hno [Mon, 29 Oct 2001 23:06:31 +0000 (23:06 +0000)]
Further cleanups of Bugzilla #255 related issues.
* Partly backed out the previous patch. We can store the status, only need
to do it slightly different.
* Same error was repeated a few lines down for another "error" case.
(out of filedescriptors)
hno [Sat, 27 Oct 2001 05:11:56 +0000 (05:11 +0000)]
Oops.. the previous bugfix to plug leaks on errors did it too well,
causing some memory to be freed twice. This confusion was caused by
some very odd management of the request buffer in clientReadRequest /
parseHttpRequest. This patch cleans this up to use a single unified
policy.
hno [Tue, 23 Oct 2001 05:35:07 +0000 (05:35 +0000)]
Removed type dependency on glib.h. This is not needed as configure
checks for all variants, and the glib wrappers crashes the build if
activated as glib.h is not always included..
hno [Sat, 20 Oct 2001 04:34:48 +0000 (04:34 +0000)]
SSL update from the "ssl" branch at SourceForge
* Several SSL tweaking options
- SSL version per https_port, no longer a global setting
- supported chipers, per https_port
- supported protocols, per https_port
- connection shutdown method
* Fix the bug reported by Noel Burton-Krahn where SSL connections
could get hung with data pending in the SSL internal buffers. Mostly
seen on large POST/PUT requests, but could in theory appear on any
request larger than 4K.
hno [Wed, 17 Oct 2001 19:43:06 +0000 (19:43 +0000)]
More SunPRO cc warnings and errors..
* Some signed/unsigned char conflicts
* Errored out on a // comment.. this is C not C++.
* cleanup of sfileno usage in "ufs".. not good to have variables named the
same as a type.. and even worse when the type differs.
hno [Wed, 17 Oct 2001 19:30:49 +0000 (19:30 +0000)]
Fixes various warnings emitted in lib by the SunPRO cc compiler, and
synchronized the MD5 prototypes with OpenSSL.
* MD5 operates on binary data, not characters. Avoid signed/unsigned issues
* The bswap16/32 macros was not signed/unsigned safe
* lib/rfc2617 had several spurious extra ;
* MD5Final wants an unsigned char *
adrian [Wed, 17 Oct 2001 16:59:08 +0000 (16:59 +0000)]
A bunch of quick patches to silence gcc warnings when compiling
with FreeBSD's BDECFLAGS. These include duplicate declarations,
declarations shadowing global declarations, implicit type size
and signed/unsigned conversions.
hno [Sat, 13 Oct 2001 00:20:23 +0000 (00:20 +0000)]
Potential fix for Bugzilla #236
Reading the code shows that if the reply contains a single truncated entry
then rfc1035AnswersUnpack will return 0 records, possibly causing assertion
failure in rfc1035RRDestroy
adrian [Wed, 10 Oct 2001 21:17:37 +0000 (21:17 +0000)]
Commit Roger Venning's TOS/source address ACL selection work.
This code allows the user to set ToS bits or the source address
of an ACL by defining an ACL and then using that ACL in
(mutiple instances of the) tcp_outgoing_address statement.
hno [Sat, 6 Oct 2001 03:54:03 +0000 (03:54 +0000)]
Be consistent with most other helpers in how to manage spaces. If there
is space characters then assume these are from the password, not the login
name.
hno [Thu, 4 Oct 2001 05:00:07 +0000 (05:00 +0000)]
Further cleanup of quick_abort description. It is wrong to say that
it can be configured to continue downloading when in fact the default
is to continue if there is less than 16 KB remaining.
hno [Thu, 4 Oct 2001 04:56:20 +0000 (04:56 +0000)]
Make the wording in quick_abort less ambigous. Lots of people seem to
think that quick_abort_min actually makes Squid to never continuing
the downloads..
hno [Wed, 3 Oct 2001 15:02:24 +0000 (15:02 +0000)]
Added maintainer mode to not have automake dependencies rebuild automake
files bo default. This is likely to be confusing and causing lots of problems
to the causal user, especially considering the rather strict requirements
we have on automake versions...
Resurrected Duanes old sleep_after_fork patch. This needed or we will
race with ourselves for swap space when restarting helpers. The hello
test is not really sufficient, we must also wait for the child to actually
exec() before forking again. In the event that we are starting many
helpers who uses CPU it is not too unlikely to see a newly started child
to be context switched before exec().
Anyway, default configuration does nothing. To activate this one must
configure sleep_after_fork in squid.conf so it is a quite harmless patch,
and if it helps some people, fine.
Move CACHE_HTTP_PORT and CACHE_ICP_PORT from config.h.in to acconfig.h.
Move config.h.in to config.h to reduce unneeded recompilations after a configure.
Regenerate autotools files.
Bugfix: Vary failed to calculate the variance when going thru a parent.
I obviously never learn to use httpState->orig_request when accessing
client data. Not the first time I have been bitten by httpState->request
being a bogus request missing most of the vital information when using
a peer proxy (only contains the request line, and how to contact the peer).
adrian [Sat, 8 Sep 2001 00:02:44 +0000 (00:02 +0000)]
Remove MEM_STMEM_BUF from the scene, and statically define the stmem
buffer into the mem_node struct. This should cut down on the number
of allocation requests made per HTTP request.