On 3/25/14, 10:25 AM, Christoph Hellwig wrote:
> On Tue, Mar 25, 2014 at 09:05:04AM -0700, Eric Sandeen wrote:
>>>> Out of curiosity, is there any major reason we don't use 0 here
>>>> unconditionally? Are we worried about I/O completing before we have a
>>>> chance to decrement the reference?
>>>
>>> I think this should unconditionally avoid the schedule, and while we're
>>> at it we should kill _xfs_buf_ioend and opencode it here and at the
>>> other callsite.
>>
>> And then remove the flag from xfs_buf_ioend which is always 0 at that
>> point ...
>
> Is it? xfs_buf_bio_end_io should stil be passing 1, the bio end_io
> handler is the place we really need the workqueue for anyway.
These are the callers of xfs_buf_ioend:
File Function Line
0 xfs_buf.c xfs_bioerror 1085 xfs_buf_ioend(bp, 0);
1 xfs_buf.c _xfs_buf_ioend 1177 xfs_buf_ioend(bp, schedule);
2 xfs_buf_item.c xfs_buf_item_unpin 494 xfs_buf_ioend(bp, 0);
3 xfs_buf_item.c xfs_buf_iodone_callbacks 1138 xfs_buf_ioend(bp, 0);
4 xfs_inode.c xfs_iflush_cluster 3015 xfs_buf_ioend(bp, 0);
5 xfs_log.c xlog_bdstrat 1644 xfs_buf_ioend(bp, 0);
6 xfs_log_recover.c xlog_recover_iodone 386 xfs_buf_ioend(bp, 0);
so only _xfs_buf_ioend *might* pass something other than 0, and:
File Function Line
0 xfs_buf.c xfs_buf_bio_end_io 1197 _xfs_buf_ioend(bp, 1);
1 xfs_buf.c xfs_buf_iorequest 1377 _xfs_buf_ioend(bp, bp->b_error ? 0 : 1);
At least up until now that was always called with "1"
>> Yeah I have a patch to do that as well; I wanted to separate the
>> bugfix from the more invasive cleanup, though - and I wanted to
>> get the fix out for review sooner.
>
> Sure, feel free to leave all the cleanups to another patch.
>
>> But yeah, I was unsure about whether or not to schedule at all here.
>> We come here from a lot of callsites and I'm honestly not sure what
>> the implications are yet.
>
> I think the the delayed completion is always wrong from the submission
> path. The error path is just a special case of a completion happening
> before _xfs_buf_ioapply returns. The combination of incredibly fast
> hardware and bad preemption could cause the same bug you observed.
I wondered about that.
I'm not sure; I don't think it was the buf_rele inside xfs_buf_iorequest
that freed it, I think it was specifically the error path afterwards -
in my case, in xfs_trans_read_buf_map():
xfs_buf_ (bp);
// xfs_buf_iorequest code below
xfs_buf_hold(bp);
atomic_set(&bp->b_io_remaining, 1);
_xfs_buf_ioapply(bp); <-- gets error
if (atomic_dec_and_test(&bp->b_io_remaining)
xfs_buf_ioend(bp, bp->b_error ? 0 : 1);
xfs_buf_rele(bp); <- releases our hold
}
error = xfs_buf_iowait(bp); <-- sees error; would have waited otherwise
if (error) {
xfs_buf_ioerror_alert(bp, __func__);
xfs_buf_relse(bp); <--- freed here ?
but my bp refcounting & lifetime knowledge is lacking :(
-Eric
_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs
