548 lines
17 KiB
C
548 lines
17 KiB
C
/*
|
|
* Copyright (c) 2015 The TCPDUMP project
|
|
* All rights reserved.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions
|
|
* are met:
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
* documentation and/or other materials provided with the distribution.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
|
|
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
|
|
* COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
|
|
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
|
|
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
|
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
|
|
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
|
|
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
|
* POSSIBILITY OF SUCH DAMAGE.
|
|
*
|
|
* Initial contribution by Andrew Darqui (andrew.darqui@gmail.com).
|
|
*/
|
|
|
|
#include <sys/cdefs.h>
|
|
#ifndef lint
|
|
__RCSID("$NetBSD: print-resp.c,v 1.4 2017/09/08 14:01:13 christos Exp $");
|
|
#endif
|
|
|
|
/* \summary: REdis Serialization Protocol (RESP) printer */
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
#include "config.h"
|
|
#endif
|
|
|
|
#include <netdissect-stdinc.h>
|
|
#include "netdissect.h"
|
|
#include <limits.h>
|
|
#include <string.h>
|
|
#include <stdlib.h>
|
|
#include <errno.h>
|
|
|
|
#include "extract.h"
|
|
|
|
static const char tstr[] = " [|RESP]";
|
|
|
|
/*
|
|
* For information regarding RESP, see: http://redis.io/topics/protocol
|
|
*/
|
|
|
|
#define RESP_SIMPLE_STRING '+'
|
|
#define RESP_ERROR '-'
|
|
#define RESP_INTEGER ':'
|
|
#define RESP_BULK_STRING '$'
|
|
#define RESP_ARRAY '*'
|
|
|
|
#define resp_print_empty(ndo) ND_PRINT((ndo, " empty"))
|
|
#define resp_print_null(ndo) ND_PRINT((ndo, " null"))
|
|
#define resp_print_length_too_large(ndo) ND_PRINT((ndo, " length too large"))
|
|
#define resp_print_length_negative(ndo) ND_PRINT((ndo, " length negative and not -1"))
|
|
#define resp_print_invalid(ndo) ND_PRINT((ndo, " invalid"))
|
|
|
|
void resp_print(netdissect_options *, const u_char *, u_int);
|
|
static int resp_parse(netdissect_options *, register const u_char *, int);
|
|
static int resp_print_string_error_integer(netdissect_options *, register const u_char *, int);
|
|
static int resp_print_simple_string(netdissect_options *, register const u_char *, int);
|
|
static int resp_print_integer(netdissect_options *, register const u_char *, int);
|
|
static int resp_print_error(netdissect_options *, register const u_char *, int);
|
|
static int resp_print_bulk_string(netdissect_options *, register const u_char *, int);
|
|
static int resp_print_bulk_array(netdissect_options *, register const u_char *, int);
|
|
static int resp_print_inline(netdissect_options *, register const u_char *, int);
|
|
static int resp_get_length(netdissect_options *, register const u_char *, int, const u_char **);
|
|
|
|
#define LCHECK2(_tot_len, _len) \
|
|
{ \
|
|
if (_tot_len < _len) \
|
|
goto trunc; \
|
|
}
|
|
|
|
#define LCHECK(_tot_len) LCHECK2(_tot_len, 1)
|
|
|
|
/*
|
|
* FIND_CRLF:
|
|
* Attempts to move our 'ptr' forward until a \r\n is found,
|
|
* while also making sure we don't exceed the buffer '_len'
|
|
* or go past the end of the captured data.
|
|
* If we exceed or go past the end of the captured data,
|
|
* jump to trunc.
|
|
*/
|
|
#define FIND_CRLF(_ptr, _len) \
|
|
for (;;) { \
|
|
LCHECK2(_len, 2); \
|
|
ND_TCHECK2(*_ptr, 2); \
|
|
if (*_ptr == '\r' && *(_ptr+1) == '\n') \
|
|
break; \
|
|
_ptr++; \
|
|
_len--; \
|
|
}
|
|
|
|
/*
|
|
* CONSUME_CRLF
|
|
* Consume a CRLF that we've just found.
|
|
*/
|
|
#define CONSUME_CRLF(_ptr, _len) \
|
|
_ptr += 2; \
|
|
_len -= 2;
|
|
|
|
/*
|
|
* FIND_CR_OR_LF
|
|
* Attempts to move our '_ptr' forward until a \r or \n is found,
|
|
* while also making sure we don't exceed the buffer '_len'
|
|
* or go past the end of the captured data.
|
|
* If we exceed or go past the end of the captured data,
|
|
* jump to trunc.
|
|
*/
|
|
#define FIND_CR_OR_LF(_ptr, _len) \
|
|
for (;;) { \
|
|
LCHECK(_len); \
|
|
ND_TCHECK(*_ptr); \
|
|
if (*_ptr == '\r' || *_ptr == '\n') \
|
|
break; \
|
|
_ptr++; \
|
|
_len--; \
|
|
}
|
|
|
|
/*
|
|
* CONSUME_CR_OR_LF
|
|
* Consume all consecutive \r and \n bytes.
|
|
* If we exceed '_len' or go past the end of the captured data,
|
|
* jump to trunc.
|
|
*/
|
|
#define CONSUME_CR_OR_LF(_ptr, _len) \
|
|
{ \
|
|
int _found_cr_or_lf = 0; \
|
|
for (;;) { \
|
|
/* \
|
|
* Have we hit the end of data? \
|
|
*/ \
|
|
if (_len == 0 || !ND_TTEST(*_ptr)) { \
|
|
/* \
|
|
* Yes. Have we seen a \r \
|
|
* or \n? \
|
|
*/ \
|
|
if (_found_cr_or_lf) { \
|
|
/* \
|
|
* Yes. Just stop. \
|
|
*/ \
|
|
break; \
|
|
} \
|
|
/* \
|
|
* No. We ran out of packet. \
|
|
*/ \
|
|
goto trunc; \
|
|
} \
|
|
if (*_ptr != '\r' && *_ptr != '\n') \
|
|
break; \
|
|
_found_cr_or_lf = 1; \
|
|
_ptr++; \
|
|
_len--; \
|
|
} \
|
|
}
|
|
|
|
/*
|
|
* SKIP_OPCODE
|
|
* Skip over the opcode character.
|
|
* The opcode has already been fetched, so we know it's there, and don't
|
|
* need to do any checks.
|
|
*/
|
|
#define SKIP_OPCODE(_ptr, _tot_len) \
|
|
_ptr++; \
|
|
_tot_len--;
|
|
|
|
/*
|
|
* GET_LENGTH
|
|
* Get a bulk string or array length.
|
|
*/
|
|
#define GET_LENGTH(_ndo, _tot_len, _ptr, _len) \
|
|
{ \
|
|
const u_char *_endp; \
|
|
_len = resp_get_length(_ndo, _ptr, _tot_len, &_endp); \
|
|
_tot_len -= (_endp - _ptr); \
|
|
_ptr = _endp; \
|
|
}
|
|
|
|
/*
|
|
* TEST_RET_LEN
|
|
* If ret_len is < 0, jump to the trunc tag which returns (-1)
|
|
* and 'bubbles up' to printing tstr. Otherwise, return ret_len.
|
|
*/
|
|
#define TEST_RET_LEN(rl) \
|
|
if (rl < 0) { goto trunc; } else { return rl; }
|
|
|
|
/*
|
|
* TEST_RET_LEN_NORETURN
|
|
* If ret_len is < 0, jump to the trunc tag which returns (-1)
|
|
* and 'bubbles up' to printing tstr. Otherwise, continue onward.
|
|
*/
|
|
#define TEST_RET_LEN_NORETURN(rl) \
|
|
if (rl < 0) { goto trunc; }
|
|
|
|
/*
|
|
* RESP_PRINT_SEGMENT
|
|
* Prints a segment in the form of: ' "<stuff>"\n"
|
|
* Assumes the data has already been verified as present.
|
|
*/
|
|
#define RESP_PRINT_SEGMENT(_ndo, _bp, _len) \
|
|
ND_PRINT((_ndo, " \"")); \
|
|
if (fn_printn(_ndo, _bp, _len, _ndo->ndo_snapend)) \
|
|
goto trunc; \
|
|
fn_print_char(_ndo, '"');
|
|
|
|
void
|
|
resp_print(netdissect_options *ndo, const u_char *bp, u_int length)
|
|
{
|
|
int ret_len = 0, length_cur = length;
|
|
|
|
if(!bp || length <= 0)
|
|
return;
|
|
|
|
ND_PRINT((ndo, ": RESP"));
|
|
while (length_cur > 0) {
|
|
/*
|
|
* This block supports redis pipelining.
|
|
* For example, multiple operations can be pipelined within the same string:
|
|
* "*2\r\n\$4\r\nINCR\r\n\$1\r\nz\r\n*2\r\n\$4\r\nINCR\r\n\$1\r\nz\r\n*2\r\n\$4\r\nINCR\r\n\$1\r\nz\r\n"
|
|
* or
|
|
* "PING\r\nPING\r\nPING\r\n"
|
|
* In order to handle this case, we must try and parse 'bp' until
|
|
* 'length' bytes have been processed or we reach a trunc condition.
|
|
*/
|
|
ret_len = resp_parse(ndo, bp, length_cur);
|
|
TEST_RET_LEN_NORETURN(ret_len);
|
|
bp += ret_len;
|
|
length_cur -= ret_len;
|
|
}
|
|
|
|
return;
|
|
|
|
trunc:
|
|
ND_PRINT((ndo, "%s", tstr));
|
|
}
|
|
|
|
static int
|
|
resp_parse(netdissect_options *ndo, register const u_char *bp, int length)
|
|
{
|
|
u_char op;
|
|
int ret_len;
|
|
|
|
LCHECK2(length, 1);
|
|
ND_TCHECK(*bp);
|
|
op = *bp;
|
|
|
|
/* bp now points to the op, so these routines must skip it */
|
|
switch(op) {
|
|
case RESP_SIMPLE_STRING: ret_len = resp_print_simple_string(ndo, bp, length); break;
|
|
case RESP_INTEGER: ret_len = resp_print_integer(ndo, bp, length); break;
|
|
case RESP_ERROR: ret_len = resp_print_error(ndo, bp, length); break;
|
|
case RESP_BULK_STRING: ret_len = resp_print_bulk_string(ndo, bp, length); break;
|
|
case RESP_ARRAY: ret_len = resp_print_bulk_array(ndo, bp, length); break;
|
|
default: ret_len = resp_print_inline(ndo, bp, length); break;
|
|
}
|
|
|
|
/*
|
|
* This gives up with a "truncated" indicator for all errors,
|
|
* including invalid packet errors; that's what we want, as
|
|
* we have to give up on further parsing in that case.
|
|
*/
|
|
TEST_RET_LEN(ret_len);
|
|
|
|
trunc:
|
|
return (-1);
|
|
}
|
|
|
|
static int
|
|
resp_print_simple_string(netdissect_options *ndo, register const u_char *bp, int length) {
|
|
return resp_print_string_error_integer(ndo, bp, length);
|
|
}
|
|
|
|
static int
|
|
resp_print_integer(netdissect_options *ndo, register const u_char *bp, int length) {
|
|
return resp_print_string_error_integer(ndo, bp, length);
|
|
}
|
|
|
|
static int
|
|
resp_print_error(netdissect_options *ndo, register const u_char *bp, int length) {
|
|
return resp_print_string_error_integer(ndo, bp, length);
|
|
}
|
|
|
|
static int
|
|
resp_print_string_error_integer(netdissect_options *ndo, register const u_char *bp, int length) {
|
|
int length_cur = length, len, ret_len;
|
|
const u_char *bp_ptr;
|
|
|
|
/* bp points to the op; skip it */
|
|
SKIP_OPCODE(bp, length_cur);
|
|
bp_ptr = bp;
|
|
|
|
/*
|
|
* bp now prints past the (+-;) opcode, so it's pointing to the first
|
|
* character of the string (which could be numeric).
|
|
* +OK\r\n
|
|
* -ERR ...\r\n
|
|
* :02912309\r\n
|
|
*
|
|
* Find the \r\n with FIND_CRLF().
|
|
*/
|
|
FIND_CRLF(bp_ptr, length_cur);
|
|
|
|
/*
|
|
* bp_ptr points to the \r\n, so bp_ptr - bp is the length of text
|
|
* preceding the \r\n. That includes the opcode, so don't print
|
|
* that.
|
|
*/
|
|
len = (bp_ptr - bp);
|
|
RESP_PRINT_SEGMENT(ndo, bp, len);
|
|
ret_len = 1 /*<opcode>*/ + len /*<string>*/ + 2 /*<CRLF>*/;
|
|
|
|
TEST_RET_LEN(ret_len);
|
|
|
|
trunc:
|
|
return (-1);
|
|
}
|
|
|
|
static int
|
|
resp_print_bulk_string(netdissect_options *ndo, register const u_char *bp, int length) {
|
|
int length_cur = length, string_len;
|
|
|
|
/* bp points to the op; skip it */
|
|
SKIP_OPCODE(bp, length_cur);
|
|
|
|
/* <length>\r\n */
|
|
GET_LENGTH(ndo, length_cur, bp, string_len);
|
|
|
|
if (string_len >= 0) {
|
|
/* Byte string of length string_len, starting at bp */
|
|
if (string_len == 0)
|
|
resp_print_empty(ndo);
|
|
else {
|
|
LCHECK2(length_cur, string_len);
|
|
ND_TCHECK2(*bp, string_len);
|
|
RESP_PRINT_SEGMENT(ndo, bp, string_len);
|
|
bp += string_len;
|
|
length_cur -= string_len;
|
|
}
|
|
|
|
/*
|
|
* Find the \r\n at the end of the string and skip past it.
|
|
* XXX - report an error if the \r\n isn't immediately after
|
|
* the item?
|
|
*/
|
|
FIND_CRLF(bp, length_cur);
|
|
CONSUME_CRLF(bp, length_cur);
|
|
} else {
|
|
/* null, truncated, or invalid for some reason */
|
|
switch(string_len) {
|
|
case (-1): resp_print_null(ndo); break;
|
|
case (-2): goto trunc;
|
|
case (-3): resp_print_length_too_large(ndo); break;
|
|
case (-4): resp_print_length_negative(ndo); break;
|
|
default: resp_print_invalid(ndo); break;
|
|
}
|
|
}
|
|
|
|
return (length - length_cur);
|
|
|
|
trunc:
|
|
return (-1);
|
|
}
|
|
|
|
static int
|
|
resp_print_bulk_array(netdissect_options *ndo, register const u_char *bp, int length) {
|
|
u_int length_cur = length;
|
|
int array_len, i, ret_len;
|
|
|
|
/* bp points to the op; skip it */
|
|
SKIP_OPCODE(bp, length_cur);
|
|
|
|
/* <array_length>\r\n */
|
|
GET_LENGTH(ndo, length_cur, bp, array_len);
|
|
|
|
if (array_len > 0) {
|
|
/* non empty array */
|
|
for (i = 0; i < array_len; i++) {
|
|
ret_len = resp_parse(ndo, bp, length_cur);
|
|
|
|
TEST_RET_LEN_NORETURN(ret_len);
|
|
|
|
bp += ret_len;
|
|
length_cur -= ret_len;
|
|
}
|
|
} else {
|
|
/* empty, null, truncated, or invalid */
|
|
switch(array_len) {
|
|
case 0: resp_print_empty(ndo); break;
|
|
case (-1): resp_print_null(ndo); break;
|
|
case (-2): goto trunc;
|
|
case (-3): resp_print_length_too_large(ndo); break;
|
|
case (-4): resp_print_length_negative(ndo); break;
|
|
default: resp_print_invalid(ndo); break;
|
|
}
|
|
}
|
|
|
|
return (length - length_cur);
|
|
|
|
trunc:
|
|
return (-1);
|
|
}
|
|
|
|
static int
|
|
resp_print_inline(netdissect_options *ndo, register const u_char *bp, int length) {
|
|
int length_cur = length;
|
|
int len;
|
|
const u_char *bp_ptr;
|
|
|
|
/*
|
|
* Inline commands are simply 'strings' followed by \r or \n or both.
|
|
* Redis will do its best to split/parse these strings.
|
|
* This feature of redis is implemented to support the ability of
|
|
* command parsing from telnet/nc sessions etc.
|
|
*
|
|
* <string><\r||\n||\r\n...>
|
|
*/
|
|
|
|
/*
|
|
* Skip forward past any leading \r, \n, or \r\n.
|
|
*/
|
|
CONSUME_CR_OR_LF(bp, length_cur);
|
|
bp_ptr = bp;
|
|
|
|
/*
|
|
* Scan forward looking for \r or \n.
|
|
*/
|
|
FIND_CR_OR_LF(bp_ptr, length_cur);
|
|
|
|
/*
|
|
* Found it; bp_ptr points to the \r or \n, so bp_ptr - bp is the
|
|
* Length of the line text that preceeds it. Print it.
|
|
*/
|
|
len = (bp_ptr - bp);
|
|
RESP_PRINT_SEGMENT(ndo, bp, len);
|
|
|
|
/*
|
|
* Skip forward past the \r, \n, or \r\n.
|
|
*/
|
|
CONSUME_CR_OR_LF(bp_ptr, length_cur);
|
|
|
|
/*
|
|
* Return the number of bytes we processed.
|
|
*/
|
|
return (length - length_cur);
|
|
|
|
trunc:
|
|
return (-1);
|
|
}
|
|
|
|
static int
|
|
resp_get_length(netdissect_options *ndo, register const u_char *bp, int len, const u_char **endp)
|
|
{
|
|
int result;
|
|
u_char c;
|
|
int saw_digit;
|
|
int neg;
|
|
int too_large;
|
|
|
|
if (len == 0)
|
|
goto trunc;
|
|
ND_TCHECK(*bp);
|
|
too_large = 0;
|
|
neg = 0;
|
|
if (*bp == '-') {
|
|
neg = 1;
|
|
bp++;
|
|
len--;
|
|
}
|
|
result = 0;
|
|
saw_digit = 0;
|
|
|
|
for (;;) {
|
|
if (len == 0)
|
|
goto trunc;
|
|
ND_TCHECK(*bp);
|
|
c = *bp;
|
|
if (!(c >= '0' && c <= '9')) {
|
|
if (!saw_digit) {
|
|
bp++;
|
|
goto invalid;
|
|
}
|
|
break;
|
|
}
|
|
c -= '0';
|
|
if (result > (INT_MAX / 10)) {
|
|
/* This will overflow an int when we multiply it by 10. */
|
|
too_large = 1;
|
|
} else {
|
|
result *= 10;
|
|
if (result == ((INT_MAX / 10) * 10) && c > (INT_MAX % 10)) {
|
|
/* This will overflow an int when we add c */
|
|
too_large = 1;
|
|
} else
|
|
result += c;
|
|
}
|
|
bp++;
|
|
len--;
|
|
saw_digit = 1;
|
|
}
|
|
|
|
/*
|
|
* OK, we found a non-digit character. It should be a \r, followed
|
|
* by a \n.
|
|
*/
|
|
if (*bp != '\r') {
|
|
bp++;
|
|
goto invalid;
|
|
}
|
|
bp++;
|
|
len--;
|
|
if (len == 0)
|
|
goto trunc;
|
|
ND_TCHECK(*bp);
|
|
if (*bp != '\n') {
|
|
bp++;
|
|
goto invalid;
|
|
}
|
|
bp++;
|
|
len--;
|
|
*endp = bp;
|
|
if (neg) {
|
|
/* -1 means "null", anything else is invalid */
|
|
if (too_large || result != 1)
|
|
return (-4);
|
|
result = -1;
|
|
}
|
|
return (too_large ? -3 : result);
|
|
|
|
trunc:
|
|
*endp = bp;
|
|
return (-2);
|
|
|
|
invalid:
|
|
*endp = bp;
|
|
return (-5);
|
|
}
|