Let’s say that I have a 4KB buffer that contains at the start a checksum. This buffer contains metadata that references other buffers. N.B. the buffers are memory mapped files. I notice that one of the buffers are invalid (either the checksum does not work out OR the checksum is fine but it references an invalid buffer). Now, I want to invalidate the buffer that references another invalid buffer. What is the mathematically best way to invalidate the buffer? Corrupt the checksum itself so that it might not match up with the data OR corrupt the data so it might not match up with the checksum. Or is there a third, better way altogether? For what its worth, this is a CRC32 variant.
Share
Assuming the CRC of your referencing buffer is valid, you can just simply subtract one from it, that will invalidate your buffer just fine. From your description, this seems to be the case since you state that the referenced buffer is the one whose CRC may be wrong.
If the referencing CRC32 wasn’t valid, the safest bet would be to calculate the CRC then subtract one. You need to know the correct CRC first because a random choice of CRC, whilst unlikely to be the correct one, still has a roughly one-in-four-billion chance of being right.
Changing the CRC seems like the safest option here. Short of doing a mathematical analysis of the CRC32 algorithm, there’s no way to guarantee that a specific change to the data won’t generate the same CRC value (since a large number of data sets may map to the same CRC).
However, it’s a certainty that a specific data set will always generate the same CRC so, by changing the CRC to something else, you can be sure that the block will now be invalid.