On 04/13/2016 09:43 AM, Martin K. Petersen wrote:
"Bart" == Bart Van Assche <bart.vanassche@xxxxxxxxxxx> writes:
Bart> Today if a BLKDISCARD ioctl passes a non-aligned start and/or end
Bart> sector to the kernel then the block layer will submit invalid
Bart> (non-aligned) REQ_DISCARD requests to the block driver the ioctl
Bart> applies to.
I do not know what you mean by "invalid".
A device that implements UNMAP can freely ignore any parts of a request
block range that are not aligned to its internal allocation units.
From SBC: "An unmap request with a number of logical blocks that is not
a multiple of this value (OPTIMAL UNMAP GRANULARITY) may result in unmap
operations on fewer LBAs than requested." and "An unmap request with a
starting LBA that is not optimal may result in unmap operations on fewer
LBAs than requested."
The same is true for SATA which has no way report granularity and
alignment at all.
Nowhere does it say that a request that is not an aligned multiple of
any reported granularity should be considered "invalid" or rejected by
the storage.
Hello Martin,
Sorry if I wasn't clear enough. Regarding the SCSI UNMAP command,
consider e.g. the following code from drivers/scsi/sd.c:
static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
{
[ ... ]
sector >>= ilog2(sdp->sector_size) - 9;
nr_sectors >>= ilog2(sdp->sector_size) - 9;
[ ... ]
case SD_LBP_UNMAP:
[ ... ]
cmd->cmd_len = 10;
cmd->cmnd[0] = UNMAP;
cmd->cmnd[8] = 24;
put_unaligned_be16(6 + 16, &buf[0]);
put_unaligned_be16(16, &buf[2]);
put_unaligned_be64(sector, &buf[8]);
put_unaligned_be32(nr_sectors, &buf[16]);
[ ... ]
}
For sdp->sector_size > 512, should we allow the block layer to pass
sector and nr_sector values to this function that are not a multiple of
sdp->sector_size? And if so, how should this code behave and if sector
and/or nr_sectors is not a multiple of sdp->sector_size? As one can see
the above code rounds down sector and nr_sectors while converting from
sectors to logical blocks. This means that if a non-aligned BLKDISCARD
is submitted from user space that one or more sectors *before* the start
of the range specified in the ioctl will be discarded. Isn't that a bug?
Bart.
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
