Hi David,
David Woodhouse <[email protected]> writes:
> [[S/MIME Signed Part:Undecided]]
> On Tue, 2023-11-21 at 22:10 +0000, Volodymyr Babchuk wrote:
>>
>> --- a/hw/xen/xen-operations.c
>> +++ b/hw/xen/xen-operations.c
>> @@ -300,6 +300,18 @@ static bool libxenstore_create(struct qemu_xs_handle
>> *h, xs_transaction_t t,
>> return false;
>> }
>>
>> + if (owner == XS_PRESERVE_OWNER) {
>> + struct xs_permissions *tmp;
>> + unsigned int num;
>> +
>> + tmp = xs_get_permissions(h->xsh, t, path, &num);
>> + if (tmp == NULL) {
>> + return false;
>> + }
>> + perms_list[0].id = tmp[0].id;
>> + free(tmp);
>> + }
>> +
>> return xs_set_permissions(h->xsh, t, path, perms_list,
>> ARRAY_SIZE(perms_list));
>> }
>
> If the existing transaction is XBT_NULL I think you want to create a
> new transaction for it, don't you?
I must say that your comment is valid even without my
changes. xenstore_mkdir() calls qemu_xen_xs_create, providing just plain
"0" (not even XBT_NULL) as a transaction, but actual xenstore interface
implementation, like xs_be_create(), issue multiple calls to lower
layer, passing "t" downwards. For example, xs_be_create() calls
xs_impl_read, xs_impl_write and xs_impl_set_perms(). If called from
xesntore_mkdir(), those three operations will be non-atomic. I don't
know if this can lead to a problem, because apparently it was so for a
long time...
--
WBR, Volodymyr
