/*************************************************************************** * osscan.cc -- Routines used for OS detection via TCP/IP fingerprinting. * * For more information on how this works in Nmap, see my paper at * * http://nmap.org/osdetect/ * * * ***********************IMPORTANT NMAP LICENSE TERMS************************ * * * The Nmap Security Scanner is (C) 1996-2008 Insecure.Com LLC. Nmap is * * also a registered trademark of Insecure.Com LLC. This program is free * * software; you may redistribute and/or modify it under the terms of the * * GNU General Public License as published by the Free Software * * Foundation; Version 2 with the clarifications and exceptions described * * below. This guarantees your right to use, modify, and redistribute * * this software under certain conditions. If you wish to embed Nmap * * technology into proprietary software, we sell alternative licenses * * (contact sales@insecure.com). Dozens of software vendors already * * license Nmap technology such as host discovery, port scanning, OS * * detection, and version detection. * * * * Note that the GPL places important restrictions on "derived works", yet * * it does not provide a detailed definition of that term. To avoid * * misunderstandings, we consider an application to constitute a * * "derivative work" for the purpose of this license if it does any of the * * following: * * o Integrates source code from Nmap * * o Reads or includes Nmap copyrighted data files, such as * * nmap-os-db or nmap-service-probes. * * o Executes Nmap and parses the results (as opposed to typical shell or * * execution-menu apps, which simply display raw Nmap output and so are * * not derivative works.) * * o Integrates/includes/aggregates Nmap into a proprietary executable * * installer, such as those produced by InstallShield. * * o Links to a library or executes a program that does any of the above * * * * The term "Nmap" should be taken to also include any portions or derived * * works of Nmap. This list is not exclusive, but is just meant to * * clarify our interpretation of derived works with some common examples. * * These restrictions only apply when you actually redistribute Nmap. For * * example, nothing stops you from writing and selling a proprietary * * front-end to Nmap. Just distribute it by itself, and point people to * * http://nmap.org to download Nmap. * * * * We don't consider these to be added restrictions on top of the GPL, but * * just a clarification of how we interpret "derived works" as it applies * * to our GPL-licensed Nmap product. This is similar to the way Linus * * Torvalds has announced his interpretation of how "derived works" * * applies to Linux kernel modules. Our interpretation refers only to * * Nmap - we don't speak for any other GPL products. * * * * If you have any questions about the GPL licensing restrictions on using * * Nmap in non-GPL works, we would be happy to help. As mentioned above, * * we also offer alternative license to integrate Nmap into proprietary * * applications and appliances. These contracts have been sold to dozens * * of software vendors, and generally include a perpetual license as well * * as providing for priority support and updates as well as helping to * * fund the continued development of Nmap technology. Please email * * sales@insecure.com for further information. * * * * As a special exception to the GPL terms, Insecure.Com LLC grants * * permission to link the code of this program with any version of the * * OpenSSL library which is distributed under a license identical to that * * listed in the included COPYING.OpenSSL file, and distribute linked * * combinations including the two. You must obey the GNU GPL in all * * respects for all of the code used other than OpenSSL. If you modify * * this file, you may extend this exception to your version of the file, * * but you are not obligated to do so. * * * * If you received these files with a written license agreement or * * contract stating terms other than the terms above, then that * * alternative license agreement takes precedence over these comments. * * * * Source is provided to this software because we believe users have a * * right to know exactly what a program is going to do before they run it. * * This also allows you to audit the software for security holes (none * * have been found so far). * * * * Source code also allows you to port Nmap to new platforms, fix bugs, * * and add new features. You are highly encouraged to send your changes * * to fyodor@insecure.org for possible incorporation into the main * * distribution. By sending these changes to Fyodor or one of the * * Insecure.Org development mailing lists, it is assumed that you are * * offering Fyodor and Insecure.Com LLC the unlimited, non-exclusive right * * to reuse, modify, and relicense the code. Nmap will always be * * available Open Source, but this is important because the inability to * * relicense code has caused devastating problems for other Free Software * * projects (such as KDE and NASM). We also occasionally relicense the * * code to third parties as discussed above. If you wish to specify * * special license conditions of your contributions, just say so when you * * send them. * * * * This program is distributed in the hope that it will be useful, but * * WITHOUT ANY WARRANTY; without even the implied warranty of * * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * * General Public License v2.0 for more details at * * http://www.gnu.org/licenses/gpl-2.0.html , or in the COPYING file * * included with Nmap. * * * ***************************************************************************/ /* $Id$ */ #include "osscan.h" #include "timing.h" #include "NmapOps.h" #include "nmap_tty.h" #include "charpool.h" #include "Target.h" #include "nmap_error.h" #include "utils.h" #if TIME_WITH_SYS_TIME # include # include #else # if HAVE_SYS_TIME_H # include # else # include # endif #endif #include extern NmapOps o; // Prints a note if observedFP has a classification and it is not in referenceFP // Returns 0 if they match, nonzero otherwise static int compareclassifications(FingerPrint *referenceFP, FingerPrint *observedFP, bool verbose) { int refclassno; struct OS_Classification *obclass, *refclass; if (observedFP->num_OS_Classifications > 0) { obclass = &(observedFP->OS_class[0]); for(refclassno = 0; refclassno < referenceFP->num_OS_Classifications; refclassno++) { refclass = &(referenceFP->OS_class[refclassno]); if (strcmp(obclass->OS_Vendor, refclass->OS_Vendor) == 0 && strcmp(obclass->OS_Family, refclass->OS_Family) == 0 && strcmp(obclass->Device_Type, refclass->Device_Type) == 0 && (obclass->OS_Generation == refclass->OS_Generation || (obclass->OS_Generation != NULL && refclass->OS_Generation != NULL && strcmp(obclass->OS_Generation, refclass->OS_Generation) == 0))) { // A match! lets get out of here return 0; } } } else { if (verbose) log_write(LOG_PLAIN, "Observed fingerprint lacks a classification\n"); return 1; } if (verbose) log_write(LOG_PLAIN, "[WARN] Classification of observed fingerprint does not appear in reference fingerprint.\n"); return 1; } static struct AVal *getattrbyname(struct AVal *AV, const char *name) { if (!AV) return NULL; do { if (!strcmp(AV->attribute, name)) return AV; AV = AV->next; } while(AV); return NULL; } static struct AVal *gettestbyname(FingerPrint *FP, const char *name) { if (!FP) return NULL; do { if (!strcmp(FP->name, name)) return FP->results; FP = FP->next; } while(FP); return NULL; } /* Returns true if perfect match -- if num_subtests & num_subtests_succeeded are non_null it ADDS THE NEW VALUES to what is already there. So initialize them to zero first if you only want to see the results from this match. if shortcircuit is zero, it does all the tests, otherwise it returns when the first one fails. If you want details of the match process printed, pass n onzero for 'verbose'. If points is non-null, it is examined to find the number of points for each test in the fprint AVal and use that the increment num_subtests and num_subtests_succeeded appropriately. If it is NULL, each test is worth 1 point. In that case, you may also pass in the group name (SEQ, T1, etc) to have that extra info printed. If you pass 0 for verbose, you might as well pass NULL for testGroupName as it won't be used. */ static int AVal_match(struct AVal *reference, struct AVal *fprint, struct AVal *points, unsigned long *num_subtests, unsigned long *num_subtests_succeeded, int shortcut, int verbose, const char *testGroupName) { struct AVal *current_ref; struct AVal *current_fp; struct AVal *current_points; unsigned int number, number1; unsigned int val; char *p, *q, *q1; /* OHHHH YEEEAAAAAHHHH!#!@#$!% */ char valcpy[512]; char *endptr; int andexp, orexp, expchar, numtrue; int testfailed; int subtests = 0, subtests_succeeded=0; int pointsThisTest = 1; for(current_ref = reference; current_ref; current_ref = current_ref->next) { current_fp = getattrbyname(fprint, current_ref->attribute); if (!current_fp) continue; /* OK, we compare an attribute value in current_fp->value to a potentially large expression in current_ref->value. The syntax uses < (less than), > (greather than), + (non-zero), | (or), - (range), and & (and). No parenthesis are allowed */ numtrue = andexp = orexp = 0; testfailed = 0; Strncpy(valcpy, current_ref->value, sizeof(valcpy)); p = valcpy; if (strchr(current_ref->value, '|')) { orexp = 1; expchar = '|'; } else { andexp = 1; expchar = '&'; } do { q = strchr(p, expchar); if (q) *q = '\0'; if (strcmp(p, "+") == 0) { if (!*current_fp->value) { if (andexp) { testfailed=1; break; } } else { val = strtol(current_fp->value, &endptr, 16); if (val == 0 || *endptr) { if (andexp) { testfailed=1; break; } } else { numtrue++; if (orexp) break; } } } else if (*p == '<' && isxdigit((int) p[1])) { if (!*current_fp->value) { if (andexp) { testfailed=1; break; } } number = strtol(p + 1, &endptr, 16); val = strtol(current_fp->value, &endptr, 16); if (val >= number || *endptr) { if (andexp) { testfailed=1; break; } } else { numtrue++; if (orexp) break; } } else if (*p == '>' && isxdigit((int) p[1])) { if (!*current_fp->value) { if (andexp) { testfailed=1; break; } } number = strtol(p + 1, &endptr, 16); val = strtol(current_fp->value, &endptr, 16); if (val <= number || *endptr) { if (andexp) { testfailed=1; break; } } else { numtrue++; if (orexp) break; } } else if (((q1 = strchr(p, '-')) != NULL) && isxdigit((int) p[0]) && isxdigit((int) q1[1])) { if (!*current_fp->value) { if (andexp) { testfailed=1; break; } } *q1 = '\0'; number = strtol(p, NULL, 16); number1 = strtol(q1 + 1, NULL, 16); if(number1 < number && o.debugging) { error("Range error in reference aval: %s=%s", current_ref->attribute, current_ref->value); } val = strtol(current_fp->value, &endptr, 16); if (val < number || val > number1 || *endptr) { if (andexp) { testfailed=1; break; } } else { numtrue++; if (orexp) break; } } else { if (strcmp(p, current_fp->value)) { if (andexp) { testfailed=1; break; } } else { numtrue++; if (orexp) break; } } if (q) p = q + 1; } while(q); if (numtrue == 0) testfailed=1; if (points) { current_points = getattrbyname(points, current_ref->attribute); if (!current_points) fatal("%s: Failed to find point amount for test %s.%s", __func__, testGroupName? testGroupName : "", current_ref->attribute); pointsThisTest = strtol(current_points->value, &endptr, 10); if (pointsThisTest < 1) fatal("%s: Got bogus point amount (%s) for test %s.%s", __func__, current_points->value, testGroupName? testGroupName : "", current_ref->attribute); } subtests += pointsThisTest; if (testfailed) { if (shortcut) { if (num_subtests) *num_subtests += subtests; return 0; } if (verbose) log_write(LOG_PLAIN, "%s.%s: \"%s\" NOMATCH \"%s\" (%d %s)\n", testGroupName, current_ref->attribute, current_fp->value, current_ref->value, pointsThisTest, (pointsThisTest == 1)? "point" : "points"); } else subtests_succeeded += pointsThisTest; /* Whew, we made it past one Attribute alive , on to the next! */ } if (num_subtests) *num_subtests += subtests; if (num_subtests_succeeded) *num_subtests_succeeded += subtests_succeeded; return (subtests == subtests_succeeded)? 1 : 0; } /* Compares 2 fingerprints -- a referenceFP (can have expression attributes) with an observed fingerprint (no expressions). If verbose is nonzero, differences will be printed. The comparison accuracy (between 0 and 1) is returned). If MatchPoints is not NULL, it is a special "fingerprints" which tells how many points each test is worth. */ double compare_fingerprints(FingerPrint *referenceFP, FingerPrint *observedFP, FingerPrint *MatchPoints, int verbose) { FingerPrint *currentReferenceTest; struct AVal *currentObservedTest; struct AVal *currentTestMatchPoints; unsigned long num_subtests = 0, num_subtests_succeeded = 0; unsigned long new_subtests, new_subtests_succeeded; assert(referenceFP); assert(observedFP); if (verbose) compareclassifications(referenceFP, observedFP, true); for(currentReferenceTest = referenceFP; currentReferenceTest; currentReferenceTest = currentReferenceTest->next) { currentObservedTest = gettestbyname(observedFP, currentReferenceTest->name); if (currentObservedTest) { new_subtests = new_subtests_succeeded = 0; if (MatchPoints) { currentTestMatchPoints = gettestbyname(MatchPoints, currentReferenceTest->name); if (!currentTestMatchPoints) fatal("%s: Failed to locate test %s in MatchPoints directive of fingerprint file", __func__, currentReferenceTest->name); } else currentTestMatchPoints = NULL; AVal_match(currentReferenceTest->results, currentObservedTest, currentTestMatchPoints, &new_subtests, &new_subtests_succeeded, 0, verbose, currentReferenceTest->name); num_subtests += new_subtests; num_subtests_succeeded += new_subtests_succeeded; } } assert(num_subtests_succeeded <= num_subtests); return (num_subtests)? (num_subtests_succeeded / (double) num_subtests) : 0; } /* Takes a fingerprint and looks for matches inside the passed in reference fingerprint DB. The results are stored in in FPR (which must point to an instantiated FingerPrintResults class) -- results will be reverse-sorted by accuracy. No results below accuracy_threshhold will be included. The max matches returned is the maximum that fits in a FingerPrintResults class. */ void match_fingerprint(FingerPrint *FP, FingerPrintResults *FPR, FingerPrintDB *DB, double accuracy_threshold) { int i; double FPR_entrance_requirement = accuracy_threshold; /* accuracy must be at least this big to be added to the list */ FingerPrint **reference_FPs = DB->prints; FingerPrint *current_os; double acc; int state; int skipfp; int max_prints = sizeof(FPR->prints) / sizeof(FingerPrint *); int idx; double tmp_acc=0.0, tmp_acc2; /* These are temp buffers for list swaps */ FingerPrint *tmp_FP=NULL, *tmp_FP2; assert(FP); assert(FPR); assert(accuracy_threshold >= 0 && accuracy_threshold <= 1); FPR->overall_results = OSSCAN_SUCCESS; for(i = 0; reference_FPs[i]; i++) { current_os = reference_FPs[i]; skipfp = 0; acc = compare_fingerprints(current_os, FP, DB->MatchPoints, 0); /* error("Comp to %s: %li/%li=%f", o.reference_FPs1[i]->OS_name, num_subtests_succeeded, num_subtests, acc); */ if (acc >= FPR_entrance_requirement || acc == 1.0) { state = 0; for(idx=0; idx < FPR->num_matches; idx++) { if (strcmp(FPR->prints[idx]->OS_name, current_os->OS_name) == 0) { if (FPR->accuracy[idx] >= acc) { skipfp = 1; /* Skip it -- a higher version is already in list */ } else { /* We must shift the list left to delete this sucker */ memmove(FPR->prints + idx, FPR->prints + idx + 1, (FPR->num_matches - 1 - idx) * sizeof(FingerPrint *)); memmove(FPR->accuracy + idx, FPR->accuracy + idx + 1, (FPR->num_matches - 1 - idx) * sizeof(double)); FPR->num_matches--; FPR->accuracy[FPR->num_matches] = 0; } break; /* There can only be 1 in the list with same name */ } } if (!skipfp) { /* First we check whether we have overflowed with perfect matches */ if (acc == 1) { /* error("DEBUG: Perfect match #%d/%d", FPR->num_perfect_matches + 1, max_prints); */ if (FPR->num_perfect_matches == max_prints) { FPR->overall_results = OSSCAN_TOOMANYMATCHES; return; } FPR->num_perfect_matches++; } /* Now we add the sucker to the list */ state = 0; /* Have not yet done the insertion */ for(idx=-1; idx < max_prints -1; idx++) { if (state == 1) { /* Push tmp_acc and tmp_FP onto the next idx */ tmp_acc2 = FPR->accuracy[idx+1]; tmp_FP2 = FPR->prints[idx+1]; FPR->accuracy[idx+1] = tmp_acc; FPR->prints[idx+1] = tmp_FP; tmp_acc = tmp_acc2; tmp_FP = tmp_FP2; } else if (FPR->accuracy[idx + 1] < acc) { /* OK, I insert the sucker into the next slot ... */ tmp_acc = FPR->accuracy[idx+1]; tmp_FP = FPR->prints[idx+1]; FPR->prints[idx+1] = current_os; FPR->accuracy[idx+1] = acc; state = 1; } } if (state != 1) { fatal("Bogus list insertion state (%d) -- num_matches = %d num_perfect_matches=%d entrance_requirement=%f", state, FPR->num_matches, FPR->num_perfect_matches, FPR_entrance_requirement); } FPR->num_matches++; /* If we are over max_prints, one was shoved off list */ if (FPR->num_matches > max_prints) FPR->num_matches = max_prints; /* Calculate the new min req. */ if (FPR->num_matches == max_prints) { FPR_entrance_requirement = FPR->accuracy[max_prints - 1] + 0.00001; } } } } if (FPR->num_matches == 0 && FPR->overall_results == OSSCAN_SUCCESS) FPR->overall_results = OSSCAN_NOMATCHES; return; } void freeFingerPrint(FingerPrint *FP) { FingerPrint *currentFP; FingerPrint *nextFP; if (!FP) return; for(currentFP = FP; currentFP; currentFP = nextFP) { nextFP = currentFP->next; if (currentFP->results) free(currentFP->results); free(currentFP); } return; } /* Writes an informational "Test" result suitable for including at the top of a fingerprint. Gives info which might be useful when the FPrint is submitted (eg Nmap version, etc). Result is written (up to ostrlen) to the ostr var passed in */ static void WriteSInfo(char *ostr, int ostrlen, bool isGoodFP, const struct in_addr * const addr, int distance, const u8 *mac, int openTcpPort, int closedTcpPort, int closedUdpPort) { struct tm *ltime; time_t timep; char dsbuf[8], otbuf[8], ctbuf[8], cubuf[8]; char macbuf[16]; timep = time(NULL); ltime = localtime(&timep); otbuf[0] = '\0'; if(openTcpPort != -1) Snprintf(otbuf, sizeof(otbuf), "%d", openTcpPort); ctbuf[0] = '\0'; if(closedTcpPort != -1) Snprintf(ctbuf, sizeof(ctbuf), "%d", closedTcpPort); cubuf[0] = '\0'; if(closedUdpPort != -1) Snprintf(cubuf, sizeof(cubuf), "%d", closedUdpPort); dsbuf[0] = '\0'; if(distance != -1) { Snprintf(dsbuf, sizeof(dsbuf), "%%DS=%d", distance); } macbuf[0] = '\0'; if (mac) Snprintf(macbuf, sizeof(macbuf), "%%M=%02X%02X%02X", mac[0], mac[1], mac[2]); Snprintf(ostr, ostrlen, "SCAN(V=%s%%D=%d/%d%%OT=%s%%CT=%s%%CU=%s%%PV=%c%s%%G=%c%s%%TM=%X%%P=%s)", NMAP_VERSION, ltime->tm_mon + 1, ltime->tm_mday, otbuf, ctbuf, cubuf, isipprivate(addr)?'Y':'N', dsbuf, isGoodFP?'Y':'N', macbuf, (int) timep, NMAP_PLATFORM); } /* Puts a textual representation of the chain of AVals beginning with AV in s. No more than n bytes will be written. Unless n is 0, the string is always null-terminated. Returns the number of bytes written, excluding the terminator. */ static int AVal2str(const struct AVal *AV, char *s, size_t n) { char *p; char *end; size_t len; if (AV == NULL) { if (n > 0) *s = '\0'; return 0; } p = s; end = s + n - 1; for ( ; AV != NULL; AV = AV->next) { if (p >= end) break; /* Put a separator in front of every attribute-value pair but the first. */ if (p != s) *p++ = '%'; len = MIN((ptrdiff_t) strlen(AV->attribute), end - p); memcpy(p, AV->attribute, len); p += len; if (p >= end) break; *p++ = '='; len = MIN((ptrdiff_t) strlen(AV->value), end - p); memcpy(p, AV->value, len); p += len; } *p = '\0'; return p - s; } static struct AVal *str2AVal(char *str) { int i = 1; int count = 1; char *q = str, *p=str; struct AVal *AVs; if (!*str) return NULL; /* count the AVals */ while((q = strchr(q, '%'))) { count++; q++; } AVs = (struct AVal *) safe_zalloc(count * sizeof(struct AVal)); for(i=0; i < count; i++) { q = strchr(p, '='); if (!q) { fatal("Parse error with AVal string (%s) in nmap-os-db file", str); } *q = '\0'; AVs[i].attribute = strdup(p); p = q+1; if (i != count - 1) { q = strchr(p, '%'); if (!q) { fatal("Parse error with AVal string (%s) in nmap-os-db file", str); } *q = '\0'; AVs[i].next = &AVs[i+1]; } Strncpy(AVs[i].value, p, sizeof(AVs[i].value)); p = q + 1; } return AVs; } /* Compare two AVal chains literally, without evaluating the value of either one as an expression. This is used by mergeFPs. Unlike with AVal_match, it is always the case that AVal_match_literal(a, b) == AVal_match_literal(b, a). */ static bool AVal_match_literal(struct AVal *a, struct AVal *b) { struct AVal *av_a, *av_b; /* Check that b contains all the AVals in a, with the same values. */ for (av_a = a; av_a != NULL; av_a = av_a->next) { av_b = getattrbyname(b, av_a->attribute); if (av_b == NULL || strcmp(av_a->value, av_b->value) != 0) return false; } /* Check that a contains all the AVals in b, with the same values. */ for (av_b = a; av_b != NULL; av_b = av_b->next) { av_a = getattrbyname(a, av_b->attribute); if (av_a == NULL || strcmp(av_a->value, av_b->value) != 0) return false; } return true; } /* This is a less-than relation predicate that establishes the preferred order of tests when they are displayed. Returns true if and only if the test a should come before the test b. */ static bool FingerTest_lessthan(const FingerTest* a, const FingerTest* b) { /* This defines the order in which test lines should appear. */ const char *TEST_ORDER[] = { "SEQ", "OPS", "WIN", "ECN", "T1", "T2", "T3", "T4", "T5", "T6", "T7", "U1", "IE" }; unsigned int i; int ia, ib; /* The indices at which the test names were found in the list. -1 means "not found." */ ia = -1; ib = -1; /* Look up the test names in the list. */ for (i = 0; i < sizeof(TEST_ORDER) / sizeof(*TEST_ORDER); i++) { if (ia == -1 && strcmp(a->name, TEST_ORDER[i]) == 0) ia = i; if (ib == -1 && strcmp(b->name, TEST_ORDER[i]) == 0) ib = i; /* Once we've found both tests we can stop searching. */ if (ia != -1 && ib != -1) break; } /* If a test name was not found, it probably indicates an error in another part of the code. */ if (ia == -1) fatal("%s received an unknown test name \"%s\".\n", __func__, a->name); if (ib == -1) fatal("%s received an unknown test name \"%s\".\n", __func__, b->name); return ia < ib; } /* Merges the tests from several fingerprints into a character string representation. Tests that are identical between more than one fingerprint are included only once. If wrapit is true, the string is wrapped for submission. */ const char *mergeFPs(FingerPrint *FPs[], int numFPs, bool isGoodFP, const struct in_addr * const addr, int distance, const u8 *mac, int openTcpPort, int closedTcpPort, int closedUdpPort, bool wrapit) { static char str[10240]; static char wrapstr[10240]; char *p; int i; char *end = str + sizeof(str) - 1; /* Last byte allowed to write into */ std::list tests; std::list::iterator iter; const FingerTest *ft; if (numFPs <= 0) return "(None)"; else if (numFPs > 32) return "(Too many)"; /* Copy the tests from each fingerprint into a flat list. */ for (i = 0; i < numFPs; i++) { for (ft = FPs[i]; ft != NULL; ft = ft->next) tests.push_back(ft); } /* Put the tests in the proper order and ensure that tests with identical names are contiguous. */ tests.sort(FingerTest_lessthan); /* Delete duplicate tests to ensure that all the tests are unique. One test is a duplicate of the other if it has the same name as the first and the two results lists match. */ for (iter = tests.begin(); iter != tests.end(); iter++) { std::list::iterator tmp_i, next; tmp_i = iter; tmp_i++; while (tmp_i != tests.end() && strcmp((*iter)->name, (*tmp_i)->name) == 0) { next = tmp_i; next++; if (AVal_match_literal((*iter)->results, (*tmp_i)->results)) { /* This is a duplicate test. Remove it. */ tests.erase(tmp_i); } tmp_i = next; } } /* A safety check to make sure that no tests were lost in merging. */ for (i = 0; i < numFPs; i++) { for (ft = FPs[i]; ft != NULL; ft = ft->next) { for (iter = tests.begin(); iter != tests.end(); iter++) { if (strcmp((*iter)->name, ft->name) == 0 && AVal_match_literal((*iter)->results, ft->results)) { break; } } if (iter == tests.end()) { char buf[200]; AVal2str(ft->results, buf, sizeof(buf)); fatal("The test %s(%s) was somehow lost in %s.\n", ft->name, buf, __func__); } } } memset(str, 0, sizeof(str)); p = str; /* Lets start by writing the fake "SCAN" test for submitting fingerprints */ WriteSInfo(p, sizeof(str), isGoodFP, addr, distance, mac, openTcpPort, closedTcpPort, closedUdpPort); p = p + strlen(str); if (!wrapit) *p++ = '\n'; assert(p <= end); /* Append the string representation of each test to the result string. */ for (iter = tests.begin(); iter != tests.end(); iter++) { size_t len; ft = *iter; len = MIN((ptrdiff_t) strlen(ft->name), (end - p)); memcpy(p, ft->name, len); p += len; if (p >= end) break; *p++ = '('; len = AVal2str(ft->results, p, end - p + 1); p += len; if (p >= end) break; *p++ = ')'; if (!wrapit) { if (p >= end) break; *p++ = '\n'; } } /* If we bailed out of the loop early it was because we ran out of space. */ if (iter != tests.end() || p > end) fatal("Merged fingerprint too long in %s.\n", __func__); *p = '\0'; if(!wrapit) { return str; } else { /* Wrap the str. */ int len; char *p1 = wrapstr; end = wrapstr + sizeof(wrapstr) - 1; p = str; while(*p && end-p1 >= 3) { len = 0; strcpy(p1, "OS:"); p1 += 3; len +=3; while(*p && len <= FP_RESULT_WRAP_LINE_LEN && end-p1 > 0) { *p1++=*p++; len++; } if(end-p1<=0) { fatal("Wrapped result too long!\n"); break; } *p1++ = '\n'; } *p1 = '\0'; return wrapstr; } } const char *fp2ascii(FingerPrint *FP) { static char str[2048]; FingerPrint *current; struct AVal *AV; char *p = str; int len; memset(str, 0, sizeof(str)); if (!FP) return "(None)"; if(FP->OS_name && *(FP->OS_name)) { len = Snprintf(str, 128, "FingerPrint %s\n", FP->OS_name); if (len < 0) fatal("OS name too long"); p += len; } for(current = FP; current ; current = current->next) { Strncpy(p, current->name, sizeof(str) - (p-str)); p += strlen(p); assert(p-str < (int) sizeof(str) - 30); *p++='('; for(AV = current->results; AV; AV = AV->next) { Strncpy(p, AV->attribute, sizeof(str) - (p-str)); p += strlen(p); assert(p-str < (int) sizeof(str) - 30); *p++='='; Strncpy(p, AV->value, sizeof(str) - (p-str)); p += strlen(p); assert(p-str < (int) sizeof(str) - 30); *p++ = '%'; } if(*(p-1) != '(') p--; /* Kill the final & */ *p++ = ')'; *p++ = '\n'; } *p = '\0'; return str; } /* Return a zero-terminated copy of the substring that starts at p and ends at q, with leading and trailing whitespace stripped. The returned string is allocated with cp_alloc. */ static char *substrstrip(const char *p, const char *q) { char *s; assert(p <= q); while (isspace(*p)) p++; while (q > p && isspace(*(q - 1))) q--; s = (char *) cp_alloc(q - p + 1); memcpy(s, p, q - p); s[q - p] = '\0'; return s; } /* Parse a 'Class' line found in the fingerprint file into the current FP. Classno is the number of 'class' lines found so far in the current fingerprint. The function quits if there is a parse error */ static void parse_classline(FingerPrint *FP, char *thisline, int lineno, int *classno) { const char *begin, *end; struct OS_Classification *os_class; if (!thisline || strncmp(thisline, "Class ", 6) != 0) fatal("Bogus line #%d (%s) passed to %s()", lineno, thisline, __func__); if (*classno >= MAX_OS_CLASSIFICATIONS_PER_FP) fatal("Too many Class lines in fingerprint (line %d: %s), remove some or increase MAX_OS_CLASSIFICATIONS_PER_FP", lineno, thisline); os_class = &FP->OS_class[*classno]; /* First let's get the vendor name. */ begin = thisline + 6; end = strchr(begin, '|'); if (end == NULL) fatal("Parse error on line %d of fingerprint: %s\n", lineno, thisline); os_class->OS_Vendor = substrstrip(begin, end); /* Next comes the OS family. */ begin = end + 1; end = strchr(begin, '|'); if (end == NULL) fatal("Parse error on line %d of fingerprint: %s\n", lineno, thisline); os_class->OS_Family = substrstrip(begin, end); /* And now the the OS generation. */ begin = end + 1; end = strchr(begin, '|'); if (end == NULL) fatal("Parse error on line %d of fingerprint: %s\n", lineno, thisline); /* OS generation is handled specially: instead of an empty string it's supposed to be NULL. */ while (isspace(*begin)) begin++; if (begin < end) os_class->OS_Generation = substrstrip(begin, end); else os_class->OS_Generation = NULL; /* And finally the device type. We look for '\0' instead of '|'. */ begin = end + 1; end = strchr(begin, '\0'); os_class->Device_Type = substrstrip(begin, end); (*classno)++; FP->num_OS_Classifications++; } /* Parses a single fingerprint from the memory region given. If a non-null fingerprint is returned, the user is in charge of freeing it when done. This function does not require the fingerprint to be 100% complete since it is used by scripts such as scripts/fingerwatch for which some partial fingerpritns are OK. */ /* This function is not currently used by Nmap, but it is present here because it is used by fingerprint utilities that link with Nmap object files. */ FingerPrint *parse_single_fingerprint(char *fprint_orig) { int lineno = 0; int classno = 0; /* Number of Class lines dealt with so far */ char *p, *q; char *thisline, *nextline; char *fprint = strdup(fprint_orig); /* Make a copy we can futz with */ FingerPrint *FP; FingerPrint *current; /* Since a fingerprint is really a linked list of FingerPrint structures */ current = FP = (FingerPrint *) safe_zalloc(sizeof(FingerPrint)); thisline = fprint; do /* 1 line at a time */ { nextline = strchr(thisline, '\n'); if (nextline) *nextline++ = '\0'; /* printf("Preparing to handle next line: %s\n", thisline); */ while(*thisline && isspace((int) *thisline)) thisline++; if (!*thisline) { fatal("Parse error on line %d of fingerprint: %s", lineno, nextline); } if (strncmp(thisline, "Fingerprint ", 12) == 0) { p = thisline + 12; while(*p && isspace((int) *p)) p++; q = strchr(p, '\n'); if (!q) q = p + strlen(p); while(q > p && isspace(*(--q))) ; FP->OS_name = (char *) cp_alloc(q - p + 2); memcpy(FP->OS_name, p, q - p + 1); FP->OS_name[q - p + 1] = '\0'; } else if (strncmp(thisline, "Class ", 6) == 0) { parse_classline(FP, thisline, lineno, &classno); } else if ((q = strchr(thisline, '('))) { *q = '\0'; if(current->name) { current->next = (FingerPrint *) safe_zalloc(sizeof(FingerPrint)); current = current->next; } current->name = strdup(thisline); p = q+1; *q = '('; q = strchr(p, ')'); if (!q) { fatal("Parse error on line %d of fingerprint: %s\n", lineno, thisline); } *q = '\0'; current->results = str2AVal(p); } else { fatal("Parse error line line #%d of fingerprint", lineno); } thisline = nextline; /* Time to handle the next line, if there is one */ lineno++; } while (thisline && *thisline); return FP; } void free_fingerprint_file(FingerPrintDB *DB) { FingerPrint **FPs = DB->prints; FingerPrint **current; FingerPrint *c, *d; struct AVal *avc; struct AVal *avd; for(current = FPs; *current != NULL; current++){ for(c = *current; c; c=d){ d = c->next; if(c->name) free((void*)c->name); //strdup if(c->results){ for(avc = c->results; avc; avc = avd) { avd = avc->next; if(avc->attribute) free(avc->attribute); } free(c->results); } free(c); } } free(FPs); if (DB->MatchPoints) { for(c = DB->MatchPoints; c; c=d){ d = c->next; if(c->name) free((void*)c->name); //strdup if(c->results){ for(avc = c->results; avc; avc = avd) { avd = avc->next; if(avc->attribute) free(avc->attribute); } free(c->results); } free(c); } } free(DB); } FingerPrintDB *parse_fingerprint_file(char *fname) { FingerPrintDB *DB = NULL; FingerPrint *current; FILE *fp; int max_records = 4096; char line[512]; int numrecords = 0; int lineno = 0; bool parsingMatchPoints = false; int classno = 0; /* Number of Class lines dealt with so far */ DB = (FingerPrintDB *) safe_zalloc(sizeof(FingerPrintDB)); char *p, *q; /* OH YEAH!!!! */ if (!DB) fatal("non-allocated DB passed to %s", __func__); DB->prints = (FingerPrint **) safe_zalloc(sizeof(FingerPrint *) * max_records); fp = fopen(fname, "r"); if (!fp) fatal("Unable to open Nmap fingerprint file: %s", fname); top: while(fgets(line, sizeof(line), fp)) { lineno++; /* Read in a record */ if (*line == '\n' || *line == '#') continue; fparse: if (strncasecmp(line, "FingerPrint", 11) == 0) { parsingMatchPoints = false; } else if (strncasecmp(line, "MatchPoints", 11) == 0) { if (DB->MatchPoints) fatal("Found MatchPoints directive on line %d of %s even though it has previously been seen in the file", lineno, fname); parsingMatchPoints = true; } else { error("Parse error on line %d of nmap-os-db file: %s", lineno, line); continue; } current = (FingerPrint *) safe_zalloc(sizeof(FingerPrint)); if (parsingMatchPoints) { current->OS_name = NULL; DB->MatchPoints = current; } else { DB->prints[numrecords] = current; p = line + 12; while(*p && isspace((int) *p)) p++; q = strpbrk(p, "\n#"); if (!p) fatal("Parse error on line %d of fingerprint: %s", lineno, line); while(isspace(*(--q))) ; if (q < p) fatal("Parse error on line %d of fingerprint: %s", lineno, line); current->OS_name = (char *) cp_alloc(q - p + 2); memcpy(current->OS_name, p, q - p + 1); current->OS_name[q - p + 1] = '\0'; } current->line = lineno; classno = 0; /* Now we read the fingerprint itself */ while(fgets(line, sizeof(line), fp)) { lineno++; if (*line == '#') continue; if (*line == '\n') break; if (!strncmp(line, "FingerPrint ",12)) { goto fparse; } else if (strncmp(line, "Class ", 6) == 0) { parse_classline(current, line, lineno, &classno); } else { p = line; q = strchr(line, '('); if (!q) { error("Parse error on line %d of nmap-os-db file: %s", lineno, line); goto top; } *q = '\0'; if(current->name) { current->next = (FingerPrint *) safe_zalloc(sizeof(FingerPrint)); current = current->next; } current->name = strdup(p); p = q+1; *q = '('; q = strchr(p, ')'); if (!q) { error("Parse error on line %d of nmap-os-db file: %s", lineno, line); goto top; } *q = '\0'; current->results = str2AVal(p); } } /* printf("Read in fingerprint:\n%s\n", fp2ascii(DB->prints[numrecords])); */ if (!parsingMatchPoints) numrecords++; if (numrecords >= max_records) fatal("Too many OS fingerprints -- 0verflow"); } fclose(fp); DB->prints[numrecords] = NULL; return DB; } FingerPrintDB *parse_fingerprint_reference_file(const char *dbname) { char filename[256]; if (nmap_fetchfile(filename, sizeof(filename), dbname) != 1){ fatal("OS scan requested but I cannot find %s file. It should be in %s, ~/.nmap/ or .", dbname, NMAPDATADIR); } /* Record where this data file was found. */ o.loaded_data_files[dbname] = filename; return parse_fingerprint_file(filename); }