github: remove sdiv crash on old armv7

.github/workflows/build.yml

@@ -100,12 +100,17 @@ jobs:
           export PKG_CONFIG_PATH=$DEPS_DIR/lib/pkgconfig
           export STAGING_DIR=$RUNNER_TEMP
           OPTIMIZE=-Oz
+          SYSMALLOC=-DLUAJIT_USE_SYSMALLOC
           case "$ARCH" in
             lexra)
               OPTIMIZE=-Os
               ;;
             arm)
-              CPU="-mcpu=cortex-a7 -mthumb"
+              CPU="-mcpu=arm1176jzf-s -mthumb"
+              ;;
+            arm64|mips64)
+              # not safe without GC64
+              SYSMALLOC=
               ;;
           esac
           MINSIZE="$OPTIMIZE $MINSIZE"
@@ -131,9 +136,10 @@ jobs:
               *)
                 HOSTCC="cc -m32"
             esac
+            echo ARCH=$ARCH SYSMALLOC=$SYSMALLOC
             (
               cd luajit2-*
-              make BUILDMODE=static XCFLAGS="-DLUAJIT_USE_SYSMALLOC -DLUAJIT_DISABLE_FFI" HOST_CC="$HOSTCC" CROSS= CC="$CC" TARGET_AR="$AR rcus" TARGET_STRIP=$STRIP TARGET_CFLAGS="$CPU $MINSIZE $CFLAGS" TARGET_LDFLAGS="$CPU $LDMINSIZE $LDFLAGS" -j$(nproc)
+              make BUILDMODE=static XCFLAGS="$SYSMALLOC -DLUAJIT_DISABLE_FFI" HOST_CC="$HOSTCC" CROSS= CC="$CC" TARGET_AR="$AR rcus" TARGET_STRIP=$STRIP TARGET_CFLAGS="$CPU $MINSIZE $CFLAGS" TARGET_LDFLAGS="$CPU $LDMINSIZE $LDFLAGS" -j$(nproc)
               make install PREFIX= DESTDIR=$DEPS_DIR
             )
             LJIT=1
@@ -236,10 +242,16 @@ jobs:
           export RANLIB=$TOOLCHAIN/bin/llvm-ranlib
           export STRIP=$TOOLCHAIN/bin/llvm-strip
           export PKG_CONFIG_PATH=$DEPS_DIR/lib/pkgconfig
+          SYSMALLOC=-DLUAJIT_USE_SYSMALLOC
           case "$ABI" in
             armeabi-v7a)
               CPU="-mthumb"
               ;;
+            arm64-v8a)
+              # not safe without GC64
+              SYSMALLOC=
+              PAGESIZE="-Wl,-z,max-page-size=16384"
+              ;;
           esac
 
           # luajit
@@ -253,7 +265,7 @@ jobs:
           esac
           (
             cd luajit2-*
-            make BUILDMODE=static XCFLAGS="-DLUAJIT_USE_SYSMALLOC -DLUAJIT_DISABLE_FFI" HOST_CC="$HOSTCC" CROSS= CC="$CC" TARGET_AR="$AR rcus" TARGET_STRIP=$STRIP TARGET_CFLAGS="$CPU $MINSIZE $CFLAGS" TARGET_LDFLAGS="$LDMINSIZE $LDFLAGS" -j$(nproc)
+            make BUILDMODE=static XCFLAGS="$SYSMALLOC -DLUAJIT_DISABLE_FFI" HOST_CC="$HOSTCC" CROSS= CC="$CC" TARGET_AR="$AR rcus" TARGET_STRIP=$STRIP TARGET_CFLAGS="$CPU $MINSIZE $CFLAGS" TARGET_LDFLAGS="$LDMINSIZE $LDFLAGS" -j$(nproc)
             make install PREFIX= DESTDIR=$DEPS_DIR
           )
           LJIT=1
@@ -279,7 +291,7 @@ jobs:
 
           # zapret2
           CFLAGS="-DZAPRET_GH_VER=${{ github.ref_name }} -DZAPRET_GH_HASH=${{ github.sha }} -I$DEPS_DIR/include $CPU" \
-          LDFLAGS="-L$DEPS_DIR/lib" \
+          LDFLAGS="-L$DEPS_DIR/lib $PAGESIZE" \
           make -C zapret2 LUA_JIT=$LJIT LUA_CFLAGS="$LCFLAGS" LUA_LIB="$LLIB" -j$(nproc) android
 
           # strip unwanted ELF sections to prevent warnings on old Android versions

docs/changes.txt

@@ -266,3 +266,10 @@ v0.9.4.3
 * github, linux-builder: reduce arm executable size by 20% - move to armv7+thumb
 * init.d: warn if hostlist/ipset files are inside zapret2 root
 * zapret-lib: do not call apply_arg_prefix in apply_execution_plan - call it right before instance execute
+
+0.9.4.5
+
+* github: rollback to lj_alloc in luajit for arm64 and mips64
+* github: use 16K page size for android arm64 build
+* nfqws2: join fragments in quic CRYPTO reconstruction. allow intersections.
+

docs/compile/builder-linux/build_deps.sh

@@ -62,11 +62,20 @@ build_lua()
 build_luajit()
 {
 	local CFL="$CFLAGS"
+	local SYSMALLOC=
 (
 	cd luajit2-*
 	CFLAGS="-Os"
 	make clean
-	make BUILDMODE=static XCFLAGS="-DLUAJIT_USE_SYSMALLOC -DLUAJIT_DISABLE_FFI $CFLAGS_PIC" HOST_CC="$HOST_CC" CROSS= CC="$CC" TARGET_AR="$AR rcus" TARGET_STRIP=$STRIP TARGET_CFLAGS="$OPTIMIZE $MINSIZE $CFL" TARGET_LDFLAGS="$CPU $LDMINSIZE $LDFLAGS"
+	case $TARGET in
+		aarch64*|mips64*)
+			# sysmalloc can cause troubles without GC64. GC64 slows down by 10-15%. better not to use sysmalloc and leave lj_alloc.
+			;;
+		*)
+			# save some exe size
+			SYSMALLOC=-DLUAJIT_USE_SYSMALLOC
+	esac
+	make BUILDMODE=static XCFLAGS="$SYSMALLOC -DLUAJIT_DISABLE_FFI $CFLAGS_PIC" HOST_CC="$HOST_CC" CROSS= CC="$CC" TARGET_AR="$AR rcus" TARGET_STRIP=$STRIP TARGET_CFLAGS="$OPTIMIZE $MINSIZE $CFL" TARGET_LDFLAGS="$CPU $LDMINSIZE $LDFLAGS"
 	make install PREFIX= DESTDIR="$STAGING_DIR"
 )
 }

lua/zapret-antidpi.lua

@@ -389,7 +389,8 @@ function syndata(ctx, desync)
 			dis.payload = blob(desync, desync.arg.blob, "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00")
 			apply_fooling(desync, dis)
 			if desync.arg.tls_mod then
-				dis.payload = tls_mod_shim(desync, dis.payload, desync.arg.tls_mod, nil)
+				local pl = tls_mod_shim(desync, dis.payload, desync.arg.tls_mod, nil)
+				if pl then dis.payload = pl end
 			end
 			if b_debug then DLOG("syndata: "..hexdump_dlog(dis.payload)) end
 			if rawsend_dissect_ipfrag(dis, desync_opts(desync)) then
@@ -449,7 +450,8 @@ function fake(ctx, desync)
 			end
 			local fake_payload = blob(desync, desync.arg.blob)
 			if desync.reasm_data and desync.arg.tls_mod then
-				fake_payload = tls_mod_shim(desync, fake_payload, desync.arg.tls_mod, desync.reasm_data)
+				local pl = tls_mod_shim(desync, fake_payload, desync.arg.tls_mod, desync.reasm_data)
+				if pl then fake_payload = pl end
 			end
 			-- check debug to save CPU
 			if b_debug then DLOG("fake: "..hexdump_dlog(fake_payload)) end

lua/zapret-obfs.lua

@@ -175,7 +175,12 @@ end
 -- arg : server=[0|1] - override server mode. by default use "--server" nfqws2 parameter
 function udp2icmp(ctx, desync)
 	local dataxor
-	local bserver = desync.arg.server and (desync.arg.server~="0") or b_server
+	local bserver
+	if desync.arg.server then
+		bserver = desync.arg.server~="0"
+	else
+		bserver = b_server
+	end
 
 	local function one_byte_arg(name)
 		if desync.arg[name] then

nfq2/Makefile

@@ -14,6 +14,7 @@ CFLAGS_UBSAN = -fsanitize=undefined,alignment -fno-sanitize-recover=undefined,al
 LDFLAGS_PIE = -pie
 LDFLAGS += -flto=auto -Wl,--gc-sections $(LDFLAGS_PIE)
 LDFLAGS_ANDROID = -llog
+LDFLAGS_CYGWIN = -Wl,--build-id=none
 STRIPP=-s
 LIBS =
 LIBS_LINUX = -lz -lnetfilter_queue -lnfnetlink -lmnl -lm
@@ -157,9 +158,9 @@ bsd: $(SRC_FILES)
 	$(CC) $(STRIPP) $(CFLAGS) $(LUA_CFL) $(CFLAGS_BSD) -o dvtws2 $(SRC_FILES) $(LIBS) $(LUA_LIB) $(LIBS_BSD) $(LDFLAGS)
 
 cygwin64:
-	$(CC) $(STRIPP) $(CFLAGS) $(LUA_CFL) $(CFLAGS_CYGWIN) $(CFLAGS_CYGWIN64) -o winws2 $(SRC_FILES) $(RES_CYGWIN64) $(LIBS) $(LUA_LIB) $(LIBS_CYGWIN) $(LIBS_CYGWIN64) $(LDFLAGS)
+	$(CC) $(STRIPP) $(CFLAGS) $(LUA_CFL) $(CFLAGS_CYGWIN) $(CFLAGS_CYGWIN64) -o winws2 $(SRC_FILES) $(RES_CYGWIN64) $(LIBS) $(LUA_LIB) $(LIBS_CYGWIN) $(LIBS_CYGWIN64) $(LDFLAGS) $(LDFLAGS_CYGWIN)
 cygwin32:
-	$(CC) $(STRIPP) $(CFLAGS) $(LUA_CFL) $(CFLAGS_CYGWIN) $(CFLAGS_CYGWIN32) -o winws2 $(SRC_FILES) $(RES_CYGWIN32) $(LIBS) $(LUA_LIB) $(LIBS_CYGWIN) $(LIBS_CYGWIN32) $(LDFLAGS)
+	$(CC) $(STRIPP) $(CFLAGS) $(LUA_CFL) $(CFLAGS_CYGWIN) $(CFLAGS_CYGWIN32) -o winws2 $(SRC_FILES) $(RES_CYGWIN32) $(LIBS) $(LUA_LIB) $(LIBS_CYGWIN) $(LIBS_CYGWIN32) $(LDFLAGS) $(LDFLAGS_CYGWIN)
 cygwin: cygwin64
 
 clean:

nfq2/helpers.c

@@ -50,7 +50,6 @@ void qsort_ssize_t(ssize_t *array, int ct)
 	qsort(array, ct, sizeof(*array), cmp_ssize_t);
 }
 
-
 int str_index(const char **strs, int count, const char *str)
 {
 	for (int i = 0; i < count; i++)

nfq2/helpers.h

@@ -12,6 +12,9 @@
 #include <fcntl.h>
 
 #define UNARY_PLUS(v) (v>0 ? "+" : "")
+//#define	MIN(v1,v2) ((v1)<(v2) ? (v1) : (v2))
+//#define	MAX(v1,v2) ((v1)<(v2) ? (v2) : (v1))
+
 
 // this saves memory. sockaddr_storage is larger than required. it can be 128 bytes. sockaddr_in6 is 28 bytes.
 typedef union

nfq2/lua.c

@@ -3350,7 +3350,6 @@ static int luacall_tls_mod(lua_State *L)
 	int argc=lua_gettop(L);
 
 	size_t fake_tls_len;
-	bool bRes;
 	const uint8_t *fake_tls = (uint8_t*)lua_reqlstring(L,1,&fake_tls_len);
 	const char *modlist = lua_reqstring(L,2);
 
@@ -3370,8 +3369,10 @@ static int luacall_tls_mod(lua_State *L)
 		uint8_t *newtls = lua_newuserdata(L, maxlen);
 
 		memcpy(newtls, fake_tls, newlen);
-		bRes = TLSMod(&mod, payload, payload_len, newtls, &newlen, maxlen);
-		lua_pushlstring(L,(char*)newtls,newlen);
+		if (TLSMod(&mod, payload, payload_len, newtls, &newlen, maxlen))
+			lua_pushlstring(L,(char*)newtls,newlen);
+		else
+			lua_pushnil(L);
 
 		lua_remove(L,-2);
 	}
@@ -3379,11 +3380,9 @@ static int luacall_tls_mod(lua_State *L)
 	{
 		// no mod. push it back
 		lua_pushlstring(L,(char*)fake_tls,fake_tls_len);
-		bRes = true;
 	}
-	lua_pushboolean(L, bRes);
 
-	LUA_STACK_GUARD_RETURN(L,2)
+	LUA_STACK_GUARD_RETURN(L,1)
 }
 
 struct userdata_zs

nfq2/protocol.c

@@ -649,9 +649,11 @@ bool TLSAdvanceToHostInSNI(const uint8_t **ext, size_t *elen, size_t *slen)
 	// u8	data+2 - server name type. 0=host_name
 	// u16	data+3 - server name length
 	if (*elen < 5 || (*ext)[2] != 0) return false;
+	uint16_t nll = pntoh16(*ext);
 	*slen = pntoh16(*ext + 3);
+	if (nll<(*slen+3) || *slen > *elen-5) return false;
 	*ext += 5; *elen -= 5;
-	return *slen <= *elen;
+	return true;
 }
 static bool TLSExtractHostFromExt(const uint8_t *ext, size_t elen, char *host, size_t len_host)
 {
@@ -1252,6 +1254,13 @@ static int cmp_range64(const void * a, const void * b)
 {
 	return (((struct range64*)a)->offset < ((struct range64*)b)->offset) ? -1 : (((struct range64*)a)->offset > ((struct range64*)b)->offset) ? 1 : 0;
 }
+/*
+static bool intersected_u64(uint64_t l1, uint64_t r1, uint64_t l2, uint64_t r2)
+{
+	return l1 <= r2 && l2 <= r1;
+}
+*/
+
 bool QUICDefragCrypto(const uint8_t *clean,size_t clean_len, uint8_t *defrag,size_t *defrag_len, bool *bFull)
 {
 	// Crypto frame can be split into multiple chunks
@@ -1265,7 +1274,7 @@ bool QUICDefragCrypto(const uint8_t *clean,size_t clean_len, uint8_t *defrag,siz
 	uint64_t offset,sz,szmax=0,zeropos=0,pos=0;
 	bool found=false;
 	struct range64 ranges[MAX_DEFRAG_PIECES];
-	int i,range=0;
+	int i,j,range=0;
 
 	while(pos<clean_len)
 	{
@@ -1287,24 +1296,54 @@ bool QUICDefragCrypto(const uint8_t *clean,size_t clean_len, uint8_t *defrag,siz
 			if ((pos+sz)>clean_len) return false;
 
 			if ((offset+sz)>defrag_data_len) return false; // defrag buf overflow
+
+			// remove exact duplicates early to save cpu
+			for(i=0;i<range;i++)
+				if (ranges[i].offset==offset && ranges[i].len==sz)
+					goto skip_range;
+
 			if (zeropos < offset)
 				// make sure no uninitialized gaps exist in case of not full fragment coverage
 				memset(defrag_data+zeropos,0,offset-zeropos);
 			if ((offset+sz) > zeropos)
 				zeropos=offset+sz;
-			memcpy(defrag_data+offset,clean+pos,sz);
-			if ((offset+sz) > szmax) szmax = offset+sz;
 
 			found=true;
-			pos+=sz;
-
+			if ((offset+sz) > szmax) szmax = offset+sz;
+			memcpy(defrag_data+offset,clean+pos,sz);
 			ranges[range].offset = offset;
 			ranges[range].len = sz;
 			range++;
+skip_range:
+			pos+=sz;
 		}
 	}
 	if (found)
 	{
+		qsort(ranges, range, sizeof(*ranges), cmp_range64);
+
+//		for(i=0 ; i<range ; i++)
+//			printf("range1 %llu-%llu\n",ranges[i].offset,ranges[i].offset+ranges[i].len);
+
+		if (range>0)
+		{
+			for (j=0,i=1; i < range; i++)
+			{
+				uint64_t current_end = ranges[j].offset + ranges[j].len;
+				uint64_t next_start = ranges[i].offset;
+				uint64_t next_end = ranges[i].offset + ranges[i].len;
+
+				if (next_start <= current_end)
+					ranges[j].len = MAX(next_end,current_end) - ranges[j].offset;
+				else
+					ranges[++j] = ranges[i];
+			}
+			range = j+1;
+		}
+
+//		for(i=0 ; i<range ; i++)
+//			printf("range2 %llu-%llu\n",ranges[i].offset,ranges[i].offset+ranges[i].len);
+
 		defrag[0] = 6;
 		defrag[1] = 0; // offset
 		// 2..9 - length 64 bit
@@ -1313,21 +1352,7 @@ bool QUICDefragCrypto(const uint8_t *clean,size_t clean_len, uint8_t *defrag,siz
 		defrag[2] |= 0xC0; // 64 bit value
 		*defrag_len = (size_t)(szmax+10);
 
-		qsort(ranges, range, sizeof(*ranges), cmp_range64);
-
-		//for(i=0 ; i<range ; i++)
-		//	printf("RANGE %zu len %zu\n",ranges[i].offset,ranges[i].len);
-
-		for(i=0,offset=0,*bFull=true ; i<range ; i++)
-		{
-			if (ranges[i].offset!=offset)
-			{
-				*bFull = false;
-				break;
-			}
-			offset += ranges[i].len;
-		}
-
+		*bFull = range==1 && !ranges[0].offset;
 		//printf("bFull=%u\n",*bFull);
 	}
 	return found;
@@ -1350,6 +1375,8 @@ bool IsQUICInitial(const uint8_t *data, size_t len)
 	if (data[offset] > QUIC_MAX_CID_LENGTH) return false;
 	offset += 1 + data[offset];
 
+	if (offset>=len) return false;
+
 	// SCID
 	if (data[offset] > QUIC_MAX_CID_LENGTH) return false;
 	offset += 1 + data[offset];