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