libgluster: Fix counting, synctask state errors

When a synctask is executed in synctask_switchto, if by the time
it goes to check the woken/sleep part a reply already comes and
__wake is called, already running task is going to be put in runq
this generates a false warning "re-running already running task".

If the reply does not come before the woken/sleep check, then the
running task is put in waitq which decrements env->runcount even
when the task is not in runq, this leads to -ve runcount everytime
a task goes from runq->switchto->waitq.

This patch fixes both of them by introducing a new state for the
task called SYNCTASK_SUSPEND just when the task is suspended before
yeild in SYNCOP.

Change-Id: Ib82182cf950f9d85b5656f6243541489a104ca3d
BUG: 816551
Signed-off-by: Pranith Kumar K <pranithk@gluster.com>
Reviewed-on: http://review.gluster.com/3249
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vijay@gluster.com>
This commit is contained in:
Pranith Kumar K 2012-04-30 07:29:16 +05:30 committed by Vijay Bellur
parent 170a3a411c
commit 9b3def01e4
2 changed files with 4 additions and 0 deletions

View File

@ -34,6 +34,7 @@ __run (struct synctask *task)
list_del_init (&task->all_tasks);
switch (task->state) {
case SYNCTASK_INIT:
case SYNCTASK_SUSPEND:
break;
case SYNCTASK_RUN:
gf_log (task->xl->name, GF_LOG_WARNING,
@ -65,6 +66,7 @@ __wait (struct synctask *task)
list_del_init (&task->all_tasks);
switch (task->state) {
case SYNCTASK_INIT:
case SYNCTASK_SUSPEND:
break;
case SYNCTASK_RUN:
env->runcount--;

View File

@ -46,6 +46,7 @@ typedef int (*synctask_fn_t) (void *opaque);
typedef enum {
SYNCTASK_INIT = 0,
SYNCTASK_RUN,
SYNCTASK_SUSPEND,
SYNCTASK_WAIT,
SYNCTASK_DONE,
} synctask_state_t;
@ -133,6 +134,7 @@ struct syncargs {
\
STACK_WIND_COOKIE (task->opframe, cbk, (void *)stb, \
subvol, op, params); \
task->state = SYNCTASK_SUSPEND; \
synctask_yield (stb->task); \
STACK_RESET (task->opframe->root); \
} while (0)