Skip to content

Commit 6e5714e

Browse files
committed
net: Compute protocol sequence numbers and fragment IDs using MD5.
Computers have become a lot faster since we compromised on the partial MD4 hash which we use currently for performance reasons. MD5 is a much safer choice, and is inline with both RFC1948 and other ISS generators (OpenBSD, Solaris, etc.) Furthermore, only having 24-bits of the sequence number be truly unpredictable is a very serious limitation. So the periodic regeneration and 8-bit counter have been removed. We compute and use a full 32-bit sequence number. For ipv6, DCCP was found to use a 32-bit truncated initial sequence number (it needs 43-bits) and that is fixed here as well. Reported-by: Dan Kaminsky <dan@doxpara.com> Tested-by: Willy Tarreau <w@1wt.eu> Signed-off-by: David S. Miller <davem@davemloft.net>
1 parent bc0b96b commit 6e5714e

File tree

14 files changed

+223
-361
lines changed

14 files changed

+223
-361
lines changed

drivers/char/random.c

Lines changed: 8 additions & 341 deletions
Original file line numberDiff line numberDiff line change
@@ -1300,363 +1300,30 @@ ctl_table random_table[] = {
13001300
};
13011301
#endif /* CONFIG_SYSCTL */
13021302

1303-
/********************************************************************
1304-
*
1305-
* Random functions for networking
1306-
*
1307-
********************************************************************/
1308-
1309-
/*
1310-
* TCP initial sequence number picking. This uses the random number
1311-
* generator to pick an initial secret value. This value is hashed
1312-
* along with the TCP endpoint information to provide a unique
1313-
* starting point for each pair of TCP endpoints. This defeats
1314-
* attacks which rely on guessing the initial TCP sequence number.
1315-
* This algorithm was suggested by Steve Bellovin.
1316-
*
1317-
* Using a very strong hash was taking an appreciable amount of the total
1318-
* TCP connection establishment time, so this is a weaker hash,
1319-
* compensated for by changing the secret periodically.
1320-
*/
1321-
1322-
/* F, G and H are basic MD4 functions: selection, majority, parity */
1323-
#define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
1324-
#define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
1325-
#define H(x, y, z) ((x) ^ (y) ^ (z))
1326-
1327-
/*
1328-
* The generic round function. The application is so specific that
1329-
* we don't bother protecting all the arguments with parens, as is generally
1330-
* good macro practice, in favor of extra legibility.
1331-
* Rotation is separate from addition to prevent recomputation
1332-
*/
1333-
#define ROUND(f, a, b, c, d, x, s) \
1334-
(a += f(b, c, d) + x, a = (a << s) | (a >> (32 - s)))
1335-
#define K1 0
1336-
#define K2 013240474631UL
1337-
#define K3 015666365641UL
1338-
1339-
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1340-
1341-
static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
1342-
{
1343-
__u32 a = buf[0], b = buf[1], c = buf[2], d = buf[3];
1344-
1345-
/* Round 1 */
1346-
ROUND(F, a, b, c, d, in[ 0] + K1, 3);
1347-
ROUND(F, d, a, b, c, in[ 1] + K1, 7);
1348-
ROUND(F, c, d, a, b, in[ 2] + K1, 11);
1349-
ROUND(F, b, c, d, a, in[ 3] + K1, 19);
1350-
ROUND(F, a, b, c, d, in[ 4] + K1, 3);
1351-
ROUND(F, d, a, b, c, in[ 5] + K1, 7);
1352-
ROUND(F, c, d, a, b, in[ 6] + K1, 11);
1353-
ROUND(F, b, c, d, a, in[ 7] + K1, 19);
1354-
ROUND(F, a, b, c, d, in[ 8] + K1, 3);
1355-
ROUND(F, d, a, b, c, in[ 9] + K1, 7);
1356-
ROUND(F, c, d, a, b, in[10] + K1, 11);
1357-
ROUND(F, b, c, d, a, in[11] + K1, 19);
1358-
1359-
/* Round 2 */
1360-
ROUND(G, a, b, c, d, in[ 1] + K2, 3);
1361-
ROUND(G, d, a, b, c, in[ 3] + K2, 5);
1362-
ROUND(G, c, d, a, b, in[ 5] + K2, 9);
1363-
ROUND(G, b, c, d, a, in[ 7] + K2, 13);
1364-
ROUND(G, a, b, c, d, in[ 9] + K2, 3);
1365-
ROUND(G, d, a, b, c, in[11] + K2, 5);
1366-
ROUND(G, c, d, a, b, in[ 0] + K2, 9);
1367-
ROUND(G, b, c, d, a, in[ 2] + K2, 13);
1368-
ROUND(G, a, b, c, d, in[ 4] + K2, 3);
1369-
ROUND(G, d, a, b, c, in[ 6] + K2, 5);
1370-
ROUND(G, c, d, a, b, in[ 8] + K2, 9);
1371-
ROUND(G, b, c, d, a, in[10] + K2, 13);
1372-
1373-
/* Round 3 */
1374-
ROUND(H, a, b, c, d, in[ 3] + K3, 3);
1375-
ROUND(H, d, a, b, c, in[ 7] + K3, 9);
1376-
ROUND(H, c, d, a, b, in[11] + K3, 11);
1377-
ROUND(H, b, c, d, a, in[ 2] + K3, 15);
1378-
ROUND(H, a, b, c, d, in[ 6] + K3, 3);
1379-
ROUND(H, d, a, b, c, in[10] + K3, 9);
1380-
ROUND(H, c, d, a, b, in[ 1] + K3, 11);
1381-
ROUND(H, b, c, d, a, in[ 5] + K3, 15);
1382-
ROUND(H, a, b, c, d, in[ 9] + K3, 3);
1383-
ROUND(H, d, a, b, c, in[ 0] + K3, 9);
1384-
ROUND(H, c, d, a, b, in[ 4] + K3, 11);
1385-
ROUND(H, b, c, d, a, in[ 8] + K3, 15);
1386-
1387-
return buf[1] + b; /* "most hashed" word */
1388-
/* Alternative: return sum of all words? */
1389-
}
1390-
#endif
1391-
1392-
#undef ROUND
1393-
#undef F
1394-
#undef G
1395-
#undef H
1396-
#undef K1
1397-
#undef K2
1398-
#undef K3
1399-
1400-
/* This should not be decreased so low that ISNs wrap too fast. */
1401-
#define REKEY_INTERVAL (300 * HZ)
1402-
/*
1403-
* Bit layout of the tcp sequence numbers (before adding current time):
1404-
* bit 24-31: increased after every key exchange
1405-
* bit 0-23: hash(source,dest)
1406-
*
1407-
* The implementation is similar to the algorithm described
1408-
* in the Appendix of RFC 1185, except that
1409-
* - it uses a 1 MHz clock instead of a 250 kHz clock
1410-
* - it performs a rekey every 5 minutes, which is equivalent
1411-
* to a (source,dest) tulple dependent forward jump of the
1412-
* clock by 0..2^(HASH_BITS+1)
1413-
*
1414-
* Thus the average ISN wraparound time is 68 minutes instead of
1415-
* 4.55 hours.
1416-
*
1417-
* SMP cleanup and lock avoidance with poor man's RCU.
1418-
* Manfred Spraul <manfred@colorfullife.com>
1419-
*
1420-
*/
1421-
#define COUNT_BITS 8
1422-
#define COUNT_MASK ((1 << COUNT_BITS) - 1)
1423-
#define HASH_BITS 24
1424-
#define HASH_MASK ((1 << HASH_BITS) - 1)
1303+
static u32 random_int_secret[MD5_MESSAGE_BYTES / 4] ____cacheline_aligned;
14251304

1426-
static struct keydata {
1427-
__u32 count; /* already shifted to the final position */
1428-
__u32 secret[12];
1429-
} ____cacheline_aligned ip_keydata[2];
1430-
1431-
static unsigned int ip_cnt;
1432-
1433-
static void rekey_seq_generator(struct work_struct *work);
1434-
1435-
static DECLARE_DELAYED_WORK(rekey_work, rekey_seq_generator);
1436-
1437-
/*
1438-
* Lock avoidance:
1439-
* The ISN generation runs lockless - it's just a hash over random data.
1440-
* State changes happen every 5 minutes when the random key is replaced.
1441-
* Synchronization is performed by having two copies of the hash function
1442-
* state and rekey_seq_generator always updates the inactive copy.
1443-
* The copy is then activated by updating ip_cnt.
1444-
* The implementation breaks down if someone blocks the thread
1445-
* that processes SYN requests for more than 5 minutes. Should never
1446-
* happen, and even if that happens only a not perfectly compliant
1447-
* ISN is generated, nothing fatal.
1448-
*/
1449-
static void rekey_seq_generator(struct work_struct *work)
1305+
static int __init random_int_secret_init(void)
14501306
{
1451-
struct keydata *keyptr = &ip_keydata[1 ^ (ip_cnt & 1)];
1452-
1453-
get_random_bytes(keyptr->secret, sizeof(keyptr->secret));
1454-
keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
1455-
smp_wmb();
1456-
ip_cnt++;
1457-
schedule_delayed_work(&rekey_work,
1458-
round_jiffies_relative(REKEY_INTERVAL));
1459-
}
1460-
1461-
static inline struct keydata *get_keyptr(void)
1462-
{
1463-
struct keydata *keyptr = &ip_keydata[ip_cnt & 1];
1464-
1465-
smp_rmb();
1466-
1467-
return keyptr;
1468-
}
1469-
1470-
static __init int seqgen_init(void)
1471-
{
1472-
rekey_seq_generator(NULL);
1307+
get_random_bytes(random_int_secret, sizeof(random_int_secret));
14731308
return 0;
14741309
}
1475-
late_initcall(seqgen_init);
1476-
1477-
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1478-
__u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
1479-
__be16 sport, __be16 dport)
1480-
{
1481-
__u32 seq;
1482-
__u32 hash[12];
1483-
struct keydata *keyptr = get_keyptr();
1484-
1485-
/* The procedure is the same as for IPv4, but addresses are longer.
1486-
* Thus we must use twothirdsMD4Transform.
1487-
*/
1488-
1489-
memcpy(hash, saddr, 16);
1490-
hash[4] = ((__force u16)sport << 16) + (__force u16)dport;
1491-
memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1492-
1493-
seq = twothirdsMD4Transform((const __u32 *)daddr, hash) & HASH_MASK;
1494-
seq += keyptr->count;
1495-
1496-
seq += ktime_to_ns(ktime_get_real());
1497-
1498-
return seq;
1499-
}
1500-
EXPORT_SYMBOL(secure_tcpv6_sequence_number);
1501-
#endif
1502-
1503-
/* The code below is shamelessly stolen from secure_tcp_sequence_number().
1504-
* All blames to Andrey V. Savochkin <saw@msu.ru>.
1505-
*/
1506-
__u32 secure_ip_id(__be32 daddr)
1507-
{
1508-
struct keydata *keyptr;
1509-
__u32 hash[4];
1510-
1511-
keyptr = get_keyptr();
1512-
1513-
/*
1514-
* Pick a unique starting offset for each IP destination.
1515-
* The dest ip address is placed in the starting vector,
1516-
* which is then hashed with random data.
1517-
*/
1518-
hash[0] = (__force __u32)daddr;
1519-
hash[1] = keyptr->secret[9];
1520-
hash[2] = keyptr->secret[10];
1521-
hash[3] = keyptr->secret[11];
1522-
1523-
return half_md4_transform(hash, keyptr->secret);
1524-
}
1525-
1526-
__u32 secure_ipv6_id(const __be32 daddr[4])
1527-
{
1528-
const struct keydata *keyptr;
1529-
__u32 hash[4];
1530-
1531-
keyptr = get_keyptr();
1532-
1533-
hash[0] = (__force __u32)daddr[0];
1534-
hash[1] = (__force __u32)daddr[1];
1535-
hash[2] = (__force __u32)daddr[2];
1536-
hash[3] = (__force __u32)daddr[3];
1537-
1538-
return half_md4_transform(hash, keyptr->secret);
1539-
}
1540-
1541-
#ifdef CONFIG_INET
1542-
1543-
__u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
1544-
__be16 sport, __be16 dport)
1545-
{
1546-
__u32 seq;
1547-
__u32 hash[4];
1548-
struct keydata *keyptr = get_keyptr();
1549-
1550-
/*
1551-
* Pick a unique starting offset for each TCP connection endpoints
1552-
* (saddr, daddr, sport, dport).
1553-
* Note that the words are placed into the starting vector, which is
1554-
* then mixed with a partial MD4 over random data.
1555-
*/
1556-
hash[0] = (__force u32)saddr;
1557-
hash[1] = (__force u32)daddr;
1558-
hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1559-
hash[3] = keyptr->secret[11];
1560-
1561-
seq = half_md4_transform(hash, keyptr->secret) & HASH_MASK;
1562-
seq += keyptr->count;
1563-
/*
1564-
* As close as possible to RFC 793, which
1565-
* suggests using a 250 kHz clock.
1566-
* Further reading shows this assumes 2 Mb/s networks.
1567-
* For 10 Mb/s Ethernet, a 1 MHz clock is appropriate.
1568-
* For 10 Gb/s Ethernet, a 1 GHz clock should be ok, but
1569-
* we also need to limit the resolution so that the u32 seq
1570-
* overlaps less than one time per MSL (2 minutes).
1571-
* Choosing a clock of 64 ns period is OK. (period of 274 s)
1572-
*/
1573-
seq += ktime_to_ns(ktime_get_real()) >> 6;
1574-
1575-
return seq;
1576-
}
1577-
1578-
/* Generate secure starting point for ephemeral IPV4 transport port search */
1579-
u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
1580-
{
1581-
struct keydata *keyptr = get_keyptr();
1582-
u32 hash[4];
1583-
1584-
/*
1585-
* Pick a unique starting offset for each ephemeral port search
1586-
* (saddr, daddr, dport) and 48bits of random data.
1587-
*/
1588-
hash[0] = (__force u32)saddr;
1589-
hash[1] = (__force u32)daddr;
1590-
hash[2] = (__force u32)dport ^ keyptr->secret[10];
1591-
hash[3] = keyptr->secret[11];
1592-
1593-
return half_md4_transform(hash, keyptr->secret);
1594-
}
1595-
EXPORT_SYMBOL_GPL(secure_ipv4_port_ephemeral);
1596-
1597-
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1598-
u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
1599-
__be16 dport)
1600-
{
1601-
struct keydata *keyptr = get_keyptr();
1602-
u32 hash[12];
1603-
1604-
memcpy(hash, saddr, 16);
1605-
hash[4] = (__force u32)dport;
1606-
memcpy(&hash[5], keyptr->secret, sizeof(__u32) * 7);
1607-
1608-
return twothirdsMD4Transform((const __u32 *)daddr, hash);
1609-
}
1610-
#endif
1611-
1612-
#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
1613-
/* Similar to secure_tcp_sequence_number but generate a 48 bit value
1614-
* bit's 32-47 increase every key exchange
1615-
* 0-31 hash(source, dest)
1616-
*/
1617-
u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
1618-
__be16 sport, __be16 dport)
1619-
{
1620-
u64 seq;
1621-
__u32 hash[4];
1622-
struct keydata *keyptr = get_keyptr();
1623-
1624-
hash[0] = (__force u32)saddr;
1625-
hash[1] = (__force u32)daddr;
1626-
hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
1627-
hash[3] = keyptr->secret[11];
1628-
1629-
seq = half_md4_transform(hash, keyptr->secret);
1630-
seq |= ((u64)keyptr->count) << (32 - HASH_BITS);
1631-
1632-
seq += ktime_to_ns(ktime_get_real());
1633-
seq &= (1ull << 48) - 1;
1634-
1635-
return seq;
1636-
}
1637-
EXPORT_SYMBOL(secure_dccp_sequence_number);
1638-
#endif
1639-
1640-
#endif /* CONFIG_INET */
1641-
1310+
late_initcall(random_int_secret_init);
16421311

16431312
/*
16441313
* Get a random word for internal kernel use only. Similar to urandom but
16451314
* with the goal of minimal entropy pool depletion. As a result, the random
16461315
* value is not cryptographically secure but for several uses the cost of
16471316
* depleting entropy is too high
16481317
*/
1649-
DEFINE_PER_CPU(__u32 [4], get_random_int_hash);
1318+
DEFINE_PER_CPU(__u32 [MD5_DIGEST_WORDS], get_random_int_hash);
16501319
unsigned int get_random_int(void)
16511320
{
1652-
struct keydata *keyptr;
16531321
__u32 *hash = get_cpu_var(get_random_int_hash);
1654-
int ret;
1322+
unsigned int ret;
16551323

1656-
keyptr = get_keyptr();
16571324
hash[0] += current->pid + jiffies + get_cycles();
1658-
1659-
ret = half_md4_transform(hash, keyptr->secret);
1325+
md5_transform(hash, random_int_secret);
1326+
ret = hash[0];
16601327
put_cpu_var(get_random_int_hash);
16611328

16621329
return ret;

include/linux/random.h

Lines changed: 0 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -57,18 +57,6 @@ extern void add_interrupt_randomness(int irq);
5757
extern void get_random_bytes(void *buf, int nbytes);
5858
void generate_random_uuid(unsigned char uuid_out[16]);
5959

60-
extern __u32 secure_ip_id(__be32 daddr);
61-
extern __u32 secure_ipv6_id(const __be32 daddr[4]);
62-
extern u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport);
63-
extern u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
64-
__be16 dport);
65-
extern __u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
66-
__be16 sport, __be16 dport);
67-
extern __u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
68-
__be16 sport, __be16 dport);
69-
extern u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
70-
__be16 sport, __be16 dport);
71-
7260
#ifndef MODULE
7361
extern const struct file_operations random_fops, urandom_fops;
7462
#endif

0 commit comments

Comments
 (0)