src/shared/discover-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <errno.h>
   4 #include <fcntl.h>
   5 #include <linux/fs.h>
   6 #include <linux/loop.h>
   7 #include <linux/magic.h>
   8 #include <stdio.h>
   9 #include <stdlib.h>
  10 #include <sys/file.h>
  11 #include <sys/ioctl.h>
  12 #include <sys/stat.h>
  13 #include <unistd.h>
  14
  15 #include "alloc-util.h"
  16 #include "btrfs-util.h"
  17 #include "chase-symlinks.h"
  18 #include "chattr-util.h"
  19 #include "copy.h"
  20 #include "dirent-util.h"
  21 #include "discover-image.h"
  22 #include "dissect-image.h"
  23 #include "env-file.h"
  24 #include "env-util.h"
  25 #include "fd-util.h"
  26 #include "fs-util.h"
  27 #include "hashmap.h"
  28 #include "hostname-setup.h"
  29 #include "id128-util.h"
  30 #include "lockfile-util.h"
  31 #include "log.h"
  32 #include "loop-util.h"
  33 #include "macro.h"
  34 #include "mkdir.h"
  35 #include "nulstr-util.h"
  36 #include "os-util.h"
  37 #include "path-util.h"
  38 #include "rm-rf.h"
  39 #include "stat-util.h"
  40 #include "string-table.h"
  41 #include "string-util.h"
  42 #include "strv.h"
  43 #include "time-util.h"
  44 #include "utf8.h"
  45 #include "xattr-util.h"
  46
  47 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  48         [IMAGE_MACHINE] =   "/etc/machines\0"              /* only place symlinks here */
  49                             "/run/machines\0"              /* and here too */
  50                             "/var/lib/machines\0"          /* the main place for images */
  51                             "/var/lib/container\0"         /* legacy */
  52                             "/usr/local/lib/machines\0"
  53                             "/usr/lib/machines\0",
  54
  55         [IMAGE_PORTABLE] =  "/etc/portables\0"             /* only place symlinks here */
  56                             "/run/portables\0"             /* and here too */
  57                             "/var/lib/portables\0"         /* the main place for images */
  58                             "/usr/local/lib/portables\0"
  59                             "/usr/lib/portables\0",
  60
  61         [IMAGE_EXTENSION] = "/etc/extensions\0"             /* only place symlinks here */
  62                             "/run/extensions\0"             /* and here too */
  63                             "/var/lib/extensions\0"         /* the main place for images */
  64                             "/usr/local/lib/extensions\0"
  65                             "/usr/lib/extensions\0",
  66 };
  67
  68 static Image *image_free(Image *i) {
  69         assert(i);
  70
  71         free(i->name);
  72         free(i->path);
  73
  74         free(i->hostname);
  75         strv_free(i->machine_info);
  76         strv_free(i->os_release);
  77         strv_free(i->extension_release);
  78
  79         return mfree(i);
  80 }
  81
  82 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  83 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
  84                                       Image, image_unref);
  85
  86 static char **image_settings_path(Image *image) {
  87         _cleanup_strv_free_ char **l = NULL;
  88         const char *fn;
  89         unsigned i = 0;
  90
  91         assert(image);
  92
  93         l = new0(char*, 4);
  94         if (!l)
  95                 return NULL;
  96
  97         fn = strjoina(image->name, ".nspawn");
  98
  99         FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
 100                 l[i] = path_join(s, fn);
 101                 if (!l[i])
 102                         return NULL;
 103
 104                 i++;
 105         }
 106
 107         l[i] = file_in_same_dir(image->path, fn);
 108         if (!l[i])
 109                 return NULL;
 110
 111         return TAKE_PTR(l);
 112 }
 113
 114 static char *image_roothash_path(Image *image) {
 115         const char *fn;
 116
 117         assert(image);
 118
 119         fn = strjoina(image->name, ".roothash");
 120
 121         return file_in_same_dir(image->path, fn);
 122 }
 123
 124 static int image_new(
 125                 ImageType t,
 126                 const char *pretty,
 127                 const char *path,
 128                 const char *filename,
 129                 bool read_only,
 130                 usec_t crtime,
 131                 usec_t mtime,
 132                 Image **ret) {
 133
 134         _cleanup_(image_unrefp) Image *i = NULL;
 135
 136         assert(t >= 0);
 137         assert(t < _IMAGE_TYPE_MAX);
 138         assert(pretty);
 139         assert(filename);
 140         assert(ret);
 141
 142         i = new(Image, 1);
 143         if (!i)
 144                 return -ENOMEM;
 145
 146         *i = (Image) {
 147                 .n_ref = 1,
 148                 .type = t,
 149                 .read_only = read_only,
 150                 .crtime = crtime,
 151                 .mtime = mtime,
 152                 .usage = UINT64_MAX,
 153                 .usage_exclusive = UINT64_MAX,
 154                 .limit = UINT64_MAX,
 155                 .limit_exclusive = UINT64_MAX,
 156         };
 157
 158         i->name = strdup(pretty);
 159         if (!i->name)
 160                 return -ENOMEM;
 161
 162         i->path = path_join(path, filename);
 163         if (!i->path)
 164                 return -ENOMEM;
 165
 166         path_simplify(i->path);
 167
 168         *ret = TAKE_PTR(i);
 169
 170         return 0;
 171 }
 172
 173 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 174         _cleanup_free_ char *name = NULL;
 175         const char *p;
 176         size_t n;
 177
 178         assert(path);
 179         assert(ret);
 180
 181         p = last_path_component(path);
 182         n = strcspn(p, "/");
 183
 184         name = strndup(p, n);
 185         if (!name)
 186                 return -ENOMEM;
 187
 188         if (suffix) {
 189                 char *e;
 190
 191                 e = endswith(name, suffix);
 192                 if (!e)
 193                         return -EINVAL;
 194
 195                 *e = 0;
 196         }
 197
 198         if (!image_name_is_valid(name))
 199                 return -EINVAL;
 200
 201         *ret = TAKE_PTR(name);
 202         return 0;
 203 }
 204
 205 static int image_make(
 206                 const char *pretty,
 207                 int dfd,
 208                 const char *path,
 209                 const char *filename,
 210                 const struct stat *st,
 211                 Image **ret) {
 212
 213         _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL;
 214         struct stat stbuf;
 215         bool read_only;
 216         int r;
 217
 218         assert(dfd >= 0 || dfd == AT_FDCWD);
 219         assert(path || dfd == AT_FDCWD);
 220         assert(filename);
 221
 222         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 223          * devices into /var/lib/machines/, and treat them normally.
 224          *
 225          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 226          * recognize. */
 227
 228         if (!st) {
 229                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 230                         return -errno;
 231
 232                 st = &stbuf;
 233         }
 234
 235         if (!path) {
 236                 if (dfd == AT_FDCWD)
 237                         (void) safe_getcwd(&parent);
 238                 else
 239                         (void) fd_get_path(dfd, &parent);
 240         }
 241
 242         read_only =
 243                 (path && path_startswith(path, "/usr")) ||
 244                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 245
 246         if (S_ISDIR(st->st_mode)) {
 247                 _cleanup_close_ int fd = -1;
 248                 unsigned file_attr = 0;
 249                 usec_t crtime = 0;
 250
 251                 if (!ret)
 252                         return 0;
 253
 254                 if (!pretty) {
 255                         r = extract_pretty(filename, NULL, &pretty_buffer);
 256                         if (r < 0)
 257                                 return r;
 258
 259                         pretty = pretty_buffer;
 260                 }
 261
 262                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 263                 if (fd < 0)
 264                         return -errno;
 265
 266                 if (btrfs_might_be_subvol(st)) {
 267
 268                         r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
 269                         if (r < 0)
 270                                 return r;
 271                         if (r) {
 272                                 BtrfsSubvolInfo info;
 273
 274                                 /* It's a btrfs subvolume */
 275
 276                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 277                                 if (r < 0)
 278                                         return r;
 279
 280                                 r = image_new(IMAGE_SUBVOLUME,
 281                                               pretty,
 282                                               path,
 283                                               filename,
 284                                               info.read_only || read_only,
 285                                               info.otime,
 286                                               0,
 287                                               ret);
 288                                 if (r < 0)
 289                                         return r;
 290
 291                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 292                                         BtrfsQuotaInfo quota;
 293
 294                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 295                                         if (r >= 0) {
 296                                                 (*ret)->usage = quota.referenced;
 297                                                 (*ret)->usage_exclusive = quota.exclusive;
 298
 299                                                 (*ret)->limit = quota.referenced_max;
 300                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 301                                         }
 302                                 }
 303
 304                                 return 0;
 305                         }
 306                 }
 307
 308                 /* Get directory creation time (not available everywhere, but that's OK */
 309                 (void) fd_getcrtime(fd, &crtime);
 310
 311                 /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not
 312                  * supported everywhere we ignore failures. */
 313                 (void) read_attr_fd(fd, &file_attr);
 314
 315                 /* It's just a normal directory. */
 316                 r = image_new(IMAGE_DIRECTORY,
 317                               pretty,
 318                               path,
 319                               filename,
 320                               read_only || (file_attr & FS_IMMUTABLE_FL),
 321                               crtime,
 322                               0, /* we don't use mtime of stat() here, since it's not the time of last change of the tree, but only of the top-level dir */
 323                               ret);
 324                 if (r < 0)
 325                         return r;
 326
 327                 return 0;
 328
 329         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 330                 usec_t crtime = 0;
 331
 332                 /* It's a RAW disk image */
 333
 334                 if (!ret)
 335                         return 0;
 336
 337                 (void) fd_getcrtime_at(dfd, filename, AT_SYMLINK_FOLLOW, &crtime);
 338
 339                 if (!pretty) {
 340                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 341                         if (r < 0)
 342                                 return r;
 343
 344                         pretty = pretty_buffer;
 345                 }
 346
 347                 r = image_new(IMAGE_RAW,
 348                               pretty,
 349                               path,
 350                               filename,
 351                               !(st->st_mode & 0222) || read_only,
 352                               crtime,
 353                               timespec_load(&st->st_mtim),
 354                               ret);
 355                 if (r < 0)
 356                         return r;
 357
 358                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 359                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 360
 361                 return 0;
 362
 363         } else if (S_ISBLK(st->st_mode)) {
 364                 _cleanup_close_ int block_fd = -1;
 365                 uint64_t size = UINT64_MAX;
 366
 367                 /* A block device */
 368
 369                 if (!ret)
 370                         return 0;
 371
 372                 if (!pretty) {
 373                         r = extract_pretty(filename, NULL, &pretty_buffer);
 374                         if (r < 0)
 375                                 return r;
 376
 377                         pretty = pretty_buffer;
 378                 }
 379
 380                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 381                 if (block_fd < 0)
 382                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 383                 else {
 384                         /* Refresh stat data after opening the node */
 385                         if (fstat(block_fd, &stbuf) < 0)
 386                                 return -errno;
 387                         st = &stbuf;
 388
 389                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 390                                 return -ENOTTY;
 391
 392                         if (!read_only) {
 393                                 int state = 0;
 394
 395                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 396                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 397                                 else if (state)
 398                                         read_only = true;
 399                         }
 400
 401                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 402                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 403
 404                         block_fd = safe_close(block_fd);
 405                 }
 406
 407                 r = image_new(IMAGE_BLOCK,
 408                               pretty,
 409                               path,
 410                               filename,
 411                               !(st->st_mode & 0222) || read_only,
 412                               0,
 413                               0,
 414                               ret);
 415                 if (r < 0)
 416                         return r;
 417
 418                 if (!IN_SET(size, 0, UINT64_MAX))
 419                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 420
 421                 return 0;
 422         }
 423
 424         return -EMEDIUMTYPE;
 425 }
 426
 427 int image_find(ImageClass class,
 428                const char *name,
 429                const char *root,
 430                Image **ret) {
 431
 432         const char *path;
 433         int r;
 434
 435         assert(class >= 0);
 436         assert(class < _IMAGE_CLASS_MAX);
 437         assert(name);
 438
 439         /* There are no images with invalid names */
 440         if (!image_name_is_valid(name))
 441                 return -ENOENT;
 442
 443         NULSTR_FOREACH(path, image_search_path[class]) {
 444                 _cleanup_free_ char *resolved = NULL;
 445                 _cleanup_closedir_ DIR *d = NULL;
 446                 struct stat st;
 447                 int flags;
 448
 449                 r = chase_symlinks_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 450                 if (r == -ENOENT)
 451                         continue;
 452                 if (r < 0)
 453                         return r;
 454
 455                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 456                  * to symlink block devices into the search path. (For now, we disable that when operating
 457                  * relative to some root directory.) */
 458                 flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 459                 if (fstatat(dirfd(d), name, &st, flags) < 0) {
 460                         _cleanup_free_ char *raw = NULL;
 461
 462                         if (errno != ENOENT)
 463                                 return -errno;
 464
 465                         raw = strjoin(name, ".raw");
 466                         if (!raw)
 467                                 return -ENOMEM;
 468
 469                         if (fstatat(dirfd(d), raw, &st, flags) < 0) {
 470                                 if (errno == ENOENT)
 471                                         continue;
 472
 473                                 return -errno;
 474                         }
 475
 476                         if (!S_ISREG(st.st_mode))
 477                                 continue;
 478
 479                         r = image_make(name, dirfd(d), resolved, raw, &st, ret);
 480
 481                 } else {
 482                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 483                                 continue;
 484
 485                         r = image_make(name, dirfd(d), resolved, name, &st, ret);
 486                 }
 487                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 488                         continue;
 489                 if (r < 0)
 490                         return r;
 491
 492                 if (ret)
 493                         (*ret)->discoverable = true;
 494
 495                 return 1;
 496         }
 497
 498         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 499                 r = image_make(".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret);
 500                 if (r < 0)
 501                         return r;
 502
 503                 if (ret)
 504                         (*ret)->discoverable = true;
 505
 506                 return r;
 507         }
 508
 509         return -ENOENT;
 510 };
 511
 512 int image_from_path(const char *path, Image **ret) {
 513
 514         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 515          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 516          * overridden by another, different image earlier in the search path */
 517
 518         if (path_equal(path, "/"))
 519                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 520
 521         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 522 }
 523
 524 int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) {
 525         if (image_name_is_valid(name_or_path))
 526                 return image_find(class, name_or_path, root, ret);
 527
 528         return image_from_path(name_or_path, ret);
 529 }
 530
 531 int image_discover(
 532                 ImageClass class,
 533                 const char *root,
 534                 Hashmap *h) {
 535
 536         const char *path;
 537         int r;
 538
 539         assert(class >= 0);
 540         assert(class < _IMAGE_CLASS_MAX);
 541         assert(h);
 542
 543         NULSTR_FOREACH(path, image_search_path[class]) {
 544                 _cleanup_free_ char *resolved = NULL;
 545                 _cleanup_closedir_ DIR *d = NULL;
 546
 547                 r = chase_symlinks_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 548                 if (r == -ENOENT)
 549                         continue;
 550                 if (r < 0)
 551                         return r;
 552
 553                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 554                         _cleanup_(image_unrefp) Image *image = NULL;
 555                         _cleanup_free_ char *truncated = NULL;
 556                         const char *pretty;
 557                         struct stat st;
 558                         int flags;
 559
 560                         if (dot_or_dot_dot(de->d_name))
 561                                 continue;
 562
 563                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to
 564                          * permit people to symlink block devices into the search path. */
 565                         flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 566                         if (fstatat(dirfd(d), de->d_name, &st, flags) < 0) {
 567                                 if (errno == ENOENT)
 568                                         continue;
 569
 570                                 return -errno;
 571                         }
 572
 573                         if (S_ISREG(st.st_mode)) {
 574                                 const char *e;
 575
 576                                 e = endswith(de->d_name, ".raw");
 577                                 if (!e)
 578                                         continue;
 579
 580                                 truncated = strndup(de->d_name, e - de->d_name);
 581                                 if (!truncated)
 582                                         return -ENOMEM;
 583
 584                                 pretty = truncated;
 585                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 586                                 pretty = de->d_name;
 587                         else
 588                                 continue;
 589
 590                         if (!image_name_is_valid(pretty))
 591                                 continue;
 592
 593                         if (hashmap_contains(h, pretty))
 594                                 continue;
 595
 596                         r = image_make(pretty, dirfd(d), resolved, de->d_name, &st, &image);
 597                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 598                                 continue;
 599                         if (r < 0)
 600                                 return r;
 601
 602                         image->discoverable = true;
 603
 604                         r = hashmap_put(h, image->name, image);
 605                         if (r < 0)
 606                                 return r;
 607
 608                         image = NULL;
 609                 }
 610         }
 611
 612         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 613                 _cleanup_(image_unrefp) Image *image = NULL;
 614
 615                 r = image_make(".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image);
 616                 if (r < 0)
 617                         return r;
 618
 619                 image->discoverable = true;
 620
 621                 r = hashmap_put(h, image->name, image);
 622                 if (r < 0)
 623                         return r;
 624
 625                 image = NULL;
 626         }
 627
 628         return 0;
 629 }
 630
 631 int image_remove(Image *i) {
 632         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 633         _cleanup_strv_free_ char **settings = NULL;
 634         _cleanup_free_ char *roothash = NULL;
 635         int r;
 636
 637         assert(i);
 638
 639         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 640                 return -EROFS;
 641
 642         settings = image_settings_path(i);
 643         if (!settings)
 644                 return -ENOMEM;
 645
 646         roothash = image_roothash_path(i);
 647         if (!roothash)
 648                 return -ENOMEM;
 649
 650         /* Make sure we don't interfere with a running nspawn */
 651         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 652         if (r < 0)
 653                 return r;
 654
 655         switch (i->type) {
 656
 657         case IMAGE_SUBVOLUME:
 658
 659                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 660                  * big guns */
 661                 if (unlink(i->path) < 0) {
 662                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 663                         if (r < 0)
 664                                 return r;
 665                 }
 666
 667                 break;
 668
 669         case IMAGE_DIRECTORY:
 670                 /* Allow deletion of read-only directories */
 671                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 672                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 673                 if (r < 0)
 674                         return r;
 675
 676                 break;
 677
 678         case IMAGE_BLOCK:
 679
 680                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 681                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 682                  * the thing (it's most likely a symlink after all). */
 683
 684                 if (path_startswith(i->path, "/dev"))
 685                         break;
 686
 687                 _fallthrough_;
 688         case IMAGE_RAW:
 689                 if (unlink(i->path) < 0)
 690                         return -errno;
 691                 break;
 692
 693         default:
 694                 return -EOPNOTSUPP;
 695         }
 696
 697         STRV_FOREACH(j, settings)
 698                 if (unlink(*j) < 0 && errno != ENOENT)
 699                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 700
 701         if (unlink(roothash) < 0 && errno != ENOENT)
 702                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 703
 704         return 0;
 705 }
 706
 707 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 708         _cleanup_free_ char *rs = NULL;
 709         const char *fn;
 710
 711         fn = strjoina(new_name, suffix);
 712
 713         rs = file_in_same_dir(path, fn);
 714         if (!rs)
 715                 return -ENOMEM;
 716
 717         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 718 }
 719
 720 int image_rename(Image *i, const char *new_name) {
 721         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 722         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 723         _cleanup_strv_free_ char **settings = NULL;
 724         unsigned file_attr = 0;
 725         int r;
 726
 727         assert(i);
 728
 729         if (!image_name_is_valid(new_name))
 730                 return -EINVAL;
 731
 732         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 733                 return -EROFS;
 734
 735         settings = image_settings_path(i);
 736         if (!settings)
 737                 return -ENOMEM;
 738
 739         roothash = image_roothash_path(i);
 740         if (!roothash)
 741                 return -ENOMEM;
 742
 743         /* Make sure we don't interfere with a running nspawn */
 744         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 745         if (r < 0)
 746                 return r;
 747
 748         /* Make sure nobody takes the new name, between the time we
 749          * checked it is currently unused in all search paths, and the
 750          * time we take possession of it */
 751         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 752         if (r < 0)
 753                 return r;
 754
 755         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 756         if (r >= 0)
 757                 return -EEXIST;
 758         if (r != -ENOENT)
 759                 return r;
 760
 761         switch (i->type) {
 762
 763         case IMAGE_DIRECTORY:
 764                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 765                 (void) read_attr_path(i->path, &file_attr);
 766
 767                 if (file_attr & FS_IMMUTABLE_FL)
 768                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 769
 770                 _fallthrough_;
 771         case IMAGE_SUBVOLUME:
 772                 new_path = file_in_same_dir(i->path, new_name);
 773                 break;
 774
 775         case IMAGE_BLOCK:
 776
 777                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 778                 if (path_startswith(i->path, "/dev"))
 779                         return -EROFS;
 780
 781                 new_path = file_in_same_dir(i->path, new_name);
 782                 break;
 783
 784         case IMAGE_RAW: {
 785                 const char *fn;
 786
 787                 fn = strjoina(new_name, ".raw");
 788                 new_path = file_in_same_dir(i->path, fn);
 789                 break;
 790         }
 791
 792         default:
 793                 return -EOPNOTSUPP;
 794         }
 795
 796         if (!new_path)
 797                 return -ENOMEM;
 798
 799         nn = strdup(new_name);
 800         if (!nn)
 801                 return -ENOMEM;
 802
 803         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 804         if (r < 0)
 805                 return r;
 806
 807         /* Restore the immutable bit, if it was set before */
 808         if (file_attr & FS_IMMUTABLE_FL)
 809                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 810
 811         free_and_replace(i->path, new_path);
 812         free_and_replace(i->name, nn);
 813
 814         STRV_FOREACH(j, settings) {
 815                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 816                 if (r < 0 && r != -ENOENT)
 817                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 818         }
 819
 820         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 821         if (r < 0 && r != -ENOENT)
 822                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 823
 824         return 0;
 825 }
 826
 827 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 828         _cleanup_free_ char *rs = NULL;
 829         const char *fn;
 830
 831         fn = strjoina(new_name, suffix);
 832
 833         rs = file_in_same_dir(path, fn);
 834         if (!rs)
 835                 return -ENOMEM;
 836
 837         return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK);
 838 }
 839
 840 int image_clone(Image *i, const char *new_name, bool read_only) {
 841         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 842         _cleanup_strv_free_ char **settings = NULL;
 843         _cleanup_free_ char *roothash = NULL;
 844         const char *new_path;
 845         int r;
 846
 847         assert(i);
 848
 849         if (!image_name_is_valid(new_name))
 850                 return -EINVAL;
 851
 852         settings = image_settings_path(i);
 853         if (!settings)
 854                 return -ENOMEM;
 855
 856         roothash = image_roothash_path(i);
 857         if (!roothash)
 858                 return -ENOMEM;
 859
 860         /* Make sure nobody takes the new name, between the time we
 861          * checked it is currently unused in all search paths, and the
 862          * time we take possession of it */
 863         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 864         if (r < 0)
 865                 return r;
 866
 867         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 868         if (r >= 0)
 869                 return -EEXIST;
 870         if (r != -ENOENT)
 871                 return r;
 872
 873         switch (i->type) {
 874
 875         case IMAGE_SUBVOLUME:
 876         case IMAGE_DIRECTORY:
 877                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 878                  * directory. */
 879
 880                 new_path = strjoina("/var/lib/machines/", new_name);
 881
 882                 r = btrfs_subvol_snapshot(i->path, new_path,
 883                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 884                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 885                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 886                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 887                                           BTRFS_SNAPSHOT_RECURSIVE |
 888                                           BTRFS_SNAPSHOT_QUOTA);
 889                 if (r >= 0)
 890                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 891                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 892
 893                 break;
 894
 895         case IMAGE_RAW:
 896                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 897
 898                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME);
 899                 break;
 900
 901         case IMAGE_BLOCK:
 902         default:
 903                 return -EOPNOTSUPP;
 904         }
 905
 906         if (r < 0)
 907                 return r;
 908
 909         STRV_FOREACH(j, settings) {
 910                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 911                 if (r < 0 && r != -ENOENT)
 912                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 913         }
 914
 915         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 916         if (r < 0 && r != -ENOENT)
 917                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 918
 919         return 0;
 920 }
 921
 922 int image_read_only(Image *i, bool b) {
 923         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 924         int r;
 925
 926         assert(i);
 927
 928         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 929                 return -EROFS;
 930
 931         /* Make sure we don't interfere with a running nspawn */
 932         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 933         if (r < 0)
 934                 return r;
 935
 936         switch (i->type) {
 937
 938         case IMAGE_SUBVOLUME:
 939
 940                 /* Note that we set the flag only on the top-level
 941                  * subvolume of the image. */
 942
 943                 r = btrfs_subvol_set_read_only(i->path, b);
 944                 if (r < 0)
 945                         return r;
 946
 947                 break;
 948
 949         case IMAGE_DIRECTORY:
 950                 /* For simple directory trees we cannot use the access
 951                    mode of the top-level directory, since it has an
 952                    effect on the container itself.  However, we can
 953                    use the "immutable" flag, to at least make the
 954                    top-level directory read-only. It's not as good as
 955                    a read-only subvolume, but at least something, and
 956                    we can read the value back. */
 957
 958                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 959                 if (r < 0)
 960                         return r;
 961
 962                 break;
 963
 964         case IMAGE_RAW: {
 965                 struct stat st;
 966
 967                 if (stat(i->path, &st) < 0)
 968                         return -errno;
 969
 970                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 971                         return -errno;
 972
 973                 /* If the images is now read-only, it's a good time to
 974                  * defrag it, given that no write patterns will
 975                  * fragment it again. */
 976                 if (b)
 977                         (void) btrfs_defrag(i->path);
 978                 break;
 979         }
 980
 981         case IMAGE_BLOCK: {
 982                 _cleanup_close_ int fd = -1;
 983                 struct stat st;
 984                 int state = b;
 985
 986                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 987                 if (fd < 0)
 988                         return -errno;
 989
 990                 if (fstat(fd, &st) < 0)
 991                         return -errno;
 992                 if (!S_ISBLK(st.st_mode))
 993                         return -ENOTTY;
 994
 995                 if (ioctl(fd, BLKROSET, &state) < 0)
 996                         return -errno;
 997
 998                 break;
 999         }
1000
1001         default:
1002                 return -EOPNOTSUPP;
1003         }
1004
1005         return 0;
1006 }
1007
1008 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
1009         _cleanup_free_ char *p = NULL;
1010         LockFile t = LOCK_FILE_INIT;
1011         struct stat st;
1012         bool exclusive;
1013         int r;
1014
1015         assert(path);
1016         assert(global);
1017         assert(local);
1018
1019         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1020          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1021          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1022          * correctly. */
1023
1024         if (!path_is_absolute(path))
1025                 return -EINVAL;
1026
1027         switch (operation & (LOCK_SH|LOCK_EX)) {
1028         case LOCK_SH:
1029                 exclusive = false;
1030                 break;
1031         case LOCK_EX:
1032                 exclusive = true;
1033                 break;
1034         default:
1035                 return -EINVAL;
1036         }
1037
1038         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1039                 *local = *global = (LockFile) LOCK_FILE_INIT;
1040                 return 0;
1041         }
1042
1043         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1044          * running off it after all, and we don't want any images to manipulate the host image. We make an
1045          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1046          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1047          * since it means changes made to the host might propagate to the container as they happen (and a
1048          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1049          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1050          * the user to do the right thing with this. */
1051         if (path_equal(path, "/")) {
1052                 if (exclusive)
1053                         return -EBUSY;
1054
1055                 *local = *global = (LockFile) LOCK_FILE_INIT;
1056                 return 0;
1057         }
1058
1059         if (stat(path, &st) >= 0) {
1060                 if (S_ISBLK(st.st_mode))
1061                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1062                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1063                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1064                 else
1065                         return -ENOTTY;
1066                 if (r < 0)
1067                         return -ENOMEM;
1068         }
1069
1070         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1071          * sufficient, since block devices are host local anyway. */
1072         if (!path_startswith(path, "/dev/")) {
1073                 r = make_lock_file_for(path, operation, &t);
1074                 if (r < 0) {
1075                         if (!exclusive && r == -EROFS)
1076                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1077                         else
1078                                 return r;
1079                 }
1080         }
1081
1082         if (p) {
1083                 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1084
1085                 r = make_lock_file(p, operation, global);
1086                 if (r < 0) {
1087                         release_lock_file(&t);
1088                         return r;
1089                 }
1090         } else
1091                 *global = (LockFile) LOCK_FILE_INIT;
1092
1093         *local = t;
1094         return 0;
1095 }
1096
1097 int image_set_limit(Image *i, uint64_t referenced_max) {
1098         assert(i);
1099
1100         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1101                 return -EROFS;
1102
1103         if (i->type != IMAGE_SUBVOLUME)
1104                 return -EOPNOTSUPP;
1105
1106         /* We set the quota both for the subvolume as well as for the
1107          * subtree. The latter is mostly for historical reasons, since
1108          * we didn't use to have a concept of subtree quota, and hence
1109          * only modified the subvolume quota. */
1110
1111         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1112         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1113         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1114 }
1115
1116 int image_read_metadata(Image *i) {
1117         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1118         int r;
1119
1120         assert(i);
1121
1122         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1123         if (r < 0)
1124                 return r;
1125
1126         switch (i->type) {
1127
1128         case IMAGE_SUBVOLUME:
1129         case IMAGE_DIRECTORY: {
1130                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
1131                 sd_id128_t machine_id = SD_ID128_NULL;
1132                 _cleanup_free_ char *hostname = NULL;
1133                 _cleanup_free_ char *path = NULL;
1134
1135                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1136                 if (r < 0 && r != -ENOENT)
1137                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1138                 else if (r >= 0) {
1139                         r = read_etc_hostname(path, &hostname);
1140                         if (r < 0)
1141                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1142                 }
1143
1144                 path = mfree(path);
1145
1146                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1147                 if (r < 0 && r != -ENOENT)
1148                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1149                 else if (r >= 0) {
1150                         _cleanup_close_ int fd = -1;
1151
1152                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1153                         if (fd < 0)
1154                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1155                         else {
1156                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1157                                 if (r < 0)
1158                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1159                         }
1160                 }
1161
1162                 path = mfree(path);
1163
1164                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1165                 if (r < 0 && r != -ENOENT)
1166                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1167                 else if (r >= 0) {
1168                         r = load_env_file_pairs(NULL, path, &machine_info);
1169                         if (r < 0)
1170                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1171                 }
1172
1173                 r = load_os_release_pairs(i->path, &os_release);
1174                 if (r < 0)
1175                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1176
1177                 r = load_extension_release_pairs(i->path, i->name, /* relax_extension_release_check= */ false, &extension_release);
1178                 if (r < 0)
1179                         log_debug_errno(r, "Failed to read extension-release in image, ignoring: %m");
1180
1181                 free_and_replace(i->hostname, hostname);
1182                 i->machine_id = machine_id;
1183                 strv_free_and_replace(i->machine_info, machine_info);
1184                 strv_free_and_replace(i->os_release, os_release);
1185                 strv_free_and_replace(i->extension_release, extension_release);
1186
1187                 break;
1188         }
1189
1190         case IMAGE_RAW:
1191         case IMAGE_BLOCK: {
1192                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1193                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1194
1195                 r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, LOCK_SH, &d);
1196                 if (r < 0)
1197                         return r;
1198
1199                 r = dissect_loop_device(
1200                                 d,
1201                                 NULL, NULL,
1202                                 DISSECT_IMAGE_GENERIC_ROOT |
1203                                 DISSECT_IMAGE_REQUIRE_ROOT |
1204                                 DISSECT_IMAGE_RELAX_VAR_CHECK |
1205                                 DISSECT_IMAGE_READ_ONLY |
1206                                 DISSECT_IMAGE_USR_NO_ROOT,
1207                                 &m);
1208                 if (r < 0)
1209                         return r;
1210
1211                 r = dissected_image_acquire_metadata(m,
1212                                                      DISSECT_IMAGE_VALIDATE_OS |
1213                                                      DISSECT_IMAGE_VALIDATE_OS_EXT);
1214                 if (r < 0)
1215                         return r;
1216
1217                 free_and_replace(i->hostname, m->hostname);
1218                 i->machine_id = m->machine_id;
1219                 strv_free_and_replace(i->machine_info, m->machine_info);
1220                 strv_free_and_replace(i->os_release, m->os_release);
1221                 strv_free_and_replace(i->extension_release, m->extension_release);
1222
1223                 break;
1224         }
1225
1226         default:
1227                 return -EOPNOTSUPP;
1228         }
1229
1230         i->metadata_valid = true;
1231
1232         return 0;
1233 }
1234
1235 int image_name_lock(const char *name, int operation, LockFile *ret) {
1236         const char *p;
1237
1238         assert(name);
1239         assert(ret);
1240
1241         /* Locks an image name, regardless of the precise path used. */
1242
1243         if (streq(name, ".host"))
1244                 return -EBUSY;
1245
1246         if (!image_name_is_valid(name))
1247                 return -EINVAL;
1248
1249         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1250                 *ret = (LockFile) LOCK_FILE_INIT;
1251                 return 0;
1252         }
1253
1254         (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1255
1256         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1257         return make_lock_file(p, operation, ret);
1258 }
1259
1260 bool image_in_search_path(
1261                 ImageClass class,
1262                 const char *root,
1263                 const char *image) {
1264
1265         const char *path;
1266
1267         assert(image);
1268
1269         NULSTR_FOREACH(path, image_search_path[class]) {
1270                 const char *p, *q;
1271                 size_t k;
1272
1273                 if (!empty_or_root(root)) {
1274                         q = path_startswith(path, root);
1275                         if (!q)
1276                                 continue;
1277                 } else
1278                         q = path;
1279
1280                 p = path_startswith(q, path);
1281                 if (!p)
1282                         continue;
1283
1284                 /* Make sure there's a filename following */
1285                 k = strcspn(p, "/");
1286                 if (k == 0)
1287                         continue;
1288
1289                 p += k;
1290
1291                 /* Accept trailing slashes */
1292                 if (p[strspn(p, "/")] == 0)
1293                         return true;
1294
1295         }
1296
1297         return false;
1298 }
1299
1300 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1301         [IMAGE_DIRECTORY] = "directory",
1302         [IMAGE_SUBVOLUME] = "subvolume",
1303         [IMAGE_RAW] = "raw",
1304         [IMAGE_BLOCK] = "block",
1305 };
1306
1307 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);