summaryrefslogtreecommitdiffstats
path: root/lib/luks1
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-17 08:06:26 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-17 08:06:26 +0000
commit1660d4b7a65d9ad2ce0deaa19d35579ca4084ac5 (patch)
tree6cf8220b628ebd2ccfc1375dd6516c6996e9abcc /lib/luks1
parentInitial commit. (diff)
downloadcryptsetup-1660d4b7a65d9ad2ce0deaa19d35579ca4084ac5.tar.xz
cryptsetup-1660d4b7a65d9ad2ce0deaa19d35579ca4084ac5.zip
Adding upstream version 2:2.6.1.upstream/2%2.6.1upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'lib/luks1')
-rw-r--r--lib/luks1/af.c170
-rw-r--r--lib/luks1/af.h67
-rw-r--r--lib/luks1/keyencryption.c268
-rw-r--r--lib/luks1/keymanage.c1300
-rw-r--r--lib/luks1/luks.h194
5 files changed, 1999 insertions, 0 deletions
diff --git a/lib/luks1/af.c b/lib/luks1/af.c
new file mode 100644
index 0000000..76afeac
--- /dev/null
+++ b/lib/luks1/af.c
@@ -0,0 +1,170 @@
+/*
+ * AFsplitter - Anti forensic information splitter
+ *
+ * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
+ * Copyright (C) 2009-2023 Red Hat, Inc. All rights reserved.
+ *
+ * AFsplitter diffuses information over a large stripe of data,
+ * therefore supporting secure data destruction.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <stddef.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+#include "internal.h"
+#include "af.h"
+
+static void XORblock(const char *src1, const char *src2, char *dst, size_t n)
+{
+ size_t j;
+
+ for (j = 0; j < n; j++)
+ dst[j] = src1[j] ^ src2[j];
+}
+
+static int hash_buf(const char *src, char *dst, uint32_t iv,
+ size_t len, const char *hash_name)
+{
+ struct crypt_hash *hd = NULL;
+ char *iv_char = (char *)&iv;
+ int r;
+
+ iv = be32_to_cpu(iv);
+ if (crypt_hash_init(&hd, hash_name))
+ return -EINVAL;
+
+ if ((r = crypt_hash_write(hd, iv_char, sizeof(uint32_t))))
+ goto out;
+
+ if ((r = crypt_hash_write(hd, src, len)))
+ goto out;
+
+ r = crypt_hash_final(hd, dst, len);
+out:
+ crypt_hash_destroy(hd);
+ return r;
+}
+
+/*
+ * diffuse: Information spreading over the whole dataset with
+ * the help of hash function.
+ */
+static int diffuse(char *src, char *dst, size_t size, const char *hash_name)
+{
+ int r, hash_size = crypt_hash_size(hash_name);
+ unsigned int digest_size;
+ unsigned int i, blocks, padding;
+
+ if (hash_size <= 0)
+ return -EINVAL;
+ digest_size = hash_size;
+
+ blocks = size / digest_size;
+ padding = size % digest_size;
+
+ for (i = 0; i < blocks; i++) {
+ r = hash_buf(src + digest_size * i,
+ dst + digest_size * i,
+ i, (size_t)digest_size, hash_name);
+ if (r < 0)
+ return r;
+ }
+
+ if (padding) {
+ r = hash_buf(src + digest_size * i,
+ dst + digest_size * i,
+ i, (size_t)padding, hash_name);
+ if (r < 0)
+ return r;
+ }
+
+ return 0;
+}
+
+/*
+ * Information splitting. The amount of data is multiplied by
+ * blocknumbers. The same blocksize and blocknumbers values
+ * must be supplied to AF_merge to recover information.
+ */
+int AF_split(struct crypt_device *ctx, const char *src, char *dst,
+ size_t blocksize, unsigned int blocknumbers, const char *hash)
+{
+ unsigned int i;
+ char *bufblock;
+ int r;
+
+ bufblock = crypt_safe_alloc(blocksize);
+ if (!bufblock)
+ return -ENOMEM;
+
+ /* process everything except the last block */
+ for (i = 0; i < blocknumbers - 1; i++) {
+ r = crypt_random_get(ctx, dst + blocksize * i, blocksize, CRYPT_RND_NORMAL);
+ if (r < 0)
+ goto out;
+
+ XORblock(dst + blocksize * i, bufblock, bufblock, blocksize);
+ r = diffuse(bufblock, bufblock, blocksize, hash);
+ if (r < 0)
+ goto out;
+ }
+ /* the last block is computed */
+ XORblock(src, bufblock, dst + blocksize * i, blocksize);
+ r = 0;
+out:
+ crypt_safe_free(bufblock);
+ return r;
+}
+
+int AF_merge(const char *src, char *dst,
+ size_t blocksize, unsigned int blocknumbers, const char *hash)
+{
+ unsigned int i;
+ char *bufblock;
+ int r;
+
+ bufblock = crypt_safe_alloc(blocksize);
+ if (!bufblock)
+ return -ENOMEM;
+
+ for (i = 0; i < blocknumbers - 1; i++) {
+ XORblock(src + blocksize * i, bufblock, bufblock, blocksize);
+ r = diffuse(bufblock, bufblock, blocksize, hash);
+ if (r < 0)
+ goto out;
+ }
+ XORblock(src + blocksize * i, bufblock, dst, blocksize);
+ r = 0;
+out:
+ crypt_safe_free(bufblock);
+ return r;
+}
+
+/* Size of final split data including sector alignment */
+size_t AF_split_sectors(size_t blocksize, unsigned int blocknumbers)
+{
+ size_t af_size;
+
+ /* data material * stripes */
+ af_size = blocksize * blocknumbers;
+
+ /* round up to sector */
+ af_size = (af_size + (SECTOR_SIZE - 1)) / SECTOR_SIZE;
+
+ return af_size;
+}
diff --git a/lib/luks1/af.h b/lib/luks1/af.h
new file mode 100644
index 0000000..8a2bceb
--- /dev/null
+++ b/lib/luks1/af.h
@@ -0,0 +1,67 @@
+/*
+ * AFsplitter - Anti forensic information splitter
+ *
+ * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
+ * Copyright (C) 2009-2023 Red Hat, Inc. All rights reserved.
+ *
+ * AFsplitter diffuses information over a large stripe of data,
+ * therefore supporting secure data destruction.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+#ifndef INCLUDED_CRYPTSETUP_LUKS_AF_H
+#define INCLUDED_CRYPTSETUP_LUKS_AF_H
+
+#include <stddef.h>
+
+struct crypt_device;
+struct volume_key;
+
+/*
+ * AF_split operates on src and produces information split data in
+ * dst. src is assumed to be of the length blocksize. The data stripe
+ * dst points to must be capable of storing blocksize*blocknumbers.
+ * blocknumbers is the data multiplication factor.
+ *
+ * AF_merge does just the opposite: reproduces the information stored in
+ * src of the length blocksize*blocknumbers into dst of the length
+ * blocksize.
+ *
+ * On error, both functions return -1, 0 otherwise.
+ */
+
+int AF_split(struct crypt_device *ctx, const char *src, char *dst,
+ size_t blocksize, unsigned int blocknumbers, const char *hash);
+int AF_merge(const char *src, char *dst, size_t blocksize,
+ unsigned int blocknumbers, const char *hash);
+size_t AF_split_sectors(size_t blocksize, unsigned int blocknumbers);
+
+int LUKS_encrypt_to_storage(
+ char *src, size_t srcLength,
+ const char *cipher,
+ const char *cipher_mode,
+ struct volume_key *vk,
+ unsigned int sector,
+ struct crypt_device *ctx);
+
+int LUKS_decrypt_from_storage(
+ char *dst, size_t dstLength,
+ const char *cipher,
+ const char *cipher_mode,
+ struct volume_key *vk,
+ unsigned int sector,
+ struct crypt_device *ctx);
+
+#endif
diff --git a/lib/luks1/keyencryption.c b/lib/luks1/keyencryption.c
new file mode 100644
index 0000000..c1c8201
--- /dev/null
+++ b/lib/luks1/keyencryption.c
@@ -0,0 +1,268 @@
+/*
+ * LUKS - Linux Unified Key Setup
+ *
+ * Copyright (C) 2004-2006 Clemens Fruhwirth <clemens@endorphin.org>
+ * Copyright (C) 2009-2023 Red Hat, Inc. All rights reserved.
+ * Copyright (C) 2012-2023 Milan Broz
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include "luks.h"
+#include "af.h"
+#include "internal.h"
+
+static void _error_hint(struct crypt_device *ctx, const char *device,
+ const char *cipher, const char *mode, size_t keyLength)
+{
+ char *c, cipher_spec[MAX_CIPHER_LEN * 3];
+
+ if (snprintf(cipher_spec, sizeof(cipher_spec), "%s-%s", cipher, mode) < 0)
+ return;
+
+ log_err(ctx, _("Failed to setup dm-crypt key mapping for device %s.\n"
+ "Check that kernel supports %s cipher (check syslog for more info)."),
+ device, cipher_spec);
+
+ if (!strncmp(mode, "xts", 3) && (keyLength != 256 && keyLength != 512))
+ log_err(ctx, _("Key size in XTS mode must be 256 or 512 bits."));
+ else if (!(c = strchr(mode, '-')) || strlen(c) < 4)
+ log_err(ctx, _("Cipher specification should be in [cipher]-[mode]-[iv] format."));
+}
+
+static int LUKS_endec_template(char *src, size_t srcLength,
+ const char *cipher, const char *cipher_mode,
+ struct volume_key *vk,
+ unsigned int sector,
+ ssize_t (*func)(int, size_t, size_t, void *, size_t),
+ int mode,
+ struct crypt_device *ctx)
+{
+ char name[PATH_MAX], path[PATH_MAX];
+ char cipher_spec[MAX_CIPHER_LEN * 3];
+ struct crypt_dm_active_device dmd = {
+ .flags = CRYPT_ACTIVATE_PRIVATE,
+ };
+ int r, devfd = -1, remove_dev = 0;
+ size_t bsize, keyslot_alignment, alignment;
+
+ log_dbg(ctx, "Using dmcrypt to access keyslot area.");
+
+ bsize = device_block_size(ctx, crypt_metadata_device(ctx));
+ alignment = device_alignment(crypt_metadata_device(ctx));
+ if (!bsize || !alignment)
+ return -EINVAL;
+
+ if (bsize > LUKS_ALIGN_KEYSLOTS)
+ keyslot_alignment = LUKS_ALIGN_KEYSLOTS;
+ else
+ keyslot_alignment = bsize;
+ dmd.size = size_round_up(srcLength, keyslot_alignment) / SECTOR_SIZE;
+
+ if (mode == O_RDONLY)
+ dmd.flags |= CRYPT_ACTIVATE_READONLY;
+
+ if (snprintf(name, sizeof(name), "temporary-cryptsetup-%d", getpid()) < 0)
+ return -ENOMEM;
+ if (snprintf(path, sizeof(path), "%s/%s", dm_get_dir(), name) < 0)
+ return -ENOMEM;
+ if (snprintf(cipher_spec, sizeof(cipher_spec), "%s-%s", cipher, cipher_mode) < 0)
+ return -ENOMEM;
+
+ r = device_block_adjust(ctx, crypt_metadata_device(ctx), DEV_OK,
+ sector, &dmd.size, &dmd.flags);
+ if (r < 0) {
+ log_err(ctx, _("Device %s does not exist or access denied."),
+ device_path(crypt_metadata_device(ctx)));
+ return -EIO;
+ }
+
+ if (mode != O_RDONLY && dmd.flags & CRYPT_ACTIVATE_READONLY) {
+ log_err(ctx, _("Cannot write to device %s, permission denied."),
+ device_path(crypt_metadata_device(ctx)));
+ return -EACCES;
+ }
+
+ r = dm_crypt_target_set(&dmd.segment, 0, dmd.size,
+ crypt_metadata_device(ctx), vk, cipher_spec, 0, sector,
+ NULL, 0, SECTOR_SIZE);
+ if (r)
+ goto out;
+
+ r = dm_create_device(ctx, name, "TEMP", &dmd);
+ if (r < 0) {
+ if (r != -EACCES && r != -ENOTSUP)
+ _error_hint(ctx, device_path(crypt_metadata_device(ctx)),
+ cipher, cipher_mode, vk->keylength * 8);
+ r = -EIO;
+ goto out;
+ }
+ remove_dev = 1;
+
+ devfd = open(path, mode | O_DIRECT | O_SYNC);
+ if (devfd == -1) {
+ log_err(ctx, _("Failed to open temporary keystore device."));
+ r = -EIO;
+ goto out;
+ }
+
+ r = func(devfd, bsize, alignment, src, srcLength);
+ if (r < 0) {
+ log_err(ctx, _("Failed to access temporary keystore device."));
+ r = -EIO;
+ } else
+ r = 0;
+ out:
+ dm_targets_free(ctx, &dmd);
+ if (devfd != -1)
+ close(devfd);
+ if (remove_dev)
+ dm_remove_device(ctx, name, CRYPT_DEACTIVATE_FORCE);
+ return r;
+}
+
+int LUKS_encrypt_to_storage(char *src, size_t srcLength,
+ const char *cipher,
+ const char *cipher_mode,
+ struct volume_key *vk,
+ unsigned int sector,
+ struct crypt_device *ctx)
+{
+ struct device *device = crypt_metadata_device(ctx);
+ struct crypt_storage *s;
+ int devfd, r = 0;
+
+ /* Only whole sector writes supported */
+ if (MISALIGNED_512(srcLength))
+ return -EINVAL;
+
+ /* Encrypt buffer */
+ r = crypt_storage_init(&s, SECTOR_SIZE, cipher, cipher_mode, vk->key, vk->keylength, false);
+
+ if (r)
+ log_dbg(ctx, "Userspace crypto wrapper cannot use %s-%s (%d).",
+ cipher, cipher_mode, r);
+
+ /* Fallback to old temporary dmcrypt device */
+ if (r == -ENOTSUP || r == -ENOENT)
+ return LUKS_endec_template(src, srcLength, cipher, cipher_mode,
+ vk, sector, write_blockwise, O_RDWR, ctx);
+
+ if (r) {
+ _error_hint(ctx, device_path(device), cipher, cipher_mode,
+ vk->keylength * 8);
+ return r;
+ }
+
+ log_dbg(ctx, "Using userspace crypto wrapper to access keyslot area.");
+
+ r = crypt_storage_encrypt(s, 0, srcLength, src);
+ crypt_storage_destroy(s);
+
+ if (r)
+ return r;
+
+ r = -EIO;
+
+ /* Write buffer to device */
+ if (device_is_locked(device))
+ devfd = device_open_locked(ctx, device, O_RDWR);
+ else
+ devfd = device_open(ctx, device, O_RDWR);
+ if (devfd < 0)
+ goto out;
+
+ if (write_lseek_blockwise(devfd, device_block_size(ctx, device),
+ device_alignment(device), src, srcLength,
+ sector * SECTOR_SIZE) < 0)
+ goto out;
+
+ r = 0;
+out:
+ device_sync(ctx, device);
+ if (r)
+ log_err(ctx, _("IO error while encrypting keyslot."));
+
+ return r;
+}
+
+int LUKS_decrypt_from_storage(char *dst, size_t dstLength,
+ const char *cipher,
+ const char *cipher_mode,
+ struct volume_key *vk,
+ unsigned int sector,
+ struct crypt_device *ctx)
+{
+ struct device *device = crypt_metadata_device(ctx);
+ struct crypt_storage *s;
+ struct stat st;
+ int devfd, r = 0;
+
+ /* Only whole sector reads supported */
+ if (MISALIGNED_512(dstLength))
+ return -EINVAL;
+
+ r = crypt_storage_init(&s, SECTOR_SIZE, cipher, cipher_mode, vk->key, vk->keylength, false);
+
+ if (r)
+ log_dbg(ctx, "Userspace crypto wrapper cannot use %s-%s (%d).",
+ cipher, cipher_mode, r);
+
+ /* Fallback to old temporary dmcrypt device */
+ if (r == -ENOTSUP || r == -ENOENT)
+ return LUKS_endec_template(dst, dstLength, cipher, cipher_mode,
+ vk, sector, read_blockwise, O_RDONLY, ctx);
+
+ if (r) {
+ _error_hint(ctx, device_path(device), cipher, cipher_mode,
+ vk->keylength * 8);
+ return r;
+ }
+
+ log_dbg(ctx, "Using userspace crypto wrapper to access keyslot area.");
+
+ /* Read buffer from device */
+ if (device_is_locked(device))
+ devfd = device_open_locked(ctx, device, O_RDONLY);
+ else
+ devfd = device_open(ctx, device, O_RDONLY);
+ if (devfd < 0) {
+ log_err(ctx, _("Cannot open device %s."), device_path(device));
+ crypt_storage_destroy(s);
+ return -EIO;
+ }
+
+ if (read_lseek_blockwise(devfd, device_block_size(ctx, device),
+ device_alignment(device), dst, dstLength,
+ sector * SECTOR_SIZE) < 0) {
+ if (!fstat(devfd, &st) && (st.st_size < (off_t)dstLength))
+ log_err(ctx, _("Device %s is too small."), device_path(device));
+ else
+ log_err(ctx, _("IO error while decrypting keyslot."));
+
+ crypt_storage_destroy(s);
+ return -EIO;
+ }
+
+ /* Decrypt buffer */
+ r = crypt_storage_decrypt(s, 0, dstLength, dst);
+ crypt_storage_destroy(s);
+
+ return r;
+}
diff --git a/lib/luks1/keymanage.c b/lib/luks1/keymanage.c
new file mode 100644
index 0000000..fe49a00
--- /dev/null
+++ b/lib/luks1/keymanage.c
@@ -0,0 +1,1300 @@
+/*
+ * LUKS - Linux Unified Key Setup
+ *
+ * Copyright (C) 2004-2006 Clemens Fruhwirth <clemens@endorphin.org>
+ * Copyright (C) 2009-2023 Red Hat, Inc. All rights reserved.
+ * Copyright (C) 2013-2023 Milan Broz
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <errno.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <ctype.h>
+#include <uuid/uuid.h>
+#include <limits.h>
+
+#include "luks.h"
+#include "af.h"
+#include "internal.h"
+
+int LUKS_keyslot_area(const struct luks_phdr *hdr,
+ int keyslot,
+ uint64_t *offset,
+ uint64_t *length)
+{
+ if(keyslot >= LUKS_NUMKEYS || keyslot < 0)
+ return -EINVAL;
+
+ *offset = (uint64_t)hdr->keyblock[keyslot].keyMaterialOffset * SECTOR_SIZE;
+ *length = AF_split_sectors(hdr->keyBytes, LUKS_STRIPES) * SECTOR_SIZE;
+
+ return 0;
+}
+
+/* insertsort: because the array has 8 elements and it's mostly sorted. that's why */
+static void LUKS_sort_keyslots(const struct luks_phdr *hdr, int *array)
+{
+ int i, j, x;
+
+ for (i = 1; i < LUKS_NUMKEYS; i++) {
+ j = i;
+ while (j > 0 && hdr->keyblock[array[j-1]].keyMaterialOffset > hdr->keyblock[array[j]].keyMaterialOffset) {
+ x = array[j];
+ array[j] = array[j-1];
+ array[j-1] = x;
+ j--;
+ }
+ }
+}
+
+static int _is_not_lower(char *str, unsigned max_len)
+{
+ for(; *str && max_len; str++, max_len--)
+ if (isupper(*str))
+ return 1;
+ return 0;
+}
+
+static int _to_lower(char *str, unsigned max_len)
+{
+ int r = 0;
+
+ for(; *str && max_len; str++, max_len--)
+ if (isupper(*str)) {
+ *str = tolower(*str);
+ r = 1;
+ }
+
+ return r;
+}
+
+size_t LUKS_device_sectors(const struct luks_phdr *hdr)
+{
+ int sorted_areas[LUKS_NUMKEYS] = { 0, 1, 2, 3, 4, 5, 6, 7 };
+
+ LUKS_sort_keyslots(hdr, sorted_areas);
+
+ return hdr->keyblock[sorted_areas[LUKS_NUMKEYS-1]].keyMaterialOffset + AF_split_sectors(hdr->keyBytes, LUKS_STRIPES);
+}
+
+size_t LUKS_keyslots_offset(const struct luks_phdr *hdr)
+{
+ int sorted_areas[LUKS_NUMKEYS] = { 0, 1, 2, 3, 4, 5, 6, 7 };
+
+ LUKS_sort_keyslots(hdr, sorted_areas);
+
+ return hdr->keyblock[sorted_areas[0]].keyMaterialOffset;
+}
+
+static int LUKS_check_device_size(struct crypt_device *ctx, const struct luks_phdr *hdr, int falloc)
+{
+ struct device *device = crypt_metadata_device(ctx);
+ uint64_t dev_sectors, hdr_sectors;
+
+ if (!hdr->keyBytes)
+ return -EINVAL;
+
+ if (device_size(device, &dev_sectors)) {
+ log_dbg(ctx, "Cannot get device size for device %s.", device_path(device));
+ return -EIO;
+ }
+
+ dev_sectors >>= SECTOR_SHIFT;
+ hdr_sectors = LUKS_device_sectors(hdr);
+ log_dbg(ctx, "Key length %u, device size %" PRIu64 " sectors, header size %"
+ PRIu64 " sectors.", hdr->keyBytes, dev_sectors, hdr_sectors);
+
+ if (hdr_sectors > dev_sectors) {
+ /* If it is header file, increase its size */
+ if (falloc && !device_fallocate(device, hdr_sectors << SECTOR_SHIFT))
+ return 0;
+
+ log_err(ctx, _("Device %s is too small. (LUKS1 requires at least %" PRIu64 " bytes.)"),
+ device_path(device), hdr_sectors * SECTOR_SIZE);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int LUKS_check_keyslots(struct crypt_device *ctx, const struct luks_phdr *phdr)
+{
+ int i, prev, next, sorted_areas[LUKS_NUMKEYS] = { 0, 1, 2, 3, 4, 5, 6, 7 };
+ uint32_t secs_per_stripes = AF_split_sectors(phdr->keyBytes, LUKS_STRIPES);
+
+ LUKS_sort_keyslots(phdr, sorted_areas);
+
+ /* Check keyslot to prevent access outside of header and keyslot area */
+ for (i = 0; i < LUKS_NUMKEYS; i++) {
+ /* enforce stripes == 4000 */
+ if (phdr->keyblock[i].stripes != LUKS_STRIPES) {
+ log_dbg(ctx, "Invalid stripes count %u in keyslot %u.",
+ phdr->keyblock[i].stripes, i);
+ log_err(ctx, _("LUKS keyslot %u is invalid."), i);
+ return -1;
+ }
+
+ /* First sectors is the header itself */
+ if (phdr->keyblock[i].keyMaterialOffset * SECTOR_SIZE < sizeof(*phdr)) {
+ log_dbg(ctx, "Invalid offset %u in keyslot %u.",
+ phdr->keyblock[i].keyMaterialOffset, i);
+ log_err(ctx, _("LUKS keyslot %u is invalid."), i);
+ return -1;
+ }
+
+ /* Ignore following check for detached header where offset can be zero. */
+ if (phdr->payloadOffset == 0)
+ continue;
+
+ if (phdr->payloadOffset <= phdr->keyblock[i].keyMaterialOffset) {
+ log_dbg(ctx, "Invalid offset %u in keyslot %u (beyond data area offset %u).",
+ phdr->keyblock[i].keyMaterialOffset, i,
+ phdr->payloadOffset);
+ log_err(ctx, _("LUKS keyslot %u is invalid."), i);
+ return -1;
+ }
+
+ if (phdr->payloadOffset < (phdr->keyblock[i].keyMaterialOffset + secs_per_stripes)) {
+ log_dbg(ctx, "Invalid keyslot size %u (offset %u, stripes %u) in "
+ "keyslot %u (beyond data area offset %u).",
+ secs_per_stripes,
+ phdr->keyblock[i].keyMaterialOffset,
+ phdr->keyblock[i].stripes,
+ i, phdr->payloadOffset);
+ log_err(ctx, _("LUKS keyslot %u is invalid."), i);
+ return -1;
+ }
+ }
+
+ /* check no keyslot overlaps with each other */
+ for (i = 1; i < LUKS_NUMKEYS; i++) {
+ prev = sorted_areas[i-1];
+ next = sorted_areas[i];
+ if (phdr->keyblock[next].keyMaterialOffset <
+ (phdr->keyblock[prev].keyMaterialOffset + secs_per_stripes)) {
+ log_dbg(ctx, "Not enough space in LUKS keyslot %d.", prev);
+ log_err(ctx, _("LUKS keyslot %u is invalid."), prev);
+ return -1;
+ }
+ }
+ /* do not check last keyslot on purpose, it must be tested in device size check */
+
+ return 0;
+}
+
+static const char *dbg_slot_state(crypt_keyslot_info ki)
+{
+ switch(ki) {
+ case CRYPT_SLOT_INACTIVE:
+ return "INACTIVE";
+ case CRYPT_SLOT_ACTIVE:
+ return "ACTIVE";
+ case CRYPT_SLOT_ACTIVE_LAST:
+ return "ACTIVE_LAST";
+ case CRYPT_SLOT_INVALID:
+ default:
+ return "INVALID";
+ }
+}
+
+int LUKS_hdr_backup(const char *backup_file, struct crypt_device *ctx)
+{
+ struct device *device = crypt_metadata_device(ctx);
+ struct luks_phdr hdr;
+ int fd, devfd, r = 0;
+ size_t hdr_size;
+ size_t buffer_size;
+ ssize_t ret;
+ char *buffer = NULL;
+
+ r = LUKS_read_phdr(&hdr, 1, 0, ctx);
+ if (r)
+ return r;
+
+ hdr_size = LUKS_device_sectors(&hdr) << SECTOR_SHIFT;
+ buffer_size = size_round_up(hdr_size, crypt_getpagesize());
+
+ buffer = malloc(buffer_size);
+ if (!buffer || hdr_size < LUKS_ALIGN_KEYSLOTS || hdr_size > buffer_size) {
+ r = -ENOMEM;
+ goto out;
+ }
+ memset(buffer, 0, buffer_size);
+
+ log_dbg(ctx, "Storing backup of header (%zu bytes) and keyslot area (%zu bytes).",
+ sizeof(hdr), hdr_size - LUKS_ALIGN_KEYSLOTS);
+
+ log_dbg(ctx, "Output backup file size: %zu bytes.", buffer_size);
+
+ devfd = device_open(ctx, device, O_RDONLY);
+ if (devfd < 0) {
+ log_err(ctx, _("Device %s is not a valid LUKS device."), device_path(device));
+ r = -EINVAL;
+ goto out;
+ }
+
+ if (read_lseek_blockwise(devfd, device_block_size(ctx, device), device_alignment(device),
+ buffer, hdr_size, 0) < (ssize_t)hdr_size) {
+ r = -EIO;
+ goto out;
+ }
+
+ /* Wipe unused area, so backup cannot contain old signatures */
+ if (hdr.keyblock[0].keyMaterialOffset * SECTOR_SIZE == LUKS_ALIGN_KEYSLOTS)
+ memset(buffer + sizeof(hdr), 0, LUKS_ALIGN_KEYSLOTS - sizeof(hdr));
+
+ fd = open(backup_file, O_CREAT|O_EXCL|O_WRONLY, S_IRUSR);
+ if (fd == -1) {
+ if (errno == EEXIST)
+ log_err(ctx, _("Requested header backup file %s already exists."), backup_file);
+ else
+ log_err(ctx, _("Cannot create header backup file %s."), backup_file);
+ r = -EINVAL;
+ goto out;
+ }
+ ret = write_buffer(fd, buffer, buffer_size);
+ close(fd);
+ if (ret < (ssize_t)buffer_size) {
+ log_err(ctx, _("Cannot write header backup file %s."), backup_file);
+ r = -EIO;
+ goto out;
+ }
+
+ r = 0;
+out:
+ crypt_safe_memzero(&hdr, sizeof(hdr));
+ crypt_safe_memzero(buffer, buffer_size);
+ free(buffer);
+ return r;
+}
+
+int LUKS_hdr_restore(
+ const char *backup_file,
+ struct luks_phdr *hdr,
+ struct crypt_device *ctx)
+{
+ struct device *device = crypt_metadata_device(ctx);
+ int fd, r = 0, devfd = -1, diff_uuid = 0;
+ ssize_t ret, buffer_size = 0;
+ char *buffer = NULL, msg[200];
+ struct luks_phdr hdr_file;
+
+ r = LUKS_read_phdr_backup(backup_file, &hdr_file, 0, ctx);
+ if (r == -ENOENT)
+ return r;
+
+ if (!r)
+ buffer_size = LUKS_device_sectors(&hdr_file) << SECTOR_SHIFT;
+
+ if (r || buffer_size < LUKS_ALIGN_KEYSLOTS) {
+ log_err(ctx, _("Backup file does not contain valid LUKS header."));
+ r = -EINVAL;
+ goto out;
+ }
+
+ buffer = malloc(buffer_size);
+ if (!buffer) {
+ r = -ENOMEM;
+ goto out;
+ }
+
+ fd = open(backup_file, O_RDONLY);
+ if (fd == -1) {
+ log_err(ctx, _("Cannot open header backup file %s."), backup_file);
+ r = -EINVAL;
+ goto out;
+ }
+
+ ret = read_buffer(fd, buffer, buffer_size);
+ close(fd);
+ if (ret < buffer_size) {
+ log_err(ctx, _("Cannot read header backup file %s."), backup_file);
+ r = -EIO;
+ goto out;
+ }
+
+ r = LUKS_read_phdr(hdr, 0, 0, ctx);
+ if (r == 0) {
+ log_dbg(ctx, "Device %s already contains LUKS header, checking UUID and offset.", device_path(device));
+ if(hdr->payloadOffset != hdr_file.payloadOffset ||
+ hdr->keyBytes != hdr_file.keyBytes) {
+ log_err(ctx, _("Data offset or key size differs on device and backup, restore failed."));
+ r = -EINVAL;
+ goto out;
+ }
+ if (memcmp(hdr->uuid, hdr_file.uuid, UUID_STRING_L))
+ diff_uuid = 1;
+ }
+
+ if (snprintf(msg, sizeof(msg), _("Device %s %s%s"), device_path(device),
+ r ? _("does not contain LUKS header. Replacing header can destroy data on that device.") :
+ _("already contains LUKS header. Replacing header will destroy existing keyslots."),
+ diff_uuid ? _("\nWARNING: real device header has different UUID than backup!") : "") < 0) {
+ r = -ENOMEM;
+ goto out;
+ }
+
+ if (!crypt_confirm(ctx, msg)) {
+ r = -EINVAL;
+ goto out;
+ }
+
+ log_dbg(ctx, "Storing backup of header (%zu bytes) and keyslot area (%zu bytes) to device %s.",
+ sizeof(*hdr), buffer_size - LUKS_ALIGN_KEYSLOTS, device_path(device));
+
+ devfd = device_open(ctx, device, O_RDWR);
+ if (devfd < 0) {
+ if (errno == EACCES)
+ log_err(ctx, _("Cannot write to device %s, permission denied."),
+ device_path(device));
+ else
+ log_err(ctx, _("Cannot open device %s."), device_path(device));
+ r = -EINVAL;
+ goto out;
+ }
+
+ if (write_lseek_blockwise(devfd, device_block_size(ctx, device), device_alignment(device),
+ buffer, buffer_size, 0) < buffer_size) {
+ r = -EIO;
+ goto out;
+ }
+
+ /* Be sure to reload new data */
+ r = LUKS_read_phdr(hdr, 1, 0, ctx);
+out:
+ device_sync(ctx, device);
+ crypt_safe_memzero(buffer, buffer_size);
+ free(buffer);
+ return r;
+}
+
+/* This routine should do some just basic recovery for known problems. */
+static int _keyslot_repair(struct luks_phdr *phdr, struct crypt_device *ctx)
+{
+ struct luks_phdr temp_phdr;
+ const unsigned char *sector = (const unsigned char*)phdr;
+ struct volume_key *vk;
+ int i, bad, r, need_write = 0;
+
+ if (phdr->keyBytes != 16 && phdr->keyBytes != 32 && phdr->keyBytes != 64) {
+ log_err(ctx, _("Non standard key size, manual repair required."));
+ return -EINVAL;
+ }
+
+ /*
+ * cryptsetup 1.0 did not align keyslots to 4k, cannot repair this one
+ * Also we cannot trust possibly broken keyslots metadata here through LUKS_keyslots_offset().
+ * Expect first keyslot is aligned, if not, then manual repair is necessary.
+ */
+ if (phdr->keyblock[0].keyMaterialOffset < (LUKS_ALIGN_KEYSLOTS / SECTOR_SIZE)) {
+ log_err(ctx, _("Non standard keyslots alignment, manual repair required."));
+ return -EINVAL;
+ }
+
+ /*
+ * ECB mode does not use IV but legacy dmcrypt silently allows it.
+ * Today device cannot be activated anyway, so we need to fix it here.
+ */
+ if (!strncmp(phdr->cipherMode, "ecb-", 4)) {
+ log_err(ctx, _("Cipher mode repaired (%s -> %s)."), phdr->cipherMode, "ecb");
+ memset(phdr->cipherMode, 0, LUKS_CIPHERMODE_L);
+ strcpy(phdr->cipherMode, "ecb");
+ need_write = 1;
+ }
+
+ /*
+ * Old cryptsetup expects "sha1", gcrypt allows case insensitive names,
+ * so always convert hash to lower case in header
+ */
+ if (_to_lower(phdr->hashSpec, LUKS_HASHSPEC_L)) {
+ log_err(ctx, _("Cipher hash repaired to lowercase (%s)."), phdr->hashSpec);
+ if (crypt_hmac_size(phdr->hashSpec) < LUKS_DIGESTSIZE) {
+ log_err(ctx, _("Requested LUKS hash %s is not supported."), phdr->hashSpec);
+ return -EINVAL;
+ }
+ need_write = 1;
+ }
+
+ r = LUKS_check_cipher(ctx, phdr->keyBytes, phdr->cipherName, phdr->cipherMode);
+ if (r < 0)
+ return -EINVAL;
+
+ vk = crypt_alloc_volume_key(phdr->keyBytes, NULL);
+ if (!vk)
+ return -ENOMEM;
+
+ log_verbose(ctx, _("Repairing keyslots."));
+
+ log_dbg(ctx, "Generating second header with the same parameters for check.");
+ /* cipherName, cipherMode, hashSpec, uuid are already null terminated */
+ /* payloadOffset - cannot check */
+ r = LUKS_generate_phdr(&temp_phdr, vk, phdr->cipherName, phdr->cipherMode,
+ phdr->hashSpec, phdr->uuid,
+ phdr->payloadOffset * SECTOR_SIZE, 0, 0, ctx);
+ if (r < 0)
+ goto out;
+
+ for(i = 0; i < LUKS_NUMKEYS; ++i) {
+ if (phdr->keyblock[i].active == LUKS_KEY_ENABLED) {
+ log_dbg(ctx, "Skipping repair for active keyslot %i.", i);
+ continue;
+ }
+
+ bad = 0;
+ if (phdr->keyblock[i].keyMaterialOffset != temp_phdr.keyblock[i].keyMaterialOffset) {
+ log_err(ctx, _("Keyslot %i: offset repaired (%u -> %u)."), i,
+ (unsigned)phdr->keyblock[i].keyMaterialOffset,
+ (unsigned)temp_phdr.keyblock[i].keyMaterialOffset);
+ phdr->keyblock[i].keyMaterialOffset = temp_phdr.keyblock[i].keyMaterialOffset;
+ bad = 1;
+ }
+
+ if (phdr->keyblock[i].stripes != temp_phdr.keyblock[i].stripes) {
+ log_err(ctx, _("Keyslot %i: stripes repaired (%u -> %u)."), i,
+ (unsigned)phdr->keyblock[i].stripes,
+ (unsigned)temp_phdr.keyblock[i].stripes);
+ phdr->keyblock[i].stripes = temp_phdr.keyblock[i].stripes;
+ bad = 1;
+ }
+
+ /* Known case - MSDOS partition table signature */
+ if (i == 6 && sector[0x1fe] == 0x55 && sector[0x1ff] == 0xaa) {
+ log_err(ctx, _("Keyslot %i: bogus partition signature."), i);
+ bad = 1;
+ }
+
+ if(bad) {
+ log_err(ctx, _("Keyslot %i: salt wiped."), i);
+ phdr->keyblock[i].active = LUKS_KEY_DISABLED;
+ memset(&phdr->keyblock[i].passwordSalt, 0x00, LUKS_SALTSIZE);
+ phdr->keyblock[i].passwordIterations = 0;
+ }
+
+ if (bad)
+ need_write = 1;
+ }
+
+ /*
+ * check repair result before writing because repair can't fix out of order
+ * keyslot offsets and would corrupt header again
+ */
+ if (LUKS_check_keyslots(ctx, phdr))
+ r = -EINVAL;
+ else if (need_write) {
+ log_verbose(ctx, _("Writing LUKS header to disk."));
+ r = LUKS_write_phdr(phdr, ctx);
+ }
+out:
+ if (r)
+ log_err(ctx, _("Repair failed."));
+ crypt_free_volume_key(vk);
+ crypt_safe_memzero(&temp_phdr, sizeof(temp_phdr));
+ return r;
+}
+
+static int _check_and_convert_hdr(const char *device,
+ struct luks_phdr *hdr,
+ int require_luks_device,
+ int repair,
+ struct crypt_device *ctx)
+{
+ int r = 0;
+ unsigned int i;
+ char luksMagic[] = LUKS_MAGIC;
+
+ hdr->version = be16_to_cpu(hdr->version);
+ if (memcmp(hdr->magic, luksMagic, LUKS_MAGIC_L)) { /* Check magic */
+ log_dbg(ctx, "LUKS header not detected.");
+ if (require_luks_device)
+ log_err(ctx, _("Device %s is not a valid LUKS device."), device);
+ return -EINVAL;
+ } else if (hdr->version != 1) {
+ log_err(ctx, _("Unsupported LUKS version %d."), hdr->version);
+ return -EINVAL;
+ }
+
+ hdr->hashSpec[LUKS_HASHSPEC_L - 1] = '\0';
+ if (crypt_hmac_size(hdr->hashSpec) < LUKS_DIGESTSIZE) {
+ log_err(ctx, _("Requested LUKS hash %s is not supported."), hdr->hashSpec);
+ r = -EINVAL;
+ }
+
+ /* Header detected */
+ hdr->payloadOffset = be32_to_cpu(hdr->payloadOffset);
+ hdr->keyBytes = be32_to_cpu(hdr->keyBytes);
+ hdr->mkDigestIterations = be32_to_cpu(hdr->mkDigestIterations);
+
+ for (i = 0; i < LUKS_NUMKEYS; ++i) {
+ hdr->keyblock[i].active = be32_to_cpu(hdr->keyblock[i].active);
+ hdr->keyblock[i].passwordIterations = be32_to_cpu(hdr->keyblock[i].passwordIterations);
+ hdr->keyblock[i].keyMaterialOffset = be32_to_cpu(hdr->keyblock[i].keyMaterialOffset);
+ hdr->keyblock[i].stripes = be32_to_cpu(hdr->keyblock[i].stripes);
+ }
+
+ if (LUKS_check_keyslots(ctx, hdr))
+ r = -EINVAL;
+
+ /* Avoid unterminated strings */
+ hdr->cipherName[LUKS_CIPHERNAME_L - 1] = '\0';
+ hdr->cipherMode[LUKS_CIPHERMODE_L - 1] = '\0';
+ hdr->uuid[UUID_STRING_L - 1] = '\0';
+
+ if (repair) {
+ if (!strncmp(hdr->cipherMode, "ecb-", 4)) {
+ log_err(ctx, _("LUKS cipher mode %s is invalid."), hdr->cipherMode);
+ r = -EINVAL;
+ }
+
+ if (_is_not_lower(hdr->hashSpec, LUKS_HASHSPEC_L)) {
+ log_err(ctx, _("LUKS hash %s is invalid."), hdr->hashSpec);
+ r = -EINVAL;
+ }
+
+ if (r == -EINVAL)
+ r = _keyslot_repair(hdr, ctx);
+ else
+ log_verbose(ctx, _("No known problems detected for LUKS header."));
+ }
+
+ return r;
+}
+
+int LUKS_read_phdr_backup(const char *backup_file,
+ struct luks_phdr *hdr,
+ int require_luks_device,
+ struct crypt_device *ctx)
+{
+ ssize_t hdr_size = sizeof(struct luks_phdr);
+ int devfd = 0, r = 0;
+
+ log_dbg(ctx, "Reading LUKS header of size %d from backup file %s",
+ (int)hdr_size, backup_file);
+
+ devfd = open(backup_file, O_RDONLY);
+ if (devfd == -1) {
+ log_err(ctx, _("Cannot open header backup file %s."), backup_file);
+ return -ENOENT;
+ }
+
+ if (read_buffer(devfd, hdr, hdr_size) < hdr_size)
+ r = -EIO;
+ else
+ r = _check_and_convert_hdr(backup_file, hdr,
+ require_luks_device, 0, ctx);
+
+ close(devfd);
+ return r;
+}
+
+int LUKS_read_phdr(struct luks_phdr *hdr,
+ int require_luks_device,
+ int repair,
+ struct crypt_device *ctx)
+{
+ int devfd, r = 0;
+ struct device *device = crypt_metadata_device(ctx);
+ ssize_t hdr_size = sizeof(struct luks_phdr);
+
+ /* LUKS header starts at offset 0, first keyslot on LUKS_ALIGN_KEYSLOTS */
+ assert(sizeof(struct luks_phdr) <= LUKS_ALIGN_KEYSLOTS);
+
+ /* Stripes count cannot be changed without additional code fixes yet */
+ assert(LUKS_STRIPES == 4000);
+
+ if (repair && !require_luks_device)
+ return -EINVAL;
+
+ log_dbg(ctx, "Reading LUKS header of size %zu from device %s",
+ hdr_size, device_path(device));
+
+ devfd = device_open(ctx, device, O_RDONLY);
+ if (devfd < 0) {
+ log_err(ctx, _("Cannot open device %s."), device_path(device));
+ return -EINVAL;
+ }
+
+ if (read_lseek_blockwise(devfd, device_block_size(ctx, device), device_alignment(device),
+ hdr, hdr_size, 0) < hdr_size)
+ r = -EIO;
+ else
+ r = _check_and_convert_hdr(device_path(device), hdr, require_luks_device,
+ repair, ctx);
+
+ if (!r)
+ r = LUKS_check_device_size(ctx, hdr, 0);
+
+ /*
+ * Cryptsetup 1.0.0 did not align keyslots to 4k (very rare version).
+ * Disable direct-io to avoid possible IO errors if underlying device
+ * has bigger sector size.
+ */
+ if (!r && hdr->keyblock[0].keyMaterialOffset * SECTOR_SIZE < LUKS_ALIGN_KEYSLOTS) {
+ log_dbg(ctx, "Old unaligned LUKS keyslot detected, disabling direct-io.");
+ device_disable_direct_io(device);
+ }
+
+ return r;
+}
+
+int LUKS_write_phdr(struct luks_phdr *hdr,
+ struct crypt_device *ctx)
+{
+ struct device *device = crypt_metadata_device(ctx);
+ ssize_t hdr_size = sizeof(struct luks_phdr);
+ int devfd = 0;
+ unsigned int i;
+ struct luks_phdr convHdr;
+ int r;
+
+ log_dbg(ctx, "Updating LUKS header of size %zu on device %s",
+ sizeof(struct luks_phdr), device_path(device));
+
+ r = LUKS_check_device_size(ctx, hdr, 1);
+ if (r)
+ return r;
+
+ devfd = device_open(ctx, device, O_RDWR);
+ if (devfd < 0) {
+ if (errno == EACCES)
+ log_err(ctx, _("Cannot write to device %s, permission denied."),
+ device_path(device));
+ else
+ log_err(ctx, _("Cannot open device %s."), device_path(device));
+ return -EINVAL;
+ }
+
+ memcpy(&convHdr, hdr, hdr_size);
+ memset(&convHdr._padding, 0, sizeof(convHdr._padding));
+
+ /* Convert every uint16/32_t item to network byte order */
+ convHdr.version = cpu_to_be16(hdr->version);
+ convHdr.payloadOffset = cpu_to_be32(hdr->payloadOffset);
+ convHdr.keyBytes = cpu_to_be32(hdr->keyBytes);
+ convHdr.mkDigestIterations = cpu_to_be32(hdr->mkDigestIterations);
+ for(i = 0; i < LUKS_NUMKEYS; ++i) {
+ convHdr.keyblock[i].active = cpu_to_be32(hdr->keyblock[i].active);
+ convHdr.keyblock[i].passwordIterations = cpu_to_be32(hdr->keyblock[i].passwordIterations);
+ convHdr.keyblock[i].keyMaterialOffset = cpu_to_be32(hdr->keyblock[i].keyMaterialOffset);
+ convHdr.keyblock[i].stripes = cpu_to_be32(hdr->keyblock[i].stripes);
+ }
+
+ r = write_lseek_blockwise(devfd, device_block_size(ctx, device), device_alignment(device),
+ &convHdr, hdr_size, 0) < hdr_size ? -EIO : 0;
+ if (r)
+ log_err(ctx, _("Error during update of LUKS header on device %s."), device_path(device));
+
+ device_sync(ctx, device);
+
+ /* Re-read header from disk to be sure that in-memory and on-disk data are the same. */
+ if (!r) {
+ r = LUKS_read_phdr(hdr, 1, 0, ctx);
+ if (r)
+ log_err(ctx, _("Error re-reading LUKS header after update on device %s."),
+ device_path(device));
+ }
+
+ return r;
+}
+
+/* Check that kernel supports requested cipher by decryption of one sector */
+int LUKS_check_cipher(struct crypt_device *ctx, size_t keylength, const char *cipher, const char *cipher_mode)
+{
+ int r;
+ struct volume_key *empty_key;
+ char buf[SECTOR_SIZE];
+
+ log_dbg(ctx, "Checking if cipher %s-%s is usable.", cipher, cipher_mode);
+
+ empty_key = crypt_alloc_volume_key(keylength, NULL);
+ if (!empty_key)
+ return -ENOMEM;
+
+ /* No need to get KEY quality random but it must avoid known weak keys. */
+ r = crypt_random_get(ctx, empty_key->key, empty_key->keylength, CRYPT_RND_NORMAL);
+ if (!r)
+ r = LUKS_decrypt_from_storage(buf, sizeof(buf), cipher, cipher_mode, empty_key, 0, ctx);
+
+ crypt_free_volume_key(empty_key);
+ crypt_safe_memzero(buf, sizeof(buf));
+ return r;
+}
+
+int LUKS_generate_phdr(struct luks_phdr *header,
+ const struct volume_key *vk,
+ const char *cipherName,
+ const char *cipherMode,
+ const char *hashSpec,
+ const char *uuid,
+ uint64_t data_offset, /* in bytes */
+ uint64_t align_offset, /* in bytes */
+ uint64_t required_alignment, /* in bytes */
+ struct crypt_device *ctx)
+{
+ int i, r;
+ size_t keyslot_sectors, header_sectors;
+ uuid_t partitionUuid;
+ struct crypt_pbkdf_type *pbkdf;
+ double PBKDF2_temp;
+ char luksMagic[] = LUKS_MAGIC;
+
+ if (data_offset % SECTOR_SIZE || align_offset % SECTOR_SIZE ||
+ required_alignment % SECTOR_SIZE)
+ return -EINVAL;
+
+ memset(header, 0, sizeof(struct luks_phdr));
+
+ keyslot_sectors = AF_split_sectors(vk->keylength, LUKS_STRIPES);
+ header_sectors = LUKS_ALIGN_KEYSLOTS / SECTOR_SIZE;
+
+ for (i = 0; i < LUKS_NUMKEYS; i++) {
+ header->keyblock[i].active = LUKS_KEY_DISABLED;
+ header->keyblock[i].keyMaterialOffset = header_sectors;
+ header->keyblock[i].stripes = LUKS_STRIPES;
+ header_sectors = size_round_up(header_sectors + keyslot_sectors,
+ LUKS_ALIGN_KEYSLOTS / SECTOR_SIZE);
+ }
+ /* In sector is now size of all keyslot material space */
+
+ /* Data offset has priority */
+ if (data_offset)
+ header->payloadOffset = data_offset / SECTOR_SIZE;
+ else if (required_alignment) {
+ header->payloadOffset = size_round_up(header_sectors, (required_alignment / SECTOR_SIZE));
+ header->payloadOffset += (align_offset / SECTOR_SIZE);
+ } else
+ header->payloadOffset = 0;
+
+ if (header->payloadOffset && header->payloadOffset < header_sectors) {
+ log_err(ctx, _("Data offset for LUKS header must be "
+ "either 0 or higher than header size."));
+ return -EINVAL;
+ }
+
+ if (crypt_hmac_size(hashSpec) < LUKS_DIGESTSIZE) {
+ log_err(ctx, _("Requested LUKS hash %s is not supported."), hashSpec);
+ return -EINVAL;
+ }
+
+ if (uuid && uuid_parse(uuid, partitionUuid) == -1) {
+ log_err(ctx, _("Wrong LUKS UUID format provided."));
+ return -EINVAL;
+ }
+ if (!uuid)
+ uuid_generate(partitionUuid);
+
+ /* Set Magic */
+ memcpy(header->magic,luksMagic,LUKS_MAGIC_L);
+ header->version=1;
+ strncpy(header->cipherName,cipherName,LUKS_CIPHERNAME_L-1);
+ strncpy(header->cipherMode,cipherMode,LUKS_CIPHERMODE_L-1);
+ strncpy(header->hashSpec,hashSpec,LUKS_HASHSPEC_L-1);
+ _to_lower(header->hashSpec, LUKS_HASHSPEC_L);
+
+ header->keyBytes=vk->keylength;
+
+ log_dbg(ctx, "Generating LUKS header version %d using hash %s, %s, %s, MK %d bytes",
+ header->version, header->hashSpec ,header->cipherName, header->cipherMode,
+ header->keyBytes);
+
+ r = crypt_random_get(ctx, header->mkDigestSalt, LUKS_SALTSIZE, CRYPT_RND_SALT);
+ if(r < 0) {
+ log_err(ctx, _("Cannot create LUKS header: reading random salt failed."));
+ return r;
+ }
+
+ /* Compute volume key digest */
+ pbkdf = crypt_get_pbkdf(ctx);
+ r = crypt_benchmark_pbkdf_internal(ctx, pbkdf, vk->keylength);
+ if (r < 0)
+ return r;
+ assert(pbkdf->iterations);
+
+ if (pbkdf->flags & CRYPT_PBKDF_NO_BENCHMARK && pbkdf->time_ms == 0)
+ PBKDF2_temp = LUKS_MKD_ITERATIONS_MIN;
+ else /* iterations per ms * LUKS_MKD_ITERATIONS_MS */
+ PBKDF2_temp = (double)pbkdf->iterations * LUKS_MKD_ITERATIONS_MS / pbkdf->time_ms;
+
+ if (PBKDF2_temp > (double)UINT32_MAX)
+ return -EINVAL;
+ header->mkDigestIterations = AT_LEAST((uint32_t)PBKDF2_temp, LUKS_MKD_ITERATIONS_MIN);
+ assert(header->mkDigestIterations);
+
+ r = crypt_pbkdf(CRYPT_KDF_PBKDF2, header->hashSpec, vk->key,vk->keylength,
+ header->mkDigestSalt, LUKS_SALTSIZE,
+ header->mkDigest,LUKS_DIGESTSIZE,
+ header->mkDigestIterations, 0, 0);
+ if (r < 0) {
+ log_err(ctx, _("Cannot create LUKS header: header digest failed (using hash %s)."),
+ header->hashSpec);
+ return r;
+ }
+
+ uuid_unparse(partitionUuid, header->uuid);
+
+ log_dbg(ctx, "Data offset %d, UUID %s, digest iterations %" PRIu32,
+ header->payloadOffset, header->uuid, header->mkDigestIterations);
+
+ return 0;
+}
+
+int LUKS_hdr_uuid_set(
+ struct luks_phdr *hdr,
+ const char *uuid,
+ struct crypt_device *ctx)
+{
+ uuid_t partitionUuid;
+
+ if (uuid && uuid_parse(uuid, partitionUuid) == -1) {
+ log_err(ctx, _("Wrong LUKS UUID format provided."));
+ return -EINVAL;
+ }
+ if (!uuid)
+ uuid_generate(partitionUuid);
+
+ uuid_unparse(partitionUuid, hdr->uuid);
+
+ return LUKS_write_phdr(hdr, ctx);
+}
+
+int LUKS_set_key(unsigned int keyIndex,
+ const char *password, size_t passwordLen,
+ struct luks_phdr *hdr, struct volume_key *vk,
+ struct crypt_device *ctx)
+{
+ struct volume_key *derived_key;
+ char *AfKey = NULL;
+ size_t AFEKSize;
+ struct crypt_pbkdf_type *pbkdf;
+ int r;
+
+ if(hdr->keyblock[keyIndex].active != LUKS_KEY_DISABLED) {
+ log_err(ctx, _("Key slot %d active, purge first."), keyIndex);
+ return -EINVAL;
+ }
+
+ /* LUKS keyslot has always at least 4000 stripes according to specification */
+ if(hdr->keyblock[keyIndex].stripes < 4000) {
+ log_err(ctx, _("Key slot %d material includes too few stripes. Header manipulation?"),
+ keyIndex);
+ return -EINVAL;
+ }
+
+ log_dbg(ctx, "Calculating data for key slot %d", keyIndex);
+ pbkdf = crypt_get_pbkdf(ctx);
+ r = crypt_benchmark_pbkdf_internal(ctx, pbkdf, vk->keylength);
+ if (r < 0)
+ return r;
+ assert(pbkdf->iterations);
+
+ /*
+ * Final iteration count is at least LUKS_SLOT_ITERATIONS_MIN
+ */
+ hdr->keyblock[keyIndex].passwordIterations =
+ AT_LEAST(pbkdf->iterations, LUKS_SLOT_ITERATIONS_MIN);
+ log_dbg(ctx, "Key slot %d use %" PRIu32 " password iterations.", keyIndex,
+ hdr->keyblock[keyIndex].passwordIterations);
+
+ derived_key = crypt_alloc_volume_key(hdr->keyBytes, NULL);
+ if (!derived_key)
+ return -ENOMEM;
+
+ r = crypt_random_get(ctx, hdr->keyblock[keyIndex].passwordSalt,
+ LUKS_SALTSIZE, CRYPT_RND_SALT);
+ if (r < 0)
+ goto out;
+
+ r = crypt_pbkdf(CRYPT_KDF_PBKDF2, hdr->hashSpec, password, passwordLen,
+ hdr->keyblock[keyIndex].passwordSalt, LUKS_SALTSIZE,
+ derived_key->key, hdr->keyBytes,
+ hdr->keyblock[keyIndex].passwordIterations, 0, 0);
+ if (r < 0) {
+ if ((crypt_backend_flags() & CRYPT_BACKEND_PBKDF2_INT) &&
+ hdr->keyblock[keyIndex].passwordIterations > INT_MAX)
+ log_err(ctx, _("PBKDF2 iteration value overflow."));
+ goto out;
+ }
+
+ /*
+ * AF splitting, the volume key stored in vk->key is split to AfKey
+ */
+ assert(vk->keylength == hdr->keyBytes);
+ AFEKSize = AF_split_sectors(vk->keylength, hdr->keyblock[keyIndex].stripes) * SECTOR_SIZE;
+ AfKey = crypt_safe_alloc(AFEKSize);
+ if (!AfKey) {
+ r = -ENOMEM;
+ goto out;
+ }
+
+ log_dbg(ctx, "Using hash %s for AF in key slot %d, %d stripes",
+ hdr->hashSpec, keyIndex, hdr->keyblock[keyIndex].stripes);
+ r = AF_split(ctx, vk->key, AfKey, vk->keylength, hdr->keyblock[keyIndex].stripes, hdr->hashSpec);
+ if (r < 0)
+ goto out;
+
+ log_dbg(ctx, "Updating key slot %d [0x%04x] area.", keyIndex,
+ hdr->keyblock[keyIndex].keyMaterialOffset << 9);
+ /* Encryption via dm */
+ r = LUKS_encrypt_to_storage(AfKey,
+ AFEKSize,
+ hdr->cipherName, hdr->cipherMode,
+ derived_key,
+ hdr->keyblock[keyIndex].keyMaterialOffset,
+ ctx);
+ if (r < 0)
+ goto out;
+
+ /* Mark the key as active in phdr */
+ r = LUKS_keyslot_set(hdr, (int)keyIndex, 1, ctx);
+ if (r < 0)
+ goto out;
+
+ r = LUKS_write_phdr(hdr, ctx);
+ if (r < 0)
+ goto out;
+
+ r = 0;
+out:
+ crypt_safe_free(AfKey);
+ crypt_free_volume_key(derived_key);
+ return r;
+}
+
+/* Check whether a volume key is invalid. */
+int LUKS_verify_volume_key(const struct luks_phdr *hdr,
+ const struct volume_key *vk)
+{
+ char checkHashBuf[LUKS_DIGESTSIZE];
+
+ if (crypt_pbkdf(CRYPT_KDF_PBKDF2, hdr->hashSpec, vk->key, vk->keylength,
+ hdr->mkDigestSalt, LUKS_SALTSIZE,
+ checkHashBuf, LUKS_DIGESTSIZE,
+ hdr->mkDigestIterations, 0, 0) < 0)
+ return -EINVAL;
+
+ if (crypt_backend_memeq(checkHashBuf, hdr->mkDigest, LUKS_DIGESTSIZE))
+ return -EPERM;
+
+ return 0;
+}
+
+/* Try to open a particular key slot */
+static int LUKS_open_key(unsigned int keyIndex,
+ const char *password,
+ size_t passwordLen,
+ struct luks_phdr *hdr,
+ struct volume_key **vk,
+ struct crypt_device *ctx)
+{
+ crypt_keyslot_info ki = LUKS_keyslot_info(hdr, keyIndex);
+ struct volume_key *derived_key;
+ char *AfKey = NULL;
+ size_t AFEKSize;
+ int r;
+
+ log_dbg(ctx, "Trying to open key slot %d [%s].", keyIndex,
+ dbg_slot_state(ki));
+
+ if (ki < CRYPT_SLOT_ACTIVE)
+ return -ENOENT;
+
+ derived_key = crypt_alloc_volume_key(hdr->keyBytes, NULL);
+ if (!derived_key)
+ return -ENOMEM;
+
+ *vk = crypt_alloc_volume_key(hdr->keyBytes, NULL);
+ if (!*vk) {
+ r = -ENOMEM;
+ goto out;
+ }
+
+ AFEKSize = AF_split_sectors(hdr->keyBytes, hdr->keyblock[keyIndex].stripes) * SECTOR_SIZE;
+ AfKey = crypt_safe_alloc(AFEKSize);
+ if (!AfKey) {
+ r = -ENOMEM;
+ goto out;
+ }
+
+ r = crypt_pbkdf(CRYPT_KDF_PBKDF2, hdr->hashSpec, password, passwordLen,
+ hdr->keyblock[keyIndex].passwordSalt, LUKS_SALTSIZE,
+ derived_key->key, hdr->keyBytes,
+ hdr->keyblock[keyIndex].passwordIterations, 0, 0);
+ if (r < 0) {
+ log_err(ctx, _("Cannot open keyslot (using hash %s)."), hdr->hashSpec);
+ goto out;
+ }
+
+ log_dbg(ctx, "Reading key slot %d area.", keyIndex);
+ r = LUKS_decrypt_from_storage(AfKey,
+ AFEKSize,
+ hdr->cipherName, hdr->cipherMode,
+ derived_key,
+ hdr->keyblock[keyIndex].keyMaterialOffset,
+ ctx);
+ if (r < 0)
+ goto out;
+
+ r = AF_merge(AfKey, (*vk)->key, (*vk)->keylength, hdr->keyblock[keyIndex].stripes, hdr->hashSpec);
+ if (r < 0)
+ goto out;
+
+ r = LUKS_verify_volume_key(hdr, *vk);
+
+ /* Allow only empty passphrase with null cipher */
+ if (!r && crypt_is_cipher_null(hdr->cipherName) && passwordLen)
+ r = -EPERM;
+out:
+ if (r < 0) {
+ crypt_free_volume_key(*vk);
+ *vk = NULL;
+ }
+ crypt_safe_free(AfKey);
+ crypt_free_volume_key(derived_key);
+ return r;
+}
+
+int LUKS_open_key_with_hdr(int keyIndex,
+ const char *password,
+ size_t passwordLen,
+ struct luks_phdr *hdr,
+ struct volume_key **vk,
+ struct crypt_device *ctx)
+{
+ unsigned int i, tried = 0;
+ int r;
+
+ if (keyIndex >= 0) {
+ r = LUKS_open_key(keyIndex, password, passwordLen, hdr, vk, ctx);
+ return (r < 0) ? r : keyIndex;
+ }
+
+ for (i = 0; i < LUKS_NUMKEYS; i++) {
+ r = LUKS_open_key(i, password, passwordLen, hdr, vk, ctx);
+ if (r == 0)
+ return i;
+
+ /* Do not retry for errors that are no -EPERM or -ENOENT,
+ former meaning password wrong, latter key slot inactive */
+ if ((r != -EPERM) && (r != -ENOENT))
+ return r;
+ if (r == -EPERM)
+ tried++;
+ }
+ /* Warning, early returns above */
+ return tried ? -EPERM : -ENOENT;
+}
+
+int LUKS_del_key(unsigned int keyIndex,
+ struct luks_phdr *hdr,
+ struct crypt_device *ctx)
+{
+ struct device *device = crypt_metadata_device(ctx);
+ unsigned int startOffset, endOffset;
+ int r;
+
+ r = LUKS_read_phdr(hdr, 1, 0, ctx);
+ if (r)
+ return r;
+
+ r = LUKS_keyslot_set(hdr, keyIndex, 0, ctx);
+ if (r) {
+ log_err(ctx, _("Key slot %d is invalid, please select keyslot between 0 and %d."),
+ keyIndex, LUKS_NUMKEYS - 1);
+ return r;
+ }
+
+ /* secure deletion of key material */
+ startOffset = hdr->keyblock[keyIndex].keyMaterialOffset;
+ endOffset = startOffset + AF_split_sectors(hdr->keyBytes, hdr->keyblock[keyIndex].stripes);
+
+ r = crypt_wipe_device(ctx, device, CRYPT_WIPE_SPECIAL, startOffset * SECTOR_SIZE,
+ (endOffset - startOffset) * SECTOR_SIZE,
+ (endOffset - startOffset) * SECTOR_SIZE, NULL, NULL);
+ if (r) {
+ if (r == -EACCES) {
+ log_err(ctx, _("Cannot write to device %s, permission denied."),
+ device_path(device));
+ r = -EINVAL;
+ } else
+ log_err(ctx, _("Cannot wipe device %s."),
+ device_path(device));
+ return r;
+ }
+
+ /* Wipe keyslot info */
+ memset(&hdr->keyblock[keyIndex].passwordSalt, 0, LUKS_SALTSIZE);
+ hdr->keyblock[keyIndex].passwordIterations = 0;
+
+ r = LUKS_write_phdr(hdr, ctx);
+
+ return r;
+}
+
+crypt_keyslot_info LUKS_keyslot_info(struct luks_phdr *hdr, int keyslot)
+{
+ int i;
+
+ if(keyslot >= LUKS_NUMKEYS || keyslot < 0)
+ return CRYPT_SLOT_INVALID;
+
+ if (hdr->keyblock[keyslot].active == LUKS_KEY_DISABLED)
+ return CRYPT_SLOT_INACTIVE;
+
+ if (hdr->keyblock[keyslot].active != LUKS_KEY_ENABLED)
+ return CRYPT_SLOT_INVALID;
+
+ for(i = 0; i < LUKS_NUMKEYS; i++)
+ if(i != keyslot && hdr->keyblock[i].active == LUKS_KEY_ENABLED)
+ return CRYPT_SLOT_ACTIVE;
+
+ return CRYPT_SLOT_ACTIVE_LAST;
+}
+
+int LUKS_keyslot_find_empty(struct luks_phdr *hdr)
+{
+ int i;
+
+ for (i = 0; i < LUKS_NUMKEYS; i++)
+ if(hdr->keyblock[i].active == LUKS_KEY_DISABLED)
+ break;
+
+ if (i == LUKS_NUMKEYS)
+ return -EINVAL;
+
+ return i;
+}
+
+int LUKS_keyslot_active_count(struct luks_phdr *hdr)
+{
+ int i, num = 0;
+
+ for (i = 0; i < LUKS_NUMKEYS; i++)
+ if(hdr->keyblock[i].active == LUKS_KEY_ENABLED)
+ num++;
+
+ return num;
+}
+
+int LUKS_keyslot_set(struct luks_phdr *hdr, int keyslot, int enable, struct crypt_device *ctx)
+{
+ crypt_keyslot_info ki = LUKS_keyslot_info(hdr, keyslot);
+
+ if (ki == CRYPT_SLOT_INVALID)
+ return -EINVAL;
+
+ hdr->keyblock[keyslot].active = enable ? LUKS_KEY_ENABLED : LUKS_KEY_DISABLED;
+ log_dbg(ctx, "Key slot %d was %s in LUKS header.", keyslot, enable ? "enabled" : "disabled");
+ return 0;
+}
+
+int LUKS1_activate(struct crypt_device *cd,
+ const char *name,
+ struct volume_key *vk,
+ uint32_t flags)
+{
+ int r;
+ struct crypt_dm_active_device dmd = {
+ .flags = flags,
+ .uuid = crypt_get_uuid(cd),
+ };
+
+ r = dm_crypt_target_set(&dmd.segment, 0, dmd.size, crypt_data_device(cd),
+ vk, crypt_get_cipher_spec(cd), crypt_get_iv_offset(cd),
+ crypt_get_data_offset(cd), crypt_get_integrity(cd),
+ crypt_get_integrity_tag_size(cd), crypt_get_sector_size(cd));
+ if (!r)
+ r = create_or_reload_device(cd, name, CRYPT_LUKS1, &dmd);
+
+ dm_targets_free(cd, &dmd);
+
+ return r;
+}
+
+int LUKS_wipe_header_areas(struct luks_phdr *hdr,
+ struct crypt_device *ctx)
+{
+ int i, r;
+ uint64_t offset, length;
+ size_t wipe_block;
+
+ r = LUKS_check_device_size(ctx, hdr, 1);
+ if (r)
+ return r;
+
+ /* Wipe complete header, keyslots and padding areas with zeroes. */
+ offset = 0;
+ length = (uint64_t)hdr->payloadOffset * SECTOR_SIZE;
+ wipe_block = 1024 * 1024;
+
+ /* On detached header or bogus header, wipe at least the first 4k */
+ if (length == 0 || length > (LUKS_MAX_KEYSLOT_SIZE * LUKS_NUMKEYS)) {
+ length = 4096;
+ wipe_block = 4096;
+ }
+
+ log_dbg(ctx, "Wiping LUKS areas (0x%06" PRIx64 " - 0x%06" PRIx64") with zeroes.",
+ offset, length + offset);
+
+ r = crypt_wipe_device(ctx, crypt_metadata_device(ctx), CRYPT_WIPE_ZERO,
+ offset, length, wipe_block, NULL, NULL);
+ if (r < 0)
+ return r;
+
+ /* Wipe keyslots areas */
+ wipe_block = 1024 * 1024;
+ for (i = 0; i < LUKS_NUMKEYS; i++) {
+ r = LUKS_keyslot_area(hdr, i, &offset, &length);
+ if (r < 0)
+ return r;
+
+ /* Ignore too big LUKS1 keyslots here */
+ if (length > LUKS_MAX_KEYSLOT_SIZE ||
+ offset > (LUKS_MAX_KEYSLOT_SIZE - length))
+ continue;
+
+ if (length == 0 || offset < 4096)
+ return -EINVAL;
+
+ log_dbg(ctx, "Wiping keyslot %i area (0x%06" PRIx64 " - 0x%06" PRIx64") with random data.",
+ i, offset, length + offset);
+
+ r = crypt_wipe_device(ctx, crypt_metadata_device(ctx), CRYPT_WIPE_RANDOM,
+ offset, length, wipe_block, NULL, NULL);
+ if (r < 0)
+ return r;
+ }
+
+ return r;
+}
+
+int LUKS_keyslot_pbkdf(struct luks_phdr *hdr, int keyslot, struct crypt_pbkdf_type *pbkdf)
+{
+ if (LUKS_keyslot_info(hdr, keyslot) < CRYPT_SLOT_ACTIVE)
+ return -EINVAL;
+
+ pbkdf->type = CRYPT_KDF_PBKDF2;
+ pbkdf->hash = hdr->hashSpec;
+ pbkdf->iterations = hdr->keyblock[keyslot].passwordIterations;
+ pbkdf->max_memory_kb = 0;
+ pbkdf->parallel_threads = 0;
+ pbkdf->time_ms = 0;
+ pbkdf->flags = 0;
+ return 0;
+}
diff --git a/lib/luks1/luks.h b/lib/luks1/luks.h
new file mode 100644
index 0000000..9c3f386
--- /dev/null
+++ b/lib/luks1/luks.h
@@ -0,0 +1,194 @@
+/*
+ * LUKS - Linux Unified Key Setup
+ *
+ * Copyright (C) 2004-2006 Clemens Fruhwirth <clemens@endorphin.org>
+ * Copyright (C) 2009-2023 Red Hat, Inc. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#ifndef INCLUDED_CRYPTSETUP_LUKS_LUKS_H
+#define INCLUDED_CRYPTSETUP_LUKS_LUKS_H
+
+/*
+ * LUKS partition header
+ */
+
+#include "libcryptsetup.h"
+
+#define LUKS_CIPHERNAME_L 32
+#define LUKS_CIPHERMODE_L 32
+#define LUKS_HASHSPEC_L 32
+#define LUKS_DIGESTSIZE 20 // since SHA1
+#define LUKS_HMACSIZE 32
+#define LUKS_SALTSIZE 32
+#define LUKS_NUMKEYS 8
+
+// Minimal number of iterations
+#define LUKS_MKD_ITERATIONS_MIN 1000
+#define LUKS_SLOT_ITERATIONS_MIN 1000
+
+// Iteration time for digest in ms
+#define LUKS_MKD_ITERATIONS_MS 125
+
+#define LUKS_KEY_DISABLED_OLD 0
+#define LUKS_KEY_ENABLED_OLD 0xCAFE
+
+#define LUKS_KEY_DISABLED 0x0000DEAD
+#define LUKS_KEY_ENABLED 0x00AC71F3
+
+#define LUKS_STRIPES 4000
+
+// partition header starts with magic
+#define LUKS_MAGIC {'L','U','K','S', 0xba, 0xbe};
+#define LUKS_MAGIC_L 6
+
+/* Actually we need only 37, but we don't want struct autoaligning to kick in */
+#define UUID_STRING_L 40
+
+/* Offset to keyslot area [in bytes] */
+#define LUKS_ALIGN_KEYSLOTS 4096
+
+/* Maximal LUKS header size, for wipe [in bytes] */
+#define LUKS_MAX_KEYSLOT_SIZE 0x1000000 /* 16 MB, up to 32768 bits key */
+
+/* Any integer values are stored in network byte order on disk and must be
+converted */
+
+struct volume_key;
+struct device_backend;
+
+struct luks_phdr {
+ char magic[LUKS_MAGIC_L];
+ uint16_t version;
+ char cipherName[LUKS_CIPHERNAME_L];
+ char cipherMode[LUKS_CIPHERMODE_L];
+ char hashSpec[LUKS_HASHSPEC_L];
+ uint32_t payloadOffset;
+ uint32_t keyBytes;
+ char mkDigest[LUKS_DIGESTSIZE];
+ char mkDigestSalt[LUKS_SALTSIZE];
+ uint32_t mkDigestIterations;
+ char uuid[UUID_STRING_L];
+
+ struct {
+ uint32_t active;
+
+ /* parameters used for password processing */
+ uint32_t passwordIterations;
+ char passwordSalt[LUKS_SALTSIZE];
+
+ /* parameters used for AF store/load */
+ uint32_t keyMaterialOffset;
+ uint32_t stripes;
+ } keyblock[LUKS_NUMKEYS];
+
+ /* Align it to 512 sector size */
+ char _padding[432];
+};
+
+int LUKS_verify_volume_key(const struct luks_phdr *hdr,
+ const struct volume_key *vk);
+
+int LUKS_check_cipher(struct crypt_device *ctx,
+ size_t keylength,
+ const char *cipher,
+ const char *cipher_mode);
+
+int LUKS_generate_phdr(struct luks_phdr *header,
+ const struct volume_key *vk,
+ const char *cipherName,
+ const char *cipherMode,
+ const char *hashSpec,
+ const char *uuid,
+ uint64_t data_offset,
+ uint64_t align_offset,
+ uint64_t required_alignment,
+ struct crypt_device *ctx);
+
+int LUKS_read_phdr(
+ struct luks_phdr *hdr,
+ int require_luks_device,
+ int repair,
+ struct crypt_device *ctx);
+
+int LUKS_read_phdr_backup(
+ const char *backup_file,
+ struct luks_phdr *hdr,
+ int require_luks_device,
+ struct crypt_device *ctx);
+
+int LUKS_hdr_uuid_set(
+ struct luks_phdr *hdr,
+ const char *uuid,
+ struct crypt_device *ctx);
+
+int LUKS_hdr_backup(
+ const char *backup_file,
+ struct crypt_device *ctx);
+
+int LUKS_hdr_restore(
+ const char *backup_file,
+ struct luks_phdr *hdr,
+ struct crypt_device *ctx);
+
+int LUKS_write_phdr(
+ struct luks_phdr *hdr,
+ struct crypt_device *ctx);
+
+int LUKS_set_key(
+ unsigned int keyIndex,
+ const char *password,
+ size_t passwordLen,
+ struct luks_phdr *hdr,
+ struct volume_key *vk,
+ struct crypt_device *ctx);
+
+int LUKS_open_key_with_hdr(
+ int keyIndex,
+ const char *password,
+ size_t passwordLen,
+ struct luks_phdr *hdr,
+ struct volume_key **vk,
+ struct crypt_device *ctx);
+
+int LUKS_del_key(
+ unsigned int keyIndex,
+ struct luks_phdr *hdr,
+ struct crypt_device *ctx);
+
+int LUKS_wipe_header_areas(struct luks_phdr *hdr,
+ struct crypt_device *ctx);
+
+crypt_keyslot_info LUKS_keyslot_info(struct luks_phdr *hdr, int keyslot);
+int LUKS_keyslot_find_empty(struct luks_phdr *hdr);
+int LUKS_keyslot_active_count(struct luks_phdr *hdr);
+int LUKS_keyslot_set(struct luks_phdr *hdr, int keyslot, int enable,
+ struct crypt_device *ctx);
+int LUKS_keyslot_area(const struct luks_phdr *hdr,
+ int keyslot,
+ uint64_t *offset,
+ uint64_t *length);
+size_t LUKS_device_sectors(const struct luks_phdr *hdr);
+size_t LUKS_keyslots_offset(const struct luks_phdr *hdr);
+int LUKS_keyslot_pbkdf(struct luks_phdr *hdr, int keyslot,
+ struct crypt_pbkdf_type *pbkdf);
+
+int LUKS1_activate(struct crypt_device *cd,
+ const char *name,
+ struct volume_key *vk,
+ uint32_t flags);
+
+#endif