Conflicts:
	bin/varnishd/cache/cache_http.c
	bin/varnishd/http1/cache_http1_fsm.c

Conflicts:
	bin/varnishd/cache/cache.h
	bin/varnishd/cache/cache_http.c
	bin/varnishd/cache/cache_http1_fsm.c
	include/tbl/sess_close.h

This is a backport of a88dbf91.

Original patch and test case by: tnt

Fixes: #1627

If a duplicate CL header is in the request, we fail the request with a
400 (Bad Request)

Fix a test case that was sending duplicate CL by misstake and would
not fail because of that.

Varnish (prior to version 4.0) was not following the standard with
regard to line separator.

Spotted and analyzed by: Régis Leroy [regilero] regis.leroy@makina-corpus.com

Fixes #1690

Backports commit 2144dc78 for Varnish 3.0.6 (by Geoff)

During the last VDD I (hastily) added a note that the empty VCL truth
value changed between 3.0.2 and 3.0.3.

As Geoff found after digging into this (see varnish-misc@ yesterday),
this happened between 3.0.3 and 3.0.4.

Found by:	Geoff

Spotted by:	scoof

Solves:		#1406

Extensive test cases often lead to a lot of log output,
which at times tripped the following assert:

Assert error in vtc_log_emit(), vtc_log.c line 122:
  Condition(vtclog_left > l) not true.

Double the amount of buffer space allowed to avoid this. Another 256KB
memory use per concurrent running test should be fine.

Solaris (and others?) need defines from it (GNU_SOURCE in particular)

Submitted by:       Clement Gautier via github

Reinstall the readline handler when we are called in order to clear the
input line. Tested in a variety of versions by scn and myself.

Fixes #1531

Reviewed by:	scoof

This is why it sucks to have the redhat/ directory inside
the source distribution.

Fixes: #1514

Round up the header space with PRNDUP in session space allocation to
fix alignment issue.

Fixes: #1327

ESI subrequests would mangle the t_open and t_resp timestamps of the
original requests, causing negative time and NAN being logged in
ReqEnd.

Fixes: #1297

Do as we have documented we do (and Apache does), printing the %D as
request time in microseconds, truncated to an integer value.

The pass bits of the oc's flags wasn't cleared before or'ing the new
pass on failing the busy object test in the ban lurker. This would
cause the pass not to be matched later, making the lurker spin on the
busy object.

Fixes: #1470

Document that our use of 0s as a magic constant had the side
effect of enabling grace if you tried to disable it in VCL, and
how to get it to work.

Merging tiny patch from:

    http://lists.freebsd.org/pipermail/freebsd-current/2010-January/015138.html

Fixes: #1448

Restart from vcl_pass and vcl_miss does not work, but was
documented that it did.

Mentioned in: #1443

Cherry pick changes from master to allow it.

Fixes: #1439

Junk after gzip body end would lead to a spinning
thread.

Fixes: #1086