void
msg_put(struct msg *msg)
{
listNode *node;
struct mbuf *mbuf;
log_debug(LOG_VVERB, "put msg %p id %"PRIu64"", msg, msg->id);
while (listLength(msg->data) > 0) {
node = listFirst(msg->data);
mbuf = listNodeValue(node);
listDelNode(msg->data, node);
mbuf_put(mbuf);
}
listRelease(msg->data);
msg->data = NULL;
if (msg->frag_seq) {
rmt_free(msg->frag_seq);
msg->frag_seq = NULL;
}
if (msg->keys) {
msg->keys->nelem = 0; /* a hack here */
array_destroy(msg->keys);
msg->keys = NULL;
}
msg->mb = NULL;
}
void dyn_parse_rsp(struct msg *r)
{
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, ":::::::::::::::::::::: In dyn_parse_rsp, start to process response :::::::::::::::::::::::: ");
msg_dump(r);
}
if (dyn_parse_core(r)) {
struct dmsg *dmsg = r->dmsg;
struct mbuf *b = STAILQ_LAST(&r->mhdr, mbuf, next);
if (dmsg->type != DMSG_UNKNOWN && dmsg->type != DMSG_RES) {
log_debug(LOG_DEBUG, "Resp parser: I got a dnode msg of type %d", dmsg->type);
r->state = 0;
r->result = MSG_PARSE_OK;
r->dyn_state = DYN_DONE;
return;
}
//check whether we need to decrypt the payload
if (dmsg->bit_field == 1) {
//dmsg->owner->owner->dnode_secured = 1;
struct mbuf *decrypted_buf = mbuf_get();
if (decrypted_buf == NULL) {
log_debug(LOG_INFO, "Unable to obtain an mbuf for dnode msg's header!");
r->result = MSG_OOM_ERROR;
return;
}
//Dont need to decrypt AES key - pull it out from the conn
dyn_aes_decrypt(dmsg->payload, dmsg->plen, decrypted_buf, r->owner->aes_key);
b->pos = b->pos + dmsg->plen;
r->pos = decrypted_buf->start;
mbuf_copy(decrypted_buf, b->pos, mbuf_length(b));
mbuf_insert(&r->mhdr, decrypted_buf);
mbuf_remove(&r->mhdr, b);
mbuf_put(b);
r->mlen = mbuf_length(decrypted_buf);
}
if (r->redis) {
return redis_parse_rsp(r);
}
return memcache_parse_rsp(r);
}
//bad case
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Resp: bad message - cannot parse"); //fix me to do something
msg_dump(r);
}
r->result = MSG_PARSE_AGAIN;
}
rstatus_t
conn_close(struct conn *conn)
{
rstatus_t status;
struct mbuf *mbuf, *nbuf; /* current and next mbuf */
if (conn->fd < 0) {
conn_put(conn);
return NC_OK;
}
if (!STAILQ_EMPTY(&conn->recv_queue)) {
log_warn("close conn %d discard data in send_queue", conn->fd);
for (mbuf = STAILQ_FIRST(&conn->recv_queue); mbuf != NULL; mbuf = nbuf) {
nbuf = STAILQ_NEXT(mbuf, next);
mbuf_remove(&conn->recv_queue, mbuf);
mbuf_put(mbuf);
}
}
if (!STAILQ_EMPTY(&conn->send_queue)) {
log_warn("close conn %d discard data in send_queue", conn->fd);
for (mbuf = STAILQ_FIRST(&conn->send_queue); mbuf != NULL; mbuf = nbuf) {
nbuf = STAILQ_NEXT(mbuf, next);
mbuf_remove(&conn->send_queue, mbuf);
mbuf_put(mbuf);
}
}
status = close(conn->fd);
if (status < 0) {
log_error("close c %d failed, ignored: %s", conn->fd, strerror(errno));
}
conn->fd = -1;
conn_put(conn);
return NC_OK;
}
static rstatus_t
conn_send_queue(struct conn *conn)
{
struct mbuf *mbuf, *nbuf; /* current and next mbuf */
size_t mlen; /* current mbuf data length */
ssize_t n;
for (mbuf = STAILQ_FIRST(&conn->send_queue); mbuf != NULL; mbuf = nbuf) {
nbuf = STAILQ_NEXT(mbuf, next);
if (mbuf_empty(mbuf)) {
continue;
}
mlen = mbuf_length(mbuf);
n = conn_send_buf(conn, mbuf->pos, mlen);
if (n < 0) {
if (n == NC_EAGAIN) {
return NC_OK;
}
return NC_ERROR;
}
mbuf->pos += n;
if (n < mlen) {
ASSERT(mbuf->pos < mbuf->end);
return NC_OK;
}
ASSERT(mbuf->pos == mbuf->last);
mbuf_remove(&conn->send_queue, mbuf);
mbuf_put(mbuf);
}
conn->send_ready = 0;
return NC_OK;
}
/*
* Pre-coalesce handler is invoked when the message is a response to
* the fragmented multi vector request - 'get' or 'gets' and all the
* responses to the fragmented request vector hasn't been received
*/
void
memcache_pre_coalesce(struct msg *r)
{
struct msg *pr = r->peer; /* peer request */
struct mbuf *mbuf;
ASSERT(!r->request);
ASSERT(pr->request);
if (pr->frag_id == 0) {
/* do nothing, if not a response to a fragmented request */
return;
}
pr->frag_owner->nfrag_done++;
switch (r->type) {
case MSG_RSP_MC_VALUE:
case MSG_RSP_MC_END:
/*
* Readjust responses of the fragmented message vector by not
* including the end marker for all
*/
ASSERT(r->end != NULL);
for (;;) {
mbuf = STAILQ_LAST(&r->mhdr, mbuf, next);
ASSERT(mbuf != NULL);
/*
* We cannot assert that end marker points to the last mbuf
* Consider a scenario where end marker points to the
* penultimate mbuf and the last mbuf only contains spaces
* and CRLF: mhdr -> [...END] -> [\r\n]
*/
if (r->end >= mbuf->pos && r->end < mbuf->last) {
/* end marker is within this mbuf */
r->mlen -= (uint32_t)(mbuf->last - r->end);
mbuf->last = r->end;
break;
}
/* end marker is not in this mbuf */
r->mlen -= mbuf_length(mbuf);
mbuf_remove(&r->mhdr, mbuf);
mbuf_put(mbuf);
}
break;
default:
/*
* Valid responses for a fragmented requests are MSG_RSP_MC_VALUE or,
* MSG_RSP_MC_END. For an invalid response, we send out SERVER_ERRROR
* with EINVAL errno
*/
mbuf = STAILQ_FIRST(&r->mhdr);
log_hexdump(LOG_ERR, mbuf->pos, mbuf_length(mbuf), "rsp fragment "
"with unknown type %d", r->type);
pr->error = 1;
pr->err = EINVAL;
break;
}
}
static rstatus_t
sentinel_proc_pub(struct context *ctx, struct msg *msg)
{
rstatus_t status;
struct string pool_name, server_name, server_ip,
tmp_string, pub_titile, pub_event;
struct mbuf *line_buf;
int server_port;
string_init(&tmp_string);
string_init(&pool_name);
string_init(&server_name);
string_init(&server_ip);
string_set_text(&pub_titile, "pmessage");
string_set_text(&pub_event, "+switch-master");
line_buf = mbuf_get();
if (line_buf == NULL) {
goto error;
}
/* get line in line num 3 for pub titile */
msg_read_line(msg, line_buf, 3);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel pmessage when skip line not used.");
goto error;
}
status = mbuf_read_string(line_buf, CR, &tmp_string);
if (status != NC_OK || string_compare(&pub_titile, &tmp_string)) {
log_error("pub title error(lineinfo %.*s)", tmp_string.len, tmp_string.data);
goto error;
}
/* get line in line num 7 for pub event */
msg_read_line(msg, line_buf, 4);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel pmessage when skip line not used.");
goto error;
}
status = mbuf_read_string(line_buf, CR, &tmp_string);
if (status != NC_OK || string_compare(&pub_event, &tmp_string)) {
log_error("pub channel error(lineinfo %.*s)", tmp_string.len, tmp_string.data);
goto error;
}
/* get line in line num 9 for pub info */
msg_read_line(msg, line_buf, 2);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel pmessage when skip line not used.");
goto error;
}
/* parse switch master info */
/* get pool name */
status = mbuf_read_string(line_buf, SENTINEL_SERVERNAME_SPLIT, &pool_name);
if (status != NC_OK) {
log_error("get pool name string failed.");
goto error;
}
/* get server name */
status = mbuf_read_string(line_buf, ' ', &server_name);
if (status != NC_OK) {
log_error("get server name string failed.");
goto error;
}
/* skip old ip and port string */
status = mbuf_read_string(line_buf, ' ', NULL);
if (status != NC_OK) {
log_error("skip old ip string failed.");
goto error;
}
status = mbuf_read_string(line_buf, ' ', NULL);
if (status != NC_OK) {
log_error("skip old port string failed.");
goto error;
}
/* get new server ip string */
status = mbuf_read_string(line_buf, ' ', &server_ip);
if (status != NC_OK) {
log_error("get new server ip string failed.");
goto error;
}
/* get new server port */
status = mbuf_read_string(line_buf, CR, &tmp_string);
if (status != NC_OK) {
log_error("get new server port string failed.");
goto error;
}
server_port = nc_atoi(tmp_string.data, tmp_string.len);
if (server_port < 0) {
log_error("tanslate server port string to int failed.");
goto error;
}
status = server_switch(ctx, &pool_name, &server_name, &server_ip, server_port);
if (status == NC_OK) {
conf_rewrite(ctx);
}
status = NC_OK;
done:
if (line_buf != NULL) {
mbuf_put(line_buf);
}
string_deinit(&tmp_string);
string_deinit(&server_ip);
string_deinit(&server_name);
string_deinit(&pool_name);
return status;
error:
status = NC_ERROR;
goto done;
}
static rstatus_t
sentinel_proc_acksub(struct context *ctx, struct msg *msg)
{
rstatus_t status;
struct string sub_titile, sub_channel, sub_ok, tmp_string;
struct mbuf *line_buf;
string_init(&tmp_string);
string_set_text(&sub_titile, "psubscribe");
string_set_text(&sub_channel, "+switch-master");
string_set_text(&sub_ok, ":1");
line_buf = mbuf_get();
if (line_buf == NULL) {
goto error;
}
/* get line in line num 3 for sub titile */
msg_read_line(msg, line_buf, 3);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel ack sub when skip line not used.");
goto error;
}
status = mbuf_read_string(line_buf, CR, &tmp_string);
if (status != NC_OK || string_compare(&sub_titile, &tmp_string)) {
goto error;
}
/* get line in line num 5 for sub channel */
msg_read_line(msg, line_buf, 2);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel ack sub when skip line not used.");
goto error;
}
status = mbuf_read_string(line_buf, CR, &tmp_string);
if (status != NC_OK || string_compare(&sub_channel, &tmp_string)) {
goto error;
}
/* get sub status */
msg_read_line(msg, line_buf, 1);
if (line_buf == 0) {
log_error("read line failed from sentinel ack sub when skip line not used.");
goto error;
}
status = mbuf_read_string(line_buf, CR, &tmp_string);
if (status != NC_OK || string_compare(&sub_ok, &tmp_string)) {
goto error;
}
log_debug(LOG_INFO, "success sub channel %.*s from sentinel", sub_channel.len, sub_channel.data);
status = NC_OK;
done:
if (line_buf != NULL) {
mbuf_put(line_buf);
}
string_deinit(&tmp_string);
return status;
error:
status = NC_ERROR;
goto done;
}
static rstatus_t
sentinel_proc_sentinel_info(struct context *ctx, struct msg *msg)
{
rstatus_t status;
int i, master_num, switch_num;
struct string pool_name, server_name, server_ip,
tmp_string, sentinel_masters_prefix, master_ok;
struct mbuf *line_buf;
int server_port;
string_init(&tmp_string);
string_init(&pool_name);
string_init(&server_name);
string_init(&server_ip);
string_set_text(&sentinel_masters_prefix, "sentinel_masters");
string_set_text(&master_ok, "status=ok");
line_buf = mbuf_get();
if (line_buf == NULL) {
goto error;
}
/* get sentinel master num at line 3 */
msg_read_line(msg, line_buf, 3);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel ack info when skip line not used.");
goto error;
}
status = mbuf_read_string(line_buf, ':', &tmp_string);
if (status != NC_OK || string_compare(&sentinel_masters_prefix, &tmp_string)) {
goto error;
}
status = mbuf_read_string(line_buf, CR, &tmp_string);
if (status != NC_OK) {
goto error;
}
master_num = nc_atoi(tmp_string.data, tmp_string.len);
if (master_num < 0) {
log_error("parse master number from sentinel ack info failed.");
goto error;
}
/* skip 3 line in ack info which is not used. */
msg_read_line(msg, line_buf, 3);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel ack info when skip line not used.");
goto error;
}
/* parse master info from sentinel ack info */
switch_num = 0;
for (i = 0; i < master_num; i++) {
msg_read_line(msg, line_buf, 1);
if (mbuf_length(line_buf) == 0) {
log_error("read line failed from sentinel ack info when parse master item.");
goto error;
}
log_debug(LOG_INFO, "master item line : %.*s", mbuf_length(line_buf), line_buf->pos);
/* skip master item prefix */
status = mbuf_read_string(line_buf, ':', NULL);
if (status != NC_OK) {
log_error("skip master item prefix failed");
goto error;
}
/* skip master item server name prefix */
status = mbuf_read_string(line_buf, '=', NULL);
if (status != NC_OK) {
log_error("skip master item server name prefix failed.");
goto error;
}
/* get server pool name */
status = mbuf_read_string(line_buf, SENTINEL_SERVERNAME_SPLIT, &pool_name);
if (status != NC_OK) {
log_error("get server pool name failed.");
goto error;
}
/* get server name */
status = mbuf_read_string(line_buf, ',', &server_name);
if (status != NC_OK) {
log_error("get server name failed.");
goto error;
}
/* get master status */
status = mbuf_read_string(line_buf, ',', &tmp_string);
if (status != NC_OK) {
log_error("get master status failed.");
goto error;
}
if (string_compare(&master_ok, &tmp_string)) {
log_error("master item status is not ok, use it anyway");
}
/* skip ip string prefix name */
status = mbuf_read_string(line_buf, '=', NULL);
if (status != NC_OK) {
log_error("skip master item address prefix failed.");
goto error;
}
/* get server ip string */
status = mbuf_read_string(line_buf, ':', &server_ip);
if (status != NC_OK) {
log_error("get server ip string failed.");
goto error;
}
/* get server port */
status = mbuf_read_string(line_buf, ',', &tmp_string);
if (status != NC_OK) {
log_error("get server port string failed.");
goto error;
}
server_port = nc_atoi(tmp_string.data, tmp_string.len);
if (server_port < 0) {
log_error("tanslate server port string to int failed.");
goto error;
}
status = server_switch(ctx, &pool_name, &server_name, &server_ip, server_port);
/* if server is switched, add switch number */
if (status == NC_OK) {
switch_num++;
}
}
if (switch_num > 0) {
conf_rewrite(ctx);
}
status = NC_OK;
done:
if (line_buf != NULL) {
mbuf_put(line_buf);
}
string_deinit(&tmp_string);
string_deinit(&pool_name);
string_deinit(&server_name);
string_deinit(&server_ip);
return status;
error:
status = NC_ERROR;
goto done;
}
/* Forward a client request over to a peer */
void dnode_peer_req_forward(struct context *ctx, struct conn *c_conn,
struct conn *p_conn, struct msg *msg,
struct rack *rack, uint8_t *key, uint32_t keylen)
{
struct server *server = p_conn->owner;
log_debug(LOG_DEBUG, "forwarding request from client conn '%s' to peer conn '%s' on rack '%.*s' dc '%.*s' ",
dn_unresolve_peer_desc(c_conn->sd), dn_unresolve_peer_desc(p_conn->sd),
rack->name->len, rack->name->data,
server->dc.len, server->dc.data);
struct string *dc = rack->dc;
rstatus_t status;
/* enqueue message (request) into client outq, if response is expected */
if (!msg->noreply && !msg->swallow) {
conn_enqueue_outq(ctx, c_conn, msg);
}
ASSERT(p_conn->type == CONN_DNODE_PEER_SERVER);
ASSERT((c_conn->type == CONN_CLIENT) ||
(c_conn->type == CONN_DNODE_PEER_CLIENT));
/* enqueue the message (request) into peer inq */
status = event_add_out(ctx->evb, p_conn);
if (status != DN_OK) {
dnode_req_forward_error(ctx, p_conn, msg);
p_conn->err = errno;
return;
}
struct mbuf *header_buf = mbuf_get();
if (header_buf == NULL) {
loga("Unable to obtain an mbuf for dnode msg's header!");
req_put(msg);
return;
}
struct server_pool *pool = c_conn->owner;
dmsg_type_t msg_type = (string_compare(&pool->dc, dc) != 0)? DMSG_REQ_FORWARD : DMSG_REQ;
if (p_conn->dnode_secured) {
//Encrypting and adding header for a request
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "AES encryption key: %s\n", base64_encode(p_conn->aes_key, AES_KEYLEN));
}
//write dnode header
if (ENCRYPTION) {
status = dyn_aes_encrypt_msg(msg, p_conn->aes_key);
if (status == DN_ERROR) {
loga("OOM to obtain an mbuf for encryption!");
mbuf_put(header_buf);
req_put(msg);
return;
}
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "#encrypted bytes : %d", status);
}
dmsg_write(header_buf, msg->id, msg_type, p_conn, msg_length(msg));
} else {
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "no encryption on the msg payload");
}
dmsg_write(header_buf, msg->id, msg_type, p_conn, msg_length(msg));
}
} else {
//write dnode header
dmsg_write(header_buf, msg->id, msg_type, p_conn, msg_length(msg));
}
mbuf_insert_head(&msg->mhdr, header_buf);
if (log_loggable(LOG_VVERB)) {
log_hexdump(LOG_VVERB, header_buf->pos, mbuf_length(header_buf), "dyn message header: ");
msg_dump(msg);
}
conn_enqueue_inq(ctx, p_conn, msg);
dnode_peer_req_forward_stats(ctx, p_conn->owner, msg);
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, "remote forward from c %d to s %d req %"PRIu64" len %"PRIu32
" type %d with key '%.*s'", c_conn->sd, p_conn->sd, msg->id,
msg->mlen, msg->type, keylen, key);
}
}
rstatus_t
dnode_peer_handshake_announcing(void *rmsg)
{
rstatus_t status;
struct ring_msg *msg = rmsg;
struct server_pool *sp = msg->sp;
log_debug(LOG_VVERB, "dyn: handshaking peers");
struct array *peers = &sp->peers;
uint32_t i,nelem;
nelem = array_n(peers);
//we assume one mbuf is enough for now - will enhance with multiple mbufs later
struct mbuf *mbuf = mbuf_get();
if (mbuf == NULL) {
log_debug(LOG_VVERB, "Too bad, not enough memory!");
return DN_ENOMEM;
}
//annoucing myself by sending msg: 'dc$rack$token,started_ts,node_state,node_dns'
mbuf_write_string(mbuf, &sp->dc);
mbuf_write_char(mbuf, '$');
mbuf_write_string(mbuf, &sp->rack);
mbuf_write_char(mbuf, '$');
struct dyn_token *token = (struct dyn_token *) array_get(&sp->tokens, 0);
if (token == NULL) {
log_debug(LOG_VVERB, "Why? This should not be null!");
mbuf_put(mbuf);
return DN_ERROR;
}
mbuf_write_uint32(mbuf, token->mag[0]);
mbuf_write_char(mbuf, ',');
int64_t cur_ts = (int64_t)time(NULL);
mbuf_write_uint64(mbuf, cur_ts);
mbuf_write_char(mbuf, ',');
mbuf_write_uint8(mbuf, sp->ctx->dyn_state);
mbuf_write_char(mbuf, ',');
char *broadcast_addr = get_broadcast_address(sp);
mbuf_write_bytes(mbuf, broadcast_addr, dn_strlen(broadcast_addr));
//for each peer, send a registered msg
for (i = 0; i < nelem; i++) {
struct server *peer = (struct server *) array_get(peers, i);
if (peer->is_local)
continue;
log_debug(LOG_VVERB, "Gossiping to node '%.*s'", peer->name.len, peer->name.data);
struct conn * conn = dnode_peer_conn(peer);
if (conn == NULL) {
//running out of connection due to memory exhaust
log_debug(LOG_DEBUG, "Unable to obtain a connection object");
return DN_ERROR;
}
status = dnode_peer_connect(sp->ctx, peer, conn);
if (status != DN_OK ) {
dnode_peer_close(sp->ctx, conn);
log_debug(LOG_DEBUG, "Error happened in connecting on conn %d", conn->sd);
return DN_ERROR;
}
//conn->
dnode_peer_gossip_forward(sp->ctx, conn, sp->redis, mbuf);
//peer_gossip_forward1(sp->ctx, conn, sp->redis, &data);
}
//free this as nobody else will do
//mbuf_put(mbuf);
return DN_OK;
}
void
dyn_parse_req(struct msg *r)
{
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, ":::::::::::::::::::::: In dyn_parse_req, start to process request :::::::::::::::::::::: ");
msg_dump(r);
}
bool done_parsing = false;
struct mbuf *b = STAILQ_LAST(&r->mhdr, mbuf, next);
if (dyn_parse_core(r)) {
struct dmsg *dmsg = r->dmsg;
struct conn *conn = r->owner;
conn->same_dc = dmsg->same_dc;
if (dmsg->type != DMSG_UNKNOWN && dmsg->type != DMSG_REQ &&
dmsg->type != DMSG_REQ_FORWARD && dmsg->type != GOSSIP_SYN) {
r->state = 0;
r->result = MSG_PARSE_OK;
r->dyn_state = DYN_DONE;
return;
}
if (r->dyn_state == DYN_DONE && dmsg->bit_field == 1) {
dmsg->owner->owner->dnode_secured = 1;
r->owner->dnode_crypto_state = 1;
r->dyn_state = DYN_POST_DONE;
r->result = MSG_PARSE_REPAIR;
if (dmsg->mlen > 1) {
//Decrypt AES key
dyn_rsa_decrypt(dmsg->data, aes_decrypted_buf);
strncpy(r->owner->aes_key, aes_decrypted_buf, strlen(aes_decrypted_buf));
}
if (dmsg->plen + b->pos <= b->last) {
struct mbuf *decrypted_buf = mbuf_get();
if (decrypted_buf == NULL) {
loga("Unable to obtain an mbuf for dnode msg's header!");
r->result = MSG_OOM_ERROR;
return;
}
dyn_aes_decrypt(b->pos, dmsg->plen, decrypted_buf, r->owner->aes_key);
b->pos = b->pos + dmsg->plen;
r->pos = decrypted_buf->start;
mbuf_copy(decrypted_buf, b->pos, mbuf_length(b));
mbuf_insert(&r->mhdr, decrypted_buf);
mbuf_remove(&r->mhdr, b);
mbuf_put(b);
r->mlen = mbuf_length(decrypted_buf);
data_store_parse_req(r);
}
//substract alraedy received bytes
dmsg->plen -= b->last - b->pos;
return;
} else if (r->dyn_state == DYN_POST_DONE) {
struct mbuf *last_buf = STAILQ_LAST(&r->mhdr, mbuf, next);
if (last_buf->read_flip == 1) {
data_store_parse_req(r);
} else {
r->result = MSG_PARSE_AGAIN;
}
return;
}
if (dmsg->type == GOSSIP_SYN) {
//TODOs: need to address multi-buffer msg later
dmsg->payload = b->pos;
b->pos = b->pos + dmsg->plen;
r->pos = b->pos;
done_parsing = true;
}
if (done_parsing)
return;
return data_store_parse_req(r);
}
//bad case
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, "Bad or splitted message"); //fix me to do something
msg_dump(r);
}
r->result = MSG_PARSE_AGAIN;
}
/* dnode sends a response back to a peer */
struct msg *
dnode_rsp_send_next(struct context *ctx, struct conn *conn)
{
rstatus_t status;
ASSERT(conn->dnode_client && !conn->dnode_server);
struct msg *rsp = rsp_send_next(ctx, conn);
if (rsp != NULL && conn->dyn_mode) {
struct msg *pmsg = rsp->peer;
//need to deal with multi-block later
uint64_t msg_id = pmsg->dmsg->id;
struct mbuf *header_buf = mbuf_get();
if (header_buf == NULL) {
loga("Unable to obtain an mbuf for header!");
return NULL; //need to address error here properly
}
dmsg_type_t msg_type = DMSG_RES;
//TODOs: need to set the outcoming conn to be secured too if the incoming conn is secured
if (pmsg->owner->dnode_secured || conn->dnode_secured) {
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, "Encrypting response ...");
loga("AES encryption key: %s\n", base64_encode(conn->aes_key, AES_KEYLEN));
}
if (ENCRYPTION) {
status = dyn_aes_encrypt_msg(rsp, conn->aes_key);
if (status == DN_ERROR) {
loga("OOM to obtain an mbuf for encryption!");
mbuf_put(header_buf);
req_put(rsp);
return NULL;
}
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "#encrypted bytes : %d", status);
}
dmsg_write(header_buf, msg_id, msg_type, conn, msg_length(rsp));
} else {
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "no encryption on the rsp payload");
}
dmsg_write(header_buf, msg_id, msg_type, conn, msg_length(rsp));
}
} else {
//write dnode header
log_info("sending dnode response with msg_id %u", msg_id);
dmsg_write(header_buf, msg_id, msg_type, conn, msg_length(rsp));
}
mbuf_insert_head(&rsp->mhdr, header_buf);
if (log_loggable(LOG_VVERB)) {
log_hexdump(LOG_VVERB, header_buf->pos, mbuf_length(header_buf), "resp dyn message - header: ");
msg_dump(rsp);
}
}
return rsp;
}
/* dnode sends a response back to a peer */
static struct msg *
dnode_rsp_send_next(struct context *ctx, struct conn *conn)
{
rstatus_t status;
// SMB: There is some non trivial thing happening here. And I think it is very
// important to read this before anything is changed in here. There is also a
// bug that exists which I will mention briefly:
// A message is a structure that has a list of mbufs which hold the actual data.
// Each mbuf has start, pos, last as pointers (amongst others) which indicate start of the
// buffer, current read position and end of the buffer respectively.
//
// Every time a message is sent to a peer within dynomite, a DNODE header is
// prepended which is created using dmsg_write. A message remembers this case
// in dnode_header_prepended, so that if the messsage is sent in parts, the
// header is not prepended again for the subsequent parts.
//
// Like I said earlier there is a pos pointer in mbuf. If a message is sent
// partially (or it is parsed partially too I think) the pos reflects that
// case such that things can be resumed where it left off.
//
// dmsg_write has a parameter which reflects the payload length following the
// dnode header calculated by msg_length. msg_length is a summation of all
// mbuf sizes (last - start). Which I think is wrong.
//
// +------------+ +---------------+
// | DC1N1 +---------> | DC2N1 |
// +------------+ +-------+-------+
// |
// |
// |
// |
// +-------v-------+
// | DC2N2 |
// +---------------+
//
// Consider the case where
// a node DC1N1 in region DC1 sends a request to DC2N1 which forwards it to
// to local token owner DC2N2. Now DC2N1 receives a response from DC2N2 which
// has to be relayed back to DC1N1. This response from DC2N2 already has a
// dnode header but for the link between DC2N1 and DC2N2. DC2N1 should strip
// this header and prepend its own header for sending it back to DC1N1. This
// gets handled in encryption case since we overwrite all mbufs in the response
// However if the encryption is off, the message length sent to dmsg_write
// consists of the header from DC2N2 also which is wrong. So this relaying
// of responses will not work for the case where encryption is disabled.
//
// So msg_length should really be from mbuf->pos and not mbuf->start. This
// is a problem only with remote region replication since that is the only
// case where we CAN have 2 hops to send the request/response. This is also
// not a problem if encryption is ON.
ASSERT(conn->type == CONN_DNODE_PEER_CLIENT);
struct msg *rsp = rsp_send_next(ctx, conn);
if (rsp != NULL && conn->dyn_mode) {
struct msg *pmsg = rsp->peer;
//need to deal with multi-block later
uint64_t msg_id = pmsg->dmsg->id;
if (rsp->dnode_header_prepended) {
return rsp;
}
struct mbuf *header_buf = mbuf_get();
if (header_buf == NULL) {
loga("Unable to obtain an mbuf for header!");
return NULL; //need to address error here properly
}
dmsg_type_t msg_type = DMSG_RES;
//TODOs: need to set the outcoming conn to be secured too if the incoming conn is secured
if (pmsg->owner->dnode_secured || conn->dnode_secured) {
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, "Encrypting response ...");
loga("AES encryption key: %s\n", base64_encode(conn->aes_key, AES_KEYLEN));
}
if (ENCRYPTION) {
status = dyn_aes_encrypt_msg(rsp, conn->aes_key);
if (status == DN_ERROR) {
loga("OOM to obtain an mbuf for encryption!");
mbuf_put(header_buf);
req_put(rsp);
return NULL;
}
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "#encrypted bytes : %d", status);
}
dmsg_write(header_buf, msg_id, msg_type, conn, msg_length(rsp));
} else {
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VERB, "no encryption on the rsp payload");
}
dmsg_write(header_buf, msg_id, msg_type, conn, msg_length(rsp));
}
} else {
//write dnode header
log_debug(LOG_VERB, "sending dnode response with msg_id %u", msg_id);
dmsg_write(header_buf, msg_id, msg_type, conn, msg_length(rsp));
}
rsp->dnode_header_prepended = 1;
mbuf_insert_head(&rsp->mhdr, header_buf);
if (log_loggable(LOG_VVERB)) {
log_hexdump(LOG_VVERB, header_buf->pos, mbuf_length(header_buf), "resp dyn message - header: ");
msg_dump(rsp);
}
}
return rsp;
}
/*
* copy one response from src to dst
* return bytes copied
* */
static rstatus_t
memcache_copy_bulk(struct msg *dst, struct msg *src)
{
struct mbuf *mbuf, *nbuf;
uint8_t *p;
uint32_t len = 0;
uint32_t bytes = 0;
uint32_t i = 0;
for (mbuf = STAILQ_FIRST(&src->mhdr);
mbuf && mbuf_empty(mbuf);
mbuf = STAILQ_FIRST(&src->mhdr)) {
mbuf_remove(&src->mhdr, mbuf);
mbuf_put(mbuf);
}
mbuf = STAILQ_FIRST(&src->mhdr);
if (mbuf == NULL) {
return NC_OK; /* key not exists */
}
p = mbuf->pos;
/* get : VALUE key 0 len\r\nval\r\n */
/* gets: VALUE key 0 len cas\r\nval\r\n */
ASSERT(*p == 'V');
for (i = 0; i < 3; i++) { /* eat 'VALUE key 0 ' */
for (; *p != ' ';) {
p++;
}
p++;
}
len = 0;
for (; p < mbuf->last && isdigit(*p); p++) {
len = len * 10 + (uint32_t)(*p - '0');
}
for (; p < mbuf->last && ('\r' != *p); p++) { /* eat cas for gets */
;
}
len += CRLF_LEN * 2;
len += (p - mbuf->pos);
bytes = len;
/* copy len bytes to dst */
for (; mbuf;) {
if (mbuf_length(mbuf) <= len) { /* steal this mbuf from src to dst */
nbuf = STAILQ_NEXT(mbuf, next);
mbuf_remove(&src->mhdr, mbuf);
mbuf_insert(&dst->mhdr, mbuf);
len -= mbuf_length(mbuf);
mbuf = nbuf;
} else { /* split it */
nbuf = mbuf_get();
if (nbuf == NULL) {
return NC_ENOMEM;
}
mbuf_copy(nbuf, mbuf->pos, len);
mbuf_insert(&dst->mhdr, nbuf);
mbuf->pos += len;
break;
}
}
dst->mlen += bytes;
src->mlen -= bytes;
log_debug(LOG_VVERB, "memcache_copy_bulk copy bytes: %d", bytes);
return NC_OK;
}
/* Forward a client request over to a peer */
void
dnode_peer_req_forward(struct context *ctx, struct conn *c_conn, struct conn *p_conn,
struct msg *msg, struct rack *rack, uint8_t *key, uint32_t keylen)
{
if (TRACING_LEVEL == LOG_VVERB) {
log_debug(LOG_VVERB, "dnode_peer_req_forward entering");
}
rstatus_t status;
/* enqueue message (request) into client outq, if response is expected */
if (!msg->noreply) {
c_conn->enqueue_outq(ctx, c_conn, msg);
}
ASSERT(!p_conn->dnode_client && !p_conn->dnode_server);
ASSERT(c_conn->client);
/* enqueue the message (request) into peer inq */
if (TAILQ_EMPTY(&p_conn->imsg_q)) {
status = event_add_out(ctx->evb, p_conn);
if (status != DN_OK) {
dnode_req_forward_error(ctx, p_conn, msg);
p_conn->err = errno;
return;
}
}
uint64_t msg_id = peer_msg_id++;
struct mbuf *header_buf = mbuf_get();
if (header_buf == NULL) {
loga("Unable to obtain an mbuf for dnode msg's header!");
return;
}
if (p_conn->dnode_secured) {
//Encrypting and adding header for a request
struct mbuf *data_buf = STAILQ_LAST(&msg->mhdr, mbuf, next);
//TODOs: need to deal with multi-block later
log_debug(LOG_VERB, "AES encryption key: %s\n", base64_encode(p_conn->aes_key, AES_KEYLEN));
struct mbuf *encrypted_buf = mbuf_get();
if (encrypted_buf == NULL) {
loga("Unable to obtain an mbuf for encryption!");
return; //TODOs: need to clean up
}
status = dyn_aes_encrypt(data_buf->pos, mbuf_length(data_buf), encrypted_buf, p_conn->aes_key);
log_debug(LOG_VERB, "#encrypted bytes : %d", status);
//write dnode header
dmsg_write(header_buf, msg_id, DMSG_REQ, p_conn, mbuf_length(encrypted_buf));
mbuf_insert_head(&msg->mhdr, header_buf);
log_hexdump(LOG_VERB, data_buf->pos, mbuf_length(data_buf), "dyn message original payload: ");
log_hexdump(LOG_VERB, encrypted_buf->pos, mbuf_length(encrypted_buf), "dyn message encrypted payload: ");
//remove the original dbuf out of the queue and insert encrypted mbuf to replace
mbuf_remove(&msg->mhdr, data_buf);
mbuf_insert(&msg->mhdr, encrypted_buf);
//free it as no one will need it again
mbuf_put(data_buf);
} else {
//write dnode header
dmsg_write(header_buf, msg_id, DMSG_REQ, p_conn, 0);
mbuf_insert_head(&msg->mhdr, header_buf);
}
if (TRACING_LEVEL == LOG_VVERB) {
log_hexdump(LOG_VVERB, header_buf->pos, mbuf_length(header_buf), "dyn message header: ");
msg_dump(msg);
}
p_conn->enqueue_inq(ctx, p_conn, msg);
dnode_peer_req_forward_stats(ctx, p_conn->owner, msg);
if (TRACING_LEVEL == LOG_VERB) {
log_debug(LOG_VERB, "remote forward from c %d to s %d req %"PRIu64" len %"PRIu32
" type %d with key '%.*s'", c_conn->sd, p_conn->sd, msg->id,
msg->mlen, msg->type, keylen, key);
}
}
static bool
dyn_parse_core(struct msg *r)
{
struct dmsg *dmsg;
struct mbuf *b;
uint8_t *p, *token;
uint8_t ch = ' ';
uint64_t num = 0;
dyn_state = r->dyn_state;
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "dyn_state: %d", r->dyn_state);
}
if (r->dyn_state == DYN_DONE || r->dyn_state == DYN_POST_DONE)
return true;
b = STAILQ_LAST(&r->mhdr, mbuf, next);
dmsg = r->dmsg;
if (dmsg == NULL) {
r->dmsg = dmsg_get();
dmsg = r->dmsg;
dmsg->owner = r;
if (dmsg == NULL) {//should track this as a dropped message
loga("unable to create a new dmsg");
goto error; //should count as OOM error
}
}
token = NULL;
for (p = r->pos; p < b->last; p++) {
ch = *p;
switch (dyn_state) {
case DYN_START:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_START");
}
if (ch != ' ' && ch != '$') {
break;
}
if (ch == ' ') {
if (token == NULL)
token = p;
break;
}
if (ch == '$') {
if (p + 5 < b->last) {
if ((*(p+1) == '2') &&
(*(p+2) == '0') &&
(*(p+3) == '1') &&
(*(p+4) == '4') &&
(*(p+5) == '$')) {
dyn_state = DYN_MAGIC_STRING;
p += 5;
} else {
//goto skip;
token = NULL; //reset
}
} else {
goto split;
}
} else {
loga("Facing a weird char %c", p);
//goto skip;
token = NULL; //reset
}
break;
case DYN_MAGIC_STRING:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_MAGIC_STRING");
}
if (ch == ' ') {
dyn_state = DYN_MSG_ID;
num = 0;
break;
} else {
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
token = NULL;
loga("Facing a weird char %c", p);
//goto skip;
dyn_state = DYN_START;
}
break;
case DYN_MSG_ID:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_MSG_ID");
log_debug(LOG_DEBUG, "num = %d", num);
}
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else if (ch == ' ' && isdigit(*(p-1))) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "MSG ID : %d", num);
}
dmsg->id = num;
dyn_state = DYN_TYPE_ID;
num = 0;
} else {
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
//goto skip;
token = NULL; //reset
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_TYPE_ID:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_TYPE_ID: num = %d", num);
}
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else if (ch == ' ' && isdigit(*(p-1))) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Type Id: %d", num);
}
dmsg->type = num;
dyn_state = DYN_BIT_FIELD;
num = 0;
} else {
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
token = NULL;
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_BIT_FIELD:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_BIT_FIELD, num = %d", num);
}
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else if (ch == ' ' && isdigit(*(p-1))) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_BIT_FIELD : %d", num);
}
dmsg->bit_field = num & 0xF;
dyn_state = DYN_VERSION;
num = 0;
} else {
token = NULL;
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_VERSION:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_VERSION: num = %d", num);
}
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else if (ch == ' ' && isdigit(*(p-1))) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "VERSION : %d", num);
}
dmsg->version = num;
dyn_state = DYN_SAME_DC;
num = 0;
} else {
token = NULL;
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_SAME_DC:
if (isdigit(ch)) {
dmsg->same_dc = ch - '0';
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_SAME_DC %d", dmsg->same_dc);
}
} else if (ch == ' ' && isdigit(*(p-1))) {
dyn_state = DYN_DATA_LEN;
num = 0;
} else {
token = NULL;
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_DATA_LEN:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_DATA_LEN: num = %d", num);
}
if (ch == '*') {
break;
} else if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else if (ch == ' ' && isdigit(*(p-1))) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Data len: %d", num);
}
dmsg->mlen = num;
dyn_state = DYN_DATA;
num = 0;
} else {
token = NULL;
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_DATA:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_DATA");
}
if (p + dmsg->mlen < b->last) {
dmsg->data = p;
p += dmsg->mlen - 1;
dyn_state = DYN_SPACES_BEFORE_PAYLOAD_LEN;
} else {
//loga("char is '%c %c %c %c'", *(p-2), *(p-1), ch, *(p+1));
goto split;
}
break;
case DYN_SPACES_BEFORE_PAYLOAD_LEN:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_SPACES_BEFORE_PAYLOAD_LEN");
}
if (ch == ' ') {
break;
} else if (ch == '*') {
dyn_state = DYN_PAYLOAD_LEN;
num = 0;
}
break;
case DYN_PAYLOAD_LEN:
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else if (ch == CR) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Payload len: %d", num);
}
dmsg->plen = num;
num = 0;
dyn_state = DYN_CRLF_BEFORE_DONE;
} else {
token = NULL;
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_CRLF_BEFORE_DONE:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_CRLF_BEFORE_DONE");
}
if (*p == LF) {
dyn_state = DYN_DONE;
} else {
token = NULL;
dyn_state = DYN_START;
if (ch == '$')
p -= 1;
}
break;
case DYN_DONE:
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "DYN_DONE");
}
r->pos = p;
dmsg->payload = p;
r->dyn_state = DYN_DONE;
b->pos = p;
goto done;
break;
default:
NOT_REACHED();
break;
}
}
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Not fully parsed yet!!!!!!");
}
split:
//this is an attempt recovery when we got a bad message
//we try to look for the start the next good one and throw away the bad part
if (r->dyn_state == DYN_START) {
r->result = MSG_PARSE_AGAIN;
if (b->last == b->end) {
struct mbuf *nbuf = mbuf_get();
if (nbuf == NULL) {
loga("Unable to obtain a new mbuf for replacement!");
mbuf_put(b);
nbuf = mbuf_get();
mbuf_insert_head(&r->mhdr, nbuf);
r->pos = nbuf->pos;
return false;
}
//replacing the bad mbuf with a new and empty mbuf
mbuf_insert(&r->mhdr, nbuf);
mbuf_remove(&r->mhdr, b);
mbuf_put(b);
r->pos = nbuf->pos;
return false;
} else { //split it and throw away the bad portion
struct mbuf *nbuf;
nbuf = mbuf_split(&r->mhdr, r->pos, NULL, NULL);
if (nbuf == NULL) {
return DN_ENOMEM;
}
mbuf_insert(&r->mhdr, nbuf);
mbuf_remove(&r->mhdr, b);
r->pos = nbuf->pos;
return false;
}
}
if (mbuf_length(b) == 0 || b->last == b->end) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Would this case ever happen?");
}
r->result = MSG_PARSE_AGAIN;
return false;
}
if (r->pos == b->last) {
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Forward to reading the new block of data");
}
r->dyn_state = DYN_START;
r->result = MSG_PARSE_AGAIN;
token = NULL;
return false;
}
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, "in split");
}
r->dyn_state = DYN_START;
r->pos = token;
r->result = MSG_PARSE_REPAIR;
if (log_loggable(LOG_VVERB)) {
log_hexdump(LOG_VVERB, b->pos, mbuf_length(b), "split and inspecting req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
r->dyn_state);
log_hexdump(LOG_VVERB, b->start, b->last - b->start, "split and inspecting full req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
r->dyn_state);
}
return false;
done:
r->pos = p;
dmsg->source_address = r->owner->addr;
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, "at done with p at %d", p);
log_hexdump(LOG_VVERB, r->pos, b->last - r->pos, "done and inspecting req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
r->dyn_state);
log_hexdump(LOG_VVERB, b->start, b->last - b->start, "inspecting req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
r->dyn_state);
}
return true;
error:
log_debug(LOG_ERR, "at error for state %d and c %c", dyn_state, *p);
r->result = MSG_PARSE_ERROR;
r->pos = p;
errno = EINVAL;
if (log_loggable(LOG_ERR)) {
log_hexdump(LOG_ERR, b->pos, mbuf_length(b), "parsed bad req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
dyn_state);
log_hexdump(LOG_ERR, p, b->last - p, "inspecting req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
dyn_state);
}
r->dyn_state = dyn_state;
return false;
}
/*
* Sending a mbuf of gossip data over the wire to a peer
*/
void
dnode_peer_gossip_forward(struct context *ctx, struct conn *conn, bool redis, struct mbuf *data_buf)
{
rstatus_t status;
struct msg *msg = msg_get(conn, 1, redis);
if (msg == NULL) {
log_debug(LOG_DEBUG, "Unable to obtain a msg");
return;
}
struct mbuf *header_buf = mbuf_get();
if (header_buf == NULL) {
log_debug(LOG_DEBUG, "Unable to obtain a data_buf");
msg_put(msg);
return;
}
uint64_t msg_id = peer_msg_id++;
if (conn->dnode_secured) {
log_debug(LOG_VERB, "Assemble a secured msg to send");
log_debug(LOG_VERB, "AES encryption key: %s\n", base64_encode(conn->aes_key, AES_KEYLEN));
struct mbuf *encrypted_buf = mbuf_get();
if (encrypted_buf == NULL) {
loga("Unable to obtain an data_buf for encryption!");
return; //TODOs: need to clean up
}
status = dyn_aes_encrypt(data_buf->pos, mbuf_length(data_buf), encrypted_buf, conn->aes_key);
log_debug(LOG_VERB, "#encrypted bytes : %d", status);
//write dnode header
dmsg_write(header_buf, msg_id, GOSSIP_SYN, conn, mbuf_length(encrypted_buf));
mbuf_insert_head(&msg->mhdr, header_buf);
if (TRACING_LEVEL == LOG_VVERB) {
log_hexdump(LOG_VVERB, data_buf->pos, mbuf_length(data_buf), "dyn message original payload: ");
log_hexdump(LOG_VVERB, encrypted_buf->pos, mbuf_length(encrypted_buf), "dyn message encrypted payload: ");
}
mbuf_insert(&msg->mhdr, encrypted_buf);
//free data_buf as no one will need it again
mbuf_put(data_buf);
} else {
log_debug(LOG_VERB, "Assemble a non-secured msg to send");
dmsg_write_mbuf(header_buf, msg_id, GOSSIP_SYN, conn, mbuf_length(data_buf));
mbuf_insert_head(&msg->mhdr, header_buf);
mbuf_insert(&msg->mhdr, data_buf);
}
if (TRACING_LEVEL == LOG_VVERB) {
log_hexdump(LOG_VVERB, header_buf->pos, mbuf_length(header_buf), "dyn gossip message header: ");
msg_dump(msg);
}
/* enqueue the message (request) into peer inq */
if (TAILQ_EMPTY(&conn->imsg_q)) {
status = event_add_out(ctx->evb, conn);
if (status != DN_OK) {
dnode_req_forward_error(ctx, conn, msg);
conn->err = errno;
return;
}
}
//need to handle a reply
//conn->enqueue_outq(ctx, conn, msg);
msg->noreply = 1;
conn->enqueue_inq(ctx, conn, msg);
}
void dyn_parse_rsp(struct msg *r)
{
if (log_loggable(LOG_VVERB)) {
log_debug(LOG_VVERB, ":::::::::::::::::::::: In dyn_parse_rsp, start to process response :::::::::::::::::::::::: ");
msg_dump(r);
}
bool done_parsing = false;
struct mbuf *b = STAILQ_LAST(&r->mhdr, mbuf, next);
if (dyn_parse_core(r)) {
struct dmsg *dmsg = r->dmsg;
struct conn *conn = r->owner;
conn->same_dc = dmsg->same_dc;
if (dmsg->type != DMSG_UNKNOWN && dmsg->type != DMSG_RES) {
log_debug(LOG_DEBUG, "Resp parser: I got a dnode msg of type %d", dmsg->type);
r->state = 0;
r->result = MSG_PARSE_OK;
r->dyn_state = DYN_DONE;
return;
}
if (r->dyn_state == DYN_DONE && dmsg->bit_field == 1) {
dmsg->owner->owner->dnode_secured = 1;
r->owner->dnode_crypto_state = 1;
r->dyn_state = DYN_POST_DONE;
r->result = MSG_PARSE_REPAIR;
if (dmsg->mlen > 1) {
//Decrypt AES key
dyn_rsa_decrypt(dmsg->data, aes_decrypted_buf);
strncpy(r->owner->aes_key, aes_decrypted_buf, strlen(aes_decrypted_buf));
}
if (dmsg->plen + b->pos <= b->last) {
struct mbuf *decrypted_buf = mbuf_get();
if (decrypted_buf == NULL) {
loga("Unable to obtain an mbuf for dnode msg's header!");
r->result = MSG_OOM_ERROR;
return;
}
dyn_aes_decrypt(b->pos, dmsg->plen, decrypted_buf, r->owner->aes_key);
b->pos = b->pos + dmsg->plen;
r->pos = decrypted_buf->start;
mbuf_copy(decrypted_buf, b->pos, mbuf_length(b));
mbuf_insert(&r->mhdr, decrypted_buf);
mbuf_remove(&r->mhdr, b);
mbuf_put(b);
r->mlen = mbuf_length(decrypted_buf);
return data_store_parse_rsp(r);
}
//Subtract already received bytes
dmsg->plen -= b->last - b->pos;
return;
} else if (r->dyn_state == DYN_POST_DONE) {
struct mbuf *last_buf = STAILQ_LAST(&r->mhdr, mbuf, next);
if (last_buf->read_flip == 1) {
data_store_parse_rsp(r);
} else {
r->result = MSG_PARSE_AGAIN;
}
return;
}
if (done_parsing)
return;
return data_store_parse_rsp(r);
}
//bad case
if (log_loggable(LOG_DEBUG)) {
log_debug(LOG_DEBUG, "Resp: bad message - cannot parse"); //fix me to do something
msg_dump(r);
}
r->result = MSG_PARSE_AGAIN;
}
static void
rsp_forward(struct context *ctx, struct conn *s_conn, struct msg *msg)
{
rstatus_t status;
struct msg *pmsg;
struct conn *c_conn;
ASSERT(!s_conn->client && !s_conn->proxy);
/* response from server implies that server is ok and heartbeating */
server_ok(ctx, s_conn);
/* dequeue peer message (request) from server */
pmsg = TAILQ_FIRST(&s_conn->omsg_q);
ASSERT(pmsg != NULL && pmsg->peer == NULL);
ASSERT(pmsg->request && !pmsg->done);
s_conn->dequeue_outq(ctx, s_conn, pmsg);
pmsg->done = 1;
/* establish msg <-> pmsg (response <-> request) link */
pmsg->peer = msg;
msg->peer = pmsg;
/*
* Readjust responses of fragmented messages by not including the end
* marker for all but the last response
*
* Valid responses for a fragmented requests are MSG_RSP_VALUE or,
* MSG_RSP_END. For an invalid response, we send out SERVER_ERRROR with
* EINVAL errno
*/
if (pmsg->frag_id != 0) {
if (msg->type != MSG_RSP_VALUE && msg->type != MSG_RSP_END) {
pmsg->error = 1;
pmsg->err = EINVAL;
} else if (!pmsg->last_fragment) {
ASSERT(msg->end != NULL);
for (;;) {
struct mbuf *mbuf;
mbuf = STAILQ_LAST(&msg->mhdr, mbuf, next);
ASSERT(mbuf != NULL);
/*
* We cannot assert that end marker points to the last mbuf
* Consider a scenario where end marker points to the
* penultimate mbuf and the last mbuf only contains spaces
* and CRLF: mhdr -> [...END] -> [\r\n]
*/
if (msg->end >= mbuf->pos && msg->end < mbuf->last) {
/* end marker is within this mbuf */
msg->mlen -= (uint32_t)(mbuf->last - msg->end);
mbuf->last = msg->end;
break;
}
/* end marker is not in this mbuf */
msg->mlen -= mbuf_length(mbuf);
mbuf_remove(&msg->mhdr, mbuf);
mbuf_put(mbuf);
}
}
}
c_conn = pmsg->owner;
ASSERT(c_conn->client && !c_conn->proxy);
if (req_done(c_conn, TAILQ_FIRST(&c_conn->omsg_q))) {
status = event_add_out(ctx->ep, c_conn);
if (status != NC_OK) {
c_conn->err = errno;
}
}
rsp_forward_stats(ctx, s_conn->owner, msg);
}
/* dnode sends a response back to a peer */
struct msg *
dnode_rsp_send_next(struct context *ctx, struct conn *conn)
{
if (TRACING_LEVEL == LOG_VVERB) {
log_debug(LOG_VVERB, "dnode_rsp_send_next entering");
}
ASSERT(conn->dnode_client && !conn->dnode_server);
struct msg *msg = rsp_send_next(ctx, conn);
if (msg != NULL && conn->dyn_mode) {
struct msg *pmsg = TAILQ_FIRST(&conn->omsg_q); //peer request's msg
//need to deal with multi-block later
uint64_t msg_id = pmsg->dmsg->id;
struct mbuf *header_buf = mbuf_get();
if (header_buf == NULL) {
loga("Unable to obtain an mbuf for header!");
return NULL; //need to address error here properly
}
//TODOs: need to set the outcoming conn to be secured too if the incoming conn is secured
if (pmsg->owner->dnode_secured || conn->dnode_secured) {
if (TRACING_LEVEL == LOG_VVERB) {
log_debug(LOG_VVERB, "Encrypting response ...");
loga("AES encryption key: %s\n", base64_encode(conn->aes_key, AES_KEYLEN));
}
struct mbuf *data_buf = STAILQ_LAST(&msg->mhdr, mbuf, next);
//if (ENCRYPTION) {
struct mbuf *encrypted_buf = mbuf_get();
if (encrypted_buf == NULL) {
loga("Unable to obtain an mbuf for encryption!");
return NULL; //TODOs: need to clean up
}
rstatus_t status = dyn_aes_encrypt(data_buf->pos, mbuf_length(data_buf),
encrypted_buf, conn->aes_key);
if (TRACING_LEVEL == LOG_VVERB) {
log_debug(LOG_VERB, "#encrypted bytes : %d", status);
}
dmsg_write(header_buf, msg_id, DMSG_RES, conn, mbuf_length(encrypted_buf));
if (TRACING_LEVEL == LOG_VVERB) {
log_hexdump(LOG_VVERB, data_buf->pos, mbuf_length(data_buf), "resp dyn message - original payload: ");
log_hexdump(LOG_VVERB, encrypted_buf->pos, mbuf_length(encrypted_buf), "dyn message encrypted payload: ");
}
mbuf_copy(header_buf, encrypted_buf->start, mbuf_length(encrypted_buf));
mbuf_insert(&msg->mhdr, header_buf);
//remove the original dbuf out of the queue and insert encrypted mbuf to replace
mbuf_remove(&msg->mhdr, data_buf);
//mbuf_insert(&msg->mhdr, encrypted_buf);
mbuf_put(data_buf);
mbuf_put(encrypted_buf);
//} else {
// log_debug(LOG_VERB, "no encryption on the response's payload");
// dmsg_write(header_buf, msg_id, DMSG_RES, conn, mbuf_length(data_buf));
//}
} else {
dmsg_write(header_buf, msg_id, DMSG_RES, conn, 0);//Dont care about 0 or the real length as we don't use that value in unencryption mode
mbuf_insert_head(&msg->mhdr, header_buf);
}
if (TRACING_LEVEL == LOG_VVERB) {
log_hexdump(LOG_VVERB, header_buf->pos, mbuf_length(header_buf), "resp dyn message - header: ");
msg_dump(msg);
}
}
return msg;
}
static void *check_thread_run(void *args)
{
int ret;
thread_data *cdata = args;
redis_group *srgroup = cdata->srgroup;
redis_group *trgroup = cdata->trgroup;
dict *nodes;
dictEntry *de;
dictIterator *di;
redis_node *rnode;
struct mbuf *mbuf;
nodes = srgroup->nodes;
di = dictGetIterator(nodes);
while ((de = dictNext(di)) != NULL) {
rnode = dictGetVal(de);
rnode->write_data = cdata;
/* remove the not used part for source redis node */
if (rnode->rdb != NULL) {
redis_rdb_deinit(rnode->rdb);
rmt_free(rnode->rdb);
rnode->rdb = NULL;
}
if (rnode->cmd_data != NULL) {
while (!mttlist_empty(rnode->cmd_data)) {
mbuf = mttlist_pop(rnode->cmd_data);
mbuf_put(mbuf);
}
mttlist_destroy(rnode->cmd_data);
rnode->cmd_data = NULL;
}
if (rnode->sockpairfds[0] > 0) {
close(rnode->sockpairfds[0]);
rnode->sockpairfds[0] = -1;
}
if (rnode->sockpairfds[1] > 0) {
close(rnode->sockpairfds[1]);
rnode->sockpairfds[1] = -1;
}
if (rnode->rr != NULL) {
redis_replication_deinit(rnode->rr);
rmt_free(rnode->rr);
rnode->rr = NULL;
}
if (rnode->piece_data != NULL) {
while ((mbuf = listPop(rnode->piece_data)) != NULL) {
mbuf_put(mbuf);
}
listRelease(rnode->piece_data);
rnode->piece_data = NULL;
}
/* add the used part for source redis node */
if (rnode->send_data == NULL) {
rnode->send_data = listCreate();
if (rnode->send_data == NULL) {
log_error("ERROR: Create msg list failed: out of memory");
return 0;
}
}
if (rnode->sent_data == NULL) {
rnode->sent_data = listCreate();
if (rnode->sent_data == NULL) {
log_error("ERROR: Create msg list failed: out of memory");
return 0;
}
}
if (rnode->sk_event < 0) {
rnode->sk_event = socket(AF_INET, SOCK_STREAM, 0);
if(rnode->sk_event < 0){
log_error("ERROR: Create sk_event for node[%s] failed: %s",
rnode->addr, strerror(errno));
return 0;
}
}
ret = aeCreateFileEvent(cdata->loop, rnode->sk_event,
AE_WRITABLE, check_begin, rnode);
if (ret != AE_OK) {
log_error("ERROR: send_data event create %ld failed: %s",
cdata->thread_id, strerror(errno));
return 0;
}
}
dictReleaseIterator(di);
nodes = trgroup->nodes;
di = dictGetIterator(nodes);
while ((de = dictNext(di)) != NULL) {
rnode = dictGetVal(de);
rnode->write_data = cdata;
}
dictReleaseIterator(di);
aeMain(cdata->loop);
return 0;
}
