Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

x86-64: Preserve RCX and R11 when calling mprotect_asm (syscall) #673

Merged
merged 1 commit into from
Jul 6, 2021

Conversation

ammarfaizi2
Copy link
Contributor

x86-64: Preserve RCX and R11 when calling mprotect_asm (syscall)

x86-64: Preserve RCX and R11 when calling mprotect_asm (syscall)

On x86-64, the syscall instruction will clobber RCX and R11. It uses
RCX to save the return address and uses R11 to save the RFLAGS.

This means we will lose our old RCX and R11 value after the syscall
returns to userspace.

Bug reproduction steps (see how RCX and R11 change):
  $ gdb any-program

  # Run it inside the GDB
  starti

  # Set RCX and R11 before set-permission
  p $rcx=0
  p $r11=0

  # Do mprotect syscall
  set-permission $sp

  # Now the value of RCX and R11 have changed
  p $rcx
  p $r11

This commit fixes the bug by preserving RCX and R11.

Signed-off-by: Ammar Faizi <[email protected]>

How Has This Been Tested?

Architecture Yes/No Comments
x86-32 ✖️
x86-64 ✔️
ARM ✖️
AARCH64 ✖️
MIPS ✖️
POWERPC ✖️
SPARC ✖️
RISC-V ✖️
make tests ✔️

Checklist

  • My PR was done against the dev branch, not master.
  • My code follows the code style of this project.
  • My change includes a change to the documentation, if required.
  • My change adds tests as appropriate.
  • I have read and agree to the CONTRIBUTING document.

@ammarfaizi2
Copy link
Contributor Author

Test Output (Linux x86_64)

ammarfaizi2@integral:~/project/now/gef$ make -j4 test
make[1]: warning: -j4 forced in submake: resetting jobserver mode.
make[1]: Entering directory '/home/ammarfaizi2/project/now/gef/tests/binaries'
[+] Building 'bss.out'
[+] Building 'canary.out'
[+] Building 'checksec-no-canary.out'
[+] Building 'checksec-no-nx.out'
[+] Building 'checksec-no-pie.out'
[+] Building 'default.out'
[+] Building 'format-string-helper.out'
[+] Building 'heap-analysis.out'
[+] Building 'heap.out'
[+] Building 'heap-fastbins.out'
[+] Building 'heap-non-main.out'
[+] Building 'heap-tcache.out'
[+] Building 'nested2.out'
[+] Building 'nested.out'
[+] Building 'pattern.out'
[+] Building 'set-permission.out'
make[1]: Leaving directory '/home/ammarfaizi2/project/now/gef/tests/binaries'
python3 tests/runtests.py
test_func_base (__main__.TestGdbFunctionsUnit) ... ok
test_func_bss (__main__.TestGdbFunctionsUnit) ... ok
test_func_got (__main__.TestGdbFunctionsUnit) ... ok
test_func_heap (__main__.TestGdbFunctionsUnit) ... ok
test_func_stack (__main__.TestGdbFunctionsUnit) ... ok
test_cmd_aliases (__main__.TestGefCommandsUnit) ... ok
test_cmd_canary (__main__.TestGefCommandsUnit) ... ok
test_cmd_capstone_disassemble (__main__.TestGefCommandsUnit) ... ok
test_cmd_checksec (__main__.TestGefCommandsUnit) ... ok
test_cmd_dereference (__main__.TestGefCommandsUnit) ... ok
test_cmd_edit_flags (__main__.TestGefCommandsUnit) ... ok
test_cmd_elf_info (__main__.TestGefCommandsUnit) ... ok
test_cmd_entry_break (__main__.TestGefCommandsUnit) ... ok
test_cmd_format_string_helper (__main__.TestGefCommandsUnit) ... ok
test_cmd_functions (__main__.TestGefCommandsUnit) ... ok
test_cmd_got (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_analysis (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_arenas (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_bins_fast (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_bins_non_main (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_bins_tcache (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_bins_tcache_all (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_chunk (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_chunks (__main__.TestGefCommandsUnit) ... ok
test_cmd_heap_set_arena (__main__.TestGefCommandsUnit) ... ok
test_cmd_hexdump (__main__.TestGefCommandsUnit) ... ok
test_cmd_highlight (__main__.TestGefCommandsUnit) ... ok
test_cmd_keystone_assemble (__main__.TestGefCommandsUnit) ... ok
test_cmd_patch (__main__.TestGefCommandsUnit) ... ok
test_cmd_patch_byte (__main__.TestGefCommandsUnit) ... ok
test_cmd_patch_dword (__main__.TestGefCommandsUnit) ... ok
test_cmd_patch_qword (__main__.TestGefCommandsUnit) ... ok
test_cmd_patch_qword_symbol (__main__.TestGefCommandsUnit) ... ok
test_cmd_patch_string (__main__.TestGefCommandsUnit) ... ok
test_cmd_patch_word (__main__.TestGefCommandsUnit) ... ok
test_cmd_pattern_create (__main__.TestGefCommandsUnit) ... ok
test_cmd_pattern_search (__main__.TestGefCommandsUnit) ... ok
test_cmd_print_format (__main__.TestGefCommandsUnit) ... ok
test_cmd_process_search (__main__.TestGefCommandsUnit) ... ok
test_cmd_process_status (__main__.TestGefCommandsUnit) ... ok
test_cmd_registers (__main__.TestGefCommandsUnit) ... ok
test_cmd_reset_cache (__main__.TestGefCommandsUnit) ... ok
test_cmd_ropper (__main__.TestGefCommandsUnit) ... ok
test_cmd_scan (__main__.TestGefCommandsUnit) ... ok
test_cmd_search_pattern (__main__.TestGefCommandsUnit) ... ok
test_cmd_set_permission (__main__.TestGefCommandsUnit) ... ok
test_cmd_shellcode (__main__.TestGefCommandsUnit) ... ok
test_cmd_shellcode_get (__main__.TestGefCommandsUnit) ... ok
test_cmd_shellcode_search (__main__.TestGefCommandsUnit) ... ok
test_cmd_stub (__main__.TestGefCommandsUnit) ... ok
test_cmd_theme (__main__.TestGefCommandsUnit) ... ok
test_cmd_trace_run (__main__.TestGefCommandsUnit) ... ok
test_cmd_unicorn_emulate (__main__.TestGefCommandsUnit) ... ok
test_cmd_vmmap (__main__.TestGefCommandsUnit) ... ok
test_cmd_xfiles (__main__.TestGefCommandsUnit) ... ok
test_cmd_xinfo (__main__.TestGefCommandsUnit) ... ok
test_cmd_xor_memory (__main__.TestGefCommandsUnit) ... ok
test_config_show_opcodes_size (__main__.TestGefConfigUnit)
Check opcodes are correctly shown ... ok
test_func_get_filepath (__main__.TestGefFunctionsUnit) ... ok
test_func_get_memory_alignment (__main__.TestGefFunctionsUnit) ... ok
test_func_get_pid (__main__.TestGefFunctionsUnit) ... ok
test_func_set_arch (__main__.TestGefFunctionsUnit) ... ok
test_func_which (__main__.TestGefFunctionsUnit) ... ok
test_context_correct_registers_refresh_with_frames (__main__.TestNonRegressionUnit)
Ensure registers are correctly refreshed when changing frame (PR #668) ... ok
test_registers_show_registers_in_correct_order (__main__.TestNonRegressionUnit)
Ensure the registers are printed in the correct order (PR #670). ... ok

----------------------------------------------------------------------
Ran 65 tests in 263.375s

OK
make[1]: warning: -j4 forced in submake: resetting jobserver mode.
make[1]: Entering directory '/home/ammarfaizi2/project/now/gef/tests/binaries'
[+] Cleaning stuff
make[1]: Leaving directory '/home/ammarfaizi2/project/now/gef/tests/binaries'

@hugsy
Copy link
Owner

hugsy commented Jul 6, 2021

Very good catch, I'll make sure we merge this quick

tests/runtests.py Outdated Show resolved Hide resolved
tests/runtests.py Outdated Show resolved Hide resolved
On x86-64, the syscall instruction will clobber RCX and R11. It uses
RCX to save the return address and uses R11 to save the RFLAGS.

This means we will lose our old RCX and R11 value after the syscall
returns to userspace.

Bug reproduction steps (see how RCX and R11 change):
  $ gdb any-program

  # Run it inside the GDB
  starti

  # Set RCX and R11 before set-permission
  p $rcx=0
  p $r11=0

  # Do mprotect syscall
  set-permission $sp

  # Now the value of RCX and R11 have changed
  p $rcx
  p $r11

This commit fixes the bug by preserving RCX and R11.

Signed-off-by: Ammar Faizi <[email protected]>
@Grazfather
Copy link
Collaborator

Thanks for this! Good find.

@Grazfather Grazfather merged commit 5eb3b24 into hugsy:dev Jul 6, 2021
ammarfaizi2 added a commit to ammarfaizi2/gef that referenced this pull request Jul 7, 2021
This commit introduces 2 changes:
1) Change "info registers" to "info registers all". This will track more
registers and make sure they are not changed due to syscall.

2) Change `gdb_start_silent_cmd` to `gdb_run_cmd`. We don't need to use
`gdb_start_silent_cmd` because our `before` commands have already
started the process. And we can't see the register before we do
`set-permission` command if the process has not been started yet.

Therefore, it makes sense not to append `"entry-break"` (calling
`gdb_start_silent_cmd`).

Before this commit the result in commands will be like this:
      before = [
        # These two do the entry-break job!
        "starti",
        "si",

        "printf \"match_before\\n\"",
        "info registers all",
        "printf \"match_before\\n\"",
        "gef config context.clear_screen False",
        "gef config context.layout '-code -stack'",

        # This is unnecessary, because we have `starti` and `si`.
        # We can't reorder it because it is appended inside the
        # `gdb_start_silent_cmd`.
        "entry-break"
      ]
      cmd = "set-permission $sp"
      after = [
        "printf \"match_after\\n\"",
        "info registers all",
        "printf \"match_after\\n\""
      ]

After this commit:
      before = [
        "entry-break",
        "printf \"match_before\\n\"",
        "info registers all",
        "printf \"match_before\\n\""
      ]
      cmd = "set-permission $sp"
      after = [
        "printf \"match_after\\n\"",
        "info registers all",
        "printf \"match_after\\n\""
      ]

Link: hugsy#673 (comment)
Fixes: 5eb3b24 ("x86-64: Preserve RCX and R11 when calling mprotect_asm (syscall)")
Cc: Grazfather <[email protected]>
Signed-off-by: Ammar Faizi <[email protected]>
ammarfaizi2 added a commit to ammarfaizi2/gef that referenced this pull request Jul 7, 2021
This commit introduces 2 changes:
1) Change "info registers" to "info registers all". This will track more
registers and make sure they are not changed due to syscall.

2) Change `gdb_start_silent_cmd` to `gdb_run_cmd`. We don't need to use
`gdb_start_silent_cmd` because our `before` commands have already
started the process. And we can't see the register before we do
`set-permission` command if the process has not been started yet.

Therefore, it makes sense not to append `"entry-break"` (calling
`gdb_start_silent_cmd`).

Before this commit the result in commands will be like this:
      before = [
        # These two do the entry-break job!
        "starti",
        "si",

        "printf \"match_before\\n\"",
        "info registers all",
        "printf \"match_before\\n\"",
        "gef config context.clear_screen False",
        "gef config context.layout '-code -stack'",

        # This is unnecessary, because we have `starti` and `si`.
        # We can't reorder it because it is appended inside the
        # `gdb_start_silent_cmd`.
        "entry-break"
      ]
      cmd = "set-permission $sp"
      after = [
        "printf \"match_after\\n\"",
        "info registers all",
        "printf \"match_after\\n\""
      ]

After this commit:
      before = [
        "entry-break",
        "printf \"match_before\\n\"",
        "info registers all",
        "printf \"match_before\\n\""
      ]
      cmd = "set-permission $sp"
      after = [
        "printf \"match_after\\n\"",
        "info registers all",
        "printf \"match_after\\n\""
      ]

Link: hugsy#673 (comment)
Fixes: 5eb3b24 ("x86-64: Preserve RCX and R11 when calling mprotect_asm (syscall)")
Cc: Grazfather <[email protected]>
Signed-off-by: Ammar Faizi <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants