Repository navigation
Conversation
This is a security bug, since it makes PBKDF2 faster for attackers than defenders. So it should be fixed from PHP 8.2 on. Fixes php#9604
0e39d92 to
690e0ac
Compare
LamentXU123
left a comment
There was a problem hiding this comment.
Have you benchmarked the difference between it? Generally we treat performance improvements as a feature to be landed on master only and only vulnerbility fixes qualifies for 8.2. @php/release-managers-82 ?
| due to partial address comparison). (CVE-2026-91768) (Alexandre Daubois) | ||
|
|
||
| - Hash: | ||
| . Improved performance of hash_pbkdf2. (Sjoerd Langkemper) |
There was a problem hiding this comment.
Feel free to add a UPGRADING entry.
|
|
|
I measured this on PHP-8.2 03e9593 (5db5b7a merges cleanly) with GCC 13.3, GCC 14.2, clang 18.1 and clang 20.1 on an i7-13700H, built with the flags of the official Docker image; PHP-8.2 does not build with Change in CPU cycles per call with the PR, relative to PHP-8.2 (negative = fewer cycles, faster); median of five alternating runs, a fixed 28-byte password and 16-byte salt (
Correctness: I compared the output with PHP-8.2 for all 44 algorithms of Code: one suggestion. ZEND_SECURE_ZERO(inner_context, ops->context_size);
ZEND_SECURE_ZERO(outer_context, ops->context_size);
efree(inner_context);
efree(outer_context);The About security: the speed-up does not make passwords stronger by itself. With the same iteration count an attacker's cost per guess is the same as before; the gain becomes strength only where an application raises the count. It also does nothing for existing hashes: PBKDF2 uses the password in every iteration, so a stored hash cannot be extended to more iterations without the original password, only recomputed with a higher count when the user next logs in. For applications concerned about password storage, I would recommend No correctness issues found. LGTM. Full tables, scripts and data: https://github.com/ArtUkrainskiy/php-src-bench/tree/main/reports/pr-24188-pbkdf2 |
- Drop const signature change, that was unnecessary - Add UPGRADING message that iteration count can be increased - Securely zero HMAC contexts before freeing
|
Ping @adoy @saundefined @ramsey as PHP 8.2 release managers. |
LamentXU123
left a comment
There was a problem hiding this comment.
I think this is correct. I agree to merge it to master at least but for lower branches I don't have strong opinions.
|
I have no objections, but please wait for the other RMs. |
|
Also no objections from me. |
|
I mailed and pinged @adoy as well, be he seems incommunicado. I think this is ready to merge. Who can merge this? |
|
@LamentXU123 Can you merge this? |
|
@saundefined Can you merge this please? |
|
@devnexen Can you take a look at this? |
|
@arnaud-lb Can you merge this? |
|
Hey. Sorry for delaying. 8.2 branch is protected so only @php/release-managers-82 can merge this. |
|
This goes currently against policy and should not be merged. Or at least I don't see on which basis this should go in. We are considering the hardenings but no final decision on that yet and think we should first agree on the policy. This is not up to RM to decide. |
|
Targeting master is fine though. |
|
PR targeting master: #25463 |
This is a security bug, since it makes PBKDF2 faster for attackers than defenders. So it should be fixed from PHP 8.2 on.