Package: bcachefs-tools
Version: 24+really1.3.4-1
Severity: important
If bcachefs-tools is installed, "mount" fails to mount any bcachefs
filesystems, saying:
.----
Unknown command /dev/nvme0n1p2
bcachefs - tool for managing bcachefs filesystems
usage: bcachefs <command> [<args>]
Superblock commands:
format Format a new filesystem
show-super Dump superblock information to stdout
set-option Set a filesystem option
[... (eleventy lines of help snipped)]
`----
Conversely, if the package is _not_ installed, everything goes fine.
Stracing "mount", I see that it tries to run /sbin/mount.bcachefs (which
is symlinked to "bcachefs") if it exists.
The culprit is:
.--====[ bcachefs.c ]
#ifndef BCACHEFS_NO_RUST
if (strstr(full_cmd, "mount"))
return cmd_mount(argc, argv);
#endif
...
#ifndef BCACHEFS_NO_RUST
if (!strcmp(cmd, "mount"))
return cmd_mount(argc, argv);
if (strstr(cmd, "completions"))
return cmd_completions(argc, argv);
#endif
`----
Ie, the "mount" subcommand (both invoked explicitly and via argv[0]) is not
compiled in our current package. It's not actually needed for basic operation
(the default logic in /sbin/mount is good enough), but if the helper exists,
"mount" will abort if it the helper fails.
I found it puzzling that no one reported this, quite fatal, bug in two weeks
since your upload (I for one have been away from this machine for a while).
However, it turns out systemd does not use this helper but reinvents it
(expect incompatible ways of handling options, but I digress). This means,
out of 21 popcon votes, most folks had boot-time mount succeed, while only
manual mount would fail.
In other words:
✓ boot-time via systemd
✗ boot-time via initscripts
✗ manual via "mount"
A simple workaround would be to just drop the /sbin/mount.bcachefs symlink
until rust pieces are back. But alas, I see that the helper is needed for
handling UUID= stanzas. A typical person would say /dev/nvme0n1p2 while
d-i prefers PARTUUID= thus that'd _usually_ work... but still not ideal.
On the other hand, at least UUID= has never worked for bcachefs on Debian,
thus at least that part is no regression.
Meow!
-- System Information:
Debian Release: trixie/sid
APT prefers testing
APT policy: (666, 'testing'), (500, 'unstable'), (500, 'stable'), (1,
'experimental')
merged-usr: no
Architecture: amd64 (x86_64)
Foreign Architectures: i386
Kernel: Linux 6.6.0-rc5-mithlond-02787-gf0f2365d02d9 (SMP w/4 CPU threads;
PREEMPT)
Kernel taint flags: TAINT_UNSIGNED_MODULE
Locale: LANG=C.UTF-8, LC_CTYPE=C.UTF-8 (charmap=UTF-8), LANGUAGE=en_US:en
Shell: /bin/sh linked to /bin/dash
Init: sysvinit (via /sbin/init)
LSM: AppArmor: enabled
Versions of packages bcachefs-tools depends on:
ii libaio1 0.3.113-5
ii libblkid1 2.39.2-6
ii libc6 2.37-12
ii libkeyutils1 1.6.3-2
ii liblz4-1 1.9.4-1
ii libsodium23 1.0.18-1
ii liburcu8 0.14.0-2
ii libuuid1 2.39.2-6
ii libzstd1 1.5.5+dfsg2-2
ii zlib1g 1:1.2.13.dfsg-3
Versions of packages bcachefs-tools recommends:
ii initramfs-tools [linux-initramfs-tool] 0.142
bcachefs-tools suggests no packages.
-- no debconf information
