This provides experimental Linux ELF userland snapshotting support based on previous work by Kasimir and scripts from Snapchange.
Move into the linux_mode/qemu_snapshot directory and run setup.sh:
user@pc:/wtf/linux_mode/qemu_snapshot$ ./setup.shThis script installs all pre-requisite tools, compiles qemu, and builds a target virtual machine consisting of a Linux kernel and disk image.
Create a subdirectory in linux_mode for your snapshot and create a bkpt.py file, like linux_mode/crash_test/bkpt.py:
# imports
import sys, os
# import fuzzing breakpoint
from gdb_fuzzbkpt import *
target_dir = 'linux_crash_test'
# address to break on, found using gdb
break_address = 'do_crash_test'
# name of the file in which to break
file_name = 'a.out'
# create the breakpoint for the executable specified
FuzzBkpt(target_dir, break_address, file_name, sym_path=file_name)target_dir- subdirectory intargetsto save the snapshot databreak_address- address to break on. This can be a hardcoded address or a symbol ifsym_pathis providedfile_name- name of the file in the target VM associated with the breakpointsym_path- optional argument if you'd like symbols to be loaded
Start the virtual machine in one tab while in the snapshot subdirectory by running ../qemu_snapshot/gdb_server.sh:
user@pc:/wtf/linux_mode/crash_test$ ../qemu_snapshot/gdb_server.shIn a separate tab, scp the target file to the target VM. With crash_test this can be done by first compiling the target file:
user@pc:/wtf/linux_mode/crash_test$ gcc test.cThen transfer the target file to the VM:
user@pc:/wtf/linux_mode/crash_test$ pushd ../qemu_snapshot/target_vm
user@pc:/wtf/linux_mode/qemu_snapshot/target_vm$ ./scp.sh ../../crash_test/a.out
a.out 100% 16KB 1.2MB/s 00:00Go back to the crash_test directory.
user@pc:/wtf/linux_mode/qemu_snapshot/target_vm$ popd
/wtf/linux_mode/crash_test
user@pc:/wtf/linux_mode/crash_test$Now, run ../qemu_snapshot/gdb_client.sh:
user@pc:/wtf/linux_mode/crash_test$ ../qemu_snapshot/gdb_client.sh In the first tab, log in to the Linux machine (user root) and run the target file:
linux login: root
Linux linux 6.7.0-rc3 #1 SMP PREEMPT_DYNAMIC Thu Nov 30 18:38:29 UTC 2023 x86_64
The programs included with the Debian GNU/Linux system are free software;
the exact distribution terms for each program are described in the
individual files in /usr/share/doc/*/copyright.
Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
permitted by applicable law.
A valid context for root could not be obtained.
Last login: Fri Dec 1 21:21:22 UTC 2023 on ttyS0
root@linux:~# ./a.out
Enter some input.
dOnce the breakpoint is hit, the second tab will start the snapshotting process:
Continuing.
In right process? True
Calling mlockall
Saving 67 bytes at 555555555146
In right process? True
Restoring 67 bytes at 0x555555555146
Restored
In the Qemu tab, press Ctrl+C, run the `cpu` commandOnce the second tab indicates to run the cpu command, press Ctrl+C and run the cpu command from the first tab:
Thread 1 "qemu-system-x86" received signal SIGINT, Interrupt.
0x00007ffff77a4ebe in __ppoll (fds=0x5555568337d0, nfds=8, timeout=<optimized out>, timeout@entry=0x7fffffffdea0, sigmask=si
42 ../sysdeps/unix/sysv/linux/ppoll.c: No such file or directory.
(gdb) cpu
cpu_state: 0x55555681e240
done...continuing debuggeeThe second tab will detect once the first tab has finished executing the cpu command and continue creating a snapshot for the target VM.
Once the second tab indicates that snapshotting is complete, the target VM can be terminated.
In the Qemu tab, press Ctrl+C, run the `cpu` command
Detected cpu registers dumped to regs.json
Connecting to Qemu monitor at localhost:55555
Connected
Instructing Qemu to dump physical memory to file raw
Done
Converting raw file raw to dump file /wtf/targets/linux_crash_test/state/mem.dmp
Done
mv regs.json /wtf/targets/linux_crash_test/state/regs.json
mv symbol-store.json /wtf/targets/linux_crash_test/state/symbol-store.json
Snapshotting complete
Breakpoint 1, 0x0000555555555189 in do_crash_test ()
(gdb)Writing harnesses is the same process as writing harnesses for Windows executables. Example harnesses for crash_test and page_fault_test are present in src/wtf/fuzzer_linux_crash_test.cc and src/wtf/fuzzer_linux_page_fault_test.cc.
Now that we have everything set up we can start our server and fuzzer:
Provide a seed input:
user@pc:/wtf/targets/linux_crash_test$ echo a>inputs/aRun the master:
user@pc:/wtf/targets/linux_crash_test$ ../../src/build/wtf master --name linux_crash_test --max_len=10Run the fuzzee and note that crashes are found quickly.
user@pc:/wtf/targets/linux_crash_test$ ../../src/build/wtf fuzz --backend=bochscpu --name linux_crash_test
Setting @fptw to 0xff'ff.
The debugger instance is loaded with 16 items
Setting debug register status to zero.
Setting debug register status to zero.
Setting mxcsr_mask to 0xffbf.
Dialing to tcp://localhost:31337/..
#113174 cov: 47 exec/s: 11.3k lastcov: 2.0s crash: 1782 timeout: 0 cr3: 0 uptime: 10.0sTo fuzz with KVM, create a coverage breakpoints file by loading the target file in IDA and running scripts/gen_linux_coveragefile_ida.py. Transfer the coverage breakpoints file to the coverage subfolder in the target's directory. For example, for linux_crash_test transfer the coverage breakpoint file to targets/linux_crash_test/coverage/a.cov. Once transferred, KVM can be used for fuzzing:
user@pc:/wtf/targets/linux_crash_test$ sudo ../../src/build/wtf fuzz --backend=kvm --name linux_crash_test
Setting @fptw to 0xff'ff.
The debugger instance is loaded with 16 items
Parsing coverage/a.cov..
Applied 44 code coverage breakpoints
Setting debug register status to zero.
Setting debug register status to zero.
Setting mxcsr_mask to 0xffbf.
Resolved breakpoint 0xffffffff82001240 at GPA 0x2001240 aka HVA 0x564428d2afe0
Resolved breakpoint 0xffffffff82000ff0 at GPA 0x2000ff0 aka HVA 0x564428d2cda0
Resolved breakpoint 0xffffffff81099dc0 at GPA 0x1099dc0 aka HVA 0x564428d2db80
Resolved breakpoint 0xffffffff810708e0 at GPA 0x10708e0 aka HVA 0x564428d2e6b0
Resolved breakpoint 0x5555555551e7 at GPA 0x972c1e7 aka HVA 0x564428d32117
Dialing to tcp://localhost:31337/..
#24348 cov: 8 exec/s: 2.4k lastcov: 3.0s crash: 871 timeout: 0 cr3: 0 uptime: 10.0sThe only current way to symbolize and debug your testcases is to use the bochscpu backend and generate a Tenet traces as per Generating Tenet traces.