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 "parse.h"
  51 #include "utils.h"
  52
  53 #ifndef O_PATH
  54 #define O_PATH      010000000
  55 #endif
  56
  57 #ifndef O_NOFOLLOW
  58 #define O_NOFOLLOW  00400000
  59 #endif
  60
  61 lxc_log_define(lxc_utils, lxc);
  62
  63 /*
  64  * if path is btrfs, tries to remove it and any subvolumes beneath it
  65  */
  66 extern bool btrfs_try_remove_subvol(const char *path);
  67
  68 static int _recursive_rmdir(const char *dirname, dev_t pdev,
  69                             const char *exclude, int level, bool onedev)
  70 {
  71         struct dirent *direntp;
  72         DIR *dir;
  73         int ret, failed=0;
  74         char pathname[MAXPATHLEN];
  75         bool hadexclude = false;
  76
  77         dir = opendir(dirname);
  78         if (!dir) {
  79                 ERROR("failed to open %s", dirname);
  80                 return -1;
  81         }
  82
  83         while ((direntp = readdir(dir))) {
  84                 struct stat mystat;
  85                 int rc;
  86
  87                 if (!direntp)
  88                         break;
  89
  90                 if (!strcmp(direntp->d_name, ".") ||
  91                     !strcmp(direntp->d_name, ".."))
  92                         continue;
  93
  94                 rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name);
  95                 if (rc < 0 || rc >= MAXPATHLEN) {
  96                         ERROR("pathname too long");
  97                         failed=1;
  98                         continue;
  99                 }
 100
 101                 if (!level && exclude && !strcmp(direntp->d_name, exclude)) {
 102                         ret = rmdir(pathname);
 103                         if (ret < 0) {
 104                                 switch(errno) {
 105                                 case ENOTEMPTY:
 106                                         INFO("Not deleting snapshot %s", pathname);
 107                                         hadexclude = true;
 108                                         break;
 109                                 case ENOTDIR:
 110                                         ret = unlink(pathname);
 111                                         if (ret)
 112                                                 INFO("Failed to remove %s", pathname);
 113                                         break;
 114                                 default:
 115                                         SYSERROR("Failed to rmdir %s", pathname);
 116                                         failed = 1;
 117                                         break;
 118                                 }
 119                         }
 120                         continue;
 121                 }
 122
 123                 ret = lstat(pathname, &mystat);
 124                 if (ret) {
 125                         ERROR("Failed to stat %s", pathname);
 126                         failed = 1;
 127                         continue;
 128                 }
 129                 if (onedev && mystat.st_dev != pdev) {
 130                         /* TODO should we be checking /proc/self/mountinfo for
 131                          * pathname and not doing this if found? */
 132                         if (btrfs_try_remove_subvol(pathname))
 133                                 INFO("Removed btrfs subvolume at %s\n", pathname);
 134                         continue;
 135                 }
 136                 if (S_ISDIR(mystat.st_mode)) {
 137                         if (_recursive_rmdir(pathname, pdev, exclude, level+1, onedev) < 0)
 138                                 failed=1;
 139                 } else {
 140                         if (unlink(pathname) < 0) {
 141                                 SYSERROR("Failed to delete %s", pathname);
 142                                 failed=1;
 143                         }
 144                 }
 145         }
 146
 147         if (rmdir(dirname) < 0 && !btrfs_try_remove_subvol(dirname) && !hadexclude) {
 148                 ERROR("Failed to delete %s", dirname);
 149                 failed=1;
 150         }
 151
 152         ret = closedir(dir);
 153         if (ret) {
 154                 ERROR("Failed to close directory %s", dirname);
 155                 failed=1;
 156         }
 157
 158         return failed ? -1 : 0;
 159 }
 160
 161 /* We have two different magic values for overlayfs, yay. */
 162 #ifndef OVERLAYFS_SUPER_MAGIC
 163 #define OVERLAYFS_SUPER_MAGIC 0x794c764f
 164 #endif
 165
 166 #ifndef OVERLAY_SUPER_MAGIC
 167 #define OVERLAY_SUPER_MAGIC 0x794c7630
 168 #endif
 169
 170 /* In overlayfs, st_dev is unreliable. So on overlayfs we don't do the
 171  * lxc_rmdir_onedev()
 172  */
 173 static bool is_native_overlayfs(const char *path)
 174 {
 175         if (has_fs_type(path, OVERLAY_SUPER_MAGIC) ||
 176             has_fs_type(path, OVERLAYFS_SUPER_MAGIC))
 177                 return true;
 178
 179         return false;
 180 }
 181
 182 /* returns 0 on success, -1 if there were any failures */
 183 extern int lxc_rmdir_onedev(const char *path, const char *exclude)
 184 {
 185         struct stat mystat;
 186         bool onedev = true;
 187
 188         if (is_native_overlayfs(path))
 189                 onedev = false;
 190
 191         if (lstat(path, &mystat) < 0) {
 192                 if (errno == ENOENT)
 193                         return 0;
 194
 195                 ERROR("Failed to stat %s", path);
 196                 return -1;
 197         }
 198
 199         return _recursive_rmdir(path, mystat.st_dev, exclude, 0, onedev);
 200 }
 201
 202 /* borrowed from iproute2 */
 203 extern int get_u16(unsigned short *val, const char *arg, int base)
 204 {
 205         unsigned long res;
 206         char *ptr;
 207
 208         if (!arg || !*arg)
 209                 return -1;
 210
 211         errno = 0;
 212         res = strtoul(arg, &ptr, base);
 213         if (!ptr || ptr == arg || *ptr || res > 0xFFFF || errno != 0)
 214                 return -1;
 215
 216         *val = res;
 217
 218         return 0;
 219 }
 220
 221 extern int mkdir_p(const char *dir, mode_t mode)
 222 {
 223         const char *tmp = dir;
 224         const char *orig = dir;
 225         char *makeme;
 226
 227         do {
 228                 dir = tmp + strspn(tmp, "/");
 229                 tmp = dir + strcspn(dir, "/");
 230                 makeme = strndup(orig, dir - orig);
 231                 if (*makeme) {
 232                         if (mkdir(makeme, mode) && errno != EEXIST) {
 233                                 SYSERROR("failed to create directory '%s'", makeme);
 234                                 free(makeme);
 235                                 return -1;
 236                         }
 237                 }
 238                 free(makeme);
 239         } while(tmp != dir);
 240
 241         return 0;
 242 }
 243
 244 char *get_rundir()
 245 {
 246         char *rundir;
 247         const char *homedir;
 248         struct stat sb;
 249
 250         if (stat(RUNTIME_PATH, &sb) < 0) {
 251                 return NULL;
 252         }
 253
 254         if (geteuid() == sb.st_uid || getegid() == sb.st_gid) {
 255                 rundir = strdup(RUNTIME_PATH);
 256                 return rundir;
 257         }
 258
 259         rundir = getenv("XDG_RUNTIME_DIR");
 260         if (rundir) {
 261                 rundir = strdup(rundir);
 262                 return rundir;
 263         }
 264
 265         INFO("XDG_RUNTIME_DIR isn't set in the environment.");
 266         homedir = getenv("HOME");
 267         if (!homedir) {
 268                 ERROR("HOME isn't set in the environment.");
 269                 return NULL;
 270         }
 271
 272         rundir = malloc(sizeof(char) * (17 + strlen(homedir)));
 273         sprintf(rundir, "%s/.cache/lxc/run/", homedir);
 274
 275         return rundir;
 276 }
 277
 278 int wait_for_pid(pid_t pid)
 279 {
 280         int status, ret;
 281
 282 again:
 283         ret = waitpid(pid, &status, 0);
 284         if (ret == -1) {
 285                 if (errno == EINTR)
 286                         goto again;
 287                 return -1;
 288         }
 289         if (ret != pid)
 290                 goto again;
 291         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0)
 292                 return -1;
 293         return 0;
 294 }
 295
 296 int lxc_wait_for_pid_status(pid_t pid)
 297 {
 298         int status, ret;
 299
 300 again:
 301         ret = waitpid(pid, &status, 0);
 302         if (ret == -1) {
 303                 if (errno == EINTR)
 304                         goto again;
 305                 return -1;
 306         }
 307         if (ret != pid)
 308                 goto again;
 309         return status;
 310 }
 311
 312 ssize_t lxc_write_nointr(int fd, const void* buf, size_t count)
 313 {
 314         ssize_t ret;
 315 again:
 316         ret = write(fd, buf, count);
 317         if (ret < 0 && errno == EINTR)
 318                 goto again;
 319         return ret;
 320 }
 321
 322 ssize_t lxc_read_nointr(int fd, void* buf, size_t count)
 323 {
 324         ssize_t ret;
 325 again:
 326         ret = read(fd, buf, count);
 327         if (ret < 0 && errno == EINTR)
 328                 goto again;
 329         return ret;
 330 }
 331
 332 ssize_t lxc_read_nointr_expect(int fd, void* buf, size_t count, const void* expected_buf)
 333 {
 334         ssize_t ret;
 335         ret = lxc_read_nointr(fd, buf, count);
 336         if (ret <= 0)
 337                 return ret;
 338         if ((size_t)ret != count)
 339                 return -1;
 340         if (expected_buf && memcmp(buf, expected_buf, count) != 0) {
 341                 errno = EINVAL;
 342                 return -1;
 343         }
 344         return ret;
 345 }
 346
 347 #if HAVE_LIBGNUTLS
 348 #include <gnutls/gnutls.h>
 349 #include <gnutls/crypto.h>
 350
 351 __attribute__((constructor))
 352 static void gnutls_lxc_init(void)
 353 {
 354         gnutls_global_init();
 355 }
 356
 357 int sha1sum_file(char *fnam, unsigned char *digest)
 358 {
 359         char *buf;
 360         int ret;
 361         FILE *f;
 362         long flen;
 363
 364         if (!fnam)
 365                 return -1;
 366         f = fopen_cloexec(fnam, "r");
 367         if (!f) {
 368                 SYSERROR("Error opening template");
 369                 return -1;
 370         }
 371         if (fseek(f, 0, SEEK_END) < 0) {
 372                 SYSERROR("Error seeking to end of template");
 373                 fclose(f);
 374                 return -1;
 375         }
 376         if ((flen = ftell(f)) < 0) {
 377                 SYSERROR("Error telling size of template");
 378                 fclose(f);
 379                 return -1;
 380         }
 381         if (fseek(f, 0, SEEK_SET) < 0) {
 382                 SYSERROR("Error seeking to start of template");
 383                 fclose(f);
 384                 return -1;
 385         }
 386         if ((buf = malloc(flen+1)) == NULL) {
 387                 SYSERROR("Out of memory");
 388                 fclose(f);
 389                 return -1;
 390         }
 391         if (fread(buf, 1, flen, f) != flen) {
 392                 SYSERROR("Failure reading template");
 393                 free(buf);
 394                 fclose(f);
 395                 return -1;
 396         }
 397         if (fclose(f) < 0) {
 398                 SYSERROR("Failre closing template");
 399                 free(buf);
 400                 return -1;
 401         }
 402         buf[flen] = '\0';
 403         ret = gnutls_hash_fast(GNUTLS_DIG_SHA1, buf, flen, (void *)digest);
 404         free(buf);
 405         return ret;
 406 }
 407 #endif
 408
 409 char** lxc_va_arg_list_to_argv(va_list ap, size_t skip, int do_strdup)
 410 {
 411         va_list ap2;
 412         size_t count = 1 + skip;
 413         char **result;
 414
 415         /* first determine size of argument list, we don't want to reallocate
 416          * constantly...
 417          */
 418         va_copy(ap2, ap);
 419         while (1) {
 420                 char* arg = va_arg(ap2, char*);
 421                 if (!arg)
 422                         break;
 423                 count++;
 424         }
 425         va_end(ap2);
 426
 427         result = calloc(count, sizeof(char*));
 428         if (!result)
 429                 return NULL;
 430         count = skip;
 431         while (1) {
 432                 char* arg = va_arg(ap, char*);
 433                 if (!arg)
 434                         break;
 435                 arg = do_strdup ? strdup(arg) : arg;
 436                 if (!arg)
 437                         goto oom;
 438                 result[count++] = arg;
 439         }
 440
 441         /* calloc has already set last element to NULL*/
 442         return result;
 443
 444 oom:
 445         free(result);
 446         return NULL;
 447 }
 448
 449 const char** lxc_va_arg_list_to_argv_const(va_list ap, size_t skip)
 450 {
 451         return (const char**)lxc_va_arg_list_to_argv(ap, skip, 0);
 452 }
 453
 454 struct lxc_popen_FILE *lxc_popen(const char *command)
 455 {
 456         int ret;
 457         int pipe_fds[2];
 458         pid_t child_pid;
 459         struct lxc_popen_FILE *fp = NULL;
 460
 461         ret = pipe2(pipe_fds, O_CLOEXEC);
 462         if (ret < 0)
 463                 return NULL;
 464
 465         child_pid = fork();
 466         if (child_pid < 0)
 467                 goto on_error;
 468
 469         if (!child_pid) {
 470                 sigset_t mask;
 471
 472                 close(pipe_fds[0]);
 473
 474                 /* duplicate stdout */
 475                 if (pipe_fds[1] != STDOUT_FILENO)
 476                         ret = dup2(pipe_fds[1], STDOUT_FILENO);
 477                 else
 478                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 479                 if (ret < 0) {
 480                         close(pipe_fds[1]);
 481                         _exit(EXIT_FAILURE);
 482                 }
 483
 484                 /* duplicate stderr */
 485                 if (pipe_fds[1] != STDERR_FILENO)
 486                         ret = dup2(pipe_fds[1], STDERR_FILENO);
 487                 else
 488                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 489                 close(pipe_fds[1]);
 490                 if (ret < 0)
 491                         _exit(EXIT_FAILURE);
 492
 493                 /* unblock all signals */
 494                 ret = sigfillset(&mask);
 495                 if (ret < 0)
 496                         _exit(EXIT_FAILURE);
 497
 498                 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
 499                 if (ret < 0)
 500                         _exit(EXIT_FAILURE);
 501
 502                 execl("/bin/sh", "sh", "-c", command, (char *)NULL);
 503                 _exit(127);
 504         }
 505
 506         close(pipe_fds[1]);
 507         pipe_fds[1] = -1;
 508
 509         fp = malloc(sizeof(*fp));
 510         if (!fp)
 511                 goto on_error;
 512         memset(fp, 0, sizeof(*fp));
 513
 514         fp->child_pid = child_pid;
 515         fp->pipe = pipe_fds[0];
 516
 517         /* From now on, closing fp->f will also close fp->pipe. So only ever
 518          * call fclose(fp->f).
 519          */
 520         fp->f = fdopen(pipe_fds[0], "r");
 521         if (!fp->f)
 522                 goto on_error;
 523
 524         return fp;
 525
 526 on_error:
 527         /* We can only close pipe_fds[0] if fdopen() didn't succeed or wasn't
 528          * called yet. Otherwise the fd belongs to the file opened by fdopen()
 529          * since it isn't dup()ed.
 530          */
 531         if (fp && !fp->f && pipe_fds[0] >= 0)
 532                 close(pipe_fds[0]);
 533
 534         if (pipe_fds[1] >= 0)
 535                 close(pipe_fds[1]);
 536
 537         if (fp && fp->f)
 538                 fclose(fp->f);
 539
 540         if (fp)
 541                 free(fp);
 542
 543         return NULL;
 544 }
 545
 546 int lxc_pclose(struct lxc_popen_FILE *fp)
 547 {
 548         pid_t wait_pid;
 549         int wstatus = 0;
 550
 551         if (!fp)
 552                 return -1;
 553
 554         do {
 555                 wait_pid = waitpid(fp->child_pid, &wstatus, 0);
 556         } while (wait_pid < 0 && errno == EINTR);
 557
 558         fclose(fp->f);
 559         free(fp);
 560
 561         if (wait_pid < 0)
 562                 return -1;
 563
 564         return wstatus;
 565 }
 566
 567 char *lxc_string_replace(const char *needle, const char *replacement, const char *haystack)
 568 {
 569         ssize_t len = -1, saved_len = -1;
 570         char *result = NULL;
 571         size_t replacement_len = strlen(replacement);
 572         size_t needle_len = strlen(needle);
 573
 574         /* should be executed exactly twice */
 575         while (len == -1 || result == NULL) {
 576                 char *p;
 577                 char *last_p;
 578                 ssize_t part_len;
 579
 580                 if (len != -1) {
 581                         result = calloc(1, len + 1);
 582                         if (!result)
 583                                 return NULL;
 584                         saved_len = len;
 585                 }
 586
 587                 len = 0;
 588
 589                 for (last_p = (char *)haystack, p = strstr(last_p, needle); p; last_p = p, p = strstr(last_p, needle)) {
 590                         part_len = (ssize_t)(p - last_p);
 591                         if (result && part_len > 0)
 592                                 memcpy(&result[len], last_p, part_len);
 593                         len += part_len;
 594                         if (result && replacement_len > 0)
 595                                 memcpy(&result[len], replacement, replacement_len);
 596                         len += replacement_len;
 597                         p += needle_len;
 598                 }
 599                 part_len = strlen(last_p);
 600                 if (result && part_len > 0)
 601                         memcpy(&result[len], last_p, part_len);
 602                 len += part_len;
 603         }
 604
 605         /* make sure we did the same thing twice,
 606          * once for calculating length, the other
 607          * time for copying data */
 608         if (saved_len != len) {
 609                 free(result);
 610                 return NULL;
 611         }
 612         /* make sure we didn't overwrite any buffer,
 613          * due to calloc the string should be 0-terminated */
 614         if (result[len] != '\0') {
 615                 free(result);
 616                 return NULL;
 617         }
 618
 619         return result;
 620 }
 621
 622 bool lxc_string_in_array(const char *needle, const char **haystack)
 623 {
 624         for (; haystack && *haystack; haystack++)
 625                 if (!strcmp(needle, *haystack))
 626                         return true;
 627         return false;
 628 }
 629
 630 char *lxc_string_join(const char *sep, const char **parts, bool use_as_prefix)
 631 {
 632         char *result;
 633         char **p;
 634         size_t sep_len = strlen(sep);
 635         size_t result_len = use_as_prefix * sep_len;
 636
 637         /* calculate new string length */
 638         for (p = (char **)parts; *p; p++)
 639                 result_len += (p > (char **)parts) * sep_len + strlen(*p);
 640
 641         result = calloc(result_len + 1, 1);
 642         if (!result)
 643                 return NULL;
 644
 645         if (use_as_prefix)
 646                 strcpy(result, sep);
 647         for (p = (char **)parts; *p; p++) {
 648                 if (p > (char **)parts)
 649                         strcat(result, sep);
 650                 strcat(result, *p);
 651         }
 652
 653         return result;
 654 }
 655
 656 char **lxc_normalize_path(const char *path)
 657 {
 658         char **components;
 659         char **p;
 660         size_t components_len = 0;
 661         size_t pos = 0;
 662
 663         components = lxc_string_split(path, '/');
 664         if (!components)
 665                 return NULL;
 666         for (p = components; *p; p++)
 667                 components_len++;
 668
 669         /* resolve '.' and '..' */
 670         for (pos = 0; pos < components_len; ) {
 671                 if (!strcmp(components[pos], ".") || (!strcmp(components[pos], "..") && pos == 0)) {
 672                         /* eat this element */
 673                         free(components[pos]);
 674                         memmove(&components[pos], &components[pos+1], sizeof(char *) * (components_len - pos));
 675                         components_len--;
 676                 } else if (!strcmp(components[pos], "..")) {
 677                         /* eat this and the previous element */
 678                         free(components[pos - 1]);
 679                         free(components[pos]);
 680                         memmove(&components[pos-1], &components[pos+1], sizeof(char *) * (components_len - pos));
 681                         components_len -= 2;
 682                         pos--;
 683                 } else {
 684                         pos++;
 685                 }
 686         }
 687
 688         return components;
 689 }
 690
 691 char *lxc_deslashify(const char *path)
 692 {
 693         char *dup, *p;
 694         char **parts = NULL;
 695         size_t n, len;
 696
 697         dup = strdup(path);
 698         if (!dup)
 699                 return NULL;
 700
 701         parts = lxc_normalize_path(dup);
 702         if (!parts) {
 703                 free(dup);
 704                 return NULL;
 705         }
 706
 707         /* We'll end up here if path == "///" or path == "". */
 708         if (!*parts) {
 709                 len = strlen(dup);
 710                 if (!len) {
 711                         lxc_free_array((void **)parts, free);
 712                         return dup;
 713                 }
 714                 n = strcspn(dup, "/");
 715                 if (n == len) {
 716                         free(dup);
 717                         lxc_free_array((void **)parts, free);
 718
 719                         p = strdup("/");
 720                         if (!p)
 721                                 return NULL;
 722
 723                         return p;
 724                 }
 725         }
 726
 727         p = lxc_string_join("/", (const char **)parts, *dup == '/');
 728         free(dup);
 729         lxc_free_array((void **)parts, free);
 730         return p;
 731 }
 732
 733 char *lxc_append_paths(const char *first, const char *second)
 734 {
 735         int ret;
 736         size_t len;
 737         char *result = NULL;
 738         const char *pattern = "%s%s";
 739
 740         len = strlen(first) + strlen(second) + 1;
 741         if (second[0] != '/') {
 742                 len += 1;
 743                 pattern = "%s/%s";
 744         }
 745
 746         result = calloc(1, len);
 747         if (!result)
 748                 return NULL;
 749
 750         ret = snprintf(result, len, pattern, first, second);
 751         if (ret < 0 || (size_t)ret >= len) {
 752                 free(result);
 753                 return NULL;
 754         }
 755
 756         return result;
 757 }
 758
 759 bool lxc_string_in_list(const char *needle, const char *haystack, char _sep)
 760 {
 761         char *token, *str, *saveptr = NULL;
 762         char sep[2] = { _sep, '\0' };
 763
 764         if (!haystack || !needle)
 765                 return 0;
 766
 767         str = alloca(strlen(haystack)+1);
 768         strcpy(str, haystack);
 769         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 770                 if (strcmp(needle, token) == 0)
 771                         return 1;
 772         }
 773
 774         return 0;
 775 }
 776
 777 char **lxc_string_split(const char *string, char _sep)
 778 {
 779         char *token, *str, *saveptr = NULL;
 780         char sep[2] = {_sep, '\0'};
 781         char **tmp = NULL, **result = NULL;
 782         size_t result_capacity = 0;
 783         size_t result_count = 0;
 784         int r, saved_errno;
 785
 786         if (!string)
 787                 return calloc(1, sizeof(char *));
 788
 789         str = alloca(strlen(string) + 1);
 790         strcpy(str, string);
 791         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 792                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 793                 if (r < 0)
 794                         goto error_out;
 795                 result[result_count] = strdup(token);
 796                 if (!result[result_count])
 797                         goto error_out;
 798                 result_count++;
 799         }
 800
 801         /* if we allocated too much, reduce it */
 802         tmp = realloc(result, (result_count + 1) * sizeof(char *));
 803         if (!tmp)
 804                 goto error_out;
 805         result = tmp;
 806         /* Make sure we don't return uninitialized memory. */
 807         if (result_count == 0)
 808                 *result = NULL;
 809         return result;
 810 error_out:
 811         saved_errno = errno;
 812         lxc_free_array((void **)result, free);
 813         errno = saved_errno;
 814         return NULL;
 815 }
 816
 817 static bool complete_word(char ***result, char *start, char *end, size_t *cap, size_t *cnt)
 818 {
 819         int r;
 820
 821         r = lxc_grow_array((void ***)result, cap, 2 + *cnt, 16);
 822         if (r < 0)
 823                 return false;
 824         (*result)[*cnt] = strndup(start, end - start);
 825         if (!(*result)[*cnt])
 826                 return false;
 827         (*cnt)++;
 828
 829         return true;
 830 }
 831
 832 /*
 833  * Given a a string 'one two "three four"', split into three words,
 834  * one, two, and "three four"
 835  */
 836 char **lxc_string_split_quoted(char *string)
 837 {
 838         char *nextword = string, *p, state;
 839         char **result = NULL;
 840         size_t result_capacity = 0;
 841         size_t result_count = 0;
 842
 843         if (!string || !*string)
 844                 return calloc(1, sizeof(char *));
 845
 846         // TODO I'm *not* handling escaped quote
 847         state = ' ';
 848         for (p = string; *p; p++) {
 849                 switch(state) {
 850                 case ' ':
 851                         if (isspace(*p))
 852                                 continue;
 853                         else if (*p == '"' || *p == '\'') {
 854                                 nextword = p;
 855                                 state = *p;
 856                                 continue;
 857                         }
 858                         nextword = p;
 859                         state = 'a';
 860                         continue;
 861                 case 'a':
 862                         if (isspace(*p)) {
 863                                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 864                                 state = ' ';
 865                                 continue;
 866                         }
 867                         continue;
 868                 case '"':
 869                 case '\'':
 870                         if (*p == state) {
 871                                 complete_word(&result, nextword+1, p, &result_capacity, &result_count);
 872                                 state = ' ';
 873                                 continue;
 874                         }
 875                         continue;
 876                 }
 877         }
 878
 879         if (state == 'a')
 880                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 881
 882         return realloc(result, (result_count + 1) * sizeof(char *));
 883 }
 884
 885 char **lxc_string_split_and_trim(const char *string, char _sep)
 886 {
 887         char *token, *str, *saveptr = NULL;
 888         char sep[2] = { _sep, '\0' };
 889         char **result = NULL;
 890         size_t result_capacity = 0;
 891         size_t result_count = 0;
 892         int r, saved_errno;
 893         size_t i = 0;
 894
 895         if (!string)
 896                 return calloc(1, sizeof(char *));
 897
 898         str = alloca(strlen(string)+1);
 899         strcpy(str, string);
 900         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 901                 while (token[0] == ' ' || token[0] == '\t')
 902                         token++;
 903                 i = strlen(token);
 904                 while (i > 0 && (token[i - 1] == ' ' || token[i - 1] == '\t')) {
 905                         token[i - 1] = '\0';
 906                         i--;
 907                 }
 908                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 909                 if (r < 0)
 910                         goto error_out;
 911                 result[result_count] = strdup(token);
 912                 if (!result[result_count])
 913                         goto error_out;
 914                 result_count++;
 915         }
 916
 917         /* if we allocated too much, reduce it */
 918         return realloc(result, (result_count + 1) * sizeof(char *));
 919 error_out:
 920         saved_errno = errno;
 921         lxc_free_array((void **)result, free);
 922         errno = saved_errno;
 923         return NULL;
 924 }
 925
 926 void lxc_free_array(void **array, lxc_free_fn element_free_fn)
 927 {
 928         void **p;
 929         for (p = array; p && *p; p++)
 930                 element_free_fn(*p);
 931         free((void*)array);
 932 }
 933
 934 int lxc_grow_array(void ***array, size_t* capacity, size_t new_size, size_t capacity_increment)
 935 {
 936         size_t new_capacity;
 937         void **new_array;
 938
 939         /* first time around, catch some trivial mistakes of the user
 940          * only initializing one of these */
 941         if (!*array || !*capacity) {
 942                 *array = NULL;
 943                 *capacity = 0;
 944         }
 945
 946         new_capacity = *capacity;
 947         while (new_size + 1 > new_capacity)
 948                 new_capacity += capacity_increment;
 949         if (new_capacity != *capacity) {
 950                 /* we have to reallocate */
 951                 new_array = realloc(*array, new_capacity * sizeof(void *));
 952                 if (!new_array)
 953                         return -1;
 954                 memset(&new_array[*capacity], 0, (new_capacity - (*capacity)) * sizeof(void *));
 955                 *array = new_array;
 956                 *capacity = new_capacity;
 957         }
 958
 959         /* array has sufficient elements */
 960         return 0;
 961 }
 962
 963 size_t lxc_array_len(void **array)
 964 {
 965         void **p;
 966         size_t result = 0;
 967
 968         for (p = array; p && *p; p++)
 969                 result++;
 970
 971         return result;
 972 }
 973
 974 int lxc_write_to_file(const char *filename, const void* buf, size_t count, bool add_newline)
 975 {
 976         int fd, saved_errno;
 977         ssize_t ret;
 978
 979         fd = open(filename, O_WRONLY | O_TRUNC | O_CREAT | O_CLOEXEC, 0666);
 980         if (fd < 0)
 981                 return -1;
 982         ret = lxc_write_nointr(fd, buf, count);
 983         if (ret < 0)
 984                 goto out_error;
 985         if ((size_t)ret != count)
 986                 goto out_error;
 987         if (add_newline) {
 988                 ret = lxc_write_nointr(fd, "\n", 1);
 989                 if (ret != 1)
 990                         goto out_error;
 991         }
 992         close(fd);
 993         return 0;
 994
 995 out_error:
 996         saved_errno = errno;
 997         close(fd);
 998         errno = saved_errno;
 999         return -1;
1000 }
1001
1002 int lxc_read_from_file(const char *filename, void* buf, size_t count)
1003 {
1004         int fd = -1, saved_errno;
1005         ssize_t ret;
1006
1007         fd = open(filename, O_RDONLY | O_CLOEXEC);
1008         if (fd < 0)
1009                 return -1;
1010
1011         if (!buf || !count) {
1012                 char buf2[100];
1013                 size_t count2 = 0;
1014                 while ((ret = read(fd, buf2, 100)) > 0)
1015                         count2 += ret;
1016                 if (ret >= 0)
1017                         ret = count2;
1018         } else {
1019                 memset(buf, 0, count);
1020                 ret = read(fd, buf, count);
1021         }
1022
1023         if (ret < 0)
1024                 ERROR("read %s: %s", filename, strerror(errno));
1025
1026         saved_errno = errno;
1027         close(fd);
1028         errno = saved_errno;
1029         return ret;
1030 }
1031
1032 void **lxc_append_null_to_array(void **array, size_t count)
1033 {
1034         void **temp;
1035
1036         /* Append NULL to the array */
1037         if (count) {
1038                 temp = realloc(array, (count + 1) * sizeof(*array));
1039                 if (!temp) {
1040                         size_t i;
1041                         for (i = 0; i < count; i++)
1042                                 free(array[i]);
1043                         free(array);
1044                         return NULL;
1045                 }
1046                 array = temp;
1047                 array[count] = NULL;
1048         }
1049         return array;
1050 }
1051
1052 int randseed(bool srand_it)
1053 {
1054         /*
1055            srand pre-seed function based on /dev/urandom
1056            */
1057         unsigned int seed = time(NULL) + getpid();
1058
1059         FILE *f;
1060         f = fopen("/dev/urandom", "r");
1061         if (f) {
1062                 int ret = fread(&seed, sizeof(seed), 1, f);
1063                 if (ret != 1)
1064                         DEBUG("unable to fread /dev/urandom, %s, fallback to time+pid rand seed", strerror(errno));
1065                 fclose(f);
1066         }
1067
1068         if (srand_it)
1069                 srand(seed);
1070
1071         return seed;
1072 }
1073
1074 uid_t get_ns_uid(uid_t orig)
1075 {
1076         char *line = NULL;
1077         size_t sz = 0;
1078         uid_t nsid, hostid, range;
1079         FILE *f = fopen("/proc/self/uid_map", "r");
1080         if (!f)
1081                 return 0;
1082
1083         while (getline(&line, &sz, f) != -1) {
1084                 if (sscanf(line, "%u %u %u", &nsid, &hostid, &range) != 3)
1085                         continue;
1086                 if (hostid <= orig && hostid + range > orig) {
1087                         nsid += orig - hostid;
1088                         goto found;
1089                 }
1090         }
1091
1092         nsid = 0;
1093 found:
1094         fclose(f);
1095         free(line);
1096         return nsid;
1097 }
1098
1099 bool dir_exists(const char *path)
1100 {
1101         struct stat sb;
1102         int ret;
1103
1104         ret = stat(path, &sb);
1105         if (ret < 0)
1106                 /* Could be something other than eexist, just say "no". */
1107                 return false;
1108         return S_ISDIR(sb.st_mode);
1109 }
1110
1111 /* Note we don't use SHA-1 here as we don't want to depend on HAVE_GNUTLS.
1112  * FNV has good anti collision properties and we're not worried
1113  * about pre-image resistance or one-way-ness, we're just trying to make
1114  * the name unique in the 108 bytes of space we have.
1115  */
1116 uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
1117 {
1118         unsigned char *bp;
1119
1120         for(bp = buf; bp < (unsigned char *)buf + len; bp++)
1121         {
1122                 /* xor the bottom with the current octet */
1123                 hval ^= (uint64_t)*bp;
1124
1125                 /* gcc optimised:
1126                  * multiply by the 64 bit FNV magic prime mod 2^64
1127                  */
1128                 hval += (hval << 1) + (hval << 4) + (hval << 5) +
1129                         (hval << 7) + (hval << 8) + (hval << 40);
1130         }
1131
1132         return hval;
1133 }
1134
1135 /*
1136  * Detect whether / is mounted MS_SHARED.  The only way I know of to
1137  * check that is through /proc/self/mountinfo.
1138  * I'm only checking for /.  If the container rootfs or mount location
1139  * is MS_SHARED, but not '/', then you're out of luck - figuring that
1140  * out would be too much work to be worth it.
1141  */
1142 int detect_shared_rootfs(void)
1143 {
1144         char buf[LXC_LINELEN], *p;
1145         FILE *f;
1146         int i;
1147         char *p2;
1148
1149         f = fopen("/proc/self/mountinfo", "r");
1150         if (!f)
1151                 return 0;
1152         while (fgets(buf, LXC_LINELEN, f)) {
1153                 for (p = buf, 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 shared? */
1163                         p = strchr(p2 + 1, ' ');
1164                         if (p && strstr(p, "shared:")) {
1165                                 fclose(f);
1166                                 return 1;
1167                         }
1168                 }
1169         }
1170         fclose(f);
1171         return 0;
1172 }
1173
1174 bool switch_to_ns(pid_t pid, const char *ns) {
1175         int fd, ret;
1176         char nspath[MAXPATHLEN];
1177
1178         /* Switch to new ns */
1179         ret = snprintf(nspath, MAXPATHLEN, "/proc/%d/ns/%s", pid, ns);
1180         if (ret < 0 || ret >= MAXPATHLEN)
1181                 return false;
1182
1183         fd = open(nspath, O_RDONLY);
1184         if (fd < 0) {
1185                 SYSERROR("failed to open %s", nspath);
1186                 return false;
1187         }
1188
1189         ret = setns(fd, 0);
1190         if (ret) {
1191                 SYSERROR("failed to set process %d to %s of %d.", pid, ns, fd);
1192                 close(fd);
1193                 return false;
1194         }
1195         close(fd);
1196         return true;
1197 }
1198
1199 /*
1200  * looking at fs/proc_namespace.c, it appears we can
1201  * actually expect the rootfs entry to very specifically contain
1202  * " - rootfs rootfs "
1203  * IIUC, so long as we've chrooted so that rootfs is not our root,
1204  * the rootfs entry should always be skipped in mountinfo contents.
1205  */
1206 bool detect_ramfs_rootfs(void)
1207 {
1208         FILE *f;
1209         char *p, *p2;
1210         char *line = NULL;
1211         size_t len = 0;
1212         int i;
1213
1214         f = fopen("/proc/self/mountinfo", "r");
1215         if (!f)
1216                 return false;
1217
1218         while (getline(&line, &len, f) != -1) {
1219                 for (p = line, i = 0; p && i < 4; i++)
1220                         p = strchr(p + 1, ' ');
1221                 if (!p)
1222                         continue;
1223                 p2 = strchr(p + 1, ' ');
1224                 if (!p2)
1225                         continue;
1226                 *p2 = '\0';
1227                 if (strcmp(p + 1, "/") == 0) {
1228                         /* This is '/'. Is it the ramfs? */
1229                         p = strchr(p2 + 1, '-');
1230                         if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) {
1231                                 free(line);
1232                                 fclose(f);
1233                                 return true;
1234                         }
1235                 }
1236         }
1237         free(line);
1238         fclose(f);
1239         return false;
1240 }
1241
1242 char *on_path(const char *cmd, const char *rootfs) {
1243         char *path = NULL;
1244         char *entry = NULL;
1245         char *saveptr = NULL;
1246         char cmdpath[MAXPATHLEN];
1247         int ret;
1248
1249         path = getenv("PATH");
1250         if (!path)
1251                 return NULL;
1252
1253         path = strdup(path);
1254         if (!path)
1255                 return NULL;
1256
1257         entry = strtok_r(path, ":", &saveptr);
1258         while (entry) {
1259                 if (rootfs)
1260                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s/%s", rootfs, entry, cmd);
1261                 else
1262                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s", entry, cmd);
1263
1264                 if (ret < 0 || ret >= MAXPATHLEN)
1265                         goto next_loop;
1266
1267                 if (access(cmdpath, X_OK) == 0) {
1268                         free(path);
1269                         return strdup(cmdpath);
1270                 }
1271
1272 next_loop:
1273                 entry = strtok_r(NULL, ":", &saveptr);
1274         }
1275
1276         free(path);
1277         return NULL;
1278 }
1279
1280 bool file_exists(const char *f)
1281 {
1282         struct stat statbuf;
1283
1284         return stat(f, &statbuf) == 0;
1285 }
1286
1287 bool cgns_supported(void)
1288 {
1289         return file_exists("/proc/self/ns/cgroup");
1290 }
1291
1292 /* historically lxc-init has been under /usr/lib/lxc and under
1293  * /usr/lib/$ARCH/lxc.  It now lives as $prefix/sbin/init.lxc.
1294  */
1295 char *choose_init(const char *rootfs)
1296 {
1297         char *retv = NULL;
1298         const char *empty = "",
1299                    *tmp;
1300         int ret, env_set = 0;
1301
1302         if (!getenv("PATH")) {
1303                 if (setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 0))
1304                         SYSERROR("Failed to setenv");
1305                 env_set = 1;
1306         }
1307
1308         retv = on_path("init.lxc", rootfs);
1309
1310         if (env_set) {
1311                 if (unsetenv("PATH"))
1312                         SYSERROR("Failed to unsetenv");
1313         }
1314
1315         if (retv)
1316                 return retv;
1317
1318         retv = malloc(PATH_MAX);
1319         if (!retv)
1320                 return NULL;
1321
1322         if (rootfs)
1323                 tmp = rootfs;
1324         else
1325                 tmp = empty;
1326
1327         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, SBINDIR, "/init.lxc");
1328         if (ret < 0 || ret >= PATH_MAX) {
1329                 ERROR("pathname too long");
1330                 goto out1;
1331         }
1332         if (access(retv, X_OK) == 0)
1333                 return retv;
1334
1335         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, LXCINITDIR, "/lxc/lxc-init");
1336         if (ret < 0 || ret >= PATH_MAX) {
1337                 ERROR("pathname too long");
1338                 goto out1;
1339         }
1340         if (access(retv, X_OK) == 0)
1341                 return retv;
1342
1343         ret = snprintf(retv, PATH_MAX, "%s/usr/lib/lxc/lxc-init", tmp);
1344         if (ret < 0 || ret >= PATH_MAX) {
1345                 ERROR("pathname too long");
1346                 goto out1;
1347         }
1348         if (access(retv, X_OK) == 0)
1349                 return retv;
1350
1351         ret = snprintf(retv, PATH_MAX, "%s/sbin/lxc-init", tmp);
1352         if (ret < 0 || ret >= PATH_MAX) {
1353                 ERROR("pathname too long");
1354                 goto out1;
1355         }
1356         if (access(retv, X_OK) == 0)
1357                 return retv;
1358
1359         /*
1360          * Last resort, look for the statically compiled init.lxc which we
1361          * hopefully bind-mounted in.
1362          * If we are called during container setup, and we get to this point,
1363          * then the init.lxc.static from the host will need to be bind-mounted
1364          * in.  So we return NULL here to indicate that.
1365          */
1366         if (rootfs)
1367                 goto out1;
1368
1369         ret = snprintf(retv, PATH_MAX, "/init.lxc.static");
1370         if (ret < 0 || ret >= PATH_MAX) {
1371                 WARN("Nonsense - name /lxc.init.static too long");
1372                 goto out1;
1373         }
1374         if (access(retv, X_OK) == 0)
1375                 return retv;
1376
1377 out1:
1378         free(retv);
1379         return NULL;
1380 }
1381
1382 int print_to_file(const char *file, const char *content)
1383 {
1384         FILE *f;
1385         int ret = 0;
1386
1387         f = fopen(file, "w");
1388         if (!f)
1389                 return -1;
1390         if (fprintf(f, "%s", content) != strlen(content))
1391                 ret = -1;
1392         fclose(f);
1393         return ret;
1394 }
1395
1396 int is_dir(const char *path)
1397 {
1398         struct stat statbuf;
1399         int ret = stat(path, &statbuf);
1400         if (ret == 0 && S_ISDIR(statbuf.st_mode))
1401                 return 1;
1402         return 0;
1403 }
1404
1405 /*
1406  * Given the '-t' template option to lxc-create, figure out what to
1407  * do.  If the template is a full executable path, use that.  If it
1408  * is something like 'sshd', then return $templatepath/lxc-sshd.
1409  * On success return the template, on error return NULL.
1410  */
1411 char *get_template_path(const char *t)
1412 {
1413         int ret, len;
1414         char *tpath;
1415
1416         if (t[0] == '/' && access(t, X_OK) == 0) {
1417                 tpath = strdup(t);
1418                 return tpath;
1419         }
1420
1421         len = strlen(LXCTEMPLATEDIR) + strlen(t) + strlen("/lxc-") + 1;
1422         tpath = malloc(len);
1423         if (!tpath)
1424                 return NULL;
1425         ret = snprintf(tpath, len, "%s/lxc-%s", LXCTEMPLATEDIR, t);
1426         if (ret < 0 || ret >= len) {
1427                 free(tpath);
1428                 return NULL;
1429         }
1430         if (access(tpath, X_OK) < 0) {
1431                 SYSERROR("bad template: %s", t);
1432                 free(tpath);
1433                 return NULL;
1434         }
1435
1436         return tpath;
1437 }
1438
1439 /*
1440  * @path:    a pathname where / replaced with '\0'.
1441  * @offsetp: pointer to int showing which path segment was last seen.
1442  *           Updated on return to reflect the next segment.
1443  * @fulllen: full original path length.
1444  * Returns a pointer to the next path segment, or NULL if done.
1445  */
1446 static char *get_nextpath(char *path, int *offsetp, int fulllen)
1447 {
1448         int offset = *offsetp;
1449
1450         if (offset >= fulllen)
1451                 return NULL;
1452
1453         while (path[offset] != '\0' && offset < fulllen)
1454                 offset++;
1455         while (path[offset] == '\0' && offset < fulllen)
1456                 offset++;
1457
1458         *offsetp = offset;
1459         return (offset < fulllen) ? &path[offset] : NULL;
1460 }
1461
1462 /*
1463  * Check that @subdir is a subdir of @dir.  @len is the length of
1464  * @dir (to avoid having to recalculate it).
1465  */
1466 static bool is_subdir(const char *subdir, const char *dir, size_t len)
1467 {
1468         size_t subdirlen = strlen(subdir);
1469
1470         if (subdirlen < len)
1471                 return false;
1472         if (strncmp(subdir, dir, len) != 0)
1473                 return false;
1474         if (dir[len-1] == '/')
1475                 return true;
1476         if (subdir[len] == '/' || subdirlen == len)
1477                 return true;
1478         return false;
1479 }
1480
1481 /*
1482  * Check if the open fd is a symlink.  Return -ELOOP if it is.  Return
1483  * -ENOENT if we couldn't fstat.  Return 0 if the fd is ok.
1484  */
1485 static int check_symlink(int fd)
1486 {
1487         struct stat sb;
1488         int ret = fstat(fd, &sb);
1489         if (ret < 0)
1490                 return -ENOENT;
1491         if (S_ISLNK(sb.st_mode))
1492                 return -ELOOP;
1493         return 0;
1494 }
1495
1496 /*
1497  * Open a file or directory, provided that it contains no symlinks.
1498  *
1499  * CAVEAT: This function must not be used for other purposes than container
1500  * setup before executing the container's init
1501  */
1502 static int open_if_safe(int dirfd, const char *nextpath)
1503 {
1504         int newfd = openat(dirfd, nextpath, O_RDONLY | O_NOFOLLOW);
1505         if (newfd >= 0) /* Was not a symlink, all good. */
1506                 return newfd;
1507
1508         if (errno == ELOOP)
1509                 return newfd;
1510
1511         if (errno == EPERM || errno == EACCES) {
1512                 /* We're not root (cause we got EPERM) so try opening with
1513                  * O_PATH.
1514                  */
1515                 newfd = openat(dirfd, nextpath, O_PATH | O_NOFOLLOW);
1516                 if (newfd >= 0) {
1517                         /* O_PATH will return an fd for symlinks. We know
1518                          * nextpath wasn't a symlink at last openat, so if fd is
1519                          * now a link, then something * fishy is going on.
1520                          */
1521                         int ret = check_symlink(newfd);
1522                         if (ret < 0) {
1523                                 close(newfd);
1524                                 newfd = ret;
1525                         }
1526                 }
1527         }
1528
1529         return newfd;
1530 }
1531
1532 /*
1533  * Open a path intending for mounting, ensuring that the final path
1534  * is inside the container's rootfs.
1535  *
1536  * CAVEAT: This function must not be used for other purposes than container
1537  * setup before executing the container's init
1538  *
1539  * @target: path to be opened
1540  * @prefix_skip: a part of @target in which to ignore symbolic links.  This
1541  * would be the container's rootfs.
1542  *
1543  * Return an open fd for the path, or <0 on error.
1544  */
1545 static int open_without_symlink(const char *target, const char *prefix_skip)
1546 {
1547         int curlen = 0, dirfd, fulllen, i;
1548         char *dup = NULL;
1549
1550         fulllen = strlen(target);
1551
1552         /* make sure prefix-skip makes sense */
1553         if (prefix_skip && strlen(prefix_skip) > 0) {
1554                 curlen = strlen(prefix_skip);
1555                 if (!is_subdir(target, prefix_skip, curlen)) {
1556                         ERROR("WHOA there - target '%s' didn't start with prefix '%s'",
1557                                 target, prefix_skip);
1558                         return -EINVAL;
1559                 }
1560                 /*
1561                  * get_nextpath() expects the curlen argument to be
1562                  * on a  (turned into \0) / or before it, so decrement
1563                  * curlen to make sure that happens
1564                  */
1565                 if (curlen)
1566                         curlen--;
1567         } else {
1568                 prefix_skip = "/";
1569                 curlen = 0;
1570         }
1571
1572         /* Make a copy of target which we can hack up, and tokenize it */
1573         if ((dup = strdup(target)) == NULL) {
1574                 SYSERROR("Out of memory checking for symbolic link");
1575                 return -ENOMEM;
1576         }
1577         for (i = 0; i < fulllen; i++) {
1578                 if (dup[i] == '/')
1579                         dup[i] = '\0';
1580         }
1581
1582         dirfd = open(prefix_skip, O_RDONLY);
1583         if (dirfd < 0)
1584                 goto out;
1585         while (1) {
1586                 int newfd, saved_errno;
1587                 char *nextpath;
1588
1589                 if ((nextpath = get_nextpath(dup, &curlen, fulllen)) == NULL)
1590                         goto out;
1591                 newfd = open_if_safe(dirfd, nextpath);
1592                 saved_errno = errno;
1593                 close(dirfd);
1594                 dirfd = newfd;
1595                 if (newfd < 0) {
1596                         errno = saved_errno;
1597                         if (errno == ELOOP)
1598                                 SYSERROR("%s in %s was a symbolic link!", nextpath, target);
1599                         goto out;
1600                 }
1601         }
1602
1603 out:
1604         free(dup);
1605         return dirfd;
1606 }
1607
1608 /*
1609  * Safely mount a path into a container, ensuring that the mount target
1610  * is under the container's @rootfs.  (If @rootfs is NULL, then the container
1611  * uses the host's /)
1612  *
1613  * CAVEAT: This function must not be used for other purposes than container
1614  * setup before executing the container's init
1615  */
1616 int safe_mount(const char *src, const char *dest, const char *fstype,
1617                 unsigned long flags, const void *data, const char *rootfs)
1618 {
1619         int destfd, ret, saved_errno;
1620         /* Only needs enough for /proc/self/fd/<fd>. */
1621         char srcbuf[50], destbuf[50];
1622         int srcfd = -1;
1623         const char *mntsrc = src;
1624
1625         if (!rootfs)
1626                 rootfs = "";
1627
1628         /* todo - allow symlinks for relative paths if 'allowsymlinks' option is passed */
1629         if (flags & MS_BIND && src && src[0] != '/') {
1630                 INFO("this is a relative bind mount");
1631                 srcfd = open_without_symlink(src, NULL);
1632                 if (srcfd < 0)
1633                         return srcfd;
1634                 ret = snprintf(srcbuf, 50, "/proc/self/fd/%d", srcfd);
1635                 if (ret < 0 || ret > 50) {
1636                         close(srcfd);
1637                         ERROR("Out of memory");
1638                         return -EINVAL;
1639                 }
1640                 mntsrc = srcbuf;
1641         }
1642
1643         destfd = open_without_symlink(dest, rootfs);
1644         if (destfd < 0) {
1645                 if (srcfd != -1) {
1646                         saved_errno = errno;
1647                         close(srcfd);
1648                         errno = saved_errno;
1649                 }
1650                 return destfd;
1651         }
1652
1653         ret = snprintf(destbuf, 50, "/proc/self/fd/%d", destfd);
1654         if (ret < 0 || ret > 50) {
1655                 if (srcfd != -1)
1656                         close(srcfd);
1657                 close(destfd);
1658                 ERROR("Out of memory");
1659                 return -EINVAL;
1660         }
1661
1662         ret = mount(mntsrc, destbuf, fstype, flags, data);
1663         saved_errno = errno;
1664         if (srcfd != -1)
1665                 close(srcfd);
1666         close(destfd);
1667         if (ret < 0) {
1668                 errno = saved_errno;
1669                 SYSERROR("Failed to mount %s onto %s", src ? src : "(null)", dest);
1670                 return ret;
1671         }
1672
1673         return 0;
1674 }
1675
1676 /*
1677  * Mount a proc under @rootfs if proc self points to a pid other than
1678  * my own.  This is needed to have a known-good proc mount for setting
1679  * up LSMs both at container startup and attach.
1680  *
1681  * @rootfs : the rootfs where proc should be mounted
1682  *
1683  * Returns < 0 on failure, 0 if the correct proc was already mounted
1684  * and 1 if a new proc was mounted.
1685  *
1686  * NOTE: not to be called from inside the container namespace!
1687  */
1688 int lxc_mount_proc_if_needed(const char *rootfs)
1689 {
1690         char path[MAXPATHLEN];
1691         int link_to_pid, linklen, mypid, ret;
1692         char link[LXC_NUMSTRLEN64] = {0};
1693
1694         ret = snprintf(path, MAXPATHLEN, "%s/proc/self", rootfs);
1695         if (ret < 0 || ret >= MAXPATHLEN) {
1696                 SYSERROR("proc path name too long");
1697                 return -1;
1698         }
1699
1700         linklen = readlink(path, link, LXC_NUMSTRLEN64);
1701
1702         ret = snprintf(path, MAXPATHLEN, "%s/proc", rootfs);
1703         if (ret < 0 || ret >= MAXPATHLEN) {
1704                 SYSERROR("proc path name too long");
1705                 return -1;
1706         }
1707
1708         /* /proc not mounted */
1709         if (linklen < 0) {
1710                 if (mkdir(path, 0755) && errno != EEXIST)
1711                         return -1;
1712                 goto domount;
1713         } else if (linklen >= LXC_NUMSTRLEN64) {
1714                 link[linklen - 1] = '\0';
1715                 ERROR("readlink returned truncated content: \"%s\"", link);
1716                 return -1;
1717         }
1718
1719         mypid = lxc_raw_getpid();
1720         INFO("I am %d, /proc/self points to \"%s\"", mypid, link);
1721
1722         if (lxc_safe_int(link, &link_to_pid) < 0)
1723                 return -1;
1724
1725         /* correct procfs is already mounted */
1726         if (link_to_pid == mypid)
1727                 return 0;
1728
1729         ret = umount2(path, MNT_DETACH);
1730         if (ret < 0)
1731                 WARN("failed to umount \"%s\" with MNT_DETACH", path);
1732
1733 domount:
1734         /* rootfs is NULL */
1735         if (!strcmp(rootfs, ""))
1736                 ret = mount("proc", path, "proc", 0, NULL);
1737         else
1738                 ret = safe_mount("proc", path, "proc", 0, NULL, rootfs);
1739         if (ret < 0)
1740                 return -1;
1741
1742         INFO("mounted /proc in container for security transition");
1743         return 1;
1744 }
1745
1746 int open_devnull(void)
1747 {
1748         int fd = open("/dev/null", O_RDWR);
1749
1750         if (fd < 0)
1751                 SYSERROR("Can't open /dev/null");
1752
1753         return fd;
1754 }
1755
1756 int set_stdfds(int fd)
1757 {
1758         int ret;
1759
1760         if (fd < 0)
1761                 return -1;
1762
1763         ret = dup2(fd, STDIN_FILENO);
1764         if (ret < 0)
1765                 return -1;
1766
1767         ret = dup2(fd, STDOUT_FILENO);
1768         if (ret < 0)
1769                 return -1;
1770
1771         ret = dup2(fd, STDERR_FILENO);
1772         if (ret < 0)
1773                 return -1;
1774
1775         return 0;
1776 }
1777
1778 int null_stdfds(void)
1779 {
1780         int ret = -1;
1781         int fd = open_devnull();
1782
1783         if (fd >= 0) {
1784                 ret = set_stdfds(fd);
1785                 close(fd);
1786         }
1787
1788         return ret;
1789 }
1790
1791 /*
1792  * Return the number of lines in file @fn, or -1 on error
1793  */
1794 int lxc_count_file_lines(const char *fn)
1795 {
1796         FILE *f;
1797         char *line = NULL;
1798         size_t sz = 0;
1799         int n = 0;
1800
1801         f = fopen_cloexec(fn, "r");
1802         if (!f)
1803                 return -1;
1804
1805         while (getline(&line, &sz, f) != -1) {
1806                 n++;
1807         }
1808         free(line);
1809         fclose(f);
1810         return n;
1811 }
1812
1813 /* Check whether a signal is blocked by a process. */
1814 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
1815 #define __PROC_STATUS_LEN (5 + (LXC_NUMSTRLEN64) + 7 + 1)
1816 bool task_blocking_signal(pid_t pid, int signal)
1817 {
1818         bool bret = false;
1819         char *line = NULL;
1820         long unsigned int sigblk = 0;
1821         size_t n = 0;
1822         int ret;
1823         FILE *f;
1824
1825         char status[__PROC_STATUS_LEN];
1826
1827         ret = snprintf(status, __PROC_STATUS_LEN, "/proc/%d/status", pid);
1828         if (ret < 0 || ret >= __PROC_STATUS_LEN)
1829                 return bret;
1830
1831         f = fopen(status, "r");
1832         if (!f)
1833                 return bret;
1834
1835         while (getline(&line, &n, f) != -1) {
1836                 if (strncmp(line, "SigBlk:\t", 8))
1837                         continue;
1838
1839                 if (sscanf(line + 8, "%lx", &sigblk) != 1)
1840                         goto out;
1841         }
1842
1843         if (sigblk & (1LU << (signal - 1)))
1844                 bret = true;
1845
1846 out:
1847         free(line);
1848         fclose(f);
1849         return bret;
1850 }
1851
1852 static int lxc_append_null_to_list(void ***list)
1853 {
1854         int newentry = 0;
1855         void **tmp;
1856
1857         if (*list)
1858                 for (; (*list)[newentry]; newentry++) {
1859                         ;
1860                 }
1861
1862         tmp = realloc(*list, (newentry + 2) * sizeof(void **));
1863         if (!tmp)
1864                 return -1;
1865
1866         *list = tmp;
1867         (*list)[newentry + 1] = NULL;
1868
1869         return newentry;
1870 }
1871
1872 int lxc_append_string(char ***list, char *entry)
1873 {
1874         char *copy;
1875         int newentry;
1876
1877         newentry = lxc_append_null_to_list((void ***)list);
1878         if (newentry < 0)
1879                 return -1;
1880
1881         copy = strdup(entry);
1882         if (!copy)
1883                 return -1;
1884
1885         (*list)[newentry] = copy;
1886
1887         return 0;
1888 }
1889
1890 int lxc_preserve_ns(const int pid, const char *ns)
1891 {
1892         int ret;
1893 /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */
1894 #define __NS_PATH_LEN 50
1895         char path[__NS_PATH_LEN];
1896
1897         /* This way we can use this function to also check whether namespaces
1898          * are supported by the kernel by passing in the NULL or the empty
1899          * string.
1900          */
1901         ret = snprintf(path, __NS_PATH_LEN, "/proc/%d/ns%s%s", pid,
1902                        !ns || strcmp(ns, "") == 0 ? "" : "/",
1903                        !ns || strcmp(ns, "") == 0 ? "" : ns);
1904         errno = EFBIG;
1905         if (ret < 0 || (size_t)ret >= __NS_PATH_LEN)
1906                 return -EFBIG;
1907
1908         return open(path, O_RDONLY | O_CLOEXEC);
1909 }
1910
1911 int lxc_safe_uint(const char *numstr, unsigned int *converted)
1912 {
1913         char *err = NULL;
1914         unsigned long int uli;
1915
1916         while (isspace(*numstr))
1917                 numstr++;
1918
1919         if (*numstr == '-')
1920                 return -EINVAL;
1921
1922         errno = 0;
1923         uli = strtoul(numstr, &err, 0);
1924         if (errno == ERANGE && uli == ULONG_MAX)
1925                 return -ERANGE;
1926
1927         if (err == numstr || *err != '\0')
1928                 return -EINVAL;
1929
1930         if (uli > UINT_MAX)
1931                 return -ERANGE;
1932
1933         *converted = (unsigned int)uli;
1934         return 0;
1935 }
1936
1937 int lxc_safe_ulong(const char *numstr, unsigned long *converted)
1938 {
1939         char *err = NULL;
1940         unsigned long int uli;
1941
1942         while (isspace(*numstr))
1943                 numstr++;
1944
1945         if (*numstr == '-')
1946                 return -EINVAL;
1947
1948         errno = 0;
1949         uli = strtoul(numstr, &err, 0);
1950         if (errno == ERANGE && uli == ULONG_MAX)
1951                 return -ERANGE;
1952
1953         if (err == numstr || *err != '\0')
1954                 return -EINVAL;
1955
1956         *converted = uli;
1957         return 0;
1958 }
1959
1960 int lxc_safe_int(const char *numstr, int *converted)
1961 {
1962         char *err = NULL;
1963         signed long int sli;
1964
1965         errno = 0;
1966         sli = strtol(numstr, &err, 0);
1967         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
1968                 return -ERANGE;
1969
1970         if (errno != 0 && sli == 0)
1971                 return -EINVAL;
1972
1973         if (err == numstr || *err != '\0')
1974                 return -EINVAL;
1975
1976         if (sli > INT_MAX || sli < INT_MIN)
1977                 return -ERANGE;
1978
1979         *converted = (int)sli;
1980         return 0;
1981 }
1982
1983 int lxc_safe_long(const char *numstr, long int *converted)
1984 {
1985         char *err = NULL;
1986         signed long int sli;
1987
1988         errno = 0;
1989         sli = strtol(numstr, &err, 0);
1990         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
1991                 return -ERANGE;
1992
1993         if (errno != 0 && sli == 0)
1994                 return -EINVAL;
1995
1996         if (err == numstr || *err != '\0')
1997                 return -EINVAL;
1998
1999         *converted = sli;
2000         return 0;
2001 }
2002
2003 int lxc_safe_long_long(const char *numstr, long long int *converted)
2004 {
2005         char *err = NULL;
2006         signed long long int sli;
2007
2008         errno = 0;
2009         sli = strtoll(numstr, &err, 0);
2010         if (errno == ERANGE && (sli == LLONG_MAX || sli == LLONG_MIN))
2011                 return -ERANGE;
2012
2013         if (errno != 0 && sli == 0)
2014                 return -EINVAL;
2015
2016         if (err == numstr || *err != '\0')
2017                 return -EINVAL;
2018
2019         *converted = sli;
2020         return 0;
2021 }
2022
2023 int lxc_switch_uid_gid(uid_t uid, gid_t gid)
2024 {
2025         if (setgid(gid) < 0) {
2026                 SYSERROR("Failed to switch to gid %d.", gid);
2027                 return -errno;
2028         }
2029         NOTICE("Switched to gid %d.", gid);
2030
2031         if (setuid(uid) < 0) {
2032                 SYSERROR("Failed to switch to uid %d.", uid);
2033                 return -errno;
2034         }
2035         NOTICE("Switched to uid %d.", uid);
2036
2037         return 0;
2038 }
2039
2040 /* Simple covenience function which enables uniform logging. */
2041 int lxc_setgroups(int size, gid_t list[])
2042 {
2043         if (setgroups(size, list) < 0) {
2044                 SYSERROR("Failed to setgroups().");
2045                 return -errno;
2046         }
2047         NOTICE("Dropped additional groups.");
2048
2049         return 0;
2050 }
2051
2052 static int lxc_get_unused_loop_dev_legacy(char *loop_name)
2053 {
2054         struct dirent *dp;
2055         struct loop_info64 lo64;
2056         DIR *dir;
2057         int dfd = -1, fd = -1, ret = -1;
2058
2059         dir = opendir("/dev");
2060         if (!dir)
2061                 return -1;
2062
2063         while ((dp = readdir(dir))) {
2064                 if (!dp)
2065                         break;
2066
2067                 if (strncmp(dp->d_name, "loop", 4) != 0)
2068                         continue;
2069
2070                 dfd = dirfd(dir);
2071                 if (dfd < 0)
2072                         continue;
2073
2074                 fd = openat(dfd, dp->d_name, O_RDWR);
2075                 if (fd < 0)
2076                         continue;
2077
2078                 ret = ioctl(fd, LOOP_GET_STATUS64, &lo64);
2079                 if (ret < 0) {
2080                         if (ioctl(fd, LOOP_GET_STATUS64, &lo64) == 0 ||
2081                             errno != ENXIO) {
2082                                 close(fd);
2083                                 fd = -1;
2084                                 continue;
2085                         }
2086                 }
2087
2088                 ret = snprintf(loop_name, LO_NAME_SIZE, "/dev/%s", dp->d_name);
2089                 if (ret < 0 || ret >= LO_NAME_SIZE) {
2090                         close(fd);
2091                         fd = -1;
2092                         continue;
2093                 }
2094
2095                 break;
2096         }
2097
2098         closedir(dir);
2099
2100         if (fd < 0)
2101                 return -1;
2102
2103         return fd;
2104 }
2105
2106 static int lxc_get_unused_loop_dev(char *name_loop)
2107 {
2108         int loop_nr, ret;
2109         int fd_ctl = -1, fd_tmp = -1;
2110
2111         fd_ctl = open("/dev/loop-control", O_RDWR | O_CLOEXEC);
2112         if (fd_ctl < 0)
2113                 return -ENODEV;
2114
2115         loop_nr = ioctl(fd_ctl, LOOP_CTL_GET_FREE);
2116         if (loop_nr < 0)
2117                 goto on_error;
2118
2119         ret = snprintf(name_loop, LO_NAME_SIZE, "/dev/loop%d", loop_nr);
2120         if (ret < 0 || ret >= LO_NAME_SIZE)
2121                 goto on_error;
2122
2123         fd_tmp = open(name_loop, O_RDWR | O_CLOEXEC);
2124         if (fd_tmp < 0)
2125                 goto on_error;
2126
2127 on_error:
2128         close(fd_ctl);
2129         return fd_tmp;
2130 }
2131
2132 int lxc_prepare_loop_dev(const char *source, char *loop_dev, int flags)
2133 {
2134         int ret;
2135         struct loop_info64 lo64;
2136         int fd_img = -1, fret = -1, fd_loop = -1;
2137
2138         fd_loop = lxc_get_unused_loop_dev(loop_dev);
2139         if (fd_loop < 0) {
2140                 if (fd_loop == -ENODEV)
2141                         fd_loop = lxc_get_unused_loop_dev_legacy(loop_dev);
2142                 else
2143                         goto on_error;
2144         }
2145
2146         fd_img = open(source, O_RDWR | O_CLOEXEC);
2147         if (fd_img < 0)
2148                 goto on_error;
2149
2150         ret = ioctl(fd_loop, LOOP_SET_FD, fd_img);
2151         if (ret < 0)
2152                 goto on_error;
2153
2154         memset(&lo64, 0, sizeof(lo64));
2155         lo64.lo_flags = flags;
2156
2157         ret = ioctl(fd_loop, LOOP_SET_STATUS64, &lo64);
2158         if (ret < 0)
2159                 goto on_error;
2160
2161         fret = 0;
2162
2163 on_error:
2164         if (fd_img >= 0)
2165                 close(fd_img);
2166
2167         if (fret < 0 && fd_loop >= 0) {
2168                 close(fd_loop);
2169                 fd_loop = -1;
2170         }
2171
2172         return fd_loop;
2173 }
2174
2175 int lxc_unstack_mountpoint(const char *path, bool lazy)
2176 {
2177         int ret;
2178         int umounts = 0;
2179
2180 pop_stack:
2181         ret = umount2(path, lazy ? MNT_DETACH : 0);
2182         if (ret < 0) {
2183                 /* We consider anything else than EINVAL deadly to prevent going
2184                  * into an infinite loop. (The other alternative is constantly
2185                  * parsing /proc/self/mountinfo which is yucky and probably
2186                  * racy.)
2187                  */
2188                 if (errno != EINVAL)
2189                         return -errno;
2190         } else {
2191                 /* Just stop counting when this happens. That'd just be so
2192                  * stupid that we won't even bother trying to report back the
2193                  * correct value anymore.
2194                  */
2195                 if (umounts != INT_MAX)
2196                         umounts++;
2197                 /* We succeeded in umounting. Make sure that there's no other
2198                  * mountpoint stacked underneath.
2199                  */
2200                 goto pop_stack;
2201         }
2202
2203         return umounts;
2204 }
2205
2206 int run_command(char *buf, size_t buf_size, int (*child_fn)(void *), void *args)
2207 {
2208         pid_t child;
2209         int ret, fret, pipefd[2];
2210         ssize_t bytes;
2211
2212         /* Make sure our callers do not receive unitialized memory. */
2213         if (buf_size > 0 && buf)
2214                 buf[0] = '\0';
2215
2216         if (pipe(pipefd) < 0) {
2217                 SYSERROR("failed to create pipe");
2218                 return -1;
2219         }
2220
2221         child = lxc_raw_clone(0);
2222         if (child < 0) {
2223                 close(pipefd[0]);
2224                 close(pipefd[1]);
2225                 SYSERROR("failed to create new process");
2226                 return -1;
2227         }
2228
2229         if (child == 0) {
2230                 /* Close the read-end of the pipe. */
2231                 close(pipefd[0]);
2232
2233                 /* Redirect std{err,out} to write-end of the
2234                  * pipe.
2235                  */
2236                 ret = dup2(pipefd[1], STDOUT_FILENO);
2237                 if (ret >= 0)
2238                         ret = dup2(pipefd[1], STDERR_FILENO);
2239
2240                 /* Close the write-end of the pipe. */
2241                 close(pipefd[1]);
2242
2243                 if (ret < 0) {
2244                         SYSERROR("failed to duplicate std{err,out} file descriptor");
2245                         exit(EXIT_FAILURE);
2246                 }
2247
2248                 /* Does not return. */
2249                 child_fn(args);
2250                 ERROR("failed to exec command");
2251                 exit(EXIT_FAILURE);
2252         }
2253
2254         /* close the write-end of the pipe */
2255         close(pipefd[1]);
2256
2257         if (buf && buf_size > 0) {
2258                 bytes = read(pipefd[0], buf, buf_size - 1);
2259                 if (bytes > 0)
2260                         buf[bytes - 1] = '\0';
2261         }
2262
2263         fret = wait_for_pid(child);
2264         /* close the read-end of the pipe */
2265         close(pipefd[0]);
2266
2267         return fret;
2268 }
2269
2270 char *must_make_path(const char *first, ...)
2271 {
2272         va_list args;
2273         char *cur, *dest;
2274         size_t full_len = strlen(first);
2275
2276         dest = must_copy_string(first);
2277
2278         va_start(args, first);
2279         while ((cur = va_arg(args, char *)) != NULL) {
2280                 full_len += strlen(cur);
2281                 if (cur[0] != '/')
2282                         full_len++;
2283                 dest = must_realloc(dest, full_len + 1);
2284                 if (cur[0] != '/')
2285                         strcat(dest, "/");
2286                 strcat(dest, cur);
2287         }
2288         va_end(args);
2289
2290         return dest;
2291 }
2292
2293 char *must_append_path(char *first, ...)
2294 {
2295         char *cur;
2296         size_t full_len;
2297         va_list args;
2298         char *dest = first;
2299
2300         full_len = strlen(first);
2301         va_start(args, first);
2302         while ((cur = va_arg(args, char *)) != NULL) {
2303                 full_len += strlen(cur);
2304
2305                 if (cur[0] != '/')
2306                         full_len++;
2307
2308                 dest = must_realloc(dest, full_len + 1);
2309
2310                 if (cur[0] != '/')
2311                         strcat(dest, "/");
2312
2313                 strcat(dest, cur);
2314         }
2315         va_end(args);
2316
2317         return dest;
2318 }
2319
2320 char *must_copy_string(const char *entry)
2321 {
2322         char *ret;
2323
2324         if (!entry)
2325                 return NULL;
2326         do {
2327                 ret = strdup(entry);
2328         } while (!ret);
2329
2330         return ret;
2331 }
2332
2333 void *must_realloc(void *orig, size_t sz)
2334 {
2335         void *ret;
2336
2337         do {
2338                 ret = realloc(orig, sz);
2339         } while (!ret);
2340
2341         return ret;
2342 }
2343
2344 bool is_fs_type(const struct statfs *fs, fs_type_magic magic_val)
2345 {
2346         return (fs->f_type == (fs_type_magic)magic_val);
2347 }
2348
2349 bool has_fs_type(const char *path, fs_type_magic magic_val)
2350 {
2351         bool has_type;
2352         int ret;
2353         struct statfs sb;
2354
2355         ret = statfs(path, &sb);
2356         if (ret < 0)
2357                 return false;
2358
2359         has_type = is_fs_type(&sb, magic_val);
2360         if (!has_type && magic_val == RAMFS_MAGIC)
2361                 WARN("When the ramfs it a tmpfs statfs() might report tmpfs");
2362
2363         return has_type;
2364 }
2365
2366 bool lxc_nic_exists(char *nic)
2367 {
2368 #define __LXC_SYS_CLASS_NET_LEN 15 + IFNAMSIZ + 1
2369         char path[__LXC_SYS_CLASS_NET_LEN];
2370         int ret;
2371         struct stat sb;
2372
2373         if (!strcmp(nic, "none"))
2374                 return true;
2375
2376         ret = snprintf(path, __LXC_SYS_CLASS_NET_LEN, "/sys/class/net/%s", nic);
2377         if (ret < 0 || (size_t)ret >= __LXC_SYS_CLASS_NET_LEN)
2378                 return false;
2379
2380         ret = stat(path, &sb);
2381         if (ret < 0)
2382                 return false;
2383
2384         return true;
2385 }
2386
2387 int lxc_make_tmpfile(char *template, bool rm)
2388 {
2389         int fd, ret;
2390
2391         fd = mkstemp(template);
2392         if (fd < 0)
2393                 return -1;
2394
2395         if (!rm)
2396                 return fd;
2397
2398         ret = unlink(template);
2399         if (ret < 0) {
2400                 close(fd);
2401                 return -1;
2402         }
2403
2404         return fd;
2405 }
2406
2407 int parse_byte_size_string(const char *s, int64_t *converted)
2408 {
2409         int ret, suffix_len;
2410         long long int conv;
2411         int64_t mltpl, overflow;
2412         char *end;
2413         char dup[LXC_NUMSTRLEN64 + 2];
2414         char suffix[3] = {0};
2415
2416         if (!s || !strcmp(s, ""))
2417                 return -EINVAL;
2418
2419         end = stpncpy(dup, s, sizeof(dup));
2420         if (*end != '\0')
2421                 return -EINVAL;
2422
2423         if (isdigit(*(end - 1)))
2424                 suffix_len = 0;
2425         else if (isalpha(*(end - 1)))
2426                 suffix_len = 1;
2427         else
2428                 return -EINVAL;
2429
2430         if (suffix_len > 0 && (end - 2) == dup && !isdigit(*(end - 2)))
2431                 return -EINVAL;
2432
2433         if (suffix_len > 0 && isalpha(*(end - 2)))
2434                 suffix_len++;
2435
2436         if (suffix_len > 0) {
2437                 memcpy(suffix, end - suffix_len, suffix_len);
2438                 *(suffix + suffix_len) = '\0';
2439                 *(end - suffix_len) = '\0';
2440         }
2441         dup[lxc_char_right_gc(dup, strlen(dup))] = '\0';
2442
2443         ret = lxc_safe_long_long(dup, &conv);
2444         if (ret < 0)
2445                 return -ret;
2446
2447         if (suffix_len != 2) {
2448                 *converted = conv;
2449                 return 0;
2450         }
2451
2452         if (!strcmp(suffix, "kB"))
2453                 mltpl = 1024;
2454         else if (!strcmp(suffix, "MB"))
2455                 mltpl = 1024 * 1024;
2456         else if (!strcmp(suffix, "GB"))
2457                 mltpl = 1024 * 1024 * 1024;
2458         else
2459                 return -EINVAL;
2460
2461         overflow = conv * mltpl;
2462         if (conv != 0 && (overflow / conv) != mltpl)
2463                 return -ERANGE;
2464
2465         *converted = overflow;
2466         return 0;
2467 }
2468
2469 uint64_t lxc_find_next_power2(uint64_t n)
2470 {
2471         /* 0 is not valid input. We return 0 to the caller since 0 is not a
2472          * valid power of two.
2473          */
2474         if (n == 0)
2475                 return 0;
2476
2477         if (!(n & (n - 1)))
2478                 return n;
2479
2480         while (n & (n - 1))
2481                 n = n & (n - 1);
2482
2483         n = n << 1;
2484         return n;
2485 }
2486
2487 int lxc_set_death_signal(int signal)
2488 {
2489         int ret;
2490         pid_t ppid;
2491
2492         ret = prctl(PR_SET_PDEATHSIG, signal, 0, 0, 0);
2493
2494         /* Check whether we have been orphaned. */
2495         ppid = (pid_t)syscall(SYS_getppid);
2496         if (ppid == 1) {
2497                 pid_t self;
2498
2499                 self = lxc_raw_getpid();
2500                 ret = kill(self, SIGKILL);
2501                 if (ret < 0)
2502                         return -1;
2503         }
2504
2505         if (ret < 0) {
2506                 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal);
2507                 return -1;
2508         }
2509
2510         return 0;
2511 }