openssl/ssl/record/ssl3_record.c

1968 lines
68 KiB
C
Raw Normal View History

/*
* Copyright 1995-2017 The OpenSSL Project Authors. All Rights Reserved.
*
* Licensed under the OpenSSL license (the "License"). You may not use
* this file except in compliance with the License. You can obtain a copy
* in the file LICENSE in the source distribution or at
* https://www.openssl.org/source/license.html
*/
#include "../ssl_locl.h"
#include "internal/constant_time_locl.h"
#include <openssl/rand.h>
#include "record_locl.h"
#include "internal/cryptlib.h"
static const unsigned char ssl3_pad_1[48] = {
0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36,
0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36,
0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36,
0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36,
0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36,
0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36, 0x36
};
static const unsigned char ssl3_pad_2[48] = {
0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c,
0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c,
0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c,
0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c,
0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c,
0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c, 0x5c
};
/*
* Clear the contents of an SSL3_RECORD but retain any memory allocated
*/
void SSL3_RECORD_clear(SSL3_RECORD *r, size_t num_recs)
{
unsigned char *comp;
size_t i;
for (i = 0; i < num_recs; i++) {
comp = r[i].comp;
memset(&r[i], 0, sizeof(*r));
r[i].comp = comp;
}
}
void SSL3_RECORD_release(SSL3_RECORD *r, size_t num_recs)
{
size_t i;
for (i = 0; i < num_recs; i++) {
OPENSSL_free(r[i].comp);
r[i].comp = NULL;
}
}
void SSL3_RECORD_set_seq_num(SSL3_RECORD *r, const unsigned char *seq_num)
{
memcpy(r->seq_num, seq_num, SEQ_NUM_SIZE);
}
/*
* Peeks ahead into "read_ahead" data to see if we have a whole record waiting
* for us in the buffer.
*/
static int ssl3_record_app_data_waiting(SSL *s)
{
SSL3_BUFFER *rbuf;
size_t left, len;
unsigned char *p;
rbuf = RECORD_LAYER_get_rbuf(&s->rlayer);
p = SSL3_BUFFER_get_buf(rbuf);
if (p == NULL)
return 0;
left = SSL3_BUFFER_get_left(rbuf);
if (left < SSL3_RT_HEADER_LENGTH)
return 0;
p += SSL3_BUFFER_get_offset(rbuf);
/*
* We only check the type and record length, we will sanity check version
* etc later
*/
if (*p != SSL3_RT_APPLICATION_DATA)
return 0;
p += 3;
n2s(p, len);
if (left < SSL3_RT_HEADER_LENGTH + len)
return 0;
return 1;
}
int early_data_count_ok(SSL *s, size_t length, size_t overhead, int send)
{
uint32_t max_early_data = s->max_early_data;
SSL_SESSION *sess = s->session;
/*
* If we are a client then we always use the max_early_data from the
* session/psksession. Otherwise we go with the lowest out of the max early
* data set in the session and the configured max_early_data.
*/
if (!s->server && sess->ext.max_early_data == 0) {
if (!ossl_assert(s->psksession != NULL
&& s->psksession->ext.max_early_data > 0)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_EARLY_DATA_COUNT_OK,
ERR_R_INTERNAL_ERROR);
return 0;
}
sess = s->psksession;
}
if (!s->server
|| (s->hit && sess->ext.max_early_data < s->max_early_data))
max_early_data = sess->ext.max_early_data;
if (max_early_data == 0) {
SSLfatal(s, send ? SSL_AD_INTERNAL_ERROR : SSL_AD_UNEXPECTED_MESSAGE,
SSL_F_EARLY_DATA_COUNT_OK, SSL_R_TOO_MUCH_EARLY_DATA);
return 0;
}
/* If we are dealing with ciphertext we need to allow for the overhead */
max_early_data += overhead;
if (s->early_data_count + length > max_early_data) {
SSLfatal(s, send ? SSL_AD_INTERNAL_ERROR : SSL_AD_UNEXPECTED_MESSAGE,
SSL_F_EARLY_DATA_COUNT_OK, SSL_R_TOO_MUCH_EARLY_DATA);
return 0;
}
s->early_data_count += length;
return 1;
}
/*
* MAX_EMPTY_RECORDS defines the number of consecutive, empty records that
* will be processed per call to ssl3_get_record. Without this limit an
* attacker could send empty records at a faster rate than we can process and
* cause ssl3_get_record to loop forever.
*/
#define MAX_EMPTY_RECORDS 32
#define SSL2_RT_HEADER_LENGTH 2
/*-
* Call this to get new input records.
* It will return <= 0 if more data is needed, normally due to an error
* or non-blocking IO.
* When it finishes, |numrpipes| records have been decoded. For each record 'i':
* rr[i].type - is the type of record
* rr[i].data, - data
* rr[i].length, - number of bytes
* Multiple records will only be returned if the record types are all
* SSL3_RT_APPLICATION_DATA. The number of records returned will always be <=
* |max_pipelines|
*/
/* used only by ssl3_read_bytes */
int ssl3_get_record(SSL *s)
{
int enc_err, rret;
int i;
size_t more, n;
SSL3_RECORD *rr, *thisrr;
SSL3_BUFFER *rbuf;
SSL_SESSION *sess;
unsigned char *p;
unsigned char md[EVP_MAX_MD_SIZE];
unsigned int version;
size_t mac_size;
int imac_size;
size_t num_recs = 0, max_recs, j;
PACKET pkt, sslv2pkt;
size_t first_rec_len;
rr = RECORD_LAYER_get_rrec(&s->rlayer);
rbuf = RECORD_LAYER_get_rbuf(&s->rlayer);
max_recs = s->max_pipelines;
if (max_recs == 0)
max_recs = 1;
sess = s->session;
do {
thisrr = &rr[num_recs];
/* check if we have the header */
if ((RECORD_LAYER_get_rstate(&s->rlayer) != SSL_ST_READ_BODY) ||
(RECORD_LAYER_get_packet_length(&s->rlayer)
< SSL3_RT_HEADER_LENGTH)) {
size_t sslv2len;
unsigned int type;
rret = ssl3_read_n(s, SSL3_RT_HEADER_LENGTH,
SSL3_BUFFER_get_len(rbuf), 0,
num_recs == 0 ? 1 : 0, &n);
if (rret <= 0)
return rret; /* error or non-blocking */
RECORD_LAYER_set_rstate(&s->rlayer, SSL_ST_READ_BODY);
p = RECORD_LAYER_get_packet(&s->rlayer);
if (!PACKET_buf_init(&pkt, RECORD_LAYER_get_packet(&s->rlayer),
RECORD_LAYER_get_packet_length(&s->rlayer))) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_SSL3_GET_RECORD,
ERR_R_INTERNAL_ERROR);
return -1;
}
sslv2pkt = pkt;
if (!PACKET_get_net_2_len(&sslv2pkt, &sslv2len)
|| !PACKET_get_1(&sslv2pkt, &type)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_SSL3_GET_RECORD,
ERR_R_INTERNAL_ERROR);
return -1;
}
/*
* The first record received by the server may be a V2ClientHello.
*/
if (s->server && RECORD_LAYER_is_first_record(&s->rlayer)
&& (sslv2len & 0x8000) != 0
&& (type == SSL2_MT_CLIENT_HELLO)) {
/*
* SSLv2 style record
*
* |num_recs| here will actually always be 0 because
* |num_recs > 0| only ever occurs when we are processing
* multiple app data records - which we know isn't the case here
* because it is an SSLv2ClientHello. We keep it using
* |num_recs| for the sake of consistency
*/
thisrr->type = SSL3_RT_HANDSHAKE;
thisrr->rec_version = SSL2_VERSION;
thisrr->length = sslv2len & 0x7fff;
if (thisrr->length > SSL3_BUFFER_get_len(rbuf)
- SSL2_RT_HEADER_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_SSL3_GET_RECORD,
SSL_R_PACKET_LENGTH_TOO_LONG);
return -1;
}
if (thisrr->length < MIN_SSL2_RECORD_LEN) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_SSL3_GET_RECORD,
SSL_R_LENGTH_TOO_SHORT);
return -1;
}
} else {
/* SSLv3+ style record */
if (s->msg_callback)
s->msg_callback(0, 0, SSL3_RT_HEADER, p, 5, s,
s->msg_callback_arg);
/* Pull apart the header into the SSL3_RECORD */
if (!PACKET_get_1(&pkt, &type)
|| !PACKET_get_net_2(&pkt, &version)
|| !PACKET_get_net_2_len(&pkt, &thisrr->length)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_SSL3_GET_RECORD,
ERR_R_INTERNAL_ERROR);
return -1;
}
thisrr->type = type;
thisrr->rec_version = version;
/*
* Lets check version. In TLSv1.3 we ignore this field. For the
* ServerHello after an HRR we haven't actually selected TLSv1.3
* yet, but we still treat it as TLSv1.3, so we must check for
* that explicitly
*/
if (!s->first_packet && !SSL_IS_TLS13(s)
&& !s->hello_retry_request
&& version != (unsigned int)s->version) {
if ((s->version & 0xFF00) == (version & 0xFF00)
&& !s->enc_write_ctx && !s->write_hash) {
if (thisrr->type == SSL3_RT_ALERT) {
/*
* The record is using an incorrect version number,
* but what we've got appears to be an alert. We
* haven't read the body yet to check whether its a
* fatal or not - but chances are it is. We probably
* shouldn't send a fatal alert back. We'll just
* end.
*/
SSLfatal(s, SSL_AD_NO_ALERT, SSL_F_SSL3_GET_RECORD,
SSL_R_WRONG_VERSION_NUMBER);
return -1;
}
/*
* Send back error using their minor version number :-)
*/
s->version = (unsigned short)version;
}
SSLfatal(s, SSL_AD_PROTOCOL_VERSION, SSL_F_SSL3_GET_RECORD,
SSL_R_WRONG_VERSION_NUMBER);
return -1;
}
if ((version >> 8) != SSL3_VERSION_MAJOR) {
if (RECORD_LAYER_is_first_record(&s->rlayer)) {
/* Go back to start of packet, look at the five bytes
* that we have. */
p = RECORD_LAYER_get_packet(&s->rlayer);
if (strncmp((char *)p, "GET ", 4) == 0 ||
strncmp((char *)p, "POST ", 5) == 0 ||
strncmp((char *)p, "HEAD ", 5) == 0 ||
strncmp((char *)p, "PUT ", 4) == 0) {
SSLfatal(s, SSL_AD_NO_ALERT, SSL_F_SSL3_GET_RECORD,
SSL_R_HTTP_REQUEST);
return -1;
} else if (strncmp((char *)p, "CONNE", 5) == 0) {
SSLfatal(s, SSL_AD_NO_ALERT, SSL_F_SSL3_GET_RECORD,
SSL_R_HTTPS_PROXY_REQUEST);
return -1;
}
/* Doesn't look like TLS - don't send an alert */
SSLfatal(s, SSL_AD_NO_ALERT, SSL_F_SSL3_GET_RECORD,
SSL_R_WRONG_VERSION_NUMBER);
return -1;
} else {
SSLfatal(s, SSL_AD_PROTOCOL_VERSION,
SSL_F_SSL3_GET_RECORD,
SSL_R_WRONG_VERSION_NUMBER);
return -1;
}
}
if (SSL_IS_TLS13(s) && s->enc_read_ctx != NULL
&& thisrr->type != SSL3_RT_APPLICATION_DATA) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE,
SSL_F_SSL3_GET_RECORD, SSL_R_BAD_RECORD_TYPE);
return -1;
}
if (thisrr->length >
SSL3_BUFFER_get_len(rbuf) - SSL3_RT_HEADER_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_SSL3_GET_RECORD,
SSL_R_PACKET_LENGTH_TOO_LONG);
return -1;
}
}
/* now s->rlayer.rstate == SSL_ST_READ_BODY */
}
if (SSL_IS_TLS13(s)) {
if (thisrr->length > SSL3_RT_MAX_TLS13_ENCRYPTED_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_SSL3_GET_RECORD,
SSL_R_ENCRYPTED_LENGTH_TOO_LONG);
return -1;
}
} else {
size_t len = SSL3_RT_MAX_ENCRYPTED_LENGTH;
#ifndef OPENSSL_NO_COMP
/*
* If OPENSSL_NO_COMP is defined then SSL3_RT_MAX_ENCRYPTED_LENGTH
* does not include the compression overhead anyway.
*/
if (s->expand == NULL)
len -= SSL3_RT_MAX_COMPRESSED_OVERHEAD;
#endif
if (thisrr->length > len) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_SSL3_GET_RECORD,
SSL_R_ENCRYPTED_LENGTH_TOO_LONG);
return -1;
}
}
/*
* s->rlayer.rstate == SSL_ST_READ_BODY, get and decode the data.
* Calculate how much more data we need to read for the rest of the
* record
*/
if (thisrr->rec_version == SSL2_VERSION) {
more = thisrr->length + SSL2_RT_HEADER_LENGTH
- SSL3_RT_HEADER_LENGTH;
} else {
more = thisrr->length;
}
if (more > 0) {
/* now s->packet_length == SSL3_RT_HEADER_LENGTH */
rret = ssl3_read_n(s, more, more, 1, 0, &n);
if (rret <= 0)
return rret; /* error or non-blocking io */
}
/* set state for later operations */
RECORD_LAYER_set_rstate(&s->rlayer, SSL_ST_READ_HEADER);
/*
* At this point, s->packet_length == SSL3_RT_HEADER_LENGTH
* + thisrr->length, or s->packet_length == SSL2_RT_HEADER_LENGTH
* + thisrr->length and we have that many bytes in s->packet
*/
if (thisrr->rec_version == SSL2_VERSION) {
thisrr->input =
&(RECORD_LAYER_get_packet(&s->rlayer)[SSL2_RT_HEADER_LENGTH]);
} else {
thisrr->input =
&(RECORD_LAYER_get_packet(&s->rlayer)[SSL3_RT_HEADER_LENGTH]);
}
/*
* ok, we can now read from 's->packet' data into 'thisrr' thisrr->input
* points at thisrr->length bytes, which need to be copied into
* thisrr->data by either the decryption or by the decompression When
* the data is 'copied' into the thisrr->data buffer, thisrr->input will
* be pointed at the new buffer
*/
/*
* We now have - encrypted [ MAC [ compressed [ plain ] ] ]
* thisrr->length bytes of encrypted compressed stuff.
*/
/* decrypt in place in 'thisrr->input' */
thisrr->data = thisrr->input;
thisrr->orig_len = thisrr->length;
/* Mark this record as not read by upper layers yet */
thisrr->read = 0;
num_recs++;
/* we have pulled in a full packet so zero things */
RECORD_LAYER_reset_packet_length(&s->rlayer);
RECORD_LAYER_clear_first_record(&s->rlayer);
} while (num_recs < max_recs
&& thisrr->type == SSL3_RT_APPLICATION_DATA
&& SSL_USE_EXPLICIT_IV(s)
&& s->enc_read_ctx != NULL
&& (EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(s->enc_read_ctx))
& EVP_CIPH_FLAG_PIPELINE)
&& ssl3_record_app_data_waiting(s));
/*
* If in encrypt-then-mac mode calculate mac from encrypted record. All
* the details below are public so no timing details can leak.
*/
if (SSL_READ_ETM(s) && s->read_hash) {
unsigned char *mac;
/* TODO(size_t): convert this to do size_t properly */
imac_size = EVP_MD_CTX_size(s->read_hash);
if (!ossl_assert(imac_size >= 0 && imac_size <= EVP_MAX_MD_SIZE)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_SSL3_GET_RECORD,
ERR_LIB_EVP);
return -1;
}
mac_size = (size_t)imac_size;
for (j = 0; j < num_recs; j++) {
thisrr = &rr[j];
if (thisrr->length < mac_size) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_SSL3_GET_RECORD,
SSL_R_LENGTH_TOO_SHORT);
return -1;
}
thisrr->length -= mac_size;
mac = thisrr->data + thisrr->length;
i = s->method->ssl3_enc->mac(s, thisrr, md, 0 /* not send */ );
if (i == 0 || CRYPTO_memcmp(md, mac, mac_size) != 0) {
SSLfatal(s, SSL_AD_BAD_RECORD_MAC, SSL_F_SSL3_GET_RECORD,
SSL_R_DECRYPTION_FAILED_OR_BAD_RECORD_MAC);
return -1;
}
}
}
first_rec_len = rr[0].length;
enc_err = s->method->ssl3_enc->enc(s, rr, num_recs, 0);
/*-
* enc_err is:
* 0: (in non-constant time) if the record is publicly invalid.
* 1: if the padding is valid
* -1: if the padding is invalid
*/
if (enc_err == 0) {
if (ossl_statem_in_error(s)) {
/* SSLfatal() already got called */
return -1;
}
if (num_recs == 1 && ossl_statem_skip_early_data(s)) {
/*
* Valid early_data that we cannot decrypt might fail here as
* publicly invalid. We treat it like an empty record.
*/
thisrr = &rr[0];
if (!early_data_count_ok(s, thisrr->length,
EARLY_DATA_CIPHERTEXT_OVERHEAD, 0)) {
/* SSLfatal() already called */
return -1;
}
thisrr->length = 0;
thisrr->read = 1;
RECORD_LAYER_set_numrpipes(&s->rlayer, 1);
RECORD_LAYER_reset_read_sequence(&s->rlayer);
return 1;
}
SSLfatal(s, SSL_AD_DECRYPTION_FAILED, SSL_F_SSL3_GET_RECORD,
SSL_R_BLOCK_CIPHER_PAD_IS_WRONG);
return -1;
}
#ifdef SSL_DEBUG
printf("dec %"OSSLzu"\n", rr[0].length);
{
size_t z;
for (z = 0; z < rr[0].length; z++)
printf("%02X%c", rr[0].data[z], ((z + 1) % 16) ? ' ' : '\n');
}
printf("\n");
#endif
/* r->length is now the compressed data plus mac */
if ((sess != NULL) &&
(s->enc_read_ctx != NULL) &&
(!SSL_READ_ETM(s) && EVP_MD_CTX_md(s->read_hash) != NULL)) {
/* s->read_hash != NULL => mac_size != -1 */
unsigned char *mac = NULL;
unsigned char mac_tmp[EVP_MAX_MD_SIZE];
mac_size = EVP_MD_CTX_size(s->read_hash);
if (!ossl_assert(mac_size <= EVP_MAX_MD_SIZE)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_SSL3_GET_RECORD,
ERR_R_INTERNAL_ERROR);
return -1;
}
for (j = 0; j < num_recs; j++) {
thisrr = &rr[j];
/*
* orig_len is the length of the record before any padding was
* removed. This is public information, as is the MAC in use,
* therefore we can safely process the record in a different amount
* of time if it's too short to possibly contain a MAC.
*/
if (thisrr->orig_len < mac_size ||
/* CBC records must have a padding length byte too. */
(EVP_CIPHER_CTX_mode(s->enc_read_ctx) == EVP_CIPH_CBC_MODE &&
thisrr->orig_len < mac_size + 1)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_SSL3_GET_RECORD,
SSL_R_LENGTH_TOO_SHORT);
return -1;
}
if (EVP_CIPHER_CTX_mode(s->enc_read_ctx) == EVP_CIPH_CBC_MODE) {
/*
* We update the length so that the TLS header bytes can be
* constructed correctly but we need to extract the MAC in
* constant time from within the record, without leaking the
* contents of the padding bytes.
*/
mac = mac_tmp;
if (!ssl3_cbc_copy_mac(mac_tmp, thisrr, mac_size)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_SSL3_GET_RECORD,
ERR_R_INTERNAL_ERROR);
return -1;
}
thisrr->length -= mac_size;
} else {
/*
* In this case there's no padding, so |rec->orig_len| equals
* |rec->length| and we checked that there's enough bytes for
* |mac_size| above.
*/
thisrr->length -= mac_size;
mac = &thisrr->data[thisrr->length];
}
i = s->method->ssl3_enc->mac(s, thisrr, md, 0 /* not send */ );
if (i == 0 || mac == NULL
|| CRYPTO_memcmp(md, mac, (size_t)mac_size) != 0)
enc_err = -1;
if (thisrr->length > SSL3_RT_MAX_COMPRESSED_LENGTH + mac_size)
enc_err = -1;
}
}
if (enc_err < 0) {
if (ossl_statem_in_error(s)) {
/* We already called SSLfatal() */
return -1;
}
if (num_recs == 1 && ossl_statem_skip_early_data(s)) {
/*
* We assume this is unreadable early_data - we treat it like an
* empty record
*/
/*
* The record length may have been modified by the mac check above
* so we use the previously saved value
*/
if (!early_data_count_ok(s, first_rec_len,
EARLY_DATA_CIPHERTEXT_OVERHEAD, 0)) {
/* SSLfatal() already called */
return -1;
}
thisrr = &rr[0];
thisrr->length = 0;
thisrr->read = 1;
RECORD_LAYER_set_numrpipes(&s->rlayer, 1);
RECORD_LAYER_reset_read_sequence(&s->rlayer);
return 1;
}
/*
* A separate 'decryption_failed' alert was introduced with TLS 1.0,
* SSL 3.0 only has 'bad_record_mac'. But unless a decryption
* failure is directly visible from the ciphertext anyway, we should
* not reveal which kind of error occurred -- this might become
* visible to an attacker (e.g. via a logfile)
*/
SSLfatal(s, SSL_AD_BAD_RECORD_MAC, SSL_F_SSL3_GET_RECORD,
SSL_R_DECRYPTION_FAILED_OR_BAD_RECORD_MAC);
return -1;
}
for (j = 0; j < num_recs; j++) {
thisrr = &rr[j];
/* thisrr->length is now just compressed */
if (s->expand != NULL) {
if (thisrr->length > SSL3_RT_MAX_COMPRESSED_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_SSL3_GET_RECORD,
SSL_R_COMPRESSED_LENGTH_TOO_LONG);
return -1;
}
if (!ssl3_do_uncompress(s, thisrr)) {
SSLfatal(s, SSL_AD_DECOMPRESSION_FAILURE, SSL_F_SSL3_GET_RECORD,
SSL_R_BAD_DECOMPRESSION);
return -1;
}
}
if (SSL_IS_TLS13(s) && s->enc_read_ctx != NULL) {
size_t end;
if (thisrr->length == 0
|| thisrr->type != SSL3_RT_APPLICATION_DATA) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE, SSL_F_SSL3_GET_RECORD,
SSL_R_BAD_RECORD_TYPE);
return -1;
}
/* Strip trailing padding */
for (end = thisrr->length - 1; end > 0 && thisrr->data[end] == 0;
end--)
continue;
thisrr->length = end;
thisrr->type = thisrr->data[end];
if (thisrr->type != SSL3_RT_APPLICATION_DATA
&& thisrr->type != SSL3_RT_ALERT
&& thisrr->type != SSL3_RT_HANDSHAKE) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE, SSL_F_SSL3_GET_RECORD,
SSL_R_BAD_RECORD_TYPE);
return -1;
}
if (s->msg_callback)
s->msg_callback(0, s->version, SSL3_RT_INNER_CONTENT_TYPE,
&thisrr->data[end], 1, s, s->msg_callback_arg);
}
/*
* TLSv1.3 alert and handshake records are required to be non-zero in
* length.
*/
if (SSL_IS_TLS13(s)
&& (thisrr->type == SSL3_RT_HANDSHAKE
|| thisrr->type == SSL3_RT_ALERT)
&& thisrr->length == 0) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE, SSL_F_SSL3_GET_RECORD,
SSL_R_BAD_LENGTH);
return -1;
}
if (thisrr->length > SSL3_RT_MAX_PLAIN_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_SSL3_GET_RECORD,
SSL_R_DATA_LENGTH_TOO_LONG);
return -1;
}
/* If received packet overflows current Max Fragment Length setting */
if (s->session != NULL && USE_MAX_FRAGMENT_LENGTH_EXT(s->session)
&& thisrr->length > GET_MAX_FRAGMENT_LENGTH(s->session)) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_SSL3_GET_RECORD,
SSL_R_DATA_LENGTH_TOO_LONG);
return -1;
}
thisrr->off = 0;
/*-
* So at this point the following is true
* thisrr->type is the type of record
* thisrr->length == number of bytes in record
* thisrr->off == offset to first valid byte
* thisrr->data == where to take bytes from, increment after use :-).
*/
/* just read a 0 length packet */
if (thisrr->length == 0) {
RECORD_LAYER_inc_empty_record_count(&s->rlayer);
if (RECORD_LAYER_get_empty_record_count(&s->rlayer)
> MAX_EMPTY_RECORDS) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE, SSL_F_SSL3_GET_RECORD,
SSL_R_RECORD_TOO_SMALL);
return -1;
}
} else {
RECORD_LAYER_reset_empty_record_count(&s->rlayer);
}
}
if (s->early_data_state == SSL_EARLY_DATA_READING) {
thisrr = &rr[0];
if (thisrr->type == SSL3_RT_APPLICATION_DATA
&& !early_data_count_ok(s, thisrr->length, 0, 0)) {
/* SSLfatal already called */
return -1;
}
}
RECORD_LAYER_set_numrpipes(&s->rlayer, num_recs);
return 1;
}
int ssl3_do_uncompress(SSL *ssl, SSL3_RECORD *rr)
{
#ifndef OPENSSL_NO_COMP
int i;
if (rr->comp == NULL) {
rr->comp = (unsigned char *)
OPENSSL_malloc(SSL3_RT_MAX_ENCRYPTED_LENGTH);
}
if (rr->comp == NULL)
return 0;
/* TODO(size_t): Convert this call */
i = COMP_expand_block(ssl->expand, rr->comp,
SSL3_RT_MAX_PLAIN_LENGTH, rr->data, (int)rr->length);
if (i < 0)
return 0;
else
rr->length = i;
rr->data = rr->comp;
#endif
return 1;
}
int ssl3_do_compress(SSL *ssl, SSL3_RECORD *wr)
{
#ifndef OPENSSL_NO_COMP
int i;
/* TODO(size_t): Convert this call */
i = COMP_compress_block(ssl->compress, wr->data,
(int)(wr->length + SSL3_RT_MAX_COMPRESSED_OVERHEAD),
wr->input, (int)wr->length);
if (i < 0)
return 0;
else
wr->length = i;
wr->input = wr->data;
#endif
return 1;
}
/*-
* ssl3_enc encrypts/decrypts |n_recs| records in |inrecs|. Will call
* SSLfatal() for internal errors, but not otherwise.
*
* Returns:
* 0: (in non-constant time) if the record is publically invalid (i.e. too
* short etc).
* 1: if the record's padding is valid / the encryption was successful.
* -1: if the record's padding is invalid or, if sending, an internal error
* occurred.
*/
int ssl3_enc(SSL *s, SSL3_RECORD *inrecs, size_t n_recs, int sending)
{
SSL3_RECORD *rec;
EVP_CIPHER_CTX *ds;
size_t l, i;
size_t bs, mac_size = 0;
int imac_size;
const EVP_CIPHER *enc;
rec = inrecs;
/*
* We shouldn't ever be called with more than one record in the SSLv3 case
*/
if (n_recs != 1)
return 0;
if (sending) {
ds = s->enc_write_ctx;
if (s->enc_write_ctx == NULL)
enc = NULL;
else
enc = EVP_CIPHER_CTX_cipher(s->enc_write_ctx);
} else {
ds = s->enc_read_ctx;
if (s->enc_read_ctx == NULL)
enc = NULL;
else
enc = EVP_CIPHER_CTX_cipher(s->enc_read_ctx);
}
if ((s->session == NULL) || (ds == NULL) || (enc == NULL)) {
memmove(rec->data, rec->input, rec->length);
rec->input = rec->data;
} else {
l = rec->length;
/* TODO(size_t): Convert this call */
bs = EVP_CIPHER_CTX_block_size(ds);
/* COMPRESS */
if ((bs != 1) && sending) {
i = bs - (l % bs);
/* we need to add 'i-1' padding bytes */
l += i;
/*
* the last of these zero bytes will be overwritten with the
* padding length.
*/
memset(&rec->input[rec->length], 0, i);
rec->length += i;
rec->input[l - 1] = (unsigned char)(i - 1);
}
if (!sending) {
if (l == 0 || l % bs != 0)
return 0;
/* otherwise, rec->length >= bs */
}
/* TODO(size_t): Convert this call */
if (EVP_Cipher(ds, rec->data, rec->input, (unsigned int)l) < 1)
return -1;
if (EVP_MD_CTX_md(s->read_hash) != NULL) {
/* TODO(size_t): convert me */
imac_size = EVP_MD_CTX_size(s->read_hash);
if (imac_size < 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_SSL3_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
}
mac_size = (size_t)imac_size;
}
if ((bs != 1) && !sending)
return ssl3_cbc_remove_padding(rec, bs, mac_size);
}
return 1;
}
#define MAX_PADDING 256
/*-
* tls1_enc encrypts/decrypts |n_recs| in |recs|. Will call SSLfatal() for
* internal errors, but not otherwise.
*
* Returns:
* 0: (in non-constant time) if the record is publically invalid (i.e. too
* short etc).
* 1: if the record's padding is valid / the encryption was successful.
* -1: if the record's padding/AEAD-authenticator is invalid or, if sending,
* an internal error occurred.
*/
int tls1_enc(SSL *s, SSL3_RECORD *recs, size_t n_recs, int sending)
{
EVP_CIPHER_CTX *ds;
size_t reclen[SSL_MAX_PIPELINES];
unsigned char buf[SSL_MAX_PIPELINES][EVP_AEAD_TLS1_AAD_LEN];
int i, pad = 0, ret, tmpr;
size_t bs, mac_size = 0, ctr, padnum, loop;
unsigned char padval;
int imac_size;
const EVP_CIPHER *enc;
if (n_recs == 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return 0;
}
if (sending) {
if (EVP_MD_CTX_md(s->write_hash)) {
int n = EVP_MD_CTX_size(s->write_hash);
if (!ossl_assert(n >= 0)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
}
}
ds = s->enc_write_ctx;
if (s->enc_write_ctx == NULL)
enc = NULL;
else {
int ivlen;
enc = EVP_CIPHER_CTX_cipher(s->enc_write_ctx);
/* For TLSv1.1 and later explicit IV */
if (SSL_USE_EXPLICIT_IV(s)
&& EVP_CIPHER_mode(enc) == EVP_CIPH_CBC_MODE)
ivlen = EVP_CIPHER_iv_length(enc);
else
ivlen = 0;
if (ivlen > 1) {
for (ctr = 0; ctr < n_recs; ctr++) {
if (recs[ctr].data != recs[ctr].input) {
/*
* we can't write into the input stream: Can this ever
* happen?? (steve)
*/
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
} else if (ssl_randbytes(s, recs[ctr].input, ivlen) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
}
}
}
}
} else {
if (EVP_MD_CTX_md(s->read_hash)) {
int n = EVP_MD_CTX_size(s->read_hash);
if (!ossl_assert(n >= 0)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
}
}
ds = s->enc_read_ctx;
if (s->enc_read_ctx == NULL)
enc = NULL;
else
enc = EVP_CIPHER_CTX_cipher(s->enc_read_ctx);
}
if ((s->session == NULL) || (ds == NULL) || (enc == NULL)) {
for (ctr = 0; ctr < n_recs; ctr++) {
memmove(recs[ctr].data, recs[ctr].input, recs[ctr].length);
recs[ctr].input = recs[ctr].data;
}
ret = 1;
} else {
bs = EVP_CIPHER_block_size(EVP_CIPHER_CTX_cipher(ds));
if (n_recs > 1) {
if (!(EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(ds))
& EVP_CIPH_FLAG_PIPELINE)) {
/*
* We shouldn't have been called with pipeline data if the
* cipher doesn't support pipelining
*/
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
SSL_R_PIPELINE_FAILURE);
return -1;
}
}
for (ctr = 0; ctr < n_recs; ctr++) {
reclen[ctr] = recs[ctr].length;
if (EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(ds))
& EVP_CIPH_FLAG_AEAD_CIPHER) {
unsigned char *seq;
seq = sending ? RECORD_LAYER_get_write_sequence(&s->rlayer)
: RECORD_LAYER_get_read_sequence(&s->rlayer);
if (SSL_IS_DTLS(s)) {
/* DTLS does not support pipelining */
unsigned char dtlsseq[9], *p = dtlsseq;
s2n(sending ? DTLS_RECORD_LAYER_get_w_epoch(&s->rlayer) :
DTLS_RECORD_LAYER_get_r_epoch(&s->rlayer), p);
memcpy(p, &seq[2], 6);
memcpy(buf[ctr], dtlsseq, 8);
} else {
memcpy(buf[ctr], seq, 8);
for (i = 7; i >= 0; i--) { /* increment */
++seq[i];
if (seq[i] != 0)
break;
}
}
buf[ctr][8] = recs[ctr].type;
buf[ctr][9] = (unsigned char)(s->version >> 8);
buf[ctr][10] = (unsigned char)(s->version);
buf[ctr][11] = (unsigned char)(recs[ctr].length >> 8);
buf[ctr][12] = (unsigned char)(recs[ctr].length & 0xff);
pad = EVP_CIPHER_CTX_ctrl(ds, EVP_CTRL_AEAD_TLS1_AAD,
EVP_AEAD_TLS1_AAD_LEN, buf[ctr]);
if (pad <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
}
if (sending) {
reclen[ctr] += pad;
recs[ctr].length += pad;
}
} else if ((bs != 1) && sending) {
padnum = bs - (reclen[ctr] % bs);
/* Add weird padding of upto 256 bytes */
if (padnum > MAX_PADDING) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
}
/* we need to add 'padnum' padding bytes of value padval */
padval = (unsigned char)(padnum - 1);
for (loop = reclen[ctr]; loop < reclen[ctr] + padnum; loop++)
recs[ctr].input[loop] = padval;
reclen[ctr] += padnum;
recs[ctr].length += padnum;
}
if (!sending) {
if (reclen[ctr] == 0 || reclen[ctr] % bs != 0)
return 0;
}
}
if (n_recs > 1) {
unsigned char *data[SSL_MAX_PIPELINES];
/* Set the output buffers */
for (ctr = 0; ctr < n_recs; ctr++) {
data[ctr] = recs[ctr].data;
}
if (EVP_CIPHER_CTX_ctrl(ds, EVP_CTRL_SET_PIPELINE_OUTPUT_BUFS,
(int)n_recs, data) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
SSL_R_PIPELINE_FAILURE);
return -1;
}
/* Set the input buffers */
for (ctr = 0; ctr < n_recs; ctr++) {
data[ctr] = recs[ctr].input;
}
if (EVP_CIPHER_CTX_ctrl(ds, EVP_CTRL_SET_PIPELINE_INPUT_BUFS,
(int)n_recs, data) <= 0
|| EVP_CIPHER_CTX_ctrl(ds, EVP_CTRL_SET_PIPELINE_INPUT_LENS,
(int)n_recs, reclen) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
SSL_R_PIPELINE_FAILURE);
return -1;
}
}
/* TODO(size_t): Convert this call */
tmpr = EVP_Cipher(ds, recs[0].data, recs[0].input,
(unsigned int)reclen[0]);
if ((EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(ds))
& EVP_CIPH_FLAG_CUSTOM_CIPHER)
? (tmpr < 0)
: (tmpr == 0))
return -1; /* AEAD can fail to verify MAC */
if (sending == 0) {
if (EVP_CIPHER_mode(enc) == EVP_CIPH_GCM_MODE) {
for (ctr = 0; ctr < n_recs; ctr++) {
recs[ctr].data += EVP_GCM_TLS_EXPLICIT_IV_LEN;
recs[ctr].input += EVP_GCM_TLS_EXPLICIT_IV_LEN;
recs[ctr].length -= EVP_GCM_TLS_EXPLICIT_IV_LEN;
}
} else if (EVP_CIPHER_mode(enc) == EVP_CIPH_CCM_MODE) {
for (ctr = 0; ctr < n_recs; ctr++) {
recs[ctr].data += EVP_CCM_TLS_EXPLICIT_IV_LEN;
recs[ctr].input += EVP_CCM_TLS_EXPLICIT_IV_LEN;
recs[ctr].length -= EVP_CCM_TLS_EXPLICIT_IV_LEN;
}
}
}
ret = 1;
if (!SSL_READ_ETM(s) && EVP_MD_CTX_md(s->read_hash) != NULL) {
imac_size = EVP_MD_CTX_size(s->read_hash);
if (imac_size < 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS1_ENC,
ERR_R_INTERNAL_ERROR);
return -1;
}
mac_size = (size_t)imac_size;
}
if ((bs != 1) && !sending) {
int tmpret;
for (ctr = 0; ctr < n_recs; ctr++) {
tmpret = tls1_cbc_remove_padding(s, &recs[ctr], bs, mac_size);
/*
* If tmpret == 0 then this means publicly invalid so we can
* short circuit things here. Otherwise we must respect constant
* time behaviour.
*/
if (tmpret == 0)
return 0;
ret = constant_time_select_int(constant_time_eq_int(tmpret, 1),
ret, -1);
}
}
if (pad && !sending) {
for (ctr = 0; ctr < n_recs; ctr++) {
recs[ctr].length -= pad;
}
}
}
return ret;
}
int n_ssl3_mac(SSL *ssl, SSL3_RECORD *rec, unsigned char *md, int sending)
{
unsigned char *mac_sec, *seq;
const EVP_MD_CTX *hash;
unsigned char *p, rec_char;
size_t md_size;
size_t npad;
int t;
if (sending) {
mac_sec = &(ssl->s3->write_mac_secret[0]);
seq = RECORD_LAYER_get_write_sequence(&ssl->rlayer);
hash = ssl->write_hash;
} else {
mac_sec = &(ssl->s3->read_mac_secret[0]);
seq = RECORD_LAYER_get_read_sequence(&ssl->rlayer);
hash = ssl->read_hash;
}
t = EVP_MD_CTX_size(hash);
if (t < 0)
return 0;
md_size = t;
npad = (48 / md_size) * md_size;
if (!sending &&
EVP_CIPHER_CTX_mode(ssl->enc_read_ctx) == EVP_CIPH_CBC_MODE &&
ssl3_cbc_record_digest_supported(hash)) {
/*
* This is a CBC-encrypted record. We must avoid leaking any
* timing-side channel information about how many blocks of data we
* are hashing because that gives an attacker a timing-oracle.
*/
/*-
* npad is, at most, 48 bytes and that's with MD5:
* 16 + 48 + 8 (sequence bytes) + 1 + 2 = 75.
*
* With SHA-1 (the largest hash speced for SSLv3) the hash size
* goes up 4, but npad goes down by 8, resulting in a smaller
* total size.
*/
unsigned char header[75];
size_t j = 0;
memcpy(header + j, mac_sec, md_size);
j += md_size;
memcpy(header + j, ssl3_pad_1, npad);
j += npad;
memcpy(header + j, seq, 8);
j += 8;
header[j++] = rec->type;
header[j++] = (unsigned char)(rec->length >> 8);
header[j++] = (unsigned char)(rec->length & 0xff);
/* Final param == is SSLv3 */
if (ssl3_cbc_digest_record(hash,
md, &md_size,
header, rec->input,
rec->length + md_size, rec->orig_len,
mac_sec, md_size, 1) <= 0)
return 0;
} else {
unsigned int md_size_u;
/* Chop the digest off the end :-) */
EVP_MD_CTX *md_ctx = EVP_MD_CTX_new();
if (md_ctx == NULL)
return 0;
rec_char = rec->type;
p = md;
s2n(rec->length, p);
if (EVP_MD_CTX_copy_ex(md_ctx, hash) <= 0
|| EVP_DigestUpdate(md_ctx, mac_sec, md_size) <= 0
|| EVP_DigestUpdate(md_ctx, ssl3_pad_1, npad) <= 0
|| EVP_DigestUpdate(md_ctx, seq, 8) <= 0
|| EVP_DigestUpdate(md_ctx, &rec_char, 1) <= 0
|| EVP_DigestUpdate(md_ctx, md, 2) <= 0
|| EVP_DigestUpdate(md_ctx, rec->input, rec->length) <= 0
|| EVP_DigestFinal_ex(md_ctx, md, NULL) <= 0
|| EVP_MD_CTX_copy_ex(md_ctx, hash) <= 0
|| EVP_DigestUpdate(md_ctx, mac_sec, md_size) <= 0
|| EVP_DigestUpdate(md_ctx, ssl3_pad_2, npad) <= 0
|| EVP_DigestUpdate(md_ctx, md, md_size) <= 0
|| EVP_DigestFinal_ex(md_ctx, md, &md_size_u) <= 0) {
EVP_MD_CTX_reset(md_ctx);
return 0;
}
EVP_MD_CTX_free(md_ctx);
}
ssl3_record_sequence_update(seq);
return 1;
}
int tls1_mac(SSL *ssl, SSL3_RECORD *rec, unsigned char *md, int sending)
{
unsigned char *seq;
EVP_MD_CTX *hash;
size_t md_size;
int i;
EVP_MD_CTX *hmac = NULL, *mac_ctx;
unsigned char header[13];
int stream_mac = (sending ? (ssl->mac_flags & SSL_MAC_FLAG_WRITE_MAC_STREAM)
: (ssl->mac_flags & SSL_MAC_FLAG_READ_MAC_STREAM));
int t;
if (sending) {
seq = RECORD_LAYER_get_write_sequence(&ssl->rlayer);
hash = ssl->write_hash;
} else {
seq = RECORD_LAYER_get_read_sequence(&ssl->rlayer);
hash = ssl->read_hash;
}
t = EVP_MD_CTX_size(hash);
if (!ossl_assert(t >= 0))
return 0;
md_size = t;
/* I should fix this up TLS TLS TLS TLS TLS XXXXXXXX */
if (stream_mac) {
mac_ctx = hash;
} else {
hmac = EVP_MD_CTX_new();
if (hmac == NULL || !EVP_MD_CTX_copy(hmac, hash))
return 0;
mac_ctx = hmac;
}
if (SSL_IS_DTLS(ssl)) {
unsigned char dtlsseq[8], *p = dtlsseq;
s2n(sending ? DTLS_RECORD_LAYER_get_w_epoch(&ssl->rlayer) :
DTLS_RECORD_LAYER_get_r_epoch(&ssl->rlayer), p);
memcpy(p, &seq[2], 6);
memcpy(header, dtlsseq, 8);
} else
memcpy(header, seq, 8);
header[8] = rec->type;
header[9] = (unsigned char)(ssl->version >> 8);
header[10] = (unsigned char)(ssl->version);
header[11] = (unsigned char)(rec->length >> 8);
header[12] = (unsigned char)(rec->length & 0xff);
if (!sending && !SSL_READ_ETM(ssl) &&
EVP_CIPHER_CTX_mode(ssl->enc_read_ctx) == EVP_CIPH_CBC_MODE &&
ssl3_cbc_record_digest_supported(mac_ctx)) {
/*
* This is a CBC-encrypted record. We must avoid leaking any
* timing-side channel information about how many blocks of data we
* are hashing because that gives an attacker a timing-oracle.
*/
/* Final param == not SSLv3 */
if (ssl3_cbc_digest_record(mac_ctx,
md, &md_size,
header, rec->input,
rec->length + md_size, rec->orig_len,
ssl->s3->read_mac_secret,
ssl->s3->read_mac_secret_size, 0) <= 0) {
EVP_MD_CTX_free(hmac);
return 0;
}
} else {
/* TODO(size_t): Convert these calls */
if (EVP_DigestSignUpdate(mac_ctx, header, sizeof(header)) <= 0
|| EVP_DigestSignUpdate(mac_ctx, rec->input, rec->length) <= 0
|| EVP_DigestSignFinal(mac_ctx, md, &md_size) <= 0) {
EVP_MD_CTX_free(hmac);
return 0;
}
}
EVP_MD_CTX_free(hmac);
#ifdef SSL_DEBUG
fprintf(stderr, "seq=");
{
int z;
for (z = 0; z < 8; z++)
fprintf(stderr, "%02X ", seq[z]);
fprintf(stderr, "\n");
}
fprintf(stderr, "rec=");
{
size_t z;
for (z = 0; z < rec->length; z++)
fprintf(stderr, "%02X ", rec->data[z]);
fprintf(stderr, "\n");
}
#endif
if (!SSL_IS_DTLS(ssl)) {
for (i = 7; i >= 0; i--) {
++seq[i];
if (seq[i] != 0)
break;
}
}
#ifdef SSL_DEBUG
{
unsigned int z;
for (z = 0; z < md_size; z++)
fprintf(stderr, "%02X ", md[z]);
fprintf(stderr, "\n");
}
#endif
return 1;
}
/*-
* ssl3_cbc_remove_padding removes padding from the decrypted, SSLv3, CBC
* record in |rec| by updating |rec->length| in constant time.
*
* block_size: the block size of the cipher used to encrypt the record.
* returns:
* 0: (in non-constant time) if the record is publicly invalid.
* 1: if the padding was valid
* -1: otherwise.
*/
int ssl3_cbc_remove_padding(SSL3_RECORD *rec,
size_t block_size, size_t mac_size)
{
size_t padding_length;
size_t good;
const size_t overhead = 1 /* padding length byte */ + mac_size;
/*
* These lengths are all public so we can test them in non-constant time.
*/
if (overhead > rec->length)
return 0;
padding_length = rec->data[rec->length - 1];
good = constant_time_ge_s(rec->length, padding_length + overhead);
/* SSLv3 requires that the padding is minimal. */
good &= constant_time_ge_s(block_size, padding_length + 1);
rec->length -= good & (padding_length + 1);
return constant_time_select_int_s(good, 1, -1);
}
/*-
* tls1_cbc_remove_padding removes the CBC padding from the decrypted, TLS, CBC
* record in |rec| in constant time and returns 1 if the padding is valid and
* -1 otherwise. It also removes any explicit IV from the start of the record
* without leaking any timing about whether there was enough space after the
* padding was removed.
*
* block_size: the block size of the cipher used to encrypt the record.
* returns:
* 0: (in non-constant time) if the record is publicly invalid.
* 1: if the padding was valid
* -1: otherwise.
*/
int tls1_cbc_remove_padding(const SSL *s,
SSL3_RECORD *rec,
size_t block_size, size_t mac_size)
{
size_t good;
size_t padding_length, to_check, i;
const size_t overhead = 1 /* padding length byte */ + mac_size;
/* Check if version requires explicit IV */
if (SSL_USE_EXPLICIT_IV(s)) {
/*
* These lengths are all public so we can test them in non-constant
* time.
*/
if (overhead + block_size > rec->length)
return 0;
/* We can now safely skip explicit IV */
rec->data += block_size;
rec->input += block_size;
rec->length -= block_size;
rec->orig_len -= block_size;
} else if (overhead > rec->length)
return 0;
padding_length = rec->data[rec->length - 1];
if (EVP_CIPHER_flags(EVP_CIPHER_CTX_cipher(s->enc_read_ctx)) &
EVP_CIPH_FLAG_AEAD_CIPHER) {
/* padding is already verified */
rec->length -= padding_length + 1;
return 1;
}
good = constant_time_ge_s(rec->length, overhead + padding_length);
/*
* The padding consists of a length byte at the end of the record and
* then that many bytes of padding, all with the same value as the length
* byte. Thus, with the length byte included, there are i+1 bytes of
* padding. We can't check just |padding_length+1| bytes because that
* leaks decrypted information. Therefore we always have to check the
* maximum amount of padding possible. (Again, the length of the record
* is public information so we can use it.)
*/
to_check = 256; /* maximum amount of padding, inc length byte. */
if (to_check > rec->length)
to_check = rec->length;
for (i = 0; i < to_check; i++) {
unsigned char mask = constant_time_ge_8_s(padding_length, i);
unsigned char b = rec->data[rec->length - 1 - i];
/*
* The final |padding_length+1| bytes should all have the value
* |padding_length|. Therefore the XOR should be zero.
*/
good &= ~(mask & (padding_length ^ b));
}
/*
* If any of the final |padding_length+1| bytes had the wrong value, one
* or more of the lower eight bits of |good| will be cleared.
*/
good = constant_time_eq_s(0xff, good & 0xff);
rec->length -= good & (padding_length + 1);
return constant_time_select_int_s(good, 1, -1);
}
/*-
* ssl3_cbc_copy_mac copies |md_size| bytes from the end of |rec| to |out| in
* constant time (independent of the concrete value of rec->length, which may
* vary within a 256-byte window).
*
* ssl3_cbc_remove_padding or tls1_cbc_remove_padding must be called prior to
* this function.
*
* On entry:
* rec->orig_len >= md_size
* md_size <= EVP_MAX_MD_SIZE
*
* If CBC_MAC_ROTATE_IN_PLACE is defined then the rotation is performed with
* variable accesses in a 64-byte-aligned buffer. Assuming that this fits into
* a single or pair of cache-lines, then the variable memory accesses don't
* actually affect the timing. CPUs with smaller cache-lines [if any] are
* not multi-core and are not considered vulnerable to cache-timing attacks.
*/
#define CBC_MAC_ROTATE_IN_PLACE
int ssl3_cbc_copy_mac(unsigned char *out,
const SSL3_RECORD *rec, size_t md_size)
{
#if defined(CBC_MAC_ROTATE_IN_PLACE)
unsigned char rotated_mac_buf[64 + EVP_MAX_MD_SIZE];
unsigned char *rotated_mac;
#else
unsigned char rotated_mac[EVP_MAX_MD_SIZE];
#endif
/*
* mac_end is the index of |rec->data| just after the end of the MAC.
*/
size_t mac_end = rec->length;
size_t mac_start = mac_end - md_size;
size_t in_mac;
/*
* scan_start contains the number of bytes that we can ignore because the
* MAC's position can only vary by 255 bytes.
*/
size_t scan_start = 0;
size_t i, j;
size_t rotate_offset;
if (!ossl_assert(rec->orig_len >= md_size
&& md_size <= EVP_MAX_MD_SIZE))
return 0;
#if defined(CBC_MAC_ROTATE_IN_PLACE)
rotated_mac = rotated_mac_buf + ((0 - (size_t)rotated_mac_buf) & 63);
#endif
/* This information is public so it's safe to branch based on it. */
if (rec->orig_len > md_size + 255 + 1)
scan_start = rec->orig_len - (md_size + 255 + 1);
in_mac = 0;
rotate_offset = 0;
memset(rotated_mac, 0, md_size);
for (i = scan_start, j = 0; i < rec->orig_len; i++) {
size_t mac_started = constant_time_eq_s(i, mac_start);
size_t mac_ended = constant_time_lt_s(i, mac_end);
unsigned char b = rec->data[i];
in_mac |= mac_started;
in_mac &= mac_ended;
rotate_offset |= j & mac_started;
rotated_mac[j++] |= b & in_mac;
j &= constant_time_lt_s(j, md_size);
}
/* Now rotate the MAC */
#if defined(CBC_MAC_ROTATE_IN_PLACE)
j = 0;
for (i = 0; i < md_size; i++) {
/* in case cache-line is 32 bytes, touch second line */
((volatile unsigned char *)rotated_mac)[rotate_offset ^ 32];
out[j++] = rotated_mac[rotate_offset++];
rotate_offset &= constant_time_lt_s(rotate_offset, md_size);
}
#else
memset(out, 0, md_size);
rotate_offset = md_size - rotate_offset;
rotate_offset &= constant_time_lt_s(rotate_offset, md_size);
for (i = 0; i < md_size; i++) {
for (j = 0; j < md_size; j++)
out[j] |= rotated_mac[i] & constant_time_eq_8_s(j, rotate_offset);
rotate_offset++;
rotate_offset &= constant_time_lt_s(rotate_offset, md_size);
}
#endif
return 1;
}
Fix DTLS replay protection The DTLS implementation provides some protection against replay attacks in accordance with RFC6347 section 4.1.2.6. A sliding "window" of valid record sequence numbers is maintained with the "right" hand edge of the window set to the highest sequence number we have received so far. Records that arrive that are off the "left" hand edge of the window are rejected. Records within the window are checked against a list of records received so far. If we already received it then we also reject the new record. If we have not already received the record, or the sequence number is off the right hand edge of the window then we verify the MAC of the record. If MAC verification fails then we discard the record. Otherwise we mark the record as received. If the sequence number was off the right hand edge of the window, then we slide the window along so that the right hand edge is in line with the newly received sequence number. Records may arrive for future epochs, i.e. a record from after a CCS being sent, can arrive before the CCS does if the packets get re-ordered. As we have not yet received the CCS we are not yet in a position to decrypt or validate the MAC of those records. OpenSSL places those records on an unprocessed records queue. It additionally updates the window immediately, even though we have not yet verified the MAC. This will only occur if currently in a handshake/renegotiation. This could be exploited by an attacker by sending a record for the next epoch (which does not have to decrypt or have a valid MAC), with a very large sequence number. This means the right hand edge of the window is moved very far to the right, and all subsequent legitimate packets are dropped causing a denial of service. A similar effect can be achieved during the initial handshake. In this case there is no MAC key negotiated yet. Therefore an attacker can send a message for the current epoch with a very large sequence number. The code will process the record as normal. If the hanshake message sequence number (as opposed to the record sequence number that we have been talking about so far) is in the future then the injected message is bufferred to be handled later, but the window is still updated. Therefore all subsequent legitimate handshake records are dropped. This aspect is not considered a security issue because there are many ways for an attacker to disrupt the initial handshake and prevent it from completing successfully (e.g. injection of a handshake message will cause the Finished MAC to fail and the handshake to be aborted). This issue comes about as a result of trying to do replay protection, but having no integrity mechanism in place yet. Does it even make sense to have replay protection in epoch 0? That issue isn't addressed here though. This addressed an OCAP Audit issue. CVE-2016-2181 Reviewed-by: Richard Levitte <levitte@openssl.org>
2016-07-01 14:20:33 +00:00
int dtls1_process_record(SSL *s, DTLS1_BITMAP *bitmap)
{
int i;
int enc_err;
SSL_SESSION *sess;
SSL3_RECORD *rr;
int imac_size;
size_t mac_size;
unsigned char md[EVP_MAX_MD_SIZE];
rr = RECORD_LAYER_get_rrec(&s->rlayer);
sess = s->session;
/*
* At this point, s->packet_length == SSL3_RT_HEADER_LNGTH + rr->length,
* and we have that many bytes in s->packet
*/
rr->input = &(RECORD_LAYER_get_packet(&s->rlayer)[DTLS1_RT_HEADER_LENGTH]);
/*
* ok, we can now read from 's->packet' data into 'rr' rr->input points
* at rr->length bytes, which need to be copied into rr->data by either
* the decryption or by the decompression When the data is 'copied' into
* the rr->data buffer, rr->input will be pointed at the new buffer
*/
/*
* We now have - encrypted [ MAC [ compressed [ plain ] ] ] rr->length
* bytes of encrypted compressed stuff.
*/
/* check is not needed I believe */
if (rr->length > SSL3_RT_MAX_ENCRYPTED_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_DTLS1_PROCESS_RECORD,
SSL_R_ENCRYPTED_LENGTH_TOO_LONG);
return 0;
}
/* decrypt in place in 'rr->input' */
rr->data = rr->input;
rr->orig_len = rr->length;
if (SSL_READ_ETM(s) && s->read_hash) {
unsigned char *mac;
mac_size = EVP_MD_CTX_size(s->read_hash);
if (!ossl_assert(mac_size <= EVP_MAX_MD_SIZE)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_DTLS1_PROCESS_RECORD,
ERR_R_INTERNAL_ERROR);
return 0;
}
if (rr->orig_len < mac_size) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_DTLS1_PROCESS_RECORD,
SSL_R_LENGTH_TOO_SHORT);
return 0;
}
rr->length -= mac_size;
mac = rr->data + rr->length;
i = s->method->ssl3_enc->mac(s, rr, md, 0 /* not send */ );
if (i == 0 || CRYPTO_memcmp(md, mac, (size_t)mac_size) != 0) {
SSLfatal(s, SSL_AD_BAD_RECORD_MAC, SSL_F_DTLS1_PROCESS_RECORD,
SSL_R_DECRYPTION_FAILED_OR_BAD_RECORD_MAC);
return 0;
}
}
enc_err = s->method->ssl3_enc->enc(s, rr, 1, 0);
/*-
* enc_err is:
* 0: (in non-constant time) if the record is publically invalid.
* 1: if the padding is valid
* -1: if the padding is invalid
*/
if (enc_err == 0) {
if (ossl_statem_in_error(s)) {
/* SSLfatal() got called */
return 0;
}
/* For DTLS we simply ignore bad packets. */
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
return 0;
}
#ifdef SSL_DEBUG
printf("dec %ld\n", rr->length);
{
size_t z;
for (z = 0; z < rr->length; z++)
printf("%02X%c", rr->data[z], ((z + 1) % 16) ? ' ' : '\n');
}
printf("\n");
#endif
/* r->length is now the compressed data plus mac */
if ((sess != NULL) && !SSL_READ_ETM(s) &&
(s->enc_read_ctx != NULL) && (EVP_MD_CTX_md(s->read_hash) != NULL)) {
/* s->read_hash != NULL => mac_size != -1 */
unsigned char *mac = NULL;
unsigned char mac_tmp[EVP_MAX_MD_SIZE];
/* TODO(size_t): Convert this to do size_t properly */
imac_size = EVP_MD_CTX_size(s->read_hash);
if (imac_size < 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_DTLS1_PROCESS_RECORD,
ERR_LIB_EVP);
return 0;
}
mac_size = (size_t)imac_size;
if (!ossl_assert(mac_size <= EVP_MAX_MD_SIZE)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_DTLS1_PROCESS_RECORD,
ERR_R_INTERNAL_ERROR);
return 0;
}
/*
* orig_len is the length of the record before any padding was
* removed. This is public information, as is the MAC in use,
* therefore we can safely process the record in a different amount
* of time if it's too short to possibly contain a MAC.
*/
if (rr->orig_len < mac_size ||
/* CBC records must have a padding length byte too. */
(EVP_CIPHER_CTX_mode(s->enc_read_ctx) == EVP_CIPH_CBC_MODE &&
rr->orig_len < mac_size + 1)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_DTLS1_PROCESS_RECORD,
SSL_R_LENGTH_TOO_SHORT);
return 0;
}
if (EVP_CIPHER_CTX_mode(s->enc_read_ctx) == EVP_CIPH_CBC_MODE) {
/*
* We update the length so that the TLS header bytes can be
* constructed correctly but we need to extract the MAC in
* constant time from within the record, without leaking the
* contents of the padding bytes.
*/
mac = mac_tmp;
if (!ssl3_cbc_copy_mac(mac_tmp, rr, mac_size)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_DTLS1_PROCESS_RECORD,
ERR_R_INTERNAL_ERROR);
return 0;
}
rr->length -= mac_size;
} else {
/*
* In this case there's no padding, so |rec->orig_len| equals
* |rec->length| and we checked that there's enough bytes for
* |mac_size| above.
*/
rr->length -= mac_size;
mac = &rr->data[rr->length];
}
i = s->method->ssl3_enc->mac(s, rr, md, 0 /* not send */ );
if (i == 0 || mac == NULL
|| CRYPTO_memcmp(md, mac, mac_size) != 0)
enc_err = -1;
if (rr->length > SSL3_RT_MAX_COMPRESSED_LENGTH + mac_size)
enc_err = -1;
}
if (enc_err < 0) {
/* decryption failed, silently discard message */
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
return 0;
}
/* r->length is now just compressed */
if (s->expand != NULL) {
if (rr->length > SSL3_RT_MAX_COMPRESSED_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_DTLS1_PROCESS_RECORD,
SSL_R_COMPRESSED_LENGTH_TOO_LONG);
return 0;
}
if (!ssl3_do_uncompress(s, rr)) {
SSLfatal(s, SSL_AD_DECOMPRESSION_FAILURE,
SSL_F_DTLS1_PROCESS_RECORD, SSL_R_BAD_DECOMPRESSION);
return 0;
}
}
if (rr->length > SSL3_RT_MAX_PLAIN_LENGTH) {
SSLfatal(s, SSL_AD_RECORD_OVERFLOW, SSL_F_DTLS1_PROCESS_RECORD,
SSL_R_DATA_LENGTH_TOO_LONG);
return 0;
}
rr->off = 0;
/*-
* So at this point the following is true
* ssl->s3->rrec.type is the type of record
* ssl->s3->rrec.length == number of bytes in record
* ssl->s3->rrec.off == offset to first valid byte
* ssl->s3->rrec.data == where to take bytes from, increment
* after use :-).
*/
/* we have pulled in a full packet so zero things */
RECORD_LAYER_reset_packet_length(&s->rlayer);
Fix DTLS replay protection The DTLS implementation provides some protection against replay attacks in accordance with RFC6347 section 4.1.2.6. A sliding "window" of valid record sequence numbers is maintained with the "right" hand edge of the window set to the highest sequence number we have received so far. Records that arrive that are off the "left" hand edge of the window are rejected. Records within the window are checked against a list of records received so far. If we already received it then we also reject the new record. If we have not already received the record, or the sequence number is off the right hand edge of the window then we verify the MAC of the record. If MAC verification fails then we discard the record. Otherwise we mark the record as received. If the sequence number was off the right hand edge of the window, then we slide the window along so that the right hand edge is in line with the newly received sequence number. Records may arrive for future epochs, i.e. a record from after a CCS being sent, can arrive before the CCS does if the packets get re-ordered. As we have not yet received the CCS we are not yet in a position to decrypt or validate the MAC of those records. OpenSSL places those records on an unprocessed records queue. It additionally updates the window immediately, even though we have not yet verified the MAC. This will only occur if currently in a handshake/renegotiation. This could be exploited by an attacker by sending a record for the next epoch (which does not have to decrypt or have a valid MAC), with a very large sequence number. This means the right hand edge of the window is moved very far to the right, and all subsequent legitimate packets are dropped causing a denial of service. A similar effect can be achieved during the initial handshake. In this case there is no MAC key negotiated yet. Therefore an attacker can send a message for the current epoch with a very large sequence number. The code will process the record as normal. If the hanshake message sequence number (as opposed to the record sequence number that we have been talking about so far) is in the future then the injected message is bufferred to be handled later, but the window is still updated. Therefore all subsequent legitimate handshake records are dropped. This aspect is not considered a security issue because there are many ways for an attacker to disrupt the initial handshake and prevent it from completing successfully (e.g. injection of a handshake message will cause the Finished MAC to fail and the handshake to be aborted). This issue comes about as a result of trying to do replay protection, but having no integrity mechanism in place yet. Does it even make sense to have replay protection in epoch 0? That issue isn't addressed here though. This addressed an OCAP Audit issue. CVE-2016-2181 Reviewed-by: Richard Levitte <levitte@openssl.org>
2016-07-01 14:20:33 +00:00
/* Mark receipt of record. */
dtls1_record_bitmap_update(s, bitmap);
return 1;
}
/*
* Retrieve a buffered record that belongs to the current epoch, i.e. processed
*/
#define dtls1_get_processed_record(s) \
dtls1_retrieve_buffered_record((s), \
&(DTLS_RECORD_LAYER_get_processed_rcds(&s->rlayer)))
/*-
* Call this to get a new input record.
* It will return <= 0 if more data is needed, normally due to an error
* or non-blocking IO.
* When it finishes, one packet has been decoded and can be found in
* ssl->s3->rrec.type - is the type of record
* ssl->s3->rrec.data, - data
* ssl->s3->rrec.length, - number of bytes
*/
/* used only by dtls1_read_bytes */
int dtls1_get_record(SSL *s)
{
int ssl_major, ssl_minor;
int rret;
size_t more, n;
SSL3_RECORD *rr;
unsigned char *p = NULL;
unsigned short version;
DTLS1_BITMAP *bitmap;
unsigned int is_next_epoch;
rr = RECORD_LAYER_get_rrec(&s->rlayer);
again:
/*
* The epoch may have changed. If so, process all the pending records.
* This is a non-blocking operation.
*/
if (!dtls1_process_buffered_records(s)) {
/* SSLfatal() already called */
return -1;
}
/* if we're renegotiating, then there may be buffered records */
if (dtls1_get_processed_record(s))
return 1;
/* get something from the wire */
/* check if we have the header */
if ((RECORD_LAYER_get_rstate(&s->rlayer) != SSL_ST_READ_BODY) ||
(RECORD_LAYER_get_packet_length(&s->rlayer) < DTLS1_RT_HEADER_LENGTH)) {
rret = ssl3_read_n(s, DTLS1_RT_HEADER_LENGTH,
SSL3_BUFFER_get_len(&s->rlayer.rbuf), 0, 1, &n);
/* read timeout is handled by dtls1_read_bytes */
if (rret <= 0) {
/* SSLfatal() already called if appropriate */
return rret; /* error or non-blocking */
}
/* this packet contained a partial record, dump it */
if (RECORD_LAYER_get_packet_length(&s->rlayer) !=
DTLS1_RT_HEADER_LENGTH) {
RECORD_LAYER_reset_packet_length(&s->rlayer);
goto again;
}
RECORD_LAYER_set_rstate(&s->rlayer, SSL_ST_READ_BODY);
p = RECORD_LAYER_get_packet(&s->rlayer);
if (s->msg_callback)
s->msg_callback(0, 0, SSL3_RT_HEADER, p, DTLS1_RT_HEADER_LENGTH,
s, s->msg_callback_arg);
/* Pull apart the header into the DTLS1_RECORD */
rr->type = *(p++);
ssl_major = *(p++);
ssl_minor = *(p++);
version = (ssl_major << 8) | ssl_minor;
/* sequence number is 64 bits, with top 2 bytes = epoch */
n2s(p, rr->epoch);
memcpy(&(RECORD_LAYER_get_read_sequence(&s->rlayer)[2]), p, 6);
p += 6;
n2s(p, rr->length);
/* Lets check version */
if (!s->first_packet) {
if (version != s->version) {
/* unexpected version, silently discard */
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
goto again;
}
}
if ((version & 0xff00) != (s->version & 0xff00)) {
/* wrong version, silently discard record */
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
goto again;
}
if (rr->length > SSL3_RT_MAX_ENCRYPTED_LENGTH) {
/* record too long, silently discard it */
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
goto again;
}
/* If received packet overflows own-client Max Fragment Length setting */
if (s->session != NULL && USE_MAX_FRAGMENT_LENGTH_EXT(s->session)
&& rr->length > GET_MAX_FRAGMENT_LENGTH(s->session)) {
/* record too long, silently discard it */
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
goto again;
}
/* now s->rlayer.rstate == SSL_ST_READ_BODY */
}
/* s->rlayer.rstate == SSL_ST_READ_BODY, get and decode the data */
if (rr->length >
RECORD_LAYER_get_packet_length(&s->rlayer) - DTLS1_RT_HEADER_LENGTH) {
/* now s->packet_length == DTLS1_RT_HEADER_LENGTH */
more = rr->length;
rret = ssl3_read_n(s, more, more, 1, 1, &n);
/* this packet contained a partial record, dump it */
if (rret <= 0 || n != more) {
if (ossl_statem_in_error(s)) {
/* ssl3_read_n() called SSLfatal() */
return -1;
}
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
goto again;
}
/*
* now n == rr->length, and s->packet_length ==
* DTLS1_RT_HEADER_LENGTH + rr->length
*/
}
/* set state for later operations */
RECORD_LAYER_set_rstate(&s->rlayer, SSL_ST_READ_HEADER);
/* match epochs. NULL means the packet is dropped on the floor */
bitmap = dtls1_get_bitmap(s, rr, &is_next_epoch);
if (bitmap == NULL) {
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer); /* dump this record */
goto again; /* get another record */
}
#ifndef OPENSSL_NO_SCTP
/* Only do replay check if no SCTP bio */
if (!BIO_dgram_is_sctp(SSL_get_rbio(s))) {
#endif
/* Check whether this is a repeat, or aged record. */
Fix DTLS replay protection The DTLS implementation provides some protection against replay attacks in accordance with RFC6347 section 4.1.2.6. A sliding "window" of valid record sequence numbers is maintained with the "right" hand edge of the window set to the highest sequence number we have received so far. Records that arrive that are off the "left" hand edge of the window are rejected. Records within the window are checked against a list of records received so far. If we already received it then we also reject the new record. If we have not already received the record, or the sequence number is off the right hand edge of the window then we verify the MAC of the record. If MAC verification fails then we discard the record. Otherwise we mark the record as received. If the sequence number was off the right hand edge of the window, then we slide the window along so that the right hand edge is in line with the newly received sequence number. Records may arrive for future epochs, i.e. a record from after a CCS being sent, can arrive before the CCS does if the packets get re-ordered. As we have not yet received the CCS we are not yet in a position to decrypt or validate the MAC of those records. OpenSSL places those records on an unprocessed records queue. It additionally updates the window immediately, even though we have not yet verified the MAC. This will only occur if currently in a handshake/renegotiation. This could be exploited by an attacker by sending a record for the next epoch (which does not have to decrypt or have a valid MAC), with a very large sequence number. This means the right hand edge of the window is moved very far to the right, and all subsequent legitimate packets are dropped causing a denial of service. A similar effect can be achieved during the initial handshake. In this case there is no MAC key negotiated yet. Therefore an attacker can send a message for the current epoch with a very large sequence number. The code will process the record as normal. If the hanshake message sequence number (as opposed to the record sequence number that we have been talking about so far) is in the future then the injected message is bufferred to be handled later, but the window is still updated. Therefore all subsequent legitimate handshake records are dropped. This aspect is not considered a security issue because there are many ways for an attacker to disrupt the initial handshake and prevent it from completing successfully (e.g. injection of a handshake message will cause the Finished MAC to fail and the handshake to be aborted). This issue comes about as a result of trying to do replay protection, but having no integrity mechanism in place yet. Does it even make sense to have replay protection in epoch 0? That issue isn't addressed here though. This addressed an OCAP Audit issue. CVE-2016-2181 Reviewed-by: Richard Levitte <levitte@openssl.org>
2016-07-01 14:20:33 +00:00
/*
* TODO: Does it make sense to have replay protection in epoch 0 where
* we have no integrity negotiated yet?
*/
if (!dtls1_record_replay_check(s, bitmap)) {
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer); /* dump this record */
goto again; /* get another record */
}
#ifndef OPENSSL_NO_SCTP
}
#endif
/* just read a 0 length packet */
if (rr->length == 0)
goto again;
/*
* If this record is from the next epoch (either HM or ALERT), and a
* handshake is currently in progress, buffer it since it cannot be
* processed at this time.
*/
if (is_next_epoch) {
if ((SSL_in_init(s) || ossl_statem_get_in_handshake(s))) {
if (dtls1_buffer_record (s,
&(DTLS_RECORD_LAYER_get_unprocessed_rcds(&s->rlayer)),
rr->seq_num) < 0) {
/* SSLfatal() already called */
return -1;
}
}
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer);
goto again;
}
Fix DTLS replay protection The DTLS implementation provides some protection against replay attacks in accordance with RFC6347 section 4.1.2.6. A sliding "window" of valid record sequence numbers is maintained with the "right" hand edge of the window set to the highest sequence number we have received so far. Records that arrive that are off the "left" hand edge of the window are rejected. Records within the window are checked against a list of records received so far. If we already received it then we also reject the new record. If we have not already received the record, or the sequence number is off the right hand edge of the window then we verify the MAC of the record. If MAC verification fails then we discard the record. Otherwise we mark the record as received. If the sequence number was off the right hand edge of the window, then we slide the window along so that the right hand edge is in line with the newly received sequence number. Records may arrive for future epochs, i.e. a record from after a CCS being sent, can arrive before the CCS does if the packets get re-ordered. As we have not yet received the CCS we are not yet in a position to decrypt or validate the MAC of those records. OpenSSL places those records on an unprocessed records queue. It additionally updates the window immediately, even though we have not yet verified the MAC. This will only occur if currently in a handshake/renegotiation. This could be exploited by an attacker by sending a record for the next epoch (which does not have to decrypt or have a valid MAC), with a very large sequence number. This means the right hand edge of the window is moved very far to the right, and all subsequent legitimate packets are dropped causing a denial of service. A similar effect can be achieved during the initial handshake. In this case there is no MAC key negotiated yet. Therefore an attacker can send a message for the current epoch with a very large sequence number. The code will process the record as normal. If the hanshake message sequence number (as opposed to the record sequence number that we have been talking about so far) is in the future then the injected message is bufferred to be handled later, but the window is still updated. Therefore all subsequent legitimate handshake records are dropped. This aspect is not considered a security issue because there are many ways for an attacker to disrupt the initial handshake and prevent it from completing successfully (e.g. injection of a handshake message will cause the Finished MAC to fail and the handshake to be aborted). This issue comes about as a result of trying to do replay protection, but having no integrity mechanism in place yet. Does it even make sense to have replay protection in epoch 0? That issue isn't addressed here though. This addressed an OCAP Audit issue. CVE-2016-2181 Reviewed-by: Richard Levitte <levitte@openssl.org>
2016-07-01 14:20:33 +00:00
if (!dtls1_process_record(s, bitmap)) {
if (ossl_statem_in_error(s)) {
/* dtls1_process_record() called SSLfatal */
return -1;
}
rr->length = 0;
RECORD_LAYER_reset_packet_length(&s->rlayer); /* dump this record */
goto again; /* get another record */
}
return 1;
}