From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:49810) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1X0UmI-0001pa-A5 for qemu-devel@nongnu.org; Fri, 27 Jun 2014 07:57:19 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1X0UmC-0001KR-80 for qemu-devel@nongnu.org; Fri, 27 Jun 2014 07:57:14 -0400 Received: from mx1.redhat.com ([209.132.183.28]:14931) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1X0UmC-0001K2-0h for qemu-devel@nongnu.org; Fri, 27 Jun 2014 07:57:08 -0400 Date: Fri, 27 Jun 2014 13:57:02 +0200 From: Kevin Wolf Message-ID: <20140627115702.GD5223@noname.redhat.com> References: <1402912857-10509-1-git-send-email-benoit.canet@irqsave.net> <1402912857-10509-4-git-send-email-benoit.canet@irqsave.net> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline In-Reply-To: <1402912857-10509-4-git-send-email-benoit.canet@irqsave.net> Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH v10 3/4] block: Add replaces argument to drive-mirror List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: =?iso-8859-1?Q?Beno=EEt?= Canet Cc: Benoit Canet , qemu-devel@nongnu.org, stefanha@redhat.com, mreitz@redhat.com Am 16.06.2014 um 12:00 hat Beno=EEt Canet geschrieben: > drive-mirror will bdrv_swap the new BDS named node-name with the one > pointed by replaces when the mirroring is finished. >=20 > Signed-off-by: Benoit Canet > Reviewed-by: Max Reitz > --- > block.c | 17 ++++++++++++++ > block/mirror.c | 60 +++++++++++++++++++++++++++++++++++++--= -------- > blockdev.c | 30 +++++++++++++++++++++++- > hmp.c | 2 +- > include/block/block.h | 4 ++++ > include/block/block_int.h | 3 +++ > qapi/block-core.json | 6 ++++- > qmp-commands.hx | 4 +++- > 8 files changed, 109 insertions(+), 17 deletions(-) >=20 > diff --git a/block.c b/block.c > index 17f763d..318f1e6 100644 > --- a/block.c > +++ b/block.c > @@ -5795,3 +5795,20 @@ bool bdrv_is_first_non_filter(BlockDriverState *= candidate) > =20 > return false; > } > + > +BlockDriverState *check_to_replace_node(const char *node_name, Error *= *errp) > +{ > + BlockDriverState *to_replace_bs =3D bdrv_find_node(node_name); > + if (!to_replace_bs) { > + error_setg(errp, "Node name '%s' not found", > + node_name); Unnecessary line break. > + return NULL; > + } > + > + if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)= ) { > + return NULL; > + } > + > + return to_replace_bs; > +} > + Empty line before EOF. > diff --git a/block/mirror.c b/block/mirror.c > index 94c8661..151167e 100644 > --- a/block/mirror.c > +++ b/block/mirror.c > @@ -32,6 +32,12 @@ typedef struct MirrorBlockJob { > RateLimit limit; > BlockDriverState *target; > BlockDriverState *base; > + /* The name of the graph node to replace */ > + char *replaces; > + /* The BDS to replace */ > + BlockDriverState *to_replace; > + /* Used to block operations on the drive-mirror-replace target */ > + Error *replace_blocker; > bool is_none_mode; > BlockdevOnError on_source_error, on_target_error; > bool synced; > @@ -490,10 +496,14 @@ immediate_exit: > bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); > bdrv_iostatus_disable(s->target); > if (s->should_complete && ret =3D=3D 0) { > - if (bdrv_get_flags(s->target) !=3D bdrv_get_flags(s->common.bs= )) { > - bdrv_reopen(s->target, bdrv_get_flags(s->common.bs), NULL)= ; > + BlockDriverState *to_replace =3D s->common.bs; > + if (s->to_replace) { > + to_replace =3D s->to_replace; > } > - bdrv_swap(s->target, s->common.bs); > + if (bdrv_get_flags(s->target) !=3D bdrv_get_flags(to_replace))= { > + bdrv_reopen(s->target, bdrv_get_flags(to_replace), NULL); > + } > + bdrv_swap(s->target, to_replace); > if (s->common.driver->job_type =3D=3D BLOCK_JOB_TYPE_COMMIT) { > /* drop the bs loop chain formed by the swap: break the lo= op then > * trigger the unref from the top one */ > @@ -502,6 +512,12 @@ immediate_exit: > bdrv_unref(p); > } > } > + if (s->to_replace) { > + bdrv_op_unblock_all(s->to_replace, s->replace_blocker); > + error_free(s->replace_blocker); > + bdrv_unref(s->to_replace); > + } > + g_free(s->replaces); > bdrv_unref(s->target); > block_job_completed(&s->common, ret); > } > @@ -540,6 +556,20 @@ static void mirror_complete(BlockJob *job, Error *= *errp) > return; > } > =20 > + /* check the target bs is not blocked and block all operations on = it */ > + if (s->replaces) { > + s->to_replace =3D check_to_replace_node(s->replaces, errp); > + This empty line looks unusual. > + if (!s->to_replace) { > + return; > + } So here is the thing that I really wanted to comment on. In the case of a REPLACE blocker being set, this is a silent failure. The completion command will return success, but s->should_complete won't actually be set, so the completion doesn't happen. The only thing that actually happens is the bdrv_open_backing_file(s->target) (which looks somewhat questionable, too...) Now I would expect that the REPLACE blocker is actually set for any backing file, because that is what bdrv_set_backing_hd() does. For quorum it does work as expected because quorum children don't get any backing_blocker (we need to check whether they should get something similar from the quorum BDS), so this is probably why it escaped your testing. We'll need a test case that tries replacing some ordinary backing file. Now I think the (accidental?) restriction to only replacing quorum nodes actually makes this patch pretty safe, so maybe it would be nice to keep this behaviour; but we need to fix it to not fail silently but return an explicit error. > + error_setg(&s->replace_blocker, > + "block device is in use by block-job-complete"); > + bdrv_op_block_all(s->to_replace, s->replace_blocker); > + bdrv_ref(s->to_replace); > + } > + > s->should_complete =3D true; > block_job_resume(job); > } > @@ -562,14 +592,15 @@ static const BlockJobDriver commit_active_job_dri= ver =3D { > }; > =20 > static void mirror_start_job(BlockDriverState *bs, BlockDriverState *t= arget, > - int64_t speed, int64_t granularity, > - int64_t buf_size, > - BlockdevOnError on_source_error, > - BlockdevOnError on_target_error, > - BlockDriverCompletionFunc *cb, > - void *opaque, Error **errp, > - const BlockJobDriver *driver, > - bool is_none_mode, BlockDriverState *base) > + const char *replaces, > + int64_t speed, int64_t granularity, > + int64_t buf_size, > + BlockdevOnError on_source_error, > + BlockdevOnError on_target_error, > + BlockDriverCompletionFunc *cb, > + void *opaque, Error **errp, > + const BlockJobDriver *driver, > + bool is_none_mode, BlockDriverState *base= ) > { > MirrorBlockJob *s; > =20 > @@ -600,6 +631,7 @@ static void mirror_start_job(BlockDriverState *bs, = BlockDriverState *target, > return; > } > =20 > + s->replaces =3D g_strdup(replaces); > s->on_source_error =3D on_source_error; > s->on_target_error =3D on_target_error; > s->target =3D target; One design question that isn't quite clear to me yet is why you resolve the device name only in mirror_complete() and not here. This means that the drive-mirror QMP command can refer to one BDS with node-name foo, which then gets removed and another BDS with node-name foo is added, and then it would refer to the new BDS on completion time. I would find it less surprising if we took a reference to the old BDS here so that you can't remove it. Perhaps setting the replace_blocker here already would be safer, too. > @@ -621,6 +653,7 @@ static void mirror_start_job(BlockDriverState *bs, = BlockDriverState *target, > } > =20 > void mirror_start(BlockDriverState *bs, BlockDriverState *target, > + const char *replaces, > int64_t speed, int64_t granularity, int64_t buf_size= , > MirrorSyncMode mode, BlockdevOnError on_source_error= , > BlockdevOnError on_target_error, > @@ -632,7 +665,8 @@ void mirror_start(BlockDriverState *bs, BlockDriver= State *target, > =20 > is_none_mode =3D mode =3D=3D MIRROR_SYNC_MODE_NONE; > base =3D mode =3D=3D MIRROR_SYNC_MODE_TOP ? bs->backing_hd : NULL; > - mirror_start_job(bs, target, speed, granularity, buf_size, > + mirror_start_job(bs, target, replaces, > + speed, granularity, buf_size, > on_source_error, on_target_error, cb, opaque, err= p, > &mirror_job_driver, is_none_mode, base); > } > @@ -680,7 +714,7 @@ void commit_active_start(BlockDriverState *bs, Bloc= kDriverState *base, > } > =20 > bdrv_ref(base); > - mirror_start_job(bs, base, speed, 0, 0, > + mirror_start_job(bs, base, NULL, speed, 0, 0, > on_error, on_error, cb, opaque, &local_err, > &commit_active_job_driver, false, base); > if (local_err) { > diff --git a/blockdev.c b/blockdev.c > index 06b14f2..237a548 100644 > --- a/blockdev.c > +++ b/blockdev.c > @@ -2107,6 +2107,7 @@ BlockDeviceInfoList *qmp_query_named_block_nodes(= Error **errp) > void qmp_drive_mirror(const char *device, const char *target, > bool has_format, const char *format, > bool has_node_name, const char *node_name, > + bool has_replaces, const char *replaces, > enum MirrorSyncMode sync, > bool has_mode, enum NewImageMode mode, > bool has_speed, int64_t speed, > @@ -2194,6 +2195,28 @@ void qmp_drive_mirror(const char *device, const = char *target, > return; > } > =20 > + if (has_replaces) { > + BlockDriverState *to_replace_bs; > + > + if (!has_node_name) { > + error_setg(errp, "a node-name must be provided when replac= ing a" > + " named node of the graph"); > + return; > + } > + > + to_replace_bs =3D check_to_replace_node(replaces, errp); > + > + if (!to_replace_bs) { > + return; > + } > + > + if (size !=3D bdrv_getlength(to_replace_bs)) { > + error_setg(errp, "cannot replace image with a mirror image= of " > + "different size"); > + return; > + } We may want to loosen some of these restrictions later, but it's good to start with more restrictions if in doubt. > + } > + > if ((sync =3D=3D MIRROR_SYNC_MODE_FULL || !source) > && mode !=3D NEW_IMAGE_MODE_EXISTING) > { Kevin