src/shared/discover-image.c

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