[Qemu-devel] Re: [PATCH 07/13] blockdev: Means to destroy blockdev only if made with drive_init()

[Markus Armbruster <armbru@redhat.com>]

Kevin Wolf <kwolf@redhat.com> writes:

> Am 02.06.2010 18:55, schrieb Markus Armbruster:
>> All drives are still made that way.  They get destroyed along with
>> their device.  That's inappropriate for the alternative way to make
>> blockdevs that will appear later in this series.  These won't have a
>> DriveInfo.
>> 
>> blockdev_detach() destroys the blockdev only if it has a DriveInfo.
>> 
>> blockdev_attach() does nothing for now.  It'll be fleshed out later.
>> 
>> Signed-off-by: Markus Armbruster <armbru@redhat.com>
>> ---
>>  blockdev.c |   35 +++++++++++++++++++++++++++++++++++
>>  blockdev.h |    7 +++++++
>>  2 files changed, 42 insertions(+), 0 deletions(-)
>> 
>> diff --git a/blockdev.c b/blockdev.c
>> index ace74e4..f90d4fc 100644
>> --- a/blockdev.c
>> +++ b/blockdev.c
>> @@ -1,8 +1,12 @@
>>  /*
>>   * QEMU host block devices
>>   *
>> + * Copyright (C) 2010 Red Hat Inc.
>>   * Copyright (c) 2003-2008 Fabrice Bellard
>>   *
>> + * Authors:
>> + *  Markus Armbruster <armbru@redhat.com>,
>> + *
>>   * This work is licensed under the terms of the GNU GPL, version 2 or
>>   * later.  See the COPYING file in the top-level directory.
>>   */
>> @@ -17,6 +21,37 @@
>>  
>>  static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
>>  
>> +static int blockdev_del_dinfo(BlockDriverState *bs)
>> +{
>> +    DriveInfo *dinfo, *next_dinfo;
>> +    int res = 0;
>> +
>> +    QTAILQ_FOREACH_SAFE(dinfo, &drives, next, next_dinfo) {
>> +        if (dinfo->bdrv == bs) {
>> +            qemu_opts_del(dinfo->opts);
>> +            QTAILQ_REMOVE(&drives, dinfo, next);
>> +            qemu_free(dinfo);
>> +            res = 1;
>> +        }
>> +    }
>> +
>> +    return res;
>
> Can it happen that a BlockDriverState belongs to multiple DriveInfos? If
> no, why not returning in the loop? Wouldn't need a FOREACH_SAFE then, too.

No, that shouldn't happen.  Defensive coding, I don't want to leave
dinfos with dangling dinfo->bdrv around.  Maybe I should put an
assert(!res) before the qemu_opts_del().  Or just forget about it, and
simplify like you suggest.

> It's not worth respinning because of this one, but there were more
> comments and I think you'll send a v2 for the actual -blockdev option
> anyway once we have decided how to do it.
>
> I have applied patches 1 to 6 now, and I think I could safely go on
> until patch 9 if the minor improvements that were mentioned in comments
> are made. I'd ignore patch 10 to 13 for now.
>
> Is this what you would have expected or should I handle anything in a
> different way?

No, that suits me fine.  I definitely need to respin from part 8 on
(commit message too terse).