Re: [Qemu-devel] Re: [RFC][PATCH v5 09/21] virtagent: add va.getdmesg RPC

[Michael Roth <mdroth@linux.vnet.ibm.com>]

On 12/08/2010 01:22 PM, Jes Sorensen wrote:
> On 12/07/10 18:32, Michael Roth wrote:
>> On 12/07/2010 08:37 AM, Jes Sorensen wrote:
>>> On 12/03/10 19:03, Michael Roth wrote:
>>>> +static xmlrpc_value *va_getdmesg(xmlrpc_env *env,
>>>> +                              xmlrpc_value *param,
>>>> +                              void *user_data)
>>>> +{
>>>> +    char *dmesg_buf = NULL, cmd[256];
>>>> +    int ret;
>>>> +    xmlrpc_value *result = NULL;
>>>> +    FILE *pipe;
>>>> +
>>>> +    SLOG("va_getdmesg()");
>>>> +
>>>> +    dmesg_buf = qemu_mallocz(VA_DMESG_LEN + 2048);
>>>> +    sprintf(cmd, "dmesg -s %d", VA_DMESG_LEN);
>>>
>>> What happens if the guest's dmesg buffer is larger than your hardcoded
>>> value?
>>
>> It'll end up getting truncated by the fread() later:
>>
>> ret = fread(dmesg_buf, sizeof(char), VA_DMESG_LEN, pipe);
>>
>> That's where the dmesg -s VA_DMESG_LEN comes into play, it should size
>> things such that we can buffer up till the end of the dmesg output.
>>
>> This param is kind of quirky though, size doesn't seem to have an affect
>> for anything below 4KB, but if we stick with VA_DMESG_LEN>= 4KB this
>> should cover us, unless it's a distro-specific. But it should blow
>> anything up, at least.
>
> I am wary of these hard coded constants. Isn't there a way to set the
> kernel's dmesg buffer size, or is that only a compile time option?
>

 From what I can tell it's a compile-time option. I originally had 
dmesg_len as a param the host could pass to the guest, but it has no 
effect if the buffer is smaller, which might cause unnecessary confusion.

> Cheers,
> Jes