diff options
author | Max Romanov <max.romanov@nginx.com> | 2020-02-03 11:20:59 +0300 |
---|---|---|
committer | Max Romanov <max.romanov@nginx.com> | 2020-02-03 11:20:59 +0300 |
commit | 7ea9ebc55a22fec53bb095975e639d4fbfebb575 (patch) | |
tree | 15925f0f0df88ea50955213ccee3b972e0a5639f | |
parent | 8c0f2cebf5eba555d104c42b556e54635f5d0890 (diff) | |
download | unit-7ea9ebc55a22fec53bb095975e639d4fbfebb575.tar.gz unit-7ea9ebc55a22fec53bb095975e639d4fbfebb575.tar.bz2 |
Fixed req_app_link reference counting on cancellation.
Re-scheduled req_app_link structures should have use_count exactly equal
to the number of references from the application and port list. However,
there's one extra usage decrement that occurs after the req_app_link is
created because the use_count is initialised as 1.
This patch removes all excess instances of the usage decrement that caused
preliminary req_app_link release and router process crash.
To reproduce the issue need to cause request rescheduling between 2 app
processes.
This issue was introduced in 61e9f23a566d.
-rw-r--r-- | src/nxt_router.c | 23 |
1 files changed, 7 insertions, 16 deletions
diff --git a/src/nxt_router.c b/src/nxt_router.c index 6a1f3792..3ff048c5 100644 --- a/src/nxt_router.c +++ b/src/nxt_router.c @@ -3750,8 +3750,6 @@ nxt_router_response_error_handler(nxt_task_t *task, nxt_port_recv_msg_t *msg, nxt_router_app_prepare_request(task, req_app_link); } - nxt_request_app_link_use(task, req_app_link, -1); - msg->port_msg.last = 0; return; @@ -4220,28 +4218,23 @@ re_ra_cancelled: if (nxt_router_port_post_select(task, &state) == NXT_OK) { /* * There should be call nxt_request_app_link_inc_use(re_ra), - * but we need to decrement use then. So, let's skip both. + * because of one more link in the queue. + * Corresponding decrement is in nxt_router_app_process_request(). */ + nxt_request_app_link_inc_use(re_ra); + nxt_work_queue_add(&task->thread->engine->fast_work_queue, nxt_router_app_process_request, &task->thread->engine->task, app, re_ra); - - } else { - /* - * This call should be unconditional, but we want to spare - * couple of CPU ticks to postpone the head death of the universe. - */ - - nxt_request_app_link_use(task, re_ra, -1); } } if (req_app_link != NULL) { /* - * Here we do the same trick as described above, - * but without conditions. - * Skip required nxt_request_app_link_inc_use(req_app_link). + * There should be call nxt_request_app_link_inc_use(req_app_link), + * because of one more link in the queue. But one link was + * recently removed from app->requests link. */ nxt_work_queue_add(&task->thread->engine->fast_work_queue, @@ -5205,8 +5198,6 @@ nxt_router_app_timeout(nxt_task_t *task, void *obj, void *data) if (nxt_router_port_post_select(task, &state) == NXT_OK) { nxt_router_app_prepare_request(task, pending_ra); } - - nxt_request_app_link_use(task, pending_ra, -1); } nxt_debug(task, "send quit to app '%V' pid %PI", &app->name, port->pid); |