Improved EPOLLRDHUP handling.

When it's known that the kernel supports EPOLLRDHUP, there is no need in
additional recv() call to get EOF or error when the flag is absent in the
event generated by the kernel.  A special runtime test is done at startup
to detect if EPOLLRDHUP is actually supported by the kernel because
epoll_ctl() silently ignores unknown flags.

With this knowledge it's now possible to drop the "ready" flag for partial
read.  Previously, the "ready" flag was kept until the recv() returned EOF
or error.  In particular, this change allows the lingering close heuristics
(which relies on the "ready" flag state) to actually work on Linux, and not
wait for more data in most cases.

The "available" flag is now used in the read event with the semantics similar
to the corresponding counter in kqueue.

src/event/modules/ngx_epoll_module.c

@@ -123,6 +123,7 @@ static ngx_int_t ngx_epoll_process_events(ngx_cycle_t *cycle, ngx_msec_t timer,
 static void ngx_epoll_eventfd_handler(ngx_event_t *ev);
 #endif
 
+static ngx_int_t ngx_epoll_module_init(ngx_cycle_t *cycle);
 static void *ngx_epoll_create_conf(ngx_cycle_t *cycle);
 static char *ngx_epoll_init_conf(ngx_cycle_t *cycle, void *conf);
 
@@ -146,6 +147,10 @@ static ngx_connection_t     ngx_eventfd_conn;
 
 #endif
 
+#if (NGX_HAVE_EPOLLRDHUP)
+ngx_uint_t                  ngx_use_epoll_rdhup;
+#endif
+
 static ngx_str_t      epoll_name = ngx_string("epoll");
 
 static ngx_command_t  ngx_epoll_commands[] = {
@@ -197,7 +202,7 @@ ngx_module_t  ngx_epoll_module = {
     ngx_epoll_commands,                  /* module directives */
     NGX_EVENT_MODULE,                    /* module type */
     NULL,                                /* init master */
-    NULL,                                /* init module */
+    ngx_epoll_module_init,               /* init module */
     NULL,                                /* init process */
     NULL,                                /* init thread */
     NULL,                                /* exit thread */
@@ -808,6 +813,8 @@ ngx_epoll_process_events(ngx_cycle_t *cycle, ngx_msec_t timer, ngx_uint_t flags)
             if (revents & EPOLLRDHUP) {
                 rev->pending_eof = 1;
             }
+
+            rev->available = 1;
 #endif
 
             rev->ready = 1;
@@ -943,6 +950,69 @@ ngx_epoll_eventfd_handler(ngx_event_t *ev)
 #endif
 
 
+static ngx_int_t
+ngx_epoll_module_init(ngx_cycle_t *cycle)
+{
+#if (NGX_HAVE_EPOLLRDHUP)
+    int                 epfd, s[2], events;
+    struct epoll_event  ee;
+
+    epfd = epoll_create(1);
+
+    if (epfd == -1) {
+        ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
+                      "epoll_create() failed");
+        return NGX_ERROR;
+    }
+
+    if (socketpair(AF_UNIX, SOCK_STREAM, 0, s) == -1) {
+        ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
+                      "socketpair() failed");
+        return NGX_ERROR;
+    }
+
+    ee.events = EPOLLET|EPOLLIN|EPOLLRDHUP;
+
+    if (epoll_ctl(epfd, EPOLL_CTL_ADD, s[0], &ee) == -1) {
+        ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
+                      "epoll_ctl() failed");
+        return NGX_ERROR;
+    }
+
+    if (close(s[1]) == -1) {
+        ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
+                      "close() failed");
+        return NGX_ERROR;
+    }
+
+    events = epoll_wait(epfd, &ee, 1, 5000);
+
+    if (events == -1) {
+        ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
+                      "epoll_wait() failed");
+        return NGX_ERROR;
+    }
+
+    (void) close(s[0]);
+    (void) close(epfd);
+
+    if (events) {
+        ngx_use_epoll_rdhup = ee.events & EPOLLRDHUP;
+
+    } else {
+        ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
+                      "epoll_wait() timedout");
+    }
+
+    ngx_log_error(NGX_LOG_NOTICE, cycle->log, 0,
+                  "testing the EPOLLRDHUP flag: %s",
+                  ngx_use_epoll_rdhup ? "success" : "fail");
+#endif
+
+    return NGX_OK;
+}
+
+
 static void *
 ngx_epoll_create_conf(ngx_cycle_t *cycle)
 {

src/event/ngx_event.h

@@ -96,6 +96,10 @@ struct ngx_event_s {
      *   write:      available space in buffer when event is ready
      *               or lowat when event is set with NGX_LOWAT_EVENT flag
      *
+     * epoll with EPOLLRDHUP:
+     *   accept:     1 if accept many, 0 otherwise
+     *   read:       1 if there can be data to read, 0 otherwise
+     *
      * iocp: TODO
      *
      * otherwise:
@@ -196,6 +200,9 @@ typedef struct {
 
 
 extern ngx_event_actions_t   ngx_event_actions;
+#if (NGX_HAVE_EPOLLRDHUP)
+extern ngx_uint_t            ngx_use_epoll_rdhup;
+#endif
 
 
 /*

src/http/ngx_http_request.c

@@ -2752,9 +2752,13 @@ ngx_http_test_reading(ngx_http_request_t *r)
 
 #if (NGX_HAVE_EPOLLRDHUP)
 
-    if ((ngx_event_flags & NGX_USE_EPOLL_EVENT) && rev->pending_eof) {
+    if ((ngx_event_flags & NGX_USE_EPOLL_EVENT) && ngx_use_epoll_rdhup) {
         socklen_t  len;
 
+        if (!rev->pending_eof) {
+            return;
+        }
+
         rev->eof = 1;
         c->error = 1;
 

src/http/ngx_http_upstream.c

@@ -1222,9 +1222,13 @@ ngx_http_upstream_check_broken_connection(ngx_http_request_t *r,
 
 #if (NGX_HAVE_EPOLLRDHUP)
 
-    if ((ngx_event_flags & NGX_USE_EPOLL_EVENT) && ev->pending_eof) {
+    if ((ngx_event_flags & NGX_USE_EPOLL_EVENT) && ngx_use_epoll_rdhup) {
         socklen_t  len;
 
+        if (!ev->pending_eof) {
+            return;
+        }
+
         ev->eof = 1;
         c->error = 1;
 

src/os/unix/ngx_readv_chain.c

@@ -51,6 +51,20 @@ ngx_readv_chain(ngx_connection_t *c, ngx_chain_t *chain, off_t limit)
         }
     }
 
+#endif
+
+#if (NGX_HAVE_EPOLLRDHUP)
+
+    if (ngx_event_flags & NGX_USE_EPOLL_EVENT) {
+        ngx_log_debug2(NGX_LOG_DEBUG_EVENT, c->log, 0,
+                       "readv: eof:%d, avail:%d",
+                       rev->pending_eof, rev->available);
+
+        if (!rev->available && !rev->pending_eof) {
+            return NGX_AGAIN;
+        }
+    }
+
 #endif
 
     prev = NULL;
@@ -149,6 +163,24 @@ ngx_readv_chain(ngx_connection_t *c, ngx_chain_t *chain, off_t limit)
                 return n;
             }
 
+#endif
+
+#if (NGX_HAVE_EPOLLRDHUP)
+
+            if ((ngx_event_flags & NGX_USE_EPOLL_EVENT)
+                && ngx_use_epoll_rdhup)
+            {
+                if (n < size) {
+                    if (!rev->pending_eof) {
+                        rev->ready = 0;
+                    }
+
+                    rev->available = 0;
+                }
+
+                return n;
+            }
+
 #endif
 
             if (n < size && !(ngx_event_flags & NGX_USE_GREEDY_EVENT)) {

src/os/unix/ngx_recv.c

@@ -48,6 +48,21 @@ ngx_unix_recv(ngx_connection_t *c, u_char *buf, size_t size)
         }
     }
 
+#endif
+
+#if (NGX_HAVE_EPOLLRDHUP)
+
+    if (ngx_event_flags & NGX_USE_EPOLL_EVENT) {
+        ngx_log_debug2(NGX_LOG_DEBUG_EVENT, c->log, 0,
+                       "recv: eof:%d, avail:%d",
+                       rev->pending_eof, rev->available);
+
+        if (!rev->available && !rev->pending_eof) {
+            rev->ready = 0;
+            return NGX_AGAIN;
+        }
+    }
+
 #endif
 
     do {
@@ -99,6 +114,24 @@ ngx_unix_recv(ngx_connection_t *c, u_char *buf, size_t size)
                 return n;
             }
 
+#endif
+
+#if (NGX_HAVE_EPOLLRDHUP)
+
+            if ((ngx_event_flags & NGX_USE_EPOLL_EVENT)
+                && ngx_use_epoll_rdhup)
+            {
+                if ((size_t) n < size) {
+                    if (!rev->pending_eof) {
+                        rev->ready = 0;
+                    }
+
+                    rev->available = 0;
+                }
+
+                return n;
+            }
+
 #endif
 
             if ((size_t) n < size