Re: [Qemu-devel] [PATCH 2/3] qmp: dump-guest-memory: don't spin if non-blocking fd would block

[Wen Congyang]

At 09/25/2012 05:01 PM, Markus Armbruster Wrote:
> Wen Congyang <address@hidden> writes:
> 
>> At 09/24/2012 09:34 PM, Luiz Capitulino Wrote:
>>> On Mon, 24 Sep 2012 14:27:17 +0800
>>> Wen Congyang <address@hidden> wrote:
>>>
>>>> At 09/22/2012 01:07 AM, Luiz Capitulino Wrote:
>>>>> fd_write_vmcore() will indefinitely spin for a non-blocking
>>>>> file-descriptor that would block. However, if the fd is non-blocking,
>>>>> how does it make sense to spin?
>>>>>
>>>>> Change this behavior to return an error instead.
>>>>>
>>>>> Note that this can only happen with an fd provided by a management
>>>>> application. The fd opened internally by dump-guest-memory is blocking.
>>>>>
>>>>> Signed-off-by: Luiz Capitulino <address@hidden>
>>>>> ---
>>>>>  dump.c | 13 +++----------
>>>>>  1 file changed, 3 insertions(+), 10 deletions(-)
>>>>>
>>>>> diff --git a/dump.c b/dump.c
>>>>> index 2bf8d8d..5eea015 100644
>>>>> --- a/dump.c
>>>>> +++ b/dump.c
>>>>> @@ -100,18 +100,11 @@ static void dump_error(DumpState *s, const char 
>>>>> *reason)
>>>>>  static int fd_write_vmcore(void *buf, size_t size, void *opaque)
>>>>>  {
>>>>>      DumpState *s = opaque;
>>>>> -    int fd = s->fd;
>>>>>      size_t writen_size;
>>>>>  
>>>>> -    /* The fd may be passed from user, and it can be non-blocked */
>>>>> -    while (size) {
>>>>> -        writen_size = qemu_write_full(fd, buf, size);
>>>>> -        if (writen_size != size && errno != EAGAIN) {
>>>>
>>>> Hmm, if the fd is a blocking fd, errno can't be EAGAIN. So the
>>>> function doesn't spin. What problems do you meet?
>>>
>>> The problem is with non-blocking fds, where spinning isn't correct, for
>>> two reasons:
>>
>> But, If the fd is non-blocking, errno can't be EAGAIN. So it doesn't spin.
> 
> I'm afraid you're confused.
> 
> http://pubs.opengroup.org/onlinepubs/009695399/functions/write.html
> 
> [EAGAIN] The O_NONBLOCK flag is set for the file descriptor and the
> thread would be delayed in the write() operation.
> 

Ahh, you are right.

Thanks
Wen Congyang