ECFS is an extension to the existing ELF core file format in Linux. Its job is to intercept the Linux core-dump handler, catch the core-dump before it is written to disk, and carefully reconstruct it into an ecfs-core file. An ecfs-core file is backwards compatible with regular core files but has been extended in such a way that they boast prolific amounts of data useful for process forensics analysis. An ecfs-file is not limited to just ELF program headers, but also contains many section headers as well as fully reconstructed relocation and symbol tables that reflect the state of code and data in runtime. ecfs-core files are also extremely straight forward to parse, moreso when using the complementary libecfs C library (Python bindings are a work in progress). See the manual page which describes how to access the different ecfs-core components in-depth [./manual/man.txt]
ECFS can also be used to take snapshots without killing the process which is often times more desirable. This can be accomplished by using the ecfsnap64 and ecfsnap32 utility (main/ecfs_snapshot.c) but at this point the utility is not quite working 100%. It relies on gcore (GDB based coredumping utility) and there are some aspects of the core files generated with this utility that are throwing off certain components of ecfs and causing failure. This will for certain be up to date and working by 4/15/2015
- Runtime malware analysis
- Process forensics
- Reversing binary protection
- Process snapshot execution (https://github.com/elfmaster/ecfs_exec)
ECFS creates high resolution snapshots of a running process. These snapshots are ideal for forensics analysis, and malware analysis. The libecfsreader API can be used to rapidly design advanced malware analysis solutions for automated detection of threats within a process. A brief example of this capability is demonstrated in POC||GTFO 0x7 https://www.alchemistowl.org/pocorgtfo/pocorgtfo07.pdf
IDA users will be happy to know that it is now possible to snapshot an entire process and view the entire thing in IDA. It is important to note that IDA does not understand that the shared library functions are included in the actual ECFS file so it doesn't make the connection reference between the calls to shared library functions and their actual location within the file. It would be worth writing a parser script for IDA to make this happen so you can click on a PLT call and have it follow the GOT pointer right to the shared library function.
In addition to the malware analysis implications of ECFS, there is an entirely different use case as well; ecfs-core snapshots can be re-executed... that's right, you can snapshot a process and re-run it later in time. Currently only a simple prototype exists for this that I wrote up in about 4 hours: https://github.com/elfmaster/ecfs_exec
Enjoy...
This source code contains the ecfs suite of software which has several components. The primary component plugs ecfs-core dump functionality into the Linux core dump routine. The secondary component being libecfs which is a library specifically designed to parse ecfs-core files with ease.
On 64bit Linux systems this software is also cross compiled for 32bit so that a 32bit-handler can be dispatched in order to handle 32bit processes. This requires that both 32bit and 64bit versions of libdwarf and libelf be installed on your system simultaneously. On ubuntu systems libelf-dev package allows for this but libdwarf package does not. To get around this you may link against the deps/libdwarf32.a and deps/libdwarf64.a files contained within.
Once compiled and installed you will notice that /opt/ecfs/bin contains these 3 files:
/opt/ecfs/bin/ecfs_handler
/opt/ecfs/bin/ecfs32
/opt/ecfs/bin/ecfs64
The ecfs_handler is more or less a stub program that detects whether a process is 32bit or 64bit, and then invokes the appropriate ecfs worker (ecfs32 or ecfs 64). ecfs-core files will be dumped in /opt/ecfs/cores and typically take up more space (especially if -t option is being used) than regular core files. The main/readecfs.c utility is the equiavalent of what readelf is for regular ELF files. readecfs parses and prints the details of an ecfs-core file to stdout. This utility uses libecfs as an example of how to use its functions.
In many cases it is not desirable to kill the program that you need to snapshot. For this you may use ecfs_snapshot (Installed into /usr/bin) as ecfsnap64 and ecfsnap32 respectively. This makes use of the /usr/bin/gcore script which is included with the GDB package. gcore creates a corefile of a running process without killing it, and ecfs will use gcore to create the bare bones core file and then build on top of that.
NOTE: ecfs will not transform a core file into an ecfs snapshot unless the process is still running in memory. In other words you cannot simply pass a core file to ecfs an expect it to create an ecfs-core file as an output because it relies heavily on what resides in /proc/$pid during the transformation phase.
$ ecfsnap64 -t -p pidof sshd -o sshd.ecfs
The following are instructions on building and installing the ecfs-core dump software suite. Which includes not only the ecfs-core functionality itself, but also libecfs (For developers to parse ecfs-core files) and the readecfs utility (Similar to readelf).
(Note: deps/libdwarf32.a and deps/libdwarf64.a can be used)
apt-get install libelf-dev libelf-dev:i386 libdwarf-dev
To build and install ecfs type:
sudo ./setup.sh
sudo make install
Which is the equivalent to:
make V=prod B=32
make V=prod B=64
sudo make install
The 'make install' will put custom line into /proc/sys/kernel/core_pattern that tells it to invoke /opt/ecfs/bin/ecfs_handler during a coredump.
!!! FOR UBUNTU USERS (AND POSSIBLY OTHER DISTROS) !!! For Ubuntu the libdwarf package will not allow you to install it for both 32bit and 64bit at the same time. To get around that simply follow these instructions for building and installation.
sudo cp deps/libdwarf32.a /usr/lib make V=prod B=32 sudo cp deps/libdwarf64.a /usr/lib make V=prod B=64 sudo make install
For the 32bit systems we don't need ecfs_handler (Which invokes either a 32bit or 64bit ./ecfs worker process). Only 32bit processes will be handled so we can use the ./ecfs binary directly (Without ecfs_handler). Follow these instructions:
Dependencies:
apt-get install libelf-dev libdwarf-dev
To build and install ecfs type:
sudo ./setup32.sh
This argument (Should generally be on) will direct the ecfs-core worker to write the entire text segment of each shared library (vs. just the first 4k). ecfs ALWAYS writes the entire text segment of the executable but limits the text of each shlib to only 4k unless you specify -t. Which means opts.text_all is set in ecfs.c. Having -t set will cause ecfs to take much longer as it has to write out sometimes hundreds of megabytes of code segments.
ecfs can perform heuristics that do things such as mark shared libraries as being DLL injected. Sometimes false positives can arise. If you want to use the heuristics feature use the -h switch. As of 3/11/2015 there are some bugs that need to be worked out in it, this README will change once they are fixed, meanwhile I would recommend not using -h.
Everytime you reboot you must run ./setup_core_pattern.sh to enable ecfs. Unless you have setup ecfs to be permanent in your sysctl.conf file or 50-coredump.conf file as discussed below. On ubuntu systems the 'apport' crash collector overrides the sysctl.conf with an init script, and a proper install packge to override this behavior has not yet been created for ecfs so it may be necessary to run setup_core_pattern.sh in ubuntu after every reboot. You may check /proc/sys/kernel/core_pattern file to see ecfs has been enabled.
To modify your sysctl.conf (For systems not using systemd)
echo 'kernel.core_pattern=|/opt/ecfs/bin/ecfs_handler -t -e %e -p %p -o /opt/ecfs/cores/%e.%p' >> /etc/sysctl.conf
For systemd based systems such as ARCH Linux
Modify:
/usr/lib/sysctl.d/50-coredump.conf
ecfs-core files will be dumped into /opt/ecfs/cores. These files take up more space than a traditional core file, especially if you are using the -t option which captures the entire text image of each loaded library. It may be desirable to write a script that periodically compresses these files. The files are named according to the executable that crashed and the pid like firefox.7737 if firefox pid 7737 were to crash, which makes them easy to identify.
To read these files you may use any tools that you would use to analyze a regular ELF file with a few Caveats. They are backwards compatible with regular core files but in order to analyze them with GDB you must flip the e_type bit in the ELF file header from ET_NONE to ET_CORE. The utility bin/et_flip will flip the bit to ET_CORE, and if you run it a second time on the file it will flip it back to ET_NONE. The reason ecfs-core files are of type ET_NONE is because it allows the objdump utility to utilize the section headers. If objdump sees that a file is of type ET_CORE it will assume it has no section headers and use the program headers which aren't nearly as useful.
Example:
# et_flip sshd.26099
# gdb -q /usr/sbin/sshd sshd.26099
Reading symbols from /usr/sbin/sshd...(no debugging symbols found)...done.
[New LWP 26099]
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Core was generated by `/usr/sbin/sshd -D'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x00007fcf37bbad83 in __select_nocancel () at ../sysdeps/unix/syscall-template.S:81
81 ../sysdeps/unix/syscall-template.S: No such file or directory.
(gdb)
Now lets flip it back before using with objdump
# et_flip sshd.26099
# objdump -d sshd.26099 | less
sshd.26099: file format elf64-x86-64
Disassembly of section .init:
00007fcf39c86c10 <.init>:
7fcf39c86c10: 48 83 ec 08 sub $0x8,%rsp
7fcf39c86c14: 48 8b 05 65 e3 2a 00 mov 0x2ae365(%rip),%rax # 7fcf39f34f80 <sub_7fcf39c88850-0x7fcf399ce8d0>
7fcf39c86c1b: 48 85 c0 test %rax,%rax
7fcf39c86c1e: 74 05 je 7fcf39c86c25 <sub_7fcf39c88850-0x7fcf39c7cc2b>
7fcf39c86c20: e8 bb 03 00 00 callq 7fcf39c86fe0 <sub_7fcf39c88850-0x7fcf39c7c870>
7fcf39c86c25: 48 83 c4 08 add $0x8,%rsp
7fcf39c86c29: c3 retq
The 'readecfs' utility is very handy very reading ecfs files and was specifically designed for this purpose. It is written using the libecfs API that also comes with this software-suite. For reading 64bit ecfs files use bin/prod/64/readecfs and use bin/prod/32/readecfs for 32bit.
The readecfs has many command line options, but -e will show most of the ecfs s pecific stuff.
# readecfs -e host.25527
- readecfs output for file cores/host.7628
- Executable path (.exepath): /home/ryan/git/ecfs/ecfs_tests/host
- Personality
dynamically linked: yes
compiled as PIE: no
local symtab reconstruction: no
malware heuristics: no
original bin had stripped section headers: no
- Thread count (.prstatus): 1
- Thread info (.prstatus)
[thread[1] pid: 7628
- Register values
r15: 0
r14: 0
r13: 7fff66b1bf40
r12: 7fff66b1bae0
rbp: ffffffff
rbx: 7fff66b1ba60
r11: 246
r10: 8
r9: 0
r8: 7fff66b1bb60
rax: fffffffffffffdfc
rcx: ffffffffffffffff
rdx: 0
rsi: 7fff66b1ba50
rdi: 7fff66b1ba50
rip: 7f3821a19f20
rsp: 7fff66b1ba48
cs: 33
ss: 2b
ds: 0
es: 0
fs: 0
gs: 0
eflags: 246
- Exited on signal (.siginfo): 11
- files/pipes/sockets (.fdinfo):
[fd: 0] path: /dev/pts/8
[fd: 1] path: /dev/pts/8
[fd: 2] path: /dev/pts/8
[fd: 3] path: /etc/passwd
[fd: 4] path: /tmp/passwd_info
- Printing shared library mappings:
shlib: libc-2.19.so.text
shlib: libc-2.19.so.undef
shlib: libc-2.19.so.relro
shlib: libc-2.19.so.data.0
shlib: ld-2.19.so.text
shlib: ld-2.19.so.relro
shlib: ld-2.19.so.data.1
- Dynamic Symbol section -
.dynsym: - 7f3821959000
.dynsym: fputs - 7f38219c7730
.dynsym: __libc_start_main - 7f382197add0
.dynsym: fgets - 7f38219c7220
.dynsym: __gmon_start__ - 0
.dynsym: fopen - 7f38219c74e0
.dynsym: sleep - 7f3821a19d00
- Symbol Table section -
.symtab: sub_4004b0 - 4004b0
.symtab: sub_400520 - 400520
.symtab: sub_40060d - 40060d
.symtab: sub_4006b0 - 4006b0
.symtab: sub_400720 - 400720
- Printing out GOT/PLT characteristics (pltgot_info_t):
gotsite gotvalue gotshlib pltval
0x601018 0x7f38219c7730 0x7f38219c7730 0x4004c6
0x601020 0x7f382197add0 0x7f382197add0 0x4004d6
0x601028 0x7f38219c7220 0x7f38219c7220 0x4004e6
0x601030 0x4004f6 0x0 0x4004f6
0x601038 0x7f38219c74e0 0x7f38219c74e0 0x400506
0x601040 0x7f3821a19d00 0x7f3821a19d00 0x400516
- Printing auxiliary vector (.auxiliary):
AT_PAGESZ: 0x1000
AT_PHDR: 0x400040
AT_PHENT: 0x38
AT_PHNUM: 0x9
AT_BASE: 0x7f3821d1e000
AT_FLAGS: 0x0
AT_ENTRY: 0x400520
AT_UID: 0x0
AT_EUID: 0x0
AT_GID: 0x0
Debug output is logged using syslog() and can be found in either /var/log/syslog or on some systems such as arch Linux you will need to look at 'journalctl -b'
elfmaster[at]zoho.com