src/shared/discover-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <errno.h>
   4 #include <fcntl.h>
   5 #include <linux/fs.h>
   6 #include <linux/loop.h>
   7 #include <linux/magic.h>
   8 #include <stdio.h>
   9 #include <stdlib.h>
  10 #include <sys/file.h>
  11 #include <sys/ioctl.h>
  12 #include <sys/stat.h>
  13 #include <unistd.h>
  14
  15 #include "alloc-util.h"
  16 #include "btrfs-util.h"
  17 #include "chase.h"
  18 #include "chattr-util.h"
  19 #include "copy.h"
  20 #include "dirent-util.h"
  21 #include "discover-image.h"
  22 #include "dissect-image.h"
  23 #include "env-file.h"
  24 #include "env-util.h"
  25 #include "extension-util.h"
  26 #include "fd-util.h"
  27 #include "fs-util.h"
  28 #include "hashmap.h"
  29 #include "hostname-setup.h"
  30 #include "id128-util.h"
  31 #include "lock-util.h"
  32 #include "log.h"
  33 #include "loop-util.h"
  34 #include "macro.h"
  35 #include "mkdir.h"
  36 #include "nulstr-util.h"
  37 #include "os-util.h"
  38 #include "path-util.h"
  39 #include "rm-rf.h"
  40 #include "stat-util.h"
  41 #include "string-table.h"
  42 #include "string-util.h"
  43 #include "strv.h"
  44 #include "time-util.h"
  45 #include "utf8.h"
  46 #include "xattr-util.h"
  47
  48 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  49         [IMAGE_MACHINE] =   "/etc/machines\0"              /* only place symlinks here */
  50                             "/run/machines\0"              /* and here too */
  51                             "/var/lib/machines\0"          /* the main place for images */
  52                             "/var/lib/container\0"         /* legacy */
  53                             "/usr/local/lib/machines\0"
  54                             "/usr/lib/machines\0",
  55
  56         [IMAGE_PORTABLE] =  "/etc/portables\0"             /* only place symlinks here */
  57                             "/run/portables\0"             /* and here too */
  58                             "/var/lib/portables\0"         /* the main place for images */
  59                             "/usr/local/lib/portables\0"
  60                             "/usr/lib/portables\0",
  61
  62         /* Note that we don't allow storing extensions under /usr/, unlike with other image types. That's
  63          * because extension images are supposed to extend /usr/, so you get into recursive races, especially
  64          * with directory-based extensions, as the kernel's OverlayFS explicitly checks for this and errors
  65          * out with -ELOOP if it finds that a lowerdir= is a child of another lowerdir=. */
  66         [IMAGE_EXTENSION] = "/etc/extensions\0"             /* only place symlinks here */
  67                             "/run/extensions\0"             /* and here too */
  68                             "/var/lib/extensions\0",        /* the main place for images */
  69 };
  70
  71 static Image *image_free(Image *i) {
  72         assert(i);
  73
  74         free(i->name);
  75         free(i->path);
  76
  77         free(i->hostname);
  78         strv_free(i->machine_info);
  79         strv_free(i->os_release);
  80         strv_free(i->extension_release);
  81
  82         return mfree(i);
  83 }
  84
  85 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  86 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
  87                                       Image, image_unref);
  88
  89 static char **image_settings_path(Image *image) {
  90         _cleanup_strv_free_ char **l = NULL;
  91         _cleanup_free_ char *fn = NULL;
  92         size_t i = 0;
  93         int r;
  94
  95         assert(image);
  96
  97         l = new0(char*, 4);
  98         if (!l)
  99                 return NULL;
 100
 101         fn = strjoin(image->name, ".nspawn");
 102         if (!fn)
 103                 return NULL;
 104
 105         FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
 106                 l[i] = path_join(s, fn);
 107                 if (!l[i])
 108                         return NULL;
 109
 110                 i++;
 111         }
 112
 113         r = file_in_same_dir(image->path, fn, l + i);
 114         if (r == -ENOMEM)
 115                 return NULL;
 116         if (r < 0)
 117                 log_debug_errno(r, "Failed to generate .nspawn settings path from image path, ignoring: %m");
 118
 119         strv_uniq(l);
 120
 121         return TAKE_PTR(l);
 122 }
 123
 124 static int image_roothash_path(Image *image, char **ret) {
 125         _cleanup_free_ char *fn = NULL;
 126
 127         assert(image);
 128
 129         fn = strjoin(image->name, ".roothash");
 130         if (!fn)
 131                 return -ENOMEM;
 132
 133         return file_in_same_dir(image->path, fn, ret);
 134 }
 135
 136 static int image_new(
 137                 ImageType t,
 138                 ImageClass c,
 139                 const char *pretty,
 140                 const char *path,
 141                 const char *filename,
 142                 bool read_only,
 143                 usec_t crtime,
 144                 usec_t mtime,
 145                 Image **ret) {
 146
 147         _cleanup_(image_unrefp) Image *i = NULL;
 148
 149         assert(t >= 0);
 150         assert(t < _IMAGE_TYPE_MAX);
 151         assert(pretty);
 152         assert(filename);
 153         assert(ret);
 154
 155         i = new(Image, 1);
 156         if (!i)
 157                 return -ENOMEM;
 158
 159         *i = (Image) {
 160                 .n_ref = 1,
 161                 .type = t,
 162                 .class = c,
 163                 .read_only = read_only,
 164                 .crtime = crtime,
 165                 .mtime = mtime,
 166                 .usage = UINT64_MAX,
 167                 .usage_exclusive = UINT64_MAX,
 168                 .limit = UINT64_MAX,
 169                 .limit_exclusive = UINT64_MAX,
 170         };
 171
 172         i->name = strdup(pretty);
 173         if (!i->name)
 174                 return -ENOMEM;
 175
 176         i->path = path_join(path, filename);
 177         if (!i->path)
 178                 return -ENOMEM;
 179
 180         path_simplify(i->path);
 181
 182         *ret = TAKE_PTR(i);
 183
 184         return 0;
 185 }
 186
 187 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 188         _cleanup_free_ char *name = NULL;
 189         const char *p;
 190
 191         assert(path);
 192         assert(ret);
 193
 194         p = last_path_component(path);
 195
 196         name = strdupcspn(p, "/");
 197         if (!name)
 198                 return -ENOMEM;
 199
 200         if (suffix) {
 201                 char *e;
 202
 203                 e = endswith(name, suffix);
 204                 if (!e)
 205                         return -EINVAL;
 206
 207                 *e = 0;
 208         }
 209
 210         if (!image_name_is_valid(name))
 211                 return -EINVAL;
 212
 213         *ret = TAKE_PTR(name);
 214         return 0;
 215 }
 216
 217 static int image_make(
 218                 ImageClass c,
 219                 const char *pretty,
 220                 int dfd,
 221                 const char *path,
 222                 const char *filename,
 223                 const struct stat *st,
 224                 Image **ret) {
 225
 226         _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL;
 227         struct stat stbuf;
 228         bool read_only;
 229         int r;
 230
 231         assert(dfd >= 0 || dfd == AT_FDCWD);
 232         assert(path || dfd == AT_FDCWD);
 233         assert(filename);
 234
 235         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 236          * devices into /var/lib/machines/, and treat them normally.
 237          *
 238          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 239          * recognize. */
 240
 241         if (!st) {
 242                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 243                         return -errno;
 244
 245                 st = &stbuf;
 246         }
 247
 248         if (!path) {
 249                 if (dfd == AT_FDCWD)
 250                         (void) safe_getcwd(&parent);
 251                 else
 252                         (void) fd_get_path(dfd, &parent);
 253         }
 254
 255         read_only =
 256                 (path && path_startswith(path, "/usr")) ||
 257                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 258
 259         if (S_ISDIR(st->st_mode)) {
 260                 _cleanup_close_ int fd = -EBADF;
 261                 unsigned file_attr = 0;
 262                 usec_t crtime = 0;
 263
 264                 if (!ret)
 265                         return 0;
 266
 267                 if (!pretty) {
 268                         r = extract_pretty(filename, NULL, &pretty_buffer);
 269                         if (r < 0)
 270                                 return r;
 271
 272                         pretty = pretty_buffer;
 273                 }
 274
 275                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 276                 if (fd < 0)
 277                         return -errno;
 278
 279                 if (btrfs_might_be_subvol(st)) {
 280
 281                         r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
 282                         if (r < 0)
 283                                 return r;
 284                         if (r) {
 285                                 BtrfsSubvolInfo info;
 286
 287                                 /* It's a btrfs subvolume */
 288
 289                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 290                                 if (r < 0)
 291                                         return r;
 292
 293                                 r = image_new(IMAGE_SUBVOLUME,
 294                                               c,
 295                                               pretty,
 296                                               path,
 297                                               filename,
 298                                               info.read_only || read_only,
 299                                               info.otime,
 300                                               0,
 301                                               ret);
 302                                 if (r < 0)
 303                                         return r;
 304
 305                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 306                                         BtrfsQuotaInfo quota;
 307
 308                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 309                                         if (r >= 0) {
 310                                                 (*ret)->usage = quota.referenced;
 311                                                 (*ret)->usage_exclusive = quota.exclusive;
 312
 313                                                 (*ret)->limit = quota.referenced_max;
 314                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 315                                         }
 316                                 }
 317
 318                                 return 0;
 319                         }
 320                 }
 321
 322                 /* Get directory creation time (not available everywhere, but that's OK */
 323                 (void) fd_getcrtime(fd, &crtime);
 324
 325                 /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not
 326                  * supported everywhere we ignore failures. */
 327                 (void) read_attr_fd(fd, &file_attr);
 328
 329                 /* It's just a normal directory. */
 330                 r = image_new(IMAGE_DIRECTORY,
 331                               c,
 332                               pretty,
 333                               path,
 334                               filename,
 335                               read_only || (file_attr & FS_IMMUTABLE_FL),
 336                               crtime,
 337                               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 */
 338                               ret);
 339                 if (r < 0)
 340                         return r;
 341
 342                 return 0;
 343
 344         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 345                 usec_t crtime = 0;
 346
 347                 /* It's a RAW disk image */
 348
 349                 if (!ret)
 350                         return 0;
 351
 352                 (void) fd_getcrtime_at(dfd, filename, AT_SYMLINK_FOLLOW, &crtime);
 353
 354                 if (!pretty) {
 355                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 356                         if (r < 0)
 357                                 return r;
 358
 359                         pretty = pretty_buffer;
 360                 }
 361
 362                 r = image_new(IMAGE_RAW,
 363                               c,
 364                               pretty,
 365                               path,
 366                               filename,
 367                               !(st->st_mode & 0222) || read_only,
 368                               crtime,
 369                               timespec_load(&st->st_mtim),
 370                               ret);
 371                 if (r < 0)
 372                         return r;
 373
 374                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 375                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 376
 377                 return 0;
 378
 379         } else if (S_ISBLK(st->st_mode)) {
 380                 _cleanup_close_ int block_fd = -EBADF;
 381                 uint64_t size = UINT64_MAX;
 382
 383                 /* A block device */
 384
 385                 if (!ret)
 386                         return 0;
 387
 388                 if (!pretty) {
 389                         r = extract_pretty(filename, NULL, &pretty_buffer);
 390                         if (r < 0)
 391                                 return r;
 392
 393                         pretty = pretty_buffer;
 394                 }
 395
 396                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 397                 if (block_fd < 0)
 398                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 399                 else {
 400                         /* Refresh stat data after opening the node */
 401                         if (fstat(block_fd, &stbuf) < 0)
 402                                 return -errno;
 403                         st = &stbuf;
 404
 405                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 406                                 return -ENOTTY;
 407
 408                         if (!read_only) {
 409                                 int state = 0;
 410
 411                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 412                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 413                                 else if (state)
 414                                         read_only = true;
 415                         }
 416
 417                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 418                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 419
 420                         block_fd = safe_close(block_fd);
 421                 }
 422
 423                 r = image_new(IMAGE_BLOCK,
 424                               c,
 425                               pretty,
 426                               path,
 427                               filename,
 428                               !(st->st_mode & 0222) || read_only,
 429                               0,
 430                               0,
 431                               ret);
 432                 if (r < 0)
 433                         return r;
 434
 435                 if (!IN_SET(size, 0, UINT64_MAX))
 436                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 437
 438                 return 0;
 439         }
 440
 441         return -EMEDIUMTYPE;
 442 }
 443
 444 int image_find(ImageClass class,
 445                const char *name,
 446                const char *root,
 447                Image **ret) {
 448
 449         int r;
 450
 451         assert(class >= 0);
 452         assert(class < _IMAGE_CLASS_MAX);
 453         assert(name);
 454
 455         /* There are no images with invalid names */
 456         if (!image_name_is_valid(name))
 457                 return -ENOENT;
 458
 459         NULSTR_FOREACH(path, image_search_path[class]) {
 460                 _cleanup_free_ char *resolved = NULL;
 461                 _cleanup_closedir_ DIR *d = NULL;
 462                 struct stat st;
 463                 int flags;
 464
 465                 r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 466                 if (r == -ENOENT)
 467                         continue;
 468                 if (r < 0)
 469                         return r;
 470
 471                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 472                  * to symlink block devices into the search path. (For now, we disable that when operating
 473                  * relative to some root directory.) */
 474                 flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 475                 if (fstatat(dirfd(d), name, &st, flags) < 0) {
 476                         _cleanup_free_ char *raw = NULL;
 477
 478                         if (errno != ENOENT)
 479                                 return -errno;
 480
 481                         raw = strjoin(name, ".raw");
 482                         if (!raw)
 483                                 return -ENOMEM;
 484
 485                         if (fstatat(dirfd(d), raw, &st, flags) < 0) {
 486                                 if (errno == ENOENT)
 487                                         continue;
 488
 489                                 return -errno;
 490                         }
 491
 492                         if (!S_ISREG(st.st_mode))
 493                                 continue;
 494
 495                         r = image_make(class, name, dirfd(d), resolved, raw, &st, ret);
 496
 497                 } else {
 498                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 499                                 continue;
 500
 501                         r = image_make(class, name, dirfd(d), resolved, name, &st, ret);
 502                 }
 503                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 504                         continue;
 505                 if (r < 0)
 506                         return r;
 507
 508                 if (ret)
 509                         (*ret)->discoverable = true;
 510
 511                 return 1;
 512         }
 513
 514         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 515                 r = image_make(class, ".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret);
 516                 if (r < 0)
 517                         return r;
 518
 519                 if (ret)
 520                         (*ret)->discoverable = true;
 521
 522                 return r;
 523         }
 524
 525         return -ENOENT;
 526 };
 527
 528 int image_from_path(const char *path, Image **ret) {
 529
 530         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 531          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 532          * overridden by another, different image earlier in the search path */
 533
 534         if (path_equal(path, "/"))
 535                 return image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, "/", NULL, ret);
 536
 537         return image_make(_IMAGE_CLASS_INVALID, NULL, AT_FDCWD, NULL, path, NULL, ret);
 538 }
 539
 540 int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) {
 541         if (image_name_is_valid(name_or_path))
 542                 return image_find(class, name_or_path, root, ret);
 543
 544         return image_from_path(name_or_path, ret);
 545 }
 546
 547 int image_discover(
 548                 ImageClass class,
 549                 const char *root,
 550                 Hashmap *h) {
 551
 552         int r;
 553
 554         assert(class >= 0);
 555         assert(class < _IMAGE_CLASS_MAX);
 556         assert(h);
 557
 558         NULSTR_FOREACH(path, image_search_path[class]) {
 559                 _cleanup_free_ char *resolved = NULL;
 560                 _cleanup_closedir_ DIR *d = NULL;
 561
 562                 r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 563                 if (r == -ENOENT)
 564                         continue;
 565                 if (r < 0)
 566                         return r;
 567
 568                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 569                         _cleanup_(image_unrefp) Image *image = NULL;
 570                         _cleanup_free_ char *truncated = NULL;
 571                         const char *pretty;
 572                         struct stat st;
 573                         int flags;
 574
 575                         if (dot_or_dot_dot(de->d_name))
 576                                 continue;
 577
 578                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to
 579                          * permit people to symlink block devices into the search path. */
 580                         flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 581                         if (fstatat(dirfd(d), de->d_name, &st, flags) < 0) {
 582                                 if (errno == ENOENT)
 583                                         continue;
 584
 585                                 return -errno;
 586                         }
 587
 588                         if (S_ISREG(st.st_mode)) {
 589                                 const char *e;
 590
 591                                 e = endswith(de->d_name, ".raw");
 592                                 if (!e)
 593                                         continue;
 594
 595                                 truncated = strndup(de->d_name, e - de->d_name);
 596                                 if (!truncated)
 597                                         return -ENOMEM;
 598
 599                                 pretty = truncated;
 600                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 601                                 pretty = de->d_name;
 602                         else
 603                                 continue;
 604
 605                         if (!image_name_is_valid(pretty))
 606                                 continue;
 607
 608                         if (hashmap_contains(h, pretty))
 609                                 continue;
 610
 611                         r = image_make(class, pretty, dirfd(d), resolved, de->d_name, &st, &image);
 612                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 613                                 continue;
 614                         if (r < 0)
 615                                 return r;
 616
 617                         image->discoverable = true;
 618
 619                         r = hashmap_put(h, image->name, image);
 620                         if (r < 0)
 621                                 return r;
 622
 623                         image = NULL;
 624                 }
 625         }
 626
 627         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 628                 _cleanup_(image_unrefp) Image *image = NULL;
 629
 630                 r = image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image);
 631                 if (r < 0)
 632                         return r;
 633
 634                 image->discoverable = true;
 635
 636                 r = hashmap_put(h, image->name, image);
 637                 if (r < 0)
 638                         return r;
 639
 640                 image = NULL;
 641         }
 642
 643         return 0;
 644 }
 645
 646 int image_remove(Image *i) {
 647         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 648         _cleanup_strv_free_ char **settings = NULL;
 649         _cleanup_free_ char *roothash = NULL;
 650         int r;
 651
 652         assert(i);
 653
 654         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 655                 return -EROFS;
 656
 657         settings = image_settings_path(i);
 658         if (!settings)
 659                 return -ENOMEM;
 660
 661         r = image_roothash_path(i, &roothash);
 662         if (r < 0)
 663                 return r;
 664
 665         /* Make sure we don't interfere with a running nspawn */
 666         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 667         if (r < 0)
 668                 return r;
 669
 670         switch (i->type) {
 671
 672         case IMAGE_SUBVOLUME:
 673
 674                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 675                  * big guns */
 676                 if (unlink(i->path) < 0) {
 677                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 678                         if (r < 0)
 679                                 return r;
 680                 }
 681
 682                 break;
 683
 684         case IMAGE_DIRECTORY:
 685                 /* Allow deletion of read-only directories */
 686                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 687                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 688                 if (r < 0)
 689                         return r;
 690
 691                 break;
 692
 693         case IMAGE_BLOCK:
 694
 695                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 696                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 697                  * the thing (it's most likely a symlink after all). */
 698
 699                 if (path_startswith(i->path, "/dev"))
 700                         break;
 701
 702                 _fallthrough_;
 703         case IMAGE_RAW:
 704                 if (unlink(i->path) < 0)
 705                         return -errno;
 706                 break;
 707
 708         default:
 709                 return -EOPNOTSUPP;
 710         }
 711
 712         STRV_FOREACH(j, settings)
 713                 if (unlink(*j) < 0 && errno != ENOENT)
 714                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 715
 716         if (unlink(roothash) < 0 && errno != ENOENT)
 717                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 718
 719         return 0;
 720 }
 721
 722 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 723         _cleanup_free_ char *fn = NULL, *rs = NULL;
 724         int r;
 725
 726         fn = strjoin(new_name, suffix);
 727         if (!fn)
 728                 return -ENOMEM;
 729
 730         r = file_in_same_dir(path, fn, &rs);
 731         if (r < 0)
 732                 return r;
 733
 734         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 735 }
 736
 737 int image_rename(Image *i, const char *new_name) {
 738         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 739         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 740         _cleanup_strv_free_ char **settings = NULL;
 741         unsigned file_attr = 0;
 742         int r;
 743
 744         assert(i);
 745
 746         if (!image_name_is_valid(new_name))
 747                 return -EINVAL;
 748
 749         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 750                 return -EROFS;
 751
 752         settings = image_settings_path(i);
 753         if (!settings)
 754                 return -ENOMEM;
 755
 756         r = image_roothash_path(i, &roothash);
 757         if (r < 0)
 758                 return r;
 759
 760         /* Make sure we don't interfere with a running nspawn */
 761         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 762         if (r < 0)
 763                 return r;
 764
 765         /* Make sure nobody takes the new name, between the time we
 766          * checked it is currently unused in all search paths, and the
 767          * time we take possession of it */
 768         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 769         if (r < 0)
 770                 return r;
 771
 772         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 773         if (r >= 0)
 774                 return -EEXIST;
 775         if (r != -ENOENT)
 776                 return r;
 777
 778         switch (i->type) {
 779
 780         case IMAGE_DIRECTORY:
 781                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 782                 (void) read_attr_path(i->path, &file_attr);
 783
 784                 if (file_attr & FS_IMMUTABLE_FL)
 785                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 786
 787                 _fallthrough_;
 788         case IMAGE_SUBVOLUME:
 789                 r = file_in_same_dir(i->path, new_name, &new_path);
 790                 break;
 791
 792         case IMAGE_BLOCK:
 793
 794                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 795                 if (path_startswith(i->path, "/dev"))
 796                         return -EROFS;
 797
 798                 r = file_in_same_dir(i->path, new_name, &new_path);
 799                 break;
 800
 801         case IMAGE_RAW: {
 802                 const char *fn;
 803
 804                 fn = strjoina(new_name, ".raw");
 805
 806                 r = file_in_same_dir(i->path, fn, &new_path);
 807                 break;
 808         }
 809
 810         default:
 811                 return -EOPNOTSUPP;
 812         }
 813         if (r < 0)
 814                 return r;
 815
 816         nn = strdup(new_name);
 817         if (!nn)
 818                 return -ENOMEM;
 819
 820         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 821         if (r < 0)
 822                 return r;
 823
 824         /* Restore the immutable bit, if it was set before */
 825         if (file_attr & FS_IMMUTABLE_FL)
 826                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 827
 828         free_and_replace(i->path, new_path);
 829         free_and_replace(i->name, nn);
 830
 831         STRV_FOREACH(j, settings) {
 832                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 833                 if (r < 0 && r != -ENOENT)
 834                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 835         }
 836
 837         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 838         if (r < 0 && r != -ENOENT)
 839                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 840
 841         return 0;
 842 }
 843
 844 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 845         _cleanup_free_ char *fn = NULL, *rs = NULL;
 846         int r;
 847
 848         fn = strjoin(new_name, suffix);
 849         if (!fn)
 850                 return -ENOMEM;
 851
 852         r = file_in_same_dir(path, fn, &rs);
 853         if (r < 0)
 854                 return r;
 855
 856         return copy_file_atomic(path, rs, 0664, COPY_REFLINK);
 857 }
 858
 859 int image_clone(Image *i, const char *new_name, bool read_only) {
 860         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 861         _cleanup_strv_free_ char **settings = NULL;
 862         _cleanup_free_ char *roothash = NULL;
 863         const char *new_path;
 864         int r;
 865
 866         assert(i);
 867
 868         if (!image_name_is_valid(new_name))
 869                 return -EINVAL;
 870
 871         settings = image_settings_path(i);
 872         if (!settings)
 873                 return -ENOMEM;
 874
 875         r = image_roothash_path(i, &roothash);
 876         if (r < 0)
 877                 return r;
 878
 879         /* Make sure nobody takes the new name, between the time we
 880          * checked it is currently unused in all search paths, and the
 881          * time we take possession of it */
 882         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 883         if (r < 0)
 884                 return r;
 885
 886         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 887         if (r >= 0)
 888                 return -EEXIST;
 889         if (r != -ENOENT)
 890                 return r;
 891
 892         switch (i->type) {
 893
 894         case IMAGE_SUBVOLUME:
 895         case IMAGE_DIRECTORY:
 896                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 897                  * directory. */
 898
 899                 new_path = strjoina("/var/lib/machines/", new_name);
 900
 901                 r = btrfs_subvol_snapshot(i->path, new_path,
 902                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 903                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 904                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 905                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 906                                           BTRFS_SNAPSHOT_RECURSIVE |
 907                                           BTRFS_SNAPSHOT_QUOTA);
 908                 if (r >= 0)
 909                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 910                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 911
 912                 break;
 913
 914         case IMAGE_RAW:
 915                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 916
 917                 r = copy_file_atomic_full(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL,
 918                                           COPY_REFLINK|COPY_CRTIME, NULL, NULL);
 919                 break;
 920
 921         case IMAGE_BLOCK:
 922         default:
 923                 return -EOPNOTSUPP;
 924         }
 925
 926         if (r < 0)
 927                 return r;
 928
 929         STRV_FOREACH(j, settings) {
 930                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 931                 if (r < 0 && r != -ENOENT)
 932                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 933         }
 934
 935         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 936         if (r < 0 && r != -ENOENT)
 937                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 938
 939         return 0;
 940 }
 941
 942 int image_read_only(Image *i, bool b) {
 943         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 944         int r;
 945
 946         assert(i);
 947
 948         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 949                 return -EROFS;
 950
 951         /* Make sure we don't interfere with a running nspawn */
 952         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 953         if (r < 0)
 954                 return r;
 955
 956         switch (i->type) {
 957
 958         case IMAGE_SUBVOLUME:
 959
 960                 /* Note that we set the flag only on the top-level
 961                  * subvolume of the image. */
 962
 963                 r = btrfs_subvol_set_read_only(i->path, b);
 964                 if (r < 0)
 965                         return r;
 966
 967                 break;
 968
 969         case IMAGE_DIRECTORY:
 970                 /* For simple directory trees we cannot use the access
 971                    mode of the top-level directory, since it has an
 972                    effect on the container itself.  However, we can
 973                    use the "immutable" flag, to at least make the
 974                    top-level directory read-only. It's not as good as
 975                    a read-only subvolume, but at least something, and
 976                    we can read the value back. */
 977
 978                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 979                 if (r < 0)
 980                         return r;
 981
 982                 break;
 983
 984         case IMAGE_RAW: {
 985                 struct stat st;
 986
 987                 if (stat(i->path, &st) < 0)
 988                         return -errno;
 989
 990                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 991                         return -errno;
 992
 993                 /* If the images is now read-only, it's a good time to
 994                  * defrag it, given that no write patterns will
 995                  * fragment it again. */
 996                 if (b)
 997                         (void) btrfs_defrag(i->path);
 998                 break;
 999         }
1000
1001         case IMAGE_BLOCK: {
1002                 _cleanup_close_ int fd = -EBADF;
1003                 struct stat st;
1004                 int state = b;
1005
1006                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
1007                 if (fd < 0)
1008                         return -errno;
1009
1010                 if (fstat(fd, &st) < 0)
1011                         return -errno;
1012                 if (!S_ISBLK(st.st_mode))
1013                         return -ENOTTY;
1014
1015                 if (ioctl(fd, BLKROSET, &state) < 0)
1016                         return -errno;
1017
1018                 break;
1019         }
1020
1021         default:
1022                 return -EOPNOTSUPP;
1023         }
1024
1025         return 0;
1026 }
1027
1028 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
1029         _cleanup_free_ char *p = NULL;
1030         LockFile t = LOCK_FILE_INIT;
1031         struct stat st;
1032         bool exclusive;
1033         int r;
1034
1035         assert(path);
1036         assert(global);
1037         assert(local);
1038
1039         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1040          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1041          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1042          * correctly. */
1043
1044         if (!path_is_absolute(path))
1045                 return -EINVAL;
1046
1047         switch (operation & (LOCK_SH|LOCK_EX)) {
1048         case LOCK_SH:
1049                 exclusive = false;
1050                 break;
1051         case LOCK_EX:
1052                 exclusive = true;
1053                 break;
1054         default:
1055                 return -EINVAL;
1056         }
1057
1058         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1059                 *local = *global = (LockFile) LOCK_FILE_INIT;
1060                 return 0;
1061         }
1062
1063         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1064          * running off it after all, and we don't want any images to manipulate the host image. We make an
1065          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1066          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1067          * since it means changes made to the host might propagate to the container as they happen (and a
1068          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1069          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1070          * the user to do the right thing with this. */
1071         if (path_equal(path, "/")) {
1072                 if (exclusive)
1073                         return -EBUSY;
1074
1075                 *local = *global = (LockFile) LOCK_FILE_INIT;
1076                 return 0;
1077         }
1078
1079         if (stat(path, &st) >= 0) {
1080                 if (S_ISBLK(st.st_mode))
1081                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1082                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1083                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1084                 else
1085                         return -ENOTTY;
1086                 if (r < 0)
1087                         return -ENOMEM;
1088         }
1089
1090         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1091          * sufficient, since block devices are host local anyway. */
1092         if (!path_startswith(path, "/dev/")) {
1093                 r = make_lock_file_for(path, operation, &t);
1094                 if (r < 0) {
1095                         if (!exclusive && r == -EROFS)
1096                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1097                         else
1098                                 return r;
1099                 }
1100         }
1101
1102         if (p) {
1103                 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1104
1105                 r = make_lock_file(p, operation, global);
1106                 if (r < 0) {
1107                         release_lock_file(&t);
1108                         return r;
1109                 }
1110         } else
1111                 *global = (LockFile) LOCK_FILE_INIT;
1112
1113         *local = t;
1114         return 0;
1115 }
1116
1117 int image_set_limit(Image *i, uint64_t referenced_max) {
1118         assert(i);
1119
1120         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1121                 return -EROFS;
1122
1123         if (i->type != IMAGE_SUBVOLUME)
1124                 return -EOPNOTSUPP;
1125
1126         /* We set the quota both for the subvolume as well as for the
1127          * subtree. The latter is mostly for historical reasons, since
1128          * we didn't use to have a concept of subtree quota, and hence
1129          * only modified the subvolume quota. */
1130
1131         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1132         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1133         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1134 }
1135
1136 int image_read_metadata(Image *i, const ImagePolicy *image_policy) {
1137         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1138         int r;
1139
1140         assert(i);
1141
1142         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1143         if (r < 0)
1144                 return r;
1145
1146         switch (i->type) {
1147
1148         case IMAGE_SUBVOLUME:
1149         case IMAGE_DIRECTORY: {
1150                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
1151                 sd_id128_t machine_id = SD_ID128_NULL;
1152                 _cleanup_free_ char *hostname = NULL;
1153                 _cleanup_free_ char *path = NULL;
1154
1155                 if (i->class == IMAGE_EXTENSION) {
1156                         r = extension_has_forbidden_content(i->path);
1157                         if (r < 0)
1158                                 return r;
1159                         if (r > 0)
1160                                 return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM),
1161                                                        "Conflicting content found in image %s, refusing.",
1162                                                        i->name);
1163                 }
1164
1165                 r = chase("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1166                 if (r < 0 && r != -ENOENT)
1167                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1168                 else if (r >= 0) {
1169                         r = read_etc_hostname(path, &hostname);
1170                         if (r < 0)
1171                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1172                 }
1173
1174                 path = mfree(path);
1175
1176                 r = id128_get_machine(i->path, &machine_id);
1177                 if (r < 0)
1178                         log_debug_errno(r, "Failed to read machine ID in image %s, ignoring: %m", i->name);
1179
1180                 r = chase("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1181                 if (r < 0 && r != -ENOENT)
1182                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1183                 else if (r >= 0) {
1184                         r = load_env_file_pairs(NULL, path, &machine_info);
1185                         if (r < 0)
1186                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1187                 }
1188
1189                 r = load_os_release_pairs(i->path, &os_release);
1190                 if (r < 0)
1191                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1192
1193                 r = load_extension_release_pairs(i->path, i->name, /* relax_extension_release_check= */ false, &extension_release);
1194                 if (r < 0)
1195                         log_debug_errno(r, "Failed to read extension-release in image, ignoring: %m");
1196
1197                 free_and_replace(i->hostname, hostname);
1198                 i->machine_id = machine_id;
1199                 strv_free_and_replace(i->machine_info, machine_info);
1200                 strv_free_and_replace(i->os_release, os_release);
1201                 strv_free_and_replace(i->extension_release, extension_release);
1202
1203                 break;
1204         }
1205
1206         case IMAGE_RAW:
1207         case IMAGE_BLOCK: {
1208                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1209                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1210
1211                 r = loop_device_make_by_path(i->path, O_RDONLY, /* sector_size= */ UINT32_MAX, LO_FLAGS_PARTSCAN, LOCK_SH, &d);
1212                 if (r < 0)
1213                         return r;
1214
1215                 r = dissect_loop_device(
1216                                 d,
1217                                 /* verity= */ NULL,
1218                                 /* mount_options= */ NULL,
1219                                 image_policy,
1220                                 DISSECT_IMAGE_GENERIC_ROOT |
1221                                 DISSECT_IMAGE_REQUIRE_ROOT |
1222                                 DISSECT_IMAGE_RELAX_VAR_CHECK |
1223                                 DISSECT_IMAGE_READ_ONLY |
1224                                 DISSECT_IMAGE_USR_NO_ROOT |
1225                                 DISSECT_IMAGE_ADD_PARTITION_DEVICES |
1226                                 DISSECT_IMAGE_PIN_PARTITION_DEVICES,
1227                                 &m);
1228                 if (r < 0)
1229                         return r;
1230
1231                 r = dissected_image_acquire_metadata(m,
1232                                                      DISSECT_IMAGE_VALIDATE_OS |
1233                                                      DISSECT_IMAGE_VALIDATE_OS_EXT);
1234                 if (r < 0)
1235                         return r;
1236
1237                 free_and_replace(i->hostname, m->hostname);
1238                 i->machine_id = m->machine_id;
1239                 strv_free_and_replace(i->machine_info, m->machine_info);
1240                 strv_free_and_replace(i->os_release, m->os_release);
1241                 strv_free_and_replace(i->extension_release, m->extension_release);
1242
1243                 break;
1244         }
1245
1246         default:
1247                 return -EOPNOTSUPP;
1248         }
1249
1250         i->metadata_valid = true;
1251
1252         return 0;
1253 }
1254
1255 int image_name_lock(const char *name, int operation, LockFile *ret) {
1256         const char *p;
1257
1258         assert(name);
1259         assert(ret);
1260
1261         /* Locks an image name, regardless of the precise path used. */
1262
1263         if (streq(name, ".host"))
1264                 return -EBUSY;
1265
1266         if (!image_name_is_valid(name))
1267                 return -EINVAL;
1268
1269         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1270                 *ret = (LockFile) LOCK_FILE_INIT;
1271                 return 0;
1272         }
1273
1274         (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1275
1276         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1277         return make_lock_file(p, operation, ret);
1278 }
1279
1280 bool image_in_search_path(
1281                 ImageClass class,
1282                 const char *root,
1283                 const char *image) {
1284
1285         assert(image);
1286
1287         NULSTR_FOREACH(path, image_search_path[class]) {
1288                 const char *p, *q;
1289                 size_t k;
1290
1291                 if (!empty_or_root(root)) {
1292                         q = path_startswith(path, root);
1293                         if (!q)
1294                                 continue;
1295                 } else
1296                         q = path;
1297
1298                 p = path_startswith(q, path);
1299                 if (!p)
1300                         continue;
1301
1302                 /* Make sure there's a filename following */
1303                 k = strcspn(p, "/");
1304                 if (k == 0)
1305                         continue;
1306
1307                 p += k;
1308
1309                 /* Accept trailing slashes */
1310                 if (p[strspn(p, "/")] == 0)
1311                         return true;
1312
1313         }
1314
1315         return false;
1316 }
1317
1318 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1319         [IMAGE_DIRECTORY] = "directory",
1320         [IMAGE_SUBVOLUME] = "subvolume",
1321         [IMAGE_RAW] = "raw",
1322         [IMAGE_BLOCK] = "block",
1323 };
1324
1325 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);
1326
1327 static const char* const image_class_table[_IMAGE_CLASS_MAX] = {
1328         [IMAGE_MACHINE] = "machine",
1329         [IMAGE_PORTABLE] = "portable",
1330         [IMAGE_EXTENSION] = "extension",
1331 };
1332
1333 DEFINE_STRING_TABLE_LOOKUP(image_class, ImageClass);