src/lxc/utils.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 #include "config.h"
  25
  26 #define __STDC_FORMAT_MACROS /* Required for PRIu64 to work. */
  27 #include <ctype.h>
  28 #include <dirent.h>
  29 #include <errno.h>
  30 #include <fcntl.h>
  31 #include <grp.h>
  32 #include <inttypes.h>
  33 #include <libgen.h>
  34 #include <stddef.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <string.h>
  38 #include <unistd.h>
  39 #include <sys/mman.h>
  40 #include <sys/mount.h>
  41 #include <sys/param.h>
  42 #include <sys/prctl.h>
  43 #include <sys/stat.h>
  44 #include <sys/types.h>
  45 #include <sys/wait.h>
  46
  47 #include "log.h"
  48 #include "lxclock.h"
  49 #include "namespace.h"
  50 #include "utils.h"
  51
  52 #ifndef PR_SET_MM
  53 #define PR_SET_MM 35
  54 #endif
  55
  56 #ifndef PR_SET_MM_MAP
  57 #define PR_SET_MM_MAP 14
  58
  59 struct prctl_mm_map {
  60         uint64_t   start_code;
  61         uint64_t   end_code;
  62         uint64_t   start_data;
  63         uint64_t   end_data;
  64         uint64_t   start_brk;
  65         uint64_t   brk;
  66         uint64_t   start_stack;
  67         uint64_t   arg_start;
  68         uint64_t   arg_end;
  69         uint64_t   env_start;
  70         uint64_t   env_end;
  71         uint64_t   *auxv;
  72         uint32_t   auxv_size;
  73         uint32_t   exe_fd;
  74 };
  75 #endif
  76
  77 #ifndef O_PATH
  78 #define O_PATH      010000000
  79 #endif
  80
  81 #ifndef O_NOFOLLOW
  82 #define O_NOFOLLOW  00400000
  83 #endif
  84
  85 lxc_log_define(lxc_utils, lxc);
  86
  87 /*
  88  * if path is btrfs, tries to remove it and any subvolumes beneath it
  89  */
  90 extern bool btrfs_try_remove_subvol(const char *path);
  91
  92 static int _recursive_rmdir(char *dirname, dev_t pdev,
  93                             const char *exclude, int level, bool onedev)
  94 {
  95         struct dirent *direntp;
  96         DIR *dir;
  97         int ret, failed=0;
  98         char pathname[MAXPATHLEN];
  99         bool hadexclude = false;
 100
 101         dir = opendir(dirname);
 102         if (!dir) {
 103                 ERROR("failed to open %s", dirname);
 104                 return -1;
 105         }
 106
 107         while ((direntp = readdir(dir))) {
 108                 struct stat mystat;
 109                 int rc;
 110
 111                 if (!direntp)
 112                         break;
 113
 114                 if (!strcmp(direntp->d_name, ".") ||
 115                     !strcmp(direntp->d_name, ".."))
 116                         continue;
 117
 118                 rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name);
 119                 if (rc < 0 || rc >= MAXPATHLEN) {
 120                         ERROR("pathname too long");
 121                         failed=1;
 122                         continue;
 123                 }
 124
 125                 if (!level && exclude && !strcmp(direntp->d_name, exclude)) {
 126                         ret = rmdir(pathname);
 127                         if (ret < 0) {
 128                                 switch(errno) {
 129                                 case ENOTEMPTY:
 130                                         INFO("Not deleting snapshot %s", pathname);
 131                                         hadexclude = true;
 132                                         break;
 133                                 case ENOTDIR:
 134                                         ret = unlink(pathname);
 135                                         if (ret)
 136                                                 INFO("Failed to remove %s", pathname);
 137                                         break;
 138                                 default:
 139                                         SYSERROR("Failed to rmdir %s", pathname);
 140                                         failed = 1;
 141                                         break;
 142                                 }
 143                         }
 144                         continue;
 145                 }
 146
 147                 ret = lstat(pathname, &mystat);
 148                 if (ret) {
 149                         ERROR("Failed to stat %s", pathname);
 150                         failed = 1;
 151                         continue;
 152                 }
 153                 if (onedev && mystat.st_dev != pdev) {
 154                         /* TODO should we be checking /proc/self/mountinfo for
 155                          * pathname and not doing this if found? */
 156                         if (btrfs_try_remove_subvol(pathname))
 157                                 INFO("Removed btrfs subvolume at %s\n", pathname);
 158                         continue;
 159                 }
 160                 if (S_ISDIR(mystat.st_mode)) {
 161                         if (_recursive_rmdir(pathname, pdev, exclude, level+1, onedev) < 0)
 162                                 failed=1;
 163                 } else {
 164                         if (unlink(pathname) < 0) {
 165                                 SYSERROR("Failed to delete %s", pathname);
 166                                 failed=1;
 167                         }
 168                 }
 169         }
 170
 171         if (rmdir(dirname) < 0 && !btrfs_try_remove_subvol(dirname) && !hadexclude) {
 172                 ERROR("Failed to delete %s", dirname);
 173                 failed=1;
 174         }
 175
 176         ret = closedir(dir);
 177         if (ret) {
 178                 ERROR("Failed to close directory %s", dirname);
 179                 failed=1;
 180         }
 181
 182         return failed ? -1 : 0;
 183 }
 184
 185 /* We have two different magic values for overlayfs, yay. */
 186 #ifndef OVERLAYFS_SUPER_MAGIC
 187 #define OVERLAYFS_SUPER_MAGIC 0x794c764f
 188 #endif
 189
 190 #ifndef OVERLAY_SUPER_MAGIC
 191 #define OVERLAY_SUPER_MAGIC 0x794c7630
 192 #endif
 193
 194 /* In overlayfs, st_dev is unreliable. So on overlayfs we don't do the
 195  * lxc_rmdir_onedev()
 196  */
 197 static bool is_native_overlayfs(const char *path)
 198 {
 199         if (has_fs_type(path, OVERLAY_SUPER_MAGIC) ||
 200             has_fs_type(path, OVERLAYFS_SUPER_MAGIC))
 201                 return true;
 202
 203         return false;
 204 }
 205
 206 /* returns 0 on success, -1 if there were any failures */
 207 extern int lxc_rmdir_onedev(char *path, const char *exclude)
 208 {
 209         struct stat mystat;
 210         bool onedev = true;
 211
 212         if (is_native_overlayfs(path)) {
 213                 onedev = false;
 214         }
 215
 216         if (lstat(path, &mystat) < 0) {
 217                 if (errno == ENOENT)
 218                         return 0;
 219                 ERROR("Failed to stat %s", path);
 220                 return -1;
 221         }
 222
 223         return _recursive_rmdir(path, mystat.st_dev, exclude, 0, onedev);
 224 }
 225
 226 /* borrowed from iproute2 */
 227 extern int get_u16(unsigned short *val, const char *arg, int base)
 228 {
 229         unsigned long res;
 230         char *ptr;
 231
 232         if (!arg || !*arg)
 233                 return -1;
 234
 235         errno = 0;
 236         res = strtoul(arg, &ptr, base);
 237         if (!ptr || ptr == arg || *ptr || res > 0xFFFF || errno != 0)
 238                 return -1;
 239
 240         *val = res;
 241
 242         return 0;
 243 }
 244
 245 extern int mkdir_p(const char *dir, mode_t mode)
 246 {
 247         const char *tmp = dir;
 248         const char *orig = dir;
 249         char *makeme;
 250
 251         do {
 252                 dir = tmp + strspn(tmp, "/");
 253                 tmp = dir + strcspn(dir, "/");
 254                 makeme = strndup(orig, dir - orig);
 255                 if (*makeme) {
 256                         if (mkdir(makeme, mode) && errno != EEXIST) {
 257                                 SYSERROR("failed to create directory '%s'", makeme);
 258                                 free(makeme);
 259                                 return -1;
 260                         }
 261                 }
 262                 free(makeme);
 263         } while(tmp != dir);
 264
 265         return 0;
 266 }
 267
 268 char *get_rundir()
 269 {
 270         char *rundir;
 271         const char *homedir;
 272
 273         if (geteuid() == 0) {
 274                 rundir = strdup(RUNTIME_PATH);
 275                 return rundir;
 276         }
 277
 278         rundir = getenv("XDG_RUNTIME_DIR");
 279         if (rundir) {
 280                 rundir = strdup(rundir);
 281                 return rundir;
 282         }
 283
 284         INFO("XDG_RUNTIME_DIR isn't set in the environment.");
 285         homedir = getenv("HOME");
 286         if (!homedir) {
 287                 ERROR("HOME isn't set in the environment.");
 288                 return NULL;
 289         }
 290
 291         rundir = malloc(sizeof(char) * (17 + strlen(homedir)));
 292         sprintf(rundir, "%s/.cache/lxc/run/", homedir);
 293
 294         return rundir;
 295 }
 296
 297 int wait_for_pid(pid_t pid)
 298 {
 299         int status, ret;
 300
 301 again:
 302         ret = waitpid(pid, &status, 0);
 303         if (ret == -1) {
 304                 if (errno == EINTR)
 305                         goto again;
 306                 return -1;
 307         }
 308         if (ret != pid)
 309                 goto again;
 310         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0)
 311                 return -1;
 312         return 0;
 313 }
 314
 315 int lxc_wait_for_pid_status(pid_t pid)
 316 {
 317         int status, ret;
 318
 319 again:
 320         ret = waitpid(pid, &status, 0);
 321         if (ret == -1) {
 322                 if (errno == EINTR)
 323                         goto again;
 324                 return -1;
 325         }
 326         if (ret != pid)
 327                 goto again;
 328         return status;
 329 }
 330
 331 ssize_t lxc_write_nointr(int fd, const void* buf, size_t count)
 332 {
 333         ssize_t ret;
 334 again:
 335         ret = write(fd, buf, count);
 336         if (ret < 0 && errno == EINTR)
 337                 goto again;
 338         return ret;
 339 }
 340
 341 ssize_t lxc_read_nointr(int fd, void* buf, size_t count)
 342 {
 343         ssize_t ret;
 344 again:
 345         ret = read(fd, buf, count);
 346         if (ret < 0 && errno == EINTR)
 347                 goto again;
 348         return ret;
 349 }
 350
 351 ssize_t lxc_read_nointr_expect(int fd, void* buf, size_t count, const void* expected_buf)
 352 {
 353         ssize_t ret;
 354         ret = lxc_read_nointr(fd, buf, count);
 355         if (ret <= 0)
 356                 return ret;
 357         if ((size_t)ret != count)
 358                 return -1;
 359         if (expected_buf && memcmp(buf, expected_buf, count) != 0) {
 360                 errno = EINVAL;
 361                 return -1;
 362         }
 363         return ret;
 364 }
 365
 366 #if HAVE_LIBGNUTLS
 367 #include <gnutls/gnutls.h>
 368 #include <gnutls/crypto.h>
 369
 370 __attribute__((constructor))
 371 static void gnutls_lxc_init(void)
 372 {
 373         gnutls_global_init();
 374 }
 375
 376 int sha1sum_file(char *fnam, unsigned char *digest)
 377 {
 378         char *buf;
 379         int ret;
 380         FILE *f;
 381         long flen;
 382
 383         if (!fnam)
 384                 return -1;
 385         f = fopen_cloexec(fnam, "r");
 386         if (!f) {
 387                 SYSERROR("Error opening template");
 388                 return -1;
 389         }
 390         if (fseek(f, 0, SEEK_END) < 0) {
 391                 SYSERROR("Error seeking to end of template");
 392                 fclose(f);
 393                 return -1;
 394         }
 395         if ((flen = ftell(f)) < 0) {
 396                 SYSERROR("Error telling size of template");
 397                 fclose(f);
 398                 return -1;
 399         }
 400         if (fseek(f, 0, SEEK_SET) < 0) {
 401                 SYSERROR("Error seeking to start of template");
 402                 fclose(f);
 403                 return -1;
 404         }
 405         if ((buf = malloc(flen+1)) == NULL) {
 406                 SYSERROR("Out of memory");
 407                 fclose(f);
 408                 return -1;
 409         }
 410         if (fread(buf, 1, flen, f) != flen) {
 411                 SYSERROR("Failure reading template");
 412                 free(buf);
 413                 fclose(f);
 414                 return -1;
 415         }
 416         if (fclose(f) < 0) {
 417                 SYSERROR("Failre closing template");
 418                 free(buf);
 419                 return -1;
 420         }
 421         buf[flen] = '\0';
 422         ret = gnutls_hash_fast(GNUTLS_DIG_SHA1, buf, flen, (void *)digest);
 423         free(buf);
 424         return ret;
 425 }
 426 #endif
 427
 428 char** lxc_va_arg_list_to_argv(va_list ap, size_t skip, int do_strdup)
 429 {
 430         va_list ap2;
 431         size_t count = 1 + skip;
 432         char **result;
 433
 434         /* first determine size of argument list, we don't want to reallocate
 435          * constantly...
 436          */
 437         va_copy(ap2, ap);
 438         while (1) {
 439                 char* arg = va_arg(ap2, char*);
 440                 if (!arg)
 441                         break;
 442                 count++;
 443         }
 444         va_end(ap2);
 445
 446         result = calloc(count, sizeof(char*));
 447         if (!result)
 448                 return NULL;
 449         count = skip;
 450         while (1) {
 451                 char* arg = va_arg(ap, char*);
 452                 if (!arg)
 453                         break;
 454                 arg = do_strdup ? strdup(arg) : arg;
 455                 if (!arg)
 456                         goto oom;
 457                 result[count++] = arg;
 458         }
 459
 460         /* calloc has already set last element to NULL*/
 461         return result;
 462
 463 oom:
 464         free(result);
 465         return NULL;
 466 }
 467
 468 const char** lxc_va_arg_list_to_argv_const(va_list ap, size_t skip)
 469 {
 470         return (const char**)lxc_va_arg_list_to_argv(ap, skip, 0);
 471 }
 472
 473 struct lxc_popen_FILE *lxc_popen(const char *command)
 474 {
 475         int ret;
 476         int pipe_fds[2];
 477         pid_t child_pid;
 478         struct lxc_popen_FILE *fp = NULL;
 479
 480         ret = pipe2(pipe_fds, O_CLOEXEC);
 481         if (ret < 0)
 482                 return NULL;
 483
 484         child_pid = fork();
 485         if (child_pid < 0)
 486                 goto on_error;
 487
 488         if (!child_pid) {
 489                 sigset_t mask;
 490
 491                 close(pipe_fds[0]);
 492
 493                 /* duplicate stdout */
 494                 if (pipe_fds[1] != STDOUT_FILENO)
 495                         ret = dup2(pipe_fds[1], STDOUT_FILENO);
 496                 else
 497                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 498                 if (ret < 0) {
 499                         close(pipe_fds[1]);
 500                         exit(EXIT_FAILURE);
 501                 }
 502
 503                 /* duplicate stderr */
 504                 if (pipe_fds[1] != STDERR_FILENO)
 505                         ret = dup2(pipe_fds[1], STDERR_FILENO);
 506                 else
 507                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 508                 close(pipe_fds[1]);
 509                 if (ret < 0)
 510                         exit(EXIT_FAILURE);
 511
 512                 /* unblock all signals */
 513                 ret = sigfillset(&mask);
 514                 if (ret < 0)
 515                         exit(EXIT_FAILURE);
 516
 517                 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
 518                 if (ret < 0)
 519                         exit(EXIT_FAILURE);
 520
 521                 execl("/bin/sh", "sh", "-c", command, (char *)NULL);
 522                 exit(127);
 523         }
 524
 525         close(pipe_fds[1]);
 526         pipe_fds[1] = -1;
 527
 528         fp = malloc(sizeof(*fp));
 529         if (!fp)
 530                 goto on_error;
 531
 532         fp->child_pid = child_pid;
 533         fp->pipe = pipe_fds[0];
 534
 535         fp->f = fdopen(pipe_fds[0], "r");
 536         if (!fp->f)
 537                 goto on_error;
 538
 539         return fp;
 540
 541 on_error:
 542         if (fp)
 543                 free(fp);
 544
 545         if (pipe_fds[0] >= 0)
 546                 close(pipe_fds[0]);
 547
 548         if (pipe_fds[1] >= 0)
 549                 close(pipe_fds[1]);
 550
 551         return NULL;
 552 }
 553
 554 int lxc_pclose(struct lxc_popen_FILE *fp)
 555 {
 556         pid_t wait_pid;
 557         int wstatus = 0;
 558
 559         if (!fp)
 560                 return -1;
 561
 562         do {
 563                 wait_pid = waitpid(fp->child_pid, &wstatus, 0);
 564         } while (wait_pid < 0 && errno == EINTR);
 565
 566         close(fp->pipe);
 567         fclose(fp->f);
 568         free(fp);
 569
 570         if (wait_pid < 0)
 571                 return -1;
 572
 573         return wstatus;
 574 }
 575
 576 char *lxc_string_replace(const char *needle, const char *replacement, const char *haystack)
 577 {
 578         ssize_t len = -1, saved_len = -1;
 579         char *result = NULL;
 580         size_t replacement_len = strlen(replacement);
 581         size_t needle_len = strlen(needle);
 582
 583         /* should be executed exactly twice */
 584         while (len == -1 || result == NULL) {
 585                 char *p;
 586                 char *last_p;
 587                 ssize_t part_len;
 588
 589                 if (len != -1) {
 590                         result = calloc(1, len + 1);
 591                         if (!result)
 592                                 return NULL;
 593                         saved_len = len;
 594                 }
 595
 596                 len = 0;
 597
 598                 for (last_p = (char *)haystack, p = strstr(last_p, needle); p; last_p = p, p = strstr(last_p, needle)) {
 599                         part_len = (ssize_t)(p - last_p);
 600                         if (result && part_len > 0)
 601                                 memcpy(&result[len], last_p, part_len);
 602                         len += part_len;
 603                         if (result && replacement_len > 0)
 604                                 memcpy(&result[len], replacement, replacement_len);
 605                         len += replacement_len;
 606                         p += needle_len;
 607                 }
 608                 part_len = strlen(last_p);
 609                 if (result && part_len > 0)
 610                         memcpy(&result[len], last_p, part_len);
 611                 len += part_len;
 612         }
 613
 614         /* make sure we did the same thing twice,
 615          * once for calculating length, the other
 616          * time for copying data */
 617         if (saved_len != len) {
 618                 free(result);
 619                 return NULL;
 620         }
 621         /* make sure we didn't overwrite any buffer,
 622          * due to calloc the string should be 0-terminated */
 623         if (result[len] != '\0') {
 624                 free(result);
 625                 return NULL;
 626         }
 627
 628         return result;
 629 }
 630
 631 bool lxc_string_in_array(const char *needle, const char **haystack)
 632 {
 633         for (; haystack && *haystack; haystack++)
 634                 if (!strcmp(needle, *haystack))
 635                         return true;
 636         return false;
 637 }
 638
 639 char *lxc_string_join(const char *sep, const char **parts, bool use_as_prefix)
 640 {
 641         char *result;
 642         char **p;
 643         size_t sep_len = strlen(sep);
 644         size_t result_len = use_as_prefix * sep_len;
 645
 646         /* calculate new string length */
 647         for (p = (char **)parts; *p; p++)
 648                 result_len += (p > (char **)parts) * sep_len + strlen(*p);
 649
 650         result = calloc(result_len + 1, 1);
 651         if (!result)
 652                 return NULL;
 653
 654         if (use_as_prefix)
 655                 strcpy(result, sep);
 656         for (p = (char **)parts; *p; p++) {
 657                 if (p > (char **)parts)
 658                         strcat(result, sep);
 659                 strcat(result, *p);
 660         }
 661
 662         return result;
 663 }
 664
 665 char **lxc_normalize_path(const char *path)
 666 {
 667         char **components;
 668         char **p;
 669         size_t components_len = 0;
 670         size_t pos = 0;
 671
 672         components = lxc_string_split(path, '/');
 673         if (!components)
 674                 return NULL;
 675         for (p = components; *p; p++)
 676                 components_len++;
 677
 678         /* resolve '.' and '..' */
 679         for (pos = 0; pos < components_len; ) {
 680                 if (!strcmp(components[pos], ".") || (!strcmp(components[pos], "..") && pos == 0)) {
 681                         /* eat this element */
 682                         free(components[pos]);
 683                         memmove(&components[pos], &components[pos+1], sizeof(char *) * (components_len - pos));
 684                         components_len--;
 685                 } else if (!strcmp(components[pos], "..")) {
 686                         /* eat this and the previous element */
 687                         free(components[pos - 1]);
 688                         free(components[pos]);
 689                         memmove(&components[pos-1], &components[pos+1], sizeof(char *) * (components_len - pos));
 690                         components_len -= 2;
 691                         pos--;
 692                 } else {
 693                         pos++;
 694                 }
 695         }
 696
 697         return components;
 698 }
 699
 700 char *lxc_deslashify(const char *path)
 701 {
 702         char *dup, *p;
 703         char **parts = NULL;
 704         size_t n, len;
 705
 706         dup = strdup(path);
 707         if (!dup)
 708                 return NULL;
 709
 710         parts = lxc_normalize_path(dup);
 711         if (!parts) {
 712                 free(dup);
 713                 return NULL;
 714         }
 715
 716         /* We'll end up here if path == "///" or path == "". */
 717         if (!*parts) {
 718                 len = strlen(dup);
 719                 if (!len) {
 720                         lxc_free_array((void **)parts, free);
 721                         return dup;
 722                 }
 723                 n = strcspn(dup, "/");
 724                 if (n == len) {
 725                         free(dup);
 726                         lxc_free_array((void **)parts, free);
 727
 728                         p = strdup("/");
 729                         if (!p)
 730                                 return NULL;
 731
 732                         return p;
 733                 }
 734         }
 735
 736         p = lxc_string_join("/", (const char **)parts, *dup == '/');
 737         free(dup);
 738         lxc_free_array((void **)parts, free);
 739         return p;
 740 }
 741
 742 char *lxc_append_paths(const char *first, const char *second)
 743 {
 744         size_t len = strlen(first) + strlen(second) + 1;
 745         const char *pattern = "%s%s";
 746         char *result = NULL;
 747
 748         if (second[0] != '/') {
 749                 len += 1;
 750                 pattern = "%s/%s";
 751         }
 752
 753         result = calloc(1, len);
 754         if (!result)
 755                 return NULL;
 756
 757         snprintf(result, len, pattern, first, second);
 758         return result;
 759 }
 760
 761 bool lxc_string_in_list(const char *needle, const char *haystack, char _sep)
 762 {
 763         char *token, *str, *saveptr = NULL;
 764         char sep[2] = { _sep, '\0' };
 765
 766         if (!haystack || !needle)
 767                 return 0;
 768
 769         str = alloca(strlen(haystack)+1);
 770         strcpy(str, haystack);
 771         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 772                 if (strcmp(needle, token) == 0)
 773                         return 1;
 774         }
 775
 776         return 0;
 777 }
 778
 779 char **lxc_string_split(const char *string, char _sep)
 780 {
 781         char *token, *str, *saveptr = NULL;
 782         char sep[2] = {_sep, '\0'};
 783         char **tmp = NULL, **result = NULL;
 784         size_t result_capacity = 0;
 785         size_t result_count = 0;
 786         int r, saved_errno;
 787
 788         if (!string)
 789                 return calloc(1, sizeof(char *));
 790
 791         str = alloca(strlen(string) + 1);
 792         strcpy(str, string);
 793         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 794                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 795                 if (r < 0)
 796                         goto error_out;
 797                 result[result_count] = strdup(token);
 798                 if (!result[result_count])
 799                         goto error_out;
 800                 result_count++;
 801         }
 802
 803         /* if we allocated too much, reduce it */
 804         tmp = realloc(result, (result_count + 1) * sizeof(char *));
 805         if (!tmp)
 806                 goto error_out;
 807         result = tmp;
 808         /* Make sure we don't return uninitialized memory. */
 809         if (result_count == 0)
 810                 *result = NULL;
 811         return result;
 812 error_out:
 813         saved_errno = errno;
 814         lxc_free_array((void **)result, free);
 815         errno = saved_errno;
 816         return NULL;
 817 }
 818
 819 char **lxc_string_split_and_trim(const char *string, char _sep)
 820 {
 821         char *token, *str, *saveptr = NULL;
 822         char sep[2] = { _sep, '\0' };
 823         char **result = NULL;
 824         size_t result_capacity = 0;
 825         size_t result_count = 0;
 826         int r, saved_errno;
 827         size_t i = 0;
 828
 829         if (!string)
 830                 return calloc(1, sizeof(char *));
 831
 832         str = alloca(strlen(string)+1);
 833         strcpy(str, string);
 834         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 835                 while (token[0] == ' ' || token[0] == '\t')
 836                         token++;
 837                 i = strlen(token);
 838                 while (i > 0 && (token[i - 1] == ' ' || token[i - 1] == '\t')) {
 839                         token[i - 1] = '\0';
 840                         i--;
 841                 }
 842                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 843                 if (r < 0)
 844                         goto error_out;
 845                 result[result_count] = strdup(token);
 846                 if (!result[result_count])
 847                         goto error_out;
 848                 result_count++;
 849         }
 850
 851         /* if we allocated too much, reduce it */
 852         return realloc(result, (result_count + 1) * sizeof(char *));
 853 error_out:
 854         saved_errno = errno;
 855         lxc_free_array((void **)result, free);
 856         errno = saved_errno;
 857         return NULL;
 858 }
 859
 860 void lxc_free_array(void **array, lxc_free_fn element_free_fn)
 861 {
 862         void **p;
 863         for (p = array; p && *p; p++)
 864                 element_free_fn(*p);
 865         free((void*)array);
 866 }
 867
 868 int lxc_grow_array(void ***array, size_t* capacity, size_t new_size, size_t capacity_increment)
 869 {
 870         size_t new_capacity;
 871         void **new_array;
 872
 873         /* first time around, catch some trivial mistakes of the user
 874          * only initializing one of these */
 875         if (!*array || !*capacity) {
 876                 *array = NULL;
 877                 *capacity = 0;
 878         }
 879
 880         new_capacity = *capacity;
 881         while (new_size + 1 > new_capacity)
 882                 new_capacity += capacity_increment;
 883         if (new_capacity != *capacity) {
 884                 /* we have to reallocate */
 885                 new_array = realloc(*array, new_capacity * sizeof(void *));
 886                 if (!new_array)
 887                         return -1;
 888                 memset(&new_array[*capacity], 0, (new_capacity - (*capacity)) * sizeof(void *));
 889                 *array = new_array;
 890                 *capacity = new_capacity;
 891         }
 892
 893         /* array has sufficient elements */
 894         return 0;
 895 }
 896
 897 size_t lxc_array_len(void **array)
 898 {
 899         void **p;
 900         size_t result = 0;
 901
 902         for (p = array; p && *p; p++)
 903                 result++;
 904
 905         return result;
 906 }
 907
 908 int lxc_write_to_file(const char *filename, const void* buf, size_t count, bool add_newline)
 909 {
 910         int fd, saved_errno;
 911         ssize_t ret;
 912
 913         fd = open(filename, O_WRONLY | O_TRUNC | O_CREAT | O_CLOEXEC, 0666);
 914         if (fd < 0)
 915                 return -1;
 916         ret = lxc_write_nointr(fd, buf, count);
 917         if (ret < 0)
 918                 goto out_error;
 919         if ((size_t)ret != count)
 920                 goto out_error;
 921         if (add_newline) {
 922                 ret = lxc_write_nointr(fd, "\n", 1);
 923                 if (ret != 1)
 924                         goto out_error;
 925         }
 926         close(fd);
 927         return 0;
 928
 929 out_error:
 930         saved_errno = errno;
 931         close(fd);
 932         errno = saved_errno;
 933         return -1;
 934 }
 935
 936 int lxc_read_from_file(const char *filename, void* buf, size_t count)
 937 {
 938         int fd = -1, saved_errno;
 939         ssize_t ret;
 940
 941         fd = open(filename, O_RDONLY | O_CLOEXEC);
 942         if (fd < 0)
 943                 return -1;
 944
 945         if (!buf || !count) {
 946                 char buf2[100];
 947                 size_t count2 = 0;
 948                 while ((ret = read(fd, buf2, 100)) > 0)
 949                         count2 += ret;
 950                 if (ret >= 0)
 951                         ret = count2;
 952         } else {
 953                 memset(buf, 0, count);
 954                 ret = read(fd, buf, count);
 955         }
 956
 957         if (ret < 0)
 958                 ERROR("read %s: %s", filename, strerror(errno));
 959
 960         saved_errno = errno;
 961         close(fd);
 962         errno = saved_errno;
 963         return ret;
 964 }
 965
 966 void **lxc_append_null_to_array(void **array, size_t count)
 967 {
 968         void **temp;
 969
 970         /* Append NULL to the array */
 971         if (count) {
 972                 temp = realloc(array, (count + 1) * sizeof(*array));
 973                 if (!temp) {
 974                         size_t i;
 975                         for (i = 0; i < count; i++)
 976                                 free(array[i]);
 977                         free(array);
 978                         return NULL;
 979                 }
 980                 array = temp;
 981                 array[count] = NULL;
 982         }
 983         return array;
 984 }
 985
 986 int randseed(bool srand_it)
 987 {
 988         /*
 989            srand pre-seed function based on /dev/urandom
 990            */
 991         unsigned int seed = time(NULL) + getpid();
 992
 993         FILE *f;
 994         f = fopen("/dev/urandom", "r");
 995         if (f) {
 996                 int ret = fread(&seed, sizeof(seed), 1, f);
 997                 if (ret != 1)
 998                         DEBUG("unable to fread /dev/urandom, %s, fallback to time+pid rand seed", strerror(errno));
 999                 fclose(f);
1000         }
1001
1002         if (srand_it)
1003                 srand(seed);
1004
1005         return seed;
1006 }
1007
1008 uid_t get_ns_uid(uid_t orig)
1009 {
1010         char *line = NULL;
1011         size_t sz = 0;
1012         uid_t nsid, hostid, range;
1013         FILE *f = fopen("/proc/self/uid_map", "r");
1014         if (!f)
1015                 return 0;
1016
1017         while (getline(&line, &sz, f) != -1) {
1018                 if (sscanf(line, "%u %u %u", &nsid, &hostid, &range) != 3)
1019                         continue;
1020                 if (hostid <= orig && hostid + range > orig) {
1021                         nsid += orig - hostid;
1022                         goto found;
1023                 }
1024         }
1025
1026         nsid = 0;
1027 found:
1028         fclose(f);
1029         free(line);
1030         return nsid;
1031 }
1032
1033 bool dir_exists(const char *path)
1034 {
1035         struct stat sb;
1036         int ret;
1037
1038         ret = stat(path, &sb);
1039         if (ret < 0)
1040                 /* Could be something other than eexist, just say "no". */
1041                 return false;
1042         return S_ISDIR(sb.st_mode);
1043 }
1044
1045 /* Note we don't use SHA-1 here as we don't want to depend on HAVE_GNUTLS.
1046  * FNV has good anti collision properties and we're not worried
1047  * about pre-image resistance or one-way-ness, we're just trying to make
1048  * the name unique in the 108 bytes of space we have.
1049  */
1050 uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
1051 {
1052         unsigned char *bp;
1053
1054         for(bp = buf; bp < (unsigned char *)buf + len; bp++)
1055         {
1056                 /* xor the bottom with the current octet */
1057                 hval ^= (uint64_t)*bp;
1058
1059                 /* gcc optimised:
1060                  * multiply by the 64 bit FNV magic prime mod 2^64
1061                  */
1062                 hval += (hval << 1) + (hval << 4) + (hval << 5) +
1063                         (hval << 7) + (hval << 8) + (hval << 40);
1064         }
1065
1066         return hval;
1067 }
1068
1069 /*
1070  * Detect whether / is mounted MS_SHARED.  The only way I know of to
1071  * check that is through /proc/self/mountinfo.
1072  * I'm only checking for /.  If the container rootfs or mount location
1073  * is MS_SHARED, but not '/', then you're out of luck - figuring that
1074  * out would be too much work to be worth it.
1075  */
1076 int detect_shared_rootfs(void)
1077 {
1078         char buf[LXC_LINELEN], *p;
1079         FILE *f;
1080         int i;
1081         char *p2;
1082
1083         f = fopen("/proc/self/mountinfo", "r");
1084         if (!f)
1085                 return 0;
1086         while (fgets(buf, LXC_LINELEN, f)) {
1087                 for (p = buf, i = 0; p && i < 4; i++)
1088                         p = strchr(p + 1, ' ');
1089                 if (!p)
1090                         continue;
1091                 p2 = strchr(p + 1, ' ');
1092                 if (!p2)
1093                         continue;
1094                 *p2 = '\0';
1095                 if (strcmp(p + 1, "/") == 0) {
1096                         /* This is '/'. Is it shared? */
1097                         p = strchr(p2 + 1, ' ');
1098                         if (p && strstr(p, "shared:")) {
1099                                 fclose(f);
1100                                 return 1;
1101                         }
1102                 }
1103         }
1104         fclose(f);
1105         return 0;
1106 }
1107
1108 bool switch_to_ns(pid_t pid, const char *ns) {
1109         int fd, ret;
1110         char nspath[MAXPATHLEN];
1111
1112         /* Switch to new ns */
1113         ret = snprintf(nspath, MAXPATHLEN, "/proc/%d/ns/%s", pid, ns);
1114         if (ret < 0 || ret >= MAXPATHLEN)
1115                 return false;
1116
1117         fd = open(nspath, O_RDONLY);
1118         if (fd < 0) {
1119                 SYSERROR("failed to open %s", nspath);
1120                 return false;
1121         }
1122
1123         ret = setns(fd, 0);
1124         if (ret) {
1125                 SYSERROR("failed to set process %d to %s of %d.", pid, ns, fd);
1126                 close(fd);
1127                 return false;
1128         }
1129         close(fd);
1130         return true;
1131 }
1132
1133 /*
1134  * looking at fs/proc_namespace.c, it appears we can
1135  * actually expect the rootfs entry to very specifically contain
1136  * " - rootfs rootfs "
1137  * IIUC, so long as we've chrooted so that rootfs is not our root,
1138  * the rootfs entry should always be skipped in mountinfo contents.
1139  */
1140 bool detect_ramfs_rootfs(void)
1141 {
1142         FILE *f;
1143         char *p, *p2;
1144         char *line = NULL;
1145         size_t len = 0;
1146         int i;
1147
1148         f = fopen("/proc/self/mountinfo", "r");
1149         if (!f)
1150                 return false;
1151
1152         while (getline(&line, &len, f) != -1) {
1153                 for (p = line, i = 0; p && i < 4; i++)
1154                         p = strchr(p + 1, ' ');
1155                 if (!p)
1156                         continue;
1157                 p2 = strchr(p + 1, ' ');
1158                 if (!p2)
1159                         continue;
1160                 *p2 = '\0';
1161                 if (strcmp(p + 1, "/") == 0) {
1162                         /* This is '/'. Is it the ramfs? */
1163                         p = strchr(p2 + 1, '-');
1164                         if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) {
1165                                 free(line);
1166                                 fclose(f);
1167                                 return true;
1168                         }
1169                 }
1170         }
1171         free(line);
1172         fclose(f);
1173         return false;
1174 }
1175
1176 char *on_path(const char *cmd, const char *rootfs) {
1177         char *path = NULL;
1178         char *entry = NULL;
1179         char *saveptr = NULL;
1180         char cmdpath[MAXPATHLEN];
1181         int ret;
1182
1183         path = getenv("PATH");
1184         if (!path)
1185                 return NULL;
1186
1187         path = strdup(path);
1188         if (!path)
1189                 return NULL;
1190
1191         entry = strtok_r(path, ":", &saveptr);
1192         while (entry) {
1193                 if (rootfs)
1194                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s/%s", rootfs, entry, cmd);
1195                 else
1196                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s", entry, cmd);
1197
1198                 if (ret < 0 || ret >= MAXPATHLEN)
1199                         goto next_loop;
1200
1201                 if (access(cmdpath, X_OK) == 0) {
1202                         free(path);
1203                         return strdup(cmdpath);
1204                 }
1205
1206 next_loop:
1207                 entry = strtok_r(NULL, ":", &saveptr);
1208         }
1209
1210         free(path);
1211         return NULL;
1212 }
1213
1214 bool file_exists(const char *f)
1215 {
1216         struct stat statbuf;
1217
1218         return stat(f, &statbuf) == 0;
1219 }
1220
1221 bool cgns_supported(void)
1222 {
1223         return file_exists("/proc/self/ns/cgroup");
1224 }
1225
1226 /* historically lxc-init has been under /usr/lib/lxc and under
1227  * /usr/lib/$ARCH/lxc.  It now lives as $prefix/sbin/init.lxc.
1228  */
1229 char *choose_init(const char *rootfs)
1230 {
1231         char *retv = NULL;
1232         const char *empty = "",
1233                    *tmp;
1234         int ret, env_set = 0;
1235
1236         if (!getenv("PATH")) {
1237                 if (setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 0))
1238                         SYSERROR("Failed to setenv");
1239                 env_set = 1;
1240         }
1241
1242         retv = on_path("init.lxc", rootfs);
1243
1244         if (env_set) {
1245                 if (unsetenv("PATH"))
1246                         SYSERROR("Failed to unsetenv");
1247         }
1248
1249         if (retv)
1250                 return retv;
1251
1252         retv = malloc(PATH_MAX);
1253         if (!retv)
1254                 return NULL;
1255
1256         if (rootfs)
1257                 tmp = rootfs;
1258         else
1259                 tmp = empty;
1260
1261         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, SBINDIR, "/init.lxc");
1262         if (ret < 0 || ret >= PATH_MAX) {
1263                 ERROR("pathname too long");
1264                 goto out1;
1265         }
1266         if (access(retv, X_OK) == 0)
1267                 return retv;
1268
1269         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, LXCINITDIR, "/lxc/lxc-init");
1270         if (ret < 0 || ret >= PATH_MAX) {
1271                 ERROR("pathname too long");
1272                 goto out1;
1273         }
1274         if (access(retv, X_OK) == 0)
1275                 return retv;
1276
1277         ret = snprintf(retv, PATH_MAX, "%s/usr/lib/lxc/lxc-init", tmp);
1278         if (ret < 0 || ret >= PATH_MAX) {
1279                 ERROR("pathname too long");
1280                 goto out1;
1281         }
1282         if (access(retv, X_OK) == 0)
1283                 return retv;
1284
1285         ret = snprintf(retv, PATH_MAX, "%s/sbin/lxc-init", tmp);
1286         if (ret < 0 || ret >= PATH_MAX) {
1287                 ERROR("pathname too long");
1288                 goto out1;
1289         }
1290         if (access(retv, X_OK) == 0)
1291                 return retv;
1292
1293         /*
1294          * Last resort, look for the statically compiled init.lxc which we
1295          * hopefully bind-mounted in.
1296          * If we are called during container setup, and we get to this point,
1297          * then the init.lxc.static from the host will need to be bind-mounted
1298          * in.  So we return NULL here to indicate that.
1299          */
1300         if (rootfs)
1301                 goto out1;
1302
1303         ret = snprintf(retv, PATH_MAX, "/init.lxc.static");
1304         if (ret < 0 || ret >= PATH_MAX) {
1305                 WARN("Nonsense - name /lxc.init.static too long");
1306                 goto out1;
1307         }
1308         if (access(retv, X_OK) == 0)
1309                 return retv;
1310
1311 out1:
1312         free(retv);
1313         return NULL;
1314 }
1315
1316 int print_to_file(const char *file, const char *content)
1317 {
1318         FILE *f;
1319         int ret = 0;
1320
1321         f = fopen(file, "w");
1322         if (!f)
1323                 return -1;
1324         if (fprintf(f, "%s", content) != strlen(content))
1325                 ret = -1;
1326         fclose(f);
1327         return ret;
1328 }
1329
1330 int is_dir(const char *path)
1331 {
1332         struct stat statbuf;
1333         int ret = stat(path, &statbuf);
1334         if (ret == 0 && S_ISDIR(statbuf.st_mode))
1335                 return 1;
1336         return 0;
1337 }
1338
1339 /*
1340  * Given the '-t' template option to lxc-create, figure out what to
1341  * do.  If the template is a full executable path, use that.  If it
1342  * is something like 'sshd', then return $templatepath/lxc-sshd.
1343  * On success return the template, on error return NULL.
1344  */
1345 char *get_template_path(const char *t)
1346 {
1347         int ret, len;
1348         char *tpath;
1349
1350         if (t[0] == '/' && access(t, X_OK) == 0) {
1351                 tpath = strdup(t);
1352                 return tpath;
1353         }
1354
1355         len = strlen(LXCTEMPLATEDIR) + strlen(t) + strlen("/lxc-") + 1;
1356         tpath = malloc(len);
1357         if (!tpath)
1358                 return NULL;
1359         ret = snprintf(tpath, len, "%s/lxc-%s", LXCTEMPLATEDIR, t);
1360         if (ret < 0 || ret >= len) {
1361                 free(tpath);
1362                 return NULL;
1363         }
1364         if (access(tpath, X_OK) < 0) {
1365                 SYSERROR("bad template: %s", t);
1366                 free(tpath);
1367                 return NULL;
1368         }
1369
1370         return tpath;
1371 }
1372
1373 /*
1374  * Sets the process title to the specified title. Note that this may fail if
1375  * the kernel doesn't support PR_SET_MM_MAP (kernels <3.18).
1376  */
1377 int setproctitle(char *title)
1378 {
1379         static char *proctitle = NULL;
1380         char buf[2048], *tmp;
1381         FILE *f;
1382         int i, len, ret = 0;
1383
1384         /* We don't really need to know all of this stuff, but unfortunately
1385          * PR_SET_MM_MAP requires us to set it all at once, so we have to
1386          * figure it out anyway.
1387          */
1388         unsigned long start_data, end_data, start_brk, start_code, end_code,
1389                         start_stack, arg_start, arg_end, env_start, env_end,
1390                         brk_val;
1391         struct prctl_mm_map prctl_map;
1392
1393         f = fopen_cloexec("/proc/self/stat", "r");
1394         if (!f) {
1395                 return -1;
1396         }
1397
1398         tmp = fgets(buf, sizeof(buf), f);
1399         fclose(f);
1400         if (!tmp) {
1401                 return -1;
1402         }
1403
1404         /* Skip the first 25 fields, column 26-28 are start_code, end_code,
1405          * and start_stack */
1406         tmp = strchr(buf, ' ');
1407         for (i = 0; i < 24; i++) {
1408                 if (!tmp)
1409                         return -1;
1410                 tmp = strchr(tmp+1, ' ');
1411         }
1412         if (!tmp)
1413                 return -1;
1414
1415         i = sscanf(tmp, "%lu %lu %lu", &start_code, &end_code, &start_stack);
1416         if (i != 3)
1417                 return -1;
1418
1419         /* Skip the next 19 fields, column 45-51 are start_data to arg_end */
1420         for (i = 0; i < 19; i++) {
1421                 if (!tmp)
1422                         return -1;
1423                 tmp = strchr(tmp+1, ' ');
1424         }
1425
1426         if (!tmp)
1427                 return -1;
1428
1429         i = sscanf(tmp, "%lu %lu %lu %*u %*u %lu %lu",
1430                 &start_data,
1431                 &end_data,
1432                 &start_brk,
1433                 &env_start,
1434                 &env_end);
1435         if (i != 5)
1436                 return -1;
1437
1438         /* Include the null byte here, because in the calculations below we
1439          * want to have room for it. */
1440         len = strlen(title) + 1;
1441
1442         proctitle = realloc(proctitle, len);
1443         if (!proctitle)
1444                 return -1;
1445
1446         arg_start = (unsigned long) proctitle;
1447         arg_end = arg_start + len;
1448
1449         brk_val = syscall(__NR_brk, 0);
1450
1451         prctl_map = (struct prctl_mm_map) {
1452                 .start_code = start_code,
1453                 .end_code = end_code,
1454                 .start_stack = start_stack,
1455                 .start_data = start_data,
1456                 .end_data = end_data,
1457                 .start_brk = start_brk,
1458                 .brk = brk_val,
1459                 .arg_start = arg_start,
1460                 .arg_end = arg_end,
1461                 .env_start = env_start,
1462                 .env_end = env_end,
1463                 .auxv = NULL,
1464                 .auxv_size = 0,
1465                 .exe_fd = -1,
1466         };
1467
1468         ret = prctl(PR_SET_MM, PR_SET_MM_MAP, (long) &prctl_map, sizeof(prctl_map), 0);
1469         if (ret == 0)
1470                 strcpy((char*)arg_start, title);
1471         else
1472                 INFO("setting cmdline failed - %s", strerror(errno));
1473
1474         return ret;
1475 }
1476
1477 /*
1478  * @path:    a pathname where / replaced with '\0'.
1479  * @offsetp: pointer to int showing which path segment was last seen.
1480  *           Updated on return to reflect the next segment.
1481  * @fulllen: full original path length.
1482  * Returns a pointer to the next path segment, or NULL if done.
1483  */
1484 static char *get_nextpath(char *path, int *offsetp, int fulllen)
1485 {
1486         int offset = *offsetp;
1487
1488         if (offset >= fulllen)
1489                 return NULL;
1490
1491         while (path[offset] != '\0' && offset < fulllen)
1492                 offset++;
1493         while (path[offset] == '\0' && offset < fulllen)
1494                 offset++;
1495
1496         *offsetp = offset;
1497         return (offset < fulllen) ? &path[offset] : NULL;
1498 }
1499
1500 /*
1501  * Check that @subdir is a subdir of @dir.  @len is the length of
1502  * @dir (to avoid having to recalculate it).
1503  */
1504 static bool is_subdir(const char *subdir, const char *dir, size_t len)
1505 {
1506         size_t subdirlen = strlen(subdir);
1507
1508         if (subdirlen < len)
1509                 return false;
1510         if (strncmp(subdir, dir, len) != 0)
1511                 return false;
1512         if (dir[len-1] == '/')
1513                 return true;
1514         if (subdir[len] == '/' || subdirlen == len)
1515                 return true;
1516         return false;
1517 }
1518
1519 /*
1520  * Check if the open fd is a symlink.  Return -ELOOP if it is.  Return
1521  * -ENOENT if we couldn't fstat.  Return 0 if the fd is ok.
1522  */
1523 static int check_symlink(int fd)
1524 {
1525         struct stat sb;
1526         int ret = fstat(fd, &sb);
1527         if (ret < 0)
1528                 return -ENOENT;
1529         if (S_ISLNK(sb.st_mode))
1530                 return -ELOOP;
1531         return 0;
1532 }
1533
1534 /*
1535  * Open a file or directory, provided that it contains no symlinks.
1536  *
1537  * CAVEAT: This function must not be used for other purposes than container
1538  * setup before executing the container's init
1539  */
1540 static int open_if_safe(int dirfd, const char *nextpath)
1541 {
1542         int newfd = openat(dirfd, nextpath, O_RDONLY | O_NOFOLLOW);
1543         if (newfd >= 0) /* Was not a symlink, all good. */
1544                 return newfd;
1545
1546         if (errno == ELOOP)
1547                 return newfd;
1548
1549         if (errno == EPERM || errno == EACCES) {
1550                 /* We're not root (cause we got EPERM) so try opening with
1551                  * O_PATH.
1552                  */
1553                 newfd = openat(dirfd, nextpath, O_PATH | O_NOFOLLOW);
1554                 if (newfd >= 0) {
1555                         /* O_PATH will return an fd for symlinks. We know
1556                          * nextpath wasn't a symlink at last openat, so if fd is
1557                          * now a link, then something * fishy is going on.
1558                          */
1559                         int ret = check_symlink(newfd);
1560                         if (ret < 0) {
1561                                 close(newfd);
1562                                 newfd = ret;
1563                         }
1564                 }
1565         }
1566
1567         return newfd;
1568 }
1569
1570 /*
1571  * Open a path intending for mounting, ensuring that the final path
1572  * is inside the container's rootfs.
1573  *
1574  * CAVEAT: This function must not be used for other purposes than container
1575  * setup before executing the container's init
1576  *
1577  * @target: path to be opened
1578  * @prefix_skip: a part of @target in which to ignore symbolic links.  This
1579  * would be the container's rootfs.
1580  *
1581  * Return an open fd for the path, or <0 on error.
1582  */
1583 static int open_without_symlink(const char *target, const char *prefix_skip)
1584 {
1585         int curlen = 0, dirfd, fulllen, i;
1586         char *dup = NULL;
1587
1588         fulllen = strlen(target);
1589
1590         /* make sure prefix-skip makes sense */
1591         if (prefix_skip && strlen(prefix_skip) > 0) {
1592                 curlen = strlen(prefix_skip);
1593                 if (!is_subdir(target, prefix_skip, curlen)) {
1594                         ERROR("WHOA there - target '%s' didn't start with prefix '%s'",
1595                                 target, prefix_skip);
1596                         return -EINVAL;
1597                 }
1598                 /*
1599                  * get_nextpath() expects the curlen argument to be
1600                  * on a  (turned into \0) / or before it, so decrement
1601                  * curlen to make sure that happens
1602                  */
1603                 if (curlen)
1604                         curlen--;
1605         } else {
1606                 prefix_skip = "/";
1607                 curlen = 0;
1608         }
1609
1610         /* Make a copy of target which we can hack up, and tokenize it */
1611         if ((dup = strdup(target)) == NULL) {
1612                 SYSERROR("Out of memory checking for symbolic link");
1613                 return -ENOMEM;
1614         }
1615         for (i = 0; i < fulllen; i++) {
1616                 if (dup[i] == '/')
1617                         dup[i] = '\0';
1618         }
1619
1620         dirfd = open(prefix_skip, O_RDONLY);
1621         if (dirfd < 0)
1622                 goto out;
1623         while (1) {
1624                 int newfd, saved_errno;
1625                 char *nextpath;
1626
1627                 if ((nextpath = get_nextpath(dup, &curlen, fulllen)) == NULL)
1628                         goto out;
1629                 newfd = open_if_safe(dirfd, nextpath);
1630                 saved_errno = errno;
1631                 close(dirfd);
1632                 dirfd = newfd;
1633                 if (newfd < 0) {
1634                         errno = saved_errno;
1635                         if (errno == ELOOP)
1636                                 SYSERROR("%s in %s was a symbolic link!", nextpath, target);
1637                         goto out;
1638                 }
1639         }
1640
1641 out:
1642         free(dup);
1643         return dirfd;
1644 }
1645
1646 /*
1647  * Safely mount a path into a container, ensuring that the mount target
1648  * is under the container's @rootfs.  (If @rootfs is NULL, then the container
1649  * uses the host's /)
1650  *
1651  * CAVEAT: This function must not be used for other purposes than container
1652  * setup before executing the container's init
1653  */
1654 int safe_mount(const char *src, const char *dest, const char *fstype,
1655                 unsigned long flags, const void *data, const char *rootfs)
1656 {
1657         int destfd, ret, saved_errno;
1658         /* Only needs enough for /proc/self/fd/<fd>. */
1659         char srcbuf[50], destbuf[50];
1660         int srcfd = -1;
1661         const char *mntsrc = src;
1662
1663         if (!rootfs)
1664                 rootfs = "";
1665
1666         /* todo - allow symlinks for relative paths if 'allowsymlinks' option is passed */
1667         if (flags & MS_BIND && src && src[0] != '/') {
1668                 INFO("this is a relative bind mount");
1669                 srcfd = open_without_symlink(src, NULL);
1670                 if (srcfd < 0)
1671                         return srcfd;
1672                 ret = snprintf(srcbuf, 50, "/proc/self/fd/%d", srcfd);
1673                 if (ret < 0 || ret > 50) {
1674                         close(srcfd);
1675                         ERROR("Out of memory");
1676                         return -EINVAL;
1677                 }
1678                 mntsrc = srcbuf;
1679         }
1680
1681         destfd = open_without_symlink(dest, rootfs);
1682         if (destfd < 0) {
1683                 if (srcfd != -1) {
1684                         saved_errno = errno;
1685                         close(srcfd);
1686                         errno = saved_errno;
1687                 }
1688                 return destfd;
1689         }
1690
1691         ret = snprintf(destbuf, 50, "/proc/self/fd/%d", destfd);
1692         if (ret < 0 || ret > 50) {
1693                 if (srcfd != -1)
1694                         close(srcfd);
1695                 close(destfd);
1696                 ERROR("Out of memory");
1697                 return -EINVAL;
1698         }
1699
1700         ret = mount(mntsrc, destbuf, fstype, flags, data);
1701         saved_errno = errno;
1702         if (srcfd != -1)
1703                 close(srcfd);
1704         close(destfd);
1705         if (ret < 0) {
1706                 errno = saved_errno;
1707                 SYSERROR("Failed to mount %s onto %s", src, dest);
1708                 return ret;
1709         }
1710
1711         return 0;
1712 }
1713
1714 /*
1715  * Mount a proc under @rootfs if proc self points to a pid other than
1716  * my own.  This is needed to have a known-good proc mount for setting
1717  * up LSMs both at container startup and attach.
1718  *
1719  * @rootfs : the rootfs where proc should be mounted
1720  *
1721  * Returns < 0 on failure, 0 if the correct proc was already mounted
1722  * and 1 if a new proc was mounted.
1723  *
1724  * NOTE: not to be called from inside the container namespace!
1725  */
1726 int lxc_mount_proc_if_needed(const char *rootfs)
1727 {
1728         char path[MAXPATHLEN];
1729         int link_to_pid, linklen, mypid, ret;
1730         char link[LXC_NUMSTRLEN64] = {0};
1731
1732         ret = snprintf(path, MAXPATHLEN, "%s/proc/self", rootfs);
1733         if (ret < 0 || ret >= MAXPATHLEN) {
1734                 SYSERROR("proc path name too long");
1735                 return -1;
1736         }
1737
1738         linklen = readlink(path, link, LXC_NUMSTRLEN64);
1739
1740         ret = snprintf(path, MAXPATHLEN, "%s/proc", rootfs);
1741         if (ret < 0 || ret >= MAXPATHLEN) {
1742                 SYSERROR("proc path name too long");
1743                 return -1;
1744         }
1745
1746         /* /proc not mounted */
1747         if (linklen < 0) {
1748                 if (mkdir(path, 0755) && errno != EEXIST)
1749                         return -1;
1750                 goto domount;
1751         } else if (linklen >= LXC_NUMSTRLEN64) {
1752                 link[linklen - 1] = '\0';
1753                 ERROR("readlink returned truncated content: \"%s\"", link);
1754                 return -1;
1755         }
1756
1757         mypid = getpid();
1758         INFO("I am %d, /proc/self points to \"%s\"", mypid, link);
1759
1760         if (lxc_safe_int(link, &link_to_pid) < 0)
1761                 return -1;
1762
1763         /* correct procfs is already mounted */
1764         if (link_to_pid == mypid)
1765                 return 0;
1766
1767         ret = umount2(path, MNT_DETACH);
1768         if (ret < 0)
1769                 WARN("failed to umount \"%s\" with MNT_DETACH", path);
1770
1771 domount:
1772         /* rootfs is NULL */
1773         if (!strcmp(rootfs, ""))
1774                 ret = mount("proc", path, "proc", 0, NULL);
1775         else
1776                 ret = safe_mount("proc", path, "proc", 0, NULL, rootfs);
1777         if (ret < 0)
1778                 return -1;
1779
1780         INFO("mounted /proc in container for security transition");
1781         return 1;
1782 }
1783
1784 int open_devnull(void)
1785 {
1786         int fd = open("/dev/null", O_RDWR);
1787
1788         if (fd < 0)
1789                 SYSERROR("Can't open /dev/null");
1790
1791         return fd;
1792 }
1793
1794 int set_stdfds(int fd)
1795 {
1796         int ret;
1797
1798         if (fd < 0)
1799                 return -1;
1800
1801         ret = dup2(fd, STDIN_FILENO);
1802         if (ret < 0)
1803                 return -1;
1804
1805         ret = dup2(fd, STDOUT_FILENO);
1806         if (ret < 0)
1807                 return -1;
1808
1809         ret = dup2(fd, STDERR_FILENO);
1810         if (ret < 0)
1811                 return -1;
1812
1813         return 0;
1814 }
1815
1816 int null_stdfds(void)
1817 {
1818         int ret = -1;
1819         int fd = open_devnull();
1820
1821         if (fd >= 0) {
1822                 ret = set_stdfds(fd);
1823                 close(fd);
1824         }
1825
1826         return ret;
1827 }
1828
1829 /*
1830  * Return the number of lines in file @fn, or -1 on error
1831  */
1832 int lxc_count_file_lines(const char *fn)
1833 {
1834         FILE *f;
1835         char *line = NULL;
1836         size_t sz = 0;
1837         int n = 0;
1838
1839         f = fopen_cloexec(fn, "r");
1840         if (!f)
1841                 return -1;
1842
1843         while (getline(&line, &sz, f) != -1) {
1844                 n++;
1845         }
1846         free(line);
1847         fclose(f);
1848         return n;
1849 }
1850
1851 void *lxc_strmmap(void *addr, size_t length, int prot, int flags, int fd,
1852                   off_t offset)
1853 {
1854         void *tmp = NULL, *overlap = NULL;
1855
1856         /* We establish an anonymous mapping that is one byte larger than the
1857          * underlying file. The pages handed to us are zero filled. */
1858         tmp = mmap(addr, length + 1, PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1859         if (tmp == MAP_FAILED)
1860                 return tmp;
1861
1862         /* Now we establish a fixed-address mapping starting at the address we
1863          * received from our anonymous mapping and replace all bytes excluding
1864          * the additional \0-byte with the file. This allows us to use normal
1865          * string-handling functions. */
1866         overlap = mmap(tmp, length, prot, MAP_FIXED | flags, fd, offset);
1867         if (overlap == MAP_FAILED)
1868                 munmap(tmp, length + 1);
1869
1870         return overlap;
1871 }
1872
1873 int lxc_strmunmap(void *addr, size_t length)
1874 {
1875         return munmap(addr, length + 1);
1876 }
1877
1878 /* Check whether a signal is blocked by a process. */
1879 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
1880 #define __PROC_STATUS_LEN (5 + (LXC_NUMSTRLEN64) + 7 + 1)
1881 bool task_blocking_signal(pid_t pid, int signal)
1882 {
1883         bool bret = false;
1884         char *line = NULL;
1885         long unsigned int sigblk = 0;
1886         size_t n = 0;
1887         int ret;
1888         FILE *f;
1889
1890         char status[__PROC_STATUS_LEN];
1891
1892         ret = snprintf(status, __PROC_STATUS_LEN, "/proc/%d/status", pid);
1893         if (ret < 0 || ret >= __PROC_STATUS_LEN)
1894                 return bret;
1895
1896         f = fopen(status, "r");
1897         if (!f)
1898                 return bret;
1899
1900         while (getline(&line, &n, f) != -1) {
1901                 if (strncmp(line, "SigBlk:\t", 8))
1902                         continue;
1903
1904                 if (sscanf(line + 8, "%lx", &sigblk) != 1)
1905                         goto out;
1906         }
1907
1908         if (sigblk & (1LU << (signal - 1)))
1909                 bret = true;
1910
1911 out:
1912         free(line);
1913         fclose(f);
1914         return bret;
1915 }
1916
1917 static int lxc_append_null_to_list(void ***list)
1918 {
1919         int newentry = 0;
1920         void **tmp;
1921
1922         if (*list)
1923                 for (; (*list)[newentry]; newentry++) {
1924                         ;
1925                 }
1926
1927         tmp = realloc(*list, (newentry + 2) * sizeof(void **));
1928         if (!tmp)
1929                 return -1;
1930
1931         *list = tmp;
1932         (*list)[newentry + 1] = NULL;
1933
1934         return newentry;
1935 }
1936
1937 int lxc_append_string(char ***list, char *entry)
1938 {
1939         char *copy;
1940         int newentry;
1941
1942         newentry = lxc_append_null_to_list((void ***)list);
1943         if (newentry < 0)
1944                 return -1;
1945
1946         copy = strdup(entry);
1947         if (!copy)
1948                 return -1;
1949
1950         (*list)[newentry] = copy;
1951
1952         return 0;
1953 }
1954
1955 int lxc_preserve_ns(const int pid, const char *ns)
1956 {
1957         int ret;
1958 /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */
1959 #define __NS_PATH_LEN 50
1960         char path[__NS_PATH_LEN];
1961
1962         /* This way we can use this function to also check whether namespaces
1963          * are supported by the kernel by passing in the NULL or the empty
1964          * string.
1965          */
1966         ret = snprintf(path, __NS_PATH_LEN, "/proc/%d/ns%s%s", pid,
1967                        !ns || strcmp(ns, "") == 0 ? "" : "/",
1968                        !ns || strcmp(ns, "") == 0 ? "" : ns);
1969         if (ret < 0 || (size_t)ret >= __NS_PATH_LEN)
1970                 return -1;
1971
1972         return open(path, O_RDONLY | O_CLOEXEC);
1973 }
1974
1975 int lxc_safe_uint(const char *numstr, unsigned int *converted)
1976 {
1977         char *err = NULL;
1978         unsigned long int uli;
1979
1980         while (isspace(*numstr))
1981                 numstr++;
1982
1983         if (*numstr == '-')
1984                 return -EINVAL;
1985
1986         errno = 0;
1987         uli = strtoul(numstr, &err, 0);
1988         if (errno == ERANGE && uli == ULONG_MAX)
1989                 return -ERANGE;
1990
1991         if (err == numstr || *err != '\0')
1992                 return -EINVAL;
1993
1994         if (uli > UINT_MAX)
1995                 return -ERANGE;
1996
1997         *converted = (unsigned int)uli;
1998         return 0;
1999 }
2000
2001 int lxc_safe_ulong(const char *numstr, unsigned long *converted)
2002 {
2003         char *err = NULL;
2004         unsigned long int uli;
2005
2006         while (isspace(*numstr))
2007                 numstr++;
2008
2009         if (*numstr == '-')
2010                 return -EINVAL;
2011
2012         errno = 0;
2013         uli = strtoul(numstr, &err, 0);
2014         if (errno == ERANGE && uli == ULONG_MAX)
2015                 return -ERANGE;
2016
2017         if (err == numstr || *err != '\0')
2018                 return -EINVAL;
2019
2020         *converted = uli;
2021         return 0;
2022 }
2023
2024 int lxc_safe_int(const char *numstr, int *converted)
2025 {
2026         char *err = NULL;
2027         signed long int sli;
2028
2029         errno = 0;
2030         sli = strtol(numstr, &err, 0);
2031         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
2032                 return -ERANGE;
2033
2034         if (errno != 0 && sli == 0)
2035                 return -EINVAL;
2036
2037         if (err == numstr || *err != '\0')
2038                 return -EINVAL;
2039
2040         if (sli > INT_MAX || sli < INT_MIN)
2041                 return -ERANGE;
2042
2043         *converted = (int)sli;
2044         return 0;
2045 }
2046
2047 int lxc_safe_long(const char *numstr, long int *converted)
2048 {
2049         char *err = NULL;
2050         signed long int sli;
2051
2052         errno = 0;
2053         sli = strtol(numstr, &err, 0);
2054         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
2055                 return -ERANGE;
2056
2057         if (errno != 0 && sli == 0)
2058                 return -EINVAL;
2059
2060         if (err == numstr || *err != '\0')
2061                 return -EINVAL;
2062
2063         *converted = sli;
2064         return 0;
2065 }
2066
2067 int lxc_switch_uid_gid(uid_t uid, gid_t gid)
2068 {
2069         if (setgid(gid) < 0) {
2070                 SYSERROR("Failed to switch to gid %d.", gid);
2071                 return -errno;
2072         }
2073         NOTICE("Switched to gid %d.", gid);
2074
2075         if (setuid(uid) < 0) {
2076                 SYSERROR("Failed to switch to uid %d.", uid);
2077                 return -errno;
2078         }
2079         NOTICE("Switched to uid %d.", uid);
2080
2081         return 0;
2082 }
2083
2084 /* Simple covenience function which enables uniform logging. */
2085 int lxc_setgroups(int size, gid_t list[])
2086 {
2087         if (setgroups(size, list) < 0) {
2088                 SYSERROR("Failed to setgroups().");
2089                 return -errno;
2090         }
2091         NOTICE("Dropped additional groups.");
2092
2093         return 0;
2094 }
2095
2096 static int lxc_get_unused_loop_dev_legacy(char *loop_name)
2097 {
2098         struct dirent *dp;
2099         struct loop_info64 lo64;
2100         DIR *dir;
2101         int dfd = -1, fd = -1, ret = -1;
2102
2103         dir = opendir("/dev");
2104         if (!dir)
2105                 return -1;
2106
2107         while ((dp = readdir(dir))) {
2108                 if (!dp)
2109                         break;
2110
2111                 if (strncmp(dp->d_name, "loop", 4) != 0)
2112                         continue;
2113
2114                 dfd = dirfd(dir);
2115                 if (dfd < 0)
2116                         continue;
2117
2118                 fd = openat(dfd, dp->d_name, O_RDWR);
2119                 if (fd < 0)
2120                         continue;
2121
2122                 ret = ioctl(fd, LOOP_GET_STATUS64, &lo64);
2123                 if (ret < 0) {
2124                         if (ioctl(fd, LOOP_GET_STATUS64, &lo64) == 0 ||
2125                             errno != ENXIO) {
2126                                 close(fd);
2127                                 fd = -1;
2128                                 continue;
2129                         }
2130                 }
2131
2132                 ret = snprintf(loop_name, LO_NAME_SIZE, "/dev/%s", dp->d_name);
2133                 if (ret < 0 || ret >= LO_NAME_SIZE) {
2134                         close(fd);
2135                         fd = -1;
2136                         continue;
2137                 }
2138
2139                 break;
2140         }
2141
2142         closedir(dir);
2143
2144         if (fd < 0)
2145                 return -1;
2146
2147         return fd;
2148 }
2149
2150 static int lxc_get_unused_loop_dev(char *name_loop)
2151 {
2152         int loop_nr, ret;
2153         int fd_ctl = -1, fd_tmp = -1;
2154
2155         fd_ctl = open("/dev/loop-control", O_RDWR | O_CLOEXEC);
2156         if (fd_ctl < 0)
2157                 return -ENODEV;
2158
2159         loop_nr = ioctl(fd_ctl, LOOP_CTL_GET_FREE);
2160         if (loop_nr < 0)
2161                 goto on_error;
2162
2163         ret = snprintf(name_loop, LO_NAME_SIZE, "/dev/loop%d", loop_nr);
2164         if (ret < 0 || ret >= LO_NAME_SIZE)
2165                 goto on_error;
2166
2167         fd_tmp = open(name_loop, O_RDWR | O_CLOEXEC);
2168         if (fd_tmp < 0)
2169                 goto on_error;
2170
2171 on_error:
2172         close(fd_ctl);
2173         return fd_tmp;
2174 }
2175
2176 int lxc_prepare_loop_dev(const char *source, char *loop_dev, int flags)
2177 {
2178         int ret;
2179         struct loop_info64 lo64;
2180         int fd_img = -1, fret = -1, fd_loop = -1;
2181
2182         fd_loop = lxc_get_unused_loop_dev(loop_dev);
2183         if (fd_loop < 0) {
2184                 if (fd_loop == -ENODEV)
2185                         fd_loop = lxc_get_unused_loop_dev_legacy(loop_dev);
2186                 else
2187                         goto on_error;
2188         }
2189
2190         fd_img = open(source, O_RDWR | O_CLOEXEC);
2191         if (fd_img < 0)
2192                 goto on_error;
2193
2194         ret = ioctl(fd_loop, LOOP_SET_FD, fd_img);
2195         if (ret < 0)
2196                 goto on_error;
2197
2198         memset(&lo64, 0, sizeof(lo64));
2199         lo64.lo_flags = flags;
2200
2201         ret = ioctl(fd_loop, LOOP_SET_STATUS64, &lo64);
2202         if (ret < 0)
2203                 goto on_error;
2204
2205         fret = 0;
2206
2207 on_error:
2208         if (fd_img >= 0)
2209                 close(fd_img);
2210
2211         if (fret < 0 && fd_loop >= 0) {
2212                 close(fd_loop);
2213                 fd_loop = -1;
2214         }
2215
2216         return fd_loop;
2217 }
2218
2219 int lxc_unstack_mountpoint(const char *path, bool lazy)
2220 {
2221         int ret;
2222         int umounts = 0;
2223
2224 pop_stack:
2225         ret = umount2(path, lazy ? MNT_DETACH : 0);
2226         if (ret < 0) {
2227                 /* We consider anything else than EINVAL deadly to prevent going
2228                  * into an infinite loop. (The other alternative is constantly
2229                  * parsing /proc/self/mountinfo which is yucky and probably
2230                  * racy.)
2231                  */
2232                 if (errno != EINVAL)
2233                         return -errno;
2234         } else {
2235                 /* Just stop counting when this happens. That'd just be so
2236                  * stupid that we won't even bother trying to report back the
2237                  * correct value anymore.
2238                  */
2239                 if (umounts != INT_MAX)
2240                         umounts++;
2241                 /* We succeeded in umounting. Make sure that there's no other
2242                  * mountpoint stacked underneath.
2243                  */
2244                 goto pop_stack;
2245         }
2246
2247         return umounts;
2248 }
2249
2250 int run_command(char *buf, size_t buf_size, int (*child_fn)(void *), void *args)
2251 {
2252         pid_t child;
2253         int ret, fret, pipefd[2];
2254         ssize_t bytes;
2255
2256         /* Make sure our callers do not receive unitialized memory. */
2257         if (buf_size > 0 && buf)
2258                 buf[0] = '\0';
2259
2260         if (pipe(pipefd) < 0) {
2261                 SYSERROR("failed to create pipe");
2262                 return -1;
2263         }
2264
2265         child = fork();
2266         if (child < 0) {
2267                 close(pipefd[0]);
2268                 close(pipefd[1]);
2269                 SYSERROR("failed to create new process");
2270                 return -1;
2271         }
2272
2273         if (child == 0) {
2274                 /* Close the read-end of the pipe. */
2275                 close(pipefd[0]);
2276
2277                 /* Redirect std{err,out} to write-end of the
2278                  * pipe.
2279                  */
2280                 ret = dup2(pipefd[1], STDOUT_FILENO);
2281                 if (ret >= 0)
2282                         ret = dup2(pipefd[1], STDERR_FILENO);
2283
2284                 /* Close the write-end of the pipe. */
2285                 close(pipefd[1]);
2286
2287                 if (ret < 0) {
2288                         SYSERROR("failed to duplicate std{err,out} file descriptor");
2289                         exit(EXIT_FAILURE);
2290                 }
2291
2292                 /* Does not return. */
2293                 child_fn(args);
2294                 ERROR("failed to exec command");
2295                 exit(EXIT_FAILURE);
2296         }
2297
2298         /* close the write-end of the pipe */
2299         close(pipefd[1]);
2300
2301         if (buf && buf_size > 0) {
2302                 bytes = read(pipefd[0], buf, buf_size - 1);
2303                 if (bytes > 0)
2304                         buf[bytes - 1] = '\0';
2305         }
2306
2307         fret = wait_for_pid(child);
2308         /* close the read-end of the pipe */
2309         close(pipefd[0]);
2310
2311         return fret;
2312 }
2313
2314 char *must_make_path(const char *first, ...)
2315 {
2316         va_list args;
2317         char *cur, *dest;
2318         size_t full_len = strlen(first);
2319
2320         dest = must_copy_string(first);
2321
2322         va_start(args, first);
2323         while ((cur = va_arg(args, char *)) != NULL) {
2324                 full_len += strlen(cur);
2325                 if (cur[0] != '/')
2326                         full_len++;
2327                 dest = must_realloc(dest, full_len + 1);
2328                 if (cur[0] != '/')
2329                         strcat(dest, "/");
2330                 strcat(dest, cur);
2331         }
2332         va_end(args);
2333
2334         return dest;
2335 }
2336
2337 char *must_copy_string(const char *entry)
2338 {
2339         char *ret;
2340
2341         if (!entry)
2342                 return NULL;
2343         do {
2344                 ret = strdup(entry);
2345         } while (!ret);
2346
2347         return ret;
2348 }
2349
2350 void *must_realloc(void *orig, size_t sz)
2351 {
2352         void *ret;
2353
2354         do {
2355                 ret = realloc(orig, sz);
2356         } while (!ret);
2357
2358         return ret;
2359 }
2360
2361 bool is_fs_type(const struct statfs *fs, fs_type_magic magic_val)
2362 {
2363         return (fs->f_type == (fs_type_magic)magic_val);
2364 }
2365
2366 bool has_fs_type(const char *path, fs_type_magic magic_val)
2367 {
2368         bool has_type;
2369         int ret;
2370         struct statfs sb;
2371
2372         ret = statfs(path, &sb);
2373         if (ret < 0)
2374                 return false;
2375
2376         has_type = is_fs_type(&sb, magic_val);
2377         if (!has_type && magic_val == RAMFS_MAGIC)
2378                 WARN("When the ramfs it a tmpfs statfs() might report tmpfs");
2379
2380         return has_type;
2381 }
2382
2383 bool lxc_nic_exists(char *nic)
2384 {
2385 #define __LXC_SYS_CLASS_NET_LEN 15 + IFNAMSIZ + 1
2386         char path[__LXC_SYS_CLASS_NET_LEN];
2387         int ret;
2388         struct stat sb;
2389
2390         if (!strcmp(nic, "none"))
2391                 return true;
2392
2393         ret = snprintf(path, __LXC_SYS_CLASS_NET_LEN, "/sys/class/net/%s", nic);
2394         if (ret < 0 || (size_t)ret >= __LXC_SYS_CLASS_NET_LEN)
2395                 return false;
2396
2397         ret = stat(path, &sb);
2398         if (ret < 0)
2399                 return false;
2400
2401         return true;
2402 }