bcachefs: Go RW before bch2_check_lrus()

btree updates before going RW are expensive if they're in random order,
since they use the list of keys for journal replay to insert, which is
just a gap buffer.

This patch improves the bucket invalidate path so that if
bch2_check_lrus() hasn't finished it only prints warnings instead of
doing an emergency shutdown, which means we can now set BCH_FS_MAY_GO_RW
before bch2_check_lrus().

Also, the filesystem state bits are reorganized a bit.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

4 files changed, 90 insertions, 55 deletions

diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index d9cf676da030..eb03b4135c3d 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -382,7 +382,8 @@ int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k,
 				return -EINVAL;
 			}
 
-			if (!a.v->io_time[READ]) {
+			if (!a.v->io_time[READ] &&
+			    test_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags)) {
 				pr_buf(err, "cached bucket with read_time == 0");
 				return -EINVAL;
 			}
@@ -588,7 +589,6 @@ int bch2_trans_mark_alloc(struct btree_trans *trans,
 	    !new_a->io_time[READ])
 		new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 
-
 	old_lru = alloc_lru_idx(old_a);
 	new_lru = alloc_lru_idx(*new_a);
 
@@ -1088,6 +1088,7 @@ static int invalidate_one_bucket(struct btree_trans *trans, struct bch_dev *ca)
 
 	bch2_trans_iter_init(trans, &lru_iter, BTREE_ID_lru,
 			     POS(ca->dev_idx, 0), 0);
+next_lru:
 	k = bch2_btree_iter_peek(&lru_iter);
 	ret = bkey_err(k);
 	if (ret)
@@ -1096,9 +1097,20 @@ static int invalidate_one_bucket(struct btree_trans *trans, struct bch_dev *ca)
 	if (!k.k || k.k->p.inode != ca->dev_idx)
 		goto out;
 
-	if (bch2_trans_inconsistent_on(k.k->type != KEY_TYPE_lru, trans,
-				       "non lru key in lru btree"))
-		goto out;
+	if (k.k->type != KEY_TYPE_lru) {
+		pr_buf(&buf, "non lru key in lru btree:\n  ");
+		bch2_bkey_val_to_text(&buf, c, k);
+
+		if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) {
+			bch_err(c, "%s", buf.buf);
+			bch2_btree_iter_advance(&lru_iter);
+			goto next_lru;
+		} else {
+			bch2_trans_inconsistent(trans, "%s", buf.buf);
+			ret = -EINVAL;
+			goto out;
+		}
+	}
 
 	idx	= k.k->p.offset;
 	bucket	= le64_to_cpu(bkey_s_c_to_lru(k).v->idx);
@@ -1111,13 +1123,19 @@ static int invalidate_one_bucket(struct btree_trans *trans, struct bch_dev *ca)
 
 	if (idx != alloc_lru_idx(a->v)) {
 		pr_buf(&buf, "alloc key does not point back to lru entry when invalidating bucket:\n  ");
-
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
 		pr_buf(&buf, "\n  ");
 		bch2_bkey_val_to_text(&buf, c, k);
-		bch2_trans_inconsistent(trans, "%s", buf.buf);
-		ret = -EINVAL;
-		goto out;
+
+		if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) {
+			bch_err(c, "%s", buf.buf);
+			bch2_btree_iter_advance(&lru_iter);
+			goto next_lru;
+		} else {
+			bch2_trans_inconsistent(trans, "%s", buf.buf);
+			ret = -EINVAL;
+			goto out;
+		}
 	}
 
 	SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index 5dda57afa802..127323b677df 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -494,11 +494,6 @@ struct bch_dev {
 
 enum {
 	/* startup: */
-	BCH_FS_CLEAN_SHUTDOWN,
-	BCH_FS_INITIAL_GC_DONE,
-	BCH_FS_INITIAL_GC_UNFIXED,
-	BCH_FS_TOPOLOGY_REPAIR_DONE,
-	BCH_FS_FSCK_DONE,
 	BCH_FS_STARTED,
 	BCH_FS_MAY_GO_RW,
 	BCH_FS_RW,
@@ -508,16 +503,22 @@ enum {
 	BCH_FS_STOPPING,
 	BCH_FS_EMERGENCY_RO,
 	BCH_FS_WRITE_DISABLE_COMPLETE,
+	BCH_FS_CLEAN_SHUTDOWN,
+
+	/* fsck passes: */
+	BCH_FS_TOPOLOGY_REPAIR_DONE,
+	BCH_FS_INITIAL_GC_DONE,		/* kill when we enumerate fsck passes */
+	BCH_FS_CHECK_LRUS_DONE,
+	BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE,
+	BCH_FS_FSCK_DONE,
+	BCH_FS_INITIAL_GC_UNFIXED,	/* kill when we enumerate fsck errors */
+	BCH_FS_NEED_ANOTHER_GC,
 
 	/* errors: */
 	BCH_FS_ERROR,
 	BCH_FS_TOPOLOGY_ERROR,
 	BCH_FS_ERRORS_FIXED,
 	BCH_FS_ERRORS_NOT_FIXED,
-
-	/* misc: */
-	BCH_FS_NEED_ANOTHER_GC,
-	BCH_FS_DELETED_NODES,
 };
 
 struct btree_debug {
diff --git a/fs/bcachefs/lru.c b/fs/bcachefs/lru.c
index fe9d15742947..ce23b38382f5 100644
--- a/fs/bcachefs/lru.c
+++ b/fs/bcachefs/lru.c
@@ -204,7 +204,9 @@ int bch2_check_lrus(struct bch_fs *c, bool initial)
 
 	for_each_btree_key(&trans, iter, BTREE_ID_lru, POS_MIN,
 			   BTREE_ITER_PREFETCH, k, ret) {
-		ret = __bch2_trans_do(&trans, NULL, NULL, 0,
+		ret = __bch2_trans_do(&trans, NULL, NULL,
+				      BTREE_INSERT_NOFAIL|
+				      BTREE_INSERT_LAZY_RW,
 			bch2_check_lru_key(&trans, &iter, initial));
 		if (ret)
 			break;
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index e2474ff99702..5831ab53a982 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -994,7 +994,6 @@ static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 	if (ret)
 		return ret;
 
-
 	bkey_subvolume_init(&root_volume.k_i);
 	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 	root_volume.v.flags	= 0;
@@ -1096,6 +1095,12 @@ int bch2_fs_recovery(struct bch_fs *c)
 		}
 	}
 
+	if (c->opts.fsck && c->opts.norecovery) {
+		bch_err(c, "cannot select both norecovery and fsck");
+		ret = -EINVAL;
+		goto err;
+	}
+
 	ret = bch2_blacklist_table_initialize(c);
 	if (ret) {
 		bch_err(c, "error initializing blacklist table");
@@ -1189,6 +1194,13 @@ use_clean:
 	if (ret)
 		goto err;
 
+	/*
+	 * Skip past versions that might have possibly been used (as nonces),
+	 * but hadn't had their pointers written:
+	 */
+	if (c->sb.encryption_type && !c->sb.clean)
+		atomic64_add(1 << 16, &c->key_version);
+
 	ret = read_btree_roots(c);
 	if (ret)
 		goto err;
@@ -1211,12 +1223,7 @@ use_clean:
 		goto err;
 	bch_verbose(c, "stripes_read done");
 
-	/*
-	 * If we're not running fsck, this ensures bch2_fsck_err() calls are
-	 * instead interpreted as bch2_inconsistent_err() calls:
-	 */
-	if (!c->opts.fsck)
-		set_bit(BCH_FS_FSCK_DONE, &c->flags);
+	bch2_stripes_heap_start(c);
 
 	if (c->opts.fsck) {
 		bool metadata_only = c->opts.norecovery;
@@ -1228,6 +1235,8 @@ use_clean:
 			goto err;
 		bch_verbose(c, "done checking allocations");
 
+		set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
+
 		bch_info(c, "checking need_discard and freespace btrees");
 		err = "error checking need_discard and freespace btrees";
 		ret = bch2_check_alloc_info(c);
@@ -1235,55 +1244,60 @@ use_clean:
 			goto err;
 		bch_verbose(c, "done checking need_discard and freespace btrees");
 
+		set_bit(BCH_FS_MAY_GO_RW, &c->flags);
+
+		bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
+		err = "journal replay failed";
+		ret = bch2_journal_replay(c);
+		if (ret)
+			goto err;
+		if (c->opts.verbose || !c->sb.clean)
+			bch_info(c, "journal replay done");
+
 		bch_info(c, "checking lrus");
 		err = "error checking lrus";
 		ret = bch2_check_lrus(c, true);
 		if (ret)
 			goto err;
 		bch_verbose(c, "done checking lrus");
-	}
 
-	bch2_stripes_heap_start(c);
+		set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
 
-	set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
-	set_bit(BCH_FS_MAY_GO_RW, &c->flags);
-
-	/*
-	 * Skip past versions that might have possibly been used (as nonces),
-	 * but hadn't had their pointers written:
-	 */
-	if (c->sb.encryption_type && !c->sb.clean)
-		atomic64_add(1 << 16, &c->key_version);
-
-	if (c->opts.norecovery)
-		goto out;
-
-	bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
-	err = "journal replay failed";
-	ret = bch2_journal_replay(c);
-	if (ret)
-		goto err;
-	if (c->opts.verbose || !c->sb.clean)
-		bch_info(c, "journal replay done");
-
-	err = "error initializing freespace";
-	ret = bch2_fs_freespace_init(c);
-	if (ret)
-		goto err;
-
-	if (c->opts.fsck) {
 		bch_info(c, "checking alloc to lru refs");
 		err = "error checking alloc to lru refs";
 		ret = bch2_check_alloc_to_lru_refs(c);
 		if (ret)
 			goto err;
+		set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
 
 		ret = bch2_check_lrus(c, true);
 		if (ret)
 			goto err;
 		bch_verbose(c, "done checking alloc to lru refs");
+	} else {
+		set_bit(BCH_FS_MAY_GO_RW, &c->flags);
+		set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
+		set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
+		set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
+		set_bit(BCH_FS_FSCK_DONE, &c->flags);
+
+		if (c->opts.norecovery)
+			goto out;
+
+		bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
+		err = "journal replay failed";
+		ret = bch2_journal_replay(c);
+		if (ret)
+			goto err;
+		if (c->opts.verbose || !c->sb.clean)
+			bch_info(c, "journal replay done");
 	}
 
+	err = "error initializing freespace";
+	ret = bch2_fs_freespace_init(c);
+	if (ret)
+		goto err;
+
 	if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
 		bch2_fs_lazy_rw(c);