After another eight week cycle was been completed, curl shipped a new release into the world. 7.65.0 brings some news and some security fixes but is primarily yet again a set of bug-fixes bundled up. Remember 7.64.1?
CURLOPT_MAXAGE_CONN is a new option that controls how long to keep a live connection in the connection cache for reuse.
This release comes with fixes for two separate security problems. Both rated low risk. Both reported via the new bug bounty program.
CVE-2019-5435 is an issue in the recently introduced URL parsing API. It is only a problem in 32 bit architectures and only if an application can be told to pass in ridiculously long (> 2GB) strings to libcurl. This bug is similar in nature to a few other bugs libcurl has had in the past, and to once and for all combat this kind of flaw libcurl now (in 7.65.0 and forward) has a “maximum string length” limit for strings that you can pass to it using its APIs. The maximum size is 8MB. (The reporter was awarded 150 USD for this find.)
CVE-2019-5436 is a problem in the TFTP code. If an application decides to uses a smaller “blksize” than 504 (default is 512), curl would overflow a buffer allocated on the heap with data received from the server. Luckily, very few people actually download data from unknown or even remote TFTP servers. Secondly, asking for a blksize smaller than 512 is rather pointless and also very rare: the primary point in changing that size is to enlarge it. (The reporter was awarded 200 USD for this find.)
Over one hundred bug-fixes landed in this release, but some of my favorites from release cycle include…
mark connection for close on TLS close_notify
close_notify is a message in the TLS protocol that means that this connection is about to close. In most circumstances that message doesn’t actually provide information to curl that is needed, but in the case the connection is closed prematurely, understanding that this message preceded the closure helps curl act appropriately. This change was done for the OpenSSL backend only as that’s where we got the bug reported and worked on it this time, but I think we might have reasons to do the same for other backends going forward!
show port in the verbose “Trying …” message
The verbose message that says “Trying 22.214.171.124…” means that curl has sent started a TCP connect attempt to that IP address. This message has now been modified to also include the target port number so when using -v with curl 7.65.0, connecting to that same host for HTTPS will instead say “Trying 126.96.36.199:443…”.
To aid debugging really. I think it gives more information faster at a place you’re already looking.
new SOCKS 4+5 test server
The test suite got a brand new SOCKS server! Previously, all SOCKS tests for both version 4 and version 5 were done by firing up ssh (typically openssh). That method was decent but made it hard to do a range of tests for bad behavior, bad protocol replies and similar. With the new custom test server, we can basically add whatever test we want and we’ve already extended the SOCKS testing to cover more code and use cases than previously.
SOCKS5 user name and passwords must be shorter than 256
curl allows user names and passwords provided in URLs and as separate options to be more or less unrestricted in size and that include if the credentials are used for SOCKS5 authentication – totally ignoring the fact that the protocol SOCKS5 has a maximum size of 255 for the fields. Starting now, curl will return an error if the credentials for SOCKS5 are too long.
Warn if curl and libcurl versions do not match
The command line tool and the library are independent and separable, as in you can run one version of the curl tool with another version of the libcurl library. The libcurl API is solid enough to allow it and the tool is independent enough to not restrict it further.
We always release curl the command line tool and libcurl the library together, using the same version number – with the code for both shipped in the same single file.
There should rarely be a good reason to actually run curl and libcurl with different versions. Starting now, curl will show a little warning if this is detected as we have learned that this is almost always a sign of an installation or setup mistake. Hopefully this message will aid people to detect the mistake earlier and easier.
Better handling of “–no-” prefixed options
curl’s command line parser allows users to switch off boolean options by prefixing them with dash dash no dash. For example we can switch off compressed responses by using “–no-compression” since there regular option “–compression” switches it on.
It turned out we stripped the “–no-” thing no regarding if the option was boolean or not and presumed the logic to handle it – which it didn’t. So users could actually pass a proxy string to curl with the regular option “–proxy” as well as “–no-proxy”. The latter of course not making much sense and was just due to an oversight.
In 7.65.0, only actual boolean command line options can be used with “–no-“. Trying it on other options will cause curl to report error for it.
Add CURLUPART_ZONEID to the URL API
Remember when we added a new URL parsing API to libcurl back in 7.62.0? It wasn’t even a year ago! When we did this, we also changed the internals to use the same code. It turned out we caused a regression when we parsed numerical IPv6 addresses that provide the zone ID within the string. Like this: “https://[ffe80::1%25eth0]/index.html”
Starting in this release, you can both set and get the zone ID in a URL using the API, but of course setting it doesn’t do anything unless the host is a numeric IPv6 address.
parse proxy with the URL parser API
We removed the separate proxy string parsing logic and instead switched that over to more appropriately use the generic URL parser for this purpose as well. This move reduced the code size, made the code simpler and makes sure we have a unified handling of URLs! Everyone is happy!
longer URL schemes
I naively wrote the URL parser to handle scheme names as long as the longest scheme we support in curl: 8 bytes. But since the parser can also be asked to parse URLs with non-supported schemes, that limit was a bit too harsh. I did a quick research, learned that the longest currently registered URI scheme is 36 characters (“microsoft.windows.camera.multipicker”). Starting in this release , curl accepts URL schemes up to 40 bytes long.
Coming up next
There’s several things brewing in the background that might be ready to show in next release. Parallel transfers in the curl tool and deprecating PolarSSL support seem likely to happen for example. Less likely for this release, but still being worked on slowly, is HTTP/3 support.
We’re also likely to get a bunch of changes and fine features we haven’t even thought about from our awesome contributors. In eight weeks I hope to write another one of these blog posts explaining what went into that release…
curl, or libcurl specifically, is probably the world’s most popular and widely used HTTP client side library counting more than six billion installs.
curl is a rock solid and feature-packed library that supports a huge amount of protocols and capabilities that surpass most competitors. But this comes at a cost: it is not the smallest library you can find.
Within a 100K
Instead of being happy with getting told that curl is “too big” for certain use cases, I set a goal for myself: make it possible to builda version of curl that can do HTTPS and fit in 100K (including the wolfSSL TLS library) on a typical 32 bit architecture.
As a comparison, the tiny-curl shared library when built on an x86-64 Linux, is smaller than 25% of the size as the default Debian shipped library is.
But let’s not stop there. Users with this kind of strict size requirements are rarely running a full Linux installation or similar OS. If you are sensitive about storage to the exact kilobyte level, you usually run a more slimmed down OS as well – so I decided that my initial tiny-curl effort should be done on FreeRTOS. That’s a fairly popular and free RTOS for the more resource constrained devices.
This port is still rough and I expect us to release follow-up releases soon that improves the FreeRTOS port and ideally also adds support for other popular RTOSes. Which RTOS would you like to support for that isn’t already supported?
Offer the libcurl API for HTTPS on FreeRTOS, within 100 kilobytes.
I strongly believe that the power of having libcurl in your embedded devices is partly powered by the libcurl API. The API that you can use for libcurl on any platform, that’s been around for a very long time and for which you can find numerous examples for on the Internet and in libcurl’s extensive documentation. Maintaining support for the API was of the highest priority.
My secondary goal was to patch as clean as possible so that we can upstream patches into the main curl source tree for the changes makes sense and that aren’t disturbing to the general code base, and for the work that we can’t we should be able to rebase on top of the curl code base with as little obstruction as possible going forward.
Keep the HTTPS basics
I just want to do HTTPS GET
That’s the mantra here. My patch disables a lot of protocols and features:
No protocols except HTTP(S) are supported
No cookie support
No date parsing
No HTTP authentication
No .netrc parsing
No HTTP multi-part formposts
No shuffled DNS support
No built-in progress meter
Although they’re all disabled individually so it is still easy to enable one or more of these for specific builds.
Most of the patches in tiny-curl are being upstreamed into curl in the #3844 pull request. I intend to upstream most, if not all, of the tiny-curl work over time.
The FreeRTOS port of tiny-curl is licensed GPLv3 and not MIT like the rest of curl. This is an experiment to see how we can do curl work like this in a sustainable way. If you want this under another license, we’re open for business over at wolfSSL!
(I will update this blog post with more links to videos and PDFs to presentations as they get published, so come back later in case your favorite isn’t linked already.)
The third curl developers conference, curl up 2019, is how history. We gathered in the lovely Charles University in central Prague where we sat down in an excellent class room. After the HTTP symposium on the Friday, we spent the weekend to dive in deeper in protocols and curl details.
I started off the Saturday by The state of the curl project (youtube). An overview of how we’re doing right now in terms of stats, graphs and numbers from different aspects and then something about what we’ve done the last year and a quick look at what’s not do good and what we could work on going forward.
James Fuller took the next session and his Newbie guide to contributing to libcurl presentation. Things to consider and general best practices to that could make your first steps into the project more likely to be pleasant!
Long term curl hacker Dan Fandrich (also known as “Daniel two” out of the three Daniels we have among our top committers) followed up with Writing an effective curl test where the detailed what different tests we have in curl, what they’re for and a little about how to write such tests.
After that I was back behind the desk in the classroom that we used for this event and I talked The Deprecation of legacy crap (Youtube). How and why we are removing things, some things we are removing and will soon remove and finally a little explainer on our new concept and handling of “experimental” features.
Our local hero organizer James Fuller then spoiled us completely when we got around to have dinner at a monastery with beer brewing monks and excellent food. Good food, good company and curl related dinner subjects. That’s almost heaven defined!
Daylight saving time morning and you could tell. I’m sure it was not at all related to the beers from the night before…
James Fuller fired off the day by talking to us about Curlpipe (github), a DSL for building http execution pipelines.
Robin Marx then put in the next gear and entertained us another hour with a protocol deep dive titled HTTP/3 (QUIC): the details (slides). For me personally this was a exactly what I needed as Robin clearly has kept up with more details and specifics in the QUIC and HTTP/3 protocols specifications than I’ve managed and his talk help the rest of the room get at least little bit more in sync with current development.
Jakub Nesetril and Lukáš Linhart from Apiary then talked us through what they’re doing and thinking around web based APIs and how they and their customers use curl: Real World curl usage at Apiary.
Jakub Klímek explained to us in very clear terms about current and existing problems in his talk IRIs and IDNs: Problems of non-ASCII countries. Some of the problems involve curl and while most of them have their clear explanations, I think we have to lessons to learn from this: URLs are still as messy and undocumented as ever before and that we might have some issues to fix in this area in curl.
To bring my fellow up to speed on the details of the new API introduced the last year I then made a presentation called The new URL API.
I ended up doing seven presentations during this single weekend. Not all of them stellar or delivered with elegance but I hope they were still valuable to some. I did not steal someone else’s time slot as I would gladly have given up time if we had other speakers wanted to say something. Let’s aim for more non-Daniel talkers next time!
A weekend like this is such a boost for inspiration, for morale and for my ego. All the friendly faces with the encouraging and appreciating comments will keep me going for a long time after this.
Thank you to our awesome and lovely event sponsors – shown in the curl up logo below! Without you, this sort of happening would not happen.
curl up 2020
I will of course want to see another curl up next year. There are no plans yet and we don’t know where to host. I think it is valuable to move it around but I think it is even more valuable that we have a friend on the ground in that particular city to help us out. Once this year’s event has sunken in properly and a month or two has passed, the case for and organization of next year’s conference will commence. Stay tuned, and if you want to help hosting us do let me know!
If you want commercial support, ports of curl to other operating systems or just instant help to fix your curl related problems, we’re here to help. Get in touch now! This is the premiere. This has not been offered by me or anyone else before.
I’m not sure I need to say it, but I personally have authored almost 60% of all commits in the curl source code during my more than twenty years in the project. I started the project, I’ve designed its architecture etc. There is simply no one around with my curl experience and knowledge of curl internals. You can’t buy better curl expertise.
curl has become one of the world’s most widely used software components and is the transfer engine doing a large chunk of all non-browser Internet transfers in the world today. curl has reached this level of success entirely without anyone offering commercial services around it. Still, not every company and product made out there has a team of curl experts and in this demanding time and age we know there are times when you rather hire the right team to help you out.
We are the curl experts that can help you and your team. Contact us for all and any support questions at email@example.com.
What about the curl project?
I’m heading into this new chapter of my life and the curl project with the full knowledge that this blurs the lines between my job and my spare time even more than before. But fear not!
The curl project is free and open and will remain independent of any commercial enterprise helping out customers. I realize me offering companies and organizations to deal with curl problems and solving curl issues for compensation creates new challenges and questions where boundaries go, if for nothing else for me personally. I still think this is worth pursuing and I’m sure we can figure out and handle whatever minor issues this can lead to.
My friends, the community, the users and harsh critiques on twitter will all help me stay true and honest. I know this. This should end up a plus for the curl project in general as well as for me personally. More focus, more work and more money involved in curl related activities should improve the project.
It is with great joy and excitement I take on this new step.
Let me start by saying thank you to all and everyone who sent me job offers or otherwise reached out with suggestions and interesting career moves. I received more than twenty different offers and almost every one of those were truly good options that I could have said yes to and still pulled home a good job. What a luxury challenge to have to select something from that! Publicly announcing me leaving Mozilla turned out a great ego-boost.
I took some time off to really reflect and contemplate on what I wanted from my next career step. What would the right next move be?
I love working on open source. Internet protocols, and transfers and doing libraries written in C are things considered pure fun for me. Can I get all that and yet keep working from home, not sacrifice my wage and perhaps integrate working on curl better in my day to day job?
I talked to different companies. Very interesting companies too, where I have friends and people who like me and who really wanted to get me working for them, but in the end there was one offer with a setup that stood out. One offer for which basically all check marks in my wish-list were checked.
On February 5, 2019 I’m starting my new job at wolfSSL. My short and sweet period as unemployed is over and now it’s full steam ahead again! (Some members of my family have expressed that they haven’t really noticed any difference between me having a job and me not having a job as I spend all work days the same way nevertheless: in front of my computer.)
Starting now, we offer commercial curl support and various services for and around curl that companies and organizations previously really haven’t been able to get. Time I do not spend on curl related activities for paying customers I will spend on other networking libraries in the wolfSSL “portfolio”. I’m sure I will be able to keep busy.
I’ve met Larry at wolfSSL physically many times over the years and every year at FOSDEM I’ve made certain to say hello to my wolfSSL friends in their booth they’ve had there for years. They’re truly old-time friends.
wolfSSL is mostly a US-based company – I’m the only Swede on the team and the only one based in Sweden. My new colleagues all of course know just as well as you that I’m prevented from traveling to the US. All coming physical meetings with my work mates will happen in other countries.
TLS 1.3 has been in development for quite some time by now, and a lot of TLS libraries already support it to some extent. At varying draft levels.
curl and libcurl has supported an explicit option to select TLS 1.3 since curl 7.52.0 (December 2016) and assuming you build curl to use a TLS library with support, you’ve been able to use TLS 1.3 with curl since at least then. The support has gradually been expanded to cover more and more libraries since then.
Today, curl and libcurl support speaking TLS 1.3 if you build it to use one of these fine TLS libraries of a recent enough version:
Secure Transport (on iOS 11 or later, and macOS 10.13 or later)