/**
* The routeQuery entry point. This is passed the query buffer
* to which the filter should be applied. Once applied the
* query should normally be passed to the downstream component
* (filter or router) in the filter chain.
*
* If my_session->residual is set then duplicate that many bytes
* and send them to the branch.
*
* If my_session->residual is zero then this must be a new request
* Extract the SQL text if possible, match against that text and forward
* the request. If the requets is not contained witin the packet we have
* then set my_session->residual to the number of outstanding bytes
*
* @param instance The filter instance data
* @param session The filter session
* @param queue The query data
*/
static int
routeQuery(FILTER *instance, void *session, GWBUF *queue)
{
TEE_INSTANCE *my_instance = (TEE_INSTANCE *)instance;
TEE_SESSION *my_session = (TEE_SESSION *)session;
char *ptr;
int rval;
GWBUF *buffer = NULL, *clone = NULL;
unsigned char command = gwbuf_length(queue) >= 5 ?
*((unsigned char*)queue->start + 4) : 1;
#ifdef SS_DEBUG
skygw_log_write(LOGFILE_TRACE,"Tee routeQuery: %d : %s",
atomic_add(&debug_seq,1),
((char*)queue->start + 5));
#endif
spinlock_acquire(&my_session->tee_lock);
if(!my_session->active)
{
skygw_log_write(LOGFILE_TRACE, "Tee: Received a reply when the session was closed.");
gwbuf_free(queue);
spinlock_release(&my_session->tee_lock);
return 0;
}
if(my_session->queue)
{
my_session->queue = gwbuf_append(my_session->queue,queue);
buffer = modutil_get_next_MySQL_packet(&my_session->queue);
}
else
{
buffer = modutil_get_next_MySQL_packet(&queue);
my_session->queue = queue;
}
if(buffer == NULL)
{
spinlock_release(&my_session->tee_lock);
return 1;
}
clone = clone_query(my_instance, my_session,buffer);
spinlock_release(&my_session->tee_lock);
/* Reset session state */
if(!reset_session_state(my_session,buffer))
return 0;
/** Route query downstream */
spinlock_acquire(&my_session->tee_lock);
rval = route_single_query(my_instance,my_session,buffer,clone);
spinlock_release(&my_session->tee_lock);
return rval;
}
/**
* We have data from the client, this is a SQL command, or other MySQL
* packet type.
*
* @param instance The router instance
* @param router_session The router session returned from the newSession call
* @param queue The queue of data buffers to route
* @return The number of bytes sent
*/
static int
execute(ROUTER *rinstance, void *router_session, GWBUF *queue)
{
INFO_INSTANCE *instance = (INFO_INSTANCE *)rinstance;
INFO_SESSION *session = (INFO_SESSION *)router_session;
uint8_t *data;
int length, len, residual;
char *sql;
if (GWBUF_TYPE(queue) == GWBUF_TYPE_HTTP)
{
return handle_url(instance, session, queue);
}
if (session->queue)
{
queue = gwbuf_append(session->queue, queue);
session->queue = NULL;
queue = gwbuf_make_contiguous(queue);
}
data = (uint8_t *)GWBUF_DATA(queue);
length = data[0] + (data[1] << 8) + (data[2] << 16);
if (length + 4 > GWBUF_LENGTH(queue))
{
// Incomplete packet, must be buffered
session->queue = queue;
return 1;
}
// We have a complete request in a signle buffer
if (modutil_MySQL_Query(queue, &sql, &len, &residual))
{
sql = strndup(sql, len);
int rc = maxinfo_execute_query(instance, session, sql);
free(sql);
return rc;
}
else
{
switch (MYSQL_COMMAND(queue))
{
case COM_PING:
return maxinfo_ping(instance, session, queue);
case COM_STATISTICS:
return maxinfo_statistics(instance, session, queue);
default:
MXS_ERROR("maxinfo: Unexpected MySQL command 0x%x",
MYSQL_COMMAND(queue));
}
}
return 1;
}
/**
* Send a "fake" format description event to the newly connected slave
*
* @param router The router instance
* @param slave The slave to send the event to
*/
static void
blr_slave_send_fde(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave)
{
BLFILE *file;
REP_HEADER hdr;
GWBUF *record, *head;
uint8_t *ptr;
uint32_t chksum;
if ((file = blr_open_binlog(router, slave->binlogfile)) == NULL)
return;
if ((record = blr_read_binlog(router, file, 4, &hdr)) == NULL)
{
blr_close_binlog(router, file);
return;
}
blr_close_binlog(router, file);
head = gwbuf_alloc(5);
ptr = GWBUF_DATA(head);
encode_value(ptr, hdr.event_size + 1, 24); // Payload length
ptr += 3;
*ptr++ = slave->seqno++;
*ptr++ = 0; // OK
head = gwbuf_append(head, record);
ptr = GWBUF_DATA(record);
encode_value(ptr, time(0), 32); // Overwrite timestamp
ptr += 13;
encode_value(ptr, 0, 32); // Set next position to 0
/*
* Since we have changed the timestamp we must recalculate the CRC
*
* Position ptr to the start of the event header,
* calculate a new checksum
* and write it into the header
*/
ptr = GWBUF_DATA(record) + hdr.event_size - 4;
chksum = crc32(0L, NULL, 0);
chksum = crc32(chksum, GWBUF_DATA(record), hdr.event_size - 4);
encode_value(ptr, chksum, 32);
slave->dcb->func.write(slave->dcb, head);
}
/**
* The clientReply entry point. This is passed the response buffer
* to which the filter should be applied. Once processed the
* query is passed to the upstream component
* (filter or router) in the filter chain.
*
* @param instance The filter instance data
* @param session The filter session
* @param reply The response data
*/
static int
clientReply(FILTER* instance, void *session, GWBUF *reply)
{
int rc, branch, eof;
TEE_SESSION *my_session = (TEE_SESSION *) session;
bool route = false, mpkt;
GWBUF *complete = NULL;
unsigned char *ptr;
uint16_t flags = 0;
int min_eof = my_session->command != 0x04 ? 2 : 1;
int more_results = 0;
#ifdef SS_DEBUG
int prev_debug_seq = atomic_add(&debug_seq, 1);
ptr = (unsigned char*) reply->start;
MXS_INFO("Tee clientReply [%s] [%s] [%s]: %d",
instance ? "parent" : "child",
my_session->active ? "open" : "closed",
PTR_IS_ERR(ptr) ? "ERR" : PTR_IS_OK(ptr) ? "OK" : "RSET",
prev_debug_seq);
#endif
spinlock_acquire(&my_session->tee_lock);
if (!my_session->active)
{
MXS_INFO("Tee: Failed to return reply, session is closed");
gwbuf_free(reply);
rc = 0;
if (my_session->waiting[PARENT])
{
GWBUF* errbuf = modutil_create_mysql_err_msg(1, 0, 1, "0000", "Session closed.");
my_session->waiting[PARENT] = false;
my_session->up.clientReply(my_session->up.instance,
my_session->up.session,
errbuf);
}
goto retblock;
}
branch = instance == NULL ? CHILD : PARENT;
my_session->tee_partials[branch] = gwbuf_append(my_session->tee_partials[branch], reply);
my_session->tee_partials[branch] = gwbuf_make_contiguous(my_session->tee_partials[branch]);
complete = modutil_get_complete_packets(&my_session->tee_partials[branch]);
if (complete == NULL)
{
/** Incomplete packet */
MXS_DEBUG("tee.c: Incomplete packet, "
"waiting for a complete packet before forwarding.");
rc = 1;
goto retblock;
}
complete = gwbuf_make_contiguous(complete);
if (my_session->tee_partials[branch] &&
GWBUF_EMPTY(my_session->tee_partials[branch]))
{
gwbuf_free(my_session->tee_partials[branch]);
my_session->tee_partials[branch] = NULL;
}
ptr = (unsigned char*) complete->start;
if (my_session->replies[branch] == 0)
{
MXS_INFO("Tee: First reply to a query for [%s].", branch == PARENT ? "PARENT" : "CHILD");
/* Reply is in a single packet if it is an OK, ERR or LOCAL_INFILE packet.
* Otherwise the reply is a result set and the amount of packets is unknown.
*/
if (PTR_IS_ERR(ptr) || PTR_IS_LOCAL_INFILE(ptr) ||
PTR_IS_OK(ptr) || !my_session->multipacket[branch])
{
my_session->waiting[branch] = false;
my_session->multipacket[branch] = false;
if (PTR_IS_OK(ptr))
{
flags = get_response_flags(ptr, true);
more_results = (flags & 0x08) && my_session->client_multistatement;
if (more_results)
{
MXS_INFO("Tee: [%s] waiting for more results.", branch == PARENT ? "PARENT" : "CHILD");
}
}
}
#ifdef SS_DEBUG
else
{
MXS_DEBUG("tee.c: [%ld] Waiting for a result set from %s session.",
my_session->d_id,
branch == PARENT ? "parent" : "child");
}
#endif
}
if (my_session->waiting[branch])
{
eof = modutil_count_signal_packets(complete, my_session->use_ok, my_session->eof[branch] > 0, &more_results);
more_results &= my_session->client_multistatement;
my_session->eof[branch] += eof;
if (my_session->eof[branch] >= min_eof)
{
#ifdef SS_DEBUG
MXS_DEBUG("tee.c [%ld] %s received last EOF packet",
my_session->d_id,
branch == PARENT ? "parent" : "child");
#endif
my_session->waiting[branch] = more_results;
if (more_results)
{
my_session->eof[branch] = 0;
}
}
}
if (branch == PARENT)
{
my_session->tee_replybuf = gwbuf_append(my_session->tee_replybuf, complete);
}
else
{
gwbuf_free(complete);
}
my_session->replies[branch]++;
rc = 1;
mpkt = my_session->multipacket[PARENT] || my_session->multipacket[CHILD];
if (my_session->tee_replybuf != NULL)
{
if (my_session->branch_session == NULL)
{
rc = 0;
gwbuf_free(my_session->tee_replybuf);
my_session->tee_replybuf = NULL;
MXS_ERROR("Tee child session was closed.");
}
if (mpkt)
{
if (my_session->waiting[PARENT])
{
route = true;
}
else if (my_session->eof[PARENT] >= min_eof &&
my_session->eof[CHILD] >= min_eof)
{
route = true;
#ifdef SS_DEBUG
MXS_DEBUG("tee.c:[%ld] Routing final packet of response set.", my_session->d_id);
#endif
}
}
else if (!my_session->waiting[PARENT] &&
!my_session->waiting[CHILD])
{
#ifdef SS_DEBUG
MXS_DEBUG("tee.c:[%ld] Routing single packet response.", my_session->d_id);
#endif
route = true;
}
}
if (route)
{
#ifdef SS_DEBUG
MXS_DEBUG("tee.c:[%ld] Routing buffer '%p' parent(waiting [%s] replies [%d] eof[%d])"
" child(waiting [%s] replies[%d] eof [%d])",
my_session->d_id,
my_session->tee_replybuf,
my_session->waiting[PARENT] ? "true" : "false",
my_session->replies[PARENT],
my_session->eof[PARENT],
my_session->waiting[CHILD] ? "true" : "false",
my_session->replies[CHILD],
my_session->eof[CHILD]);
#endif
rc = my_session->up.clientReply(my_session->up.instance,
my_session->up.session,
my_session->tee_replybuf);
my_session->tee_replybuf = NULL;
}
if (my_session->queue &&
!my_session->waiting[PARENT] &&
!my_session->waiting[CHILD])
{
GWBUF* buffer = modutil_get_next_MySQL_packet(&my_session->queue);
GWBUF* clone = clone_query(my_session->instance, my_session, buffer);
reset_session_state(my_session, buffer);
route_single_query(my_session->instance, my_session, buffer, clone);
MXS_INFO("tee: routing queued query");
}
retblock:
spinlock_release(&my_session->tee_lock);
return rc;
}
/**
* We have a registered slave that is behind the current leading edge of the
* binlog. We must replay the log entries to bring this node up to speed.
*
* There may be a large number of records to send to the slave, the process
* is triggered by the slave COM_BINLOG_DUMP message and all the events must
* be sent without receiving any new event. This measn there is no trigger into
* MaxScale other than this initial message. However, if we simply send all the
* events we end up with an extremely long write queue on the DCB and risk
* running the server out of resources.
*
* The slave catchup routine will send a burst of replication events per single
* call. The paramter "long" control the number of events in the burst. The
* short burst is intended to be used when the master receive an event and
* needs to put the slave into catchup mode. This prevents the slave taking
* too much tiem away from the thread that is processing the master events.
*
* At the end of the burst a fake EPOLLOUT event is added to the poll event
* queue. This ensures that the slave callback for processing DCB write drain
* will be called and future catchup requests will be handled on another thread.
*
* @param router The binlog router
* @param slave The slave that is behind
* @param large Send a long or short burst of events
* @return The number of bytes written
*/
int
blr_slave_catchup(ROUTER_INSTANCE *router, ROUTER_SLAVE *slave, bool large)
{
GWBUF *head, *record;
REP_HEADER hdr;
int written, rval = 1, burst;
int rotating;
unsigned long burst_size;
uint8_t *ptr;
if (large)
burst = router->long_burst;
else
burst = router->short_burst;
burst_size = router->burst_size;
spinlock_acquire(&slave->catch_lock);
if (slave->cstate & CS_BUSY)
{
spinlock_release(&slave->catch_lock);
return 0;
}
slave->cstate |= CS_BUSY;
spinlock_release(&slave->catch_lock);
if (slave->file == NULL)
{
rotating = router->rotating;
if ((slave->file = blr_open_binlog(router, slave->binlogfile)) == NULL)
{
if (rotating)
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
slave->cstate &= ~CS_BUSY;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
return rval;
}
LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR,
"blr_slave_catchup failed to open binlog file %s",
slave->binlogfile)));
slave->cstate &= ~CS_BUSY;
slave->state = BLRS_ERRORED;
dcb_close(slave->dcb);
return 0;
}
}
slave->stats.n_bursts++;
while (burst-- && burst_size > 0 &&
(record = blr_read_binlog(router, slave->file, slave->binlog_pos, &hdr)) != NULL)
{
head = gwbuf_alloc(5);
ptr = GWBUF_DATA(head);
encode_value(ptr, hdr.event_size + 1, 24);
ptr += 3;
*ptr++ = slave->seqno++;
*ptr++ = 0; // OK
head = gwbuf_append(head, record);
if (hdr.event_type == ROTATE_EVENT)
{
unsigned long beat1 = hkheartbeat;
blr_close_binlog(router, slave->file);
if (hkheartbeat - beat1 > 1) LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR, "blr_close_binlog took %d beats",
hkheartbeat - beat1)));
blr_slave_rotate(slave, GWBUF_DATA(record));
beat1 = hkheartbeat;
if ((slave->file = blr_open_binlog(router, slave->binlogfile)) == NULL)
{
if (rotating)
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
slave->cstate &= ~CS_BUSY;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
return rval;
}
LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR,
"blr_slave_catchup failed to open binlog file %s",
slave->binlogfile)));
slave->state = BLRS_ERRORED;
dcb_close(slave->dcb);
break;
}
if (hkheartbeat - beat1 > 1) LOGIF(LE, (skygw_log_write(
LOGFILE_ERROR, "blr_open_binlog took %d beats",
hkheartbeat - beat1)));
}
slave->stats.n_bytes += gwbuf_length(head);
written = slave->dcb->func.write(slave->dcb, head);
if (written && hdr.event_type != ROTATE_EVENT)
{
slave->binlog_pos = hdr.next_pos;
}
rval = written;
slave->stats.n_events++;
burst_size -= hdr.event_size;
}
if (record == NULL)
slave->stats.n_failed_read++;
spinlock_acquire(&slave->catch_lock);
slave->cstate &= ~CS_BUSY;
spinlock_release(&slave->catch_lock);
if (record)
{
slave->stats.n_flows++;
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
}
else if (slave->binlog_pos == router->binlog_position &&
strcmp(slave->binlogfile, router->binlog_name) == 0)
{
int state_change = 0;
spinlock_acquire(&router->binlog_lock);
spinlock_acquire(&slave->catch_lock);
/*
* Now check again since we hold the router->binlog_lock
* and slave->catch_lock.
*/
if (slave->binlog_pos != router->binlog_position ||
strcmp(slave->binlogfile, router->binlog_name) != 0)
{
slave->cstate &= ~CS_UPTODATE;
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
spinlock_release(&router->binlog_lock);
poll_fake_write_event(slave->dcb);
}
else
{
if ((slave->cstate & CS_UPTODATE) == 0)
{
slave->stats.n_upd++;
slave->cstate |= CS_UPTODATE;
spinlock_release(&slave->catch_lock);
spinlock_release(&router->binlog_lock);
state_change = 1;
}
}
if (state_change)
{
slave->stats.n_caughtup++;
if (slave->stats.n_caughtup == 1)
{
LOGIF(LM, (skygw_log_write(LOGFILE_MESSAGE,
"%s: Slave %s is up to date %s, %u.",
router->service->name,
slave->dcb->remote,
slave->binlogfile, slave->binlog_pos)));
}
else if ((slave->stats.n_caughtup % 50) == 0)
{
LOGIF(LM, (skygw_log_write(LOGFILE_MESSAGE,
"%s: Slave %s is up to date %s, %u.",
router->service->name,
slave->dcb->remote,
slave->binlogfile, slave->binlog_pos)));
}
}
}
else
{
if (slave->binlog_pos >= blr_file_size(slave->file)
&& router->rotating == 0
&& strcmp(router->binlog_name, slave->binlogfile) != 0
&& blr_master_connected(router))
{
/* We may have reached the end of file of a non-current
* binlog file.
*
* Note if the master is rotating there is a window during
* which the rotate event has been written to the old binlog
* but the new binlog file has not yet been created. Therefore
* we ignore these issues during the rotate processing.
*/
LOGIF(LE, (skygw_log_write(LOGFILE_ERROR,
"Slave reached end of file for binlong file %s at %u "
"which is not the file currently being downloaded. "
"Master binlog is %s, %lu.",
slave->binlogfile, slave->binlog_pos,
router->binlog_name, router->binlog_position)));
if (blr_slave_fake_rotate(router, slave))
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
}
else
{
slave->state = BLRS_ERRORED;
dcb_close(slave->dcb);
}
}
else
{
spinlock_acquire(&slave->catch_lock);
slave->cstate |= CS_EXPECTCB;
spinlock_release(&slave->catch_lock);
poll_fake_write_event(slave->dcb);
}
}
return rval;
}
/**
* Read event for EPOLLIN on the httpd protocol module.
*
* @param dcb The descriptor control block
* @return
*/
static int
httpd_read_event(DCB *dcb)
{
SESSION *session = dcb->session;
GWBUF *buf = NULL;
char *ptr, *sol;
HTTPD_session *client_data = NULL;
int n;
// Read all the available data
if ((n = dcb_read(dcb, &buf)) != -1)
{
client_data = dcb->data;
if (client_data->saved)
{
buf = gwbuf_append(client_data->saved, buf);
client_data->saved = NULL;
}
buf = gwbuf_make_contiguous(buf);
ptr = GWBUF_DATA(buf);
if (strncasecmp(ptr, "POST", 4))
{
client_data->method = METHOD_POST;
gwbuf_add_property(buf, "Method", "POST");
ptr = ptr + 4;
}
else if (strncasecmp(ptr, "PUT", 3))
{
client_data->method = METHOD_PUT;
gwbuf_add_property(buf, "Method", "PUT");
ptr = ptr + 3;
}
else if (strncasecmp(ptr, "GET", 3))
{
client_data->method = METHOD_GET;
gwbuf_add_property(buf, "Method", "GET");
ptr = ptr + 3;
}
else if (strncasecmp(ptr, "HEAD", 4))
{
client_data->method = METHOD_HEAD;
gwbuf_add_property(buf, "Method", "HEAD");
ptr = ptr + 4;
}
while (ptr < (char *)(buf->end) && isspace(*ptr))
ptr++;
sol = ptr;
while (ptr < (char *)(buf->end) && isspace(*ptr) == 0)
ptr++;
client_data->url = strndup(sol, ptr - sol);
gwbuf_add_property(buf, "URL", client_data->url);
while ((sol = httpd_nextline(buf, ptr)) != NULL &&
*sol != '\n' && *sol != '\r')
{
httpd_process_header(buf, sol, client_data);
ptr = sol;
}
/*
* We have read all the headers, or run out of data to
* examine.
*/
if (sol == NULL)
{
client_data->saved = buf;
return 0;
}
else
{
if (((char *)(buf->end) - sol)
< client_data->request_len)
{
client_data->saved = buf;
}
else
{
LOGIF(LT, (skygw_log_write(
LOGFILE_TRACE,
"HTTPD: request %s.\n", client_data->url)));
SESSION_ROUTE_QUERY(session, buf);
if (client_data->url)
{
free(client_data->url);
client_data->url = NULL;
}
}
}
}
return 0;
}
