-
-
Notifications
You must be signed in to change notification settings - Fork 10k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fix leakage when the cacheline is 32-bytes in CBC_MAC_ROTATE_IN_PLACE #18033
Conversation
This will require a CLA to be submitted: |
@mattcaswell do you agree this should be cherry picked to all branches? |
rotated_mac is a 64-byte aligned buffer of size 64 and rotate_offset is secret. Consider a weaker leakage model(CL) where only cacheline base address is leaked, i.e address/32 for 32-byte cacheline(CL32). Previous code used to perform two loads 1. rotated_mac[rotate_offset ^ 32] and 2. rotated_mac[rotate_offset++] which would leak 2q + 1, 2q for 0 <= rotate_offset < 32 and 2q, 2q + 1 for 32 <= rotate_offset < 64 The proposed fix performs load operations which will always leak 2q, 2q + 1 and selects the appropriate value in constant-time.
Yes - although I strongly suspect it won't cherry-pick to 1.1.1 cleanly so we will probably need another PR there. |
@mattcaswell @t8m |
Ah, missed that #18050 wasn't the main PR. Reposting my comment here, to keep the main discussion in one place: Rather than making assumptions about cachelines, we switched to a simpler cacheline-independent O(N log N) algorithm in BoringSSL. It removes the need for the I think that would be a better strategy. It is more straightforwardly constant-time. CVE-2016-0702 is an example where assumptions about cacheline behavior don't quite match how CPUs behave. The O(N log N) approach avoids these assumptions. |
I prefer @davidben's suggestion. Making assumptions about cache lines is less than ideal. |
24 hours has passed since 'approval: done' was set, but as this PR has been updated in that time the label 'approval: ready to merge' is not being automatically set. Please review the updates and set the label manually. |
Merged to master and 3.0 branches. Thank you for your contribution. |
rotated_mac is a 64-byte aligned buffer of size 64 and rotate_offset is secret. Consider a weaker leakage model(CL) where only cacheline base address is leaked, i.e address/32 for 32-byte cacheline(CL32). Previous code used to perform two loads 1. rotated_mac[rotate_offset ^ 32] and 2. rotated_mac[rotate_offset++] which would leak 2q + 1, 2q for 0 <= rotate_offset < 32 and 2q, 2q + 1 for 32 <= rotate_offset < 64 The proposed fix performs load operations which will always leak 2q, 2q + 1 and selects the appropriate value in constant-time. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org> (Merged from #18033) (cherry picked from commit 3b83638)
rotated_mac is a 64-byte aligned buffer of size 64 and rotate_offset is secret. Consider a weaker leakage model(CL) where only cacheline base address is leaked, i.e address/32 for 32-byte cacheline(CL32). Previous code used to perform two loads 1. rotated_mac[rotate_offset ^ 32] and 2. rotated_mac[rotate_offset++] which would leak 2q + 1, 2q for 0 <= rotate_offset < 32 and 2q, 2q + 1 for 32 <= rotate_offset < 64 The proposed fix performs load operations which will always leak 2q, 2q + 1 and selects the appropriate value in constant-time. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org> (Merged from #18033)
rotated_mac is a 64-byte aligned buffer of size 64 and rotate_offset is secret.
Consider a weaker leakage model(CL) where only cacheline base address is leaked,
i.e address/32 for 32-byte cacheline(CL32).
Previous code used to perform two loads
1. rotated_mac[rotate_offset ^ 32] and
2. rotated_mac[rotate_offset++]
which would leak 2q + 1, 2q for 0 <= rotate_offset < 32
and 2q, 2q + 1 for 32 <= rotate_offset < 64
The proposed fix performs load operations which will always leak 2q, 2q + 1 and
selects the appropriate value in constant-time.