src/shared/machine-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <dirent.h>
   4 #include <errno.h>
   5 #include <fcntl.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <string.h>
   9 #include <sys/file.h>
  10 #include <sys/ioctl.h>
  11 #include <sys/stat.h>
  12 #include <unistd.h>
  13 #include <linux/fs.h>
  14
  15 #include "alloc-util.h"
  16 #include "btrfs-util.h"
  17 #include "chattr-util.h"
  18 #include "copy.h"
  19 #include "dirent-util.h"
  20 #include "dissect-image.h"
  21 #include "env-file.h"
  22 #include "env-util.h"
  23 #include "fd-util.h"
  24 #include "fs-util.h"
  25 #include "hashmap.h"
  26 #include "hostname-util.h"
  27 #include "id128-util.h"
  28 #include "lockfile-util.h"
  29 #include "log.h"
  30 #include "loop-util.h"
  31 #include "machine-image.h"
  32 #include "macro.h"
  33 #include "mkdir.h"
  34 #include "nulstr-util.h"
  35 #include "os-util.h"
  36 #include "path-util.h"
  37 #include "rm-rf.h"
  38 #include "string-table.h"
  39 #include "string-util.h"
  40 #include "strv.h"
  41 #include "time-util.h"
  42 #include "utf8.h"
  43 #include "xattr-util.h"
  44
  45 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  46         [IMAGE_MACHINE] =  "/etc/machines\0"              /* only place symlinks here */
  47                            "/run/machines\0"              /* and here too */
  48                            "/var/lib/machines\0"          /* the main place for images */
  49                            "/var/lib/container\0"         /* legacy */
  50                            "/usr/local/lib/machines\0"
  51                            "/usr/lib/machines\0",
  52
  53         [IMAGE_PORTABLE] = "/etc/portables\0"             /* only place symlinks here */
  54                            "/run/portables\0"             /* and here too */
  55                            "/var/lib/portables\0"         /* the main place for images */
  56                            "/usr/local/lib/portables\0"
  57                            "/usr/lib/portables\0",
  58 };
  59
  60 static Image *image_free(Image *i) {
  61         assert(i);
  62
  63         free(i->name);
  64         free(i->path);
  65
  66         free(i->hostname);
  67         strv_free(i->machine_info);
  68         strv_free(i->os_release);
  69
  70         return mfree(i);
  71 }
  72
  73 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  74 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
  75                                       Image, image_unref);
  76
  77 static char **image_settings_path(Image *image) {
  78         _cleanup_strv_free_ char **l = NULL;
  79         const char *fn, *s;
  80         unsigned i = 0;
  81
  82         assert(image);
  83
  84         l = new0(char*, 4);
  85         if (!l)
  86                 return NULL;
  87
  88         fn = strjoina(image->name, ".nspawn");
  89
  90         FOREACH_STRING(s, "/etc/systemd/nspawn/", "/run/systemd/nspawn/") {
  91                 l[i] = strappend(s, fn);
  92                 if (!l[i])
  93                         return NULL;
  94
  95                 i++;
  96         }
  97
  98         l[i] = file_in_same_dir(image->path, fn);
  99         if (!l[i])
 100                 return NULL;
 101
 102         return TAKE_PTR(l);
 103 }
 104
 105 static char *image_roothash_path(Image *image) {
 106         const char *fn;
 107
 108         assert(image);
 109
 110         fn = strjoina(image->name, ".roothash");
 111
 112         return file_in_same_dir(image->path, fn);
 113 }
 114
 115 static int image_new(
 116                 ImageType t,
 117                 const char *pretty,
 118                 const char *path,
 119                 const char *filename,
 120                 bool read_only,
 121                 usec_t crtime,
 122                 usec_t mtime,
 123                 Image **ret) {
 124
 125         _cleanup_(image_unrefp) Image *i = NULL;
 126
 127         assert(t >= 0);
 128         assert(t < _IMAGE_TYPE_MAX);
 129         assert(pretty);
 130         assert(filename);
 131         assert(ret);
 132
 133         i = new0(Image, 1);
 134         if (!i)
 135                 return -ENOMEM;
 136
 137         i->n_ref = 1;
 138         i->type = t;
 139         i->read_only = read_only;
 140         i->crtime = crtime;
 141         i->mtime = mtime;
 142         i->usage = i->usage_exclusive = (uint64_t) -1;
 143         i->limit = i->limit_exclusive = (uint64_t) -1;
 144
 145         i->name = strdup(pretty);
 146         if (!i->name)
 147                 return -ENOMEM;
 148
 149         if (path)
 150                 i->path = strjoin(path, "/", filename);
 151         else
 152                 i->path = strdup(filename);
 153         if (!i->path)
 154                 return -ENOMEM;
 155
 156         path_simplify(i->path, false);
 157
 158         *ret = TAKE_PTR(i);
 159
 160         return 0;
 161 }
 162
 163 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 164         _cleanup_free_ char *name = NULL;
 165         const char *p;
 166         size_t n;
 167
 168         assert(path);
 169         assert(ret);
 170
 171         p = last_path_component(path);
 172         n = strcspn(p, "/");
 173
 174         name = strndup(p, n);
 175         if (!name)
 176                 return -ENOMEM;
 177
 178         if (suffix) {
 179                 char *e;
 180
 181                 e = endswith(name, suffix);
 182                 if (!e)
 183                         return -EINVAL;
 184
 185                 *e = 0;
 186         }
 187
 188         if (!image_name_is_valid(name))
 189                 return -EINVAL;
 190
 191         *ret = TAKE_PTR(name);
 192         return 0;
 193 }
 194
 195 static int image_make(
 196                 const char *pretty,
 197                 int dfd,
 198                 const char *path,
 199                 const char *filename,
 200                 const struct stat *st,
 201                 Image **ret) {
 202
 203         _cleanup_free_ char *pretty_buffer = NULL;
 204         struct stat stbuf;
 205         bool read_only;
 206         int r;
 207
 208         assert(dfd >= 0 || dfd == AT_FDCWD);
 209         assert(filename);
 210
 211         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 212          * devices into /var/lib/machines/, and treat them normally.
 213          *
 214          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 215          * recognize. */
 216
 217         if (!st) {
 218                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 219                         return -errno;
 220
 221                 st = &stbuf;
 222         }
 223
 224         read_only =
 225                 (path && path_startswith(path, "/usr")) ||
 226                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 227
 228         if (S_ISDIR(st->st_mode)) {
 229                 _cleanup_close_ int fd = -1;
 230                 unsigned file_attr = 0;
 231
 232                 if (!ret)
 233                         return 0;
 234
 235                 if (!pretty) {
 236                         r = extract_pretty(filename, NULL, &pretty_buffer);
 237                         if (r < 0)
 238                                 return r;
 239
 240                         pretty = pretty_buffer;
 241                 }
 242
 243                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 244                 if (fd < 0)
 245                         return -errno;
 246
 247                 /* btrfs subvolumes have inode 256 */
 248                 if (st->st_ino == 256) {
 249
 250                         r = btrfs_is_filesystem(fd);
 251                         if (r < 0)
 252                                 return r;
 253                         if (r) {
 254                                 BtrfsSubvolInfo info;
 255
 256                                 /* It's a btrfs subvolume */
 257
 258                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 259                                 if (r < 0)
 260                                         return r;
 261
 262                                 r = image_new(IMAGE_SUBVOLUME,
 263                                               pretty,
 264                                               path,
 265                                               filename,
 266                                               info.read_only || read_only,
 267                                               info.otime,
 268                                               0,
 269                                               ret);
 270                                 if (r < 0)
 271                                         return r;
 272
 273                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 274                                         BtrfsQuotaInfo quota;
 275
 276                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 277                                         if (r >= 0) {
 278                                                 (*ret)->usage = quota.referenced;
 279                                                 (*ret)->usage_exclusive = quota.exclusive;
 280
 281                                                 (*ret)->limit = quota.referenced_max;
 282                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 283                                         }
 284                                 }
 285
 286                                 return 0;
 287                         }
 288                 }
 289
 290                 /* If the IMMUTABLE bit is set, we consider the
 291                  * directory read-only. Since the ioctl is not
 292                  * supported everywhere we ignore failures. */
 293                 (void) read_attr_fd(fd, &file_attr);
 294
 295                 /* It's just a normal directory. */
 296                 r = image_new(IMAGE_DIRECTORY,
 297                               pretty,
 298                               path,
 299                               filename,
 300                               read_only || (file_attr & FS_IMMUTABLE_FL),
 301                               0,
 302                               0,
 303                               ret);
 304                 if (r < 0)
 305                         return r;
 306
 307                 return 0;
 308
 309         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 310                 usec_t crtime = 0;
 311
 312                 /* It's a RAW disk image */
 313
 314                 if (!ret)
 315                         return 0;
 316
 317                 (void) fd_getcrtime_at(dfd, filename, &crtime, 0);
 318
 319                 if (!pretty) {
 320                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 321                         if (r < 0)
 322                                 return r;
 323
 324                         pretty = pretty_buffer;
 325                 }
 326
 327                 r = image_new(IMAGE_RAW,
 328                               pretty,
 329                               path,
 330                               filename,
 331                               !(st->st_mode & 0222) || read_only,
 332                               crtime,
 333                               timespec_load(&st->st_mtim),
 334                               ret);
 335                 if (r < 0)
 336                         return r;
 337
 338                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 339                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 340
 341                 return 0;
 342
 343         } else if (S_ISBLK(st->st_mode)) {
 344                 _cleanup_close_ int block_fd = -1;
 345                 uint64_t size = UINT64_MAX;
 346
 347                 /* A block device */
 348
 349                 if (!ret)
 350                         return 0;
 351
 352                 if (!pretty) {
 353                         r = extract_pretty(filename, NULL, &pretty_buffer);
 354                         if (r < 0)
 355                                 return r;
 356
 357                         pretty = pretty_buffer;
 358                 }
 359
 360                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 361                 if (block_fd < 0)
 362                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path, filename);
 363                 else {
 364                         /* Refresh stat data after opening the node */
 365                         if (fstat(block_fd, &stbuf) < 0)
 366                                 return -errno;
 367                         st = &stbuf;
 368
 369                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 370                                 return -ENOTTY;
 371
 372                         if (!read_only) {
 373                                 int state = 0;
 374
 375                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 376                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path, filename);
 377                                 else if (state)
 378                                         read_only = true;
 379                         }
 380
 381                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 382                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path, filename);
 383
 384                         block_fd = safe_close(block_fd);
 385                 }
 386
 387                 r = image_new(IMAGE_BLOCK,
 388                               pretty,
 389                               path,
 390                               filename,
 391                               !(st->st_mode & 0222) || read_only,
 392                               0,
 393                               0,
 394                               ret);
 395                 if (r < 0)
 396                         return r;
 397
 398                 if (size != 0 && size != UINT64_MAX)
 399                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 400
 401                 return 0;
 402         }
 403
 404         return -EMEDIUMTYPE;
 405 }
 406
 407 int image_find(ImageClass class, const char *name, Image **ret) {
 408         const char *path;
 409         int r;
 410
 411         assert(class >= 0);
 412         assert(class < _IMAGE_CLASS_MAX);
 413         assert(name);
 414
 415         /* There are no images with invalid names */
 416         if (!image_name_is_valid(name))
 417                 return -ENOENT;
 418
 419         NULSTR_FOREACH(path, image_search_path[class]) {
 420                 _cleanup_closedir_ DIR *d = NULL;
 421                 struct stat st;
 422
 423                 d = opendir(path);
 424                 if (!d) {
 425                         if (errno == ENOENT)
 426                                 continue;
 427
 428                         return -errno;
 429                 }
 430
 431                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to
 432                  * symlink block devices into the search path */
 433                 if (fstatat(dirfd(d), name, &st, 0) < 0) {
 434                         _cleanup_free_ char *raw = NULL;
 435
 436                         if (errno != ENOENT)
 437                                 return -errno;
 438
 439                         raw = strappend(name, ".raw");
 440                         if (!raw)
 441                                 return -ENOMEM;
 442
 443                         if (fstatat(dirfd(d), raw, &st, 0) < 0) {
 444
 445                                 if (errno == ENOENT)
 446                                         continue;
 447
 448                                 return -errno;
 449                         }
 450
 451                         if (!S_ISREG(st.st_mode))
 452                                 continue;
 453
 454                         r = image_make(name, dirfd(d), path, raw, &st, ret);
 455
 456                 } else {
 457                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 458                                 continue;
 459
 460                         r = image_make(name, dirfd(d), path, name, &st, ret);
 461                 }
 462                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 463                         continue;
 464                 if (r < 0)
 465                         return r;
 466
 467                 if (ret)
 468                         (*ret)->discoverable = true;
 469
 470                 return 1;
 471         }
 472
 473         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 474                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 475                 if (r < 0)
 476                         return r;
 477
 478                 if (ret)
 479                         (*ret)->discoverable = true;
 480
 481                 return r;
 482         }
 483
 484         return -ENOENT;
 485 };
 486
 487 int image_from_path(const char *path, Image **ret) {
 488
 489         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 490          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 491          * overridden by another, different image earlier in the search path */
 492
 493         if (path_equal(path, "/"))
 494                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 495
 496         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 497 }
 498
 499 int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) {
 500         if (image_name_is_valid(name_or_path))
 501                 return image_find(class, name_or_path, ret);
 502
 503         return image_from_path(name_or_path, ret);
 504 }
 505
 506 int image_discover(ImageClass class, Hashmap *h) {
 507         const char *path;
 508         int r;
 509
 510         assert(class >= 0);
 511         assert(class < _IMAGE_CLASS_MAX);
 512         assert(h);
 513
 514         NULSTR_FOREACH(path, image_search_path[class]) {
 515                 _cleanup_closedir_ DIR *d = NULL;
 516                 struct dirent *de;
 517
 518                 d = opendir(path);
 519                 if (!d) {
 520                         if (errno == ENOENT)
 521                                 continue;
 522
 523                         return -errno;
 524                 }
 525
 526                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 527                         _cleanup_(image_unrefp) Image *image = NULL;
 528                         _cleanup_free_ char *truncated = NULL;
 529                         const char *pretty;
 530                         struct stat st;
 531
 532                         if (dot_or_dot_dot(de->d_name))
 533                                 continue;
 534
 535                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 536                          * to symlink block devices into the search path */
 537                         if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) {
 538                                 if (errno == ENOENT)
 539                                         continue;
 540
 541                                 return -errno;
 542                         }
 543
 544                         if (S_ISREG(st.st_mode)) {
 545                                 const char *e;
 546
 547                                 e = endswith(de->d_name, ".raw");
 548                                 if (!e)
 549                                         continue;
 550
 551                                 truncated = strndup(de->d_name, e - de->d_name);
 552                                 if (!truncated)
 553                                         return -ENOMEM;
 554
 555                                 pretty = truncated;
 556                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 557                                 pretty = de->d_name;
 558                         else
 559                                 continue;
 560
 561                         if (!image_name_is_valid(pretty))
 562                                 continue;
 563
 564                         if (hashmap_contains(h, pretty))
 565                                 continue;
 566
 567                         r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image);
 568                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 569                                 continue;
 570                         if (r < 0)
 571                                 return r;
 572
 573                         image->discoverable = true;
 574
 575                         r = hashmap_put(h, image->name, image);
 576                         if (r < 0)
 577                                 return r;
 578
 579                         image = NULL;
 580                 }
 581         }
 582
 583         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 584                 _cleanup_(image_unrefp) Image *image = NULL;
 585
 586                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image);
 587                 if (r < 0)
 588                         return r;
 589
 590                 image->discoverable = true;
 591
 592                 r = hashmap_put(h, image->name, image);
 593                 if (r < 0)
 594                         return r;
 595
 596                 image = NULL;
 597         }
 598
 599         return 0;
 600 }
 601
 602 int image_remove(Image *i) {
 603         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 604         _cleanup_strv_free_ char **settings = NULL;
 605         _cleanup_free_ char *roothash = NULL;
 606         char **j;
 607         int r;
 608
 609         assert(i);
 610
 611         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 612                 return -EROFS;
 613
 614         settings = image_settings_path(i);
 615         if (!settings)
 616                 return -ENOMEM;
 617
 618         roothash = image_roothash_path(i);
 619         if (!roothash)
 620                 return -ENOMEM;
 621
 622         /* Make sure we don't interfere with a running nspawn */
 623         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 624         if (r < 0)
 625                 return r;
 626
 627         switch (i->type) {
 628
 629         case IMAGE_SUBVOLUME:
 630
 631                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 632                  * big guns */
 633                 if (unlink(i->path) < 0) {
 634                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 635                         if (r < 0)
 636                                 return r;
 637                 }
 638
 639                 break;
 640
 641         case IMAGE_DIRECTORY:
 642                 /* Allow deletion of read-only directories */
 643                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 644                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 645                 if (r < 0)
 646                         return r;
 647
 648                 break;
 649
 650         case IMAGE_BLOCK:
 651
 652                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 653                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 654                  * the thing (it's most likely a symlink after all). */
 655
 656                 if (path_startswith(i->path, "/dev"))
 657                         break;
 658
 659                 _fallthrough_;
 660         case IMAGE_RAW:
 661                 if (unlink(i->path) < 0)
 662                         return -errno;
 663                 break;
 664
 665         default:
 666                 return -EOPNOTSUPP;
 667         }
 668
 669         STRV_FOREACH(j, settings) {
 670                 if (unlink(*j) < 0 && errno != ENOENT)
 671                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 672         }
 673
 674         if (unlink(roothash) < 0 && errno != ENOENT)
 675                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 676
 677         return 0;
 678 }
 679
 680 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 681         _cleanup_free_ char *rs = NULL;
 682         const char *fn;
 683
 684         fn = strjoina(new_name, suffix);
 685
 686         rs = file_in_same_dir(path, fn);
 687         if (!rs)
 688                 return -ENOMEM;
 689
 690         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 691 }
 692
 693 int image_rename(Image *i, const char *new_name) {
 694         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 695         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 696         _cleanup_strv_free_ char **settings = NULL;
 697         unsigned file_attr = 0;
 698         char **j;
 699         int r;
 700
 701         assert(i);
 702
 703         if (!image_name_is_valid(new_name))
 704                 return -EINVAL;
 705
 706         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 707                 return -EROFS;
 708
 709         settings = image_settings_path(i);
 710         if (!settings)
 711                 return -ENOMEM;
 712
 713         roothash = image_roothash_path(i);
 714         if (!roothash)
 715                 return -ENOMEM;
 716
 717         /* Make sure we don't interfere with a running nspawn */
 718         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 719         if (r < 0)
 720                 return r;
 721
 722         /* Make sure nobody takes the new name, between the time we
 723          * checked it is currently unused in all search paths, and the
 724          * time we take possession of it */
 725         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 726         if (r < 0)
 727                 return r;
 728
 729         r = image_find(IMAGE_MACHINE, new_name, NULL);
 730         if (r >= 0)
 731                 return -EEXIST;
 732         if (r != -ENOENT)
 733                 return r;
 734
 735         switch (i->type) {
 736
 737         case IMAGE_DIRECTORY:
 738                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 739                 (void) read_attr_path(i->path, &file_attr);
 740
 741                 if (file_attr & FS_IMMUTABLE_FL)
 742                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 743
 744                 _fallthrough_;
 745         case IMAGE_SUBVOLUME:
 746                 new_path = file_in_same_dir(i->path, new_name);
 747                 break;
 748
 749         case IMAGE_BLOCK:
 750
 751                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 752                 if (path_startswith(i->path, "/dev"))
 753                         return -EROFS;
 754
 755                 new_path = file_in_same_dir(i->path, new_name);
 756                 break;
 757
 758         case IMAGE_RAW: {
 759                 const char *fn;
 760
 761                 fn = strjoina(new_name, ".raw");
 762                 new_path = file_in_same_dir(i->path, fn);
 763                 break;
 764         }
 765
 766         default:
 767                 return -EOPNOTSUPP;
 768         }
 769
 770         if (!new_path)
 771                 return -ENOMEM;
 772
 773         nn = strdup(new_name);
 774         if (!nn)
 775                 return -ENOMEM;
 776
 777         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 778         if (r < 0)
 779                 return r;
 780
 781         /* Restore the immutable bit, if it was set before */
 782         if (file_attr & FS_IMMUTABLE_FL)
 783                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 784
 785         free_and_replace(i->path, new_path);
 786         free_and_replace(i->name, nn);
 787
 788         STRV_FOREACH(j, settings) {
 789                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 790                 if (r < 0 && r != -ENOENT)
 791                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 792         }
 793
 794         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 795         if (r < 0 && r != -ENOENT)
 796                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 797
 798         return 0;
 799 }
 800
 801 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 802         _cleanup_free_ char *rs = NULL;
 803         const char *fn;
 804
 805         fn = strjoina(new_name, suffix);
 806
 807         rs = file_in_same_dir(path, fn);
 808         if (!rs)
 809                 return -ENOMEM;
 810
 811         return copy_file_atomic(path, rs, 0664, 0, COPY_REFLINK);
 812 }
 813
 814 int image_clone(Image *i, const char *new_name, bool read_only) {
 815         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 816         _cleanup_strv_free_ char **settings = NULL;
 817         _cleanup_free_ char *roothash = NULL;
 818         const char *new_path;
 819         char **j;
 820         int r;
 821
 822         assert(i);
 823
 824         if (!image_name_is_valid(new_name))
 825                 return -EINVAL;
 826
 827         settings = image_settings_path(i);
 828         if (!settings)
 829                 return -ENOMEM;
 830
 831         roothash = image_roothash_path(i);
 832         if (!roothash)
 833                 return -ENOMEM;
 834
 835         /* Make sure nobody takes the new name, between the time we
 836          * checked it is currently unused in all search paths, and the
 837          * time we take possession of it */
 838         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 839         if (r < 0)
 840                 return r;
 841
 842         r = image_find(IMAGE_MACHINE, new_name, NULL);
 843         if (r >= 0)
 844                 return -EEXIST;
 845         if (r != -ENOENT)
 846                 return r;
 847
 848         switch (i->type) {
 849
 850         case IMAGE_SUBVOLUME:
 851         case IMAGE_DIRECTORY:
 852                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 853                  * directory. */
 854
 855                 new_path = strjoina("/var/lib/machines/", new_name);
 856
 857                 r = btrfs_subvol_snapshot(i->path, new_path,
 858                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 859                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 860                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 861                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 862                                           BTRFS_SNAPSHOT_RECURSIVE |
 863                                           BTRFS_SNAPSHOT_QUOTA);
 864                 if (r >= 0)
 865                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 866                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 867
 868                 break;
 869
 870         case IMAGE_RAW:
 871                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 872
 873                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME);
 874                 break;
 875
 876         case IMAGE_BLOCK:
 877         default:
 878                 return -EOPNOTSUPP;
 879         }
 880
 881         if (r < 0)
 882                 return r;
 883
 884         STRV_FOREACH(j, settings) {
 885                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 886                 if (r < 0 && r != -ENOENT)
 887                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 888         }
 889
 890         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 891         if (r < 0 && r != -ENOENT)
 892                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 893
 894         return 0;
 895 }
 896
 897 int image_read_only(Image *i, bool b) {
 898         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 899         int r;
 900
 901         assert(i);
 902
 903         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 904                 return -EROFS;
 905
 906         /* Make sure we don't interfere with a running nspawn */
 907         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 908         if (r < 0)
 909                 return r;
 910
 911         switch (i->type) {
 912
 913         case IMAGE_SUBVOLUME:
 914
 915                 /* Note that we set the flag only on the top-level
 916                  * subvolume of the image. */
 917
 918                 r = btrfs_subvol_set_read_only(i->path, b);
 919                 if (r < 0)
 920                         return r;
 921
 922                 break;
 923
 924         case IMAGE_DIRECTORY:
 925                 /* For simple directory trees we cannot use the access
 926                    mode of the top-level directory, since it has an
 927                    effect on the container itself.  However, we can
 928                    use the "immutable" flag, to at least make the
 929                    top-level directory read-only. It's not as good as
 930                    a read-only subvolume, but at least something, and
 931                    we can read the value back. */
 932
 933                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 934                 if (r < 0)
 935                         return r;
 936
 937                 break;
 938
 939         case IMAGE_RAW: {
 940                 struct stat st;
 941
 942                 if (stat(i->path, &st) < 0)
 943                         return -errno;
 944
 945                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 946                         return -errno;
 947
 948                 /* If the images is now read-only, it's a good time to
 949                  * defrag it, given that no write patterns will
 950                  * fragment it again. */
 951                 if (b)
 952                         (void) btrfs_defrag(i->path);
 953                 break;
 954         }
 955
 956         case IMAGE_BLOCK: {
 957                 _cleanup_close_ int fd = -1;
 958                 struct stat st;
 959                 int state = b;
 960
 961                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 962                 if (fd < 0)
 963                         return -errno;
 964
 965                 if (fstat(fd, &st) < 0)
 966                         return -errno;
 967                 if (!S_ISBLK(st.st_mode))
 968                         return -ENOTTY;
 969
 970                 if (ioctl(fd, BLKROSET, &state) < 0)
 971                         return -errno;
 972
 973                 break;
 974         }
 975
 976         default:
 977                 return -EOPNOTSUPP;
 978         }
 979
 980         return 0;
 981 }
 982
 983 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
 984         _cleanup_free_ char *p = NULL;
 985         LockFile t = LOCK_FILE_INIT;
 986         struct stat st;
 987         int r;
 988
 989         assert(path);
 990         assert(global);
 991         assert(local);
 992
 993         /* Locks an image path. This actually creates two locks: one
 994          * "local" one, next to the image path itself, which might be
 995          * shared via NFS. And another "global" one, in /run, that
 996          * uses the device/inode number. This has the benefit that we
 997          * can even lock a tree that is a mount point, correctly. */
 998
 999         if (!path_is_absolute(path))
1000                 return -EINVAL;
1001
1002         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1003                 *local = *global = (LockFile) LOCK_FILE_INIT;
1004                 return 0;
1005         }
1006
1007         if (path_equal(path, "/"))
1008                 return -EBUSY;
1009
1010         if (stat(path, &st) >= 0) {
1011                 if (S_ISBLK(st.st_mode))
1012                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1013                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1014                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1015                 else
1016                         return -ENOTTY;
1017
1018                 if (r < 0)
1019                         return -ENOMEM;
1020         }
1021
1022         /* For block devices we don't need the "local" lock, as the major/minor lock above should be sufficient, since
1023          * block devices are device local anyway. */
1024         if (!path_startswith(path, "/dev")) {
1025                 r = make_lock_file_for(path, operation, &t);
1026                 if (r < 0) {
1027                         if ((operation & LOCK_SH) && r == -EROFS)
1028                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1029                         else
1030                                 return r;
1031                 }
1032         }
1033
1034         if (p) {
1035                 mkdir_p("/run/systemd/nspawn/locks", 0700);
1036
1037                 r = make_lock_file(p, operation, global);
1038                 if (r < 0) {
1039                         release_lock_file(&t);
1040                         return r;
1041                 }
1042         } else
1043                 *global = (LockFile) LOCK_FILE_INIT;
1044
1045         *local = t;
1046         return 0;
1047 }
1048
1049 int image_set_limit(Image *i, uint64_t referenced_max) {
1050         assert(i);
1051
1052         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1053                 return -EROFS;
1054
1055         if (i->type != IMAGE_SUBVOLUME)
1056                 return -EOPNOTSUPP;
1057
1058         /* We set the quota both for the subvolume as well as for the
1059          * subtree. The latter is mostly for historical reasons, since
1060          * we didn't use to have a concept of subtree quota, and hence
1061          * only modified the subvolume quota. */
1062
1063         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1064         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1065         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1066 }
1067
1068 int image_read_metadata(Image *i) {
1069         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1070         int r;
1071
1072         assert(i);
1073
1074         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1075         if (r < 0)
1076                 return r;
1077
1078         switch (i->type) {
1079
1080         case IMAGE_SUBVOLUME:
1081         case IMAGE_DIRECTORY: {
1082                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1083                 sd_id128_t machine_id = SD_ID128_NULL;
1084                 _cleanup_free_ char *hostname = NULL;
1085                 _cleanup_free_ char *path = NULL;
1086
1087                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1088                 if (r < 0 && r != -ENOENT)
1089                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1090                 else if (r >= 0) {
1091                         r = read_etc_hostname(path, &hostname);
1092                         if (r < 0)
1093                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1094                 }
1095
1096                 path = mfree(path);
1097
1098                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1099                 if (r < 0 && r != -ENOENT)
1100                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1101                 else if (r >= 0) {
1102                         _cleanup_close_ int fd = -1;
1103
1104                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1105                         if (fd < 0)
1106                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1107                         else {
1108                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1109                                 if (r < 0)
1110                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1111                         }
1112                 }
1113
1114                 path = mfree(path);
1115
1116                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1117                 if (r < 0 && r != -ENOENT)
1118                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1119                 else if (r >= 0) {
1120                         r = load_env_file_pairs(NULL, path, &machine_info);
1121                         if (r < 0)
1122                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1123                 }
1124
1125                 r = load_os_release_pairs(i->path, &os_release);
1126                 if (r < 0)
1127                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1128
1129                 free_and_replace(i->hostname, hostname);
1130                 i->machine_id = machine_id;
1131                 strv_free_and_replace(i->machine_info, machine_info);
1132                 strv_free_and_replace(i->os_release, os_release);
1133
1134                 break;
1135         }
1136
1137         case IMAGE_RAW:
1138         case IMAGE_BLOCK: {
1139                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1140                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1141
1142                 r = loop_device_make_by_path(i->path, O_RDONLY, &d);
1143                 if (r < 0)
1144                         return r;
1145
1146                 r = dissect_image(d->fd, NULL, 0, DISSECT_IMAGE_REQUIRE_ROOT, &m);
1147                 if (r < 0)
1148                         return r;
1149
1150                 r = dissected_image_acquire_metadata(m);
1151                 if (r < 0)
1152                         return r;
1153
1154                 free_and_replace(i->hostname, m->hostname);
1155                 i->machine_id = m->machine_id;
1156                 strv_free_and_replace(i->machine_info, m->machine_info);
1157                 strv_free_and_replace(i->os_release, m->os_release);
1158
1159                 break;
1160         }
1161
1162         default:
1163                 return -EOPNOTSUPP;
1164         }
1165
1166         i->metadata_valid = true;
1167
1168         return 0;
1169 }
1170
1171 int image_name_lock(const char *name, int operation, LockFile *ret) {
1172         const char *p;
1173
1174         assert(name);
1175         assert(ret);
1176
1177         /* Locks an image name, regardless of the precise path used. */
1178
1179         if (!image_name_is_valid(name))
1180                 return -EINVAL;
1181
1182         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1183                 *ret = (LockFile) LOCK_FILE_INIT;
1184                 return 0;
1185         }
1186
1187         if (streq(name, ".host"))
1188                 return -EBUSY;
1189
1190         mkdir_p("/run/systemd/nspawn/locks", 0700);
1191         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1192
1193         return make_lock_file(p, operation, ret);
1194 }
1195
1196 bool image_name_is_valid(const char *s) {
1197         if (!filename_is_valid(s))
1198                 return false;
1199
1200         if (string_has_cc(s, NULL))
1201                 return false;
1202
1203         if (!utf8_is_valid(s))
1204                 return false;
1205
1206         /* Temporary files for atomically creating new files */
1207         if (startswith(s, ".#"))
1208                 return false;
1209
1210         return true;
1211 }
1212
1213 bool image_in_search_path(ImageClass class, const char *image) {
1214         const char *path;
1215
1216         assert(image);
1217
1218         NULSTR_FOREACH(path, image_search_path[class]) {
1219                 const char *p;
1220                 size_t k;
1221
1222                 p = path_startswith(image, path);
1223                 if (!p)
1224                         continue;
1225
1226                 /* Make sure there's a filename following */
1227                 k = strcspn(p, "/");
1228                 if (k == 0)
1229                         continue;
1230
1231                 p += k;
1232
1233                 /* Accept trailing slashes */
1234                 if (p[strspn(p, "/")] == 0)
1235                         return true;
1236
1237         }
1238
1239         return false;
1240 }
1241
1242 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1243         [IMAGE_DIRECTORY] = "directory",
1244         [IMAGE_SUBVOLUME] = "subvolume",
1245         [IMAGE_RAW] = "raw",
1246         [IMAGE_BLOCK] = "block",
1247 };
1248
1249 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);