Files
AFLplusplus/include/afl-as.h
van Hauser 602eceed8b push to stable (#1983)
* Output afl-clang-fast stuffs only if necessary (#1912)

* afl-cc header

* afl-cc common declarations

 - Add afl-cc-state.c
 - Strip includes, find_object, debug/be_quiet/have_*/callname setting from afl-cc.c
 - Use debugf_args in main
 - Modify execvp stuffs to fit new aflcc struct

* afl-cc show usage

* afl-cc mode selecting

1. compiler_mode by callname in argv[0]
2. compiler_mode by env "AFL_CC_COMPILER"
3. compiler_mode/instrument_mode by command line options "--afl-..."
4. instrument_mode/compiler_mode by various env vars including "AFL_LLVM_INSTRUMENT"
5. final checking steps
6. print "... - mode: %s-%s\n"
7. determine real argv[0] according to compiler_mode

* afl-cc macro defs

* afl-cc linking behaviors

* afl-cc fsanitize behaviors

* afl-cc misc

* afl-cc body update

* afl-cc all-in-one

formated with custom-format.py

* nits

---------

Co-authored-by: vanhauser-thc <vh@thc.org>

* changelog

* update grammar mutator

* lto llvm 12+

* docs(custom_mutators): fix missing ':' (#1953)

* Fix broken LTO mode and response file support (#1948)

* Strip `-Wl,-no-undefined` during compilation (#1952)

Make the compiler wrapper stripping `-Wl,-no-undefined` in addition to `-Wl,--no-undefined`.
Both versions of the flag are accepted by clang and, therefore, used by building systems in the wild (e.g., samba will not build without this fix).

* Remove dead code in write_to_testcase (#1955)

The custom_mutators_count check in if case is duplicate with if condition.
The else case is custom_mutators_count == 0, neither custom_mutator_list iteration nor sent check needed.

Signed-off-by: Xeonacid <h.dwwwwww@gmail.com>

* update qemuafl

* WIP: Add ability to generate drcov trace using QEMU backend (#1956)

* Document new drcov QEMU plugin

* Add link to lightkeeper for QEMU drcov file loading

---------

Co-authored-by: Jean-Romain Garnier <jean-romain.garnier@airbus.com>

* code format

* changelog

* sleep on uid != 0 afl-system-config

* fix segv about skip_next, warn on unsupported cases of linking options (#1958)

* todos

* ensure afl-cc only allows available compiler modes

* update grammar mutator

* disable aslr on apple

* fix for arm64

* help selective instrumentation

* typos

* macos

* add compiler test script

* apple fixes

* bump nyx submodules (#1963)

* fix docs

* update changelog

* update grammar mutator

* improve compiler test script

* gcc asan workaround (#1966)

* fix github merge fuckup

* fix

* Fix afl-cc (#1968)

- Check if too many cmdline params here, each time before insert a new param.
 - Check if it is "-fsanitize=..." before we do sth.
 - Remove improper param_st transfer.

* Avoid adding llvmnative instrumentation when linking rust sanitizer runtime (#1969)

* Dynamic instrumentation filtering for LLVM native (#1971)

* Add two dynamic instrumentation filter methods to runtime

* Always use pc-table with native pcguard

* Add make_symbol_list.py and README

* changelog

* todos

* new forkserver check

* fix

* nyx test for CI

* improve nyx docs

* Fixes to afl-cc and documentation (#1974)

* Always compile with -ldl when building for CODE_COVERAGE

When building with CODE_COVERAGE, the afl runtime contains code that
calls `dladdr` which requires -ldl. Under most circumstances, clang
already adds this (e.g. when building with pc-table), but there are some
circumstances where it isn't added automatically.

* Add visibility declaration to __afl_connected

When building with hidden visibility, the use of __AFL_LOOP inside such
code can cause linker errors due to __afl_connected being declared
"hidden".

* Update docs to clarify that CODE_COVERAGE=1 is required for dynamic_covfilter

* nits

* nyx build script updates

* test error output

* debug ci

* debug ci

* Improve afl-cc (#1975)

* update response file support

 - full support of rsp file
 - fix some segv issues

* Improve afl-cc

 - remove dead code about allow/denylist options of sancov
 - missing `if (!aflcc->have_msan)`
 - add docs for each function
 - typo

* enable nyx

* debug ci

* debug ci

* debug ci

* debug ci

* debug ci

* debug ci

* debug ci

* debug ci

* fix ci

* clean test script

* NO_NYX

* NO_NYX

* fix ci

* debug ci

* fix ci

* finalize ci fix

* Enhancement on Deterministic stage (#1972)

* fuzzer: init commit based on aflpp 60dc37a8cf

* fuzzers: adding the skip variables and initialize

* log: profile the det/havoc finding

* log: add profile log output

* fuzzers: sperate log/skipdet module

* fuzzers: add quick eff_map calc

* fuzzers: add skip_eff_map in fuzz_one

* fuzzers: mark whole input space in eff_map

* fuzzers: add undet bit threshold to skip some seeds

* fuzzers: fix one byte overflow

* fuzzers: fix overflow

* fix code format

* add havoc only again

* code format

* remove log to INTROSPECTION, rename skipdet module

* rename skipdet module

* remove log to stats

* clean redundant code

* code format

* remove redundant code format check

* remove redundant doc

* remove redundant objects

* clean files

* change -d to default skipdet

* disable deterministic when using CUSTOM_MUTATOR

* revert fix

* final touches for skipdet

* remove unused var

* remove redundant eff struct (#1977)

* update QEMU-Nyx submodule (#1978)

* update QEMU-Nyx submodule (#1980)

* Fix type in AFL_NOOPT env variable in afl-cc help message (#1982)

* nits

* 2024 v4.10c release

* fixes

---------

Signed-off-by: Xeonacid <h.dwwwwww@gmail.com>
Co-authored-by: Sonic <50692172+SonicStark@users.noreply.github.com>
Co-authored-by: Xeonacid <h.dwwwwww@gmail.com>
Co-authored-by: Nils Bars <nils.bars@rub.de>
Co-authored-by: Jean-Romain Garnier <7504819+JRomainG@users.noreply.github.com>
Co-authored-by: Jean-Romain Garnier <jean-romain.garnier@airbus.com>
Co-authored-by: Sergej Schumilo <sergej@schumilo.de>
Co-authored-by: Christian Holler (:decoder) <choller@mozilla.com>
Co-authored-by: Han Zheng <35988108+kdsjZh@users.noreply.github.com>
Co-authored-by: Khaled Yakdan <yakdan@code-intelligence.com>
2024-02-03 10:55:51 +00:00

776 lines
23 KiB
C

/*
american fuzzy lop++ - injectable parts
---------------------------------------
Originally written by Michal Zalewski
Now maintained by Marc Heuse <mh@mh-sec.de>,
Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
Andrea Fioraldi <andreafioraldi@gmail.com>,
Dominik Maier <mail@dmnk.co>
Copyright 2016, 2017 Google Inc. All rights reserved.
Copyright 2019-2024 AFLplusplus Project. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at:
https://www.apache.org/licenses/LICENSE-2.0
This file houses the assembly-level instrumentation injected into fuzzed
programs. The instrumentation stores XORed pairs of data: identifiers of the
currently executing branch and the one that executed immediately before.
TL;DR: the instrumentation does shm_trace_map[cur_loc ^ prev_loc]++
The code is designed for 32-bit and 64-bit x86 systems. Both modes should
work everywhere except for Apple systems. Apple does relocations differently
from everybody else, so since their OSes have been 64-bit for a longer while,
I didn't go through the mental effort of porting the 32-bit code.
In principle, similar code should be easy to inject into any well-behaved
binary-only code (e.g., using DynamoRIO). Conditional jumps offer natural
targets for instrumentation, and should offer comparable probe density.
*/
#ifndef _HAVE_AFL_AS_H
#define _HAVE_AFL_AS_H
#include "config.h"
#include "types.h"
/*
------------------
Performances notes
------------------
Contributions to make this code faster are appreciated! Here are some
rough notes that may help with the task:
- Only the trampoline_fmt and the non-setup __afl_maybe_log code paths are
really worth optimizing; the setup / fork server stuff matters a lot less
and should be mostly just kept readable.
- We're aiming for modern CPUs with out-of-order execution and large
pipelines; the code is mostly follows intuitive, human-readable
instruction ordering, because "textbook" manual reorderings make no
substantial difference.
- Interestingly, instrumented execution isn't a lot faster if we store a
variable pointer to the setup, log, or return routine and then do a reg
call from within trampoline_fmt. It does speed up non-instrumented
execution quite a bit, though, since that path just becomes
push-call-ret-pop.
- There is also not a whole lot to be gained by doing SHM attach at a
fixed address instead of retrieving __afl_area_ptr. Although it allows us
to have a shorter log routine inserted for conditional jumps and jump
labels (for a ~10% perf gain), there is a risk of bumping into other
allocations created by the program or by tools such as ASAN.
- popf is *awfully* slow, which is why we're doing the lahf / sahf +
overflow test trick. Unfortunately, this forces us to taint eax / rax, but
this dependency on a commonly-used register still beats the alternative of
using pushf / popf.
One possible optimization is to avoid touching flags by using a circular
buffer that stores just a sequence of current locations, with the XOR stuff
happening offline. Alas, this doesn't seem to have a huge impact:
https://groups.google.com/d/msg/afl-users/MsajVf4fRLo/2u6t88ntUBIJ
- Preforking one child a bit sooner, and then waiting for the "go" command
from within the child, doesn't offer major performance gains; fork() seems
to be relatively inexpensive these days. Preforking multiple children does
help, but badly breaks the "~1 core per fuzzer" design, making it harder to
scale up. Maybe there is some middle ground.
Perhaps of note: in the 64-bit version for all platforms except for Apple,
the instrumentation is done slightly differently than on 32-bit, with
__afl_prev_loc and __afl_area_ptr being local to the object file (.lcomm),
rather than global (.comm). This is to avoid GOTRELPC lookups in the critical
code path, which AFAICT, are otherwise unavoidable if we want gcc -shared to
work; simple relocations between .bss and .text won't work on most 64-bit
platforms in such a case.
(Fun fact: on Apple systems, .lcomm can segfault the linker.)
The side effect is that state transitions are measured in a somewhat
different way, with previous tuple being recorded separately within the scope
of every .c file. This should have no impact in any practical sense.
Another side effect of this design is that getenv() will be called once per
every .o file when running in non-instrumented mode; and since getenv() tends
to be optimized in funny ways, we need to be very careful to save every
oddball register it may touch.
*/
static const u8 *trampoline_fmt_32 =
"\n"
"/* --- AFL TRAMPOLINE (32-BIT) --- */\n"
"\n"
".align 4\n"
"\n"
"leal -16(%%esp), %%esp\n"
"movl %%edi, 0(%%esp)\n"
"movl %%edx, 4(%%esp)\n"
"movl %%ecx, 8(%%esp)\n"
"movl %%eax, 12(%%esp)\n"
"movl $0x%08x, %%ecx\n"
"call __afl_maybe_log\n"
"movl 12(%%esp), %%eax\n"
"movl 8(%%esp), %%ecx\n"
"movl 4(%%esp), %%edx\n"
"movl 0(%%esp), %%edi\n"
"leal 16(%%esp), %%esp\n"
"\n"
"/* --- END --- */\n"
"\n";
static const u8 *trampoline_fmt_64 =
"\n"
"/* --- AFL TRAMPOLINE (64-BIT) --- */\n"
"\n"
".align 4\n"
"\n"
"leaq -(128+24)(%%rsp), %%rsp\n"
"movq %%rdx, 0(%%rsp)\n"
"movq %%rcx, 8(%%rsp)\n"
"movq %%rax, 16(%%rsp)\n"
"movq $0x%08x, %%rcx\n"
"call __afl_maybe_log\n"
"movq 16(%%rsp), %%rax\n"
"movq 8(%%rsp), %%rcx\n"
"movq 0(%%rsp), %%rdx\n"
"leaq (128+24)(%%rsp), %%rsp\n"
"\n"
"/* --- END --- */\n"
"\n";
static const u8 *main_payload_32 =
"\n"
"/* --- AFL MAIN PAYLOAD (32-BIT) --- */\n"
"\n"
".text\n"
".att_syntax\n"
".code32\n"
".align 8\n"
"\n"
"__afl_maybe_log:\n"
"\n"
" lahf\n"
" seto %al\n"
"\n"
" /* Check if SHM region is already mapped. */\n"
"\n"
" movl __afl_area_ptr, %edx\n"
" testl %edx, %edx\n"
" je __afl_setup\n"
"\n"
"__afl_store:\n"
"\n"
" /* Calculate and store hit for the code location specified in ecx. There\n"
" is a double-XOR way of doing this without tainting another register,\n"
" and we use it on 64-bit systems; but it's slower for 32-bit ones. */\n"
"\n"
#ifndef COVERAGE_ONLY
" movl __afl_prev_loc, %edi\n"
" xorl %ecx, %edi\n"
" shrl $1, %ecx\n"
" movl %ecx, __afl_prev_loc\n"
#else
" movl %ecx, %edi\n"
#endif /* ^!COVERAGE_ONLY */
"\n"
#ifdef SKIP_COUNTS
" orb $1, (%edx, %edi, 1)\n"
#else
" addb $1, (%edx, %edi, 1)\n"
" adcb $0, (%edx, %edi, 1)\n" // never zero counter implementation. slightly better path discovery and little performance impact
#endif /* ^SKIP_COUNTS */
"\n"
"__afl_return:\n"
"\n"
" addb $127, %al\n"
" sahf\n"
" ret\n"
"\n"
".align 8\n"
"\n"
"__afl_setup:\n"
"\n"
" /* Do not retry setup if we had previous failures. */\n"
"\n"
" cmpb $0, __afl_setup_failure\n"
" jne __afl_return\n"
"\n"
" /* Map SHM, jumping to __afl_setup_abort if something goes wrong.\n"
" We do not save FPU/MMX/SSE registers here, but hopefully, nobody\n"
" will notice this early in the game. */\n"
"\n"
" pushl %eax\n"
" pushl %ecx\n"
"\n"
" pushl $.AFL_SHM_ENV\n"
" call getenv\n"
" addl $4, %esp\n"
"\n"
" testl %eax, %eax\n"
" je __afl_setup_abort\n"
"\n"
#ifdef USEMMAP
" pushl $384 /* shm_open mode 0600 */\n"
" pushl $2 /* flags O_RDWR */\n"
" pushl %eax /* SHM file path */\n"
" call shm_open\n"
" addl $12, %esp\n"
"\n"
" cmpl $-1, %eax\n"
" je __afl_setup_abort\n"
"\n"
" pushl $0 /* mmap off */\n"
" pushl %eax /* shm fd */\n"
" pushl $1 /* mmap flags */\n"
" pushl $3 /* mmap prot */\n"
" pushl $"STRINGIFY(MAP_SIZE)" /* mmap len */\n"
" pushl $0 /* mmap addr */\n"
" call mmap\n"
" addl $12, %esp\n"
"\n"
" cmpl $-1, %eax\n"
" je __afl_setup_abort\n"
"\n"
#else
" pushl %eax\n"
" call atoi\n"
" addl $4, %esp\n"
"\n"
" pushl $0 /* shmat flags */\n"
" pushl $0 /* requested addr */\n"
" pushl %eax /* SHM ID */\n"
" call shmat\n"
" addl $12, %esp\n"
"\n"
" cmpl $-1, %eax\n"
" je __afl_setup_abort\n"
"\n"
#endif
" movb $1, (%eax)\n"
" /* Store the address of the SHM region. */\n"
"\n"
" movl %eax, __afl_area_ptr\n"
" movl %eax, %edx\n"
"\n"
" popl %ecx\n"
" popl %eax\n"
"\n"
"__afl_forkserver:\n"
"\n"
" /* Enter the fork server mode to avoid the overhead of execve() calls. */\n"
"\n"
" pushl %eax\n"
" pushl %ecx\n"
" pushl %edx\n"
"\n"
" /* Phone home and tell the parent that we're OK. (Note that signals with\n"
" no SA_RESTART will mess it up). If this fails, assume that the fd is\n"
" closed because we were execve()d from an instrumented binary, or because\n"
" the parent doesn't want to use the fork server. */\n"
"\n"
" pushl $4 /* length */\n"
" pushl $__afl_temp /* data */\n"
" pushl $" STRINGIFY((FORKSRV_FD + 1)) " /* file desc */\n"
" call write\n"
" addl $12, %esp\n"
"\n"
" cmpl $4, %eax\n"
" jne __afl_fork_resume\n"
"\n"
"__afl_fork_wait_loop:\n"
"\n"
" /* Wait for parent by reading from the pipe. Abort if read fails. */\n"
"\n"
" pushl $4 /* length */\n"
" pushl $__afl_temp /* data */\n"
" pushl $" STRINGIFY(FORKSRV_FD) " /* file desc */\n"
" call read\n"
" addl $12, %esp\n"
"\n"
" cmpl $4, %eax\n"
" jne __afl_die\n"
"\n"
" /* Once woken up, create a clone of our process. This is an excellent use\n"
" case for syscall(__NR_clone, 0, CLONE_PARENT), but glibc boneheadedly\n"
" caches getpid() results and offers no way to update the value, breaking\n"
" abort(), raise(), and a bunch of other things :-( */\n"
"\n"
" call fork\n"
"\n"
" cmpl $0, %eax\n"
" jl __afl_die\n"
" je __afl_fork_resume\n"
"\n"
" /* In parent process: write PID to pipe, then wait for child. */\n"
"\n"
" movl %eax, __afl_fork_pid\n"
"\n"
" pushl $4 /* length */\n"
" pushl $__afl_fork_pid /* data */\n"
" pushl $" STRINGIFY((FORKSRV_FD + 1)) " /* file desc */\n"
" call write\n"
" addl $12, %esp\n"
"\n"
" pushl $0 /* no flags */\n"
" pushl $__afl_temp /* status */\n"
" pushl __afl_fork_pid /* PID */\n"
" call waitpid\n"
" addl $12, %esp\n"
"\n"
" cmpl $0, %eax\n"
" jle __afl_die\n"
"\n"
" /* Relay wait status to pipe, then loop back. */\n"
"\n"
" pushl $4 /* length */\n"
" pushl $__afl_temp /* data */\n"
" pushl $" STRINGIFY((FORKSRV_FD + 1)) " /* file desc */\n"
" call write\n"
" addl $12, %esp\n"
"\n"
" jmp __afl_fork_wait_loop\n"
"\n"
"__afl_fork_resume:\n"
"\n"
" /* In child process: close fds, resume execution. */\n"
"\n"
" pushl $" STRINGIFY(FORKSRV_FD) "\n"
" call close\n"
"\n"
" pushl $" STRINGIFY((FORKSRV_FD + 1)) "\n"
" call close\n"
"\n"
" addl $8, %esp\n"
"\n"
" popl %edx\n"
" popl %ecx\n"
" popl %eax\n"
" jmp __afl_store\n"
"\n"
"__afl_die:\n"
"\n"
" xorl %eax, %eax\n"
" call _exit\n"
"\n"
"__afl_setup_abort:\n"
"\n"
" /* Record setup failure so that we don't keep calling\n"
" shmget() / shmat() over and over again. */\n"
"\n"
" incb __afl_setup_failure\n"
" popl %ecx\n"
" popl %eax\n"
" jmp __afl_return\n"
"\n"
".AFL_VARS:\n"
"\n"
" .comm __afl_area_ptr, 4, 32\n"
" .comm __afl_setup_failure, 1, 32\n"
#ifndef COVERAGE_ONLY
" .comm __afl_prev_loc, 4, 32\n"
#endif /* !COVERAGE_ONLY */
" .comm __afl_final_loc, 4, 32\n"
" .comm __afl_fork_pid, 4, 32\n"
" .comm __afl_temp, 4, 32\n"
"\n"
".AFL_SHM_ENV:\n"
" .asciz \"" SHM_ENV_VAR "\"\n"
"\n"
"/* --- END --- */\n"
"\n";
/* The OpenBSD hack is due to lahf and sahf not being recognized by some
versions of binutils: https://marc.info/?l=openbsd-cvs&m=141636589924400
The Apple code is a bit different when calling libc functions because
they are doing relocations differently from everybody else. We also need
to work around the crash issue with .lcomm and the fact that they don't
recognize .string. */
#ifdef __APPLE__
#define CALL_L64(str) "call _" str "\n"
#else
#define CALL_L64(str) "call " str "@PLT\n"
#endif /* ^__APPLE__ */
static const u8 *main_payload_64 =
"\n"
"/* --- AFL MAIN PAYLOAD (64-BIT) --- */\n"
"\n"
".text\n"
".att_syntax\n"
".code64\n"
".align 8\n"
"\n"
"__afl_maybe_log:\n"
"\n"
#if defined(__OpenBSD__) || (defined(__FreeBSD__) && (__FreeBSD__ < 9))
" .byte 0x9f /* lahf */\n"
#else
" lahf\n"
#endif /* ^__OpenBSD__, etc */
" seto %al\n"
"\n"
" /* Check if SHM region is already mapped. */\n"
"\n"
" movq __afl_area_ptr(%rip), %rdx\n"
" testq %rdx, %rdx\n"
" je __afl_setup\n"
"\n"
"__afl_store:\n"
"\n"
" /* Calculate and store hit for the code location specified in rcx. */\n"
"\n"
#ifndef COVERAGE_ONLY
" xorq __afl_prev_loc(%rip), %rcx\n"
" xorq %rcx, __afl_prev_loc(%rip)\n"
" shrq $1, __afl_prev_loc(%rip)\n"
#endif /* ^!COVERAGE_ONLY */
"\n"
#ifdef SKIP_COUNTS
" orb $1, (%rdx, %rcx, 1)\n"
#else
" addb $1, (%rdx, %rcx, 1)\n"
" adcb $0, (%rdx, %rcx, 1)\n" // never zero counter implementation. slightly better path discovery and little performance impact
#endif /* ^SKIP_COUNTS */
"\n"
"__afl_return:\n"
"\n"
" addb $127, %al\n"
#if defined(__OpenBSD__) || (defined(__FreeBSD__) && (__FreeBSD__ < 9))
" .byte 0x9e /* sahf */\n"
#else
" sahf\n"
#endif /* ^__OpenBSD__, etc */
" ret\n"
"\n"
".align 8\n"
"\n"
"__afl_setup:\n"
"\n"
" /* Do not retry setup if we had previous failures. */\n"
"\n"
" cmpb $0, __afl_setup_failure(%rip)\n"
" jne __afl_return\n"
"\n"
" /* Check out if we have a global pointer on file. */\n"
"\n"
#ifndef __APPLE__
" movq __afl_global_area_ptr@GOTPCREL(%rip), %rdx\n"
" movq (%rdx), %rdx\n"
#else
" movq __afl_global_area_ptr(%rip), %rdx\n"
#endif /* !^__APPLE__ */
" testq %rdx, %rdx\n"
" je __afl_setup_first\n"
"\n"
" movq %rdx, __afl_area_ptr(%rip)\n"
" jmp __afl_store\n"
"\n"
"__afl_setup_first:\n"
"\n"
" /* Save everything that is not yet saved and that may be touched by\n"
" getenv() and several other libcalls we'll be relying on. */\n"
"\n"
" leaq -352(%rsp), %rsp\n"
"\n"
" movq %rax, 0(%rsp)\n"
" movq %rcx, 8(%rsp)\n"
" movq %rdi, 16(%rsp)\n"
" movq %rsi, 32(%rsp)\n"
" movq %r8, 40(%rsp)\n"
" movq %r9, 48(%rsp)\n"
" movq %r10, 56(%rsp)\n"
" movq %r11, 64(%rsp)\n"
"\n"
" movq %xmm0, 96(%rsp)\n"
" movq %xmm1, 112(%rsp)\n"
" movq %xmm2, 128(%rsp)\n"
" movq %xmm3, 144(%rsp)\n"
" movq %xmm4, 160(%rsp)\n"
" movq %xmm5, 176(%rsp)\n"
" movq %xmm6, 192(%rsp)\n"
" movq %xmm7, 208(%rsp)\n"
" movq %xmm8, 224(%rsp)\n"
" movq %xmm9, 240(%rsp)\n"
" movq %xmm10, 256(%rsp)\n"
" movq %xmm11, 272(%rsp)\n"
" movq %xmm12, 288(%rsp)\n"
" movq %xmm13, 304(%rsp)\n"
" movq %xmm14, 320(%rsp)\n"
" movq %xmm15, 336(%rsp)\n"
"\n"
" /* Map SHM, jumping to __afl_setup_abort if something goes wrong. */\n"
"\n"
" /* The 64-bit ABI requires 16-byte stack alignment. We'll keep the\n"
" original stack ptr in the callee-saved r12. */\n"
"\n"
" pushq %r12\n"
" movq %rsp, %r12\n"
" subq $16, %rsp\n"
" andq $0xfffffffffffffff0, %rsp\n"
"\n"
" leaq .AFL_SHM_ENV(%rip), %rdi\n"
CALL_L64("getenv")
"\n"
" testq %rax, %rax\n"
" je __afl_setup_abort\n"
"\n"
#ifdef USEMMAP
" movl $384, %edx /* shm_open mode 0600 */\n"
" movl $2, %esi /* flags O_RDWR */\n"
" movq %rax, %rdi /* SHM file path */\n"
CALL_L64("shm_open")
"\n"
" cmpq $-1, %rax\n"
" je __afl_setup_abort\n"
"\n"
" movl $0, %r9d\n"
" movl %eax, %r8d\n"
" movl $1, %ecx\n"
" movl $3, %edx\n"
" movl $"STRINGIFY(MAP_SIZE)", %esi\n"
" movl $0, %edi\n"
CALL_L64("mmap")
"\n"
" cmpq $-1, %rax\n"
" je __afl_setup_abort\n"
"\n"
#else
" movq %rax, %rdi\n"
CALL_L64("atoi")
"\n"
" xorq %rdx, %rdx /* shmat flags */\n"
" xorq %rsi, %rsi /* requested addr */\n"
" movq %rax, %rdi /* SHM ID */\n"
CALL_L64("shmat")
"\n"
" cmpq $-1, %rax\n"
" je __afl_setup_abort\n"
"\n"
#endif
" movb $1, (%rax)\n"
" /* Store the address of the SHM region. */\n"
"\n"
" movq %rax, %rdx\n"
" movq %rax, __afl_area_ptr(%rip)\n"
"\n"
#ifdef __APPLE__
" movq %rax, __afl_global_area_ptr(%rip)\n"
#else
" movq __afl_global_area_ptr@GOTPCREL(%rip), %rdx\n"
" movq %rax, (%rdx)\n"
#endif /* ^__APPLE__ */
" movq %rax, %rdx\n"
"\n"
"__afl_forkserver:\n"
"\n"
" /* Enter the fork server mode to avoid the overhead of execve() calls. We\n"
" push rdx (area ptr) twice to keep stack alignment neat. */\n"
"\n"
" pushq %rdx\n"
" pushq %rdx\n"
"\n"
" /* Phone home and tell the parent that we're OK. (Note that signals with\n"
" no SA_RESTART will mess it up). If this fails, assume that the fd is\n"
" closed because we were execve()d from an instrumented binary, or because\n"
" the parent doesn't want to use the fork server. */\n"
"\n"
" movq $4, %rdx /* length */\n"
" leaq __afl_temp(%rip), %rsi /* data */\n"
" movq $" STRINGIFY((FORKSRV_FD + 1)) ", %rdi /* file desc */\n"
CALL_L64("write")
"\n"
" cmpq $4, %rax\n"
" jne __afl_fork_resume\n"
"\n"
"__afl_fork_wait_loop:\n"
"\n"
" /* Wait for parent by reading from the pipe. Abort if read fails. */\n"
"\n"
" movq $4, %rdx /* length */\n"
" leaq __afl_temp(%rip), %rsi /* data */\n"
" movq $" STRINGIFY(FORKSRV_FD) ", %rdi /* file desc */\n"
CALL_L64("read")
" cmpq $4, %rax\n"
" jne __afl_die\n"
"\n"
" /* Once woken up, create a clone of our process. This is an excellent use\n"
" case for syscall(__NR_clone, 0, CLONE_PARENT), but glibc boneheadedly\n"
" caches getpid() results and offers no way to update the value, breaking\n"
" abort(), raise(), and a bunch of other things :-( */\n"
"\n"
CALL_L64("fork")
" cmpq $0, %rax\n"
" jl __afl_die\n"
" je __afl_fork_resume\n"
"\n"
" /* In parent process: write PID to pipe, then wait for child. */\n"
"\n"
" movl %eax, __afl_fork_pid(%rip)\n"
"\n"
" movq $4, %rdx /* length */\n"
" leaq __afl_fork_pid(%rip), %rsi /* data */\n"
" movq $" STRINGIFY((FORKSRV_FD + 1)) ", %rdi /* file desc */\n"
CALL_L64("write")
"\n"
" movq $0, %rdx /* no flags */\n"
" leaq __afl_temp(%rip), %rsi /* status */\n"
" movq __afl_fork_pid(%rip), %rdi /* PID */\n"
CALL_L64("waitpid")
" cmpq $0, %rax\n"
" jle __afl_die\n"
"\n"
" /* Relay wait status to pipe, then loop back. */\n"
"\n"
" movq $4, %rdx /* length */\n"
" leaq __afl_temp(%rip), %rsi /* data */\n"
" movq $" STRINGIFY((FORKSRV_FD + 1)) ", %rdi /* file desc */\n"
CALL_L64("write")
"\n"
" jmp __afl_fork_wait_loop\n"
"\n"
"__afl_fork_resume:\n"
"\n"
" /* In child process: close fds, resume execution. */\n"
"\n"
" movq $" STRINGIFY(FORKSRV_FD) ", %rdi\n"
CALL_L64("close")
"\n"
" movq $" STRINGIFY((FORKSRV_FD + 1)) ", %rdi\n"
CALL_L64("close")
"\n"
" popq %rdx\n"
" popq %rdx\n"
"\n"
" movq %r12, %rsp\n"
" popq %r12\n"
"\n"
" movq 0(%rsp), %rax\n"
" movq 8(%rsp), %rcx\n"
" movq 16(%rsp), %rdi\n"
" movq 32(%rsp), %rsi\n"
" movq 40(%rsp), %r8\n"
" movq 48(%rsp), %r9\n"
" movq 56(%rsp), %r10\n"
" movq 64(%rsp), %r11\n"
"\n"
" movq 96(%rsp), %xmm0\n"
" movq 112(%rsp), %xmm1\n"
" movq 128(%rsp), %xmm2\n"
" movq 144(%rsp), %xmm3\n"
" movq 160(%rsp), %xmm4\n"
" movq 176(%rsp), %xmm5\n"
" movq 192(%rsp), %xmm6\n"
" movq 208(%rsp), %xmm7\n"
" movq 224(%rsp), %xmm8\n"
" movq 240(%rsp), %xmm9\n"
" movq 256(%rsp), %xmm10\n"
" movq 272(%rsp), %xmm11\n"
" movq 288(%rsp), %xmm12\n"
" movq 304(%rsp), %xmm13\n"
" movq 320(%rsp), %xmm14\n"
" movq 336(%rsp), %xmm15\n"
"\n"
" leaq 352(%rsp), %rsp\n"
"\n"
" jmp __afl_store\n"
"\n"
"__afl_die:\n"
"\n"
" xorq %rax, %rax\n"
CALL_L64("_exit")
"\n"
"__afl_setup_abort:\n"
"\n"
" /* Record setup failure so that we don't keep calling\n"
" shmget() / shmat() over and over again. */\n"
"\n"
" incb __afl_setup_failure(%rip)\n"
"\n"
" movq %r12, %rsp\n"
" popq %r12\n"
"\n"
" movq 0(%rsp), %rax\n"
" movq 8(%rsp), %rcx\n"
" movq 16(%rsp), %rdi\n"
" movq 32(%rsp), %rsi\n"
" movq 40(%rsp), %r8\n"
" movq 48(%rsp), %r9\n"
" movq 56(%rsp), %r10\n"
" movq 64(%rsp), %r11\n"
"\n"
" movq 96(%rsp), %xmm0\n"
" movq 112(%rsp), %xmm1\n"
" movq 128(%rsp), %xmm2\n"
" movq 144(%rsp), %xmm3\n"
" movq 160(%rsp), %xmm4\n"
" movq 176(%rsp), %xmm5\n"
" movq 192(%rsp), %xmm6\n"
" movq 208(%rsp), %xmm7\n"
" movq 224(%rsp), %xmm8\n"
" movq 240(%rsp), %xmm9\n"
" movq 256(%rsp), %xmm10\n"
" movq 272(%rsp), %xmm11\n"
" movq 288(%rsp), %xmm12\n"
" movq 304(%rsp), %xmm13\n"
" movq 320(%rsp), %xmm14\n"
" movq 336(%rsp), %xmm15\n"
"\n"
" leaq 352(%rsp), %rsp\n"
"\n"
" jmp __afl_return\n"
"\n"
".AFL_VARS:\n"
"\n"
#ifdef __APPLE__
" .comm __afl_area_ptr, 8\n"
#ifndef COVERAGE_ONLY
" .comm __afl_prev_loc, 8\n"
#endif /* !COVERAGE_ONLY */
" .comm __afl_fork_pid, 4\n"
" .comm __afl_temp, 4\n"
" .comm __afl_setup_failure, 1\n"
#else
" .lcomm __afl_area_ptr, 8\n"
#ifndef COVERAGE_ONLY
" .lcomm __afl_prev_loc, 8\n"
#endif /* !COVERAGE_ONLY */
" .lcomm __afl_fork_pid, 4\n"
" .lcomm __afl_temp, 4\n"
" .lcomm __afl_setup_failure, 1\n"
#endif /* ^__APPLE__ */
" .comm __afl_global_area_ptr, 8, 8\n"
"\n"
".AFL_SHM_ENV:\n"
" .asciz \"" SHM_ENV_VAR "\"\n"
"\n"
"/* --- END --- */\n"
"\n";
#endif /* !_HAVE_AFL_AS_H */