src/shared/machine-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2013 Lennart Poettering
   6 ***/
   7
   8 #include <dirent.h>
   9 #include <errno.h>
  10 #include <fcntl.h>
  11 #include <stdio.h>
  12 #include <stdlib.h>
  13 #include <string.h>
  14 #include <sys/file.h>
  15 #include <sys/stat.h>
  16 #include <unistd.h>
  17 #include <linux/fs.h>
  18
  19 #include "alloc-util.h"
  20 #include "btrfs-util.h"
  21 #include "chattr-util.h"
  22 #include "copy.h"
  23 #include "dirent-util.h"
  24 #include "dissect-image.h"
  25 #include "env-util.h"
  26 #include "fd-util.h"
  27 #include "fileio.h"
  28 #include "fs-util.h"
  29 #include "hashmap.h"
  30 #include "hostname-util.h"
  31 #include "id128-util.h"
  32 #include "lockfile-util.h"
  33 #include "log.h"
  34 #include "loop-util.h"
  35 #include "machine-image.h"
  36 #include "macro.h"
  37 #include "mkdir.h"
  38 #include "path-util.h"
  39 #include "rm-rf.h"
  40 #include "string-table.h"
  41 #include "string-util.h"
  42 #include "strv.h"
  43 #include "time-util.h"
  44 #include "utf8.h"
  45 #include "util.h"
  46 #include "xattr-util.h"
  47
  48 static const char image_search_path[] =
  49         "/var/lib/machines\0"
  50         "/var/lib/container\0" /* legacy */
  51         "/usr/local/lib/machines\0"
  52         "/usr/lib/machines\0";
  53
  54 Image *image_unref(Image *i) {
  55         if (!i)
  56                 return NULL;
  57
  58         free(i->name);
  59         free(i->path);
  60
  61         free(i->hostname);
  62         strv_free(i->machine_info);
  63         strv_free(i->os_release);
  64
  65         return mfree(i);
  66 }
  67
  68 static char **image_settings_path(Image *image) {
  69         _cleanup_strv_free_ char **l = NULL;
  70         const char *fn, *s;
  71         unsigned i = 0;
  72
  73         assert(image);
  74
  75         l = new0(char*, 4);
  76         if (!l)
  77                 return NULL;
  78
  79         fn = strjoina(image->name, ".nspawn");
  80
  81         FOREACH_STRING(s, "/etc/systemd/nspawn/", "/run/systemd/nspawn/") {
  82                 l[i] = strappend(s, fn);
  83                 if (!l[i])
  84                         return NULL;
  85
  86                 i++;
  87         }
  88
  89         l[i] = file_in_same_dir(image->path, fn);
  90         if (!l[i])
  91                 return NULL;
  92
  93         return TAKE_PTR(l);
  94 }
  95
  96 static char *image_roothash_path(Image *image) {
  97         const char *fn;
  98
  99         assert(image);
 100
 101         fn = strjoina(image->name, ".roothash");
 102
 103         return file_in_same_dir(image->path, fn);
 104 }
 105
 106 static int image_new(
 107                 ImageType t,
 108                 const char *pretty,
 109                 const char *path,
 110                 const char *filename,
 111                 bool read_only,
 112                 usec_t crtime,
 113                 usec_t mtime,
 114                 Image **ret) {
 115
 116         _cleanup_(image_unrefp) Image *i = NULL;
 117
 118         assert(t >= 0);
 119         assert(t < _IMAGE_TYPE_MAX);
 120         assert(pretty);
 121         assert(filename);
 122         assert(ret);
 123
 124         i = new0(Image, 1);
 125         if (!i)
 126                 return -ENOMEM;
 127
 128         i->type = t;
 129         i->read_only = read_only;
 130         i->crtime = crtime;
 131         i->mtime = mtime;
 132         i->usage = i->usage_exclusive = (uint64_t) -1;
 133         i->limit = i->limit_exclusive = (uint64_t) -1;
 134
 135         i->name = strdup(pretty);
 136         if (!i->name)
 137                 return -ENOMEM;
 138
 139         if (path)
 140                 i->path = strjoin(path, "/", filename);
 141         else
 142                 i->path = strdup(filename);
 143
 144         if (!i->path)
 145                 return -ENOMEM;
 146
 147         path_kill_slashes(i->path);
 148
 149         *ret = TAKE_PTR(i);
 150
 151         return 0;
 152 }
 153
 154 static int image_make(
 155                 const char *pretty,
 156                 int dfd,
 157                 const char *path,
 158                 const char *filename,
 159                 Image **ret) {
 160
 161         struct stat st;
 162         bool read_only;
 163         int r;
 164
 165         assert(filename);
 166
 167         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 168          * devices into /var/lib/machines/, and treat them normally. */
 169
 170         if (fstatat(dfd, filename, &st, 0) < 0)
 171                 return -errno;
 172
 173         read_only =
 174                 (path && path_startswith(path, "/usr")) ||
 175                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 176
 177         if (S_ISDIR(st.st_mode)) {
 178                 _cleanup_close_ int fd = -1;
 179                 unsigned file_attr = 0;
 180
 181                 if (!ret)
 182                         return 1;
 183
 184                 if (!pretty)
 185                         pretty = filename;
 186
 187                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 188                 if (fd < 0)
 189                         return -errno;
 190
 191                 /* btrfs subvolumes have inode 256 */
 192                 if (st.st_ino == 256) {
 193
 194                         r = btrfs_is_filesystem(fd);
 195                         if (r < 0)
 196                                 return r;
 197                         if (r) {
 198                                 BtrfsSubvolInfo info;
 199
 200                                 /* It's a btrfs subvolume */
 201
 202                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 203                                 if (r < 0)
 204                                         return r;
 205
 206                                 r = image_new(IMAGE_SUBVOLUME,
 207                                               pretty,
 208                                               path,
 209                                               filename,
 210                                               info.read_only || read_only,
 211                                               info.otime,
 212                                               0,
 213                                               ret);
 214                                 if (r < 0)
 215                                         return r;
 216
 217                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 218                                         BtrfsQuotaInfo quota;
 219
 220                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 221                                         if (r >= 0) {
 222                                                 (*ret)->usage = quota.referenced;
 223                                                 (*ret)->usage_exclusive = quota.exclusive;
 224
 225                                                 (*ret)->limit = quota.referenced_max;
 226                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 227                                         }
 228                                 }
 229
 230                                 return 1;
 231                         }
 232                 }
 233
 234                 /* If the IMMUTABLE bit is set, we consider the
 235                  * directory read-only. Since the ioctl is not
 236                  * supported everywhere we ignore failures. */
 237                 (void) read_attr_fd(fd, &file_attr);
 238
 239                 /* It's just a normal directory. */
 240                 r = image_new(IMAGE_DIRECTORY,
 241                               pretty,
 242                               path,
 243                               filename,
 244                               read_only || (file_attr & FS_IMMUTABLE_FL),
 245                               0,
 246                               0,
 247                               ret);
 248                 if (r < 0)
 249                         return r;
 250
 251                 return 1;
 252
 253         } else if (S_ISREG(st.st_mode) && endswith(filename, ".raw")) {
 254                 usec_t crtime = 0;
 255
 256                 /* It's a RAW disk image */
 257
 258                 if (!ret)
 259                         return 1;
 260
 261                 fd_getcrtime_at(dfd, filename, &crtime, 0);
 262
 263                 if (!pretty)
 264                         pretty = strndupa(filename, strlen(filename) - 4);
 265
 266                 r = image_new(IMAGE_RAW,
 267                               pretty,
 268                               path,
 269                               filename,
 270                               !(st.st_mode & 0222) || read_only,
 271                               crtime,
 272                               timespec_load(&st.st_mtim),
 273                               ret);
 274                 if (r < 0)
 275                         return r;
 276
 277                 (*ret)->usage = (*ret)->usage_exclusive = st.st_blocks * 512;
 278                 (*ret)->limit = (*ret)->limit_exclusive = st.st_size;
 279
 280                 return 1;
 281
 282         } else if (S_ISBLK(st.st_mode)) {
 283                 _cleanup_close_ int block_fd = -1;
 284                 uint64_t size = UINT64_MAX;
 285
 286                 /* A block device */
 287
 288                 if (!ret)
 289                         return 1;
 290
 291                 if (!pretty)
 292                         pretty = filename;
 293
 294                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 295                 if (block_fd < 0)
 296                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path, filename);
 297                 else {
 298                         if (fstat(block_fd, &st) < 0)
 299                                 return -errno;
 300                         if (!S_ISBLK(st.st_mode)) /* Verify that what we opened is actually what we think it is */
 301                                 return -ENOTTY;
 302
 303                         if (!read_only) {
 304                                 int state = 0;
 305
 306                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 307                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path, filename);
 308                                 else if (state)
 309                                         read_only = true;
 310                         }
 311
 312                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 313                                 log_debug_errno(errno, "Failed to issue BLKFLSBUF on device %s/%s, ignoring: %m", path, filename);
 314
 315                         block_fd = safe_close(block_fd);
 316                 }
 317
 318                 r = image_new(IMAGE_BLOCK,
 319                               pretty,
 320                               path,
 321                               filename,
 322                               !(st.st_mode & 0222) || read_only,
 323                               0,
 324                               0,
 325                               ret);
 326                 if (r < 0)
 327                         return r;
 328
 329                 if (size != 0 && size != UINT64_MAX)
 330                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 331
 332                 return 1;
 333         }
 334
 335         return 0;
 336 }
 337
 338 int image_find(const char *name, Image **ret) {
 339         const char *path;
 340         int r;
 341
 342         assert(name);
 343
 344         /* There are no images with invalid names */
 345         if (!image_name_is_valid(name))
 346                 return 0;
 347
 348         NULSTR_FOREACH(path, image_search_path) {
 349                 _cleanup_closedir_ DIR *d = NULL;
 350
 351                 d = opendir(path);
 352                 if (!d) {
 353                         if (errno == ENOENT)
 354                                 continue;
 355
 356                         return -errno;
 357                 }
 358
 359                 r = image_make(NULL, dirfd(d), path, name, ret);
 360                 if (IN_SET(r, 0, -ENOENT)) {
 361                         _cleanup_free_ char *raw = NULL;
 362
 363                         raw = strappend(name, ".raw");
 364                         if (!raw)
 365                                 return -ENOMEM;
 366
 367                         r = image_make(NULL, dirfd(d), path, raw, ret);
 368                         if (IN_SET(r, 0, -ENOENT))
 369                                 continue;
 370                 }
 371                 if (r < 0)
 372                         return r;
 373
 374                 return 1;
 375         }
 376
 377         if (streq(name, ".host"))
 378                 return image_make(".host", AT_FDCWD, NULL, "/", ret);
 379
 380         return 0;
 381 };
 382
 383 int image_discover(Hashmap *h) {
 384         const char *path;
 385         int r;
 386
 387         assert(h);
 388
 389         NULSTR_FOREACH(path, image_search_path) {
 390                 _cleanup_closedir_ DIR *d = NULL;
 391                 struct dirent *de;
 392
 393                 d = opendir(path);
 394                 if (!d) {
 395                         if (errno == ENOENT)
 396                                 continue;
 397
 398                         return -errno;
 399                 }
 400
 401                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 402                         _cleanup_(image_unrefp) Image *image = NULL;
 403
 404                         if (!image_name_is_valid(de->d_name))
 405                                 continue;
 406
 407                         if (hashmap_contains(h, de->d_name))
 408                                 continue;
 409
 410                         r = image_make(NULL, dirfd(d), path, de->d_name, &image);
 411                         if (IN_SET(r, 0, -ENOENT))
 412                                 continue;
 413                         if (r < 0)
 414                                 return r;
 415
 416                         r = hashmap_put(h, image->name, image);
 417                         if (r < 0)
 418                                 return r;
 419
 420                         image = NULL;
 421                 }
 422         }
 423
 424         if (!hashmap_contains(h, ".host")) {
 425                 _cleanup_(image_unrefp) Image *image = NULL;
 426
 427                 r = image_make(".host", AT_FDCWD, NULL, "/", &image);
 428                 if (r < 0)
 429                         return r;
 430
 431                 r = hashmap_put(h, image->name, image);
 432                 if (r < 0)
 433                         return r;
 434
 435                 image = NULL;
 436
 437         }
 438
 439         return 0;
 440 }
 441
 442 int image_remove(Image *i) {
 443         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 444         _cleanup_strv_free_ char **settings = NULL;
 445         _cleanup_free_ char *roothash = NULL;
 446         char **j;
 447         int r;
 448
 449         assert(i);
 450
 451         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 452                 return -EROFS;
 453
 454         settings = image_settings_path(i);
 455         if (!settings)
 456                 return -ENOMEM;
 457
 458         roothash = image_roothash_path(i);
 459         if (!roothash)
 460                 return -ENOMEM;
 461
 462         /* Make sure we don't interfere with a running nspawn */
 463         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 464         if (r < 0)
 465                 return r;
 466
 467         switch (i->type) {
 468
 469         case IMAGE_SUBVOLUME:
 470
 471                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 472                  * big guns */
 473                 if (unlink(i->path) < 0) {
 474                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 475                         if (r < 0)
 476                                 return r;
 477                 }
 478
 479                 break;
 480
 481         case IMAGE_DIRECTORY:
 482                 /* Allow deletion of read-only directories */
 483                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL);
 484                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 485                 if (r < 0)
 486                         return r;
 487
 488                 break;
 489
 490         case IMAGE_BLOCK:
 491
 492                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 493                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 494                  * the thing (it's most likely a symlink after all). */
 495
 496                 if (path_startswith(i->path, "/dev"))
 497                         break;
 498
 499                 _fallthrough_;
 500         case IMAGE_RAW:
 501                 if (unlink(i->path) < 0)
 502                         return -errno;
 503                 break;
 504
 505         default:
 506                 return -EOPNOTSUPP;
 507         }
 508
 509         STRV_FOREACH(j, settings) {
 510                 if (unlink(*j) < 0 && errno != ENOENT)
 511                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 512         }
 513
 514         if (unlink(roothash) < 0 && errno != ENOENT)
 515                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 516
 517         return 0;
 518 }
 519
 520 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 521         _cleanup_free_ char *rs = NULL;
 522         const char *fn;
 523
 524         fn = strjoina(new_name, suffix);
 525
 526         rs = file_in_same_dir(path, fn);
 527         if (!rs)
 528                 return -ENOMEM;
 529
 530         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 531 }
 532
 533 int image_rename(Image *i, const char *new_name) {
 534         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 535         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 536         _cleanup_strv_free_ char **settings = NULL;
 537         unsigned file_attr = 0;
 538         char **j;
 539         int r;
 540
 541         assert(i);
 542
 543         if (!image_name_is_valid(new_name))
 544                 return -EINVAL;
 545
 546         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 547                 return -EROFS;
 548
 549         settings = image_settings_path(i);
 550         if (!settings)
 551                 return -ENOMEM;
 552
 553         roothash = image_roothash_path(i);
 554         if (!roothash)
 555                 return -ENOMEM;
 556
 557         /* Make sure we don't interfere with a running nspawn */
 558         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 559         if (r < 0)
 560                 return r;
 561
 562         /* Make sure nobody takes the new name, between the time we
 563          * checked it is currently unused in all search paths, and the
 564          * time we take possession of it */
 565         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 566         if (r < 0)
 567                 return r;
 568
 569         r = image_find(new_name, NULL);
 570         if (r < 0)
 571                 return r;
 572         if (r > 0)
 573                 return -EEXIST;
 574
 575         switch (i->type) {
 576
 577         case IMAGE_DIRECTORY:
 578                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 579                 (void) read_attr_path(i->path, &file_attr);
 580
 581                 if (file_attr & FS_IMMUTABLE_FL)
 582                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL);
 583
 584                 _fallthrough_;
 585         case IMAGE_SUBVOLUME:
 586                 new_path = file_in_same_dir(i->path, new_name);
 587                 break;
 588
 589         case IMAGE_BLOCK:
 590
 591                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 592                 if (path_startswith(i->path, "/dev"))
 593                         return -EROFS;
 594
 595                 new_path = file_in_same_dir(i->path, new_name);
 596                 break;
 597
 598         case IMAGE_RAW: {
 599                 const char *fn;
 600
 601                 fn = strjoina(new_name, ".raw");
 602                 new_path = file_in_same_dir(i->path, fn);
 603                 break;
 604         }
 605
 606         default:
 607                 return -EOPNOTSUPP;
 608         }
 609
 610         if (!new_path)
 611                 return -ENOMEM;
 612
 613         nn = strdup(new_name);
 614         if (!nn)
 615                 return -ENOMEM;
 616
 617         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 618         if (r < 0)
 619                 return r;
 620
 621         /* Restore the immutable bit, if it was set before */
 622         if (file_attr & FS_IMMUTABLE_FL)
 623                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL);
 624
 625         free_and_replace(i->path, new_path);
 626         free_and_replace(i->name, nn);
 627
 628         STRV_FOREACH(j, settings) {
 629                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 630                 if (r < 0 && r != -ENOENT)
 631                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 632         }
 633
 634         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 635         if (r < 0 && r != -ENOENT)
 636                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 637
 638         return 0;
 639 }
 640
 641 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 642         _cleanup_free_ char *rs = NULL;
 643         const char *fn;
 644
 645         fn = strjoina(new_name, suffix);
 646
 647         rs = file_in_same_dir(path, fn);
 648         if (!rs)
 649                 return -ENOMEM;
 650
 651         return copy_file_atomic(path, rs, 0664, 0, COPY_REFLINK);
 652 }
 653
 654 int image_clone(Image *i, const char *new_name, bool read_only) {
 655         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 656         _cleanup_strv_free_ char **settings = NULL;
 657         _cleanup_free_ char *roothash = NULL;
 658         const char *new_path;
 659         char **j;
 660         int r;
 661
 662         assert(i);
 663
 664         if (!image_name_is_valid(new_name))
 665                 return -EINVAL;
 666
 667         settings = image_settings_path(i);
 668         if (!settings)
 669                 return -ENOMEM;
 670
 671         roothash = image_roothash_path(i);
 672         if (!roothash)
 673                 return -ENOMEM;
 674
 675         /* Make sure nobody takes the new name, between the time we
 676          * checked it is currently unused in all search paths, and the
 677          * time we take possession of it */
 678         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 679         if (r < 0)
 680                 return r;
 681
 682         r = image_find(new_name, NULL);
 683         if (r < 0)
 684                 return r;
 685         if (r > 0)
 686                 return -EEXIST;
 687
 688         switch (i->type) {
 689
 690         case IMAGE_SUBVOLUME:
 691         case IMAGE_DIRECTORY:
 692                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 693                  * directory. */
 694
 695                 new_path = strjoina("/var/lib/machines/", new_name);
 696
 697                 r = btrfs_subvol_snapshot(i->path, new_path,
 698                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 699                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 700                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 701                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 702                                           BTRFS_SNAPSHOT_RECURSIVE |
 703                                           BTRFS_SNAPSHOT_QUOTA);
 704                 if (r >= 0)
 705                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 706                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 707
 708                 break;
 709
 710         case IMAGE_RAW:
 711                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 712
 713                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, COPY_REFLINK);
 714                 break;
 715
 716         case IMAGE_BLOCK:
 717         default:
 718                 return -EOPNOTSUPP;
 719         }
 720
 721         if (r < 0)
 722                 return r;
 723
 724         STRV_FOREACH(j, settings) {
 725                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 726                 if (r < 0 && r != -ENOENT)
 727                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 728         }
 729
 730         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 731         if (r < 0 && r != -ENOENT)
 732                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 733
 734         return 0;
 735 }
 736
 737 int image_read_only(Image *i, bool b) {
 738         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 739         int r;
 740
 741         assert(i);
 742
 743         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 744                 return -EROFS;
 745
 746         /* Make sure we don't interfere with a running nspawn */
 747         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 748         if (r < 0)
 749                 return r;
 750
 751         switch (i->type) {
 752
 753         case IMAGE_SUBVOLUME:
 754
 755                 /* Note that we set the flag only on the top-level
 756                  * subvolume of the image. */
 757
 758                 r = btrfs_subvol_set_read_only(i->path, b);
 759                 if (r < 0)
 760                         return r;
 761
 762                 break;
 763
 764         case IMAGE_DIRECTORY:
 765                 /* For simple directory trees we cannot use the access
 766                    mode of the top-level directory, since it has an
 767                    effect on the container itself.  However, we can
 768                    use the "immutable" flag, to at least make the
 769                    top-level directory read-only. It's not as good as
 770                    a read-only subvolume, but at least something, and
 771                    we can read the value back. */
 772
 773                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL);
 774                 if (r < 0)
 775                         return r;
 776
 777                 break;
 778
 779         case IMAGE_RAW: {
 780                 struct stat st;
 781
 782                 if (stat(i->path, &st) < 0)
 783                         return -errno;
 784
 785                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 786                         return -errno;
 787
 788                 /* If the images is now read-only, it's a good time to
 789                  * defrag it, given that no write patterns will
 790                  * fragment it again. */
 791                 if (b)
 792                         (void) btrfs_defrag(i->path);
 793                 break;
 794         }
 795
 796         case IMAGE_BLOCK: {
 797                 _cleanup_close_ int fd = -1;
 798                 struct stat st;
 799                 int state = b;
 800
 801                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 802                 if (fd < 0)
 803                         return -errno;
 804
 805                 if (fstat(fd, &st) < 0)
 806                         return -errno;
 807                 if (!S_ISBLK(st.st_mode))
 808                         return -ENOTTY;
 809
 810                 if (ioctl(fd, BLKROSET, &state) < 0)
 811                         return -errno;
 812
 813                 break;
 814         }
 815
 816         default:
 817                 return -EOPNOTSUPP;
 818         }
 819
 820         return 0;
 821 }
 822
 823 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
 824         _cleanup_free_ char *p = NULL;
 825         LockFile t = LOCK_FILE_INIT;
 826         struct stat st;
 827         int r;
 828
 829         assert(path);
 830         assert(global);
 831         assert(local);
 832
 833         /* Locks an image path. This actually creates two locks: one
 834          * "local" one, next to the image path itself, which might be
 835          * shared via NFS. And another "global" one, in /run, that
 836          * uses the device/inode number. This has the benefit that we
 837          * can even lock a tree that is a mount point, correctly. */
 838
 839         if (!path_is_absolute(path))
 840                 return -EINVAL;
 841
 842         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
 843                 *local = *global = (LockFile) LOCK_FILE_INIT;
 844                 return 0;
 845         }
 846
 847         if (path_equal(path, "/"))
 848                 return -EBUSY;
 849
 850         if (stat(path, &st) >= 0) {
 851                 if (S_ISBLK(st.st_mode))
 852                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
 853                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
 854                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
 855                 else
 856                         return -ENOTTY;
 857
 858                 if (r < 0)
 859                         return -ENOMEM;
 860         }
 861
 862         /* For block devices we don't need the "local" lock, as the major/minor lock above should be sufficient, since
 863          * block devices are device local anyway. */
 864         if (!path_startswith(path, "/dev")) {
 865                 r = make_lock_file_for(path, operation, &t);
 866                 if (r < 0) {
 867                         if ((operation & LOCK_SH) && r == -EROFS)
 868                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
 869                         else
 870                                 return r;
 871                 }
 872         }
 873
 874         if (p) {
 875                 mkdir_p("/run/systemd/nspawn/locks", 0700);
 876
 877                 r = make_lock_file(p, operation, global);
 878                 if (r < 0) {
 879                         release_lock_file(&t);
 880                         return r;
 881                 }
 882         } else
 883                 *global = (LockFile) LOCK_FILE_INIT;
 884
 885         *local = t;
 886         return 0;
 887 }
 888
 889 int image_set_limit(Image *i, uint64_t referenced_max) {
 890         assert(i);
 891
 892         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 893                 return -EROFS;
 894
 895         if (i->type != IMAGE_SUBVOLUME)
 896                 return -EOPNOTSUPP;
 897
 898         /* We set the quota both for the subvolume as well as for the
 899          * subtree. The latter is mostly for historical reasons, since
 900          * we didn't use to have a concept of subtree quota, and hence
 901          * only modified the subvolume quota. */
 902
 903         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
 904         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
 905         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
 906 }
 907
 908 int image_read_metadata(Image *i) {
 909         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 910         int r;
 911
 912         assert(i);
 913
 914         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
 915         if (r < 0)
 916                 return r;
 917
 918         switch (i->type) {
 919
 920         case IMAGE_SUBVOLUME:
 921         case IMAGE_DIRECTORY: {
 922                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
 923                 sd_id128_t machine_id = SD_ID128_NULL;
 924                 _cleanup_free_ char *hostname = NULL;
 925                 _cleanup_free_ char *path = NULL;
 926
 927                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
 928                 if (r < 0 && r != -ENOENT)
 929                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
 930                 else if (r >= 0) {
 931                         r = read_etc_hostname(path, &hostname);
 932                         if (r < 0)
 933                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
 934                 }
 935
 936                 path = mfree(path);
 937
 938                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
 939                 if (r < 0 && r != -ENOENT)
 940                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
 941                 else if (r >= 0) {
 942                         _cleanup_close_ int fd = -1;
 943
 944                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
 945                         if (fd < 0)
 946                                 log_debug_errno(errno, "Failed to open %s: %m", path);
 947                         else {
 948                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
 949                                 if (r < 0)
 950                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
 951                         }
 952                 }
 953
 954                 path = mfree(path);
 955
 956                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
 957                 if (r < 0 && r != -ENOENT)
 958                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
 959                 else if (r >= 0) {
 960                         r = load_env_file_pairs(NULL, path, NULL, &machine_info);
 961                         if (r < 0)
 962                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
 963                 }
 964
 965                 path = mfree(path);
 966
 967                 r = chase_symlinks("/etc/os-release", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
 968                 if (r == -ENOENT)
 969                         r = chase_symlinks("/usr/lib/os-release", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
 970                 if (r < 0 && r != -ENOENT)
 971                         log_debug_errno(r, "Failed to chase os-release in image: %m");
 972                 else if (r >= 0) {
 973                         r = load_env_file_pairs(NULL, path, NULL, &os_release);
 974                         if (r < 0)
 975                                 log_debug_errno(r, "Failed to parse os-release data of %s: %m", i->name);
 976                 }
 977
 978                 free_and_replace(i->hostname, hostname);
 979                 i->machine_id = machine_id;
 980                 strv_free_and_replace(i->machine_info, machine_info);
 981                 strv_free_and_replace(i->os_release, os_release);
 982
 983                 break;
 984         }
 985
 986         case IMAGE_RAW:
 987         case IMAGE_BLOCK: {
 988                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
 989                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
 990
 991                 r = loop_device_make_by_path(i->path, O_RDONLY, &d);
 992                 if (r < 0)
 993                         return r;
 994
 995                 r = dissect_image(d->fd, NULL, 0, DISSECT_IMAGE_REQUIRE_ROOT, &m);
 996                 if (r < 0)
 997                         return r;
 998
 999                 r = dissected_image_acquire_metadata(m);
1000                 if (r < 0)
1001                         return r;
1002
1003                 free_and_replace(i->hostname, m->hostname);
1004                 i->machine_id = m->machine_id;
1005                 strv_free_and_replace(i->machine_info, m->machine_info);
1006                 strv_free_and_replace(i->os_release, m->os_release);
1007
1008                 break;
1009         }
1010
1011         default:
1012                 return -EOPNOTSUPP;
1013         }
1014
1015         i->metadata_valid = true;
1016
1017         return 0;
1018 }
1019
1020 int image_name_lock(const char *name, int operation, LockFile *ret) {
1021         const char *p;
1022
1023         assert(name);
1024         assert(ret);
1025
1026         /* Locks an image name, regardless of the precise path used. */
1027
1028         if (!image_name_is_valid(name))
1029                 return -EINVAL;
1030
1031         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1032                 *ret = (LockFile) LOCK_FILE_INIT;
1033                 return 0;
1034         }
1035
1036         if (streq(name, ".host"))
1037                 return -EBUSY;
1038
1039         mkdir_p("/run/systemd/nspawn/locks", 0700);
1040         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1041
1042         return make_lock_file(p, operation, ret);
1043 }
1044
1045 bool image_name_is_valid(const char *s) {
1046         if (!filename_is_valid(s))
1047                 return false;
1048
1049         if (string_has_cc(s, NULL))
1050                 return false;
1051
1052         if (!utf8_is_valid(s))
1053                 return false;
1054
1055         /* Temporary files for atomically creating new files */
1056         if (startswith(s, ".#"))
1057                 return false;
1058
1059         return true;
1060 }
1061
1062 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1063         [IMAGE_DIRECTORY] = "directory",
1064         [IMAGE_SUBVOLUME] = "subvolume",
1065         [IMAGE_RAW] = "raw",
1066         [IMAGE_BLOCK] = "block",
1067 };
1068
1069 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);