Re: [PATCH v3] cifs: fix pagecache leak when do writepages

[yangerkun <yangerkun@huawei.com>]

在 2025/9/11 18:53, Greg KH 写道:
> On Thu, Sep 11, 2025 at 11:22:57AM +0800, yangerkun wrote:
>> Hello,
>>
>> In stable version 6.6, IO operations for CIFS cause system memory leaks
>> shortly after starting; our test case triggers this issue, and other users
>> have reported it as well [1].
>>
>> This problem does not occur in the mainline kernel after commit 3ee1a1fc3981
>> ("cifs: Cut over to using netfslib") (v6.10-rc1), but backporting this fix
>> to stable versions 6.6 through 6.9 is challenging. Therefore, I have decided
>> to address the issue with a separate patch.
>>
>> Hi Greg,
>>
>> I have reviewed [2] to understand the process for submitting patches to
>> stable branches. However, this patch may not fit their criteria since it is
>> not a backport from mainline. Is there anything else I should do to make
>> this patch appear more formal?
> 
> Yes, please include the info as to why this is not a backport from
> upstream, and why it can only go into this one specific tree and get the
> developers involved to agree with this.

Alright, the reason I favor this single patch is that the mainline 
solution involves a major refactor [1] to change the I/O path to 
netfslib. Backporting it would cause many conflicts, and such a large 
patch set would introduce numerous KABI changes. Therefore, this single 
patch is provided here instead...

[1]. 
https://lore.kernel.org/all/20240328165845.2782259-1-dhowells@redhat.com/

> 
> But why not submit the upstream changes instead?  That should be much
> simpler and is always preferred as that way the code can be maintained
> easier over time.  Whenever we have these one-off changes, they are
> almost always wrong and incur additional development efforts for future
> changes in the same area.
> 
> So please, do the backports first.
> 
> thanks,
> 
> greg k-h
>