Merge tag 'bcachefs-2025-02-12' of git://evilpiepirate.org/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Just small stuff.

  As a general announcement, on disk format is now frozen in my master
  branch - future on disk format changes will be optional, not required.

   - More fixes for going read-only: the previous fix was insufficient,
     but with more work on ordering journal reclaim flushing (and a
     btree node accounting fix so we don't split until we have to) the
     tiering_replication test now consistently goes read-only in less
     than a second.

   - fix for fsck when we have reflink pointers to missing indirect
     extents

   - some transaction restart handling fixes from Alan; the "Pass
     _orig_restart_count to trans_was_restarted" likely fixes some rare
     undefined behaviour heisenbugs"

* tag 'bcachefs-2025-02-12' of git://evilpiepirate.org/bcachefs:
  bcachefs: Reuse transaction
  bcachefs: Pass _orig_restart_count to trans_was_restarted
  bcachefs: CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS
  bcachefs: Fix want_new_bset() so we write until the end of the btree node
  bcachefs: Split out journal pins by btree level
  bcachefs: Fix use after free
  bcachefs: Fix marking reflink pointers to missing indirect extents

Author: torvalds

fs/bcachefs/Kconfig

@@ -61,6 +61,13 @@ config BCACHEFS_DEBUG
 	The resulting code will be significantly slower than normal; you
 	probably shouldn't select this option unless you're a developer.
 
+config BCACHEFS_INJECT_TRANSACTION_RESTARTS
+	bool "Randomly inject transaction restarts"
+	depends on BCACHEFS_DEBUG
+	help
+	Randomly inject transaction restarts in a few core paths - may have a
+	significant performance penalty
+
 config BCACHEFS_TESTS
 	bool "bcachefs unit and performance tests"
 	depends on BCACHEFS_FS

fs/bcachefs/btree_iter.c

@@ -2357,6 +2357,12 @@ struct bkey_s_c bch2_btree_iter_peek_max(struct btree_iter *iter, struct bpos en
 	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bkey_eq(end, POS_MAX));
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret)) {
+		k = bkey_s_c_err(ret);
+		goto out_no_locked;
+	}
+
 	if (iter->update_path) {
 		bch2_path_put_nokeep(trans, iter->update_path,
 				     iter->flags & BTREE_ITER_intent);
@@ -2622,6 +2628,12 @@ struct bkey_s_c bch2_btree_iter_peek_prev_min(struct btree_iter *iter, struct bp
 	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bpos_eq(end, POS_MIN));
 
+	int ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret)) {
+		k = bkey_s_c_err(ret);
+		goto out_no_locked;
+	}
+
 	while (1) {
 		k = __bch2_btree_iter_peek_prev(iter, search_key);
 		if (unlikely(!k.k))
@@ -2749,6 +2761,12 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
 	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_with_key_cache));
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret)) {
+		k = bkey_s_c_err(ret);
+		goto out_no_locked;
+	}
+
 	/* extents can't span inode numbers: */
 	if ((iter->flags & BTREE_ITER_is_extents) &&
 	    unlikely(iter->pos.offset == KEY_OFFSET_MAX)) {
@@ -3106,6 +3124,10 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 
 	WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX);
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (ret)
+		return ERR_PTR(ret);
+
 	struct btree_transaction_stats *s = btree_trans_stats(trans);
 	s->max_mem = max(s->max_mem, new_bytes);
 
@@ -3163,7 +3185,8 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 
 	if (old_bytes) {
 		trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes);
-		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
+		return ERR_PTR(btree_trans_restart_ip(trans,
+					BCH_ERR_transaction_restart_mem_realloced, _RET_IP_));
 	}
 out_change_top:
 	p = trans->mem + trans->mem_top;
@@ -3271,6 +3294,14 @@ u32 bch2_trans_begin(struct btree_trans *trans)
 
 	trans->last_begin_ip = _RET_IP_;
 
+#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS
+	if (trans->restarted) {
+		trans->restart_count_this_trans++;
+	} else {
+		trans->restart_count_this_trans = 0;
+	}
+#endif
+
 	trans_set_locked(trans, false);
 
 	if (trans->restarted) {

fs/bcachefs/btree_iter.h

@@ -355,6 +355,18 @@ static int btree_trans_restart(struct btree_trans *trans, int err)
 	return btree_trans_restart_ip(trans, err, _THIS_IP_);
 }
 
+static inline int trans_maybe_inject_restart(struct btree_trans *trans, unsigned long ip)
+{
+#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS
+	if (!(ktime_get_ns() & ~(~0ULL << min(63, (10 + trans->restart_count_this_trans))))) {
+		trace_and_count(trans->c, trans_restart_injected, trans, ip);
+		return btree_trans_restart_ip(trans,
+					BCH_ERR_transaction_restart_fault_inject, ip);
+	}
+#endif
+	return 0;
+}
+
 bool bch2_btree_node_upgrade(struct btree_trans *,
 			     struct btree_path *, unsigned);
 
@@ -739,7 +751,7 @@ transaction_restart:							\
 	if (!_ret2)							\
 		bch2_trans_verify_not_restarted(_trans, _restart_count);\
 									\
-	_ret2 ?: trans_was_restarted(_trans, _restart_count);		\
+	_ret2 ?: trans_was_restarted(_trans, _orig_restart_count);		\
 })
 
 #define for_each_btree_key_max_continue(_trans, _iter,			\

fs/bcachefs/btree_trans_commit.c

@@ -999,6 +999,10 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 
 	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
+	ret = trans_maybe_inject_restart(trans, _RET_IP_);
+	if (unlikely(ret))
+		goto out_reset;
+
 	if (!trans->nr_updates &&
 	    !trans->journal_entries_u64s)
 		goto out_reset;

fs/bcachefs/btree_types.h

@@ -509,6 +509,9 @@ struct btree_trans {
 	bool			notrace_relock_fail:1;
 	enum bch_errcode	restarted:16;
 	u32			restart_count;
+#ifdef CONFIG_BCACHEFS_INJECT_TRANSACTION_RESTARTS
+	u32			restart_count_this_trans;
+#endif
 
 	u64			last_begin_time;
 	unsigned long		last_begin_ip;

fs/bcachefs/btree_update_interior.h

@@ -278,12 +278,12 @@ static inline struct btree_node_entry *want_new_bset(struct bch_fs *c, struct bt
 {
 	struct bset_tree *t = bset_tree_last(b);
 	struct btree_node_entry *bne = max(write_block(b),
-			(void *) btree_bkey_last(b, bset_tree_last(b)));
+			(void *) btree_bkey_last(b, t));
 	ssize_t remaining_space =
 		__bch2_btree_u64s_remaining(b, bne->keys.start);
 
 	if (unlikely(bset_written(b, bset(b, t)))) {
-		if (remaining_space > (ssize_t) (block_bytes(c) >> 3))
+		if (b->written + block_sectors(c) <= btree_sectors(c))
 			return bne;
 	} else {
 		if (unlikely(bset_u64s(t) * sizeof(u64) > btree_write_set_buffer(b)) &&

fs/bcachefs/disk_accounting.h

@@ -210,11 +210,13 @@ static inline void bch2_accounting_mem_read_counters(struct bch_accounting_mem *
 static inline void bch2_accounting_mem_read(struct bch_fs *c, struct bpos p,
 					    u64 *v, unsigned nr)
 {
+	percpu_down_read(&c->mark_lock);
 	struct bch_accounting_mem *acc = &c->accounting;
 	unsigned idx = eytzinger0_find(acc->k.data, acc->k.nr, sizeof(acc->k.data[0]),
 				       accounting_pos_cmp, &p);
 
 	bch2_accounting_mem_read_counters(acc, idx, v, nr, false);
+	percpu_up_read(&c->mark_lock);
 }
 
 static inline struct bversion journal_pos_to_bversion(struct journal_res *res, unsigned offset)

fs/bcachefs/io_write.c

@@ -411,6 +411,16 @@ void bch2_write_op_error(struct printbuf *out, struct bch_write_op *op)
 	__bch2_write_op_error(out, op, op->pos.offset);
 }
 
+static void bch2_write_op_error_trans(struct btree_trans *trans, struct printbuf *out,
+				      struct bch_write_op *op, u64 offset)
+{
+	bch2_inum_offset_err_msg_trans(trans, out,
+				       (subvol_inum) { op->subvol, op->pos.inode, },
+				       offset << 9);
+	prt_printf(out, "write error%s: ",
+		   op->flags & BCH_WRITE_MOVE ? "(internal move)" : "");
+}
+
 void bch2_submit_wbio_replicas(struct bch_write_bio *wbio, struct bch_fs *c,
 			       enum bch_data_type type,
 			       const struct bkey_i *k,
@@ -1193,7 +1203,7 @@ static void bch2_nocow_write_convert_unwritten(struct bch_write_op *op)
 			struct bkey_i *insert = bch2_keylist_front(&op->insert_keys);
 
 			struct printbuf buf = PRINTBUF;
-			__bch2_write_op_error(&buf, op, bkey_start_offset(&insert->k));
+			bch2_write_op_error_trans(trans, &buf, op, bkey_start_offset(&insert->k));
 			prt_printf(&buf, "btree update error: %s", bch2_err_str(ret));
 			bch_err_ratelimited(c, "%s", buf.buf);
 			printbuf_exit(&buf);

fs/bcachefs/journal_reclaim.c

@@ -384,12 +384,16 @@ void bch2_journal_pin_drop(struct journal *j,
 	spin_unlock(&j->lock);
 }
 
-static enum journal_pin_type journal_pin_type(journal_pin_flush_fn fn)
+static enum journal_pin_type journal_pin_type(struct journal_entry_pin *pin,
+					      journal_pin_flush_fn fn)
 {
 	if (fn == bch2_btree_node_flush0 ||
-	    fn == bch2_btree_node_flush1)
-		return JOURNAL_PIN_TYPE_btree;
-	else if (fn == bch2_btree_key_cache_journal_flush)
+	    fn == bch2_btree_node_flush1) {
+		unsigned idx = fn == bch2_btree_node_flush1;
+		struct btree *b = container_of(pin, struct btree, writes[idx].journal);
+
+		return JOURNAL_PIN_TYPE_btree0 - b->c.level;
+	} else if (fn == bch2_btree_key_cache_journal_flush)
 		return JOURNAL_PIN_TYPE_key_cache;
 	else
 		return JOURNAL_PIN_TYPE_other;
@@ -441,7 +445,7 @@ void bch2_journal_pin_copy(struct journal *j,
 
 	bool reclaim = __journal_pin_drop(j, dst);
 
-	bch2_journal_pin_set_locked(j, seq, dst, flush_fn, journal_pin_type(flush_fn));
+	bch2_journal_pin_set_locked(j, seq, dst, flush_fn, journal_pin_type(dst, flush_fn));
 
 	if (reclaim)
 		bch2_journal_reclaim_fast(j);
@@ -465,7 +469,7 @@ void bch2_journal_pin_set(struct journal *j, u64 seq,
 
 	bool reclaim = __journal_pin_drop(j, pin);
 
-	bch2_journal_pin_set_locked(j, seq, pin, flush_fn, journal_pin_type(flush_fn));
+	bch2_journal_pin_set_locked(j, seq, pin, flush_fn, journal_pin_type(pin, flush_fn));
 
 	if (reclaim)
 		bch2_journal_reclaim_fast(j);
@@ -587,7 +591,7 @@ static size_t journal_flush_pins(struct journal *j,
 		spin_lock(&j->lock);
 		/* Pin might have been dropped or rearmed: */
 		if (likely(!err && !j->flush_in_progress_dropped))
-			list_move(&pin->list, &journal_seq_pin(j, seq)->flushed[journal_pin_type(flush_fn)]);
+			list_move(&pin->list, &journal_seq_pin(j, seq)->flushed[journal_pin_type(pin, flush_fn)]);
 		j->flush_in_progress = NULL;
 		j->flush_in_progress_dropped = false;
 		spin_unlock(&j->lock);
@@ -869,18 +873,13 @@ static int journal_flush_done(struct journal *j, u64 seq_to_flush,
 
 	mutex_lock(&j->reclaim_lock);
 
-	if (journal_flush_pins_or_still_flushing(j, seq_to_flush,
-			       BIT(JOURNAL_PIN_TYPE_key_cache)|
-			       BIT(JOURNAL_PIN_TYPE_other))) {
-		*did_work = true;
-		goto unlock;
-	}
-
-	if (journal_flush_pins_or_still_flushing(j, seq_to_flush,
-			       BIT(JOURNAL_PIN_TYPE_btree))) {
-		*did_work = true;
-		goto unlock;
-	}
+	for (int type = JOURNAL_PIN_TYPE_NR - 1;
+	     type >= 0;
+	     --type)
+		if (journal_flush_pins_or_still_flushing(j, seq_to_flush, BIT(type))) {
+			*did_work = true;
+			goto unlock;
+		}
 
 	if (seq_to_flush > journal_cur_seq(j))
 		bch2_journal_entry_close(j);

fs/bcachefs/journal_types.h

@@ -53,7 +53,10 @@ struct journal_buf {
  */
 
 enum journal_pin_type {
-	JOURNAL_PIN_TYPE_btree,
+	JOURNAL_PIN_TYPE_btree3,
+	JOURNAL_PIN_TYPE_btree2,
+	JOURNAL_PIN_TYPE_btree1,
+	JOURNAL_PIN_TYPE_btree0,
 	JOURNAL_PIN_TYPE_key_cache,
 	JOURNAL_PIN_TYPE_other,
 	JOURNAL_PIN_TYPE_NR,