src/shared/machine-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 <stdio.h>
   8 #include <stdlib.h>
   9 #include <sys/file.h>
  10 #include <sys/ioctl.h>
  11 #include <sys/stat.h>
  12 #include <unistd.h>
  13
  14 #include "alloc-util.h"
  15 #include "btrfs-util.h"
  16 #include "chattr-util.h"
  17 #include "copy.h"
  18 #include "dirent-util.h"
  19 #include "dissect-image.h"
  20 #include "env-file.h"
  21 #include "env-util.h"
  22 #include "fd-util.h"
  23 #include "fs-util.h"
  24 #include "hashmap.h"
  25 #include "hostname-setup.h"
  26 #include "id128-util.h"
  27 #include "lockfile-util.h"
  28 #include "log.h"
  29 #include "loop-util.h"
  30 #include "machine-image.h"
  31 #include "macro.h"
  32 #include "mkdir.h"
  33 #include "nulstr-util.h"
  34 #include "os-util.h"
  35 #include "path-util.h"
  36 #include "rm-rf.h"
  37 #include "string-table.h"
  38 #include "string-util.h"
  39 #include "strv.h"
  40 #include "time-util.h"
  41 #include "utf8.h"
  42 #include "xattr-util.h"
  43
  44 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  45         [IMAGE_MACHINE] =   "/etc/machines\0"              /* only place symlinks here */
  46                             "/run/machines\0"              /* and here too */
  47                             "/var/lib/machines\0"          /* the main place for images */
  48                             "/var/lib/container\0"         /* legacy */
  49                             "/usr/local/lib/machines\0"
  50                             "/usr/lib/machines\0",
  51
  52         [IMAGE_PORTABLE] =  "/etc/portables\0"             /* only place symlinks here */
  53                             "/run/portables\0"             /* and here too */
  54                             "/var/lib/portables\0"         /* the main place for images */
  55                             "/usr/local/lib/portables\0"
  56                             "/usr/lib/portables\0",
  57
  58         [IMAGE_EXTENSION] = "/etc/extensions\0"             /* only place symlinks here */
  59                             "/run/extensions\0"             /* and here too */
  60                             "/var/lib/extensions\0"         /* the main place for images */
  61                             "/usr/local/lib/extensions\0"
  62                             "/usr/lib/extensions\0",
  63 };
  64
  65 static Image *image_free(Image *i) {
  66         assert(i);
  67
  68         free(i->name);
  69         free(i->path);
  70
  71         free(i->hostname);
  72         strv_free(i->machine_info);
  73         strv_free(i->os_release);
  74
  75         return mfree(i);
  76 }
  77
  78 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  79 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
  80                                       Image, image_unref);
  81
  82 static char **image_settings_path(Image *image) {
  83         _cleanup_strv_free_ char **l = NULL;
  84         const char *fn, *s;
  85         unsigned i = 0;
  86
  87         assert(image);
  88
  89         l = new0(char*, 4);
  90         if (!l)
  91                 return NULL;
  92
  93         fn = strjoina(image->name, ".nspawn");
  94
  95         FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
  96                 l[i] = path_join(s, fn);
  97                 if (!l[i])
  98                         return NULL;
  99
 100                 i++;
 101         }
 102
 103         l[i] = file_in_same_dir(image->path, fn);
 104         if (!l[i])
 105                 return NULL;
 106
 107         return TAKE_PTR(l);
 108 }
 109
 110 static char *image_roothash_path(Image *image) {
 111         const char *fn;
 112
 113         assert(image);
 114
 115         fn = strjoina(image->name, ".roothash");
 116
 117         return file_in_same_dir(image->path, fn);
 118 }
 119
 120 static int image_new(
 121                 ImageType t,
 122                 const char *pretty,
 123                 const char *path,
 124                 const char *filename,
 125                 bool read_only,
 126                 usec_t crtime,
 127                 usec_t mtime,
 128                 Image **ret) {
 129
 130         _cleanup_(image_unrefp) Image *i = NULL;
 131
 132         assert(t >= 0);
 133         assert(t < _IMAGE_TYPE_MAX);
 134         assert(pretty);
 135         assert(filename);
 136         assert(ret);
 137
 138         i = new(Image, 1);
 139         if (!i)
 140                 return -ENOMEM;
 141
 142         *i = (Image) {
 143                 .n_ref = 1,
 144                 .type = t,
 145                 .read_only = read_only,
 146                 .crtime = crtime,
 147                 .mtime = mtime,
 148                 .usage = UINT64_MAX,
 149                 .usage_exclusive = UINT64_MAX,
 150                 .limit = UINT64_MAX,
 151                 .limit_exclusive = UINT64_MAX,
 152         };
 153
 154         i->name = strdup(pretty);
 155         if (!i->name)
 156                 return -ENOMEM;
 157
 158         i->path = path_join(path, filename);
 159         if (!i->path)
 160                 return -ENOMEM;
 161
 162         path_simplify(i->path, false);
 163
 164         *ret = TAKE_PTR(i);
 165
 166         return 0;
 167 }
 168
 169 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 170         _cleanup_free_ char *name = NULL;
 171         const char *p;
 172         size_t n;
 173
 174         assert(path);
 175         assert(ret);
 176
 177         p = last_path_component(path);
 178         n = strcspn(p, "/");
 179
 180         name = strndup(p, n);
 181         if (!name)
 182                 return -ENOMEM;
 183
 184         if (suffix) {
 185                 char *e;
 186
 187                 e = endswith(name, suffix);
 188                 if (!e)
 189                         return -EINVAL;
 190
 191                 *e = 0;
 192         }
 193
 194         if (!image_name_is_valid(name))
 195                 return -EINVAL;
 196
 197         *ret = TAKE_PTR(name);
 198         return 0;
 199 }
 200
 201 static int image_make(
 202                 const char *pretty,
 203                 int dfd,
 204                 const char *path,
 205                 const char *filename,
 206                 const struct stat *st,
 207                 Image **ret) {
 208
 209         _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL;
 210         struct stat stbuf;
 211         bool read_only;
 212         int r;
 213
 214         assert(dfd >= 0 || dfd == AT_FDCWD);
 215         assert(path || dfd == AT_FDCWD);
 216         assert(filename);
 217
 218         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 219          * devices into /var/lib/machines/, and treat them normally.
 220          *
 221          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 222          * recognize. */
 223
 224         if (!st) {
 225                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 226                         return -errno;
 227
 228                 st = &stbuf;
 229         }
 230
 231         if (!path) {
 232                 if (dfd == AT_FDCWD)
 233                         (void) safe_getcwd(&parent);
 234                 else
 235                         (void) fd_get_path(dfd, &parent);
 236         }
 237
 238         read_only =
 239                 (path && path_startswith(path, "/usr")) ||
 240                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 241
 242         if (S_ISDIR(st->st_mode)) {
 243                 _cleanup_close_ int fd = -1;
 244                 unsigned file_attr = 0;
 245                 usec_t crtime = 0;
 246
 247                 if (!ret)
 248                         return 0;
 249
 250                 if (!pretty) {
 251                         r = extract_pretty(filename, NULL, &pretty_buffer);
 252                         if (r < 0)
 253                                 return r;
 254
 255                         pretty = pretty_buffer;
 256                 }
 257
 258                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 259                 if (fd < 0)
 260                         return -errno;
 261
 262                 /* btrfs subvolumes have inode 256 */
 263                 if (st->st_ino == 256) {
 264
 265                         r = btrfs_is_filesystem(fd);
 266                         if (r < 0)
 267                                 return r;
 268                         if (r) {
 269                                 BtrfsSubvolInfo info;
 270
 271                                 /* It's a btrfs subvolume */
 272
 273                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 274                                 if (r < 0)
 275                                         return r;
 276
 277                                 r = image_new(IMAGE_SUBVOLUME,
 278                                               pretty,
 279                                               path,
 280                                               filename,
 281                                               info.read_only || read_only,
 282                                               info.otime,
 283                                               0,
 284                                               ret);
 285                                 if (r < 0)
 286                                         return r;
 287
 288                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 289                                         BtrfsQuotaInfo quota;
 290
 291                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 292                                         if (r >= 0) {
 293                                                 (*ret)->usage = quota.referenced;
 294                                                 (*ret)->usage_exclusive = quota.exclusive;
 295
 296                                                 (*ret)->limit = quota.referenced_max;
 297                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 298                                         }
 299                                 }
 300
 301                                 return 0;
 302                         }
 303                 }
 304
 305                 /* Get directory creation time (not available everywhere, but that's OK */
 306                 (void) fd_getcrtime(dfd, &crtime);
 307
 308                 /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not
 309                  * supported everywhere we ignore failures. */
 310                 (void) read_attr_fd(fd, &file_attr);
 311
 312                 /* It's just a normal directory. */
 313                 r = image_new(IMAGE_DIRECTORY,
 314                               pretty,
 315                               path,
 316                               filename,
 317                               read_only || (file_attr & FS_IMMUTABLE_FL),
 318                               crtime,
 319                               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 */
 320                               ret);
 321                 if (r < 0)
 322                         return r;
 323
 324                 return 0;
 325
 326         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 327                 usec_t crtime = 0;
 328
 329                 /* It's a RAW disk image */
 330
 331                 if (!ret)
 332                         return 0;
 333
 334                 (void) fd_getcrtime_at(dfd, filename, &crtime, 0);
 335
 336                 if (!pretty) {
 337                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 338                         if (r < 0)
 339                                 return r;
 340
 341                         pretty = pretty_buffer;
 342                 }
 343
 344                 r = image_new(IMAGE_RAW,
 345                               pretty,
 346                               path,
 347                               filename,
 348                               !(st->st_mode & 0222) || read_only,
 349                               crtime,
 350                               timespec_load(&st->st_mtim),
 351                               ret);
 352                 if (r < 0)
 353                         return r;
 354
 355                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 356                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 357
 358                 return 0;
 359
 360         } else if (S_ISBLK(st->st_mode)) {
 361                 _cleanup_close_ int block_fd = -1;
 362                 uint64_t size = UINT64_MAX;
 363
 364                 /* A block device */
 365
 366                 if (!ret)
 367                         return 0;
 368
 369                 if (!pretty) {
 370                         r = extract_pretty(filename, NULL, &pretty_buffer);
 371                         if (r < 0)
 372                                 return r;
 373
 374                         pretty = pretty_buffer;
 375                 }
 376
 377                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 378                 if (block_fd < 0)
 379                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 380                 else {
 381                         /* Refresh stat data after opening the node */
 382                         if (fstat(block_fd, &stbuf) < 0)
 383                                 return -errno;
 384                         st = &stbuf;
 385
 386                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 387                                 return -ENOTTY;
 388
 389                         if (!read_only) {
 390                                 int state = 0;
 391
 392                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 393                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 394                                 else if (state)
 395                                         read_only = true;
 396                         }
 397
 398                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 399                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 400
 401                         block_fd = safe_close(block_fd);
 402                 }
 403
 404                 r = image_new(IMAGE_BLOCK,
 405                               pretty,
 406                               path,
 407                               filename,
 408                               !(st->st_mode & 0222) || read_only,
 409                               0,
 410                               0,
 411                               ret);
 412                 if (r < 0)
 413                         return r;
 414
 415                 if (!IN_SET(size, 0, UINT64_MAX))
 416                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 417
 418                 return 0;
 419         }
 420
 421         return -EMEDIUMTYPE;
 422 }
 423
 424 int image_find(ImageClass class,
 425                const char *name,
 426                const char *root,
 427                Image **ret) {
 428
 429         const char *path;
 430         int r;
 431
 432         assert(class >= 0);
 433         assert(class < _IMAGE_CLASS_MAX);
 434         assert(name);
 435
 436         /* There are no images with invalid names */
 437         if (!image_name_is_valid(name))
 438                 return -ENOENT;
 439
 440         NULSTR_FOREACH(path, image_search_path[class]) {
 441                 _cleanup_free_ char *resolved = NULL;
 442                 _cleanup_closedir_ DIR *d = NULL;
 443                 struct stat st;
 444                 int flags;
 445
 446                 r = chase_symlinks_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 447                 if (r == -ENOENT)
 448                         continue;
 449                 if (r < 0)
 450                         return r;
 451
 452                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 453                  * to symlink block devices into the search path. (For now, we disable that when operating
 454                  * relative to some root directory.) */
 455                 flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 456                 if (fstatat(dirfd(d), name, &st, flags) < 0) {
 457                         _cleanup_free_ char *raw = NULL;
 458
 459                         if (errno != ENOENT)
 460                                 return -errno;
 461
 462                         raw = strjoin(name, ".raw");
 463                         if (!raw)
 464                                 return -ENOMEM;
 465
 466                         if (fstatat(dirfd(d), raw, &st, flags) < 0) {
 467                                 if (errno == ENOENT)
 468                                         continue;
 469
 470                                 return -errno;
 471                         }
 472
 473                         if (!S_ISREG(st.st_mode))
 474                                 continue;
 475
 476                         r = image_make(name, dirfd(d), resolved, raw, &st, ret);
 477
 478                 } else {
 479                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 480                                 continue;
 481
 482                         r = image_make(name, dirfd(d), resolved, name, &st, ret);
 483                 }
 484                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 485                         continue;
 486                 if (r < 0)
 487                         return r;
 488
 489                 if (ret)
 490                         (*ret)->discoverable = true;
 491
 492                 return 1;
 493         }
 494
 495         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 496                 r = image_make(".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret);
 497                 if (r < 0)
 498                         return r;
 499
 500                 if (ret)
 501                         (*ret)->discoverable = true;
 502
 503                 return r;
 504         }
 505
 506         return -ENOENT;
 507 };
 508
 509 int image_from_path(const char *path, Image **ret) {
 510
 511         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 512          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 513          * overridden by another, different image earlier in the search path */
 514
 515         if (path_equal(path, "/"))
 516                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 517
 518         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 519 }
 520
 521 int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) {
 522         if (image_name_is_valid(name_or_path))
 523                 return image_find(class, name_or_path, root, ret);
 524
 525         return image_from_path(name_or_path, ret);
 526 }
 527
 528 int image_discover(
 529                 ImageClass class,
 530                 const char *root,
 531                 Hashmap *h) {
 532
 533         const char *path;
 534         int r;
 535
 536         assert(class >= 0);
 537         assert(class < _IMAGE_CLASS_MAX);
 538         assert(h);
 539
 540         NULSTR_FOREACH(path, image_search_path[class]) {
 541                 _cleanup_free_ char *resolved = NULL;
 542                 _cleanup_closedir_ DIR *d = NULL;
 543                 struct dirent *de;
 544
 545                 r = chase_symlinks_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d);
 546                 if (r == -ENOENT)
 547                         continue;
 548                 if (r < 0)
 549                         return r;
 550
 551                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 552                         _cleanup_(image_unrefp) Image *image = NULL;
 553                         _cleanup_free_ char *truncated = NULL;
 554                         const char *pretty;
 555                         struct stat st;
 556                         int flags;
 557
 558                         if (dot_or_dot_dot(de->d_name))
 559                                 continue;
 560
 561                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to
 562                          * permit people to symlink block devices into the search path. */
 563                         flags = root ? AT_SYMLINK_NOFOLLOW : 0;
 564                         if (fstatat(dirfd(d), de->d_name, &st, flags) < 0) {
 565                                 if (errno == ENOENT)
 566                                         continue;
 567
 568                                 return -errno;
 569                         }
 570
 571                         if (S_ISREG(st.st_mode)) {
 572                                 const char *e;
 573
 574                                 e = endswith(de->d_name, ".raw");
 575                                 if (!e)
 576                                         continue;
 577
 578                                 truncated = strndup(de->d_name, e - de->d_name);
 579                                 if (!truncated)
 580                                         return -ENOMEM;
 581
 582                                 pretty = truncated;
 583                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 584                                 pretty = de->d_name;
 585                         else
 586                                 continue;
 587
 588                         if (!image_name_is_valid(pretty))
 589                                 continue;
 590
 591                         if (hashmap_contains(h, pretty))
 592                                 continue;
 593
 594                         r = image_make(pretty, dirfd(d), resolved, de->d_name, &st, &image);
 595                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 596                                 continue;
 597                         if (r < 0)
 598                                 return r;
 599
 600                         image->discoverable = true;
 601
 602                         r = hashmap_put(h, image->name, image);
 603                         if (r < 0)
 604                                 return r;
 605
 606                         image = NULL;
 607                 }
 608         }
 609
 610         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 611                 _cleanup_(image_unrefp) Image *image = NULL;
 612
 613                 r = image_make(".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image);
 614                 if (r < 0)
 615                         return r;
 616
 617                 image->discoverable = true;
 618
 619                 r = hashmap_put(h, image->name, image);
 620                 if (r < 0)
 621                         return r;
 622
 623                 image = NULL;
 624         }
 625
 626         return 0;
 627 }
 628
 629 int image_remove(Image *i) {
 630         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 631         _cleanup_strv_free_ char **settings = NULL;
 632         _cleanup_free_ char *roothash = NULL;
 633         char **j;
 634         int r;
 635
 636         assert(i);
 637
 638         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 639                 return -EROFS;
 640
 641         settings = image_settings_path(i);
 642         if (!settings)
 643                 return -ENOMEM;
 644
 645         roothash = image_roothash_path(i);
 646         if (!roothash)
 647                 return -ENOMEM;
 648
 649         /* Make sure we don't interfere with a running nspawn */
 650         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 651         if (r < 0)
 652                 return r;
 653
 654         switch (i->type) {
 655
 656         case IMAGE_SUBVOLUME:
 657
 658                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 659                  * big guns */
 660                 if (unlink(i->path) < 0) {
 661                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 662                         if (r < 0)
 663                                 return r;
 664                 }
 665
 666                 break;
 667
 668         case IMAGE_DIRECTORY:
 669                 /* Allow deletion of read-only directories */
 670                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 671                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 672                 if (r < 0)
 673                         return r;
 674
 675                 break;
 676
 677         case IMAGE_BLOCK:
 678
 679                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 680                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 681                  * the thing (it's most likely a symlink after all). */
 682
 683                 if (path_startswith(i->path, "/dev"))
 684                         break;
 685
 686                 _fallthrough_;
 687         case IMAGE_RAW:
 688                 if (unlink(i->path) < 0)
 689                         return -errno;
 690                 break;
 691
 692         default:
 693                 return -EOPNOTSUPP;
 694         }
 695
 696         STRV_FOREACH(j, settings) {
 697                 if (unlink(*j) < 0 && errno != ENOENT)
 698                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 699         }
 700
 701         if (unlink(roothash) < 0 && errno != ENOENT)
 702                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 703
 704         return 0;
 705 }
 706
 707 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 708         _cleanup_free_ char *rs = NULL;
 709         const char *fn;
 710
 711         fn = strjoina(new_name, suffix);
 712
 713         rs = file_in_same_dir(path, fn);
 714         if (!rs)
 715                 return -ENOMEM;
 716
 717         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 718 }
 719
 720 int image_rename(Image *i, const char *new_name) {
 721         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 722         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 723         _cleanup_strv_free_ char **settings = NULL;
 724         unsigned file_attr = 0;
 725         char **j;
 726         int r;
 727
 728         assert(i);
 729
 730         if (!image_name_is_valid(new_name))
 731                 return -EINVAL;
 732
 733         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 734                 return -EROFS;
 735
 736         settings = image_settings_path(i);
 737         if (!settings)
 738                 return -ENOMEM;
 739
 740         roothash = image_roothash_path(i);
 741         if (!roothash)
 742                 return -ENOMEM;
 743
 744         /* Make sure we don't interfere with a running nspawn */
 745         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 746         if (r < 0)
 747                 return r;
 748
 749         /* Make sure nobody takes the new name, between the time we
 750          * checked it is currently unused in all search paths, and the
 751          * time we take possession of it */
 752         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 753         if (r < 0)
 754                 return r;
 755
 756         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 757         if (r >= 0)
 758                 return -EEXIST;
 759         if (r != -ENOENT)
 760                 return r;
 761
 762         switch (i->type) {
 763
 764         case IMAGE_DIRECTORY:
 765                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 766                 (void) read_attr_path(i->path, &file_attr);
 767
 768                 if (file_attr & FS_IMMUTABLE_FL)
 769                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 770
 771                 _fallthrough_;
 772         case IMAGE_SUBVOLUME:
 773                 new_path = file_in_same_dir(i->path, new_name);
 774                 break;
 775
 776         case IMAGE_BLOCK:
 777
 778                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 779                 if (path_startswith(i->path, "/dev"))
 780                         return -EROFS;
 781
 782                 new_path = file_in_same_dir(i->path, new_name);
 783                 break;
 784
 785         case IMAGE_RAW: {
 786                 const char *fn;
 787
 788                 fn = strjoina(new_name, ".raw");
 789                 new_path = file_in_same_dir(i->path, fn);
 790                 break;
 791         }
 792
 793         default:
 794                 return -EOPNOTSUPP;
 795         }
 796
 797         if (!new_path)
 798                 return -ENOMEM;
 799
 800         nn = strdup(new_name);
 801         if (!nn)
 802                 return -ENOMEM;
 803
 804         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 805         if (r < 0)
 806                 return r;
 807
 808         /* Restore the immutable bit, if it was set before */
 809         if (file_attr & FS_IMMUTABLE_FL)
 810                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 811
 812         free_and_replace(i->path, new_path);
 813         free_and_replace(i->name, nn);
 814
 815         STRV_FOREACH(j, settings) {
 816                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 817                 if (r < 0 && r != -ENOENT)
 818                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 819         }
 820
 821         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 822         if (r < 0 && r != -ENOENT)
 823                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 824
 825         return 0;
 826 }
 827
 828 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 829         _cleanup_free_ char *rs = NULL;
 830         const char *fn;
 831
 832         fn = strjoina(new_name, suffix);
 833
 834         rs = file_in_same_dir(path, fn);
 835         if (!rs)
 836                 return -ENOMEM;
 837
 838         return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK);
 839 }
 840
 841 int image_clone(Image *i, const char *new_name, bool read_only) {
 842         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 843         _cleanup_strv_free_ char **settings = NULL;
 844         _cleanup_free_ char *roothash = NULL;
 845         const char *new_path;
 846         char **j;
 847         int r;
 848
 849         assert(i);
 850
 851         if (!image_name_is_valid(new_name))
 852                 return -EINVAL;
 853
 854         settings = image_settings_path(i);
 855         if (!settings)
 856                 return -ENOMEM;
 857
 858         roothash = image_roothash_path(i);
 859         if (!roothash)
 860                 return -ENOMEM;
 861
 862         /* Make sure nobody takes the new name, between the time we
 863          * checked it is currently unused in all search paths, and the
 864          * time we take possession of it */
 865         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 866         if (r < 0)
 867                 return r;
 868
 869         r = image_find(IMAGE_MACHINE, new_name, NULL, NULL);
 870         if (r >= 0)
 871                 return -EEXIST;
 872         if (r != -ENOENT)
 873                 return r;
 874
 875         switch (i->type) {
 876
 877         case IMAGE_SUBVOLUME:
 878         case IMAGE_DIRECTORY:
 879                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 880                  * directory. */
 881
 882                 new_path = strjoina("/var/lib/machines/", new_name);
 883
 884                 r = btrfs_subvol_snapshot(i->path, new_path,
 885                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 886                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 887                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 888                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 889                                           BTRFS_SNAPSHOT_RECURSIVE |
 890                                           BTRFS_SNAPSHOT_QUOTA);
 891                 if (r >= 0)
 892                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 893                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 894
 895                 break;
 896
 897         case IMAGE_RAW:
 898                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 899
 900                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME);
 901                 break;
 902
 903         case IMAGE_BLOCK:
 904         default:
 905                 return -EOPNOTSUPP;
 906         }
 907
 908         if (r < 0)
 909                 return r;
 910
 911         STRV_FOREACH(j, settings) {
 912                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 913                 if (r < 0 && r != -ENOENT)
 914                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 915         }
 916
 917         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 918         if (r < 0 && r != -ENOENT)
 919                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 920
 921         return 0;
 922 }
 923
 924 int image_read_only(Image *i, bool b) {
 925         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 926         int r;
 927
 928         assert(i);
 929
 930         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 931                 return -EROFS;
 932
 933         /* Make sure we don't interfere with a running nspawn */
 934         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 935         if (r < 0)
 936                 return r;
 937
 938         switch (i->type) {
 939
 940         case IMAGE_SUBVOLUME:
 941
 942                 /* Note that we set the flag only on the top-level
 943                  * subvolume of the image. */
 944
 945                 r = btrfs_subvol_set_read_only(i->path, b);
 946                 if (r < 0)
 947                         return r;
 948
 949                 break;
 950
 951         case IMAGE_DIRECTORY:
 952                 /* For simple directory trees we cannot use the access
 953                    mode of the top-level directory, since it has an
 954                    effect on the container itself.  However, we can
 955                    use the "immutable" flag, to at least make the
 956                    top-level directory read-only. It's not as good as
 957                    a read-only subvolume, but at least something, and
 958                    we can read the value back. */
 959
 960                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 961                 if (r < 0)
 962                         return r;
 963
 964                 break;
 965
 966         case IMAGE_RAW: {
 967                 struct stat st;
 968
 969                 if (stat(i->path, &st) < 0)
 970                         return -errno;
 971
 972                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 973                         return -errno;
 974
 975                 /* If the images is now read-only, it's a good time to
 976                  * defrag it, given that no write patterns will
 977                  * fragment it again. */
 978                 if (b)
 979                         (void) btrfs_defrag(i->path);
 980                 break;
 981         }
 982
 983         case IMAGE_BLOCK: {
 984                 _cleanup_close_ int fd = -1;
 985                 struct stat st;
 986                 int state = b;
 987
 988                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 989                 if (fd < 0)
 990                         return -errno;
 991
 992                 if (fstat(fd, &st) < 0)
 993                         return -errno;
 994                 if (!S_ISBLK(st.st_mode))
 995                         return -ENOTTY;
 996
 997                 if (ioctl(fd, BLKROSET, &state) < 0)
 998                         return -errno;
 999
1000                 break;
1001         }
1002
1003         default:
1004                 return -EOPNOTSUPP;
1005         }
1006
1007         return 0;
1008 }
1009
1010 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
1011         _cleanup_free_ char *p = NULL;
1012         LockFile t = LOCK_FILE_INIT;
1013         struct stat st;
1014         bool exclusive;
1015         int r;
1016
1017         assert(path);
1018         assert(global);
1019         assert(local);
1020
1021         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1022          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1023          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1024          * correctly. */
1025
1026         if (!path_is_absolute(path))
1027                 return -EINVAL;
1028
1029         switch (operation & (LOCK_SH|LOCK_EX)) {
1030         case LOCK_SH:
1031                 exclusive = false;
1032                 break;
1033         case LOCK_EX:
1034                 exclusive = true;
1035                 break;
1036         default:
1037                 return -EINVAL;
1038         }
1039
1040         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1041                 *local = *global = (LockFile) LOCK_FILE_INIT;
1042                 return 0;
1043         }
1044
1045         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1046          * running off it after all, and we don't want any images to manipulate the host image. We make an
1047          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1048          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1049          * since it means changes made to the host might propagate to the container as they happen (and a
1050          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1051          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1052          * the user to do the right thing with this. */
1053         if (path_equal(path, "/")) {
1054                 if (exclusive)
1055                         return -EBUSY;
1056
1057                 *local = *global = (LockFile) LOCK_FILE_INIT;
1058                 return 0;
1059         }
1060
1061         if (stat(path, &st) >= 0) {
1062                 if (S_ISBLK(st.st_mode))
1063                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1064                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1065                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1066                 else
1067                         return -ENOTTY;
1068
1069                 if (r < 0)
1070                         return -ENOMEM;
1071         }
1072
1073         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1074          * sufficient, since block devices are host local anyway. */
1075         if (!path_startswith(path, "/dev/")) {
1076                 r = make_lock_file_for(path, operation, &t);
1077                 if (r < 0) {
1078                         if (!exclusive && r == -EROFS)
1079                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1080                         else
1081                                 return r;
1082                 }
1083         }
1084
1085         if (p) {
1086                 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1087
1088                 r = make_lock_file(p, operation, global);
1089                 if (r < 0) {
1090                         release_lock_file(&t);
1091                         return r;
1092                 }
1093         } else
1094                 *global = (LockFile) LOCK_FILE_INIT;
1095
1096         *local = t;
1097         return 0;
1098 }
1099
1100 int image_set_limit(Image *i, uint64_t referenced_max) {
1101         assert(i);
1102
1103         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1104                 return -EROFS;
1105
1106         if (i->type != IMAGE_SUBVOLUME)
1107                 return -EOPNOTSUPP;
1108
1109         /* We set the quota both for the subvolume as well as for the
1110          * subtree. The latter is mostly for historical reasons, since
1111          * we didn't use to have a concept of subtree quota, and hence
1112          * only modified the subvolume quota. */
1113
1114         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1115         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1116         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1117 }
1118
1119 int image_read_metadata(Image *i) {
1120         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1121         int r;
1122
1123         assert(i);
1124
1125         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1126         if (r < 0)
1127                 return r;
1128
1129         switch (i->type) {
1130
1131         case IMAGE_SUBVOLUME:
1132         case IMAGE_DIRECTORY: {
1133                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1134                 sd_id128_t machine_id = SD_ID128_NULL;
1135                 _cleanup_free_ char *hostname = NULL;
1136                 _cleanup_free_ char *path = NULL;
1137
1138                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1139                 if (r < 0 && r != -ENOENT)
1140                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1141                 else if (r >= 0) {
1142                         r = read_etc_hostname(path, &hostname);
1143                         if (r < 0)
1144                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1145                 }
1146
1147                 path = mfree(path);
1148
1149                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1150                 if (r < 0 && r != -ENOENT)
1151                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1152                 else if (r >= 0) {
1153                         _cleanup_close_ int fd = -1;
1154
1155                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1156                         if (fd < 0)
1157                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1158                         else {
1159                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1160                                 if (r < 0)
1161                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1162                         }
1163                 }
1164
1165                 path = mfree(path);
1166
1167                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1168                 if (r < 0 && r != -ENOENT)
1169                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1170                 else if (r >= 0) {
1171                         r = load_env_file_pairs(NULL, path, &machine_info);
1172                         if (r < 0)
1173                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1174                 }
1175
1176                 r = load_os_release_pairs(i->path, &os_release);
1177                 if (r < 0)
1178                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1179
1180                 free_and_replace(i->hostname, hostname);
1181                 i->machine_id = machine_id;
1182                 strv_free_and_replace(i->machine_info, machine_info);
1183                 strv_free_and_replace(i->os_release, os_release);
1184
1185                 break;
1186         }
1187
1188         case IMAGE_RAW:
1189         case IMAGE_BLOCK: {
1190                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1191                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1192
1193                 r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, &d);
1194                 if (r < 0)
1195                         return r;
1196
1197                 r = dissect_image(d->fd, NULL, NULL, DISSECT_IMAGE_REQUIRE_ROOT|DISSECT_IMAGE_RELAX_VAR_CHECK, &m);
1198                 if (r < 0)
1199                         return r;
1200
1201                 r = dissected_image_acquire_metadata(m);
1202                 if (r < 0)
1203                         return r;
1204
1205                 free_and_replace(i->hostname, m->hostname);
1206                 i->machine_id = m->machine_id;
1207                 strv_free_and_replace(i->machine_info, m->machine_info);
1208                 strv_free_and_replace(i->os_release, m->os_release);
1209
1210                 break;
1211         }
1212
1213         default:
1214                 return -EOPNOTSUPP;
1215         }
1216
1217         i->metadata_valid = true;
1218
1219         return 0;
1220 }
1221
1222 int image_name_lock(const char *name, int operation, LockFile *ret) {
1223         assert(name);
1224         assert(ret);
1225
1226         /* Locks an image name, regardless of the precise path used. */
1227
1228         if (!image_name_is_valid(name))
1229                 return -EINVAL;
1230
1231         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1232                 *ret = (LockFile) LOCK_FILE_INIT;
1233                 return 0;
1234         }
1235
1236         if (streq(name, ".host"))
1237                 return -EBUSY;
1238
1239         const char *p = strjoina("/run/systemd/nspawn/locks/name-", name);
1240         (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1241         return make_lock_file(p, operation, ret);
1242 }
1243
1244 bool image_name_is_valid(const char *s) {
1245         if (!filename_is_valid(s))
1246                 return false;
1247
1248         if (string_has_cc(s, NULL))
1249                 return false;
1250
1251         if (!utf8_is_valid(s))
1252                 return false;
1253
1254         /* Temporary files for atomically creating new files */
1255         if (startswith(s, ".#"))
1256                 return false;
1257
1258         return true;
1259 }
1260
1261 bool image_in_search_path(
1262                 ImageClass class,
1263                 const char *root,
1264                 const char *image) {
1265
1266         const char *path;
1267
1268         assert(image);
1269
1270         NULSTR_FOREACH(path, image_search_path[class]) {
1271                 const char *p, *q;
1272                 size_t k;
1273
1274                 if (!empty_or_root(root)) {
1275                         q = path_startswith(path, root);
1276                         if (!q)
1277                                 continue;
1278                 } else
1279                         q = path;
1280
1281                 p = path_startswith(q, path);
1282                 if (!p)
1283                         continue;
1284
1285                 /* Make sure there's a filename following */
1286                 k = strcspn(p, "/");
1287                 if (k == 0)
1288                         continue;
1289
1290                 p += k;
1291
1292                 /* Accept trailing slashes */
1293                 if (p[strspn(p, "/")] == 0)
1294                         return true;
1295
1296         }
1297
1298         return false;
1299 }
1300
1301 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1302         [IMAGE_DIRECTORY] = "directory",
1303         [IMAGE_SUBVOLUME] = "subvolume",
1304         [IMAGE_RAW] = "raw",
1305         [IMAGE_BLOCK] = "block",
1306 };
1307
1308 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);