Fix out-of-bound (OOB) input read in AES-XTS Decrypt in AVX-512 implementation

crypto/fipsmodule/aes/asm/aesni-xts-avx512.pl

@@ -2493,7 +2493,7 @@
   vmovdqu8 	 0x40($input),%zmm2
   vmovdqu8 	 0x80($input),%zmm3
   vmovdqu8 	 0xc0($input),%zmm4
-  vmovdqu8 	 0xf0($input),%zmm5
+  vmovdqu8 	 0xf0($input),%xmm5
   add 	 \$0x100,$input
 ___
   }

crypto/fipsmodule/modes/xts_test.cc

@@ -23,7 +23,9 @@
 #include "internal.h"
 #include "../../test/test_util.h"
 
-
+#if defined(OPENSSL_LINUX)
+#include <sys/mman.h>
+#endif
 struct XTSTestCase {
   const char *key_hex;
   const char *iv_hex;
@@ -995,8 +997,30 @@ static const XTSTestCase kXTSTestCases[] = {
     },
 };
 
+#if defined(OPENSSL_LINUX)
+static uint8_t *get_buffer_end(int pagesize) {
+  uint8_t *two_pages_p = (uint8_t *)mmap(NULL, 2*pagesize, PROT_READ|PROT_WRITE,
+                                      MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
+  EXPECT_TRUE(two_pages_p != NULL) << "mmap returned NULL.";
+
+  int ret = mprotect(two_pages_p + pagesize, pagesize, PROT_NONE);
+  EXPECT_TRUE(ret == 0) << "mprotect failed.";
+
+  return two_pages_p + pagesize;
+}
+
+static void free_memory(uint8_t *addr, int pagesize) {
+  munmap(addr - pagesize, 2 * pagesize);
+}
+#endif
+
 TEST(XTSTest, TestVectors) {
   unsigned test_num = 0;
+#if defined(OPENSSL_LINUX)
+  int pagesize = sysconf(_SC_PAGE_SIZE);
+  uint8_t *in_buffer_end = get_buffer_end(pagesize);
+  uint8_t *out_buffer_end = get_buffer_end(pagesize);
+#endif
   for (const auto &test : kXTSTestCases) {
     test_num++;
     SCOPED_TRACE(test_num);
@@ -1013,45 +1037,57 @@ TEST(XTSTest, TestVectors) {
     ASSERT_EQ(EVP_CIPHER_iv_length(cipher), iv.size());
     ASSERT_EQ(plaintext.size(), ciphertext.size());
 
+    int len;
+    uint8_t *in_p, *out_p;
+  #if defined(OPENSSL_LINUX)
+    ASSERT_GE(pagesize, (int)plaintext.size());
+    in_p = in_buffer_end - plaintext.size();
+    out_p = out_buffer_end - plaintext.size();
+    OPENSSL_memset(in_p, 0x00, plaintext.size());
+    OPENSSL_memset(out_p, 0x00, plaintext.size());
+  #else
+    std::unique_ptr<uint8_t[]> in(new uint8_t[plaintext.size()]);
+    std::unique_ptr<uint8_t[]> out(new uint8_t[plaintext.size()]);
+    in_p = in.get();
+    out_p = out.get();
+  #endif
+
     // Note XTS doesn't support streaming, so we only test single-shot inputs.
     for (bool in_place : {false, true}) {
       SCOPED_TRACE(in_place);
 
       // Test encryption.
-      bssl::Span<const uint8_t> in = plaintext;
-      std::vector<uint8_t> out(plaintext.size());
+
+      OPENSSL_memcpy(in_p, plaintext.data(), plaintext.size());
       if (in_place) {
-        out = plaintext;
-        in = out;
+        out_p = in_p;
       }
 
       bssl::ScopedEVP_CIPHER_CTX ctx;
       ASSERT_TRUE(EVP_EncryptInit_ex(ctx.get(), cipher, nullptr, key.data(),
                                      iv.data()));
-      int len;
       ASSERT_TRUE(
-          EVP_EncryptUpdate(ctx.get(), out.data(), &len, in.data(), in.size()));
-      out.resize(len);
-      EXPECT_EQ(Bytes(ciphertext), Bytes(out));
+          EVP_EncryptUpdate(ctx.get(), out_p, &len, in_p, plaintext.size()));
+      EXPECT_EQ(Bytes(ciphertext), Bytes(out_p, static_cast<size_t>(len)));
 
       // Test decryption.
-      in = ciphertext;
-      out.clear();
-      out.resize(plaintext.size());
-      if (in_place) {
-        out = ciphertext;
-        in = out;
+
+      if (!in_place) {
+        OPENSSL_memset(in_p, 0, len);
       }
 
       ctx.Reset();
       ASSERT_TRUE(EVP_DecryptInit_ex(ctx.get(), cipher, nullptr, key.data(),
                                      iv.data()));
       ASSERT_TRUE(
-          EVP_DecryptUpdate(ctx.get(), out.data(), &len, in.data(), in.size()));
-      out.resize(len);
-      EXPECT_EQ(Bytes(plaintext), Bytes(out));
+          EVP_DecryptUpdate(ctx.get(), in_p, &len, out_p, ciphertext.size()));
+      EXPECT_EQ(Bytes(plaintext), Bytes(in_p, static_cast<size_t>(len)));
     }
   }
+#if defined(OPENSSL_LINUX)
+  free_memory(in_buffer_end, pagesize);
+  free_memory(out_buffer_end, pagesize);
+#endif
 }
 
 // Negative test for key1 = key2

generated-src/linux-x86_64/crypto/fipsmodule/aesni-xts-avx512.S

@@ -3271,7 +3271,7 @@ aes_hw_xts_decrypt_avx512:
 	vmovdqu8	64(%rdi),%zmm2
 	vmovdqu8	128(%rdi),%zmm3
 	vmovdqu8	192(%rdi),%zmm4
-	vmovdqu8	240(%rdi),%zmm5
+	vmovdqu8	240(%rdi),%xmm5
 	addq	$0x100,%rdi
 	vpxorq	%zmm9,%zmm1,%zmm1
 	vpxorq	%zmm10,%zmm2,%zmm2

generated-src/mac-x86_64/crypto/fipsmodule/aesni-xts-avx512.S

@@ -3271,7 +3271,7 @@ L$_main_loop_run_16_amivrujEyduiFoi:
 	vmovdqu8	64(%rdi),%zmm2
 	vmovdqu8	128(%rdi),%zmm3
 	vmovdqu8	192(%rdi),%zmm4
-	vmovdqu8	240(%rdi),%zmm5
+	vmovdqu8	240(%rdi),%xmm5
 	addq	$0x100,%rdi
 	vpxorq	%zmm9,%zmm1,%zmm1
 	vpxorq	%zmm10,%zmm2,%zmm2

generated-src/win-x86_64/crypto/fipsmodule/aesni-xts-avx512.asm

@@ -3330,7 +3330,7 @@ $L$_main_loop_run_16_amivrujEyduiFoi:
 	vmovdqu8	zmm2,ZMMWORD[64+rcx]
 	vmovdqu8	zmm3,ZMMWORD[128+rcx]
 	vmovdqu8	zmm4,ZMMWORD[192+rcx]
-	vmovdqu8	zmm5,ZMMWORD[240+rcx]
+	vmovdqu8	xmm5,XMMWORD[240+rcx]
 	add	rcx,0x100
 	vpxorq	zmm1,zmm1,zmm9
 	vpxorq	zmm2,zmm2,zmm10