[Linux-cachefs] Another Bug with cachefs during underlaying nfs server timeout

[David Howells]

Romain DEGEZ <romain.degez at smartjog.com> wrote:

> [1365947.770498] page:ffffea000cb91cf8 flags:0200000000001008 count:0 mapcount:0 mapping:(null) index:7052

I *think* that means that PG_private_2 (PG_fscache) is still set on the page
(the '1' three places from the end of the flags being bit 12).

Now this looks very odd...  Ext4 doesn't use fscache yet (as far as I know), so
this bit shouldn't be set in Ext4 pages.  The pages can't have had this bit set
when they were allocated, because get_page_from_freelist() checks for that.

I'm not sure what's going on here.  PG_private_2 should only be set on pages
that NFS hands to fscache to read into.  At the point they're set, NFS should
still own them:-/

Are you using anything other than NFS that could be doing caching?  AFS of 9PFS
for example?

David