bool b2b_forward_item_vbucket(conn *uc, downstream *d, item *it,
conn *c, bool self, int vbucket) {
(void)self;
assert(d != NULL);
assert(d->ptd != NULL);
assert(uc != NULL);
assert(uc->next == NULL);
assert(uc->noreply == false);
assert(c != NULL);
// Assuming we're already connected to downstream.
//
// TODO: Optimize to self codepath.
//
if (settings.verbose > 2) {
moxi_log_write("%d: b2b_forward_item_vbucket %x to %d, vbucket %d\n",
uc->sfd, uc->cmd, c->sfd, vbucket);
}
protocol_binary_request_header *req =
(protocol_binary_request_header *) ITEM_data(it);
if (vbucket >= 0) {
req->request.reserved = htons(vbucket);
}
if (add_conn_item(c, it) == true) {
// The caller keeps its refcount, and we need our own.
//
it->refcount++;
if (add_iov(c, ITEM_data(it), it->nbytes) == 0) {
conn_set_state(c, conn_mwrite);
c->write_and_go = conn_new_cmd;
if (update_event(c, EV_WRITE | EV_PERSIST)) {
if (settings.verbose > 2) {
moxi_log_write("%d: b2b_forward %x to %d success\n",
uc->sfd, uc->cmd, c->sfd);
}
return true;
}
}
}
d->ptd->stats.stats.err_oom++;
cproxy_close_conn(c);
return false;
}
void cproxy_process_a2a_downstream(conn *c, char *line) {
assert(c != NULL);
assert(c->next == NULL);
assert(c->extra != NULL);
assert(c->cmd == -1);
assert(c->item == NULL);
assert(line != NULL);
assert(line == c->rcurr);
assert(IS_ASCII(c->protocol));
assert(IS_PROXY(c->protocol));
if (settings.verbose > 1)
fprintf(stderr, "<%d cproxy_process_a2a_downstream %s\n",
c->sfd, line);
downstream *d = c->extra;
assert(d != NULL);
assert(d->ptd != NULL);
assert(d->ptd->proxy != NULL);
if (strncmp(line, "VALUE ", 6) == 0) {
token_t tokens[MAX_TOKENS];
size_t ntokens;
unsigned int flags;
int clen = 0;
int vlen;
uint64_t cas = CPROXY_NOT_CAS;
ntokens = scan_tokens(line, tokens, MAX_TOKENS, &clen);
if (ntokens >= 5 && // Accounts for extra termimation token.
ntokens <= 6 &&
tokens[KEY_TOKEN].length <= KEY_MAX_LENGTH &&
safe_strtoul(tokens[2].value, (uint32_t *) &flags) &&
safe_strtoul(tokens[3].value, (uint32_t *) &vlen)) {
char *key = tokens[KEY_TOKEN].value;
size_t nkey = tokens[KEY_TOKEN].length;
item *it = item_alloc(key, nkey, flags, 0, vlen + 2);
if (it != NULL) {
if (ntokens == 5 ||
safe_strtoull(tokens[4].value, &cas)) {
ITEM_set_cas(it, cas);
c->item = it;
c->ritem = ITEM_data(it);
c->rlbytes = it->nbytes;
c->cmd = -1;
conn_set_state(c, conn_nread);
return; // Success.
} else {
if (settings.verbose > 1)
fprintf(stderr, "cproxy could not parse cas\n");
}
} else {
if (settings.verbose > 1)
fprintf(stderr, "cproxy could not item_alloc size %u\n",
vlen + 2);
}
if (it != NULL)
item_remove(it);
it = NULL;
c->sbytes = vlen + 2; // Number of bytes to swallow.
conn_set_state(c, conn_swallow);
// Note, eventually, we'll see an END later.
} else {
// We don't know how much to swallow, so close the downstream.
// The conn_closing should release the downstream,
// which should write a suffix/error to the upstream.
//
conn_set_state(c, conn_closing);
}
} else if (strncmp(line, "END", 3) == 0) {
conn_set_state(c, conn_pause);
} else if (strncmp(line, "OK", 2) == 0) {
conn_set_state(c, conn_pause);
// TODO: Handle flush_all's expiration parameter against
// the front_cache.
//
// TODO: We flush the front_cache too often, inefficiently
// on every downstream flush_all OK response, rather than
// on just the last flush_all OK response.
//
conn *uc = d->upstream_conn;
if (uc != NULL &&
uc->cmd_curr == PROTOCOL_BINARY_CMD_FLUSH) {
mcache_flush_all(&d->ptd->proxy->front_cache, 0);
}
} else if (strncmp(line, "STAT ", 5) == 0 ||
strncmp(line, "ITEM ", 5) == 0 ||
strncmp(line, "PREFIX ", 7) == 0) {
assert(d->merger != NULL);
conn *uc = d->upstream_conn;
if (uc != NULL) {
assert(uc->next == NULL);
if (protocol_stats_merge_line(d->merger, line) == false) {
// Forward the line as-is if we couldn't merge it.
//
int nline = strlen(line);
item *it = item_alloc("s", 1, 0, 0, nline + 2);
if (it != NULL) {
strncpy(ITEM_data(it), line, nline);
strncpy(ITEM_data(it) + nline, "\r\n", 2);
if (add_conn_item(uc, it)) {
add_iov(uc, ITEM_data(it), nline + 2);
it = NULL;
}
if (it != NULL)
item_remove(it);
}
}
}
conn_set_state(c, conn_new_cmd);
} else {
conn_set_state(c, conn_pause);
// The upstream conn might be NULL when closed already
// or while handling a noreply.
//
conn *uc = d->upstream_conn;
if (uc != NULL) {
assert(uc->next == NULL);
out_string(uc, line);
if (!update_event(uc, EV_WRITE | EV_PERSIST)) {
if (settings.verbose > 1)
fprintf(stderr,
"Can't update upstream write event\n");
d->ptd->stats.stats.err_oom++;
cproxy_close_conn(uc);
}
cproxy_del_front_cache_key_ascii_response(d, line,
uc->cmd_start);
}
}
}
/* We reach here after nread'ing a header+body into an item.
*/
void cproxy_process_b2b_downstream_nread(conn *c) {
conn *uc;
item *it;
downstream *d;
protocol_binary_response_header *header;
int extlen;
int keylen;
uint32_t bodylen;
int status;
int opcode;
cb_assert(c != NULL);
cb_assert(c->cmd >= 0);
cb_assert(c->next == NULL);
cb_assert(c->cmd_start == NULL);
cb_assert(IS_BINARY(c->protocol));
cb_assert(IS_PROXY(c->protocol));
header = (protocol_binary_response_header *) &c->binary_header;
extlen = header->response.extlen;
keylen = header->response.keylen;
bodylen = header->response.bodylen;
status = ntohs(header->response.status);
opcode = header->response.opcode;
if (settings.verbose > 2) {
moxi_log_write("<%d cproxy_process_b2b_downstream_nread %x %x %d %d %u %d %x\n",
c->sfd, c->cmd, opcode, extlen, keylen, bodylen, c->noreply, status);
}
d = c->extra;
cb_assert(d != NULL);
cb_assert(d->ptd != NULL);
cb_assert(d->ptd->proxy != NULL);
/* TODO: Need to handle quiet binary command error response, */
/* in the right order. */
/* TODO: Need to handle not-my-vbucket error during a quiet cmd. */
uc = d->upstream_conn;
it = c->item;
/* Clear c->item because we either move it to the upstream or */
/* item_remove() it on error. */
c->item = NULL;
cb_assert(it != NULL);
cb_assert(it->refcount == 1);
if (cproxy_binary_ignore_reply(c, header, it)) {
return;
}
if (c->noreply) {
conn_set_state(c, conn_new_cmd);
} else {
conn_set_state(c, conn_pause);
if (opcode == PROTOCOL_BINARY_CMD_NOOP ||
opcode == PROTOCOL_BINARY_CMD_FLUSH) {
goto done;
}
if (opcode == PROTOCOL_BINARY_CMD_STAT) {
if (status == PROTOCOL_BINARY_RESPONSE_SUCCESS) {
if (keylen > 0) {
if (d->merger != NULL) {
char *key = (ITEM_data(it)) + sizeof(*header) + extlen;
char *val = key + keylen;
protocol_stats_merge_name_val(d->merger, "STAT", 4,
key, keylen,
val, bodylen - keylen - extlen);
}
conn_set_state(c, conn_new_cmd); /* Get next STATS response. */
}
}
goto done;
}
/* If the client is still there, we should handle */
/* a not-my-vbucket error with a possible retry. */
if (uc != NULL &&
status == PROTOCOL_BINARY_RESPONSE_NOT_MY_VBUCKET) {
int max_retries;
protocol_binary_request_header *req;
int vbucket;
int sindex;
if (settings.verbose > 2) {
moxi_log_write("<%d cproxy_process_b2b_downstream_nread not-my-vbucket, "
"cmd: %x %d\n",
c->sfd, header->response.opcode, uc->item != NULL);
}
cb_assert(uc->item != NULL);
req = (protocol_binary_request_header *)ITEM_data((item*)uc->item);
vbucket = ntohs(req->request.reserved);
sindex = downstream_conn_index(d, c);
if (settings.verbose > 2) {
moxi_log_write("<%d cproxy_process_b2b_downstream_nread not-my-vbucket, "
"cmd: %x not multi-key get, sindex %d, vbucket %d, retries %d\n",
c->sfd, header->response.opcode,
sindex, vbucket, uc->cmd_retries);
}
mcs_server_invalid_vbucket(&d->mst, sindex, vbucket);
/* As long as the upstream is still open and we haven't */
/* retried too many times already. */
max_retries = cproxy_max_retries(d);
if (uc->cmd_retries < max_retries) {
uc->cmd_retries++;
d->upstream_retry++;
d->ptd->stats.stats.tot_retry_vbucket++;
goto done;
}
if (settings.verbose > 2) {
moxi_log_write("%d: cproxy_process_b2b_downstream_nread not-my-vbucket, "
"cmd: %x skipping retry %d >= %d\n",
c->sfd, header->response.opcode, uc->cmd_retries,
max_retries);
}
}
}
/* Write the response to the upstream connection. */
if (uc != NULL) {
if (settings.verbose > 2) {
moxi_log_write("<%d cproxy_process_b2b_downstream_nread got %u\n",
c->sfd, it->nbytes);
cproxy_dump_header(c->sfd, ITEM_data(it));
}
if (add_conn_item(uc, it) == true) {
it->refcount++;
if (add_iov(uc, ITEM_data(it), it->nbytes) == 0) {
/* If we got a quiet response, however, don't change the */
/* upstream connection's state (should be in paused state), */
/* as we expect the downstream server to provide a */
/* verbal/non-quiet response that moves the downstream */
/* conn through the conn_pause countdown codepath. */
if (c->noreply == false) {
cproxy_update_event_write(d, uc);
conn_set_state(uc, conn_mwrite);
}
goto done;
}
}
d->ptd->stats.stats.err_oom++;
cproxy_close_conn(uc);
}
done:
if (it != NULL) {
item_remove(it);
}
}
/* Used for broadcast commands, like no-op, flush_all or stats.
*/
bool cproxy_broadcast_b2b_downstream(downstream *d, conn *uc) {
int nwrite = 0;
int nconns;
int i;
cb_assert(d != NULL);
cb_assert(d->ptd != NULL);
cb_assert(d->ptd->proxy != NULL);
cb_assert(d->downstream_conns != NULL);
cb_assert(uc != NULL);
cb_assert(uc->next == NULL);
cb_assert(uc->noreply == false);
nconns = mcs_server_count(&d->mst);
for (i = 0; i < nconns; i++) {
conn *c = d->downstream_conns[i];
if (c != NULL &&
c != NULL_CONN &&
b2b_forward_item_vbucket(uc, d, uc->item, c, -1) == true) {
nwrite++;
}
}
if (settings.verbose > 2) {
moxi_log_write("%d: b2b broadcast nwrite %d out of %d\n",
uc->sfd, nwrite, nconns);
}
if (nwrite > 0) {
/* TODO: Handle binary 'stats reset' sub-command. */
item *it;
if (uc->cmd == PROTOCOL_BINARY_CMD_STAT &&
d->merger == NULL) {
d->merger = genhash_init(128, skeyhash_ops);
}
it = item_alloc("h", 1, 0, 0,
sizeof(protocol_binary_response_header));
if (it != NULL) {
protocol_binary_response_header *header =
(protocol_binary_response_header *) ITEM_data(it);
memset(ITEM_data(it), 0, it->nbytes);
header->response.magic = (uint8_t) PROTOCOL_BINARY_RES;
header->response.opcode = uc->binary_header.request.opcode;
header->response.opaque = uc->opaque;
if (add_conn_item(uc, it)) {
d->upstream_suffix = ITEM_data(it);
d->upstream_suffix_len = it->nbytes;
d->upstream_status = PROTOCOL_BINARY_RESPONSE_SUCCESS;
d->target_host_ident = NULL;
if (settings.verbose > 2) {
moxi_log_write("%d: b2b broadcast upstream_suffix", uc->sfd);
cproxy_dump_header(uc->sfd, ITEM_data(it));
}
/* TODO: Handle FLUSHQ (quiet binary flush_all). */
d->downstream_used_start = nwrite;
d->downstream_used = nwrite;
cproxy_start_downstream_timeout(d, NULL);
return true;
}
item_remove(it);
}
}
return false;
}
/**
* @param cas_emit 1: emit CAS.
* 0: do not emit CAS.
* -1: data driven.
*/
void cproxy_upstream_ascii_item_response(item *it, conn *uc,
int cas_emit) {
assert(it != NULL);
assert(uc != NULL);
assert(uc->state == conn_pause);
assert(uc->funcs != NULL);
assert(IS_ASCII(uc->protocol));
assert(IS_PROXY(uc->protocol));
if (settings.verbose > 2) {
char key[KEY_MAX_LENGTH + 10];
assert(it->nkey <= KEY_MAX_LENGTH);
memcpy(key, ITEM_key(it), it->nkey);
key[it->nkey] = '\0';
moxi_log_write("<%d cproxy ascii item response, key %s\n",
uc->sfd, key);
}
if (strncmp(ITEM_data(it) + it->nbytes - 2, "\r\n", 2) == 0) {
// TODO: Need to clean up half-written add_iov()'s.
// Consider closing the upstream_conns?
//
uint64_t cas = ITEM_get_cas(it);
if ((cas_emit == 0) ||
(cas_emit < 0 &&
cas == CPROXY_NOT_CAS)) {
if (add_conn_item(uc, it)) {
it->refcount++;
if (add_iov(uc, "VALUE ", 6) == 0 &&
add_iov(uc, ITEM_key(it), it->nkey) == 0 &&
add_iov(uc, ITEM_suffix(it),
it->nsuffix + it->nbytes) == 0) {
if (settings.verbose > 2) {
moxi_log_write("<%d cproxy ascii item response success\n",
uc->sfd);
}
}
}
} else {
char *suffix = add_conn_suffix(uc);
if (suffix != NULL) {
sprintf(suffix, " %llu\r\n", (unsigned long long) cas);
if (add_conn_item(uc, it)) {
it->refcount++;
if (add_iov(uc, "VALUE ", 6) == 0 &&
add_iov(uc, ITEM_key(it), it->nkey) == 0 &&
add_iov(uc, ITEM_suffix(it),
it->nsuffix - 2) == 0 &&
add_iov(uc, suffix, strlen(suffix)) == 0 &&
add_iov(uc, ITEM_data(it), it->nbytes) == 0) {
if (settings.verbose > 2) {
moxi_log_write("<%d cproxy ascii item response ok\n",
uc->sfd);
}
}
}
}
}
} else {
if (settings.verbose > 1) {
moxi_log_write("ERROR: unexpected downstream data block");
}
}
}
void protocol_stats_foreach_write(const void *key,
const void *value,
void *user_data) {
char *line = (char *) value;
conn *uc = (conn *) user_data;
int nline;
cb_assert(line != NULL);
cb_assert(uc != NULL);
(void)key;
nline = strlen(line);
if (nline > 0) {
item *it;
if (settings.verbose > 2) {
moxi_log_write("%d: cproxy_stats writing: %s\n", uc->sfd, line);
}
if (IS_BINARY(uc->protocol)) {
token_t line_tokens[MAX_TOKENS];
size_t line_ntokens = scan_tokens(line, line_tokens, MAX_TOKENS, NULL);
if (line_ntokens == 4) {
uint16_t key_len = line_tokens[NAME_TOKEN].length;
uint32_t data_len = line_tokens[VALUE_TOKEN].length;
it = item_alloc("s", 1, 0, 0,
sizeof(protocol_binary_response_stats) + key_len + data_len);
if (it != NULL) {
protocol_binary_response_stats *header =
(protocol_binary_response_stats *) ITEM_data(it);
memset(ITEM_data(it), 0, it->nbytes);
header->message.header.response.magic = (uint8_t) PROTOCOL_BINARY_RES;
header->message.header.response.opcode = uc->binary_header.request.opcode;
header->message.header.response.keylen = (uint16_t) htons(key_len);
header->message.header.response.bodylen = htonl(key_len + data_len);
header->message.header.response.opaque = uc->opaque;
memcpy((ITEM_data(it)) + sizeof(protocol_binary_response_stats),
line_tokens[NAME_TOKEN].value, key_len);
memcpy((ITEM_data(it)) + sizeof(protocol_binary_response_stats) + key_len,
line_tokens[VALUE_TOKEN].value, data_len);
if (add_conn_item(uc, it)) {
add_iov(uc, ITEM_data(it), it->nbytes);
if (settings.verbose > 2) {
moxi_log_write("%d: cproxy_stats writing binary", uc->sfd);
cproxy_dump_header(uc->sfd, ITEM_data(it));
}
return;
}
item_remove(it);
}
}
return;
}
it = item_alloc("s", 1, 0, 0, nline + 2);
if (it != NULL) {
strncpy(ITEM_data(it), line, nline);
strncpy(ITEM_data(it) + nline, "\r\n", 2);
if (add_conn_item(uc, it)) {
add_iov(uc, ITEM_data(it), nline + 2);
return;
}
item_remove(it);
}
}
}