Re: [RFC PATCH v5] mm, pmem, xfs: Introduce MF_MEM_REMOVE for unbind

From: Darrick J. Wong
Date: Fri Jul 08 2022 - 11:28:41 EST


On Fri, Jul 08, 2022 at 05:42:22AM +0000, ruansy.fnst@xxxxxxxxxxx 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() # was pmem driver ->remove() in v1
> -> kill_dax()
> -> dax_holder_notify_failure(dax_dev, 0, U64_MAX, MF_MEM_REMOVE)
> -> xfs_dax_notify_failure()
>
> Introduce MF_MEM_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 v4:
> 1. sync_filesystem() at the beginning when MF_MEM_REMOVE
> 2. Rebased on next-20220706
>
> Changes since v3:
> 1. Flush dirty files and logs when pmem is about to be removed.
> 2. Rebased on next-20220701
>
> Changes since v2:
> 1. Rebased on next-20220615
>
> Changes since v1:
> 1. Drop the needless change of moving {kill,put}_dax()
> 2. Rebased on '[PATCHSETS] v14 fsdax-rmap + v11 fsdax-reflink'[2]
>
> [1]: https://lore.kernel.org/linux-mm/161604050314.1463742.14151665140035795571.stgit@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> [2]: https://lore.kernel.org/linux-xfs/20220508143620.1775214-1-ruansy.fnst@xxxxxxxxxxx/
>
> Signed-off-by: Shiyang Ruan <ruansy.fnst@xxxxxxxxxxx>
> ---
> drivers/dax/super.c | 2 +-
> fs/xfs/xfs_notify_failure.c | 16 ++++++++++++++++
> 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..d4bc83159d46 100644
> --- a/drivers/dax/super.c
> +++ b/drivers/dax/super.c
> @@ -323,7 +323,7 @@ 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_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 aa8dc27c599c..728b0c1d0ddf 100644
> --- a/fs/xfs/xfs_notify_failure.c
> +++ b/fs/xfs/xfs_notify_failure.c
> @@ -18,6 +18,7 @@
> #include "xfs_rmap_btree.h"
> #include "xfs_rtalloc.h"
> #include "xfs_trans.h"
> +#include "xfs_log.h"
>
> #include <linux/mm.h>
> #include <linux/dax.h>
> @@ -75,6 +76,10 @@ 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_REMOVE) {
> + return 0;
> + }

Nit: no curly braces needed here.

> xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_ONDISK);
> return -EFSCORRUPTED;
> }
> @@ -168,6 +173,14 @@ xfs_dax_notify_failure(
> struct xfs_mount *mp = dax_holder(dax_dev);
> u64 ddev_start;
> u64 ddev_end;
> + int error;
> +
> + if (mf_flags & MF_MEM_REMOVE) {
> + xfs_info(mp, "device is about to be removed!");
> + error = sync_filesystem(mp->m_super);

sync_filesystem requires callers to hold s_umount. Does the dax media
failure code take that lock for us, or is this missing a lock?

Also, I'm not sure it's a good idea to sync_filesystem() before checking
if SB_BORN has been set.

> + if (error)
> + return error;
> + }
>
> if (!(mp->m_sb.sb_flags & SB_BORN)) {
> xfs_warn(mp, "filesystem is not ready for notify_failure()!");
> @@ -182,6 +195,9 @@ 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_REMOVE) {
> + return 0;
> + }

Same nit about not needing curly braces.

> 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 794ad19b57f8..3eab2d7ba884 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -3240,6 +3240,7 @@ enum mf_flags {
> MF_UNPOISON = 1 << 4,
> MF_SW_SIMULATED = 1 << 5,
> MF_NO_RETRY = 1 << 6,
> + MF_MEM_REMOVE = 1 << 7,

This is more of a pre-removal notification, right? I think the flag
value ought to be named that way too (MF_MEM_PRE_REMOVE).

--D

> };
> int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
> unsigned long count, int mf_flags);
> --
> 2.37.0