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.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 "lock-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         /* Note that we don't allow storing extensions under /usr/, unlike with other image types. That's
  62          * because extension images are supposed to extend /usr/, so you get into recursive races, especially
  63          * with directory-based extensions, as the kernel's OverlayFS explicitly checks for this and errors
  64          * out with -ELOOP if it finds that a lowerdir= is a child of another lowerdir=. */
  65         [IMAGE_EXTENSION] = "/etc/extensions\0"             /* only place symlinks here */
  66                             "/run/extensions\0"             /* and here too */
  67                             "/var/lib/extensions\0",        /* the main place for images */
  68 };
  69
  70 static Image *image_free(Image *i) {
  71         assert(i);
  72
  73         free(i->name);
  74         free(i->path);
  75
  76         free(i->hostname);
  77         strv_free(i->machine_info);
  78         strv_free(i->os_release);
  79         strv_free(i->extension_release);
  80
  81         return mfree(i);
  82 }
  83
  84 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  85 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
  86                                       Image, image_unref);
  87
  88 static char **image_settings_path(Image *image) {
  89         _cleanup_strv_free_ char **l = NULL;
  90         _cleanup_free_ char *fn = NULL;
  91         size_t i = 0;
  92         int r;
  93
  94         assert(image);
  95
  96         l = new0(char*, 4);
  97         if (!l)
  98                 return NULL;
  99
 100         fn = strjoin(image->name, ".nspawn");
 101         if (!fn)
 102                 return NULL;
 103
 104         FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
 105                 l[i] = path_join(s, fn);
 106                 if (!l[i])
 107                         return NULL;
 108
 109                 i++;
 110         }
 111
 112         r = file_in_same_dir(image->path, fn, l + i);
 113         if (r == -ENOMEM)
 114                 return NULL;
 115         if (r < 0)
 116                 log_debug_errno(r, "Failed to generate .nspawn settings path from image path, ignoring: %m");
 117
 118         strv_uniq(l);
 119
 120         return TAKE_PTR(l);
 121 }
 122
 123 static int image_roothash_path(Image *image, char **ret) {
 124         _cleanup_free_ char *fn = NULL;
 125
 126         assert(image);
 127
 128         fn = strjoin(image->name, ".roothash");
 129         if (!fn)
 130                 return -ENOMEM;
 131
 132         return file_in_same_dir(image->path, fn, ret);
 133 }
 134
 135 static int image_new(
 136                 ImageType t,
 137                 ImageClass c,
 138                 const char *pretty,
 139                 const char *path,
 140                 const char *filename,
 141                 bool read_only,
 142                 usec_t crtime,
 143                 usec_t mtime,
 144                 Image **ret) {
 145
 146         _cleanup_(image_unrefp) Image *i = NULL;
 147
 148         assert(t >= 0);
 149         assert(t < _IMAGE_TYPE_MAX);
 150         assert(pretty);
 151         assert(filename);
 152         assert(ret);
 153
 154         i = new(Image, 1);
 155         if (!i)
 156                 return -ENOMEM;
 157
 158         *i = (Image) {
 159                 .n_ref = 1,
 160                 .type = t,
 161                 .class = c,
 162                 .read_only = read_only,
 163                 .crtime = crtime,
 164                 .mtime = mtime,
 165                 .usage = UINT64_MAX,
 166                 .usage_exclusive = UINT64_MAX,
 167                 .limit = UINT64_MAX,
 168                 .limit_exclusive = UINT64_MAX,
 169         };
 170
 171         i->name = strdup(pretty);
 172         if (!i->name)
 173                 return -ENOMEM;
 174
 175         i->path = path_join(path, filename);
 176         if (!i->path)
 177                 return -ENOMEM;
 178
 179         path_simplify(i->path);
 180
 181         *ret = TAKE_PTR(i);
 182
 183         return 0;
 184 }
 185
 186 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 187         _cleanup_free_ char *name = NULL;
 188         const char *p;
 189
 190         assert(path);
 191         assert(ret);
 192
 193         p = last_path_component(path);
 194
 195         name = strdupcspn(p, "/");
 196         if (!name)
 197                 return -ENOMEM;
 198
 199         if (suffix) {
 200                 char *e;
 201
 202                 e = endswith(name, suffix);
 203                 if (!e)
 204                         return -EINVAL;
 205
 206                 *e = 0;
 207         }
 208
 209         if (!image_name_is_valid(name))
 210                 return -EINVAL;
 211
 212         *ret = TAKE_PTR(name);
 213         return 0;
 214 }
 215
 216 static int image_make(
 217                 ImageClass c,
 218                 const char *pretty,
 219                 int dfd,
 220                 const char *path,
 221                 const char *filename,
 222                 const struct stat *st,
 223                 Image **ret) {
 224
 225         _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL;
 226         struct stat stbuf;
 227         bool read_only;
 228         int r;
 229
 230         assert(dfd >= 0 || dfd == AT_FDCWD);
 231         assert(path || dfd == AT_FDCWD);
 232         assert(filename);
 233
 234         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 235          * devices into /var/lib/machines/, and treat them normally.
 236          *
 237          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 238          * recognize. */
 239
 240         if (!st) {
 241                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 242                         return -errno;
 243
 244                 st = &stbuf;
 245         }
 246
 247         if (!path) {
 248                 if (dfd == AT_FDCWD)
 249                         (void) safe_getcwd(&parent);
 250                 else
 251                         (void) fd_get_path(dfd, &parent);
 252         }
 253
 254         read_only =
 255                 (path && path_startswith(path, "/usr")) ||
 256                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 257
 258         if (S_ISDIR(st->st_mode)) {
 259                 _cleanup_close_ int fd = -EBADF;
 260                 unsigned file_attr = 0;
 261                 usec_t crtime = 0;
 262
 263                 if (!ret)
 264                         return 0;
 265
 266                 if (!pretty) {
 267                         r = extract_pretty(filename, NULL, &pretty_buffer);
 268                         if (r < 0)
 269                                 return r;
 270
 271                         pretty = pretty_buffer;
 272                 }
 273
 274                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 275                 if (fd < 0)
 276                         return -errno;
 277
 278                 if (btrfs_might_be_subvol(st)) {
 279
 280                         r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
 281                         if (r < 0)
 282                                 return r;
 283                         if (r) {
 284                                 BtrfsSubvolInfo info;
 285
 286                                 /* It's a btrfs subvolume */
 287
 288                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 289                                 if (r < 0)
 290                                         return r;
 291
 292                                 r = image_new(IMAGE_SUBVOLUME,
 293                                               c,
 294                                               pretty,
 295                                               path,
 296                                               filename,
 297                                               info.read_only || read_only,
 298                                               info.otime,
 299                                               0,
 300                                               ret);
 301                                 if (r < 0)
 302                                         return r;
 303
 304                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 305                                         BtrfsQuotaInfo quota;
 306
 307                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 308                                         if (r >= 0) {
 309                                                 (*ret)->usage = quota.referenced;
 310                                                 (*ret)->usage_exclusive = quota.exclusive;
 311
 312                                                 (*ret)->limit = quota.referenced_max;
 313                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 314                                         }
 315                                 }
 316
 317                                 return 0;
 318                         }
 319                 }
 320
 321                 /* Get directory creation time (not available everywhere, but that's OK */
 322                 (void) fd_getcrtime(fd, &crtime);
 323
 324                 /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not
 325                  * supported everywhere we ignore failures. */
 326                 (void) read_attr_fd(fd, &file_attr);
 327
 328                 /* It's just a normal directory. */
 329                 r = image_new(IMAGE_DIRECTORY,
 330                               c,
 331                               pretty,
 332                               path,
 333                               filename,
 334                               read_only || (file_attr & FS_IMMUTABLE_FL),
 335                               crtime,
 336                               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 */
 337                               ret);
 338                 if (r < 0)
 339                         return r;
 340
 341                 return 0;
 342
 343         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 344                 usec_t crtime = 0;
 345
 346                 /* It's a RAW disk image */
 347
 348                 if (!ret)
 349                         return 0;
 350
 351                 (void) fd_getcrtime_at(dfd, filename, AT_SYMLINK_FOLLOW, &crtime);
 352
 353                 if (!pretty) {
 354                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 355                         if (r < 0)
 356                                 return r;
 357
 358                         pretty = pretty_buffer;
 359                 }
 360
 361                 r = image_new(IMAGE_RAW,
 362                               c,
 363                               pretty,
 364                               path,
 365                               filename,
 366                               !(st->st_mode & 0222) || read_only,
 367                               crtime,
 368                               timespec_load(&st->st_mtim),
 369                               ret);
 370                 if (r < 0)
 371                         return r;
 372
 373                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 374                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 375
 376                 return 0;
 377
 378         } else if (S_ISBLK(st->st_mode)) {
 379                 _cleanup_close_ int block_fd = -EBADF;
 380                 uint64_t size = UINT64_MAX;
 381
 382                 /* A block device */
 383
 384                 if (!ret)
 385                         return 0;
 386
 387                 if (!pretty) {
 388                         r = extract_pretty(filename, NULL, &pretty_buffer);
 389                         if (r < 0)
 390                                 return r;
 391
 392                         pretty = pretty_buffer;
 393                 }
 394
 395                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 396                 if (block_fd < 0)
 397                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 398                 else {
 399                         /* Refresh stat data after opening the node */
 400                         if (fstat(block_fd, &stbuf) < 0)
 401                                 return -errno;
 402                         st = &stbuf;
 403
 404                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 405                                 return -ENOTTY;
 406
 407                         if (!read_only) {
 408                                 int state = 0;
 409
 410                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 411                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 412                                 else if (state)
 413                                         read_only = true;
 414                         }
 415
 416                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 417                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 418
 419                         block_fd = safe_close(block_fd);
 420                 }
 421
 422                 r = image_new(IMAGE_BLOCK,
 423                               c,
 424                               pretty,
 425                               path,
 426                               filename,
 427                               !(st->st_mode & 0222) || read_only,
 428                               0,
 429                               0,
 430                               ret);
 431                 if (r < 0)
 432                         return r;
 433
 434                 if (!IN_SET(size, 0, UINT64_MAX))
 435                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 436
 437                 return 0;
 438         }
 439
 440         return -EMEDIUMTYPE;
 441 }
 442
 443 int image_find(ImageClass class,
 444                const char *name,
 445                const char *root,
 446                Image **ret) {
 447
 448         int r;
 449
 450         assert(class >= 0);
 451         assert(class < _IMAGE_CLASS_MAX);
 452         assert(name);
 453
 454         /* There are no images with invalid names */
 455         if (!image_name_is_valid(name))
 456                 return -ENOENT;
 457
 458         NULSTR_FOREACH(path, image_search_path[class]) {
 459                 _cleanup_free_ char *resolved = NULL;
 460                 _cleanup_closedir_ DIR *d = NULL;
 461                 struct stat st;
 462                 int flags;
 463
 464                 r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 465                 if (r == -ENOENT)
 466                         continue;
 467                 if (r < 0)
 468                         return r;
 469
 470                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 471                  * to symlink block devices into the search path. (For now, we disable that when operating
 472                  * relative to some root directory.) */
 473                 flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 474                 if (fstatat(dirfd(d), name, &st, flags) < 0) {
 475                         _cleanup_free_ char *raw = NULL;
 476
 477                         if (errno != ENOENT)
 478                                 return -errno;
 479
 480                         raw = strjoin(name, ".raw");
 481                         if (!raw)
 482                                 return -ENOMEM;
 483
 484                         if (fstatat(dirfd(d), raw, &st, flags) < 0) {
 485                                 if (errno == ENOENT)
 486                                         continue;
 487
 488                                 return -errno;
 489                         }
 490
 491                         if (!S_ISREG(st.st_mode))
 492                                 continue;
 493
 494                         r = image_make(class, name, dirfd(d), resolved, raw, &st, ret);
 495
 496                 } else {
 497                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 498                                 continue;
 499
 500                         r = image_make(class, name, dirfd(d), resolved, name, &st, ret);
 501                 }
 502                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 503                         continue;
 504                 if (r < 0)
 505                         return r;
 506
 507                 if (ret)
 508                         (*ret)->discoverable = true;
 509
 510                 return 1;
 511         }
 512
 513         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 514                 r = image_make(class, ".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret);
 515                 if (r < 0)
 516                         return r;
 517
 518                 if (ret)
 519                         (*ret)->discoverable = true;
 520
 521                 return r;
 522         }
 523
 524         return -ENOENT;
 525 };
 526
 527 int image_from_path(const char *path, Image **ret) {
 528
 529         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 530          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 531          * overridden by another, different image earlier in the search path */
 532
 533         if (path_equal(path, "/"))
 534                 return image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, "/", NULL, ret);
 535
 536         return image_make(_IMAGE_CLASS_INVALID, NULL, AT_FDCWD, NULL, path, NULL, ret);
 537 }
 538
 539 int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) {
 540         if (image_name_is_valid(name_or_path))
 541                 return image_find(class, name_or_path, root, ret);
 542
 543         return image_from_path(name_or_path, ret);
 544 }
 545
 546 int image_discover(
 547                 ImageClass class,
 548                 const char *root,
 549                 Hashmap *h) {
 550
 551         int r;
 552
 553         assert(class >= 0);
 554         assert(class < _IMAGE_CLASS_MAX);
 555         assert(h);
 556
 557         NULSTR_FOREACH(path, image_search_path[class]) {
 558                 _cleanup_free_ char *resolved = NULL;
 559                 _cleanup_closedir_ DIR *d = NULL;
 560
 561                 r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 562                 if (r == -ENOENT)
 563                         continue;
 564                 if (r < 0)
 565                         return r;
 566
 567                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 568                         _cleanup_(image_unrefp) Image *image = NULL;
 569                         _cleanup_free_ char *truncated = NULL;
 570                         const char *pretty;
 571                         struct stat st;
 572                         int flags;
 573
 574                         if (dot_or_dot_dot(de->d_name))
 575                                 continue;
 576
 577                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to
 578                          * permit people to symlink block devices into the search path. */
 579                         flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 580                         if (fstatat(dirfd(d), de->d_name, &st, flags) < 0) {
 581                                 if (errno == ENOENT)
 582                                         continue;
 583
 584                                 return -errno;
 585                         }
 586
 587                         if (S_ISREG(st.st_mode)) {
 588                                 const char *e;
 589
 590                                 e = endswith(de->d_name, ".raw");
 591                                 if (!e)
 592                                         continue;
 593
 594                                 truncated = strndup(de->d_name, e - de->d_name);
 595                                 if (!truncated)
 596                                         return -ENOMEM;
 597
 598                                 pretty = truncated;
 599                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 600                                 pretty = de->d_name;
 601                         else
 602                                 continue;
 603
 604                         if (!image_name_is_valid(pretty))
 605                                 continue;
 606
 607                         if (hashmap_contains(h, pretty))
 608                                 continue;
 609
 610                         r = image_make(class, pretty, dirfd(d), resolved, de->d_name, &st, &image);
 611                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 612                                 continue;
 613                         if (r < 0)
 614                                 return r;
 615
 616                         image->discoverable = true;
 617
 618                         r = hashmap_put(h, image->name, image);
 619                         if (r < 0)
 620                                 return r;
 621
 622                         image = NULL;
 623                 }
 624         }
 625
 626         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 627                 _cleanup_(image_unrefp) Image *image = NULL;
 628
 629                 r = image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image);
 630                 if (r < 0)
 631                         return r;
 632
 633                 image->discoverable = true;
 634
 635                 r = hashmap_put(h, image->name, image);
 636                 if (r < 0)
 637                         return r;
 638
 639                 image = NULL;
 640         }
 641
 642         return 0;
 643 }
 644
 645 int image_remove(Image *i) {
 646         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 647         _cleanup_strv_free_ char **settings = NULL;
 648         _cleanup_free_ char *roothash = NULL;
 649         int r;
 650
 651         assert(i);
 652
 653         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 654                 return -EROFS;
 655
 656         settings = image_settings_path(i);
 657         if (!settings)
 658                 return -ENOMEM;
 659
 660         r = image_roothash_path(i, &roothash);
 661         if (r < 0)
 662                 return r;
 663
 664         /* Make sure we don't interfere with a running nspawn */
 665         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 666         if (r < 0)
 667                 return r;
 668
 669         switch (i->type) {
 670
 671         case IMAGE_SUBVOLUME:
 672
 673                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 674                  * big guns */
 675                 if (unlink(i->path) < 0) {
 676                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 677                         if (r < 0)
 678                                 return r;
 679                 }
 680
 681                 break;
 682
 683         case IMAGE_DIRECTORY:
 684                 /* Allow deletion of read-only directories */
 685                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 686                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 687                 if (r < 0)
 688                         return r;
 689
 690                 break;
 691
 692         case IMAGE_BLOCK:
 693
 694                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 695                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 696                  * the thing (it's most likely a symlink after all). */
 697
 698                 if (path_startswith(i->path, "/dev"))
 699                         break;
 700
 701                 _fallthrough_;
 702         case IMAGE_RAW:
 703                 if (unlink(i->path) < 0)
 704                         return -errno;
 705                 break;
 706
 707         default:
 708                 return -EOPNOTSUPP;
 709         }
 710
 711         STRV_FOREACH(j, settings)
 712                 if (unlink(*j) < 0 && errno != ENOENT)
 713                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 714
 715         if (unlink(roothash) < 0 && errno != ENOENT)
 716                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 717
 718         return 0;
 719 }
 720
 721 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 722         _cleanup_free_ char *fn = NULL, *rs = NULL;
 723         int r;
 724
 725         fn = strjoin(new_name, suffix);
 726         if (!fn)
 727                 return -ENOMEM;
 728
 729         r = file_in_same_dir(path, fn, &rs);
 730         if (r < 0)
 731                 return r;
 732
 733         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 734 }
 735
 736 int image_rename(Image *i, const char *new_name) {
 737         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 738         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 739         _cleanup_strv_free_ char **settings = NULL;
 740         unsigned file_attr = 0;
 741         int r;
 742
 743         assert(i);
 744
 745         if (!image_name_is_valid(new_name))
 746                 return -EINVAL;
 747
 748         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 749                 return -EROFS;
 750
 751         settings = image_settings_path(i);
 752         if (!settings)
 753                 return -ENOMEM;
 754
 755         r = image_roothash_path(i, &roothash);
 756         if (r < 0)
 757                 return r;
 758
 759         /* Make sure we don't interfere with a running nspawn */
 760         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 761         if (r < 0)
 762                 return r;
 763
 764         /* Make sure nobody takes the new name, between the time we
 765          * checked it is currently unused in all search paths, and the
 766          * time we take possession of it */
 767         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 768         if (r < 0)
 769                 return r;
 770
 771         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 772         if (r >= 0)
 773                 return -EEXIST;
 774         if (r != -ENOENT)
 775                 return r;
 776
 777         switch (i->type) {
 778
 779         case IMAGE_DIRECTORY:
 780                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 781                 (void) read_attr_path(i->path, &file_attr);
 782
 783                 if (file_attr & FS_IMMUTABLE_FL)
 784                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 785
 786                 _fallthrough_;
 787         case IMAGE_SUBVOLUME:
 788                 r = file_in_same_dir(i->path, new_name, &new_path);
 789                 break;
 790
 791         case IMAGE_BLOCK:
 792
 793                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 794                 if (path_startswith(i->path, "/dev"))
 795                         return -EROFS;
 796
 797                 r = file_in_same_dir(i->path, new_name, &new_path);
 798                 break;
 799
 800         case IMAGE_RAW: {
 801                 const char *fn;
 802
 803                 fn = strjoina(new_name, ".raw");
 804
 805                 r = file_in_same_dir(i->path, fn, &new_path);
 806                 break;
 807         }
 808
 809         default:
 810                 return -EOPNOTSUPP;
 811         }
 812         if (r < 0)
 813                 return r;
 814
 815         nn = strdup(new_name);
 816         if (!nn)
 817                 return -ENOMEM;
 818
 819         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 820         if (r < 0)
 821                 return r;
 822
 823         /* Restore the immutable bit, if it was set before */
 824         if (file_attr & FS_IMMUTABLE_FL)
 825                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 826
 827         free_and_replace(i->path, new_path);
 828         free_and_replace(i->name, nn);
 829
 830         STRV_FOREACH(j, settings) {
 831                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 832                 if (r < 0 && r != -ENOENT)
 833                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 834         }
 835
 836         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 837         if (r < 0 && r != -ENOENT)
 838                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 839
 840         return 0;
 841 }
 842
 843 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 844         _cleanup_free_ char *fn = NULL, *rs = NULL;
 845         int r;
 846
 847         fn = strjoin(new_name, suffix);
 848         if (!fn)
 849                 return -ENOMEM;
 850
 851         r = file_in_same_dir(path, fn, &rs);
 852         if (r < 0)
 853                 return r;
 854
 855         return copy_file_atomic(path, rs, 0664, COPY_REFLINK);
 856 }
 857
 858 int image_clone(Image *i, const char *new_name, bool read_only) {
 859         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 860         _cleanup_strv_free_ char **settings = NULL;
 861         _cleanup_free_ char *roothash = NULL;
 862         const char *new_path;
 863         int r;
 864
 865         assert(i);
 866
 867         if (!image_name_is_valid(new_name))
 868                 return -EINVAL;
 869
 870         settings = image_settings_path(i);
 871         if (!settings)
 872                 return -ENOMEM;
 873
 874         r = image_roothash_path(i, &roothash);
 875         if (r < 0)
 876                 return r;
 877
 878         /* Make sure nobody takes the new name, between the time we
 879          * checked it is currently unused in all search paths, and the
 880          * time we take possession of it */
 881         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 882         if (r < 0)
 883                 return r;
 884
 885         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 886         if (r >= 0)
 887                 return -EEXIST;
 888         if (r != -ENOENT)
 889                 return r;
 890
 891         switch (i->type) {
 892
 893         case IMAGE_SUBVOLUME:
 894         case IMAGE_DIRECTORY:
 895                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 896                  * directory. */
 897
 898                 new_path = strjoina("/var/lib/machines/", new_name);
 899
 900                 r = btrfs_subvol_snapshot(i->path, new_path,
 901                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 902                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 903                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 904                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 905                                           BTRFS_SNAPSHOT_RECURSIVE |
 906                                           BTRFS_SNAPSHOT_QUOTA);
 907                 if (r >= 0)
 908                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 909                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 910
 911                 break;
 912
 913         case IMAGE_RAW:
 914                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 915
 916                 r = copy_file_atomic_full(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL,
 917                                           COPY_REFLINK|COPY_CRTIME, NULL, NULL);
 918                 break;
 919
 920         case IMAGE_BLOCK:
 921         default:
 922                 return -EOPNOTSUPP;
 923         }
 924
 925         if (r < 0)
 926                 return r;
 927
 928         STRV_FOREACH(j, settings) {
 929                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 930                 if (r < 0 && r != -ENOENT)
 931                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 932         }
 933
 934         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 935         if (r < 0 && r != -ENOENT)
 936                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 937
 938         return 0;
 939 }
 940
 941 int image_read_only(Image *i, bool b) {
 942         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 943         int r;
 944
 945         assert(i);
 946
 947         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 948                 return -EROFS;
 949
 950         /* Make sure we don't interfere with a running nspawn */
 951         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 952         if (r < 0)
 953                 return r;
 954
 955         switch (i->type) {
 956
 957         case IMAGE_SUBVOLUME:
 958
 959                 /* Note that we set the flag only on the top-level
 960                  * subvolume of the image. */
 961
 962                 r = btrfs_subvol_set_read_only(i->path, b);
 963                 if (r < 0)
 964                         return r;
 965
 966                 break;
 967
 968         case IMAGE_DIRECTORY:
 969                 /* For simple directory trees we cannot use the access
 970                    mode of the top-level directory, since it has an
 971                    effect on the container itself.  However, we can
 972                    use the "immutable" flag, to at least make the
 973                    top-level directory read-only. It's not as good as
 974                    a read-only subvolume, but at least something, and
 975                    we can read the value back. */
 976
 977                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 978                 if (r < 0)
 979                         return r;
 980
 981                 break;
 982
 983         case IMAGE_RAW: {
 984                 struct stat st;
 985
 986                 if (stat(i->path, &st) < 0)
 987                         return -errno;
 988
 989                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 990                         return -errno;
 991
 992                 /* If the images is now read-only, it's a good time to
 993                  * defrag it, given that no write patterns will
 994                  * fragment it again. */
 995                 if (b)
 996                         (void) btrfs_defrag(i->path);
 997                 break;
 998         }
 999
1000         case IMAGE_BLOCK: {
1001                 _cleanup_close_ int fd = -EBADF;
1002                 struct stat st;
1003                 int state = b;
1004
1005                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
1006                 if (fd < 0)
1007                         return -errno;
1008
1009                 if (fstat(fd, &st) < 0)
1010                         return -errno;
1011                 if (!S_ISBLK(st.st_mode))
1012                         return -ENOTTY;
1013
1014                 if (ioctl(fd, BLKROSET, &state) < 0)
1015                         return -errno;
1016
1017                 break;
1018         }
1019
1020         default:
1021                 return -EOPNOTSUPP;
1022         }
1023
1024         return 0;
1025 }
1026
1027 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
1028         _cleanup_free_ char *p = NULL;
1029         LockFile t = LOCK_FILE_INIT;
1030         struct stat st;
1031         bool exclusive;
1032         int r;
1033
1034         assert(path);
1035         assert(global);
1036         assert(local);
1037
1038         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1039          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1040          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1041          * correctly. */
1042
1043         if (!path_is_absolute(path))
1044                 return -EINVAL;
1045
1046         switch (operation & (LOCK_SH|LOCK_EX)) {
1047         case LOCK_SH:
1048                 exclusive = false;
1049                 break;
1050         case LOCK_EX:
1051                 exclusive = true;
1052                 break;
1053         default:
1054                 return -EINVAL;
1055         }
1056
1057         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1058                 *local = *global = (LockFile) LOCK_FILE_INIT;
1059                 return 0;
1060         }
1061
1062         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1063          * running off it after all, and we don't want any images to manipulate the host image. We make an
1064          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1065          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1066          * since it means changes made to the host might propagate to the container as they happen (and a
1067          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1068          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1069          * the user to do the right thing with this. */
1070         if (path_equal(path, "/")) {
1071                 if (exclusive)
1072                         return -EBUSY;
1073
1074                 *local = *global = (LockFile) LOCK_FILE_INIT;
1075                 return 0;
1076         }
1077
1078         if (stat(path, &st) >= 0) {
1079                 if (S_ISBLK(st.st_mode))
1080                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1081                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1082                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1083                 else
1084                         return -ENOTTY;
1085                 if (r < 0)
1086                         return -ENOMEM;
1087         }
1088
1089         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1090          * sufficient, since block devices are host local anyway. */
1091         if (!path_startswith(path, "/dev/")) {
1092                 r = make_lock_file_for(path, operation, &t);
1093                 if (r < 0) {
1094                         if (!exclusive && r == -EROFS)
1095                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1096                         else
1097                                 return r;
1098                 }
1099         }
1100
1101         if (p) {
1102                 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1103
1104                 r = make_lock_file(p, operation, global);
1105                 if (r < 0) {
1106                         release_lock_file(&t);
1107                         return r;
1108                 }
1109         } else
1110                 *global = (LockFile) LOCK_FILE_INIT;
1111
1112         *local = t;
1113         return 0;
1114 }
1115
1116 int image_set_limit(Image *i, uint64_t referenced_max) {
1117         assert(i);
1118
1119         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1120                 return -EROFS;
1121
1122         if (i->type != IMAGE_SUBVOLUME)
1123                 return -EOPNOTSUPP;
1124
1125         /* We set the quota both for the subvolume as well as for the
1126          * subtree. The latter is mostly for historical reasons, since
1127          * we didn't use to have a concept of subtree quota, and hence
1128          * only modified the subvolume quota. */
1129
1130         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1131         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1132         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1133 }
1134
1135 int image_read_metadata(Image *i) {
1136         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1137         int r;
1138
1139         assert(i);
1140
1141         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1142         if (r < 0)
1143                 return r;
1144
1145         switch (i->type) {
1146
1147         case IMAGE_SUBVOLUME:
1148         case IMAGE_DIRECTORY: {
1149                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
1150                 sd_id128_t machine_id = SD_ID128_NULL;
1151                 _cleanup_free_ char *hostname = NULL;
1152                 _cleanup_free_ char *path = NULL;
1153
1154                 r = chase("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1155                 if (r < 0 && r != -ENOENT)
1156                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1157                 else if (r >= 0) {
1158                         r = read_etc_hostname(path, &hostname);
1159                         if (r < 0)
1160                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1161                 }
1162
1163                 path = mfree(path);
1164
1165                 r = chase("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1166                 if (r < 0 && r != -ENOENT)
1167                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1168                 else if (r >= 0) {
1169                         _cleanup_close_ int fd = -EBADF;
1170
1171                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1172                         if (fd < 0)
1173                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1174                         else {
1175                                 r = id128_read_fd(fd, ID128_FORMAT_PLAIN, &machine_id);
1176                                 if (r < 0)
1177                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1178                         }
1179                 }
1180
1181                 path = mfree(path);
1182
1183                 r = chase("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1184                 if (r < 0 && r != -ENOENT)
1185                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1186                 else if (r >= 0) {
1187                         r = load_env_file_pairs(NULL, path, &machine_info);
1188                         if (r < 0)
1189                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1190                 }
1191
1192                 r = load_os_release_pairs(i->path, &os_release);
1193                 if (r < 0)
1194                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1195
1196                 r = load_extension_release_pairs(i->path, i->name, /* relax_extension_release_check= */ false, &extension_release);
1197                 if (r < 0)
1198                         log_debug_errno(r, "Failed to read extension-release in image, ignoring: %m");
1199
1200                 free_and_replace(i->hostname, hostname);
1201                 i->machine_id = machine_id;
1202                 strv_free_and_replace(i->machine_info, machine_info);
1203                 strv_free_and_replace(i->os_release, os_release);
1204                 strv_free_and_replace(i->extension_release, extension_release);
1205
1206                 break;
1207         }
1208
1209         case IMAGE_RAW:
1210         case IMAGE_BLOCK: {
1211                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1212                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1213
1214                 r = loop_device_make_by_path(i->path, O_RDONLY, /* sector_size= */ UINT32_MAX, LO_FLAGS_PARTSCAN, LOCK_SH, &d);
1215                 if (r < 0)
1216                         return r;
1217
1218                 r = dissect_loop_device(
1219                                 d,
1220                                 NULL, NULL,
1221                                 DISSECT_IMAGE_GENERIC_ROOT |
1222                                 DISSECT_IMAGE_REQUIRE_ROOT |
1223                                 DISSECT_IMAGE_RELAX_VAR_CHECK |
1224                                 DISSECT_IMAGE_READ_ONLY |
1225                                 DISSECT_IMAGE_USR_NO_ROOT |
1226                                 DISSECT_IMAGE_ADD_PARTITION_DEVICES |
1227                                 DISSECT_IMAGE_PIN_PARTITION_DEVICES,
1228                                 &m);
1229                 if (r < 0)
1230                         return r;
1231
1232                 r = dissected_image_acquire_metadata(m,
1233                                                      DISSECT_IMAGE_VALIDATE_OS |
1234                                                      DISSECT_IMAGE_VALIDATE_OS_EXT);
1235                 if (r < 0)
1236                         return r;
1237
1238                 free_and_replace(i->hostname, m->hostname);
1239                 i->machine_id = m->machine_id;
1240                 strv_free_and_replace(i->machine_info, m->machine_info);
1241                 strv_free_and_replace(i->os_release, m->os_release);
1242                 strv_free_and_replace(i->extension_release, m->extension_release);
1243
1244                 break;
1245         }
1246
1247         default:
1248                 return -EOPNOTSUPP;
1249         }
1250
1251         i->metadata_valid = true;
1252
1253         return 0;
1254 }
1255
1256 int image_name_lock(const char *name, int operation, LockFile *ret) {
1257         const char *p;
1258
1259         assert(name);
1260         assert(ret);
1261
1262         /* Locks an image name, regardless of the precise path used. */
1263
1264         if (streq(name, ".host"))
1265                 return -EBUSY;
1266
1267         if (!image_name_is_valid(name))
1268                 return -EINVAL;
1269
1270         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1271                 *ret = (LockFile) LOCK_FILE_INIT;
1272                 return 0;
1273         }
1274
1275         (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1276
1277         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1278         return make_lock_file(p, operation, ret);
1279 }
1280
1281 bool image_in_search_path(
1282                 ImageClass class,
1283                 const char *root,
1284                 const char *image) {
1285
1286         assert(image);
1287
1288         NULSTR_FOREACH(path, image_search_path[class]) {
1289                 const char *p, *q;
1290                 size_t k;
1291
1292                 if (!empty_or_root(root)) {
1293                         q = path_startswith(path, root);
1294                         if (!q)
1295                                 continue;
1296                 } else
1297                         q = path;
1298
1299                 p = path_startswith(q, path);
1300                 if (!p)
1301                         continue;
1302
1303                 /* Make sure there's a filename following */
1304                 k = strcspn(p, "/");
1305                 if (k == 0)
1306                         continue;
1307
1308                 p += k;
1309
1310                 /* Accept trailing slashes */
1311                 if (p[strspn(p, "/")] == 0)
1312                         return true;
1313
1314         }
1315
1316         return false;
1317 }
1318
1319 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1320         [IMAGE_DIRECTORY] = "directory",
1321         [IMAGE_SUBVOLUME] = "subvolume",
1322         [IMAGE_RAW] = "raw",
1323         [IMAGE_BLOCK] = "block",
1324 };
1325
1326 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);
1327
1328 static const char* const image_class_table[_IMAGE_CLASS_MAX] = {
1329         [IMAGE_MACHINE] = "machine",
1330         [IMAGE_PORTABLE] = "portable",
1331         [IMAGE_EXTENSION] = "extension",
1332 };
1333
1334 DEFINE_STRING_TABLE_LOOKUP(image_class, ImageClass);