transmission/libtransmission/tracker.c

1104 lines
30 KiB
C

/******************************************************************************
* $Id$
*
* Copyright (c) 2005-2006 Transmission authors and contributors
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*****************************************************************************/
#include "transmission.h"
#include "shared.h"
typedef struct tr_announce_list_ptr_s tr_announce_list_ptr_t;
struct tr_announce_list_ptr_s
{
tr_tracker_info_t * item;
tr_announce_list_ptr_t * nextItem;
};
struct tr_tracker_s
{
tr_torrent_t * tor;
char * id;
char * trackerid;
tr_tracker_info_t * info;
tr_announce_list_ptr_t ** trackerAnnounceListPtr;
#define TC_CHANGE_NO 0
#define TC_CHANGE_NEXT 1
#define TC_CHANGE_NONEXT 2
#define TC_CHANGE_REDIRECT 4
int shouldChangeAnnounce;
int announceTier;
int announceTierLast;
char * redirectAddress;
int redirectAddressLen;
char * redirectScrapeAddress;
int redirectScrapeAddressLen;
char started;
char completed;
char stopped;
int interval;
int minInterval;
int scrapeInterval;
int seeders;
int leechers;
int hasManyPeers;
int complete;
int randOffset;
int completelyUnconnectable;
int allUnreachIfError;
int lastError;
uint64_t dateTry;
uint64_t dateOk;
uint64_t dateScrape;
int lastScrapeFailed;
int scrapeNeeded;
tr_http_t * http;
tr_http_t * httpScrape;
int publicPort;
};
static void setAnnounce ( tr_tracker_t * tc, tr_announce_list_ptr_t * announceItem );
static void failureAnnouncing( tr_tracker_t * tc );
static tr_http_t * getQuery ( tr_tracker_t * tc );
static tr_http_t * getScrapeQuery ( tr_tracker_t * tc );
static void readAnswer ( tr_tracker_t * tc, const char *, int,
int * peerCount, uint8_t ** peerCompact );
static void readScrapeAnswer ( tr_tracker_t * tc, const char *, int );
static void killHttp ( tr_http_t ** http );
static int shouldChangePort( tr_tracker_t * tc );
tr_tracker_t * tr_trackerInit( tr_torrent_t * tor )
{
tr_info_t * inf = &tor->info;
tr_tracker_t * tc;
tr_announce_list_ptr_t * prev, * cur;
int ii, jj;
tc = calloc( 1, sizeof( tr_tracker_t ) );
tc->tor = tor;
tc->id = tor->id;
tc->started = 1;
tc->shouldChangeAnnounce = TC_CHANGE_NO;
tc->redirectAddress = NULL;
tc->interval = 300;
tc->scrapeInterval = 1200;
tc->lastError = 0;
tc->allUnreachIfError = 1;
tc->publicPort = tor->publicPort;
tc->trackerAnnounceListPtr = calloc( sizeof( int ), inf->trackerTiers );
for( ii = 0; ii < inf->trackerTiers; ii++ )
{
prev = NULL;
for( jj = 0; jj < inf->trackerList[ii].count; jj++ )
{
cur = calloc( sizeof( tr_announce_list_ptr_t ), 1 );
cur->item = &inf->trackerList[ii].list[jj];
if( NULL == prev )
{
tc->trackerAnnounceListPtr[ii] = cur;
}
else
{
prev->nextItem = cur;
}
prev = cur;
}
}
setAnnounce( tc, tc->trackerAnnounceListPtr[0] );
return tc;
}
static void setAnnounce( tr_tracker_t * tc, tr_announce_list_ptr_t * announcePtr )
{
tc->info = announcePtr->item;
/* Needs a new scrape */
tc->seeders = -1;
tc->leechers = -1;
tc->complete = -1;
tc->dateScrape = 0;
}
static void failureAnnouncing( tr_tracker_t * tc )
{
tr_info_t * inf = &tc->tor->info;
tc->shouldChangeAnnounce = tc->announceTier + 1 < inf->trackerTiers
|| tc->announceTierLast + 1 < inf->trackerList[tc->announceTier].count
? TC_CHANGE_NEXT : TC_CHANGE_NONEXT;
if( tc->shouldChangeAnnounce == TC_CHANGE_NONEXT )
{
tc->completelyUnconnectable = 1;
}
}
static int shouldConnect( tr_tracker_t * tc )
{
tr_torrent_t * tor = tc->tor;
uint64_t now;
/* Last tracker failed, try next */
if( tc->shouldChangeAnnounce == TC_CHANGE_NEXT
|| tc->shouldChangeAnnounce == TC_CHANGE_REDIRECT )
{
return 1;
}
now = tr_date();
/* If last attempt was an error and it did not change trackers, then all must have been errors */
if( tc->lastError )
{
/* Unreachable trackers, wait 10 seconds + random value before trying again */
if( tc->allUnreachIfError )
{
if( now < tc->dateTry + tc->randOffset + 10000 )
{
return 0;
}
}
/* The tracker rejected us (like 4XX code, unauthorized IP...),
don't hammer it - we'll probably get the same answer next time anyway */
else
{
if( now < tc->dateTry + 1000 * tc->interval + tc->randOffset )
{
return 0;
}
else
{
tc->allUnreachIfError = 1; //since starting at the top of the list, reset if any were reached previously
}
}
}
/* Do we need to send an event? */
if( tc->started || tc->completed || tc->stopped || shouldChangePort( tc ) )
{
return 1;
}
/* Should we try and get more peers? */
if( now > tc->dateOk + 1000 * tc->interval + tc->randOffset )
{
return 1;
}
/* If there is quite a lot of people on this torrent, stress
the tracker a bit until we get a decent number of peers */
if( tc->hasManyPeers && !tr_cpIsSeeding( tor->completion ) )
{
/* reannounce in 10 seconds if we have less than 5 peers */
if( tor->peerCount < 5 )
{
if( now > tc->dateOk + 1000 * MAX( 10, tc->minInterval ) )
{
return 1;
}
}
/* reannounce in 20 seconds if we have less than 10 peers */
else if( tor->peerCount < 10 )
{
if( now > tc->dateOk + 1000 * MAX( 20, tc->minInterval ) )
{
return 1;
}
}
/* reannounce in 30 seconds if we have less than 15 peers */
else if( tor->peerCount < 15 )
{
if( now > tc->dateOk + 1000 * MAX( 30, tc->minInterval ) )
{
return 1;
}
}
}
return 0;
}
static int shouldScrape( tr_tracker_t * tc )
{
uint64_t now, interval;
/* in process of changing tracker or scrape not supported */
if( tc->shouldChangeAnnounce != TC_CHANGE_NO ||
NULL == tc->info->scrape || tc->stopped )
{
return 0;
}
now = tr_date();
interval = 1000 * MAX( tc->scrapeInterval, 600 );
/* scrape more often if needed */
if( tc->scrapeNeeded || tc->lastScrapeFailed )
{
interval /= 2;
}
return now > tc->dateScrape + interval;
}
void tr_trackerAnnouncePulse( tr_tracker_t * tc, int * peerCount,
uint8_t ** peerCompact, int manual )
{
tr_torrent_t * tor = tc->tor;
tr_info_t * inf = &tor->info;
const char * data;
char * address, * announce;
int len, i, port;
tr_announce_list_ptr_t * announcePtr, * prevAnnouncePtr;
*peerCount = 0;
*peerCompact = NULL;
if( ( NULL == tc->http ) && ( manual || shouldConnect( tc ) ) )
{
//if announcing manually, don't consider not reaching a tracker an error
if( manual )
{
tc->allUnreachIfError = 0;
}
tc->completelyUnconnectable = 0;
tc->randOffset = tr_rand( 60000 );
tc->dateTry = tr_date();
/* Use redirected address */
if( tc->shouldChangeAnnounce == TC_CHANGE_REDIRECT )
{
if( !tr_httpParseUrl( tc->redirectAddress, tc->redirectAddressLen,
&address, &port, &announce ) )
{
tr_err( "Tracker: redirected URL: %s:%d", address, port );
tc->http = tr_httpClient( TR_HTTP_GET, address, port, "%s", announce );
free( address );
free( announce );
}
free( tc->redirectAddress );
tc->redirectAddress = NULL;
}
else
{
/* Need to change to next address in list */
if( tc->shouldChangeAnnounce == TC_CHANGE_NEXT )
{
tr_inf( "Tracker: failed to connect to %s, trying next",
tc->info->address );
if( tc->announceTierLast + 1 < inf->trackerList[tc->announceTier].count )
{
tc->announceTierLast++;
announcePtr = tc->trackerAnnounceListPtr[tc->announceTier];
for( i = 0; i < tc->announceTierLast; i++ )
{
announcePtr = announcePtr->nextItem;
}
}
else
{
tc->announceTierLast = 0;
tc->announceTier++;
announcePtr = tc->trackerAnnounceListPtr[tc->announceTier];
}
tr_inf( "Tracker: tracker address set to %s", tc->trackerAnnounceListPtr[tc->announceTier]->item->address );
setAnnounce( tc, announcePtr );
}
/* Need to change to first in list */
else if( tc->announceTier != 0 || tc->announceTierLast != 0 )
{
/* Check if the last announce was successful and wasn't the first in the sublist */
if( tc->shouldChangeAnnounce == TC_CHANGE_NO && tc->announceTierLast != 0 )
{
announcePtr = tc->trackerAnnounceListPtr[tc->announceTier];
prevAnnouncePtr = NULL;
for( i = 0; i < tc->announceTierLast; i++ )
{
prevAnnouncePtr = announcePtr;
announcePtr = announcePtr->nextItem;
}
/* Move address to front of tier in announce list */
prevAnnouncePtr->nextItem = announcePtr->nextItem;
announcePtr->nextItem = tc->trackerAnnounceListPtr[tc->announceTier];
tc->trackerAnnounceListPtr[tc->announceTier] = announcePtr;
}
setAnnounce( tc, tc->trackerAnnounceListPtr[0] );
tc->announceTier = 0;
tc->announceTierLast = 0;
}
tc->http = getQuery( tc );
tr_inf( "Tracker: connecting to %s:%d (%s)",
tc->info->address, tc->info->port,
tc->started ? "sending 'started'" :
( tc->completed ? "sending 'completed'" :
( tc->stopped ? "sending 'stopped'" :
( shouldChangePort( tc ) ? "sending 'stopped' to change port" :
"getting peers" ) ) ) );
}
tc->shouldChangeAnnounce = TC_CHANGE_NO;
}
if( NULL != tc->http )
{
switch( tr_httpPulse( tc->http, &data, &len ) )
{
case TR_NET_WAIT:
break;
case TR_NET_ERROR:
killHttp( &tc->http );
tc->dateTry = tr_date();
failureAnnouncing( tc );
tc->lastError = 1;
break;
case TR_NET_OK:
readAnswer( tc, data, len, peerCount, peerCompact );
killHttp( &tc->http );
break;
}
}
if( ( NULL == tc->httpScrape ) && shouldScrape( tc ) )
{
tc->dateScrape = tr_date();
if ( tc->redirectScrapeAddress != NULL )
{
/* Use redirected address */
if( !tr_httpParseUrl( tc->redirectScrapeAddress, tc->redirectScrapeAddressLen,
&address, &port, &announce ) )
{
tr_err( "Scrape: redirected URL: %s:%d", address, port );
tc->httpScrape = tr_httpClient( TR_HTTP_GET, address, port, "%s", announce );
free( address );
free( announce );
}
free( tc->redirectScrapeAddress );
tc->redirectScrapeAddress = NULL;
}
else
{
tc->httpScrape = getScrapeQuery( tc );
tr_inf( "Scrape: sent HTTP request to %s:%d%s",
tc->info->address, tc->info->port, tc->info->scrape );
}
}
if( NULL != tc->httpScrape )
{
switch( tr_httpPulse( tc->httpScrape, &data, &len ) )
{
case TR_NET_WAIT:
break;
case TR_NET_ERROR:
killHttp( &tc->httpScrape );
tc->lastScrapeFailed = 1;
break;
case TR_NET_OK:
readScrapeAnswer( tc, data, len );
killHttp( &tc->httpScrape );
break;
}
}
return;
}
void tr_trackerCompleted( tr_tracker_t * tc )
{
tc->started = 0;
tc->completed = 1;
tc->stopped = 0;
}
void tr_trackerStopped( tr_tracker_t * tc )
{
/* If we are already sending a query at the moment, we need to
reconnect */
killHttp( &tc->http );
tc->started = 0;
tc->completed = 0;
tc->stopped = 1;
/* Even if we have connected recently, reconnect right now */
tc->dateTry = 0;
}
void tr_trackerClose( tr_tracker_t * tc )
{
tr_torrent_t * tor = tc->tor;
tr_info_t * inf = &tor->info;
tr_announce_list_ptr_t * cur, * curFree;
int ii;
killHttp( &tc->http );
killHttp( &tc->httpScrape );
for( ii = 0; ii < inf->trackerTiers; ii++ )
{
for( cur = tc->trackerAnnounceListPtr[ii]; cur; )
{
curFree = cur;
cur = cur->nextItem;
free( curFree );
}
}
free( tc->trackerAnnounceListPtr );
free( tc->trackerid );
free( tc );
}
static tr_http_t * getQuery( tr_tracker_t * tc )
{
tr_torrent_t * tor = tc->tor;
char * event, * trackerid, * idparam;
uint64_t left;
uint64_t down;
uint64_t up;
char start;
int numwant = 50;
down = tor->downloadedCur;
up = tor->uploadedCur;
if( tc->started )
{
event = "&event=started";
down = 0;
up = 0;
if( shouldChangePort( tc ) )
{
tc->publicPort = tor->publicPort;
}
}
else if( tc->completed )
{
event = "&event=completed";
}
else if( tc->stopped || shouldChangePort( tc ) )
{
event = "&event=stopped";
numwant = 0;
}
else
{
event = "";
}
if( NULL == tc->trackerid )
{
trackerid = "";
idparam = "";
}
else
{
trackerid = tc->trackerid;
idparam = "&trackerid=";
}
start = ( strchr( tc->info->announce, '?' ) ? '&' : '?' );
left = tr_cpLeftBytes( tor->completion );
return tr_httpClient( TR_HTTP_GET, tc->info->address, tc->info->port,
"%s%c"
"info_hash=%s&"
"peer_id=%s&"
"port=%d&"
"uploaded=%"PRIu64"&"
"downloaded=%"PRIu64"&"
"left=%"PRIu64"&"
"compact=1&"
"numwant=%d&"
"key=%s"
"%s%s"
"%s",
tc->info->announce, start, tor->escapedHashString,
tc->id, tc->publicPort, up, down, left, numwant,
tor->key, idparam, trackerid, event );
}
static tr_http_t * getScrapeQuery( tr_tracker_t * tc )
{
tr_torrent_t * tor = tc->tor;
char start;
start = ( strchr( tc->info->scrape, '?' ) ? '&' : '?' );
return tr_httpClient( TR_HTTP_GET, tc->info->address, tc->info->port,
"%s%c"
"info_hash=%s",
tc->info->scrape, start, tor->escapedHashString );
}
static void readAnswer( tr_tracker_t * tc, const char * data, int len,
int * _peerCount, uint8_t ** _peerCompact )
{
tr_torrent_t * tor = tc->tor;
int i;
int code;
benc_val_t beAll;
benc_val_t * bePeers, * beFoo;
const uint8_t * body;
int bodylen, shouldfree, scrapeNeeded;
char * address;
int peerCount;
uint8_t * peerCompact;
*_peerCount = peerCount = 0;
*_peerCompact = peerCompact = NULL;
tc->dateTry = tr_date();
code = tr_httpResponseCode( data, len );
if( 0 > code )
{
/* We don't have a valid HTTP status line */
tr_inf( "Tracker: invalid HTTP status line" );
tc->lastError = 1;
failureAnnouncing( tc );
return;
}
if( code == 301 || code == 302 )
{
tr_http_header_t hdr[] = { { "Location", NULL, 0 }, { NULL, NULL, 0 } };
tr_err( "Tracker: HTTP status code: %i", code );
tr_httpParse( data, len, hdr );
address = calloc( sizeof( char ), hdr->len+1 );
snprintf( address, hdr->len+1, "%s", hdr->data );
tc->shouldChangeAnnounce = TC_CHANGE_REDIRECT;
tc->redirectAddress = address;
tc->redirectAddressLen = hdr->len;
return;
}
if( !TR_HTTP_STATUS_OK( code ) )
{
/* we didn't get a 2xx status code */
tr_err( "Tracker: invalid HTTP status code: %i", code );
tc->lastError = 1;
tc->allUnreachIfError = 0;
failureAnnouncing( tc );
return;
}
/* find the end of the http headers */
body = (uint8_t *) tr_httpParse( data, len, NULL );
if( NULL == body )
{
tr_err( "Tracker: could not find end of HTTP headers" );
tc->lastError = 1;
failureAnnouncing( tc );
return;
}
bodylen = len - ( body - (const uint8_t*)data );
/* Find and load the dictionary */
shouldfree = 0;
for( i = 0; i < bodylen; i++ )
{
if( !tr_bencLoad( &body[i], bodylen - i, &beAll, NULL ) )
{
shouldfree = 1;
break;
}
}
if( i >= bodylen )
{
if( tc->stopped || shouldChangePort( tc ) )
{
tc->lastError = 0;
goto nodict;
}
tr_err( "Tracker: no valid dictionary found in answer" );
tc->lastError = 1;
tc->allUnreachIfError = 0;
failureAnnouncing( tc );
return;
}
// tr_bencPrint( &beAll );
if( ( bePeers = tr_bencDictFind( &beAll, "failure reason" ) ) )
{
tr_err( "Tracker: Error - %s", bePeers->val.s.s );
tor->error = TR_ERROR_TC_ERROR;
snprintf( tor->errorString, sizeof( tor->errorString ),
"%s", bePeers->val.s.s );
tc->lastError = 1;
tc->allUnreachIfError = 0;
failureAnnouncing( tc );
goto cleanup;
}
else if( ( bePeers = tr_bencDictFind( &beAll, "warning message" ) ) )
{
tr_err( "Tracker: Warning - %s", bePeers->val.s.s );
tor->error = TR_ERROR_TC_WARNING;
snprintf( tor->errorString, sizeof( tor->errorString ),
"%s", bePeers->val.s.s );
}
else if( tor->error & TR_ERROR_TC_MASK )
{
tor->error = TR_OK;
}
tc->lastError = 0;
tc->allUnreachIfError = 0;
/* Get the tracker interval, force to between
10 sec and 5 mins */
beFoo = tr_bencDictFind( &beAll, "interval" );
if( !beFoo || TYPE_INT != beFoo->type )
{
tr_err( "Tracker: no 'interval' field" );
goto cleanup;
}
tc->interval = beFoo->val.i;
tr_inf( "Tracker: interval = %d seconds", tc->interval );
tc->interval = MAX( 10, tc->interval );
/* Get the tracker minimum interval, force to between
10 sec and 5 mins */
beFoo = tr_bencDictFind( &beAll, "min interval" );
if( beFoo && TYPE_INT == beFoo->type )
{
tc->minInterval = beFoo->val.i;
tr_inf( "Tracker: min interval = %d seconds", tc->minInterval );
tc->minInterval = MAX( 10, tc->minInterval );
if( tc->interval < tc->minInterval )
{
tc->interval = tc->minInterval;
tr_inf( "Tracker: 'interval' less than 'min interval', "
"using 'min interval'" );
}
}
else
{
tc->minInterval = 0;
tr_inf( "Tracker: no 'min interval' field" );
}
scrapeNeeded = 0;
beFoo = tr_bencDictFind( &beAll, "complete" );
if( beFoo && TYPE_INT == beFoo->type )
{
tc->seeders = beFoo->val.i;
}
else
{
scrapeNeeded = 1;
}
beFoo = tr_bencDictFind( &beAll, "incomplete" );
if( beFoo && TYPE_INT == beFoo->type )
{
tc->leechers = beFoo->val.i;
}
else
{
scrapeNeeded = 1;
}
tc->scrapeNeeded = scrapeNeeded;
if( !scrapeNeeded )
{
tc->hasManyPeers = ( tc->seeders + tc->leechers >= 50 );
}
beFoo = tr_bencDictFind( &beAll, "tracker id" );
if( beFoo )
{
free( tc->trackerid );
tc->trackerid = strdup( beFoo->val.s.s );
tr_inf( "Tracker: tracker id = %s", tc->trackerid );
}
bePeers = tr_bencDictFind( &beAll, "peers" );
if( !bePeers )
{
if( tc->stopped || shouldChangePort( tc ) )
{
goto nodict;
}
tr_err( "Tracker: no \"peers\" field" );
failureAnnouncing( tc );
goto cleanup;
}
if( bePeers->type & TYPE_LIST )
{
/* Original protocol */
if( bePeers->val.l.count > 0 )
{
struct in_addr addr;
in_port_t port;
peerCount = 0;
peerCompact = malloc( 6 * bePeers->val.l.count );
/* Convert to compact form */
for( i = 0; i < bePeers->val.l.count; i++ )
{
if( !( beFoo = tr_bencDictFind(
&bePeers->val.l.vals[i], "ip" ) ) )
continue;
if( tr_netResolve( beFoo->val.s.s, &addr ) )
continue;
memcpy( &peerCompact[6 * peerCount], &addr, 4 );
if( !( beFoo = tr_bencDictFind(
&bePeers->val.l.vals[i], "port" ) ) )
continue;
port = htons( beFoo->val.i );
memcpy( &peerCompact[6 * peerCount + 4], &port, 2 );
peerCount++;
}
}
}
else if( bePeers->type & TYPE_STR )
{
/* "Compact" extension */
if( bePeers->val.s.i >= 6 )
{
peerCount = bePeers->val.s.i / 6;
peerCompact = malloc( bePeers->val.s.i );
memcpy( peerCompact, bePeers->val.s.s, bePeers->val.s.i );
}
}
if( peerCount > 0 )
{
tr_inf( "Tracker: got %d peers", peerCount );
if( peerCount >= 50 )
{
tc->hasManyPeers = 1;
}
*_peerCount = peerCount;
*_peerCompact = peerCompact;
}
nodict:
/* Success */
tc->started = 0;
tc->completed = 0;
tc->dateOk = tr_date();
if( tc->stopped )
{
tor->status = TR_STATUS_STOPPED;
tc->stopped = 0;
}
else if( shouldChangePort( tc ) )
{
tc->started = 1;
}
cleanup:
if( shouldfree )
{
tr_bencFree( &beAll );
}
}
static void readScrapeAnswer( tr_tracker_t * tc, const char * data, int len )
{
int code;
const uint8_t * body;
int bodylen, ii;
benc_val_t scrape, * val1, * val2;
char * address;
code = tr_httpResponseCode( data, len );
if( 0 > code )
{
/* We don't have a valid HTTP status line */
tr_inf( "Scrape: invalid HTTP status line" );
tc->lastScrapeFailed = 1;
return;
}
if( code == 301 || code == 302 )
{
tr_http_header_t hdr[] = { { "Location", NULL, 0 }, { NULL, NULL, 0 } };
tr_err( "Scrape: HTTP status code: %i", code );
tr_httpParse( data, len, hdr );
address = calloc( sizeof( char ), hdr->len+1 );
snprintf( address, hdr->len+1, "%s", hdr->data );
/* Needs a new scrape */
tc->dateScrape = 0;
tc->redirectScrapeAddress = address;
tc->redirectScrapeAddressLen = hdr->len;
return;
}
if( !TR_HTTP_STATUS_OK( code ) )
{
/* we didn't get a 2xx status code */
tr_err( "Scrape: invalid HTTP status code: %i", code );
if( TR_HTTP_STATUS_FAIL_CLIENT( code ) )
{
free( tc->info->scrape );
tc->info->scrape = NULL;
}
tc->lastScrapeFailed = 1;
return;
}
/* find the end of the http headers */
body = (uint8_t *) tr_httpParse( data, len, NULL );
if( NULL == body )
{
tr_err( "Scrape: could not find end of HTTP headers" );
tc->lastScrapeFailed = 1;
return;
}
tc->lastScrapeFailed = 0;
bodylen = len - ( body - (const uint8_t*)data );
for( ii = 0; ii < bodylen; ii++ )
{
if( !tr_bencLoad( body + ii, bodylen - ii, &scrape, NULL ) )
{
break;
}
}
if( ii >= bodylen )
{
return;
}
val1 = tr_bencDictFind( &scrape, "files" );
if( !val1 || val1->type != TYPE_DICT || val1->val.l.count < 1 )
{
tr_bencFree( &scrape );
return;
}
val1 = &val1->val.l.vals[1];
val2 = tr_bencDictFind( val1, "complete" );
if( !val2 )
{
tr_bencFree( &scrape );
return;
}
tc->seeders = val2->val.i;
val2 = tr_bencDictFind( val1, "incomplete" );
if( !val2 )
{
tr_bencFree( &scrape );
return;
}
tc->leechers = val2->val.i;
val2 = tr_bencDictFind( val1, "downloaded" );
if( !val2 )
{
tr_bencFree( &scrape );
return;
}
tc->complete = val2->val.i;
val2 = tr_bencDictFind( &scrape, "flags" );
if( val2 )
{
val2 = tr_bencDictFind( val2, "min_request_interval" );
if( val2 )
{
tc->scrapeInterval = val2->val.i;
tr_inf( "Scrape: min_request_interval = %d seconds", tc->scrapeInterval );
}
}
tc->hasManyPeers = ( tc->seeders + tc->leechers >= 50 );
tr_bencFree( &scrape );
}
int tr_trackerSeeders( tr_tracker_t * tc )
{
if( !tc )
{
return -1;
}
return tc->seeders;
}
int tr_trackerLeechers( tr_tracker_t * tc )
{
if( !tc )
{
return -1;
}
return tc->leechers;
}
int tr_trackerDownloaded( tr_tracker_t * tc )
{
if( !tc )
{
return -1;
}
return tc->complete;
}
tr_tracker_info_t * tr_trackerGet( tr_tracker_t * tc )
{
if( !tc )
{
return NULL;
}
return tc->info;
}
int tr_trackerCannotConnect( tr_tracker_t * tc )
{
if( !tc )
{
return 0;
}
return tc->completelyUnconnectable;
}
/* Blocking version */
int tr_trackerScrape( tr_torrent_t * tor, int * s, int * l, int * d )
{
tr_tracker_t * tc;
tr_http_t * http;
const char * data;
int len;
int ret;
tc = tr_trackerInit( tor );
if( NULL == tc->info->scrape )
{
return 1;
}
http = getScrapeQuery( tc );
for( data = NULL; !data; tr_wait( 10 ) )
{
switch( tr_httpPulse( http, &data, &len ) )
{
case TR_NET_WAIT:
break;
case TR_NET_ERROR:
goto scrapeDone;
case TR_NET_OK:
readScrapeAnswer( tc, data, len );
goto scrapeDone;
}
}
scrapeDone:
tr_httpClose( http );
ret = 1;
if( tc->seeders > -1 && tc->leechers > -1 && tc->complete > -1 )
{
*s = tc->seeders;
*l = tc->leechers;
*d = tc->complete;
ret = 0;
}
tr_trackerClose( tc );
return ret;
}
static void killHttp( tr_http_t ** http )
{
if( NULL != *http )
{
tr_httpClose( *http );
*http = NULL;
}
}
static int shouldChangePort( tr_tracker_t * tc )
{
tr_torrent_t * tor = tc->tor;
return ( tor->publicPort != tc->publicPort );
}