* server/mpm/event/event.c (listener_thread): EVENT_FUDGE_FACTOR is used to
limit wakeups but still expiring timers should be compared with the real
apr_time_now().
mpm_event: clear APR_INCOMPLETE_READ on lingering close.
* server/mpm/event/event.c (process_lingering_close): APR_INCOMPLETE_READ may
have been set anytime while processing the connection, clear it before
nonblocking apr_socket_recv() to prevent internal apr_poll().
* server/connection.c (ap_lingering_close): Do apr_socket_close() on
ap_start_lingering_close() failure.
On failure ap_start_lingering_close() did not consistently close the socket,
so the caller had to call apr_socket_close() too with possible/unreliable
EBADF. The only upstream callers of ap_start_lingering_close() are MPM event
and ap_start_lingering_close(), but any third-party user has this issue so the
change should't break anyone.
mpm_event: handle lingering close fully in process_lingering_close().
* server/mpm/event/event.c (start_lingering_close_blocking): Axe the function
and put the code directly in process_lingering_close() on the first call when
cs->pub.state == CONN_STATE_LINGER, this simplifies lingering close handling
from process_socket() and event_resume_suspended().
mpm_event: Fix queues' maintenance and linger timeouts on graceful restart/stop
* server/mpm/event/event.c (abort_socket_nonblocking): Renamed to
close_socket_nonblocking_() with close_socket_nonblocking() as a macro to
APLOG_TRACE8 the caller __FUNCTION__ and __LINE__. Do nothing if the socket
is closed already, e.g. from close_worker_sockets().
* server/mpm/event/event.c (close_worker_sockets): Let defer_linger_chain be
cleaned by the terminating workers.
* server/mpm/event/event.c (wakeup_listener): Add a trace8 of the call.
* server/mpm/event/event.c (decrement_connection_count): Add a trace8 of the
call, decrement lingering_count from CONN_STATE_LINGER too which may happen
from start_lingering_close_blocking() on ap_start_lingering_close() failure,
and when dying call ap_queue_interrupt_one() from there instead of
close_connection() so that we don't miss a worker_thread_should_exit_early()
case.
* server/mpm/event/event.c (start_lingering_close_blocking): Don't increment
lingering_count when it's already done by defer_lingering_close() (i.e.
cs->deferred_linger is set), set socket timeout to 2s (SECONDS_TO_LINGER)
before pre_close hooks and then after let socket timeout handling to
process_lingering_close().
* server/mpm/event/event.c (start_lingering_close_nonblocking): Rename to
defer_lingering_close(), add trace6 of the call, set CONN_STATE_LINGER so
that the connection is accounted for until handled by a worker, and let
start_lingering_close_blocking() know by setting cs->deferred_linger.
* server/mpm/event/event.c (stop_lingering_close): Rename to close_connection()
and log the call at trace6 level instead of trace4 (for consistency), and use
it everywhere apr_socket_close()+ap_queue_info_push_pool() is open coded.
* server/mpm/event/event.c (shutdown_connection): This helper can be called
anywhere where nonblocking is required to either initiate short lingering
close or directly close if the connection is lingering already. It's used
as a generic process_timeout_queue() callback to terminate expired entries
in any state.
* server/mpm/event/event.c (update_reqevents_from_sense): Allow to specify the
sense as argument where -1 means cs->pub.state as before, and call it from
everywhere before adding to the pollset such that APR_POLL* events are set
fully/appropritely regardless of the underlying pollset implementation.
* server/mpm/event/event.c (process_socket): Gracefuly stop the process on
pollset errors, and don't forcibly close the connection if listener_may_exit
when ap_run_input_pending() determines that the next request is already here.
* server/mpm/event/event.c (close_listeners): Add trace6 of the call, and
return whether it's the first close or not.
* server/mpm/event/event.c (push2worker): Call shutdown_connection() on failure
when cs is available so that pre_close hooks (h2 flush, ssl close notify) are
still called.
* server/mpm/event/event.c (process_lingering_close): Add trace6 of the call,
always set timeout to zero (hence nonblocking) before reading pending data,
and terminate the process gracefuly on pollset errors.
* server/mpm/event/event.c (listener_thread): Add some trace7 messages around
poll()ing for debug, round poll()ing timeout to the upper millisecond to work
around the rounding down done by some pollset implementations, and on the
first call to close_listeners() when listener_may_exit->dying let's force the
maintenance of the queues to kill keep alive connections and apply shorter
lingering close timeout to free worker threads faster for the next generation.
* server/mpm/event/event.c (join_workers): Rewrite the !dying loop to avoid
"listener has not stopped accepting yet" if sleeping 500ms once did it.
Stefan Eissing [Tue, 20 Jul 2021 12:44:31 +0000 (12:44 +0000)]
*) mod_md:
- Domain names in `<MDomain ...>` can now appear in quoted form.
- Fixed a failure in ACME challenge selection that aborted further searches
when the tls-alpn-01 method did not seem to be suitable.
- Changed the tls-alpn-01 setup to only become unsuitable when none of the
dns names showed support for a configured 'Protocols ... acme-tls/1'. This
allows use of tls-alpn-01 for dns names that are not mapped to a VirtualHost.
Stefan Eissing [Thu, 15 Jul 2021 08:29:19 +0000 (08:29 +0000)]
* mod_http2:
- Added a timeout to h2 worker cleanup to exit latest after 5 seconds of
waiting on idle workers to terminate. This happens after all connections
have been processed. a WARNING is logged in case workers lagged behind.
mod_proxy: Fix icomplete initialization of BalancerMember(s) from the manager.
Clear the workers created in ap_proxy_sync_balancer(), notably ->local_status
for below ap_proxy_initialize_worker() to initialize all the child structures
like ->cp and ->cp->reslist, avoiding a possible crash when the workers are
used at runtime.
Stefan Eissing [Tue, 6 Jul 2021 13:06:00 +0000 (13:06 +0000)]
*) mod_http2:
- Aborting requests via RST_STREAM no longer affect the available
resources of a connection when the first chunk of the response
body has been sent.
- H2Min/MaxWorkers behave as intended again. The module will initially
create H2MinWorkers threads and add up to H2MaxWorkers when needed. These
additional workers time out when idle after H2MaxWorkerIdleSeconds and
disappear again.
- When the shutdown of a child is detected (e.g. graceful shutdown), the
module will terminate all idle workers above H2MinWorkers right away.
This detection currently only happens when a HTTP/2 connection is active.
mod_proxy: Avoid confusion of prefix/regex matching workers at loading. PR 65429.
ap_proxy_get_worker() needs to know whether it should lookup for prefix or
match or both matching workers, depending on the context.
For instance <Proxy[Match]> or ProxyPass[Match] directives need to lookup for
an existing worker with the same type as the directive (*Match or not), because
they will define one with that matching type if none exists.
On the contrary, "ProxySet <url>" at load time or ap_proxy_pre_request() at run
time need to find a worker matching an url whether it's by prefix or by regex.
So this commit adds ap_proxy_get_worker_ex() which takes a bitmask for the
matching type and calls it appropriately where needed.
For consistency, ap_proxy_define_worker_ex() is also added, using the same
bitmask flags, deprecating ap_proxy_define_match_worker().
mpm_proxy: Fix possible reuse/merging of Proxy(Pass)Match workers. PR 65419.
We can't truncate ProxyMatch's worker name/url to the first '$' substitution
without possibly colliding with other workers. This also makes the matching
done at runtime by ap_proxy_strcmp_ematch() completely pointless.
To fix this and still address r1878467 (i.e. make http://host:port$1 a "valid"
URL), we need to remove '$' substitutions from the :port part of the URL only
since it's allowed anywhere else by apr_uri_parse().
So let's strip them before apr_uri_parse() and prepend them back in the path
before apr_uri_unparse() to restore the original URL. Non-matchable workers are
not concerned so ap_proxy_define_worker() is made a local helper (w/o the ap_
prefix) which takes "matchable" as argument and can then be called by both
ap_proxy_define_[match_]worker() functions.
Yann Ylavic [Tue, 29 Jun 2021 21:16:21 +0000 (21:16 +0000)]
core: Write Completion (WC) bucket type.
A WC bucket is meant to prevent buffering/coalescing filters from retaining
data, but unlike a FLUSH bucket it won't cause the core output filter to
block trying to flush anything before.
It can be passed by async handlers which want to never block, followed by
ap_filter_should_yield() to check for pending data and eventually suspend
processing until MPM/asynchronous write completion finishes.
In this commit it's used that way by the tunneling loop of mod_proxy to
prevent SSL coaslescing.
Graham Leggett [Thu, 24 Jun 2021 10:27:49 +0000 (10:27 +0000)]
dbm: Split the loading of a dbm driver from the opening of a dbm file. When
an attempt to load a dbm driver fails, log clearly which driver triggered
the error (not "default"), and what the error was.
Stefan Eissing [Fri, 11 Jun 2021 10:45:25 +0000 (10:45 +0000)]
*) mod_ssl: tighten the handling of ALPN for outgoing (proxy)
connections. If ALPN protocols are provided and sent to the
remote server, the received protocol selected is inspected
and checked for a match. Without match, the peer handshake
fails.
An exception is the proposal of "http/1.1" where it is
accepted if the remote server did not answer ALPN with
a selected protocol. This accomodates for hosts that do
not observe/support ALPN and speak http/1.x be default.
Stefan Eissing [Tue, 8 Jun 2021 14:37:44 +0000 (14:37 +0000)]
*) core/mod_proxy/mod_ssl:
Adding `outgoing` flag to conn_rec, indicating a connection is
initiated by the server to somewhere, in contrast to incoming
connections from clients.
Adding 'ap_ssl_bind_outgoing()` function that marks a connection
as outgoing and is used by mod_proxy instead of the previous
optional function `ssl_engine_set`. This enables other SSL
module to secure proxy connections.
The optional functions `ssl_engine_set`, `ssl_engine_disable` and
`ssl_proxy_enable` are now provided by the core to have backward
compatibility with non-httpd modules that might use them. mod_ssl
itself no longer registers these functions, but keeps them in its
header for backward compatibility.
The core provided optional function wrap any registered function
like it was done for `ssl_is_ssl`.
Yann Ylavic [Fri, 4 Jun 2021 13:21:28 +0000 (13:21 +0000)]
mpm_prefork: mask signals during ap_run_child_init().
This prevents threads potentially created from the child_init hooks (e.g.
mod_watchdog workers) to catch signals needed by the MPM, like here:
https://travis-ci.com/github/apache/httpd/jobs/510821148#L5356.
Joe Orton [Fri, 21 May 2021 09:58:14 +0000 (09:58 +0000)]
mod_ssl: Switch to using OpenSSL's automatic internal DH parameter
generation from OpenSSL 1.1.0 and later. The
SSL_set_tmp_dh_callback() API is deprecated from OpenSSL 3.0 onwards.
Should not be a user-visible change (except mod_ssl gets smaller).
* modules/ssl/ssl_private.h,
modules/ssl/ssl_engine_kernel.c,
modules/ssl/ssl_engine_init.c (ssl_init_ctx_callbacks):
Drop internal DH parameter generation and callback for OpenSSL 1.1+,
use SSL_CTX_set_dh_auto(, 1) instead.
Stefan Eissing [Tue, 18 May 2021 14:42:52 +0000 (14:42 +0000)]
* mod_log_config/mod_ssl: moved the log_handlers registered by mod_ssl
into mod_log_config itself. These now use the global `ap_ssl_var_lookup()`
functions and work for all running SSL modules.
The dependency from mod_ssl to mod_log_config and its header is removed.
mod_ssl now provides the content of "{errstr}c" as variable "SSL_CLIENT_VERIFY_ERRSTR".
This change should be fully compatible to all deployed configurations.
Joe Orton [Mon, 17 May 2021 11:56:32 +0000 (11:56 +0000)]
* server/mpm/event/event.c,
server/mpm/simple/simple_run.c,
support/htcacheclean.c: Adjust use of APR_RING macros to
ensure the APR_RING_HEAD is always embedded in a containing
structure, to avoid warnings with GCC 11. (apr_ring.h also
suggests this is best practice)
See also: https://bugzilla.redhat.com/show_bug.cgi?id=1957353
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98503
msgid <814cca2a-1992-3fae-bb0e-c84b3d09b73b@gmail.com>
Github: closes #186
Yann Ylavic [Sun, 16 May 2021 21:49:49 +0000 (21:49 +0000)]
mod_ssl: coalesce using a bucket brigade and the setaside/reinstate mechanism.
ssl_io_filter_coalesce() now uses apr_brigade_write() to save its retained data
in a heap bucket, and ap_filter_{setaside,reinstate}_brigade() to declare them
to the output filters' write completion mechanism.
This prevents MPM event to miss them when it enters write completion state, and
will allow the tunneling loop of mod_proxy to flush them in a following commit
too.
Stefan Eissing [Thu, 22 Apr 2021 10:17:02 +0000 (10:17 +0000)]
mod_md: make sure its post-config hook runs before mod_watchdog. if
not, mod_watchdog thinks it has nothing to do, if no one else installs
any watchdog instance.
Stefan Eissing [Tue, 20 Apr 2021 12:16:05 +0000 (12:16 +0000)]
core/ap_ssl_*: changes after review by rpluem
- removed no longer needed (char*) casts when looking
up ssl variables.
- move 'goto cleanup;' on separate source line
- fixed check for wrong optional function in ap_run_ssl_var_lookup
- remove ap_bytes_t again from httpd.h and passes now ocsp
identifier as separate const char* and apr_size_t. This
follows more how such data is passed in the rest of the
server.
Stefan Eissing [Tue, 13 Apr 2021 11:12:00 +0000 (11:12 +0000)]
*) core/mod_ssl/mod_md: adding OCSP response provisioning as core feature. This
allows modules to access and provide OCSP response data without being tied
of each other. The data is exchanged in standard, portable formats (PEM encoded
certificates and DER encoded responses), so that the actual SSL/crypto
implementations used by the modules are independant of each other.
Registration and retrieval happen in the context of a server (server_rec)
which modules may use to decide if they are configured for this or not.
The area of changes:
1. core: defines 2 functions in include/http_ssl.h, so that modules may
register a certificate, together with its issuer certificate for OCSP
response provisioning and ask for current response data (DER bytes) later.
Also, 2 hooks are defined that allow modules to implement this OCSP
provisioning.
2. mod_ssl uses the new functions, in addition to what it did already, to
register its certificates this way. If no one is interested in providing
OCSP, it falls back to its own (if configured) stapling implementation.
3. mod_md registers itself at the core hooks for OCSP provisioning. Depending
on configuration, it will accept registrations of its own certificates only,
all certficates or none.
mod_socache_shmcb: be safe from socache_shmcb_destroy() late call.
ssl_init_Module() in post_config early registers ssl_init_ModuleKill(), which
will then run after all the next cleanups registered later in post_config, thus
any shm_cleanup() registered from ssl_scache_init::socache_shmcb_init().
This can cause a double SHM cleanup when apr_shm_destroy() is called from
ssl_init_ModuleKill() as pconf is cleared.
Fix this in mod_socache_shmcb by registering a socache_shmcb_cleanup() after
the SHM is created, and by letting socache_shmcb_destroy() run the cleanup,
such that shm_cleanup() is always and ever called only once.
Ideally apr_shm_create() would be consistent accross platforms to register its
shm_cleanup() on the pool but that's not the case for now (I'm on it), so httpd
has to call apr_shm_destroy() explicitely from several places (we'll be able to
remove ssl_scache_kill() and other similar cleanups once the minimal APR
version required by httpd is fixed..).
We could also fix this by registering ssl_init_ModuleKill() late(r) in
ssl_init_Module(), though the more robust mod_socache_shmcb the better for
all the modules..
projects / thirdparty / apache / httpd.git / log
