Bart,
On 2/16/17 01:42, Bart Van Assche wrote:
> On Wed, 2017-02-15 at 11:12 +0900, Damien Le Moal wrote:
>> diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
>> index 1f5d92a..d05a328 100644
>> --- a/drivers/scsi/sd.c
>> +++ b/drivers/scsi/sd.c
>> @@ -1790,6 +1790,8 @@ static int sd_done(struct scsi_cmnd *SCpnt)
>> {
>> int result = SCpnt->result;
>> unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
>> + unsigned int sector_size = SCpnt->device->sector_size;
>> + unsigned int resid;
>> struct scsi_sense_hdr sshdr;
>> struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
>> struct request *req = SCpnt->request;
>> @@ -1820,6 +1822,24 @@ static int sd_done(struct scsi_cmnd *SCpnt)
>> scsi_set_resid(SCpnt, blk_rq_bytes(req));
>> }
>> break;
>> + default:
>> + /*
>> + * In case of bogus fw or device, we could end up having
>> + * an unaligned partial completion. Check this here and force
>> + * alignment.
>> + */
>> + resid = scsi_get_resid(SCpnt);
>> + if (resid & (sector_size - 1)) {
>> + SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
>> + "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
>> + resid, sector_size));
>> + resid = round_up(resid, sector_size);
>> + if (resid < good_bytes)
>> + good_bytes -= resid;
>> + else
>> + good_bytes = 0;
>> + scsi_set_resid(SCpnt, resid);
>> + }
>> }
>>
>> if (result) {
>
> An additional concern: what if the size of the Data-Out buffer is not a
> multiple of the logical block size? Shouldn't we round down (good_bytes -
> resid) instead of rounding up resid?
The only REQ_TYPE_FS request that may not operate on LBA size aligned
payloads is REQ_OP_ZONE_REPORT which is handles in a different case of
the switch statement. So I think it is safe.
Best regards.
--
Damien Le Moal, Ph.D.
Sr. Manager, System Software Research Group,
Western Digital Corporation
Damien.LeMoal@xxxxxxx
(+81) 0466-98-3593 (ext. 513593)
1 kirihara-cho, Fujisawa,
Kanagawa, 252-0888 Japan
www.wdc.com, www.hgst.com
