在 2024/11/7 22:32, Chuck Lever 写道:
On Thu, Nov 07, 2024 at 09:22:39AM +0800, yangerkun wrote:
在 2024/11/6 21:35, Chuck Lever 写道:
On Tue, Nov 05, 2024 at 07:03:14PM +0800, Yang Erkun wrote:
From: Yang Erkun <yangerkun@xxxxxxxxxx>
Add nfs4_openowner_unhashed to help found unhashed nfs4_openowner, and
break nfsd4_open process to fix this problem.
Cc: stable@xxxxxxxxxxxxxxx # 2.6
Hi -
Questions about the "stable@" tag:
- You refer above to a leak of nfsd_file objects, but the nfsd_file
cache was added in v5.4. Any thoughts about what might be leaked,
if anything, in kernels earlier than v5.4?
From the above analysis, actually openowner is leaked, and all object
associated with it has been leaked too, include nfsd_file, and openowner
seems already been there since 2.6....
Before v5.4, openowners are leaked. After, openowners and nfsd_file
objects are leaked. Got it.
Yes
- Have you tried applying this patch to LTS kernels?
I have not try to apply this to LTS, what I think is all kernel after 2.6
has this bug...
Understood.
Is "2.6" a guess, or do you know of a specific kernel version where
this problem started to appear? Generally if a problem goes back far
enough or there isn't sufficient evidence about where the problem
started, we don't want a "# xx.yy" annotation.
Thanks for pointing that out! Yes, 2.6 is just a guess, and it's really
not appropriate to say that 2.6 is involved in the beginning.
I expect the stable folks will pull this fix into LTS kernels
automatically, and I have nightly CI running on all of those. That
can catch problems with applying recent fixes to old code bases, but
it ain't perfect.
