Ephemeral source port not random #31
Assignees
Labels
Comments
This number is uninitialized, and is checked and increased by upstream lwIP (ref) I understand the issue, thanks for the links, and we can in this repository apply the same patch as in lwIP-2.0.3 used by rtos-sdk. |
d-a-v
added a commit
to d-a-v/Arduino
that referenced
this issue
Mar 22, 2019
d-a-v
added a commit
to esp8266/Arduino
that referenced
this issue
Apr 5, 2019
ref: d-a-v/esp82xx-nonos-linklayer#31 origin: #5902 me-no-dev/ESPAsyncTCP#108 Following the links above is instructive. To summarize: * currently and from a long time lwIP tcp client connections always uses the same tcp source port number right after boot * this port number is increased everytime a new one is needed (= new tcp client connection) (to be noted, linux has the same increasing behavior) * when connecting to the same server (right after boot), the triplet (esp-ip-address, source port, destination port) are the same, and may hit remote server list of sockets in time-wait-state (previous connection unproperly closed from the same esp). Consequently the new connection fails when it happens. * this is happening only when debugging (esp reboots often, in less time than time-wait expiration), so the nasty effect is amplified especially when bugs are being chased * efforts had been done when espressif's lwIP implementation wasn't open source, with WiFiClient::setLocalPortStart() #632 but it must be explicitely called with a different random number at every reboot. Efficient but not ideal. This PR uses espressif firmware's r_rand() everytime a new local source port is needed. A different source port number is now showed by tcpdump right after boot. Source port range and duplication is verified everytime in lwIP's src/core/tcp.c:tcp_new_port(). It is implemented as a local patch for upstream lwIP so it is valid not only with WiFiClient but also with @me-no-dev's Async libraries (they don't use WiFiClient). WiFiClient::setLocalPortStart() is still usable with the same effects as before.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I happened to stumble upon another difference between lwip1.4 and lwip2 used by the Arduino Core - source port randomization. After reset, lwip1 always selects a random port when doing outgoing connections, while lwip2 increments port values starting from the same number. Which causes weird issues from time to time.
Some sources:
esp8266/Arduino#632
me-no-dev/ESPAsyncTCP#108
https://blog.voneicken.com/2018/lp-wifi-esp8266-2/
Core's WiFiClient has an option to select source port, but current versions of AsyncTCP does not expose such method. IDK whether this is a issue for async lib or lwip but! current versions of ESP32 and ESP8266_RTOS modify lwip to randomize it by default:
idf: espressif/esp-lwip@4cc89a0
https://github.com/espressif/esp-idf/blob/1aa8e8d38ebe688c391359b5962707c1d135d07e/components/lwip/port/esp32/include/lwipopts.h#L749
esp8266: espressif/ESP8266_RTOS_SDK@a66d130#diff-d4ffea296f65048ca90617f4680cf389
The text was updated successfully, but these errors were encountered: