[Yaffs] Lost blocks

Top Page
Attachments:
Message as email
+ (text/plain)
Delete this message
Reply to this message
Author: John M Cavallo
Date:  
To: Yaffs
Subject: [Yaffs] Lost blocks
I am having problems with portions of a flash becoming increasingly
unavailable during usage. After a few cycles of filling and emptying the
flash, with occasional reboots/remounts thrown in (see the script at the
end of this message), the majority of the flash will become lost. For
example, I had two partitions on one device, with 160 Mbytes on each,
one was stressed, otherwise they are identical. When I get the disk
usage of the two, there is approximately the same amount of information
on both:

# du -sk /nand /nand1
22544 /nand
22560 /nand1

However, when I look at the available space on the two, none is left on
the one that had been stressed.

# df /nand /nand1
Filesystem           1k-blocks      Used Available Use% Mounted on
/dev/mtdblock/5         163840     24856    138984  16% /nand
/dev/mtdblock/4         163840    163840         0 100% /nand1


This effect is repeatable. We haven't been able to track down the exact
cause of the problem, but it looks like it is connected to the initial
scan not properly disposing of remnants of files that have been deleted.

-------------------------

This is a stripped down version of the script that we used to stress
tests a partition mounted at /nand1.

tar -cf /nand1/usr.tar /usr

dirnum=1

while true
  # fill and empty 5 times
  for i in 1 2 3 4 5
  do
    # create a sub directory and try filling it
    mkdir -p /nand1/test/$dirnum
    while cp -a /nand1/usr.tar /usr /nand1/test/$dirnum
    do
      dirnum=`expr $dirnum + 1`
      mkdir -p /nand1/test/$dirnum
    done
    # the copy failed, so remove the test directory
    rm -r /nand1/test
  done
  # unmount and remount. Could also reboot at this point,
  # but that requires more than a single script.
  umount /nand1
  mount /nand1
done