Hi, With GCC 9 coming soon, its new default warning, -Waddress-of-packed-member, is already causing a lot of btrfs-progs warnings. It's pretty sure kernel will just suppress this warning, but this makes me to think about the proper way to use packed attribute. To my poor understanding, we should only use packed attribute handling: - On-disk format Obviously. And also needs extra handlers to do the endian convert. - Ioctl parameters To make sure the format doesn't change. - Fixed format packages For network packages. But then this means, we should have two copies of data for every such structures. One for the fixed format one, and one for the compiler aligned one, with enough helper to convert them (along with needed endian convert). Is that the correct practice? And for a btrfs specific question, why we have packed attribute for btrfs_key? I see no specific reason to make a CPU native structure packed, not to mention we already have btrfs_disk_key. And no ioctl structure is using btrfs_key in its parameter. Thanks, Qu
Attachment:
signature.asc
Description: OpenPGP digital signature
