sock_dns: fix out-of-bound errors [backport 2018.10]

[miri64]

Backport of #10740

Contribution description

Fixes #10739

Testing procedure

I compiled tests/gnrc_sock_dns with the following patch:

diff --git a/tests/gnrc_sock_dns/Makefile b/tests/gnrc_sock_dns/Makefile
index 4ef2c75..0b3824e 100644
--- a/tests/gnrc_sock_dns/Makefile
+++ b/tests/gnrc_sock_dns/Makefile
@@ -11,7 +11,11 @@ USEMODULE += sock_dns
 USEMODULE += gnrc_sock_udp
 USEMODULE += gnrc_ipv6_default
 USEMODULE += gnrc_ipv6_nib_dns
-USEMODULE += gnrc_netdev_default
+USEMODULE += ethos
+
+# ethos baudrate can be configured from make command
+ETHOS_BAUDRATE ?= 115200
+CFLAGS += -DETHOS_BAUDRATE=$(ETHOS_BAUDRATE) -DUSE_ETHOS_FOR_STDIO
 USEMODULE += auto_init_gnrc_netif
 
 USEMODULE += shell_commands
diff --git a/tests/gnrc_sock_dns/main.c b/tests/gnrc_sock_dns/main.c
index 52700fe..b8c2ad5 100644
--- a/tests/gnrc_sock_dns/main.c
+++ b/tests/gnrc_sock_dns/main.c
@@ -40,7 +40,7 @@ int main(void)
     uint8_t addr[16] = {0};
 
     puts("waiting for router advertisement...");
-    xtimer_usleep(1U*1000000);
+    xtimer_usleep(5U*1000000);
 
     /* print network addresses */
     puts("Configured network interfaces:");

flashed a samr21-xpro (alternatively pba-d-01-kw2x as described in #10739 is also possible), and connected an ethos terminal to the node

sudo ./dist/tools/ethos/start_network.sh /dev/ttyACM0 tap0 affe:abe::/48

I reset the node to get its link-local address for later.

Then I started a RADVD with the following config:

interface tap0 {
        AdvSendAdvert on;
        MinRtrAdvInterval 3;
        MaxRtrAdvInterval 10;
        prefix affe:abe::/64 {
                AdvOnLink off;
                AdvAutonomous on;
                AdvRouterAddr on;
        };
        RDNSS <a global address on your host> {
            AdvRDNSSLifetime 60;
        };
};

and reconfigured the correct route to the host (the preconfigured route by start_network.sh does not work, since uhcpc is not available on the node):

sudo ip route del affe:abe::/64
sudo ip route add affe:abe::/64 via "<link-local address of node>" dev tap0

Then I start the exploit script described in #10739 (or provided by https://github.com/beduino-project/exploit-riot-dns) and reset the node again to start the test. Without this PR the node will crash (note that the exploit described in #10739 does not work but only crashes the node since due to the usage of ethos the binary is different), withit it will just report error resolving example.org.

I also tested expected operation as described in the application's README.

Issues/PRs references

Fixes #10739.

[maribu]

The code is identical to #10740, so only testing should be required.

[kaspar030]

tests/gnrc_sock_dns still working. ACK.

[kaspar030]

ACK.

[kaspar030]

We have a problem, though: the release branch doesn't build with our current docker image, thus CI doesn't pass.

[miri64]

Dafuq?

[kaspar030]

Dafuq?

Well, that's how interpret the unrelated Murdock errors.

[miri64]

#10759 was merged. Please rebase.

[miri64]

Rebased to current 2018.10-branch.

[jia200x]

@kaspar030 Re-ACK?

[miri64]

As suspected in #10765 Travis still isn't able to build.

[miri64]

@jia200x could you review?

[jia200x]

yes, sure

[miri64]

Should we close this, now that 2019.01 is imminent?

[jia200x]

Should we close this, now that 2019.01 is imminent?

If we can ignore Travis output, I can test it now and merge

[miri64]

If we can ignore Travis output, I can test it now and merge

We basically have to, since we would have to backport #10765 for that. Since Travis is performing the same checks as Murdock does, this seems unnecessary to me.

[jia200x]

I was not able to reproduce the crash on my machine, but I think it's because of my local setup. The fix is evident and the patch is the same as in #10740. So, let's merge it

[jia200x]

ACK

[miri64]

So now we can publish 2018.10.1 just before 2019.01 ^^