Search Results: "rra"

31 May 2022

Russ Allbery: Review: Maskerade

Review: Maskerade, by Terry Pratchett
Series: Discworld #18
Publisher: Harper
Copyright: 1995
Printing: February 2014
ISBN: 0-06-227552-6
Format: Mass market
Pages: 360
Maskerade is the 18th book of the Discworld series, but you probably could start here. You'd miss the introduction of Granny Weatherwax and Nanny Ogg, which might be a bit confusing, but I suspect you could pick it up as you went if you wanted. This is a sequel of sorts to Lords and Ladies, but not in a very immediate sense. Granny is getting distracted and less interested in day-to-day witching in Lancre. This is not good; Granny is incredibly powerful, and bored and distracted witches can go to dark places. Nanny is concerned. Granny needs something to do, and their coven needs a third. It's not been the same since they lost their maiden member. Nanny's solution to this problem is two-pronged. First, they'd had their eye on a local girl named Agnes, who had magic but who wasn't interested in being a witch. Perhaps it was time to recruit her anyway, even though she'd left Lancre for Ankh-Morpork. And second, Granny needs something to light a fire under her, something that will get her outraged and ready to engage with the world. Something like a cookbook of aphrodisiac recipes attributed to the Witch of Lancre. Agnes, meanwhile, is auditioning for the opera. She's a sensible person, cursed her whole life by having a wonderful personality, but a part of her deep inside wants to be called Perdita X. Dream and have a dramatic life. Having a wonderful personality can be very frustrating, but no one in Lancre took either that desire or her name seriously. Perhaps the opera is somewhere where she can find the life she's looking for, along with another opportunity to try on the Perdita name. One thing she can do is sing; that's where all of her magic went. The Ankh-Morpork opera is indeed dramatic. It's also losing an astounding amount of money for its new owner, who foolishly thought owning an opera would be a good retirement project after running a cheese business. And it's haunted by a ghost, a very tangible ghost who has started killing people. I think this is my favorite Discworld novel to date (although with a caveat about the ending that I'll get to in a moment). It's certainly the one that had me laughing out loud the most. Agnes (including her Perdita personality aspect) shot to the top of my list of favorite Discworld characters, in part because I found her sensible personality so utterly relatable. She is fascinated by drama, she wants to be in the middle of it and let her inner Perdita goth character revel in it, and yet she cannot help being practical and unflappable even when surrounded by people who use far too many exclamation points. It's one thing to want drama in the abstract; it's quite another to be heedlessly dramatic in the moment, when there's an obviously reasonable thing to do instead. Pratchett writes this wonderfully. The other half of the story follows Granny and Nanny, who are unstoppable forces of nature and a wonderful team. They have the sort of long-standing, unshakable adult friendship between very unlike people that's full of banter and minor irritations layered on top of a deep mutual understanding and respect. Once they decide to start investigating this supposed opera ghost, they divvy up the investigative work with hardly a word exchanged. Planning isn't necessary; they both know each other's strengths. We've gotten a lot of Granny's skills in previous books. Maskerade gives Nanny a chance to show off her skills, and it's a delight. She effortlessly becomes the sort of friendly grandmother who blends in so well that no one questions why she's there, and thus manages to be in the middle of every important event. Granny watches and thinks and theorizes; Nanny simply gets into the middle of everything and talks to everyone until people tell her what she wants to know. There's no real doubt that the two of them are going to get to the bottom of anything they want to get to the bottom of, but watching how they get there is a delight. I love how Pratchett handles that sort of magical power from a world-building perspective. Ankh-Morpork is the Big City, the center of political power in most of the Discworld books, and Granny and Nanny are from the boondocks. By convention, that means they should either be awed or confused by the city, or gain power in the city by transforming it in some way to match their area of power. This isn't how Pratchett writes witches at all. Their magic is in understanding people, and the people in Ankh-Morpork are just as much people as the people in Lancre. The differences of the city may warrant an occasional grumpy aside, but the witches are fully as capable of navigating the city as they are their home town. Maskerade is, of course, a parody of opera and musicals, with Phantom of the Opera playing the central role in much the same way that Macbeth did in Wyrd Sisters. Agnes ends up doing the singing for a beautiful, thin actress named Christine, who can't sing at all despite being an opera star, uses a truly astonishing excess of exclamation points, and strategically faints at the first sign of danger. (And, despite all of this, is still likable in that way that it's impossible to be really upset at a puppy.) She is the special chosen focus of the ghost, whose murderous taunting is a direct parody of the Phantom. That was a sufficiently obvious reference that even I picked up on it, despite being familiar with Phantom of the Opera only via the soundtrack. Apart from that, though, the references were lost on me, since I'm neither a musical nor an opera fan. That didn't hurt my enjoyment of the book in the slightest; in fact, I suspect it's part of why it's in my top tier of Discworld books. One of my complaints about Discworld to date is that Pratchett often overdoes the parody to the extent that it gets in the way of his own (excellent) characters and story. Maybe it's better to read Discworld novels where one doesn't recognize the material being parodied and thus doesn't keep getting distracted by references. It's probably worth mentioning that Agnes is a large woman and there are several jokes about her weight in Maskerade. I think they're the good sort of jokes, about how absurd human bodies can be, not the mean sort? Pratchett never implies her weight is any sort of moral failing or something she should change; quite the contrary, Nanny considers it a sign of solid Lancre genes. But there is some fat discrimination in the opera itself, since one of the things Pratchett is commenting on is the switch from full-bodied female opera singers to thin actresses matching an idealized beauty standard. Christine is the latter, but she can't sing, and the solution is for Agnes to sing for her from behind, something that was also done in real opera. I'm not a good judge of how well this plot line was handled; be aware, going in, if this may bother you. What did bother me was the ending, and more generally the degree to which Granny and Nanny felt comfortable making decisions about Agnes's life without consulting her or appearing to care what she thought of their conclusions. Pratchett seemed to be on their side, emphasizing how well they know people. But Agnes left Lancre and avoided the witches for a reason, and that reason is not honored in much the same way that Lancre refused to honor her desire to go by Perdita. This doesn't seem to be malicious, and Agnes herself is a little uncertain about her choice of identity, but it still rubbed me the wrong way. I felt like Agnes got steamrolled by both the other characters and by Pratchett, and it's the one thing about this book that I didn't like. Hopefully future Discworld books about these characters revisit Agnes's agency. Overall, though, this was great, and a huge improvement over Interesting Times. I'm excited for the next witches book. Followed in publication order by Feet of Clay, and later by Carpe Jugulum in the thematic sense. Rating: 8 out of 10

18 May 2022

Reproducible Builds: Supporter spotlight: Jan Nieuwenhuizen on Bootstrappable Builds, GNU Mes and GNU Guix

The Reproducible Builds project relies on several projects, supporters and sponsors for financial support, but they are also valued as ambassadors who spread the word about our project and the work that we do. This is the fourth instalment in a series featuring the projects, companies and individuals who support the Reproducible Builds project. We started this series by featuring the Civil Infrastructure Platform project and followed this up with a post about the Ford Foundation as well as a recent ones about ARDC and the Google Open Source Security Team (GOSST). Today, however, we will be talking with Jan Nieuwenhuizen about Bootstrappable Builds, GNU Mes and GNU Guix.
Chris Lamb: Hi Jan, thanks for taking the time to talk with us today. First, could you briefly tell me about yourself? Jan: Thanks for the chat; it s been a while! Well, I ve always been trying to find something new and interesting that is just asking to be created but is mostly being overlooked. That s how I came to work on GNU Guix and create GNU Mes to address the bootstrapping problem that we have in free software. It s also why I have been working on releasing Dezyne, a programming language and set of tools to specify and formally verify concurrent software systems as free software. Briefly summarised, compilers are often written in the language they are compiling. This creates a chicken-and-egg problem which leads users and distributors to rely on opaque, pre-built binaries of those compilers that they use to build newer versions of the compiler. To gain trust in our computing platforms, we need to be able to tell how each part was produced from source, and opaque binaries are a threat to user security and user freedom since they are not auditable. The goal of bootstrappability (and the bootstrappable.org project in particular) is to minimise the amount of these bootstrap binaries. Anyway, after studying Physics at Eindhoven University of Technology (TU/e), I worked for digicash.com, a startup trying to create a digital and anonymous payment system sadly, however, a traditional account-based system won. Separate to this, as there was no software (either free or proprietary) to automatically create beautiful music notation, together with Han-Wen Nienhuys, I created GNU LilyPond. Ten years ago, I took the initiative to co-found a democratic school in Eindhoven based on the principles of sociocracy. And last Christmas I finally went vegan, after being mostly vegetarian for about 20 years!
Chris: For those who have not heard of it before, what is GNU Guix? What are the key differences between Guix and other Linux distributions? Jan: GNU Guix is both a package manager and a full-fledged GNU/Linux distribution. In both forms, it provides state-of-the-art package management features such as transactional upgrades and package roll-backs, hermetical-sealed build environments, unprivileged package management as well as per-user profiles. One obvious difference is that Guix forgoes the usual Filesystem Hierarchy Standard (ie. /usr, /lib, etc.), but there are other significant differences too, such as Guix being scriptable using Guile/Scheme, as well as Guix s dedication and focus on free software.
Chris: How does GNU Guix relate to GNU Mes? Or, rather, what problem is Mes attempting to solve? Jan: GNU Mes was created to address the security concerns that arise from bootstrapping an operating system such as Guix. Even if this process entirely involves free software (i.e. the source code is, at least, available), this commonly uses large and unauditable binary blobs. Mes is a Scheme interpreter written in a simple subset of C and a C compiler written in Scheme, and it comes with a small, bootstrappable C library. Twice, the Mes bootstrap has halved the size of opaque binaries that were needed to bootstrap GNU Guix. These reductions were achieved by first replacing GNU Binutils, GNU GCC and the GNU C Library with Mes, and then replacing Unix utilities such as awk, bash, coreutils, grep sed, etc., by Gash and Gash-Utils. The final goal of Mes is to help create a full-source bootstrap for any interested UNIX-like operating system.
Chris: What is the current status of Mes? Jan: Mes supports all that is needed from R5RS and GNU Guile to run MesCC with Nyacc, the C parser written for Guile, for 32-bit x86 and ARM. The next step for Mes would be more compatible with Guile, e.g., have guile-module support and support running Gash and Gash Utils. In working to create a full-source bootstrap, I have disregarded the kernel and Guix build system for now, but otherwise, all packages should be built from source, and obviously, no binary blobs should go in. We still need a Guile binary to execute some scripts, and it will take at least another one to two years to remove that binary. I m using the 80/20 approach, cutting corners initially to get something working and useful early. Another metric would be how many architectures we have. We are quite a way with ARM, tinycc now works, but there are still problems with GCC and Glibc. RISC-V is coming, too, which could be another metric. Someone has looked into picking up NixOS this summer. How many distros do anything about reproducibility or bootstrappability? The bootstrappability community is so small that we don t need metrics, sadly. The number of bytes of binary seed is a nice metric, but running the whole thing on a full-fledged Linux system is tough to put into a metric. Also, it is worth noting that I m developing on a modern Intel machine (ie. a platform with a management engine), that s another key component that doesn t have metrics.
Chris: From your perspective as a Mes/Guix user and developer, what does reproducibility mean to you? Are there any related projects? Jan: From my perspective, I m more into the problem of bootstrapping, and reproducibility is a prerequisite for bootstrappability. Reproducibility clearly takes a lot of effort to achieve, however. It s relatively easy to install some Linux distribution and be happy, but if you look at communities that really care about security, they are investing in reproducibility and other ways of improving the security of their supply chain. Projects I believe are complementary to Guix and Mes include NixOS, Debian and on the hardware side the RISC-V platform shares many of our core principles and goals.
Chris: Well, what are these principles and goals? Jan: Reproducibility and bootstrappability often feel like the next step in the frontier of free software. If you have all the sources and you can t reproduce a binary, that just doesn t feel right anymore. We should start to desire (and demand) transparent, elegant and auditable software stacks. To a certain extent, that s always been a low-level intent since the beginning of free software, but something clearly got lost along the way. On the other hand, if you look at the NPM or Rust ecosystems, we see a world where people directly install binaries. As they are not as supportive of copyleft as the rest of the free software community, you can see that movement and people in our area are doing more as a response to that so that what we have continues to remain free, and to prevent us from falling asleep and waking up in a couple of years and see, for example, Rust in the Linux kernel and (more importantly) we require big binary blobs to use our systems. It s an excellent time to advance right now, so we should get a foothold in and ensure we don t lose any more.
Chris: What would be your ultimate reproducibility goal? And what would the key steps or milestones be to reach that? Jan: The ultimate goal would be to have a system built with open hardware, with all software on it fully bootstrapped from its source. This bootstrap path should be easy to replicate and straightforward to inspect and audit. All fully reproducible, of course! In essence, we would have solved the supply chain security problem. Our biggest challenge is ignorance. There is much unawareness about the importance of what we are doing. As it is rather technical and doesn t really affect everyday computer use, that is not surprising. This unawareness can be a great force driving us in the opposite direction. Think of Rust being allowed in the Linux kernel, or Python being required to build a recent GNU C library (glibc). Also, the fact that companies like Google/Apple still want to play us vs them , not willing to to support GPL software. Not ready yet to truly support user freedom. Take the infamous log4j bug everyone is using open source these days, but nobody wants to take responsibility and help develop or nurture the community. Not ecosystem , as that s how it s being approached right now: live and let live/die: see what happens without taking any responsibility. We are growing and we are strong and we can do a lot but if we have to work against those powers, it can become problematic. So, let s spread our great message and get more people involved!
Chris: What has been your biggest win? Jan: From a technical point of view, the full-source bootstrap has have been our biggest win. A talk by Carl Dong at the 2019 Breaking Bitcoin conference stated that connecting Jeremiah Orian s Stage0 project to Mes would be the holy grail of bootstrapping, and we recently managed to achieve just that: in other words, starting from hex0, 357-byte binary, we can now build the entire Guix system. This past year we have not made significant visible progress, however, as our funding was unfortunately not there. The Stage0 project has advanced in RISC-V. A month ago, though, I secured NLnet funding for another year, and thanks to NLnet, Ekaitz Zarraga and Timothy Sample will work on GNU Mes and the Guix bootstrap as well. Separate to this, the bootstrappable community has grown a lot from two people it was six years ago: there are now currently over 100 people in the #bootstrappable IRC channel, for example. The enlarged community is possibly an even more important win going forward.
Chris: How realistic is a 100% bootstrappable toolchain? And from someone who has been working in this area for a while, is solving Trusting Trust) actually feasible in reality? Jan: Two answers: Yes and no, it really depends on your definition. One great thing is that the whole Stage0 project can also run on the Knight virtual machine, a hardware platform that was designed, I think, in the 1970s. I believe we can and must do better than we are doing today, and that there s a lot of value in it. The core issue is not the trust; we can probably all trust each other. On the other hand, we don t want to trust each other or even ourselves. I am not, personally, going to inspect my RISC-V laptop, and other people create the hardware and do probably not want to inspect the software. The answer comes back to being conscientious and doing what is right. Inserting GCC as a binary blob is not right. I think we can do better, and that s what I d like to do. The security angle is interesting, but I don t like the paranoid part of that; I like the beauty of what we are creating together and stepwise improving on that.
Chris: Thanks for taking the time to talk to us today. If someone wanted to get in touch or learn more about GNU Guix or Mes, where might someone go? Jan: Sure! First, check out: I m also on Twitter (@janneke_gnu) and on octodon.social (@janneke@octodon.social).
Chris: Thanks for taking the time to talk to us today. Jan: No problem. :)


For more information about the Reproducible Builds project, please see our website at reproducible-builds.org. If you are interested in ensuring the ongoing security of the software that underpins our civilisation and wish to sponsor the Reproducible Builds project, please reach out to the project by emailing contact@reproducible-builds.org.

16 May 2022

Matthew Garrett: Can we fix bearer tokens?

Last month I wrote about how bearer tokens are just awful, and a week later Github announced that someone had managed to exfiltrate bearer tokens from Heroku that gave them access to, well, a lot of Github repositories. This has inevitably resulted in a whole bunch of discussion about a number of things, but people seem to be largely ignoring the fundamental issue that maybe we just shouldn't have magical blobs that grant you access to basically everything even if you've copied them from a legitimate holder to Honest John's Totally Legitimate API Consumer.

To make it clearer what the problem is here, let's use an analogy. You have a safety deposit box. To gain access to it, you simply need to be able to open it with a key you were given. Anyone who turns up with the key can open the box and do whatever they want with the contents. Unfortunately, the key is extremely easy to copy - anyone who is able to get hold of your keyring for a moment is in a position to duplicate it, and then they have access to the box. Wouldn't it be better if something could be done to ensure that whoever showed up with a working key was someone who was actually authorised to have that key?

To achieve that we need some way to verify the identity of the person holding the key. In the physical world we have a range of ways to achieve this, from simply checking whether someone has a piece of ID that associates them with the safety deposit box all the way up to invasive biometric measurements that supposedly verify that they're definitely the same person. But computers don't have passports or fingerprints, so we need another way to identify them.

When you open a browser and try to connect to your bank, the bank's website provides a TLS certificate that lets your browser know that you're talking to your bank instead of someone pretending to be your bank. The spec allows this to be a bi-directional transaction - you can also prove your identity to the remote website. This is referred to as "mutual TLS", or mTLS, and a successful mTLS transaction ends up with both ends knowing who they're talking to, as long as they have a reason to trust the certificate they were presented with.

That's actually a pretty big constraint! We have a reasonable model for the server - it's something that's issued by a trusted third party and it's tied to the DNS name for the server in question. Clients don't tend to have stable DNS identity, and that makes the entire thing sort of awkward. But, thankfully, maybe we don't need to? We don't need the client to be able to prove its identity to arbitrary third party sites here - we just need the client to be able to prove it's a legitimate holder of whichever bearer token it's presenting to that site. And that's a much easier problem.

Here's the simple solution - clients generate a TLS cert. This can be self-signed, because all we want to do here is be able to verify whether the machine talking to us is the same one that had a token issued to it. The client contacts a service that's going to give it a bearer token. The service requests mTLS auth without being picky about the certificate that's presented. The service embeds a hash of that certificate in the token before handing it back to the client. Whenever the client presents that token to any other service, the service ensures that the mTLS cert the client presented matches the hash in the bearer token. Copy the token without copying the mTLS certificate and the token gets rejected. Hurrah hurrah hats for everyone.

Well except for the obvious problem that if you're in a position to exfiltrate the bearer tokens you can probably just steal the client certificates and keys as well, and now you can pretend to be the original client and this is not adding much additional security. Fortunately pretty much everything we care about has the ability to store the private half of an asymmetric key in hardware (TPMs on Linux and Windows systems, the Secure Enclave on Macs and iPhones, either a piece of magical hardware or Trustzone on Android) in a way that avoids anyone being able to just steal the key.

How do we know that the key is actually in hardware? Here's the fun bit - it doesn't matter. If you're issuing a bearer token to a system then you're already asserting that the system is trusted. If the system is lying to you about whether or not the key it's presenting is hardware-backed then you've already lost. If it lied and the system is later compromised then sure all your apes get stolen, but maybe don't run systems that lie and avoid that situation as a result?

Anyway. This is covered in RFC 8705 so why aren't we all doing this already? From the client side, the largest generic issue is that TPMs are astonishingly slow in comparison to doing a TLS handshake on the CPU. RSA signing operations on TPMs can take around half a second, which doesn't sound too bad, except your browser is probably establishing multiple TLS connections to subdomains on the site it's connecting to and performance is going to tank. Fixing this involves doing whatever's necessary to convince the browser to pipe everything over a single TLS connection, and that's just not really where the web is right at the moment. Using EC keys instead helps a lot (~0.1 seconds per signature on modern TPMs), but it's still going to be a bottleneck.

The other problem, of course, is that ecosystem support for hardware-backed certificates is just awful. Windows lets you stick them into the standard platform certificate store, but the docs for this are hidden in a random PDF in a Github repo. Macs require you to do some weird bridging between the Secure Enclave API and the keychain API. Linux? Well, the standard answer is to do PKCS#11, and I have literally never met anybody who likes PKCS#11 and I have spent a bunch of time in standards meetings with the sort of people you might expect to like PKCS#11 and even they don't like it. It turns out that loading a bunch of random C bullshit that has strong feelings about function pointers into your security critical process is not necessarily something that is going to improve your quality of life, so instead you should use something like this and just have enough C to bridge to a language that isn't secretly plotting to kill your pets the moment you turn your back.

And, uh, obviously none of this matters at all unless people actually support it. Github has no support at all for validating the identity of whoever holds a bearer token. Most issuers of bearer tokens have no support for embedding holder identity into the token. This is not good! As of last week, all three of the big cloud providers support virtualised TPMs in their VMs - we should be running CI on systems that can do that, and tying any issued tokens to the VMs that are supposed to be making use of them.

So sure this isn't trivial. But it's also not impossible, and making this stuff work would improve the security of, well, everything. We literally have the technology to prevent attacks like Github suffered. What do we have to do to get people to actually start working on implementing that?

comment count unavailable comments

13 May 2022

Antoine Beaupr : BTRFS notes

I'm not a fan of BTRFS. This page serves as a reminder of why, but also a cheat sheet to figure out basic tasks in a BTRFS environment because those are not obvious to me, even after repeatedly having to deal with them. Content warning: there might be mentions of ZFS.

Stability concerns I'm worried about BTRFS stability, which has been historically ... changing. RAID-5 and RAID-6 are still marked unstable, for example. It's kind of a lucky guess whether your current kernel will behave properly with your planned workload. For example, in Linux 4.9, RAID-1 and RAID-10 were marked as "mostly OK" with a note that says:
Needs to be able to create two copies always. Can get stuck in irreversible read-only mode if only one copy can be made.
Even as of now, RAID-1 and RAID-10 has this note:
The simple redundancy RAID levels utilize different mirrors in a way that does not achieve the maximum performance. The logic can be improved so the reads will spread over the mirrors evenly or based on device congestion.
Granted, that's not a stability concern anymore, just performance. A reviewer of a draft of this article actually claimed that BTRFS only reads from one of the drives, which hopefully is inaccurate, but goes to show how confusing all this is. There are other warnings in the Debian wiki that are quite scary. Even the legendary Arch wiki has a warning on top of their BTRFS page, still. Even if those issues are now fixed, it can be hard to tell when they were fixed. There is a changelog by feature but it explicitly warns that it doesn't know "which kernel version it is considered mature enough for production use", so it's also useless for this. It would have been much better if BTRFS was released into the world only when those bugs were being completely fixed. Or that, at least, features were announced when they were stable, not just "we merged to mainline, good luck". Even now, we get mixed messages even in the official BTRFS documentation which says "The Btrfs code base is stable" (main page) while at the same time clearly stating unstable parts in the status page (currently RAID56). There are much harsher BTRFS critics than me out there so I will stop here, but let's just say that I feel a little uncomfortable trusting server data with full RAID arrays to BTRFS. But surely, for a workstation, things should just work smoothly... Right? Well, let's see the snags I hit.

My BTRFS test setup Before I go any further, I should probably clarify how I am testing BTRFS in the first place. The reason I tried BTRFS is that I was ... let's just say "strongly encouraged" by the LWN editors to install Fedora for the terminal emulators series. That, in turn, meant the setup was done with BTRFS, because that was somewhat the default in Fedora 27 (or did I want to experiment? I don't remember, it's been too long already). So Fedora was setup on my 1TB HDD and, with encryption, the partition table looks like this:
NAME                   MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINT
sda                      8:0    0 931,5G  0 disk  
 sda1                   8:1    0   200M  0 part  /boot/efi
 sda2                   8:2    0     1G  0 part  /boot
 sda3                   8:3    0   7,8G  0 part  
   fedora_swap        253:5    0   7.8G  0 crypt [SWAP]
 sda4                   8:4    0 922,5G  0 part  
   fedora_crypt       253:4    0 922,5G  0 crypt /
(This might not entirely be accurate: I rebuilt this from the Debian side of things.) This is pretty straightforward, except for the swap partition: normally, I just treat swap like any other logical volume and create it in a logical volume. This is now just speculation, but I bet it was setup this way because "swap" support was only added in BTRFS 5.0. I fully expect BTRFS experts to yell at me now because this is an old setup and BTRFS is so much better now, but that's exactly the point here. That setup is not that old (2018? old? really?), and migrating to a new partition scheme isn't exactly practical right now. But let's move on to more practical considerations.

No builtin encryption BTRFS aims at replacing the entire mdadm, LVM, and ext4 stack with a single entity, and adding new features like deduplication, checksums and so on. Yet there is one feature it is critically missing: encryption. See, my typical stack is actually mdadm, LUKS, and then LVM and ext4. This is convenient because I have only a single volume to decrypt. If I were to use BTRFS on servers, I'd need to have one LUKS volume per-disk. For a simple RAID-1 array, that's not too bad: one extra key. But for large RAID-10 arrays, this gets really unwieldy. The obvious BTRFS alternative, ZFS, supports encryption out of the box and mixes it above the disks so you only have one passphrase to enter. The main downside of ZFS encryption is that it happens above the "pool" level so you can typically see filesystem names (and possibly snapshots, depending on how it is built), which is not the case with a more traditional stack.

Subvolumes, filesystems, and devices I find BTRFS's architecture to be utterly confusing. In the traditional LVM stack (which is itself kind of confusing if you're new to that stuff), you have those layers:
  • disks: let's say /dev/nvme0n1 and nvme1n1
  • RAID arrays with mdadm: let's say the above disks are joined in a RAID-1 array in /dev/md1
  • volume groups or VG with LVM: the above RAID device (technically a "physical volume" or PV) is assigned into a VG, let's call it vg_tbbuild05 (multiple PVs can be added to a single VG which is why there is that abstraction)
  • LVM logical volumes: out of that volume group actually "virtual partitions" or "logical volumes" are created, that is where your filesystem lives
  • filesystem, typically with ext4: that's your normal filesystem, which treats the logical volume as just another block device
A typical server setup would look like this:
NAME                      MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINT
nvme0n1                   259:0    0   1.7T  0 disk  
 nvme0n1p1               259:1    0     8M  0 part  
 nvme0n1p2               259:2    0   512M  0 part  
   md0                     9:0    0   511M  0 raid1 /boot
 nvme0n1p3               259:3    0   1.7T  0 part  
   md1                     9:1    0   1.7T  0 raid1 
     crypt_dev_md1       253:0    0   1.7T  0 crypt 
       vg_tbbuild05-root 253:1    0    30G  0 lvm   /
       vg_tbbuild05-swap 253:2    0 125.7G  0 lvm   [SWAP]
       vg_tbbuild05-srv  253:3    0   1.5T  0 lvm   /srv
 nvme0n1p4               259:4    0     1M  0 part
I stripped the other nvme1n1 disk because it's basically the same. Now, if we look at my BTRFS-enabled workstation, which doesn't even have RAID, we have the following:
  • disk: /dev/sda with, again, /dev/sda4 being where BTRFS lives
  • filesystem: fedora_crypt, which is, confusingly, kind of like a volume group. it's where everything lives. i think.
  • subvolumes: home, root, /, etc. those are actually the things that get mounted. you'd think you'd mount a filesystem, but no, you mount a subvolume. that is backwards.
It looks something like this to lsblk:
NAME                   MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINT
sda                      8:0    0 931,5G  0 disk  
 sda1                   8:1    0   200M  0 part  /boot/efi
 sda2                   8:2    0     1G  0 part  /boot
 sda3                   8:3    0   7,8G  0 part  [SWAP]
 sda4                   8:4    0 922,5G  0 part  
   fedora_crypt       253:4    0 922,5G  0 crypt /srv
Notice how we don't see all the BTRFS volumes here? Maybe it's because I'm mounting this from the Debian side, but lsblk definitely gets confused here. I frankly don't quite understand what's going on, even after repeatedly looking around the rather dismal documentation. But that's what I gather from the following commands:
root@curie:/home/anarcat# btrfs filesystem show
Label: 'fedora'  uuid: 5abb9def-c725-44ef-a45e-d72657803f37
    Total devices 1 FS bytes used 883.29GiB
    devid    1 size 922.47GiB used 916.47GiB path /dev/mapper/fedora_crypt
root@curie:/home/anarcat# btrfs subvolume list /srv
ID 257 gen 108092 top level 5 path home
ID 258 gen 108094 top level 5 path root
ID 263 gen 108020 top level 258 path root/var/lib/machines
I only got to that point through trial and error. Notice how I use an existing mountpoint to list the related subvolumes. If I try to use the filesystem path, the one that's listed in filesystem show, I fail:
root@curie:/home/anarcat# btrfs subvolume list /dev/mapper/fedora_crypt 
ERROR: not a btrfs filesystem: /dev/mapper/fedora_crypt
ERROR: can't access '/dev/mapper/fedora_crypt'
Maybe I just need to use the label? Nope:
root@curie:/home/anarcat# btrfs subvolume list fedora
ERROR: cannot access 'fedora': No such file or directory
ERROR: can't access 'fedora'
This is really confusing. I don't even know if I understand this right, and I've been staring at this all afternoon. Hopefully, the lazyweb will correct me eventually. (As an aside, why are they called "subvolumes"? If something is a "sub" of "something else", that "something else" must exist right? But no, BTRFS doesn't have "volumes", it only has "subvolumes". Go figure. Presumably the filesystem still holds "files" though, at least empirically it doesn't seem like it lost anything so far. In any case, at least I can refer to this section in the future, the next time I fumble around the btrfs commandline, as I surely will. I will possibly even update this section as I get better at it, or based on my reader's judicious feedback.

Mounting BTRFS subvolumes So how did I even get to that point? I have this in my /etc/fstab, on the Debian side of things:
UUID=5abb9def-c725-44ef-a45e-d72657803f37   /srv    btrfs  defaults 0   2
This thankfully ignores all the subvolume nonsense because it relies on the UUID. mount tells me that's actually the "root" (? /?) subvolume:
root@curie:/home/anarcat# mount   grep /srv
/dev/mapper/fedora_crypt on /srv type btrfs (rw,relatime,space_cache,subvolid=5,subvol=/)
Let's see if I can mount the other volumes I have on there. Remember that subvolume list showed I had home, root, and var/lib/machines. Let's try root:
mount -o subvol=root /dev/mapper/fedora_crypt /mnt
Interestingly, root is not the same as /, it's a different subvolume! It seems to be the Fedora root (/, really) filesystem. No idea what is happening here. I also have a home subvolume, let's mount it too, for good measure:
mount -o subvol=home /dev/mapper/fedora_crypt /mnt/home
Note that lsblk doesn't notice those two new mountpoints, and that's normal: it only lists block devices and subvolumes (rather inconveniently, I'd say) do not show up as devices:
root@curie:/home/anarcat# lsblk 
NAME                   MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINT
sda                      8:0    0 931,5G  0 disk  
 sda1                   8:1    0   200M  0 part  
 sda2                   8:2    0     1G  0 part  
 sda3                   8:3    0   7,8G  0 part  
 sda4                   8:4    0 922,5G  0 part  
   fedora_crypt       253:4    0 922,5G  0 crypt /srv
This is really, really confusing. Maybe I did something wrong in the setup. Maybe it's because I'm mounting it from outside Fedora. Either way, it just doesn't feel right.

No disk usage per volume If you want to see what's taking up space in one of those subvolumes, tough luck:
root@curie:/home/anarcat# df -h  /srv /mnt /mnt/home
Filesystem                Size  Used Avail Use% Mounted on
/dev/mapper/fedora_crypt  923G  886G   31G  97% /srv
/dev/mapper/fedora_crypt  923G  886G   31G  97% /mnt
/dev/mapper/fedora_crypt  923G  886G   31G  97% /mnt/home
(Notice, in passing, that it looks like the same filesystem is mounted in different places. In that sense, you'd expect /srv and /mnt (and /mnt/home?!) to be exactly the same, but no: they are entirely different directory structures, which I will not call "filesystems" here because everyone's head will explode in sparks of confusion.) Yes, disk space is shared (that's the Size and Avail columns, makes sense). But nope, no cookie for you: they all have the same Used columns, so you need to actually walk the entire filesystem to figure out what each disk takes. (For future reference, that's basically:
root@curie:/home/anarcat# time du -schx /mnt/home /mnt /srv
124M    /mnt/home
7.5G    /mnt
875G    /srv
883G    total
real    2m49.080s
user    0m3.664s
sys 0m19.013s
And yes, that was painfully slow.) ZFS actually has some oddities in that regard, but at least it tells me how much disk each volume (and snapshot) takes:
root@tubman:~# time df -t zfs -h
Filesystem         Size  Used Avail Use% Mounted on
rpool/ROOT/debian  3.5T  1.4G  3.5T   1% /
rpool/var/tmp      3.5T  384K  3.5T   1% /var/tmp
rpool/var/spool    3.5T  256K  3.5T   1% /var/spool
rpool/var/log      3.5T  2.0G  3.5T   1% /var/log
rpool/home/root    3.5T  2.2G  3.5T   1% /root
rpool/home         3.5T  256K  3.5T   1% /home
rpool/srv          3.5T   80G  3.5T   3% /srv
rpool/var/cache    3.5T  114M  3.5T   1% /var/cache
bpool/BOOT/debian  571M   90M  481M  16% /boot
real    0m0.003s
user    0m0.002s
sys 0m0.000s
That's 56360 times faster, by the way. But yes, that's not fair: those in the know will know there's a different command to do what df does with BTRFS filesystems, the btrfs filesystem usage command:
root@curie:/home/anarcat# time btrfs filesystem usage /srv
Overall:
    Device size:         922.47GiB
    Device allocated:        916.47GiB
    Device unallocated:        6.00GiB
    Device missing:          0.00B
    Used:            884.97GiB
    Free (estimated):         30.84GiB  (min: 27.84GiB)
    Free (statfs, df):        30.84GiB
    Data ratio:               1.00
    Metadata ratio:           2.00
    Global reserve:      512.00MiB  (used: 0.00B)
    Multiple profiles:              no
Data,single: Size:906.45GiB, Used:881.61GiB (97.26%)
   /dev/mapper/fedora_crypt  906.45GiB
Metadata,DUP: Size:5.00GiB, Used:1.68GiB (33.58%)
   /dev/mapper/fedora_crypt   10.00GiB
System,DUP: Size:8.00MiB, Used:128.00KiB (1.56%)
   /dev/mapper/fedora_crypt   16.00MiB
Unallocated:
   /dev/mapper/fedora_crypt    6.00GiB
real    0m0,004s
user    0m0,000s
sys 0m0,004s
Almost as fast as ZFS's df! Good job. But wait. That doesn't actually tell me usage per subvolume. Notice it's filesystem usage, not subvolume usage, which unhelpfully refuses to exist. That command only shows that one "filesystem" internal statistics that are pretty opaque.. You can also appreciate that it's wasting 6GB of "unallocated" disk space there: I probably did something Very Wrong and should be punished by Hacker News. I also wonder why it has 1.68GB of "metadata" used... At this point, I just really want to throw that thing out of the window and restart from scratch. I don't really feel like learning the BTRFS internals, as they seem oblique and completely bizarre to me. It feels a little like the state of PHP now: it's actually pretty solid, but built upon so many layers of cruft that I still feel it corrupts my brain every time I have to deal with it (needle or haystack first? anyone?)...

Conclusion I find BTRFS utterly confusing and I'm worried about its reliability. I think a lot of work is needed on usability and coherence before I even consider running this anywhere else than a lab, and that's really too bad, because there are really nice features in BTRFS that would greatly help my workflow. (I want to use filesystem snapshots as high-performance, high frequency backups.) So now I'm experimenting with OpenZFS. It's so much simpler, just works, and it's rock solid. After this 8 minute read, I had a good understanding of how ZFS worked. Here's the 30 seconds overview:
  • vdev: a RAID array
  • vpool: a volume group of vdevs
  • datasets: normal filesystems (or block device, if you want to use another filesystem on top of ZFS)
There's also other special volumes like caches and logs that you can (really easily, compared to LVM caching) use to tweak your setup. You might also want to look at recordsize or ashift to tweak the filesystem to fit better your workload (or deal with drives lying about their sector size, I'm looking at you Samsung), but that's it. Running ZFS on Linux currently involves building kernel modules from scratch on every host, which I think is pretty bad. But I was able to setup a ZFS-only server using this excellent documentation without too much problem. I'm hoping some day the copyright issues are resolved and we can at least ship binary packages, but the politics (e.g. convincing Debian that is the right thing to do) and the logistics (e.g. DKMS auto-builders? is that even a thing? how about signed DKMS packages? fun-fun-fun!) seem really impractical. Who knows, maybe hell will freeze over (again) and Oracle will fix the CDDL. I personally think that we should just completely ignore this problem (which wasn't even supposed to be a problem) and ship binary packages directly, but I'm a pragmatic and do not always fit well with the free software fundamentalists. All of this to say that, short term, we don't have a reliable, advanced filesystem/logical disk manager in Linux. And that's really too bad.

10 May 2022

Russell Coker: Elon and Free Speech

Elon Musk has made the news for spending billions to buy a share of Twitter for the alleged purpose of providing free speech. The problem with this claim is that having any company controlling a large portion of the world s communication is inherently bad for free speech. The same applies for Facebook, but that s not a hot news item at the moment. If Elon wanted to provide free speech he would want to have decentralised messaging systems so that someone who breaks rules on one platform could find another with different rules. Among other things free speech ideally permits people to debate issues with residents of another country on issues related to different laws. If advocates for the Russian government get kicked off Twitter as part of the American sanctions against Russia then American citizens can t debate the issue with Russian citizens via Twitter. Mastodon is one example of a federated competitor to Twitter [1]. With a federated messaging system each host could make independent decisions about interpretation of sanctions. Someone who used a Mastodon instance based in the US could get a second account in another country if they wanted to communicate with people in countries that are sanctioned by the US. The problem with Mastodon at the moment is lack of use. It s got a good set of features and support for different platforms, there are apps for Android and iPhone as well as lots of other software using the API. But if the people you want to communicate with aren t on it then it s less useful. Elon could solve that problem by creating a Tesla Mastodon server and give a free account to everyone who buys a new Tesla, which is the sort of thing that a lot of Tesla buyers would like. It s quite likely that other companies selling prestige products would follow that example. Everyone has seen evidence of people sharing photos on social media with someone else s expensive car, a Mastodon account on ferrari.com or mercedes.com would be proof of buying the cars in question. The number of people who buy expensive cars new is a very small portion of the world population, but it s a group of people who are more influential than average and others would join Mastodon servers to follow them. The next thing that Elon could do to kill Twitter would be to have all his companies (which have something more than a dozen verified Twitter accounts) use Mastodon accounts for their primary PR releases and then send the same content to Twitter with a 48 hour delay. That would force journalists and people who want to discuss those companies on social media to follow the Mastodon accounts. Again this wouldn t be a significant number of people, but they would be influential people. Getting journalists to use a communications system increases it s importance. The question is whether Elon is lacking the vision necessary to plan a Mastodon deployment or whether he just wants to allow horrible people to run wild on Twitter. The Verge has an interesting article from 2019 about Gab using Mastodon [2]. The fact that over the last 2.5 years I didn t even hear of Gab using Mastodon suggests that the fears of some people significantly exceeded the problem. I m sure that some Gab users managed to harass some Mastodon users, but generally they were apparently banned quickly. As an aside the Mastodon server I use doesn t appear to ban Gab, a search for Gab on it gave me a user posting about being pureblood at the top of the list. Gab claims to have 4 million accounts and has an estimated 100,000 active users. If 5.5% of Tesla owners became active users on a hypothetical Tesla server that would be the largest Mastodon server. Elon could demonstrate his commitment to free speech by refusing to ban Gab in any way. The Wikipedia page about Gab [3] has a long list of horrible people and activities associated with it. Is that the free speech to associate with Tesla? Polestar makes some nice electric cars that appear quite luxurious [4] and doesn t get negative PR from the behaviour of it s owner, that s something Elon might want to consider. Is this really about bragging rights? Buying a controlling interest in a company that has a partial monopoly on Internet communication is something to boast about. Could users of commercial social media be considered serfs who serve their billionaire overlord?

Melissa Wen: Multiple syncobjs support for V3D(V) (Part 2)

In the previous post, I described how we enable multiple syncobjs capabilities in the V3D kernel driver. Now I will tell you what was changed on the userspace side, where we reworked the V3DV sync mechanisms to use Vulkan multiple wait and signal semaphores directly. This change represents greater adherence to the Vulkan submission framework. I was not used to Vulkan concepts and the V3DV driver. Fortunately, I counted on the guidance of the Igalia s Graphics team, mainly Iago Toral (thanks!), to understand the Vulkan Graphics Pipeline, sync scopes, and submission order. Therefore, we changed the original V3DV implementation for vkQueueSubmit and all related functions to allow direct mapping of multiple semaphores from V3DV to the V3D-kernel interface. Disclaimer: Here s a brief and probably inaccurate background, which we ll go into more detail later on. In Vulkan, GPU work submissions are described as command buffers. These command buffers, with GPU jobs, are grouped in a command buffer submission batch, specified by vkSubmitInfo, and submitted to a queue for execution. vkQueueSubmit is the command called to submit command buffers to a queue. Besides command buffers, vkSubmitInfo also specifies semaphores to wait before starting the batch execution and semaphores to signal when all command buffers in the batch are complete. Moreover, a fence in vkQueueSubmit can be signaled when all command buffer batches have completed execution. From this sequence, we can see some implicit ordering guarantees. Submission order defines the start order of execution between command buffers, in other words, it is determined by the order in which pSubmits appear in VkQueueSubmit and pCommandBuffers appear in VkSubmitInfo. However, we don t have any completion guarantees for jobs submitted to different GPU queue, which means they may overlap and complete out of order. Of course, jobs submitted to the same GPU engine follow start and finish order. A fence is ordered after all semaphores signal operations for signal operation order. In addition to implicit sync, we also have some explicit sync resources, such as semaphores, fences, and events. Considering these implicit and explicit sync mechanisms, we rework the V3DV implementation of queue submissions to better use multiple syncobjs capabilities from the kernel. In this merge request, you can find this work: v3dv: add support to multiple wait and signal semaphores. In this blog post, we run through each scope of change of this merge request for a V3D driver-guided description of the multisync support implementation.

Groundwork and basic code clean-up: As the original V3D-kernel interface allowed only one semaphore, V3DV resorted to booleans to translate multiple semaphores into one. Consequently, if a command buffer batch had at least one semaphore, it needed to wait on all jobs submitted complete before starting its execution. So, instead of just boolean, we created and changed structs that store semaphores information to accept the actual list of wait semaphores.

Expose multisync kernel interface to the driver: In the two commits below, we basically updated the DRM V3D interface from that one defined in the kernel and verified if the multisync capability is available for use.

Handle multiple semaphores for all GPU job types: At this point, we were only changing the submission design to consider multiple wait semaphores. Before supporting multisync, V3DV was waiting for the last job submitted to be signaled when at least one wait semaphore was defined, even when serialization wasn t required. V3DV handle GPU jobs according to the GPU queue in which they are submitted:
  • Control List (CL) for binning and rendering
  • Texture Formatting Unit (TFU)
  • Compute Shader Dispatch (CSD)
Therefore, we changed their submission setup to do jobs submitted to any GPU queues able to handle more than one wait semaphores. These commits created all mechanisms to set arrays of wait and signal semaphores for GPU job submissions:
  • Checking the conditions to define the wait_stage.
  • Wrapping them in a multisync extension.
  • According to the kernel interface (described in the previous blog post), configure the generic extension as a multisync extension.
Finally, we extended the ability of GPU jobs to handle multiple signal semaphores, but at this point, no GPU job is actually in charge of signaling them. With this in place, we could rework part of the code that tracks CPU and GPU job completions by verifying the GPU status and threads spawned by Event jobs.

Rework the QueueWaitIdle mechanism to track the syncobj of the last job submitted in each queue: As we had only single in/out syncobj interfaces for semaphores, we used a single last_job_sync to synchronize job dependencies of the previous submission. Although the DRM scheduler guarantees the order of starting to execute a job in the same queue in the kernel space, the order of completion isn t predictable. On the other hand, we still needed to use syncobjs to follow job completion since we have event threads on the CPU side. Therefore, a more accurate implementation requires last_job syncobjs to track when each engine (CL, TFU, and CSD) is idle. We also needed to keep the driver working on previous versions of v3d kernel-driver with single semaphores, then we kept tracking ANY last_job_sync to preserve the previous implementation.

Rework synchronization and submission design to let the jobs handle wait and signal semaphores: With multiple semaphores support, the conditions for waiting and signaling semaphores changed accordingly to the particularities of each GPU job (CL, CSD, TFU) and CPU job restrictions (Events, CSD indirect, etc.). In this sense, we redesigned V3DV semaphores handling and job submissions for command buffer batches in vkQueueSubmit. We scrutinized possible scenarios for submitting command buffer batches to change the original implementation carefully. It resulted in three commits more: We keep track of whether we have submitted a job to each GPU queue (CSD, TFU, CL) and a CPU job for each command buffer. We use syncobjs to track the last job submitted to each GPU queue and a flag that indicates if this represents the beginning of a command buffer. The first GPU job submitted to a GPU queue in a command buffer should wait on wait semaphores. The first CPU job submitted in a command buffer should call v3dv_QueueWaitIdle() to do the waiting and ignore semaphores (because it is waiting for everything). If the job is not the first but has the serialize flag set, it should wait on the completion of all last job submitted to any GPU queue before running. In practice, it means using syncobjs to track the last job submitted by queue and add these syncobjs as job dependencies of this serialized job. If this job is the last job of a command buffer batch, it may be used to signal semaphores if this command buffer batch has only one type of GPU job (because we have guarantees of execution ordering). Otherwise, we emit a no-op job just to signal semaphores. It waits on the completion of all last jobs submitted to any GPU queue and then signal semaphores. Note: We changed this approach to correctly deal with ordering changes caused by event threads at some point. Whenever we have an event job in the command buffer, we cannot use the last job in the last command buffer assumption. We have to wait all event threads complete to signal After submitting all command buffers, we emit a no-op job to wait on all last jobs by queue completion and signal fence. Note: at some point, we changed this approach to correct deal with ordering changes caused by event threads, as mentioned before.

Final considerations With many changes and many rounds of reviews, the patchset was merged. After more validations and code review, we polished and fixed the implementation together with external contributions: Also, multisync capabilities enabled us to add new features to V3DV and switch the driver to the common synchronization and submission framework:
  • v3dv: expose support for semaphore imports
    This was waiting for multisync support in the v3d kernel, which is already available. Exposing this feature however enabled a few more CTS tests that exposed pre-existing bugs in the user-space driver so we fix those here before exposing the feature.
  • v3dv: Switch to the common submit framework
    This should give you emulated timeline semaphores for free and kernel-assisted sharable timeline semaphores for cheap once you have the kernel interface wired in.
We used a set of games to ensure no performance regression in the new implementation. For this, we used GFXReconstruct to capture Vulkan API calls when playing those games. Then, we compared results with and without multisync caps in the kernelspace and also enabling multisync on v3dv. We didn t observe any compromise in performance, but improvements when replaying scenes of vkQuake game.

Melissa Wen: Multiple syncobjs support for V3D(V) (Part 1)

As you may already know, we at Igalia have been working on several improvements to the 3D rendering drivers of Broadcom Videocore GPU, found in Raspberry Pi 4 devices. One of our recent works focused on improving V3D(V) drivers adherence to Vulkan submission and synchronization framework. We had to cross various layers from the Linux Graphics stack to add support for multiple syncobjs to V3D(V), from the Linux/DRM kernel to the Vulkan driver. We have delivered bug fixes, a generic gate to extend job submission interfaces, and a more direct sync mapping of the Vulkan framework. These changes did not impact the performance of the tested games and brought greater precision to the synchronization mechanisms. Ultimately, support for multiple syncobjs opened the door to new features and other improvements to the V3DV submission framework.

DRM Syncobjs But, first, what are DRM sync objs?
* DRM synchronization objects (syncobj, see struct &drm_syncobj) provide a
* container for a synchronization primitive which can be used by userspace
* to explicitly synchronize GPU commands, can be shared between userspace
* processes, and can be shared between different DRM drivers.
* Their primary use-case is to implement Vulkan fences and semaphores.
[...]
* At it's core, a syncobj is simply a wrapper around a pointer to a struct
* &dma_fence which may be NULL.
And Jason Ekstrand well-summarized dma_fence features in a talk at the Linux Plumbers Conference 2021:
A struct that represents a (potentially future) event:
  • Has a boolean signaled state
  • Has a bunch of useful utility helpers/concepts, such as refcount, callback wait mechanisms, etc.
Provides two guarantees:
  • One-shot: once signaled, it will be signaled forever
  • Finite-time: once exposed, is guaranteed signal in a reasonable amount of time

What does multiple semaphores support mean for Raspberry Pi 4 GPU drivers? For our main purpose, the multiple syncobjs support means that V3DV can submit jobs with more than one wait and signal semaphore. In the kernel space, wait semaphores become explicit job dependencies to wait on before executing the job. Signal semaphores (or post dependencies), in turn, work as fences to be signaled when the job completes its execution, unlocking following jobs that depend on its completion. The multisync support development comprised of many decision-making points and steps summarized as follow:
  • added to the v3d kernel-driver capabilities to handle multiple syncobj;
  • exposed multisync capabilities to the userspace through a generic extension; and
  • reworked synchronization mechanisms of the V3DV driver to benefit from this feature
  • enabled simulator to work with multiple semaphores
  • tested on Vulkan games to verify the correctness and possible performance enhancements.
We decided to refactor parts of the V3D(V) submission design in kernel-space and userspace during this development. We improved job scheduling on V3D-kernel and the V3DV job submission design. We also delivered more accurate synchronizing mechanisms and further updates in the Broadcom Vulkan driver running on Raspberry Pi 4. Therefore, we summarize here changes in the kernel space, describing the previous state of the driver, taking decisions, side improvements, and fixes.

From single to multiple binary in/out syncobjs: Initially, V3D was very limited in the numbers of syncobjs per job submission. V3D job interfaces (CL, CSD, and TFU) only supported one syncobj (in_sync) to be added as an execution dependency and one syncobj (out_sync) to be signaled when a submission completes. Except for CL submission, which accepts two in_syncs: one for binner and another for render job, it didn t change the limited options. Meanwhile in the userspace, the V3DV driver followed alternative paths to meet Vulkan s synchronization and submission framework. It needed to handle multiple wait and signal semaphores, but the V3D kernel-driver interface only accepts one in_sync and one out_sync. In short, V3DV had to fit multiple semaphores into one when submitting every GPU job.

Generic ioctl extension The first decision was how to extend the V3D interface to accept multiple in and out syncobjs. We could extend each ioctl with two entries of syncobj arrays and two entries for their counters. We could create new ioctls with multiple in/out syncobj. But after examining other drivers solutions to extend their submission s interface, we decided to extend V3D ioctls (v3d_cl_submit_ioctl, v3d_csd_submit_ioctl, v3d_tfu_submit_ioctl) by a generic ioctl extension. I found a curious commit message when I was examining how other developers handled the issue in the past:
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Mar 22 09:23:22 2019 +0000
    drm/i915: Introduce the i915_user_extension_method
    
    An idea for extending uABI inspired by Vulkan's extension chains.
    Instead of expanding the data struct for each ioctl every time we need
    to add a new feature, define an extension chain instead. As we add
    optional interfaces to control the ioctl, we define a new extension
    struct that can be linked into the ioctl data only when required by the
    user. The key advantage being able to ignore large control structs for
    optional interfaces/extensions, while being able to process them in a
    consistent manner.
    
    In comparison to other extensible ioctls, the key difference is the
    use of a linked chain of extension structs vs an array of tagged
    pointers. For example,
    
    struct drm_amdgpu_cs_chunk  
    	__u32		chunk_id;
        __u32		length_dw;
        __u64		chunk_data;
     ;
[...]
So, inspired by amdgpu_cs_chunk and i915_user_extension, we opted to extend the V3D interface through a generic interface. After applying some suggestions from Iago Toral (Igalia) and Daniel Vetter, we reached the following struct:
struct drm_v3d_extension  
	__u64 next;
	__u32 id;
#define DRM_V3D_EXT_ID_MULTI_SYNC		0x01
	__u32 flags; /* mbz */
 ;
This generic extension has an id to identify the feature/extension we are adding to an ioctl (that maps the related struct type), a pointer to the next extension, and flags (if needed). Whenever we need to extend the V3D interface again for another specific feature, we subclass this generic extension into the specific one instead of extending ioctls indefinitely.

Multisync extension For the multiple syncobjs extension, we define a multi_sync extension struct that subclasses the generic extension struct. It has arrays of in and out syncobjs, the respective number of elements in each of them, and a wait_stage value used in CL submissions to determine which job needs to wait for syncobjs before running.
struct drm_v3d_multi_sync  
	struct drm_v3d_extension base;
	/* Array of wait and signal semaphores */
	__u64 in_syncs;
	__u64 out_syncs;
	/* Number of entries */
	__u32 in_sync_count;
	__u32 out_sync_count;
	/* set the stage (v3d_queue) to sync */
	__u32 wait_stage;
	__u32 pad; /* mbz */
 ;
And if a multisync extension is defined, the V3D driver ignores the previous interface of single in/out syncobjs. Once we had the interface to support multiple in/out syncobjs, v3d kernel-driver needed to handle it. As V3D uses the DRM scheduler for job executions, changing from single syncobj to multiples is quite straightforward. V3D copies from userspace the in syncobjs and uses drm_syncobj_find_fence()+ drm_sched_job_add_dependency() to add all in_syncs (wait semaphores) as job dependencies, i.e. syncobjs to be checked by the scheduler before running the job. On CL submissions, we have the bin and render jobs, so V3D follows the value of wait_stage to determine which job depends on those in_syncs to start its execution. When V3D defines the last job in a submission, it replaces dma_fence of out_syncs with the done_fence from this last job. It uses drm_syncobj_find() + drm_syncobj_replace_fence() to do that. Therefore, when a job completes its execution and signals done_fence, all out_syncs are signaled too.

Other improvements to v3d kernel driver This work also made possible some improvements in the original implementation. Following Iago s suggestions, we refactored the job s initialization code to allocate memory and initialize a job in one go. With this, we started to clean up resources more cohesively, clearly distinguishing cleanups in case of failure from job completion. We also fixed the resource cleanup when a job is aborted before the DRM scheduler arms it - at that point, drm_sched_job_arm() had recently been introduced to job initialization. Finally, we prepared the semaphore interface to implement timeline syncobjs in the future.

Going Up The patchset that adds multiple syncobjs support and improvements to V3D is available here and comprises four patches:
  • drm/v3d: decouple adding job dependencies steps from job init
  • drm/v3d: alloc and init job in one shot
  • drm/v3d: add generic ioctl extension
  • drm/v3d: add multiple syncobjs support
After extending the V3D kernel interface to accept multiple syncobjs, we worked on V3DV to benefit from V3D multisync capabilities. In the next post, I will describe a little of this work.

9 May 2022

Robert McQueen: Evolving a strategy for 2022 and beyond

As a board, we have been working on several initiatives to make the Foundation a better asset for the GNOME Project. We re working on a number of threads in parallel, so I wanted to explain the big picture a bit more to try and connect together things like the new ED search and the bylaw changes. We re all here to see free and open source software succeed and thrive, so that people can be be truly empowered with agency over their technology, rather than being passive consumers. We want to bring GNOME to as many people as possible so that they have computing devices that they can inspect, trust, share and learn from. In previous years we ve tried to boost the relevance of GNOME (or technologies such as GTK) or solicit donations from businesses and individuals with existing engagement in FOSS ideology and technology. The problem with this approach is that we re mostly addressing people and organisations who are already supporting or contributing FOSS in some way. To truly scale our impact, we need to look to the outside world, build better awareness of GNOME outside of our current user base, and find opportunities to secure funding to invest back into the GNOME project. The Foundation supports the GNOME project with infrastructure, arranging conferences, sponsoring hackfests and travel, design work, legal support, managing sponsorships, advisory board, being the fiscal sponsor of GNOME, GTK, Flathub and we will keep doing all of these things. What we re talking about here are additional ways for the Foundation to support the GNOME project we want to go beyond these activities, and invest into GNOME to grow its adoption amongst people who need it. This has a cost, and that means in parallel with these initiatives, we need to find partners to fund this work. Neil has previously talked about themes such as education, advocacy, privacy, but we ve not previously translated these into clear specific initiatives that we would establish in addition to the Foundation s existing work. This is all a work in progress and we welcome any feedback from the community about refining these ideas, but here are the current strategic initiatives the board is working on. We ve been thinking about growing our community by encouraging and retaining diverse contributors, and addressing evolving computing needs which aren t currently well served on the desktop. Initiative 1. Welcoming newcomers. The community is already spending a lot of time welcoming newcomers and teaching them the best practices. Those activities are as time consuming as they are important, but currently a handful of individuals are running initiatives such as GSoC, Outreachy and outreach to Universities. These activities help bring diverse individuals and perspectives into the community, and helps them develop skills and experience of collaborating to create Open Source projects. We want to make those efforts more sustainable by finding sponsors for these activities. With funding, we can hire people to dedicate their time to operating these programs, including paid mentors and creating materials to support newcomers in future, such as developer documentation, examples and tutorials. This is the initiative that needs to be refined the most before we can turn it into something real. Initiative 2: Diverse and sustainable Linux app ecosystem. I spoke at the Linux App Summit about the work that GNOME and Endless has been supporting in Flathub, but this is an example of something which has a great overlap between commercial, technical and mission-based advantages. The key goal here is to improve the financial sustainability of participating in our community, which in turn has an impact on the diversity of who we can expect to afford to enter and remain in our community. We believe the existence of this is critically important for individual developers and contributors to unlock earning potential from our ecosystem, through donations or app sales. In turn, a healthy app ecosystem also improves the usefulness of the Linux desktop as a whole for potential users. We believe that we can build a case for commercial vendors in the space to join an advisory board alongside with GNOME, KDE, etc to input into the governance and contribute to the costs of growing Flathub. Initiative 3: Local-first applications for the GNOME desktop. This is what Thib has been starting to discuss on Discourse, in this thread. There are many different threats to free access to computing and information in today s world. The GNOME desktop and apps need to give users convenient and reliable access to technology which works similarly to the tools they already use everyday, but keeps them and their data safe from surveillance, censorship, filtering or just being completely cut off from the Internet. We believe that we can seek both philanthropic and grant funding for this work. It will make GNOME a more appealing and comprehensive offering for the many people who want to protect their privacy. The idea is that these initiatives all sit on the boundary between the GNOME community and the outside world. If the Foundation can grow and deliver these kinds of projects, we are reaching to new people, new contributors and new funding. These contributions and investments back into GNOME represent a true win-win for the newcomers and our existing community. (Originally posted to GNOME Discourse, please feel free to join the discussion there.)

Russ Allbery: rra-c-util 10.2

rra-c-util is my collection of utility functions, mostly but not entirely for C, that I use with my various software releases. There are two major changes in this release. The first is Autoconf support for PCRE2, the new version of the Perl-Compatible Regular Expression library (PCRE1 is now deprecated), which was the motivation for a new release. The second is a huge update to the Perl formatting rules due to lots of work by Julien LIE for INN. This release also tags deallocation functions, similar to the change mentioned for C TAP Harness 4.8, for all the utility libraries provided by rra-c-util, and fixes an issue with the systemd support. You can get the latest version from the rra-c-util distribution page.

29 April 2022

Russ Allbery: Review: Interesting Times

Review: Interesting Times, by Terry Pratchett
Series: Discworld #17
Publisher: Harper
Copyright: 1994
Printing: February 2014
ISBN: 0-06-227629-8
Format: Mass market
Pages: 399
Interesting Times is the seventeenth Discworld novel and certainly not the place to start. At the least, you will probably want to read The Colour of Magic and The Light Fantastic before this book, since it's a sequel to those (although Rincewind has had some intervening adventures). Lord Vetinari has received a message from the Counterweight Continent, the first in ten years, cryptically demanding the Great Wizzard be sent immediately. The Agatean Empire is one of the most powerful states on the Disc. Thankfully for everyone else, it normally suits its rulers to believe that the lands outside their walls are inhabited only by ghosts. No one is inclined to try to change their minds or otherwise draw their attention. Accordingly, the Great Wizard must be sent, a task that Vetinari efficiently delegates to the Archchancellor. There is only the small matter of determining who the Great Wizzard is, and why it was spelled with two z's. Discworld readers with a better memory than I will recall Rincewind's hat. Why the Counterweight Continent would demanding a wizard notorious for his near-total inability to perform magic is a puzzle for other people. Rincewind is promptly located by a magical computer, and nearly as promptly transported across the Disc, swapping him for an unnecessarily exciting object of roughly equivalent mass and hurling him into an unexpected rescue of Cohen the Barbarian. Rincewind predictably reacts by running away, although not fast or far enough to keep him from being entangled in a glorious popular uprising. Or, well, something that has aspirations of being glorious, and popular, and an uprising. I hate to say this, because Pratchett is an ethically thoughtful writer to whom I am willing to give the benefit of many doubts, but this book was kind of racist. The Agatean Empire is modeled after China, and the Rincewind books tend to be the broadest and most obvious parodies, so that was already a recipe for some trouble. Some of the social parody is not too objectionable, albeit not my thing. I find ethnic stereotypes and making fun of funny-sounding names in other languages (like a city named Hunghung) to be in poor taste, but Pratchett makes fun of everyone's names and cultures rather equally. (Also, I admit that some of the water buffalo jokes, despite the stereotypes, were pretty good.) If it had stopped there, it would have prompted some eye-rolling but not much comment. Unfortunately, a significant portion of the plot depends on the idea that the population of the Agatean Empire has been so brainwashed into obedience that they have a hard time even imagining resistance, and even their revolutionaries are so polite that the best they can manage for slogans are things like "Timely Demise to All Enemies!" What they need are a bunch of outsiders, such as Rincewind or Cohen and his gang. More details would be spoilers, but there are several deliberate uses of Ankh-Morpork as a revolutionary inspiration and a great deal of narrative hand-wringing over how awful it is to so completely convince people they are slaves that you don't need chains. There is a depressingly tedious tendency of western writers, even otherwise thoughtful and well-meaning ones like Pratchett, to adopt a simplistic ranking of political systems on a crude measure of freedom. That analysis immediately encounters the problem that lots of people who live within systems that rate poorly on this one-dimensional scale seem inadequately upset about circumstances that are "obviously" horrific oppression. This should raise questions about the validity of the assumptions, but those assumptions are so unquestionable that the writer instead decides the people who are insufficiently upset about their lack of freedom must be defective. The more racist writers attribute that defectiveness to racial characteristics. The less racist writers, like Pratchett, attribute that defectiveness to brainwashing and systemic evil, which is not quite as bad as overt racism but still rests on a foundation of smug cultural superiority. Krister Stendahl, a bishop of the Church of Sweden, coined three famous rules for understanding other religions:
  1. When you are trying to understand another religion, you should ask the adherents of that religion and not its enemies.
  2. Don't compare your best to their worst.
  3. Leave room for "holy envy."
This is excellent advice that should also be applied to politics. Most systems exist for some reason. The differences from your preferred system are easy to see, particularly those that strike you as horrible. But often there are countervailing advantages that are less obvious, and those are more psychologically difficult to understand and objectively analyze. You might find they have something that you wish your system had, which causes discomfort if you're convinced you have the best political system in the world, or are making yourself feel better about the abuses of your local politics by assuring yourself that at least you're better than those people. I was particularly irritated to see this sort of simplistic stereotyping in Discworld given that Ankh-Morpork, the setting of most of the Discworld novels, is an authoritarian dictatorship. Vetinari quite capably maintains his hold on power, and yet this is not taken as a sign that the city's inhabitants have been brainwashed into considering themselves slaves. Instead, he's shown as adept at maintaining the stability of a precarious system with a lot of competing forces and a high potential for destructive chaos. Vetinari is an awful person, but he may be better than anyone who would replace him. Hmm. This sort of complexity is permitted in the "local" city, but as soon as we end up in an analog of China, the rulers are evil, the system lacks any justification, and the peasants only don't revolt because they've been trained to believe they can't. Gah. I was muttering about this all the way through Interesting Times, which is a shame because, outside of the ham-handed political plot, it has some great Pratchett moments. Rincewind's approach to any and all danger is a running (sorry) gag that keeps working, and Cohen and his gang of absurdly competent decrepit barbarians are both funnier here than they have been in any previous book and the rare highly-positive portrayal of old people in fantasy adventures who are not wizards or crones. Pretty Butterfly is a great character who deserved to be in a better plot. And I loved the trouble that Rincewind had with the Agatean tonal language, which is an excuse for Pratchett to write dialog full of frustrated non-sequiturs when Rincewind mispronounces a word. I do have to grumble about the Luggage, though. From a world-building perspective its subplot makes sense, but the Luggage was always the best character in the Rincewind stories, and the way it lost all of its specialness here was oddly sad and depressing. Pratchett also failed to convince me of the drastic retcon of The Colour of Magic and The Light Fantastic that he does here (and which I can't talk about in detail due to spoilers), in part because it's entangled in the orientalism of the plot. I'm not sure Pratchett could write a bad book, and I still enjoyed reading Interesting Times, but I don't think he gave the politics his normal care, attention, and thoughtful humanism. I hope later books in this part of the Disc add more nuance, and are less confident and judgmental. I can't really recommend this one, even though it has some merits. Also, just for the record, "may you live in interesting times" is not a Chinese curse. It's an English saying that likely was attributed to China to make it sound exotic, which is the sort of landmine that good-natured parody of other people's cultures needs to be wary of. Followed in publication order by Maskerade, and in Rincewind's personal timeline by The Last Continent. Rating: 6 out of 10

28 April 2022

Bits from Debian: DebConf22 bursary applications and call for papers are closing in less than 72 hours!

If you intend to apply for a DebConf22 bursary and/or submit an event proposal and have not yet done so, please proceed as soon as possible! Bursary applications for DebConf22 will be accepted until May 1st at 23:59 UTC. Applications submitted after this deadline will not be considered. You can apply for a bursary when you register for the conference. Remember that giving a talk or organising an event is considered towards your bursary; if you have a submission to make, submit it even if it is only sketched-out. You will be able to detail it later. DebCamp plans can be entered in the usual Sprints page at the Debian wiki. Please make sure to double-check your accommodation choices (dates and venue). Details about accommodation arrangements can be found on the accommodation page. Event proposals will be accepted until May 1st at 23:59 UTC too. Events are not limited to traditional presentations or informal sessions (BoFs): we welcome submissions of tutorials, performances, art installations, debates, or any other format of event that you think would be of interest to the Debian community. Regular sessions may either be 20 or 45 minutes long (including time for questions), other kinds of sessions (workshops, demos, lightning talks, and so on) could have different durations. Please choose the most suitable duration for your event and explain any special requests. You can submit it here. The the 23rd edition of DebConf will take place from July 17th to 24th, 2022 at the Innovation and Training Park (ITP) in Prizren, Kosovo, and will be preceded by DebCamp, from July 10th to 16th. See you in Prizren! DebConf22 banner open registration

22 April 2022

Ritesh Raj Sarraf: Systemd Service Hang

Finally, TIL, what can all be the reason for systemd services to hang indefinitely. The internet is flooded with numerous reports on this topic but no clear answers. So no more uselessly marked workarounds like: systemctl daemon-reload and systemctl-daemon-reexec for this scenario. The scene would be something along the lines of:
rrs         6467  0.0  0.0  23088 15852 pts/1    Ss   12:53   0:00          \_ /bin/bash
rrs        11512  0.0  0.0  14876  4608 pts/1    S+   13:18   0:00              \_ systemctl restart snapper-timeline.timer
rrs        11513  0.0  0.0  14984  3076 pts/1    S+   13:18   0:00                  \_ /bin/systemd-tty-ask-password-agent --watch
rrs        11514  0.0  0.0 234756  6752 pts/1    Sl+  13:18   0:00                  \_ /usr/bin/pkttyagent --notify-fd 5 --fallback
The snapper-timeline service is important to me and it not running for months is a complete failure. Disappointingly, commands like systemctl --failed do not report of this oddity. The overall system status is reported to be fine, which is completely incorrect. Thankfully, a kind soul s comment gave the hint. The problem is that you could be having certain services in Activating status, which thus blocks all other services; quietly. So much for the unnecessary fun. Looking further, in my case, it was:
rrs@priyasi:~$ systemctl list-jobs 
JOB  UNIT                           TYPE  STATE  
81   timers.target                  start waiting
85   man-db.timer                   start waiting
88   fstrim.timer                   start waiting
3832 snapper-timeline.service       start waiting
83   snapper-timeline.timer         start waiting
39   systemd-time-wait-sync.service start running
87   logrotate.timer                start waiting
84   debspawn-clear-caches.timer    start waiting
89   plocate-updatedb.timer         start waiting
91   dpkg-db-backup.timer           start waiting
93   e2scrub_all.timer              start waiting
40   time-sync.target               start waiting
86   apt-listbugs.timer             start waiting

13 jobs listed.
13:12                      
That was it. I knew the systemd-timesyncd service, in the past, had given me enough headaches. And so was it this time, just quietly doing it all again.
rrs@priyasi:~$ systemctl status systemd-time-wait-sync.service
  systemd-time-wait-sync.service - Wait Until Kernel Time Synchronized
     Loaded: loaded (/lib/systemd/system/systemd-time-wait-sync.service; enabled; vendor preset>
     Active: activating (start) since Fri 2022-04-22 13:14:25 IST; 1min 38s ago
       Docs: man:systemd-time-wait-sync.service(8)
   Main PID: 11090 (systemd-time-wa)
      Tasks: 1 (limit: 37051)
     Memory: 836.0K
        CPU: 7ms
     CGroup: /system.slice/systemd-time-wait-sync.service
              11090 /lib/systemd/systemd-time-wait-sync

Apr 22 13:14:25 priyasi systemd[1]: Starting Wait Until Kernel Time Synchronized...
Apr 22 13:14:25 priyasi systemd-time-wait-sync[11090]: adjtime state 5 status 40 time Fri 2022->
13:16                   => 3  
Dear LazyWeb, anybody knows of why the systemd-time-wait-sync service would hang indefinitely? I ve had identical setups on many machines, in the same network, where others don t exhibit this problem.
rrs@priyasi:~$ systemctl cat systemd-time-wait-sync.service

...snipped...

[Service]
Type=oneshot
ExecStart=/lib/systemd/systemd-time-wait-sync
TimeoutStartSec=infinity
RemainAfterExit=yes

[Install]
WantedBy=sysinit.target
The TimeoutStartSec=infinity is definitely an attribute that shouldn t be shipped in any system services. There are use cases for it but that should be left for local admins to explicitly decide. Hanging for infinity is not a desired behavior for a system service. In figuring all this out, today I learnt the handy systemctl list-jobs command, which will give the list of active running/blocked/waiting jobs.

20 April 2022

Ritesh Raj Sarraf: Btrfs Subvol Fix

There surely is need for better tooling on the BTRFS File System side. While migrating my setup from one machine to another, this is one issue I came to be aware of, only today, when my backup tool (btrbk) complained about it. Following the pointers, I see the below snippet in btrfs-subvolume manual page.
       A snapshot that was created by send/receive will be read-only, with different last change generation, read-only and with set received_uuid which identifies the subvolume on the
       filesystem that produced the stream. The usecase relies on matching data on both sides. Changing the subvolume to read-write after it has been received requires to reset the
       received_uuid. As this is a notable change and could potentially break the incremental send use case, performing it by btrfs property set requires force if that is really desired by
       user.

           Note
           The safety checks have been implemented in 5.14.2, any subvolumes previously received (with a valid received_uuid) and read-write status may exist and could still lead to
           problems with send/receive. You can use btrfs subvolume show to identify them. Flipping the flags to read-only and back to read-write will reset the received_uuid manually. There
           may exist a convenience tool in the future.
Fixing the Received UUID: flag meant running the below:
rrs@priyasi:.../spool$ sudo btrfs sub show /
WARNING: the subvolume is read-write and has received_uuid set,
         don't use it for incremental send. Please see section
         'SUBVOLUME FLAGS' in manual page btrfs-subvolume for
         further information.
ROOTVOL
        Name:                   ROOTVOL
        UUID:                   122b0de1-e6f2-6845-aba0-6bf766c16526
        Parent UUID:            -
        Received UUID:          34772967-c709-5146-bf20-898f7dbc2c1f
        Creation time:          2021-12-02 19:59:29 +0530
        Subvolume ID:           256
        Generation:             138473
        Gen at creation:        7
        Parent ID:              5
        Top level ID:           5
        Flags:                  -
        Send transid:           35245
        Send time:              2021-12-02 19:59:29 +0530
        Receive transid:        34
        Receive time:           2021-12-02 20:13:11 +0530
        Snapshot(s):
                                ROOTVOL/.snapshots/1/snapshot
                                ROOTVOL/.snapshots/2/snapshot
22:40                      


rrs@priyasi:.../spool$ sudo btrfs property set / ro true
WARNING: read-write subvolume with received_uuid, this is bad
22:40                      



rrs@priyasi:.../spool$ sudo btrfs property set -f / ro false
22:40                      



rrs@priyasi:.../spool$ sudo btrfs sub show /
ROOTVOL
        Name:                   ROOTVOL
        UUID:                   122b0de1-e6f2-6845-aba0-6bf766c16526
        Parent UUID:            -
        Received UUID:          -
        Creation time:          2021-12-02 19:59:29 +0530
        Subvolume ID:           256
        Generation:             138473
        Gen at creation:        7
        Parent ID:              5
        Top level ID:           5
        Flags:                  -
        Send transid:           0
        Send time:              2021-12-02 19:59:29 +0530
        Receive transid:        138480
        Receive time:           2022-04-20 22:40:43 +0530
        Snapshot(s):
                                ROOTVOL/.snapshots/1/snapshot
                                ROOTVOL/.snapshots/2/snapshot
22:40                      
Hoping there won t be surprises in the coming months.

17 April 2022

Matthew Garrett: The Freedom Phone is not great at privacy

The Freedom Phone advertises itself as a "Free speech and privacy first focused phone". As documented on the features page, it runs ClearOS, an Android-based OS produced by Clear United (or maybe one of the bewildering array of associated companies, we'll come back to that later). It's advertised as including Signal, but what's shipped is not the version available from the Signal website or any official app store - instead it's this fork called "ClearSignal".

The first thing to note about ClearSignal is that the privacy policy link from that page 404s, which is not a great start. The second thing is that it has a version number of 5.8.14, which is strange because upstream went from 5.8.10 to 5.9.0. The third is that, despite Signal being GPL 3, there's no source code available. So, I grabbed jadx and started looking for differences between ClearSignal and the upstream 5.8.10 release. The results were, uh, surprising.

First up is that they seem to have integrated ACRA, a crash reporting framework. This feels a little odd - in the absence of a privacy policy, it's unclear what information this gathers or how it'll be stored. Having a piece of privacy software automatically uploading information about what you were doing in the event of a crash with no notification other than a toast that appears saying "Crash Report" feels a little dubious.

Next is that Signal (for fairly obvious reasons) warns you if your version is out of date and eventually refuses to work unless you upgrade. ClearSignal has dealt with this problem by, uh, simply removing that code. The MacOS version of the desktop app they provide for download seems to be derived from a release from last September, which for an Electron-based app feels like a pretty terrible idea. Weirdly, for Windows they link to an official binary release from February 2021, and for Linux they tell you how to use the upstream repo properly. I have no idea what's going on here.

They've also added support for network backups of your Signal data. This involves the backups being pushed to an S3 bucket using credentials that are statically available in the app. It's ok, though, each upload has some sort of nominally unique identifier associated with it, so it's not trivial to just download other people's backups. But, uh, where does this identifier come from? It turns out that Clear Center, another of the Clear family of companies, employs a bunch of people to work on a ClearID[1], some sort of decentralised something or other that seems to be based on KERI. There's an overview slide deck here which didn't really answer any of my questions and as far as I can tell this is entirely lacking any sort of peer review, but hey it's only the one thing that stops anyone on the internet being able to grab your Signal backups so how important can it be.

The final thing, though? They've extended Signal's invitation support to encourage users to get others to sign up for Clear United. There's an exposed API endpoint called "get_user_email_by_mobile_number" which does exactly what you'd expect - if you give it a registered phone number, it gives you back the associated email address. This requires no authentication. But it gets better! The API to generate a referral link to send to others sends the name and phone number of everyone in your phone's contact list. There does not appear to be any indication that this is going to happen.

So, from a privacy perspective, going to go with things being some distance from ideal. But what's going on with all these Clear companies anyway? They all seem to be related to Michael Proper, who founded the Clear Foundation in 2009. They are, perhaps unsurprisingly, heavily invested in blockchain stuff, while Clear United also appears to be some sort of multi-level marketing scheme which has a membership agreement that includes the somewhat astonishing claim that:

Specifically, the initial focus of the Association will provide members with supplements and technologies for:

9a. Frequency Evaluation, Scans, Reports;

9b. Remote Frequency Health Tuning through Quantum Entanglement;

9c. General and Customized Frequency Optimizations;


- there's more discussion of this and other weirdness here. Clear Center, meanwhile, has a Chief Physics Officer? I have a lot of questions.

Anyway. We have a company that seems to be combining blockchain and MLM, has some opinions about Quantum Entanglement, bases the security of its platform on a set of novel cryptographic primitives that seem to have had no external review, has implemented an API that just hands out personal information without any authentication and an app that appears more than happy to upload all your contact details without telling you first, has failed to update this app to keep up with upstream security updates, and is violating the upstream license. If this is their idea of "privacy first", I really hate to think what their code looks like when privacy comes further down the list.

[1] Pointed out to me here

comment count unavailable comments

16 April 2022

Petter Reinholdtsen: Playing and encoding AV1 in Debian Bullseye

Inspired by the recent news of AV1 hardware encoding support from Intel, I decided to look into the state of AV1 on Linux today. AV1 is a free and open standard as defined by Digistan without any royalty payment requirement, unlike its much used competitor encoding H.264. While looking, I came across an 5 year old question on askubuntu.com which in turn inspired me to check out how things are in Debian Stable regarding AV1. The test file listed in the question (askubuntu_test_aom.mp4) did not exist any more, so I tracked down a different set of test files on av1.webmfiles.org to test them with the various video tools I had installed on my machine. I was happy to discover that AV1 decoding and playback worked with almost every tool I tested:
mediainfo ok
dragonplayer ok
ffmpeg / ffplay ok
gnome-mplayer fail
mplayer ok
mpv ok
parole ok
vlc ok
firefox ok
chromium ok
AV1 encoding is available in Debian Stable from the aom-tools version 1.0.0.errata1-3 package, using the aomenc tool. The encoding using the package in Debian Stable is quite slow, with the frame rate for my 10 second test video at around 0.25 fps. My 10 second video test took 16 minutes and 11 seconds on my test machine. I tested by first running ffmpeg and then aomenc using the recipe provided by the askubuntu recipe above. I had to remove the '--row-mt=1' option, as it was not supported in my 1.0.0 version. The encoding only used a single thread, according to top.
ffmpeg -i some-old-video.ogv -t 10 -pix_fmt yuv420p video.y4m
aomenc --fps=24/1 -u 0 --codec=av1 --target-bitrate=1000 \
  --lag-in-frames=25 --auto-alt-ref=1 -t 24 --cpu-used=8 \
  --tile-columns=2 --tile-rows=2 -o output.webm video.y4m
As version 1.0.0 currently have several unsolved security issues in Debian Stable, and to see if the recent backport provided in Debian is any quicker, I ran apt -t bullseye-backports install aom-tools to fetch the backported version and re-encoded the video using the latest version. This time the '--row-mt=1' option worked, and the encoding was done in 46 seconds with a frame rate of around 5.22 fps. This time it seem to be using all my four cores to encode. Encoding speed is still too low for streaming and real time, which would require frame rates above 25 fps, but might be good enough for offline encoding. I am very happy to see AV1 playback working so well with the default tools in Debian Stable. I hope the encoding situation improve too, allowing even a slow old computer like my 10 year old laptop to be used for encoding. As usual, if you use Bitcoin and want to show your support of my activities, please send Bitcoin donations to my address 15oWEoG9dUPovwmUL9KWAnYRtNJEkP1u1b.

5 April 2022

Kees Cook: security things in Linux v5.10

Previously: v5.9 Linux v5.10 was released in December, 2020. Here s my summary of various security things that I found interesting: AMD SEV-ES
While guest VM memory encryption with AMD SEV has been supported for a while, Joerg Roedel, Thomas Lendacky, and others added register state encryption (SEV-ES). This means it s even harder for a VM host to reconstruct a guest VM s state. x86 static calls
Josh Poimboeuf and Peter Zijlstra implemented static calls for x86, which operates very similarly to the static branch infrastructure in the kernel. With static branches, an if/else choice can be hard-coded, instead of being run-time evaluated every time. Such branches can be updated too (the kernel just rewrites the code to switch around the branch ). All these principles apply to static calls as well, but they re for replacing indirect function calls (i.e. a call through a function pointer) with a direct call (i.e. a hard-coded call address). This eliminates the need for Spectre mitigations (e.g. RETPOLINE) for these indirect calls, and avoids a memory lookup for the pointer. For hot-path code (like the scheduler), this has a measurable performance impact. It also serves as a kind of Control Flow Integrity implementation: an indirect call got removed, and the potential destinations have been explicitly identified at compile-time. network RNG improvements
In an effort to improve the pseudo-random number generator used by the network subsystem (for things like port numbers and packet sequence numbers), Linux s home-grown pRNG has been replaced by the SipHash round function, and perturbed by (hopefully) hard-to-predict internal kernel states. This should make it very hard to brute force the internal state of the pRNG and make predictions about future random numbers just from examining network traffic. Similarly, ICMP s global rate limiter was adjusted to avoid leaking details of network state, as a start to fixing recent DNS Cache Poisoning attacks. SafeSetID handles GID
Thomas Cedeno improved the SafeSetID LSM to handle group IDs (which required teaching the kernel about which syscalls were actually performing setgid.) Like the earlier setuid policy, this lets the system owner define an explicit list of allowed group ID transitions under CAP_SETGID (instead of to just any group), providing a way to keep the power of granting this capability much more limited. (This isn t complete yet, though, since handling setgroups() is still needed.) improve kernel s internal checking of file contents
The kernel provides LSMs (like the Integrity subsystem) with details about files as they re loaded. (For example, loading modules, new kernel images for kexec, and firmware.) There wasn t very good coverage for cases where the contents were coming from things that weren t files. To deal with this, new hooks were added that allow the LSMs to introspect the contents directly, and to do partial reads. This will give the LSMs much finer grain visibility into these kinds of operations. set_fs removal continues
With the earlier work landed to free the core kernel code from set_fs(), Christoph Hellwig made it possible for set_fs() to be optional for an architecture. Subsequently, he then removed set_fs() entirely for x86, riscv, and powerpc. These architectures will now be free from the entire class of kernel address limit attacks that only needed to corrupt a single value in struct thead_info. sysfs_emit() replaces sprintf() in /sys
Joe Perches tackled one of the most common bug classes with sprintf() and snprintf() in /sys handlers by creating a new helper, sysfs_emit(). This will handle the cases where kernel code was not correctly dealing with the length results from sprintf() calls, which might lead to buffer overflows in the PAGE_SIZE buffer that /sys handlers operate on. With the helper in place, it was possible to start the refactoring of the many sprintf() callers. nosymfollow mount option
Mattias Nissler and Ross Zwisler implemented the nosymfollow mount option. This entirely disables symlink resolution for the given filesystem, similar to other mount options where noexec disallows execve(), nosuid disallows setid bits, and nodev disallows device files. Quoting the patch, it is useful as a defensive measure for systems that need to deal with untrusted file systems in privileged contexts. (i.e. for when /proc/sys/fs/protected_symlinks isn t a big enough hammer.) Chrome OS uses this option for its stateful filesystem, as symlink traversal as been a common attack-persistence vector. ARMv8.5 Memory Tagging Extension support
Vincenzo Frascino added support to arm64 for the coming Memory Tagging Extension, which will be available for ARMv8.5 and later chips. It provides 4 bits of tags (covering multiples of 16 byte spans of the address space). This is enough to deterministically eliminate all linear heap buffer overflow flaws (1 tag for free , and then rotate even values and odd values for neighboring allocations), which is probably one of the most common bugs being currently exploited. It also makes use-after-free and over/under indexing much more difficult for attackers (but still possible if the target s tag bits can be exposed). Maybe some day we can switch to 128 bit virtual memory addresses and have fully versioned allocations. But for now, 16 tag values is better than none, though we do still need to wait for anyone to actually be shipping ARMv8.5 hardware. fixes for flaws found by UBSAN
The work to make UBSAN generally usable under syzkaller continues to bear fruit, with various fixes all over the kernel for stuff like shift-out-of-bounds, divide-by-zero, and integer overflow. Seeing these kinds of patches land reinforces the the rationale of shifting the burden of these kinds of checks to the toolchain: these run-time bugs continue to pop up. flexible array conversions
The work on flexible array conversions continues. Gustavo A. R. Silva and others continued to grind on the conversions, getting the kernel ever closer to being able to enable the -Warray-bounds compiler flag and clear the path for saner bounds checking of array indexes and memcpy() usage. That s it for now! Please let me know if you think anything else needs some attention. Next up is Linux v5.11.

2022, Kees Cook. This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 License.
CC BY-SA 4.0

2 April 2022

Ian Jackson: Otter (game server) 1.0.0 released

I have just released Otter 1.0.0. Recap: what is Otter Otter is my game server for arbitrary board games. Unlike most online game systems. It does not know (nor does it need to know) the rules of the game you are playing. Instead, it lets you and your friends play with common tabletop/boardgame elements such as hands of cards, boards, and so on. So it s something like a tabletop simulator (but it does not have any 3D, or a physics engine, or anything like that). There are provided game materials and templates for Penultima, Mao, and card games in general. Otter also supports uploadable game bundles, which allows users to add support for additional games - and this can be done without programming. For more information, see the online documentation. There are a longer intro and some screenshots in my 2021 introductory blog post about Otter Releasing 1.0.0 I m calling this release 1.0.0 because I think I can now say that its quality, reliability and stability is suitable for general use. In particular, Otter now builds on Stable Rust, which makes it a lot easier to install and maintain. Switching web framework, and async Rust I switched Otter from the Rocket web framework to Actix. There are things to prefer about both systems, and I still have a soft spot for Rocket. But ultimately I needed a framework which was fully released and supported for use with Stable Rust. There are few if any Rust web frameworks that are not async. This is rather a shame. Async Rust is a considerably more awkward programming environment than ordinary non-async Rust. I don t want to digress into a litany of complaints, but suffice it to say that while I really love Rust, my views on async Rust are considerably more mixed. Future plans In the near future I plan to add a couple of features to better support some particular games: currency-like resources, and a better UI for dice-like randomness. In the longer term, Otter s, installation and account management arrangements are rather unsophisticated and un-webby. There is not currently any publicly available instance for you to try it out without installing it on a machine of your own. There s not even any provided binaries: you must built Otter yourself. I hope to be able to improve this situation but it involves dealing with cloud CI and containers and so-on, which can all be rather unpleasant. Users on chiark will find an instance of Otter there.

comment count unavailable comments

1 April 2022

Russ Allbery: Review: Princess Floralinda and the Forty-Flight Tower

Review: Princess Floralinda and the Forty-Flight Tower, by Tamsyn Muir
Publisher: Subterranean Press
Copyright: 2020
ISBN: 1-59606-992-9
Format: Kindle
Pages: 111
A witch put Princess Floralinda at the top of a forty-flight tower, but it wasn't personal. This is just what witches do, particularly with princesses with butter-coloured curls and sapphire-blue eyes. Princes would come from miles around to battle up the floors of the tower and rescue the princess. The witch even helpfully provided a golden sword, in case a prince didn't care that much about princesses. Floralinda was provided with water and milk, two loaves of bread, and an orange, all of them magically renewing, to sustain her while she waited. In retrospect, the dragon with diamond-encrusted scales on the first floor may have been a mistake. None of the princely endeavors ever saw the second floor. The diary that Floralinda found in her room indicated that she may not be the first princess to have failed to be rescued from this tower. Floralinda finally reaches the rather astonishing conclusion that she might have to venture down the tower herself, despite the goblins she was warned were on the 39th floor (not to mention all the other monsters). The result of that short adventure, after some fast thinking, a great deal of luck, and an unforeseen assist from her magical food, is a surprising number of dead goblins. Also seriously infected hand wounds, because it wouldn't be a Tamsyn Muir story without wasting illness and body horror. That probably would have been the end of Floralinda, except a storm blew a bottom-of-the-garden fairy in through the window, sufficiently injured that she and Floralinda were stuck with each other, at least temporarily. Cobweb, the fairy, is neither kind nor inclined to help Floralinda (particularly given that Floralinda is not a child whose mother is currently in hospital), but it is an amateur chemist and finds both Floralinda's tears and magical food intriguing. Cobweb's magic is also based on wishes, and after a few failed attempts, Floralinda manages to make a wish that takes hold. Whether she'll regret the results is another question. This is a fairly short novella by the same author as Gideon the Ninth, but it's in a different universe and quite different in tone. This summary doesn't capture the writing style, which is a hard-to-describe mix of fairy tale, children's story, and slightly archaic and long-winded sentence construction. This is probably easier to show with a quote:
"You are displaying a very small-minded attitude," said the fairy, who seemed genuinely grieved by this. "Consider the orange-peel, which by itself has many very nice properties. Now, if you had a more educated brain (I cannot consider myself educated; I have only attempted to better my situation) you would have immediately said, 'Why, if I had some liquor, or even very hot water, I could extract some oil from this orange-peel, which as everyone knows is antibacterial; that may well do my hands some good,' and you wouldn't be in such a stupid predicament."
On balance, I think this style worked. It occasionally annoyed me, but it has some charm. About halfway through, I was finding the story lightly entertaining, although I would have preferred a bit less grime, illness, and physical injury. Unfortunately, the rest of the story didn't work for me. The dynamic between Floralinda and Cobweb turns into a sort of D&D progression through monster fights, and while there are some creative twists to those fights, they become all of a sameness. And while I won't spoil the ending, it didn't work for me. I think I see what Muir was trying to do, and I have some intellectual appreciation for the idea, but it wasn't emotionally satisfying. I think my root problem with this story is that Muir sets up a rather interesting world, one in which witches artistically imprison princesses, and particularly bright princesses (with the help of amateur chemist fairies) can use the trappings of a magical tower in ways the witch never intended. I liked that; it has a lot of potential. But I didn't feel like that potential went anywhere satisfying. There is some relationship and characterization work, and it reached some resolution, but it didn't go as far as I wanted. And, most significantly, I found the end point the characters reached in relation to the world to be deeply unsatisfying and vaguely irritating. I wanted to like this more than I did. I think there's a story idea in here that I would have enjoyed more. Unfortunately, it's not the one that Muir wrote, and since so much of my problem is with the ending, I can't provide much guidance on whether someone else would like this story better (and why). But if the idea of taking apart a fairy-tale tower and repurposing the pieces sounds appealing, and if you get along better with Muir's illness motif than I do, you may enjoy this more than I did. Rating: 5 out of 10

28 March 2022

Dirk Eddelbuettel: RcppCNPy 0.2.11

A minor maintenance release of the RcppCNPy package arrived on CRAN three days ago, but we skipped announcing it right then. RcppCNPy provides R with read and write access to NumPy files thanks to the cnpy library by Carl Rogers along with Rcpp for the glue to R. One of the vignettes created an issue at CRAN with one of the Python modules used, so we simply switched to pre-made vignettes just as we do for a few other packages. Other small changes that had accumulated since the previous release were a new section in the reticulate vignette as well as some more documentation on types as well as some updates to continuous integration and bdages.

Changes in version 0.2.11 (2022-03-24)
  • The reticulate vignette has new section on 3d-arrays.
  • Added clarification to the manual page that the default types are 32-bit integer and 64-bit double (as we are working with R here).
  • Several updates have been made to the CI system; it now runs r-ci.
  • The README.md was updated with some new badges.
  • The vignettes are now pre-made to avoid any external dependencies.

CRANberries also provides a diffstat report for the latest release. As always, feedback is welcome and the best place to start a discussion may be the GitHub issue tickets page.

This post by Dirk Eddelbuettel originated on his Thinking inside the box blog. Please report excessive re-aggregation in third-party for-profit settings.

21 March 2022

Antoine Beaupr : 20+ years of Emacs

I enjoyed reading this article named "22 years of Emacs" recently. It's kind of fascinating, because I realised I don't exactly know for how long I've been using Emacs. It's lost in the mists of history. If I would have to venture a guess, it was back in the "early days", which in that history is mapped around 1996-1997, when I installed my very own "PC" with FreeBSD 2.2.x and painstakingly managed to make XFree86 run on it. Modelines. Those were the days... But I digress.

I am old... The only formal timestamp I can put is that my rebuilt .emacs.d git repository has its first commit in 2002. Some people reading this may be born after that time. This means I'm at least significantly older than those people, to put things gently. Clever history nerds will notice that the commit is obviously fake: Git itself did not exist until 2005. But ah-ah! I was already managing my home directory with CVS in 2001! I converted that repository into git some time in 2009, and therefore you can see all my embarrassing history, including changes from two decades ago. That includes my first known .emacs file which is just bizarre to read right now: 200 lines, most of which are "customize" stuff. Compare with the current, 1000+ lines init.el which is also still kind of a mess, but actually shares very little with the original, thankfully. All this to say that in those years (decades, really) of using Emacs, I have had a very different experience than credmp who wrote packages, sent patches, and got name dropping from other developers. My experience is just struggling to keep up with everything, in general, but also in Emacs.

... and Emacs is too fast for me It might sound odd to say, but Emacs is actually moving pretty fast right now. A lot of new packages are coming out, and I can hardly keep up.
  • I am not using org mode, but did use it for time (and task) tracking for a while (and for invoicing too, funky stuff).
  • I am not using mu4e, but maybe I'm using something better (notmuch) and yes, I am reading my mail in Emacs, which I find questionable from a security perspective. (Sandboxing untrusted inputs? Anyone?)
  • I am using magit, but only when coding, so I do end up using git on the command line quite a bit anyways.
  • I do have which-key enabled, and reading about it reminded me I wanted to turn it off because it's kind of noisy and I never remember I can actually use it for anything. Or, in other words, I don't even remember the prefix key or, when I do, there's too many possible commands after for it to be useful.
  • I haven't setup lsp-mode, let alone Eglot, which I just learned about reading the article. I thought I would be super shiny and cool by setting up LSP instead of the (dying?) elpy package, but I never got around to it. And now it seems lsp-mode is uncool and I should really do eglot instead, and that doesn't help. UPDATE: I finally got tired and switched to lsp-mode. The main reason for choosing it over eglot is that it's in Debian (and eglot is not). (Apparently, eglot has more chance of being upstreamed, "when it's done", but I guess I'll cross that bridge when I get there.) lsp-mode feels slower than elpy but I haven't done any of the performance tuning and this will improve even more with native compilation (see below). I already had lsp-mode partially setup in Emacs so I only had to do this small tweak to switch and change the prefix key (because s-l or mod is used by my window manager). I also had to pin LSP packages to bookworm here and here.
  • I am not using projectile. It's on some of my numerous todo lists somewhere, surely. I suspect it's important to getting my projects organised, but I still live halfway between the terminal and Emacs, so it's not quite clear what I would gain.
  • I had to ask what native compilation was or why it mattered the first time I heard of it. And when I saw it again in the article, I had to click through to remember.
Overall, I feel there's a lot of cool stuff in Emacs out there. But I can't quite tell what's the best of which. I can barely remember which completion mechanism I use (company, maybe?) or what makes my mini-buffer completion work the way it does. Everything is lost in piles of customize and .emacs hacks that is constantly changing. Because a lot is in third-party packages, there are often many different options and it's hard to tell which one we should be using.

... or at least fast enough And really, Emacs feels fast enough for me. When I started, I was running Emacs on a Pentium I, 166MHz, with 8MB of RAM (eventually upgraded to 32MB, whoohoo!). Back in those days, the joke was that EMACS was an acronym for "Eight Megs, Always Scratching" and now that I write this down, I realize it's actually "Eight Megs, and Constantly Swapping", which doesn't sound as nice because you could actually hear Emacs running on those old hard drives back in the days. It would make a "scratching" noise as the hard drive heads would scramble maniacally to swap pages in and out of swap to make room for the memory-hungry editor. Now Emacs is pretty far down the list of processes in top(1) regardless of how you look at it. It's using 97MB of resident memory and close to 400MB of virtual memory, which does sound like an awful lot compared to my first computer... But it's absolutely nothing compared to things like Signal-desktop, which somehow manages to map a whopping 20.5GB virtual memory. (That's twenty Gigabytes of memory for old timers or time travelers from the past, and yes, that is now a thing.) I'm not exactly sure how much resident memory it uses (because it forks multiple processes), probably somewhere around 300MB of resident memory. Firefox also uses gigabytes of that good stuff, also spread around the multiple processes, per tab. Emacs "feels" super fast. Typing latency is noticeably better in Emacs than my web browser, and even beats most terminal emulators. It gets a little worse when font-locking is enabled, unfortunately, but it's still feels much better. And all my old stuff still works in Emacs, amazingly. (Good luck with your old Netscape or ICQ configuration from 2000.) I feel like an oldie, using Emacs, but I'm really happy to see younger people using it, and learning it, and especially improving it. If anything, one direction I would like to see it go is closer to what web browsers are doing (yes, I know how bad that sounds) and get better isolation between tasks. An attack on my email client shouldn't be able to edit my Puppet code, and/or all files on my system, for example. And I know, fundamentally, that's a really hard challenge in Emacs. But if you're going to treat your editor as your operating system (or vice versa, I lost track of where we are now that there's an Emacs Window Manager, which I do not use), at least we should get that kind of security. Otherwise I'll have to find a new mail client, and that's really something I try to limit to once a decade or so.

Next.

Previous.