Replace GC implementation with bdwgc (envoyproxy#73)

buildtools/bdwgc/Dockerfile

@@ -0,0 +1,22 @@
+# Copyright The OWASP Coraza contributors
+# SPDX-License-Identifier: Apache-2.0
+
+FROM ghcr.io/corazawaf/coraza-proxy-wasm/buildtools-wasi-sdk:main
+
+RUN apt-get install -y autogen autoconf automake libtool patch
+
+RUN mkdir -p /bdwgc && curl -L https://github.com/ivmai/bdwgc/archive/refs/tags/v8.2.2.tar.gz | tar -xz --strip-components 1 -C /bdwgc
+WORKDIR /bdwgc
+ADD bdwgc.patch bdwgc.patch
+RUN patch -p1 < bdwgc.patch
+RUN ./autogen.sh
+
+# -D_WASI_EMULATED_MMAN allows wasi-libc's mmap header files to be included. We do not actually use
+# the library itself, instead reimplementing in mmap.go
+ENV CFLAGS ${CFLAGS} -D_WASI_EMULATED_MMAN
+
+# host is required by configure but not used so set it arbitrarily
+RUN ./configure --disable-threads --enable-mmap --disable-shared --disable-gcj-support --disable-java-finalization --disable-atomic-uncollectible --host=i686-pc-linux-gnu
+RUN make
+
+CMD ["cp", "/bdwgc/.libs/libgc.a", "/out/"]

buildtools/bdwgc/bdwgc.patch

@@ -0,0 +1,109 @@
+diff --git a/alloc.c b/alloc.c
+index a53e12c0..6e7be764 100644
+--- a/alloc.c
++++ b/alloc.c
+@@ -19,7 +19,7 @@
+ #include "private/gc_priv.h"
+
+ #include <stdio.h>
+-#if !defined(MACOS) && !defined(MSWINCE)
++#if !defined(MACOS) && !defined(MSWINCE) && !defined(WASI)
+ # include <signal.h>
+ # if !defined(GC_NO_TYPES) && !defined(SN_TARGET_PSP2) \
+      && !defined(__CC_ARM)
+diff --git a/include/private/gc_priv.h b/include/private/gc_priv.h
+index 2ffd4dd0..28ed5100 100644
+--- a/include/private/gc_priv.h
++++ b/include/private/gc_priv.h
+@@ -838,7 +838,9 @@ EXTERN_C_END
+ # endif
+ #endif /* DARWIN */
+
+-#include <setjmp.h>
++#if !defined(GC_NO_SIGSETJMP)
++# include <setjmp.h>
++#endif
+
+ #if __STDC_VERSION__ >= 201112L
+ # include <assert.h> /* for static_assert */
+diff --git a/include/private/gcconfig.h b/include/private/gcconfig.h
+index 2de01afe..cf25e78c 100644
+--- a/include/private/gcconfig.h
++++ b/include/private/gcconfig.h
+@@ -703,6 +703,11 @@ EXTERN_C_BEGIN
+ #   define I386
+ #   define mach_type_known
+ # endif
++# if defined(__wasi__)
++#   define WASI
++#   define I386
++#   define mach_type_known
++#endif
+
+ /* Feel free to add more clauses here */
+
+@@ -1425,6 +1430,21 @@ EXTERN_C_BEGIN
+ #     define USE_MMAP_ANON      /* avoid /dev/zero, not supported */
+ #     define STACK_GROWS_DOWN
+ #   endif
++#   ifdef WASI
++      extern unsigned char __global_base;
++      extern unsigned char __heap_base;
++#     define OS_TYPE "WASI"
++#     define DATASTART ((ptr_t)&__global_base)
++#     define DATAEND ((ptr_t)&__heap_base)
++#     define STACKBOTTOM ((ptr_t)&__global_base)
++#     define GETPAGESIZE() 65536
++#     define USE_MMAP_ANON      /* avoid /dev/zero, not supported */
++#     define GC_NO_SIGSETJMP 1
++#     define NO_CLOCK 1
++#     ifndef HBLKSIZE
++#       define HBLKSIZE 4096
++#     endif
++#   endif
+ #   if defined(__QNX__)
+ #     define OS_TYPE "QNX"
+ #     define SA_RESTART 0
+diff --git a/misc.c b/misc.c
+index 64eeb7e4..273a1fa7 100644
+--- a/misc.c
++++ b/misc.c
+@@ -20,7 +20,7 @@
+ #include <limits.h>
+ #include <stdarg.h>
+
+-#ifndef MSWINCE
++#if !defined(MSWINCE) && !defined(WASI)
+ # include <signal.h>
+ #endif
+
+diff --git a/os_dep.c b/os_dep.c
+index e116ad02..4bd9031c 100644
+--- a/os_dep.c
++++ b/os_dep.c
+@@ -27,14 +27,14 @@
+ #endif
+
+ #include <stdio.h>
+-#if defined(MSWINCE) || defined(SN_TARGET_PS3)
++#if defined(MSWINCE) || defined(SN_TARGET_PS3) || defined(WASI)
+ # define SIGSEGV 0 /* value is irrelevant */
+ #else
+ # include <signal.h>
+ #endif
+
+ #if defined(UNIX_LIKE) || defined(CYGWIN32) || defined(NACL) \
+-    || defined(SYMBIAN)
++    || defined(SYMBIAN) || defined(WASI)
+ # include <fcntl.h>
+ #endif
+
+@@ -2668,7 +2668,7 @@ static void block_unmap_inner(ptr_t start_addr, size_t len)
+             if (madvise(start_addr, len, MADV_DONTNEED) == -1)
+               ABORT_ON_REMAP_FAIL("unmap: madvise", start_addr, len);
+ #         endif
+-#       elif defined(EMSCRIPTEN)
++#       elif defined(EMSCRIPTEN) || defined(WASI)
+           /* Nothing to do, mmap(PROT_NONE) is not supported and        */
+           /* mprotect() is just a no-op.                                */
+ #       else

buildtools/tinygo/wasi-libc.Dockerfile

@@ -5,6 +5,7 @@ FROM ghcr.io/corazawaf/coraza-proxy-wasm/buildtools-wasi-sdk:main
 
 RUN apt-get install -y git
 
+# https://github.com/tinygo-org/tinygo/pull/3280
 RUN git clone https://github.com/anuraaga/tinygo --branch wasm-stacks-nogc
 WORKDIR /tinygo
 RUN git fetch origin wasm-stacks-nogc && git reset --hard e2da8f6f0f5cf5a75a384bbddae80c1b8a84eca7

init_tinygo.go

@@ -7,5 +7,5 @@ package main
 
 import _ "github.com/corazawaf/coraza-proxy-wasm/internal/agc"
 
-// #cgo LDFLAGS: lib/libinjection.a lib/libre2.a lib/libcre2.a lib/libc++.a lib/libc++abi.a lib/libclang_rt.builtins-wasm32.a lib/libaho_corasick.a lib/libmimalloc.a
+// #cgo LDFLAGS: lib/libinjection.a lib/libre2.a lib/libcre2.a lib/libc++.a lib/libc++abi.a lib/libclang_rt.builtins-wasm32.a lib/libaho_corasick.a lib/libmimalloc.a lib/libgc.a
 import "C"

internal/agc/doc.go

@@ -1,15 +1,12 @@
 // Copyright The OWASP Coraza contributors
 // SPDX-License-Identifier: Apache-2.0
 
-// Package agc is a custom gargabe gollector for TinyGo. The main difference is instead of taking
-// ownership of the entire process heap, it uses malloc to allocate blocks for the GC to then
-// assign to allocated objects.
+// Package agc is a custom garbage collector for TinyGo. It delegates to bdwgc for actual
+// allocation and collection.
 //
 // Unfortunately, we must rely on a package init() method for initializing the heap because we
 // cannot override TinyGo's initHeap function that normally does it. This means initialization
 // order matters, this package should be the first package to be initialized - any packages
 // initialized before cannot allocate memory. For that reason, we have named this agc instead
 // of gc.
-//
-// Currently, only one block can be allocated meaning this has a fixed-size heap.
 package agc