bcachefs: Move 'btree_transactions' debug to debugs

This moves btree_transactions from sysfs to debugfs, and makes it more
verbose: now we also include the backtrace of each task, since we
generally need this for debugging deadlocks.

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

5 files changed, 123 insertions, 64 deletions

diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 5c5e14d1d360..a1512eb06914 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -3273,11 +3273,14 @@ void __bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
 		       const char *fn)
 	__acquires(&c->btree_trans_barrier)
 {
+	struct btree_trans *pos;
+
 	BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
 
 	memset(trans, 0, sizeof(*trans));
 	trans->c		= c;
 	trans->fn		= fn;
+	trans->task		= current;
 
 	bch2_trans_alloc_paths(trans, c);
 
@@ -3293,9 +3296,15 @@ void __bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
 
 	trans->srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
 
-	trans->pid = current->pid;
 	mutex_lock(&c->btree_trans_lock);
-	list_add(&trans->list, &c->btree_trans_list);
+	list_for_each_entry(pos, &c->btree_trans_list, list) {
+		if (trans->task->pid < pos->task->pid) {
+			list_add_tail(&trans->list, &pos->list);
+			goto list_add_done;
+		}
+	}
+	list_add_tail(&trans->list, &c->btree_trans_list);
+list_add_done:
 	mutex_unlock(&c->btree_trans_lock);
 }
 
@@ -3383,73 +3392,55 @@ bch2_btree_path_node_to_text(struct printbuf *out,
 	bch2_bpos_to_text(out, btree_node_pos(_b, cached));
 }
 
-static bool trans_has_locks(struct btree_trans *trans)
-{
-	struct btree_path *path;
-
-	trans_for_each_path(trans, path)
-		if (path->nodes_locked)
-			return true;
-	return false;
-}
-
-void bch2_btree_trans_to_text(struct printbuf *out, struct bch_fs *c)
+void bch2_btree_trans_to_text(struct printbuf *out, struct btree_trans *trans)
 {
-	struct btree_trans *trans;
 	struct btree_path *path;
 	struct btree *b;
 	static char lock_types[] = { 'r', 'i', 'w' };
 	unsigned l;
 
-	mutex_lock(&c->btree_trans_lock);
-	list_for_each_entry(trans, &c->btree_trans_list, list) {
-		if (!trans_has_locks(trans))
-			continue;
-
-		prt_printf(out, "%i %s\n", trans->pid, trans->fn);
+	prt_printf(out, "%i %s\n", trans->task->pid, trans->fn);
 
-		trans_for_each_path(trans, path) {
-			if (!path->nodes_locked)
-				continue;
+	trans_for_each_path(trans, path) {
+		if (!path->nodes_locked)
+			continue;
 
-			prt_printf(out, "  path %u %c l=%u %s:",
-			       path->idx,
-			       path->cached ? 'c' : 'b',
-			       path->level,
-			       bch2_btree_ids[path->btree_id]);
-			bch2_bpos_to_text(out, path->pos);
-			prt_printf(out, "\n");
-
-			for (l = 0; l < BTREE_MAX_DEPTH; l++) {
-				if (btree_node_locked(path, l)) {
-					prt_printf(out, "    %s l=%u ",
-					       btree_node_intent_locked(path, l) ? "i" : "r", l);
-					bch2_btree_path_node_to_text(out,
-							(void *) path->l[l].b,
-							path->cached);
-					prt_printf(out, "\n");
-				}
+		prt_printf(out, "  path %u %c l=%u %s:",
+		       path->idx,
+		       path->cached ? 'c' : 'b',
+		       path->level,
+		       bch2_btree_ids[path->btree_id]);
+		bch2_bpos_to_text(out, path->pos);
+		prt_printf(out, "\n");
+
+		for (l = 0; l < BTREE_MAX_DEPTH; l++) {
+			if (btree_node_locked(path, l)) {
+				prt_printf(out, "    %s l=%u ",
+				       btree_node_intent_locked(path, l) ? "i" : "r", l);
+				bch2_btree_path_node_to_text(out,
+						(void *) path->l[l].b,
+						path->cached);
+				prt_printf(out, "\n");
 			}
 		}
+	}
 
-		b = READ_ONCE(trans->locking);
-		if (b) {
-			path = &trans->paths[trans->locking_path_idx];
-			prt_printf(out, "  locking path %u %c l=%u %c %s:",
-			       trans->locking_path_idx,
-			       path->cached ? 'c' : 'b',
-			       trans->locking_level,
-			       lock_types[trans->locking_lock_type],
-			       bch2_btree_ids[trans->locking_btree_id]);
-			bch2_bpos_to_text(out, trans->locking_pos);
-
-			prt_printf(out, " node ");
-			bch2_btree_path_node_to_text(out,
-					(void *) b, path->cached);
-			prt_printf(out, "\n");
-		}
+	b = READ_ONCE(trans->locking);
+	if (b) {
+		path = &trans->paths[trans->locking_path_idx];
+		prt_printf(out, "  locking path %u %c l=%u %c %s:",
+		       trans->locking_path_idx,
+		       path->cached ? 'c' : 'b',
+		       trans->locking_level,
+		       lock_types[trans->locking_lock_type],
+		       bch2_btree_ids[trans->locking_btree_id]);
+		bch2_bpos_to_text(out, trans->locking_pos);
+
+		prt_printf(out, " node ");
+		bch2_btree_path_node_to_text(out,
+				(void *) b, path->cached);
+		prt_printf(out, "\n");
 	}
-	mutex_unlock(&c->btree_trans_lock);
 }
 
 void bch2_fs_btree_iter_exit(struct bch_fs *c)
diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h
index f2b302c8150c..9da0a4152ce4 100644
--- a/fs/bcachefs/btree_iter.h
+++ b/fs/bcachefs/btree_iter.h
@@ -403,7 +403,7 @@ void bch2_trans_exit(struct btree_trans *);
 
 #define bch2_trans_init(...)	__bch2_trans_init(__VA_ARGS__, __func__)
 
-void bch2_btree_trans_to_text(struct printbuf *, struct bch_fs *);
+void bch2_btree_trans_to_text(struct printbuf *, struct btree_trans *);
 
 void bch2_fs_btree_iter_exit(struct bch_fs *);
 int bch2_fs_btree_iter_init(struct bch_fs *);
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index 5382f2b85e19..7e26e7584f1d 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -388,7 +388,7 @@ struct btree_trans {
 	u8			locking_btree_id;
 	u8			locking_level;
 	u8			locking_lock_type;
-	pid_t			pid;
+	struct task_struct	*task;
 	int			srcu_idx;
 
 	u8			nr_sorted;
diff --git a/fs/bcachefs/debug.c b/fs/bcachefs/debug.c
index bdc50d5ba3a2..05cae0ed41ae 100644
--- a/fs/bcachefs/debug.c
+++ b/fs/bcachefs/debug.c
@@ -529,6 +529,76 @@ static const struct file_operations cached_btree_nodes_ops = {
 	.read		= bch2_cached_btree_nodes_read,
 };
 
+static int prt_backtrace(struct printbuf *out, struct task_struct *task)
+{
+	unsigned long entries[32];
+	unsigned i, nr_entries;
+	int ret;
+
+	ret = down_read_killable(&task->signal->exec_update_lock);
+	if (ret)
+		return ret;
+
+	nr_entries = stack_trace_save_tsk(task, entries, ARRAY_SIZE(entries), 0);
+	for (i = 0; i < nr_entries; i++) {
+		prt_printf(out, "[<0>] %pB", (void *)entries[i]);
+		prt_newline(out);
+	}
+
+	up_read(&task->signal->exec_update_lock);
+	return 0;
+}
+
+static ssize_t bch2_btree_transactions_read(struct file *file, char __user *buf,
+					    size_t size, loff_t *ppos)
+{
+	struct dump_iter *i = file->private_data;
+	struct bch_fs *c = i->c;
+	struct btree_trans *trans;
+	int err;
+
+	i->ubuf = buf;
+	i->size	= size;
+	i->ret	= 0;
+
+	mutex_lock(&c->btree_trans_lock);
+	list_for_each_entry(trans, &c->btree_trans_list, list) {
+		if (trans->task->pid <= i->iter)
+			continue;
+
+		err = flush_buf(i);
+		if (err)
+			return err;
+
+		if (!i->size)
+			break;
+
+		bch2_btree_trans_to_text(&i->buf, trans);
+
+		prt_printf(&i->buf, "backtrace:");
+		prt_newline(&i->buf);
+		printbuf_indent_add(&i->buf, 2);
+		prt_backtrace(&i->buf, trans->task);
+		printbuf_indent_sub(&i->buf, 2);
+		prt_newline(&i->buf);
+
+		i->iter = trans->task->pid;
+	}
+	mutex_unlock(&c->btree_trans_lock);
+
+	if (i->buf.allocation_failure)
+		return -ENOMEM;
+
+	return i->ret;
+}
+
+static const struct file_operations btree_transactions_ops = {
+	.owner		= THIS_MODULE,
+	.open		= bch2_dump_open,
+	.release	= bch2_dump_release,
+	.read		= bch2_btree_transactions_read,
+};
+
 static ssize_t bch2_journal_pins_read(struct file *file, char __user *buf,
 				      size_t size, loff_t *ppos)
 {
@@ -588,6 +658,9 @@ void bch2_fs_debug_init(struct bch_fs *c)
 	debugfs_create_file("cached_btree_nodes", 0400, c->fs_debug_dir,
 			    c->btree_debug, &cached_btree_nodes_ops);
 
+	debugfs_create_file("btree_transactions", 0400, c->fs_debug_dir,
+			    c->btree_debug, &btree_transactions_ops);
+
 	debugfs_create_file("journal_pins", 0400, c->fs_debug_dir,
 			    c->btree_debug, &journal_pins_ops);
 
diff --git a/fs/bcachefs/sysfs.c b/fs/bcachefs/sysfs.c
index d72ec0629a37..d1a3ce747fda 100644
--- a/fs/bcachefs/sysfs.c
+++ b/fs/bcachefs/sysfs.c
@@ -182,7 +182,6 @@ read_attribute(journal_debug);
 read_attribute(btree_updates);
 read_attribute(btree_cache);
 read_attribute(btree_key_cache);
-read_attribute(btree_transactions);
 read_attribute(stripes_heap);
 read_attribute(open_buckets);
 
@@ -420,9 +419,6 @@ SHOW(bch2_fs)
 	if (attr == &sysfs_btree_key_cache)
 		bch2_btree_key_cache_to_text(out, &c->btree_key_cache);
 
-	if (attr == &sysfs_btree_transactions)
-		bch2_btree_trans_to_text(out, c);
-
 	if (attr == &sysfs_stripes_heap)
 		bch2_stripes_heap_to_text(out, c);
 
@@ -617,7 +613,6 @@ struct attribute *bch2_fs_internal_files[] = {
 	&sysfs_btree_updates,
 	&sysfs_btree_cache,
 	&sysfs_btree_key_cache,
-	&sysfs_btree_transactions,
 	&sysfs_new_stripes,
 	&sysfs_stripes_heap,
 	&sysfs_open_buckets,