a comment why it is not necessary.

Only call vcl_hash{} if the initial hash is still intact when we get there.

This lifts internal restrictions on the hashkey production,
but does not change how things work from a VCL perspective.

	expect	resp.http.content-length -le 1024

And use that for logging purposes when a successfully opened h2 session
ends. RX_JUNK is still the default session close reason when existing
reasons aren't accurate enough.

Fixes #3393

In the same fashion as include/tbl/params.h for legibility.

with manual polish for cstyle and un-doing some return ((x)).

Either flexelint or me, probabbly the latter, do not understand the
code.

As far as I am concerned, I would have thought that flexelint could
follow for

	AN(y);
	REPLACE(x, y);

x being guranteed to be non-NULL

This reverts commit 897e5313.
This reverts commit 6f4d0ec1.

to slim down one of my PRs with a trivial addition which, I hope, will
not be controversial.

Missing include for REPLACE() / memset() added manually.

After edb0ca98, the ladybug understands
our code better.

Move patches there which are not intended to be applied again, but which
we used once and want to keep for reference.

Also another coccinelle script which I found useful at one point.

We cannot :ref: the rest of the online documentation from here

... regarding multiple VRT_call()s from vmods.

Dridi noticed that we need to consider the handling for VRT_call(),
thank you!

Based on feedback from Dridi

VMODs must not use SUB references across VCLs. This test checks for a
panic if they do.

This test case checks dynamic calls, and, in particular, the runtime
context checks and recursion detection. Recursions have to be caught at
the first instance.

Also test interop with the PRIV_TOP compile time restriction

Ref #3498
Ref 6d49b18f

The test has been manually annotated with information regarding the
results expted from VCCs call tree walk.

Explanation:

.called == call, but every walk counts (for instance, barbarbar can be
reached twice, via bar and barbar)

.nref == sum(refs): Each VCL ref gains a ref, and each call from a sub
which is itself dynamic (has nref > called).

This allows for VCL subs to be called dynamically at runtime: vmods
may store references to VCL_SUBs and call them some time later.

Calling a VCL_SUB pointer requires to conduct two fundamental checks
also at runtime, which previously were implemented by VCC at compile
time only:

* checking the allowed context
* checking for recursive calls

The foundation for both has been laid in previous commits and has
been made available through VPI_Call_*().

Note that we do not change the VCC checks in any way for static calls,
we only add runtime checks for dynamic calls.

This commit adds a VRT interface for dynamic calls and changes VGC to
ensure proper runtime checking (as suggested by phk).

We add to the vcl_func_f signature
- a vcl_func_call_e argument denoting the type of call and
- a vcl_func_fail_e value argument to optionally return a failure code

On vcl_func_call_e:

VSUB_CHECK allows runtime callers to preflight a "can this sub be
called" check, which is the basis for VRT_check_call().

VSUB_DYNAMIC must be used by any calls outside VGC.

VSUB_STATIC is for VGC only and, strictly speaking, would not be
required. It is added for clarity and safety and used for "call"
actions from VCL.

On vcl_func_fail_e:

If the argument is present, any error will be returned in
it. Otherwise, the generated code fails the VCL.

On the implementation:

To minimize the overhead for runtime context and recursion checks, we
only add them where needed.

In previous commits, we changed the VCC walks over SUBs to ensure that
any dynamically referenced SUBs and all SUBs reachable from these via
"call" can be identified by having more references than calls
(p->sym->nref > p->called).

So we now know, at compile time, for which SUBs we need to add runtime
checks. Read
https://github.com/varnishcache/varnish-cache/pull/3163#issuecomment-770266098
for more details.

Depending on whether the SUB is dynamically callable or not, we either
add runtime checks or keep the code identical to before (except for a
changed signature and some assertions).

For the dynamic case, we need to wrap the actual SUB in order to clear
the recursion check bit when it returns. Note that this happens
independend of whether the actuall call is static or dynamic - because
it has to. The recursion check for dynamic calls requires any previous
static calls to be registered.

The VGC for the dynamic case looks like this (slightly edited for
clarity):

static void
VGC_function_foo_checked(VRT_CTX)
{
  assert(ctx->method & (VCL_MET_DELIVER|VCL_MET_SYNTH));
  { { VPI_count(ctx, 1);
      // ... as before

void v_matchproto_(vcl_func_f)
VGC_function_foo(VRT_CTX, enum vcl_func_call_e call,
    enum vcl_func_fail_e *failp)
{
  enum vcl_func_fail_e fail;

  fail = VPI_Call_Check(ctx, &VCL_conf, 0x300, 0);
  if (failp)
    *failp = fail;
  else if (fail == VSUB_E_METHOD)
    VRT_fail(ctx, "call to \"sub foo{}\" not allowed from here");
  else if (fail == VSUB_E_RECURSE)
    VRT_fail(ctx, "Recursive call to \"sub foo{}\" not allowed from here");
  else
    assert(fail == VSUB_E_OK);

  if (fail != VSUB_E_OK || call == VSUB_CHECK)
    return;
  VPI_Call_Begin(ctx, 0);
  VGC_function_foo_checked(ctx);
  VPI_Call_End(ctx, 0);
}

The actual function body remains unchanged, but is contained in the
static function named "*_checked". The vcl_func_f is now a wrapper.

The wrapper

- checks for context errors using VPI_Call_Check()
- either returns any error in the failp argument or fails the VCL
- encoses the call to the "_checked" function with VPI_Call_Begin /
  VPI_Call_End which set/clear the recursion marker bit.

This implements the first bit of phks suggestion in
https://github.com/varnishcache/varnish-cache/pull/3163#issuecomment-769718143

These VPI functions implement the operations on the recursion check
bitmask for dynamic SUB calls. Strictly speaking, the VPI subs would
not need to be concerned with the methods bitmask check and the
assertion that calls originate from the right VCL, but we centralise
these aspects for simplicity.

We also add an enum to denote check failures.

To allow dynamic SUB calls, we need to check for recursions at
runtime, reflecting the compile time recursion check.

We create a bitmask to note each dynamically called SUB by its unique
number. To check for recursions, we only need to test if the
respective bit for the sub number is already set.

We only conduct this check if the track_call argument of
vcl_call_method() is true. By setting this argument to the number of
SUB references, we enable call tracking only for VCLs with dynamic SUB
references (when (struct VCL_conf).nsub is non-zero).

VCC already guarantees that static calls are loop free - so loops can
only be introduced through dynamic SUB references.

Call tracking with a bitmask was suggested by phk

Before the addition of the okmask, we could only selectively turn off
optimizations for the built-in housekeeping subs (vcl_init {} /
vcl_fini {}), not for any custom subs called from there.

The previous commit changed this and turned off optimizations for any
custom subs called from there.

We pull this tighter again by only turning off optimizations for subs
which can _only_ be called from housekeeping.

Ref f1c47e48
Ref #3230

Improve safety by asserting that a sub is never called from a context
which is not allowed as per its okmask.

This is to ensure correctness of our compile- and runtime-checks.

We previously missed to run the static recursion check on SUBs which
are only referenced symbolically (and thus might be called
dynamically).

This changes the error reporting tested in v00021.vtc because the
recursion is detected on the recursing subs alone before the call tree
from vcl_recv {} is walked.

In follow up commits, we will need a marker for subs which can be
reachable via dynamically referenced SUBs.

We use "more references than calls" is this marker.  When walking
the call trees, we add one reference for all subs which are called from
a SUB having this marker, thus marking those, too.

The .called attribute was not used.

We now increment it only for custom subs which are actually called
statically (with "call"), that is, not for the builtin subs. This is
to simplify and clarify the use case in the next commits.

The increment happens for each VCC walk of the SUBs.

We also add the number of symbolic references and calls to (struct
vcl_sub) of the VGC to facilitate reviews and debugging.

For follow up commits, we will need to know if a VCL contains any
references to the SUB type, that is, if the SUB type was used as a vmod
function or method argument.

We change the semantics of the rname and lname attributes of SUB
symbols: rname now is the name of the (struct vcl_sub), lname is the
cname, the name of the function to be called directly from VGC.

We call those subs impossible which use a combination of variables not
being available together from any of the built-in subroutines (aka
methods), for example::

        sub foo {
                set req.http.impossible = beresp.reason;
        }

For this example, there is no built-in subroutine for which both req
and beresp were available, so it can not possibly be valid in any
context.

As long as subs were actually called directly or indirectly from a
built-in sub, VCC did already detect the problem. For the above
example, if `sub vcl_recv { call foo; }` existed, VCC would complain
about beresp.reason not being accessible from vcl_recv.

Except, when vcc_err_unref was disabled, the example would pass
without error, and likewise it will with calls which we are about to
add.

We now add a check to detect impossible subroutines. It happens in a
second walk over the VCL's SUB symbols in order to keep the more
precise error messages for calls rooting in one of the built-in subs.

This change also changes the order of error reporting slightly and
incorporates, indirectly, the vcc_CheckUses() check moved in
ed36b638.

v00020.vtc now has the "Impossible Subroutine with vcc_err_unref off"
check, while the newly added m00053.vtc has been changed to test
failure of the impossible sub at compile time rather than at runtime.

This change was triggered by feedback from Dridi

Until now, the VCC SUB type represented the literal name of a
vcc-generated C function, which was used to expand the VCL "call"
action.

We now add a struct to describe VCL subs and calculate a bitmask which
represents the bilt-in subs which a SUB may be called from (the
"okmask"),

This is in preparation of the goal to add a VCL_SUB vmod type which
will allow VMODs to call into vcl subs.

We add to the vcl shared object a struct vcl_sub for each sub, which
contains:

	- a methods bitmask defining which built-in subs this sub may
	  be called from (directly or indirectly)
	- the name as seen from vcl
	- a pointer back to the VCL_conf
	- the function pointer to the vcl_func_f
	- a unique number

About the methods bitmask:

It contains the VCL_MET_* bits of of built-in subs (traditionally
called methods) which are allowed to call this sub.

It is intended for checks like

	if (sub->methods & ctx->method) {
		sub->func(ctx);
		return;
	} else {
		VRT_fail(ctx, "not allowed here");
		return;
	}

In existing compile-time calls, we already check if used objects and
returned actions are allowed within the respective calling context.

To build the methods bitmask, we begin with a VCL_MET_TASK_ALL
bitfield and clear the bits of methods which would not be allowed for
any used object or returned action.

About the VCL_conf pointer:

Each VCL SUB belongs to a VCL, this pointer will allow to check that
it is only ever used from the right VCL.

About the unique number:

By numbering vcl subs (per VCL), we can later implement a recursion
check using a bitmask.

In struct VCL_conf, we record the total number of subs.