We are fragmenting too aggressively

[g-oikonomou]

I observed that DIOs are now being sent fragmented without having grown in size!

The problem is in L1463 in core/net/sicslowpan.c:

if((int)uip_len - (int)uncomp_hdr_len > (int)MAC_MAX_PAYLOAD - framer_hdrlen - (int)rime_hdr_len) {

We are leaving too little space for the L2 payload. We reduce it by 25 first inside MAC_MAX_PAYLOAD (which defaults to 127 - 25) and then we further subtract framer_hdrlen.

There was nothing wrong with the old condition, so am I right to suspect it may be related with crypto at L2?

What we should be doing is subtract the highest of 25 or framer_hdrlen, like so (obviously formatted correctly etc):

(int)127 - (framer_hdrlen > 25 ? framer_hdrlen : 25) - (int)rime_hdr_len)

There are two more instances of MAC_MAX_PAYLOAD - framer_hdrlen which would need adjustments as well.

One could argue that this is not a bug however it needs changed, since at the moment we don't have RPL in platforms which don't support 6lowpan fragmentation.

[aguirrem]

I just noticed this too! I was trying to figure out what had changed.
Could the same be said for platforms that don't have crypto or have it disabled?

[aguirrem]

👍
I have tested the changes proposed here with good results

[adamdunkels]

Nice! Does anyone have a patch/pull request?

[g-oikonomou]

This was more like a bug report.

If the proposed change works it's only a fluke, it was off the top of my head

On 18 Nov 2013, at 06:23, Adam Dunkels [email protected] wrote:

Nice! Does anyone have a patch/pull request?

—
Reply to this email directly or view it on GitHub.

[aguirrem]

I am not very familiar with that part of the code or the reasons why it changed.
Can anyone shed into @g-oikonomou speculations above? I can put a pull request together if the fix is acceptable

[laurentderu]

MAC_MAX_PAYLOAD should be replaced by the maximum PHY frame size. There is one caveat though, if the framer appends bytes after the payload, they are not included in framer_hdrlen. In case of 802.15.4 the FCS is appended, so you have to replace MAC_MAX_PAYLOAD by 127 - 2. By doing that there is no need to use max(framer_hdrlen, 25)

ce99c02 from #557 fixes this as a side effect

Maybe a cleaner solution would be to add a function in framer to return the len of appended data.