ninharp/ninlight-w600
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

fix: Fix-Runde 3 für Task 4 - WLAN-Aussetzer bleibt resumable statt sofortigem Abbruch

Browse Source
This commit is contained in:
Michael Sauer
2026-09-12 02:11:56 +02:00
parent 2d5095574d
commit 7474cd77e1
2 changed files with 56 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files
sdk/Src/App/OTA/wm_http_fwup.c
+27 -9
View File
@@ -147,6 +147,20 @@ int http_fwup(HTTPParameters ClientParams)
now_state = SHUTDOWN_LINK;
break;
}
if (nSize == 0) {
/* ninlight: a zero-length read is not a fwup rejection - it's what
* HTTPClient.c's HTTPIntrnRecv()/nSocketEvents<0 paths report for a server
* that closed mid-body or a transient socket hiccup (WLAN dropout), while
* still returning HTTP_CLIENT_SUCCESS/HTTP_CLIENT_EOS. Passing an empty
* packet into socket_fwup_recv() would turn it into the exact same ERR_ABRT
* as a genuine checksum/trailer rejection (wm_socket_fwup.c: datalen==0 goes
* to the same "goto err" as a failed tls_fwup_request_sync()) - the QUIT_OTA-
* on-ERR_ABRT fix below must not see this case, or a routine WLAN dropout
* would abort the update instead of resuming via SHUTDOWN_LINK/PREPARE_PACKET
* like before. See sdk/PATCHES.md point 1 (Fix-Runde 3). */
now_state = SHUTDOWN_LINK;
break;
}
#if USE_BREAK_POINT_RESUME == 0
if(breakFlag == 1){
breakLen += nSize;
@@ -181,15 +195,19 @@ int http_fwup(HTTPParameters ClientParams)
// Send received data to fwup thread and deal with socket issues.
s8 ret = socket_fwup_recv(0, p, ERR_OK);
if(ret != ERR_OK){
/* ninlight: socket_fwup_recv() rejected this chunk (the fwup task hit a
* checksum/trailer mismatch or another fatal error, see wm_fwup.c) - recvLen
* never advances past this point, so going through SHUTDOWN_LINK used to start
* a full Range-resume retry (up to NL_OTA_MAX_RETRY_ROUNDS rounds) whose data
* the fwup task would just keep discarding. Bail out to QUIT_OTA directly so
* the caller gets the real error immediately instead of after minutes. Force
* nRetCode nonzero so the socket_fwup_err() cleanup below still runs. See
* sdk/PATCHES.md point 1. */
nRetCode = ret;
/* ninlight: nSize > 0 is guaranteed here (see the nSize==0 check above), so
* this ERR_ABRT is NOT the "empty packet" case - it means
* wm_socket_fwup.c's socket_fwup_recv() actually rejected non-empty data:
* either tls_fwup_enter() refused the session (e.g. still busy from a
* previous attempt) on the first chunk, or tls_fwup_request_sync() reported a
* checksum/trailer/hash mismatch on a later one (see wm_fwup.c). Either way
* recvLen never advances past this point, so going through SHUTDOWN_LINK used
* to start a full Range-resume retry (up to NL_OTA_MAX_RETRY_ROUNDS rounds)
* whose data the fwup task would just keep discarding. Bail out to QUIT_OTA
* directly so the caller gets the real error immediately instead of after
* minutes. Force nRetCode nonzero so the socket_fwup_err() cleanup below
* still runs. See sdk/PATCHES.md point 1. */
nRetCode = HTTP_CLIENT_ERROR_BAD_STATE;
now_state = QUIT_OTA;
break;
}