[PATCH 0/2 V7] Resubmit items failed during writeback

[PATCH 0/2 V7] Resubmit items failed during writeback

[Carlos Maiolino]

Hi,

V7 version of this patchset,

just removed an 'unused variable' warning from an ASSERT in xfs_inode_item_error
in patch 2/2

no changes in patch 1/2

keeping Brian's reviewed-by on both patches

cheers

Carlos Maiolino (2):
  xfs: Add infrastructure needed for error propagation during buffer IO
    failure
  xfs: Properly retry failed inode items in case of error during buffer
    writeback

 fs/xfs/xfs_buf_item.c   | 60 ++++++++++++++++++++++++++++++++++++++++++++++++-
 fs/xfs/xfs_buf_item.h   |  3 +++
 fs/xfs/xfs_inode_item.c | 47 ++++++++++++++++++++++++++++++++++----
 fs/xfs/xfs_trans.h      |  8 +++++--
 fs/xfs/xfs_trans_ail.c  |  3 ++-
 fs/xfs/xfs_trans_priv.h | 31 +++++++++++++++++++++++++
 6 files changed, 144 insertions(+), 8 deletions(-)

-- 
2.9.4

[PATCH 1/2 V7] xfs: Add infrastructure needed for error propagation during buffer IO failure

[Carlos Maiolino]

With the current code, XFS never re-submit a failed buffer for IO,
because the failed item in the buffer is kept in the flush locked state
forever.

To be able to resubmit an log item for IO, we need a way to mark an item
as failed, if, for any reason the buffer which the item belonged to
failed during writeback.

Add a new log item callback to be used after an IO completion failure
and make the needed clean ups.

Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
---
V2:
	- Update commit log to include a better description of why this
	  patch is needed and fix spelling mistakes
	- Move xfs_buf_do_callbacks_fail() call into
	  xfs_buf_iodone_callback_error, so the callbacks can be executed
	  before the buffer is released, and only after it has been
	  retried once

V3:
	- fix some loops according to hch suggestion
	- whitespace cleanup

V4:
	- Invoke failure callbacks before reset the I/O error
	- Remove bflags field from iop_error callback
	- move spin_lock/unlock xa_lock up in the stack, handling all
	  log items in the same buffer into a single lock

V5:
	- Reorganize variable declarations
	  in fxs_buf_do_callbacks_fail

V6:
	- Add a comment to xfs_buf_do_callbacks_fail

 fs/xfs/xfs_buf_item.c | 32 +++++++++++++++++++++++++++++++-
 fs/xfs/xfs_trans.h    |  7 +++++--
 2 files changed, 36 insertions(+), 3 deletions(-)

diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
index f6a8422..7573a1f 100644
--- a/fs/xfs/xfs_buf_item.c
+++ b/fs/xfs/xfs_buf_item.c
@@ -29,6 +29,7 @@
 #include "xfs_error.h"
 #include "xfs_trace.h"
 #include "xfs_log.h"
+#include "xfs_inode.h"
 
 
 kmem_zone_t	*xfs_buf_item_zone;
@@ -1054,6 +1055,31 @@ xfs_buf_do_callbacks(
 	}
 }
 
+/*
+ * Invoke the error state callback for each log item affected by the failed I/O.
+ *
+ * If a metadata buffer write fails with a non-permanent error, the buffer is
+ * eventually resubmitted and so the completion callbacks are not run. The error
+ * state may need to be propagated to the log items attached to the buffer,
+ * however, so the next AIL push of the item knows hot to handle it correctly.
+ */
+STATIC void
+xfs_buf_do_callbacks_fail(
+	struct xfs_buf		*bp)
+{
+	struct xfs_log_item	*next;
+	struct xfs_log_item	*lip = bp->b_fspriv;
+	struct xfs_ail		*ailp = lip->li_ailp;
+
+	spin_lock(&ailp->xa_lock);
+	for (; lip; lip = next) {
+		next = lip->li_bio_list;
+		if (lip->li_ops->iop_error)
+			lip->li_ops->iop_error(lip, bp);
+	}
+	spin_unlock(&ailp->xa_lock);
+}
+
 static bool
 xfs_buf_iodone_callback_error(
 	struct xfs_buf		*bp)
@@ -1123,7 +1149,11 @@ xfs_buf_iodone_callback_error(
 	if ((mp->m_flags & XFS_MOUNT_UNMOUNTING) && mp->m_fail_unmount)
 		goto permanent_error;
 
-	/* still a transient error, higher layers will retry */
+	/*
+	 * Still a transient error, run IO completion failure callbacks and let
+	 * the higher layers retry the buffer.
+	 */
+	xfs_buf_do_callbacks_fail(bp);
 	xfs_buf_ioerror(bp, 0);
 	xfs_buf_relse(bp);
 	return true;
diff --git a/fs/xfs/xfs_trans.h b/fs/xfs/xfs_trans.h
index 6bdad6f..442d679 100644
--- a/fs/xfs/xfs_trans.h
+++ b/fs/xfs/xfs_trans.h
@@ -64,11 +64,13 @@ typedef struct xfs_log_item {
 } xfs_log_item_t;
 
 #define	XFS_LI_IN_AIL	0x1
-#define XFS_LI_ABORTED	0x2
+#define	XFS_LI_ABORTED	0x2
+#define	XFS_LI_FAILED	0x4
 
 #define XFS_LI_FLAGS \
 	{ XFS_LI_IN_AIL,	"IN_AIL" }, \
-	{ XFS_LI_ABORTED,	"ABORTED" }
+	{ XFS_LI_ABORTED,	"ABORTED" }, \
+	{ XFS_LI_FAILED,	"FAILED" }
 
 struct xfs_item_ops {
 	void (*iop_size)(xfs_log_item_t *, int *, int *);
@@ -79,6 +81,7 @@ struct xfs_item_ops {
 	void (*iop_unlock)(xfs_log_item_t *);
 	xfs_lsn_t (*iop_committed)(xfs_log_item_t *, xfs_lsn_t);
 	void (*iop_committing)(xfs_log_item_t *, xfs_lsn_t);
+	void (*iop_error)(xfs_log_item_t *, xfs_buf_t *);
 };
 
 void	xfs_log_item_init(struct xfs_mount *mp, struct xfs_log_item *item,
-- 
2.9.4

Re: [PATCH 1/2 V7] xfs: Add infrastructure needed for error propagation during buffer IO failure

[Christoph Hellwig]

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH 2/2 V7] xfs: Properly retry failed inode items in case of error during buffer writeback

[Carlos Maiolino]

When a buffer has been failed during writeback, the inode items into it
are kept flush locked, and are never resubmitted due the flush lock, so,
if any buffer fails to be written, the items in AIL are never written to
disk and never unlocked.

This causes unmount operation to hang due these items flush locked in AIL,
but this also causes the items in AIL to never be written back, even when
the IO device comes back to normal.

I've been testing this patch with a DM-thin device, creating a
filesystem larger than the real device.

When writing enough data to fill the DM-thin device, XFS receives ENOSPC
errors from the device, and keep spinning on xfsaild (when 'retry
forever' configuration is set).

At this point, the filesystem can not be unmounted because of the flush locked
items in AIL, but worse, the items in AIL are never retried at all
(once xfs_inode_item_push() will skip the items that are flush locked),
even if the underlying DM-thin device is expanded to the proper size.

This patch fixes both cases, retrying any item that has been failed
previously, using the infra-structure provided by the previous patch.

Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
---
V2:
	- Fix XFS_LI_FAILED flag removal
	- Use atomic operations to set and clear XFS_LI_FAILED flag
	- Remove check for XBF_WRITE_FAIL in xfs_inode_item_push
	- Add more comments to the code
	- Add a helper function to resubmit the failed buffers, so this
	  can be also used in dquot system without duplicating code

V3:
	- kill xfs_imap_to_bp call using a pointer in the log item to
	  hold the buffer address
	- use xa_lock instead of atomic operations to handle log item
	  flags
	- Add a hold to the buffer for each log item failed
	- move buffer resubmission up in xfs_inode_item_push()

V4:
	- Remove bflags argument from iop_error callback
	- Remove ip argument from xfs_buf_resubmit_failed_buffers
	- Use helpers to set/clear XFS_LI_FAILED flag
	- remove ->xa_lock from the iop->error callback and move it up
	  on the stack, so all log items are processed into a single
	  pair of lock/unlock

V5:
	- fix comments
	- move buf_lock from xfs_buf_resubmit_failed_buffers() up to
	  xfs_inode_item_push, and use trylock instead
	- assert xa_lock is held in xfs_set/clear_li_failed helpers
	  and move such helpers into xfs_trans_priv.h, once
	  lockdep_assert_held() usage will require xfs_ail definition
	- ASSERT XFS_LI_IN_AIL into xfs_clear_li_failed()
	- assert inode is flush locked in xfs_inode_item_error
	- fix bitwise ops mess

V6:
	- Better comment for xfs_inode_item_error
	- Indentation fixes

V7:
	- Fix unused variable warning

 fs/xfs/xfs_buf_item.c   | 28 ++++++++++++++++++++++++++++
 fs/xfs/xfs_buf_item.h   |  3 +++
 fs/xfs/xfs_inode_item.c | 47 +++++++++++++++++++++++++++++++++++++++++++----
 fs/xfs/xfs_trans.h      |  1 +
 fs/xfs/xfs_trans_ail.c  |  3 ++-
 fs/xfs/xfs_trans_priv.h | 31 +++++++++++++++++++++++++++++++
 6 files changed, 108 insertions(+), 5 deletions(-)

diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
index 7573a1f..573fc72 100644
--- a/fs/xfs/xfs_buf_item.c
+++ b/fs/xfs/xfs_buf_item.c
@@ -1234,3 +1234,31 @@ xfs_buf_iodone(
 	xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
 	xfs_buf_item_free(BUF_ITEM(lip));
 }
+
+/*
+ * Requeue a failed buffer for writeback
+ *
+ * Return true if the buffer has been re-queued properly, false otherwise
+ */
+bool
+xfs_buf_resubmit_failed_buffers(
+	struct xfs_buf		*bp,
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
+{
+	struct xfs_log_item	*next;
+
+	/*
+	 * Clear XFS_LI_FAILED flag from all items before resubmit
+	 *
+	 * XFS_LI_FAILED set/clear is protected by xa_lock, caller  this
+	 * function already have it acquired
+	 */
+	for (; lip; lip = next) {
+		next = lip->li_bio_list;
+		xfs_clear_li_failed(lip);
+	}
+
+	/* Add this buffer back to the delayed write list */
+	return xfs_buf_delwri_queue(bp, buffer_list);
+}
diff --git a/fs/xfs/xfs_buf_item.h b/fs/xfs/xfs_buf_item.h
index f7eba99..530686e 100644
--- a/fs/xfs/xfs_buf_item.h
+++ b/fs/xfs/xfs_buf_item.h
@@ -70,6 +70,9 @@ void	xfs_buf_attach_iodone(struct xfs_buf *,
 			      xfs_log_item_t *);
 void	xfs_buf_iodone_callbacks(struct xfs_buf *);
 void	xfs_buf_iodone(struct xfs_buf *, struct xfs_log_item *);
+bool	xfs_buf_resubmit_failed_buffers(struct xfs_buf *,
+					struct xfs_log_item *,
+					struct list_head *);
 
 extern kmem_zone_t	*xfs_buf_item_zone;
 
diff --git a/fs/xfs/xfs_inode_item.c b/fs/xfs/xfs_inode_item.c
index 08cb7d1..9491574 100644
--- a/fs/xfs/xfs_inode_item.c
+++ b/fs/xfs/xfs_inode_item.c
@@ -27,6 +27,7 @@
 #include "xfs_error.h"
 #include "xfs_trace.h"
 #include "xfs_trans_priv.h"
+#include "xfs_buf_item.h"
 #include "xfs_log.h"
 
 
@@ -475,6 +476,23 @@ xfs_inode_item_unpin(
 		wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
 }
 
+/*
+ * Callback used to mark a buffer with XFS_LI_FAILED when items in the buffer
+ * have been failed during writeback
+ *
+ * This informs the AIL that the inode is already flush locked on the next push,
+ * and acquires a hold on the buffer to ensure that it isn't reclaimed before
+ * dirty data makes it to disk.
+ */
+STATIC void
+xfs_inode_item_error(
+	struct xfs_log_item	*lip,
+	struct xfs_buf		*bp)
+{
+	ASSERT(xfs_isiflocked(INODE_ITEM(lip)->ili_inode));
+	xfs_set_li_failed(lip, bp);
+}
+
 STATIC uint
 xfs_inode_item_push(
 	struct xfs_log_item	*lip,
@@ -484,13 +502,28 @@ xfs_inode_item_push(
 {
 	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 	struct xfs_inode	*ip = iip->ili_inode;
-	struct xfs_buf		*bp = NULL;
+	struct xfs_buf		*bp = lip->li_buf;
 	uint			rval = XFS_ITEM_SUCCESS;
 	int			error;
 
 	if (xfs_ipincount(ip) > 0)
 		return XFS_ITEM_PINNED;
 
+	/*
+	 * The buffer containing this item failed to be written back
+	 * previously. Resubmit the buffer for IO.
+	 */
+	if (lip->li_flags & XFS_LI_FAILED) {
+		if (!xfs_buf_trylock(bp))
+			return XFS_ITEM_LOCKED;
+
+		if (!xfs_buf_resubmit_failed_buffers(bp, lip, buffer_list))
+			rval = XFS_ITEM_FLUSHING;
+
+		xfs_buf_unlock(bp);
+		return rval;
+	}
+
 	if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
 		return XFS_ITEM_LOCKED;
 
@@ -622,7 +655,8 @@ static const struct xfs_item_ops xfs_inode_item_ops = {
 	.iop_unlock	= xfs_inode_item_unlock,
 	.iop_committed	= xfs_inode_item_committed,
 	.iop_push	= xfs_inode_item_push,
-	.iop_committing = xfs_inode_item_committing
+	.iop_committing = xfs_inode_item_committing,
+	.iop_error	= xfs_inode_item_error
 };
 
 
@@ -710,7 +744,8 @@ xfs_iflush_done(
 		 * the AIL lock.
 		 */
 		iip = INODE_ITEM(blip);
-		if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
+		if ((iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn) ||
+		    lip->li_flags & XFS_LI_FAILED)
 			need_ail++;
 
 		blip = next;
@@ -718,7 +753,8 @@ xfs_iflush_done(
 
 	/* make sure we capture the state of the initial inode. */
 	iip = INODE_ITEM(lip);
-	if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
+	if ((iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn) ||
+	    lip->li_flags & XFS_LI_FAILED)
 		need_ail++;
 
 	/*
@@ -739,6 +775,9 @@ xfs_iflush_done(
 			if (INODE_ITEM(blip)->ili_logged &&
 			    blip->li_lsn == INODE_ITEM(blip)->ili_flush_lsn)
 				mlip_changed |= xfs_ail_delete_one(ailp, blip);
+			else {
+				xfs_clear_li_failed(blip);
+			}
 		}
 
 		if (mlip_changed) {
diff --git a/fs/xfs/xfs_trans.h b/fs/xfs/xfs_trans.h
index 442d679..7d62772 100644
--- a/fs/xfs/xfs_trans.h
+++ b/fs/xfs/xfs_trans.h
@@ -49,6 +49,7 @@ typedef struct xfs_log_item {
 	struct xfs_ail			*li_ailp;	/* ptr to AIL */
 	uint				li_type;	/* item type */
 	uint				li_flags;	/* misc flags */
+	struct xfs_buf			*li_buf;	/* real buffer pointer */
 	struct xfs_log_item		*li_bio_list;	/* buffer item list */
 	void				(*li_cb)(struct xfs_buf *,
 						 struct xfs_log_item *);
diff --git a/fs/xfs/xfs_trans_ail.c b/fs/xfs/xfs_trans_ail.c
index 9056c0f..70f5ab0 100644
--- a/fs/xfs/xfs_trans_ail.c
+++ b/fs/xfs/xfs_trans_ail.c
@@ -687,12 +687,13 @@ xfs_trans_ail_update_bulk(
 bool
 xfs_ail_delete_one(
 	struct xfs_ail		*ailp,
-	struct xfs_log_item 	*lip)
+	struct xfs_log_item	*lip)
 {
 	struct xfs_log_item	*mlip = xfs_ail_min(ailp);
 
 	trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
 	xfs_ail_delete(ailp, lip);
+	xfs_clear_li_failed(lip);
 	lip->li_flags &= ~XFS_LI_IN_AIL;
 	lip->li_lsn = 0;
 
diff --git a/fs/xfs/xfs_trans_priv.h b/fs/xfs/xfs_trans_priv.h
index d91706c..b317a36 100644
--- a/fs/xfs/xfs_trans_priv.h
+++ b/fs/xfs/xfs_trans_priv.h
@@ -164,4 +164,35 @@ xfs_trans_ail_copy_lsn(
 	*dst = *src;
 }
 #endif
+
+static inline void
+xfs_clear_li_failed(
+	struct xfs_log_item	*lip)
+{
+	struct xfs_buf	*bp = lip->li_buf;
+
+	ASSERT(lip->li_flags & XFS_LI_IN_AIL);
+	lockdep_assert_held(&lip->li_ailp->xa_lock);
+
+	if (lip->li_flags & XFS_LI_FAILED) {
+		lip->li_flags &= ~XFS_LI_FAILED;
+		lip->li_buf = NULL;
+		xfs_buf_rele(bp);
+	}
+}
+
+static inline void
+xfs_set_li_failed(
+	struct xfs_log_item	*lip,
+	struct xfs_buf		*bp)
+{
+	lockdep_assert_held(&lip->li_ailp->xa_lock);
+
+	if (!(lip->li_flags & XFS_LI_FAILED)) {
+		xfs_buf_hold(bp);
+		lip->li_flags |= XFS_LI_FAILED;
+		lip->li_buf = bp;
+	}
+}
+
 #endif	/* __XFS_TRANS_PRIV_H__ */
-- 
2.9.4

Re: [PATCH 2/2 V7] xfs: Properly retry failed inode items in case of error during buffer writeback

[Christoph Hellwig]

>  }
>  #endif
> +
> +static inline void
> +xfs_clear_li_failed(
> +	struct xfs_log_item	*lip)
> +{
> +	struct xfs_buf	*bp = lip->li_buf;
> +
> +	ASSERT(lip->li_flags & XFS_LI_IN_AIL);
> +	lockdep_assert_held(&lip->li_ailp->xa_lock);
> +
> +	if (lip->li_flags & XFS_LI_FAILED) {
> +		lip->li_flags &= ~XFS_LI_FAILED;
> +		lip->li_buf = NULL;
> +		xfs_buf_rele(bp);

This means we now nest pag->pag_buf_lock and bp->b_lock inside
ailp->xa_lock.  I couldn't find any place where we take
it in a different order, but it makes me feel a little nervous.

Also I think the set/clear failed helper might better be out of line
instead of bloating the callers.

Except for that the patch looks fine to me.

Re: [PATCH 2/2 V7] xfs: Properly retry failed inode items in case of error during buffer writeback

[Carlos Maiolino]

Hi Christoph,

On Fri, Aug 11, 2017 at 06:09:57AM -0700, Christoph Hellwig wrote:
> >  }
> >  #endif
> > +
> > +static inline void
> > +xfs_clear_li_failed(
> > +	struct xfs_log_item	*lip)
> > +{
> > +	struct xfs_buf	*bp = lip->li_buf;
> > +
> > +	ASSERT(lip->li_flags & XFS_LI_IN_AIL);
> > +	lockdep_assert_held(&lip->li_ailp->xa_lock);
> > +
> > +	if (lip->li_flags & XFS_LI_FAILED) {
> > +		lip->li_flags &= ~XFS_LI_FAILED;
> > +		lip->li_buf = NULL;
> > +		xfs_buf_rele(bp);
> 
> This means we now nest pag->pag_buf_lock and bp->b_lock inside
> ailp->xa_lock.  I couldn't find any place where we take
> it in a different order, but it makes me feel a little nervous.
> 

If it makes you feel less nervous, I plan to remove the xa_lock there, and use
atomic operations instead, to set/clear the lip->li_flags, as discussed in
previous versions.

> Also I think the set/clear failed helper might better be out of line
> instead of bloating the callers.
> 

Sure, I have no problem in making it a not inline function, My goal in making it
inlined was to not add another call in the stack just to remove the flags, but,
maybe a stupid question, but, what would be the benefit in making it a regular
function instead of inline?


> Except for that the patch looks fine to me.

Thanks for the review.

Cheers

-- 
Carlos