]> git.ipfire.org Git - people/mlorenz/ipfire-2.x.git/commitdiff
apache2: Import patch for PR61382
authorMichael Tremer <michael.tremer@ipfire.org>
Thu, 7 Sep 2017 11:27:43 +0000 (12:27 +0100)
committerMichael Tremer <michael.tremer@ipfire.org>
Thu, 7 Sep 2017 11:27:43 +0000 (12:27 +0100)
We usually do not download patches, but rather ship them with
our source.

Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
lfs/apache2
src/patches/apache-2.4.27-PR61382-fix.patch [new file with mode: 0644]

index 72a9dfad1a12a3d9ea13fbc5f60a039dfeaa6e6a..e434d1b244203ce8b920a26e2100d7bb972f6681 100644 (file)
@@ -41,14 +41,11 @@ DEPS       = "aprutil pcre"
 # Top-level Rules
 ###############################################################################
 
-objects = $(DL_FILE) \
-       PR61382-Fix.patch
+objects = $(DL_FILE)
 
 $(DL_FILE) = $(DL_FROM)/$(DL_FILE)
-PR61382-Fix.patch = ${DL_FROM}/patches/apply_to_2.4.27/PR61382-Fix.patch
 
 $(DL_FILE)_MD5 = 97b6bbfa83c866dbe20ef317e3afd108
-PR61382-Fix.patch_MD5 = 5c1107bb1f399419574d983ce103c99a
 
 install : $(TARGET)
 
@@ -78,7 +75,7 @@ $(subst %,%_MD5,$(objects)) :
 $(TARGET) : $(patsubst %,$(DIR_DL)/%,$(objects))
        @$(PREBUILD)
        @rm -rf $(DIR_APP) && cd $(DIR_SRC) && tar jxf $(DIR_DL)/$(DL_FILE)
-       cd $(DIR_APP) && patch -Np0 -i $(DIR_DL)/PR61382-Fix.patch
+       cd $(DIR_APP) && patch -Np0 -i $(DIR_SRC)/src/patches/apache-2.4.27-PR61382-fix.patch
 
        ### Add IPFire's layout, too
        echo "# IPFire layout" >> $(DIR_APP)/config.layout
diff --git a/src/patches/apache-2.4.27-PR61382-fix.patch b/src/patches/apache-2.4.27-PR61382-fix.patch
new file mode 100644 (file)
index 0000000..128621a
--- /dev/null
@@ -0,0 +1,783 @@
+Index: modules/http2/h2_bucket_beam.c
+===================================================================
+--- modules/http2/h2_bucket_beam.c     (revision 1804645)
++++ modules/http2/h2_bucket_beam.c     (working copy)
+@@ -287,7 +287,7 @@
+             /* do not count */
+         }
+         else if (APR_BUCKET_IS_FILE(b)) {
+-            /* if unread, has no real mem footprint. how to test? */
++            /* if unread, has no real mem footprint. */
+         }
+         else {
+             len += b->length;
+@@ -316,32 +316,80 @@
+     return APR_SIZE_MAX;
+ }
+-static apr_status_t wait_cond(h2_bucket_beam *beam, apr_thread_mutex_t *lock)
++static int buffer_is_empty(h2_bucket_beam *beam)
+ {
+-    if (beam->timeout > 0) {
+-        return apr_thread_cond_timedwait(beam->cond, lock, beam->timeout);
++    return ((!beam->recv_buffer || APR_BRIGADE_EMPTY(beam->recv_buffer))
++            && H2_BLIST_EMPTY(&beam->send_list));
++}
++
++static apr_status_t wait_empty(h2_bucket_beam *beam, apr_read_type_e block,  
++                               apr_thread_mutex_t *lock)
++{
++    apr_status_t rv = APR_SUCCESS;
++    
++    while (!buffer_is_empty(beam) && APR_SUCCESS == rv) {
++        if (APR_BLOCK_READ != block || !lock) {
++            rv = APR_EAGAIN;
++        }
++        else if (beam->timeout > 0) {
++            rv = apr_thread_cond_timedwait(beam->change, lock, beam->timeout);
++        }
++        else {
++            rv = apr_thread_cond_wait(beam->change, lock);
++        }
+     }
+-    else {
+-        return apr_thread_cond_wait(beam->cond, lock);
++    return rv;
++}
++
++static apr_status_t wait_not_empty(h2_bucket_beam *beam, apr_read_type_e block,  
++                                   apr_thread_mutex_t *lock)
++{
++    apr_status_t rv = APR_SUCCESS;
++    
++    while (buffer_is_empty(beam) && APR_SUCCESS == rv) {
++        if (beam->aborted) {
++            rv = APR_ECONNABORTED;
++        }
++        else if (beam->closed) {
++            rv = APR_EOF;
++        }
++        else if (APR_BLOCK_READ != block || !lock) {
++            rv = APR_EAGAIN;
++        }
++        else if (beam->timeout > 0) {
++            rv = apr_thread_cond_timedwait(beam->change, lock, beam->timeout);
++        }
++        else {
++            rv = apr_thread_cond_wait(beam->change, lock);
++        }
+     }
++    return rv;
+ }
+-static apr_status_t r_wait_space(h2_bucket_beam *beam, apr_read_type_e block,
+-                                 h2_beam_lock *pbl, apr_size_t *premain) 
++static apr_status_t wait_not_full(h2_bucket_beam *beam, apr_read_type_e block, 
++                                  apr_size_t *pspace_left, h2_beam_lock *bl)
+ {
+-    *premain = calc_space_left(beam);
+-    while (!beam->aborted && *premain <= 0 
+-           && (block == APR_BLOCK_READ) && pbl->mutex) {
+-        apr_status_t status;
+-        report_prod_io(beam, 1, pbl);
+-        status = wait_cond(beam, pbl->mutex);
+-        if (APR_STATUS_IS_TIMEUP(status)) {
+-            return status;
++    apr_status_t rv = APR_SUCCESS;
++    apr_size_t left;
++    
++    while (0 == (left = calc_space_left(beam)) && APR_SUCCESS == rv) {
++        if (beam->aborted) {
++            rv = APR_ECONNABORTED;
+         }
+-        r_purge_sent(beam);
+-        *premain = calc_space_left(beam);
++        else if (block != APR_BLOCK_READ || !bl->mutex) {
++            rv = APR_EAGAIN;
++        }
++        else {
++            if (beam->timeout > 0) {
++                rv = apr_thread_cond_timedwait(beam->change, bl->mutex, beam->timeout);
++            }
++            else {
++                rv = apr_thread_cond_wait(beam->change, bl->mutex);
++            }
++        }
+     }
+-    return beam->aborted? APR_ECONNABORTED : APR_SUCCESS;
++    *pspace_left = left;
++    return rv;
+ }
+ static void h2_beam_emitted(h2_bucket_beam *beam, h2_beam_proxy *proxy)
+@@ -404,8 +452,8 @@
+         if (!bl.mutex) {
+             r_purge_sent(beam);
+         }
+-        else if (beam->cond) {
+-            apr_thread_cond_broadcast(beam->cond);
++        else {
++            apr_thread_cond_broadcast(beam->change);
+         }
+         leave_yellow(beam, &bl);
+     }
+@@ -425,9 +473,7 @@
+ {
+     if (!beam->closed) {
+         beam->closed = 1;
+-        if (beam->cond) {
+-            apr_thread_cond_broadcast(beam->cond);
+-        }
++        apr_thread_cond_broadcast(beam->change);
+     }
+     return APR_SUCCESS;
+ }
+@@ -582,7 +628,7 @@
+                             apr_interval_time_t timeout)
+ {
+     h2_bucket_beam *beam;
+-    apr_status_t status = APR_SUCCESS;
++    apr_status_t rv = APR_SUCCESS;
+     
+     beam = apr_pcalloc(pool, sizeof(*beam));
+     if (!beam) {
+@@ -601,16 +647,15 @@
+     beam->max_buf_size = max_buf_size;
+     beam->timeout = timeout;
+-    status = apr_thread_mutex_create(&beam->lock, APR_THREAD_MUTEX_DEFAULT, 
+-                                     pool);
+-    if (status == APR_SUCCESS) {
+-        status = apr_thread_cond_create(&beam->cond, pool);
+-        if (status == APR_SUCCESS) {
++    rv = apr_thread_mutex_create(&beam->lock, APR_THREAD_MUTEX_DEFAULT, pool);
++    if (APR_SUCCESS == rv) {
++        rv = apr_thread_cond_create(&beam->change, pool);
++        if (APR_SUCCESS == rv) {
+             apr_pool_pre_cleanup_register(pool, beam, beam_cleanup);
+             *pbeam = beam;
+         }
+     }
+-    return status;
++    return rv;
+ }
+ void h2_beam_buffer_size_set(h2_bucket_beam *beam, apr_size_t buffer_size)
+@@ -691,9 +736,7 @@
+             h2_blist_cleanup(&beam->send_list);
+             report_consumption(beam, &bl);
+         }
+-        if (beam->cond) {
+-            apr_thread_cond_broadcast(beam->cond);
+-        }
++        apr_thread_cond_broadcast(beam->change);
+         leave_yellow(beam, &bl);
+     }
+ }
+@@ -730,18 +773,7 @@
+     h2_beam_lock bl;
+     
+     if ((status = enter_yellow(beam, &bl)) == APR_SUCCESS) {
+-        while (status == APR_SUCCESS
+-               && !H2_BLIST_EMPTY(&beam->send_list)
+-               && !H2_BPROXY_LIST_EMPTY(&beam->proxies)) {
+-            if (block == APR_NONBLOCK_READ || !bl.mutex) {
+-                status = APR_EAGAIN;
+-                break;
+-            }
+-            if (beam->cond) {
+-                apr_thread_cond_broadcast(beam->cond);
+-            }
+-            status = wait_cond(beam, bl.mutex);
+-        }
++        status = wait_empty(beam, block, bl.mutex);
+         leave_yellow(beam, &bl);
+     }
+     return status;
+@@ -761,13 +793,18 @@
+ static apr_status_t append_bucket(h2_bucket_beam *beam, 
+                                   apr_bucket *b,
+                                   apr_read_type_e block,
++                                  apr_size_t *pspace_left,
+                                   h2_beam_lock *pbl)
+ {
+     const char *data;
+     apr_size_t len;
+-    apr_size_t space_left = 0;
+     apr_status_t status;
++    int can_beam, check_len;
+     
++    if (beam->aborted) {
++        return APR_ECONNABORTED;
++    }
++    
+     if (APR_BUCKET_IS_METADATA(b)) {
+         if (APR_BUCKET_IS_EOS(b)) {
+             beam->closed = 1;
+@@ -777,11 +814,31 @@
+         return APR_SUCCESS;
+     }
+     else if (APR_BUCKET_IS_FILE(b)) {
+-        /* file bucket lengths do not really count */
++        /* 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 only from the sender thread.
++         *
++         * Currently, we do not handle file bucket with refcount > 1 as
++         * the beam is then not in complete control of the file's lifetime.
++         * Which results in the bug that a file get closed by the receiver
++         * while the sender or the beam still have buckets using it. 
++         * 
++         * 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_bucket_file *bf = b->data;
++        apr_file_t *fd = bf->fd;
++        can_beam = (bf->refcount.refcount == 1);
++        if (can_beam && beam->can_beam_fn) {
++            can_beam = beam->can_beam_fn(beam->can_beam_ctx, beam, fd);
++        }
++        check_len = !can_beam;
+     }
+     else {
+-        space_left = calc_space_left(beam);
+-        if (space_left > 0 && b->length == ((apr_size_t)-1)) {
++        if (b->length == ((apr_size_t)-1)) {
+             const char *data;
+             status = apr_bucket_read(b, &data, &len, APR_BLOCK_READ);
+             if (status != APR_SUCCESS) {
+@@ -788,19 +845,15 @@
+                 return status;
+             }
+         }
+-        
+-        if (space_left <= 0) {
+-            status = r_wait_space(beam, block, pbl, &space_left);
+-            if (status != APR_SUCCESS) {
+-                return status;
+-            }
+-            if (space_left <= 0) {
+-                return APR_EAGAIN;
+-            }
++        check_len = 1;
++    }
++    
++    if (check_len) {
++        if (b->length > *pspace_left) {
++            apr_bucket_split(b, *pspace_left);
+         }
+-        /* space available, maybe need bucket split */
++        *pspace_left -= b->length;
+     }
+-    
+     /* The fundamental problem is that reading a sender bucket from
+      * a receiver thread is a total NO GO, because the bucket might use
+@@ -830,32 +883,8 @@
+             apr_bucket_heap_make(b, data, len, NULL);
+         }
+     }
+-    else if (APR_BUCKET_IS_FILE(b)) {
+-        /* 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 only from the sender thread.
+-         *
+-         * Currently, we do not handle file bucket with refcount > 1 as
+-         * the beam is then not in complete control of the file's lifetime.
+-         * Which results in the bug that a file get closed by the receiver
+-         * while the sender or the beam still have buckets using it. 
+-         * 
+-         * 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_bucket_file *bf = b->data;
+-        apr_file_t *fd = bf->fd;
+-        int can_beam = (bf->refcount.refcount == 1);
+-        if (can_beam && beam->can_beam_fn) {
+-            can_beam = beam->can_beam_fn(beam->can_beam_ctx, beam, fd);
+-        }
+-        if (can_beam) {
+-            status = apr_bucket_setaside(b, beam->send_pool);
+-        }
+-        /* else: enter ENOTIMPL case below */
++    else if (APR_BUCKET_IS_FILE(b) && can_beam) {
++        status = apr_bucket_setaside(b, beam->send_pool);
+     }
+     
+     if (status == APR_ENOTIMPL) {
+@@ -865,12 +894,6 @@
+          * a counter example).
+          * We do the read while in the sender thread, so that the bucket may
+          * use pools/allocators safely. */
+-        if (space_left < APR_BUCKET_BUFF_SIZE) {
+-            space_left = APR_BUCKET_BUFF_SIZE;
+-        }
+-        if (space_left < b->length) {
+-            apr_bucket_split(b, space_left);
+-        }
+         status = apr_bucket_read(b, &data, &len, APR_BLOCK_READ);
+         if (status == APR_SUCCESS) {
+             status = apr_bucket_setaside(b, beam->send_pool);
+@@ -884,7 +907,7 @@
+     APR_BUCKET_REMOVE(b);
+     H2_BLIST_INSERT_TAIL(&beam->send_list, b);
+     beam->sent_bytes += b->length;
+-    
++
+     return APR_SUCCESS;
+ }
+@@ -904,7 +927,8 @@
+                           apr_read_type_e block)
+ {
+     apr_bucket *b;
+-    apr_status_t status = APR_SUCCESS;
++    apr_status_t rv = APR_SUCCESS;
++    apr_size_t space_left = 0;
+     h2_beam_lock bl;
+     /* Called from the sender thread to add buckets to the beam */
+@@ -914,23 +938,31 @@
+         
+         if (beam->aborted) {
+             move_to_hold(beam, sender_bb);
+-            status = APR_ECONNABORTED;
++            rv = APR_ECONNABORTED;
+         }
+         else if (sender_bb) {
+-            int force_report = !APR_BRIGADE_EMPTY(sender_bb); 
+-            while (!APR_BRIGADE_EMPTY(sender_bb) && status == APR_SUCCESS) {
++            int force_report = !APR_BRIGADE_EMPTY(sender_bb);
++            
++            space_left = calc_space_left(beam);
++            while (!APR_BRIGADE_EMPTY(sender_bb) && APR_SUCCESS == rv) {
++                if (space_left <= 0) {
++                    report_prod_io(beam, force_report, &bl);
++                    rv = wait_not_full(beam, block, &space_left, &bl);
++                    if (APR_SUCCESS != rv) {
++                        break;
++                    }
++                }
+                 b = APR_BRIGADE_FIRST(sender_bb);
+-                status = append_bucket(beam, b, block, &bl);
++                rv = append_bucket(beam, b, block, &space_left, &bl);
+             }
++            
+             report_prod_io(beam, force_report, &bl);
+-            if (beam->cond) {
+-                apr_thread_cond_broadcast(beam->cond);
+-            }
++            apr_thread_cond_broadcast(beam->change);
+         }
+         report_consumption(beam, &bl);
+         leave_yellow(beam, &bl);
+     }
+-    return status;
++    return rv;
+ }
+ apr_status_t h2_beam_receive(h2_bucket_beam *beam, 
+@@ -942,11 +974,16 @@
+     apr_bucket *bsender, *brecv, *ng;
+     int transferred = 0;
+     apr_status_t status = APR_SUCCESS;
+-    apr_off_t remain = readbytes;
++    apr_off_t remain;
+     int transferred_buckets = 0;
+     
+     /* Called from the receiver thread to take buckets from the beam */
+     if (enter_yellow(beam, &bl) == APR_SUCCESS) {
++        if (readbytes <= 0) {
++            readbytes = APR_SIZE_MAX;
++        }
++        remain = readbytes;
++        
+ transfer:
+         if (beam->aborted) {
+             recv_buffer_cleanup(beam, &bl);
+@@ -955,11 +992,12 @@
+         }
+         /* transfer enough buckets from our receiver brigade, if we have one */
+-        while (beam->recv_buffer
+-               && !APR_BRIGADE_EMPTY(beam->recv_buffer)
+-               && (readbytes <= 0 || remain >= 0)) {
++        while (remain >= 0 
++               && beam->recv_buffer 
++               && !APR_BRIGADE_EMPTY(beam->recv_buffer)) {
++               
+             brecv = APR_BRIGADE_FIRST(beam->recv_buffer);
+-            if (readbytes > 0 && brecv->length > 0 && remain <= 0) {
++            if (brecv->length > 0 && remain <= 0) {
+                 break;
+             }            
+             APR_BUCKET_REMOVE(brecv);
+@@ -970,11 +1008,11 @@
+         /* transfer from our sender brigade, transforming sender buckets to
+          * receiver ones until we have enough */
+-        while (!H2_BLIST_EMPTY(&beam->send_list) && (readbytes <= 0 || remain >= 0)) {
+-            bsender = H2_BLIST_FIRST(&beam->send_list);
++        while (remain >= 0 && !H2_BLIST_EMPTY(&beam->send_list)) {
++               
+             brecv = NULL;
+-            
+-            if (readbytes > 0 && bsender->length > 0 && remain <= 0) {
++            bsender = H2_BLIST_FIRST(&beam->send_list);            
++            if (bsender->length > 0 && remain <= 0) {
+                 break;
+             }
+                         
+@@ -1020,11 +1058,12 @@
+                  * been handed out. See also PR 59348 */
+                 apr_bucket_file_enable_mmap(ng, 0);
+ #endif
+-                remain -= bsender->length;
+-                ++transferred;
+                 APR_BUCKET_REMOVE(bsender);
+                 H2_BLIST_INSERT_TAIL(&beam->hold_list, bsender);
++
++                remain -= bsender->length;
+                 ++transferred;
++                ++transferred_buckets;
+                 continue;
+             }
+             else {
+@@ -1041,6 +1080,7 @@
+              * receiver bucket references it any more. */
+             APR_BUCKET_REMOVE(bsender);
+             H2_BLIST_INSERT_TAIL(&beam->hold_list, bsender);
++            
+             beam->received_bytes += bsender->length;
+             ++transferred_buckets;
+             
+@@ -1063,8 +1103,8 @@
+             }
+         }
+-        if (readbytes > 0 && remain < 0) {
+-            /* too much, put some back */
++        if (remain < 0) {
++            /* too much, put some back into out recv_buffer */
+             remain = readbytes;
+             for (brecv = APR_BRIGADE_FIRST(bb);
+                  brecv != APR_BRIGADE_SENTINEL(bb);
+@@ -1081,15 +1121,7 @@
+             }
+         }
+-        if (transferred_buckets > 0) {
+-           if (beam->cons_ev_cb) { 
+-               beam->cons_ev_cb(beam->cons_ctx, beam);
+-            }
+-        }
+-        
+-        if (beam->closed 
+-            && (!beam->recv_buffer || APR_BRIGADE_EMPTY(beam->recv_buffer))
+-            && H2_BLIST_EMPTY(&beam->send_list)) {
++        if (beam->closed && buffer_is_empty(beam)) {
+             /* beam is closed and we have nothing more to receive */ 
+             if (!beam->close_sent) {
+                 apr_bucket *b = apr_bucket_eos_create(bb->bucket_alloc);
+@@ -1100,28 +1132,23 @@
+             }
+         }
+         
++        if (transferred_buckets > 0) {
++           if (beam->cons_ev_cb) { 
++               beam->cons_ev_cb(beam->cons_ctx, beam);
++            }
++        }
++        
+         if (transferred) {
+-            if (beam->cond) {
+-                apr_thread_cond_broadcast(beam->cond);
+-            }
++            apr_thread_cond_broadcast(beam->change);
+             status = APR_SUCCESS;
+         }
+-        else if (beam->closed) {
+-            status = APR_EOF;
+-        }
+-        else if (block == APR_BLOCK_READ && bl.mutex && beam->cond) {
+-            status = wait_cond(beam, bl.mutex);
++        else {
++            status = wait_not_empty(beam, block, bl.mutex);
+             if (status != APR_SUCCESS) {
+                 goto leave;
+             }
+             goto transfer;
+         }
+-        else {
+-            if (beam->cond) {
+-                apr_thread_cond_broadcast(beam->cond);
+-            }
+-            status = APR_EAGAIN;
+-        }
+ leave:        
+         leave_yellow(beam, &bl);
+     }
+Index: modules/http2/h2_bucket_beam.h
+===================================================================
+--- modules/http2/h2_bucket_beam.h     (revision 1804645)
++++ modules/http2/h2_bucket_beam.h     (working copy)
+@@ -190,7 +190,7 @@
+     unsigned int tx_mem_limits : 1; /* only memory size counts on transfers */
+     struct apr_thread_mutex_t *lock;
+-    struct apr_thread_cond_t *cond;
++    struct apr_thread_cond_t *change;
+     void *m_ctx;
+     h2_beam_mutex_enter *m_enter;
+     
+Index: modules/http2/h2_stream.c
+===================================================================
+--- modules/http2/h2_stream.c  (revision 1804645)
++++ modules/http2/h2_stream.c  (working copy)
+@@ -774,20 +774,20 @@
+     return NULL;
+ }
+-static apr_status_t add_data(h2_stream *stream, apr_off_t requested,
+-                             apr_off_t *plen, int *peos, int *complete, 
+-                             h2_headers **pheaders)
++static apr_status_t add_buffered_data(h2_stream *stream, apr_off_t requested,
++                                      apr_off_t *plen, int *peos, int *is_all, 
++                                      h2_headers **pheaders)
+ {
+     apr_bucket *b, *e;
+     
+     *peos = 0;
+     *plen = 0;
+-    *complete = 0;
++    *is_all = 0;
+     if (pheaders) {
+         *pheaders = NULL;
+     }
+-    H2_STREAM_OUT_LOG(APLOG_TRACE2, stream, "add_data");
++    H2_STREAM_OUT_LOG(APLOG_TRACE2, stream, "add_buffered_data");
+     b = APR_BRIGADE_FIRST(stream->out_buffer);
+     while (b != APR_BRIGADE_SENTINEL(stream->out_buffer)) {
+         e = APR_BUCKET_NEXT(b);
+@@ -833,7 +833,7 @@
+         }
+         b = e;
+     }
+-    *complete = 1;
++    *is_all = 1;
+     return APR_SUCCESS;
+ }
+@@ -865,7 +865,7 @@
+     requested = (*plen > 0)? H2MIN(*plen, max_chunk) : max_chunk;
+     
+     /* count the buffered data until eos or a headers bucket */
+-    status = add_data(stream, requested, plen, peos, &complete, pheaders);
++    status = add_buffered_data(stream, requested, plen, peos, &complete, pheaders);
+     
+     if (status == APR_EAGAIN) {
+         /* TODO: ugly, someone needs to retrieve the response first */
+@@ -882,29 +882,39 @@
+         return APR_SUCCESS;
+     }
+     
++    /* If there we do not have enough buffered data to satisfy the requested
++     * length *and* we counted the _complete_ buffer (and did not stop in the middle
++     * because of meta data there), lets see if we can read more from the
++     * output beam */
+     missing = H2MIN(requested, stream->max_mem) - *plen;
+     if (complete && !*peos && missing > 0) {
++        apr_status_t rv = APR_EOF;
++        
+         if (stream->output) {
+             H2_STREAM_OUT_LOG(APLOG_TRACE2, stream, "pre");
+-            status = h2_beam_receive(stream->output, stream->out_buffer, 
+-                                     APR_NONBLOCK_READ, 
+-                                     stream->max_mem - *plen);
++            rv = h2_beam_receive(stream->output, stream->out_buffer, 
++                                 APR_NONBLOCK_READ, stream->max_mem - *plen);
+             H2_STREAM_OUT_LOG(APLOG_TRACE2, stream, "post");
+         }
+-        else {
+-            status = APR_EOF;
++        
++        if (rv == APR_SUCCESS) {
++            /* count the buffer again, now that we have read output */
++            status = add_buffered_data(stream, requested, plen, peos, &complete, pheaders);
+         }
+-        
+-        if (APR_STATUS_IS_EOF(status)) {
++        else if (APR_STATUS_IS_EOF(rv)) {
+             apr_bucket *eos = apr_bucket_eos_create(c->bucket_alloc);
+             APR_BRIGADE_INSERT_TAIL(stream->out_buffer, eos);
+             *peos = 1;
+-            status = APR_SUCCESS;
+         }
+-        else if (status == APR_SUCCESS) {
+-            /* do it again, now that we have gotten more */
+-            status = add_data(stream, requested, plen, peos, &complete, pheaders);
++        else if (APR_STATUS_IS_EAGAIN(rv)) {
++            /* we set this is the status of this call only if there
++             * is no buffered data, see check below */
+         }
++        else {
++            /* real error reading. Give this back directly, even though
++             * we may have something buffered. */
++            status = rv;
++        }
+     }
+     
+     if (status == APR_SUCCESS) {
+Index: modules/http2/h2_task.c
+===================================================================
+--- modules/http2/h2_task.c    (revision 1804645)
++++ modules/http2/h2_task.c    (working copy)
+@@ -129,7 +129,7 @@
+                               apr_bucket_brigade* bb)
+ {
+     apr_bucket *b;
+-    apr_status_t status = APR_SUCCESS;
++    apr_status_t rv = APR_SUCCESS;
+     int flush = 0, blocking;
+     
+     if (task->frozen) {
+@@ -148,17 +148,16 @@
+         return APR_SUCCESS;
+     }
++send:
+     /* we send block once we opened the output, so someone is there
+      * reading it *and* the task is not assigned to a h2_req_engine */
+     blocking = (!task->assigned && task->output.opened);
+-    if (!task->output.opened) {
+-        for (b = APR_BRIGADE_FIRST(bb);
+-             b != APR_BRIGADE_SENTINEL(bb);
+-             b = APR_BUCKET_NEXT(b)) {
+-            if (APR_BUCKET_IS_FLUSH(b)) {
+-                flush = 1;
+-                break;
+-            }
++    for (b = APR_BRIGADE_FIRST(bb);
++         b != APR_BRIGADE_SENTINEL(bb);
++         b = APR_BUCKET_NEXT(b)) {
++        if (APR_BUCKET_IS_FLUSH(b) || APR_BUCKET_IS_EOS(b) || AP_BUCKET_IS_EOR(b)) {
++            flush = 1;
++            break;
+         }
+     }
+     
+@@ -166,32 +165,48 @@
+         /* still have data buffered from previous attempt.
+          * setaside and append new data and try to pass the complete data */
+         if (!APR_BRIGADE_EMPTY(bb)) {
+-            status = ap_save_brigade(f, &task->output.bb, &bb, task->pool);
++            if (APR_SUCCESS != (rv = ap_save_brigade(f, &task->output.bb, &bb, task->pool))) {
++                goto out;
++            }
+         }
+-        if (status == APR_SUCCESS) {
+-            status = send_out(task, task->output.bb, blocking);
+-        } 
++        rv = send_out(task, task->output.bb, blocking);
+     }
+     else {
+-        /* no data buffered here, try to pass the brigade directly */
+-        status = send_out(task, bb, blocking); 
+-        if (status == APR_SUCCESS && !APR_BRIGADE_EMPTY(bb)) {
+-            /* could not write all, buffer the rest */
+-            ap_log_cerror(APLOG_MARK, APLOG_DEBUG, status, task->c, APLOGNO(03405)
+-                          "h2_slave_out(%s): saving brigade", 
+-                          task->id);
+-            status = ap_save_brigade(f, &task->output.bb, &bb, task->pool);
+-            flush = 1;
++        /* no data buffered previously, pass brigade directly */
++        rv = send_out(task, bb, blocking);
++
++        if (APR_SUCCESS == rv && !APR_BRIGADE_EMPTY(bb)) {
++            /* output refused to buffer it all, time to open? */
++            if (!task->output.opened && APR_SUCCESS == (rv = open_output(task))) {
++                /* Make another attempt to send the data. With the output open,
++                 * the call might be blocking and send all data, so we do not need
++                 * to save the brigade */
++                goto send;
++            }
++            else if (blocking && flush) {
++                /* Need to keep on doing this. */
++                goto send;
++            }
++            
++            if (APR_SUCCESS == rv) {
++                /* could not write all, buffer the rest */
++                ap_log_cerror(APLOG_MARK, APLOG_DEBUG, rv, task->c, APLOGNO(03405)
++                              "h2_slave_out(%s): saving brigade", task->id);
++                ap_assert(NULL);
++                rv = ap_save_brigade(f, &task->output.bb, &bb, task->pool);
++                flush = 1;
++            }
+         }
+     }
+     
+-    if (status == APR_SUCCESS && !task->output.opened && flush) {
++    if (APR_SUCCESS == rv && !task->output.opened && flush) {
+         /* got a flush or could not write all, time to tell someone to read */
+-        status = open_output(task);
++        rv = open_output(task);
+     }
+-    ap_log_cerror(APLOG_MARK, APLOG_TRACE2, status, task->c, 
++out:
++    ap_log_cerror(APLOG_MARK, APLOG_TRACE2, rv, task->c, 
+                   "h2_slave_out(%s): slave_out leave", task->id);    
+-    return status;
++    return rv;
+ }
+ static apr_status_t output_finish(h2_task *task)
+Index: modules/http2/h2_version.h
+===================================================================
+--- modules/http2/h2_version.h (revision 1804645)
++++ modules/http2/h2_version.h (working copy)
+@@ -26,7 +26,7 @@
+  * @macro
+  * Version number of the http2 module as c string
+  */
+-#define MOD_HTTP2_VERSION "1.10.7"
++#define MOD_HTTP2_VERSION "1.10.10"
+ /**
+  * @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 0x010a06
++#define MOD_HTTP2_VERSION_NUM 0x010a0a
+ #endif /* mod_h2_h2_version_h */
+Index: modules/http2
+===================================================================
+--- modules/http2      (revision 1804645)
++++ modules/http2      (working copy)
+
+Property changes on: modules/http2
+___________________________________________________________________
+Modified: svn:mergeinfo
+## -0,0 +0,1 ##
+   Merged /httpd/httpd/trunk/modules/http2:r1803420,1803454,1804090
+Index: .
+===================================================================
+--- .  (revision 1804645)
++++ .  (working copy)
+
+Property changes on: .
+___________________________________________________________________
+Modified: svn:mergeinfo
+## -0,0 +0,1 ##
+   Merged /httpd/httpd/trunk:r1803420,1803454,1804090