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