Comment 8 for bug 444915

@pitti -- that original fix in 2.6.28 timeframe appears to have been adding readahead to FAT chain scanning in fat_count_free_clusters() in the commit below:

  commit 9f966be8996f2829406324c68e4c67c2d64d864b
  Author: OGAWA Hirofumi <email address hidden>
  Date: Tue Jan 8 15:32:41 2008 -0800

    fat: optimize fat_count_free_clusters()

That change does not appear to have been undone specifically still appearing in the Karmic kernel.

Doing a quick test here with a 500GB USB drive I am not seeing any particularly spectacularly huge delay (this is with 2.6.31-13.43-generic):

    # echo 3 >/proc/sys/vm/drop_caches
    # mount /dev/sdc1 /mnt; time df -h
    [...]
    /dev/sdc1 459G 45G 391G 11% /mnt

    real 0m0.012s
    user 0m0.004s
    sys 0m0.000s