From: Chengguang Xu <cgxu519@mykernel.net>
To: "Amir Goldstein" <amir73il@gmail.com>
Cc: "miklos" <miklos@szeredi.hu>, "jack" <jack@suse.cz>,
"linux-unionfs" <linux-unionfs@vger.kernel.org>,
"linux-fsdevel" <linux-fsdevel@vger.kernel.org>
Subject: Re: [RFC PATCH v3 09/10] ovl: introduce helper of syncfs writeback inode waiting
Date: Mon, 09 Nov 2020 20:06:56 +0800 [thread overview]
Message-ID: <175ace6edde.c8001f892368.3152350117841492998@mykernel.net> (raw)
In-Reply-To: <CAOQ4uxgfi26HDp6YWx3Tgc1tY_EMrfcW_hz5FMG8vXeHLdycBw@mail.gmail.com>
---- 在 星期一, 2020-11-09 18:07:18 Amir Goldstein <amir73il@gmail.com> 撰写 ----
> On Mon, Nov 9, 2020 at 10:34 AM Chengguang Xu <cgxu519@mykernel.net> wrote:
> >
> > ---- 在 星期一, 2020-11-09 15:07:18 Amir Goldstein <amir73il@gmail.com> 撰写 ----
> > > On Mon, Nov 9, 2020 at 5:34 AM Chengguang Xu <cgxu519@mykernel.net> wrote:
> > > >
> > > > ---- 在 星期日, 2020-11-08 22:03:06 Chengguang Xu <cgxu519@mykernel.net> 撰写 ----
> > > > > Introduce a helper ovl_wait_wb_inodes() to wait until all
> > > > > target upper inodes finish writeback.
> > > > >
> > > > > Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
> > > > > ---
> > > > > fs/overlayfs/super.c | 30 ++++++++++++++++++++++++++++++
> > > > > 1 file changed, 30 insertions(+)
> > > > >
> > > > > diff --git a/fs/overlayfs/super.c b/fs/overlayfs/super.c
> > > > > index e5607a908d82..9a535fc11221 100644
> > > > > --- a/fs/overlayfs/super.c
> > > > > +++ b/fs/overlayfs/super.c
> > > > > @@ -255,6 +255,36 @@ static void ovl_put_super(struct super_block *sb)
> > > > > ovl_free_fs(ofs);
> > > > > }
> > > > >
> > > > > +void ovl_wait_wb_inodes(struct ovl_fs *ofs)
> > > > > +{
> > > > > + LIST_HEAD(tmp_list);
> > > > > + struct ovl_inode *oi;
> > > > > + struct inode *upper;
> > > > > +
> > > > > + spin_lock(&ofs->syncfs_wait_list_lock);
> > > > > + list_splice_init(&ofs->syncfs_wait_list, &tmp_list);
> > > > > +
> > > > > + while (!list_empty(&tmp_list)) {
> > > > > + oi = list_first_entry(&tmp_list, struct ovl_inode, wait_list);
> > > > > + list_del_init(&oi->wait_list);
> > > > > + ihold(&oi->vfs_inode);
> > > >
> > > > Maybe I overlooked race condition with inode eviction, so still need to introduce
> > > > OVL_EVICT_PENDING flag just like we did in old syncfs efficiency patch series.
> > > >
> > >
> > > I am not sure why you added the ovl wait list.
> > >
> > > I think you misunderstood Jan's suggestion.
> > > I think what Jan meant is that ovl_sync_fs() should call
> > > wait_sb_inodes(upper_sb)
> > > to wait for writeback of ALL upper inodes after sync_filesystem()
> > > started writeback
> > > only on this ovl instance upper inodes.
> > >
> >
> >
> > Maybe you are right, the wait list is just for accuracy that can completely
> > avoid interferes between ovl instances, otherwise we may need to face
> > waiting interferes in high density environment.
> >
> >
> > > I am not sure if this is acceptable or not - it is certainly an improvement over
> > > current situation, but I have a feeling that on a large scale (many
> > > containers) it
> > > won't be enough.
> > >
> >
> > The same as your thought.
> >
> >
> > > The idea was to keep it simple without over optimizing, since anyway
> > > you are going for the "correct" solution long term (ovl inode aops),
> > > so I wouldn't
> > > add the wait list.
> > >
> >
> > Maybe, I think it depends on how to implement ovl page-cache, so at current
> > stage I have no idea for the wait list.
> >
> >
> > > As long as the upper inode is still dirty, we can keep the ovl inode in cache,
> > > so the worst outcome is that drop_caches needs to get called twice before the
> > > ovl inode can be evicted, no?
> > >
> >
> > IIUC, since currently ovl does not have it's own page-cache, so there is no affect to page-cache reclaim,
> > also there is no ovl shrinker to reclaim slab because we drop ovl inode directly after final iput.
> > So should we add a shrinker in this series?
> >
>
> Would that add a lot of complexity?
Sorry, don't need any other shrinker because inode and dentry use common vfs shrinker.
> Thinking out loud: maybe we follow Jan's suggestion and fix remaining
> performance with followup series?
>
Okay, so let's leave it as homework.
Thanks,
Chengguang
next prev parent reply other threads:[~2020-11-09 12:07 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-11-08 14:02 [RFC PATCH v3 00/10] implement containerized syncfs for overlayfs Chengguang Xu
2020-11-08 14:02 ` [RFC PATCH v3 01/10] ovl: setup overlayfs' private bdi Chengguang Xu
2020-11-08 14:02 ` [RFC PATCH v3 02/10] ovl: introduce waiting list for syncfs Chengguang Xu
2020-11-08 14:03 ` [RFC PATCH v3 03/10] ovl: implement ->writepages operation Chengguang Xu
2020-11-08 14:03 ` [RFC PATCH v3 04/10] ovl: implement overlayfs' ->evict_inode operation Chengguang Xu
2020-11-08 14:03 ` [RFC PATCH v3 05/10] ovl: mark overlayfs' inode dirty on modification Chengguang Xu
2020-11-08 14:03 ` [RFC PATCH v3 06/10] ovl: mark overlayfs' inode dirty on shared mmap Chengguang Xu
2020-11-11 13:05 ` 回复:[RFC " Chengguang Xu
2020-11-11 15:20 ` [RFC " Amir Goldstein
2020-11-11 16:09 ` Chengguang Xu
2020-11-08 14:03 ` [RFC PATCH v3 07/10] ovl: implement overlayfs' ->write_inode operation Chengguang Xu
2020-11-10 13:45 ` Jan Kara
2020-11-10 15:12 ` Chengguang Xu
2020-11-11 10:54 ` Jan Kara
2020-11-10 16:18 ` Amir Goldstein
2020-11-08 14:03 ` [RFC PATCH v3 08/10] ovl: cache dirty overlayfs' inode Chengguang Xu
2020-11-08 14:03 ` [RFC PATCH v3 09/10] ovl: introduce helper of syncfs writeback inode waiting Chengguang Xu
2020-11-09 3:33 ` 回复:[RFC " Chengguang Xu
2020-11-09 7:07 ` [RFC " Amir Goldstein
2020-11-09 8:34 ` Chengguang Xu
2020-11-09 10:07 ` Amir Goldstein
2020-11-09 12:06 ` Chengguang Xu [this message]
2020-11-08 14:03 ` [RFC PATCH v3 10/10] ovl: implement containerized syncfs for overlayfs Chengguang Xu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=175ace6edde.c8001f892368.3152350117841492998@mykernel.net \
--to=cgxu519@mykernel.net \
--cc=amir73il@gmail.com \
--cc=jack@suse.cz \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-unionfs@vger.kernel.org \
--cc=miklos@szeredi.hu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).