bcachefs: Go RW before check_alloc_info()

It's possible to do btree updates before going RW by adding them to the
list of updates for journal replay to do, but this is limited by what
fits in RAM. This patch switches the second alloc info phase to run
after going RW - btree_gc has already ensured the alloc btree itself is
correct - and tweaks the allocation path to deal with the potential
small inconsistencies.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/alloc_background.c

@@ -583,6 +583,7 @@ static int bch2_bucket_do_index(struct btree_trans *trans,
 		goto err;
 
 	if (ca->mi.freespace_initialized &&
+	    test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags) &&
 	    bch2_trans_inconsistent_on(old.k->type != old_type, trans,
 			"incorrect key when %s %s btree (got %s should be %s)\n"
 			"  for %s",
@@ -1028,21 +1029,28 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 		goto write;
 	}
 
-	if (bch2_trans_inconsistent_on(a->v.journal_seq > c->journal.flushed_seq_ondisk, trans,
+	if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
+		if (test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags)) {
+			bch2_trans_inconsistent(trans,
 				"clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
 				"%s",
 				a->v.journal_seq,
 				c->journal.flushed_seq_ondisk,
-			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 			ret = -EIO;
+		}
 		goto out;
 	}
 
-	if (bch2_trans_inconsistent_on(a->v.data_type != BCH_DATA_need_discard, trans,
+	if (a->v.data_type != BCH_DATA_need_discard) {
+		if (test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags)) {
+			bch2_trans_inconsistent(trans,
 				"bucket incorrectly set in need_discard btree\n"
 				"%s",
-			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 			ret = -EIO;
+		}
+
 		goto out;
 	}
 

fs/bcachefs/alloc_foreground.c

@@ -316,28 +316,34 @@ static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bc
 
 	a = bch2_alloc_to_v4(k, &a_convert);
 
-	if (genbits != (alloc_freespace_genbits(*a) >> 56)) {
-		prt_printf(&buf, "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
-		       "  freespace key ",
-		       genbits, alloc_freespace_genbits(*a) >> 56);
+	if (a->data_type != BCH_DATA_free) {
+		if (!test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags)) {
+			ob = NULL;
+			goto err;
+		}
+
+		prt_printf(&buf, "non free bucket in freespace btree\n"
+		       "  freespace key ");
 		bch2_bkey_val_to_text(&buf, c, freespace_k);
 		prt_printf(&buf, "\n  ");
 		bch2_bkey_val_to_text(&buf, c, k);
 		bch2_trans_inconsistent(trans, "%s", buf.buf);
 		ob = ERR_PTR(-EIO);
 		goto err;
-
 	}
 
-	if (a->data_type != BCH_DATA_free) {
-		prt_printf(&buf, "non free bucket in freespace btree\n"
-		       "  freespace key ");
+	if (genbits != (alloc_freespace_genbits(*a) >> 56) &&
+	    test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags)) {
+		prt_printf(&buf, "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
+		       "  freespace key ",
+		       genbits, alloc_freespace_genbits(*a) >> 56);
 		bch2_bkey_val_to_text(&buf, c, freespace_k);
 		prt_printf(&buf, "\n  ");
 		bch2_bkey_val_to_text(&buf, c, k);
 		bch2_trans_inconsistent(trans, "%s", buf.buf);
 		ob = ERR_PTR(-EIO);
 		goto err;
+
 	}
 
 	ob = __try_alloc_bucket(c, ca, b, reserve, a, s, cl);
@@ -505,6 +511,7 @@ static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
 {
 	struct bch_fs *c = trans->c;
 	struct open_bucket *ob = NULL;
+	bool freespace = READ_ONCE(ca->mi.freespace_initialized);
 	u64 avail;
 	struct bucket_alloc_state s = { 0 };
 	bool waiting = false;
@@ -543,13 +550,18 @@ static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
 		if (ob)
 			return ob;
 	}
-
-	ob = likely(ca->mi.freespace_initialized)
+alloc:
+	ob = likely(freespace)
 		? bch2_bucket_alloc_freelist(trans, ca, reserve, &s, cl)
 		: bch2_bucket_alloc_early(trans, ca, reserve, &s, cl);
 
 	if (s.skipped_need_journal_commit * 2 > avail)
 		bch2_journal_flush_async(&c->journal, NULL);
+
+	if (!ob && freespace && !test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags)) {
+		freespace = false;
+		goto alloc;
+	}
 err:
 	if (!ob)
 		ob = ERR_PTR(-BCH_ERR_no_buckets_found);

fs/bcachefs/bcachefs.h

@@ -549,6 +549,7 @@ enum {
 	/* fsck passes: */
 	BCH_FS_TOPOLOGY_REPAIR_DONE,
 	BCH_FS_INITIAL_GC_DONE,		/* kill when we enumerate fsck passes */
+	BCH_FS_CHECK_ALLOC_DONE,
 	BCH_FS_CHECK_LRUS_DONE,
 	BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE,
 	BCH_FS_FSCK_DONE,

fs/bcachefs/recovery.c

@@ -1260,13 +1260,6 @@ int bch2_fs_recovery(struct bch_fs *c)
 
 		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);
-		if (ret)
-			goto err;
-		bch_verbose(c, "done checking need_discard and freespace btrees");
-
 		if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
 			err = "error creating root snapshot node";
 			ret = bch2_fs_initialize_subvolumes(c);
@@ -1291,6 +1284,15 @@ int bch2_fs_recovery(struct bch_fs *c)
 		if (c->opts.verbose || !c->sb.clean)
 			bch_info(c, "journal replay done");
 
+		bch_info(c, "checking need_discard and freespace btrees");
+		err = "error checking need_discard and freespace btrees";
+		ret = bch2_check_alloc_info(c);
+		if (ret)
+			goto err;
+		bch_verbose(c, "done checking need_discard and freespace btrees");
+
+		set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
+
 		bch_info(c, "checking lrus");
 		err = "error checking lrus";
 		ret = bch2_check_lrus(c);
@@ -1308,6 +1310,7 @@ int bch2_fs_recovery(struct bch_fs *c)
 		set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
 	} else {
 		set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
+		set_bit(BCH_FS_CHECK_ALLOC_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);