https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=256712
Bug ID: 256712
Summary: UFS: kernel panic: ffs_blkfree_cg: freeing free frag –
in response to pkg-delete(8) soon after login
Product: Base System
Version: 13.0-RELEASE
Hardware: amd64
OS: Any
Status: New
Severity: Affects Only Me
Priority: ---
Component: misc
Assignee: [email protected]
Reporter: [email protected]
I began with a fresh installation of FreeBSD 13.0 to UFS, then (in a nutshell):
1. single user mode
2. tunefs -n disable
3. ee /etc/fstab
4. ee /etc/sysctl.conf
5. shutdown -r now
5. an update to the OS
6. shutdown -r now
7. /etc/pkg/FreeBSD.conf set to latest
8. tests for file system resilience.
----
# Device Mountpoint FStype Options Dump Pass#
/dev/ada0s1a / ufs rw,sync 1 1
/dev/ada0s1b none swap sw 0 0
----
kern.filedelay=10
kern.dirdelay=9
kern.metadelay=8
hw.acpi.power_button_state="NONE"
----
Tests typically involved pressing and holding the power button whilst
installing packages:
* specifically, timing the press for interruption to occur
during or immediately after extraction of devel/gdb
– most, if not all, results were OK. Dozens of tests yesterday and this
morning.
----
For just one test, after-effects were remarkable:
pkg: sqlite error while executing INSERT INTO files (path, sha256, package_id)
VALUES (?1, ?2, ?3) in file pkgdb.c:1825: database disk image is malformed
<https://forums.freebsd.org/threads/80925/> – the broken local.sqlite was no
surprise, given the nature of the tests and honestly, I'm not certain that the
breakage was a _direct_ consequence. I mention this incident solely as a point
of reference; unless advised otherwise, I'll treat it as negligible.
======================================================================
For a next round, this morning (with the file system reportedly clean) I
_enabled_ soft updates then tested in much the same way – observing automated
file system checks and repairs at startup time.
Shortly after a login, I ran:
pkg delete -fy python38 gdb
The system crashed. I awaited completion of the dump, suppressed automatic
reboot and took a photograph of the backtrace.
The round of testing that led to this situation was very brief – maybe only two
or three tests (interruptions) – so I hope to make things reproducible with not
too much difficulty.
--
You are receiving this mail because:
You are the assignee for the bug.
