/**
* nginx post-read-request handler to feed request line and headers to Ironbee.
*
* @param[in] r The nginx request object.
* @return Declined (ignoreme) or error status.
*/
static ngx_int_t ironbee_post_read_request(ngx_http_request_t *r)
{
ngx_log_t *prev_log;
ngxib_req_ctx *ctx;
ib_conn_t *iconn;
ib_parsed_req_line_t *rline;
ib_parsed_header_wrapper_t *ibhdrs;
ib_status_t rc;
ngx_list_part_t *part;
ngx_table_elt_t *hdr;
unsigned int i;
/* Don't process internal requests */
if (r->internal)
return NGX_DECLINED;
prev_log = ngxib_log(r->connection->log);
ngx_regex_malloc_init(r->pool);
ctx = ngx_pcalloc(r->pool, sizeof(ngxib_req_ctx));
ctx->r = r;
ngx_http_set_ctx(r, ctx, ngx_ironbee_module);
iconn = ngxib_conn_get(ctx, ironbee);
ib_tx_create(&ctx->tx, iconn, ctx);
/* Notify Ironbee of request line and headers */
rc = ib_parsed_req_line_create(ctx->tx, &rline,
(const char*)r->request_line.data,
r->request_line.len,
(const char*)r->method_name.data,
r->method_name.len,
(const char*)r->unparsed_uri.data,
r->unparsed_uri.len,
(const char*)r->http_protocol.data,
r->http_protocol.len);
if (rc != IB_OK)
cleanup_return(prev_log) NGX_ERROR;
ib_state_notify_request_started(ironbee, ctx->tx, rline);
rc = ib_parsed_name_value_pair_list_wrapper_create(&ibhdrs, ctx->tx);
if (rc != IB_OK)
cleanup_return(prev_log) NGX_ERROR;
for (part = &r->headers_in.headers.part; part != NULL; part = part->next) {
hdr = part->elts;
for (i = 0; i < part->nelts; ++i) {
ib_parsed_name_value_pair_list_add(ibhdrs,
(const char*)hdr->key.data,
hdr->key.len,
(const char*)hdr->value.data,
hdr->value.len);
++hdr;
}
}
rc = ib_state_notify_request_header_data(ironbee, ctx->tx, ibhdrs);
if (rc != IB_OK)
cleanup_return(prev_log) NGX_ERROR;
rc = ib_state_notify_request_header_finished(ironbee, ctx->tx);
if (rc != IB_OK)
cleanup_return(prev_log) NGX_ERROR;
if (!ngxib_has_request_body(r, ctx)) {
rc = ib_state_notify_request_finished(ironbee, ctx->tx);
if (rc != IB_OK)
cleanup_return(prev_log) NGX_ERROR;
}
ctx->hdrs_in = 1;
if (STATUS_IS_ERROR(ctx->status)) {
ctx->internal_errordoc = 1;
cleanup_return(prev_log) ctx->status;
}
cleanup_return(prev_log) NGX_DECLINED;
}
static void ironbee_plugin_pre_remap(TSCont contp, TSHttpTxn txnp)
{
assert(contp != NULL);
assert(txnp != NULL);
tsib_txn_ctx *txndata;
tsib_hdr_outcome status;
int request_inspection_finished = 0;
txndata = TSContDataGet(contp);
assert ((txndata != NULL) && (txndata->tx != NULL));
status = process_hdr(txndata, txnp, &tsib_direction_client_req);
if (HDR_OUTCOME_IS_HTTP_OR_ERROR(status, txndata)) {
if (status == HDR_HTTP_STATUS) {
ib_log_debug_tx(txndata->tx,
"HTTP code %d contp=%p", txndata->status, contp);
}
else {
/* Ironbee set a status we don't handle.
* We returned EINVAL, but we also need housekeeping to
* avoid a crash in modhtp and log something bad.
*/
ib_log_debug_tx(txndata->tx,
"Internal error %d contp=%p", txndata->status, contp);
/* Ugly hack: notifications to stop modhtp bombing out */
request_inspection_finished = 1;
}
}
else {
/* Other nonzero statuses not supported */
switch(status) {
case HDR_OK:
/* If we're not inspecting the Request body,
* we can bring forward notification of end-request
* so any header-only tests run on Request phase
* can abort the tx before opening a backend connection.
*/
if (!ib_flags_all(txndata->tx->flags, IB_TX_FINSPECT_REQBODY)) {
request_inspection_finished = 1;
}
break; /* All's well */
case HDR_HTTP_STATUS:
// FIXME: should we take the initiative here and return 500?
ib_log_error_tx(txndata->tx,
"Internal error: ts-ironbee requested error but no error response set.");
break;
case HDR_HTTP_100:
/* This can't actually happen with current Trafficserver
* versions, as TS will generate a 400 error without
* reference to us. But in case that changes in future ...
*/
ib_log_error_tx(txndata->tx,
"No request headers found.");
break;
default:
ib_log_error_tx(txndata->tx,
"Unhandled state arose in handling request headers.");
break;
}
}
if (request_inspection_finished) {
if (!ib_flags_all(txndata->tx->flags, IB_TX_FREQ_STARTED) ) {
ib_state_notify_request_started(txndata->tx->ib, txndata->tx, NULL);
}
if (!ib_flags_all(txndata->tx->flags, IB_TX_FREQ_FINISHED) ) {
ib_state_notify_request_finished(txndata->tx->ib, txndata->tx);
}
}
else {
/* hook an input filter to watch data */
TSHttpTxnHookAdd(txnp, TS_HTTP_REQUEST_TRANSFORM_HOOK,
txndata->in_data_cont);
}
/* Flag that we can no longer prevent a request going to backend */
ib_tx_flags_set(txndata->tx, IB_TX_FSERVERREQ_STARTED);
/* Check whether Ironbee told us to block the request.
* This could now come not just from process_hdr, but also
* from a brought-forward notification if we aren't inspecting
* a request body and notified request_finished.
*/
if (HTTP_CODE(txndata->status)) {
TSHttpTxnHookAdd(txnp, TS_HTTP_SEND_RESPONSE_HDR_HOOK, contp);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_ERROR);
}
else {
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
}
}
/**
* Plugin for the IronBee ATS.
*
* Handles some ATS events.
*
* @param[in,out] contp Pointer to the continuation
* @param[in,out] event Event from ATS
* @param[in,out] edata Event data
*
* @returns status
*/
int ironbee_plugin(TSCont contp, TSEvent event, void *edata)
{
ib_status_t rc;
TSCont mycont;
TSHttpTxn txnp = (TSHttpTxn) edata;
TSHttpSsn ssnp = (TSHttpSsn) edata;
tsib_txn_ctx *txndata;
tsib_ssn_ctx *ssndata;
tsib_hdr_outcome status;
TSMutex ts_mutex = NULL;
TSDebug("ironbee", "Entering ironbee_plugin with %d", event);
switch (event) {
/* CONNECTION */
case TS_EVENT_HTTP_SSN_START:
/* start of connection */
/* But we can't initialize conn stuff here, because there's
* no API to get the connection stuff required by ironbee
* at this point. So instead, intercept the first TXN
*
* what we can and must do: create a new contp whose
* lifetime is our ssn
*/
ts_mutex = TSMutexCreate();
mycont = TSContCreate(ironbee_plugin, ts_mutex);
TSHttpSsnHookAdd (ssnp, TS_HTTP_TXN_START_HOOK, mycont);
ssndata = TSmalloc(sizeof(*ssndata));
memset(ssndata, 0, sizeof(*ssndata));
/* The only failure here is EALLOC, and if that happens
* we're ****ed anyway
*/
rc = ib_lock_create_malloc(&(ssndata->mutex));
assert(rc == IB_OK);
ssndata->contp = mycont;
ssndata->ts_mutex = ts_mutex;
TSContDataSet(mycont, ssndata);
TSHttpSsnHookAdd (ssnp, TS_HTTP_SSN_CLOSE_HOOK, mycont);
TSHttpSsnReenable (ssnp, TS_EVENT_HTTP_CONTINUE);
break;
case TS_EVENT_HTTP_TXN_START:
{
/* start of Request */
/* First req on a connection, we set up conn stuff */
ib_status_t rc;
ib_engine_t *ib = NULL;
ssndata = TSContDataGet(contp);
ib_lock_lock(ssndata->mutex);
if (ssndata->iconn == NULL) {
rc = tsib_manager_engine_acquire(&ib);
if (rc == IB_DECLINED) {
/* OK, this means the manager is disabled deliberately,
* but otherwise all's well. So this TXN
* gets processed without intervention from Ironbee
* and is invisble when our SSN_CLOSE hook runs.
*/
ib_lock_unlock(ssndata->mutex);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
TSDebug("ironbee", "Decline from engine manager");
break;
}
else if (rc != IB_OK) {
TSError("[ironbee] Failed to acquire engine: %s",
ib_status_to_string(rc));
goto noib_error;
}
if (ib != NULL) {
rc = ib_conn_create(ib, &ssndata->iconn, contp);
if (rc != IB_OK) {
TSError("[ironbee] ib_conn_create: %s",
ib_status_to_string(rc));
tsib_manager_engine_release(ib);
goto noib_error;
}
/* In the normal case, release the engine when the
* connection's memory pool is destroyed */
rc = ib_mm_register_cleanup(ssndata->iconn->mm,
cleanup_ib_connection,
ib);
if (rc != IB_OK) {
TSError("[ironbee] ib_mm_register_cleanup: %s",
ib_status_to_string(rc));
tsib_manager_engine_release(ib);
goto noib_error;
}
TSDebug("ironbee", "CONN CREATE: conn=%p", ssndata->iconn);
ssndata->txnp = txnp;
ssndata->txn_count = ssndata->closing = 0;
rc = ironbee_conn_init(ssndata);
if (rc != IB_OK) {
TSError("[ironbee] ironbee_conn_init: %s",
ib_status_to_string(rc));
goto noib_error;
}
TSContDataSet(contp, ssndata);
TSDebug("ironbee",
"ironbee_plugin: ib_state_notify_conn_opened()");
rc = ib_state_notify_conn_opened(ib, ssndata->iconn);
if (rc != IB_OK) {
TSError("[ironbee] Failed to notify connection opened: %s",
ib_status_to_string(rc));
}
}
else {
/* Use TSError where there's no ib or tx */
TSError("Ironbee: No ironbee engine!");
goto noib_error;
}
}
/* create a txn cont (request ctx) and tx */
txndata = TSmalloc(sizeof(*txndata));
memset(txndata, 0, sizeof(*txndata));
txndata->ssn = ssndata;
txndata->txnp = txnp;
rc = ib_tx_create(&txndata->tx, ssndata->iconn, txndata);
if (rc != IB_OK) {
TSError("[ironbee] Failed to create tx: %d", rc);
tsib_manager_engine_release(ib);
TSfree(txndata);
goto noib_error;
}
++ssndata->txn_count;
ib_lock_unlock(ssndata->mutex);
ib_log_debug_tx(txndata->tx,
"TX CREATE: conn=%p tx=%p id=%s txn_count=%d",
ssndata->iconn, txndata->tx, txndata->tx->id,
txndata->ssn->txn_count);
mycont = TSContCreate(ironbee_plugin, ssndata->ts_mutex);
TSContDataSet(mycont, txndata);
TSHttpTxnHookAdd(txnp, TS_HTTP_TXN_CLOSE_HOOK, mycont);
/* Hook to process responses */
TSHttpTxnHookAdd(txnp, TS_HTTP_READ_RESPONSE_HDR_HOOK, mycont);
/* Hook to process requests */
TSHttpTxnHookAdd(txnp, TS_HTTP_READ_REQUEST_HDR_HOOK, mycont);
/* Create continuations for input and output filtering
* to give them txn lifetime.
*/
txndata->in_data_cont = TSTransformCreate(in_data_event, txnp);
TSContDataSet(txndata->in_data_cont, txndata);
txndata->out_data_cont = TSTransformCreate(out_data_event, txnp);
TSContDataSet(txndata->out_data_cont, txndata);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
noib_error:
ib_lock_unlock(ssndata->mutex);
/* NULL txndata signals this to SEND_RESPONSE */
mycont = TSContCreate(ironbee_plugin, ssndata->ts_mutex);
TSContDataSet(mycont, NULL);
TSError("[ironbee] Internal error initialising for transaction");
TSHttpTxnHookAdd(txnp, TS_HTTP_SEND_RESPONSE_HDR_HOOK, mycont);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_ERROR);
break;
}
/* HTTP RESPONSE */
case TS_EVENT_HTTP_READ_RESPONSE_HDR:
txndata = TSContDataGet(contp);
if (txndata->tx == NULL) {
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
}
/* Feed ironbee the headers if not done already. */
if (!ib_flags_all(txndata->tx->flags, IB_TX_FRES_STARTED)) {
status = process_hdr(txndata, txnp, &tsib_direction_server_resp);
/* OK, if this was an HTTP 100 response, it's not the
* response we're interested in. No headers have been
* sent yet, and no data will be sent until we've
* reached here again with the final response.
*/
if (status == HDR_HTTP_100) {
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
}
// FIXME: Need to know if this fails as it (I think) means
// that the response did not come from the server and
// that ironbee should ignore it.
/* I've not seen a fail here. AFAICT if either the origin
* isn't responding or we're responding from cache. we
* never reach here in the first place.
*/
}
/* If ironbee signalled an error while processing request body data,
* this is the first opportunity to divert to an errordoc
*/
if (HTTP_CODE(txndata->status)) {
ib_log_debug_tx(txndata->tx,
"HTTP code %d contp=%p", txndata->status, contp);
TSHttpTxnHookAdd(txnp, TS_HTTP_SEND_RESPONSE_HDR_HOOK, contp);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_ERROR);
break;
}
/* If we're not going to inspect response body data
* we can bring forward notification of response-end
* so we're in time to respond with an errordoc if Ironbee
* wants to block in the response phase.
*
* This currently fails. However, that appears to be because I
* can't unset IB_TX_FINSPECT_RESBODY with InspectionEngineOptions
*/
if (!ib_flags_all(txndata->tx->flags, IB_TX_FINSPECT_RESBODY)) {
if (!ib_flags_all(txndata->tx->flags, IB_TX_FRES_STARTED) ) {
ib_state_notify_response_started(txndata->tx->ib, txndata->tx, NULL);
}
if (!ib_flags_all(txndata->tx->flags, IB_TX_FRES_FINISHED) ) {
ib_state_notify_response_finished(txndata->tx->ib, txndata->tx);
}
/* Test again for Ironbee telling us to block */
if (HTTP_CODE(txndata->status)) {
TSHttpTxnHookAdd(txnp, TS_HTTP_SEND_RESPONSE_HDR_HOOK, contp);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_ERROR);
break;
}
}
/* Flag that we're too late to divert to an error response */
ib_tx_flags_set(txndata->tx, IB_TX_FCLIENTRES_STARTED);
/* Normal execution. Add output filter to inspect response. */
TSHttpTxnHookAdd(txnp, TS_HTTP_RESPONSE_TRANSFORM_HOOK,
txndata->out_data_cont);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
/* Hook for processing response headers. */
case TS_EVENT_HTTP_SEND_RESPONSE_HDR:
txndata = TSContDataGet(contp);
if (txndata == NULL) {
/* Ironbee is unavailable to help with our response. */
internal_error_response(txnp);
/* This contp isn't going through the normal flow. */
TSContDestroy(contp);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
}
/* If ironbee has sent us into an error response then
* we came here in our error path, with nonzero status.
*/
if (txndata->status != 0) {
error_response(txnp, txndata);
}
/* Feed ironbee the headers if not done already. */
if (!ib_flags_all(txndata->tx->flags, IB_TX_FRES_STARTED)) {
if (process_hdr(txndata, txnp, &tsib_direction_client_resp) != HDR_OK) {
/* I think this is a shouldn't happen event, and that
* if it does we have an ironbee bug or misconfiguration.
* Log an error to catch if it happens in practice.
*/
ib_log_error_tx(txndata->tx, "process_hdr returned error in send_response_hdr event");
}
}
/* If there is an ironbee-generated response body, notify ironbee.
*
* NOTE: I do not see anywhere else to put this as the error body is
* just a buffer and not delivered via normal IO channels, so
* the error body will never get caught by an event.
*/
if ((txndata->status != 0) && (txndata->err_body != NULL)) {
const char *data = txndata->err_body;
size_t data_length = txndata->err_body_len;
ib_log_debug_tx(txndata->tx,
"error_response: calling ib_state_notify_response_body_data() %s:%d",
__FILE__, __LINE__);
ib_state_notify_response_body_data(txndata->tx->ib,
txndata->tx,
data, data_length);
}
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
/* HTTP REQUEST */
case TS_EVENT_HTTP_READ_REQUEST_HDR:
/* hook to examine output headers. They're not available yet */
TSHttpTxnHookAdd(txnp, TS_HTTP_PRE_REMAP_HOOK, contp);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
/* hook for processing incoming request/headers
* The OS_DNS hook is an alternative here.
*/
case TS_EVENT_HTTP_PRE_REMAP:
{
int request_inspection_finished = 0;
txndata = TSContDataGet(contp);
assert ((txndata != NULL) && (txndata->tx != NULL));
status = process_hdr(txndata, txnp, &tsib_direction_client_req);
if (HDR_OUTCOME_IS_HTTP_OR_ERROR(status, txndata)) {
if (status == HDR_HTTP_STATUS) {
ib_log_debug_tx(txndata->tx,
"HTTP code %d contp=%p", txndata->status, contp);
}
else {
/* Ironbee set a status we don't handle.
* We returned EINVAL, but we also need housekeeping to
* avoid a crash in modhtp and log something bad.
*/
ib_log_debug_tx(txndata->tx,
"Internal error %d contp=%p", txndata->status, contp);
/* Ugly hack: notifications to stop modhtp bombing out */
request_inspection_finished = 1;
}
}
else {
/* Other nonzero statuses not supported */
switch(status) {
case HDR_OK:
/* If we're not inspecting the Request body,
* we can bring forward notification of end-request
* so any header-only tests run on Request phase
* can abort the tx before opening a backend connection.
*/
if (!ib_flags_all(txndata->tx->flags, IB_TX_FINSPECT_REQBODY)) {
request_inspection_finished = 1;
}
break; /* All's well */
case HDR_HTTP_STATUS:
// FIXME: should we take the initiative here and return 500?
ib_log_error_tx(txndata->tx,
"Internal error: ts-ironbee requested error but no error response set.");
break;
case HDR_HTTP_100:
/* This can't actually happen with current Trafficserver
* versions, as TS will generate a 400 error without
* reference to us. But in case that changes in future ...
*/
ib_log_error_tx(txndata->tx,
"No request headers found.");
break;
default:
ib_log_error_tx(txndata->tx,
"Unhandled state arose in handling request headers.");
break;
}
}
if (request_inspection_finished) {
if (!ib_flags_all(txndata->tx->flags, IB_TX_FREQ_STARTED) ) {
ib_state_notify_request_started(txndata->tx->ib, txndata->tx, NULL);
}
if (!ib_flags_all(txndata->tx->flags, IB_TX_FREQ_FINISHED) ) {
ib_state_notify_request_finished(txndata->tx->ib, txndata->tx);
}
}
else {
/* hook an input filter to watch data */
TSHttpTxnHookAdd(txnp, TS_HTTP_REQUEST_TRANSFORM_HOOK,
txndata->in_data_cont);
}
/* Flag that we can no longer prevent a request going to backend */
ib_tx_flags_set(txndata->tx, IB_TX_FSERVERREQ_STARTED);
/* Check whether Ironbee told us to block the request.
* This could now come not just from process_hdr, but also
* from a brought-forward notification if we aren't inspecting
* a request body and notified request_finished.
*/
if (HTTP_CODE(txndata->status)) {
TSHttpTxnHookAdd(txnp, TS_HTTP_SEND_RESPONSE_HDR_HOOK, contp);
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_ERROR);
}
else {
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
}
break;
}
/* CLEANUP EVENTS */
case TS_EVENT_HTTP_TXN_CLOSE:
{
txndata = TSContDataGet(contp);
TSContDestroy(txndata->out_data_cont);
TSContDestroy(txndata->in_data_cont);
TSContDataSet(contp, NULL);
TSContDestroy(contp);
if ( (txndata != NULL) && (txndata->tx != NULL) ) {
ib_log_debug_tx(txndata->tx,
"TXN Close: %p", (void *)contp);
tsib_txn_ctx_destroy(txndata);
}
TSHttpTxnReenable(txnp, TS_EVENT_HTTP_CONTINUE);
break;
}
case TS_EVENT_HTTP_SSN_CLOSE:
TSDebug("ironbee", "SSN Close: %p", (void *)contp);
tsib_ssn_ctx_destroy(TSContDataGet(contp));
tsib_manager_engine_cleanup();
TSHttpSsnReenable(ssnp, TS_EVENT_HTTP_CONTINUE);
break;
case TS_EVENT_MGMT_UPDATE:
{
TSDebug("ironbee", "Management update");
ib_status_t rc;
rc = tsib_manager_engine_create();
if (rc != IB_OK) {
TSError("[ironbee] Error creating new engine: %s",
ib_status_to_string(rc));
}
break;
}
/* if we get here we've got a bug */
default:
TSError("[ironbee] *** Unhandled event %d in ironbee_plugin.", event);
break;
}
return 0;
}