On 2026/5/12 15:46, zhaoyifan (H) wrote:
On 2026/5/12 15:33, Gao Xiang wrote:
On 2026/5/12 15:16, Yifan Zhao wrote:
If packed_nid carries the metabox NID bit, loading the packed
inode first redirects its inode metadata lookup through the
metabox inode. Initializing that metabox inode can then read
metadata that refers back through the packed inode, forming a
recursive packed inode -> metabox inode -> packed inode path.
Reject such images while parsing the superblock, matching the
format rule that the special packed inode itself is not stored
inside the metabox.
Reproducible image (base64-encoded gzipped blob):
H4sIAAAAAAAAA2NgGAWjYBSMVPDo4dcHrY0KwsxANg8jAwMLFjVMSGzPx/7Lzj3zXb3jSFTh
5iN7v6CrbQSa8f8/gq8GoRpARHErYxYDEh8EVAm4jw2Il4AMFYDoB7IYmDGVNRhAzf8PBMh+
yEjNyclXKM8vyklRIILNRcA5o2AUjIJRMApGwSgYBaNgFAxpAGorv3VkYtBgQLSfQW3sF8wv
kJvZDSqIXkCDKpANlWxQZ2Bk0NPTS8RlPkgXqP0Oa5/DxNDNB7XvR8EoGAWjYBSMglEwCkbB
KBgFo2AUjIJRQBsAAEZO6n4AIAAA
Assisted-by: Codex:GPT-5.5
Signed-off-by: Yifan Zhao <[email protected]>
Can you fix the kernel instead first and backport to
erofs-utils? like below,
OK. Should we do the same for [PATCH 1/3] too?
Basically for incompatble features, they should assume to 0.
So I don't think [PATCH 1/3] is needed.
Thanks,
Gao Xiang
