Re: [PATCH v7] mm, pmem, xfs: Introduce MF_MEM_REMOVE for unbind

[Shiyang Ruan <ruansy.fnst@xxxxxxxxxxx>]

在 2022/8/27 5:35, Dan Williams 写道:
> Shiyang Ruan wrote:
> > This patch is inspired by Dan's "mm, dax, pmem: Introduce
> > dev_pagemap_failure()"[1].  With the help of dax_holder and
> > ->notify_failure() mechanism, the pmem driver is able to ask filesystem
> > (or mapped device) on it to unmap all files in use and notify processes
> > who are using those files.
> >
> > Call trace:
> > trigger unbind
> >    -> unbind_store()
> >     -> ... (skip)
> >      -> devres_release_all()
> >       -> kill_dax()
> >        -> dax_holder_notify_failure(dax_dev, 0, U64_MAX, MF_MEM_PRE_REMOVE)
> >         -> xfs_dax_notify_failure()
> >
> > Introduce MF_MEM_PRE_REMOVE to let filesystem know this is a remove
> > event.  So do not shutdown filesystem directly if something not
> > supported, or if failure range includes metadata area.  Make sure all
> > files and processes are handled correctly.
> >
> > ==
> > Changes since v6:
> >     1. Rebase on 6.0-rc2 and Darrick's patch[2].
> >
> > Changes since v5:
> >     1. Renamed MF_MEM_REMOVE to MF_MEM_PRE_REMOVE
> >     2. hold s_umount before sync_filesystem()
> >     3. do sync_filesystem() after SB_BORN check
> >     4. Rebased on next-20220714
> >
> > [1]:
> > https://lore.kernel.org/linux-mm/161604050314.1463742.14151665140035795571.stgit@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> > [2]: https://lore.kernel.org/linux-xfs/Yv5wIa2crHioYeRr@magnolia/
> >
> > Signed-off-by: Shiyang Ruan <ruansy.fnst@xxxxxxxxxxx>
> > Reviewed-by: Darrick J. Wong <djwong@xxxxxxxxxx>
> > ---
> >    drivers/dax/super.c         |  3 ++-
> >    fs/xfs/xfs_notify_failure.c | 15 +++++++++++++++
> >    include/linux/mm.h          |  1 +
> >    3 files changed, 18 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/dax/super.c b/drivers/dax/super.c
> > index 9b5e2a5eb0ae..cf9a64563fbe 100644
> > --- a/drivers/dax/super.c
> > +++ b/drivers/dax/super.c
> > @@ -323,7 +323,8 @@ void kill_dax(struct dax_device *dax_dev)
> >    		return;
> >     	if (dax_dev->holder_data != NULL)
> > -		dax_holder_notify_failure(dax_dev, 0, U64_MAX, 0);
> > +		dax_holder_notify_failure(dax_dev, 0, U64_MAX,
> > +				MF_MEM_PRE_REMOVE);
> >     	clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
> >    	synchronize_srcu(&dax_srcu);
> > diff --git a/fs/xfs/xfs_notify_failure.c b/fs/xfs/xfs_notify_failure.c
> > index 65d5eb20878e..a9769f17e998 100644
> > --- a/fs/xfs/xfs_notify_failure.c
> > +++ b/fs/xfs/xfs_notify_failure.c
> > @@ -77,6 +77,9 @@ xfs_dax_failure_fn(
> >     	if (XFS_RMAP_NON_INODE_OWNER(rec->rm_owner) ||
> >    	    (rec->rm_flags & (XFS_RMAP_ATTR_FORK | XFS_RMAP_BMBT_BLOCK))) {
> > +		/* Do not shutdown so early when device is to be removed */
> > +		if (notify->mf_flags & MF_MEM_PRE_REMOVE)
> > +			return 0;
> >    		notify->want_shutdown = true;
> >    		return 0;
> >    	}
> > @@ -182,12 +185,22 @@ xfs_dax_notify_failure(
> >    	struct xfs_mount	*mp = dax_holder(dax_dev);
> >    	u64			ddev_start;
> >    	u64			ddev_end;
> > +	int			error;
> >     	if (!(mp->m_sb.sb_flags & SB_BORN)) {
>
> How are you testing the SB_BORN interactions? I have a fix for this
> pending here:
>
> https://lore.kernel.org/nvdimm/166153428094.2758201.7936572520826540019.stgit@xxxxxxxxxxxxxxxxxxxxxxxxx/

That was my mistake.  Yes, it should be mp->m_super->s_flags.

(I remember my testcase did pass in my dev version, but now that seems 
impossible.  I think something was wrong when I did the test.)

> >    		xfs_warn(mp, "filesystem is not ready for notify_failure()!");
> >    		return -EIO;
> >    	}
> >    +	if (mf_flags & MF_MEM_PRE_REMOVE) {
>
> It appears this patch is corrupted here. I confirmed that b4 sees the
> same when trying to apply it.

Can't this patch be applied?  It is based on 6.0-rc2 + Darrick's patch. 
It's also ok to rebase on 6.0-rc3 + Darrick's patch.

> > +		xfs_info(mp, "device is about to be removed!");
> > +		down_write(&mp->m_super->s_umount);
> > +		error = sync_filesystem(mp->m_super);
>
> This syncs to make data persistent, but for DAX this also needs to get
> invalidate all current DAX mappings. I do not see that in these changes.

I'll add it.


--
Thanks,
Ruan.

> > +		up_write(&mp->m_super->s_umount);
> > +		if (error)
> > +			return error;
> > +	}
> > +
> >    	if (mp->m_rtdev_targp && mp->m_rtdev_targp->bt_daxdev == dax_dev) {
> >    		xfs_warn(mp,
> >    			 "notify_failure() not supported on realtime device!");
> > @@ -196,6 +209,8 @@ xfs_dax_notify_failure(
> >     	if (mp->m_logdev_targp && mp->m_logdev_targp->bt_daxdev == dax_dev &&
> >    	    mp->m_logdev_targp != mp->m_ddev_targp) {
> > +		if (mf_flags & MF_MEM_PRE_REMOVE)
> > +			return 0;
> >    		xfs_err(mp, "ondisk log corrupt, shutting down fs!");
> >    		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_ONDISK);
> >    		return -EFSCORRUPTED;
> > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > index 982f2607180b..2c7c132e6512 100644
> > --- a/include/linux/mm.h
> > +++ b/include/linux/mm.h
> > @@ -3176,6 +3176,7 @@ enum mf_flags {
> >    	MF_UNPOISON = 1 << 4,
> >    	MF_SW_SIMULATED = 1 << 5,
> >    	MF_NO_RETRY = 1 << 6,
> > +	MF_MEM_PRE_REMOVE = 1 << 7,
> >    };
> >    int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
> >    		      unsigned long count, int mf_flags);
> > --
> > 2.37.2