From: Mateusz Moscicki Date: Thu, 27 Oct 2022 17:53:27 +0000 (+0200) Subject: Fix the race in data_finished() and resp_finished() X-Git-Tag: accepted/tools/devbase/tools/legacy/20240422.110856^0 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=refs%2Fchanges%2F70%2F283570%2F1;p=tools%2Flthor.git Fix the race in data_finished() and resp_finished() Setting the data_finished and resp_finished flags too early sometimes led to a situation in which another chunk was sent, even though the functions had not finished executing, which resulted in a flashing error. For example: When data_finished() was called, the flag data_finished was set. Then after entering aio_error/aio_return the thread would be put to sleep and the program would start executing the resp_finished() function. The function was executed to the end and because data_finished flag was set, the check_next_chunk() was called. After that, the thread with data_finished() would return from sleep and continue its execution, even though the data it was pointing to at that moment is out of date. This led, among other things, to a mismatch between numbers of chunk and the response. Change-Id: I404b3723cb9e671d13221bff2b7823f1fdc43906 Signed-off-by: Mateusz Moscicki (cherry picked from commit 400428531184493e9d025af29286d071f84c1115) --- diff --git a/libthor/thor_net.c b/libthor/thor_net.c index 29cb295..9da314c 100644 --- a/libthor/thor_net.c +++ b/libthor/thor_net.c @@ -242,8 +242,6 @@ static void data_finished(sigval_t sigval) struct t_thor_net_transfer *transfer = chunk->user_data; int ret; - chunk->data_finished = 1; - ret = aio_error(&chunk->data_transfer); if (ret == ECANCELED || transfer->ret) { return; @@ -260,6 +258,7 @@ static void data_finished(sigval_t sigval) return; } + chunk->data_finished = 1; if (chunk->resp_finished) check_next_chunk(chunk, transfer); } @@ -270,7 +269,6 @@ static void resp_finished(sigval_t sigval) struct t_thor_net_transfer *transfer = chunk->user_data; int ret; - chunk->resp_finished = 1; transfer->data_in_progress -= chunk->useful_size; ret = aio_error(&chunk->resp_transfer); @@ -309,6 +307,7 @@ static void resp_finished(sigval_t sigval) chunk->chunk_number, transfer->user_data); + chunk->resp_finished = 1; if (chunk->data_finished) check_next_chunk(chunk, transfer); }