NOTES

OBSERVATIONS about 'bmap' in action.

(This document has been largely superceded by application notes contained in
 the bmap documentation)

* ext2 reports file blockcounts overzealously. stat() returns a blockcount
  measured in sectors on the author's system. However, stat claims that the
  the stat(2) manpage suggests that this blockcount may not be in the same
  units reported by bmap system call, but instead may be expressed in 
  'optimal' transfer units. The author has trouble buying an optimal transfer
  blocksize of 512 bytes.

  The kernel correctly reports the bmap blocksize with ioctl(FIGETBSZ,...). 

* stat _seems_ to report extra blocks for ext2 files over 12k. Does stat
  get caught up in the size of metadata affiliated with a file? 
  UPDATE: Walter Hofmann reported that indirect blocks count against the
  stat-reported size in blocks.

* holes can be tricky to deal with