blob: 68c15d13e4906f3c4cfe94b0d236e2fe5f5efcfa [file] [log] [blame]
/* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You 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_strings.h>
#include <ap_mpm.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_bucket_eos.h"
#include "h2_config.h"
#include "h2_conn_io.h"
#include "h2_h2.h"
#include "h2_session.h"
#include "h2_util.h"
#define TLS_DATA_MAX (16*1024)
/* Calculated like this: assuming MTU 1500 bytes
* 1500 - 40 (IP) - 20 (TCP) - 40 (TCP options)
* - TLS overhead (60-100)
* ~= 1300 bytes */
#define WRITE_SIZE_INITIAL 1300
/* The maximum we'd like to write in one chunk is
* the max size of a TLS record. When pushing
* many frames down the h2 connection, this might
* align differently because of headers and other
* frames or simply as not sufficient data is
* in a response body.
* However keeping frames at or below this limit
* should make optimizations at the layer that writes
* to TLS easier.
*/
#define WRITE_SIZE_MAX (TLS_DATA_MAX)
#define BUF_REMAIN ((apr_size_t)(bmax-off))
static void h2_conn_io_bb_log(conn_rec *c, int stream_id, int level,
const char *tag, apr_bucket_brigade *bb)
{
char buffer[16 * 1024];
const char *line = "(null)";
int bmax = sizeof(buffer)/sizeof(buffer[0]);
int off = 0;
apr_bucket *b;
(void)stream_id;
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, BUF_REMAIN, "eos ");
}
else if (APR_BUCKET_IS_FLUSH(b)) {
off += apr_snprintf(buffer+off, BUF_REMAIN, "flush ");
}
else if (AP_BUCKET_IS_EOR(b)) {
off += apr_snprintf(buffer+off, BUF_REMAIN, "eor ");
}
else if (H2_BUCKET_IS_H2EOS(b)) {
off += apr_snprintf(buffer+off, BUF_REMAIN, "h2eos ");
}
else {
off += apr_snprintf(buffer+off, BUF_REMAIN, "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, BUF_REMAIN, "%s[%ld] ",
btype,
(long)(b->length == ((apr_size_t)-1)? -1UL : b->length));
}
}
line = *buffer? buffer : "(empty)";
}
/* Intentional no APLOGNO */
ap_log_cerror(APLOG_MARK, level, 0, c, "h2_session(%ld)-%s: %s",
c->id, tag, line);
}
apr_status_t h2_conn_io_init(h2_conn_io *io, conn_rec *c, server_rec *s)
{
io->c = c;
io->output = apr_brigade_create(c->pool, c->bucket_alloc);
io->is_tls = h2_h2_is_tls(c);
io->buffer_output = io->is_tls;
io->flush_threshold = (apr_size_t)h2_config_sgeti64(s, H2_CONF_STREAM_MAX_MEM);
if (io->is_tls) {
/* This is what we start with,
* see https://issues.apache.org/jira/browse/TS-2503
*/
io->warmup_size = h2_config_sgeti64(s, H2_CONF_TLS_WARMUP_SIZE);
io->cooldown_usecs = (h2_config_sgeti(s, H2_CONF_TLS_COOLDOWN_SECS)
* APR_USEC_PER_SEC);
io->write_size = (io->cooldown_usecs > 0?
WRITE_SIZE_INITIAL : WRITE_SIZE_MAX);
}
else {
io->warmup_size = 0;
io->cooldown_usecs = 0;
io->write_size = 0;
}
if (APLOGctrace1(c)) {
ap_log_cerror(APLOG_MARK, APLOG_TRACE4, 0, io->c,
"h2_conn_io(%ld): init, buffering=%d, warmup_size=%ld, "
"cd_secs=%f", io->c->id, io->buffer_output,
(long)io->warmup_size,
((double)io->cooldown_usecs/APR_USEC_PER_SEC));
}
return APR_SUCCESS;
}
static void append_scratch(h2_conn_io *io)
{
if (io->scratch && io->slen > 0) {
apr_bucket *b = apr_bucket_heap_create(io->scratch, io->slen,
apr_bucket_free,
io->c->bucket_alloc);
APR_BRIGADE_INSERT_TAIL(io->output, b);
io->scratch = NULL;
io->slen = io->ssize = 0;
}
}
static apr_size_t assure_scratch_space(h2_conn_io *io) {
apr_size_t remain = io->ssize - io->slen;
if (io->scratch && remain == 0) {
append_scratch(io);
}
if (!io->scratch) {
/* we control the size and it is larger than what buckets usually
* allocate. */
io->scratch = apr_bucket_alloc(io->write_size, io->c->bucket_alloc);
io->ssize = io->write_size;
io->slen = 0;
remain = io->ssize;
}
return remain;
}
static apr_status_t read_to_scratch(h2_conn_io *io, apr_bucket *b)
{
apr_status_t status;
const char *data;
apr_size_t len;
if (!b->length) {
return APR_SUCCESS;
}
ap_assert(b->length <= (io->ssize - io->slen));
if (APR_BUCKET_IS_FILE(b)) {
apr_bucket_file *f = (apr_bucket_file *)b->data;
apr_file_t *fd = f->fd;
apr_off_t offset = b->start;
len = b->length;
/* file buckets will either mmap (which we do not want) or
* read 8000 byte chunks and split themself. However, we do
* know *exactly* how many bytes we need where.
*/
status = apr_file_seek(fd, APR_SET, &offset);
if (status != APR_SUCCESS) {
return status;
}
status = apr_file_read(fd, io->scratch + io->slen, &len);
if (status != APR_SUCCESS && status != APR_EOF) {
return status;
}
io->slen += len;
}
else {
status = apr_bucket_read(b, &data, &len, APR_BLOCK_READ);
if (status == APR_SUCCESS) {
memcpy(io->scratch+io->slen, data, len);
io->slen += len;
}
}
return status;
}
static void check_write_size(h2_conn_io *io)
{
if (io->write_size > WRITE_SIZE_INITIAL
&& (io->cooldown_usecs > 0)
&& (apr_time_now() - io->last_write) >= io->cooldown_usecs) {
/* long time not written, reset write size */
io->write_size = WRITE_SIZE_INITIAL;
io->bytes_written = 0;
}
else if (io->write_size < WRITE_SIZE_MAX
&& io->bytes_written >= io->warmup_size) {
/* connection is hot, use max size */
io->write_size = WRITE_SIZE_MAX;
}
}
static apr_status_t pass_output(h2_conn_io *io, int flush)
{
conn_rec *c = io->c;
apr_bucket_brigade *bb = io->output;
apr_bucket *b;
apr_off_t bblen;
apr_status_t status;
append_scratch(io);
if (flush && !io->is_flushed) {
b = apr_bucket_flush_create(c->bucket_alloc);
APR_BRIGADE_INSERT_TAIL(bb, b);
}
if (APR_BRIGADE_EMPTY(bb)) {
return APR_SUCCESS;
}
ap_update_child_status(c->sbh, SERVER_BUSY_WRITE, NULL);
apr_brigade_length(bb, 0, &bblen);
h2_conn_io_bb_log(c, 0, APLOG_TRACE2, "out", bb);
status = ap_pass_brigade(c->output_filters, bb);
if (status == APR_SUCCESS) {
io->bytes_written += (apr_size_t)bblen;
io->last_write = apr_time_now();
if (flush) {
io->is_flushed = 1;
}
}
apr_brigade_cleanup(bb);
if (status != APR_SUCCESS) {
ap_log_cerror(APLOG_MARK, APLOG_DEBUG, status, c, APLOGNO(03044)
"h2_conn_io(%ld): pass_out brigade %ld bytes",
c->id, (long)bblen);
}
return status;
}
int h2_conn_io_needs_flush(h2_conn_io *io)
{
if (!io->is_flushed) {
apr_off_t len = h2_brigade_mem_size(io->output);
if (len > (apr_off_t)io->flush_threshold) {
return 1;
}
/* if we do not exceed flush length due to memory limits,
* we want at least flush when we have that amount of data. */
apr_brigade_length(io->output, 0, &len);
return len > (apr_off_t)(4 * io->flush_threshold);
}
return 0;
}
apr_status_t h2_conn_io_flush(h2_conn_io *io)
{
apr_status_t status;
status = pass_output(io, 1);
check_write_size(io);
return status;
}
apr_status_t h2_conn_io_write(h2_conn_io *io, const char *data, size_t length)
{
apr_status_t status = APR_SUCCESS;
apr_size_t remain;
if (length > 0) {
io->is_flushed = 0;
}
if (io->buffer_output) {
while (length > 0) {
remain = assure_scratch_space(io);
if (remain >= length) {
memcpy(io->scratch + io->slen, data, length);
io->slen += length;
length = 0;
}
else {
memcpy(io->scratch + io->slen, data, remain);
io->slen += remain;
data += remain;
length -= remain;
}
}
}
else {
status = apr_brigade_write(io->output, NULL, NULL, data, length);
}
return status;
}
apr_status_t h2_conn_io_pass(h2_conn_io *io, apr_bucket_brigade *bb)
{
apr_bucket *b;
apr_status_t status = APR_SUCCESS;
if (!APR_BRIGADE_EMPTY(bb)) {
io->is_flushed = 0;
}
while (!APR_BRIGADE_EMPTY(bb) && status == APR_SUCCESS) {
b = APR_BRIGADE_FIRST(bb);
if (APR_BUCKET_IS_METADATA(b)) {
/* need to finish any open scratch bucket, as meta data
* needs to be forward "in order". */
append_scratch(io);
APR_BUCKET_REMOVE(b);
APR_BRIGADE_INSERT_TAIL(io->output, b);
}
else if (io->buffer_output) {
apr_size_t remain = assure_scratch_space(io);
if (b->length > remain) {
apr_bucket_split(b, remain);
if (io->slen == 0) {
/* complete write_size bucket, append unchanged */
APR_BUCKET_REMOVE(b);
APR_BRIGADE_INSERT_TAIL(io->output, b);
continue;
}
}
else {
/* bucket fits in remain, copy to scratch */
status = read_to_scratch(io, b);
apr_bucket_delete(b);
continue;
}
}
else {
/* no buffering, forward buckets setaside on flush */
if (APR_BUCKET_IS_TRANSIENT(b)) {
apr_bucket_setaside(b, io->c->pool);
}
APR_BUCKET_REMOVE(b);
APR_BRIGADE_INSERT_TAIL(io->output, b);
}
}
return status;
}