Update bcachefs sources

11 files changed, 125 insertions, 66 deletions

diff --git a/.bcachefs_revision b/.bcachefs_revision
index 398cb4ac..f0ec32ef 100644
--- a/.bcachefs_revision
+++ b/.bcachefs_revision
@@ -1 +1 @@
-217509029551f8b40024d7fcd6e311be1fb6410c
+d37e2bb6dca5e4decf6b5918737c719fa6f17cc6
diff --git a/libbcachefs/btree_gc.c b/libbcachefs/btree_gc.c
index 0883b9b4..3620c29e 100644
--- a/libbcachefs/btree_gc.c
+++ b/libbcachefs/btree_gc.c
@@ -98,7 +98,7 @@ u8 bch2_btree_key_recalc_oldest_gen(struct bch_fs *c, struct bkey_s_c k)
 			struct bch_dev *ca = c->devs[ptr->dev];
 			size_t b = PTR_BUCKET_NR(ca, ptr);
 
-			if (__gen_after(ca->oldest_gens[b], ptr->gen))
+			if (gen_after(ca->oldest_gens[b], ptr->gen))
 				ca->oldest_gens[b] = ptr->gen;
 
 			max_stale = max(max_stale, ptr_stale(ca, ptr));
@@ -126,14 +126,40 @@ static u8 bch2_btree_mark_key(struct bch_fs *c, enum bkey_type type,
 	}
 }
 
-u8 bch2_btree_mark_key_initial(struct bch_fs *c, enum bkey_type type,
-			       struct bkey_s_c k)
+int bch2_btree_mark_key_initial(struct bch_fs *c, enum bkey_type type,
+				struct bkey_s_c k)
 {
+	int ret;
+
+	switch (k.k->type) {
+	case BCH_EXTENT:
+	case BCH_EXTENT_CACHED: {
+		struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
+		const struct bch_extent_ptr *ptr;
+
+		extent_for_each_ptr(e, ptr) {
+			struct bch_dev *ca = c->devs[ptr->dev];
+			struct bucket *g = PTR_BUCKET(ca, ptr);
+
+			unfixable_fsck_err_on(gen_cmp(ptr->gen, g->mark.gen) > 0, c,
+					      "%s ptr gen in the future: %u > %u",
+					      type == BKEY_TYPE_BTREE
+					      ? "btree" : "data",
+					      ptr->gen, g->mark.gen);
+
+		}
+		break;
+	}
+	}
+
 	atomic64_set(&c->key_version,
 		     max_t(u64, k.k->version.lo,
 			   atomic64_read(&c->key_version)));
 
-	return bch2_btree_mark_key(c, type, k);
+	bch2_btree_mark_key(c, type, k);
+	return 0;
+fsck_err:
+	return ret;
 }
 
 static bool btree_gc_mark_node(struct bch_fs *c, struct btree *b)
@@ -890,16 +916,22 @@ int bch2_gc_thread_start(struct bch_fs *c)
 
 /* Initial GC computes bucket marks during startup */
 
-static void bch2_initial_gc_btree(struct bch_fs *c, enum btree_id id)
+static int bch2_initial_gc_btree(struct bch_fs *c, enum btree_id id)
 {
 	struct btree_iter iter;
 	struct btree *b;
 	struct range_checks r;
+	int ret = 0;
 
 	btree_node_range_checks_init(&r, 0);
 
 	if (!c->btree_roots[id].b)
-		return;
+		return 0;
+
+	ret = bch2_btree_mark_key_initial(c, BKEY_TYPE_BTREE,
+			   bkey_i_to_s_c(&c->btree_roots[id].b->key));
+	if (ret)
+		return ret;
 
 	/*
 	 * We have to hit every btree node before starting journal replay, in
@@ -915,28 +947,37 @@ static void bch2_initial_gc_btree(struct bch_fs *c, enum btree_id id)
 
 			for_each_btree_node_key_unpack(b, k, &node_iter,
 						       btree_node_is_extents(b),
-						       &unpacked)
-				bch2_btree_mark_key_initial(c, btree_node_type(b), k);
+						       &unpacked) {
+				ret = bch2_btree_mark_key_initial(c,
+							btree_node_type(b), k);
+				if (ret)
+					goto err;
+			}
 		}
 
 		bch2_btree_iter_cond_resched(&iter);
 	}
-
+err:
 	bch2_btree_iter_unlock(&iter);
-
-	bch2_btree_mark_key(c, BKEY_TYPE_BTREE,
-			   bkey_i_to_s_c(&c->btree_roots[id].b->key));
+	return ret;
 }
 
 int bch2_initial_gc(struct bch_fs *c, struct list_head *journal)
 {
 	enum btree_id id;
+	int ret;
 
-	for (id = 0; id < BTREE_ID_NR; id++)
-		bch2_initial_gc_btree(c, id);
+	for (id = 0; id < BTREE_ID_NR; id++) {
+		ret = bch2_initial_gc_btree(c, id);
+		if (ret)
+			return ret;
+	}
 
-	if (journal)
-		bch2_journal_mark(c, journal);
+	if (journal) {
+		ret = bch2_journal_mark(c, journal);
+		if (ret)
+			return ret;
+	}
 
 	bch2_mark_metadata(c);
 
diff --git a/libbcachefs/btree_gc.h b/libbcachefs/btree_gc.h
index 07210d33..1d461518 100644
--- a/libbcachefs/btree_gc.h
+++ b/libbcachefs/btree_gc.h
@@ -11,7 +11,7 @@ void bch2_gc_thread_stop(struct bch_fs *);
 int bch2_gc_thread_start(struct bch_fs *);
 int bch2_initial_gc(struct bch_fs *, struct list_head *);
 u8 bch2_btree_key_recalc_oldest_gen(struct bch_fs *, struct bkey_s_c);
-u8 bch2_btree_mark_key_initial(struct bch_fs *, enum bkey_type,
+int bch2_btree_mark_key_initial(struct bch_fs *, enum bkey_type,
 				struct bkey_s_c);
 void bch2_mark_dev_metadata(struct bch_fs *, struct bch_dev *);
 
diff --git a/libbcachefs/buckets.c b/libbcachefs/buckets.c
index 396251d5..f9d41ffc 100644
--- a/libbcachefs/buckets.c
+++ b/libbcachefs/buckets.c
@@ -353,38 +353,45 @@ void bch2_mark_alloc_bucket(struct bch_dev *ca, struct bucket *g,
 	}));
 }
 
+#define saturated_add(ca, dst, src, max)			\
+do {								\
+	BUG_ON((int) (dst) + (src) < 0);			\
+	if ((dst) == (max))					\
+		;						\
+	else if ((dst) + (src) <= (max))			\
+		dst += (src);					\
+	else {							\
+		dst = (max);					\
+		trace_sectors_saturated(ca);		\
+	}							\
+} while (0)
+
 void bch2_mark_metadata_bucket(struct bch_dev *ca, struct bucket *g,
-			      enum bucket_data_type type,
-			      bool may_make_unavailable)
+			       enum bucket_data_type type,
+			       bool may_make_unavailable)
 {
 	struct bucket_mark old, new;
 
 	BUG_ON(!type);
 
 	old = bucket_data_cmpxchg(ca, g, new, ({
+		saturated_add(ca, new.dirty_sectors, ca->mi.bucket_size,
+			      GC_MAX_SECTORS_USED);
 		new.data_type = type;
 		new.had_metadata = 1;
 	}));
 
-	BUG_ON(old.cached_sectors);
-	BUG_ON(old.dirty_sectors);
+	if (old.data_type != type &&
+	    (old.data_type ||
+	     old.cached_sectors ||
+	     old.dirty_sectors))
+		bch_err(ca->fs, "bucket %zu has multiple types of data (%u, %u)",
+			g - ca->buckets, old.data_type, new.data_type);
+
 	BUG_ON(!may_make_unavailable &&
 	       bucket_became_unavailable(ca->fs, old, new));
 }
 
-#define saturated_add(ca, dst, src, max)			\
-do {								\
-	BUG_ON((int) (dst) + (src) < 0);			\
-	if ((dst) == (max))					\
-		;						\
-	else if ((dst) + (src) <= (max))			\
-		dst += (src);					\
-	else {							\
-		dst = (max);					\
-		trace_sectors_saturated(ca);		\
-	}							\
-} while (0)
-
 #if 0
 /* Reverting this until the copygc + compression issue is fixed: */
 
@@ -431,6 +438,8 @@ static void bch2_mark_pointer(struct bch_fs *c,
 	unsigned saturated;
 	struct bch_dev *ca = c->devs[ptr->dev];
 	struct bucket *g = ca->buckets + PTR_BUCKET_NR(ca, ptr);
+	unsigned data_type = type == S_META
+		? BUCKET_BTREE : BUCKET_DATA;
 	unsigned old_sectors, new_sectors;
 	int disk_sectors, compressed_sectors;
 
@@ -494,13 +503,19 @@ static void bch2_mark_pointer(struct bch_fs *c,
 				new.journal_seq = journal_seq;
 			}
 		} else {
-			new.data_type = type == S_META
-				? BUCKET_BTREE : BUCKET_DATA;
+			new.data_type = data_type;
 		}
 
 		new.had_metadata |= is_meta_bucket(new);
 	}));
 
+	if (old.data_type != data_type &&
+	    (old.data_type ||
+	     old.cached_sectors ||
+	     old.dirty_sectors))
+		bch_err(ca->fs, "bucket %zu has multiple types of data (%u, %u)",
+			g - ca->buckets, old.data_type, new.data_type);
+
 	BUG_ON(!may_make_unavailable &&
 	       bucket_became_unavailable(c, old, new));
 
diff --git a/libbcachefs/buckets.h b/libbcachefs/buckets.h
index 9c77304f..5303cdc7 100644
--- a/libbcachefs/buckets.h
+++ b/libbcachefs/buckets.h
@@ -73,20 +73,16 @@ static inline struct bucket *PTR_BUCKET(const struct bch_dev *ca,
 	return ca->buckets + PTR_BUCKET_NR(ca, ptr);
 }
 
-static inline u8 __gen_after(u8 a, u8 b)
+static inline int gen_cmp(u8 a, u8 b)
 {
-	u8 r = a - b;
-
-	return r > 128U ? 0 : r;
+	return (s8) (a - b);
 }
 
-static inline u8 gen_after(u8 a, u8 b)
+static inline int gen_after(u8 a, u8 b)
 {
-	u8 r = a - b;
-
-	BUG_ON(r > 128U);
+	int r = gen_cmp(a, b);
 
-	return r;
+	return r > 0 ? r : 0;
 }
 
 /**
@@ -239,7 +235,7 @@ void bch2_invalidate_bucket(struct bch_dev *, struct bucket *);
 void bch2_mark_free_bucket(struct bch_dev *, struct bucket *);
 void bch2_mark_alloc_bucket(struct bch_dev *, struct bucket *, bool);
 void bch2_mark_metadata_bucket(struct bch_dev *, struct bucket *,
-			      enum bucket_data_type, bool);
+			       enum bucket_data_type, bool);
 
 void __bch2_gc_mark_key(struct bch_fs *, struct bkey_s_c, s64, bool,
 		       struct bch_fs_usage *);
diff --git a/libbcachefs/extents.c b/libbcachefs/extents.c
index 26f9352a..c80da362 100644
--- a/libbcachefs/extents.c
+++ b/libbcachefs/extents.c
@@ -602,12 +602,6 @@ bch2_btree_pick_ptr(struct bch_fs *c, const struct btree *b)
 				PTR_BUCKET_NR(ca, ptr)))
 			break;
 
-		if (bch2_dev_inconsistent_on(ptr_stale(ca, ptr), ca,
-				"stale btree node pointer at btree %u level %u/%u bucket %zu",
-				b->btree_id, b->level, root ? root->level : -1,
-				PTR_BUCKET_NR(ca, ptr)))
-			continue;
-
 		if (ca->mi.state == BCH_MEMBER_STATE_FAILED)
 			continue;
 
diff --git a/libbcachefs/fs-gc.c b/libbcachefs/fs-gc.c
index 1a03b7ea..03370c0e 100644
--- a/libbcachefs/fs-gc.c
+++ b/libbcachefs/fs-gc.c
@@ -134,18 +134,22 @@ static int check_extents(struct bch_fs *c)
 			"extent type %u for missing inode %llu",
 			k.k->type, k.k->p.inode);
 
-		unfixable_fsck_err_on(w.first_this_inode && w.have_inode &&
+		unfixable_fsck_err_on(w.have_inode &&
+			!S_ISREG(w.inode.i_mode) && !S_ISLNK(w.inode.i_mode), c,
+			"extent type %u for non regular file, inode %llu mode %o",
+			k.k->type, k.k->p.inode, w.inode.i_mode);
+
+		unfixable_fsck_err_on(w.first_this_inode &&
+			w.have_inode &&
+			!(w.inode.i_flags & BCH_INODE_I_SECTORS_DIRTY) &&
 			w.inode.i_sectors !=
 			(i_sectors = bch2_count_inode_sectors(c, w.cur_inum)),
 			c, "i_sectors wrong: got %llu, should be %llu",
 			w.inode.i_sectors, i_sectors);
 
 		unfixable_fsck_err_on(w.have_inode &&
-			!S_ISREG(w.inode.i_mode) && !S_ISLNK(w.inode.i_mode), c,
-			"extent type %u for non regular file, inode %llu mode %o",
-			k.k->type, k.k->p.inode, w.inode.i_mode);
-
-		unfixable_fsck_err_on(k.k->type != BCH_RESERVATION &&
+			!(w.inode.i_flags & BCH_INODE_I_SIZE_DIRTY) &&
+			k.k->type != BCH_RESERVATION &&
 			k.k->p.offset > round_up(w.inode.i_size, PAGE_SIZE) >> 9, c,
 			"extent type %u offset %llu past end of inode %llu, i_size %llu",
 			k.k->type, k.k->p.offset, k.k->p.inode, w.inode.i_size);
diff --git a/libbcachefs/io.c b/libbcachefs/io.c
index 212a5a65..da06845c 100644
--- a/libbcachefs/io.c
+++ b/libbcachefs/io.c
@@ -475,7 +475,7 @@ static int bch2_write_extent(struct bch_write_op *op,
 			nonce = extent_nonce(op->version,
 					     crc_nonce,
 					     src_len >> 9,
-					     compression_type),
+					     fragment_compression_type),
 
 			bch2_encrypt_bio(c, csum_type, nonce, bio);
 
diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index 60c5c9b0..76a3b465 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -1077,20 +1077,26 @@ fsck_err:
 	return ret;
 }
 
-void bch2_journal_mark(struct bch_fs *c, struct list_head *list)
+int bch2_journal_mark(struct bch_fs *c, struct list_head *list)
 {
 	struct bkey_i *k, *n;
 	struct jset_entry *j;
 	struct journal_replay *r;
+	int ret;
 
 	list_for_each_entry(r, list, list)
 		for_each_jset_key(k, n, j, &r->j) {
 			enum bkey_type type = bkey_type(j->level, j->btree_id);
 			struct bkey_s_c k_s_c = bkey_i_to_s_c(k);
 
-			if (btree_type_has_ptrs(type))
-				bch2_btree_mark_key_initial(c, type, k_s_c);
+			if (btree_type_has_ptrs(type)) {
+				ret = bch2_btree_mark_key_initial(c, type, k_s_c);
+				if (ret)
+					return ret;
+			}
 		}
+
+	return 0;
 }
 
 static bool journal_entry_is_open(struct journal *j)
diff --git a/libbcachefs/journal.h b/libbcachefs/journal.h
index f5fc465a..3825f0dc 100644
--- a/libbcachefs/journal.h
+++ b/libbcachefs/journal.h
@@ -336,7 +336,7 @@ static inline bool journal_flushes_device(struct bch_dev *ca)
 }
 
 void bch2_journal_start(struct bch_fs *);
-void bch2_journal_mark(struct bch_fs *, struct list_head *);
+int bch2_journal_mark(struct bch_fs *, struct list_head *);
 void bch2_journal_entries_free(struct list_head *);
 int bch2_journal_read(struct bch_fs *, struct list_head *);
 int bch2_journal_replay(struct bch_fs *, struct list_head *);
diff --git a/libbcachefs/super.c b/libbcachefs/super.c
index 1977bdc4..366fa672 100644
--- a/libbcachefs/super.c
+++ b/libbcachefs/super.c
@@ -727,7 +727,8 @@ static const char *__bch2_fs_start(struct bch_fs *c)
 		bch_verbose(c, "starting mark and sweep:");
 
 		err = "error in recovery";
-		if (bch2_initial_gc(c, &journal))
+		ret = bch2_initial_gc(c, &journal);
+		if (ret)
 			goto err;
 
 		if (c->opts.noreplay)
@@ -777,7 +778,9 @@ static const char *__bch2_fs_start(struct bch_fs *c)
 
 		bch_notice(c, "initializing new filesystem");
 
-		bch2_initial_gc(c, NULL);
+		ret = bch2_initial_gc(c, &journal);
+		if (ret)
+			goto err;
 
 		err = "unable to allocate journal buckets";
 		for_each_rw_member(ca, c, i)