Merge tag 'vfs-6.18-rc1.writeback' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs

Pull vfs writeback updates from Christian Brauner:
 "This contains work adressing lockups reported by users when a systemd
  unit reading lots of files from a filesystem mounted with the lazytime
  mount option exits.

  With the lazytime mount option enabled we can be switching many dirty
  inodes on cgroup exit to the parent cgroup. The numbers observed in
  practice when systemd slice of a large cron job exits can easily reach
  hundreds of thousands or millions.

  The logic in inode_do_switch_wbs() which sorts the inode into
  appropriate place in b_dirty list of the target wb however has linear
  complexity in the number of dirty inodes thus overall time complexity
  of switching all the inodes is quadratic leading to workers being
  pegged for hours consuming 100% of the CPU and switching inodes to the
  parent wb.

  Simple reproducer of the issue:

      FILES=10000
      # Filesystem mounted with lazytime mount option
      MNT=/mnt/
      echo "Creating files and switching timestamps"
      for (( j = 0; j < 50; j ++ )); do
          mkdir $MNT/dir$j
          for (( i = 0; i < $FILES; i++ )); do
              echo "foo" >$MNT/dir$j/file$i
          done
          touch -a -t 202501010000 $MNT/dir$j/file*
      done
      wait
      echo "Syncing and flushing"
      sync
      echo 3 >/proc/sys/vm/drop_caches

      echo "Reading all files from a cgroup"
      mkdir /sys/fs/cgroup/unified/mycg1 || exit
      echo $$ >/sys/fs/cgroup/unified/mycg1/cgroup.procs || exit
      for (( j = 0; j < 50; j ++ )); do
          cat /mnt/dir$j/file* >/dev/null &
      done
      wait
      echo "Switching wbs"
      # Now rmdir the cgroup after the script exits

  This can be solved by:

   - Avoiding contention on the wb->list_lock when switching inodes by
     running a single work item per wb and managing a queue of items
     switching to the wb

   - Allowing rescheduling when switching inodes over to a different
     cgroup to avoid softlockups

   - Maintaining the b_dirty list ordering instead of sorting it"

* tag 'vfs-6.18-rc1.writeback' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
  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

1 files changed, 86 insertions, 47 deletions

diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
index 815be2bbb1c4..2b35e80037fe 100644
--- a/fs/fs-writeback.c
+++ b/fs/fs-writeback.c
@@ -368,7 +368,8 @@ static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode)
 }
 
 struct inode_switch_wbs_context {
-	struct rcu_work		work;
+	/* List of queued switching contexts for the wb */
+	struct llist_node	list;
 
 	/*
 	 * Multiple inodes can be switched at once.  The switching procedure
@@ -378,7 +379,6 @@ struct inode_switch_wbs_context {
 	 * array embedded into struct inode_switch_wbs_context.  Otherwise
 	 * an inode could be left in a non-consistent state.
 	 */
-	struct bdi_writeback	*new_wb;
 	struct inode		*inodes[];
 };
 
@@ -445,22 +445,23 @@ static bool inode_do_switch_wbs(struct inode *inode,
 	 * Transfer to @new_wb's IO list if necessary.  If the @inode is dirty,
 	 * the specific list @inode was on is ignored and the @inode is put on
 	 * ->b_dirty which is always correct including from ->b_dirty_time.
-	 * The transfer preserves @inode->dirtied_when ordering.  If the @inode
-	 * was clean, it means it was on the b_attached list, so move it onto
-	 * the b_attached list of @new_wb.
+	 * If the @inode was clean, it means it was on the b_attached list, so
+	 * move it onto the b_attached list of @new_wb.
 	 */
 	if (!list_empty(&inode->i_io_list)) {
 		inode->i_wb = new_wb;
 
 		if (inode->i_state & I_DIRTY_ALL) {
-			struct inode *pos;
-
-			list_for_each_entry(pos, &new_wb->b_dirty, i_io_list)
-				if (time_after_eq(inode->dirtied_when,
-						  pos->dirtied_when))
-					break;
+			/*
+			 * We need to keep b_dirty list sorted by
+			 * dirtied_time_when. However properly sorting the
+			 * inode in the list gets too expensive when switching
+			 * many inodes. So just attach inode at the end of the
+			 * dirty list and clobber the dirtied_time_when.
+			 */
+			inode->dirtied_time_when = jiffies;
 			inode_io_list_move_locked(inode, new_wb,
-						  pos->i_io_list.prev);
+						  &new_wb->b_dirty);
 		} else {
 			inode_cgwb_move_to_attached(inode, new_wb);
 		}
@@ -486,13 +487,11 @@ skip_switch:
 	return switched;
 }
 
-static void inode_switch_wbs_work_fn(struct work_struct *work)
+static void process_inode_switch_wbs(struct bdi_writeback *new_wb,
+				     struct inode_switch_wbs_context *isw)
 {
-	struct inode_switch_wbs_context *isw =
-		container_of(to_rcu_work(work), struct inode_switch_wbs_context, work);
 	struct backing_dev_info *bdi = inode_to_bdi(isw->inodes[0]);
 	struct bdi_writeback *old_wb = isw->inodes[0]->i_wb;
-	struct bdi_writeback *new_wb = isw->new_wb;
 	unsigned long nr_switched = 0;
 	struct inode **inodep;
 
@@ -502,6 +501,7 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
 	 */
 	down_read(&bdi->wb_switch_rwsem);
 
+	inodep = isw->inodes;
 	/*
 	 * By the time control reaches here, RCU grace period has passed
 	 * since I_WB_SWITCH assertion and all wb stat update transactions
@@ -512,6 +512,7 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
 	 * gives us exclusion against all wb related operations on @inode
 	 * including IO list manipulations and stat updates.
 	 */
+relock:
 	if (old_wb < new_wb) {
 		spin_lock(&old_wb->list_lock);
 		spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING);
@@ -520,10 +521,17 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
 		spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING);
 	}
 
-	for (inodep = isw->inodes; *inodep; inodep++) {
+	while (*inodep) {
 		WARN_ON_ONCE((*inodep)->i_wb != old_wb);
 		if (inode_do_switch_wbs(*inodep, old_wb, new_wb))
 			nr_switched++;
+		inodep++;
+		if (*inodep && need_resched()) {
+			spin_unlock(&new_wb->list_lock);
+			spin_unlock(&old_wb->list_lock);
+			cond_resched();
+			goto relock;
+		}
 	}
 
 	spin_unlock(&new_wb->list_lock);
@@ -543,6 +551,38 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
 	atomic_dec(&isw_nr_in_flight);
 }
 
+void inode_switch_wbs_work_fn(struct work_struct *work)
+{
+	struct bdi_writeback *new_wb = container_of(work, struct bdi_writeback,
+						    switch_work);
+	struct inode_switch_wbs_context *isw, *next_isw;
+	struct llist_node *list;
+
+	/*
+	 * Grab out reference to wb so that it cannot get freed under us
+	 * after we process all the isw items.
+	 */
+	wb_get(new_wb);
+	while (1) {
+		list = llist_del_all(&new_wb->switch_wbs_ctxs);
+		/* Nothing to do? */
+		if (!list)
+			break;
+		/*
+		 * In addition to synchronizing among switchers, I_WB_SWITCH
+		 * tells the RCU protected stat update paths to grab the i_page
+		 * lock so that stat transfer can synchronize against them.
+		 * Let's continue after I_WB_SWITCH is guaranteed to be
+		 * visible.
+		 */
+		synchronize_rcu();
+
+		llist_for_each_entry_safe(isw, next_isw, list, list)
+			process_inode_switch_wbs(new_wb, isw);
+	}
+	wb_put(new_wb);
+}
+
 static bool inode_prepare_wbs_switch(struct inode *inode,
 				     struct bdi_writeback *new_wb)
 {
@@ -572,6 +612,13 @@ static bool inode_prepare_wbs_switch(struct inode *inode,
 	return true;
 }
 
+static void wb_queue_isw(struct bdi_writeback *wb,
+			 struct inode_switch_wbs_context *isw)
+{
+	if (llist_add(&isw->list, &wb->switch_wbs_ctxs))
+		queue_work(isw_wq, &wb->switch_work);
+}
+
 /**
  * inode_switch_wbs - change the wb association of an inode
  * @inode: target inode
@@ -585,6 +632,7 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
 	struct backing_dev_info *bdi = inode_to_bdi(inode);
 	struct cgroup_subsys_state *memcg_css;
 	struct inode_switch_wbs_context *isw;
+	struct bdi_writeback *new_wb = NULL;
 
 	/* noop if seems to be already in progress */
 	if (inode->i_state & I_WB_SWITCH)
@@ -609,40 +657,35 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
 	if (!memcg_css)
 		goto out_free;
 
-	isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
+	new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
 	css_put(memcg_css);
-	if (!isw->new_wb)
+	if (!new_wb)
 		goto out_free;
 
-	if (!inode_prepare_wbs_switch(inode, isw->new_wb))
+	if (!inode_prepare_wbs_switch(inode, new_wb))
 		goto out_free;
 
 	isw->inodes[0] = inode;
 
-	/*
-	 * In addition to synchronizing among switchers, I_WB_SWITCH tells
-	 * the RCU protected stat update paths to grab the i_page
-	 * lock so that stat transfer can synchronize against them.
-	 * Let's continue after I_WB_SWITCH is guaranteed to be visible.
-	 */
-	INIT_RCU_WORK(&isw->work, inode_switch_wbs_work_fn);
-	queue_rcu_work(isw_wq, &isw->work);
+	trace_inode_switch_wbs_queue(inode->i_wb, new_wb, 1);
+	wb_queue_isw(new_wb, isw);
 	return;
 
 out_free:
 	atomic_dec(&isw_nr_in_flight);
-	if (isw->new_wb)
-		wb_put(isw->new_wb);
+	if (new_wb)
+		wb_put(new_wb);
 	kfree(isw);
 }
 
-static bool isw_prepare_wbs_switch(struct inode_switch_wbs_context *isw,
+static bool isw_prepare_wbs_switch(struct bdi_writeback *new_wb,
+				   struct inode_switch_wbs_context *isw,
 				   struct list_head *list, int *nr)
 {
 	struct inode *inode;
 
 	list_for_each_entry(inode, list, i_io_list) {
-		if (!inode_prepare_wbs_switch(inode, isw->new_wb))
+		if (!inode_prepare_wbs_switch(inode, new_wb))
 			continue;
 
 		isw->inodes[*nr] = inode;
@@ -666,6 +709,7 @@ bool cleanup_offline_cgwb(struct bdi_writeback *wb)
 {
 	struct cgroup_subsys_state *memcg_css;
 	struct inode_switch_wbs_context *isw;
+	struct bdi_writeback *new_wb;
 	int nr;
 	bool restart = false;
 
@@ -678,12 +722,12 @@ bool cleanup_offline_cgwb(struct bdi_writeback *wb)
 
 	for (memcg_css = wb->memcg_css->parent; memcg_css;
 	     memcg_css = memcg_css->parent) {
-		isw->new_wb = wb_get_create(wb->bdi, memcg_css, GFP_KERNEL);
-		if (isw->new_wb)
+		new_wb = wb_get_create(wb->bdi, memcg_css, GFP_KERNEL);
+		if (new_wb)
 			break;
 	}
-	if (unlikely(!isw->new_wb))
-		isw->new_wb = &wb->bdi->wb; /* wb_get() is noop for bdi's wb */
+	if (unlikely(!new_wb))
+		new_wb = &wb->bdi->wb; /* wb_get() is noop for bdi's wb */
 
 	nr = 0;
 	spin_lock(&wb->list_lock);
@@ -695,27 +739,22 @@ bool cleanup_offline_cgwb(struct bdi_writeback *wb)
 	 * bandwidth restrictions, as writeback of inode metadata is not
 	 * accounted for.
 	 */
-	restart = isw_prepare_wbs_switch(isw, &wb->b_attached, &nr);
+	restart = isw_prepare_wbs_switch(new_wb, isw, &wb->b_attached, &nr);
 	if (!restart)
-		restart = isw_prepare_wbs_switch(isw, &wb->b_dirty_time, &nr);
+		restart = isw_prepare_wbs_switch(new_wb, isw, &wb->b_dirty_time,
+						 &nr);
 	spin_unlock(&wb->list_lock);
 
 	/* no attached inodes? bail out */
 	if (nr == 0) {
 		atomic_dec(&isw_nr_in_flight);
-		wb_put(isw->new_wb);
+		wb_put(new_wb);
 		kfree(isw);
 		return restart;
 	}
 
-	/*
-	 * In addition to synchronizing among switchers, I_WB_SWITCH tells
-	 * the RCU protected stat update paths to grab the i_page
-	 * lock so that stat transfer can synchronize against them.
-	 * Let's continue after I_WB_SWITCH is guaranteed to be visible.
-	 */
-	INIT_RCU_WORK(&isw->work, inode_switch_wbs_work_fn);
-	queue_rcu_work(isw_wq, &isw->work);
+	trace_inode_switch_wbs_queue(wb, new_wb, nr);
+	wb_queue_isw(new_wb, isw);
 
 	return restart;
 }