NetBSD-Bugs archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: kern/60621 (panic: ufs_inactive: dirty filesystem?)



> Date: Tue, 17 Nov 2026 20:58:53 +0000
> From: "J. Hannken-Illjes" <hannken%netbsd.org@localhost>
> 
> On Tue, Nov 17, 2026 at 07:13:19PM +0000, Taylor R Campbell wrote:
> > This suggests a (purely speculative, unconfirmed) hypothesis about the
> > `ufs_inactive: dirty filesystem?' panic, that the fast path of writing
> > to unlinked internal snapshots doesn't work with wapbl because it
> > allocates blocks and updates i_blocks without grouping them into a
> > single atomic transaction.
> 
> I suppose it is ffs_snapblkfree() claiming the freed block.  The
> block moves to the snapshot and its i_blocks increases.  Don't know
> how to group the snapshot inode update into this transaction yet.

Perhaps, if that's not convenient to do, a cheap alternative would be
relax the panic so it doesn't apply to snapshots?

-	if (ip->i_mode == 0 && (DIP(ip, size) != 0 || DIP(ip, blocks) != 0)) {
+	if (ip->i_mode == 0 &&
+	    (DIP(ip, size) != 0 ||
+		(DIP(ip, blocks) != 0 &&
+		    (ip->i_flags & (SF_SNAPSHOT|SF_SNAPINVAL)) != 0))) {

This should avoid the troublesome symptom (machine crashes
persistently on reboot, requires manual intervention of running fsck
from single-user mode to repair), without losing data (snapshot is
unlinked anyway so it's garbage to be tossed).

I say `cheap' both because:

(a) this is a small change, but also

(b) this doesn't address the underlying ickiness of
    - i_blocks going negative, or
    - (if my hypothesis is correct) non-atomic updates of the inode,
      violating ffs invariants in spite of having wapbl atomic
      transactions available.



Home | Main Index | Thread Index | Old Index