Re: [Qemu-devel] [PATCH 09/10] block: Reuse success path from bdrv_open()

[Max Reitz <mreitz@redhat.com>]

On 27.01.2014 18:44, Jeff Cody wrote:
> On Sun, Jan 26, 2014 at 08:02:42PM +0100, Max Reitz wrote:
>> The fail and success paths of bdrv_file_open() may be further shortened
>> by reusing code already existent in bdrv_open(). This includes
>> bdrv_file_open() not taking the reference to options which allows the
>> removal of QDECREF(options) in that function.
>>
>> Signed-off-by: Max Reitz <mreitz@redhat.com>
>> ---
>>   block.c | 33 ++++++++++++---------------------
>>   1 file changed, 12 insertions(+), 21 deletions(-)
>>
>> diff --git a/block.c b/block.c
>> index f847c4b..b1bae23 100644
>> --- a/block.c
>> +++ b/block.c
>> @@ -943,9 +943,7 @@ free_and_fail:
>>    * Opens a file using a protocol (file, host_device, nbd, ...)
>>    *
>>    * options is a QDict of options to pass to the block drivers, or NULL for an
>> - * empty set of options. The reference to the QDict belongs to the block layer
>> - * after the call (even on failure), so if the caller intends to reuse the
>> - * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
>> + * empty set of options.
>>    */
>>   static int bdrv_file_open(BlockDriverState *bs, const char *filename,
>>                             QDict *options, int flags, Error **errp)
>> @@ -1010,8 +1008,8 @@ static int bdrv_file_open(BlockDriverState *bs, const char *filename,
>>       }
>>   
>>       if (!drv->bdrv_file_open) {
>> +        QINCREF(options);
>>           ret = bdrv_open(&bs, filename, NULL, options, flags, drv, &local_err);
>> -        options = NULL;
>>       } else {
>>           ret = bdrv_open_common(bs, NULL, options, flags, drv, &local_err);
>>       }
>> @@ -1020,21 +1018,10 @@ static int bdrv_file_open(BlockDriverState *bs, const char *filename,
>>           goto fail;
>>       }
>>   
>> -    /* Check if any unknown options were used */
>> -    if (options && (qdict_size(options) != 0)) {
>> -        const QDictEntry *entry = qdict_first(options);
>> -        error_setg(errp, "Block protocol '%s' doesn't support the option '%s'",
>> -                   drv->format_name, entry->key);
>> -        ret = -EINVAL;
>> -        goto fail;
>> -    }
>> -    QDECREF(options);
>> -
>>       bs->growable = 1;
>>       return 0;
>>   
>>   fail:
>> -    QDECREF(options);
>>       return ret;
>>   }
>>   
>> @@ -1238,7 +1225,6 @@ int bdrv_open(BlockDriverState **pbs, const char *filename,
>>           assert(!drv);
>>           ret = bdrv_file_open(bs, filename, options, flags & ~BDRV_O_PROTOCOL,
>>                                &local_err);
>> -        options = NULL;
>>           if (ret) {
>>               if (bs->drv) {
>>                   goto close_and_fail;
>> @@ -1246,8 +1232,7 @@ int bdrv_open(BlockDriverState **pbs, const char *filename,
>>                   goto fail;
>>               }
>>           }
>> -        *pbs = bs;
>> -        return 0;
>> +        goto done;
>>       }
>>   
>>       /* For snapshot=on, create a temporary qcow2 overlay */
>> @@ -1377,12 +1362,18 @@ int bdrv_open(BlockDriverState **pbs, const char *filename,
>>           }
>>       }
>>   
>> +done:
>>       /* Check if any unknown options were used */
>>       if (qdict_size(options) != 0) {
>>           const QDictEntry *entry = qdict_first(options);
>> -        error_setg(errp, "Block format '%s' used by device '%s' doesn't "
>> -                   "support the option '%s'", drv->format_name, bs->device_name,
>> -                   entry->key);
>> +        if (flags & BDRV_O_PROTOCOL) {
>> +            error_setg(errp, "Block protocol '%s' doesn't support the option "
>> +                       "'%s'", drv->format_name, entry->key);
> Tests 071 and 072 segfault, and gdb shows that it occurs here.  More
> investigation shows that entry is NULL, although qdict_size() is
> returning 1.

Hm, and I thought I'd run the tests, but obviously I did not for this 
final version… The problem is probably the QINCREF(). Hm, I really want 
to reuse this test; the only way I see then is to make the options 
parameter of bdrv_file_open() an indirect pointer so it can be set to 
NULL after bdrv_open() (in bdrv_file_open()). That would not be very 
pretty, but it would work and there wouldn't be two tests whether all 
options could be handled.

Thanks for catching this,

Max

>> +        } else {
>> +            error_setg(errp, "Block format '%s' used by device '%s' doesn't "
>> +                       "support the option '%s'", drv->format_name,
>> +                       bs->device_name, entry->key);
>> +        }
>>   
>>           ret = -EINVAL;
>>           goto close_and_fail;
>> -- 
>> 1.8.5.3
>>
>>