Re: git: 112c453ba910 - stable/15 - geom(9): struct bio KBI fix

On Mon, Dec 29, 2025 at 10:22:50AM +0100, Kristof Provost wrote:
> On 22 Dec 2025, at 1:38, Ka Ho Ng wrote:
> > The branch stable/15 has been updated by khng:
> > 
> > URL: https://cgit.FreeBSD.org/src/commit/?id=112c453ba91012f19ed140c56a3ac8fc929ddabb
> > 
> > commit 112c453ba91012f19ed140c56a3ac8fc929ddabb
> > Author:     Ka Ho Ng <khng@FreeBSD.org>
> > AuthorDate: 2025-12-21 22:45:03 +0000
> > Commit:     Ka Ho Ng <khng@FreeBSD.org>
> > CommitDate: 2025-12-22 00:37:40 +0000
> > 
> >     geom(9): struct bio KBI fix
> > 
> >     The struct bio was changed after cb85c2e2e995 on the branch. To fix
> >     this, move BIO_ERROR flag to another value, and now BIO_ERROR_COMPAT
> >     occupies 0x1 instead. Also, introduce b_error_compat field at the
> > place
> >     where the old bio_error was.
> > 
> >     This allows non-CAM(9) disk drivers and software volume manager
> > modules
> >     compiled against 15.0-RELEASE kernel to work on 15-STABLE kernel
> > again.
> > 
> >     Reviewed by:    kib
> >     Differential Revision:  https://reviews.freebsd.org/D54327
> > 
> >     Approved by:    re (cperciva)
> 
> I’m seeing panics on boot now:
> 
> 	vtnet0: Ethernet address: 58:9c:fc:0c:60:85
> 	vtnet0: netmap queues/slots: TX 1/1024, RX 1panic: Memory modified after
> free 0xfffff800038b51a8 (424, g_bio) + 0 = deadc0dedeacc0de
> 
> 	cpuid = 6
> 	time = 1
> 	KDB: stack backtrace:
> 	db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame
> 0xfffffe008d5679a0
> 	vpanic() at vpanic+0x136/frame 0xfffffe008d567ad0
> 	panic() at panic+0x43/frame 0xfffffe008d567b30
> 	trash_ctor() at trash_ctor+0x74/frame 0xfffffe008d567b40
> 	item_ctor() at item_ctor+0x59/frame 0xfffffe008d567b90
> 	g_clone_bio() at g_clone_bio+0x1d/frame 0xfffffe008d567bb0
> 	g_part_start() at g_part_start+0x96/frame 0xfffffe008d567c40
> 	g_io_request() at g_io_request+0x2ca/frame 0xfffffe008d567c70
> 	g_read_data() at g_read_data+0x95/frame 0xfffffe008d567cb0
> 	g_raid_md_taste_intel() at g_raid_md_taste_intel+0x1d8/frame
> 0xfffffe008d567e50
> 	g_raid_taste() at g_raid_taste+0x18e/frame 0xfffffe008d567ea0
> 	g_new_provider_event() at g_new_provider_event+0xaa/frame
> 0xfffffe008d567ec0
> 	g_run_events() at g_run_events+0x104/frame 0xfffffe008d567ef0
> 	fork_exit() at fork_exit+0x82/frame 0xfffffe008d567f30
> 	fork_trampoline() at fork_trampoline+0xe/frame 0xfffffe008d567f30
> 	--- trap 0, rip = 0, rsp = 0, rbp = 0 ---
> 	KDB: enter: panic
> 	[ thread pid 13 tid 100069 ]
> 	Stopped at      kdb_enter+0x33: movq    $0,0x121aa92(%rip)
> 	db>
> 
> With this commit (and the follow-up “geom(9): unset the BIO_ERROR_COMPAT
> flag correctly”) reverted the VM boots again.

Does the following help?

diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index dc90bd3add22..2368f863a248 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -4523,8 +4523,6 @@ biodone(struct bio *bp)
 			bp->bio_flags |= BIO_ERROR_COMPAT;
 		}
 		done(bp);
-		bp->bio_error_compat = 0;
-		bp->bio_flags &= ~BIO_ERROR_COMPAT;
 	}
 }