bcachefs: Make sure __bch2_run_explicit_recovery_pass() signals to rewind

We should always signal to rewind if the requested pass hasn't been run,
even if called multiple times.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

2 files changed, 14 insertions, 13 deletions

diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index b12c9c78beec..e6cd93e1ed0f 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -1044,6 +1044,7 @@ struct bch_fs {
 	 * for signaling to the toplevel code which pass we want to run now.
 	 */
 	enum bch_recovery_pass	curr_recovery_pass;
+	enum bch_recovery_pass	next_recovery_pass;
 	/* bitmask of recovery passes that we actually ran */
 	u64			recovery_passes_complete;
 	/* never rewinds version of curr_recovery_pass */
diff --git a/fs/bcachefs/recovery_passes.c b/fs/bcachefs/recovery_passes.c
index f6d3a99cb63e..34d4299c2365 100644
--- a/fs/bcachefs/recovery_passes.c
+++ b/fs/bcachefs/recovery_passes.c
@@ -103,12 +103,12 @@ u64 bch2_recovery_passes_from_stable(u64 v)
 static int __bch2_run_explicit_recovery_pass(struct bch_fs *c,
 					     enum bch_recovery_pass pass)
 {
-	if (c->opts.recovery_passes & BIT_ULL(pass))
-		return 0;
-
 	if (c->curr_recovery_pass == ARRAY_SIZE(recovery_pass_fns))
 		return -BCH_ERR_not_in_recovery;
 
+	if (c->recovery_passes_complete & BIT_ULL(pass))
+		return 0;
+
 	if (pass < BCH_RECOVERY_PASS_set_may_go_rw &&
 	    c->curr_recovery_pass >= BCH_RECOVERY_PASS_set_may_go_rw) {
 		bch_info(c, "need recovery pass %s (%u), but already rw",
@@ -122,8 +122,8 @@ static int __bch2_run_explicit_recovery_pass(struct bch_fs *c,
 
 	c->opts.recovery_passes |= BIT_ULL(pass);
 
-	if (c->curr_recovery_pass >= pass) {
-		c->curr_recovery_pass = pass;
+	if (c->curr_recovery_pass > pass) {
+		c->next_recovery_pass = pass;
 		c->recovery_passes_complete &= (1ULL << pass) >> 1;
 		return -BCH_ERR_restart_recovery;
 	} else {
@@ -265,6 +265,8 @@ int bch2_run_recovery_passes(struct bch_fs *c)
 	c->opts.recovery_passes_exclude &= ~BCH_RECOVERY_PASS_set_may_go_rw;
 
 	while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
+		c->next_recovery_pass = c->curr_recovery_pass + 1;
+
 		spin_lock_irq(&c->recovery_pass_lock);
 		unsigned pass = c->curr_recovery_pass;
 
@@ -286,27 +288,25 @@ int bch2_run_recovery_passes(struct bch_fs *c)
 			bch2_journal_flush(&c->journal);
 
 		spin_lock_irq(&c->recovery_pass_lock);
-		if (c->curr_recovery_pass < pass) {
+		if (c->next_recovery_pass < c->curr_recovery_pass) {
 			/*
 			 * bch2_run_explicit_recovery_pass() was called: we
 			 * can't always catch -BCH_ERR_restart_recovery because
 			 * it may have been called from another thread (btree
 			 * node read completion)
 			 */
-			spin_unlock_irq(&c->recovery_pass_lock);
-			continue;
-		} else if (c->curr_recovery_pass == pass) {
-			c->curr_recovery_pass++;
+			ret = 0;
+			c->recovery_passes_complete &= ~(~0ULL << c->curr_recovery_pass);
 		} else {
-			BUG();
+			c->recovery_passes_complete |= BIT_ULL(pass);
+			c->recovery_pass_done = max(c->recovery_pass_done, pass);
 		}
+		c->curr_recovery_pass = c->next_recovery_pass;
 		spin_unlock_irq(&c->recovery_pass_lock);
 
 		if (ret)
 			break;
 
-		c->recovery_passes_complete |= BIT_ULL(pass);
-		c->recovery_pass_done = max(c->recovery_pass_done, pass);
 
 		if (!test_bit(BCH_FS_error, &c->flags))
 			bch2_clear_recovery_pass_required(c, pass);