Tag Archives: Security

6 hours of bliss

I sent out the release announcement for curl 7.52.0 exactly 07:59 in the morning of December 21, 2016. A Wednesday. We typically  release curl on Wednesdays out of old habit. It is a good release day.

curl 7.52.0 was just as any other release. Perhaps with a slightly larger set of new features than what’s typical for us. We introduce TLS 1.3 support, we now provide HTTPS-proxy support and the command line tool has this option called –fail-early that I think users will start to appreciate once they start to discover it. We also  announced three fixed security vulnerabilities. And some other good things.

I pushed the code to git, signed and uploaded the tarballs, I updated the info on the web site and I sent off that release announcement email and I felt good. Release-time good. That short feeling of relief and starting over on a new slate that I often experience these release days. Release days make me happy.

Any bets?

It is not unusual for someone to find a bug really fast after a release has shipped. As I was feeling good, I had to joke in the #curl IRC channel (42 minutes after that email):

08:41 <bagder> any bets on when the first bug report on the new release shows up? =)

Hours passed and maybe, just maybe there was not going to be any quick bugs filed on this release?

But of course. I wouldn’t write this blog post if it all had been nice and dandy. At 14:03, I got the email. 6 hours and 4 minutes since I wrote the 7.52.0 announcement email.

The email was addressed to the curl project security email list and included a very short patch and explanation how the existing code is wrong and needs “this fix” to work correctly. And it was entirely correct!

Now I didn’t feel that sense of happiness anymore. For some reason it was now completely gone and instead I felt something that involved sensations like rage, embarrassment and general tiredness. How the [beep] could this slip through like this?

I’ve done releases in the past that were broken to various extents but this is a sort of a new record and an unprecedented event. Enough time had passed that I couldn’t just yank the package from the download page either. I had to take it through the correct procedures.

What happened?

As part of a general code cleanup during this last development round, I changed all the internals to use a proper internal API to get random data and if libcurl is built with a TLS library it uses its provided API to get secure and safe random data. As a move to improve our use of random internally. We use this internal API for getting the nonce in authentication mechanisms such as Digest and NTLM and also for generating the boundary string in HTTP multipart formposts and more. (It is not used for any TLS or SSH level protocol stuff though.)

I did the largest part of the random overhaul of this in commit f682156a4f, just a little over a month ago.

Of course I made sure that all test cases kept working and there were no valgrind reports or anything, the code didn’t cause any compiler warnings. It did not generate any reports in the many clang-analyzer or Coverity static code analyzer runs we’ve done since. We run clang-analyzer daily and Coverity perhaps weekly.

But there’s a valgrind report just here!

Kamil Dudka, who sent the 14:03 email, got a valgrind error and that’s what set him off – but how come he got that and I didn’t?

The explanation consists of the following two conditions that together worked to hide the problem for us quite successfully:

  1. I (and I suppose several of the other curl hackers) usually build curl and libcurl “debug enabled”. This allows me to run more tests, do more diagnostics and debug it easier when I run into problems. It also provides a system with “fake random” so that we can actually verify that functions that otherwise use real random values generate the correct output when given a known random value… and yeah, this debug system prevented valgrind from detecting any problem!
  2. In the curl test suite we once had a problem with valgrind generating reports on third party libraries etc which then ended up as false positives. We then introduced a “valgrind report parser” that would detect if the report concerns curl or something else. It turns out this parser doesn’t detect the errors if curl is compiled without the cc’s -g command line option. And of course… curl and libcurl both build without -g by default!

The patch?

The vulnerable function basically uses this simple prototype. It is meant to get an “int” worth of random value stored in the buffer ‘rnd’ points to. That’s 4 bytes.

randit(struct Curl_easy *data, unsigned int *rnd)

But due to circumstances I can’t explain on anything other than my sloppy programming, I managed to write the function store random value in the actual pointer instead of the buffer it points to. So when the function returns, there’s nothing stored in the buffer. No 4 bytes of random. Just the uninitialized value of whatever happened to be there, on the stack.

The patch that fixes this problem looks like this (with some names shortened to simplify but keep the idea):

- res = random(data, (char *)&rnd, sizeof(rnd));
+ res = random(data, (char *)rnd, sizeof(*rnd));

So yeah. I introduced this security flaw in 7.52.0. We had it fixed in 7.52.1, released roughly 48 hours later.

(I really do not need comments on what other languages that wouldn’t have allowed this mistake or otherwise would’ve brought us world peace a long time ago.)

Make it not happen again

The primary way to make this same mistake not happen again easily, is that I’m removing the valgrind report parsing function from the test suite and we will now instead assume that valgrind reports will be legitimate and if not, work on suppressing the false positives in a better way.


This flaw is officially known as CVE-2016-9594

The real commit that fixed this problem is here, or as stand-alone patch.

The full security advisory for this flaw is here: https://curl.haxx.se/docs/adv_20161223.html

Facepalm photo by Alex E. Proimos.

xkcd: 221


curl security audit

“the overall impression of the state of security and robustness
of the cURL library was positive.”

I asked for, and we were granted a security audit of curl from the Mozilla Secure Open Source program a while ago. This was done by Mozilla getting a 3rd party company involved to do the job and footing the bill for it. The auditing company is called Cure53.

good_curl_logoI applied for the security audit because I feel that we’ve had some security related issues lately and I’ve had the feeling that we might be missing something so it would be really good to get some experts’ eyes on the code. Also, as curl is one of the most used software components in the world a serious problem in curl could have a serious impact on tools, devices and applications everywhere. We don’t want that to happen.

Scans and tests and all

We run static analyzers on the code frequently with a zero warnings tolerance. The daily clang-analyzer scan hasn’t found a problem in a long time and the Coverity once-every-few-weeks occasionally finds something suspicious but we always fix those immediately.

We have  thousands of tests and unit tests that we run non-stop on the code on multiple platforms running multiple build combinations. We also use valgrind when running tests to verify memory use and check for potential memory leaks.


The audit itself. The report and the work on fixing the issues were all done on closed mailing lists without revealing to the world what was really going on. All as our fine security process describes.

There are several downsides with fixing things secretly. One of the primary ones is that we get much fewer eyes on the fixes and there aren’t that many people involved when discussing solutions or approaches to the issues at hand. Another is that our test infrastructure is made for and runs only public code so the code can’t really be fully tested until it is merged into the public git repository.

The report

We got the report on September 23, 2016 and it certainly gave us a lot of work.

The audit report has now been made public and is a very interesting work if you’re into security, C code and curl hacking. I find the report very clear, well written and it spells out each problem very accurately and even shows proof of concept code snippets and exploit examples to drive the points home.

Quoted from the report intro:

As for the approach, the test was rooted in the public availability of the source code belonging to the cURL software and the investigation involved five testers of the Cure53 team. The tool was tested over the course of twenty days in August and September of 2016 and main efforts were focused on examining cURL 7.50.1. and later versions of cURL. It has to be noted that rather than employ fuzzing or similar approaches to validate the robustness of the build of the application and library, the latter goal was pursued through a classic source code audit. Sources covering authentication, various protocols, and, partly, SSL/TLS, were analyzed in considerable detail. A rationale behind this type of scoping pointed to these parts of the cURL tool that were most likely to be prone and exposed to real-life attack scenarios. Rounding up the methodology of the classic code audit, Cure53 benefited from certain tools, which included ASAN targeted with detecting memory errors, as well as Helgrind, which was tasked with pinpointing synchronization errors with the threading model.

They identified no less than twenty-three (23) potential problems in the code, out of which nine were deemed security vulnerabilities. But I’d also like to emphasize that they did also actually say this:

At the same time, the overall impression of the state of security and robustness of the cURL library was positive.

Resolving problems

In the curl security team we decided to downgrade one of the 9 vulnerabilities to a “plain bug” since the required attack scenario was very complicated and the risk deemed small, and two of the issues we squashed into treating them as a single one. That left us with 7 security vulnerabilities. Whoa, that’s a lot. The largest amount we’ve ever fixed in a single release before was 4.

I consider handling security issues in the project to be one of my most important tasks; pretty much all other jobs are down-prioritized in comparison. So with a large queue of security work, a lot of bug fixing and work on features basically had to halt.

You can get a fairly detailed description of our work on fixing the issues in the fix and validation log. The report, the log and the advisories we’ve already posted should cover enough details about these problems and associated fixes that I don’t feel a need to write about them much further.

More problems

Just because we got our hands full with an audit report doesn’t mean that the world stops, right? While working on the issues one by one to have them fixed we also ended up getting an additional 4 security issues to add to the set, by three independent individuals.

All these issues gave me a really busy period and it felt great when we finally shipped 7.51.0 and announced all those eleven fixes to the world and I could get a short period of relief until the next tsunami hits.

a single byte write opened a root execution exploit

Thursday, September 22nd 2016. An email popped up in my inbox.

Subject: ares_create_query OOB write

As one of the maintainers of the c-ares project I’m receiving mails for suspected security problems in c-ares and this was such a one. In this case, the email with said subject came from an individual who had reported a ChromeOS exploit to Google.

It turned out that this particular c-ares flaw was one important step in a sequence of necessary procedures that when followed could let the user execute code on ChromeOS from JavaScript – as the root user. I suspect that is pretty much the worst possible exploit of ChromeOS that can be done. I presume the reporter will get a fair amount of bug bounty reward for this.

The setup and explanation on how this was accomplished is very complicated and I am deeply impressed by how this was figured out, tracked down and eventually exploited in a repeatable fashion. But bear with me. Here comes a very simplified explanation on how a single byte buffer overwrite with a fixed value could end up aiding running exploit code as root.

The main Google bug for this problem is still not open since they still have pending mitigations to perform, but since the c-ares issue has been fixed I’ve been told that it is fine to talk about this publicly.

c-ares writes a 1 outside its buffer

c-ares has a function called ares_create_query. It was added in 1.10 (released in May 2013) as an updated version of the older function ares_mkquery. This detail is mostly interesting because Google uses an older version than 1.10 of c-ares so in their case the flaw is in the old function. This is the two functions that contain the problem we’re discussing today. It used to be in the ares_mkquery function but was moved over to ares_create_query a few years ago (and the new function got an additional argument). The code was mostly unchanged in the move so the bug was just carried over. This bug was actually already present in the original ares project that I forked and created c-ares from, back in October 2003. It just took this long for someone to figure it out and report it!

I won’t bore you with exactly what these functions do, but we can stick to the simple fact that they take a name string as input, allocate a memory area for the outgoing packet with DNS protocol data and return that newly allocated memory area and its length.

Due to a logic mistake in the function, you could trick the function to allocate a too short buffer by passing in a string with an escaped trailing dot. An input string like “one.two.three\.” would then cause the allocated memory area to be one byte too small and the last byte would be written outside of the allocated memory area. A buffer overflow if you want. The single byte written outside of the memory area is most commonly a 1 due to how the DNS protocol data is laid out in that packet.

This flaw was given the name CVE-2016-5180 and was fixed and announced to the world in the end of September 2016 when c-ares 1.12.0 shipped. The actual commit that fixed it is here.

What to do with a 1?

Ok, so a function can be made to write a single byte to the value of 1 outside of its allocated buffer. How do you turn that into your advantage?

The Redhat security team deemed this problem to be of “Moderate security impact” so they clearly do not think you can do a lot of harm with it. But behold, with the right amount of imagination and luck you certainly can!

Back to ChromeOS we go.

First, we need to know that ChromeOS runs an internal HTTP proxy which is very liberal in what it accepts – this is the software that uses c-ares. This proxy is a key component that the attacker needed to tickle really badly. So by figuring out how you can send the correctly crafted request to the proxy, it would send the right string to c-ares and write a 1 outside its heap buffer.

ChromeOS uses dlmalloc for managing the heap memory. Each time the program allocates memory, it will get a pointer back to the request memory region, and dlmalloc will put a small header of its own just before that memory region for its own purpose. If you ask for N bytes with malloc, dlmalloc will use ( header size + N ) and return the pointer to the N bytes the application asked for. Like this:


With a series of cleverly crafted HTTP requests of various sizes to the proxy, the attacker managed to create a hole of freed memory where he then reliably makes the c-ares allocated memory to end up. He knows exactly how the ChromeOS dlmalloc system works and its best-fit allocator, how big the c-ares malloc will be and thus where the overwritten 1 will end up. When the byte 1 is written after the memory, it is written into the header of the next memory chunk handled by dlmalloc:


The specific byte of that following dlmalloc header that it writes to, is used for flags and the lowest bits of size of that allocated chunk of memory.

Writing 1 to that byte clears 2 flags, sets one flag and clears the lowest bits of the chunk size. The important flag it sets is called prev_inuse and is used by dlmalloc to tell if it can merge adjacent areas on free. (so, if the value 1 simply had been a 2 instead, this flaw could not have been exploited this way!)

When the c-ares buffer that had overflowed is then freed again, dlmalloc gets fooled into consolidating that buffer with the subsequent one in memory (since it had toggled that bit) and thus the larger piece of assumed-to-be-free memory is partly still being in use. Open for manipulations!


Using that memory buffer mess

This freed memory area whose end part is actually still being used opened up the play-field for more “fun”. With doing another creative HTTP request, that memory block would be allocated and used to store new data into.

The attacker managed to insert the right data in that further end of the data block, the one that was still used by another part of the program, mostly since the proxy pretty much allowed anything to get crammed into the request. The attacker managed to put his own code to execute in there and after a few more steps he ran whatever he wanted as root. Well, the user would have to get tricked into running a particular JavaScript but still…

I cannot even imagine how long time it must have taken to make this exploit and how much work and sweat that were spent. The report I read on this was 37 very detailed pages. And it was one of the best things I’ve read in a long while! When this goes public in the future, I hope at least parts of that description will become available for you as well.

A lesson to take away from this?

No matter how limited or harmless a flaw may appear at a first glance, it can serve a malicious purpose and serve as one little step in a long chain of events to attack a system. And there are skilled people out there, ready to figure out all the necessary steps.

Update: A detailed write-up about this flaw (pretty much the report I refer to above) by the researcher who found it was posted on Google’s Project Zero blog on December 14:
Chrome OS exploit: one byte overflow and symlinks.

No more heartbleeds please

caution-quarantine-areaAs a reaction to the whole Heartbleed thing two years ago, The Linux Foundation started its Core Infrastructure Initiative (CII for short) with the intention to help track down well used but still poorly maintained projects or at least detect which projects that might need help. Where the next Heartbleed might occur.

A bunch of companies putting in money to improve projects that need help. Sounds almost like a fairy tale to me!


In order to identify which projects to help, they run their Census Project: “The Census represents CII’s current view of the open source ecosystem and which projects are at risk.

The Census automatically extracts a lot of different meta data about open source projects in order to deduce a “Risk Index” for each project. Once you’ve assembled such a great data trove for a busload of projects, you can sort them all based on that risk index number and then you basically end up with a list of projects in a priority order that you can go through and throw code at. Or however they deem the help should be offered.

Which projects will fail?

The old blog post How you know your Free or Open Source Software Project is doomed to FAIL provides such a way, but it isn’t that easy to follow programmatically. The foundation has its own 88 page white paper detailing its methods and algorithm.

Risk Index

  • A project without a web site gets a point
  • If the project has had four or more CVEs (publicly disclosed security vulnerabilities) since 2010, it receives 3 points and if fewer than four there’s a diminishing scale.
  • The number of contributors the last 12 months is a rather heavy factor, which thus could make the index grow old fairly quick. 3 contributors still give 4 points.
  • Popular packages based on Debian’s popcon get points.
  • If the project’s main language is C or C++, it gets two points.
  • Network “exposed” projects get points.
  • some additional details like dependencies and how many outstanding patches not accepted upstream that exist

All combined, this grades projects’ “risk” between 0 and 15.

Not high enough resolution

Assuming that a larger number of CVEs means anything bad is just wrong. Even the most careful and active projects can potentially have large amounts of CVEs. It means they disclose what they find and that people are actually reviewing code, finding problems and are reporting problems. All good things.

Sure, security problems are not good but the absence of CVEs in a project doesn’t say that the project is one bit more secure. It could just mean that nobody ever looked closely enough or that the project doesn’t deal with responsible disclosure of the problems.

When I look through the projects they have right now, I get the feeling the resolution (0-15) is too low and they’ve shied away from more aggressively handing out penalty based on factors we all recognize in abandoned/dead projects (some of which are decently specified in Tom Calloway’s blog post mentioned above).

The result being that the projects get a score that is mostly based on what kind of project it is.

But this said, they have several improvements to their algorithm already suggested in their issue tracker. I firmly believe this will improve over time.

The riskiest ?

The top three projects, the only ones that scores 13 right now are expat, procmail and unzip. All of them really small projects (source code wise) that have been around since a very long time.

curl, being the project I of course look out for, scores a 9: many CVEs (3), written in C (2), network exposure (2), 5+ apps depend on it (2). Seriously, based on these factors, how would you say the project is situated?

In the sorted list with a little over 400 projects, curl is rated #73 (at the time of this writing at least). Just after reportbug but before libattr1. [curl summary – which is mentioning a very old curl release]

But the list of projects mysteriously lack many projects. Like I couldn’t find neither c-ares nor libssh2. They may not be super big, but they’re used by a bunch of smaller and bigger projects at least, including curl itself.

The full list of projects, their meta-data and scores are hosted in their repository on github.

Benefits for projects near me

I can see how projects in my own backyard have gotten some good out of this effort.

I’ve received some really great bug reports and gotten handed security problems in curl by an individual who did his digging funded by this project.

I’ve seen how the foundation sponsored a test suite for c-ares since the project lacked one. Now it doesn’t anymore!


In addition to that, the Linux Foundation has also just launched the CII Best Practices Badge Program, to allow open source projects to fill in a bunch of questions and if meeting enough requirements, they will get a “badge” to boast to the world as a “well run project” that meets current open source project best practices.

I’ve joined their mailing list and provided some of my thoughts on the current set of questions, as I consider a few of them to be, well, lets call them “less than optimal”. But then again, which project doesn’t have bugs? We can fix them!

curl is just now marked as “100% compliance” with all the best practices listed. I hope to be able to keep it like that even with future and more best practices added.

The most popular curl download – by a malware

During October 2015 the curl web site sent out 1127 gigabytes of data. This was the first time we crossed the terabyte limit within a single month.

Looking at the stats a little closer, I noticed that in July 2015 a particular single package started to get very popular. The exact URL was


Curious. In October it alone was downloaded more than 300,000 times, accounting for over 70% of the site’s bandwidth. Why?

The downloads came from what appears to be different locations. They don’t use any HTTP referer headers and they used different User-agent headers. I couldn’t really see a search bot gone haywire or a malicious robot stuck in a crazy mode.

After I shared some of this data over in our IRC channel (#curl on freenode), Björn Stenberg stumbled over this AVG slide set, describing how a particular malware works when it infects a computer. Downloading that particular file is thus a step in its procedures to create a trojan that will run on the host system – see slide 11 for the curl details. The slide also mentions that an updated version of the malware comes bundled with the curl library already, which then I guess makes the hits we see on the curl site being done by the older versions still being run.

Of course, we can’t be completely sure this is the source for the increased download of this particular file but it seems highly likely.

I renamed the file just now to see what happens.

Evil use of good code

We can of course not prevent evil uses of our code. We provide source code and we even host some binaries of curl and libcurl and both good and bad actors are able to take advantage of our offers.

This rename won’t prevent a dedicated hacker, but hopefully it can prevent a few new victims from getting this malware running on their machines.

Update: the hacker news discussion about this post.


SSL padlockI’ve written the http2 explained document and I’ve done several talks about HTTP/2. I’ve gotten a lot of questions about TLS in association with HTTP/2 due to this, and I want to address some of them here.

TLS is not mandatory

In the HTTP/2 specification that has been approved and that is about to become an official RFC any day now, there is no language that mandates the use of TLS for securing the protocol. On the contrary, the spec clearly explains how to use it both in clear text (over plain TCP) as well as over TLS. TLS is not mandatory for HTTP/2.

TLS mandatory in effect

While the spec doesn’t force anyone to implement HTTP/2 over TLS but allows you to do it over clear text TCP, representatives from both the Firefox and the Chrome development teams have expressed their intents to only implement HTTP/2 over TLS. This means HTTPS:// URLs are the only ones that will enable HTTP/2 for these browsers. Internet Explorer people have expressed that they intend to also support the new protocol without TLS, but when they shipped their first test version as part of the Windows 10 tech preview, that browser also only supported HTTP/2 over TLS. As of this writing, there has been no browser released to the public that speaks clear text HTTP/2. Most existing servers only speak HTTP/2 over TLS.

The difference between what the spec allows and what browsers will provide is the key here, and browsers and all other user-agents are all allowed and expected to each select their own chosen path forward.

If you’re implementing and deploying a server for HTTP/2, you pretty much have to do it for HTTPS to get users. And your clear text implementation will not be as tested…

A valid remark would be that browsers are not the only HTTP/2 user-agents and there are several such non-browser implementations that implement the non-TLS version of the protocol, but I still believe that the browsers’ impact on this will be notable.

Stricter TLS

When opting to speak HTTP/2 over TLS, the spec mandates stricter TLS requirements than what most clients ever have enforced for normal HTTP 1.1 over TLS.

It says TLS 1.2 or later is a MUST. It forbids compression and renegotiation. It specifies fairly detailed “worst acceptable” key sizes and cipher suites. HTTP/2 will simply use safer TLS.

Another detail here is that HTTP/2 over TLS requires the use of ALPN which is a relatively new TLS extension, RFC 7301, which helps us negotiate the new HTTP version without losing valuable time or network packet round-trips.

TLS-only encourages more HTTPS

Since browsers only speak HTTP/2 over TLS (so far at least), sites that want HTTP/2 enabled must do it over HTTPS to get users. It provides a gentle pressure on sites to offer proper HTTPS. It pushes more people over to end-to-end TLS encrypted connections.

This (more HTTPS) is generally considered a good thing by me and us who are concerned about users and users’ right to privacy and right to avoid mass surveillance.

Why not mandatory TLS?

The fact that it didn’t get in the spec as mandatory was because quite simply there was never a consensus that it was a good idea for the protocol. A large enough part of the working group’s participants spoke up against the notion of mandatory TLS for HTTP/2. TLS was not mandatory before so the starting point was without mandatory TLS and we didn’t manage to get to another stand-point.

When I mention this in discussions with people the immediate follow-up question is…

No really, why not mandatory TLS?

The motivations why anyone would be against TLS for HTTP/2 are plentiful. Let me address the ones I hear most commonly, in an order that I think shows the importance of the arguments from those who argued them.

1. A desire to inspect HTTP traffic

looking-glassThere is a claimed “need” to inspect or intercept HTTP traffic for various reasons. Prisons, schools, anti-virus, IPR-protection, local law requirements, whatever are mentioned. The absolute requirement to cache things in a proxy is also often bundled with this, saying that you can never build a decent network on an airplane or with a satellite link etc without caching that has to be done with intercepts.

Of course, MITMing proxies that terminate SSL traffic are not even rare these days and HTTP/2 can’t do much about limiting the use of such mechanisms.

2. Think of the little ones

small-big-dogSmall devices cannot handle the extra TLS burden“. Either because of the extra CPU load that comes with TLS or because of the cert management in a billion printers/fridges/routers etc. Certificates also expire regularly and need to be updated in the field.

Of course there will be a least acceptable system performance required to do TLS decently and there will always be systems that fall below that threshold.

3. Certificates are too expensive

The price of certificates for servers are historically often brought up as an argument against TLS even it isn’t really HTTP/2 related and I don’t think it was ever an argument that was particularly strong against TLS within HTTP/2. Several CAs now offer zero-cost or very close to zero-cost certificates these days and with the upcoming efforts like letsencrypt.com, chances are it’ll become even better in the not so distant future.

pile-of-moneyRecently someone even claimed that HTTPS limits the freedom of users since you need to give personal information away (he said) in order to get a certificate for your server. This was not a price he was willing to pay apparently. This is however simply not true for the simplest kinds of certificates. For Domain Validated (DV) certificates you usually only have to prove that you “control” the domain in question in some way. Usually by being able to receive email to a specific receiver within the domain.

4. The CA system is broken

TLS of today requires a PKI system where there are trusted certificate authorities that sign certificates and this leads to a situation where all modern browsers trust several hundred CAs to do this right. I don’t think a lot of people are happy with this and believe this is the ultimate security solution. There’s a portion of the Internet that advocates for DANE (DNSSEC) to address parts of the problem, while others work on gradual band-aids like Certificate Transparency and OCSP stapling to make it suck less.

please trust me

My personal belief is that rejecting TLS on the grounds that it isn’t good enough or not perfect is a weak argument. TLS and HTTPS are the best way we currently have to secure web sites. I wouldn’t mind seeing it improved in all sorts of ways but I don’t believe running protocols clear text until we have designed and deployed the next generation secure protocol is a good idea – and I think it will take a long time (if ever) until we see a TLS replacement.

Who were against mandatory TLS?

Yeah, lots of people ask me this, but I will refrain from naming specific people or companies here since I have no plans on getting into debates with them about details and subtleties in the way I portrait their arguments. You can find them yourself if you just want to and you can most certainly make educated guesses without even doing so.

What about opportunistic security?

A text about TLS in HTTP/2 can’t be complete without mentioning this part. A lot of work in the IETF these days are going on around introducing and making sure opportunistic security is used for protocols. It was also included in the HTTP/2 draft for a while but was moved out from the core spec in the name of simplification and because it could be done anyway without being part of the spec. Also, far from everyone believes opportunistic security is a good idea. The opponents tend to say that it will hinder the adoption of “real” HTTPS for sites. I don’t believe that, but I respect that opinion because it is a guess as to how users will act just as well as my guess is they won’t act like that!

Opportunistic security for HTTP is now being pursued outside of the HTTP/2 spec and allows clients to upgrade plain TCP connections to instead do “unauthenticated TLS” connections. And yes, it should always be emphasized: with opportunistic security, there should never be a “padlock” symbol or anything that would suggest that the connection is “secure”.

Firefox supports opportunistic security for HTTP and it will be enabled by default from Firefox 37.


Пост доступен на сайте softdroid.net: Восстановление: TLS в HTTP/2. (Russian)

TLS in HTTP/2 (Kazakh)

Bug finding is slow in spite of many eyeballs

“given enough eyeballs, all bugs are shallow”

The saying (also known as Linus’ law) doesn’t say that the bugs are found fast and neither does it say who finds them. My version of the law would be much more cynical, something like: “eventually, bugs are found“, emphasizing the ‘eventually’ part.

(Jim Zemlin apparently said the other day that it can work the Linus way, if we just fund the eyeballs to watch. I don’t think that’s the way the saying originally intended.)

Because in reality, many many bugs are never really found by all those given “eyeballs” in the first place. They are found when someone trips over a problem and is annoyed enough to go searching for the culprit, the reason for the malfunction. Even if the code is open and has been around for years it doesn’t necessarily mean that any of all the people who casually read the code or single-stepped over it will actually ever discover the flaws in the logic. The last few years several world-shaking bugs turned out to have existed for decades until discovered. In code that had been read by lots of people – over and over.

So sure, in the end the bugs were found and fixed. I would argue though that it wasn’t because the projects or problems were given enough eyeballs. Some of those problems were found in extremely popular and widely used projects. They were found because eventually someone accidentally ran into a problem and started digging for the reason.

Time until discovery in the curl project

I decided to see how it looks in the curl project. A project near and dear to me. To take it up a notch, we’ll look only at security flaws. Not only because they are the probably most important bugs we’ve had but also because those are the ones we have the most carefully noted meta-data for. Like when they were reported, when they were introduced and when they were fixed.

We have no less than 30 logged vulnerabilities for curl and libcurl so far through-out our history, spread out over the past 16 years. I’ve spent some time going through them to see if there’s a pattern or something that sticks out that we should put some extra attention to in order to improve our processes and code. While doing this I gathered some random info about what we’ve found so far.

On average, each security problem had been present in the code for 2100 days when fixed – that’s more than five and a half years. On average! That means they survived about 30 releases each. If bugs truly are shallow, it is still certainly not a fast processes.

Perhaps you think these 30 bugs are really tricky, deeply hidden and complicated logic monsters that would explain the time they took to get found? Nope, I would say that every single one of them are pretty obvious once you spot them and none of them take a very long time for a reviewer to understand.

Vulnerability ages

This first graph (click it for the large version) shows the period each problem remained in the code for the 30 different problems, in number of days. The leftmost bar is the most recent flaw and the bar on the right the oldest vulnerability. The red line shows the trend and the green is the average.

The trend is clearly that the bugs are around longer before they are found, but since the project is also growing older all the time it sort of comes naturally and isn’t necessarily a sign of us getting worse at finding them. The average age of flaws is aging slower than the project itself.

Reports per year

How have the reports been distributed over the years? We have a  fairly linear increase in number of lines of code but yet the reports were submitted like this (now it goes from oldest to the left and most recent on the right – click for the large version):


Compare that to this chart below over lines of code added in the project (chart from openhub and shows blanks in green, comments in grey and code in blue, click it for the large version):

curl source code growth

We received twice as many security reports in 2014 as in 2013 and we got half of all our reports during the last two years. Clearly we have gotten more eyes on the code or perhaps users pay more attention to problems or are generally more likely to see the security angle of problems? It is hard to say but clearly the frequency of security reports has increased a lot lately. (Note that I here count the report year, not the year we announced the particular problems, as they sometimes were done on the following year if the report happened late in the year.)

On average, we publish information about a found flaw 19 days after it was reported to us. We seem to have became slightly worse at this over time, the last two years the average has been 25 days.

Did people find the problems by reading code?

In general, no. Sure people read code but the typical pattern seems to be that people run into some sort of problem first, then dive in to investigate the root of it and then eventually they spot or learn about the security problem.

(This conclusion is based on my understanding from how people have reported the problems, I have not explicitly asked them about these details.)

Common patterns among the problems?

I went over the bugs and marked them with a bunch of descriptive keywords for each flaw, and then I wrote up a script to see how the frequent the keywords are used. This turned out to describe the flaws more than how they ended up in the code. Out of the 30 flaws, the 10 most used keywords ended up like this, showing number of flaws and the keyword:

8 cert-check
8 buffer-overflow

6 info-leak
3 URL-parsing
3 openssl
3 http-headers
3 cookie

I don’t think it is surprising that TLS, HTTP or certificate checking are common areas of security problems. TLS and certs are complicated, HTTP is huge and not easy to get right. curl is mostly C so buffer overflows is a mistake that sneaks in, and I don’t think 27% of the problems tells us that this is a problem we need to handle better. Also, only 2 of the last 15 flaws (13%) were buffer overflows.

The discussion following this blog post is on hacker news.

curl 7.40.0: unix domain sockets and smb

curl and libcurl curl dot-to-dot7.40.0 was just released this morning. There’s a closer look at some of the perhaps more noteworthy changes. As usual, you can find the entire changelog on the curl web site.

HTTP over unix domain sockets

So just before the feature window closed for the pending 7.40.0 release of curl, Peter Wu’s patch series was merged that brings the ability to curl and libcurl to do HTTP over unix domain sockets. This is a feature that’s been mentioned many times through the history of curl but never previously truly implemented. Peter also very nicely adjusted the test server and made two test cases that verify the functionality.

To use this with the curl command line, you specify the socket path to the new –unix-domain option and assuming your local HTTP server listens on that socket, you’ll get the response back just as with an ordinary TCP connection.

Doing the operation from libcurl means using the new CURLOPT_UNIX_SOCKET_PATH option.

This feature is actually not limited to HTTP, you can do all the TCP-based protocols except FTP over the unix domain socket, but it is to my knowledge only HTTP that is regularly used this way. The reason FTP isn’t supported is of course its use of two connections which would be even weirder to do like this.


SMB is also known as CIFS and is an old network protocol from the Microsoft world access files. curl and libcurl now support this protocol with SMB:// URLs thanks to work by Bill Nagel and Steve Holme.

Security Advisories

Last year we had a large amount of security advisories published (eight to be precise), and this year we start out with two fresh ones already on the 8th day… The ones this time were of course discovered and researched already last year.

CVE-2014-8151 is a way we accidentally allowed an application to bypass the TLS server certificate check if a TLS Session-ID was already cached for a non-checked session – when using the Mac OS SecureTransport SSL backend.

CVE-2014-8150 is a URL request injection. When letting curl or libcurl speak over a HTTP proxy, it would copy the URL verbatim into the HTTP request going to the proxy, which means that if you craft the URL and insert CRLFs (carriage returns and linefeed characters) you can insert your own second request or even custom headers into the request that goes to the proxy.

You may enjoy taking a look at the curl vulnerabilities table.

Bugs bugs bugs

The release notes mention no less than 120 specific bug fixes, which in comparison to other releases is more than average.


Can curl avoid to be in a future funnily named exploit that shakes the world?

During this year we’ve seen heartbleed and shellshock strike (and a  few more big flaws that I’ll skip for now). Two really eye opening recent vulnerabilities in projects with many similarities:

  1. Popular corner stones of open source stacks and internet servers
  2. Mostly run and maintained by volunteers
  3. Mature projects that have been around since “forever”
  4. Projects believed to be fairly stable and relatively trustworthy by now
  5. A myriad of features, switches and code that build on many platforms, with some parts of code only running on a rare few
  6. Written in C in a portable style

Does it sound like the curl project to you too? It does to me. Sure, this description also matches a slew of other projects but I lead the curl development so let me stay here and focus on this project.

cURLAre we in jeopardy? I honestly don’t know, but I want to explain what we do in our project in order to minimize the risk and maximize our ability to find problems on our own before they become serious attack vectors somewhere!

previous flaws

There’s no secret that we have let security problems slip through at times. We’re right now working toward our 143rd release during our around 16 years of life-time. We have found and announced 28 security problems over the years. Looking at these found problems, it is clear that very few security problems are discovered quickly after introduction. Most of them linger around for several years until found and fixed. So, realistically speaking based on history: there are security bugs still in the code, and they have probably been present for a while already.

code reviews and code standards

We try to review all patches from people without push rights in the project. It would probably be a good idea to review all patches before they go in for real, but that just wouldn’t work with the (lack of) man power we have in the project while we at the same time want to develop curl, move it forward and introduce new things and features.

We maintain code standards and formatting to keep code easy to understand and follow. We keep individual commits smallish for easier review now or in the future.

test cases

As simple as it is, we test that the basic stuff works. We don’t and can’t test everything but having test cases for most things give us the confidence to change code when we see problems as we then remain fairly sure things keep working the same way as long as the test go through. In projects with much less test coverage, you become much more conservative with what you dare to change and that also makes you more vulnerable.

We always want more test cases and we want to improve on how we always add test cases when we add new features and ideally we should also add new test cases when we fix bugs so that we know that we don’t introduce any such bug again in the future.

static code analyzes

We regularly scan our code base using static code analyzers. Both clang-analyzer and coverity are good tools, and they help us by pointing out code that look wrong or suspicious. By making sure we have very few or no such flaws left in the code, we minimize the risk. A static code analyzer is better than run-time tools for cases where they can check code flows that are hard to repeat in my local environment.


bike helmetValgrind is an awesome tool to detect memory problems in run-time. Leaks or just stupid uses of memory or related functions. We have our test suite automatically use valgrind when it runs tests in case it is present and it helps us make sure that all situations we test for are also error-free from valgrind’s point of view.


Building and testing curl on a plethora of platforms non-stop is also useful to make sure we don’t depend on behaviors of particular library implementations or non-standard features and more. Testing it all is basically the only way to make sure everything keeps working over the years while we continue to develop and fix bugs. We would course be even better off with more platforms that would test automatically and with more developers keeping an eye on problems that show up there…

code complexity

Arguably, one of the best ways to avoid security flaws and bugs in general, is to keep the source code as simple as possible. Complex functions need to be broken down into smaller functions that are possible to read and understand. A good way to identify functions suitable for fixing is pmccabe,

essential third parties

curl and libcurl are usually built to use a whole bunch of third party libraries in order to perform all the functionality. In order to not have any of those uses turn into a source for trouble we must of course also participate in those projects and help them stay strong and make sure that we use them the proper way that doesn’t lead to any bad side-effects.

You can help!

All this takes time, energy and system resources. Your contributions and help will be appreciated where ever among these tasks that you can insert any. We could do more of all this, more often and more thorough if we only were more people involved!

curl is no POODLE

Once again the internet flooded over with reports and alerts about a vulnerability using a funny name: POODLE. If you have even the slightest interest in this sort of stuff you’ve already grown tired and bored about everything that’s been written about this so why on earth do I have to pile on and add to the pain?

This is my way of explaining how POODLE affects or doesn’t affect curl, libcurl and the huge amount of existing applications using libcurl.

Is my application using HTTPS with libcurl or curl vulnerable to POODLE?

No. POODLE really is a browser-attack.


The POODLE attack is a combination of several separate pieces that when combined allow attackers to exploit it. The individual pieces are not enough stand-alone.

SSLv3 is getting a lot of heat now since POODLE must be able to downgrade a connection to SSLv3 from TLS to work. Downgrade in a fairly crude way – in libcurl, only libcurl built to use NSS as its TLS backend supports this way of downgrading the protocol level.

Then, if an attacker manages to downgrade to SSLv3 (both the client and server must thus allow this) and get to use the sensitive block cipher of that protocol, it must maintain a connection to the server and then retry many similar requests to the server in order to try to work out details of the request – to figure out secrets it shouldn’t be able to. This would typically be made using javascript in a browser and really only HTTPS allows this so no other SSL-using protocol can be exploited like this.

For the typical curl user or a libcurl user, there’s A) no javascript and B) the application already knows the request it is doing and normally doesn’t inject random stuff from 3rd party sources that could be allowed to steal secrets. There’s really no room for any outsider here to steal secrets or cookies or whatever.

How will curl change

There’s no immediate need to do anything as curl and libcurl are not vulnerable to POODLE.

Still, SSLv3 is long overdue and is not really a modern protocol (TLS 1.0, the successor, had its RFC published 1999) so in order to really avoid the risk that it will be possible exploit this protocol one way or another now or later using curl/libcurl, we will disable SSLv3 by default in the next curl release. For all TLS backends.

Why? Just to be extra super cautious and because this attack helped us remember that SSLv3 is old and should be let down to die.

If possible, explicitly requesting SSLv3 should still be possible so that users can still work with their legacy systems in dire need of upgrade but placed in corners of the world that every sensible human has since long forgotten or just ignored.

In-depth explanations of POODLE

I especially like the ones provided by PolarSSL and GnuTLS, possibly due to their clear “distance” from browsers.