Change the session handling logic, so that we hand any session back to
the herder, if the session_linger timer fires.

This means that a new session, will get a worker thread, but if nothing
happens within session_linger, it disembarks and lets the herder handle
the assembly of the request.

This actually simplifies the state engine, because the "again" state
(now renamed "wait) can also be used both for the first request.



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4079 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4078 d4fa192b-c00b-0410-8231-f00ffab90ce4

It turns out that you could turn off ESI processing for subrequest and compeltely hose the response. Ignore the flag if you are in a subrequest.

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4077 d4fa192b-c00b-0410-8231-f00ffab90ce4

workspace before we start using it in miss/pass/pipe.  



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4076 d4fa192b-c00b-0410-8231-f00ffab90ce4

we short-cut through STP_AGAIN.

Expected to fix #487



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4075 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4074 d4fa192b-c00b-0410-8231-f00ffab90ce4

doing an esi-include.

This should reduce the pressure on session workspace from nested
ESI includes considerably.



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4073 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4072 d4fa192b-c00b-0410-8231-f00ffab90ce4

Instead of the acceptor thread doing it when reusing the session, have the
worker threads clean it out before putting it on the free list.

It could be, and probably was, argued that this is a performance
pessimization, but having thought much about it, I can't spot the
argument any longer, and certainly moving load off the acceptor thread
to the massively parallel worker threads is a good idea.



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4071 d4fa192b-c00b-0410-8231-f00ffab90ce4

drop new connections summarily.

The parameter is session_max, default is 100k and the stats variable
client_drop counts how many sessions were dropped.

This is mostly an anti-DoS measure and your feedback and experience
with it is most welcome.




git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4070 d4fa192b-c00b-0410-8231-f00ffab90ce4

happening again:

We optimize session allocation, SES_New(), to minimize the amount
of locking the VCA_thread participates in with flip-flop lists of
free sessions: Allocate from one list, free to the other, flip lists
only when necessary.

This scheme only works if nobody but VCA_thread calls SES_New().

Background worker threads need dummy sessions and thus called into
SES_New() as well.

These calls all happen during startup, so they do in fact not mess
up the locking, but made mockery of a number of very stern comments.

Add SES_Alloc() for such "other uses" and enforce VCA_thread monopoly
with an assert.


git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4069 d4fa192b-c00b-0410-8231-f00ffab90ce4

There is probably a bug if an ESI subrequest turns ESI delivery back on -- add reminder so I can circle back and extend the test for it

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4068 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4067 d4fa192b-c00b-0410-8231-f00ffab90ce4

Allows you to turn off esi processing for a specific request using set req.esi = off even if the object has been esi processed -- this is useful if you have a cache heirarchy of varnish machines as well as debugging. Only change to processing is to store a copy of the attribute value instead of mangling to original. Committed from Oahu.

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4066 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4065 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4064 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4063 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4062 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4061 d4fa192b-c00b-0410-8231-f00ffab90ce4

Platform-breakage to be expected I pressume.

Apply gross hacks to POSIX-shortages, trying to get a result barely as
usable as what sensible operating systems have provided since at least
1985.



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4060 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4059 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4058 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4057 d4fa192b-c00b-0410-8231-f00ffab90ce4

Maxim Sobolev <sobomax@FreeBSD.org>



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4056 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4055 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4054 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4053 d4fa192b-c00b-0410-8231-f00ffab90ce4

	HTTP/1.1 1000\n\r\n\r
would panic us trying to find a suitable message for 1000.

Now we 503 the response instead.

Fixes #506



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4052 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4050 d4fa192b-c00b-0410-8231-f00ffab90ce4

This is in anticipation for set req.esi = off;


git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4049 d4fa192b-c00b-0410-8231-f00ffab90ce4

Fixes #502



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4048 d4fa192b-c00b-0410-8231-f00ffab90ce4

broke pipehandling.

Fixes #505



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4047 d4fa192b-c00b-0410-8231-f00ffab90ce4

Previously we always used SO_LINGER to make sure that all queued data
got transmitted, no matter under which circumstances we closed the
client connection.

Change this so that SO_LINGER is only activated for orderly connection
closure (ie: "Connection: close" from client or error handling), in
all other cases (usually the client connecting on us, abandon any data
queued for transmission.

This _may_ reduce the tendency of worker threads to get hung up on
network failures a little bit.



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4046 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4045 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4044 d4fa192b-c00b-0410-8231-f00ffab90ce4

Fixes #329



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4043 d4fa192b-c00b-0410-8231-f00ffab90ce4

and this prevented users from implementing "purge on POST" semantics
by looking up the POST requests and ditching the object found.

Now we only overwrite the request with "GET" if it is a HEAD, (the
internal "wantbody" flag tells hos how the response should be composed.)

The testcase shold probably be elevated to VCL example, with suitable
footnotes about URL sanitizing etc.

Fixes #444



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4042 d4fa192b-c00b-0410-8231-f00ffab90ce4

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4041 d4fa192b-c00b-0410-8231-f00ffab90ce4

add "-Wl,--as-needed" to EXTRA_DEVELOPER_CFLAGS to try to prevent
future unordering.

Fixes #486



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4039 d4fa192b-c00b-0410-8231-f00ffab90ce4

fix for #498 -- if a return header overran the obj_workspace (but was small enough to fit in sess_workspace) the header is dropped and an empty slot is put in there -- rest of the code now detects this empty slot. It should probably ask VLC what to do if a header is dropped

git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@4038 d4fa192b-c00b-0410-8231-f00ffab90ce4