src/home/homework-luks.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <linux/loop.h>
   4 #include <poll.h>
   5 #include <sys/file.h>
   6 #include <sys/ioctl.h>
   7 #include <sys/xattr.h>
   8
   9 #if HAVE_VALGRIND_MEMCHECK_H
  10 #include <valgrind/memcheck.h>
  11 #endif
  12
  13 #include "sd-daemon.h"
  14 #include "sd-device.h"
  15 #include "sd-event.h"
  16 #include "sd-id128.h"
  17
  18 #include "blkid-util.h"
  19 #include "blockdev-util.h"
  20 #include "btrfs-util.h"
  21 #include "chattr-util.h"
  22 #include "device-util.h"
  23 #include "devnum-util.h"
  24 #include "dm-util.h"
  25 #include "env-util.h"
  26 #include "errno-util.h"
  27 #include "fd-util.h"
  28 #include "fdisk-util.h"
  29 #include "fileio.h"
  30 #include "filesystems.h"
  31 #include "fs-util.h"
  32 #include "fsck-util.h"
  33 #include "glyph-util.h"
  34 #include "gpt.h"
  35 #include "home-util.h"
  36 #include "homework-luks.h"
  37 #include "homework-mount.h"
  38 #include "io-util.h"
  39 #include "keyring-util.h"
  40 #include "memory-util.h"
  41 #include "missing_magic.h"
  42 #include "mkdir.h"
  43 #include "mkfs-util.h"
  44 #include "mount-util.h"
  45 #include "openssl-util.h"
  46 #include "parse-util.h"
  47 #include "path-util.h"
  48 #include "process-util.h"
  49 #include "random-util.h"
  50 #include "resize-fs.h"
  51 #include "strv.h"
  52 #include "sync-util.h"
  53 #include "tmpfile-util.h"
  54 #include "udev-util.h"
  55 #include "user-util.h"
  56
  57 /* Round down to the nearest 4K size. Given that newer hardware generally prefers 4K sectors, let's align our
  58  * partitions to that too. In the worst case we'll waste 3.5K per partition that way, but I think I can live
  59  * with that. */
  60 #define DISK_SIZE_ROUND_DOWN(x) ((x) & ~UINT64_C(4095))
  61
  62 /* Rounds up to the nearest 4K boundary. Returns UINT64_MAX on overflow */
  63 #define DISK_SIZE_ROUND_UP(x)                                           \
  64         ({                                                              \
  65                 uint64_t _x = (x);                                      \
  66                 _x > UINT64_MAX - 4095U ? UINT64_MAX : (_x + 4095U) & ~UINT64_C(4095); \
  67         })
  68
  69 /* How much larger will the image on disk be than the fs inside it, i.e. the space we pay for the GPT and
  70  * LUKS2 envelope. (As measured on cryptsetup 2.4.1) */
  71 #define GPT_LUKS2_OVERHEAD UINT64_C(18874368)
  72
  73 static int resize_image_loop(UserRecord *h, HomeSetup *setup, uint64_t old_image_size, uint64_t new_image_size, uint64_t *ret_image_size);
  74
  75 int run_mark_dirty(int fd, bool b) {
  76         char x = '1';
  77         int r, ret;
  78
  79         /* Sets or removes the 'user.home-dirty' xattr on the specified file. We use this to detect when a
  80          * home directory was not properly unmounted. */
  81
  82         assert(fd >= 0);
  83
  84         r = fd_verify_regular(fd);
  85         if (r < 0)
  86                 return r;
  87
  88         if (b) {
  89                 ret = fsetxattr(fd, "user.home-dirty", &x, 1, XATTR_CREATE);
  90                 if (ret < 0 && errno != EEXIST)
  91                         return log_debug_errno(errno, "Could not mark home directory as dirty: %m");
  92
  93         } else {
  94                 r = fsync_full(fd);
  95                 if (r < 0)
  96                         return log_debug_errno(r, "Failed to synchronize image before marking it clean: %m");
  97
  98                 ret = fremovexattr(fd, "user.home-dirty");
  99                 if (ret < 0 && !ERRNO_IS_XATTR_ABSENT(errno))
 100                         return log_debug_errno(errno, "Could not mark home directory as clean: %m");
 101         }
 102
 103         r = fsync_full(fd);
 104         if (r < 0)
 105                 return log_debug_errno(r, "Failed to synchronize dirty flag to disk: %m");
 106
 107         return ret >= 0;
 108 }
 109
 110 int run_mark_dirty_by_path(const char *path, bool b) {
 111         _cleanup_close_ int fd = -EBADF;
 112
 113         assert(path);
 114
 115         fd = open(path, O_RDWR|O_CLOEXEC|O_NOCTTY);
 116         if (fd < 0)
 117                 return log_debug_errno(errno, "Failed to open %s to mark dirty or clean: %m", path);
 118
 119         return run_mark_dirty(fd, b);
 120 }
 121
 122 static int probe_file_system_by_fd(
 123                 int fd,
 124                 char **ret_fstype,
 125                 sd_id128_t *ret_uuid) {
 126
 127         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
 128         _cleanup_free_ char *s = NULL;
 129         const char *fstype = NULL, *uuid = NULL;
 130         sd_id128_t id;
 131         int r;
 132
 133         assert(fd >= 0);
 134         assert(ret_fstype);
 135         assert(ret_uuid);
 136
 137         b = blkid_new_probe();
 138         if (!b)
 139                 return -ENOMEM;
 140
 141         errno = 0;
 142         r = blkid_probe_set_device(b, fd, 0, 0);
 143         if (r != 0)
 144                 return errno_or_else(ENOMEM);
 145
 146         (void) blkid_probe_enable_superblocks(b, 1);
 147         (void) blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_UUID);
 148
 149         errno = 0;
 150         r = blkid_do_safeprobe(b);
 151         if (r == _BLKID_SAFEPROBE_ERROR)
 152                 return errno_or_else(EIO);
 153         if (IN_SET(r, _BLKID_SAFEPROBE_AMBIGUOUS, _BLKID_SAFEPROBE_NOT_FOUND))
 154                 return -ENOPKG;
 155
 156         assert(r == _BLKID_SAFEPROBE_FOUND);
 157
 158         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
 159         if (!fstype)
 160                 return -ENOPKG;
 161
 162         (void) blkid_probe_lookup_value(b, "UUID", &uuid, NULL);
 163         if (!uuid)
 164                 return -ENOPKG;
 165
 166         r = sd_id128_from_string(uuid, &id);
 167         if (r < 0)
 168                 return r;
 169
 170         s = strdup(fstype);
 171         if (!s)
 172                 return -ENOMEM;
 173
 174         *ret_fstype = TAKE_PTR(s);
 175         *ret_uuid = id;
 176
 177         return 0;
 178 }
 179
 180 static int probe_file_system_by_path(const char *path, char **ret_fstype, sd_id128_t *ret_uuid) {
 181         _cleanup_close_ int fd = -EBADF;
 182
 183         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
 184         if (fd < 0)
 185                 return negative_errno();
 186
 187         return probe_file_system_by_fd(fd, ret_fstype, ret_uuid);
 188 }
 189
 190 static int block_get_size_by_fd(int fd, uint64_t *ret) {
 191         struct stat st;
 192
 193         assert(fd >= 0);
 194         assert(ret);
 195
 196         if (fstat(fd, &st) < 0)
 197                 return -errno;
 198
 199         if (!S_ISBLK(st.st_mode))
 200                 return -ENOTBLK;
 201
 202         return blockdev_get_device_size(fd, ret);
 203 }
 204
 205 static int block_get_size_by_path(const char *path, uint64_t *ret) {
 206         _cleanup_close_ int fd = -EBADF;
 207
 208         fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
 209         if (fd < 0)
 210                 return -errno;
 211
 212         return block_get_size_by_fd(fd, ret);
 213 }
 214
 215 static int run_fsck(const char *node, const char *fstype) {
 216         int r, exit_status;
 217         pid_t fsck_pid;
 218
 219         assert(node);
 220         assert(fstype);
 221
 222         r = fsck_exists_for_fstype(fstype);
 223         if (r < 0)
 224                 return log_error_errno(r, "Failed to check if fsck for file system %s exists: %m", fstype);
 225         if (r == 0) {
 226                 log_warning("No fsck for file system %s installed, ignoring.", fstype);
 227                 return 0;
 228         }
 229
 230         r = safe_fork("(fsck)",
 231                       FORK_RESET_SIGNALS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG_SIGTERM|FORK_LOG|FORK_STDOUT_TO_STDERR|FORK_CLOSE_ALL_FDS,
 232                       &fsck_pid);
 233         if (r < 0)
 234                 return r;
 235         if (r == 0) {
 236                 /* Child */
 237                 execlp("fsck", "fsck", "-aTl", node, NULL);
 238                 log_open();
 239                 log_error_errno(errno, "Failed to execute fsck: %m");
 240                 _exit(FSCK_OPERATIONAL_ERROR);
 241         }
 242
 243         exit_status = wait_for_terminate_and_check("fsck", fsck_pid, WAIT_LOG_ABNORMAL);
 244         if (exit_status < 0)
 245                 return exit_status;
 246         if ((exit_status & ~FSCK_ERROR_CORRECTED) != 0) {
 247                 log_warning("fsck failed with exit status %i.", exit_status);
 248
 249                 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
 250                         return log_error_errno(SYNTHETIC_ERRNO(EIO), "File system is corrupted, refusing.");
 251
 252                 log_warning("Ignoring fsck error.");
 253         }
 254
 255         log_info("File system check completed.");
 256
 257         return 1;
 258 }
 259
 260 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(key_serial_t, keyring_unlink, -1);
 261
 262 static int upload_to_keyring(
 263                 UserRecord *h,
 264                 const char *password,
 265                 key_serial_t *ret_key_serial) {
 266
 267         _cleanup_free_ char *name = NULL;
 268         key_serial_t serial;
 269
 270         assert(h);
 271         assert(password);
 272
 273         /* If auto-shrink-on-logout is turned on, we need to keep the key we used to unlock the LUKS volume
 274          * around, since we'll need it when automatically resizing (since we can't ask the user there
 275          * again). We do this by uploading it into the kernel keyring, specifically the "session" one. This
 276          * is done under the assumption systemd-homed gets its private per-session keyring (i.e. default
 277          * service behaviour, given that KeyringMode=private is the default). It will survive between our
 278          * systemd-homework invocations that way.
 279          *
 280          * If auto-shrink-on-logout is disabled we'll skip this step, to be frugal with sensitive data. */
 281
 282         if (user_record_auto_resize_mode(h) != AUTO_RESIZE_SHRINK_AND_GROW) {  /* Won't need it */
 283                 if (ret_key_serial)
 284                         *ret_key_serial = -1;
 285                 return 0;
 286         }
 287
 288         name = strjoin("homework-user-", h->user_name);
 289         if (!name)
 290                 return -ENOMEM;
 291
 292         serial = add_key("user", name, password, strlen(password), KEY_SPEC_SESSION_KEYRING);
 293         if (serial == -1)
 294                 return -errno;
 295
 296         if (ret_key_serial)
 297                 *ret_key_serial = serial;
 298
 299         return 1;
 300 }
 301
 302 static int luks_try_passwords(
 303                 UserRecord *h,
 304                 struct crypt_device *cd,
 305                 char **passwords,
 306                 void *volume_key,
 307                 size_t *volume_key_size,
 308                 key_serial_t *ret_key_serial) {
 309
 310         int r;
 311
 312         assert(h);
 313         assert(cd);
 314
 315         STRV_FOREACH(pp, passwords) {
 316                 size_t vks = *volume_key_size;
 317
 318                 r = sym_crypt_volume_key_get(
 319                                 cd,
 320                                 CRYPT_ANY_SLOT,
 321                                 volume_key,
 322                                 &vks,
 323                                 *pp,
 324                                 strlen(*pp));
 325                 if (r >= 0) {
 326                         if (ret_key_serial) {
 327                                 /* If ret_key_serial is non-NULL, let's try to upload the password that
 328                                  * worked, and return its serial. */
 329                                 r = upload_to_keyring(h, *pp, ret_key_serial);
 330                                 if (r < 0) {
 331                                         log_debug_errno(r, "Failed to upload LUKS password to kernel keyring, ignoring: %m");
 332                                         *ret_key_serial = -1;
 333                                 }
 334                         }
 335
 336                         *volume_key_size = vks;
 337                         return 0;
 338                 }
 339
 340                 log_debug_errno(r, "Password %zu didn't work for unlocking LUKS superblock: %m", (size_t) (pp - passwords));
 341         }
 342
 343         return -ENOKEY;
 344 }
 345
 346 static int luks_setup(
 347                 UserRecord *h,
 348                 const char *node,
 349                 const char *dm_name,
 350                 sd_id128_t uuid,
 351                 const char *cipher,
 352                 const char *cipher_mode,
 353                 uint64_t volume_key_size,
 354                 char **passwords,
 355                 const PasswordCache *cache,
 356                 bool discard,
 357                 struct crypt_device **ret,
 358                 sd_id128_t *ret_found_uuid,
 359                 void **ret_volume_key,
 360                 size_t *ret_volume_key_size,
 361                 key_serial_t *ret_key_serial) {
 362
 363         _cleanup_(keyring_unlinkp) key_serial_t key_serial = -1;
 364         _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
 365         _cleanup_(erase_and_freep) void *vk = NULL;
 366         sd_id128_t p;
 367         size_t vks;
 368         char **list;
 369         int r;
 370
 371         assert(h);
 372         assert(node);
 373         assert(dm_name);
 374         assert(ret);
 375
 376         r = sym_crypt_init(&cd, node);
 377         if (r < 0)
 378                 return log_error_errno(r, "Failed to allocate libcryptsetup context: %m");
 379
 380         cryptsetup_enable_logging(cd);
 381
 382         r = sym_crypt_load(cd, CRYPT_LUKS2, NULL);
 383         if (r < 0)
 384                 return log_error_errno(r, "Failed to load LUKS superblock: %m");
 385
 386         r = sym_crypt_get_volume_key_size(cd);
 387         if (r <= 0)
 388                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine LUKS volume key size");
 389         vks = (size_t) r;
 390
 391         if (!sd_id128_is_null(uuid) || ret_found_uuid) {
 392                 const char *s;
 393
 394                 s = sym_crypt_get_uuid(cd);
 395                 if (!s)
 396                         return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock has no UUID.");
 397
 398                 r = sd_id128_from_string(s, &p);
 399                 if (r < 0)
 400                         return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock has invalid UUID.");
 401
 402                 /* Check that the UUID matches, if specified */
 403                 if (!sd_id128_is_null(uuid) &&
 404                     !sd_id128_equal(uuid, p))
 405                         return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock has wrong UUID.");
 406         }
 407
 408         if (cipher && !streq_ptr(cipher, sym_crypt_get_cipher(cd)))
 409                 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock declares wrong cipher.");
 410
 411         if (cipher_mode && !streq_ptr(cipher_mode, sym_crypt_get_cipher_mode(cd)))
 412                 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock declares wrong cipher mode.");
 413
 414         if (volume_key_size != UINT64_MAX && vks != volume_key_size)
 415                 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock declares wrong volume key size.");
 416
 417         vk = malloc(vks);
 418         if (!vk)
 419                 return log_oom();
 420
 421         r = -ENOKEY;
 422         FOREACH_POINTER(list,
 423                         cache ? cache->keyring_passswords : NULL,
 424                         cache ? cache->pkcs11_passwords : NULL,
 425                         cache ? cache->fido2_passwords : NULL,
 426                         passwords) {
 427                 r = luks_try_passwords(h, cd, list, vk, &vks, ret_key_serial ? &key_serial : NULL);
 428                 if (r != -ENOKEY)
 429                         break;
 430         }
 431         if (r == -ENOKEY)
 432                 return log_error_errno(r, "No valid password for LUKS superblock.");
 433         if (r < 0)
 434                 return log_error_errno(r, "Failed to unlock LUKS superblock: %m");
 435
 436         r = sym_crypt_activate_by_volume_key(
 437                         cd,
 438                         dm_name,
 439                         vk, vks,
 440                         discard ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0);
 441         if (r < 0)
 442                 return log_error_errno(r, "Failed to unlock LUKS superblock: %m");
 443
 444         log_info("Setting up LUKS device /dev/mapper/%s completed.", dm_name);
 445
 446         *ret = TAKE_PTR(cd);
 447
 448         if (ret_found_uuid) /* Return the UUID actually found if the caller wants to know */
 449                 *ret_found_uuid = p;
 450         if (ret_volume_key)
 451                 *ret_volume_key = TAKE_PTR(vk);
 452         if (ret_volume_key_size)
 453                 *ret_volume_key_size = vks;
 454         if (ret_key_serial)
 455                 *ret_key_serial = TAKE_KEY_SERIAL(key_serial);
 456
 457         return 0;
 458 }
 459
 460 static int make_dm_names(UserRecord *h, HomeSetup *setup) {
 461         assert(h);
 462         assert(h->user_name);
 463         assert(setup);
 464
 465         if (!setup->dm_name) {
 466                 setup->dm_name = strjoin("home-", h->user_name);
 467                 if (!setup->dm_name)
 468                         return log_oom();
 469         }
 470
 471         if (!setup->dm_node) {
 472                 setup->dm_node = path_join("/dev/mapper/", setup->dm_name);
 473                 if (!setup->dm_node)
 474                         return log_oom();
 475         }
 476
 477         return 0;
 478 }
 479
 480 static int acquire_open_luks_device(
 481                 UserRecord *h,
 482                 HomeSetup *setup,
 483                 bool graceful) {
 484
 485         _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
 486         int r;
 487
 488         assert(h);
 489         assert(setup);
 490         assert(!setup->crypt_device);
 491
 492         r = dlopen_cryptsetup();
 493         if (r < 0)
 494                 return r;
 495
 496         r = make_dm_names(h, setup);
 497         if (r < 0)
 498                 return r;
 499
 500         r = sym_crypt_init_by_name(&cd, setup->dm_name);
 501         if ((ERRNO_IS_NEG_DEVICE_ABSENT(r) || r == -EINVAL) && graceful)
 502                 return 0;
 503         if (r < 0)
 504                 return log_error_errno(r, "Failed to initialize cryptsetup context for %s: %m", setup->dm_name);
 505
 506         cryptsetup_enable_logging(cd);
 507
 508         setup->crypt_device = TAKE_PTR(cd);
 509         return 1;
 510 }
 511
 512 static int luks_open(
 513                 UserRecord *h,
 514                 HomeSetup *setup,
 515                 const PasswordCache *cache,
 516                 sd_id128_t *ret_found_uuid,
 517                 void **ret_volume_key,
 518                 size_t *ret_volume_key_size) {
 519
 520         _cleanup_(erase_and_freep) void *vk = NULL;
 521         sd_id128_t p;
 522         char **list;
 523         size_t vks;
 524         int r;
 525
 526         assert(h);
 527         assert(setup);
 528         assert(!setup->crypt_device);
 529
 530         /* Opens a LUKS device that is already set up. Re-validates the password while doing so (which also
 531          * provides us with the volume key, which we want). */
 532
 533         r = acquire_open_luks_device(h, setup, /* graceful= */ false);
 534         if (r < 0)
 535                 return r;
 536
 537         r = sym_crypt_load(setup->crypt_device, CRYPT_LUKS2, NULL);
 538         if (r < 0)
 539                 return log_error_errno(r, "Failed to load LUKS superblock: %m");
 540
 541         r = sym_crypt_get_volume_key_size(setup->crypt_device);
 542         if (r <= 0)
 543                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine LUKS volume key size");
 544         vks = (size_t) r;
 545
 546         if (ret_found_uuid) {
 547                 const char *s;
 548
 549                 s = sym_crypt_get_uuid(setup->crypt_device);
 550                 if (!s)
 551                         return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock has no UUID.");
 552
 553                 r = sd_id128_from_string(s, &p);
 554                 if (r < 0)
 555                         return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "LUKS superblock has invalid UUID.");
 556         }
 557
 558         vk = malloc(vks);
 559         if (!vk)
 560                 return log_oom();
 561
 562         r = -ENOKEY;
 563         FOREACH_POINTER(list,
 564                         cache ? cache->keyring_passswords : NULL,
 565                         cache ? cache->pkcs11_passwords : NULL,
 566                         cache ? cache->fido2_passwords : NULL,
 567                         h->password) {
 568                 r = luks_try_passwords(h, setup->crypt_device, list, vk, &vks, NULL);
 569                 if (r != -ENOKEY)
 570                         break;
 571         }
 572         if (r == -ENOKEY)
 573                 return log_error_errno(r, "No valid password for LUKS superblock.");
 574         if (r < 0)
 575                 return log_error_errno(r, "Failed to unlock LUKS superblock: %m");
 576
 577         log_info("Discovered used LUKS device /dev/mapper/%s, and validated password.", setup->dm_name);
 578
 579         /* This is needed so that crypt_resize() can operate correctly for pre-existing LUKS devices. We need
 580          * to tell libcryptsetup the volume key explicitly, so that it is in the kernel keyring. */
 581         r = sym_crypt_activate_by_volume_key(setup->crypt_device, NULL, vk, vks, CRYPT_ACTIVATE_KEYRING_KEY);
 582         if (r < 0)
 583                 return log_error_errno(r, "Failed to upload volume key again: %m");
 584
 585         log_info("Successfully re-activated LUKS device.");
 586
 587         if (ret_found_uuid)
 588                 *ret_found_uuid = p;
 589         if (ret_volume_key)
 590                 *ret_volume_key = TAKE_PTR(vk);
 591         if (ret_volume_key_size)
 592                 *ret_volume_key_size = vks;
 593
 594         return 0;
 595 }
 596
 597 static int fs_validate(
 598                 const char *dm_node,
 599                 sd_id128_t uuid,
 600                 char **ret_fstype,
 601                 sd_id128_t *ret_found_uuid) {
 602
 603         _cleanup_free_ char *fstype = NULL;
 604         sd_id128_t u = SD_ID128_NULL; /* avoid false maybe-unitialized warning */
 605         int r;
 606
 607         assert(dm_node);
 608         assert(ret_fstype);
 609
 610         r = probe_file_system_by_path(dm_node, &fstype, &u);
 611         if (r < 0)
 612                 return log_error_errno(r, "Failed to probe file system: %m");
 613
 614         /* Limit the set of supported file systems a bit, as protection against little tested kernel file
 615          * systems. Also, we only support the resize ioctls for these file systems. */
 616         if (!supported_fstype(fstype))
 617                 return log_error_errno(SYNTHETIC_ERRNO(EPROTONOSUPPORT), "Image contains unsupported file system: %s", strna(fstype));
 618
 619         if (!sd_id128_is_null(uuid) &&
 620             !sd_id128_equal(uuid, u))
 621                 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE), "File system has wrong UUID.");
 622
 623         log_info("Probing file system completed (found %s).", fstype);
 624
 625         *ret_fstype = TAKE_PTR(fstype);
 626
 627         if (ret_found_uuid) /* Return the UUID actually found if the caller wants to know */
 628                 *ret_found_uuid = u;
 629
 630         return 0;
 631 }
 632
 633 static int luks_validate(
 634                 int fd,
 635                 const char *label,
 636                 sd_id128_t partition_uuid,
 637                 sd_id128_t *ret_partition_uuid,
 638                 uint64_t *ret_offset,
 639                 uint64_t *ret_size) {
 640
 641         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
 642         sd_id128_t found_partition_uuid = SD_ID128_NULL;
 643         const char *fstype = NULL, *pttype = NULL;
 644         blkid_loff_t offset = 0, size = 0;
 645         blkid_partlist pl;
 646         bool found = false;
 647         int r, n;
 648
 649         assert(fd >= 0);
 650         assert(label);
 651         assert(ret_offset);
 652         assert(ret_size);
 653
 654         b = blkid_new_probe();
 655         if (!b)
 656                 return -ENOMEM;
 657
 658         errno = 0;
 659         r = blkid_probe_set_device(b, fd, 0, 0);
 660         if (r != 0)
 661                 return errno_or_else(ENOMEM);
 662
 663         (void) blkid_probe_enable_superblocks(b, 1);
 664         (void) blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
 665         (void) blkid_probe_enable_partitions(b, 1);
 666         (void) blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
 667
 668         errno = 0;
 669         r = blkid_do_safeprobe(b);
 670         if (r == _BLKID_SAFEPROBE_ERROR)
 671                 return errno_or_else(EIO);
 672         if (IN_SET(r, _BLKID_SAFEPROBE_AMBIGUOUS, _BLKID_SAFEPROBE_NOT_FOUND))
 673                 return -ENOPKG;
 674
 675         assert(r == _BLKID_SAFEPROBE_FOUND);
 676
 677         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
 678         if (streq_ptr(fstype, "crypto_LUKS")) {
 679                 /* Directly a LUKS image */
 680                 *ret_offset = 0;
 681                 *ret_size = UINT64_MAX; /* full disk */
 682                 *ret_partition_uuid = SD_ID128_NULL;
 683                 return 0;
 684         } else if (fstype)
 685                 return -ENOPKG;
 686
 687         (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
 688         if (!streq_ptr(pttype, "gpt"))
 689                 return -ENOPKG;
 690
 691         errno = 0;
 692         pl = blkid_probe_get_partitions(b);
 693         if (!pl)
 694                 return errno_or_else(ENOMEM);
 695
 696         errno = 0;
 697         n = blkid_partlist_numof_partitions(pl);
 698         if (n < 0)
 699                 return errno_or_else(EIO);
 700
 701         for (int i = 0; i < n; i++) {
 702                 sd_id128_t id = SD_ID128_NULL;
 703                 blkid_partition pp;
 704
 705                 errno = 0;
 706                 pp = blkid_partlist_get_partition(pl, i);
 707                 if (!pp)
 708                         return errno_or_else(EIO);
 709
 710                 if (sd_id128_string_equal(blkid_partition_get_type_string(pp), SD_GPT_USER_HOME) <= 0)
 711                         continue;
 712
 713                 if (!streq_ptr(blkid_partition_get_name(pp), label))
 714                         continue;
 715
 716
 717                 r = blkid_partition_get_uuid_id128(pp, &id);
 718                 if (r < 0)
 719                         log_debug_errno(r, "Failed to read partition UUID, ignoring: %m");
 720                 else if (!sd_id128_is_null(partition_uuid) && !sd_id128_equal(id, partition_uuid))
 721                         continue;
 722
 723                 if (found)
 724                         return -ENOPKG;
 725
 726                 offset = blkid_partition_get_start(pp);
 727                 size = blkid_partition_get_size(pp);
 728                 found_partition_uuid = id;
 729
 730                 found = true;
 731         }
 732
 733         if (!found)
 734                 return -ENOPKG;
 735
 736         if (offset < 0)
 737                 return -EINVAL;
 738         if ((uint64_t) offset > UINT64_MAX / 512U)
 739                 return -EINVAL;
 740         if (size <= 0)
 741                 return -EINVAL;
 742         if ((uint64_t) size > UINT64_MAX / 512U)
 743                 return -EINVAL;
 744
 745         *ret_offset = offset * 512U;
 746         *ret_size = size * 512U;
 747         *ret_partition_uuid = found_partition_uuid;
 748
 749         return 0;
 750 }
 751
 752 static int crypt_device_to_evp_cipher(struct crypt_device *cd, const EVP_CIPHER **ret) {
 753         _cleanup_free_ char *cipher_name = NULL;
 754         const char *cipher, *cipher_mode, *e;
 755         size_t key_size, key_bits;
 756         const EVP_CIPHER *cc;
 757         int r;
 758
 759         assert(cd);
 760
 761         /* Let's find the right OpenSSL EVP_CIPHER object that matches the encryption settings of the LUKS
 762          * device */
 763
 764         cipher = sym_crypt_get_cipher(cd);
 765         if (!cipher)
 766                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot get cipher from LUKS device.");
 767
 768         cipher_mode = sym_crypt_get_cipher_mode(cd);
 769         if (!cipher_mode)
 770                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot get cipher mode from LUKS device.");
 771
 772         e = strchr(cipher_mode, '-');
 773         if (e)
 774                 cipher_mode = strndupa_safe(cipher_mode, e - cipher_mode);
 775
 776         r = sym_crypt_get_volume_key_size(cd);
 777         if (r <= 0)
 778                 return log_error_errno(r < 0 ? r : SYNTHETIC_ERRNO(EINVAL), "Cannot get volume key size from LUKS device.");
 779
 780         key_size = r;
 781         key_bits = key_size * 8;
 782         if (streq(cipher_mode, "xts"))
 783                 key_bits /= 2;
 784
 785         if (asprintf(&cipher_name, "%s-%zu-%s", cipher, key_bits, cipher_mode) < 0)
 786                 return log_oom();
 787
 788         cc = EVP_get_cipherbyname(cipher_name);
 789         if (!cc)
 790                 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "Selected cipher mode '%s' not supported, can't encrypt JSON record.", cipher_name);
 791
 792         /* Verify that our key length calculations match what OpenSSL thinks */
 793         r = EVP_CIPHER_key_length(cc);
 794         if (r < 0 || (uint64_t) r != key_size)
 795                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Key size of selected cipher doesn't meet our expectations.");
 796
 797         *ret = cc;
 798         return 0;
 799 }
 800
 801 static int luks_validate_home_record(
 802                 struct crypt_device *cd,
 803                 UserRecord *h,
 804                 const void *volume_key,
 805                 PasswordCache *cache,
 806                 UserRecord **ret_luks_home_record) {
 807
 808         int r;
 809
 810         assert(cd);
 811         assert(h);
 812
 813         for (int token = 0; token < sym_crypt_token_max(CRYPT_LUKS2); token++) {
 814                 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL, *rr = NULL;
 815                 _cleanup_(EVP_CIPHER_CTX_freep) EVP_CIPHER_CTX *context = NULL;
 816                 _cleanup_(user_record_unrefp) UserRecord *lhr = NULL;
 817                 _cleanup_free_ void *encrypted = NULL, *iv = NULL;
 818                 size_t decrypted_size, encrypted_size, iv_size;
 819                 int decrypted_size_out1, decrypted_size_out2;
 820                 _cleanup_free_ char *decrypted = NULL;
 821                 const char *text, *type;
 822                 crypt_token_info state;
 823                 JsonVariant *jr, *jiv;
 824                 unsigned line, column;
 825                 const EVP_CIPHER *cc;
 826
 827                 state = sym_crypt_token_status(cd, token, &type);
 828                 if (state == CRYPT_TOKEN_INACTIVE) /* First unconfigured token, give up */
 829                         break;
 830                 if (IN_SET(state, CRYPT_TOKEN_INTERNAL, CRYPT_TOKEN_INTERNAL_UNKNOWN, CRYPT_TOKEN_EXTERNAL))
 831                         continue;
 832                 if (state != CRYPT_TOKEN_EXTERNAL_UNKNOWN)
 833                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unexpected token state of token %i: %i", token, (int) state);
 834
 835                 if (!streq(type, "systemd-homed"))
 836                         continue;
 837
 838                 r = sym_crypt_token_json_get(cd, token, &text);
 839                 if (r < 0)
 840                         return log_error_errno(r, "Failed to read LUKS token %i: %m", token);
 841
 842                 r = json_parse(text, JSON_PARSE_SENSITIVE, &v, &line, &column);
 843                 if (r < 0)
 844                         return log_error_errno(r, "Failed to parse LUKS token JSON data %u:%u: %m", line, column);
 845
 846                 jr = json_variant_by_key(v, "record");
 847                 if (!jr)
 848                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "LUKS token lacks 'record' field.");
 849                 jiv = json_variant_by_key(v, "iv");
 850                 if (!jiv)
 851                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "LUKS token lacks 'iv' field.");
 852
 853                 r = json_variant_unbase64(jr, &encrypted, &encrypted_size);
 854                 if (r < 0)
 855                         return log_error_errno(r, "Failed to base64 decode record: %m");
 856
 857                 r = json_variant_unbase64(jiv, &iv, &iv_size);
 858                 if (r < 0)
 859                         return log_error_errno(r, "Failed to base64 decode IV: %m");
 860
 861                 r = crypt_device_to_evp_cipher(cd, &cc);
 862                 if (r < 0)
 863                         return r;
 864                 if (iv_size > INT_MAX || EVP_CIPHER_iv_length(cc) != (int) iv_size)
 865                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "IV size doesn't match.");
 866
 867                 context = EVP_CIPHER_CTX_new();
 868                 if (!context)
 869                         return log_oom();
 870
 871                 if (EVP_DecryptInit_ex(context, cc, NULL, volume_key, iv) != 1)
 872                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to initialize decryption context.");
 873
 874                 decrypted_size = encrypted_size + EVP_CIPHER_key_length(cc) * 2;
 875                 decrypted = new(char, decrypted_size);
 876                 if (!decrypted)
 877                         return log_oom();
 878
 879                 if (EVP_DecryptUpdate(context, (uint8_t*) decrypted, &decrypted_size_out1, encrypted, encrypted_size) != 1)
 880                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to decrypt JSON record.");
 881
 882                 assert((size_t) decrypted_size_out1 <= decrypted_size);
 883
 884                 if (EVP_DecryptFinal_ex(context, (uint8_t*) decrypted + decrypted_size_out1, &decrypted_size_out2) != 1)
 885                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to finish decryption of JSON record.");
 886
 887                 assert((size_t) decrypted_size_out1 + (size_t) decrypted_size_out2 < decrypted_size);
 888                 decrypted_size = (size_t) decrypted_size_out1 + (size_t) decrypted_size_out2;
 889
 890                 if (memchr(decrypted, 0, decrypted_size))
 891                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Inner NUL byte in JSON record, refusing.");
 892
 893                 decrypted[decrypted_size] = 0;
 894
 895                 r = json_parse(decrypted, JSON_PARSE_SENSITIVE, &rr, NULL, NULL);
 896                 if (r < 0)
 897                         return log_error_errno(r, "Failed to parse decrypted JSON record, refusing.");
 898
 899                 lhr = user_record_new();
 900                 if (!lhr)
 901                         return log_oom();
 902
 903                 r = user_record_load(lhr, rr, USER_RECORD_LOAD_EMBEDDED|USER_RECORD_PERMISSIVE);
 904                 if (r < 0)
 905                         return log_error_errno(r, "Failed to parse user record: %m");
 906
 907                 if (!user_record_compatible(h, lhr))
 908                         return log_error_errno(SYNTHETIC_ERRNO(EREMCHG), "LUKS home record not compatible with host record, refusing.");
 909
 910                 r = user_record_authenticate(lhr, h, cache, /* strict_verify= */ true);
 911                 if (r < 0)
 912                         return r;
 913                 assert(r > 0); /* Insist that a password was verified */
 914
 915                 *ret_luks_home_record = TAKE_PTR(lhr);
 916                 return 0;
 917         }
 918
 919         return log_error_errno(SYNTHETIC_ERRNO(EBADMSG), "Couldn't find home record in LUKS2 header, refusing.");
 920 }
 921
 922 static int format_luks_token_text(
 923                 struct crypt_device *cd,
 924                 UserRecord *hr,
 925                 const void *volume_key,
 926                 char **ret) {
 927
 928         int r, encrypted_size_out1 = 0, encrypted_size_out2 = 0, iv_size, key_size;
 929         _cleanup_(EVP_CIPHER_CTX_freep) EVP_CIPHER_CTX *context = NULL;
 930         _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
 931         _cleanup_free_ void *iv = NULL, *encrypted = NULL;
 932         size_t text_length, encrypted_size;
 933         _cleanup_free_ char *text = NULL;
 934         const EVP_CIPHER *cc;
 935
 936         assert(cd);
 937         assert(hr);
 938         assert(volume_key);
 939         assert(ret);
 940
 941         r = crypt_device_to_evp_cipher(cd, &cc);
 942         if (r < 0)
 943                 return r;
 944
 945         key_size = EVP_CIPHER_key_length(cc);
 946         iv_size = EVP_CIPHER_iv_length(cc);
 947
 948         if (iv_size > 0) {
 949                 iv = malloc(iv_size);
 950                 if (!iv)
 951                         return log_oom();
 952
 953                 r = crypto_random_bytes(iv, iv_size);
 954                 if (r < 0)
 955                         return log_error_errno(r, "Failed to generate IV: %m");
 956         }
 957
 958         context = EVP_CIPHER_CTX_new();
 959         if (!context)
 960                 return log_oom();
 961
 962         if (EVP_EncryptInit_ex(context, cc, NULL, volume_key, iv) != 1)
 963                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to initialize encryption context.");
 964
 965         r = json_variant_format(hr->json, 0, &text);
 966         if (r < 0)
 967                 return log_error_errno(r, "Failed to format user record for LUKS: %m");
 968
 969         text_length = strlen(text);
 970         encrypted_size = text_length + 2*key_size - 1;
 971
 972         encrypted = malloc(encrypted_size);
 973         if (!encrypted)
 974                 return log_oom();
 975
 976         if (EVP_EncryptUpdate(context, encrypted, &encrypted_size_out1, (uint8_t*) text, text_length) != 1)
 977                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to encrypt JSON record.");
 978
 979         assert((size_t) encrypted_size_out1 <= encrypted_size);
 980
 981         if (EVP_EncryptFinal_ex(context, (uint8_t*) encrypted + encrypted_size_out1, &encrypted_size_out2) != 1)
 982                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to finish encryption of JSON record. ");
 983
 984         assert((size_t) encrypted_size_out1 + (size_t) encrypted_size_out2 <= encrypted_size);
 985
 986         r = json_build(&v,
 987                        JSON_BUILD_OBJECT(
 988                                        JSON_BUILD_PAIR("type", JSON_BUILD_CONST_STRING("systemd-homed")),
 989                                        JSON_BUILD_PAIR("keyslots", JSON_BUILD_EMPTY_ARRAY),
 990                                        JSON_BUILD_PAIR("record", JSON_BUILD_BASE64(encrypted, encrypted_size_out1 + encrypted_size_out2)),
 991                                        JSON_BUILD_PAIR("iv", JSON_BUILD_BASE64(iv, iv_size))));
 992         if (r < 0)
 993                 return log_error_errno(r, "Failed to prepare LUKS JSON token object: %m");
 994
 995         r = json_variant_format(v, 0, ret);
 996         if (r < 0)
 997                 return log_error_errno(r, "Failed to format encrypted user record for LUKS: %m");
 998
 999         return 0;
1000 }
1001
1002 int home_store_header_identity_luks(
1003                 UserRecord *h,
1004                 HomeSetup *setup,
1005                 UserRecord *old_home) {
1006
1007         _cleanup_(user_record_unrefp) UserRecord *header_home = NULL;
1008         _cleanup_free_ char *text = NULL;
1009         int r;
1010
1011         assert(h);
1012
1013         if (!setup->crypt_device)
1014                 return 0;
1015
1016         assert(setup->volume_key);
1017
1018         /* Let's store the user's identity record in the LUKS2 "token" header data fields, in an encrypted
1019          * fashion. Why that? If we'd rely on the record being embedded in the payload file system itself we
1020          * would have to mount the file system before we can validate the JSON record, its signatures and
1021          * whether it matches what we are looking for. However, kernel file system implementations are
1022          * generally not ready to be used on untrusted media. Hence let's store the record independently of
1023          * the file system, so that we can validate it first, and only then mount the file system. To keep
1024          * things simple we use the same encryption settings for this record as for the file system itself. */
1025
1026         r = user_record_clone(h, USER_RECORD_EXTRACT_EMBEDDED|USER_RECORD_PERMISSIVE, &header_home);
1027         if (r < 0)
1028                 return log_error_errno(r, "Failed to determine new header record: %m");
1029
1030         if (old_home && user_record_equal(old_home, header_home)) {
1031                 log_debug("Not updating header home record.");
1032                 return 0;
1033         }
1034
1035         r = format_luks_token_text(setup->crypt_device, header_home, setup->volume_key, &text);
1036         if (r < 0)
1037                 return r;
1038
1039         for (int token = 0; token < sym_crypt_token_max(CRYPT_LUKS2); token++) {
1040                 crypt_token_info state;
1041                 const char *type;
1042
1043                 state = sym_crypt_token_status(setup->crypt_device, token, &type);
1044                 if (state == CRYPT_TOKEN_INACTIVE) /* First unconfigured token, we are done */
1045                         break;
1046                 if (IN_SET(state, CRYPT_TOKEN_INTERNAL, CRYPT_TOKEN_INTERNAL_UNKNOWN, CRYPT_TOKEN_EXTERNAL))
1047                         continue; /* Not ours */
1048                 if (state != CRYPT_TOKEN_EXTERNAL_UNKNOWN)
1049                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unexpected token state of token %i: %i", token, (int) state);
1050
1051                 if (!streq(type, "systemd-homed"))
1052                         continue;
1053
1054                 r = sym_crypt_token_json_set(setup->crypt_device, token, text);
1055                 if (r < 0)
1056                         return log_error_errno(r, "Failed to set JSON token for slot %i: %m", token);
1057
1058                 /* Now, let's free the text so that for all further matching tokens we all crypt_json_token_set()
1059                  * with a NULL text in order to invalidate the tokens. */
1060                 text = mfree(text);
1061         }
1062
1063         if (text)
1064                 return log_error_errno(SYNTHETIC_ERRNO(EBADMSG), "Didn't find any record token to update.");
1065
1066         log_info("Wrote LUKS header user record.");
1067
1068         return 1;
1069 }
1070
1071 int run_fitrim(int root_fd) {
1072         struct fstrim_range range = {
1073                 .len = UINT64_MAX,
1074         };
1075
1076         /* If discarding is on, discard everything right after mounting, so that the discard setting takes
1077          * effect on activation. (Also, optionally, trim on logout) */
1078
1079         assert(root_fd >= 0);
1080
1081         if (ioctl(root_fd, FITRIM, &range) < 0) {
1082                 if (ERRNO_IS_NOT_SUPPORTED(errno) || errno == EBADF) {
1083                         log_debug_errno(errno, "File system does not support FITRIM, not trimming.");
1084                         return 0;
1085                 }
1086
1087                 return log_warning_errno(errno, "Failed to invoke FITRIM, ignoring: %m");
1088         }
1089
1090         log_info("Discarded unused %s.", FORMAT_BYTES(range.len));
1091         return 1;
1092 }
1093
1094 int run_fallocate(int backing_fd, const struct stat *st) {
1095         struct stat stbuf;
1096
1097         assert(backing_fd >= 0);
1098
1099         /* If discarding is off, let's allocate the whole image before mounting, so that the setting takes
1100          * effect on activation */
1101
1102         if (!st) {
1103                 if (fstat(backing_fd, &stbuf) < 0)
1104                         return log_error_errno(errno, "Failed to fstat(): %m");
1105
1106                 st = &stbuf;
1107         }
1108
1109         if (!S_ISREG(st->st_mode))
1110                 return 0;
1111
1112         if (st->st_blocks >= DIV_ROUND_UP(st->st_size, 512)) {
1113                 log_info("Backing file is fully allocated already.");
1114                 return 0;
1115         }
1116
1117         if (fallocate(backing_fd, FALLOC_FL_KEEP_SIZE, 0, st->st_size) < 0) {
1118
1119                 if (ERRNO_IS_NOT_SUPPORTED(errno)) {
1120                         log_debug_errno(errno, "fallocate() not supported on file system, ignoring.");
1121                         return 0;
1122                 }
1123
1124                 if (ERRNO_IS_DISK_SPACE(errno)) {
1125                         log_debug_errno(errno, "Not enough disk space to fully allocate home.");
1126                         return -ENOSPC; /* make recognizable */
1127                 }
1128
1129                 return log_error_errno(errno, "Failed to allocate backing file blocks: %m");
1130         }
1131
1132         log_info("Allocated additional %s.",
1133                  FORMAT_BYTES((DIV_ROUND_UP(st->st_size, 512) - st->st_blocks) * 512));
1134         return 1;
1135 }
1136
1137 int run_fallocate_by_path(const char *backing_path) {
1138         _cleanup_close_ int backing_fd = -EBADF;
1139
1140         backing_fd = open(backing_path, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
1141         if (backing_fd < 0)
1142                 return log_error_errno(errno, "Failed to open '%s' for fallocate(): %m", backing_path);
1143
1144         return run_fallocate(backing_fd, NULL);
1145 }
1146
1147 static int lock_image_fd(int image_fd, const char *ip) {
1148         int r;
1149
1150         /* If the $SYSTEMD_LUKS_LOCK environment variable is set we'll take an exclusive BSD lock on the
1151          * image file, and send it to our parent. homed will keep it open to ensure no other instance of
1152          * homed (across the network or such) will also mount the file. */
1153
1154         assert(image_fd >= 0);
1155         assert(ip);
1156
1157         r = getenv_bool("SYSTEMD_LUKS_LOCK");
1158         if (r == -ENXIO)
1159                 return 0;
1160         if (r < 0)
1161                 return log_error_errno(r, "Failed to parse $SYSTEMD_LUKS_LOCK environment variable: %m");
1162         if (r == 0)
1163                 return 0;
1164
1165         if (flock(image_fd, LOCK_EX|LOCK_NB) < 0) {
1166
1167                 if (errno == EAGAIN)
1168                         log_error_errno(errno, "Image file '%s' already locked, can't use.", ip);
1169                 else
1170                         log_error_errno(errno, "Failed to lock image file '%s': %m", ip);
1171
1172                 return errno != EAGAIN ? -errno : -EADDRINUSE; /* Make error recognizable */
1173         }
1174
1175         log_info("Successfully locked image file '%s'.", ip);
1176
1177         /* Now send it to our parent to keep safe while the home dir is active */
1178         r = sd_pid_notify_with_fds(0, false, "SYSTEMD_LUKS_LOCK_FD=1", &image_fd, 1);
1179         if (r < 0)
1180                 log_warning_errno(r, "Failed to send LUKS lock fd to parent, ignoring: %m");
1181
1182         return 0;
1183 }
1184
1185 static int open_image_file(
1186                 UserRecord *h,
1187                 const char *force_image_path,
1188                 struct stat *ret_stat) {
1189
1190         _cleanup_close_ int image_fd = -EBADF;
1191         struct stat st;
1192         const char *ip;
1193         int r;
1194
1195         assert(h || force_image_path);
1196
1197         ip = force_image_path ?: user_record_image_path(h);
1198
1199         image_fd = open(ip, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
1200         if (image_fd < 0)
1201                 return log_error_errno(errno, "Failed to open image file %s: %m", ip);
1202
1203         if (fstat(image_fd, &st) < 0)
1204                 return log_error_errno(errno, "Failed to fstat() image file: %m");
1205         if (!S_ISREG(st.st_mode) && !S_ISBLK(st.st_mode))
1206                 return log_error_errno(
1207                                 S_ISDIR(st.st_mode) ? SYNTHETIC_ERRNO(EISDIR) : SYNTHETIC_ERRNO(EBADFD),
1208                                 "Image file %s is not a regular file or block device: %m", ip);
1209
1210         /* Locking block devices doesn't really make sense, as this might interfere with
1211          * udev's workings, and these locks aren't network propagated anyway, hence not what
1212          * we are after here. */
1213         if (S_ISREG(st.st_mode)) {
1214                 r = lock_image_fd(image_fd, ip);
1215                 if (r < 0)
1216                         return r;
1217         }
1218
1219         if (ret_stat)
1220                 *ret_stat = st;
1221
1222         return TAKE_FD(image_fd);
1223 }
1224
1225 int home_setup_luks(
1226                 UserRecord *h,
1227                 HomeSetupFlags flags,
1228                 const char *force_image_path,
1229                 HomeSetup *setup,
1230                 PasswordCache *cache,
1231                 UserRecord **ret_luks_home) {
1232
1233         sd_id128_t found_partition_uuid, found_fs_uuid = SD_ID128_NULL, found_luks_uuid = SD_ID128_NULL;
1234         _cleanup_(user_record_unrefp) UserRecord *luks_home = NULL;
1235         _cleanup_(erase_and_freep) void *volume_key = NULL;
1236         size_t volume_key_size = 0;
1237         uint64_t offset, size;
1238         struct stat st;
1239         int r;
1240
1241         assert(h);
1242         assert(setup);
1243         assert(user_record_storage(h) == USER_LUKS);
1244
1245         r = dlopen_cryptsetup();
1246         if (r < 0)
1247                 return r;
1248
1249         r = make_dm_names(h, setup);
1250         if (r < 0)
1251                 return r;
1252
1253         /* Reuse the image fd if it has already been opened by an earlier step */
1254         if (setup->image_fd < 0) {
1255                 setup->image_fd = open_image_file(h, force_image_path, &st);
1256                 if (setup->image_fd < 0)
1257                         return setup->image_fd;
1258         } else if (fstat(setup->image_fd, &st) < 0)
1259                 return log_error_errno(errno, "Failed to stat image: %m");
1260
1261         if (FLAGS_SET(flags, HOME_SETUP_ALREADY_ACTIVATED)) {
1262                 struct loop_info64 info;
1263                 const char *n;
1264
1265                 if (!setup->crypt_device) {
1266                         r = luks_open(h,
1267                                       setup,
1268                                       cache,
1269                                       &found_luks_uuid,
1270                                       &volume_key,
1271                                       &volume_key_size);
1272                         if (r < 0)
1273                                 return r;
1274                 }
1275
1276                 if (ret_luks_home) {
1277                         r = luks_validate_home_record(setup->crypt_device, h, volume_key, cache, &luks_home);
1278                         if (r < 0)
1279                                 return r;
1280                 }
1281
1282                 n = sym_crypt_get_device_name(setup->crypt_device);
1283                 if (!n)
1284                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine backing device for DM %s.", setup->dm_name);
1285
1286                 if (!setup->loop) {
1287                         r = loop_device_open_from_path(n, O_RDWR, LOCK_UN, &setup->loop);
1288                         if (r < 0)
1289                                 return log_error_errno(r, "Failed to open loopback device %s: %m", n);
1290                 }
1291
1292                 if (ioctl(setup->loop->fd, LOOP_GET_STATUS64, &info) < 0) {
1293                         _cleanup_free_ char *sysfs = NULL;
1294
1295                         if (!IN_SET(errno, ENOTTY, EINVAL))
1296                                 return log_error_errno(errno, "Failed to get block device metrics of %s: %m", n);
1297
1298                         if (fstat(setup->loop->fd, &st) < 0)
1299                                 return log_error_errno(r, "Failed to stat block device %s: %m", n);
1300                         assert(S_ISBLK(st.st_mode));
1301
1302                         if (asprintf(&sysfs, "/sys/dev/block/" DEVNUM_FORMAT_STR "/partition", DEVNUM_FORMAT_VAL(st.st_rdev)) < 0)
1303                                 return log_oom();
1304
1305                         if (access(sysfs, F_OK) < 0) {
1306                                 if (errno != ENOENT)
1307                                         return log_error_errno(errno, "Failed to determine whether %s exists: %m", sysfs);
1308
1309                                 offset = 0;
1310                         } else {
1311                                 _cleanup_free_ char *buffer = NULL;
1312
1313                                 if (asprintf(&sysfs, "/sys/dev/block/" DEVNUM_FORMAT_STR "/start", DEVNUM_FORMAT_VAL(st.st_rdev)) < 0)
1314                                         return log_oom();
1315
1316                                 r = read_one_line_file(sysfs, &buffer);
1317                                 if (r < 0)
1318                                         return log_error_errno(r, "Failed to read partition start offset: %m");
1319
1320                                 r = safe_atou64(buffer, &offset);
1321                                 if (r < 0)
1322                                         return log_error_errno(r, "Failed to parse partition start offset: %m");
1323
1324                                 if (offset > UINT64_MAX / 512U)
1325                                         return log_error_errno(SYNTHETIC_ERRNO(E2BIG), "Offset too large for 64 byte range, refusing.");
1326
1327                                 offset *= 512U;
1328                         }
1329
1330                         size = setup->loop->device_size;
1331                 } else {
1332 #if HAVE_VALGRIND_MEMCHECK_H
1333                         VALGRIND_MAKE_MEM_DEFINED(&info, sizeof(info));
1334 #endif
1335
1336                         offset = info.lo_offset;
1337                         size = info.lo_sizelimit;
1338                 }
1339
1340                 found_partition_uuid = found_fs_uuid = SD_ID128_NULL;
1341
1342                 log_info("Discovered used loopback device %s.", setup->loop->node);
1343
1344                 if (setup->root_fd < 0) {
1345                         setup->root_fd = open(user_record_home_directory(h), O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
1346                         if (setup->root_fd < 0)
1347                                 return log_error_errno(errno, "Failed to open home directory: %m");
1348                 }
1349         } else {
1350                 _cleanup_free_ char *fstype = NULL, *subdir = NULL;
1351                 const char *ip;
1352
1353                 /* When we aren't reopening the home directory we are allocating it fresh, hence the relevant
1354                  * objects can't be allocated yet. */
1355                 assert(setup->root_fd < 0);
1356                 assert(!setup->crypt_device);
1357                 assert(!setup->loop);
1358
1359                 ip = force_image_path ?: user_record_image_path(h);
1360
1361                 subdir = path_join(HOME_RUNTIME_WORK_DIR, user_record_user_name_and_realm(h));
1362                 if (!subdir)
1363                         return log_oom();
1364
1365                 r = luks_validate(setup->image_fd, user_record_user_name_and_realm(h), h->partition_uuid, &found_partition_uuid, &offset, &size);
1366                 if (r < 0)
1367                         return log_error_errno(r, "Failed to validate disk label: %m");
1368
1369                 /* Everything before this point left the image untouched. We are now starting to make
1370                  * changes, hence mark the image dirty */
1371                 if (run_mark_dirty(setup->image_fd, true) > 0)
1372                         setup->do_mark_clean = true;
1373
1374                 if (!user_record_luks_discard(h)) {
1375                         r = run_fallocate(setup->image_fd, &st);
1376                         if (r < 0)
1377                                 return r;
1378                 }
1379
1380                 r = loop_device_make(
1381                                 setup->image_fd,
1382                                 O_RDWR,
1383                                 offset,
1384                                 size,
1385                                 h->luks_sector_size == UINT64_MAX ? UINT32_MAX : user_record_luks_sector_size(h), /* if sector size is not specified, select UINT32_MAX, i.e. auto-probe */
1386                                 /* loop_flags= */ 0,
1387                                 LOCK_UN,
1388                                 &setup->loop);
1389                 if (r == -ENOENT) {
1390                         log_error_errno(r, "Loopback block device support is not available on this system.");
1391                         return -ENOLINK; /* make recognizable */
1392                 }
1393                 if (r < 0)
1394                         return log_error_errno(r, "Failed to allocate loopback context: %m");
1395
1396                 log_info("Setting up loopback device %s completed.", setup->loop->node ?: ip);
1397
1398                 r = luks_setup(h,
1399                                setup->loop->node ?: ip,
1400                                setup->dm_name,
1401                                h->luks_uuid,
1402                                h->luks_cipher,
1403                                h->luks_cipher_mode,
1404                                h->luks_volume_key_size,
1405                                h->password,
1406                                cache,
1407                                user_record_luks_discard(h) || user_record_luks_offline_discard(h),
1408                                &setup->crypt_device,
1409                                &found_luks_uuid,
1410                                &volume_key,
1411                                &volume_key_size,
1412                                &setup->key_serial);
1413                 if (r < 0)
1414                         return r;
1415
1416                 setup->undo_dm = true;
1417
1418                 if (ret_luks_home) {
1419                         r = luks_validate_home_record(setup->crypt_device, h, volume_key, cache, &luks_home);
1420                         if (r < 0)
1421                                 return r;
1422                 }
1423
1424                 r = fs_validate(setup->dm_node, h->file_system_uuid, &fstype, &found_fs_uuid);
1425                 if (r < 0)
1426                         return r;
1427
1428                 r = run_fsck(setup->dm_node, fstype);
1429                 if (r < 0)
1430                         return r;
1431
1432                 r = home_unshare_and_mount(setup->dm_node, fstype, user_record_luks_discard(h), user_record_mount_flags(h), h->luks_extra_mount_options);
1433                 if (r < 0)
1434                         return r;
1435
1436                 setup->undo_mount = true;
1437
1438                 setup->root_fd = open(subdir, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
1439                 if (setup->root_fd < 0)
1440                         return log_error_errno(errno, "Failed to open home directory: %m");
1441
1442                 if (user_record_luks_discard(h))
1443                         (void) run_fitrim(setup->root_fd);
1444
1445                 setup->do_offline_fallocate = !(setup->do_offline_fitrim = user_record_luks_offline_discard(h));
1446         }
1447
1448         if (!sd_id128_is_null(found_partition_uuid))
1449                 setup->found_partition_uuid = found_partition_uuid;
1450         if (!sd_id128_is_null(found_luks_uuid))
1451                 setup->found_luks_uuid = found_luks_uuid;
1452         if (!sd_id128_is_null(found_fs_uuid))
1453                 setup->found_fs_uuid = found_fs_uuid;
1454
1455         setup->partition_offset = offset;
1456         setup->partition_size = size;
1457
1458         if (volume_key) {
1459                 erase_and_free(setup->volume_key);
1460                 setup->volume_key = TAKE_PTR(volume_key);
1461                 setup->volume_key_size = volume_key_size;
1462         }
1463
1464         if (ret_luks_home)
1465                 *ret_luks_home = TAKE_PTR(luks_home);
1466
1467         return 0;
1468 }
1469
1470 static void print_size_summary(uint64_t host_size, uint64_t encrypted_size, const struct statfs *sfs) {
1471         assert(sfs);
1472
1473         log_info("Image size is %s, file system size is %s, file system payload size is %s, file system free is %s.",
1474                  FORMAT_BYTES(host_size),
1475                  FORMAT_BYTES(encrypted_size),
1476                  FORMAT_BYTES((uint64_t) sfs->f_blocks * (uint64_t) sfs->f_frsize),
1477                  FORMAT_BYTES((uint64_t) sfs->f_bfree * (uint64_t) sfs->f_frsize));
1478 }
1479
1480 static int home_auto_grow_luks(
1481                 UserRecord *h,
1482                 HomeSetup *setup,
1483                 PasswordCache *cache) {
1484
1485         struct statfs sfs;
1486
1487         assert(h);
1488         assert(setup);
1489
1490         if (!IN_SET(user_record_auto_resize_mode(h), AUTO_RESIZE_GROW, AUTO_RESIZE_SHRINK_AND_GROW))
1491                 return 0;
1492
1493         assert(setup->root_fd >= 0);
1494
1495         if (fstatfs(setup->root_fd, &sfs) < 0)
1496                 return log_error_errno(errno, "Failed to statfs home directory: %m");
1497
1498         if (!fs_can_online_shrink_and_grow(sfs.f_type)) {
1499                 log_debug("Not auto-grow file system, since selected file system cannot do both online shrink and grow.");
1500                 return 0;
1501         }
1502
1503         log_debug("Initiating auto-grow...");
1504
1505         return home_resize_luks(
1506                         h,
1507                         HOME_SETUP_ALREADY_ACTIVATED|
1508                         HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES|
1509                         HOME_SETUP_RESIZE_DONT_SHRINK|
1510                         HOME_SETUP_RESIZE_DONT_UNDO,
1511                         setup,
1512                         cache,
1513                         NULL);
1514 }
1515
1516 int home_activate_luks(
1517                 UserRecord *h,
1518                 HomeSetupFlags flags,
1519                 HomeSetup *setup,
1520                 PasswordCache *cache,
1521                 UserRecord **ret_home) {
1522
1523         _cleanup_(user_record_unrefp) UserRecord *new_home = NULL, *luks_home_record = NULL;
1524         uint64_t host_size, encrypted_size;
1525         const char *hdo, *hd;
1526         struct statfs sfs;
1527         int r;
1528
1529         assert(h);
1530         assert(user_record_storage(h) == USER_LUKS);
1531         assert(setup);
1532         assert(ret_home);
1533
1534         r = dlopen_cryptsetup();
1535         if (r < 0)
1536                 return r;
1537
1538         assert_se(hdo = user_record_home_directory(h));
1539         hd = strdupa_safe(hdo); /* copy the string out, since it might change later in the home record object */
1540
1541         r = home_get_state_luks(h, setup);
1542         if (r < 0)
1543                 return r;
1544         if (r > 0)
1545                 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Device mapper device %s already exists, refusing.", setup->dm_node);
1546
1547         r = home_setup_luks(
1548                         h,
1549                         0,
1550                         NULL,
1551                         setup,
1552                         cache,
1553                         &luks_home_record);
1554         if (r < 0)
1555                 return r;
1556
1557         r = home_auto_grow_luks(h, setup, cache);
1558         if (r < 0)
1559                 return r;
1560
1561         r = block_get_size_by_fd(setup->loop->fd, &host_size);
1562         if (r < 0)
1563                 return log_error_errno(r, "Failed to get loopback block device size: %m");
1564
1565         r = block_get_size_by_path(setup->dm_node, &encrypted_size);
1566         if (r < 0)
1567                 return log_error_errno(r, "Failed to get LUKS block device size: %m");
1568
1569         r = home_refresh(
1570                         h,
1571                         flags,
1572                         setup,
1573                         luks_home_record,
1574                         cache,
1575                         &sfs,
1576                         &new_home);
1577         if (r < 0)
1578                 return r;
1579
1580         r = home_extend_embedded_identity(new_home, h, setup);
1581         if (r < 0)
1582                 return r;
1583
1584         setup->root_fd = safe_close(setup->root_fd);
1585
1586         r = home_move_mount(user_record_user_name_and_realm(h), hd);
1587         if (r < 0)
1588                 return r;
1589
1590         setup->undo_mount = false;
1591         setup->do_offline_fitrim = false;
1592
1593         loop_device_relinquish(setup->loop);
1594
1595         r = sym_crypt_deactivate_by_name(NULL, setup->dm_name, CRYPT_DEACTIVATE_DEFERRED);
1596         if (r < 0)
1597                 log_warning_errno(r, "Failed to relinquish DM device, ignoring: %m");
1598
1599         setup->undo_dm = false;
1600         setup->do_offline_fallocate = false;
1601         setup->do_mark_clean = false;
1602         setup->do_drop_caches = false;
1603         TAKE_KEY_SERIAL(setup->key_serial); /* Leave key in kernel keyring */
1604
1605         log_info("Activation completed.");
1606
1607         print_size_summary(host_size, encrypted_size, &sfs);
1608
1609         *ret_home = TAKE_PTR(new_home);
1610         return 1;
1611 }
1612
1613 int home_deactivate_luks(UserRecord *h, HomeSetup *setup) {
1614         bool we_detached = false;
1615         int r;
1616
1617         assert(h);
1618         assert(setup);
1619
1620         /* Note that the DM device and loopback device are set to auto-detach, hence strictly speaking we
1621          * don't have to explicitly have to detach them. However, we do that nonetheless (in case of the DM
1622          * device), to avoid races: by explicitly detaching them we know when the detaching is complete. We
1623          * don't bother about the loopback device because unlike the DM device it doesn't have a fixed
1624          * name. */
1625
1626         if (!setup->crypt_device) {
1627                 r = acquire_open_luks_device(h, setup, /* graceful= */ true);
1628                 if (r < 0)
1629                         return log_error_errno(r, "Failed to initialize cryptsetup context for %s: %m", setup->dm_name);
1630                 if (r == 0)
1631                         log_debug("LUKS device %s has already been detached.", setup->dm_name);
1632         }
1633
1634         if (setup->crypt_device) {
1635                 log_info("Discovered used LUKS device %s.", setup->dm_node);
1636
1637                 cryptsetup_enable_logging(setup->crypt_device);
1638
1639                 r = sym_crypt_deactivate_by_name(setup->crypt_device, setup->dm_name, 0);
1640                 if (ERRNO_IS_NEG_DEVICE_ABSENT(r) || r == -EINVAL)
1641                         log_debug_errno(r, "LUKS device %s is already detached.", setup->dm_node);
1642                 else if (r < 0)
1643                         return log_info_errno(r, "LUKS device %s couldn't be deactivated: %m", setup->dm_node);
1644                 else {
1645                         log_info("LUKS device detaching completed.");
1646                         we_detached = true;
1647                 }
1648         }
1649
1650         (void) wait_for_block_device_gone(setup, USEC_PER_SEC * 30);
1651         setup->undo_dm = false;
1652
1653         if (user_record_luks_offline_discard(h))
1654                 log_debug("Not allocating on logout.");
1655         else
1656                 (void) run_fallocate_by_path(user_record_image_path(h));
1657
1658         run_mark_dirty_by_path(user_record_image_path(h), false);
1659         return we_detached;
1660 }
1661
1662 int home_trim_luks(UserRecord *h, HomeSetup *setup) {
1663         assert(h);
1664         assert(setup);
1665         assert(setup->root_fd >= 0);
1666
1667         if (!user_record_luks_offline_discard(h)) {
1668                 log_debug("Not trimming on logout.");
1669                 return 0;
1670         }
1671
1672         (void) run_fitrim(setup->root_fd);
1673         return 0;
1674 }
1675
1676 static struct crypt_pbkdf_type* build_good_pbkdf(struct crypt_pbkdf_type *buffer, UserRecord *hr) {
1677         assert(buffer);
1678         assert(hr);
1679
1680         bool benchmark = user_record_luks_pbkdf_force_iterations(hr) == UINT64_MAX;
1681
1682         *buffer = (struct crypt_pbkdf_type) {
1683                 .hash = user_record_luks_pbkdf_hash_algorithm(hr),
1684                 .type = user_record_luks_pbkdf_type(hr),
1685                 .time_ms = benchmark ? user_record_luks_pbkdf_time_cost_usec(hr) / USEC_PER_MSEC : 0,
1686                 .iterations = benchmark ? 0 : user_record_luks_pbkdf_force_iterations(hr),
1687                 .max_memory_kb = user_record_luks_pbkdf_memory_cost(hr) / 1024,
1688                 .parallel_threads = user_record_luks_pbkdf_parallel_threads(hr),
1689                 .flags = benchmark ? 0 : CRYPT_PBKDF_NO_BENCHMARK,
1690         };
1691
1692         return buffer;
1693 }
1694
1695 static struct crypt_pbkdf_type* build_minimal_pbkdf(struct crypt_pbkdf_type *buffer, UserRecord *hr) {
1696         assert(buffer);
1697         assert(hr);
1698
1699         /* For PKCS#11 derived keys (which are generated randomly and are of high quality already) we use a
1700          * minimal PBKDF */
1701         *buffer = (struct crypt_pbkdf_type) {
1702                 .hash = user_record_luks_pbkdf_hash_algorithm(hr),
1703                 .type = CRYPT_KDF_PBKDF2,
1704                 .iterations = 1,
1705                 .time_ms = 1,
1706         };
1707
1708         return buffer;
1709 }
1710
1711 static int luks_format(
1712                 const char *node,
1713                 const char *dm_name,
1714                 sd_id128_t uuid,
1715                 const char *label,
1716                 const PasswordCache *cache,
1717                 char **effective_passwords,
1718                 bool discard,
1719                 UserRecord *hr,
1720                 struct crypt_device **ret) {
1721
1722         _cleanup_(user_record_unrefp) UserRecord *reduced = NULL;
1723         _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1724         _cleanup_(erase_and_freep) void *volume_key = NULL;
1725         struct crypt_pbkdf_type good_pbkdf, minimal_pbkdf;
1726         _cleanup_free_ char *text = NULL;
1727         size_t volume_key_size;
1728         int slot = 0, r;
1729
1730         assert(node);
1731         assert(dm_name);
1732         assert(hr);
1733         assert(ret);
1734
1735         r = sym_crypt_init(&cd, node);
1736         if (r < 0)
1737                 return log_error_errno(r, "Failed to allocate libcryptsetup context: %m");
1738
1739         cryptsetup_enable_logging(cd);
1740
1741         /* Normally we'd, just leave volume key generation to libcryptsetup. However, we can't, since we
1742          * can't extract the volume key from the library again, but we need it in order to encrypt the JSON
1743          * record. Hence, let's generate it on our own, so that we can keep track of it. */
1744
1745         volume_key_size = user_record_luks_volume_key_size(hr);
1746         volume_key = malloc(volume_key_size);
1747         if (!volume_key)
1748                 return log_oom();
1749
1750         r = crypto_random_bytes(volume_key, volume_key_size);
1751         if (r < 0)
1752                 return log_error_errno(r, "Failed to generate volume key: %m");
1753
1754 #if HAVE_CRYPT_SET_METADATA_SIZE
1755         /* Increase the metadata space to 4M, the largest LUKS2 supports */
1756         r = sym_crypt_set_metadata_size(cd, 4096U*1024U, 0);
1757         if (r < 0)
1758                 return log_error_errno(r, "Failed to change LUKS2 metadata size: %m");
1759 #endif
1760
1761         build_good_pbkdf(&good_pbkdf, hr);
1762         build_minimal_pbkdf(&minimal_pbkdf, hr);
1763
1764         r = sym_crypt_format(
1765                         cd,
1766                         CRYPT_LUKS2,
1767                         user_record_luks_cipher(hr),
1768                         user_record_luks_cipher_mode(hr),
1769                         SD_ID128_TO_UUID_STRING(uuid),
1770                         volume_key,
1771                         volume_key_size,
1772                         &(struct crypt_params_luks2) {
1773                                 .label = label,
1774                                 .subsystem = "systemd-home",
1775                                 .sector_size = user_record_luks_sector_size(hr),
1776                                 .pbkdf = &good_pbkdf,
1777                         });
1778         if (r < 0)
1779                 return log_error_errno(r, "Failed to format LUKS image: %m");
1780
1781         log_info("LUKS formatting completed.");
1782
1783         STRV_FOREACH(pp, effective_passwords) {
1784
1785                 if (password_cache_contains(cache, *pp)) { /* is this a fido2 or pkcs11 password? */
1786                         log_debug("Using minimal PBKDF for slot %i", slot);
1787                         r = sym_crypt_set_pbkdf_type(cd, &minimal_pbkdf);
1788                 } else {
1789                         log_debug("Using good PBKDF for slot %i", slot);
1790                         r = sym_crypt_set_pbkdf_type(cd, &good_pbkdf);
1791                 }
1792                 if (r < 0)
1793                         return log_error_errno(r, "Failed to tweak PBKDF for slot %i: %m", slot);
1794
1795                 r = sym_crypt_keyslot_add_by_volume_key(
1796                                 cd,
1797                                 slot,
1798                                 volume_key,
1799                                 volume_key_size,
1800                                 *pp,
1801                                 strlen(*pp));
1802                 if (r < 0)
1803                         return log_error_errno(r, "Failed to set up LUKS password for slot %i: %m", slot);
1804
1805                 log_info("Writing password to LUKS keyslot %i completed.", slot);
1806                 slot++;
1807         }
1808
1809         r = sym_crypt_activate_by_volume_key(
1810                         cd,
1811                         dm_name,
1812                         volume_key,
1813                         volume_key_size,
1814                         discard ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0);
1815         if (r < 0)
1816                 return log_error_errno(r, "Failed to activate LUKS superblock: %m");
1817
1818         log_info("LUKS activation by volume key succeeded.");
1819
1820         r = user_record_clone(hr, USER_RECORD_EXTRACT_EMBEDDED|USER_RECORD_PERMISSIVE, &reduced);
1821         if (r < 0)
1822                 return log_error_errno(r, "Failed to prepare home record for LUKS: %m");
1823
1824         r = format_luks_token_text(cd, reduced, volume_key, &text);
1825         if (r < 0)
1826                 return r;
1827
1828         r = sym_crypt_token_json_set(cd, CRYPT_ANY_TOKEN, text);
1829         if (r < 0)
1830                 return log_error_errno(r, "Failed to set LUKS JSON token: %m");
1831
1832         log_info("Writing user record as LUKS token completed.");
1833
1834         if (ret)
1835                 *ret = TAKE_PTR(cd);
1836
1837         return 0;
1838 }
1839
1840 static int make_partition_table(
1841                 int fd,
1842                 uint32_t sector_size,
1843                 const char *label,
1844                 sd_id128_t uuid,
1845                 uint64_t *ret_offset,
1846                 uint64_t *ret_size,
1847                 sd_id128_t *ret_disk_uuid) {
1848
1849         _cleanup_(fdisk_unref_partitionp) struct fdisk_partition *p = NULL, *q = NULL;
1850         _cleanup_(fdisk_unref_parttypep) struct fdisk_parttype *t = NULL;
1851         _cleanup_(fdisk_unref_contextp) struct fdisk_context *c = NULL;
1852         _cleanup_free_ char *disk_uuid_as_string = NULL;
1853         uint64_t offset, size, first_lba, start, last_lba, end;
1854         sd_id128_t disk_uuid;
1855         int r;
1856
1857         assert(fd >= 0);
1858         assert(label);
1859         assert(ret_offset);
1860         assert(ret_size);
1861
1862         t = fdisk_new_parttype();
1863         if (!t)
1864                 return log_oom();
1865
1866         r = fdisk_parttype_set_typestr(t, SD_GPT_USER_HOME_STR);
1867         if (r < 0)
1868                 return log_error_errno(r, "Failed to initialize partition type: %m");
1869
1870         r = fdisk_new_context_at(fd, /* path= */ NULL, /* read_only= */ false, sector_size, &c);
1871         if (r < 0)
1872                 return log_error_errno(r, "Failed to open device: %m");
1873
1874         r = fdisk_create_disklabel(c, "gpt");
1875         if (r < 0)
1876                 return log_error_errno(r, "Failed to create GPT disk label: %m");
1877
1878         p = fdisk_new_partition();
1879         if (!p)
1880                 return log_oom();
1881
1882         r = fdisk_partition_set_type(p, t);
1883         if (r < 0)
1884                 return log_error_errno(r, "Failed to set partition type: %m");
1885
1886         r = fdisk_partition_partno_follow_default(p, 1);
1887         if (r < 0)
1888                 return log_error_errno(r, "Failed to place partition at first free partition index: %m");
1889
1890         first_lba = fdisk_get_first_lba(c); /* Boundary where usable space starts */
1891         assert(first_lba <= UINT64_MAX/512);
1892         start = DISK_SIZE_ROUND_UP(first_lba * 512); /* Round up to multiple of 4K */
1893
1894         log_debug("Starting partition at offset %" PRIu64, start);
1895
1896         if (start == UINT64_MAX)
1897                 return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "Overflow while rounding up start LBA.");
1898
1899         last_lba = fdisk_get_last_lba(c); /* One sector before boundary where usable space ends */
1900         assert(last_lba < UINT64_MAX/512);
1901         end = DISK_SIZE_ROUND_DOWN((last_lba + 1) * 512); /* Round down to multiple of 4K */
1902
1903         if (end <= start)
1904                 return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "Resulting partition size zero or negative.");
1905
1906         r = fdisk_partition_set_start(p, start / 512);
1907         if (r < 0)
1908                 return log_error_errno(r, "Failed to place partition at offset %" PRIu64 ": %m", start);
1909
1910         r = fdisk_partition_set_size(p, (end - start) / 512);
1911         if (r < 0)
1912                 return log_error_errno(r, "Failed to end partition at offset %" PRIu64 ": %m", end);
1913
1914         r = fdisk_partition_set_name(p, label);
1915         if (r < 0)
1916                 return log_error_errno(r, "Failed to set partition name: %m");
1917
1918         r = fdisk_partition_set_uuid(p, SD_ID128_TO_UUID_STRING(uuid));
1919         if (r < 0)
1920                 return log_error_errno(r, "Failed to set partition UUID: %m");
1921
1922         r = fdisk_add_partition(c, p, NULL);
1923         if (r < 0)
1924                 return log_error_errno(r, "Failed to add partition: %m");
1925
1926         r = fdisk_write_disklabel(c);
1927         if (r < 0)
1928                 return log_error_errno(r, "Failed to write disk label: %m");
1929
1930         r = fdisk_get_disklabel_id(c, &disk_uuid_as_string);
1931         if (r < 0)
1932                 return log_error_errno(r, "Failed to determine disk label UUID: %m");
1933
1934         r = sd_id128_from_string(disk_uuid_as_string, &disk_uuid);
1935         if (r < 0)
1936                 return log_error_errno(r, "Failed to parse disk label UUID: %m");
1937
1938         r = fdisk_get_partition(c, 0, &q);
1939         if (r < 0)
1940                 return log_error_errno(r, "Failed to read created partition metadata: %m");
1941
1942         assert(fdisk_partition_has_start(q));
1943         offset = fdisk_partition_get_start(q);
1944         if (offset > UINT64_MAX / 512U)
1945                 return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "Partition offset too large.");
1946
1947         assert(fdisk_partition_has_size(q));
1948         size = fdisk_partition_get_size(q);
1949         if (size > UINT64_MAX / 512U)
1950                 return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "Partition size too large.");
1951
1952         *ret_offset = offset * 512U;
1953         *ret_size = size * 512U;
1954         *ret_disk_uuid = disk_uuid;
1955
1956         return 0;
1957 }
1958
1959 static bool supported_fs_size(const char *fstype, uint64_t host_size) {
1960         uint64_t m;
1961
1962         m = minimal_size_by_fs_name(fstype);
1963         if (m == UINT64_MAX)
1964                 return false;
1965
1966         return host_size >= m;
1967 }
1968
1969 static int wait_for_devlink(const char *path) {
1970         _cleanup_close_ int inotify_fd = -EBADF;
1971         usec_t until;
1972         int r;
1973
1974         /* let's wait for a device link to show up in /dev, with a timeout. This is good to do since we
1975          * return a /dev/disk/by-uuid/… link to our callers and they likely want to access it right-away,
1976          * hence let's wait until udev has caught up with our changes, and wait for the symlink to be
1977          * created. */
1978
1979         until = usec_add(now(CLOCK_MONOTONIC), 45 * USEC_PER_SEC);
1980
1981         for (;;) {
1982                 _cleanup_free_ char *dn = NULL;
1983                 usec_t w;
1984
1985                 if (laccess(path, F_OK) < 0) {
1986                         if (errno != ENOENT)
1987                                 return log_error_errno(errno, "Failed to determine whether %s exists: %m", path);
1988                 } else
1989                         return 0; /* Found it */
1990
1991                 if (inotify_fd < 0) {
1992                         /* We need to wait for the device symlink to show up, let's create an inotify watch for it */
1993                         inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1994                         if (inotify_fd < 0)
1995                                 return log_error_errno(errno, "Failed to allocate inotify fd: %m");
1996                 }
1997
1998                 r = path_extract_directory(path, &dn);
1999                 if (r < 0)
2000                         return log_error_errno(r, "Failed to extract directory from device node path '%s': %m", path);
2001                 for (;;) {
2002                         _cleanup_free_ char *ndn = NULL;
2003
2004                         log_info("Watching %s", dn);
2005
2006                         if (inotify_add_watch(inotify_fd, dn, IN_CREATE|IN_MOVED_TO|IN_ONLYDIR|IN_DELETE_SELF|IN_MOVE_SELF) < 0) {
2007                                 if (errno != ENOENT)
2008                                         return log_error_errno(errno, "Failed to add watch on %s: %m", dn);
2009                         } else
2010                                 break;
2011
2012                         r = path_extract_directory(dn, &ndn);
2013                         if (r == -EADDRNOTAVAIL) /* Arrived at the top? */
2014                                 break;
2015                         if (r < 0)
2016                                 return log_error_errno(r, "Failed to extract directory from device node path '%s': %m", dn);
2017
2018                         free_and_replace(dn, ndn);
2019                 }
2020
2021                 w = now(CLOCK_MONOTONIC);
2022                 if (w >= until)
2023                         return log_error_errno(SYNTHETIC_ERRNO(ETIMEDOUT), "Device link %s still hasn't shown up, giving up.", path);
2024
2025                 r = fd_wait_for_event(inotify_fd, POLLIN, until - w);
2026                 if (ERRNO_IS_NEG_TRANSIENT(r))
2027                         continue;
2028                 if (r < 0)
2029                         return log_error_errno(r, "Failed to watch inotify: %m");
2030
2031                 (void) flush_fd(inotify_fd);
2032         }
2033 }
2034
2035 static int calculate_initial_image_size(UserRecord *h, int image_fd, const char *fstype, uint64_t *ret) {
2036         uint64_t upper_boundary, lower_boundary;
2037         struct statfs sfs;
2038
2039         assert(h);
2040         assert(image_fd >= 0);
2041         assert(ret);
2042
2043         if (fstatfs(image_fd, &sfs) < 0)
2044                 return log_error_errno(errno, "statfs() on image failed: %m");
2045
2046         upper_boundary = DISK_SIZE_ROUND_DOWN((uint64_t) sfs.f_bsize * sfs.f_bavail);
2047
2048         if (h->disk_size != UINT64_MAX)
2049                 *ret = MIN(DISK_SIZE_ROUND_DOWN(h->disk_size), upper_boundary);
2050         else if (h->disk_size_relative == UINT64_MAX) {
2051
2052                 if (upper_boundary > UINT64_MAX / USER_DISK_SIZE_DEFAULT_PERCENT)
2053                         return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Disk size too large.");
2054
2055                 *ret = DISK_SIZE_ROUND_DOWN(upper_boundary * USER_DISK_SIZE_DEFAULT_PERCENT / 100);
2056
2057                 log_info("Sizing home to %u%% of available disk space, which is %s.",
2058                          USER_DISK_SIZE_DEFAULT_PERCENT,
2059                          FORMAT_BYTES(*ret));
2060         } else {
2061                 *ret = DISK_SIZE_ROUND_DOWN((uint64_t) ((double) upper_boundary * (double) CLAMP(h->disk_size_relative, 0U, UINT32_MAX) / (double) UINT32_MAX));
2062
2063                 log_info("Sizing home to %" PRIu64 ".%01" PRIu64 "%% of available disk space, which is %s.",
2064                          (h->disk_size_relative * 100) / UINT32_MAX,
2065                          ((h->disk_size_relative * 1000) / UINT32_MAX) % 10,
2066                          FORMAT_BYTES(*ret));
2067         }
2068
2069         lower_boundary = minimal_size_by_fs_name(fstype);
2070         if (lower_boundary != UINT64_MAX) {
2071                 assert(GPT_LUKS2_OVERHEAD < UINT64_MAX - lower_boundary);
2072                 lower_boundary += GPT_LUKS2_OVERHEAD;
2073         }
2074         if (lower_boundary == UINT64_MAX || lower_boundary < USER_DISK_SIZE_MIN)
2075                 lower_boundary = USER_DISK_SIZE_MIN;
2076
2077         if (*ret < lower_boundary)
2078                 *ret = lower_boundary;
2079
2080         return 0;
2081 }
2082
2083 static int home_truncate(
2084                 UserRecord *h,
2085                 int fd,
2086                 uint64_t size) {
2087
2088         bool trunc;
2089         int r;
2090
2091         assert(h);
2092         assert(fd >= 0);
2093
2094         trunc = user_record_luks_discard(h);
2095         if (!trunc) {
2096                 r = fallocate(fd, 0, 0, size);
2097                 if (r < 0 && ERRNO_IS_NOT_SUPPORTED(errno)) {
2098                         /* Some file systems do not support fallocate(), let's gracefully degrade
2099                          * (ZFS, reiserfs, …) and fall back to truncation */
2100                         log_notice_errno(errno, "Backing file system does not support fallocate(), falling back to ftruncate(), i.e. implicitly using non-discard mode.");
2101                         trunc = true;
2102                 }
2103         }
2104
2105         if (trunc)
2106                 r = ftruncate(fd, size);
2107
2108         if (r < 0) {
2109                 if (ERRNO_IS_DISK_SPACE(errno)) {
2110                         log_debug_errno(errno, "Not enough disk space to allocate home of size %s.", FORMAT_BYTES(size));
2111                         return -ENOSPC; /* make recognizable */
2112                 }
2113
2114                 return log_error_errno(errno, "Failed to truncate home image: %m");
2115         }
2116
2117         return !trunc; /* Return == 0 if we managed to truncate, > 0 if we managed to allocate */
2118 }
2119
2120 int home_create_luks(
2121                 UserRecord *h,
2122                 HomeSetup *setup,
2123                 const PasswordCache *cache,
2124                 char **effective_passwords,
2125                 UserRecord **ret_home) {
2126
2127         _cleanup_free_ char *subdir = NULL, *disk_uuid_path = NULL;
2128         uint64_t encrypted_size,
2129                 host_size = 0, partition_offset = 0, partition_size = 0; /* Unnecessary initialization to appease gcc */
2130         _cleanup_(user_record_unrefp) UserRecord *new_home = NULL;
2131         sd_id128_t partition_uuid, fs_uuid, luks_uuid, disk_uuid;
2132         _cleanup_close_ int mount_fd = -EBADF;
2133         const char *fstype, *ip;
2134         struct statfs sfs;
2135         int r;
2136         _cleanup_strv_free_ char **extra_mkfs_options = NULL;
2137
2138         assert(h);
2139         assert(h->storage < 0 || h->storage == USER_LUKS);
2140         assert(setup);
2141         assert(!setup->temporary_image_path);
2142         assert(setup->image_fd < 0);
2143         assert(ret_home);
2144
2145         r = dlopen_cryptsetup();
2146         if (r < 0)
2147                 return r;
2148
2149         assert_se(ip = user_record_image_path(h));
2150
2151         fstype = user_record_file_system_type(h);
2152         if (!supported_fstype(fstype))
2153                 return log_error_errno(SYNTHETIC_ERRNO(EPROTONOSUPPORT), "Unsupported file system type: %s", fstype);
2154
2155         r = mkfs_exists(fstype);
2156         if (r < 0)
2157                 return log_error_errno(r, "Failed to check if mkfs binary for %s exists: %m", fstype);
2158         if (r == 0) {
2159                 if (h->file_system_type || streq(fstype, "ext4") || !supported_fstype("ext4"))
2160                         return log_error_errno(SYNTHETIC_ERRNO(EPROTONOSUPPORT), "mkfs binary for file system type %s does not exist.", fstype);
2161
2162                 /* If the record does not explicitly declare a file system to use, and the compiled-in
2163                  * default does not actually exist, than do an automatic fallback onto ext4, as the baseline
2164                  * fs of Linux. We won't search for a working fs type here beyond ext4, i.e. nothing fancier
2165                  * than a single, conservative fallback to baseline. This should be useful in minimal
2166                  * environments where mkfs.btrfs or so are not made available, but mkfs.ext4 as Linux' most
2167                  * boring, most basic fs is. */
2168                 log_info("Formatting tool for compiled-in default file system %s not available, falling back to ext4 instead.", fstype);
2169                 fstype = "ext4";
2170         }
2171
2172         if (sd_id128_is_null(h->partition_uuid)) {
2173                 r = sd_id128_randomize(&partition_uuid);
2174                 if (r < 0)
2175                         return log_error_errno(r, "Failed to acquire partition UUID: %m");
2176         } else
2177                 partition_uuid = h->partition_uuid;
2178
2179         if (sd_id128_is_null(h->luks_uuid)) {
2180                 r = sd_id128_randomize(&luks_uuid);
2181                 if (r < 0)
2182                         return log_error_errno(r, "Failed to acquire LUKS UUID: %m");
2183         } else
2184                 luks_uuid = h->luks_uuid;
2185
2186         if (sd_id128_is_null(h->file_system_uuid)) {
2187                 r = sd_id128_randomize(&fs_uuid);
2188                 if (r < 0)
2189                         return log_error_errno(r, "Failed to acquire file system UUID: %m");
2190         } else
2191                 fs_uuid = h->file_system_uuid;
2192
2193         r = make_dm_names(h, setup);
2194         if (r < 0)
2195                 return r;
2196
2197         r = access(setup->dm_node, F_OK);
2198         if (r < 0) {
2199                 if (errno != ENOENT)
2200                         return log_error_errno(errno, "Failed to determine whether %s exists: %m", setup->dm_node);
2201         } else
2202                 return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Device mapper device %s already exists, refusing.", setup->dm_node);
2203
2204         if (path_startswith(ip, "/dev/")) {
2205                 _cleanup_free_ char *sysfs = NULL;
2206                 uint64_t block_device_size;
2207                 struct stat st;
2208
2209                 /* Let's place the home directory on a real device, i.e. a USB stick or such */
2210
2211                 setup->image_fd = open_image_file(h, ip, &st);
2212                 if (setup->image_fd < 0)
2213                         return setup->image_fd;
2214
2215                 if (!S_ISBLK(st.st_mode))
2216                         return log_error_errno(SYNTHETIC_ERRNO(ENOTBLK), "Device is not a block device, refusing.");
2217
2218                 if (asprintf(&sysfs, "/sys/dev/block/" DEVNUM_FORMAT_STR "/partition", DEVNUM_FORMAT_VAL(st.st_rdev)) < 0)
2219                         return log_oom();
2220                 if (access(sysfs, F_OK) < 0) {
2221                         if (errno != ENOENT)
2222                                 return log_error_errno(errno, "Failed to check whether %s exists: %m", sysfs);
2223                 } else
2224                         return log_error_errno(SYNTHETIC_ERRNO(ENOTBLK), "Operating on partitions is currently not supported, sorry. Please specify a top-level block device.");
2225
2226                 if (flock(setup->image_fd, LOCK_EX) < 0) /* make sure udev doesn't read from it while we operate on the device */
2227                         return log_error_errno(errno, "Failed to lock block device %s: %m", ip);
2228
2229                 r = blockdev_get_device_size(setup->image_fd, &block_device_size);
2230                 if (r < 0)
2231                         return log_error_errno(r, "Failed to read block device size: %m");
2232
2233                 if (h->disk_size == UINT64_MAX) {
2234
2235                         /* If a relative disk size is requested, apply it relative to the block device size */
2236                         if (h->disk_size_relative < UINT32_MAX)
2237                                 host_size = CLAMP(DISK_SIZE_ROUND_DOWN(block_device_size * h->disk_size_relative / UINT32_MAX),
2238                                                   USER_DISK_SIZE_MIN, USER_DISK_SIZE_MAX);
2239                         else
2240                                 host_size = block_device_size; /* Otherwise, take the full device */
2241
2242                 } else if (h->disk_size > block_device_size)
2243                         return log_error_errno(SYNTHETIC_ERRNO(EMSGSIZE), "Selected disk size larger than backing block device, refusing.");
2244                 else
2245                         host_size = DISK_SIZE_ROUND_DOWN(h->disk_size);
2246
2247                 if (!supported_fs_size(fstype, LESS_BY(host_size, GPT_LUKS2_OVERHEAD)))
2248                         return log_error_errno(SYNTHETIC_ERRNO(ERANGE),
2249                                                "Selected file system size too small for %s.", fstype);
2250
2251                 /* After creation we should reference this partition by its UUID instead of the block
2252                  * device. That's preferable since the user might have specified a device node such as
2253                  * /dev/sdb to us, which might look very different when replugged. */
2254                 if (asprintf(&disk_uuid_path, "/dev/disk/by-uuid/" SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(luks_uuid)) < 0)
2255                         return log_oom();
2256
2257                 if (user_record_luks_discard(h) || user_record_luks_offline_discard(h)) {
2258                         /* If we want online or offline discard, discard once before we start using things. */
2259
2260                         if (ioctl(setup->image_fd, BLKDISCARD, (uint64_t[]) { 0, block_device_size }) < 0)
2261                                 log_full_errno(errno == EOPNOTSUPP ? LOG_DEBUG : LOG_WARNING, errno,
2262                                                "Failed to issue full-device BLKDISCARD on device, ignoring: %m");
2263                         else
2264                                 log_info("Full device discard completed.");
2265                 }
2266         } else {
2267                 _cleanup_free_ char *t = NULL;
2268
2269                 r = mkdir_parents(ip, 0755);
2270                 if (r < 0)
2271                         return log_error_errno(r, "Failed to create parent directory of %s: %m", ip);
2272
2273                 r = tempfn_random(ip, "homework", &t);
2274                 if (r < 0)
2275                         return log_error_errno(r, "Failed to derive temporary file name for %s: %m", ip);
2276
2277                 setup->image_fd = open(t, O_RDWR|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, 0600);
2278                 if (setup->image_fd < 0)
2279                         return log_error_errno(errno, "Failed to create home image %s: %m", t);
2280
2281                 setup->temporary_image_path = TAKE_PTR(t);
2282
2283                 r = chattr_full(setup->image_fd, NULL, FS_NOCOW_FL|FS_NOCOMP_FL, FS_NOCOW_FL|FS_NOCOMP_FL, NULL, NULL, CHATTR_FALLBACK_BITWISE);
2284                 if (r < 0 && r != -ENOANO) /* ENOANO → some bits didn't work; which we skip logging about because chattr_full() already debug logs about those flags */
2285                         log_full_errno(ERRNO_IS_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r,
2286                                        "Failed to set file attributes on %s, ignoring: %m", setup->temporary_image_path);
2287
2288                 r = calculate_initial_image_size(h, setup->image_fd, fstype, &host_size);
2289                 if (r < 0)
2290                         return r;
2291
2292                 r = resize_image_loop(h, setup, 0, host_size, &host_size);
2293                 if (r < 0)
2294                         return r;
2295
2296                 log_info("Allocating image file completed.");
2297         }
2298
2299         r = make_partition_table(
2300                         setup->image_fd,
2301                         user_record_luks_sector_size(h),
2302                         user_record_user_name_and_realm(h),
2303                         partition_uuid,
2304                         &partition_offset,
2305                         &partition_size,
2306                         &disk_uuid);
2307         if (r < 0)
2308                 return r;
2309
2310         log_info("Writing of partition table completed.");
2311
2312         r = loop_device_make(
2313                         setup->image_fd,
2314                         O_RDWR,
2315                         partition_offset,
2316                         partition_size,
2317                         user_record_luks_sector_size(h),
2318                         0,
2319                         LOCK_EX,
2320                         &setup->loop);
2321         if (r < 0) {
2322                 if (r == -ENOENT) { /* this means /dev/loop-control doesn't exist, i.e. we are in a container
2323                                      * or similar and loopback bock devices are not available, return a
2324                                      * recognizable error in this case. */
2325                         log_error_errno(r, "Loopback block device support is not available on this system.");
2326                         return -ENOLINK; /* Make recognizable */
2327                 }
2328
2329                 return log_error_errno(r, "Failed to set up loopback device for %s: %m", setup->temporary_image_path);
2330         }
2331
2332         log_info("Setting up loopback device %s completed.", setup->loop->node ?: ip);
2333
2334         r = luks_format(setup->loop->node,
2335                         setup->dm_name,
2336                         luks_uuid,
2337                         user_record_user_name_and_realm(h),
2338                         cache,
2339                         effective_passwords,
2340                         user_record_luks_discard(h) || user_record_luks_offline_discard(h),
2341                         h,
2342                         &setup->crypt_device);
2343         if (r < 0)
2344                 return r;
2345
2346         setup->undo_dm = true;
2347
2348         r = block_get_size_by_path(setup->dm_node, &encrypted_size);
2349         if (r < 0)
2350                 return log_error_errno(r, "Failed to get encrypted block device size: %m");
2351
2352         log_info("Setting up LUKS device %s completed.", setup->dm_node);
2353
2354         r = mkfs_options_from_env("HOME", fstype, &extra_mkfs_options);
2355         if (r < 0)
2356                 return log_error_errno(r, "Failed to determine mkfs command line options for '%s': %m", fstype);
2357
2358         r = make_filesystem(setup->dm_node,
2359                             fstype,
2360                             user_record_user_name_and_realm(h),
2361                             /* root = */ NULL,
2362                             fs_uuid,
2363                             user_record_luks_discard(h),
2364                             /* quiet = */ true,
2365                             /* sector_size = */ 0,
2366                             extra_mkfs_options);
2367         if (r < 0)
2368                 return r;
2369
2370         log_info("Formatting file system completed.");
2371
2372         r = home_unshare_and_mount(setup->dm_node, fstype, user_record_luks_discard(h), user_record_mount_flags(h), h->luks_extra_mount_options);
2373         if (r < 0)
2374                 return r;
2375
2376         setup->undo_mount = true;
2377
2378         subdir = path_join(HOME_RUNTIME_WORK_DIR, user_record_user_name_and_realm(h));
2379         if (!subdir)
2380                 return log_oom();
2381
2382         /* Prefer using a btrfs subvolume if we can, fall back to directory otherwise */
2383         r = btrfs_subvol_make_fallback(AT_FDCWD, subdir, 0700);
2384         if (r < 0)
2385                 return log_error_errno(r, "Failed to create user directory in mounted image file: %m");
2386
2387         setup->root_fd = open(subdir, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
2388         if (setup->root_fd < 0)
2389                 return log_error_errno(errno, "Failed to open user directory in mounted image file: %m");
2390
2391         (void) home_shift_uid(setup->root_fd, NULL, UID_NOBODY, h->uid, &mount_fd);
2392
2393         if (mount_fd >= 0) {
2394                 /* If we have established a new mount, then we can use that as new root fd to our home directory. */
2395                 safe_close(setup->root_fd);
2396
2397                 setup->root_fd = fd_reopen(mount_fd, O_RDONLY|O_CLOEXEC|O_DIRECTORY);
2398                 if (setup->root_fd < 0)
2399                         return log_error_errno(setup->root_fd, "Unable to convert mount fd into proper directory fd: %m");
2400
2401                 mount_fd = safe_close(mount_fd);
2402         }
2403
2404         r = home_populate(h, setup->root_fd);
2405         if (r < 0)
2406                 return r;
2407
2408         r = home_sync_and_statfs(setup->root_fd, &sfs);
2409         if (r < 0)
2410                 return r;
2411
2412         r = user_record_clone(h, USER_RECORD_LOAD_MASK_SECRET|USER_RECORD_LOG|USER_RECORD_PERMISSIVE, &new_home);
2413         if (r < 0)
2414                 return log_error_errno(r, "Failed to clone record: %m");
2415
2416         r = user_record_add_binding(
2417                         new_home,
2418                         USER_LUKS,
2419                         disk_uuid_path ?: ip,
2420                         partition_uuid,
2421                         luks_uuid,
2422                         fs_uuid,
2423                         sym_crypt_get_cipher(setup->crypt_device),
2424                         sym_crypt_get_cipher_mode(setup->crypt_device),
2425                         luks_volume_key_size_convert(setup->crypt_device),
2426                         fstype,
2427                         NULL,
2428                         h->uid,
2429                         (gid_t) h->uid);
2430         if (r < 0)
2431                 return log_error_errno(r, "Failed to add binding to record: %m");
2432
2433         if (user_record_luks_offline_discard(h)) {
2434                 r = run_fitrim(setup->root_fd);
2435                 if (r < 0)
2436                         return r;
2437         }
2438
2439         setup->root_fd = safe_close(setup->root_fd);
2440
2441         r = home_setup_undo_mount(setup, LOG_ERR);
2442         if (r < 0)
2443                 return r;
2444
2445         r = home_setup_undo_dm(setup, LOG_ERR);
2446         if (r < 0)
2447                 return r;
2448
2449         setup->loop = loop_device_unref(setup->loop);
2450
2451         if (!user_record_luks_offline_discard(h)) {
2452                 r= run_fallocate(setup->image_fd, NULL /* refresh stat() data */);
2453                 if (r < 0)
2454                         return r;
2455         }
2456
2457         /* Sync everything to disk before we move things into place under the final name. */
2458         if (fsync(setup->image_fd) < 0)
2459                 return log_error_errno(r, "Failed to synchronize image to disk: %m");
2460
2461         if (disk_uuid_path)
2462                 /* Reread partition table if this is a block device */
2463                 (void) ioctl(setup->image_fd, BLKRRPART, 0);
2464         else {
2465                 assert(setup->temporary_image_path);
2466
2467                 if (rename(setup->temporary_image_path, ip) < 0)
2468                         return log_error_errno(errno, "Failed to rename image file: %m");
2469
2470                 setup->temporary_image_path = mfree(setup->temporary_image_path);
2471
2472                 /* If we operate on a file, sync the containing directory too. */
2473                 r = fsync_directory_of_file(setup->image_fd);
2474                 if (r < 0)
2475                         return log_error_errno(r, "Failed to synchronize directory of image file to disk: %m");
2476
2477                 log_info("Moved image file into place.");
2478         }
2479
2480         /* Let's close the image fd now. If we are operating on a real block device this will release the BSD
2481          * lock that ensures udev doesn't interfere with what we are doing */
2482         setup->image_fd = safe_close(setup->image_fd);
2483
2484         if (disk_uuid_path)
2485                 (void) wait_for_devlink(disk_uuid_path);
2486
2487         log_info("Creation completed.");
2488
2489         print_size_summary(host_size, encrypted_size, &sfs);
2490
2491         log_debug("GPT + LUKS2 overhead is %" PRIu64 " (expected %" PRIu64 ")", host_size - encrypted_size, GPT_LUKS2_OVERHEAD);
2492
2493         *ret_home = TAKE_PTR(new_home);
2494         return 0;
2495 }
2496
2497 int home_get_state_luks(UserRecord *h, HomeSetup *setup) {
2498         int r;
2499
2500         assert(h);
2501         assert(setup);
2502
2503         r = make_dm_names(h, setup);
2504         if (r < 0)
2505                 return r;
2506
2507         r = access(setup->dm_node, F_OK);
2508         if (r < 0 && errno != ENOENT)
2509                 return log_error_errno(errno, "Failed to determine whether %s exists: %m", setup->dm_node);
2510
2511         return r >= 0;
2512 }
2513
2514 enum {
2515         CAN_RESIZE_ONLINE,
2516         CAN_RESIZE_OFFLINE,
2517 };
2518
2519 static int can_resize_fs(int fd, uint64_t old_size, uint64_t new_size) {
2520         struct statfs sfs;
2521
2522         assert(fd >= 0);
2523
2524         /* Filter out bogus requests early */
2525         if (old_size == 0 || old_size == UINT64_MAX ||
2526             new_size == 0 || new_size == UINT64_MAX)
2527                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid resize parameters.");
2528
2529         if ((old_size & 511) != 0 || (new_size & 511) != 0)
2530                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Resize parameters not multiple of 512.");
2531
2532         if (fstatfs(fd, &sfs) < 0)
2533                 return log_error_errno(errno, "Failed to fstatfs() file system: %m");
2534
2535         if (is_fs_type(&sfs, BTRFS_SUPER_MAGIC)) {
2536
2537                 if (new_size < BTRFS_MINIMAL_SIZE)
2538                         return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "New file system size too small for btrfs (needs to be 256M at least.");
2539
2540                 /* btrfs can grow and shrink online */
2541
2542         } else if (is_fs_type(&sfs, XFS_SB_MAGIC)) {
2543
2544                 if (new_size < XFS_MINIMAL_SIZE)
2545                         return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "New file system size too small for xfs (needs to be 14M at least).");
2546
2547                 /* XFS can grow, but not shrink */
2548                 if (new_size < old_size)
2549                         return log_error_errno(SYNTHETIC_ERRNO(EMSGSIZE), "Shrinking this type of file system is not supported.");
2550
2551         } else if (is_fs_type(&sfs, EXT4_SUPER_MAGIC)) {
2552
2553                 if (new_size < EXT4_MINIMAL_SIZE)
2554                         return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "New file system size too small for ext4 (needs to be 1M at least).");
2555
2556                 /* ext4 can grow online, and shrink offline */
2557                 if (new_size < old_size)
2558                         return CAN_RESIZE_OFFLINE;
2559
2560         } else
2561                 return log_error_errno(SYNTHETIC_ERRNO(ESOCKTNOSUPPORT), "Resizing this type of file system is not supported.");
2562
2563         return CAN_RESIZE_ONLINE;
2564 }
2565
2566 static int ext4_offline_resize_fs(
2567                 HomeSetup *setup,
2568                 uint64_t new_size,
2569                 bool discard,
2570                 unsigned long flags,
2571                 const char *extra_mount_options) {
2572
2573         _cleanup_free_ char *size_str = NULL;
2574         bool re_open = false, re_mount = false;
2575         pid_t resize_pid, fsck_pid;
2576         int r, exit_status;
2577
2578         assert(setup);
2579         assert(setup->dm_node);
2580
2581         /* First, unmount the file system */
2582         if (setup->root_fd >= 0) {
2583                 setup->root_fd = safe_close(setup->root_fd);
2584                 re_open = true;
2585         }
2586
2587         if (setup->undo_mount) {
2588                 r = home_setup_undo_mount(setup, LOG_ERR);
2589                 if (r < 0)
2590                         return r;
2591
2592                 re_mount = true;
2593         }
2594
2595         log_info("Temporary unmounting of file system completed.");
2596
2597         /* resize2fs requires that the file system is force checked first, do so. */
2598         r = safe_fork("(e2fsck)",
2599                       FORK_RESET_SIGNALS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG_SIGTERM|FORK_LOG|FORK_STDOUT_TO_STDERR|FORK_CLOSE_ALL_FDS,
2600                       &fsck_pid);
2601         if (r < 0)
2602                 return r;
2603         if (r == 0) {
2604                 /* Child */
2605                 execlp("e2fsck" ,"e2fsck", "-fp", setup->dm_node, NULL);
2606                 log_open();
2607                 log_error_errno(errno, "Failed to execute e2fsck: %m");
2608                 _exit(EXIT_FAILURE);
2609         }
2610
2611         exit_status = wait_for_terminate_and_check("e2fsck", fsck_pid, WAIT_LOG_ABNORMAL);
2612         if (exit_status < 0)
2613                 return exit_status;
2614         if ((exit_status & ~FSCK_ERROR_CORRECTED) != 0) {
2615                 log_warning("e2fsck failed with exit status %i.", exit_status);
2616
2617                 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
2618                         return log_error_errno(SYNTHETIC_ERRNO(EIO), "File system is corrupted, refusing.");
2619
2620                 log_warning("Ignoring fsck error.");
2621         }
2622
2623         log_info("Forced file system check completed.");
2624
2625         /* We use 512 sectors here, because resize2fs doesn't do byte sizes */
2626         if (asprintf(&size_str, "%" PRIu64 "s", new_size / 512) < 0)
2627                 return log_oom();
2628
2629         /* Resize the thing */
2630         r = safe_fork("(e2resize)",
2631                       FORK_RESET_SIGNALS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG_SIGTERM|FORK_LOG|FORK_WAIT|FORK_STDOUT_TO_STDERR|FORK_CLOSE_ALL_FDS,
2632                       &resize_pid);
2633         if (r < 0)
2634                 return r;
2635         if (r == 0) {
2636                 /* Child */
2637                 execlp("resize2fs" ,"resize2fs", setup->dm_node, size_str, NULL);
2638                 log_open();
2639                 log_error_errno(errno, "Failed to execute resize2fs: %m");
2640                 _exit(EXIT_FAILURE);
2641         }
2642
2643         log_info("Offline file system resize completed.");
2644
2645         /* Re-establish mounts and reopen the directory */
2646         if (re_mount) {
2647                 r = home_mount_node(setup->dm_node, "ext4", discard, flags, extra_mount_options);
2648                 if (r < 0)
2649                         return r;
2650
2651                 setup->undo_mount = true;
2652         }
2653
2654         if (re_open) {
2655                 setup->root_fd = open(HOME_RUNTIME_WORK_DIR, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
2656                 if (setup->root_fd < 0)
2657                         return log_error_errno(errno, "Failed to reopen file system: %m");
2658         }
2659
2660         log_info("File system mounted again.");
2661
2662         return 0;
2663 }
2664
2665 static int prepare_resize_partition(
2666                 int fd,
2667                 uint64_t partition_offset,
2668                 uint64_t old_partition_size,
2669                 sd_id128_t *ret_disk_uuid,
2670                 struct fdisk_table **ret_table,
2671                 struct fdisk_partition **ret_partition) {
2672
2673         _cleanup_(fdisk_unref_contextp) struct fdisk_context *c = NULL;
2674         _cleanup_(fdisk_unref_tablep) struct fdisk_table *t = NULL;
2675         _cleanup_free_ char *disk_uuid_as_string = NULL;
2676         struct fdisk_partition *found = NULL;
2677         sd_id128_t disk_uuid;
2678         size_t n_partitions;
2679         int r;
2680
2681         assert(fd >= 0);
2682         assert(ret_disk_uuid);
2683         assert(ret_table);
2684
2685         assert((partition_offset & 511) == 0);
2686         assert((old_partition_size & 511) == 0);
2687         assert(UINT64_MAX - old_partition_size >= partition_offset);
2688
2689         if (partition_offset == 0) {
2690                 /* If the offset is at the beginning we assume no partition table, let's exit early. */
2691                 log_debug("Not rewriting partition table, operating on naked device.");
2692                 *ret_disk_uuid = SD_ID128_NULL;
2693                 *ret_table = NULL;
2694                 *ret_partition = NULL;
2695                 return 0;
2696         }
2697
2698         r = fdisk_new_context_at(fd, /* path= */ NULL, /* read_only= */ false, UINT32_MAX, &c);
2699         if (r < 0)
2700                 return log_error_errno(r, "Failed to open device: %m");
2701
2702         if (!fdisk_is_labeltype(c, FDISK_DISKLABEL_GPT))
2703                 return log_error_errno(SYNTHETIC_ERRNO(ENOMEDIUM), "Disk has no GPT partition table.");
2704
2705         r = fdisk_get_disklabel_id(c, &disk_uuid_as_string);
2706         if (r < 0)
2707                 return log_error_errno(r, "Failed to acquire disk UUID: %m");
2708
2709         r = sd_id128_from_string(disk_uuid_as_string, &disk_uuid);
2710         if (r < 0)
2711                 return log_error_errno(r, "Failed parse disk UUID: %m");
2712
2713         r = fdisk_get_partitions(c, &t);
2714         if (r < 0)
2715                 return log_error_errno(r, "Failed to acquire partition table: %m");
2716
2717         n_partitions = fdisk_table_get_nents(t);
2718         for (size_t i = 0; i < n_partitions; i++)  {
2719                 struct fdisk_partition *p;
2720
2721                 p = fdisk_table_get_partition(t, i);
2722                 if (!p)
2723                         return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to read partition metadata: %m");
2724
2725                 if (fdisk_partition_is_used(p) <= 0)
2726                         continue;
2727                 if (fdisk_partition_has_start(p) <= 0 || fdisk_partition_has_size(p) <= 0 || fdisk_partition_has_end(p) <= 0)
2728                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Found partition without a size.");
2729
2730                 if (fdisk_partition_get_start(p) == partition_offset / 512U &&
2731                     fdisk_partition_get_size(p) == old_partition_size / 512U) {
2732
2733                         if (found)
2734                                 return log_error_errno(SYNTHETIC_ERRNO(ENOTUNIQ), "Partition found twice, refusing.");
2735
2736                         found = p;
2737                 } else if (fdisk_partition_get_end(p) > partition_offset / 512U)
2738                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Can't extend, not last partition in image.");
2739         }
2740
2741         if (!found)
2742                 return log_error_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to find matching partition to resize.");
2743
2744         *ret_disk_uuid = disk_uuid;
2745         *ret_table = TAKE_PTR(t);
2746         *ret_partition = found;
2747
2748         return 1;
2749 }
2750
2751 static int get_maximum_partition_size(
2752                 int fd,
2753                 struct fdisk_partition *p,
2754                 uint64_t *ret_maximum_partition_size) {
2755
2756         _cleanup_(fdisk_unref_contextp) struct fdisk_context *c = NULL;
2757         uint64_t start_lba, start, last_lba, end;
2758         int r;
2759
2760         assert(fd >= 0);
2761         assert(p);
2762         assert(ret_maximum_partition_size);
2763
2764         r = fdisk_new_context_at(fd, /* path= */ NULL, /* read_only= */ true, /* sector_size= */ UINT32_MAX, &c);
2765         if (r < 0)
2766                 return log_error_errno(r, "Failed to create fdisk context: %m");
2767
2768         start_lba = fdisk_partition_get_start(p);
2769         assert(start_lba <= UINT64_MAX/512);
2770         start = start_lba * 512;
2771
2772         last_lba = fdisk_get_last_lba(c); /* One sector before boundary where usable space ends */
2773         assert(last_lba < UINT64_MAX/512);
2774         end = DISK_SIZE_ROUND_DOWN((last_lba + 1) * 512); /* Round down to multiple of 4K */
2775
2776         if (start > end)
2777                 return log_error_errno(SYNTHETIC_ERRNO(EBADMSG), "Last LBA is before partition start.");
2778
2779         *ret_maximum_partition_size = DISK_SIZE_ROUND_DOWN(end - start);
2780
2781         return 1;
2782 }
2783
2784 static int ask_cb(struct fdisk_context *c, struct fdisk_ask *ask, void *userdata) {
2785         char *result;
2786
2787         assert(c);
2788
2789         switch (fdisk_ask_get_type(ask)) {
2790
2791         case FDISK_ASKTYPE_STRING:
2792                 result = new(char, 37);
2793                 if (!result)
2794                         return log_oom();
2795
2796                 fdisk_ask_string_set_result(ask, sd_id128_to_uuid_string(*(sd_id128_t*) userdata, result));
2797                 break;
2798
2799         default:
2800                 log_debug("Unexpected question from libfdisk, ignoring.");
2801         }
2802
2803         return 0;
2804 }
2805
2806 static int apply_resize_partition(
2807                 int fd,
2808                 sd_id128_t disk_uuids,
2809                 struct fdisk_table *t,
2810                 struct fdisk_partition *p,
2811                 size_t new_partition_size) {
2812
2813         _cleanup_(fdisk_unref_contextp) struct fdisk_context *c = NULL;
2814         _cleanup_free_ void *two_zero_lbas = NULL;
2815         uint32_t ssz;
2816         ssize_t n;
2817         int r;
2818
2819         assert(fd >= 0);
2820         assert(!t == !p);
2821
2822         if (!t) /* no partition table to apply, exit early */
2823                 return 0;
2824
2825         assert(p);
2826
2827         /* Before writing our partition patch the final size in */
2828         r = fdisk_partition_size_explicit(p, 1);
2829         if (r < 0)
2830                 return log_error_errno(r, "Failed to enable explicit partition size: %m");
2831
2832         r = fdisk_partition_set_size(p, new_partition_size / 512U);
2833         if (r < 0)
2834                 return log_error_errno(r, "Failed to change partition size: %m");
2835
2836         r = probe_sector_size(fd, &ssz);
2837         if (r < 0)
2838                 return log_error_errno(r, "Failed to determine current sector size: %m");
2839
2840         two_zero_lbas = malloc0(ssz * 2);
2841         if (!two_zero_lbas)
2842                 return log_oom();
2843
2844         /* libfdisk appears to get confused by the existing PMBR. Let's explicitly flush it out. */
2845         n = pwrite(fd, two_zero_lbas, ssz * 2, 0);
2846         if (n < 0)
2847                 return log_error_errno(errno, "Failed to wipe partition table: %m");
2848         if ((size_t) n != ssz * 2)
2849                 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while wiping partition table.");
2850
2851         r = fdisk_new_context_at(fd, /* path= */ NULL, /* read_only= */ false, ssz, &c);
2852         if (r < 0)
2853                 return log_error_errno(r, "Failed to open device: %m");
2854
2855         r = fdisk_create_disklabel(c, "gpt");
2856         if (r < 0)
2857                 return log_error_errno(r, "Failed to create GPT disk label: %m");
2858
2859         r = fdisk_apply_table(c, t);
2860         if (r < 0)
2861                 return log_error_errno(r, "Failed to apply partition table: %m");
2862
2863         r = fdisk_set_ask(c, ask_cb, &disk_uuids);
2864         if (r < 0)
2865                 return log_error_errno(r, "Failed to set libfdisk query function: %m");
2866
2867         r = fdisk_set_disklabel_id(c);
2868         if (r < 0)
2869                 return log_error_errno(r, "Failed to change disklabel ID: %m");
2870
2871         r = fdisk_write_disklabel(c);
2872         if (r < 0)
2873                 return log_error_errno(r, "Failed to write disk label: %m");
2874
2875         return 1;
2876 }
2877
2878 /* Always keep at least 16M free, so that we can safely log in and update the user record while doing so */
2879 #define HOME_MIN_FREE (16U*1024U*1024U)
2880
2881 static int get_smallest_fs_size(int fd, uint64_t *ret) {
2882         uint64_t minsz, needed;
2883         struct statfs sfs;
2884
2885         assert(fd >= 0);
2886         assert(ret);
2887
2888         /* Determines the minimal disk size we might be able to shrink the file system referenced by the fd to. */
2889
2890         if (syncfs(fd) < 0) /* let's sync before we query the size, so that the values returned are accurate */
2891                 return log_error_errno(errno, "Failed to synchronize home file system: %m");
2892
2893         if (fstatfs(fd, &sfs) < 0)
2894                 return log_error_errno(errno, "Failed to statfs() home file system: %m");
2895
2896         /* Let's determine the minimal file system size of the used fstype */
2897         minsz = minimal_size_by_fs_magic(sfs.f_type);
2898         if (minsz == UINT64_MAX)
2899                 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "Don't know minimum file system size of file system type '%s' of home directory.", fs_type_to_string(sfs.f_type));
2900
2901         if (minsz < USER_DISK_SIZE_MIN)
2902                 minsz = USER_DISK_SIZE_MIN;
2903
2904         if (sfs.f_bfree > sfs.f_blocks)
2905                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Detected amount of free blocks is greater than the total amount of file system blocks. Refusing.");
2906
2907         /* Calculate how much disk space is currently in use. */
2908         needed = sfs.f_blocks - sfs.f_bfree;
2909         if (needed > UINT64_MAX / sfs.f_bsize)
2910                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "File system size out of range.");
2911
2912         needed *= sfs.f_bsize;
2913
2914         /* Add some safety margin of free space we'll always keep */
2915         if (needed > UINT64_MAX - HOME_MIN_FREE) /* Check for overflow */
2916                 needed = UINT64_MAX;
2917         else
2918                 needed += HOME_MIN_FREE;
2919
2920         *ret = DISK_SIZE_ROUND_UP(MAX(needed, minsz));
2921         return 0;
2922 }
2923
2924 static int get_largest_image_size(int fd, const struct stat *st, uint64_t *ret) {
2925         uint64_t used, avail, sum;
2926         struct statfs sfs;
2927         int r;
2928
2929         assert(fd >= 0);
2930         assert(st);
2931         assert(ret);
2932
2933         /* Determines the maximum file size we might be able to grow the image file referenced by the fd to. */
2934
2935         r = stat_verify_regular(st);
2936         if (r < 0)
2937                 return log_error_errno(r, "Image file is not a regular file, refusing: %m");
2938
2939         if (syncfs(fd) < 0)
2940                 return log_error_errno(errno, "Failed to synchronize file system backing image file: %m");
2941
2942         if (fstatfs(fd, &sfs) < 0)
2943                 return log_error_errno(errno, "Failed to statfs() image file: %m");
2944
2945         used = (uint64_t) st->st_blocks * 512;
2946         avail = (uint64_t) sfs.f_bsize * sfs.f_bavail;
2947
2948         if (avail > UINT64_MAX - used)
2949                 sum = UINT64_MAX;
2950         else
2951                 sum = avail + used;
2952
2953         *ret = DISK_SIZE_ROUND_DOWN(MIN(sum, USER_DISK_SIZE_MAX));
2954         return 0;
2955 }
2956
2957 static int resize_fs_loop(
2958                 UserRecord *h,
2959                 HomeSetup *setup,
2960                 int resize_type,
2961                 uint64_t old_fs_size,
2962                 uint64_t new_fs_size,
2963                 uint64_t *ret_fs_size) {
2964
2965         uint64_t current_fs_size;
2966         unsigned n_iterations = 0;
2967         int r;
2968
2969         assert(h);
2970         assert(setup);
2971         assert(setup->root_fd >= 0);
2972
2973         /* A bisection loop trying to find the closest size to what the user asked for. (Well, we bisect like
2974          * this only when we *shrink* the fs — if we grow the fs there's no need to bisect.) */
2975
2976         current_fs_size = old_fs_size;
2977         for (uint64_t lower_boundary = new_fs_size, upper_boundary = old_fs_size, try_fs_size = new_fs_size;;) {
2978                 bool worked;
2979
2980                 n_iterations++;
2981
2982                 /* Now resize the file system */
2983                 if (resize_type == CAN_RESIZE_ONLINE) {
2984                         r = resize_fs(setup->root_fd, try_fs_size, NULL);
2985                         if (r < 0) {
2986                                 if (!ERRNO_IS_DISK_SPACE(r) || new_fs_size > old_fs_size) /* Not a disk space issue? Not trying to shrink? */
2987                                         return log_error_errno(r, "Failed to resize file system: %m");
2988
2989                                 log_debug_errno(r, "Shrinking from %s to %s didn't work, not enough space for contained data.", FORMAT_BYTES(current_fs_size), FORMAT_BYTES(try_fs_size));
2990                                 worked = false;
2991                         } else {
2992                                 log_debug("Successfully resized from %s to %s.", FORMAT_BYTES(current_fs_size), FORMAT_BYTES(try_fs_size));
2993                                 current_fs_size = try_fs_size;
2994                                 worked = true;
2995                         }
2996
2997                         /* If we hit a disk space issue and are shrinking the fs, then maybe it helps to
2998                          * increase the image size. */
2999                 } else {
3000                         r = ext4_offline_resize_fs(setup, try_fs_size, user_record_luks_discard(h), user_record_mount_flags(h), h->luks_extra_mount_options);
3001                         if (r < 0)
3002                                 return r;
3003
3004                         /* For now, when we fail to shrink an ext4 image we'll not try again via the
3005                          * bisection logic. We might add that later, but given this involves shelling out
3006                          * multiple programs, it's a bit too cumbersome for my taste. */
3007
3008                         worked = true;
3009                         current_fs_size = try_fs_size;
3010                 }
3011
3012                 if (new_fs_size > old_fs_size) /* If we are growing we are done after one iteration */
3013                         break;
3014
3015                 /* If we are shrinking then let's adjust our bisection boundaries and try again. */
3016                 if (worked)
3017                         upper_boundary = MIN(upper_boundary, try_fs_size);
3018                 else
3019                         lower_boundary = MAX(lower_boundary, try_fs_size);
3020
3021                 /* OK, this attempt to shrink didn't work. Let's try between the old size and what worked. */
3022                 if (lower_boundary >= upper_boundary) {
3023                         log_debug("Image can't be shrunk further (range to try is empty).");
3024                         break;
3025                 }
3026
3027                 /* Let's find a new value to try half-way between the lower boundary and the upper boundary
3028                  * to try now. */
3029                 try_fs_size = DISK_SIZE_ROUND_DOWN(lower_boundary + (upper_boundary - lower_boundary) / 2);
3030                 if (try_fs_size <= lower_boundary || try_fs_size >= upper_boundary) {
3031                         log_debug("Image can't be shrunk further (remaining range to try too small).");
3032                         break;
3033                 }
3034         }
3035
3036         log_debug("Bisection loop completed after %u iterations.", n_iterations);
3037
3038         if (ret_fs_size)
3039                 *ret_fs_size = current_fs_size;
3040
3041         return 0;
3042 }
3043
3044 static int resize_image_loop(
3045                 UserRecord *h,
3046                 HomeSetup *setup,
3047                 uint64_t old_image_size,
3048                 uint64_t new_image_size,
3049                 uint64_t *ret_image_size) {
3050
3051         uint64_t current_image_size;
3052         unsigned n_iterations = 0;
3053         int r;
3054
3055         assert(h);
3056         assert(setup);
3057         assert(setup->image_fd >= 0);
3058
3059         /* A bisection loop trying to find the closest size to what the user asked for. (Well, we bisect like
3060          * this only when we *grow* the image — if we shrink the image then there's no need to bisect.) */
3061
3062         current_image_size = old_image_size;
3063         for (uint64_t lower_boundary = old_image_size, upper_boundary = new_image_size, try_image_size = new_image_size;;) {
3064                 bool worked;
3065
3066                 n_iterations++;
3067
3068                 r = home_truncate(h, setup->image_fd, try_image_size);
3069                 if (r < 0) {
3070                         if (!ERRNO_IS_DISK_SPACE(r) || new_image_size < old_image_size) /* Not a disk space issue? Not trying to grow? */
3071                                 return r;
3072
3073                         log_debug_errno(r, "Growing from %s to %s didn't work, not enough space on backing disk.", FORMAT_BYTES(current_image_size), FORMAT_BYTES(try_image_size));
3074                         worked = false;
3075                 } else if (r > 0) { /* Success: allocation worked */
3076                         log_debug("Resizing from %s to %s via allocation worked successfully.", FORMAT_BYTES(current_image_size), FORMAT_BYTES(try_image_size));
3077                         current_image_size = try_image_size;
3078                         worked = true;
3079                 } else { /* Success, but through truncation, not allocation. */
3080                         log_debug("Resizing from %s to %s via truncation worked successfully.", FORMAT_BYTES(old_image_size), FORMAT_BYTES(try_image_size));
3081                         current_image_size = try_image_size;
3082                         break; /* there's no point in the bisection logic if this was plain truncation and
3083                                 * not allocation, let's exit immediately. */
3084                 }
3085
3086                 if (new_image_size < old_image_size) /* If we are shrinking we are done after one iteration */
3087                         break;
3088
3089                 /* If we are growing then let's adjust our bisection boundaries and try again */
3090                 if (worked)
3091                         lower_boundary = MAX(lower_boundary, try_image_size);
3092                 else
3093                         upper_boundary = MIN(upper_boundary, try_image_size);
3094
3095                 if (lower_boundary >= upper_boundary) {
3096                         log_debug("Image can't be grown further (range to try is empty).");
3097                         break;
3098                 }
3099
3100                 try_image_size = DISK_SIZE_ROUND_DOWN(lower_boundary + (upper_boundary - lower_boundary) / 2);
3101                 if (try_image_size <= lower_boundary || try_image_size >= upper_boundary) {
3102                         log_debug("Image can't be grown further (remaining range to try too small).");
3103                         break;
3104                 }
3105         }
3106
3107         log_debug("Bisection loop completed after %u iterations.", n_iterations);
3108
3109         if (ret_image_size)
3110                 *ret_image_size = current_image_size;
3111
3112         return 0;
3113 }
3114
3115 int home_resize_luks(
3116                 UserRecord *h,
3117                 HomeSetupFlags flags,
3118                 HomeSetup *setup,
3119                 PasswordCache *cache,
3120                 UserRecord **ret_home) {
3121
3122         uint64_t old_image_size, new_image_size, old_fs_size, new_fs_size, crypto_offset, crypto_offset_bytes,
3123                 new_partition_size, smallest_fs_size, resized_fs_size;
3124         _cleanup_(user_record_unrefp) UserRecord *header_home = NULL, *embedded_home = NULL, *new_home = NULL;
3125         _cleanup_(fdisk_unref_tablep) struct fdisk_table *table = NULL;
3126         struct fdisk_partition *partition = NULL;
3127         _cleanup_close_ int opened_image_fd = -EBADF;
3128         _cleanup_free_ char *whole_disk = NULL;
3129         int r, resize_type, image_fd = -EBADF;
3130         sd_id128_t disk_uuid;
3131         const char *ip, *ipo;
3132         struct statfs sfs;
3133         struct stat st;
3134         enum {
3135                 INTENTION_DONT_KNOW = 0,    /* These happen to match the return codes of CMP() */
3136                 INTENTION_SHRINK = -1,
3137                 INTENTION_GROW = 1,
3138         } intention = INTENTION_DONT_KNOW;
3139
3140         assert(h);
3141         assert(user_record_storage(h) == USER_LUKS);
3142         assert(setup);
3143
3144         r = dlopen_cryptsetup();
3145         if (r < 0)
3146                 return r;
3147
3148         assert_se(ipo = user_record_image_path(h));
3149         ip = strdupa_safe(ipo); /* copy out since original might change later in home record object */
3150
3151         if (setup->image_fd < 0) {
3152                 setup->image_fd = open_image_file(h, NULL, &st);
3153                 if (setup->image_fd < 0)
3154                         return setup->image_fd;
3155         } else {
3156                 if (fstat(setup->image_fd, &st) < 0)
3157                         return log_error_errno(errno, "Failed to stat image file %s: %m", ip);
3158         }
3159
3160         image_fd = setup->image_fd;
3161
3162         if (S_ISBLK(st.st_mode)) {
3163                 dev_t parent;
3164
3165                 r = block_get_whole_disk(st.st_rdev, &parent);
3166                 if (r < 0)
3167                         return log_error_errno(r, "Failed to acquire whole block device for %s: %m", ip);
3168                 if (r > 0) {
3169                         /* If we shall resize a file system on a partition device, then let's figure out the
3170                          * whole disk device and operate on that instead, since we need to rewrite the
3171                          * partition table to resize the partition. */
3172
3173                         log_info("Operating on partition device %s, using parent device.", ip);
3174
3175                         opened_image_fd = r = device_open_from_devnum(S_IFBLK, parent, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK, &whole_disk);
3176                         if (r < 0)
3177                                 return log_error_errno(r, "Failed to open whole block device for %s: %m", ip);
3178
3179                         image_fd = opened_image_fd;
3180
3181                         if (fstat(image_fd, &st) < 0)
3182                                 return log_error_errno(errno, "Failed to stat whole block device %s: %m", whole_disk);
3183                 } else
3184                         log_info("Operating on whole block device %s.", ip);
3185
3186                 r = blockdev_get_device_size(image_fd, &old_image_size);
3187                 if (r < 0)
3188                         return log_error_errno(r, "Failed to determine size of original block device: %m");
3189
3190                 if (flock(image_fd, LOCK_EX) < 0) /* make sure udev doesn't read from it while we operate on the device */
3191                         return log_error_errno(errno, "Failed to lock block device %s: %m", ip);
3192
3193                 new_image_size = old_image_size; /* we can't resize physical block devices */
3194         } else {
3195                 r = stat_verify_regular(&st);
3196                 if (r < 0)
3197                         return log_error_errno(r, "Image %s is not a block device nor regular file: %m", ip);
3198
3199                 old_image_size = st.st_size;
3200
3201                 /* Note an asymmetry here: when we operate on loopback files the specified disk size we get we
3202                  * apply onto the loopback file as a whole. When we operate on block devices we instead apply
3203                  * to the partition itself only. */
3204
3205                 if (FLAGS_SET(flags, HOME_SETUP_RESIZE_MINIMIZE)) {
3206                         new_image_size = 0;
3207                         intention = INTENTION_SHRINK;
3208                 } else {
3209                         uint64_t new_image_size_rounded;
3210
3211                         new_image_size_rounded = DISK_SIZE_ROUND_DOWN(h->disk_size);
3212
3213                         if (old_image_size >= new_image_size_rounded && old_image_size <= h->disk_size) {
3214                                 /* If exact match, or a match after we rounded down, don't do a thing */
3215                                 log_info("Image size already matching, skipping operation.");
3216                                 return 0;
3217                         }
3218
3219                         new_image_size = new_image_size_rounded;
3220                         intention = CMP(new_image_size, old_image_size); /* Is this a shrink */
3221                 }
3222         }
3223
3224         r = home_setup_luks(
3225                         h,
3226                         flags,
3227                         whole_disk,
3228                         setup,
3229                         cache,
3230                         FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES) ? NULL : &header_home);
3231         if (r < 0)
3232                 return r;
3233
3234         if (!FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES)) {
3235                 r = home_load_embedded_identity(h, setup->root_fd, header_home, USER_RECONCILE_REQUIRE_NEWER_OR_EQUAL, cache, &embedded_home, &new_home);
3236                 if (r < 0)
3237                         return r;
3238         }
3239
3240         r = home_maybe_shift_uid(h, flags, setup);
3241         if (r < 0)
3242                 return r;
3243
3244         log_info("offset = %" PRIu64 ", size = %" PRIu64 ", image = %" PRIu64, setup->partition_offset, setup->partition_size, old_image_size);
3245
3246         if ((UINT64_MAX - setup->partition_offset) < setup->partition_size ||
3247             setup->partition_offset + setup->partition_size > old_image_size)
3248                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Old partition doesn't fit in backing storage, refusing.");
3249
3250         /* Get target partition information in here for new_partition_size calculation */
3251         r = prepare_resize_partition(
3252                         image_fd,
3253                         setup->partition_offset,
3254                         setup->partition_size,
3255                         &disk_uuid,
3256                         &table,
3257                         &partition);
3258         if (r < 0)
3259                 return r;
3260
3261         if (S_ISREG(st.st_mode)) {
3262                 uint64_t partition_table_extra, largest_size;
3263
3264                 partition_table_extra = old_image_size - setup->partition_size;
3265
3266                 r = get_largest_image_size(setup->image_fd, &st, &largest_size);
3267                 if (r < 0)
3268                         return r;
3269                 if (new_image_size > largest_size)
3270                         new_image_size = largest_size;
3271
3272                 if (new_image_size < partition_table_extra)
3273                         new_image_size = partition_table_extra;
3274
3275                 new_partition_size = DISK_SIZE_ROUND_DOWN(new_image_size - partition_table_extra);
3276         } else {
3277                 assert(S_ISBLK(st.st_mode));
3278
3279                 if (FLAGS_SET(flags, HOME_SETUP_RESIZE_MINIMIZE)) {
3280                         new_partition_size = 0;
3281                         intention = INTENTION_SHRINK;
3282                 } else {
3283                         uint64_t new_partition_size_rounded = DISK_SIZE_ROUND_DOWN(h->disk_size);
3284
3285                         if (h->disk_size == UINT64_MAX && partition) {
3286                                 r = get_maximum_partition_size(image_fd, partition, &new_partition_size_rounded);
3287                                 if (r < 0)
3288                                         return r;
3289                         }
3290
3291                         if (setup->partition_size >= new_partition_size_rounded &&
3292                             setup->partition_size <= h->disk_size) {
3293                                 log_info("Partition size already matching, skipping operation.");
3294                                 return 0;
3295                         }
3296
3297                         new_partition_size = new_partition_size_rounded;
3298                         intention = CMP(new_partition_size, setup->partition_size);
3299                 }
3300         }
3301
3302         if ((UINT64_MAX - setup->partition_offset) < new_partition_size ||
3303             setup->partition_offset + new_partition_size > new_image_size)
3304                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "New partition doesn't fit into backing storage, refusing.");
3305
3306         crypto_offset = sym_crypt_get_data_offset(setup->crypt_device);
3307         if (crypto_offset > UINT64_MAX/512U)
3308                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "LUKS2 data offset out of range, refusing.");
3309         crypto_offset_bytes = (uint64_t) crypto_offset * 512U;
3310         if (setup->partition_size <= crypto_offset_bytes)
3311                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Weird, old crypto payload offset doesn't actually fit in partition size?");
3312
3313         /* Make sure at least the LUKS header fit in */
3314         if (new_partition_size <= crypto_offset_bytes) {
3315                 uint64_t add;
3316
3317                 add = DISK_SIZE_ROUND_UP(crypto_offset_bytes) - new_partition_size;
3318                 new_partition_size += add;
3319                 if (S_ISREG(st.st_mode))
3320                         new_image_size += add;
3321         }
3322
3323         old_fs_size = setup->partition_size - crypto_offset_bytes;
3324         new_fs_size = DISK_SIZE_ROUND_DOWN(new_partition_size - crypto_offset_bytes);
3325
3326         r = get_smallest_fs_size(setup->root_fd, &smallest_fs_size);
3327         if (r < 0)
3328                 return r;
3329
3330         if (new_fs_size < smallest_fs_size) {
3331                 uint64_t add;
3332
3333                 add = DISK_SIZE_ROUND_UP(smallest_fs_size) - new_fs_size;
3334                 new_fs_size += add;
3335                 new_partition_size += add;
3336                 if (S_ISREG(st.st_mode))
3337                         new_image_size += add;
3338         }
3339
3340         if (new_fs_size == old_fs_size) {
3341                 log_info("New file system size identical to old file system size, skipping operation.");
3342                 return 0;
3343         }
3344
3345         if (FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_GROW) && new_fs_size > old_fs_size) {
3346                 log_info("New file system size would be larger than old, but shrinking requested, skipping operation.");
3347                 return 0;
3348         }
3349
3350         if (FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_SHRINK) && new_fs_size < old_fs_size) {
3351                 log_info("New file system size would be smaller than old, but growing requested, skipping operation.");
3352                 return 0;
3353         }
3354
3355         if (CMP(new_fs_size, old_fs_size) != intention) {
3356                 if (intention < 0)
3357                         log_info("Shrink operation would enlarge file system, skipping operation.");
3358                 else {
3359                         assert(intention > 0);
3360                         log_info("Grow operation would shrink file system, skipping operation.");
3361                 }
3362                 return 0;
3363         }
3364
3365         /* Before we start doing anything, let's figure out if we actually can */
3366         resize_type = can_resize_fs(setup->root_fd, old_fs_size, new_fs_size);
3367         if (resize_type < 0)
3368                 return resize_type;
3369         if (resize_type == CAN_RESIZE_OFFLINE && FLAGS_SET(flags, HOME_SETUP_ALREADY_ACTIVATED))
3370                 return log_error_errno(SYNTHETIC_ERRNO(ETXTBSY), "File systems of this type can only be resized offline, but is currently online.");
3371
3372         log_info("Ready to resize image size %s %s %s, partition size %s %s %s, file system size %s %s %s.",
3373                  FORMAT_BYTES(old_image_size),
3374                  special_glyph(SPECIAL_GLYPH_ARROW_RIGHT),
3375                  FORMAT_BYTES(new_image_size),
3376                  FORMAT_BYTES(setup->partition_size),
3377                  special_glyph(SPECIAL_GLYPH_ARROW_RIGHT),
3378                  FORMAT_BYTES(new_partition_size),
3379                  FORMAT_BYTES(old_fs_size),
3380                  special_glyph(SPECIAL_GLYPH_ARROW_RIGHT),
3381                  FORMAT_BYTES(new_fs_size));
3382
3383         if (new_fs_size > old_fs_size) { /* → Grow */
3384
3385                 if (S_ISREG(st.st_mode)) {
3386                         uint64_t resized_image_size;
3387
3388                         /* Grow file size */
3389                         r = resize_image_loop(h, setup, old_image_size, new_image_size, &resized_image_size);
3390                         if (r < 0)
3391                                 return r;
3392
3393                         if (resized_image_size == old_image_size) {
3394                                 log_info("Couldn't change image size.");
3395                                 return 0;
3396                         }
3397
3398                         assert(resized_image_size > old_image_size);
3399
3400                         log_info("Growing of image file from %s to %s completed.", FORMAT_BYTES(old_image_size), FORMAT_BYTES(resized_image_size));
3401
3402                         if (resized_image_size < new_image_size) {
3403                                 uint64_t sub;
3404
3405                                 /* If the growing we managed to do is smaller than what we wanted we need to
3406                                  * adjust the partition/file system sizes we are going for, too */
3407                                 sub = new_image_size - resized_image_size;
3408                                 assert(new_partition_size >= sub);
3409                                 new_partition_size -= sub;
3410                                 assert(new_fs_size >= sub);
3411                                 new_fs_size -= sub;
3412                         }
3413
3414                         new_image_size = resized_image_size;
3415                 } else {
3416                         assert(S_ISBLK(st.st_mode));
3417                         assert(new_image_size == old_image_size);
3418                 }
3419
3420                 /* Make sure loopback device sees the new bigger size */
3421                 r = loop_device_refresh_size(setup->loop, UINT64_MAX, new_partition_size);
3422                 if (r == -ENOTTY)
3423                         log_debug_errno(r, "Device is not a loopback device, not refreshing size.");
3424                 else if (r < 0)
3425                         return log_error_errno(r, "Failed to refresh loopback device size: %m");
3426                 else
3427                         log_info("Refreshing loop device size completed.");
3428
3429                 r = apply_resize_partition(image_fd, disk_uuid, table, partition, new_partition_size);
3430                 if (r < 0)
3431                         return r;
3432                 if (r > 0)
3433                         log_info("Growing of partition completed.");
3434
3435                 if (S_ISBLK(st.st_mode) && ioctl(image_fd, BLKRRPART, 0) < 0)
3436                         log_debug_errno(errno, "BLKRRPART failed on block device, ignoring: %m");
3437
3438                 /* Tell LUKS about the new bigger size too */
3439                 r = sym_crypt_resize(setup->crypt_device, setup->dm_name, new_fs_size / 512U);
3440                 if (r < 0)
3441                         return log_error_errno(r, "Failed to grow LUKS device: %m");
3442
3443                 log_info("LUKS device growing completed.");
3444         } else {
3445                 /* → Shrink */
3446
3447                 if (!FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES)) {
3448                         r = home_store_embedded_identity(new_home, setup->root_fd, h->uid, embedded_home);
3449                         if (r < 0)
3450                                 return r;
3451                 }
3452
3453                 if (S_ISREG(st.st_mode)) {
3454                         if (user_record_luks_discard(h))
3455                                 /* Before we shrink, let's trim the file system, so that we need less space on disk during the shrinking */
3456                                 (void) run_fitrim(setup->root_fd);
3457                         else {
3458                                 /* If discard is off, let's ensure all backing blocks are allocated, so that our resize operation doesn't fail half-way */
3459                                 r = run_fallocate(image_fd, &st);
3460                                 if (r < 0)
3461                                         return r;
3462                         }
3463                 }
3464         }
3465
3466         /* Now try to resize the file system. The requested size might not always be possible, in which case
3467          * we'll try to get as close as we can get. The result is returned in 'resized_fs_size' */
3468         r = resize_fs_loop(h, setup, resize_type, old_fs_size, new_fs_size, &resized_fs_size);
3469         if (r < 0)
3470                 return r;
3471
3472         if (resized_fs_size == old_fs_size) {
3473                 log_info("Couldn't change file system size.");
3474                 return 0;
3475         }
3476
3477         log_info("File system resizing from %s to %s completed.", FORMAT_BYTES(old_fs_size), FORMAT_BYTES(resized_fs_size));
3478
3479         if (resized_fs_size > new_fs_size) {
3480                 uint64_t add;
3481
3482                 /* If the shrinking we managed to do is larger than what we wanted we need to adjust the partition/image sizes. */
3483                 add = resized_fs_size - new_fs_size;
3484                 new_partition_size += add;
3485                 if (S_ISREG(st.st_mode))
3486                         new_image_size += add;
3487         }
3488
3489         new_fs_size = resized_fs_size;
3490
3491         /* Immediately sync afterwards */
3492         r = home_sync_and_statfs(setup->root_fd, NULL);
3493         if (r < 0)
3494                 return r;
3495
3496         if (new_fs_size < old_fs_size) { /* → Shrink */
3497
3498                 /* Shrink the LUKS device now, matching the new file system size */
3499                 r = sym_crypt_resize(setup->crypt_device, setup->dm_name, new_fs_size / 512);
3500                 if (r < 0)
3501                         return log_error_errno(r, "Failed to shrink LUKS device: %m");
3502
3503                 log_info("LUKS device shrinking completed.");
3504
3505                 /* Refresh the loop devices size */
3506                 r = loop_device_refresh_size(setup->loop, UINT64_MAX, new_partition_size);
3507                 if (r == -ENOTTY)
3508                         log_debug_errno(r, "Device is not a loopback device, not refreshing size.");
3509                 else if (r < 0)
3510                         return log_error_errno(r, "Failed to refresh loopback device size: %m");
3511                 else
3512                         log_info("Refreshing loop device size completed.");
3513
3514                 if (S_ISREG(st.st_mode)) {
3515                         /* Shrink the image file */
3516                         if (ftruncate(image_fd, new_image_size) < 0)
3517                                 return log_error_errno(errno, "Failed to shrink image file %s: %m", ip);
3518
3519                         log_info("Shrinking of image file completed.");
3520                 } else {
3521                         assert(S_ISBLK(st.st_mode));
3522                         assert(new_image_size == old_image_size);
3523                 }
3524
3525                 r = apply_resize_partition(image_fd, disk_uuid, table, partition, new_partition_size);
3526                 if (r < 0)
3527                         return r;
3528                 if (r > 0)
3529                         log_info("Shrinking of partition completed.");
3530
3531                 if (S_ISBLK(st.st_mode) && ioctl(image_fd, BLKRRPART, 0) < 0)
3532                         log_debug_errno(errno, "BLKRRPART failed on block device, ignoring: %m");
3533
3534         } else { /* → Grow */
3535                 if (!FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES)) {
3536                         r = home_store_embedded_identity(new_home, setup->root_fd, h->uid, embedded_home);
3537                         if (r < 0)
3538                                 return r;
3539                 }
3540         }
3541
3542         if (!FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES)) {
3543                 r = home_store_header_identity_luks(new_home, setup, header_home);
3544                 if (r < 0)
3545                         return r;
3546
3547                 r = home_extend_embedded_identity(new_home, h, setup);
3548                 if (r < 0)
3549                         return r;
3550         }
3551
3552         if (user_record_luks_discard(h))
3553                 (void) run_fitrim(setup->root_fd);
3554
3555         r = home_sync_and_statfs(setup->root_fd, &sfs);
3556         if (r < 0)
3557                 return r;
3558
3559         if (!FLAGS_SET(flags, HOME_SETUP_RESIZE_DONT_UNDO)) {
3560                 r = home_setup_done(setup);
3561                 if (r < 0)
3562                         return r;
3563         }
3564
3565         log_info("Resizing completed.");
3566
3567         print_size_summary(new_image_size, new_fs_size, &sfs);
3568
3569         if (ret_home)
3570                 *ret_home = TAKE_PTR(new_home);
3571
3572         return 0;
3573 }
3574
3575 int home_passwd_luks(
3576                 UserRecord *h,
3577                 HomeSetupFlags flags,
3578                 HomeSetup *setup,
3579                 const PasswordCache *cache, /* the passwords acquired via PKCS#11/FIDO2 security tokens */
3580                 char **effective_passwords  /* new passwords */) {
3581
3582         size_t volume_key_size, max_key_slots, n_effective;
3583         _cleanup_(erase_and_freep) void *volume_key = NULL;
3584         struct crypt_pbkdf_type good_pbkdf, minimal_pbkdf;
3585         const char *type;
3586         char **list;
3587         int r;
3588
3589         assert(h);
3590         assert(user_record_storage(h) == USER_LUKS);
3591         assert(setup);
3592
3593         r = dlopen_cryptsetup();
3594         if (r < 0)
3595                 return r;
3596
3597         type = sym_crypt_get_type(setup->crypt_device);
3598         if (!type)
3599                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine crypto device type.");
3600
3601         r = sym_crypt_keyslot_max(type);
3602         if (r <= 0)
3603                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine number of key slots.");
3604         max_key_slots = r;
3605
3606         r = sym_crypt_get_volume_key_size(setup->crypt_device);
3607         if (r <= 0)
3608                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine volume key size.");
3609         volume_key_size = (size_t) r;
3610
3611         volume_key = malloc(volume_key_size);
3612         if (!volume_key)
3613                 return log_oom();
3614
3615         r = -ENOKEY;
3616         FOREACH_POINTER(list,
3617                         cache ? cache->keyring_passswords : NULL,
3618                         cache ? cache->pkcs11_passwords : NULL,
3619                         cache ? cache->fido2_passwords : NULL,
3620                         h->password) {
3621
3622                 r = luks_try_passwords(h, setup->crypt_device, list, volume_key, &volume_key_size, NULL);
3623                 if (r != -ENOKEY)
3624                         break;
3625         }
3626         if (r == -ENOKEY)
3627                 return log_error_errno(SYNTHETIC_ERRNO(ENOKEY), "Failed to unlock LUKS superblock with supplied passwords.");
3628         if (r < 0)
3629                 return log_error_errno(r, "Failed to unlock LUKS superblock: %m");
3630
3631         n_effective = strv_length(effective_passwords);
3632
3633         build_good_pbkdf(&good_pbkdf, h);
3634         build_minimal_pbkdf(&minimal_pbkdf, h);
3635
3636         for (size_t i = 0; i < max_key_slots; i++) {
3637                 r = sym_crypt_keyslot_destroy(setup->crypt_device, i);
3638                 if (r < 0 && !IN_SET(r, -ENOENT, -EINVAL)) /* Returns EINVAL or ENOENT if there's no key in this slot already */
3639                         return log_error_errno(r, "Failed to destroy LUKS password: %m");
3640
3641                 if (i >= n_effective) {
3642                         if (r >= 0)
3643                                 log_info("Destroyed LUKS key slot %zu.", i);
3644                         continue;
3645                 }
3646
3647                 if (password_cache_contains(cache, effective_passwords[i])) { /* Is this a FIDO2 or PKCS#11 password? */
3648                         log_debug("Using minimal PBKDF for slot %zu", i);
3649                         r = sym_crypt_set_pbkdf_type(setup->crypt_device, &minimal_pbkdf);
3650                 } else {
3651                         log_debug("Using good PBKDF for slot %zu", i);
3652                         r = sym_crypt_set_pbkdf_type(setup->crypt_device, &good_pbkdf);
3653                 }
3654                 if (r < 0)
3655                         return log_error_errno(r, "Failed to tweak PBKDF for slot %zu: %m", i);
3656
3657                 r = sym_crypt_keyslot_add_by_volume_key(
3658                                 setup->crypt_device,
3659                                 i,
3660                                 volume_key,
3661                                 volume_key_size,
3662                                 effective_passwords[i],
3663                                 strlen(effective_passwords[i]));
3664                 if (r < 0)
3665                         return log_error_errno(r, "Failed to set up LUKS password: %m");
3666
3667                 log_info("Updated LUKS key slot %zu.", i);
3668
3669                 /* If we changed the password, then make sure to update the copy in the keyring, so that
3670                  * auto-rebalance continues to work. We only do this if we operate on an active home dir. */
3671                 if (i == 0 && FLAGS_SET(flags, HOME_SETUP_ALREADY_ACTIVATED))
3672                         upload_to_keyring(h, effective_passwords[i], NULL);
3673         }
3674
3675         return 1;
3676 }
3677
3678 int home_lock_luks(UserRecord *h, HomeSetup *setup) {
3679         const char *p;
3680         int r;
3681
3682         assert(h);
3683         assert(setup);
3684         assert(setup->root_fd < 0);
3685         assert(!setup->crypt_device);
3686
3687         r = acquire_open_luks_device(h, setup, /* graceful= */ false);
3688         if (r < 0)
3689                 return r;
3690
3691         log_info("Discovered used LUKS device %s.", setup->dm_node);
3692
3693         assert_se(p = user_record_home_directory(h));
3694         r = syncfs_path(AT_FDCWD, p);
3695         if (r < 0) /* Snake oil, but let's better be safe than sorry */
3696                 return log_error_errno(r, "Failed to synchronize file system %s: %m", p);
3697
3698         log_info("File system synchronized.");
3699
3700         /* Note that we don't invoke FIFREEZE here, it appears libcryptsetup/device-mapper already does that on its own for us */
3701
3702         r = sym_crypt_suspend(setup->crypt_device, setup->dm_name);
3703         if (r < 0)
3704                 return log_error_errno(r, "Failed to suspend cryptsetup device: %s: %m", setup->dm_node);
3705
3706         log_info("LUKS device suspended.");
3707         return 0;
3708 }
3709
3710 static int luks_try_resume(
3711                 struct crypt_device *cd,
3712                 const char *dm_name,
3713                 char **password) {
3714
3715         int r;
3716
3717         assert(cd);
3718         assert(dm_name);
3719
3720         STRV_FOREACH(pp, password) {
3721                 r = sym_crypt_resume_by_passphrase(
3722                                 cd,
3723                                 dm_name,
3724                                 CRYPT_ANY_SLOT,
3725                                 *pp,
3726                                 strlen(*pp));
3727                 if (r >= 0) {
3728                         log_info("Resumed LUKS device %s.", dm_name);
3729                         return 0;
3730                 }
3731
3732                 log_debug_errno(r, "Password %zu didn't work for resuming device: %m", (size_t) (pp - password));
3733         }
3734
3735         return -ENOKEY;
3736 }
3737
3738 int home_unlock_luks(UserRecord *h, HomeSetup *setup, const PasswordCache *cache) {
3739         char **list;
3740         int r;
3741
3742         assert(h);
3743         assert(setup);
3744         assert(!setup->crypt_device);
3745
3746         r = acquire_open_luks_device(h, setup, /* graceful= */ false);
3747         if (r < 0)
3748                 return r;
3749
3750         log_info("Discovered used LUKS device %s.", setup->dm_node);
3751
3752         r = -ENOKEY;
3753         FOREACH_POINTER(list,
3754                         cache ? cache->pkcs11_passwords : NULL,
3755                         cache ? cache->fido2_passwords : NULL,
3756                         h->password) {
3757                 r = luks_try_resume(setup->crypt_device, setup->dm_name, list);
3758                 if (r != -ENOKEY)
3759                         break;
3760         }
3761         if (r == -ENOKEY)
3762                 return log_error_errno(r, "No valid password for LUKS superblock.");
3763         if (r < 0)
3764                 return log_error_errno(r, "Failed to resume LUKS superblock: %m");
3765
3766         log_info("LUKS device resumed.");
3767         return 0;
3768 }
3769
3770 static int device_is_gone(HomeSetup *setup) {
3771         _cleanup_(sd_device_unrefp) sd_device *d = NULL;
3772         struct stat st;
3773         int r;
3774
3775         assert(setup);
3776
3777         if (!setup->dm_node)
3778                 return true;
3779
3780         if (stat(setup->dm_node, &st) < 0) {
3781                 if (errno != ENOENT)
3782                         return log_error_errno(errno, "Failed to stat block device node %s: %m", setup->dm_node);
3783
3784                 return true;
3785         }
3786
3787         r = sd_device_new_from_stat_rdev(&d, &st);
3788         if (r < 0) {
3789                 if (r != -ENODEV)
3790                         return log_error_errno(errno, "Failed to allocate device object from block device node %s: %m", setup->dm_node);
3791
3792                 return true;
3793         }
3794
3795         return false;
3796 }
3797
3798 static int device_monitor_handler(sd_device_monitor *monitor, sd_device *device, void *userdata) {
3799         HomeSetup *setup = ASSERT_PTR(userdata);
3800         int r;
3801
3802         if (!device_for_action(device, SD_DEVICE_REMOVE))
3803                 return 0;
3804
3805         /* We don't really care for the device object passed to us, we just check if the device node still
3806          * exists */
3807
3808         r = device_is_gone(setup);
3809         if (r < 0)
3810                 return r;
3811         if (r > 0) /* Yay! we are done! */
3812                 (void) sd_event_exit(sd_device_monitor_get_event(monitor), 0);
3813
3814         return 0;
3815 }
3816
3817 int wait_for_block_device_gone(HomeSetup *setup, usec_t timeout_usec) {
3818         _cleanup_(sd_device_monitor_unrefp) sd_device_monitor *m = NULL;
3819         _cleanup_(sd_event_unrefp) sd_event *event = NULL;
3820         int r;
3821
3822         assert(setup);
3823
3824         /* So here's the thing: we enable "deferred deactivation" on our dm-crypt volumes. This means they
3825          * are automatically torn down once not used anymore (i.e. once unmounted). Which is great. It also
3826          * means that when we deactivate a home directory and try to tear down the volume that backs it, it
3827          * possibly is already torn down or in the process of being torn down, since we race against the
3828          * automatic tearing down. Which is fine, we handle errors from that. However, we lose the ability to
3829          * naturally wait for the tear down operation to complete: if we are not the ones who tear down the
3830          * device we are also not the ones who naturally block on that operation. Hence let's add some code
3831          * to actively wait for the device to go away, via sd-device. We'll call this whenever tearing down a
3832          * LUKS device, to ensure the device is really really gone before we proceed. Net effect: "homectl
3833          * deactivate foo && homectl activate foo" will work reliably, i.e. deactivation immediately followed
3834          * by activation will work. Also, by the time deactivation completes we can guarantee that all data
3835          * is sync'ed down to the lowest block layer as all higher levels are fully and entirely
3836          * destructed. */
3837
3838         if (!setup->dm_name)
3839                 return 0;
3840
3841         assert(setup->dm_node);
3842         log_debug("Waiting until %s disappears.", setup->dm_node);
3843
3844         r = sd_event_new(&event);
3845         if (r < 0)
3846                 return log_error_errno(r, "Failed to allocate event loop: %m");
3847
3848         r = sd_device_monitor_new(&m);
3849         if (r < 0)
3850                 return log_error_errno(r, "Failed to allocate device monitor: %m");
3851
3852         r = sd_device_monitor_filter_add_match_subsystem_devtype(m, "block", "disk");
3853         if (r < 0)
3854                 return log_error_errno(r, "Failed to configure device monitor match: %m");
3855
3856         r = sd_device_monitor_attach_event(m, event);
3857         if (r < 0)
3858                 return log_error_errno(r, "Failed to attach device monitor to event loop: %m");
3859
3860         r = sd_device_monitor_start(m, device_monitor_handler, setup);
3861         if (r < 0)
3862                 return log_error_errno(r, "Failed to start device monitor: %m");
3863
3864         r = device_is_gone(setup);
3865         if (r < 0)
3866                 return r;
3867         if (r > 0) {
3868                 log_debug("%s has already disappeared before entering wait loop.", setup->dm_node);
3869                 return 0; /* gone already */
3870         }
3871
3872         if (timeout_usec != USEC_INFINITY) {
3873                 r = sd_event_add_time_relative(event, NULL, CLOCK_MONOTONIC, timeout_usec, 0, NULL, NULL);
3874                 if (r < 0)
3875                         return log_error_errno(r, "Failed to add timer event: %m");
3876         }
3877
3878         r = sd_event_loop(event);
3879         if (r < 0)
3880                 return log_error_errno(r, "Failed to run event loop: %m");
3881
3882         r = device_is_gone(setup);
3883         if (r < 0)
3884                 return r;
3885         if (r == 0)
3886                 return log_error_errno(r, "Device %s still around.", setup->dm_node);
3887
3888         log_debug("Successfully waited until device %s disappeared.", setup->dm_node);
3889         return 0;
3890 }
3891
3892 int home_auto_shrink_luks(UserRecord *h, HomeSetup *setup, PasswordCache *cache) {
3893         struct statfs sfs;
3894         int r;
3895
3896         assert(h);
3897         assert(user_record_storage(h) == USER_LUKS);
3898         assert(setup);
3899         assert(setup->root_fd >= 0);
3900
3901         if (user_record_auto_resize_mode(h) != AUTO_RESIZE_SHRINK_AND_GROW)
3902                 return 0;
3903
3904         if (fstatfs(setup->root_fd, &sfs) < 0)
3905                 return log_error_errno(errno, "Failed to statfs home directory: %m");
3906
3907         if (!fs_can_online_shrink_and_grow(sfs.f_type)) {
3908                 log_debug("Not auto-shrinking file system, since selected file system cannot do both online shrink and grow.");
3909                 return 0;
3910         }
3911
3912         r = home_resize_luks(
3913                         h,
3914                         HOME_SETUP_ALREADY_ACTIVATED|
3915                         HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES|
3916                         HOME_SETUP_RESIZE_MINIMIZE|
3917                         HOME_SETUP_RESIZE_DONT_GROW|
3918                         HOME_SETUP_RESIZE_DONT_UNDO,
3919                         setup,
3920                         cache,
3921                         NULL);
3922         if (r < 0)
3923                 return r;
3924
3925         return 1;
3926 }