e980214bc4 serialization: prevent int overflow for big Coin::nHeight (pierrenn)
Pull request description:
This is an attempt to fix fuzzer issues 1,2,8 reported by practicalswift here : https://github.com/bitcoin/bitcoin/issues/18046
The fuzzer harness doesn't prevent deserialization of unrealistic high values for `Coin::nHeight`. In the [provided examples](https://github.com/bitcoin/bitcoin/issues/18046), we have :
- `blockundo_deserialize` : the varint `0x8DD88DD700` is deserialized as `3944983552` in `Coin::nHeight` (`TxInOutFormatter::Unser`)
- `coins_deserialize` : the varint `0x8DD5D5EC40` is deserialized as `3939874496` similarly
- `txundo_deserialize`: the varint `0x8DCD828F01` is deserialized as `3921725441` in `Coin::nHeight` (`Coin::Unserialize`)
Since `Coin::nHeight` is 31 bit long, multiplying a large value by 2 triggers the fuzzer.
AFAIK those values are unrealistic (~70k years for the smallest..). I've looked a bit a reducing the range of values the fuzzer can deserialize, but this seems to be too much code change for not much.
Hence this PR chooses to static cast `nHeight` when re-serializing; it seems to be the less intrusive/safest way to prevent the fuzzer output.
Another more "upstream" approach would be to limit `Coin::nHeight` values to something more realistic, e.g. `0xFFFFFFF` (~5k years) :
de3a30bab2/src/undo.h (L39) and de3a30bab2/src/coins.h (L71)
Thanks !
NB: i was also not sure about the component/area to prefix the PR/commit with.. ?
ACKs for top commit:
practicalswift:
ACK e980214bc4 -- patch looks correct
promag:
ACK e980214bc4.
sipa:
utACK e980214bc4
MarcoFalke:
re-ACK e980214bc4🎑
ryanofsky:
Code review ACK e980214bc4. Just removed ternary ? 1 : 0 and replaced / 2 with >> 1 since last review
Tree-SHA512: 905fc9e5e52a6857abee4a1c863751767835965804bb8c39474f27a120f65399ff4ba7a49ef1da0ba565379f8c12095bd384b6c492cf06776f01b2db68d522b8
6a11d9e330