Newer
Older
file exists or not. Instead we now directly check for the existence of
those esoteric ones.
[Ralf S. Engelschall] PR#2093, PR#2361, PR#2377, PR#2434,
PR#2524, PR#2525, PR#2533, PR#2569
*) mod_setenvif (BrowserMatch* and friends) will now match a missing
field with "^$". [Ken Coar]
*) Set the RTLD_GLOBAL dlopen mode parameter to allow dynamically loaded
modules to load their own modules dynamically. This improves mod_perl
and mod_php3 when these modules are loaded dynamically into Apache.
[Rasmus Lerdorf]
*) Cache a proxied request in the event that the client cancels the
transfer, provided that the configured percentage of the file has
already been transferred. It works for HTTP transfers only. The
new configuration directive is called CacheForceCompletion.
*) Add the "<!DOCTYPE HTML" magic cookie used by modern documents (and
required by HTML 3.2 and later) to mod_mime_magic's conf/magic.
*) Fix yet another signal-based race condition involving nested timers.
Signals suck. [Dean Gaudet]
*) suexec's error messages have been clarified a little bit. [Ken Coar]
*) Clean up some, but perhaps not all, 8-bit character set problems
with config file parsing, and URL parsing. We now define
ap_isdigit(), ap_isupper(), ... which cast to an (unsigned char).
This should work on most modern unixes.
[Dean Gaudet] PR#800, 2282, 2553 (and others)
*) The "handler not found" error was issued in cases where the handler
really did exist, but was just declining to serve the request.
*) Add Dynamic Shared Object (DSO) support for SCO5 (OpenServer 5.0.x).
*) The APACI libexecdir was not extended with an "apache/" subdir
if the installation prefix didn't already contain "apache", but
it should be because the DSO files are Apache-specific. Now
libexecdir is treated the same way sysconfdir, datadir, localstatedir
and includedir are already treated.
*) The <Limit> parsing routine was incorrectly treating methods as
case-insensitive. [Ken Coar]
*) The ap_bprintf() code neglected to test if there was an error on
the connection. ap_bflush() misdiagnosed a failure as a success.
[Dean Gaudet]
*) add support for #perl arg interpolation in mod_include
[Doug MacEachern]
*) API: Name changes of table_elts to ap_table_elts, is_table_empty
to ap_is_table_empty and bgetflag to ap_bgetflag. [Ben Laurie]
*) PORT: Add UnixWare 7 support
10064
10065
10066
10067
10068
10069
10070
10071
10072
10073
10074
10075
10076
10077
10078
10079
10080
10081
*) Fix the Guess-DSO-flags-from-Perl stuff in src/Configure: "perl" was
used instead of "$PERL" which contains the correctly determined Perl
interpreter (important for instance on systems where "perl" and "perl5"
exists, like BSDI or FreeBSD, etc).
[Ralf S. Engelschall] PR#2505
*) Move the initial suEXEC-related startup message from plain
fprintf()/stderr to a delayed ap_log_error()-based one to avoid problems
when Apache is started from inetd (instead of standalone). Under this
situation startup messages on stderr lead to problems (the line is sent
to the client in front of the requested document).
[Ralf S. Engelschall] PR#871, PR#1318
*) Add a flag so ap_fnmatch() can be used for case-blind pattern matching.
[Ken Coar, Dean Gaudet]
*) WIN32: Don't collapse multiple slashes in PATH_INFO.
*) WIN32 SECURITY: Eliminate trailing "."s in path components. These are
ignored by the Windows filesystem, and so can be used to bypass security.
[Ben Laurie, Alexei Kosut].
*) We now attempt to dump core when we get SIGILL. [Jim Jagielski]
*) PORT: remove broken test for MAP_FILE in http_main.c.
10092
10093
10094
10095
10096
10097
10098
10099
10100
10101
10102
10103
10104
10105
10106
10107
10108
10109
10110
10111
10112
10113
10114
10115
10116
10117
*) PORT: Change support/apachectl to use "kill -0 $pid" to test if the
httpd is running. This should be more portable than figuring out
which of three dozen different versions of "ps" are installed.
[a cast of dozens]
*) WIN32: If we can't figure out how to execute a file in a script
directory, bail out of the request with an error message. [W G Stoddard]
*) WIN32 SECURITY: Eliminate directories consisting of three or more dots;
these are treated by Win32 as if they are ".." but are not detected by
other machinery within Apache. This is something of a kludge but
eliminates a security hole. [Manoj Kasichainula, Ben Laurie]
*) Move ap_escape_quotes() from src/ap to src/main/util.c; it uses
pools and thus pollutes libap (until the pool stuff is moved there).
[Ken Coar]
*) IndexIgnore should be case-blind on Win32 (and any other case-aware
but case-insensitive platforms). New #define for this added to conf.h
(CASE_BLIND_FILESYSTEM). [Ken Coar] PR#2455
*) Enable DSO support for OpenBSD in general, not only for 2.x, because it
also works for OpenBSD 1.x. [Ralf S. Engelschall]
*) PORT: Fix compilation problem on ARM Linux.
10119
10120
10121
10122
10123
10124
10125
10126
10127
10128
10129
10130
10131
10132
10133
10134
10135
10136
10137
10138
10139
10140
10141
10142
10143
10144
10145
10146
*) Let APACI's configure script determine some configuration parameters
(Group, Port, ServerAdmin, ServerName) via some intelligent tests to
remove some of the classical hurdles for new users when setting up
Apache. This is done per default because it is useful for the average
user. Package authors can use the --without-confadjust option to disable
these configuration adjustments.
[Ralf S. Engelschall]
*) Added an EXTRA_DEPS configuration parameter which can be used
to add an extra Makefile dependency for the httpd target, for instance
to external third-party libraries, etc.
[Ralf S. Engelschall]
*) Add <IfDefine>..</IfDefine> sections to the core module (with same spirit
as <IfModule>..</IfModule> sections) which can be used to skip or process
contained commands dependend of ``-D PARAMETER'' options on the command
line. This can be used to achieve logical conditions like <IfDefine
ReverseProxy> instead of physically ones (e.g. <IfModule mod_proxy.c>)
and thus especially can be used for conditionally loading DSO-based
modules via LoadModule, etc. [Ralf S. Engelschall]
*) PORT: clean up a warning in mod_status for OS/2. [Brian Havard]
*) Make table elements const. This may prevent obscure errors. [Ben Laurie]
*) Fix parsing of FTP `SIZE' responses in proxy module: The newline was not
truncated which forced following HTTP headers to be data in the HTTP
PR#2412, 2367
*) Portability fix for APACI shadow tree support: Swap order of awk and sed
in top-level configure script to avoid sed fails on some platforms (for
instance SunOS 4.1.3 and NCR SysV) because of the non-newline-termined
10154
10155
10156
10157
10158
10159
10160
10161
10162
10163
10164
10165
10166
10167
10168
10169
10170
10171
10172
10173
10174
10175
10176
10177
10178
10179
10180
10181
10182
10183
10184
10185
10186
10187
*) Improve performance of directory listings (mod_autoindex) by comparing
integer keys (last-modified and size) as integers rather than converting
them to strings first. Also use a set of explicit byte tests rather
than strcmp() to check for parent directory-ness of an entry. Oh, and
make sure the parent directory (if displayed) is *always* listed first
regardless of the sort key. Overall performance winnage should be good
in CPU time, instruction cache, and memory usage, particularly for large
directories. [Ken Coar]
*) Add a tiny but useful goody to APACI's configure script: The generation
of a config.status script (as GNU Autoconf does) which remembers the used
configure command and hence can be used to restore the configuration by
just re-running this script or for remembering the configuration between
releases.
[Ralf S. Engelschall]
*) Add httpd -t (test) option for running configuration syntax tests only.
If something is broken it complains and exits with a return code
non-equal to 0. This can be used manually by the user to check the Apache
configuration after editing and is also automatically used by apachectl
on (graceful) restart command to make sure Apache doesn't die on restarts
because of a configuration which is now broken since the last (re)start.
This way `apachectl restart' can be used inside cronjobs without having
to expect Apache to be falling down. Additionally the httpd -t can be run
via `apachectl configtest'.
[Ralf S. Engelschall] PR#2393
*) Minor display fix for "install" target of top-level Makefile:
the displayed installation command was incorrect although the
executed command was correct. Now they are in sync.
[Ralf S. Engelschall] PR#2402
*) Correct initialization of variable `allowed_globals' in http_main.c
*) Apache would incorrectly downcase the entire Content-Type passed from
CGIs. This affected server-push scripts and such which use
multipart/x-mixed-replace;boundary=ThisRandomString.
[Dean Gaudet] PR#2394
*) PORT: QNX update to properly guess 32-bit systems.
10197
10198
10199
10200
10201
10202
10203
10204
10205
10206
10207
10208
10209
10210
10211
10212
10213
10214
10215
*) Make sure the DSO emulation code for HPUX finds the proprietary shl_xxx()
functions which are in libdld under HPUX 9/10.
[Ralf S. Engelschall] PR#2378
*) Make sure the "install" target of the top-level Makefile doesn't break
because of a return code of 1 from an "if" (for instance under braindead
Ultrix the result code of an "if" construct is 1 if the "then" clause
didn't match). [Ralf S. Engelschall]
*) Add an additional "dummy" target to the "$(LIB)" target in generated
modules/xxx/Makefile's to avoid problems with SVR4 Make under "full-DSO"
situation (no libxxx.a built, only mod_xxx.so's) where LIB and OBJS are
empty. [Ralf S. Engelschall, Dean Gaudet, Martin Kraemer]
*) Replace two bad sprintf() calls with ap_snprintf() variants in
mod_rewrite. [Ralf S. Engelschall]
*) Fix missing usage description for MetaFiles directive.
*) mod_log_config wouldn't let vhosts use log formats defined in the
*) mod_usertrack was corrupting the client hostname. As part of the
fix, the cookie values were slightly extended to include the
fully qualified hostname of the client.
[Dean Gaudet] PR#2190, 2229, 2366
*) Fix a typo in pool debugging code. [Alvaro Martinez Echevarria]
*) mod_unique_id did not work on alpha linux (in general on any
architecture that has 64-bit time_t).
[Alvaro Martinez Echevarria]
*) PORT: Make SCO 5 (and probably 3) compile again. [Ben Laurie]
*) PORT: NCR MPRAS systems have the same bug with SIGHUP restart that
Solaris systems experience. So define WORKAROUND_SOLARIS_BUG.
*) Change "Options None" to "Options FollowSymLinks" in the
<Directory /> section of the default access.conf-dist
(and -win even though it doesn't matter there). This has better
performance, and more intuitive semantics. [Dean Gaudet]
*) PORT: Updated support for UTS 2.1.2.
*) Fix symbol export list (src/support/httpd.exp) after recent
API changes in the child spawning area.
*) Workaround for configure script and old `test' commands which do not
support the -x flag (for instance under platforms like Ultrix). This is
solved by another helper script findprg.sh which searches for Perl and
Awk like PrintPath but _via different names_.
[Ralf S. Engelschall]
*) Remove the system() call from htpasswd.c, which eliminates a system
*) PORT: Fix compilation failures on NEXTSTEP.
*) PORT: F_NDELAY is a typo, should have been FNDELAY. There's also
O_NDELAY on various systems. [Dave Dykstra <dwd bell-labs.com>] PR#2313
*) PORT: helpers/GuessOS updates for various versions for NCR SVR4.
[juerg schreiner <j.schreiner zh.ch>,
Bill Houle <Bill.Houle SanDiegoCA.NCR.COM>] PR#2310
10268
10269
10270
10271
10272
10273
10274
10275
10276
10277
10278
10279
10280
10281
10282
10283
10284
*) Fix recently introduced Win32 child spawning code in mod_rewrite.c which
was broken because of invalid ap_pstrcat() -> strcat() transformation.
[Ralf S. Engelschall]
*) Proxy Cache Fixes: account for directory sizes, fork off garbage collection
to continue in background, use predefined types (off_t, size_t, time_t),
log the current cache usage percentage at LogLevel debug
[Martin Kraemer, based on discussion between Dean Gaudet & Dirk vanGulik]
Changes with Apache 1.3.0
*) Using a type map file as a custom error document was not possible.
[Lars Eilebrecht] PR#1031
*) Avoid problems with braindead Awks by additionally searching for gawk
and nawk in APACI's configure script.
10286
10287
10288
10289
10290
10291
10292
10293
10294
10295
10296
10297
10298
10299
10300
10301
10302
*) Rename md5.h to ap_md5.h to avoid conflicts with native MD5 on
some systems. [Randy Terbush]
*) Change usage of perror()+fprintf(stderr,...) in mod_rewrite to
more proper ap_log_error() variants.
[Ralf S. Engelschall]
*) Make sure the argument for the --add-module option to APACI's configure
script is of type [path/to/]mod_xxx.c because all calculations inside
configure and src/Configure depend on this.
[Ralf S. Engelschall] PR#2307
*) Changes usage of perror/fprintf to stderr to more proper ap_log_error
in mod_mime, mod_log_referer, mod_log_agent, and mod_log_config.
[Brian Behlendorf]
*) Various OS/2 cleanups ["Brian Havard" <brianh kheldar.apana.org.au>]
*) PORT: QNX needed a #include <sys/mman.h>; and now it uses flock
serialized accept to handle multiple sockets.
*) Have NT properly set the directory for CGI scripts
(& other spawned children)
*) Propagate environment to CGI scripts correctly in Win32.
10315
10316
10317
10318
10319
10320
10321
10322
10323
10324
10325
10326
10327
10328
10329
10330
10331
10332
10333
10334
10335
10336
10337
10338
10339
10340
10341
10342
10343
10344
10345
10346
10347
10348
10349
10350
10351
10352
10353
10354
10355
10356
10357
10358
10359
10360
10361
10362
*) Some symbol renaming:
ap_spawn_child_err became ap_spawn_child
ap_spawn_child_err_buff became ap_bspawn_child
spawn_child was obsoleted and moved to compat.h
[Brian Behlendorf]
*) Upgrade the child spawning code in mod_rewrite for the RewriteMap
programs: ap_spawn_child_err() is used and the Win32 case now uses
CreateProcess() instead of a low-level execl() (which caused problems in
the past under Win32).
[Ralf S. Engelschall]
*) A few cosmetics and trivial enhancements to APXS to make the
generated Makefile more user friendly. [Ralf S. Engelschall]
*) Proxy Fix: The proxy special failure routine ap_proxyerror()
was updated to use the normal apache error processing, thereby allowing
proxy errors to be treated by ErrorDocument's as well. For this
purpose, a new module-to-core communication variable "error-notes"
was introduced; the proxy (and possibly other modules) communicates
its error text using this variable. Its content is copied to a new
cgi-env-var REDIRECT_ERROR_NOTES for use by ErrorDocuments.
The old proxy special error routine ap_proxy_log_uerror()
was replaced by regular ap_log_error() calls, many messages were made
more informative.
[Martin Kraemer] PR#494, 1259
*) SECURITY: A possible buffer overflow in the ftp proxy was fixed.
[Martin Kraemer]
*) Transform the configure message "You need root privileges for suEXEC"
from a fatal error into a (more friendly) warning because the building
("make") of Apache we can allow, of course. Root privileges are needed
only for the installation step ("make install"). So make sure the
user is aware of this fact but let him proceed as long as he can.
[Ralf S. Engelschall] PR#2288
*) Renamed three more functions to common ap_ prefix which we missed at the
Big Symbol Renaming because they're #defines and not real C functions:
is_default_port(), default_port(), http_method().
[Ralf S. Engelschall]
*) A zero-length name after a $ in an SSI document should cause
just the $ to be in the expansion. This was broken during the
security fixes in 1.2.5. [Dean Gaudet] PR#1921, 2249
*) Call ap_destroy_sub_req() in ap_add_cgi_vars() to reclaim some
10364
10365
10366
10367
10368
10369
10370
10371
10372
10373
10374
10375
10376
10377
10378
10379
10380
10381
10382
10383
10384
*) Fix src/support/httpd.exp (DSO export file which is currently only
used under AIX) because of recent changes to function names.
[Ralf S. Engelschall]
Changes with Apache 1.3b7
*) Make sure a MIME-type can be forced via a RewriteRule even when no
substitution takes place, for instance via the following rule:
``RewriteRule ^myscript$ - [T=application/x-httpd-cgi]'' This was often
requested by users in the past to force a single script without a .cgi
extension and outside any cgi-bin dirs to be executed as a CGI program.
[Ralf S. Engelschall] PR#2254
*) A fix for protocol issues surrounding 400, 408, and
414 responses. [Ed Korthof]
*) Ignore MaxRequestsPerChild on WIN32. [Brian Behlendorf]
*) Fix discrepancy in proxy_ftp.c which was causing failures when
trying to connect to certain ftpd's, such as anonftpd.
*) Make mod_rewrite use ap_open_piped_log() for RewriteLog directive's
logfile instead of fiddling around itself with child spawning stuff.
[Ralf S. Engelschall]
*) Made RefererIgnore case-insensitive.
*) Mod_log_agent, mod_log_referer now use ap_open_piped_log for piped logs.
[Brian Behlendorf]
*) Replace use of spawn_child with ap_spawn_child_err_buff, to make everything
"safe" under Win32. In: mod_include.c, mod_mime_magic.c
[Brian Behlendorf]
*) Improve RFC1413 support. [Bob Beck <beck bofh.ucs.ualberta.ca>]
*) Fix support script `dbmmanage': It was unable to handle some sort
of passwords, especially passwords with "0" chars.
[Ralf S. Engelschall] PR#2242
*) WIN32: Clicking on "Last Modified" in a fancy index caused a crash. Fixed.
[Ben Laurie] PR#2238
*) WIN32: CGIs could cause a hang (because of a deadlock in the standard C
library), so CGI handling has been changed to use Win32 native handles
instead of C file descriptors.
[Ben Laurie and Bill Stoddard <wgstodda us.ibm.com>] PR#1129, 1607
*) The proxy cache would store an incorrect content-length in the cached
file copy after a cache update. That resulted in repeated fetching
of the original copy instead of using the cached copy.
10418
10419
10420
10421
10422
10423
10424
10425
10426
10427
10428
10429
10430
10431
10432
10433
10434
*) The Makefiles assumed that DSO files are build via $(LD). This
is broken for two reasons: First we never defined at least LD=ld
somewhere to make sure this works (it was silently assumed that most Make
provide a built-in LD definition - ARGL!) and second using the generic LD
variable is not the truth. Instead a special variable named LD_SHLIB is
reasonable because although "ld" is usually the default, the command for
building DSO files can be "libtool" or even "cc" on some systems.
[Ralf S. Engelschall]
*) Replace the AddVersionPlatform directive with ServerTokens which
provides for more control over the format of the Server:
header line. SERVER_SUBVERSION is no longer supported;
all module should use the ap_add_version_component()
API function instead. [Jim Jagielski]
*) Support for the NCR MP/RAS 3.0
10436
10437
10438
10439
10440
10441
10442
10443
10444
10445
10446
10447
10448
10449
10450
10451
10452
10453
10454
10455
10456
10457
10458
10459
10460
10461
10462
10463
10464
10465
10466
10467
10468
10469
10470
10471
10472
10473
10474
10475
*) The LDFLAGS_SHLIB_EXPORT variable of src/Configuration[.tmpl] was
not retrieved in src/Configure and thus was not useable.
[Ralf S. Engelschall]
*) Various Makefile consistency cleanups:
- make OSDIR also automatically be relative to src/ like INCDIR
- SUBDIRS is now generated in src/Makefile only and not in
Makefile.config because it is a local define for this location.
- remove BROKEN_BPRINTF_FLAGS because is it no longer used inside
any Makefile but make sure that at least the "-K inline" is kept in
CFLAGS for SCO 5.
- update the "depend" targets in Makefile.tmpl files to use $(OSDIR), too.
- updated the dependencies theirself
- removed not existing SHLIB variable from "clean" targets
- replaced SHLIB_OBJS/SHLIBS_OBJ consistently with OBJS_PIC because OBJS
already exists and OBJS_PIC are also just plain objects and have not
directly to do with "shared" things. The only difference is that they
contain PIC. So OBJS_PIC is the more canonical name.
- Updated the Makefile-dependency lines for OBJS_PIC
- Removed the Makefile-dependency line in Configure to avoid double
definitions
- replaced ugly xx-so.o/xx.so-o hack with a clean and consistent usage
of xxx.lo as GNU libtool does with its PIC objects
- reduce local complexity in modules Makefile.tmpl by moving the last
existing target "depend" to the generation section in Configure, too.
- removed the historical $(SPACER) which was used in the past together
with BROKEN_BPRINTF_FLAGS to avoid zig-zags in the build process. This
is no longer needed.
- force the build and run of the gen_xxx programs under main/ as the
first step before building the objects because it looks cleaner
[Ralf S. Engelschall]
*) WIN32: Make Win32 work again after the /dev/null DoS fix.
[Ben Laurie]
*) WIN32: Check for buffer overflows in ap_os_canonical_filename.
[Ben Laurie]
*) WIN32: Don't force ISAPI headers to finish with \n.
10477
10478
10479
10480
10481
10482
10483
10484
10485
10486
10487
10488
10489
10490
10491
10492
10493
10494
10495
10496
10497
10498
10499
10500
10501
10502
10503
10504
10505
10506
10507
10508
10509
10510
10511
10512
10513
10514
10515
10516
10517
10518
10519
10520
10521
10522
10523
10524
10525
10526
10527
10528
10529
10530
10531
10532
10533
10534
10535
10536
10537
10538
10539
10540
10541
10542
10543
10544
10545
10546
10547
10548
10549
10550
10551
10552
10553
10554
10555
10556
10557
10558
10559
10560
10561
10562
10563
10564
10565
10566
10567
10568
10569
10570
10571
10572
10573
10574
10575
10576
10577
*) When opening "configuration" files (like httpd.conf, htaccess
and htpasswd), Apache will not allow them to be non-/dev/null
device files. This closes a DoS hole. At the same time,
we use ap_pfopen to open these files to handle timeouts.
[Jim Jagielski, Martin Kraemer]
*) Apache will now log the reason its httpd children exit if they exit
due to an unexpected signal. (It requires a new porting define,
SYS_SIGLIST, which if defined should point to a list of text
descriptions of the signals available. See PORTING.) [Dean Gaudet]
*) WIN32: chdir() doesn't make sense in a multithreaded environment
like WIN32. Before, Win32 CGI's could have had sporadic failures
if a chdir call from one thread was made between another chdir call
and a spawn in another thread. So, for now don't chdir for CGI scripts
in WIN32. The current CGI "spec" is unclear as to whether it's
necessary. Long-term fix is to either serialize the chdir/spawn combo
or use WIN32 native calls to spawn a process. This temp fix was
necessary to remove this as a showstopper for 1.3's release.
[Brian Behlendorf]
*) Cleanup the suEXEC support in APACI and make it more safe:
1. Add big fat hint in INSTALL about risks and to read the
htdocs/manual/suexec.html document before using the suexec-related
configure options.
2. Make sure the user has at least provided one --suexec-xxxx option
(specifies suEXEC parameters) in addition to --enable-suexec option.
If only --enable-suexec is given APACI stops with a hint to INSTALL
and htdocs/manual/suexec.html documents.
3. Provide two additional --suexec-xxxx options to make the suEXEC
configuration complete (especially for package maintainers who else
had to patch the source tree) by providing ways to configure minimal
UID/GID and safe PATH, too.
[Ralf S. Engelschall]
*) Cleanup of the `configure --shadow' process:
- make sure the configure script creates its temporary files in the
shadow tree to avoid conflicts with parallel configure runs
- removed unnecessary option "-r" from "rm" call for Makefiles
- make sure the configure scripts creates the shadow-wrapper Makefile
only when no shadow trees already exists
- make sure "make distclean" removes the shadow-wrapper Makefile but only
when no more shadow trees exists
- overhauled mkshadow.sh script: now its more IFS-safe and approx. twice
as fast (in the past it needed 70sec, now it runs just 38sec)
- make sure CVS does not complain about the created files
Makefille.<gnutriple> and directories src.<gnutriple>
[Ralf S. Engelschall]
*) Added the ap_add_version_component() API routine and the
AddVersionPlatform core directive. The first allows modules to
declare themselves in the Server response header field value,
augmenting the SERVER_SUBVERSION define in the Configuration file
with run-time settings (more useful in a loadable-module environment).
AddVersionPlatform inserts a comment such as "(UNIX)" or "(Win32)"
into the server version string. [Ken Coar] PR#2056
*) Minor stability tweaks to avoid core dumps in ap_snprintf.
[Martin Kraemer]
*) Emit the "Accept-Range" header for the default handler.
[Brian Behlendorf] PR#1464
*) Add a note to httpd.conf-dist that apache will on some systems fail
to start when the Group # is set to a negative or large positive value.
[Martin Kraemer]
*) Make sure the module execution order is correct even when some modules
are loaded under runtime (`LoadModule') via the DSO mechanism:
1. The list of loaded modules is now a dynamically allocated one
and not the original statically list from modules.c
2. The loaded modules are now correctly setup by LoadModule for
later use by the AddModule command.
3. When the DSO mechanism for modules is used APACI's `install'
target now enables all created `LoadModule' lines per default because
this is both already expected by the user _and_ needed to avoid
confusion with the next point and reduces the Makefile.tmpl complexity
4. When the DSO mechanism for modules is used, APACI's `install'
target now additionally makes sure the module list is reconstructed
via a complete `ClearModuleList+AddModule...' entry.
5. The support tool `apxs' now also makes sure an AddModule command
is added in addition to the LoadModule command.
6. The modules.c generation was extended to now contain two
comments to make sure no one is confused by the confusing terminology
of loading/linking (we use load=link+load & link=activate instead of
the obvious load=activate & link=link :-( )
This way now there is no longer a difference under execution time between
statically and dynamically linked modules.
[Ralf S. Engelschall]
*) Fix the generated mod_xxx.c from "apxs -g -f xxx" after the
Big Symbol Renaming. [Ralf S. Engelschall]
*) Add a comment to mod_example.c showing the format of a FLAG command
handler. [Ken Coar]
*) Standardized the time format in mod_status to match that of other
places in the code (e.g. DATE_GMT). PR#1551
*) Fix handling of %Z in timefmt strings for those platforms with no time
zone information in their tm struct. [Paul Eggert <eggert twinsun.com>]
10579
10580
10581
10582
10583
10584
10585
10586
10587
10588
10589
10590
10591
10592
10593
10594
10595
10596
10597
10598
10599
10600
10601
10602
10603
10604
10605
PR#754
*) Makes mod_rewrite, mod_log_config, mod_status and the ServerSignature
feature compatible with 'UseCanonicalName off' by changing
r->server->server_hostname to ap_get_server_name(). And I changed some
functions which use r->server->port to use ap_get_server_port() instead,
because if there's no Port directive in the config r->server->port is 0.
[Lars Eilebrecht]
*) get/set_module_config are trivial enough to be better off inline. Worth
1.5% performance boost. [Dean Gaudet]
*) Fix off-by-one error in ap_proxy_date_canon() in proxy_util.c
when ensuring 'x' is at least 30-chars big. [Jim Jagielski,
Brian Behlendorf]
*) [BS2000 security] BS2000 needs an extra authentication to initialize
the task environment to the unprivileged User id. Otherwise CGI scripts
would have a way to gain super user access. [Martin Kraemer]
*) Fix debug log messages for BS2000/OSD: instead of logging the whole
absolute path, only log base name of logging source as is done
in unix. [Martin Kraemer]
*) Ronald Tschalaer's Accept-Encoding patch - preserve the "x-" in
the encoding type from the Accept-Encoding header (if it's there)
and use it in the response, as that's probably what it'll be expecting.
*) Fix to mod_alias: translate_alias_redir is dealing with
a URI, not a filename, so the check for drive letters for win32
and emx is not necessary. [Dean Gaudet]
*) WIN32: Allow .cmd as an executable extension.
*) Make Apache header files, and some variables, C++ friendly.
10617
10618
10619
10620
10621
10622
10623
10624
10625
10626
10627
10628
10629
10630
10631
10632
10633
10634
10635
10636
10637
10638
10639
10640
10641
10642
*) Child processes can now "signal" (by exiting with a status
of APEXIT_CHILDFATAL) the parent process to abort and
shutdown the server if the error in the child process was
fatal enough. [Jim Jagielski]
*) mod_autoindex's find_itme() was sensitive to MIME type case.
[Jim Jagielski] PR#2112
*) Make sure the referer_log and agent_log entries in the default httpd.conf
file are also adjusted for the actual relative installation paths.
[Ralf S. Engelschall] PR#2175
*) WIN32: Extensive overhaul of the way UNCs are handled. [Ben Laurie]
*) WIN32: Make roots of filesystems (e.g. c:/) work. [Ben Laurie]
PR#1558
*) PORT: Various porting changes to support AIX 3.2, 4.1.5, 4.2 and 4.3.
Additionally the checks for finding the vendor DSO library were moved
from mod_so.c to Configure because first it needs $PLAT etc. and second
mod_so already uses an abstraction layer and does not fiddle with the
vendor functions itself.
[Jens-Uwe Mager, Ralf S. Engelschall]
*) PORT: Some optimization defines for NetBSD
*) PORT: Dynamic Shared Object (DSO) support for NetBSD.
[Jaromir Dolecek <dolecek ics.muni.cz>, Ralf S. Engelschall] PR#2158
10647
10648
10649
10650
10651
10652
10653
10654
10655
10656
10657
10658
10659
10660
10661
10662
10663
10664
10665
10666
*) Add Dynamic Shared Object (DSO) support for AIX (at least 4.2 but older
AIX variants should work fine, too. Even AIX 3.x should work). This is
accomplished by using the free DSO emulation code from Jens-Uwe Mager
which we put into a os/unix/os-dso-aix.c file.
[Ralf S. Engelschall]
*) PORT: Fix compiler warnings under AIX >= 4.2 where the manual pages imply
that we should use NET_SIZE_T == int but the include files force size_t.
[Ralf S. Engelschall]
*) Fix two bugs in select() handling in http_main.c.
[Roy Fielding]
*) Suppress "error(0)" messages for ap_log_error() when the APLOG_NOERRNO
is unset (as it is in situations like timeouts) where it is unclear
whether errno is set or not. [Martin Kraemer]
*) Just having APACI's localstatedir is too general and not enough for most
of the systems. 1.3b6 again required manual APACI patches by package
maintainers from Red Hat and FreeBSD because for their filesystem layout a
little bit more flexibility in configuring the paths is needed. Hence we
provide three additional configure options (--runtimedir, --logfiledir,
--proxycachedir) which now can be used for more granular adjustments if
--localstatedir is not enough to fit the particular needs. As a nice
side-effect this reduces some subdir fiddling in configure+Makefile.tmpl.
[Ralf S. Engelschall]
*) Make the install root for "make install" in APACI's Makefile overrideable
by package authors. This way we are even more friendly to package
maintainers (especially Debian and Red Hat) who build for the real prefix
10678
10679
10680
10681
10682
10683
10684
10685
10686
10687
10688
10689
10690
10691
10692
10693
10694
10695
10696
10697
10698
10699
10700
10701
10702
10703
10704
10705
10706
10707
10708
10709
10710
10711
10712
10713
10714
10715
10716
10717
10718
10719
10720
10721
10722
10723
10724
10725
10726
10727
10728
10729
10730
10731
10732
10733
10734
10735
10736
10737
10738
10739
10740
10741
10742
10743
10744
10745
10746
via "configure --prefix=/<real>" but use a different local prefix via
"make root=/tmp/apache install" for rolling the package without bristling
the target location on their system.
[Ralf S. Engelschall]
*) Workaround sed limitations in APACI's configure script by now
substituting in chunks of 50 commands (because for instance HPUX's vendor
sed has a limit of max. 98 commands)
[Ralf S. Engelschall] PR#2136
*) Adding SOCKS5 support and fixing existing SOCKS4 support.
[Ralf S. Engelschall] PR#2140
*) Manually fix some symbols which were not renamed to prefix ap_ in the BIG
RENAMING process because they are defined as pre-processor macros instead
of real functions: bputc, bgetc, piped_log_write_fd, piped_log_read_fd
[Ralf S. Engelschall]
*) Workaround braindead AWK's when generating ap_config.h: The split() and
substr() functions cannot be nested under vendor AWK from Solaris 2.6.
[Ralf S. Engelschall] PR#2139
*) Various bugfixes and cleanups for the APACI configure script:
o fix IFS handling for _nested_ situation
o fix Perl interpreter search: take first one found instead of last one
o fix DSO consistency check
o print error messages to stderr instead of stdout
o add install-quiet for --shadow situation to Makefile stub
o reduce complexity by avoiding sed-hacks for rule and module list loops
[Ralf S. Engelschall]
*) Fix DEBUG_CGI situation in mod_cgi.c [David MacKenzie] PR#2114
*) Make sure the input field separator (IFS) shell variable is explicitly
initialized correctly before _every_ `for' loop and also restored after
the loops. [Ralf S. Engelschall]
*) Make sure that "make install" doesn't overwrite the `mime.types' and
`magic' files from an existing Apache installation. Because people often
customize these for own MIME and content types.
[Ralf S. Engelschall]
*) PORT: Dynamic Shared Object (DSO) support for OpenBSD 2.x
[Peter Galbavy, Ralf S. Engelschall] PR#2109
*) Fix the path to the ScoreBoardFile in the install-config target, too.
[Ralf S. Engelschall] PR#2105
*) Let "configure" clear out the users parameters (provided as shell
variables) to avoid side-effects in "src/Configure" when the user
exported them (which is not needed, but some users do it).
[Ralf S. Engelschall] PR#2101
*) Provide backward compatibility from some old src/Configuration.tmpl
parameter names to the canonical Autoconf-style shell variable names. For
instance CFLAGS vs. EXTRA_CFLAGS. The EXTRA_xxx variants are accepted now
but a hint message is displayed. [Ralf S. Engelschall]
*) Make sure that "make install" doesn't overwrite the DocumentRoot and
CGI scripts from an existing Apache installation.
[Ralf S. Engelschall, Jim Jagielski] PR#2084
*) Make `configure --compat' more "compatible" by first
let the libexecdir default to EPREFIX/libexec instead of EPREFIX/bin and
second by making sure the "avoid-bristling-suffix" /apache is not
appended to sysconfdir, datadir, localstatedir and includedir when
--compat is used. [Ralf S. Engelschall, Lars Eilebrecht]
*) NeXT required strdup() in support/logresolve.c
*) AIX required sys/select.h in support/ab.c
*) Fix the path to the MimeMagicFile in the install-config target, too.
[Ralf S. Engelschall] PR#2089
*) PORT: Added HP-UX 11 patches [Jeff Earickson <jaearick colby.edu>]
10756
10757
10758
10759
10760
10761
10762
10763
10764
10765
10766
10767
10768
10769
10770
10771
10772
10773
10774
10775
10776
10777
10778
10779
10780
10781
10782
10783
10784
10785
10786
10787
10788
10789
10790
10791
10792
10793
10794
10795
10796
10797
10798
10799
10800
10801
10802
10803
10804
10805
10806
10807
10808
10809
10810
10811
10812
10813
10814
10815
10816
10817
10818
10819
10820
10821
10822
10823
10824
10825
10826
10827
10828
10829
10830
10831
10832
10833
10834
10835
10836
10837
10838
10839
10840
10841
10842
10843
10844
10845
10846
10847
10848
10849
10850
10851
10852
*) If you start apache with the -S command line option it will dump
out the parsed vhost settings. This is useful for folks trying
to figure out what is wrong with their vhost configuration.
(Other dumps may be added in the future.) [Dean Gaudet]
*) Add %pA, %pI, and %pp codes to ap_vformatter (and hence ap_bprintf,
ap_snprintf, and ap_psprintf). See include/ap.h for docs.
[Dean Gaudet]
*) Because /usr/local/apache is the default prefix the ``configure
--compat'' option no longer has to set prefix, again. This way the
--compat option honors a leading --prefix option. [Lars Eilebrecht]
*) PORT: Cast the first argument of dlopen() in ap_os_dso_load()
to `char *' under OSF1 and FreeBSD 2.x where it is defined this way
to avoid "discard const" warnings. [Ralf S. Engelschall]
*) If a specific handler is set for a file yet the request still
ends up being handled by the default handler, log an error
message before handling it. This catches things such as trying
to use SSIs without mod_include enabled. [Marc Slemko]
*) Fix error logging for the startup case where ap_log_error() still uses
stderr as the target. Now the default log level is honored here, too.
[Ralf S. Engelschall]
*) PORT: Make sure some AWK's don't fail in src/Configure with "string too
long" errors when generating the MODULES entry for src/Makefile
[Ben Hyde, Ralf S. Engelschall]
*) Make sure src/Configure doesn't complain about the old directory
/usr/local/etc/httpd/ when APACI is used. [Lars Eilebrecht]
Changes with Apache 1.3b6
*) PORT: Clean up warnings on Ultrix and HPUX. [Ben Hyde]
*) Adding DSO support for the HP/UX platform by emulating the dlopen-style
interface via the similar but proprietary HP/UX shl_xxx-style system
calls. [Ralf S. Engelschall]
*) PORT: Updated UnixWare 2.0.x and 2.1.x entries for DSO support and made
APACI Makefile.tmpl "install" target more robust for sensible UnixWare
Make. [Ralf S. Engelschall]
*) ++++ THE BIG SYMBOL RENAMING ++++
To avoid symbol clashes with third-party code compiled into the server,
we globally applied the prefix "ap_" to the following classes of
functions:
- Apache provided general functions (e.g., ap_cpystrn)
- Public API functions (e.g., palloc, bgets)
- Private functions which we can't make static (because of
cross-object usage) but should be (e.g., new_connection)
For backward source compatibility a new header file named compat.h was
created which provides defines for the old symbol names and can be used
by third-party module authors.
[The Apache Group]
*) Added dynamic shared object (DSO) support for SVR4-derivates: The
problem under SVR4 is that there is no command flag to force the linker
to export the global symbols of the httpd executable therewith they are
available to the DSO's. Instead of problematic hacks like creating a
dummy.so file (containing dummy references to all global symbols) the
httpd binary is linked against, we use a clean trick stolen from Perl 5:
Placing the Apache core code itself into a DSO library named libhttpd.so.
This way the global symbols _HAVE_ to be exported and thus are available
to any manually loaded DSO's under runtime. To reduce the impact to the
user to null we go even further and create a stub httpd executable which
automatically keeps track of the DSO library loading itself and thus
hides the complete mechanism from the user. Although the generation of
this DSO library is automatically triggered for platforms which
essentially need it (mostly all SVR4-derivates) it can be also enabled
manually via the Rule SHARED_CORE. This can be interesting in the future
where we perhaps exploit this libhttpd.so mechanism for providing nifty
features like graceful upgrades, or whatever.
[Ralf S. Engelschall, Martin Kraemer]
*) Build the libraries before building the rest of the tools. [Ben Hyde]
*) Add "distclean" target to src/-Makefiles to provide "make distclean" also
inside the src subtree (i.e. for non-APACI users). Following GNU Makefile
conventions while "clean" removes only stuff created by "all" targets,
"distclean" additionally removes the stuff from the configuration
process. This way "make distclean" (hence the name) provides a fresh
source tree as it was for distribution.
[Ralf S. Engelschall]
*) Allow top-level (APACI) Makefile to break on build errors
the same way the src/ subtree Makefiles breaks on them by replacing the
initial APACI sed-subdir-display-kludge with a more clean
variable-passing-solution: variable SDP can optionally hold the subdir
prefix which is consistently used for displaying the subdir movement.
This way even the top-level Makefile can stop correctly on errors as the
user expects. [Ralf S. Engelschall]
*) Fixed ordering of argument checks for RewriteBase directive.
10854
10855
10856
10857
10858
10859
10860
10861
10862
10863
10864
10865
10866
10867
10868
10869
10870
10871
10872
10873
10874
10875
10876
10877
10878
10879
10880
10881
10882
10883
10884
10885
10886
10887
10888
10889
10890
10891
10892
10893
10894
10895
10896
10897
10898
10899
10900
10901
10902
10903
10904
10905
10906
10907
*) Change Win32 IS_MODULE to SHARED_MODULE to match Unix' method of
indicating that a module is being compiled for dynamic loading. Also
remove #define IS_MODULE from modules and add SHARED_MODULE define
to the mak/dsp files. [Alexei Kosut]
*) Reduce logging level of "normal" warning messages to APLOG_INFO,
since we are now logging APLOG_WARNING by default. [Roy Fielding]
*) PORT: OS/2 tweak to deal with multiple .exe targets. [Brian Havard]
*) Add documentation file and src/Configuration.tmpl entry for the
experimental mod_mmap_static module. Because although it is and marked as
an experimental one it is distributed and thus should be documented and
prepared for configuration the same way as all others modules.
[Ralf S. Engelschall]
*) Add query (-q) option to apxs support tool to be able to manually query
specific settings from apxs. This is needed for instance when you
manually want to access Apache's header files and you need to assemble
the -I option. Now you can do -I`apxs -q INCLUDEDIR`.
[Ralf S. Engelschall]
*) Now src/Configure uses a fallback strategy for the shared object support
on platforms where no explicit information is available: If a Perl
installation exists we ask it about its shared object support and if it's
the dlopen-style one we shamelessly guess the compiler and linker flags
for creating shared objects from Perls knowledge. Of course, the user is
warning about what we are doing and informed that he should send us
the guessed flags when they work. [Ralf S. Engelschall]
*) Provide APACI --without-support option to be able to disable the build
and installation of the support tools from the src/support/ area.
Although its useful to have these installed per default we should provide
a way to compile and install without them for backward-compatibility.
[Ralf S. Engelschall]
*) Add of the new APache eXtenSion (apxs) support tool for building and
installing modules into an _already installed_ Apache package through the
dynamic shared object (DSO) mechanism [mod_so.c]. The trick here is that
this approach actually doesn't need the Apache source tree. The
(APACI-installed) server package is enough, because this now includes the
Apache C header files (PREFIX/include) and the new APXS tool
(SBINDIR/apxs). The intend is to provide a handy tool for third-party
module authors to build their Apache modules _OUTSIDE_ the Apache source
tree while avoiding them to fiddle around with the totally platform
dependend way of compiling DSO files. The tool supports all ranges of
modules, from trivial ones (single mod_foo.c) to complex ones (like PHP3
which has a mod_php3.c plus a pre-built libmodphp3-so.a) and even can
on-the-fly generate a minimalistic Makefile and sample module for the
first step to provide both a quick success event and to demonstrate the
APXS mechanism to module authors. [Ralf S. Engelschall]
*) Fix core dumps in use of CONNECT in proxy.
10909
10910
10911
10912
10913
10914
10915
10916
10917
10918
10919
10920
10921
10922
10923
10924
10925
10926
10927
10928
10929
10930
10931
10932
10933
10934
10935
10936
10937
10938
10939
10940
10941
10942
10943
10944
10945
10946
10947
10948
10949
10950
10951
10952
10953
10954
10955
*) Modify the log directives in httpd.conf-dist files to use CustomLog
so that users have examples of how CustomLog can be used.
[Lars Eilebrecht]
*) Add the new Apache Autoconf-style Interface (APACI) for the top-level of
the Apache distribution tree. Until Apache 1.3 there was no real
out-of-the-box batch-capable build and installation procedure for the
complete Apache package. This is now provided by a top-level "configure"
script and a corresponding top-level "Makefile.tmpl" file. The goal is
to provide a GNU Autoconf-style frontend which is capable to both drive
the old src/Configure stuff in batch and additionally installs the
package with a GNU-conforming directory layout. Any options from the old
configuration scheme are available plus a lot of new options for flexibly
customizing Apache. [Ralf S. Engelschall]
*) The floating point ap_snprintf code wasn't threadsafe.
Had to remove the HAVE_CVT macro in order to do threadsafe
calling of the ?cvt() floating point routines. [Dean Gaudet]
*) PORT: Add the SCO_SV port. [Jim Jagielski] PR#1962
*) PORT: IRIX needs the -n32 flag iff using the 'cc' compiler
[Jim Jagielski] PR#1901
*) BUG: Configure was using TCC and CC inconsistently. Make sure
Configure knows which CC we are using. [Jim Jagielski]
*) "Options +Includes" wasn't correctly merged if "+IncludesNoExec"
was defined in a parent directory. [Lars Eilebrecht]
*) API: ap_snprintf() code mutated into ap_vformatter(), which is
a generic printf-style routine that can call arbitrary output
routines. Use this to replace http_bprintf.c. Add new routines
psprintf(), pvsprintf() which allocate the exact amount of memory
required for a string from a pool. Use psprintf() to clean up
various bits of code which used ap_snprintf()/pstrdup().
[Dean Gaudet]
*) PORT: HAVE_SNPRINTF doesn't do anything any longer. This is because
ap_snprintf() has different semantics and formatting codes than
snprintf(). [Dean Gaudet]
*) SIGXCPU and SIGXFSZ are now reset to SIG_DFL at boot-time. This
is necessary on at least Solaris where the /etc/rc?.d scripts
are run with these signals ignored, and "SIG_IGN" settings are
maintained across exec().
*) Fix the check for symbolic links in ``RewriteCond ... -l'': stat() was
used instead of lstat() and thus this flag didn't work as expected.
10961
10962
10963
10964
10965
10966
10967
10968
10969
10970
10971
10972
10973
10974
10975
10976
10977
10978
10979
10980
*) Fix the proxy pass-through feature of mod_rewrite for the case of
existing QUERY_STRING now that mod_proxy was recently changed because of
the new URL parsing stuff. [Ralf S. Engelschall]
*) A few changes to scoreboard definitions which helps gcc generate
better code. [Dean Gaudet]
*) ANSI C doesn't guarantee that "int foo : 2" in a structure will
be a signed bitfield. So mark a few bitfields as signed to
ensure correct code. [Dean Gaudet]
*) The default for HostnameLookups was changed to Off, but there
was a problem and it wasn't taking effect. [Dean Gaudet]
*) PORT: Clean up undefined signals on some platforms (SCO, BeOS).
[Dean Gaudet]
*) After a SIGHUP the listening sockets in the parent weren't
properly marked for closure on fork().
[Jürgen Keil <jk tools.de>] PR#2000
10982
10983
10984
10985
10986
10987
10988
10989
10990
10991
10992
10993
10994
10995
10996
10997
10998
*) Allow %2F in two situations: 1) it is in the query part of the URI,
therefore not exposed to %2F -> '/' translations and 2) the request
is a proxy request, so we're not dealing with a local resource anyway.
Without this, the proxy would fail to work for any URL's with
%2f in them (occurs quite often in
http://.../cgi-bin/...?http%3A%2F%2F... references) [Martin Kraemer]
*) Protect against FD_SETSIZE mismatches. [Dean Gaudet]
*) Make the shared object compilation command more portable by avoiding
the direct combination of `-c' & `-o' which is not honored by some
compilers like UnixWare's cc. [Ralf S. Engelschall]
*) WIN32: the proxy was creating filenames missing the last four
characters. While this normally doesn't stop anything from
working, it can result in extra collisions.