In association with the release of curl 8.4.0, we publish a security advisory and all the details for CVE-2023-38545. This problem is the worst security problem found in curl in a long time. We set it to severity HIGH.

While the advisory contains all the necessary details. I figured I would use a few additional words and expand the explanations for anyone who cares to understand how this flaw works and how it happened.

Background

curl has supported SOCKS5 since August 2002.

SOCKS5 is a proxy protocol. It is a rather simple protocol for setting up network communication via a dedicated “middle man”. The protocol is for example typically used when setting up communication to get done over Tor but also for accessing Internet from within organizations and companies.

SOCKS5 has two different host name resolver modes. Either the client resolves the host name locally and passes on the destination as a resolved address, or the client passes on the entire host name to the proxy and lets the proxy itself resolve the host remotely.

In early 2020 I assigned myself an old long-standing curl issue: to convert the function that connects to a SOCKS5 proxy from a blocking call into a non-blocking state machine. This is for example much noticeable when an application performs a large amount of parallel transfers that all go over SOCKS5.

On February 14 2020 I landed the main commit for this change in master. It shipped in 7.69.0 as the first release featuring this enhancement. And by extension also the first release vulnerable to CVE-2023-38545.

A less wise decision

The state machine is called repeatedly when there is more network data to work on until it is done: when the connection is established.

At the top of the function I made this:

bool socks5_resolve_local =
  (proxytype == CURLPROXY_SOCKS5) ? TRUE : FALSE;

This boolean variable holds information about whether curl should resolve the host or just pass on the name to the proxy. This assignment is done at the top and thus for every invocation while the state machine is running.

The state machine starts in the INIT state, in which the main bug for today’s story time lies. The flaw is inherited from the function from before it was turned into a state-machine.

if(!socks5_resolve_local && hostname_len > 255) {
  socks5_resolve_local = TRUE;
}

SOCKS5 allows the host name field to be up to 255 bytes long, meaning a SOCKS5 proxy cannot resolve a longer host name. On finding a too long host name. the curl code makes the bad decision to instead switch over to local resolve mode. It sets the local variable for that purpose to TRUE. (This condition is a leftover from code added ages ago. I think it was downright wrong to switch mode like this, since the user asked for remote resolve curl should stick to that or fail. It is not even likely to work to just switch, even in “good” situations.)

The state machine then switches state and continues.

The issue triggers

If the state machine cannot continue because it has no more data to work with, like if the SOCKS5 server is not fast enough, it returns. It gets called again when there is data available to continue working on. Moments later.

But now, look at the local variable socks5_resolve_local at the top of the function again. It again gets set to a value depending on proxy mode – not remembering the changed value because of the too long host name. Now it again holds a value that says the proxy should resolve the name remotely. But the name is too long…

curl builds a protocol frame in a memory buffer, and it copies the destination to that buffer. Since the code wrongly thinks it should pass on the host name, even though the host name is too long to fit, the memory copy can overflow the allocated target buffer. Of course depending on the length of the host name and the size of the target buffer.

Target buffer

The allocated memory area curl uses to build the protocol frame in to send to the proxy, is the same as the regular download buffer. It is simply reused for this purpose before the transfer starts. The download buffer is 16kB by default but can also be set to use a different size at the request of the application. The curl tool sets the buffer size to 100kB. The minimum accepted size is 1024 bytes.

If the buffer size is set smaller than 65541 bytes this overflow is possible. The smaller the size, the larger the possible overflow.

Host name length

A host name in a URL has no real size limit, but libcurl’s URL parser refuses to accept names longer than 65535 bytes. DNS only accepts host names up 253 bytes. So, a legitimate name that is longer than 253 bytes is unusual. A real name that is longer than 1024 is virtually unheard of.

Thus it pretty much requires a malicious actor to feed a super-long host name into this equation to trigger this flaw. To use it in an attack. The name needs to be longer than the target buffer to make the memory copy overwrite heap memory.

Host name contents

The host name field of a URL can only contain a subset of octets. A range of byte values are plain invalid and would cause the URL parser to reject it. If libcurl is built to use an IDN library, that one might also reject invalid host names. This bug can therefore only trigger if the right set of bytes are used in the host name.

Attack

An attacker that controls an HTTPS server that a libcurl using client accesses over a SOCKS5 proxy (using the proxy-resolver-mode) can make it return a crafted redirect to the application via a HTTP 30x response.

Such a 30x redirect would then contain a Location: header in the style of:

Location: https://aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/

… where the host name is longer than 16kB and up to 64kB

If the libcurl using client has automatic redirect-following enabled, and the SOCKS5 proxy is “slow enough” to trigger the local variable bug, it will copy the crafted host name into the too small allocated buffer and into the adjacent heap memory.

A heap buffer overflow has then occurred.

The fix

curl should not switch mode from remote resolve to local resolve due to too long host name. It should rather return an error and starting in curl 8.4.0, it does.

We now also have a dedicated test case for this scenario.

Credits

This issue was reported, analyzed and patched by Jay Satiro.

This is the largest curl bug-bounty paid to date: 4,660 USD (plus 1,165 USD to the curl project, as per IBB policy)

Rewrite it?

Yes, this family of flaws would have been impossible if curl had been written in a memory-safe language instead of C, but porting curl to another language is not on the agenda. I am sure the news about this vulnerability will trigger a new flood of questions about and calls for that and I can sigh, roll my eyes and try to answer this again.

The only approach in that direction I consider viable and sensible is to:

1. allow, use and support more dependencies written in memory-safe languages and
2. potentially and gradually replace parts of curl piecemeal, like with the introduction of hyper.

Such development is however currently happening in a near glacial speed and shows with painful clarity the challenges involved. curl will remain written in C for the foreseeable future.

Everyone not happy about this are of course welcome to roll up their sleeves and get working.

Including the latest two CVEs reported for curl 8.4.0, the accumulated total says that 41% of the security vulnerabilities ever found in curl would likely not have happened should we have used a memory-safe language. But also: the rust language was not even a possibility for practical use for this purpose during the time in which we introduced maybe the first 80% of the C related problems.

It burns in my soul

Reading the code now it is impossible not to see the bug. Yes, it truly aches having to accept the fact that I did this mistake without noticing and that the flaw then remained undiscovered in code for 1315 days. I apologize. I am but a human.

It could have been detected with a better set of tests. We repeatedly run several static code analyzers on the code and none of them have spotted any problems in this function.

In hindsight, shipping a heap overflow in code installed in over twenty billion instances is not an experience I would recommend.

Behind the scenes

To learn how this flaw was reported and we worked on the issue before it was made public. Go check the Hackerone report.

On Scott Adams

I use his “I’m going to write myself a minivan”-strip above because it’s a classic. Adams himself has turned out to be a questionable person with questionable opinions and I do not condone or agree with what he says.

On August 26 I posted details here on my blog about the bogus curl issue CVE-2020-19909. Luckily, it got a lot of attention and triggered discussions widely. Maybe I helped shed light on the brittleness of this system. This was not a unique instance and it was not the first time it happened. This has been going on for years.

For example, the PostgreSQL peeps got a similarly bogus CVE almost at the same time. Not a real problem either.

UB

Some people did in discussions following my blog post insist that a signed long overflow is undefined behavior in C and since it truly is undefined all bets are off so this could be a security issue.

I am not a fan of philosophical thought exercises around vulnerabilities. They are distractions from the real matters and I find them rather pointless. It is easy to test how this flaw plays out on numerous platforms using numerous compilers. It’s not a security problem on any of them. Even if the retry time would go down to 0, it is not a DOS to do multiple internet transfers next to each other. If it would be, then a browser makes a DOS every time you visit a website – and curl does it when you give it two URLs on the same command line. Unexpected behavior and a silly bug, yes. Security flaw: no.

Not to even mention that no person or script actually would use such numbers on the command line that trigger the bug. This is the bug that never triggered for anyone who did not deliberately try to make it happen.

Who gets to report a CVE

A CVE Id is basically just a bug tracker identifier. You can ask for a number for a problem you found, and by the nature of this you don’t even have to present many details when you ask for it. You are however supposed to provide them when the issue goes public. CVE-2020-19909 certainly failed this latter step. You could perhaps ask why MITRE even allows for CVEs to get published at all if there isn’t more information available.

When you request a CVE Id, there are a few hundred distributed CVE number authorities (CNAs) you can get the number from.

CNA

A CNA can (apparently) require to be consulted if the ask is in regards to their scope. To their products.

I have initiated work on having curl become a CNA; to take responsibility for all CVEs that concern curl or libcurl. I personally think it is rather problematic that this basically is the only available means to prevent this level of crap, but I still intend to attempt this route.

Lots of other Open Source projects (Apache, OpenSSL, Python, etc) already are CNAs, many of them presumably because they realized these things much faster than me. I suspect curl is a smaller shop than most of the existing CNAs, but I don’t think size matters here.

It does of course add administration and work for us which is unfortunate, because that time and energy could be spent on more productive activities, but it still feels like a reasonable thing to try. Hopefully this can also help combat inflated severity levels.

MITRE: Reject 2020-19909 please

The main care-taker of the CVE database is MITRE. They provide a means for anyone to ask for updates to already published CVEs. I submitted my request for rejection using their online form, explaining that this issue is not a security problem and thus should be removed again. With all the info I had, including a link to my blog post.

They denied my request.

Their motivation? The exact quote:

After review there are multiple perspectives on whether the issue information is helpful to consumers of the CVE List, our current preference is in the direction of keeping the CVE ID assignment. There is a valid weakness (integer overflow) that can lead to a valid security impact (denial of service, based on retrying network traffic much more often than is documented/requested).

This is signed CVE Assignment Team, no identifiable human involved.

It is not a Denial of Service. It’s blatant scaremongering and it helps absolutely no one to say that it is.

NVD: Rescore 2020-19909 please

NVD maintains some sort of database and index with all the CVE Ids. They score them and as you may remember, they decided this non-issue is a 9.8 out of 10 severity.

The same evening I learned about this CVE I contacted NVD (nvd@nist.gov), pointed out this silliness and asked them to act on it.

A few days later they responded. From the National Vulnerability Database Team, no identifiable human.

First, they of course think their initial score was a fair assessment:

At the time of analysis, information for this CVE record was particularly sparse. Per NVD policy, if any information is lacking, unclear or conflicts between sources, the NVD policy is to represent the worst-case scenario. The NVD takes this conservative approach to avoid under reporting the possible severity of a given vulnerability.

Meaning: when they don’t know, they make up the worst possible scenario. That is literally what they say and how they act. Panic by default.

However, at this time there are more information sources and clarifications provided that make amendment of the CVSS vector string appropriate. We will break down the various amendments for transparency and to assist in confirmation of the appropriateness of these changes by you or a member of your team.

Meaning: their messed up system requires us to react and provide additional information to them to trigger a reassessment and them updating the score. To lower the panic level.

NVD themselves will not remove any CVE. As long as MITRE says it is a genuine issue, NVD shows it.

Rescore

As a rejection of the CVE was denied, the secondary goal would be to lower NVD’s severity score as much as possible.

The NVD response helped me understand how their research into and analyses of issues are performed.

Their email highlighted three specific items in the CVSS score and how they could maybe re-evaluate them to a lower setting. How do they base and motivate that? Providing links to specific reddit discussions for details of the CVE-2020-19909 problems. Postings done by the ordinary set of pseudonymous persons. Don’t get me wrong, some of those posts are excellent and clearly written by skilled and knowledgeable persons. It just seems a bit… arbitrary.

I did not participate in those discussions and neither did any curl core contributor to my knowledge. There are simply no domain experts involved there.

No reading or pointing to code. No discussions on data input and what changing retry delays mean and not a single word on UB. No mention on why no or a short delay would imply DOS.

Eventually, their proposed updated CVSS ends up with a new score of…

3.3

To be honest, I find CVSS calculations exhausting and I did not care to evaluate their proposal very closely. Again, this is not a security problem.

If you concur with the assessment above based on publicly available information this would change the vector string as follows:

AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H - 9.8
AV:L/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:L - 3.3

In my response to NVD I wrote:

I have not personally checked the scoring very closely. I do not agree with the premise that this is a security problem. It is bad, wrong and hostile to insist that non-issues are security problems. It is counter-productive for the entire industry.

As they set it to 3.3 it at least takes the edge off this silliness and I feel a little bit better. They now also link to my blog post from their CVE-2020-19909 info page, which I think is grand of them.

This is a story consisting of several little building blocks and they occurred spread out in time and in different places. It is a story that shows with clarity how our current system with CVE Ids and lots of power given to NVD is a completely broken system.

CVE-2020-19909

On August 25 2023, we got an email to the curl-library mailing list from Samuel Henrique that informed us that “someone” had recently created a CVE, a security vulnerability identification number and report really, for a curl problem.

I wanted to let you know that there's a recent curl CVE published and it doesn't look like it was acknowledged by the curl authors since it's not mentioned in the curl website: CVE-2020-19909

We can’t tell who filed it. We just know that it is now there.

We own our curl issues

In the curl project we work hard and fierce on security and we always work with security researchers who report problems. We file our own CVEs, we document them and we make sure to tell the world about them. We list over 140 of them with every imaginable detail about them provided. We aim at providing gold-level documentation for everything and that includes our past security vulnerabilities.

That someone else suddenly has submitted a CVE for curl is a surprise. We have not been told about this and we would really have liked to. Now there is a new CVE out there reporting a curl issue and we have no details to say about it on the website. Not good.

I bet curl users soon would like to know the details about this.

Wait 2020?

The new CVE has an ID containing 2020 and that is weird. When you register a CVE you typically get it with the year you request it. Unless you get an ID for an old problem of the past. Is that what they did?

Sources seem to indicate that this was published just days ago.

What is this CVE?

Of course the top link when you search for this CVE is to NVD. Not the most reliable organization, but now we can’t be too picky. On their site they explain this with very few details:

Integer overflow vulnerability in tool_operate.c in curl 7.65.2 via crafted value as the retry delay.

And then the craziest statement of the year. They grade it a 9.8 CRITICAL issue. With 10 as a maximum, this is close to the worst case possible, right?

The code

Let’s pause NVD in their panic state for a moment because I immediately recognized this description. Brief as it is.

I spend a lot of time in the curl security team receiving reports, reviewing reports, reviewing source code, assessing claims and figuring out curl security issues. I had seen this claim before!

On July 27, 2019, a Jason Lee file an issue on hackerone, where he reported that there was an integer overflow problem in curl’s --retry-delay command line option. The option accepts number of seconds and then internally converts to milliseconds by multiplying the value by 1000. The option sets how long time curl should wait until it makes a retry if the previous transfer failed with a transient error.

This means that on a 64 bit machine, if you write

curl --retry-delay 18446744073709552 ...

The number will overflow the math and instead of waiting until the end of the universe, it might retry again within the next few seconds. The above example apparently made it 384 seconds instead. On Windows, which uses 32 bit longs, you can get the problem already by asking for more than two million seconds (roughly 25 days).

A bug, sure. Security problem? No. I told Jason that in 2019 and then we closed the security report. I then filed a pull-request and fixed the bug. Credits to Jason for the report. We moved on. The fix was shipped in curl 7.66.0, released in September 2019.

Grading issues

In previous desperate attempts from me to reason with NVD and stop their scaremongering and their grossly inflating the severity level of issues, they have insisted that they take in all publicly available data about the problem and make an assessment.

It was obvious already before that NVD really does not try very hard to actually understand or figure out the problem they grade. In this case it is quite impossible for me to understand how they could come up with this severity level. It’s like they saw “integer overflow” and figure that wow, yeah that is the most horrible flaw we can imagine, but clearly nobody at NVD engaged their brains nor looked at the “vulnerable” code or the patch that fixed the bug. Anyone that looks can see that this is not a security problem.

The issue listed by NVD even links to my pull request I mention above. There is no doubt that it is the exact same bug they refer to.

Spreading like a virus

NVD hosts a CVE database and there is an entire world and eco system now that pulls the records from them.

NVD now has this CVE-2020-19909 entry in there, rated 9.8 CRITICAL and now this disinformation spreads across the world. Now when we search for this CVE number we find numerous sites that repeats the same data. “This is a 9.8 CRITICAL problem in curl” – when it is not.

I will object

I learned about this slap in my face just a few hours ago (and I write this past Friday midnight), but I intend to do what I can to reject this CVE.

Update: I’m glad to see the Ubuntu took the lead and marked it as not-affected.

Update2: MITRE denied my request to reject the CVE. The full reason from them is now included in my description of CVE-2020-19909 on the curl site.

Update3: a follow-up post.

Discuss

Hacker news. Reddit. Lobste.rs.