_file);
 118                 return NULL;
 119         }
 120
 121         found = false;
 122         while (getline(&line, &line_bufsz, proc_file) != -1) {
 123                 ret = sscanf(line, "CapBnd: %llx", &info->capability_mask);
 124                 if (ret != EOF && ret == 1) {
 125                         found = true;
 126                         break;
 127                 }
 128         }
 129
 130         free(line);
 131         fclose(proc_file);
 132
 133         if (!found) {
 134                 SYSERROR("Could not read capability bounding set from %s.",
 135                          proc_fn);
 136                 errno = ENOENT;
 137                 goto on_error;
 138         }
 139
 140         info->lsm_label = lsm_process_label_get(pid);
 141         info->ns_inherited = 0;
 142         memset(info->ns_fd, -1, sizeof(int) * LXC_NS_MAX);
 143
 144         return info;
 145
 146 on_error:
 147         free(info);
 148         return NULL;
 149 }
 150
 151 static inline void lxc_proc_close_ns_fd(struct lxc_proc_context_info *ctx)
 152 {
 153         int i;
 154
 155         for (i = 0; i < LXC_NS_MAX; i++) {
 156                 if (ctx->ns_fd[i] < 0)
 157                         continue;
 158                 close(ctx->ns_fd[i]);
 159                 ctx->ns_fd[i] = -EBADF;
 160         }
 161 }
 162
 163 static void lxc_proc_put_context_info(struct lxc_proc_context_info *ctx)
 164 {
 165         free(ctx->lsm_label);
 166         ctx->lsm_label = NULL;
 167
 168         if (ctx->container) {
 169                 lxc_container_put(ctx->container);
 170                 ctx->container = NULL;
 171         }
 172
 173         lxc_proc_close_ns_fd(ctx);
 174         free(ctx);
 175         ctx = NULL;
 176 }
 177
 178 /**
 179  * in_same_namespace - Check whether two processes are in the same namespace.
 180  * @pid1 - PID of the first process.
 181  * @pid2 - PID of the second process.
 182  * @ns   - Name of the namespace to check. Must correspond to one of the names
 183  *         for the namespaces as shown in /proc/<pid/ns/
 184  *
 185  * If the two processes are not in the same namespace returns an fd to the
 186  * namespace of the second process identified by @pid2. If the two processes are
 187  * in the same namespace returns -EINVAL, -1 if an error occurred.
 188  */
 189 static int in_same_namespace(pid_t pid1, pid_t pid2, const char *ns)
 190 {
 191         int ns_fd1 = -1, ns_fd2 = -1, ret = -1;
 192         struct stat ns_st1, ns_st2;
 193
 194         ns_fd1 = lxc_preserve_ns(pid1, ns);
 195         if (ns_fd1 < 0) {
 196                 /* The kernel does not support this namespace. This is not an
 197                  * error.
 198                  */
 199                 if (errno == ENOENT)
 200                         return -EINVAL;
 201
 202                 goto out;
 203         }
 204
 205         ns_fd2 = lxc_preserve_ns(pid2, ns);
 206         if (ns_fd2 < 0)
 207                 goto out;
 208
 209         ret = fstat(ns_fd1, &ns_st1);
 210         if (ret < 0)
 211                 goto out;
 212
 213         ret = fstat(ns_fd2, &ns_st2);
 214         if (ret < 0)
 215                 goto out;
 216
 217         /* processes are in the same namespace */
 218         ret = -EINVAL;
 219         if ((ns_st1.st_dev == ns_st2.st_dev ) && (ns_st1.st_ino == ns_st2.st_ino))
 220                 goto out;
 221
 222         /* processes are in different namespaces */
 223         ret = ns_fd2;
 224         ns_fd2 = -1;
 225
 226 out:
 227
 228         if (ns_fd1 >= 0)
 229                 close(ns_fd1);
 230         if (ns_fd2 >= 0)
 231                 close(ns_fd2);
 232
 233         return ret;
 234 }
 235
 236 static int lxc_attach_to_ns(pid_t pid, struct lxc_proc_context_info *ctx)
 237 {
 238         int i, ret;
 239
 240         for (i = 0; i < LXC_NS_MAX; i++) {
 241                 if (ctx->ns_fd[i] < 0)
 242                         continue;
 243
 244                 ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag);
 245                 if (ret < 0) {
 246                         SYSERROR("Failed to attach to %s namespace of %d",
 247                                  ns_info[i].proc_name, pid);
 248                         return -1;
 249                 }
 250
 251                 DEBUG("Attached to %s namespace of %d", ns_info[i].proc_name, pid);
 252         }
 253
 254         return 0;
 255 }
 256
 257 static int lxc_attach_remount_sys_proc(void)
 258 {
 259         int ret;
 260
 261         ret = unshare(CLONE_NEWNS);
 262         if (ret < 0) {
 263                 SYSERROR("Failed to unshare mount namespace.");
 264                 return -1;
 265         }
 266
 267         if (detect_shared_rootfs()) {
 268                 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) {
 269                         SYSERROR("Failed to make / rslave.");
 270                         ERROR("Continuing...");
 271                 }
 272         }
 273
 274         /* Assume /proc is always mounted, so remount it. */
 275         ret = umount2("/proc", MNT_DETACH);
 276         if (ret < 0) {
 277                 SYSERROR("Failed to unmount /proc.");
 278                 return -1;
 279         }
 280
 281         ret = mount("none", "/proc", "proc", 0, NULL);
 282         if (ret < 0) {
 283                 SYSERROR("Failed to remount /proc.");
 284                 return -1;
 285         }
 286
 287         /* Try to umount /sys. If it's not a mount point, we'll get EINVAL, then
 288          * we ignore it because it may not have been mounted in the first place.
 289          */
 290         ret = umount2("/sys", MNT_DETACH);
 291         if (ret < 0 && errno != EINVAL) {
 292                 SYSERROR("Failed to unmount /sys.");
 293                 return -1;
 294         } else if (ret == 0) {
 295                 /* Remount it. */
 296                 ret = mount("none", "/sys", "sysfs", 0, NULL);
 297                 if (ret < 0) {
 298                         SYSERROR("Failed to remount /sys.");
 299                         return -1;
 300                 }
 301         }
 302
 303         return 0;
 304 }
 305
 306 static int lxc_attach_drop_privs(struct lxc_proc_context_info *ctx)
 307 {
 308         int cap, last_cap;
 309
 310         last_cap = lxc_caps_last_cap();
 311         for (cap = 0; cap <= last_cap; cap++) {
 312                 if (ctx->capability_mask & (1LL << cap))
 313                         continue;
 314
 315                 if (prctl(PR_CAPBSET_DROP, cap, 0, 0, 0)) {
 316                         SYSERROR("Failed to drop capability %d", cap);
 317                         return -1;
 318                 }
 319                 TRACE("Dropped capability %d", cap);
 320         }
 321
 322         return 0;
 323 }
 324
 325 static int lxc_attach_set_environment(struct lxc_proc_context_info *init_ctx,
 326                                       enum lxc_attach_env_policy_t policy,
 327                                       char **extra_env, char **extra_keep)
 328 {
 329         int ret;
 330         struct lxc_list *iterator;
 331
 332         if (policy == LXC_ATTACH_CLEAR_ENV) {
 333                 int path_kept = 0;
 334                 char **extra_keep_store = NULL;
 335
 336                 if (extra_keep) {
 337                         size_t count, i;
 338
 339                         for (count = 0; extra_keep[count]; count++)
 340                                 ;
 341
 342                         extra_keep_store = calloc(count, sizeof(char *));
 343                         if (!extra_keep_store)
 344                                 return -1;
 345
 346                         for (i = 0; i < count; i++) {
 347                                 char *v = getenv(extra_keep[i]);
 348                                 if (v) {
 349                                         extra_keep_store[i] = strdup(v);
 350                                         if (!extra_keep_store[i]) {
 351                                                 while (i > 0)
 352                                                         free(extra_keep_store[--i]);
 353                                                 free(extra_keep_store);
 354                                                 return -1;
 355                                         }
 356
 357                                         if (strcmp(extra_keep[i], "PATH") == 0)
 358                                                 path_kept = 1;
 359                                 }
 360                         }
 361                 }
 362
 363                 if (clearenv()) {
 364                         if (extra_keep_store) {
 365                                 char **p;
 366
 367                                 for (p = extra_keep_store; *p; p++)
 368                                         free(*p);
 369
 370                                 free(extra_keep_store);
 371                         }
 372
 373                         SYSERROR("Failed to clear environment");
 374                         return -1;
 375                 }
 376
 377                 if (extra_keep_store) {
 378                         size_t i;
 379
 380                         for (i = 0; extra_keep[i]; i++) {
 381                                 if (extra_keep_store[i]) {
 382                                         ret = setenv(extra_keep[i], extra_keep_store[i], 1);
 383                                         if (ret < 0)
 384                                                 WARN("%s - Failed to set environment variable", strerror(errno));
 385                                 }
 386                                 free(extra_keep_store[i]);
 387                         }
 388                         free(extra_keep_store);
 389                 }
 390
 391                 /* Always set a default path; shells and execlp tend to be fine
 392                  * without it, but there is a disturbing number of C programs
 393                  * out there that just assume that getenv("PATH") is never NULL
 394                  * and then die a painful segfault death.
 395                  */
 396                 if (!path_kept) {
 397                         ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1);
 398                         if (ret < 0)
 399                                 WARN("%s - Failed to set environment variable", strerror(errno));
 400                 }
 401         }
 402
 403         ret = putenv("container=lxc");
 404         if (ret < 0) {
 405                 WARN("%s - Failed to set environment variable", strerror(errno));
 406                 return -1;
 407         }
 408
 409         /* Set container environment variables.*/
 410         if (init_ctx && init_ctx->container && init_ctx->container->lxc_conf) {
 411                 lxc_list_for_each(iterator, &init_ctx->container->lxc_conf->environment) {
 412                         char *env_tmp;
 413
 414                         env_tmp = strdup((char *)iterator->elem);
 415                         if (!env_tmp)
 416                                 return -1;
 417
 418                         ret = putenv(env_tmp);
 419                         if (ret < 0) {
 420                                 SYSERROR("Failed to set environment variable: %s", (char *)iterator->elem);
 421                                 return -1;
 422                         }
 423                 }
 424         }
 425
 426         /* Set extra environment variables. */
 427         if (extra_env) {
 428                 for (; *extra_env; extra_env++) {
 429                         char *p;
 430                         /* We just assume the user knows what they are doing, so
 431                          * we don't do any checks.
 432                          */
 433                         p = strdup(*extra_env);
 434                         if (!p)
 435                                 return -1;
 436
 437                         ret = putenv(p);
 438                         if (ret < 0)
 439                                 WARN("%s - Failed to set environment variable", strerror(errno));
 440                 }
 441         }
 442
 443         return 0;
 444 }
 445
 446 static char *lxc_attach_getpwshell(uid_t uid)
 447 {
 448         int fd, ret;
 449         pid_t pid;
 450         int pipes[2];
 451         char *result = NULL;
 452
 453         /* We need to fork off a process that runs the getent program, and we
 454          * need to capture its output, so we use a pipe for that purpose.
 455          */
 456         ret = pipe(pipes);
 457         if (ret < 0)
 458                 return NULL;
 459
 460         pid = fork();
 461         if (pid < 0) {
 462                 close(pipes[0]);
 463                 close(pipes[1]);
 464                 return NULL;
 465         }
 466
 467         if (pid) {
 468                 int status;
 469                 FILE *pipe_f;
 470                 int found = 0;
 471                 size_t line_bufsz = 0;
 472                 char *line = NULL;
 473
 474                 close(pipes[1]);
 475
 476                 pipe_f = fdopen(pipes[0], "r");
 477                 while (getline(&line, &line_bufsz, pipe_f) != -1) {
 478                         int i;
 479                         long value;
 480                         char *token;
 481                         char *endptr = NULL, *saveptr = NULL;
 482
 483                         /* If we already found something, just continue to read
 484                          * until the pipe doesn't deliver any more data, but
 485                          * don't modify the existing data structure.
 486                          */
 487                         if (found)
 488                                 continue;
 489
 490                         /* Trim line on the right hand side. */
 491                         for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i)
 492                                 line[i - 1] = '\0';
 493
 494                         /* Split into tokens: first: user name. */
 495                         token = strtok_r(line, ":", &saveptr);
 496                         if (!token)
 497                                 continue;
 498                         /* next: dummy password field */
 499                         token = strtok_r(NULL, ":", &saveptr);
 500                         if (!token)
 501                                 continue;
 502                         /* next: user id */
 503                         token = strtok_r(NULL, ":", &saveptr);
 504                         value = token ? strtol(token, &endptr, 10) : 0;
 505                         if (!token || !endptr || *endptr || value == LONG_MIN || value == LONG_MAX)
 506                                 continue;
 507                         /* dummy sanity check: user id matches */
 508                         if ((uid_t) value != uid)
 509                                 continue;
 510                         /* skip fields: gid, gecos, dir, go to next field 'shell' */
 511                         for (i = 0; i < 4; i++) {
 512                                 token = strtok_r(NULL, ":", &saveptr);
 513                                 if (!token)
 514                                         break;
 515                         }
 516                         if (!token)
 517                                 continue;
 518                         free(result);
 519                         result = strdup(token);
 520
 521                         /* Sanity check that there are no fields after that. */
 522                         token = strtok_r(NULL, ":", &saveptr);
 523                         if (token)
 524                                 continue;
 525
 526                         found = 1;
 527                 }
 528
 529                 free(line);
 530                 fclose(pipe_f);
 531         again:
 532                 if (waitpid(pid, &status, 0) < 0) {
 533                         if (errno == EINTR)
 534                                 goto again;
 535                         free(result);
 536                         return NULL;
 537                 }
 538
 539                 /* Some sanity checks. If anything even hinted at going wrong,
 540                  * we can't be sure we have a valid result, so we assume we
 541                  * don't.
 542                  */
 543
 544                 if (!WIFEXITED(status)) {
 545                         free(result);
 546                         return NULL;
 547                 }
 548
 549                 if (WEXITSTATUS(status) != 0) {
 550                         free(result);
 551                         return NULL;
 552                 }
 553
 554                 if (!found) {
 555                         free(result);
 556                         return NULL;
 557                 }
 558
 559                 return result;
 560         } else {
 561                 char uid_buf[32];
 562                 char *arguments[] = {
 563                         "getent",
 564                         "passwd",
 565                         uid_buf,
 566                         NULL
 567                 };
 568
 569                 close(pipes[0]);
 570
 571                 /* We want to capture stdout. */
 572                 dup2(pipes[1], 1);
 573                 close(pipes[1]);
 574
 575                 /* Get rid of stdin/stderr, so we try to associate it with
 576                  * /dev/null.
 577                  */
 578                 fd = open("/dev/null", O_RDWR);
 579                 if (fd < 0) {
 580                         close(0);
 581                         close(2);
 582                 } else {
 583                         dup2(fd, 0);
 584                         dup2(fd, 2);
 585                         close(fd);
 586                 }
 587
 588                 /* Finish argument list. */
 589                 ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long) uid);
 590                 if (ret <= 0)
 591                         exit(-1);
 592
 593                 /* Try to run getent program. */
 594                 (void) execvp("getent", arguments);
 595                 exit(-1);
 596         }
 597 }
 598
 599 static void lxc_attach_get_init_uidgid(uid_t *init_uid, gid_t *init_gid)
 600 {
 601         FILE *proc_file;
 602         char proc_fn[__PROC_STATUS_LEN];
 603         int ret;
 604         char *line = NULL;
 605         size_t line_bufsz = 0;
 606         long value = -1;
 607         uid_t uid = (uid_t)-1;
 608         gid_t gid = (gid_t)-1;
 609
 610         /* Read capabilities. */
 611         snprintf(proc_fn, __PROC_STATUS_LEN, "/proc/%d/status", 1);
 612
 613         proc_file = fopen(proc_fn, "r");
 614         if (!proc_file)
 615                 return;
 616
 617         while (getline(&line, &line_bufsz, proc_file) != -1) {
 618                 /* Format is: real, effective, saved set user, fs we only care
 619                  * about real uid.
 620                  */
 621                 ret = sscanf(line, "Uid: %ld", &value);
 622                 if (ret != EOF && ret == 1) {
 623                         uid = (uid_t)value;
 624                 } else {
 625                         ret = sscanf(line, "Gid: %ld", &value);
 626                         if (ret != EOF && ret == 1)
 627                                 gid = (gid_t)value;
 628                 }
 629                 if (uid != (uid_t)-1 && gid != (gid_t)-1)
 630                         break;
 631         }
 632
 633         fclose(proc_file);
 634         free(line);
 635
 636         /* Only override arguments if we found something. */
 637         if (uid != (uid_t)-1)
 638                 *init_uid = uid;
 639         if (gid != (gid_t)-1)
 640                 *init_gid = gid;
 641
 642         /* TODO: we should also parse supplementary groups and use
 643          * setgroups() to set them.
 644          */
 645 }
 646
 647 /* Help the optimizer along if it doesn't know that exit always exits. */
 648 #define rexit(c)                                                               \
 649         do {                                                                   \
 650                 int __c = (c);                                                 \
 651                 _exit(__c);                                                    \
 652                 return __c;                                                    \
 653         } while (0)
 654
 655 /* Define default options if no options are supplied by the user. */
 656 static lxc_attach_options_t attach_static_default_options = LXC_ATTACH_OPTIONS_DEFAULT;
 657
 658 static bool fetch_seccomp(struct lxc_container *c,
 659                           lxc_attach_options_t *options)
 660 {
 661         char *path;
 662
 663         if (!(options->namespaces & CLONE_NEWNS) ||
 664             !(options->attach_flags & LXC_ATTACH_LSM)) {
 665                 free(c->lxc_conf->seccomp);
 666                 c->lxc_conf->seccomp = NULL;
 667                 return true;
 668         }
 669
 670         /* Remove current setting. */
 671         if (!c->set_config_item(c, "lxc.seccomp", "") &&
 672             !c->set_config_item(c, "lxc.seccomp.profile", "")) {
 673                 return false;
 674         }
 675
 676         /* Fetch the current profile path over the cmd interface. */
 677         path = c->get_running_config_item(c, "lxc.seccomp.profile");
 678         if (!path) {
 679                 INFO("Failed to get running config item for lxc.seccomp.profile");
 680                 path = c->get_running_config_item(c, "lxc.seccomp");
 681         }
 682         if (!path) {
 683                 INFO("Failed to get running config item for lxc.seccomp");
 684                 return true;
 685         }
 686
 687         /* Copy the value into the new lxc_conf. */
 688         if (!c->set_config_item(c, "lxc.seccomp.profile", path)) {
 689                 free(path);
 690                 return false;
 691         }
 692         free(path);
 693
 694         /* Attempt to parse the resulting config. */
 695         if (lxc_read_seccomp_config(c->lxc_conf) < 0) {
 696                 ERROR("Error reading seccomp policy.");
 697                 return false;
 698         }
 699
 700         INFO("Retrieved seccomp policy.");
 701         return true;
 702 }
 703
 704 static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options)
 705 {
 706         char *val;
 707
 708         /* Remove current setting. */
 709         if (!c->set_config_item(c, "lxc.no_new_privs", ""))
 710                 return false;
 711
 712         /* Retrieve currently active setting. */
 713         val = c->get_running_config_item(c, "lxc.no_new_privs");
 714         if (!val) {
 715                 INFO("Failed to get running config item for lxc.no_new_privs.");
 716                 return false;
 717         }
 718
 719         /* Set currently active setting. */
 720         if (!c->set_config_item(c, "lxc.no_new_privs", val)) {
 721                 free(val);
 722                 return false;
 723         }
 724         free(val);
 725
 726         return true;
 727 }
 728
 729 static signed long get_personality(const char *name, const char *lxcpath)
 730 {
 731         char *p;
 732         signed long ret;
 733
 734         p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath);
 735         if (!p)
 736                 return -1;
 737
 738         ret = lxc_config_parse_arch(p);
 739         free(p);
 740
 741         return ret;
 742 }
 743
 744 struct attach_clone_payload {
 745         int ipc_socket;
 746         int terminal_slave_fd;
 747         lxc_attach_options_t *options;
 748         struct lxc_proc_context_info *init_ctx;
 749         lxc_attach_exec_t exec_function;
 750         void *exec_payload;
 751 };
 752
 753 static void lxc_put_attach_clone_payload(struct attach_clone_payload *p)
 754 {
 755         if (p->ipc_socket >= 0) {
 756                 shutdown(p->ipc_socket, SHUT_RDWR);
 757                 close(p->ipc_socket);
 758                 p->ipc_socket = -EBADF;
 759         }
 760
 761         if (p->terminal_slave_fd >= 0) {
 762                 close(p->terminal_slave_fd);
 763                 p->terminal_slave_fd = -EBADF;
 764         }
 765
 766         if (p->init_ctx) {
 767                 lxc_proc_put_context_info(p->init_ctx);
 768                 p->init_ctx = NULL;
 769         }
 770 }
 771
 772 static int attach_child_main(struct attach_clone_payload *payload)
 773 {
 774         int fd, lsm_fd, ret;
 775         uid_t new_uid;
 776         gid_t new_gid;
 777         lxc_attach_options_t* options = payload->options;
 778         struct lxc_proc_context_info* init_ctx = payload->init_ctx;
 779         bool needs_lsm = (options->namespaces & CLONE_NEWNS) &&
 780                          (options->attach_flags & LXC_ATTACH_LSM) &&
 781                          init_ctx->lsm_label;
 782
 783         /* A description of the purpose of this functionality is provided in the
 784          * lxc-attach(1) manual page. We have to remount here and not in the
 785          * parent process, otherwise /proc may not properly reflect the new pid
 786          * namespace.
 787          */
 788         if (!(options->namespaces & CLONE_NEWNS) &&
 789             (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) {
 790                 ret = lxc_attach_remount_sys_proc();
 791                 if (ret < 0)
 792                         goto on_error;
 793                 TRACE("Remounted \"/proc\" and \"/sys\"");
 794         }
 795
 796 /* Now perform additional attachments. */
 797 #if HAVE_SYS_PERSONALITY_H
 798         if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) {
 799                 long new_personality;
 800
 801                 if (options->personality < 0)
 802                         new_personality = init_ctx->personality;
 803                 else
 804                         new_personality = options->personality;
 805                 ret = personality(new_personality);
 806                 if (ret < 0)
 807                         goto on_error;
 808                 TRACE("Set new personality");
 809         }
 810 #endif
 811
 812         if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) {
 813                 ret = lxc_attach_drop_privs(init_ctx);
 814                 if (ret < 0)
 815                         goto on_error;
 816                 TRACE("Dropped capabilities");
 817         }
 818
 819         /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL)
 820          * if you want this to be a no-op).
 821          */
 822         ret = lxc_attach_set_environment(init_ctx,
 823                                          options->env_policy,
 824                                          options->extra_env_vars,
 825                                          options->extra_keep_env);
 826         if (ret < 0)
 827                 goto on_error;
 828         TRACE("Set up environment");
 829
 830         /* This remark only affects fully unprivileged containers:
 831          * Receive fd for LSM security module before we set{g,u}id(). The reason
 832          * is that on set{g,u}id() the kernel will a) make us undumpable and b)
 833          * we will change our effective uid. This means our effective uid will
 834          * be different from the effective uid of the process that created us
 835          * which means that this processs no longer has capabilities in our
 836          * namespace including CAP_SYS_PTRACE. This means we will not be able to
 837          * read and /proc/<pid> files for the process anymore when /proc is
 838          * mounted with hidepid={1,2}. So let's get the lsm label fd before the
 839          * set{g,u}id().
 840          */
 841         if (needs_lsm) {
 842                 ret = lxc_abstract_unix_recv_fds(payload->ipc_socket, &lsm_fd, 1, NULL, 0);
 843                 if (ret <= 0)
 844                         goto on_error;
 845                 TRACE("Received LSM label file descriptor %d from parent", lsm_fd);
 846         }
 847
 848         if (options->stdin_fd > 0 && isatty(options->stdin_fd)) {
 849                 ret = lxc_make_controlling_terminal(options->stdin_fd);
 850                 if (ret < 0)
 851                         goto on_error;
 852         }
 853
 854         /* Set {u,g}id. */
 855         new_uid = 0;
 856         new_gid = 0;
 857         /* Ignore errors, we will fall back to root in that case (/proc was not
 858          * mounted etc.).
 859          */
 860         if (options->namespaces & CLONE_NEWUSER)
 861                 lxc_attach_get_init_uidgid(&new_uid, &new_gid);
 862
 863         if (options->uid != (uid_t)-1)
 864                 new_uid = options->uid;
 865         if (options->gid != (gid_t)-1)
 866                 new_gid = options->gid;
 867
 868         /* Try to set the {u,g}id combination. */
 869         if (new_uid != 0 || new_gid != 0 || options->namespaces & CLONE_NEWUSER) {
 870                 ret = lxc_switch_uid_gid(new_uid, new_gid);
 871                 if (ret < 0)
 872                         goto on_error;
 873         }
 874
 875         ret = lxc_setgroups(0, NULL);
 876         if (ret < 0)
 877                 goto on_error;
 878
 879         if ((init_ctx->container && init_ctx->container->lxc_conf &&
 880              init_ctx->container->lxc_conf->no_new_privs) ||
 881             (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) {
 882                 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 883                 if (ret < 0)
 884                         goto on_error;
 885                 TRACE("Set PR_SET_NO_NEW_PRIVS");
 886         }
 887
 888         if (needs_lsm) {
 889                 bool on_exec;
 890
 891                 /* Change into our new LSM profile. */
 892                 on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false;
 893                 ret = lsm_process_label_set_at(lsm_fd, init_ctx->lsm_label, on_exec);
 894                 close(lsm_fd);
 895                 if (ret < 0)
 896                         goto on_error;
 897                 TRACE("Set %s LSM label to \"%s\"", lsm_name(), init_ctx->lsm_label);
 898         }
 899
 900         if (init_ctx->container && init_ctx->container->lxc_conf &&
 901             init_ctx->container->lxc_conf->seccomp) {
 902                 ret = lxc_seccomp_load(init_ctx->container->lxc_conf);
 903                 if (ret < 0)
 904                         goto on_error;
 905                 TRACE("Loaded seccomp profile");
 906         }
 907         shutdown(payload->ipc_socket, SHUT_RDWR);
 908         close(payload->ipc_socket);
 909         payload->ipc_socket = -EBADF;
 910         lxc_proc_put_context_info(init_ctx);
 911
 912         /* The following is done after the communication socket is shut down.
 913          * That way, all errors that might (though unlikely) occur up until this
 914          * point will have their messages printed to the original stderr (if
 915          * logging is so configured) and not the fd the user supplied, if any.
 916          */
 917
 918         /* Fd handling for stdin, stdout and stderr; ignore errors here, user
 919          * may want to make sure the fds are closed, for example.
 920          */
 921         if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO)
 922                 dup2(options->stdin_fd, STDIN_FILENO);
 923
 924         if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO)
 925                 dup2(options->stdout_fd, STDOUT_FILENO);
 926
 927         if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO)
 928                 dup2(options->stderr_fd, STDERR_FILENO);
 929
 930         /* close the old fds */
 931         if (options->stdin_fd > STDERR_FILENO)
 932                 close(options->stdin_fd);
 933
 934         if (options->stdout_fd > STDERR_FILENO)
 935                 close(options->stdout_fd);
 936
 937         if (options->stderr_fd > STDERR_FILENO)
 938                 close(options->stderr_fd);
 939
 940         /* Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also
 941          * here, ignore errors.
 942          */
 943         for (fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) {
 944                 int flags;
 945
 946                 flags = fcntl(fd, F_GETFL);
 947                 if (flags < 0)
 948                         continue;
 949
 950                 if ((flags & FD_CLOEXEC) == 0)
 951                         continue;
 952
 953                 ret = fcntl(fd, F_SETFL, flags & ~FD_CLOEXEC);
 954                 if (ret < 0) {
 955                         SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd);
 956                         goto on_error;
 957                 }
 958         }
 959
 960         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
 961                 ret = lxc_terminal_prepare_login(payload->terminal_slave_fd);
 962                 if (ret < 0) {
 963                         SYSERROR("Failed to prepare terminal file descriptor %d", payload->terminal_slave_fd);
 964                         goto on_error;
 965                 }
 966                 TRACE("Prepared terminal file descriptor %d", payload->terminal_slave_fd);
 967         }
 968
 969         /* We're done, so we can now do whatever the user intended us to do. */
 970         rexit(payload->exec_function(payload->exec_payload));
 971
 972 on_error:
 973         lxc_put_attach_clone_payload(payload);
 974         rexit(EXIT_FAILURE);
 975 }
 976
 977 static int lxc_attach_terminal(struct lxc_conf *conf,
 978                                struct lxc_terminal *terminal)
 979 {
 980         int ret;
 981
 982         lxc_terminal_init(terminal);
 983
 984         ret = lxc_terminal_create(terminal);
 985         if (ret < 0) {
 986                 SYSERROR("Failed to create terminal");
 987                 return -1;
 988         }
 989
 990         /* Shift ttys to container. */
 991         ret = lxc_terminal_map_ids(conf, terminal);
 992         if (ret < 0) {
 993                 ERROR("Failed to chown terminal");
 994                 goto on_error;
 995         }
 996
 997         return 0;
 998
 999 on_error:
1000         lxc_terminal_delete(terminal);
1001         lxc_terminal_conf_free(terminal);
1002         return -1;
1003 }
1004
1005 static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal,
1006                                              struct lxc_epoll_descr *descr)
1007 {
1008         int ret;
1009
1010         ret = lxc_mainloop_open(descr);
1011         if (ret < 0) {
1012                 ERROR("Failed to create mainloop");
1013                 return -1;
1014         }
1015
1016         ret = lxc_terminal_mainloop_add(descr, terminal);
1017         if (ret < 0) {
1018                 ERROR("Failed to add handlers to mainloop");
1019                 lxc_mainloop_close(descr);
1020                 return -1;
1021         }
1022
1023         return 0;
1024 }
1025
1026 static inline void lxc_attach_terminal_close_master(struct lxc_terminal *terminal)
1027 {
1028         if (terminal->master < 0)
1029                 return;
1030
1031         close(terminal->master);
1032         terminal->master = -EBADF;
1033 }
1034
1035 static inline void lxc_attach_terminal_close_slave(struct lxc_terminal *terminal)
1036 {
1037         if (terminal->slave < 0)
1038                 return;
1039
1040         close(terminal->slave);
1041         terminal->slave = -EBADF;
1042 }
1043
1044 static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal)
1045 {
1046         if (terminal->peer < 0)
1047                 return;
1048
1049         close(terminal->peer);
1050         terminal->peer = -EBADF;
1051 }
1052
1053 static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal)
1054 {
1055         if (terminal->log_fd < 0)
1056                 return;
1057
1058         close(terminal->log_fd);
1059         terminal->log_fd = -EBADF;
1060 }
1061
1062 int lxc_attach(const char *name, const char *lxcpath,
1063                lxc_attach_exec_t exec_function, void *exec_payload,
1064                lxc_attach_options_t *options, pid_t *attached_process)
1065 {
1066         int i, ret, status;
1067         int ipc_sockets[2];
1068         char *cwd, *new_cwd;
1069         signed long personality;
1070         pid_t attached_pid, init_pid, pid;
1071         struct lxc_proc_context_info *init_ctx;
1072         struct lxc_terminal terminal;
1073         struct lxc_conf *conf;
1074         struct attach_clone_payload payload = {0};
1075
1076         ret = access("/proc/self/ns", X_OK);
1077         if (ret) {
1078                 ERROR("Does this kernel version support namespaces?");
1079                 return -1;
1080         }
1081
1082         if (!options)
1083                 options = &attach_static_default_options;
1084
1085         init_pid = lxc_cmd_get_init_pid(name, lxcpath);
1086         if (init_pid < 0) {
1087                 ERROR("Failed to get init pid.");
1088                 return -1;
1089         }
1090
1091         init_ctx = lxc_proc_get_context_info(init_pid);
1092         if (!init_ctx) {
1093                 ERROR("Failed to get context of init process: %ld", (long)init_pid);
1094                 return -1;
1095         }
1096
1097         personality = get_personality(name, lxcpath);
1098         if (init_ctx->personality < 0) {
1099                 ERROR("Failed to get personality of the container");
1100                 lxc_proc_put_context_info(init_ctx);
1101                 return -1;
1102         }
1103         init_ctx->personality = personality;
1104
1105         init_ctx->container = lxc_container_new(name, lxcpath);
1106         if (!init_ctx->container) {
1107                 lxc_proc_put_context_info(init_ctx);
1108                 return -1;
1109         }
1110
1111         if (!init_ctx->container->lxc_conf) {
1112                 init_ctx->container->lxc_conf = lxc_conf_init();
1113                 if (!init_ctx->container->lxc_conf) {
1114                         lxc_proc_put_context_info(init_ctx);
1115                         return -ENOMEM;
1116                 }
1117         }
1118         conf = init_ctx->container->lxc_conf;
1119
1120         if (!fetch_seccomp(init_ctx->container, options))
1121                 WARN("Failed to get seccomp policy.");
1122
1123         if (!no_new_privs(init_ctx->container, options))
1124                 WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set.");
1125
1126         cwd = getcwd(NULL, 0);
1127
1128         /* Determine which namespaces the container was created with
1129          * by asking lxc-start, if necessary.
1130          */
1131         if (options->namespaces == -1) {
1132                 options->namespaces = lxc_cmd_get_clone_flags(name, lxcpath);
1133                 /* call failed */
1134                 if (options->namespaces == -1) {
1135                         ERROR("Failed to automatically determine the "
1136                               "namespaces which the container uses");
1137                         free(cwd);
1138                         lxc_proc_put_context_info(init_ctx);
1139                         return -1;
1140                 }
1141
1142                 for (i = 0; i < LXC_NS_MAX; i++) {
1143                         if (ns_info[i].clone_flag & CLONE_NEWCGROUP)
1144                                 if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) ||
1145                                     !cgns_supported())
1146                                         continue;
1147
1148                         if (ns_info[i].clone_flag & options->namespaces)
1149                                 continue;
1150
1151                         init_ctx->ns_inherited |= ns_info[i].clone_flag;
1152                 }
1153         }
1154
1155         pid = lxc_raw_getpid();
1156         for (i = 0; i < LXC_NS_MAX; i++) {
1157                 int j, saved_errno;
1158
1159                 if (options->namespaces & ns_info[i].clone_flag)
1160                         init_ctx->ns_fd[i] = lxc_preserve_ns(init_pid, ns_info[i].proc_name);
1161                 else if (init_ctx->ns_inherited & ns_info[i].clone_flag)
1162                         init_ctx->ns_fd[i] = in_same_namespace(pid, init_pid, ns_info[i].proc_name);
1163                 else
1164                         continue;
1165                 if (init_ctx->ns_fd[i] >= 0)
1166                         continue;
1167
1168                 if (init_ctx->ns_fd[i] == -EINVAL) {
1169                         DEBUG("Inheriting %s namespace from %d",
1170                               ns_info[i].proc_name, pid);
1171                         init_ctx->ns_inherited &= ~ns_info[i].clone_flag;
1172                         continue;
1173                 }
1174
1175                 /* We failed to preserve the namespace. */
1176                 saved_errno = errno;
1177                 /* Close all already opened file descriptors before we return an
1178                  * error, so we don't leak them.
1179                  */
1180                 for (j = 0; j < i; j++)
1181                         close(init_ctx->ns_fd[j]);
1182
1183                 errno = saved_errno;
1184                 SYSERROR("Failed to attach to %s namespace of %d",
1185                          ns_info[i].proc_name, pid);
1186                 free(cwd);
1187                 lxc_proc_put_context_info(init_ctx);
1188                 return -1;
1189         }
1190
1191         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1192                 ret = lxc_attach_terminal(conf, &terminal);
1193                 if (ret < 0) {
1194                         ERROR("Failed to setup new terminal");
1195                         free(cwd);
1196                         lxc_proc_put_context_info(init_ctx);
1197                         return -1;
1198                 }
1199
1200                 terminal.log_fd = options->log_fd;
1201         } else {
1202                 lxc_terminal_init(&terminal);
1203         }
1204
1205         /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order
1206          * to make sure we don't irritate other threads that want to fork+exec
1207          * away
1208          *
1209          * IMPORTANT: if the initial process is multithreaded and another call
1210          * just fork()s away without exec'ing directly after, the socket fd will
1211          * exist in the forked process from the other thread and any close() in
1212          * our own child process will not really cause the socket to close
1213          * properly, potentiall causing the parent to hang.
1214          *
1215          * For this reason, while IPC is still active, we have to use shutdown()
1216          * if the child exits prematurely in order to signal that the socket is
1217          * closed and cannot assume that the child exiting will automatically do
1218          * that.
1219          *
1220          * IPC mechanism: (X is receiver)
1221          *   initial process        intermediate          attached
1222          *        X           <---  send pid of
1223          *                          attached proc,
1224          *                          then exit
1225          *    send 0 ------------------------------------>    X
1226          *                                              [do initialization]
1227          *        X  <------------------------------------  send 1
1228          *   [add to cgroup, ...]
1229          *    send 2 ------------------------------------>    X
1230          *                                              [set LXC_ATTACH_NO_NEW_PRIVS]
1231          *        X  <------------------------------------  send 3
1232          *   [open LSM label fd]
1233          *    send 4 ------------------------------------>    X
1234          *                                              [set LSM label]
1235          *   close socket                                 close socket
1236          *                                                run program
1237          */
1238         ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets);
1239         if (ret < 0) {
1240                 SYSERROR("Could not set up required IPC mechanism for attaching.");
1241                 free(cwd);
1242                 lxc_proc_put_context_info(init_ctx);
1243                 return -1;
1244         }
1245
1246         /* Create intermediate subprocess, two reasons:
1247          *       1. We can't setns() in the child itself, since we want to make
1248          *          sure we are properly attached to the pidns.
1249          *       2. Also, the initial thread has to put the attached process
1250          *          into the cgroup, which we can only do if we didn't already
1251          *          setns() (otherwise, user namespaces will hate us).
1252          */
1253         pid = fork();
1254         if (pid < 0) {
1255                 SYSERROR("Failed to create first subprocess.");
1256                 free(cwd);
1257                 lxc_proc_put_context_info(init_ctx);
1258                 return -1;
1259         }
1260
1261         if (pid) {
1262                 int ret_parent = -1;
1263                 pid_t to_cleanup_pid = pid;
1264                 struct lxc_epoll_descr descr = {0};
1265
1266                 /* close unneeded file descriptors */
1267                 close(ipc_sockets[1]);
1268                 free(cwd);
1269                 lxc_proc_close_ns_fd(init_ctx);
1270                 if (options->attach_flags & LXC_ATTACH_TERMINAL)
1271                         lxc_attach_terminal_close_slave(&terminal);
1272
1273                 /* Attach to cgroup, if requested. */
1274                 if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) {
1275                         if (!cgroup_attach(name, lxcpath, pid))
1276                                 goto on_error;
1277                         TRACE("Moved intermediate process %d into container's "
1278                               "cgroups", pid);
1279                 }
1280
1281                 /* Setup /proc limits */
1282                 if (!lxc_list_empty(&conf->procs)) {
1283                         ret = setup_proc_filesystem(&conf->procs, pid);
1284                         if (ret < 0)
1285                                 goto on_error;
1286                 }
1287
1288                 /* Setup resource limits */
1289                 if (!lxc_list_empty(&conf->limits)) {
1290                         ret = setup_resource_limits(&conf->limits, pid);
1291                         if (ret < 0)
1292                                 goto on_error;
1293                 }
1294
1295                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1296                         ret = lxc_attach_terminal_mainloop_init(&terminal, &descr);
1297                         if (ret < 0)
1298                                 goto on_error;
1299                         TRACE("Initialized terminal mainloop");
1300                 }
1301
1302                 /* Let the child process know to go ahead. */
1303                 status = 0;
1304                 ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status));
1305                 if (ret != sizeof(status))
1306                         goto close_mainloop;
1307                 TRACE("Told intermediate process to start initializing");
1308
1309                 /* Get pid of attached process from intermediate process. */
1310                 ret = lxc_read_nointr(ipc_sockets[0], &attached_pid, sizeof(attached_pid));
1311                 if (ret != sizeof(attached_pid))
1312                         goto close_mainloop;
1313                 TRACE("Received pid %d of attached process in parent pid namespace", attached_pid);
1314
1315                 /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */
1316                 if (options->stdin_fd == 0) {
1317                         signal(SIGINT, SIG_IGN);
1318                         signal(SIGQUIT, SIG_IGN);
1319                 }
1320
1321                 /* Reap intermediate process. */
1322                 ret = wait_for_pid(pid);
1323                 if (ret < 0)
1324                         goto close_mainloop;
1325                 TRACE("Intermediate process %d exited", pid);
1326
1327                 /* We will always have to reap the attached process now. */
1328                 to_cleanup_pid = attached_pid;
1329
1330                 /* Open LSM fd and send it to child. */
1331                 if ((options->namespaces & CLONE_NEWNS) &&
1332                     (options->attach_flags & LXC_ATTACH_LSM) &&
1333                     init_ctx->lsm_label) {
1334                         int ret = -1;
1335                         int labelfd;
1336                         bool on_exec;
1337
1338                         on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false;
1339                         labelfd = lsm_process_label_fd_get(attached_pid, on_exec);
1340                         if (labelfd < 0)
1341                                 goto close_mainloop;
1342                         TRACE("Opened LSM label file descriptor %d", labelfd);
1343
1344                         /* Send child fd of the LSM security module to write to. */
1345                         ret = lxc_abstract_unix_send_fds(ipc_sockets[0], &labelfd, 1, NULL, 0);
1346                         close(labelfd);
1347                         if (ret <= 0) {
1348                                 SYSERROR("%d", (int)ret);
1349                                 goto close_mainloop;
1350                         }
1351                         TRACE("Sent LSM label file descriptor %d to child", labelfd);
1352                 }
1353
1354                 /* We're done, the child process should now execute whatever it
1355                  * is that the user requested. The parent can now track it with
1356                  * waitpid() or similar.
1357                  */
1358
1359                 *attached_process = attached_pid;
1360
1361                 /* Now shut down communication with child, we're done. */
1362                 shutdown(ipc_sockets[0], SHUT_RDWR);
1363                 close(ipc_sockets[0]);
1364                 ipc_sockets[0] = -1;
1365
1366                 ret_parent = 0;
1367                 to_cleanup_pid = -1;
1368                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1369                         ret = lxc_mainloop(&descr, -1);
1370                         if (ret < 0) {
1371                                 ret_parent = -1;
1372                                 to_cleanup_pid = attached_pid;
1373                         }
1374                 }
1375
1376         close_mainloop:
1377                 if (options->attach_flags & LXC_ATTACH_TERMINAL)
1378                         lxc_mainloop_close(&descr);
1379
1380         on_error:
1381                 if (ipc_sockets[0] >= 0) {
1382                         shutdown(ipc_sockets[0], SHUT_RDWR);
1383                         close(ipc_sockets[0]);
1384                 }
1385
1386                 if (to_cleanup_pid > 0)
1387                         (void)wait_for_pid(to_cleanup_pid);
1388
1389                 if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1390                         lxc_terminal_delete(&terminal);
1391                         lxc_terminal_conf_free(&terminal);
1392                 }
1393                 lxc_proc_put_context_info(init_ctx);
1394                 return ret_parent;
1395         }
1396
1397         /* close unneeded file descriptors */
1398         close(ipc_sockets[0]);
1399         ipc_sockets[0] = -EBADF;
1400         if (options->attach_flags & LXC_ATTACH_TERMINAL) {
1401                 lxc_attach_terminal_close_master(&terminal);
1402                 lxc_attach_terminal_close_peer(&terminal);
1403                 lxc_attach_terminal_close_log(&terminal);
1404         }
1405
1406         /* Wait for the parent to have setup cgroups. */
1407         ret = lxc_read_nointr(ipc_sockets[1], &status, sizeof(status));
1408         if (ret != sizeof(status)) {
1409                 shutdown(ipc_sockets[1], SHUT_RDWR);
1410                 lxc_proc_put_context_info(init_ctx);
1411                 rexit(-1);
1412         }
1413         TRACE("Intermediate process starting to initialize");
1414
1415         /* Attach now, create another subprocess later, since pid namespaces
1416          * only really affect the children of the current process.
1417          */
1418         ret = lxc_attach_to_ns(init_pid, init_ctx);
1419         if (ret < 0) {
1420                 ERROR("Failed to enter namespaces");
1421                 shutdown(ipc_sockets[1], SHUT_RDWR);
1422                 lxc_proc_put_context_info(init_ctx);
1423                 rexit(-1);
1424         }
1425         /* close namespace file descriptors */
1426         lxc_proc_close_ns_fd(init_ctx);
1427
1428         /* Attach succeeded, try to cwd. */
1429         if (options->initial_cwd)
1430                 new_cwd = options->initial_cwd;
1431         else
1432                 new_cwd = cwd;
1433         if (new_cwd) {
1434                 ret = chdir(new_cwd);
1435                 if (ret < 0)
1436                         WARN("Could not change directory to \"%s\"", new_cwd);
1437         }
1438         free(cwd);
1439
1440         /* Create attached process. */
1441         payload.ipc_socket = ipc_sockets[1];
1442         payload.options = options;
1443         payload.init_ctx = init_ctx;
1444         payload.terminal_slave_fd = terminal.slave;
1445         payload.exec_function = exec_function;
1446         payload.exec_payload = exec_payload;
1447
1448         pid = lxc_raw_clone(CLONE_PARENT);
1449         if (pid < 0) {
1450                 SYSERROR("Failed to clone attached process");
1451                 shutdown(ipc_sockets[1], SHUT_RDWR);
1452                 lxc_proc_put_context_info(init_ctx);
1453                 rexit(-1);
1454         }
1455
1456         if (pid == 0) {
1457                 ret = attach_child_main(&payload);
1458                 if (ret < 0)
1459                         ERROR("Failed to exec");
1460                 _exit(EXIT_FAILURE);
1461         }
1462         if (options->attach_flags & LXC_ATTACH_TERMINAL)
1463                 lxc_attach_terminal_close_slave(&terminal);
1464
1465         /* Tell grandparent the pid of the pid of the newly created child. */
1466         ret = lxc_write_nointr(ipc_sockets[1], &pid, sizeof(pid));
1467         if (ret != sizeof(pid)) {
1468                 /* If this really happens here, this is very unfortunate, since
1469                  * the parent will not know the pid of the attached process and
1470                  * will not be able to wait for it (and we won't either due to
1471                  * CLONE_PARENT) so the parent won't be able to reap it and the
1472                  * attached process will remain a zombie.
1473                  */
1474                 shutdown(ipc_sockets[1], SHUT_RDWR);
1475                 lxc_proc_put_context_info(init_ctx);
1476                 rexit(-1);
1477         }
1478         TRACE("Sending pid %d of attached process", pid);
1479
1480         /* The rest is in the hands of the initial and the attached process. */
1481         lxc_proc_put_context_info(init_ctx);
1482         rexit(0);
1483 }
1484
1485 int lxc_attach_run_command(void* payload)
1486 {
1487         lxc_attach_command_t* cmd = (lxc_attach_command_t*)payload;
1488
1489         execvp(cmd->program, cmd->argv);
1490         SYSERROR("Failed to exec \"%s\".", cmd->program);
1491         return -1;
1492 }
1493
1494 int lxc_attach_run_shell(void* payload)
1495 {
1496         uid_t uid;
1497         struct passwd *passwd;
1498         char *user_shell;
1499
1500         /* Ignore payload parameter. */
1501         (void)payload;
1502
1503         uid = getuid();
1504         passwd = getpwuid(uid);
1505
1506         /* This probably happens because of incompatible nss implementations in
1507          * host and container (remember, this code is still using the host's
1508          * glibc but our mount namespace is in the container) we may try to get
1509          * the information by spawning a [getent passwd uid] process and parsing
1510          * the result.
1511          */
1512         if (!passwd)
1513                 user_shell = lxc_attach_getpwshell(uid);
1514         else
1515                 user_shell = passwd->pw_shell;
1516         if (user_shell)
1517                 execlp(user_shell, user_shell, (char *)NULL);
1518
1519         /* Executed if either no passwd entry or execvp fails, we will fall back
1520          * on /bin/sh as a default shell.
1521          */
1522         execlp("/bin/sh", "/bin/sh", (char *)NULL);
1523         SYSERROR("Failed to execute shell");
1524         if (!passwd)
1525                 free(user_shell);
1526         return -1;
1527 }