]> git.ipfire.org Git - thirdparty/apache/httpd.git/commitdiff
mod_http2: new bucket beams for tranporting buckets across threads without buffer...
authorStefan Eissing <icing@apache.org>
Fri, 15 Apr 2016 13:50:46 +0000 (13:50 +0000)
committerStefan Eissing <icing@apache.org>
Fri, 15 Apr 2016 13:50:46 +0000 (13:50 +0000)
git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@1739303 13f79535-47bb-0310-9956-ffa450edef68

40 files changed:
CHANGES
CMakeLists.txt
modules/http2/NWGNUmod_http2
modules/http2/config2.m4
modules/http2/h2_bucket_beam.c [new file with mode: 0644]
modules/http2/h2_bucket_beam.h [new file with mode: 0644]
modules/http2/h2_bucket_eos.c
modules/http2/h2_filter.c
modules/http2/h2_filter.h
modules/http2/h2_from_h1.c
modules/http2/h2_int_queue.c [deleted file]
modules/http2/h2_int_queue.h [deleted file]
modules/http2/h2_io.c
modules/http2/h2_io.h
modules/http2/h2_io_set.c [deleted file]
modules/http2/h2_io_set.h [deleted file]
modules/http2/h2_mplx.c
modules/http2/h2_mplx.h
modules/http2/h2_ngn_shed.c
modules/http2/h2_proxy_session.c
modules/http2/h2_proxy_session.h
modules/http2/h2_session.c
modules/http2/h2_session.h
modules/http2/h2_stream.c
modules/http2/h2_stream.h
modules/http2/h2_task.c
modules/http2/h2_task.h
modules/http2/h2_task_input.c [deleted file]
modules/http2/h2_task_input.h [deleted file]
modules/http2/h2_task_output.c [deleted file]
modules/http2/h2_task_output.h [deleted file]
modules/http2/h2_util.c
modules/http2/h2_util.h
modules/http2/h2_version.h
modules/http2/h2_worker.c
modules/http2/h2_worker.h
modules/http2/h2_workers.c
modules/http2/mod_http2.c
modules/http2/mod_http2.dsp
modules/http2/mod_proxy_http2.c

diff --git a/CHANGES b/CHANGES
index 4506788baf7cfcc5211ee973164f902f47c05ca4..66d8811832cf9008a9f98d4cdf5cefdc580587c7 100644 (file)
--- a/CHANGES
+++ b/CHANGES
@@ -1,6 +1,9 @@
                                                          -*- coding: utf-8 -*-
 Changes with Apache 2.5.0
 
+  *) mod_http2: new "bucket beam" technology to transport buckets across
+     threads without buffer copy.
+     
   *) http: Respond with "408 Request Timeout" when a timeout occurs while
      reading the request body.  [Yann Ylavic]
 
index e68076630e55b1b9b512a08abbb95d96556a73b7..b33f11caaba1f56f8b03775756e1cf0856fc3a0b 100644 (file)
@@ -403,15 +403,13 @@ SET(mod_http2_extra_sources
   modules/http2/h2_conn.c            modules/http2/h2_conn_io.c
   modules/http2/h2_ctx.c             modules/http2/h2_filter.c
   modules/http2/h2_from_h1.c         modules/http2/h2_h2.c
-  modules/http2/h2_io.c              modules/http2/h2_io_set.c
+  modules/http2/h2_io.c              modules/http2/h2_bucket_beam.c
   modules/http2/h2_mplx.c            modules/http2/h2_push.c
   modules/http2/h2_request.c         modules/http2/h2_response.c
   modules/http2/h2_session.c         modules/http2/h2_stream.c 
   modules/http2/h2_switch.c          modules/http2/h2_ngn_shed.c 
-  modules/http2/h2_task.c            modules/http2/h2_task_input.c
-  modules/http2/h2_task_output.c     modules/http2/h2_int_queue.c
-  modules/http2/h2_util.c            modules/http2/h2_worker.c
-  modules/http2/h2_workers.c
+  modules/http2/h2_task.c            modules/http2/h2_util.c
+  modules/http2/h2_worker.c          modules/http2/h2_workers.c
 )
 SET(mod_ldap_extra_defines           LDAP_DECLARE_EXPORT)
 SET(mod_ldap_extra_libs              wldap32)
index da470ecdc0af201e6f68f733242a56d6f1fe050e..dd855d172af195915e321116be25c142f145ee23 100644 (file)
@@ -185,6 +185,7 @@ TARGET_lib = \
 #
 FILES_nlm_objs = \
        $(OBJDIR)/h2_alt_svc.o \
+       $(OBJDIR)/h2_bucket_beam.o \
        $(OBJDIR)/h2_bucket_eoc.o \
        $(OBJDIR)/h2_bucket_eos.o \
        $(OBJDIR)/h2_config.o \
@@ -194,9 +195,7 @@ FILES_nlm_objs = \
        $(OBJDIR)/h2_filter.o \
        $(OBJDIR)/h2_from_h1.o \
        $(OBJDIR)/h2_h2.o \
-       $(OBJDIR)/h2_int_queue.o \
        $(OBJDIR)/h2_io.o \
-       $(OBJDIR)/h2_io_set.o \
        $(OBJDIR)/h2_mplx.o \
        $(OBJDIR)/h2_ngn_shed.o \
        $(OBJDIR)/h2_push.o \
@@ -206,8 +205,6 @@ FILES_nlm_objs = \
        $(OBJDIR)/h2_stream.o \
        $(OBJDIR)/h2_switch.o \
        $(OBJDIR)/h2_task.o \
-       $(OBJDIR)/h2_task_input.o \
-       $(OBJDIR)/h2_task_output.o \
        $(OBJDIR)/h2_util.o \
        $(OBJDIR)/h2_worker.o \
        $(OBJDIR)/h2_workers.o \
index 1515c4dbcf55f8da9970bd25fcea5df6d5ab6909..b4ba6da98ac16c3270198c82e9362e1e2be55640 100644 (file)
@@ -20,6 +20,7 @@ dnl #  list of module object files
 http2_objs="dnl
 mod_http2.lo dnl
 h2_alt_svc.lo dnl
+h2_bucket_beam.lo dnl
 h2_bucket_eoc.lo dnl
 h2_bucket_eos.lo dnl
 h2_config.lo dnl
@@ -29,9 +30,7 @@ h2_ctx.lo dnl
 h2_filter.lo dnl
 h2_from_h1.lo dnl
 h2_h2.lo dnl
-h2_int_queue.lo dnl
 h2_io.lo dnl
-h2_io_set.lo dnl
 h2_mplx.lo dnl
 h2_ngn_shed.lo dnl
 h2_push.lo dnl
@@ -41,8 +40,6 @@ h2_session.lo dnl
 h2_stream.lo dnl
 h2_switch.lo dnl
 h2_task.lo dnl
-h2_task_input.lo dnl
-h2_task_output.lo dnl
 h2_util.lo dnl
 h2_worker.lo dnl
 h2_workers.lo dnl
@@ -209,7 +206,6 @@ is usually linked shared and requires loading. ], $http2_objs, , most, [
 dnl #  list of module object files
 proxy_http2_objs="dnl
 mod_proxy_http2.lo dnl
-h2_int_queue.lo dnl
 h2_proxy_session.lo dnl
 h2_request.lo dnl
 h2_util.lo dnl
diff --git a/modules/http2/h2_bucket_beam.c b/modules/http2/h2_bucket_beam.c
new file mode 100644 (file)
index 0000000..5406176
--- /dev/null
@@ -0,0 +1,850 @@
+/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <apr_lib.h>
+#include <apr_strings.h>
+#include <apr_time.h>
+#include <apr_buckets.h>
+#include <apr_thread_mutex.h>
+#include <apr_thread_cond.h>
+
+#include <httpd.h>
+
+#include "h2_util.h"
+#include "h2_bucket_beam.h"
+
+static void h2_beam_emitted(h2_bucket_beam *beam, apr_bucket *bred);
+
+/*******************************************************************************
+ * beam bucket with reference to beam and bucket it represents
+ ******************************************************************************/
+
+extern const apr_bucket_type_t h2_bucket_type_beam;
+
+#define H2_BUCKET_IS_BEAM(e)     (e->type == &h2_bucket_type_beam)
+
+typedef struct {
+    apr_bucket_refcount refcount;
+    h2_bucket_beam *beam;
+    apr_bucket *bred;
+} h2_beam_bucket;
+
+static const char Dummy = '\0';
+
+static apr_status_t beam_bucket_read(apr_bucket *b, const char **str, 
+                                     apr_size_t *len, apr_read_type_e block)
+{
+    h2_beam_bucket *d = b->data;
+    if (d->bred) {
+        const char *data;
+        apr_status_t status = apr_bucket_read(d->bred, &data, len, block);
+        if (status == APR_SUCCESS) {
+            *str = data + b->start;
+            *len = b->length;
+        }
+        return status;
+    }
+    *str = &Dummy;
+    *len = 0;
+    return APR_SUCCESS;
+}
+
+static void beam_bucket_destroy(void *data)
+{
+    h2_beam_bucket *d = data;
+
+    if (apr_bucket_shared_destroy(d)) {
+        if (d->bred) {
+            h2_beam_emitted(d->beam, d->bred);
+        }
+        apr_bucket_free(d);
+    }
+}
+
+static apr_bucket * h2_beam_bucket_make(apr_bucket *b, 
+                                        h2_bucket_beam *beam,
+                                        apr_bucket *bred)
+{
+    h2_beam_bucket *d;
+
+    d = apr_bucket_alloc(sizeof(*d), b->list);
+    d->beam = beam;
+    d->bred = bred;
+
+    b = apr_bucket_shared_make(b, d, 0, bred? bred->length : 0);
+    b->type = &h2_bucket_type_beam;
+
+    return b;
+}
+
+static apr_bucket *h2_beam_bucket_create(h2_bucket_beam *beam,
+                                         apr_bucket *bred,
+                                         apr_bucket_alloc_t *list)
+{
+    apr_bucket *b = apr_bucket_alloc(sizeof(*b), list);
+
+    APR_BUCKET_INIT(b);
+    b->free = apr_bucket_free;
+    b->list = list;
+    return h2_beam_bucket_make(b, beam, bred);
+}
+
+APU_DECLARE_DATA const apr_bucket_type_t h2_bucket_type_beam = {
+    "BEAM", 5, APR_BUCKET_DATA,
+    beam_bucket_destroy,
+    beam_bucket_read,
+    apr_bucket_setaside_noop,
+    apr_bucket_shared_split,
+    apr_bucket_shared_copy
+};
+
+/*******************************************************************************
+ * h2_blist, a brigade without allocations
+ ******************************************************************************/
+apr_size_t h2_util_bl_print(char *buffer, apr_size_t bmax, 
+                            const char *tag, const char *sep, 
+                            h2_blist *bl)
+{
+    apr_size_t off = 0;
+    const char *sp = "";
+    apr_bucket *b;
+    
+    if (bl) {
+        memset(buffer, 0, bmax--);
+        off += apr_snprintf(buffer+off, bmax-off, "%s(", tag);
+        for (b = H2_BLIST_FIRST(bl); 
+             bmax && (b != H2_BLIST_SENTINEL(bl));
+             b = APR_BUCKET_NEXT(b)) {
+            
+            off += h2_util_bucket_print(buffer+off, bmax-off, b, sp);
+            sp = " ";
+        }
+        off += apr_snprintf(buffer+off, bmax-off, ")%s", sep);
+    }
+    else {
+        off += apr_snprintf(buffer+off, bmax-off, "%s(null)%s", tag, sep);
+    }
+    return off;
+}
+
+
+
+/*******************************************************************************
+ * bucket beam that can transport buckets across threads
+ ******************************************************************************/
+
+static apr_status_t enter_yellow(h2_bucket_beam *beam, 
+                                 apr_thread_mutex_t **plock, int *pacquired)
+{
+    if (beam->m_enter) {
+        return beam->m_enter(beam->m_ctx, plock, pacquired);
+    }
+    *plock = NULL;
+    *pacquired = 0;
+    return APR_SUCCESS;
+}
+
+static void leave_yellow(h2_bucket_beam *beam, 
+                         apr_thread_mutex_t *lock, int acquired)
+{
+    if (acquired && beam->m_leave) {
+        beam->m_leave(beam->m_ctx, lock, acquired);
+    }
+}
+
+static apr_off_t calc_buffered(h2_bucket_beam *beam)
+{
+    apr_off_t len = 0;
+    apr_bucket *b;
+    for (b = H2_BLIST_FIRST(&beam->red); 
+         b != H2_BLIST_SENTINEL(&beam->red);
+         b = APR_BUCKET_NEXT(b)) {
+        if (b->length == ((apr_size_t)-1)) {
+            /* do not count */
+        }
+        else if (APR_BUCKET_IS_FILE(b)) {
+            /* if unread, has no real mem footprint. how to test? */
+        }
+        else {
+            len += b->length;
+        }
+    }
+    return len;
+}
+
+static void r_purge_reds(h2_bucket_beam *beam)
+{
+    apr_bucket *bred;
+    /* delete all red buckets in purge brigade, needs to be called
+     * from red thread only */
+    while (!H2_BLIST_EMPTY(&beam->purge)) {
+        bred = H2_BLIST_FIRST(&beam->purge);
+        apr_bucket_delete(bred);
+    }
+}
+
+static apr_size_t calc_space_left(h2_bucket_beam *beam)
+{
+    if (beam->max_buf_size > 0) {
+        apr_off_t len = calc_buffered(beam);
+        return (beam->max_buf_size > len? (beam->max_buf_size - len) : 0);
+    }
+    return APR_SIZE_MAX;
+}
+
+static apr_status_t wait_cond(h2_bucket_beam *beam, apr_thread_mutex_t *lock)
+{
+    if (beam->timeout > 0) {
+        return apr_thread_cond_timedwait(beam->m_cond, lock, beam->timeout);
+    }
+    else {
+        return apr_thread_cond_wait(beam->m_cond, lock);
+    }
+}
+
+static apr_status_t r_wait_space(h2_bucket_beam *beam, apr_read_type_e block,
+                                 apr_thread_mutex_t *lock, apr_off_t *premain) 
+{
+    *premain = calc_space_left(beam);
+    while (!beam->aborted && *premain <= 0 
+           && (block == APR_BLOCK_READ) && lock) {
+        apr_status_t status = wait_cond(beam, lock);
+        if (APR_STATUS_IS_TIMEUP(status)) {
+            return status;
+        }
+        r_purge_reds(beam);
+        *premain = calc_space_left(beam);
+    }
+    return beam->aborted? APR_ECONNABORTED : APR_SUCCESS;
+}
+
+static void h2_beam_prep_purge(h2_bucket_beam *beam, apr_bucket *bred)
+{
+    APR_BUCKET_REMOVE(bred);
+    H2_BLIST_INSERT_TAIL(&beam->purge, bred);
+}
+
+static void h2_beam_emitted(h2_bucket_beam *beam, apr_bucket *bred)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        /* even when beam buckets are split, only the one where
+         * refcount drops to 0 will call us */
+        --beam->live_beam_buckets;
+        /* invoked from green thread, the last beam bucket for the red
+         * bucket bred is about to be destroyed.
+         * remove it from the hold, where it should be now */
+        h2_beam_prep_purge(beam, bred);
+        /* notify anyone waiting on space to become available */
+        if (!lock) {
+            r_purge_reds(beam);
+        }
+        else if (beam->m_cond) {
+            apr_thread_cond_broadcast(beam->m_cond);
+        }
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+static void report_consumption(h2_bucket_beam *beam)
+{
+    if (beam->consumed_fn && (beam->received_bytes != beam->reported_bytes)) {
+        beam->consumed_fn(beam->consumed_ctx, beam, 
+                          beam->received_bytes - beam->reported_bytes);
+        beam->reported_bytes = beam->received_bytes;
+    }
+}
+
+static void h2_blist_cleanup(h2_blist *bl)
+{
+    apr_bucket *e;
+
+    while (!H2_BLIST_EMPTY(bl)) {
+        e = H2_BLIST_FIRST(bl);
+        apr_bucket_delete(e);
+    }
+}
+
+static apr_status_t beam_cleanup(void *data)
+{
+    h2_bucket_beam *beam = data;
+
+    AP_DEBUG_ASSERT(beam->live_beam_buckets == 0);
+    h2_blist_cleanup(&beam->red);
+    h2_blist_cleanup(&beam->purge);
+    h2_blist_cleanup(&beam->hold);
+    return APR_SUCCESS;
+}
+
+apr_status_t h2_beam_destroy(h2_bucket_beam *beam)
+{
+    apr_pool_cleanup_kill(beam->life_pool, beam, beam_cleanup);
+    return beam_cleanup(beam);
+}
+
+apr_status_t h2_beam_create(h2_bucket_beam **pbeam, apr_pool_t *life_pool, 
+                            int id, const char *tag, 
+                            apr_size_t max_buf_size)
+{
+    h2_bucket_beam *beam;
+    apr_status_t status = APR_SUCCESS;
+    
+    beam = apr_pcalloc(life_pool, sizeof(*beam));
+    if (!beam) {
+        return APR_ENOMEM;
+    }
+
+    beam->id = id;
+    beam->tag = tag;
+    H2_BLIST_INIT(&beam->red);
+    H2_BLIST_INIT(&beam->hold);
+    H2_BLIST_INIT(&beam->purge);
+    beam->life_pool = life_pool;
+    beam->max_buf_size = max_buf_size;
+
+    apr_pool_cleanup_register(life_pool, beam, beam_cleanup, 
+                              apr_pool_cleanup_null);
+    *pbeam = beam;
+    
+    return status;
+}
+
+void h2_beam_buffer_size_set(h2_bucket_beam *beam, apr_size_t buffer_size)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        beam->max_buf_size = buffer_size;
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+apr_size_t h2_beam_buffer_size_get(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    apr_size_t buffer_size = 0;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        buffer_size = beam->max_buf_size;
+        leave_yellow(beam, lock, acquired);
+    }
+    return buffer_size;
+}
+
+void h2_beam_mutex_set(h2_bucket_beam *beam, 
+                       h2_beam_mutex_enter m_enter,
+                       h2_beam_mutex_leave m_leave,
+                       apr_thread_cond_t *cond,
+                       void *m_ctx)
+{
+    apr_thread_mutex_t *lock;
+    h2_beam_mutex_leave *prev_leave;
+    void *prev_ctx;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        prev_ctx = beam->m_ctx;
+        prev_leave = beam->m_leave;
+        beam->m_enter = m_enter;
+        beam->m_leave = m_leave;
+        beam->m_ctx   = m_ctx;
+        beam->m_cond  = cond;
+        if (acquired && prev_leave) {
+            /* special tactics when NULLing a lock */
+            prev_leave(prev_ctx, lock, acquired);
+        }
+    }
+}
+
+void h2_beam_timeout_set(h2_bucket_beam *beam, apr_interval_time_t timeout)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        beam->timeout = timeout;
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+apr_interval_time_t h2_beam_timeout_get(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    apr_interval_time_t timeout = 0;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        timeout = beam->timeout;
+        leave_yellow(beam, lock, acquired);
+    }
+    return timeout;
+}
+
+void h2_beam_abort(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        r_purge_reds(beam);
+        h2_blist_cleanup(&beam->red);
+        beam->aborted = 1;
+        report_consumption(beam);
+        if (beam->m_cond) {
+            apr_thread_cond_broadcast(beam->m_cond);
+        }
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+static apr_status_t beam_close(h2_bucket_beam *beam)
+{
+    if (!beam->closed) {
+        beam->closed = 1;
+        if (beam->m_cond) {
+            apr_thread_cond_broadcast(beam->m_cond);
+        }
+    }
+    return APR_SUCCESS;
+}
+
+apr_status_t h2_beam_close(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        r_purge_reds(beam);
+        beam_close(beam);
+        report_consumption(beam);
+        leave_yellow(beam, lock, acquired);
+    }
+    return beam->aborted? APR_ECONNABORTED : APR_SUCCESS;
+}
+
+void h2_beam_shutdown(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        r_purge_reds(beam);
+        h2_blist_cleanup(&beam->red);
+        beam_close(beam);
+        report_consumption(beam);
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+void h2_beam_reset(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        beam_cleanup(beam);
+        beam->closed = beam->close_sent = 0;
+        beam->sent_bytes = beam->received_bytes = beam->reported_bytes = 0;
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+static apr_status_t append_bucket(h2_bucket_beam *beam, 
+                                  apr_bucket *bred,
+                                  apr_read_type_e block,
+                                  apr_pool_t *pool,
+                                  apr_thread_mutex_t *lock)
+{
+    const char *data;
+    apr_size_t len;
+    apr_off_t space_left = 0;
+    apr_status_t status;
+    
+    if (APR_BUCKET_IS_METADATA(bred)) {
+        if (APR_BUCKET_IS_EOS(bred)) {
+            beam->closed = 1;
+        }
+        APR_BUCKET_REMOVE(bred);
+        H2_BLIST_INSERT_TAIL(&beam->red, bred);
+        return APR_SUCCESS;
+    }
+    else if (APR_BUCKET_IS_FILE(bred)) {
+        /* file bucket lengths do not really count */
+    }
+    else {
+        space_left = calc_space_left(beam);
+        if (space_left > 0 && bred->length == ((apr_size_t)-1)) {
+            const char *data;
+            status = apr_bucket_read(bred, &data, &len, APR_BLOCK_READ);
+            if (status != APR_SUCCESS) {
+                return status;
+            }
+        }
+        
+        if (space_left < bred->length) {
+            status = r_wait_space(beam, block, lock, &space_left);
+            if (status != APR_SUCCESS) {
+                return status;
+            }
+            if (space_left <= 0) {
+                return APR_EAGAIN;
+            }
+        }
+        /* space available, maybe need bucket split */
+    }
+    
+
+    /* The fundamental problem is that reading a red bucket from
+     * a green thread is a total NO GO, because the bucket might use
+     * its pool/bucket_alloc from a foreign thread and that will
+     * corrupt. */
+    status = APR_ENOTIMPL;
+    if (beam->closed && bred->length > 0) {
+        status = APR_EOF;
+    }
+    else if (APR_BUCKET_IS_TRANSIENT(bred)) {
+        /* this takes care of transient buckets and converts them
+         * into heap ones. Other bucket types might or might not be
+         * affected by this. */
+        status = apr_bucket_setaside(bred, pool);
+    }
+    else if (APR_BUCKET_IS_HEAP(bred) || APR_BUCKET_IS_POOL(bred)) {
+        /* For heap/pool buckets read from a green thread is fine. The
+         * data will be there and live until the bucket itself is
+         * destroyed. */
+        status = APR_SUCCESS;
+    }
+    else if (APR_BUCKET_IS_FILE(bred)) {
+        /* For file buckets the problem is their internal readpool that
+         * is used on the first read to allocate buffer/mmap.
+         * Since setting aside a file bucket will de-register the
+         * file cleanup function from the previous pool, we need to
+         * call that from a red thread. Do it now and make our
+         * yellow pool the owner. 
+         * Additionally, we allow callbacks to prevent beaming file
+         * handles across. The use case for this is to limit the number 
+         * of open file handles and rather use a less efficient beam
+         * transport. */ 
+        apr_file_t *fd = ((apr_bucket_file *)bred->data)->fd;
+        int can_beam = 1;
+        if (beam->last_beamed != fd && beam->can_beam_fn) {
+            can_beam = beam->can_beam_fn(beam->can_beam_ctx, beam, fd);
+        }
+        if (can_beam) {
+            beam->last_beamed = fd;
+            status = apr_bucket_setaside(bred, pool);
+        }
+    }
+    
+    if (status == APR_ENOTIMPL) {
+        /* we have no knowledge about the internals of this bucket,
+         * but on read, it needs to make the data available somehow.
+         * So we do this while still in a red thread. The data will
+         * live at least os long as the red bucket itself. */
+        if (space_left < APR_BUCKET_BUFF_SIZE) {
+            space_left = APR_BUCKET_BUFF_SIZE;
+        }
+        if (space_left < bred->length) {
+            apr_bucket_split(bred, space_left);
+        }
+        status = apr_bucket_read(bred, &data, &len, APR_BLOCK_READ);
+        if (status == APR_SUCCESS) {
+            status = apr_bucket_setaside(bred, pool);
+        }
+    }
+    
+    if (status != APR_SUCCESS && status != APR_ENOTIMPL) {
+        return status;
+    }
+    
+    APR_BUCKET_REMOVE(bred);
+    H2_BLIST_INSERT_TAIL(&beam->red, bred);
+    beam->sent_bytes += bred->length;
+    
+    return APR_SUCCESS;
+}
+
+apr_status_t h2_beam_send(h2_bucket_beam *beam, 
+                          apr_bucket_brigade *red_brigade, 
+                          apr_read_type_e block)
+{
+    apr_thread_mutex_t *lock;
+    apr_bucket *bred;
+    apr_status_t status = APR_SUCCESS;
+    int acquired;
+
+    /* Called from the red thread to add buckets to the beam */
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        r_purge_reds(beam);
+        
+        if (beam->aborted) {
+            status = APR_ECONNABORTED;
+        }
+        else if (red_brigade) {
+            while (!APR_BRIGADE_EMPTY(red_brigade)
+                   && status == APR_SUCCESS) {
+                bred = APR_BRIGADE_FIRST(red_brigade);
+                status = append_bucket(beam, bred, block, red_brigade->p, lock);
+            }
+            if (beam->m_cond) {
+                apr_thread_cond_broadcast(beam->m_cond);
+            }
+        }
+        report_consumption(beam);
+        leave_yellow(beam, lock, acquired);
+    }
+    return status;
+}
+
+apr_status_t h2_beam_receive(h2_bucket_beam *beam, 
+                             apr_bucket_brigade *bb, 
+                             apr_read_type_e block,
+                             apr_off_t readbytes)
+{
+    apr_thread_mutex_t *lock;
+    apr_bucket *bred, *bgreen;
+    int acquired, transferred = 0;
+    apr_status_t status = APR_SUCCESS;
+    apr_off_t remain = readbytes;
+    
+    /* Called from the green thread to take buckets from the beam */
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+transfer:
+        if (beam->aborted) {
+            status = APR_ECONNABORTED;
+            goto leave;
+        }
+
+        /* transfer enough buckets from our green brigade, if we have one */
+        while (beam->green
+               && !APR_BRIGADE_EMPTY(beam->green)
+               && (readbytes <= 0 || remain >= 0)) {
+            bgreen = APR_BRIGADE_FIRST(beam->green);
+            if (readbytes > 0 && bgreen->length > 0 && remain <= 0) {
+                break;
+            }            
+            APR_BUCKET_REMOVE(bgreen);
+            APR_BRIGADE_INSERT_TAIL(bb, bgreen);
+            remain -= bgreen->length;
+            ++transferred;
+        }
+
+        /* transfer from our red brigade, transforming red buckets to
+         * green ones until we have enough */
+        while (!H2_BLIST_EMPTY(&beam->red) && (readbytes <= 0 || remain >= 0)) {
+            bred = H2_BLIST_FIRST(&beam->red);
+            bgreen = NULL;
+            
+            if (readbytes > 0 && bred->length > 0 && remain <= 0) {
+                break;
+            }
+                        
+            if (APR_BUCKET_IS_METADATA(bred)) {
+                if (APR_BUCKET_IS_EOS(bred)) {
+                    beam->close_sent = 1;
+                    bgreen = apr_bucket_eos_create(bb->bucket_alloc);
+                }
+                else if (APR_BUCKET_IS_FLUSH(bred)) {
+                    bgreen = apr_bucket_flush_create(bb->bucket_alloc);
+                }
+                else {
+                    /* put red into hold, no green sent out */
+                }
+            }
+            else if (APR_BUCKET_IS_FILE(bred)) {
+                /* This is set aside into the target brigade pool so that 
+                 * any read operation messes with that pool and not 
+                 * the red one. */
+                apr_bucket_file *f = (apr_bucket_file *)bred->data;
+                apr_file_t *fd = f->fd;
+                int setaside = (f->readpool != bb->p);
+                
+                if (setaside) {
+                    status = apr_file_setaside(&fd, fd, bb->p);
+                    if (status != APR_SUCCESS) {
+                        goto leave;
+                    }
+                    ++beam->files_beamed;
+                }
+                apr_brigade_insert_file(bb, fd, bred->start, bred->length, 
+                                        bb->p);
+                remain -= bred->length;
+                ++transferred;
+            }
+            else {
+                /* create a "green" standin bucket. we took care about the
+                 * underlying red bucket and its data when we placed it into
+                 * the red brigade.
+                 * the beam bucket will notify us on destruction that bred is
+                 * no longer needed. */
+                bgreen = h2_beam_bucket_create(beam, bred, bb->bucket_alloc);
+                ++beam->live_beam_buckets;
+            }
+            
+            /* Place the red bucket into our hold, to be destroyed when no
+             * green bucket references it any more. */
+            APR_BUCKET_REMOVE(bred);
+            H2_BLIST_INSERT_TAIL(&beam->hold, bred);
+            beam->received_bytes += bred->length;
+            if (bgreen) {
+                APR_BRIGADE_INSERT_TAIL(bb, bgreen);
+                remain -= bgreen->length;
+                ++transferred;
+            }
+        }
+
+        if (readbytes > 0 && remain < 0) {
+            /* too much, put some back */
+            remain = readbytes;
+            for (bgreen = APR_BRIGADE_FIRST(bb);
+                 bgreen != APR_BRIGADE_SENTINEL(bb);
+                 bgreen = APR_BUCKET_NEXT(bgreen)) {
+                 remain -= bgreen->length;
+                 if (remain < 0) {
+                     apr_bucket_split(bgreen, bgreen->length+remain);
+                     beam->green = apr_brigade_split_ex(bb, 
+                                                        APR_BUCKET_NEXT(bgreen), 
+                                                        beam->green);
+                     break;
+                 }
+            }
+        }
+                
+        if (transferred) {
+            status = APR_SUCCESS;
+        }
+        else if (beam->closed) {
+            if (!beam->close_sent) {
+                apr_bucket *b = apr_bucket_eos_create(bb->bucket_alloc);
+                APR_BRIGADE_INSERT_TAIL(bb, b);
+                beam->close_sent = 1;
+                status = APR_SUCCESS;
+            }
+            else {
+                status = APR_EOF;
+            }
+        }
+        else if (block == APR_BLOCK_READ && lock && beam->m_cond) {
+            status = wait_cond(beam, lock);
+            if (status != APR_SUCCESS) {
+                goto leave;
+            }
+            goto transfer;
+        }
+        else {
+            status = APR_EAGAIN;
+        }
+leave:        
+        leave_yellow(beam, lock, acquired);
+    }
+    return status;
+}
+
+void h2_beam_on_consumed(h2_bucket_beam *beam, 
+                         h2_beam_consumed_callback *cb, void *ctx)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        beam->consumed_fn = cb;
+        beam->consumed_ctx = ctx;
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+void h2_beam_on_file_beam(h2_bucket_beam *beam, 
+                          h2_beam_can_beam_callback *cb, void *ctx)
+{
+    apr_thread_mutex_t *lock;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        beam->can_beam_fn = cb;
+        beam->can_beam_ctx = ctx;
+        leave_yellow(beam, lock, acquired);
+    }
+}
+
+
+apr_off_t h2_beam_get_buffered(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    apr_bucket *b;
+    apr_off_t l = 0;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        for (b = H2_BLIST_FIRST(&beam->red); 
+            b != H2_BLIST_SENTINEL(&beam->red);
+            b = APR_BUCKET_NEXT(b)) {
+            /* should all have determinate length */
+            l += b->length;
+        }
+        leave_yellow(beam, lock, acquired);
+    }
+    return l;
+}
+
+int h2_beam_empty(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int empty = 1;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        empty = (H2_BLIST_EMPTY(&beam->red) 
+                 && (!beam->green || APR_BRIGADE_EMPTY(beam->green)));
+        leave_yellow(beam, lock, acquired);
+    }
+    return empty;
+}
+
+int h2_beam_closed(h2_bucket_beam *beam)
+{
+    return beam->closed;
+}
+
+int h2_beam_was_received(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    int happend = 0;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        happend = (beam->received_bytes > 0);
+        leave_yellow(beam, lock, acquired);
+    }
+    return happend;
+}
+
+apr_size_t h2_beam_get_files_beamed(h2_bucket_beam *beam)
+{
+    apr_thread_mutex_t *lock;
+    apr_size_t n = 0;
+    int acquired;
+    
+    if (enter_yellow(beam, &lock, &acquired) == APR_SUCCESS) {
+        n = beam->files_beamed;
+        leave_yellow(beam, lock, acquired);
+    }
+    return n;
+}
+
diff --git a/modules/http2/h2_bucket_beam.h b/modules/http2/h2_bucket_beam.h
new file mode 100644 (file)
index 0000000..45d98b2
--- /dev/null
@@ -0,0 +1,297 @@
+/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef h2_bucket_beam_h
+#define h2_bucket_beam_h
+
+struct apr_thread_mutex_t;
+struct apr_thread_cond_t;
+
+/*******************************************************************************
+ * apr_bucket list without bells and whistles
+ ******************************************************************************/
+/**
+ * h2_blist can hold a list of buckets just like apr_bucket_brigade, but
+ * does not to any allocations or related features.
+ */
+typedef struct {
+    APR_RING_HEAD(h2_bucket_list, apr_bucket) list;
+} h2_blist;
+
+#define H2_BLIST_INIT(b)        APR_RING_INIT(&(b)->list, apr_bucket, link);
+#define H2_BLIST_SENTINEL(b)    APR_RING_SENTINEL(&(b)->list, apr_bucket, link)
+#define H2_BLIST_EMPTY(b)       APR_RING_EMPTY(&(b)->list, apr_bucket, link)
+#define H2_BLIST_FIRST(b)       APR_RING_FIRST(&(b)->list)
+#define H2_BLIST_LAST(b)       APR_RING_LAST(&(b)->list)
+#define H2_BLIST_INSERT_HEAD(b, e) do {                                \
+       apr_bucket *ap__b = (e);                                        \
+       APR_RING_INSERT_HEAD(&(b)->list, ap__b, apr_bucket, link);      \
+    } while (0)
+#define H2_BLIST_INSERT_TAIL(b, e) do {                                \
+       apr_bucket *ap__b = (e);                                        \
+       APR_RING_INSERT_TAIL(&(b)->list, ap__b, apr_bucket, link);      \
+    } while (0)
+#define H2_BLIST_CONCAT(a, b) do {                                     \
+        APR_RING_CONCAT(&(a)->list, &(b)->list, apr_bucket, link);     \
+    } while (0)
+#define H2_BLIST_PREPEND(a, b) do {                                    \
+        APR_RING_PREPEND(&(a)->list, &(b)->list, apr_bucket, link);    \
+    } while (0)
+
+/**
+ * Print the buckets in the list into the buffer (type and lengths).
+ * @param buffer the buffer to print into
+ * @param bmax max number of characters to place in buffer, incl. trailing 0
+ * @param tag tag string for this bucket list
+ * @param sep separator to use
+ * @param bl the bucket list to print
+ * @return number of characters printed
+ */
+apr_size_t h2_util_bl_print(char *buffer, apr_size_t bmax, 
+                            const char *tag, const char *sep, 
+                            h2_blist *bl);
+
+/*******************************************************************************
+ * h2_bucket_beam
+ ******************************************************************************/
+
+/**
+ * A h2_bucket_beam solves the task of transferring buckets, esp. their data,
+ * across threads with zero buffer copies.
+ *
+ * When a thread, let's call it the red thread, wants to send buckets to
+ * another, the green thread, it creates a h2_bucket_beam and adds buckets
+ * via the h2_beam_send(). It gives the beam to the green thread which then
+ * can receive buckets into its own brigade via h2_beam_receive().
+ *
+ * Sending and receiving can happen concurrently, if a thread mutex is set
+ * for the beam, see h2_beam_mutex_set.
+ *
+ * The beam can limit the amount of data it accepts via the buffer_size. This
+ * can also be adjusted during its lifetime. When the beam not only gets a 
+ * mutex but als a condition variable (in h2_beam_mutex_set()), sends and
+ * receives can be done blocking. A timeout can be set for such blocks.
+ *
+ * Care needs to be taken when terminating the beam. The beam registers at
+ * the pool it was created with and will cleanup after itself. However, if
+ * received buckets do still exist, already freed memory might be accessed.
+ * The beam does a AP_DEBUG_ASSERT on this condition.
+ * 
+ * The proper way of shutting down a beam is to first make sure there are no
+ * more green buckets out there, then cleanup the beam to purge eventually
+ * still existing red buckets and then, possibly, terminate the beam itself
+ * (or the pool it was created with).
+ *
+ * The following restrictions apply to bucket transport:
+ * - only EOS and FLUSH meta buckets are copied through. All other meta buckets
+ *   are kept in the beams hold.
+ * - all kind of data buckets are transported through:
+ *   - transient buckets are converted to heap ones on send
+ *   - heap and pool buckets require no extra handling
+ *   - buckets with indeterminate length are read on send
+ *   - file buckets will transfer the file itself into a new bucket, if allowed
+ *   - all other buckets are read on send to make sure data is present
+ *
+ * This assures that when the red thread sends its red buckets, the data
+ * is made accessible while still on the red side. The red bucket then enters
+ * the beams hold storage.
+ * When the green thread calls receive, red buckets in the hold are wrapped
+ * into special beam buckets. Beam buckets on read present the data directly
+ * from the internal red one, but otherwise live on the green side. When a
+ * beam bucket gets destroyed, it notifies its beam that the corresponding
+ * red bucket from the hold may be destroyed.
+ * Since the destruction of green buckets happens in the green thread, any
+ * corresponding red bucket can not immediately be destroyed, as that would
+ * result in race conditions.
+ * Instead, the beam transfers such red buckets from the hold to the purge
+ * storage. Next time there is a call from the red side, the buckets in
+ * purge will be deleted.
+ *
+ * There are callbacks that can be registered with a beam:
+ * - a "consumed" callback that gets called on the red side with the
+ *   amount of data that has been received by the green side. The amount
+ *   is a delta from the last callback invocation. The red side can trigger
+ *   these callbacks by calling h2_beam_send() with a NULL brigade.
+ * - a "can_beam_file" callback that can prohibit the transfer of file handles
+ *   through the beam. This will cause file buckets to be read on send and
+ *   its data buffer will then be transports just like a heap bucket would.
+ *   When no callback is registered, no restrictions apply and all files are
+ *   passed through.
+ *   File handles transferred to the green side will stay there until the
+ *   receiving brigade's pool is destroyed/cleared. If the pool lives very
+ *   long or if many different files are beamed, the process might run out
+ *   of available file handles.
+ *
+ * The name "beam" of course is inspired by good old transporter
+ * technology where humans are kept inside the transporter's memory
+ * buffers until the transmission is complete. Star gates use a similar trick.
+ */
+typedef struct h2_bucket_beam h2_bucket_beam;
+
+typedef apr_status_t h2_beam_mutex_enter(void *ctx, 
+                                         struct apr_thread_mutex_t **plock, 
+                                         int *acquired);
+typedef void h2_beam_mutex_leave(void *ctx, 
+                                 struct apr_thread_mutex_t *lock, 
+                                 int acquired);
+typedef void h2_beam_consumed_callback(void *ctx, h2_bucket_beam *beam,
+                                       apr_off_t bytes);
+
+typedef int h2_beam_can_beam_callback(void *ctx, h2_bucket_beam *beam,
+                                      apr_file_t *file);
+
+struct h2_bucket_beam {
+    int id;
+    const char *tag;
+    h2_blist red;
+    h2_blist hold;
+    h2_blist purge;
+    apr_bucket_brigade *green;
+    apr_pool_t *life_pool;
+    
+    apr_size_t max_buf_size;
+    apr_size_t live_beam_buckets;
+    apr_size_t files_beamed;  /* how many file handles have been set aside */
+    apr_file_t *last_beamed;  /* last file beamed */
+    apr_off_t sent_bytes;     /* amount of bytes send */
+    apr_off_t received_bytes; /* amount of bytes received */
+    apr_off_t reported_bytes; /* amount of bytes reported as consumed */
+    
+    unsigned int aborted : 1;
+    unsigned int closed : 1;
+    unsigned int close_sent : 1;
+
+    void *m_ctx;
+    h2_beam_mutex_enter *m_enter;
+    h2_beam_mutex_leave *m_leave;
+    struct apr_thread_cond_t *m_cond;
+    apr_interval_time_t timeout;
+    
+    h2_beam_consumed_callback *consumed_fn;
+    void *consumed_ctx;
+    h2_beam_can_beam_callback *can_beam_fn;
+    void *can_beam_ctx;
+};
+
+/**
+ * Creates a new bucket beam for transfer of buckets across threads.
+ *
+ * The pool the beam is created with will be protected by the given 
+ * mutex and will be used in multiple threads. It needs a pool allocator
+ * that is only used inside that same mutex.
+ *
+ * @param pbeam will hold the created beam on return
+ * @param life_pool     pool for allocating initial structure and cleanups
+ * @param buffer_size   maximum memory footprint of buckets buffered in beam, or
+ *                      0 for no limitation
+ */
+apr_status_t h2_beam_create(h2_bucket_beam **pbeam,
+                            apr_pool_t *life_pool, 
+                            int id, const char *tag, 
+                            apr_size_t buffer_size);
+
+apr_status_t h2_beam_destroy(h2_bucket_beam *beam);
+
+/**
+ * Send buckets from the given brigade through the beam. Will hold buckets 
+ * internally as long as they have not been processed by the receiving side.
+ * All accepted buckets are removed from the given brigade. Will return with
+ * APR_EAGAIN on non-blocking sends when not all buckets could be accepted.
+ */
+apr_status_t h2_beam_send(h2_bucket_beam *beam,  
+                          apr_bucket_brigade *red_buckets, 
+                          apr_read_type_e block);
+
+/**
+ * Receive buckets from the beam into the given brigade. Will return APR_EOF
+ * when reading past an EOS bucket. Reads can be blocking until data is 
+ * available or the beam has been closed. Non-blocking calls return APR_EAGAIN
+ * if no data is available.
+ */
+apr_status_t h2_beam_receive(h2_bucket_beam *beam, 
+                             apr_bucket_brigade *green_buckets, 
+                             apr_read_type_e block,
+                             apr_off_t readbytes);
+
+void h2_beam_abort(h2_bucket_beam *beam);
+
+/**
+ * Close the beam. Does not need to be invoked if certain that an EOS bucket
+ * has been sent. 
+ */
+apr_status_t h2_beam_close(h2_bucket_beam *beam);
+
+/**
+ * Empty the buffer and close.
+ */
+void h2_beam_shutdown(h2_bucket_beam *beam);
+
+/**
+ * Reset the beam to its intial, empty state.
+ */
+void h2_beam_reset(h2_bucket_beam *beam);
+
+void h2_beam_mutex_set(h2_bucket_beam *beam, 
+                       h2_beam_mutex_enter m_enter,
+                       h2_beam_mutex_leave m_leave,
+                       struct apr_thread_cond_t *cond,
+                       void *m_ctx);
+
+/** 
+ * Set/get the timeout for blocking read/write operations. Only works
+ * if a mutex has been set for the beam.
+ */
+void h2_beam_timeout_set(h2_bucket_beam *beam, 
+                         apr_interval_time_t timeout);
+apr_interval_time_t h2_beam_timeout_get(h2_bucket_beam *beam);
+
+/**
+ * Set/get the maximum buffer size for beam data (memory footprint).
+ */
+void h2_beam_buffer_size_set(h2_bucket_beam *beam, 
+                             apr_size_t buffer_size);
+apr_size_t h2_beam_buffer_size_get(h2_bucket_beam *beam);
+
+/**
+ * Register a callback to be invoked on the red side with the
+ * amount of bytes that have been consumed by the red side, since the
+ * last callback invocation or reset.
+ * @param beam the beam to set the callback on
+ * @param cb   the callback or NULL
+ * @param ctx  the context to use in callback invocation
+ */
+void h2_beam_on_consumed(h2_bucket_beam *beam, 
+                         h2_beam_consumed_callback *cb, void *ctx);
+
+void h2_beam_on_file_beam(h2_bucket_beam *beam, 
+                          h2_beam_can_beam_callback *cb, void *ctx);
+
+/**
+ * Get the amount of bytes currently buffered in the beam (unread).
+ */
+apr_off_t h2_beam_get_buffered(h2_bucket_beam *beam);
+
+int h2_beam_closed(h2_bucket_beam *beam);
+int h2_beam_empty(h2_bucket_beam *beam);
+
+/**
+ * Return != 0 iff (some) data from the beam has been received.
+ */
+int h2_beam_was_received(h2_bucket_beam *beam);
+
+apr_size_t h2_beam_get_files_beamed(h2_bucket_beam *beam);
+
+#endif /* h2_bucket_beam_h */
index 98a0b365c689899f3ed14097a9337e8e665e540f..3a5b1a570a6327ebdc759f507b42e6a4ba6fefc6 100644 (file)
@@ -92,7 +92,7 @@ static void bucket_destroy(void *data)
     if (apr_bucket_shared_destroy(h)) {
         h2_stream *stream = h->stream;
         if (stream) {
-            h2_stream_cleanup(stream);
+            h2_stream_eos_destroy(stream);
         }
         apr_bucket_free(h);
     }
index 8bf7fbcb40ca4fcae95975e1a2b924a85e454e32..30ccb2c43ed6bba7f509e6e350b26854fad15fb8 100644 (file)
@@ -208,18 +208,27 @@ static apr_status_t bbout(apr_bucket_brigade *bb, const char *fmt, ...)
     return rv;
 }
 
-static apr_status_t h2_sos_h2_status_buffer(h2_sos *sos, apr_bucket_brigade *bb)
+static apr_status_t h2_status_stream_filter(h2_stream *stream)
 {
-    h2_stream *stream = sos->stream;
     h2_session *session = stream->session;
     h2_mplx *mplx = session->mplx;
+    conn_rec *c = session->c;
     h2_push_diary *diary;
+    apr_bucket_brigade *bb;
     apr_status_t status;
     
-    if (!bb) {
-        bb = apr_brigade_create(stream->pool, session->c->bucket_alloc);
+    if (!stream->response) {
+        return APR_EINVAL;
     }
     
+    if (!stream->buffer) {
+        stream->buffer = apr_brigade_create(stream->pool, c->bucket_alloc);
+    }
+    bb = stream->buffer;
+    
+    apr_table_unset(stream->response->headers, "Content-Length");
+    stream->response->content_length = -1;
+    
     bbout(bb, "{\n");
     bbout(bb, "  \"HTTP2\": \"on\",\n");
     bbout(bb, "  \"H2PUSH\": \"%s\",\n", h2_session_push_enabled(session)? "on" : "off");
@@ -266,57 +275,15 @@ static apr_status_t h2_sos_h2_status_buffer(h2_sos *sos, apr_bucket_brigade *bb)
     bbout(bb, "  \"bytes_sent\": %"APR_UINT64_T_FMT"\n", session->io.bytes_written);
     bbout(bb, "}\n");
     
-    return sos->prev->buffer(sos->prev, bb);
-}
-
-static apr_status_t h2_sos_h2_status_read_to(h2_sos *sos, apr_bucket_brigade *bb, 
-                                             apr_off_t *plen, int *peos)
-{
-    return sos->prev->read_to(sos->prev, bb, plen, peos);
-}
-
-static apr_status_t h2_sos_h2_status_prepare(h2_sos *sos, apr_off_t *plen, int *peos)
-{
-    return sos->prev->prepare(sos->prev, plen, peos);
-}
-
-static apr_status_t h2_sos_h2_status_readx(h2_sos *sos, h2_io_data_cb *cb, void *ctx,
-                                           apr_off_t *plen, int *peos)
-{
-    return sos->prev->readx(sos->prev, cb, ctx, plen, peos);
-}
-
-static apr_table_t *h2_sos_h2_status_get_trailers(h2_sos *sos)
-{
-    return sos->prev->get_trailers(sos->prev);
-}
-
-static h2_sos *h2_sos_h2_status_create(h2_sos *prev) 
-{
-    h2_sos *sos;
-    h2_response *response = prev->response;
-    
-    apr_table_unset(response->headers, "Content-Length");
-    response->content_length = -1;
-
-    sos = apr_pcalloc(prev->stream->pool, sizeof(*sos));
-    sos->prev         = prev;
-    sos->response     = response;
-    sos->stream       = prev->stream;
-    sos->buffer       = h2_sos_h2_status_buffer;
-    sos->prepare      = h2_sos_h2_status_prepare;
-    sos->readx        = h2_sos_h2_status_readx;
-    sos->read_to      = h2_sos_h2_status_read_to;
-    sos->get_trailers = h2_sos_h2_status_get_trailers;
-    
-    return sos;
+    return APR_SUCCESS;
 }
 
-h2_sos *h2_filter_sos_create(const char *name, struct h2_sos *prev)
+apr_status_t h2_stream_filter(h2_stream *stream)
 {
-    if (!strcmp(H2_SOS_H2_STATUS, name)) {
-        return h2_sos_h2_status_create(prev);
+    const char *fname = stream->response? stream->response->sos_filter : NULL; 
+    if (fname && !strcmp(H2_SOS_H2_STATUS, fname)) {
+        return h2_status_stream_filter(stream);
     }
-    return prev;
+    return APR_SUCCESS;
 }
 
index 2f281f8be1b8a31a3b0a83b52671c7e740df6c47..5ba7d1581b924d5d4a268fca9c73a4ca73c4e4a6 100644 (file)
@@ -43,35 +43,9 @@ apr_status_t h2_filter_core_input(ap_filter_t* filter,
                                   apr_read_type_e block,
                                   apr_off_t readbytes);
 
-typedef struct h2_sos h2_sos;
-typedef apr_status_t h2_sos_data_cb(void *ctx, const char *data, apr_off_t len);
-
-typedef apr_status_t h2_sos_buffer(h2_sos *sos, apr_bucket_brigade *bb);
-typedef apr_status_t h2_sos_prepare(h2_sos *sos, apr_off_t *plen, int *peos);
-typedef apr_status_t h2_sos_readx(h2_sos *sos, h2_sos_data_cb *cb, 
-                                  void *ctx, apr_off_t *plen, int *peos);
-typedef apr_status_t h2_sos_read_to(h2_sos *sos, apr_bucket_brigade *bb, 
-                                    apr_off_t *plen, int *peos);
-typedef apr_table_t *h2_sos_get_trailers(h2_sos *sos);
-
-
 #define H2_RESP_SOS_NOTE     "h2-sos-filter"
 
-struct h2_sos {
-    struct h2_stream *stream;
-    h2_sos           *prev;
-    struct h2_response *response;
-    void             *ctx;
-    h2_sos_buffer    *buffer;
-    h2_sos_prepare   *prepare;
-    h2_sos_readx     *readx;
-    h2_sos_read_to   *read_to;
-    h2_sos_get_trailers *get_trailers;
-};
-
-h2_sos *h2_filter_sos_create(const char *name, struct h2_sos *prev); 
-
+apr_status_t h2_stream_filter(struct h2_stream *stream);
 int h2_filter_h2_status_handler(request_rec *r);
 
-
 #endif /* __mod_h2__h2_filter__ */
index 8e1f163a2aaeb7555af120f3ce5a26e31cabb584..eb866a7911ef3ecfc92ae4d1c44211fbe8ec0e48 100644 (file)
@@ -31,7 +31,6 @@
 #include "h2_response.h"
 #include "h2_from_h1.h"
 #include "h2_task.h"
-#include "h2_task_output.h"
 #include "h2_util.h"
 
 
@@ -473,7 +472,7 @@ static h2_response *create_response(h2_from_h1 *from_h1, request_rec *r)
 apr_status_t h2_response_output_filter(ap_filter_t *f, apr_bucket_brigade *bb)
 {
     h2_task *task = f->ctx;
-    h2_from_h1 *from_h1 = task->output? task->output->from_h1 : NULL;
+    h2_from_h1 *from_h1 = task->output.from_h1;
     request_rec *r = f->r;
     apr_bucket *b;
     ap_bucket_error *eb = NULL;
@@ -483,7 +482,7 @@ apr_status_t h2_response_output_filter(ap_filter_t *f, apr_bucket_brigade *bb)
     ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
                   "h2_from_h1(%d): output_filter called", from_h1->stream_id);
     
-    if (r->header_only && task->output && from_h1->response) {
+    if (r->header_only && from_h1->response) {
         /* throw away any data after we have compiled the response */
         apr_brigade_cleanup(bb);
         return OK;
@@ -552,7 +551,7 @@ apr_status_t h2_response_output_filter(ap_filter_t *f, apr_bucket_brigade *bb)
 apr_status_t h2_response_trailers_filter(ap_filter_t *f, apr_bucket_brigade *bb)
 {
     h2_task *task = f->ctx;
-    h2_from_h1 *from_h1 = task->output? task->output->from_h1 : NULL;
+    h2_from_h1 *from_h1 = task->output.from_h1;
     request_rec *r = f->r;
     apr_bucket *b;
  
diff --git a/modules/http2/h2_int_queue.c b/modules/http2/h2_int_queue.c
deleted file mode 100644 (file)
index 472ae34..0000000
+++ /dev/null
@@ -1,187 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include <assert.h>
-#include <stddef.h>
-#include <apr_pools.h>
-
-#include "h2_int_queue.h"
-
-
-static void tq_grow(h2_int_queue *q, int nlen);
-static void tq_swap(h2_int_queue *q, int i, int j);
-static int tq_bubble_up(h2_int_queue *q, int i, int top, 
-                        h2_iq_cmp *cmp, void *ctx);
-static int tq_bubble_down(h2_int_queue *q, int i, int bottom, 
-                          h2_iq_cmp *cmp, void *ctx);
-
-h2_int_queue *h2_iq_create(apr_pool_t *pool, int capacity)
-{
-    h2_int_queue *q = apr_pcalloc(pool, sizeof(h2_int_queue));
-    if (q) {
-        q->pool = pool;
-        tq_grow(q, capacity);
-        q->nelts = 0;
-    }
-    return q;
-}
-
-int h2_iq_empty(h2_int_queue *q)
-{
-    return q->nelts == 0;
-}
-
-int h2_iq_size(h2_int_queue *q)
-{
-    return q->nelts;
-}
-
-
-void h2_iq_add(h2_int_queue *q, int sid, h2_iq_cmp *cmp, void *ctx)
-{
-    int i;
-    
-    if (q->nelts >= q->nalloc) {
-        tq_grow(q, q->nalloc * 2);
-    }
-    
-    i = (q->head + q->nelts) % q->nalloc;
-    q->elts[i] = sid;
-    ++q->nelts;
-    
-    if (cmp) {
-        /* bubble it to the front of the queue */
-        tq_bubble_up(q, i, q->head, cmp, ctx);
-    }
-}
-
-int h2_iq_remove(h2_int_queue *q, int sid)
-{
-    int i;
-    for (i = 0; i < q->nelts; ++i) {
-        if (sid == q->elts[(q->head + i) % q->nalloc]) {
-            break;
-        }
-    }
-    
-    if (i < q->nelts) {
-        ++i;
-        for (; i < q->nelts; ++i) {
-            q->elts[(q->head+i-1)%q->nalloc] = q->elts[(q->head+i)%q->nalloc];
-        }
-        --q->nelts;
-        return 1;
-    }
-    return 0;
-}
-
-void h2_iq_clear(h2_int_queue *q)
-{
-    q->nelts = 0;
-}
-
-void h2_iq_sort(h2_int_queue *q, h2_iq_cmp *cmp, void *ctx)
-{
-    /* Assume that changes in ordering are minimal. This needs,
-     * best case, q->nelts - 1 comparisions to check that nothing
-     * changed.
-     */
-    if (q->nelts > 0) {
-        int i, ni, prev, last;
-        
-        /* Start at the end of the queue and create a tail of sorted
-         * entries. Make that tail one element longer in each iteration.
-         */
-        last = i = (q->head + q->nelts - 1) % q->nalloc;
-        while (i != q->head) {
-            prev = (q->nalloc + i - 1) % q->nalloc;
-            
-            ni = tq_bubble_up(q, i, prev, cmp, ctx);
-            if (ni == prev) {
-                /* i bubbled one up, bubble the new i down, which
-                 * keeps all tasks below i sorted. */
-                tq_bubble_down(q, i, last, cmp, ctx);
-            }
-            i = prev;
-        };
-    }
-}
-
-
-int h2_iq_shift(h2_int_queue *q)
-{
-    int sid;
-    
-    if (q->nelts <= 0) {
-        return 0;
-    }
-    
-    sid = q->elts[q->head];
-    q->head = (q->head + 1) % q->nalloc;
-    q->nelts--;
-    
-    return sid;
-}
-
-static void tq_grow(h2_int_queue *q, int nlen)
-{
-    if (nlen > q->nalloc) {
-        int *nq = apr_pcalloc(q->pool, sizeof(int) * nlen);
-        if (q->nelts > 0) {
-            int l = ((q->head + q->nelts) % q->nalloc) - q->head;
-            
-            memmove(nq, q->elts + q->head, sizeof(int) * l);
-            if (l < q->nelts) {
-                /* elts wrapped, append elts in [0, remain] to nq */
-                int remain = q->nelts - l;
-                memmove(nq + l, q->elts, sizeof(int) * remain);
-            }
-        }
-        q->elts = nq;
-        q->nalloc = nlen;
-        q->head = 0;
-    }
-}
-
-static void tq_swap(h2_int_queue *q, int i, int j)
-{
-    int x = q->elts[i];
-    q->elts[i] = q->elts[j];
-    q->elts[j] = x;
-}
-
-static int tq_bubble_up(h2_int_queue *q, int i, int top, 
-                        h2_iq_cmp *cmp, void *ctx) 
-{
-    int prev;
-    while (((prev = (q->nalloc + i - 1) % q->nalloc), i != top) 
-           && (*cmp)(q->elts[i], q->elts[prev], ctx) < 0) {
-        tq_swap(q, prev, i);
-        i = prev;
-    }
-    return i;
-}
-
-static int tq_bubble_down(h2_int_queue *q, int i, int bottom, 
-                          h2_iq_cmp *cmp, void *ctx)
-{
-    int next;
-    while (((next = (q->nalloc + i + 1) % q->nalloc), i != bottom) 
-           && (*cmp)(q->elts[i], q->elts[next], ctx) > 0) {
-        tq_swap(q, next, i);
-        i = next;
-    }
-    return i;
-}
diff --git a/modules/http2/h2_int_queue.h b/modules/http2/h2_int_queue.h
deleted file mode 100644 (file)
index 69f1e1c..0000000
+++ /dev/null
@@ -1,108 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef __mod_h2__h2_int_queue__
-#define __mod_h2__h2_int_queue__
-
-/**
- * h2_int_queue keeps a list of sorted h2_task* in ascending order.
- */
-typedef struct h2_int_queue h2_int_queue;
-
-struct h2_int_queue {
-    int *elts;
-    int head;
-    int nelts;
-    int nalloc;
-    apr_pool_t *pool;
-};
-
-/**
- * Comparator for two task to determine their order.
- *
- * @param s1 stream id to compare
- * @param s2 stream id to compare
- * @param ctx provided user data
- * @return value is the same as for strcmp() and has the effect:
- *    == 0: s1 and s2 are treated equal in ordering
- *     < 0: s1 should be sorted before s2
- *     > 0: s2 should be sorted before s1
- */
-typedef int h2_iq_cmp(int s1, int s2, void *ctx);
-
-
-/**
- * Allocate a new queue from the pool and initialize.
- * @param id the identifier of the queue
- * @param pool the memory pool
- */
-h2_int_queue *h2_iq_create(apr_pool_t *pool, int capacity);
-
-/**
- * Return != 0 iff there are no tasks in the queue.
- * @param q the queue to check
- */
-int h2_iq_empty(h2_int_queue *q);
-
-/**
- * Return the number of int in the queue.
- * @param q the queue to get size on
- */
-int h2_iq_size(h2_int_queue *q);
-
-/**
- * Add a stream idto the queue. 
- *
- * @param q the queue to append the task to
- * @param sid the stream id to add
- * @param cmp the comparator for sorting
- * @param ctx user data for comparator 
- */
-void h2_iq_add(h2_int_queue *q, int sid, h2_iq_cmp *cmp, void *ctx);
-
-/**
- * Remove the stream id from the queue. Return != 0 iff task
- * was found in queue.
- * @param q the task queue
- * @param sid the stream id to remove
- * @return != 0 iff task was found in queue
- */
-int h2_iq_remove(h2_int_queue *q, int sid);
-
-/**
- * Remove all entries in the queue.
- */
-void h2_iq_clear(h2_int_queue *q);
-
-/**
- * Sort the stream idqueue again. Call if the task ordering
- * has changed.
- *
- * @param q the queue to sort
- * @param cmp the comparator for sorting
- * @param ctx user data for the comparator 
- */
-void h2_iq_sort(h2_int_queue *q, h2_iq_cmp *cmp, void *ctx);
-
-/**
- * Get the first stream id from the queue or NULL if the queue is empty. 
- * The task will be removed.
- *
- * @param q the queue to get the first task from
- * @return the first stream id of the queue, 0 if empty
- */
-int h2_iq_shift(h2_int_queue *q);
-
-#endif /* defined(__mod_h2__h2_int_queue__) */
index 6c140a08933bba96e45e450f5273d18ca1eb555c..395c34e50218e19e6536344f91db2e3fa863bf8c 100644 (file)
@@ -26,6 +26,7 @@
 #include <http_request.h>
 
 #include "h2_private.h"
+#include "h2_bucket_beam.h"
 #include "h2_h2.h"
 #include "h2_io.h"
 #include "h2_mplx.h"
 #include "h2_task.h"
 #include "h2_util.h"
 
-h2_io *h2_io_create(int id, apr_pool_t *pool, 
-                    apr_bucket_alloc_t *bucket_alloc,
-                    const h2_request *request)
+h2_io *h2_io_create(int id, apr_pool_t *pool, const h2_request *request)
 {
     h2_io *io = apr_pcalloc(pool, sizeof(*io));
     if (io) {
         io->id = id;
         io->pool = pool;
-        io->bucket_alloc = bucket_alloc;
         io->request = request;
+        if (request->body) {
+            h2_beam_create(&io->beam_in, pool, id, "input", 0);
+        }
     }
     return io;
 }
 
-static void check_bbin(h2_io *io)
-{
-    if (!io->bbin) {
-        io->bbin = apr_brigade_create(io->pool, io->bucket_alloc);
-    }
-}
-
-static void check_bbout(h2_io *io)
-{
-    if (!io->bbout) {
-        io->bbout = apr_brigade_create(io->pool, io->bucket_alloc);
-    }
-}
-
-static void check_bbtmp(h2_io *io)
-{
-    if (!io->bbtmp) {
-        io->bbtmp = apr_brigade_create(io->pool, io->bucket_alloc);
-    }
-}
-
-static void append_eos(h2_io *io, apr_bucket_brigade *bb)
-{
-    APR_BRIGADE_INSERT_TAIL(bb, apr_bucket_eos_create(io->bucket_alloc));
-}
-
 void h2_io_redo(h2_io *io)
 {
     io->worker_started = 0;
     io->response = NULL;
     io->rst_error = 0;
-    if (io->bbin) {
-        apr_brigade_cleanup(io->bbin);
-    }
-    if (io->bbout) {
-        apr_brigade_cleanup(io->bbout);
-    }
-    if (io->bbtmp) {
-        apr_brigade_cleanup(io->bbtmp);
-    }
     io->started_at = io->done_at = 0;
 }
 
-int h2_io_is_repeatable(h2_io *io) {
+int h2_io_can_redo(h2_io *io) {
     if (io->submitted
-        || io->input_consumed > 0 
+        || (io->beam_in && h2_beam_was_received(io->beam_in)) 
         || !io->request) {
         /* cannot repeat that. */
         return 0;
@@ -103,351 +69,40 @@ int h2_io_is_repeatable(h2_io *io) {
             || !strcmp("OPTIONS", io->request->method));
 }
 
-void h2_io_set_response(h2_io *io, h2_response *response) 
+void h2_io_set_response(h2_io *io, h2_response *response, 
+                        h2_bucket_beam *output) 
 {
-    AP_DEBUG_ASSERT(io->pool);
     AP_DEBUG_ASSERT(response);
     AP_DEBUG_ASSERT(!io->response);
-    /* we used to clone the response into the io->pool. But
+    /* we used to clone the response into out own pool. But
      * we have much tighter control over the EOR bucket nowadays,
      * so just use the instance given */
     io->response = response;
+    if (output) {
+        io->beam_out = output;
+    }
     if (response->rst_error) {
         h2_io_rst(io, response->rst_error);
     }
-    else if (response->content_length == 0) {
-        io->eos_out = 1;
-    }
 }
 
 void h2_io_rst(h2_io *io, int error)
 {
     io->rst_error = error;
-    io->eos_in = 1;
-}
-
-int h2_io_out_has_data(h2_io *io)
-{
-    return io->bbout && h2_util_bb_has_data_or_eos(io->bbout);
-}
-
-apr_off_t h2_io_out_length(h2_io *io)
-{
-    if (io->bbout) {
-        apr_off_t len = 0;
-        apr_brigade_length(io->bbout, 0, &len);
-        return (len > 0)? len : 0;
-    }
-    return 0;
-}
-
-apr_status_t h2_io_in_shutdown(h2_io *io)
-{
-    if (io->bbin) {
-        apr_off_t end_len = 0;
-        apr_brigade_length(io->bbin, 1, &end_len);
-        io->input_consumed += end_len;
-        apr_brigade_cleanup(io->bbin);
-    }
-    return h2_io_in_close(io);
-}
-
-
-void h2_io_signal_init(h2_io *io, h2_io_op op, apr_interval_time_t timeout, 
-                       apr_thread_cond_t *cond)
-{
-    io->timed_op = op;
-    io->timed_cond = cond;
-    if (timeout > 0) {
-        io->timeout_at = apr_time_now() + timeout;
-    }
-    else {
-        io->timeout_at = 0; 
-    }
-}
-
-void h2_io_signal_exit(h2_io *io)
-{
-    io->timed_cond = NULL;
-    io->timeout_at = 0; 
-}
-
-apr_status_t h2_io_signal_wait(h2_mplx *m, h2_io *io)
-{
-    apr_status_t status;
-    
-    if (io->timeout_at != 0) {
-        status = apr_thread_cond_timedwait(io->timed_cond, m->lock, io->timeout_at);
-        if (APR_STATUS_IS_TIMEUP(status)) {
-            ap_log_cerror(APLOG_MARK, APLOG_DEBUG, status, m->c, APLOGNO(03055)  
-                          "h2_mplx(%ld-%d): stream timeout expired: %s",
-                          m->id, io->id, 
-                          (io->timed_op == H2_IO_READ)? "read" : "write");
-            h2_io_rst(io, H2_ERR_CANCEL);
-        }
-    }
-    else {
-        apr_thread_cond_wait(io->timed_cond, m->lock);
-        status = APR_SUCCESS;
-    }
-    if (io->orphaned && status == APR_SUCCESS) {
-        return APR_ECONNABORTED;
-    }
-    return status;
-}
-
-void h2_io_signal(h2_io *io, h2_io_op op)
-{
-    if (io->timed_cond && (io->timed_op == op || H2_IO_ANY == op)) {
-        apr_thread_cond_signal(io->timed_cond);
-    }
-}
-
-void h2_io_make_orphaned(h2_io *io, int error)
-{
-    io->orphaned = 1;
-    if (error) {
-        h2_io_rst(io, error);
-    }
-    /* if someone is waiting, wake him up */
-    h2_io_signal(io, H2_IO_ANY);
-}
-
-static int add_trailer(void *ctx, const char *key, const char *value)
-{
-    apr_bucket_brigade *bb = ctx;
-    apr_status_t status;
-    
-    status = apr_brigade_printf(bb, NULL, NULL, "%s: %s\r\n", 
-                                key, value);
-    return (status == APR_SUCCESS);
-}
-
-static apr_status_t in_append_eos(h2_io *io, apr_bucket_brigade *bb, 
-                                  apr_table_t *trailers)
-{
-    apr_status_t status = APR_SUCCESS;
-    apr_table_t *t = io->request->trailers;
-
-    if (trailers && t && !apr_is_empty_table(trailers)) {
-        /* trailers passed in, transfer directly. */
-        apr_table_overlap(trailers, t, APR_OVERLAP_TABLES_SET);
-        t = NULL;
-    }
-    
-    if (io->request->chunked) {
-        if (t && !apr_is_empty_table(t)) {
-            /* no trailers passed in, transfer via chunked */
-            status = apr_brigade_puts(bb, NULL, NULL, "0\r\n");
-            apr_table_do(add_trailer, bb, t, NULL);
-            status = apr_brigade_puts(bb, NULL, NULL, "\r\n");
-        }
-        else {
-            status = apr_brigade_puts(bb, NULL, NULL, "0\r\n\r\n");
-        }
-    }
-    append_eos(io, bb);
-    return status;
-}
-
-apr_status_t h2_io_in_read(h2_io *io, apr_bucket_brigade *bb, 
-                           apr_size_t maxlen, apr_table_t *trailers)
-{
-    apr_off_t start_len = 0;
-    apr_status_t status;
-
-    if (io->rst_error) {
-        return APR_ECONNABORTED;
-    }
-    
-    if (!io->bbin || APR_BRIGADE_EMPTY(io->bbin)) {
-        if (io->eos_in) {
-            if (!io->eos_in_written) {
-                status = in_append_eos(io, bb, trailers);
-                io->eos_in_written = 1;
-                return status;
-            }
-            return APR_EOF;
-        }
-        return APR_EAGAIN;
-    }
-    
-    if (io->request->chunked) {
-        /* the reader expects HTTP/1.1 chunked encoding */
-        check_bbtmp(io);
-        status = h2_util_move(io->bbtmp, io->bbin, maxlen, NULL, "h2_io_in_read_chunk");
-        if (status == APR_SUCCESS) {
-            apr_off_t tmp_len = 0;
-            
-            apr_brigade_length(io->bbtmp, 1, &tmp_len);
-            if (tmp_len > 0) {
-                io->input_consumed += tmp_len;
-                status = apr_brigade_printf(bb, NULL, NULL, "%lx\r\n", 
-                                            (unsigned long)tmp_len);
-                if (status == APR_SUCCESS) {
-                    status = h2_util_move(bb, io->bbtmp, -1, NULL, "h2_io_in_read_tmp1");
-                    if (status == APR_SUCCESS) {
-                        status = apr_brigade_puts(bb, NULL, NULL, "\r\n");
-                    }
-                }
-            }
-            else {
-                status = h2_util_move(bb, io->bbtmp, -1, NULL, "h2_io_in_read_tmp2");
-            }
-            apr_brigade_cleanup(io->bbtmp);
-        }
-    }
-    else {
-        apr_brigade_length(bb, 1, &start_len);
-        
-        status = h2_util_move(bb, io->bbin, maxlen, NULL, "h2_io_in_read");
-        if (status == APR_SUCCESS) {
-            apr_off_t end_len = 0;
-            apr_brigade_length(bb, 1, &end_len);
-            io->input_consumed += (end_len - start_len);
-        }
-    }
-    
-    if (status == APR_SUCCESS && (!io->bbin || APR_BRIGADE_EMPTY(io->bbin))) {
-        if (io->eos_in) {
-            if (!io->eos_in_written) {
-                status = in_append_eos(io, bb, trailers);
-                io->eos_in_written = 1;
-            }
-        }
-    }
-    
-    if (status == APR_SUCCESS && APR_BRIGADE_EMPTY(bb)) {
-        return APR_EAGAIN;
-    }
-    return status;
-}
-
-apr_status_t h2_io_in_write(h2_io *io, const char *d, apr_size_t len, int eos)
-{
-    if (io->rst_error) {
-        return APR_ECONNABORTED;
-    }
-    
-    if (io->eos_in) {
-        return APR_EOF;
+    if (io->beam_in) {
+        h2_beam_abort(io->beam_in);
     }
-    if (eos) {
-        io->eos_in = 1;
-    }
-    if (len > 0) {
-        check_bbin(io);
-        return apr_brigade_write(io->bbin, NULL, NULL, d, len);
-    }
-    return APR_SUCCESS;
-}
-
-apr_status_t h2_io_in_close(h2_io *io)
-{
-    if (io->rst_error) {
-        return APR_ECONNABORTED;
+    if (io->beam_out) {
+        h2_beam_abort(io->beam_out);
     }
-    
-    io->eos_in = 1;
-    return APR_SUCCESS;
 }
 
-apr_status_t h2_io_out_get_brigade(h2_io *io, apr_bucket_brigade *bb, 
-                                   apr_off_t len)
+void h2_io_shutdown(h2_io *io)
 {
-    if (io->rst_error) {
-        return APR_ECONNABORTED;
-    }
-    if (io->eos_out_read) {
-        return APR_EOF;
-    }
-    else if (!io->bbout || APR_BRIGADE_EMPTY(io->bbout)) {
-        return APR_EAGAIN;
+    if (io->beam_in) {
+        h2_beam_shutdown(io->beam_in);
     }
-    else {
-        apr_status_t status;
-        apr_off_t pre_len, post_len;
-        /* Allow file handles pass through without limits. If they
-         * already have the lifetime of this stream, we might as well
-         * pass them on to the master connection */
-        apr_size_t files = INT_MAX;
-        
-        apr_brigade_length(bb, 0, &pre_len);
-        status = h2_util_move(bb, io->bbout, len, &files, "h2_io_read_to");
-        if (status == APR_SUCCESS && io->eos_out 
-            && APR_BRIGADE_EMPTY(io->bbout)) {
-            io->eos_out_read = 1;
-        }
-        apr_brigade_length(bb, 0, &post_len);
-        io->output_consumed += (post_len - pre_len);
-        return status;
-    }
-}
-
-apr_status_t h2_io_out_write(h2_io *io, apr_bucket_brigade *bb, 
-                             apr_size_t maxlen, 
-                             apr_size_t *pfile_buckets_allowed)
-{
-    apr_status_t status;
-    apr_bucket *b;
-    int start_allowed;
-    
-    if (io->rst_error) {
-        return APR_ECONNABORTED;
-    }
-
-    /* Filter the EOR bucket and set it aside. We prefer to tear down
-     * the request when the whole h2 stream is done */
-    for (b = APR_BRIGADE_FIRST(bb);
-         b != APR_BRIGADE_SENTINEL(bb);
-         b = APR_BUCKET_NEXT(b))
-    {
-        if (AP_BUCKET_IS_EOR(b)) {
-            APR_BUCKET_REMOVE(b);
-            io->eor = b;
-            break;
-        }
-        else if (APR_BUCKET_IS_EOS(b)) {
-            io->eos_out = 1;
-            break;
-        }
-    }     
-    
-    /* Let's move the buckets from the request processing in here, so
-     * that the main thread can read them when it has time/capacity.
-     *
-     * Move at most "maxlen" memory bytes. If buckets remain, it is
-     * the caller's responsibility to take care of this.
-     *
-     * We allow passing of file buckets as long as we do not have too
-     * many open files already buffered. Otherwise we will run out of
-     * file handles.
-     */
-    check_bbout(io);
-    start_allowed = *pfile_buckets_allowed;
-    status = h2_util_move(io->bbout, bb, maxlen, pfile_buckets_allowed, 
-                          "h2_io_out_write");
-    /* track # file buckets moved into our pool */
-    if (start_allowed != *pfile_buckets_allowed) {
-        io->files_handles_owned += (start_allowed - *pfile_buckets_allowed);
-    }
-    return status;
-}
-
-
-apr_status_t h2_io_out_close(h2_io *io)
-{
-    if (io->rst_error) {
-        return APR_ECONNABORTED;
-    }
-    if (!io->eos_out_read) { /* EOS has not been read yet */
-        if (!io->eos_out) {
-            check_bbout(io);
-            io->eos_out = 1;
-            if (!h2_util_has_eos(io->bbout, -1)) {
-                append_eos(io, io->bbout);
-            }
-        }
+    if (io->beam_out) {
+        h2_beam_shutdown(io->beam_out);
     }
-    return APR_SUCCESS;
 }
index d700f6f32203794a58215b92e6792ff21939bc27..bfd130eababb5b6635faa9cb3d94f8a11cb8b7ba 100644 (file)
@@ -16,6 +16,7 @@
 #ifndef __mod_h2__h2_io__
 #define __mod_h2__h2_io__
 
+struct h2_bucket_beam;
 struct h2_response;
 struct apr_thread_cond_t;
 struct h2_mplx;
@@ -37,40 +38,23 @@ typedef struct h2_io h2_io;
 
 struct h2_io {
     int id;                          /* stream identifier */
-     apr_pool_t *pool;                /* stream pool */
-    apr_bucket_alloc_t *bucket_alloc;
+    apr_pool_t *pool;                /* io pool */
     
-    const struct h2_request *request;/* request on this io */
-    struct h2_response *response;    /* response to request */
-    int rst_error;                   /* h2 related stream abort error */
+    const struct h2_request *request;/* request to process */
+    struct h2_response *response;    /* response to submit */
+
+    struct h2_bucket_beam *beam_in;  /* request body buckets */
+    struct h2_bucket_beam *beam_out; /* response body buckets */
 
-    apr_bucket *eor;                 /* the EOR bucket, set aside */
     struct h2_task *task;            /* the task once started */
+    apr_time_t started_at;           /* when processing started */
+    apr_time_t done_at;              /* when processing was done */
     
-    apr_bucket_brigade *bbin;        /* input data for stream */
-    apr_bucket_brigade *bbout;       /* output data from stream */
-    apr_bucket_brigade *bbtmp;       /* temporary data for chunking */
-
+    int rst_error;                   /* h2 related stream abort error */
     unsigned int orphaned       : 1; /* h2_stream is gone for this io */    
+    unsigned int submitted      : 1; /* response has been submitted to client */
     unsigned int worker_started : 1; /* h2_worker started processing for this io */
     unsigned int worker_done    : 1; /* h2_worker finished for this io */
-    unsigned int submitted      : 1; /* response has been submitted to client */
-    unsigned int request_body   : 1; /* iff request has body */
-    unsigned int eos_in         : 1; /* input eos has been seen */
-    unsigned int eos_in_written : 1; /* input eos has been forwarded */
-    unsigned int eos_out        : 1; /* output eos is present */
-    unsigned int eos_out_read   : 1; /* output eos has been forwarded */
-    
-    h2_io_op timed_op;               /* which operation is waited on, if any */
-    struct apr_thread_cond_t *timed_cond; /* condition to wait on, maybe NULL */
-    apr_time_t timeout_at;           /* when IO wait will time out */
-    
-    apr_time_t started_at;           /* when processing started */
-    apr_time_t done_at;              /* when processing was done */
-    apr_size_t input_consumed;       /* how many bytes have been read */
-    apr_size_t output_consumed;      /* how many bytes have been written out */
-        
-    int files_handles_owned;
 };
 
 /*******************************************************************************
@@ -80,96 +64,25 @@ struct h2_io {
 /**
  * Creates a new h2_io for the given stream id. 
  */
-h2_io *h2_io_create(int id, apr_pool_t *pool, 
-                    apr_bucket_alloc_t *bucket_alloc, 
-                    const struct h2_request *request);
+h2_io *h2_io_create(int id, apr_pool_t *pool, const struct h2_request *request);
 
 /**
  * Set the response of this stream.
  */
-void h2_io_set_response(h2_io *io, struct h2_response *response);
+void h2_io_set_response(h2_io *io, struct h2_response *response,
+                        struct h2_bucket_beam *output);
 
 /**
  * Reset the stream with the given error code.
  */
 void h2_io_rst(h2_io *io, int error);
 
-int h2_io_is_repeatable(h2_io *io);
+int h2_io_can_redo(h2_io *io);
 void h2_io_redo(h2_io *io);
 
 /**
- * Output data is available.
- */
-int h2_io_out_has_data(h2_io *io);
-
-void h2_io_signal(h2_io *io, h2_io_op op);
-void h2_io_signal_init(h2_io *io, h2_io_op op, apr_interval_time_t timeout, 
-                       struct apr_thread_cond_t *cond);
-void h2_io_signal_exit(h2_io *io);
-apr_status_t h2_io_signal_wait(struct h2_mplx *m, h2_io *io);
-
-void h2_io_make_orphaned(h2_io *io, int error);
-
-/*******************************************************************************
- * Input handling of streams.
- ******************************************************************************/
-/**
- * Reads the next bucket from the input. Returns APR_EAGAIN if none
- * is currently available, APR_EOF if end of input has been reached.
- */
-apr_status_t h2_io_in_read(h2_io *io, apr_bucket_brigade *bb, 
-                           apr_size_t maxlen, apr_table_t *trailers);
-
-/**
- * Appends given bucket to the input.
- */
-apr_status_t h2_io_in_write(h2_io *io, const char *d, apr_size_t len, int eos);
-
-/**
- * Closes the input. After existing data has been read, APR_EOF will
- * be returned.
+ * Shuts all input/output down. Clears any buckets buffered and closes.
  */
-apr_status_t h2_io_in_close(h2_io *io);
-
-/**
- * Shuts all input down. Will close input and mark any data buffered
- * as consumed.
- */
-apr_status_t h2_io_in_shutdown(h2_io *io);
-
-/*******************************************************************************
- * Output handling of streams.
- ******************************************************************************/
-
-/**
- * Read a bucket from the output head. Return APR_EAGAIN if non is available,
- * APR_EOF if none available and output has been closed. 
- * May be called with buffer == NULL in order to find out how much data
- * is available.
- * @param io the h2_io to read output from
- * @param buffer the buffer to copy the data to, may be NULL
- * @param plen the requested max len, set to amount of data on return
- * @param peos != 0 iff the end of stream has been reached
- */
-apr_status_t h2_io_out_get_brigade(h2_io *io, 
-                                   apr_bucket_brigade *bb, 
-                                   apr_off_t len);
-
-apr_status_t h2_io_out_write(h2_io *io, apr_bucket_brigade *bb, 
-                             apr_size_t maxlen, 
-                             apr_size_t *pfile_buckets_allowed);
-
-/**
- * Closes the input. After existing data has been read, APR_EOF will
- * be returned.
- */
-apr_status_t h2_io_out_close(h2_io *io);
-
-/**
- * Gives the overall length of the data that is currently queued for
- * output.
- */
-apr_off_t h2_io_out_length(h2_io *io);
-
+void h2_io_shutdown(h2_io *io);
 
 #endif /* defined(__mod_h2__h2_io__) */
diff --git a/modules/http2/h2_io_set.c b/modules/http2/h2_io_set.c
deleted file mode 100644 (file)
index e094979..0000000
+++ /dev/null
@@ -1,159 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include <assert.h>
-#include <stddef.h>
-
-#include <apr_strings.h>
-
-#include <httpd.h>
-#include <http_core.h>
-#include <http_connection.h>
-#include <http_log.h>
-
-#include "h2_private.h"
-#include "h2_io.h"
-#include "h2_io_set.h"
-
-#define h2_io_IDX(list, i) ((h2_io**)(list)->elts)[i]
-
-struct h2_io_set {
-    apr_array_header_t *list;
-};
-
-h2_io_set *h2_io_set_create(apr_pool_t *pool)
-{
-    h2_io_set *sp = apr_pcalloc(pool, sizeof(h2_io_set));
-    if (sp) {
-        sp->list = apr_array_make(pool, 100, sizeof(h2_io*));
-        if (!sp->list) {
-            return NULL;
-        }
-    }
-    return sp;
-}
-
-static int h2_stream_id_cmp(const void *s1, const void *s2)
-{
-    h2_io **pio1 = (h2_io **)s1;
-    h2_io **pio2 = (h2_io **)s2;
-    return (*pio1)->id - (*pio2)->id;
-}
-
-h2_io *h2_io_set_get(h2_io_set *sp, int stream_id)
-{
-    /* we keep the array sorted by id, so lookup can be done
-     * by bsearch.
-     */
-    h2_io **ps;
-    h2_io key;
-    h2_io *pkey = &key;
-
-    memset(&key, 0, sizeof(key));
-    key.id = stream_id;
-    ps = bsearch(&pkey, sp->list->elts, sp->list->nelts, 
-                         sp->list->elt_size, h2_stream_id_cmp);
-    return ps? *ps : NULL;
-}
-
-static void h2_io_set_sort(h2_io_set *sp)
-{
-    qsort(sp->list->elts, sp->list->nelts, sp->list->elt_size, 
-          h2_stream_id_cmp);
-}
-
-apr_status_t h2_io_set_add(h2_io_set *sp, h2_io *io)
-{
-    h2_io *existing = h2_io_set_get(sp, io->id);
-    if (!existing) {
-        int last;
-        APR_ARRAY_PUSH(sp->list, h2_io*) = io;
-        /* Normally, streams get added in ascending order if id. We
-         * keep the array sorted, so we just need to check if the newly
-         * appended stream has a lower id than the last one. if not,
-         * sorting is not necessary.
-         */
-        last = sp->list->nelts - 1;
-        if (last > 0 
-            && (h2_io_IDX(sp->list, last)->id 
-                < h2_io_IDX(sp->list, last-1)->id)) {
-                h2_io_set_sort(sp);
-            }
-    }
-    return APR_SUCCESS;
-}
-
-static void remove_idx(h2_io_set *sp, int idx)
-{
-    int n;
-    --sp->list->nelts;
-    n = sp->list->nelts - idx;
-    if (n > 0) {
-        /* There are n h2_io* behind idx. Move the rest down */
-        h2_io **selts = (h2_io**)sp->list->elts;
-        memmove(selts + idx, selts + idx + 1, n * sizeof(h2_io*));
-    }
-}
-
-h2_io *h2_io_set_remove(h2_io_set *sp, h2_io *io)
-{
-    int i;
-    for (i = 0; i < sp->list->nelts; ++i) {
-        h2_io *e = h2_io_IDX(sp->list, i);
-        if (e->id == io->id) {
-            remove_idx(sp, i);
-            return e;
-        }
-    }
-    return NULL;
-}
-
-h2_io *h2_io_set_shift(h2_io_set *set)
-{
-    /* For now, this just removes the first element in the set.
-     * the name is misleading...
-     */
-    if (set->list->nelts > 0) {
-        h2_io *io = h2_io_IDX(set->list, 0);
-        remove_idx(set, 0);
-        return io;
-    }
-    return NULL;
-}
-
-int h2_io_set_is_empty(h2_io_set *sp)
-{
-    AP_DEBUG_ASSERT(sp);
-    return sp->list->nelts == 0;
-}
-
-int h2_io_set_iter(h2_io_set *sp,
-                        h2_io_set_iter_fn *iter, void *ctx)
-{
-    int i;
-    for (i = 0; i < sp->list->nelts; ++i) {
-        h2_io *s = h2_io_IDX(sp->list, i);
-        if (!iter(ctx, s)) {
-            return 0;
-        }
-    }
-    return 1;
-}
-
-apr_size_t h2_io_set_size(h2_io_set *sp)
-{
-    return sp->list->nelts;
-}
-
diff --git a/modules/http2/h2_io_set.h b/modules/http2/h2_io_set.h
deleted file mode 100644 (file)
index 936e725..0000000
+++ /dev/null
@@ -1,53 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef __mod_h2__h2_io_set__
-#define __mod_h2__h2_io_set__
-
-struct h2_io;
-
-/**
- * A set of h2_io instances. Allows lookup by stream id
- * and other criteria.
- */
-typedef struct h2_io_set h2_io_set;
-
-h2_io_set *h2_io_set_create(apr_pool_t *pool);
-
-apr_status_t h2_io_set_add(h2_io_set *set, struct h2_io *io);
-h2_io *h2_io_set_get(h2_io_set *set, int stream_id);
-h2_io *h2_io_set_remove(h2_io_set *set, struct h2_io *io);
-
-int h2_io_set_is_empty(h2_io_set *set);
-apr_size_t h2_io_set_size(h2_io_set *set);
-
-
-typedef int h2_io_set_iter_fn(void *ctx, struct h2_io *io);
-
-/**
- * Iterator over all h2_io* in the set or until a
- * callback returns 0. It is not safe to add or remove
- * set members during iteration.
- *
- * @param set the set of h2_io to iterate over
- * @param iter the function to call for each io
- * @param ctx user data for the callback
- * @return 1 iff iteration completed for all members
- */
-int h2_io_set_iter(h2_io_set *set, h2_io_set_iter_fn *iter, void *ctx);
-
-h2_io *h2_io_set_shift(h2_io_set *set);
-
-#endif /* defined(__mod_h2__h2_io_set__) */
index 6513a754152109957d838d4989392c8326fe5ca6..d75188a78e03fa9e244274238c2d203761a99a9b 100644 (file)
 #include "mod_http2.h"
 
 #include "h2_private.h"
+#include "h2_bucket_beam.h"
 #include "h2_config.h"
 #include "h2_conn.h"
 #include "h2_ctx.h"
 #include "h2_h2.h"
-#include "h2_int_queue.h"
 #include "h2_io.h"
-#include "h2_io_set.h"
 #include "h2_response.h"
 #include "h2_mplx.h"
 #include "h2_ngn_shed.h"
 #include "h2_request.h"
 #include "h2_stream.h"
 #include "h2_task.h"
-#include "h2_task_input.h"
-#include "h2_task_output.h"
 #include "h2_worker.h"
 #include "h2_workers.h"
 #include "h2_util.h"
 
 
-#define H2_MPLX_IO_OUT(lvl,m,io,msg) \
-    do { \
-        if (APLOG_C_IS_LEVEL((m)->c,lvl)) \
-        h2_util_bb_log((m)->c,(io)->id,lvl,msg,(io)->bbout); \
-    } while(0)
-    
-#define H2_MPLX_IO_IN(lvl,m,io,msg) \
-    do { \
-        if (APLOG_C_IS_LEVEL((m)->c,lvl)) \
-        h2_util_bb_log((m)->c,(io)->id,lvl,msg,(io)->bbin); \
-    } while(0)
-
+static void h2_beam_log(h2_bucket_beam *beam, int id, const char *msg, 
+                        conn_rec *c, int level)
+{
+    if (beam && APLOG_C_IS_LEVEL(c,level)) {
+        char buffer[2048];
+        apr_size_t off = 0;
+        
+        off += apr_snprintf(buffer+off, H2_ALEN(buffer)-off, "cl=%d, ", beam->closed);
+        off += h2_util_bl_print(buffer+off, H2_ALEN(buffer)-off, "red", ", ", &beam->red);
+        off += h2_util_bb_print(buffer+off, H2_ALEN(buffer)-off, "green", ", ", beam->green);
+        off += h2_util_bl_print(buffer+off, H2_ALEN(buffer)-off, "hold", ", ", &beam->hold);
+        off += h2_util_bl_print(buffer+off, H2_ALEN(buffer)-off, "purge", "", &beam->purge);
+
+        ap_log_cerror(APLOG_MARK, level, 0, c, "beam(%ld-%d): %s %s", 
+                      c->id, id, msg, buffer);
+    }
+}
 
 /* NULL or the mutex hold by this thread, used for recursive calls
  */
@@ -104,13 +106,51 @@ static void leave_mutex(h2_mplx *m, int acquired)
     }
 }
 
-static int is_aborted(h2_mplx *m, apr_status_t *pstatus)
+static apr_status_t io_mutex_enter(void *ctx, 
+                                   apr_thread_mutex_t **plock, int *acquired)
 {
-    AP_DEBUG_ASSERT(m);
-    if (m->aborted) {
-        *pstatus = APR_ECONNABORTED;
+    h2_mplx *m = ctx;
+    *plock = m->lock;
+    return enter_mutex(m, acquired);
+}
+
+static void io_mutex_leave(void *ctx, apr_thread_mutex_t *lock, int acquired)
+{
+    h2_mplx *m = ctx;
+    leave_mutex(m, acquired);
+}
+
+static void stream_output_consumed(void *ctx, 
+                                   h2_bucket_beam *beam, apr_off_t length)
+{
+    h2_io *io = ctx;
+    if (length > 0 && io->task && io->task->assigned) {
+        h2_req_engine_out_consumed(io->task->assigned, io->task->c, length); 
+    }
+}
+
+static void stream_input_consumed(void *ctx, 
+                                  h2_bucket_beam *beam, apr_off_t length)
+{
+    h2_mplx *m = ctx;
+    if (m->input_consumed && length) {
+        m->input_consumed(m->input_consumed_ctx, beam->id, length);
+    }
+}
+
+static int can_beam_file(void *ctx, h2_bucket_beam *beam,  apr_file_t *file)
+{
+    h2_mplx *m = ctx;
+    if (m->tx_handles_reserved > 0) {
+        --m->tx_handles_reserved;
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE3, 0, m->c,
+                      "h2_mplx(%ld-%d): beaming file %s, tx_avail %d", 
+                      m->id, beam->id, beam->tag, m->tx_handles_reserved);
         return 1;
     }
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE3, 0, m->c,
+                  "h2_mplx(%ld-%d): can_beam_file denied on %s", 
+                  m->id, beam->id, beam->tag);
     return 0;
 }
 
@@ -118,9 +158,9 @@ static void have_out_data_for(h2_mplx *m, int stream_id);
 
 static void check_tx_reservation(h2_mplx *m) 
 {
-    if (m->tx_handles_reserved == 0) {
+    if (m->tx_handles_reserved <= 0) {
         m->tx_handles_reserved += h2_workers_tx_reserve(m->workers, 
-            H2MIN(m->tx_chunk_size, h2_io_set_size(m->stream_ios)));
+            H2MIN(m->tx_chunk_size, h2_ilist_count(m->stream_ios)));
     }
 }
 
@@ -132,7 +172,7 @@ static void check_tx_free(h2_mplx *m)
         h2_workers_tx_free(m->workers, count);
     }
     else if (m->tx_handles_reserved 
-             && (!m->stream_ios || h2_io_set_is_empty(m->stream_ios))) {
+             && (!m->stream_ios || h2_ilist_empty(m->stream_ios))) {
         h2_workers_tx_free(m->workers, m->tx_handles_reserved);
         m->tx_handles_reserved = 0;
     }
@@ -143,7 +183,7 @@ static void h2_mplx_destroy(h2_mplx *m)
     AP_DEBUG_ASSERT(m);
     ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, m->c,
                   "h2_mplx(%ld): destroy, ios=%d", 
-                  m->id, (int)h2_io_set_size(m->stream_ios));
+                  m->id, (int)h2_ilist_count(m->stream_ios));
     check_tx_free(m);
     if (m->pool) {
         apr_pool_destroy(m->pool);
@@ -205,8 +245,8 @@ h2_mplx *h2_mplx_create(conn_rec *c, apr_pool_t *parent,
         m->max_streams = h2_config_geti(conf, H2_CONF_MAX_STREAMS);
         m->stream_max_mem = h2_config_geti(conf, H2_CONF_STREAM_MAX_MEM);
         m->q = h2_iq_create(m->pool, m->max_streams);
-        m->stream_ios = h2_io_set_create(m->pool);
-        m->ready_ios = h2_io_set_create(m->pool);
+        m->stream_ios = h2_ilist_create(m->pool);
+        m->ready_ios = h2_ilist_create(m->pool);
         m->stream_timeout = stream_timeout;
         m->workers = workers;
         m->workers_max = workers->max_workers;
@@ -240,49 +280,29 @@ apr_uint32_t h2_mplx_shutdown(h2_mplx *m)
     return max_stream_started;
 }
 
-static void workers_register(h2_mplx *m)
-{
-    /* h2_workers is only a hub for all the h2_worker instances.
-     * At the end-of-life of this h2_mplx, we always unregister at
-     * the workers. The thing to manage are all the h2_worker instances
-     * out there. Those may hold a reference to this h2_mplx and we cannot
-     * call them to unregister.
-     * 
-     * Therefore: ref counting for h2_workers in not needed, ref counting
-     * for h2_worker using this is critical.
-     */
-    m->need_registration = 0;
-    h2_workers_register(m->workers, m);
-}
-
-static int io_in_consumed_signal(h2_mplx *m, h2_io *io)
+static void io_in_consumed_signal(h2_mplx *m, h2_io *io)
 {
-    if (io->input_consumed && m->input_consumed) {
-        m->input_consumed(m->input_consumed_ctx, 
-                          io->id, io->input_consumed);
-        io->input_consumed = 0;
-        return 1;
+    if (io->beam_in && io->worker_started) {
+        h2_beam_send(io->beam_in, NULL, 0); /* trigger updates */
     }
-    return 0;
 }
 
 static int io_out_consumed_signal(h2_mplx *m, h2_io *io)
 {
-    if (io->output_consumed && io->task && io->task->assigned) {
-        h2_req_engine_out_consumed(io->task->assigned, io->task->c, 
-                                   io->output_consumed);
-        io->output_consumed = 0;
-        return 1;
+    if (io->beam_out && io->worker_started && io->task && io->task->assigned) {
+        h2_beam_send(io->beam_out, NULL, 0); /* trigger updates */
     }
     return 0;
 }
 
+
 static void io_destroy(h2_mplx *m, h2_io *io, int events)
 {
+    conn_rec *slave = NULL;
     int reuse_slave;
     
     /* cleanup any buffered input */
-    h2_io_in_shutdown(io);
+    h2_io_shutdown(io);
     if (events) {
         /* Process outstanding events before destruction */
         io_in_consumed_signal(m, io);
@@ -291,24 +311,37 @@ static void io_destroy(h2_mplx *m, h2_io *io, int events)
     /* The pool is cleared/destroyed which also closes all
      * allocated file handles. Give this count back to our
      * file handle pool. */
-    m->tx_handles_reserved += io->files_handles_owned;
+    if (io->beam_in) {
+        m->tx_handles_reserved += h2_beam_get_files_beamed(io->beam_in);
+    }
+    if (io->beam_out) {
+        m->tx_handles_reserved += h2_beam_get_files_beamed(io->beam_out);
+    }
 
-    h2_io_set_remove(m->stream_ios, io);
-    h2_io_set_remove(m->ready_ios, io);
+    h2_ilist_remove(m->stream_ios, io->id);
+    h2_ilist_remove(m->ready_ios, io->id);
     if (m->redo_ios) {
-        h2_io_set_remove(m->redo_ios, io);
+        h2_ilist_remove(m->redo_ios, io->id);
     }
 
     reuse_slave = ((m->spare_slaves->nelts < m->spare_slaves->nalloc)
-                    && !io->rst_error && io->eor);
+                    && !io->rst_error);
     if (io->task) {
-        conn_rec *slave = io->task->c;
+        slave = io->task->c;
         h2_task_destroy(io->task);
         io->task = NULL;
-        
+    }
+
+    if (io->pool) {
+        if (m->spare_io_pool) {
+            apr_pool_destroy(m->spare_io_pool);
+        }
+        apr_pool_clear(io->pool);
+        m->spare_io_pool = io->pool;
+    }
+
+    if (slave) {
         if (reuse_slave && slave->keepalive == AP_CONN_KEEPALIVE) {
-            apr_bucket_delete(io->eor);
-            io->eor = NULL;
             APR_ARRAY_PUSH(m->spare_slaves, conn_rec*) = slave;
         }
         else {
@@ -316,18 +349,14 @@ static void io_destroy(h2_mplx *m, h2_io *io, int events)
             h2_slave_destroy(slave, NULL);
         }
     }
-
-    if (io->pool) {
-        apr_pool_destroy(io->pool);
-    }
-
+    
     check_tx_free(m);
 }
 
 static int io_stream_done(h2_mplx *m, h2_io *io, int rst_error) 
 {
     /* Remove io from ready set, we will never submit it */
-    h2_io_set_remove(m->ready_ios, io);
+    h2_ilist_remove(m->ready_ios, io->id);
     if (!io->worker_started || io->worker_done) {
         /* already finished or not even started yet */
         h2_iq_remove(m->q, io->id);
@@ -336,39 +365,41 @@ static int io_stream_done(h2_mplx *m, h2_io *io, int rst_error)
     }
     else {
         /* cleanup once task is done */
-        h2_io_make_orphaned(io, rst_error);
+        io->orphaned = 1;
+        if (rst_error) {
+            h2_io_rst(io, rst_error);
+        }
         return 1;
     }
 }
 
-static int stream_done_iter(void *ctx, h2_io *io)
+static int stream_done_iter(void *ctx, void *val)
 {
-    return io_stream_done((h2_mplx*)ctx, io, 0);
+    return io_stream_done((h2_mplx*)ctx, val, 0);
 }
 
-static int stream_print(void *ctx, h2_io *io)
+static int stream_print(void *ctx, void *val)
 {
     h2_mplx *m = ctx;
+    h2_io *io = val;
     if (io && io->request) {
         ap_log_cerror(APLOG_MARK, APLOG_WARNING, 0, m->c, /* NO APLOGNO */
                       "->03198: h2_stream(%ld-%d): %s %s %s -> %s %d"
-                      "[orph=%d/started=%d/done=%d/eos_in=%d/eos_out=%d]", 
+                      "[orph=%d/started=%d/done=%d]", 
                       m->id, io->id, 
                       io->request->method, io->request->authority, io->request->path,
                       io->response? "http" : (io->rst_error? "reset" : "?"),
                       io->response? io->response->http_status : io->rst_error,
-                      io->orphaned, io->worker_started, io->worker_done,
-                      io->eos_in, io->eos_out);
+                      io->orphaned, io->worker_started, io->worker_done);
     }
     else if (io) {
         ap_log_cerror(APLOG_MARK, APLOG_WARNING, 0, m->c, /* NO APLOGNO */
                       "->03198: h2_stream(%ld-%d): NULL -> %s %d"
-                      "[orph=%d/started=%d/done=%d/eos_in=%d/eos_out=%d]", 
+                      "[orph=%d/started=%d/done=%d]", 
                       m->id, io->id, 
                       io->response? "http" : (io->rst_error? "reset" : "?"),
                       io->response? io->response->http_status : io->rst_error,
-                      io->orphaned, io->worker_started, io->worker_done,
-                      io->eos_in, io->eos_out);
+                      io->orphaned, io->worker_started, io->worker_done);
     }
     else {
         ap_log_cerror(APLOG_MARK, APLOG_WARNING, 0, m->c, /* NO APLOGNO */
@@ -392,7 +423,7 @@ apr_status_t h2_mplx_release_and_join(h2_mplx *m, apr_thread_cond_t *wait)
         
         h2_iq_clear(m->q);
         apr_thread_cond_broadcast(m->task_thawed);
-        while (!h2_io_set_iter(m->stream_ios, stream_done_iter, m)) {
+        while (!h2_ilist_iter(m->stream_ios, stream_done_iter, m)) {
             /* iterate until all ios have been orphaned or destroyed */
         }
     
@@ -407,9 +438,13 @@ apr_status_t h2_mplx_release_and_join(h2_mplx *m, apr_thread_cond_t *wait)
             m->join_wait = wait;
             ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, m->c,
                           "h2_mplx(%ld): release_join, waiting on %d worker to report back", 
-                          m->id, (int)h2_io_set_size(m->stream_ios));
+                          m->id, (int)h2_ilist_count(m->stream_ios));
                           
             status = apr_thread_cond_timedwait(wait, m->lock, apr_time_from_sec(wait_secs));
+            
+            while (!h2_ilist_iter(m->stream_ios, stream_done_iter, m)) {
+                /* iterate until all ios have been orphaned or destroyed */
+            }
             if (APR_STATUS_IS_TIMEUP(status)) {
                 if (i > 0) {
                     /* Oh, oh. Still we wait for assigned  workers to report that 
@@ -421,9 +456,9 @@ apr_status_t h2_mplx_release_and_join(h2_mplx *m, apr_thread_cond_t *wait)
                                   "h2_mplx(%ld): release, waiting for %d seconds now for "
                                   "%d h2_workers to return, have still %d requests outstanding", 
                                   m->id, i*wait_secs, m->workers_busy,
-                                  (int)h2_io_set_size(m->stream_ios));
+                                  (int)h2_ilist_count(m->stream_ios));
                     if (i == 1) {
-                        h2_io_set_iter(m->stream_ios, stream_print, m);
+                        h2_ilist_iter(m->stream_ios, stream_print, m);
                     }
                 }
                 h2_mplx_abort(m);
@@ -431,10 +466,10 @@ apr_status_t h2_mplx_release_and_join(h2_mplx *m, apr_thread_cond_t *wait)
             }
         }
         
-        if (!h2_io_set_is_empty(m->stream_ios)) {
+        if (!h2_ilist_empty(m->stream_ios)) {
             ap_log_cerror(APLOG_MARK, APLOG_INFO, 0, m->c, 
                           "h2_mplx(%ld): release_join, %d streams still open", 
-                          m->id, (int)h2_io_set_size(m->stream_ios));
+                          m->id, (int)h2_ilist_count(m->stream_ios));
         }
         ap_log_cerror(APLOG_MARK, APLOG_DEBUG, 0, m->c, APLOGNO(03056)
                       "h2_mplx(%ld): release_join -> destroy", m->id);
@@ -468,7 +503,7 @@ apr_status_t h2_mplx_stream_done(h2_mplx *m, int stream_id, int rst_error)
      */
     AP_DEBUG_ASSERT(m);
     if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
+        h2_io *io = h2_ilist_get(m->stream_ios, stream_id);
 
         /* there should be an h2_io, once the stream has been scheduled
          * for processing, e.g. when we received all HEADERs. But when
@@ -484,107 +519,16 @@ apr_status_t h2_mplx_stream_done(h2_mplx *m, int stream_id, int rst_error)
     return status;
 }
 
-apr_status_t h2_mplx_in_read(h2_mplx *m, apr_read_type_e block,
-                             int stream_id, apr_bucket_brigade *bb, 
-                             apr_table_t *trailers,
-                             struct apr_thread_cond_t *iowait)
-{
-    apr_status_t status; 
-    int acquired;
-    
-    AP_DEBUG_ASSERT(m);
-    if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-        if (io && !io->orphaned) {
-            H2_MPLX_IO_IN(APLOG_TRACE2, m, io, "h2_mplx_in_read_pre");
-            
-            h2_io_signal_init(io, H2_IO_READ, m->stream_timeout, iowait);
-            status = h2_io_in_read(io, bb, -1, trailers);
-            while (APR_STATUS_IS_EAGAIN(status) 
-                   && !is_aborted(m, &status)
-                   && block == APR_BLOCK_READ) {
-                ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, m->c,
-                              "h2_mplx(%ld-%d): wait on in data (BLOCK_READ)", 
-                              m->id, stream_id);
-                status = h2_io_signal_wait(m, io);
-                if (status == APR_SUCCESS) {
-                    status = h2_io_in_read(io, bb, -1, trailers);
-                }
-            }
-            H2_MPLX_IO_IN(APLOG_TRACE2, m, io, "h2_mplx_in_read_post");
-            h2_io_signal_exit(io);
-        }
-        else {
-            status = APR_EOF;
-        }
-        leave_mutex(m, acquired);
-    }
-    return status;
-}
-
-apr_status_t h2_mplx_in_write(h2_mplx *m, int stream_id, 
-                              const char *data, apr_size_t len, int eos)
-{
-    apr_status_t status;
-    int acquired;
-    
-    AP_DEBUG_ASSERT(m);
-    if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-        if (io && !io->orphaned) {
-            H2_MPLX_IO_IN(APLOG_TRACE2, m, io, "h2_mplx_in_write_pre");
-            status = h2_io_in_write(io, data, len, eos);
-            H2_MPLX_IO_IN(APLOG_TRACE2, m, io, "h2_mplx_in_write_post");
-            h2_io_signal(io, H2_IO_READ);
-            io_in_consumed_signal(m, io);
-        }
-        else {
-            status = APR_ECONNABORTED;
-        }
-        leave_mutex(m, acquired);
-    }
-    return status;
-}
-
-apr_status_t h2_mplx_in_close(h2_mplx *m, int stream_id)
-{
-    apr_status_t status;
-    int acquired;
-    
-    AP_DEBUG_ASSERT(m);
-    if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-        if (io && !io->orphaned) {
-            status = h2_io_in_close(io);
-            H2_MPLX_IO_IN(APLOG_TRACE2, m, io, "h2_mplx_in_close");
-            h2_io_signal(io, H2_IO_READ);
-            io_in_consumed_signal(m, io);
-        }
-        else {
-            status = APR_ECONNABORTED;
-        }
-        leave_mutex(m, acquired);
-    }
-    return status;
-}
-
 void h2_mplx_set_consumed_cb(h2_mplx *m, h2_mplx_consumed_cb *cb, void *ctx)
 {
     m->input_consumed = cb;
     m->input_consumed_ctx = ctx;
 }
 
-typedef struct {
-    h2_mplx * m;
-    int streams_updated;
-} update_ctx;
-
-static int update_window(void *ctx, h2_io *io)
+static int update_window(void *ctx, void *val)
 {
-    update_ctx *uctx = (update_ctx*)ctx;
-    if (io_in_consumed_signal(uctx->m, io)) {
-        ++uctx->streams_updated;
-    }
+    h2_mplx *m = ctx;
+    io_in_consumed_signal(m, val);
     return 1;
 }
 
@@ -598,46 +542,11 @@ apr_status_t h2_mplx_in_update_windows(h2_mplx *m)
         return APR_ECONNABORTED;
     }
     if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        update_ctx ctx;
+        h2_ilist_iter(m->stream_ios, update_window, m);
         
-        ctx.m               = m;
-        ctx.streams_updated = 0;
-
-        status = APR_EAGAIN;
-        h2_io_set_iter(m->stream_ios, update_window, &ctx);
-        
-        if (ctx.streams_updated) {
-            status = APR_SUCCESS;
-        }
-        leave_mutex(m, acquired);
-    }
-    return status;
-}
-
-apr_status_t h2_mplx_out_get_brigade(h2_mplx *m, int stream_id, 
-                                     apr_bucket_brigade *bb, 
-                                     apr_off_t len, apr_table_t **ptrailers)
-{
-    apr_status_t status;
-    int acquired;
-
-    AP_DEBUG_ASSERT(m);
-    if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-        if (io && !io->orphaned) {
-            H2_MPLX_IO_OUT(APLOG_TRACE2, m, io, "h2_mplx_out_get_brigade_pre");
-            
-            status = h2_io_out_get_brigade(io, bb, len);
-            
-            H2_MPLX_IO_OUT(APLOG_TRACE2, m, io, "h2_mplx_out_get_brigade_post");
-            if (status == APR_SUCCESS) {
-                h2_io_signal(io, H2_IO_WRITE);
-            }
-        }
-        else {
-            status = APR_ECONNABORTED;
-        }
-        *ptrailers = io->response? io->response->trailers : NULL;
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE2, 0, m->c, 
+                      "h2_session(%ld): windows updated", m->id);
+        status = APR_SUCCESS;
         leave_mutex(m, acquired);
     }
     return status;
@@ -651,7 +560,7 @@ h2_stream *h2_mplx_next_submit(h2_mplx *m, h2_ihash_t *streams)
 
     AP_DEBUG_ASSERT(m);
     if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_shift(m->ready_ios);
+        h2_io *io = h2_ilist_shift(m->ready_ios);
         if (io && !m->aborted) {
             stream = h2_ihash_get(streams, io->id);
             if (stream) {
@@ -661,9 +570,7 @@ h2_stream *h2_mplx_next_submit(h2_mplx *m, h2_ihash_t *streams)
                 }
                 else {
                     AP_DEBUG_ASSERT(io->response);
-                    H2_MPLX_IO_OUT(APLOG_TRACE2, m, io, "h2_mplx_next_submit_pre");
-                    h2_stream_set_response(stream, io->response, io->bbout);
-                    H2_MPLX_IO_OUT(APLOG_TRACE2, m, io, "h2_mplx_next_submit_post");
+                    h2_stream_set_response(stream, io->response, io->beam_out);
                 }
             }
             else {
@@ -675,114 +582,62 @@ h2_stream *h2_mplx_next_submit(h2_mplx *m, h2_ihash_t *streams)
                               "h2_mplx(%ld): stream for response %d closed, "
                               "resetting io to close request processing",
                               m->id, io->id);
-                h2_io_make_orphaned(io, H2_ERR_STREAM_CLOSED);
+                io->orphaned = 1;
+                h2_io_rst(io, H2_ERR_STREAM_CLOSED);
                 if (!io->worker_started || io->worker_done) {
                     io_destroy(m, io, 1);
                 }
                 else {
                     /* hang around until the h2_task is done, but
-                     * shutdown input and send out any events (e.g. window
-                     * updates) asap. */
-                    h2_io_in_shutdown(io);
+                     * shutdown input/output and send out any events asap. */
+                    h2_io_shutdown(io);
                     io_in_consumed_signal(m, io);
                 }
             }
-            
-            h2_io_signal(io, H2_IO_WRITE);
         }
         leave_mutex(m, acquired);
     }
     return stream;
 }
 
-static apr_status_t out_write(h2_mplx *m, h2_io *io, 
-                              ap_filter_t* f, int blocking,
-                              apr_bucket_brigade *bb,
-                              struct apr_thread_cond_t *iowait)
+static apr_status_t out_open(h2_mplx *m, int stream_id, h2_response *response,
+                             h2_bucket_beam *output)
 {
     apr_status_t status = APR_SUCCESS;
-    /* We check the memory footprint queued for this stream_id
-     * and block if it exceeds our configured limit.
-     * We will not split buckets to enforce the limit to the last
-     * byte. After all, the bucket is already in memory.
-     */
-    while (status == APR_SUCCESS 
-           && !APR_BRIGADE_EMPTY(bb) 
-           && !is_aborted(m, &status)) {
-        
-        status = h2_io_out_write(io, bb, blocking? m->stream_max_mem : INT_MAX, 
-                                 &m->tx_handles_reserved);
-        io_out_consumed_signal(m, io);
-        
-        /* Wait for data to drain until there is room again or
-         * stream timeout expires */
-        h2_io_signal_init(io, H2_IO_WRITE, m->stream_timeout, iowait);
-        while (status == APR_SUCCESS
-               && !APR_BRIGADE_EMPTY(bb) 
-               && iowait
-               && (m->stream_max_mem <= h2_io_out_length(io))
-               && !is_aborted(m, &status)) {
-            if (!blocking) {
-                ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
-                              "h2_mplx(%ld-%d): incomplete write", 
-                              m->id, io->id);
-                return APR_INCOMPLETE;
-            }
-            if (f) {
-                ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                              "h2_mplx(%ld-%d): waiting for out drain", 
-                              m->id, io->id);
-            }
-            status = h2_io_signal_wait(m, io);
-        }
-        h2_io_signal_exit(io);
+    
+    h2_io *io = h2_ilist_get(m->stream_ios, stream_id);
+    if (!io || io->orphaned) {
+        return APR_ECONNABORTED;
     }
-    apr_brigade_cleanup(bb);
     
-    return status;
-}
-
-static apr_status_t out_open(h2_mplx *m, int stream_id, h2_response *response,
-                             ap_filter_t* f, apr_bucket_brigade *bb,
-                             struct apr_thread_cond_t *iowait)
-{
-    apr_status_t status = APR_SUCCESS;
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, m->c,
+                  "h2_mplx(%ld-%d): open response: %d, rst=%d",
+                  m->id, stream_id, response->http_status, 
+                  response->rst_error);
     
-    h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-    if (io && !io->orphaned) {
-        if (f) {
-            ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
-                          "h2_mplx(%ld-%d): open response: %d, rst=%d",
-                          m->id, stream_id, response->http_status, 
-                          response->rst_error);
-        }
-        
-        h2_io_set_response(io, response);
-        h2_io_set_add(m->ready_ios, io);
-        if (response && response->http_status < 300) {
-            /* we might see some file buckets in the output, see
-             * if we have enough handles reserved. */
-            check_tx_reservation(m);
-        }
-        if (bb) {
-            status = out_write(m, io, f, 0, bb, iowait);
-            if (status == APR_INCOMPLETE) {
-                /* write will have transferred as much data as possible.
-                   caller has to deal with non-empty brigade */
-                status = APR_SUCCESS;
-            }
-        }
-        have_out_data_for(m, stream_id);
+    if (output) {
+        h2_beam_buffer_size_set(output, m->stream_max_mem);
+        h2_beam_timeout_set(output, m->stream_timeout);
+        h2_beam_on_consumed(output, stream_output_consumed, io);
+        m->tx_handles_reserved -= h2_beam_get_files_beamed(output);
+        h2_beam_on_file_beam(output, can_beam_file, m);
+        h2_beam_mutex_set(output, io_mutex_enter, io_mutex_leave, 
+                          io->task->cond, m);
     }
-    else {
-        status = APR_ECONNABORTED;
+    h2_io_set_response(io, response, output);
+    
+    h2_ilist_add(m->ready_ios, io);
+    if (response && response->http_status < 300) {
+        /* we might see some file buckets in the output, see
+         * if we have enough handles reserved. */
+        check_tx_reservation(m);
     }
+    have_out_data_for(m, stream_id);
     return status;
 }
 
 apr_status_t h2_mplx_out_open(h2_mplx *m, int stream_id, h2_response *response,
-                              ap_filter_t* f, apr_bucket_brigade *bb,
-                              struct apr_thread_cond_t *iowait)
+                              h2_bucket_beam *output)
 {
     apr_status_t status;
     int acquired;
@@ -793,37 +648,7 @@ apr_status_t h2_mplx_out_open(h2_mplx *m, int stream_id, h2_response *response,
             status = APR_ECONNABORTED;
         }
         else {
-            status = out_open(m, stream_id, response, f, bb, iowait);
-            if (APLOGctrace1(m->c)) {
-                h2_util_bb_log(m->c, stream_id, APLOG_TRACE1, "h2_mplx_out_open", bb);
-            }
-        }
-        leave_mutex(m, acquired);
-    }
-    return status;
-}
-
-apr_status_t h2_mplx_out_write(h2_mplx *m, int stream_id, 
-                               ap_filter_t* f, int blocking,
-                               apr_bucket_brigade *bb,
-                               struct apr_thread_cond_t *iowait)
-{
-    apr_status_t status;
-    int acquired;
-    
-    AP_DEBUG_ASSERT(m);
-    if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-        if (io && !io->orphaned) {
-            status = out_write(m, io, f, blocking, bb, iowait);
-            ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, m->c,
-                          "h2_mplx(%ld-%d): write", m->id, io->id);
-            H2_MPLX_IO_OUT(APLOG_TRACE2, m, io, "h2_mplx_out_write");
-            
-            have_out_data_for(m, stream_id);
-        }
-        else {
-            status = APR_ECONNABORTED;
+            status = out_open(m, stream_id, response, output);
         }
         leave_mutex(m, acquired);
     }
@@ -837,7 +662,7 @@ apr_status_t h2_mplx_out_close(h2_mplx *m, int stream_id)
     
     AP_DEBUG_ASSERT(m);
     if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
+        h2_io *io = h2_ilist_get(m->stream_ios, stream_id);
         if (io && !io->orphaned) {
             if (!io->response && !io->rst_error) {
                 /* In case a close comes before a response was created,
@@ -846,45 +671,20 @@ apr_status_t h2_mplx_out_close(h2_mplx *m, int stream_id)
                  */
                 h2_response *r = h2_response_die(stream_id, APR_EGENERAL, 
                                                  io->request, m->pool);
-                status = out_open(m, stream_id, r, NULL, NULL, NULL);
+                status = out_open(m, stream_id, r, NULL);
                 ap_log_cerror(APLOG_MARK, APLOG_DEBUG, status, m->c,
                               "h2_mplx(%ld-%d): close, no response, no rst", 
                               m->id, io->id);
             }
             ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, m->c,
-                          "h2_mplx(%ld-%d): close with eor=%s", 
-                          m->id, io->id, io->eor? "yes" : "no");
-            status = h2_io_out_close(io);
-            H2_MPLX_IO_OUT(APLOG_TRACE2, m, io, "h2_mplx_out_close");
-            io_out_consumed_signal(m, io);
-            
-            have_out_data_for(m, stream_id);
-        }
-        else {
-            status = APR_ECONNABORTED;
-        }
-        leave_mutex(m, acquired);
-    }
-    return status;
-}
-
-apr_status_t h2_mplx_out_rst(h2_mplx *m, int stream_id, int error)
-{
-    apr_status_t status;
-    int acquired;
-    
-    AP_DEBUG_ASSERT(m);
-    if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-        if (io && !io->rst_error && !io->orphaned) {
-            h2_io_rst(io, error);
-            if (!io->response) {
-                h2_io_set_add(m->ready_ios, io);
+                          "h2_mplx(%ld-%d): close", m->id, io->id);
+            if (io->beam_out) {
+                status = h2_beam_close(io->beam_out);
+                h2_beam_log(io->beam_out, stream_id, "out_close", m->c, 
+                            APLOG_TRACE2);
             }
-            H2_MPLX_IO_OUT(APLOG_TRACE2, m, io, "h2_mplx_out_rst");
-            
+            io_out_consumed_signal(m, io);
             have_out_data_for(m, stream_id);
-            h2_io_signal(io, H2_IO_WRITE);
         }
         else {
             status = APR_ECONNABORTED;
@@ -894,26 +694,6 @@ apr_status_t h2_mplx_out_rst(h2_mplx *m, int stream_id, int error)
     return status;
 }
 
-int h2_mplx_out_has_data_for(h2_mplx *m, int stream_id)
-{
-    apr_status_t status;
-    int has_data = 0;
-    int acquired;
-    
-    AP_DEBUG_ASSERT(m);
-    if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, stream_id);
-        if (io && !io->orphaned) {
-            has_data = h2_io_out_has_data(io);
-        }
-        else {
-            has_data = 0;
-        }
-        leave_mutex(m, acquired);
-    }
-    return has_data;
-}
-
 apr_status_t h2_mplx_out_trywait(h2_mplx *m, apr_interval_time_t timeout,
                                  apr_thread_cond_t *iowait)
 {
@@ -969,22 +749,7 @@ apr_status_t h2_mplx_reprioritize(h2_mplx *m, h2_stream_pri_cmp *cmp, void *ctx)
     return status;
 }
 
-static h2_io *open_io(h2_mplx *m, int stream_id, const h2_request *request)
-{
-    apr_pool_t *io_pool;
-    h2_io *io;
-    
-    apr_pool_create(&io_pool, m->pool);
-    apr_pool_tag(io_pool, "h2_io");
-    io = h2_io_create(stream_id, io_pool, m->bucket_alloc, request);
-    h2_io_set_add(m->stream_ios, io);
-    
-    return io;
-}
-
-
-apr_status_t h2_mplx_process(h2_mplx *m, int stream_id, 
-                             const h2_request *req, 
+apr_status_t h2_mplx_process(h2_mplx *m, struct h2_stream *stream, 
                              h2_stream_pri_cmp *cmp, void *ctx)
 {
     apr_status_t status;
@@ -997,24 +762,38 @@ apr_status_t h2_mplx_process(h2_mplx *m, int stream_id,
             status = APR_ECONNABORTED;
         }
         else {
-            h2_io *io = open_io(m, stream_id, req);
+            apr_pool_t *io_pool;
+            h2_io *io;
             
-            if (!io->request->body) {
-                status = h2_io_in_close(io);
+            if (!m->need_registration) {
+                m->need_registration = h2_iq_empty(m->q);
             }
-            
-            m->need_registration = m->need_registration || h2_iq_empty(m->q);
-            do_registration = (m->need_registration && m->workers_busy < m->workers_max);
+            if (m->workers_busy < m->workers_max) {
+                do_registration = m->need_registration;
+            }
+
+            io_pool = m->spare_io_pool;
+            if (io_pool) {
+                m->spare_io_pool = NULL;
+            }
+            else {
+                apr_pool_create(&io_pool, m->pool);
+                apr_pool_tag(io_pool, "h2_io");
+            }
+            io = h2_io_create(stream->id, io_pool, stream->request);
+            h2_ilist_add(m->stream_ios, io);            
             h2_iq_add(m->q, io->id, cmp, ctx);
             
+            stream->input = io->beam_in;
             ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, m->c,
-                          "h2_mplx(%ld-%d): process", m->c->id, stream_id);
-            H2_MPLX_IO_IN(APLOG_TRACE2, m, io, "h2_mplx_process");
+                          "h2_mplx(%ld-%d): process, body=%d", 
+                          m->c->id, stream->id, io->request->body);
         }
         leave_mutex(m, acquired);
     }
-    if (status == APR_SUCCESS && do_registration) {
-        workers_register(m);
+    if (do_registration) {
+        m->need_registration = 0;
+        h2_workers_register(m->workers, m);
     }
     return status;
 }
@@ -1022,20 +801,24 @@ apr_status_t h2_mplx_process(h2_mplx *m, int stream_id,
 static h2_task *pop_task(h2_mplx *m)
 {
     h2_task *task = NULL;
+    h2_io *io;
     int sid;
-    while (!m->aborted && !task 
-        && (m->workers_busy < m->workers_limit)
-        && (sid = h2_iq_shift(m->q)) > 0) {
-        h2_io *io = h2_io_set_get(m->stream_ios, sid);
-        if (io && io->orphaned) {
-            io_destroy(m, io, 0);
-            if (m->join_wait) {
-                apr_thread_cond_signal(m->join_wait);
-            }
-        }
-        else if (io) {
+    while (!m->aborted && !task  && (m->workers_busy < m->workers_limit)
+           && (sid = h2_iq_shift(m->q)) > 0) {
+        
+        io = h2_ilist_get(m->stream_ios, sid);
+        if (io) {
             conn_rec *slave, **pslave;
             
+            if (io->orphaned) {
+                /* TODO: add to purge list */
+                io_destroy(m, io, 0);
+                if (m->join_wait) {
+                    apr_thread_cond_signal(m->join_wait);
+                }
+                continue;
+            }
+            
             pslave = (conn_rec **)apr_array_pop(m->spare_slaves);
             if (pslave) {
                 slave = *pslave;
@@ -1046,12 +829,21 @@ static h2_task *pop_task(h2_mplx *m)
             }
             
             slave->sbh = m->c->sbh;
-            io->task = task = h2_task_create(m->id, io->request, slave, m);
+            io->task = task = h2_task_create(slave, io->request, 
+                                             io->beam_in, m);
             m->c->keepalives++;
             apr_table_setn(slave->notes, H2_TASK_ID_NOTE, task->id);
             
             io->worker_started = 1;
             io->started_at = apr_time_now();
+            
+            if (io->beam_in) {
+                h2_beam_timeout_set(io->beam_in, m->stream_timeout);
+                h2_beam_on_consumed(io->beam_in, stream_input_consumed, m);
+                h2_beam_on_file_beam(io->beam_in, can_beam_file, m);
+                h2_beam_mutex_set(io->beam_in, io_mutex_enter, 
+                                  io_mutex_leave, task->cond, m);
+            }
             if (sid > m->max_stream_started) {
                 m->max_stream_started = sid;
             }
@@ -1088,7 +880,7 @@ h2_task *h2_mplx_pop_task(h2_mplx *m, int *has_more)
 static void task_done(h2_mplx *m, h2_task *task, h2_req_engine *ngn)
 {
     if (task) {
-        h2_io *io = h2_io_set_get(m->stream_ios, task->stream_id);
+        h2_io *io = h2_ilist_get(m->stream_ios, task->stream_id);
         
         if (task->frozen) {
             /* this task was handed over to an engine for processing 
@@ -1112,14 +904,17 @@ static void task_done(h2_mplx *m, h2_task *task, h2_req_engine *ngn)
             h2_mplx_out_close(m, task->stream_id);
             
             if (ngn && io) {
-                apr_off_t bytes = io->output_consumed + h2_io_out_length(io);
+                apr_off_t bytes = 0;
+                if (io->beam_out) {
+                    h2_beam_send(io->beam_out, NULL, APR_NONBLOCK_READ);
+                    bytes += h2_beam_get_buffered(io->beam_out);
+                }
                 if (bytes > 0) {
                     /* we need to report consumed and current buffered output
                      * to the engine. The request will be streamed out or cancelled,
                      * no more data is coming from it and the engine should update
                      * its calculations before we destroy this information. */
                     h2_req_engine_out_consumed(ngn, task->c, bytes);
-                    io->output_consumed = 0;
                 }
             }
             
@@ -1136,10 +931,10 @@ static void task_done(h2_mplx *m, h2_task *task, h2_req_engine *ngn)
             if (io) {
                 apr_time_t now = apr_time_now();
                 if (!io->orphaned && m->redo_ios
-                    && h2_io_set_get(m->redo_ios, io->id)) {
+                    && h2_ilist_get(m->redo_ios, io->id)) {
                     /* reset and schedule again */
                     h2_io_redo(io);
-                    h2_io_set_remove(m->redo_ios, io);
+                    h2_ilist_remove(m->redo_ios, io->id);
                     h2_iq_add(m->q, io->id, NULL, NULL);
                 }
                 else {
@@ -1168,6 +963,7 @@ static void task_done(h2_mplx *m, h2_task *task, h2_req_engine *ngn)
                 }
                 
                 if (io->orphaned) {
+                    /* TODO: add to purge list */
                     io_destroy(m, io, 0);
                     if (m->join_wait) {
                         apr_thread_cond_signal(m->join_wait);
@@ -1211,12 +1007,12 @@ typedef struct {
     apr_time_t now;
 } io_iter_ctx;
 
-static int latest_repeatable_busy_unsubmitted_iter(void *data, h2_io *io)
+static int latest_repeatable_busy_unsubmitted_iter(void *data, void *val)
 {
     io_iter_ctx *ctx = data;
+    h2_io *io = val;
     if (io->worker_started && !io->worker_done
-        && h2_io_is_repeatable(io)
-        && !h2_io_set_get(ctx->m->redo_ios, io->id)) {
+        && h2_io_can_redo(io) && !h2_ilist_get(ctx->m->redo_ios, io->id)) {
         /* this io occupies a worker, the response has not been submitted yet,
          * not been cancelled and it is a repeatable request
          * -> it can be re-scheduled later */
@@ -1233,13 +1029,14 @@ static h2_io *get_latest_repeatable_busy_unsubmitted_io(h2_mplx *m)
     io_iter_ctx ctx;
     ctx.m = m;
     ctx.io = NULL;
-    h2_io_set_iter(m->stream_ios, latest_repeatable_busy_unsubmitted_iter, &ctx);
+    h2_ilist_iter(m->stream_ios, latest_repeatable_busy_unsubmitted_iter, &ctx);
     return ctx.io;
 }
 
-static int timed_out_busy_iter(void *data, h2_io *io)
+static int timed_out_busy_iter(void *data, void *val)
 {
     io_iter_ctx *ctx = data;
+    h2_io *io = val;
     if (io->worker_started && !io->worker_done
         && (ctx->now - io->started_at) > ctx->m->stream_timeout) {
         /* timed out stream occupying a worker, found */
@@ -1254,7 +1051,7 @@ static h2_io *get_timed_out_busy_stream(h2_mplx *m)
     ctx.m = m;
     ctx.io = NULL;
     ctx.now = apr_time_now();
-    h2_io_set_iter(m->stream_ios, timed_out_busy_iter, &ctx);
+    h2_ilist_iter(m->stream_ios, timed_out_busy_iter, &ctx);
     return ctx.io;
 }
 
@@ -1264,19 +1061,19 @@ static apr_status_t unschedule_slow_ios(h2_mplx *m)
     int n;
     
     if (!m->redo_ios) {
-        m->redo_ios = h2_io_set_create(m->pool);
+        m->redo_ios = h2_ilist_create(m->pool);
     }
     /* Try to get rid of streams that occupy workers. Look for safe requests
      * that are repeatable. If none found, fail the connection.
      */
-    n = (m->workers_busy - m->workers_limit - h2_io_set_size(m->redo_ios));
+    n = (m->workers_busy - m->workers_limit - h2_ilist_count(m->redo_ios));
     while (n > 0 && (io = get_latest_repeatable_busy_unsubmitted_io(m))) {
-        h2_io_set_add(m->redo_ios, io);
+        h2_ilist_add(m->redo_ios, io);
         h2_io_rst(io, H2_ERR_CANCEL);
         --n;
     }
     
-    if ((m->workers_busy - h2_io_set_size(m->redo_ios)) > m->workers_limit) {
+    if ((m->workers_busy - h2_ilist_count(m->redo_ios)) > m->workers_limit) {
         io = get_timed_out_busy_stream(m);
         if (io) {
             /* Too many busy workers, unable to cancel enough streams
@@ -1295,7 +1092,7 @@ apr_status_t h2_mplx_idle(h2_mplx *m)
     int acquired;
     
     if (enter_mutex(m, &acquired) == APR_SUCCESS) {
-        apr_size_t scount = h2_io_set_size(m->stream_ios);
+        apr_size_t scount = h2_ilist_count(m->stream_ios);
         if (scount > 0 && m->workers_busy) {
             /* If we have streams in connection state 'IDLE', meaning
              * all streams are ready to sent data out, but lack
@@ -1350,9 +1147,10 @@ typedef struct {
     int streams_updated;
 } ngn_update_ctx;
 
-static int ngn_update_window(void *ctx, h2_io *io)
+static int ngn_update_window(void *ctx, void *val)
 {
     ngn_update_ctx *uctx = ctx;
+    h2_io *io = val;
     if (io && io->task && io->task->assigned == uctx->ngn
         && io_out_consumed_signal(uctx->m, io)) {
         ++uctx->streams_updated;
@@ -1367,7 +1165,7 @@ static apr_status_t ngn_out_update_windows(h2_mplx *m, h2_req_engine *ngn)
     ctx.m = m;
     ctx.ngn = ngn;
     ctx.streams_updated = 0;
-    h2_io_set_iter(m->stream_ios, ngn_update_window, &ctx);
+    h2_ilist_iter(m->stream_ios, ngn_update_window, &ctx);
     
     return ctx.streams_updated? APR_SUCCESS : APR_EAGAIN;
 }
@@ -1389,7 +1187,7 @@ apr_status_t h2_mplx_req_engine_push(const char *ngn_type,
     task->r = r;
     
     if ((status = enter_mutex(m, &acquired)) == APR_SUCCESS) {
-        h2_io *io = h2_io_set_get(m->stream_ios, task->stream_id);
+        h2_io *io = h2_ilist_get(m->stream_ios, task->stream_id);
         if (!io || io->orphaned) {
             status = APR_ECONNABORTED;
         }
index 40298476789e7d3db546501a64f2a0201d8d7fcd..7e6f5bceadfcd878be24438ffa5a825e15475736 100644 (file)
 struct apr_pool_t;
 struct apr_thread_mutex_t;
 struct apr_thread_cond_t;
+struct h2_bucket_beam;
 struct h2_config;
 struct h2_ihash_t;
+struct h2_ilist_t;
 struct h2_response;
 struct h2_task;
 struct h2_stream;
 struct h2_request;
-struct h2_io_set;
 struct apr_thread_cond_t;
 struct h2_workers;
-struct h2_int_queue;
+struct h2_iqueue;
 struct h2_ngn_shed;
 struct h2_req_engine;
 
@@ -72,10 +73,10 @@ struct h2_mplx {
     unsigned int aborted : 1;
     unsigned int need_registration : 1;
 
-    struct h2_int_queue *q;
-    struct h2_io_set *stream_ios;
-    struct h2_io_set *ready_ios;
-    struct h2_io_set *redo_ios;
+    struct h2_iqueue *q;
+    struct h2_ilist_t *stream_ios;
+    struct h2_ilist_t *ready_ios;
+    struct h2_ilist_t *redo_ios;
     
     apr_uint32_t max_streams;        /* max # of concurrent streams */
     apr_uint32_t max_stream_started; /* highest stream id that started processing */
@@ -96,10 +97,11 @@ struct h2_mplx {
     apr_size_t stream_max_mem;
     apr_interval_time_t stream_timeout;
     
+    apr_pool_t *spare_io_pool;
     apr_array_header_t *spare_slaves; /* spare slave connections */
     
     struct h2_workers *workers;
-    apr_size_t tx_handles_reserved;
+    int tx_handles_reserved;
     apr_size_t tx_chunk_size;
     
     h2_mplx_consumed_cb *input_consumed;
@@ -166,10 +168,6 @@ apr_uint32_t h2_mplx_shutdown(h2_mplx *m);
  */
 apr_status_t h2_mplx_stream_done(h2_mplx *m, int stream_id, int rst_error);
 
-/* Return != 0 iff the multiplexer has output data for the given stream. 
- */
-int h2_mplx_out_has_data_for(h2_mplx *m, int stream_id);
-
 /**
  * Waits on output data from any stream in this session to become available. 
  * Returns APR_TIMEUP if no data arrived in the given time.
@@ -190,8 +188,7 @@ apr_status_t h2_mplx_out_trywait(h2_mplx *m, apr_interval_time_t timeout,
  * @param cmp the stream priority compare function
  * @param ctx context data for the compare function
  */
-apr_status_t h2_mplx_process(h2_mplx *m, int stream_id, 
-                             const struct h2_request *r, 
+apr_status_t h2_mplx_process(h2_mplx *m, struct h2_stream *stream, 
                              h2_stream_pri_cmp *cmp, void *ctx);
 
 /**
@@ -218,38 +215,12 @@ void h2_mplx_set_consumed_cb(h2_mplx *m, h2_mplx_consumed_cb *cb, void *ctx);
  * Input handling of streams.
  ******************************************************************************/
 
-/**
- * Reads a buckets for the given stream_id. Will return ARP_EAGAIN when
- * called with APR_NONBLOCK_READ and no data present. Will return APR_EOF
- * when the end of the stream input has been reached.
- * The condition passed in will be used for blocking/signalling and will
- * be protected by the mplx's own mutex.
- */
-apr_status_t h2_mplx_in_read(h2_mplx *m, apr_read_type_e block,
-                             int stream_id, apr_bucket_brigade *bb,
-                             apr_table_t *trailers, 
-                             struct apr_thread_cond_t *iowait);
-
-/**
- * Appends data to the input of the given stream. Storage of input data is
- * not subject to flow control.
- */
-apr_status_t h2_mplx_in_write(h2_mplx *m, int stream_id, 
-                              const char *data, apr_size_t len, int eos);
-
-/**
- * Closes the input for the given stream_id.
- */
-apr_status_t h2_mplx_in_close(h2_mplx *m, int stream_id);
-
 /**
  * Invoke the consumed callback for all streams that had bytes read since the 
  * last call to this function. If no stream had input data consumed, the 
  * callback is not invoked.
  * The consumed callback may also be invoked at other times whenever
  * the need arises.
- * Returns APR_SUCCESS when an update happened, APR_EAGAIN if no update
- * happened.
  */
 apr_status_t h2_mplx_in_update_windows(h2_mplx *m);
 
@@ -266,45 +237,18 @@ apr_status_t h2_mplx_in_update_windows(h2_mplx *m);
 struct h2_stream *h2_mplx_next_submit(h2_mplx *m, 
                                       struct h2_ihash_t *streams);
 
-/**
- * Reads output data into the given brigade. Will never block, but
- * return APR_EAGAIN until data arrives or the stream is closed.
- */
-apr_status_t h2_mplx_out_get_brigade(h2_mplx *mplx, int stream_id, 
-                                     apr_bucket_brigade *bb, 
-                                     apr_off_t len, apr_table_t **ptrailers);
-
 /**
  * Opens the output for the given stream with the specified response.
  */
 apr_status_t h2_mplx_out_open(h2_mplx *mplx, int stream_id,
                               struct h2_response *response,
-                              ap_filter_t* filter, apr_bucket_brigade *bb,
-                              struct apr_thread_cond_t *iowait);
-
-/**
- * Append the brigade to the stream output. Might block if amount
- * of bytes buffered reaches configured max.
- * @param stream_id the stream identifier
- * @param filter the apache filter context of the data
- * @param blocking == 0 iff call should return with APR_INCOMPLETE if
- *                 the full brigade cannot be written at once
- * @param bb the bucket brigade to append
- * @param iowait a conditional used for block/signalling in h2_mplx
- */
-apr_status_t h2_mplx_out_write(h2_mplx *mplx, int stream_id, 
-                               ap_filter_t* filter, 
-                               int blocking,
-                               apr_bucket_brigade *bb,
-                               struct apr_thread_cond_t *iowait);
+                              struct h2_bucket_beam *output);
 
 /**
  * Closes the output for stream stream_id. 
  */
 apr_status_t h2_mplx_out_close(h2_mplx *m, int stream_id);
 
-apr_status_t h2_mplx_out_rst(h2_mplx *m, int stream_id, int error);
-
 /*******************************************************************************
  * h2_mplx list Manipulation.
  ******************************************************************************/
index 32483d9332a3e047592c59c45863dc73f496c652..48ac09fae63d4d008cd6f4924267abd55e607cfa 100644 (file)
 #include "h2_conn.h"
 #include "h2_ctx.h"
 #include "h2_h2.h"
-#include "h2_int_queue.h"
 #include "h2_mplx.h"
 #include "h2_response.h"
 #include "h2_request.h"
 #include "h2_task.h"
-#include "h2_task_output.h"
 #include "h2_util.h"
 #include "h2_ngn_shed.h"
 
index 6f0298e2ef1da404d0771fddc83e92b235394bab..19aff5bf6f896b2473a706e3f8823ddae92fa4db 100644 (file)
@@ -23,7 +23,6 @@
 
 #include "mod_http2.h"
 #include "h2.h"
-#include "h2_int_queue.h"
 #include "h2_request.h"
 #include "h2_util.h"
 #include "h2_proxy_session.h"
index 7078981c7a61ba6c473ce391683f7dbdacc9b64b..52be5c6b37b6207085921e0b509ddb81cc25f718 100644 (file)
@@ -20,7 +20,7 @@
 
 #include <nghttp2/nghttp2.h>
 
-struct h2_int_queue;
+struct h2_iqueue;
 struct h2_ihash_t;
 
 typedef enum {
@@ -74,7 +74,7 @@ struct h2_proxy_session {
     apr_interval_time_t wait_timeout;
 
     struct h2_ihash_t *streams;
-    struct h2_int_queue *suspended;
+    struct h2_iqueue *suspended;
     apr_size_t remote_max_concurrent;
     int last_stream_id;     /* last stream id processed by backend, or 0 */
     
index 8da2e9f2bf7e5fec42bd2c3c4b4ee054dfc8c49b..e491f1f1b7ef22c9ccf360f50018f07b63adbe60 100644 (file)
@@ -112,7 +112,7 @@ static void cleanup_streams(h2_session *session)
     while (1) {
         h2_ihash_iter(session->streams, find_cleanup_stream, &ctx);
         if (ctx.candidate) {
-            h2_session_stream_destroy(session, ctx.candidate);
+            h2_session_stream_done(session, ctx.candidate);
             ctx.candidate = NULL;
         }
         else {
@@ -1277,8 +1277,9 @@ static apr_status_t submit_response(h2_session *session, h2_stream *stream)
         const h2_priority *prio;
         
         ap_log_cerror(APLOG_MARK, APLOG_DEBUG, 0, session->c, APLOGNO(03073)
-                      "h2_stream(%ld-%d): submit response %d",
-                      session->id, stream->id, response->http_status);
+                      "h2_stream(%ld-%d): submit response %d, REMOTE_WINDOW_SIZE=%u",
+                      session->id, stream->id, response->http_status,
+                      (unsigned int)nghttp2_session_get_stream_remote_window_size(session->ngh2, stream->id));
         
         if (response->content_length != 0) {
             memset(&provider, 0, sizeof(provider));
@@ -1504,21 +1505,24 @@ apr_status_t h2_session_set_prio(h2_session *session, h2_stream *stream,
     return status;
 }
 
-apr_status_t h2_session_stream_destroy(h2_session *session, h2_stream *stream)
+apr_status_t h2_session_stream_done(h2_session *session, h2_stream *stream)
 {
     apr_pool_t *pool = h2_stream_detach_pool(stream);
-
+    int stream_id = stream->id;
+    int rst_error = stream->rst_error;
+    
     ap_log_cerror(APLOG_MARK, APLOG_TRACE2, 0, session->c,
                   "h2_stream(%ld-%d): cleanup by EOS bucket destroy", 
-                  session->id, stream->id);
+                  session->id, stream_id);
+    if (session->streams) {
+        h2_ihash_remove(session->streams, stream_id);
+    }
+    
+    h2_stream_cleanup(stream);
     /* this may be called while the session has already freed
      * some internal structures or even when the mplx is locked. */
     if (session->mplx) {
-        h2_mplx_stream_done(session->mplx, stream->id, stream->rst_error);
-    }
-    
-    if (session->streams) {
-        h2_ihash_remove(session->streams, stream->id);
+        h2_mplx_stream_done(session->mplx, stream_id, rst_error);
     }
     h2_stream_destroy(stream);
     
@@ -1529,6 +1533,7 @@ apr_status_t h2_session_stream_destroy(h2_session *session, h2_stream *stream)
         }
         session->spare = pool;
     }
+
     return APR_SUCCESS;
 }
 
@@ -2217,9 +2222,10 @@ apr_status_t h2_session_process(h2_session *session, int async)
                     }
                     /* send out window updates for our inputs */
                     status = h2_mplx_in_update_windows(session->mplx);
-                    if (status != APR_SUCCESS && status != APR_EAGAIN) {
+                    if (status != APR_SUCCESS) {
                         dispatch_event(session, H2_SESSION_EV_CONN_ERROR, 
-                                         H2_ERR_INTERNAL_ERROR, "window update error");
+                                       H2_ERR_INTERNAL_ERROR, 
+                                       "window update error");
                         break;
                     }
                 }
index 28eedba843f1d693c1665408b84efeb58b2da619..bf4ded338a807c619b26d91b4531a7abba5b4e94 100644 (file)
@@ -218,8 +218,8 @@ int h2_session_push_enabled(h2_session *session);
  * @param session the session to which the stream belongs
  * @param stream the stream to destroy
  */
-apr_status_t h2_session_stream_destroy(h2_session *session, 
-                                       struct h2_stream *stream);
+apr_status_t h2_session_stream_done(h2_session *session, 
+                                    struct h2_stream *stream);
 
 /**
  * Submit a push promise on the stream and schedule the new steam for
index a7979e3ffad37fd81cdb033fda27b2714e3df41b..ae980194569a64ffb7b44ac2979be15ba968837d 100644 (file)
@@ -24,6 +24,7 @@
 #include <nghttp2/nghttp2.h>
 
 #include "h2_private.h"
+#include "h2_bucket_beam.h"
 #include "h2_conn.h"
 #include "h2_config.h"
 #include "h2_h2.h"
@@ -36,7 +37,6 @@
 #include "h2_stream.h"
 #include "h2_task.h"
 #include "h2_ctx.h"
-#include "h2_task_input.h"
 #include "h2_task.h"
 #include "h2_util.h"
 
@@ -52,6 +52,13 @@ static int state_transition[][7] = {
 /*CL*/{  1, 1, 0, 0, 1, 1, 1 },
 };
 
+#define H2_STREAM_OUT_LOG(lvl,s,msg) \
+    do { \
+        if (APLOG_C_IS_LEVEL((s)->session->c,lvl)) \
+        h2_util_bb_log((s)->session->c,(s)->session->id,lvl,msg,(s)->buffer); \
+    } while(0)
+    
+
 static int set_state(h2_stream *stream, h2_stream_state_t state)
 {
     int allowed = state_transition[state][stream->state];
@@ -135,8 +142,6 @@ static int output_open(h2_stream *stream)
     }
 }
 
-static h2_sos *h2_sos_mplx_create(h2_stream *stream, h2_response *response);
-
 h2_stream *h2_stream_open(int id, apr_pool_t *pool, h2_session *session,
                           int initiated_on, const h2_request *creq)
 {
@@ -166,18 +171,26 @@ h2_stream *h2_stream_open(int id, apr_pool_t *pool, h2_session *session,
     return stream;
 }
 
-apr_status_t h2_stream_destroy(h2_stream *stream)
+void h2_stream_cleanup(h2_stream *stream)
 {
     AP_DEBUG_ASSERT(stream);
+    if (stream->buffer) {
+        apr_brigade_cleanup(stream->buffer);
+    }
+}
+
+void h2_stream_destroy(h2_stream *stream)
+{
+    AP_DEBUG_ASSERT(stream);
+    h2_stream_cleanup(stream);
     if (stream->pool) {
         apr_pool_destroy(stream->pool);
     }
-    return APR_SUCCESS;
 }
 
-void h2_stream_cleanup(h2_stream *stream)
+void h2_stream_eos_destroy(h2_stream *stream)
 {
-    h2_session_stream_destroy(stream->session, stream);
+    h2_session_stream_done(stream->session, stream);
     /* stream is gone */
 }
 
@@ -200,33 +213,7 @@ void h2_stream_rst(h2_stream *stream, int error_code)
 
 struct h2_response *h2_stream_get_response(h2_stream *stream)
 {
-    return stream->sos? stream->sos->response : NULL;
-}
-
-apr_status_t h2_stream_set_response(h2_stream *stream, h2_response *response,
-                                    apr_bucket_brigade *bb)
-{
-    apr_status_t status = APR_SUCCESS;
-    h2_sos *sos;
-    
-    if (!output_open(stream)) {
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, stream->session->c,
-                      "h2_stream(%ld-%d): output closed", 
-                      stream->session->id, stream->id);
-        return APR_ECONNRESET;
-    }
-    
-    sos = h2_sos_mplx_create(stream, response);
-    if (sos->response->sos_filter) {
-        sos = h2_filter_sos_create(sos->response->sos_filter, sos); 
-    }
-    stream->sos = sos;
-    
-    status = stream->sos->buffer(stream->sos, bb);
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, stream->session->c,
-                  "h2_stream(%ld-%d): set_response(%d)", 
-                  stream->session->id, stream->id, stream->sos->response->http_status);
-    return status;
+    return stream->response;
 }
 
 apr_status_t h2_stream_set_request(h2_stream *stream, request_rec *r)
@@ -286,15 +273,11 @@ apr_status_t h2_stream_schedule(h2_stream *stream, int eos, int push_enabled,
     status = h2_request_end_headers(stream->request, stream->pool, 
                                     eos, push_enabled);
     if (status == APR_SUCCESS) {
-        if (!eos) {
-            stream->request->body = 1;
-        }
-        stream->input_remaining = stream->request->content_length;
-        
-        status = h2_mplx_process(stream->session->mplx, stream->id, 
-                                 stream->request, cmp, ctx);
+        stream->request->body = !eos;
         stream->scheduled = 1;
+        stream->input_remaining = stream->request->content_length;
         
+        status = h2_mplx_process(stream->session->mplx, stream, cmp, ctx);
         ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, stream->session->c,
                       "h2_stream(%ld-%d): scheduled %s %s://%s%s",
                       stream->session->id, stream->id,
@@ -331,8 +314,8 @@ apr_status_t h2_stream_close_input(h2_stream *stream)
         return APR_ECONNRESET;
     }
     
-    if (close_input(stream)) {
-        status = h2_mplx_in_close(stream->session->mplx, stream->id);
+    if (close_input(stream) && stream->input) {
+        status = h2_beam_close(stream->input);
     }
     return status;
 }
@@ -340,25 +323,29 @@ apr_status_t h2_stream_close_input(h2_stream *stream)
 apr_status_t h2_stream_write_data(h2_stream *stream,
                                   const char *data, size_t len, int eos)
 {
+    conn_rec *c = stream->session->c;
     apr_status_t status = APR_SUCCESS;
     
     AP_DEBUG_ASSERT(stream);
+    if (!stream->input) {
+        return APR_EOF;
+    }
     if (input_closed(stream) || !stream->request->eoh) {
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, stream->session->c,
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, c,
                       "h2_stream(%ld-%d): writing denied, closed=%d, eoh=%d", 
                       stream->session->id, stream->id, input_closed(stream),
                       stream->request->eoh);
         return APR_EINVAL;
     }
 
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, stream->session->c,
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, c,
                   "h2_stream(%ld-%d): add %ld input bytes", 
                   stream->session->id, stream->id, (long)len);
 
     if (!stream->request->chunked) {
         stream->input_remaining -= len;
         if (stream->input_remaining < 0) {
-            ap_log_cerror(APLOG_MARK, APLOG_WARNING, 0, stream->session->c,
+            ap_log_cerror(APLOG_MARK, APLOG_WARNING, 0, c,
                           APLOGNO(02961) 
                           "h2_stream(%ld-%d): got %ld more content bytes than announced "
                           "in content-length header: %ld", 
@@ -370,10 +357,18 @@ apr_status_t h2_stream_write_data(h2_stream *stream,
         }
     }
     
-    status = h2_mplx_in_write(stream->session->mplx, stream->id, data, len, eos);
+    if (!stream->tmp) {
+        stream->tmp = apr_brigade_create(stream->pool, c->bucket_alloc);
+    }
+    apr_brigade_write(stream->tmp, NULL, NULL, data, len);
     if (eos) {
+        APR_BRIGADE_INSERT_TAIL(stream->tmp, 
+                                apr_bucket_eos_create(c->bucket_alloc)); 
         close_input(stream);
     }
+    
+    status = h2_beam_send(stream->input, stream->tmp, APR_BLOCK_READ);
+    apr_brigade_cleanup(stream->tmp);
     return status;
 }
 
@@ -392,44 +387,122 @@ int h2_stream_is_suspended(const h2_stream *stream)
     return stream->suspended;
 }
 
-apr_status_t h2_stream_out_prepare(h2_stream *stream, 
+static apr_status_t fill_buffer(h2_stream *stream, apr_size_t amount)
+{
+    if (!stream->output) {
+        return APR_EOF;
+    }
+    return h2_beam_receive(stream->output, stream->buffer, 
+                           APR_NONBLOCK_READ, amount);
+}
+
+apr_status_t h2_stream_set_response(h2_stream *stream, h2_response *response,
+                                    h2_bucket_beam *output)
+{
+    apr_status_t status = APR_SUCCESS;
+    conn_rec *c = stream->session->c;
+    
+    if (!output_open(stream)) {
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, c,
+                      "h2_stream(%ld-%d): output closed", 
+                      stream->session->id, stream->id);
+        return APR_ECONNRESET;
+    }
+    
+    stream->response = response;
+    stream->output = output;
+    stream->buffer = apr_brigade_create(stream->pool, c->bucket_alloc);
+    
+    h2_stream_filter(stream);
+    if (stream->output) {
+        status = fill_buffer(stream, 0);
+    }
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, c,
+                  "h2_stream(%ld-%d): set_response(%d)", 
+                  stream->session->id, stream->id, 
+                  stream->response->http_status);
+    return status;
+}
+
+apr_status_t h2_stream_out_prepare(h2_stream *stream,
                                    apr_off_t *plen, int *peos)
 {
+    conn_rec *c = stream->session->c;
+    apr_status_t status = APR_SUCCESS;
+    apr_off_t requested = (*plen > 0)? *plen : 32*1024;
+
     if (stream->rst_error) {
         *plen = 0;
         *peos = 1;
         return APR_ECONNRESET;
     }
 
-    AP_DEBUG_ASSERT(stream->sos);
-    return stream->sos->prepare(stream->sos, plen, peos);
+    H2_STREAM_OUT_LOG(APLOG_TRACE2, stream, "h2_stream_out_prepare_pre");
+    h2_util_bb_avail(stream->buffer, plen, peos);
+    if (!*peos && !*plen) {
+        /* try to get more data */
+        status = fill_buffer(stream, H2MIN(requested, 32*1024));
+        if (APR_STATUS_IS_EOF(status)) {
+            apr_bucket *eos = apr_bucket_eos_create(c->bucket_alloc);
+            APR_BRIGADE_INSERT_TAIL(stream->buffer, eos);
+            status = APR_SUCCESS;
+        }
+        h2_util_bb_avail(stream->buffer, plen, peos);
+    }
+    H2_STREAM_OUT_LOG(APLOG_TRACE2, stream, "h2_stream_out_prepare_post");
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, c,
+                  "h2_stream(%ld-%d): prepare, len=%ld eos=%d, trailers=%s",
+                  c->id, stream->id, (long)*plen, *peos,
+                  (stream->response && stream->response->trailers)? 
+                  "yes" : "no");
+    if (!*peos && !*plen && status == APR_SUCCESS) {
+        return APR_EAGAIN;
+    }
+    return status;
 }
 
+
 apr_status_t h2_stream_readx(h2_stream *stream, 
                              h2_io_data_cb *cb, void *ctx,
                              apr_off_t *plen, int *peos)
 {
+    conn_rec *c = stream->session->c;
+    apr_status_t status = APR_SUCCESS;
+
     if (stream->rst_error) {
         return APR_ECONNRESET;
     }
-    if (!stream->sos) {
-        return APR_EGENERAL;
+    status = h2_util_bb_readx(stream->buffer, cb, ctx, plen, peos);
+    if (status == APR_SUCCESS && !*peos && !*plen) {
+        status = APR_EAGAIN;
     }
-    return stream->sos->readx(stream->sos, cb, ctx, plen, peos);
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, c,
+                  "h2_stream(%ld-%d): readx, len=%ld eos=%d",
+                  c->id, stream->id, (long)*plen, *peos);
+    return status;
 }
 
+
 apr_status_t h2_stream_read_to(h2_stream *stream, apr_bucket_brigade *bb, 
                                apr_off_t *plen, int *peos)
 {
+    conn_rec *c = stream->session->c;
+    apr_status_t status = APR_SUCCESS;
+
     if (stream->rst_error) {
         return APR_ECONNRESET;
     }
-    if (!stream->sos) {
-        return APR_EGENERAL;
+    status = h2_append_brigade(bb, stream->buffer, plen, peos);
+    if (status == APR_SUCCESS && !*peos && !*plen) {
+        status = APR_EAGAIN;
     }
-    return stream->sos->read_to(stream->sos, bb, plen, peos);
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, c,
+                  "h2_stream(%ld-%d): read_to, len=%ld eos=%d",
+                  c->id, stream->id, (long)*plen, *peos);
+    return status;
 }
 
+
 int h2_stream_input_is_open(const h2_stream *stream) 
 {
     return input_open(stream);
@@ -474,7 +547,7 @@ apr_status_t h2_stream_submit_pushes(h2_stream *stream)
 
 apr_table_t *h2_stream_get_trailers(h2_stream *stream)
 {
-    return stream->sos? stream->sos->get_trailers(stream->sos) : NULL;
+    return stream->response? stream->response->trailers : NULL;
 }
 
 const h2_priority *h2_stream_get_priority(h2_stream *stream)
@@ -491,147 +564,3 @@ const h2_priority *h2_stream_get_priority(h2_stream *stream)
     return NULL;
 }
 
-/*******************************************************************************
- * h2_sos_mplx
- ******************************************************************************/
-
-typedef struct h2_sos_mplx {
-    h2_mplx *m;
-    apr_bucket_brigade *bb;
-    apr_bucket_brigade *tmp;
-    apr_table_t *trailers;
-    apr_off_t  buffer_size;
-} h2_sos_mplx;
-
-#define H2_SOS_MPLX_OUT(lvl,msos,msg) \
-    do { \
-        if (APLOG_C_IS_LEVEL((msos)->m->c,lvl)) \
-        h2_util_bb_log((msos)->m->c,(msos)->m->id,lvl,msg,(msos)->bb); \
-    } while(0)
-    
-
-static apr_status_t mplx_transfer(h2_sos_mplx *msos, int stream_id, 
-                                  apr_pool_t *pool)
-{
-    apr_status_t status;
-    apr_table_t *trailers = NULL;
-    
-    if (!msos->tmp) {
-        msos->tmp = apr_brigade_create(msos->bb->p, msos->bb->bucket_alloc);
-    }
-    status = h2_mplx_out_get_brigade(msos->m, stream_id, msos->tmp, 
-                                     msos->buffer_size-1, &trailers);
-    if (!APR_BRIGADE_EMPTY(msos->tmp)) {
-        h2_transfer_brigade(msos->bb, msos->tmp, pool);
-    }
-    if (trailers) {
-        msos->trailers = trailers;
-    }
-    return status;
-}
-static apr_status_t h2_sos_mplx_read_to(h2_sos *sos, apr_bucket_brigade *bb, 
-                                        apr_off_t *plen, int *peos)
-{
-    h2_sos_mplx *msos = sos->ctx;
-    apr_status_t status;
-
-    status = h2_append_brigade(bb, msos->bb, plen, peos);
-    if (status == APR_SUCCESS && !*peos && !*plen) {
-        status = APR_EAGAIN;
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, msos->m->c,
-                      "h2_stream(%ld-%d): read_to, len=%ld eos=%d",
-                      msos->m->id, sos->stream->id, (long)*plen, *peos);
-    }
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, msos->m->c,
-                  "h2_stream(%ld-%d): read_to, len=%ld eos=%d",
-                  msos->m->id, sos->stream->id, (long)*plen, *peos);
-    return status;
-}
-
-static apr_status_t h2_sos_mplx_readx(h2_sos *sos, h2_io_data_cb *cb, void *ctx,
-                                      apr_off_t *plen, int *peos)
-{
-    h2_sos_mplx *msos = sos->ctx;
-    apr_status_t status = APR_SUCCESS;
-    
-    status = h2_util_bb_readx(msos->bb, cb, ctx, plen, peos);
-    if (status == APR_SUCCESS && !*peos && !*plen) {
-        status = APR_EAGAIN;
-    }
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, msos->m->c,
-                  "h2_stream(%ld-%d): readx, len=%ld eos=%d",
-                  msos->m->id, sos->stream->id, (long)*plen, *peos);
-    return status;
-}
-
-static apr_status_t h2_sos_mplx_prepare(h2_sos *sos, apr_off_t *plen, int *peos)
-{
-    h2_sos_mplx *msos = sos->ctx;
-    apr_status_t status = APR_SUCCESS;
-    
-    H2_SOS_MPLX_OUT(APLOG_TRACE2, msos, "h2_sos_mplx prepare_pre");
-    
-    if (APR_BRIGADE_EMPTY(msos->bb)) {
-        status = mplx_transfer(msos, sos->stream->id, sos->stream->pool);
-    }
-    h2_util_bb_avail(msos->bb, plen, peos);
-    
-    H2_SOS_MPLX_OUT(APLOG_TRACE2, msos, "h2_sos_mplx prepare_post");
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, msos->m->c,
-                  "h2_stream(%ld-%d): prepare, len=%ld eos=%d, trailers=%s",
-                  msos->m->id, sos->stream->id, (long)*plen, *peos,
-                  msos->trailers? "yes" : "no");
-    if (!*peos && !*plen) {
-        status = APR_EAGAIN;
-    }
-    
-    return status;
-}
-
-static apr_table_t *h2_sos_mplx_get_trailers(h2_sos *sos)
-{
-    h2_sos_mplx *msos = sos->ctx;
-
-    return msos->trailers;
-}
-
-static apr_status_t h2_sos_mplx_buffer(h2_sos *sos, apr_bucket_brigade *bb) 
-{
-    h2_sos_mplx *msos = sos->ctx;
-    apr_status_t status = APR_SUCCESS;
-
-    if (bb && !APR_BRIGADE_EMPTY(bb)) {
-        H2_SOS_MPLX_OUT(APLOG_TRACE2, msos, "h2_sos_mplx set_response_pre");
-        status = mplx_transfer(msos, sos->stream->id, sos->stream->pool);
-        H2_SOS_MPLX_OUT(APLOG_TRACE2, msos, "h2_sos_mplx set_response_post");
-    }
-    return status;
-}
-
-static h2_sos *h2_sos_mplx_create(h2_stream *stream, h2_response *response)
-{
-    h2_sos *sos;
-    h2_sos_mplx *msos;
-    
-    msos = apr_pcalloc(stream->pool, sizeof(*msos));
-    msos->m = stream->session->mplx;
-    msos->bb = apr_brigade_create(stream->pool, msos->m->c->bucket_alloc);
-    msos->buffer_size = 32 * 1024;
-    
-    sos = apr_pcalloc(stream->pool, sizeof(*sos));
-    sos->stream = stream;
-    sos->response = response;
-    
-    sos->ctx = msos;
-    sos->buffer = h2_sos_mplx_buffer;
-    sos->prepare = h2_sos_mplx_prepare;
-    sos->readx = h2_sos_mplx_readx;
-    sos->read_to = h2_sos_mplx_read_to;
-    sos->get_trailers = h2_sos_mplx_get_trailers;
-    
-    sos->response = response;
-
-    return sos;
-}
-
index 6f142d9dc0f37e90e35d5e5f142216d437ae61f9..c851f7e64b1bcae98841b068d4baecf9f5734c47 100644 (file)
@@ -38,6 +38,7 @@ struct h2_request;
 struct h2_response;
 struct h2_session;
 struct h2_sos;
+struct h2_bucket_beam;
 
 typedef struct h2_stream h2_stream;
 
@@ -48,8 +49,14 @@ struct h2_stream {
     
     apr_pool_t *pool;           /* the memory pool for this stream */
     struct h2_request *request; /* the request made in this stream */
+    struct h2_bucket_beam *input;
+
+    struct h2_response *response;
+    struct h2_bucket_beam *output;
+    apr_bucket_brigade *buffer;
+    apr_bucket_brigade *tmp;
+
     int rst_error;              /* stream error for RST_STREAM */
-    
     unsigned int aborted   : 1; /* was aborted */
     unsigned int suspended : 1; /* DATA sending has been suspended */
     unsigned int scheduled : 1; /* stream has been scheduled */
@@ -57,7 +64,6 @@ struct h2_stream {
     
     apr_off_t input_remaining;  /* remaining bytes on input as advertised via content-length */
 
-    struct h2_sos *sos;         /* stream output source, e.g. to read output from */
     apr_off_t data_frames_sent; /* # of DATA frames sent out for this stream */
 };
 
@@ -75,12 +81,14 @@ h2_stream *h2_stream_open(int id, apr_pool_t *pool, struct h2_session *session,
                           int initiated_on, const struct h2_request *req);
 
 /**
- * Destroy any resources held by this stream. Will destroy memory pool
- * if still owned by the stream.
- *
- * @param stream the stream to destroy
+ * Cleanup any resources still held by the stream, called by last bucket.
+ */
+void h2_stream_eos_destroy(h2_stream *stream);
+
+/**
+ * Destroy memory pool if still owned by the stream.
  */
-apr_status_t h2_stream_destroy(h2_stream *stream);
+void h2_stream_destroy(h2_stream *stream);
 
 /**
  * Removes stream from h2_session and destroys it.
@@ -179,7 +187,7 @@ struct h2_response *h2_stream_get_response(h2_stream *stream);
  */
 apr_status_t h2_stream_set_response(h2_stream *stream, 
                                     struct h2_response *response,
-                                    apr_bucket_brigade *bb);
+                                    struct h2_bucket_beam *output);
 
 /**
  * Do a speculative read on the stream output to determine the 
index e764fcc832d3779080ed7cb67927de4a2e8d0109..ba67b4198f375a18f4dea279a583773b272a5028 100644 (file)
@@ -33,6 +33,7 @@
 #include <scoreboard.h>
 
 #include "h2_private.h"
+#include "h2_bucket_beam.h"
 #include "h2_conn.h"
 #include "h2_config.h"
 #include "h2_ctx.h"
 #include "h2_request.h"
 #include "h2_session.h"
 #include "h2_stream.h"
-#include "h2_task_input.h"
-#include "h2_task_output.h"
 #include "h2_task.h"
-#include "h2_ctx.h"
 #include "h2_worker.h"
+#include "h2_util.h"
+
+/*******************************************************************************
+ * task input handling
+ ******************************************************************************/
+
+static int input_ser_header(void *ctx, const char *name, const char *value) 
+{
+    h2_task *task = ctx;
+    apr_brigade_printf(task->input.bb, NULL, NULL, "%s: %s\r\n", name, value);
+    return 1;
+}
+
+static apr_status_t input_append_eos(h2_task *task, request_rec *r)
+{
+    apr_status_t status = APR_SUCCESS;
+    apr_bucket_brigade *bb = task->input.bb;
+    apr_table_t *t = task->request->trailers;
+
+    if (task->input.chunked) {
+        if (t && !apr_is_empty_table(t)) {
+            status = apr_brigade_puts(bb, NULL, NULL, "0\r\n");
+            apr_table_do(input_ser_header, task, t, NULL);
+            status = apr_brigade_puts(bb, NULL, NULL, "\r\n");
+        }
+        else {
+            status = apr_brigade_puts(bb, NULL, NULL, "0\r\n\r\n");
+        }
+    }
+    else if (r && t && !apr_is_empty_table(t)){
+        /* trailers passed in directly. */
+        apr_table_overlap(r->trailers_in, t, APR_OVERLAP_TABLES_SET);
+    }
+    task->input.eos_written = 1;
+    APR_BRIGADE_INSERT_TAIL(bb, apr_bucket_eos_create(bb->bucket_alloc));
+    return status;
+}
+
+static apr_status_t input_read(h2_task *task, ap_filter_t* f,
+                               apr_bucket_brigade* bb, ap_input_mode_t mode,
+                               apr_read_type_e block, apr_off_t readbytes)
+{
+    apr_status_t status = APR_SUCCESS;
+    apr_bucket *b, *next;
+    apr_off_t bblen = 0;
+    
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
+                  "h2_task(%s): read, mode=%d, block=%d, readbytes=%ld", 
+                  task->id, mode, block, (long)readbytes);
+    
+    if (mode == AP_MODE_INIT) {
+        return ap_get_brigade(f->c->input_filters, bb, mode, block, readbytes);
+    }
+    
+    if (f->c->aborted) {
+        return APR_ECONNABORTED;
+    }
+    
+    if (task->input.bb) {
+        /* Cleanup brigades from those nasty 0 length non-meta buckets
+         * that apr_brigade_split_line() sometimes produces. */
+        for (b = APR_BRIGADE_FIRST(task->input.bb); 
+             b != APR_BRIGADE_SENTINEL(task->input.bb); b = next) {
+            next = APR_BUCKET_NEXT(b);
+            if (b->length == 0 && !APR_BUCKET_IS_METADATA(b)) {
+                apr_bucket_delete(b);
+            } 
+        }
+        apr_brigade_length(task->input.bb, 0, &bblen);
+    }
+    
+    if (bblen == 0) {
+        if (task->input.eos_written) {
+            ap_log_cerror(APLOG_MARK, APLOG_TRACE1, APR_EOF, f->c,
+                          "h2_task(%s): read no data", task->id); 
+            return APR_EOF;
+        }
+        else if (task->input.eos) {
+            input_append_eos(task, f->r);
+        }
+    }
+    
+    while (APR_BRIGADE_EMPTY(task->input.bb)) {
+        /* Get more input data for our request. */
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
+                      "h2_task(%s): get more data from mplx, block=%d, "
+                      "readbytes=%ld, queued=%ld",
+                      task->id, block, (long)readbytes, (long)bblen);
+        
+        /* Override the block mode we get called with depending on the input's
+         * setting. */
+        if (task->input.beam) {
+            status = h2_beam_receive(task->input.beam, task->input.bb, block, 
+                                     H2MIN(readbytes, 32*1024));
+        }
+        else {
+            status = APR_EOF;
+        }
+        
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
+                      "h2_task(%s): read returned", task->id);
+        if (APR_STATUS_IS_EAGAIN(status) 
+            && (mode == AP_MODE_GETLINE || block == APR_BLOCK_READ)) {
+            /* chunked input handling does not seem to like it if we
+             * return with APR_EAGAIN from a GETLINE read... 
+             * upload 100k test on test-ser.example.org hangs */
+            status = APR_SUCCESS;
+        }
+        else if (APR_STATUS_IS_EOF(status) && !task->input.eos_written) {
+            task->input.eos = 1;
+        }
+        else if (status != APR_SUCCESS) {
+            return status;
+        }
+        
+        apr_brigade_length(task->input.bb, 0, &bblen);
+        if (bblen > 0 && task->input.chunked) {
+            /* need to add chunks since request processing expects it */
+            char buffer[128];
+            apr_bucket *b;
+            int len;
+            
+            len = apr_snprintf(buffer, H2_ALEN(buffer), "%lx\r\n", 
+                               (unsigned long)bblen);
+            b = apr_bucket_heap_create(buffer, len, NULL, 
+                                       task->input.bb->bucket_alloc);
+            APR_BRIGADE_INSERT_HEAD(task->input.bb, b);
+            status = apr_brigade_puts(task->input.bb, NULL, NULL, "\r\n");
+        }
+        
+        if (h2_util_has_eos(task->input.bb, -1)) {
+            task->input.eos = 1;
+        }
+        
+        if (task->input.eos && !task->input.eos_written) {
+            input_append_eos(task, f->r);
+        }
+        
+        if (h2_task_logio_add_bytes_in) {
+            h2_task_logio_add_bytes_in(f->c, bblen);
+        }
+    }
+    
+    h2_util_bb_log(f->c, task->stream_id, APLOG_TRACE2, 
+                   "task_input.bb", task->input.bb);
+           
+    if (APR_BRIGADE_EMPTY(task->input.bb)) {
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
+                      "h2_task(%s): no data", task->id);
+        return (block == APR_NONBLOCK_READ)? APR_EAGAIN : APR_EOF;
+    }
+    
+    if (mode == AP_MODE_EXHAUSTIVE) {
+        /* return all we have */
+        APR_BRIGADE_CONCAT(bb, task->input.bb);
+    }
+    else if (mode == AP_MODE_READBYTES) {
+        status = h2_brigade_concat_length(bb, task->input.bb, readbytes);
+    }
+    else if (mode == AP_MODE_SPECULATIVE) {
+        status = h2_brigade_copy_length(bb, task->input.bb, readbytes);
+    }
+    else if (mode == AP_MODE_GETLINE) {
+        /* we are reading a single LF line, e.g. the HTTP headers. 
+         * this has the nasty side effect to split the bucket, even
+         * though it ends with CRLF and creates a 0 length bucket */
+        status = apr_brigade_split_line(bb, task->input.bb, block, 
+                                        HUGE_STRING_LEN);
+        if (APLOGctrace1(f->c)) {
+            char buffer[1024];
+            apr_size_t len = sizeof(buffer)-1;
+            apr_brigade_flatten(bb, buffer, &len);
+            buffer[len] = 0;
+            ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
+                          "h2_task(%s): getline: %s",
+                          task->id, buffer);
+        }
+    }
+    else {
+        /* Hmm, well. There is mode AP_MODE_EATCRLF, but we chose not
+         * to support it. Seems to work. */
+        ap_log_cerror(APLOG_MARK, APLOG_ERR, APR_ENOTIMPL, f->c,
+                      APLOGNO(02942) 
+                      "h2_task, unsupported READ mode %d", mode);
+        status = APR_ENOTIMPL;
+    }
+    
+    if (APLOGctrace1(f->c)) {
+        apr_brigade_length(bb, 0, &bblen);
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
+                      "h2_task(%s): return %ld data bytes",
+                      task->id, (long)bblen);
+    }
+    return status;
+}
 
+/*******************************************************************************
+ * task input handling
+ ******************************************************************************/
+
+static apr_status_t open_response(h2_task *task)
+{
+    h2_response *response;
+    response = h2_from_h1_get_response(task->output.from_h1);
+    if (!response) {
+        /* This happens currently when ap_die(status, r) is invoked
+         * by a read request filter. */
+        ap_log_cerror(APLOG_MARK, APLOG_INFO, 0, task->c, APLOGNO(03204)
+                      "h2_task(%s): write without response for %s %s %s",
+                      task->id, 
+                      task->request->method, 
+                      task->request->authority, 
+                      task->request->path);
+        task->c->aborted = 1;
+        return APR_ECONNABORTED;
+    }
+    
+    if (h2_task_logio_add_bytes_out) {
+        /* count headers as if we'd do a HTTP/1.1 serialization */
+        task->output.written = h2_util_table_bytes(response->headers, 3)+1;
+        h2_task_logio_add_bytes_out(task->c, task->output.written);
+    }
+    ap_log_cerror(APLOG_MARK, APLOG_DEBUG, 0, task->c, APLOGNO(03348)
+                  "h2_task(%s): open response to %s %s %s",
+                  task->id, task->request->method, 
+                  task->request->authority, 
+                  task->request->path);
+    return h2_mplx_out_open(task->mplx, task->stream_id, 
+                            response, task->output.beam);
+}
+
+static apr_status_t send_out(h2_task *task, apr_bucket_brigade* bb)
+{
+    apr_off_t written, left;
+    apr_status_t status;
+
+    apr_brigade_length(bb, 0, &written);
+    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, task->c,
+                  "h2_task(%s): write response body (%ld bytes)", 
+                  task->id, (long)written);
+    
+    status = h2_beam_send(task->output.beam, bb, 
+                          task->blocking? APR_BLOCK_READ
+                          : APR_NONBLOCK_READ);
+    if (APR_STATUS_IS_EAGAIN(status)) {
+        apr_brigade_length(bb, 0, &left);
+        written -= left;
+        status = APR_SUCCESS;
+    }
+    if (status == APR_SUCCESS) {
+        task->output.written += written;
+        if (h2_task_logio_add_bytes_out) {
+            h2_task_logio_add_bytes_out(task->c, written);
+        }
+    }
+    return status;
+}
+
+/* Bring the data from the brigade (which represents the result of the
+ * request_rec out filter chain) into the h2_mplx for further sending
+ * on the master connection. 
+ */
+static apr_status_t output_write(h2_task *task, ap_filter_t* f, 
+                                 apr_bucket_brigade* bb)
+{
+    apr_bucket *b;
+    apr_status_t status = APR_SUCCESS;
+    
+    if (APR_BRIGADE_EMPTY(bb)) {
+        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, task->c,
+                      "h2_task(%s): empty write", task->id);
+        return APR_SUCCESS;
+    }
+    
+    if (task->frozen) {
+        h2_util_bb_log(task->c, task->stream_id, APLOG_TRACE2,
+                       "frozen task output write, ignored", bb);
+        while (!APR_BRIGADE_EMPTY(bb)) {
+            b = APR_BRIGADE_FIRST(bb);
+            if (AP_BUCKET_IS_EOR(b)) {
+                APR_BUCKET_REMOVE(b);
+                task->eor = b;
+            }
+            else {
+                apr_bucket_delete(b);
+            }
+        }
+        return APR_SUCCESS;
+    }
+    
+    if (!task->output.beam) {
+        h2_beam_create(&task->output.beam, task->pool, 
+                       task->stream_id, "output", 0); 
+    }
+    
+    /* Attempt to write saved brigade first */
+    if (task->output.bb && !APR_BRIGADE_EMPTY(task->output.bb)) {
+        status = send_out(task, task->output.bb); 
+        if (status != APR_SUCCESS) {
+            return status;
+        }
+    }
+    
+    /* If there is nothing saved (anymore), try to write the brigade passed */
+    if ((!task->output.bb || APR_BRIGADE_EMPTY(task->output.bb)) && !APR_BRIGADE_EMPTY(bb)) {
+        status = send_out(task, bb); 
+        if (status != APR_SUCCESS) {
+            return status;
+        }
+    }
+    
+    /* If the passed brigade is not empty, save it before return */
+    if (!APR_BRIGADE_EMPTY(bb)) {
+        ap_log_cerror(APLOG_MARK, APLOG_DEBUG, status, task->c,
+                      "h2_task(%s): could not write all, saving brigade", 
+                      task->id);
+        if (!task->output.bb) {
+            task->output.bb = apr_brigade_create(task->pool, 
+                                          task->c->bucket_alloc);
+        }
+        return ap_save_brigade(f, &task->output.bb, &bb, task->pool);
+    }
+    
+    if (!task->output.response_open) {
+        /* data is in the output beam, if we have not opened the response,
+         * do so now. */
+        status = open_response(task);
+        task->output.response_open = 1;
+    }
+    
+    return status;
+}
+
+/*******************************************************************************
+ * task slave connection filters
+ ******************************************************************************/
 
 static apr_status_t h2_filter_stream_input(ap_filter_t* filter,
                                            apr_bucket_brigade* brigade,
@@ -57,11 +390,7 @@ static apr_status_t h2_filter_stream_input(ap_filter_t* filter,
 {
     h2_task *task = filter->ctx;
     AP_DEBUG_ASSERT(task);
-    if (!task->input) {
-        return APR_ECONNABORTED;
-    }
-    return h2_task_input_read(task->input, filter, brigade,
-                              mode, block, readbytes);
+    return input_read(task, filter, brigade, mode, block, readbytes);
 }
 
 static apr_status_t h2_filter_stream_output(ap_filter_t* filter,
@@ -69,10 +398,7 @@ static apr_status_t h2_filter_stream_output(ap_filter_t* filter,
 {
     h2_task *task = filter->ctx;
     AP_DEBUG_ASSERT(task);
-    if (!task->output) {
-        return APR_ECONNABORTED;
-    }
-    return h2_task_output_write(task->output, filter, brigade);
+    return output_write(task, filter, brigade);
 }
 
 static apr_status_t h2_filter_read_response(ap_filter_t* f,
@@ -80,10 +406,10 @@ static apr_status_t h2_filter_read_response(ap_filter_t* f,
 {
     h2_task *task = f->ctx;
     AP_DEBUG_ASSERT(task);
-    if (!task->output || !task->output->from_h1) {
+    if (!task->output.from_h1) {
         return APR_ECONNABORTED;
     }
-    return h2_from_h1_read_response(task->output->from_h1, f, bb);
+    return h2_from_h1_read_response(task->output.from_h1, f, bb);
 }
 
 /*******************************************************************************
@@ -123,8 +449,8 @@ apr_status_t h2_task_init(apr_pool_t *pool, server_rec *s)
     return APR_SUCCESS;
 }
 
-h2_task *h2_task_create(long session_id, const h2_request *req, 
-                        conn_rec *c, h2_mplx *mplx)
+h2_task *h2_task_create(conn_rec *c, const h2_request *req, 
+                        h2_bucket_beam *input, h2_mplx *mplx)
 {
     apr_pool_t *pool;
     h2_task *task;
@@ -134,21 +460,23 @@ h2_task *h2_task_create(long session_id, const h2_request *req,
     if (task == NULL) {
         ap_log_cerror(APLOG_MARK, APLOG_ERR, APR_ENOMEM, c,
                       APLOGNO(02941) "h2_task(%ld-%d): create stream task", 
-                      session_id, req->id);
+                      c->id, req->id);
         h2_mplx_out_close(mplx, req->id);
         return NULL;
     }
     
-    task->id          = apr_psprintf(pool, "%ld-%d", session_id, req->id);
+    task->id          = apr_psprintf(pool, "%ld-%d", c->id, req->id);
     task->stream_id   = req->id;
     task->c           = c;
     task->mplx        = mplx;
     task->c->keepalives = mplx->c->keepalives;
     task->pool        = pool;
     task->request     = req;
-    task->input_eos   = !req->body;
     task->ser_headers = req->serialize;
     task->blocking    = 1;
+    task->input.beam  = input;
+    
+    apr_thread_cond_create(&task->cond, pool);
 
     h2_ctx_create_for(c, task);
     /* Add our own, network level in- and output filters. */
@@ -162,6 +490,9 @@ void h2_task_destroy(h2_task *task)
 {
     ap_remove_input_filter_byhandle(task->c->input_filters, "H2_TO_H1");
     ap_remove_output_filter_byhandle(task->c->output_filters, "H1_TO_H2");
+    if (task->eor) {
+        apr_bucket_destroy(task->eor);
+    }
     if (task->pool) {
         apr_pool_destroy(task->pool);
     }
@@ -172,14 +503,39 @@ void h2_task_set_io_blocking(h2_task *task, int blocking)
     task->blocking = blocking;
 }
 
-apr_status_t h2_task_do(h2_task *task, apr_thread_cond_t *cond)
+apr_status_t h2_task_do(h2_task *task)
 {
     apr_status_t status;
     
     AP_DEBUG_ASSERT(task);
-    task->io = cond;
-    task->input = h2_task_input_create(task, task->c);
-    task->output = h2_task_output_create(task, task->c);
+    
+    task->input.block = APR_BLOCK_READ;
+    task->input.chunked = task->request->chunked;
+    task->input.eos = !task->request->body;
+    if (task->input.eos && !task->input.chunked && !task->ser_headers) {
+        /* We do not serialize/chunk and have eos already, no need to
+         * create a bucket brigade. */
+        task->input.bb = NULL;
+        task->input.eos_written = 1;
+    }
+    else {
+        task->input.bb = apr_brigade_create(task->pool, task->c->bucket_alloc);
+        if (task->ser_headers) {
+            ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, task->c,
+                          "h2_task(%s): serialize request %s %s", 
+                          task->id, task->request->method, task->request->path);
+            apr_brigade_printf(task->input.bb, NULL, 
+                               NULL, "%s %s HTTP/1.1\r\n", 
+                               task->request->method, task->request->path);
+            apr_table_do(input_ser_header, task, task->request->headers, NULL);
+            apr_brigade_puts(task->input.bb, NULL, NULL, "\r\n");
+        }
+        if (task->input.eos) {
+            input_append_eos(task, NULL);
+        }
+    }
+    
+    task->output.from_h1 = h2_from_h1_create(task->stream_id, task->pool);
     
     ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, task->c,
                   "h2_task(%s): process connection", task->id);
@@ -195,6 +551,7 @@ apr_status_t h2_task_do(h2_task *task, apr_thread_cond_t *cond)
     else {
         ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, task->c,
                       "h2_task(%s): processing done", task->id);
+        h2_mplx_out_close(task->mplx, task->stream_id);
         status = APR_SUCCESS;
     }
     
index 15a1d3cb2c0ec4ded6d7459dc4336a7b9038ca2a..450763ee40cf62742ecd0c2eba80252e9df808ca 100644 (file)
@@ -38,6 +38,7 @@
  */
 
 struct apr_thread_cond_t;
+struct h2_bucket_beam;
 struct h2_conn;
 struct h2_mplx;
 struct h2_task;
@@ -55,29 +56,42 @@ struct h2_task {
     struct h2_mplx *mplx;
     apr_pool_t *pool;
     const struct h2_request *request;
+    apr_bucket *eor;
+    struct apr_thread_cond_t *cond;
     
     unsigned int filters_set : 1;
-    unsigned int input_eos   : 1;
     unsigned int ser_headers : 1;
     unsigned int frozen      : 1;
     unsigned int blocking    : 1;
     unsigned int detached    : 1;
     
-    struct h2_task_input *input;
-    struct h2_task_output *output;
-    struct apr_thread_cond_t *io;   /* used to wait for events on */
+    struct {
+        struct h2_bucket_beam *beam;
+        apr_bucket_brigade *bb;
+        apr_read_type_e block;
+        unsigned int chunked : 1;
+        unsigned int eos : 1;
+        unsigned int eos_written : 1;
+    } input;
+    struct {
+        struct h2_bucket_beam *beam;
+        struct h2_from_h1 *from_h1;
+        unsigned int response_open : 1;
+        apr_off_t written;
+        apr_bucket_brigade *bb;
+    } output;
     
     struct h2_req_engine *engine;   /* engine hosted by this task */
     struct h2_req_engine *assigned; /* engine that task has been assigned to */
     request_rec *r;                 /* request being processed in this task */
 };
 
-h2_task *h2_task_create(long session_id, const struct h2_request *req, 
-                        conn_rec *c, struct h2_mplx *mplx);
+h2_task *h2_task_create(conn_rec *c, const struct h2_request *req, 
+                        struct h2_bucket_beam *input, struct h2_mplx *mplx);
 
 void h2_task_destroy(h2_task *task);
 
-apr_status_t h2_task_do(h2_task *task, struct apr_thread_cond_t *cond);
+apr_status_t h2_task_do(h2_task *task);
 
 void h2_task_register_hooks(void);
 /*
diff --git a/modules/http2/h2_task_input.c b/modules/http2/h2_task_input.c
deleted file mode 100644 (file)
index 3993b6b..0000000
+++ /dev/null
@@ -1,228 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include <assert.h>
-
-#include <httpd.h>
-#include <http_core.h>
-#include <http_log.h>
-#include <http_connection.h>
-
-#include "h2_private.h"
-#include "h2_conn.h"
-#include "h2_mplx.h"
-#include "h2_request.h"
-#include "h2_session.h"
-#include "h2_stream.h"
-#include "h2_task_input.h"
-#include "h2_task.h"
-#include "h2_util.h"
-
-
-static int is_aborted(ap_filter_t *f)
-{
-    return (f->c->aborted);
-}
-
-static int ser_header(void *ctx, const char *name, const char *value) 
-{
-    h2_task_input *input = (h2_task_input*)ctx;
-    apr_brigade_printf(input->bb, NULL, NULL, "%s: %s\r\n", name, value);
-    return 1;
-}
-
-h2_task_input *h2_task_input_create(h2_task *task, conn_rec *c)
-{
-    h2_task_input *input = apr_pcalloc(task->pool, sizeof(h2_task_input));
-    if (input) {
-        input->task = task;
-        input->bb = NULL;
-        input->block = APR_BLOCK_READ;
-        
-        if (task->ser_headers) {
-            ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, c,
-                          "h2_task_input(%s): serialize request %s %s", 
-                          task->id, task->request->method, task->request->path);
-            input->bb = apr_brigade_create(task->pool, c->bucket_alloc);
-            apr_brigade_printf(input->bb, NULL, NULL, "%s %s HTTP/1.1\r\n", 
-                               task->request->method, task->request->path);
-            apr_table_do(ser_header, input, task->request->headers, NULL);
-            apr_brigade_puts(input->bb, NULL, NULL, "\r\n");
-            if (input->task->input_eos) {
-                APR_BRIGADE_INSERT_TAIL(input->bb, apr_bucket_eos_create(c->bucket_alloc));
-            }
-        }
-        else if (!input->task->input_eos) {
-            input->bb = apr_brigade_create(task->pool, c->bucket_alloc);
-        }
-        else {
-            /* We do not serialize and have eos already, no need to
-             * create a bucket brigade. */
-        }
-    }
-    return input;
-}
-
-void h2_task_input_block_set(h2_task_input *input, apr_read_type_e block)
-{
-    input->block = block;
-}
-
-apr_status_t h2_task_input_read(h2_task_input *input,
-                                ap_filter_t* f,
-                                apr_bucket_brigade* bb,
-                                ap_input_mode_t mode,
-                                apr_read_type_e block,
-                                apr_off_t readbytes)
-{
-    apr_status_t status = APR_SUCCESS;
-    apr_off_t bblen = 0;
-    
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
-                  "h2_task_input(%s): read, block=%d, mode=%d, readbytes=%ld", 
-                  input->task->id, block, mode, (long)readbytes);
-    
-    if (mode == AP_MODE_INIT) {
-        return ap_get_brigade(f->c->input_filters, bb, mode, block, readbytes);
-    }
-    
-    if (is_aborted(f)) {
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
-                      "h2_task_input(%s): is aborted", input->task->id);
-        return APR_ECONNABORTED;
-    }
-    
-    if (input->bb) {
-        status = apr_brigade_length(input->bb, 1, &bblen);
-        if (status != APR_SUCCESS) {
-            ap_log_cerror(APLOG_MARK, APLOG_WARNING, status, f->c,
-                          APLOGNO(02958) "h2_task_input(%s): brigade length fail", 
-                          input->task->id);
-            return status;
-        }
-    }
-    
-    if ((bblen == 0) && input->task->input_eos) {
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, f->c,
-                      "h2_task_input(%s): eos", input->task->id);
-        return APR_EOF;
-    }
-    
-    while (bblen == 0) {
-        /* Get more data for our stream from mplx.
-         */
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                      "h2_task_input(%s): get more data from mplx, block=%d, "
-                      "readbytes=%ld, queued=%ld",
-                      input->task->id, block, 
-                      (long)readbytes, (long)bblen);
-        
-        /* Override the block mode we get called with depending on the input's
-         * setting. 
-         */
-        status = h2_mplx_in_read(input->task->mplx, block,
-                                 input->task->stream_id, input->bb, 
-                                 f->r? f->r->trailers_in : NULL, 
-                                 input->task->io);
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                      "h2_task_input(%s): mplx in read returned",
-                      input->task->id);
-        if (APR_STATUS_IS_EAGAIN(status) 
-            && (mode == AP_MODE_GETLINE || block == APR_BLOCK_READ)) {
-            /* chunked input handling does not seem to like it if we
-             * return with APR_EAGAIN from a GETLINE read... 
-             * upload 100k test on test-ser.example.org hangs */
-            status = APR_SUCCESS;
-        }
-        else if (status != APR_SUCCESS) {
-            return status;
-        }
-        
-        status = apr_brigade_length(input->bb, 1, &bblen);
-        if (status != APR_SUCCESS) {
-            return status;
-        }
-        
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                      "h2_task_input(%s): mplx in read, %ld bytes in brigade",
-                      input->task->id, (long)bblen);
-        if (h2_task_logio_add_bytes_in) {
-            h2_task_logio_add_bytes_in(f->c, bblen);
-        }
-    }
-    
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                  "h2_task_input(%s): read, mode=%d, block=%d, "
-                  "readbytes=%ld, queued=%ld",
-                  input->task->id, mode, block, 
-                  (long)readbytes, (long)bblen);
-           
-    if (!APR_BRIGADE_EMPTY(input->bb)) {
-        if (mode == AP_MODE_EXHAUSTIVE) {
-            /* return all we have */
-            status = h2_util_move(bb, input->bb, readbytes, NULL, 
-                                  "task_input_read(exhaustive)");
-        }
-        else if (mode == AP_MODE_READBYTES) {
-            status = h2_util_move(bb, input->bb, readbytes, NULL, 
-                                  "task_input_read(readbytes)");
-        }
-        else if (mode == AP_MODE_SPECULATIVE) {
-            /* return not more than was asked for */
-            status = h2_util_copy(bb, input->bb, readbytes,  
-                                  "task_input_read(speculative)");
-        }
-        else if (mode == AP_MODE_GETLINE) {
-            /* we are reading a single LF line, e.g. the HTTP headers */
-            status = apr_brigade_split_line(bb, input->bb, block, 
-                                            HUGE_STRING_LEN);
-            if (APLOGctrace1(f->c)) {
-                char buffer[1024];
-                apr_size_t len = sizeof(buffer)-1;
-                apr_brigade_flatten(bb, buffer, &len);
-                buffer[len] = 0;
-                ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                              "h2_task_input(%s): getline: %s",
-                              input->task->id, buffer);
-            }
-        }
-        else {
-            /* Hmm, well. There is mode AP_MODE_EATCRLF, but we chose not
-             * to support it. Seems to work. */
-            ap_log_cerror(APLOG_MARK, APLOG_ERR, APR_ENOTIMPL, f->c,
-                          APLOGNO(02942) 
-                          "h2_task_input, unsupported READ mode %d", mode);
-            status = APR_ENOTIMPL;
-        }
-        
-        if (APLOGctrace1(f->c)) {
-            apr_brigade_length(bb, 0, &bblen);
-            ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                          "h2_task_input(%s): return %ld data bytes",
-                          input->task->id, (long)bblen);
-        }
-        return status;
-    }
-    
-    if (is_aborted(f)) {
-        return APR_ECONNABORTED;
-    }
-    
-    status = (block == APR_NONBLOCK_READ)? APR_EAGAIN : APR_EOF;
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, status, f->c,
-                  "h2_task_input(%s): no data", input->task->id);
-    return status;
-}
-
diff --git a/modules/http2/h2_task_input.h b/modules/http2/h2_task_input.h
deleted file mode 100644 (file)
index c8913ca..0000000
+++ /dev/null
@@ -1,46 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef __mod_h2__h2_task_input__
-#define __mod_h2__h2_task_input__
-
-/* h2_task_input places the HEADER+DATA, formatted in HTTP/1.1, into
- * a bucket brigade. The brigade is setup as the input brigade for our
- * pseudo httpd conn_rec that is handling a specific h2_task.
- */
-struct apr_thread_cond_t;
-struct h2_mplx;
-struct h2_task;
-
-typedef struct h2_task_input h2_task_input;
-struct h2_task_input {
-    struct h2_task *task;
-    apr_bucket_brigade *bb;
-    apr_read_type_e block;
-};
-
-
-h2_task_input *h2_task_input_create(struct h2_task *task, conn_rec *c);
-
-apr_status_t h2_task_input_read(h2_task_input *input,
-                                  ap_filter_t* filter,
-                                  apr_bucket_brigade* brigade,
-                                  ap_input_mode_t mode,
-                                  apr_read_type_e block,
-                                  apr_off_t readbytes);
-
-void h2_task_input_block_set(h2_task_input *input, apr_read_type_e block);
-
-#endif /* defined(__mod_h2__h2_task_input__) */
diff --git a/modules/http2/h2_task_output.c b/modules/http2/h2_task_output.c
deleted file mode 100644 (file)
index 80938d1..0000000
+++ /dev/null
@@ -1,176 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include <assert.h>
-
-#include <apr_thread_cond.h>
-#include <httpd.h>
-#include <http_core.h>
-#include <http_log.h>
-#include <http_connection.h>
-#include <http_request.h>
-
-#include "h2_private.h"
-#include "h2_conn.h"
-#include "h2_mplx.h"
-#include "h2_request.h"
-#include "h2_session.h"
-#include "h2_stream.h"
-#include "h2_from_h1.h"
-#include "h2_response.h"
-#include "h2_task_output.h"
-#include "h2_task.h"
-#include "h2_util.h"
-
-
-h2_task_output *h2_task_output_create(h2_task *task, conn_rec *c)
-{
-    h2_task_output *output = apr_pcalloc(task->pool, sizeof(h2_task_output));
-    if (output) {
-        output->task = task;
-        output->from_h1 = h2_from_h1_create(task->stream_id, task->pool);
-    }
-    return output;
-}
-
-static apr_status_t open_response(h2_task_output *output, ap_filter_t *f,
-                                  apr_bucket_brigade *bb, const char *caller)
-{
-    h2_response *response;
-    response = h2_from_h1_get_response(output->from_h1);
-    if (!response) {
-        if (f) {
-            /* This happens currently when ap_die(status, r) is invoked
-             * by a read request filter. */
-            ap_log_cerror(APLOG_MARK, APLOG_INFO, 0, output->task->c, APLOGNO(03204)
-                          "h2_task_output(%s): write without response by %s "
-                          "for %s %s %s",
-                          output->task->id, caller, 
-                          output->task->request->method, 
-                          output->task->request->authority, 
-                          output->task->request->path);
-            output->task->c->aborted = 1;
-        }
-        if (output->task->io) {
-            apr_thread_cond_broadcast(output->task->io);
-        }
-        return APR_ECONNABORTED;
-    }
-    
-    if (h2_task_logio_add_bytes_out) {
-        /* count headers as if we'd do a HTTP/1.1 serialization */
-        output->written = h2_util_table_bytes(response->headers, 3)+1;
-        h2_task_logio_add_bytes_out(output->task->c, output->written);
-    }
-    ap_log_cerror(APLOG_MARK, APLOG_DEBUG, 0, output->task->c, APLOGNO(03348)
-                  "h2_task(%s): open response to %s %s %s",
-                  output->task->id, output->task->request->method, 
-                  output->task->request->authority, 
-                  output->task->request->path);
-    return h2_mplx_out_open(output->task->mplx, output->task->stream_id, 
-                            response, f, bb, output->task->io);
-}
-
-static apr_status_t write_brigade_raw(h2_task_output *output, 
-                                      ap_filter_t* f, apr_bucket_brigade* bb)
-{
-    apr_off_t written, left;
-    apr_status_t status;
-
-    apr_brigade_length(bb, 0, &written);
-    ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, output->task->c,
-                  "h2_task(%s): write response body (%ld bytes)", 
-                  output->task->id, (long)written);
-    
-    status = h2_mplx_out_write(output->task->mplx, output->task->stream_id, 
-                               f, output->task->blocking, bb, output->task->io);
-    if (status == APR_INCOMPLETE) {
-        apr_brigade_length(bb, 0, &left);
-        written -= left;
-        status = APR_SUCCESS;
-    }
-
-    if (status == APR_SUCCESS) {
-        output->written += written;
-        if (h2_task_logio_add_bytes_out) {
-            h2_task_logio_add_bytes_out(output->task->c, written);
-        }
-    }
-    return status;
-}
-
-/* Bring the data from the brigade (which represents the result of the
- * request_rec out filter chain) into the h2_mplx for further sending
- * on the master connection. 
- */
-apr_status_t h2_task_output_write(h2_task_output *output,
-                                  ap_filter_t* f, apr_bucket_brigade* bb)
-{
-    apr_bucket *b;
-    apr_status_t status = APR_SUCCESS;
-    
-    if (APR_BRIGADE_EMPTY(bb)) {
-        ap_log_cerror(APLOG_MARK, APLOG_TRACE1, 0, output->task->c,
-                      "h2_task(%s): empty write", output->task->id);
-        return APR_SUCCESS;
-    }
-    
-    if (output->task->frozen) {
-        h2_util_bb_log(output->task->c, output->task->stream_id, APLOG_TRACE2,
-                       "frozen task output write, ignored", bb);
-        while (!APR_BRIGADE_EMPTY(bb)) {
-            b = APR_BRIGADE_FIRST(bb);
-            if (AP_BUCKET_IS_EOR(b)) {
-                /* TODO: keep it */
-                APR_BUCKET_REMOVE(b);
-            }
-            else {
-                apr_bucket_delete(b);
-            }
-        }
-        return APR_SUCCESS;
-    }
-    
-    if (!output->response_open) {
-        status = open_response(output, f, bb, "write");
-        output->response_open = 1;
-    }
-    
-    /* Attempt to write saved brigade first */
-    if (status == APR_SUCCESS && output->bb && !APR_BRIGADE_EMPTY(output->bb)) {
-        status = write_brigade_raw(output, f, output->bb);
-    }
-    
-    /* If there is nothing saved (anymore), try to write the brigade passed */
-    if (status == APR_SUCCESS
-        && (!output->bb || APR_BRIGADE_EMPTY(output->bb))
-        && !APR_BRIGADE_EMPTY(bb)) {
-        status = write_brigade_raw(output, f, bb);
-    }
-    
-    /* If the passed brigade is not empty, save it before return */
-    if (status == APR_SUCCESS && !APR_BRIGADE_EMPTY(bb)) {
-        ap_log_cerror(APLOG_MARK, APLOG_DEBUG, status, output->task->c,
-                      "h2_task(%s): could not write all, saving brigade", 
-                      output->task->id);
-        if (!output->bb) {
-            output->bb = apr_brigade_create(output->task->pool, output->task->c->bucket_alloc);
-        }
-        return ap_save_brigade(f, &output->bb, &bb, output->task->pool);
-    }
-    
-    return status;
-}
-
diff --git a/modules/http2/h2_task_output.h b/modules/http2/h2_task_output.h
deleted file mode 100644 (file)
index 3135bc4..0000000
+++ /dev/null
@@ -1,50 +0,0 @@
-/* Copyright 2015 greenbytes GmbH (https://www.greenbytes.de)
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef __mod_h2__h2_task_output__
-#define __mod_h2__h2_task_output__
-
-/* h2_task_output reads a HTTP/1 response from the brigade and applies
- * them to a h2_output_converter. The brigade is setup as the output brigade
- * for our pseudo httpd conn_rec that is handling a specific h2_task.
- * 
- */
-struct apr_thread_cond_t;
-struct h2_mplx;
-struct h2_task;
-struct h2_from_h1;
-
-typedef struct h2_task_output h2_task_output;
-
-struct h2_task_output {
-    struct h2_task *task;
-    struct h2_from_h1 *from_h1;
-    
-    unsigned int response_open : 1;
-
-    apr_off_t written;
-    apr_bucket_brigade *bb;
-};
-
-h2_task_output *h2_task_output_create(struct h2_task *task, conn_rec *c);
-
-apr_status_t h2_task_output_write(h2_task_output *output,
-                                  ap_filter_t* filter,
-                                  apr_bucket_brigade* brigade);
-
-apr_status_t h2_task_output_freeze(h2_task_output *output);
-apr_status_t h2_task_output_thaw(h2_task_output *output);
-
-#endif /* defined(__mod_h2__h2_task_output__) */
index 06472425f229e9e27e814eb20f4391304d9aba2d..9beaeae0480336102f970ec540ce12cc1fbd19de 100644 (file)
@@ -331,6 +331,301 @@ void h2_ihash_clear(h2_ihash_t *ih)
     apr_hash_clear(ih->hash);
 }
 
+/*******************************************************************************
+ * ilist - sorted list for structs with int identifier
+ ******************************************************************************/
+
+#define h2_ilist_IDX(list, i) ((int**)(list)->elts)[i]
+
+struct h2_ilist_t {
+    apr_array_header_t *l;
+};
+
+h2_ilist_t *h2_ilist_create(apr_pool_t *pool)
+{
+    h2_ilist_t *list = apr_pcalloc(pool, sizeof(h2_ilist_t));
+    if (list) {
+        list->l = apr_array_make(pool, 100, sizeof(int*));
+        if (!list->l) {
+            return NULL;
+        }
+    }
+    return list;
+}
+
+static int h2_ilist_cmp(const void *s1, const void *s2)
+{
+    int **pi1 = (int **)s1;
+    int **pi2 = (int **)s2;
+    return *(*pi1) - *(*pi2);
+}
+
+void *h2_ilist_get(h2_ilist_t *list, int id)
+{
+    /* we keep the array sorted by id, so lookup can be done
+     * by bsearch.
+     */
+    int **pi;
+    int *pkey = &id;
+
+    pi = bsearch(&pkey, list->l->elts, list->l->nelts, 
+                 list->l->elt_size, h2_ilist_cmp);
+    return pi? *pi : NULL;
+}
+
+static void h2_ilist_sort(h2_ilist_t *list)
+{
+    qsort(list->l->elts, list->l->nelts, list->l->elt_size, h2_ilist_cmp);
+}
+
+apr_status_t h2_ilist_add(h2_ilist_t *list, void *val)
+{
+    int *pi = val;
+    void *existing = h2_ilist_get(list, *pi);
+    if (!existing) {
+        int last;
+        APR_ARRAY_PUSH(list->l, void*) = val;
+        /* Often, values get added in ascending order of id. We
+         * keep the array sorted, so we just need to check if the newly
+         * appended stream has a lower id than the last one. if not,
+         * sorting is not necessary.
+         */
+        last = list->l->nelts - 1;
+        if (last > 0 
+            && *h2_ilist_IDX(list->l, last) < *h2_ilist_IDX(list->l, last-1)) {
+            h2_ilist_sort(list);
+        }
+    }
+    return APR_SUCCESS;
+}
+
+static void remove_idx(h2_ilist_t *list, int idx)
+{
+    int n;
+    --list->l->nelts;
+    n = list->l->nelts - idx;
+    if (n > 0) {
+        /* There are n h2_io* behind idx. Move the rest down */
+        int **selts = (int**)list->l->elts;
+        memmove(selts + idx, selts + idx + 1, n * sizeof(int*));
+    }
+}
+
+void *h2_ilist_remove(h2_ilist_t *list, int id)
+{
+    int i;
+    for (i = 0; i < list->l->nelts; ++i) {
+        int *e = h2_ilist_IDX(list->l, i);
+        if (id == *e) {
+            remove_idx(list, i);
+            return e;
+        }
+    }
+    return NULL;
+}
+
+void *h2_ilist_shift(h2_ilist_t *list)
+{
+    if (list->l->nelts > 0) {
+        int *pi = h2_ilist_IDX(list->l, 0);
+        remove_idx(list, 0);
+        return pi;
+    }
+    return NULL;
+}
+
+int h2_ilist_empty(h2_ilist_t *list)
+{
+    return list->l->nelts == 0;
+}
+
+int h2_ilist_iter(h2_ilist_t *list, h2_ilist_iter_t *iter, void *ctx)
+{
+    int i;
+    for (i = 0; i < list->l->nelts; ++i) {
+        int *pi = h2_ilist_IDX(list->l, i);
+        if (!iter(ctx, pi)) {
+            return 0;
+        }
+    }
+    return 1;
+}
+
+apr_size_t h2_ilist_count(h2_ilist_t *list)
+{
+    return list->l->nelts;
+}
+
+/*******************************************************************************
+ * iqueue - sorted list of int
+ ******************************************************************************/
+
+static void iq_grow(h2_iqueue *q, int nlen);
+static void iq_swap(h2_iqueue *q, int i, int j);
+static int iq_bubble_up(h2_iqueue *q, int i, int top, 
+                        h2_iq_cmp *cmp, void *ctx);
+static int iq_bubble_down(h2_iqueue *q, int i, int bottom, 
+                          h2_iq_cmp *cmp, void *ctx);
+
+h2_iqueue *h2_iq_create(apr_pool_t *pool, int capacity)
+{
+    h2_iqueue *q = apr_pcalloc(pool, sizeof(h2_iqueue));
+    if (q) {
+        q->pool = pool;
+        iq_grow(q, capacity);
+        q->nelts = 0;
+    }
+    return q;
+}
+
+int h2_iq_empty(h2_iqueue *q)
+{
+    return q->nelts == 0;
+}
+
+int h2_iq_count(h2_iqueue *q)
+{
+    return q->nelts;
+}
+
+
+void h2_iq_add(h2_iqueue *q, int sid, h2_iq_cmp *cmp, void *ctx)
+{
+    int i;
+    
+    if (q->nelts >= q->nalloc) {
+        iq_grow(q, q->nalloc * 2);
+    }
+    
+    i = (q->head + q->nelts) % q->nalloc;
+    q->elts[i] = sid;
+    ++q->nelts;
+    
+    if (cmp) {
+        /* bubble it to the front of the queue */
+        iq_bubble_up(q, i, q->head, cmp, ctx);
+    }
+}
+
+int h2_iq_remove(h2_iqueue *q, int sid)
+{
+    int i;
+    for (i = 0; i < q->nelts; ++i) {
+        if (sid == q->elts[(q->head + i) % q->nalloc]) {
+            break;
+        }
+    }
+    
+    if (i < q->nelts) {
+        ++i;
+        for (; i < q->nelts; ++i) {
+            q->elts[(q->head+i-1)%q->nalloc] = q->elts[(q->head+i)%q->nalloc];
+        }
+        --q->nelts;
+        return 1;
+    }
+    return 0;
+}
+
+void h2_iq_clear(h2_iqueue *q)
+{
+    q->nelts = 0;
+}
+
+void h2_iq_sort(h2_iqueue *q, h2_iq_cmp *cmp, void *ctx)
+{
+    /* Assume that changes in ordering are minimal. This needs,
+     * best case, q->nelts - 1 comparisions to check that nothing
+     * changed.
+     */
+    if (q->nelts > 0) {
+        int i, ni, prev, last;
+        
+        /* Start at the end of the queue and create a tail of sorted
+         * entries. Make that tail one element longer in each iteration.
+         */
+        last = i = (q->head + q->nelts - 1) % q->nalloc;
+        while (i != q->head) {
+            prev = (q->nalloc + i - 1) % q->nalloc;
+            
+            ni = iq_bubble_up(q, i, prev, cmp, ctx);
+            if (ni == prev) {
+                /* i bubbled one up, bubble the new i down, which
+                 * keeps all tasks below i sorted. */
+                iq_bubble_down(q, i, last, cmp, ctx);
+            }
+            i = prev;
+        };
+    }
+}
+
+
+int h2_iq_shift(h2_iqueue *q)
+{
+    int sid;
+    
+    if (q->nelts <= 0) {
+        return 0;
+    }
+    
+    sid = q->elts[q->head];
+    q->head = (q->head + 1) % q->nalloc;
+    q->nelts--;
+    
+    return sid;
+}
+
+static void iq_grow(h2_iqueue *q, int nlen)
+{
+    if (nlen > q->nalloc) {
+        int *nq = apr_pcalloc(q->pool, sizeof(int) * nlen);
+        if (q->nelts > 0) {
+            int l = ((q->head + q->nelts) % q->nalloc) - q->head;
+            
+            memmove(nq, q->elts + q->head, sizeof(int) * l);
+            if (l < q->nelts) {
+                /* elts wrapped, append elts in [0, remain] to nq */
+                int remain = q->nelts - l;
+                memmove(nq + l, q->elts, sizeof(int) * remain);
+            }
+        }
+        q->elts = nq;
+        q->nalloc = nlen;
+        q->head = 0;
+    }
+}
+
+static void iq_swap(h2_iqueue *q, int i, int j)
+{
+    int x = q->elts[i];
+    q->elts[i] = q->elts[j];
+    q->elts[j] = x;
+}
+
+static int iq_bubble_up(h2_iqueue *q, int i, int top, 
+                        h2_iq_cmp *cmp, void *ctx) 
+{
+    int prev;
+    while (((prev = (q->nalloc + i - 1) % q->nalloc), i != top) 
+           && (*cmp)(q->elts[i], q->elts[prev], ctx) < 0) {
+        iq_swap(q, prev, i);
+        i = prev;
+    }
+    return i;
+}
+
+static int iq_bubble_down(h2_iqueue *q, int i, int bottom, 
+                          h2_iq_cmp *cmp, void *ctx)
+{
+    int next;
+    while (((next = (q->nalloc + i + 1) % q->nalloc), i != bottom) 
+           && (*cmp)(q->elts[i], q->elts[next], ctx) > 0) {
+        iq_swap(q, next, i);
+        i = next;
+    }
+    return i;
+}
+
 /*******************************************************************************
  * h2_util for apt_table_t
  ******************************************************************************/
@@ -368,15 +663,6 @@ apr_size_t h2_util_table_bytes(apr_table_t *t, apr_size_t pair_extra)
  * h2_util for bucket brigades
  ******************************************************************************/
 
-/* DEEP_COPY==0 crashes under load. I think the setaside is fine, 
- * however buckets moved to another thread will still be
- * free'd against the old bucket_alloc. *And* if the old
- * pool gets destroyed too early, the bucket disappears while
- * still needed.
- */
-static const int DEEP_COPY = 1;
-static const int FILE_MOVE = 1;
-
 static apr_status_t last_not_included(apr_bucket_brigade *bb, 
                                       apr_off_t maxlen, 
                                       int same_alloc,
@@ -434,200 +720,95 @@ static apr_status_t last_not_included(apr_bucket_brigade *bb,
     return status;
 }
 
-#define LOG_BUCKETS     0
-#define LOG_LEVEL APLOG_INFO
-
-apr_status_t h2_util_move(apr_bucket_brigade *to, apr_bucket_brigade *from, 
-                          apr_off_t maxlen, apr_size_t *pfile_buckets_allowed, 
-                          const char *msg)
+apr_status_t h2_brigade_concat_length(apr_bucket_brigade *dest, 
+                                      apr_bucket_brigade *src,
+                                      apr_off_t length)
 {
+    apr_bucket *b, *next;
+    apr_off_t remain = length;
     apr_status_t status = APR_SUCCESS;
-    int same_alloc;
     
-    AP_DEBUG_ASSERT(to);
-    AP_DEBUG_ASSERT(from);
-    same_alloc = (to->bucket_alloc == from->bucket_alloc 
-                  || to->p == from->p);
-
-    if (!FILE_MOVE) {
-        pfile_buckets_allowed = NULL;
-    }
-    
-    if (!APR_BRIGADE_EMPTY(from)) {
-        apr_bucket *b, *end;
+    for (b = APR_BRIGADE_FIRST(src); 
+         b != APR_BRIGADE_SENTINEL(src);
+         b = next) {
+        next = APR_BUCKET_NEXT(b);
         
-        status = last_not_included(from, maxlen, same_alloc,
-                                   pfile_buckets_allowed, &end);
-        if (status != APR_SUCCESS) {
-            return status;
+        if (APR_BUCKET_IS_METADATA(b)) {
+            /* fall through */
         }
-        
-        while (!APR_BRIGADE_EMPTY(from) && status == APR_SUCCESS) {
-            b = APR_BRIGADE_FIRST(from);
-            if (b == end) {
-                break;
+        else {
+            if (remain == b->length) {
+                /* fall through */
             }
-            
-            if (same_alloc || (b->list == to->bucket_alloc)) {
-                /* both brigades use the same bucket_alloc and auto-cleanups
-                 * have the same life time. It's therefore safe to just move
-                 * directly. */
-                APR_BUCKET_REMOVE(b);
-                APR_BRIGADE_INSERT_TAIL(to, b);
-#if LOG_BUCKETS
-                ap_log_perror(APLOG_MARK, LOG_LEVEL, 0, to->p, APLOGNO(03205)
-                              "h2_util_move: %s, passed bucket(same bucket_alloc) "
-                              "%ld-%ld, type=%s",
-                              msg, (long)b->start, (long)b->length, 
-                              APR_BUCKET_IS_METADATA(b)? 
-                              (APR_BUCKET_IS_EOS(b)? "EOS": 
-                               (APR_BUCKET_IS_FLUSH(b)? "FLUSH" : "META")) : 
-                              (APR_BUCKET_IS_FILE(b)? "FILE" : "DATA"));
-#endif
+            else if (remain <= 0) {
+                return status;
             }
-            else if (DEEP_COPY) {
-                /* we have not managed the magic of passing buckets from
-                 * one thread to another. Any attempts result in
-                 * cleanup of pools scrambling memory.
-                 */
-                if (APR_BUCKET_IS_METADATA(b)) {
-                    if (APR_BUCKET_IS_EOS(b)) {
-                        APR_BRIGADE_INSERT_TAIL(to, apr_bucket_eos_create(to->bucket_alloc));
-                    }
-                    else {
-                        /* ignore */
-                    }
-                }
-                else if (pfile_buckets_allowed 
-                         && *pfile_buckets_allowed > 0 
-                         && APR_BUCKET_IS_FILE(b)) {
-                    /* We do not want to read files when passing buckets, if
-                     * we can avoid it. However, what we've come up so far
-                     * is not working corrently, resulting either in crashes or
-                     * too many open file descriptors.
-                     */
-                    apr_bucket_file *f = (apr_bucket_file *)b->data;
-                    apr_file_t *fd = f->fd;
-                    int setaside = (f->readpool != to->p);
-#if LOG_BUCKETS
-                    ap_log_perror(APLOG_MARK, LOG_LEVEL, 0, to->p, APLOGNO(03206)
-                                  "h2_util_move: %s, moving FILE bucket %ld-%ld "
-                                  "from=%lx(p=%lx) to=%lx(p=%lx), setaside=%d",
-                                  msg, (long)b->start, (long)b->length, 
-                                  (long)from, (long)from->p, 
-                                  (long)to, (long)to->p, setaside);
-#endif
-                    if (setaside) {
-                        status = apr_file_setaside(&fd, fd, to->p);
-                        if (status != APR_SUCCESS) {
-                            ap_log_perror(APLOG_MARK, APLOG_ERR, status, to->p,
-                                          APLOGNO(02947) "h2_util: %s, setaside FILE", 
-                                          msg);
-                            return status;
-                        }
+            else {
+                if (b->length == ((apr_size_t)-1)) {
+                    const char *ign;
+                    apr_size_t ilen;
+                    status = apr_bucket_read(b, &ign, &ilen, APR_BLOCK_READ);
+                    if (status != APR_SUCCESS) {
+                        return status;
                     }
-                    apr_brigade_insert_file(to, fd, b->start, b->length, 
-                                            to->p);
-                    --(*pfile_buckets_allowed);
                 }
-                else {
-                    const char *data;
-                    apr_size_t len;
-
-                    status = apr_bucket_read(b, &data, &len, APR_BLOCK_READ);
-                    if (status == APR_SUCCESS && len > 0) {
-                        status = apr_brigade_write(to, NULL, NULL, data, len);
-#if LOG_BUCKETS
-                        ap_log_perror(APLOG_MARK, LOG_LEVEL, 0, to->p, APLOGNO(03207)
-                                      "h2_util_move: %s, copied bucket %ld-%ld "
-                                      "from=%lx(p=%lx) to=%lx(p=%lx)",
-                                      msg, (long)b->start, (long)b->length, 
-                                      (long)from, (long)from->p, 
-                                      (long)to, (long)to->p);
-#endif
-                    }
+            
+                if (remain < b->length) {
+                    apr_bucket_split(b, remain);
                 }
-                apr_bucket_delete(b);
-            }
-            else {
-                apr_bucket_setaside(b, to->p);
-                APR_BUCKET_REMOVE(b);
-                APR_BRIGADE_INSERT_TAIL(to, b);
-#if LOG_BUCKETS
-                ap_log_perror(APLOG_MARK, LOG_LEVEL, 0, to->p, APLOGNO(03208)
-                              "h2_util_move: %s, passed setaside bucket %ld-%ld "
-                              "from=%lx(p=%lx) to=%lx(p=%lx)",
-                              msg, (long)b->start, (long)b->length, 
-                              (long)from, (long)from->p, 
-                              (long)to, (long)to->p);
-#endif
             }
         }
+        APR_BUCKET_REMOVE(b);
+        APR_BRIGADE_INSERT_TAIL(dest, b);
+        remain -= b->length;
     }
-    
     return status;
 }
 
-apr_status_t h2_util_copy(apr_bucket_brigade *to, apr_bucket_brigade *from, 
-                          apr_off_t maxlen, const char *msg)
+apr_status_t h2_brigade_copy_length(apr_bucket_brigade *dest, 
+                                    apr_bucket_brigade *src,
+                                    apr_off_t length)
 {
+    apr_bucket *b, *next;
+    apr_off_t remain = length;
     apr_status_t status = APR_SUCCESS;
-    int same_alloc;
-
-    (void)msg;
-    AP_DEBUG_ASSERT(to);
-    AP_DEBUG_ASSERT(from);
-    same_alloc = (to->bucket_alloc == from->bucket_alloc);
-
-    if (!APR_BRIGADE_EMPTY(from)) {
-        apr_bucket *b, *end, *cpy;
+    
+    for (b = APR_BRIGADE_FIRST(src); 
+         b != APR_BRIGADE_SENTINEL(src);
+         b = next) {
+        next = APR_BUCKET_NEXT(b);
         
-        status = last_not_included(from, maxlen, 0, 0, &end);
-        if (status != APR_SUCCESS) {
-            return status;
+        if (APR_BUCKET_IS_METADATA(b)) {
+            /* fall through */
         }
-
-        for (b = APR_BRIGADE_FIRST(from);
-             b != APR_BRIGADE_SENTINEL(from) && b != end;
-             b = APR_BUCKET_NEXT(b))
-        {
-            if (same_alloc) {
-                status = apr_bucket_copy(b, &cpy);
-                if (status != APR_SUCCESS) {
-                    break;
-                }
-                APR_BRIGADE_INSERT_TAIL(to, cpy);
+        else {
+            if (remain == b->length) {
+                /* fall through */
+            }
+            else if (remain <= 0) {
+                return status;
             }
             else {
-                if (APR_BUCKET_IS_METADATA(b)) {
-                    if (APR_BUCKET_IS_EOS(b)) {
-                        APR_BRIGADE_INSERT_TAIL(to, apr_bucket_eos_create(to->bucket_alloc));
-                    }
-                    else if (APR_BUCKET_IS_FLUSH(b)) {
-                        APR_BRIGADE_INSERT_TAIL(to, apr_bucket_flush_create(to->bucket_alloc));
-                    }
-                    else {
-                        /* ignore */
+                if (b->length == ((apr_size_t)-1)) {
+                    const char *ign;
+                    apr_size_t ilen;
+                    status = apr_bucket_read(b, &ign, &ilen, APR_BLOCK_READ);
+                    if (status != APR_SUCCESS) {
+                        return status;
                     }
                 }
-                else {
-                    const char *data;
-                    apr_size_t len;
-                    status = apr_bucket_read(b, &data, &len, APR_BLOCK_READ);
-                    if (status == APR_SUCCESS && len > 0) {
-                        status = apr_brigade_write(to, NULL, NULL, data, len);
-#if LOG_BUCKETS                        
-                        ap_log_perror(APLOG_MARK, LOG_LEVEL, 0, to->p, APLOGNO(03209)
-                                      "h2_util_copy: %s, copied bucket %ld-%ld "
-                                      "from=%lx(p=%lx) to=%lx(p=%lx)",
-                                      msg, (long)b->start, (long)b->length, 
-                                      (long)from, (long)from->p, 
-                                      (long)to, (long)to->p);
-#endif
-                    }
+            
+                if (remain < b->length) {
+                    apr_bucket_split(b, remain);
                 }
             }
         }
+        status = apr_bucket_copy(b, &b);
+        if (status != APR_SUCCESS) {
+            return status;
+        }
+        APR_BRIGADE_INSERT_TAIL(dest, b);
+        remain -= b->length;
     }
     return status;
 }
@@ -666,25 +847,6 @@ int h2_util_bb_has_data(apr_bucket_brigade *bb)
     return 0;
 }
 
-int h2_util_bb_has_data_or_eos(apr_bucket_brigade *bb)
-{
-    apr_bucket *b;
-    for (b = APR_BRIGADE_FIRST(bb);
-         b != APR_BRIGADE_SENTINEL(bb);
-         b = APR_BUCKET_NEXT(b))
-    {
-        if (APR_BUCKET_IS_METADATA(b)) {
-            if (APR_BUCKET_IS_EOS(b)) {
-                return 1;
-            }
-        }
-        else {
-            return 1;
-        }
-    }
-    return 0;
-}
-
 apr_status_t h2_util_bb_avail(apr_bucket_brigade *bb, 
                               apr_off_t *plen, int *peos)
 {
@@ -789,186 +951,102 @@ apr_status_t h2_util_bb_readx(apr_bucket_brigade *bb,
     return status;
 }
 
-void h2_util_bb_log(conn_rec *c, int stream_id, int level
-                    const char *tag, apr_bucket_brigade *bb)
+apr_size_t h2_util_bucket_print(char *buffer, apr_size_t bmax
+                                apr_bucket *b, const char *sep)
 {
-    char buffer[16 * 1024];
-    const char *line = "(null)";
-    apr_size_t bmax = sizeof(buffer)/sizeof(buffer[0]);
-    int off = 0;
-    apr_bucket *b;
+    apr_size_t off = 0;
+    if (sep && *sep) {
+        off += apr_snprintf(buffer+off, bmax-off, "%s", sep);
+    }
     
-    if (bb) {
-        memset(buffer, 0, bmax--);
-        for (b = APR_BRIGADE_FIRST(bb); 
-             bmax && (b != APR_BRIGADE_SENTINEL(bb));
-             b = APR_BUCKET_NEXT(b)) {
-            
-            if (APR_BUCKET_IS_METADATA(b)) {
-                if (APR_BUCKET_IS_EOS(b)) {
-                    off += apr_snprintf(buffer+off, bmax-off, "eos ");
-                }
-                else if (APR_BUCKET_IS_FLUSH(b)) {
-                    off += apr_snprintf(buffer+off, bmax-off, "flush ");
-                }
-                else if (AP_BUCKET_IS_EOR(b)) {
-                    off += apr_snprintf(buffer+off, bmax-off, "eor ");
-                }
-                else {
-                    off += apr_snprintf(buffer+off, bmax-off, "meta(unknown) ");
-                }
-            }
-            else {
-                const char *btype = "data";
-                if (APR_BUCKET_IS_FILE(b)) {
-                    btype = "file";
-                }
-                else if (APR_BUCKET_IS_PIPE(b)) {
-                    btype = "pipe";
-                }
-                else if (APR_BUCKET_IS_SOCKET(b)) {
-                    btype = "socket";
-                }
-                else if (APR_BUCKET_IS_HEAP(b)) {
-                    btype = "heap";
-                }
-                else if (APR_BUCKET_IS_TRANSIENT(b)) {
-                    btype = "transient";
-                }
-                else if (APR_BUCKET_IS_IMMORTAL(b)) {
-                    btype = "immortal";
-                }
-#if APR_HAS_MMAP
-                else if (APR_BUCKET_IS_MMAP(b)) {
-                    btype = "mmap";
-                }
-#endif
-                else if (APR_BUCKET_IS_POOL(b)) {
-                    btype = "pool";
-                }
-                
-                off += apr_snprintf(buffer+off, bmax-off, "%s[%ld] ", 
-                                    btype, 
-                                    (long)(b->length == ((apr_size_t)-1)? 
-                                           -1 : b->length));
-            }
+    if (APR_BUCKET_IS_METADATA(b)) {
+        if (APR_BUCKET_IS_EOS(b)) {
+            off += apr_snprintf(buffer+off, bmax-off, "eos");
+        }
+        else if (APR_BUCKET_IS_FLUSH(b)) {
+            off += apr_snprintf(buffer+off, bmax-off, "flush");
+        }
+        else if (AP_BUCKET_IS_EOR(b)) {
+            off += apr_snprintf(buffer+off, bmax-off, "eor");
+        }
+        else {
+            off += apr_snprintf(buffer+off, bmax-off, "meta(unknown)");
         }
-        line = *buffer? buffer : "(empty)";
     }
-    /* Intentional no APLOGNO */
-    ap_log_cerror(APLOG_MARK, level, 0, c, "bb_dump(%ld-%d)-%s: %s", 
-                  c->id, stream_id, tag, line);
-
-}
-
-apr_status_t h2_ltransfer_brigade(apr_bucket_brigade *to,
-                                 apr_bucket_brigade *from, 
-                                 apr_pool_t *p,
-                                 apr_off_t *plen,
-                                 int *peos)
-{
-    apr_bucket *e;
-    apr_off_t len = 0, remain = *plen;
-    apr_status_t rv;
-
-    *peos = 0;
-    
-    while (!APR_BRIGADE_EMPTY(from)) {
-        e = APR_BRIGADE_FIRST(from);
-        
-        if (APR_BUCKET_IS_METADATA(e)) {
-            if (APR_BUCKET_IS_EOS(e)) {
-                *peos = 1;
-            }
+    else {
+        const char *btype = "data";
+        if (APR_BUCKET_IS_FILE(b)) {
+            btype = "file";
         }
-        else {        
-            if (remain > 0 && e->length == ((apr_size_t)-1)) {
-                const char *ign;
-                apr_size_t ilen;
-                rv = apr_bucket_read(e, &ign, &ilen, APR_BLOCK_READ);
-                if (rv != APR_SUCCESS) {
-                    return rv;
-                }
-            }
-            
-            if (remain < e->length) {
-                if (remain <= 0) {
-                    return APR_SUCCESS;
-                }
-                apr_bucket_split(e, remain);
-            }
+        else if (APR_BUCKET_IS_PIPE(b)) {
+            btype = "pipe";
         }
-        
-        rv = apr_bucket_setaside(e, p);
-        
-        /* If the bucket type does not implement setaside, then
-         * (hopefully) morph it into a bucket type which does, and set
-         * *that* aside... */
-        if (rv == APR_ENOTIMPL) {
-            const char *s;
-            apr_size_t n;
-            
-            rv = apr_bucket_read(e, &s, &n, APR_BLOCK_READ);
-            if (rv == APR_SUCCESS) {
-                rv = apr_bucket_setaside(e, p);
-            }
+        else if (APR_BUCKET_IS_SOCKET(b)) {
+            btype = "socket";
         }
-        
-        if (rv != APR_SUCCESS) {
-            /* Return an error but still save the brigade if
-             * ->setaside() is really not implemented. */
-            if (rv != APR_ENOTIMPL) {
-                return rv;
-            }
+        else if (APR_BUCKET_IS_HEAP(b)) {
+            btype = "heap";
+        }
+        else if (APR_BUCKET_IS_TRANSIENT(b)) {
+            btype = "transient";
+        }
+        else if (APR_BUCKET_IS_IMMORTAL(b)) {
+            btype = "immortal";
+        }
+#if APR_HAS_MMAP
+        else if (APR_BUCKET_IS_MMAP(b)) {
+            btype = "mmap";
+        }
+#endif
+        else if (APR_BUCKET_IS_POOL(b)) {
+            btype = "pool";
         }
         
-        APR_BUCKET_REMOVE(e);
-        APR_BRIGADE_INSERT_TAIL(to, e);
-        len += e->length;
-        remain -= e->length;
+        off += apr_snprintf(buffer+off, bmax-off, "%s[%ld]", 
+                            btype, 
+                            (long)(b->length == ((apr_size_t)-1)? 
+                                   -1 : b->length));
     }
-    
-    *plen = len;
-    return APR_SUCCESS;
+    return off;
 }
 
-apr_status_t h2_transfer_brigade(apr_bucket_brigade *to,
-                                 apr_bucket_brigade *from
-                                 apr_pool_t *p)
+apr_size_t h2_util_bb_print(char *buffer, apr_size_t bmax, 
+                            const char *tag, const char *sep
+                            apr_bucket_brigade *bb)
 {
-    apr_bucket *e;
-    apr_status_t rv;
-
-    while (!APR_BRIGADE_EMPTY(from)) {
-        e = APR_BRIGADE_FIRST(from);
-        
-        rv = apr_bucket_setaside(e, p);
-        
-        /* If the bucket type does not implement setaside, then
-         * (hopefully) morph it into a bucket type which does, and set
-         * *that* aside... */
-        if (rv == APR_ENOTIMPL) {
-            const char *s;
-            apr_size_t n;
+    apr_size_t off = 0;
+    const char *sp = "";
+    apr_bucket *b;
+    
+    if (bb) {
+        memset(buffer, 0, bmax--);
+        off += apr_snprintf(buffer+off, bmax-off, "%s(", tag);
+        for (b = APR_BRIGADE_FIRST(bb); 
+             bmax && (b != APR_BRIGADE_SENTINEL(bb));
+             b = APR_BUCKET_NEXT(b)) {
             
-            rv = apr_bucket_read(e, &s, &n, APR_BLOCK_READ);
-            if (rv == APR_SUCCESS) {
-                rv = apr_bucket_setaside(e, p);
-            }
-        }
-        
-        if (rv != APR_SUCCESS) {
-            /* Return an error but still save the brigade if
-             * ->setaside() is really not implemented. */
-            if (rv != APR_ENOTIMPL) {
-                return rv;
-            }
+            off += h2_util_bucket_print(buffer+off, bmax-off, b, sp);
+            sp = " ";
         }
-        
-        APR_BUCKET_REMOVE(e);
-        APR_BRIGADE_INSERT_TAIL(to, e);
+        off += apr_snprintf(buffer+off, bmax-off, ")%s", sep);
     }
-    return APR_SUCCESS;
+    else {
+        off += apr_snprintf(buffer+off, bmax-off, "%s(null)%s", tag, sep);
+    }
+    return off;
+}
+
+void h2_util_bb_log(conn_rec *c, int stream_id, int level, 
+                    const char *tag, apr_bucket_brigade *bb)
+{
+    char buffer[4 * 1024];
+    const char *line = "(null)";
+    apr_size_t len, bmax = sizeof(buffer)/sizeof(buffer[0]);
+    
+    len = h2_util_bb_print(buffer, bmax, tag, "", bb);
+    /* Intentional no APLOGNO */
+    ap_log_cerror(APLOG_MARK, level, 0, c, "bb_dump(%ld-%d): %s", 
+                  c->id, stream_id, len? buffer : line);
 }
 
 apr_status_t h2_append_brigade(apr_bucket_brigade *to,
index 4ca2f9b65b68f9181ca3b120f173a19aa2253346..e191c1ee1d4055c40deb3af3bbdc2c985ea12012 100644 (file)
@@ -16,6 +16,8 @@
 #ifndef __mod_h2__h2_util__
 #define __mod_h2__h2_util__
 
+#include <nghttp2/nghttp2.h>
+
 /*******************************************************************************
  * some debugging/format helpers
  ******************************************************************************/
@@ -63,7 +65,120 @@ int h2_ihash_iter(h2_ihash_t *ih, h2_ihash_iter_t *fn, void *ctx);
 void h2_ihash_add(h2_ihash_t *ih, void *val);
 void h2_ihash_remove(h2_ihash_t *ih, int id);
 void h2_ihash_clear(h2_ihash_t *ih);
+
+/*******************************************************************************
+ * ilist - sorted list for structs with int identifier as first member
+ ******************************************************************************/
+typedef struct h2_ilist_t h2_ilist_t;
+typedef int h2_ilist_iter_t(void *ctx, void *val);
+
+h2_ilist_t *h2_ilist_create(apr_pool_t *pool);
+
+apr_status_t h2_ilist_add(h2_ilist_t *list, void *val);
+void *h2_ilist_get(h2_ilist_t *list, int id);
+void *h2_ilist_shift(h2_ilist_t *list);
+void *h2_ilist_remove(h2_ilist_t *list, int id);
+
+int h2_ilist_empty(h2_ilist_t *list);
+apr_size_t h2_ilist_count(h2_ilist_t *list);
+
+/* Iterator over all h2_io* in the set or until a
+ * callback returns 0. It is not safe to add or remove
+ * set members during iteration.
+ *
+ * @param set the set of h2_io to iterate over
+ * @param iter the function to call for each io
+ * @param ctx user data for the callback
+ * @return 1 iff iteration completed for all members
+ */
+int h2_ilist_iter(h2_ilist_t *lis, h2_ilist_iter_t *iter, void *ctx);
+
+/*******************************************************************************
+ * iqueue - sorted list of int with user defined ordering
+ ******************************************************************************/
+typedef struct h2_iqueue {
+    int *elts;
+    int head;
+    int nelts;
+    int nalloc;
+    apr_pool_t *pool;
+} h2_iqueue;
+
+/**
+ * Comparator for two int to determine their order.
+ *
+ * @param i1 first int to compare
+ * @param i2 second int to compare
+ * @param ctx provided user data
+ * @return value is the same as for strcmp() and has the effect:
+ *    == 0: s1 and s2 are treated equal in ordering
+ *     < 0: s1 should be sorted before s2
+ *     > 0: s2 should be sorted before s1
+ */
+typedef int h2_iq_cmp(int i1, int i2, void *ctx);
+
+/**
+ * Allocate a new queue from the pool and initialize.
+ * @param id the identifier of the queue
+ * @param pool the memory pool
+ */
+h2_iqueue *h2_iq_create(apr_pool_t *pool, int capacity);
+
+/**
+ * Return != 0 iff there are no tasks in the queue.
+ * @param q the queue to check
+ */
+int h2_iq_empty(h2_iqueue *q);
+
+/**
+ * Return the number of int in the queue.
+ * @param q the queue to get size on
+ */
+int h2_iq_count(h2_iqueue *q);
+
+/**
+ * Add a stream idto the queue. 
+ *
+ * @param q the queue to append the task to
+ * @param sid the stream id to add
+ * @param cmp the comparator for sorting
+ * @param ctx user data for comparator 
+ */
+void h2_iq_add(h2_iqueue *q, int i, h2_iq_cmp *cmp, void *ctx);
+
+/**
+ * Remove the stream id from the queue. Return != 0 iff task
+ * was found in queue.
+ * @param q the task queue
+ * @param sid the stream id to remove
+ * @return != 0 iff task was found in queue
+ */
+int h2_iq_remove(h2_iqueue *q, int i);
+
+/**
+ * Remove all entries in the queue.
+ */
+void h2_iq_clear(h2_iqueue *q);
+
+/**
+ * Sort the stream idqueue again. Call if the task ordering
+ * has changed.
+ *
+ * @param q the queue to sort
+ * @param cmp the comparator for sorting
+ * @param ctx user data for the comparator 
+ */
+void h2_iq_sort(h2_iqueue *q, h2_iq_cmp *cmp, void *ctx);
+
+/**
+ * Get the first stream id from the queue or NULL if the queue is empty. 
+ * The task will be removed.
+ *
+ * @param q the queue to get the first task from
+ * @return the first stream id of the queue, 0 if empty
+ */
+int h2_iq_shift(h2_iqueue *q);
+
 /*******************************************************************************
  * common helpers
  ******************************************************************************/
@@ -164,33 +279,23 @@ h2_ngheader *h2_util_ngheader_make_req(apr_pool_t *p,
 /*******************************************************************************
  * apr brigade helpers
  ******************************************************************************/
+
 /**
- * Moves data from one brigade into another. If maxlen > 0, it only
- * moves up to maxlen bytes into the target brigade, making bucket splits
- * if needed.
- * @param to the brigade to move the data to
- * @param from the brigade to get the data from
- * @param maxlen of bytes to move, <= 0 for all
- * @param pfile_buckets_allowed how many file buckets may be moved, 
- *        may be 0 or NULL
- * @param msg message for use in logging
+ * Concatenate at most length bytes from src to dest brigade, splitting
+ * buckets if necessary and reading buckets of indeterminate length.
  */
-apr_status_t h2_util_move(apr_bucket_brigade *to, apr_bucket_brigade *from
-                          apr_off_t maxlen, apr_size_t *pfile_buckets_allowed, 
-                          const char *msg);
-
+apr_status_t h2_brigade_concat_length(apr_bucket_brigade *dest
+                                      apr_bucket_brigade *src,
+                                      apr_off_t length);
+                                
 /**
- * Copies buckets from one brigade into another. If maxlen > 0, it only
- * copies up to maxlen bytes into the target brigade, making bucket splits
- * if needed.
- * @param to the brigade to copy the data to
- * @param from the brigade to get the data from
- * @param maxlen of bytes to copy, <= 0 for all
- * @param msg message for use in logging
+ * Copy at most length bytes from src to dest brigade, splitting
+ * buckets if necessary and reading buckets of indeterminate length.
  */
-apr_status_t h2_util_copy(apr_bucket_brigade *to, apr_bucket_brigade *from, 
-                          apr_off_t maxlen, const char *msg);
-
+apr_status_t h2_brigade_copy_length(apr_bucket_brigade *dest, 
+                                    apr_bucket_brigade *src,
+                                    apr_off_t length);
+                                
 /**
  * Return != 0 iff there is a FLUSH or EOS bucket in the brigade.
  * @param bb the brigade to check on
@@ -198,7 +303,6 @@ apr_status_t h2_util_copy(apr_bucket_brigade *to, apr_bucket_brigade *from,
  */
 int h2_util_has_eos(apr_bucket_brigade *bb, apr_off_t len);
 int h2_util_bb_has_data(apr_bucket_brigade *bb);
-int h2_util_bb_has_data_or_eos(apr_bucket_brigade *bb);
 
 /**
  * Check how many bytes of the desired amount are available and if the
@@ -229,6 +333,21 @@ apr_status_t h2_util_bb_readx(apr_bucket_brigade *bb,
                               h2_util_pass_cb *cb, void *ctx, 
                               apr_off_t *plen, int *peos);
 
+/**
+ * Print a bucket's meta data (type and length) to the buffer.
+ * @return number of characters printed
+ */
+apr_size_t h2_util_bucket_print(char *buffer, apr_size_t bmax, 
+                                apr_bucket *b, const char *sep);
+                                
+/**
+ * Prints the brigade bucket types and lengths into the given buffer
+ * up to bmax.
+ * @return number of characters printed
+ */
+apr_size_t h2_util_bb_print(char *buffer, apr_size_t bmax, 
+                            const char *tag, const char *sep, 
+                            apr_bucket_brigade *bb);
 /**
  * Logs the bucket brigade (which bucket types with what length)
  * to the log at the given level.
@@ -241,33 +360,6 @@ apr_status_t h2_util_bb_readx(apr_bucket_brigade *bb,
 void h2_util_bb_log(conn_rec *c, int stream_id, int level, 
                     const char *tag, apr_bucket_brigade *bb);
 
-/**
- * Transfer buckets from one brigade to another with a limit on the 
- * maximum amount of bytes transfered. Sets aside the buckets to
- * pool p.
- * @param to   brigade to transfer buckets to
- * @param from brigades to remove buckets from
- * @param p    pool that buckets should be setaside to
- * @param plen maximum bytes to transfer, actual bytes transferred
- * @param peos if an EOS bucket was transferred
- */
-apr_status_t h2_ltransfer_brigade(apr_bucket_brigade *to,
-                                  apr_bucket_brigade *from, 
-                                  apr_pool_t *p,
-                                  apr_off_t *plen,
-                                  int *peos);
-
-/**
- * Transfer all buckets from one brigade to another. Sets aside the buckets to
- * pool p.
- * @param to   brigade to transfer buckets to
- * @param from brigades to remove buckets from
- * @param p    pool that buckets should be setaside to
- */
-apr_status_t h2_transfer_brigade(apr_bucket_brigade *to,
-                                 apr_bucket_brigade *from, 
-                                 apr_pool_t *p);
-
 /**
  * Transfer buckets from one brigade to another with a limit on the 
  * maximum amount of bytes transfered. Does no setaside magic, lifetime
index a9bd29e895a2d486bff13f22d76f33176fcd130f..191af8be4c8200776fe0313d49f810ae7a587d75 100644 (file)
@@ -26,7 +26,7 @@
  * @macro
  * Version number of the http2 module as c string
  */
-#define MOD_HTTP2_VERSION "1.4.7-DEV"
+#define MOD_HTTP2_VERSION "1.5.0-DEV"
 
 /**
  * @macro
@@ -34,7 +34,7 @@
  * release. This is a 24 bit number with 8 bits for major number, 8 bits
  * for minor and 8 bits for patch. Version 1.2.3 becomes 0x010203.
  */
-#define MOD_HTTP2_VERSION_NUM 0x010407
+#define MOD_HTTP2_VERSION_NUM 0x010500
 
 
 #endif /* mod_h2_h2_version_h */
index ca6ce3a2f289e3731b899313b33dfbf865af20b7..e394298e7da05d8383199c48323835aacc7a4af8 100644 (file)
@@ -42,10 +42,8 @@ static void* APR_THREAD_FUNC execute(apr_thread_t *thread, void *wctx)
         /* Get a h2_task from the main workers queue. */
         worker->get_next(worker, worker->ctx, &task, &sticky);
         while (task) {
-            h2_task_do(task, worker->io);
-            
-            /* if someone was waiting on this task, time to wake up */
-            apr_thread_cond_signal(worker->io);
+        
+            h2_task_do(task);
             /* report the task done and maybe get another one from the same
              * mplx (= master connection), if we can be sticky. 
              */
@@ -64,40 +62,20 @@ static void* APR_THREAD_FUNC execute(apr_thread_t *thread, void *wctx)
 }
 
 h2_worker *h2_worker_create(int id,
-                            apr_pool_t *parent_pool,
+                            apr_pool_t *pool,
                             apr_threadattr_t *attr,
                             h2_worker_mplx_next_fn *get_next,
                             h2_worker_done_fn *worker_done,
                             void *ctx)
 {
-    apr_allocator_t *allocator = NULL;
-    apr_pool_t *pool = NULL;
-    h2_worker *w;
-    apr_status_t status;
-    
-    apr_allocator_create(&allocator);
-    apr_allocator_max_free_set(allocator, ap_max_mem_free);
-    apr_pool_create_ex(&pool, parent_pool, NULL, allocator);
-    apr_pool_tag(pool, "h2_worker");
-    apr_allocator_owner_set(allocator, pool);
-
-    w = apr_pcalloc(pool, sizeof(h2_worker));
+    h2_worker *w = apr_pcalloc(pool, sizeof(h2_worker));
     if (w) {
-        APR_RING_ELEM_INIT(w, link);
-        
         w->id = id;
-        w->pool = pool;
-
+        APR_RING_ELEM_INIT(w, link);
         w->get_next = get_next;
         w->worker_done = worker_done;
         w->ctx = ctx;
-        
-        status = apr_thread_cond_create(&w->io, w->pool);
-        if (status != APR_SUCCESS) {
-            return NULL;
-        }
-        
-        apr_thread_create(&w->thread, attr, execute, w, w->pool);
+        apr_thread_create(&w->thread, attr, execute, w, pool);
     }
     return w;
 }
@@ -109,22 +87,9 @@ apr_status_t h2_worker_destroy(h2_worker *worker)
         apr_thread_join(&status, worker->thread);
         worker->thread = NULL;
     }
-    if (worker->io) {
-        apr_thread_cond_destroy(worker->io);
-        worker->io = NULL;
-    }
-    if (worker->pool) {
-        apr_pool_destroy(worker->pool);
-        /* worker is gone */
-    }
     return APR_SUCCESS;
 }
 
-int h2_worker_get_id(h2_worker *worker)
-{
-    return worker->id;
-}
-
 void h2_worker_abort(h2_worker *worker)
 {
     worker->aborted = 1;
index 7a8c254f5d7b492a67427c4609558300e98fd89c..04ff5703616e7f911401b892fc3050a784d3ed35 100644 (file)
@@ -16,7 +16,6 @@
 #ifndef __mod_h2__h2_worker__
 #define __mod_h2__h2_worker__
 
-struct apr_thread_cond_t;
 struct h2_mplx;
 struct h2_request;
 struct h2_task;
@@ -39,19 +38,14 @@ typedef void h2_worker_done_fn(h2_worker *worker, void *ctx);
 
 
 struct h2_worker {
+    int id;
     /** Links to the rest of the workers */
     APR_RING_ENTRY(h2_worker) link;
-    
-    int id;
     apr_thread_t *thread;
-    apr_pool_t *pool;
-    struct apr_thread_cond_t *io;
-    
     h2_worker_mplx_next_fn *get_next;
     h2_worker_done_fn *worker_done;
     void *ctx;
-    
-    unsigned int aborted : 1;
+    int aborted;
 };
 
 /**
@@ -136,8 +130,6 @@ apr_status_t h2_worker_destroy(h2_worker *worker);
 
 void h2_worker_abort(h2_worker *worker);
 
-int h2_worker_get_id(h2_worker *worker);
-
 int h2_worker_is_aborted(h2_worker *worker);
 
 #endif /* defined(__mod_h2__h2_worker__) */
index 2c1dc8dab42d5add5aa4fcb918f92b39a4fd46d4..2a1599914c65f3fd8aba0d34b2f531bc22a0115d 100644 (file)
@@ -116,7 +116,7 @@ static apr_status_t get_mplx_next(h2_worker *worker, void *ctx,
     if (status == APR_SUCCESS) {
         ++workers->idle_workers;
         ap_log_error(APLOG_MARK, APLOG_TRACE3, 0, workers->s,
-                     "h2_worker(%d): looking for work", h2_worker_get_id(worker));
+                     "h2_worker(%d): looking for work", worker->id);
         
         while (!h2_worker_is_aborted(worker) && !workers->aborted
                && !(task = next_task(workers))) {
@@ -195,7 +195,7 @@ static void worker_done(h2_worker *worker, void *ctx)
     apr_status_t status = apr_thread_mutex_lock(workers->lock);
     if (status == APR_SUCCESS) {
         ap_log_error(APLOG_MARK, APLOG_TRACE3, 0, workers->s,
-                     "h2_worker(%d): done", h2_worker_get_id(worker));
+                     "h2_worker(%d): done", worker->id);
         H2_WORKER_REMOVE(worker);
         --workers->worker_count;
         H2_WORKER_LIST_INSERT_TAIL(&workers->zombies, worker);
@@ -213,7 +213,7 @@ static apr_status_t add_worker(h2_workers *workers)
         return APR_ENOMEM;
     }
     ap_log_error(APLOG_MARK, APLOG_TRACE3, 0, workers->s,
-                 "h2_workers: adding worker(%d)", h2_worker_get_id(w));
+                 "h2_workers: adding worker(%d)", w->id);
     ++workers->worker_count;
     H2_WORKER_LIST_INSERT_TAIL(&workers->workers, w);
     return APR_SUCCESS;
index 0d33969161d6dd7ac617aab57e3d0fca4b7eb6a5..480917a4191687975fdc70fc0f7d3d8947fa6470 100644 (file)
@@ -57,6 +57,13 @@ AP_DECLARE_MODULE(http2) = {
 
 static int h2_h2_fixups(request_rec *r);
 
+typedef struct {
+    unsigned int change_prio : 1;
+    unsigned int sha256 : 1;
+} features;
+
+static features myfeats;
+
 /* The module initialization. Called once as apache hook, before any multi
  * processing (threaded or not) happens. It is typically at least called twice, 
  * see
@@ -77,7 +84,16 @@ static int h2_post_config(apr_pool_t *p, apr_pool_t *plog,
     const char *mod_h2_init_key = "mod_http2_init_counter";
     nghttp2_info *ngh2;
     apr_status_t status;
+    const char *sep = "";
+    
     (void)plog;(void)ptemp;
+#ifdef H2_NG2_CHANGE_PRIO
+    myfeats.change_prio = 1;
+    sep = "+";
+#endif
+#ifdef H2_OPENSSL
+    myfeats.sha256 = 1;
+#endif
     
     apr_pool_userdata_get(&data, mod_h2_init_key, s->process->pool);
     if ( data == NULL ) {
@@ -90,8 +106,11 @@ static int h2_post_config(apr_pool_t *p, apr_pool_t *plog,
     
     ngh2 = nghttp2_version(0);
     ap_log_error( APLOG_MARK, APLOG_INFO, 0, s, APLOGNO(03090)
-                 "mod_http2 (v%s, nghttp2 %s), initializing...",
-                 MOD_HTTP2_VERSION, ngh2? ngh2->version_str : "unknown");
+                 "mod_http2 (v%s, feats=%s%s%s, nghttp2 %s), initializing...",
+                 MOD_HTTP2_VERSION, 
+                 myfeats.change_prio? "CHPRIO" : "", sep, 
+                 myfeats.sha256?      "SHA256" : "",
+                 ngh2?                ngh2->version_str : "unknown");
     
     switch (h2_conn_mpm_type()) {
         case H2_MPM_SIMPLE:
index eb55028a06744b835d95b99c2dfff63d4e6d5ab6..cc8f66638712df75820a40f26baebf09d712d784 100644 (file)
@@ -105,6 +105,10 @@ SOURCE=./h2_alt_svc.c
 # End Source File
 # Begin Source File
 
+SOURCE=./h2_bucket_beam.c
+# End Source File
+# Begin Source File
+
 SOURCE=./h2_bucket_eoc.c
 # End Source File
 # Begin Source File
@@ -141,18 +145,10 @@ SOURCE=./h2_h2.c
 # End Source File
 # Begin Source File
 
-SOURCE=./h2_int_queue.c
-# End Source File
-# Begin Source File
-
 SOURCE=./h2_io.c
 # End Source File
 # Begin Source File
 
-SOURCE=./h2_io_set.c
-# End Source File
-# Begin Source File
-
 SOURCE=./h2_mplx.c
 # End Source File
 # Begin Source File
@@ -189,14 +185,6 @@ SOURCE=./h2_task.c
 # End Source File
 # Begin Source File
 
-SOURCE=./h2_task_input.c
-# End Source File
-# Begin Source File
-
-SOURCE=./h2_task_output.c
-# End Source File
-# Begin Source File
-
 SOURCE=./h2_util.c
 # End Source File
 # Begin Source File
index 7335e9527ff6d9ad640d52a6c0613b8781d346bb..f64326549742060c523e4f54513f61d4c348dcdb 100644 (file)
@@ -21,7 +21,6 @@
 
 
 #include "mod_proxy_http2.h"
-#include "h2_int_queue.h"
 #include "h2_request.h"
 #include "h2_util.h"
 #include "h2_version.h"