Resolve the stdcall/cdecl ambiguity that caused Win2K to crash with the
set service description patch. Also lighten the load (don't allocate
a handle) to get at the advapi32.dll we already statically link to.
PR: 7471
Submitted by: Roman Neuhauser <cynic@mail.cz>
Solved a very serious threading problem with WinNT/2K Services. The
moment master_main told that the shutdown was complete, the parent
control thread exited Apache, leaving mod_jserv's Java process running
and alternately invoking mod_perl's cleanups from the correct thread
or the service control thread. [William Rowe]
Clean out a bunch of stale code. This patch should effectively be a
noop, simply making the code easier to follow before committing the
real fix to our thread-jumping mod_perl/mod_jserv parent process bug.
Jeff Trawick [Fri, 16 Mar 2001 14:13:40 +0000 (14:13 +0000)]
Change any slashes in the "uname -m" output to underbars. HP machines
have slashes in the name, and a slash in the machine name, and thus the
name of a binary distribution, messes up binbuild.sh (among other things).
Chuck Murcko [Fri, 16 Mar 2001 07:01:23 +0000 (07:01 +0000)]
Allow empty headers in mod_proxy. See below:
Empty headers are allowed by RFC2068 section 4.2. The presence or
absence of an empty header can be significant. The current mod_proxy
of httpd 1.3.x removes empty headers. Change mod_proxy to preserve
empty headers.
Submitted by: Christian von Roques <roques@mti.ag>
Reviewed by: Graham Leggett, Chuck Murcko
Jeff Trawick [Wed, 14 Mar 2001 04:43:04 +0000 (04:43 +0000)]
Fix a typo in a comment.
PR: 7404 (which had a patch for the hostname lookup core dump
which was already fixed as well as this typo)
Submitted by: Kunihiro Tanaka <tanaka@apache.or.jp>
Ken Coar [Tue, 13 Mar 2001 20:05:05 +0000 (20:05 +0000)]
Improve customisability of rotatelogs' behaviour. Allow the
logfile name to be formatted according to strftime(3), and
let the user specify an offset from UTC.
Martin Kraemer [Tue, 13 Mar 2001 10:22:21 +0000 (10:22 +0000)]
Ignasi writes,...
The first correction is not important, only removes a redundant line,
the ap_pstrdup(a, "127.0.0.1") is done anyway at the end of the function.
The second correction is really a bug the p-> can be NULL.
Martin Kraemer [Mon, 12 Mar 2001 10:23:34 +0000 (10:23 +0000)]
Brad writes,
Even though I don't run AIX, I was reading BUILD.NOTES and went to the URL
specified in the AIX section about the PTFs to find the URL does not work
anymore, I have found this URL to find the PTFs at
http://techsupport.services.ibm.com/rs6k/fixdb.html
Martin Kraemer [Fri, 9 Mar 2001 10:17:48 +0000 (10:17 +0000)]
David McCreedy writes:
> I've update the text files in os/tpf/samples, in part due to these ebcdic
> patches.
> I'd like to
>
> - remove os/tpf/samples/linkdll.jcl
> - add os/tpf/samples/linkhttp.jcl
Submitted by: "David McCreedy" <mccreedy@us.ibm.com>
Martin Kraemer [Fri, 9 Mar 2001 10:10:55 +0000 (10:10 +0000)]
I added configurable EBCDIC configuration functionality to http_core.c;
because conversion is a base feature, this looked like the only sensible
place to me. Other basic features like Resource limits live there too.
* make EBCDIC conversion truly configurable. Up to know, there was a
hardwired assumption that only files of MIME type text/* (and a few
exceptions) could be EBCDIC files; and everything else HAD TO BE
binary. This assumption breaks (again and again) for MIME types like
application/postscript, model/vrml, application/x-javascript and
others, which can never be stored as EBCDIC text files.
I now implemented two new directives for defining the conversion
based on MIME type or file extension. The conversion can be set for
upload and download independently, on or off, for any file.
* make EBCDIC conversion symmetric. Up to now, there was no clean way
to deal with uploaded content (POST with Content-Type:, PUT with
Content-Type:) because a) there was no MIME checker attached to the
header-reader, and b) you could not actually define a conversion
based on MIME type. Both deficiencies have now been addressed (the
ap_checkconv_in() routine is called upon parsing of an uploaded
Content-Type: line, and it can set a flag which is different from
the download conversion flag. Also, its change is triggered at a
different phase in the request handling: the upload conversion
is set as soon as a Content-Type: for an uploaded document is
detected, but the download conversion is only set when the
response is about to be returned). All this was impossible in the
past.
* unify the EBCDIC tables to be in a single common file
src/ap/ap_ebcdic.c (and src/include/ap_ebcdic.h) instead
of having redundant copies under src/os/{tpf,bs2000,os390}/ebcdic.{c,h}
The common file will have the correct translation table
conditionally compiled based on #ifdef TPF/OS390/_OSD_POSIX.
Up to now, each EBCDIC had its own copy, cloned and slightly
adapted.
* Also, put the conversion checker ap_checkconv() to http_core,
because is is an essential central cog in the whole conversion
clockwork. Until now, three almost incompatible copies were
spread throughout the EBCDIC platforms' os.c files.
Martin Kraemer [Fri, 2 Mar 2001 12:35:53 +0000 (12:35 +0000)]
Ignasi writes:
Both mod_ssl and KAME-ipv6 patch sets are adding the
if (fd >= 0) in the FD_SET(fd, &listenfds) on http_main.c.
I think this patch doesn't concert neither to mod_ssl nor to IPv6.
I agree, and I think it avoids a potential core dump too.
Submitted by: "Roca Carrio, Ignasi" <Ignasi.Roca@fujitsu-siemens.com>
Clean up this mess ... stripped out all the bogus absolute paths
from win32 the .mak structure. Our immunity has been that our master
makefile.win doesn't recurse the .mak projects, and we simply build
all .mak targets in the order we choose. Use src/helpers/fixwin32mak.pl
when exporting these .mak files to keep up with it.
Introduce NUL device pipes for stdin and stdout for the parent Win32
service process. This solves bugs such as PR7198 that report
"dup2(stdin) failed" when trying to use piped logs.
Martin Kraemer [Mon, 26 Feb 2001 14:09:04 +0000 (14:09 +0000)]
Gertjan writes:
I have fixed the problem described in bug report 7114 regarding the
missing WWW-Authenticate header in responses to directory URL's,
while using the DirectoryIndex directive.
PR: 7114
Submitted by: Gertjan van Wingerde <gwingerde@home.nl>
Reviewed by: William A. Rowe, Jim Jagielski, Greg Stein
*) Fixed a potential bug in the 1.3.18 candidate that had the possibility
of corrupting a module's string space if they replaced the r->hostname
argument provided by the Apache core prior to http_vhost unescaping
the hostname string. [Greg Stein, William Rowe]
*) Fixed system shutdown on Windows 2000 to assure that the modules have
an opportunity to clean up. Note there is a _very_ limited amount of
time in which to execute all cleanups [see MSKB Q146092] so all of
the modules may still not be given an opportunity to complete their
cleanups if they require more than 20 seconds total. [William Rowe]
*) Tidied up os/win32/service.c adding several info level debugging
messages that helped resolve the second bullet point, and eliminated
the service_cd() function since we 'do the right thing' in main() and
now get our serverroot from the module's path instead of the current
directory on Win32.
Martin Kraemer [Sat, 17 Feb 2001 20:02:57 +0000 (20:02 +0000)]
If I see it correctly, we don't have an open "core dump candidate" right
now. I've got a good feeling at this point for Jim's proposed release date
for 1.3.18
Martin Kraemer [Sat, 17 Feb 2001 14:24:00 +0000 (14:24 +0000)]
This patch works around dereferencing a zero server pointer.
The assumption is that these come from a misconfiguration, and don't
hurt the running server (because the vhost resolution works in the
opposite direction, from server to vhost). It only hurts in the "-S"
command, and that can be avoided by printing a WARNING about the
misconfiguration.
Martin Kraemer [Sat, 17 Feb 2001 14:07:04 +0000 (14:07 +0000)]
Avoid potential core dump. This patch is a small part of a larger
patch for the ap_unparse_uri_components() routine, and can be safely
applied independently.
Submitted by: "Jeffrey W. Baker" <jwbaker@acm.org>
Reviewed by: Jim Jagielski, Martin Kraemer
dgaudet [Sat, 17 Feb 2001 11:17:45 +0000 (11:17 +0000)]
we have to unescape the hostname at some point... this seems to be the
easiest. (having just gone through all the parsing code again i'm
thinking it would have been nice to have all the parsing and validity
checks in one place.)
PR:
Obtained from:
Submitted by:
Reviewed by:
Very obscure bug; once installed or configured with -w argument, any
error stopping and uninstalling the Apache/Win32 service always prompts
the user to press escape, even if the stop or uninstall command didn't
specify the -w argument. This is the fix.
Reported by: Ryan Gilfillan <rgilfill@covalent.net>
Jim Jagielski [Fri, 16 Feb 2001 19:50:41 +0000 (19:50 +0000)]
Add some detail on which is which. I prefer
#1 since I don't like the idea of 'assuming' http. #3 is a bit
much for 1.3.18, but should be looked at after we release
A quick fix from Brad Nicholes <bnicholes@novell.com> for mod_userdir
to recognize the user's home directory, since Netware doesn't implement
the get home directory mechanics.
Jim Jagielski [Fri, 16 Feb 2001 14:53:24 +0000 (14:53 +0000)]
Get rid of bogus warnings when using '-Wall -Wshadow -Wmissing-prototypes
-Wmissing-declarations':
http_protocol.c: In function `ap_set_byterange':
http_protocol.c:274: warning: `one_start' might be used uninitialized
in this function
http_protocol.c:274: warning: `one_end' might be used uninitialized in
this function
Jim Jagielski [Fri, 16 Feb 2001 14:36:36 +0000 (14:36 +0000)]
Add Jeffrey W. Baker proposed ap_unparse_uri_components() rewrite
for possible inclusion in 1.3.18. At present, I think we should
investigate it, with the knowledge that including it will most
likely push out the release. It's a big change. But I'm not
ready to -1 it for 1.3.18 at present.
Martin Kraemer [Fri, 16 Feb 2001 08:25:54 +0000 (08:25 +0000)]
Brad says: [...] This patch just
cleans up the OpenBSD section, removes an unnecessary/unused flag which is
used only when creating a shared library/module from an archive which is
not being done thus it's not needed.
From ld man page:
-Bforcearchive
Force all members of archives to be loaded, whether or not such
members contribute a definition to any plain object files. Useful
for making a shared library from an archive of PIC objects with-
out having to unpack the archive.
*) mod_user on Win32 bug introduced in 1.3.17 is corrected. The parsing
is completely rewritten so all platforms share the same file/redirect
logic interpreting UserDir directives. Specifying a Win32 or Netware
relative path or a relative path on any platform using an '*' now
emits an error when httpd.conf is first parsed. [William Rowe]
Add third state to track the -w flag. This will prevent command line
users and the installer from having the console held open, while the
shortcuts will add the -w flag as the first option.
This patch also splits the Netware and Win32 getopt arg lists.
Martin Kraemer [Tue, 13 Feb 2001 22:26:07 +0000 (22:26 +0000)]
Observed a core dump because a ipaddr_chain struct contains
a NULL-"server" pointer when being dereferenced by invoking "httpd -S".
See Message-ID: <20010213231854.A20932@deejai2.mch.fsc.net>