quick is meant to be the fast option, but it only changes how repeated
attempts are handled - it still works through every variation of a
strategy regardless. For strategies built around a fake TLS handshake,
that means testing dozens of small tweaks (different TTLs, different
TCP tricks, extra padding, split positions) well past the point where
an early exit would have been possible.
The new veryquick level stops as soon as a strategy succeeds, instead
of continuing through the rest just to be thorough. It's a separate
option rather than a change to quick, so quick, standard, and force
behave exactly as they did.
The pad file, PARALLEL_OUT, and HDRTEMP are all already removed on
a normal exit - Ctrl+C, closing the terminal, a plain kill all clean
up after themselves. kill -9 and crashes skip that entirely, same as
for any script, and a later run has no way to know about a different
PID's leftovers - so all three can pile up in /tmp indefinitely.
Before each is used, the script now checks for existing files matching
its pattern and removes any whose PID is confirmed no longer running.
A file still owned by a live process - a second instance running
at the same time - is left alone.
The "16 KB" block triggers based on how much data passes through the
connection, not on what that data actually is. Testing for it usually
means pointing at a page you know is big enough to download - but that
only works if your target actually has one.
Adds CURL_PAD_KB: set it to a number and every request gets padded
with that many KB of junk data, so the block can be triggered against
any target, even a bare server with nothing to download. Off by
default, and doesn't change anything unless you turn it on.
The padding is written to a temp file and attached from there, not
built inline: a large inline value can hit low, easy-to-reach limits
on some platforms (Cygwin among them).
When testing several targets at once, the results only ever show
strategies that worked for every single one. If your targets are
a bunch of unrelated servers, that list can easily come up empty
even when several strategies each work for most of them.
Adds a second summary that lists every strategy that worked at all,
with how many targets it worked for, best first - so you can see
which one comes closest instead of getting an all-or-nothing answer.
bol-van