all PEAR packages to php?-pear-* and all Apache packages to ap13-* or
ap2-* respectively. Add new variables to simplify the Makefile
handling. Add CONFLICTS on the old names. Reset revisions of bumped
packages. ap-php will now depend on the default Apache and PHP version.
All programs using it have an implicit option of the Apache version
as well.
OK from jlam@ and adrianp@.
rest through a Makefile.common entity. It will help with incoming
ap2-fastcgi addition.
Changes: lots of!
2.4.2
*) [WIN] Fix handle leaks in the process manager.
Sascha Schumann [sascha@schumann.cx]
*) [WIN] Use a permanent pool for allocating the SystemRoot
environment variable. Sakamoto [ringring@zb4.so-net.ne.jp]
*) [WIN] Fix starting of scripts under AP2.
*) Do the connect() to the application *after* collecting a chunk
of client data. This reinstates the pre-2.4.0 behaviour.
James Jurach [muaddib@erf.net].
*) Remove an assert that was triggering on WIN when spawn() failed.
*) Provide the NO_SUEXEC_FOR_AP_USER_N_GROUP macro for building
mod_fastcgi with the AP13 suexec behaviour (don't use suexec
if httpd's user and group match that needed for the application).
*) Prevent the use of all but the "auth" directives from being used
anywhere but in global scope. Prevent more than one instance
of the FastCgiWrapper directive.
*) Return NOT_FOUND (404) or FORBIDDEN (403) instead of
INTERNAL_SERVER_ERROR (500) when there are configuration issues
or the script does't exist. Suggested by Jeff Lawson [bovine@ud.com].
2.4.0
*) [*nix Security] - When FastCgiWrapper (FastCgiSuexec) was in use
and a vhost configured to use the same uid/gid as the main
server, mod_fastcgi would not bother using the wrapper (suexec)
because its effective uid/gid was already appropriate. This is
consistent with Apache's v1.3 mod_cgi behaviour. There are two
problems with this approach: 1) when FastCgWrapper is in use
mod_fastcgi's process manager keeps its root privileges (as its
real uid/gid) so it can terminate the applications its starts -
this privilege was being passed to applications when the use of
the wrapper was bypassed 2) wrappers are often employed to
perform functionality beyond setting the uid/gid - by not calling
the wrapper under certain circumstances, application invocation
environments were inconsistent. With this change, the wrapper is
always used (when enabled) under both Apache 1.3 and 2.
Reported by Michael Richards [michael@fastmail.ca].
*) [*nix/AP2] Use the vhost uid/gid instead of the server uid/gid
for dynamic application invocation when the FastCgiWrapper is in use.
Reported by Michael Richards [michael@fastmail.ca]
*) [*nix] Fix handling of FastCgiWrapper when passed a real path,
i.e. other than "on" or "off". Michael Richards [michael@fastmail.ca]
*) Eliminate the logging of "incomplete headers (0 bytes) received from
server" when a client aborts.
*) [WIN32] Fix a delay in handling large POSTs to named pipe based
servers. Philip Gladstone [philip@okena.com]
*) [*nix/AP2] Prevent the module from being initalized twice at startup
(resulting in confusing error messages to the log).
*) Eliminate the need for SetHandler or AddHandler with static or
external applications.
*) Limit PM requests to start a dynamic application
to 5sec to prevent endless spinning (this is a drop-dead
limit that should only occur if the socket/named_pipe directory
is removed out from under a running server).
*) [*nix] Change the default socket directory from /tmp/fcgi to:
Apache - logs/fastcgi
Apache2 - RUNTIMEDIR/fastcgi
*) Add -user & -group args to FastCgiServer and FastCgiExternalServer
for use with wrappers (in lieu of finding the user/group associated
with a virtual host - under Apache2 this isn't accomodated).
*) [WIN32] Under Apache2, require v2.0.41 or later in order to pickup my
apr_proc_create() changes.
*) Log when invoking and restoring the restart backoff policy.
*) [WIN32] Prevent intermittent ReadFile() failures (properly initialize the
OVERLAPPED structure).
*) Eliminate need for dummy files for external servers under Apache2
*) Fix auth compatibility mode handling for access checker and authorizer
*) Fix HEAD request handling. Based on a patch by
Chris Lightfoot [chris@ex-parrot.com]
*) [*nix] When autoupdate is enabled touch the socket when restarting
the processes to prevent further requests.
Eckebrecht von Pappenheim [evp@heise.de]
*) Apache 2.0 support.
*) [WIN32] Don't read from a potentially closed named pipe.
Philip Gladstone [philip@okena.com]
*) Require the Apache version 1.3.6 or later to eliminate some signal
handling funk.
*) [WIN32] Use asyncronous io with named pipes instead of polled
nonblocking io. This should eliminate the last of the npipe issues.
*) Handle an application returning a complete and valid response without
having consumed all of the data sent to it.
*) Consume remaining client data (RESPONDERs only) if any.
*) Add support for backing off attempts to start applications that continuously
fail to start. Three new macros defined in mod_fastcgi.h control this
behaviour: MAX_FAILED_STARTS, RUNTIME_SUCCESS_INTERVAL, FAILED_STARTS_DELAY
*) [WIN32] Add (back) support for use of TerminateProcess() to accomodate
applications that do not (properly) support the shutdown event (this
feature was introduced in fcgi2 2.2.2 and improved in 2.2.4). The
new macro WIN32_SHUTDOWN_GRACEFUL_WAIT in mod_fastcgi.h conrols the
interval between signaling a proper shutdown and wacking the process(s)
with a TerminateProcess().
*) [WIN32] Don't set the OVERLAPPED_IO flag on NamedPipe listen HANDLEs -
setting it was just plain broken.
*) [WIN32] Fix the accept mutex - all applications were sharing one!?
*) Fix 'FastCgiConfig -autoUpdate'.
*) Fix 'FastCgiConfig -flush'.
*) Prevent silly maxProcesses and processSlack combinations.
Dmitry Dorofeev [dima@yasp.com]
*) Properly handle the killing of idle processes when one takes a long time
to exit once signaled down (or the config is funky).
Dmitry Dorofeev [dima@yasp.com]
2.2.12
*) Delay the logging of write errors to the pm to account for shutdown/restart.
*) (Win32) An assortment of fixes.
*) Fix some broken casts that were likely the cause of an assert.
*) Win32. Eliminate forward slashes from the named pipe path name.
Gerald Richter [richter@ecos.de]
*) SIGUSR2 is no longer blocked in the process manager and the fastcgi
applications it spawns. [ryans@amazon.com]
*) Added support for the -flush argument to FastCgiConfig.
Eric Sit [esit@alum.mit.edu]
*) Change the "which call to module_init() is this" check to a more
reliable approach. Doru Petrescu [pdoru@kappa.ro]
*) Close the old pipe file descriptor in apache main on USR1/HUP
(elimnates a small leak). James E. Jurach Jr. jjurach@fundsxpress.com
*) Fix a bug in fcgi_config_set_authoritative_slot(). Tetsuya Furukawa
[tetsuya@secom-sis.co.jp]
*) Eliminate the use of locks to assist in the clean shutdown of
applications. Instead, it is assumed that applications handle
termination signals properly (this is now embedded in the C
application lib).
*) Fix Win32 process termination. Proper operation requires the use of
an updated application lib (termination is now signalled with an
Event and handled by specialized thread).
*) Docs cleanup. Andrew Benham [adsb@bigfoot.com]
*) Added code so if the last instance of a dynamic application died without
provocation, then don't restart it if singleThreshold > 0 (i.e. if the
configuration allows the last instance to be killed, then allow it to die).
Andrew Benham [adsb@bigfoot.com]
*) Fix the loadFactor calculation used to determine when dyanmic
applications could be killed off due to low demand [adsb@bigfoot.com].
*) Fix a deadlock condition that could occur with Win32 named
pipes (dynamic).
*) Fix a potential deadlock condition when FastCGI application
sent responses while still reading the client request (POST data).
first component is now a package name+version/pattern, no more
executable/patchname/whatnot.
While there, introduce BUILD_USES_MSGFMT as shorthand to pull in
devel/gettext unless /usr/bin/msgfmt exists (i.e. on post-1.5 -current).
Patch by Alistair Crooks <agc@netbsd.org>
*) Allow absolute pathnames in the -socket argument.
*) Don't invoke suexec when the user/group for the fastcgi application
is the same as the apache main server. This is consistent with
apache's suexec handling.
*) Reset the apache drop dead timer upon successful read or writes
to/from the client. This eliminates timeouts that were occuring
during the large file transfers to/from slow clients.
*) Support generic wrappers such as cgiwrap by eliminating dependencies
on Apache's SUEXEC, renaming the FastCgiSuexec directive
FastCgiWrapper and eliminating any checks regarding the target
application (this is the repsonibility of the wrapper).
*) Fix a nasty bug that occurred when a client aborted a POST request
before the connection to a dynamic FastCGI application was opened.
Changes with mod_fastcgi 2.2.8
*) Eliminate the concept of disabled applications. If a failure
occurs trying to setup an application (e.g. bind() error) its
tried repeatedly every init-start-delay seconds.
*) Tweak to Makefile.tmpl to support DSOs. Dave Hill [ddhill@zk3.dec.com]
Changes with mod_fastcgi 2.2.6
*) Shutdown the PM when Apache appears to have disappeared.
*) seteuid() tweak for HP-UX 11. Milton L. Hankins [mlh@swl.msd.ray.com]
*) Increase the number of open FDs we look to close when spawning apps.
*) Prevent an assert from popping unnecessarily.
*) Added support for blocking connect()s by setting appConnTimeout to 0.
Non-blocking connect()s (the default) can be troublesome on some platforms.
*) Wrap the SIGPIPE handler manipulation code such that it is only applicable
to Apache releases prior to 1.3.6.
*) Minor tweaks for RUSSIAN_APACHE.
*) Dynamic updates:
Always restart a failed dynamic application if it is the last instance.
This means there once started there will always be at least one process
instance of a dynamic application.
Send PLEASE_START to the PM when a connect() results in an ECONNREFUSED.
ECONNREFUSED means the listen queue is full (or there isn't one). Asking
the PM to start (another) application instance may help empty it faster.
Change two sleep() calls to select() based snoozes because alarm() is in
effect and sleep() and alarm() don't always play nice together.
*) Fix -listen-queue-depth arg on FastCgiConfig (dynamic). Previously it
was ignored and the default was always used.
*) Allow the -initial-env argument to be used to pass variables from the
Apache process environment to the FastCGI server (by specifying a
variable name without the "=" or a value).
*) Improved script stderr handling.
*) Changed the default listen-queue-depth (FCGI_DEFAULT_LISTEN_Q) from
5 to 100. Its still configurable with the -listen-queue-depth option.
*) Fix a bug in FastCgiExternalServer that broke support for external servers
on other hosts.
