robertc [Sat, 16 Aug 2003 06:19:55 +0000 (06:19 +0000)]
Summary: Use SQUID_C[XX]FLAGS throughout configure.
Keywords:
Where we are setting defines etc. that are required to build squid correctly, set them via SQUID_C[XX]FLAGS, not C[XX]FLAGS, so that the user can override the user-related flags (i.e. optimisation / profiling / debugging) without breaking the squid build.
robertc [Fri, 15 Aug 2003 19:06:34 +0000 (19:06 +0000)]
Summary: Prevent pushing persistent connections with outstanding writes.
Keywords:
Adrians' recent post to squid-dev looks like it's triggered by persistent connections being pushed inappropriately... this change adds an assert if a persistent connection is pushed while a write is outstanding.
robertc [Thu, 14 Aug 2003 18:15:04 +0000 (18:15 +0000)]
Summary: Swallow request bodies even when denying access.
Keywords:
* Assert when destroying requests with body connections.
* Move ClientBody class definition to client_side.h
* Encapsulate comm_close calls from ClientSocketContext.
* When closing a client socket, if there is a body, swallow it.
* When swalling a request body, if the connection needs to close, do so.
hno [Thu, 14 Aug 2003 03:49:45 +0000 (03:49 +0000)]
Bug #718: accel with nonstandard cache_peer port uses standard port
The recent fix for vhost acceleration and persistent connections
confusing certain servers did a little bit too much, accidently using
the port of the URL rather than the cache_peer in case of origin class
peers.
The algorithm that calculates the timeout for a set of ICP
queries ignores multicast neighbors. It also ignores the
expected number of replies because "*exprep" is always set
equal to parent_exprep + sibling_exprep.
This patch adds multicast group counts to the expected number
of ICP replies. It also sets the query timeout to the
multicast peer rtt (x2) if there were no parents query. If
there are parents, the expected parents timeout is used.
robertc [Wed, 13 Aug 2003 06:17:24 +0000 (06:17 +0000)]
Summary: Add %<sS and %<sH to custom log formats.
Keywords:
Bugfix to access logging - the last request in a connection was racing with releasing of the store entry.
Add new high offset and available offset counters.
robertc [Mon, 11 Aug 2003 19:07:56 +0000 (19:07 +0000)]
Summary: Cachemgr page length - Guido.
Keywords:
Hi,
On Windows native port, cachemgr parses the squid output in TEXT mode, with
a wrong resultant Content-Length in html pages, see this thread for details:
hno [Mon, 11 Aug 2003 00:57:08 +0000 (00:57 +0000)]
Bug #663: Username not logged into ACCESS.LOG in case of /407
The Basic auth module did not implement the direction call correctly,
indication authentication scheme failure on unsuccessful helper lookup
rather than "completed".
Note: This API and it's interactions with checklist->auth_user_request &
request->auth_user_request looks overly complex for the job. Should be
possible to simplify this a lot I think.
robertc [Sun, 10 Aug 2003 15:53:49 +0000 (15:53 +0000)]
Summary: Fixup remaining MSVC issues.
Keywords:
Addresses -
protos.h:
c:\work\nt-3.0\src\protos.h(443) : warning C4190: '<Unknown>' has C-linkage
specified, but returns UDT 'String' which is incompatible with C
c:\work\nt-3.0\src\squidstring.h(77) : see declaration of 'String'
c:\work\nt-3.0\src\protos.h(444) : warning C4190: '<Unknown>' has C-linkage
specified, but returns UDT 'String' which is incompatible with C
c:\work\nt-3.0\src\squidstring.h(77) : see declaration of 'String'
c:\work\nt-3.0\src\protos.h(445) : warning C4190: '<Unknown>' has C-linkage
specified, but returns UDT 'String' which is incompatible with C
c:\work\nt-3.0\src\squidstring.h(77) : see declaration of 'String'
c:\work\nt-3.0\src\protos.h(446) : warning C4190: '<Unknown>' has C-linkage
specified, but returns UDT 'String' which is incompatible with C
c:\work\nt-3.0\src\squidstring.h(77) : see declaration of 'String'
c:\work\nt-3.0\src\protos.h(447) : warning C4190: '<Unknown>' has C-linkage
specified, but returns UDT 'String' which is incompatible with C
c:\work\nt-3.0\src\squidstring.h(77) : see declaration of 'String'
protos.h:
c:\work\nt-3.0\src\protos.h(740) : error C2526: 'storeCreate' : C linkage
function cannot return C++ class 'RefCount<class storeIOState>'
c:\work\nt-3.0\src\protos.h(741) : error C2526: 'storeOpen' : C linkage
function cannot return C++ class 'RefCount<class storeIOState>'
typedefs.h:
c:\work\nt-3.0\src\httprequest.h(56) : warning C4099: 'HttpRequest' : type
name first seen using 'struct' now seen using 'class'
c:\work\nt-3.0\src\typedefs.h(202) : see declaration of 'HttpRequest'
robertc [Sun, 10 Aug 2003 09:59:19 +0000 (09:59 +0000)]
Summary: Fix range offsets when the full object is retrieved.
Keywords:
* Extend StoreIOBuffer to provide Range details.
* Extend ClientSocketContext::lengthToSend to be offset aware and adjust throughout.
* ClientSocketContext::packRange - tidy up the calling interface, make private, and skip unwanted data at the beginning as well as the end of ranges.
robertc [Wed, 6 Aug 2003 03:39:59 +0000 (03:39 +0000)]
NetBSD defines bswap16() and bswap32(). The re-definitions in [squid]/include/ntlmauth.c cause compile errors. I'll send a patch to the squid-dev list soon. - Alan Barrett (bug #724)
hno [Mon, 4 Aug 2003 04:53:47 +0000 (04:53 +0000)]
Cut out unwanted regex interfaces from GNUregex.c
- emacs related stuff
- BSD 4.2 regex interface
- old GNU regex interface
leaving only the POSIX interface which is what we use
robertc [Sun, 3 Aug 2003 16:37:30 +0000 (16:37 +0000)]
From email:
> > MemObject.cc
>
>Looks like another limitation. taking the first one: there is a for_each
>template that is explicitly for dlink_list. You might try
>for_each<StoreClientsStats>(clients, statsVisitor);
robertc [Sun, 3 Aug 2003 16:05:12 +0000 (16:05 +0000)]
Hi,
On Windows (native and Cygwin) and OS/2, when running
UFSSwapDir::closeTmpSwapLog() in store_dir_ufs.cc, Squid can be fail if the
target file for a rename operation was already deleted.
Because xrename() already try to remove the destination target on Windows,
a previous unlink() is not needed.
This patch extended the native Windows xrename() behaviour to Cygwin and
OS/2 too and remove not needed unlink().
robertc [Sun, 3 Aug 2003 14:09:00 +0000 (14:09 +0000)]
Various patches from Pawel Worach
[patch] make trans. ipfilter compile
one else too many?
[patch] update .cvsignore files
Make all .cvsignore files up-to-date and add some
new ones to libTrie.
[patch] make ntlm_auth compile on freebsd 5.x
I have no idea how portable this is but freebsd doesn't
use malloc.h anymore.
# grep error /usr/include/malloc.h
#error "<malloc.h> has been replaced by <stdlib.h>"
[patch] make auth_ntlm compile if NTLM_FAIL_OPEN is defined
ntlm/auth_ntlm.cc: In function `stateful_helper_callback_t
authenticateNTLMHandleReply(void*, void*, char*)':
ntlm/auth_ntlm.cc:601: error: invalid conversion from `void*' to `
ntlm_helper_state_t*'
[patch] make useragent and referer log stuff compile
Can't use class RefCount where a bool is expected says gcc 3.3.1
I guess this is the right fix :)
[patch] pam_auth has to be installed setuid root
pam_auth has to be setuid root for it to work.
this could apply to 2.5 too.
Summary: Various fixes - comms and diskio.
Keywords:
Comm accept logic was not indicating that it was not cbdata.
Also, mark eventAdd as extern, not C extern.
UFS IO module was not reporting failure on failed opens.
AUFS - log when reads are scheduled, as queued reads are scheduled after the read_ call.
Fix a race condition with synchronous store opens, that previously lead to double-callbacks.
This fixes many issues in mgr:config dump of custom access log format
specifications. Most notably it no longer hangs, but should now also
be capable to return the same specification as originally given in
squid.conf.
An unfortunate bug. The hdr->entries array is initialized
such that count is set to zero. httpHeaderClean() seems to
be called both when 'hdr' is created, and destroyed. Thus,
we accumulate a large number of zero counts for 'hdr' before
it is ever used. Can't think of a good way to fix it, except
adding a state variable that indicates whether or not 'hdr'
has been used. As a hack, just never count zero-sized header
arrays.
Remove bitfield definition of quote. Having it as a bitfield only marginally
saves memory, but probably both confuses compilers and make more inefficient
code. Robert reported issues with using GCC-2.95 related to this.
Summary: More MSVC compatability tweaks.
Keywords:
More pointer construction to pointer assignment replacements.
Add a test case for delete - checking that the correct delete operator is used in a class hierarchy.
Partly fix g++ 2.95 builds - there sstream includes the system assert, which breaks our custom assert macro. So, a couple of undefs and we have an ugly fix.
Give HttpRequest it's own mempool.
Give storeSwapLogData it's own file, mem pool.
Provide a file_write adapter for objects.
Give helper_stateful_request it's own mem pool.
Move into new file helper.h
Give helper_request it's own mem pool.
Move into new file helper.h
Make all arrays typesafe.
Remove C bindinds to Array logic (we can instate new ones easily if desired later on).
Make all users of HttpHeader classes: htcpReplyData.
Ensure all class contianing HttpHeader are constructed and destructed:
htcpReplyData, request_t, HttpReply.
rename request_t to HttpRequest.
Move htcp prototypes to htcp.h.
projects / thirdparty / squid.git / log
