--- old/ngx_http_request.c 2009-04-05 20:41:18.000000000 +0300
+++ new/ngx_http_request.c 2009-04-05 20:38:33.000000000 +0300
@@ -1403,16 +1403,6 @@
return NGX_ERROR;
}
- if (r->headers_in.transfer_encoding
- && ngx_strcasestrn(r->headers_in.transfer_encoding->value.data,
- "chunked", 7 - 1))
- {
- ngx_log_error(NGX_LOG_INFO, r->connection->log, 0,
- "client sent \"Transfer-Encoding: chunked\" header");
- ngx_http_finalize_request(r, NGX_HTTP_LENGTH_REQUIRED);
- return NGX_ERROR;
- }
-
if (r->headers_in.connection_type == NGX_HTTP_CONNECTION_KEEP_ALIVE) {
if (r->headers_in.keep_alive) {
r->headers_in.keep_alive_n =
--- old/ngx_http_request_body.c 2009-04-05 20:41:39.000000000 +0300
+++ new/ngx_http_request_body.c 2009-04-05 20:34:45.000000000 +0300
@@ -54,11 +54,6 @@
r->request_body = rb;
- if (r->headers_in.content_length_n < 0) {
- post_handler(r);
- return NGX_OK;
- }
-
clcf = ngx_http_get_module_loc_conf(r, ngx_http_core_module);
if (r->headers_in.content_length_n == 0) {
@@ -180,7 +175,8 @@
} else {
b = NULL;
- rb->rest = r->headers_in.content_length_n;
+ rb->rest = r->headers_in.content_length_n >= 0 ? r->headers_in.content_length_n
+ : NGX_MAX_SIZE_T_VALUE;
next = &rb->bufs;
}In the mean time I've sorted out for myself the server's internals. What can I say? First of all the event-driven architecture is complicated. Especially, when there's no high-level description, who is calling what. And especially under virtually complete absence of comments in the code.
And about C and Lisp (based on examining the source of the two HTTP-servers: nginx and Hunchentoot). Lisp is clay, shape whatever you want with all the connected advantages and shortcomings. C... There's a children's construction set from metal bars and corners. It's fun to play with them. And, generally, everything is quite clear, it's possible to assemble good, solid constructions. But making a vase with rounded corners — it's for the geeks...
2 comments:
Alas, this solution was a premature one: turned out, that it relied on nginx handling of preread data, which worked only on fast connections (for example, in my testing environment on localhost :).
Below is the more adequate one:
--- old/ngx_http_request_body.c 2009-04-05 20:44:46.000000000 +0300
+++ new/ngx_http_request_body.c 2009-04-09 12:03:56.000000000 +0300
@@ -54,7 +54,7 @@
r->request_body = rb;
- if (r->headers_in.content_length_n < 0) {
+ if (r->headers_in.content_length_n < 0 && !(r->method & NGX_HTTP_POST)) {
post_handler(r);
return NGX_OK;
}
@@ -137,7 +137,8 @@
rb->buf = b;
- if ((off_t) preread >= r->headers_in.content_length_n) {
+ if (r->headers_in.content_length_n >= 0 &&
+ (off_t) preread >= r->headers_in.content_length_n) {
/* the whole request body was pre-read */
@@ -163,9 +164,12 @@
r->request_length += preread;
- rb->rest = r->headers_in.content_length_n - preread;
+ if (r->headers_in.content_length_n >= 0) {
+ rb->rest = r->headers_in.content_length_n - preread;
+ }
- if (rb->rest <= (off_t) (b->end - b->last)) {
+ if (r->headers_in.content_length_n >= 0 &&
+ rb->rest <= (off_t) (b->end - b->last)) {
/* the whole request body may be placed in r->header_in */
@@ -180,14 +184,14 @@
} else {
b = NULL;
- rb->rest = r->headers_in.content_length_n;
+ rb->rest = r->headers_in.content_length_n >= 0 ?r->headers_in.content_length_n : 1024;
next = &rb->bufs;
}
size = clcf->client_body_buffer_size;
size += size >> 2;
- if (rb->rest < size) {
+ if (rb->rest >=0 && rb->rest < size) {
size = (ssize_t) rb->rest;
if (r->request_body_in_single_buf) {
@@ -224,13 +228,15 @@
*next = cl;
- if (r->request_body_in_file_only || r->request_body_in_single_buf) {
+ if (r->request_body_in_file_only || r->request_body_in_single_buf || r->headers_in.content_length_n < 0) {
rb->to_write = rb->bufs;
} else {
rb->to_write = rb->bufs->next ? rb->bufs->next : rb->bufs;
}
+ // if (r->headers_in.content_length_n < 0) {
+
r->read_event_handler = ngx_http_read_client_request_body_handler;
return ngx_http_do_read_client_request_body(r);
@@ -310,7 +316,13 @@
}
rb->buf->last += n;
- rb->rest -= n;
+
+ if ( ngx_memcmp(rb->buf->last - 7, (u_char *) CRLF "0" CRLF CRLF, sizeof(u_char)) ) {
+ rb->rest -= n;
+ } else {
+ rb->rest = 0;
+ }
+
r->request_length += n;
if (rb->rest == 0) {
Thank you for such a detailed and insightful post! The explanation of how to tackle the Transfer-Encoding: chunked issue in Nginx was clear and well-structured. Your step-by-step approach and the inclusion of practical examples really helped me understand the root cause of the problem and how to address it effectively. It's great to see complex topics broken down in a way that's accessible to developers at all levels. Keep up the fantastic work! Global Talent Pathway
Post a Comment