Archive for the ‘Work’ Category

Tightening Firefox’s HTTP framing – again

Thursday, February 12th, 2015

An old http1.1 frameCall me crazy, but I’m at it again. First a little resume from our previous episodes in this exciting saga:

Chapter 1: I closed the 10+ year old bug that made the Firefox download manager not detect failed downloads, simply because Firefox didn’t care if the HTTP 1.1 Content-Length was larger than what was actually saved – after the connection potentially was cut off for example. There were additional details, but that was the bigger part.

Chapter 2: After having been included all the way to public release, we got a whole slew of bug reports immediately when Firefox 33 shipped and we had to revert parts of the fix I did.

Chapter 3.

Will it land before it turns 11 years old? The bug was originally submitted 2004-03-16.

Since chapter two of this drama brought back the original bugs again we still have to do something about them. I fully understand if not that many readers of this can even keep up of all this back and forth and juggling of HTTP protocol details, but this time we’re putting back the stricter frame checks with a few extra conditions to allow a few violations to remain but detect and react on others!

Here’s how I addressed this issue. I wanted to make the checks stricter but still allow some common protocol violations.

In particular I needed to allow two particular flaws that have proven to be somewhat common in the wild and were the reasons for the previous fix being backed out again:

A – HTTP chunk-encoded responses that lack the final 0-sized chunk.

B – HTTP gzipped responses where the Content-Length is not the same as the actual contents.

So, in order to allow A + B and yet be able to detect prematurely cut off transfers I decided to:

  1. Detect incomplete chunks then the transfer has ended. So, if a chunk-encoded transfer ends on exactly a chunk boundary we consider that fine. Good: This will allow case (A) to be considered fine. Bad: It will make us not detect a certain amount of cut-offs.
  2. When receiving a gzipped response, we consider a gzip stream that doesn’t end fine according to the gzip decompressing state machine to be a partial transfer. IOW: if a gzipped transfer ends fine according to the decompressor, we do not check for size misalignment. This allows case (B) as long as the content could be decoded.
  3. When receiving HTTP that isn’t content-encoded/compressed (like in case 2) and not chunked (like in case 1), perform the size comparison between Content-Length: and the actual size received and consider a mismatch to mean a NS_ERROR_NET_PARTIAL_TRANSFER error.

Firefox BallPrefs

When my first fix was backed out, it was actually not removed but was just put behind a config string (pref as we call it) named “network.http.enforce-framing.http1“. If you set that to true, Firefox will behave as it did with my original fix applied. It makes the HTTP1.1 framing fairly strict and standard compliant. In order to not mess with that setting that now has been around for a while (and I’ve also had it set to true for a while in my browser and I have not seen any problems with doing it this way), I decided to introduce my new changes pref’ed behind a separate variable.

network.http.enforce-framing.soft” is the new pref that is set to true by default with my patch. It will make Firefox do the detections outlined in 1 – 3 and setting it to false will disable those checks again.

Now I only hope there won’t ever be any chapter 4 in this story… If things go well, this will appear in Firefox 38.

Chromium

But how do they solve these problems in the Chromium project? They have slightly different heuristics (with the small disclaimer that I haven’t read their code for this in a while so details may have changed). First of all, they do not allow a missing final 0-chunk. Then, they basically allow any sort of misaligned size when the content is gzipped.

Update: this patch was subsequently backed out again due to several bug reports about it. I have yet to analyze exactly what went wrong.

HTTP/2 is at 5%

Tuesday, February 10th, 2015

http2 logoHere follow some numbers extracted from my recent HTTP/2 presentation.

First: HTTP/2 is not finalized yet and it is not yet in RFC status, even though things are progressing nicely within the IETF. With some luck we reach RFC status within Q1 this year.

On January 13th 2015, Firefox 35 was released with HTTP/2 enabled by default. Firefox was already running it enabled before that in beta and development versions.

Chrome has also been sporting HTTP/2 support in development versions since many moths back where it could easily be manually enabled. Chrome 40 was the first main release shipped with HTTP/2 enabled by default, but it has so far only been enabled for a very small fraction of the user-base.

On January 28th 2015, Google reported to me by email that they saw HTTP/2 being used in 5% of their global traffic (que all relevant disclaimers that this is not statistically safe numbers). This, close after a shaky period with Google having had their HTTP/2 services disabled through parts of the Christmas holidays (due to bugs) – and as explained above, there’s been no time for any mainstream browser to use HTTP/2 by default for very long!

Further data points: Mozilla collects telemetry data from Firefox users who opted-in to it, and it collects numbers on “HTTP Protocol Version Used on Response”. On February 10, it reports that Firefox 35 users have got their responses to report HTTP/2 in 9% of all responses (out of more than 340 billion reported responses). The Telemetry for Firefox Nightly 38 even reports HTTP/2 in 14% of all responses (based on a much smaller sample collection), which I guess could very well be because users on such a bleeding edge version are more experimental by nature.

In these Firefox stats we see that recently, the number of HTTP/2 responses outnumber the HTTP/1.0 responses 9 to 1.

Http2 right now

Sunday, February 1st, 2015

I talked in the Mozilla devroom at FOSDEM 2015. Here are the slides from it. It was recorded on video and I will post a suitable link to that once it becomes available. The talk was meant to be 20 minutes, I think I did it on 22 or something.

Changing networks with Linux

Friday, January 16th, 2015

A rather long time ago I blogged about my work to better deal with changing networks while Firefox is running, and the change was then pushed for Android and I subsequently pushed the same functionality for Firefox on Mac.

Today I’ve landed yet another change, which detects network changes on Firefox OS and Linux.

Firefox Nightly screenshotAs Firefox OS uses a Linux kernel, I ended up doing the same fix for both the Firefox OS devices as for Firefox on Linux desktop: I open a socket in the AF_NETLINK family and listen on the stream of messages the kernel sends when there are network updates. This way we’re told when the routing tables update or when we get a new IP address etc. I consider this way better than the NotifyIpInterfaceChange() API Windows provides, as this allows us to filter what we’re interested in. The windows API makes that rather complicated and in fact a lot of the times when we get the notification on windows it isn’t clear to me why!

The Mac API way is what I would consider even more obscure, but then I’m not at all used to their way of doing things and how you add things to the event handlers etc.

The journey to the landing of this particular patch was once again long and bumpy and full of sweat in this tradition that seem seems to be my destiny, and this time I ran into problems with the Firefox OS emulator which seems to have some interesting bugs that cause my code to not work properly and as a result of that our automated tests failed: occasionally data sent over a pipe or socketpair doesn’t end up in the receiving end. In my case this means that my signal to the child thread to die would sometimes not be noticed and thus the thread wouldn’t exit and die as intended.

I ended up implementing a work-around that makes it work even if the emulator eats the data by also checking a shared should-I-shutdown-now flag every once in a while. For more specific details on that, see the bug.

My talks at FOSDEM 2015

Wednesday, January 14th, 2015

fosdem

Sunday 13:00, embedded room (Lameere)

Tile: Internet all the things – using curl in your device

Embedded devices are very often network connected these days. Network connected embedded devices often need to transfer data to and from them as clients, using one or more of the popular internet protocols.

libcurl is the world’s most used and most popular internet transfer library, already used in every imaginable sort of embedded device out there. How did this happen and how do you use libcurl to transfer data to or from your device?

Note that this talk was originally scheduled to be at a different time!

Sunday, 09:00 Mozilla room (UD2.218A)

Title: HTTP/2 right now

HTTP/2 is the new version of the web’s most important and used protocol. Version 2 is due to be out very soon after FOSDEM and I want to inform the audience about what’s going on with the protocol, why it matters to most web developers and users and not the last what its status is at the time of FOSDEM.

My first year at Mozilla

Tuesday, January 13th, 2015

January 13th 2014 I started my fiMozilla dinosaur head logorst day at Mozilla. One year ago exactly today.

It still feels like it was just a very short while ago and I keep having this sense of being a beginner at the company, in the source tree and all over.

One year of networking code work that really at least during periods has not progressed as quickly as I would’ve wished for, and I’ve had some really hair-tearing problems and challenges that have taken me sweat and tears to get through. But I am getting through and I’m enjoying every (oh well, let’s say almost every) moment.

During the year I’ve had the chance to meetup with my team mates twice (in Paris and in Portland) and I’ve managed to attend one IETF (in London) and two special HTTP2 design meetings (in London and NYC).

openhub.net counts 47 commits by me in Firefox and that feels like counting high. bugzilla has tracked activity by me in 107 bug reports through the year.

I’ve barely started. I’ll spend the next year as well improving Firefox networking, hopefully with a higher turnout this year. (I don’t mean to make this sound as if Firefox networking is just me, I’m just speaking for my particular part of the networking team and effort and I let the others speak for themselves!)

Onwards and upwards!

Keyboard key frequency

Wednesday, November 12th, 2014

A while ago I wrote about my hunt for a new keyboard, and in my follow-up conversations with friends around that subject I quickly came to the conclusion I should get myself better analysis and data on how I actually use a keyboard and the individual keys on it. And if you know me, you know I like (useless) statistics.

Func KB-460 keyboardSo, I tried out the popular and widely used Linux key-logger software ‘logkeys‘ and immediately figured out that it doesn’t really support the precision and detail level I wanted so I forked the project and modified the code to work the way I want it: keyfreq was born. Code on github. (I forked it because I couldn’t find any way to send back my modifications to the upstream project, I don’t really feel a need for another project.)

Then I fired up the logging process and it has been running in the background for a while now, logging every key stroke with a time stamp.

Counting key frequency and how it gets distributed very quickly turns into basically seeing when I’m active in front of the computer and it also gave me thoughts around what a high key frequency actually means in terms of activity and productivity. Does a really high key frequency really mean that I was working intensely or isn’t that purpose more a sign of mail sending time? When I debug problems or research details, won’t those periods result in slower key activity?

In the end I guess that over time, the key frequency chart basically says that if I have pressed a lot of keys during a period, I was working on something then. Hours or days with a very low average key frequency are probably times when I don’t work as much.

The weekend key frequency is bound to be slightly wrong due to me sometimes doing weekend hacking on other computers where I don’t log the keys since my results are recorded from a single specific keyboard only.

Conclusions

So what did I learn? Here are some conclusions and results from 1276614 keystrokes done over a period of the most recent 52 calendar days.

I have a 105-key keyboard, but during this period I only pressed 90 unique keys. Out of the 90 keys I pressed, 3 were pressed more than 5% of the time – each. In fact, those 3 keys are more than 20% of all keystrokes. Those keys are: <Space>, <Backspace> and the letter ‘e’.

<Space> stands out from all the rest as it has been used more than 10%.

Only 29 keys were used more than 1% of the presses, giving this a really long tail with lots of keys hardly ever used.

Over this logged time, I have registered key strokes during 46% of all hours. Counting only the hours in which I actually used the keyboard, the average number of key strokes were 2185/hour, 36 keys/minute.

The average week day (excluding weekend days), I registered 32486 key presses. The most active sinngle minute during this logging period, I hit 405 keys. The most active single hour I managed to do 7937 key presses. During weekends my activity is much lower, and then I average at 5778 keys/day (7.2% of all activity were weekends).

When counting most active hours over the day, there are 14 hours that have more than 1% activity and there are 5 with less than 1%, leaving 5 hours with no keyboard activity at all (02:00- 06:59). Interestingly, the hour between 23-24 at night is the single most busy hour for me, with 12.5% of all keypresses during the period.

Random “anecdotes”

Longest contiguous time without keys: 26.4 hours

Longest key sequence without backspace: 946

There are 7 keys I only pressed once during this period; 4 of them are on the numerical keypad and the other three are F10, F3 and <Pause>.

More

I’ll try to keep the logging going and see if things change over time or if there later might end up things that can be seen in the data when looked over a longer period.

Changing networks on Mac with Firefox

Thursday, October 30th, 2014

Not too long ago I blogged about my work to better deal with changing networks while Firefox is running. That job was basically two parts.

A) generic code to handle receiving such a network-changed event and then

B) a platform specific part that was for Windows that detected such a network change and sent the event

Today I’ve landed yet another fix for part B called bug 1079385, which detects network changes for Firefox on Mac OS X.

mac miniI’ve never programmed anything before on the Mac so this was sort of my christening in this environment. I mean, I’ve written countless of POSIX compliant programs including curl and friends that certainly builds and runs on Mac OS just fine, but I never before used the Mac-specific APIs to do things.

I got a mac mini just two weeks ago to work on this. Getting it up, prepared and my first Firefox built from source took all-in-all less than three hours. Learning the details of the mac API world was much more trouble and can’t say that I’m mastering it now either but I did find myself at least figuring out how to detect when IP addresses on the interfaces change and a changed address is a pretty good signal that the network changed somehow.

daniel.haxx.se episode 8

Monday, October 27th, 2014

Today I hesitated to make my new weekly video episode. I looked at the viewers number and how they basically have dwindled the last few weeks. I’m not making this video series interesting enough for a very large crowd of people. I’m re-evaluating if I should do them at all, or if I can do something to spice them up…

… or perhaps just not look at the viewers numbers at all and just do what think is fun?

I decided I’ll go with the latter for now. After all, I enjoy making these and they usually give me some interesting feedback and discussions even if the numbers are really low. What good is a number anyway?

This week’s episode:

Personal

Firefox

Fun

HTTP/2

TALKS

  • I’m offering two talks for FOSDEM

curl

  • release next Wednesday
  • bug fixing period
  • security advisory is pending

wget

Stricter HTTP 1.1 framing good bye

Sunday, October 26th, 2014

I worked on a patch for Firefox bug 237623 to make sure Firefox would use a stricter check for “HTTP 1.1 framing”, checking that Content-Length is correct and that there’s no broken chunked encoding pieces. I was happy to close an over 10 years old bug when the fix landed in June 2014.

The fix landed and has not caused any grief all the way since June through to the actual live release (Nightlies, Aurora, Beta etc). This change finally shipped in Firefox 33 and I had more or less already started to forget about it, and now things went south really fast.

The amount of broken servers ended up too massive for us and we had to backpedal. The largest amount of problems can be split up in these two categories:

  1. Servers that deliver gzipped content and sends a Content-Length: for the uncompressed data. This seems to be commonly done with old mod_deflate and mod_fastcgi versions on Apache, but we also saw people using IIS reporting this symptom.
  2. Servers that deliver chunked-encoding but who skip the final zero-size chunk so that the stream actually never really ends.

We recognize that not everyone can have the servers fixed – even if all these servers should still be fixed! We now make these HTTP 1.1 framing problems get detected but only cause a problem if a certain pref variable is set (network.http.enforce-framing.http1), and since that is disabled by default they will be silently ignored much like before. The Internet is a more broken and more sad place than I want to accept at times.

We haven’t fully worked out how to also make the download manager (ie the thing that downloads things directly to disk, without showing it in the browser) happy, which was the original reason for bug 237623…

Although the code may now no longer alert anything about HTTP 1.1 framing problems, it will now at least mark the connection not due for re-use which will be a big boost compared to before since these broken framing cases really hurt persistent connections use. The partial transfer return codes for broken SPDY and HTTP/2 transfers remain though and I hope to be able to remain stricter with these newer protocols.

This partial reversion will land ASAP and get merged into patch releases of Firefox 33 and later.

Finally, to top this off. Here’s a picture of an old HTTP 1.1 frame so that you know what we’re talking about.

An old http1.1 frame