aboutsummaryrefslogtreecommitdiff
path: root/tests/test-ecdsa.py
blob: f50cf5978221049b6a881c25c25eb218382b6cb7 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
# Test vectors from "Suite B Implementer's Guide to FIPS 186-3".
#
# e is given in decimal, all other values are hex, because that's how
# these were given in the paper

p256_d    = 0x70a12c2db16845ed56ff68cfc21a472b3f04d7d6851bf6349f2d7d5b3452b38a
p256_Qx   = 0x8101ece47464a6ead70cf69a6e2bd3d88691a3262d22cba4f7635eaff26680a8
p256_Qy   = 0xd8a12ba61d599235f67d9cb4d58f1783d3ca43e78f0a5abaa624079936c0c3a9
p256_k    = 0x580ec00d856434334cef3f71ecaed4965b12ae37fa47055b1965c7b134ee45d0
p256_kinv = 0x6a664fa115356d33f16331b54c4e7ce967965386c7dcbf2904604d0c132b4a74
p256_Rx   = 0x7214bc9647160bbd39ff2f80533f5dc6ddd70ddf86bb815661e805d5d4e6f27c
p256_Ry   = 0x8b81e3e977597110c7cf2633435b2294b72642987defd3d4007e1cfc5df84541
p256_r    = p256_Rx
p256_M    = 0x54686973206973206f6e6c7920612074657374206d6573736167652e204974206973203438206279746573206c6f6e67
p256_H    = 0x7c3e883ddc8bd688f96eac5e9324222c8f30f9d6bb59e9c5f020bd39ba2b8377
p256_e    = 56197278047627432394583341962843287937266210957576322469816113796290471232375
p256_s    = 0x7d1ff961980f961bdaa3233b6209f4013317d3e3f9e1493592dbeaa1af2bc367
p256_w    = 0xd69be75f67ee5394cabb6c286f3610cf62d722cba9eea70faee770a6b2ed72dc
p256_u1   = 0xbb252401d6fb322bb747184cf2ac52bf8d54b95a1515062a2f6141f2e2092ed8
p256_u2   = 0xaae7d1c7f2c232dfc641948af3dba141d4de8634e571cf84c486301b510cfc04
p256_v    = 0x7214bc9647160bbd39ff2f80533f5dc6ddd70ddf86bb815661e805d5d4e6f27c

p384_d    = 0xc838b85253ef8dc7394fa5808a5183981c7deef5a69ba8f4f2117ffea39cfcd90e95f6cbc854abacab701d50c1f3cf24
p384_Qx   = 0x1fbac8eebd0cbf35640b39efe0808dd774debff20a2a329e91713baf7d7f3c3e81546d883730bee7e48678f857b02ca0
p384_Qy   = 0xeb213103bd68ce343365a8a4c3d4555fa385f5330203bdd76ffad1f3affb95751c132007e1b240353cb0a4cf1693bdf9
p384_k    = 0xdc6b44036989a196e39d1cdac000812f4bdd8b2db41bb33af51372585ebd1db63f0ce8275aa1fd45e2d2a735f8749359
p384_kinv = 0x7436f03088e65c37ba8e7b33887fbc87757514d611f7d1fbdf6d2104a297ad318cdbf7404e4ba37e599666df37b8d8be
p384_Rx   = 0xa0c27ec893092dea1e1bd2ccfed3cf945c8134ed0c9f81311a0f4a05942db8dbed8dd59f267471d5462aa14fe72de856
p384_Ry   = 0x855649409815bb91424eaca5fd76c97375d575d1422ec53d343bd33b847fdf0c11569685b528ab25493015428d7cf72b
p384_r    = p384_Rx
p384_M    = 0x54686973206973206f6e6c7920612074657374206d6573736167652e204974206973203438206279746573206c6f6e67
p384_H    = 0xb9210c9d7e20897ab86597266a9d5077e8db1b06f7220ed6ee75bd8b45db37891f8ba5550304004159f4453dc5b3f5a1
p384_e    = 28493976155450475404302482243066463769180620629462008675793884393889401828800663731864240088367206094074919580333473
p384_s    = 0x20ab3f45b74f10b6e11f96a2c8eb694d206b9dda86d3c7e331c26b22c987b7537726577667adadf168ebbe803794a402
p384_w    = 0x1798845cd0a6cea5327c501a71a4baf2f7be882cfbc303750a7c861af8fe8225467a257f5bf91a4aaa5a79a8637d218a
p384_u1   = 0x6ce25649d42d223e020c11140fe772326612bb11b686d35ee98ed4550e0635d9dd3a2afbca0cf2c4baedcd23313b189e
p384_u2   = 0xf3b240751d5d8ed394a4b5bf8e2a4c0e1e21aa51f2620a08b8c55a2bc334c9689923162648f06e5f4659fc526d9c1fd6
p384_v    = 0xa0c27ec893092dea1e1bd2ccfed3cf945c8134ed0c9f81311a0f4a05942db8dbed8dd59f267471d5462aa14fe72de856

from textwrap                   import TextWrapper
from os.path                    import basename
from sys                        import argv
from pyasn1.type.univ           import Sequence, Choice, Integer, OctetString, ObjectIdentifier, BitString
from pyasn1.type.namedtype      import NamedTypes, NamedType, OptionalNamedType
from pyasn1.type.namedval       import NamedValues
from pyasn1.type.tag            import Tag, tagClassContext, tagFormatSimple
from pyasn1.type.constraint     import SingleValueConstraint
from pyasn1.codec.der.encoder   import encode as DER_Encode
from pyasn1.codec.der.decoder   import decode as DER_Decode

wrapper = TextWrapper(width = 78, initial_indent = " " * 2, subsequent_indent = " " * 2)

def long_to_bytes(number, order):
  #
  # This is just plain nasty.
  #
  s = "%x" % number
  s = ("0" * (order/8 - len(s))) + s
  return s.decode("hex")

def bytes_to_bits(bytes):
  #
  # This, on the other hand, is not just plain nasty, this is fancy nasty.
  # This is nasty with raisins in it.
  #
  s = bin(long(bytes.encode("hex"), 16))[2:]
  if len(s) % 8:
    s = ("0" * (8 - len(s) % 8)) + s
  return tuple(int(i) for i in s)

###

def encode_sig(r, s, order):
  return long_to_bytes(r, order) + long_to_bytes(s, order)

p256_sig = encode_sig(p256_r, p256_s, 256)
p384_sig = encode_sig(p384_r, p384_s, 384)

###

class ECPrivateKey(Sequence):
  componentType = NamedTypes(
    NamedType("version", Integer(namedValues = NamedValues(("ecPrivkeyVer1", 1))
                                 ).subtype(subtypeSpec = Integer.subtypeSpec + SingleValueConstraint(1))),
    NamedType("privateKey", OctetString()),
    OptionalNamedType("parameters", ObjectIdentifier().subtype(explicitTag = Tag(tagClassContext, tagFormatSimple, 0))),
    OptionalNamedType("publicKey", BitString().subtype(explicitTag = Tag(tagClassContext, tagFormatSimple, 1))))

def encode_key(d, Qx, Qy, order, oid):
  private_key = long_to_bytes(d, order)
  public_key  = bytes_to_bits(chr(0x04) + long_to_bytes(Qx, order) + long_to_bytes(Qy, order))
  parameters = oid
  key = ECPrivateKey()
  key["version"]    = 1
  key["privateKey"] = private_key
  key["parameters"] = parameters
  key["publicKey"]  = public_key
  return DER_Encode(key)

p256_key = encode_key(p256_d, p256_Qx, p256_Qy, 256, "1.2.840.10045.3.1.7")
p384_key = encode_key(p384_d, p384_Qx, p384_Qy, 384, "1.3.132.0.34")

###

print "/*"
print " * ECDSA test data."
print " * File automatically generated by", basename(argv[0])
print " */"

curves = ("p256", "p384")
vars   = set()

for name in dir():
  head, sep, tail = name.partition("_")
  if head in curves:
    vars.add(tail)

vars = sorted(vars)

for curve in curves:
  order = int(curve[1:])
  for var in vars:
    name = curve + "_" + var
    value = globals().get(name, None)
    if isinstance(value, (int, long)):
      value = long_to_bytes(value, order)
    if value is not None:
      print
      print "static const uint8_t %s[] = { /* %d bytes */" % (name, len(value))
      print wrapper.fill(", ".join("0x%02x" % ord(v) for v in value))
      print "};"

print
print "typedef struct {"
print "  hal_curve_name_t curve;"
for var in vars:
  print "  const uint8_t *%8s; size_t %8s_len;" % (var, var)
print "} ecdsa_tc_t;"
print
print "static const ecdsa_tc_t ecdsa_tc[] = {"
for curve in curves:
  print "  { HAL_CURVE_%s," % curve.upper()
  for var in vars:
    name = curve + "_" + var
    if name in globals():
      print "    %-14s sizeof(%s)," % (name + ",", name)
    else:
      print "    %-14s 0," % "NULL,"
  print "  },"
print "};"
='#n1047'>1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761
/*
 * ks_flash.c
 * ----------
 * Keystore implementation in flash memory.
 *
 * Authors: Rob Austein, Fredrik Thulin
 * Copyright (c) 2015-2016, NORDUnet A/S All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are
 * met:
 * - Redistributions of source code must retain the above copyright notice,
 *   this list of conditions and the following disclaimer.
 *
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 *
 * - Neither the name of the NORDUnet nor the names of its contributors may
 *   be used to endorse or promote products derived from this software
 *   without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
 * IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include <stddef.h>
#include <string.h>
#include <assert.h>

#include "hal.h"
#include "hal_internal.h"

#include "last_gasp_pin_internal.h"

#define HAL_OK CMIS_HAL_OK
#include "stm-keystore.h"
#undef HAL_OK

/*
 * Known block states.
 *
 * C does not guarantee any particular representation for enums, so
 * including enums directly in the block header isn't safe.  Instead,
 * we use an access method which casts when reading from the header.
 * Writing to the header isn't a problem, because C does guarantee
 * that enum is compatible with *some* integer type, it just doesn't
 * specify which one.
 */

typedef enum {
  BLOCK_TYPE_ERASED  = 0xFF, /* Pristine erased block (candidate for reuse) */
  BLOCK_TYPE_ZEROED  = 0x00, /* Zeroed block (recently used) */
  BLOCK_TYPE_KEY     = 0x55, /* Block contains key material */
  BLOCK_TYPE_ATTR    = 0x66, /* Block contains key attributes (overflow from key block) */
  BLOCK_TYPE_PIN     = 0xAA, /* Block contains PINs */
  BLOCK_TYPE_UNKNOWN = -1,   /* Internal code for "I have no clue what this is" */
} flash_block_type_t;

/*
 * Block status.
 */

typedef enum {
  BLOCK_STATUS_LIVE      = 0x66, /* This is a live flash block */
  BLOCK_STATUS_TOMBSTONE = 0x44, /* This is a tombstone left behind during an update  */
  BLOCK_STATUS_UNKNOWN   = -1,   /* Internal code for "I have no clue what this is" */
} flash_block_status_t;

/*
 * Common header for all flash block types.
 * A few of these fields are deliberately omitted from the CRC.
 */

typedef struct {
  uint8_t               block_type;
  uint8_t               block_status;
  uint8_t               total_chunks;
  uint8_t               this_chunk;
  hal_crc32_t           crc;
} flash_block_header_t;

/*
 * Key block.  Tail end of "der" field (after der_len) used for attributes.
 */

typedef struct {
  flash_block_header_t  header;
  hal_uuid_t            name;
  hal_key_type_t        type;
  hal_curve_name_t      curve;
  hal_key_flags_t       flags;
  size_t                der_len;
  unsigned              attributes_len;
  uint8_t               der[];  /* Must be last field -- C99 "flexible array member" */
} flash_key_block_t;

#define SIZEOF_FLASH_KEY_BLOCK_DER \
  (KEYSTORE_SUBSECTOR_SIZE - offsetof(flash_key_block_t, der))

/*
 * Key attribute overflow block (attributes which don't fit in der field of key block).
 */

typedef struct {
  flash_block_header_t  header;
  hal_uuid_t            name;
  unsigned              attributes_len;
  uint8_t               attributes[]; /* Must be last field -- C99 "flexible array member" */
} flash_attributes_block_t;

#define SIZEOF_FLASH_ATTRIBUTE_BLOCK_ATTRIBUTES \
  (KEYSTORE_SUBSECTOR_SIZE - offsetof(flash_attributes_block_t, attributes))

/*
 * PIN block.  Also includes space for backing up the KEK when
 * HAL_MKM_FLASH_BACKUP_KLUDGE is enabled.
 */

typedef struct {
  flash_block_header_t  header;
  hal_ks_pin_t          wheel_pin;
  hal_ks_pin_t          so_pin;
  hal_ks_pin_t          user_pin;
#if HAL_MKM_FLASH_BACKUP_KLUDGE
  uint32_t              kek_set;
  uint8_t               kek[KEK_LENGTH];
#endif
} flash_pin_block_t;

#define FLASH_KEK_SET   0x33333333

/*
 * One flash block.
 */

typedef union {
  uint8_t                       bytes[KEYSTORE_SUBSECTOR_SIZE];
  flash_block_header_t          header;
  flash_key_block_t             key;
  flash_attributes_block_t      attr;
  flash_pin_block_t             pin;
} flash_block_t;

/*
 * In-memory cache.
 */

typedef struct {
  unsigned            blockno;
  uint32_t            lru;
  flash_block_t       block;
} cache_block_t;

/*
 * In-memory database.
 *
 * The top-level structure is a static variable; the arrays are allocated at runtime
 * using hal_allocate_static_memory() because they can get kind of large.
 */

#ifndef KS_FLASH_CACHE_SIZE
#define KS_FLASH_CACHE_SIZE 4
#endif

#define NUM_FLASH_BLOCKS        KEYSTORE_NUM_SUBSECTORS

typedef struct {
  hal_ks_t              ks;                  /* Must be first (C "subclassing") */
  hal_ks_index_t        ksi;
  hal_ks_pin_t          wheel_pin;
  hal_ks_pin_t          so_pin;
  hal_ks_pin_t          user_pin;
  uint32_t              cache_lru;
  cache_block_t         *cache;
} db_t;

/*
 * PIN block gets the all-zeros UUID, which will never be returned by
 * the UUID generation code (by definition -- it's not a version 4 UUID).
 */

const static hal_uuid_t pin_uuid = {{0}};

/*
 * The in-memory database structure itself is small, but the arrays it
 * points to are large enough that they come from SDRAM allocated at
 * startup.
 */

static db_t db;

/*
 * Type safe casts.
 */

static inline flash_block_type_t block_get_type(const flash_block_t * const block)
{
  assert(block != NULL);
  return (flash_block_type_t) block->header.block_type;
}

static inline flash_block_status_t block_get_status(const flash_block_t * const block)
{
  assert(block != NULL);
  return (flash_block_status_t) block->header.block_status;
}

/*
 * Pick unused or least-recently-used slot in our in-memory cache.
 *
 * Updating lru values is caller's problem: if caller is using a cache
 * slot as a temporary buffer and there's no point in caching the
 * result, leave the lru values alone and the right thing will happen.
 */

static inline flash_block_t *cache_pick_lru(void)
{
  uint32_t best_delta = 0;
  int      best_index = 0;

  for (int i = 0; i < KS_FLASH_CACHE_SIZE; i++) {

    if (db.cache[i].blockno == ~0)
      return &db.cache[i].block;

    const uint32_t delta = db.cache_lru - db.cache[i].lru;
    if (delta > best_delta) {
      best_delta = delta;
      best_index = i;
    }

  }

  db.cache[best_index].blockno = ~0;
  return &db.cache[best_index].block;
}

/*
 * Find a block in our in-memory cache; return block or NULL if not present.
 */

static inline flash_block_t *cache_find_block(const unsigned blockno)
{
  for (int i = 0; i < KS_FLASH_CACHE_SIZE; i++)
    if (db.cache[i].blockno == blockno)
      return &db.cache[i].block;
  return NULL;
}

/*
 * Mark a block in our in-memory cache as being in current use.
 */

static inline void cache_mark_used(const flash_block_t * const block, const unsigned blockno)
{
  for (int i = 0; i < KS_FLASH_CACHE_SIZE; i++) {
    if (&db.cache[i].block == block) {
      db.cache[i].blockno = blockno;
      db.cache[i].lru = ++db.cache_lru;
      return;
    }
  }
}

/*
 * Release a block from the in-memory cache.
 */

static inline void cache_release(const flash_block_t * const block)
{
  if (block != NULL)
    cache_mark_used(block, ~0);
}

/*
 * Generate CRC-32 for a block.
 *
 * This function needs to understand the structure of
 * flash_block_header_t, so that it can skip over fields that
 * shouldn't be included in the CRC.
 */

static hal_crc32_t calculate_block_crc(const flash_block_t * const block)
{
  assert(block != NULL);

  hal_crc32_t crc = hal_crc32_init();

  crc = hal_crc32_update(crc, &block->header.block_type,
                         sizeof(block->header.block_type));

  crc = hal_crc32_update(crc, &block->header.total_chunks,
                         sizeof(block->header.total_chunks));

  crc = hal_crc32_update(crc, &block->header.this_chunk,
                         sizeof(block->header.this_chunk));

  crc = hal_crc32_update(crc, block->bytes + sizeof(flash_block_header_t),
                         sizeof(*block) - sizeof(flash_block_header_t));

  return hal_crc32_finalize(crc);
}

/*
 * Calculate block offset.
 */

static uint32_t block_offset(const unsigned blockno)
{
  return blockno * KEYSTORE_SUBSECTOR_SIZE;
}

/*
 * Read a flash block.
 *
 * Flash read on the Alpha is slow enough that it pays to check the
 * first page before reading the rest of the block.
 */

static hal_error_t block_read(const unsigned blockno, flash_block_t *block)
{
  if (block == NULL || blockno >= NUM_FLASH_BLOCKS || sizeof(*block) != KEYSTORE_SUBSECTOR_SIZE)
    return HAL_ERROR_IMPOSSIBLE;

  /* Sigh, magic numeric return codes */
  if (keystore_read_data(block_offset(blockno),
                         block->bytes,
                         KEYSTORE_PAGE_SIZE) != 1)
    return HAL_ERROR_KEYSTORE_ACCESS;

  switch (block_get_type(block)) {
  case BLOCK_TYPE_ERASED:
  case BLOCK_TYPE_ZEROED:
    return HAL_OK;
  case BLOCK_TYPE_KEY:
  case BLOCK_TYPE_PIN:
  case BLOCK_TYPE_ATTR:
    break;
  default:
    return HAL_ERROR_KEYSTORE_BAD_BLOCK_TYPE;
  }

  switch (block_get_status(block)) {
  case BLOCK_STATUS_LIVE:
  case BLOCK_STATUS_TOMBSTONE:
    break;
  default:
    return HAL_ERROR_KEYSTORE_BAD_BLOCK_TYPE;
  }

  /* Sigh, magic numeric return codes */
  if (keystore_read_data(block_offset(blockno) + KEYSTORE_PAGE_SIZE,
                         block->bytes + KEYSTORE_PAGE_SIZE,
                         sizeof(*block) - KEYSTORE_PAGE_SIZE) != 1)
    return HAL_ERROR_KEYSTORE_ACCESS;

  if (calculate_block_crc(block) != block->header.crc)
    return HAL_ERROR_KEYSTORE_BAD_CRC;

  return HAL_OK;
}

/*
 * Read a block using the cache.  Marking the block as used is left
 * for the caller, so we can avoid blowing out the cache when we
 * perform a ks_list() operation.
 */

static hal_error_t block_read_cached(const unsigned blockno, flash_block_t **block)
{
  if (block == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  if ((*block = cache_find_block(blockno)) != NULL)
    return HAL_OK;

  if ((*block = cache_pick_lru()) == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  return block_read(blockno, *block);
}

/*
 * Convert a live block into a tombstone.  Caller is responsible for
 * making sure that the block being converted is valid; since we don't
 * need to update the CRC for this, we just modify the first page.
 */

static hal_error_t block_deprecate(const unsigned blockno)
{
  if (blockno >= NUM_FLASH_BLOCKS)
    return HAL_ERROR_IMPOSSIBLE;

  uint8_t page[KEYSTORE_PAGE_SIZE];
  flash_block_header_t *header = (void *) page;
  uint32_t offset = block_offset(blockno);

  /* Sigh, magic numeric return codes */
  if (keystore_read_data(offset, page, sizeof(page)) != 1)
    return HAL_ERROR_KEYSTORE_ACCESS;

  header->block_status = BLOCK_STATUS_TOMBSTONE;

  /* Sigh, magic numeric return codes */
  if (keystore_write_data(offset, page, sizeof(page)) != 1)
    return HAL_ERROR_KEYSTORE_ACCESS;

  return HAL_OK;
}

/*
 * Zero (not erase) a flash block.  Just need to zero the first page.
 */

static hal_error_t block_zero(const unsigned blockno)
{
  if (blockno >= NUM_FLASH_BLOCKS)
    return HAL_ERROR_IMPOSSIBLE;

  uint8_t page[KEYSTORE_PAGE_SIZE] = {0};

  /* Sigh, magic numeric return codes */
  if (keystore_write_data(block_offset(blockno), page, sizeof(page)) != 1)
    return HAL_ERROR_KEYSTORE_ACCESS;

  return HAL_OK;
}

/*
 * Erase a flash block.  Also see block_erase_maybe(), below.
 */

static hal_error_t block_erase(const unsigned blockno)
{
  if (blockno >= NUM_FLASH_BLOCKS)
    return HAL_ERROR_IMPOSSIBLE;

  /* Sigh, magic numeric return codes */
  if (keystore_erase_subsectors(blockno, blockno) != 1)
    return HAL_ERROR_KEYSTORE_ACCESS;

  return HAL_OK;
}

/*
 * Erase a flash block if it hasn't already been erased.
 * May not be necessary, trying to avoid unnecessary wear.
 *
 * Unclear whether there's any sane reason why this needs to be
 * constant time, given how slow erasure is.  But side channel attacks
 * can be tricky things, and it's theoretically possible that we could
 * leak information about, eg, key length, so we do constant time.
 */

static hal_error_t block_erase_maybe(const unsigned blockno)
{
  if (blockno >= NUM_FLASH_BLOCKS)
    return HAL_ERROR_IMPOSSIBLE;

  uint8_t mask = 0xFF;

  for (uint32_t a = block_offset(blockno); a < block_offset(blockno + 1); a += KEYSTORE_PAGE_SIZE) {
    uint8_t page[KEYSTORE_PAGE_SIZE];
    if (keystore_read_data(a, page, sizeof(page)) != 1)
      return HAL_ERROR_KEYSTORE_ACCESS;
    for (int i = 0; i < KEYSTORE_PAGE_SIZE; i++)
      mask &= page[i];
  }

  return mask == 0xFF ? HAL_OK : block_erase(blockno);
}

/*
 * Write a flash block, calculating CRC when appropriate.
 */

static hal_error_t block_write(const unsigned blockno, flash_block_t *block)
{
  if (block == NULL || blockno >= NUM_FLASH_BLOCKS || sizeof(*block) != KEYSTORE_SUBSECTOR_SIZE)
    return HAL_ERROR_IMPOSSIBLE;

  hal_error_t err = block_erase_maybe(blockno);

  if (err != HAL_OK)
    return err;

  switch (block_get_type(block)) {
  case BLOCK_TYPE_KEY:
  case BLOCK_TYPE_PIN:
  case BLOCK_TYPE_ATTR:
    block->header.crc = calculate_block_crc(block);
    break;
  default:
    break;
  }

  /* Sigh, magic numeric return codes */
  if (keystore_write_data(block_offset(blockno), block->bytes, sizeof(*block)) != 1)
    return HAL_ERROR_KEYSTORE_ACCESS;

  return HAL_OK;
}

/*
 * Update one flash block, including zombie jamboree.
 */

static hal_error_t block_update(const unsigned b1, flash_block_t *block,
                                const hal_uuid_t * const uuid, const unsigned chunk, int *hint)
{
  if (block == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  if (db.ksi.used == db.ksi.size)
    return HAL_ERROR_NO_KEY_INDEX_SLOTS;

  cache_release(block);

  hal_error_t err;
  unsigned b2;

  if ((err = block_deprecate(b1))                                       != HAL_OK ||
      (err = hal_ks_index_replace(&db.ksi, uuid, chunk, &b2, hint))     != HAL_OK ||
      (err = block_write(b2, block))                                    != HAL_OK ||
      (err = block_zero(b1))                                            != HAL_OK)
    return err;

  cache_mark_used(block, b2);

  return block_erase_maybe(db.ksi.index[db.ksi.used]);
}

/*
 * Forward reference.
 */

static hal_error_t fetch_pin_block(unsigned *b, flash_block_t **block);

/*
 * Initialize keystore.  This includes various tricky bits, some of
 * which attempt to preserve the free list ordering across reboots, to
 * improve our simplistic attempt at wear leveling, others attempt to
 * recover from unclean shutdown.
 */

static inline void *gnaw(uint8_t **mem, size_t *len, const size_t size)
{
  if (mem == NULL || *mem == NULL || len == NULL || size > *len)
    return NULL;
  void *ret = *mem;
  *mem += size;
  *len -= size;
  return ret;
}

static hal_error_t ks_init(const hal_ks_driver_t * const driver)
{
  /*
   * Initialize the in-memory database.
   */

  size_t len = (sizeof(*db.ksi.index) * NUM_FLASH_BLOCKS +
                sizeof(*db.ksi.names) * NUM_FLASH_BLOCKS +
                sizeof(*db.cache)     * KS_FLASH_CACHE_SIZE);

  uint8_t *mem = hal_allocate_static_memory(len);

  if (mem == NULL)
    return HAL_ERROR_ALLOCATION_FAILURE;

  memset(&db, 0, sizeof(db));
  memset(mem, 0, len);

  db.ksi.index = gnaw(&mem, &len, sizeof(*db.ksi.index) * NUM_FLASH_BLOCKS);
  db.ksi.names = gnaw(&mem, &len, sizeof(*db.ksi.names) * NUM_FLASH_BLOCKS);
  db.cache     = gnaw(&mem, &len, sizeof(*db.cache)     * KS_FLASH_CACHE_SIZE);
  db.ksi.size  = NUM_FLASH_BLOCKS;
  db.ksi.used  = 0;

  if (db.ksi.index == NULL || db.ksi.names == NULL || db.cache == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  for (int i = 0; i < KS_FLASH_CACHE_SIZE; i++)
    db.cache[i].blockno = ~0;

  /*
   * Scan existing content of flash to figure out what we've got.
   * This gets a bit involved due to the need to recover from things
   * like power failures at inconvenient times.
   */

  flash_block_type_t   block_types[NUM_FLASH_BLOCKS];
  flash_block_status_t block_status[NUM_FLASH_BLOCKS];
  flash_block_t *block = cache_pick_lru();
  int first_erased = -1;
  hal_error_t err;
  uint16_t n = 0;

  if (block == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  for (int i = 0; i < NUM_FLASH_BLOCKS; i++) {

    /*
     * Read one block.  If the CRC is bad or the block type is
     * unknown, it's old data we don't understand, something we were
     * writing when we crashed, or bad flash; in any of these cases,
     * we want the block to ends up near the end of the free list.
     */

    err = block_read(i, block);

    if (err == HAL_ERROR_KEYSTORE_BAD_CRC || err == HAL_ERROR_KEYSTORE_BAD_BLOCK_TYPE)
      block_types[i] = BLOCK_TYPE_UNKNOWN;

    else if (err == HAL_OK)
      block_types[i] = block_get_type(block);

    else
      return err;

    switch (block_types[i]) {
    case BLOCK_TYPE_KEY:
    case BLOCK_TYPE_PIN:
    case BLOCK_TYPE_ATTR:
      block_status[i] = block_get_status(block);
      break;
    default:
      block_status[i] = BLOCK_STATUS_UNKNOWN;
    }

    /*
     * First erased block we see is head of the free list.
     */

    if (block_types[i] == BLOCK_TYPE_ERASED && first_erased < 0)
      first_erased = i;

    /*
     * If it's a valid data block, include it in the index.  We remove
     * tombstones (if any) below, for now it's easiest to include them
     * in the index, so we can look them up by name if we must.
     */

    const hal_uuid_t *uuid = NULL;

    switch (block_types[i]) {
    case BLOCK_TYPE_KEY:        uuid = &block->key.name;        break;
    case BLOCK_TYPE_ATTR:       uuid = &block->attr.name;       break;
    case BLOCK_TYPE_PIN:        uuid = &pin_uuid;               break;
    default:                    /* Keep GCC happy */            break;
    }

    if (uuid != NULL) {
      db.ksi.names[i].name = *uuid;
      db.ksi.names[i].chunk = block->header.this_chunk;
      db.ksi.index[n++] = i;
    }
  }

  db.ksi.used = n;

  assert(db.ksi.used <= db.ksi.size);

  /*
   * At this point we've built the (unsorted) index from all the valid
   * blocks.  Now we need to insert free and unrecognized blocks into
   * the free list in our preferred order.  It's possible that there's
   * a better way to do this than linear scan, but this is just
   * integer comparisons in a fairly small data set, so it's probably
   * not worth trying to optimize.
   */

  if (n < db.ksi.size)
    for (int i = 0; i < NUM_FLASH_BLOCKS; i++)
      if (block_types[i] == BLOCK_TYPE_ERASED)
        db.ksi.index[n++] = i;

  if (n < db.ksi.size)
    for (int i = first_erased; i < NUM_FLASH_BLOCKS; i++)
      if (block_types[i] == BLOCK_TYPE_ZEROED)
        db.ksi.index[n++] = i;

  if (n < db.ksi.size)
    for (int i = 0; i < first_erased; i++)
      if (block_types[i] == BLOCK_TYPE_ZEROED)
        db.ksi.index[n++] = i;

  if (n < db.ksi.size)
    for (int i = 0; i < NUM_FLASH_BLOCKS; i++)
      if (block_types[i] == BLOCK_TYPE_UNKNOWN)
        db.ksi.index[n++] = i;

  assert(n == db.ksi.size);

  /*
   * Initialize the index.
   */

  if ((err = hal_ks_index_setup(&db.ksi)) != HAL_OK)
    return err;

  /*
   * We might want to call hal_ks_index_fsck() here, if we can figure
   * out some safe set of recovery actions we can take.
   */

  /*
   * Deal with tombstones.  These are blocks left behind when
   * something bad (like a power failure) happened while we updating.
   * The sequence of operations while updating is designed so that,
   * barring a bug or a hardware failure, we should never lose data.
   *
   * For any tombstone we find, we start by looking for all the blocks
   * with a matching UUID, then see what valid sequences we can
   * construct from what we found.
   *
   * If we can construct a valid sequence of live blocks, the complete
   * update was written out, and we just need to zero the tombstones.
   *
   * Otherwise, if we can construct a complete sequence of tombstone
   * blocks, the update failed before it was completely written, so we
   * have to zero the incomplete sequence of live blocks then restore
   * from the tombstones.
   *
   * Otherwise, if the live and tombstone blocks taken together form a
   * valid sequence, the update failed while deprecating the old live
   * blocks, and the update itself was not written, so we need to
   * restore the tombstones and leave the live blocks alone.
   *
   * If none of the above applies, we don't understand what happened,
   * which is a symptom of either a bug or a hardware failure more
   * serious than simple loss of power or reboot at an inconvenient
   * time, so we error out to avoid accidental loss of data.
   */

  for (int i = 0; i < NUM_FLASH_BLOCKS; i++) {

    if (block_status[i] != BLOCK_STATUS_TOMBSTONE)
      continue;

    hal_uuid_t name = db.ksi.names[i].name;
    unsigned n_blocks;
    int where = -1;

    if ((err = hal_ks_index_find_range(&db.ksi, &name, 0, &n_blocks, NULL, &where)) != HAL_OK)
      return err;

    while (where > 0 && !hal_uuid_cmp(&name, &db.ksi.names[db.ksi.index[where - 1]].name)) {
      where--;
      n_blocks++;
    }

    int live_ok = 1, tomb_ok = 1, join_ok = 1;
    unsigned n_live = 0, n_tomb = 0;
    unsigned i_live = 0, i_tomb = 0;

    for (int j = 0; j < n_blocks; j++) {
      unsigned b = db.ksi.index[where + j];
      switch (block_status[b]) {
      case BLOCK_STATUS_LIVE:           n_live++;       break;
      case BLOCK_STATUS_TOMBSTONE:      n_tomb++;       break;
      default:                          return HAL_ERROR_IMPOSSIBLE;
      }
    }

    uint16_t live_blocks[n_live], tomb_blocks[n_tomb];

    for (int j = 0; j < n_blocks; j++) {
      unsigned b = db.ksi.index[where + j];

      if ((err = block_read(b, block)) != HAL_OK)
        return err;

      join_ok &= block->header.this_chunk == j && block->header.total_chunks == n_blocks;

      switch (block_status[b]) {
      case BLOCK_STATUS_LIVE:
        live_blocks[i_live] = b;
        live_ok &= block->header.this_chunk == i_live++ && block->header.total_chunks == n_live;
        break;
      case BLOCK_STATUS_TOMBSTONE:
        tomb_blocks[i_tomb] = b;
        tomb_ok &= block->header.this_chunk == i_tomb++ && block->header.total_chunks == n_tomb;
        break;
      default:
        return HAL_ERROR_IMPOSSIBLE;
      }
    }

    if (!live_ok && !tomb_ok && !join_ok)
      return HAL_ERROR_KEYSTORE_LOST_DATA;

    if (live_ok) {
      for (int j = 0; j < n_tomb; j++) {
        const unsigned b = tomb_blocks[j];
        if ((err = block_zero(b)) != HAL_OK)
          return err;
        block_types[b]  = BLOCK_TYPE_ZEROED;
        block_status[b] = BLOCK_STATUS_UNKNOWN;
      }
    }

    else if (tomb_ok) {
      for (int j = 0; j < n_live; j++) {
        const unsigned b = live_blocks[j];
        if ((err = block_zero(b)) != HAL_OK)
          return err;
        block_types[b]  = BLOCK_TYPE_ZEROED;
        block_status[b] = BLOCK_STATUS_UNKNOWN;
      }
    }

    if (live_ok) {
      memcpy(&db.ksi.index[where], live_blocks, n_live * sizeof(*db.ksi.index));
      memmove(&db.ksi.index[where + n_live], &db.ksi.index[where + n_blocks],
              (db.ksi.size - where - n_blocks) * sizeof(*db.ksi.index));
      memcpy(&db.ksi.index[db.ksi.size - n_tomb], tomb_blocks, n_tomb * sizeof(*db.ksi.index));
      db.ksi.used -= n_tomb;
      n_blocks = n_live;
    }

    else if (tomb_ok) {
      memcpy(&db.ksi.index[where], tomb_blocks, n_tomb * sizeof(*db.ksi.index));
      memmove(&db.ksi.index[where + n_tomb], &db.ksi.index[where + n_blocks],
              (db.ksi.size - where - n_blocks) * sizeof(*db.ksi.index));
      memcpy(&db.ksi.index[db.ksi.size - n_live], live_blocks, n_live * sizeof(*db.ksi.index));
      db.ksi.used -= n_live;
      n_blocks = n_tomb;
    }

    for (int j = 0; j < n_blocks; j++) {
      int hint = where + j;
      unsigned b1 = db.ksi.index[hint], b2;
      if (block_status[b1] != BLOCK_STATUS_TOMBSTONE)
        continue;
      if ((err = block_read(b1, block)) != HAL_OK)
        return err;
      block->header.block_status = BLOCK_STATUS_LIVE;
      if ((err = hal_ks_index_replace(&db.ksi, &name, j, &b2, &hint)) != HAL_OK ||
          (err = block_write(b2, block)) != HAL_OK)
        return err;
      block_types[b1]  = BLOCK_TYPE_ZEROED;
      block_status[b1] = BLOCK_STATUS_UNKNOWN;
      block_status[b2] = BLOCK_STATUS_LIVE;
    }
  }

  err = fetch_pin_block(NULL, &block);

  if (err == HAL_OK) {
    db.wheel_pin = block->pin.wheel_pin;
    db.so_pin    = block->pin.so_pin;
    db.user_pin  = block->pin.user_pin;
  }

  else if (err != HAL_ERROR_KEY_NOT_FOUND)
    return err;

  else {
    /*
     * We found no PIN block, so create one, with the user and so PINs
     * cleared and the wheel PIN set to the last-gasp value.  The
     * last-gasp WHEEL PIN is a terrible answer, but we need some kind
     * of bootstrapping mechanism when all else fails.  If you have a
     * better suggestion, we'd love to hear it.
     */

    unsigned b;

    memset(block, 0xFF, sizeof(*block));

    block->header.block_type   = BLOCK_TYPE_PIN;
    block->header.block_status = BLOCK_STATUS_LIVE;
    block->header.total_chunks = 1;
    block->header.this_chunk   = 0;

    block->pin.wheel_pin = db.wheel_pin = hal_last_gasp_pin;
    block->pin.so_pin    = db.so_pin;
    block->pin.user_pin  = db.user_pin;

    if ((err = hal_ks_index_add(&db.ksi, &pin_uuid, 0, &b, NULL)) != HAL_OK)
      return err;

    cache_mark_used(block, b);

    err = block_write(b, block);

    cache_release(block);

    if (err != HAL_OK)
      return err;
  }

  /*
   * Erase first block on free list if it's not already erased.
   */

  if (db.ksi.used < db.ksi.size &&
      (err = block_erase_maybe(db.ksi.index[db.ksi.used])) != HAL_OK)
    return err;

  /*
   * And we're finally done.
   */

  db.ks.driver = driver;

  return HAL_OK;
}

static hal_error_t ks_shutdown(const hal_ks_driver_t * const driver)
{
  if (db.ks.driver != driver)
    return HAL_ERROR_KEYSTORE_ACCESS;
  return HAL_OK;
}

static hal_error_t ks_open(const hal_ks_driver_t * const driver,
                                    hal_ks_t **ks)
{
  if (driver != hal_ks_token_driver || ks == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  *ks = &db.ks;
  return HAL_OK;
}

static hal_error_t ks_close(hal_ks_t *ks)
{
  if (ks != NULL && ks != &db.ks)
    return HAL_ERROR_BAD_ARGUMENTS;

  return HAL_OK;
}

static inline int acceptable_key_type(const hal_key_type_t type)
{
  switch (type) {
  case HAL_KEY_TYPE_RSA_PRIVATE:
  case HAL_KEY_TYPE_EC_PRIVATE:
  case HAL_KEY_TYPE_RSA_PUBLIC:
  case HAL_KEY_TYPE_EC_PUBLIC:
    return 1;
  default:
    return 0;
  }
}

static hal_error_t ks_store(hal_ks_t *ks,
                            hal_pkey_slot_t *slot,
                            const uint8_t * const der, const size_t der_len)
{
  if (ks != &db.ks || slot == NULL || der == NULL || der_len == 0 || !acceptable_key_type(slot->type))
    return HAL_ERROR_BAD_ARGUMENTS;

  flash_block_t *block = cache_pick_lru();
  flash_key_block_t *k = &block->key;
  uint8_t kek[KEK_LENGTH];
  size_t kek_len;
  hal_error_t err;
  unsigned b;

  if (block == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  if ((err = hal_ks_index_add(&db.ksi, &slot->name, 0, &b, &slot->hint)) != HAL_OK)
    return err;

  cache_mark_used(block, b);

  memset(block, 0xFF, sizeof(*block));

  block->header.block_type   = BLOCK_TYPE_KEY;
  block->header.block_status = BLOCK_STATUS_LIVE;
  block->header.total_chunks = 1;
  block->header.this_chunk   = 0;

  k->name    = slot->name;
  k->type    = slot->type;
  k->curve   = slot->curve;
  k->flags   = slot->flags;
  k->der_len = SIZEOF_FLASH_KEY_BLOCK_DER;
  k->attributes_len = 0;

  if ((err = hal_mkm_get_kek(kek, &kek_len, sizeof(kek))) == HAL_OK)
    err = hal_aes_keywrap(NULL, kek, kek_len, der, der_len, k->der, &k->der_len);

  memset(kek, 0, sizeof(kek));

  if (err == HAL_OK &&
      (err = block_write(b, block)) == HAL_OK)
    return HAL_OK;

  memset(block, 0, sizeof(*block));
  cache_release(block);
  (void) hal_ks_index_delete(&db.ksi, &slot->name, 0, NULL, &slot->hint);
  return err;
}

static hal_error_t ks_fetch(hal_ks_t *ks,
                            hal_pkey_slot_t *slot,
                            uint8_t *der, size_t *der_len, const size_t der_max)
{
  if (ks != &db.ks || slot == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  flash_block_t *block;
  hal_error_t err;
  unsigned b;

  if ((err = hal_ks_index_find(&db.ksi, &slot->name, 0, &b, &slot->hint)) != HAL_OK ||
      (err = block_read_cached(b, &block))                                != HAL_OK)
    return err;

  if (block_get_type(block) != BLOCK_TYPE_KEY)
    return HAL_ERROR_KEYSTORE_WRONG_BLOCK_TYPE; /* HAL_ERROR_KEY_NOT_FOUND */

  cache_mark_used(block, b);

  flash_key_block_t *k = &block->key;

  slot->type  = k->type;
  slot->curve = k->curve;
  slot->flags = k->flags;

  if (der == NULL && der_len != NULL)
    *der_len = k->der_len;

  if (der != NULL) {

    uint8_t kek[KEK_LENGTH];
    size_t kek_len, der_len_;
    hal_error_t err;

    if (der_len == NULL)
      der_len = &der_len_;

    *der_len = der_max;

    if ((err = hal_mkm_get_kek(kek, &kek_len, sizeof(kek))) == HAL_OK)
      err = hal_aes_keyunwrap(NULL, kek, kek_len, k->der, k->der_len, der, der_len);

    memset(kek, 0, sizeof(kek));

    if (err != HAL_OK)
      return err;
  }

  return HAL_OK;
}

static hal_error_t ks_delete(hal_ks_t *ks,
                             hal_pkey_slot_t *slot)
{
  if (ks != &db.ks || slot == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  hal_error_t err;
  unsigned n;

  if ((err = hal_ks_index_delete_range(&db.ksi, &slot->name, 0, &n, NULL, &slot->hint)) != HAL_OK)
    return err;

  unsigned b[n];

  if ((err = hal_ks_index_delete_range(&db.ksi, &slot->name, n, NULL, b, &slot->hint)) != HAL_OK)
    return err;

  for (int i = 0; i < n; i++)
    cache_release(cache_find_block(b[i]));

  for (int i = 0; i < n; i++)
    if ((err = block_zero(b[i])) != HAL_OK)
      return err;

  return block_erase_maybe(db.ksi.index[db.ksi.used]);
}

static hal_error_t ks_list(hal_ks_t *ks,
                           const hal_client_handle_t client,
                           const hal_session_handle_t session,
                           hal_pkey_info_t *result,
                           unsigned *result_len,
                           const unsigned result_max)
{
  if (ks != &db.ks || result == NULL || result_len == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  flash_block_t *block;
  hal_error_t err;

  *result_len = 0;

  for (int i = 0; i < db.ksi.used; i++) {
    unsigned b = db.ksi.index[i];

    if (*result_len >= result_max)
      return HAL_ERROR_RESULT_TOO_LONG;

    if ((err = block_read_cached(b, &block)) != HAL_OK)
      return err;

    if (block_get_type(block) != BLOCK_TYPE_KEY || block->header.this_chunk > 0)
      continue;

    result[*result_len].type  = block->key.type;
    result[*result_len].curve = block->key.curve;
    result[*result_len].flags = block->key.flags;
    result[*result_len].name  = block->key.name;
    ++ *result_len;
  }

  return HAL_OK;
}

static inline hal_error_t locate_attributes(flash_block_t *block, const unsigned chunk,
                                            uint8_t **bytes, size_t *bytes_len,
                                            unsigned **attrs_len)
{
  if (block == NULL || bytes == NULL || bytes_len == NULL || attrs_len == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  if (chunk == 0) {
    if (block_get_type(block) != BLOCK_TYPE_KEY)
      return HAL_ERROR_KEYSTORE_WRONG_BLOCK_TYPE; /* HAL_ERROR_KEY_NOT_FOUND */
    *attrs_len = &block->key.attributes_len;
    *bytes = block->key.der + block->key.der_len;
    *bytes_len = SIZEOF_FLASH_KEY_BLOCK_DER - block->key.der_len;
  }

  else {
    if (block_get_type(block) != BLOCK_TYPE_ATTR)
      return HAL_ERROR_KEYSTORE_WRONG_BLOCK_TYPE; /* HAL_ERROR_KEY_NOT_FOUND */
    *attrs_len = &block->attr.attributes_len;
    *bytes = block->attr.attributes;
    *bytes_len = SIZEOF_FLASH_ATTRIBUTE_BLOCK_ATTRIBUTES;
  }

  return HAL_OK;
}

static hal_error_t ks_match(hal_ks_t *ks,
                            const hal_client_handle_t client,
                            const hal_session_handle_t session,
                            const hal_key_type_t type,
                            const hal_curve_name_t curve,
                            const hal_key_flags_t flags,
                            hal_rpc_pkey_attribute_t *attributes,
                            const unsigned attributes_len,
                            hal_uuid_t *result,
                            unsigned *result_len,
                            const unsigned result_max,
                            const hal_uuid_t * const previous_uuid)
{
  if (ks == NULL || attributes == NULL ||
      result == NULL || result_len == NULL || previous_uuid == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  uint8_t need_attr[attributes_len > 0 ? attributes_len : 1];
  flash_block_t *block;
  int possible = 0;
  hal_error_t err;
  int i = -1;

  *result_len = 0;

  err = hal_ks_index_find(&db.ksi, previous_uuid, 0, NULL, &i);

  if (err == HAL_ERROR_KEY_NOT_FOUND)
    i--;
  else if (err != HAL_OK)
    return err;

  while (*result_len < result_max && ++i < db.ksi.used) {

    unsigned b = db.ksi.index[i];

    if (db.ksi.names[b].chunk == 0)
      possible = 1;

    if (!possible)
      continue;

    if ((err = block_read_cached(b, &block)) != HAL_OK)
      return err;

    if (db.ksi.names[b].chunk == 0) {
      memset(need_attr, 1, sizeof(need_attr));
      possible = ((type == HAL_KEY_TYPE_NONE || type  == block->key.type) &&
                  (curve == HAL_CURVE_NONE   || curve == block->key.curve));
    }

    if (!possible)
      continue;

    if (attributes_len > 0) {
      uint8_t *bytes = NULL;
      size_t bytes_len = 0;
      unsigned *attrs_len;

      if ((err = locate_attributes(block, db.ksi.names[b].chunk,
                                   &bytes, &bytes_len, &attrs_len)) != HAL_OK)
        return err;

      if (*attrs_len > 0) {
        hal_rpc_pkey_attribute_t attrs[*attrs_len];

        if ((err = hal_ks_attribute_scan(bytes, bytes_len, attrs, *attrs_len, NULL)) != HAL_OK)
          return err;

        for (int j = 0; possible && j < attributes_len; j++) {

          if (!need_attr[j])
            continue;

          for (hal_rpc_pkey_attribute_t *a = attrs; a < attrs + *attrs_len; a++) {
            if (a->type != attributes[j].type)
              continue;
            need_attr[j] = 0;
            possible = (a->length == attributes[j].length &&
                        !memcmp(a->value, attributes[j].value, a->length));
            break;
          }
        }
      }
    }

    if (!possible)
      continue;

    if (attributes_len > 0 && memchr(need_attr, 1, sizeof(need_attr)) != NULL)
      continue;

    result[*result_len] = db.ksi.names[b].name;
    ++*result_len;
    possible = 0;
  }

  return HAL_OK;
}

static  hal_error_t ks_set_attribute(hal_ks_t *ks,
                                     hal_pkey_slot_t *slot,
                                     const uint32_t type,
                                     const uint8_t * const value,
                                     const size_t value_len)
{
  if (ks != &db.ks || slot == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  /*
   * Try to add new attribute as a single-block update.
   */

  flash_block_t *block;
  unsigned chunk = 0;
  hal_error_t err;

  do {
    int hint = slot->hint + chunk;
    unsigned b;

    if ((err = hal_ks_index_find(&db.ksi, &slot->name, chunk, &b, &hint)) != HAL_OK ||
        (err = block_read_cached(b, &block))                              != HAL_OK)
      return err;

    if (block->header.this_chunk != chunk)
      return HAL_ERROR_IMPOSSIBLE;

    if (chunk == 0)
      slot->hint = hint;

    uint8_t *bytes = NULL;
    size_t bytes_len = 0;
    unsigned *attrs_len;

    if ((err = locate_attributes(block, chunk, &bytes, &bytes_len, &attrs_len)) != HAL_OK)
      return err;

    cache_mark_used(block, b);

    hal_rpc_pkey_attribute_t attrs[*attrs_len + 1];
    const unsigned old_attrs_len = *attrs_len;
    size_t total;

    if ((err = hal_ks_attribute_scan(bytes, bytes_len, attrs, *attrs_len, &total)) != HAL_OK)
      return err;

    err = hal_ks_attribute_insert(bytes, bytes_len, attrs, attrs_len, &total, type, value, value_len);

    if (*attrs_len != old_attrs_len && err != HAL_OK)
      cache_release(block);

    if (err == HAL_ERROR_RESULT_TOO_LONG)
      continue;

    if (err != HAL_OK)
      return err;

    return block_update(b, block, &slot->name, chunk, &hint);

  } while (++chunk < block->header.total_chunks);

  /*
   * If we get here, we have to add a new block, which requires
   * rewriting all the others to bump the total_blocks count.  We need
   * to keep track of all the old chunks so we can zero them at the
   * end, and because we can't zero them until we've written out the
   * new chunks, we need enough free blocks for the entire new object.
   */

  const unsigned total_chunks = block->header.total_chunks;
  unsigned b, blocks[total_chunks];

  if (db.ksi.used + total_chunks + 1 > db.ksi.size)
    return HAL_ERROR_NO_KEY_INDEX_SLOTS;

  /*
   * Phase 1: Deprecate all the old chunks, remember where the were.
   */

  for (chunk = 0; chunk < total_chunks; chunk++) {
    int hint = slot->hint + chunk;
    if ((err = hal_ks_index_find(&db.ksi, &slot->name, chunk, &blocks[chunk], &hint)) != HAL_OK ||
        (err = block_deprecate(blocks[chunk]))                                        != HAL_OK)
      return err;
  }

  /*
   * Phase 2: Rewrite all the existing chunks with the new total_chunks value.
   */

  for (chunk = 0; chunk < total_chunks; chunk++) {
    int hint = slot->hint + chunk;
    if ((err = block_read_cached(blocks[chunk], &block)) != HAL_OK)
      return err;
    if (block->header.this_chunk != chunk)
      return HAL_ERROR_IMPOSSIBLE;
    block->header.block_status = BLOCK_STATUS_LIVE;
    block->header.total_chunks = total_chunks + 1;

    uint8_t *bytes = NULL;
    size_t bytes_len = 0;
    unsigned *attrs_len;
    size_t total;

    if ((err = locate_attributes(block, chunk, &bytes, &bytes_len, &attrs_len)) != HAL_OK)
      return err;

    if (*attrs_len > 0) {
      hal_rpc_pkey_attribute_t attrs[*attrs_len];
      if ((err = hal_ks_attribute_scan(bytes, bytes_len, attrs, *attrs_len, &total)) != HAL_OK ||
          (err = hal_ks_attribute_delete(bytes, bytes_len, attrs, attrs_len, &total, type)) != HAL_OK)
        return err;
    }

    unsigned b;

    if ((err = hal_ks_index_replace(&db.ksi, &slot->name, chunk, &b, &hint))    != HAL_OK ||
        (err = block_write(b, block))                                           != HAL_OK)
      return err;
  }

  /*
   * Phase 3: Write the new chunk.
   */

  {
    block = cache_pick_lru();

    memset(block, 0xFF, sizeof(*block));

    block->header.block_type   = BLOCK_TYPE_ATTR;
    block->header.block_status = BLOCK_STATUS_LIVE;
    block->header.total_chunks = total_chunks + 1;
    block->header.this_chunk   = total_chunks;
    block->attr.name = slot->name;
    block->attr.attributes_len = 0;

    hal_rpc_pkey_attribute_t attrs[1];
    size_t total = 0;

    if ((err = hal_ks_attribute_insert(block->attr.attributes,
                                       SIZEOF_FLASH_ATTRIBUTE_BLOCK_ATTRIBUTES,
                                       attrs, &block->attr.attributes_len, &total,
                                       type, value, value_len))                         != HAL_OK ||
        (err = hal_ks_index_add(&db.ksi, &slot->name, total_chunks, &b, NULL))          != HAL_OK ||
        (err = block_write(b, block))                                                   != HAL_OK)
      return err;

    cache_mark_used(block, b);
  }

  /*
   * Phase 4: Zero the old chunks we deprecated in phase 1.
   */

  for (chunk = 0; chunk < total_chunks; chunk++)
    if ((err = block_zero(blocks[chunk])) != HAL_OK)
      return err;

  return HAL_OK;
}

static  hal_error_t ks_get_attribute(hal_ks_t *ks,
                                     hal_pkey_slot_t *slot,
                                     const uint32_t type,
                                     uint8_t *value,
                                     size_t *value_len,
                                     const size_t value_max)
{
  if (ks != &db.ks || slot == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  flash_block_t *block;
  hal_error_t err;
  unsigned b;

  hal_rpc_pkey_attribute_t a = {0, 0, NULL};
  unsigned chunk = 0;

  do {
    int hint = slot->hint + chunk;

    if ((err = hal_ks_index_find(&db.ksi, &slot->name, chunk, &b, &hint)) != HAL_OK ||
        (err = block_read_cached(b, &block))                              != HAL_OK)
      return err;

    if (block->header.this_chunk != chunk)
      return HAL_ERROR_IMPOSSIBLE;

    if (chunk == 0)
      slot->hint = hint;

    cache_mark_used(block, b);

    uint8_t *bytes = NULL;
    size_t bytes_len = 0;
    unsigned *attrs_len;

    if ((err = locate_attributes(block, chunk, &bytes, &bytes_len, &attrs_len)) != HAL_OK)
      return err;

    if (*attrs_len == 0)
      continue;

    hal_rpc_pkey_attribute_t attrs[*attrs_len];

    if ((err = hal_ks_attribute_scan(bytes, bytes_len, attrs, *attrs_len, NULL)) != HAL_OK)
      return err;

    for (int i = 0; a.value == NULL && i < *attrs_len; ++i)
      if (attrs[i].type == type)
        a = attrs[i];

  } while (a.value == NULL && ++chunk < block->header.total_chunks);

  if (a.value == NULL)
    return HAL_ERROR_ATTRIBUTE_NOT_FOUND;

  if (a.length > value_max && value != NULL)
    return HAL_ERROR_RESULT_TOO_LONG;

  if (value != NULL)
    memcpy(value, a.value, a.length);

  if (value_len != NULL)
    *value_len = a.length;

  return HAL_OK;
}

static hal_error_t ks_delete_attribute(hal_ks_t *ks,
                                       hal_pkey_slot_t *slot,
                                       const uint32_t type)
{
  /*
   * For extra credit, we could handle attribute compaction here, but
   * in practice we expect attribute deletion without deleting the
   * entire object to be a rare enough event that it may not be worth
   * it.  Certainly it's not a high priority, so later, if ever.
   */

  if (ks != &db.ks || slot == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  flash_block_t *block;
  unsigned chunk = 0;

  do {
    int hint = slot->hint + chunk;
    hal_error_t err;
    unsigned b;

    if ((err = hal_ks_index_find(&db.ksi, &slot->name, chunk, &b, &hint)) != HAL_OK ||
        (err = block_read_cached(b, &block))                              != HAL_OK)
      return err;

    if (block->header.this_chunk != chunk)
      return HAL_ERROR_IMPOSSIBLE;

    if (chunk == 0)
      slot->hint = hint;

    uint8_t *bytes = NULL;
    size_t bytes_len = 0;
    unsigned *attrs_len;

    if ((err = locate_attributes(block, chunk, &bytes, &bytes_len, &attrs_len)) != HAL_OK)
      return err;

    cache_mark_used(block, b);

    if (*attrs_len == 0)
      continue;

    hal_rpc_pkey_attribute_t attrs[*attrs_len];
    const unsigned old_attrs_len = *attrs_len;
    size_t total;

    if ((err = hal_ks_attribute_scan(  bytes, bytes_len, attrs, *attrs_len, &total))       != HAL_OK ||
        (err = hal_ks_attribute_delete(bytes, bytes_len, attrs,  attrs_len, &total, type)) != HAL_OK)
      return err;

    if (*attrs_len == old_attrs_len)
      continue;

    if ((err = block_update(b, block, &slot->name, chunk, &hint)) != HAL_OK)
      return err;

  } while (++chunk < block->header.total_chunks);

  return HAL_OK;
}

const hal_ks_driver_t hal_ks_token_driver[1] = {{
  ks_init,
  ks_shutdown,
  ks_open,
  ks_close,
  ks_store,
  ks_fetch,
  ks_delete,
  ks_list,
  ks_match,
  ks_set_attribute,
  ks_get_attribute,
  ks_delete_attribute
}};

/*
 * The remaining functions aren't really part of the keystore API per se,
 * but they all involve non-key data which we keep in the keystore
 * because it's the flash we've got.
 */

/*
 * Fetch PIN.  This is always cached, so just returned cached value.
 */

hal_error_t hal_get_pin(const hal_user_t user,
                        const hal_ks_pin_t **pin)
{
  if (pin == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  switch (user) {
  case HAL_USER_WHEEL:  *pin = &db.wheel_pin;  break;
  case HAL_USER_SO:     *pin = &db.so_pin;     break;
  case HAL_USER_NORMAL: *pin = &db.user_pin;   break;
  default:              return HAL_ERROR_BAD_ARGUMENTS;
  }

  return HAL_OK;
}

/*
 * Fetch PIN block.  hint = 0 because we know that the all-zeros UUID
 * should always sort to first slot in the index.
 */

static hal_error_t fetch_pin_block(unsigned *b, flash_block_t **block)
{
  if (block == NULL)
    return HAL_ERROR_IMPOSSIBLE;

  hal_error_t err;
  int hint = 0;
  unsigned b_;

  if (b == NULL)
    b = &b_;

  if ((err = hal_ks_index_find(&db.ksi, &pin_uuid, 0, b, &hint)) != HAL_OK ||
      (err = block_read_cached(*b, block))                       != HAL_OK)
    return err;

  cache_mark_used(*block, *b);

  if (block_get_type(*block) != BLOCK_TYPE_PIN)
    return HAL_ERROR_IMPOSSIBLE;

  return HAL_OK;
}

/*
 * Update the PIN block.  This block should always be present, but we
 * have to do the zombie jamboree to make sure we write the new PIN
 * block before destroying the old one.  hint = 0 because we know that
 * the all-zeros UUID should always sort to first slot in the index.
 */

static hal_error_t update_pin_block(const unsigned b,
                                    flash_block_t *block,
                                    const flash_pin_block_t * const new_data)
{
  if (block == NULL || new_data == NULL || block_get_type(block) != BLOCK_TYPE_PIN)
    return HAL_ERROR_IMPOSSIBLE;

  int hint = 0;

  block->pin = *new_data;

  return block_update(b, block, &pin_uuid, 0, &hint);
}

/*
 * Change a PIN.
 */

hal_error_t hal_set_pin(const hal_user_t user,
                        const hal_ks_pin_t * const pin)
{
  if (pin == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  flash_block_t *block;
  hal_error_t err;
  unsigned b;

  if ((err = fetch_pin_block(&b, &block)) != HAL_OK)
    return err;

  flash_pin_block_t new_data = block->pin;
  hal_ks_pin_t *dp, *bp;

  switch (user) {
  case HAL_USER_WHEEL:  bp = &new_data.wheel_pin; dp = &db.wheel_pin; break;
  case HAL_USER_SO:     bp = &new_data.so_pin;    dp = &db.so_pin;    break;
  case HAL_USER_NORMAL: bp = &new_data.user_pin;  dp = &db.user_pin;  break;
  default:              return HAL_ERROR_BAD_ARGUMENTS;
  }

  const hal_ks_pin_t old_pin = *dp;
  *dp = *bp = *pin;

  if ((err = update_pin_block(b, block, &new_data)) != HAL_OK)
    *dp = old_pin;

  return err;
}

#if HAL_MKM_FLASH_BACKUP_KLUDGE

/*
 * Horrible insecure kludge in lieu of a battery for the MKM.
 *
 * API here is a little strange: all calls pass a length parameter,
 * but any length other than the compiled in constant just returns an
 * immediate error, there's no notion of buffer max length vs buffer
 * used length, querying for the size of buffer really needed, or
 * anything like that.
 *
 * We might want to rewrite this some day, if we don't replace it with
 * a battery first.  For now we just preserve the API as we found it
 * while re-implementing it on top of the new keystore.
 */

hal_error_t hal_mkm_flash_read(uint8_t *buf, const size_t len)
{
  if (buf != NULL && len != KEK_LENGTH)
    return HAL_ERROR_MASTERKEY_BAD_LENGTH;

  flash_block_t *block;
  hal_error_t err;
  unsigned b;

  if ((err = fetch_pin_block(&b, &block)) != HAL_OK)
    return err;

  if (block->pin.kek_set != FLASH_KEK_SET)
    return HAL_ERROR_MASTERKEY_NOT_SET;

  if (buf != NULL)
    memcpy(buf, block->pin.kek, len);

  return HAL_OK;
}

hal_error_t hal_mkm_flash_write(const uint8_t * const buf, const size_t len)
{
  if (buf == NULL)
    return HAL_ERROR_BAD_ARGUMENTS;

  if (len != KEK_LENGTH)
    return HAL_ERROR_MASTERKEY_BAD_LENGTH;

  flash_block_t *block;
  hal_error_t err;
  unsigned b;

  if ((err = fetch_pin_block(&b, &block)) != HAL_OK)
    return err;

  flash_pin_block_t new_data = block->pin;

  new_data.kek_set = FLASH_KEK_SET;
  memcpy(new_data.kek, buf, len);

  return update_pin_block(b, block, &new_data);
}

hal_error_t hal_mkm_flash_erase(const size_t len)
{
  if (len != KEK_LENGTH)
    return HAL_ERROR_MASTERKEY_BAD_LENGTH;

  flash_block_t *block;
  hal_error_t err;
  unsigned b;

  if ((err = fetch_pin_block(&b, &block)) != HAL_OK)
    return err;

  flash_pin_block_t new_data = block->pin;

  new_data.kek_set = FLASH_KEK_SET;
  memset(new_data.kek, 0, len);

  return update_pin_block(b, block, &new_data);
}

#endif /* HAL_MKM_FLASH_BACKUP_KLUDGE */


/*
 * Local variables:
 * indent-tabs-mode: nil
 * End:
 */