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                 ImageClass c,
 127                 const char *pretty,
 128                 const char *path,
 129                 const char *filename,
 130                 bool read_only,
 131                 usec_t crtime,
 132                 usec_t mtime,
 133                 Image **ret) {
 134
 135         _cleanup_(image_unrefp) Image *i = NULL;
 136
 137         assert(t >= 0);
 138         assert(t < _IMAGE_TYPE_MAX);
 139         assert(pretty);
 140         assert(filename);
 141         assert(ret);
 142
 143         i = new(Image, 1);
 144         if (!i)
 145                 return -ENOMEM;
 146
 147         *i = (Image) {
 148                 .n_ref = 1,
 149                 .type = t,
 150                 .class = c,
 151                 .read_only = read_only,
 152                 .crtime = crtime,
 153                 .mtime = mtime,
 154                 .usage = UINT64_MAX,
 155                 .usage_exclusive = UINT64_MAX,
 156                 .limit = UINT64_MAX,
 157                 .limit_exclusive = UINT64_MAX,
 158         };
 159
 160         i->name = strdup(pretty);
 161         if (!i->name)
 162                 return -ENOMEM;
 163
 164         i->path = path_join(path, filename);
 165         if (!i->path)
 166                 return -ENOMEM;
 167
 168         path_simplify(i->path);
 169
 170         *ret = TAKE_PTR(i);
 171
 172         return 0;
 173 }
 174
 175 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 176         _cleanup_free_ char *name = NULL;
 177         const char *p;
 178         size_t n;
 179
 180         assert(path);
 181         assert(ret);
 182
 183         p = last_path_component(path);
 184         n = strcspn(p, "/");
 185
 186         name = strndup(p, n);
 187         if (!name)
 188                 return -ENOMEM;
 189
 190         if (suffix) {
 191                 char *e;
 192
 193                 e = endswith(name, suffix);
 194                 if (!e)
 195                         return -EINVAL;
 196
 197                 *e = 0;
 198         }
 199
 200         if (!image_name_is_valid(name))
 201                 return -EINVAL;
 202
 203         *ret = TAKE_PTR(name);
 204         return 0;
 205 }
 206
 207 static int image_make(
 208                 ImageClass c,
 209                 const char *pretty,
 210                 int dfd,
 211                 const char *path,
 212                 const char *filename,
 213                 const struct stat *st,
 214                 Image **ret) {
 215
 216         _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL;
 217         struct stat stbuf;
 218         bool read_only;
 219         int r;
 220
 221         assert(dfd >= 0 || dfd == AT_FDCWD);
 222         assert(path || dfd == AT_FDCWD);
 223         assert(filename);
 224
 225         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 226          * devices into /var/lib/machines/, and treat them normally.
 227          *
 228          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 229          * recognize. */
 230
 231         if (!st) {
 232                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 233                         return -errno;
 234
 235                 st = &stbuf;
 236         }
 237
 238         if (!path) {
 239                 if (dfd == AT_FDCWD)
 240                         (void) safe_getcwd(&parent);
 241                 else
 242                         (void) fd_get_path(dfd, &parent);
 243         }
 244
 245         read_only =
 246                 (path && path_startswith(path, "/usr")) ||
 247                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 248
 249         if (S_ISDIR(st->st_mode)) {
 250                 _cleanup_close_ int fd = -EBADF;
 251                 unsigned file_attr = 0;
 252                 usec_t crtime = 0;
 253
 254                 if (!ret)
 255                         return 0;
 256
 257                 if (!pretty) {
 258                         r = extract_pretty(filename, NULL, &pretty_buffer);
 259                         if (r < 0)
 260                                 return r;
 261
 262                         pretty = pretty_buffer;
 263                 }
 264
 265                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 266                 if (fd < 0)
 267                         return -errno;
 268
 269                 if (btrfs_might_be_subvol(st)) {
 270
 271                         r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
 272                         if (r < 0)
 273                                 return r;
 274                         if (r) {
 275                                 BtrfsSubvolInfo info;
 276
 277                                 /* It's a btrfs subvolume */
 278
 279                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 280                                 if (r < 0)
 281                                         return r;
 282
 283                                 r = image_new(IMAGE_SUBVOLUME,
 284                                               c,
 285                                               pretty,
 286                                               path,
 287                                               filename,
 288                                               info.read_only || read_only,
 289                                               info.otime,
 290                                               0,
 291                                               ret);
 292                                 if (r < 0)
 293                                         return r;
 294
 295                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 296                                         BtrfsQuotaInfo quota;
 297
 298                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 299                                         if (r >= 0) {
 300                                                 (*ret)->usage = quota.referenced;
 301                                                 (*ret)->usage_exclusive = quota.exclusive;
 302
 303                                                 (*ret)->limit = quota.referenced_max;
 304                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 305                                         }
 306                                 }
 307
 308                                 return 0;
 309                         }
 310                 }
 311
 312                 /* Get directory creation time (not available everywhere, but that's OK */
 313                 (void) fd_getcrtime(fd, &crtime);
 314
 315                 /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not
 316                  * supported everywhere we ignore failures. */
 317                 (void) read_attr_fd(fd, &file_attr);
 318
 319                 /* It's just a normal directory. */
 320                 r = image_new(IMAGE_DIRECTORY,
 321                               c,
 322                               pretty,
 323                               path,
 324                               filename,
 325                               read_only || (file_attr & FS_IMMUTABLE_FL),
 326                               crtime,
 327                               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 */
 328                               ret);
 329                 if (r < 0)
 330                         return r;
 331
 332                 return 0;
 333
 334         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 335                 usec_t crtime = 0;
 336
 337                 /* It's a RAW disk image */
 338
 339                 if (!ret)
 340                         return 0;
 341
 342                 (void) fd_getcrtime_at(dfd, filename, AT_SYMLINK_FOLLOW, &crtime);
 343
 344                 if (!pretty) {
 345                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 346                         if (r < 0)
 347                                 return r;
 348
 349                         pretty = pretty_buffer;
 350                 }
 351
 352                 r = image_new(IMAGE_RAW,
 353                               c,
 354                               pretty,
 355                               path,
 356                               filename,
 357                               !(st->st_mode & 0222) || read_only,
 358                               crtime,
 359                               timespec_load(&st->st_mtim),
 360                               ret);
 361                 if (r < 0)
 362                         return r;
 363
 364                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 365                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 366
 367                 return 0;
 368
 369         } else if (S_ISBLK(st->st_mode)) {
 370                 _cleanup_close_ int block_fd = -EBADF;
 371                 uint64_t size = UINT64_MAX;
 372
 373                 /* A block device */
 374
 375                 if (!ret)
 376                         return 0;
 377
 378                 if (!pretty) {
 379                         r = extract_pretty(filename, NULL, &pretty_buffer);
 380                         if (r < 0)
 381                                 return r;
 382
 383                         pretty = pretty_buffer;
 384                 }
 385
 386                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 387                 if (block_fd < 0)
 388                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 389                 else {
 390                         /* Refresh stat data after opening the node */
 391                         if (fstat(block_fd, &stbuf) < 0)
 392                                 return -errno;
 393                         st = &stbuf;
 394
 395                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 396                                 return -ENOTTY;
 397
 398                         if (!read_only) {
 399                                 int state = 0;
 400
 401                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 402                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 403                                 else if (state)
 404                                         read_only = true;
 405                         }
 406
 407                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 408                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 409
 410                         block_fd = safe_close(block_fd);
 411                 }
 412
 413                 r = image_new(IMAGE_BLOCK,
 414                               c,
 415                               pretty,
 416                               path,
 417                               filename,
 418                               !(st->st_mode & 0222) || read_only,
 419                               0,
 420                               0,
 421                               ret);
 422                 if (r < 0)
 423                         return r;
 424
 425                 if (!IN_SET(size, 0, UINT64_MAX))
 426                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 427
 428                 return 0;
 429         }
 430
 431         return -EMEDIUMTYPE;
 432 }
 433
 434 int image_find(ImageClass class,
 435                const char *name,
 436                const char *root,
 437                Image **ret) {
 438
 439         int r;
 440
 441         assert(class >= 0);
 442         assert(class < _IMAGE_CLASS_MAX);
 443         assert(name);
 444
 445         /* There are no images with invalid names */
 446         if (!image_name_is_valid(name))
 447                 return -ENOENT;
 448
 449         NULSTR_FOREACH(path, image_search_path[class]) {
 450                 _cleanup_free_ char *resolved = NULL;
 451                 _cleanup_closedir_ DIR *d = NULL;
 452                 struct stat st;
 453                 int flags;
 454
 455                 r = chase_symlinks_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 456                 if (r == -ENOENT)
 457                         continue;
 458                 if (r < 0)
 459                         return r;
 460
 461                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 462                  * to symlink block devices into the search path. (For now, we disable that when operating
 463                  * relative to some root directory.) */
 464                 flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 465                 if (fstatat(dirfd(d), name, &st, flags) < 0) {
 466                         _cleanup_free_ char *raw = NULL;
 467
 468                         if (errno != ENOENT)
 469                                 return -errno;
 470
 471                         raw = strjoin(name, ".raw");
 472                         if (!raw)
 473                                 return -ENOMEM;
 474
 475                         if (fstatat(dirfd(d), raw, &st, flags) < 0) {
 476                                 if (errno == ENOENT)
 477                                         continue;
 478
 479                                 return -errno;
 480                         }
 481
 482                         if (!S_ISREG(st.st_mode))
 483                                 continue;
 484
 485                         r = image_make(class, name, dirfd(d), resolved, raw, &st, ret);
 486
 487                 } else {
 488                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 489                                 continue;
 490
 491                         r = image_make(class, name, dirfd(d), resolved, name, &st, ret);
 492                 }
 493                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 494                         continue;
 495                 if (r < 0)
 496                         return r;
 497
 498                 if (ret)
 499                         (*ret)->discoverable = true;
 500
 501                 return 1;
 502         }
 503
 504         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 505                 r = image_make(class, ".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret);
 506                 if (r < 0)
 507                         return r;
 508
 509                 if (ret)
 510                         (*ret)->discoverable = true;
 511
 512                 return r;
 513         }
 514
 515         return -ENOENT;
 516 };
 517
 518 int image_from_path(const char *path, Image **ret) {
 519
 520         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 521          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 522          * overridden by another, different image earlier in the search path */
 523
 524         if (path_equal(path, "/"))
 525                 return image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, "/", NULL, ret);
 526
 527         return image_make(_IMAGE_CLASS_INVALID, NULL, AT_FDCWD, NULL, path, NULL, ret);
 528 }
 529
 530 int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) {
 531         if (image_name_is_valid(name_or_path))
 532                 return image_find(class, name_or_path, root, ret);
 533
 534         return image_from_path(name_or_path, ret);
 535 }
 536
 537 int image_discover(
 538                 ImageClass class,
 539                 const char *root,
 540                 Hashmap *h) {
 541
 542         int r;
 543
 544         assert(class >= 0);
 545         assert(class < _IMAGE_CLASS_MAX);
 546         assert(h);
 547
 548         NULSTR_FOREACH(path, image_search_path[class]) {
 549                 _cleanup_free_ char *resolved = NULL;
 550                 _cleanup_closedir_ DIR *d = NULL;
 551
 552                 r = chase_symlinks_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 553                 if (r == -ENOENT)
 554                         continue;
 555                 if (r < 0)
 556                         return r;
 557
 558                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 559                         _cleanup_(image_unrefp) Image *image = NULL;
 560                         _cleanup_free_ char *truncated = NULL;
 561                         const char *pretty;
 562                         struct stat st;
 563                         int flags;
 564
 565                         if (dot_or_dot_dot(de->d_name))
 566                                 continue;
 567
 568                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to
 569                          * permit people to symlink block devices into the search path. */
 570                         flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 571                         if (fstatat(dirfd(d), de->d_name, &st, flags) < 0) {
 572                                 if (errno == ENOENT)
 573                                         continue;
 574
 575                                 return -errno;
 576                         }
 577
 578                         if (S_ISREG(st.st_mode)) {
 579                                 const char *e;
 580
 581                                 e = endswith(de->d_name, ".raw");
 582                                 if (!e)
 583                                         continue;
 584
 585                                 truncated = strndup(de->d_name, e - de->d_name);
 586                                 if (!truncated)
 587                                         return -ENOMEM;
 588
 589                                 pretty = truncated;
 590                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 591                                 pretty = de->d_name;
 592                         else
 593                                 continue;
 594
 595                         if (!image_name_is_valid(pretty))
 596                                 continue;
 597
 598                         if (hashmap_contains(h, pretty))
 599                                 continue;
 600
 601                         r = image_make(class, pretty, dirfd(d), resolved, de->d_name, &st, &image);
 602                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 603                                 continue;
 604                         if (r < 0)
 605                                 return r;
 606
 607                         image->discoverable = true;
 608
 609                         r = hashmap_put(h, image->name, image);
 610                         if (r < 0)
 611                                 return r;
 612
 613                         image = NULL;
 614                 }
 615         }
 616
 617         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 618                 _cleanup_(image_unrefp) Image *image = NULL;
 619
 620                 r = image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image);
 621                 if (r < 0)
 622                         return r;
 623
 624                 image->discoverable = true;
 625
 626                 r = hashmap_put(h, image->name, image);
 627                 if (r < 0)
 628                         return r;
 629
 630                 image = NULL;
 631         }
 632
 633         return 0;
 634 }
 635
 636 int image_remove(Image *i) {
 637         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 638         _cleanup_strv_free_ char **settings = NULL;
 639         _cleanup_free_ char *roothash = NULL;
 640         int r;
 641
 642         assert(i);
 643
 644         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 645                 return -EROFS;
 646
 647         settings = image_settings_path(i);
 648         if (!settings)
 649                 return -ENOMEM;
 650
 651         roothash = image_roothash_path(i);
 652         if (!roothash)
 653                 return -ENOMEM;
 654
 655         /* Make sure we don't interfere with a running nspawn */
 656         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 657         if (r < 0)
 658                 return r;
 659
 660         switch (i->type) {
 661
 662         case IMAGE_SUBVOLUME:
 663
 664                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 665                  * big guns */
 666                 if (unlink(i->path) < 0) {
 667                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 668                         if (r < 0)
 669                                 return r;
 670                 }
 671
 672                 break;
 673
 674         case IMAGE_DIRECTORY:
 675                 /* Allow deletion of read-only directories */
 676                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 677                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 678                 if (r < 0)
 679                         return r;
 680
 681                 break;
 682
 683         case IMAGE_BLOCK:
 684
 685                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 686                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 687                  * the thing (it's most likely a symlink after all). */
 688
 689                 if (path_startswith(i->path, "/dev"))
 690                         break;
 691
 692                 _fallthrough_;
 693         case IMAGE_RAW:
 694                 if (unlink(i->path) < 0)
 695                         return -errno;
 696                 break;
 697
 698         default:
 699                 return -EOPNOTSUPP;
 700         }
 701
 702         STRV_FOREACH(j, settings)
 703                 if (unlink(*j) < 0 && errno != ENOENT)
 704                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 705
 706         if (unlink(roothash) < 0 && errno != ENOENT)
 707                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 708
 709         return 0;
 710 }
 711
 712 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 713         _cleanup_free_ char *rs = NULL;
 714         const char *fn;
 715
 716         fn = strjoina(new_name, suffix);
 717
 718         rs = file_in_same_dir(path, fn);
 719         if (!rs)
 720                 return -ENOMEM;
 721
 722         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 723 }
 724
 725 int image_rename(Image *i, const char *new_name) {
 726         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 727         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 728         _cleanup_strv_free_ char **settings = NULL;
 729         unsigned file_attr = 0;
 730         int r;
 731
 732         assert(i);
 733
 734         if (!image_name_is_valid(new_name))
 735                 return -EINVAL;
 736
 737         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 738                 return -EROFS;
 739
 740         settings = image_settings_path(i);
 741         if (!settings)
 742                 return -ENOMEM;
 743
 744         roothash = image_roothash_path(i);
 745         if (!roothash)
 746                 return -ENOMEM;
 747
 748         /* Make sure we don't interfere with a running nspawn */
 749         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 750         if (r < 0)
 751                 return r;
 752
 753         /* Make sure nobody takes the new name, between the time we
 754          * checked it is currently unused in all search paths, and the
 755          * time we take possession of it */
 756         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 757         if (r < 0)
 758                 return r;
 759
 760         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 761         if (r >= 0)
 762                 return -EEXIST;
 763         if (r != -ENOENT)
 764                 return r;
 765
 766         switch (i->type) {
 767
 768         case IMAGE_DIRECTORY:
 769                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 770                 (void) read_attr_path(i->path, &file_attr);
 771
 772                 if (file_attr & FS_IMMUTABLE_FL)
 773                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 774
 775                 _fallthrough_;
 776         case IMAGE_SUBVOLUME:
 777                 new_path = file_in_same_dir(i->path, new_name);
 778                 break;
 779
 780         case IMAGE_BLOCK:
 781
 782                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 783                 if (path_startswith(i->path, "/dev"))
 784                         return -EROFS;
 785
 786                 new_path = file_in_same_dir(i->path, new_name);
 787                 break;
 788
 789         case IMAGE_RAW: {
 790                 const char *fn;
 791
 792                 fn = strjoina(new_name, ".raw");
 793                 new_path = file_in_same_dir(i->path, fn);
 794                 break;
 795         }
 796
 797         default:
 798                 return -EOPNOTSUPP;
 799         }
 800
 801         if (!new_path)
 802                 return -ENOMEM;
 803
 804         nn = strdup(new_name);
 805         if (!nn)
 806                 return -ENOMEM;
 807
 808         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 809         if (r < 0)
 810                 return r;
 811
 812         /* Restore the immutable bit, if it was set before */
 813         if (file_attr & FS_IMMUTABLE_FL)
 814                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 815
 816         free_and_replace(i->path, new_path);
 817         free_and_replace(i->name, nn);
 818
 819         STRV_FOREACH(j, settings) {
 820                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 821                 if (r < 0 && r != -ENOENT)
 822                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 823         }
 824
 825         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 826         if (r < 0 && r != -ENOENT)
 827                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 828
 829         return 0;
 830 }
 831
 832 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 833         _cleanup_free_ char *rs = NULL;
 834         const char *fn;
 835
 836         fn = strjoina(new_name, suffix);
 837
 838         rs = file_in_same_dir(path, fn);
 839         if (!rs)
 840                 return -ENOMEM;
 841
 842         return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK);
 843 }
 844
 845 int image_clone(Image *i, const char *new_name, bool read_only) {
 846         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 847         _cleanup_strv_free_ char **settings = NULL;
 848         _cleanup_free_ char *roothash = NULL;
 849         const char *new_path;
 850         int r;
 851
 852         assert(i);
 853
 854         if (!image_name_is_valid(new_name))
 855                 return -EINVAL;
 856
 857         settings = image_settings_path(i);
 858         if (!settings)
 859                 return -ENOMEM;
 860
 861         roothash = image_roothash_path(i);
 862         if (!roothash)
 863                 return -ENOMEM;
 864
 865         /* Make sure nobody takes the new name, between the time we
 866          * checked it is currently unused in all search paths, and the
 867          * time we take possession of it */
 868         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 869         if (r < 0)
 870                 return r;
 871
 872         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 873         if (r >= 0)
 874                 return -EEXIST;
 875         if (r != -ENOENT)
 876                 return r;
 877
 878         switch (i->type) {
 879
 880         case IMAGE_SUBVOLUME:
 881         case IMAGE_DIRECTORY:
 882                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 883                  * directory. */
 884
 885                 new_path = strjoina("/var/lib/machines/", new_name);
 886
 887                 r = btrfs_subvol_snapshot(i->path, new_path,
 888                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 889                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 890                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 891                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 892                                           BTRFS_SNAPSHOT_RECURSIVE |
 893                                           BTRFS_SNAPSHOT_QUOTA);
 894                 if (r >= 0)
 895                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 896                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 897
 898                 break;
 899
 900         case IMAGE_RAW:
 901                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 902
 903                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME);
 904                 break;
 905
 906         case IMAGE_BLOCK:
 907         default:
 908                 return -EOPNOTSUPP;
 909         }
 910
 911         if (r < 0)
 912                 return r;
 913
 914         STRV_FOREACH(j, settings) {
 915                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 916                 if (r < 0 && r != -ENOENT)
 917                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 918         }
 919
 920         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 921         if (r < 0 && r != -ENOENT)
 922                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 923
 924         return 0;
 925 }
 926
 927 int image_read_only(Image *i, bool b) {
 928         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 929         int r;
 930
 931         assert(i);
 932
 933         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 934                 return -EROFS;
 935
 936         /* Make sure we don't interfere with a running nspawn */
 937         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 938         if (r < 0)
 939                 return r;
 940
 941         switch (i->type) {
 942
 943         case IMAGE_SUBVOLUME:
 944
 945                 /* Note that we set the flag only on the top-level
 946                  * subvolume of the image. */
 947
 948                 r = btrfs_subvol_set_read_only(i->path, b);
 949                 if (r < 0)
 950                         return r;
 951
 952                 break;
 953
 954         case IMAGE_DIRECTORY:
 955                 /* For simple directory trees we cannot use the access
 956                    mode of the top-level directory, since it has an
 957                    effect on the container itself.  However, we can
 958                    use the "immutable" flag, to at least make the
 959                    top-level directory read-only. It's not as good as
 960                    a read-only subvolume, but at least something, and
 961                    we can read the value back. */
 962
 963                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 964                 if (r < 0)
 965                         return r;
 966
 967                 break;
 968
 969         case IMAGE_RAW: {
 970                 struct stat st;
 971
 972                 if (stat(i->path, &st) < 0)
 973                         return -errno;
 974
 975                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 976                         return -errno;
 977
 978                 /* If the images is now read-only, it's a good time to
 979                  * defrag it, given that no write patterns will
 980                  * fragment it again. */
 981                 if (b)
 982                         (void) btrfs_defrag(i->path);
 983                 break;
 984         }
 985
 986         case IMAGE_BLOCK: {
 987                 _cleanup_close_ int fd = -EBADF;
 988                 struct stat st;
 989                 int state = b;
 990
 991                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 992                 if (fd < 0)
 993                         return -errno;
 994
 995                 if (fstat(fd, &st) < 0)
 996                         return -errno;
 997                 if (!S_ISBLK(st.st_mode))
 998                         return -ENOTTY;
 999
1000                 if (ioctl(fd, BLKROSET, &state) < 0)
1001                         return -errno;
1002
1003                 break;
1004         }
1005
1006         default:
1007                 return -EOPNOTSUPP;
1008         }
1009
1010         return 0;
1011 }
1012
1013 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
1014         _cleanup_free_ char *p = NULL;
1015         LockFile t = LOCK_FILE_INIT;
1016         struct stat st;
1017         bool exclusive;
1018         int r;
1019
1020         assert(path);
1021         assert(global);
1022         assert(local);
1023
1024         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1025          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1026          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1027          * correctly. */
1028
1029         if (!path_is_absolute(path))
1030                 return -EINVAL;
1031
1032         switch (operation & (LOCK_SH|LOCK_EX)) {
1033         case LOCK_SH:
1034                 exclusive = false;
1035                 break;
1036         case LOCK_EX:
1037                 exclusive = true;
1038                 break;
1039         default:
1040                 return -EINVAL;
1041         }
1042
1043         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1044                 *local = *global = (LockFile) LOCK_FILE_INIT;
1045                 return 0;
1046         }
1047
1048         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1049          * running off it after all, and we don't want any images to manipulate the host image. We make an
1050          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1051          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1052          * since it means changes made to the host might propagate to the container as they happen (and a
1053          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1054          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1055          * the user to do the right thing with this. */
1056         if (path_equal(path, "/")) {
1057                 if (exclusive)
1058                         return -EBUSY;
1059
1060                 *local = *global = (LockFile) LOCK_FILE_INIT;
1061                 return 0;
1062         }
1063
1064         if (stat(path, &st) >= 0) {
1065                 if (S_ISBLK(st.st_mode))
1066                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1067                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1068                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1069                 else
1070                         return -ENOTTY;
1071                 if (r < 0)
1072                         return -ENOMEM;
1073         }
1074
1075         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1076          * sufficient, since block devices are host local anyway. */
1077         if (!path_startswith(path, "/dev/")) {
1078                 r = make_lock_file_for(path, operation, &t);
1079                 if (r < 0) {
1080                         if (!exclusive && r == -EROFS)
1081                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1082                         else
1083                                 return r;
1084                 }
1085         }
1086
1087         if (p) {
1088                 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1089
1090                 r = make_lock_file(p, operation, global);
1091                 if (r < 0) {
1092                         release_lock_file(&t);
1093                         return r;
1094                 }
1095         } else
1096                 *global = (LockFile) LOCK_FILE_INIT;
1097
1098         *local = t;
1099         return 0;
1100 }
1101
1102 int image_set_limit(Image *i, uint64_t referenced_max) {
1103         assert(i);
1104
1105         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1106                 return -EROFS;
1107
1108         if (i->type != IMAGE_SUBVOLUME)
1109                 return -EOPNOTSUPP;
1110
1111         /* We set the quota both for the subvolume as well as for the
1112          * subtree. The latter is mostly for historical reasons, since
1113          * we didn't use to have a concept of subtree quota, and hence
1114          * only modified the subvolume quota. */
1115
1116         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1117         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1118         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1119 }
1120
1121 int image_read_metadata(Image *i) {
1122         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1123         int r;
1124
1125         assert(i);
1126
1127         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1128         if (r < 0)
1129                 return r;
1130
1131         switch (i->type) {
1132
1133         case IMAGE_SUBVOLUME:
1134         case IMAGE_DIRECTORY: {
1135                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
1136                 sd_id128_t machine_id = SD_ID128_NULL;
1137                 _cleanup_free_ char *hostname = NULL;
1138                 _cleanup_free_ char *path = NULL;
1139
1140                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1141                 if (r < 0 && r != -ENOENT)
1142                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1143                 else if (r >= 0) {
1144                         r = read_etc_hostname(path, &hostname);
1145                         if (r < 0)
1146                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1147                 }
1148
1149                 path = mfree(path);
1150
1151                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1152                 if (r < 0 && r != -ENOENT)
1153                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1154                 else if (r >= 0) {
1155                         _cleanup_close_ int fd = -EBADF;
1156
1157                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1158                         if (fd < 0)
1159                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1160                         else {
1161                                 r = id128_read_fd(fd, ID128_FORMAT_PLAIN, &machine_id);
1162                                 if (r < 0)
1163                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1164                         }
1165                 }
1166
1167                 path = mfree(path);
1168
1169                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1170                 if (r < 0 && r != -ENOENT)
1171                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1172                 else if (r >= 0) {
1173                         r = load_env_file_pairs(NULL, path, &machine_info);
1174                         if (r < 0)
1175                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1176                 }
1177
1178                 r = load_os_release_pairs(i->path, &os_release);
1179                 if (r < 0)
1180                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1181
1182                 r = load_extension_release_pairs(i->path, i->name, /* relax_extension_release_check= */ false, &extension_release);
1183                 if (r < 0)
1184                         log_debug_errno(r, "Failed to read extension-release in image, ignoring: %m");
1185
1186                 free_and_replace(i->hostname, hostname);
1187                 i->machine_id = machine_id;
1188                 strv_free_and_replace(i->machine_info, machine_info);
1189                 strv_free_and_replace(i->os_release, os_release);
1190                 strv_free_and_replace(i->extension_release, extension_release);
1191
1192                 break;
1193         }
1194
1195         case IMAGE_RAW:
1196         case IMAGE_BLOCK: {
1197                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1198                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1199
1200                 r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, LOCK_SH, &d);
1201                 if (r < 0)
1202                         return r;
1203
1204                 r = dissect_loop_device(
1205                                 d,
1206                                 NULL, NULL,
1207                                 DISSECT_IMAGE_GENERIC_ROOT |
1208                                 DISSECT_IMAGE_REQUIRE_ROOT |
1209                                 DISSECT_IMAGE_RELAX_VAR_CHECK |
1210                                 DISSECT_IMAGE_READ_ONLY |
1211                                 DISSECT_IMAGE_USR_NO_ROOT |
1212                                 DISSECT_IMAGE_ADD_PARTITION_DEVICES |
1213                                 DISSECT_IMAGE_PIN_PARTITION_DEVICES,
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         assert(image);
1273
1274         NULSTR_FOREACH(path, image_search_path[class]) {
1275                 const char *p, *q;
1276                 size_t k;
1277
1278                 if (!empty_or_root(root)) {
1279                         q = path_startswith(path, root);
1280                         if (!q)
1281                                 continue;
1282                 } else
1283                         q = path;
1284
1285                 p = path_startswith(q, path);
1286                 if (!p)
1287                         continue;
1288
1289                 /* Make sure there's a filename following */
1290                 k = strcspn(p, "/");
1291                 if (k == 0)
1292                         continue;
1293
1294                 p += k;
1295
1296                 /* Accept trailing slashes */
1297                 if (p[strspn(p, "/")] == 0)
1298                         return true;
1299
1300         }
1301
1302         return false;
1303 }
1304
1305 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1306         [IMAGE_DIRECTORY] = "directory",
1307         [IMAGE_SUBVOLUME] = "subvolume",
1308         [IMAGE_RAW] = "raw",
1309         [IMAGE_BLOCK] = "block",
1310 };
1311
1312 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);
1313
1314 static const char* const image_class_table[_IMAGE_CLASS_MAX] = {
1315         [IMAGE_MACHINE] = "machine",
1316         [IMAGE_PORTABLE] = "portable",
1317         [IMAGE_EXTENSION] = "extension",
1318 };
1319
1320 DEFINE_STRING_TABLE_LOOKUP(image_class, ImageClass);