kern/60523: mount_lfs can break inode freelist consistency in ifile

["Shinichi Doyashiki via gnats" <gnats-admin%NetBSD.org@localhost>]

>Number:         60523
>Category:       kern
>Synopsis:       mount_lfs can break inode freelist consistency in ifile
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Aug 01 11:50:00 +0000 2026
>Originator:     Shinichi Doyashiki
>Release:        NetBSD 11.99.7 around 2026-07
>Organization:
	at home
>Environment:
System: NetBSD kanade2.hq.csel.org 11.99.7 NetBSD 11.99.7 (KANADE2_12) #60: Sat Aug 1 19:49:26 JST 2026 clare%kanade2.hq.csel.org@localhost:/export/netbsd/debuglfs/src/sys/arch/amd64/compile/KANADE2_12 amd64
Architecture: x86_64
Machine: amd64
>Description:
mount_lfs can break inode freelist consistency in ifile.

*** PREPARE TEST ***
===> create lfs
Creating a version 2 LFS32 with roll-forward ident 0x7e39d1b3
1023.0MB in 1023 segments of size 1048576
super-block backups (for fsck -b #) at:
16, 208896, 417792, 626688, 835584, 1044480, 1253376, 1462272, 1671168, 
1880064.
===> mount lfs
===> create files
===> umount lfs
===> fsck lfs
** /dev/rdk3 (NO WRITE)
** File system is already clean
** Last Mounted on /lfs
** Phase 0 - Check Inode Free List
** Phase 1 - Check Blocks and Sizes
** Phase 2 - Check Pathnames
** Phase 3 - Check Connectivity
** Phase 4 - Check Reference Counts
** Phase 5 - Check Segment Block Accounting
2 files, 41 used, 942948 free
===> dumplfs lfs
*** RUN TEST ***
===> mount lfs
===> umount lfs
===> fsck lfs
** /dev/rdk3 (NO WRITE)
** File system is already clean
** Last Mounted on /lfs
** Phase 0 - Check Inode Free List
INO 18 FREE BUT NOT ON FREE LIST

FIX? no

INO 17 FREE BUT NOT ON FREE LIST

FIX? no

INO 16 FREE BUT NOT ON FREE LIST

FIX? no

INO 15 FREE BUT NOT ON FREE LIST

FIX? no

INO 14 FREE BUT NOT ON FREE LIST

FIX? no

INO 13 FREE BUT NOT ON FREE LIST

FIX? no

INO 12 FREE BUT NOT ON FREE LIST

FIX? no

INO 11 FREE BUT NOT ON FREE LIST

FIX? no

INO 10 FREE BUT NOT ON FREE LIST

FIX? no

INO 9 FREE BUT NOT ON FREE LIST

FIX? no

INO 8 FREE BUT NOT ON FREE LIST

FIX? no

INO 7 FREE BUT NOT ON FREE LIST

FIX? no

INO 6 FREE BUT NOT ON FREE LIST

FIX? no

INO 5 FREE BUT NOT ON FREE LIST

FIX? no

INO 4 FREE BUT NOT ON FREE LIST

FIX? no

** Phase 1 - Check Blocks and Sizes
** Phase 2 - Check Pathnames
** Phase 3 - Check Connectivity
** Phase 4 - Check Reference Counts
** Phase 5 - Check Segment Block Accounting
2 files, 41 used, 942946 free
===> dumplfs lfs
	
>How-To-Repeat:
#!/bin/sh
echo "*** PREPARE TEST ***"
mkdir -p /lfs
echo "===> create lfs"
newfs_lfs /dev/rdk3
echo "===> mount lfs"
mount_lfs -n /dev/dk3 /lfs
echo "===> create files"
(
  cd /lfs
  for i in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15; do
    touch $i
  done
  sync
  sync
  for i in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15; do
    rm $i
  done
)
echo "===> umount lfs"
umount /lfs
echo "===> fsck lfs"
fsck_lfs -nf /dev/rdk3 | tee before.fsck
echo "===> dumplfs lfs"
dumplfs -i /dev/rdk3 > before.dumplfs
echo "*** RUN TEST ***"
echo "===> mount lfs"
mount_lfs -n /dev/dk3 /lfs
echo "===> umount lfs"
umount /lfs
echo "===> fsck lfs"
fsck_lfs -nf /dev/rdk3 | tee after.fsck
echo "===> dumplfs lfs"
dumplfs -i /dev/rdk3 > after.dumplfs

>Fix:
the following patch is diagnostic workaround; not the fix.

Index: lfs_bio.c
===================================================================
RCS file: /cvsroot/src/sys/ufs/lfs/lfs_bio.c,v
retrieving revision 1.153
diff -u -1 -0 -r1.153 lfs_bio.c
--- lfs_bio.c	2 Dec 2025 01:23:09 -0000	1.153
+++ lfs_bio.c	1 Aug 2026 10:47:08 -0000
@@ -426,21 +426,22 @@
 	KASSERT((bp->b_flags & B_LOCKED) || !(bp->b_oflags & BO_DELWRI));
 	KASSERT(!(flags & BW_CLEAN) || lfs_cleanerlock_held(fs));
 
 	
 	/*
 	 * Don't write *any* blocks if we're mounted read-only, or
 	 * if we are "already unmounted".
 	 *
 	 * In particular the cleaner can't write blocks either.
 	 */
-	if (fs->lfs_ronly || (lfs_sb_getpflags(fs) & LFS_PF_CLEAN)) {
+	if (fs->lfs_ronly ||
+            ((lfs_sb_getpflags(fs) & LFS_PF_CLEAN) && !(fs->lfs_flags & LFS_NOTYET))) {
 		bp->b_oflags &= ~BO_DELWRI;
 		bp->b_flags |= B_READ; /* XXX is this right? --ks */
 		bp->b_error = 0;
 		mutex_enter(&bufcache_lock);
 		LFS_UNLOCK_BUF(bp);
 		if (LFS_IS_MALLOC_BUF(bp))
 			bp->b_cflags &= ~BC_BUSY;
 		else
 			brelsel(bp, 0);
 		mutex_exit(&bufcache_lock);