From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:52126) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Zf1u9-00052s-M1 for qemu-devel@nongnu.org; Thu, 24 Sep 2015 04:29:26 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1Zf1u5-0007ah-KY for qemu-devel@nongnu.org; Thu, 24 Sep 2015 04:29:25 -0400 Received: from mx1.redhat.com ([209.132.183.28]:47900) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Zf1u5-0007aJ-EG for qemu-devel@nongnu.org; Thu, 24 Sep 2015 04:29:21 -0400 Date: Thu, 24 Sep 2015 16:29:16 +0800 From: Fam Zheng Message-ID: <20150924082902.GA24855@localhost.nay.redhat.com> References: <1442889976-8733-1-git-send-email-famz@redhat.com> <1442889976-8733-10-git-send-email-famz@redhat.com> <5601A767.5060109@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <5601A767.5060109@redhat.com> Subject: Re: [Qemu-devel] [PATCH v7 09/14] block: Add block job transactions List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: John Snow Cc: Kevin Wolf , Jeff Cody , qemu-devel@nongnu.org, Max Reitz , vsementsov@parallels.com, stefanha@redhat.com 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 > > [Rewrite the implementation which is now contained in block_job_completed. > > --Fam] > > Signed-off-by: Fam Zheng > > Reviewed-by: Max Reitz > > --- > > 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