--- gvpe/src/vpn_dns.C 2005/03/06 21:52:15 1.23 +++ gvpe/src/vpn_dns.C 2007/11/10 05:14:22 1.42 @@ -16,9 +16,13 @@ You should have received a copy of the GNU General Public License along with gvpe; if not, write to the Free Software - Foundation, Inc. 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + Foundation, Inc. 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +// TODO: EDNS0 option to increase dns mtu? +// TODO: re-write dns packet parsing/creation using a safe mem-buffer +// to ensure no buffer overflows or similar problems. + #include "config.h" #if ENABLE_DNS @@ -46,23 +50,20 @@ #include "vpn.h" -#define MIN_POLL_INTERVAL .02 // how often to poll minimally when the server has data -#define MAX_POLL_INTERVAL 6. // how often to poll minimally when the server has no data +#define MAX_POLL_INTERVAL 5. // how often to poll minimally when the server has no data #define ACTIVITY_INTERVAL 5. #define INITIAL_TIMEOUT 0.1 // retry timeouts -#define INITIAL_SYN_TIMEOUT 10. // retry timeout for initial syn +#define INITIAL_SYN_TIMEOUT 2. // retry timeout for initial syn -#define MIN_SEND_INTERVAL 0.01 // wait at least this time between sending requests -#define MAX_SEND_INTERVAL 0.5 // optimistic? +#define MAX_SEND_INTERVAL 2. // optimistic? -#define MAX_OUTSTANDING 10 // max. outstanding requests #define MAX_WINDOW 1000 // max. for MAX_OUTSTANDING, and backlog -#define MAX_BACKLOG (100*1024) // size of gvpe protocol backlog (bytes), must be > MAXSIZE +#define MAX_BACKLOG (64*1024) // size of gvpe protocol backlog (bytes), must be > MAXSIZE -#define MAX_DOMAIN_SIZE 220 // 255 is legal limit, but bind doesn't compress well +#define MAX_DOMAIN_SIZE 240 // 255 is legal limit, but bind doesn't compress well // 240 leaves about 4 bytes of server reply data -// every two request bytes less give room for one reply byte +// every request byte less give room for two reply bytes #define SEQNO_MASK 0x3fff #define SEQNO_EQ(a,b) ( 0 == ( ((a) ^ (b)) & SEQNO_MASK) ) @@ -267,7 +268,7 @@ ///////////////////////////////////////////////////////////////////////////// #define HDRSIZE 6 - + inline void encode_header (char *data, int clientid, int seqno, int retry = 0) { seqno &= SEQNO_MASK; @@ -411,17 +412,24 @@ static int next_uid; u8 id1, id2, id3, id4; + u8 version; - u8 rrtype; u8 flags; + u8 rrtype; u8 def_ttl; - u8 rcv_cdc; - u8 snd_cdc; - u16 max_size; + u16 client; u16 uid; // to make request unique - u8 reserved[8]; + u16 max_size; + u8 seq_cdc; + u8 req_cdc; + + u8 rep_cdc; + u8 delay; // time in 0.01s units that the server may delay replying packets + u8 r3, r4; + + u8 r5, r6, r7, r8; void reset (int clientid); bool valid (); @@ -440,27 +448,31 @@ rrtype = RR_TYPE_TXT; flags = 0; - def_ttl = 1; - rcv_cdc = 0; - snd_cdc = 62; - max_size = ntohs (MAX_PKT_SIZE); - client = ntohs (clientid); + def_ttl = 0; + seq_cdc = 26; + req_cdc = 62; + rep_cdc = 0; + max_size = htons (MAX_PKT_SIZE); + client = htons (clientid); uid = next_uid++; + delay = 0; - memset (reserved, 0, 8); + r3 = r4 = 0; + r4 = r5 = r6 = r7 = 0; } bool dns_cfg::valid () { + // although the protocol itself allows for some configurability, + // only the following encoding/decoding settings are implemented. return id1 == 'G' && id2 == 'V' && id3 == 'P' && id4 == 'E' - && version == 1 - && flags == 0 - && rcv_cdc == 0 - && snd_cdc == 62 - && max_size == ntohs (MAX_PKT_SIZE); + && seq_cdc == 26 + && req_cdc == 62 + && rep_cdc == 0 + && version == 1; } struct dns_packet : net_packet @@ -469,7 +481,7 @@ u16 flags; // QR:1 Opcode:4 AA:1 TC:1 RD:1 RA:1 Z:3 RCODE:4 u16 qdcount, ancount, nscount, arcount; - u8 data[MAXSIZE - 6 * 2]; + u8 data [MAXSIZE - 6 * 2]; int decode_label (char *data, int size, int &offs); }; @@ -541,15 +553,15 @@ tstamp last_received; tstamp last_sent; - double last_latency; + double min_latency; double poll_interval, send_interval; vector rcvpq; - byte_stream rcvdq; int rcvseq; + byte_stream rcvdq; int rcvseq; int repseq; byte_stream snddq; int sndseq; - void time_cb (time_watcher &w); time_watcher tw; + void time_cb (ev::timer &w, int revents); ev::timer tw; void receive_rep (dns_rcv *r); dns_connection (connection *c); @@ -578,7 +590,7 @@ timeout = 0; retry = 0; seqno = 0; - sent = NOW; + sent = ev::ev_now (); stdhdr = false; pkt = new dns_packet; @@ -619,7 +631,7 @@ stdhdr = true; this->seqno = seqno; - timeout = NOW + INITIAL_TIMEOUT; + timeout = ev::ev_now () + INITIAL_TIMEOUT; pkt->flags = htons (DEFAULT_CLIENT_FLAGS); pkt->qdcount = htons (1); @@ -664,7 +676,7 @@ void dns_snd::gen_syn_req () { - timeout = NOW + INITIAL_SYN_TIMEOUT; + timeout = ev::ev_now () + INITIAL_SYN_TIMEOUT; pkt->flags = htons (DEFAULT_CLIENT_FLAGS); pkt->qdcount = htons (1); @@ -713,19 +725,19 @@ dns_connection::dns_connection (connection *c) : c (c) , rcvdq (MAX_BACKLOG * 2) -, snddq (MAX_BACKLOG * 2) +, snddq (MAX_BACKLOG) , tw (this, &dns_connection::time_cb) { vpn = c->vpn; established = false; - rcvseq = sndseq = 0; + rcvseq = repseq = sndseq = 0; last_sent = last_received = 0; - poll_interval = MIN_POLL_INTERVAL; + poll_interval = 0.5; // starting here send_interval = 0.5; // starting rate - last_latency = INITIAL_TIMEOUT; + min_latency = INITIAL_TIMEOUT; } dns_connection::~dns_connection () @@ -740,14 +752,15 @@ { if (r->datalen) { - last_received = NOW; - tw.trigger (); + last_received = ev::ev_now (); + tw (); poll_interval = send_interval; } else { poll_interval *= 1.5; + if (poll_interval > MAX_POLL_INTERVAL) poll_interval = MAX_POLL_INTERVAL; } @@ -760,6 +773,7 @@ for (vector::iterator i = rcvpq.end (); i-- != rcvpq.begin (); ) if (SEQNO_EQ (rcvseq, (*i)->seqno)) { + //printf ("seqno eq %x %x\n", rcvseq, (*i)->seqno);//D // enter the packet into our input stream r = *i; @@ -767,6 +781,7 @@ for (vector::iterator j = rcvpq.begin (); j != rcvpq.end (); ++j) if (SEQNO_EQ ((*j)->seqno, rcvseq - MAX_WINDOW)) { + //printf ("seqno RR %x %x\n", (*j)->seqno, rcvseq - MAX_WINDOW);//D delete *j; rcvpq.erase (j); break; @@ -783,7 +798,7 @@ while (vpn_packet *pkt = rcvdq.get ()) { sockinfo si; - si.host = 0x01010101; si.port = htons (c->conf->id); si.prot = PROT_DNSv4; + si.host = htonl (c->conf->id); si.port = 0; si.prot = PROT_DNSv4; vpn->recv_vpn_packet (pkt, si); @@ -807,7 +822,7 @@ if (0 == (flags & (FLAG_RESPONSE | FLAG_OP_MASK)) && pkt.qdcount == htons (1)) { - char qname[MAXSIZE]; + char qname [MAXSIZE]; int qlen = pkt.decode_label ((char *)qname, MAXSIZE - offs, offs); u16 qtype = pkt [offs++] << 8; qtype |= pkt [offs++]; @@ -824,7 +839,7 @@ if (qclass == RR_CLASS_IN && qlen > dlen + 1 - && !memcmp (qname + qlen - dlen - 1, THISNODE->domain, dlen)) + && !memcmp (qname + qlen - (dlen + 1), THISNODE->domain, dlen)) { // now generate reply pkt.ancount = htons (1); // one answer RR @@ -847,7 +862,6 @@ connection *c = conns [client - 1]; dns_connection *dns = c->dns; dns_rcv *rcv; - bool in_seq; if (dns) { @@ -868,8 +882,6 @@ goto duplicate_request; } - in_seq = dns->rcvseq == seqno; - // new packet, queue rcv = new dns_rcv (seqno, data, datalen); dns->receive_rep (rcv); @@ -886,31 +898,38 @@ int rdlen_offs = offs += 2; - int dlen = (dns ? ntohs (dns->cfg.max_size) : MAX_PKT_SIZE) - offs; - // bind doesn't compress well, so reduce further by one label length - dlen -= qlen; - if (dns) { + int dlen = ntohs (dns->cfg.max_size) - offs; + + // bind doesn't compress well, so reduce further by one label length + dlen -= qlen; + // only put data into in-order sequence packets, if // we receive out-of-order packets we generate empty // replies - while (dlen > 1 && !dns->snddq.empty () && in_seq) + //printf ("%d - %d & %x (=%d) < %d\n", seqno, dns->repseq, SEQNO_MASK, (seqno - dns->repseq) & SEQNO_MASK, MAX_WINDOW);//D + if (((seqno - dns->repseq) & SEQNO_MASK) <= MAX_WINDOW) { - int txtlen = dlen <= 255 ? dlen - 1 : 255; - - if (txtlen > dns->snddq.size ()) - txtlen = dns->snddq.size (); + dns->repseq = seqno; - pkt[offs++] = txtlen; - memcpy (pkt.at (offs), dns->snddq.begin (), txtlen); - offs += txtlen; - dns->snddq.remove (txtlen); + while (dlen > 1 && !dns->snddq.empty ()) + { + int txtlen = dlen <= 255 ? dlen - 1 : 255; + + if (txtlen > dns->snddq.size ()) + txtlen = dns->snddq.size (); + + pkt[offs++] = txtlen; + memcpy (pkt.at (offs), dns->snddq.begin (), txtlen); + offs += txtlen; + dns->snddq.remove (txtlen); - dlen -= txtlen + 1; + dlen -= txtlen + 1; + } } - // avoid empty TXT rdata + // avoid completely empty TXT rdata if (offs == rdlen_offs) pkt[offs++] = 0; @@ -956,7 +975,7 @@ pkt [offs++] = 0; pkt [offs++] = cfg.def_ttl; // TTL pkt [offs++] = 0; pkt [offs++] = 4; // rdlength - slog (L_INFO, _("DNS: client %d tries to connect"), client); + slog (L_INFO, _("DNS: client %d connects"), client); pkt [offs++] = CMD_IP_1; pkt [offs++] = CMD_IP_2; pkt [offs++] = CMD_IP_3; pkt [offs++] = CMD_IP_REJ; @@ -1000,6 +1019,7 @@ connection *c = dns->c; int seqno = (*i)->seqno; u8 data[MAXSIZE], *datap = data; + //printf ("rcv pkt %x\n", seqno);//D if ((*i)->retry) { @@ -1009,19 +1029,21 @@ } else { -#if 1 +#if 0 dns->send_interval *= 0.999; #endif - if (dns->send_interval < MIN_SEND_INTERVAL) - dns->send_interval = MIN_SEND_INTERVAL; - // the latency surely puts an upper bound on // the minimum send interval - double latency = NOW - (*i)->sent; - dns->last_latency = latency; + double latency = ev::ev_now () - (*i)->sent; + + if (latency < dns->min_latency) + dns->min_latency = latency; - if (dns->send_interval > latency) - dns->send_interval = latency; + if (dns->send_interval > dns->min_latency * conf.dns_overlap_factor) + dns->send_interval = dns->min_latency * conf.dns_overlap_factor; + + if (dns->send_interval < conf.dns_send_interval) + dns->send_interval = conf.dns_send_interval; } delete *i; @@ -1136,9 +1158,9 @@ } void -vpn::dnsv4_ev (io_watcher &w, short revents) +vpn::dnsv4_ev (ev::io &w, int revents) { - if (revents & EVENT_READ) + if (revents & EV_READ) { dns_packet *pkt = new dns_packet; struct sockaddr_in sa; @@ -1148,13 +1170,13 @@ if (pkt->len > 0) { - if (THISNODE->dns_port) + if (ntohs (pkt->flags) & FLAG_RESPONSE) + dnsv4_client (*pkt); + else { dnsv4_server (*pkt); sendto (w.fd, pkt->at (0), pkt->len, 0, (sockaddr *)&sa, sa_len); } - else - dnsv4_client (*pkt); delete pkt; } @@ -1164,7 +1186,7 @@ bool vpn::send_dnsv4_packet (vpn_packet *pkt, const sockinfo &si, int tos) { - int client = ntohs (si.port); + int client = ntohl (si.host); assert (0 < client && client <= conns.size ()); @@ -1173,11 +1195,10 @@ if (!c->dns) c->dns = new dns_connection (c); - if (!c->dns->snddq.put (pkt)) - return false; - - c->dns->tw.trigger (); + if (c->dns->snddq.put (pkt)) + c->dns->tw (); + // always return true even if the buffer overflows return true; } @@ -1190,14 +1211,14 @@ #define NEXT(w) do { if (next > (w)) next = w; } while (0) void -dns_connection::time_cb (time_watcher &w) +dns_connection::time_cb (ev::timer &w, int revents) { // servers have to be polled if (THISNODE->dns_port) return; // check for timeouts and (re)transmit - tstamp next = NOW + poll_interval; + tstamp next = ev::now () + poll_interval; dns_snd *send = 0; for (vector::iterator i = vpn->dns_sndpq.begin (); @@ -1206,85 +1227,86 @@ { dns_snd *r = *i; - if (r->timeout <= NOW) + if (r->timeout <= ev::ev_now ()) { if (!send) { send = r; r->retry++; - r->timeout = NOW + (r->retry * last_latency * 8.); + r->timeout = ev::ev_now () + (r->retry * min_latency * conf.dns_timeout_factor); + //printf ("RETRY %x (%d, %f)\n", r->seqno, r->retry, r->timeout - ev::ev_now ());//D // the following code changes the query section a bit, forcing // the forwarder to generate a new request if (r->stdhdr) - { - //printf ("reencoded header for ID %d retry %d:%d:%d\n", htons (r->pkt->id), THISNODE->id, r->seqno, r->retry);printf ("reencoded header for ID %d retry %d:%d:%d\n", htons (r->pkt->id), THISNODE->id, r->seqno, r->retry); - //encode_header ((char *)r->pkt->at (6 * 2 + 1), THISNODE->id, r->seqno, r->retry); - } + encode_header ((char *)r->pkt->at (6 * 2 + 1), THISNODE->id, r->seqno, r->retry); } } else NEXT (r->timeout); } - if (last_sent + send_interval <= NOW) + if (!send) { - if (!send) - { - // generate a new packet, if wise + // generate a new packet, if wise - if (!established) + if (!established) + { + if (vpn->dns_sndpq.empty ()) { - if (vpn->dns_sndpq.empty ()) - { - send = new dns_snd (this); + send = new dns_snd (this); - cfg.reset (THISNODE->id); - send->gen_syn_req (); - } + cfg.reset (THISNODE->id); + send->gen_syn_req (); } - else if (vpn->dns_sndpq.size () < MAX_OUTSTANDING - && !SEQNO_EQ (rcvseq, sndseq - (MAX_WINDOW - 1))) + } + else if (vpn->dns_sndpq.size () < conf.dns_max_outstanding + && !SEQNO_EQ (rcvseq, sndseq - (MAX_WINDOW - 1))) + { + if (last_sent + send_interval <= ev::ev_now ()) { //printf ("sending data request etc.\n"); //D - if (!snddq.empty ()) + if (!snddq.empty () || last_received + 1. > ev::ev_now ()) { poll_interval = send_interval; - NEXT (NOW + send_interval); + NEXT (ev::ev_now () + send_interval); } send = new dns_snd (this); send->gen_stream_req (sndseq, snddq); - send->timeout = NOW + last_latency * 8.; + send->timeout = ev::ev_now () + min_latency * conf.dns_timeout_factor; + //printf ("SEND %x (%f)\n", send->seqno, send->timeout - ev::ev_now (), min_latency, conf.dns_timeout_factor);//D sndseq = (sndseq + 1) & SEQNO_MASK; } - - if (send) - vpn->dns_sndpq.push_back (send); + else + NEXT (last_sent + send_interval); } if (send) - { - last_sent = NOW; - sendto (vpn->dnsv4_fd, - send->pkt->at (0), send->pkt->len, 0, - vpn->dns_forwarder.sav4 (), vpn->dns_forwarder.salenv4 ()); - } + vpn->dns_sndpq.push_back (send); } - else - NEXT (last_sent + send_interval); - slog (L_NOISE, "DNS: pi %f si %f N %f (%d:%d)", - poll_interval, send_interval, next - NOW, - vpn->dns_sndpq.size (), snddq.size ()); + if (send) + { + last_sent = ev::ev_now (); + sendto (vpn->dnsv4_fd, + send->pkt->at (0), send->pkt->len, 0, + vpn->dns_forwarder.sav4 (), vpn->dns_forwarder.salenv4 ()); + } - // TODO: no idea when this happens, but when next < NOW, we have a problem - if (next < NOW + 0.0001) - next = NOW + 0.1; + slog (L_NOISE, "DNS: pi %f si %f N %f (%d:%d %d)", + poll_interval, send_interval, next - ev::ev_now (), + vpn->dns_sndpq.size (), snddq.size (), + rcvpq.size ()); + + // TODO: no idea when this happens, but when next < ev::ev_now (), we have a problem + // doesn't seem to happen anymore + if (next < ev::ev_now () + 0.001) + next = ev::ev_now () + 0.1; - w.start (next); + w.start (next - ev::ev_now ()); } #endif