Vendor Description
“ZTE Corporation is a global leader in telecommunications and information technology. Founded in 1985 and listed on both the Hong Kong and Shenzhen Stock Exchanges, the company has been committed to providing integrated end-to-end innovations to deliver excellence and value to consumers, carriers, businesses and public sector customers from over 160 countries around the world to enable increased connectivity and productivity.”
Source: https://www.zte.com.cn/global/about/corporate_information
Business Recommandation
The vendor recommends to change the hardware and use a newer product. SEC Consult recommends to remove the device from productive environments.
Vulnerability Overview / Description
1) Hard-coded Administrator Password
The hard-coded administrator password was found in the ZTE mobile hotspot MS910S firmware version “CN_EUV1.00.01”, which is available at
http://devicedownload.zte.com.cn/support/product/201701161506340/soft/20170116151106465.zip
2) Known BusyBox Vulnerabilities
The used BusyBox toolkit in version 1.15.0 is outdated and contains multiple known vulnerabilities. The outdated version was found by IoT Inspector. One of the discovered vulnerabilities (CVE-2017-16544) was verified by using the MEDUSA scalable firmware run-time.
3) Known Backdoor in GoAhead Webserver
An unusual “telnetd” port was identified on an emulated device which led to the assumption that a backdoor can be opened via the GoAhead web-server. This conclusion was done because of a blog post from another researcher. By partially reverse engineering the binaries of the GoAhead webserver, the functionality described in the corresponding blog post can be underpinned.
Proof Of Concept
1) Hard-coded Administrator Password
The firmware file (ZTE_MF910SV1.0.1B09.bin) is using the JFFS2 filesystem which was extracted. The hardcoded password can be found in the /etc/shadow file within the firmware:
/_DL_MF910S_CN_EUV1.00.01.exe.extracted/Data/version/ _ZTE_MF910SV1.0.1B09.bin.extracted/jffs2-root/fs_1/etc/shadow
The file content is shown below:
root:$1$$zdlNHiCDxYDfeF4MZL.H3/:10933:0:99999:7::: Admin:$1$$zdlNHiCDxYDfeF4MZL.H3/:10933:0:99999:7::: bin::10933:0:99999:7::: daemon::10933:0:99999:7::: adm::10933:0:99999:7::: lp:*:10933:0:99999:7::: sync:*:10933:0:99999:7::: shutdown:*:10933:0:99999:7::: halt:*:10933:0:99999:7::: uucp:*:10933:0:99999:7::: operator:*:10933:0:99999:7::: nobody::10933:0:99999:7::: ap71::10933:0:99999:7:::
Both the user “root” and “admin” are using the same weak hardcoded password 5up.
2) Known BusyBox Vulnerabilities
BusyBox version 1.12.0 contains multiple CVEs like: CVE-2013-1813, CVE-2016-2148, CVE-2016-6301, CVE-2011-2716, CVE-2011-5325, CVE-2015-9261, CVE-2016-2147 and more. The BusyBox shell autocompletion vulnerability (CVE-2017-16544) was verified on an emulated device. A file with the name “\ectest\n\e]55;test.txt\a” was crea- ted to trigger the vulnerability.
# ls "pressing " test ]55;test.txt #
3) Known Backdoor in GoAhead Webserver
Starting the telnet daemon on the emulated device leads to a listener on a very unusual port:
[...] # telnetd #netstat -tulen Active Internet connections (only servers) Proto Recv-Q Send-Q Local Address Foreign Address State tcp 0 0 0.0.0.0:4719 0.0.0.0:* LISTEN [...]
Because this seems to be not configured on the system by any file, the BusyBox binary was inspected. The pseudocode snippet of BusyBox’ telnetd function that was generated by Hex- Rays ARM Decompiler, indicated that this port was hard-coded:
<snip> dword_788DC = (int)"/bin/login"; dword_788E0 = (int)"/etc/issue.net"; v3 = sub_5DBFC(a2, "f:l:Kip:b:F", &dword_788E0, &dword_788DC, &v76, &v75); v4 = v3 & 8; v5 = v3; if ( !(v3 & 8) && !(v3 & 0x40) ) sub_64A10(v3 & 8); if ( (v5 & 0x48) != 64 ) { openlog((const char *)dword_798DC, 1, 24); dword_78630 = 2; } if ( v5 & 0x10 ) v6 = sub_64FF8(v76); else v6 = 4719; <------------------------------------- Port "4719" if ( !v4 ) { v2 = sub_65578(v75, v6); sub_E394(v2, 1); sub_D9B0(v2); goto LABEL_13; } dword_788D8 = (int)sub_1C480(0); if ( dword_788D8 ) { <snip>
This led to the assumption that the GoAhead webserver was modified like described in the following blog post: http://blog.asiantuntijakaveri.fi/2017/03/backdoor-and-root-shell-on-zte-mf286.html
Inspecting the GoAhead webserver binary reinforces this assumption. The pseudocode was generated with Hex-Rays ARM Decompiler, like for the prior BusyBox binary:
<snip> int __fastcall sub_21D48(int a1) { int v1; // r5 const char *v2; // r4 v1 = a1; cfg_get("debug_level"); v2 = (const char *)sub_17DF0(v1, "change_mode", ""); cfg_set("login_9527", "1"); if ( !strcmp("1", v2) ) { cfg_set("change_mode", "1"); cfg_get("debug_level"); system("mode_change 1"); } else if ( !strcmp("2", v2) ) <--- change mode "2" { cfg_get("debug_level"); system("telnetd &"); <--------- telnet daemon started } else if ( !strcmp("3", v2) ) { cfg_get("debug_level"); system("rem_start.sh &"); } else if ( !strcmp("4", v2) ) { cfg_get("debug_level"); system("rem_kill.sh &"); } else { cfg_set("change_mode", "0"); cfg_get("debug_level"); system("mode_change 0"); } return sub_34374(v1, "success"); } <snip>
Other scripts could also be started via the webserver, like “rem_start.sh”. This script contains the following lines:
#!/bin/sh if ps|grep remserial then echo "remserial is running.." else remserial -p 10005 -r 192.168.1.10 -s "115200 raw" /dev/ttyUSB0 & fi
That means, that a serial console with the speed of 115200 Baud on port 10005 is started.
Vulnerable / Tested Versions
The following firmware has been tested, which was the latest version available during the time of the test:
- DL_MF910S_CN_EUV1.00.01