[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH v2] qcow2: Unlock during COW
From: |
Stefan Hajnoczi |
Subject: |
Re: [Qemu-devel] [PATCH v2] qcow2: Unlock during COW |
Date: |
Tue, 15 Nov 2011 12:07:21 +0000 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
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