nsapi_dns: Provide 2 IPv6-hosted default servers #3166
Conversation
|
Looks good to me. |
|
Looks good to me! This may be a separate pr, but do you think we should add the logic to try the ipv6 servers first if the requested version is ipv6? |
|
All DNS server are capable of answering AAAA resource types (should be) so it does not matter if we use IPv4 or IPv6 to contact the server. |
|
I was thinking of the case of Nanostack, where the ipv4 requests would always fail. Requesting an AAAA record implies the stack at least supports ipv6. I don't know if it's a worthwhile optimization though. |
Not sure if that's really worthwhile. If we're really dual-stack, the IPv4 ones should work fine anyway. If we're not dual stack, I expect sendto() to fail immediately if asked to send to the wrong address type. I guess there could be cases when IPv4 disappears into a black hole (or we get ICMP "no route" which we ignore) and we time out. So may need to address it then. Anyway, in a dual-stack scenario we would ideally be doing A and AAAA queries simultaneously to the same DNS server ("Happy Eyeballs"). So the "send to server type matching query" logic is a bit specific to our limited "only attempt one type" setup. |
Replace Comodo and OpenDNS IPv4 servers with Google and DNS.WATCH IPv6 servers, so IPv6-only devices (eg 6LoWPAN) have a default. 3 IPv4 resolvers should be plenty - existing code doesn't remember which one last worked, so if early list entries were unreachable performance would be consistently bad anyway. Replacing two entries avoids increasing image size and RAM consumption. On an IPv6-only or IPv4-only system, the sendto() for the wrong type of address should fail immediately - change loop to move on to the next server for any sendto() error.
kjbracey
force-pushed
the
ipv6_dns_server
branch
from
5c2e519
to
aa583bd
Compare
|
/morph test-nightly |
Result: FAILUREYour command has finished executing! Here's what you wrote!
Outputmbed Build Number: 984 Test failed! |
|
/morph test-nightly |
|
I reran the tcp-echo-hello-world test locally on iar and it passed fine. This specific test relies on developer.mbed.org being available, and the site was having issues earlier. Interesting that the downtime failed a random test. This pr should be good to go 👍 |
|
@geky thanks for rerunning, should be fine in that case. |
Release mbed OS 5.2.2 and mbed lib v129 Known issues in this release There is currently a DNS resolution failure in Thread mode with this release. This causes a failure in the mbed-os-example-client. This will be fixed in a subsequent release. This can be worked around by reverting to mbed-os-5.2.0 Ports for Upcoming Targets 3011: Add u-blox Sara-N target. ARMmbed#3011 3099: MAX32625 ARMmbed#3099 3151: Add support for FRDM-K82F ARMmbed#3151 3177: New mcu k22512 fixing pr 3136 ARMmbed#3177 Fixes and Changes 2990: [tools] Parallel building of tests ARMmbed#2990 3008: NUCLEO_F072RB: Fix wrong timer channel number on pwm PB_5 pin ARMmbed#3008 3013: STM32xx - Change how the ADC internal pins are checked before pinmap_ ARMmbed#3013 3023: digital_loop tests update for STM32 ARMmbed#3023 3041: [nRF5] - added implementation of API of serial port flow control configuration. ARMmbed#3041 3092: [tools + tests] Adding parallelized build option for iar and uvision exporters ARMmbed#3092 3084: [nrf5] fix in Digital I/O : a gpioe pin was uninitialized badly ARMmbed#3084 3009: TRNG enabled. TRNG APIs implemented. REV A/B/C/D flags removed. Warnings removed ARMmbed#3009 3139: Handle [NOT_SUPPORTED] exception in make.py ARMmbed#3139 3074: Target stm init gcc alignement ARMmbed#3074 3140: [tests] Replacing getchar with RawSerial getc in greentea-client ARMmbed#3140 3158: Added support for 6lowpan PAN ID filter to mbed mesh api configuration ARMmbed#3158 2988: Update of can_api.c fixing ARMmbed#2987 ARMmbed#2988 3175: Updating IAR definition for the NCS36510 for IAR EW v7.8 ARMmbed#3175 3170: [tests] Preventing test from printing before Greentea __sync ARMmbed#3170 3169: [Update of ARMmbed#3014] Usb updates ARMmbed#3169 3143: CFStore fix needed for the Cloud Client ARMmbed#3143 3135: lwip - Fix memory leak in k64f cyclic-buffer overflow ARMmbed#3135 3048: Make update.py test compile examples prior to updating mbed-os version. ARMmbed#3048 3162: lwip/nsapi - Clean up warnings in network code ARMmbed#3162 3161: nsapi - Add better heuristic for the default record of DNS queries ARMmbed#3161 3173: [Exporters] Add a device_name to microbit entry in targets.json ARMmbed#3173 3072: i2c_loop tests update for STM32 ARMmbed#3072 2958: Allowing mbed_app.json files to be discovered for tests. ARMmbed#2958 2969: [nRF52] - switch irq priorities of driver handlers to the lowest level ARMmbed#2969 3078: lwip: Allow several configuration macros to be set externally (bis) ARMmbed#3078 3165: Add address type checks to NanostackInterface ARMmbed#3165 3166: nsapi_dns: Provide 2 IPv6-hosted default servers ARMmbed#3166 3171: [tools] Fixing project.py -S printing problem ARMmbed#3171 3172: [Exporters] New export-build tests ARMmbed#3172 3184: ARMmbed#3183 Compiler warning in trng_api.c with K64F ARMmbed#3184 3185: Update tests to fix build failures. Also make the code similar to oth ARMmbed#3185 3104: [NuMaker] Support CAN and fix PWM CLK error ARMmbed#3104 3182: Exporter documentation ARMmbed#3182 3186: MultiTech mDot - add back SPI3 pins ARMmbed#3186 3187: [Export-Make] Use internal class variable for resolving templates in makefiles ARMmbed#3187 3195: [Exporters - Make-based] Quote the shell call in mkdir and rmdir ARMmbed#3195 3204: [Export build-test] Directory traversal error ARMmbed#3204 3189: [Exporters - Make-based] Force make exporter to search PATH for compilers ARMmbed#3189 3200: Using Popen for uVision and unifying the structure of the build function ARMmbed#3200 3075: nsapi - Add standardized return types for size and errors ARMmbed#3075 3221: u-blox odin w2 drivers update ARMmbed#3221
Replace Comodo and OpenDNS IPv4 servers with Google and DNS.WATCH IPv6
servers, so IPv6-only devices (eg 6LoWPAN) have a default.
3 IPv4 resolvers should be plenty - existing code doesn't remember which
one last worked, so if early list entries were unreachable performance
would be consistently bad anyway. Replacing two entries avoids
increasing image size and RAM consumption.
For this to fully work with Nanostack, #3146 needs to be fixed - Nanostack is currently attempting to send to the IPv4 resolvers. (Other socket providers like the Wifi module may need equivalent checks to not interpret an IPv6 address as IPv4.)
@SeppoTakalo, @geky, @c1728p9, @mikaleppanen