src/shared/machine-image.c

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