Per W. Stoddard, Greg Marr and my research, let's get this right.
This change is a decorative NOOP. It accurately reflects what the
MSVC compiler actuall turned into bytecode. We hope this change
makes the actual fns clearer to the developer.
However, there is no effective difference. The _stdcall declaration
for these functions was _ignored_ by MSVC, which treated the fn's
as _cdecl (our API_EXPORT_NONSTD macro) because they use varargs.
This time, buff.c:ap_bvputs() was out of sync (correct to this patch)
ap_rprintf is implemented as API_EXPORT() in http_protocol.c.
Therefore, make the prototype and implementation consistant.
The question remains, why no compiler warning/emit? Because MSVC [my
version, at least] must have changed all of the ap_fn(foo, ...) decl
from _stdcall to _cdecl, on it's own!
I suggest our declarations are still borked, but they have worked only
because MSVC ignored our poor judgement :)
Several final namespace changes. Exports (with an ap_ prefix!) the
useful sendwithtimeout/recvwithtimeout symbols. How useful? SSL needs
them, and proxy_connect should probably use them as well. And corrects
two newly exported (never released) symbols since they did not require
the _NONSTD semantics.
Dispatch 26 compiler emits into oblivion. Vetting is desired, please
post to the list if you participate. They are all blindingly obvious,
but extra eyes always help
This eliminates all but the regex emits and MSVC's borked misdeclaration
of FD_SET.
Bill Stoddard [Thu, 17 Jan 2002 21:21:09 +0000 (21:21 +0000)]
On HPUX 11.x, the 'ENOBUFS, No buffer space available'
error occures because the accept() cannot complete.
You will not see ENOBUFS at 10.20 because the kernel
hides any occurrence from being returned from user space.
ENOBUFS at 11.0 TCP/IP is quite possible, and could
occur intermittently. As a work-around, we are going to
ingnore ENOBUFS.
Submitted by: madhusudan_mathihalli@hp.com
Reviewed by: Bill Stoddard
Jim Jagielski [Thu, 17 Jan 2002 13:20:51 +0000 (13:20 +0000)]
PR:
Obtained from:
Submitted by: Stipe Tolj <tolj@wapme-systems.de>
Reviewed by:
Add in Stipe's Cygwin changes.... Have not folded in the Cygwin
timeout kill signaling patch yet... waiting for feedback.
Here are the changes:
* src/include/ap_config.h: adding HAVE_PTHREAD_SERIALIZED_ACCEPT to
Cygwin block and defaulting to it.
* src/main/http_main.c: exluding pthread_mutexattr_setpshared() call
for Cygwin platform. This calls seems yet not fully functional on this
platform. Forwarding problem to Cygwin core developers.
* src/Configure: changed OS_MODULE_INCLUDE var for the Cygwin
platform block to refer to the (absolute) relative path for the
Makefile.Cygwin file.
* src/modules/proxy/Makefile.tmpl: target libproxy.dll is hardcoded
for OS/2. That's not good. We have the same target on Cygwin, but use
other calls for it, so if construct here.
* src/modules/standard/Makefile.Cygwin: adding a dummy target %.def
: %.c to satisfy OS/2 related dependacies on Cygwin too. Changed the
"run make twice" screen slightly.
Jeff Trawick [Wed, 16 Jan 2002 15:52:15 +0000 (15:52 +0000)]
fix a syntax error in an initializer which prevented http_main.c from
compiling on at least OS/390, AIX w/ xlc, and HP-UX with HP compiler;
Tru64 cc previously complained but generated code anyway; gcc -Wall
didn't complain before
unfortunately, this introduces a warning with gcc -Wall on Solaris since
in_addr is a union
Bill Stoddard [Tue, 15 Jan 2002 17:27:32 +0000 (17:27 +0000)]
This patch fixes my previous change for argument passing, put in place
to support rotatelogs. It only affects TPF. I was stupidly displacing past
the end of the "args" array while setting up to fork to rotatelogs (or a CGI).
Tony Finch [Mon, 14 Jan 2002 04:49:44 +0000 (04:49 +0000)]
Following taunts from Alfred Perlstein on IRC, use "httpready"
accept filter rather than "dataready" on FreeBSD after 4.1.1-RELEASE
where it works correctly.
Ken Coar [Wed, 9 Jan 2002 19:47:03 +0000 (19:47 +0000)]
Whoops, forgot to bump MMN. Major bump because of a change
to the semi-private core_dir_config structure. (Also fix a
stale comment from an earlier version of the FileETag patch.)
It's clear from recent security reports that including configuration
examples for non-core add-ins with our default .conf files causes some
admins to bypass actually researching -how-to-configure- the given
add-in module. While this example was nice, we need to find another
useful example, in addition to x-tar perhaps, to make this clearer.
Martin Kraemer [Tue, 8 Jan 2002 15:09:48 +0000 (15:09 +0000)]
The 416 "range not satisfiable" response would include a
Content-Length header set to the size of the resource, but no body was
actually returned (r->header_only was set to 1). This appeared as a
premature EOF to the client.
Submitted by: Joe Orton <joe@manyfish.co.uk>
Reviewed by: Martin Kraemer
Ken Coar [Sat, 5 Jan 2002 17:13:03 +0000 (17:13 +0000)]
Replace automatic ETag generation with configurable algorithm
defined by FileETag directive. Keywords are All, None,
[+|-]INode, [+|-]Size, [+|-]MTime. Server farms with content
fanout may want to use 'FileETag MTime Size' to keep the system-
variant inode from being included and busting caches.
PR: 7010
Submitted by: Based on an idea proposed by Phil Dietz
Fixed the problem on NetWare when accessing an empty directory which
has option indexes specified produces an access forbidden message.
Submitted by: Charles Goldman, Guenter Knauf
Fixed the problem on NetWare when accessing an empty directory which
has option indexes specified produces an access forbidden message.
Submitted by: Charles Goldman, Guenter Knauf
Normalize symbol exports for Win32/Netware to the httpd.exp reference.
Diff tags pre_win_nw_syms/post_win_nw_syms for complete edit.
Note I've corrected _SEVERAL_ badly declared symbols on Win32 into
API_EXPORT_NONSTD flavors (e.g. those using (...) args). This may,
or may not, break binary compatibility depending on how those args
are addressed, and if those functions were used.
I've further tested by setting aside the .def file and rebuilding,
and validated that our symbols list corresponds to the API_DECLARE()
macros at this moment.
Submitted by: Thomas Eibner <thomas@stderr.net>
Reviewed by: Stoddard, Rowe
Already patched in the 2.0 tree - make all .conf's reflect the applicable
DNS rfc's that underbars aren't permitted - why should our examples show
administrators otherwise?!?
One of two gdawful bugs in Apache/Win32, start capturing child errors.
The other [significant] error is the tendancy for the parent to restart
instantly on init errors, which are never corrected. But that's a patch
for another day.
I know Stoddard fixed this other bullet - would someone with a long
memory (Jim? Marc?) be willing to get rid of a few more now-dead items?
I'm not certain on the others, but there are several bullets that look
like they've been addressed, or are no longer relevant.
Fix a very irritating nit - this was the 1.3.20 code - there was no
call for changing the formatting when the patch to add another platform
in the #if case was modified.
Revert the code causing lost path_info/query_args.
The potentially correct solution is to test the URI with recombined
path_info and args, but I'd rather see this reverted for 1.3 and
corrected in 2.0 with appropriate optimizations and user testing.
Bill Stoddard [Fri, 30 Nov 2001 14:08:44 +0000 (14:08 +0000)]
Modify buff.h and buff.c to enable modules to intercept the
output byte stream for dynamic page caching. A pointer to a
'filter callback' function is added to the end of buff.h.
This function, if registered by a module, is called
at the top of buff_write() and writev_it_all().
MMN Minor bumped.
Obtained from: [Kevin Mallory <kmallory@spidercache.com>
Reviewed by: Bill Stoddard
Correct mod_autoindex's test of the query string's requested ordering.
This patch is a bit more paranoid than Kestutis's, but this is user
supplied input, so a bit of paranoia is healthy.
dgaudet [Sun, 11 Nov 2001 19:46:41 +0000 (19:46 +0000)]
rev 1.248 of this file removed the default locking mechanism
from unixware 7.0. instead it should have just defined
SINGLE_LISTEN_UNSERIALIZED_ACCEPT (since locking is always required for
multiple sockets).
Same IDE enforcement of InstallBin as the 'default' project from the 2.0
tree, so when the user first confronts Apache sources they have the right
top-level project.
Prevent an Apache module from being loaded or added twice due
to duplicate LoadModule or AddModule directives (or a missing
ClearModuleList directive).
LoadModule dupcheck (for 2.0) by Brian Pane <bpane@pacbell.net>
Ported and added AddModule dupcheck by Will Rowe
Identified by an old collegue of Will's who tripped over this.
Ken Coar [Fri, 26 Oct 2001 18:05:26 +0000 (18:05 +0000)]
Some platforms varf on a setgid(-1) and hence httpd will fall
over immediately after being started. However, since
'Group #-1' is syntactically correct, apachectl won't catch
this and will assume the server started successfully. This
checkgid app will return -1 if any of the Apache-understandable
group values (i.e., name or "#n") are invalid. apachestl still
needs to be enhanced to use this.
Jeff Trawick [Thu, 18 Oct 2001 15:25:26 +0000 (15:25 +0000)]
HP-UX requires DSOs to be executable. "make install" takes care of
this, but binbuild-install.sh does not. This patch tweaks
install-bindist.sh to make the DSOs executable. As with "make
install", the DSOs will be marked executable for all platforms, not
just HP-UX. (Actually, install-sh is invoked by "make install" a
little differently for DSOs on rhapsody/darwin, but there is no
special processing for rhapsody/darwin in install-bindist.sh
currently so they shouldn't be hurt by the change below.)
Bill Stoddard [Wed, 17 Oct 2001 14:45:29 +0000 (14:45 +0000)]
Win32: The Apache Win32 developers generally recommend that
MaxRequestsPerChild be set to 0 to prevent the child process
from ever recycling. However, for those that do require a
non-zero setting, this patch fixes a serious bug that can cause
an apparent 'server-hang' condition where the server stops
responding to requests for a period of time. Prior to this
fix, when the child process handled MaxRequestsPerChild
connnections, the child process would stop accepting new
connections and begin allowing inactive threads to exit. The
problem was that a new process would not be created to begin
handling requests until the old process fully exited. The old
process can take an indeterminate amount of time to exit because
it may be sending large responses to clients connected over slow
links, or it may have threads blocked in read awaiting requests
(eg, one attack mode of the Nimda worm is to establish a
connection to the server but not send an HTTP request. This
connection will be timed out according to the setting of the
Timeout directive, 300 seconds). This fix allows the new process
to be immediately started and begin accepting requests when the
old child process reaches MaxRequestsPerChild.
Jeff Trawick [Tue, 9 Oct 2001 19:29:46 +0000 (19:29 +0000)]
AIX 5L has socklen_t, so use it so we avoid some warnings
AIX 4.3.3 has socklen_t too, but
a) I don't know if all 4.3 (e.g., 4.3.0) has it
b) I don't see the warning on 4.3
c) it isn't a possible 64-bit vs. 32-bit problem, so I'm not
worried about run-time breakage
Parse individual paths listed in the UserDir directive for correctness.
This also addresses the 'quote' problem, where quoted paths would fail
the is absolute redirect or absolute path test on Netware/Win32, given
a "/somepath" designation.
Mark J. Cox [Mon, 8 Oct 2001 19:34:38 +0000 (19:34 +0000)]
Add CVE candidate names to the announcement mail and changes file to allow
them to be cross-referenced with other security publications easily
PR:
Obtained from:
Submitted by:
Reviewed by:
projects / thirdparty / apache / httpd.git / log
