[PATCH] xfs_db: fix multiple dblock commands

[PATCH] xfs_db: fix multiple dblock commands

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Tom Samstag reported that running the following sequence of commands no
longer works quite right:

> inode [inodenum]
> dblock 0
> p
> dblock 1
> p
> dblock 2
> p
> [etc]

Mr. Samstag looked into the source code and discovered that the
dblock_f is incorrectly accessing iocur_top->data outside of the
push_cur -> set_cur_inode -> pop_cur sequence that this function uses to
compute the type of the file data.  In other words, it's using
whatever's on top of the stack at the start of the function.  For the
"dblock 0" case above this is the inode, but for the "dblock 1" case
this is the contents of file data block 0, not an inode.

Fix this by relocating the check to the correct place.

Reported-by: tom.samstag@netrise.io
Tested-by: Tom Samstag <tom.samstag@netrise.io>
Cc: <linux-xfs@vger.kernel.org> # v6.12.0
Fixes: b05a31722f5d4c ("xfs_db: access realtime file blocks")
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
---
 db/block.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/db/block.c b/db/block.c
index 00830a3d57e1df..2f1978c41f3094 100644
--- a/db/block.c
+++ b/db/block.c
@@ -246,6 +246,7 @@ dblock_f(
 	int		nb;
 	xfs_extnum_t	nex;
 	char		*p;
+	bool		isrt;
 	typnm_t		type;
 
 	bno = (xfs_fileoff_t)strtoull(argv[1], &p, 0);
@@ -255,6 +256,7 @@ dblock_f(
 	}
 	push_cur();
 	set_cur_inode(iocur_top->ino);
+	isrt = is_rtfile(iocur_top->data);
 	type = inode_next_type();
 	pop_cur();
 	if (type == TYP_NONE) {
@@ -273,7 +275,7 @@ dblock_f(
 	ASSERT(typtab[type].typnm == type);
 	if (nex > 1)
 		make_bbmap(&bbmap, nex, bmp);
-	if (is_rtfile(iocur_top->data))
+	if (isrt)
 		set_rt_cur(&typtab[type], xfs_rtb_to_daddr(mp, dfsbno),
 				nb * blkbb, DB_RING_ADD,
 				nex > 1 ? &bbmap : NULL);

Re: [PATCH] xfs_db: fix multiple dblock commands

[Christoph Hellwig]

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>