Am 14.08.2012 13:13, schrieb Supriya Kannery:
> On 08/10/2012 07:15 PM, Corey Bryant wrote:
>>
>>
>> On 07/30/2012 05:34 PM, Supriya Kannery wrote:
>
>>> +static int raw_reopen_prepare(BlockDriverState *bs, BDRVReopenState
>>> **prs,
>>> + int flags)
>>> +{
>>> + BDRVRawReopenState *raw_rs = g_malloc0(sizeof(BDRVRawReopenState));
>>> + BDRVRawState *s = bs->opaque;
>>> + int ret = 0;
>>> +
>>> + raw_rs->reopen_state.bs = bs;
>>> +
>>> + /* stash state before reopen */
>>> + raw_rs->stash_s = g_malloc0(sizeof(BDRVRawState));
>>> + raw_stash_state(raw_rs->stash_s, s);
>>> + s->fd = dup3(raw_rs->stash_s->fd, s->fd, O_CLOEXEC);
>>> +
>>> + *prs = &(raw_rs->reopen_state);
>>> +
>>> + /* Flags that can be set using fcntl */
>>> + int fcntl_flags = BDRV_O_NOCACHE;
>>> +
>>> + if ((bs->open_flags & ~fcntl_flags) == (flags & ~fcntl_flags)) {
>>> + if ((flags & BDRV_O_NOCACHE)) {
>>> + s->open_flags |= O_DIRECT;
>>> + } else {
>>> + s->open_flags &= ~O_DIRECT;
>>> + }
>>> + ret = fcntl_setfl(s->fd, s->open_flags);
>>> + } else {
>>> +
>>> + /* close and reopen using new flags */
>>> + bs->drv->bdrv_close(bs);
>>> + ret = bs->drv->bdrv_file_open(bs, bs->filename, flags);
>>
>> Will this allow the fdset refcount to get to zero? I was hoping your
>> patches would prevent that from happening. Perhaps Kevin or Eric can
>> weigh in. qemu_open() increments the refcount for an fdset when an fd
>> from it is used, and qemu_close() decrements it. I think if you were
>> able to perform the open before the close here that refcount wouldn't
>> get to zero.
>>
>
> Since we are duping the file descriptor before reaching this bdrv_close(),
> refcount for fd won't become zero.
We need to use a qemu_dup() here, so that the fdset implementation can
keep track of the new fd.
Kevin
