Byteswap fixes

* Add byteswap test cases * Fix asm_parse.cpp to allow le* operations Signed-off-by: Dave Thaler <[email protected]>
vbpf · Dec 28, 2023 · 2d83ab9 · 2d83ab9
1 parent 81a834d
commit 2d83ab9
Show file tree

Hide file tree

Showing 4 changed files with 93 additions and 8 deletions.
diff --git a/src/asm_parse.cpp b/src/asm_parse.cpp
@@ -33,7 +33,7 @@ using crab::linear_expression_t;
 #define OPASSIGN R"_(\s*(\S*)=\s*)_"
 #define ASSIGN R"_(\s*=\s*)_"
 #define LONGLONG R"_(\s*(ll|)\s*)_"
-#define UNOP R"_((-|be16|be32|be64))_"
+#define UNOP R"_((-|be16|be32|be64|le16|le32|le64))_"
 
 #define PLUSMINUS R"_((\s*[+-])\s*)_"
 #define LPAREN R"_(\s*\(\s*)_"

diff --git a/src/asm_unmarshal.cpp b/src/asm_unmarshal.cpp
@@ -122,14 +122,15 @@ struct Unmarshaller {
                 note("arsh32 is not allowed");
             return Bin::Op::ARSH;
         case INST_ALU_OP_END:
+            if ((inst.opcode & INST_CLS_MASK) == INST_CLS_ALU64) {
+                std::string error_message =
+                    "invalid endian immediate " + std::to_string(inst.imm) + " for 64 bit instruction";
+                throw InvalidInstruction(pc, error_message.c_str());
+            }
             switch (inst.imm) {
             case 16: return (inst.opcode & INST_END_BE) ? Un::Op::BE16 : Un::Op::LE16;
-            case 32:
-                if ((inst.opcode & INST_CLS_MASK) == INST_CLS_ALU64)
-                    throw InvalidInstruction(pc, "invalid endian immediate 32 for 64 bit instruction");
-                return (inst.opcode & INST_END_BE) ? Un::Op::BE32 : Un::Op::LE32;
-            case 64:
-                return (inst.opcode & INST_END_BE) ? Un::Op::BE64 : Un::Op::LE64;
+            case 32: return (inst.opcode & INST_END_BE) ? Un::Op::BE32 : Un::Op::LE32;
+            case 64: return (inst.opcode & INST_END_BE) ? Un::Op::BE64 : Un::Op::LE64;
             default:
                 throw InvalidInstruction(pc, "invalid endian immediate");
             }

diff --git a/src/test/test_marshal.cpp b/src/test/test_marshal.cpp
@@ -47,7 +47,9 @@ static void check_unmarshal_fail(ebpf_inst inst, std::string expected_error_mess
     program_info info{.platform = &g_ebpf_platform_linux,
                       .type = g_ebpf_platform_linux.get_program_type("unspec", "unspec")};
     std::vector<ebpf_inst> insns = {inst};
-    std::string error_message = std::get<std::string>(unmarshal(raw_program{"", "", insns, info}));
+    auto result = unmarshal(raw_program{"", "", insns, info});
+    REQUIRE(std::holds_alternative<std::string>(result));
+    std::string error_message = std::get<std::string>(result);
     REQUIRE(error_message == expected_error_message);
 }
 
@@ -270,4 +272,6 @@ TEST_CASE("fail unmarshal", "[disasm][marshal]") {
     check_unmarshal_fail(ebpf_inst{.opcode = INST_CLS_JMP32}, "0: jump out of bounds\n");
     check_unmarshal_fail(ebpf_inst{.opcode = 0x90 | INST_CLS_JMP32}, "0: Bad instruction\n");
     check_unmarshal_fail(ebpf_inst{.opcode = 0x10 | INST_CLS_JMP32}, "0: jump out of bounds\n");
+    check_unmarshal_fail(ebpf_inst{.opcode = INST_ALU_OP_END | INST_CLS_ALU64, .imm = 0}, "0: invalid endian immediate 0 for 64 bit instruction\n");
+    check_unmarshal_fail(ebpf_inst{.opcode = INST_ALU_OP_END | INST_CLS_ALU64, .imm = 16}, "0: invalid endian immediate 16 for 64 bit instruction\n");
 }
diff --git a/test-data/unop.yaml b/test-data/unop.yaml
@@ -40,3 +40,83 @@ code:
     r1 = -r1
 
 post: ["r1.type=number", "r1.svalue=[-5, 5]"]
+---
+test-case: be16 singleton
+
+pre: ["r1.type=number", "r1.svalue=6636321", "r1.uvalue=6636321"]
+
+code:
+  <start>: |
+    r1 = be16 r1 ; 0x654321 -> 0x2143
+
+post: ["r1.type=number", "r1.svalue=8515", "r1.uvalue=8515"]
+---
+test-case: be32 singleton
+
+pre: ["r1.type=number", "r1.svalue=40926266145", "r1.uvalue=40926266145"]
+
+code:
+  <start>: |
+    r1 = be32 r1 ; 0x0987654321 -> 0x21436587
+
+post: ["r1.type=number", "r1.svalue=558065031", "r1.uvalue=558065031"]
+---
+test-case: be64 singleton
+
+pre: ["r1.type=number", "r1.svalue=40926266145", "r1.uvalue=40926266145"]
+
+code:
+  <start>: |
+    r1 = be64 r1 ; 0x0987654321 -> 0x2143658709000000
+
+post: ["r1.type=number", "r1.svalue=2396871057337221120", "r1.uvalue=2396871057337221120"]
+---
+test-case: le16 singleton
+
+pre: ["r1.type=number", "r1.svalue=6636321", "r1.uvalue=6636321"]
+
+code:
+  <start>: |
+    r1 = le16 r1 ; 0x654321 -> 0x4321
+
+post: ["r1.type=number", "r1.svalue=17185", "r1.uvalue=17185"]
+---
+test-case: le32 singleton
+
+pre: ["r1.type=number", "r1.svalue=40926266145", "r1.uvalue=40926266145"]
+
+code:
+  <start>: |
+    r1 = le32 r1 ; 0x0987654321 -> 0x87654321
+
+post: ["r1.type=number", "r1.svalue=2271560481", "r1.uvalue=2271560481"]
+---
+test-case: le64 singleton
+
+pre: ["r1.type=number", "r1.svalue=40926266145", "r1.uvalue=40926266145"]
+
+code:
+  <start>: |
+    r1 = le64 r1 ; 0x0987654321 -> 0x2143658709000000
+
+post: ["r1.type=number", "r1.svalue=40926266145", "r1.uvalue=40926266145"]
+---
+test-case: le16 range
+
+pre: ["r1.type=number", "r1.svalue=[0, 2]", "r1.uvalue=[0, 2]", "r1.svalue=r1.uvalue"]
+
+code:
+  <start>: |
+    r1 = le16 r1 ; this could preserve the range but we don't support that yet
+
+post: ["r1.type=number"]
+---
+test-case: be16 range
+
+pre: ["r1.type=number", "r1.svalue=[0, 2]", "r1.uvalue=[0, 2]", "r1.svalue=r1.uvalue"]
+
+code:
+  <start>: |
+    r1 = be16 r1 ; [0x0000, 0x0002] -> [0x0000, 0x2000] but currently we just lose the range
+
+post: ["r1.type=number"]