src/shared/machine-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <dirent.h>
   4 #include <errno.h>
   5 #include <fcntl.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <string.h>
   9 #include <sys/file.h>
  10 #include <sys/stat.h>
  11 #include <unistd.h>
  12 #include <linux/fs.h>
  13
  14 #include "alloc-util.h"
  15 #include "btrfs-util.h"
  16 #include "chattr-util.h"
  17 #include "copy.h"
  18 #include "dirent-util.h"
  19 #include "dissect-image.h"
  20 #include "env-util.h"
  21 #include "fd-util.h"
  22 #include "fileio.h"
  23 #include "fs-util.h"
  24 #include "hashmap.h"
  25 #include "hostname-util.h"
  26 #include "id128-util.h"
  27 #include "lockfile-util.h"
  28 #include "log.h"
  29 #include "loop-util.h"
  30 #include "machine-image.h"
  31 #include "macro.h"
  32 #include "mkdir.h"
  33 #include "os-util.h"
  34 #include "path-util.h"
  35 #include "rm-rf.h"
  36 #include "string-table.h"
  37 #include "string-util.h"
  38 #include "strv.h"
  39 #include "time-util.h"
  40 #include "utf8.h"
  41 #include "util.h"
  42 #include "xattr-util.h"
  43
  44 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  45         [IMAGE_MACHINE] =  "/etc/machines\0"              /* only place symlinks here */
  46                            "/run/machines\0"              /* and here too */
  47                            "/var/lib/machines\0"          /* the main place for images */
  48                            "/var/lib/container\0"         /* legacy */
  49                            "/usr/local/lib/machines\0"
  50                            "/usr/lib/machines\0",
  51
  52         [IMAGE_PORTABLE] = "/etc/portables\0"             /* only place symlinks here */
  53                            "/run/portables\0"             /* and here too */
  54                            "/var/lib/portables\0"         /* the main place for images */
  55                            "/usr/local/lib/portables\0"
  56                            "/usr/lib/portables\0",
  57 };
  58
  59 static Image *image_free(Image *i) {
  60         assert(i);
  61
  62         free(i->name);
  63         free(i->path);
  64
  65         free(i->hostname);
  66         strv_free(i->machine_info);
  67         strv_free(i->os_release);
  68
  69         return mfree(i);
  70 }
  71
  72 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  73
  74 static char **image_settings_path(Image *image) {
  75         _cleanup_strv_free_ char **l = NULL;
  76         const char *fn, *s;
  77         unsigned i = 0;
  78
  79         assert(image);
  80
  81         l = new0(char*, 4);
  82         if (!l)
  83                 return NULL;
  84
  85         fn = strjoina(image->name, ".nspawn");
  86
  87         FOREACH_STRING(s, "/etc/systemd/nspawn/", "/run/systemd/nspawn/") {
  88                 l[i] = strappend(s, fn);
  89                 if (!l[i])
  90                         return NULL;
  91
  92                 i++;
  93         }
  94
  95         l[i] = file_in_same_dir(image->path, fn);
  96         if (!l[i])
  97                 return NULL;
  98
  99         return TAKE_PTR(l);
 100 }
 101
 102 static char *image_roothash_path(Image *image) {
 103         const char *fn;
 104
 105         assert(image);
 106
 107         fn = strjoina(image->name, ".roothash");
 108
 109         return file_in_same_dir(image->path, fn);
 110 }
 111
 112 static int image_new(
 113                 ImageType t,
 114                 const char *pretty,
 115                 const char *path,
 116                 const char *filename,
 117                 bool read_only,
 118                 usec_t crtime,
 119                 usec_t mtime,
 120                 Image **ret) {
 121
 122         _cleanup_(image_unrefp) Image *i = NULL;
 123
 124         assert(t >= 0);
 125         assert(t < _IMAGE_TYPE_MAX);
 126         assert(pretty);
 127         assert(filename);
 128         assert(ret);
 129
 130         i = new0(Image, 1);
 131         if (!i)
 132                 return -ENOMEM;
 133
 134         i->n_ref = 1;
 135         i->type = t;
 136         i->read_only = read_only;
 137         i->crtime = crtime;
 138         i->mtime = mtime;
 139         i->usage = i->usage_exclusive = (uint64_t) -1;
 140         i->limit = i->limit_exclusive = (uint64_t) -1;
 141
 142         i->name = strdup(pretty);
 143         if (!i->name)
 144                 return -ENOMEM;
 145
 146         if (path)
 147                 i->path = strjoin(path, "/", filename);
 148         else
 149                 i->path = strdup(filename);
 150         if (!i->path)
 151                 return -ENOMEM;
 152
 153         path_simplify(i->path, false);
 154
 155         *ret = TAKE_PTR(i);
 156
 157         return 0;
 158 }
 159
 160 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 161         _cleanup_free_ char *name = NULL;
 162         const char *p;
 163         size_t n;
 164
 165         assert(path);
 166         assert(ret);
 167
 168         p = last_path_component(path);
 169         n = strcspn(p, "/");
 170
 171         name = strndup(p, n);
 172         if (!name)
 173                 return -ENOMEM;
 174
 175         if (suffix) {
 176                 char *e;
 177
 178                 e = endswith(name, suffix);
 179                 if (!e)
 180                         return -EINVAL;
 181
 182                 *e = 0;
 183         }
 184
 185         if (!image_name_is_valid(name))
 186                 return -EINVAL;
 187
 188         *ret = TAKE_PTR(name);
 189         return 0;
 190 }
 191
 192 static int image_make(
 193                 const char *pretty,
 194                 int dfd,
 195                 const char *path,
 196                 const char *filename,
 197                 const struct stat *st,
 198                 Image **ret) {
 199
 200         _cleanup_free_ char *pretty_buffer = NULL;
 201         struct stat stbuf;
 202         bool read_only;
 203         int r;
 204
 205         assert(dfd >= 0 || dfd == AT_FDCWD);
 206         assert(filename);
 207
 208         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 209          * devices into /var/lib/machines/, and treat them normally.
 210          *
 211          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 212          * recognize. */
 213
 214         if (!st) {
 215                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 216                         return -errno;
 217
 218                 st = &stbuf;
 219         }
 220
 221         read_only =
 222                 (path && path_startswith(path, "/usr")) ||
 223                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 224
 225         if (S_ISDIR(st->st_mode)) {
 226                 _cleanup_close_ int fd = -1;
 227                 unsigned file_attr = 0;
 228
 229                 if (!ret)
 230                         return 0;
 231
 232                 if (!pretty) {
 233                         r = extract_pretty(filename, NULL, &pretty_buffer);
 234                         if (r < 0)
 235                                 return r;
 236
 237                         pretty = pretty_buffer;
 238                 }
 239
 240                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 241                 if (fd < 0)
 242                         return -errno;
 243
 244                 /* btrfs subvolumes have inode 256 */
 245                 if (st->st_ino == 256) {
 246
 247                         r = btrfs_is_filesystem(fd);
 248                         if (r < 0)
 249                                 return r;
 250                         if (r) {
 251                                 BtrfsSubvolInfo info;
 252
 253                                 /* It's a btrfs subvolume */
 254
 255                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 256                                 if (r < 0)
 257                                         return r;
 258
 259                                 r = image_new(IMAGE_SUBVOLUME,
 260                                               pretty,
 261                                               path,
 262                                               filename,
 263                                               info.read_only || read_only,
 264                                               info.otime,
 265                                               0,
 266                                               ret);
 267                                 if (r < 0)
 268                                         return r;
 269
 270                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 271                                         BtrfsQuotaInfo quota;
 272
 273                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 274                                         if (r >= 0) {
 275                                                 (*ret)->usage = quota.referenced;
 276                                                 (*ret)->usage_exclusive = quota.exclusive;
 277
 278                                                 (*ret)->limit = quota.referenced_max;
 279                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 280                                         }
 281                                 }
 282
 283                                 return 0;
 284                         }
 285                 }
 286
 287                 /* If the IMMUTABLE bit is set, we consider the
 288                  * directory read-only. Since the ioctl is not
 289                  * supported everywhere we ignore failures. */
 290                 (void) read_attr_fd(fd, &file_attr);
 291
 292                 /* It's just a normal directory. */
 293                 r = image_new(IMAGE_DIRECTORY,
 294                               pretty,
 295                               path,
 296                               filename,
 297                               read_only || (file_attr & FS_IMMUTABLE_FL),
 298                               0,
 299                               0,
 300                               ret);
 301                 if (r < 0)
 302                         return r;
 303
 304                 return 0;
 305
 306         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 307                 usec_t crtime = 0;
 308
 309                 /* It's a RAW disk image */
 310
 311                 if (!ret)
 312                         return 0;
 313
 314                 (void) fd_getcrtime_at(dfd, filename, &crtime, 0);
 315
 316                 if (!pretty) {
 317                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 318                         if (r < 0)
 319                                 return r;
 320
 321                         pretty = pretty_buffer;
 322                 }
 323
 324                 r = image_new(IMAGE_RAW,
 325                               pretty,
 326                               path,
 327                               filename,
 328                               !(st->st_mode & 0222) || read_only,
 329                               crtime,
 330                               timespec_load(&st->st_mtim),
 331                               ret);
 332                 if (r < 0)
 333                         return r;
 334
 335                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 336                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 337
 338                 return 0;
 339
 340         } else if (S_ISBLK(st->st_mode)) {
 341                 _cleanup_close_ int block_fd = -1;
 342                 uint64_t size = UINT64_MAX;
 343
 344                 /* A block device */
 345
 346                 if (!ret)
 347                         return 0;
 348
 349                 if (!pretty) {
 350                         r = extract_pretty(filename, NULL, &pretty_buffer);
 351                         if (r < 0)
 352                                 return r;
 353
 354                         pretty = pretty_buffer;
 355                 }
 356
 357                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 358                 if (block_fd < 0)
 359                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path, filename);
 360                 else {
 361                         /* Refresh stat data after opening the node */
 362                         if (fstat(block_fd, &stbuf) < 0)
 363                                 return -errno;
 364                         st = &stbuf;
 365
 366                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 367                                 return -ENOTTY;
 368
 369                         if (!read_only) {
 370                                 int state = 0;
 371
 372                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 373                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path, filename);
 374                                 else if (state)
 375                                         read_only = true;
 376                         }
 377
 378                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 379                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path, filename);
 380
 381                         block_fd = safe_close(block_fd);
 382                 }
 383
 384                 r = image_new(IMAGE_BLOCK,
 385                               pretty,
 386                               path,
 387                               filename,
 388                               !(st->st_mode & 0222) || read_only,
 389                               0,
 390                               0,
 391                               ret);
 392                 if (r < 0)
 393                         return r;
 394
 395                 if (size != 0 && size != UINT64_MAX)
 396                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 397
 398                 return 0;
 399         }
 400
 401         return -EMEDIUMTYPE;
 402 }
 403
 404 int image_find(ImageClass class, const char *name, Image **ret) {
 405         const char *path;
 406         int r;
 407
 408         assert(class >= 0);
 409         assert(class < _IMAGE_CLASS_MAX);
 410         assert(name);
 411
 412         /* There are no images with invalid names */
 413         if (!image_name_is_valid(name))
 414                 return -ENOENT;
 415
 416         NULSTR_FOREACH(path, image_search_path[class]) {
 417                 _cleanup_closedir_ DIR *d = NULL;
 418                 struct stat st;
 419
 420                 d = opendir(path);
 421                 if (!d) {
 422                         if (errno == ENOENT)
 423                                 continue;
 424
 425                         return -errno;
 426                 }
 427
 428                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to
 429                  * symlink block devices into the search path */
 430                 if (fstatat(dirfd(d), name, &st, 0) < 0) {
 431                         _cleanup_free_ char *raw = NULL;
 432
 433                         if (errno != ENOENT)
 434                                 return -errno;
 435
 436                         raw = strappend(name, ".raw");
 437                         if (!raw)
 438                                 return -ENOMEM;
 439
 440                         if (fstatat(dirfd(d), raw, &st, 0) < 0) {
 441
 442                                 if (errno == ENOENT)
 443                                         continue;
 444
 445                                 return -errno;
 446                         }
 447
 448                         if (!S_ISREG(st.st_mode))
 449                                 continue;
 450
 451                         r = image_make(name, dirfd(d), path, raw, &st, ret);
 452
 453                 } else {
 454                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 455                                 continue;
 456
 457                         r = image_make(name, dirfd(d), path, name, &st, ret);
 458                 }
 459                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 460                         continue;
 461                 if (r < 0)
 462                         return r;
 463
 464                 if (ret)
 465                         (*ret)->discoverable = true;
 466
 467                 return 1;
 468         }
 469
 470         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 471                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 472                 if (r < 0)
 473                         return r;
 474
 475                 if (ret)
 476                         (*ret)->discoverable = true;
 477
 478                 return r;
 479         }
 480
 481         return -ENOENT;
 482 };
 483
 484 int image_from_path(const char *path, Image **ret) {
 485
 486         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 487          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 488          * overridden by another, different image earlier in the search path */
 489
 490         if (path_equal(path, "/"))
 491                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 492
 493         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 494 }
 495
 496 int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) {
 497         if (image_name_is_valid(name_or_path))
 498                 return image_find(class, name_or_path, ret);
 499
 500         return image_from_path(name_or_path, ret);
 501 }
 502
 503 int image_discover(ImageClass class, Hashmap *h) {
 504         const char *path;
 505         int r;
 506
 507         assert(class >= 0);
 508         assert(class < _IMAGE_CLASS_MAX);
 509         assert(h);
 510
 511         NULSTR_FOREACH(path, image_search_path[class]) {
 512                 _cleanup_closedir_ DIR *d = NULL;
 513                 struct dirent *de;
 514
 515                 d = opendir(path);
 516                 if (!d) {
 517                         if (errno == ENOENT)
 518                                 continue;
 519
 520                         return -errno;
 521                 }
 522
 523                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 524                         _cleanup_(image_unrefp) Image *image = NULL;
 525                         _cleanup_free_ char *truncated = NULL;
 526                         const char *pretty;
 527                         struct stat st;
 528
 529                         if (dot_or_dot_dot(de->d_name))
 530                                 continue;
 531
 532                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 533                          * to symlink block devices into the search path */
 534                         if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) {
 535                                 if (errno == ENOENT)
 536                                         continue;
 537
 538                                 return -errno;
 539                         }
 540
 541                         if (S_ISREG(st.st_mode)) {
 542                                 const char *e;
 543
 544                                 e = endswith(de->d_name, ".raw");
 545                                 if (!e)
 546                                         continue;
 547
 548                                 truncated = strndup(de->d_name, e - de->d_name);
 549                                 if (!truncated)
 550                                         return -ENOMEM;
 551
 552                                 pretty = truncated;
 553                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 554                                 pretty = de->d_name;
 555                         else
 556                                 continue;
 557
 558                         if (!image_name_is_valid(pretty))
 559                                 continue;
 560
 561                         if (hashmap_contains(h, pretty))
 562                                 continue;
 563
 564                         r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image);
 565                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 566                                 continue;
 567                         if (r < 0)
 568                                 return r;
 569
 570                         image->discoverable = true;
 571
 572                         r = hashmap_put(h, image->name, image);
 573                         if (r < 0)
 574                                 return r;
 575
 576                         image = NULL;
 577                 }
 578         }
 579
 580         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 581                 _cleanup_(image_unrefp) Image *image = NULL;
 582
 583                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image);
 584                 if (r < 0)
 585                         return r;
 586
 587                 image->discoverable = true;
 588
 589                 r = hashmap_put(h, image->name, image);
 590                 if (r < 0)
 591                         return r;
 592
 593                 image = NULL;
 594         }
 595
 596         return 0;
 597 }
 598
 599 int image_remove(Image *i) {
 600         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 601         _cleanup_strv_free_ char **settings = NULL;
 602         _cleanup_free_ char *roothash = NULL;
 603         char **j;
 604         int r;
 605
 606         assert(i);
 607
 608         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 609                 return -EROFS;
 610
 611         settings = image_settings_path(i);
 612         if (!settings)
 613                 return -ENOMEM;
 614
 615         roothash = image_roothash_path(i);
 616         if (!roothash)
 617                 return -ENOMEM;
 618
 619         /* Make sure we don't interfere with a running nspawn */
 620         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 621         if (r < 0)
 622                 return r;
 623
 624         switch (i->type) {
 625
 626         case IMAGE_SUBVOLUME:
 627
 628                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 629                  * big guns */
 630                 if (unlink(i->path) < 0) {
 631                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 632                         if (r < 0)
 633                                 return r;
 634                 }
 635
 636                 break;
 637
 638         case IMAGE_DIRECTORY:
 639                 /* Allow deletion of read-only directories */
 640                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 641                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 642                 if (r < 0)
 643                         return r;
 644
 645                 break;
 646
 647         case IMAGE_BLOCK:
 648
 649                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 650                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 651                  * the thing (it's most likely a symlink after all). */
 652
 653                 if (path_startswith(i->path, "/dev"))
 654                         break;
 655
 656                 _fallthrough_;
 657         case IMAGE_RAW:
 658                 if (unlink(i->path) < 0)
 659                         return -errno;
 660                 break;
 661
 662         default:
 663                 return -EOPNOTSUPP;
 664         }
 665
 666         STRV_FOREACH(j, settings) {
 667                 if (unlink(*j) < 0 && errno != ENOENT)
 668                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 669         }
 670
 671         if (unlink(roothash) < 0 && errno != ENOENT)
 672                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 673
 674         return 0;
 675 }
 676
 677 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 678         _cleanup_free_ char *rs = NULL;
 679         const char *fn;
 680
 681         fn = strjoina(new_name, suffix);
 682
 683         rs = file_in_same_dir(path, fn);
 684         if (!rs)
 685                 return -ENOMEM;
 686
 687         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 688 }
 689
 690 int image_rename(Image *i, const char *new_name) {
 691         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 692         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 693         _cleanup_strv_free_ char **settings = NULL;
 694         unsigned file_attr = 0;
 695         char **j;
 696         int r;
 697
 698         assert(i);
 699
 700         if (!image_name_is_valid(new_name))
 701                 return -EINVAL;
 702
 703         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 704                 return -EROFS;
 705
 706         settings = image_settings_path(i);
 707         if (!settings)
 708                 return -ENOMEM;
 709
 710         roothash = image_roothash_path(i);
 711         if (!roothash)
 712                 return -ENOMEM;
 713
 714         /* Make sure we don't interfere with a running nspawn */
 715         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 716         if (r < 0)
 717                 return r;
 718
 719         /* Make sure nobody takes the new name, between the time we
 720          * checked it is currently unused in all search paths, and the
 721          * time we take possession of it */
 722         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 723         if (r < 0)
 724                 return r;
 725
 726         r = image_find(IMAGE_MACHINE, new_name, NULL);
 727         if (r >= 0)
 728                 return -EEXIST;
 729         if (r != -ENOENT)
 730                 return r;
 731
 732         switch (i->type) {
 733
 734         case IMAGE_DIRECTORY:
 735                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 736                 (void) read_attr_path(i->path, &file_attr);
 737
 738                 if (file_attr & FS_IMMUTABLE_FL)
 739                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 740
 741                 _fallthrough_;
 742         case IMAGE_SUBVOLUME:
 743                 new_path = file_in_same_dir(i->path, new_name);
 744                 break;
 745
 746         case IMAGE_BLOCK:
 747
 748                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 749                 if (path_startswith(i->path, "/dev"))
 750                         return -EROFS;
 751
 752                 new_path = file_in_same_dir(i->path, new_name);
 753                 break;
 754
 755         case IMAGE_RAW: {
 756                 const char *fn;
 757
 758                 fn = strjoina(new_name, ".raw");
 759                 new_path = file_in_same_dir(i->path, fn);
 760                 break;
 761         }
 762
 763         default:
 764                 return -EOPNOTSUPP;
 765         }
 766
 767         if (!new_path)
 768                 return -ENOMEM;
 769
 770         nn = strdup(new_name);
 771         if (!nn)
 772                 return -ENOMEM;
 773
 774         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 775         if (r < 0)
 776                 return r;
 777
 778         /* Restore the immutable bit, if it was set before */
 779         if (file_attr & FS_IMMUTABLE_FL)
 780                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 781
 782         free_and_replace(i->path, new_path);
 783         free_and_replace(i->name, nn);
 784
 785         STRV_FOREACH(j, settings) {
 786                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 787                 if (r < 0 && r != -ENOENT)
 788                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 789         }
 790
 791         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 792         if (r < 0 && r != -ENOENT)
 793                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 794
 795         return 0;
 796 }
 797
 798 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 799         _cleanup_free_ char *rs = NULL;
 800         const char *fn;
 801
 802         fn = strjoina(new_name, suffix);
 803
 804         rs = file_in_same_dir(path, fn);
 805         if (!rs)
 806                 return -ENOMEM;
 807
 808         return copy_file_atomic(path, rs, 0664, 0, COPY_REFLINK);
 809 }
 810
 811 int image_clone(Image *i, const char *new_name, bool read_only) {
 812         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 813         _cleanup_strv_free_ char **settings = NULL;
 814         _cleanup_free_ char *roothash = NULL;
 815         const char *new_path;
 816         char **j;
 817         int r;
 818
 819         assert(i);
 820
 821         if (!image_name_is_valid(new_name))
 822                 return -EINVAL;
 823
 824         settings = image_settings_path(i);
 825         if (!settings)
 826                 return -ENOMEM;
 827
 828         roothash = image_roothash_path(i);
 829         if (!roothash)
 830                 return -ENOMEM;
 831
 832         /* Make sure nobody takes the new name, between the time we
 833          * checked it is currently unused in all search paths, and the
 834          * time we take possession of it */
 835         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 836         if (r < 0)
 837                 return r;
 838
 839         r = image_find(IMAGE_MACHINE, new_name, NULL);
 840         if (r >= 0)
 841                 return -EEXIST;
 842         if (r != -ENOENT)
 843                 return r;
 844
 845         switch (i->type) {
 846
 847         case IMAGE_SUBVOLUME:
 848         case IMAGE_DIRECTORY:
 849                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 850                  * directory. */
 851
 852                 new_path = strjoina("/var/lib/machines/", new_name);
 853
 854                 r = btrfs_subvol_snapshot(i->path, new_path,
 855                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 856                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 857                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 858                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 859                                           BTRFS_SNAPSHOT_RECURSIVE |
 860                                           BTRFS_SNAPSHOT_QUOTA);
 861                 if (r >= 0)
 862                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 863                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 864
 865                 break;
 866
 867         case IMAGE_RAW:
 868                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 869
 870                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, COPY_REFLINK);
 871                 break;
 872
 873         case IMAGE_BLOCK:
 874         default:
 875                 return -EOPNOTSUPP;
 876         }
 877
 878         if (r < 0)
 879                 return r;
 880
 881         STRV_FOREACH(j, settings) {
 882                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 883                 if (r < 0 && r != -ENOENT)
 884                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 885         }
 886
 887         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 888         if (r < 0 && r != -ENOENT)
 889                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 890
 891         return 0;
 892 }
 893
 894 int image_read_only(Image *i, bool b) {
 895         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 896         int r;
 897
 898         assert(i);
 899
 900         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 901                 return -EROFS;
 902
 903         /* Make sure we don't interfere with a running nspawn */
 904         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 905         if (r < 0)
 906                 return r;
 907
 908         switch (i->type) {
 909
 910         case IMAGE_SUBVOLUME:
 911
 912                 /* Note that we set the flag only on the top-level
 913                  * subvolume of the image. */
 914
 915                 r = btrfs_subvol_set_read_only(i->path, b);
 916                 if (r < 0)
 917                         return r;
 918
 919                 break;
 920
 921         case IMAGE_DIRECTORY:
 922                 /* For simple directory trees we cannot use the access
 923                    mode of the top-level directory, since it has an
 924                    effect on the container itself.  However, we can
 925                    use the "immutable" flag, to at least make the
 926                    top-level directory read-only. It's not as good as
 927                    a read-only subvolume, but at least something, and
 928                    we can read the value back. */
 929
 930                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 931                 if (r < 0)
 932                         return r;
 933
 934                 break;
 935
 936         case IMAGE_RAW: {
 937                 struct stat st;
 938
 939                 if (stat(i->path, &st) < 0)
 940                         return -errno;
 941
 942                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 943                         return -errno;
 944
 945                 /* If the images is now read-only, it's a good time to
 946                  * defrag it, given that no write patterns will
 947                  * fragment it again. */
 948                 if (b)
 949                         (void) btrfs_defrag(i->path);
 950                 break;
 951         }
 952
 953         case IMAGE_BLOCK: {
 954                 _cleanup_close_ int fd = -1;
 955                 struct stat st;
 956                 int state = b;
 957
 958                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 959                 if (fd < 0)
 960                         return -errno;
 961
 962                 if (fstat(fd, &st) < 0)
 963                         return -errno;
 964                 if (!S_ISBLK(st.st_mode))
 965                         return -ENOTTY;
 966
 967                 if (ioctl(fd, BLKROSET, &state) < 0)
 968                         return -errno;
 969
 970                 break;
 971         }
 972
 973         default:
 974                 return -EOPNOTSUPP;
 975         }
 976
 977         return 0;
 978 }
 979
 980 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
 981         _cleanup_free_ char *p = NULL;
 982         LockFile t = LOCK_FILE_INIT;
 983         struct stat st;
 984         int r;
 985
 986         assert(path);
 987         assert(global);
 988         assert(local);
 989
 990         /* Locks an image path. This actually creates two locks: one
 991          * "local" one, next to the image path itself, which might be
 992          * shared via NFS. And another "global" one, in /run, that
 993          * uses the device/inode number. This has the benefit that we
 994          * can even lock a tree that is a mount point, correctly. */
 995
 996         if (!path_is_absolute(path))
 997                 return -EINVAL;
 998
 999         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1000                 *local = *global = (LockFile) LOCK_FILE_INIT;
1001                 return 0;
1002         }
1003
1004         if (path_equal(path, "/"))
1005                 return -EBUSY;
1006
1007         if (stat(path, &st) >= 0) {
1008                 if (S_ISBLK(st.st_mode))
1009                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1010                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1011                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1012                 else
1013                         return -ENOTTY;
1014
1015                 if (r < 0)
1016                         return -ENOMEM;
1017         }
1018
1019         /* For block devices we don't need the "local" lock, as the major/minor lock above should be sufficient, since
1020          * block devices are device local anyway. */
1021         if (!path_startswith(path, "/dev")) {
1022                 r = make_lock_file_for(path, operation, &t);
1023                 if (r < 0) {
1024                         if ((operation & LOCK_SH) && r == -EROFS)
1025                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1026                         else
1027                                 return r;
1028                 }
1029         }
1030
1031         if (p) {
1032                 mkdir_p("/run/systemd/nspawn/locks", 0700);
1033
1034                 r = make_lock_file(p, operation, global);
1035                 if (r < 0) {
1036                         release_lock_file(&t);
1037                         return r;
1038                 }
1039         } else
1040                 *global = (LockFile) LOCK_FILE_INIT;
1041
1042         *local = t;
1043         return 0;
1044 }
1045
1046 int image_set_limit(Image *i, uint64_t referenced_max) {
1047         assert(i);
1048
1049         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1050                 return -EROFS;
1051
1052         if (i->type != IMAGE_SUBVOLUME)
1053                 return -EOPNOTSUPP;
1054
1055         /* We set the quota both for the subvolume as well as for the
1056          * subtree. The latter is mostly for historical reasons, since
1057          * we didn't use to have a concept of subtree quota, and hence
1058          * only modified the subvolume quota. */
1059
1060         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1061         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1062         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1063 }
1064
1065 int image_read_metadata(Image *i) {
1066         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1067         int r;
1068
1069         assert(i);
1070
1071         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1072         if (r < 0)
1073                 return r;
1074
1075         switch (i->type) {
1076
1077         case IMAGE_SUBVOLUME:
1078         case IMAGE_DIRECTORY: {
1079                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1080                 sd_id128_t machine_id = SD_ID128_NULL;
1081                 _cleanup_free_ char *hostname = NULL;
1082                 _cleanup_free_ char *path = NULL;
1083
1084                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1085                 if (r < 0 && r != -ENOENT)
1086                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1087                 else if (r >= 0) {
1088                         r = read_etc_hostname(path, &hostname);
1089                         if (r < 0)
1090                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1091                 }
1092
1093                 path = mfree(path);
1094
1095                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1096                 if (r < 0 && r != -ENOENT)
1097                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1098                 else if (r >= 0) {
1099                         _cleanup_close_ int fd = -1;
1100
1101                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1102                         if (fd < 0)
1103                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1104                         else {
1105                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1106                                 if (r < 0)
1107                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1108                         }
1109                 }
1110
1111                 path = mfree(path);
1112
1113                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1114                 if (r < 0 && r != -ENOENT)
1115                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1116                 else if (r >= 0) {
1117                         r = load_env_file_pairs(NULL, path, NULL, &machine_info);
1118                         if (r < 0)
1119                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1120                 }
1121
1122                 r = load_os_release_pairs(i->path, &os_release);
1123                 if (r < 0)
1124                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1125
1126                 free_and_replace(i->hostname, hostname);
1127                 i->machine_id = machine_id;
1128                 strv_free_and_replace(i->machine_info, machine_info);
1129                 strv_free_and_replace(i->os_release, os_release);
1130
1131                 break;
1132         }
1133
1134         case IMAGE_RAW:
1135         case IMAGE_BLOCK: {
1136                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1137                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1138
1139                 r = loop_device_make_by_path(i->path, O_RDONLY, &d);
1140                 if (r < 0)
1141                         return r;
1142
1143                 r = dissect_image(d->fd, NULL, 0, DISSECT_IMAGE_REQUIRE_ROOT, &m);
1144                 if (r < 0)
1145                         return r;
1146
1147                 r = dissected_image_acquire_metadata(m);
1148                 if (r < 0)
1149                         return r;
1150
1151                 free_and_replace(i->hostname, m->hostname);
1152                 i->machine_id = m->machine_id;
1153                 strv_free_and_replace(i->machine_info, m->machine_info);
1154                 strv_free_and_replace(i->os_release, m->os_release);
1155
1156                 break;
1157         }
1158
1159         default:
1160                 return -EOPNOTSUPP;
1161         }
1162
1163         i->metadata_valid = true;
1164
1165         return 0;
1166 }
1167
1168 int image_name_lock(const char *name, int operation, LockFile *ret) {
1169         const char *p;
1170
1171         assert(name);
1172         assert(ret);
1173
1174         /* Locks an image name, regardless of the precise path used. */
1175
1176         if (!image_name_is_valid(name))
1177                 return -EINVAL;
1178
1179         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1180                 *ret = (LockFile) LOCK_FILE_INIT;
1181                 return 0;
1182         }
1183
1184         if (streq(name, ".host"))
1185                 return -EBUSY;
1186
1187         mkdir_p("/run/systemd/nspawn/locks", 0700);
1188         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1189
1190         return make_lock_file(p, operation, ret);
1191 }
1192
1193 bool image_name_is_valid(const char *s) {
1194         if (!filename_is_valid(s))
1195                 return false;
1196
1197         if (string_has_cc(s, NULL))
1198                 return false;
1199
1200         if (!utf8_is_valid(s))
1201                 return false;
1202
1203         /* Temporary files for atomically creating new files */
1204         if (startswith(s, ".#"))
1205                 return false;
1206
1207         return true;
1208 }
1209
1210 bool image_in_search_path(ImageClass class, const char *image) {
1211         const char *path;
1212
1213         assert(image);
1214
1215         NULSTR_FOREACH(path, image_search_path[class]) {
1216                 const char *p;
1217                 size_t k;
1218
1219                 p = path_startswith(image, path);
1220                 if (!p)
1221                         continue;
1222
1223                 /* Make sure there's a filename following */
1224                 k = strcspn(p, "/");
1225                 if (k == 0)
1226                         continue;
1227
1228                 p += k;
1229
1230                 /* Accept trailing slashes */
1231                 if (p[strspn(p, "/")] == 0)
1232                         return true;
1233
1234         }
1235
1236         return false;
1237 }
1238
1239 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1240         [IMAGE_DIRECTORY] = "directory",
1241         [IMAGE_SUBVOLUME] = "subvolume",
1242         [IMAGE_RAW] = "raw",
1243         [IMAGE_BLOCK] = "block",
1244 };
1245
1246 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);