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
 513         fp->child_pid = child_pid;
 514         fp->pipe = pipe_fds[0];
 515
 516         fp->f = fdopen(pipe_fds[0], "r");
 517         if (!fp->f)
 518                 goto on_error;
 519
 520         return fp;
 521
 522 on_error:
 523         if (fp)
 524                 free(fp);
 525
 526         if (pipe_fds[0] >= 0)
 527                 close(pipe_fds[0]);
 528
 529         if (pipe_fds[1] >= 0)
 530                 close(pipe_fds[1]);
 531
 532         return NULL;
 533 }
 534
 535 int lxc_pclose(struct lxc_popen_FILE *fp)
 536 {
 537         pid_t wait_pid;
 538         int wstatus = 0;
 539
 540         if (!fp)
 541                 return -1;
 542
 543         do {
 544                 wait_pid = waitpid(fp->child_pid, &wstatus, 0);
 545         } while (wait_pid < 0 && errno == EINTR);
 546
 547         close(fp->pipe);
 548         fclose(fp->f);
 549         free(fp);
 550
 551         if (wait_pid < 0)
 552                 return -1;
 553
 554         return wstatus;
 555 }
 556
 557 char *lxc_string_replace(const char *needle, const char *replacement, const char *haystack)
 558 {
 559         ssize_t len = -1, saved_len = -1;
 560         char *result = NULL;
 561         size_t replacement_len = strlen(replacement);
 562         size_t needle_len = strlen(needle);
 563
 564         /* should be executed exactly twice */
 565         while (len == -1 || result == NULL) {
 566                 char *p;
 567                 char *last_p;
 568                 ssize_t part_len;
 569
 570                 if (len != -1) {
 571                         result = calloc(1, len + 1);
 572                         if (!result)
 573                                 return NULL;
 574                         saved_len = len;
 575                 }
 576
 577                 len = 0;
 578
 579                 for (last_p = (char *)haystack, p = strstr(last_p, needle); p; last_p = p, p = strstr(last_p, needle)) {
 580                         part_len = (ssize_t)(p - last_p);
 581                         if (result && part_len > 0)
 582                                 memcpy(&result[len], last_p, part_len);
 583                         len += part_len;
 584                         if (result && replacement_len > 0)
 585                                 memcpy(&result[len], replacement, replacement_len);
 586                         len += replacement_len;
 587                         p += needle_len;
 588                 }
 589                 part_len = strlen(last_p);
 590                 if (result && part_len > 0)
 591                         memcpy(&result[len], last_p, part_len);
 592                 len += part_len;
 593         }
 594
 595         /* make sure we did the same thing twice,
 596          * once for calculating length, the other
 597          * time for copying data */
 598         if (saved_len != len) {
 599                 free(result);
 600                 return NULL;
 601         }
 602         /* make sure we didn't overwrite any buffer,
 603          * due to calloc the string should be 0-terminated */
 604         if (result[len] != '\0') {
 605                 free(result);
 606                 return NULL;
 607         }
 608
 609         return result;
 610 }
 611
 612 bool lxc_string_in_array(const char *needle, const char **haystack)
 613 {
 614         for (; haystack && *haystack; haystack++)
 615                 if (!strcmp(needle, *haystack))
 616                         return true;
 617         return false;
 618 }
 619
 620 char *lxc_string_join(const char *sep, const char **parts, bool use_as_prefix)
 621 {
 622         char *result;
 623         char **p;
 624         size_t sep_len = strlen(sep);
 625         size_t result_len = use_as_prefix * sep_len;
 626
 627         /* calculate new string length */
 628         for (p = (char **)parts; *p; p++)
 629                 result_len += (p > (char **)parts) * sep_len + strlen(*p);
 630
 631         result = calloc(result_len + 1, 1);
 632         if (!result)
 633                 return NULL;
 634
 635         if (use_as_prefix)
 636                 strcpy(result, sep);
 637         for (p = (char **)parts; *p; p++) {
 638                 if (p > (char **)parts)
 639                         strcat(result, sep);
 640                 strcat(result, *p);
 641         }
 642
 643         return result;
 644 }
 645
 646 char **lxc_normalize_path(const char *path)
 647 {
 648         char **components;
 649         char **p;
 650         size_t components_len = 0;
 651         size_t pos = 0;
 652
 653         components = lxc_string_split(path, '/');
 654         if (!components)
 655                 return NULL;
 656         for (p = components; *p; p++)
 657                 components_len++;
 658
 659         /* resolve '.' and '..' */
 660         for (pos = 0; pos < components_len; ) {
 661                 if (!strcmp(components[pos], ".") || (!strcmp(components[pos], "..") && pos == 0)) {
 662                         /* eat this element */
 663                         free(components[pos]);
 664                         memmove(&components[pos], &components[pos+1], sizeof(char *) * (components_len - pos));
 665                         components_len--;
 666                 } else if (!strcmp(components[pos], "..")) {
 667                         /* eat this and the previous element */
 668                         free(components[pos - 1]);
 669                         free(components[pos]);
 670                         memmove(&components[pos-1], &components[pos+1], sizeof(char *) * (components_len - pos));
 671                         components_len -= 2;
 672                         pos--;
 673                 } else {
 674                         pos++;
 675                 }
 676         }
 677
 678         return components;
 679 }
 680
 681 char *lxc_deslashify(const char *path)
 682 {
 683         char *dup, *p;
 684         char **parts = NULL;
 685         size_t n, len;
 686
 687         dup = strdup(path);
 688         if (!dup)
 689                 return NULL;
 690
 691         parts = lxc_normalize_path(dup);
 692         if (!parts) {
 693                 free(dup);
 694                 return NULL;
 695         }
 696
 697         /* We'll end up here if path == "///" or path == "". */
 698         if (!*parts) {
 699                 len = strlen(dup);
 700                 if (!len) {
 701                         lxc_free_array((void **)parts, free);
 702                         return dup;
 703                 }
 704                 n = strcspn(dup, "/");
 705                 if (n == len) {
 706                         free(dup);
 707                         lxc_free_array((void **)parts, free);
 708
 709                         p = strdup("/");
 710                         if (!p)
 711                                 return NULL;
 712
 713                         return p;
 714                 }
 715         }
 716
 717         p = lxc_string_join("/", (const char **)parts, *dup == '/');
 718         free(dup);
 719         lxc_free_array((void **)parts, free);
 720         return p;
 721 }
 722
 723 char *lxc_append_paths(const char *first, const char *second)
 724 {
 725         int ret;
 726         size_t len;
 727         char *result = NULL;
 728         const char *pattern = "%s%s";
 729
 730         len = strlen(first) + strlen(second) + 1;
 731         if (second[0] != '/') {
 732                 len += 1;
 733                 pattern = "%s/%s";
 734         }
 735
 736         result = calloc(1, len);
 737         if (!result)
 738                 return NULL;
 739
 740         ret = snprintf(result, len, pattern, first, second);
 741         if (ret < 0 || (size_t)ret >= len) {
 742                 free(result);
 743                 return NULL;
 744         }
 745
 746         return result;
 747 }
 748
 749 bool lxc_string_in_list(const char *needle, const char *haystack, char _sep)
 750 {
 751         char *token, *str, *saveptr = NULL;
 752         char sep[2] = { _sep, '\0' };
 753
 754         if (!haystack || !needle)
 755                 return 0;
 756
 757         str = alloca(strlen(haystack)+1);
 758         strcpy(str, haystack);
 759         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 760                 if (strcmp(needle, token) == 0)
 761                         return 1;
 762         }
 763
 764         return 0;
 765 }
 766
 767 char **lxc_string_split(const char *string, char _sep)
 768 {
 769         char *token, *str, *saveptr = NULL;
 770         char sep[2] = {_sep, '\0'};
 771         char **tmp = NULL, **result = NULL;
 772         size_t result_capacity = 0;
 773         size_t result_count = 0;
 774         int r, saved_errno;
 775
 776         if (!string)
 777                 return calloc(1, sizeof(char *));
 778
 779         str = alloca(strlen(string) + 1);
 780         strcpy(str, string);
 781         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 782                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 783                 if (r < 0)
 784                         goto error_out;
 785                 result[result_count] = strdup(token);
 786                 if (!result[result_count])
 787                         goto error_out;
 788                 result_count++;
 789         }
 790
 791         /* if we allocated too much, reduce it */
 792         tmp = realloc(result, (result_count + 1) * sizeof(char *));
 793         if (!tmp)
 794                 goto error_out;
 795         result = tmp;
 796         /* Make sure we don't return uninitialized memory. */
 797         if (result_count == 0)
 798                 *result = NULL;
 799         return result;
 800 error_out:
 801         saved_errno = errno;
 802         lxc_free_array((void **)result, free);
 803         errno = saved_errno;
 804         return NULL;
 805 }
 806
 807 static bool complete_word(char ***result, char *start, char *end, size_t *cap, size_t *cnt)
 808 {
 809         int r;
 810
 811         r = lxc_grow_array((void ***)result, cap, 2 + *cnt, 16);
 812         if (r < 0)
 813                 return false;
 814         (*result)[*cnt] = strndup(start, end - start);
 815         if (!(*result)[*cnt])
 816                 return false;
 817         (*cnt)++;
 818
 819         return true;
 820 }
 821
 822 /*
 823  * Given a a string 'one two "three four"', split into three words,
 824  * one, two, and "three four"
 825  */
 826 char **lxc_string_split_quoted(char *string)
 827 {
 828         char *nextword = string, *p, state;
 829         char **result = NULL;
 830         size_t result_capacity = 0;
 831         size_t result_count = 0;
 832
 833         if (!string || !*string)
 834                 return calloc(1, sizeof(char *));
 835
 836         // TODO I'm *not* handling escaped quote
 837         state = ' ';
 838         for (p = string; *p; p++) {
 839                 switch(state) {
 840                 case ' ':
 841                         if (isspace(*p))
 842                                 continue;
 843                         else if (*p == '"' || *p == '\'') {
 844                                 nextword = p;
 845                                 state = *p;
 846                                 continue;
 847                         }
 848                         nextword = p;
 849                         state = 'a';
 850                         continue;
 851                 case 'a':
 852                         if (isspace(*p)) {
 853                                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 854                                 state = ' ';
 855                                 continue;
 856                         }
 857                         continue;
 858                 case '"':
 859                 case '\'':
 860                         if (*p == state) {
 861                                 complete_word(&result, nextword+1, p, &result_capacity, &result_count);
 862                                 state = ' ';
 863                                 continue;
 864                         }
 865                         continue;
 866                 }
 867         }
 868
 869         if (state == 'a')
 870                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 871
 872         return realloc(result, (result_count + 1) * sizeof(char *));
 873 }
 874
 875 char **lxc_string_split_and_trim(const char *string, char _sep)
 876 {
 877         char *token, *str, *saveptr = NULL;
 878         char sep[2] = { _sep, '\0' };
 879         char **result = NULL;
 880         size_t result_capacity = 0;
 881         size_t result_count = 0;
 882         int r, saved_errno;
 883         size_t i = 0;
 884
 885         if (!string)
 886                 return calloc(1, sizeof(char *));
 887
 888         str = alloca(strlen(string)+1);
 889         strcpy(str, string);
 890         for (; (token = strtok_r(str, sep, &saveptr)); str = NULL) {
 891                 while (token[0] == ' ' || token[0] == '\t')
 892                         token++;
 893                 i = strlen(token);
 894                 while (i > 0 && (token[i - 1] == ' ' || token[i - 1] == '\t')) {
 895                         token[i - 1] = '\0';
 896                         i--;
 897                 }
 898                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 899                 if (r < 0)
 900                         goto error_out;
 901                 result[result_count] = strdup(token);
 902                 if (!result[result_count])
 903                         goto error_out;
 904                 result_count++;
 905         }
 906
 907         /* if we allocated too much, reduce it */
 908         return realloc(result, (result_count + 1) * sizeof(char *));
 909 error_out:
 910         saved_errno = errno;
 911         lxc_free_array((void **)result, free);
 912         errno = saved_errno;
 913         return NULL;
 914 }
 915
 916 void lxc_free_array(void **array, lxc_free_fn element_free_fn)
 917 {
 918         void **p;
 919         for (p = array; p && *p; p++)
 920                 element_free_fn(*p);
 921         free((void*)array);
 922 }
 923
 924 int lxc_grow_array(void ***array, size_t* capacity, size_t new_size, size_t capacity_increment)
 925 {
 926         size_t new_capacity;
 927         void **new_array;
 928
 929         /* first time around, catch some trivial mistakes of the user
 930          * only initializing one of these */
 931         if (!*array || !*capacity) {
 932                 *array = NULL;
 933                 *capacity = 0;
 934         }
 935
 936         new_capacity = *capacity;
 937         while (new_size + 1 > new_capacity)
 938                 new_capacity += capacity_increment;
 939         if (new_capacity != *capacity) {
 940                 /* we have to reallocate */
 941                 new_array = realloc(*array, new_capacity * sizeof(void *));
 942                 if (!new_array)
 943                         return -1;
 944                 memset(&new_array[*capacity], 0, (new_capacity - (*capacity)) * sizeof(void *));
 945                 *array = new_array;
 946                 *capacity = new_capacity;
 947         }
 948
 949         /* array has sufficient elements */
 950         return 0;
 951 }
 952
 953 size_t lxc_array_len(void **array)
 954 {
 955         void **p;
 956         size_t result = 0;
 957
 958         for (p = array; p && *p; p++)
 959                 result++;
 960
 961         return result;
 962 }
 963
 964 int lxc_write_to_file(const char *filename, const void* buf, size_t count, bool add_newline)
 965 {
 966         int fd, saved_errno;
 967         ssize_t ret;
 968
 969         fd = open(filename, O_WRONLY | O_TRUNC | O_CREAT | O_CLOEXEC, 0666);
 970         if (fd < 0)
 971                 return -1;
 972         ret = lxc_write_nointr(fd, buf, count);
 973         if (ret < 0)
 974                 goto out_error;
 975         if ((size_t)ret != count)
 976                 goto out_error;
 977         if (add_newline) {
 978                 ret = lxc_write_nointr(fd, "\n", 1);
 979                 if (ret != 1)
 980                         goto out_error;
 981         }
 982         close(fd);
 983         return 0;
 984
 985 out_error:
 986         saved_errno = errno;
 987         close(fd);
 988         errno = saved_errno;
 989         return -1;
 990 }
 991
 992 int lxc_read_from_file(const char *filename, void* buf, size_t count)
 993 {
 994         int fd = -1, saved_errno;
 995         ssize_t ret;
 996
 997         fd = open(filename, O_RDONLY | O_CLOEXEC);
 998         if (fd < 0)
 999                 return -1;
1000
1001         if (!buf || !count) {
1002                 char buf2[100];
1003                 size_t count2 = 0;
1004                 while ((ret = read(fd, buf2, 100)) > 0)
1005                         count2 += ret;
1006                 if (ret >= 0)
1007                         ret = count2;
1008         } else {
1009                 memset(buf, 0, count);
1010                 ret = read(fd, buf, count);
1011         }
1012
1013         if (ret < 0)
1014                 ERROR("read %s: %s", filename, strerror(errno));
1015
1016         saved_errno = errno;
1017         close(fd);
1018         errno = saved_errno;
1019         return ret;
1020 }
1021
1022 void **lxc_append_null_to_array(void **array, size_t count)
1023 {
1024         void **temp;
1025
1026         /* Append NULL to the array */
1027         if (count) {
1028                 temp = realloc(array, (count + 1) * sizeof(*array));
1029                 if (!temp) {
1030                         size_t i;
1031                         for (i = 0; i < count; i++)
1032                                 free(array[i]);
1033                         free(array);
1034                         return NULL;
1035                 }
1036                 array = temp;
1037                 array[count] = NULL;
1038         }
1039         return array;
1040 }
1041
1042 int randseed(bool srand_it)
1043 {
1044         /*
1045            srand pre-seed function based on /dev/urandom
1046            */
1047         unsigned int seed = time(NULL) + getpid();
1048
1049         FILE *f;
1050         f = fopen("/dev/urandom", "r");
1051         if (f) {
1052                 int ret = fread(&seed, sizeof(seed), 1, f);
1053                 if (ret != 1)
1054                         DEBUG("unable to fread /dev/urandom, %s, fallback to time+pid rand seed", strerror(errno));
1055                 fclose(f);
1056         }
1057
1058         if (srand_it)
1059                 srand(seed);
1060
1061         return seed;
1062 }
1063
1064 uid_t get_ns_uid(uid_t orig)
1065 {
1066         char *line = NULL;
1067         size_t sz = 0;
1068         uid_t nsid, hostid, range;
1069         FILE *f = fopen("/proc/self/uid_map", "r");
1070         if (!f)
1071                 return 0;
1072
1073         while (getline(&line, &sz, f) != -1) {
1074                 if (sscanf(line, "%u %u %u", &nsid, &hostid, &range) != 3)
1075                         continue;
1076                 if (hostid <= orig && hostid + range > orig) {
1077                         nsid += orig - hostid;
1078                         goto found;
1079                 }
1080         }
1081
1082         nsid = 0;
1083 found:
1084         fclose(f);
1085         free(line);
1086         return nsid;
1087 }
1088
1089 bool dir_exists(const char *path)
1090 {
1091         struct stat sb;
1092         int ret;
1093
1094         ret = stat(path, &sb);
1095         if (ret < 0)
1096                 /* Could be something other than eexist, just say "no". */
1097                 return false;
1098         return S_ISDIR(sb.st_mode);
1099 }
1100
1101 /* Note we don't use SHA-1 here as we don't want to depend on HAVE_GNUTLS.
1102  * FNV has good anti collision properties and we're not worried
1103  * about pre-image resistance or one-way-ness, we're just trying to make
1104  * the name unique in the 108 bytes of space we have.
1105  */
1106 uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
1107 {
1108         unsigned char *bp;
1109
1110         for(bp = buf; bp < (unsigned char *)buf + len; bp++)
1111         {
1112                 /* xor the bottom with the current octet */
1113                 hval ^= (uint64_t)*bp;
1114
1115                 /* gcc optimised:
1116                  * multiply by the 64 bit FNV magic prime mod 2^64
1117                  */
1118                 hval += (hval << 1) + (hval << 4) + (hval << 5) +
1119                         (hval << 7) + (hval << 8) + (hval << 40);
1120         }
1121
1122         return hval;
1123 }
1124
1125 /*
1126  * Detect whether / is mounted MS_SHARED.  The only way I know of to
1127  * check that is through /proc/self/mountinfo.
1128  * I'm only checking for /.  If the container rootfs or mount location
1129  * is MS_SHARED, but not '/', then you're out of luck - figuring that
1130  * out would be too much work to be worth it.
1131  */
1132 int detect_shared_rootfs(void)
1133 {
1134         char buf[LXC_LINELEN], *p;
1135         FILE *f;
1136         int i;
1137         char *p2;
1138
1139         f = fopen("/proc/self/mountinfo", "r");
1140         if (!f)
1141                 return 0;
1142         while (fgets(buf, LXC_LINELEN, f)) {
1143                 for (p = buf, i = 0; p && i < 4; i++)
1144                         p = strchr(p + 1, ' ');
1145                 if (!p)
1146                         continue;
1147                 p2 = strchr(p + 1, ' ');
1148                 if (!p2)
1149                         continue;
1150                 *p2 = '\0';
1151                 if (strcmp(p + 1, "/") == 0) {
1152                         /* This is '/'. Is it shared? */
1153                         p = strchr(p2 + 1, ' ');
1154                         if (p && strstr(p, "shared:")) {
1155                                 fclose(f);
1156                                 return 1;
1157                         }
1158                 }
1159         }
1160         fclose(f);
1161         return 0;
1162 }
1163
1164 bool switch_to_ns(pid_t pid, const char *ns) {
1165         int fd, ret;
1166         char nspath[MAXPATHLEN];
1167
1168         /* Switch to new ns */
1169         ret = snprintf(nspath, MAXPATHLEN, "/proc/%d/ns/%s", pid, ns);
1170         if (ret < 0 || ret >= MAXPATHLEN)
1171                 return false;
1172
1173         fd = open(nspath, O_RDONLY);
1174         if (fd < 0) {
1175                 SYSERROR("failed to open %s", nspath);
1176                 return false;
1177         }
1178
1179         ret = setns(fd, 0);
1180         if (ret) {
1181                 SYSERROR("failed to set process %d to %s of %d.", pid, ns, fd);
1182                 close(fd);
1183                 return false;
1184         }
1185         close(fd);
1186         return true;
1187 }
1188
1189 /*
1190  * looking at fs/proc_namespace.c, it appears we can
1191  * actually expect the rootfs entry to very specifically contain
1192  * " - rootfs rootfs "
1193  * IIUC, so long as we've chrooted so that rootfs is not our root,
1194  * the rootfs entry should always be skipped in mountinfo contents.
1195  */
1196 bool detect_ramfs_rootfs(void)
1197 {
1198         FILE *f;
1199         char *p, *p2;
1200         char *line = NULL;
1201         size_t len = 0;
1202         int i;
1203
1204         f = fopen("/proc/self/mountinfo", "r");
1205         if (!f)
1206                 return false;
1207
1208         while (getline(&line, &len, f) != -1) {
1209                 for (p = line, i = 0; p && i < 4; i++)
1210                         p = strchr(p + 1, ' ');
1211                 if (!p)
1212                         continue;
1213                 p2 = strchr(p + 1, ' ');
1214                 if (!p2)
1215                         continue;
1216                 *p2 = '\0';
1217                 if (strcmp(p + 1, "/") == 0) {
1218                         /* This is '/'. Is it the ramfs? */
1219                         p = strchr(p2 + 1, '-');
1220                         if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) {
1221                                 free(line);
1222                                 fclose(f);
1223                                 return true;
1224                         }
1225                 }
1226         }
1227         free(line);
1228         fclose(f);
1229         return false;
1230 }
1231
1232 char *on_path(const char *cmd, const char *rootfs) {
1233         char *path = NULL;
1234         char *entry = NULL;
1235         char *saveptr = NULL;
1236         char cmdpath[MAXPATHLEN];
1237         int ret;
1238
1239         path = getenv("PATH");
1240         if (!path)
1241                 return NULL;
1242
1243         path = strdup(path);
1244         if (!path)
1245                 return NULL;
1246
1247         entry = strtok_r(path, ":", &saveptr);
1248         while (entry) {
1249                 if (rootfs)
1250                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s/%s", rootfs, entry, cmd);
1251                 else
1252                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s", entry, cmd);
1253
1254                 if (ret < 0 || ret >= MAXPATHLEN)
1255                         goto next_loop;
1256
1257                 if (access(cmdpath, X_OK) == 0) {
1258                         free(path);
1259                         return strdup(cmdpath);
1260                 }
1261
1262 next_loop:
1263                 entry = strtok_r(NULL, ":", &saveptr);
1264         }
1265
1266         free(path);
1267         return NULL;
1268 }
1269
1270 bool file_exists(const char *f)
1271 {
1272         struct stat statbuf;
1273
1274         return stat(f, &statbuf) == 0;
1275 }
1276
1277 bool cgns_supported(void)
1278 {
1279         return file_exists("/proc/self/ns/cgroup");
1280 }
1281
1282 /* historically lxc-init has been under /usr/lib/lxc and under
1283  * /usr/lib/$ARCH/lxc.  It now lives as $prefix/sbin/init.lxc.
1284  */
1285 char *choose_init(const char *rootfs)
1286 {
1287         char *retv = NULL;
1288         const char *empty = "",
1289                    *tmp;
1290         int ret, env_set = 0;
1291
1292         if (!getenv("PATH")) {
1293                 if (setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 0))
1294                         SYSERROR("Failed to setenv");
1295                 env_set = 1;
1296         }
1297
1298         retv = on_path("init.lxc", rootfs);
1299
1300         if (env_set) {
1301                 if (unsetenv("PATH"))
1302                         SYSERROR("Failed to unsetenv");
1303         }
1304
1305         if (retv)
1306                 return retv;
1307
1308         retv = malloc(PATH_MAX);
1309         if (!retv)
1310                 return NULL;
1311
1312         if (rootfs)
1313                 tmp = rootfs;
1314         else
1315                 tmp = empty;
1316
1317         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, SBINDIR, "/init.lxc");
1318         if (ret < 0 || ret >= PATH_MAX) {
1319                 ERROR("pathname too long");
1320                 goto out1;
1321         }
1322         if (access(retv, X_OK) == 0)
1323                 return retv;
1324
1325         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, LXCINITDIR, "/lxc/lxc-init");
1326         if (ret < 0 || ret >= PATH_MAX) {
1327                 ERROR("pathname too long");
1328                 goto out1;
1329         }
1330         if (access(retv, X_OK) == 0)
1331                 return retv;
1332
1333         ret = snprintf(retv, PATH_MAX, "%s/usr/lib/lxc/lxc-init", tmp);
1334         if (ret < 0 || ret >= PATH_MAX) {
1335                 ERROR("pathname too long");
1336                 goto out1;
1337         }
1338         if (access(retv, X_OK) == 0)
1339                 return retv;
1340
1341         ret = snprintf(retv, PATH_MAX, "%s/sbin/lxc-init", tmp);
1342         if (ret < 0 || ret >= PATH_MAX) {
1343                 ERROR("pathname too long");
1344                 goto out1;
1345         }
1346         if (access(retv, X_OK) == 0)
1347                 return retv;
1348
1349         /*
1350          * Last resort, look for the statically compiled init.lxc which we
1351          * hopefully bind-mounted in.
1352          * If we are called during container setup, and we get to this point,
1353          * then the init.lxc.static from the host will need to be bind-mounted
1354          * in.  So we return NULL here to indicate that.
1355          */
1356         if (rootfs)
1357                 goto out1;
1358
1359         ret = snprintf(retv, PATH_MAX, "/init.lxc.static");
1360         if (ret < 0 || ret >= PATH_MAX) {
1361                 WARN("Nonsense - name /lxc.init.static too long");
1362                 goto out1;
1363         }
1364         if (access(retv, X_OK) == 0)
1365                 return retv;
1366
1367 out1:
1368         free(retv);
1369         return NULL;
1370 }
1371
1372 int print_to_file(const char *file, const char *content)
1373 {
1374         FILE *f;
1375         int ret = 0;
1376
1377         f = fopen(file, "w");
1378         if (!f)
1379                 return -1;
1380         if (fprintf(f, "%s", content) != strlen(content))
1381                 ret = -1;
1382         fclose(f);
1383         return ret;
1384 }
1385
1386 int is_dir(const char *path)
1387 {
1388         struct stat statbuf;
1389         int ret = stat(path, &statbuf);
1390         if (ret == 0 && S_ISDIR(statbuf.st_mode))
1391                 return 1;
1392         return 0;
1393 }
1394
1395 /*
1396  * Given the '-t' template option to lxc-create, figure out what to
1397  * do.  If the template is a full executable path, use that.  If it
1398  * is something like 'sshd', then return $templatepath/lxc-sshd.
1399  * On success return the template, on error return NULL.
1400  */
1401 char *get_template_path(const char *t)
1402 {
1403         int ret, len;
1404         char *tpath;
1405
1406         if (t[0] == '/' && access(t, X_OK) == 0) {
1407                 tpath = strdup(t);
1408                 return tpath;
1409         }
1410
1411         len = strlen(LXCTEMPLATEDIR) + strlen(t) + strlen("/lxc-") + 1;
1412         tpath = malloc(len);
1413         if (!tpath)
1414                 return NULL;
1415         ret = snprintf(tpath, len, "%s/lxc-%s", LXCTEMPLATEDIR, t);
1416         if (ret < 0 || ret >= len) {
1417                 free(tpath);
1418                 return NULL;
1419         }
1420         if (access(tpath, X_OK) < 0) {
1421                 SYSERROR("bad template: %s", t);
1422                 free(tpath);
1423                 return NULL;
1424         }
1425
1426         return tpath;
1427 }
1428
1429 /*
1430  * @path:    a pathname where / replaced with '\0'.
1431  * @offsetp: pointer to int showing which path segment was last seen.
1432  *           Updated on return to reflect the next segment.
1433  * @fulllen: full original path length.
1434  * Returns a pointer to the next path segment, or NULL if done.
1435  */
1436 static char *get_nextpath(char *path, int *offsetp, int fulllen)
1437 {
1438         int offset = *offsetp;
1439
1440         if (offset >= fulllen)
1441                 return NULL;
1442
1443         while (path[offset] != '\0' && offset < fulllen)
1444                 offset++;
1445         while (path[offset] == '\0' && offset < fulllen)
1446                 offset++;
1447
1448         *offsetp = offset;
1449         return (offset < fulllen) ? &path[offset] : NULL;
1450 }
1451
1452 /*
1453  * Check that @subdir is a subdir of @dir.  @len is the length of
1454  * @dir (to avoid having to recalculate it).
1455  */
1456 static bool is_subdir(const char *subdir, const char *dir, size_t len)
1457 {
1458         size_t subdirlen = strlen(subdir);
1459
1460         if (subdirlen < len)
1461                 return false;
1462         if (strncmp(subdir, dir, len) != 0)
1463                 return false;
1464         if (dir[len-1] == '/')
1465                 return true;
1466         if (subdir[len] == '/' || subdirlen == len)
1467                 return true;
1468         return false;
1469 }
1470
1471 /*
1472  * Check if the open fd is a symlink.  Return -ELOOP if it is.  Return
1473  * -ENOENT if we couldn't fstat.  Return 0 if the fd is ok.
1474  */
1475 static int check_symlink(int fd)
1476 {
1477         struct stat sb;
1478         int ret = fstat(fd, &sb);
1479         if (ret < 0)
1480                 return -ENOENT;
1481         if (S_ISLNK(sb.st_mode))
1482                 return -ELOOP;
1483         return 0;
1484 }
1485
1486 /*
1487  * Open a file or directory, provided that it contains no symlinks.
1488  *
1489  * CAVEAT: This function must not be used for other purposes than container
1490  * setup before executing the container's init
1491  */
1492 static int open_if_safe(int dirfd, const char *nextpath)
1493 {
1494         int newfd = openat(dirfd, nextpath, O_RDONLY | O_NOFOLLOW);
1495         if (newfd >= 0) /* Was not a symlink, all good. */
1496                 return newfd;
1497
1498         if (errno == ELOOP)
1499                 return newfd;
1500
1501         if (errno == EPERM || errno == EACCES) {
1502                 /* We're not root (cause we got EPERM) so try opening with
1503                  * O_PATH.
1504                  */
1505                 newfd = openat(dirfd, nextpath, O_PATH | O_NOFOLLOW);
1506                 if (newfd >= 0) {
1507                         /* O_PATH will return an fd for symlinks. We know
1508                          * nextpath wasn't a symlink at last openat, so if fd is
1509                          * now a link, then something * fishy is going on.
1510                          */
1511                         int ret = check_symlink(newfd);
1512                         if (ret < 0) {
1513                                 close(newfd);
1514                                 newfd = ret;
1515                         }
1516                 }
1517         }
1518
1519         return newfd;
1520 }
1521
1522 /*
1523  * Open a path intending for mounting, ensuring that the final path
1524  * is inside the container's rootfs.
1525  *
1526  * CAVEAT: This function must not be used for other purposes than container
1527  * setup before executing the container's init
1528  *
1529  * @target: path to be opened
1530  * @prefix_skip: a part of @target in which to ignore symbolic links.  This
1531  * would be the container's rootfs.
1532  *
1533  * Return an open fd for the path, or <0 on error.
1534  */
1535 static int open_without_symlink(const char *target, const char *prefix_skip)
1536 {
1537         int curlen = 0, dirfd, fulllen, i;
1538         char *dup = NULL;
1539
1540         fulllen = strlen(target);
1541
1542         /* make sure prefix-skip makes sense */
1543         if (prefix_skip && strlen(prefix_skip) > 0) {
1544                 curlen = strlen(prefix_skip);
1545                 if (!is_subdir(target, prefix_skip, curlen)) {
1546                         ERROR("WHOA there - target '%s' didn't start with prefix '%s'",
1547                                 target, prefix_skip);
1548                         return -EINVAL;
1549                 }
1550                 /*
1551                  * get_nextpath() expects the curlen argument to be
1552                  * on a  (turned into \0) / or before it, so decrement
1553                  * curlen to make sure that happens
1554                  */
1555                 if (curlen)
1556                         curlen--;
1557         } else {
1558                 prefix_skip = "/";
1559                 curlen = 0;
1560         }
1561
1562         /* Make a copy of target which we can hack up, and tokenize it */
1563         if ((dup = strdup(target)) == NULL) {
1564                 SYSERROR("Out of memory checking for symbolic link");
1565                 return -ENOMEM;
1566         }
1567         for (i = 0; i < fulllen; i++) {
1568                 if (dup[i] == '/')
1569                         dup[i] = '\0';
1570         }
1571
1572         dirfd = open(prefix_skip, O_RDONLY);
1573         if (dirfd < 0)
1574                 goto out;
1575         while (1) {
1576                 int newfd, saved_errno;
1577                 char *nextpath;
1578
1579                 if ((nextpath = get_nextpath(dup, &curlen, fulllen)) == NULL)
1580                         goto out;
1581                 newfd = open_if_safe(dirfd, nextpath);
1582                 saved_errno = errno;
1583                 close(dirfd);
1584                 dirfd = newfd;
1585                 if (newfd < 0) {
1586                         errno = saved_errno;
1587                         if (errno == ELOOP)
1588                                 SYSERROR("%s in %s was a symbolic link!", nextpath, target);
1589                         goto out;
1590                 }
1591         }
1592
1593 out:
1594         free(dup);
1595         return dirfd;
1596 }
1597
1598 /*
1599  * Safely mount a path into a container, ensuring that the mount target
1600  * is under the container's @rootfs.  (If @rootfs is NULL, then the container
1601  * uses the host's /)
1602  *
1603  * CAVEAT: This function must not be used for other purposes than container
1604  * setup before executing the container's init
1605  */
1606 int safe_mount(const char *src, const char *dest, const char *fstype,
1607                 unsigned long flags, const void *data, const char *rootfs)
1608 {
1609         int destfd, ret, saved_errno;
1610         /* Only needs enough for /proc/self/fd/<fd>. */
1611         char srcbuf[50], destbuf[50];
1612         int srcfd = -1;
1613         const char *mntsrc = src;
1614
1615         if (!rootfs)
1616                 rootfs = "";
1617
1618         /* todo - allow symlinks for relative paths if 'allowsymlinks' option is passed */
1619         if (flags & MS_BIND && src && src[0] != '/') {
1620                 INFO("this is a relative bind mount");
1621                 srcfd = open_without_symlink(src, NULL);
1622                 if (srcfd < 0)
1623                         return srcfd;
1624                 ret = snprintf(srcbuf, 50, "/proc/self/fd/%d", srcfd);
1625                 if (ret < 0 || ret > 50) {
1626                         close(srcfd);
1627                         ERROR("Out of memory");
1628                         return -EINVAL;
1629                 }
1630                 mntsrc = srcbuf;
1631         }
1632
1633         destfd = open_without_symlink(dest, rootfs);
1634         if (destfd < 0) {
1635                 if (srcfd != -1) {
1636                         saved_errno = errno;
1637                         close(srcfd);
1638                         errno = saved_errno;
1639                 }
1640                 return destfd;
1641         }
1642
1643         ret = snprintf(destbuf, 50, "/proc/self/fd/%d", destfd);
1644         if (ret < 0 || ret > 50) {
1645                 if (srcfd != -1)
1646                         close(srcfd);
1647                 close(destfd);
1648                 ERROR("Out of memory");
1649                 return -EINVAL;
1650         }
1651
1652         ret = mount(mntsrc, destbuf, fstype, flags, data);
1653         saved_errno = errno;
1654         if (srcfd != -1)
1655                 close(srcfd);
1656         close(destfd);
1657         if (ret < 0) {
1658                 errno = saved_errno;
1659                 SYSERROR("Failed to mount %s onto %s", src ? src : "(null)", dest);
1660                 return ret;
1661         }
1662
1663         return 0;
1664 }
1665
1666 /*
1667  * Mount a proc under @rootfs if proc self points to a pid other than
1668  * my own.  This is needed to have a known-good proc mount for setting
1669  * up LSMs both at container startup and attach.
1670  *
1671  * @rootfs : the rootfs where proc should be mounted
1672  *
1673  * Returns < 0 on failure, 0 if the correct proc was already mounted
1674  * and 1 if a new proc was mounted.
1675  *
1676  * NOTE: not to be called from inside the container namespace!
1677  */
1678 int lxc_mount_proc_if_needed(const char *rootfs)
1679 {
1680         char path[MAXPATHLEN];
1681         int link_to_pid, linklen, mypid, ret;
1682         char link[LXC_NUMSTRLEN64] = {0};
1683
1684         ret = snprintf(path, MAXPATHLEN, "%s/proc/self", rootfs);
1685         if (ret < 0 || ret >= MAXPATHLEN) {
1686                 SYSERROR("proc path name too long");
1687                 return -1;
1688         }
1689
1690         linklen = readlink(path, link, LXC_NUMSTRLEN64);
1691
1692         ret = snprintf(path, MAXPATHLEN, "%s/proc", rootfs);
1693         if (ret < 0 || ret >= MAXPATHLEN) {
1694                 SYSERROR("proc path name too long");
1695                 return -1;
1696         }
1697
1698         /* /proc not mounted */
1699         if (linklen < 0) {
1700                 if (mkdir(path, 0755) && errno != EEXIST)
1701                         return -1;
1702                 goto domount;
1703         } else if (linklen >= LXC_NUMSTRLEN64) {
1704                 link[linklen - 1] = '\0';
1705                 ERROR("readlink returned truncated content: \"%s\"", link);
1706                 return -1;
1707         }
1708
1709         mypid = lxc_raw_getpid();
1710         INFO("I am %d, /proc/self points to \"%s\"", mypid, link);
1711
1712         if (lxc_safe_int(link, &link_to_pid) < 0)
1713                 return -1;
1714
1715         /* correct procfs is already mounted */
1716         if (link_to_pid == mypid)
1717                 return 0;
1718
1719         ret = umount2(path, MNT_DETACH);
1720         if (ret < 0)
1721                 WARN("failed to umount \"%s\" with MNT_DETACH", path);
1722
1723 domount:
1724         /* rootfs is NULL */
1725         if (!strcmp(rootfs, ""))
1726                 ret = mount("proc", path, "proc", 0, NULL);
1727         else
1728                 ret = safe_mount("proc", path, "proc", 0, NULL, rootfs);
1729         if (ret < 0)
1730                 return -1;
1731
1732         INFO("mounted /proc in container for security transition");
1733         return 1;
1734 }
1735
1736 int open_devnull(void)
1737 {
1738         int fd = open("/dev/null", O_RDWR);
1739
1740         if (fd < 0)
1741                 SYSERROR("Can't open /dev/null");
1742
1743         return fd;
1744 }
1745
1746 int set_stdfds(int fd)
1747 {
1748         int ret;
1749
1750         if (fd < 0)
1751                 return -1;
1752
1753         ret = dup2(fd, STDIN_FILENO);
1754         if (ret < 0)
1755                 return -1;
1756
1757         ret = dup2(fd, STDOUT_FILENO);
1758         if (ret < 0)
1759                 return -1;
1760
1761         ret = dup2(fd, STDERR_FILENO);
1762         if (ret < 0)
1763                 return -1;
1764
1765         return 0;
1766 }
1767
1768 int null_stdfds(void)
1769 {
1770         int ret = -1;
1771         int fd = open_devnull();
1772
1773         if (fd >= 0) {
1774                 ret = set_stdfds(fd);
1775                 close(fd);
1776         }
1777
1778         return ret;
1779 }
1780
1781 /*
1782  * Return the number of lines in file @fn, or -1 on error
1783  */
1784 int lxc_count_file_lines(const char *fn)
1785 {
1786         FILE *f;
1787         char *line = NULL;
1788         size_t sz = 0;
1789         int n = 0;
1790
1791         f = fopen_cloexec(fn, "r");
1792         if (!f)
1793                 return -1;
1794
1795         while (getline(&line, &sz, f) != -1) {
1796                 n++;
1797         }
1798         free(line);
1799         fclose(f);
1800         return n;
1801 }
1802
1803 void *lxc_strmmap(void *addr, size_t length, int prot, int flags, int fd,
1804                   off_t offset)
1805 {
1806         void *tmp = NULL, *overlap = NULL;
1807
1808         /* We establish an anonymous mapping that is one byte larger than the
1809          * underlying file. The pages handed to us are zero filled. */
1810         tmp = mmap(addr, length + 1, PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1811         if (tmp == MAP_FAILED)
1812                 return tmp;
1813
1814         /* Now we establish a fixed-address mapping starting at the address we
1815          * received from our anonymous mapping and replace all bytes excluding
1816          * the additional \0-byte with the file. This allows us to use normal
1817          * string-handling functions. */
1818         overlap = mmap(tmp, length, prot, MAP_FIXED | flags, fd, offset);
1819         if (overlap == MAP_FAILED)
1820                 munmap(tmp, length + 1);
1821
1822         return overlap;
1823 }
1824
1825 int lxc_strmunmap(void *addr, size_t length)
1826 {
1827         return munmap(addr, length + 1);
1828 }
1829
1830 /* Check whether a signal is blocked by a process. */
1831 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
1832 #define __PROC_STATUS_LEN (5 + (LXC_NUMSTRLEN64) + 7 + 1)
1833 bool task_blocking_signal(pid_t pid, int signal)
1834 {
1835         bool bret = false;
1836         char *line = NULL;
1837         long unsigned int sigblk = 0;
1838         size_t n = 0;
1839         int ret;
1840         FILE *f;
1841
1842         char status[__PROC_STATUS_LEN];
1843
1844         ret = snprintf(status, __PROC_STATUS_LEN, "/proc/%d/status", pid);
1845         if (ret < 0 || ret >= __PROC_STATUS_LEN)
1846                 return bret;
1847
1848         f = fopen(status, "r");
1849         if (!f)
1850                 return bret;
1851
1852         while (getline(&line, &n, f) != -1) {
1853                 if (strncmp(line, "SigBlk:\t", 8))
1854                         continue;
1855
1856                 if (sscanf(line + 8, "%lx", &sigblk) != 1)
1857                         goto out;
1858         }
1859
1860         if (sigblk & (1LU << (signal - 1)))
1861                 bret = true;
1862
1863 out:
1864         free(line);
1865         fclose(f);
1866         return bret;
1867 }
1868
1869 static int lxc_append_null_to_list(void ***list)
1870 {
1871         int newentry = 0;
1872         void **tmp;
1873
1874         if (*list)
1875                 for (; (*list)[newentry]; newentry++) {
1876                         ;
1877                 }
1878
1879         tmp = realloc(*list, (newentry + 2) * sizeof(void **));
1880         if (!tmp)
1881                 return -1;
1882
1883         *list = tmp;
1884         (*list)[newentry + 1] = NULL;
1885
1886         return newentry;
1887 }
1888
1889 int lxc_append_string(char ***list, char *entry)
1890 {
1891         char *copy;
1892         int newentry;
1893
1894         newentry = lxc_append_null_to_list((void ***)list);
1895         if (newentry < 0)
1896                 return -1;
1897
1898         copy = strdup(entry);
1899         if (!copy)
1900                 return -1;
1901
1902         (*list)[newentry] = copy;
1903
1904         return 0;
1905 }
1906
1907 int lxc_preserve_ns(const int pid, const char *ns)
1908 {
1909         int ret;
1910 /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */
1911 #define __NS_PATH_LEN 50
1912         char path[__NS_PATH_LEN];
1913
1914         /* This way we can use this function to also check whether namespaces
1915          * are supported by the kernel by passing in the NULL or the empty
1916          * string.
1917          */
1918         ret = snprintf(path, __NS_PATH_LEN, "/proc/%d/ns%s%s", pid,
1919                        !ns || strcmp(ns, "") == 0 ? "" : "/",
1920                        !ns || strcmp(ns, "") == 0 ? "" : ns);
1921         errno = EFBIG;
1922         if (ret < 0 || (size_t)ret >= __NS_PATH_LEN)
1923                 return -EFBIG;
1924
1925         return open(path, O_RDONLY | O_CLOEXEC);
1926 }
1927
1928 int lxc_safe_uint(const char *numstr, unsigned int *converted)
1929 {
1930         char *err = NULL;
1931         unsigned long int uli;
1932
1933         while (isspace(*numstr))
1934                 numstr++;
1935
1936         if (*numstr == '-')
1937                 return -EINVAL;
1938
1939         errno = 0;
1940         uli = strtoul(numstr, &err, 0);
1941         if (errno == ERANGE && uli == ULONG_MAX)
1942                 return -ERANGE;
1943
1944         if (err == numstr || *err != '\0')
1945                 return -EINVAL;
1946
1947         if (uli > UINT_MAX)
1948                 return -ERANGE;
1949
1950         *converted = (unsigned int)uli;
1951         return 0;
1952 }
1953
1954 int lxc_safe_ulong(const char *numstr, unsigned long *converted)
1955 {
1956         char *err = NULL;
1957         unsigned long int uli;
1958
1959         while (isspace(*numstr))
1960                 numstr++;
1961
1962         if (*numstr == '-')
1963                 return -EINVAL;
1964
1965         errno = 0;
1966         uli = strtoul(numstr, &err, 0);
1967         if (errno == ERANGE && uli == ULONG_MAX)
1968                 return -ERANGE;
1969
1970         if (err == numstr || *err != '\0')
1971                 return -EINVAL;
1972
1973         *converted = uli;
1974         return 0;
1975 }
1976
1977 int lxc_safe_int(const char *numstr, int *converted)
1978 {
1979         char *err = NULL;
1980         signed long int sli;
1981
1982         errno = 0;
1983         sli = strtol(numstr, &err, 0);
1984         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
1985                 return -ERANGE;
1986
1987         if (errno != 0 && sli == 0)
1988                 return -EINVAL;
1989
1990         if (err == numstr || *err != '\0')
1991                 return -EINVAL;
1992
1993         if (sli > INT_MAX || sli < INT_MIN)
1994                 return -ERANGE;
1995
1996         *converted = (int)sli;
1997         return 0;
1998 }
1999
2000 int lxc_safe_long(const char *numstr, long int *converted)
2001 {
2002         char *err = NULL;
2003         signed long int sli;
2004
2005         errno = 0;
2006         sli = strtol(numstr, &err, 0);
2007         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
2008                 return -ERANGE;
2009
2010         if (errno != 0 && sli == 0)
2011                 return -EINVAL;
2012
2013         if (err == numstr || *err != '\0')
2014                 return -EINVAL;
2015
2016         *converted = sli;
2017         return 0;
2018 }
2019
2020 int lxc_safe_long_long(const char *numstr, long long int *converted)
2021 {
2022         char *err = NULL;
2023         signed long long int sli;
2024
2025         errno = 0;
2026         sli = strtoll(numstr, &err, 0);
2027         if (errno == ERANGE && (sli == LLONG_MAX || sli == LLONG_MIN))
2028                 return -ERANGE;
2029
2030         if (errno != 0 && sli == 0)
2031                 return -EINVAL;
2032
2033         if (err == numstr || *err != '\0')
2034                 return -EINVAL;
2035
2036         *converted = sli;
2037         return 0;
2038 }
2039
2040 int lxc_switch_uid_gid(uid_t uid, gid_t gid)
2041 {
2042         if (setgid(gid) < 0) {
2043                 SYSERROR("Failed to switch to gid %d.", gid);
2044                 return -errno;
2045         }
2046         NOTICE("Switched to gid %d.", gid);
2047
2048         if (setuid(uid) < 0) {
2049                 SYSERROR("Failed to switch to uid %d.", uid);
2050                 return -errno;
2051         }
2052         NOTICE("Switched to uid %d.", uid);
2053
2054         return 0;
2055 }
2056
2057 /* Simple covenience function which enables uniform logging. */
2058 int lxc_setgroups(int size, gid_t list[])
2059 {
2060         if (setgroups(size, list) < 0) {
2061                 SYSERROR("Failed to setgroups().");
2062                 return -errno;
2063         }
2064         NOTICE("Dropped additional groups.");
2065
2066         return 0;
2067 }
2068
2069 static int lxc_get_unused_loop_dev_legacy(char *loop_name)
2070 {
2071         struct dirent *dp;
2072         struct loop_info64 lo64;
2073         DIR *dir;
2074         int dfd = -1, fd = -1, ret = -1;
2075
2076         dir = opendir("/dev");
2077         if (!dir)
2078                 return -1;
2079
2080         while ((dp = readdir(dir))) {
2081                 if (!dp)
2082                         break;
2083
2084                 if (strncmp(dp->d_name, "loop", 4) != 0)
2085                         continue;
2086
2087                 dfd = dirfd(dir);
2088                 if (dfd < 0)
2089                         continue;
2090
2091                 fd = openat(dfd, dp->d_name, O_RDWR);
2092                 if (fd < 0)
2093                         continue;
2094
2095                 ret = ioctl(fd, LOOP_GET_STATUS64, &lo64);
2096                 if (ret < 0) {
2097                         if (ioctl(fd, LOOP_GET_STATUS64, &lo64) == 0 ||
2098                             errno != ENXIO) {
2099                                 close(fd);
2100                                 fd = -1;
2101                                 continue;
2102                         }
2103                 }
2104
2105                 ret = snprintf(loop_name, LO_NAME_SIZE, "/dev/%s", dp->d_name);
2106                 if (ret < 0 || ret >= LO_NAME_SIZE) {
2107                         close(fd);
2108                         fd = -1;
2109                         continue;
2110                 }
2111
2112                 break;
2113         }
2114
2115         closedir(dir);
2116
2117         if (fd < 0)
2118                 return -1;
2119
2120         return fd;
2121 }
2122
2123 static int lxc_get_unused_loop_dev(char *name_loop)
2124 {
2125         int loop_nr, ret;
2126         int fd_ctl = -1, fd_tmp = -1;
2127
2128         fd_ctl = open("/dev/loop-control", O_RDWR | O_CLOEXEC);
2129         if (fd_ctl < 0)
2130                 return -ENODEV;
2131
2132         loop_nr = ioctl(fd_ctl, LOOP_CTL_GET_FREE);
2133         if (loop_nr < 0)
2134                 goto on_error;
2135
2136         ret = snprintf(name_loop, LO_NAME_SIZE, "/dev/loop%d", loop_nr);
2137         if (ret < 0 || ret >= LO_NAME_SIZE)
2138                 goto on_error;
2139
2140         fd_tmp = open(name_loop, O_RDWR | O_CLOEXEC);
2141         if (fd_tmp < 0)
2142                 goto on_error;
2143
2144 on_error:
2145         close(fd_ctl);
2146         return fd_tmp;
2147 }
2148
2149 int lxc_prepare_loop_dev(const char *source, char *loop_dev, int flags)
2150 {
2151         int ret;
2152         struct loop_info64 lo64;
2153         int fd_img = -1, fret = -1, fd_loop = -1;
2154
2155         fd_loop = lxc_get_unused_loop_dev(loop_dev);
2156         if (fd_loop < 0) {
2157                 if (fd_loop == -ENODEV)
2158                         fd_loop = lxc_get_unused_loop_dev_legacy(loop_dev);
2159                 else
2160                         goto on_error;
2161         }
2162
2163         fd_img = open(source, O_RDWR | O_CLOEXEC);
2164         if (fd_img < 0)
2165                 goto on_error;
2166
2167         ret = ioctl(fd_loop, LOOP_SET_FD, fd_img);
2168         if (ret < 0)
2169                 goto on_error;
2170
2171         memset(&lo64, 0, sizeof(lo64));
2172         lo64.lo_flags = flags;
2173
2174         ret = ioctl(fd_loop, LOOP_SET_STATUS64, &lo64);
2175         if (ret < 0)
2176                 goto on_error;
2177
2178         fret = 0;
2179
2180 on_error:
2181         if (fd_img >= 0)
2182                 close(fd_img);
2183
2184         if (fret < 0 && fd_loop >= 0) {
2185                 close(fd_loop);
2186                 fd_loop = -1;
2187         }
2188
2189         return fd_loop;
2190 }
2191
2192 int lxc_unstack_mountpoint(const char *path, bool lazy)
2193 {
2194         int ret;
2195         int umounts = 0;
2196
2197 pop_stack:
2198         ret = umount2(path, lazy ? MNT_DETACH : 0);
2199         if (ret < 0) {
2200                 /* We consider anything else than EINVAL deadly to prevent going
2201                  * into an infinite loop. (The other alternative is constantly
2202                  * parsing /proc/self/mountinfo which is yucky and probably
2203                  * racy.)
2204                  */
2205                 if (errno != EINVAL)
2206                         return -errno;
2207         } else {
2208                 /* Just stop counting when this happens. That'd just be so
2209                  * stupid that we won't even bother trying to report back the
2210                  * correct value anymore.
2211                  */
2212                 if (umounts != INT_MAX)
2213                         umounts++;
2214                 /* We succeeded in umounting. Make sure that there's no other
2215                  * mountpoint stacked underneath.
2216                  */
2217                 goto pop_stack;
2218         }
2219
2220         return umounts;
2221 }
2222
2223 int run_command(char *buf, size_t buf_size, int (*child_fn)(void *), void *args)
2224 {
2225         pid_t child;
2226         int ret, fret, pipefd[2];
2227         ssize_t bytes;
2228
2229         /* Make sure our callers do not receive unitialized memory. */
2230         if (buf_size > 0 && buf)
2231                 buf[0] = '\0';
2232
2233         if (pipe(pipefd) < 0) {
2234                 SYSERROR("failed to create pipe");
2235                 return -1;
2236         }
2237
2238         child = lxc_raw_clone(0);
2239         if (child < 0) {
2240                 close(pipefd[0]);
2241                 close(pipefd[1]);
2242                 SYSERROR("failed to create new process");
2243                 return -1;
2244         }
2245
2246         if (child == 0) {
2247                 /* Close the read-end of the pipe. */
2248                 close(pipefd[0]);
2249
2250                 /* Redirect std{err,out} to write-end of the
2251                  * pipe.
2252                  */
2253                 ret = dup2(pipefd[1], STDOUT_FILENO);
2254                 if (ret >= 0)
2255                         ret = dup2(pipefd[1], STDERR_FILENO);
2256
2257                 /* Close the write-end of the pipe. */
2258                 close(pipefd[1]);
2259
2260                 if (ret < 0) {
2261                         SYSERROR("failed to duplicate std{err,out} file descriptor");
2262                         exit(EXIT_FAILURE);
2263                 }
2264
2265                 /* Does not return. */
2266                 child_fn(args);
2267                 ERROR("failed to exec command");
2268                 exit(EXIT_FAILURE);
2269         }
2270
2271         /* close the write-end of the pipe */
2272         close(pipefd[1]);
2273
2274         if (buf && buf_size > 0) {
2275                 bytes = read(pipefd[0], buf, buf_size - 1);
2276                 if (bytes > 0)
2277                         buf[bytes - 1] = '\0';
2278         }
2279
2280         fret = wait_for_pid(child);
2281         /* close the read-end of the pipe */
2282         close(pipefd[0]);
2283
2284         return fret;
2285 }
2286
2287 char *must_make_path(const char *first, ...)
2288 {
2289         va_list args;
2290         char *cur, *dest;
2291         size_t full_len = strlen(first);
2292
2293         dest = must_copy_string(first);
2294
2295         va_start(args, first);
2296         while ((cur = va_arg(args, char *)) != NULL) {
2297                 full_len += strlen(cur);
2298                 if (cur[0] != '/')
2299                         full_len++;
2300                 dest = must_realloc(dest, full_len + 1);
2301                 if (cur[0] != '/')
2302                         strcat(dest, "/");
2303                 strcat(dest, cur);
2304         }
2305         va_end(args);
2306
2307         return dest;
2308 }
2309
2310 char *must_copy_string(const char *entry)
2311 {
2312         char *ret;
2313
2314         if (!entry)
2315                 return NULL;
2316         do {
2317                 ret = strdup(entry);
2318         } while (!ret);
2319
2320         return ret;
2321 }
2322
2323 void *must_realloc(void *orig, size_t sz)
2324 {
2325         void *ret;
2326
2327         do {
2328                 ret = realloc(orig, sz);
2329         } while (!ret);
2330
2331         return ret;
2332 }
2333
2334 bool is_fs_type(const struct statfs *fs, fs_type_magic magic_val)
2335 {
2336         return (fs->f_type == (fs_type_magic)magic_val);
2337 }
2338
2339 bool has_fs_type(const char *path, fs_type_magic magic_val)
2340 {
2341         bool has_type;
2342         int ret;
2343         struct statfs sb;
2344
2345         ret = statfs(path, &sb);
2346         if (ret < 0)
2347                 return false;
2348
2349         has_type = is_fs_type(&sb, magic_val);
2350         if (!has_type && magic_val == RAMFS_MAGIC)
2351                 WARN("When the ramfs it a tmpfs statfs() might report tmpfs");
2352
2353         return has_type;
2354 }
2355
2356 bool lxc_nic_exists(char *nic)
2357 {
2358 #define __LXC_SYS_CLASS_NET_LEN 15 + IFNAMSIZ + 1
2359         char path[__LXC_SYS_CLASS_NET_LEN];
2360         int ret;
2361         struct stat sb;
2362
2363         if (!strcmp(nic, "none"))
2364                 return true;
2365
2366         ret = snprintf(path, __LXC_SYS_CLASS_NET_LEN, "/sys/class/net/%s", nic);
2367         if (ret < 0 || (size_t)ret >= __LXC_SYS_CLASS_NET_LEN)
2368                 return false;
2369
2370         ret = stat(path, &sb);
2371         if (ret < 0)
2372                 return false;
2373
2374         return true;
2375 }
2376
2377 int lxc_make_tmpfile(char *template, bool rm)
2378 {
2379         int fd, ret;
2380
2381         fd = mkstemp(template);
2382         if (fd < 0)
2383                 return -1;
2384
2385         if (!rm)
2386                 return fd;
2387
2388         ret = unlink(template);
2389         if (ret < 0) {
2390                 close(fd);
2391                 return -1;
2392         }
2393
2394         return fd;
2395 }
2396
2397 int parse_byte_size_string(const char *s, int64_t *converted)
2398 {
2399         int ret, suffix_len;
2400         long long int conv;
2401         int64_t mltpl, overflow;
2402         char *end;
2403         char dup[LXC_NUMSTRLEN64 + 2];
2404         char suffix[3] = {0};
2405
2406         if (!s || !strcmp(s, ""))
2407                 return -EINVAL;
2408
2409         end = stpncpy(dup, s, sizeof(dup));
2410         if (*end != '\0')
2411                 return -EINVAL;
2412
2413         if (isdigit(*(end - 1)))
2414                 suffix_len = 0;
2415         else if (isalpha(*(end - 1)))
2416                 suffix_len = 1;
2417         else
2418                 return -EINVAL;
2419
2420         if (suffix_len > 0 && (end - 2) == dup && !isdigit(*(end - 2)))
2421                 return -EINVAL;
2422
2423         if (suffix_len > 0 && isalpha(*(end - 2)))
2424                 suffix_len++;
2425
2426         if (suffix_len > 0) {
2427                 memcpy(suffix, end - suffix_len, suffix_len);
2428                 *(suffix + suffix_len) = '\0';
2429                 *(end - suffix_len) = '\0';
2430         }
2431         dup[lxc_char_right_gc(dup, strlen(dup))] = '\0';
2432
2433         ret = lxc_safe_long_long(dup, &conv);
2434         if (ret < 0)
2435                 return -ret;
2436
2437         if (suffix_len != 2) {
2438                 *converted = conv;
2439                 return 0;
2440         }
2441
2442         if (!strcmp(suffix, "kB"))
2443                 mltpl = 1024;
2444         else if (!strcmp(suffix, "MB"))
2445                 mltpl = 1024 * 1024;
2446         else if (!strcmp(suffix, "GB"))
2447                 mltpl = 1024 * 1024 * 1024;
2448         else
2449                 return -EINVAL;
2450
2451         overflow = conv * mltpl;
2452         if (conv != 0 && (overflow / conv) != mltpl)
2453                 return -ERANGE;
2454
2455         *converted = overflow;
2456         return 0;
2457 }
2458
2459 uint64_t lxc_find_next_power2(uint64_t n)
2460 {
2461         /* 0 is not valid input. We return 0 to the caller since 0 is not a
2462          * valid power of two.
2463          */
2464         if (n == 0)
2465                 return 0;
2466
2467         if (!(n & (n - 1)))
2468                 return n;
2469
2470         while (n & (n - 1))
2471                 n = n & (n - 1);
2472
2473         n = n << 1;
2474         return n;
2475 }