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 <pthread.h>
  35 #include <stddef.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <string.h>
  39 #include <unistd.h>
  40 #include <sys/mman.h>
  41 #include <sys/mount.h>
  42 #include <sys/param.h>
  43 #include <sys/prctl.h>
  44 #include <sys/stat.h>
  45 #include <sys/types.h>
  46 #include <sys/wait.h>
  47
  48 #include "log.h"
  49 #include "lxclock.h"
  50 #include "namespace.h"
  51 #include "parse.h"
  52 #include "utils.h"
  53
  54 #ifndef O_PATH
  55 #define O_PATH      010000000
  56 #endif
  57
  58 #ifndef O_NOFOLLOW
  59 #define O_NOFOLLOW  00400000
  60 #endif
  61
  62 lxc_log_define(lxc_utils, lxc);
  63
  64 /*
  65  * if path is btrfs, tries to remove it and any subvolumes beneath it
  66  */
  67 extern bool btrfs_try_remove_subvol(const char *path);
  68
  69 static int _recursive_rmdir(const char *dirname, dev_t pdev,
  70                             const char *exclude, int level, bool onedev)
  71 {
  72         struct dirent *direntp;
  73         DIR *dir;
  74         int ret, failed=0;
  75         char pathname[MAXPATHLEN];
  76         bool hadexclude = false;
  77
  78         dir = opendir(dirname);
  79         if (!dir) {
  80                 ERROR("failed to open %s", dirname);
  81                 return -1;
  82         }
  83
  84         while ((direntp = readdir(dir))) {
  85                 struct stat mystat;
  86                 int rc;
  87
  88                 if (!direntp)
  89                         break;
  90
  91                 if (!strcmp(direntp->d_name, ".") ||
  92                     !strcmp(direntp->d_name, ".."))
  93                         continue;
  94
  95                 rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name);
  96                 if (rc < 0 || rc >= MAXPATHLEN) {
  97                         ERROR("pathname too long");
  98                         failed=1;
  99                         continue;
 100                 }
 101
 102                 if (!level && exclude && !strcmp(direntp->d_name, exclude)) {
 103                         ret = rmdir(pathname);
 104                         if (ret < 0) {
 105                                 switch(errno) {
 106                                 case ENOTEMPTY:
 107                                         INFO("Not deleting snapshot %s", pathname);
 108                                         hadexclude = true;
 109                                         break;
 110                                 case ENOTDIR:
 111                                         ret = unlink(pathname);
 112                                         if (ret)
 113                                                 INFO("Failed to remove %s", pathname);
 114                                         break;
 115                                 default:
 116                                         SYSERROR("Failed to rmdir %s", pathname);
 117                                         failed = 1;
 118                                         break;
 119                                 }
 120                         }
 121                         continue;
 122                 }
 123
 124                 ret = lstat(pathname, &mystat);
 125                 if (ret) {
 126                         ERROR("Failed to stat %s", pathname);
 127                         failed = 1;
 128                         continue;
 129                 }
 130                 if (onedev && mystat.st_dev != pdev) {
 131                         /* TODO should we be checking /proc/self/mountinfo for
 132                          * pathname and not doing this if found? */
 133                         if (btrfs_try_remove_subvol(pathname))
 134                                 INFO("Removed btrfs subvolume at %s\n", pathname);
 135                         continue;
 136                 }
 137                 if (S_ISDIR(mystat.st_mode)) {
 138                         if (_recursive_rmdir(pathname, pdev, exclude, level+1, onedev) < 0)
 139                                 failed=1;
 140                 } else {
 141                         if (unlink(pathname) < 0) {
 142                                 SYSERROR("Failed to delete %s", pathname);
 143                                 failed=1;
 144                         }
 145                 }
 146         }
 147
 148         if (rmdir(dirname) < 0 && !btrfs_try_remove_subvol(dirname) && !hadexclude) {
 149                 ERROR("Failed to delete %s", dirname);
 150                 failed=1;
 151         }
 152
 153         ret = closedir(dir);
 154         if (ret) {
 155                 ERROR("Failed to close directory %s", dirname);
 156                 failed=1;
 157         }
 158
 159         return failed ? -1 : 0;
 160 }
 161
 162 /* We have two different magic values for overlayfs, yay. */
 163 #ifndef OVERLAYFS_SUPER_MAGIC
 164 #define OVERLAYFS_SUPER_MAGIC 0x794c764f
 165 #endif
 166
 167 #ifndef OVERLAY_SUPER_MAGIC
 168 #define OVERLAY_SUPER_MAGIC 0x794c7630
 169 #endif
 170
 171 /* In overlayfs, st_dev is unreliable. So on overlayfs we don't do the
 172  * lxc_rmdir_onedev()
 173  */
 174 static bool is_native_overlayfs(const char *path)
 175 {
 176         if (has_fs_type(path, OVERLAY_SUPER_MAGIC) ||
 177             has_fs_type(path, OVERLAYFS_SUPER_MAGIC))
 178                 return true;
 179
 180         return false;
 181 }
 182
 183 /* returns 0 on success, -1 if there were any failures */
 184 extern int lxc_rmdir_onedev(const char *path, const char *exclude)
 185 {
 186         struct stat mystat;
 187         bool onedev = true;
 188
 189         if (is_native_overlayfs(path))
 190                 onedev = false;
 191
 192         if (lstat(path, &mystat) < 0) {
 193                 if (errno == ENOENT)
 194                         return 0;
 195
 196                 ERROR("Failed to stat %s", path);
 197                 return -1;
 198         }
 199
 200         return _recursive_rmdir(path, mystat.st_dev, exclude, 0, onedev);
 201 }
 202
 203 /* borrowed from iproute2 */
 204 extern int get_u16(unsigned short *val, const char *arg, int base)
 205 {
 206         unsigned long res;
 207         char *ptr;
 208
 209         if (!arg || !*arg)
 210                 return -1;
 211
 212         errno = 0;
 213         res = strtoul(arg, &ptr, base);
 214         if (!ptr || ptr == arg || *ptr || res > 0xFFFF || errno != 0)
 215                 return -1;
 216
 217         *val = res;
 218
 219         return 0;
 220 }
 221
 222 extern int mkdir_p(const char *dir, mode_t mode)
 223 {
 224         const char *tmp = dir;
 225         const char *orig = dir;
 226         char *makeme;
 227
 228         do {
 229                 dir = tmp + strspn(tmp, "/");
 230                 tmp = dir + strcspn(dir, "/");
 231                 makeme = strndup(orig, dir - orig);
 232                 if (*makeme) {
 233                         if (mkdir(makeme, mode) && errno != EEXIST) {
 234                                 SYSERROR("failed to create directory '%s'", makeme);
 235                                 free(makeme);
 236                                 return -1;
 237                         }
 238                 }
 239                 free(makeme);
 240         } while(tmp != dir);
 241
 242         return 0;
 243 }
 244
 245 char *get_rundir()
 246 {
 247         char *rundir;
 248         const char *homedir;
 249         struct stat sb;
 250
 251         if (stat(RUNTIME_PATH, &sb) < 0) {
 252                 return NULL;
 253         }
 254
 255         if (geteuid() == sb.st_uid || getegid() == sb.st_gid) {
 256                 rundir = strdup(RUNTIME_PATH);
 257                 return rundir;
 258         }
 259
 260         rundir = getenv("XDG_RUNTIME_DIR");
 261         if (rundir) {
 262                 rundir = strdup(rundir);
 263                 return rundir;
 264         }
 265
 266         INFO("XDG_RUNTIME_DIR isn't set in the environment.");
 267         homedir = getenv("HOME");
 268         if (!homedir) {
 269                 ERROR("HOME isn't set in the environment.");
 270                 return NULL;
 271         }
 272
 273         rundir = malloc(sizeof(char) * (17 + strlen(homedir)));
 274         sprintf(rundir, "%s/.cache/lxc/run/", homedir);
 275
 276         return rundir;
 277 }
 278
 279 int wait_for_pid(pid_t pid)
 280 {
 281         int status, ret;
 282
 283 again:
 284         ret = waitpid(pid, &status, 0);
 285         if (ret == -1) {
 286                 if (errno == EINTR)
 287                         goto again;
 288                 return -1;
 289         }
 290         if (ret != pid)
 291                 goto again;
 292         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0)
 293                 return -1;
 294         return 0;
 295 }
 296
 297 int lxc_wait_for_pid_status(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         return status;
 311 }
 312
 313 ssize_t lxc_write_nointr(int fd, const void* buf, size_t count)
 314 {
 315         ssize_t ret;
 316 again:
 317         ret = write(fd, buf, count);
 318         if (ret < 0 && errno == EINTR)
 319                 goto again;
 320         return ret;
 321 }
 322
 323 ssize_t lxc_read_nointr(int fd, void* buf, size_t count)
 324 {
 325         ssize_t ret;
 326 again:
 327         ret = read(fd, buf, count);
 328         if (ret < 0 && errno == EINTR)
 329                 goto again;
 330         return ret;
 331 }
 332
 333 ssize_t lxc_read_nointr_expect(int fd, void* buf, size_t count, const void* expected_buf)
 334 {
 335         ssize_t ret;
 336         ret = lxc_read_nointr(fd, buf, count);
 337         if (ret <= 0)
 338                 return ret;
 339         if ((size_t)ret != count)
 340                 return -1;
 341         if (expected_buf && memcmp(buf, expected_buf, count) != 0) {
 342                 errno = EINVAL;
 343                 return -1;
 344         }
 345         return ret;
 346 }
 347
 348 #if HAVE_LIBGNUTLS
 349 #include <gnutls/gnutls.h>
 350 #include <gnutls/crypto.h>
 351
 352 __attribute__((constructor))
 353 static void gnutls_lxc_init(void)
 354 {
 355         gnutls_global_init();
 356 }
 357
 358 int sha1sum_file(char *fnam, unsigned char *digest)
 359 {
 360         char *buf;
 361         int ret;
 362         FILE *f;
 363         long flen;
 364
 365         if (!fnam)
 366                 return -1;
 367         f = fopen_cloexec(fnam, "r");
 368         if (!f) {
 369                 SYSERROR("Error opening template");
 370                 return -1;
 371         }
 372         if (fseek(f, 0, SEEK_END) < 0) {
 373                 SYSERROR("Error seeking to end of template");
 374                 fclose(f);
 375                 return -1;
 376         }
 377         if ((flen = ftell(f)) < 0) {
 378                 SYSERROR("Error telling size of template");
 379                 fclose(f);
 380                 return -1;
 381         }
 382         if (fseek(f, 0, SEEK_SET) < 0) {
 383                 SYSERROR("Error seeking to start of template");
 384                 fclose(f);
 385                 return -1;
 386         }
 387         if ((buf = malloc(flen+1)) == NULL) {
 388                 SYSERROR("Out of memory");
 389                 fclose(f);
 390                 return -1;
 391         }
 392         if (fread(buf, 1, flen, f) != flen) {
 393                 SYSERROR("Failure reading template");
 394                 free(buf);
 395                 fclose(f);
 396                 return -1;
 397         }
 398         if (fclose(f) < 0) {
 399                 SYSERROR("Failre closing template");
 400                 free(buf);
 401                 return -1;
 402         }
 403         buf[flen] = '\0';
 404         ret = gnutls_hash_fast(GNUTLS_DIG_SHA1, buf, flen, (void *)digest);
 405         free(buf);
 406         return ret;
 407 }
 408 #endif
 409
 410 char** lxc_va_arg_list_to_argv(va_list ap, size_t skip, int do_strdup)
 411 {
 412         va_list ap2;
 413         size_t count = 1 + skip;
 414         char **result;
 415
 416         /* first determine size of argument list, we don't want to reallocate
 417          * constantly...
 418          */
 419         va_copy(ap2, ap);
 420         while (1) {
 421                 char* arg = va_arg(ap2, char*);
 422                 if (!arg)
 423                         break;
 424                 count++;
 425         }
 426         va_end(ap2);
 427
 428         result = calloc(count, sizeof(char*));
 429         if (!result)
 430                 return NULL;
 431         count = skip;
 432         while (1) {
 433                 char* arg = va_arg(ap, char*);
 434                 if (!arg)
 435                         break;
 436                 arg = do_strdup ? strdup(arg) : arg;
 437                 if (!arg)
 438                         goto oom;
 439                 result[count++] = arg;
 440         }
 441
 442         /* calloc has already set last element to NULL*/
 443         return result;
 444
 445 oom:
 446         free(result);
 447         return NULL;
 448 }
 449
 450 const char** lxc_va_arg_list_to_argv_const(va_list ap, size_t skip)
 451 {
 452         return (const char**)lxc_va_arg_list_to_argv(ap, skip, 0);
 453 }
 454
 455 struct lxc_popen_FILE *lxc_popen(const char *command)
 456 {
 457         int ret;
 458         int pipe_fds[2];
 459         pid_t child_pid;
 460         struct lxc_popen_FILE *fp = NULL;
 461
 462         ret = pipe2(pipe_fds, O_CLOEXEC);
 463         if (ret < 0)
 464                 return NULL;
 465
 466         child_pid = fork();
 467         if (child_pid < 0)
 468                 goto on_error;
 469
 470         if (!child_pid) {
 471                 sigset_t mask;
 472
 473                 close(pipe_fds[0]);
 474
 475                 /* duplicate stdout */
 476                 if (pipe_fds[1] != STDOUT_FILENO)
 477                         ret = dup2(pipe_fds[1], STDOUT_FILENO);
 478                 else
 479                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 480                 if (ret < 0) {
 481                         close(pipe_fds[1]);
 482                         _exit(EXIT_FAILURE);
 483                 }
 484
 485                 /* duplicate stderr */
 486                 if (pipe_fds[1] != STDERR_FILENO)
 487                         ret = dup2(pipe_fds[1], STDERR_FILENO);
 488                 else
 489                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 490                 close(pipe_fds[1]);
 491                 if (ret < 0)
 492                         _exit(EXIT_FAILURE);
 493
 494                 /* unblock all signals */
 495                 ret = sigfillset(&mask);
 496                 if (ret < 0)
 497                         _exit(EXIT_FAILURE);
 498
 499                 ret = pthread_sigmask(SIG_UNBLOCK, &mask, NULL);
 500                 if (ret < 0)
 501                         _exit(EXIT_FAILURE);
 502
 503                 execl("/bin/sh", "sh", "-c", command, (char *)NULL);
 504                 _exit(127);
 505         }
 506
 507         close(pipe_fds[1]);
 508         pipe_fds[1] = -1;
 509
 510         fp = malloc(sizeof(*fp));
 511         if (!fp)
 512                 goto on_error;
 513         memset(fp, 0, sizeof(*fp));
 514
 515         fp->child_pid = child_pid;
 516         fp->pipe = pipe_fds[0];
 517
 518         /* From now on, closing fp->f will also close fp->pipe. So only ever
 519          * call fclose(fp->f).
 520          */
 521         fp->f = fdopen(pipe_fds[0], "r");
 522         if (!fp->f)
 523                 goto on_error;
 524
 525         return fp;
 526
 527 on_error:
 528         /* We can only close pipe_fds[0] if fdopen() didn't succeed or wasn't
 529          * called yet. Otherwise the fd belongs to the file opened by fdopen()
 530          * since it isn't dup()ed.
 531          */
 532         if (fp && !fp->f && pipe_fds[0] >= 0)
 533                 close(pipe_fds[0]);
 534
 535         if (pipe_fds[1] >= 0)
 536                 close(pipe_fds[1]);
 537
 538         if (fp && fp->f)
 539                 fclose(fp->f);
 540
 541         if (fp)
 542                 free(fp);
 543
 544         return NULL;
 545 }
 546
 547 int lxc_pclose(struct lxc_popen_FILE *fp)
 548 {
 549         pid_t wait_pid;
 550         int wstatus = 0;
 551
 552         if (!fp)
 553                 return -1;
 554
 555         do {
 556                 wait_pid = waitpid(fp->child_pid, &wstatus, 0);
 557         } while (wait_pid < 0 && errno == EINTR);
 558
 559         fclose(fp->f);
 560         free(fp);
 561
 562         if (wait_pid < 0)
 563                 return -1;
 564
 565         return wstatus;
 566 }
 567
 568 char *lxc_string_replace(const char *needle, const char *replacement, const char *haystack)
 569 {
 570         ssize_t len = -1, saved_len = -1;
 571         char *result = NULL;
 572         size_t replacement_len = strlen(replacement);
 573         size_t needle_len = strlen(needle);
 574
 575         /* should be executed exactly twice */
 576         while (len == -1 || result == NULL) {
 577                 char *p;
 578                 char *last_p;
 579                 ssize_t part_len;
 580
 581                 if (len != -1) {
 582                         result = calloc(1, len + 1);
 583                         if (!result)
 584                                 return NULL;
 585                         saved_len = len;
 586                 }
 587
 588                 len = 0;
 589
 590                 for (last_p = (char *)haystack, p = strstr(last_p, needle); p; last_p = p, p = strstr(last_p, needle)) {
 591                         part_len = (ssize_t)(p - last_p);
 592                         if (result && part_len > 0)
 593                                 memcpy(&result[len], last_p, part_len);
 594                         len += part_len;
 595                         if (result && replacement_len > 0)
 596                                 memcpy(&result[len], replacement, replacement_len);
 597                         len += replacement_len;
 598                         p += needle_len;
 599                 }
 600                 part_len = strlen(last_p);
 601                 if (result && part_len > 0)
 602                         memcpy(&result[len], last_p, part_len);
 603                 len += part_len;
 604         }
 605
 606         /* make sure we did the same thing twice,
 607          * once for calculating length, the other
 608          * time for copying data */
 609         if (saved_len != len) {
 610                 free(result);
 611                 return NULL;
 612         }
 613         /* make sure we didn't overwrite any buffer,
 614          * due to calloc the string should be 0-terminated */
 615         if (result[len] != '\0') {
 616                 free(result);
 617                 return NULL;
 618         }
 619
 620         return result;
 621 }
 622
 623 bool lxc_string_in_array(const char *needle, const char **haystack)
 624 {
 625         for (; haystack && *haystack; haystack++)
 626                 if (!strcmp(needle, *haystack))
 627                         return true;
 628         return false;
 629 }
 630
 631 char *lxc_string_join(const char *sep, const char **parts, bool use_as_prefix)
 632 {
 633         char *result;
 634         char **p;
 635         size_t sep_len = strlen(sep);
 636         size_t result_len = use_as_prefix * sep_len;
 637
 638         /* calculate new string length */
 639         for (p = (char **)parts; *p; p++)
 640                 result_len += (p > (char **)parts) * sep_len + strlen(*p);
 641
 642         result = calloc(result_len + 1, 1);
 643         if (!result)
 644                 return NULL;
 645
 646         if (use_as_prefix)
 647                 strcpy(result, sep);
 648         for (p = (char **)parts; *p; p++) {
 649                 if (p > (char **)parts)
 650                         strcat(result, sep);
 651                 strcat(result, *p);
 652         }
 653
 654         return result;
 655 }
 656
 657 char **lxc_normalize_path(const char *path)
 658 {
 659         char **components;
 660         char **p;
 661         size_t components_len = 0;
 662         size_t pos = 0;
 663
 664         components = lxc_string_split(path, '/');
 665         if (!components)
 666                 return NULL;
 667         for (p = components; *p; p++)
 668                 components_len++;
 669
 670         /* resolve '.' and '..' */
 671         for (pos = 0; pos < components_len; ) {
 672                 if (!strcmp(components[pos], ".") || (!strcmp(components[pos], "..") && pos == 0)) {
 673                         /* eat this element */
 674                         free(components[pos]);
 675                         memmove(&components[pos], &components[pos+1], sizeof(char *) * (components_len - pos));
 676                         components_len--;
 677                 } else if (!strcmp(components[pos], "..")) {
 678                         /* eat this and the previous element */
 679                         free(components[pos - 1]);
 680                         free(components[pos]);
 681                         memmove(&components[pos-1], &components[pos+1], sizeof(char *) * (components_len - pos));
 682                         components_len -= 2;
 683                         pos--;
 684                 } else {
 685                         pos++;
 686                 }
 687         }
 688
 689         return components;
 690 }
 691
 692 char *lxc_deslashify(const char *path)
 693 {
 694         char *dup, *p;
 695         char **parts = NULL;
 696         size_t n, len;
 697
 698         dup = strdup(path);
 699         if (!dup)
 700                 return NULL;
 701
 702         parts = lxc_normalize_path(dup);
 703         if (!parts) {
 704                 free(dup);
 705                 return NULL;
 706         }
 707
 708         /* We'll end up here if path == "///" or path == "". */
 709         if (!*parts) {
 710                 len = strlen(dup);
 711                 if (!len) {
 712                         lxc_free_array((void **)parts, free);
 713                         return dup;
 714                 }
 715                 n = strcspn(dup, "/");
 716                 if (n == len) {
 717                         free(dup);
 718                         lxc_free_array((void **)parts, free);
 719
 720                         p = strdup("/");
 721                         if (!p)
 722                                 return NULL;
 723
 724                         return p;
 725                 }
 726         }
 727
 728         p = lxc_string_join("/", (const char **)parts, *dup == '/');
 729         free(dup);
 730         lxc_free_array((void **)parts, free);
 731         return p;
 732 }
 733
 734 char *lxc_append_paths(const char *first, const char *second)
 735 {
 736         int ret;
 737         size_t len;
 738         char *result = NULL;
 739         const char *pattern = "%s%s";
 740
 741         len = strlen(first) + strlen(second) + 1;
 742         if (second[0] != '/') {
 743                 len += 1;
 744                 pattern = "%s/%s";
 745         }
 746
 747         result = calloc(1, len);
 748         if (!result)
 749                 return NULL;
 750
 751         ret = snprintf(result, len, pattern, first, second);
 752         if (ret < 0 || (size_t)ret >= len) {
 753                 free(result);
 754                 return NULL;
 755         }
 756
 757         return result;
 758 }
 759
 760 bool lxc_string_in_list(const char *needle, const char *haystack, char _sep)
 761 {
 762         char *token, *str, *saveptr = NULL;
 763         char sep[2] = { _sep, '\0' };
 764
 765         if (!haystack || !needle)
 766                 return 0;
 767
 768         str = alloca(strlen(haystack)+1);
 769         strcpy(str, haystack);
 770         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 771                 if (strcmp(needle, token) == 0)
 772                         return 1;
 773         }
 774
 775         return 0;
 776 }
 777
 778 char **lxc_string_split(const char *string, char _sep)
 779 {
 780         char *token, *str, *saveptr = NULL;
 781         char sep[2] = {_sep, '\0'};
 782         char **tmp = NULL, **result = NULL;
 783         size_t result_capacity = 0;
 784         size_t result_count = 0;
 785         int r, saved_errno;
 786
 787         if (!string)
 788                 return calloc(1, sizeof(char *));
 789
 790         str = alloca(strlen(string) + 1);
 791         strcpy(str, string);
 792         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 793                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 794                 if (r < 0)
 795                         goto error_out;
 796                 result[result_count] = strdup(token);
 797                 if (!result[result_count])
 798                         goto error_out;
 799                 result_count++;
 800         }
 801
 802         /* if we allocated too much, reduce it */
 803         tmp = realloc(result, (result_count + 1) * sizeof(char *));
 804         if (!tmp)
 805                 goto error_out;
 806         result = tmp;
 807         /* Make sure we don't return uninitialized memory. */
 808         if (result_count == 0)
 809                 *result = NULL;
 810         return result;
 811 error_out:
 812         saved_errno = errno;
 813         lxc_free_array((void **)result, free);
 814         errno = saved_errno;
 815         return NULL;
 816 }
 817
 818 static bool complete_word(char ***result, char *start, char *end, size_t *cap, size_t *cnt)
 819 {
 820         int r;
 821
 822         r = lxc_grow_array((void ***)result, cap, 2 + *cnt, 16);
 823         if (r < 0)
 824                 return false;
 825         (*result)[*cnt] = strndup(start, end - start);
 826         if (!(*result)[*cnt])
 827                 return false;
 828         (*cnt)++;
 829
 830         return true;
 831 }
 832
 833 /*
 834  * Given a a string 'one two "three four"', split into three words,
 835  * one, two, and "three four"
 836  */
 837 char **lxc_string_split_quoted(char *string)
 838 {
 839         char *nextword = string, *p, state;
 840         char **result = NULL;
 841         size_t result_capacity = 0;
 842         size_t result_count = 0;
 843
 844         if (!string || !*string)
 845                 return calloc(1, sizeof(char *));
 846
 847         // TODO I'm *not* handling escaped quote
 848         state = ' ';
 849         for (p = string; *p; p++) {
 850                 switch(state) {
 851                 case ' ':
 852                         if (isspace(*p))
 853                                 continue;
 854                         else if (*p == '"' || *p == '\'') {
 855                                 nextword = p;
 856                                 state = *p;
 857                                 continue;
 858                         }
 859                         nextword = p;
 860                         state = 'a';
 861                         continue;
 862                 case 'a':
 863                         if (isspace(*p)) {
 864                                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 865                                 state = ' ';
 866                                 continue;
 867                         }
 868                         continue;
 869                 case '"':
 870                 case '\'':
 871                         if (*p == state) {
 872                                 complete_word(&result, nextword+1, p, &result_capacity, &result_count);
 873                                 state = ' ';
 874                                 continue;
 875                         }
 876                         continue;
 877                 }
 878         }
 879
 880         if (state == 'a')
 881                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 882
 883         return realloc(result, (result_count + 1) * sizeof(char *));
 884 }
 885
 886 char **lxc_string_split_and_trim(const char *string, char _sep)
 887 {
 888         char *token, *str, *saveptr = NULL;
 889         char sep[2] = { _sep, '\0' };
 890         char **result = NULL;
 891         size_t result_capacity = 0;
 892         size_t result_count = 0;
 893         int r, saved_errno;
 894         size_t i = 0;
 895
 896         if (!string)
 897                 return calloc(1, sizeof(char *));
 898
 899         str = alloca(strlen(string)+1);
 900         strcpy(str, string);
 901         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 902                 while (token[0] == ' ' || token[0] == '\t')
 903                         token++;
 904                 i = strlen(token);
 905                 while (i > 0 && (token[i - 1] == ' ' || token[i - 1] == '\t')) {
 906                         token[i - 1] = '\0';
 907                         i--;
 908                 }
 909                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 910                 if (r < 0)
 911                         goto error_out;
 912                 result[result_count] = strdup(token);
 913                 if (!result[result_count])
 914                         goto error_out;
 915                 result_count++;
 916         }
 917
 918         /* if we allocated too much, reduce it */
 919         return realloc(result, (result_count + 1) * sizeof(char *));
 920 error_out:
 921         saved_errno = errno;
 922         lxc_free_array((void **)result, free);
 923         errno = saved_errno;
 924         return NULL;
 925 }
 926
 927 void lxc_free_array(void **array, lxc_free_fn element_free_fn)
 928 {
 929         void **p;
 930         for (p = array; p && *p; p++)
 931                 element_free_fn(*p);
 932         free((void*)array);
 933 }
 934
 935 int lxc_grow_array(void ***array, size_t* capacity, size_t new_size, size_t capacity_increment)
 936 {
 937         size_t new_capacity;
 938         void **new_array;
 939
 940         /* first time around, catch some trivial mistakes of the user
 941          * only initializing one of these */
 942         if (!*array || !*capacity) {
 943                 *array = NULL;
 944                 *capacity = 0;
 945         }
 946
 947         new_capacity = *capacity;
 948         while (new_size + 1 > new_capacity)
 949                 new_capacity += capacity_increment;
 950         if (new_capacity != *capacity) {
 951                 /* we have to reallocate */
 952                 new_array = realloc(*array, new_capacity * sizeof(void *));
 953                 if (!new_array)
 954                         return -1;
 955                 memset(&new_array[*capacity], 0, (new_capacity - (*capacity)) * sizeof(void *));
 956                 *array = new_array;
 957                 *capacity = new_capacity;
 958         }
 959
 960         /* array has sufficient elements */
 961         return 0;
 962 }
 963
 964 size_t lxc_array_len(void **array)
 965 {
 966         void **p;
 967         size_t result = 0;
 968
 969         for (p = array; p && *p; p++)
 970                 result++;
 971
 972         return result;
 973 }
 974
 975 int lxc_write_to_file(const char *filename, const void *buf, size_t count,
 976                       bool add_newline, mode_t mode)
 977 {
 978         int fd, saved_errno;
 979         ssize_t ret;
 980
 981         fd = open(filename, O_WRONLY | O_TRUNC | O_CREAT | O_CLOEXEC, mode);
 982         if (fd < 0)
 983                 return -1;
 984         ret = lxc_write_nointr(fd, buf, count);
 985         if (ret < 0)
 986                 goto out_error;
 987         if ((size_t)ret != count)
 988                 goto out_error;
 989         if (add_newline) {
 990                 ret = lxc_write_nointr(fd, "\n", 1);
 991                 if (ret != 1)
 992                         goto out_error;
 993         }
 994         close(fd);
 995         return 0;
 996
 997 out_error:
 998         saved_errno = errno;
 999         close(fd);
1000         errno = saved_errno;
1001         return -1;
1002 }
1003
1004 int lxc_read_from_file(const char *filename, void* buf, size_t count)
1005 {
1006         int fd = -1, saved_errno;
1007         ssize_t ret;
1008
1009         fd = open(filename, O_RDONLY | O_CLOEXEC);
1010         if (fd < 0)
1011                 return -1;
1012
1013         if (!buf || !count) {
1014                 char buf2[100];
1015                 size_t count2 = 0;
1016                 while ((ret = read(fd, buf2, 100)) > 0)
1017                         count2 += ret;
1018                 if (ret >= 0)
1019                         ret = count2;
1020         } else {
1021                 memset(buf, 0, count);
1022                 ret = read(fd, buf, count);
1023         }
1024
1025         if (ret < 0)
1026                 ERROR("read %s: %s", filename, strerror(errno));
1027
1028         saved_errno = errno;
1029         close(fd);
1030         errno = saved_errno;
1031         return ret;
1032 }
1033
1034 void **lxc_append_null_to_array(void **array, size_t count)
1035 {
1036         void **temp;
1037
1038         /* Append NULL to the array */
1039         if (count) {
1040                 temp = realloc(array, (count + 1) * sizeof(*array));
1041                 if (!temp) {
1042                         size_t i;
1043                         for (i = 0; i < count; i++)
1044                                 free(array[i]);
1045                         free(array);
1046                         return NULL;
1047                 }
1048                 array = temp;
1049                 array[count] = NULL;
1050         }
1051         return array;
1052 }
1053
1054 int randseed(bool srand_it)
1055 {
1056         /*
1057            srand pre-seed function based on /dev/urandom
1058            */
1059         unsigned int seed = time(NULL) + getpid();
1060
1061         FILE *f;
1062         f = fopen("/dev/urandom", "r");
1063         if (f) {
1064                 int ret = fread(&seed, sizeof(seed), 1, f);
1065                 if (ret != 1)
1066                         DEBUG("unable to fread /dev/urandom, %s, fallback to time+pid rand seed", strerror(errno));
1067                 fclose(f);
1068         }
1069
1070         if (srand_it)
1071                 srand(seed);
1072
1073         return seed;
1074 }
1075
1076 uid_t get_ns_uid(uid_t orig)
1077 {
1078         char *line = NULL;
1079         size_t sz = 0;
1080         uid_t nsid, hostid, range;
1081         FILE *f = fopen("/proc/self/uid_map", "r");
1082         if (!f)
1083                 return 0;
1084
1085         while (getline(&line, &sz, f) != -1) {
1086                 if (sscanf(line, "%u %u %u", &nsid, &hostid, &range) != 3)
1087                         continue;
1088                 if (hostid <= orig && hostid + range > orig) {
1089                         nsid += orig - hostid;
1090                         goto found;
1091                 }
1092         }
1093
1094         nsid = 0;
1095 found:
1096         fclose(f);
1097         free(line);
1098         return nsid;
1099 }
1100
1101 bool dir_exists(const char *path)
1102 {
1103         struct stat sb;
1104         int ret;
1105
1106         ret = stat(path, &sb);
1107         if (ret < 0)
1108                 /* Could be something other than eexist, just say "no". */
1109                 return false;
1110         return S_ISDIR(sb.st_mode);
1111 }
1112
1113 /* Note we don't use SHA-1 here as we don't want to depend on HAVE_GNUTLS.
1114  * FNV has good anti collision properties and we're not worried
1115  * about pre-image resistance or one-way-ness, we're just trying to make
1116  * the name unique in the 108 bytes of space we have.
1117  */
1118 uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
1119 {
1120         unsigned char *bp;
1121
1122         for(bp = buf; bp < (unsigned char *)buf + len; bp++)
1123         {
1124                 /* xor the bottom with the current octet */
1125                 hval ^= (uint64_t)*bp;
1126
1127                 /* gcc optimised:
1128                  * multiply by the 64 bit FNV magic prime mod 2^64
1129                  */
1130                 hval += (hval << 1) + (hval << 4) + (hval << 5) +
1131                         (hval << 7) + (hval << 8) + (hval << 40);
1132         }
1133
1134         return hval;
1135 }
1136
1137 /*
1138  * Detect whether / is mounted MS_SHARED.  The only way I know of to
1139  * check that is through /proc/self/mountinfo.
1140  * I'm only checking for /.  If the container rootfs or mount location
1141  * is MS_SHARED, but not '/', then you're out of luck - figuring that
1142  * out would be too much work to be worth it.
1143  */
1144 int detect_shared_rootfs(void)
1145 {
1146         char buf[LXC_LINELEN], *p;
1147         FILE *f;
1148         int i;
1149         char *p2;
1150
1151         f = fopen("/proc/self/mountinfo", "r");
1152         if (!f)
1153                 return 0;
1154         while (fgets(buf, LXC_LINELEN, f)) {
1155                 for (p = buf, i = 0; p && i < 4; i++)
1156                         p = strchr(p + 1, ' ');
1157                 if (!p)
1158                         continue;
1159                 p2 = strchr(p + 1, ' ');
1160                 if (!p2)
1161                         continue;
1162                 *p2 = '\0';
1163                 if (strcmp(p + 1, "/") == 0) {
1164                         /* This is '/'. Is it shared? */
1165                         p = strchr(p2 + 1, ' ');
1166                         if (p && strstr(p, "shared:")) {
1167                                 fclose(f);
1168                                 return 1;
1169                         }
1170                 }
1171         }
1172         fclose(f);
1173         return 0;
1174 }
1175
1176 bool switch_to_ns(pid_t pid, const char *ns) {
1177         int fd, ret;
1178         char nspath[MAXPATHLEN];
1179
1180         /* Switch to new ns */
1181         ret = snprintf(nspath, MAXPATHLEN, "/proc/%d/ns/%s", pid, ns);
1182         if (ret < 0 || ret >= MAXPATHLEN)
1183                 return false;
1184
1185         fd = open(nspath, O_RDONLY);
1186         if (fd < 0) {
1187                 SYSERROR("failed to open %s", nspath);
1188                 return false;
1189         }
1190
1191         ret = setns(fd, 0);
1192         if (ret) {
1193                 SYSERROR("failed to set process %d to %s of %d.", pid, ns, fd);
1194                 close(fd);
1195                 return false;
1196         }
1197         close(fd);
1198         return true;
1199 }
1200
1201 /*
1202  * looking at fs/proc_namespace.c, it appears we can
1203  * actually expect the rootfs entry to very specifically contain
1204  * " - rootfs rootfs "
1205  * IIUC, so long as we've chrooted so that rootfs is not our root,
1206  * the rootfs entry should always be skipped in mountinfo contents.
1207  */
1208 bool detect_ramfs_rootfs(void)
1209 {
1210         FILE *f;
1211         char *p, *p2;
1212         char *line = NULL;
1213         size_t len = 0;
1214         int i;
1215
1216         f = fopen("/proc/self/mountinfo", "r");
1217         if (!f)
1218                 return false;
1219
1220         while (getline(&line, &len, f) != -1) {
1221                 for (p = line, i = 0; p && i < 4; i++)
1222                         p = strchr(p + 1, ' ');
1223                 if (!p)
1224                         continue;
1225                 p2 = strchr(p + 1, ' ');
1226                 if (!p2)
1227                         continue;
1228                 *p2 = '\0';
1229                 if (strcmp(p + 1, "/") == 0) {
1230                         /* This is '/'. Is it the ramfs? */
1231                         p = strchr(p2 + 1, '-');
1232                         if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) {
1233                                 free(line);
1234                                 fclose(f);
1235                                 return true;
1236                         }
1237                 }
1238         }
1239         free(line);
1240         fclose(f);
1241         return false;
1242 }
1243
1244 char *on_path(const char *cmd, const char *rootfs) {
1245         char *path = NULL;
1246         char *entry = NULL;
1247         char *saveptr = NULL;
1248         char cmdpath[MAXPATHLEN];
1249         int ret;
1250
1251         path = getenv("PATH");
1252         if (!path)
1253                 return NULL;
1254
1255         path = strdup(path);
1256         if (!path)
1257                 return NULL;
1258
1259         entry = strtok_r(path, ":", &saveptr);
1260         while (entry) {
1261                 if (rootfs)
1262                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s/%s", rootfs, entry, cmd);
1263                 else
1264                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s", entry, cmd);
1265
1266                 if (ret < 0 || ret >= MAXPATHLEN)
1267                         goto next_loop;
1268
1269                 if (access(cmdpath, X_OK) == 0) {
1270                         free(path);
1271                         return strdup(cmdpath);
1272                 }
1273
1274 next_loop:
1275                 entry = strtok_r(NULL, ":", &saveptr);
1276         }
1277
1278         free(path);
1279         return NULL;
1280 }
1281
1282 bool file_exists(const char *f)
1283 {
1284         struct stat statbuf;
1285
1286         return stat(f, &statbuf) == 0;
1287 }
1288
1289 bool cgns_supported(void)
1290 {
1291         return file_exists("/proc/self/ns/cgroup");
1292 }
1293
1294 /* historically lxc-init has been under /usr/lib/lxc and under
1295  * /usr/lib/$ARCH/lxc.  It now lives as $prefix/sbin/init.lxc.
1296  */
1297 char *choose_init(const char *rootfs)
1298 {
1299         char *retv = NULL;
1300         const char *empty = "",
1301                    *tmp;
1302         int ret, env_set = 0;
1303
1304         if (!getenv("PATH")) {
1305                 if (setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 0))
1306                         SYSERROR("Failed to setenv");
1307                 env_set = 1;
1308         }
1309
1310         retv = on_path("init.lxc", rootfs);
1311
1312         if (env_set) {
1313                 if (unsetenv("PATH"))
1314                         SYSERROR("Failed to unsetenv");
1315         }
1316
1317         if (retv)
1318                 return retv;
1319
1320         retv = malloc(PATH_MAX);
1321         if (!retv)
1322                 return NULL;
1323
1324         if (rootfs)
1325                 tmp = rootfs;
1326         else
1327                 tmp = empty;
1328
1329         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, SBINDIR, "/init.lxc");
1330         if (ret < 0 || ret >= PATH_MAX) {
1331                 ERROR("pathname too long");
1332                 goto out1;
1333         }
1334         if (access(retv, X_OK) == 0)
1335                 return retv;
1336
1337         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, LXCINITDIR, "/lxc/lxc-init");
1338         if (ret < 0 || ret >= PATH_MAX) {
1339                 ERROR("pathname too long");
1340                 goto out1;
1341         }
1342         if (access(retv, X_OK) == 0)
1343                 return retv;
1344
1345         ret = snprintf(retv, PATH_MAX, "%s/usr/lib/lxc/lxc-init", tmp);
1346         if (ret < 0 || ret >= PATH_MAX) {
1347                 ERROR("pathname too long");
1348                 goto out1;
1349         }
1350         if (access(retv, X_OK) == 0)
1351                 return retv;
1352
1353         ret = snprintf(retv, PATH_MAX, "%s/sbin/lxc-init", tmp);
1354         if (ret < 0 || ret >= PATH_MAX) {
1355                 ERROR("pathname too long");
1356                 goto out1;
1357         }
1358         if (access(retv, X_OK) == 0)
1359                 return retv;
1360
1361         /*
1362          * Last resort, look for the statically compiled init.lxc which we
1363          * hopefully bind-mounted in.
1364          * If we are called during container setup, and we get to this point,
1365          * then the init.lxc.static from the host will need to be bind-mounted
1366          * in.  So we return NULL here to indicate that.
1367          */
1368         if (rootfs)
1369                 goto out1;
1370
1371         ret = snprintf(retv, PATH_MAX, "/init.lxc.static");
1372         if (ret < 0 || ret >= PATH_MAX) {
1373                 WARN("Nonsense - name /lxc.init.static too long");
1374                 goto out1;
1375         }
1376         if (access(retv, X_OK) == 0)
1377                 return retv;
1378
1379 out1:
1380         free(retv);
1381         return NULL;
1382 }
1383
1384 int print_to_file(const char *file, const char *content)
1385 {
1386         FILE *f;
1387         int ret = 0;
1388
1389         f = fopen(file, "w");
1390         if (!f)
1391                 return -1;
1392         if (fprintf(f, "%s", content) != strlen(content))
1393                 ret = -1;
1394         fclose(f);
1395         return ret;
1396 }
1397
1398 int is_dir(const char *path)
1399 {
1400         struct stat statbuf;
1401         int ret = stat(path, &statbuf);
1402         if (ret == 0 && S_ISDIR(statbuf.st_mode))
1403                 return 1;
1404         return 0;
1405 }
1406
1407 /*
1408  * Given the '-t' template option to lxc-create, figure out what to
1409  * do.  If the template is a full executable path, use that.  If it
1410  * is something like 'sshd', then return $templatepath/lxc-sshd.
1411  * On success return the template, on error return NULL.
1412  */
1413 char *get_template_path(const char *t)
1414 {
1415         int ret, len;
1416         char *tpath;
1417
1418         if (t[0] == '/' && access(t, X_OK) == 0) {
1419                 tpath = strdup(t);
1420                 return tpath;
1421         }
1422
1423         len = strlen(LXCTEMPLATEDIR) + strlen(t) + strlen("/lxc-") + 1;
1424         tpath = malloc(len);
1425         if (!tpath)
1426                 return NULL;
1427         ret = snprintf(tpath, len, "%s/lxc-%s", LXCTEMPLATEDIR, t);
1428         if (ret < 0 || ret >= len) {
1429                 free(tpath);
1430                 return NULL;
1431         }
1432         if (access(tpath, X_OK) < 0) {
1433                 SYSERROR("bad template: %s", t);
1434                 free(tpath);
1435                 return NULL;
1436         }
1437
1438         return tpath;
1439 }
1440
1441 /*
1442  * @path:    a pathname where / replaced with '\0'.
1443  * @offsetp: pointer to int showing which path segment was last seen.
1444  *           Updated on return to reflect the next segment.
1445  * @fulllen: full original path length.
1446  * Returns a pointer to the next path segment, or NULL if done.
1447  */
1448 static char *get_nextpath(char *path, int *offsetp, int fulllen)
1449 {
1450         int offset = *offsetp;
1451
1452         if (offset >= fulllen)
1453                 return NULL;
1454
1455         while (path[offset] != '\0' && offset < fulllen)
1456                 offset++;
1457         while (path[offset] == '\0' && offset < fulllen)
1458                 offset++;
1459
1460         *offsetp = offset;
1461         return (offset < fulllen) ? &path[offset] : NULL;
1462 }
1463
1464 /*
1465  * Check that @subdir is a subdir of @dir.  @len is the length of
1466  * @dir (to avoid having to recalculate it).
1467  */
1468 static bool is_subdir(const char *subdir, const char *dir, size_t len)
1469 {
1470         size_t subdirlen = strlen(subdir);
1471
1472         if (subdirlen < len)
1473                 return false;
1474         if (strncmp(subdir, dir, len) != 0)
1475                 return false;
1476         if (dir[len-1] == '/')
1477                 return true;
1478         if (subdir[len] == '/' || subdirlen == len)
1479                 return true;
1480         return false;
1481 }
1482
1483 /*
1484  * Check if the open fd is a symlink.  Return -ELOOP if it is.  Return
1485  * -ENOENT if we couldn't fstat.  Return 0 if the fd is ok.
1486  */
1487 static int check_symlink(int fd)
1488 {
1489         struct stat sb;
1490         int ret = fstat(fd, &sb);
1491         if (ret < 0)
1492                 return -ENOENT;
1493         if (S_ISLNK(sb.st_mode))
1494                 return -ELOOP;
1495         return 0;
1496 }
1497
1498 /*
1499  * Open a file or directory, provided that it contains no symlinks.
1500  *
1501  * CAVEAT: This function must not be used for other purposes than container
1502  * setup before executing the container's init
1503  */
1504 static int open_if_safe(int dirfd, const char *nextpath)
1505 {
1506         int newfd = openat(dirfd, nextpath, O_RDONLY | O_NOFOLLOW);
1507         if (newfd >= 0) /* Was not a symlink, all good. */
1508                 return newfd;
1509
1510         if (errno == ELOOP)
1511                 return newfd;
1512
1513         if (errno == EPERM || errno == EACCES) {
1514                 /* We're not root (cause we got EPERM) so try opening with
1515                  * O_PATH.
1516                  */
1517                 newfd = openat(dirfd, nextpath, O_PATH | O_NOFOLLOW);
1518                 if (newfd >= 0) {
1519                         /* O_PATH will return an fd for symlinks. We know
1520                          * nextpath wasn't a symlink at last openat, so if fd is
1521                          * now a link, then something * fishy is going on.
1522                          */
1523                         int ret = check_symlink(newfd);
1524                         if (ret < 0) {
1525                                 close(newfd);
1526                                 newfd = ret;
1527                         }
1528                 }
1529         }
1530
1531         return newfd;
1532 }
1533
1534 /*
1535  * Open a path intending for mounting, ensuring that the final path
1536  * is inside the container's rootfs.
1537  *
1538  * CAVEAT: This function must not be used for other purposes than container
1539  * setup before executing the container's init
1540  *
1541  * @target: path to be opened
1542  * @prefix_skip: a part of @target in which to ignore symbolic links.  This
1543  * would be the container's rootfs.
1544  *
1545  * Return an open fd for the path, or <0 on error.
1546  */
1547 static int open_without_symlink(const char *target, const char *prefix_skip)
1548 {
1549         int curlen = 0, dirfd, fulllen, i;
1550         char *dup = NULL;
1551
1552         fulllen = strlen(target);
1553
1554         /* make sure prefix-skip makes sense */
1555         if (prefix_skip && strlen(prefix_skip) > 0) {
1556                 curlen = strlen(prefix_skip);
1557                 if (!is_subdir(target, prefix_skip, curlen)) {
1558                         ERROR("WHOA there - target '%s' didn't start with prefix '%s'",
1559                                 target, prefix_skip);
1560                         return -EINVAL;
1561                 }
1562                 /*
1563                  * get_nextpath() expects the curlen argument to be
1564                  * on a  (turned into \0) / or before it, so decrement
1565                  * curlen to make sure that happens
1566                  */
1567                 if (curlen)
1568                         curlen--;
1569         } else {
1570                 prefix_skip = "/";
1571                 curlen = 0;
1572         }
1573
1574         /* Make a copy of target which we can hack up, and tokenize it */
1575         if ((dup = strdup(target)) == NULL) {
1576                 SYSERROR("Out of memory checking for symbolic link");
1577                 return -ENOMEM;
1578         }
1579         for (i = 0; i < fulllen; i++) {
1580                 if (dup[i] == '/')
1581                         dup[i] = '\0';
1582         }
1583
1584         dirfd = open(prefix_skip, O_RDONLY);
1585         if (dirfd < 0)
1586                 goto out;
1587         while (1) {
1588                 int newfd, saved_errno;
1589                 char *nextpath;
1590
1591                 if ((nextpath = get_nextpath(dup, &curlen, fulllen)) == NULL)
1592                         goto out;
1593                 newfd = open_if_safe(dirfd, nextpath);
1594                 saved_errno = errno;
1595                 close(dirfd);
1596                 dirfd = newfd;
1597                 if (newfd < 0) {
1598                         errno = saved_errno;
1599                         if (errno == ELOOP)
1600                                 SYSERROR("%s in %s was a symbolic link!", nextpath, target);
1601                         goto out;
1602                 }
1603         }
1604
1605 out:
1606         free(dup);
1607         return dirfd;
1608 }
1609
1610 /*
1611  * Safely mount a path into a container, ensuring that the mount target
1612  * is under the container's @rootfs.  (If @rootfs is NULL, then the container
1613  * uses the host's /)
1614  *
1615  * CAVEAT: This function must not be used for other purposes than container
1616  * setup before executing the container's init
1617  */
1618 int safe_mount(const char *src, const char *dest, const char *fstype,
1619                 unsigned long flags, const void *data, const char *rootfs)
1620 {
1621         int destfd, ret, saved_errno;
1622         /* Only needs enough for /proc/self/fd/<fd>. */
1623         char srcbuf[50], destbuf[50];
1624         int srcfd = -1;
1625         const char *mntsrc = src;
1626
1627         if (!rootfs)
1628                 rootfs = "";
1629
1630         /* todo - allow symlinks for relative paths if 'allowsymlinks' option is passed */
1631         if (flags & MS_BIND && src && src[0] != '/') {
1632                 INFO("this is a relative bind mount");
1633                 srcfd = open_without_symlink(src, NULL);
1634                 if (srcfd < 0)
1635                         return srcfd;
1636                 ret = snprintf(srcbuf, 50, "/proc/self/fd/%d", srcfd);
1637                 if (ret < 0 || ret > 50) {
1638                         close(srcfd);
1639                         ERROR("Out of memory");
1640                         return -EINVAL;
1641                 }
1642                 mntsrc = srcbuf;
1643         }
1644
1645         destfd = open_without_symlink(dest, rootfs);
1646         if (destfd < 0) {
1647                 if (srcfd != -1) {
1648                         saved_errno = errno;
1649                         close(srcfd);
1650                         errno = saved_errno;
1651                 }
1652                 return destfd;
1653         }
1654
1655         ret = snprintf(destbuf, 50, "/proc/self/fd/%d", destfd);
1656         if (ret < 0 || ret > 50) {
1657                 if (srcfd != -1)
1658                         close(srcfd);
1659                 close(destfd);
1660                 ERROR("Out of memory");
1661                 return -EINVAL;
1662         }
1663
1664         ret = mount(mntsrc, destbuf, fstype, flags, data);
1665         saved_errno = errno;
1666         if (srcfd != -1)
1667                 close(srcfd);
1668         close(destfd);
1669         if (ret < 0) {
1670                 errno = saved_errno;
1671                 SYSERROR("Failed to mount %s onto %s", src ? src : "(null)", dest);
1672                 return ret;
1673         }
1674
1675         return 0;
1676 }
1677
1678 /*
1679  * Mount a proc under @rootfs if proc self points to a pid other than
1680  * my own.  This is needed to have a known-good proc mount for setting
1681  * up LSMs both at container startup and attach.
1682  *
1683  * @rootfs : the rootfs where proc should be mounted
1684  *
1685  * Returns < 0 on failure, 0 if the correct proc was already mounted
1686  * and 1 if a new proc was mounted.
1687  *
1688  * NOTE: not to be called from inside the container namespace!
1689  */
1690 int lxc_mount_proc_if_needed(const char *rootfs)
1691 {
1692         char path[MAXPATHLEN];
1693         int link_to_pid, linklen, mypid, ret;
1694         char link[LXC_NUMSTRLEN64] = {0};
1695
1696         ret = snprintf(path, MAXPATHLEN, "%s/proc/self", rootfs);
1697         if (ret < 0 || ret >= MAXPATHLEN) {
1698                 SYSERROR("proc path name too long");
1699                 return -1;
1700         }
1701
1702         linklen = readlink(path, link, LXC_NUMSTRLEN64);
1703
1704         ret = snprintf(path, MAXPATHLEN, "%s/proc", rootfs);
1705         if (ret < 0 || ret >= MAXPATHLEN) {
1706                 SYSERROR("proc path name too long");
1707                 return -1;
1708         }
1709
1710         /* /proc not mounted */
1711         if (linklen < 0) {
1712                 if (mkdir(path, 0755) && errno != EEXIST)
1713                         return -1;
1714                 goto domount;
1715         } else if (linklen >= LXC_NUMSTRLEN64) {
1716                 link[linklen - 1] = '\0';
1717                 ERROR("readlink returned truncated content: \"%s\"", link);
1718                 return -1;
1719         }
1720
1721         mypid = lxc_raw_getpid();
1722         INFO("I am %d, /proc/self points to \"%s\"", mypid, link);
1723
1724         if (lxc_safe_int(link, &link_to_pid) < 0)
1725                 return -1;
1726
1727         /* correct procfs is already mounted */
1728         if (link_to_pid == mypid)
1729                 return 0;
1730
1731         ret = umount2(path, MNT_DETACH);
1732         if (ret < 0)
1733                 WARN("failed to umount \"%s\" with MNT_DETACH", path);
1734
1735 domount:
1736         /* rootfs is NULL */
1737         if (!strcmp(rootfs, ""))
1738                 ret = mount("proc", path, "proc", 0, NULL);
1739         else
1740                 ret = safe_mount("proc", path, "proc", 0, NULL, rootfs);
1741         if (ret < 0)
1742                 return -1;
1743
1744         INFO("mounted /proc in container for security transition");
1745         return 1;
1746 }
1747
1748 int open_devnull(void)
1749 {
1750         int fd = open("/dev/null", O_RDWR);
1751
1752         if (fd < 0)
1753                 SYSERROR("Can't open /dev/null");
1754
1755         return fd;
1756 }
1757
1758 int set_stdfds(int fd)
1759 {
1760         int ret;
1761
1762         if (fd < 0)
1763                 return -1;
1764
1765         ret = dup2(fd, STDIN_FILENO);
1766         if (ret < 0)
1767                 return -1;
1768
1769         ret = dup2(fd, STDOUT_FILENO);
1770         if (ret < 0)
1771                 return -1;
1772
1773         ret = dup2(fd, STDERR_FILENO);
1774         if (ret < 0)
1775                 return -1;
1776
1777         return 0;
1778 }
1779
1780 int null_stdfds(void)
1781 {
1782         int ret = -1;
1783         int fd = open_devnull();
1784
1785         if (fd >= 0) {
1786                 ret = set_stdfds(fd);
1787                 close(fd);
1788         }
1789
1790         return ret;
1791 }
1792
1793 /*
1794  * Return the number of lines in file @fn, or -1 on error
1795  */
1796 int lxc_count_file_lines(const char *fn)
1797 {
1798         FILE *f;
1799         char *line = NULL;
1800         size_t sz = 0;
1801         int n = 0;
1802
1803         f = fopen_cloexec(fn, "r");
1804         if (!f)
1805                 return -1;
1806
1807         while (getline(&line, &sz, f) != -1) {
1808                 n++;
1809         }
1810         free(line);
1811         fclose(f);
1812         return n;
1813 }
1814
1815 /* Check whether a signal is blocked by a process. */
1816 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
1817 #define __PROC_STATUS_LEN (6 + (LXC_NUMSTRLEN64) + 7 + 1)
1818 bool task_blocking_signal(pid_t pid, int signal)
1819 {
1820         int ret;
1821         char status[__PROC_STATUS_LEN];
1822         FILE *f;
1823         long unsigned int sigblk = 0;
1824         size_t n = 0;
1825         bool bret = false;
1826         char *line = NULL;
1827
1828         ret = snprintf(status, __PROC_STATUS_LEN, "/proc/%d/status", pid);
1829         if (ret < 0 || ret >= __PROC_STATUS_LEN)
1830                 return bret;
1831
1832         f = fopen(status, "r");
1833         if (!f)
1834                 return bret;
1835
1836         while (getline(&line, &n, f) != -1) {
1837                 if (strncmp(line, "SigBlk:", 7))
1838                         continue;
1839
1840                 if (sscanf(line + 7, "%lx", &sigblk) != 1)
1841                         goto out;
1842         }
1843
1844         if (sigblk & (1LU << (signal - 1)))
1845                 bret = true;
1846
1847 out:
1848         free(line);
1849         fclose(f);
1850         return bret;
1851 }
1852
1853 static int lxc_append_null_to_list(void ***list)
1854 {
1855         int newentry = 0;
1856         void **tmp;
1857
1858         if (*list)
1859                 for (; (*list)[newentry]; newentry++) {
1860                         ;
1861                 }
1862
1863         tmp = realloc(*list, (newentry + 2) * sizeof(void **));
1864         if (!tmp)
1865                 return -1;
1866
1867         *list = tmp;
1868         (*list)[newentry + 1] = NULL;
1869
1870         return newentry;
1871 }
1872
1873 int lxc_append_string(char ***list, char *entry)
1874 {
1875         char *copy;
1876         int newentry;
1877
1878         newentry = lxc_append_null_to_list((void ***)list);
1879         if (newentry < 0)
1880                 return -1;
1881
1882         copy = strdup(entry);
1883         if (!copy)
1884                 return -1;
1885
1886         (*list)[newentry] = copy;
1887
1888         return 0;
1889 }
1890
1891 int lxc_preserve_ns(const int pid, const char *ns)
1892 {
1893         int ret;
1894 /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */
1895 #define __NS_PATH_LEN 50
1896         char path[__NS_PATH_LEN];
1897
1898         /* This way we can use this function to also check whether namespaces
1899          * are supported by the kernel by passing in the NULL or the empty
1900          * string.
1901          */
1902         ret = snprintf(path, __NS_PATH_LEN, "/proc/%d/ns%s%s", pid,
1903                        !ns || strcmp(ns, "") == 0 ? "" : "/",
1904                        !ns || strcmp(ns, "") == 0 ? "" : ns);
1905         errno = EFBIG;
1906         if (ret < 0 || (size_t)ret >= __NS_PATH_LEN)
1907                 return -EFBIG;
1908
1909         return open(path, O_RDONLY | O_CLOEXEC);
1910 }
1911
1912 int lxc_safe_uint(const char *numstr, unsigned int *converted)
1913 {
1914         char *err = NULL;
1915         unsigned long int uli;
1916
1917         while (isspace(*numstr))
1918                 numstr++;
1919
1920         if (*numstr == '-')
1921                 return -EINVAL;
1922
1923         errno = 0;
1924         uli = strtoul(numstr, &err, 0);
1925         if (errno == ERANGE && uli == ULONG_MAX)
1926                 return -ERANGE;
1927
1928         if (err == numstr || *err != '\0')
1929                 return -EINVAL;
1930
1931         if (uli > UINT_MAX)
1932                 return -ERANGE;
1933
1934         *converted = (unsigned int)uli;
1935         return 0;
1936 }
1937
1938 int lxc_safe_ulong(const char *numstr, unsigned long *converted)
1939 {
1940         char *err = NULL;
1941         unsigned long int uli;
1942
1943         while (isspace(*numstr))
1944                 numstr++;
1945
1946         if (*numstr == '-')
1947                 return -EINVAL;
1948
1949         errno = 0;
1950         uli = strtoul(numstr, &err, 0);
1951         if (errno == ERANGE && uli == ULONG_MAX)
1952                 return -ERANGE;
1953
1954         if (err == numstr || *err != '\0')
1955                 return -EINVAL;
1956
1957         *converted = uli;
1958         return 0;
1959 }
1960
1961 int lxc_safe_uint64(const char *numstr, uint64_t *converted)
1962 {
1963         char *err = NULL;
1964         uint64_t u;
1965
1966         while (isspace(*numstr))
1967                 numstr++;
1968
1969         if (*numstr == '-')
1970                 return -EINVAL;
1971
1972         errno = 0;
1973         u = strtoull(numstr, &err, 0);
1974         if (errno == ERANGE && u == ULLONG_MAX)
1975                 return -ERANGE;
1976
1977         if (err == numstr || *err != '\0')
1978                 return -EINVAL;
1979
1980         *converted = u;
1981         return 0;
1982 }
1983
1984 int lxc_safe_int(const char *numstr, int *converted)
1985 {
1986         char *err = NULL;
1987         signed long int sli;
1988
1989         errno = 0;
1990         sli = strtol(numstr, &err, 0);
1991         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
1992                 return -ERANGE;
1993
1994         if (errno != 0 && sli == 0)
1995                 return -EINVAL;
1996
1997         if (err == numstr || *err != '\0')
1998                 return -EINVAL;
1999
2000         if (sli > INT_MAX || sli < INT_MIN)
2001                 return -ERANGE;
2002
2003         *converted = (int)sli;
2004         return 0;
2005 }
2006
2007 int lxc_safe_long(const char *numstr, long int *converted)
2008 {
2009         char *err = NULL;
2010         signed long int sli;
2011
2012         errno = 0;
2013         sli = strtol(numstr, &err, 0);
2014         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
2015                 return -ERANGE;
2016
2017         if (errno != 0 && sli == 0)
2018                 return -EINVAL;
2019
2020         if (err == numstr || *err != '\0')
2021                 return -EINVAL;
2022
2023         *converted = sli;
2024         return 0;
2025 }
2026
2027 int lxc_safe_long_long(const char *numstr, long long int *converted)
2028 {
2029         char *err = NULL;
2030         signed long long int sli;
2031
2032         errno = 0;
2033         sli = strtoll(numstr, &err, 0);
2034         if (errno == ERANGE && (sli == LLONG_MAX || sli == LLONG_MIN))
2035                 return -ERANGE;
2036
2037         if (errno != 0 && sli == 0)
2038                 return -EINVAL;
2039
2040         if (err == numstr || *err != '\0')
2041                 return -EINVAL;
2042
2043         *converted = sli;
2044         return 0;
2045 }
2046
2047 int lxc_switch_uid_gid(uid_t uid, gid_t gid)
2048 {
2049         if (setgid(gid) < 0) {
2050                 SYSERROR("Failed to switch to gid %d.", gid);
2051                 return -errno;
2052         }
2053         NOTICE("Switched to gid %d.", gid);
2054
2055         if (setuid(uid) < 0) {
2056                 SYSERROR("Failed to switch to uid %d.", uid);
2057                 return -errno;
2058         }
2059         NOTICE("Switched to uid %d.", uid);
2060
2061         return 0;
2062 }
2063
2064 /* Simple covenience function which enables uniform logging. */
2065 int lxc_setgroups(int size, gid_t list[])
2066 {
2067         if (setgroups(size, list) < 0) {
2068                 SYSERROR("Failed to setgroups().");
2069                 return -errno;
2070         }
2071         NOTICE("Dropped additional groups.");
2072
2073         return 0;
2074 }
2075
2076 static int lxc_get_unused_loop_dev_legacy(char *loop_name)
2077 {
2078         struct dirent *dp;
2079         struct loop_info64 lo64;
2080         DIR *dir;
2081         int dfd = -1, fd = -1, ret = -1;
2082
2083         dir = opendir("/dev");
2084         if (!dir)
2085                 return -1;
2086
2087         while ((dp = readdir(dir))) {
2088                 if (!dp)
2089                         break;
2090
2091                 if (strncmp(dp->d_name, "loop", 4) != 0)
2092                         continue;
2093
2094                 dfd = dirfd(dir);
2095                 if (dfd < 0)
2096                         continue;
2097
2098                 fd = openat(dfd, dp->d_name, O_RDWR);
2099                 if (fd < 0)
2100                         continue;
2101
2102                 ret = ioctl(fd, LOOP_GET_STATUS64, &lo64);
2103                 if (ret < 0) {
2104                         if (ioctl(fd, LOOP_GET_STATUS64, &lo64) == 0 ||
2105                             errno != ENXIO) {
2106                                 close(fd);
2107                                 fd = -1;
2108                                 continue;
2109                         }
2110                 }
2111
2112                 ret = snprintf(loop_name, LO_NAME_SIZE, "/dev/%s", dp->d_name);
2113                 if (ret < 0 || ret >= LO_NAME_SIZE) {
2114                         close(fd);
2115                         fd = -1;
2116                         continue;
2117                 }
2118
2119                 break;
2120         }
2121
2122         closedir(dir);
2123
2124         if (fd < 0)
2125                 return -1;
2126
2127         return fd;
2128 }
2129
2130 static int lxc_get_unused_loop_dev(char *name_loop)
2131 {
2132         int loop_nr, ret;
2133         int fd_ctl = -1, fd_tmp = -1;
2134
2135         fd_ctl = open("/dev/loop-control", O_RDWR | O_CLOEXEC);
2136         if (fd_ctl < 0)
2137                 return -ENODEV;
2138
2139         loop_nr = ioctl(fd_ctl, LOOP_CTL_GET_FREE);
2140         if (loop_nr < 0)
2141                 goto on_error;
2142
2143         ret = snprintf(name_loop, LO_NAME_SIZE, "/dev/loop%d", loop_nr);
2144         if (ret < 0 || ret >= LO_NAME_SIZE)
2145                 goto on_error;
2146
2147         fd_tmp = open(name_loop, O_RDWR | O_CLOEXEC);
2148         if (fd_tmp < 0)
2149                 goto on_error;
2150
2151 on_error:
2152         close(fd_ctl);
2153         return fd_tmp;
2154 }
2155
2156 int lxc_prepare_loop_dev(const char *source, char *loop_dev, int flags)
2157 {
2158         int ret;
2159         struct loop_info64 lo64;
2160         int fd_img = -1, fret = -1, fd_loop = -1;
2161
2162         fd_loop = lxc_get_unused_loop_dev(loop_dev);
2163         if (fd_loop < 0) {
2164                 if (fd_loop == -ENODEV)
2165                         fd_loop = lxc_get_unused_loop_dev_legacy(loop_dev);
2166                 else
2167                         goto on_error;
2168         }
2169
2170         fd_img = open(source, O_RDWR | O_CLOEXEC);
2171         if (fd_img < 0)
2172                 goto on_error;
2173
2174         ret = ioctl(fd_loop, LOOP_SET_FD, fd_img);
2175         if (ret < 0)
2176                 goto on_error;
2177
2178         memset(&lo64, 0, sizeof(lo64));
2179         lo64.lo_flags = flags;
2180
2181         ret = ioctl(fd_loop, LOOP_SET_STATUS64, &lo64);
2182         if (ret < 0)
2183                 goto on_error;
2184
2185         fret = 0;
2186
2187 on_error:
2188         if (fd_img >= 0)
2189                 close(fd_img);
2190
2191         if (fret < 0 && fd_loop >= 0) {
2192                 close(fd_loop);
2193                 fd_loop = -1;
2194         }
2195
2196         return fd_loop;
2197 }
2198
2199 int lxc_unstack_mountpoint(const char *path, bool lazy)
2200 {
2201         int ret;
2202         int umounts = 0;
2203
2204 pop_stack:
2205         ret = umount2(path, lazy ? MNT_DETACH : 0);
2206         if (ret < 0) {
2207                 /* We consider anything else than EINVAL deadly to prevent going
2208                  * into an infinite loop. (The other alternative is constantly
2209                  * parsing /proc/self/mountinfo which is yucky and probably
2210                  * racy.)
2211                  */
2212                 if (errno != EINVAL)
2213                         return -errno;
2214         } else {
2215                 /* Just stop counting when this happens. That'd just be so
2216                  * stupid that we won't even bother trying to report back the
2217                  * correct value anymore.
2218                  */
2219                 if (umounts != INT_MAX)
2220                         umounts++;
2221                 /* We succeeded in umounting. Make sure that there's no other
2222                  * mountpoint stacked underneath.
2223                  */
2224                 goto pop_stack;
2225         }
2226
2227         return umounts;
2228 }
2229
2230 int run_command(char *buf, size_t buf_size, int (*child_fn)(void *), void *args)
2231 {
2232         pid_t child;
2233         int ret, fret, pipefd[2];
2234         ssize_t bytes;
2235
2236         /* Make sure our callers do not receive unitialized memory. */
2237         if (buf_size > 0 && buf)
2238                 buf[0] = '\0';
2239
2240         if (pipe(pipefd) < 0) {
2241                 SYSERROR("failed to create pipe");
2242                 return -1;
2243         }
2244
2245         child = lxc_raw_clone(0);
2246         if (child < 0) {
2247                 close(pipefd[0]);
2248                 close(pipefd[1]);
2249                 SYSERROR("failed to create new process");
2250                 return -1;
2251         }
2252
2253         if (child == 0) {
2254                 /* Close the read-end of the pipe. */
2255                 close(pipefd[0]);
2256
2257                 /* Redirect std{err,out} to write-end of the
2258                  * pipe.
2259                  */
2260                 ret = dup2(pipefd[1], STDOUT_FILENO);
2261                 if (ret >= 0)
2262                         ret = dup2(pipefd[1], STDERR_FILENO);
2263
2264                 /* Close the write-end of the pipe. */
2265                 close(pipefd[1]);
2266
2267                 if (ret < 0) {
2268                         SYSERROR("failed to duplicate std{err,out} file descriptor");
2269                         _exit(EXIT_FAILURE);
2270                 }
2271
2272                 /* Does not return. */
2273                 child_fn(args);
2274                 ERROR("failed to exec command");
2275                 _exit(EXIT_FAILURE);
2276         }
2277
2278         /* close the write-end of the pipe */
2279         close(pipefd[1]);
2280
2281         if (buf && buf_size > 0) {
2282                 bytes = read(pipefd[0], buf, buf_size - 1);
2283                 if (bytes > 0)
2284                         buf[bytes - 1] = '\0';
2285         }
2286
2287         fret = wait_for_pid(child);
2288         /* close the read-end of the pipe */
2289         close(pipefd[0]);
2290
2291         return fret;
2292 }
2293
2294 char *must_make_path(const char *first, ...)
2295 {
2296         va_list args;
2297         char *cur, *dest;
2298         size_t full_len = strlen(first);
2299
2300         dest = must_copy_string(first);
2301
2302         va_start(args, first);
2303         while ((cur = va_arg(args, char *)) != NULL) {
2304                 full_len += strlen(cur);
2305                 if (cur[0] != '/')
2306                         full_len++;
2307                 dest = must_realloc(dest, full_len + 1);
2308                 if (cur[0] != '/')
2309                         strcat(dest, "/");
2310                 strcat(dest, cur);
2311         }
2312         va_end(args);
2313
2314         return dest;
2315 }
2316
2317 char *must_append_path(char *first, ...)
2318 {
2319         char *cur;
2320         size_t full_len;
2321         va_list args;
2322         char *dest = first;
2323
2324         full_len = strlen(first);
2325         va_start(args, first);
2326         while ((cur = va_arg(args, char *)) != NULL) {
2327                 full_len += strlen(cur);
2328
2329                 if (cur[0] != '/')
2330                         full_len++;
2331
2332                 dest = must_realloc(dest, full_len + 1);
2333
2334                 if (cur[0] != '/')
2335                         strcat(dest, "/");
2336
2337                 strcat(dest, cur);
2338         }
2339         va_end(args);
2340
2341         return dest;
2342 }
2343
2344 char *must_copy_string(const char *entry)
2345 {
2346         char *ret;
2347
2348         if (!entry)
2349                 return NULL;
2350         do {
2351                 ret = strdup(entry);
2352         } while (!ret);
2353
2354         return ret;
2355 }
2356
2357 void *must_realloc(void *orig, size_t sz)
2358 {
2359         void *ret;
2360
2361         do {
2362                 ret = realloc(orig, sz);
2363         } while (!ret);
2364
2365         return ret;
2366 }
2367
2368 bool is_fs_type(const struct statfs *fs, fs_type_magic magic_val)
2369 {
2370         return (fs->f_type == (fs_type_magic)magic_val);
2371 }
2372
2373 bool has_fs_type(const char *path, fs_type_magic magic_val)
2374 {
2375         bool has_type;
2376         int ret;
2377         struct statfs sb;
2378
2379         ret = statfs(path, &sb);
2380         if (ret < 0)
2381                 return false;
2382
2383         has_type = is_fs_type(&sb, magic_val);
2384         if (!has_type && magic_val == RAMFS_MAGIC)
2385                 WARN("When the ramfs it a tmpfs statfs() might report tmpfs");
2386
2387         return has_type;
2388 }
2389
2390 bool lxc_nic_exists(char *nic)
2391 {
2392 #define __LXC_SYS_CLASS_NET_LEN 15 + IFNAMSIZ + 1
2393         char path[__LXC_SYS_CLASS_NET_LEN];
2394         int ret;
2395         struct stat sb;
2396
2397         if (!strcmp(nic, "none"))
2398                 return true;
2399
2400         ret = snprintf(path, __LXC_SYS_CLASS_NET_LEN, "/sys/class/net/%s", nic);
2401         if (ret < 0 || (size_t)ret >= __LXC_SYS_CLASS_NET_LEN)
2402                 return false;
2403
2404         ret = stat(path, &sb);
2405         if (ret < 0)
2406                 return false;
2407
2408         return true;
2409 }
2410
2411 int lxc_make_tmpfile(char *template, bool rm)
2412 {
2413         int fd, ret;
2414
2415         fd = mkstemp(template);
2416         if (fd < 0)
2417                 return -1;
2418
2419         if (!rm)
2420                 return fd;
2421
2422         ret = unlink(template);
2423         if (ret < 0) {
2424                 close(fd);
2425                 return -1;
2426         }
2427
2428         return fd;
2429 }
2430
2431 int parse_byte_size_string(const char *s, int64_t *converted)
2432 {
2433         int ret, suffix_len;
2434         long long int conv;
2435         int64_t mltpl, overflow;
2436         char *end;
2437         char dup[LXC_NUMSTRLEN64 + 2];
2438         char suffix[3] = {0};
2439
2440         if (!s || !strcmp(s, ""))
2441                 return -EINVAL;
2442
2443         end = stpncpy(dup, s, sizeof(dup) - 1);
2444         if (*end != '\0')
2445                 return -EINVAL;
2446
2447         if (isdigit(*(end - 1)))
2448                 suffix_len = 0;
2449         else if (isalpha(*(end - 1)))
2450                 suffix_len = 1;
2451         else
2452                 return -EINVAL;
2453
2454         if (suffix_len > 0 && (end - 2) == dup && !isdigit(*(end - 2)))
2455                 return -EINVAL;
2456
2457         if (suffix_len > 0 && isalpha(*(end - 2)))
2458                 suffix_len++;
2459
2460         if (suffix_len > 0) {
2461                 memcpy(suffix, end - suffix_len, suffix_len);
2462                 *(suffix + suffix_len) = '\0';
2463                 *(end - suffix_len) = '\0';
2464         }
2465         dup[lxc_char_right_gc(dup, strlen(dup))] = '\0';
2466
2467         ret = lxc_safe_long_long(dup, &conv);
2468         if (ret < 0)
2469                 return -ret;
2470
2471         if (suffix_len != 2) {
2472                 *converted = conv;
2473                 return 0;
2474         }
2475
2476         if (strcasecmp(suffix, "KB") == 0)
2477                 mltpl = 1024;
2478         else if (strcasecmp(suffix, "MB") == 0)
2479                 mltpl = 1024 * 1024;
2480         else if (strcasecmp(suffix, "GB") == 0)
2481                 mltpl = 1024 * 1024 * 1024;
2482         else
2483                 return -EINVAL;
2484
2485         overflow = conv * mltpl;
2486         if (conv != 0 && (overflow / conv) != mltpl)
2487                 return -ERANGE;
2488
2489         *converted = overflow;
2490         return 0;
2491 }
2492
2493 uint64_t lxc_find_next_power2(uint64_t n)
2494 {
2495         /* 0 is not valid input. We return 0 to the caller since 0 is not a
2496          * valid power of two.
2497          */
2498         if (n == 0)
2499                 return 0;
2500
2501         if (!(n & (n - 1)))
2502                 return n;
2503
2504         while (n & (n - 1))
2505                 n = n & (n - 1);
2506
2507         n = n << 1;
2508         return n;
2509 }
2510
2511 int lxc_set_death_signal(int signal)
2512 {
2513         int ret;
2514         pid_t ppid;
2515
2516         ret = prctl(PR_SET_PDEATHSIG, signal, 0, 0, 0);
2517
2518         /* Check whether we have been orphaned. */
2519         ppid = (pid_t)syscall(SYS_getppid);
2520         if (ppid == 1) {
2521                 pid_t self;
2522
2523                 self = lxc_raw_getpid();
2524                 ret = kill(self, SIGKILL);
2525                 if (ret < 0)
2526                         return -1;
2527         }
2528
2529         if (ret < 0) {
2530                 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal);
2531                 return -1;
2532         }
2533
2534         return 0;
2535 }
2536
2537 void remove_trailing_newlines(char *l)
2538 {
2539         char *p = l;
2540
2541         while (*p)
2542                 p++;
2543
2544         while (--p >= l && *p == '\n')
2545                 *p = '\0';
2546 }