Merge remote-tracking branch 'asf/9.2.x' into 9.2.x

* asf/9.2.x:
  Updated ChangeLog
  Fix length bug in validate_unmapped_url_path (apache#8080)
  crash fix (apache#8268)
  test_MMH: fix memory leak in unit test (apache#8357)
  Doc: Add proxy.config.cacvhe.mutex_retry_delay (apache#8376)
  Add thread safety to PendingAction operations. (apache#8443)
  Report an error if configure can't find zlib (apache#8446)
  Update roadmap doc with latest releases (apache#8977)
  Setup UA consumer only if ua_entry is not nullptr (apache#8949)
  Update slice to only prefetch when first block is miss/hit-stale (apache#8890)
  Add RangeTransform::m_write_vio state checks (apache#8980)
  Fix compile on M1 Mac (apache#8999)
  Add stack guard pages (apache#8996)
  Fail fast on HTTP/2 header validation (apache#9009)
  Restrict unknown scheme of HTTP/2 request (apache#9010)
  Add content length mismatch check on handling HEADERS frame and CONTINUATION frame (apache#9012)
  Ignore POST request case from a check for background fill (apache#9013)
  Add back validatation that the scheme matches the wire protocol (apache#9005)
  Pin flask to version 2.1.3 (apache#9008)

CHANGELOG-9.2.0

@@ -199,6 +199,7 @@ Changes with Apache Traffic Server 9.2.0
   #8075 - Add plugin parent_select reloading
   #8076 - Fix plugin parent_select missing null checks.
   #8078 - Clean up constructors and initialization in LogAccess class.
+  #8080 - Fix length bug in validate_unmapped_url_path
   #8083 - Cancel active timeout when releasing session
   #8084 - Defer canceling UA inactivity timeout for chunked bodies too
   #8085 - Add cache condition in header_rewrite plugin for lookup results
@@ -226,7 +227,6 @@ Changes with Apache Traffic Server 9.2.0
   #8187 - Ran clang-tidy over the master branch
   #8188 - lua: add ts.is_debug_tag_set function
   #8190 - Add TSMgmtConfigFileAdd api doc
-  #8191 - Fix error connection logging crash
   #8194 - Remove global g_rec_config_contents_llq
   #8201 - Fix H2 logic when setting EOS flag on DATA frame
   #8204 - transaction data sink: only stream body bytes
@@ -260,6 +260,7 @@ Changes with Apache Traffic Server 9.2.0
   #8264 - Prefetch plugin: add  --fetch-query option
   #8266 - Do not turn off cache for internal requests
   #8267 - Align TS API Au test with Yahoo internal version.
+  #8268 - Avoid chunking in the post redirect case
   #8269 - Fix leaks in ConfigManager::configName
   #8271 - Add warning message about why enhance your calm is being sent
   #8274 - Use better done file locations for some AuTests
@@ -303,6 +304,7 @@ Changes with Apache Traffic Server 9.2.0
   #8348 - remove unused RecConfigFileEntry from RecConfigParse
   #8354 - Adds support for TCP_NOTSENT_LOWAT sockopt
   #8356 - free sessions when timeout
+  #8357 - test_MMH: fix memory leak in unit test
   #8358 - Locking around SSLSecret::secret_map access
   #8359 - Fix regex_remap plugin redirection.
   #8360 - Stabilize regex_revalidate Au test.
@@ -313,6 +315,7 @@ Changes with Apache Traffic Server 9.2.0
   #8372 - Add Au test for pqsi and pqsp log fields.
   #8374 - 9.2.x autest fixes
   #8375 - Add a set-http-cntl operator for header_rewrite
+  #8376 - Doc: Add proxy.config.cache.mutex_retry_delay
   #8381 - Fix a potential H2 stall
   #8392 - Update the feature lists available for autest
   #8395 - Added metrics to the rate limit plugin and document the new options
@@ -334,7 +337,9 @@ Changes with Apache Traffic Server 9.2.0
   #8435 - Fix unused-but-set-variable warnings
   #8440 - Update the Multiplexer Docs for Multplexed HTTPS Connections
   #8441 - bigobj: use automake to build test utilities
+  #8443 - Add thread safety to PendingAction operations.
   #8445 - Include CPPFLAGS when building CompileParseRules
+  #8446 - Report an error if configure can't find zlib
   #8449 - Add AuTest for HTTP CONNECT
   #8450 - Document proxy.config.memory.max_usage
   #8453 - Ignore ECONNABORTED on blocking accept
@@ -363,7 +368,6 @@ Changes with Apache Traffic Server 9.2.0
   #8518 - Add 5xx's to be allowed to be used for simple retries
   #8520 - Extend milestone api time tracking to remap.
   #8523 - url_sig: change cleanup msg from TSError to TSDebug
-  #8524 - Set an appropriate callback function for OpenSSL 3
   #8525 - Remove unused function "ats_mallopt"
   #8531 - Destroy ssl context after use.
   #8532 - Adds autest to test error cases loading ssl_multicert.config
@@ -485,6 +489,7 @@ Changes with Apache Traffic Server 9.2.0
   #8882 - Make ats_scoped_obj a synonym for std::unique_ptr .
   #8884 - Upgrade to Proxy Verifier 2.4.0.
   #8886 - Add escape json for logging
+  #8890 - Update slice to only prefetch when first block is miss/hit-stale
   #8892 - Cleanup SNIConfig
   #8897 - Removed references to the throttle option from the slice plugin. (#8373)
   #8898 - slice and cache_range_requests: allow header override (#8666)
@@ -493,7 +498,6 @@ Changes with Apache Traffic Server 9.2.0
   #8909 - Move HKDF to OpenSSL 3 interfaces
   #8924 - Update location for core rule set in modsecurity example
   #8925 - Cleanup. Remove unnecessary use of a memory arena when logging.
-  #8926 - Allows errors from plugin initialization to bubble up
   #8927 - Add docs for remap_stats plugin
   #8928 - proxy_serve_stale: Test updates
   #8931 - Fix clang-format installation with multiple threads
@@ -502,8 +506,20 @@ Changes with Apache Traffic Server 9.2.0
   #8942 - Fix doc formatting for plugin remap_stats
   #8943 - Fix doc formatting for rate_limit plugin
   #8944 - Make the autopep8 clang-format targets quieter
+  #8949 - Setup UA consumer only if ua_entry is not nullptr
   #8950 - .fit/fmt/.clang-format-installed prerequisite
   #8965 - Proxy Verifier: Update to version 2.4.1
   #8970 - add a metric to track how often the range seek bug is detected
   #8971 - Fixes issue with file size calculation for existing logs
   #8975 - Fix reverting PR#7302
+  #8977 - Update roadmap doc with latest releases
+  #8980 - Add RangeTransform::m_write_vio state checks
+  #8996 - Add stack guard pages
+  #8999 - Fix compile on M1 Mac
+  #9005 - Add back validatation that the scheme matches the wire protocol
+  #9008 - Pin flask to version 2.1.3
+  #9009 - Fail fast on HTTP/2 header validation
+  #9010 - Restrict unknown scheme of HTTP/2 request
+  #9011 - Add control char check in MIME Parser
+  #9012 - Add content length mismatch check on handling HEADERS frame and CONTINUATION frame
+  #9013 - Ignore POST request case from a check for background fill

configure.ac

@@ -1390,6 +1390,9 @@ AM_CONDITIONAL([BUILD_JA3_PLUGIN], [test "x${enable_ja3_plugin}" = "xyes"])
 #
 # Check for zlib presence and usability
 TS_CHECK_ZLIB
+if test "x${enable_zlib}" != "xyes"; then
+  AC_MSG_ERROR([Cannot find zlib library. Configure --with-zlib=DIR])
+fi
 
 #
 # Check for lzma presence and usability

doc/admin-guide/files/records.config.en.rst

@@ -2428,6 +2428,17 @@ Cache Control
    write vector. For further details on cache write vectors, refer to the
    developer documentation for :cpp:class:`CacheVC`.
 
+.. ts::cv:: CONFIG proxy.config.cache.mutex_retry_delay INT 2
+   :reloadable:
+   :units: milliseconds
+
+   The retry delay for missing a lock on a mutex in the cache component. This is used generically
+   for most locks, except those that have an explicit configuration for the retry delay. For
+   instance, if the cache component is notifying another continuation of a cache event and fails to
+   get the lock for that continuation, it will use this as the delay for the retry. This is also
+   used from the asynchronous IO threads when IO finishes and the ``CacheVC`` lock or stripe lock is
+   required.
+
 RAM Cache
 =========
 

doc/release-notes/roadmap.en.rst

@@ -26,8 +26,10 @@ ATS Release Roadmap
    :maxdepth: 1
 
 Apache Traffic Server is currently on a two year major release cycle. At any
-given time, we guarantee to have two major versions supported. Currently,
-this means v7.1.x and v8.0.x, with a possible v8.1.x coming.
+given time, we guarantee to have two major versions supported. Please refer to
+our `download page <https://trafficserver.apache.org/downloads>`_ to see the
+current supported versions.
+
 
 For details on the actual developer release process, See
 

include/tscore/PendingAction.h

@@ -73,13 +73,15 @@ class PendingAction
    *
    * @param action @c Action to check.
    *
+   * @return @c true if the action was cleared, @c false if not.
+   *
    * This clears the internal pointer without any side effect. it is used when the @c Action
    * is handled and therefore should no longer be canceled.
    */
-  void clear_if_action_is(Action *action);
+  bool clear_if_action_is(Action *action);
 
 private:
-  Action *pending_action = nullptr;
+  std::atomic<Action *> pending_action = nullptr;
 };
 
 inline bool
@@ -91,22 +93,32 @@ PendingAction::empty() const
 inline PendingAction &
 PendingAction::operator=(Action *action)
 {
-  // Apparently HttpSM depends on not canceling the previous action if anew
+  // Apparently @c HttpSM depends on not canceling the previous action if a new
   // one completes immediately. Canceling the contained action in that case
-  // cause the HttpSm to permanently stall.
+  // cause the @c HttpSM to permanently stall.
   if (ACTION_RESULT_DONE != action) {
-    if (action != pending_action && pending_action != nullptr) {
-      pending_action->cancel();
+    Action *expected; // Need for exchange, and to load @a pending_action only once.
+    // Avoid race conditions - for each assigned action, ensure exactly one thread
+    // cancels it. Assigning @a expected in the @c while expression avoids potential
+    // races if two calls to this method have the same @a action.
+    while ((expected = pending_action) != action) {
+      if (pending_action.compare_exchange_strong(expected, action)) {
+        // This thread did the swap and now no other thread can get @a expected which
+        // must be canceled if it's not @c nullptr and then this thread is done.
+        if (expected != nullptr) {
+          expected->cancel();
+        }
+        break;
+      }
     }
-    pending_action = action;
   }
   return *this;
 }
 
 inline Continuation *
 PendingAction::get_continuation() const
 {
-  return pending_action ? pending_action->continuation : nullptr;
+  return pending_action ? pending_action.load()->continuation : nullptr;
 }
 
 inline Action *
@@ -118,14 +130,20 @@ PendingAction::get() const
 inline PendingAction::~PendingAction()
 {
   if (pending_action) {
-    pending_action->cancel();
+    pending_action.load()->cancel();
   }
 }
 
-inline void
+inline bool
 PendingAction::clear_if_action_is(Action *action)
 {
-  if (action == pending_action) {
-    pending_action = nullptr;
+  if (action != nullptr) {
+    while (action == pending_action) {
+      if (pending_action.compare_exchange_strong(action, nullptr)) {
+        // do NOT cancel - this is called when the event is handled.
+        return true;
+      }
+    }
   }
+  return false;
 }

iocore/eventsystem/UnixEventProcessor.cc

@@ -47,15 +47,16 @@ class ThreadAffinityInitializer : public Continuation
   void init();
   /// Set the affinity for the current thread.
   int set_affinity(int, Event *);
-  /// Allocate a stack.
+  /// Allocate a stack and set guard pages.
   /// @internal This is the external entry point and is different depending on
   /// whether HWLOC is enabled.
   void *alloc_stack(EThread *t, size_t stacksize);
 
 protected:
   /// Allocate a hugepage stack.
   /// If huge pages are not enable, allocate a basic stack.
-  void *alloc_hugepage_stack(size_t stacksize);
+  void *do_alloc_stack(size_t stacksize);
+  void setup_stack_guard(void *stack, int stackguard_pages);
 
 #if TS_USE_HWLOC
 
@@ -136,10 +137,55 @@ class ThreadInitByFunc : public Continuation
 } Thread_Init_Func;
 } // namespace
 
+void
+ThreadAffinityInitializer::setup_stack_guard(void *stack, int stackguard_pages)
+{
+#if !(defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__arm64__) || defined(__aarch64__) || \
+      defined(__mips__))
+#error Unknown stack growth direction.  Determine the stack growth direction of your platform.
+// If your stack grows upwards, you need to change this function and the calculation of stack_begin in do_alloc_stack.
+#endif
+  // Assumption: stack grows down
+  if (stackguard_pages <= 0) {
+    return;
+  }
+
+  size_t pagesize  = ats_hugepage_enabled() ? ats_hugepage_size() : ats_pagesize();
+  size_t guardsize = stackguard_pages * pagesize;
+  int ret          = mprotect(stack, guardsize, 0);
+  if (ret != 0) {
+    Fatal("Failed to set up stack guard pages: %s (%d)", strerror(errno), errno);
+  }
+}
+
 void *
-ThreadAffinityInitializer::alloc_hugepage_stack(size_t stacksize)
+ThreadAffinityInitializer::do_alloc_stack(size_t stacksize)
 {
-  return ats_hugepage_enabled() ? ats_alloc_hugepage(stacksize) : ats_memalign(ats_pagesize(), stacksize);
+  size_t pagesize = ats_hugepage_enabled() ? ats_hugepage_size() : ats_pagesize();
+  int stackguard_pages;
+  REC_ReadConfigInteger(stackguard_pages, "proxy.config.thread.default.stackguard_pages");
+  ink_release_assert(stackguard_pages >= 0);
+
+  size_t size    = INK_ALIGN(stacksize + stackguard_pages * pagesize, pagesize);
+  int mmap_flags = MAP_PRIVATE | MAP_ANONYMOUS;
+#ifdef MAP_HUGETLB
+  if (ats_hugepage_enabled()) {
+    mmap_flags |= MAP_HUGETLB;
+  }
+#endif
+  void *stack_and_guard = mmap(nullptr, size, PROT_READ | PROT_WRITE, mmap_flags, -1, 0);
+  if (stack_and_guard == MAP_FAILED) {
+    Error("Failed to allocate stack pages: size = %zu", size);
+    return nullptr;
+  }
+
+  setup_stack_guard(stack_and_guard, stackguard_pages);
+
+  void *stack_begin = static_cast<char *>(stack_and_guard) + stackguard_pages * pagesize;
+  Debug("iocore_thread", "Allocated %zu bytes (%zu bytes in guard pages) for stack {%p-%p guard, %p-%p stack}", size,
+        stackguard_pages * pagesize, stack_and_guard, stack_begin, stack_begin, static_cast<char *>(stack_begin) + stacksize);
+
+  return stack_begin;
 }
 
 #if TS_USE_HWLOC
@@ -239,7 +285,7 @@ ThreadAffinityInitializer::alloc_numa_stack(EThread *t, size_t stacksize)
   }
 
   // Alloc our stack
-  stack = this->alloc_hugepage_stack(stacksize);
+  stack = this->do_alloc_stack(stacksize);
 
   if (mem_policy != HWLOC_MEMBIND_DEFAULT) {
     // Now let's set it back to default for this thread.
@@ -260,7 +306,7 @@ ThreadAffinityInitializer::alloc_numa_stack(EThread *t, size_t stacksize)
 void *
 ThreadAffinityInitializer::alloc_stack(EThread *t, size_t stacksize)
 {
-  return this->obj_count > 0 ? this->alloc_numa_stack(t, stacksize) : this->alloc_hugepage_stack(stacksize);
+  return this->obj_count > 0 ? this->alloc_numa_stack(t, stacksize) : this->do_alloc_stack(stacksize);
 }
 
 #else
@@ -279,7 +325,7 @@ ThreadAffinityInitializer::set_affinity(int, Event *)
 void *
 ThreadAffinityInitializer::alloc_stack(EThread *, size_t stacksize)
 {
-  return this->alloc_hugepage_stack(stacksize);
+  return this->do_alloc_stack(stacksize);
 }
 
 #endif // TS_USE_HWLOC

mgmt/RecordsConfig.cc

@@ -125,6 +125,8 @@ static const RecordElement RecordsConfig[] =
   ,
   {RECT_CONFIG, "proxy.config.thread.default.stacksize", RECD_INT, "1048576", RECU_RESTART_TS, RR_NULL, RECC_INT, "[131072-104857600]", RECA_READ_ONLY}
   ,
+  {RECT_CONFIG, "proxy.config.thread.default.stackguard_pages", RECD_INT, "1", RECU_RESTART_TS, RR_NULL, RECC_INT, "[1-256]", RECA_READ_ONLY}
+  ,
   {RECT_CONFIG, "proxy.config.restart.active_client_threshold", RECD_INT, "0", RECU_DYNAMIC, RR_NULL, RECC_NULL, nullptr, RECA_NULL}
   ,
   {RECT_CONFIG, "proxy.config.restart.stop_listening", RECD_INT, "0", RECU_DYNAMIC, RR_NULL, RECC_INT, "[0-1]", RECA_NULL}

proxy/Transform.cc

@@ -791,7 +791,9 @@ RangeTransform::handle_event(int event, void *edata)
   } else {
     switch (event) {
     case VC_EVENT_ERROR:
-      m_write_vio.cont->handleEvent(VC_EVENT_ERROR, &m_write_vio);
+      if (m_write_vio.cont) {
+        m_write_vio.cont->handleEvent(VC_EVENT_ERROR, &m_write_vio);
+      }
       break;
     case VC_EVENT_WRITE_COMPLETE:
       ink_assert(m_output_vio == (VIO *)edata);
@@ -814,6 +816,10 @@ RangeTransform::handle_event(int event, void *edata)
         }
       }
 
+      if (!m_write_vio.mutex) {
+        return 0;
+      }
+
       MUTEX_TRY_LOCK(trylock, m_write_vio.mutex, this_ethread());
       if (!trylock.is_locked()) {
         retry(10);

proxy/http/HttpSM.cc

@@ -6049,6 +6049,7 @@ HttpSM::do_setup_post_tunnel(HttpVC_t to_vc_type)
     // When redirect in process is true and redirection is enabled
     // add http server as the consumer
     if (post_redirect) {
+      chunked = false;
       HTTP_SM_SET_DEFAULT_HANDLER(&HttpSM::tunnel_handler_for_partial_post);
       tunnel.add_consumer(server_entry->vc, HTTP_TUNNEL_STATIC_PRODUCER, &HttpSM::tunnel_handler_post_server, HT_HTTP_SERVER,
                           "redirect http server post");
@@ -6683,18 +6684,22 @@ HttpSM::setup_internal_transfer(HttpSMHandler handler_arg)
 
   HTTP_SM_SET_DEFAULT_HANDLER(handler_arg);
 
-  // Clear the decks before we setup the new producers
-  // As things stand, we cannot have two static producers operating at
-  // once
-  tunnel.reset();
+  if (ua_entry && ua_entry->vc) {
+    // Clear the decks before we setup the new producers
+    // As things stand, we cannot have two static producers operating at
+    // once
+    tunnel.reset();
 
-  // Setup the tunnel to the client
-  HttpTunnelProducer *p =
-    tunnel.add_producer(HTTP_TUNNEL_STATIC_PRODUCER, nbytes, buf_start, (HttpProducerHandler) nullptr, HT_STATIC, "internal msg");
-  tunnel.add_consumer(ua_entry->vc, HTTP_TUNNEL_STATIC_PRODUCER, &HttpSM::tunnel_handler_ua, HT_HTTP_CLIENT, "user agent");
+    // Setup the tunnel to the client
+    HttpTunnelProducer *p =
+      tunnel.add_producer(HTTP_TUNNEL_STATIC_PRODUCER, nbytes, buf_start, (HttpProducerHandler) nullptr, HT_STATIC, "internal msg");
+    tunnel.add_consumer(ua_entry->vc, HTTP_TUNNEL_STATIC_PRODUCER, &HttpSM::tunnel_handler_ua, HT_HTTP_CLIENT, "user agent");
 
-  ua_entry->in_tunnel = true;
-  tunnel.tunnel_run(p);
+    ua_entry->in_tunnel = true;
+    tunnel.tunnel_run(p);
+  } else {
+    (this->*default_handler)(HTTP_TUNNEL_EVENT_DONE, &tunnel);
+  }
 }
 
 // int HttpSM::find_http_resp_buffer_size(int cl)