src/shared/machine-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   Copyright 2013 Lennart Poettering
   4 ***/
   5
   6 #include <dirent.h>
   7 #include <errno.h>
   8 #include <fcntl.h>
   9 #include <stdio.h>
  10 #include <stdlib.h>
  11 #include <string.h>
  12 #include <sys/file.h>
  13 #include <sys/stat.h>
  14 #include <unistd.h>
  15 #include <linux/fs.h>
  16
  17 #include "alloc-util.h"
  18 #include "btrfs-util.h"
  19 #include "chattr-util.h"
  20 #include "copy.h"
  21 #include "dirent-util.h"
  22 #include "dissect-image.h"
  23 #include "env-util.h"
  24 #include "fd-util.h"
  25 #include "fileio.h"
  26 #include "fs-util.h"
  27 #include "hashmap.h"
  28 #include "hostname-util.h"
  29 #include "id128-util.h"
  30 #include "lockfile-util.h"
  31 #include "log.h"
  32 #include "loop-util.h"
  33 #include "machine-image.h"
  34 #include "macro.h"
  35 #include "mkdir.h"
  36 #include "os-util.h"
  37 #include "path-util.h"
  38 #include "rm-rf.h"
  39 #include "string-table.h"
  40 #include "string-util.h"
  41 #include "strv.h"
  42 #include "time-util.h"
  43 #include "utf8.h"
  44 #include "util.h"
  45 #include "xattr-util.h"
  46
  47 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  48         [IMAGE_MACHINE] =  "/etc/machines\0"              /* only place symlinks here */
  49                            "/run/machines\0"              /* and here too */
  50                            "/var/lib/machines\0"          /* the main place for images */
  51                            "/var/lib/container\0"         /* legacy */
  52                            "/usr/local/lib/machines\0"
  53                            "/usr/lib/machines\0",
  54
  55         [IMAGE_PORTABLE] = "/etc/portables\0"             /* only place symlinks here */
  56                            "/run/portables\0"             /* and here too */
  57                            "/var/lib/portables\0"         /* the main place for images */
  58                            "/usr/local/lib/portables\0"
  59                            "/usr/lib/portables\0",
  60 };
  61
  62 Image *image_unref(Image *i) {
  63         if (!i)
  64                 return NULL;
  65
  66         assert(i->n_ref > 0);
  67         i->n_ref--;
  68
  69         if (i->n_ref > 0)
  70                 return NULL;
  71
  72         free(i->name);
  73         free(i->path);
  74
  75         free(i->hostname);
  76         strv_free(i->machine_info);
  77         strv_free(i->os_release);
  78
  79         return mfree(i);
  80 }
  81
  82 Image *image_ref(Image *i) {
  83         if (!i)
  84                 return NULL;
  85
  86         assert(i->n_ref > 0);
  87         i->n_ref++;
  88
  89         return i;
  90 }
  91
  92 static char **image_settings_path(Image *image) {
  93         _cleanup_strv_free_ char **l = NULL;
  94         const char *fn, *s;
  95         unsigned i = 0;
  96
  97         assert(image);
  98
  99         l = new0(char*, 4);
 100         if (!l)
 101                 return NULL;
 102
 103         fn = strjoina(image->name, ".nspawn");
 104
 105         FOREACH_STRING(s, "/etc/systemd/nspawn/", "/run/systemd/nspawn/") {
 106                 l[i] = strappend(s, fn);
 107                 if (!l[i])
 108                         return NULL;
 109
 110                 i++;
 111         }
 112
 113         l[i] = file_in_same_dir(image->path, fn);
 114         if (!l[i])
 115                 return NULL;
 116
 117         return TAKE_PTR(l);
 118 }
 119
 120 static char *image_roothash_path(Image *image) {
 121         const char *fn;
 122
 123         assert(image);
 124
 125         fn = strjoina(image->name, ".roothash");
 126
 127         return file_in_same_dir(image->path, fn);
 128 }
 129
 130 static int image_new(
 131                 ImageType t,
 132                 const char *pretty,
 133                 const char *path,
 134                 const char *filename,
 135                 bool read_only,
 136                 usec_t crtime,
 137                 usec_t mtime,
 138                 Image **ret) {
 139
 140         _cleanup_(image_unrefp) Image *i = NULL;
 141
 142         assert(t >= 0);
 143         assert(t < _IMAGE_TYPE_MAX);
 144         assert(pretty);
 145         assert(filename);
 146         assert(ret);
 147
 148         i = new0(Image, 1);
 149         if (!i)
 150                 return -ENOMEM;
 151
 152         i->n_ref = 1;
 153         i->type = t;
 154         i->read_only = read_only;
 155         i->crtime = crtime;
 156         i->mtime = mtime;
 157         i->usage = i->usage_exclusive = (uint64_t) -1;
 158         i->limit = i->limit_exclusive = (uint64_t) -1;
 159
 160         i->name = strdup(pretty);
 161         if (!i->name)
 162                 return -ENOMEM;
 163
 164         if (path)
 165                 i->path = strjoin(path, "/", filename);
 166         else
 167                 i->path = strdup(filename);
 168         if (!i->path)
 169                 return -ENOMEM;
 170
 171         path_simplify(i->path, false);
 172
 173         *ret = TAKE_PTR(i);
 174
 175         return 0;
 176 }
 177
 178 static int extract_pretty(const char *path, const char *suffix, char **ret) {
 179         _cleanup_free_ char *name = NULL;
 180         const char *p;
 181         size_t n;
 182
 183         assert(path);
 184         assert(ret);
 185
 186         p = last_path_component(path);
 187         n = strcspn(p, "/");
 188
 189         name = strndup(p, n);
 190         if (!name)
 191                 return -ENOMEM;
 192
 193         if (suffix) {
 194                 char *e;
 195
 196                 e = endswith(name, suffix);
 197                 if (!e)
 198                         return -EINVAL;
 199
 200                 *e = 0;
 201         }
 202
 203         if (!image_name_is_valid(name))
 204                 return -EINVAL;
 205
 206         *ret = TAKE_PTR(name);
 207         return 0;
 208 }
 209
 210 static int image_make(
 211                 const char *pretty,
 212                 int dfd,
 213                 const char *path,
 214                 const char *filename,
 215                 const struct stat *st,
 216                 Image **ret) {
 217
 218         _cleanup_free_ char *pretty_buffer = NULL;
 219         struct stat stbuf;
 220         bool read_only;
 221         int r;
 222
 223         assert(dfd >= 0 || dfd == AT_FDCWD);
 224         assert(filename);
 225
 226         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 227          * devices into /var/lib/machines/, and treat them normally.
 228          *
 229          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 230          * recognize. */
 231
 232         if (!st) {
 233                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 234                         return -errno;
 235
 236                 st = &stbuf;
 237         }
 238
 239         read_only =
 240                 (path && path_startswith(path, "/usr")) ||
 241                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 242
 243         if (S_ISDIR(st->st_mode)) {
 244                 _cleanup_close_ int fd = -1;
 245                 unsigned file_attr = 0;
 246
 247                 if (!ret)
 248                         return 0;
 249
 250                 if (!pretty) {
 251                         r = extract_pretty(filename, NULL, &pretty_buffer);
 252                         if (r < 0)
 253                                 return r;
 254
 255                         pretty = pretty_buffer;
 256                 }
 257
 258                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 259                 if (fd < 0)
 260                         return -errno;
 261
 262                 /* btrfs subvolumes have inode 256 */
 263                 if (st->st_ino == 256) {
 264
 265                         r = btrfs_is_filesystem(fd);
 266                         if (r < 0)
 267                                 return r;
 268                         if (r) {
 269                                 BtrfsSubvolInfo info;
 270
 271                                 /* It's a btrfs subvolume */
 272
 273                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 274                                 if (r < 0)
 275                                         return r;
 276
 277                                 r = image_new(IMAGE_SUBVOLUME,
 278                                               pretty,
 279                                               path,
 280                                               filename,
 281                                               info.read_only || read_only,
 282                                               info.otime,
 283                                               0,
 284                                               ret);
 285                                 if (r < 0)
 286                                         return r;
 287
 288                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 289                                         BtrfsQuotaInfo quota;
 290
 291                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 292                                         if (r >= 0) {
 293                                                 (*ret)->usage = quota.referenced;
 294                                                 (*ret)->usage_exclusive = quota.exclusive;
 295
 296                                                 (*ret)->limit = quota.referenced_max;
 297                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 298                                         }
 299                                 }
 300
 301                                 return 0;
 302                         }
 303                 }
 304
 305                 /* If the IMMUTABLE bit is set, we consider the
 306                  * directory read-only. Since the ioctl is not
 307                  * supported everywhere we ignore failures. */
 308                 (void) read_attr_fd(fd, &file_attr);
 309
 310                 /* It's just a normal directory. */
 311                 r = image_new(IMAGE_DIRECTORY,
 312                               pretty,
 313                               path,
 314                               filename,
 315                               read_only || (file_attr & FS_IMMUTABLE_FL),
 316                               0,
 317                               0,
 318                               ret);
 319                 if (r < 0)
 320                         return r;
 321
 322                 return 0;
 323
 324         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 325                 usec_t crtime = 0;
 326
 327                 /* It's a RAW disk image */
 328
 329                 if (!ret)
 330                         return 0;
 331
 332                 (void) fd_getcrtime_at(dfd, filename, &crtime, 0);
 333
 334                 if (!pretty) {
 335                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 336                         if (r < 0)
 337                                 return r;
 338
 339                         pretty = pretty_buffer;
 340                 }
 341
 342                 r = image_new(IMAGE_RAW,
 343                               pretty,
 344                               path,
 345                               filename,
 346                               !(st->st_mode & 0222) || read_only,
 347                               crtime,
 348                               timespec_load(&st->st_mtim),
 349                               ret);
 350                 if (r < 0)
 351                         return r;
 352
 353                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 354                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 355
 356                 return 0;
 357
 358         } else if (S_ISBLK(st->st_mode)) {
 359                 _cleanup_close_ int block_fd = -1;
 360                 uint64_t size = UINT64_MAX;
 361
 362                 /* A block device */
 363
 364                 if (!ret)
 365                         return 0;
 366
 367                 if (!pretty) {
 368                         r = extract_pretty(filename, NULL, &pretty_buffer);
 369                         if (r < 0)
 370                                 return r;
 371
 372                         pretty = pretty_buffer;
 373                 }
 374
 375                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 376                 if (block_fd < 0)
 377                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path, filename);
 378                 else {
 379                         /* Refresh stat data after opening the node */
 380                         if (fstat(block_fd, &stbuf) < 0)
 381                                 return -errno;
 382                         st = &stbuf;
 383
 384                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 385                                 return -ENOTTY;
 386
 387                         if (!read_only) {
 388                                 int state = 0;
 389
 390                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 391                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path, filename);
 392                                 else if (state)
 393                                         read_only = true;
 394                         }
 395
 396                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 397                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path, filename);
 398
 399                         block_fd = safe_close(block_fd);
 400                 }
 401
 402                 r = image_new(IMAGE_BLOCK,
 403                               pretty,
 404                               path,
 405                               filename,
 406                               !(st->st_mode & 0222) || read_only,
 407                               0,
 408                               0,
 409                               ret);
 410                 if (r < 0)
 411                         return r;
 412
 413                 if (size != 0 && size != UINT64_MAX)
 414                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 415
 416                 return 0;
 417         }
 418
 419         return -EMEDIUMTYPE;
 420 }
 421
 422 int image_find(ImageClass class, const char *name, Image **ret) {
 423         const char *path;
 424         int r;
 425
 426         assert(class >= 0);
 427         assert(class < _IMAGE_CLASS_MAX);
 428         assert(name);
 429
 430         /* There are no images with invalid names */
 431         if (!image_name_is_valid(name))
 432                 return -ENOENT;
 433
 434         NULSTR_FOREACH(path, image_search_path[class]) {
 435                 _cleanup_closedir_ DIR *d = NULL;
 436                 struct stat st;
 437
 438                 d = opendir(path);
 439                 if (!d) {
 440                         if (errno == ENOENT)
 441                                 continue;
 442
 443                         return -errno;
 444                 }
 445
 446                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to
 447                  * symlink block devices into the search path */
 448                 if (fstatat(dirfd(d), name, &st, 0) < 0) {
 449                         _cleanup_free_ char *raw = NULL;
 450
 451                         if (errno != ENOENT)
 452                                 return -errno;
 453
 454                         raw = strappend(name, ".raw");
 455                         if (!raw)
 456                                 return -ENOMEM;
 457
 458                         if (fstatat(dirfd(d), raw, &st, 0) < 0) {
 459
 460                                 if (errno == ENOENT)
 461                                         continue;
 462
 463                                 return -errno;
 464                         }
 465
 466                         if (!S_ISREG(st.st_mode))
 467                                 continue;
 468
 469                         r = image_make(name, dirfd(d), path, raw, &st, ret);
 470
 471                 } else {
 472                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 473                                 continue;
 474
 475                         r = image_make(name, dirfd(d), path, name, &st, ret);
 476                 }
 477                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 478                         continue;
 479                 if (r < 0)
 480                         return r;
 481
 482                 if (ret)
 483                         (*ret)->discoverable = true;
 484
 485                 return 1;
 486         }
 487
 488         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 489                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 490                 if (r < 0)
 491                         return r;
 492
 493                 if (ret)
 494                         (*ret)->discoverable = true;
 495
 496                 return r;
 497         }
 498
 499         return -ENOENT;
 500 };
 501
 502 int image_from_path(const char *path, Image **ret) {
 503
 504         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 505          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 506          * overriden by another, different image earlier in the search path */
 507
 508         if (path_equal(path, "/"))
 509                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 510
 511         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 512 }
 513
 514 int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) {
 515         if (image_name_is_valid(name_or_path))
 516                 return image_find(class, name_or_path, ret);
 517
 518         return image_from_path(name_or_path, ret);
 519 }
 520
 521 int image_discover(ImageClass class, Hashmap *h) {
 522         const char *path;
 523         int r;
 524
 525         assert(class >= 0);
 526         assert(class < _IMAGE_CLASS_MAX);
 527         assert(h);
 528
 529         NULSTR_FOREACH(path, image_search_path[class]) {
 530                 _cleanup_closedir_ DIR *d = NULL;
 531                 struct dirent *de;
 532
 533                 d = opendir(path);
 534                 if (!d) {
 535                         if (errno == ENOENT)
 536                                 continue;
 537
 538                         return -errno;
 539                 }
 540
 541                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 542                         _cleanup_(image_unrefp) Image *image = NULL;
 543                         _cleanup_free_ char *truncated = NULL;
 544                         const char *pretty;
 545                         struct stat st;
 546
 547                         if (dot_or_dot_dot(de->d_name))
 548                                 continue;
 549
 550                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 551                          * to symlink block devices into the search path */
 552                         if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) {
 553                                 if (errno == ENOENT)
 554                                         continue;
 555
 556                                 return -errno;
 557                         }
 558
 559                         if (S_ISREG(st.st_mode)) {
 560                                 const char *e;
 561
 562                                 e = endswith(de->d_name, ".raw");
 563                                 if (!e)
 564                                         continue;
 565
 566                                 truncated = strndup(de->d_name, e - de->d_name);
 567                                 if (!truncated)
 568                                         return -ENOMEM;
 569
 570                                 pretty = truncated;
 571                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 572                                 pretty = de->d_name;
 573                         else
 574                                 continue;
 575
 576                         if (!image_name_is_valid(pretty))
 577                                 continue;
 578
 579                         if (hashmap_contains(h, pretty))
 580                                 continue;
 581
 582                         r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image);
 583                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 584                                 continue;
 585                         if (r < 0)
 586                                 return r;
 587
 588                         image->discoverable = true;
 589
 590                         r = hashmap_put(h, image->name, image);
 591                         if (r < 0)
 592                                 return r;
 593
 594                         image = NULL;
 595                 }
 596         }
 597
 598         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 599                 _cleanup_(image_unrefp) Image *image = NULL;
 600
 601                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image);
 602                 if (r < 0)
 603                         return r;
 604
 605                 image->discoverable = true;
 606
 607                 r = hashmap_put(h, image->name, image);
 608                 if (r < 0)
 609                         return r;
 610
 611                 image = NULL;
 612         }
 613
 614         return 0;
 615 }
 616
 617 int image_remove(Image *i) {
 618         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 619         _cleanup_strv_free_ char **settings = NULL;
 620         _cleanup_free_ char *roothash = NULL;
 621         char **j;
 622         int r;
 623
 624         assert(i);
 625
 626         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 627                 return -EROFS;
 628
 629         settings = image_settings_path(i);
 630         if (!settings)
 631                 return -ENOMEM;
 632
 633         roothash = image_roothash_path(i);
 634         if (!roothash)
 635                 return -ENOMEM;
 636
 637         /* Make sure we don't interfere with a running nspawn */
 638         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 639         if (r < 0)
 640                 return r;
 641
 642         switch (i->type) {
 643
 644         case IMAGE_SUBVOLUME:
 645
 646                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 647                  * big guns */
 648                 if (unlink(i->path) < 0) {
 649                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 650                         if (r < 0)
 651                                 return r;
 652                 }
 653
 654                 break;
 655
 656         case IMAGE_DIRECTORY:
 657                 /* Allow deletion of read-only directories */
 658                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL);
 659                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 660                 if (r < 0)
 661                         return r;
 662
 663                 break;
 664
 665         case IMAGE_BLOCK:
 666
 667                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 668                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 669                  * the thing (it's most likely a symlink after all). */
 670
 671                 if (path_startswith(i->path, "/dev"))
 672                         break;
 673
 674                 _fallthrough_;
 675         case IMAGE_RAW:
 676                 if (unlink(i->path) < 0)
 677                         return -errno;
 678                 break;
 679
 680         default:
 681                 return -EOPNOTSUPP;
 682         }
 683
 684         STRV_FOREACH(j, settings) {
 685                 if (unlink(*j) < 0 && errno != ENOENT)
 686                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 687         }
 688
 689         if (unlink(roothash) < 0 && errno != ENOENT)
 690                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 691
 692         return 0;
 693 }
 694
 695 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 696         _cleanup_free_ char *rs = NULL;
 697         const char *fn;
 698
 699         fn = strjoina(new_name, suffix);
 700
 701         rs = file_in_same_dir(path, fn);
 702         if (!rs)
 703                 return -ENOMEM;
 704
 705         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 706 }
 707
 708 int image_rename(Image *i, const char *new_name) {
 709         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 710         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 711         _cleanup_strv_free_ char **settings = NULL;
 712         unsigned file_attr = 0;
 713         char **j;
 714         int r;
 715
 716         assert(i);
 717
 718         if (!image_name_is_valid(new_name))
 719                 return -EINVAL;
 720
 721         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 722                 return -EROFS;
 723
 724         settings = image_settings_path(i);
 725         if (!settings)
 726                 return -ENOMEM;
 727
 728         roothash = image_roothash_path(i);
 729         if (!roothash)
 730                 return -ENOMEM;
 731
 732         /* Make sure we don't interfere with a running nspawn */
 733         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 734         if (r < 0)
 735                 return r;
 736
 737         /* Make sure nobody takes the new name, between the time we
 738          * checked it is currently unused in all search paths, and the
 739          * time we take possession of it */
 740         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 741         if (r < 0)
 742                 return r;
 743
 744         r = image_find(IMAGE_MACHINE, new_name, NULL);
 745         if (r >= 0)
 746                 return -EEXIST;
 747         if (r != -ENOENT)
 748                 return r;
 749
 750         switch (i->type) {
 751
 752         case IMAGE_DIRECTORY:
 753                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 754                 (void) read_attr_path(i->path, &file_attr);
 755
 756                 if (file_attr & FS_IMMUTABLE_FL)
 757                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL);
 758
 759                 _fallthrough_;
 760         case IMAGE_SUBVOLUME:
 761                 new_path = file_in_same_dir(i->path, new_name);
 762                 break;
 763
 764         case IMAGE_BLOCK:
 765
 766                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 767                 if (path_startswith(i->path, "/dev"))
 768                         return -EROFS;
 769
 770                 new_path = file_in_same_dir(i->path, new_name);
 771                 break;
 772
 773         case IMAGE_RAW: {
 774                 const char *fn;
 775
 776                 fn = strjoina(new_name, ".raw");
 777                 new_path = file_in_same_dir(i->path, fn);
 778                 break;
 779         }
 780
 781         default:
 782                 return -EOPNOTSUPP;
 783         }
 784
 785         if (!new_path)
 786                 return -ENOMEM;
 787
 788         nn = strdup(new_name);
 789         if (!nn)
 790                 return -ENOMEM;
 791
 792         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 793         if (r < 0)
 794                 return r;
 795
 796         /* Restore the immutable bit, if it was set before */
 797         if (file_attr & FS_IMMUTABLE_FL)
 798                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL);
 799
 800         free_and_replace(i->path, new_path);
 801         free_and_replace(i->name, nn);
 802
 803         STRV_FOREACH(j, settings) {
 804                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 805                 if (r < 0 && r != -ENOENT)
 806                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 807         }
 808
 809         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 810         if (r < 0 && r != -ENOENT)
 811                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 812
 813         return 0;
 814 }
 815
 816 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 817         _cleanup_free_ char *rs = NULL;
 818         const char *fn;
 819
 820         fn = strjoina(new_name, suffix);
 821
 822         rs = file_in_same_dir(path, fn);
 823         if (!rs)
 824                 return -ENOMEM;
 825
 826         return copy_file_atomic(path, rs, 0664, 0, COPY_REFLINK);
 827 }
 828
 829 int image_clone(Image *i, const char *new_name, bool read_only) {
 830         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 831         _cleanup_strv_free_ char **settings = NULL;
 832         _cleanup_free_ char *roothash = NULL;
 833         const char *new_path;
 834         char **j;
 835         int r;
 836
 837         assert(i);
 838
 839         if (!image_name_is_valid(new_name))
 840                 return -EINVAL;
 841
 842         settings = image_settings_path(i);
 843         if (!settings)
 844                 return -ENOMEM;
 845
 846         roothash = image_roothash_path(i);
 847         if (!roothash)
 848                 return -ENOMEM;
 849
 850         /* Make sure nobody takes the new name, between the time we
 851          * checked it is currently unused in all search paths, and the
 852          * time we take possession of it */
 853         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 854         if (r < 0)
 855                 return r;
 856
 857         r = image_find(IMAGE_MACHINE, new_name, NULL);
 858         if (r >= 0)
 859                 return -EEXIST;
 860         if (r != -ENOENT)
 861                 return r;
 862
 863         switch (i->type) {
 864
 865         case IMAGE_SUBVOLUME:
 866         case IMAGE_DIRECTORY:
 867                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 868                  * directory. */
 869
 870                 new_path = strjoina("/var/lib/machines/", new_name);
 871
 872                 r = btrfs_subvol_snapshot(i->path, new_path,
 873                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 874                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 875                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 876                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 877                                           BTRFS_SNAPSHOT_RECURSIVE |
 878                                           BTRFS_SNAPSHOT_QUOTA);
 879                 if (r >= 0)
 880                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 881                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 882
 883                 break;
 884
 885         case IMAGE_RAW:
 886                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 887
 888                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, COPY_REFLINK);
 889                 break;
 890
 891         case IMAGE_BLOCK:
 892         default:
 893                 return -EOPNOTSUPP;
 894         }
 895
 896         if (r < 0)
 897                 return r;
 898
 899         STRV_FOREACH(j, settings) {
 900                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 901                 if (r < 0 && r != -ENOENT)
 902                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 903         }
 904
 905         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 906         if (r < 0 && r != -ENOENT)
 907                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 908
 909         return 0;
 910 }
 911
 912 int image_read_only(Image *i, bool b) {
 913         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 914         int r;
 915
 916         assert(i);
 917
 918         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 919                 return -EROFS;
 920
 921         /* Make sure we don't interfere with a running nspawn */
 922         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 923         if (r < 0)
 924                 return r;
 925
 926         switch (i->type) {
 927
 928         case IMAGE_SUBVOLUME:
 929
 930                 /* Note that we set the flag only on the top-level
 931                  * subvolume of the image. */
 932
 933                 r = btrfs_subvol_set_read_only(i->path, b);
 934                 if (r < 0)
 935                         return r;
 936
 937                 break;
 938
 939         case IMAGE_DIRECTORY:
 940                 /* For simple directory trees we cannot use the access
 941                    mode of the top-level directory, since it has an
 942                    effect on the container itself.  However, we can
 943                    use the "immutable" flag, to at least make the
 944                    top-level directory read-only. It's not as good as
 945                    a read-only subvolume, but at least something, and
 946                    we can read the value back. */
 947
 948                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL);
 949                 if (r < 0)
 950                         return r;
 951
 952                 break;
 953
 954         case IMAGE_RAW: {
 955                 struct stat st;
 956
 957                 if (stat(i->path, &st) < 0)
 958                         return -errno;
 959
 960                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 961                         return -errno;
 962
 963                 /* If the images is now read-only, it's a good time to
 964                  * defrag it, given that no write patterns will
 965                  * fragment it again. */
 966                 if (b)
 967                         (void) btrfs_defrag(i->path);
 968                 break;
 969         }
 970
 971         case IMAGE_BLOCK: {
 972                 _cleanup_close_ int fd = -1;
 973                 struct stat st;
 974                 int state = b;
 975
 976                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 977                 if (fd < 0)
 978                         return -errno;
 979
 980                 if (fstat(fd, &st) < 0)
 981                         return -errno;
 982                 if (!S_ISBLK(st.st_mode))
 983                         return -ENOTTY;
 984
 985                 if (ioctl(fd, BLKROSET, &state) < 0)
 986                         return -errno;
 987
 988                 break;
 989         }
 990
 991         default:
 992                 return -EOPNOTSUPP;
 993         }
 994
 995         return 0;
 996 }
 997
 998 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
 999         _cleanup_free_ char *p = NULL;
1000         LockFile t = LOCK_FILE_INIT;
1001         struct stat st;
1002         int r;
1003
1004         assert(path);
1005         assert(global);
1006         assert(local);
1007
1008         /* Locks an image path. This actually creates two locks: one
1009          * "local" one, next to the image path itself, which might be
1010          * shared via NFS. And another "global" one, in /run, that
1011          * uses the device/inode number. This has the benefit that we
1012          * can even lock a tree that is a mount point, correctly. */
1013
1014         if (!path_is_absolute(path))
1015                 return -EINVAL;
1016
1017         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1018                 *local = *global = (LockFile) LOCK_FILE_INIT;
1019                 return 0;
1020         }
1021
1022         if (path_equal(path, "/"))
1023                 return -EBUSY;
1024
1025         if (stat(path, &st) >= 0) {
1026                 if (S_ISBLK(st.st_mode))
1027                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1028                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1029                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1030                 else
1031                         return -ENOTTY;
1032
1033                 if (r < 0)
1034                         return -ENOMEM;
1035         }
1036
1037         /* For block devices we don't need the "local" lock, as the major/minor lock above should be sufficient, since
1038          * block devices are device local anyway. */
1039         if (!path_startswith(path, "/dev")) {
1040                 r = make_lock_file_for(path, operation, &t);
1041                 if (r < 0) {
1042                         if ((operation & LOCK_SH) && r == -EROFS)
1043                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1044                         else
1045                                 return r;
1046                 }
1047         }
1048
1049         if (p) {
1050                 mkdir_p("/run/systemd/nspawn/locks", 0700);
1051
1052                 r = make_lock_file(p, operation, global);
1053                 if (r < 0) {
1054                         release_lock_file(&t);
1055                         return r;
1056                 }
1057         } else
1058                 *global = (LockFile) LOCK_FILE_INIT;
1059
1060         *local = t;
1061         return 0;
1062 }
1063
1064 int image_set_limit(Image *i, uint64_t referenced_max) {
1065         assert(i);
1066
1067         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1068                 return -EROFS;
1069
1070         if (i->type != IMAGE_SUBVOLUME)
1071                 return -EOPNOTSUPP;
1072
1073         /* We set the quota both for the subvolume as well as for the
1074          * subtree. The latter is mostly for historical reasons, since
1075          * we didn't use to have a concept of subtree quota, and hence
1076          * only modified the subvolume quota. */
1077
1078         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1079         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1080         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1081 }
1082
1083 int image_read_metadata(Image *i) {
1084         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1085         int r;
1086
1087         assert(i);
1088
1089         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1090         if (r < 0)
1091                 return r;
1092
1093         switch (i->type) {
1094
1095         case IMAGE_SUBVOLUME:
1096         case IMAGE_DIRECTORY: {
1097                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1098                 sd_id128_t machine_id = SD_ID128_NULL;
1099                 _cleanup_free_ char *hostname = NULL;
1100                 _cleanup_free_ char *path = NULL;
1101
1102                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1103                 if (r < 0 && r != -ENOENT)
1104                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1105                 else if (r >= 0) {
1106                         r = read_etc_hostname(path, &hostname);
1107                         if (r < 0)
1108                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1109                 }
1110
1111                 path = mfree(path);
1112
1113                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1114                 if (r < 0 && r != -ENOENT)
1115                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1116                 else if (r >= 0) {
1117                         _cleanup_close_ int fd = -1;
1118
1119                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1120                         if (fd < 0)
1121                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1122                         else {
1123                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1124                                 if (r < 0)
1125                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1126                         }
1127                 }
1128
1129                 path = mfree(path);
1130
1131                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path);
1132                 if (r < 0 && r != -ENOENT)
1133                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1134                 else if (r >= 0) {
1135                         r = load_env_file_pairs(NULL, path, NULL, &machine_info);
1136                         if (r < 0)
1137                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1138                 }
1139
1140                 r = load_os_release_pairs(i->path, &os_release);
1141                 if (r < 0)
1142                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1143
1144                 free_and_replace(i->hostname, hostname);
1145                 i->machine_id = machine_id;
1146                 strv_free_and_replace(i->machine_info, machine_info);
1147                 strv_free_and_replace(i->os_release, os_release);
1148
1149                 break;
1150         }
1151
1152         case IMAGE_RAW:
1153         case IMAGE_BLOCK: {
1154                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1155                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1156
1157                 r = loop_device_make_by_path(i->path, O_RDONLY, &d);
1158                 if (r < 0)
1159                         return r;
1160
1161                 r = dissect_image(d->fd, NULL, 0, DISSECT_IMAGE_REQUIRE_ROOT, &m);
1162                 if (r < 0)
1163                         return r;
1164
1165                 r = dissected_image_acquire_metadata(m);
1166                 if (r < 0)
1167                         return r;
1168
1169                 free_and_replace(i->hostname, m->hostname);
1170                 i->machine_id = m->machine_id;
1171                 strv_free_and_replace(i->machine_info, m->machine_info);
1172                 strv_free_and_replace(i->os_release, m->os_release);
1173
1174                 break;
1175         }
1176
1177         default:
1178                 return -EOPNOTSUPP;
1179         }
1180
1181         i->metadata_valid = true;
1182
1183         return 0;
1184 }
1185
1186 int image_name_lock(const char *name, int operation, LockFile *ret) {
1187         const char *p;
1188
1189         assert(name);
1190         assert(ret);
1191
1192         /* Locks an image name, regardless of the precise path used. */
1193
1194         if (!image_name_is_valid(name))
1195                 return -EINVAL;
1196
1197         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1198                 *ret = (LockFile) LOCK_FILE_INIT;
1199                 return 0;
1200         }
1201
1202         if (streq(name, ".host"))
1203                 return -EBUSY;
1204
1205         mkdir_p("/run/systemd/nspawn/locks", 0700);
1206         p = strjoina("/run/systemd/nspawn/locks/name-", name);
1207
1208         return make_lock_file(p, operation, ret);
1209 }
1210
1211 bool image_name_is_valid(const char *s) {
1212         if (!filename_is_valid(s))
1213                 return false;
1214
1215         if (string_has_cc(s, NULL))
1216                 return false;
1217
1218         if (!utf8_is_valid(s))
1219                 return false;
1220
1221         /* Temporary files for atomically creating new files */
1222         if (startswith(s, ".#"))
1223                 return false;
1224
1225         return true;
1226 }
1227
1228 bool image_in_search_path(ImageClass class, const char *image) {
1229         const char *path;
1230
1231         assert(image);
1232
1233         NULSTR_FOREACH(path, image_search_path[class]) {
1234                 const char *p;
1235                 size_t k;
1236
1237                 p = path_startswith(image, path);
1238                 if (!p)
1239                         continue;
1240
1241                 /* Make sure there's a filename following */
1242                 k = strcspn(p, "/");
1243                 if (k == 0)
1244                         continue;
1245
1246                 p += k;
1247
1248                 /* Accept trailing slashes */
1249                 if (p[strspn(p, "/")] == 0)
1250                         return true;
1251
1252         }
1253
1254         return false;
1255 }
1256
1257 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1258         [IMAGE_DIRECTORY] = "directory",
1259         [IMAGE_SUBVOLUME] = "subvolume",
1260         [IMAGE_RAW] = "raw",
1261         [IMAGE_BLOCK] = "block",
1262 };
1263
1264 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);