[Qemu-devel] [PATCH 4/6] file-posix: Support BDRV_REQ_NO_FALLBACK for zero writes

[Kevin Wolf]

We know that the kernel implements a slow fallback code path for
BLKZEROOUT, so if BDRV_REQ_NO_FALLBACK is given, we shouldn't call it.
The other operations we call in the context of .bdrv_co_pwrite_zeroes
should usually be quick, so no modification should be needed for them.
If we ever notice that there are additional problematic cases, we can
still make these conditional as well.

Signed-off-by: Kevin Wolf <address@hidden>
---
 include/block/raw-aio.h |  1 +
 block/file-posix.c      | 24 ++++++++++++++++--------
 2 files changed, 17 insertions(+), 8 deletions(-)

diff --git a/include/block/raw-aio.h b/include/block/raw-aio.h
index 6799614e56..ba223dd1f1 100644
--- a/include/block/raw-aio.h
+++ b/include/block/raw-aio.h
@@ -40,6 +40,7 @@
 /* AIO flags */
 #define QEMU_AIO_MISALIGNED   0x1000
 #define QEMU_AIO_BLKDEV       0x2000
+#define QEMU_AIO_NO_FALLBACK  0x4000
 
 
 /* linux-aio.c - Linux native implementation */
diff --git a/block/file-posix.c b/block/file-posix.c
index d102f3b222..db4cccbe51 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -652,7 +652,7 @@ static int raw_open_common(BlockDriverState *bs, QDict 
*options,
     }
 #endif
 
-    bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
+    bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK;
     ret = 0;
 fail:
     if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
@@ -1500,14 +1500,19 @@ static ssize_t 
handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
     }
 
 #ifdef BLKZEROOUT
-    do {
-        uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
-        if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
-            return 0;
-        }
-    } while (errno == EINTR);
+    /* The BLKZEROOUT implementation in the kernel doesn't set
+     * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
+     * fallbacks. */
+    if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) {
+        do {
+            uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
+            if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
+                return 0;
+            }
+        } while (errno == EINTR);
 
-    ret = translate_err(-errno);
+        ret = translate_err(-errno);
+    }
 #endif
 
     if (ret == -ENOTSUP) {
@@ -2659,6 +2664,9 @@ raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t 
offset, int bytes,
     if (blkdev) {
         acb.aio_type |= QEMU_AIO_BLKDEV;
     }
+    if (flags & BDRV_REQ_NO_FALLBACK) {
+        acb.aio_type |= QEMU_AIO_NO_FALLBACK;
+    }
 
     if (flags & BDRV_REQ_MAY_UNMAP) {
         acb.aio_type |= QEMU_AIO_DISCARD;
-- 
2.20.1