bup-damage(1) Bup 0.33.3

Avery Pennarun



bup-damage - randomly destroy blocks of a file


bup damage [-n count] [-s maxsize] [--percent pct] [-S seed] [--equal] <filenames…>


Use bup damage to deliberately destroy blocks in a .pack or .idx file (from .bup/objects/pack) to test the recovery features of bup-fsck(1) or other programs.


bup damage is primarily useful for automated or manual tests of data recovery tools, to reassure yourself that the tools actually work.

Note that the details of the current behavior may change (particularly the details not documented here). For example the moment, the damage is strictly probabilistic, and so may or may not actually alter any given block. With a block size of 1, there should be a 1/256 chance that the block won’t actually change. This behavior may change.


-n, --num=numblocks
the number of separate blocks to damage in each file (default 10). Note that it’s possible for more than one damaged segment to fall in the same bup-fsck(1) recovery block, so you might not damage as many recovery blocks as you expect. If this is a problem, use --equal.
-s, --size=maxblocksize
the maximum size, in bytes, of each damaged block (default 1 unless --percent is specified). Note that because of the way bup-fsck(1) works, a multi-byte block could fall on the boundary between two recovery blocks, and thus damaging two separate recovery blocks. In small files, it’s also possible for a damaged block to be larger than a recovery block. If these issues might be a problem, you should use the default damage size of one byte.
the maximum size, in percent of the original file, of each damaged block. If both --size and --percent are given, the maximum block size is the minimum of the two restrictions. You can use this to ensure that a given block will never damage more than one or two git-fsck(1) recovery blocks.
-S, --seed=randomseed
seed the random number generator with the given value. If you use this option, your tests will be repeatable, since the damaged block offsets, sizes, and contents will be the same every time. By default, the random numbers are different every time (so you can run tests in a loop and repeatedly test with different damage each time).
instead of choosing random offsets for each damaged block, space the blocks equally throughout the file, starting at offset 0. If you also choose a correct maximum block size, this can guarantee that any given damage block never damages more than one git-fsck(1) recovery block. (This is also guaranteed if you use -s 1.)


# make a backup in case things go horribly wrong
cp -pPR ~/.bup/objects/pack ~/bup-packs.bak

# generate recovery blocks for all packs
bup fsck -g

# deliberately damage the packs
bup damage -n 10 -s 1 -S 0 ~/.bup/objects/pack/*.{pack,idx}

# recover from the damage
bup fsck -r


bup-fsck(1), par2(1)


Part of the bup(1) suite.