Re: [RFC PATCH v2 1/3] fs/drop_caches: skip filesystems without page cache

[Huang Shijie]

On Mon, Sep 21, 2026 at 10:41:26PM -0700, Christoph Hellwig wrote:
> On Sun, Sep 20, 2026 at 03:28:09PM +0800, Huang Shijie wrote:
> > Add a new flag SB_I_NO_PAGECACHE for superblock.
> > 
> > Skip scanning the inode lists of filesystems that have no page cache
> > in drop_pagecache_sb(), as indicated by the SB_I_NO_PAGECACHE flag.
> 
> I don't think we need a new flag.  This can check for
> 
> 	sb->s_bdi == &noop_backing_dev_info
> 
Okay. I can use this method.

> which is actually the more relevant concept - not if a file system
> uses the page cache, but if it supports page cache writeback.
> E.g. for ramfs there is page cache, but it is the only data store.
> For shmemfs, there is writeback, but it is driven in a different
> way (through the anonymous memory code and not fs writeback).
> 
> And most importantly it just work, no need to mark file systems,
> as you're currently missing a lot of those that can have a fair
> number of inodes, e.g., sysfs, kernfs, cgroupfs or configfs.
> 
I ever added support for these file systems, but I deleted the code before
I sent out this patch.

To Jan Kara,
   thanks for the information about the issue.
   I will check the inode iteration issue firstly.

Thanks
Huang Shijie