struct command *conn_get_cmd(struct connection *client)
{
struct command *cmd;
int reuse = 0;
if (!STAILQ_EMPTY(&client->info->cmd_queue)) {
cmd = STAILQ_LAST(&client->info->cmd_queue, command, cmd_next);
if (!cmd->parse_done) {
reuse = 1;
}
}
if (!reuse) {
cmd = cmd_create(client->ctx);
STAILQ_INSERT_TAIL(&client->info->cmd_queue, cmd, cmd_next);
}
return cmd;
}
elf_strtab_entry *
elf_strtab_append_str(elf_strtab_head *strtab, const char *str)
{
elf_strtab_entry *last, *entry;
if (strtab == NULL)
yasm_internal_error("strtab is null");
if (STAILQ_EMPTY(strtab))
yasm_internal_error("strtab is missing initial dummy entry");
last = STAILQ_LAST(strtab, elf_strtab_entry, qlink);
entry = elf_strtab_entry_create(str);
entry->index = last->index + (unsigned long)strlen(last->str) + 1;
STAILQ_INSERT_TAIL(strtab, entry, qlink);
return entry;
}
/* Push a new entry into the slow log.
* This function will make sure to trim the slow log accordingly to the
* configured max length.
* The unit of duration is microseconds */
void slowlog_push_entry_if_needed(struct msg *r, long long duration) {
if (slowlog_log_slower_than < 0) return; /* Slowlog disabled */
if (duration >= slowlog_log_slower_than) {
slowlog_entry *se = slowlog_create_entry(r,duration);
pthread_rwlock_wrlock(&rwlocker);
se->id = slowlog_entry_id++;
STAILQ_INSERT_HEAD(&slowlog, se, next);
if (slowlog_len >= slowlog_max_len) {
se = STAILQ_LAST(&slowlog, slowlog_entry, next);
STAILQ_REMOVE(&slowlog, se, slowlog_entry, next);
slowlog_free_entry(se);
} else {
slowlog_len ++;
}
pthread_rwlock_unlock(&rwlocker);
slowlog_statistics_input();
}
}
static void
direct_reply(struct context *ctx, struct conn *conn, struct msg *smsg, char *_msg) {
struct mbuf *mbuf;
int n;
struct msg *msg = msg_get(conn, true, conn->redis);
if (msg == NULL) {
conn->err = errno;
conn->done = 1;
return;
}
mbuf = STAILQ_LAST(&msg->mhdr, mbuf, next);
if (mbuf == NULL || mbuf_full(mbuf)) {
mbuf = mbuf_get();
if (mbuf != NULL) {
mbuf_insert(&msg->mhdr, mbuf);
msg->pos = mbuf->pos;
}
}
if (mbuf == NULL) {
conn->err = errno;
conn->done = 1;
msg_put(msg);
return;
}
smsg->peer = msg;
msg->peer = smsg;
msg->request = 0;
n = (int)strlen(_msg);
memcpy(mbuf->last, _msg, (size_t)n);
mbuf->last += n;
msg->mlen += (uint32_t)n;
smsg->done = 1;
event_add_out(ctx->evb, conn);
conn->enqueue_outq(ctx, conn, smsg);
}
/*
* Split mbuf h into h and t by copying data from h to t. Before
* the copy, we invoke a precopy handler cb that will copy a predefined
* string to the head of t.
*
* Return new mbuf t, if the split was successful.
*/
struct mbuf *
mbuf_split(struct mhdr *h, uint8_t *pos, func_mbuf_copy_t cb, void *cbarg)
{
struct mbuf *mbuf, *nbuf;
size_t size;
ASSERT(!STAILQ_EMPTY(h));
mbuf = STAILQ_LAST(h, mbuf, next);
//ASSERT(pos >= mbuf->pos && pos <= mbuf->last);
if (pos < mbuf->pos || pos > mbuf->last)
return NULL;
nbuf = mbuf_get();
if (nbuf == NULL) {
return NULL;
}
if (cb != NULL) {
/* precopy nbuf */
cb(nbuf, cbarg);
}
/* copy data from mbuf to nbuf */
size = (size_t)(mbuf->last - pos);
mbuf_copy(nbuf, pos, size);
/* adjust mbuf */
mbuf->last = pos;
log_debug(LOG_VVERB, "split into mbuf %p len %"PRIu32" and nbuf %p len "
"%"PRIu32" copied %zu bytes", mbuf, mbuf_length(mbuf), nbuf,
mbuf_length(nbuf), size);
return nbuf;
}
/*
* note: we should not call conn_add_out here.
* because we may call append many times.
*/
rstatus_t
conn_sendq_append(struct conn *conn, char *pos, size_t n)
{
struct mbuf *mbuf;
size_t bytes = 0;
size_t len;
while (bytes < n) {
mbuf = STAILQ_LAST(&conn->send_queue, mbuf, next);
if ((mbuf == NULL) || mbuf_full(mbuf)) {
mbuf = mbuf_get();
if (mbuf == NULL) {
return NC_ENOMEM;
}
mbuf_insert(&conn->send_queue, mbuf);
}
len = MIN(mbuf_size(mbuf), n - bytes);
mbuf_copy(mbuf, (uint8_t *)pos + bytes, len);
bytes += len;
}
return NC_OK;
}
void
memcache_parse_rsp(struct msg *r)
{
struct mbuf *b;
uint8_t *p, *m;
uint8_t ch;
enum {
SW_START,
SW_RSP_NUM,
SW_RSP_STR,
SW_SPACES_BEFORE_KEY,
SW_KEY,
SW_SPACES_BEFORE_FLAGS, /* 5 */
SW_FLAGS,
SW_SPACES_BEFORE_VLEN,
SW_VLEN,
SW_RUNTO_VAL,
SW_VAL, /* 10 */
SW_VAL_LF,
SW_END,
SW_RUNTO_CRLF,
SW_CRLF,
SW_ALMOST_DONE, /* 15 */
SW_SENTINEL
} state;
state = r->state;
b = STAILQ_LAST(&r->mhdr, mbuf, next);
ASSERT(!r->request);
ASSERT(!r->redis);
ASSERT(state >= SW_START && state < SW_SENTINEL);
ASSERT(b != NULL);
ASSERT(b->pos <= b->last);
/* validate the parsing marker */
ASSERT(r->pos != NULL);
ASSERT(r->pos >= b->pos && r->pos <= b->last);
for (p = r->pos; p < b->last; p++) {
ch = *p;
switch (state) {
case SW_START:
if (isdigit(ch)) {
state = SW_RSP_NUM;
} else {
state = SW_RSP_STR;
}
p = p - 1; /* go back by 1 byte */
break;
case SW_RSP_NUM:
if (r->token == NULL) {
/* rsp_start <- p; type_start <- p */
r->token = p;
}
if (isdigit(ch)) {
/* num <- num * 10 + (ch - '0') */
;
} else if (ch == ' ' || ch == CR) {
/* type_end <- p - 1 */
r->token = NULL;
r->type = MSG_RSP_MC_NUM;
p = p - 1; /* go back by 1 byte */
state = SW_CRLF;
} else {
goto error;
}
break;
case SW_RSP_STR:
if (r->token == NULL) {
/* rsp_start <- p; type_start <- p */
r->token = p;
}
if (ch == ' ' || ch == CR) {
/* type_end <- p - 1 */
m = r->token;
/* r->token = NULL; */
r->type = MSG_UNKNOWN;
switch (p - m) {
case 3:
if (str4cmp(m, 'E', 'N', 'D', '\r')) {
r->type = MSG_RSP_MC_END;
/* end_start <- m; end_end <- p - 1 */
r->end = m;
break;
}
break;
case 5:
if (str5cmp(m, 'V', 'A', 'L', 'U', 'E')) {
/*
* Encompasses responses for 'get', 'gets' and
* 'cas' command.
*/
r->type = MSG_RSP_MC_VALUE;
break;
}
if (str5cmp(m, 'E', 'R', 'R', 'O', 'R')) {
r->type = MSG_RSP_MC_ERROR;
break;
}
break;
case 6:
if (str6cmp(m, 'S', 'T', 'O', 'R', 'E', 'D')) {
r->type = MSG_RSP_MC_STORED;
break;
}
if (str6cmp(m, 'E', 'X', 'I', 'S', 'T', 'S')) {
r->type = MSG_RSP_MC_EXISTS;
break;
}
break;
case 7:
if (str7cmp(m, 'D', 'E', 'L', 'E', 'T', 'E', 'D')) {
r->type = MSG_RSP_MC_DELETED;
break;
}
break;
case 9:
if (str9cmp(m, 'N', 'O', 'T', '_', 'F', 'O', 'U', 'N', 'D')) {
r->type = MSG_RSP_MC_NOT_FOUND;
break;
}
break;
case 10:
if (str10cmp(m, 'N', 'O', 'T', '_', 'S', 'T', 'O', 'R', 'E', 'D')) {
r->type = MSG_RSP_MC_NOT_STORED;
break;
}
break;
case 12:
if (str12cmp(m, 'C', 'L', 'I', 'E', 'N', 'T', '_', 'E', 'R', 'R', 'O', 'R')) {
r->type = MSG_RSP_MC_CLIENT_ERROR;
break;
}
if (str12cmp(m, 'S', 'E', 'R', 'V', 'E', 'R', '_', 'E', 'R', 'R', 'O', 'R')) {
r->type = MSG_RSP_MC_SERVER_ERROR;
break;
}
break;
}
switch (r->type) {
case MSG_UNKNOWN:
goto error;
case MSG_RSP_MC_STORED:
case MSG_RSP_MC_NOT_STORED:
case MSG_RSP_MC_EXISTS:
case MSG_RSP_MC_NOT_FOUND:
case MSG_RSP_MC_DELETED:
state = SW_CRLF;
break;
case MSG_RSP_MC_END:
state = SW_CRLF;
break;
case MSG_RSP_MC_VALUE:
state = SW_SPACES_BEFORE_KEY;
break;
case MSG_RSP_MC_ERROR:
state = SW_CRLF;
break;
case MSG_RSP_MC_CLIENT_ERROR:
case MSG_RSP_MC_SERVER_ERROR:
state = SW_RUNTO_CRLF;
break;
default:
NOT_REACHED();
}
p = p - 1; /* go back by 1 byte */
}
break;
case SW_SPACES_BEFORE_KEY:
if (ch != ' ') {
state = SW_KEY;
p = p - 1; /* go back by 1 byte */
}
break;
case SW_KEY:
if (ch == ' ') {
/* r->token = NULL; */
state = SW_SPACES_BEFORE_FLAGS;
}
break;
case SW_SPACES_BEFORE_FLAGS:
if (ch != ' ') {
if (!isdigit(ch)) {
goto error;
}
state = SW_FLAGS;
p = p - 1; /* go back by 1 byte */
}
break;
case SW_FLAGS:
if (r->token == NULL) {
/* flags_start <- p */
/* r->token = p; */
}
if (isdigit(ch)) {
/* flags <- flags * 10 + (ch - '0') */
;
} else if (ch == ' ') {
/* flags_end <- p - 1 */
/* r->token = NULL; */
state = SW_SPACES_BEFORE_VLEN;
} else {
goto error;
}
break;
case SW_SPACES_BEFORE_VLEN:
if (ch != ' ') {
if (!isdigit(ch)) {
goto error;
}
p = p - 1; /* go back by 1 byte */
state = SW_VLEN;
r->vlen = 0;
}
break;
case SW_VLEN:
if (isdigit(ch)) {
r->vlen = r->vlen * 10 + (uint32_t)(ch - '0');
} else if (ch == ' ' || ch == CR) {
/* vlen_end <- p - 1 */
p = p - 1; /* go back by 1 byte */
/* r->token = NULL; */
state = SW_RUNTO_CRLF;
} else {
goto error;
}
break;
case SW_RUNTO_VAL:
switch (ch) {
case LF:
/* val_start <- p + 1 */
state = SW_VAL;
r->token = NULL;
break;
default:
goto error;
}
break;
case SW_VAL:
m = p + r->vlen;
if (m >= b->last) {
ASSERT(r->vlen >= (uint32_t)(b->last - p));
r->vlen -= (uint32_t)(b->last - p);
m = b->last - 1;
p = m; /* move forward by vlen bytes */
break;
}
switch (*m) {
case CR:
/* val_end <- p - 1 */
p = m; /* move forward by vlen bytes */
state = SW_VAL_LF;
break;
default:
goto error;
}
break;
case SW_VAL_LF:
switch (ch) {
case LF:
/* state = SW_END; */
state = SW_RSP_STR;
break;
default:
goto error;
}
break;
case SW_END:
if (r->token == NULL) {
if (ch != 'E') {
goto error;
}
/* end_start <- p */
r->token = p;
} else if (ch == CR) {
/* end_end <- p */
m = r->token;
r->token = NULL;
switch (p - m) {
case 3:
if (str4cmp(m, 'E', 'N', 'D', '\r')) {
r->end = m;
state = SW_ALMOST_DONE;
}
break;
default:
goto error;
}
}
break;
case SW_RUNTO_CRLF:
switch (ch) {
case CR:
if (r->type == MSG_RSP_MC_VALUE) {
state = SW_RUNTO_VAL;
} else {
state = SW_ALMOST_DONE;
}
break;
default:
break;
}
break;
case SW_CRLF:
switch (ch) {
case ' ':
break;
case CR:
state = SW_ALMOST_DONE;
break;
default:
goto error;
}
break;
case SW_ALMOST_DONE:
switch (ch) {
case LF:
/* rsp_end <- p */
goto done;
default:
goto error;
}
break;
case SW_SENTINEL:
default:
NOT_REACHED();
break;
}
}
ASSERT(p == b->last);
r->pos = p;
r->state = state;
if (b->last == b->end && r->token != NULL) {
if (state <= SW_RUNTO_VAL || state == SW_CRLF || state == SW_ALMOST_DONE) {
r->state = SW_START;
}
r->pos = r->token;
r->token = NULL;
r->result = MSG_PARSE_REPAIR;
} else {
r->result = MSG_PARSE_AGAIN;
}
log_hexdump(LOG_VERB, b->pos, mbuf_length(b), "parsed rsp %"PRIu64" res %d "
"type %d state %d rpos %d of %d", r->id, r->result, r->type,
r->state, r->pos - b->pos, b->last - b->pos);
return;
done:
ASSERT(r->type > MSG_UNKNOWN && r->type < MSG_SENTINEL);
r->pos = p + 1;
ASSERT(r->pos <= b->last);
r->state = SW_START;
r->token = NULL;
r->result = MSG_PARSE_OK;
log_hexdump(LOG_VERB, b->pos, mbuf_length(b), "parsed rsp %"PRIu64" res %d "
"type %d state %d rpos %d of %d", r->id, r->result, r->type,
r->state, r->pos - b->pos, b->last - b->pos);
return;
error:
r->result = MSG_PARSE_ERROR;
r->state = state;
errno = EINVAL;
log_hexdump(LOG_INFO, b->pos, mbuf_length(b), "parsed bad rsp %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
r->state);
}
/* 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;
}
vt_buf_entry_t* vt_buf_last_entry(vt_buf_t *bp){
return STAILQ_LAST(&bp->entries, vt_buf_entry, entry);
}
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;
}
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;
}
void
memcache_parse_req(struct msg *r)
{
struct mbuf *b;
uint8_t *p, *m;
uint8_t ch;
enum {
SW_START,
SW_REQ_TYPE,
SW_SPACES_BEFORE_KEY,
SW_KEY,
SW_SPACES_BEFORE_KEYS,
SW_SPACES_BEFORE_FLAGS,
SW_FLAGS,
SW_SPACES_BEFORE_EXPIRY,
SW_EXPIRY,
SW_SPACES_BEFORE_VLEN,
SW_VLEN,
SW_SPACES_BEFORE_CAS,
SW_CAS,
SW_RUNTO_VAL,
SW_VAL,
SW_SPACES_BEFORE_NUM,
SW_NUM,
SW_RUNTO_CRLF,
SW_CRLF,
SW_NOREPLY,
SW_AFTER_NOREPLY,
SW_ALMOST_DONE,
SW_SENTINEL
} state;
state = r->state;
b = STAILQ_LAST(&r->mhdr, mbuf, next);
ASSERT(r->request);
ASSERT(state >= SW_START && state < SW_SENTINEL);
ASSERT(b != NULL);
ASSERT(b->pos <= b->last);
/* validate the parsing maker */
ASSERT(r->pos != NULL);
ASSERT(r->pos >= b->pos && r->pos <= b->last);
for (p = r->pos; p < b->last; p++) {
ch = *p;
switch (state) {
case SW_START:
if (ch == ' ') {
break;
}
if (!islower(ch)) {
goto error;
}
/* req_start <- p; type_start <- p */
r->token = p;
state = SW_REQ_TYPE;
break;
case SW_REQ_TYPE:
if (ch == ' ' || ch == CR) {
/* type_end = p - 1 */
m = r->token;
r->token = NULL;
r->type = MSG_UNKNOWN;
switch (p - m) {
case 3:
if (str4cmp(m, 'g', 'e', 't', ' ')) {
r->type = MSG_REQ_GET;
break;
}
if (str4cmp(m, 's', 'e', 't', ' ')) {
r->type = MSG_REQ_SET;
break;
}
if (str4cmp(m, 'a', 'd', 'd', ' ')) {
r->type = MSG_REQ_ADD;
break;
}
if (str4cmp(m, 'c', 'a', 's', ' ')) {
r->type = MSG_REQ_CAS;
break;
}
break;
case 4:
if (str4cmp(m, 'g', 'e', 't', 's')) {
r->type = MSG_REQ_GETS;
break;
}
if (str4cmp(m, 'i', 'n', 'c', 'r')) {
r->type = MSG_REQ_INCR;
break;
}
if (str4cmp(m, 'd', 'e', 'c', 'r')) {
r->type = MSG_REQ_DECR;
break;
}
if (str4cmp(m, 'q', 'u', 'i', 't')) {
r->type = MSG_REQ_QUIT;
r->quit = 1;
break;
}
break;
case 6:
if (str6cmp(m, 'a', 'p', 'p', 'e', 'n', 'd')) {
r->type = MSG_REQ_APPEND;
break;
}
if (str6cmp(m, 'd', 'e', 'l', 'e', 't', 'e')) {
r->type = MSG_REQ_DELETE;
break;
}
break;
case 7:
if (str7cmp(m, 'p', 'r', 'e', 'p', 'e', 'n', 'd')) {
r->type = MSG_REQ_PREPEND;
break;
}
if (str7cmp(m, 'r', 'e', 'p', 'l', 'a', 'c', 'e')) {
r->type = MSG_REQ_REPLACE;
break;
}
if (str7cmp(m, 'v', 'e', 'r', 's', 'i', 'o', 'n')) {
r->type = MSG_REQ_VERSION;
break;
}
break;
}
if (memcache_key(r)) {
if (ch == CR) {
goto error;
}
state = SW_SPACES_BEFORE_KEY;
} else if (memcache_quit(r) || memcache_version(r)) {
p = p - 1; /* go back by 1 byte */
state = SW_CRLF;
} else {
goto error;
}
} else if (!islower(ch)) {
goto error;
}
break;
case SW_SPACES_BEFORE_KEY:
if (ch != ' ') {
p = p - 1; /* go back by 1 byte */
state = SW_KEY;
}
break;
case SW_KEY:
if (r->token == NULL) {
r->token = p;
r->key_start = p;
}
if (ch == ' ' || ch == CR) {
if ((p - r->key_start) > MEMCACHE_MAX_KEY_LENGTH) {
log_error("parsed bad req %"PRIu64" of type %d with key "
"prefix '%.*s...' and length %d that exceeds "
"maximum key length", r->id, r->type, 16,
r->key_start, p - r->key_start);
goto error;
}
r->key_end = p;
r->token = NULL;
/* get next state */
if (memcache_storage(r)) {
state = SW_SPACES_BEFORE_FLAGS;
} else if (memcache_arithmetic(r)) {
state = SW_SPACES_BEFORE_NUM;
} else if (memcache_delete(r)) {
state = SW_RUNTO_CRLF;
} else if (memcache_retrieval(r)) {
state = SW_SPACES_BEFORE_KEYS;
} else {
state = SW_RUNTO_CRLF;
}
if (ch == CR) {
if (memcache_storage(r) || memcache_arithmetic(r)) {
goto error;
}
p = p - 1; /* go back by 1 byte */
}
}
break;
case SW_SPACES_BEFORE_KEYS:
ASSERT(memcache_retrieval(r));
switch (ch) {
case ' ':
break;
case CR:
state = SW_ALMOST_DONE;
break;
default:
r->token = p;
goto fragment;
}
break;
case SW_SPACES_BEFORE_FLAGS:
if (ch != ' ') {
if (!isdigit(ch)) {
goto error;
}
p = p - 1; /* go back by 1 byte */
state = SW_FLAGS;
}
break;
case SW_FLAGS:
if (r->token == NULL) {
/* flags_start <- p */
r->token = p;
r->flags = 0;
}
if (isdigit(ch)) {
r->flags = r->flags * 10 + (uint32_t)(ch - '0');
} else if (ch == ' ') {
/* flags_end <- p - 1 */
r->token = NULL;
state = SW_SPACES_BEFORE_EXPIRY;
} else {
goto error;
}
break;
case SW_SPACES_BEFORE_EXPIRY:
if (ch != ' ') {
if (!isdigit(ch)) {
goto error;
}
p = p - 1; /* go back by 1 byte */
state = SW_EXPIRY;
}
break;
case SW_EXPIRY:
if (r->token == NULL) {
/* expiry_start <- p */
r->token = p;
r->expiry = 0;
}
if (isdigit(ch)) {
r->expiry = r->expiry * 10 + (uint32_t)(ch - '0');
} else if (ch == ' ') {
/* expiry_end <- p - 1 */
r->token = NULL;
state = SW_SPACES_BEFORE_VLEN;
} else {
goto error;
}
break;
case SW_SPACES_BEFORE_VLEN:
if (ch != ' ') {
if (!isdigit(ch)) {
goto error;
}
p = p - 1; /* go back by 1 byte */
state = SW_VLEN;
}
break;
case SW_VLEN:
if (r->token == NULL) {
/* vlen_start <- p */
r->token = p;
r->vlen = 0;
}
if (isdigit(ch)) {
r->vlen = r->vlen * 10 + (uint32_t)(ch - '0');
} else if (memcache_cas(r)) {
if (ch != ' ') {
goto error;
}
/* vlen_end <- p - 1 */
r->rvlen = r->vlen;
p = p - 1; /* go back by 1 byte */
r->token = NULL;
state = SW_SPACES_BEFORE_CAS;
} else if (ch == ' ' || ch == CR) {
/* vlen_end <- p - 1 */
r->rvlen = r->vlen;
p = p - 1; /* go back by 1 byte */
r->token = NULL;
state = SW_RUNTO_CRLF;
} else {
goto error;
}
break;
case SW_SPACES_BEFORE_CAS:
if (ch != ' ') {
if (!isdigit(ch)) {
goto error;
}
p = p - 1; /* go back by 1 byte */
state = SW_CAS;
}
break;
case SW_CAS:
if (r->token == NULL) {
/* cas_start <- p */
r->token = p;
r->cas = 0;
}
if (isdigit(ch)) {
r->cas = r->cas * 10ULL + (uint64_t)(ch - '0');
} else if (ch == ' ' || ch == CR) {
/* cas_end <- p - 1 */
p = p - 1; /* go back by 1 byte */
r->token = NULL;
state = SW_RUNTO_CRLF;
} else {
goto error;
}
break;
case SW_RUNTO_VAL:
switch (ch) {
case LF:
/* val_start <- p + 1 */
state = SW_VAL;
break;
default:
goto error;
}
break;
case SW_VAL:
if (r->value == NULL) {
r->value = p;
}
m = p + r->rvlen;
if (m >= b->last) {
ASSERT(r->rvlen >= (uint32_t)(b->last - p));
r->rvlen -= (uint32_t)(b->last - p);
m = b->last - 1;
p = m; /* move forward by vlen bytes */
break;
}
switch (*m) {
case CR:
/* val_end <- p - 1 */
p = m; /* move forward by vlen bytes */
state = SW_ALMOST_DONE;
break;
default:
goto error;
}
break;
case SW_SPACES_BEFORE_NUM:
if (ch != ' ') {
if (!isdigit(ch)) {
goto error;
}
p = p - 1; /* go back by 1 byte */
state = SW_NUM;
}
break;
case SW_NUM:
if (r->token == NULL) {
/* num_start <- p */
r->token = p;
r->num = 0;
}
if (isdigit(ch)) {
r->num = r->num * 10ULL + (uint64_t)(ch - '0');
} else if (ch == ' ' || ch == CR) {
r->token = NULL;
/* num_end <- p - 1 */
p = p - 1; /* go back by 1 byte */
state = SW_RUNTO_CRLF;
} else {
goto error;
}
break;
case SW_RUNTO_CRLF:
switch (ch) {
case ' ':
break;
case 'n':
if (memcache_storage(r) || memcache_arithmetic(r) || memcache_delete(r)) {
p = p - 1; /* go back by 1 byte */
state = SW_NOREPLY;
} else {
goto error;
}
break;
case CR:
if (memcache_storage(r)) {
state = SW_RUNTO_VAL;
} else {
state = SW_ALMOST_DONE;
}
break;
default:
goto error;
}
break;
case SW_NOREPLY:
if (r->token == NULL) {
/* noreply_start <- p */
r->token = p;
}
switch (ch) {
case ' ':
case CR:
m = r->token;
if (((p - m) == 7) && str7cmp(m, 'n', 'o', 'r', 'e', 'p', 'l', 'y')) {
ASSERT(memcache_storage(r) || memcache_arithmetic(r) || memcache_delete(r));
r->token = NULL;
/* noreply_end <- p - 1 */
r->noreply = 1;
state = SW_AFTER_NOREPLY;
p = p - 1; /* go back by 1 byte */
} else {
goto error;
}
}
break;
case SW_AFTER_NOREPLY:
switch (ch) {
case ' ':
break;
case CR:
if (memcache_storage(r)) {
state = SW_RUNTO_VAL;
} else {
state = SW_ALMOST_DONE;
}
break;
default:
goto error;
}
break;
case SW_CRLF:
switch (ch) {
case ' ':
break;
case CR:
state = SW_ALMOST_DONE;
break;
default:
goto error;
}
break;
case SW_ALMOST_DONE:
switch (ch) {
case LF:
/* req_end <- p */
goto done;
default:
goto error;
}
break;
case SW_SENTINEL:
default:
NOT_REACHED();
break;
}
}
/*
* At this point, buffer from b->pos to b->last has been parsed completely
* but we haven't been able to reach to any conclusion. Normally, this
* means that we have to parse again starting from the state we are in
* after more data has been read. The newly read data is either read into
* a new mbuf, if existing mbuf is full (b->last == b->end) or into the
* existing mbuf.
*
* The only exception to this is when the existing mbuf is full (b->last
* is at b->end) and token marker is set, which means that we have to
* copy the partial token into a new mbuf and parse again with more data
* read into new mbuf.
*/
ASSERT(p == b->last);
r->pos = p;
r->state = state;
if (b->last == b->end && r->token != NULL) {
r->pos = r->token;
r->token = NULL;
r->result = MSG_PARSE_REPAIR;
} else {
r->result = MSG_PARSE_AGAIN;
}
log_hexdump(LOG_VERB, b->pos, mbuf_length(b), "parsed req %"PRIu64" res %d "
"type %d state %d rpos %d of %d", r->id, r->result, r->type,
r->state, r->pos - b->pos, b->last - b->pos);
return;
fragment:
ASSERT(p != b->last);
ASSERT(r->token != NULL);
r->pos = r->token;
r->token = NULL;
r->state = state;
r->result = MSG_PARSE_FRAGMENT;
log_hexdump(LOG_VERB, b->pos, mbuf_length(b), "parsed req %"PRIu64" res %d "
"type %d state %d rpos %d of %d", r->id, r->result, r->type,
r->state, r->pos - b->pos, b->last - b->pos);
return;
done:
ASSERT(r->type > MSG_UNKNOWN && r->type < MSG_SENTINEL);
r->pos = p + 1;
ASSERT(r->pos <= b->last);
r->state = SW_START;
r->result = MSG_PARSE_OK;
log_hexdump(LOG_VERB, b->pos, mbuf_length(b), "parsed req %"PRIu64" res %d "
"type %d state %d rpos %d of %d", r->id, r->result, r->type,
r->state, r->pos - b->pos, b->last - b->pos);
return;
error:
r->result = MSG_PARSE_ERROR;
r->state = state;
errno = EINVAL;
log_hexdump(LOG_INFO, b->pos, mbuf_length(b), "parsed bad req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
r->state);
}
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;
}
/*
* 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;
}
}
void
dyn_parse_req(struct msg *r)
{
#ifdef DN_DEBUG_LOG
log_debug(LOG_VVERB, "In dyn_parse_req, start to process request :::::::::::::::::::::: ");
msg_dump(r);
#endif
bool done_parsing = false;
if (dyn_parse_core(r)) {
struct dmsg *dmsg = r->dmsg;
if (dmsg->type != DMSG_UNKNOWN && dmsg->type != DMSG_REQ && dmsg->type != GOSSIP_SYN) {
log_debug(LOG_DEBUG, "Req 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 (dmsg->type == GOSSIP_SYN) {
#ifdef DN_DEBUG_LOG
log_debug(LOG_DEBUG, "Req parser: I got a GOSSIP_SYN msg");
#endif
//TODOs: need to address multi-buffer msg later
struct mbuf *b = STAILQ_LAST(&r->mhdr, mbuf, next);
dmsg->payload = b->pos;
b->pos = b->pos + dmsg->plen;
r->pos = b->pos;
done_parsing = true;
}
//check whether we need to decrypt the payload
if (dmsg->bit_field == 1) {
dmsg->owner->owner->dnode_secured = 1;
r->owner->dnode_crypto_state = 1;
struct mbuf *decrypted_buf = mbuf_get();
if (decrypted_buf == NULL) {
loga("Unable to obtain an mbuf for dnode msg's header!");
return;
}
#ifdef DN_DEBUG_LOG
log_debug(LOG_DEBUG, "data or encrypted aes key length : %d", dmsg->plen);
#endif
if (dmsg->mlen > 1) {
#ifdef DN_DEBUG_LOG
log_debug(LOG_DEBUG, "dmsg->mlen is something: %d, need to process it", dmsg->plen);
#endif
//Decrypt AES key
dyn_rsa_decrypt(dmsg->data, aes_decrypted_buf);
strncpy(r->owner->aes_key, aes_decrypted_buf, strlen(aes_decrypted_buf));
//Decrypt payload
dyn_aes_decrypt(dmsg->payload, dmsg->plen, decrypted_buf, aes_decrypted_buf);
} else {
#ifdef DN_DEBUG_LOG
log_debug(LOG_DEBUG, "dmsg->mlen is a dummy: %d, NO need to process it", dmsg->plen);
#endif
dyn_aes_decrypt(dmsg->payload, dmsg->plen, decrypted_buf, r->owner->aes_key);
}
#ifdef DN_DEBUG_LOG
loga("AES encryption key: %s\n", base64_encode(aes_decrypted_buf, AES_KEYLEN));
log_hexdump(LOG_VERB, decrypted_buf->pos, mbuf_length(decrypted_buf), "dyn message decrypted payload: ");
#endif
struct mbuf *b = STAILQ_LAST(&r->mhdr, mbuf, next);
b->last = b->pos;
r->pos = decrypted_buf->start;
mbuf_insert(&r->mhdr, decrypted_buf);
//reset these variables
dmsg->payload = decrypted_buf->start;
dmsg->plen = mbuf_length(decrypted_buf);
}
if (done_parsing)
return;
if (r->redis)
return redis_parse_req(r);
return memcache_parse_req(r);
}
//bad case
log_debug(LOG_DEBUG, "Bad message - cannot parse"); //fix me to do something
msg_dump(r);
}
/*
* Load an ELF section table and create a list of Elf_Scn structures.
*/
static int
_libelf_load_scn(Elf *e, void *ehdr)
{
int ec, swapbytes;
size_t fsz, i, shnum;
uint64_t shoff;
char *src;
Elf32_Ehdr *eh32;
Elf64_Ehdr *eh64;
Elf_Scn *scn;
void (*xlator)(char *_d, char *_s, size_t _c, int _swap);
assert(e != NULL);
assert(ehdr != NULL);
assert((e->e_flags & LIBELF_F_SHDRS_LOADED) == 0);
#define CHECK_EHDR(E,EH) do { \
if (fsz != (EH)->e_shentsize || \
shoff + fsz * shnum > e->e_rawsize) { \
LIBELF_SET_ERROR(HEADER, 0); \
return (0); \
} \
} while (0)
ec = e->e_class;
fsz = _libelf_fsize(ELF_T_SHDR, ec, e->e_version, (size_t) 1);
assert(fsz > 0);
shnum = e->e_u.e_elf.e_nscn;
if (ec == ELFCLASS32) {
eh32 = (Elf32_Ehdr *) ehdr;
shoff = (uint64_t) eh32->e_shoff;
CHECK_EHDR(e, eh32);
} else {
eh64 = (Elf64_Ehdr *) ehdr;
shoff = eh64->e_shoff;
CHECK_EHDR(e, eh64);
}
xlator = _libelf_get_translator(ELF_T_SHDR, ELF_TOMEMORY, ec);
swapbytes = e->e_byteorder != LIBELF_PRIVATE(byteorder);
src = e->e_rawfile + shoff;
/*
* If the file is using extended numbering then section #0
* would have already been read in.
*/
i = 0;
if (!STAILQ_EMPTY(&e->e_u.e_elf.e_scn)) {
assert(STAILQ_FIRST(&e->e_u.e_elf.e_scn) ==
STAILQ_LAST(&e->e_u.e_elf.e_scn, _Elf_Scn, s_next));
i = 1;
src += fsz;
}
for (; i < shnum; i++, src += fsz) {
if ((scn = _libelf_allocate_scn(e, i)) == NULL)
return (0);
(*xlator)((char *) &scn->s_shdr, src, (size_t) 1, swapbytes);
if (ec == ELFCLASS32) {
scn->s_offset = scn->s_rawoff =
scn->s_shdr.s_shdr32.sh_offset;
scn->s_size = scn->s_shdr.s_shdr32.sh_size;
} else {
scn->s_offset = scn->s_rawoff =
scn->s_shdr.s_shdr64.sh_offset;
scn->s_size = scn->s_shdr.s_shdr64.sh_size;
}
}
e->e_flags |= LIBELF_F_SHDRS_LOADED;
return (1);
}
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);
}
/* 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;
uint8_t ch;
uint64_t num = 0;
state = r->dyn_state;
b = STAILQ_LAST(&r->mhdr, mbuf, next);
dmsg = r->dmsg;
if (dmsg == NULL) {
r->dmsg = dmsg_get();
dmsg = r->dmsg;
if (dmsg == NULL) {//should track this as a dropped message
goto error; //should count as OOM error
}
}
//log_hexdump(LOG_VERB, b->pos, mbuf_length(b), "dyn parser: parsed req %"PRIu64" res %d type %d", r->id, r->result, r->type, r->dyn_state);
for (p = r->pos; p < b->last; p++) {
ch = *p;
switch (state) {
case DYN_START:
//log_debug(LOG_DEBUG, "DYN_START");
if (ch == ' ') {
break;
} else if (isdigit(ch)) {
num = ch - '0';
state = DYN_MAGIC_NUMBER;
} else {
goto skip;
}
break;
case DYN_MAGIC_NUMBER:
//log_debug(LOG_DEBUG, "DYN_MAGIC_NUMBER");
//log_debug(LOG_DEBUG, "num = %d", num);
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else {
if (num == MAGIC_NUMBER) {
state = DYN_SPACES_BEFORE_MSG_ID;
} else {
goto error;
}
}
break;
case DYN_SPACES_BEFORE_MSG_ID:
//log_debug(LOG_DEBUG, "DYN_SPACES_BEFORE_MSG_ID");
if (ch == ' ') {
break;
} else if (isdigit(ch)) {
num = ch - '0';
state = DYN_MSG_ID;
} else {
goto error;
}
break;
case DYN_MSG_ID:
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 != ' ') {
goto error;
} else {
//if (num >= 0) {
//log_debug(LOG_DEBUG, "MSG ID : %d", num);
dmsg->id = num;
state = DYN_SPACES_BEFORE_TYPE_ID;
//} else {
// goto error;
//}
}
break;
case DYN_SPACES_BEFORE_TYPE_ID:
log_debug(LOG_DEBUG, "DYN_SPACES_BEFORE_TYPE_ID");
if (ch == ' ') {
break;
} else if (isdigit(ch)) {
num = ch - '0';
state = DYN_TYPE_ID;
} else {
goto error;
}
break;
case DYN_TYPE_ID:
log_debug(LOG_DEBUG, "DYN_TYPE_ID");
log_debug(LOG_DEBUG, "num = %d", num);
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else {
if (num > 0) {
log_debug(LOG_DEBUG, "Type Id: %d", num);
dmsg->type = num;
//state = DYN_SPACES_BEFORE_VERSION;
state = DYN_SPACES_BEFORE_BIT_FIELD;
} else {
goto error;
}
}
break;
case DYN_SPACES_BEFORE_BIT_FIELD:
if (ch == ' ') {
break;
} else if (isdigit(ch)) {
num = ch - '0';
state = DYN_BIT_FIELD;
} else {
goto error;
}
break;
case DYN_BIT_FIELD:
log_debug(LOG_DEBUG, "DYN_BIT_FIELD");
log_debug(LOG_DEBUG, "num = %d", num);
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else {
if (ch == ' ') {
log_debug(LOG_DEBUG, "DYN_BIT_FIELD : %d", num);
dmsg->bit_field = num & 0xF;
state = DYN_SPACES_BEFORE_VERSION;
} else {
goto error;
}
}
log_debug(LOG_DEBUG, "Post DYN_BIT_FIELD");
log_debug(LOG_DEBUG, "num = %d", num);
break;
case DYN_SPACES_BEFORE_VERSION:
log_debug(LOG_DEBUG, "DYN_SPACES_BEFORE_VERSION");
if (ch == ' ') {
break;
} else if (isdigit(ch)) {
num = ch - '0';
state = DYN_VERSION;
} else {
goto error;
}
break;
case DYN_VERSION:
log_debug(LOG_DEBUG, "DYN_VERSION");
log_debug(LOG_DEBUG, "num = %d", num);
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else {
if (ch == ' ') {
//log_debug(LOG_DEBUG, "VERSION : %d", num);
dmsg->version = num;
state = DYN_SPACES_BEFORE_STAR;
} else {
goto error;
}
}
break;
case DYN_SPACES_BEFORE_STAR:
//log_debug(LOG_DEBUG, "DYN_CRLF_BEFORE_STAR");
if (ch == ' ') {
break;
} else if (ch == '*') {
state = DYN_DATA_LEN;
num = 0;
} else {
goto error;
}
//else {
// state = DYN_STAR;
//}
break;
//case DYN_STAR:
//log_debug(LOG_DEBUG, "DYN_STAR");
// if (ch == '*') {
// state = DYN_DATA_LEN;
// num = 0;
// } else {
// goto error;
// }
// break;
case DYN_DATA_LEN:
log_debug(LOG_DEBUG, "DYN_DATA_LEN");
log_debug(LOG_DEBUG, "num = %d", num);
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else {
if (ch == ' ') {
log_debug(LOG_DEBUG, "Data len: %d", num);
dmsg->mlen = num;
state = DYN_SPACE_BEFORE_DATA;
num = 0;
} else {
goto error;
}
}
break;
case DYN_SPACE_BEFORE_DATA:
log_debug(LOG_DEBUG, "DYN_SPACE_BEFORE_DATA");
state = DYN_DATA;
break;
case DYN_DATA:
log_debug(LOG_DEBUG, "DYN_DATA");
p -= 1;
if (dmsg->mlen > 0) {
dmsg->data = p;
p += dmsg->mlen - 1;
state = DYN_SPACES_BEFORE_PAYLOAD_LEN;
} else {
goto error;
}
break;
case DYN_SPACES_BEFORE_PAYLOAD_LEN: //this only need in dynomite's custome msg
log_debug(LOG_DEBUG, "DYN_SPACES_BEFORE_PAYLOAD_LEN");
if (ch == ' ') {
break;
} else if (ch == '*') {
state = DYN_PAYLOAD_LEN;
num = 0;
} else {
goto error;
}
break;
case DYN_PAYLOAD_LEN:
if (isdigit(ch)) {
num = num*10 + (ch - '0');
} else {
if (ch == CR) {
log_debug(LOG_DEBUG, "Payload len: %d", num);
dmsg->plen = num;
state = DYN_CRLF_BEFORE_DONE;
num = 0;
} else {
goto error;
}
}
break;
case DYN_CRLF_BEFORE_DONE:
//log_debug(LOG_DEBUG, "DYN_CRLF_BEFORE_DONE");
if (*p == LF) {
state = DYN_DONE;
} else {
goto error;
}
break;
case DYN_DONE:
//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;
}
}
done:
dmsg->owner = r;
dmsg->source_address = r->owner->addr;
//r->mlen = mbuf_length(b);
//log_debug(LOG_DEBUG, "at done with p at %d", p);
dmsg_dump(r->dmsg);
log_hexdump(LOG_VERB, b->pos, mbuf_length(b), "dyn: parsed req %"PRIu64" res %d "
"type %d state %d rpos %d of %d", r->id, r->result, r->type,
r->dyn_state, r->pos - b->pos, b->last - b->pos);
return true;
skip:
//log_debug(LOG_DEBUG, "This is not a dyn message");
dmsg->type = DMSG_UNKNOWN;
dmsg->owner = r;
dmsg->source_address = r->owner->addr;
return true;
error:
log_debug(LOG_ERR, "at error");
r->result = MSG_PARSE_ERROR;
r->state = state;
errno = EINVAL;
log_hexdump(LOG_INFO, b->pos, mbuf_length(b), "parsed bad req %"PRIu64" "
"res %d type %d state %d", r->id, r->result, r->type,
r->state);
return false;
return true; //fix me
}
void dyn_parse_rsp(struct msg *r)
{
#ifdef DN_DEBUG_LOG
log_debug(LOG_VERB, "In dyn_parse_rsp, start to process response :::::::::::::::::::::::: ");
msg_dump(r);
#endif
if (dyn_parse_core(r)) {
struct dmsg *dmsg = r->dmsg;
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!");
return;
}
#ifdef DN_DEBUG_LOG
log_debug(LOG_VERB, "encrypted aes key length : %d", dmsg->mlen);
loga("AES encryption key from conn: %s\n", base64_encode(r->owner->aes_key, AES_KEYLEN));
#endif
//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);
#ifdef DN_DEBUG_LOG
log_hexdump(LOG_VERB, decrypted_buf->pos, mbuf_length(decrypted_buf), "dyn message decrypted payload: ");
#endif
struct mbuf *b = STAILQ_LAST(&r->mhdr, mbuf, next);
b->last = b->pos;
r->pos = decrypted_buf->start;
mbuf_insert(&r->mhdr, decrypted_buf);
}
if (r->redis)
return redis_parse_rsp(r);
return memcache_parse_rsp(r);
}
#ifdef DN_DEBUG_LOG
//bad case
log_debug(LOG_DEBUG, "Bad message - cannot parse"); //fix me to do something
msg_dump(r);
#endif
//r->state = 0;
//r->result = MSG_PARSE_OK;
}
