push down vector linearization to posix-aio-compat.c (Christoph Hellwig)

Make all AIO requests vectored and defer linearization until the actual
I/O thread.  This prepares for using native preadv/pwritev.

Also enables asynchronous direct I/O by handling that case in the I/O thread.

Qcow and qcow2 propably want to be adopted to directly deal with multi-segment
requests, but that can be implemented later.


Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>


git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@7020 c046a42c-6fe2-441c-8c8c-71466251a162

7 files changed, 317 insertions, 284 deletions

diff --git a/block-qcow.c b/block-qcow.c
index 2decd13e76..b60f4c1921 100644
--- a/block-qcow.c
+++ b/block-qcow.c
@@ -525,7 +525,9 @@ static int qcow_write(BlockDriverState *bs, int64_t sector_num,
 typedef struct QCowAIOCB {
     BlockDriverAIOCB common;
     int64_t sector_num;
+    QEMUIOVector *qiov;
     uint8_t *buf;
+    void *orig_buf;
     int nb_sectors;
     int n;
     uint64_t cluster_offset;
@@ -543,12 +545,8 @@ static void qcow_aio_read_cb(void *opaque, int ret)
     int index_in_cluster;
 
     acb->hd_aiocb = NULL;
-    if (ret < 0) {
-    fail:
-        acb->common.cb(acb->common.opaque, ret);
-        qemu_aio_release(acb);
-        return;
-    }
+    if (ret < 0)
+        goto done;
 
  redo:
     /* post process the read buffer */
@@ -570,9 +568,8 @@ static void qcow_aio_read_cb(void *opaque, int ret)
 
     if (acb->nb_sectors == 0) {
         /* request completed */
-        acb->common.cb(acb->common.opaque, 0);
-        qemu_aio_release(acb);
-        return;
+        ret = 0;
+        goto done;
     }
 
     /* prepare next AIO request */
@@ -592,7 +589,7 @@ static void qcow_aio_read_cb(void *opaque, int ret)
             acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
                 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
             if (acb->hd_aiocb == NULL)
-                goto fail;
+                goto done;
         } else {
             /* Note: in this case, no need to wait */
             memset(acb->buf, 0, 512 * acb->n);
@@ -601,14 +598,14 @@ static void qcow_aio_read_cb(void *opaque, int ret)
     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
         /* add AIO support for compressed blocks ? */
         if (decompress_cluster(s, acb->cluster_offset) < 0)
-            goto fail;
+            goto done;
         memcpy(acb->buf,
                s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
         goto redo;
     } else {
         if ((acb->cluster_offset & 511) != 0) {
             ret = -EIO;
-            goto fail;
+            goto done;
         }
         acb->hd_iov.iov_base = acb->buf;
         acb->hd_iov.iov_len = acb->n * 512;
@@ -617,12 +614,22 @@ static void qcow_aio_read_cb(void *opaque, int ret)
                             (acb->cluster_offset >> 9) + index_in_cluster,
                             &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
         if (acb->hd_aiocb == NULL)
-            goto fail;
+            goto done;
+    }
+
+    return;
+
+done:
+    if (acb->qiov->niov > 1) {
+        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
+        qemu_vfree(acb->orig_buf);
     }
+    acb->common.cb(acb->common.opaque, ret);
+    qemu_aio_release(acb);
 }
 
-static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
     QCowAIOCB *acb;
@@ -632,7 +639,11 @@ static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
         return NULL;
     acb->hd_aiocb = NULL;
     acb->sector_num = sector_num;
-    acb->buf = buf;
+    acb->qiov = qiov;
+    if (qiov->niov > 1)
+        acb->buf = acb->orig_buf = qemu_memalign(512, qiov->size);
+    else
+        acb->buf = qiov->iov->iov_base;
     acb->nb_sectors = nb_sectors;
     acb->n = 0;
     acb->cluster_offset = 0;
@@ -652,12 +663,8 @@ static void qcow_aio_write_cb(void *opaque, int ret)
 
     acb->hd_aiocb = NULL;
 
-    if (ret < 0) {
-    fail:
-        acb->common.cb(acb->common.opaque, ret);
-        qemu_aio_release(acb);
-        return;
-    }
+    if (ret < 0)
+        goto done;
 
     acb->nb_sectors -= acb->n;
     acb->sector_num += acb->n;
@@ -665,9 +672,8 @@ static void qcow_aio_write_cb(void *opaque, int ret)
 
     if (acb->nb_sectors == 0) {
         /* request completed */
-        acb->common.cb(acb->common.opaque, 0);
-        qemu_aio_release(acb);
-        return;
+        ret = 0;
+        goto done;
     }
 
     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
@@ -679,14 +685,14 @@ static void qcow_aio_write_cb(void *opaque, int ret)
                                         index_in_cluster + acb->n);
     if (!cluster_offset || (cluster_offset & 511) != 0) {
         ret = -EIO;
-        goto fail;
+        goto done;
     }
     if (s->crypt_method) {
         if (!acb->cluster_data) {
             acb->cluster_data = qemu_mallocz(s->cluster_size);
             if (!acb->cluster_data) {
                 ret = -ENOMEM;
-                goto fail;
+                goto done;
             }
         }
         encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
@@ -704,11 +710,18 @@ static void qcow_aio_write_cb(void *opaque, int ret)
                                     &acb->hd_qiov, acb->n,
                                     qcow_aio_write_cb, acb);
     if (acb->hd_aiocb == NULL)
-        goto fail;
+        goto done;
+    return;
+
+done:
+    if (acb->qiov->niov > 1)
+        qemu_vfree(acb->orig_buf);
+    acb->common.cb(acb->common.opaque, ret);
+    qemu_aio_release(acb);
 }
 
-static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
     BDRVQcowState *s = bs->opaque;
@@ -721,7 +734,12 @@ static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
         return NULL;
     acb->hd_aiocb = NULL;
     acb->sector_num = sector_num;
-    acb->buf = (uint8_t *)buf;
+    acb->qiov = qiov;
+    if (qiov->niov > 1) {
+        acb->buf = acb->orig_buf = qemu_memalign(512, qiov->size);
+        qemu_iovec_to_buffer(qiov, acb->buf);
+    } else
+        acb->buf = qiov->iov->iov_base;
     acb->nb_sectors = nb_sectors;
     acb->n = 0;
 
@@ -909,8 +927,8 @@ BlockDriver bdrv_qcow = {
     .bdrv_is_allocated	= qcow_is_allocated,
     .bdrv_set_key	= qcow_set_key,
     .bdrv_make_empty	= qcow_make_empty,
-    .bdrv_aio_read	= qcow_aio_read,
-    .bdrv_aio_write	= qcow_aio_write,
+    .bdrv_aio_readv	= qcow_aio_readv,
+    .bdrv_aio_writev	= qcow_aio_writev,
     .bdrv_aio_cancel	= qcow_aio_cancel,
     .aiocb_size		= sizeof(QCowAIOCB),
     .bdrv_write_compressed = qcow_write_compressed,
diff --git a/block-qcow2.c b/block-qcow2.c
index dd28c28e5f..3bd38b0d9d 100644
--- a/block-qcow2.c
+++ b/block-qcow2.c
@@ -1264,7 +1264,9 @@ static int qcow_write(BlockDriverState *bs, int64_t sector_num,
 typedef struct QCowAIOCB {
     BlockDriverAIOCB common;
     int64_t sector_num;
+    QEMUIOVector *qiov;
     uint8_t *buf;
+    void *orig_buf;
     int nb_sectors;
     int n;
     uint64_t cluster_offset;
@@ -1307,12 +1309,8 @@ static void qcow_aio_read_cb(void *opaque, int ret)
     int index_in_cluster, n1;
 
     acb->hd_aiocb = NULL;
-    if (ret < 0) {
-fail:
-        acb->common.cb(acb->common.opaque, ret);
-        qemu_aio_release(acb);
-        return;
-    }
+    if (ret < 0)
+        goto done;
 
     /* post process the read buffer */
     if (!acb->cluster_offset) {
@@ -1333,9 +1331,8 @@ fail:
 
     if (acb->nb_sectors == 0) {
         /* request completed */
-        acb->common.cb(acb->common.opaque, 0);
-        qemu_aio_release(acb);
-        return;
+        ret = 0;
+        goto done;
     }
 
     /* prepare next AIO request */
@@ -1356,32 +1353,32 @@ fail:
                                     &acb->hd_qiov, acb->n,
 				    qcow_aio_read_cb, acb);
                 if (acb->hd_aiocb == NULL)
-                    goto fail;
+                    goto done;
             } else {
                 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
                 if (ret < 0)
-                    goto fail;
+                    goto done;
             }
         } else {
             /* Note: in this case, no need to wait */
             memset(acb->buf, 0, 512 * acb->n);
             ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
             if (ret < 0)
-                goto fail;
+                goto done;
         }
     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
         /* add AIO support for compressed blocks ? */
         if (decompress_cluster(s, acb->cluster_offset) < 0)
-            goto fail;
+            goto done;
         memcpy(acb->buf,
                s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
         ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
         if (ret < 0)
-            goto fail;
+            goto done;
     } else {
         if ((acb->cluster_offset & 511) != 0) {
             ret = -EIO;
-            goto fail;
+            goto done;
         }
 
         acb->hd_iov.iov_base = acb->buf;
@@ -1391,13 +1388,22 @@ fail:
                             (acb->cluster_offset >> 9) + index_in_cluster,
                             &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
         if (acb->hd_aiocb == NULL)
-            goto fail;
+            goto done;
+    }
+
+    return;
+done:
+    if (acb->qiov->niov > 1) {
+        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
+        qemu_vfree(acb->orig_buf);
     }
+    acb->common.cb(acb->common.opaque, ret);
+    qemu_aio_release(acb);
 }
 
 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
-        BlockDriverCompletionFunc *cb, void *opaque)
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
+        BlockDriverCompletionFunc *cb, void *opaque, int is_write)
 {
     QCowAIOCB *acb;
 
@@ -1406,7 +1412,13 @@ static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
         return NULL;
     acb->hd_aiocb = NULL;
     acb->sector_num = sector_num;
-    acb->buf = buf;
+    acb->qiov = qiov;
+    if (qiov->niov > 1) {
+        acb->buf = acb->orig_buf = qemu_memalign(512, qiov->size);
+        if (is_write)
+            qemu_iovec_to_buffer(qiov, acb->buf);
+    } else
+        acb->buf = qiov->iov->iov_base;
     acb->nb_sectors = nb_sectors;
     acb->n = 0;
     acb->cluster_offset = 0;
@@ -1414,13 +1426,13 @@ static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
     return acb;
 }
 
-static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
     QCowAIOCB *acb;
 
-    acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
+    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
     if (!acb)
         return NULL;
 
@@ -1439,16 +1451,12 @@ static void qcow_aio_write_cb(void *opaque, int ret)
 
     acb->hd_aiocb = NULL;
 
-    if (ret < 0) {
-    fail:
-        acb->common.cb(acb->common.opaque, ret);
-        qemu_aio_release(acb);
-        return;
-    }
+    if (ret < 0)
+        goto done;
 
     if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
         free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
-        goto fail;
+        goto done;
     }
 
     acb->nb_sectors -= acb->n;
@@ -1457,9 +1465,8 @@ static void qcow_aio_write_cb(void *opaque, int ret)
 
     if (acb->nb_sectors == 0) {
         /* request completed */
-        acb->common.cb(acb->common.opaque, 0);
-        qemu_aio_release(acb);
-        return;
+        ret = 0;
+        goto done;
     }
 
     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
@@ -1473,7 +1480,7 @@ static void qcow_aio_write_cb(void *opaque, int ret)
                                           n_end, &acb->n, &acb->l2meta);
     if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
         ret = -EIO;
-        goto fail;
+        goto done;
     }
     if (s->crypt_method) {
         if (!acb->cluster_data) {
@@ -1494,11 +1501,19 @@ static void qcow_aio_write_cb(void *opaque, int ret)
                                     &acb->hd_qiov, acb->n,
                                     qcow_aio_write_cb, acb);
     if (acb->hd_aiocb == NULL)
-        goto fail;
+        goto done;
+
+    return;
+
+done:
+    if (acb->qiov->niov > 1)
+        qemu_vfree(acb->orig_buf);
+    acb->common.cb(acb->common.opaque, ret);
+    qemu_aio_release(acb);
 }
 
-static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
     BDRVQcowState *s = bs->opaque;
@@ -1506,7 +1521,7 @@ static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
 
     s->cluster_cache_offset = -1; /* disable compressed cache */
 
-    acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
+    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
     if (!acb)
         return NULL;
 
@@ -2771,8 +2786,8 @@ BlockDriver bdrv_qcow2 = {
     .bdrv_set_key	= qcow_set_key,
     .bdrv_make_empty	= qcow_make_empty,
 
-    .bdrv_aio_read	= qcow_aio_read,
-    .bdrv_aio_write	= qcow_aio_write,
+    .bdrv_aio_readv	= qcow_aio_readv,
+    .bdrv_aio_writev	= qcow_aio_writev,
     .bdrv_aio_cancel	= qcow_aio_cancel,
     .aiocb_size		= sizeof(QCowAIOCB),
     .bdrv_write_compressed = qcow_write_compressed,
diff --git a/block-raw-posix.c b/block-raw-posix.c
index 4da5ae4397..822839fb8f 100644
--- a/block-raw-posix.c
+++ b/block-raw-posix.c
@@ -599,8 +599,8 @@ static int posix_aio_init(void)
     return 0;
 }
 
-static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
+static RawAIOCB *raw_aio_setup(BlockDriverState *bs, int64_t sector_num,
+        QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
     BDRVRawState *s = bs->opaque;
@@ -614,24 +614,25 @@ static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
         return NULL;
     acb->aiocb.aio_fildes = s->fd;
     acb->aiocb.ev_signo = SIGUSR2;
-    acb->aiocb.aio_buf = buf;
-    if (nb_sectors < 0)
-        acb->aiocb.aio_nbytes = -nb_sectors;
-    else
-        acb->aiocb.aio_nbytes = nb_sectors * 512;
+    acb->aiocb.aio_iov = qiov->iov;
+    acb->aiocb.aio_niov = qiov->niov;
+    acb->aiocb.aio_nbytes = nb_sectors * 512;
     acb->aiocb.aio_offset = sector_num * 512;
+    acb->aiocb.aio_flags = 0;
+
+    /*
+     * If O_DIRECT is used the buffer needs to be aligned on a sector
+     * boundary. Tell the low level code to ensure that in case it's
+     * not done yet.
+     */
+    if (s->aligned_buf)
+        acb->aiocb.aio_flags |= QEMU_AIO_SECTOR_ALIGNED;
+
     acb->next = posix_aio_state->first_aio;
     posix_aio_state->first_aio = acb;
     return acb;
 }
 
-static void raw_aio_em_cb(void* opaque)
-{
-    RawAIOCB *acb = opaque;
-    acb->common.cb(acb->common.opaque, acb->ret);
-    qemu_aio_release(acb);
-}
-
 static void raw_aio_remove(RawAIOCB *acb)
 {
     RawAIOCB **pacb;
@@ -651,28 +652,13 @@ static void raw_aio_remove(RawAIOCB *acb)
     }
 }
 
-static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
     RawAIOCB *acb;
 
-    /*
-     * If O_DIRECT is used and the buffer is not aligned fall back
-     * to synchronous IO.
-     */
-    BDRVRawState *s = bs->opaque;
-
-    if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
-        QEMUBH *bh;
-        acb = qemu_aio_get(bs, cb, opaque);
-        acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
-        bh = qemu_bh_new(raw_aio_em_cb, acb);
-        qemu_bh_schedule(bh);
-        return &acb->common;
-    }
-
-    acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
+    acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
     if (!acb)
         return NULL;
     if (qemu_paio_read(&acb->aiocb) < 0) {
@@ -682,28 +668,13 @@ static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
     return &acb->common;
 }
 
-static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
     RawAIOCB *acb;
 
-    /*
-     * If O_DIRECT is used and the buffer is not aligned fall back
-     * to synchronous IO.
-     */
-    BDRVRawState *s = bs->opaque;
-
-    if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
-        QEMUBH *bh;
-        acb = qemu_aio_get(bs, cb, opaque);
-        acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
-        bh = qemu_bh_new(raw_aio_em_cb, acb);
-        qemu_bh_schedule(bh);
-        return &acb->common;
-    }
-
-    acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
+    acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
     if (!acb)
         return NULL;
     if (qemu_paio_write(&acb->aiocb) < 0) {
@@ -887,8 +858,8 @@ BlockDriver bdrv_raw = {
     .bdrv_flush = raw_flush,
 
 #ifdef CONFIG_AIO
-    .bdrv_aio_read = raw_aio_read,
-    .bdrv_aio_write = raw_aio_write,
+    .bdrv_aio_readv = raw_aio_readv,
+    .bdrv_aio_writev = raw_aio_writev,
     .bdrv_aio_cancel = raw_aio_cancel,
     .aiocb_size = sizeof(RawAIOCB),
 #endif
@@ -1215,12 +1186,24 @@ static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
         unsigned long int req, void *buf,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
+    BDRVRawState *s = bs->opaque;
     RawAIOCB *acb;
 
-    acb = raw_aio_setup(bs, 0, buf, 0, cb, opaque);
+    if (fd_open(bs) < 0)
+        return NULL;
+
+    acb = qemu_aio_get(bs, cb, opaque);
     if (!acb)
         return NULL;
+    acb->aiocb.aio_fildes = s->fd;
+    acb->aiocb.ev_signo = SIGUSR2;
+    acb->aiocb.aio_offset = 0;
+    acb->aiocb.aio_flags = 0;
+
+    acb->next = posix_aio_state->first_aio;
+    posix_aio_state->first_aio = acb;
 
+    acb->aiocb.aio_ioctl_buf = buf;
     acb->aiocb.aio_ioctl_cmd = req;
     if (qemu_paio_ioctl(&acb->aiocb) < 0) {
         raw_aio_remove(acb);
@@ -1424,8 +1407,8 @@ BlockDriver bdrv_host_device = {
     .bdrv_flush		= raw_flush,
 
 #ifdef CONFIG_AIO
-    .bdrv_aio_read	= raw_aio_read,
-    .bdrv_aio_write	= raw_aio_write,
+    .bdrv_aio_readv	= raw_aio_readv,
+    .bdrv_aio_writev	= raw_aio_writev,
     .bdrv_aio_cancel	= raw_aio_cancel,
     .aiocb_size		= sizeof(RawAIOCB),
 #endif
diff --git a/block.c b/block.c
index 916f08e76f..8a78f14ed5 100644
--- a/block.c
+++ b/block.c
@@ -47,25 +47,21 @@
 #define SECTOR_BITS 9
 #define SECTOR_SIZE (1 << SECTOR_BITS)
 
-static AIOPool vectored_aio_pool;
-
 typedef struct BlockDriverAIOCBSync {
     BlockDriverAIOCB common;
     QEMUBH *bh;
     int ret;
+    /* vector translation state */
+    QEMUIOVector *qiov;
+    uint8_t *bounce;
+    int is_write;
 } BlockDriverAIOCBSync;
 
-static BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque);
-static BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
-        BlockDriverCompletionFunc *cb, void *opaque);
-static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
-        BlockDriverCompletionFunc *cb, void *opaque);
-static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque);
 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
@@ -144,10 +140,10 @@ void path_combine(char *dest, int dest_size,
 
 static void bdrv_register(BlockDriver *bdrv)
 {
-    if (!bdrv->bdrv_aio_read) {
+    if (!bdrv->bdrv_aio_readv) {
         /* add AIO emulation layer */
-        bdrv->bdrv_aio_read = bdrv_aio_read_em;
-        bdrv->bdrv_aio_write = bdrv_aio_write_em;
+        bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
+        bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
     } else if (!bdrv->bdrv_read) {
@@ -1295,91 +1291,10 @@ char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
 /**************************************************************/
 /* async I/Os */
 
-typedef struct VectorTranslationAIOCB {
-    BlockDriverAIOCB common;
-    QEMUIOVector *iov;
-    uint8_t *bounce;
-    int is_write;
-    BlockDriverAIOCB *aiocb;
-} VectorTranslationAIOCB;
-
-static void bdrv_aio_cancel_vector(BlockDriverAIOCB *_acb)
-{
-    VectorTranslationAIOCB *acb
-        = container_of(_acb, VectorTranslationAIOCB, common);
-
-    bdrv_aio_cancel(acb->aiocb);
-}
-
-static void bdrv_aio_rw_vector_cb(void *opaque, int ret)
-{
-    VectorTranslationAIOCB *s = (VectorTranslationAIOCB *)opaque;
-
-    if (!s->is_write) {
-        qemu_iovec_from_buffer(s->iov, s->bounce, s->iov->size);
-    }
-    qemu_vfree(s->bounce);
-    s->common.cb(s->common.opaque, ret);
-    qemu_aio_release(s);
-}
-
-static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
-                                            int64_t sector_num,
-                                            QEMUIOVector *iov,
-                                            int nb_sectors,
-                                            BlockDriverCompletionFunc *cb,
-                                            void *opaque,
-                                            int is_write)
-
-{
-    VectorTranslationAIOCB *s = qemu_aio_get_pool(&vectored_aio_pool, bs,
-                                                  cb, opaque);
-
-    s->iov = iov;
-    s->bounce = qemu_memalign(512, nb_sectors * 512);
-    s->is_write = is_write;
-    if (is_write) {
-        qemu_iovec_to_buffer(s->iov, s->bounce);
-        s->aiocb = bdrv_aio_write(bs, sector_num, s->bounce, nb_sectors,
-                                  bdrv_aio_rw_vector_cb, s);
-    } else {
-        s->aiocb = bdrv_aio_read(bs, sector_num, s->bounce, nb_sectors,
-                                 bdrv_aio_rw_vector_cb, s);
-    }
-    if (!s->aiocb) {
-        qemu_vfree(s->bounce);
-        qemu_aio_release(s);
-        return NULL;
-    }
-    return &s->common;
-}
-
 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
-                                 QEMUIOVector *iov, int nb_sectors,
+                                 QEMUIOVector *qiov, int nb_sectors,
                                  BlockDriverCompletionFunc *cb, void *opaque)
 {
-    if (bdrv_check_request(bs, sector_num, nb_sectors))
-        return NULL;
-
-    return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
-                              cb, opaque, 0);
-}
-
-BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
-                                  QEMUIOVector *iov, int nb_sectors,
-                                  BlockDriverCompletionFunc *cb, void *opaque)
-{
-    if (bdrv_check_request(bs, sector_num, nb_sectors))
-        return NULL;
-
-    return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
-                              cb, opaque, 1);
-}
-
-static BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
-                                uint8_t *buf, int nb_sectors,
-                                BlockDriverCompletionFunc *cb, void *opaque)
-{
     BlockDriver *drv = bs->drv;
     BlockDriverAIOCB *ret;
 
@@ -1388,7 +1303,8 @@ static BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
     if (bdrv_check_request(bs, sector_num, nb_sectors))
         return NULL;
 
-    ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
+    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
+                              cb, opaque);
 
     if (ret) {
 	/* Update stats even though technically transfer has not happened. */
@@ -1399,9 +1315,9 @@ static BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
     return ret;
 }
 
-static BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
-                                 const uint8_t *buf, int nb_sectors,
-                                 BlockDriverCompletionFunc *cb, void *opaque)
+BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
+                                  QEMUIOVector *qiov, int nb_sectors,
+                                  BlockDriverCompletionFunc *cb, void *opaque)
 {
     BlockDriver *drv = bs->drv;
     BlockDriverAIOCB *ret;
@@ -1413,7 +1329,8 @@ static BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num
     if (bdrv_check_request(bs, sector_num, nb_sectors))
         return NULL;
 
-    ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
+    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
+                               cb, opaque);
 
     if (ret) {
 	/* Update stats even though technically transfer has not happened. */
@@ -1436,42 +1353,62 @@ void bdrv_aio_cancel(BlockDriverAIOCB *acb)
 static void bdrv_aio_bh_cb(void *opaque)
 {
     BlockDriverAIOCBSync *acb = opaque;
+
+    qemu_vfree(acb->bounce);
+
+    if (!acb->is_write)
+        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
     acb->common.cb(acb->common.opaque, acb->ret);
+
     qemu_aio_release(acb);
 }
 
-static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
-        BlockDriverCompletionFunc *cb, void *opaque)
+static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
+                                            int64_t sector_num,
+                                            QEMUIOVector *qiov,
+                                            int nb_sectors,
+                                            BlockDriverCompletionFunc *cb,
+                                            void *opaque,
+                                            int is_write)
+
 {
     BlockDriverAIOCBSync *acb;
-    int ret;
 
     acb = qemu_aio_get(bs, cb, opaque);
+    acb->is_write = is_write;
+    acb->qiov = qiov;
+    acb->bounce = qemu_memalign(512, qiov->size);
+
     if (!acb->bh)
         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
-    ret = bdrv_read(bs, sector_num, buf, nb_sectors);
-    acb->ret = ret;
+
+    if (is_write) {
+        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
+        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
+    } else {
+        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
+    }
+
     qemu_bh_schedule(acb->bh);
+
     return &acb->common;
 }
 
-static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
-    BlockDriverAIOCBSync *acb;
-    int ret;
+    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
+}
 
-    acb = qemu_aio_get(bs, cb, opaque);
-    if (!acb->bh)
-        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
-    ret = bdrv_write(bs, sector_num, buf, nb_sectors);
-    acb->ret = ret;
-    qemu_bh_schedule(acb->bh);
-    return &acb->common;
+static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
+        BlockDriverCompletionFunc *cb, void *opaque)
+{
+    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
 }
 
+
 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
 {
     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
@@ -1494,10 +1431,15 @@ static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
 {
     int async_ret;
     BlockDriverAIOCB *acb;
+    struct iovec iov;
+    QEMUIOVector qiov;
 
     async_ret = NOT_DONE;
-    acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
-                        bdrv_rw_em_cb, &async_ret);
+    iov.iov_base = buf;
+    iov.iov_len = nb_sectors * 512;
+    qemu_iovec_init_external(&qiov, &iov, 1);
+    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
+        bdrv_rw_em_cb, &async_ret);
     if (acb == NULL)
         return -1;
 
@@ -1513,10 +1455,15 @@ static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
 {
     int async_ret;
     BlockDriverAIOCB *acb;
+    struct iovec iov;
+    QEMUIOVector qiov;
 
     async_ret = NOT_DONE;
-    acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
-                         bdrv_rw_em_cb, &async_ret);
+    iov.iov_base = (void *)buf;
+    iov.iov_len = nb_sectors * 512;
+    qemu_iovec_init_external(&qiov, &iov, 1);
+    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
+        bdrv_rw_em_cb, &async_ret);
     if (acb == NULL)
         return -1;
     while (async_ret == NOT_DONE) {
@@ -1527,9 +1474,6 @@ static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
 
 void bdrv_init(void)
 {
-    aio_pool_init(&vectored_aio_pool, sizeof(VectorTranslationAIOCB),
-                  bdrv_aio_cancel_vector);
-
     bdrv_register(&bdrv_raw);
     bdrv_register(&bdrv_host_device);
 #ifndef _WIN32
diff --git a/block_int.h b/block_int.h
index dc4335982c..3e78997c71 100644
--- a/block_int.h
+++ b/block_int.h
@@ -54,11 +54,11 @@ struct BlockDriver {
     int (*bdrv_set_key)(BlockDriverState *bs, const char *key);
     int (*bdrv_make_empty)(BlockDriverState *bs);
     /* aio */
-    BlockDriverAIOCB *(*bdrv_aio_read)(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
+    BlockDriverAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque);
-    BlockDriverAIOCB *(*bdrv_aio_write)(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+    BlockDriverAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque);
     void (*bdrv_aio_cancel)(BlockDriverAIOCB *acb);
     int aiocb_size;
diff --git a/posix-aio-compat.c b/posix-aio-compat.c
index 65c80ecc3e..0eb77a50d3 100644
--- a/posix-aio-compat.c
+++ b/posix-aio-compat.c
@@ -20,6 +20,7 @@
 #include <stdlib.h>
 #include <stdio.h>
 #include "osdep.h"
+#include "qemu-common.h"
 
 #include "posix-aio-compat.h"
 
@@ -76,45 +77,110 @@ static void thread_create(pthread_t *thread, pthread_attr_t *attr,
     if (ret) die2(ret, "pthread_create");
 }
 
-static size_t handle_aiocb_readwrite(struct qemu_paiocb *aiocb)
+static size_t handle_aiocb_ioctl(struct qemu_paiocb *aiocb)
+{
+	int ret;
+
+	ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
+	if (ret == -1)
+		return -errno;
+	return ret;
+}
+
+/*
+ * Check if we need to copy the data in the aiocb into a new
+ * properly aligned buffer.
+ */
+static int aiocb_needs_copy(struct qemu_paiocb *aiocb)
+{
+    if (aiocb->aio_flags & QEMU_AIO_SECTOR_ALIGNED) {
+        int i;
+
+        for (i = 0; i < aiocb->aio_niov; i++)
+            if ((uintptr_t) aiocb->aio_iov[i].iov_base % 512)
+                return 1;
+    }
+
+    return 0;
+}
+
+static size_t handle_aiocb_rw_linear(struct qemu_paiocb *aiocb, char *buf)
 {
     size_t offset = 0;
-    ssize_t len;
+    size_t len;
 
     while (offset < aiocb->aio_nbytes) {
-        if (aiocb->aio_type == QEMU_PAIO_WRITE)
-            len = pwrite(aiocb->aio_fildes,
-                         (const char *)aiocb->aio_buf + offset,
+         if (aiocb->aio_type == QEMU_PAIO_WRITE)
+             len = pwrite(aiocb->aio_fildes,
+                          (const char *)buf + offset,
+                          aiocb->aio_nbytes - offset,
+                          aiocb->aio_offset + offset);
+         else
+             len = pread(aiocb->aio_fildes,
+                         buf + offset,
                          aiocb->aio_nbytes - offset,
                          aiocb->aio_offset + offset);
-        else
-            len = pread(aiocb->aio_fildes,
-                        (char *)aiocb->aio_buf + offset,
-                        aiocb->aio_nbytes - offset,
-                        aiocb->aio_offset + offset);
-
-        if (len == -1 && errno == EINTR)
-            continue;
-        else if (len == -1) {
-            offset = -errno;
-            break;
-        } else if (len == 0)
-            break;
 
-        offset += len;
+         if (len == -1 && errno == EINTR)
+             continue;
+         else if (len == -1) {
+             offset = -errno;
+             break;
+         } else if (len == 0)
+             break;
+
+         offset += len;
     }
 
     return offset;
 }
 
-static size_t handle_aiocb_ioctl(struct qemu_paiocb *aiocb)
+static size_t handle_aiocb_rw(struct qemu_paiocb *aiocb)
 {
-	int ret;
+    size_t nbytes;
+    char *buf;
+
+    if (!aiocb_needs_copy(aiocb) && aiocb->aio_niov == 1) {
+        /*
+         * If there is just a single buffer, and it is properly aligned
+         * we can just use plain pread/pwrite without any problems.
+         */
+        return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
+    }
 
-	ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_buf);
-	if (ret == -1)
-		return -errno;
-	return ret;
+    /*
+     * Ok, we have to do it the hard way, copy all segments into
+     * a single aligned buffer.
+     */
+    buf = qemu_memalign(512, aiocb->aio_nbytes);
+    if (aiocb->aio_type == QEMU_PAIO_WRITE) {
+        char *p = buf;
+        int i;
+
+        for (i = 0; i < aiocb->aio_niov; ++i) {
+            memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
+            p += aiocb->aio_iov[i].iov_len;
+        }
+    }
+
+    nbytes = handle_aiocb_rw_linear(aiocb, buf);
+    if (aiocb->aio_type != QEMU_PAIO_WRITE) {
+        char *p = buf;
+        size_t count = aiocb->aio_nbytes, copy;
+        int i;
+
+        for (i = 0; i < aiocb->aio_niov && count; ++i) {
+            copy = count;
+            if (copy > aiocb->aio_iov[i].iov_len)
+                copy = aiocb->aio_iov[i].iov_len;
+            memcpy(aiocb->aio_iov[i].iov_base, p, copy);
+            p     += copy;
+            count -= copy;
+        }
+    }
+    qemu_vfree(buf);
+
+    return nbytes;
 }
 
 static void *aio_thread(void *unused)
@@ -157,7 +223,7 @@ static void *aio_thread(void *unused)
         switch (aiocb->aio_type) {
         case QEMU_PAIO_READ:
         case QEMU_PAIO_WRITE:
-		ret = handle_aiocb_readwrite(aiocb);
+		ret = handle_aiocb_rw(aiocb);
 		break;
         case QEMU_PAIO_IOCTL:
 		ret = handle_aiocb_ioctl(aiocb);
diff --git a/posix-aio-compat.h b/posix-aio-compat.h
index a1cdfd7f1f..1c5dcbd920 100644
--- a/posix-aio-compat.h
+++ b/posix-aio-compat.h
@@ -27,11 +27,18 @@
 struct qemu_paiocb
 {
     int aio_fildes;
-    void *aio_buf;
+    union {
+        struct iovec *aio_iov;
+	void *aio_ioctl_buf;
+    };
+    int aio_niov;
     size_t aio_nbytes;
 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_PAIO_IOCTL */
     int ev_signo;
     off_t aio_offset;
+    unsigned aio_flags;
+/* 512 byte alignment required for buffer, offset and length */
+#define QEMU_AIO_SECTOR_ALIGNED	0x01
 
     /* private */
     TAILQ_ENTRY(qemu_paiocb) node;