summaryrefslogtreecommitdiff
path: root/lib/tevent/tevent_req.c
diff options
context:
space:
mode:
Diffstat (limited to 'lib/tevent/tevent_req.c')
-rw-r--r--lib/tevent/tevent_req.c338
1 files changed, 138 insertions, 200 deletions
diff --git a/lib/tevent/tevent_req.c b/lib/tevent/tevent_req.c
index db47bd93ed..e0cd0bec0f 100644
--- a/lib/tevent/tevent_req.c
+++ b/lib/tevent/tevent_req.c
@@ -2,6 +2,7 @@
Unix SMB/CIFS implementation.
Infrastructure for async requests
Copyright (C) Volker Lendecke 2008
+ Copyright (C) Stefan Metzmacher 2009
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@@ -17,35 +18,32 @@
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
-#include "includes.h"
-#include "lib/tevent/tevent.h"
-#include "lib/talloc/talloc.h"
-#include "lib/util/dlinklist.h"
-#include "lib/async_req/async_req.h"
-
-#ifndef TALLOC_FREE
-#define TALLOC_FREE(ctx) do { talloc_free(ctx); ctx=NULL; } while(0)
-#endif
+#include "replace.h"
+#include "tevent.h"
+#include "tevent_internal.h"
+#include "tevent_util.h"
/**
- * @brief Print an async_req structure
+ * @brief Print an tevent_req structure in debug messages
* @param[in] mem_ctx The memory context for the result
* @param[in] req The request to be printed
* @retval Text representation of req
*
- * This is a default print function for async requests. Implementations should
- * override this with more specific information.
- *
- * This function should not be used by async API users, this is non-static
- * only to allow implementations to easily provide default information in
- * their specific functions.
*/
-char *async_req_print(TALLOC_CTX *mem_ctx, struct async_req *req)
+char *tevent_req_print(TALLOC_CTX *mem_ctx, struct tevent_req *req)
{
- return talloc_asprintf(mem_ctx, "async_req: state=%d, error=%d, "
- "priv=%s", req->state, (int)req->error,
- talloc_get_name(req->private_data));
+ return talloc_asprintf(mem_ctx,
+ "tevent_req[%p/%s]: state[%d] error[%lld (0x%llX)] "
+ " state[%s (%p)] timer[%p]",
+ req, req->internal.location,
+ req->internal.state,
+ (unsigned long long)req->internal.error,
+ (unsigned long long)req->internal.error,
+ talloc_get_name(req->private_state),
+ req->private_state,
+ req->internal.timer
+ );
}
/**
@@ -57,22 +55,40 @@ char *async_req_print(TALLOC_CTX *mem_ctx, struct async_req *req)
* The new async request will be initialized in state ASYNC_REQ_IN_PROGRESS
*/
-struct async_req *async_req_new(TALLOC_CTX *mem_ctx)
+struct tevent_req *_tevent_req_create(TALLOC_CTX *mem_ctx,
+ void *pstate,
+ size_t state_size,
+ const char *type,
+ const char *location)
{
- struct async_req *result;
+ struct tevent_req *req;
+ void **ppstate = (void **)pstate;
+ void *state;
+
+ req = talloc_zero(mem_ctx, struct tevent_req);
+ if (req == NULL) {
+ return NULL;
+ }
+ req->internal.private_type = type;
+ req->internal.location = location;
+ req->internal.state = TEVENT_REQ_IN_PROGRESS;
- result = talloc_zero(mem_ctx, struct async_req);
- if (result == NULL) {
+ state = talloc_size(req, state_size);
+ if (state == NULL) {
+ talloc_free(req);
return NULL;
}
- result->state = ASYNC_REQ_IN_PROGRESS;
- result->print = async_req_print;
- return result;
+ talloc_set_name_const(state, type);
+
+ req->private_state = state;
+
+ *ppstate = state;
+ return req;
}
-static void async_req_finish(struct async_req *req, enum async_req_state state)
+static void tevent_req_finish(struct tevent_req *req, enum tevent_req_state state)
{
- req->state = state;
+ req->internal.state = state;
if (req->async.fn != NULL) {
req->async.fn(req);
}
@@ -87,9 +103,9 @@ static void async_req_finish(struct async_req *req, enum async_req_state state)
* function.
*/
-void async_req_done(struct async_req *req)
+void tevent_req_done(struct tevent_req *req)
{
- async_req_finish(req, ASYNC_REQ_DONE);
+ tevent_req_finish(req, TEVENT_REQ_DONE);
}
/**
@@ -97,37 +113,38 @@ void async_req_done(struct async_req *req)
* @param[in] req The request with an error
* @param[in] error The error code
*
- * async_req_done is to be used by implementors of async requests. When a
+ * tevent_req_done is to be used by implementors of async requests. When a
* request can not successfully completed, the implementation should call this
* function with the appropriate status code.
+ *
+ * If error is 0 the function returns false and does nothing more.
+ *
+ * Call pattern would be
+ * \code
+ * int error = first_function();
+ * if (tevent_req_error(req, error)) {
+ * return;
+ * }
+ *
+ * error = second_function();
+ * if (tevent_req_error(req, error)) {
+ * return;
+ * }
+ *
+ * tevent_req_done(req);
+ * return;
+ * \endcode
*/
-void async_req_error(struct async_req *req, uint64_t error)
-{
- req->error = error;
- async_req_finish(req, ASYNC_REQ_USER_ERROR);
-}
-
-/**
- * @brief Timed event callback
- * @param[in] ev Event context
- * @param[in] te The timed event
- * @param[in] now zero time
- * @param[in] priv The async request to be finished
- */
-
-static void async_trigger(struct tevent_context *ev, struct tevent_timer *te,
- struct timeval now, void *priv)
+bool tevent_req_error(struct tevent_req *req, uint64_t error)
{
- struct async_req *req = talloc_get_type_abort(priv, struct async_req);
-
- TALLOC_FREE(te);
- if (req->error == 0) {
- async_req_done(req);
- }
- else {
- async_req_error(req, req->error);
+ if (error == 0) {
+ return false;
}
+
+ req->internal.error = error;
+ tevent_req_finish(req, TEVENT_REQ_USER_ERROR);
+ return true;
}
/**
@@ -141,25 +158,48 @@ static void async_trigger(struct tevent_context *ev, struct tevent_timer *te,
* Call pattern would be
* \code
* p = talloc(mem_ctx, bla);
- * if (async_req_ntnomem(p, req)) {
+ * if (tevent_req_nomem(p, req)) {
* return;
* }
* \endcode
*/
-bool async_req_nomem(const void *p, struct async_req *req)
+bool tevent_req_nomem(const void *p, struct tevent_req *req)
{
if (p != NULL) {
return false;
}
- async_req_finish(req, ASYNC_REQ_NO_MEMORY);
+ tevent_req_finish(req, TEVENT_REQ_NO_MEMORY);
return true;
}
/**
- * @brief Finish a request before it started processing
+ * @brief Timed event callback
+ * @param[in] ev Event context
+ * @param[in] te The timed event
+ * @param[in] now zero time
+ * @param[in] priv The async request to be finished
+ */
+static void tevent_req_trigger(struct tevent_context *ev,
+ struct tevent_timer *te,
+ struct timeval zero,
+ void *private_data)
+{
+ struct tevent_req *req = talloc_get_type(private_data,
+ struct tevent_req);
+
+ talloc_free(req->internal.trigger);
+ req->internal.trigger = NULL;
+
+ tevent_req_finish(req, req->internal.state);
+}
+
+/**
+ * @brief Finish a request before the caller had the change to set the callback
* @param[in] req The finished request
- * @param[in] status The success code
+ * @param[in] ev The tevent_context for the timed event
+ * @retval On success req will be returned,
+ * on failure req will be destroyed
*
* An implementation of an async request might find that it can either finish
* the request without waiting for an external event, or it can't even start
@@ -169,170 +209,68 @@ bool async_req_nomem(const void *p, struct async_req *req)
* conventions, independent of whether the request was actually deferred.
*/
-bool async_post_error(struct async_req *req, struct tevent_context *ev,
- uint64_t error)
+struct tevent_req *tevent_req_post(struct tevent_req *req,
+ struct tevent_context *ev)
{
- req->error = error;
-
- if (tevent_add_timer(ev, req, timeval_zero(),
- async_trigger, req) == NULL) {
- return false;
+ req->internal.trigger = tevent_add_timer(ev, req, ev_timeval_zero(),
+ tevent_req_trigger, req);
+ if (!req->internal.trigger) {
+ talloc_free(req);
+ return NULL;
}
- return true;
-}
-bool async_req_is_error(struct async_req *req, enum async_req_state *state,
- uint64_t *error)
-{
- if (req->state == ASYNC_REQ_DONE) {
- return false;
- }
- if (req->state == ASYNC_REQ_USER_ERROR) {
- *error = req->error;
- }
- *state = req->state;
- return true;
+ return req;
}
-static void async_req_timedout(struct tevent_context *ev,
- struct tevent_timer *te,
- struct timeval now,
- void *priv)
+bool tevent_req_is_in_progress(struct tevent_req *req)
{
- struct async_req *req = talloc_get_type_abort(priv, struct async_req);
- TALLOC_FREE(te);
- async_req_finish(req, ASYNC_REQ_TIMED_OUT);
-}
+ if (req->internal.state == TEVENT_REQ_IN_PROGRESS) {
+ return true;
+ }
-bool async_req_set_timeout(struct async_req *req, struct tevent_context *ev,
- struct timeval to)
-{
- return (tevent_add_timer(
- ev, req, timeval_current_ofs(to.tv_sec, to.tv_usec),
- async_req_timedout, req)
- != NULL);
+ return false;
}
-struct async_req *async_wait_send(TALLOC_CTX *mem_ctx,
- struct tevent_context *ev,
- struct timeval to)
+bool tevent_req_is_error(struct tevent_req *req, enum tevent_req_state *state,
+ uint64_t *error)
{
- struct async_req *result;
-
- result = async_req_new(mem_ctx);
- if (result == NULL) {
- return result;
+ if (req->internal.state == TEVENT_REQ_DONE) {
+ return false;
}
- if (!async_req_set_timeout(result, ev, to)) {
- TALLOC_FREE(result);
- return NULL;
+ if (req->internal.state == TEVENT_REQ_USER_ERROR) {
+ *error = req->internal.error;
}
- return result;
-}
-
-bool async_wait_recv(struct async_req *req)
-{
+ *state = req->internal.state;
return true;
}
-struct async_queue_entry {
- struct async_queue_entry *prev, *next;
- struct async_req_queue *queue;
- struct async_req *req;
- void (*trigger)(struct async_req *req);
-};
-
-struct async_req_queue {
- struct async_queue_entry *queue;
-};
-
-struct async_req_queue *async_req_queue_init(TALLOC_CTX *mem_ctx)
-{
- return talloc_zero(mem_ctx, struct async_req_queue);
-}
-
-static int async_queue_entry_destructor(struct async_queue_entry *e)
+static void tevent_req_timedout(struct tevent_context *ev,
+ struct tevent_timer *te,
+ struct timeval now,
+ void *private_data)
{
- struct async_req_queue *queue = e->queue;
+ struct tevent_req *req = talloc_get_type(private_data,
+ struct tevent_req);
- DLIST_REMOVE(queue->queue, e);
+ talloc_free(req->internal.timer);
+ req->internal.timer = NULL;
- if (queue->queue != NULL) {
- queue->queue->trigger(queue->queue->req);
- }
-
- return 0;
+ tevent_req_finish(req, TEVENT_REQ_TIMED_OUT);
}
-static void async_req_immediate_trigger(struct tevent_context *ev,
- struct tevent_timer *te,
- struct timeval now,
- void *priv)
+bool tevent_req_set_timeout(struct tevent_req *req,
+ struct tevent_context *ev,
+ struct timeval endtime)
{
- struct async_queue_entry *e = talloc_get_type_abort(
- priv, struct async_queue_entry);
-
- TALLOC_FREE(te);
- e->trigger(e->req);
-}
-
-bool async_req_enqueue(struct async_req_queue *queue, struct tevent_context *ev,
- struct async_req *req,
- void (*trigger)(struct async_req *req))
-{
- struct async_queue_entry *e;
- bool busy;
-
- busy = (queue->queue != NULL);
+ talloc_free(req->internal.timer);
- e = talloc(req, struct async_queue_entry);
- if (e == NULL) {
+ req->internal.timer = tevent_add_timer(ev, req, endtime,
+ tevent_req_timedout,
+ req);
+ if (tevent_req_nomem(req->internal.timer, req)) {
return false;
}
- e->req = req;
- e->trigger = trigger;
- e->queue = queue;
-
- DLIST_ADD_END(queue->queue, e, struct async_queue_entry *);
- talloc_set_destructor(e, async_queue_entry_destructor);
-
- if (!busy) {
- struct tevent_timer *te;
-
- te = tevent_add_timer(ev, e, timeval_zero(),
- async_req_immediate_trigger,
- e);
- if (te == NULL) {
- TALLOC_FREE(e);
- return false;
- }
- }
-
return true;
}
-bool _async_req_setup(TALLOC_CTX *mem_ctx, struct async_req **preq,
- void *pstate, size_t state_size, const char *typename)
-{
- struct async_req *req;
- void **ppstate = (void **)pstate;
- void *state;
-
- req = async_req_new(mem_ctx);
- if (req == NULL) {
- return false;
- }
- state = talloc_size(req, state_size);
- if (state == NULL) {
- TALLOC_FREE(req);
- return false;
- }
- talloc_set_name_const(state, typename);
- req->private_data = state;
-
- *preq = req;
- *ppstate = state;
-
- return true;
-}