#!/bin/sh

# Log total disk activity.
# This script logs disk activity, with the aim of helping to guage flash disk 
# lifetimes.  (Flash disks supposedly support limited numbers of write cycles.)

# The kernel records the total number of sectors read and written for each disk 
# since the last boot for us.  We need to accumulate this data.  But we don't want 
# to rely on the system always being shut down cleanly in order to save the data.  
# Instead we'll run this script from cron, so that in the event of an unclean 
# shutdown we have lost only a limited amount of data.  We can also create an init
# script to run it on shutdown.

# We'll record the time of the last boot with each record written.  In that way, 
# when computing the totals we can discard all but the last record with each boot 
# time.

# Log file(s) where the records will be written.  There's not much point in keeping 
# this only on the disk that you're monitoring as the data is most interesting when 
# the disk has failed.  You can therefore specify multiple log files, all of which 
# will get all of the data.
hostname=`hostname`
logfiles="/var/log/diskstats-${hostname} /home/diskstats-${hostname}.log"


# Get the time of the last boot, in seconds-since-the-epoch.
# Is "last" the best way to do this?  Its output is not straightforward to parse.  I 
# had hoped to find something useful in /proc, but I can't find what I need.
# The output of last looks like this:

# reboot   system boot  2.6.19           Tue Dec 12 08:56 - 20:00  (11:04)    
# 
# wtmp begins Sat Apr  1 01:20:03 2006

# This gets the "Tue Dec 12 08:56" portion:
boottime_fmt=`last -1 reboot | awk 'NR==1 {print $5,$6,$7,$8}'`

# Now convert to seconds-since-the-epoch:
boottime_secs=`date +%s -d "${boottime_fmt}"`


# Get the disk statistics.
# See Documentation/iostats.txt for details of the formats.

{

case `uname -r` in
2.6.*)

# For kernel 2.6, typical data from /proc/diskstats is:

#   3    0 hda 23013 2383 430466 63692 7483 5929 107344 646004 0 229336 709696
#   3    1 hda1 25379 430266 13418 107344

# The first two fields are the device major/minor numbers.  Whole disks have more 
# fields than partitions, and we're interested in whole disks.  We're also not 
# interested in ram disks and mtd blocks, which are filtered out based on their 
# major numbers of 1 and 31 respectively.

# The number of sectors read and written are in the 6th and 10th fields respectively.

cat /proc/diskstats | awk '$1!=1 && $1!=31 && NF==14 { print $3, $6, $10; }' ;;

2.4.*)

For kernel 2.4, typical data from /proc/partitions is:

#   3     0   39082680 hda 446216 784926 9550688 4382310 424847 312726 5922052 19310380 0 3376340 23705160
#   3     1    9221278 hda1 35486 0 35496 38030 0 0 0 0 0 38030 38030

# This is similar to the 2.6 format but with an extra field that we have to skip, 
# and with equally-many fields for disks and partitions.  I think that the best way 
# to find only whole disks is using the device minor number.

# I have not tested this.  The only 2.4 machine I have is a UML virtual machine, and 
# it doesn't seem to have any of these fields in /proc/partitions.

cat /proc/partitions | awk '$1!=1 && $1!=31 && ($2%8)==0 { print $4, $7, $11; }' ;;

*)

echo "Sorry, your kernel is not supported."
exit 1

esac

} | {

# Read that data disk at a time

while read disk reads writes
do

  # We really need to map from disk names to IDs of some sort, since the names can 
  # change from one boot to the next depending on driver startup order.
  # We have a choice of UUIDs or IDs.  I'm not sure of the difference.  IDs look 
  # more inteligable, but are they unique?  If you choose to use UUIDs instead, make 
  # sure you keep an independent record of what disks they correspond to.
  # I can't find a better way of getting this data than searching the links in 
  # /dev/disk: where does udev get the IDs from when it makes the links?

  devnode="/dev/${disk}"
  disk_id=`find -L /dev/disk/by-id -samefile "${devnode}" -printf '%f\n'`

  # Put together a log file entry:
  logline="${boottime_secs} \"${disk_id}\" ${reads} ${writes}"

  # Append it to each logfile:
  for f in ${logfiles}
  do
    echo ${logline} >> $f
  done

done

}