Re: [Qemu-devel] [PATCH v7 09/14] block: Add block job transactions

[John Snow <jsnow@redhat.com>]

On 09/21/2015 10:46 PM, Fam Zheng wrote:
> Sometimes block jobs must execute as a transaction group.  Finishing
> jobs wait until all other jobs are ready to complete successfully.
> Failure or cancellation of one job cancels the other jobs in the group.
> 
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> [Rewrite the implementation which is now contained in block_job_completed.
> --Fam]
> Signed-off-by: Fam Zheng <famz@redhat.com>
> Reviewed-by: Max Reitz <mreitz@redhat.com>
> ---
>  blockjob.c               | 135 ++++++++++++++++++++++++++++++++++++++++++++++-
>  include/block/block.h    |   1 +
>  include/block/blockjob.h |  38 +++++++++++++
>  3 files changed, 172 insertions(+), 2 deletions(-)
> 
> diff --git a/blockjob.c b/blockjob.c
> index 36c18e0..91e8d3c 100644
> --- a/blockjob.c
> +++ b/blockjob.c
> @@ -36,6 +36,19 @@
>  #include "qemu/timer.h"
>  #include "qapi-event.h"
>  
> +/* Transactional group of block jobs */
> +struct BlockJobTxn {
> +
> +    /* Is this txn being cancelled? */
> +    bool aborting;
> +
> +    /* List of jobs */
> +    QLIST_HEAD(, BlockJob) jobs;
> +
> +    /* Reference count */
> +    int refcnt;
> +};
> +
>  void *block_job_create(const BlockJobDriver *driver, BlockDriverState *bs,
>                         int64_t speed, BlockCompletionFunc *cb,
>                         void *opaque, Error **errp)
> @@ -90,6 +103,86 @@ void block_job_unref(BlockJob *job)
>      }
>  }
>  
> +static void block_job_completed_single(BlockJob *job)
> +{
> +    if (!job->ret) {
> +        if (job->driver->commit) {
> +            job->driver->commit(job);
> +        }
> +    } else {
> +        if (job->driver->abort) {
> +            job->driver->abort(job);
> +        }
> +    }
> +    job->cb(job->opaque, job->ret);
> +    if (job->txn) {
> +        block_job_txn_unref(job->txn);
> +    }
> +    block_job_unref(job);
> +}
> +
> +static void block_job_completed_txn_abort(BlockJob *job)
> +{
> +    AioContext *ctx;
> +    BlockJobTxn *txn = job->txn;
> +    BlockJob *other_job, *next;
> +
> +    if (txn->aborting) {
> +        /*
> +         * We are cancelled by another job, which will handle everything.
> +         */
> +        return;
> +    }
> +    txn->aborting = true;
> +    /* We are the first failed job. Cancel other jobs. */
> +    QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
> +        ctx = bdrv_get_aio_context(other_job->bs);
> +        aio_context_acquire(ctx);
> +    }
> +    QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
> +        if (other_job == job || other_job->completed) {
> +            /* Other jobs are "effectively" cancelled by us, set the status for
> +             * them; this job, however, may or may not be cancelled, depending
> +             * on the caller, so leave it. */
> +            if (other_job != job) {
> +                other_job->cancelled = true;
> +            }
> +            continue;
> +        }
> +        block_job_cancel_sync(other_job);
> +        assert(other_job->completed);
> +    }
> +    QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
> +        ctx = bdrv_get_aio_context(other_job->bs);
> +        block_job_completed_single(other_job);
> +        aio_context_release(ctx);
> +    }
> +}
> +
> +static void block_job_completed_txn_success(BlockJob *job)
> +{
> +    AioContext *ctx;
> +    BlockJobTxn *txn = job->txn;
> +    BlockJob *other_job, *next;
> +    /*
> +     * Successful completion, see if there are other running jobs in this
> +     * txn.
> +     */
> +    QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
> +        if (!other_job->completed) {
> +            return;
> +        }
> +    }
> +    /* We are the last completed job, commit the transaction. */
> +    QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
> +        ctx = bdrv_get_aio_context(other_job->bs);
> +        aio_context_acquire(ctx);
> +        assert(other_job->ret == 0);

Sorry for being a dense noggin about this, but is it documented anywhere
(through an assertion or otherwise) that we will never return a
positive, non-zero return code for a block job?

I just don't want to get into a situation where, in the future, someone
decides to do so and then mysteriously something breaks a version or two
later.

> +        block_job_completed_single(other_job);
> +        aio_context_release(ctx);
> +    }
> +}
> +
>  void block_job_completed(BlockJob *job, int ret)
>  {
>      BlockDriverState *bs = job->bs;
> @@ -98,8 +191,13 @@ void block_job_completed(BlockJob *job, int ret)
>      assert(!job->completed);
>      job->completed = true;
>      job->ret = ret;
> -    job->cb(job->opaque, ret);
> -    block_job_unref(job);
> +    if (!job->txn) {
> +        block_job_completed_single(job);
> +    } else if (ret < 0 || block_job_is_cancelled(job)) {
> +        block_job_completed_txn_abort(job);
> +    } else {
> +        block_job_completed_txn_success(job);
> +    }
>  }
>  
>  void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
> @@ -398,3 +496,36 @@ void block_job_defer_to_main_loop(BlockJob *job,
>  
>      qemu_bh_schedule(data->bh);
>  }
> +
> +BlockJobTxn *block_job_txn_new(void)
> +{
> +    BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
> +    QLIST_INIT(&txn->jobs);
> +    txn->refcnt = 1;
> +    return txn;
> +}
> +
> +static void block_job_txn_ref(BlockJobTxn *txn)
> +{
> +    txn->refcnt++;
> +}
> +
> +void block_job_txn_unref(BlockJobTxn *txn)
> +{
> +    if (--txn->refcnt == 0) {
> +        g_free(txn);
> +    }
> +}
> +
> +void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
> +{
> +    if (!txn) {
> +        return;
> +    }
> +
> +    assert(!job->txn);
> +    job->txn = txn;
> +
> +    QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
> +    block_job_txn_ref(txn);
> +}
> diff --git a/include/block/block.h b/include/block/block.h
> index ca56021..94548cb 100644
> --- a/include/block/block.h
> +++ b/include/block/block.h
> @@ -14,6 +14,7 @@ typedef struct BlockDriver BlockDriver;
>  typedef struct BlockJob BlockJob;
>  typedef struct BdrvChild BdrvChild;
>  typedef struct BdrvChildRole BdrvChildRole;
> +typedef struct BlockJobTxn BlockJobTxn;
>  
>  typedef struct BlockDriverInfo {
>      /* in bytes, 0 if irrelevant */
> diff --git a/include/block/blockjob.h b/include/block/blockjob.h
> index a77799f..fa77189 100644
> --- a/include/block/blockjob.h
> +++ b/include/block/blockjob.h
> @@ -154,6 +154,9 @@ struct BlockJob {
>       */
>      int ret;
>  
> +    /** Non-NULL if this job is part of a transaction */
> +    BlockJobTxn *txn;
> +    QLIST_ENTRY(BlockJob) txn_list;
>  };
>  
>  /**
> @@ -397,4 +400,39 @@ void block_job_defer_to_main_loop(BlockJob *job,
>                                    BlockJobDeferToMainLoopFn *fn,
>                                    void *opaque);
>  
> +/**
> + * block_job_txn_new:
> + *
> + * Allocate and return a new block job transaction.  Jobs can be added to the
> + * transaction using block_job_txn_add_job().
> + *
> + * The transaction is automatically freed when the last job completes or is
> + * cancelled.
> + *
> + * All jobs in the transaction either complete successfully or fail/cancel as a
> + * group.  Jobs wait for each other before completing.  Cancelling one job
> + * cancels all jobs in the transaction.
> + */
> +BlockJobTxn *block_job_txn_new(void);
> +
> +/**
> + * block_job_txn_unref:
> + *
> + * Release a reference that was previously acquired with block_job_txn_add_job
> + * or block_job_txn_new. If it's the last reference to the object, it will be
> + * freed.
> + */
> +void block_job_txn_unref(BlockJobTxn *txn);
> +
> +/**
> + * block_job_txn_add_job:
> + * @txn: The transaction (may be NULL)
> + * @job: Job to add to the transaction
> + *
> + * Add @job to the transaction.  The @job must not already be in a transaction.
> + * The caller must call either block_job_txn_unref() or block_job_completed()
> + * to release the reference that is automatically grabbed here.
> + */
> +void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job);
> +
>  #endif
> 

-- 
—js