Before, the origin time for the objects was defined as the time we
received the full request from the client that initiated the
fetch. Now we use the time we received the response headers from the
backend.

The EXP_Touch and the Age header will be based off the last timestamp
taken. This can be slightly inaccurate, but saves us from having to do
another timestamp for them.

Fixes: #1475

backend. Set it on the object if missing.

Previously we would add an explicit Date header on each response with
the time and date of the response, overwriting any Date header on the
object that came from the backend.

Now we will add a Date header if the backend didn't supply one just
before vcl_backend_response() is run, and not add any Date headers for
the replies.

This change means Date headers sent by Varnish is approximately the
time and date this object was generated by the backend, and not the
current date and time.

While here fmt(1) the text.

Introduce new section for new parameters

Spotted by:	Ruben

We need req.backend_hint in vcl_hit to enable different
grace times depending on backend health.

size.

Fixes: #1469

Fixes: #1468

Fixes: #1462

Spotted by: Martin Boer

imposed by the kernel.

Spotted by:	ingvar (on PPC64)

From Martin Boer on #varnish

Spotted by Martin Boer on #varnish

The oc's busyobj isn't cleared in HSH_Fail (like it is in HSH_Complete).
This allows a thread to get a pointer to an already free'd busyobj.

Fixes: #1467

This file was named wrong, and the test does not make sense anymore, since the backend health information is in the fetch thread

This triggers 'shadows global' build errors.

This makes sure no extra 'transfer-encoding' headers should pop up
changing the byte counts.

This triggers 'shadows global' build errors.

This triggers 'shadows global' build errors.