[Libguestfs] [nbdkit PATCH] nbd: Opt in to libnbd pread_initialize speedup

[Laszlo Ersek]

On 02/11/22 15:35, Eric Blake wrote:
> On Fri, Feb 11, 2022 at 07:14:30AM +0100, Laszlo Ersek wrote:
>> On 02/10/22 16:38, Eric Blake wrote:
>>> Our nbd plugin has always properly checked for asynch errors (and thus
>>> has never been at risk of a vulnerability similar to CVE-2022-0485
>>> just fixed in nbdcopy).  What's more, the nbdkit core guarantees
>>> (since commit b1ce255e in 2019, v1.13.1) that the buffer handed to a
>>> .pread callback has been pre-sanitized to not leak heap contents,
>>> regardless of how buggy a plugin might be (see
>>> server/protocol.c:protocol_recv_request_send_reply(), which uses
>>> server/threadlocal.c:threadlocal_buffer() for a safe buffer).  Thus,
>>> we do not need libnbd to spend time re-initializing the buffer, if
>>> libnbd is new enough to give us that knob.
>>> ---
>>>
> 
>>> +#if LIBNBD_HAVE_NBD_SET_PREAD_INITIALIZE
>>> +  /* nbdkit guarantees that the buffers passed to our .pread callback
>>> +   * are pre-initialized; and we in turn ensure that the buffer is not
>>> +   * dereferenced if the NBD server replied with an error.  Thus, we
>>> +   * are safe opting in to this libnbd speedup.
>>> +   */
>>> +  if (nbd_set_pread_initialize (h->nbd, false) == -1)
>>> +    goto errnbd;
>>>  #endif
>>>    if (dynamic_export && uri) {
>>>  #if LIBNBD_HAVE_NBD_SET_OPT_MODE
>>>
>>
>> hahaha, this was real quick ;)
> 
> I don't know if the speedy pun was intentional ;)

Heh, no, I didn't notice that coincidence :)

> 
> Now pushed as 31997b25.
>