From 50dc0f692a0dbe3e6a95d3f8e5c7e718bc9f021d Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@gmail.com>
Date: Fri, 19 Mar 2021 20:29:11 -0400
Subject: [PATCH] bcachefs: Require all btree iterators to be freed

We keep running into occasional bugs with btree transaction iterators
overflowing - this will make those bugs more visible.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/acl.c              | 20 ++++++++------
 fs/bcachefs/alloc_background.c |  4 +--
 fs/bcachefs/btree_gc.c         | 12 ++++++---
 fs/bcachefs/btree_io.c         |  1 +
 fs/bcachefs/btree_iter.c       | 17 ++++++++++++
 fs/bcachefs/debug.c            |  4 +++
 fs/bcachefs/dirent.c           |  3 +++
 fs/bcachefs/ec.c               |  4 ++-
 fs/bcachefs/extents.c          |  2 ++
 fs/bcachefs/fs-io.c            | 49 +++++++++++++++++++---------------
 fs/bcachefs/fs.c               |  3 +++
 fs/bcachefs/fsck.c             |  3 ++-
 fs/bcachefs/inode.c            |  1 +
 fs/bcachefs/io.c               | 38 +++++++++++++-------------
 fs/bcachefs/migrate.c          |  9 +++++--
 fs/bcachefs/move.c             |  3 +++
 fs/bcachefs/quota.c            |  7 ++++-
 fs/bcachefs/reflink.c          | 26 +++++++-----------
 fs/bcachefs/tests.c            | 28 ++++++++++++++-----
 fs/bcachefs/xattr.c            | 18 +++++++------
 20 files changed, 163 insertions(+), 89 deletions(-)

diff --git a/fs/bcachefs/acl.c b/fs/bcachefs/acl.c
index 3879815bcede..afb9562be2b2 100644
--- a/fs/bcachefs/acl.c
+++ b/fs/bcachefs/acl.c
@@ -241,12 +241,12 @@ retry:
 	}
 
 	xattr = bkey_s_c_to_xattr(bch2_btree_iter_peek_slot(iter));
-
 	acl = bch2_acl_from_disk(xattr_val(xattr.v),
 			le16_to_cpu(xattr.v->x_val_len));
 
 	if (!IS_ERR(acl))
 		set_cached_acl(&inode->v, type, acl);
+	bch2_trans_iter_put(&trans, iter);
 out:
 	bch2_trans_exit(&trans);
 	return acl;
@@ -313,7 +313,7 @@ retry:
 	if (type == ACL_TYPE_ACCESS) {
 		ret = posix_acl_update_mode(idmap, &inode->v, &mode, &acl);
 		if (ret)
-			goto err;
+			goto btree_err;
 	}
 
 	hash_info = bch2_hash_info_init(c, &inode_u);
@@ -330,6 +330,8 @@ retry:
 				  &inode->ei_journal_seq,
 				  BTREE_INSERT_NOUNLOCK);
 btree_err:
+	bch2_trans_iter_put(&trans, inode_iter);
+
 	if (ret == -EINTR)
 		goto retry;
 	if (unlikely(ret))
@@ -356,21 +358,22 @@ int bch2_acl_chmod(struct btree_trans *trans,
 	struct bkey_s_c_xattr xattr;
 	struct bkey_i_xattr *new;
 	struct posix_acl *acl;
-	int ret = 0;
+	int ret;
 
 	iter = bch2_hash_lookup(trans, bch2_xattr_hash_desc,
 			&hash_info, inode->bi_inum,
 			&X_SEARCH(KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS, "", 0),
 			BTREE_ITER_INTENT);
-	if (IS_ERR(iter))
-		return PTR_ERR(iter) != -ENOENT ? PTR_ERR(iter) : 0;
+	ret = PTR_ERR_OR_ZERO(iter);
+	if (ret)
+		return ret == -ENOENT ? 0 : ret;
 
 	xattr = bkey_s_c_to_xattr(bch2_btree_iter_peek_slot(iter));
-
 	acl = bch2_acl_from_disk(xattr_val(xattr.v),
 			le16_to_cpu(xattr.v->x_val_len));
-	if (IS_ERR_OR_NULL(acl))
-		return PTR_ERR(acl);
+	ret = PTR_ERR_OR_ZERO(acl);
+	if (ret || !acl)
+		goto err;
 
 	ret = __posix_acl_chmod(&acl, GFP_KERNEL, mode);
 	if (ret)
@@ -387,6 +390,7 @@ int bch2_acl_chmod(struct btree_trans *trans,
 	*new_acl = acl;
 	acl = NULL;
 err:
+	bch2_trans_iter_put(trans, iter);
 	kfree(acl);
 	return ret;
 }
diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index add04dcb849b..e2200cedecca 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -385,7 +385,6 @@ int bch2_alloc_write(struct bch_fs *c, unsigned flags)
 	int ret = 0;
 
 	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
 	iter = bch2_trans_get_iter(&trans, BTREE_ID_alloc, POS_MIN,
 				   BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 
@@ -405,6 +404,7 @@ int bch2_alloc_write(struct bch_fs *c, unsigned flags)
 		}
 	}
 err:
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -926,7 +926,6 @@ static int bch2_invalidate_buckets(struct bch_fs *c, struct bch_dev *ca)
 	int ret = 0;
 
 	bch2_trans_init(&trans, c, 0, 0);
-
 	iter = bch2_trans_get_iter(&trans, BTREE_ID_alloc,
 				   POS(ca->dev_idx, 0),
 				   BTREE_ITER_CACHED|
@@ -942,6 +941,7 @@ static int bch2_invalidate_buckets(struct bch_fs *c, struct bch_dev *ca)
 				(!fifo_empty(&ca->free_inc)
 				 ? BTREE_INSERT_NOWAIT : 0));
 
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 
 	/* If we used NOWAIT, don't return the error: */
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index e8cdc82d3451..a303cd376d4b 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -456,6 +456,8 @@ static int bch2_gc_btree(struct bch_fs *c, enum btree_id btree_id,
 
 		bch2_trans_cond_resched(&trans);
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	ret = bch2_trans_exit(&trans) ?: ret;
 	if (ret)
 		return ret;
@@ -1212,6 +1214,7 @@ static int bch2_gc_btree_gens(struct bch_fs *c, enum btree_id btree_id)
 
 		bch2_btree_iter_next(iter);
 	}
+	bch2_trans_iter_put(&trans, iter);
 
 	bch2_trans_exit(&trans);
 	bch2_bkey_buf_exit(&sk, c);
@@ -1509,6 +1512,7 @@ static int bch2_coalesce_btree(struct bch_fs *c, enum btree_id btree_id)
 	struct btree *b;
 	bool kthread = (current->flags & PF_KTHREAD) != 0;
 	unsigned i;
+	int ret = 0;
 
 	/* Sliding window of adjacent btree nodes */
 	struct btree *merge[GC_MERGE_NODES];
@@ -1557,8 +1561,8 @@ static int bch2_coalesce_btree(struct bch_fs *c, enum btree_id btree_id)
 		lock_seq[0] = merge[0]->c.lock.state.seq;
 
 		if (kthread && kthread_should_stop()) {
-			bch2_trans_exit(&trans);
-			return -ESHUTDOWN;
+			ret = -ESHUTDOWN;
+			break;
 		}
 
 		bch2_trans_cond_resched(&trans);
@@ -1573,7 +1577,9 @@ static int bch2_coalesce_btree(struct bch_fs *c, enum btree_id btree_id)
 			memset(merge + 1, 0,
 			       (GC_MERGE_NODES - 1) * sizeof(merge[0]));
 	}
-	return bch2_trans_exit(&trans);
+	bch2_trans_iter_put(&trans, iter);
+
+	return bch2_trans_exit(&trans) ?: ret;
 }
 
 /**
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index d547bfabf09f..7ec14cd8f02b 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -1208,6 +1208,7 @@ retry:
 	if (ret)
 		goto err;
 out:
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	bch2_bkey_buf_exit(&k, c);
 	bio_put(&wbio->wbio.bio);
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 02a486e83881..00140ae50cb9 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -9,6 +9,7 @@
 #include "btree_locking.h"
 #include "btree_update.h"
 #include "debug.h"
+#include "error.h"
 #include "extents.h"
 #include "journal.h"
 #include "trace.h"
@@ -2116,6 +2117,7 @@ struct btree_iter *bch2_trans_get_node_iter(struct btree_trans *trans,
 	for (i = 0; i < ARRAY_SIZE(iter->l); i++)
 		iter->l[i].b		= NULL;
 	iter->l[iter->level].b		= BTREE_ITER_NO_NODE_INIT;
+	iter->ip_allocated = _RET_IP_;
 
 	return iter;
 }
@@ -2224,6 +2226,8 @@ void bch2_trans_reset(struct btree_trans *trans, unsigned flags)
 		       (void *) &trans->fs_usage_deltas->memset_start);
 	}
 
+	bch2_trans_cond_resched(trans);
+
 	if (!(flags & TRANS_RESET_NOTRAVERSE))
 		bch2_btree_iter_traverse_all(trans);
 }
@@ -2290,6 +2294,19 @@ int bch2_trans_exit(struct btree_trans *trans)
 	bch2_trans_unlock(trans);
 
 #ifdef CONFIG_BCACHEFS_DEBUG
+	if (trans->iters_live) {
+		struct btree_iter *iter;
+
+		bch_err(c, "btree iterators leaked!");
+		trans_for_each_iter(trans, iter)
+			if (btree_iter_live(trans, iter))
+				printk(KERN_ERR "  btree %s allocated at %pS\n",
+				       bch2_btree_ids[iter->btree_id],
+				       (void *) iter->ip_allocated);
+		/* Be noisy about this: */
+		bch2_fatal_error(c);
+	}
+
 	mutex_lock(&trans->c->btree_trans_lock);
 	list_del(&trans->list);
 	mutex_unlock(&trans->c->btree_trans_lock);
diff --git a/fs/bcachefs/debug.c b/fs/bcachefs/debug.c
index 3ac700dc72d5..2c2d58514c68 100644
--- a/fs/bcachefs/debug.c
+++ b/fs/bcachefs/debug.c
@@ -242,6 +242,8 @@ static ssize_t bch2_read_btree(struct file *file, char __user *buf,
 		if (!i->size)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 
 	return err < 0 ? err : i->ret;
@@ -294,6 +296,8 @@ static ssize_t bch2_read_btree_formats(struct file *file, char __user *buf,
 		if (!i->size)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 
 	return err < 0 ? err : i->ret;
diff --git a/fs/bcachefs/dirent.c b/fs/bcachefs/dirent.c
index b0625176ab35..592dd80cf963 100644
--- a/fs/bcachefs/dirent.c
+++ b/fs/bcachefs/dirent.c
@@ -321,6 +321,7 @@ u64 bch2_dirent_lookup(struct bch_fs *c, u64 dir_inum,
 
 	k = bch2_btree_iter_peek_slot(iter);
 	inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
+	bch2_trans_iter_put(&trans, iter);
 out:
 	bch2_trans_exit(&trans);
 	return inum;
@@ -379,6 +380,8 @@ int bch2_readdir(struct bch_fs *c, u64 inum, struct dir_context *ctx)
 			break;
 		ctx->pos = dirent.k->p.offset + 1;
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	ret = bch2_trans_exit(&trans) ?: ret;
 
 	return ret;
diff --git a/fs/bcachefs/ec.c b/fs/bcachefs/ec.c
index f61d4c873a82..bdce37981c5c 100644
--- a/fs/bcachefs/ec.c
+++ b/fs/bcachefs/ec.c
@@ -873,6 +873,7 @@ static int ec_stripe_update_ptrs(struct bch_fs *c,
 		if (ret)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
 
 	bch2_trans_exit(&trans);
 	bch2_bkey_buf_exit(&sk, c);
@@ -1663,12 +1664,13 @@ int bch2_ec_mem_alloc(struct bch_fs *c, bool gc)
 	int ret = 0;
 
 	bch2_trans_init(&trans, c, 0, 0);
-
 	iter = bch2_trans_get_iter(&trans, BTREE_ID_stripes, POS(0, U64_MAX), 0);
 
 	k = bch2_btree_iter_prev(iter);
 	if (!IS_ERR_OR_NULL(k.k))
 		idx = k.k->p.offset + 1;
+
+	bch2_trans_iter_put(&trans, iter);
 	ret = bch2_trans_exit(&trans);
 	if (ret)
 		return ret;
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 3fe9ef50f5c0..8cf45b7b9459 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -687,6 +687,8 @@ bool bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size,
 			break;
 		}
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 
 	return ret;
diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 332795eb9ae8..8891207c46a9 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -866,7 +866,6 @@ void bch2_readahead(struct readahead_control *ractl)
 	BUG_ON(ret);
 
 	bch2_trans_init(&trans, c, 0, 0);
-
 	iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, POS_MIN,
 				   BTREE_ITER_SLOTS);
 
@@ -895,6 +894,7 @@ void bch2_readahead(struct readahead_control *ractl)
 
 	bch2_pagecache_add_put(&inode->ei_pagecache_lock);
 
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	kfree(readpages_iter.pages);
 }
@@ -918,6 +918,7 @@ static void __bchfs_readpage(struct bch_fs *c, struct bch_read_bio *rbio,
 
 	bchfs_read(&trans, iter, rbio, inum, NULL);
 
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 }
 
@@ -2155,6 +2156,7 @@ static inline int range_has_data(struct bch_fs *c,
 			break;
 		}
 	}
+	bch2_trans_iter_put(&trans, iter);
 
 	return bch2_trans_exit(&trans) ?: ret;
 }
@@ -2325,6 +2327,7 @@ int bch2_truncate(struct bch_inode_info *inode, struct iattr *iattr)
 	bch2_trans_init(&trans, c, 0, 0);
 	iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino, 0);
 	ret = PTR_ERR_OR_ZERO(iter);
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 
 	if (ret)
@@ -2459,14 +2462,11 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 	struct btree_iter *src, *dst, *del;
 	loff_t shift, new_size;
 	u64 src_start;
-	int ret;
+	int ret = 0;
 
 	if ((offset | len) & (block_bytes(c) - 1))
 		return -EINVAL;
 
-	bch2_bkey_buf_init(&copy);
-	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 256);
-
 	/*
 	 * We need i_mutex to keep the page cache consistent with the extents
 	 * btree, and the btree consistent with i_size - we don't need outside
@@ -2522,13 +2522,15 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 			goto err;
 	}
 
+	bch2_bkey_buf_init(&copy);
+	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 256);
 	src = bch2_trans_get_iter(&trans, BTREE_ID_extents,
 			POS(inode->v.i_ino, src_start >> 9),
 			BTREE_ITER_INTENT);
 	dst = bch2_trans_copy_iter(&trans, src);
 	del = bch2_trans_copy_iter(&trans, src);
 
-	while (1) {
+	while (ret == 0 || ret == -EINTR) {
 		struct disk_reservation disk_res =
 			bch2_disk_reservation_init(c, 0);
 		struct bkey_i delete;
@@ -2542,7 +2544,7 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 			? bch2_btree_iter_peek_prev(src)
 			: bch2_btree_iter_peek(src);
 		if ((ret = bkey_err(k)))
-			goto bkey_err;
+			continue;
 
 		if (!k.k || k.k->p.inode != inode->v.i_ino)
 			break;
@@ -2562,7 +2564,7 @@ reassemble:
 
 		ret = bch2_extent_atomic_end(dst, copy.k, &atomic_end);
 		if (ret)
-			goto bkey_err;
+			continue;
 
 		if (bkey_cmp(atomic_end, copy.k->k.p)) {
 			if (insert) {
@@ -2605,18 +2607,18 @@ reassemble:
 					  &inode->ei_journal_seq,
 					  BTREE_INSERT_NOFAIL);
 		bch2_disk_reservation_put(c, &disk_res);
-bkey_err:
+
 		if (!ret)
 			bch2_btree_iter_set_pos(src, next_pos);
-
-		if (ret == -EINTR)
-			ret = 0;
-		if (ret)
-			goto err;
-
-		bch2_trans_cond_resched(&trans);
 	}
-	bch2_trans_unlock(&trans);
+	bch2_trans_iter_put(&trans, del);
+	bch2_trans_iter_put(&trans, dst);
+	bch2_trans_iter_put(&trans, src);
+	bch2_trans_exit(&trans);
+	bch2_bkey_buf_exit(&copy, c);
+
+	if (ret)
+		goto err;
 
 	if (!insert) {
 		i_size_write(&inode->v, new_size);
@@ -2626,8 +2628,6 @@ bkey_err:
 		mutex_unlock(&inode->ei_update_lock);
 	}
 err:
-	bch2_trans_exit(&trans);
-	bch2_bkey_buf_exit(&copy, c);
 	bch2_pagecache_block_put(&inode->ei_pagecache_lock);
 	inode_unlock(&inode->v);
 	return ret;
@@ -2682,7 +2682,7 @@ static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
 			BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 	end_pos = POS(inode->v.i_ino, block_end >> 9);
 
-	while (bkey_cmp(iter->pos, end_pos) < 0) {
+	while (!ret && bkey_cmp(iter->pos, end_pos) < 0) {
 		s64 i_sectors_delta = 0;
 		struct disk_reservation disk_res = { 0 };
 		struct quota_res quota_res = { 0 };
@@ -2746,9 +2746,11 @@ bkey_err:
 		bch2_disk_reservation_put(c, &disk_res);
 		if (ret == -EINTR)
 			ret = 0;
-		if (ret)
-			goto err;
 	}
+	bch2_trans_iter_put(&trans, iter);
+
+	if (ret)
+		goto err;
 
 	/*
 	 * Do we need to extend the file?
@@ -2770,6 +2772,7 @@ bkey_err:
 			ret = PTR_ERR_OR_ZERO(inode_iter);
 		} while (ret == -EINTR);
 
+		bch2_trans_iter_put(&trans, inode_iter);
 		bch2_trans_unlock(&trans);
 
 		if (ret)
@@ -3015,6 +3018,7 @@ static loff_t bch2_seek_data(struct file *file, u64 offset)
 		} else if (k.k->p.offset >> 9 > isize)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
 
 	ret = bch2_trans_exit(&trans) ?: ret;
 	if (ret)
@@ -3118,6 +3122,7 @@ static loff_t bch2_seek_hole(struct file *file, u64 offset)
 			offset = max(offset, bkey_start_offset(k.k) << 9);
 		}
 	}
+	bch2_trans_iter_put(&trans, iter);
 
 	ret = bch2_trans_exit(&trans) ?: ret;
 	if (ret)
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 1fafd393912c..3acda0389da8 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -734,6 +734,8 @@ retry:
 				  BTREE_INSERT_NOUNLOCK|
 				  BTREE_INSERT_NOFAIL);
 btree_err:
+	bch2_trans_iter_put(&trans, inode_iter);
+
 	if (ret == -EINTR)
 		goto retry;
 	if (unlikely(ret))
@@ -961,6 +963,7 @@ retry:
 		ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
 				       FIEMAP_EXTENT_LAST);
 
+	bch2_trans_iter_put(&trans, iter);
 	ret = bch2_trans_exit(&trans) ?: ret;
 	bch2_bkey_buf_exit(&cur, c);
 	bch2_bkey_buf_exit(&prev, c);
diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index 033d37891c60..f8e0b24d087a 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -1485,11 +1485,12 @@ int bch2_fsck_walk_inodes_only(struct bch_fs *c)
 		     BCH_INODE_I_SECTORS_DIRTY|
 		     BCH_INODE_UNLINKED)) {
 			ret = check_inode(&trans, NULL, iter, inode, NULL);
-			BUG_ON(ret == -EINTR);
 			if (ret)
 				break;
 		}
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	BUG_ON(ret == -EINTR);
 
 	return bch2_trans_exit(&trans) ?: ret;
diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index a3d2bae0a652..aec0fc9228a3 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -620,6 +620,7 @@ retry:
 
 	ret = bch2_trans_commit(&trans, NULL, NULL,
 				BTREE_INSERT_NOFAIL);
+	bch2_trans_iter_put(&trans, iter);
 err:
 	if (ret == -EINTR)
 		goto retry;
diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c
index 425502f7b1b8..b841b3da2510 100644
--- a/fs/bcachefs/io.c
+++ b/fs/bcachefs/io.c
@@ -414,6 +414,8 @@ int bch2_fpunch(struct bch_fs *c, u64 inum, u64 start, u64 end,
 
 	ret = bch2_fpunch_at(&trans, iter, POS(inum, end),
 			     journal_seq, i_sectors_delta);
+
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 
 	if (ret == -EINTR)
@@ -460,6 +462,7 @@ int bch2_write_index_default(struct bch_write_op *op)
 			bch2_keylist_pop_front(keys);
 	} while (!bch2_keylist_empty(keys));
 
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	bch2_bkey_buf_exit(&sk, c);
 
@@ -1659,6 +1662,7 @@ retry:
 		goto err;
 out:
 	bch2_rbio_done(rbio);
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	bch2_bkey_buf_exit(&sk, c);
 	return;
@@ -2259,7 +2263,7 @@ retry:
 		k = bch2_btree_iter_peek_slot(iter);
 		ret = bkey_err(k);
 		if (ret)
-			goto err;
+			break;
 
 		offset_into_extent = iter->pos.offset -
 			bkey_start_offset(k.k);
@@ -2270,7 +2274,7 @@ retry:
 		ret = bch2_read_indirect_extent(&trans, &data_btree,
 					&offset_into_extent, &sk);
 		if (ret)
-			goto err;
+			break;
 
 		k = bkey_i_to_s_c(sk.k);
 
@@ -2295,12 +2299,8 @@ retry:
 		ret = __bch2_read_extent(&trans, rbio, bvec_iter, iter->pos,
 					 data_btree, k,
 					 offset_into_extent, failed, flags);
-		switch (ret) {
-		case READ_RETRY:
-			goto retry;
-		case READ_ERR:
-			goto err;
-		};
+		if (ret)
+			break;
 
 		if (flags & BCH_READ_LAST_FRAGMENT)
 			break;
@@ -2308,19 +2308,19 @@ retry:
 		swap(bvec_iter.bi_size, bytes);
 		bio_advance_iter(&rbio->bio, &bvec_iter, bytes);
 	}
-out:
-	bch2_trans_exit(&trans);
-	bch2_bkey_buf_exit(&sk, c);
-	return;
-err:
-	if (ret == -EINTR)
+	bch2_trans_iter_put(&trans, iter);
+
+	if (ret == -EINTR || ret == READ_RETRY || ret == READ_RETRY_AVOID)
 		goto retry;
 
-	bch_err_inum_ratelimited(c, inode,
-				 "read error %i from btree lookup", ret);
-	rbio->bio.bi_status = BLK_STS_IOERR;
-	bch2_rbio_done(rbio);
-	goto out;
+	if (ret) {
+		bch_err_inum_ratelimited(c, inode,
+					 "read error %i from btree lookup", ret);
+		rbio->bio.bi_status = BLK_STS_IOERR;
+		bch2_rbio_done(rbio);
+	}
+	bch2_trans_exit(&trans);
+	bch2_bkey_buf_exit(&sk, c);
 }
 
 void bch2_fs_io_exit(struct bch_fs *c)
diff --git a/fs/bcachefs/migrate.c b/fs/bcachefs/migrate.c
index 1db2c2d6b970..4d8b4169923d 100644
--- a/fs/bcachefs/migrate.c
+++ b/fs/bcachefs/migrate.c
@@ -88,6 +88,7 @@ static int __bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags
 		if (ret)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
 
 	ret = bch2_trans_exit(&trans) ?: ret;
 	bch2_bkey_buf_exit(&sk, c);
@@ -135,20 +136,24 @@ retry:
 					    dev_idx, flags, true);
 			if (ret) {
 				bch_err(c, "Cannot drop device without losing data");
-				goto err;
+				break;
 			}
 
 			ret = bch2_btree_node_update_key(c, iter, b, k.k);
 			if (ret == -EINTR) {
 				b = bch2_btree_iter_peek_node(iter);
+				ret = 0;
 				goto retry;
 			}
 			if (ret) {
 				bch_err(c, "Error updating btree node key: %i", ret);
-				goto err;
+				break;
 			}
 		}
 		bch2_trans_iter_free(&trans, iter);
+
+		if (ret)
+			goto err;
 	}
 
 	/* flush relevant btree updates */
diff --git a/fs/bcachefs/move.c b/fs/bcachefs/move.c
index 3ff31d25f396..f7b0764d9c98 100644
--- a/fs/bcachefs/move.c
+++ b/fs/bcachefs/move.c
@@ -195,6 +195,7 @@ nomatch:
 		goto next;
 	}
 out:
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	bch2_bkey_buf_exit(&_insert, c);
 	bch2_bkey_buf_exit(&_new, c);
@@ -641,6 +642,8 @@ next_nondata:
 		bch2_trans_cond_resched(&trans);
 	}
 out:
+
+	bch2_trans_iter_put(&trans, iter);
 	ret = bch2_trans_exit(&trans) ?: ret;
 	bch2_bkey_buf_exit(&sk, c);
 
diff --git a/fs/bcachefs/quota.c b/fs/bcachefs/quota.c
index 8e272519ce0e..35b409e0f366 100644
--- a/fs/bcachefs/quota.c
+++ b/fs/bcachefs/quota.c
@@ -372,6 +372,7 @@ static int bch2_quota_init_type(struct bch_fs *c, enum quota_types type)
 		if (ret)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
 
 	return bch2_trans_exit(&trans) ?: ret;
 }
@@ -449,6 +450,8 @@ int bch2_fs_quota_read(struct bch_fs *c)
 					KEY_TYPE_QUOTA_NOCHECK);
 		}
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	return bch2_trans_exit(&trans) ?: ret;
 }
 
@@ -739,7 +742,9 @@ static int bch2_set_quota_trans(struct btree_trans *trans,
 	if (qdq->d_fieldmask & QC_INO_HARD)
 		new_quota->v.c[Q_INO].hardlimit = cpu_to_le64(qdq->d_ino_hardlimit);
 
-	return bch2_trans_update(trans, iter, &new_quota->k_i, 0);
+	ret = bch2_trans_update(trans, iter, &new_quota->k_i, 0);
+	bch2_trans_iter_put(trans, iter);
+	return ret;
 }
 
 static int bch2_set_quota(struct super_block *sb, struct kqid qid,
diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c
index e9a6a5f639b4..0978ad92614c 100644
--- a/fs/bcachefs/reflink.c
+++ b/fs/bcachefs/reflink.c
@@ -223,20 +223,18 @@ s64 bch2_remap_range(struct bch_fs *c,
 	dst_iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, dst_start,
 				       BTREE_ITER_INTENT);
 
-	while (1) {
+	while (ret == 0 || ret == -EINTR) {
 		bch2_trans_begin(&trans);
 
-		trans.mem_top = 0;
-
 		if (fatal_signal_pending(current)) {
 			ret = -EINTR;
-			goto err;
+			break;
 		}
 
 		src_k = get_next_src(src_iter, src_end);
 		ret = bkey_err(src_k);
 		if (ret)
-			goto btree_err;
+			continue;
 
 		src_done = bpos_min(src_iter->pos, src_end).offset -
 			src_start.offset;
@@ -245,8 +243,6 @@ s64 bch2_remap_range(struct bch_fs *c,
 		if (bkey_cmp(dst_iter->pos, dst_want) < 0) {
 			ret = bch2_fpunch_at(&trans, dst_iter, dst_want,
 					     journal_seq, i_sectors_delta);
-			if (ret)
-				goto btree_err;
 			continue;
 		}
 
@@ -265,7 +261,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 			ret = bch2_make_extent_indirect(&trans, src_iter,
 						new_src.k);
 			if (ret)
-				goto btree_err;
+				continue;
 
 			BUG_ON(src_k.k->type != KEY_TYPE_reflink_p);
 		}
@@ -294,20 +290,16 @@ s64 bch2_remap_range(struct bch_fs *c,
 					 NULL, journal_seq,
 					 new_i_size, i_sectors_delta);
 		if (ret)
-			goto btree_err;
+			continue;
 
 		dst_done = dst_iter->pos.offset - dst_start.offset;
 		src_want = POS(src_start.inode, src_start.offset + dst_done);
 		bch2_btree_iter_set_pos(src_iter, src_want);
-btree_err:
-		if (ret == -EINTR)
-			ret = 0;
-		if (ret)
-			goto err;
 	}
+	bch2_trans_iter_put(&trans, dst_iter);
+	bch2_trans_iter_put(&trans, src_iter);
 
-	BUG_ON(bkey_cmp(dst_iter->pos, dst_end));
-err:
+	BUG_ON(!ret && bkey_cmp(dst_iter->pos, dst_end));
 	BUG_ON(bkey_cmp(dst_iter->pos, dst_end) > 0);
 
 	dst_done = dst_iter->pos.offset - dst_start.offset;
@@ -329,6 +321,8 @@ err:
 			ret2  = bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 				bch2_trans_commit(&trans, NULL, journal_seq, 0);
 		}
+
+		bch2_trans_iter_put(&trans, inode_iter);
 	} while (ret2 == -EINTR);
 
 	ret = bch2_trans_exit(&trans) ?: ret;
diff --git a/fs/bcachefs/tests.c b/fs/bcachefs/tests.c
index f25a27f26202..286587a118fe 100644
--- a/fs/bcachefs/tests.c
+++ b/fs/bcachefs/tests.c
@@ -67,6 +67,7 @@ static int test_delete(struct bch_fs *c, u64 nr)
 		goto err;
 	}
 err:
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -106,6 +107,7 @@ static int test_delete_written(struct bch_fs *c, u64 nr)
 		goto err;
 	}
 err:
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -113,7 +115,7 @@ err:
 static int test_iterate(struct bch_fs *c, u64 nr)
 {
 	struct btree_trans trans;
-	struct btree_iter *iter;
+	struct btree_iter *iter = NULL;
 	struct bkey_s_c k;
 	u64 i;
 	int ret = 0;
@@ -159,6 +161,7 @@ static int test_iterate(struct bch_fs *c, u64 nr)
 
 	BUG_ON(i);
 err:
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -166,7 +169,7 @@ err:
 static int test_iterate_extents(struct bch_fs *c, u64 nr)
 {
 	struct btree_trans trans;
-	struct btree_iter *iter;
+	struct btree_iter *iter = NULL;
 	struct bkey_s_c k;
 	u64 i;
 	int ret = 0;
@@ -213,6 +216,7 @@ static int test_iterate_extents(struct bch_fs *c, u64 nr)
 
 	BUG_ON(i);
 err:
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -257,7 +261,7 @@ static int test_iterate_slots(struct bch_fs *c, u64 nr)
 		BUG_ON(k.k->p.offset != i);
 		i += 2;
 	}
-	bch2_trans_iter_free(&trans, iter);
+	bch2_trans_iter_put(&trans, iter);
 
 	BUG_ON(i != nr * 2);
 
@@ -274,6 +278,7 @@ static int test_iterate_slots(struct bch_fs *c, u64 nr)
 		if (i == nr * 2)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
 err:
 	bch2_trans_exit(&trans);
 	return ret;
@@ -318,7 +323,7 @@ static int test_iterate_slots_extents(struct bch_fs *c, u64 nr)
 		BUG_ON(k.k->size != 8);
 		i += 16;
 	}
-	bch2_trans_iter_free(&trans, iter);
+	bch2_trans_iter_put(&trans, iter);
 
 	BUG_ON(i != nr);
 
@@ -337,6 +342,7 @@ static int test_iterate_slots_extents(struct bch_fs *c, u64 nr)
 		if (i == nr)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
 err:
 	bch2_trans_exit(&trans);
 	return 0;
@@ -362,6 +368,8 @@ static int test_peek_end(struct bch_fs *c, u64 nr)
 	k = bch2_btree_iter_peek(iter);
 	BUG_ON(k.k);
 
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 	return 0;
 }
@@ -382,6 +390,8 @@ static int test_peek_end_extents(struct bch_fs *c, u64 nr)
 	k = bch2_btree_iter_peek(iter);
 	BUG_ON(k.k);
 
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 	return 0;
 }
@@ -508,7 +518,7 @@ static int rand_lookup(struct bch_fs *c, u64 nr)
 		}
 	}
 
-	bch2_trans_iter_free(&trans, iter);
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -549,7 +559,7 @@ static int rand_mixed(struct bch_fs *c, u64 nr)
 		}
 	}
 
-	bch2_trans_iter_free(&trans, iter);
+	bch2_trans_iter_put(&trans, iter);
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -630,6 +640,8 @@ static int seq_insert(struct bch_fs *c, u64 nr)
 		if (++i == nr)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -645,6 +657,8 @@ static int seq_lookup(struct bch_fs *c, u64 nr)
 
 	for_each_btree_key(&trans, iter, BTREE_ID_xattrs, POS_MIN, 0, k, ret)
 		;
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 	return ret;
 }
@@ -671,6 +685,8 @@ static int seq_overwrite(struct bch_fs *c, u64 nr)
 			break;
 		}
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	bch2_trans_exit(&trans);
 	return ret;
 }
diff --git a/fs/bcachefs/xattr.c b/fs/bcachefs/xattr.c
index 5692b47eb3c9..f18a795620d8 100644
--- a/fs/bcachefs/xattr.c
+++ b/fs/bcachefs/xattr.c
@@ -133,12 +133,9 @@ int bch2_xattr_get(struct bch_fs *c, struct bch_inode_info *inode,
 				inode->v.i_ino,
 				&X_SEARCH(type, name, strlen(name)),
 				0);
-	if (IS_ERR(iter)) {
-		bch2_trans_exit(&trans);
-		BUG_ON(PTR_ERR(iter) == -EINTR);
-
-		return PTR_ERR(iter) == -ENOENT ? -ENODATA : PTR_ERR(iter);
-	}
+	ret = PTR_ERR_OR_ZERO(iter);
+	if (ret)
+		goto err;
 
 	xattr = bkey_s_c_to_xattr(bch2_btree_iter_peek_slot(iter));
 	ret = le16_to_cpu(xattr.v->x_val_len);
@@ -148,9 +145,12 @@ int bch2_xattr_get(struct bch_fs *c, struct bch_inode_info *inode,
 		else
 			memcpy(buffer, xattr_val(xattr.v), ret);
 	}
-
+	bch2_trans_iter_put(&trans, iter);
+err:
 	bch2_trans_exit(&trans);
-	return ret;
+
+	BUG_ON(ret == -EINTR);
+	return ret == -ENOENT ? -ENODATA : ret;
 }
 
 int bch2_xattr_set(struct btree_trans *trans, u64 inum,
@@ -294,6 +294,8 @@ ssize_t bch2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 		if (ret)
 			break;
 	}
+	bch2_trans_iter_put(&trans, iter);
+
 	ret = bch2_trans_exit(&trans) ?: ret;
 
 	if (ret)
-- 
2.20.1