Resets counters about nethogs HOT 5 CLOSED

Interesting, might be an overflow - we count bytes - let's see if we can widen that type :)

from nethogs.

I'm pretty sure this is an overflow since we're using uint32_t to count bytes, and maximum for that type (2^32 - 1) works out to be approximately 4.2GB (and regardless of what view you specify, the totals are stored as bytes and converted to the unit you asked for). That's stored per connection, but it's also summed per process... so I think it's quite likely that (especially on a busy server or a desktop with a process that's doing a lot of network IO like a game, torrent client, single-process browser, or streaming video) it can wrap around - that will happen with any process that sends or receives more than 4.2 GB.

If we bumped all of the total bytes counters from uint32_t to uint64_t, we'd end up consuming a bit more memory (an additional 64 bytes per connection, plus some program-wide overhead for summing) and change the public API of libnethogs, but we'd be able to count 2^64 - 1 bytes, or 18.4 exabytes (that's 18,446,744.1 terabytes) before wrapping around.

from nethogs.

Sounds like a good idea to me!

from nethogs.

Ok, I'll give a try at a PR for this one once I get some feedback on my other open PRs; this is going to change a lot of variable types, so it would be a rather painful rebase.

from nethogs.

Thanks for this fix @jantman

from nethogs.