shred: overwrite inode storage used by some file systems

* doc/coreutils.texi (shred invocation): Mention some reasons
why clearing slack space might be useful.
* src/shred.c (do_wipefd): Add initial writes for each pass
for small regular files in case the storage for those is
in the inode, and thus a larger write up to a block size would
bypass that.  Move the direct I/O control to...
(dopass): ... here so we can avoid enabling it for these small
initial writes.  It's better to retry direct I/O for each pass
anyway to handle the case where direct I/O is disabled for only
the last portion of a file when the size is not a multiple of
the block size.  Note we don't avoid the sync for the initial
write as it will be small but more importantly could be on a
different part of the disk and so worth doing independently
to ensure the write is not discarded.
* tests/misc/shred-exact.sh: Check some more direct I/O cases.
* NEWS: Mention the improvements.
The inode storage issue was mentioned by Paul Eggert.

1 files changed, 3 insertions, 2 deletions

diff --git a/doc/coreutils.texi b/doc/coreutils.texi
index 75875d88c..6c49385e1 100644
--- a/doc/coreutils.texi
+++ b/doc/coreutils.texi
@@ -9634,8 +9634,9 @@ Display to standard error all status updates as sterilization proceeds.
 @opindex -x
 @opindex --exact
 By default, @command{shred} rounds the size of a regular file up to the next
-multiple of the file system block size to fully erase the last block
-of the file.
+multiple of the file system block size to fully erase the slack space in
+the last block of the file.  This space may contain portions of the current
+system memory on some systems for example.
 Use @option{--exact} to suppress that behavior.
 Thus, by default if you shred a 10-byte regular file on a system with 512-byte
 blocks, the resulting file will be 512 bytes long.  With this option,