[musl] NILFS2 root mounted read-only on Void Linux musl libc system #12

atweiden · 2019-01-12T03:10:10Z

Hi,

I'm really enjoying using NILFS+LVM in my Void Linux installer, and it shows a lot of promise. However, when musl libc is used instead of glibc, the root NILFS2 filesystem fails to mount rw. The boot process essentially fails, and the root NILFS2 filesystem is mounted read-only:

NILFS (dm-1): the device already has a read-only mount.
mount.nilfs2: Error while mounting /dev/mapper/vg0-root on /: Resource busy

Cannot continue due to errors above, starting emergency shell.

On shutdown, I receive this error message:

Assertion failed: cxt->syscall_status == 1 (libmount/src/context_umount.c: mnt_context_do_umount: 903)

edit: I have now also tried using nilfs-utils git HEAD, but it makes no difference. The startup and shutdown errors are the same.

Does anything come to mind re:musl libc?

The text was updated successfully, but these errors were encountered:

- root NILFS2 filesystem is mounted read-only for unknown reasons - behavior appeared twice in a row on fresh installs - may have something to do with `libmount` - see: nilfs-dev/nilfs-utils#12

- mark void-live-x86_64-musl-20181111.iso as failing - root NILFS2 filesystem is mounted read-only for unknown reasons - behavior appeared twice in a row on fresh installs - may have something to do with `libmount` - see: nilfs-dev/nilfs-utils#12 - implement customizable pool names - LVM volume group name = "pool name" - add `PoolName` type - validated against `man 8 lvm` naming rules - most of them, anyway - we do not validate dynamically against existing entries in `/dev/` - add pool name var to examples - add crude warning about naming rules for pool - as opposed to ensuring `$vault-name` and `$pool-name` are unique - which wouldn't ensure `$pool-name` is unique in `/dev/` anyway - we forego an incomplete solution for a crude non-solution and trust the user - simplifies it quite a bit - use `/dev/$pool-name/$lv`, not `/dev/mapper/$pool-name-$lv` - as recommended by `man 8 lvm` line 162: - Links or nodes in /dev/mapper are intended only for internal use and the precise format and escaping might change between releases and distributions - Other software and scripts should use the /dev/VolumeGroupName/LogicalVolumeName format to reduce the chance of needing amendment when the software is updated - configure nilfs_cleanerd to reduce overhead - credit: [xte on HN](https://news.ycombinator.com/item?id=18754205) - xte's posts were the impetus for experimenting with nilfs - fix scripts/* for nilfs+lvm - translate btrfs-administration.md to NILFS+LVM - rm doc/guides/btrfs-administration.md - set GRUB_PRELOAD_MODULES=lvm in /etc/default/grub - doesn't appear to make any difference, but done out of thoroughness

- mark void-live-x86_64-musl-20181111.iso as failing - root NILFS2 filesystem is mounted read-only for unknown reasons - behavior appeared twice in a row on fresh installs - may have something to do with `libmount` - see: nilfs-dev/nilfs-utils#12 - implement customizable pool names - LVM volume group name = "pool name" - add `PoolName` type - validated against `man 8 lvm` naming rules - most of them, anyway - we do not validate dynamically against existing entries in `/dev/` - add pool name var to examples - add crude warning about naming rules for pool - as opposed to ensuring `$vault-name` and `$pool-name` are unique - which wouldn't ensure `$pool-name` is unique in `/dev/` anyway - we forego an incomplete solution for a crude non-solution and trust the user - simplifies it quite a bit - use `/dev/$pool-name/$lv`, not `/dev/mapper/$pool-name-$lv` - as recommended by `man 8 lvm` line 162: - Links or nodes in /dev/mapper are intended only for internal use and the precise format and escaping might change between releases and distributions - Other software and scripts should use the /dev/VolumeGroupName/LogicalVolumeName format to reduce the chance of needing amendment when the software is updated - configure nilfs_cleanerd to reduce overhead - credit: [xte on HN](https://news.ycombinator.com/item?id=18754205) - xte's posts were the impetus for experimenting with nilfs - fix scripts/* for nilfs+lvm - translate btrfs-administration.md to NILFS+LVM - rm doc/guides/btrfs-administration.md - set GRUB_PRELOAD_MODULES=lvm in /etc/default/grub - doesn't appear to make any difference, but done out of thoroughness - lvcreate with mainly absolute --size values - set minimum size of lv to 200M - nilfs has minimum size of 134217728 bytes

nilfsuser5678 · 2024-11-25T22:45:50Z

This also happens in chimera linux: chimera-linux/cports#3214
I get the same error message but with a different line number for the assert since the code has been added to, but it's basically the same error.
Compiling with --without-libmount fixes it.
Remounting also works fine without nilfs-utils, specifically when mount.nilfs2 isn't present. So, this bug can be worked around by simply deleting mount.nilfs2.

nilfsuser5678 · 2024-12-07T01:14:09Z

This has been fixed/worked-around in util-linux/util-linux#3309. The issue is that mount.nilfs2 compiled against musl libc can't parse option arguments that appear after non-option arguments, because of the posix-standard getopt provided by musl. The change in util-linux makes mount call mount helpers with the posix argument order accepted by posix getopt, which improves general compatibility with non-gnu systems and as a consequence fixes this bug.
I'm not sure if there is anything that can/should be done in nilfs-utils about this. It seems a bit silly to expect nilfs-utils to second-guess the intentions of the designers of a system and patch in gnu getopt in order to behave non-posixly on systems that deliberately provide posix getopt instead of the gnu version. The mount program provided on such a system may well only call mount helpers with posix argument order, (as suckless ubase always did and util-linux now does) making gnu getopt pointless. There is no way for nilfs-utils to know either at compile or runtime whether it will be affected by this bug or whether a gnu getopt is really needed. It should really be the job of the person who decided to combine implementations of libc, mount, etc. that mandate/expect posixly correct behaviour with implementations that don't, to deal with this.

nilfsuser5678 mentioned this issue Nov 29, 2024

libmount: exec mount helpers with posixly correct argument order util-linux/util-linux#3309

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[musl] NILFS2 root mounted read-only on Void Linux musl libc system #12

[musl] NILFS2 root mounted read-only on Void Linux musl libc system #12

atweiden commented Jan 12, 2019 •

edited

Loading

nilfsuser5678 commented Nov 25, 2024 •

edited

Loading

nilfsuser5678 commented Dec 7, 2024

[musl] NILFS2 root mounted read-only on Void Linux musl libc system #12

[musl] NILFS2 root mounted read-only on Void Linux musl libc system #12

Comments

atweiden commented Jan 12, 2019 • edited Loading

nilfsuser5678 commented Nov 25, 2024 • edited Loading

nilfsuser5678 commented Dec 7, 2024

atweiden commented Jan 12, 2019 •

edited

Loading

nilfsuser5678 commented Nov 25, 2024 •

edited

Loading