Merge patch series "writeback: Avoid lockups when switching inodes"

Jan Kara <jack@suse.cz> says:

This patch series addresses lockups reported by users when systemd unit reading
lots of files from a filesystem mounted with lazytime mount option exits. See
patch 3 for more details about the reproducer.

There are two main issues why switching many inodes between wbs:

1) Multiple workers will be spawned to do the switching but they all contend
on the same wb->list_lock making all the parallelism pointless and just
wasting time.

2) Sorting of wb->b_dirty list by dirtied_time_when is inherently slow.

Patches 1-3 address these problems, patch 4 adds a tracepoint for better
observability of inode writeback switching.

* patches from https://lore.kernel.org/20250912103522.2935-1-jack@suse.cz:
  writeback: Add tracepoint to track pending inode switches
  writeback: Avoid excessively long inode switching times
  writeback: Avoid softlockup when switching many inodes
  writeback: Avoid contention on wb->list_lock when switching inodes

Signed-off-by: Christian Brauner <brauner@kernel.org>

3 files changed, 35 insertions, 0 deletions

diff --git a/include/linux/backing-dev-defs.h b/include/linux/backing-dev-defs.h
index 2ad261082bba..c5c9d89c73ed 100644
--- a/include/linux/backing-dev-defs.h
+++ b/include/linux/backing-dev-defs.h
@@ -152,6 +152,10 @@ struct bdi_writeback {
 	struct list_head blkcg_node;	/* anchored at blkcg->cgwb_list */
 	struct list_head b_attached;	/* attached inodes, protected by list_lock */
 	struct list_head offline_node;	/* anchored at offline_cgwbs */
+	struct work_struct switch_work;	/* work used to perform inode switching
+					 * to this wb */
+	struct llist_head switch_wbs_ctxs;	/* queued contexts for
+						 * writeback switching */
 
 	union {
 		struct work_struct release_work;
diff --git a/include/linux/writeback.h b/include/linux/writeback.h
index a2848d731a46..15a4bc4ab819 100644
--- a/include/linux/writeback.h
+++ b/include/linux/writeback.h
@@ -265,6 +265,8 @@ static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
 		bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
 }
 
+void inode_switch_wbs_work_fn(struct work_struct *work);
+
 #else	/* CONFIG_CGROUP_WRITEBACK */
 
 static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
diff --git a/include/trace/events/writeback.h b/include/trace/events/writeback.h
index 1e23919c0da9..c08aff044e80 100644
--- a/include/trace/events/writeback.h
+++ b/include/trace/events/writeback.h
@@ -213,6 +213,35 @@ TRACE_EVENT(inode_foreign_history,
 	)
 );
 
+TRACE_EVENT(inode_switch_wbs_queue,
+
+	TP_PROTO(struct bdi_writeback *old_wb, struct bdi_writeback *new_wb,
+		 unsigned int count),
+
+	TP_ARGS(old_wb, new_wb, count),
+
+	TP_STRUCT__entry(
+		__array(char,		name, 32)
+		__field(ino_t,		old_cgroup_ino)
+		__field(ino_t,		new_cgroup_ino)
+		__field(unsigned int,	count)
+	),
+
+	TP_fast_assign(
+		strscpy_pad(__entry->name, bdi_dev_name(old_wb->bdi), 32);
+		__entry->old_cgroup_ino	= __trace_wb_assign_cgroup(old_wb);
+		__entry->new_cgroup_ino	= __trace_wb_assign_cgroup(new_wb);
+		__entry->count		= count;
+	),
+
+	TP_printk("bdi %s: old_cgroup_ino=%lu new_cgroup_ino=%lu count=%u",
+		__entry->name,
+		(unsigned long)__entry->old_cgroup_ino,
+		(unsigned long)__entry->new_cgroup_ino,
+		__entry->count
+	)
+);
+
 TRACE_EVENT(inode_switch_wbs,
 
 	TP_PROTO(struct inode *inode, struct bdi_writeback *old_wb,