WiFiUDP sending large udp packet only works once after powerup

[Rodmg]

Hello, I'm trying to send large UDP packets of (for example) 1345 bytes over my local network, but for some reason it only works once after powering on the ESP8266, after that, I cannot send any other packet of that size until the module is restarted.

Functions udp.beginPacket(), udp.write() and udp.endPacket() return always success.

However smaller packets are still working reliably (tried with a 204 byte packet), even after failing to send those large packets.

I'm using "Packet Sender" (https://packetsender.com/) app for receiving those packets in my PC.

I know that there could be problems with the MTU size, however I find it strange that it actually works the first time.

I'm using the Staging version of the libraries in Arduino 1.6.5.

[chaeplin]

Check this :

Arduino/libraries/ESP8266WiFi/src/include/UdpContext.h

Line 308 in 342c4ae

const size_t pbuf_unit_size = 512;

512 bytes is max.

[igrr]

512 bytes is the maximum size of single pbuf, but multiple pbufs will be allocated and chained if required.

[chaeplin]

I have tested pbuf_unit_size, changed to 1024, 1500, and 512.
https://gist.github.com/chaeplin/1b346a6c0bf6e2b56b28

Using udp.write, can't send more than pbuf_unit_size ?

[me-no-dev]

I could not send anything over 512Bytes as well. changing the unit size to 1460 helped

[manitou48]

I hacked pbuf_unit_size to 1024 and that fixed my problem with sending 1000-byte packets, but the other UDP write problem, I'm having is that I send 10 back-to-back 512-byte packets, but only 5 show up at the linux box (verified on linux box with tcpdump as well as receiving application) ... any thoughts?

[me-no-dev]

do you yield() after each 512 packet?

[manitou48]

just added yield() after endPacket(), didn't help . also confirmed that the first 5 packets make it on to the wire. endPacket() returns 1 every time, saying it thinks all is well.

unsurprisingly, If i add delay(1) between each write() all 10 packets go

[me-no-dev]

@igrr I think the problem with pbuf_unit_size is that when you add a new pbuf to the chain through _reserve, the pbuf added has 0 len so tot_len of the parent pbuf does not change.
Also if there are more pbufs before that they also do not know the new size of the child ones.
A fix to that is to hold the current TX pbuf until we are done adding stuff to it or we have filled it up, after which the pbuf can be cat to the current stack of pbufs and a new one be reserved. That in theory should ensure that all pbufs in a packet chain know what's going on.

[Rodmg]

@me-no-dev I'm currently debugging the problem, the _reserve seems to be working ok, I'm printing tot_len before it returns and it is correct. (i'm using @chaeplin sketch https://gist.github.com/chaeplin/1b346a6c0bf6e2b56b28 for testing).

I will continue debugging and keep you updated.

[Rodmg]

send() seems to be OK too, checked sizes of all the pbuf chain and the adjustment.

I'm thinking the problem could be inside udp_sendto() function, but I cannot find the implementation of that function, do you know where it is?

[me-no-dev]

oh if you are correct the we are screwed with this approach.
udp_sendto() is part of lwip.a library and we do not have the source.
I guess the method can accept only a single pbuf as a packet and does not expect the pbuf to have any other pbufs chained to it? that would be horrible...