# Randomize hash-function in php - Thread: https://externals.io/message/58934 - Original archive: https://news-web.php.net/php.internals/58934 - Started by Simon Schick on 2012-03-14T20:26:56+00:00 - Messages: 10 - Reply by email to internals@lists.php.net - More for agents: search with https://externals.io/search.md?q={query} · JSON API at https://externals.io/api · documentation at https://externals.io/llms.txt --- ## #58934 — Simon Schick — 2012-03-14T20:26:56+00:00 URL: https://externals.io/message/58934#58934 Hi, All I just came around that talk a couple of days ago .. http://www.youtube.com/watch?v=R2Cq3CLI6H8 I don't know much about hash-maps and internal php-stuff at all, but they say that the fix provided in 5.3.9 (and 5.4.0) is more a work-around than a fix ... Would it be an option to provide a real fix in PHP 6.0? They got the feedback that this will take some time and is not trivial, but we have a good time before PHP6 and can also break backwards compatibility for php-plugins if really necessary. As they said in the movie, PHP seems to have the algorithm DJBX33A implemented as Ruby. So as they're so proud of the fix provided by the Ruby-Team, may we can use that for PHP as well :) https://bugzilla.redhat.com/show_bug.cgi?id=CVE-2011-4815 This is not much because some attacker can do something, but what if you have a real-world-application that (for some reason) build up an array that just will blow up because of that? I haven't experienced that until now, but it's possible ... Bye Simon --- ## #58987 — Simon Schick — 2012-03-17T23:17:30+00:00 URL: https://externals.io/message/58934#58987 · In reply to: #58934 Hi, All I just came around that talk a couple of days ago .. http://www.youtube.com/watch?v=R2Cq3CLI6H8 I don't know much about hash-maps and internal php-stuff at all, but they say that the fix provided in 5.3.9 (and 5.4.0) is more a work-around than a fix ... Would it be an option to provide a real fix in PHP 6.0? They got the feedback that this will take some time and is not trivial, but we have a good time before PHP6 and can also break backwards compatibility for php-plugins if really necessary. As they said in the movie, PHP seems to have the algorithm DJBX33A implemented as Ruby. So as they're so proud of the fix provided by the Ruby-Team, may we can use that for PHP as well :) https://bugzilla.redhat.com/show_bug.cgi?id=CVE-2011-4815 This is not much because some attacker can do something, but what if you have a real-world-application that (for some reason) build up an array that just will blow up because of that? I haven't experienced that until now, but it's possible ... Bye Simon --- ## #58988 — Stas Malyshev — 2012-03-17T23:45:42+00:00 URL: https://externals.io/message/58934#58988 · In reply to: #58987 Hi! > I don't know much about hash-maps and internal php-stuff at all, but > they say that the fix provided in 5.3.9 (and 5.4.0) is more a > work-around than a fix ... This is true, it is a workaround in a meaning that the hash stays the same, but the fix prevents one from using excessive amounts of data that makes the hash unusable. > Would it be an option to provide a real fix in PHP 6.0? They got the Definitely. However doing this has some challenges. We're working on it. -- Stanislav Malyshev, Software Architect SugarCRM: http://www.sugarcrm.com/ (408)454-6900 ext. 227 --- ## #58989 — sam.e.giles@gmail.com — 2012-03-18T00:04:13+00:00 URL: https://externals.io/message/58934#58989 · In reply to: #58987 > Hi, All > > I just came around that talk a couple of days ago .. > http://www.youtube.com/watch?v=R2Cq3CLI6H8 > > I don't know much about hash-maps and internal php-stuff at all, but > they say that the fix provided in 5.3.9 (and 5.4.0) is more a > work-around than a fix ... > Would it be an option to provide a real fix in PHP 6.0? They got the > feedback that this will take some time and is not trivial, but we have > a good time before PHP6 and can also break backwards compatibility for > php-plugins if really necessary. > > As they said in the movie, PHP seems to have the algorithm DJBX33A > implemented as Ruby. So as they're so proud of the fix provided by the > Ruby-Team, may we can use that for PHP as well :) > https://bugzilla.redhat.com/show_bug.cgi?id=CVE-2011-4815 > > This is not much because some attacker can do something, but what if > you have a real-world-application that (for some reason) build up an > array that just will blow up because of that? I haven't experienced > that until now, but it's possible ... > > Bye > Simon > Hi, Fairly new to this list so go easy :P.. Anyway I was looking at the hash function in PHP the other week, and was playing around with some different implementations. DJBX33A is fast, which I guess is why PHP uses it as it is hit so many times in the execution. However I tried and benchmarked a few different algorithms, I didn't try the patch you mentioned, however the only algorithm that came anywhere close to matching the DJBX33A method is Paul Hseih's Super Fast Hash algorithm: http://www.azillionmonkeys.com/qed/hash.html I benchmarked the DJBX33A against Hseih's algorithm and compared the results using an Intel x86 architecture and an AMD 64 machine and both algorithms performed similarly. The benchmarks weren't robust as they were only quick, 'let's hack and see' tests. Would be interesting if anyone else has had a go. If I can find it I'll post a patch so you could? Cheers Sam --- ## #58990 — Stas Malyshev — 2012-03-18T00:12:24+00:00 URL: https://externals.io/message/58934#58990 · In reply to: #58989 Hi! > Anyway I was looking at the hash function in PHP the other week, and was > playing around with some different implementations. DJBX33A is fast, > which I guess is why PHP uses it as it is hit so many times in the > execution. Some time ago we've checked various implementations of hash functions and the result was none produces better results consistently than one we already have. Note that you have to account not only for the function itself but for the usage patterns - e.g., distribution of key sizes for variables, functions, classes, etc. However bigger question is - wouldn't another hash function be as vulnerable? Unless we have a perfect hash we'll still have collisions, and that means it still can be attacked if collisions are easy to generate. Obvious solution would be to use a salt for the hash, which prevents blind pre-computing of hash collisions. However, due to the fact that PHP hash values can be reused in different processes by bytecode caches, implementing it properly is not trivial. -- Stanislav Malyshev, Software Architect SugarCRM: http://www.sugarcrm.com/ (408)454-6900 ext. 227 --- ## #58993 — Tjerk Anne Meesters — 2012-03-18T05:56:33+00:00 URL: https://externals.io/message/58934#58993 · In reply to: #58990 > Obvious solution would be to use a salt for the hash, which prevents blind > pre-computing of hash collisions. However, due to the fact that PHP hash > values can be reused in different processes by bytecode caches, implementing > it properly is not trivial. What if php uses salts for specific hashes only, such as GPC (or all hashes whose lifetime is limited to the current reuqest), and use a zero-value salt for all others? --- ## #58994 — Xinchen Hui — 2012-03-18T06:32:43+00:00 URL: https://externals.io/message/58934#58994 · In reply to: #58993 Sent from my iPhone 在 2012-3-18，13:57，Tjerk Anne Meesters 写道： > > > […] > > What if php uses salts for specific hashes only, such as GPC (or all > hashes whose lifetime is limited to the current reuqest), and use a > zero-value salt for all others? definitely no，thinking of pre-calculated hash. Or Ajax which use json_decode parse input json. IMO, this Make no sense but mess things up. Thanks --- ## #58995 — Tjerk Meesters — 2012-03-18T07:05:42+00:00 URL: https://externals.io/message/58934#58995 · In reply to: #58994 > > […] > definitely no，thinking of pre-calculated hash. Pre-calculated hash of what? You mean binary serialisation? > Or Ajax which use > json_decode parse input json. That would be considered a request lifetime hash and therefore could be salted. > > IMO, this Make no sense but mess things up. We all have opinions. If a clear distinction between vulnerable and non vulnerable data can be reliably made, in my limited knowledge of the whole ecosystem I genuinely think this has a shot :) --- ## #58996 — Xinchen Hui — 2012-03-18T07:17:58+00:00 URL: https://externals.io/message/58934#58996 · In reply to: #58995 Sent from my iPhone 在 2012-3-18，15:05，Tjerk Meesters 写道： > > > > […] > > Pre-calculated hash of what? You mean binary serialisation? > > > […] > > That would be considered a request lifetime hash and therefore could be salted. > > > […] > > We all have opinions. If a clear distinction between vulnerable and non vulnerable data can be reliably made, in my limited knowledge of the whole ecosystem I genuinely think this has a shot :) > Ha, sorry for my rude words, I am not meaning you， but the point self;) And it's also why I am not usually saying words at internal@ , my poor English :) --- ## #59006 — keisial@gmail.com — 2012-03-18T16:04:30+00:00 URL: https://externals.io/message/58934#59006 · In reply to: #58993 > > > […] > What if php uses salts for specific hashes only, such as GPC (or all > hashes whose lifetime is limited to the current reuqest), and use a > zero-value salt for all others? We'll need to have at least two kind of hashes, at that point, I think it makes sense to place the salt as a member of the HashTable struct. Bytecode caches would just store the salt with the hash. We can also `mt_rand()` the salt of each hash, for further randomization.