[flagged]
[flagged]
OMG
So what's the timeline here? About a week ago (damn UIs that don't easily let you see the actual date) someone submitted a PR, but only two days ago the CVE was public. The PR is still not merged. Was the team let know in advance but ran out of time to patch or something?
I think that’s enough Internet for me this week.
What about LibreSSL, BoringSSL, WolfSSL, BearSSL? Are those vial replacement for OpenSSL. I know there was a lot of fuzz back then after HeartBleed, but how about these days?
if your threat model includes someone setting a 1.6GB proxy password, the base64 encoder isn't where you've lost
[flagged]
Is this demonstrable? or a finding, that can be exploited?
[flagged]
[flagged]
[flagged]
Assessment: https://tldrcve.com/openssl-encodeblock/
Bookmarked this, thanks for sharing!
This is what the scores should have been, I'm tired of these 9.0+ vulns every other day.
Thanks for sharing!
I like the site: concise explanations, commands to check if you’re affected (if available), and low traffic RSS subscriptions as they only process vital CVEs. Hope it continues to operate!
Is this going to be the 39th 9.8 CVE that results in 0 websites getting hacked?
I need some of you people to get hacked so I have something to point to, or I won't be able to convince management that this is worth taking seriously for much longer.
Since none of the enterprise distros have shipped OpenSSL 4 yet, this is not as serious as people are making it seem.
A CVE score doesn't determine your chance of getting hacked. It doesn't even determine how badly fucked you are if you don't update. You can have a CVE 10.0 that only affects you if your computer is set to a North Korean locale locate in Antarctica.
It's a number to guide your risk assessment process. If you don't have one of those, you shouldn't worry about them and read what people with risk assessment processes write about the CVEs.
Some people processing a lot of data are probably panicking right now. I doubt any website will get hacked.
Oh god, the discussion about fix is the most C thing ever, it makes the recent musl issue much more understandable:
- Let's return -1 for error
- People might not expect new -1 value, let's return 0
- How are clients supposed to differ between old 0 and new 0
It's not even C, most of the Windows API solves this problem by operating on handles error codes are clear and error checking becomes easier (though it still goes wrong). Good old winerror even has bit patterns to see what sort of error you're dealing with.
The Unix flavour of C, which relies on passing magical numbers, is making C's lack of proper error handling patterns a lot worse.
If the first replies I got from maintainers were those I'd tell them to fuck off then and keep having a broken library. They are like stackoverflow curmudgeons and it's not surprising other libraries appeared - who would want to deal with this.
What in particular do you find disagreeable here?
> Calling EVP_EncodeBlock() with an input length close to INT_MAX may overflow the calculation of the encoded output size and write past the end of the caller-supplied output buffer.
Haven't we suffered enough yet with lack of a basic feature in C? Why are enhancements going into C that never address this fundamental problem?
I wrote this proposal for C 17 years ago:
file:///C:/digitalmars/articles/C-biggest-mistake.html
It's a simple addition, it breaks 0 code, and provides a path forward to stop array overflows for good.
It's a foundational feature for D. We don't have buffer overflows.
Might wanna double check that url
> file:///C:/digitalmars/articles/C-biggest-mistake.html
Is that URL supposed to be a joke?
Web URL for that article:
> We don't have buffer overflows.
You don't have that class of buffer overflows, but for as long as D allows raw pointer arithmetic, you still got buffer overflows, i.e. here's a simple snippet triggering ASAN: https://godbolt.org/z/qPhMKzM1z
The more I see these posts, the more I'm convinced that "never reinvent the wheel" mentality we had so far in software engineering basically covered up a lot of lazy engineering, and we just pretended everything was fine. It's like these findings cross-pollinate.
Whaat? You want me to implement left-pad on my own? /s
that bug likely came from reinventing the wheel
It was introduced a year ago, so yes.
Not reinventing the wheel is fine. But one should change tires every now and then!
More like, we used to kick the tires and do manual inspections with a flashlight, now we have an automated tire checking system with multiple sensors that's uncovering stuff that manual inspections didn't see.
:/ I really need a day off from patching servers.
Sorry buddy, back to the server mines for you!
And yea, I'd really like a day where I'm not writing an email to customers about what they do and don't need to update regarding our software.
call in the agents
"Baldrick, you can have a day off when you're dead"
Same! Just did a whole bunch of out of band patching, more on the weekend. Fun times.
Don't worry. No enterprise distro have shipped OpenSSL 4.
If you stopped patching servers, you wouldn't be affected. Always use trusted, 5-10+ years old code.
Just stick to windows xp sp1 indefinitely. Those were simpler times. /s
So, uh, like, any word on what the "9.8" actually means? Is this "an obscure feature nobody has on could with a bit of luck and a cosmic ray strike allow arbitrary code execution if you're one of the few people using it" or is this "if you have a socket with OpenSSL exposed to the internet you're already compromised" or what?
So this might be very similar to Java4shell, though will need to be verified.
This is another base64 encoder where it takes a blob and coverts it to base64. I'm assuming an attacker control bit of data could be passed to the function which then overflows. Not sure if it can attempt to control a valid address or not, but unlocked AI likely could pretty quick.
This was possibly a rhetorical question but apparently it goes something like this - just a simple formula someone came up with that tries to model badness:
# Hardcoded weights for AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
attack_vector = 0.85 # Network
attack_complexity = 0.77 # Low
privileges_required = 0.85 # None
user_interaction = 0.85 # None
confidentiality = 0.56 # High
integrity = 0.56 # High
availability = 0.56 # High
impact_sub_score = 1 - (1 - confidentiality) \* (1 - integrity) \* (1 - availability) # -> 0.91
impact = 6.42 \* impact_sub_score # -> 5.87
exploitability = (
8.22 \* attack_vector \* attack_complexity \* privileges_required \* user_interaction) # -> 3.88
score = impact + exploitability # -> 9.76
(using Github's values from https://github.com/advisories/GHSA-83xj-7xh4-4p7x)Interesting that the impact calculation treats C, I, A as equal. And of course a lot of those inputs are just human(?) supplied vibes and/or context dependent in your use case.
Not quite what I meant. Having been on the end of trying to establish CVE scores for bugs myself I am well aware that even two bugs of the "same score" can be very different in nature. A "push button, pop every website on the internet" is a 9.8-ish, but so is "nginx with this particular couple of config settings will yield a root shell for a particular request" even if it "only" affects 5% of the world, because "affecting 5% of every nginx install" is still fairly justifiably a high rating for all the scores. The scale tops out at the upper end.
Plus we're seeing "AI has found this bug and it looks super duper scary everyone panic" followed by two weeks later "well actually it took sustained and direct assault on a website and sometimes it didn't work and it was actually more like a 6.4" lately.
Geezus_42's "you probably haven't actually shipped this code anywhere anyhow" is more like what I was looking for.
CIA psyop confirmed? (/s)
Since no enterprise distro has shipped OpenSSL 4 yet, it will only be an issue if you are running an upstream distro on a personal machine.
This should probably be in the headline. I was trying to figure out why Debian didn't have an update.
For real.
Seems like only OpenSSL 4 is impacted? What distros and systems use that?
I feel like in the post heartbleed era, a lot of people are not including new features of upstream openssl as much as they used to.
Considering OpenSSL 4.2 is the first proper 4.x LTS release, I think most distros will still be on OpenSSL 3.x so not impacted.
Bleeding edge distros like Arch/Fedora are already on 4.x, so they'll be impacted, but they should also receive a patch relatively quickly.
Well, Fedora 45 (with which I am composing this message).
Fortunately it doesn't look like I need to patch anything at work though.
Arch
pacman -Q | grep openssl
openssl 4.1.0-1
openssl-3 3.5.9-2Yeh but Arch is more of a vibe than an actual distribution.
I think that better describes omarchy then Arch proper
I see the OpenBSD guys are all over the PR (deservedly) shitting all over the proposed fixes so I assume LibreSSL may be impacted? Or are they just providing free advice and color commentary?
I think the second, LibreSSL was forked before OpenSSL v3 and this only impacts OpenSSLv4.
Nobody is shitting over anything in that PR thread, please stop catastrophizing.
Isnit a cultural, maybe: i found bob to be constructive and to the point.
Describing the comments on this PR as "shitting all over the proposed fixes" seems wildly out of proportion? This is direct, productive feedback on a PR that led to a consensus and approval on an appropriate fix in ~a week, we don't need to make the OS process sound worse than it is.
To me it looked like entirely reasonable discussion. Return one value or return an other value. And which one should be chosen based on what is in general expected. And then also consider what parts of code can this sort of change impact.
I really find sad how far we have gone. Purely technical merits of a fix should always be an open conversation.
I'm not really involved in OpenSSL, aside from a few tiny changes and commenting on a few things that affected GNU coreutils. However, I have seen Bob, whom I assume you are referring to based on the profile picture, participate in many discussions. I wouldn't infer too much from it.
Bob is a member of the Engineering & Standards team at the OpenSSL Corporation.
Do the projects collaborate closely since the split? My history is admittedly fuzzy.
Sometimes the fix can be worse than the bug. Like patching a hole in the wall.
Yes, they collaborate. BoringSSL and libressl especially, which IIRC share some of the same developers. But OpenSSL has it's own priorities and vision, which sometimes (often?) clashes with the sensibilities of the BoringSSL and libressl developers. Because OpenSSL is still the de facto standard bearer and developer target, it can get frustrating for the libressl and BoringSSL developers, especially when they zig and OpenSSL decides to zag.
I think CVSS 9.8 is exaggerated. This is not remotely anywhere near as serious as Heartbleed. It's an internal function which will fail if its input is far too long, but exploitability rests on how untrusted input can reach it; the key word being "untrusted". E.g. if you operate a base64 encoding service on the Internet that just reads all the data into memory at once and calls EVP_EncodeBlock(), that's exploitable. If you have a protocol that involves a block of data with a 16-bit length field which is passed to EVP_EncodeBlock(), that's 100% safe.
For those wondering about TLS: speaking as someone who has written a TLS implementation, AFAIK neither base64 encoding nor decoding occurs in the protocol itself, and the longest length field is 24 bits (Hello stream). Certificate processing involves ASN.1 which does support arbitrarily-long integers, but that's a big hint to check that they're within a sensical range.
Time is a construct. CVSS is a vibe.
"Internal"? It's documented in the `libssl-dev` manpages.
(Although I agree it should be an internal function.)
THIS!!!!!!!!!!!!!!! THIS times 10000!
I'm starting to get annoyed about the OMG OMG BREAKING BREAKING ALARM-EMOJI crap that's not a real emergency.
Four times in the last month, I dropped everything and started investigating whether a serious CVE related to any of our dependencies warranted an emergency bump, only to be frustrated that images, even from AWS, did not have the fix yet. So then I had to worry about whether I should build from source. We actually built PHP from source and deployed it to prod, just to get ahead of the PHP2Shell vulnerability
But yesterday's PHP vulnerability, CVE-2026-115560, didn't even apply to us or most people running PHP, and I don't think there are any realistic working exploits out there. I'm not saying that it's not possible, but I'm starting to feel that there's a large difference between what's theoretically possible and what's practically possible. And given the storm of patches and vulnerabilities found, this will become increasingly important to ask.
I think everything above a CVE score of 9.0 should warrant some level of “how exploitable is this in real life,” and anything above 9.5 should be proven to be easy for an attacker to execute, not just “possible in theory.” I don't see how we will survive as developers if there isn't some sort of better triage for vulnerability severity assessment. If everything is an emergency, then nothing is an emergency.
Yes, and this also causes inertia, as if everything is super highly critical it means nothing is, and people will stop caring when something is actually serious.
Heartbleed, Spectre, log4shell, they were all worthy of a 9.8.
This isn’t.
Yes. Lots of this. E.g. the nginx flap yesterday depended on http/2 being enabled, which it isn't by default. AI can be helpful with the analysis though.
No one runs nginx in the default configuration
There are more values in the CVSS vector you can fill in yourself: environmental and threat scores
So you can take e.g. a 9.3 critical down to 6.5 medium by setting CIA requirements to low and exploit maturity to Unreported.
I’m not saying CVSS is perfect but if you want to have a score that applies to your environment, you have to input some info about your environment into it.
“how exploitable is this in real life” is answered by the exploitability score, which is calculated from Attack Vector, Attack Complexity, Attack Requirements, Privileges Required, and User Interaction scores.
Good info, thanks.
(I'm a broken record on this so I'll keep it brief). The problem here is that the CVSS score is in the headline, as if it meant anything. But CVSS doesn't mean anything. It's a Ouija board; VR people jack it up to get attention, vendors nerf it to chill their customers out.
Someone should design something better, and less malleable.
I think tiny pieces of isolated code like this function simply cannot be assigned CVSS scores, because it's missing all the context and threat model around how it is used.
Then again, given that I once had to respond to a report which assigned a similar amount of severity to the strlen() function claiming that it will read beyond its bounds if its input is missing a null terminator (DUH!), I'm not surprised at the lack of critical thought among a certain population in this industry.
CVSS would be fine if they shared full vector rather than adding up the score. The vector is representative of what. Going from what to "how bad" through making it a single number is where they messed up because the weights usually don't make sense from case to case.
One major issue with CVSS is "attack complexity". It only has two options, high and low, and they're entirely dependent on the time in which it's evaluated.
Something that was super complex a few years ago might be super simple today. And I'm not talking about LLMs, I'm talking about stuff like if some CVE was dependent on you breaking e.g. MD5 in 1998, then it'd be "high", but in 2026 it's most certainly not that. And it doesn't need to take 28 years for something like that to change. It can theoretically happen over night or as the result of some other vulnerability. It even goes the other way too e.g. when ASLR was introduced.
I don't see that being at all helpful in this case. Which part of the vector conveys "this is a library function, so everything from the attack vector to authentication requirements to impact depends on how your application is utilising it, if it is at all"? For a bug like this, anything you select for into those fields is a stab in the dark.
A few years ago I participated in a study[0] about CVSS and apparently was very much in the minority in feeling that these scores are essentially useless. I have no idea what would actually shift the majority’s opinion on this, but I will at least echo your sentiment that everyone deserves something better.
It's not an out-there opinion among vulnerability researchers, though.
OpenSSL (I think, or maybe it was Bouncy Castle) no longer submits CVSS values because the final CVEs just end up with some arbitrary value that some random person made up rather than the actual value that the people who wrote the code and know it inside out assigned to it. The scores are actually literally useless in many cases.
At the end of the day, sysadmins care about "how fucked am I" from an email subject line. A replacement system would be "let me run a daemon on your network telling me what you run" with vendors telling me about the vulnerabilies and I'll tell you if you (the sysadmin) give a shit. Technically that's trivial, but who's gonna trust me to run a daemon like that on their network. Maybe you'd have luck with your public history, but that seems like quite the uphill battle.
There are several products like that (Tenable comes to mind). Having used them, I would disagree that "telling you how fucked you are" is straightforward, because they all do a terrible job!
They will statically determine (e.g. by SSHing in and rummaging around) that a library with a known vulnerability exists on the filesystem, but determining the reachability of that bug in the system as a whole (can this be exploited by the entire internet, or 3 sysadmins, or no one at all?) is not really an automatable task.
What if we standardized some sort of representation of known vulnerable combinations instead of a numeric score? As well as a standardized way to record system state regarding which libraries are used together and which of the resultant application stacks have network access or whatever else. That sort of data could be readily analyzed by automated tooling.
CVSS scores do not indicate priority. They give a baseline so you can pick what CVEs affecting you you should do risk analysis for first.
The score is just fine. The people misinterpreting everything with a CVSS score higher than 8 as the end of the world are panicking over nothing.
> The people misinterpreting everything with a CVSS score higher than 8 as the end of the world are panicking over nothing.
They are indeed, but that means CVSS has turned into a trigger for The Boy Who Cried Wolf. So we need a different metric. The problem, of course, is that adding a new score creates an xkcd 927 issue.
It's impossible to create such a score if you want to represent it somewhat objectively in a somewhat simple number.
If you want a quick overview of the risk: CVE-2026-121481 is listed as CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H which is what calculates the CVE score. This is missing the temporal/environmental components (which rarely ever seems to get entered).
This. It's a calculated score based on pretty objective measures. Yes, the CVE[0] notes that "The severity is Low because no attacker-controlled path to the overflow is known."
But risk or severity is not what CVSS measures. It's a score intended for security professionals who know the context, graded against a fixed algorithm, with a fixed definition for each answer. The nvd.nist.gov site isn't always the easiest to use, so grab a good CVSS calculator[1] and look at what it means. That's why the alphabet soup vector string is there: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
I'm sorry, CVSS wasn't meant to eliminate your need to think or understand your environment. Yes, if it's a 9.8 CVSS you should look at it. Triggering you to look at it is all the CVSS is meant to do!
Now, whether you think CVSS scores mark too many things as High or Critical is a little secondary. There is a reason we're at CVSS v4. Defining metrics and keeping them useful while remaining objective is difficult. This vulnerability, however, is exactly the kind a chain exploit would take advantage of.
[0]: https://www.cve.org/CVERecord?id=CVE-2026-121481 [1]: https://www.metaeffekt.com/security/cvss/calculator/
> But risk or severity is not what CVSS measures
Risk is not what the CVSS score measures, correct. However the CVSS base score is first and foremost a measure of severity (of the vulnerability). I think they were being loose with their terminology when they wrote "severity is low" and probably meant to say "risk".
The spec talks about this in depth. Paraphrasing: The base score captures intrinsic characteristics of the vulnerability assuming a (reasonable) worst-case configuration.
It's a measure of "How screwed am I if I'm using this software in this specific-but-reasonable manner?", it's not a measure of "How screwed is everyone using this software?".
P.S. I broadly agree, just adding a minor correction on the terminology.
Exactly, yes.
And CVSS v4 does add criteria that let's the provider specify how urgent it is, if it can be automated, etc.
I think the bigger problem is that the scoring system lets you specify a small number of factors and still reach a 10.0, and all the optional and supplemental factors reduce it. Like if you try to enter this vulnerability into v4 and fill out more factors, you easily drop it into the 8s.
So essentially everyone submitting the vulnerability is doing the bare minimum, contributing to the perception that the scores are inflated.
And then people don't use the current version of the score. CVSS 3.1 came out in 2021. CVSS 4.0 came out in 2023. Why the hell is everyone still using 3.1?
It feels to me the only way to beat LLMs finding vulnerabilities is to have LLMs (or LLMs paired with other AI methods) constantly monitor your infrastructure, traffic, database, etc. to ensure that nothing out of the ordinary is happening
Or use defense in depth. There will be a vulnerability in every layer with some non-trivial probability, but rarely in multiple layers at the same time.
When Plex got hacked and my linux server had a cryptominer installed, I had no clue where to start fixing it. I loaded up Claude to analyze the system logs, and it told me what commands to run (and why) and the infection was gone in about 10 minutes. There's no way I would have known the many things I needed to do to fix and secure the system. I'm fairly new to Linux, and without the AI I would be helpless with some of the tricky things.
You're still helpless, since you didn't do what you should have done (format).
The right thing to do in that case is to save whatever data you need from the server and then nuke it thoroughly, which means all the disks should be wiped, don't keep the old UEFI partition around, and reinstall. It would be preferable to skip the first step and get the data from a source that wasn't compromised, but sometimes you have bring out the hazmat suits to recover stuff from compromised systems.
This is very reactive though -- imagine you were a business. You'd be screwed. Your data is already gone.
I think so too; I think apart from pushing vulnerability research forward (which is important on its own merits just for computer science reasons), the major engineering challenge of this moment is figuring out how to make platforms resilient without rewriting everything; that could be better live-patching, it could be some weird monitoring thing, it could be automatic segregation and workload isolation.
It's not "just keep up with the patches". That's going to be a losing strat.
maybe today, you cant keep up, 2 year later, AI rewrite everything safe in rust, you wont need patch everday
Hahahahaha
I'm not sure how cost effective or even viable that option would be. I think the only practical defense is some kind of baseline hardening at the OS or application level. Something like eBPF rules i.e. what Cilium Tetragon policies provide. Basically it seems like there's no way to guarantee that your application code can't be exploited in some novel way so your best option is to restrict them as much as possible.
> there's no way to guarantee that your application code can't be exploited in some novel way
As a thought experiment if this hypothetical application code were written in something resembling haskell I suspect it would actually prove fairly difficult if not impossible to find novel ways to exploit it. I think at that point an attackers only options are logical inconsistencies in the underlying business logic (see that one NASA rocket) or bugs in the compiler and runtime.
I'm really just talking as a layman here.
My feeling is that if we use traditional security approaches that seems like a losing strategy. You cannot just keep patching. I mean obviously you should, but that is not enough.
LLMs are finding zero days in operating systems, language runtimes, everything. They will definitely be finding holes in your API. It is not a question of 'if'.
I agree that patching is just plugging holes in the dam with your fingers. I think the only viable defense is something more low level in the kernel. SELinux, eBPF, or something new and more sophisticated than those. A lightweight agent passively monitoring system calls and detecting anomalies perhaps?
And what would that solve?
That's why it's recommended to use size_t for passing sizes of buffers. It's unlikely to get overflow with it. In this particular case input buffer needs to have size larger than half of address space in order to trigger overflow, which isn't possible (64-bit systems usually have only 48 bits of actual address space, 32-bit systems don't allow allocating so large blocks).
never say never :)
> That's why it's recommended to use size_t for passing sizes of buffers.
I have always been told it is a bad idea to use unsigned integers in your interfaces. Why? An accidental negative (signed) value will be cast into an enormous unsigned value. For example, if you look at the vast majority of the Win32 API, inputs are normally signed integers. They rarely use unsigned integers.> An accidental negative (signed) value will
If you use mostly unsigned values it's not a huge problem, as long as you don't perform subtractions or operations with possibility of overflow. If obtaining a negative value or getting an overflow may occur, you should perform proper checks before calling an API function. And in case of WinAPI or something similar usually no extra computations before a call are necessary, arguments are already precomputed (like with buffer sizes).
Not everybody recommends size_t. There's a sizeable community of people absolutely convinced most integers, including length and capacity integers, should be signed. And they stick to their guns even for C and C++. They have some very compelling logical arguments on their side, but IMO in languages like C and C++ they're just fatally unrealistic.
ssize_t
This is anachronistic imo. There is no point of using signed or larger than needed integers when programming in a modern way.
You have fuzzing, test coverage checks, overflow checks, static analysis, llm review. There is no conceivable reason why you might have a problem with using an u16 for a buffer length.
Even reading the PR comments shows that these libraries are just really low quality.
They are discussing if they should check the error return on callsites now that they made a function fallible.
Even a library like openssl using raw arithmetic operators all over the place is just weird imo. Every arithmetic operation should be checked version of the operation unless there is a provable performance hit. Languages should just not have operators
Why unrealistic? I was in the unsigned camp a long time ago and switched to signed types for indexing and sizes, and this made many problems go a way.
There was just a nginx issue that highlights how problematic the use of unsigned is in this context: CVE -2026-123007
Why would anyone prefer signed for indexing and sizes?
With unsigned you only need one check: 'i < bound', it works always, regardless how i is computed, and it doesn't matter if the calculation overflowed or wrapped around, still works.
With signed you need: 'i >= 0 && i < bound', and you need to be very careful how you calculate i because overflow is undefined behavior.
The issue with accidental overflow leading to wrap-around is also relevant to the value of `bound`.
If `bound` is unsigned and experiences accidental wrap-around due to underflow, the simple `i < bound` check fails open: most values of i pass against the unintentionally-huge `bound`. The signed version fails closed (no i passes both i >= 0 and i < bound), which is usually safer.
That’s how the argument goes, anyway. You’re right that using signed means a missing `>= 0` check can be a problem, as well as all the risks of signed-overflow UB.
No perfect solution in C.
The N >= 0 check you often get automatically using sanitizers, e.g. when using an array [N] somewhere, or by concerting to size_t at last moment and then getting a size which is so big that allocation is guaranteed to fail.
Not that are wraparound CVEs in Rust as well, so there is no perfect solution even outside of C.
From a brief look, I'd say there were several wrong turns made that led to the CVE-2026-123007 overflow bug.
For why I think it's unrealistic to go all-in on signed integers in C, it's because sizeof is unsigned, and most APIs dealing with lengths use size_t, and there are many pitfalls when mixing signed and unsigned types.
But I agree that there's a good case for using signed for the task behind that CVE. As I said, there are compelling reasons.
OpenSSL is a case study in the pitfalls with mixing signed and unsigned in C code. A major reason is the careful dance it has to do mixing signed and unsigned, and it stumbles. There's no substitute for being careful with arithmetic, and at least for me, when I expend that effort I also tend to use overflow checking APIs, clamping, etc, because it's less prone to error. Do that and the benefits of signed start to disappear, or at least the debate starts to lose relevance. Why OpenSSL doesn't create and use a set of overflow checking routines is beyond me as it would simplify an incredible amount of very hairy and brittle code.
The way to deal with APIs which use size_t and which works for me is to convert at the last moment.
I'll have to more carefully read some of your code sometime. Any links to good exemplars? I'll admit I should be more amenable to making use of signed, and I wouldn't have said that several years ago.
At the same time, the more I employ overflow checking routines[1], the less interest I have in the whole debate. Those routines cut both ways as they also make it safer to mix signed and unsigned.
[1] GCC/clang builtins, which IIUC the new C2y API mirrors. The return value semantic is unfortunate, especially in C2y where they lack the "overflow" name prefix, but it's understandable why C2y had to copy it.
>>I am really worried about the negative return value being added. Existing third party callers might not be ready to handle this correctly (i.e. if they cast the return to size_t without checking). Could we instead return 0? (In all the added error returns in this PR.)
Rust fixes it.
Sometimes I like to think of how little performance would be lost and how much would be gained by just collectively deciding to use Option/Result instead of sentinel values.
Things like the dumb (not like I'm any individual being dumb. Just collective accumulated stupidity) a*hole vulnerability some weeks ago. That wasn't even performance sensitive.
Rust does not fix existing third party callers.
yes, it does. Please refer Rust's return types
Existing third party callers don't use Rust, so Rust's return types can't possibly help here.
Be sure to let us know when your equivalent Rust crypto library is available.
You rang?
If I am not mistaken, the default backend is aws-lc [1] which is... a C library (still better than having the frontend in C as well, though).
Meanwhile, ring [2] carries an explicit disclaimer of "This is an experiment".
You can also use a formally proven https://github.com/ctz/graviola/
aws-lc is a lot of assembly, which you kinda need anyway. Neither Rust nor C are fully suitable for cryptography because common optimizing compiler backends (LLVM, GCC, MSVC) can introduce branches (even if you wrote it in a branchless manner!), making the crypto code vulnerable to side-channel attacks.
So the options are
- write it in assembly
- use a high-level language, then verify the generated assembly the compiler generates
- write a new or modify an existing backend to support timing-independent code
aws-lc does the first, some other crypto libs try to do the 2nd but I don't know how solid that is and the 3rd one takes time.base64 encode hopefully isn't used in places where it would be a timing channel, so it could still benefit from a safe Rust impl. And at least it's API could benefit from slices and Result
The crypto primitives in OpenSSL have been pretty solid (although timing attacks come up from time to time).
Where OpenSSL has been bad to terrible over decades is protocol handling, api management, and lately locking.
Managing protocol in a more restrictive language is a good thing. Bonus if it has the potential to require less work from application developers around version updates. Maybe they got better, I was using <= 1.0.2; back then there was a chance some things would break between different letter releases on the same version number even though it still compiles, and going to a new version meant a lot of work.
we don't need many crypto libraries, maybe just one for the whole world, written in Rust and then all functions are math-proofed.
It's not just "math-proofed". You need a lot more to be a crypto library.
Like time-constant, and it's hard to be time-constant in a performance optimised language like Rust. LLVM is doing some magic, and you don't want any magic in a crypto library.
You'd probably need a second one for FIPS.
Does the alpha-stage WIP code you call 'Rust' compile for 32-bit yet? No? Ok.
A crypto library written in Rust would not be available everywhere, so can not be the only one. And if you formally verify the code, you can also formally verify that it is memory safe, so there is not much point to have it in Rust. Crypto has properties which may be difficult to verify though.
If you want a safe interface that does not require passing random sizes around in extra arguments, this can also be done in C.
it's too late for C/C++ developers.
Rust and C developer there. C++ it's an unrelated beast, don't mix C99 users with these. On C, I compiled Gplaces (A barebones Gemini -protocol- client with Clang hardening CFLAGS and I linked it against BoehmGC. That in 32 bit.
On Rust, it's pure love in 64 bit, I acknowledge that (I use Cosmic DE with Pop!_OS so I can relate) and evcode it's my main IDE, Crabcraft being second. But compiling 'simple' TUI tools on Rust it's impossible on 32 bit, even with ZRAM and a total of 1.5GB of RAM.
If you want half-secure TUI tools on 32 bit Golang it's a saner option than C/C++ and the only actual altual alternative to Rust. For instance AERC can be safer than Mutt to read email, even under a Maildir mailbox and email fetched/sent with isync/msmtp.
I am not a system administrator, but I am getting a serious vulnerability fatigue, every single day there's at least two new vulnerabilities in major pieces of software, I'm just tired of seeing that everywhere and hearing that everywhere. Fellow system administrators must be having it much worse.
The bugs were always there. The fixes are coming in fast today.
This will lead to LLMs being embedded 24/7 into critical infrastructure. A "living" firewall and sysadmin.
Already happening. My org has shifted agentic vuln triage/patching to p0.
We could it call it a Blackwall
Don't tell me. I work for a bank and we have automated vuln checks on our code and dependencies and we can't release in pre-prod if we don't fix the cves. Basically since last year is 1 day you deploy new code and 2 days you deploy cve fixes.
That's a great process for a bank!
The problem with "must fix CVEs" is that it's stupid, as most CVEs will never be a problem unless somebody alredy broke into your server. Yes, yes, security comes in layers, but the security circus is also a reality.
It's also a denial of service attack on your brain because you have to do the work to prove to some beurocrat that the vulnerability can't be exploited in your environment. It usually turns out to be easier to just take the fix. And then of course we get the production downtime events because the fixed versions have other new bugs.
The solution is {Open,Net}BSD
No it isn't.
From OpenBSD, LibreSSL might be.
Until enterprise distros switch or companies are convinced to switch distros, OpenSSL will remain the default. Neither seems likely to me. Even if they switched to LibreSSL, it still has vulnerabilities.
I see bugs being patched, and software improving. Bring em on.
Those aren't mutually exclusive... it's good that these bugs are getting fixed, but each critical severity bug that gets fixed is a headache for all of us who administer servers.
You're seeing bugs get band-aided. Clanker-generated band-aids for clanker-surfaced issues aren't necessarily a proper fix, and may even be a step backwards if they introduce new issues.
I see the Kessler Syndrome play out at work.
Are you implying that patching vulnerabilities will lead to more vulnerabilities? How so?
Not necessarily more vulnerabilities, but I think the fear is that the continuous churn of security patches will lead to increased instability and reduced availability.
Security patch intentionally or inadvertently removes a behavior that systems X and Y depend on, so they must be fixed, and then that change cascades to dependent systems A, B, C, and D. And some of those changes will be incorrect and introduce new vulnerabilities and the cycle of life continues..
this bug was introduced by clankers most likely...
likely based on what
- version 4 got released this year
- mid 2025 a lot of people appeared contributing out of nowhere
- https://github.com/openssl/openssl/pull/29178 this rewrote the code for base64 - you can delve into it to check if this was human generated.
That contributor is definitely a vibe coder these days; just look at their profile's readme: https://github.com/Nick-Nuon, the only place I have ever seen that emoji list style where every bullet point starts with an emoji is from chat bots, but chat bots really really like it.
I don't see any obvious tells from the code diff though, I would've expected chatbot code to be more commented and nothing immediately stands out.
The commit messages and PR description don't read at AI-generated at all, there are small typos and grammatical issues. For example, the sentence "Originally, the OpenSSL base64 encoding kernel includes a buffer of ctx->length" confuses word tense a bit, correct grammar would've said "Originally, the OpenSSL base64 encoding kernel included ...". Chat bots are terrible at writing but typically impeccable when it comes to word tense and spelling.
Good to see that I'm not alone in seeing very weird things there - I also don't think it was vibecoded (back then, Claude wasn't there yet...) but they surely used an LLM (the passage you mentioned likely can easily be produced by pointing the LLM at 2 folders... - also scripts outputting emojis are a tell imo...).
I guess the PR back then was a bit of a CV-padding action, where they integrated their research into a widely used OSS-project. That the project maintainers just accepted this in good faith (tbh the PR is better than the last vibecoder I saw pushing into a rust-numerics projects: "enable AVX only on AMD because my broken 10 y.o. Intel clocks down when executing AVX...") bodes ill that our shared infrastructure survives this in usable "shape".
On a very very quick read, the bug is that EVP_EncodeBlock malfunctions if the input is larger than ~1.5GB. EVP_EncodeBlock is a function that encodes base64.
https://docs.openssl.org/1.0.2/man3/EVP_EncodeInit/
As a seasoned C programmer, I can immediately observe that the whole API is bad: the caller is expected to calculate the buffer length on their own, using a nontrivial formula, and the caller does not pass in the output buffer length, so the implementation cannot validate it. But apparently OpenSSL 4 also implemented it wrong and wrote (possibly far) past the end of the buffer.
You should probably assume that, if you are running anything that can be convinced to use this API to encode a very long blob as base64, then you are in trouble. I don’t know why a modern server would do this.
BTW, OpenSSL seems to be a rather low quality implementation of a very low quality set of APIs. Stay away.
edit: Fixed my size threshold, hopefully. I was thinking of unsigned ints, but this is a signed int. I have not looked at the code, just the description.
Many OpenSSL APIs are terribly designed.
>As a seasoned C programmer, I can immediately observe that the whole API is bad: the caller is expected to calculate the buffer length in their own, using a nontrivial formula, and the caller does not pass in the output buffer length, so the implementation cannot validate it. But apparently OpenSSL 4 also implemented it wrong and wrote (possibly far) past the end of the buffer.
Many such cases.
I expect we'll see some fun libyuv bugs soon in applications which decode media. The APIs are begging for misuse.
https://chromium.googlesource.com/libyuv/libyuv/+/refs/heads...
I've been annoyed by writing code against OpenSSL's APIs before, precisely because it expects me to provide "appropriately sized buffers" without documenting the correct way to compute the correct size very well.
> I can immediately observe that the whole API is bad
I see this is your first time reading OpenSSL code
> BTW, OpenSSL seems to be a rather low quality implementation of a very low quality set of APIs. Stay away.
Yep, this is not news. https://en.wikipedia.org/wiki/LibreSSL was forked off OpenSSL now more than ten years ago to address exactly these kind of issues, and more (at the price of occasionally taking some drastic measures)
But somehow we're still here.
See: Lindy effect
> BTW, OpenSSL seems to be a rather low quality implementation of a very low quality set of APIs. Stay away.
Well, it depends. if in the coming weeks many more Heartbleed-level vulnerabilities would be exposed, {Amazon, Google, OpenAI, Anthropic} can just rewrite it in Rust, formally verified.
There does not exist such implementation as of now as OpenSSL was good enough, AI may expose so many security flaws it would no longer be good enough.
[dead]
> There does not exist such implementation as of now as OpenSSL was good enough
A fair amount of such implementation do exist.
This is really embarrassing because this isn't even ancient code.
OpenSSL has been around a long time and has a lot of legacy code, but this is in OpenSSL 4.x only -- stuff that was written in the last ~year.
If they don't have the manpower to write new code safely they should just stick to maintaining what they've already written.
This function has been there since the first commit in git history, from 1998.
and now guess who helped writing this code. the chicken are coming home to roost, but people who are happy about their youthful productivity or stock packages will not be able to notice...
the code before this PR included the check. But that was old code!
I like to declare that industry coding standards such as MISRA exist precisely because code bases like OpenSSL (specifically) exist, as it's such a good example of breaking every sensible rule in the standards.
MISRA's not great itself, though. Half of it is obvious and could be summarised as 'don't invoke undefined behaviour' a quarter is rules to get around absolutely terrible ancient compiler problems, and a quarter are arbitrary restrictions that don't really help improve code quality much. For example, the interface of this function wouldn't break any MISRA rules in a way that would force it to be any better.
Even the discussion in the PR about not handling the errors in all call-sites is revealing imo.
What is it revealing? I read it and the assertion seemed like a reasonable middle ground.
One of the ‘rewrite it in rust’ projects that should really cause anyone to be offended IMHO.
The problem with "rewrite this in Rust" is more that Rust has laughable amount of Tier 1 platforms.
Now, rewrite in Ada, given that GNAT supports the same targets as rest of GCC...
Completely agreed but I’ll take any improvement at this point however small
There's gcc-rs too but at a slower pace than mainline/official Rust.
On security, both clang and gcc have security flags for this and Boehmgc (or even mimalloc) can be temporally used as a replacement to avoid nasty surprises.
What's obnoxious is the WG14 standard committee could have at any time in the last 40 years provided a standard slice and buffer type that would go a long way to discourage trash API's like this one.
I swear it's de rigueur in C that if an operation requires 5 parameters to insure the operation is error free than the function shall only take 4 parameters.
Rewriting the function with idiot mode disabled
int EVP_EncodeUpdate(EVP_ENCODE_CTX ctx, buffer_t out, const slice_t in);
How dare you give your parameters names of multiple characters!
How dare you pass a struct by value to a function. Heresy!
> What's obnoxious is the WG14 standard committee could have at any time in the last 40 years
Or just use bloody C++ that solved that years ago with std::span.
Unfortunately, many C dev prefers whining about "How horribly bloated" C++ is instead of using the part that makes sense.
The world would be better off without yet another newly slopped C/C++ app.
There are very few reasons to start a greenfield project in C++ at this point.
There are plenty of reasons to start a greenfield C++ project in 2026.
Providing C compatible APIs is one of them.
However, I do not see any reasons to start a plain C user space project in 2026. Excepted if you are embedded / kernel dev.
Everything advantage you do have in C (portability, compatibility, memory control).... you have them in Modern C++, but better.
> embedded
Nintendo has been writing their OSes (3DS and Switch 1/Switch 2) fully in C++ (microkernel included), so not even this is an excuse to use C over C++.
if the input is larger than ~1.5GB.
...on a system where ints are 32 bits. The bottom line is that this doesn't appear to be easily exploitable and IMHO it should be the callers of EVP_EncodeBlock which need to be inspected to determine the actual risk.
using a nontrivial formula
It's base64. 4/3 expansion; round it up and add padding. What's "nontrivial" about that?
and the caller does not pass in the output buffer length, so the implementation cannot validate it
Not necessary, because the caller is responsible for allocation.
I am confused why you make out 32 bit int as being rare. Almost all devices that you'd reasonably connect to the internet at this point have int as 32 bits
The question can be reduce to "who do the length validation".
In my case, if i write a public-facing API, i never trust the caller will do the correct thing, thus the caller must provide all the critical parameters, and i must validate all of them.
The question can be reduce to "who do the length validation".
At the boundary where untrusted input becomes trusted. This is commonly known as the sanitisation/validation layer.
> It's base64. 4/3 expansion; round it up and add padding. What's "nontrivial" about that?
If you mean ceil(inlen * 4/3) + 1 (even if you use exact arithmetic), then you are passing too small a buffer and you will overflow. Maybe your “add padding” part fixes it? In any case, I think that the fact that this confusion is even possible proves my point.
You should not need to read the manpage of a function you are calling in order to determine whether you have passed a large enough buffer to avoid overflow.
Reading the manpage to get the function to not return an error would be a different story.
If you mean ceil(inlen * 4/3) + 1 (even if you use exact arithmetic), then you are passing too small a buffer and you will overflow.
If you let untrusted input reach inlen and overflow that computation, you have already failed:
inlen isn't an untrusted input and it's not this calculation that's overflowing (integer overflow). It's the buffer that oveflows when you allow the operation to proceed based on faulty validation.
The caller shouldn't be responsible for enforcing preconditions of your function. They should just provide the input buffer, output buffer, their sizes and your implementation should verify whether it's safe to proceed.
Are you saying that any code that base64’s a string of attacker-controlled length is a mistake?
I think userbinator thought that you meant that the calculation would overflow, while I think you meant that the calculation, even if not overflowing, would return the wrong value under some conditions and that subsequent code would then overflow the buffer. (As for what I think about the actual issue, I’m too sleepy to even try. Which is probably an indication of something. =) )
> BTW, OpenSSL seems to be a rather low quality implementation of a very low quality set of APIs.
Fully agree with your assessment above that line, but damn, this is a strong statement for a library this ubiquitous. You wonder; how could this happen? And yet, as an open source developer myself, I know how little people care about code quality.
If you've ever tried using OpenSSL you'll know that it's true.
The API is a fractal of bad design. Not just that, but because of the API refusing to make sensible choices, it forces everyone using it to be an expert in cryptography, with predictable results for those programs that use it.
I'd like to point you to: https://www.haproxy.com/blog/state-of-ssl-stacks
Among the deeply technical and highly credible and detailed articles I've read in my life, it's definitely the spiciest. At the end:
> It is often said that fixing a problem requires smarter minds than those who created that problem. When the problem was architected by a team with strong convictions about the solution‘s correctness, it seems extremely unlikely that the resolution will come from the team that created that problem in the first place. The lack of progress in the latest releases tends to confirm these unfortunate hypotheses. The only path forward seems to be for the team to revert some of the major changes that plague the 3.x versions, but discussions suggest that this is out of the equation for them.
IOW this extremely strong claim does seem to be supported by extremely credible actors.
Many people are using forks that are mostly API compatible. So the ubiquity is not as huge a problem as it used to be.
Because you don't roll your own SSL.
How it happened? A lot of features in OpenSSL was written as part of a PhD and dumped in the repo without any thought about long term maintainability or quality. Now, during the last 5 years or so, people have been trying to straighten it out, but this is a pretty daunting task.
A related problem is that the IETF's RFC production center (that's its actual name) is manufacturing new RFCs far beyond the capability of anyone to keep up, and every half-assed idea anyone ever has or that some vendor pays an IETF consultant to turn into an RFC for them ends up getting bolted onto OpenSSL, which serves as the AB-positive universal recipient that everyone tests their ideas on, typically in a half-finished, barely-tested form of which Heartbleed was the prime example. It's not even a Red Queen problem because no matter how hard you run you're still falling behind each time.
> And yet, as an open source developer myself, I know how little people care about code quality.
This is a weird statement. I am someone who has worked on closed-source internal enterprise software for my entire career. Universally, the internal source is worse that the numerous open source libraries that we use. If anything, open source has "more eyeballs" than closed source, so the code quality will be better.I don’t know. If you are using all the tools at your disposal like linting, static analysis, dynamic analysis, DAST (if web), fuzzing, SCA, etc and correcting your code for the findings, dispositioning and correcting them can that really be said? Open source doesn’t always
> You wonder; how could this happen?
It all came from a bad place. OpenSSL grew from primitives created by cryptographers. Cryptographers are great, and they make fabulous things. But almost none of the understand software development or care a great deal about the mundane matters of API design, or the obnoxious limitations of naively designed processing elements.
Others have cited the documentation for EVP_EncodeBlock(). In the mind of a cryptographer, the text is perfectly clear and trivial, and they likely haven't exposed themselves for many years to anyone that can't read that exactly once and not immediately and correctly visualize the necessary formulae. Nothing wrong here at all. The fact that on real computers, numbers don't just scale to whatever value a computation requires is merely a defect in the design of the machine.
Mathematician: "Um, just ignore that infinity symbol there"
SSL/TLS has always been a wonky and complicated standard built upon other wonky standards like x509 and ASN.1 (for PKI) that nobody would touch unless they are forced to.
There is a reason why people did not do a lot of TLS (outside of anything involving monetary transactions) before Firesheep, the NSA leaks were the final nail in the coffin for unencrypted HTTP. On top of that Let's Encrypt and ACME did not exist, so TLS was behind a paywall on top of that.
And then everything proliferated relatively quick even for tech standards. I guess people grabbed onto the next available tooling just like a drowning person would take any piece of driftwood.
Not so fun fact: AFAIK the openssl binary was initially explicitly meant to be only used for testing. That is why its command line is so strange. It was initially meant to expose a couple of low-level primitives, not to implement workflows. One example would be the openssl-ca man page, which explicitly mentions this. People didn't care because unfortunately you could somehow do everything you could ever want with the openssl binary and I guess that is also something how the impression stuck that openssl is the most advanced library.
> Not so fun fact: AFAIK the openssl binary was initially explicitly meant to be only used for testing.
I'm in shock ! Someone noticed ! Thanks man ! :)
In the meantime in real world IT :>>
line 57: https://github.com/acmesh-official/acme.sh/blob/master/acme....
line 306: "This script requires openssl binary" https://github.com/dehydrated-io/dehydrated/blob/master/dehy...
Pretty recommended scripts :)
And I'm sure there are more and not only for ACME, for example I'm sure I saw few examples in production setup docs from IBM and Redhat... :>
Yes, openssl library is wholesale backdoor, subtype: footgun.
Google forked OpenSSL in 2014 (BoringSSL) with a pretty similar rationale, stated more diplomatically.
I just wanted to check what became of the original LibreSSL fork. Google forked it into Boring SSL, that's what happened.
If you use OpenBSD that's what you get in the base system.
LibreSSL is still going strong.
I remember a long time ago reading openssl was dumpster file code quality wise. I guess it was just built that way and never improved.
I wanted to see how OpenSSL handles the x64 instructions to accelerate SHA1 hashing and it turns out the assembly code is generated by a perl script.
That's not automatically bad, in fact it can be better than just inlining the instructions. I won't judge this instance of the pattern, but if faced with (perhaps incomplete at the time) intrinsics, or hand coded inline assembly, vs generating with an external script (or "mini compiler" if you will), it might be less error prone to generate.
That's normal, if outdated - you'd use Python in anything written since 2010
Assuming you mean x86, that part was originally written around 30 years ago via this funky setup that did MD5 and SHA-1 from the same code. It's not really any kind of example of current practice.
No, the SHA-NI instructions
We have known this since at least heartbleed. The answer was that basically no one funds the project and it had only one developer.
All these trillion dollar tech companies have been leaching off open source for free and never contributed back to the projects they rely on.
https://boringssl.googlesource.com/boringssl/
All yours! Use it to your heart's content.
If you think about it, giant corporations are making gobbles of money from people who wrote software for free and in their free time. And they still complain when it doesn't work as they expected. Why they didn't fund a better version of OpenSSL?
Not allo companies loot OS projects with no contributions. M$ has contributes back to many OS projects and Google has dozens of OS projects themselves.
And when Redis changed their license for such reasons, companies and people united to create ValKey or even the Centos/Rocky linux situation.
Redis cloud would cost you arm and leg compared to self hosting. I gives me an idea why big tech was not keen on making business with them
AWS has 90% profit margins and still millions of customers.
big tech can spare a few arms and legs, though
It's insane. I maintain a small open source project with only about 100k downloads per week and so, in terms of funding, it's crickets for me but I assumed famous projects like OpenSSL would get some decent funding.
Then I read articles about billionaires donating millions or even billions to various (sometimes frivolous) charities and I wonder WTF.
The numbers just don't add up. It's like there is an invisible force field preventing people from donating money to open source?
Crazy to think that the people who once posted a dumb viral 30-second meme video on YouTube probably got more money than the author of a complex, world-famous open source project which has been at the top for decades and which quite literally intermediates every aspect of almost every human's life.
Does not compute.
AFAIK donation distribution is still unsolved problem. That is there is no single org business can donate $1M to and have it distributed among all the projects business identified it relies on.
I would very much like for my employer, who benefits massively, to donate. However, the people in charge of the money decisions don't know anything about software and don't want to be lectured by me. They have _important business stuff_ to do. Sigh.
Plus it would bring literally no benefit to your company .
I understand why you would say that, nut it's not even true. We've had plenty of bugreports that have had crippling effects. For a minuscule fraction of these, I have been able to provide fixes or workarounds. But mostly, I would have been thrilled to get paid support there!
[flagged]
What's this about?
I'm not the parent poster, but I believe it was a reference to https://en.wikipedia.org/wiki/Omarchy#Funding_and_controvers...
OpenSSL receives millions of dollars of funding each year. The corporation that runs the foundation reported 5.5M USD funding in 2024[1], and 5.0M USD funding in 2025[2]. These days, OpenSSL is well-funded.
From your HN bio, I guess this is your project: https://socketcluster.io/
From the home page, I see that you are sponsored by two different companies. To me, "in terms of funding, it's crickets for me" isn't an honest statement.
If funding is such a problem for you, why don't you use a non-open source license, e.g., Business Source License? This would allow non-commercial projects to use your software for free, but commercial users need to pay. There are a bunch of big, important open source projects that already changed: HashiCorp, CockroachDB, MariaDB, etc.
[1] https://openssl-corporation.org/blog/2025-02-14-annual-repor...
[2] https://www.openssl-corporation.org/assets/Annual-Report-202...
I don't believe BSL = commerical issues have to pay. Typically it means that the code can be used internally, but not distributed or resold as a service.
Interesting, three years no new Release necessary of SocketCluster? That must be quite stable, congrats!
From my understanding the problem was never that OpenSSL had no money, but that all the money it had was tied to services and features that didn't help improve the libraries overall security and in some cases where even outright harmful.
Historically that was true, but it has improved in the last few years.
This is really the same for most enterprise software. Customers pay for features , I guess they think the security is implied, but more features at a rapid pace typically ruins long term security.
Honest question: shall free software licenses be changed to force large corporations to sponsor free software projects adequately?
Then it wouldn't be free software anymore.
This is backwards imo, it is related to people that chose to build these projects. Everyone might have a different decision on it.
Yes, they should do this. Howvee, they'd stop being considered free software and open source.
Alternatively they can put such large share-back requirements that corporations don't want to use it and would rather pay for an alternate license. SSPL and to a lesser extent AGPL achieve this.
A free software license is by definition a license that doesn't place requirements on anyone to pay for it, sponsor it, contribute or otherwise. If a license didn't grant these freedoms, it wouldn't be a free software license.
What you're asking is, should widely used libraries be proprietary software sold for money? And that is indeed an honest question. There's a huge ecosystem of developers out there who say yes, there's nothing special about that idea at all.
For example, on Apple, Google and Microsoft platforms you don't use OpenSSL when writing native apps. You use the vendor's cryptography APIs. The licenses of Windows and macOS force all users large or small to pay for them. On Android, the native crypto provider is Conscrypt:
https://github.com/google/conscrypt/
... which is open source, but you will notice that it doesn't accept outside contributions and is funded by revenue from Android's proprietary components licensing. So that's very close to the model you're suggesting: it's "open source" but in practice large corporations pay Google lots of money for the privilege of shipping it.
Moreover Google fund BoringSSL and other forks of OpenSSL. The issue here isn't funding levels (anymore). The issue is that OpenSSL is written in C, and not very good C at that.
I don't see any restrictions in CONTRIBUTING (https://github.com/google/conscrypt/blob/master/CONTRIBUTING...).
Minor nitpick, but at least as far as the GNU GPL goes, it explicitly allows you to charge for software distribution when the software came bundled with a tangible item (like the firmware of an IoT device). The distributor can require you to pay them to get it. This was more common in the old days when software mostly came on CD-ROMs and you'd be getting another physical item in the mail, but the provision remains in the GPLv3 under section 6(b)(1) as an option that the distributor can require.
It has to be the reasonable costs of distribution. It can't just be an arbitrary fee.
Not exactly. You can charge whatever you like for the software - but whoever buys it from you is then allowed to give it away for free to anyone else, so this is only a theoretical possibility.
The thing you're allowed to charge for but only a reasonable amount is to provide the sources to software you have previously distributed, for free or otherwise. If I get some binaries from you and then request the sources too, under the GPL, you are forced to provide them, but you can ask for a fee to cover the transfer costs.
Many important open source projects have already switched to Business Source License. However, they only switched after they became famous. I have not seen any projects that became famous that started with BSL.
I often hear this criticism and "bait-and-switch" is also mentioned. At the same time, when you have a small project or tool only used by a handful of people the risk of it being hacker or not responding in time is very different to a project which is essential to the whole world. I don't now what a realistic solution would look like though.
> Many important open source projects have already switched to Business Source License.
No: first not "many", but very few. And second they are proprietary projects.
Some of us have the radical opinion that the definition of open source isn't what a specific organization (with no trademark on the term) dictates, but whether the source is open.
Other people express strong opposition to the idea that "open" might just mean the same thing as "available" in the context of open source. None of those people have proposed a more coherent definition that I've seen of what that word means in that specific phrase though.
"Open Source" and "Free Software" are both well-established terms with well-documented histories of debate and consensus-forming among huge and healthy communities over a span of decades, and organizations that steward the meanings of those terms. Just because you don't like them doesn't mean you get to pretend you're the first person to try to define the word "open" for software.
[dead]
I certainly think so, but we who think this get shouted down because "It's not Open Source(TM)".
Have you ever wondered who's in charge of choosing what counts as Open Source(TM)? Look no farther: https://opensource.org/sponsors
Yes, that Automattic. Yes, that Google, AWS and Cisco. Yes, that Microsoft. There are one or two entities in this list I don't mind but... damn.
It's time to retire the phrase "open source" and go back to "free software". Not many people know "open source" was originally coined by corporations who didn't want anything to do with freedom.
The FSF would say the exact same. So this point is moot, unless you think Richard Stallman is a corporate shill.
Not many people seem to know that the definition of "open source" they use (the OSD) is actually the definition of "free software" Debian uses (the DFSG) with the serial numbers filed off.
The problem with "free software" lies exclusively with the English language; other languages don't have the same issue due to having separate words for "freedom" and "zero price".
That's why some prefer calling it "libre software."
Also see LibreOffice.
Debian is free to use any inclusion criteria it wants - especially since it still includes software thst doesn't meet the criteria. It shouldn't become the global default definition of "open source" though.
> BTW, OpenSSL seems to be a rather low quality implementation of a very low quality set of APIs. Stay away.
Well sure, but the world doesn't seem to be able to agree on, or put resources into, a strong contender.
I'm usually for heterogeneity and diversity in the software ecosystem, but for cryptography I'd be happy with a single high-quality "anointed path". Because cryptography.
other commenters mentioned boringssl (used by Google everywhere it applies), LibreSSL (used in BSD), rustls (used in Rust). Go has its own implementation just called crypto/tls.
> boringssl
"Although BoringSSL is an open source project, it is not intended for general use, as OpenSSL is. We don't recommend that third parties depend upon it. Doing so is likely to be frustrating because there are no guarantees of API or ABI stability." (Their README).
> LibreSSL
That does indeed still seem like the most promising avenue. But it is "just" a fork of OpenSSL.
> rustls
It relies on one of two crates for cryptographic primitives, both of which have zero regard for long term support. "Latest release or bust" doesn't cut it for crypto.
You do need to keep TLS up to date as the network changes around you. Just recently you needed to block eMudhra intermediate certificates.
> You do need to keep TLS up to date as the network changes around you.
Yes. Which is precisely why you need cryptographic libraries that take long term support seriously and don't just fix whatever the latest version is.
> Just recently you needed to block eMudhra intermediate certificates.
Why on earth is your TLS library responsible for your CA certificate database?!
Because it's a TLS feature? That's what people want from TLS libraries? They say connect, and if it succeeds it's secure?
But the database of trusted CAs is kept and maintained independently of the TLS library, obviously! It's data, not code. And it's agnostic as to which library uses it.
Any comments on my clarification about your misunderstanding of long term support?
> BTW, OpenSSL seems to be a rather low quality implementation of a very low quality set of APIs. Stay away.
We’ve known this for well over two decade since heartbleed yet for some reason Linux distros haven’t migrated off of it
Not breaking userspace is more important than security………
The forks of OpenSSL are fairly API compatible with upstream, especially if you're not using new and exotic features. So for many projects you can just link to a better maintained fork and be done.
Yeah, we all know users would rather be reliably hacked than unreliably safe.
Users would rather risk getting hacked than have a computer that doesn't boot or doesn't go on the internet. If they preferred the latter they'd unplug it.
Correct. You can plan around reliable insecurity and stay safe. You cannot ensure safety (or security) of unreliable software.
"not breaking userspace" refers to kernel development. OpenSSL is userspace.
> OpenSSL is userspace.
If only it kept itself to that…
Paraphrasing to "don't make breaking API/ABI changes for projects consuming your API/ABI" doesn't materially change a reasonable point. (And I generally agree with that point, although I confess openssl specifically really strains my view)
And this is why the Linux Desktop will never go mainstream. You should never break apps at all, the OS's API surface goes beyond just the kernel.
The Linux desktop is already mainstream. It has been a first class GUI experience for awhile now. Not to mention the endless papercuts in the proprietary OS's stemming from enshittification (Windows generally) or absurdly poor design (hello Finder).
For those too young to remember 2014, Heartbleed led to both BoringSSL and LibreSSL being forked from OpenSSL.
TIL it's 2036
To be fair, frustrations with OpenSSL long preceded Heartbleed. Heartbleed was just the final straw for those who forked libressl and BoringSSL.
And to be fair to OpenSSL, it was one of the first widely used crypto libraries, along with Netscape's NSS. Setting aside the reasonablebess of the initial choices, the rapid uptake meant those choices were difficult if not impossible to completely unwind, particularly the style and typing of parameters and error reporting. Even today debates rage about signed vs unsigned, etc, in C and other languages. (But suffice it to say there are still alot of unforced errors, IMO, such as implicit output buffer sizes.)
And also led to Amazon's s2n-tls and libcrypto being released around 2015 too as I call.
EDIT: found the release announcement - June 2015 - https://aws.amazon.com/blogs/security/introducing-s2n-a-new-...
AWS-LC[0] itself based on BoringSSL, it makes it feel safer for those behind ALBs (but looks like this CVE may not be TLS related).
[0]: https://github.com/aws/aws-lc
Edit: corrected fork to based on, I don't actually know if it was a fork and a repo just says based on
I wonder. Realistically, Boring and Libre haven't taking off. I measure that by most use and default deploys. But with LLMs, we may be entering a new world.
In that world, certainly some projects may be impervious to fixes. We've all seen projects that basically eschew all change, and ignore issue trackers, and are basically dead.
But the ones which still live? Eventually this first wave of LLM derived fixes will get through the old churn. Presumably, code will improve dramatically. If that's so, OpenSSL will become "good code". And the same with old legacy projects which are still hanging around.
Meaning... will legacy simply exist forever? Our two forks above will likely fade away, if openssl's code becomes clean as can be.
We used BoringSSL at Cloudflare. I imagine there’s a lot of usage in places you can’t see despite it not seeing more adoption. One place you can see it is all browsers use it since Chromium bundles it everywhere except Apple but Apple’s OSes use it to underpin their Network.framework which underpins WkWebview so all Apple devices use it.
> all browsers use it since Chromium bundles it
Doesn't Firefox use NSS instead?
Seriously, at this point BoringSSL is surely more widely deployed than OpenSSL? A billion Apple devices, plus every Chrome installation?
BoringSSL is somewhat widely used because Google baked it into their projects (and partially by Microsoft, I believe they made a FIPS variant of it and used it in Go?)
LibreSSL had, IIRC, problem known as OpenBSD-itis. Or to put it bluntly, the developers didn't really care about any other platform. Plus I will admit that being the first and loud "we will fix openssl" they got linked with some loud PR that showed less understanding of code they mocked than OpenSSL developer who wrote it, and that's really impressive when you consider some of the mocked code fragments were results of developers never reading foundational documentation for a platform
> LibreSSL had, IIRC, problem known as OpenBSD-itis. Or to put it bluntly, the developers didn't really care about any other platform.
Well put. This is the crux of the issue.
If you ever try to use libreSSL you rub up against so much friction that most projects would hammer out early.
There's LibreSSL portable. It worked better in legacy machines I own, go figure.
I just don't get it; this is a task to hand to an LLM, "go rebuild OpenSSL, use Rust (no mistakes)" and have it churn through a greenfield implementation. Then open a PR =). Why hasn't this happened yet? The fact that the LLM will write tests and deal with all the compiler errors probably puts it lightyears ahead.
you don't get credit for that because it is clear you are not very skillful at what AI-bros value most, namely manipulating other nerds...
if you look at the commit history of OpenSSL, you can easily notice how a lot of people started contributing once their clanker got up to speed. They are very happy to rip out weird C and replace it with assembler or intrinsics. Rip out lists of characters formatted as a single string and replace them with array syntax, each char quoted.
LLMs have a bottleneck, human review. And human review is absolutely required at this stage (the future is unknown). A rust built LLM only reviewed SSL library would be a recipe for disaster.
Beyond that, when it comes to exceptionally complex APIs, forward thinking implementation, and revising the 'way things are done'? These are weak points for LLMs. If you don't know how to write OpenSSL(or its replacement), and don't understand all its quirks, and the entire industry surrounding it, you cannot be the human driving the LLM successfully.
There is no way for a human to catch all the design mistakes an LLM will make, otherwise.
People seem to get upset, when I say this, for reasons I cannot fathom. However, we need to have a mental model of how we trust LLM output, and the 'new junior' is great for that. Especially considering that every day, is a new day for an LLM, its context window always loses yesterday.
So consider LLMs as an eager, intelligent, incredibly well read junior, just out of college. And on their third day of work.
Zero real world experience. Constant review required.
Would you want that junior to design, build, and write test scripts for a library to underpin the entire internet?
> A rust built LLM only reviewed SSL library would be a recipe for disaster.
There was a great quote at this year's OpenSSL conference that went something like "LLMs if left to their own devices will reinvent sendmail".
Is that Zawinsky's law of LLMs?
There’s already a Rust TLS library, rustls, that’s relatively mature (10 years in development, a couple of paid maintainers, robust development and release processes). It’s not a drop-in replacement for OpenSSL, though. There’s a companion project to build an OpenSSL-compatible wrapper API for it, but that’s not as far along. Interestingly, while rustls has a plug-in API that lets you choose your favorite Rust encryption library for the low-level operations (aws-lc-rs, Ring, Graviola, etc), those libraries all tend to contain a lot of assembly … not just for speed, but because it’s easier to avoid side channel timing vulnerabilities in assembly.
The side channels issue is where rust crypto libraries are currently the weakest
Can you explain why it is easier to avoid side-channel timing attacks in assembly?
I guess it's easier to control the timing of your operations if you write them directly in assembly. Otherwise, the compiler optimizations could change the code in unexpected ways.
Compilers like to make code go brrr as much as they can. With crypto you want the code to run at a deterministic speed (execution time), regardless of input. With hand written assembly you can (more easily) make sure you have the same cycle count regardless of what the input is.
At a guess, because it is easier to control timing in assembly than in a higher level language that will compile your source code into whatever shape it feels like
It's still impossible to control actual timing, but easy to ensure control flow isn't data-dependent, which is what you need. Classic example is memcmp. Although you can write the or-then-check-zero memcmp in any high level language and it's very unlikely to compile to data-dependent timing unless your compiler is AGI.
> Although you can write the or-then-check-zero memcmp in any high level language and it's very unlikely to compile to data-dependent timing unless your compiler is AGI.
No need for "AGI", the compiler just needs to have enough value tracking to know that the value saturates (once a bit is set, it never gets unset), and an heuristic to move the comparison inside the loop when it might allow it to do less loop iterations.
(I wrote a Rust crate to do precisely that kind of "constant time memcmp", and the only truly reliable way to do so without help of the compiler, is by using inline assembly as a "black box" to make a few things invisible to the compiler's optimizer. It was inspired by code in the Linux kernel which did the same thing also using inline assembly as a "black box".)
> EVP_EncodeBlock() encodes a full block of input data in f and of length dlen and stores it in t. For every 3 bytes of input provided 4 bytes of output data will be produced. If dlen is not divisible by 3 then the block is encoded as a final block of data and the output is padded such that it is always divisible by 4. Additionally a NUL terminator character will be added. For example if 16 bytes of input data is provided then 24 bytes of encoded data is created plus 1 byte for a NUL terminator (i.e. 25 bytes in total). The length of the data generated without the NUL terminator is returned from the function.
My god, you only see this type of horrendous API in C don't you.
"No way to prevent this" say users of only language where this regularly happens
It makes lot of sense when you think what encoding here does. Changes 3 arbitrary bytes to 4 bytes with only some values being possible. Null termination is one issue, but the output is a string such only length of the text itself makes sense.
> My god, you only see this type of horrendous API in C don't you.
This is the appropriate place to put a link to this pair of classics on API design:
"How Do I Make This Hard to Misuse?" https://ozlabs.org/~rusty/index.cgi/tech/2008-03-30.html
"What If I Don't Actually Like My Users?" https://ozlabs.org/~rusty/index.cgi/tech/2008-04-01.html
gets belongs in category -10
What's "horrendous"? I don't see anything which stands out as being wrong. It specifies exactly the lengths of input and output.
gets(), on the other hand, that is "horrendous".
The thing that jumps out to me is that it returns “the size of the output data minus one” but if I’m passing in a pointer it’s on me to make sure there’s enough memory allocated isn’t it? So the caller must make sure it’s calculated this correctly - in which case why return the size? Taking the size as an argument could be safer as you’d be checking the caller calculated enough space too.
(Honest questions in this it’s probably been 20 years since I did c)
That part of the API seems particularly random indeed, especially when comparing it to other EVP_{En,De}code* functions which use the returned int for success/error.
The only reason for such a design I could think of is that the caller would be able to "detect" the overflow by comparing the returned value against the length of the buffer. But that is of course pointless, as the compiler is free to assume a buffer overflow cannot happen and thus optimize the check away...
Even in C, if you have a function that takes an input buffer and output buffer of differing lengths, the right API has the caller pass both lengths.
But yes, gets is worse.
takes an input buffer and output buffer of differing lengths, the right API has the caller pass both lengths.
If there is no fixed relationship between the two, then that makes sense; but in this case there is a very simple and well-defined relationship between them. But to specify two lengths which must always agree; and if not, then what? You've effectively invented an additional artificial error condition, an additional branch, and burdened all the callers up the chain to do something, which is a treacherous path to complexity and the production of even more bugs.
Or are you just preaching the same mindless idiocy that lead to Microsoft's memcpy_s() (the "s" is for "stupid")?
The artificial error condition is the whole point and the hoops you have to jump through to be able to reach it as you say are reasons to not use C because it’s hard to enforce invariants like this and easy to ignore them with mostly working code. Mostly working was fine last year, it isn’t anymore.
Turning this into an attack on the language with otherwise nonsense arguments is how I know you're an ideological bully not worth arguing with, just like the ones who called this nothingburger a 9.8.
Perhaps I'm just used to having programmed in C# professionally for a little bit, but these sorts of APIs where you have to do a calculation to determine the destination buffer size make me feel like a caveman. A minuscule typo and you get a CVE like this.
I know, I know, C is low level and high performance, but surely in the year of our lord 2026, we could have come out with something simpler.
Horrendous is a strong word, you're right.
Not even that is usable as excuse, only those that never tried out the systems programming languages that either predated C, or where equally developed during the 1970's outside Bell Labs and UNIX ecosystem.
You could have come with something simpler in PL/I.
You should be comparing with assembly language, not C#.
The proper way to do it is to factor out the length calc into a utility function that the API provides. Also the calculated buffer length is passed in to the function, and an error is returned if it's too small.
One can abstract buffer handling away in C just fine, eliminating all those problems. I don't know why programmers insist on open-coding low-level buffer management.
Yes, of course, this isn’t C, this is just another case of people writing C wrong.
Perhaps we could use another language, where it's not so easy (or possible) to write it wrong?
It is also easy to write it wrong in other languages. The coreutils rewrite had hundreds of issues, cloudflare brought down half of the internet because in the magic language where nothing can go wrong due to the advanced type system someone used an unchecked .unwrap (but this does not count, because this was Rust and not C - so could not possibly be the fault of the language, so only the programmer is to blame), and just recently Microsoft found a couple of memory safety issues in Rust's standard library. And let's better not talk about the supply chain.