// Slightly modified from getTextEntity
short get_iana_charset(char *cs, int len)
{
if (!s_isInitialized){
// set up the hash table
if ( ! s_table.set ( 8,4,4096,NULL,0,false,0,"ianatbl") )
return log("build: Could not init table of "
"IANA Charsets.");
// now add in all the charset entries
long n = (long)sizeof(s_charsets) / (long)sizeof(IANACharset);
// turn off quickpolling
char saved = g_conf.m_useQuickpoll;
g_conf.m_useQuickpoll = false;
for ( long i = 0 ; i < n ; i++ ) {
long long h = hash64Lower_a ( s_charsets[i].name, strlen(s_charsets[i].name) );
// store the charset index in the hash table as score
if ( ! s_table.addTerm(&h, i+1) )
return log("build: add term failed");
}
g_conf.m_useQuickpoll = saved;
s_isInitialized = true;
}
long long h = hash64Lower_a ( cs , len );
// get the entity index from table (stored in the score field)
long i = (long) s_table.getScore ( &h );
// return 0 if no match
if ( i == 0 ) return csUnknown;
// return the iso character
return (short)s_charsets[i-1].mib_enum;
}
// Slightly modified from getTextEntity
int16_t get_iana_charset(const char *cs, int len)
{
if (!s_isInitialized){
// set up the hash table
if ( ! s_table.set ( 8,4,4096,NULL,0,false,"ianatbl") ) {
log(LOG_WARN, "build: Could not init table of IANA Charsets.");
return csUnknown;
}
// now add in all the charset entries
int32_t n = (int32_t)sizeof(s_charsets) / (int32_t)sizeof(IANACharset);
// turn off quickpolling
char saved = g_conf.m_useQuickpoll;
g_conf.m_useQuickpoll = false;
for ( int32_t i = 0 ; i < n ; i++ ) {
int64_t h = hash64Lower_a ( s_charsets[i].name, strlen(s_charsets[i].name) );
// store the charset index in the hash table as score
if ( ! s_table.addTerm(h, i+1) ) {
log(LOG_WARN, "build: add term failed");
return csUnknown;
}
}
g_conf.m_useQuickpoll = saved;
s_isInitialized = true;
}
int64_t h = hash64Lower_a ( cs , len );
// get the entity index from table (stored in the score field)
int32_t i = (int32_t) s_table.getScore(h);
// return 0 if no match
if ( i == 0 ) return csUnknown;
// return the iso character
return (int16_t)s_charsets[i-1].mib_enum;
}
static bool loadSpiderProxyStats() {
initProxyTables();
// take this out for now since i was seeing dups in s_iptab for
// some reason. was causing an infinite loop bug calling goto redo:
// all the time above.
// save hashtable
//s_proxyBannedTable.load(g_hostdb.m_dir,"proxybantable.dat");
//s_banCountTable.load(g_hostdb.m_dir,"proxybancounttable.dat");
// save hash table. this also returns false if does not exist.
//if ( ! s_iptab.load(g_hostdb.m_dir,"spiderproxystats.dat") )
// return false;
// unset some flags
for ( int32_t i = 0 ; i < s_iptab.getNumSlots() ; i++ ) {
// skip empty slots
if ( ! s_iptab.m_flags[i] ) continue;
SpiderProxy *sp = (SpiderProxy *)s_iptab.getValueFromSlot(i);
sp->m_isWaiting = false;
}
return true;
}
// . init s_mimeTable in this call
// . called from HttpServer::init
// . returns false and sets g_errno on error
bool HttpMime::init ( ) {
// only need to call once
if ( s_init ) return true;
// make sure only called once
s_init = true;
//s_mimeTable.set ( 256 );
//s_mimeTable.setLabel("mimetbl");
if ( ! s_mimeTable.set(4,sizeof(char *),256,NULL,0,false,1,"mimetbl"))
return false;
// set table from internal list
for ( uint32_t i = 0 ; i < sizeof(s_ext)/sizeof(char *) ; i+=2 ) {
int32_t key = hash32n ( s_ext[i] );
if ( ! s_mimeTable.addKey ( &key , &s_ext[i+1] ) )
return log("HttpMime::init: failed to set table.");
}
// quick text
const char *tt = getContentTypeFromExtension ( "zip" );
if ( strcmp(tt,"application/zip") != 0 ) {
g_errno = EBADENGINEER;
return log("http: Failed to init mime table correctly.");
}
// a more thorough test
for ( uint32_t i = 0 ; i < sizeof(s_ext)/sizeof(char *) ; i+=2) {
tt = getContentTypeFromExtension ( s_ext[i] );
if ( strcmp(tt,s_ext[i+1]) == 0 ) continue;
g_errno = EBADENGINEER;
return log("http: Failed to do mime table correctly. i=%" PRId32,i);
}
// TODO: set it from a user supplied file here
return true;
}
bool AdultBit::isDirty ( char *s , int32_t len ) {
static bool s_isInitialized = false;
static char *s_dirty[] = {
"anal",
"analsex",
"b*****b",
"blowjobs",
"boob",
"boobs",
"clitoris",
"c**k",
"cocks",
"cum",
"dick",
"dicks",
"g******g",
"gangbangs",
"gangbanging",
"movie",
"movies",
"oral",
"oralsex",
"p**n",
"porno",
"pussy",
"pussies",
"sex",
"sexy",
"tit",
"t**s",
"video",
"videos",
"xxx",
"xxxx",
"xxxx"
};
if ( ! s_isInitialized ) {
// set up the hash table
if ( ! s_dtable.set ( 8,4,sizeof(s_dirty )*2,NULL,0,false,0,
"adulttab"))
return log("build: Error initializing "
"dirty word hash table." );
// now add in all the dirty words
int32_t n = (int32_t)sizeof(s_dirty)/ sizeof(char *);
for ( int32_t i = 0 ; i < n ; i++ ) {
int64_t h = hash64b ( s_dirty [i] );
if ( ! s_dtable.addTerm (&h, i+1) ) return false;
}
s_isInitialized = true;
}
// compute the hash of the word "s"
int64_t h = hash64Lower_a ( s , len );
// get from table
return s_dtable.getScore ( &h );
}
static bool initProxyTables() {
// initialize proxy/urlip ban table?
if ( ! s_init ) return true;
s_init = false;
s_proxyBannedTable.set(8,0,0,NULL,0,false,"proxban");
s_banCountTable.set(4,4,0,NULL,0,false,"bancnt");
return true;
}
// save the stats
bool saveSpiderProxyStats ( ) {
// save hashtable
s_proxyBannedTable.save(g_hostdb.m_dir,"proxybantable.dat");
s_banCountTable.save(g_hostdb.m_dir,"proxybancounttable.dat");
// save hash table
return s_iptab.save(g_hostdb.m_dir,"spiderproxystats.dat");
}
//
// . new code for saving hashtablex in a thread
// . so Process.cpp's call to g_spiderCache.save() can save the doleiptable
// without blocking...
//
static void *saveWrapper ( void *state , class ThreadEntry *t ) {
// get this class
HashTableX *THIS = (HashTableX *)state;
// this returns false and sets g_errno on error
THIS->save( THIS->m_dir ,
THIS->m_filename ,
THIS->m_tbuf ,
THIS->m_tsize );
// now exit the thread, bogus return
return NULL;
}
bool resetProxyStats ( ) {
// s_proxyBannedTable.reset();
// s_banCountTable.reset();
// s_iptab.reset();
s_iptab.set(8,sizeof(SpiderProxy),0,NULL,0,false,"siptab",true);
// skip port part of key magic, and get LSB of the IP as key magic
s_iptab.m_maskKeyOffset = 5;
s_proxyBannedTable.set(8,0,0,NULL,0,false,"proxban");
s_banCountTable.set(4,4,0,NULL,0,false,"bancnt");
return buildProxyTable();
}
SpiderProxy *getSpiderProxyByIpPort ( int32_t ip , uint16_t port ) {
for ( int32_t i = 0 ; i < s_iptab.getNumSlots() ; i++ ) {
// skip empty slots
if ( ! s_iptab.m_flags[i] ) continue;
SpiderProxy *sp = (SpiderProxy *)s_iptab.getValueFromSlot(i);
if ( sp->m_ip != ip ) continue;
if ( sp->m_port != port ) continue;
return sp;
}
return NULL;
}
void gbiconv_reset(){
for (int32_t i=0;i<s_convTable.getNumSlots();i++){
//int32_t key = *(int32_t *)s_convTable.getKey(i);
//if (!key) continue;
if ( ! s_convTable.m_flags[i] ) continue;
iconv_t *pconv = (iconv_t *)s_convTable.getValueFromSlot(i);
if (! pconv) continue;
iconv_t iconv = *pconv;
//logf(LOG_DEBUG, "iconv: freeing iconv: 0x%x", (int)iconv);
g_mem.rmMem((void*)iconv, 52, "iconv");
libiconv_close(iconv);
}
s_convTable.reset();
}
static bool initEntityTable(){
if ( ! s_isInitialized ) {
// set up the hash table
if ( ! s_table.set ( 8,4,255,NULL,0,false,0,"enttbl" ) )
return log("build: Could not init table of "
"HTML entities.");
// now add in all the stop words
int32_t n = (int32_t)sizeof(s_entities) / (int32_t)sizeof(Entity);
for ( int32_t i = 0 ; i < n ; i++ ) {
int64_t h = hash64b ( s_entities[i].entity );
// grab the unicode code point
UChar32 up = s_entities[i].unicode;
// now we are 100% up
if ( ! up ) { char *xx=NULL;*xx=0; }
// point to it
char *buf = (char *)s_entities[i].utf8;
// if uchar32 not 0 then set the utf8 with it
int32_t len = utf8Encode(up,buf);
//
// make my own mods to make parsing easier
//
if ( up == 160 ) { // nbsp
buf[0] = ' ';
len = 1;
}
//
// end custom mods
//
// set length
s_entities[i].utf8Len = len;
// check it
if ( len == 0 ) { char *xx=NULL;*xx=0; }
// must not exist!
if ( s_table.isInTable(&h) ) { char*xx=NULL;*xx=0;}
// store the entity index in the hash table as score
if ( ! s_table.addTerm ( &h, i+1 ) ) return false;
}
s_isInitialized = true;
}
return true;
}
iconv_t gbiconv_open( char *tocode, char *fromcode) {
// get hash for to/from
uint32_t hash1 = hash32Lower_a(tocode, gbstrlen(tocode), 0);
uint32_t hash2 = hash32Lower_a(fromcode, gbstrlen(fromcode),0);
uint32_t hash = hash32h(hash1, hash2);
g_errno = 0;
iconv_t *convp = (iconv_t *)s_convTable.getValue(&hash);
iconv_t conv = NULL;
if ( convp ) conv = *convp;
//log(LOG_DEBUG, "uni: convertor %s -> %s from hash 0x%"XINT32": 0x%"XINT32"",
// fromcode, tocode,
// hash, conv);
if (!conv){
//log(LOG_DEBUG, "uni: Allocating new convertor for "
// "%s to %s (hash: 0x%"XINT32")",
// fromcode, tocode,hash);
conv = iconv_open(tocode, fromcode);
if (conv == (iconv_t) -1) {
log(LOG_WARN, "uni: failed to open converter for "
"%s to %s: %s (%d)", fromcode, tocode,
strerror(errno), errno);
// need to stop if necessary converters don't open
//char *xx=NULL; *xx = 0;
g_errno = errno;
if (errno == EINVAL)
g_errno = EBADCHARSET;
return conv;
}
// add mem to table to keep track
g_mem.addMem((void*)conv, 52, "iconv", 1);
// cache convertor
s_convTable.addKey(&hash, &conv);
//log(LOG_DEBUG, "uni: Saved convertor 0x%"INT32" under hash 0x%"XINT32"",
// conv, hash);
}
else{
// reset convertor
char *dummy = NULL;
size_t dummy2 = 0;
// JAB: warning abatement
//size_t res = iconv(conv,NULL,NULL,&dummy,&dummy2);
iconv(conv,NULL,NULL,&dummy,&dummy2);
}
return conv;
}
bool saveHashTable ( ) {
if ( s_ht.m_numSlotsUsed <= 0 ) return true;
SafeBuf fn;
fn.safePrintf("%s/qa/",g_hostdb.m_dir);
log("qa: saving crctable.dat");
s_ht.save ( fn.getBufStart() , "crctable.dat" );
return true;
}
bool AdultBit::isObscene ( char *s , int32_t len ) {
static bool s_isInitialized = false;
static char *s_obscene[] = {
"c**t",
"clits",
// "cum", magna cum laude
"cums",
"cumshot",
"c**t",
"cunts",
"milf",
"rimjob",
"felch",
"f**k",
"f****d",
"f****r",
"f*****g",
"f***s",
"w***e",
"w****s"
};
if ( ! s_isInitialized ) {
// set up the hash table
if ( ! s_otable.set ( 8,4,sizeof(s_obscene)*2,NULL,0,false,0,
"obscenetab") )
return log("build: Error initializing "
"obscene word hash table." );
// now add in all the stop words
int32_t n = sizeof(s_obscene) / sizeof(char *);
for ( int32_t i = 0 ; i < n ; i++ ) {
int64_t h = hash64b ( s_obscene[i] );
if ( ! s_otable.addTerm ( &h, i+1 ) ) return false;
}
s_isInitialized = true;
}
// compute the hash of the word "s"
int64_t h = hash64Lower_a ( s , len );
// get from table
return s_otable.getScore ( &h );
}
nodeid_t getTagId ( char *s , NodeType **retp ) {
// init table?
static bool s_init = false;
static HashTableX s_ht;
static char s_buf[10000];
if ( ! s_init ) {
s_init = true;
s_ht.set ( 4 ,4,1024,s_buf,10000,false,0,"tagids");//niceness=0
// how many NodeTypes do we have in g_nodes?
static int32_t nn = sizeof(g_nodes) / sizeof(NodeType);
// set the hash table
for ( int32_t i = 0 ; i < nn ; i++ ) {
char *name = g_nodes[i].m_nodeName;
int32_t nlen = gbstrlen(name);
int64_t h = hash64Upper_a ( name,nlen,0LL );
NodeType *nt = &g_nodes[i];
if ( ! s_ht.addKey(&h,&nt) ) {
char *xx=NULL;*xx=0; }
}
// sanity
if ( s_ht.m_numSlots != 1024 ) { char *xx=NULL;*xx=0; }
// sanity test
nodeid_t tt = getTagId ( "br" );
if ( tt != TAG_BR ) { char *xx=NULL;*xx=0; }
}
// find end of tag name. hyphens are ok to be in name.
// facebook uses underscores like <start_time>
char *e = s; for ( ; *e && (is_alnum_a(*e) || *e=='-'|| *e=='_'); e++);
// hash it for lookup
int64_t h = hash64Upper_a ( s , e - s , 0 );
// look it up
NodeType **ntp = (NodeType **)s_ht.getValue(&h);
// assume none
if ( retp ) *retp = NULL;
// none?
if ( ! ntp ) return 0;
// got one
if ( retp ) *retp = *ntp;
// get id otherwise
return (*ntp)->m_nodeId;
}
const char *extensionToContentTypeStr2 ( const char *ext , int32_t elen ) {
// assume text/html if no extension provided
if ( ! ext || ! ext[0] ) return NULL;
if ( elen <= 0 ) return NULL;
// get hash for table look up
int32_t key = hash32 ( ext , elen );
char **pp = (char **)s_mimeTable.getValue ( &key );
if ( ! pp ) return NULL;
return *pp;
}
// . list of types is on: http://www.duke.edu/websrv/file-extensions.html
// . i copied it to the bottom of this file though
const char *HttpMime::getContentTypeFromExtension ( const char *ext ) {
// assume text/html if no extension provided
if ( ! ext || ! ext[0] ) return "text/html";
// get hash for table look up
int32_t key = hash32n ( ext );
char **pp = (char **)s_mimeTable.getValue ( &key );
// if not found in table, assume text/html
if ( ! pp ) return "text/html";
return *pp;
}
// call this at startup to register the handlers
bool initSpiderProxyStuff() {
// do this for all hosts in case host #0 goes dead, then everyone
// will, according to Msg13.cpp, send to host #1, the next in line
// if she is alive
//if ( g_hostdb.m_myHostId != 0 ) return true;
// only host #0 has handlers
if ( ! g_udpServer.registerHandler ( msg_type_54, handleRequest54 ))
return false;
// key is ip/port
s_iptab.set(8,sizeof(SpiderProxy),0,NULL,0,false,"siptab",true);
// skip port part of key magic, and get LSB of the IP as key magic
s_iptab.m_maskKeyOffset = 5;
loadSpiderProxyStats();
// build the s_iptab hashtable for the first time
buildProxyTable ();
// reset spider proxy stats every hour to alleviate false positives (moved from Process.cpp)
if (!g_loop.registerSleepCallback(3600000, NULL, resetProxyStatWrapper, 0)) {
gbshutdownResourceError();
}
// make the loadtable hashtable
static bool s_flag = 0;
if ( s_flag ) return true;
s_flag = true;
return s_loadTable.set(4,
sizeof(LoadBucket),
128,
NULL,
0,
// this slows us down
true, // allow dups?
"lbtab",
true); // use key magic to mix things up
}
// . use msg 0x55 to say you are done using the proxy
// . we now use the top part of the Msg13Request as the proxy request
void returnProxy ( Msg13Request *preq , UdpSlot *udpSlot ) {
//char *p = request;
//int32_t proxyIp = *(int32_t *)p; p += 4;
//int16_t proxyPort = *(int16_t *)p; p += 2;
//int32_t lbId = *(int32_t *)p; p += 4;
int32_t urlIp = preq->m_urlIp;
//
// update the load bucket
//
// scan over all that match to find lbid
int32_t hslot = s_loadTable.getSlot ( &urlIp );
// scan all proxies that have this urlip outstanding
int32_t i;for (i=hslot ; i >= 0 ; i = s_loadTable.getNextSlot(i,&urlIp)){
// get the bucket
LoadBucket *lb= (LoadBucket *)s_loadTable.getValueFromSlot(i);
// is it the right id?
if ( lb->m_id != preq->m_lbId ) continue;
if ( lb->m_proxyIp != preq->m_proxyIp ) continue;
if ( lb->m_proxyPort != preq->m_proxyPort ) continue;
// that's it. set the download end time
int64_t nowms = gettimeofdayInMillisecondsLocal();
lb->m_downloadEndTimeMS = nowms;
break;
}
if ( i < 0 )
log("sproxy: could not find load bucket id #%" PRId32,preq->m_lbId);
// if no slot provided, return to called without sending back reply,
// they are banning a proxy and need to also return it before
// we send them back another proxy to try.
if ( ! udpSlot ) return;
// gotta send reply back
g_udpServer.sendReply(0, 0, 0, 0, udpSlot);
}
// get the id from a 2 character country code
uint8_t getCountryId ( char *cc ) {
static bool s_init = false;
static char buf[2000];
static HashTableX ht;
char tmp[4];
if ( ! s_init ) {
s_init = true;
// hash them up
ht.set ( 4 , 1 , -1,buf,2000,false,MAX_NICENESS,"ctryids");
// now add in all the country codes
long n = (long) sizeof(s_countryCode) / sizeof(char *);
for ( long i = 0 ; i < n ; i++ ) {
char *s = (char *)s_countryCode[i];
//long slen = gbstrlen ( s );
// sanity check
if ( !s[0] || !s[1] || s[2]) { char *xx=NULL;*xx=0; }
// map it to a 4 byte key
tmp[0]=s[0];
tmp[1]=s[1];
tmp[2]=0;
tmp[3]=0;
// a val of 0 does not mean empty in HashTableX,
// that is an artifact of HashTableT
uint8_t val = i; // +1;
// add 1 cuz 0 means lang unknown
if ( ! ht.addKey ( tmp , &val ) ) {
char *xx=NULL;*xx=0; }
}
}
// lookup
tmp[0]=to_lower_a(cc[0]);
tmp[1]=to_lower_a(cc[1]);
tmp[2]=0;
tmp[3]=0;
long slot = ht.getSlot ( tmp );
if ( slot < 0 ) return 0;
void *val = ht.getValueFromSlot ( slot );
return *(uint8_t *)val ;
}
// . how many keys are dups
// . returns -1 on error
long HashTableX::getNumDups() {
if ( ! m_allowDups ) return 0;
HashTableX tmp;
if ( ! tmp.set ( m_ks, 0, m_numSlots, NULL , 0 , false , m_niceness,
"htxtmp") )
return -1;
// put into that table
for ( long i = 0 ; i < m_numSlots ; i++ ) {
// skip empty bucket
if ( ! m_flags[i] ) continue;
// get the key
char *kp = (char *)getKeyFromSlot(i);
// add to new table
if ( ! tmp.addKey ( kp ) ) return -1;
}
// the unqieus
long uniques = tmp.m_numSlotsUsed;
// the dups
long dups = m_numSlotsUsed - uniques;
// that's it
return dups;
}
// we come here after thread exits
static void threadDoneWrapper ( void *state , class ThreadEntry *t ) {
// get this class
HashTableX *THIS = (HashTableX *)state;
// store save error into g_errno
//g_errno = THIS->m_saveErrno;
// log it
log("db: done saving %s/%s",THIS->m_dir,THIS->m_filename);
// . resume adding to the hashtable
// . this will also allow other threads to be queued
// . if we did this at the end of the thread we could end up with
// an overflow of queued SAVETHREADs
THIS->m_isSaving = false;
// we do not need to be saved now?
THIS->m_needsSave = false;
// g_errno should be preserved from the thread so if threadSave()
// had an error it will be set
if ( g_errno )
log("db: Had error saving hashtable to disk for %s: %s.",
THIS->m_allocName,mstrerror(g_errno));
// . call callback
if ( THIS->m_callback ) THIS->m_callback ( THIS->m_state );
}
static bool initEntityTable(){
if ( ! s_isInitialized ) {
// set up the hash table
if ( ! s_table.set ( 8,4,4096,NULL,0,false,"enttbl" ) ) {
log("build: Could not init table of HTML entities.");
return false;
}
// now add in all the html entities
const int32_t n = (int32_t)sizeof(s_entities) / (int32_t)sizeof(Entity);
for ( int32_t i = 0 ; i < n ; i++ ) {
int64_t h = hash64b ( s_entities[i].entity );
// convert the unicode codepoints to an utf8 string
char *buf = (char *)s_entities[i].utf8;
for(int j=0; j<s_entities[i].codepoints; j++) {
UChar32 codepoint = s_entities[i].codepoint[j];
int32_t len = utf8Encode(codepoint,buf);
if ( len == 0 ) { g_process.shutdownAbort(true); }
// make modification to make parsing easier
if ( codepoint == 160 ) { // nbsp
buf[0] = ' ';
len = 1;
}
buf += len;
}
s_entities[i].utf8Len = (size_t)(buf-s_entities[i].utf8);
// must not exist!
if ( s_table.isInTable(&h) ) { g_process.shutdownAbort(true);}
// store the entity index in the hash table as score
if ( ! s_table.addTerm(h, i+1) ) return false;
}
s_isInitialized = true;
}
return true;
}
// . is "s" an HTML entity? (ascii representative of an iso char)
// . return the 32-bit unicode char it represents
// . returns 0 if none
// . JAB: const-ness for optimizer...
static const Entity *getTextEntity ( const char *s , int32_t len ) {
if ( !initEntityTable()) return 0;
// take the ; off, if any
if ( s[len-1] == ';' ) len--;
// compute the hash of the entity including &, but not ;
int64_t h = hash64 ( s , len );
// get the entity index from table (stored in the score field)
int32_t i = (int32_t) s_table.getScore(h);
// return 0 if no match
if ( i == 0 )
return NULL;
// point to the utf8 char. these is 1 or 2 bytes it seems
return s_entities+i-1;
}
// . is "s" an HTML entity? (ascii representative of an iso char)
// . return the 32-bit unicode char it represents
// . returns 0 if none
// . JAB: const-ness for optimizer...
uint32_t getTextEntity ( const char *s , int32_t len ) {
if ( !initEntityTable()) return 0;
// take the ; off, if any
if ( s[len-1] == ';' ) len--;
// compute the hash of the entity including &, but not ;
int64_t h = hash64 ( s , len );
// get the entity index from table (stored in the score field)
int32_t i = (int32_t) s_table.getScore ( &h );
// return 0 if no match
if ( i == 0 ) return 0;
// point to the utf8 char. these is 1 or 2 bytes it seems
char *p = (char *)s_entities[i-1].utf8;
// encode into unicode
uint32_t c = utf8Decode ( p );
// return that
return c;
}
// . sets *current to how many downloads are CURRENTly outstanding for
// this proxy
// . returns total # of load points, i.e. downloads, in the last
// LOADPOINT_EXPIRE_MS milliseconds (currently 10 minutes)
static int32_t getNumLoadPoints(SpiderProxy *sp, int32_t *current) {
// currently outstanding downloads on proxy
*current = 0;
int32_t count = 0;
// scan all proxies that have this urlip outstanding
for ( int32_t i = 0 ; i < s_loadTable.m_numSlots ; i++ ) {
// skip if empty
if ( ! s_loadTable.m_flags[i] ) continue;
// get the bucket
LoadBucket *lb= (LoadBucket *)s_loadTable.getValueFromSlot(i);
// get the spider proxy this load point was for
if ( lb->m_proxyIp != sp->m_ip ) continue;
if ( lb->m_proxyPort != sp->m_port ) continue;
// currently outstanding downloads on proxy
if ( lb->m_downloadEndTimeMS == 0LL )
*current = *current + 1;
count++;
}
return count;
}
// . is "s" an HTML entity? (ascii representative of an iso char)
// . return the 32-bit unicode char it represents
// . returns 0 if none
// . JAB: const-ness for optimizer...
uint32_t getTextEntity ( char *s , int32_t len ) {
if ( !initEntityTable()) return 0;
// take the ; off, if any
if ( s[len-1] == ';' ) len--;
// compute the hash of the entity including &, but not ;
int64_t h = hash64 ( s , len );
// get the entity index from table (stored in the score field)
int32_t i = (int32_t) s_table.getScore ( &h );
// return 0 if no match
if ( i == 0 ) return 0;
// point to the utf8 char. these is 1 or 2 bytes it seems
char *p = (char *)s_entities[i-1].utf8;
// encode into unicode
uint32_t c = utf8Decode ( p );
// return that
return c;
// return the iso character
//printf("Converted text entity \"");
//for(int si=0;si<len;si++)putchar(s[si]);
//printf("\" to 0x%x(%d)\"%c\"\n",s_entities[i-1].c,s_entities[i-1].c,
// s_entities[i-1].c);
//return (uint32_t)s_entities[i-1].c;
}
// . when the Conf::m_proxyIps parm is updated we call this to rebuild
// s_iptab, our table of SpiderProxy instances, which has the proxies and
// their performance statistics.
// . we try to maintain stats of ip/ports that did NOT change when rebuilding.
bool buildProxyTable ( ) {
// scan the NEW list of proxy ip/port pairs in g_conf
char *p = g_conf.m_proxyIps.getBufStart();
HashTableX tmptab;
tmptab.set(8,0,16,NULL,0,false,"tmptab");
// scan the user inputted space-separated list of ip:ports
// (optional username:password@ip:port)
for ( ; *p ; ) {
// skip white space
if ( is_wspace_a(*p) ) { p++; continue; }
// skip http://
if ( strncasecmp(p,"http://",7) == 0 ) { p += 7; continue; }
// scan in an ip:port
char *s = p; char *portStr = NULL;
int32_t dc = 0, pc = 0, gc = 0, bc = 0;
const char *msg;
char *usernamePwd = NULL;
int32_t usernamePwdLen = 0;
char *ipStart = p;
// scan all characters until we hit \0 or another whitespace
for ( ; *s && !is_wspace_a(*s); s++) {
if ( *s == '@' ) {
// must be username:pwd
if ( pc != 1 ) {
msg = "bad username:password";
goto hadError;
}
usernamePwd = p;
usernamePwdLen = s - p;
if ( usernamePwdLen >= MAXUSERNAMEPWD-2 ) {
msg = "username:password too long";
goto hadError;
}
dc = 0;
gc = 0;
bc = 0;
pc = 0;
portStr = NULL;
ipStart = s+1;
continue;
}
if ( *s == '.' ) { dc++; continue; }
if ( *s == ':' ) { portStr=s; pc++; continue; }
if ( is_digit(*s) ) { gc++; continue; }
bc++;
continue;
}
// ensure it is a legit ip:port combo
msg = NULL;
if ( gc < 4 )
msg = "not enough digits for an ip";
if ( pc > 1 )
msg = "too many colons";
if ( dc != 3 )
msg = "need 3 dots for an ip address";
if ( bc )
msg = "got illegal char in ip:port listing";
if ( msg ) {
hadError:
char c = *s;
*s = '\0';
log("buf: %s for %s",msg,p);
*s = c;
return false;
}
// convert it
int32_t iplen = s - ipStart;
if ( portStr ) iplen = portStr - ipStart;
int32_t ip = atoip(ipStart,iplen);
// another sanity check
if ( ip == 0 || ip == -1 ) {
log("spider: got bad proxy ip for %s",p);
return false;
}
// and the port default is 80
int32_t port = 80;
if ( portStr ) port = atol2(portStr+1,s-portStr-1);
if ( port < 0 || port > 65535 ) {
log("spider: got bad proxy port for %s",p);
return false;
}
// . we got a legit ip:port
// . see if already in our table
uint64_t ipKey = (uint32_t)ip;
ipKey <<= 16;
ipKey |= (uint16_t)(port & 0xffff);
// also store into tmptable to see what we need to remove
tmptab.addKey(&ipKey);
// see if in table
int32_t islot = s_iptab.getSlot( &ipKey);
// advance p
p = s;
// if in there, keep it as is
if ( islot >= 0 ) continue;
// otherwise add new entry
SpiderProxy newThing;
memset ( &newThing , 0 , sizeof(SpiderProxy));
newThing.m_ip = ip;
newThing.m_port = port;
newThing.m_lastDownloadTookMS = -1;
newThing.m_lastSuccessfulTestMS = -1;
gbmemcpy(newThing.m_usernamePwd,usernamePwd,usernamePwdLen);
// ensure it is NULL terminated
newThing.m_usernamePwd[usernamePwdLen] = '\0';
if ( ! s_iptab.addKey ( &ipKey, &newThing ) )
return false;
}
redo:
int32_t removed = 0;
// scan all SpiderProxies in tmptab
for ( int32_t i = 0 ; i < s_iptab.getNumSlots() ; i++ ) {
// skip empty buckets in hashtable s_iptab
if ( ! s_iptab.m_flags[i] ) continue;
// get the key
int64_t key = *(int64_t *)s_iptab.getKeyFromSlot(i);
// must also exist in tmptab, otherwise it got removed by user
if ( tmptab.isInTable ( &key ) ) continue;
// skip if not in table
if ( s_iptab.getSlot ( &key ) < 0 ) {
log("sproxy: iptable hashing messed up");
continue;
}
// shoot, it got removed. not in the new list of ip:ports
s_iptab.removeKey ( &key );
removed++;
// hashtable is messed up now, start over
//goto redo;
}
if ( removed ) goto redo;
return true;
}
// a host is asking us (host #0) what proxy to use?
static void handleRequest54(UdpSlot *udpSlot, int32_t niceness) {
char *request = udpSlot->m_readBuf;
int32_t requestSize = udpSlot->m_readBufSize;
// we now use the top part of the Msg13Request as the ProxyRequest
Msg13Request *preq = (Msg13Request *)request;
// sanity check
if ( requestSize != preq->getProxyRequestSize() ) {
log("db: Got bad request 0x54 size of %" PRId32" bytes. bad",
requestSize );
g_udpServer.sendErrorReply ( udpSlot , EBADREQUESTSIZE );
return;
}
// is the request telling us it is done downloading through a proxy?
if ( preq->m_opCode == OP_RETPROXY ) {
returnProxy ( preq , udpSlot );
return;
}
// if sender is asking for a new proxy and wants us to ban
// the previous proxy we sent for this urlIp...
if ( preq->m_banProxyIp ) {
// don't core if misses sanity. it seems we don't always
// NULLify these or something.
// these must match
if(preq->m_banProxyIp != preq->m_proxyIp ||
preq->m_banProxyPort != preq->m_proxyPort){
log("db: proxy: banproxyip != proxyip. mismatch!");
g_udpServer.sendErrorReply ( udpSlot , EBADENGINEER);
return;
}
// this will "return" the banned proxy
returnProxy ( preq , NULL );
// now add it to the banned table
int64_t uip = preq->m_urlIp;
int64_t pip = preq->m_banProxyIp;
int64_t h64 = hash64h ( uip , pip );
if ( ! s_proxyBannedTable.isInTable ( &h64 ) ) {
s_proxyBannedTable.addKey ( &h64 );
// for stats counting. each proxy ip maps to #
// of unique website IPs that have banned it.
s_banCountTable.addTerm32((uint32_t)pip);
}
}
// shortcut
int32_t urlIp = preq->m_urlIp;
// send to a proxy that is up and has the least amount
// of LoadBuckets with this urlIp, if tied, go to least loaded.
// clear counts for this url ip for scoring the best proxy to use
for ( int32_t i = 0 ; i < s_iptab.getNumSlots() ; i++ ) {
// skip empty slots
if ( ! s_iptab.m_flags[i] ) continue;
SpiderProxy *sp = (SpiderProxy *)s_iptab.getValueFromSlot(i);
sp->m_countForThisIp = 0;
sp->m_lastTimeUsedForThisIp = 0LL;
}
// this table maps a url's current IP to a possibly MULTIPLE slots
// which tell us what proxy is downloading a page from that IP.
// so we can try to find a proxy that is not download a url from
// this IP currently, or hasn't been for the longest time...
int32_t hslot = s_loadTable.getSlot ( &urlIp );
// scan all proxies that have this urlip outstanding
for ( int32_t i = hslot ; i >= 0 ; i = s_loadTable.getNextSlot(i,&urlIp)){
// get the bucket
LoadBucket *lb;
lb = (LoadBucket *)s_loadTable.getValueFromSlot(i);
// get the spider proxy this load point was for
uint64_t key = (uint32_t)lb->m_proxyIp;
key <<= 16;
key |= (uint16_t)lb->m_proxyPort;
SpiderProxy *sp = (SpiderProxy *)s_iptab.getValue(&key);
// must be there unless user remove it from the list
if ( ! sp ) continue;
// count it up
if ( lb->m_downloadEndTimeMS == 0LL )
sp->m_countForThisIp++;
// set the last time used to the most recently downloaded time
// that this proxy has downloaded from this ip
if ( lb->m_downloadEndTimeMS &&
lb->m_downloadEndTimeMS > sp->m_lastTimeUsedForThisIp )
sp->m_lastTimeUsedForThisIp = lb->m_downloadEndTimeMS;
}
// first try to get a spider proxy that is not "dead"
bool skipDead = true;
int32_t numBannedProxies = 0;
int32_t aliveProxyCandidates = 0;
redo:
// get the min of the counts
int32_t minCount = 999999;
for ( int32_t i = 0 ; i < s_iptab.getNumSlots() ; i++ ) {
// skip empty slots
if ( ! s_iptab.m_flags[i] ) continue;
// get the spider proxy
SpiderProxy *sp = (SpiderProxy *)s_iptab.getValueFromSlot(i);
// if this proxy was banned by the url's ip... skip it. it is
// not a candidate...
if ( skipDead ) {
int64_t uip = preq->m_urlIp;
int64_t pip = sp->m_ip;
int64_t h64 = hash64h ( uip , pip );
if ( s_proxyBannedTable.isInTable ( &h64 ) ) {
numBannedProxies++;
continue;
}
}
// if it failed the last test, skip it
if ( skipDead && sp->m_lastDownloadError ) continue;
if ( skipDead ) aliveProxyCandidates++;
if ( sp->m_countForThisIp >= minCount ) continue;
minCount = sp->m_countForThisIp;
}
// all dead? then get the best dead one
if ( minCount == 999999 ) {
skipDead = false;
goto redo;
}
// . we only use one proxy if none are banned by this IP
// . when that gets banned, we will use the next 2 proxies with
// a higher backoff/crawlDelay, etc.
int32_t threshHold;
if ( numBannedProxies <= 0 ) threshHold = 1;
// if first proxy gets banned, try next 2 proxies until both get ban'd
else if ( numBannedProxies == 1 ) threshHold = 2;
else if ( numBannedProxies < 1+2) threshHold = 3 - numBannedProxies;
// if next two proxies got banned, try next 4 proxies until banned
else if ( numBannedProxies == 3 ) threshHold = 4;
else if ( numBannedProxies < 3+4) threshHold = 7 - numBannedProxies;
// if next 4 proxies got banned, try next 8 proxies until they get band
else if ( numBannedProxies == 7 ) threshHold = 8;
else if ( numBannedProxies < 7+8) threshHold = 15 - numBannedProxies;
else if ( numBannedProxies == 15) threshHold = 16;
else if ( numBannedProxies < 15+16 ) threshHold = 31-numBannedProxies;
else if ( numBannedProxies == 31 ) threshHold = 32;
else if ( numBannedProxies < 31+32)threshHold=63-numBannedProxies;
else if ( numBannedProxies == 63 ) threshHold = 64;
else if ( numBannedProxies < 63+64)threshHold=127-numBannedProxies;
else if ( numBannedProxies == 127 ) threshHold = 128;
else if ( numBannedProxies < 127+128)threshHold=255-numBannedProxies;
else if ( numBannedProxies == 255 ) threshHold = 256;
else if ( numBannedProxies < 255+256)threshHold=512-numBannedProxies;
else if ( numBannedProxies == 511 ) threshHold = 512;
else if ( numBannedProxies < 511+512)threshHold=1024-numBannedProxies;
else threshHold = 1024;
if ( threshHold <= 0 ) {
log("proxy: spiderproxy error in threshold of %" PRId32" "
"for banned=%" PRId32,threshHold,numBannedProxies);
threshHold = 1;
}
// reset minCount so we can take the min over those we check here
minCount = -1;
int64_t oldest = 0x7fffffffffffffffLL;
SpiderProxy *winnersp = NULL;
int32_t count = 0;
// start at a random slot based on url's IP so we don't
// overload the first proxy
int32_t start = ((uint32_t)urlIp) % s_iptab.getNumSlots();
int32_t slotCount = s_iptab.getNumSlots();
// . now find the best proxy wih the minCount
for ( int32_t i = start ; ; i++ ) {
// scan all slots in hash table, then stop
if ( slotCount-- <= 0 ) break;
// wrap around to zero if we hit the end
if ( i == s_iptab.getNumSlots() ) i = 0;
// skip empty slots
if ( ! s_iptab.m_flags[i] ) continue;
// get the spider proxy
SpiderProxy *sp = (SpiderProxy *)s_iptab.getValueFromSlot(i);
// if it failed the last test, skip it... not here...
if ( skipDead && sp->m_lastDownloadError ) continue;
// if this proxy was banned by the url's ip... skip it. it is
// not a candidate...
if ( skipDead ) {
int64_t uip = preq->m_urlIp;
int64_t pip = sp->m_ip;
int64_t h64 = hash64h ( uip , pip );
if ( s_proxyBannedTable.isInTable ( &h64 ) ) continue;
}
// if some proxies are "alive" then only pick from
// the first half of the proxies that are alive (i.e. still
// work). that way, when one of those goes dead we will inc
// the backoff (crawldelay) and a new proxy that we haven't
// used for this url's IP will take it's place. and such
// new proxies will only have the new backoff count used
// through them. that way, we don't get ALL of our proxies
// banned at about the same time since we do somewhat uniform
// load balancing over them.
if ( skipDead && count >= threshHold)//aliveProxyCandidates/2 )
continue;
// count the alive/non-banned candidates
count++;
// if all hosts were "dead" because they all had
// m_lastDownloadError set then minCount will be 999999
// and nobody should continue from this statement:
if ( sp->m_countForThisIp > minCount && minCount>=0 ) continue;
// then go by last download time for this ip
if ( sp->m_countForThisIp == minCount && minCount>=0 &&
sp->m_lastTimeUsedForThisIp >= oldest )
continue;
// pick the spider proxy used longest ago
oldest = sp->m_lastTimeUsedForThisIp;
minCount = sp->m_countForThisIp;
// got a new winner
winnersp = sp;
}
// we must have a winner
if ( ! winnersp ) { g_process.shutdownAbort(true); }
int64_t nowms = gettimeofdayInMillisecondsLocal();
// add a new load bucket then!
LoadBucket bb;
bb.m_urlIp = urlIp;
// the time it started
bb.m_downloadStartTimeMS = nowms;
// download has not ended yet
bb.m_downloadEndTimeMS = 0LL;
// the host using the proxy
bb.m_hostId = udpSlot->getHostId();
// key is this for m_prTable
bb.m_proxyIp = winnersp->m_ip;
bb.m_proxyPort = winnersp->m_port;
// a new id. we use this to update the downloadEndTime when done
static int32_t s_lbid = 0;
// add it now
bb.m_id = s_lbid++;
s_loadTable.addKey ( &urlIp , &bb );
// winner count update
winnersp->m_timesUsed++;
// sanity
if ( (int32_t)sizeof(ProxyReply) > TMPBUFSIZE ){g_process.shutdownAbort(true);}
// and give proxy ip/port back to the requester so they can
// use that to download their url
ProxyReply *prep = (ProxyReply *)udpSlot->m_tmpBuf;
prep->m_proxyIp = winnersp->m_ip;
prep->m_proxyPort = winnersp->m_port;
// this is just '\0' if none
strcpy(prep->m_usernamePwd,winnersp->m_usernamePwd);
// do not count the proxy we are returning as "more"
prep->m_hasMoreProxiesToTry = ( aliveProxyCandidates > 1 );
// and the loadbucket id, so requester can tell us it is done
// downloading through the proxy and we can update the LoadBucket
// for this transaction (m_lbId)
prep->m_lbId = bb.m_id;
// requester wants to know how many proxies have been banned by the
// urlIp so it can increase a self-imposed crawl-delay to be more
// sensitive to the spider policy.
prep->m_numBannedProxies = numBannedProxies;
//char *p = udpSlot->m_tmpBuf;
//*(int32_t *)p = winnersp->m_ip ; p += 4;
//*(int16_t *)p = winnersp->m_port; p += 2;
// and the loadbucket id
//*(int32_t *)p = bb.m_id; p += 4;
// with dup keys we end up with long chains of crap and this
// takes forever. so just flush the whole thing every 2 minutes AND
// when 20000+ entries are in there
static time_t s_lastTime = 0;
time_t now = nowms / 1000;
if ( s_lastTime == 0 ) s_lastTime = now;
time_t elapsed = now - s_lastTime;
if ( elapsed > 120 && s_loadTable.getNumSlots() > 10000 ) {
log("sproxy: flushing %i entries from proxy loadtable that "
"have accumulated since %i seconds ago",
(int)s_loadTable.m_numSlotsUsed,(int)elapsed);
s_loadTable.clear();
// only do this one per minute
s_lastTime = now;
}
int32_t sanityCount = 0;//s_loadTable.getNumSlots();
// top:
// now remove old entries from the load table. entries that
// have completed and have a download end time more than 10 mins ago.
for ( int32_t i = s_loadTable.getNumSlots() - 1 ; i >= 0 ; i-- ) {
// skip if empty
if ( ! s_loadTable.m_flags[i] ) continue;
// get the bucket
LoadBucket *pp =(LoadBucket *)s_loadTable.getValueFromSlot(i);
// skip if still active
if ( pp->m_downloadEndTimeMS == 0LL ) continue;
// delta t
int64_t took = nowms - pp->m_downloadEndTimeMS;
// < 10 mins? now it's < 15 seconds to prevent clogging.
if ( took < LOADPOINT_EXPIRE_MS ) continue;
// 100 at a time so we don't slam cpu
if ( sanityCount++ > 100 ) break;
// ok, its too old, nuke it to save memory
s_loadTable.removeSlot(i);
// the keys might have buried us but we really should not
// mis out on analyzing any keys if we just keep looping here
// should we? TODO: figure it out. if we miss a few it's not
// a big deal.
//i--;
//goto top;
}
// send the proxy ip/port/LBid back to user
g_udpServer.sendReply(udpSlot->m_tmpBuf, sizeof(ProxyReply), udpSlot->m_tmpBuf, sizeof(ProxyReply), udpSlot);
}
