Re: [RFC/BUG] Use a bounce buffer for mds client decryption

On Thu, Jul 2, 2026 at 4:40 AM Alex Markuze <[email protected]> wrote:
>
> Hi Sam,
>
> NACK for now. Thanks for the patch, but I have some concerns
> that need a v2:

Hi Alex,

This RFC was only to get feedback on my (ultimately improper) approach
of fixing the issue inside parse_reply_info_readdir(), not intended to
be a kernel-worthy patch.

I see you've already found/reviewed the superseding v2 [1], so I
suspect you later stumbled on this one while working on the backlog in
Patchwork and didn't notice that it can be closed. :)

Cheers,
Sam

[1] https://lore.kernel.org/ceph-devel/[email protected]/T/

>
>   1. [major] fs/ceph/mds_client.c:578: BUG_ON on wire-controlled length panics the client
>   2. [major] fs/ceph/mds_client.c:580: Patch may not build against current tree
>   3. [major] fs/ceph/inode.c: Trace/LOOKUPNAME path has the same vmalloc/fscrypt bug
>   4. [minor] fs/ceph/mds_client.c:497: Assignment-in-if and missing braces
>   5. [nit] fs/ceph/mds_client.c:494: Typo: fscript_str → fscrypt_str
>   6. [nit] fs/ceph/mds_client.c:574: Missing unlikely() on inner-loop bounce check
>
>
> --
> Alex Markuze
>