src/shared/machine-image.c

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