Re: [Qemu-devel] [PATCH v3 3/9] block: add image streaming block job

[Stefan Hajnoczi <stefanha@gmail.com>]

On Tue, Dec 13, 2011 at 2:14 PM, Marcelo Tosatti <mtosatti@redhat.com> wrote:
> On Tue, Dec 13, 2011 at 01:52:25PM +0000, Stefan Hajnoczi wrote:
>> Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
>> ---
>>  Makefile.objs  |    1 +
>>  block/stream.c |  121 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>>  block_int.h    |    3 +
>>  trace-events   |    4 ++
>>  4 files changed, 129 insertions(+), 0 deletions(-)
>>  create mode 100644 block/stream.c
>>
>> diff --git a/Makefile.objs b/Makefile.objs
>> index d737d81..025cc08 100644
>> --- a/Makefile.objs
>> +++ b/Makefile.objs
>> @@ -34,6 +34,7 @@ block-nested-y += qcow2.o qcow2-refcount.o qcow2-cluster.o qcow2-snapshot.o qcow
>>  block-nested-y += qed.o qed-gencb.o qed-l2-cache.o qed-table.o qed-cluster.o
>>  block-nested-y += qed-check.o
>>  block-nested-y += parallels.o nbd.o blkdebug.o sheepdog.o blkverify.o
>> +block-nested-y += stream.o
>>  block-nested-$(CONFIG_WIN32) += raw-win32.o
>>  block-nested-$(CONFIG_POSIX) += raw-posix.o
>>  block-nested-$(CONFIG_LIBISCSI) += iscsi.o
>> diff --git a/block/stream.c b/block/stream.c
>> new file mode 100644
>> index 0000000..7d362ab
>> --- /dev/null
>> +++ b/block/stream.c
>> @@ -0,0 +1,121 @@
>> +/*
>> + * Image streaming
>> + *
>> + * Copyright IBM, Corp. 2011
>> + *
>> + * Authors:
>> + *  Stefan Hajnoczi   <stefanha@linux.vnet.ibm.com>
>> + *
>> + * This work is licensed under the terms of the GNU LGPL, version 2 or later.
>> + * See the COPYING.LIB file in the top-level directory.
>> + *
>> + */
>> +
>> +#include "trace.h"
>> +#include "block_int.h"
>> +
>> +enum {
>> +    /*
>> +     * Size of data buffer for populating the image file.  This should be large
>> +     * enough to process multiple clusters in a single call, so that populating
>> +     * contiguous regions of the image is efficient.
>> +     */
>> +    STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
>> +};
>> +
>> +typedef struct StreamBlockJob {
>> +    BlockJob common;
>> +    BlockDriverState *base;
>> +} StreamBlockJob;
>> +
>> +static int coroutine_fn stream_populate(BlockDriverState *bs,
>> +                                        int64_t sector_num, int nb_sectors,
>> +                                        void *buf)
>> +{
>> +    struct iovec iov = {
>> +        .iov_base = buf,
>> +        .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
>> +    };
>> +    QEMUIOVector qiov;
>> +
>> +    qemu_iovec_init_external(&qiov, &iov, 1);
>> +
>> +    /* Copy-on-read the unallocated clusters */
>> +    return bdrv_co_readv(bs, sector_num, nb_sectors, &qiov);
>> +}
>> +
>> +static void coroutine_fn stream_run(void *opaque)
>> +{
>> +    StreamBlockJob *s = opaque;
>> +    BlockDriverState *bs = s->common.bs;
>> +    int64_t sector_num, end;
>> +    int ret = 0;
>> +    int n;
>> +    void *buf;
>> +
>> +    buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
>> +    s->common.len = bdrv_getlength(bs);
>> +    bdrv_get_geometry(bs, (uint64_t *)&end);
>> +
>> +    bdrv_enable_zero_detection(bs);
>> +    bdrv_enable_copy_on_read(bs);
>> +
>> +    for (sector_num = 0; sector_num < end; sector_num += n) {
>> +        if (block_job_is_cancelled(&s->common)) {
>> +            break;
>> +        }
>> +
>> +        /* TODO rate-limit */
>> +        /* Note that even when no rate limit is applied we need to yield with
>> +         * no pending I/O here so that qemu_aio_flush() is able to return.
>> +         */
>> +        co_sleep_ns(rt_clock, 0);
>> +
>> +        ret = bdrv_co_is_allocated(bs, sector_num,
>> +                                   STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
>> +        trace_stream_one_iteration(s, sector_num, n, ret);
>> +        if (ret == 0) {
>> +            ret = stream_populate(bs, sector_num, n, buf);
>> +        }
>> +        if (ret < 0) {
>> +            break;
>> +        }
>> +
>> +        /* Publish progress */
>> +        s->common.offset += n * BDRV_SECTOR_SIZE;
>> +    }
>> +
>> +    bdrv_disable_copy_on_read(bs);
>> +    bdrv_disable_zero_detection(bs);
>> +
>> +    if (sector_num == end && ret == 0) {
>> +        bdrv_change_backing_file(bs, NULL, NULL);
>> +    }
>
> In fact there is no need for a new command to switch backing file,
> it can be done here for the shared base case also.

Agreed.

Stefan