Re: [Qemu-devel] [Qemu-trivial] [PATCH 1/1] Do not hang on full PTY

[Don Slutz <dslutz@verizon.com>]

On 12/29/14 04:59, Michael Tokarev wrote:
> 22.12.2014 18:04, Don Slutz wrote:
> 
>> --- a/qemu-char.c
>> +++ b/qemu-char.c
>> @@ -1387,6 +1387,7 @@ static CharDriverState *qemu_chr_open_pty(const char *id,
>>      }
>>  
>>      close(slave_fd);
>> +    qemu_set_nonblock(master_fd);
>>  
>>      chr = qemu_chr_alloc();
> 
> 
> Hm.  I'm not sure at all this is a trivial change.  While the
> patch itself is trivial indeed, it changes behavour of the file
> descriptor significantly.  Are all the places where this fd is
> subsequently used prepared for it being non-blocking?  Oh well... ;)

I was not sure on this being trivial also, but it looked like it could
be to me.  The uses of this FD all looked that they handle non-blocking.

Here are the calls to qemu_set_nonblock in qemu-char.c (including this
add) :

b qemu-char.c             qemu_chr_open_fd           1070
qemu_set_nonblock(fd_out);
c qemu-char.c             qemu_chr_open_stdio        1166
qemu_set_nonblock(0);
d qemu-char.c             qemu_chr_open_pty          1390
qemu_set_nonblock(master_fd);
e qemu-char.c             tcp_chr_add_client         2952
qemu_set_nonblock(fd);
f qemu-char.c             qmp_chardev_open_serial    4060
qemu_set_nonblock(fd);
g qemu-char.c             qmp_chardev_open_socket    4172
qemu_set_nonblock(s->listen_fd);

So there are many cases where the FD is non-blocking already.

Hope this info helps.

> 
> Thanks,
> 
> /mjt
>