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 "blockdev-util.h"
  17 #include "btrfs-util.h"
  18 #include "chase.h"
  19 #include "chattr-util.h"
  20 #include "copy.h"
  21 #include "dirent-util.h"
  22 #include "discover-image.h"
  23 #include "dissect-image.h"
  24 #include "env-file.h"
  25 #include "env-util.h"
  26 #include "extension-util.h"
  27 #include "fd-util.h"
  28 #include "fs-util.h"
  29 #include "hashmap.h"
  30 #include "hostname-setup.h"
  31 #include "id128-util.h"
  32 #include "initrd-util.h"
  33 #include "lock-util.h"
  34 #include "log.h"
  35 #include "loop-util.h"
  36 #include "macro.h"
  37 #include "mkdir.h"
  38 #include "nulstr-util.h"
  39 #include "os-util.h"
  40 #include "path-util.h"
  41 #include "rm-rf.h"
  42 #include "stat-util.h"
  43 #include "string-table.h"
  44 #include "string-util.h"
  45 #include "strv.h"
  46 #include "time-util.h"
  47 #include "utf8.h"
  48 #include "vpick.h"
  49 #include "xattr-util.h"
  50
  51 const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  52         [IMAGE_MACHINE] =   "/etc/machines\0"              /* only place symlinks here */
  53                             "/run/machines\0"              /* and here too */
  54                             "/var/lib/machines\0"          /* the main place for images */
  55                             "/var/lib/container\0"         /* legacy */
  56                             "/usr/local/lib/machines\0"
  57                             "/usr/lib/machines\0",
  58
  59         [IMAGE_PORTABLE] =  "/etc/portables\0"             /* only place symlinks here */
  60                             "/run/portables\0"             /* and here too */
  61                             "/var/lib/portables\0"         /* the main place for images */
  62                             "/usr/local/lib/portables\0"
  63                             "/usr/lib/portables\0",
  64
  65         /* Note that we don't allow storing extensions under /usr/, unlike with other image types. That's
  66          * because extension images are supposed to extend /usr/, so you get into recursive races, especially
  67          * with directory-based extensions, as the kernel's OverlayFS explicitly checks for this and errors
  68          * out with -ELOOP if it finds that a lowerdir= is a child of another lowerdir=. */
  69         [IMAGE_SYSEXT] =    "/etc/extensions\0"            /* only place symlinks here */
  70                             "/run/extensions\0"            /* and here too */
  71                             "/var/lib/extensions\0",       /* the main place for images */
  72
  73         [IMAGE_CONFEXT] =   "/run/confexts\0"              /* only place symlinks here */
  74                             "/var/lib/confexts\0"          /* the main place for images */
  75                             "/usr/local/lib/confexts\0"
  76                             "/usr/lib/confexts\0",
  77 };
  78
  79 /* Inside the initrd, use a slightly different set of search path (i.e. include .extra/sysext/ and
  80  * .extra/confext/ in extension search dir) */
  81 static const char* const image_search_path_initrd[_IMAGE_CLASS_MAX] = {
  82         /* (entries that aren't listed here will get the same search path as for the non initrd-case) */
  83
  84         [IMAGE_SYSEXT] =    "/etc/extensions\0"            /* only place symlinks here */
  85                             "/run/extensions\0"            /* and here too */
  86                             "/var/lib/extensions\0"        /* the main place for images */
  87                             "/.extra/sysext\0",            /* put sysext picked up by systemd-stub last, since not trusted */
  88
  89         [IMAGE_CONFEXT] =   "/run/confexts\0"              /* only place symlinks here */
  90                             "/var/lib/confexts\0"          /* the main place for images */
  91                             "/usr/local/lib/confexts\0"
  92                             "/.extra/confext\0",           /* put confext picked up by systemd-stub last, since not trusted */
  93 };
  94
  95 static const char* image_class_suffix_table[_IMAGE_CLASS_MAX] = {
  96         [IMAGE_SYSEXT]  = ".sysext",
  97         [IMAGE_CONFEXT] = ".confext",
  98 };
  99
 100 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(image_class_suffix, ImageClass);
 101
 102 static const char *const image_root_table[_IMAGE_CLASS_MAX] = {
 103         [IMAGE_MACHINE]  = "/var/lib/machines",
 104         [IMAGE_PORTABLE] = "/var/lib/portables",
 105         [IMAGE_SYSEXT]   = "/var/lib/extensions",
 106         [IMAGE_CONFEXT]  = "/var/lib/confexts",
 107 };
 108
 109 DEFINE_STRING_TABLE_LOOKUP_TO_STRING(image_root, ImageClass);
 110
 111 static Image *image_free(Image *i) {
 112         assert(i);
 113
 114         free(i->name);
 115         free(i->path);
 116
 117         free(i->hostname);
 118         strv_free(i->machine_info);
 119         strv_free(i->os_release);
 120         strv_free(i->sysext_release);
 121         strv_free(i->confext_release);
 122
 123         return mfree(i);
 124 }
 125
 126 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
 127 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
 128                                       Image, image_unref);
 129
 130 static char **image_settings_path(Image *image) {
 131         _cleanup_strv_free_ char **l = NULL;
 132         _cleanup_free_ char *fn = NULL;
 133         size_t i = 0;
 134         int r;
 135
 136         assert(image);
 137
 138         l = new0(char*, 4);
 139         if (!l)
 140                 return NULL;
 141
 142         fn = strjoin(image->name, ".nspawn");
 143         if (!fn)
 144                 return NULL;
 145
 146         FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
 147                 l[i] = path_join(s, fn);
 148                 if (!l[i])
 149                         return NULL;
 150
 151                 i++;
 152         }
 153
 154         r = file_in_same_dir(image->path, fn, l + i);
 155         if (r == -ENOMEM)
 156                 return NULL;
 157         if (r < 0)
 158                 log_debug_errno(r, "Failed to generate .nspawn settings path from image path, ignoring: %m");
 159
 160         strv_uniq(l);
 161
 162         return TAKE_PTR(l);
 163 }
 164
 165 static int image_roothash_path(Image *image, char **ret) {
 166         _cleanup_free_ char *fn = NULL;
 167
 168         assert(image);
 169
 170         fn = strjoin(image->name, ".roothash");
 171         if (!fn)
 172                 return -ENOMEM;
 173
 174         return file_in_same_dir(image->path, fn, ret);
 175 }
 176
 177 static int image_new(
 178                 ImageType t,
 179                 ImageClass c,
 180                 const char *pretty,
 181                 const char *path,
 182                 const char *filename,
 183                 bool read_only,
 184                 usec_t crtime,
 185                 usec_t mtime,
 186                 Image **ret) {
 187
 188         _cleanup_(image_unrefp) Image *i = NULL;
 189
 190         assert(t >= 0);
 191         assert(t < _IMAGE_TYPE_MAX);
 192         assert(pretty);
 193         assert(filename);
 194         assert(ret);
 195
 196         i = new(Image, 1);
 197         if (!i)
 198                 return -ENOMEM;
 199
 200         *i = (Image) {
 201                 .n_ref = 1,
 202                 .type = t,
 203                 .class = c,
 204                 .read_only = read_only,
 205                 .crtime = crtime,
 206                 .mtime = mtime,
 207                 .usage = UINT64_MAX,
 208                 .usage_exclusive = UINT64_MAX,
 209                 .limit = UINT64_MAX,
 210                 .limit_exclusive = UINT64_MAX,
 211         };
 212
 213         i->name = strdup(pretty);
 214         if (!i->name)
 215                 return -ENOMEM;
 216
 217         i->path = path_join(path, filename);
 218         if (!i->path)
 219                 return -ENOMEM;
 220
 221         path_simplify(i->path);
 222
 223         *ret = TAKE_PTR(i);
 224
 225         return 0;
 226 }
 227
 228 static int extract_image_basename(
 229                 const char *path,
 230                 const char *class_suffix,  /* e.g. ".sysext" (this is an optional suffix) */
 231                 char **format_suffixes,    /* e.g. ".raw"    (one of these will be required) */
 232                 char **ret_basename,
 233                 char **ret_suffix) {
 234
 235         _cleanup_free_ char *name = NULL, *suffix = NULL;
 236         int r;
 237
 238         assert(path);
 239
 240         r = path_extract_filename(path, &name);
 241         if (r < 0)
 242                 return r;
 243
 244         if (format_suffixes) {
 245                 char *e = endswith_strv(name, format_suffixes);
 246                 if (!e) /* Format suffix is required */
 247                         return -EINVAL;
 248
 249                 if (ret_suffix) {
 250                         suffix = strdup(e);
 251                         if (!suffix)
 252                                 return -ENOMEM;
 253                 }
 254
 255                 *e = 0;
 256         }
 257
 258         if (class_suffix) {
 259                 char *e = endswith(name, class_suffix);
 260                 if (e) { /* Class suffix is optional */
 261                         if (ret_suffix) {
 262                                 _cleanup_free_ char *j = strjoin(e, suffix);
 263                                 if (!j)
 264                                         return -ENOMEM;
 265
 266                                 free_and_replace(suffix, j);
 267                         }
 268
 269                         *e = 0;
 270                 }
 271         }
 272
 273         if (!image_name_is_valid(name))
 274                 return -EINVAL;
 275
 276         if (ret_suffix)
 277                 *ret_suffix = TAKE_PTR(suffix);
 278
 279         if (ret_basename)
 280                 *ret_basename = TAKE_PTR(name);
 281
 282         return 0;
 283 }
 284
 285 static int image_update_quota(Image *i, int fd) {
 286         _cleanup_close_ int fd_close = -EBADF;
 287         int r;
 288
 289         assert(i);
 290
 291         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 292                 return -EROFS;
 293
 294         if (i->type != IMAGE_SUBVOLUME)
 295                 return -EOPNOTSUPP;
 296
 297         if (fd < 0) {
 298                 fd_close = open(i->path, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 299                 if (fd_close < 0)
 300                         return -errno;
 301                 fd = fd_close;
 302         }
 303
 304         r = btrfs_quota_scan_ongoing(fd);
 305         if (r < 0)
 306                 return r;
 307         if (r > 0)
 308                 return 0;
 309
 310         BtrfsQuotaInfo quota;
 311         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 312         if (r < 0)
 313                 return r;
 314
 315         i->usage = quota.referenced;
 316         i->usage_exclusive = quota.exclusive;
 317         i->limit = quota.referenced_max;
 318         i->limit_exclusive = quota.exclusive_max;
 319
 320         return 1;
 321 }
 322
 323 static int image_make(
 324                 ImageClass c,
 325                 const char *pretty,
 326                 int dfd,
 327                 const char *path,
 328                 const char *filename,
 329                 const struct stat *st,
 330                 Image **ret) {
 331
 332         _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL;
 333         struct stat stbuf;
 334         bool read_only;
 335         int r;
 336
 337         assert(dfd >= 0 || dfd == AT_FDCWD);
 338         assert(path || dfd == AT_FDCWD);
 339         assert(filename);
 340
 341         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 342          * devices into /var/lib/machines/, and treat them normally.
 343          *
 344          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 345          * recognize. */
 346
 347         if (!st) {
 348                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 349                         return -errno;
 350
 351                 st = &stbuf;
 352         }
 353
 354         if (!path) {
 355                 if (dfd == AT_FDCWD)
 356                         (void) safe_getcwd(&parent);
 357                 else
 358                         (void) fd_get_path(dfd, &parent);
 359         }
 360
 361         read_only =
 362                 (path && path_startswith(path, "/usr")) ||
 363                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 364
 365         if (S_ISDIR(st->st_mode)) {
 366                 _cleanup_close_ int fd = -EBADF;
 367                 unsigned file_attr = 0;
 368                 usec_t crtime = 0;
 369
 370                 if (!ret)
 371                         return 0;
 372
 373                 if (!pretty) {
 374                         r = extract_image_basename(
 375                                         filename,
 376                                         image_class_suffix_to_string(c),
 377                                         /* format_suffix= */ NULL,
 378                                         &pretty_buffer,
 379                                         /* ret_suffix= */ NULL);
 380                         if (r < 0)
 381                                 return r;
 382
 383                         pretty = pretty_buffer;
 384                 }
 385
 386                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 387                 if (fd < 0)
 388                         return -errno;
 389
 390                 if (btrfs_might_be_subvol(st)) {
 391
 392                         r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
 393                         if (r < 0)
 394                                 return r;
 395                         if (r > 0) {
 396                                 BtrfsSubvolInfo info;
 397
 398                                 /* It's a btrfs subvolume */
 399
 400                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 401                                 if (r < 0)
 402                                         return r;
 403
 404                                 r = image_new(IMAGE_SUBVOLUME,
 405                                               c,
 406                                               pretty,
 407                                               path,
 408                                               filename,
 409                                               info.read_only || read_only,
 410                                               info.otime,
 411                                               0,
 412                                               ret);
 413                                 if (r < 0)
 414                                         return r;
 415
 416                                 (void) image_update_quota(*ret, fd);
 417                                 return 0;
 418                         }
 419                 }
 420
 421                 /* Get directory creation time (not available everywhere, but that's OK */
 422                 (void) fd_getcrtime(fd, &crtime);
 423
 424                 /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not
 425                  * supported everywhere we ignore failures. */
 426                 (void) read_attr_fd(fd, &file_attr);
 427
 428                 /* It's just a normal directory. */
 429                 r = image_new(IMAGE_DIRECTORY,
 430                               c,
 431                               pretty,
 432                               path,
 433                               filename,
 434                               read_only || (file_attr & FS_IMMUTABLE_FL),
 435                               crtime,
 436                               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 */
 437                               ret);
 438                 if (r < 0)
 439                         return r;
 440
 441                 return 0;
 442
 443         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 444                 usec_t crtime = 0;
 445
 446                 /* It's a RAW disk image */
 447
 448                 if (!ret)
 449                         return 0;
 450
 451                 (void) fd_getcrtime_at(dfd, filename, AT_SYMLINK_FOLLOW, &crtime);
 452
 453                 if (!pretty) {
 454                         r = extract_image_basename(
 455                                         filename,
 456                                         image_class_suffix_to_string(c),
 457                                         STRV_MAKE(".raw"),
 458                                         &pretty_buffer,
 459                                         /* ret_suffix= */ NULL);
 460                         if (r < 0)
 461                                 return r;
 462
 463                         pretty = pretty_buffer;
 464                 }
 465
 466                 r = image_new(IMAGE_RAW,
 467                               c,
 468                               pretty,
 469                               path,
 470                               filename,
 471                               !(st->st_mode & 0222) || read_only,
 472                               crtime,
 473                               timespec_load(&st->st_mtim),
 474                               ret);
 475                 if (r < 0)
 476                         return r;
 477
 478                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 479                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 480
 481                 return 0;
 482
 483         } else if (S_ISBLK(st->st_mode)) {
 484                 _cleanup_close_ int block_fd = -EBADF;
 485                 uint64_t size = UINT64_MAX;
 486
 487                 /* A block device */
 488
 489                 if (!ret)
 490                         return 0;
 491
 492                 if (!pretty) {
 493                         r = extract_image_basename(
 494                                         filename,
 495                                         /* class_suffix= */ NULL,
 496                                         /* format_suffix= */ NULL,
 497                                         &pretty_buffer,
 498                                         /* ret_suffix= */ NULL);
 499                         if (r < 0)
 500                                 return r;
 501
 502                         pretty = pretty_buffer;
 503                 }
 504
 505                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 506                 if (block_fd < 0)
 507                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 508                 else {
 509                         /* Refresh stat data after opening the node */
 510                         if (fstat(block_fd, &stbuf) < 0)
 511                                 return -errno;
 512                         st = &stbuf;
 513
 514                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 515                                 return -ENOTTY;
 516
 517                         if (!read_only) {
 518                                 int state = 0;
 519
 520                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 521                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 522                                 else if (state)
 523                                         read_only = true;
 524                         }
 525
 526                         r = blockdev_get_device_size(block_fd, &size);
 527                         if (r < 0)
 528                                 log_debug_errno(r, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 529
 530                         block_fd = safe_close(block_fd);
 531                 }
 532
 533                 r = image_new(IMAGE_BLOCK,
 534                               c,
 535                               pretty,
 536                               path,
 537                               filename,
 538                               !(st->st_mode & 0222) || read_only,
 539                               0,
 540                               0,
 541                               ret);
 542                 if (r < 0)
 543                         return r;
 544
 545                 if (!IN_SET(size, 0, UINT64_MAX))
 546                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 547
 548                 return 0;
 549         }
 550
 551         return -EMEDIUMTYPE;
 552 }
 553
 554 static const char *pick_image_search_path(ImageClass class) {
 555         if (class < 0 || class >= _IMAGE_CLASS_MAX)
 556                 return NULL;
 557
 558         /* Use the initrd search path if there is one, otherwise use the common one */
 559         return in_initrd() && image_search_path_initrd[class] ? image_search_path_initrd[class] : image_search_path[class];
 560 }
 561
 562 static char **make_possible_filenames(ImageClass class, const char *image_name) {
 563         _cleanup_strv_free_ char **l = NULL;
 564
 565         assert(image_name);
 566
 567         FOREACH_STRING(v_suffix, "", ".v")
 568                 FOREACH_STRING(format_suffix, "", ".raw") {
 569                         _cleanup_free_ char *j = NULL;
 570                         const char *class_suffix;
 571
 572                         class_suffix = image_class_suffix_to_string(class);
 573                         if (class_suffix) {
 574                                 j = strjoin(image_name, class_suffix, format_suffix, v_suffix);
 575                                 if (!j)
 576                                         return NULL;
 577
 578                                 if (strv_consume(&l, TAKE_PTR(j)) < 0)
 579                                         return NULL;
 580                         }
 581
 582                         j = strjoin(image_name, format_suffix, v_suffix);
 583                         if (!j)
 584                                 return NULL;
 585
 586                         if (strv_consume(&l, TAKE_PTR(j)) < 0)
 587                                 return NULL;
 588                 }
 589
 590         return TAKE_PTR(l);
 591 }
 592
 593 int image_find(ImageClass class,
 594                const char *name,
 595                const char *root,
 596                Image **ret) {
 597
 598         int r;
 599
 600         assert(class >= 0);
 601         assert(class < _IMAGE_CLASS_MAX);
 602         assert(name);
 603
 604         /* There are no images with invalid names */
 605         if (!image_name_is_valid(name))
 606                 return -ENOENT;
 607
 608         _cleanup_strv_free_ char **names = make_possible_filenames(class, name);
 609         if (!names)
 610                 return -ENOMEM;
 611
 612         NULSTR_FOREACH(path, pick_image_search_path(class)) {
 613                 _cleanup_free_ char *resolved = NULL;
 614                 _cleanup_closedir_ DIR *d = NULL;
 615                 struct stat st;
 616                 int flags;
 617
 618                 r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 619                 if (r == -ENOENT)
 620                         continue;
 621                 if (r < 0)
 622                         return r;
 623
 624                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 625                  * to symlink block devices into the search path. (For now, we disable that when operating
 626                  * relative to some root directory.) */
 627                 flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 628
 629                 STRV_FOREACH(n, names) {
 630                         _cleanup_free_ char *fname_buf = NULL;
 631                         const char *fname = *n;
 632
 633                         if (fstatat(dirfd(d), fname, &st, flags) < 0) {
 634                                 if (errno != ENOENT)
 635                                         return -errno;
 636
 637                                 continue; /* Vanished while we were looking at it */
 638                         }
 639
 640                         if (endswith(fname, ".raw")) {
 641                                 if (!S_ISREG(st.st_mode)) {
 642                                         log_debug("Ignoring non-regular file '%s' with .raw suffix.", fname);
 643                                         continue;
 644                                 }
 645
 646                         } else if (endswith(fname, ".v")) {
 647
 648                                 if (!S_ISDIR(st.st_mode)) {
 649                                         log_debug("Ignoring non-directory file '%s' with .v suffix.", fname);
 650                                         continue;
 651                                 }
 652
 653                                 _cleanup_free_ char *suffix = NULL;
 654                                 suffix = strdup(ASSERT_PTR(startswith(fname, name)));
 655                                 if (!suffix)
 656                                         return -ENOMEM;
 657
 658                                 *ASSERT_PTR(endswith(suffix, ".v")) = 0;
 659
 660                                 _cleanup_free_ char *vp = path_join(resolved, fname);
 661                                 if (!vp)
 662                                         return -ENOMEM;
 663
 664                                 PickFilter filter = {
 665                                         .type_mask = endswith(suffix, ".raw") ? (UINT32_C(1) << DT_REG) | (UINT32_C(1) << DT_BLK) : (UINT32_C(1) << DT_DIR),
 666                                         .basename = name,
 667                                         .architecture = _ARCHITECTURE_INVALID,
 668                                         .suffix = STRV_MAKE(suffix),
 669                                 };
 670
 671                                 _cleanup_(pick_result_done) PickResult result = PICK_RESULT_NULL;
 672                                 r = path_pick(root,
 673                                               /* toplevel_fd= */ AT_FDCWD,
 674                                               vp,
 675                                               &filter,
 676                                               PICK_ARCHITECTURE|PICK_TRIES,
 677                                               &result);
 678                                 if (r < 0) {
 679                                         log_debug_errno(r, "Failed to pick versioned image on '%s', skipping: %m", vp);
 680                                         continue;
 681                                 }
 682                                 if (!result.path) {
 683                                         log_debug("Found versioned directory '%s', without matching entry, skipping: %m", vp);
 684                                         continue;
 685                                 }
 686
 687                                 /* Refresh the stat data for the discovered target */
 688                                 st = result.st;
 689
 690                                 _cleanup_free_ char *bn = NULL;
 691                                 r = path_extract_filename(result.path, &bn);
 692                                 if (r < 0) {
 693                                         log_debug_errno(r, "Failed to extract basename of image path '%s', skipping: %m", result.path);
 694                                         continue;
 695                                 }
 696
 697                                 fname_buf = path_join(fname, bn);
 698                                 if (!fname_buf)
 699                                         return log_oom();
 700
 701                                 fname = fname_buf;
 702
 703                         } else if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode)) {
 704                                 log_debug("Ignoring non-directory and non-block device file '%s' without suffix.", fname);
 705                                 continue;
 706                         }
 707
 708                         r = image_make(class, name, dirfd(d), resolved, fname, &st, ret);
 709                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 710                                 continue;
 711                         if (r < 0)
 712                                 return r;
 713
 714                         if (ret)
 715                                 (*ret)->discoverable = true;
 716
 717                         return 1;
 718                 }
 719         }
 720
 721         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 722                 r = image_make(class, ".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret);
 723                 if (r < 0)
 724                         return r;
 725
 726                 if (ret)
 727                         (*ret)->discoverable = true;
 728
 729                 return 1;
 730         }
 731
 732         return -ENOENT;
 733 };
 734
 735 int image_from_path(const char *path, Image **ret) {
 736
 737         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 738          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 739          * overridden by another, different image earlier in the search path */
 740
 741         if (path_equal(path, "/"))
 742                 return image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, "/", NULL, ret);
 743
 744         return image_make(_IMAGE_CLASS_INVALID, NULL, AT_FDCWD, NULL, path, NULL, ret);
 745 }
 746
 747 int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) {
 748         if (image_name_is_valid(name_or_path))
 749                 return image_find(class, name_or_path, root, ret);
 750
 751         return image_from_path(name_or_path, ret);
 752 }
 753
 754 int image_discover(
 755                 ImageClass class,
 756                 const char *root,
 757                 Hashmap *h) {
 758
 759         int r;
 760
 761         assert(class >= 0);
 762         assert(class < _IMAGE_CLASS_MAX);
 763         assert(h);
 764
 765         NULSTR_FOREACH(path, pick_image_search_path(class)) {
 766                 _cleanup_free_ char *resolved = NULL;
 767                 _cleanup_closedir_ DIR *d = NULL;
 768
 769                 r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 770                 if (r == -ENOENT)
 771                         continue;
 772                 if (r < 0)
 773                         return r;
 774
 775                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 776                         _cleanup_free_ char *pretty = NULL, *fname_buf = NULL;
 777                         _cleanup_(image_unrefp) Image *image = NULL;
 778                         const char *fname = de->d_name;
 779                         struct stat st;
 780                         int flags;
 781
 782                         if (dot_or_dot_dot(fname))
 783                                 continue;
 784
 785                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to
 786                          * permit people to symlink block devices into the search path. */
 787                         flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 788                         if (fstatat(dirfd(d), fname, &st, flags) < 0) {
 789                                 if (errno == ENOENT)
 790                                         continue;
 791
 792                                 return -errno;
 793                         }
 794
 795                         if (S_ISREG(st.st_mode)) {
 796                                 r = extract_image_basename(
 797                                                 fname,
 798                                                 image_class_suffix_to_string(class),
 799                                                 STRV_MAKE(".raw"),
 800                                                 &pretty,
 801                                                 /* suffix= */ NULL);
 802                                 if (r < 0) {
 803                                         log_debug_errno(r, "Skipping directory entry '%s', which doesn't look like an image.", fname);
 804                                         continue;
 805                                 }
 806                         } else if (S_ISDIR(st.st_mode)) {
 807                                 const char *v;
 808
 809                                 v = endswith(fname, ".v");
 810                                 if (v) {
 811                                         _cleanup_free_ char *suffix = NULL, *nov = NULL;
 812
 813                                         nov = strndup(fname, v - fname); /* Chop off the .v */
 814                                         if (!nov)
 815                                                 return -ENOMEM;
 816
 817                                         r = extract_image_basename(
 818                                                         nov,
 819                                                         image_class_suffix_to_string(class),
 820                                                         STRV_MAKE(".raw", ""),
 821                                                         &pretty,
 822                                                         &suffix);
 823                                         if (r < 0) {
 824                                                 log_debug_errno(r, "Skipping directory entry '%s', which doesn't look like a versioned image.", fname);
 825                                                 continue;
 826                                         }
 827
 828                                         _cleanup_free_ char *vp = path_join(resolved, fname);
 829                                         if (!vp)
 830                                                 return -ENOMEM;
 831
 832                                         PickFilter filter = {
 833                                                 .type_mask = endswith(suffix, ".raw") ? (UINT32_C(1) << DT_REG) | (UINT32_C(1) << DT_BLK) : (UINT32_C(1) << DT_DIR),
 834                                                 .basename = pretty,
 835                                                 .architecture = _ARCHITECTURE_INVALID,
 836                                                 .suffix = STRV_MAKE(suffix),
 837                                         };
 838
 839                                         _cleanup_(pick_result_done) PickResult result = PICK_RESULT_NULL;
 840                                         r = path_pick(root,
 841                                                       /* toplevel_fd= */ AT_FDCWD,
 842                                                       vp,
 843                                                       &filter,
 844                                                       PICK_ARCHITECTURE|PICK_TRIES,
 845                                                       &result);
 846                                         if (r < 0) {
 847                                                 log_debug_errno(r, "Failed to pick versioned image on '%s', skipping: %m", vp);
 848                                                 continue;
 849                                         }
 850                                         if (!result.path) {
 851                                                 log_debug("Found versioned directory '%s', without matching entry, skipping: %m", vp);
 852                                                 continue;
 853                                         }
 854
 855                                         /* Refresh the stat data for the discovered target */
 856                                         st = result.st;
 857
 858                                         _cleanup_free_ char *bn = NULL;
 859                                         r = path_extract_filename(result.path, &bn);
 860                                         if (r < 0) {
 861                                                 log_debug_errno(r, "Failed to extract basename of image path '%s', skipping: %m", result.path);
 862                                                 continue;
 863                                         }
 864
 865                                         fname_buf = path_join(fname, bn);
 866                                         if (!fname_buf)
 867                                                 return log_oom();
 868
 869                                         fname = fname_buf;
 870                                 } else {
 871                                         r = extract_image_basename(
 872                                                         fname,
 873                                                         image_class_suffix_to_string(class),
 874                                                         /* format_suffix= */ NULL,
 875                                                         &pretty,
 876                                                         /* ret_suffix= */ NULL);
 877                                         if (r < 0) {
 878                                                 log_debug_errno(r, "Skipping directory entry '%s', which doesn't look like an image.", fname);
 879                                                 continue;
 880                                         }
 881                                 }
 882
 883                         } else if (S_ISBLK(st.st_mode)) {
 884                                 r = extract_image_basename(
 885                                                 fname,
 886                                                 /* class_suffix= */ NULL,
 887                                                 /* format_suffix= */ NULL,
 888                                                 &pretty,
 889                                                 /* ret_v_suffix= */ NULL);
 890                                 if (r < 0) {
 891                                         log_debug_errno(r, "Skipping directory entry '%s', which doesn't look like an image.", fname);
 892                                         continue;
 893                                 }
 894                         } else {
 895                                 log_debug("Skipping directory entry '%s', which is neither regular file, directory nor block device.", fname);
 896                                 continue;
 897                         }
 898
 899                         if (hashmap_contains(h, pretty))
 900                                 continue;
 901
 902                         r = image_make(class, pretty, dirfd(d), resolved, fname, &st, &image);
 903                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 904                                 continue;
 905                         if (r < 0)
 906                                 return r;
 907
 908                         image->discoverable = true;
 909
 910                         r = hashmap_put(h, image->name, image);
 911                         if (r < 0)
 912                                 return r;
 913
 914                         TAKE_PTR(image);
 915                 }
 916         }
 917
 918         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 919                 _cleanup_(image_unrefp) Image *image = NULL;
 920
 921                 r = image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image);
 922                 if (r < 0)
 923                         return r;
 924
 925                 image->discoverable = true;
 926
 927                 r = hashmap_put(h, image->name, image);
 928                 if (r < 0)
 929                         return r;
 930
 931                 image = NULL;
 932         }
 933
 934         return 0;
 935 }
 936
 937 int image_remove(Image *i) {
 938         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 939         _cleanup_strv_free_ char **settings = NULL;
 940         _cleanup_free_ char *roothash = NULL;
 941         int r;
 942
 943         assert(i);
 944
 945         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 946                 return -EROFS;
 947
 948         settings = image_settings_path(i);
 949         if (!settings)
 950                 return -ENOMEM;
 951
 952         r = image_roothash_path(i, &roothash);
 953         if (r < 0)
 954                 return r;
 955
 956         /* Make sure we don't interfere with a running nspawn */
 957         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 958         if (r < 0)
 959                 return r;
 960
 961         switch (i->type) {
 962
 963         case IMAGE_SUBVOLUME:
 964
 965                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 966                  * big guns */
 967                 if (unlink(i->path) < 0) {
 968                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 969                         if (r < 0)
 970                                 return r;
 971                 }
 972
 973                 break;
 974
 975         case IMAGE_DIRECTORY:
 976                 /* Allow deletion of read-only directories */
 977                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 978                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 979                 if (r < 0)
 980                         return r;
 981
 982                 break;
 983
 984         case IMAGE_BLOCK:
 985
 986                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 987                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 988                  * the thing (it's most likely a symlink after all). */
 989
 990                 if (path_startswith(i->path, "/dev"))
 991                         break;
 992
 993                 _fallthrough_;
 994         case IMAGE_RAW:
 995                 if (unlink(i->path) < 0)
 996                         return -errno;
 997                 break;
 998
 999         default:
1000                 return -EOPNOTSUPP;
1001         }
1002
1003         STRV_FOREACH(j, settings)
1004                 if (unlink(*j) < 0 && errno != ENOENT)
1005                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
1006
1007         if (unlink(roothash) < 0 && errno != ENOENT)
1008                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
1009
1010         return 0;
1011 }
1012
1013 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
1014         _cleanup_free_ char *fn = NULL, *rs = NULL;
1015         int r;
1016
1017         fn = strjoin(new_name, suffix);
1018         if (!fn)
1019                 return -ENOMEM;
1020
1021         r = file_in_same_dir(path, fn, &rs);
1022         if (r < 0)
1023                 return r;
1024
1025         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
1026 }
1027
1028 int image_rename(Image *i, const char *new_name) {
1029         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
1030         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
1031         _cleanup_strv_free_ char **settings = NULL;
1032         unsigned file_attr = 0;
1033         int r;
1034
1035         assert(i);
1036
1037         if (!image_name_is_valid(new_name))
1038                 return -EINVAL;
1039
1040         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1041                 return -EROFS;
1042
1043         settings = image_settings_path(i);
1044         if (!settings)
1045                 return -ENOMEM;
1046
1047         r = image_roothash_path(i, &roothash);
1048         if (r < 0)
1049                 return r;
1050
1051         /* Make sure we don't interfere with a running nspawn */
1052         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
1053         if (r < 0)
1054                 return r;
1055
1056         /* Make sure nobody takes the new name, between the time we
1057          * checked it is currently unused in all search paths, and the
1058          * time we take possession of it */
1059         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
1060         if (r < 0)
1061                 return r;
1062
1063         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
1064         if (r >= 0)
1065                 return -EEXIST;
1066         if (r != -ENOENT)
1067                 return r;
1068
1069         switch (i->type) {
1070
1071         case IMAGE_DIRECTORY:
1072                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
1073                 (void) read_attr_at(AT_FDCWD, i->path, &file_attr);
1074
1075                 if (file_attr & FS_IMMUTABLE_FL)
1076                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
1077
1078                 _fallthrough_;
1079         case IMAGE_SUBVOLUME:
1080                 r = file_in_same_dir(i->path, new_name, &new_path);
1081                 break;
1082
1083         case IMAGE_BLOCK:
1084
1085                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
1086                 if (path_startswith(i->path, "/dev"))
1087                         return -EROFS;
1088
1089                 r = file_in_same_dir(i->path, new_name, &new_path);
1090                 break;
1091
1092         case IMAGE_RAW: {
1093                 const char *fn;
1094
1095                 fn = strjoina(new_name, ".raw");
1096
1097                 r = file_in_same_dir(i->path, fn, &new_path);
1098                 break;
1099         }
1100
1101         default:
1102                 return -EOPNOTSUPP;
1103         }
1104         if (r < 0)
1105                 return r;
1106
1107         nn = strdup(new_name);
1108         if (!nn)
1109                 return -ENOMEM;
1110
1111         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
1112         if (r < 0)
1113                 return r;
1114
1115         /* Restore the immutable bit, if it was set before */
1116         if (file_attr & FS_IMMUTABLE_FL)
1117                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
1118
1119         free_and_replace(i->path, new_path);
1120         free_and_replace(i->name, nn);
1121
1122         STRV_FOREACH(j, settings) {
1123                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
1124                 if (r < 0 && r != -ENOENT)
1125                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
1126         }
1127
1128         r = rename_auxiliary_file(roothash, new_name, ".roothash");
1129         if (r < 0 && r != -ENOENT)
1130                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
1131
1132         return 0;
1133 }
1134
1135 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
1136         _cleanup_free_ char *fn = NULL, *rs = NULL;
1137         int r;
1138
1139         fn = strjoin(new_name, suffix);
1140         if (!fn)
1141                 return -ENOMEM;
1142
1143         r = file_in_same_dir(path, fn, &rs);
1144         if (r < 0)
1145                 return r;
1146
1147         return copy_file_atomic(path, rs, 0664, COPY_REFLINK);
1148 }
1149
1150 int image_clone(Image *i, const char *new_name, bool read_only) {
1151         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
1152         _cleanup_strv_free_ char **settings = NULL;
1153         _cleanup_free_ char *roothash = NULL;
1154         const char *new_path;
1155         int r;
1156
1157         assert(i);
1158
1159         if (!image_name_is_valid(new_name))
1160                 return -EINVAL;
1161
1162         settings = image_settings_path(i);
1163         if (!settings)
1164                 return -ENOMEM;
1165
1166         r = image_roothash_path(i, &roothash);
1167         if (r < 0)
1168                 return r;
1169
1170         /* Make sure nobody takes the new name, between the time we
1171          * checked it is currently unused in all search paths, and the
1172          * time we take possession of it */
1173         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
1174         if (r < 0)
1175                 return r;
1176
1177         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
1178         if (r >= 0)
1179                 return -EEXIST;
1180         if (r != -ENOENT)
1181                 return r;
1182
1183         switch (i->type) {
1184
1185         case IMAGE_SUBVOLUME:
1186         case IMAGE_DIRECTORY:
1187                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
1188                  * directory. */
1189
1190                 new_path = strjoina("/var/lib/machines/", new_name);
1191
1192                 r = btrfs_subvol_snapshot_at(AT_FDCWD, i->path, AT_FDCWD, new_path,
1193                                              (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
1194                                              BTRFS_SNAPSHOT_FALLBACK_COPY |
1195                                              BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
1196                                              BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
1197                                              BTRFS_SNAPSHOT_RECURSIVE |
1198                                              BTRFS_SNAPSHOT_QUOTA);
1199                 if (r >= 0)
1200                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
1201                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
1202
1203                 break;
1204
1205         case IMAGE_RAW:
1206                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
1207
1208                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644,
1209                                      COPY_REFLINK|COPY_CRTIME|COPY_NOCOW_AFTER);
1210                 break;
1211
1212         case IMAGE_BLOCK:
1213         default:
1214                 return -EOPNOTSUPP;
1215         }
1216
1217         if (r < 0)
1218                 return r;
1219
1220         STRV_FOREACH(j, settings) {
1221                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
1222                 if (r < 0 && r != -ENOENT)
1223                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
1224         }
1225
1226         r = clone_auxiliary_file(roothash, new_name, ".roothash");
1227         if (r < 0 && r != -ENOENT)
1228                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
1229
1230         return 0;
1231 }
1232
1233 int image_read_only(Image *i, bool b) {
1234         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1235         int r;
1236
1237         assert(i);
1238
1239         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1240                 return -EROFS;
1241
1242         /* Make sure we don't interfere with a running nspawn */
1243         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
1244         if (r < 0)
1245                 return r;
1246
1247         switch (i->type) {
1248
1249         case IMAGE_SUBVOLUME:
1250
1251                 /* Note that we set the flag only on the top-level
1252                  * subvolume of the image. */
1253
1254                 r = btrfs_subvol_set_read_only(i->path, b);
1255                 if (r < 0)
1256                         return r;
1257
1258                 break;
1259
1260         case IMAGE_DIRECTORY:
1261                 /* For simple directory trees we cannot use the access
1262                    mode of the top-level directory, since it has an
1263                    effect on the container itself.  However, we can
1264                    use the "immutable" flag, to at least make the
1265                    top-level directory read-only. It's not as good as
1266                    a read-only subvolume, but at least something, and
1267                    we can read the value back. */
1268
1269                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
1270                 if (r < 0)
1271                         return r;
1272
1273                 break;
1274
1275         case IMAGE_RAW: {
1276                 struct stat st;
1277
1278                 if (stat(i->path, &st) < 0)
1279                         return -errno;
1280
1281                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
1282                         return -errno;
1283
1284                 /* If the images is now read-only, it's a good time to
1285                  * defrag it, given that no write patterns will
1286                  * fragment it again. */
1287                 if (b)
1288                         (void) btrfs_defrag(i->path);
1289                 break;
1290         }
1291
1292         case IMAGE_BLOCK: {
1293                 _cleanup_close_ int fd = -EBADF;
1294                 struct stat st;
1295                 int state = b;
1296
1297                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
1298                 if (fd < 0)
1299                         return -errno;
1300
1301                 if (fstat(fd, &st) < 0)
1302                         return -errno;
1303                 if (!S_ISBLK(st.st_mode))
1304                         return -ENOTTY;
1305
1306                 if (ioctl(fd, BLKROSET, &state) < 0)
1307                         return -errno;
1308
1309                 break;
1310         }
1311
1312         default:
1313                 return -EOPNOTSUPP;
1314         }
1315
1316         i->read_only = b;
1317         return 0;
1318 }
1319
1320 static void make_lock_dir(void) {
1321         (void) mkdir_p("/run/systemd/nspawn", 0755);
1322         (void) mkdir("/run/systemd/nspawn/locks", 0700);
1323 }
1324
1325 int image_path_lock(
1326                 const char *path,
1327                 int operation,
1328                 LockFile *ret_global,
1329                 LockFile *ret_local) {
1330
1331         _cleanup_free_ char *p = NULL;
1332         LockFile t = LOCK_FILE_INIT;
1333         struct stat st;
1334         bool exclusive;
1335         int r;
1336
1337         assert(path);
1338         assert(ret_local);
1339
1340         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1341          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1342          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1343          * correctly. */
1344
1345         if (!path_is_absolute(path))
1346                 return -EINVAL;
1347
1348         switch (operation & (LOCK_SH|LOCK_EX)) {
1349         case LOCK_SH:
1350                 exclusive = false;
1351                 break;
1352         case LOCK_EX:
1353                 exclusive = true;
1354                 break;
1355         default:
1356                 return -EINVAL;
1357         }
1358
1359         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1360                 *ret_local = LOCK_FILE_INIT;
1361                 if (ret_global)
1362                         *ret_global = LOCK_FILE_INIT;
1363                 return 0;
1364         }
1365
1366         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1367          * running off it after all, and we don't want any images to manipulate the host image. We make an
1368          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1369          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1370          * since it means changes made to the host might propagate to the container as they happen (and a
1371          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1372          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1373          * the user to do the right thing with this. */
1374         if (path_equal(path, "/")) {
1375                 if (exclusive)
1376                         return -EBUSY;
1377
1378                 *ret_local = LOCK_FILE_INIT;
1379                 if (ret_global)
1380                         *ret_global = LOCK_FILE_INIT;
1381                 return 0;
1382         }
1383
1384         if (ret_global) {
1385                 if (stat(path, &st) >= 0) {
1386                         if (S_ISBLK(st.st_mode))
1387                                 r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1388                         else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1389                                 r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1390                         else
1391                                 return -ENOTTY;
1392                         if (r < 0)
1393                                 return -ENOMEM;
1394                 }
1395         }
1396
1397         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1398          * sufficient, since block devices are host local anyway. */
1399         if (!path_startswith(path, "/dev/")) {
1400                 r = make_lock_file_for(path, operation, &t);
1401                 if (r < 0) {
1402                         if (!exclusive && r == -EROFS)
1403                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1404                         else
1405                                 return r;
1406                 }
1407         }
1408
1409         if (p) {
1410                 make_lock_dir();
1411
1412                 r = make_lock_file(p, operation, ret_global);
1413                 if (r < 0) {
1414                         release_lock_file(&t);
1415                         return r;
1416                 }
1417         } else if (ret_global)
1418                 *ret_global = LOCK_FILE_INIT;
1419
1420         *ret_local = t;
1421         return 0;
1422 }
1423
1424 int image_set_limit(Image *i, uint64_t referenced_max) {
1425         int r;
1426
1427         assert(i);
1428
1429         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1430                 return -EROFS;
1431
1432         if (i->type != IMAGE_SUBVOLUME)
1433                 return -EOPNOTSUPP;
1434
1435         /* We set the quota both for the subvolume as well as for the
1436          * subtree. The latter is mostly for historical reasons, since
1437          * we didn't use to have a concept of subtree quota, and hence
1438          * only modified the subvolume quota. */
1439
1440         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1441         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1442         r = btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1443         if (r < 0)
1444                 return r;
1445
1446         (void) image_update_quota(i, -EBADF);
1447         return 0;
1448 }
1449
1450 int image_read_metadata(Image *i, const ImagePolicy *image_policy) {
1451         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1452         int r;
1453
1454         assert(i);
1455
1456         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1457         if (r < 0)
1458                 return r;
1459
1460         switch (i->type) {
1461
1462         case IMAGE_SUBVOLUME:
1463         case IMAGE_DIRECTORY: {
1464                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **sysext_release = NULL, **confext_release = NULL;
1465                 _cleanup_free_ char *hostname = NULL, *path = NULL;
1466                 sd_id128_t machine_id = SD_ID128_NULL;
1467
1468                 if (i->class == IMAGE_SYSEXT) {
1469                         r = extension_has_forbidden_content(i->path);
1470                         if (r < 0)
1471                                 return r;
1472                         if (r > 0)
1473                                 return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM),
1474                                                        "Conflicting content found in image %s, refusing.",
1475                                                        i->name);
1476                 }
1477
1478                 r = chase("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1479                 if (r < 0 && r != -ENOENT)
1480                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1481                 else if (r >= 0) {
1482                         r = read_etc_hostname(path, &hostname);
1483                         if (r < 0)
1484                                 log_debug_errno(r, "Failed to read /etc/hostname of image %s: %m", i->name);
1485                 }
1486
1487                 path = mfree(path);
1488
1489                 r = id128_get_machine(i->path, &machine_id);
1490                 if (r < 0)
1491                         log_debug_errno(r, "Failed to read machine ID in image %s, ignoring: %m", i->name);
1492
1493                 r = chase("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1494                 if (r < 0 && r != -ENOENT)
1495                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1496                 else if (r >= 0) {
1497                         r = load_env_file_pairs(NULL, path, &machine_info);
1498                         if (r < 0)
1499                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1500                 }
1501
1502                 r = load_os_release_pairs(i->path, &os_release);
1503                 if (r < 0)
1504                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1505
1506                 r = load_extension_release_pairs(i->path, IMAGE_SYSEXT, i->name, /* relax_extension_release_check= */ false, &sysext_release);
1507                 if (r < 0)
1508                         log_debug_errno(r, "Failed to read sysext-release in image, ignoring: %m");
1509
1510                 r = load_extension_release_pairs(i->path, IMAGE_CONFEXT, i->name, /* relax_extension_release_check= */ false, &confext_release);
1511                 if (r < 0)
1512                         log_debug_errno(r, "Failed to read confext-release in image, ignoring: %m");
1513
1514                 free_and_replace(i->hostname, hostname);
1515                 i->machine_id = machine_id;
1516                 strv_free_and_replace(i->machine_info, machine_info);
1517                 strv_free_and_replace(i->os_release, os_release);
1518                 strv_free_and_replace(i->sysext_release, sysext_release);
1519                 strv_free_and_replace(i->confext_release, confext_release);
1520                 break;
1521         }
1522
1523         case IMAGE_RAW:
1524         case IMAGE_BLOCK: {
1525                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1526                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1527                 DissectImageFlags flags =
1528                         DISSECT_IMAGE_GENERIC_ROOT |
1529                         DISSECT_IMAGE_REQUIRE_ROOT |
1530                         DISSECT_IMAGE_RELAX_VAR_CHECK |
1531                         DISSECT_IMAGE_READ_ONLY |
1532                         DISSECT_IMAGE_USR_NO_ROOT |
1533                         DISSECT_IMAGE_ADD_PARTITION_DEVICES |
1534                         DISSECT_IMAGE_PIN_PARTITION_DEVICES |
1535                         DISSECT_IMAGE_VALIDATE_OS |
1536                         DISSECT_IMAGE_VALIDATE_OS_EXT |
1537                         DISSECT_IMAGE_ALLOW_USERSPACE_VERITY;
1538
1539                 r = loop_device_make_by_path(
1540                                 i->path,
1541                                 O_RDONLY,
1542                                 /* sector_size= */ UINT32_MAX,
1543                                 LO_FLAGS_PARTSCAN,
1544                                 LOCK_SH,
1545                                 &d);
1546                 if (r < 0)
1547                         return r;
1548
1549                 r = dissect_loop_device(
1550                                 d,
1551                                 /* verity= */ NULL,
1552                                 /* mount_options= */ NULL,
1553                                 image_policy,
1554                                 flags,
1555                                 &m);
1556                 if (r < 0)
1557                         return r;
1558
1559                 r = dissected_image_acquire_metadata(
1560                                 m,
1561                                 /* userns_fd= */ -EBADF,
1562                                 flags);
1563                 if (r < 0)
1564                         return r;
1565
1566                 free_and_replace(i->hostname, m->hostname);
1567                 i->machine_id = m->machine_id;
1568                 strv_free_and_replace(i->machine_info, m->machine_info);
1569                 strv_free_and_replace(i->os_release, m->os_release);
1570                 strv_free_and_replace(i->sysext_release, m->sysext_release);
1571                 strv_free_and_replace(i->confext_release, m->confext_release);
1572
1573                 break;
1574         }
1575
1576         default:
1577                 return -EOPNOTSUPP;
1578         }
1579
1580         i->metadata_valid = true;
1581
1582         return 0;
1583 }
1584
1585 int image_name_lock(const char *name, int operation, LockFile *ret) {
1586         const char *p;
1587
1588         assert(name);
1589         assert(ret);
1590
1591         /* Locks an image name, regardless of the precise path used. */
1592
1593         if (streq(name, ".host"))
1594                 return -EBUSY;
1595
1596         if (!image_name_is_valid(name))
1597                 return -EINVAL;
1598
1599         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1600                 *ret = (LockFile) LOCK_FILE_INIT;
1601                 return 0;
1602         }
1603
1604         make_lock_dir();
1605
1606         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1607         return make_lock_file(p, operation, ret);
1608 }
1609
1610 bool image_in_search_path(
1611                 ImageClass class,
1612                 const char *root,
1613                 const char *image) {
1614
1615         assert(image);
1616
1617         NULSTR_FOREACH(path, pick_image_search_path(class)) {
1618                 const char *p, *q;
1619                 size_t k;
1620
1621                 if (!empty_or_root(root)) {
1622                         q = path_startswith(path, root);
1623                         if (!q)
1624                                 continue;
1625                 } else
1626                         q = path;
1627
1628                 p = path_startswith(q, path);
1629                 if (!p)
1630                         continue;
1631
1632                 /* Make sure there's a filename following */
1633                 k = strcspn(p, "/");
1634                 if (k == 0)
1635                         continue;
1636
1637                 p += k;
1638
1639                 /* Accept trailing slashes */
1640                 if (p[strspn(p, "/")] == 0)
1641                         return true;
1642         }
1643
1644         return false;
1645 }
1646
1647 int image_to_json(const struct Image *img, sd_json_variant **ret) {
1648         assert(img);
1649
1650         return sd_json_buildo(
1651                         ret,
1652                         SD_JSON_BUILD_PAIR_STRING("Type", image_type_to_string(img->type)),
1653                         SD_JSON_BUILD_PAIR_STRING("Class", image_class_to_string(img->class)),
1654                         SD_JSON_BUILD_PAIR_STRING("Name", img->name),
1655                         SD_JSON_BUILD_PAIR_CONDITION(!!img->path, "Path", SD_JSON_BUILD_STRING(img->path)),
1656                         SD_JSON_BUILD_PAIR_BOOLEAN("ReadOnly", img->read_only),
1657                         SD_JSON_BUILD_PAIR_CONDITION(img->crtime != 0, "CreationTimestamp", SD_JSON_BUILD_UNSIGNED(img->crtime)),
1658                         SD_JSON_BUILD_PAIR_CONDITION(img->mtime != 0, "ModificationTimestamp", SD_JSON_BUILD_UNSIGNED(img->mtime)),
1659                         SD_JSON_BUILD_PAIR_CONDITION(img->usage != UINT64_MAX, "Usage", SD_JSON_BUILD_UNSIGNED(img->usage)),
1660                         SD_JSON_BUILD_PAIR_CONDITION(img->usage_exclusive != UINT64_MAX, "UsageExclusive", SD_JSON_BUILD_UNSIGNED(img->usage_exclusive)),
1661                         SD_JSON_BUILD_PAIR_CONDITION(img->limit != UINT64_MAX, "Limit", SD_JSON_BUILD_UNSIGNED(img->limit)),
1662                         SD_JSON_BUILD_PAIR_CONDITION(img->limit_exclusive != UINT64_MAX, "LimitExclusive", SD_JSON_BUILD_UNSIGNED(img->limit_exclusive)));
1663 }
1664
1665 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1666         [IMAGE_DIRECTORY] = "directory",
1667         [IMAGE_SUBVOLUME] = "subvolume",
1668         [IMAGE_RAW]       = "raw",
1669         [IMAGE_BLOCK]     = "block",
1670 };
1671
1672 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);