[dm-crypt] dm-crypt perf related changes and cryptsetup LUKS format failing

[dm-crypt] dm-crypt perf related changes and cryptsetup LUKS format failing

[Redwood Hyd]

Hi All ,

I did an experimental kernel change in linux/drivers/md/dm-crypt.c for
improving crypto HW performance.
Here is brief of it
static void crypt_io_hints(struct dm_target *ti,
                            struct queue_limits *limits)
{
  limits->physical_block_size = 32768;
  blk_limits_io_min(limits, 32768);
  blk_limits_io_opt(limits, 32768);
}

It works (passed iozone -a,  robocopy from samba client) and doubled
performance of dm-crypt WRITES !

Problem -
After above change cryptsetup luks format started failing disk
validations so additionally in
Cryptsetup-1.1.3  I  did  following changes to pass disk validation:
========================================================
/* Change-1 to overcome validation failure from dm-table.c that
"adding target device  caused an alignment inconsistency"  */
#define LUKS_STRIPES increased from  4000 to 524288

/* Change-2 to overcome validation failure in dm-table.c
"device-mapper: table: 253:1 start=8 not aligned to h/w logical block
size 32768 of dm-0" */
#define LUKS_ALIGN_KEYSLOTS  increased from  4096 to 32768

/* Change-3 Additionally did following change */
#define DEFAULT_ALIGNMENT    increased from  4096 to 32768

Question - Can someone recommend any later  version of cryptsetup
package where I don't need to change
                 cryptsetup and luks format can still pass
validate_hardware_logical_block_alignment() in dm-table.c  for 32k
encrypt block device ?

Performance related changes are experimental - If someone can suggest
things around this I will post my results to forum.

Regards and Thanks in advance

Re: [dm-crypt] dm-crypt perf related changes and cryptsetup LUKS format failing

[Milan Broz]

On 02/07/2014 02:53 PM, Redwood Hyd wrote:
> Hi All ,
> 
> I did an experimental kernel change in linux/drivers/md/dm-crypt.c for
> improving crypto HW performance.
> Here is brief of it
> static void crypt_io_hints(struct dm_target *ti,
>                             struct queue_limits *limits)
> {
>   limits->physical_block_size = 32768;
>   blk_limits_io_min(limits, 32768);
>   blk_limits_io_opt(limits, 32768);
> }
> 
> It works (passed iozone -a,  robocopy from samba client) and doubled
> performance of dm-crypt WRITES !

Well, increasing encryption block size was discussed several times.

Please search archive, e.g. here is some patch which increased block size:
http://thread.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/6441/focus=6587

(yes, your patch is just modification of io limits, which is simpler but has some
side effects - random io read/write will be worse I guess)


> Problem -
> After above change cryptsetup luks format started failing disk
> validations so additionally in
> Cryptsetup-1.1.3  I  did  following changes to pass disk validation:

Why are you using version released 4 years ago? :)

Try recent 1.6.3. It works without changes.

(Btw your changes to cryptsetup are not correct, it will just paperback over
real problem and cause LUKS format incompatibility.)

Actually, this is what I see after you patch (x is activated LUKS device):

# lsblk -t /dev/sdb

NAME ALIGNMENT MIN-IO OPT-IO PHY-SEC LOG-SEC ROTA SCHED RQ-SIZE  RA WSAME
sdb          0    512      0     512     512    1 cfq       128 128   32M
`-x          0  32768  32768   32768     512    1           128 128    0B

# lsblk -f /dev/sdb
NAME FSTYPE      LABEL UUID                                 MOUNTPOINT
sdb  crypto_LUKS       fc534366-6be5-46f3-9349-44dbe47f4b80 
`-x  ext4              6dfa13d0-2d0e-4955-a5c4-bd0848b07b47 /mnt/tst

so IO limits works...

Milan