*: make bcrypt-cost configurable #9633

jxuan · 2018-04-26T01:10:31Z

This change is to implement the feature request in #9615

heyitsanthony · 2018-04-26T01:43:00Z

@jxuan There's no consensus enforcing the cost; snapshot exchange between members with different costs would break auth. Keep separate buckets for each bcrypt level? Store initial cost and always use that? Presumably not something to plumb through raft. @mitake?

mitake · 2018-04-26T05:34:07Z

@heyitsanthony @jxuan I see, the possible inconsistent state (the diverged bcrypt costs) is similar to #9475 . How about adding a new option for configuring the cost to etcdctl auth enable? The RPC of enabling auth goes through Raft so the inconsistent state can be avoided.

heyitsanthony · 2018-04-26T06:41:09Z

@mitake I checked the bcrypt code and the problem I was worried about isn't an issue. It can check bcrypt.Cost to see if it should recompute/write the new hash if CompareHashAndPassword succeeds (outside the raft path). Different members may have different costs / password hashes, but the cluster wouldn't break. Setting cost via RPC could get bit equivalence across members, but cost upgrade/downgrade for existing passwords seems impractical for that case.

mitake · 2018-04-26T07:43:23Z

@heyitsanthony Ah I see. Then the inconsistent state is acceptable. Thanks for checking!

mitake

The changes are good, thanks! Could you fix the trivial problem I pointed out? Also please fix the commit title for following the standard style e.g. *: make bcrypt cost configurable

mitake · 2018-04-26T07:43:27Z

embed/serve_test.go

@@ -36,3 +37,33 @@ func TestStartEtcdWrongToken(t *testing.T) {
 		t.Fatalf("expected %v, got %v", auth.ErrInvalidAuthOpts, err)
 	}
 }
+
+// TestStartEtcdLargeBcryptCost ensures that StartEtcd with good configs returns with error.


trivial nit: good configs should be bad configs?

jxuan · 2018-04-26T18:33:58Z

@mitake the code is updated based on your feedback. Please check

@heyitsanthony Thank you for checking.

jxuan · 2018-04-26T22:51:07Z

@mitake The CI errors are not related to my code change. I ran those tests manually on my mac and they all pass. Anyway to re-trigger these tests?

gyuho

Also please reformat the git commit title to *: make bcrypt .... And rebase from current master branch. Defer to @mitake @heyitsanthony

gyuho · 2018-04-27T21:12:14Z

auth/store.go

+		plog.Errorf("Invalid bcrypt-cost: %d", bcryptCost)
+		return nil, ErrInvalidAuthOpts
+	}
+	BcryptCost = bcryptCost


auth should pass around bcyptCost values rather than sharing a global BcryptCost (racey as CI fails).

Will make it private. BcryptCost is only referenced in clientv3/main_test.go and clientv3/main_test.go does not trigger anything related to bcrypt.

gyuho · 2018-04-27T21:13:42Z

auth/store.go

 	tokenType, typeSpecificOpts, err := decomposeOpts(tokenOpts)
 	if err != nil {
 		return nil, ErrInvalidAuthOpts
 	}

+	if bcryptCost < bcrypt.MinCost || bcryptCost > bcrypt.MaxCost {
+		plog.Errorf("Invalid bcrypt-cost: %d", bcryptCost)


if lg != nil { lg.Warn( "invalid bcrypt cost", zap.Int("min-cost", bcrypt.MinCost), zap.Int("max-cost", bcrypt.MaxCost), zap.Int("given-cost", bcryptCost), ) }

gyuho · 2018-04-27T21:14:47Z

embed/serve_test.go

@@ -36,3 +37,33 @@ func TestStartEtcdWrongToken(t *testing.T) {
 		t.Fatalf("expected %v, got %v", auth.ErrInvalidAuthOpts, err)
 	}
 }
+
+// TestStartEtcdLargeBcryptCost ensures that StartEtcd with invalid large bcrypt-cost returns with error.


Tests in embed are no needed. Just unit tests around NewTokenProvider should be enough.

gyuho · 2018-04-27T21:15:51Z

etcdmain/config.go

@@ -237,6 +237,7 @@ func newConfig() *config {

 	// auth
 	fs.StringVar(&cfg.ec.AuthToken, "auth-token", cfg.ec.AuthToken, "Specify auth token specific options.")
+	fs.UintVar(&cfg.ec.BcryptCost, "bcrypt-cost", cfg.ec.BcryptCost, "Bcrypt algorithm cost / strength for hashing auth passwords")


"Specify bcrypt algorithm cost factor for auth password hashing."

gyuho · 2018-04-27T21:16:41Z

etcdmain/help.go

@@ -148,6 +148,8 @@ Security:
 Auth:
  --auth-token 'simple'
    Specify a v3 authentication token type and its options ('simple' or 'jwt').
+  --bcrypt-cost 10


+ fmt.Sprintf("%d", bcrypt.DefaultCost) +?

gyuho · 2018-04-27T21:17:06Z

tests/e2e/cluster_test.go

@@ -123,6 +123,7 @@ type etcdProcessClusterConfig struct {
 	noStrictReconfig    bool
 	initialCorruptCheck bool
 	authTokenOpts       string
+	bcryptCost          int


Please remove changes from tests/e2e.

gyuho · 2018-04-27T21:17:24Z

integration/cluster.go

@@ -605,6 +606,7 @@ func mustNewMember(t *testing.T, mcfg memberConfig) *member {
 		m.MaxRequestBytes = embed.DefaultMaxRequestBytes
 	}
 	m.AuthToken = "simple" // for the purpose of integration testing, simple token is enough
+	m.BcryptCost = uint(bcrypt.MinCost)


Not needed when we have unit tests inside auth package.

jxuan

Thanks for the feedback. Fixed in new version.

jxuan · 2018-04-30T18:41:51Z

auth/store.go

+		plog.Errorf("Invalid bcrypt-cost: %d", bcryptCost)
+		return nil, ErrInvalidAuthOpts
+	}
+	BcryptCost = bcryptCost


Will make it private. BcryptCost is only referenced in clientv3/main_test.go and clientv3/main_test.go does not trigger anything related to bcrypt.

jxuan · 2018-04-30T20:50:59Z

@mitake @gyuho Code updated based on your feedback. Please review again. Thanks

gyuho · 2018-04-30T21:27:19Z

@jxuan Can you squash commits? Or separate per package levels?

jxuan · 2018-04-30T21:38:10Z

Sure. Once all tests pass, I will squash everything into one commit.

Signed-off-by: Gyuho Lee <gyuhox@gmail.com>

*: highlight "--log-package-levels" deprecation in v3.5

Documentation: binding listeners must be IP.

…t-cost-config

jxuan · 2018-05-02T19:48:32Z

@mitake @gyuho Commit d923b1d is a squashed version of everything. Can you please review it?

gyuho · 2018-05-02T20:03:11Z

@jxuan gofmt -w *.go should work?

jxuan · 2018-05-02T20:08:41Z

To squash commits, do you recommend starting a new branch?

gyuho · 2018-05-02T20:14:28Z

@jxuan You can just cherry-pick commits to your local branch and force push. Or just keep rebasing from latest master branch, and force push to your local branch. We don't include merge commits.

jxuan · 2018-05-02T23:15:37Z

Right now it only fails on the title format check

'commit_title' started at Wed May 2 21:50:52 UTC 2018
e1d05d0 Fix format...
Expected commit title format '{", "}: '

Can you start to review to code and see if there is anything else that you would like to change? In the mean time, I am working on a squashed version for submission. Thanks,

jxuan · 2018-05-03T18:58:21Z

I created a squashed version in #9687

gyuho · 2018-05-03T19:39:32Z

@jxuan Thanks. Let's close this in favor of #9687.

Make bcrypt-cost configurable

334263c

mitake suggested changes Apr 26, 2018

View reviewed changes

*: Update based on CR comments

ea3b601

jxuan changed the title ~~Make bcrypt-cost configurable~~ *: Make bcrypt-cost configurable Apr 26, 2018

jxuan changed the title *: Make bcrypt-cost configurable *: make bcrypt-cost configurable Apr 26, 2018

*: add missing line in config.go for bcrypt-cost

d18422e

Remove comments to trigger rebuild

9005333

gyuho suggested changes Apr 27, 2018

View reviewed changes

Jiang Xuan added 2 commits April 30, 2018 12:13

Move testcases and address comments

d70369e

Merge branch 'master' into bcrypt-cost-config

e159331

jxuan commented Apr 30, 2018

View reviewed changes

Jiang Xuan added 2 commits April 30, 2018 14:13

*: remove duplicates in Documentation/op-guide/configuration.md

94dd681

merge

3f23c99

Jiang Xuan and others added 8 commits April 30, 2018 20:49

*: move bcryptCost into authStore

596e9c0

Fix help message and trigger rebuild

5b51eb9

CHANGELOG-3.5: deprecate "--log-package-levels" in v3.5

21fdde5

Signed-off-by: Gyuho Lee <gyuhox@gmail.com>

Documentation/upgrades: highlight "--log-package-levels"

9e7a3e5

Signed-off-by: Gyuho Lee <gyuhox@gmail.com>

etcd.conf.yml.sample: remove "--log-package-levels"

d80ef1f

Signed-off-by: Gyuho Lee <gyuhox@gmail.com>

Documentation: binding listeners must be IP.

2afd827

*: update bcrypt cost for integration tests

5e06955

Merge branch 'master' into bcrypt-cost-config

9e1b58f

gyuho and others added 8 commits May 2, 2018 09:39

Merge pull request etcd-io#9676 from gyuho/logger

53f51b2

*: highlight "--log-package-levels" deprecation in v3.5

Merge pull request etcd-io#9638 from hexfusion/df

859172e

Documentation: binding listeners must be IP.

*: make bcrypt-cost configurable (squashed)

8893c66

Merge branch 'bcrypt-cost-config' of github.com:jxuan/etcd into bcryp…

ad1be4e

…t-cost-config

Fix format

e1d05d0

*: make bcrypt-cost configurable (squashed)

d923b1d

Merge branch 'bcrypt-cost-config' of github.com:jxuan/etcd into bcryp…

c16a8d9

…t-cost-config

*: make bcrypt-cost configurable (squashed)

afcf163

*: use proper constant for help message

7328865

jxuan mentioned this pull request May 3, 2018

*: make bcrypt-cost configurable #9687

Merged

gyuho closed this May 3, 2018

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

*: make bcrypt-cost configurable #9633

*: make bcrypt-cost configurable #9633

jxuan commented Apr 26, 2018

heyitsanthony commented Apr 26, 2018

mitake commented Apr 26, 2018

heyitsanthony commented Apr 26, 2018 •

edited

Loading

mitake commented Apr 26, 2018

mitake left a comment

mitake Apr 26, 2018

jxuan Apr 26, 2018

jxuan commented Apr 26, 2018

jxuan commented Apr 26, 2018

gyuho left a comment

gyuho Apr 27, 2018

jxuan Apr 30, 2018

gyuho Apr 27, 2018

gyuho Apr 27, 2018

gyuho Apr 27, 2018

gyuho Apr 27, 2018

gyuho Apr 27, 2018

gyuho Apr 27, 2018

jxuan left a comment

jxuan Apr 30, 2018

jxuan commented Apr 30, 2018

gyuho commented Apr 30, 2018

jxuan commented Apr 30, 2018

jxuan commented May 2, 2018

gyuho commented May 2, 2018

jxuan commented May 2, 2018

gyuho commented May 2, 2018

jxuan commented May 2, 2018

jxuan commented May 3, 2018 •

edited

Loading

gyuho commented May 3, 2018

*: make bcrypt-cost configurable #9633

*: make bcrypt-cost configurable #9633

Conversation

jxuan commented Apr 26, 2018

heyitsanthony commented Apr 26, 2018

mitake commented Apr 26, 2018

heyitsanthony commented Apr 26, 2018 • edited Loading

mitake commented Apr 26, 2018

mitake left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jxuan commented Apr 26, 2018

jxuan commented Apr 26, 2018

gyuho left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jxuan left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jxuan commented Apr 30, 2018

gyuho commented Apr 30, 2018

jxuan commented Apr 30, 2018

jxuan commented May 2, 2018

gyuho commented May 2, 2018

jxuan commented May 2, 2018

gyuho commented May 2, 2018

jxuan commented May 2, 2018

jxuan commented May 3, 2018 • edited Loading

gyuho commented May 3, 2018

heyitsanthony commented Apr 26, 2018 •

edited

Loading

jxuan commented May 3, 2018 •

edited

Loading