search for: checksum1

...tch mode means block collisions are reproducible and predictable. Thus, some files will be permanently "unlucky" in batch mode and will always need retransmission. Also, it means a malicious adversary can perform a performance degradation attack by injecting a pair of blocks with the same checksum1 and checksum2 (so that, for example, if you use rsync to backup your website database, someone could slow down your backups by posting carefully crafted comments). The latter issue also exists in non-batch mode, since time() is often predictable. The right thing to do is to always use a really rand...

...was a great deal of code duplication involved with reading and writing the sum_struct fields. I have eliminated the duplication by consolidating that transmission in the new (read|write)_sum_head functions. A added bonus is that write_sum_head replaces the send_null_sums function. The Adleresque checksum1 is in no way affected. The varsumlen patch builds on that groundwork by implementing a simple heuristic to generate the per-file sum2 lengths. It remains two bytes until the file is 8193 blocks at which it increments and continues incrementing each time the block count quadruples. This builds on...