[crypto] Rework AES CTR/XTS streaming and squash heap churn (#2782)

AES Updates:
Replaced heap churn with stack scratch buffers tail handling now stays in-place, no more recursive transcode detours.
CTR/XTS modes read in larger, aligned chunks and still handle odd offsets cleanly.
XTS prefetches a few sectors ahead to reduce extra reads.
AesCtrStorage writer now uses the pooled buffer properly one stack slab, chunk forward, bump counter, repeat.
Result: less malloc noise, fewer watchdog spikes at startup (though mbedtls still sets the pace).
This should make the loading speed slightly better than before. Make sure to test.

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/2782
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
Reviewed-by: crueter <crueter@eden-emu.dev>
Co-authored-by: godpow <thesaviorsrule@yahoo.com>
Co-committed-by: godpow <thesaviorsrule@yahoo.com>

1 files changed, 75 insertions, 25 deletions

diff --git a/src/core/crypto/ctr_encryption_layer.cpp b/src/core/crypto/ctr_encryption_layer.cpp
index 6769754413..82473f57d7 100644
--- a/src/core/crypto/ctr_encryption_layer.cpp
+++ b/src/core/crypto/ctr_encryption_layer.cpp
@@ -5,6 +5,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
+#include <array>
 #include <cstring>
 #include "core/crypto/ctr_encryption_layer.h"
 
@@ -18,35 +19,84 @@ std::size_t CTREncryptionLayer::Read(u8* data, std::size_t length, std::size_t o
     if (length == 0)
         return 0;
 
+    constexpr std::size_t BlockSize = 0x10;
+    constexpr std::size_t MaxChunkSize = 0x10000;
+
     std::size_t total_read = 0;
-    // Handle an initial misaligned portion if needed.
-    if (auto const sector_offset = offset & 0xF; sector_offset != 0) {
-        const std::size_t aligned_off = offset - sector_offset;
-        std::array<u8, 0x10> block{};
-        if (auto const got = base->Read(block.data(), block.size(), aligned_off); got != 0) {
-            UpdateIV(base_offset + aligned_off);
-            cipher.Transcode(block.data(), got, block.data(), Op::Decrypt);
-            auto const to_copy = std::min<std::size_t>(length, got > sector_offset ? got - sector_offset : 0);
-            if (to_copy > 0) {
-                std::memcpy(data, block.data() + sector_offset, to_copy);
-                data += to_copy;
-                offset += to_copy;
-                length -= to_copy;
-                total_read += to_copy;
-            }
-        } else {
-            return 0;
+    auto* out = data;
+    std::size_t remaining = length;
+    std::size_t current_offset = offset;
+
+    const auto read_exact = [this](u8* dst, std::size_t bytes, std::size_t src_offset) {
+        std::size_t filled = 0;
+        while (filled < bytes) {
+            const std::size_t got = base->Read(dst + filled, bytes - filled, src_offset + filled);
+            if (got == 0)
+                break;
+            filled += got;
         }
+        return filled;
+    };
+
+    if (const std::size_t intra_block = current_offset & (BlockSize - 1); intra_block != 0) {
+        std::array<u8, BlockSize> block{};
+        const std::size_t aligned_offset = current_offset - intra_block;
+        const std::size_t got = read_exact(block.data(), BlockSize, aligned_offset);
+        if (got <= intra_block)
+            return total_read;
+
+        UpdateIV(base_offset + aligned_offset);
+        cipher.Transcode(block.data(), got, block.data(), Op::Decrypt);
+
+        const std::size_t available = got - intra_block;
+        const std::size_t to_copy = std::min<std::size_t>(remaining, available);
+        std::memcpy(out, block.data() + intra_block, to_copy);
+
+        out += to_copy;
+        current_offset += to_copy;
+        remaining -= to_copy;
+        total_read += to_copy;
+
+        if (to_copy != available)
+            return total_read;
     }
-    if (length > 0) {
-        // Now aligned to 0x10
-        UpdateIV(base_offset + offset);
-        const std::size_t got = base->Read(data, length, offset);
-        if (got > 0) {
-            cipher.Transcode(data, got, data, Op::Decrypt);
-            total_read += got;
-        }
+
+    while (remaining >= BlockSize) {
+        const std::size_t chunk_request = std::min<std::size_t>(remaining, MaxChunkSize);
+        const std::size_t aligned_request = chunk_request - (chunk_request % BlockSize);
+        if (aligned_request == 0)
+            break;
+
+        const std::size_t got = read_exact(out, aligned_request, current_offset);
+        if (got == 0)
+            break;
+
+        UpdateIV(base_offset + current_offset);
+        cipher.Transcode(out, got, out, Op::Decrypt);
+
+        out += got;
+        current_offset += got;
+        remaining -= got;
+        total_read += got;
+
+        if (got < aligned_request)
+            return total_read;
     }
+
+    if (remaining > 0) {
+        std::array<u8, BlockSize> block{};
+        const std::size_t got = read_exact(block.data(), BlockSize, current_offset);
+        if (got == 0)
+            return total_read;
+
+        UpdateIV(base_offset + current_offset);
+        cipher.Transcode(block.data(), got, block.data(), Op::Decrypt);
+
+        const std::size_t to_copy = std::min<std::size_t>(remaining, got);
+        std::memcpy(out, block.data(), to_copy);
+        total_read += to_copy;
+    }
+
     return total_read;
 }