Re: [RFC PATCH v5 2/4] ocfs2: switch DIO read path from buffer_head to iomap

From: Heming Zhao

Date: Fri Nov 06 2026 - 09:18:16 EST


On 11/6/26 18:48, Christoph Hellwig wrote:
On Thu, Nov 05, 2026 at 03:19:26PM +0800, Heming Zhao wrote:
Migrates the DIO read path from the legacy buffer_head infrastructure
to the modern and more efficient iomap framework.

Under the current code, ocfs2 holds a cluster lock during DIO via
ocfs2_iocb_set_rw_locked(), and releases the cluster lock in
ocfs2_dio_end_io() via ocfs2_iocb_clear_rw_locked(). The lock level is
stashed in iocb->private.

However, iomap DIO owns that field (it stores a bio there for polled I/O
and unconditionally clears it before calling ->end_io). Any side-channel
state packed into iocb->private is corrupted, which trips the end_io
lock-state check.

Instead, encode the unlock level by passing distinct iomap_dio_ops methods
(ocfs2_iomap_dio_ops_pr for PRMODE or ocfs2_iomap_dio_ops_ex for EXMODE).
Additionally, the *_iter() functions use the split __iomap_dio_rw() +
iomap_dio_complete() API so they can determine unambiguously, from the
return value, whether the completion handler already dropped the lock
(real I/O, sync or -EIOCBQUEUED) or whether the caller must drop it
(no I/O issued, or buffered fallback).

The new function ocfs2_iomap_begin() performs a similar role to
ocfs2_get_block(). While ocfs2_get_block() retrieves only one block per
call, ocfs2_iomap_begin() can handle multiple blocks at once. Limit the
map offset and length in ocfs2_iomap_begin() to OCFS2_MAX_LOGICAL_BLOCK
to avoid overflow issues.

Co-developed-by: Joseph Qi <joseph.qi@xxxxxxxxxxxxxxxxx>
Signed-off-by: Joseph Qi <joseph.qi@xxxxxxxxxxxxxxxxx>
Signed-off-by: Heming Zhao <heming.zhao@xxxxxxxx>
---
fs/ocfs2/Kconfig | 1 +
fs/ocfs2/aops.c | 111 ++++++++++++++++++++++++++++++++++++++++++++
fs/ocfs2/aops.h | 4 ++
fs/ocfs2/file.c | 62 +++++++++++++++++++++----
fs/ocfs2/ocfs2_fs.h | 3 ++
5 files changed, 172 insertions(+), 9 deletions(-)

diff --git a/fs/ocfs2/Kconfig b/fs/ocfs2/Kconfig
index 2514d36cbe01..bf1678a5eb01 100644
--- a/fs/ocfs2/Kconfig
+++ b/fs/ocfs2/Kconfig
@@ -7,6 +7,7 @@ config OCFS2_FS
select CRC32
select QUOTA
select QUOTA_TREE
+ select FS_IOMAP
select FS_POSIX_ACL
select LEGACY_DIRECT_IO
help
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 0f7502de9b8d..8e4d31cb85db 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -4,6 +4,7 @@
*/
#include <linux/fs.h>
+#include <linux/iomap.h>
#include <linux/slab.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>
@@ -2534,6 +2535,116 @@ static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
ocfs2_dio_end_io, 0);
}
+static int ocfs2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
+ unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
+{
+ int ret;
+ struct ocfs2_map_block map;
+ struct ocfs2_inode_info *oi = OCFS2_I(inode);
+ u8 blkbits = inode->i_blkbits;
+
+ if ((offset >> blkbits) > OCFS2_MAX_LOGICAL_BLOCK)
+ return -EINVAL;
+
+ /*
+ * Calculate the first and last logical blocks respectively.
+ */
+ map.lblk = offset >> blkbits;
+ map.len = min_t(loff_t, (offset + length - 1) >> blkbits,
+ OCFS2_MAX_LOGICAL_BLOCK) - map.lblk + 1;
+ map.flags = 0;
+
+ if (flags & IOMAP_WRITE) {
+ return -EOPNOTSUPP;
+ } else {

No need for an else after a return..

OK.


+ /*
+ * Before returning to iomap, let's ensure the allocated mapping
+ * covers the entire requested length for atomic writes.
+ */
+ if (flags & IOMAP_ATOMIC) {

IOMAP_ATOMIC is never set for reads.


Thanks for your review, I will remove it on next version.

Thanks,
Heming