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, *s;
  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         char **j;
 636         int r;
 637
 638         assert(i);
 639
 640         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 641                 return -EROFS;
 642
 643         settings = image_settings_path(i);
 644         if (!settings)
 645                 return -ENOMEM;
 646
 647         roothash = image_roothash_path(i);
 648         if (!roothash)
 649                 return -ENOMEM;
 650
 651         /* Make sure we don't interfere with a running nspawn */
 652         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 653         if (r < 0)
 654                 return r;
 655
 656         switch (i->type) {
 657
 658         case IMAGE_SUBVOLUME:
 659
 660                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 661                  * big guns */
 662                 if (unlink(i->path) < 0) {
 663                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 664                         if (r < 0)
 665                                 return r;
 666                 }
 667
 668                 break;
 669
 670         case IMAGE_DIRECTORY:
 671                 /* Allow deletion of read-only directories */
 672                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 673                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 674                 if (r < 0)
 675                         return r;
 676
 677                 break;
 678
 679         case IMAGE_BLOCK:
 680
 681                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 682                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 683                  * the thing (it's most likely a symlink after all). */
 684
 685                 if (path_startswith(i->path, "/dev"))
 686                         break;
 687
 688                 _fallthrough_;
 689         case IMAGE_RAW:
 690                 if (unlink(i->path) < 0)
 691                         return -errno;
 692                 break;
 693
 694         default:
 695                 return -EOPNOTSUPP;
 696         }
 697
 698         STRV_FOREACH(j, settings) {
 699                 if (unlink(*j) < 0 && errno != ENOENT)
 700                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 701         }
 702
 703         if (unlink(roothash) < 0 && errno != ENOENT)
 704                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 705
 706         return 0;
 707 }
 708
 709 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 710         _cleanup_free_ char *rs = NULL;
 711         const char *fn;
 712
 713         fn = strjoina(new_name, suffix);
 714
 715         rs = file_in_same_dir(path, fn);
 716         if (!rs)
 717                 return -ENOMEM;
 718
 719         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 720 }
 721
 722 int image_rename(Image *i, const char *new_name) {
 723         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 724         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 725         _cleanup_strv_free_ char **settings = NULL;
 726         unsigned file_attr = 0;
 727         char **j;
 728         int r;
 729
 730         assert(i);
 731
 732         if (!image_name_is_valid(new_name))
 733                 return -EINVAL;
 734
 735         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 736                 return -EROFS;
 737
 738         settings = image_settings_path(i);
 739         if (!settings)
 740                 return -ENOMEM;
 741
 742         roothash = image_roothash_path(i);
 743         if (!roothash)
 744                 return -ENOMEM;
 745
 746         /* Make sure we don't interfere with a running nspawn */
 747         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 748         if (r < 0)
 749                 return r;
 750
 751         /* Make sure nobody takes the new name, between the time we
 752          * checked it is currently unused in all search paths, and the
 753          * time we take possession of it */
 754         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 755         if (r < 0)
 756                 return r;
 757
 758         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 759         if (r >= 0)
 760                 return -EEXIST;
 761         if (r != -ENOENT)
 762                 return r;
 763
 764         switch (i->type) {
 765
 766         case IMAGE_DIRECTORY:
 767                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 768                 (void) read_attr_path(i->path, &file_attr);
 769
 770                 if (file_attr & FS_IMMUTABLE_FL)
 771                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 772
 773                 _fallthrough_;
 774         case IMAGE_SUBVOLUME:
 775                 new_path = file_in_same_dir(i->path, new_name);
 776                 break;
 777
 778         case IMAGE_BLOCK:
 779
 780                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 781                 if (path_startswith(i->path, "/dev"))
 782                         return -EROFS;
 783
 784                 new_path = file_in_same_dir(i->path, new_name);
 785                 break;
 786
 787         case IMAGE_RAW: {
 788                 const char *fn;
 789
 790                 fn = strjoina(new_name, ".raw");
 791                 new_path = file_in_same_dir(i->path, fn);
 792                 break;
 793         }
 794
 795         default:
 796                 return -EOPNOTSUPP;
 797         }
 798
 799         if (!new_path)
 800                 return -ENOMEM;
 801
 802         nn = strdup(new_name);
 803         if (!nn)
 804                 return -ENOMEM;
 805
 806         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 807         if (r < 0)
 808                 return r;
 809
 810         /* Restore the immutable bit, if it was set before */
 811         if (file_attr & FS_IMMUTABLE_FL)
 812                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 813
 814         free_and_replace(i->path, new_path);
 815         free_and_replace(i->name, nn);
 816
 817         STRV_FOREACH(j, settings) {
 818                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 819                 if (r < 0 && r != -ENOENT)
 820                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 821         }
 822
 823         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 824         if (r < 0 && r != -ENOENT)
 825                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 826
 827         return 0;
 828 }
 829
 830 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 831         _cleanup_free_ char *rs = NULL;
 832         const char *fn;
 833
 834         fn = strjoina(new_name, suffix);
 835
 836         rs = file_in_same_dir(path, fn);
 837         if (!rs)
 838                 return -ENOMEM;
 839
 840         return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK);
 841 }
 842
 843 int image_clone(Image *i, const char *new_name, bool read_only) {
 844         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 845         _cleanup_strv_free_ char **settings = NULL;
 846         _cleanup_free_ char *roothash = NULL;
 847         const char *new_path;
 848         char **j;
 849         int r;
 850
 851         assert(i);
 852
 853         if (!image_name_is_valid(new_name))
 854                 return -EINVAL;
 855
 856         settings = image_settings_path(i);
 857         if (!settings)
 858                 return -ENOMEM;
 859
 860         roothash = image_roothash_path(i);
 861         if (!roothash)
 862                 return -ENOMEM;
 863
 864         /* Make sure nobody takes the new name, between the time we
 865          * checked it is currently unused in all search paths, and the
 866          * time we take possession of it */
 867         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 868         if (r < 0)
 869                 return r;
 870
 871         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 872         if (r >= 0)
 873                 return -EEXIST;
 874         if (r != -ENOENT)
 875                 return r;
 876
 877         switch (i->type) {
 878
 879         case IMAGE_SUBVOLUME:
 880         case IMAGE_DIRECTORY:
 881                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 882                  * directory. */
 883
 884                 new_path = strjoina("/var/lib/machines/", new_name);
 885
 886                 r = btrfs_subvol_snapshot(i->path, new_path,
 887                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 888                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 889                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 890                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 891                                           BTRFS_SNAPSHOT_RECURSIVE |
 892                                           BTRFS_SNAPSHOT_QUOTA);
 893                 if (r >= 0)
 894                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 895                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 896
 897                 break;
 898
 899         case IMAGE_RAW:
 900                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 901
 902                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME);
 903                 break;
 904
 905         case IMAGE_BLOCK:
 906         default:
 907                 return -EOPNOTSUPP;
 908         }
 909
 910         if (r < 0)
 911                 return r;
 912
 913         STRV_FOREACH(j, settings) {
 914                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 915                 if (r < 0 && r != -ENOENT)
 916                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 917         }
 918
 919         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 920         if (r < 0 && r != -ENOENT)
 921                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 922
 923         return 0;
 924 }
 925
 926 int image_read_only(Image *i, bool b) {
 927         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 928         int r;
 929
 930         assert(i);
 931
 932         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 933                 return -EROFS;
 934
 935         /* Make sure we don't interfere with a running nspawn */
 936         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 937         if (r < 0)
 938                 return r;
 939
 940         switch (i->type) {
 941
 942         case IMAGE_SUBVOLUME:
 943
 944                 /* Note that we set the flag only on the top-level
 945                  * subvolume of the image. */
 946
 947                 r = btrfs_subvol_set_read_only(i->path, b);
 948                 if (r < 0)
 949                         return r;
 950
 951                 break;
 952
 953         case IMAGE_DIRECTORY:
 954                 /* For simple directory trees we cannot use the access
 955                    mode of the top-level directory, since it has an
 956                    effect on the container itself.  However, we can
 957                    use the "immutable" flag, to at least make the
 958                    top-level directory read-only. It's not as good as
 959                    a read-only subvolume, but at least something, and
 960                    we can read the value back. */
 961
 962                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 963                 if (r < 0)
 964                         return r;
 965
 966                 break;
 967
 968         case IMAGE_RAW: {
 969                 struct stat st;
 970
 971                 if (stat(i->path, &st) < 0)
 972                         return -errno;
 973
 974                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 975                         return -errno;
 976
 977                 /* If the images is now read-only, it's a good time to
 978                  * defrag it, given that no write patterns will
 979                  * fragment it again. */
 980                 if (b)
 981                         (void) btrfs_defrag(i->path);
 982                 break;
 983         }
 984
 985         case IMAGE_BLOCK: {
 986                 _cleanup_close_ int fd = -1;
 987                 struct stat st;
 988                 int state = b;
 989
 990                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 991                 if (fd < 0)
 992                         return -errno;
 993
 994                 if (fstat(fd, &st) < 0)
 995                         return -errno;
 996                 if (!S_ISBLK(st.st_mode))
 997                         return -ENOTTY;
 998
 999                 if (ioctl(fd, BLKROSET, &state) < 0)
1000                         return -errno;
1001
1002                 break;
1003         }
1004
1005         default:
1006                 return -EOPNOTSUPP;
1007         }
1008
1009         return 0;
1010 }
1011
1012 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
1013         _cleanup_free_ char *p = NULL;
1014         LockFile t = LOCK_FILE_INIT;
1015         struct stat st;
1016         bool exclusive;
1017         int r;
1018
1019         assert(path);
1020         assert(global);
1021         assert(local);
1022
1023         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1024          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1025          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1026          * correctly. */
1027
1028         if (!path_is_absolute(path))
1029                 return -EINVAL;
1030
1031         switch (operation & (LOCK_SH|LOCK_EX)) {
1032         case LOCK_SH:
1033                 exclusive = false;
1034                 break;
1035         case LOCK_EX:
1036                 exclusive = true;
1037                 break;
1038         default:
1039                 return -EINVAL;
1040         }
1041
1042         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1043                 *local = *global = (LockFile) LOCK_FILE_INIT;
1044                 return 0;
1045         }
1046
1047         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1048          * running off it after all, and we don't want any images to manipulate the host image. We make an
1049          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1050          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1051          * since it means changes made to the host might propagate to the container as they happen (and a
1052          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1053          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1054          * the user to do the right thing with this. */
1055         if (path_equal(path, "/")) {
1056                 if (exclusive)
1057                         return -EBUSY;
1058
1059                 *local = *global = (LockFile) LOCK_FILE_INIT;
1060                 return 0;
1061         }
1062
1063         if (stat(path, &st) >= 0) {
1064                 if (S_ISBLK(st.st_mode))
1065                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1066                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1067                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1068                 else
1069                         return -ENOTTY;
1070                 if (r < 0)
1071                         return -ENOMEM;
1072         }
1073
1074         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1075          * sufficient, since block devices are host local anyway. */
1076         if (!path_startswith(path, "/dev/")) {
1077                 r = make_lock_file_for(path, operation, &t);
1078                 if (r < 0) {
1079                         if (!exclusive && r == -EROFS)
1080                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1081                         else
1082                                 return r;
1083                 }
1084         }
1085
1086         if (p) {
1087                 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1088
1089                 r = make_lock_file(p, operation, global);
1090                 if (r < 0) {
1091                         release_lock_file(&t);
1092                         return r;
1093                 }
1094         } else
1095                 *global = (LockFile) LOCK_FILE_INIT;
1096
1097         *local = t;
1098         return 0;
1099 }
1100
1101 int image_set_limit(Image *i, uint64_t referenced_max) {
1102         assert(i);
1103
1104         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1105                 return -EROFS;
1106
1107         if (i->type != IMAGE_SUBVOLUME)
1108                 return -EOPNOTSUPP;
1109
1110         /* We set the quota both for the subvolume as well as for the
1111          * subtree. The latter is mostly for historical reasons, since
1112          * we didn't use to have a concept of subtree quota, and hence
1113          * only modified the subvolume quota. */
1114
1115         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1116         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1117         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1118 }
1119
1120 int image_read_metadata(Image *i) {
1121         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1122         int r;
1123
1124         assert(i);
1125
1126         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1127         if (r < 0)
1128                 return r;
1129
1130         switch (i->type) {
1131
1132         case IMAGE_SUBVOLUME:
1133         case IMAGE_DIRECTORY: {
1134                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
1135                 sd_id128_t machine_id = SD_ID128_NULL;
1136                 _cleanup_free_ char *hostname = NULL;
1137                 _cleanup_free_ char *path = NULL;
1138
1139                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1140                 if (r < 0 && r != -ENOENT)
1141                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1142                 else if (r >= 0) {
1143                         r = read_etc_hostname(path, &hostname);
1144                         if (r < 0)
1145                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1146                 }
1147
1148                 path = mfree(path);
1149
1150                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1151                 if (r < 0 && r != -ENOENT)
1152                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1153                 else if (r >= 0) {
1154                         _cleanup_close_ int fd = -1;
1155
1156                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1157                         if (fd < 0)
1158                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1159                         else {
1160                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1161                                 if (r < 0)
1162                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1163                         }
1164                 }
1165
1166                 path = mfree(path);
1167
1168                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1169                 if (r < 0 && r != -ENOENT)
1170                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1171                 else if (r >= 0) {
1172                         r = load_env_file_pairs(NULL, path, &machine_info);
1173                         if (r < 0)
1174                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1175                 }
1176
1177                 r = load_os_release_pairs(i->path, &os_release);
1178                 if (r < 0)
1179                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1180
1181                 r = load_extension_release_pairs(i->path, i->name, &extension_release);
1182                 if (r < 0)
1183                         log_debug_errno(r, "Failed to read extension-release in image, ignoring: %m");
1184
1185                 free_and_replace(i->hostname, hostname);
1186                 i->machine_id = machine_id;
1187                 strv_free_and_replace(i->machine_info, machine_info);
1188                 strv_free_and_replace(i->os_release, os_release);
1189                 strv_free_and_replace(i->extension_release, extension_release);
1190
1191                 break;
1192         }
1193
1194         case IMAGE_RAW:
1195         case IMAGE_BLOCK: {
1196                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1197                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1198
1199                 r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, &d);
1200                 if (r < 0)
1201                         return r;
1202
1203                 r = dissect_image(
1204                                 d->fd,
1205                                 NULL, NULL,
1206                                 d->diskseq,
1207                                 d->uevent_seqnum_not_before,
1208                                 d->timestamp_not_before,
1209                                 DISSECT_IMAGE_GENERIC_ROOT |
1210                                 DISSECT_IMAGE_REQUIRE_ROOT |
1211                                 DISSECT_IMAGE_RELAX_VAR_CHECK |
1212                                 DISSECT_IMAGE_READ_ONLY |
1213                                 DISSECT_IMAGE_USR_NO_ROOT,
1214                                 &m);
1215                 if (r < 0)
1216                         return r;
1217
1218                 r = dissected_image_acquire_metadata(m,
1219                                                      DISSECT_IMAGE_VALIDATE_OS |
1220                                                      DISSECT_IMAGE_VALIDATE_OS_EXT);
1221                 if (r < 0)
1222                         return r;
1223
1224                 free_and_replace(i->hostname, m->hostname);
1225                 i->machine_id = m->machine_id;
1226                 strv_free_and_replace(i->machine_info, m->machine_info);
1227                 strv_free_and_replace(i->os_release, m->os_release);
1228                 strv_free_and_replace(i->extension_release, m->extension_release);
1229
1230                 break;
1231         }
1232
1233         default:
1234                 return -EOPNOTSUPP;
1235         }
1236
1237         i->metadata_valid = true;
1238
1239         return 0;
1240 }
1241
1242 int image_name_lock(const char *name, int operation, LockFile *ret) {
1243         const char *p;
1244
1245         assert(name);
1246         assert(ret);
1247
1248         /* Locks an image name, regardless of the precise path used. */
1249
1250         if (streq(name, ".host"))
1251                 return -EBUSY;
1252
1253         if (!image_name_is_valid(name))
1254                 return -EINVAL;
1255
1256         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1257                 *ret = (LockFile) LOCK_FILE_INIT;
1258                 return 0;
1259         }
1260
1261         (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1262
1263         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1264         return make_lock_file(p, operation, ret);
1265 }
1266
1267 bool image_in_search_path(
1268                 ImageClass class,
1269                 const char *root,
1270                 const char *image) {
1271
1272         const char *path;
1273
1274         assert(image);
1275
1276         NULSTR_FOREACH(path, image_search_path[class]) {
1277                 const char *p, *q;
1278                 size_t k;
1279
1280                 if (!empty_or_root(root)) {
1281                         q = path_startswith(path, root);
1282                         if (!q)
1283                                 continue;
1284                 } else
1285                         q = path;
1286
1287                 p = path_startswith(q, path);
1288                 if (!p)
1289                         continue;
1290
1291                 /* Make sure there's a filename following */
1292                 k = strcspn(p, "/");
1293                 if (k == 0)
1294                         continue;
1295
1296                 p += k;
1297
1298                 /* Accept trailing slashes */
1299                 if (p[strspn(p, "/")] == 0)
1300                         return true;
1301
1302         }
1303
1304         return false;
1305 }
1306
1307 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1308         [IMAGE_DIRECTORY] = "directory",
1309         [IMAGE_SUBVOLUME] = "subvolume",
1310         [IMAGE_RAW] = "raw",
1311         [IMAGE_BLOCK] = "block",
1312 };
1313
1314 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);