[Qemu-devel] [PATCH v2 14/20] nbd: Support BDRV_REQ_FUA

[Kevin Wolf <kwolf@redhat.com>]

The NBD server already used to send a FUA flag when the writethrough
mode was set. This code was a remnant from the times where protocol
drivers actually had to implement writethrough modes. Since nowadays the
block layer sends flushes in writethrough mode and non-root nodes are
always writeback, this was mostly dead code - only mostly because if NBD
was configured to be used without a format, we sent _both_ FUA and an
explicit flush afterwards, which makes the code not technically dead,
but useless overhead.

This patch changes the code so that the block layer's FUA flag is
recognised and translated into a NBD FUA flag. The additional flush is
avoided now.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 block/nbd-client.c | 13 +++++++------
 block/nbd-client.h |  2 +-
 block/nbd.c        | 27 ++++++++++++++++++++++++++-
 3 files changed, 34 insertions(+), 8 deletions(-)

diff --git a/block/nbd-client.c b/block/nbd-client.c
index 6a9b4c7..021a88b 100644
--- a/block/nbd-client.c
+++ b/block/nbd-client.c
@@ -243,15 +243,15 @@ static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 
 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
                            int nb_sectors, QEMUIOVector *qiov,
-                           int offset)
+                           int offset, int *flags)
 {
     NbdClientSession *client = nbd_get_client_session(bs);
     struct nbd_request request = { .type = NBD_CMD_WRITE };
     struct nbd_reply reply;
     ssize_t ret;
 
-    if (!bdrv_enable_write_cache(bs) &&
-        (client->nbdflags & NBD_FLAG_SEND_FUA)) {
+    if ((*flags & BDRV_REQ_FUA) && (client->nbdflags & NBD_FLAG_SEND_FUA)) {
+        *flags &= ~BDRV_REQ_FUA;
         request.type |= NBD_CMD_FLAG_FUA;
     }
 
@@ -291,12 +291,13 @@ int nbd_client_co_readv(BlockDriverState *bs, int64_t sector_num,
 }
 
 int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num,
-                         int nb_sectors, QEMUIOVector *qiov)
+                         int nb_sectors, QEMUIOVector *qiov, int *flags)
 {
     int offset = 0;
     int ret;
     while (nb_sectors > NBD_MAX_SECTORS) {
-        ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
+        ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset,
+                              flags);
         if (ret < 0) {
             return ret;
         }
@@ -304,7 +305,7 @@ int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num,
         sector_num += NBD_MAX_SECTORS;
         nb_sectors -= NBD_MAX_SECTORS;
     }
-    return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
+    return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset, flags);
 }
 
 int nbd_client_co_flush(BlockDriverState *bs)
diff --git a/block/nbd-client.h b/block/nbd-client.h
index 53f116d..bc7aec0 100644
--- a/block/nbd-client.h
+++ b/block/nbd-client.h
@@ -48,7 +48,7 @@ int nbd_client_co_discard(BlockDriverState *bs, int64_t sector_num,
                           int nb_sectors);
 int nbd_client_co_flush(BlockDriverState *bs);
 int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num,
-                         int nb_sectors, QEMUIOVector *qiov);
+                         int nb_sectors, QEMUIOVector *qiov, int *flags);
 int nbd_client_co_readv(BlockDriverState *bs, int64_t sector_num,
                         int nb_sectors, QEMUIOVector *qiov);
 
diff --git a/block/nbd.c b/block/nbd.c
index 836424c..7b1eb5f 100644
--- a/block/nbd.c
+++ b/block/nbd.c
@@ -354,10 +354,29 @@ static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
     return nbd_client_co_readv(bs, sector_num, nb_sectors, qiov);
 }
 
+static int nbd_co_writev_flags(BlockDriverState *bs, int64_t sector_num,
+                               int nb_sectors, QEMUIOVector *qiov, int flags)
+{
+    int ret;
+
+    ret = nbd_client_co_writev(bs, sector_num, nb_sectors, qiov, &flags);
+    if (ret < 0) {
+        return ret;
+    }
+
+    /* The flag wasn't sent to the server, so we need to emulate it with an
+     * explicit flush */
+    if (flags & BDRV_REQ_FUA) {
+        ret = nbd_client_co_flush(bs);
+    }
+
+    return ret;
+}
+
 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
                          int nb_sectors, QEMUIOVector *qiov)
 {
-    return nbd_client_co_writev(bs, sector_num, nb_sectors, qiov);
+    return nbd_co_writev_flags(bs, sector_num, nb_sectors, qiov, 0);
 }
 
 static int nbd_co_flush(BlockDriverState *bs)
@@ -457,6 +476,8 @@ static BlockDriver bdrv_nbd = {
     .bdrv_file_open             = nbd_open,
     .bdrv_co_readv              = nbd_co_readv,
     .bdrv_co_writev             = nbd_co_writev,
+    .bdrv_co_writev_flags       = nbd_co_writev_flags,
+    .supported_write_flags      = BDRV_REQ_FUA,
     .bdrv_close                 = nbd_close,
     .bdrv_co_flush_to_os        = nbd_co_flush,
     .bdrv_co_discard            = nbd_co_discard,
@@ -475,6 +496,8 @@ static BlockDriver bdrv_nbd_tcp = {
     .bdrv_file_open             = nbd_open,
     .bdrv_co_readv              = nbd_co_readv,
     .bdrv_co_writev             = nbd_co_writev,
+    .bdrv_co_writev_flags       = nbd_co_writev_flags,
+    .supported_write_flags      = BDRV_REQ_FUA,
     .bdrv_close                 = nbd_close,
     .bdrv_co_flush_to_os        = nbd_co_flush,
     .bdrv_co_discard            = nbd_co_discard,
@@ -493,6 +516,8 @@ static BlockDriver bdrv_nbd_unix = {
     .bdrv_file_open             = nbd_open,
     .bdrv_co_readv              = nbd_co_readv,
     .bdrv_co_writev             = nbd_co_writev,
+    .bdrv_co_writev_flags       = nbd_co_writev_flags,
+    .supported_write_flags      = BDRV_REQ_FUA,
     .bdrv_close                 = nbd_close,
     .bdrv_co_flush_to_os        = nbd_co_flush,
     .bdrv_co_discard            = nbd_co_discard,
-- 
1.8.3.1