LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Jaegeuk Kim <jaegeuk@kernel.org>
To: Chao Yu <chao@kernel.org>
Cc: Yangtao Li <frank.li@vivo.com>,
	linux-kernel@vger.kernel.org,
	linux-f2fs-devel@lists.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH] f2fs: reset free segment to prefree status when do_checkpoint() fail
Date: Mon, 2 Aug 2021 18:44:48 -0700	[thread overview]
Message-ID: <YQifkGIYCHfaQ/AJ@google.com> (raw)
In-Reply-To: <3d3ef5c9-fbb6-df85-26f7-e493da594b22@kernel.org>

On 08/03, Chao Yu wrote:
> On 2021/8/3 1:59, Jaegeuk Kim wrote:
> > On 08/01, Chao Yu wrote:
> > > On 2021/7/31 6:18, Jaegeuk Kim wrote:
> > > > On 07/20, Chao Yu wrote:
> > > > > On 2021/7/20 2:25, Jaegeuk Kim wrote:
> > > > > > On 07/19, Chao Yu wrote:
> > > > > > > On 2021/4/27 20:37, Chao Yu wrote:
> > > > > > > > I think just reverting dirty/free bitmap is not enough if checkpoint fails,
> > > > > > > > due to we have updated sbi->cur_cp_pack and nat/sit bitmap, next CP tries
> > > > > > > > to overwrite last valid meta/node/data, then filesystem will be corrupted.
> > > > > > > > 
> > > > > > > > So I suggest to set cp_error if do_checkpoint() fails until we can handle
> > > > > > > > all cases, which is not so easy.
> > > > > > > > 
> > > > > > > > How do you think?
> > > > > > > 
> > > > > > > Let's add below patch first before you figure out the patch which covers all
> > > > > > > things.
> > > > > > > 
> > > > > > >    From 3af957c98e9e04259f8bb93ca0b74ba164f3f27e Mon Sep 17 00:00:00 2001
> > > > > > > From: Chao Yu <chao@kernel.org>
> > > > > > > Date: Mon, 19 Jul 2021 16:37:44 +0800
> > > > > > > Subject: [PATCH] f2fs: fix to stop filesystem update once CP failed
> > > > > > > 
> > > > > > > During f2fs_write_checkpoint(), once we failed in
> > > > > > > f2fs_flush_nat_entries() or do_checkpoint(), metadata of filesystem
> > > > > > > such as prefree bitmap, nat/sit version bitmap won't be recovered,
> > > > > > > it may cause f2fs image to be inconsistent, let's just set CP error
> > > > > > > flag to avoid further updates until we figure out a scheme to rollback
> > > > > > > all metadatas in such condition.
> > > > > > > 
> > > > > > > Reported-by: Yangtao Li <frank.li@vivo.com>
> > > > > > > Signed-off-by: Yangtao Li <frank.li@vivo.com>
> > > > > > > Signed-off-by: Chao Yu <chao@kernel.org>
> > > > > > > ---
> > > > > > >     fs/f2fs/checkpoint.c | 10 +++++++---
> > > > > > >     1 file changed, 7 insertions(+), 3 deletions(-)
> > > > > > > 
> > > > > > > diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> > > > > > > index 6c208108d69c..096c85022f62 100644
> > > > > > > --- a/fs/f2fs/checkpoint.c
> > > > > > > +++ b/fs/f2fs/checkpoint.c
> > > > > > > @@ -1639,8 +1639,10 @@ int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
> > > > > > > 
> > > > > > >     	/* write cached NAT/SIT entries to NAT/SIT area */
> > > > > > >     	err = f2fs_flush_nat_entries(sbi, cpc);
> > > > > > > -	if (err)
> > > > > > > +	if (err) {
> > > > > > > +		f2fs_stop_checkpoint(sbi, false);
> > > > > > 
> > > > > > I think we should abuse this, since we can get any known ENOMEM as well.
> > > > > 
> > > > > Yup, but one critical issue here is it can break A/B update of NAT area,
> > > > > so, in order to fix this hole, how about using NOFAIL memory allocation
> > > > > in f2fs_flush_nat_entries() first until we figure out the finial scheme?
> > > > 
> > > > NOFAIL is risky, so how about adding a retry logic on ENOMEM with a message
> > > > and then giving up if we can't get the memory? BTW, what about EIO or other
> > > > family?
> > > 
> > > How about this?
> > 
> > Hmm, it seems we won't get ENOMEM.
> > 
> > __flush_nat_entry_set
> >   -> get_next_nat_page
> >     -> ...
> >      -> __get_meta_page
> >        -> repeat on ENOMEM, but stop_checkpoint on EIO
> 
> Correct, I missed to check __get_meta_page() and f2fs_get_meta_page_retry().
> 
> > 
> > If we have an error here, we should have stopped checkpoint. Have you seen other
> > issue?
> 
> Still we should fix the case from below path?
> 
> - f2fs_write_checkpoint
>  - do_checkpoint
>   - f2fs_flush_device_cache failed

What about adding a retry logic to deal with EIO in __submit_flush_wait()?
We probably need to retry submitting FLUSH commands, and then give up
with f2fs_stop_checkpoint(). And, then how about adding f2fs_bug_on() if
f2fs_flush_nat_entries() returns error without f2fs_cp_error()?

> 
> Thanks,

  reply	other threads:[~2021-08-03  1:44 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-04-27  8:21 Yangtao Li
2021-04-27 12:37 ` [f2fs-dev] " Chao Yu
2021-07-19  8:54   ` Chao Yu
2021-07-19 18:25     ` Jaegeuk Kim
2021-07-20  0:04       ` Chao Yu
2021-07-29  1:42         ` Chao Yu
2021-07-30 22:18         ` Jaegeuk Kim
2021-08-01  9:59           ` Chao Yu
2021-08-02 17:59             ` Jaegeuk Kim
2021-08-03  1:00               ` Chao Yu
2021-08-03  1:44                 ` Jaegeuk Kim [this message]
2021-08-03  2:57                   ` Chao Yu
2021-08-03 18:14                     ` Jaegeuk Kim
2021-08-03 18:15                     ` Jaegeuk Kim

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=YQifkGIYCHfaQ/AJ@google.com \
    --to=jaegeuk@kernel.org \
    --cc=chao@kernel.org \
    --cc=frank.li@vivo.com \
    --cc=linux-f2fs-devel@lists.sourceforge.net \
    --cc=linux-kernel@vger.kernel.org \
    --subject='Re: [f2fs-dev] [PATCH] f2fs: reset free segment to prefree status when do_checkpoint() fail' \
    /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

This is a public inbox, see mirroring instructions
on how to clone and mirror all data and code used for this inbox