bcachefs: Fix read retry path for indirect extents

In the read path, for retry of indirect extents to work we need to
differentiate between the location in the btree the read was for, vs.
the location where we found the data. This patch adds that plumbing to
bch_read_bio.

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

6 files changed, 62 insertions, 39 deletions

diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 4ccc9318a924..8584b90a3df9 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -788,6 +788,7 @@ retry:
 	while (1) {
 		struct bkey_s_c k;
 		unsigned bytes, sectors, offset_into_extent;
+		enum btree_id data_btree = BTREE_ID_extents;
 
 		bch2_btree_iter_set_pos(iter,
 				POS(inum, rbio->bio.bi_iter.bi_sector));
@@ -803,7 +804,7 @@ retry:
 
 		bch2_bkey_buf_reassemble(&sk, c, k);
 
-		ret = bch2_read_indirect_extent(trans,
+		ret = bch2_read_indirect_extent(trans, &data_btree,
 					&offset_into_extent, &sk);
 		if (ret)
 			break;
@@ -827,7 +828,8 @@ retry:
 		if (bkey_extent_is_allocation(k.k))
 			bch2_add_page_sectors(&rbio->bio, k);
 
-		bch2_read_extent(trans, rbio, k, offset_into_extent, flags);
+		bch2_read_extent(trans, rbio, iter->pos,
+				 data_btree, k, offset_into_extent, flags);
 
 		if (flags & BCH_READ_LAST_FRAGMENT)
 			break;
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index ef8505da7391..1fafd393912c 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -908,6 +908,8 @@ retry:
 	while ((k = bch2_btree_iter_peek(iter)).k &&
 	       !(ret = bkey_err(k)) &&
 	       bkey_cmp(iter->pos, end) < 0) {
+		enum btree_id data_btree = BTREE_ID_extents;
+
 		if (!bkey_extent_is_data(k.k) &&
 		    k.k->type != KEY_TYPE_reservation) {
 			bch2_btree_iter_next(iter);
@@ -920,7 +922,7 @@ retry:
 
 		bch2_bkey_buf_reassemble(&cur, c, k);
 
-		ret = bch2_read_indirect_extent(&trans,
+		ret = bch2_read_indirect_extent(&trans, &data_btree,
 					&offset_into_extent, &cur);
 		if (ret)
 			break;
diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c
index b402fc2e51d6..425502f7b1b8 100644
--- a/fs/bcachefs/io.c
+++ b/fs/bcachefs/io.c
@@ -1627,8 +1627,8 @@ static void bch2_read_retry_nodecode(struct bch_fs *c, struct bch_read_bio *rbio
 	bch2_bkey_buf_init(&sk);
 	bch2_trans_init(&trans, c, 0, 0);
 
-	iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
-				   rbio->pos, BTREE_ITER_SLOTS);
+	iter = bch2_trans_get_iter(&trans, rbio->data_btree,
+				   rbio->read_pos, BTREE_ITER_SLOTS);
 retry:
 	rbio->bio.bi_status = 0;
 
@@ -1642,14 +1642,17 @@ retry:
 
 	if (!bch2_bkey_matches_ptr(c, k,
 				   rbio->pick.ptr,
-				   rbio->pos.offset -
+				   rbio->data_pos.offset -
 				   rbio->pick.crc.offset)) {
 		/* extent we wanted to read no longer exists: */
 		rbio->hole = true;
 		goto out;
 	}
 
-	ret = __bch2_read_extent(&trans, rbio, bvec_iter, k, 0, failed, flags);
+	ret = __bch2_read_extent(&trans, rbio, bvec_iter,
+				 rbio->read_pos,
+				 rbio->data_btree,
+				 k, 0, failed, flags);
 	if (ret == READ_RETRY)
 		goto retry;
 	if (ret)
@@ -1671,7 +1674,7 @@ static void bch2_rbio_retry(struct work_struct *work)
 	struct bch_fs *c	= rbio->c;
 	struct bvec_iter iter	= rbio->bvec_iter;
 	unsigned flags		= rbio->flags;
-	u64 inode		= rbio->pos.inode;
+	u64 inode		= rbio->read_pos.inode;
 	struct bch_io_failures failed = { .nr = 0 };
 
 	trace_read_retry(&rbio->bio);
@@ -1719,7 +1722,7 @@ static int __bch2_rbio_narrow_crcs(struct btree_trans *trans,
 				   struct bch_read_bio *rbio)
 {
 	struct bch_fs *c = rbio->c;
-	u64 data_offset = rbio->pos.offset - rbio->pick.crc.offset;
+	u64 data_offset = rbio->data_pos.offset - rbio->pick.crc.offset;
 	struct bch_extent_crc_unpacked new_crc;
 	struct btree_iter *iter = NULL;
 	struct bkey_i *new;
@@ -1729,7 +1732,7 @@ static int __bch2_rbio_narrow_crcs(struct btree_trans *trans,
 	if (crc_is_compressed(rbio->pick.crc))
 		return 0;
 
-	iter = bch2_trans_get_iter(trans, BTREE_ID_extents, rbio->pos,
+	iter = bch2_trans_get_iter(trans, rbio->data_btree, rbio->data_pos,
 				   BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 	k = bch2_btree_iter_peek_slot(iter);
 	if ((ret = bkey_err(k)))
@@ -1862,14 +1865,14 @@ csum_err:
 		return;
 	}
 
-	bch2_dev_inum_io_error(ca, rbio->pos.inode, (u64) rbio->bvec_iter.bi_sector,
+	bch2_dev_inum_io_error(ca, rbio->read_pos.inode, (u64) rbio->bvec_iter.bi_sector,
 		"data checksum error: expected %0llx:%0llx got %0llx:%0llx (type %u)",
 		rbio->pick.crc.csum.hi, rbio->pick.crc.csum.lo,
 		csum.hi, csum.lo, crc.csum_type);
 	bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
 	return;
 decompression_err:
-	bch_err_inum_ratelimited(c, rbio->pos.inode,
+	bch_err_inum_ratelimited(c, rbio->read_pos.inode,
 				 "decompression error");
 	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
 	return;
@@ -1892,13 +1895,9 @@ static void bch2_read_endio(struct bio *bio)
 	if (!rbio->split)
 		rbio->bio.bi_end_io = rbio->end_io;
 
-	/*
-	 * XXX: rbio->pos is not what we want here when reading from indirect
-	 * extents
-	 */
 	if (bch2_dev_inum_io_err_on(bio->bi_status, ca,
-				    rbio->pos.inode,
-				    rbio->pos.offset,
+				    rbio->read_pos.inode,
+				    rbio->read_pos.offset,
 				    "data read error: %s",
 			       bch2_blk_status_to_str(bio->bi_status))) {
 		bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status);
@@ -1963,7 +1962,8 @@ err:
 }
 
 int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
-		       struct bvec_iter iter, struct bkey_s_c k,
+		       struct bvec_iter iter, struct bpos read_pos,
+		       enum btree_id data_btree, struct bkey_s_c k,
 		       unsigned offset_into_extent,
 		       struct bch_io_failures *failed, unsigned flags)
 {
@@ -1973,7 +1973,7 @@ int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
 	struct bch_dev *ca;
 	struct promote_op *promote = NULL;
 	bool bounce = false, read_full = false, narrow_crcs = false;
-	struct bpos pos = bkey_start_pos(k.k);
+	struct bpos data_pos = bkey_start_pos(k.k);
 	int pick_ret;
 
 	if (bkey_extent_is_inline_data(k.k)) {
@@ -2049,7 +2049,7 @@ int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
 			 pick.crc.offset ||
 			 offset_into_extent));
 
-		pos.offset += offset_into_extent;
+		data_pos.offset += offset_into_extent;
 		pick.ptr.offset += pick.crc.offset +
 			offset_into_extent;
 		offset_into_extent		= 0;
@@ -2123,7 +2123,9 @@ get_bio:
 	/* XXX: only initialize this if needed */
 	rbio->devs_have		= bch2_bkey_devs(k);
 	rbio->pick		= pick;
-	rbio->pos		= pos;
+	rbio->read_pos		= read_pos;
+	rbio->data_btree	= data_btree;
+	rbio->data_pos		= data_pos;
 	rbio->version		= k.k->version;
 	rbio->promote		= promote;
 	INIT_WORK(&rbio->work, NULL);
@@ -2249,6 +2251,7 @@ retry:
 				   BTREE_ITER_SLOTS);
 	while (1) {
 		unsigned bytes, sectors, offset_into_extent;
+		enum btree_id data_btree = BTREE_ID_extents;
 
 		bch2_btree_iter_set_pos(iter,
 				POS(inode, bvec_iter.bi_sector));
@@ -2264,7 +2267,7 @@ retry:
 
 		bch2_bkey_buf_reassemble(&sk, c, k);
 
-		ret = bch2_read_indirect_extent(&trans,
+		ret = bch2_read_indirect_extent(&trans, &data_btree,
 					&offset_into_extent, &sk);
 		if (ret)
 			goto err;
@@ -2289,7 +2292,8 @@ retry:
 		if (bvec_iter.bi_size == bytes)
 			flags |= BCH_READ_LAST_FRAGMENT;
 
-		ret = __bch2_read_extent(&trans, rbio, bvec_iter, k,
+		ret = __bch2_read_extent(&trans, rbio, bvec_iter, iter->pos,
+					 data_btree, k,
 					 offset_into_extent, failed, flags);
 		switch (ret) {
 		case READ_RETRY:
diff --git a/fs/bcachefs/io.h b/fs/bcachefs/io.h
index 1c0a444ea325..ccbd8c3e6642 100644
--- a/fs/bcachefs/io.h
+++ b/fs/bcachefs/io.h
@@ -121,12 +121,15 @@ int __bch2_read_indirect_extent(struct btree_trans *, unsigned *,
 				struct bkey_buf *);
 
 static inline int bch2_read_indirect_extent(struct btree_trans *trans,
+					    enum btree_id *data_btree,
 					    unsigned *offset_into_extent,
 					    struct bkey_buf *k)
 {
-	return k->k->k.type == KEY_TYPE_reflink_p
-		? __bch2_read_indirect_extent(trans, offset_into_extent, k)
-		: 0;
+	if (k->k->k.type != KEY_TYPE_reflink_p)
+		return 0;
+
+	*data_btree = BTREE_ID_reflink;
+	return __bch2_read_indirect_extent(trans, offset_into_extent, k);
 }
 
 enum bch_read_flags {
@@ -143,17 +146,17 @@ enum bch_read_flags {
 };
 
 int __bch2_read_extent(struct btree_trans *, struct bch_read_bio *,
-		       struct bvec_iter, struct bkey_s_c, unsigned,
+		       struct bvec_iter, struct bpos, enum btree_id,
+		       struct bkey_s_c, unsigned,
 		       struct bch_io_failures *, unsigned);
 
 static inline void bch2_read_extent(struct btree_trans *trans,
-				    struct bch_read_bio *rbio,
-				    struct bkey_s_c k,
-				    unsigned offset_into_extent,
-				    unsigned flags)
+			struct bch_read_bio *rbio, struct bpos read_pos,
+			enum btree_id data_btree, struct bkey_s_c k,
+			unsigned offset_into_extent, unsigned flags)
 {
-	__bch2_read_extent(trans, rbio, rbio->bio.bi_iter, k,
-			   offset_into_extent, NULL, flags);
+	__bch2_read_extent(trans, rbio, rbio->bio.bi_iter, read_pos,
+			   data_btree, k, offset_into_extent, NULL, flags);
 }
 
 void __bch2_read(struct bch_fs *, struct bch_read_bio *, struct bvec_iter,
diff --git a/fs/bcachefs/io_types.h b/fs/bcachefs/io_types.h
index 65969eeac253..99b4b4c4a53b 100644
--- a/fs/bcachefs/io_types.h
+++ b/fs/bcachefs/io_types.h
@@ -58,8 +58,18 @@ struct bch_read_bio {
 	struct bch_devs_list	devs_have;
 
 	struct extent_ptr_decoded pick;
-	/* start pos of data we read (may not be pos of data we want) */
-	struct bpos		pos;
+
+	/*
+	 * pos we read from - different from data_pos for indirect extents:
+	 */
+	struct bpos		read_pos;
+
+	/*
+	 * start pos of data we read (may not be pos of data we want) - for
+	 * promote, narrow extents paths:
+	 */
+	enum btree_id		data_btree;
+	struct bpos		data_pos;
 	struct bversion		version;
 
 	struct promote_op	*promote;
diff --git a/fs/bcachefs/move.c b/fs/bcachefs/move.c
index dfe7f05f39e9..3ff31d25f396 100644
--- a/fs/bcachefs/move.c
+++ b/fs/bcachefs/move.c
@@ -208,9 +208,9 @@ void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
 	BUG_ON(!m->op.wbio.bio.bi_vcnt);
 
 	m->ptr		= rbio->pick.ptr;
-	m->offset	= rbio->pos.offset - rbio->pick.crc.offset;
+	m->offset	= rbio->data_pos.offset - rbio->pick.crc.offset;
 	m->op.devs_have	= rbio->devs_have;
-	m->op.pos	= rbio->pos;
+	m->op.pos	= rbio->data_pos;
 	m->op.version	= rbio->version;
 	m->op.crc	= rbio->pick.crc;
 	m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
@@ -492,7 +492,9 @@ static int bch2_move_extent(struct btree_trans *trans,
 	 * ctxt when doing wakeup
 	 */
 	closure_get(&ctxt->cl);
-	bch2_read_extent(trans, &io->rbio, k, 0,
+	bch2_read_extent(trans, &io->rbio,
+			 bkey_start_pos(k.k),
+			 btree_id, k, 0,
 			 BCH_READ_NODECODE|
 			 BCH_READ_LAST_FRAGMENT);
 	return 0;