Re: [Bug-ddrescue] Fill mode does not work if input_position is greater than file size

[Christian Franke]

Hello Antonio,

> Of course ddrescue cannot work if input_position is greater than input 
> file size. Just remember that in fill mode the input file is the file 
> from which the fill data is read.

In fill mode, the input file reading already wraps around at EOF. 
Starting at (input_position % size) would be IMO consistent with this 
behaviour.


> Try this:
> # ddrescue -i 10000000b -s 2000000b -o 0 bad_disk file.img file.log
> # ddrescue -F - fill_file file.img file.log

This does not work because file.log records the original offset from 
bad_disk, not the resulting offset in file.img.

The fill command has no information about the former difference between 
-i and -o. Therefore it will write the output file starting at offset 
10000000b, not at 0.


Testcase: Extract 3 blocks from a file, fill second block.

# ddrescue -i 16b -s 3b -o 0 COPYING outfile logfile

logfile:
# Rescue Logfile. Created by GNU ddrescue version 1.7
# current_pos  current_status
0x00002200     +
#      pos        size  status
0x00000000  0x00002000  ?
0x00002000  0x00000600  +
0x00002600  0x000062FC  ?


Now edit logfile to simulate a bad block:

# Rescue Logfile. Created by GNU ddrescue version 1.7
# current_pos  current_status
0x00002200     -
#      pos        size  status
0x00000000  0x00002000  ?
0x00002000  0x00000200  +
0x00002200  0x00000200  -
0x00002400  0x00000200  +
0x00002600  0x000062FC  ?

(Hmm...a --simulate-error option would be useful for such testing :-)


# cp outfile outfile.orig

# echo "BAD BLOCK" >fillfile


Now fill bad block as suggested above:

# ddrescue -F - fillfile outfile logfile


Observed result: "BAD BLOCK " appended at the original (infile) position:

# ls -l outfile.orig outfile
-rw-r--r-- 1 ... 9216 Jan  5 14:42 outfile
-rw-r--r-- 1 ... 1536 Jan  5 14:41 outfile.orig

# cmp outfile.orig outfile
cmp: EOF on outfile.orig

# diff -u <(xxd outfile.orig) <(xxd outfile)
--- /dev/fd/63    2006-12-01 01:00:00.000000000 +0100
+++ /dev/fd/62    2006-12-01 01:00:00.000000000 +0100
@@ -94,3 +94,483 @@
00005d0: 7572 206f 7220 7468 6972 6420 7061 7274  ur or third part
00005e0: 6965 7327 206c 6567 616c 2072 6967 6874  ies' legal right
00005f0: 7320 746f 2066 6f72 6269 6420 6369 7263  s to forbid circ
+0000600: 0000 0000 0000 0000 0000 0000 0000 0000  ................
+...
+00021f0: 0000 0000 0000 0000 0000 0000 0000 0000  ................
+0002200: 4241 4420 424c 4f43 4b0a 4241 4420 424c  BAD BLOCK.BAD BL
+0002210: 4f43 4b0a 4241 4420 424c 4f43 4b0a 4241  OCK.BAD BLOCK.BA
+...
+00023e0: 4241 4420 424c 4f43 4b0a 4241 4420 424c  BAD BLOCK.BAD BL
+00023f0: 4f43 4b0a 4241 4420 424c 4f43 4b0a 4241  OCK.BAD BLOCK.BA


Expected result: "BAD BLOCK..." should be written to area 0x200-0x3ff of 
outfile. Outfile size should not change.


> > The -i is necessary, because ddrescue records the original physical 
> > offsets in the logfile.
>
> The -i is needed only if you want to read the fill file from a given 
> position.

It also needed to specify the difference between the (infile) positions 
recorded in the logfile and actual positions in outfile.


> > Hmm...would it make sense the store the input/output_position offset 
> > in the logfile also?
>
> And what is ddrescue supposed to do with them?

E.g. fail if difference between -i and -o is not identical to the first run.


Regards,
   Christian