Ref: #2693

This is the case for the ctx created for director operations as first
introduced with 5536f102

Alternatively, we could always set ctx->handling to some location
with the same scope as the ctx, but as we are not inside VCL (and
just happen to use the same context for simplicity and reusability
of VRT/vmod functions), adding specific handling appears to be the
cleaner solution.

Fixes #2692

Properly shut down streams where we could not process a req
body (e.g. via a failing cache_req_body). The rxthread may be waiting on
this stream to consume a DATA frame - let it know before we go away to
avoid it being stuck forever.

If we have one or more streams set up because of a PRIORITY frame _and_
for some reason a tcp FIN is lost (or a client merely keeps a connection
open), this transaction would be stalled indefinitely.

Avoid failing sessions with pipelined data.

match syslog-output.

Submitted by: Frédéric Lécaille <flecaille@haproxy.com>

Hacked to pieces by:	phk

We already emitted Debug log records for accept failures, but for
all practical purposes this is useless for forensics.

These counters will point directly to the root cause for the most
common issues with sess_fail (accept failures). sess_fail is preserved
as an overall counter as it will most likely be already monitored for
many installations.

Ref	#2622

Conflicts:
	doc/changes.rst

With that both -d and -1 options report the same output, with the former
limited to the terminal height. While this increases the risk of "never
ending" dumps of log records showing only once, the assumption is that
varnishtop is useful when someone is looking for something in particular.
Restricting the output to the records someone is interested in mitigates
the risk.

Fixes #2686

Following the documentation, the -d option implies that once processing
is done the process should exit. We give it the time to do one last
refresh of the screen in curses mode.

Refs #2686

For the case that http_max_hdr is not a multiple of the pointer
size.

Ref #2690

Use pthread_cleanup_pop(0) and an explict call to the cleanup
function to make things visible to FlexeLint.

of the source tree.

We do not want to cause a panic from the panic handler.

It could be argued that bo->vfc cannot possibly be NULL, except
when it is - and that's what we got the panic handler for.

vfc->failed is output by pan_vfp now so I should have removed
that duplication when introducing pan_vfp

See comment on how we could auto-generate the documentation from the
include - any less hackish ideas welcome

Fixes #2685

Fixes #2647

The h/2 request body VFP would drop data when the input buffer was too
small to fit the data in the received frame.

With this fix we have the VFP code call us again with a fresh buffer
when we run out.

Fixes: #2679

Ref #2683