On Tue, 09/22 15:09, John Snow wrote:
>
>
> 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 <stefa...@redhat.com>
> > [Rewrite the implementation which is now contained in block_job_completed.
> > --Fam]
> > Signed-off-by: Fam Zheng <f...@redhat.com>
> > Reviewed-by: Max Reitz <mre...@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.
No, I don't think it's documented. And yes we should document it.
Anyway if someone decides to do so he will hit the assertion here immediately,
instead of a version or two later.
Fam
