tls: Use SHA1 for sessionIdContext in FIPS mode #3755

stefanmb · 2015-11-10T23:55:16Z

By default, a call to tls.createServer() without a sessionIdContext will use a “MD5 hash value generated from command-line” as per documentation.

In FIPS mode MD5 is not allowed, however createServer is often called without specifying an explicit sessionIdContext. A significant number of test cases and applications break. The simple solution is to to use an allowed hash function. I have chosen SHA1 and truncated the output to 128 bits (which is the hardcoded length required by OpenSSL’s SSL_MAX_SID_CTX_LENGTH).

Note that I have opted to maintain the use of MD5 in non-FIPS mode, and updated the documentation accordingly.

mscdex · 2015-11-11T00:03:16Z

This should target tls instead of the crypto subsystem.

stefanmb · 2015-11-11T16:47:40Z

@mscdex I updated the commit message accordingly. Thanks.

indutny · 2015-11-11T22:50:46Z

lib/_tls_wrap.js

+  if (process.versions.openssl && process.versions.openssl.endsWith("-fips")) {
+    return crypto.createHash('sha1')
+      .update(defaultText)
+      .digest('hex').substring(0,32); /* SSL_MAX_SID_CTX_LENGTH is 128 bits */


Let's use .slice() here, this is what we try to do everywhere else.

indutny · 2015-11-11T22:51:40Z

I think that in general it may be a good idea to use sha256 on both fips and non-fips node. What do you think?

mhdawson · 2015-11-12T17:34:45Z

The main concern would be interoperability. Would changing to sha256 for non-FIPs mean that there would be problems inter-operating with with earlier Node.js releases ?

stefanmb · 2015-11-12T23:39:19Z

@indutny @mhdawson

I don't particularly like this commit. What's fed into the hash function is the process name and arguments. I maintained MD5 for non-FIPS because I was afraid someone somewhere might be serializing (caching) sessions across process lifetimes.

I've looked a bit more into the OpenSSL caching mechanism, specifically:
https://www.openssl.org/docs/manmaster/ssl/SSL_CTX_set_session_cache_mode.html
https://www.openssl.org/docs/manmaster/ssl/SSL_CTX_sess_set_get_cb.html

The only place where we provide external callbacks for session management is in node_crypto.cc GetSessionCallback and NewSessionCallback. In turn, these store data using CRYPTO_get_ex_data and CRYPTO_set_ex_data which I believe means the data only live as long as the process.

Therefore, I suggest we don't use a hash at all, and instead simply call crypto.randomBytes(16) once in this file and use the resulting value for the lifetime of a given Node process, in both FIPS and non-FIPS mode.

Edit: We still have to clarify why MD5 was mentioned in the docs, is that just internals leaking into the documentation or did it have some meaning to users?

jasnell · 2015-11-14T16:18:02Z

@stefanmb ... just a quick note, I see this PR has a few conflicts now, can you please take a moment to rebase and update. Thank you!

stefanmb · 2015-11-14T17:17:23Z

@jasnell Fixed, thanks!

stefanmb · 2015-11-14T18:00:23Z

@indutny @shigeki
Any thoughts on my comment above regarding the use of crypto.randomBytes instead of a hash?
(1) Is it okay to provide a randomized session ID for the lifespan of each process invocation, instead of deriving the session ID deterministically from the program arguments? In particular, I'm not sure how this affects the cluster module. It seems execArgv is passed to the child workers in cluster.js.
(2) Is it okay to remove the mention of MD5 from the documentation? Quote:

sessionIdContext: A string containing an opaque identifier for session resumption. If requestCert is true, the default is MD5 hash value generated from command-line. Otherwise, the default is not provided.
From documentation for tls.createServer(options[, secureConnectionListener])

If so, I can update this PR. Thank you for your help and advice!

indutny · 2015-11-14T18:03:55Z

@stefanmb the idea, originally was to provide the same session context to clustered processes (if I remember it right)

stefanmb · 2015-11-14T18:12:38Z

@indutny Okay, that's what I started to suspect. It's kind of clunky to do it through argv though. Anyway, perhaps the simplest thing is to move to a truncated SHA1 for both FIPS and non-FIPS. OpenSSL requires 128 bits, so I don't think there's a point in using a bigger hash.

indutny · 2015-11-14T18:14:06Z

Alright, this makes sense to me.

indutny · 2015-11-15T06:23:33Z

lib/_tls_wrap.js

+      .update(defaultText)
+      .digest('hex').slice(0, 32); /* SSL_MAX_SID_CTX_LENGTH is 128 bits */
+  } else {
+    return crypto.createHash('md5')


Hm... Let's use sha1 for both cases, and remove condition?

stefanmb · 2015-11-16T21:10:16Z

@indutny Okay, changed this to do SHA1 in all cases, the change is now pretty trivial.

indutny · 2015-11-16T21:27:33Z

@stefanmb I was thinking about it for a bit now, and I think we should handle it this way. Let's split this PR into two commits:

First with both md5 and sha1 (as you originally did). This commit will be backported as semver-patch
And the one that removes md5 entirely. This commit is going to be a semver-major

The best way to do it is probably to submit a second PR for the second half. @stefanmb does it sound reasonable?

Thank you very much!

FIPS 140-2 disallows use of MD5, which is used to derive the default sessionIdContext for tls.createServer().

FIPS 140-2 disallows use of MD5, which is used to derive the default sessionIdContext for tls.createServer(). PR-URL: #3755 Reviewed-By: Fedor Indutny <fedor@indutny.com>

Notable changes: * buffer: The `noAssert` option for many buffer functions will now silently drop invalid write values rather than crashing (P.S.V.R) nodejs/node#3767. - This makes the behavior match what the docs suggest. * child_process: `child.send()` now properly returns a boolean like the docs suggest (Rich Trott) nodejs/node#3577. * doc: All of the API docs have been re-ordered so as to read in alphabetical order (Tristian Flanagan) nodejs/node#3662. * http_parser: update http-parser to 2.6.0 from 2.5.0 (James M Snell) nodejs/node#3569. - Now supports the following HTTP methods: `LINK`, `UNLINK`, `BIND`, `REBIND`, `UNBIND`. - Also added ACL and IPv6 Zone ID support. * npm: upgrade npm to 3.3.12 from v3.3.6 (Rebecca Turner) nodejs/node#3685. - See the release notes for https://github.com/npm/npm/releases/tag/v3.3.7, https://github.com/npm/npm/releases/tag/v3.3.8, https://github.com/npm/npm/releases/tag/v3.3.9, https://github.com/npm/npm/releases/tag/v3.3.10, https://github.com/npm/npm/releases/tag/v3.3.11, and https://github.com/npm/npm/releases/tag/v3.3.12 for more details. * repl: The REPL no longer crashes if the persistent history file cannot be opened (Evan Lucas) nodejs/node#3630. * tls: The default `sessionIdContext` now uses SHA1 in FIPS mode rather than MD5 (Stefan Budeanu) nodejs/node#3755. * v8: Added some more useful post-mortem data (Fedor Indutny) nodejs/node#3779. PR-URL: nodejs/node#3736

sneak · 2016-04-07T19:57:12Z

This broke my app:

krs_1 | 
krs_1 | > key-recovery-service@0.0.4 test /var/app
krs_1 | > node ./node_modules/mocha/bin/_mocha
krs_1 | 
krs_1 | _tls_wrap.js:22
krs_1 |   if (process.config.variables.openssl_fips) {
krs_1 |                               ^
krs_1 | 
krs_1 | TypeError: Cannot read property 'openssl_fips' of undefined
krs_1 |   at getDefaultSessionIdContext (_tls_wrap.js:22:31)
krs_1 |   at _tls_wrap.js:17:33
krs_1 |   at NativeModule.compile (node.js:954:5)
krs_1 |   at NativeModule.require (node.js:902:18)
krs_1 |   at tls.js:221:21
krs_1 |   at NativeModule.compile (node.js:954:5)
krs_1 |   at NativeModule.require (node.js:902:18)
krs_1 |   at https.js:3:13
krs_1 |   at NativeModule.compile (node.js:954:5)
krs_1 |   at Function.NativeModule.require (node.js:902:18)
krs_1 |   at Function.Module._load (module.js:298:25)
krs_1 |   at Module.require (module.js:366:17)
krs_1 |   at require (module.js:385:17)
krs_1 |   at Object.<anonymous> (/var/app/node_modules/supertest/node_modules/superagent/node_modules/form-data/lib/form_data.js:5:13)
krs_1 |   at Module._compile (module.js:435:26)
krs_1 |   at Object.Module._extensions..js (module.js:442:10)
krs_1 |   at Module.load (/var/app/node_modules/coffee-script/lib/coffee-script/register.js:45:36)
krs_1 |   at Function.Module._load (module.js:313:12)
krs_1 |   at Module.require (module.js:366:17)
krs_1 |   at require (module.js:385:17)
krs_1 |   at Object.<anonymous> (/var/app/node_modules/supertest/node_modules/superagent/lib/node/index.js:8:16)
krs_1 |   at Module._compile (module.js:435:26)
krs_1 |   at Object.Module._extensions..js (module.js:442:10)
krs_1 |   at Module.load (/var/app/node_modules/coffee-script/lib/coffee-script/register.js:45:36)
krs_1 |   at Function.Module._load (module.js:313:12)
krs_1 |   at Module.require (module.js:366:17)
krs_1 |   at require (module.js:385:17)
krs_1 |   at Object.<anonymous> (/var/app/node_modules/supertest/lib/test.js:5:15)
krs_1 |   at Module._compile (module.js:435:26)
krs_1 |   at Object.Module._extensions..js (module.js:442:10)
krs_1 |   at Module.load (/var/app/node_modules/coffee-script/lib/coffee-script/register.js:45:36)
krs_1 |   at Function.Module._load (module.js:313:12)
krs_1 |   at Module.require (module.js:366:17)
krs_1 |   at require (module.js:385:17)
krs_1 |   at Object.<anonymous> (/var/app/node_modules/supertest/index.js:7:12)
krs_1 |   at Module._compile (module.js:435:26)
krs_1 |   at Object.Module._extensions..js (module.js:442:10)
krs_1 |   at Module.load (/var/app/node_modules/coffee-script/lib/coffee-script/register.js:45:36)
krs_1 |   at Function.Module._load (module.js:313:12)
krs_1 |   at Module.require (module.js:366:17)
krs_1 |   at require (module.js:385:17)
krs_1 |   at Object.<anonymous> (/var/app/node_modules/supertest-as-promised/index.js:3:17)
krs_1 |   at Module._compile (module.js:435:26)
krs_1 |   at Object.Module._extensions..js (module.js:442:10)
krs_1 |   at Module.load (/var/app/node_modules/coffee-script/lib/coffee-script/register.js:45:36)
krs_1 |   at Function.Module._load (module.js:313:12)
krs_1 |   at Module.require (module.js:366:17)
krs_1 |   at require (module.js:385:17)
krs_1 |   at Object.<anonymous> (/var/app/test/app.coffee:3:11)
krs_1 |   at Object.<anonymous> (/var/app/test/app.coffee:2:1)
krs_1 |   at Module._compile (module.js:435:26)
krs_1 |   at Object.loadFile (/var/app/node_modules/coffee-script/lib/coffee-script/register.js:16:19)
krs_1 |   at Module.load (/var/app/node_modules/coffee-script/lib/coffee-script/register.js:45:36)
krs_1 |   at Function.Module._load (module.js:313:12)
krs_1 |   at Module.require (module.js:366:17)
krs_1 |   at require (module.js:385:17)
krs_1 |   at /var/app/node_modules/mocha/lib/mocha.js:219:27
krs_1 |   at Array.forEach (native)
krs_1 |   at Mocha.loadFiles (/var/app/node_modules/mocha/lib/mocha.js:216:14)
krs_1 |   at Mocha.run (/var/app/node_modules/mocha/lib/mocha.js:468:10)
krs_1 |   at Object.<anonymous> (/var/app/node_modules/mocha/bin/_mocha:403:18)
krs_1 |   at Module._compile (module.js:435:26)
krs_1 |   at Object.Module._extensions..js (module.js:442:10)
krs_1 |   at Module.load (module.js:356:32)
krs_1 |   at Function.Module._load (module.js:313:12)
krs_1 |   at Function.Module.runMain (module.js:467:10)
krs_1 |   at startup (node.js:136:18)
krs_1 |   at node.js:963:3
krs_1 | 
krs_1 | npm ERR! Test failed.  See above for more details.

It works on 4.2.3 but fails on 4.2.4. It still fails on 4.4.2 (current) and latest 5.x.

Isn't the point of LTS to not add features and not break compatibility?

MylesBorins · 2016-04-07T20:08:39Z

@sneak are you 100% there isn't some monkey patching of the process object happening in your app?

The error you are getting is suggesting that process.config.variables does not exist at all.

here is a screen shot of me sanity checking this in v4

Here is v5

sneak · 2016-04-07T20:12:29Z

I'm not; I have not read every line of the testing framework involved, I'm just letting you know that a test suite that passed on 4.2.3 now no longer runs on 4.2.4.

I thought the point of LTS was to not break stuff and cost people time; why not keep this FIPS stuff to 5.x? I don't know what ill-advised stuff this app or its deps was doing, but I do now know that it doesn't work anymore and that this was the change that broke it.

sneak · 2016-04-07T20:13:41Z

PS: When posting text snippets, you don't need to push bitmaps, simply wrapping them in triple backticks is sufficient.

MylesBorins · 2016-04-07T20:18:26Z

/cc @nodejs/crypto @nodejs/lts

fwiw we extensively smoke test LTS, we do occasionally hit edges like this though. Thank you for reporting this, we'll dig in and see what we can find.

sneak · 2016-04-07T20:20:46Z

FWIW:

  "devDependencies": {
    "chai": "3.5.0",
    "istanbul": "0.4.3",
    "mocha": "2.4.5",
    "supertest": "1.2.0",
    "supertest-as-promised": "3.1.0",
    "sqlite3": "*"
  },
  "dependencies": {
    "argparse": "1.0.2",
    "bitcoinjs-lib": "1.5.6",
    "body-parser": "1.13.2",
    "coffee-script": "1.10.0",
    "express": "4.13.1",
    "js-beautify": "1.5.10",
    "lodash": "^4.0.0",
    "moment": "2.10.6",
    "morgan": "1.6.1",
    "node-jsrender": "1.0.6",
    "nodemailer": "1.4.0",
    "nodemailer-smtp-transport": "1.0.3",
    "pg": "4.4.2",
    "pg-hstore": "2.3.2",
    "prompt-sync": "1.0.0",
    "q": "1.4.1",
    "sequelize": "3.17.0",
    "sequelize-cli": "1.9.1",
    "validator": "4.0.2",
    "winston": "2.2.0"
  }

I bumped the versions of the stuff mentioned in the stack trace to latest, but still no love.

jasnell · 2016-04-07T21:46:22Z

Yeah, the goal of LTS is to keep things as stable as absolutely possible but regressions do creep in from time to time. To help keep things easier to maintain we'll go ahead and pull certain changes back that appear innocuous but every once in a while we'll hit an edge case that needs to be looked at. For this, a quick existence check on process.config.variables and process.config (just to be safe) should be sufficient. Interested in opening a PR? ;-) Once it's reviewed we can get it landed in master, v5 and v4 so it goes out with the next release runs.

sneak · 2016-04-08T00:48:54Z

There you go, a PR against 4.x. I haven't tried getting 5.x to work yet, I can do that next.

sneak · 2016-04-08T01:06:04Z

It looks like the change is exactly the same in 5.x, but master has a different file without any FIPS stuff in it. (I haven't tested the breaking app against master, but can if you need me to.)

mscdex added the tls Issues and PRs related to the tls subsystem. label Nov 11, 2015

stefanmb mentioned this pull request Nov 11, 2015

crypto: Resolve FIPS test failures and API issues #3760

Closed

stefanmb force-pushed the fips-cs4-tls-wrap branch from 2687e79 to 7a0d7d0 Compare November 11, 2015 16:47

stefanmb changed the title ~~crypto: Use SHA1 for sessionIdContext in FIPS mode~~ tls: Use SHA1 for sessionIdContext in FIPS mode Nov 11, 2015

mhdawson added the crypto Issues and PRs related to the crypto subsystem. label Nov 11, 2015

indutny reviewed Nov 11, 2015
View reviewed changes

jasnell added the lts-watch-v4.x label Nov 12, 2015

stefanmb force-pushed the fips-cs4-tls-wrap branch from 7a0d7d0 to d3140c4 Compare November 13, 2015 00:13

stefanmb force-pushed the fips-cs4-tls-wrap branch from d3140c4 to 2b01c3c Compare November 14, 2015 17:17

indutny reviewed Nov 15, 2015
View reviewed changes

stefanmb force-pushed the fips-cs4-tls-wrap branch 2 times, most recently from 042f645 to ab64b8b Compare November 16, 2015 21:09

stefanmb changed the title ~~tls: Use SHA1 for sessionIdContext in FIPS mode~~ tls: Use SHA1 for sessionIdContext Nov 16, 2015

stefanmb changed the title ~~tls: Use SHA1 for sessionIdContext~~ tls: Use SHA1 for sessionIdContext in FIPS mode Nov 16, 2015

tls: Use SHA1 for sessionIdContext in FIPS mode

532857b

FIPS 140-2 disallows use of MD5, which is used to derive the default sessionIdContext for tls.createServer().

stefanmb force-pushed the fips-cs4-tls-wrap branch from ab64b8b to 532857b Compare November 16, 2015 21:40

stefanmb mentioned this pull request Nov 16, 2015

tls: Use SHA1 for sessionIdContext #3866

Closed

MylesBorins added lts-landed-on-v4.x and removed land-on-v4.x labels Nov 30, 2015

MylesBorins added land-on-v4.x and removed lts-landed-on-v4.x labels Dec 17, 2015

jasnell mentioned this pull request Dec 17, 2015

Release 4.2.4 Planning #4321

Closed

sneak mentioned this pull request Apr 8, 2016

fix FIPS regression in 4.x #6110

Closed

This was referenced Apr 8, 2016

Defensive check for process.config.variables (v4-staging) #6114

Closed

Defensive fix for process.config.variables regression (v5) #6115

Closed

msimerson mentioned this pull request Dec 13, 2016

set tls.sessionIdContext property haraka/Haraka#1740

Merged

2 tasks

Clemens-git76 mentioned this pull request Apr 16, 2024

[Snyk] Upgrade d3 from 5.16.0 to 7.9.0 Clemens-git76/node#5

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

tls: Use SHA1 for sessionIdContext in FIPS mode #3755

tls: Use SHA1 for sessionIdContext in FIPS mode #3755

stefanmb commented Nov 10, 2015

mscdex commented Nov 11, 2015

stefanmb commented Nov 11, 2015

indutny Nov 11, 2015

indutny commented Nov 11, 2015

mhdawson commented Nov 12, 2015

stefanmb commented Nov 12, 2015

jasnell commented Nov 14, 2015

stefanmb commented Nov 14, 2015

stefanmb commented Nov 14, 2015

indutny commented Nov 14, 2015

stefanmb commented Nov 14, 2015

indutny commented Nov 14, 2015

indutny Nov 15, 2015

stefanmb commented Nov 16, 2015

indutny commented Nov 16, 2015

sneak commented Apr 7, 2016

MylesBorins commented Apr 7, 2016

sneak commented Apr 7, 2016

sneak commented Apr 7, 2016

MylesBorins commented Apr 7, 2016

sneak commented Apr 7, 2016

jasnell commented Apr 7, 2016

sneak commented Apr 8, 2016

sneak commented Apr 8, 2016

tls: Use SHA1 for sessionIdContext in FIPS mode #3755

tls: Use SHA1 for sessionIdContext in FIPS mode #3755

Conversation

stefanmb commented Nov 10, 2015

mscdex commented Nov 11, 2015

stefanmb commented Nov 11, 2015

indutny Nov 11, 2015

Choose a reason for hiding this comment

indutny commented Nov 11, 2015

mhdawson commented Nov 12, 2015

stefanmb commented Nov 12, 2015

jasnell commented Nov 14, 2015

stefanmb commented Nov 14, 2015

stefanmb commented Nov 14, 2015

indutny commented Nov 14, 2015

stefanmb commented Nov 14, 2015

indutny commented Nov 14, 2015

indutny Nov 15, 2015

Choose a reason for hiding this comment

stefanmb commented Nov 16, 2015

indutny commented Nov 16, 2015

sneak commented Apr 7, 2016

MylesBorins commented Apr 7, 2016

sneak commented Apr 7, 2016

sneak commented Apr 7, 2016

MylesBorins commented Apr 7, 2016

sneak commented Apr 7, 2016

jasnell commented Apr 7, 2016

sneak commented Apr 8, 2016

sneak commented Apr 8, 2016