void Field::set_no_copy_byte_string(ByteString value) const
{
Internal::check_type(BYTE_STRING, type());
Internal::set_value_no_copy(
ib(), ib_ftype_bytestr_mutable_in(value.ib())
);
}
Field Field::create_no_copy_byte_string(
MemoryPool pool,
const char* name,
size_t name_length,
ByteString value
)
{
return Internal::create_no_copy(
pool,
name, name_length,
Field::BYTE_STRING,
ib_ftype_bytestr_mutable_in(value.ib())
);
}
ib_status_t ib_field_create_bytestr_alias(
ib_field_t **pf,
ib_mpool_t *mp,
const char *name,
size_t nlen,
uint8_t *val,
size_t vlen
)
{
ib_status_t rc;
ib_bytestr_t *bs;
rc = ib_bytestr_alias_mem(&bs, mp, val, vlen);
if (rc != IB_OK) {
goto failed;
}
rc = ib_field_create_no_copy(
pf, mp, name, nlen,
IB_FTYPE_BYTESTR,
ib_ftype_bytestr_mutable_in(bs)
);
if (rc != IB_OK) {
goto failed;
}
ib_field_util_log_debug("FIELD_CREATE_BYTESTR_ALIAS", (*pf));
return IB_OK;
failed:
/* Make sure everything is cleaned up on failure. */
*pf = NULL;
return rc;
}
static
ib_status_t sqltfn_normalize_pg_tfn(ib_engine_t *ib,
ib_mpool_t *mp,
void *tfn_data,
const ib_field_t *field_in,
const ib_field_t **field_out,
ib_flags_t *pflags)
{
assert(ib != NULL);
assert(mp != NULL);
assert(field_in != NULL);
assert(field_out != NULL);
assert(pflags != NULL);
ib_bytestr_t *bs_in;
ib_bytestr_t *bs_out;
const char *buf_in;
const char *buf_in_start;
size_t buf_in_len;
char *buf_out;
char *buf_out_end;
size_t buf_out_len;
size_t lead_len = 0;
ib_status_t rc;
int ret;
ib_field_t *field_new;
/* Currently only bytestring types are supported.
* Other types will just get passed through. */
if (field_in->type != IB_FTYPE_BYTESTR) {
*field_out = field_in;
return IB_OK;
}
/* Extract the underlying incoming value. */
rc = ib_field_value(field_in, ib_ftype_bytestr_mutable_out(&bs_in));
if (rc != IB_OK) {
return rc;
}
/* Create a buffer for normalization. */
buf_out = buf_out_end = (char *)ib_mpool_alloc(mp, ib_bytestr_length(bs_in));
if (buf_out == NULL) {
return IB_EALLOC;
}
/* As SQL can be injected into a string, the normalization
* needs to start after the first quote character if one
* exists.
*
* First try single quote, then double, then none.
*
* TODO: Handle returning multiple transformations:
* 1) Straight normalization
* 2) Normalization as if with single quotes (starting point
* should be based on straight normalization)
* 3) Normalization as if with double quotes (starting point
* should be based on straight normalization)
*/
buf_in = (const char *)ib_bytestr_const_ptr(bs_in);
buf_in_start = memchr(buf_in, '\'', ib_bytestr_length(bs_in));
if (buf_in_start == NULL) {
buf_in_start = memchr(buf_in, '"', ib_bytestr_length(bs_in));
}
if (buf_in_start == NULL) {
buf_in_start = buf_in;
buf_in_len = ib_bytestr_length(bs_in);
}
else {
++buf_in_start; /* After the quote. */
buf_in_len = ib_bytestr_length(bs_in) - (buf_in_start - buf_in);
}
/* Copy the leading string if one exists. */
if (buf_in_start != buf_in) {
lead_len = buf_in_start - buf_in;
memcpy(buf_out, buf_in, lead_len);
buf_out_end += lead_len;
}
/* Normalize. */
ret = sqltfn_normalize_pg_ex(buf_in_start, buf_in_len,
&buf_out_end, &buf_out_len);
if (ret < 0) {
return IB_EALLOC;
}
else if (ret > 0) {
/* Mark as modified. */
*pflags = IB_TFN_FMODIFIED;
}
/* Create the output field wrapping bs_out. */
buf_out_len += lead_len;
rc = ib_bytestr_alias_mem(&bs_out, mp, (uint8_t *)buf_out, buf_out_len);
if (rc != IB_OK) {
return rc;
}
rc =ib_field_create(&field_new, mp,
field_in->name, field_in->nlen,
IB_FTYPE_BYTESTR,
ib_ftype_bytestr_mutable_in(bs_out));
if (rc == IB_OK) {
*field_out = field_new;
}
return rc;
}
static
ib_status_t sqli_normalize_tfn(ib_mpool_t *mp,
const ib_field_t *field_in,
const ib_field_t **field_out,
void *tfn_data)
{
assert(mp != NULL);
assert(field_in != NULL);
assert(field_out != NULL);
const sqli_pattern_set_t *ps = (const sqli_pattern_set_t *)tfn_data;
sfilter sf;
ib_bytestr_t *bs_in;
ib_bytestr_t *bs_out;
const char *buf_in;
char *buf_in_start;
size_t buf_in_len;
char *buf_out;
char *buf_out_end;
size_t buf_out_len;
size_t lead_len = 0;
char prev_token_type;
ib_field_t *field_new;
ib_status_t rc;
size_t fingerprint_len;
/* Currently only bytestring types are supported.
* Other types will just get passed through. */
if (field_in->type != IB_FTYPE_BYTESTR) {
*field_out = field_in;
return IB_OK;
}
/* Extract the underlying incoming value. */
rc = ib_field_value(field_in, ib_ftype_bytestr_mutable_out(&bs_in));
if (rc != IB_OK) {
return rc;
}
/* Create a buffer big enough (double) to allow for normalization. */
buf_in = (const char *)ib_bytestr_const_ptr(bs_in);
buf_out = buf_out_end = (char *)ib_mpool_calloc(mp, 2, ib_bytestr_length(bs_in));
if (buf_out == NULL) {
return IB_EALLOC;
}
/* TODO: With the latest libinjection, we will need to do something like the
* following, but more robust, instead of just calling is_sqli. This seems
* to be because folding is now called, which removes some tokens.
*/
#if 0
/* As SQL can be injected into a string, the normalization
* needs to start after the first quote character if one
* exists.
*
* First try single quote, then double, then none.
*
* TODO: Handle returning multiple transformations:
* 1) Straight normalization
* 2) Normalization as if with single quotes (starting point
* should be based on straight normalization)
* 3) Normalization as if with double quotes (starting point
* should be based on straight normalization)
*/
buf_in_start = memchr(buf_in, CHAR_SINGLE, ib_bytestr_length(bs_in));
if (buf_in_start == NULL) {
buf_in_start = memchr(buf_in, CHAR_DOUBLE, ib_bytestr_length(bs_in));
}
if (buf_in_start == NULL) {
buf_in_start = (char *)buf_in;
buf_in_len = ib_bytestr_length(bs_in);
}
else {
++buf_in_start; /* After the quote. */
buf_in_len = ib_bytestr_length(bs_in) - (buf_in_start - buf_in);
}
/* Copy the leading string if one exists. */
if (buf_in_start != buf_in) {
lead_len = buf_in_start - buf_in;
memcpy(buf_out, buf_in, lead_len);
buf_out_end += lead_len;
}
#endif
buf_in_start = (char *)buf_in;
buf_in_len = ib_bytestr_length(bs_in);
/* Copy the normalized tokens as a space separated list. Since
* the tokenizer does not backtrack, and the normalized values
* are always equal to or less than the original length, the
* tokens are written back to the beginning of the original
* buffer.
*/
libinjection_sqli_init(&sf,buf_in_start, buf_in_len, FLAG_NONE);
libinjection_sqli_callback(&sf, sqli_lookup_word, (void *)ps);
/* NOTE: We do not care if it is sqli, but just want the tokens. */
libinjection_is_sqli(&sf);
if (strlen(sf.fingerprint) == 0) {
*field_out = field_in;
return IB_OK;
}
buf_out_len = 0;
prev_token_type = 0;
fingerprint_len = strlen(sf.fingerprint);
for (size_t i = 0; i < fingerprint_len; ++i) {
stoken_t current = sf.tokenvec[i];
size_t token_len = strlen(current.val);
/* Add in the space if required. */
if ((buf_out_end != buf_out) &&
(current.type != 'o') &&
(prev_token_type != 'o') &&
(current.type != ',') &&
(*(buf_out_end - 1) != ','))
{
*buf_out_end = ' ';
buf_out_end += 1;
++buf_out_len;
}
/* Copy the token value. */
memcpy(buf_out_end, current.val, token_len);
buf_out_end += token_len;
buf_out_len += token_len;
prev_token_type = current.type;
}
/* Create the output field wrapping bs_out. */
buf_out_len += lead_len;
rc = ib_bytestr_alias_mem(&bs_out, mp, (uint8_t *)buf_out, buf_out_len);
if (rc != IB_OK) {
return rc;
}
rc = ib_field_create(&field_new, mp,
field_in->name, field_in->nlen,
IB_FTYPE_BYTESTR,
ib_ftype_bytestr_mutable_in(bs_out));
if (rc == IB_OK) {
*field_out = field_new;
}
return rc;
}
/**
* Set the matches into the given field name as .0, .1, .2 ... .9.
*
* @param[in] ib The IronBee engine to log to.
* @param[in] tx The transaction to store the values into (tx->dpi).
* @param[in] field_name The field to populate with Regex matches.
* @param[in] ovector The vector of integer pairs of matches from PCRE.
* @param[in] matches The number of matches.
* @param[in] subject The matched-against string data.
*
* @returns IB_OK or IB_EALLOC.
*/
static ib_status_t pcre_set_matches(ib_engine_t *ib,
ib_tx_t *tx,
const char* field_name,
int *ovector,
int matches,
const char *subject)
{
IB_FTRACE_INIT();
/* IronBee status. */
ib_status_t rc;
/* Iterator. */
int i;
/* Length of field_name. */
const int field_name_sz = strlen(field_name);
/* The length of the match. */
size_t match_len;
/* The first character in the match. */
const char* match_start;
/* +3 = '.', [digit], and \0. */
char *full_field_name = malloc(field_name_sz+3);
/* Holder to build an optional debug message in. */
char *debug_msg;
/* Holder for a copy of the field value when creating a new field. */
ib_bytestr_t *field_value;
/* Field holder. */
ib_field_t *ib_field;
/* Ensure the above allocations happened. */
if (full_field_name==NULL) {
IB_FTRACE_RET_STATUS(IB_EALLOC);
}
rc = ensure_field_exists(ib, tx, field_name);
if (rc != IB_OK) {
ib_log_alert_tx(tx, "Could not ensure that field %s was a list.",
field_name);
free(full_field_name);
IB_FTRACE_RET_STATUS(IB_EINVAL);
}
/* We have a match! Now populate TX.0-9 in tx->dpi. */
for (i=0; i<matches; i++)
{
/* Build the field name. Typically TX.0, TX.1 ... TX.9 */
sprintf(full_field_name, "%s.%d", field_name, i);
/* Readability. Mark the start and length of the string. */
match_start = subject+ovector[i*2];
match_len = ovector[i*2+1] - ovector[i*2];
/* If debugging this, copy the string value out and print it to the
* log. This could be dangerous as there could be non-character
* values in the match. */
if (ib_log_get_level(ib) >= 7) {
debug_msg = malloc(match_len+1);
/* Notice: Don't provoke a crash if malloc fails. */
if (debug_msg != NULL) {
memcpy(debug_msg, match_start, match_len);
debug_msg[match_len] = '\0';
ib_log_debug2_tx(tx, "REGEX Setting %s=%s",
full_field_name,
debug_msg);
free(debug_msg);
}
}
ib_data_get(tx->dpi, full_field_name, &ib_field);
if (ib_field == NULL) {
ib_data_add_bytestr(tx->dpi,
full_field_name,
(uint8_t*)subject+ovector[i*2],
match_len,
NULL);
}
else {
ib_bytestr_dup_mem(&field_value,
tx->mp,
(const uint8_t*)match_start,
match_len);
ib_field_setv_no_copy(
ib_field,
ib_ftype_bytestr_mutable_in(field_value)
);
}
}
IB_FTRACE_RET_STATUS(IB_OK);
}
static
ib_status_t sqli_normalize_tfn(ib_engine_t *ib,
ib_mpool_t *mp,
void *tfn_data,
const ib_field_t *field_in,
const ib_field_t **field_out,
ib_flags_t *pflags)
{
assert(ib != NULL);
assert(mp != NULL);
assert(field_in != NULL);
assert(field_out != NULL);
assert(pflags != NULL);
sfilter sf;
ib_bytestr_t *bs_in;
ib_bytestr_t *bs_out;
const char *buf_in;
char *buf_in_start;
size_t buf_in_len;
char *buf_out;
char *buf_out_end;
size_t buf_out_len;
size_t lead_len = 0;
char prev_token_type;
ib_field_t *field_new;
ib_status_t rc;
size_t pat_len;
/* Currently only bytestring types are supported.
* Other types will just get passed through. */
if (field_in->type != IB_FTYPE_BYTESTR) {
*field_out = field_in;
return IB_OK;
}
/* Extract the underlying incoming value. */
rc = ib_field_value(field_in, ib_ftype_bytestr_mutable_out(&bs_in));
if (rc != IB_OK) {
return rc;
}
/* Create a buffer big enough (double) to allow for normalization. */
buf_in = (const char *)ib_bytestr_const_ptr(bs_in);
buf_out = buf_out_end = (char *)ib_mpool_calloc(mp, 2, ib_bytestr_length(bs_in));
if (buf_out == NULL) {
return IB_EALLOC;
}
/* As SQL can be injected into a string, the normalization
* needs to start after the first quote character if one
* exists.
*
* First try single quote, then double, then none.
*
* TODO: Handle returning multiple transformations:
* 1) Straight normalization
* 2) Normalization as if with single quotes (starting point
* should be based on straight normalization)
* 3) Normalization as if with double quotes (starting point
* should be based on straight normalization)
*/
buf_in_start = memchr(buf_in, CHAR_SINGLE, ib_bytestr_length(bs_in));
if (buf_in_start == NULL) {
buf_in_start = memchr(buf_in, CHAR_DOUBLE, ib_bytestr_length(bs_in));
}
if (buf_in_start == NULL) {
buf_in_start = (char *)buf_in;
buf_in_len = ib_bytestr_length(bs_in);
}
else {
++buf_in_start; /* After the quote. */
buf_in_len = ib_bytestr_length(bs_in) - (buf_in_start - buf_in);
}
/* Copy the leading string if one exists. */
if (buf_in_start != buf_in) {
lead_len = buf_in_start - buf_in;
memcpy(buf_out, buf_in, lead_len);
buf_out_end += lead_len;
}
/* Copy the normalized tokens as a space separated list. Since
* the tokenizer does not backtrack, and the normalized values
* are always equal to or less than the original length, the
* tokens are written back to the beginning of the original
* buffer.
*/
libinjection_is_sqli(&sf, buf_in_start, buf_in_len, NULL, NULL);
buf_out_len = 0;
prev_token_type = 0;
pat_len = strlen(sf.pat);
for (size_t i = 0; i < pat_len; ++i) {
stoken_t current = sf.tokenvec[i];
size_t token_len = strlen(current.val);
ib_log_debug2(ib, "SQLi TOKEN: %c \"%s\"", current.type, current.val);
/* Add in the space if required. */
if ((buf_out_end != buf_out) &&
(current.type != 'o') &&
(prev_token_type != 'o') &&
(current.type != ',') &&
(*(buf_out_end - 1) != ','))
{
*buf_out_end = ' ';
buf_out_end += 1;
++buf_out_len;
}
/* Copy the token value. */
memcpy(buf_out_end, current.val, token_len);
buf_out_end += token_len;
buf_out_len += token_len;
prev_token_type = current.type;
}
/* Mark as modified. */
*pflags = IB_TFN_FMODIFIED;
/* Create the output field wrapping bs_out. */
buf_out_len += lead_len;
rc = ib_bytestr_alias_mem(&bs_out, mp, (uint8_t *)buf_out, buf_out_len);
if (rc != IB_OK) {
return rc;
}
rc = ib_field_create(&field_new, mp,
field_in->name, field_in->nlen,
IB_FTYPE_BYTESTR,
ib_ftype_bytestr_mutable_in(bs_out));
if (rc == IB_OK) {
*field_out = field_new;
}
return rc;
}
