I still haven’t upgraded from 6.3 because the entire opensmtpd configuration language changed quite a bit. Maybe with 6.6 I can find some better tutorials out there, but I still need to build out a full test environment for it. E-mail is not something I want to fuck up :-P
Yeah I had a lot of problems with that. Took me awhile to get back to a working configuration and I made a few mistakes! As my configuration is pretty simple I probably would have been better just starting from scratch and reading the manpage without any prior assumptions than trying to adapt what I had.
Wow .. that’s .. long .. gotta skip way down to the code.
I mean .. I’d rather just have a tutorial that tells me how to convert my old config to the new config. Like here’s the ansible template for my old one:
I don’t think the old one had filter implemented yet. pki and action look mostly the same, but the accept have been replaced by action. I’ve read through the official doc before and got confused and couldn’t find a straight upgrade guide. I’ll probably write one myself whenever I get around to it.
Slowly start implementing tagging support for man(7) pages: tag alphabetic arguments of .IP, .TP, and .TQ macros.
This is the kind of secret sauce that makes OpenBSD such a pleasure to me. Try navigating the tmux man page with tags and mark points if you want to feel spoiled by the thoughtfulness of Nicholas and whoever else contributed to that document. It takes documentation that dense & complex to truly appreciate this work.
@cmb just upgraded our OpenBSD netboot installer to the 6.6 release. As a VPS provider, prgmr.com has done a lot of upgrades, patching and mitigation for the various CPU vulnerabilities discovered or announced over the past couple years. I’m particularly glad then to see the following in the release notes:
Further and improved mitigations against Spectre side-channel vulnerability in Intel CPUs built since 2012.
Mitigations for Intel’s Microarchitectural Data Sampling vulnerability, using the new CPU VERW behavior if available or by using the proper sequence from Intel’s “Deep Dive” doc in the return-to-userspace and enter-VMM-guest paths. Updated vmm(4) to pass through the MSR bits so that guests can apply the optimal mitigation.
Anyone running spaCy or PyTorch code (or the like) on the GPU have anything to share (success, failure, exploration) about trying to move their work onto OpenBSD, particularly in the context of amdgpu(4)?
My workstation is still Pop_OS! for Nvidia hardware and the Unity Editor, but I dream of a better future (and am on OpenBSD 6.5 everywhere else).
For AMD’s modern compute stack (ROCm) you need amdkfd(4), which is currently still not working on FreeBSD - I would guess it’s not ported on OpenBSD either. Without it, you can only have OpenCL via Clover, which is… not good.
IMO the usage of these “special” compute APIs is just infuriating. Why can’t everyone just use Vulkan compute? >_<
Well, for deep learning ncnn does use Vulkan, but seems like that’s only inference, not training. And some googlers are working on an OpenCL-to-Vulkan-SPIR-V compiler. So clearly I’m not alone with the “just use normal APIs” sentiment, but we’re still in the minority, while the GPGPU community at large is fine with using even proprietary APIs, not just special but standard ones…
But practically speaking, the reason my code needs to run fast is usually the same reason I need to use higher-level libraries stacked upon one proprietary API or another. I’m rapidly iterating and exploring results in a continuous loop, with 99% of the code I write being to explore something that was just uncovered by the last code I ran; code which I know I’ll throw out not merely because it is prototype code, but because the entire idea will be a dead end.
I’m willing to reorient nearly everything I do with technology to live by my principles (and mostly have), but this one is an untenable sacrifice because it’s how I do the principled work more important than my principles about technology.
The chain of dependencies to do this with such a high-level interface is stacked incredibly high (and far back in time), but I’m excited to see the continuous progress being made as good people diligently chip away at every level.
Thank you @tekknolagi. I went ahead and used the top-level domain for the story URL which redirects to the link in your comment. The submission by itself is a store selling computing-focused t-shirts, but is not itself computing-focused: I merged it in to the OpenBSD 6.6 release announcement.
I would assume that it is because it makes a relicense easier. Like if something more appropriate than a BSD 3 Clause is released. If he becomes evil, we have the right to form an new organization and fork. So we have all the guarantees of the current license plus the ability to move forward in a case where it is appropriate.
Think about the current state of Linux. Let’s say BSD or Apache were considered? It would be impossible to move because of all the people who would have to agree to the move or the rewrite of the code they own.
As far as I can tell, he just holds the copyright for the overall distribution I guess. All the code has copyright notices by a bunch of other people. It really doesn’t matter who has it, in that sense.
And.. one can upgrade from 6.5 using sysupgrade
Sooooo nice!
I still haven’t upgraded from 6.3 because the entire opensmtpd configuration language changed quite a bit. Maybe with 6.6 I can find some better tutorials out there, but I still need to build out a full test environment for it. E-mail is not something I want to fuck up :-P
Yeah I had a lot of problems with that. Took me awhile to get back to a working configuration and I made a few mistakes! As my configuration is pretty simple I probably would have been better just starting from scratch and reading the manpage without any prior assumptions than trying to adapt what I had.
You mean like this one: https://poolp.org/posts/2019-09-14/setting-up-a-mail-server-with-opensmtpd-dovecot-and-rspamd/ ?
Wow .. that’s .. long .. gotta skip way down to the code.
I mean .. I’d rather just have a tutorial that tells me how to convert my old config to the new config. Like here’s the ansible template for my old one:
https://github.com/sumdog/bee2/blob/master/ansible/roles/openbsd-email/templates/smtpd.conf.j2
I don’t think the old one had filter implemented yet. pki and action look mostly the same, but the accept have been replaced by action. I’ve read through the official doc before and got confused and couldn’t find a straight upgrade guide. I’ll probably write one myself whenever I get around to it.
You can contact the author on twitter, maybe he knows of such a resource. https://twitter.com/PoolpOrg
just upgraded from 6.5 to 6.6 using syspatch/sysupgrade. Went very smooth.
Everything now feels snappier on my laptop, especially browsing with firefox. Have to say, very polished experience overall with this upgrade.
This is the kind of secret sauce that makes OpenBSD such a pleasure to me. Try navigating the tmux man page with tags and mark points if you want to feel spoiled by the thoughtfulness of Nicholas and whoever else contributed to that document. It takes documentation that dense & complex to truly appreciate this work.
“Released XXX, 2019”
Well, at least there’s an e-mail to confirm it.
Here’s the changelog:
https://www.openbsd.org/plus66.html
@cmb just upgraded our OpenBSD netboot installer to the 6.6 release. As a VPS provider, prgmr.com has done a lot of upgrades, patching and mitigation for the various CPU vulnerabilities discovered or announced over the past couple years. I’m particularly glad then to see the following in the release notes:
Anyone running spaCy or PyTorch code (or the like) on the GPU have anything to share (success, failure, exploration) about trying to move their work onto OpenBSD, particularly in the context of amdgpu(4)?
My workstation is still Pop_OS! for Nvidia hardware and the Unity Editor, but I dream of a better future (and am on OpenBSD 6.5 everywhere else).
For AMD’s modern compute stack (ROCm) you need amdkfd(4), which is currently still not working on FreeBSD - I would guess it’s not ported on OpenBSD either. Without it, you can only have OpenCL via Clover, which is… not good.
IMO the usage of these “special” compute APIs is just infuriating. Why can’t everyone just use Vulkan compute? >_<
Well, for deep learning ncnn does use Vulkan, but seems like that’s only inference, not training. And some googlers are working on an OpenCL-to-Vulkan-SPIR-V compiler. So clearly I’m not alone with the “just use normal APIs” sentiment, but we’re still in the minority, while the GPGPU community at large is fine with using even proprietary APIs, not just special but standard ones…
Philosophically, I couldn’t agree more.
But practically speaking, the reason my code needs to run fast is usually the same reason I need to use higher-level libraries stacked upon one proprietary API or another. I’m rapidly iterating and exploring results in a continuous loop, with 99% of the code I write being to explore something that was just uncovered by the last code I ran; code which I know I’ll throw out not merely because it is prototype code, but because the entire idea will be a dead end.
I’m willing to reorient nearly everything I do with technology to live by my principles (and mostly have), but this one is an untenable sacrifice because it’s how I do the principled work more important than my principles about technology.
The chain of dependencies to do this with such a high-level interface is stacked incredibly high (and far back in time), but I’m excited to see the continuous progress being made as good people diligently chip away at every level.
Why link to reddit instead of the teespring target? https://teespring.com/stores/openbsd
Thank you @tekknolagi. I went ahead and used the top-level domain for the story URL which redirects to the link in your comment. The submission by itself is a store selling computing-focused t-shirts, but is not itself computing-focused: I merged it in to the OpenBSD 6.6 release announcement.
Honest question: Do you know why Theo holds the copyright? Is it better than having the Foundation holding it?
“Copyright 1997-2019, Theo de Raadt.”
I would assume that it is because it makes a relicense easier. Like if something more appropriate than a BSD 3 Clause is released. If he becomes evil, we have the right to form an new organization and fork. So we have all the guarantees of the current license plus the ability to move forward in a case where it is appropriate.
Think about the current state of Linux. Let’s say BSD or Apache were considered? It would be impossible to move because of all the people who would have to agree to the move or the rewrite of the code they own.
Or am I totally off base?
As far as I can tell, he just holds the copyright for the overall distribution I guess. All the code has copyright notices by a bunch of other people. It really doesn’t matter who has it, in that sense.