[1/4] block/quorum.c: stable children names
diff mbox series

Message ID ce10f8cf2bb9ae8a1505b59bbc2199f7b4966990.1574356137.git.lukasstraub2@web.de
State New
Headers show
Series
  • colo: Introduce resource agent and high-level test
Related show

Commit Message

Lukas Straub Nov. 21, 2019, 5:49 p.m. UTC
If we remove the child with the highest index from the quorum,
decrement s->next_child_index. This way we get stable children
names as long as we only remove the last child.

Signed-off-by: Lukas Straub <lukasstraub2@web.de>
---
 block/quorum.c | 6 ++++++
 1 file changed, 6 insertions(+)

--
2.20.1

Comments

Eric Blake Nov. 21, 2019, 6:04 p.m. UTC | #1
On 11/21/19 11:49 AM, Lukas Straub wrote:
> If we remove the child with the highest index from the quorum,
> decrement s->next_child_index. This way we get stable children
> names as long as we only remove the last child.
> 
> Signed-off-by: Lukas Straub <lukasstraub2@web.de>
> ---
>   block/quorum.c | 6 ++++++
>   1 file changed, 6 insertions(+)
> 
> diff --git a/block/quorum.c b/block/quorum.c
> index df68adcfaa..6100d4108a 100644
> --- a/block/quorum.c
> +++ b/block/quorum.c
> @@ -1054,6 +1054,12 @@ static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
>       /* We know now that num_children > threshold, so blkverify must be false */
>       assert(!s->is_blkverify);
> 
> +    unsigned child_id;
> +    sscanf(child->name, "children.%u", &child_id);

sscanf() cannot detect overflow. Do we trust our input enough to ignore 
this shortfall in the interface, or should we be using saner interfaces 
like qemu_strtoul()?  For that matter, why do we have to reparse 
something; is it not already available somewhere in numerical form?

> +    if (child_id == s->next_child_index - 1) {
> +        s->next_child_index--;
> +    }
> +
>       bdrv_drained_begin(bs);
> 
>       /* We can safely remove this child now */
> --
> 2.20.1
> 
>
Lukas Straub Nov. 21, 2019, 6:34 p.m. UTC | #2
On Thu, 21 Nov 2019 12:04:58 -0600
Eric Blake <eblake@redhat.com> wrote:

> On 11/21/19 11:49 AM, Lukas Straub wrote:
> > If we remove the child with the highest index from the quorum,
> > decrement s->next_child_index. This way we get stable children
> > names as long as we only remove the last child.
> >
> > Signed-off-by: Lukas Straub <lukasstraub2@web.de>
> > ---
> >   block/quorum.c | 6 ++++++
> >   1 file changed, 6 insertions(+)
> >
> > diff --git a/block/quorum.c b/block/quorum.c
> > index df68adcfaa..6100d4108a 100644
> > --- a/block/quorum.c
> > +++ b/block/quorum.c
> > @@ -1054,6 +1054,12 @@ static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
> >       /* We know now that num_children > threshold, so blkverify must be false */
> >       assert(!s->is_blkverify);
> >
> > +    unsigned child_id;
> > +    sscanf(child->name, "children.%u", &child_id);
>
> sscanf() cannot detect overflow. Do we trust our input enough to ignore
> this shortfall in the interface, or should we be using saner interfaces
> like qemu_strtoul()?  For that matter, why do we have to reparse
> something; is it not already available somewhere in numerical form?

Hi,
Yes, I wondered about that too, but found no other way. But the input
is trusted, AFAIK the only way to add child nodes is trough quorum_add_child
above and quorum_open and there already are adequate checks there.

> > +    if (child_id == s->next_child_index - 1) {
> > +        s->next_child_index--;
> > +    }
> > +
> >       bdrv_drained_begin(bs);
> >
> >       /* We can safely remove this child now */
> > --
> > 2.20.1
> >
> >
>
Alberto Garcia Nov. 26, 2019, 2:21 p.m. UTC | #3
On Thu 21 Nov 2019 07:34:45 PM CET, Lukas Straub wrote:
>> > diff --git a/block/quorum.c b/block/quorum.c
>> > index df68adcfaa..6100d4108a 100644
>> > --- a/block/quorum.c
>> > +++ b/block/quorum.c
>> > @@ -1054,6 +1054,12 @@ static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
>> >       /* We know now that num_children > threshold, so blkverify must be false */
>> >       assert(!s->is_blkverify);
>> >
>> > +    unsigned child_id;
>> > +    sscanf(child->name, "children.%u", &child_id);
>>
>> sscanf() cannot detect overflow. Do we trust our input enough to
>> ignore this shortfall in the interface, or should we be using saner
>> interfaces like qemu_strtoul()?  For that matter, why do we have to
>> reparse something; is it not already available somewhere in numerical
>> form?
>
> Yes, I wondered about that too, but found no other way. But the input
> is trusted, AFAIK the only way to add child nodes is trough
> quorum_add_child above and quorum_open and there already are adequate
> checks there.

I also don't see any other way to get that value, unless we change
BDRVQuorumState to store that information (e.g. instead of children
being a list of pointers BdrvChild ** it could be a list of {pointer,
index}, or something like that).

There's another (more convoluted) alternative if we don't want to parse
child->name. Since we only want to know if the child number equals
s->next_child_index - 1, we can do it the other way around:

   snprintf(str, 32, "children.%u", s->next_child_index - 1);

and then compare str and child->name.

Berto
Lukas Straub Nov. 27, 2019, 9:20 p.m. UTC | #4
On Tue, 26 Nov 2019 15:21:37 +0100
Alberto Garcia <berto@igalia.com> wrote:

> On Thu 21 Nov 2019 07:34:45 PM CET, Lukas Straub wrote:
> >> > diff --git a/block/quorum.c b/block/quorum.c
> >> > index df68adcfaa..6100d4108a 100644
> >> > --- a/block/quorum.c
> >> > +++ b/block/quorum.c
> >> > @@ -1054,6 +1054,12 @@ static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
> >> >       /* We know now that num_children > threshold, so blkverify must be false */
> >> >       assert(!s->is_blkverify);
> >> >
> >> > +    unsigned child_id;
> >> > +    sscanf(child->name, "children.%u", &child_id);
> >>
> >> sscanf() cannot detect overflow. Do we trust our input enough to
> >> ignore this shortfall in the interface, or should we be using saner
> >> interfaces like qemu_strtoul()?  For that matter, why do we have to
> >> reparse something; is it not already available somewhere in numerical
> >> form?
> >
> > Yes, I wondered about that too, but found no other way. But the input
> > is trusted, AFAIK the only way to add child nodes is trough
> > quorum_add_child above and quorum_open and there already are adequate
> > checks there.
>
> I also don't see any other way to get that value, unless we change
> BDRVQuorumState to store that information (e.g. instead of children
> being a list of pointers BdrvChild ** it could be a list of {pointer,
> index}, or something like that).
>
> There's another (more convoluted) alternative if we don't want to parse
> child->name. Since we only want to know if the child number equals
> s->next_child_index - 1, we can do it the other way around:
>
>    snprintf(str, 32, "children.%u", s->next_child_index - 1);
>
> and then compare str and child->name.
>
> Berto

Hi,
I will do it your way, then it's also more consistent with the name
creation in quorum_add and quorum_open.

Regards,
Lukas Straub

Patch
diff mbox series

diff --git a/block/quorum.c b/block/quorum.c
index df68adcfaa..6100d4108a 100644
--- a/block/quorum.c
+++ b/block/quorum.c
@@ -1054,6 +1054,12 @@  static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
     /* We know now that num_children > threshold, so blkverify must be false */
     assert(!s->is_blkverify);

+    unsigned child_id;
+    sscanf(child->name, "children.%u", &child_id);
+    if (child_id == s->next_child_index - 1) {
+        s->next_child_index--;
+    }
+
     bdrv_drained_begin(bs);

     /* We can safely remove this child now */