Arch Linux User Repository

If there are bugs related to the PKGBUILD or config, report here.

Otherwise, please report here:

https://github.com/koverstreet/bcachefs/issues

This needs to be updated to use the 5.18.x arch kernel since bcachefs master has moved to use kernel 5.18.x

@RX14 the bcachefs github upgraded to linux 5.17, I found editing the PKGBUILD to clone linux 5.17 before cloning and merging bcachefs-linux instead of 5.16 works fine.

Plenty of merge conflicts :(

@benjumanji

Cheers ~ I've fixed it now. :)

Gotta remember to update the sha512sums before pushing...

Current package sums are broken for me. This fixed it:

diff --git a/PKGBUILD b/PKGBUILD
index 4c40381..cf1ba3c 100755
--- a/PKGBUILD
+++ b/PKGBUILD
@@ -117,7 +117,7 @@ sha512sums=('SKIP'
             'SKIP'
             'SKIP'
             'SKIP'
-            '90f8c978bde10c80770431c3d8686b6f8c10e2c9645dddb9e8772312eb0cb1129bcfc6af5b86477c5616b723da25f629859092b1af178da7bd71c7b89e1de97d')
+            'd154f68d39cbe0dc25f9a721e987b7b8fbbebe5d08ed1a50c4ef64174814caf13fc8c4cc64d7d057e3d157d72ab1e88543cab397bff47bcca71bc916ec4da189')

 export KBUILD_BUILD_HOST=archlinux
 export KBUILD_BUILD_USER=$pkgbase

Recent changes allow this to build on aarch64. Please update this for those of us on ARM devices. :)

@hwaiting

Cannot reproduce.

Please report upstream.

I'm getting the following error when building with makepkg -sc

make[4]: *** No rule to make target 'libbpf_util.h', needed by '/home/gamer/Source/linux-bcachefs-git/src/linux-bcachefs/tools/bpf/resolve_btfids/libbpf/staticobjs/xsk.o'.  Stop.
make[3]: *** [Makefile:158: /home/gamer/Source/linux-bcachefs-git/src/linux-bcachefs/tools/bpf/resolve_btfids/libbpf/staticobjs/libbpf-in.o] Error 2
make[2]: *** [Makefile:44: /home/gamer/Source/linux-bcachefs-git/src/linux-bcachefs/tools/bpf/resolve_btfids//libbpf/libbpf.a] Error 2
make[1]: *** [Makefile:72: bpf/resolve_btfids] Error 2
make: *** [Makefile:1974: tools/bpf/resolve_btfids] Error 2

Looks like the latest changes have introduced a merge conflict?

@QuartzDragon: I thought yay had pulled, but it looks like it didn't. For some reason it did not detect the package update and i just forced a rebuild. Sorry for the trouble. With a pull, and a clean rebuild, it builds fine now. Thanks for the help!

I did remove the checksum check for config though, since there was a conflict. It is part of this AUR repo, so i don't see why it needs to be checked.

@clonejo

Actually... maybe I didn't forget...

Did you pull the repo for the latest updates?

What are you terminal output results?

@clonejo

Whoops ~ I'd forgot to push the URL changes after updating them.

Please try again. :)

Is git.archlinux.org gone? The domain does not resolve anymore.

The repo the linux package uses is on Github: https://github.com/archlinux/linux.git

@RealGecko

Not with this package, at least.

You'll be better off asking upstream about this.

Is there a way to install bcachefs as kernel module? DKMS maybe?

if anyone cares, i will support 5.10 lts zen at linux-bcachefs-510-zen and i ll have a kpatch script available in the next week for that.

@Hamled

Fixed. :)

Patch file got renamed again :P

s/more-uarches-for-gcc-v11-and-kernel-5.8+.patch/more-uarches-for-kernel-5.8+.patch/

Looks like the Graysky GCC patch file was renamed as of this commit: https://github.com/graysky2/kernel_gcc_patch/commit/cced8de3e769833194ee2bf6da1a7fafe39806de

So the PKGBUILD should reference "more-uarches-for-gcc-v10-and-kernel-5.8+.patch" instead of "enable_additional_cpu_optimizations_for_gcc_v10.1+_kernel_v5.8+".

@bobpaul

Hmmm. Thanks. Will do that.

~~make defconfig should create a default .config and then you can compare it with the config you want to use and it probably won't be very fun. ~~ (Edit: Actually, I don't think that's a thing. Things that aren't defined as Y, N, or M in the config are just left as "undefined". But the suggestion below is good, and it looks like linux-git and other linux packages on aur do that:)

But in this case, it looks like the config you're providing has a lot of commented lines. You could simply delete all comments and blank lines. Lines like # CONFIG_KERNEL_GZIP is not set represent a feature that's available in the code base, but not configured. Removing all empty and commented lines reduces the file to 205K. Something like grep -v '^#' big_config | grep -v '^$' > config

Also, for a single file it's somewhat preferable to avoid tar and use gzip directly to produce a config.gz, since tools like vim/emacs, zdiff, zgrep, zless, zcat, etc can directly read a gzipped text file. That's why you never see *.tar.gz archives in /var/log/

@bobpaul

So I suppose also not ideal, but another option would be to remove some kernel config options that match the default config.

Don't have the technical knowledge on how to do that.

I'm not sure which options are default config, considering that the Arch devs have many non-default kernel options.

So I suppose also not ideal, but another option would be to remove some kernel config options that match the default config.

Compressed the config file, so it fits within the 250KiB blob limit... :|

Currently cannot update due to https://bugs.archlinux.org/task/68997

@bobpaul

Please report upstream ~ I'm unable to do anything about this.

Has anyone tried building virtualbox-host-dkms?

I get this:

==> dkms install vboxhost/6.1.4_OSE -k 5.3.18-1-bcachefs-git-196109-g3d16f5d33e21
Error! Bad return status for module build on kernel: 5.3.18-1-bcachefs-git-196109-g3d16f5d33e21 (x86_64)
Consult /var/lib/dkms/vboxhost/6.1.4_OSE/build/make.log for more information.
[bobpaul@celestra ~]$ cat /var/lib/dkms/vboxhost/6.1.4_OSE/build/make.log
DKMS make.log for vboxhost-6.1.4_OSE for kernel 5.3.18-1-bcachefs-git-196109-g3d16f5d33e21 (x86_64)
Sun 22 Mar 2020 06:51:12 PM CDT
make: Entering directory '/usr/lib/modules/5.3.18-1-bcachefs-git-196109-g3d16f5d33e21/build'
  CC [M]  /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetadp/linux/VBoxNetAdp-linux.o
  CC [M]  /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetadp/VBoxNetAdp.o
  CC [M]  /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetflt/linux/VBoxNetFlt-linux.o
cc1: error: incompatible gcc/plugin versions
cc1: error: fail to initialize plugin ./scripts/gcc-plugins/structleak_plugin.so
make[2]: *** [scripts/Makefile.build:281: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetadp/linux/VBoxNetAdp-linux.o] Error 1
make[2]: *** Waiting for unfinished jobs....
  CC [M]  /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/linux/SUPDrv-linux.o
cc1: error: incompatible gcc/plugin versions
cc1: error: fail to initialize plugin ./scripts/gcc-plugins/structleak_plugin.so
make[2]: *** [scripts/Makefile.build:281: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetadp/VBoxNetAdp.o] Error 1
make[1]: *** [scripts/Makefile.build:497: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetadp] Error 2
make[1]: *** Waiting for unfinished jobs....
cc1: error: incompatible gcc/plugin versions
cc1: error: fail to initialize plugin ./scripts/gcc-plugins/structleak_plugin.so
make[2]: *** [scripts/Makefile.build:281: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetflt/linux/VBoxNetFlt-linux.o] Error 1
make[1]: *** [scripts/Makefile.build:497: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxnetflt] Error 2
cc1: error: incompatible gcc/plugin versions
cc1: error: fail to initialize plugin ./scripts/gcc-plugins/structleak_plugin.so
make[2]: *** [scripts/Makefile.build:280: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/linux/SUPDrv-linux.o] Error 1
make[2]: *** Waiting for unfinished jobs....
  CC [M]  /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/SUPDrv.o
  CC [M]  /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/SUPDrvGip.o
  CC [M]  /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/SUPDrvSem.o
cc1: error: incompatible gcc/plugin versions
cc1: error: fail to initialize plugin ./scripts/gcc-plugins/structleak_plugin.so
make[2]: *** [scripts/Makefile.build:281: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/SUPDrv.o] Error 1
cc1: error: incompatible gcc/plugin versions
cc1: error: fail to initialize plugin ./scripts/gcc-plugins/structleak_plugin.so
make[2]: *** [scripts/Makefile.build:281: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/SUPDrvGip.o] Error 1
cc1: error: incompatible gcc/plugin versions
cc1: error: fail to initialize plugin ./scripts/gcc-plugins/structleak_plugin.so
make[2]: *** [scripts/Makefile.build:281: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv/SUPDrvSem.o] Error 1
make[1]: *** [scripts/Makefile.build:497: /var/lib/dkms/vboxhost/6.1.4_OSE/build/vboxdrv] Error 2
make: *** [Makefile:1626: _module_/var/lib/dkms/vboxhost/6.1.4_OSE/build] Error 2
make: Leaving directory '/usr/lib/modules/5.3.18-1-bcachefs-git-196109-g3d16f5d33e21/build'

The latest kernel seems to be able to build the latest kernel, atleast in my configuration, so it seems the issue with bcachefs root has been fixed.

guys can someone confirm that makepkg works with bcachefs root so i can close the issue in github

Thanks!

I'd missed added the updated config file to my update. :/

I got this today:

:: Downloading the latest sources for a devel package linux-bcachefs-git...
:: error: failed to retrieve latest dev sources:
==> Making package: linux-bcachefs-git v5.3.15.arch1.r859832.ac127f6e3fa5-1 (Sat 07 Dec 2019 10:50:43 AM CST)
==> WARNING: Skipping dependency checks.
==> Retrieving sources...
  -> Cloning bcachefs git repo...
Cloning into bare repository '/home/bobpaul/.cache/pikaur/build/linux-bcachefs-git/bcachefs'...
  -> Cloning kernel_gcc_patch git repo...
Cloning into bare repository '/home/bobpaul/.cache/pikaur/build/linux-bcachefs-git/kernel_gcc_patch'...
  -> Found config
  -> Found d938db454691d69590eb081db22c507c9ebca5f8.patch
  -> Found 4932fe6b8e63b92f8252591b5381423c01bac552.patch
==> Validating source files with sha512sums...
    bcachefs ... Skipped
    kernel_gcc_patch ... Skipped
    config ... FAILED
    d938db454691d69590eb081db22c507c9ebca5f8.patch ... Passed
    4932fe6b8e63b92f8252591b5381423c01bac552.patch ... Passed
==> ERROR: One or more files did not pass the validity check!

I had to use makepkg -g to generate new hashes. It's still building.

@QuartzDragon updated everything to work with the new changes - just without docs

running 5.4.0-rc8-1-mainline-bcachefs-00303-ga29c11e13cf5-dirty atm, everything is as expected

Ah, no, that was an error somehow caused by my previous fixing of EXTRAVERSION. Worked for 5.2, but is broken and also obsolete for 5.3.

So, in a way, yes, fixed by upstream ages ago, probably.

bcachefs had been stuck on 5.2 for ages.

I've tested it a few times but my first issue continues while the second issue has gone away with the newest PKGBUILD. So I guess upstream fixed itself?

@nicman23

Yes, you have. :P

I maintain my package by following all of the changes at https://git.archlinux.org/svntogit/packages.git/log/trunk?h=packages/linux

60-linux.hook, 90-linux.hook, linux.install, linux.preset are all obsolete now, upstream. Instead, Pacman now has builtin hooks that replace all that.

I've also updated your PKGBUILD to match my own: https://invent.kde.org/snippets/574

changes? what changes o.O ? i have been living under a rock in the past weeks

@nicman23

Can you sync your package with mine, where appropriate, to reflect the latest Arch kernel package changes? Some pretty significant ones recently that I notice you haven't seem to have been aware of.

hi, 2 things:

• i maintain a version with the latest tag of mainline linux here https://aur.archlinux.org/packages/linux-mainline-bcachefs/

• at least on the above version, compiling a package with makepkg in bcachefs root, no longer corrupts the binaries in the resulting pkg/ or pkg.tar.xz

Ah, I see... my fixing of EXTRAVERSION, which was working for 5.2, isn't needed anymore.

Weird... really weird.

I could reproduce it multiple times, but after I made that comment... suddenly, I've not been able to reproduce it.

Please try building it again now.

@tscs37

Thanks for reporting. :)

Cannot reproduce your first issue, but can reproduce the second.

Please report a bug for bcachefs, while I figure out a stopgap fix.

When building from a clean copy of the repo, I get an error that a fast forward could not be performed. It occurs when building on bcachefs but also on xfs/ext4 and on cleanbuilds or using a (previously failed with this error) repository. I'm using yay to run the build.

Adding the "--no-ff" Option to line 125 and 135 fixes this.

After that I get a build failure from sphinx as seen in the second snippet.

==> Starting prepare()...
  -> Setting version...
  -> Adding patches from Linux upstream kernel repository...
remote: Enumerating objects: 10276, done.
remote: Counting objects: 100% (10276/10276), done.
remote: Compressing objects: 100% (3296/3296), done.
remote: Total 10276 (delta 7095), reused 9673 (delta 6956)
Receiving objects: 100% (10276/10276), 13.17 MiB | 13.03 MiB/s, done.
Resolving deltas: 100% (7095/7095), completed with 732 local objects.
From <https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux>
 * tag                         v5.3.11    -> FETCH_HEAD
fatal: Not possible to fast-forward, aborting.
==> ERROR: A failure occurred in prepare().
    Aborting...
Error making: linux-bcachefs-git (linux-bcachefs-git linux-bcachefs-git-docs linux-bcachefs-git-headers)



  CC      arch/x86/events/probe.o
usage: sphinx-build [OPTIONS] SOURCEDIR OUTPUTDIR [FILENAMES...]
sphinx-build: error: argument -a: ignored explicit argument 'rch1'
  CC      arch/x86/events/amd/core.o
make[1]: *** [Documentation/Makefile:81: htmldocs] Fehler 2
make: *** [Makefile:1579: htmldocs] Error 2
make: *** Waiting for unfinished jobs....

@nicman23 yeah that was pretty much the issue

creating a ext4 /tmp and patching makepkg fixes the issue afaict.

@tscs37

this is a known issue when packaging anything on a bcachefs root. either compile the package on some other fs (tmpfs is also a choice) or patch your makepkg like so

--- /usr/bin/makepkg    2019-03-01 03:41:56.000000000 +0200
+++ /home/nikos/.local/bin/makepkg      2019-07-21 11:47:22.353595608 +0300
@@ -1465,10 +1465,10 @@

 if [[ $BUILDDIR -ef "$startdir" ]]; then
        srcdir="$BUILDDIR/src"
-       pkgdirbase="$BUILDDIR/pkg"
+       pkgdirbase="/tmp/$BUILDDIR/pkg"
 else
        srcdir="$BUILDDIR/$pkgbase/src"
-       pkgdirbase="$BUILDDIR/$pkgbase/pkg"
+       pkgdirbase="/tmp/$BUILDDIR/$pkgbase/pkg"

 fi

@tscs37

Hmmm. Cannot reproduce, sorry. :/

How are you building this? What PKGBUILD settings are you using?

I'm getting a bunch of errors from mkinitcpio after installing the latest version of this kernel;

depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/btrfs.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/trusted.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/scsi_mod.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/sd_mod.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/sr_mod.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/cxgb4.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/usb-storage.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/mmc_block.ko.xz: Exec format error
depmod: ERROR: failed to load symbols from /tmp/mkinitcpio.5iO0rN/root/lib/modules/5.2.0-arch1-1-bcachefs-git-180243-gb83dba0d3fde/kernel/virtio.ko.xz: Exec format error

I've changed the package version scheme to better reflect the commit of the build, so if you need to report a bug upstream, you can see what commit you're using at a glance.

I think I might cautiously wait until Kent fixes up bcachefs proper.

I guess I don't quite trust the conflict handler to do a proper fixup that won't cause issues...

https://pastebin.com/zDhCzA9r

diff to the pkgbuild, it includes:

• automatic latest x.x.y detection ie if you set _srcver_tag=5.1 then it goes to 5.1.19 + it correctly sets pkgver

• automatic conflict handler (the below code did not work because git merge is silly with tags)

• removal of bluetooth patch - it is included iirc E: it is

• having the upstream git repo fetched in source=() from makepkg and added as a local filesystem remote

• renaming prepare to actual_prepare because makepkg ordering is silly

compiling right now will update you within the hour for any possible issues but there ought to be none

E: running it since a couple of hours ago, nothing wrong to report

Implemented the first suggestion.

Second suggestion is borked:

merge: upstream_stable/v5.1.18 - not something we can merge
fatal: '--ours/--theirs' cannot be used with switching branches

(echo "$_subarch"; yes) | make oldconfig

is better

also for simple conflicts, such as the one that is blocking 5.1.17, remove the git pull and add this

git fetch --tags upstream_stable
if ! git merge upstream_stable/v${_srcver_tag}
  then if ! git diff --name-only --diff-filter=U | xargs git checkout --theirs
    then git diff --name-only --diff-filter=U |
    while read line
    do sleep 0.1; if ! git checkout --theirs $line
      then sleep 0.1; git rm $line
      fi
    done
  fi
fi

Stuck at 5.1.15 for now.

5.1.16 produces an auto-merge conflict.

5.1 is a go it seams https://github.com/koverstreet/bcachefs/issues/46#issuecomment-508420107

Reverted back to 5.0.21 for now.

Will revert.

do not update, it kernel panics as per the latest issue in the project's github

If there are bugs related to the PKGBUILD or config, report here.

Otherwise, please report here:

https://github.com/koverstreet/bcachefs/issues

Done. :)

please add || true on line 109

and the other remote add

Should be fixed now. :)

Forgot to update it. :P

bad checksum on the config? Is anyone else seeing that?

@tscs37

This commit may have fixed that:

https://aur.archlinux.org/cgit/aur.git/commit/?h=linux-bcachefs-git&id=ad464d630a76

I had used backticks without realizing that Bash will execute this blindly, as a command, even when encased in normal quotes.

@QuartzDragon the latest commit fixed the issue and the package builds headless again.

@tscs37

I've implemented your patch. Thanks. :)

Can't help you with the headless build issues, sorry. :/

The current PKGBUILD doesn't work when building headlessly, I made a small patch to prevent git revert from opening vi when running, the commit message here doesn't matter anyway

I'm however still having an issue with the build; after the message "Prepared linux-bcachefs-git version 5.0.6-arch1-1-bcachefs-git-151713-g2ee4babf9fa8-dirty", I get the error "rm: cannot remove '/logdest/logpipe.ink8OBNo': No such file or directory" I'm gonna look into that and see if I can fix that too...

diff --git a/PKGBUILD b/PKGBUILD
index [`55d15f1`](https://aur.archlinux.org/cgit/aur.git/commit/?h=linux-bcachefs-git&id=55d15f1)..c76edf8 100755
--- a/PKGBUILD
+++ b/PKGBUILD
@@ -99,7 +99,7 @@ prepare() {
     git pull arch_stable v$_srcver_arch

     msg2 "Reverting 'block: remove the unused bio_iov_iter_get_pages export' which causes `make modules` to fail"
-    git revert 0374e1132217711bc2e920cde877dd7fc3dbd2d9
+    git revert --no-edit 0374e1132217711bc2e920cde877dd7fc3dbd2d9

     # <https://github.com/graysky2/kernel_gcc_patch>
     msg2 "Patching to enabled additional gcc CPU optimizatons..."

5.0.6 update at last. :)

May not update for the time being, as the rebase is failing to build.

Bug report here:

https://github.com/koverstreet/bcachefs/issues/27

Thanks for the notification!

I'll do some testing to make sure it works, and then I'll push the update.

https://github.com/koverstreet/bcachefs/issues/13#issuecomment-479570397

5.0 rebase done :D

I really hope this becomes mainstream. This file system has so much potential!

Will just track master from now on.

If there are problems ~ contact Kent about it. :)

works with new upstream tag :)

4.20.3-arch1-1-bcachefs-arch-30730-gf86045ca260b-dirty

If you're feeling so inclined, to help with debugging the OOPS, alter the PKGBUILD to remove the chosen commit to pull the latest master code.

Kent needs a kernel panic backtrace for it, but I don't understand how to get one myself. :/

Because I'd rather not hold this back any longer, I'm pushing it to last known working commit for bcachefs, being 303e27810eaf58a0f52518681c90422e50113d63.

Still trying to bisect why it panics beyond that.

@nicman23

Yeah, I know. :/

Please email Kent the results of your OOPS ~ that should help him isolate the issue. :)

it does compile but it oops at like 1.2 secs :/

I should have checked before... bcachefs doesn't build properly with 4.20 at the moment. :|

4.20 version is out now. :)

That's my tired mistake.

For 4.19, the Arch kernel only goes up to .12, because they've now started using 4.20.

A proper update will have to wait until Kent rebases onto 4.20, sorry.

I have problem with building ==> Starting prepare()... -> Adding patches from Arch Linux kernel repository... fatal: Couldn't find remote ref v4.19.13-arch1

I've added a patch from https://bugzilla.kernel.org/show_bug.cgi?id=201685 that fix the EXT4 corruption bugs caused by BLK-MQ.

Will probably be in the next stable release, but I've added it early. :)

Fixed. :)

nconfig can now be enabled by enabling a variable in the PKGBUILD. :)

I did that just in case anyone wants to modify the config before building.

I guess I should take a leaf from linux-ck's PKGBUILD... hold on.

Atm the compile process is interrupted by nconfig, that's probably not intentional, removing the line with "make nconfig" should do it.