bdev: add function to return aio's errno

This helps user to locate whether bdev_io fails in
spdk bdev layer or inside Linux AIO.
SPDK_BDEV_IO_STATUS_AIO_ERROR indicates bdev_io fails
due to Linux AIO or its lower layer's failure.

New functions spdk_bdev_io_complete_aio_status and
spdk_bdev_io_get_aio_status can be used to report out
the errno from Linux AIO.

Change-Id: I32640e4a0459cca057278c02ea5a7522f3408a02
Signed-off-by: Liu Xiaodong <xiaodong.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5690
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>

CHANGELOG.md

@@ -15,6 +15,11 @@ examine process. Along with corresponding `bdev_wait_for_examine` RPC, which
 is now always called during `spdk_bdev_subsystem_config_json` making sure
 every bdev is ready to be used.
 
+A new API `spdk_bdev_io_get_aio_status` was added for getting the status of
+bdev_io as Linux AIO errno. Also `spdk_bdev_io_complete_aio_status` function
+and `SPDK_BDEV_IO_STATUS_AIO_ERROR` were added for bdev module to complete
+a bdev_io with Linux AIO errno.
+
 ### blob
 
 An `opts_size` element was added in the `spdk_bs_opts` structure to solve the

include/spdk/bdev.h

@@ -1678,6 +1678,14 @@ void spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint
 void spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
 				  int *sc, int *sk, int *asc, int *ascq);
 
+/**
+ * Get the status of bdev_io as aio errno.
+ *
+ * \param bdev_io I/O to get the status from.
+ * \param aio_result Negative errno returned from AIO.
+ */
+void spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result);
+
 /**
  * Get the iovec describing the data buffer of a bdev_io.
  *

include/spdk/bdev_module.h

@@ -226,6 +226,7 @@ struct spdk_bdev_fn_table {
 
 /** bdev I/O completion status */
 enum spdk_bdev_io_status {
+	SPDK_BDEV_IO_STATUS_AIO_ERROR = -8,
 	SPDK_BDEV_IO_STATUS_ABORTED = -7,
 	SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED = -6,
 	SPDK_BDEV_IO_STATUS_MISCOMPARE = -5,
@@ -643,6 +644,8 @@ struct spdk_bdev_io {
 				/** SCSI additional sense code qualifier */
 				uint8_t ascq;
 			} scsi;
+			/** Only valid when status is SPDK_BDEV_IO_STATUS_AIO_ERROR */
+			int aio_result;
 		} error;
 
 		/**
@@ -934,6 +937,14 @@ void spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cd
 void spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
 				       enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq);
 
+/**
+ * Complete a bdev_io with AIO errno.
+ *
+ * \param bdev_io I/O to complete.
+ * \param aio_result Negative errno returned from AIO.
+ */
+void spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result);
+
 /**
  * Get a thread that given bdev_io was submitted on.
  *

lib/bdev/bdev.c

@@ -5278,6 +5278,34 @@ spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
 	}
 }
 
+void
+spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
+{
+	if (aio_result == 0) {
+		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
+	} else {
+		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
+	}
+
+	bdev_io->internal.error.aio_result = aio_result;
+
+	spdk_bdev_io_complete(bdev_io, bdev_io->internal.status);
+}
+
+void
+spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
+{
+	assert(aio_result != NULL);
+
+	if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
+		*aio_result = bdev_io->internal.error.aio_result;
+	} else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
+		*aio_result = 0;
+	} else {
+		*aio_result = -EIO;
+	}
+}
+
 void
 spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
 {

lib/bdev/spdk_bdev.map

@@ -87,6 +87,7 @@
 	spdk_bdev_io_get_nvme_status;
 	spdk_bdev_io_get_nvme_fused_status;
 	spdk_bdev_io_get_scsi_status;
+	spdk_bdev_io_get_aio_status;
 	spdk_bdev_io_get_iovec;
 	spdk_bdev_io_get_md_buf;
 	spdk_bdev_io_get_cb_arg;
@@ -116,6 +117,7 @@
 	spdk_bdev_io_complete;
 	spdk_bdev_io_complete_nvme_status;
 	spdk_bdev_io_complete_scsi_status;
+	spdk_bdev_io_complete_aio_status;
 	spdk_bdev_io_get_thread;
 	spdk_bdev_io_get_io_channel;
 	spdk_bdev_notify_blockcnt_change;

module/bdev/aio/bdev_aio.c

@@ -191,7 +191,7 @@ bdev_aio_readv(struct file_disk *fdisk, struct spdk_io_channel *ch,
 		if (rc == -EAGAIN) {
 			spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), SPDK_BDEV_IO_STATUS_NOMEM);
 		} else {
-			spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), SPDK_BDEV_IO_STATUS_FAILED);
+			spdk_bdev_io_complete_aio_status(spdk_bdev_io_from_ctx(aio_task), rc);
 			SPDK_ERRLOG("%s: io_submit returned %d\n", __func__, rc);
 		}
 		return -1;
@@ -225,7 +225,7 @@ bdev_aio_writev(struct file_disk *fdisk, struct spdk_io_channel *ch,
 		if (rc == -EAGAIN) {
 			spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), SPDK_BDEV_IO_STATUS_NOMEM);
 		} else {
-			spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), SPDK_BDEV_IO_STATUS_FAILED);
+			spdk_bdev_io_complete_aio_status(spdk_bdev_io_from_ctx(aio_task), rc);
 			SPDK_ERRLOG("%s: io_submit returned %d\n", __func__, rc);
 		}
 		return -1;
@@ -239,8 +239,11 @@ bdev_aio_flush(struct file_disk *fdisk, struct bdev_aio_task *aio_task)
 {
 	int rc = fsync(fdisk->fd);
 
-	spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task),
+	if (rc == 0) {
-			      rc == 0 ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
+		spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), SPDK_BDEV_IO_STATUS_SUCCESS);
+	} else {
+		spdk_bdev_io_complete_aio_status(spdk_bdev_io_from_ctx(aio_task), -errno);
+	}
 }
 
 static int
@@ -326,9 +329,9 @@ static int
 bdev_aio_io_channel_poll(struct bdev_aio_io_channel *io_ch)
 {
 	int nr, i = 0;
-	enum spdk_bdev_io_status status;
 	struct bdev_aio_task *aio_task;
 	struct io_event events[SPDK_AIO_QUEUE_DEPTH];
+	uint64_t io_result;
 
 	nr = bdev_user_io_getevents(io_ch->io_ctx, SPDK_AIO_QUEUE_DEPTH, events);
 
@@ -336,16 +339,23 @@ bdev_aio_io_channel_poll(struct bdev_aio_io_channel *io_ch)
 		return 0;
 	}
 
+#define MAX_AIO_ERRNO 256
 	for (i = 0; i < nr; i++) {
 		aio_task = events[i].data;
-		if (events[i].res != aio_task->len) {
-			status = SPDK_BDEV_IO_STATUS_FAILED;
-		} else {
-			status = SPDK_BDEV_IO_STATUS_SUCCESS;
-		}
-
 		aio_task->ch->io_inflight--;
-		spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), status);
+		io_result = events[i].res;
+		if (io_result == aio_task->len) {
+			spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), SPDK_BDEV_IO_STATUS_SUCCESS);
+		} else if (io_result < MAX_AIO_ERRNO) {
+			/* Linux AIO will return its errno to io_event.res */
+			int aio_errno = io_result;
+
+			spdk_bdev_io_complete_aio_status(spdk_bdev_io_from_ctx(aio_task), -aio_errno);
+		} else {
+			SPDK_ERRLOG("failed to complete aio: requested len is %lu, but completed len is %lu.\n",
+				    aio_task->len, io_result);
+			spdk_bdev_io_complete(spdk_bdev_io_from_ctx(aio_task), SPDK_BDEV_IO_STATUS_FAILED);
+		}
 	}
 
 	return nr;