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