On Sat, Aug 4, 2012 at 2:44 AM, Bharata B Rao
<bhar...@linux.vnet.ibm.com> wrote:
> On Fri, Aug 03, 2012 at 03:57:20PM +0000, Blue Swirl wrote:
>> >> > +static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void
>> >> > *arg)
>> >> > +{
>> >> > + GlusterAIOCB *acb = (GlusterAIOCB *)arg;
>> >> > + BDRVGlusterState *s = acb->common.bs->opaque;
>> >> > +
>> >> > + acb->ret = ret;
>> >> > + if (qemu_gluster_send_pipe(s, acb) < 0) {
>> >> > + error_report("Could not complete read/write/flush from
>> >> > gluster");
>> >> > + abort();
>> >>
>> >> Aborting is a bit drastic, it would be nice to save and exit gracefully.
>> >
>> > I am not sure if there is an easy way to recover sanely and exit from this
>> > kind of error.
>> >
>> > Here the non-QEMU thread (gluster thread) failed to notify the QEMU thread
>> > on the read side of the pipe about the IO completion. So essentially
>> > bdrv_read or bdrv_write will never complete if this error happens.
>> >
>> > Do you have any suggestions on how to exit gracefully here ?
>>
>> Ignore but set the callback return to -EIO, see for example curl.c:249.
>
> I see the precedence for how I am handling this in
> posix-aio-compat.c:posix_aio_notify_event().
>
> So instead of aborting, I could do acb->common.cb(acb->common.opaque, -EIO)
> as you suggest, but that would not help because, the thread at the read side
> of the pipe is still waiting and user will see the read/write failure as hang.
Probably the other side needs to be informed somehow.
Maybe it's enough for 1.2 to just use abort() and add a FIXME comment.
>
> [root@bharata qemu]# gdb ./x86_64-softmmu/qemu-system-x86_64
> Starting program: ./x86_64-softmmu/qemu-system-x86_64 --enable-kvm
> --nographic -m 1024 -smp 4 -drive
> file=gluster://bharata/test/F16,if=virtio,cache=none
> [New Thread 0x7ffff4c7f700 (LWP 6537)]
> [New Thread 0x7ffff447e700 (LWP 6538)]
> [New Thread 0x7ffff3420700 (LWP 6539)]
> [New Thread 0x7ffff1407700 (LWP 6540)]
> qemu-system-x86_64: -drive
> file=gluster://bharata/test/F16,if=virtio,cache=none: Could not complete
> read/write/flush from gluster
> ^C
> Program received signal SIGINT, Interrupt.
> 0x00007ffff60e9403 in select () from /lib64/libc.so.6
> (gdb) bt
> #0 0x00007ffff60e9403 in select () from /lib64/libc.so.6
> #1 0x00005555555baee3 in qemu_aio_wait () at aio.c:158
> #2 0x00005555555cf57b in bdrv_rw_co (bs=0x5555564cfa50, sector_num=0, buf=
> 0x7fffffffb640 "\353c\220", nb_sectors=4, is_write=false) at block.c:1623
> #3 0x00005555555cf5e1 in bdrv_read (bs=0x5555564cfa50, sector_num=0, buf=
> 0x7fffffffb640 "\353c\220", nb_sectors=4) at block.c:1633
> #4 0x00005555555cf9d0 in bdrv_pread (bs=0x5555564cfa50, offset=0,
> buf=0x7fffffffb640,
> count1=2048) at block.c:1720
> #5 0x00005555555cc8d4 in find_image_format (filename=
> 0x5555564cc290 "gluster://bharata/test/F16", pdrv=0x7fffffffbe60) at
> block.c:529
> #6 0x00005555555cd303 in bdrv_open (bs=0x5555564cef20, filename=
> 0x5555564cc290 "gluster://bharata/test/F16", flags=98, drv=0x0) at
> block.c:800
> #7 0x0000555555609f69 in drive_init (opts=0x5555564cf900, default_to_scsi=0)
> at blockdev.c:608
> #8 0x0000555555711b6c in drive_init_func (opts=0x5555564cc1e0,
> opaque=0x555555c357a0)
> at vl.c:775
> #9 0x000055555574ceda in qemu_opts_foreach (list=0x555555c319e0, func=
> 0x555555711b31 <drive_init_func>, opaque=0x555555c357a0,
> abort_on_failure=1)
> at qemu-option.c:1094
> #10 0x0000555555719d78 in main (argc=9, argv=0x7fffffffe468,
> envp=0x7fffffffe4b8)
> at vl.c:3430
>
