src/lxc/utils.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #include "config.h"
  25
  26 #define __STDC_FORMAT_MACROS /* Required for PRIu64 to work. */
  27 #include <ctype.h>
  28 #include <dirent.h>
  29 #include <errno.h>
  30 #include <fcntl.h>
  31 #include <grp.h>
  32 #include <inttypes.h>
  33 #include <libgen.h>
  34 #include <pthread.h>
  35 #include <stddef.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <string.h>
  39 #include <unistd.h>
  40 #include <sys/mman.h>
  41 #include <sys/mount.h>
  42 #include <sys/param.h>
  43 #include <sys/prctl.h>
  44 #include <sys/stat.h>
  45 #include <sys/types.h>
  46 #include <sys/wait.h>
  47
  48 #include "log.h"
  49 #include "lxclock.h"
  50 #include "namespace.h"
  51 #include "parse.h"
  52 #include "utils.h"
  53
  54 #ifndef HAVE_STRLCPY
  55 #include "include/strlcpy.h"
  56 #endif
  57
  58 #ifndef HAVE_STRLCAT
  59 #include "include/strlcat.h"
  60 #endif
  61
  62 #ifndef O_PATH
  63 #define O_PATH      010000000
  64 #endif
  65
  66 #ifndef O_NOFOLLOW
  67 #define O_NOFOLLOW  00400000
  68 #endif
  69
  70 lxc_log_define(utils, lxc);
  71
  72 /*
  73  * if path is btrfs, tries to remove it and any subvolumes beneath it
  74  */
  75 extern bool btrfs_try_remove_subvol(const char *path);
  76
  77 static int _recursive_rmdir(const char *dirname, dev_t pdev,
  78                             const char *exclude, int level, bool onedev)
  79 {
  80         struct dirent *direntp;
  81         DIR *dir;
  82         int ret, failed=0;
  83         char pathname[MAXPATHLEN];
  84         bool hadexclude = false;
  85
  86         dir = opendir(dirname);
  87         if (!dir) {
  88                 ERROR("failed to open %s", dirname);
  89                 return -1;
  90         }
  91
  92         while ((direntp = readdir(dir))) {
  93                 struct stat mystat;
  94                 int rc;
  95
  96                 if (!strcmp(direntp->d_name, ".") ||
  97                     !strcmp(direntp->d_name, ".."))
  98                         continue;
  99
 100                 rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name);
 101                 if (rc < 0 || rc >= MAXPATHLEN) {
 102                         ERROR("pathname too long");
 103                         failed=1;
 104                         continue;
 105                 }
 106
 107                 if (!level && exclude && !strcmp(direntp->d_name, exclude)) {
 108                         ret = rmdir(pathname);
 109                         if (ret < 0) {
 110                                 switch(errno) {
 111                                 case ENOTEMPTY:
 112                                         INFO("Not deleting snapshot %s", pathname);
 113                                         hadexclude = true;
 114                                         break;
 115                                 case ENOTDIR:
 116                                         ret = unlink(pathname);
 117                                         if (ret)
 118                                                 INFO("Failed to remove %s", pathname);
 119                                         break;
 120                                 default:
 121                                         SYSERROR("Failed to rmdir %s", pathname);
 122                                         failed = 1;
 123                                         break;
 124                                 }
 125                         }
 126                         continue;
 127                 }
 128
 129                 ret = lstat(pathname, &mystat);
 130                 if (ret) {
 131                         ERROR("Failed to stat %s", pathname);
 132                         failed = 1;
 133                         continue;
 134                 }
 135
 136                 if (onedev && mystat.st_dev != pdev) {
 137                         /* TODO should we be checking /proc/self/mountinfo for
 138                          * pathname and not doing this if found? */
 139                         if (btrfs_try_remove_subvol(pathname))
 140                                 INFO("Removed btrfs subvolume at %s\n", pathname);
 141                         continue;
 142                 }
 143
 144                 if (S_ISDIR(mystat.st_mode)) {
 145                         if (_recursive_rmdir(pathname, pdev, exclude, level+1, onedev) < 0)
 146                                 failed=1;
 147                 } else {
 148                         if (unlink(pathname) < 0) {
 149                                 SYSERROR("Failed to delete %s", pathname);
 150                                 failed=1;
 151                         }
 152                 }
 153         }
 154
 155         if (rmdir(dirname) < 0 && !btrfs_try_remove_subvol(dirname) && !hadexclude) {
 156                 ERROR("Failed to delete %s", dirname);
 157                 failed=1;
 158         }
 159
 160         ret = closedir(dir);
 161         if (ret) {
 162                 ERROR("Failed to close directory %s", dirname);
 163                 failed=1;
 164         }
 165
 166         return failed ? -1 : 0;
 167 }
 168
 169 /* We have two different magic values for overlayfs, yay. */
 170 #ifndef OVERLAYFS_SUPER_MAGIC
 171 #define OVERLAYFS_SUPER_MAGIC 0x794c764f
 172 #endif
 173
 174 #ifndef OVERLAY_SUPER_MAGIC
 175 #define OVERLAY_SUPER_MAGIC 0x794c7630
 176 #endif
 177
 178 /* In overlayfs, st_dev is unreliable. So on overlayfs we don't do the
 179  * lxc_rmdir_onedev()
 180  */
 181 static bool is_native_overlayfs(const char *path)
 182 {
 183         if (has_fs_type(path, OVERLAY_SUPER_MAGIC) ||
 184             has_fs_type(path, OVERLAYFS_SUPER_MAGIC))
 185                 return true;
 186
 187         return false;
 188 }
 189
 190 /* returns 0 on success, -1 if there were any failures */
 191 extern int lxc_rmdir_onedev(const char *path, const char *exclude)
 192 {
 193         struct stat mystat;
 194         bool onedev = true;
 195
 196         if (is_native_overlayfs(path))
 197                 onedev = false;
 198
 199         if (lstat(path, &mystat) < 0) {
 200                 if (errno == ENOENT)
 201                         return 0;
 202
 203                 ERROR("Failed to stat %s", path);
 204                 return -1;
 205         }
 206
 207         return _recursive_rmdir(path, mystat.st_dev, exclude, 0, onedev);
 208 }
 209
 210 /* borrowed from iproute2 */
 211 extern int get_u16(unsigned short *val, const char *arg, int base)
 212 {
 213         unsigned long res;
 214         char *ptr;
 215
 216         if (!arg || !*arg)
 217                 return -1;
 218
 219         errno = 0;
 220         res = strtoul(arg, &ptr, base);
 221         if (!ptr || ptr == arg || *ptr || res > 0xFFFF || errno != 0)
 222                 return -1;
 223
 224         *val = res;
 225
 226         return 0;
 227 }
 228
 229 extern int mkdir_p(const char *dir, mode_t mode)
 230 {
 231         const char *tmp = dir;
 232         const char *orig = dir;
 233         char *makeme;
 234
 235         do {
 236                 dir = tmp + strspn(tmp, "/");
 237                 tmp = dir + strcspn(dir, "/");
 238
 239                 makeme = strndup(orig, dir - orig);
 240                 if (*makeme) {
 241                         if (mkdir(makeme, mode) && errno != EEXIST) {
 242                                 SYSERROR("failed to create directory '%s'", makeme);
 243                                 free(makeme);
 244                                 return -1;
 245                         }
 246                 }
 247                 free(makeme);
 248         } while(tmp != dir);
 249
 250         return 0;
 251 }
 252
 253 char *get_rundir()
 254 {
 255         char *rundir;
 256         const char *homedir;
 257         struct stat sb;
 258
 259         if (stat(RUNTIME_PATH, &sb) < 0)
 260                 return NULL;
 261
 262         if (geteuid() == sb.st_uid || getegid() == sb.st_gid) {
 263                 rundir = strdup(RUNTIME_PATH);
 264                 return rundir;
 265         }
 266
 267         rundir = getenv("XDG_RUNTIME_DIR");
 268         if (rundir) {
 269                 rundir = strdup(rundir);
 270                 return rundir;
 271         }
 272
 273         INFO("XDG_RUNTIME_DIR isn't set in the environment.");
 274         homedir = getenv("HOME");
 275         if (!homedir) {
 276                 ERROR("HOME isn't set in the environment.");
 277                 return NULL;
 278         }
 279
 280         rundir = malloc(sizeof(char) * (17 + strlen(homedir)));
 281         if (!rundir)
 282                 return NULL;
 283
 284         sprintf(rundir, "%s/.cache/lxc/run/", homedir);
 285
 286         return rundir;
 287 }
 288
 289 int wait_for_pid(pid_t pid)
 290 {
 291         int status, ret;
 292
 293 again:
 294         ret = waitpid(pid, &status, 0);
 295         if (ret == -1) {
 296                 if (errno == EINTR)
 297                         goto again;
 298
 299                 return -1;
 300         }
 301
 302         if (ret != pid)
 303                 goto again;
 304
 305         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0)
 306                 return -1;
 307
 308         return 0;
 309 }
 310
 311 int lxc_wait_for_pid_status(pid_t pid)
 312 {
 313         int status, ret;
 314
 315 again:
 316         ret = waitpid(pid, &status, 0);
 317         if (ret == -1) {
 318                 if (errno == EINTR)
 319                         goto again;
 320
 321                 return -1;
 322         }
 323
 324         if (ret != pid)
 325                 goto again;
 326
 327         return status;
 328 }
 329
 330 ssize_t lxc_write_nointr(int fd, const void *buf, size_t count)
 331 {
 332         ssize_t ret;
 333 again:
 334         ret = write(fd, buf, count);
 335         if (ret < 0 && errno == EINTR)
 336                 goto again;
 337
 338         return ret;
 339 }
 340
 341 ssize_t lxc_read_nointr(int fd, void *buf, size_t count)
 342 {
 343         ssize_t ret;
 344 again:
 345         ret = read(fd, buf, count);
 346         if (ret < 0 && errno == EINTR)
 347                 goto again;
 348
 349         return ret;
 350 }
 351
 352 ssize_t lxc_read_nointr_expect(int fd, void *buf, size_t count, const void *expected_buf)
 353 {
 354         ssize_t ret;
 355
 356         ret = lxc_read_nointr(fd, buf, count);
 357         if (ret <= 0)
 358                 return ret;
 359
 360         if ((size_t)ret != count)
 361                 return -1;
 362
 363         if (expected_buf && memcmp(buf, expected_buf, count) != 0) {
 364                 errno = EINVAL;
 365                 return -1;
 366         }
 367
 368         return ret;
 369 }
 370
 371 #if HAVE_LIBGNUTLS
 372 #include <gnutls/gnutls.h>
 373 #include <gnutls/crypto.h>
 374
 375 __attribute__((constructor))
 376 static void gnutls_lxc_init(void)
 377 {
 378         gnutls_global_init();
 379 }
 380
 381 int sha1sum_file(char *fnam, unsigned char *digest)
 382 {
 383         char *buf;
 384         int ret;
 385         FILE *f;
 386         long flen;
 387
 388         if (!fnam)
 389                 return -1;
 390
 391         f = fopen_cloexec(fnam, "r");
 392         if (!f) {
 393                 SYSERROR("Error opening template");
 394                 return -1;
 395         }
 396
 397         if (fseek(f, 0, SEEK_END) < 0) {
 398                 SYSERROR("Error seeking to end of template");
 399                 fclose(f);
 400                 return -1;
 401         }
 402
 403         if ((flen = ftell(f)) < 0) {
 404                 SYSERROR("Error telling size of template");
 405                 fclose(f);
 406                 return -1;
 407         }
 408
 409         if (fseek(f, 0, SEEK_SET) < 0) {
 410                 SYSERROR("Error seeking to start of template");
 411                 fclose(f);
 412                 return -1;
 413         }
 414
 415         if ((buf = malloc(flen+1)) == NULL) {
 416                 SYSERROR("Out of memory");
 417                 fclose(f);
 418                 return -1;
 419         }
 420
 421         if (fread(buf, 1, flen, f) != flen) {
 422                 SYSERROR("Failure reading template");
 423                 free(buf);
 424                 fclose(f);
 425                 return -1;
 426         }
 427
 428         if (fclose(f) < 0) {
 429                 SYSERROR("Failre closing template");
 430                 free(buf);
 431                 return -1;
 432         }
 433
 434         buf[flen] = '\0';
 435         ret = gnutls_hash_fast(GNUTLS_DIG_SHA1, buf, flen, (void *)digest);
 436         free(buf);
 437         return ret;
 438 }
 439 #endif
 440
 441 char** lxc_va_arg_list_to_argv(va_list ap, size_t skip, int do_strdup)
 442 {
 443         va_list ap2;
 444         size_t count = 1 + skip;
 445         char **result;
 446
 447         /* first determine size of argument list, we don't want to reallocate
 448          * constantly...
 449          */
 450         va_copy(ap2, ap);
 451         while (1) {
 452                 char* arg = va_arg(ap2, char*);
 453                 if (!arg)
 454                         break;
 455                 count++;
 456         }
 457         va_end(ap2);
 458
 459         result = calloc(count, sizeof(char*));
 460         if (!result)
 461                 return NULL;
 462
 463         count = skip;
 464         while (1) {
 465                 char* arg = va_arg(ap, char*);
 466                 if (!arg)
 467                         break;
 468                 arg = do_strdup ? strdup(arg) : arg;
 469                 if (!arg)
 470                         goto oom;
 471                 result[count++] = arg;
 472         }
 473
 474         /* calloc has already set last element to NULL*/
 475         return result;
 476
 477 oom:
 478         free(result);
 479         return NULL;
 480 }
 481
 482 const char** lxc_va_arg_list_to_argv_const(va_list ap, size_t skip)
 483 {
 484         return (const char**)lxc_va_arg_list_to_argv(ap, skip, 0);
 485 }
 486
 487 struct lxc_popen_FILE *lxc_popen(const char *command)
 488 {
 489         int ret;
 490         int pipe_fds[2];
 491         pid_t child_pid;
 492         struct lxc_popen_FILE *fp = NULL;
 493
 494         ret = pipe2(pipe_fds, O_CLOEXEC);
 495         if (ret < 0)
 496                 return NULL;
 497
 498         child_pid = fork();
 499         if (child_pid < 0)
 500                 goto on_error;
 501
 502         if (!child_pid) {
 503                 sigset_t mask;
 504
 505                 close(pipe_fds[0]);
 506
 507                 /* duplicate stdout */
 508                 if (pipe_fds[1] != STDOUT_FILENO)
 509                         ret = dup2(pipe_fds[1], STDOUT_FILENO);
 510                 else
 511                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 512                 if (ret < 0) {
 513                         close(pipe_fds[1]);
 514                         _exit(EXIT_FAILURE);
 515                 }
 516
 517                 /* duplicate stderr */
 518                 if (pipe_fds[1] != STDERR_FILENO)
 519                         ret = dup2(pipe_fds[1], STDERR_FILENO);
 520                 else
 521                         ret = fcntl(pipe_fds[1], F_SETFD, 0);
 522                 close(pipe_fds[1]);
 523                 if (ret < 0)
 524                         _exit(EXIT_FAILURE);
 525
 526                 /* unblock all signals */
 527                 ret = sigfillset(&mask);
 528                 if (ret < 0)
 529                         _exit(EXIT_FAILURE);
 530
 531                 ret = pthread_sigmask(SIG_UNBLOCK, &mask, NULL);
 532                 if (ret < 0)
 533                         _exit(EXIT_FAILURE);
 534
 535                 execl("/bin/sh", "sh", "-c", command, (char *)NULL);
 536                 _exit(127);
 537         }
 538
 539         close(pipe_fds[1]);
 540         pipe_fds[1] = -1;
 541
 542         fp = malloc(sizeof(*fp));
 543         if (!fp)
 544                 goto on_error;
 545
 546         memset(fp, 0, sizeof(*fp));
 547
 548         fp->child_pid = child_pid;
 549         fp->pipe = pipe_fds[0];
 550
 551         /* From now on, closing fp->f will also close fp->pipe. So only ever
 552          * call fclose(fp->f).
 553          */
 554         fp->f = fdopen(pipe_fds[0], "r");
 555         if (!fp->f)
 556                 goto on_error;
 557
 558         return fp;
 559
 560 on_error:
 561         /* We can only close pipe_fds[0] if fdopen() didn't succeed or wasn't
 562          * called yet. Otherwise the fd belongs to the file opened by fdopen()
 563          * since it isn't dup()ed.
 564          */
 565         if (fp && !fp->f && pipe_fds[0] >= 0)
 566                 close(pipe_fds[0]);
 567
 568         if (pipe_fds[1] >= 0)
 569                 close(pipe_fds[1]);
 570
 571         if (fp && fp->f)
 572                 fclose(fp->f);
 573
 574         if (fp)
 575                 free(fp);
 576
 577         return NULL;
 578 }
 579
 580 int lxc_pclose(struct lxc_popen_FILE *fp)
 581 {
 582         pid_t wait_pid;
 583         int wstatus = 0;
 584
 585         if (!fp)
 586                 return -1;
 587
 588         do {
 589                 wait_pid = waitpid(fp->child_pid, &wstatus, 0);
 590         } while (wait_pid < 0 && errno == EINTR);
 591
 592         fclose(fp->f);
 593         free(fp);
 594
 595         if (wait_pid < 0)
 596                 return -1;
 597
 598         return wstatus;
 599 }
 600
 601 char *lxc_string_replace(const char *needle, const char *replacement, const char *haystack)
 602 {
 603         ssize_t len = -1, saved_len = -1;
 604         char *result = NULL;
 605         size_t replacement_len = strlen(replacement);
 606         size_t needle_len = strlen(needle);
 607
 608         /* should be executed exactly twice */
 609         while (len == -1 || result == NULL) {
 610                 char *p;
 611                 char *last_p;
 612                 ssize_t part_len;
 613
 614                 if (len != -1) {
 615                         result = calloc(1, len + 1);
 616                         if (!result)
 617                                 return NULL;
 618
 619                         saved_len = len;
 620                 }
 621
 622                 len = 0;
 623
 624                 for (last_p = (char *)haystack, p = strstr(last_p, needle); p; last_p = p, p = strstr(last_p, needle)) {
 625                         part_len = (ssize_t)(p - last_p);
 626                         if (result && part_len > 0)
 627                                 memcpy(&result[len], last_p, part_len);
 628
 629                         len += part_len;
 630
 631                         if (result && replacement_len > 0)
 632                                 memcpy(&result[len], replacement, replacement_len);
 633
 634                         len += replacement_len;
 635                         p += needle_len;
 636                 }
 637
 638                 part_len = strlen(last_p);
 639                 if (result && part_len > 0)
 640                         memcpy(&result[len], last_p, part_len);
 641
 642                 len += part_len;
 643         }
 644
 645         /* make sure we did the same thing twice,
 646          * once for calculating length, the other
 647          * time for copying data */
 648         if (saved_len != len) {
 649                 free(result);
 650                 return NULL;
 651         }
 652
 653         /* make sure we didn't overwrite any buffer,
 654          * due to calloc the string should be 0-terminated */
 655         if (result[len] != '\0') {
 656                 free(result);
 657                 return NULL;
 658         }
 659
 660         return result;
 661 }
 662
 663 bool lxc_string_in_array(const char *needle, const char **haystack)
 664 {
 665         for (; haystack && *haystack; haystack++)
 666                 if (!strcmp(needle, *haystack))
 667                         return true;
 668
 669         return false;
 670 }
 671
 672 char *lxc_string_join(const char *sep, const char **parts, bool use_as_prefix)
 673 {
 674         char *result;
 675         char **p;
 676         size_t sep_len = strlen(sep);
 677         size_t result_len = use_as_prefix * sep_len;
 678         size_t buf_len;
 679
 680         /* calculate new string length */
 681         for (p = (char **)parts; *p; p++)
 682                 result_len += (p > (char **)parts) * sep_len + strlen(*p);
 683
 684         buf_len = result_len + 1;
 685         result = calloc(buf_len, 1);
 686         if (!result)
 687                 return NULL;
 688
 689         if (use_as_prefix)
 690                 (void)strlcpy(result, sep, buf_len);
 691
 692         for (p = (char **)parts; *p; p++) {
 693                 if (p > (char **)parts)
 694                         (void)strlcat(result, sep, buf_len);
 695
 696                 (void)strlcat(result, *p, buf_len);
 697         }
 698
 699         return result;
 700 }
 701
 702 char **lxc_normalize_path(const char *path)
 703 {
 704         char **components;
 705         char **p;
 706         size_t components_len = 0;
 707         size_t pos = 0;
 708
 709         components = lxc_string_split(path, '/');
 710         if (!components)
 711                 return NULL;
 712
 713         for (p = components; *p; p++)
 714                 components_len++;
 715
 716         /* resolve '.' and '..' */
 717         for (pos = 0; pos < components_len; ) {
 718                 if (!strcmp(components[pos], ".") || (!strcmp(components[pos], "..") && pos == 0)) {
 719                         /* eat this element */
 720                         free(components[pos]);
 721                         memmove(&components[pos], &components[pos+1], sizeof(char *) * (components_len - pos));
 722                         components_len--;
 723                 } else if (!strcmp(components[pos], "..")) {
 724                         /* eat this and the previous element */
 725                         free(components[pos - 1]);
 726                         free(components[pos]);
 727                         memmove(&components[pos-1], &components[pos+1], sizeof(char *) * (components_len - pos));
 728                         components_len -= 2;
 729                         pos--;
 730                 } else {
 731                         pos++;
 732                 }
 733         }
 734
 735         return components;
 736 }
 737
 738 char *lxc_deslashify(const char *path)
 739 {
 740         char *dup, *p;
 741         char **parts = NULL;
 742         size_t n, len;
 743
 744         dup = strdup(path);
 745         if (!dup)
 746                 return NULL;
 747
 748         parts = lxc_normalize_path(dup);
 749         if (!parts) {
 750                 free(dup);
 751                 return NULL;
 752         }
 753
 754         /* We'll end up here if path == "///" or path == "". */
 755         if (!*parts) {
 756                 len = strlen(dup);
 757                 if (!len) {
 758                         lxc_free_array((void **)parts, free);
 759                         return dup;
 760                 }
 761
 762                 n = strcspn(dup, "/");
 763                 if (n == len) {
 764                         free(dup);
 765                         lxc_free_array((void **)parts, free);
 766
 767                         p = strdup("/");
 768                         if (!p)
 769                                 return NULL;
 770
 771                         return p;
 772                 }
 773         }
 774
 775         p = lxc_string_join("/", (const char **)parts, *dup == '/');
 776         free(dup);
 777         lxc_free_array((void **)parts, free);
 778         return p;
 779 }
 780
 781 char *lxc_append_paths(const char *first, const char *second)
 782 {
 783         int ret;
 784         size_t len;
 785         char *result = NULL;
 786         const char *pattern = "%s%s";
 787
 788         len = strlen(first) + strlen(second) + 1;
 789         if (second[0] != '/') {
 790                 len += 1;
 791                 pattern = "%s/%s";
 792         }
 793
 794         result = calloc(1, len);
 795         if (!result)
 796                 return NULL;
 797
 798         ret = snprintf(result, len, pattern, first, second);
 799         if (ret < 0 || (size_t)ret >= len) {
 800                 free(result);
 801                 return NULL;
 802         }
 803
 804         return result;
 805 }
 806
 807 bool lxc_string_in_list(const char *needle, const char *haystack, char _sep)
 808 {
 809         char *token, *str;
 810         char sep[2] = { _sep, '\0' };
 811         size_t len;
 812
 813         if (!haystack || !needle)
 814                 return 0;
 815
 816         len = strlen(haystack);
 817         str = alloca(len + 1);
 818         (void)strlcpy(str, haystack, len + 1);
 819
 820         lxc_iterate_parts(token, str, sep)
 821                 if (strcmp(needle, token) == 0)
 822                         return 1;
 823
 824         return 0;
 825 }
 826
 827 char **lxc_string_split(const char *string, char _sep)
 828 {
 829         char *token, *str;
 830         char sep[2] = {_sep, '\0'};
 831         char **tmp = NULL, **result = NULL;
 832         size_t result_capacity = 0;
 833         size_t result_count = 0;
 834         int r, saved_errno;
 835         size_t len;
 836
 837         if (!string)
 838                 return calloc(1, sizeof(char *));
 839
 840         len = strlen(string);
 841         str = alloca(len + 1);
 842         (void)strlcpy(str, string, len + 1);
 843
 844         lxc_iterate_parts(token, str, sep) {
 845                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 846                 if (r < 0)
 847                         goto error_out;
 848
 849                 result[result_count] = strdup(token);
 850                 if (!result[result_count])
 851                         goto error_out;
 852
 853                 result_count++;
 854         }
 855
 856         /* if we allocated too much, reduce it */
 857         tmp = realloc(result, (result_count + 1) * sizeof(char *));
 858         if (!tmp)
 859                 goto error_out;
 860
 861         result = tmp;
 862
 863         /* Make sure we don't return uninitialized memory. */
 864         if (result_count == 0)
 865                 *result = NULL;
 866
 867         return result;
 868
 869 error_out:
 870         saved_errno = errno;
 871         lxc_free_array((void **)result, free);
 872         errno = saved_errno;
 873         return NULL;
 874 }
 875
 876 static bool complete_word(char ***result, char *start, char *end, size_t *cap, size_t *cnt)
 877 {
 878         int r;
 879
 880         r = lxc_grow_array((void ***)result, cap, 2 + *cnt, 16);
 881         if (r < 0)
 882                 return false;
 883
 884         (*result)[*cnt] = strndup(start, end - start);
 885         if (!(*result)[*cnt])
 886                 return false;
 887
 888         (*cnt)++;
 889
 890         return true;
 891 }
 892
 893 /*
 894  * Given a a string 'one two "three four"', split into three words,
 895  * one, two, and "three four"
 896  */
 897 char **lxc_string_split_quoted(char *string)
 898 {
 899         char *nextword = string, *p, state;
 900         char **result = NULL;
 901         size_t result_capacity = 0;
 902         size_t result_count = 0;
 903
 904         if (!string || !*string)
 905                 return calloc(1, sizeof(char *));
 906
 907         // TODO I'm *not* handling escaped quote
 908         state = ' ';
 909         for (p = string; *p; p++) {
 910                 switch(state) {
 911                 case ' ':
 912                         if (isspace(*p))
 913                                 continue;
 914                         else if (*p == '"' || *p == '\'') {
 915                                 nextword = p;
 916                                 state = *p;
 917                                 continue;
 918                         }
 919                         nextword = p;
 920                         state = 'a';
 921                         continue;
 922                 case 'a':
 923                         if (isspace(*p)) {
 924                                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 925                                 state = ' ';
 926                                 continue;
 927                         }
 928                         continue;
 929                 case '"':
 930                 case '\'':
 931                         if (*p == state) {
 932                                 complete_word(&result, nextword+1, p, &result_capacity, &result_count);
 933                                 state = ' ';
 934                                 continue;
 935                         }
 936                         continue;
 937                 }
 938         }
 939
 940         if (state == 'a')
 941                 complete_word(&result, nextword, p, &result_capacity, &result_count);
 942
 943         return realloc(result, (result_count + 1) * sizeof(char *));
 944 }
 945
 946 char **lxc_string_split_and_trim(const char *string, char _sep)
 947 {
 948         char *token, *str;
 949         char sep[2] = { _sep, '\0' };
 950         char **result = NULL;
 951         size_t result_capacity = 0;
 952         size_t result_count = 0;
 953         int r, saved_errno;
 954         size_t i = 0;
 955         size_t len;
 956
 957         if (!string)
 958                 return calloc(1, sizeof(char *));
 959
 960         len = strlen(string);
 961         str = alloca(len + 1);
 962         (void)strlcpy(str, string, len + 1);
 963
 964         lxc_iterate_parts(token, str, sep) {
 965                 while (token[0] == ' ' || token[0] == '\t')
 966                         token++;
 967
 968                 i = strlen(token);
 969                 while (i > 0 && (token[i - 1] == ' ' || token[i - 1] == '\t')) {
 970                         token[i - 1] = '\0';
 971                         i--;
 972                 }
 973
 974                 r = lxc_grow_array((void ***)&result, &result_capacity, result_count + 1, 16);
 975                 if (r < 0)
 976                         goto error_out;
 977
 978                 result[result_count] = strdup(token);
 979                 if (!result[result_count])
 980                         goto error_out;
 981
 982                 result_count++;
 983         }
 984
 985         /* if we allocated too much, reduce it */
 986         return realloc(result, (result_count + 1) * sizeof(char *));
 987
 988 error_out:
 989         saved_errno = errno;
 990         lxc_free_array((void **)result, free);
 991         errno = saved_errno;
 992         return NULL;
 993 }
 994
 995 void lxc_free_array(void **array, lxc_free_fn element_free_fn)
 996 {
 997         void **p;
 998
 999         for (p = array; p && *p; p++)
1000                 element_free_fn(*p);
1001
1002         free((void*)array);
1003 }
1004
1005 int lxc_grow_array(void ***array, size_t *capacity, size_t new_size, size_t capacity_increment)
1006 {
1007         size_t new_capacity;
1008         void **new_array;
1009
1010         /* first time around, catch some trivial mistakes of the user
1011          * only initializing one of these */
1012         if (!*array || !*capacity) {
1013                 *array = NULL;
1014                 *capacity = 0;
1015         }
1016
1017         new_capacity = *capacity;
1018         while (new_size + 1 > new_capacity)
1019                 new_capacity += capacity_increment;
1020
1021         if (new_capacity != *capacity) {
1022                 /* we have to reallocate */
1023                 new_array = realloc(*array, new_capacity * sizeof(void *));
1024                 if (!new_array)
1025                         return -1;
1026
1027                 memset(&new_array[*capacity], 0, (new_capacity - (*capacity)) * sizeof(void *));
1028                 *array = new_array;
1029                 *capacity = new_capacity;
1030         }
1031
1032         /* array has sufficient elements */
1033         return 0;
1034 }
1035
1036 size_t lxc_array_len(void **array)
1037 {
1038         void **p;
1039         size_t result = 0;
1040
1041         for (p = array; p && *p; p++)
1042                 result++;
1043
1044         return result;
1045 }
1046
1047 int lxc_write_to_file(const char *filename, const void *buf, size_t count,
1048                       bool add_newline, mode_t mode)
1049 {
1050         int fd, saved_errno;
1051         ssize_t ret;
1052
1053         fd = open(filename, O_WRONLY | O_TRUNC | O_CREAT | O_CLOEXEC, mode);
1054         if (fd < 0)
1055                 return -1;
1056
1057         ret = lxc_write_nointr(fd, buf, count);
1058         if (ret < 0)
1059                 goto out_error;
1060
1061         if ((size_t)ret != count)
1062                 goto out_error;
1063
1064         if (add_newline) {
1065                 ret = lxc_write_nointr(fd, "\n", 1);
1066                 if (ret != 1)
1067                         goto out_error;
1068         }
1069
1070         close(fd);
1071         return 0;
1072
1073 out_error:
1074         saved_errno = errno;
1075         close(fd);
1076         errno = saved_errno;
1077         return -1;
1078 }
1079
1080 int lxc_read_from_file(const char *filename, void *buf, size_t count)
1081 {
1082         int fd = -1, saved_errno;
1083         ssize_t ret;
1084
1085         fd = open(filename, O_RDONLY | O_CLOEXEC);
1086         if (fd < 0)
1087                 return -1;
1088
1089         if (!buf || !count) {
1090                 char buf2[100];
1091                 size_t count2 = 0;
1092
1093                 while ((ret = read(fd, buf2, 100)) > 0)
1094                         count2 += ret;
1095
1096                 if (ret >= 0)
1097                         ret = count2;
1098         } else {
1099                 memset(buf, 0, count);
1100                 ret = read(fd, buf, count);
1101         }
1102
1103         if (ret < 0)
1104                 SYSERROR("Read %s", filename);
1105
1106         saved_errno = errno;
1107         close(fd);
1108         errno = saved_errno;
1109         return ret;
1110 }
1111
1112 void **lxc_append_null_to_array(void **array, size_t count)
1113 {
1114         void **temp;
1115
1116         /* Append NULL to the array */
1117         if (count) {
1118                 temp = realloc(array, (count + 1) * sizeof(*array));
1119                 if (!temp) {
1120                         size_t i;
1121                         for (i = 0; i < count; i++)
1122                                 free(array[i]);
1123                         free(array);
1124                         return NULL;
1125                 }
1126
1127                 array = temp;
1128                 array[count] = NULL;
1129         }
1130
1131         return array;
1132 }
1133
1134 int randseed(bool srand_it)
1135 {
1136         /*
1137            srand pre-seed function based on /dev/urandom
1138            */
1139         unsigned int seed = time(NULL) + getpid();
1140
1141         FILE *f;
1142         f = fopen("/dev/urandom", "r");
1143         if (f) {
1144                 int ret = fread(&seed, sizeof(seed), 1, f);
1145                 if (ret != 1)
1146                         SYSDEBUG("unable to fread /dev/urandom, fallback to time+pid rand seed");
1147
1148                 fclose(f);
1149         }
1150
1151         if (srand_it)
1152                 srand(seed);
1153
1154         return seed;
1155 }
1156
1157 uid_t get_ns_uid(uid_t orig)
1158 {
1159         char *line = NULL;
1160         size_t sz = 0;
1161         uid_t nsid, hostid, range;
1162         FILE *f = fopen("/proc/self/uid_map", "r");
1163         if (!f)
1164                 return 0;
1165
1166         while (getline(&line, &sz, f) != -1) {
1167                 if (sscanf(line, "%u %u %u", &nsid, &hostid, &range) != 3)
1168                         continue;
1169
1170                 if (hostid <= orig && hostid + range > orig) {
1171                         nsid += orig - hostid;
1172                         goto found;
1173                 }
1174         }
1175
1176         nsid = 0;
1177
1178 found:
1179         fclose(f);
1180         free(line);
1181         return nsid;
1182 }
1183
1184 bool dir_exists(const char *path)
1185 {
1186         struct stat sb;
1187         int ret;
1188
1189         ret = stat(path, &sb);
1190         if (ret < 0)
1191                 /* Could be something other than eexist, just say "no". */
1192                 return false;
1193
1194         return S_ISDIR(sb.st_mode);
1195 }
1196
1197 /* Note we don't use SHA-1 here as we don't want to depend on HAVE_GNUTLS.
1198  * FNV has good anti collision properties and we're not worried
1199  * about pre-image resistance or one-way-ness, we're just trying to make
1200  * the name unique in the 108 bytes of space we have.
1201  */
1202 uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
1203 {
1204         unsigned char *bp;
1205
1206         for(bp = buf; bp < (unsigned char *)buf + len; bp++)
1207         {
1208                 /* xor the bottom with the current octet */
1209                 hval ^= (uint64_t)*bp;
1210
1211                 /* gcc optimised:
1212                  * multiply by the 64 bit FNV magic prime mod 2^64
1213                  */
1214                 hval += (hval << 1) + (hval << 4) + (hval << 5) +
1215                         (hval << 7) + (hval << 8) + (hval << 40);
1216         }
1217
1218         return hval;
1219 }
1220
1221 bool is_shared_mountpoint(const char *path)
1222 {
1223         char buf[LXC_LINELEN];
1224         FILE *f;
1225         int i;
1226         char *p, *p2;
1227
1228         f = fopen("/proc/self/mountinfo", "r");
1229         if (!f)
1230                 return 0;
1231
1232         while (fgets(buf, LXC_LINELEN, f)) {
1233                 for (p = buf, i = 0; p && i < 4; i++)
1234                         p = strchr(p + 1, ' ');
1235                 if (!p)
1236                         continue;
1237
1238                 p2 = strchr(p + 1, ' ');
1239                 if (!p2)
1240                         continue;
1241
1242                 *p2 = '\0';
1243                 if (strcmp(p + 1, path) == 0) {
1244                         /* This is the path. Is it shared? */
1245                         p = strchr(p2 + 1, ' ');
1246                         if (p && strstr(p, "shared:")) {
1247                                 fclose(f);
1248                                 return true;
1249                         }
1250                 }
1251         }
1252
1253         fclose(f);
1254         return false;
1255 }
1256
1257 /*
1258  * Detect whether / is mounted MS_SHARED.  The only way I know of to
1259  * check that is through /proc/self/mountinfo.
1260  * I'm only checking for /.  If the container rootfs or mount location
1261  * is MS_SHARED, but not '/', then you're out of luck - figuring that
1262  * out would be too much work to be worth it.
1263  */
1264 int detect_shared_rootfs(void)
1265 {
1266         if (is_shared_mountpoint("/"))
1267                 return 1;
1268         return 0;
1269 }
1270
1271 bool switch_to_ns(pid_t pid, const char *ns) {
1272         int fd, ret;
1273         char nspath[MAXPATHLEN];
1274
1275         /* Switch to new ns */
1276         ret = snprintf(nspath, MAXPATHLEN, "/proc/%d/ns/%s", pid, ns);
1277         if (ret < 0 || ret >= MAXPATHLEN)
1278                 return false;
1279
1280         fd = open(nspath, O_RDONLY);
1281         if (fd < 0) {
1282                 SYSERROR("Failed to open %s", nspath);
1283                 return false;
1284         }
1285
1286         ret = setns(fd, 0);
1287         if (ret) {
1288                 SYSERROR("Failed to set process %d to %s of %d.", pid, ns, fd);
1289                 close(fd);
1290                 return false;
1291         }
1292
1293         close(fd);
1294         return true;
1295 }
1296
1297 /*
1298  * looking at fs/proc_namespace.c, it appears we can
1299  * actually expect the rootfs entry to very specifically contain
1300  * " - rootfs rootfs "
1301  * IIUC, so long as we've chrooted so that rootfs is not our root,
1302  * the rootfs entry should always be skipped in mountinfo contents.
1303  */
1304 bool detect_ramfs_rootfs(void)
1305 {
1306         FILE *f;
1307         char *p, *p2;
1308         char *line = NULL;
1309         size_t len = 0;
1310         int i;
1311
1312         f = fopen("/proc/self/mountinfo", "r");
1313         if (!f)
1314                 return false;
1315
1316         while (getline(&line, &len, f) != -1) {
1317                 for (p = line, i = 0; p && i < 4; i++)
1318                         p = strchr(p + 1, ' ');
1319                 if (!p)
1320                         continue;
1321
1322                 p2 = strchr(p + 1, ' ');
1323                 if (!p2)
1324                         continue;
1325
1326                 *p2 = '\0';
1327                 if (strcmp(p + 1, "/") == 0) {
1328                         /* This is '/'. Is it the ramfs? */
1329                         p = strchr(p2 + 1, '-');
1330                         if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) {
1331                                 free(line);
1332                                 fclose(f);
1333                                 return true;
1334                         }
1335                 }
1336         }
1337
1338         free(line);
1339         fclose(f);
1340         return false;
1341 }
1342
1343 char *on_path(const char *cmd, const char *rootfs) {
1344         char *entry = NULL, *path = NULL;
1345         char cmdpath[MAXPATHLEN];
1346         int ret;
1347
1348         path = getenv("PATH");
1349         if (!path)
1350                 return NULL;
1351
1352         path = strdup(path);
1353         if (!path)
1354                 return NULL;
1355
1356         lxc_iterate_parts(entry, path, ":") {
1357                 if (rootfs)
1358                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s/%s", rootfs, entry, cmd);
1359                 else
1360                         ret = snprintf(cmdpath, MAXPATHLEN, "%s/%s", entry, cmd);
1361                 if (ret < 0 || ret >= MAXPATHLEN)
1362                         continue;
1363
1364                 if (access(cmdpath, X_OK) == 0) {
1365                         free(path);
1366                         return strdup(cmdpath);
1367                 }
1368         }
1369
1370         free(path);
1371         return NULL;
1372 }
1373
1374 bool file_exists(const char *f)
1375 {
1376         struct stat statbuf;
1377
1378         return stat(f, &statbuf) == 0;
1379 }
1380
1381 bool cgns_supported(void)
1382 {
1383         return file_exists("/proc/self/ns/cgroup");
1384 }
1385
1386 /* historically lxc-init has been under /usr/lib/lxc and under
1387  * /usr/lib/$ARCH/lxc.  It now lives as $prefix/sbin/init.lxc.
1388  */
1389 char *choose_init(const char *rootfs)
1390 {
1391         char *retv = NULL;
1392         const char *empty = "",
1393                    *tmp;
1394         int ret, env_set = 0;
1395
1396         if (!getenv("PATH")) {
1397                 if (setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 0))
1398                         SYSERROR("Failed to setenv");
1399
1400                 env_set = 1;
1401         }
1402
1403         retv = on_path("init.lxc", rootfs);
1404
1405         if (env_set) {
1406                 if (unsetenv("PATH"))
1407                         SYSERROR("Failed to unsetenv");
1408         }
1409
1410         if (retv)
1411                 return retv;
1412
1413         retv = malloc(PATH_MAX);
1414         if (!retv)
1415                 return NULL;
1416
1417         if (rootfs)
1418                 tmp = rootfs;
1419         else
1420                 tmp = empty;
1421
1422         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, SBINDIR, "/init.lxc");
1423         if (ret < 0 || ret >= PATH_MAX) {
1424                 ERROR("pathname too long");
1425                 goto out1;
1426         }
1427
1428         if (access(retv, X_OK) == 0)
1429                 return retv;
1430
1431         ret = snprintf(retv, PATH_MAX, "%s/%s/%s", tmp, LXCINITDIR, "/lxc/lxc-init");
1432         if (ret < 0 || ret >= PATH_MAX) {
1433                 ERROR("pathname too long");
1434                 goto out1;
1435         }
1436
1437         if (access(retv, X_OK) == 0)
1438                 return retv;
1439
1440         ret = snprintf(retv, PATH_MAX, "%s/usr/lib/lxc/lxc-init", tmp);
1441         if (ret < 0 || ret >= PATH_MAX) {
1442                 ERROR("pathname too long");
1443                 goto out1;
1444         }
1445
1446         if (access(retv, X_OK) == 0)
1447                 return retv;
1448
1449         ret = snprintf(retv, PATH_MAX, "%s/sbin/lxc-init", tmp);
1450         if (ret < 0 || ret >= PATH_MAX) {
1451                 ERROR("pathname too long");
1452                 goto out1;
1453         }
1454
1455         if (access(retv, X_OK) == 0)
1456                 return retv;
1457
1458         /*
1459          * Last resort, look for the statically compiled init.lxc which we
1460          * hopefully bind-mounted in.
1461          * If we are called during container setup, and we get to this point,
1462          * then the init.lxc.static from the host will need to be bind-mounted
1463          * in.  So we return NULL here to indicate that.
1464          */
1465         if (rootfs)
1466                 goto out1;
1467
1468         ret = snprintf(retv, PATH_MAX, "/init.lxc.static");
1469         if (ret < 0 || ret >= PATH_MAX) {
1470                 WARN("Nonsense - name /lxc.init.static too long");
1471                 goto out1;
1472         }
1473
1474         if (access(retv, X_OK) == 0)
1475                 return retv;
1476
1477 out1:
1478         free(retv);
1479         return NULL;
1480 }
1481
1482 int print_to_file(const char *file, const char *content)
1483 {
1484         FILE *f;
1485         int ret = 0;
1486
1487         f = fopen(file, "w");
1488         if (!f)
1489                 return -1;
1490
1491         if (fprintf(f, "%s", content) != strlen(content))
1492                 ret = -1;
1493
1494         fclose(f);
1495         return ret;
1496 }
1497
1498 int is_dir(const char *path)
1499 {
1500         struct stat statbuf;
1501         int ret;
1502
1503         ret = stat(path, &statbuf);
1504         if (ret == 0 && S_ISDIR(statbuf.st_mode))
1505                 return 1;
1506
1507         return 0;
1508 }
1509
1510 /*
1511  * Given the '-t' template option to lxc-create, figure out what to
1512  * do.  If the template is a full executable path, use that.  If it
1513  * is something like 'sshd', then return $templatepath/lxc-sshd.
1514  * On success return the template, on error return NULL.
1515  */
1516 char *get_template_path(const char *t)
1517 {
1518         int ret, len;
1519         char *tpath;
1520
1521         if (t[0] == '/' && access(t, X_OK) == 0) {
1522                 tpath = strdup(t);
1523                 return tpath;
1524         }
1525
1526         len = strlen(LXCTEMPLATEDIR) + strlen(t) + strlen("/lxc-") + 1;
1527
1528         tpath = malloc(len);
1529         if (!tpath)
1530                 return NULL;
1531
1532         ret = snprintf(tpath, len, "%s/lxc-%s", LXCTEMPLATEDIR, t);
1533         if (ret < 0 || ret >= len) {
1534                 free(tpath);
1535                 return NULL;
1536         }
1537
1538         if (access(tpath, X_OK) < 0) {
1539                 SYSERROR("bad template: %s", t);
1540                 free(tpath);
1541                 return NULL;
1542         }
1543
1544         return tpath;
1545 }
1546
1547 /*
1548  * @path:    a pathname where / replaced with '\0'.
1549  * @offsetp: pointer to int showing which path segment was last seen.
1550  *           Updated on return to reflect the next segment.
1551  * @fulllen: full original path length.
1552  * Returns a pointer to the next path segment, or NULL if done.
1553  */
1554 static char *get_nextpath(char *path, int *offsetp, int fulllen)
1555 {
1556         int offset = *offsetp;
1557
1558         if (offset >= fulllen)
1559                 return NULL;
1560
1561         while (path[offset] != '\0' && offset < fulllen)
1562                 offset++;
1563
1564         while (path[offset] == '\0' && offset < fulllen)
1565                 offset++;
1566
1567         *offsetp = offset;
1568         return (offset < fulllen) ? &path[offset] : NULL;
1569 }
1570
1571 /*
1572  * Check that @subdir is a subdir of @dir.  @len is the length of
1573  * @dir (to avoid having to recalculate it).
1574  */
1575 static bool is_subdir(const char *subdir, const char *dir, size_t len)
1576 {
1577         size_t subdirlen = strlen(subdir);
1578
1579         if (subdirlen < len)
1580                 return false;
1581
1582         if (strncmp(subdir, dir, len) != 0)
1583                 return false;
1584
1585         if (dir[len-1] == '/')
1586                 return true;
1587
1588         if (subdir[len] == '/' || subdirlen == len)
1589                 return true;
1590
1591         return false;
1592 }
1593
1594 /*
1595  * Check if the open fd is a symlink.  Return -ELOOP if it is.  Return
1596  * -ENOENT if we couldn't fstat.  Return 0 if the fd is ok.
1597  */
1598 static int check_symlink(int fd)
1599 {
1600         struct stat sb;
1601         int ret;
1602
1603         ret = fstat(fd, &sb);
1604         if (ret < 0)
1605                 return -ENOENT;
1606
1607         if (S_ISLNK(sb.st_mode))
1608                 return -ELOOP;
1609
1610         return 0;
1611 }
1612
1613 /*
1614  * Open a file or directory, provided that it contains no symlinks.
1615  *
1616  * CAVEAT: This function must not be used for other purposes than container
1617  * setup before executing the container's init
1618  */
1619 static int open_if_safe(int dirfd, const char *nextpath)
1620 {
1621         int newfd = openat(dirfd, nextpath, O_RDONLY | O_NOFOLLOW);
1622         if (newfd >= 0) /* Was not a symlink, all good. */
1623                 return newfd;
1624
1625         if (errno == ELOOP)
1626                 return newfd;
1627
1628         if (errno == EPERM || errno == EACCES) {
1629                 /* We're not root (cause we got EPERM) so try opening with
1630                  * O_PATH.
1631                  */
1632                 newfd = openat(dirfd, nextpath, O_PATH | O_NOFOLLOW);
1633                 if (newfd >= 0) {
1634                         /* O_PATH will return an fd for symlinks. We know
1635                          * nextpath wasn't a symlink at last openat, so if fd is
1636                          * now a link, then something * fishy is going on.
1637                          */
1638                         int ret = check_symlink(newfd);
1639                         if (ret < 0) {
1640                                 close(newfd);
1641                                 newfd = ret;
1642                         }
1643                 }
1644         }
1645
1646         return newfd;
1647 }
1648
1649 /*
1650  * Open a path intending for mounting, ensuring that the final path
1651  * is inside the container's rootfs.
1652  *
1653  * CAVEAT: This function must not be used for other purposes than container
1654  * setup before executing the container's init
1655  *
1656  * @target: path to be opened
1657  * @prefix_skip: a part of @target in which to ignore symbolic links.  This
1658  * would be the container's rootfs.
1659  *
1660  * Return an open fd for the path, or <0 on error.
1661  */
1662 static int open_without_symlink(const char *target, const char *prefix_skip)
1663 {
1664         int curlen = 0, dirfd, fulllen, i;
1665         char *dup = NULL;
1666
1667         fulllen = strlen(target);
1668
1669         /* make sure prefix-skip makes sense */
1670         if (prefix_skip && strlen(prefix_skip) > 0) {
1671                 curlen = strlen(prefix_skip);
1672                 if (!is_subdir(target, prefix_skip, curlen)) {
1673                         ERROR("WHOA there - target '%s' didn't start with prefix '%s'",
1674                                 target, prefix_skip);
1675                         return -EINVAL;
1676                 }
1677
1678                 /*
1679                  * get_nextpath() expects the curlen argument to be
1680                  * on a  (turned into \0) / or before it, so decrement
1681                  * curlen to make sure that happens
1682                  */
1683                 if (curlen)
1684                         curlen--;
1685         } else {
1686                 prefix_skip = "/";
1687                 curlen = 0;
1688         }
1689
1690         /* Make a copy of target which we can hack up, and tokenize it */
1691         if ((dup = strdup(target)) == NULL) {
1692                 SYSERROR("Out of memory checking for symbolic link");
1693                 return -ENOMEM;
1694         }
1695
1696         for (i = 0; i < fulllen; i++) {
1697                 if (dup[i] == '/')
1698                         dup[i] = '\0';
1699         }
1700
1701         dirfd = open(prefix_skip, O_RDONLY);
1702         if (dirfd < 0)
1703                 goto out;
1704
1705         while (1) {
1706                 int newfd, saved_errno;
1707                 char *nextpath;
1708
1709                 if ((nextpath = get_nextpath(dup, &curlen, fulllen)) == NULL)
1710                         goto out;
1711
1712                 newfd = open_if_safe(dirfd, nextpath);
1713                 saved_errno = errno;
1714                 close(dirfd);
1715
1716                 dirfd = newfd;
1717                 if (newfd < 0) {
1718                         errno = saved_errno;
1719                         if (errno == ELOOP)
1720                                 SYSERROR("%s in %s was a symbolic link!", nextpath, target);
1721
1722                         goto out;
1723                 }
1724         }
1725
1726 out:
1727         free(dup);
1728         return dirfd;
1729 }
1730
1731 /*
1732  * Safely mount a path into a container, ensuring that the mount target
1733  * is under the container's @rootfs.  (If @rootfs is NULL, then the container
1734  * uses the host's /)
1735  *
1736  * CAVEAT: This function must not be used for other purposes than container
1737  * setup before executing the container's init
1738  */
1739 int safe_mount(const char *src, const char *dest, const char *fstype,
1740                 unsigned long flags, const void *data, const char *rootfs)
1741 {
1742         int destfd, ret, saved_errno;
1743         /* Only needs enough for /proc/self/fd/<fd>. */
1744         char srcbuf[50], destbuf[50];
1745         int srcfd = -1;
1746         const char *mntsrc = src;
1747
1748         if (!rootfs)
1749                 rootfs = "";
1750
1751         /* todo - allow symlinks for relative paths if 'allowsymlinks' option is passed */
1752         if (flags & MS_BIND && src && src[0] != '/') {
1753                 INFO("this is a relative bind mount");
1754
1755                 srcfd = open_without_symlink(src, NULL);
1756                 if (srcfd < 0)
1757                         return srcfd;
1758
1759                 ret = snprintf(srcbuf, 50, "/proc/self/fd/%d", srcfd);
1760                 if (ret < 0 || ret > 50) {
1761                         close(srcfd);
1762                         ERROR("Out of memory");
1763                         return -EINVAL;
1764                 }
1765                 mntsrc = srcbuf;
1766         }
1767
1768         destfd = open_without_symlink(dest, rootfs);
1769         if (destfd < 0) {
1770                 if (srcfd != -1) {
1771                         saved_errno = errno;
1772                         close(srcfd);
1773                         errno = saved_errno;
1774                 }
1775
1776                 return destfd;
1777         }
1778
1779         ret = snprintf(destbuf, 50, "/proc/self/fd/%d", destfd);
1780         if (ret < 0 || ret > 50) {
1781                 if (srcfd != -1)
1782                         close(srcfd);
1783
1784                 close(destfd);
1785                 ERROR("Out of memory");
1786                 return -EINVAL;
1787         }
1788
1789         ret = mount(mntsrc, destbuf, fstype, flags, data);
1790         saved_errno = errno;
1791         if (srcfd != -1)
1792                 close(srcfd);
1793
1794         close(destfd);
1795         if (ret < 0) {
1796                 errno = saved_errno;
1797                 SYSERROR("Failed to mount %s onto %s", src ? src : "(null)", dest);
1798                 return ret;
1799         }
1800
1801         return 0;
1802 }
1803
1804 /*
1805  * Mount a proc under @rootfs if proc self points to a pid other than
1806  * my own.  This is needed to have a known-good proc mount for setting
1807  * up LSMs both at container startup and attach.
1808  *
1809  * @rootfs : the rootfs where proc should be mounted
1810  *
1811  * Returns < 0 on failure, 0 if the correct proc was already mounted
1812  * and 1 if a new proc was mounted.
1813  *
1814  * NOTE: not to be called from inside the container namespace!
1815  */
1816 int lxc_mount_proc_if_needed(const char *rootfs)
1817 {
1818         char path[MAXPATHLEN];
1819         int link_to_pid, linklen, mypid, ret;
1820         char link[LXC_NUMSTRLEN64] = {0};
1821
1822         ret = snprintf(path, MAXPATHLEN, "%s/proc/self", rootfs);
1823         if (ret < 0 || ret >= MAXPATHLEN) {
1824                 SYSERROR("proc path name too long");
1825                 return -1;
1826         }
1827
1828         linklen = readlink(path, link, LXC_NUMSTRLEN64);
1829
1830         ret = snprintf(path, MAXPATHLEN, "%s/proc", rootfs);
1831         if (ret < 0 || ret >= MAXPATHLEN) {
1832                 SYSERROR("proc path name too long");
1833                 return -1;
1834         }
1835
1836         /* /proc not mounted */
1837         if (linklen < 0) {
1838                 if (mkdir(path, 0755) && errno != EEXIST)
1839                         return -1;
1840
1841                 goto domount;
1842         } else if (linklen >= LXC_NUMSTRLEN64) {
1843                 link[linklen - 1] = '\0';
1844                 ERROR("readlink returned truncated content: \"%s\"", link);
1845                 return -1;
1846         }
1847
1848         mypid = lxc_raw_getpid();
1849         INFO("I am %d, /proc/self points to \"%s\"", mypid, link);
1850
1851         if (lxc_safe_int(link, &link_to_pid) < 0)
1852                 return -1;
1853
1854         /* correct procfs is already mounted */
1855         if (link_to_pid == mypid)
1856                 return 0;
1857
1858         ret = umount2(path, MNT_DETACH);
1859         if (ret < 0)
1860                 WARN("failed to umount \"%s\" with MNT_DETACH", path);
1861
1862 domount:
1863         /* rootfs is NULL */
1864         if (!strcmp(rootfs, ""))
1865                 ret = mount("proc", path, "proc", 0, NULL);
1866         else
1867                 ret = safe_mount("proc", path, "proc", 0, NULL, rootfs);
1868         if (ret < 0)
1869                 return -1;
1870
1871         INFO("mounted /proc in container for security transition");
1872         return 1;
1873 }
1874
1875 int open_devnull(void)
1876 {
1877         int fd = open("/dev/null", O_RDWR);
1878
1879         if (fd < 0)
1880                 SYSERROR("Can't open /dev/null");
1881
1882         return fd;
1883 }
1884
1885 int set_stdfds(int fd)
1886 {
1887         int ret;
1888
1889         if (fd < 0)
1890                 return -1;
1891
1892         ret = dup2(fd, STDIN_FILENO);
1893         if (ret < 0)
1894                 return -1;
1895
1896         ret = dup2(fd, STDOUT_FILENO);
1897         if (ret < 0)
1898                 return -1;
1899
1900         ret = dup2(fd, STDERR_FILENO);
1901         if (ret < 0)
1902                 return -1;
1903
1904         return 0;
1905 }
1906
1907 int null_stdfds(void)
1908 {
1909         int ret = -1;
1910         int fd;
1911
1912         fd = open_devnull();
1913         if (fd >= 0) {
1914                 ret = set_stdfds(fd);
1915                 close(fd);
1916         }
1917
1918         return ret;
1919 }
1920
1921 /*
1922  * Return the number of lines in file @fn, or -1 on error
1923  */
1924 int lxc_count_file_lines(const char *fn)
1925 {
1926         FILE *f;
1927         char *line = NULL;
1928         size_t sz = 0;
1929         int n = 0;
1930
1931         f = fopen_cloexec(fn, "r");
1932         if (!f)
1933                 return -1;
1934
1935         while (getline(&line, &sz, f) != -1) {
1936                 n++;
1937         }
1938
1939         free(line);
1940         fclose(f);
1941         return n;
1942 }
1943
1944 /* Check whether a signal is blocked by a process. */
1945 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
1946 #define __PROC_STATUS_LEN (6 + (LXC_NUMSTRLEN64) + 7 + 1)
1947 bool task_blocks_signal(pid_t pid, int signal)
1948 {
1949         int ret;
1950         char status[__PROC_STATUS_LEN];
1951         FILE *f;
1952         uint64_t sigblk = 0, one = 1;
1953         size_t n = 0;
1954         bool bret = false;
1955         char *line = NULL;
1956
1957         ret = snprintf(status, __PROC_STATUS_LEN, "/proc/%d/status", pid);
1958         if (ret < 0 || ret >= __PROC_STATUS_LEN)
1959                 return bret;
1960
1961         f = fopen(status, "r");
1962         if (!f)
1963                 return bret;
1964
1965         while (getline(&line, &n, f) != -1) {
1966                 char *numstr;
1967
1968                 if (strncmp(line, "SigBlk:", 7))
1969                         continue;
1970
1971                 numstr = lxc_trim_whitespace_in_place(line + 7);
1972                 ret = lxc_safe_uint64(numstr, &sigblk, 16);
1973                 if (ret < 0)
1974                         goto out;
1975
1976                 break;
1977         }
1978
1979         if (sigblk & (one << (signal - 1)))
1980                 bret = true;
1981
1982 out:
1983         free(line);
1984         fclose(f);
1985         return bret;
1986 }
1987
1988 static int lxc_append_null_to_list(void ***list)
1989 {
1990         int newentry = 0;
1991         void **tmp;
1992
1993         if (*list)
1994                 for (; (*list)[newentry]; newentry++) {
1995                         ;
1996                 }
1997
1998         tmp = realloc(*list, (newentry + 2) * sizeof(void **));
1999         if (!tmp)
2000                 return -1;
2001
2002         *list = tmp;
2003         (*list)[newentry + 1] = NULL;
2004
2005         return newentry;
2006 }
2007
2008 int lxc_append_string(char ***list, char *entry)
2009 {
2010         char *copy;
2011         int newentry;
2012
2013         newentry = lxc_append_null_to_list((void ***)list);
2014         if (newentry < 0)
2015                 return -1;
2016
2017         copy = strdup(entry);
2018         if (!copy)
2019                 return -1;
2020
2021         (*list)[newentry] = copy;
2022
2023         return 0;
2024 }
2025
2026 int lxc_preserve_ns(const int pid, const char *ns)
2027 {
2028         int ret;
2029 /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */
2030 #define __NS_PATH_LEN 50
2031         char path[__NS_PATH_LEN];
2032
2033         /* This way we can use this function to also check whether namespaces
2034          * are supported by the kernel by passing in the NULL or the empty
2035          * string.
2036          */
2037         ret = snprintf(path, __NS_PATH_LEN, "/proc/%d/ns%s%s", pid,
2038                        !ns || strcmp(ns, "") == 0 ? "" : "/",
2039                        !ns || strcmp(ns, "") == 0 ? "" : ns);
2040         if (ret < 0 || (size_t)ret >= __NS_PATH_LEN) {
2041                 errno = EFBIG;
2042                 return -1;
2043         }
2044
2045         return open(path, O_RDONLY | O_CLOEXEC);
2046 }
2047
2048 int lxc_safe_uint(const char *numstr, unsigned int *converted)
2049 {
2050         char *err = NULL;
2051         unsigned long int uli;
2052
2053         while (isspace(*numstr))
2054                 numstr++;
2055
2056         if (*numstr == '-')
2057                 return -EINVAL;
2058
2059         errno = 0;
2060         uli = strtoul(numstr, &err, 0);
2061         if (errno == ERANGE && uli == ULONG_MAX)
2062                 return -ERANGE;
2063
2064         if (err == numstr || *err != '\0')
2065                 return -EINVAL;
2066
2067         if (uli > UINT_MAX)
2068                 return -ERANGE;
2069
2070         *converted = (unsigned int)uli;
2071         return 0;
2072 }
2073
2074 int lxc_safe_ulong(const char *numstr, unsigned long *converted)
2075 {
2076         char *err = NULL;
2077         unsigned long int uli;
2078
2079         while (isspace(*numstr))
2080                 numstr++;
2081
2082         if (*numstr == '-')
2083                 return -EINVAL;
2084
2085         errno = 0;
2086         uli = strtoul(numstr, &err, 0);
2087         if (errno == ERANGE && uli == ULONG_MAX)
2088                 return -ERANGE;
2089
2090         if (err == numstr || *err != '\0')
2091                 return -EINVAL;
2092
2093         *converted = uli;
2094         return 0;
2095 }
2096
2097 int lxc_safe_uint64(const char *numstr, uint64_t *converted, int base)
2098 {
2099         char *err = NULL;
2100         uint64_t u;
2101
2102         while (isspace(*numstr))
2103                 numstr++;
2104
2105         if (*numstr == '-')
2106                 return -EINVAL;
2107
2108         errno = 0;
2109         u = strtoull(numstr, &err, base);
2110         if (errno == ERANGE && u == ULLONG_MAX)
2111                 return -ERANGE;
2112
2113         if (err == numstr || *err != '\0')
2114                 return -EINVAL;
2115
2116         *converted = u;
2117         return 0;
2118 }
2119
2120 int lxc_safe_int(const char *numstr, int *converted)
2121 {
2122         char *err = NULL;
2123         signed long int sli;
2124
2125         errno = 0;
2126         sli = strtol(numstr, &err, 0);
2127         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
2128                 return -ERANGE;
2129
2130         if (errno != 0 && sli == 0)
2131                 return -EINVAL;
2132
2133         if (err == numstr || *err != '\0')
2134                 return -EINVAL;
2135
2136         if (sli > INT_MAX || sli < INT_MIN)
2137                 return -ERANGE;
2138
2139         *converted = (int)sli;
2140         return 0;
2141 }
2142
2143 int lxc_safe_long(const char *numstr, long int *converted)
2144 {
2145         char *err = NULL;
2146         signed long int sli;
2147
2148         errno = 0;
2149         sli = strtol(numstr, &err, 0);
2150         if (errno == ERANGE && (sli == LONG_MAX || sli == LONG_MIN))
2151                 return -ERANGE;
2152
2153         if (errno != 0 && sli == 0)
2154                 return -EINVAL;
2155
2156         if (err == numstr || *err != '\0')
2157                 return -EINVAL;
2158
2159         *converted = sli;
2160         return 0;
2161 }
2162
2163 int lxc_safe_long_long(const char *numstr, long long int *converted)
2164 {
2165         char *err = NULL;
2166         signed long long int sli;
2167
2168         errno = 0;
2169         sli = strtoll(numstr, &err, 0);
2170         if (errno == ERANGE && (sli == LLONG_MAX || sli == LLONG_MIN))
2171                 return -ERANGE;
2172
2173         if (errno != 0 && sli == 0)
2174                 return -EINVAL;
2175
2176         if (err == numstr || *err != '\0')
2177                 return -EINVAL;
2178
2179         *converted = sli;
2180         return 0;
2181 }
2182
2183 int lxc_switch_uid_gid(uid_t uid, gid_t gid)
2184 {
2185         if (setgid(gid) < 0) {
2186                 SYSERROR("Failed to switch to gid %d.", gid);
2187                 return -errno;
2188         }
2189         NOTICE("Switched to gid %d.", gid);
2190
2191         if (setuid(uid) < 0) {
2192                 SYSERROR("Failed to switch to uid %d.", uid);
2193                 return -errno;
2194         }
2195         NOTICE("Switched to uid %d.", uid);
2196
2197         return 0;
2198 }
2199
2200 /* Simple covenience function which enables uniform logging. */
2201 int lxc_setgroups(int size, gid_t list[])
2202 {
2203         if (setgroups(size, list) < 0) {
2204                 SYSERROR("Failed to setgroups().");
2205                 return -errno;
2206         }
2207         NOTICE("Dropped additional groups.");
2208
2209         return 0;
2210 }
2211
2212 static int lxc_get_unused_loop_dev_legacy(char *loop_name)
2213 {
2214         struct dirent *dp;
2215         struct loop_info64 lo64;
2216         DIR *dir;
2217         int dfd = -1, fd = -1, ret = -1;
2218
2219         dir = opendir("/dev");
2220         if (!dir)
2221                 return -1;
2222
2223         while ((dp = readdir(dir))) {
2224                 if (strncmp(dp->d_name, "loop", 4) != 0)
2225                         continue;
2226
2227                 dfd = dirfd(dir);
2228                 if (dfd < 0)
2229                         continue;
2230
2231                 fd = openat(dfd, dp->d_name, O_RDWR);
2232                 if (fd < 0)
2233                         continue;
2234
2235                 ret = ioctl(fd, LOOP_GET_STATUS64, &lo64);
2236                 if (ret < 0) {
2237                         if (ioctl(fd, LOOP_GET_STATUS64, &lo64) == 0 ||
2238                             errno != ENXIO) {
2239                                 close(fd);
2240                                 fd = -1;
2241                                 continue;
2242                         }
2243                 }
2244
2245                 ret = snprintf(loop_name, LO_NAME_SIZE, "/dev/%s", dp->d_name);
2246                 if (ret < 0 || ret >= LO_NAME_SIZE) {
2247                         close(fd);
2248                         fd = -1;
2249                         continue;
2250                 }
2251
2252                 break;
2253         }
2254
2255         closedir(dir);
2256
2257         if (fd < 0)
2258                 return -1;
2259
2260         return fd;
2261 }
2262
2263 static int lxc_get_unused_loop_dev(char *name_loop)
2264 {
2265         int loop_nr, ret;
2266         int fd_ctl = -1, fd_tmp = -1;
2267
2268         fd_ctl = open("/dev/loop-control", O_RDWR | O_CLOEXEC);
2269         if (fd_ctl < 0)
2270                 return -ENODEV;
2271
2272         loop_nr = ioctl(fd_ctl, LOOP_CTL_GET_FREE);
2273         if (loop_nr < 0)
2274                 goto on_error;
2275
2276         ret = snprintf(name_loop, LO_NAME_SIZE, "/dev/loop%d", loop_nr);
2277         if (ret < 0 || ret >= LO_NAME_SIZE)
2278                 goto on_error;
2279
2280         fd_tmp = open(name_loop, O_RDWR | O_CLOEXEC);
2281         if (fd_tmp < 0)
2282                 goto on_error;
2283
2284 on_error:
2285         close(fd_ctl);
2286         return fd_tmp;
2287 }
2288
2289 int lxc_prepare_loop_dev(const char *source, char *loop_dev, int flags)
2290 {
2291         int ret;
2292         struct loop_info64 lo64;
2293         int fd_img = -1, fret = -1, fd_loop = -1;
2294
2295         fd_loop = lxc_get_unused_loop_dev(loop_dev);
2296         if (fd_loop < 0) {
2297                 if (fd_loop == -ENODEV)
2298                         fd_loop = lxc_get_unused_loop_dev_legacy(loop_dev);
2299                 else
2300                         goto on_error;
2301         }
2302
2303         fd_img = open(source, O_RDWR | O_CLOEXEC);
2304         if (fd_img < 0)
2305                 goto on_error;
2306
2307         ret = ioctl(fd_loop, LOOP_SET_FD, fd_img);
2308         if (ret < 0)
2309                 goto on_error;
2310
2311         memset(&lo64, 0, sizeof(lo64));
2312         lo64.lo_flags = flags;
2313
2314         ret = ioctl(fd_loop, LOOP_SET_STATUS64, &lo64);
2315         if (ret < 0)
2316                 goto on_error;
2317
2318         fret = 0;
2319
2320 on_error:
2321         if (fd_img >= 0)
2322                 close(fd_img);
2323
2324         if (fret < 0 && fd_loop >= 0) {
2325                 close(fd_loop);
2326                 fd_loop = -1;
2327         }
2328
2329         return fd_loop;
2330 }
2331
2332 int lxc_unstack_mountpoint(const char *path, bool lazy)
2333 {
2334         int ret;
2335         int umounts = 0;
2336
2337 pop_stack:
2338         ret = umount2(path, lazy ? MNT_DETACH : 0);
2339         if (ret < 0) {
2340                 /* We consider anything else than EINVAL deadly to prevent going
2341                  * into an infinite loop. (The other alternative is constantly
2342                  * parsing /proc/self/mountinfo which is yucky and probably
2343                  * racy.)
2344                  */
2345                 if (errno != EINVAL)
2346                         return -errno;
2347         } else {
2348                 /* Just stop counting when this happens. That'd just be so
2349                  * stupid that we won't even bother trying to report back the
2350                  * correct value anymore.
2351                  */
2352                 if (umounts != INT_MAX)
2353                         umounts++;
2354
2355                 /* We succeeded in umounting. Make sure that there's no other
2356                  * mountpoint stacked underneath.
2357                  */
2358                 goto pop_stack;
2359         }
2360
2361         return umounts;
2362 }
2363
2364 int run_command(char *buf, size_t buf_size, int (*child_fn)(void *), void *args)
2365 {
2366         pid_t child;
2367         int ret, fret, pipefd[2];
2368         ssize_t bytes;
2369
2370         /* Make sure our callers do not receive uninitialized memory. */
2371         if (buf_size > 0 && buf)
2372                 buf[0] = '\0';
2373
2374         if (pipe(pipefd) < 0) {
2375                 SYSERROR("failed to create pipe");
2376                 return -1;
2377         }
2378
2379         child = lxc_raw_clone(0);
2380         if (child < 0) {
2381                 close(pipefd[0]);
2382                 close(pipefd[1]);
2383                 SYSERROR("failed to create new process");
2384                 return -1;
2385         }
2386
2387         if (child == 0) {
2388                 /* Close the read-end of the pipe. */
2389                 close(pipefd[0]);
2390
2391                 /* Redirect std{err,out} to write-end of the
2392                  * pipe.
2393                  */
2394                 ret = dup2(pipefd[1], STDOUT_FILENO);
2395                 if (ret >= 0)
2396                         ret = dup2(pipefd[1], STDERR_FILENO);
2397
2398                 /* Close the write-end of the pipe. */
2399                 close(pipefd[1]);
2400
2401                 if (ret < 0) {
2402                         SYSERROR("failed to duplicate std{err,out} file descriptor");
2403                         _exit(EXIT_FAILURE);
2404                 }
2405
2406                 /* Does not return. */
2407                 child_fn(args);
2408                 ERROR("failed to exec command");
2409                 _exit(EXIT_FAILURE);
2410         }
2411
2412         /* close the write-end of the pipe */
2413         close(pipefd[1]);
2414
2415         if (buf && buf_size > 0) {
2416                 bytes = read(pipefd[0], buf, buf_size - 1);
2417                 if (bytes > 0)
2418                         buf[bytes - 1] = '\0';
2419         }
2420
2421         fret = wait_for_pid(child);
2422         /* close the read-end of the pipe */
2423         close(pipefd[0]);
2424
2425         return fret;
2426 }
2427
2428 char *must_concat(const char *first, ...)
2429 {
2430         va_list args;
2431         char *cur, *dest;
2432         size_t cur_len, it_len;
2433
2434         dest = must_copy_string(first);
2435         cur_len = it_len = strlen(first);
2436
2437         va_start(args, first);
2438         while ((cur = va_arg(args, char *)) != NULL) {
2439                 it_len = strlen(cur);
2440
2441                 dest = must_realloc(dest, cur_len + it_len + 1);
2442
2443                 (void)memcpy(dest + cur_len, cur, it_len);
2444                 cur_len += it_len;
2445         }
2446         va_end(args);
2447
2448         dest[cur_len] = 0;
2449         return dest;
2450 }
2451
2452 char *must_make_path(const char *first, ...)
2453 {
2454         va_list args;
2455         char *cur, *dest;
2456         size_t full_len = strlen(first);
2457         size_t buf_len;
2458
2459         dest = must_copy_string(first);
2460
2461         va_start(args, first);
2462         while ((cur = va_arg(args, char *)) != NULL) {
2463                 full_len += strlen(cur);
2464                 if (cur[0] != '/')
2465                         full_len++;
2466
2467                 buf_len = full_len + 1;
2468                 dest = must_realloc(dest, buf_len);
2469
2470                 if (cur[0] != '/')
2471                         (void)strlcat(dest, "/", buf_len);
2472                 (void)strlcat(dest, cur, buf_len);
2473         }
2474         va_end(args);
2475
2476         return dest;
2477 }
2478
2479 char *must_append_path(char *first, ...)
2480 {
2481         char *cur;
2482         size_t full_len;
2483         va_list args;
2484         char *dest = first;
2485         size_t buf_len;
2486
2487         full_len = strlen(first);
2488         va_start(args, first);
2489         while ((cur = va_arg(args, char *)) != NULL) {
2490                 full_len += strlen(cur);
2491                 if (cur[0] != '/')
2492                         full_len++;
2493
2494                 buf_len = full_len + 1;
2495                 dest = must_realloc(dest, buf_len);
2496
2497                 if (cur[0] != '/')
2498                         (void)strlcat(dest, "/", buf_len);
2499                 (void)strlcat(dest, cur, buf_len);
2500         }
2501         va_end(args);
2502
2503         return dest;
2504 }
2505
2506 char *must_copy_string(const char *entry)
2507 {
2508         char *ret;
2509
2510         if (!entry)
2511                 return NULL;
2512
2513         do {
2514                 ret = strdup(entry);
2515         } while (!ret);
2516
2517         return ret;
2518 }
2519
2520 void *must_realloc(void *orig, size_t sz)
2521 {
2522         void *ret;
2523
2524         do {
2525                 ret = realloc(orig, sz);
2526         } while (!ret);
2527
2528         return ret;
2529 }
2530
2531 bool is_fs_type(const struct statfs *fs, fs_type_magic magic_val)
2532 {
2533         return (fs->f_type == (fs_type_magic)magic_val);
2534 }
2535
2536 bool has_fs_type(const char *path, fs_type_magic magic_val)
2537 {
2538         bool has_type;
2539         int ret;
2540         struct statfs sb;
2541
2542         ret = statfs(path, &sb);
2543         if (ret < 0)
2544                 return false;
2545
2546         has_type = is_fs_type(&sb, magic_val);
2547         if (!has_type && magic_val == RAMFS_MAGIC)
2548                 WARN("When the ramfs it a tmpfs statfs() might report tmpfs");
2549
2550         return has_type;
2551 }
2552
2553 bool lxc_nic_exists(char *nic)
2554 {
2555 #define __LXC_SYS_CLASS_NET_LEN 15 + IFNAMSIZ + 1
2556         char path[__LXC_SYS_CLASS_NET_LEN];
2557         int ret;
2558         struct stat sb;
2559
2560         if (!strcmp(nic, "none"))
2561                 return true;
2562
2563         ret = snprintf(path, __LXC_SYS_CLASS_NET_LEN, "/sys/class/net/%s", nic);
2564         if (ret < 0 || (size_t)ret >= __LXC_SYS_CLASS_NET_LEN)
2565                 return false;
2566
2567         ret = stat(path, &sb);
2568         if (ret < 0)
2569                 return false;
2570
2571         return true;
2572 }
2573
2574 int lxc_make_tmpfile(char *template, bool rm)
2575 {
2576         int fd, ret;
2577         mode_t msk;
2578
2579         msk = umask(0022);
2580         fd = mkstemp(template);
2581         umask(msk);
2582         if (fd < 0)
2583                 return -1;
2584
2585         if (!rm)
2586                 return fd;
2587
2588         ret = unlink(template);
2589         if (ret < 0) {
2590                 close(fd);
2591                 return -1;
2592         }
2593
2594         return fd;
2595 }
2596
2597 int parse_byte_size_string(const char *s, int64_t *converted)
2598 {
2599         int ret, suffix_len;
2600         long long int conv;
2601         int64_t mltpl, overflow;
2602         char *end;
2603         char dup[LXC_NUMSTRLEN64 + 2];
2604         char suffix[3] = {0};
2605
2606         if (!s || !strcmp(s, ""))
2607                 return -EINVAL;
2608
2609         end = stpncpy(dup, s, sizeof(dup) - 1);
2610         if (*end != '\0')
2611                 return -EINVAL;
2612
2613         if (isdigit(*(end - 1)))
2614                 suffix_len = 0;
2615         else if (isalpha(*(end - 1)))
2616                 suffix_len = 1;
2617         else
2618                 return -EINVAL;
2619
2620         if (suffix_len > 0 && (end - 2) == dup && !isdigit(*(end - 2)))
2621                 return -EINVAL;
2622
2623         if (suffix_len > 0 && isalpha(*(end - 2)))
2624                 suffix_len++;
2625
2626         if (suffix_len > 0) {
2627                 memcpy(suffix, end - suffix_len, suffix_len);
2628                 *(suffix + suffix_len) = '\0';
2629                 *(end - suffix_len) = '\0';
2630         }
2631         dup[lxc_char_right_gc(dup, strlen(dup))] = '\0';
2632
2633         ret = lxc_safe_long_long(dup, &conv);
2634         if (ret < 0)
2635                 return -ret;
2636
2637         if (suffix_len != 2) {
2638                 *converted = conv;
2639                 return 0;
2640         }
2641
2642         if (strcasecmp(suffix, "KB") == 0)
2643                 mltpl = 1024;
2644         else if (strcasecmp(suffix, "MB") == 0)
2645                 mltpl = 1024 * 1024;
2646         else if (strcasecmp(suffix, "GB") == 0)
2647                 mltpl = 1024 * 1024 * 1024;
2648         else
2649                 return -EINVAL;
2650
2651         overflow = conv * mltpl;
2652         if (conv != 0 && (overflow / conv) != mltpl)
2653                 return -ERANGE;
2654
2655         *converted = overflow;
2656         return 0;
2657 }
2658
2659 uint64_t lxc_find_next_power2(uint64_t n)
2660 {
2661         /* 0 is not valid input. We return 0 to the caller since 0 is not a
2662          * valid power of two.
2663          */
2664         if (n == 0)
2665                 return 0;
2666
2667         if (!(n & (n - 1)))
2668                 return n;
2669
2670         while (n & (n - 1))
2671                 n = n & (n - 1);
2672
2673         n = n << 1;
2674         return n;
2675 }
2676
2677 int lxc_set_death_signal(int signal)
2678 {
2679         int ret;
2680         pid_t ppid;
2681
2682         ret = prctl(PR_SET_PDEATHSIG, signal, 0, 0, 0);
2683
2684         /* Check whether we have been orphaned. */
2685         ppid = (pid_t)syscall(SYS_getppid);
2686         if (ppid == 1) {
2687                 pid_t self;
2688
2689                 self = lxc_raw_getpid();
2690                 ret = kill(self, SIGKILL);
2691                 if (ret < 0)
2692                         return -1;
2693         }
2694
2695         if (ret < 0) {
2696                 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal);
2697                 return -1;
2698         }
2699
2700         return 0;
2701 }
2702
2703 void remove_trailing_newlines(char *l)
2704 {
2705         char *p = l;
2706
2707         while (*p)
2708                 p++;
2709
2710         while (--p >= l && *p == '\n')
2711                 *p = '\0';
2712 }
2713
2714 int fd_cloexec(int fd, bool cloexec)
2715 {
2716         int oflags, nflags;
2717
2718         oflags = fcntl(fd, F_GETFD, 0);
2719         if (oflags < 0)
2720                 return -errno;
2721
2722         if (cloexec)
2723                 nflags = oflags | FD_CLOEXEC;
2724         else
2725                 nflags = oflags & ~FD_CLOEXEC;
2726
2727         if (nflags == oflags)
2728                 return 0;
2729
2730         if (fcntl(fd, F_SETFD, nflags) < 0)
2731                 return -errno;
2732
2733         return 0;
2734 }
2735
2736 int recursive_destroy(char *dirname)
2737 {
2738         int ret;
2739         struct dirent *direntp;
2740         DIR *dir;
2741         int r = 0;
2742
2743         dir = opendir(dirname);
2744         if (!dir)
2745                 return -1;
2746
2747         while ((direntp = readdir(dir))) {
2748                 char *pathname;
2749                 struct stat mystat;
2750
2751                 if (!strcmp(direntp->d_name, ".") ||
2752                     !strcmp(direntp->d_name, ".."))
2753                         continue;
2754
2755                 pathname = must_make_path(dirname, direntp->d_name, NULL);
2756
2757                 ret = lstat(pathname, &mystat);
2758                 if (ret < 0) {
2759                         if (!r)
2760                                 WARN("Failed to stat \"%s\"", pathname);
2761
2762                         r = -1;
2763                         goto next;
2764                 }
2765
2766                 if (!S_ISDIR(mystat.st_mode))
2767                         goto next;
2768
2769                 ret = recursive_destroy(pathname);
2770                 if (ret < 0)
2771                         r = -1;
2772
2773         next:
2774                 free(pathname);
2775         }
2776
2777         ret = rmdir(dirname);
2778         if (ret < 0) {
2779                 if (!r)
2780                         SYSWARN("Failed to delete \"%s\"", dirname);
2781
2782                 r = -1;
2783         }
2784
2785         ret = closedir(dir);
2786         if (ret < 0) {
2787                 if (!r)
2788                         SYSWARN("Failed to delete \"%s\"", dirname);
2789
2790                 r = -1;
2791         }
2792
2793         return r;
2794 }