diff --git a/CHANGELOG.md b/CHANGELOG.md
index d0a9d58b6..6a3c19690 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,8 @@ Both of interleaved and separated metadata are now supported by the malloc bdev
 
 Protection information is now supported by the malloc bdev module.
 
+A new API `spkd_bdev_part_submit_request_ext` was added to specify a custom completion callback.
+
 ### scheduler
 
 Changing scheduler from dynamic back to static is no longer possible,
diff --git a/include/spdk/bdev_module.h b/include/spdk/bdev_module.h
index 8e13e0b39..d598d05d6 100644
--- a/include/spdk/bdev_module.h
+++ b/include/spdk/bdev_module.h
@@ -1287,6 +1287,25 @@ int spdk_bdev_part_construct(struct spdk_bdev_part *part, struct spdk_bdev_part_
  */
 int spdk_bdev_part_submit_request(struct spdk_bdev_part_channel *ch, struct spdk_bdev_io *bdev_io);
 
+/**
+ * Forwards I/O from an spdk_bdev_part to the underlying base bdev.
+ *
+ * This function will apply the offset_blocks the user provided to
+ * spdk_bdev_part_construct to the I/O. The user should not manually
+ * apply this offset before submitting any I/O through this function.
+ *
+ * This function enables user to specify a completion callback. It is required that
+ * the completion callback calls spdk_bdev_io_complete() for the forwarded I/O.
+ *
+ * \param ch The I/O channel associated with the spdk_bdev_part.
+ * \param bdev_io The I/O to be submitted to the underlying bdev.
+ * \param cb Called when the forwarded I/O completes.
+ * \return 0 on success or non-zero if submit request failed.
+ */
+int spdk_bdev_part_submit_request_ext(struct spdk_bdev_part_channel *ch,
+				      struct spdk_bdev_io *bdev_io,
+				      spdk_bdev_io_completion_cb cb);
+
 /**
  * Return a pointer to this part's spdk_bdev.
  *
diff --git a/lib/bdev/part.c b/lib/bdev/part.c
index f9fda7fcc..c186eb530 100644
--- a/lib/bdev/part.c
+++ b/lib/bdev/part.c
@@ -238,6 +238,7 @@ bdev_part_complete_io(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
 {
 	struct spdk_bdev_io *part_io = cb_arg;
 	uint32_t offset, remapped_offset;
+	spdk_bdev_io_completion_cb cb;
 	int rc, status;
 
 	switch (bdev_io->type) {
@@ -260,14 +261,22 @@ bdev_part_complete_io(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
 		break;
 	}
 
-	status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
 
-	spdk_bdev_io_complete(part_io, status);
+	cb = part_io->u.bdev.stored_user_cb;
+	if (cb != NULL) {
+		cb(part_io, success, NULL);
+	} else {
+		status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
+
+		spdk_bdev_io_complete(part_io, status);
+	}
+
 	spdk_bdev_free_io(bdev_io);
 }
 
 int
-spdk_bdev_part_submit_request(struct spdk_bdev_part_channel *ch, struct spdk_bdev_io *bdev_io)
+spdk_bdev_part_submit_request_ext(struct spdk_bdev_part_channel *ch, struct spdk_bdev_io *bdev_io,
+				  spdk_bdev_io_completion_cb cb)
 {
 	struct spdk_bdev_part *part = ch->part;
 	struct spdk_io_channel *base_ch = ch->base_ch;
@@ -275,6 +284,8 @@ spdk_bdev_part_submit_request(struct spdk_bdev_part_channel *ch, struct spdk_bde
 	uint64_t offset, remapped_offset, remapped_src_offset;
 	int rc = 0;
 
+	bdev_io->u.bdev.stored_user_cb = cb;
+
 	offset = bdev_io->u.bdev.offset_blocks;
 	remapped_offset = offset + part->internal.offset_blocks;
 
@@ -382,6 +393,12 @@ spdk_bdev_part_submit_request(struct spdk_bdev_part_channel *ch, struct spdk_bde
 	return rc;
 }
 
+int
+spdk_bdev_part_submit_request(struct spdk_bdev_part_channel *ch, struct spdk_bdev_io *bdev_io)
+{
+	return spdk_bdev_part_submit_request_ext(ch, bdev_io, NULL);
+}
+
 static int
 bdev_part_channel_create_cb(void *io_device, void *ctx_buf)
 {
diff --git a/lib/bdev/spdk_bdev.map b/lib/bdev/spdk_bdev.map
index 8e708e59f..c04cc7dc6 100644
--- a/lib/bdev/spdk_bdev.map
+++ b/lib/bdev/spdk_bdev.map
@@ -149,6 +149,7 @@
 	spdk_bdev_part_base_construct_ext;
 	spdk_bdev_part_construct;
 	spdk_bdev_part_submit_request;
+	spdk_bdev_part_submit_request_ext;
 	spdk_bdev_part_get_bdev;
 	spdk_bdev_part_get_base;
 	spdk_bdev_part_get_base_bdev;