Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

1138 lines (979 sloc) 33.204 kb
/*
Onion HTTP server library
Copyright (C) 2010-2014 David Moreno Montero and othes
This library is free software; you can redistribute it and/or
modify it under the terms of, at your choice:
a. the Apache License Version 2.0.
b. the GNU General Public License as published by the
Free Software Foundation; either version 2.0 of the License,
or (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of both libraries, if not see
<http://www.gnu.org/licenses/> and
<http://www.apache.org/licenses/LICENSE-2.0>.
*/
#include <string.h>
#include <stdlib.h>
#include <libgen.h>
#include <ctype.h>
#include <fcntl.h>
#include <unistd.h>
#include <assert.h>
#include "dict.h"
#include "request.h"
#include "types_internal.h"
#include "codecs.h"
#include "log.h"
#include "block.h"
#include "low.h"
#include "ptr_list.h"
/**
* @short Known token types. This is merged with onion_connection_status as return value at token readers.
* @private
*/
typedef enum{
STRING=1001, // I start on 1000, to prevent collission with error codes
KEY=1002,
LINE=1003,
NEW_LINE=1004,
URLENCODE=1005,
MULTIPART_BOUNDARY=1006,
MULTIPART_END=1007,
MULTIPART_NEXT=1008,
STRING_NEW_LINE=1009,
}onion_token_token;
/// @private
typedef struct onion_token_s{
char str[256*4*8];
off_t pos;
char *extra; // Only used when need some previous data, like at header value, i need the key
size_t extra_size;
}onion_token;
/// @private
typedef struct onion_buffer_s{
const char *data;
size_t size;
off_t pos;
}onion_buffer;
/**
* @short This struct is mapped at token->extra. The token, _start and data are maped to token->extra area too.
* @private
*
* This is a bit juggling with pointers, but needed to keep the deallocation simple, and the token object minimal in most cases.
*
* token->pos is the pointer to the free area and token->extra_size is the real size.
*/
typedef struct onion_multipart_buffer_s{
char *boundary; /// Pointer to where the boundary is stored
size_t size; /// Size of the boundary
off_t pos; /// Pos at the boundarycheck
off_t startpos; /// Where did the boundary started (\n\r vs \n)
char *data; /// When need a new buffered data (always from stream), get it from here. This advances as used.
char *name; /// point to a in data position
char *filename; /// point to a in data position
size_t file_total_size; /// Total size of all file read data
size_t post_total_size; /// Total size of post, to check limits
int fd; /// If file, the file descriptor.
}onion_multipart_buffer;
static void onion_request_parse_query_to_dict(onion_dict *dict, char *p);
static int onion_request_parse_query(onion_request *req);
static onion_connection_status prepare_POST(onion_request *req);
static onion_connection_status prepare_CONTENT_LENGTH(onion_request *req);
static onion_connection_status prepare_PUT(onion_request *req);
/// Reads a string until a non-string char. Returns an onion_token
int token_read_STRING(onion_token *token, onion_buffer *data){
if (data->pos>=data->size)
return OCS_NEED_MORE_DATA;
char c=data->data[data->pos++];
while (!isspace(c)){
token->str[token->pos++]=c;
if (data->pos>=data->size)
return OCS_NEED_MORE_DATA;
if (token->pos>=(sizeof(token->str)-1)){
char tmp[16];
strncpy(tmp, token->str, 16);
tmp[15]='\0';
ONION_ERROR("Token too long to parse it. Part read start as %s (%d bytes)",tmp,token->pos);
return OCS_INTERNAL_ERROR;
}
c=data->data[data->pos++];
}
int ret;
if (c=='\n')
ret=STRING_NEW_LINE;
else
ret=STRING;
token->str[token->pos]='\0';
token->pos=0;
//ONION_DEBUG0("Found STRING token %s",token->str);
return ret;
}
/**
* @short Reads a string until a delimiter is found. Returns an onion_token. Also detects an empty line.
*
* @returns error code | STRING | STRING_NEW_LINE | NEW_LINE
*
* STRING is a delimited string, string new line is that the string finished witha new line, new line is that I fuond only a new line char.
*/
///
int token_read_until(onion_token *token, onion_buffer *data, char delimiter){
if (data->pos>=data->size)
return OCS_NEED_MORE_DATA;
//ONION_DEBUG0("Read data %d bytes, at token pos %d",data->size-data->pos, token->pos);
char c=data->data[data->pos++];
while (c!=delimiter && c!='\n'){
if (c!='\r') // Just ignore it here
token->str[token->pos++]=c;
if (data->pos>=data->size){
return OCS_NEED_MORE_DATA;
}
if (token->pos>=(sizeof(token->str)-1)){
ONION_ERROR("Token too long to parse it. Part read is %s (%d bytes)",token->str,token->pos);
return OCS_INTERNAL_ERROR;
}
c=data->data[data->pos++];
}
int ret=STRING;
token->str[token->pos]='\0';
if (c!=delimiter){
if ( token->pos==0 && c=='\n' ){
ret=NEW_LINE;
}
else{ // only option left.
ret=STRING_NEW_LINE;
}
}
token->pos=0;
//ONION_DEBUG0("Found KEY token %s",token->str);
return ret;
}
/// Reads a key, that is a string ended with ':'.
int token_read_KEY(onion_token *token, onion_buffer *data){
int res=token_read_until(token, data, ':');
if (res==STRING)
return KEY;
if (res==STRING_NEW_LINE){
ONION_ERROR("When parsing header, found a non valid delimited string token: '%s'",token->str);
return OCS_INTERNAL_ERROR;
}
return res;
}
/// Reads a string until a '\n|\r\n' is found. Returns an onion_token.
int token_read_LINE(onion_token *token, onion_buffer *data){
if (data->pos>=data->size)
return OCS_NEED_MORE_DATA;
char c=data->data[data->pos++];
int ignore_to_end=0;
while (c!='\n'){
if (!ignore_to_end && (token->pos>=(sizeof(token->str)-1))){
ONION_WARNING("Token too long to parse it. Ignoring remaining. ");
#ifdef __DEBUG__
char tmp[16];
strncpy(tmp, token->str, 16);
tmp[15]='\0';
ONION_DEBUG("Long token starts with: %s...",tmp);
#endif
ignore_to_end=1;
}
if (!ignore_to_end)
token->str[token->pos++]=c;
if (data->pos>=data->size)
return OCS_NEED_MORE_DATA;
c=data->data[data->pos++];
}
if (token->str[token->pos-1]=='\r')
token->str[token->pos-1]='\0';
else
token->str[token->pos]='\0';
//token->pos=0;
//ONION_DEBUG0("Found LINE token %s",token->str);
return LINE;
}
/// Reads a string until a '\n|\r\n' is found. Returns an onion_token.
int token_read_URLENCODE(onion_token *token, onion_buffer *data){
int l=data->size-data->pos;
if (l > (token->extra_size-token->pos))
l=token->extra_size-token->pos;
//ONION_DEBUG0("Feed %d bytes",l);
memcpy(&token->extra[token->pos], &data->data[data->pos], l);
token->pos+=l;
data->size-=l;
if (token->extra_size==token->pos){
token->extra[token->pos]='\0';
return URLENCODE;
}
return OCS_NEED_MORE_DATA;
}
/// Reads as much as possible from the boundary.
int token_read_MULTIPART_BOUNDARY(onion_token *token, onion_buffer *data){
onion_multipart_buffer *multipart=(onion_multipart_buffer*)token->extra;
while (multipart->boundary[multipart->pos] && multipart->boundary[multipart->pos]==data->data[data->pos]){
multipart->pos++;
data->pos++;
if (data->pos>=data->size)
return OCS_NEED_MORE_DATA;
}
if (multipart->boundary[multipart->pos]){
ONION_ERROR("Expecting multipart boundary, but not here (pos %d) (%c!=%c)", multipart->pos, data->data[data->pos], multipart->boundary[multipart->pos]);
return OCS_INTERNAL_ERROR;
}
return MULTIPART_BOUNDARY;
}
/// Reads as much as possible from the boundary.
int token_read_MULTIPART_next(onion_token *token, onion_buffer *data){
//onion_multipart_buffer *multipart=(onion_multipart_buffer*)token->extra;
while (token->pos<2 && (data->size-data->pos)>0){
token->str[token->pos++]=data->data[data->pos++];
if (token->pos==1 && token->str[0]=='\n'){
token->pos=0;
return MULTIPART_NEXT;
}
}
//ONION_DEBUG("pos %d, %d %d (%c %c)",token->pos, token->str[0], token->str[1], token->str[0], token->str[1]);
if (token->pos==2){
token->pos=0;
if (token->str[0]=='-' && token->str[1]=='-')
return MULTIPART_END;
if (token->str[0]=='\r' && token->str[1]=='\n')
return MULTIPART_NEXT;
return OCS_INTERNAL_ERROR;
}
return OCS_NEED_MORE_DATA;
}
int token_read_NEW_LINE(onion_token *token, onion_buffer *data){
while (token->pos<2 && (data->size-data->pos)>0){
token->str[token->pos++]=data->data[data->pos++];
if (token->pos==1 && token->str[0]=='\n'){
token->pos=0;
return NEW_LINE;
}
}
//ONION_DEBUG("pos %d, %d %d (%c %c)",token->pos, token->str[0], token->str[1], token->str[0], token->str[1]);
int res=OCS_NEED_MORE_DATA;
if (token->pos==2){
token->pos=0;
if (token->str[0]=='\r' && token->str[1]=='\n')
res=NEW_LINE;
else
res=OCS_INTERNAL_ERROR;
}
token->pos=0;
return res;
}
static char token_peek_next_char(onion_buffer *data){
if (data->pos>=data->size)
return '\0'; // If not enought data, go to safe side.
return data->data[data->pos];
}
static onion_connection_status parse_POST_multipart_next(onion_request *req, onion_buffer *data);
/**
* @short Reads from the data to fulfill content-length data.
*/
static onion_connection_status parse_CONTENT_LENGTH(onion_request *req, onion_buffer *data){
ONION_DEBUG0("Adding data to request->data (non form POST)");
onion_token *token=req->parser_data;
int length=data->size-data->pos;
int exit=0;
if (length>=token->extra_size-token->pos){
exit=1;
length=token->extra_size-token->pos;
}
onion_block_add_data(req->data, &data->data[data->pos], length);
data->pos+=length;
if (exit)
return onion_request_process(req);
return OCS_NEED_MORE_DATA;
}
/**
* @short Reads from the data to fulfill content-length data.
*
* All data is writen a temporal file, which will be removed later.
*/
static onion_connection_status parse_PUT(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int length=data->size-data->pos;
int exit=0;
if (length>=token->extra_size-token->pos){
exit=1;
length=token->extra_size-token->pos;
}
//ONION_DEBUG0("Writing %d. %d / %d bytes", length, token->pos+length, token->extra_size);
int *fd=(int*)token->extra;
ssize_t w=write(*fd, &data->data[data->pos], length);
if (w<0){
ONION_ERROR("Could not write all data to temporal file.");
return OCS_INTERNAL_ERROR;
}
data->pos+=length;
token->pos+=length;
#if __DEBUG__
const char *filename=onion_block_data(req->data);
ONION_DEBUG0("Done with PUT. Created %s (%d bytes)", filename, token->pos);
#endif
if (exit){
close (*fd);
onion_low_free(token->extra);
token->extra=NULL;
return onion_request_process(req);
}
return OCS_NEED_MORE_DATA;
}
/**
* Hard parser as I must set into the file as I read, until i found the boundary token (or start), try to parse, and if fail,
* write to the file.
*
* The boundary may be in two or N parts.
*/
static onion_connection_status parse_POST_multipart_file(onion_request *req, onion_buffer *data){
char tmp[1024]; /// Read 1024 bytes fo the read on this call, and write. If read less write as read.
int tmppos=0;
onion_token *token=req->parser_data;
onion_multipart_buffer *multipart=(onion_multipart_buffer*)token->extra;
const char *p=data->data+data->pos;
for (;data->pos<data->size;data->pos++){
//ONION_DEBUG("*p %d boundary %d (%s)",*p,multipart->boundary[multipart->pos],multipart->boundary);
if (multipart->pos==0){
if (*p=='\n') // \r is optional.
multipart->startpos=multipart->pos=1;
else
multipart->startpos=0;
}
if (*p==multipart->boundary[multipart->pos]){
multipart->pos++;
if (multipart->pos==multipart->size){
multipart->startpos=multipart->pos=0;
data->pos++; // Not sure why this is needed. FIXME.
int w=write(multipart->fd,tmp,tmppos);
if (w!=tmppos){
ONION_ERROR("Error writing multipart data to file. Check permissions on temp directory, and availabe disk.");
close(multipart->fd);
return OCS_INTERNAL_ERROR;
}
close(multipart->fd);
req->parser=parse_POST_multipart_next;
return OCS_NEED_MORE_DATA;
}
}
else{
if (multipart->file_total_size>req->connection.listen_point->server->max_file_size){
ONION_ERROR("Files on this post too big. Aborting.");
return OCS_INTERNAL_ERROR;
}
if (multipart->pos!=0){
multipart->file_total_size+=multipart->pos;
int r=multipart->pos-multipart->startpos;
//ONION_DEBUG0("Write %d bytes",r);
int w=write(multipart->fd,tmp,tmppos);
if (w!=tmppos){
ONION_ERROR("Error writing multipart data to file. Check permissions on temp directory, and availabe disk.");
close(multipart->fd);
return OCS_INTERNAL_ERROR;
}
tmppos=0;
w=write(multipart->fd, multipart->boundary+multipart->startpos, r);
if (w!=r){
ONION_ERROR("Error writing multipart data to file. Check permissions on temp directory, and availabe disk.");
close(multipart->fd);
return OCS_INTERNAL_ERROR;
}
multipart->startpos=multipart->pos=0;
data->pos--; // Ignore read charater, try again. May be start of boundary.
continue;
}
//ONION_DEBUG0("Write 1 byte");
tmp[tmppos++]=*p;
if (tmppos==sizeof(tmp)){
int w=write(multipart->fd,tmp,tmppos);
if (w!=tmppos){
ONION_ERROR("Error writing multipart data to file. Check permissions on temp directory, and availabe disk.");
close(multipart->fd);
return OCS_INTERNAL_ERROR;
}
tmppos=0;
}
}
multipart->file_total_size++;
p++;
}
int w=write(multipart->fd,tmp,tmppos);
if (w!=tmppos){
ONION_ERROR("Error writing multipart data to file. Check permissions on temp directory, and availabe disk.");
close(multipart->fd);
return OCS_INTERNAL_ERROR;
}
return OCS_NEED_MORE_DATA;
}
/**
* Hard parser as I must set into the file as I read, until i found the boundary token (or start), try to parse, and if fail,
* write to the file.
*
* The boundary may be in two or N parts.
*/
static onion_connection_status parse_POST_multipart_data(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
onion_multipart_buffer *multipart=(onion_multipart_buffer*)token->extra;
const char *p=data->data+data->pos;
char *d=&multipart->data[token->pos];
for (;data->pos<data->size;data->pos++){
if (multipart->pos==0 && *p=='\n') // \r is optional.
multipart->pos=1;
if (*p==multipart->boundary[multipart->pos]){ // Check boundary
multipart->pos++;
if (multipart->pos==multipart->size){
multipart->pos=0;
data->pos++; // Not sure why this is needed. FIXME.
if (token->pos>0 && *(d-1)=='\n'){
token->pos--;
d--;
}
if (token->pos>0 && *(d-1)=='\r')
d--;
*d='\0';
ONION_DEBUG0("Adding POST data '%s'",multipart->name);
onion_dict_add(req->POST, multipart->name, multipart->data, 0);
multipart->data=multipart->data+token->pos+1;
token->pos=0;
req->parser=parse_POST_multipart_next;
return OCS_NEED_MORE_DATA;
}
}
else{ // No boundary
if (multipart->pos!=0){// Maybe i was checking before... so I need to copy as much as I wrongly thought I got.
if (multipart->post_total_size<=multipart->pos){
ONION_ERROR("No space left for this post.");
return OCS_INTERNAL_ERROR;
}
multipart->post_total_size-=multipart->pos;
memcpy(d,p,multipart->pos);
d+=multipart->pos;
token->pos+=multipart->pos;
multipart->pos=0;
}
if (multipart->post_total_size<=0){
ONION_ERROR("No space left for this post.");
return OCS_INTERNAL_ERROR;
}
multipart->post_total_size--;
//ONION_DEBUG0("%d %d",multipart->post_total_size,token->extra_size - (int) (d-token->extra));
*d=*p;
d++;
token->pos++;
}
p++;
}
return OCS_NEED_MORE_DATA;
}
static onion_connection_status parse_POST_multipart_headers_key(onion_request *req, onion_buffer *data);
static onion_connection_status parse_POST_multipart_content_type(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_until(token, data,';');
if (res<=1000)
return res;
//ONION_DEBUG("Got content type %s",token->str);
onion_multipart_buffer *multipart=(onion_multipart_buffer*)token->extra;
const char *name;
name=strstr(token->str, "filename=");
if (name){
int l=strlen(token->str)-9;
if (l>multipart->post_total_size){
ONION_ERROR("Post buffer exhausted. content-Length wrong passed.");
return OCS_INTERNAL_ERROR;
}
multipart->filename=multipart->data;
memcpy(multipart->filename, name+9, l);
multipart->filename[l]=0;
multipart->data=multipart->data+l+1;
if (*multipart->filename=='"' && multipart->filename[l-2]=='"'){
multipart->filename[l-2]='\0';
multipart->filename++;
}
ONION_DEBUG0("Set filename '%s'",multipart->filename);
}
else{
name=strstr(token->str, "name=");
if (name){
int l=strlen(token->str)-5;
if (l>multipart->post_total_size){
ONION_ERROR("Post buffer exhausted. Content-Length had wrong size.");
return OCS_INTERNAL_ERROR;
}
multipart->name=multipart->data;
memcpy(multipart->name, name+5, l);
multipart->name[l]=0;
if (*multipart->name=='"' && multipart->name[l-2]=='"'){
l-=2;
multipart->name[l]='\0';
multipart->name++;
}
multipart->data=multipart->name+l+1;
ONION_DEBUG0("Set field name '%s'",multipart->name);
}
}
if (res==STRING_NEW_LINE){
req->parser=parse_POST_multipart_headers_key;
return OCS_NEED_MORE_DATA;
}
return OCS_NEED_MORE_DATA;
}
static onion_connection_status parse_POST_multipart_ignore_header(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_LINE(token, data);
if (res<=1000)
return res;
token->pos=0;
req->parser=parse_POST_multipart_headers_key;
return OCS_NEED_MORE_DATA;
}
static onion_connection_status parse_POST_multipart_headers_key(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_KEY(token, data);
if (res<=1000)
return res;
if (res==NEW_LINE){
if (!req->POST)
req->POST=onion_dict_new();
ONION_DEBUG("New line");
onion_multipart_buffer *multipart=(onion_multipart_buffer*)token->extra;
multipart->pos=0;
if (multipart->filename){
char filename[]="/tmp/onion-XXXXXX";
multipart->fd=mkstemp(filename);
if (multipart->fd<0)
ONION_ERROR("Could not create temporal file at %s.", filename);
if (!req->FILES)
req->FILES=onion_dict_new();
onion_dict_add(req->POST,multipart->name,multipart->filename, 0);
onion_dict_add(req->FILES,multipart->name, filename, OD_DUP_VALUE);
ONION_DEBUG0("Created temporal file %s",filename);
req->parser=parse_POST_multipart_file;
return parse_POST_multipart_file(req, data);
}
else{
req->parser=parse_POST_multipart_data;
return parse_POST_multipart_data(req, data);
}
}
// Only interested in one header
if (strcmp(token->str,"Content-Disposition")==0){
req->parser=parse_POST_multipart_content_type;
return parse_POST_multipart_content_type(req,data);
}
ONION_DEBUG("Not interested in header '%s'",token->str);
req->parser=parse_POST_multipart_ignore_header;
return parse_POST_multipart_ignore_header(req,data);
}
static onion_connection_status parse_POST_multipart_next(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_MULTIPART_next(token, data);
if (res<=1000)
return res;
//ONION_DEBUG("Found next token: %d",res);
if (res==MULTIPART_END)
return onion_request_process(req);
onion_multipart_buffer *multipart=(onion_multipart_buffer*)token->extra;
multipart->filename=NULL;
multipart->name=NULL;
req->parser=parse_POST_multipart_headers_key;
return parse_POST_multipart_headers_key(req, data);
}
static onion_connection_status parse_POST_multipart_start(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_MULTIPART_BOUNDARY(token, data);
if (res<=1000)
return res;
req->parser=parse_POST_multipart_next;
return parse_POST_multipart_next(req, data);
}
static onion_connection_status parse_POST_urlencode(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_URLENCODE(token, data);
if (res<=1000)
return res;
req->POST=onion_dict_new();
onion_request_parse_query_to_dict(req->POST, token->extra);
token->extra=NULL; // At query to dict, it keeps the pointer and free it when POST is freed.
return onion_request_process(req);
}
static onion_connection_status parse_headers_KEY(onion_request *req, onion_buffer *data);
static onion_connection_status parse_headers_VALUE(onion_request *req, onion_buffer *data);
static onion_connection_status parse_headers_VALUE_skip_leading_whitespace(onion_request *req, onion_buffer *data){
while (data->pos<data->size){
char peek=data->data[data->pos++];
if (peek!='\t' && peek!=' '){
onion_token *token=req->parser_data;
if (peek=='\n')
return OCS_NEED_MORE_DATA;
token->str[token->pos++]=' ';
token->str[token->pos++]=peek;
req->parser=parse_headers_VALUE;
return parse_headers_VALUE(req, data);
}
}
return OCS_NEED_MORE_DATA;
}
static onion_connection_status parse_headers_VALUE_multiline_if_space(onion_request *req, onion_buffer *data){
char peek=token_peek_next_char(data);
ONION_DEBUG0("Multiline if space, peek %d",peek);
if (peek==0){
req->parser=parse_headers_VALUE_multiline_if_space;
return OCS_NEED_MORE_DATA;
}
if (peek==' ' || peek=='\t'){
req->parser=parse_headers_VALUE_skip_leading_whitespace;
return parse_headers_VALUE_skip_leading_whitespace(req, data);
}
onion_token *token=req->parser_data;
token->pos=0;
char *p=token->str; // skips leading spaces
while (isspace(*p)) p++;
ONION_DEBUG0("Adding header %s : %s",token->extra,p);
onion_dict_add(req->headers,token->extra,p, OD_DUP_VALUE|OD_FREE_KEY);
token->extra=NULL; // It will be freed at onion_dict_free.
req->parser=parse_headers_KEY;
return OCS_NEED_MORE_DATA; // Get back recursion if any, to prevent too long callstack (on long headers) and stack overflow.
}
static onion_connection_status parse_headers_VALUE(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_LINE(token, data);
if (res<=1000)
return res;
return parse_headers_VALUE_multiline_if_space(req, data);
}
static onion_connection_status parse_headers_KEY(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_KEY(token, data);
if (res<=1000)
return res;
ONION_DEBUG0("Got %d at KEY",res);
if ( res == NEW_LINE ){
if ((req->flags&OR_METHODS)==OR_POST){
const char *content_type=onion_request_get_header(req, "Content-Type");
if (content_type && (strstr(content_type,"application/x-www-form-urlencoded") || strstr(content_type, "boundary")))
return prepare_POST(req);
}
if ((req->flags&OR_METHODS)==OR_PUT)
return prepare_PUT(req);
if (onion_request_get_header(req, "Content-Length")){ // Some length, not POST, get data.
int n=atoi(onion_request_get_header(req, "Content-Length"));
if (n>0)
return prepare_CONTENT_LENGTH(req);
}
return onion_request_process(req);
}
assert(token->extra==NULL);
token->extra=onion_low_strdup(token->str);
req->parser=parse_headers_VALUE;
return parse_headers_VALUE(req, data);
}
static onion_connection_status parse_headers_KEY_skip_NL(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int r=token_read_NEW_LINE(token,data);
if (r<=1000)
return r;
req->parser=parse_headers_KEY;
return parse_headers_KEY(req, data);
}
static onion_connection_status parse_headers_VERSION(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_STRING(token, data);
if (res<=1000)
return res;
if (strcmp(token->str,"HTTP/1.1")==0)
req->flags|=OR_HTTP11;
if (!req->GET)
req->GET=onion_dict_new();
if (res==STRING){
req->parser=parse_headers_KEY_skip_NL;
return parse_headers_KEY_skip_NL(req, data);
}
else{ // STRING_NEW_LINE, only when \n as STRING separator, not \r\n.
req->parser=parse_headers_KEY;
return parse_headers_KEY(req, data);
}
}
static onion_connection_status parse_headers_URL(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_STRING(token, data);
if (res<=1000)
return res;
req->fullpath=onion_low_strdup(token->str);
onion_request_parse_query(req);
ONION_DEBUG0("URL path is %s", req->fullpath);
if (res==STRING_NEW_LINE){ // Old HTTP/0? or lazy user, dont set the HTTP version, straigth new line.
req->parser=parse_headers_KEY;
return parse_headers_KEY(req, data);
}
else{
req->parser=parse_headers_VERSION;
return parse_headers_VERSION(req, data);
}
}
static onion_connection_status parse_headers_GET(onion_request *req, onion_buffer *data){
onion_token *token=req->parser_data;
int res=token_read_STRING(token, data);
if (res<=1000)
return res;
int i;
for (i=0;i<16;i++){
if (!onion_request_methods[i]){
ONION_ERROR("Unknown method '%s' (%d known methods)",token->str, i);
return OCS_NOT_IMPLEMENTED;
}
if (strcmp(onion_request_methods[i], token->str)==0){
ONION_DEBUG0("Method is %s", token->str);
req->flags=(req->flags&~0x0F)+i;
break;
}
}
req->parser=parse_headers_URL;
return parse_headers_URL(req, data);
}
/**
* @short Write some data into the request, and passes it line by line to onion_request_fill
*
* It features a state machine, from req->parse_state.
*
* Depending on the state input is redirected to a diferent parser, one for headers, POST url encoded data...
*
* @return Returns the number of bytes writen, or <=0 if connection should close, according to onion_connection_status
* @see onion_connection_status
*/
onion_connection_status onion_request_write(onion_request *req, const char *data, size_t size){
if (!req->parser_data){
req->parser_data=onion_low_calloc(1, sizeof(onion_token));
req->parser=parse_headers_GET;
}
onion_connection_status (*parse)(onion_request *req, onion_buffer *data);
parse=req->parser;
if (parse){
onion_buffer odata={ data, size, 0};
while (odata.size>odata.pos){
int r=parse(req, &odata);
if (r!=OCS_NEED_MORE_DATA){
return r;
}
parse=req->parser;
}
return OCS_NEED_MORE_DATA;
}
return OCS_INTERNAL_ERROR;
}
/**
* @short Parses the query to unquote the path and get the query.
*/
static int onion_request_parse_query(onion_request *req){
if (!req->fullpath)
return 0;
if (req->GET) // already done
return 1;
char *p=req->fullpath;
char have_query=0;
while(*p){
if (*p=='?'){
have_query=1;
break;
}
p++;
}
*p='\0';
onion_unquote_inplace(req->fullpath);
if (have_query){ // There are querys.
p++;
req->GET=onion_dict_new();
onion_request_parse_query_to_dict(req->GET, p);
}
return 1;
}
/**
* @short Parses the query part to a given dictionary.
*
* The data is overwriten as necessary. It is NOT dupped, so if you free this char *p, please free the tree too.
*/
static void onion_request_parse_query_to_dict(onion_dict *dict, char *p){
ONION_DEBUG0("Query to dict %s",p);
char *key=NULL, *value=NULL;
int state=0; // 0 key, 1 value
key=p;
while(*p){
if (state==0){
if (*p=='='){
*p='\0';
value=p+1;
state=1;
}
else if (*p=='&'){
*p='\0';
onion_unquote_inplace(key);
ONION_DEBUG0("Adding key %s",key);
onion_dict_add(dict, key, "", 0);
key=p+1;
state=0;
}
}
else{
if (*p=='&'){
*p='\0';
onion_unquote_inplace(key);
onion_unquote_inplace(value);
ONION_DEBUG0("Adding key %s=%-16s",key,value);
onion_dict_add(dict, key, value, 0);
key=p+1;
state=0;
}
}
p++;
}
if (state==0){
if (key[0]!='\0'){
onion_unquote_inplace(key);
ONION_DEBUG0("Adding key %s",key);
onion_dict_add(dict, key, "", 0);
}
}
else{
onion_unquote_inplace(key);
onion_unquote_inplace(value);
ONION_DEBUG0("Adding key %s=%-16s",key,value);
onion_dict_add(dict, key, value, 0);
}
}
/**
* @short Prepares the POST
*/
static onion_connection_status prepare_POST(onion_request *req){
// ok post
onion_token *token=req->parser_data;
const char *content_type=onion_dict_get(req->headers, "Content-Type");
const char *content_size=onion_dict_get(req->headers, "Content-Length");
if (!content_size){
ONION_ERROR("I need the content size header to support POST data");
return OCS_INTERNAL_ERROR;
}
size_t cl=atol(content_size);
if (cl==0)
return onion_request_process(req);
//ONION_DEBUG("Content type %s",content_type);
if (!content_type || (strstr(content_type, "application/x-www-form-urlencoded"))){
if (cl>req->connection.listen_point->server->max_post_size){
ONION_ERROR("Asked to send much POST data. Limit %d. Failing.",req->connection.listen_point->server->max_post_size);
return OCS_INTERNAL_ERROR;
}
assert(token->extra==NULL);
token->extra=onion_low_scalar_malloc(cl+1); // Cl + \0
token->extra_size=cl;
req->free_list=onion_ptr_list_add(req->free_list, token->extra); // Free when the request is freed.
req->parser=parse_POST_urlencode;
return OCS_NEED_MORE_DATA;
}
// multipart.
const char *mp_token=strstr(content_type, "boundary=");
if (!mp_token){
ONION_ERROR("No boundary set at content-type");
return OCS_INTERNAL_ERROR;
}
mp_token+=9;
if (cl>req->connection.listen_point->server->max_post_size) // I hope the missing part is files, else error later.
cl=req->connection.listen_point->server->max_post_size;
int mp_token_size=strlen(mp_token);
token->extra_size=cl; // Max size of the multipart->data
onion_multipart_buffer *multipart=onion_low_malloc(token->extra_size+sizeof(onion_multipart_buffer)+mp_token_size+2);
assert(token->extra==NULL);
token->extra=(char*)multipart;
multipart->boundary=(char*)multipart+sizeof(onion_multipart_buffer)+1;
multipart->size=mp_token_size+4;
multipart->pos=2; // First boundary already have [\r]\n readen
multipart->post_total_size=cl;
multipart->file_total_size=0;
multipart->boundary[0]='\r';
multipart->boundary[1]='\n';
multipart->boundary[2]='-';
multipart->boundary[3]='-';
strcpy(&multipart->boundary[4],mp_token);
multipart->data=(char*)multipart+sizeof(onion_multipart_buffer)+multipart->size+1;
//ONION_DEBUG("Multipart POST boundary '%s'",multipart->boundary);
req->parser=parse_POST_multipart_start;
return OCS_NEED_MORE_DATA;
}
/**
* @short Prepares the CONTENT LENGTH
*/
static onion_connection_status prepare_CONTENT_LENGTH(onion_request *req){
onion_token *token=req->parser_data;
const char *content_size=onion_dict_get(req->headers, "Content-Length");
if (!content_size){
ONION_ERROR("I need the Content-Length header to get data");
return OCS_INTERNAL_ERROR;
}
size_t cl=atol(content_size);
if (cl>req->connection.listen_point->server->max_post_size){
ONION_ERROR("Trying to set more data at server than allowed %d", req->connection.listen_point->server->max_post_size);
return OCS_INTERNAL_ERROR;
}
req->data=onion_block_new();
assert(token->extra==NULL);
//token->extra=NULL; // Should be already null, as should have no data.
token->extra_size=cl;
token->pos=0;
req->parser=parse_CONTENT_LENGTH;
return OCS_NEED_MORE_DATA;
}
/**
* @short Prepares the PUT
*
* It saves the data to a temporal file, which name is stored at data.
*/
static onion_connection_status prepare_PUT(onion_request *req){
onion_token *token=req->parser_data;
const char *content_size=onion_dict_get(req->headers, "Content-Length");
if (!content_size){
ONION_ERROR("I need the Content-Length header to get data");
return OCS_INTERNAL_ERROR;
}
size_t cl=atol(content_size);
if (cl>req->connection.listen_point->server->max_file_size){
ONION_ERROR("Trying to PUT a file bigger than allowed size");
return OCS_INTERNAL_ERROR;
}
req->data=onion_block_new();
char filename[]="/tmp/onion-XXXXXX";
int fd=mkstemp(filename);
if (fd<0)
ONION_ERROR("Could not create temporary file at %s.", filename);
onion_block_add_str(req->data, filename);
ONION_DEBUG0("Creating PUT file %s (%d bytes long)", filename, token->extra_size);
if (!req->FILES){
req->FILES=onion_dict_new();
}
{
const char *filename=onion_block_data(req->data);
onion_dict_add(req->FILES,"filename", filename, 0);
}
if (cl==0){
ONION_DEBUG0("Created 0 length file");
close(fd);
return onion_request_process(req);
}
int *pfd=onion_low_scalar_malloc(sizeof(fd));
*pfd=fd;
assert(token->extra==NULL);
token->extra=(char*)pfd;
token->extra_size=cl;
token->pos=0;
req->parser=parse_PUT;
return OCS_NEED_MORE_DATA;
}
/**
* @short Frees the parser data.
*/
void onion_request_parser_data_free(void *t){
ONION_DEBUG0("Free parser data");
onion_token *token=t;
if (token->extra){
onion_low_free(token->extra);
token->extra=NULL;
}
onion_low_free(token);
}
Jump to Line
Something went wrong with that request. Please try again.