On Fri, Oct 07, 2016 at 11:53:07AM +0100, Sitsofe Wheeler wrote:

On 6 October 2016 at 22:20, Shaohua Li [off-list ref] wrote:

quoted

Sitsofe,
could you try if this patch fixes the issue too please?

As with with the previous patch the BUG_ON/crash is fixed and as
before dmesg only says this now:
[   17.884175] sd 0:0:1:0: [sdb] tag#0 FAILED Result: hostbyte=DID_OK
driverbyte=DRIVER_SENSE
[   17.884180] sd 0:0:1:0: [sdb] tag#0 Sense Key : Illegal Request [current]
[   17.884185] sd 0:0:1:0: [sdb] tag#0 Add. Sense: Invalid command
operation code
[   17.884190] sd 0:0:1:0: [sdb] tag#0 CDB: Write same(16) 93 08 00 00
00 00 00 00 40 80 00 00 08 00 00 00
[   17.884196] blk_update_request: critical target error, dev sdb, sector 16512
[   17.884205] sd 0:0:2:0: [sdc] tag#1 FAILED Result: hostbyte=DID_OK
driverbyte=DRIVER_SENSE
[   17.884208] sd 0:0:2:0: [sdc] tag#1 Sense Key : Illegal Request [current]
[   17.884211] sd 0:0:2:0: [sdc] tag#1 Add. Sense: Invalid command
operation code
[   17.884215] sd 0:0:2:0: [sdc] tag#1 CDB: Write same(16) 93 08 00 00
00 00 00 00 40 80 00 00 08 00 00 00
[   17.884217] blk_update_request: critical target error, dev sdc, sector 16512
[   17.924716] Buffer I/O error on dev dm-0, logical block 0, async page read

(sd[bc] are the devices the RAID is using)

However this time around blkdiscard doesn't return an error (despite
the discard not being done) and is quiet. It's hard to say if this is
desirable given that the underlying devices switch to saying they
don't support discard but perhaps that's part of a different
discussion.

Yep, I think the spec doesn't require discard IO returns error. We probably can
make this smarter later in MD

Tested-by: Sitsofe Wheeler <redacted>

Thanks for the testing, I'll queue the two patches.

Thanks,
Shaohua