dtrace discuss - Feb 2008 - Mapping device major, minor numbers to the related link under /dev/[r]dsk/

When using DTrace to track I/Os to disks under the sd/ssd driver, it''s
easy enough to  get the major/minor device number of the disk being targeted for
I/O:

fbt:ssd:ssdstrategy:entry,
fbt:sd:sdstrategy:entry
{
  io_start[(struct buf *)arg0] = timestamp;
}

fbt:ssd:ssdintr:entry,
fbt:sd:sdintr:entry
/ io_start[(this->buf = (struct buf *)((struct scsi_pkt
*)arg0)->pkt_private)]
  != 0 /
{
  this->un = ((struct sd_xbuf *) this->buf->b_private)->xb_un;
  this->major = getmajor(this->buf->b_edev);
  this->minor = getminor(this->buf->b_edev);

  @max_ms[this->major, this->minor]     max((timestamp -
io_start[this->buf])/1000000); /* convert ns to ms */

  io_start[this->buf] = 0;
}

We can even get device names in the form of ''ssdXXX'', which
unfortunately aren''t that helpful.

How can we map major/minor numbers or device names to their /dev/[r]dsk/...
symlinks in DTrace?   That would make the output more intuitive, at least to us.

Gordon Marler
gmarler at gmarler.com


--
This message posted from opensolaris.org

Gordon Marler writes:
 > When using DTrace to track I/Os to disks under the sd/ssd driver,
it''s easy enough to  get the major/minor device number of the disk
being targeted for I/O:
 > 
 > fbt:ssd:ssdstrategy:entry,
 > fbt:sd:sdstrategy:entry
 > {
 >   io_start[(struct buf *)arg0] = timestamp;
 > }
 > 
 > fbt:ssd:ssdintr:entry,
 > fbt:sd:sdintr:entry
 > / io_start[(this->buf = (struct buf *)((struct scsi_pkt
*)arg0)->pkt_private)]
 >   != 0 /
 > {
 >   this->un = ((struct sd_xbuf *) this->buf->b_private)->xb_un;
 >   this->major = getmajor(this->buf->b_edev);
 >   this->minor = getminor(this->buf->b_edev);
 > 
 >   @max_ms[this->major, this->minor]  >     max((timestamp -
io_start[this->buf])/1000000); /* convert ns to ms */
 > 
 >   io_start[this->buf] = 0;
 > }
 > 
 > We can even get device names in the form of ''ssdXXX'',
which unfortunately aren''t that helpful.
 > 
 > How can we map major/minor numbers or device names to their
/dev/[r]dsk/... symlinks in DTrace?   That would make the output more intuitive,
at least to us.
 > 
 > Gordon Marler
 > gmarler at gmarler.com
 > 
 > 
 > --
 > This message posted from opensolaris.org
 > _______________________________________________
 > dtrace-discuss mailing list
 > dtrace-discuss at opensolaris.org


Here is an example script where I got this working :


#!/usr/bin/ksh
#
# iolat - print avg latency and distribution of each disk
#
# This scripts prints average latency of each disk over the length of the
# run. With -d, results includes an lquantization of the result. Scripts
# runs until interrupted. Results are in usec (actually nsec >>10).
#
# If you get "DIF program exceeds maximum program size"
# Try : echo dtrace_dof_maxsize/Z1000000 | mdb -kw 
#

opt_dist=0; min=0; max=50000; bin=5000;

### process options
while getopts dh name
do
	case $name in
	d)	opt_dist=1 ;;
	h|?)	cat <<-END >&2
		USAGE: iolat [-d [min max bucket]]

		            -d              # report per disk latency distribution.
		            min max bucket  # arguments to dtrace lquantize in usec.
		                            # Default : 0, 5000, 500.
		END
		exit 1
	esac
done

shift $(( $OPTIND - 1 ))

### option logic
if [[ ! "$1" == "" ]]; then
        min=$1; shift
fi
if [[ ! "$1" == "" ]]; then
        max=$1; shift
fi
if [[ ! "$1" == "" ]]; then
        bin=$1; shift
fi

#Build a string to be parsed by dtrace
#Contains entry of the form 
#          disks[major, minor] = "c0t0d0s0"; 
#
disks_arr_string=`/bin/ls -1lL /dev/dsk/* 2>/dev/null | /usr/bin/nawk
-F''[ ,\t/]*'' ''{print "disks[" $5 ",
" $6 "] = \"" $12 "\";"}''`

if [[ $opt_dist == 1 ]]; then
    quantize_gather="@d[this->disk] = lquantize(this->lat, $min,
$max, $bin);"
    quantize_print="printf(\"\n\nLatency Distribution:\n\n\");
printa(\"%20s   %@10d\n\", @d);"
else
    quantize_gather=""
    quantize_print=""
fi


script_str=''

BEGIN {
    ''$disks_arr_string'';
    printf("Ctrl-C to terminate and get results\n");
}

io:::start{
	started[args[0]->b_edev, args[0]->b_blkno] = timestamp;
}

io:::done
/started[args[0]->b_edev, args[0]->b_blkno] != 0/
{

 
	this->major = args[1]->dev_major;
	this->minor = args[1]->dev_minor;
	this->disk = (disks[this->major, this->minor] == "" ?
"unknown" : disks[this->major, this->minor]);
	this->lat = (timestamp - started[args[0]->b_edev,
args[0]->b_blkno])>>10;

	@a[this->disk] = avg(this->lat); /* First is sort key */

	@b[this->disk] = count();
	@c[this->disk] = sum(args[0]->b_bcount);
	''$quantize_gather''

	started[args[0]->b_edev, args[0]->b_blkno] = 0;
}

END {
	printf("%20s   %10s %10s %10s\n", "Disk", "Avg
Lat(us)", "IO cnt", "Bytes");
        printf("%20s   %10s %10s %10s\n", "----",
"----------", "------", "-----");
	printa("%20s   %@10d %@10d %@10d\n", @a, @b, @c);
	''$quantize_print''
}
''

/usr/bin/pfexec /usr/sbin/dtrace  -qn "$script_str"

Very nice example.  Thanks for the follow up.

Seems to work well, just had to bump dtrace_dof_maxsize up to 512K with mdb to
get it to work on Solaris 10.


--
This message posted from opensolaris.org