src/shared/machine-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <fcntl.h>
   5 #include <linux/fs.h>
   6 #include <linux/loop.h>
   7 #include <stdio.h>
   8 #include <stdlib.h>
   9 #include <sys/file.h>
  10 #include <sys/ioctl.h>
  11 #include <sys/stat.h>
  12 #include <unistd.h>
  13
  14 #include "alloc-util.h"
  15 #include "btrfs-util.h"
  16 #include "chattr-util.h"
  17 #include "copy.h"
  18 #include "dirent-util.h"
  19 #include "dissect-image.h"
  20 #include "env-file.h"
  21 #include "env-util.h"
  22 #include "fd-util.h"
  23 #include "fs-util.h"
  24 #include "hashmap.h"
  25 #include "hostname-util.h"
  26 #include "id128-util.h"
  27 #include "lockfile-util.h"
  28 #include "log.h"
  29 #include "loop-util.h"
  30 #include "machine-image.h"
  31 #include "macro.h"
  32 #include "mkdir.h"
  33 #include "nulstr-util.h"
  34 #include "os-util.h"
  35 #include "path-util.h"
  36 #include "rm-rf.h"
  37 #include "string-table.h"
  38 #include "string-util.h"
  39 #include "strv.h"
  40 #include "time-util.h"
  41 #include "utf8.h"
  42 #include "xattr-util.h"
  43
  44 static const char* const image_search_path[_IMAGE_CLASS_MAX] = {
  45         [IMAGE_MACHINE] =  "/etc/machines\0"              /* only place symlinks here */
  46                            "/run/machines\0"              /* and here too */
  47                            "/var/lib/machines\0"          /* the main place for images */
  48                            "/var/lib/container\0"         /* legacy */
  49                            "/usr/local/lib/machines\0"
  50                            "/usr/lib/machines\0",
  51
  52         [IMAGE_PORTABLE] = "/etc/portables\0"             /* only place symlinks here */
  53                            "/run/portables\0"             /* and here too */
  54                            "/var/lib/portables\0"         /* the main place for images */
  55                            "/usr/local/lib/portables\0"
  56                            "/usr/lib/portables\0",
  57 };
  58
  59 static Image *image_free(Image *i) {
  60         assert(i);
  61
  62         free(i->name);
  63         free(i->path);
  64
  65         free(i->hostname);
  66         strv_free(i->machine_info);
  67         strv_free(i->os_release);
  68
  69         return mfree(i);
  70 }
  71
  72 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free);
  73 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func,
  74                                       Image, image_unref);
  75
  76 static char **image_settings_path(Image *image) {
  77         _cleanup_strv_free_ char **l = NULL;
  78         const char *fn, *s;
  79         unsigned i = 0;
  80
  81         assert(image);
  82
  83         l = new0(char*, 4);
  84         if (!l)
  85                 return NULL;
  86
  87         fn = strjoina(image->name, ".nspawn");
  88
  89         FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
  90                 l[i] = path_join(s, fn);
  91                 if (!l[i])
  92                         return NULL;
  93
  94                 i++;
  95         }
  96
  97         l[i] = file_in_same_dir(image->path, fn);
  98         if (!l[i])
  99                 return NULL;
 100
 101         return TAKE_PTR(l);
 102 }
 103
 104 static char *image_roothash_path(Image *image) {
 105         const char *fn;
 106
 107         assert(image);
 108
 109         fn = strjoina(image->name, ".roothash");
 110
 111         return file_in_same_dir(image->path, fn);
 112 }
 113
 114 static int image_new(
 115                 ImageType t,
 116                 const char *pretty,
 117                 const char *path,
 118                 const char *filename,
 119                 bool read_only,
 120                 usec_t crtime,
 121                 usec_t mtime,
 122                 Image **ret) {
 123
 124         _cleanup_(image_unrefp) Image *i = NULL;
 125
 126         assert(t >= 0);
 127         assert(t < _IMAGE_TYPE_MAX);
 128         assert(pretty);
 129         assert(filename);
 130         assert(ret);
 131
 132         i = new0(Image, 1);
 133         if (!i)
 134                 return -ENOMEM;
 135
 136         i->n_ref = 1;
 137         i->type = t;
 138         i->read_only = read_only;
 139         i->crtime = crtime;
 140         i->mtime = mtime;
 141         i->usage = i->usage_exclusive = (uint64_t) -1;
 142         i->limit = i->limit_exclusive = (uint64_t) -1;
 143
 144         i->name = strdup(pretty);
 145         if (!i->name)
 146                 return -ENOMEM;
 147
 148         i->path = path_join(path, filename);
 149         if (!i->path)
 150                 return -ENOMEM;
 151
 152         path_simplify(i->path, false);
 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, *parent = NULL;
 200         struct stat stbuf;
 201         bool read_only;
 202         int r;
 203
 204         assert(dfd >= 0 || dfd == AT_FDCWD);
 205         assert(path || dfd == AT_FDCWD);
 206         assert(filename);
 207
 208         /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
 209          * devices into /var/lib/machines/, and treat them normally.
 210          *
 211          * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
 212          * recognize. */
 213
 214         if (!st) {
 215                 if (fstatat(dfd, filename, &stbuf, 0) < 0)
 216                         return -errno;
 217
 218                 st = &stbuf;
 219         }
 220
 221         if (!path) {
 222                 if (dfd == AT_FDCWD)
 223                         (void) safe_getcwd(&parent);
 224                 else
 225                         (void) fd_get_path(dfd, &parent);
 226         }
 227
 228         read_only =
 229                 (path && path_startswith(path, "/usr")) ||
 230                 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
 231
 232         if (S_ISDIR(st->st_mode)) {
 233                 _cleanup_close_ int fd = -1;
 234                 unsigned file_attr = 0;
 235
 236                 if (!ret)
 237                         return 0;
 238
 239                 if (!pretty) {
 240                         r = extract_pretty(filename, NULL, &pretty_buffer);
 241                         if (r < 0)
 242                                 return r;
 243
 244                         pretty = pretty_buffer;
 245                 }
 246
 247                 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
 248                 if (fd < 0)
 249                         return -errno;
 250
 251                 /* btrfs subvolumes have inode 256 */
 252                 if (st->st_ino == 256) {
 253
 254                         r = btrfs_is_filesystem(fd);
 255                         if (r < 0)
 256                                 return r;
 257                         if (r) {
 258                                 BtrfsSubvolInfo info;
 259
 260                                 /* It's a btrfs subvolume */
 261
 262                                 r = btrfs_subvol_get_info_fd(fd, 0, &info);
 263                                 if (r < 0)
 264                                         return r;
 265
 266                                 r = image_new(IMAGE_SUBVOLUME,
 267                                               pretty,
 268                                               path,
 269                                               filename,
 270                                               info.read_only || read_only,
 271                                               info.otime,
 272                                               0,
 273                                               ret);
 274                                 if (r < 0)
 275                                         return r;
 276
 277                                 if (btrfs_quota_scan_ongoing(fd) == 0) {
 278                                         BtrfsQuotaInfo quota;
 279
 280                                         r = btrfs_subvol_get_subtree_quota_fd(fd, 0, &quota);
 281                                         if (r >= 0) {
 282                                                 (*ret)->usage = quota.referenced;
 283                                                 (*ret)->usage_exclusive = quota.exclusive;
 284
 285                                                 (*ret)->limit = quota.referenced_max;
 286                                                 (*ret)->limit_exclusive = quota.exclusive_max;
 287                                         }
 288                                 }
 289
 290                                 return 0;
 291                         }
 292                 }
 293
 294                 /* If the IMMUTABLE bit is set, we consider the
 295                  * directory read-only. Since the ioctl is not
 296                  * supported everywhere we ignore failures. */
 297                 (void) read_attr_fd(fd, &file_attr);
 298
 299                 /* It's just a normal directory. */
 300                 r = image_new(IMAGE_DIRECTORY,
 301                               pretty,
 302                               path,
 303                               filename,
 304                               read_only || (file_attr & FS_IMMUTABLE_FL),
 305                               0,
 306                               0,
 307                               ret);
 308                 if (r < 0)
 309                         return r;
 310
 311                 return 0;
 312
 313         } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) {
 314                 usec_t crtime = 0;
 315
 316                 /* It's a RAW disk image */
 317
 318                 if (!ret)
 319                         return 0;
 320
 321                 (void) fd_getcrtime_at(dfd, filename, &crtime, 0);
 322
 323                 if (!pretty) {
 324                         r = extract_pretty(filename, ".raw", &pretty_buffer);
 325                         if (r < 0)
 326                                 return r;
 327
 328                         pretty = pretty_buffer;
 329                 }
 330
 331                 r = image_new(IMAGE_RAW,
 332                               pretty,
 333                               path,
 334                               filename,
 335                               !(st->st_mode & 0222) || read_only,
 336                               crtime,
 337                               timespec_load(&st->st_mtim),
 338                               ret);
 339                 if (r < 0)
 340                         return r;
 341
 342                 (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512;
 343                 (*ret)->limit = (*ret)->limit_exclusive = st->st_size;
 344
 345                 return 0;
 346
 347         } else if (S_ISBLK(st->st_mode)) {
 348                 _cleanup_close_ int block_fd = -1;
 349                 uint64_t size = UINT64_MAX;
 350
 351                 /* A block device */
 352
 353                 if (!ret)
 354                         return 0;
 355
 356                 if (!pretty) {
 357                         r = extract_pretty(filename, NULL, &pretty_buffer);
 358                         if (r < 0)
 359                                 return r;
 360
 361                         pretty = pretty_buffer;
 362                 }
 363
 364                 block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
 365                 if (block_fd < 0)
 366                         log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 367                 else {
 368                         /* Refresh stat data after opening the node */
 369                         if (fstat(block_fd, &stbuf) < 0)
 370                                 return -errno;
 371                         st = &stbuf;
 372
 373                         if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */
 374                                 return -ENOTTY;
 375
 376                         if (!read_only) {
 377                                 int state = 0;
 378
 379                                 if (ioctl(block_fd, BLKROGET, &state) < 0)
 380                                         log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 381                                 else if (state)
 382                                         read_only = true;
 383                         }
 384
 385                         if (ioctl(block_fd, BLKGETSIZE64, &size) < 0)
 386                                 log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename);
 387
 388                         block_fd = safe_close(block_fd);
 389                 }
 390
 391                 r = image_new(IMAGE_BLOCK,
 392                               pretty,
 393                               path,
 394                               filename,
 395                               !(st->st_mode & 0222) || read_only,
 396                               0,
 397                               0,
 398                               ret);
 399                 if (r < 0)
 400                         return r;
 401
 402                 if (!IN_SET(size, 0, UINT64_MAX))
 403                         (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size;
 404
 405                 return 0;
 406         }
 407
 408         return -EMEDIUMTYPE;
 409 }
 410
 411 int image_find(ImageClass class, const char *name, Image **ret) {
 412         const char *path;
 413         int r;
 414
 415         assert(class >= 0);
 416         assert(class < _IMAGE_CLASS_MAX);
 417         assert(name);
 418
 419         /* There are no images with invalid names */
 420         if (!image_name_is_valid(name))
 421                 return -ENOENT;
 422
 423         NULSTR_FOREACH(path, image_search_path[class]) {
 424                 _cleanup_closedir_ DIR *d = NULL;
 425                 struct stat st;
 426
 427                 d = opendir(path);
 428                 if (!d) {
 429                         if (errno == ENOENT)
 430                                 continue;
 431
 432                         return -errno;
 433                 }
 434
 435                 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to
 436                  * symlink block devices into the search path */
 437                 if (fstatat(dirfd(d), name, &st, 0) < 0) {
 438                         _cleanup_free_ char *raw = NULL;
 439
 440                         if (errno != ENOENT)
 441                                 return -errno;
 442
 443                         raw = strjoin(name, ".raw");
 444                         if (!raw)
 445                                 return -ENOMEM;
 446
 447                         if (fstatat(dirfd(d), raw, &st, 0) < 0) {
 448
 449                                 if (errno == ENOENT)
 450                                         continue;
 451
 452                                 return -errno;
 453                         }
 454
 455                         if (!S_ISREG(st.st_mode))
 456                                 continue;
 457
 458                         r = image_make(name, dirfd(d), path, raw, &st, ret);
 459
 460                 } else {
 461                         if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode))
 462                                 continue;
 463
 464                         r = image_make(name, dirfd(d), path, name, &st, ret);
 465                 }
 466                 if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 467                         continue;
 468                 if (r < 0)
 469                         return r;
 470
 471                 if (ret)
 472                         (*ret)->discoverable = true;
 473
 474                 return 1;
 475         }
 476
 477         if (class == IMAGE_MACHINE && streq(name, ".host")) {
 478                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 479                 if (r < 0)
 480                         return r;
 481
 482                 if (ret)
 483                         (*ret)->discoverable = true;
 484
 485                 return r;
 486         }
 487
 488         return -ENOENT;
 489 };
 490
 491 int image_from_path(const char *path, Image **ret) {
 492
 493         /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
 494          * the image is in the image search path. And if it is we don't know if the path we used is actually not
 495          * overridden by another, different image earlier in the search path */
 496
 497         if (path_equal(path, "/"))
 498                 return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret);
 499
 500         return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret);
 501 }
 502
 503 int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) {
 504         if (image_name_is_valid(name_or_path))
 505                 return image_find(class, name_or_path, ret);
 506
 507         return image_from_path(name_or_path, ret);
 508 }
 509
 510 int image_discover(ImageClass class, Hashmap *h) {
 511         const char *path;
 512         int r;
 513
 514         assert(class >= 0);
 515         assert(class < _IMAGE_CLASS_MAX);
 516         assert(h);
 517
 518         NULSTR_FOREACH(path, image_search_path[class]) {
 519                 _cleanup_closedir_ DIR *d = NULL;
 520                 struct dirent *de;
 521
 522                 d = opendir(path);
 523                 if (!d) {
 524                         if (errno == ENOENT)
 525                                 continue;
 526
 527                         return -errno;
 528                 }
 529
 530                 FOREACH_DIRENT_ALL(de, d, return -errno) {
 531                         _cleanup_(image_unrefp) Image *image = NULL;
 532                         _cleanup_free_ char *truncated = NULL;
 533                         const char *pretty;
 534                         struct stat st;
 535
 536                         if (dot_or_dot_dot(de->d_name))
 537                                 continue;
 538
 539                         /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
 540                          * to symlink block devices into the search path */
 541                         if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) {
 542                                 if (errno == ENOENT)
 543                                         continue;
 544
 545                                 return -errno;
 546                         }
 547
 548                         if (S_ISREG(st.st_mode)) {
 549                                 const char *e;
 550
 551                                 e = endswith(de->d_name, ".raw");
 552                                 if (!e)
 553                                         continue;
 554
 555                                 truncated = strndup(de->d_name, e - de->d_name);
 556                                 if (!truncated)
 557                                         return -ENOMEM;
 558
 559                                 pretty = truncated;
 560                         } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode))
 561                                 pretty = de->d_name;
 562                         else
 563                                 continue;
 564
 565                         if (!image_name_is_valid(pretty))
 566                                 continue;
 567
 568                         if (hashmap_contains(h, pretty))
 569                                 continue;
 570
 571                         r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image);
 572                         if (IN_SET(r, -ENOENT, -EMEDIUMTYPE))
 573                                 continue;
 574                         if (r < 0)
 575                                 return r;
 576
 577                         image->discoverable = true;
 578
 579                         r = hashmap_put(h, image->name, image);
 580                         if (r < 0)
 581                                 return r;
 582
 583                         image = NULL;
 584                 }
 585         }
 586
 587         if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) {
 588                 _cleanup_(image_unrefp) Image *image = NULL;
 589
 590                 r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image);
 591                 if (r < 0)
 592                         return r;
 593
 594                 image->discoverable = true;
 595
 596                 r = hashmap_put(h, image->name, image);
 597                 if (r < 0)
 598                         return r;
 599
 600                 image = NULL;
 601         }
 602
 603         return 0;
 604 }
 605
 606 int image_remove(Image *i) {
 607         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 608         _cleanup_strv_free_ char **settings = NULL;
 609         _cleanup_free_ char *roothash = NULL;
 610         char **j;
 611         int r;
 612
 613         assert(i);
 614
 615         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 616                 return -EROFS;
 617
 618         settings = image_settings_path(i);
 619         if (!settings)
 620                 return -ENOMEM;
 621
 622         roothash = image_roothash_path(i);
 623         if (!roothash)
 624                 return -ENOMEM;
 625
 626         /* Make sure we don't interfere with a running nspawn */
 627         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 628         if (r < 0)
 629                 return r;
 630
 631         switch (i->type) {
 632
 633         case IMAGE_SUBVOLUME:
 634
 635                 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
 636                  * big guns */
 637                 if (unlink(i->path) < 0) {
 638                         r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA);
 639                         if (r < 0)
 640                                 return r;
 641                 }
 642
 643                 break;
 644
 645         case IMAGE_DIRECTORY:
 646                 /* Allow deletion of read-only directories */
 647                 (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 648                 r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 649                 if (r < 0)
 650                         return r;
 651
 652                 break;
 653
 654         case IMAGE_BLOCK:
 655
 656                 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
 657                  * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
 658                  * the thing (it's most likely a symlink after all). */
 659
 660                 if (path_startswith(i->path, "/dev"))
 661                         break;
 662
 663                 _fallthrough_;
 664         case IMAGE_RAW:
 665                 if (unlink(i->path) < 0)
 666                         return -errno;
 667                 break;
 668
 669         default:
 670                 return -EOPNOTSUPP;
 671         }
 672
 673         STRV_FOREACH(j, settings) {
 674                 if (unlink(*j) < 0 && errno != ENOENT)
 675                         log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j);
 676         }
 677
 678         if (unlink(roothash) < 0 && errno != ENOENT)
 679                 log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash);
 680
 681         return 0;
 682 }
 683
 684 static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 685         _cleanup_free_ char *rs = NULL;
 686         const char *fn;
 687
 688         fn = strjoina(new_name, suffix);
 689
 690         rs = file_in_same_dir(path, fn);
 691         if (!rs)
 692                 return -ENOMEM;
 693
 694         return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs);
 695 }
 696
 697 int image_rename(Image *i, const char *new_name) {
 698         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
 699         _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL;
 700         _cleanup_strv_free_ char **settings = NULL;
 701         unsigned file_attr = 0;
 702         char **j;
 703         int r;
 704
 705         assert(i);
 706
 707         if (!image_name_is_valid(new_name))
 708                 return -EINVAL;
 709
 710         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 711                 return -EROFS;
 712
 713         settings = image_settings_path(i);
 714         if (!settings)
 715                 return -ENOMEM;
 716
 717         roothash = image_roothash_path(i);
 718         if (!roothash)
 719                 return -ENOMEM;
 720
 721         /* Make sure we don't interfere with a running nspawn */
 722         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 723         if (r < 0)
 724                 return r;
 725
 726         /* Make sure nobody takes the new name, between the time we
 727          * checked it is currently unused in all search paths, and the
 728          * time we take possession of it */
 729         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 730         if (r < 0)
 731                 return r;
 732
 733         r = image_find(IMAGE_MACHINE, new_name, NULL);
 734         if (r >= 0)
 735                 return -EEXIST;
 736         if (r != -ENOENT)
 737                 return r;
 738
 739         switch (i->type) {
 740
 741         case IMAGE_DIRECTORY:
 742                 /* Turn of the immutable bit while we rename the image, so that we can rename it */
 743                 (void) read_attr_path(i->path, &file_attr);
 744
 745                 if (file_attr & FS_IMMUTABLE_FL)
 746                         (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL);
 747
 748                 _fallthrough_;
 749         case IMAGE_SUBVOLUME:
 750                 new_path = file_in_same_dir(i->path, new_name);
 751                 break;
 752
 753         case IMAGE_BLOCK:
 754
 755                 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
 756                 if (path_startswith(i->path, "/dev"))
 757                         return -EROFS;
 758
 759                 new_path = file_in_same_dir(i->path, new_name);
 760                 break;
 761
 762         case IMAGE_RAW: {
 763                 const char *fn;
 764
 765                 fn = strjoina(new_name, ".raw");
 766                 new_path = file_in_same_dir(i->path, fn);
 767                 break;
 768         }
 769
 770         default:
 771                 return -EOPNOTSUPP;
 772         }
 773
 774         if (!new_path)
 775                 return -ENOMEM;
 776
 777         nn = strdup(new_name);
 778         if (!nn)
 779                 return -ENOMEM;
 780
 781         r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
 782         if (r < 0)
 783                 return r;
 784
 785         /* Restore the immutable bit, if it was set before */
 786         if (file_attr & FS_IMMUTABLE_FL)
 787                 (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL);
 788
 789         free_and_replace(i->path, new_path);
 790         free_and_replace(i->name, nn);
 791
 792         STRV_FOREACH(j, settings) {
 793                 r = rename_auxiliary_file(*j, new_name, ".nspawn");
 794                 if (r < 0 && r != -ENOENT)
 795                         log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j);
 796         }
 797
 798         r = rename_auxiliary_file(roothash, new_name, ".roothash");
 799         if (r < 0 && r != -ENOENT)
 800                 log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash);
 801
 802         return 0;
 803 }
 804
 805 static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) {
 806         _cleanup_free_ char *rs = NULL;
 807         const char *fn;
 808
 809         fn = strjoina(new_name, suffix);
 810
 811         rs = file_in_same_dir(path, fn);
 812         if (!rs)
 813                 return -ENOMEM;
 814
 815         return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK);
 816 }
 817
 818 int image_clone(Image *i, const char *new_name, bool read_only) {
 819         _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT;
 820         _cleanup_strv_free_ char **settings = NULL;
 821         _cleanup_free_ char *roothash = NULL;
 822         const char *new_path;
 823         char **j;
 824         int r;
 825
 826         assert(i);
 827
 828         if (!image_name_is_valid(new_name))
 829                 return -EINVAL;
 830
 831         settings = image_settings_path(i);
 832         if (!settings)
 833                 return -ENOMEM;
 834
 835         roothash = image_roothash_path(i);
 836         if (!roothash)
 837                 return -ENOMEM;
 838
 839         /* Make sure nobody takes the new name, between the time we
 840          * checked it is currently unused in all search paths, and the
 841          * time we take possession of it */
 842         r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
 843         if (r < 0)
 844                 return r;
 845
 846         r = image_find(IMAGE_MACHINE, new_name, NULL);
 847         if (r >= 0)
 848                 return -EEXIST;
 849         if (r != -ENOENT)
 850                 return r;
 851
 852         switch (i->type) {
 853
 854         case IMAGE_SUBVOLUME:
 855         case IMAGE_DIRECTORY:
 856                 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
 857                  * directory. */
 858
 859                 new_path = strjoina("/var/lib/machines/", new_name);
 860
 861                 r = btrfs_subvol_snapshot(i->path, new_path,
 862                                           (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
 863                                           BTRFS_SNAPSHOT_FALLBACK_COPY |
 864                                           BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
 865                                           BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
 866                                           BTRFS_SNAPSHOT_RECURSIVE |
 867                                           BTRFS_SNAPSHOT_QUOTA);
 868                 if (r >= 0)
 869                         /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
 870                         (void) btrfs_subvol_auto_qgroup(new_path, 0, true);
 871
 872                 break;
 873
 874         case IMAGE_RAW:
 875                 new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 876
 877                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME);
 878                 break;
 879
 880         case IMAGE_BLOCK:
 881         default:
 882                 return -EOPNOTSUPP;
 883         }
 884
 885         if (r < 0)
 886                 return r;
 887
 888         STRV_FOREACH(j, settings) {
 889                 r = clone_auxiliary_file(*j, new_name, ".nspawn");
 890                 if (r < 0 && r != -ENOENT)
 891                         log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j);
 892         }
 893
 894         r = clone_auxiliary_file(roothash, new_name, ".roothash");
 895         if (r < 0 && r != -ENOENT)
 896                 log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash);
 897
 898         return 0;
 899 }
 900
 901 int image_read_only(Image *i, bool b) {
 902         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
 903         int r;
 904
 905         assert(i);
 906
 907         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
 908                 return -EROFS;
 909
 910         /* Make sure we don't interfere with a running nspawn */
 911         r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
 912         if (r < 0)
 913                 return r;
 914
 915         switch (i->type) {
 916
 917         case IMAGE_SUBVOLUME:
 918
 919                 /* Note that we set the flag only on the top-level
 920                  * subvolume of the image. */
 921
 922                 r = btrfs_subvol_set_read_only(i->path, b);
 923                 if (r < 0)
 924                         return r;
 925
 926                 break;
 927
 928         case IMAGE_DIRECTORY:
 929                 /* For simple directory trees we cannot use the access
 930                    mode of the top-level directory, since it has an
 931                    effect on the container itself.  However, we can
 932                    use the "immutable" flag, to at least make the
 933                    top-level directory read-only. It's not as good as
 934                    a read-only subvolume, but at least something, and
 935                    we can read the value back. */
 936
 937                 r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL);
 938                 if (r < 0)
 939                         return r;
 940
 941                 break;
 942
 943         case IMAGE_RAW: {
 944                 struct stat st;
 945
 946                 if (stat(i->path, &st) < 0)
 947                         return -errno;
 948
 949                 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
 950                         return -errno;
 951
 952                 /* If the images is now read-only, it's a good time to
 953                  * defrag it, given that no write patterns will
 954                  * fragment it again. */
 955                 if (b)
 956                         (void) btrfs_defrag(i->path);
 957                 break;
 958         }
 959
 960         case IMAGE_BLOCK: {
 961                 _cleanup_close_ int fd = -1;
 962                 struct stat st;
 963                 int state = b;
 964
 965                 fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY);
 966                 if (fd < 0)
 967                         return -errno;
 968
 969                 if (fstat(fd, &st) < 0)
 970                         return -errno;
 971                 if (!S_ISBLK(st.st_mode))
 972                         return -ENOTTY;
 973
 974                 if (ioctl(fd, BLKROSET, &state) < 0)
 975                         return -errno;
 976
 977                 break;
 978         }
 979
 980         default:
 981                 return -EOPNOTSUPP;
 982         }
 983
 984         return 0;
 985 }
 986
 987 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
 988         _cleanup_free_ char *p = NULL;
 989         LockFile t = LOCK_FILE_INIT;
 990         struct stat st;
 991         bool exclusive;
 992         int r;
 993
 994         assert(path);
 995         assert(global);
 996         assert(local);
 997
 998         /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
 999          * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1000          * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1001          * correctly. */
1002
1003         if (!path_is_absolute(path))
1004                 return -EINVAL;
1005
1006         switch (operation & (LOCK_SH|LOCK_EX)) {
1007         case LOCK_SH:
1008                 exclusive = false;
1009                 break;
1010         case LOCK_EX:
1011                 exclusive = true;
1012                 break;
1013         default:
1014                 return -EINVAL;
1015         }
1016
1017         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1018                 *local = *global = (LockFile) LOCK_FILE_INIT;
1019                 return 0;
1020         }
1021
1022         /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1023          * running off it after all, and we don't want any images to manipulate the host image. We make an
1024          * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1025          * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1026          * since it means changes made to the host might propagate to the container as they happen (and a
1027          * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1028          * useful not to allow read-only containers off the host root, hence let's support this, and trust
1029          * the user to do the right thing with this. */
1030         if (path_equal(path, "/")) {
1031                 if (exclusive)
1032                         return -EBUSY;
1033
1034                 *local = *global = (LockFile) LOCK_FILE_INIT;
1035                 return 0;
1036         }
1037
1038         if (stat(path, &st) >= 0) {
1039                 if (S_ISBLK(st.st_mode))
1040                         r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev));
1041                 else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode))
1042                         r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino);
1043                 else
1044                         return -ENOTTY;
1045
1046                 if (r < 0)
1047                         return -ENOMEM;
1048         }
1049
1050         /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1051          * sufficient, since block devices are host local anyway. */
1052         if (!path_startswith(path, "/dev/")) {
1053                 r = make_lock_file_for(path, operation, &t);
1054                 if (r < 0) {
1055                         if (!exclusive && r == -EROFS)
1056                                 log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path);
1057                         else
1058                                 return r;
1059                 }
1060         }
1061
1062         if (p) {
1063                 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1064
1065                 r = make_lock_file(p, operation, global);
1066                 if (r < 0) {
1067                         release_lock_file(&t);
1068                         return r;
1069                 }
1070         } else
1071                 *global = (LockFile) LOCK_FILE_INIT;
1072
1073         *local = t;
1074         return 0;
1075 }
1076
1077 int image_set_limit(Image *i, uint64_t referenced_max) {
1078         assert(i);
1079
1080         if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i))
1081                 return -EROFS;
1082
1083         if (i->type != IMAGE_SUBVOLUME)
1084                 return -EOPNOTSUPP;
1085
1086         /* We set the quota both for the subvolume as well as for the
1087          * subtree. The latter is mostly for historical reasons, since
1088          * we didn't use to have a concept of subtree quota, and hence
1089          * only modified the subvolume quota. */
1090
1091         (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max);
1092         (void) btrfs_subvol_auto_qgroup(i->path, 0, true);
1093         return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max);
1094 }
1095
1096 int image_read_metadata(Image *i) {
1097         _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
1098         int r;
1099
1100         assert(i);
1101
1102         r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock);
1103         if (r < 0)
1104                 return r;
1105
1106         switch (i->type) {
1107
1108         case IMAGE_SUBVOLUME:
1109         case IMAGE_DIRECTORY: {
1110                 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1111                 sd_id128_t machine_id = SD_ID128_NULL;
1112                 _cleanup_free_ char *hostname = NULL;
1113                 _cleanup_free_ char *path = NULL;
1114
1115                 r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1116                 if (r < 0 && r != -ENOENT)
1117                         log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name);
1118                 else if (r >= 0) {
1119                         r = read_etc_hostname(path, &hostname);
1120                         if (r < 0)
1121                                 log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name);
1122                 }
1123
1124                 path = mfree(path);
1125
1126                 r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1127                 if (r < 0 && r != -ENOENT)
1128                         log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name);
1129                 else if (r >= 0) {
1130                         _cleanup_close_ int fd = -1;
1131
1132                         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1133                         if (fd < 0)
1134                                 log_debug_errno(errno, "Failed to open %s: %m", path);
1135                         else {
1136                                 r = id128_read_fd(fd, ID128_PLAIN, &machine_id);
1137                                 if (r < 0)
1138                                         log_debug_errno(r, "Image %s contains invalid machine ID.", i->name);
1139                         }
1140                 }
1141
1142                 path = mfree(path);
1143
1144                 r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL);
1145                 if (r < 0 && r != -ENOENT)
1146                         log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name);
1147                 else if (r >= 0) {
1148                         r = load_env_file_pairs(NULL, path, &machine_info);
1149                         if (r < 0)
1150                                 log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name);
1151                 }
1152
1153                 r = load_os_release_pairs(i->path, &os_release);
1154                 if (r < 0)
1155                         log_debug_errno(r, "Failed to read os-release in image, ignoring: %m");
1156
1157                 free_and_replace(i->hostname, hostname);
1158                 i->machine_id = machine_id;
1159                 strv_free_and_replace(i->machine_info, machine_info);
1160                 strv_free_and_replace(i->os_release, os_release);
1161
1162                 break;
1163         }
1164
1165         case IMAGE_RAW:
1166         case IMAGE_BLOCK: {
1167                 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1168                 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1169
1170                 r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, &d);
1171                 if (r < 0)
1172                         return r;
1173
1174                 r = dissect_image(d->fd, NULL, 0, NULL, DISSECT_IMAGE_REQUIRE_ROOT|DISSECT_IMAGE_RELAX_VAR_CHECK, &m);
1175                 if (r < 0)
1176                         return r;
1177
1178                 r = dissected_image_acquire_metadata(m);
1179                 if (r < 0)
1180                         return r;
1181
1182                 free_and_replace(i->hostname, m->hostname);
1183                 i->machine_id = m->machine_id;
1184                 strv_free_and_replace(i->machine_info, m->machine_info);
1185                 strv_free_and_replace(i->os_release, m->os_release);
1186
1187                 break;
1188         }
1189
1190         default:
1191                 return -EOPNOTSUPP;
1192         }
1193
1194         i->metadata_valid = true;
1195
1196         return 0;
1197 }
1198
1199 int image_name_lock(const char *name, int operation, LockFile *ret) {
1200         assert(name);
1201         assert(ret);
1202
1203         /* Locks an image name, regardless of the precise path used. */
1204
1205         if (!image_name_is_valid(name))
1206                 return -EINVAL;
1207
1208         if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1209                 *ret = (LockFile) LOCK_FILE_INIT;
1210                 return 0;
1211         }
1212
1213         if (streq(name, ".host"))
1214                 return -EBUSY;
1215
1216         const char *p = strjoina("/run/systemd/nspawn/locks/name-", name);
1217         (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1218         return make_lock_file(p, operation, ret);
1219 }
1220
1221 bool image_name_is_valid(const char *s) {
1222         if (!filename_is_valid(s))
1223                 return false;
1224
1225         if (string_has_cc(s, NULL))
1226                 return false;
1227
1228         if (!utf8_is_valid(s))
1229                 return false;
1230
1231         /* Temporary files for atomically creating new files */
1232         if (startswith(s, ".#"))
1233                 return false;
1234
1235         return true;
1236 }
1237
1238 bool image_in_search_path(ImageClass class, const char *image) {
1239         const char *path;
1240
1241         assert(image);
1242
1243         NULSTR_FOREACH(path, image_search_path[class]) {
1244                 const char *p;
1245                 size_t k;
1246
1247                 p = path_startswith(image, path);
1248                 if (!p)
1249                         continue;
1250
1251                 /* Make sure there's a filename following */
1252                 k = strcspn(p, "/");
1253                 if (k == 0)
1254                         continue;
1255
1256                 p += k;
1257
1258                 /* Accept trailing slashes */
1259                 if (p[strspn(p, "/")] == 0)
1260                         return true;
1261
1262         }
1263
1264         return false;
1265 }
1266
1267 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
1268         [IMAGE_DIRECTORY] = "directory",
1269         [IMAGE_SUBVOLUME] = "subvolume",
1270         [IMAGE_RAW] = "raw",
1271         [IMAGE_BLOCK] = "block",
1272 };
1273
1274 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);