[Bug-ddrescue] Preallocating space, if possible

[Matthew Ruffalo]

Hello Antonio-

First, thank you for the wonderful tool. I've used ddrescue many times 
to recover data from failing hard drives, as well as for manually 
imaging working drives (the progress/speed display is very nice).

I'm curious whether it's possible for ddrescue to preallocate space for 
the file that it writes using posix_fallocate. I don't envision this as 
standard behavior; I remember reading a message that you posted to the 
mailing list saying that on some platforms, the size information is not 
available at all. Additionally, I've seen failing hard drives report 
erroneous sizes (such as an 80GB drive showing as 2.1TB -- it apparently 
reported 0xFFFFFFFF when queried).

If the size is available (which would be a compile-time option) and 
correct (which would be a runtime option that defaults to false), would 
you support using a preallocation system call in order to improve 
filesystem performance and reduce fragmentation? Modern filesystems such 
as ext4, xfs, ocfs2, etc. support very efficient preallocation of large 
files and it seems worth it to take advantage of this capability.

I did a quick grep through the source code and didn't find any instance 
of 'allocate'. My understanding (which is not all that likely to be 
correct) is that this behavior is distinct from the --sparse option. A 
sparse file's size is reported to be much larger than the allocated size 
on disk, whereas the new behavior would tell the filesystem in advance 
that e.g. the image file will eventually contain 250GB. The filesystem 
will then allocate this space as contiguously as possible.


Semi-related question: is there a publicly accessible source code 
repository for ddrescue? I created a repository on GitHub ( 
http://github.com/mruffalo/ddrescue ) that I will use to publish my 
changes in the event that I implement this feature myself. It contains 
all released versions of ddrescue back to version 0.6 and the authorship 
timestamps match the modification times of the archives on the GNU 
mirrors. For example:

Author: Antonio Diaz Diaz <address@hidden>  2005-01-04 05:39:00
Committer: Matthew Ruffalo <address@hidden>  2010-05-30 14:16:34
Tags: v0.9

Thank you,
--
Matt Ruffalo...
Department of Electrical Engineering and Computer Science
Case Western Reserve University