Re: [Qemu-devel] [PATCH v2] qcow2: Unlock during COW

[Stefan Hajnoczi]

On Mon, Nov 14, 2011 at 06:55:18PM +0100, Kevin Wolf wrote:
> Unlocking during COW allows for more parallelism. One change it requires is
> that buffers are dynamically allocated instead of just using a per-image
> buffer.
> 
> While touching the code, drop the synchronous qcow2_read() function and 
> replace
> it by a bdrv_read() call.
> 
> Signed-off-by: Kevin Wolf <address@hidden>
> ---
>  block/qcow2-cluster.c |  104 ++++++++++++++++--------------------------------
>  1 files changed, 35 insertions(+), 69 deletions(-)

This should be safe because dependent requests are queued so we can
perform copy_sectors() in parallel with the non-dependent requests.

> -static int qcow2_read(BlockDriverState *bs, int64_t sector_num,
> -                      uint8_t *buf, int nb_sectors)
> -{
> -    BDRVQcowState *s = bs->opaque;
> -    int ret, index_in_cluster, n, n1;
> -    uint64_t cluster_offset;
> -    struct iovec iov;
> -    QEMUIOVector qiov;
> -
> -    while (nb_sectors > 0) {
> -        n = nb_sectors;
> -
> -        ret = qcow2_get_cluster_offset(bs, sector_num << 9, &n,
> -            &cluster_offset);
> -        if (ret < 0) {
> -            return ret;
> -        }
> -
> -        index_in_cluster = sector_num & (s->cluster_sectors - 1);
> -        if (!cluster_offset) {
> -            if (bs->backing_hd) {
> -                /* read from the base image */
> -                iov.iov_base = buf;
> -                iov.iov_len = n * 512;
> -                qemu_iovec_init_external(&qiov, &iov, 1);
> -
> -                n1 = qcow2_backing_read1(bs->backing_hd, &qiov, sector_num, 
> n);
> -                if (n1 > 0) {
> -                    BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING);
> -                    ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
> -                    if (ret < 0)
> -                        return -1;
> -                }
> -            } else {
> -                memset(buf, 0, 512 * n);
> -            }
> -        } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
> -            if (qcow2_decompress_cluster(bs, cluster_offset) < 0)
> -                return -1;
> -            memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
> -        } else {
> -            BLKDBG_EVENT(bs->file, BLKDBG_READ);

BLKDBG_READ is now unused, there is no other occurrence in QEMU.  Not
sure if you want to remove it from blkdebug now?

> -            ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 
> 512, buf, n * 512);
> -            if (ret != n * 512)
> -                return -1;
> -            if (s->crypt_method) {
> -                qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
> -                                &s->aes_decrypt_key);
> -            }
> -        }
> -        nb_sectors -= n;
> -        sector_num += n;
> -        buf += n * 512;
> -    }
> -    return 0;
> -}
> -
>  static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
>                          uint64_t cluster_offset, int n_start, int n_end)
>  {
>      BDRVQcowState *s = bs->opaque;
>      int n, ret;
> +    void *buf;
> +
> +    /*
> +     * If this is the last cluster and it is only partially used, we must 
> only
> +     * copy until the end of the image, or bdrv_check_request will fail for 
> the
> +     * bdrv_read/write calls below.
> +     */
> +    if (start_sect + n_end > bs->total_sectors) {
> +        n_end = bs->total_sectors - start_sect;
> +    }
>  
>      n = n_end - n_start;
> -    if (n <= 0)
> +    if (n <= 0) {
>          return 0;
> +    }
> +
> +    buf = qemu_blockalign(bs, n * BDRV_SECTOR_SIZE);
> +
>      BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
> -    ret = qcow2_read(bs, start_sect + n_start, s->cluster_data, n);
> -    if (ret < 0)
> -        return ret;
> +    ret = bdrv_read(bs, start_sect + n_start, buf, n);
> +    if (ret < 0) {
> +        goto out;
> +    }
> +
>      if (s->crypt_method) {
>          qcow2_encrypt_sectors(s, start_sect + n_start,
> -                        s->cluster_data,
> -                        s->cluster_data, n, 1,
> +                        buf, buf, n, 1,
>                          &s->aes_encrypt_key);
>      }
> +
>      BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
> -    ret = bdrv_write(bs->file, (cluster_offset >> 9) + n_start,
> -        s->cluster_data, n);
> -    if (ret < 0)
> -        return ret;
> -    return 0;
> +    ret = bdrv_write(bs->file, (cluster_offset >> 9) + n_start, buf, n);
> +    if (ret < 0) {
> +        goto out;
> +    }
> +
> +    ret = 0;

bdrv_write() returns 0 on success or -errno on failure.  We don't need
to check for ret < 0 or set ret = 0 here.  Just continuing execution
should be fine unless I missed something.

Stefan