transmission/libtransmission/tr-dht.c

/*
Copyright (c) 2009 by Juliusz Chroboczek

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.
*/

/* ansi */
#include <errno.h>
#include <stdio.h>

/* posix */
#include <netinet/in.h> /* sockaddr_in */
#include <signal.h> /* sig_atomic_t */
#include <sys/time.h>
#include <sys/types.h>
#include <sys/socket.h> /* socket(), bind() */
#include <netdb.h>
#include <unistd.h> /* close() */

/* third party */
#include <event.h>
#include <dht/dht.h>

/* libT */
#include "transmission.h"
#include "bencode.h"
#include "crypto.h"
#include "net.h"
#include "peer-mgr.h" /* tr_peerMgrCompactToPex() */
#include "platform.h" /* tr_threadNew() */
#include "session.h"
#include "torrent.h" /* tr_torrentFindFromHash() */
#include "tr-dht.h"
#include "trevent.h" /* tr_runInEventThread() */
#include "utils.h"
#include "version.h"

static int dht_socket = -1, dht6_socket = -1;
static struct event dht_event, dht6_event;
static tr_port dht_port;
static unsigned char myid[20];
static tr_session *session = NULL;

static void event_callback(int s, short type, void *ignore);

struct bootstrap_closure {
    tr_session *session;
    uint8_t *nodes;
    uint8_t *nodes6;
    size_t len, len6;
};

static int
bootstrap_done( tr_session *session, int af )
{
    int status;

    if(af == 0)
        return
            bootstrap_done(session, AF_INET) &&
            bootstrap_done(session, AF_INET6);

    status = tr_dhtStatus(session, af, NULL);
    return status == TR_DHT_STOPPED || status >= TR_DHT_FIREWALLED;
}

static void
nap( int roughly )
{
    struct timeval tv;
    tv.tv_sec = roughly / 2 + tr_cryptoWeakRandInt( roughly );
    tv.tv_usec = tr_cryptoWeakRandInt( 1000000 );
    select( 0, NULL, NULL, NULL, &tv );
}

static int
bootstrap_af(tr_session *session)
{
    if( bootstrap_done(session, AF_INET6) )
        return AF_INET;
    else if ( bootstrap_done(session, AF_INET) )
        return AF_INET6;
    else
        return 0;
}

static void
bootstrap_from_name( const char *name, short int port, int af )
{
    struct addrinfo hints, *info, *infop;
    char pp[10];
    int rc;

    memset(&hints, 0, sizeof(hints));
    hints.ai_socktype = SOCK_DGRAM;
    hints.ai_family = af;
    /* No, just passing p + 1 to gai won't work. */
    tr_snprintf(pp, sizeof(pp), "%d", port);

    rc = getaddrinfo(name, pp, &hints, &info);
    if(rc != 0) {
        tr_nerr("DHT", "%s:%s: %s", name, pp, gai_strerror(rc));
        return;
    }

    infop = info;
    while(infop) {
        dht_ping_node(infop->ai_addr, infop->ai_addrlen);

        nap(15);

        if(bootstrap_done(session, af))
            break;
        infop = infop->ai_next;
    }
    freeaddrinfo(info);
}

static void
dht_bootstrap(void *closure)
{
    struct bootstrap_closure *cl = closure;
    int i;
    int num = cl->len / 6, num6 = cl->len6 / 18;

    if(session != cl->session)
        return;

    if(cl->len > 0)
        tr_ninf( "DHT", "Bootstrapping from %d nodes", num );

    if(cl->len6 > 0)
        tr_ninf( "DHT", "Bootstrapping from %d IPv6 nodes", num6 );


    for(i = 0; i < MAX(num, num6); i++) {
        if( i < num && !bootstrap_done(cl->session, AF_INET) ) {
            tr_port port;
            struct tr_address addr;

            memset(&addr, 0, sizeof(addr));
            addr.type = TR_AF_INET;
            memcpy(&addr.addr.addr4, &cl->nodes[i * 6], 4);
            memcpy(&port, &cl->nodes[i * 6 + 4], 2);
            port = ntohs(port);
            tr_dhtAddNode(cl->session, &addr, port, 1);
        }
        if( i < num6 && !bootstrap_done(cl->session, AF_INET6) ) {
            tr_port port;
            struct tr_address addr;

            memset(&addr, 0, sizeof(addr));
            addr.type = TR_AF_INET6;
            memcpy(&addr.addr.addr6, &cl->nodes6[i * 18], 16);
            memcpy(&port, &cl->nodes6[i * 18 + 16], 2);
            port = ntohs(port);
            tr_dhtAddNode(cl->session, &addr, port, 1);
        }

        /* Our DHT code is able to take up to 9 nodes in a row without
           dropping any.  After that, it takes some time to split buckets.
           So ping the first 8 nodes quickly, then slow down. */
        if(i < 8)
            nap(2);
        else
            nap(15);

        if(bootstrap_done( session, 0 ))
            break;
    }

    if(!bootstrap_done(cl->session, 0)) {
        char *bootstrap_file;
        FILE *f;

        bootstrap_file =
            tr_buildPath(cl->session->configDir, "dht.bootstrap", NULL);

        if(bootstrap_file)
            f = fopen(bootstrap_file, "r");
        if(f) {
            tr_ninf("DHT", "Attempting manual bootstrap");
            while(1) {
                char buf[201];
                char *p;
                int port = 0;

                p = fgets(buf, 200, f);
                if( p == NULL )
                    break;

                p = memchr(buf, ' ', strlen(buf));
                if(p != NULL)
                    port = atoi(p + 1);
                if(p == NULL || port <= 0 || port >= 0x10000) {
                    tr_nerr("DHT", "Couldn't parse %s", buf);
                    continue;
                }

                *p = '\0';

                bootstrap_from_name( buf, port, bootstrap_af(session) );

                if(bootstrap_done(cl->session, 0))
                    break;
            }
        }

        tr_free( bootstrap_file );
    }

    /* We really don't want to abuse our bootstrap nodes.
       Be glacially slow. */
    if(!bootstrap_done(cl->session, 0))
        nap(30);

    if(!bootstrap_done(cl->session, 0)) {
        tr_ninf("DHT", "Attempting bootstrap from dht.transmissionbt.com");
        bootstrap_from_name( "dht.transmissionbt.com", 6881,
                             bootstrap_af(session) );
    }

    if( cl->nodes )
        tr_free( cl->nodes );
    if( cl->nodes6 )
        tr_free( cl->nodes6 );
    tr_free( closure );
    tr_ndbg( "DHT", "Finished bootstrapping" );
}

/* BEP-32 has a rather nice explanation of why we need to bind to one
   IPv6 address, if I may say so myself. */

static int
rebind_ipv6(int force)
{
    struct sockaddr_in6 sin6;
    const unsigned char *ipv6 = tr_globalIPv6();
    static unsigned char *last_bound = NULL;
    int rc;

    if(dht6_socket < 0)
        return 0;

    if(!force &&
       ((ipv6 == NULL && last_bound == NULL) ||
        (ipv6 != NULL && last_bound != NULL &&
         memcmp(ipv6, last_bound, 16) == 0)))
        return 0;

    memset(&sin6, 0, sizeof(sin6));
    sin6.sin6_family = AF_INET6;
    if(ipv6)
        memcpy(&sin6.sin6_addr, ipv6, 16);
    sin6.sin6_port = htons(dht_port);
    rc = bind(dht6_socket, (struct sockaddr*)&sin6, sizeof(sin6));

    if(last_bound)
        free(last_bound);
    last_bound = NULL;

    if(rc >= 0 && ipv6) {
        last_bound = malloc(16);
        if(last_bound)
            memcpy(last_bound, ipv6, 16);
    }
    return rc;
}

int
tr_dhtInit(tr_session *ss, const tr_address * tr_addr)
{
    struct sockaddr_in sin;
    tr_benc benc;
    int rc;
    tr_bool have_id = FALSE;
    char * dat_file;
    uint8_t * nodes = NULL, * nodes6 = NULL;
    const uint8_t * raw;
    size_t len, len6;
    char v[5];
    struct bootstrap_closure * cl;

    if( session ) /* already initialized */
        return -1;

    dht_port = tr_sessionGetPeerPort(ss);
    if(dht_port <= 0)
        return -1;

    tr_ndbg( "DHT", "Initializing DHT" );

    dht_socket = socket(PF_INET, SOCK_DGRAM, 0);
    if(dht_socket < 0)
        goto fail;

    memset(&sin, 0, sizeof(sin));
    sin.sin_family = AF_INET;
    memcpy(&sin.sin_addr, &tr_addr->addr.addr4, sizeof (struct in_addr));
    sin.sin_port = htons(dht_port);
    rc = bind(dht_socket, (struct sockaddr*)&sin, sizeof(sin));
    if(rc < 0)
        goto fail;

    if(tr_globalIPv6()) {
        int one = 1;
        dht6_socket = socket(PF_INET6, SOCK_DGRAM, 0);
        if(dht6_socket < 0)
            goto fail;

#ifdef IPV6_V6ONLY
        /* Since we always open an IPv4 socket on the same port, this
           shouldn't matter.  But I'm superstitious. */
        setsockopt(dht6_socket, IPPROTO_IPV6, IPV6_V6ONLY,
                   &one, sizeof(one));
#endif

        rebind_ipv6(1);

    }

    if( getenv( "TR_DHT_VERBOSE" ) != NULL )
        dht_debug = stderr;

    dat_file = tr_buildPath( ss->configDir, "dht.dat", NULL );
    rc = tr_bencLoadFile( &benc, TR_FMT_BENC, dat_file );
    tr_free( dat_file );
    if(rc == 0) {
        have_id = tr_bencDictFindRaw(&benc, "id", &raw, &len);
        if( have_id && len==20 )
            memcpy( myid, raw, len );
        if( dht_socket >= 0 &&
            tr_bencDictFindRaw( &benc, "nodes", &raw, &len ) && !(len%6) ) {
                nodes = tr_memdup( raw, len );
        }
        if( dht6_socket > 0 &&
            tr_bencDictFindRaw( &benc, "nodes6", &raw, &len6 ) && !(len6%18) ) {
            nodes6 = tr_memdup( raw, len6 );
        }
        tr_bencFree( &benc );
    }

    if(nodes == NULL)
        len = 0;
    if(nodes6 == NULL)
        len6 = 0;

    if( have_id )
        tr_ninf( "DHT", "Reusing old id" );
    else {
        /* Note that DHT ids need to be distributed uniformly,
         * so it should be something truly random. */
        tr_ninf( "DHT", "Generating new id" );
        tr_cryptoRandBuf( myid, 20 );
    }

    v[0] = 'T';
    v[1] = 'R';
    v[2] = (SVN_REVISION_NUM >> 8) & 0xFF;
    v[3] = SVN_REVISION_NUM & 0xFF;
    rc = dht_init( dht_socket, dht6_socket, myid, (const unsigned char*)v );
    if(rc < 0)
        goto fail;

    session = ss;

    cl = tr_new( struct bootstrap_closure, 1 );
    cl->session = session;
    cl->nodes = nodes;
    cl->nodes6 = nodes6;
    cl->len = len;
    cl->len6 = len6;
    tr_threadNew( dht_bootstrap, cl );

    event_set( &dht_event, dht_socket, EV_READ, event_callback, NULL );
    tr_timerAdd( &dht_event, 0, tr_cryptoWeakRandInt( 1000000 ) );

    if( dht6_socket >= 0 )
    {
        event_set( &dht6_event, dht6_socket, EV_READ, event_callback, NULL );
        tr_timerAdd( &dht6_event, 0, tr_cryptoWeakRandInt( 1000000 ) );
    }

    tr_ndbg( "DHT", "DHT initialized" );

    return 1;

    fail:
    {
        const int save = errno;
        close(dht_socket);
        if( dht6_socket >= 0 )
            close(dht6_socket);
        dht_socket = dht6_socket = -1;
        session = NULL;
        tr_ndbg( "DHT", "DHT initialization failed (errno = %d)", save );
        errno = save;
    }

    return -1;
}

void
tr_dhtUninit(tr_session *ss)
{
    if(session != ss)
        return;

    tr_ndbg( "DHT", "Uninitializing DHT" );

    event_del( &dht_event );

    if( dht6_socket >= 0 )
        event_del( &dht6_event );

    /* Since we only save known good nodes, avoid erasing older data if we
       don't know enough nodes. */
    if(tr_dhtStatus(ss, AF_INET, NULL) < TR_DHT_FIREWALLED)
        tr_ninf( "DHT", "Not saving nodes, DHT not ready" );
    else {
        tr_benc benc;
        struct sockaddr_in sins[300];
        struct sockaddr_in6 sins6[300];
        char compact[300 * 6], compact6[300 * 18];
        char *dat_file;
        int i, j, num = 300, num6 = 300;
        int n = dht_get_nodes(sins, &num, sins6, &num6);

        tr_ninf( "DHT", "Saving %d (%d + %d) nodes", n, num, num6 );

        j = 0;
        for( i=0; i<num; ++i ) {
            memcpy( compact + j, &sins[i].sin_addr, 4 );
            memcpy( compact + j + 4, &sins[i].sin_port, 2 );
            j += 6;
        }
        j = 0;
        for( i=0; i<num6; ++i ) {
            memcpy( compact6 + j, &sins6[i].sin6_addr, 16 );
            memcpy( compact6 + j + 16, &sins6[i].sin6_port, 2 );
            j += 18;
        }
        tr_bencInitDict( &benc, 3 );
        tr_bencDictAddRaw( &benc, "id", myid, 20 );
        if(num > 0)
            tr_bencDictAddRaw( &benc, "nodes", compact, num * 6 );
        if(num6 > 0)
            tr_bencDictAddRaw( &benc, "nodes6", compact6, num6 * 18 );
        dat_file = tr_buildPath( ss->configDir, "dht.dat", NULL );
        tr_bencToFile( &benc, TR_FMT_BENC, dat_file );
        tr_bencFree( &benc );
        tr_free( dat_file );
    }

    dht_uninit( 1 );
    tr_netCloseSocket( dht_socket );
    if(dht6_socket > 0)
        tr_netCloseSocket( dht6_socket );

    tr_ndbg("DHT", "Done uninitializing DHT");

    session = NULL;
}

tr_bool
tr_dhtEnabled( tr_session * ss )
{
    return ss && ( ss == session );
}

struct getstatus_closure
{
    int af;
    sig_atomic_t status;
    sig_atomic_t count;
};

static void
getstatus( void * cl )
{
    struct getstatus_closure * closure = cl;
    int good, dubious, incoming;

    dht_nodes( closure->af, &good, &dubious, NULL, &incoming );

    closure->count = good + dubious;

    if( good < 4 || good + dubious <= 8 )
        closure->status = TR_DHT_BROKEN;
    else if( good < 40 )
        closure->status = TR_DHT_POOR;
    else if( incoming < 8 )
        closure->status = TR_DHT_FIREWALLED;
    else
        closure->status = TR_DHT_GOOD;
}

int
tr_dhtStatus( tr_session * ss, int af, int * nodes_return )
{
    struct getstatus_closure closure = { af, -1, -1 };

    if( !tr_dhtEnabled( ss ) ||
        (af == AF_INET && dht_socket < 0) ||
        (af == AF_INET6 && dht6_socket < 0) ) {
        if( nodes_return )
            *nodes_return = 0;
        return TR_DHT_STOPPED;
    }

    tr_runInEventThread( ss, getstatus, &closure );
    while( closure.status < 0 )
        tr_wait( 10 /*msec*/ );

    if( nodes_return )
        *nodes_return = closure.count;

    return closure.status;
}

tr_port
tr_dhtPort( tr_session *ss )
{
    return tr_dhtEnabled( ss ) ? dht_port : 0;
}

int
tr_dhtAddNode( tr_session       * ss,
               const tr_address * address,
               tr_port            port,
               tr_bool            bootstrap )
{
    int af = address->type == TR_AF_INET ? AF_INET : AF_INET6;

    if( !tr_dhtEnabled( ss ) )
        return 0;

    /* Since we don't want to abuse our bootstrap nodes,
     * we don't ping them if the DHT is in a good state. */

    if(bootstrap) {
        if(tr_dhtStatus(ss, af, NULL) >= TR_DHT_FIREWALLED)
            return 0;
    }

    if( address->type == TR_AF_INET ) {
        struct sockaddr_in sin;
        memset(&sin, 0, sizeof(sin));
        sin.sin_family = AF_INET;
        memcpy(&sin.sin_addr, &address->addr.addr4, 4);
        sin.sin_port = htons(port);
        dht_ping_node((struct sockaddr*)&sin, sizeof(sin));
        return 1;
    } else if( address->type == TR_AF_INET6 ) {
        struct sockaddr_in6 sin6;
        memset(&sin6, 0, sizeof(sin6));
        sin6.sin6_family = AF_INET6;
        memcpy(&sin6.sin6_addr, &address->addr.addr6, 16);
        sin6.sin6_port = htons(port);
        dht_ping_node((struct sockaddr*)&sin6, sizeof(sin6));
        return 1;
    }

    return 0;
}

const char *
tr_dhtPrintableStatus(int status)
{
    switch(status) {
    case TR_DHT_STOPPED: return "stopped";
    case TR_DHT_BROKEN: return "broken";
    case TR_DHT_POOR: return "poor";
    case TR_DHT_FIREWALLED: return "firewalled";
    case TR_DHT_GOOD: return "good";
    default: return "???";
    }
}

static void
callback( void *ignore UNUSED, int event,
          unsigned char *info_hash, void *data, size_t data_len )
{
    if( event == DHT_EVENT_VALUES || event == DHT_EVENT_VALUES6 ) {
        tr_torrent *tor;
        tr_globalLock( session );
        tor = tr_torrentFindFromHash( session, info_hash );
        if( tor && tr_torrentAllowsDHT( tor ))
        {
            size_t i, n;
            tr_pex * pex;
            if( event == DHT_EVENT_VALUES )
                pex = tr_peerMgrCompactToPex(data, data_len, NULL, 0, &n);
            else
                pex = tr_peerMgrCompact6ToPex(data, data_len, NULL, 0, &n);
            for( i=0; i<n; ++i )
                tr_peerMgrAddPex( tor, TR_PEER_FROM_DHT, pex+i );
            tr_free(pex);
            tr_torinf(tor, "Learned %d%s peers from DHT",
                      (int)n,
                      event == DHT_EVENT_VALUES6 ? " IPv6" : "");
        }
        tr_globalUnlock( session );
    } else if( event == DHT_EVENT_SEARCH_DONE ||
               event == DHT_EVENT_SEARCH_DONE6) {
        tr_torrent * tor = tr_torrentFindFromHash( session, info_hash );
        if( tor ) {
            if( event == DHT_EVENT_SEARCH_DONE ) {
                tr_torinf(tor, "DHT announce done");
                tor->dhtAnnounceInProgress = 0;
            } else {
                tr_torinf(tor, "IPv6 DHT announce done");
                tor->dhtAnnounce6InProgress = 0;
            }
        }
    }
}

int
tr_dhtAnnounce(tr_torrent *tor, int af, tr_bool announce)
{
    int rc, status, numnodes, ret = 0;

    if( !tr_torrentAllowsDHT( tor ) )
        return -1;

    status = tr_dhtStatus( tor->session, af, &numnodes );

    if( status == TR_DHT_STOPPED ) {
        /* Let the caller believe everything is all right. */
        return 1;
    }

    if(status >= TR_DHT_POOR ) {
        rc = dht_search( tor->info.hash,
                         announce ? tr_sessionGetPeerPort(session) : 0,
                         af, callback, NULL);
        if( rc >= 1 ) {
            tr_torinf(tor, "Starting%s DHT announce (%s, %d nodes)",
                      af == AF_INET6 ? " IPv6" : "",
                      tr_dhtPrintableStatus(status), numnodes);
            if(af == AF_INET)
                tor->dhtAnnounceInProgress = TRUE;
            else
                tor->dhtAnnounce6InProgress = TRUE;
            ret = 1;
        } else {
            tr_torerr(tor, "%sDHT announce failed, errno = %d (%s, %d nodes)",
                      af == AF_INET6 ? "IPv6 " : "",
                      errno, tr_dhtPrintableStatus(status), numnodes);
        }
    } else {
        tr_tordbg(tor, "%sDHT not ready (%s, %d nodes)",
                  af == AF_INET6 ? "IPv6 " : "",
                  tr_dhtPrintableStatus(status), numnodes);
    }

    return ret;
}

static void
event_callback(int s, short type, void *ignore UNUSED )
{
    struct event *event = (s == dht_socket) ? &dht_event : &dht6_event;
    time_t tosleep;
    static int count = 0;

    if( dht_periodic( type == EV_READ, &tosleep, callback, NULL) < 0 ) {
        if(errno == EINTR) {
            tosleep = 0;
        } else {
            tr_nerr("DHT", "dht_periodic failed (errno = %d)", errno);
            if(errno == EINVAL || errno == EFAULT)
                    abort();
            tosleep = 1;
        }
    }

    /* Only do this once in a while.  Counting rather than measuring time
       avoids a system call. */
    count++;
    if(count >= 128) {
        rebind_ipv6(0);
        count = 0;
    }

    /* Being slightly late is fine,
       and has the added benefit of adding some jitter. */
    tr_timerAdd( event, tosleep, tr_cryptoWeakRandInt( 1000000 ) );
}

void
dht_hash(void *hash_return, int hash_size,
         const void *v1, int len1,
         const void *v2, int len2,
         const void *v3, int len3)
{
    unsigned char sha1[SHA_DIGEST_LENGTH];
    tr_sha1( sha1, v1, len1, v2, len2, v3, len3, NULL );
    memset( hash_return, 0, hash_size );
    memcpy( hash_return, sha1, MIN( hash_size, SHA_DIGEST_LENGTH ) );
}

int
dht_random_bytes( void * buf, size_t size )
{
    tr_cryptoRandBuf( buf, size );
    return size;
}