[RFC PATCH] jffs2: disable empty summaries instead of calling BUG()

[RFC PATCH] jffs2: disable empty summaries instead of calling BUG()

[eltavine]

With CONFIG_JFFS2_SUMMARY enabled, jffs2_do_reserve_space() tries to
write out the current block's summary when a node and its summary
overhead do not fit. jffs2_sum_write_sumnode() assumes that at least
one summary entry has been collected and calls BUG() otherwise.

An empty summary is a valid state. For example, the syzbot reproducer
mounts an image whose selected nextblock contains only a cleanmarker
and padding, leaving 140 bytes free and no summary entries. Creating
a symlink needs 128 bytes for its inode node, but reserving another
60 bytes for the summary reaches the BUG():

  kernel BUG at fs/jffs2/summary.c:868!

Large pristine-node GC reservations and xattr updates that fail after
reserving space can also leave a newly selected block with an empty
summary and reach the same check.

Disable summary collection for the block and return through the
existing exit path. jffs2_do_reserve_space() then retries the
reservation without summary overhead, following the existing fallback
for a summary that does not fit. Use dbg_summary() for this valid
condition.

The block is scanned normally on the next mount. The on-flash format
and xattr size limits are unchanged.

Fixes: e631ddba5887 ("[JFFS2] Add erase block summary support (mount time improvement)")
Reported-by: syzbot+0727f46a0571de01bf93@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=0727f46a0571de01bf93
Reported-by: syzbot+48f2d47f83a4c223ad14@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=48f2d47f83a4c223ad14
Reported-by: syzbot+e4484dcf0982d86ca2e1@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=e4484dcf0982d86ca2e1
Cc: stable@vger.kernel.org
Signed-off-by: eltavine <me@eltavine.com>
---
Validation: Built fs/jffs2/ in an arm64 Linux VM with GCC 13.3 (W=1),
with and without summary support; the latter warns about unused vecs
in writev.c. Not runtime-tested.

I'm working from an arm64 Mac and don't have a physical Linux test
machine available. Sorry for the limited testing; I'd appreciate
review and help running the reproducer:)

 fs/jffs2/summary.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/jffs2/summary.c b/fs/jffs2/summary.c
index 4521a7723f30..4a07b7911b21 100644
--- a/fs/jffs2/summary.c
+++ b/fs/jffs2/summary.c
@@ -864,8 +864,11 @@ int jffs2_sum_write_sumnode(struct jffs2_sb_info *c)
 		goto out;
 
 	if (!c->summary->sum_num || !c->summary->sum_list_head) {
-		JFFS2_WARNING("Empty summary info!!!\n");
-		BUG();
+		/* Nothing collected yet, e.g. freshly opened jeb */
+		dbg_summary("empty summary, disabling it for jeb at 0x%08x\n",
+			    jeb->offset);
+		jffs2_sum_disable_collecting(c->summary);
+		goto out;
 	}
 
 	datasize = c->summary->sum_size + sizeof(struct jffs2_sum_marker);

base-commit: 0fcb05b72344f40806cbe810de8872f1698bf053