g2_node_drop(const char *routine, gnutella_node_t *n, const g2_tree_t *t,
const char *fmt, ...)
{
if (GNET_PROPERTY(g2_debug) || GNET_PROPERTY(log_dropped_g2)) {
va_list args;
char buf[256];
va_start(args, fmt);
if (fmt != NULL)
str_vbprintf(ARYLEN(buf), fmt, args);
else
buf[0] = '\0';
g_debug("%s(): dropping /%s from %s%s%s",
routine, g2_tree_name(t), node_infostr(n),
NULL == fmt ? "" : ": ", buf);
va_end(args);
}
gnet_stats_count_dropped(n, MSG_DROP_G2_UNEXPECTED);
if (GNET_PROPERTY(log_dropped_g2)) {
g2_tfmt_tree_dump(t, stderr, G2FMT_O_PAYLEN);
}
}
/**
* Handle reception of a /PI
*/
static void
g2_node_handle_ping(gnutella_node_t *n, const g2_tree_t *t)
{
g2_tree_t *c;
/*
* Throttle pings received from UDP.
*/
if (NODE_IS_UDP(n)) {
if (aging_lookup(g2_udp_pings, &n->addr)) {
gnet_stats_count_dropped(n, MSG_DROP_THROTTLE);
return;
}
aging_record(g2_udp_pings, WCOPY(&n->addr));
/* FALL THROUGH */
}
c = g2_tree_first_child(t);
/*
* If there is no payload, it's a keep-alive ping, send back a pong.
*/
if (NULL == c) {
g2_node_send_pong(n);
return;
}
/*
* There are children.
*
* If there is a /PI/UDP present, drop the message: we're not a hub,
* we don't have to relay this message to its UDP target (we're only
* connected to hubs, and the hub which got it should only forward that
* message it its neighbouring hubs, not to leaves).
*
* If there is a /PI/RELAY, the ping was relayed by a hub, but it made
* a mistake because we are a leaf node.
*/
g2_node_drop(G_STRFUNC, n, t, "has children and we are a leaf");
}
/**
* Look whether the datagram we received is a valid Gnutella packet.
*
* The routine also handles traffic statistics (reception and dropping).
*
* If ``n'' is not NULL, then ``s'' may be NULL. If ``n'' is NULL, then
* ``s'' must not be NULL.
*
* @param n the pseudo UDP reception node (NULL if invalid IP:port)
* @param s the socket on which we got the UDP datagram
* @param truncated whether datagram was truncated during reception
* @param header header of message
* @param payload payload of message (maybe not contiguous with header)
* @param len total length of message (header + payload)
*
* @return TRUE if valid, FALSE otherwise.
*/
bool
udp_is_valid_gnet_split(gnutella_node_t *n, const gnutella_socket_t *s,
bool truncated, const void *header, const void *payload, size_t len)
{
const char *msg;
uint16 size; /**< Payload size, from the Gnutella message */
g_assert(s != NULL || n != NULL);
/*
* If we can't get a proper UDP node for this address/port combination,
* ignore the message.
*/
if (NULL == n) {
msg = "Invalid address/port combination";
goto not;
}
if (len < GTA_HEADER_SIZE) {
msg = "Too short";
goto not;
}
/*
* We have enough to account for packet reception.
* Note that packet could be garbage at this point.
*/
memcpy(n->header, header, sizeof n->header);
n->size = len - GTA_HEADER_SIZE; /* Payload size if Gnutella msg */
gnet_stats_count_received_header(n);
gnet_stats_count_received_payload(n, payload);
/*
* If the message was truncated, then there is also going to be a
* size mismatch, but we want to flag truncated messages as being
* "too large" because this is mainly why we reject them. They may
* be legitimate Gnutella packets, too bad.
*/
if (truncated) {
msg = "Truncated (too large?)";
goto too_large;
}
/*
* Message sizes are architecturally limited to 64K bytes.
*
* We don't ensure the leading bits are zero in the size field because
* this constraint we put allows us to use those bits for flags in
* future extensions.
*
* The downside is that we have only 3 bytes (2 bytes for the size and
* 1 byte for the function type) to identify a valid Gnutella packet.
*/
switch (gmsg_size_valid(header, &size)) {
case GMSG_VALID:
case GMSG_VALID_MARKED:
break;
case GMSG_VALID_NO_PROCESS:
msg = "Header flags undefined for now";
goto drop;
case GMSG_INVALID:
msg = "Invalid size (greater than 64 KiB without flags)";
goto not; /* Probably just garbage */
}
if ((size_t) size + GTA_HEADER_SIZE != len) {
msg = "Size mismatch";
goto not;
}
/*
* We only support a subset of Gnutella message from UDP. In particular,
* messages like HSEP data, BYE or QRP are not expected!
*/
switch (gnutella_header_get_function(header)) {
case GTA_MSG_INIT:
case GTA_MSG_INIT_RESPONSE:
case GTA_MSG_VENDOR:
case GTA_MSG_STANDARD:
case GTA_MSG_PUSH_REQUEST:
case GTA_MSG_SEARCH_RESULTS:
case GTA_MSG_RUDP:
case GTA_MSG_DHT:
return TRUE;
case GTA_MSG_SEARCH:
if (settings_is_ultra() && GNET_PROPERTY(enable_guess)) {
return TRUE; /* GUESS query accepted */
}
msg = "Query from UDP refused";
goto drop;
}
msg = "Gnutella message not processed from UDP";
drop:
gnet_stats_count_dropped(n, MSG_DROP_UNEXPECTED);
gnet_stats_inc_general(GNR_UDP_UNPROCESSED_MESSAGE);
goto log;
too_large:
gnet_stats_count_dropped(n, MSG_DROP_TOO_LARGE);
gnet_stats_inc_general(GNR_UDP_UNPROCESSED_MESSAGE);
goto log;
not:
gnet_stats_inc_general(GNR_UDP_ALIEN_MESSAGE);
/* FALL THROUGH */
log:
if (GNET_PROPERTY(udp_debug)) {
g_warning("UDP got invalid %sGnutella packet (%zu byte%s) "
"\"%s\" %sfrom %s: %s",
socket_udp_is_old(s) ? "OLD " : "",
len, 1 == len ? "" : "s",
len >= GTA_HEADER_SIZE ?
gmsg_infostr_full_split(header, payload, len - GTA_HEADER_SIZE)
: "<incomplete Gnutella header>",
truncated ? "(truncated) " : "",
NULL == n ?
host_addr_port_to_string(s->addr, s->port) :
node_infostr(n),
msg);
if (len != 0) {
iovec_t iov[2];
iovec_set(&iov[0], header, GTA_HEADER_SIZE);
iovec_set(&iov[1], payload, len - GTA_HEADER_SIZE);
dump_hex_vec(stderr, "UDP datagram", iov, G_N_ELEMENTS(iov));
}
}
return FALSE; /* Dropped */
}
/**
* Handle reception of a /Q2
*/
static void
g2_node_handle_q2(gnutella_node_t *n, const g2_tree_t *t)
{
const guid_t *muid;
size_t paylen;
const g2_tree_t *c;
char *dn = NULL;
char *md = NULL;
uint32 iflags = 0;
search_request_info_t sri;
bool has_interest = FALSE;
node_inc_rx_query(n);
/*
* As a G2 leaf, we cannot handle queries coming from UDP because we
* are not supposed to get any!
*/
if (NODE_IS_UDP(n)) {
g2_node_drop(G_STRFUNC, n, t, "coming from UDP");
return;
}
/*
* The MUID of the query is the payload of the root node.
*/
muid = g2_tree_node_payload(t, &paylen);
if (paylen != GUID_RAW_SIZE) {
g2_node_drop(G_STRFUNC, n, t, "missing MUID");
return;
}
/*
* Make sure we have never seen this query already.
*
* To be able to leverage on Gnutella's routing table to detect duplicates
* over a certain lifespan, we are going to fake a minimal Gnutella header
* with a message type of GTA_MSG_G2_SEARCH, which is never actually used
* on the network.
*
* The TTL and hops are set to 1 and 0 initially, so that the message seems
* to come from a neighbouring host and cannot be forwarded.
*
* When that is done, we will be able to call route_message() and have
* all the necessary bookkeeping done for us.
*/
{
struct route_dest dest;
gnutella_header_set_muid(&n->header, muid);
gnutella_header_set_function(&n->header, GTA_MSG_G2_SEARCH);
gnutella_header_set_ttl(&n->header, 1);
gnutella_header_set_hops(&n->header, 0);
if (!route_message(&n, &dest))
return; /* Already accounted as duplicated, and logged */
}
/*
* Setup request information so that we can call search_request()
* to process our G2 query.
*/
ZERO(&sri);
sri.magic = SEARCH_REQUEST_INFO_MAGIC;
/*
* Handle the children of /Q2.
*/
G2_TREE_CHILD_FOREACH(t, c) {
enum g2_q2_child ct = TOKENIZE(g2_tree_name(c), g2_q2_children);
const char *payload;
switch (ct) {
case G2_Q2_DN:
payload = g2_tree_node_payload(c, &paylen);
if (payload != NULL && NULL == dn) {
uint off = 0;
/* Not NUL-terminated, need to h_strndup() it */
dn = h_strndup(payload, paylen);
if (!query_utf8_decode(dn, &off)) {
gnet_stats_count_dropped(n, MSG_DROP_MALFORMED_UTF_8);
goto done; /* Drop the query */
}
sri.extended_query = dn + off;
sri.search_len = paylen - off; /* In bytes */
}
break;
case G2_Q2_I:
if (!has_interest)
iflags = g2_node_extract_interest(c);
has_interest = TRUE;
break;
case G2_Q2_MD:
payload = g2_tree_node_payload(c, &paylen);
if (payload != NULL && NULL == md) {
/* Not NUL-terminated, need to h_strndup() it */
md = h_strndup(payload, paylen);
}
break;
case G2_Q2_NAT:
sri.flags |= QUERY_F_FIREWALLED;
break;
case G2_Q2_SZR: /* Size limits */
if (g2_node_extract_size_request(c, &sri.minsize, &sri.maxsize))
sri.size_restrictions = TRUE;
break;
case G2_Q2_UDP:
if (!sri.oob)
g2_node_extract_udp(c, &sri, n);
break;
case G2_Q2_URN:
g2_node_extract_urn(c, &sri);
break;
}
}
/*
* When there is no /Q2/I, return a default set of information.
*/
if (!has_interest)
iflags = G2_Q2_F_DFLT;
/*
* If there are meta-data, try to intuit which media types there are
* looking for.
*
* The payload is XML looking like "<audio/>" or "<video/>" but there
* can be attributes and we don't want to do a full XML parsing there.
* Hence we'll base our analysis on simple lexical parsing, which is
* why we call a routine to "intuit", not to "extract".
*
* Also, this is poorer than Gnutella's GGEP "M" because apparently there
* can be only one single type, since the XML payload must obey some
* kind of schema and there is an audio schema, a video schema, etc...
* XML was just a wrong design choice there.
*/
if (md != NULL)
sri.media_types = g2_node_intuit_media_type(md);
/*
* Validate the return address if OOB hit delivery is configured.
*/
if (sri.oob && !search_oob_is_allowed(n, &sri))
goto done;
/*
* Update statistics, as done in search_request_preprocess() for Gnutella.
*/
if (sri.exv_sha1cnt) {
gnet_stats_inc_general(GNR_QUERY_G2_SHA1);
if (NULL == dn) {
int i;
for (i = 0; i < sri.exv_sha1cnt; i++) {
search_request_listener_emit(QUERY_SHA1,
sha1_base32(&sri.exv_sha1[i].sha1), n->addr, n->port);
}
}
}
if (dn != NULL && !is_ascii_string(dn))
gnet_stats_inc_general(GNR_QUERY_G2_UTF8);
if (dn != NULL)
search_request_listener_emit(QUERY_STRING, dn, n->addr, n->port);
if (!search_is_valid(n, 0, &sri))
goto done;
/*
* Perform the query.
*/
sri.g2_query = TRUE;
sri.partials = booleanize(iflags & G2_Q2_F_PFS);
sri.g2_wants_url = booleanize(iflags & G2_Q2_F_URL);
sri.g2_wants_alt = booleanize(iflags & G2_Q2_F_A);
sri.g2_wants_dn = booleanize(iflags & G2_Q2_F_DN);
search_request(n, &sri, NULL);
done:
HFREE_NULL(dn);
HFREE_NULL(md);
}
/**
* Route query hits from one node to the other.
*/
void
dh_route(gnutella_node_t *src, gnutella_node_t *dest, int count)
{
pmsg_t *mb;
struct dh_pmsg_info *pmi;
const struct guid *muid;
dqhit_t *dh;
mqueue_t *mq;
g_assert(
gnutella_header_get_function(&src->header) == GTA_MSG_SEARCH_RESULTS);
g_assert(count >= 0);
if (!NODE_IS_WRITABLE(dest))
goto drop_shutdown;
muid = gnutella_header_get_muid(&src->header);
dh = dh_locate(muid);
g_assert(dh != NULL); /* Must have called dh_got_results() first! */
if (GNET_PROPERTY(dh_debug) > 19) {
g_debug("DH #%s got %d hit%s: "
"msg=%u, hits_recv=%u, hits_sent=%u, hits_queued=%u",
guid_hex_str(muid), count, plural(count),
dh->msg_recv, dh->hits_recv, dh->hits_sent,
dh->hits_queued);
}
mq = dest->outq;
/*
* Can we forward the message?
*/
switch (dh_can_forward(dh, mq, FALSE)) {
case DH_DROP_FC:
goto drop_flow_control;
case DH_DROP_THROTTLE:
goto drop_throttle;
case DH_DROP_TRANSIENT:
goto drop_transient;
case DH_FORWARD:
default:
break;
}
/*
* Allow message through.
*/
WALLOC(pmi);
pmi->hits = count;
dh->hits_queued += count;
dh->msg_queued++;
g_assert(dh->hits_queued >= UNSIGNED(count));
/*
* Magic: we create an extended version of a pmsg_t that contains a
* free routine, which will be invoked when the message queue frees
* the message.
*
* This enables us to track how much results we already queued/sent.
*/
if (NODE_IS_UDP(dest)) {
gnet_host_t to;
pmsg_t *mbe;
gnet_host_set(&to, dest->addr, dest->port);
/*
* With GUESS we may route back a query hit to an UDP node.
*/
if (GNET_PROPERTY(guess_server_debug) > 19) {
g_debug("GUESS sending %d hit%s (%s) for #%s to %s",
count, plural(count),
NODE_CAN_SR_UDP(dest) ? "reliably" :
NODE_CAN_INFLATE(dest) ? "possibly deflated" : "uncompressed",
guid_hex_str(muid), node_infostr(dest));
}
/*
* Attempt to compress query hit if the destination supports it.
*
* If we're going to send the hit using semi-reliable UDP, there's
* no need to compress beforehand, since the transport layer will
* attempt its own compression anyway.
*/
if (!NODE_CAN_SR_UDP(dest) && NODE_CAN_INFLATE(dest)) {
mb = gmsg_split_to_deflated_pmsg(&src->header, src->data,
src->size + GTA_HEADER_SIZE);
if (gnutella_header_get_ttl(pmsg_start(mb)) & GTA_UDP_DEFLATED)
gnet_stats_inc_general(GNR_UDP_TX_COMPRESSED);
} else {
mb = gmsg_split_to_pmsg(&src->header, src->data,
src->size + GTA_HEADER_SIZE);
}
mbe = pmsg_clone_extend(mb, dh_pmsg_free, pmi);
pmsg_free(mb);
if (NODE_CAN_SR_UDP(dest))
pmsg_mark_reliable(mbe);
mq_udp_putq(mq, mbe, &to);
} else {
mb = gmsg_split_to_pmsg_extend(&src->header, src->data,
src->size + GTA_HEADER_SIZE, dh_pmsg_free, pmi);
mq_tcp_putq(mq, mb, src);
if (GNET_PROPERTY(dh_debug) > 19) {
g_debug("DH enqueued %d hit%s for #%s to %s",
count, plural(count), guid_hex_str(muid),
node_infostr(dest));
}
}
return;
drop_shutdown:
gnet_stats_count_dropped(src, MSG_DROP_SHUTDOWN);
return;
drop_flow_control:
gnet_stats_count_dropped(src, MSG_DROP_FLOW_CONTROL);
gnet_stats_count_flowc(&src->header, TRUE);
return;
drop_throttle:
gnet_stats_count_dropped(src, MSG_DROP_THROTTLE);
return;
drop_transient:
gnet_stats_count_dropped(src, MSG_DROP_TRANSIENT);
return;
}
