On 4/12/23 15:16, Matthew Wilcox wrote:
On Wed, Apr 12, 2023 at 12:18:29PM +0200, Hannes Reinecke wrote:
Ceterum censeo ...
Having looked at implementing large blocksizes I constantly get bogged down
by buffer_heads and being as they are intricately linked into filesystems
and mm.
And also everyone seems to have agreed to phase out buffer_heads eventually.
So maybe it's time to start discussing exactly _how_ this could be done.
And LSF/MM seems to be the idea location for it.
So far I've came across the following issues:
- reading superblocks / bread(): maybe convert to ->read_folio() ?
- bh_lru and friends (maybe pointless once bread() has been converted)
- How to handle legacy filesystems still running on buffer_heads
I'm sure this is an incomplete list, and I'm equally sure that several
people have their own ideas what should or need to be done.
So this BOF will be about collecting these ideas and coming up with a design
how we can deprecated buffer_heads.
Might be worth reviewing this thread:
https://lore.kernel.org/linux-fsdevel/20230129044645.3cb2ayyxwxvxzhah@garbanzo/
A-ha. So I'm not alone.
Please count me in in that session.
Cheers,
Hannes
--
Dr. Hannes Reinecke Kernel Storage Architect
hare@xxxxxxx +49 911 74053 688
SUSE Software Solutions Germany GmbH, Frankenstr. 146, 90461 Nürnberg
Managing Directors: I. Totev, A. Myers, A. McDonald, M. B. Moerman
(HRB 36809, AG Nürnberg)
