x32 platform segfaults on `algparse -ta`

[dkg]

the debian x32 buildd shows this failure:

debian/tests/cryptocheck: line 17: 16108 Segmentation fault      $algparse -ta
algorithm test failed
make[1]: *** [debian/rules:56: override_dh_auto_test] Error 1

that doesn't happen on any other platform that i can see.

[letoams]

can you add -v ? It will hopefully clarify what we were doing when things blew up

[dkg]

you want -v or -d?

[letoams]

On Tue, 12 Jan 2021, dkg wrote:

you want -v or -d?

you can do both.

[dkg]

whoops, sorry, i uploaded 4.1-2 with just -v -- should be a result soon on the x32 buildds for you to take a look at. let me know if you want -d as well, i can do another upload.

[cagney]

algparse: testing AES_GCM_16: algparse: empty string algparse: one block debian/tests/cryptocheck: line 17: 18432 Segmentation fault $algparse -v -ta

[dkg]

i've added --debug (see #398) to those tests for 4.1-3 -- it should show up soon on the debian buildd network.

[dkg]

algparse:   one block
| decode_to_chunk: raw_key: input "0xe98b72a9881a84ca6b76e0f43e68647a"
| decode_to_chunk: output: 
|   e9 8b 72 a9  88 1a 84 ca  6b 76 e0 f4  3e 68 64 7a
| CONCATENATE_DATA_AND_BASE:
|     target: EXTRACT_KEY_FROM_KEY
|     base: base-key@0x58453bd0 (16-bytes, AES_KEY_GEN)
|     params: 8-bytes@0xfff11fd4
|     result: newref symkey-key@0x58456010 (32-bytes, EXTRACT_KEY_FROM_KEY)(in merge_symkey_bytes() at crypt_symkey.c:224)
|     key-offset: 0, key-size: 16
| EXTRACT_KEY_FROM_KEY:
|     target: AES_GCM
|     flags: ENCRYPT+DECRYPT
|     key_size: 16-bytes
|     base: base-key@0x58456010 (32-bytes, EXTRACT_KEY_FROM_KEY)
|     operation: FLAGS_ONLY
|     params: 4-bytes@0xfff11fd0
|     result: newref symkey-key@0x58454d80 (16-bytes, AES_GCM)(in decode_to_key() at test_buffer.c:121)
| symkey: delref tmp-key@0x58456010
| decode_to_chunk: salted IV: input "0x8b23299fde174053f3d652ba"
| decode_to_chunk: output: 
|   8b 23 29 9f  de 17 40 53  f3 d6 52 ba
| decode_to_chunk: AAD: input ""
| decode_to_chunk: output: 
| 
| decode_to_chunk: plaintext: input "0x28286a321293253c3e0aa2704a278032"
| decode_to_chunk: output: 
|   28 28 6a 32  12 93 25 3c  3e 0a a2 70  4a 27 80 32
| decode_to_chunk: ciphertext: input "0x5a3c1cf1985dbb8bed818036fdd5ab42"
| decode_to_chunk: output: 
|   5a 3c 1c f1  98 5d bb 8b  ed 81 80 36  fd d5 ab 42
| decode_to_chunk: tag: input "0x23c7ab0f952b7091cd324835043b5eb5"
| decode_to_chunk: output: 
|   23 c7 ab 0f  95 2b 70 91  cd 32 48 35  04 3b 5e b5
| test_gcm_vector: decrypt: aad-size=0 salt-size=4 wire-IV-size=8 text-size=16 tag-size=16
| test_gcm_vector: text+tag on call
|   5a 3c 1c f1  98 5d bb 8b  ed 81 80 36  fd d5 ab 42
|   23 c7 ab 0f  95 2b 70 91  cd 32 48 35  04 3b 5e b5
debian/tests/cryptocheck: line 19: 22163 Segmentation fault      $algparse -v --debug -ta

[cagney]

Which version of NSS?

[dkg]

the most recent x32 buildd attempt shows libnss3-dev version 2:3.56-1

(that's upstream version 3.56, if we ignore the debian epoch prefix and revision suffix)

[dkg]

I do note that there are weird build and test failures for recent versions of nss on the debian x32 buildds as well, so maybe this is an x32 NSS problem, and not a libreswan problem? not sure how to pursue that line of inquiry further, since i don't think i have direct access to any x32 machines.

[cagney]

https://bugzilla.mozilla.org/show_bug.cgi?id=1062903 scary

[cagney]

The problem isn't building 32-bit (I just built on a 32-bit debian ARM (aka pi) and algparse ran fine. I'd assume the problem is in NSS and the lack of love for the x32 frankenabi.

[cagney]

Can re-open if latest code shows same problems.