Enable all concurrent tests.

[govint]

Retain calloc of response buffer, verify original issue is reproduced. Minor fix in vmci_client.c code where the request sent is kept within MAXBUF size limit.

The build is successful and original panic isn't reproduced. Most likely it may be a GO/C run time issue that was addressed with the recent upgrade of Go run time.

[pdhamdhere]

Remove this comment.

[pdhamdhere]

Great this fixes the issue but I am not sure if I understand the fix. Original code had "+1" to accomodate terminating null byte. Does it mean either c-go or vmci has issues when buffers crosses 1M boundry and another way to fix is make MAXBUF = (1024 * 1024) - 1 ?

[pdhamdhere]

@govint ?

[govint]

Sorry missed updating, the original code was creating a string which was MAXBUF + 1, the strnlen() can create a string that can be at most MAXBUF length in size and over that it adds the NULL byte. The change is to create a string thats MAXBUF -1 (1MB -1) plus the NULL byte which makes a total of MAXBUF. We may never have a request size of 1MB but the code is not correct as is.

[pdhamdhere]

I am still puzzled how this change automagically fixes crash in concurrency test. Did we have a case where req.mlen is (MAXBUF+1)?

[govint]

This fix is from reviewing the code, the original panic isn't reproducible anymore with the GO env. upgrade.

[msterin]

@govint - thanks for pointing to this. While this +- issue could not have been a reason for test failing (see below) , the buffer size control needs fixing, albeit in a separate PR.

Here is the summary :

the original code was creating a string which was MAXBUF + 1,

Actually it was not - the string was created in GO with whatever size, and the code tried to make sure that strings > 1M were truncated before being passed to VMCI but the actual truncation is missing (indeed a bug).
However, this was never hit, because vmdk_opsd.py MAX_JSON_SIZE is set to 4K, and for vmci message > 4K the vmci_server.c receive() code will reject the message and drop the connection with "message too large" , without even trying to receive the message body !

The bottom line:

• Please remove this line from the PR and simply enable the test. LGTM from me on that
• I also suggest a separate PR that cleans up the inconsistency. I suggest making MAXBUF consistent with MAX_JSON_SIZE, put comments in the code with cross-reference, and then limit strings to MAXBUF-1 plus NULL termination -

In this new PR this specific text would change to

   req.mlen = strnlen(json_request, MAXBUF - 1) + 1;
   req.msg = json_request;       // sizeof string, with null temintation
   req.msg[req.mlen] = NULL;  // just in case we hit the MAXBUF limit

[govint]

Absolutely, I was just responding on the string sizing logic of what I'd made.