Skip to contentSkip to conversation

writeback: Lazytime handling fix and cleanups

PATCH Superseded Jan Kara 3 participants 13 messages Jun 1, 2020 – Jun 10, 2020

Subsystems (6)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Jan Kara

author maintainer
#0 lore raw

cover writeback: Lazytime handling fix and cleanups

Hello,

this patch series fixes an issue with handling of lazy inode timestamp writeback which could result in missed background writeback of an inode or missed update of inode timestamps during sync(2). It also somewhat simplifies the writeback code handling of lazy inode timestamp updates. Review is welcome!

Honza

author maintainer
#1 lore raw

PATCH 1/3 writeback: Avoid skipping inode writeback

Inode's i_io_list list head is used to attach inode to several different lists - wb->{b_dirty, b_dirty_time, b_io, b_more_io}. When flush worker prepares a list of inodes to writeback e.g. for sync(2), it moves inodes to b_io list. Thus it is critical for sync(2) data integrity guarantees that inode is not requeued to any other writeback list when inode is queued for processing by flush worker. That's the reason why writeback_single_inode() does not touch i_io_list (unless the inode is completely clean) and why __mark_inode_dirty() does not touch i_io_list if I_SYNC flag is set.

However there are two flaws in the current logic:

1) When inode has only I_DIRTY_TIME set but it is already queued in b_io list due to sync(2), concurrent __mark_inode_dirty(inode, I_DIRTY_SYNC) can still move inode back to b_dirty list resulting in skipping writeback of inode time stamps during sync(2).

2) When inode is on b_dirty_time list and writeback_single_inode() races with __mark_inode_dirty() like:

writeback_single_inode() __mark_inode_dirty(inode, I_DIRTY_PAGES)
inode->i_state |= I_SYNC
__writeback_single_inode()

					  inode->i_state |= I_DIRTY_PAGES;
					  if (inode->i_state & I_SYNC)
					    bail

if (!(inode->i_state & I_DIRTY_ALL))
- not true so nothing done

We end up with I_DIRTY_PAGES inode on b_dirty_time list and thus standard background writeback will not writeback this inode leading to possible dirty throttling stalls etc. (thanks to Martijn Coenen for this analysis).

Fix these problems by tracking whether inode is queued in b_io or b_more_io lists in a new I_SYNC_QUEUED flag. When this flag is set, we know flush worker has queued inode and we should not touch i_io_list. On the other hand we also know that once flush worker is done with the inode it will requeue the inode to appropriate dirty list. When I_SYNC_QUEUED is not set, __mark_inode_dirty() can (and must) move inode to appropriate dirty list.

Fixes 0ae45f63d4ef ("vfs: add support for a lazytime mount option")
---
 fs/fs-writeback.c  | 39 +++++++++++++++++++++++++++++----------
 include/linux/fs.h |  8 ++++++--
 2 files changed, 35 insertions(+), 12 deletions(-)
fs/fs-writeback.c +29 -10
index 76ac9c7d32ec..855c6611710a 100644
@@ -144,7 +144,9 @@ static void inode_io_list_del_locked(struct inode *inode,
 				     struct bdi_writeback *wb)
 {
 	assert_spin_locked(&wb->list_lock);
+	assert_spin_locked(&inode->i_lock);
 
+	inode->i_state &= ~I_SYNC_QUEUED;
 	list_del_init(&inode->i_io_list);
 	wb_io_lists_depopulated(wb);
 }
@@ -1123,7 +1125,9 @@ void inode_io_list_del(struct inode *inode)
 	struct bdi_writeback *wb;
 
 	wb = inode_to_wb_and_lock_list(inode);
+	spin_lock(&inode->i_lock);
 	inode_io_list_del_locked(inode, wb);
+	spin_unlock(&inode->i_lock);
 	spin_unlock(&wb->list_lock);
 }
 
@@ -1172,8 +1176,9 @@ void sb_clear_inode_writeback(struct inode *inode)
  * the case then the inode must have been redirtied while it was being written
  * out and we don't reset its dirtied_when.
  */
-static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
+static void __redirty_tail(struct inode *inode, struct bdi_writeback *wb)
 {
+	assert_spin_locked(&inode->i_lock);
 	if (!list_empty(&wb->b_dirty)) {
 		struct inode *tail;
 
@@ -1182,6 +1187,14 @@ static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
 			inode->dirtied_when = jiffies;
 	}
 	inode_io_list_move_locked(inode, wb, &wb->b_dirty);
+	inode->i_state &= ~I_SYNC_QUEUED;
+}
+
+static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
+{
+	spin_lock(&inode->i_lock);
+	__redirty_tail(inode, wb);
+	spin_unlock(&inode->i_lock);
 }
 
 /*
@@ -1250,8 +1263,11 @@ static int move_expired_inodes(struct list_head *delaying_queue,
 			break;
 		list_move(&inode->i_io_list, &tmp);
 		moved++;
+		spin_lock(&inode->i_lock);
 		if (flags & EXPIRE_DIRTY_ATIME)
-			set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
+			inode->i_state |= I_DIRTY_TIME_EXPIRED;
+		inode->i_state |= I_SYNC_QUEUED;
+		spin_unlock(&inode->i_lock);
 		if (sb_is_blkdev_sb(inode->i_sb))
 			continue;
 		if (sb && sb != inode->i_sb)
@@ -1394,7 +1410,7 @@ static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
 		 * writeback is not making progress due to locked
 		 * buffers. Skip this inode for now.
 		 */
-		redirty_tail(inode, wb);
+		__redirty_tail(inode, wb);
 		return;
 	}
 
@@ -1414,7 +1430,7 @@ static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
 			 * retrying writeback of the dirty page/inode
 			 * that cannot be performed immediately.
 			 */
-			redirty_tail(inode, wb);
+			__redirty_tail(inode, wb);
 		}
 	} else if (inode->i_state & I_DIRTY) {
 		/*
@@ -1422,10 +1438,11 @@ static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
 		 * such as delayed allocation during submission or metadata
 		 * updates after data IO completion.
 		 */
-		redirty_tail(inode, wb);
+		__redirty_tail(inode, wb);
 	} else if (inode->i_state & I_DIRTY_TIME) {
 		inode->dirtied_when = jiffies;
 		inode_io_list_move_locked(inode, wb, &wb->b_dirty_time);
+		inode->i_state &= ~I_SYNC_QUEUED;
 	} else {
 		/* The inode is clean. Remove from writeback lists. */
 		inode_io_list_del_locked(inode, wb);
@@ -1669,8 +1686,9 @@ static long writeback_sb_inodes(struct super_block *sb,
 		 */
 		spin_lock(&inode->i_lock);
 		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
+			inode->i_state &= ~I_SYNC_QUEUED;
+			__redirty_tail(inode, wb);
 			spin_unlock(&inode->i_lock);
-			redirty_tail(inode, wb);
 			continue;
 		}
 		if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
@@ -2289,11 +2307,12 @@ void __mark_inode_dirty(struct inode *inode, int flags)
 		inode->i_state |= flags;
 
 		/*
-		 * If the inode is being synced, just update its dirty state.
-		 * The unlocker will place the inode on the appropriate
-		 * superblock list, based upon its state.
+		 * If the inode is queued for writeback by flush worker, just
+		 * update its dirty state. Once the flush worker is done with
+		 * the inode it will place it on the appropriate superblock
+		 * list, based upon its state.
 		 */
-		if (inode->i_state & I_SYNC)
+		if (inode->i_state & I_SYNC_QUEUED)
 			goto out_unlock_inode;
 
 		/*
include/linux/fs.h +6 -2
index 45cc10cdf6dd..b02290d19edd 100644
@@ -2156,6 +2156,10 @@ static inline void kiocb_clone(struct kiocb *kiocb, struct kiocb *kiocb_src,
  *
  * I_CREATING		New object's inode in the middle of setting up.
  *
+ * I_SYNC_QUEUED	Inode is queued in b_io or b_more_io writeback lists.
+ *			Used to detect that mark_inode_dirty() should not move
+ * 			inode between dirty lists.
+ *
  * Q: What is the difference between I_WILL_FREE and I_FREEING?
  */
 #define I_DIRTY_SYNC		(1 << 0)
@@ -2173,11 +2177,11 @@ static inline void kiocb_clone(struct kiocb *kiocb, struct kiocb *kiocb_src,
 #define I_DIO_WAKEUP		(1 << __I_DIO_WAKEUP)
 #define I_LINKABLE		(1 << 10)
 #define I_DIRTY_TIME		(1 << 11)
-#define __I_DIRTY_TIME_EXPIRED	12
-#define I_DIRTY_TIME_EXPIRED	(1 << __I_DIRTY_TIME_EXPIRED)
+#define I_DIRTY_TIME_EXPIRED	(1 << 12)
 #define I_WB_SWITCH		(1 << 13)
 #define I_OVL_INUSE		(1 << 14)
 #define I_CREATING		(1 << 15)
+#define I_SYNC_QUEUED		(1 << 16)
 
 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
 #define I_DIRTY (I_DIRTY_INODE | I_DIRTY_PAGES)
Signature · 1 line
2.16.4
maintainer
#2 lore raw

Re: [PATCH 1/3] writeback: Avoid skipping inode writeback

Hi

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag fixing commit: 0ae45f63d4ef ("vfs: add support for a lazytime mount option").

The bot has tested the following trees: v5.6.15, v5.4.43, v4.19.125, v4.14.182, v4.9.225, v4.4.225.

v5.6.15: Build OK!
v5.4.43: Build OK!
v4.19.125: Build OK!
v4.14.182: Failed to apply! Possible dependencies:

    80ea09a002bf ("vfs: factor out inode_insert5()")
    c2b6d621c4ff ("new primitive: discard_new_inode()")

v4.9.225: Failed to apply! Possible dependencies:

    09d8b586731b ("ovl: move __upperdentry to ovl_inode")
    13c72075ac9f ("ovl: move impure to ovl_inode")
    2aff4534b6c4 ("ovl: check lower existence when removing")
    2b8c30e9ef14 ("ovl: use d_is_dir()")
    32a3d848eb91 ("ovl: clean up kstat usage")
    370e55ace59c ("ovl: rename: simplify handling of lower/merged directory")
    38e813db61c3 ("ovl: get rid of PURE type")
    3ee23ff1025a ("ovl: check lower existence of rename target")
    42f269b92540 ("ovl: rearrange code in ovl_copy_up_locked()")
    5cf5b477f0ca ("ovl: opaque cleanup")
    6c02cb59e6fe ("ovl: rename ovl_rename2() to ovl_rename()")
    804032fabb3b ("ovl: don't check rename to self")
    80ea09a002bf ("vfs: factor out inode_insert5()")
    8ee6059c58ea ("ovl: simplify lookup")
    a6c606551141 ("ovl: redirect on rename-dir")
    ad0af7104dad ("vfs: introduce inode 'inuse' lock")
    bbb1e54dd53c ("ovl: split super.c")
    c2b6d621c4ff ("new primitive: discard_new_inode()")
    c412ce498396 ("ovl: add ovl_dentry_is_whiteout()")
    ca4c8a3a8000 ("ovl: treat special files like a regular fs")
    d8514d8edb5b ("ovl: copy up regular file using O_TMPFILE")

v4.4.225: Failed to apply! Possible dependencies:

    09d8b586731b ("ovl: move __upperdentry to ovl_inode")
    1175b6b8d963 ("ovl: do operations on underlying file system in mounter's context")
    13c72075ac9f ("ovl: move impure to ovl_inode")
    2864f3014242 ("iget_locked et.al.: make sure we don't return bad inodes")
    32a3d848eb91 ("ovl: clean up kstat usage")
    38b256973ea9 ("ovl: handle umask and posix_acl_default correctly on creation")
    42f269b92540 ("ovl: rearrange code in ovl_copy_up_locked()")
    6b2553918d8b ("replace ->follow_link() with new method that could stay in RCU mode")
    80ea09a002bf ("vfs: factor out inode_insert5()")
    aa80deab33a8 ("namei: page_getlink() and page_follow_link_light() are the same thing")
    ad0af7104dad ("vfs: introduce inode 'inuse' lock")
    bb0d2b8ad296 ("ovl: fix sgid on directory")
    c2b6d621c4ff ("new primitive: discard_new_inode()")
    d6335d77a762 ("security: Make inode argument of inode_getsecid non-const")
    d8514d8edb5b ("ovl: copy up regular file using O_TMPFILE")
    d8ad8b496184 ("security, overlayfs: provide copy up security hook for unioned files")
    fceef393a538 ("switch ->get_link() to delayed_call, kill ->put_link()")

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

Signature · 2 lines
Thanks
Sasha
#3 lore raw

Re: [PATCH 1/3] writeback: Avoid skipping inode writeback

In reply to Jan Kara

This generall looks ok, but a few nitpicks below:

fs/fs-writeback.c void sb_clear_inode_writeback(struct inode *inode) · line 1179

Quoted 2 lines
fs/fs-writeback.c +1 -1
-static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
+static void __redirty_tail(struct inode *inode, struct bdi_writeback *wb)

I think redirty_tail_locked would be a more decriptive name, and also fit other uses in this file (e.g. inode_io_list_move_locked and inode_io_list_del_locked).

fs/fs-writeback.c void sb_clear_inode_writeback(struct inode *inode) · line 1182

Quoted 3 lines

{

fs/fs-writeback.c +1 -0
+	assert_spin_locked(&inode->i_lock);
 	if (!list_empty(&wb->b_dirty)) {

Nit: I find an empty line after asserts and before the real code starts nice on the eye.

fs/fs-writeback.c static int move_expired_inodes(struct list_head *delaying_queue, · line 1270

Quoted 9 lines
fs/fs-writeback.c +4 -1
 			break;
 		list_move(&inode->i_io_list, &tmp);
 		moved++;
+		spin_lock(&inode->i_lock);
 		if (flags & EXPIRE_DIRTY_ATIME)
-			set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
+			inode->i_state |= I_DIRTY_TIME_EXPIRED;
+		inode->i_state |= I_SYNC_QUEUED;
+		spin_unlock(&inode->i_lock);

I wonder if the locking changes should go into a prep patch vs the actual logic changes related to I_SYNC_QUEUED? That would untangle the patch quite a bit and make it easier to follow.

Quoted 4 lines

#define I_WB_SWITCH (1 << 13)
#define I_OVL_INUSE (1 << 14)
#define I_CREATING (1 << 15)
+#define I_SYNC_QUEUED (1 << 16)

FYI, this conflicts with the I_DONTCAT addition in mainline now.

author maintainer
#4 lore raw

Re: [PATCH 1/3] writeback: Avoid skipping inode writeback

Quoted 8 lines · Christoph Hellwig

This generall looks ok, but a few nitpicks below:

fs/fs-writeback.c void sb_clear_inode_writeback(struct inode *inode) · line 1179

Quoted 2 lines
fs/fs-writeback.c +1 -1
-static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
+static void __redirty_tail(struct inode *inode, struct bdi_writeback *wb)

I think redirty_tail_locked would be a more decriptive name, and also fit other uses in this file (e.g. inode_io_list_move_locked and inode_io_list_del_locked).

Fair enough, will do.

Quoted 6 lines

fs/fs-writeback.c void sb_clear_inode_writeback(struct inode *inode) · line 1182

Quoted 3 lines

{

fs/fs-writeback.c +1 -0
+	assert_spin_locked(&inode->i_lock);
 	if (!list_empty(&wb->b_dirty)) {

Nit: I find an empty line after asserts and before the real code starts nice on the eye.

Sure.

Quoted 5 lines
Quoted 5 lines
fs/fs-writeback.c +1 -0
 			break;
 		list_move(&inode->i_io_list, &tmp);
 		moved++;
+		spin_lock(&inode->i_lock);
 		if (flags & EXPIRE_DIRTY_ATIME)
Quoted 8 lines

fs/fs-writeback.c static int move_expired_inodes(struct list_head *delaying_queue, · line 1270

Quoted 4 lines
fs/fs-writeback.c +3 -1
-			set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
+			inode->i_state |= I_DIRTY_TIME_EXPIRED;
+		inode->i_state |= I_SYNC_QUEUED;
+		spin_unlock(&inode->i_lock);

I wonder if the locking changes should go into a prep patch vs the actual logic changes related to I_SYNC_QUEUED? That would untangle the patch quite a bit and make it easier to follow.

OK, will do.

Quoted 6 lines
Quoted 4 lines

#define I_WB_SWITCH (1 << 13)
#define I_OVL_INUSE (1 << 14)
#define I_CREATING (1 << 15)
+#define I_SYNC_QUEUED (1 << 16)

FYI, this conflicts with the I_DONTCAT addition in mainline now.

Yup, I've already found out when rebasing...

Thanks for review!

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#5 lore raw

PATCH 2/3 writeback: Fix sync livelock due to b_dirty_time processing

In reply to Jan Kara

When we are processing writeback for sync(2), move_expired_inodes() didn't set any inode expiry value (older_than_this). This can result in writeback never completing if there's steady stream of inodes added to b_dirty_time list as writeback rechecks dirty lists after each writeback round whether there's more work to be done. Fix the problem by using sync(2) start time is inode expiry value when processing b_dirty_time list similarly as for ordinarily dirtied inodes. This requires some refactoring of older_than_this handling which simplifies the code noticeably as a bonus.

---
 fs/fs-writeback.c                | 37 ++++++++++++++-----------------------
 include/trace/events/writeback.h |  9 ++++-----
 2 files changed, 18 insertions(+), 28 deletions(-)
fs/fs-writeback.c +14 -23
index 855c6611710a..b9aa4ff83bbd 100644
@@ -42,7 +42,6 @@
 struct wb_writeback_work {
 	long nr_pages;
 	struct super_block *sb;
-	unsigned long *older_than_this;
 	enum writeback_sync_modes sync_mode;
 	unsigned int tagged_writepages:1;
 	unsigned int for_kupdate:1;
@@ -1233,16 +1232,14 @@ static bool inode_dirtied_after(struct inode *inode, unsigned long t)
 #define EXPIRE_DIRTY_ATIME 0x0001
 
 /*
- * Move expired (dirtied before work->older_than_this) dirty inodes from
+ * Move expired (dirtied before older_than_this) dirty inodes from
  * @delaying_queue to @dispatch_queue.
  */
 static int move_expired_inodes(struct list_head *delaying_queue,
 			       struct list_head *dispatch_queue,
 			       int flags,
-			       struct wb_writeback_work *work)
+			       unsigned long older_than_this)
 {
-	unsigned long *older_than_this = NULL;
-	unsigned long expire_time;
 	LIST_HEAD(tmp);
 	struct list_head *pos, *node;
 	struct super_block *sb = NULL;
@@ -1250,16 +1247,9 @@ static int move_expired_inodes(struct list_head *delaying_queue,
 	int do_sb_sort = 0;
 	int moved = 0;
 
-	if ((flags & EXPIRE_DIRTY_ATIME) == 0)
-		older_than_this = work->older_than_this;
-	else if (!work->for_sync) {
-		expire_time = jiffies - (dirtytime_expire_interval * HZ);
-		older_than_this = &expire_time;
-	}
 	while (!list_empty(delaying_queue)) {
 		inode = wb_inode(delaying_queue->prev);
-		if (older_than_this &&
-		    inode_dirtied_after(inode, *older_than_this))
+		if (inode_dirtied_after(inode, older_than_this))
 			break;
 		list_move(&inode->i_io_list, &tmp);
 		moved++;
@@ -1305,18 +1295,22 @@ static int move_expired_inodes(struct list_head *delaying_queue,
  *                                           |
  *                                           +--> dequeue for IO
  */
-static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
+static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work,
+		     unsigned long older_than_this)
 {
 	int moved;
+	unsigned long time_expire_jif = older_than_this;
 
 	assert_spin_locked(&wb->list_lock);
 	list_splice_init(&wb->b_more_io, &wb->b_io);
-	moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work);
+	moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, older_than_this);
+	if (!work->for_sync)
+		time_expire_jif = jiffies - dirtytime_expire_interval * HZ;
 	moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io,
-				     EXPIRE_DIRTY_ATIME, work);
+				     EXPIRE_DIRTY_ATIME, time_expire_jif);
 	if (moved)
 		wb_io_lists_populated(wb);
-	trace_writeback_queue_io(wb, work, moved);
+	trace_writeback_queue_io(wb, work, older_than_this, moved);
 }
 
 static int write_inode(struct inode *inode, struct writeback_control *wbc)
@@ -1829,7 +1823,7 @@ static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
 	blk_start_plug(&plug);
 	spin_lock(&wb->list_lock);
 	if (list_empty(&wb->b_io))
-		queue_io(wb, &work);
+		queue_io(wb, &work, jiffies);
 	__writeback_inodes_wb(wb, &work);
 	spin_unlock(&wb->list_lock);
 	blk_finish_plug(&plug);
@@ -1857,14 +1851,11 @@ static long wb_writeback(struct bdi_writeback *wb,
 {
 	unsigned long wb_start = jiffies;
 	long nr_pages = work->nr_pages;
-	unsigned long oldest_jif;
+	unsigned long oldest_jif = jiffies;
 	struct inode *inode;
 	long progress;
 	struct blk_plug plug;
 
-	oldest_jif = jiffies;
-	work->older_than_this = &oldest_jif;
-
 	blk_start_plug(&plug);
 	spin_lock(&wb->list_lock);
 	for (;;) {
@@ -1905,7 +1896,7 @@ static long wb_writeback(struct bdi_writeback *wb,
 
 		trace_writeback_start(wb, work);
 		if (list_empty(&wb->b_io))
-			queue_io(wb, work);
+			queue_io(wb, work, oldest_jif);
 		if (work->sb)
 			progress = writeback_sb_inodes(work->sb, wb, work);
 		else
include/trace/events/writeback.h +4 -5
index 85a33bea76f1..b4500bdee2c0 100644
@@ -498,8 +498,9 @@ DEFINE_WBC_EVENT(wbc_writepage);
 TRACE_EVENT(writeback_queue_io,
 	TP_PROTO(struct bdi_writeback *wb,
 		 struct wb_writeback_work *work,
+		 unsigned long older_than_this,
 		 int moved),
-	TP_ARGS(wb, work, moved),
+	TP_ARGS(wb, work, older_than_this, moved),
 	TP_STRUCT__entry(
 		__array(char,		name, 32)
 		__field(unsigned long,	older)
@@ -509,11 +510,9 @@ TRACE_EVENT(writeback_queue_io,
 		__field(ino_t,		cgroup_ino)
 	),
 	TP_fast_assign(
-		unsigned long *older_than_this = work->older_than_this;
 		strscpy_pad(__entry->name, bdi_dev_name(wb->bdi), 32);
-		__entry->older	= older_than_this ?  *older_than_this : 0;
-		__entry->age	= older_than_this ?
-				  (jiffies - *older_than_this) * 1000 / HZ : -1;
+		__entry->older	= older_than_this;
+		__entry->age	= (jiffies - older_than_this) * 1000 / HZ;
 		__entry->moved	= moved;
 		__entry->reason	= work->reason;
 		__entry->cgroup_ino	= __trace_wb_assign_cgroup(wb);
Signature · 1 line
2.16.4
#6 lore raw

Re: [PATCH 2/3] writeback: Fix sync livelock due to b_dirty_time processing

Quoted 9 lines · Jan Kara

When we are processing writeback for sync(2), move_expired_inodes() didn't set any inode expiry value (older_than_this). This can result in writeback never completing if there's steady stream of inodes added to b_dirty_time list as writeback rechecks dirty lists after each writeback round whether there's more work to be done. Fix the problem by using sync(2) start time is inode expiry value when processing b_dirty_time list similarly as for ordinarily dirtied inodes. This requires some refactoring of older_than_this handling which simplifies the code noticeably as a bonus.

Looks sane, but if you touch all the older_than_this users can we rename it to something more reasonable like oldest or oldest_jif?

author maintainer
#7 lore raw

Re: [PATCH 2/3] writeback: Fix sync livelock due to b_dirty_time processing

Quoted 11 lines · Christoph Hellwig
Quoted 9 lines · Jan Kara

When we are processing writeback for sync(2), move_expired_inodes() didn't set any inode expiry value (older_than_this). This can result in writeback never completing if there's steady stream of inodes added to b_dirty_time list as writeback rechecks dirty lists after each writeback round whether there's more work to be done. Fix the problem by using sync(2) start time is inode expiry value when processing b_dirty_time list similarly as for ordinarily dirtied inodes. This requires some refactoring of older_than_this handling which simplifies the code noticeably as a bonus.

Quoted 2 lines

Looks sane, but if you touch all the older_than_this users can we rename it to something more reasonable like oldest or oldest_jif?

OK, I can certainly rename this. I've just realized that 'oldest' is really misleading since we are in fact processing inodes that were dirtied before the given time. So maybe name that 'dirtied_before'?

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#8 lore raw

Re: [PATCH 2/3] writeback: Fix sync livelock due to b_dirty_time processing

Quoted 12 lines · Jan Kara
Quoted 11 lines · Christoph Hellwig
Quoted 9 lines · Jan Kara

When we are processing writeback for sync(2), move_expired_inodes() didn't set any inode expiry value (older_than_this). This can result in writeback never completing if there's steady stream of inodes added to b_dirty_time list as writeback rechecks dirty lists after each writeback round whether there's more work to be done. Fix the problem by using sync(2) start time is inode expiry value when processing b_dirty_time list similarly as for ordinarily dirtied inodes. This requires some refactoring of older_than_this handling which simplifies the code noticeably as a bonus.

Quoted 6 lines
Quoted 2 lines

Looks sane, but if you touch all the older_than_this users can we rename it to something more reasonable like oldest or oldest_jif?

OK, I can certainly rename this. I've just realized that 'oldest' is really misleading since we are in fact processing inodes that were dirtied before the given time. So maybe name that 'dirtied_before'?

Sounds good to me.

author maintainer
#9 lore raw

PATCH 3/3 writeback: Drop I_DIRTY_TIME_EXPIRE

In reply to Jan Kara

The only use of I_DIRTY_TIME_EXPIRE is to detect in __writeback_single_inode() that inode got there because flush worker decided it's time to writeback the dirty inode time stamps (either because we are syncing or because of age). However we can detect this directly in __writeback_single_inode() and there's no need for the strange propagation with I_DIRTY_TIME_EXPIRE flag.

---
 fs/ext4/inode.c                  |  2 +-
 fs/fs-writeback.c                | 15 +++++----------
 fs/xfs/libxfs/xfs_trans_inode.c  |  4 ++--
 include/linux/fs.h               |  1 -
 include/trace/events/writeback.h |  1 -
 5 files changed, 8 insertions(+), 15 deletions(-)
fs/ext4/inode.c +1 -1
index 2a4aae6acdcb..038cfe85f9cb 100644
@@ -4881,7 +4881,7 @@ static int other_inode_match(struct inode * inode, unsigned long ino,
 	    (inode->i_state & I_DIRTY_TIME)) {
 		struct ext4_inode_info	*ei = EXT4_I(inode);
 
-		inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
+		inode->i_state &= ~I_DIRTY_TIME;
 		spin_unlock(&inode->i_lock);
 
 		spin_lock(&ei->i_raw_lock);
fs/fs-writeback.c +5 -10
index b9aa4ff83bbd..62e2e1ed345a 100644
@@ -1237,7 +1237,6 @@ static bool inode_dirtied_after(struct inode *inode, unsigned long t)
  */
 static int move_expired_inodes(struct list_head *delaying_queue,
 			       struct list_head *dispatch_queue,
-			       int flags,
 			       unsigned long older_than_this)
 {
 	LIST_HEAD(tmp);
@@ -1254,8 +1253,6 @@ static int move_expired_inodes(struct list_head *delaying_queue,
 		list_move(&inode->i_io_list, &tmp);
 		moved++;
 		spin_lock(&inode->i_lock);
-		if (flags & EXPIRE_DIRTY_ATIME)
-			inode->i_state |= I_DIRTY_TIME_EXPIRED;
 		inode->i_state |= I_SYNC_QUEUED;
 		spin_unlock(&inode->i_lock);
 		if (sb_is_blkdev_sb(inode->i_sb))
@@ -1303,11 +1300,11 @@ static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work,
 
 	assert_spin_locked(&wb->list_lock);
 	list_splice_init(&wb->b_more_io, &wb->b_io);
-	moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, older_than_this);
+	moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
 	if (!work->for_sync)
 		time_expire_jif = jiffies - dirtytime_expire_interval * HZ;
 	moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io,
-				     EXPIRE_DIRTY_ATIME, time_expire_jif);
+				     time_expire_jif);
 	if (moved)
 		wb_io_lists_populated(wb);
 	trace_writeback_queue_io(wb, work, older_than_this, moved);
@@ -1485,16 +1482,14 @@ __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
 	dirty = inode->i_state & I_DIRTY;
 	if (inode->i_state & I_DIRTY_TIME) {
 		if ((dirty & I_DIRTY_INODE) ||
-		    wbc->sync_mode == WB_SYNC_ALL ||
-		    unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
+		    wbc->sync_mode == WB_SYNC_ALL || wbc->for_sync ||
 		    unlikely(time_after(jiffies,
 					(inode->dirtied_time_when +
 					 dirtytime_expire_interval * HZ)))) {
-			dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
+			dirty |= I_DIRTY_TIME;
 			trace_writeback_lazytime(inode);
 		}
-	} else
-		inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
+	}
 	inode->i_state &= ~dirty;
 
 	/*
fs/xfs/libxfs/xfs_trans_inode.c +2 -2
index 2b8ccb5b975d..3c903f2c498d 100644
@@ -96,9 +96,9 @@ xfs_trans_log_inode(
 	 * to log the timestamps, or will clear already cleared fields in the
 	 * worst case.
 	 */
-	if (inode->i_state & (I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED)) {
+	if (inode->i_state & I_DIRTY_TIME) {
 		spin_lock(&inode->i_lock);
-		inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
+		inode->i_state &= ~I_DIRTY_TIME;
 		spin_unlock(&inode->i_lock);
 	}
 
include/linux/fs.h +0 -1
index b02290d19edd..2c141d53949f 100644
@@ -2177,7 +2177,6 @@ static inline void kiocb_clone(struct kiocb *kiocb, struct kiocb *kiocb_src,
 #define I_DIO_WAKEUP		(1 << __I_DIO_WAKEUP)
 #define I_LINKABLE		(1 << 10)
 #define I_DIRTY_TIME		(1 << 11)
-#define I_DIRTY_TIME_EXPIRED	(1 << 12)
 #define I_WB_SWITCH		(1 << 13)
 #define I_OVL_INUSE		(1 << 14)
 #define I_CREATING		(1 << 15)
include/trace/events/writeback.h +0 -1
index b4500bdee2c0..a48e1f83fcdc 100644
@@ -20,7 +20,6 @@
 		{I_CLEAR,		"I_CLEAR"},		\
 		{I_SYNC,		"I_SYNC"},		\
 		{I_DIRTY_TIME,		"I_DIRTY_TIME"},	\
-		{I_DIRTY_TIME_EXPIRED,	"I_DIRTY_TIME_EXPIRED"}, \
 		{I_REFERENCED,		"I_REFERENCED"}		\
 	)
 
Signature · 1 line
2.16.4
Reviewed-by on v1 3/3
#10 lore raw

Re: [PATCH 3/3] writeback: Drop I_DIRTY_TIME_EXPIRE

Quoted 6 lines · Jan Kara

The only use of I_DIRTY_TIME_EXPIRE is to detect in __writeback_single_inode() that inode got there because flush worker decided it's time to writeback the dirty inode time stamps (either because we are syncing or because of age). However we can detect this directly in __writeback_single_inode() and there's no need for the strange propagation with I_DIRTY_TIME_EXPIRE flag.

Looks good:

One nit below:

Quoted 5 lines
fs/fs-writeback.c +1 -2
 	if (inode->i_state & I_DIRTY_TIME) {
 		if ((dirty & I_DIRTY_INODE) ||
-		    wbc->sync_mode == WB_SYNC_ALL ||
-		    unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
+		    wbc->sync_mode == WB_SYNC_ALL || wbc->for_sync ||

fs/fs-writeback.c __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) · line 1492

Quoted 10 lines
fs/fs-writeback.c +2 -3
 		    unlikely(time_after(jiffies,
 					(inode->dirtied_time_when +
 					 dirtytime_expire_interval * HZ)))) {
-			dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
+			dirty |= I_DIRTY_TIME;
 			trace_writeback_lazytime(inode);
 		}
-	} else
-		inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
+	}

We can also drop some indentation here. And remove the totally silly unlikely, something like:

	if ((inode->i_state & I_DIRTY_TIME) &&
	    ((dirty & I_DIRTY_INODE) ||
	     wbc->sync_mode == WB_SYNC_ALL || wbc->for_sync ||
	     time_after(jiffies, inode->dirtied_time_when +
			dirtytime_expire_interval * HZ)))) {
		dirty |= I_DIRTY_TIME;
		trace_writeback_lazytime(inode);
	}
author maintainer
#11 lore raw

Re: [PATCH 3/3] writeback: Drop I_DIRTY_TIME_EXPIRE

Quoted 33 lines · Christoph Hellwig
Quoted 6 lines · Jan Kara

The only use of I_DIRTY_TIME_EXPIRE is to detect in __writeback_single_inode() that inode got there because flush worker decided it's time to writeback the dirty inode time stamps (either because we are syncing or because of age). However we can detect this directly in __writeback_single_inode() and there's no need for the strange propagation with I_DIRTY_TIME_EXPIRE flag.

Looks good:

One nit below:

Quoted 5 lines
fs/fs-writeback.c +1 -2
 	if (inode->i_state & I_DIRTY_TIME) {
 		if ((dirty & I_DIRTY_INODE) ||
-		    wbc->sync_mode == WB_SYNC_ALL ||
-		    unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
+		    wbc->sync_mode == WB_SYNC_ALL || wbc->for_sync ||

fs/fs-writeback.c __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) · line 1492

Quoted 10 lines
fs/fs-writeback.c +2 -3
 		    unlikely(time_after(jiffies,
 					(inode->dirtied_time_when +
 					 dirtytime_expire_interval * HZ)))) {
-			dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
+			dirty |= I_DIRTY_TIME;
 			trace_writeback_lazytime(inode);
 		}
-	} else
-		inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
+	}

We can also drop some indentation here. And remove the totally silly unlikely, something like:

Quoted 8 lines
	if ((inode->i_state & I_DIRTY_TIME) &&
	    ((dirty & I_DIRTY_INODE) ||
	     wbc->sync_mode == WB_SYNC_ALL || wbc->for_sync ||
	     time_after(jiffies, inode->dirtied_time_when +
			dirtytime_expire_interval * HZ)))) {
		dirty |= I_DIRTY_TIME;
		trace_writeback_lazytime(inode);
	}

Sure, I've done this. Once fstests run passes, I'll send v2 (likely tomorrow).

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#12 lore raw
In reply to Jan Kara
Quoted 7 lines · Jan Kara

Hello,

this patch series fixes an issue with handling of lazy inode timestamp writeback which could result in missed background writeback of an inode or missed update of inode timestamps during sync(2). It also somewhat simplifies the writeback code handling of lazy inode timestamp updates. Review is welcome!

OK, nobody has replied to this and I have positive feedback from Martijn so I guess I'll pickup the patches to my tree so that they get some testing in linux-next and push them to Linus sometime next week.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR