Deliver nonpadding events instead of calling token removal functions.

This reverts commit ab8b80944967ee5a6a0c45dbf61839cf257bfe44.

Some of these functions are now public and cpath-specific so their name should
signify the fact they are part of the cpath module:

assert_cpath_layer_ok -> cpath_assert_layer_ok
assert_cpath_ok -> cpath_assert_ok
onion_append_hop -> cpath_append_hop
circuit_init_cpath_crypto -> cpath_init_circuit_crypto
circuit_free_cpath_node -> cpath_free
onion_append_to_cpath -> cpath_extend_linked_list

We are using an opaque pointer so the structure needs to be allocated on the
heap. This means we now need a constructor for crypt_path_t.

Also modify all places initializing a crypt_path_t to use the constructor.

This commit only moves code.

This reverts commit 387d9448.

This test was disabled in 0.4.0 and later, but the fix in #29298 was only
merged to 0.4.1. So this test will never be re-enabled in 0.4.0.

Part of 29500.

Mike Perry authored Apr 05, 2019 and teor committed Apr 05, 2019

Our monotime mocking forces us to call monotime_init() *before* we set the
mocked time value. monotime_init() thus stores the first ratchet value at
whatever the platform is at, and then we set fake mocked time to some later
value.

If monotime_init() gets a value from the host that is greater than what we
choose to mock time at for our unittests, all subsequent monotime_abosolute()
calls return zero, which breaks all unittests that depend on time moving
forward by updating mocked monotime values.

So, we need to adjust our mocked time to take the weird monotime_init() time
into account, when we set fake time.

Mike Perry authored Mar 04, 2019 and teor committed Apr 05, 2019

Hopefully this will stop monotime_absolute_usec() from returning 0 on some
platforms in the tests.

Mike Perry authored Feb 27, 2019 and teor committed Apr 05, 2019

Cancel the padding timer by changing order of sent vs recv (sent cancels).

George Kadianakis authored Mar 13, 2019 and teor committed Mar 22, 2019

The name of circpad_machine_state_t was very confusing since it was conflicting
with circpad_state_t and circpad_circuit_state_t.

Right now here is the current meaning of these structs:

      circpad_state_t -> A state of the state machine.
      circpad_machine_runtime_t -> The current mutable runtime info of the state machine.
      circpad_circuit_state_t -> Circuit conditions based on which we should apply a machine to the circuit

George Kadianakis authored Mar 13, 2019 and teor committed Mar 21, 2019

This is something we should think about harder, but we probably want dormant
mode to be more powerful than padding in case a client has been inactive for a
day or so. After all, there are probably no circuits open at this point and
dormant mode will not allow the client to open more circuits.

Furthermore, padding should not block dormant mode from being activated, since
dormant mode relies on SocksPort activity, and circuit padding does not mess
with that.

David Goulet authored Jan 29, 2019 and Roger Dingledine committed Mar 14, 2019

They are simply not used apart from assigning a pointer and asserting on the
pointer depending on the cell direction.

Closes #29196.

Signed-off-by: David Goulet <dgoulet@torproject.org>

George Kadianakis authored Feb 25, 2019 and Nick Mathewson committed Mar 13, 2019

They were causing the following warnings in circuitpadding/circuitpadding_sample_distribution:

     src/lib/math/prob_distr.c:1311:17: runtime error: division by zero
     SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior src/lib/math/prob_distr.c:1311:17 in
     src/lib/math/prob_distr.c:1219:49: runtime error: division by zero
     SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior src/lib/math/prob_distr.c:1219:49 in

because the distributions were called with erroneous parameters (e.g. geometric
distribution with p=0).

We now defined these test probability distributions with more realistic
parameters.

As far as the circuitpadding_sample_distribution() test is concerned, it
doesn't matter if the distributions return values outside of [0,10] since we
already restrict the values into that interval using min=0 and max=10 (and RTT
estimate is disabled).

Add a basic validation function for the histograms. It can be a building block
for the future

- All histogram tests were using start_usec/range_usec. We now manually specify
  the edges.
- Also add a test for histogram_get_bin_upper_bound().

until #29298 is implemented.

Fixes bug 29169.

George Kadianakis authored Jan 21, 2019 and Nick Mathewson committed Jan 22, 2019

We fix it by disabling the scheduling of actual padding.
Fixes #29122.

Co-authored-by: Mike Perry <mikeperry-git@torproject.org>

Co-authored-by: George Kadianakis <desnacked@riseup.net>