Merge branch 'io_ring'

.gitignore

@@ -3,5 +3,11 @@ file_hasher.ilk
 file_hasher.rdi
 file_hasher.exe
 file_hashes.txt
+Binaries/file_hashes.txt
 file_list.txt
 temp_code.c
+/.cache/clangd/index
+/file_hasher
+/io_uring_test
+/file_hasher
+/io_uring_test

README.md

@@ -1,24 +1,221 @@
 # filehasher
 
-Collects some metadata and hashes files.
+## Presentation  
+Collects some metadata and hashes files. It outputs the path, hash, size, creation and 
+last modification dates and the author in file_hasher.txt.  
+Creation and modification dates and author can be disabled in the config file.  
 
-## Building:  
+It is a high performance cross platform Windows and Linux compatible program, it uses:  
-### Windows:  
+  * Multiple threads for scanning and hashing (multi-threading can be disabled in the config file).  
-#### Release:  
+  * Stores the generated data in thread local configurable arenas that support growing 
-clang-cl /O3 file_hasher.c xxh_x86dispatch.c advapi32.lib  
+  by committing more memory and chaining blocks.  
-clang -O3 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
+  * Two Multi Producer Multi Consumer queues, one for the scanners and one between the scanners and hashers.  
-gcc -O3 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
+  * xxh3_128bits algorithm from xxhash, that supports SIMD instruction sets (SSE2, AVX2, AVX512)
+  and uses a runtime dispatcher to select the best available instruction set.  
+  * IO Ring for asynchronous I/O in Windows and the equivalent io_uring in Linux. 
+  The implementation is event driven, thread local, uses DMA and direct disk I/O, 
+  bypassing the OS cache completely, registered buffers (and registered files in io_uring), 
+  it supports bashing multiple submissions and can handle multiple files at the same time.  
+  It can be disabled in the config file.
+  * Fallback to buffered I/O if there is errors in the IO Ring path.  
 
-#### Debug:  
+## Building  
-clang-cl /Zi /Od file_hasher.c xxh_x86dispatch.c advapi32.lib  
+### Windows  
-clang -g -O0 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
+#### Release  
-gcc -g -O0 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
 
-### Linux:  
+**Note**: Make sur to use UCRT64 environment from MSYS2 instead of the standard MinGW environment. 
-#### Release:  
+UCRT64 uses the modern Universal C Runtime (ucrtbase.dll), which supports the newest APIs, 
-clang -O3 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
+the standard MSYS2 uses the legacy msvcrt.dll and does not support IO Ring.  
-gcc -O3 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
+To install:  
+pacman -S mingw-w64-ucrt-x86_64-gcc  
+pacman -S mingw-w64-ucrt-x86_64-clang
+pacman -Syu
+And add to path:  
+C:\msys64\ucrt64\bin  
 
-#### Debug:  
+gcc -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -o file_hasher  
-clang -g -O0 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
+clang -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -o file_hasher  
-gcc -g -O0 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
+clang-cl /O2 file_hasher.c xxhash.c xxh_x86dispatch.c  
+
+#### Debug  
+gcc -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -o file_hasher  
+clang -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -o file_hasher  
+clang-cl /Zi /Od file_hasher.c xxhash.c xxh_x86dispatch.c  
+
+### Linux  
+#### Release  
+gcc -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o file_hasher  
+clang -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o file_hasher  
+
+#### Debug  
+gcc -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o file_hasher  
+clang -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o file_hasher  
+
+## Notes about the IO Ring implementations  
+### IO Ring  
+
+#### File registration  
+Registering files is a performance optimization that allows the kernel to allocate an array 
+of descriptors/handles to pre-validate and maintain long-term references to file handles. 
+Instead of passing a standard file descriptor/handle with every I/O request, you pass a simple integer 
+index into a pre-registered table.  
+
+The Linux implementation has io_uring_register_files_scarse() to create an empty array of descriptors 
+(initialized with -1) without having to create and initialize it in the user space, and we can 
+use io_uring_register_files_update() to update one or more entries. Windows on the other hand 
+is limited to BuildIoRingRegisterFileHandles() only, so we need to re register the entire array of handles 
+each time. This is why there is a provided macro in config.h to disable or enable it.  
+
+##### Why Register Files? (The Benefits)  
+When you use a standard file descriptor in a high-frequency I/O loop, 
+the kernel must perform several "hidden" tasks for every single operation:
+  * Permission Checks: Validating that the process still has the right to read/write 
+  that specific file.  
+  * Reference Counting: Incrementing the file's internal reference count at the start of 
+  the I/O and decrementing it at the end to ensure the file isn't closed while in use.  
+  * Object Lookup: Traversing the internal "file descriptor table" to find the actual 
+  kernel object associated with your integer ID.  
+
+Registering the files performs these checks once at registration time. Subsequent 
+I/O operations skip these steps, significantly reducing CPU overhead and latency, 
+especially when handling thousands of small I/O operations per second.  
+
+##### Comparison: Linux vs. Windows Implementation  
+While both systems share the same core concept, their APIs and management styles differ significantly. 
+Feature 	Linux (io_uring)	Windows (IoRing)  
+API Call	io_uring_register	BuildIoRingRegisterFileHandles  
+Registration Method	Synchronous system call that blocks until the table is set up.	Asynchronous request submitted to the ring just like a read/write operation.
+Partial Updates	Supports IORING_REGISTER_FILES_UPDATE to swap specific indices without a full reset.	Does not support partial updates; a new registration call replaces the entire existing table.
+Memory Mapping	User must manually mmap() the queues into their address space.	The kernel handles memory mapping automatically when the ring is created.
+Scope of Operations	Extremely broad (files, sockets, timers, signals, even other rings).	Primarily focused on file storage (read, write, flush).
+
+#### Completion Wait count  
+To avoid busy waiting when receiving CQEs, we can use io_uring_submit_and_wait() in Linux by entering a wait count, 
+the threads sleeps until the count of CQEs are received, in windows the wait_count is present in SubmitIoRing() 
+but is not implemented yet, so we wait with a completion event for a single completion. Another limitation on the completion 
+event is that the kernel will waik up the thread only when receiving the first CQE, after that we need to drain the completion
+queue completely before sleeping again, or we enter an eternal slumber. And my config, each time the thread wakes up 
+it receives rarely more than 3 to 5 CQEs and most of the time only one CQE.
+
+#### Filtering CQEs  
+
+Unlike Linux, The Windows implementation treats buffer and file registration 
+as an asynchronous operation that we submit to the ring, similar to a read or write. 
+Those operations produce CQEs (completion queue entries) that we filter here using 
+cqe.UserData == USERDATA_REGISTER  
+```c
+    if (win_cqe.UserData == USERDATA_REGISTER)
+      continue;
+```
+
+### io_uring  
+
+#### Creation flags  
+io_uring provides a lot of configuration flags compared to IO Ring, some 
+of them are at the creation and others during the operations, here what 
+we use in this implementation at creation time and is lacking in the 
+IO Ring implementation.  
+
+  * IORING_SETUP_SINGLE_ISSUER: Since we are using a thread local io_uring, we can 
+  set this flag to remove the atomic operations.  
+  * IORING_SETUP_DEFER_TASKRUN: By default, the kernel sends an interrupts when a CQE 
+  is ready, we use this flag to disable this syscall and wait for a specific number of 
+  CQEs to be ready to group them, this reduces the number of syscall.  
+
+#### Memlock limit warning  
+
+```c
+     "WARNING: Buffer registration failed due to memlock limits (ENOMEM).\n"
+     "Increase the limit to solve this warning.\n");
+```
+
+The Memlock limit in Linux restricts the amount of memory a process can
+"lock" into physical RAM using the mlock() family of system calls. This
+prevents the operating system from swapping that memory out to disk.  
+And registering buffers will lock the buffers memory so the hardware
+can access it directly without kernel intervention and prevents the kernel from 
+swapping it to the SSD or HDD. Increase the limit to be able to register the buffers.  
+
+##### Modifying the Limit:  
+The method for changing the memlock limit depends on whether you are
+managing a user session or a system service.  
+1. For Users and Interactive Sessions  
+To permanently increase the limit for a specific user or group, modify
+the /etc/security/limits.conf file. Add the following lines:  
+
+```conf
+ # Example for a specific user (replace 'username'), unlimited or a custom value in KB
+ username   soft    memlock    unlimited
+ username   hard    memlock    unlimited
+
+ # Example for all users
+ *          soft    memlock    unlimited
+ *          hard    memlock    unlimited
+```
+
+Soft Limit: The value the user starts with; can be raised up to the
+hard limit.  
+
+Hard Limit: The absolute maximum; only a privileged user
+(root) can increase this. Values: Can be set in Kilobytes (KB) or as
+unlimited.  
+
+2. For Systemd Services
+Settings in limits.conf do not affect background services managed by
+systemd. To increase the limit for a service, edit its service file
+(e.g., /etc/systemd/system/myservice.service) and add:  
+```conf
+  [Service]
+  LimitMEMLOCK=infinity
+```
+
+##### Why Register Buffers?  
+In a standard "unregistered" I/O operation, the kernel must perform several 
+expensive steps for every single read or write:  
+  * Virtual-to-Physical Mapping: The kernel has to translate your application's 
+  virtual memory addresses into physical RAM addresses.  
+  * Page Pinning: The kernel must "pin" the memory pages (using get_user_pages) 
+  to prevent them from being swapped to disk or moved while the hardware 
+  (like your SSD) is writing to them.  
+  * TLB Overhead: Constant mapping and unmapping put pressure on the Translation 
+  Lookaside Buffer (TLB), which can slow down the CPU.  
+
+Registering the buffers performs all of this "pinning" and "mapping" once.  
+
+#### Direct I/O: O_DIRECT (Linux) and FILE_FLAG_NO_BUFFERING (Windows)
+
+Modern operating systems normally use a page cache when reading files. This means file 
+data is first loaded into kernel memory and then copied to user space. While this improves 
+performance for many workloads, it introduces extra memory usage and copy overhead.  
+
+Both Linux and Windows provide a way to bypass this cache and perform direct I/O:  
+
+Linux: O_DIRECT
+Windows: FILE_FLAG_NO_BUFFERING
+
+These flags instruct the OS to transfer data directly between disk and user-provided buffers, avoiding the page cache.
+
+##### Benefits
+1. Reduced memory overhead
+Avoids polluting the OS page cache
+Especially useful for large sequential reads (e.g. hashing, backups)
+2. Lower CPU usage
+Eliminates extra memory copies between kernel and user space
+3. Predictable performance
+No interference from cache eviction or readahead heuristics
+More consistent throughput for streaming workloads
+4. Better scalability
+Ideal for high-throughput, multi-threaded I/O pipelines
+Prevents cache contention between threads
+5. Avoids double caching
+Important when the application already manages its own buffering
+
+##### File system compatibility
+Not all file systems are compatible with O_DIRECT, if we try to open files residing in an NTFS partition, 
+most of the time it will fail, and some times it opens but the CQEs return with an error code bad 
+descriptor, and it causes some lags.
+
+To address this issue the program falls back to sequential read when the open fails, and falls back to
+buffered sequential hashing if we receive an error in the CQEs. There is also a file system detection 
+that we can enable in the config file, it will enable the collection of the file system in scan_folder() 
+and the file will be opened accordingly, but it costs one additional syscall / directory.

arena.c

@@ -437,12 +437,14 @@ void *arena_push(mem_arena **arena_ptr, u64 size, bool zero) { // mk push
      Commit memory if needed
      ------------------------------------------------------------ */
 
-  if (local_post > selected->commit_pos) {
+  if (local_post > selected->commit_pos -
-    u64 new_commit = ALIGN_UP_POW2(local_post, arena_pagesize());
+                       ALIGN_UP_POW2(sizeof(mem_arena), selected->align)) {
+    u64 new_commit = ALIGN_UP_POW2(local_post + ALIGN_UP_POW2(sizeof(mem_arena), selected->align), arena_pagesize());
     new_commit = MIN(new_commit, selected->reserve_size);
 
     if (!plat_mem_commit((u8 *)selected + selected->commit_pos,
                          new_commit - selected->commit_pos)) {
+        printf("ERROR: Could not commit memory!\n");
       return NULL;
     }
 

base.h

@@ -1,9 +1,49 @@
 #pragma once
 
+#define _CRT_SECURE_NO_WARNINGS
+
+#if defined(_WIN32) || defined(_WIN64)
+
+#if defined(_MSC_VER)
+#pragma comment(lib, "advapi32.lib")
+#endif
+
+#include <aclapi.h>
+#include <fcntl.h>
+#include <io.h>
+#include <ioringapi.h> // Needs to be included before stdatomic to avoid errors
+#include <ntioring_x.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <windows.h>
+#include <winerror.h>
+
+#elif defined(__linux__)
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+
+#include <dirent.h>
+#include <fcntl.h>
+#include <liburing.h>
+#include <poll.h>
+#include <pthread.h>
+#include <pwd.h>
+#include <sys/eventfd.h>
+#include <sys/resource.h>
+#include <sys/stat.h>
+#include <sys/vfs.h>
+#include <unistd.h>
+#endif
+
 #include <assert.h>
+#include <ctype.h>
 #include <immintrin.h>
+#include <limits.h>
 #include <stdatomic.h>
 #include <stdbool.h>
+#include <stddef.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -11,25 +51,6 @@
 #include <time.h>
 #include <errno.h>
 
-#if defined(_WIN32) || defined(_WIN64)
-#define PLATFORM_WINDOWS 1
-#include <aclapi.h>
-#include <fcntl.h>
-#include <io.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <windows.h>
-
-#define strdup _strdup
-#else
-#include <dirent.h>
-#include <fcntl.h>
-#include <pthread.h>
-#include <pwd.h>
-#include <sys/stat.h>
-#include <unistd.h>
-#endif
-
 /* ------------------------------------------------------------
    Base types
    ------------------------------------------------------------ */

binaries/changelog.txt

@@ -14,7 +14,7 @@ v3.2: Making the lock free MPMC queue growable
 Add padding to avoir false sharing
 Add sleep() and SwitchToThread() to limit spinning
 
-v3.3: Fix bug slots used before initialization,compare and swap is protecting updating committed, but it is not protecting the memory initialization. Adding atomic_flag commit_lock to protect against that
+v3.3: Fix bug slots used before initialization, compare and swap is protecting updating committed, but it is not protecting the memory initialization. Adding atomic_flag commit_lock to protect against that
 Fix bug multiple threads committing at the same time, fixed by using atomic_flag commit_lock and re-checking committed after acquiring the lock
 Reorder helper functions
 
@@ -49,3 +49,10 @@ Fixing user prompt parsing
 4.5: Porting to linux
 Reorganising the code
 Improving the scan function
+
+5.0: Implementing the IO Ring for windows and ui_uring for linux instead of buffered hashing, huge performance gains. The IO Ring is event driven, thread local, uses DMA and direct disk I/O, bypassing the OS cache completely, registered buffers (and registered files in io_uring), it supports bashing multiple submissions and can handle multiple files at the same time.
+Hashing small files using XXH3_128bits() instead of the streaming pipeline(XXH3_128bits_reset(), XXH3_128bits_update(), XXH3_128bits_digest()), this reduses the overhead of creating a state and digest, coupled with the IO Ring it improves the hashing of small files whose size is inferior to the size of IO Ring buffers
+fixing the xxh_x86dispatch warnings
+Updating the progress printing function
+Implementing a config file
+Writing the README file

config.h

@@ -0,0 +1,31 @@
+
+#define FILE_HASHES_TXT "file_hashes.txt"
+
+// Metadata selection
+#define FILE_TIMES 1 // created and modified time
+#define FILE_OWNER 1
+
+#define MULTI_THREADING 1
+#define READ_BLOCK KiB(64)
+
+// -------------------- IO Ring Configuration ----------------------
+#define USE_IORING 1
+
+#if USE_IORING
+#define IORING_BUFFER_SIZE KiB(256)
+#define NUM_BUFFERS_PER_THREAD 32
+#define MAX_ACTIVE_FILES 16
+
+#define SUBMIT_TIMEOUT_MS 10000
+#define IORING_DEBUG_PRINTS 0
+#define IORING_DEBUG_STATS 0
+
+#if defined(_WIN32) || defined(_WIN64)
+#define USE_REGISTERED_FILES 1
+
+#elif defined(__linux__)
+#define USE_REGISTERED_FILES 1
+#define CHECK_FILE_SYSTEM 0
+
+#endif
+#endif

file_hasher.c

@@ -74,21 +74,44 @@ int main(int argc, char **argv) {
   mem_arena *gp_arena = arena_create(&params);
 
   // -------------------------------
-  // Detect hardware threads
+  // Detect hardware
   // -------------------------------
   // --- Windows: detect PHYSICAL cores (not logical threads) ---
-  size_t hw_threads = platform_physical_cores();
+  uint32_t cpu_cores = platform_physical_cores();
 
   // Logical threads = CPU cores * 2
-  size_t num_threads = hw_threads * 2;
+  uint32_t cpu_threads = cpu_cores * 2;
 
-  printf("Starting thread pool: %zu threads (CPU cores: %zu)\n", num_threads,
+#if MULTI_THREADING
-         hw_threads);
+  uint32_t num_scan_threads = cpu_threads;
-  printf("  Selected instruction set: %s\n", get_xxhash_instruction_set());
+  uint32_t num_hash_threads = cpu_threads;
+
+  printf("%d cores %d threads CPU detected with %s instruction set\n"
+         "Starting thread pool: %d scanning and %d hashing threads\n",
+         cpu_cores, cpu_threads, get_xxhash_instruction_set(), num_scan_threads,
+         num_hash_threads);
+#else
+  uint32_t num_scan_threads = 1;
+  uint32_t num_hash_threads = 1;
+
+  printf(
+      "%d cores %d threads CPU detected with %s instruction set\n"
+      "Starting thread pool: %d scanning and %d hashing threads(Debug mode)\n",
+      cpu_cores, cpu_threads, get_xxhash_instruction_set(), num_scan_threads,
+      num_hash_threads);
+
+#endif
+
+  // Align IO Ring block size to the system page size
+#if USE_IORING
+  g_ioring_buffer_size = ALIGN_UP_POW2(g_ioring_buffer_size, g_pagesize);
+#endif
 
   // -------------------------------
   // Scanning and hashing
   // -------------------------------
+
+  // test_io_ring();
   MPMCQueue dir_queue;
   mpmc_init(&dir_queue, MiB(1));
 
@@ -96,8 +119,6 @@ int main(int argc, char **argv) {
   mpmc_init(&file_queue, MiB(1));
 
   // Starting hash threads
-  size_t num_hash_threads = num_threads;
-
   WorkerContext workers[num_hash_threads];
   Thread *hash_threads =
       arena_push(&gp_arena, sizeof(Thread) * num_hash_threads, true);
@@ -106,8 +127,14 @@ int main(int argc, char **argv) {
     workers[i].arena = arena_create(&params);
     workers[i].file_queue = &file_queue;
 
+#if USE_IORING
+    if (thread_create(&hash_threads[i], (ThreadFunc)hash_worker_ioring,
+                      &workers[i]) != 0)
+#else
     if (thread_create(&hash_threads[i], (ThreadFunc)hash_worker, &workers[i]) !=
-        0) {
+        0)
+#endif
+    {
       fprintf(stderr, "Failed to create hash thread %zu\n", i);
       exit(1);
     }
@@ -122,8 +149,6 @@ int main(int argc, char **argv) {
   }
 
   // Starting scan threads
-  size_t num_scan_threads = num_threads;
-
   ScannerContext scanners[num_scan_threads];
   Thread *scan_threads =
       arena_push(&gp_arena, sizeof(Thread) * num_scan_threads, true);
@@ -197,7 +222,7 @@ int main(int argc, char **argv) {
 
   FILE *f = fopen(FILE_HASHES_TXT, "wb");
 
-  for (int i = 0; i < num_threads; i++) {
+  for (int i = 0; i < num_hash_threads; i++) {
     mem_arena *arena = workers[i].arena;
     u8 *arena_base =
         (u8 *)arena + ALIGN_UP_POW2(sizeof(mem_arena), arena->align);
@@ -209,6 +234,15 @@ int main(int argc, char **argv) {
   // -------------------------------
   // Print summary
   // -------------------------------
+#if USE_IORING
+  uint64_t incomplete = atomic_load(&g_io_ring_fallbacks);
+  if (incomplete > 0) {
+    printf("\nWARNING: I/O Ring incomplete files: %llu (fallback to buffered "
+           "I/O used)\n",
+           (unsigned long long)incomplete);
+  }
+#endif
+
   double total_seconds = timer_elapsed(&total_timer);
 
   printf("Completed hashing %zu files\n", total_found);

io_ring_test.c

@@ -0,0 +1,147 @@
+#pragma once
+
+#include <ioringapi.h>
+#include <ntioring_x.h>
+// #include "ioringapi.c"
+#include <winerror.h>
+
+// Initialize I/O Ring
+HIORING io_ring_init(void) {
+
+  // if (!io_ring_load_functions()) {
+  //   printf("[I/O Ring] Failed to load functions\n");
+  //   return NULL;
+  // }
+
+  IORING_CAPABILITIES caps;
+  ZeroMemory(&caps, sizeof(caps));
+
+  HRESULT hr = QueryIoRingCapabilities(&caps);
+  if (FAILED(hr)) {
+    printf("[I/O Ring] QueryIoRingCapabilities failed: 0x%08lx\n", hr);
+    return NULL;
+  }
+
+  // printf("[I/O Ring] MaxVersion=%d, MaxSubmission=%u, MaxCompletion=%u\n",
+  //        (int)caps.MaxVersion, caps.MaxSubmissionQueueSize,
+  // caps.MaxCompletionQueueSize);
+
+  if (caps.MaxVersion < IORING_VERSION_1) {
+    printf("[I/O Ring] Version too old\n");
+    return NULL;
+  }
+
+  IORING_CREATE_FLAGS flags = {0};
+  HIORING ring = NULL;
+
+  // hr = CreateIoRing(IORING_VERSION_1, flags, 256, 512, &ring);
+  hr = CreateIoRing(caps.MaxVersion, flags, 256, 512, &ring);
+  if (FAILED(hr)) {
+    printf("[I/O Ring] CreateIoRing failed: 0x%08lx\n", hr);
+    return NULL;
+  }
+
+  // printf("[I/O Ring] Created successfully\n");
+
+  // Check if read operation is supported
+
+  // HRESULT io_ring_support = IsIoRingOpSupported(ring, IORING_OP_READ);
+  // if (io_ring_support == S_FALSE) {
+  //   printf("[I/O Ring] Not supported, %ld /n", io_ring_support);
+  // }
+
+  // Get ring info
+  IORING_INFO info;
+  ZeroMemory(&info, sizeof(info));
+  GetIoRingInfo(ring, &info);
+  // printf("[I/O Ring] Submission: %u, Completion: %u\n",
+  // info.SubmissionQueueSize, info.CompletionQueueSize);
+
+  return ring;
+}
+
+void io_ring_cleanup(HIORING ring) {
+  if (ring) {
+    CloseIoRing(ring);
+    // printf("[I/O Ring] Closed\n");
+  }
+}
+
+// Read file using I/O Ring
+int io_ring_read_file(HIORING ring, HANDLE hFile, void *buffer, DWORD size,
+                      UINT64 offset) {
+
+  IORING_HANDLE_REF file_ref = IoRingHandleRefFromHandle(hFile);
+  IORING_BUFFER_REF buf_ref = IoRingBufferRefFromPointer(buffer);
+
+  HRESULT hr = BuildIoRingReadFile(ring, file_ref, buf_ref, size, offset,
+                                   (UINT_PTR)buffer, IOSQE_FLAGS_NONE);
+
+  if (FAILED(hr))
+    return -1;
+
+  UINT32 submitted = 0;
+  hr = SubmitIoRing(ring, 1, INFINITE, &submitted);
+  if (FAILED(hr) || submitted == 0)
+    return -1;
+
+  for (;;) {
+    IORING_CQE cqe;
+    hr = PopIoRingCompletion(ring, &cqe);
+
+    if (FAILED(hr))
+      continue;
+
+    if (cqe.UserData != (UINT_PTR)buffer)
+      continue;
+
+    if (FAILED(cqe.ResultCode))
+      return -1;
+
+    return (int)cqe.Information;
+  }
+}
+
+// Test function
+void test_io_ring(void) {
+  printf("\n=== Testing I/O Ring ===\n");
+
+  HIORING ring = io_ring_init();
+  if (!ring) {
+    printf("I/O Ring not available\n");
+    return;
+  }
+
+  // Create test file
+  HANDLE hFile = CreateFileA("test.txt", GENERIC_READ | GENERIC_WRITE, 0, NULL,
+                             CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);
+  if (hFile != INVALID_HANDLE_VALUE) {
+    char test_data[] =
+        "Hello, I/O Ring! This is a test of the Windows I/O Ring API.";
+    DWORD written;
+    WriteFile(hFile, test_data, sizeof(test_data), &written, NULL);
+    CloseHandle(hFile);
+  }
+
+  // Read using I/O Ring
+  hFile = CreateFileA("test.txt", GENERIC_READ, FILE_SHARE_READ, NULL,
+                      OPEN_EXISTING, FILE_FLAG_OVERLAPPED, NULL);
+  if (hFile != INVALID_HANDLE_VALUE) {
+    char buffer[512] = {0};
+    int bytes = io_ring_read_file(ring, hFile, buffer, sizeof(buffer), 0);
+    if (bytes > 0) {
+      printf("Read %d bytes: %s\n", bytes, buffer);
+    } else {
+      printf("Failed to read file\n");
+    }
+    CloseHandle(hFile);
+  } else {
+    printf("Failed to open test file\n");
+  }
+
+  // Cleanup
+  DeleteFileA("test.txt");
+  io_ring_cleanup(ring);
+
+  printf("=== Test complete ===\n\n");
+}

io_uring_test.c

@@ -0,0 +1,397 @@
+/*
+# Compile
+gcc -o io_uring_test io_uring_test.c -luring
+
+# Run
+./io_uring_test
+   */
+#include "base.h"
+#include <stdint.h>
+#define _GNU_SOURCE
+#include <errno.h>
+#include <fcntl.h>
+#include <liburing.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+#define BUFFER_SIZE 4096
+#define NUM_BUFFERS 4
+#define NUM_REGISTERED_FILES 3 // Test with 3 files
+
+// Colors for output
+#define COLOR_GREEN "\033[0;32m"
+#define COLOR_RED "\033[0;31m"
+#define COLOR_YELLOW "\033[0;33m"
+#define COLOR_BLUE "\033[0;34m"
+#define COLOR_RESET "\033[0m"
+
+// Test result tracking
+typedef struct {
+  int passed;
+  int failed;
+} TestResults;
+
+static void print_success(const char *step) {
+  printf(COLOR_GREEN "[✓] SUCCESS: %s" COLOR_RESET "\n", step);
+}
+
+static void print_failure(const char *step, const char *error) {
+  printf(COLOR_RED "[✗] FAILED: %s - %s" COLOR_RESET "\n", step, error);
+}
+
+static void print_info(const char *msg) {
+  printf(COLOR_BLUE "[i] INFO: %s" COLOR_RESET "\n", msg);
+}
+
+static void print_step(const char *step) {
+  printf(COLOR_YELLOW "\n>>> Testing: %s" COLOR_RESET "\n", step);
+}
+
+static int create_test_file(const char *filename, const char *content) {
+  FILE *f = fopen(filename, "w");
+  if (!f) {
+    perror("Failed to create test file");
+    return -1;
+  }
+
+  fprintf(f, "%s", content);
+  fclose(f);
+
+  printf("  Created test file: %s\n", filename);
+  return 0;
+}
+
+// Test 1: Create io_uring instance
+static int test_io_uring_create(struct io_uring *ring, TestResults *results) {
+  print_step("io_uring creation");
+
+  int ret = io_uring_queue_init(256, ring, 0);
+  if (ret < 0) {
+    print_failure("io_uring_queue_init", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("io_uring instance created");
+  results->passed++;
+  return 0;
+}
+
+// Test 2: Register buffers
+static int test_register_buffers(struct io_uring *ring, void **buffers,
+                                 struct iovec *iovs, TestResults *results) {
+  print_step("Buffer registration");
+
+  size_t total_size = BUFFER_SIZE * NUM_BUFFERS;
+  *buffers = aligned_alloc(4096, total_size);
+  if (!*buffers) {
+    print_failure("Buffer allocation", strerror(errno));
+    results->failed++;
+    return -1;
+  }
+
+  for (int i = 0; i < NUM_BUFFERS; i++) {
+    iovs[i].iov_base = (char *)*buffers + (i * BUFFER_SIZE);
+    iovs[i].iov_len = BUFFER_SIZE;
+    memset(iovs[i].iov_base, 0, BUFFER_SIZE);
+  }
+
+  int ret = io_uring_register_buffers(ring, iovs, NUM_BUFFERS);
+  if (ret < 0) {
+    print_failure("io_uring_register_buffers", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Buffers registered successfully");
+  results->passed++;
+  return 0;
+}
+
+// Test 3: Register files sparse (empty table)
+static int test_register_files_sparse(struct io_uring *ring, unsigned nr_files,
+                                      TestResults *results) {
+  print_step("Sparse file registration (empty table)");
+
+  int ret = io_uring_register_files_sparse(ring, nr_files);
+  if (ret < 0) {
+    if (ret == -EINVAL) {
+      print_info(
+          "io_uring_register_files_sparse not supported (kernel < 5.19)");
+      print_info("Trying regular file registration with invalid fds...");
+
+      // Fallback: register with invalid fds
+      int *invalid_fds = calloc(nr_files, sizeof(int));
+      if (!invalid_fds) {
+        print_failure("Allocating invalid fds array", "Out of memory");
+        results->failed++;
+        return -1;
+      }
+
+      for (int i = 0; i < nr_files; i++) {
+        invalid_fds[i] = -1; // Mark all as invalid
+      }
+
+      ret = io_uring_register_files(ring, invalid_fds, nr_files);
+      free(invalid_fds);
+
+      if (ret < 0) {
+        print_failure("Regular file registration also failed", strerror(-ret));
+        results->failed++;
+        return -1;
+      }
+      print_success("File table registered (regular, with invalid fds)");
+    } else {
+      print_failure("io_uring_register_files_sparse", strerror(-ret));
+      results->failed++;
+      return -1;
+    }
+  } else {
+    printf("  Registered empty file table with %u slots\n", nr_files);
+    print_success("Sparse file table created");
+  }
+
+  results->passed++;
+  return 0;
+}
+
+// Test 4: Update file slot and read from it
+static int test_file_read_loop(struct io_uring *ring, struct iovec *iovs,
+                               const char **filenames,
+                               const char **expected_contents, int num_files,
+                               TestResults *results) {
+  print_step("File slot update and read loop");
+
+  int *fds = calloc(num_files, sizeof(int));
+  if (!fds) {
+    print_failure("Allocating fd array", "Out of memory");
+    results->failed++;
+    return -1;
+  }
+
+  // Open all files first
+  for (int i = 0; i < num_files; i++) {
+    fds[i] = open(filenames[i], O_RDONLY);
+    if (fds[i] < 0) {
+      print_failure("Opening file", filenames[i]);
+      results->failed++;
+      // Close already opened files
+      for (int j = 0; j < i; j++)
+        close(fds[j]);
+      free(fds);
+      return -1;
+    }
+    printf("  Opened %s (fd=%d)\n", filenames[i], fds[i]);
+  }
+
+  // Test loop: update slot, submit read, verify
+  for (int slot = 0; slot < num_files; slot++) {
+    printf("\n  --- Testing slot %d with file '%s' ---\n", slot,
+           filenames[slot]);
+
+    // Update the file registration for this slot
+    printf("  Updating slot %d with fd %d...\n", slot, fds[slot]);
+    int ret = io_uring_register_files_update(ring, slot, &fds[slot], 1);
+
+    if (ret < 0) {
+      print_failure("File registration update", strerror(-ret));
+      results->failed++;
+      continue;
+    }
+    printf("  Slot update result: %d (expected 1)\n", ret);
+
+    // Get file size for read size calculation
+    struct stat st;
+    if (fstat(fds[slot], &st) != 0) {
+      print_failure("fstat", strerror(errno));
+      results->failed++;
+      continue;
+    }
+
+    size_t file_size = st.st_size;
+    size_t read_size = BUFFER_SIZE;
+
+    // Adjust read size for O_DIRECT if needed
+    int page_size = plat_get_pagesize();
+    if (read_size > file_size) {
+      read_size = ALIGN_UP_POW2(file_size, page_size);
+    }
+
+    printf("  File size: %zu, read size: %zu\n", file_size, read_size);
+
+    // Clear buffer for this test
+    memset(iovs[0].iov_base, 0, BUFFER_SIZE);
+
+    // Submit read using registered file
+    struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
+    if (!sqe) {
+      print_failure("Getting SQE", "No available SQE");
+      results->failed++;
+      continue;
+    }
+
+    // Use slot index with fixed file flag
+    io_uring_prep_read_fixed(sqe, slot, iovs[0].iov_base, read_size, 0, 0);
+    sqe->flags |= IOSQE_FIXED_FILE;
+    io_uring_sqe_set_data64(sqe, 100 + slot); // Unique user_data per slot
+
+    ret = io_uring_submit(ring);
+    if (ret < 0) {
+      print_failure("Submitting read", strerror(-ret));
+      results->failed++;
+      continue;
+    }
+    printf("  Submitted read (1 SQE)\n");
+
+    // Wait for completion
+    struct io_uring_cqe *cqe;
+    ret = io_uring_wait_cqe(ring, &cqe);
+    if (ret < 0) {
+      print_failure("Waiting for completion", strerror(-ret));
+      results->failed++;
+      continue;
+    }
+
+    // Process completion
+    uint64_t user_data = io_uring_cqe_get_data64(cqe);
+    int bytes_read = cqe->res;
+
+    printf("  Completion: user_data=%lu, result=%d\n", (unsigned long)user_data,
+           bytes_read);
+
+    if (bytes_read < 0) {
+      print_failure("Read operation", strerror(-bytes_read));
+      results->failed++;
+      io_uring_cqe_seen(ring, cqe);
+      continue;
+    }
+
+    if (user_data != 100 + slot) {
+      print_failure("User data mismatch", "Wrong user_data value");
+      results->failed++;
+      io_uring_cqe_seen(ring, cqe);
+      continue;
+    }
+
+    // Verify the data
+    char *data = (char *)iovs[0].iov_base;
+    printf("  Data read (%d bytes): %.*s\n", bytes_read,
+           bytes_read < 100 ? bytes_read : 100, data);
+
+    if (strstr(data, expected_contents[slot]) == NULL) {
+      print_failure("Data verification",
+                    "Expected content not found in read data");
+      results->failed++;
+    } else {
+      print_success("Data verified successfully");
+      results->passed++;
+    }
+
+    io_uring_cqe_seen(ring, cqe);
+
+    // Invalidate the slot after use (mark as -1)
+    int invalid_fd = -1;
+    ret = io_uring_register_files_update(ring, slot, &invalid_fd, 1);
+    if (ret < 0) {
+      printf("  Warning: Could not invalidate slot %d: %s\n", slot,
+             strerror(-ret));
+    }
+  }
+
+  // Close all files
+  for (int i = 0; i < num_files; i++) {
+    if (fds[i] >= 0)
+      close(fds[i]);
+  }
+  free(fds);
+
+  return 0;
+}
+
+int main() {
+  TestResults results = {0, 0};
+  struct io_uring ring;
+  void *buffers = NULL;
+  struct iovec iovs[NUM_BUFFERS];
+
+  printf(COLOR_BLUE "\n========================================\n");
+  printf("    io_uring Sparse File Registration Test\n");
+  printf("========================================\n" COLOR_RESET);
+
+  // Define test files and their content
+  const char *filenames[] = {"test_file_0.txt", "test_file_1.txt",
+                             "test_file_2.txt"};
+
+  const char *contents[] = {
+      "This is file 0: Hello World! The quick brown fox jumps over the lazy "
+      "dog.",
+      "This is file 1: io_uring is awesome for async I/O operations!",
+      "This is file 2: Testing sparse file registration with multiple files."};
+
+  const char *expected_substrings[] = {"Hello World", "io_uring is awesome",
+                                       "sparse file registration"};
+
+  int num_files = 3;
+
+  // Create all test files
+  print_info("Creating test files...");
+  for (int i = 0; i < num_files; i++) {
+    if (create_test_file(filenames[i], contents[i]) != 0) {
+      return 1;
+    }
+  }
+
+  // Test 1: Create io_uring
+  if (test_io_uring_create(&ring, &results) != 0) {
+    goto cleanup_files;
+  }
+
+  // Test 2: Register buffers
+  if (test_register_buffers(&ring, &buffers, iovs, &results) != 0) {
+    io_uring_queue_exit(&ring);
+    goto cleanup_files;
+  }
+
+  // Test 3: Register empty file table (sparse)
+  if (test_register_files_sparse(&ring, num_files, &results) != 0) {
+    io_uring_unregister_buffers(&ring);
+    free(buffers);
+    io_uring_queue_exit(&ring);
+    goto cleanup_files;
+  }
+
+  // Test 4: Loop through files, update slots, read and verify
+  test_file_read_loop(&ring, iovs, filenames, expected_substrings, num_files,
+                      &results);
+
+  // Cleanup
+  io_uring_unregister_files(&ring);
+  io_uring_unregister_buffers(&ring);
+  free(buffers);
+  io_uring_queue_exit(&ring);
+
+cleanup_files:
+  // Remove test files
+  for (int i = 0; i < num_files; i++) {
+    remove(filenames[i]);
+  }
+
+  // Print summary
+  int total = results.passed + results.failed;
+  printf(COLOR_BLUE "\n========================================\n");
+  printf("           TEST SUMMARY\n");
+  printf("========================================\n" COLOR_RESET);
+  printf("  Total tests: %d\n", total);
+  printf(COLOR_GREEN "  Passed: %d\n" COLOR_RESET, results.passed);
+  if (results.failed > 0) {
+    printf(COLOR_RED "  Failed: %d\n" COLOR_RESET, results.failed);
+    printf(COLOR_RED "\n  ✗ SOME TESTS FAILED!\n" COLOR_RESET);
+  } else {
+    printf(COLOR_GREEN "\n  ✓ ALL TESTS PASSED!\n" COLOR_RESET);
+  }
+
+  return results.failed > 0 ? 1 : 0;
+}

ioringapi.c

@@ -0,0 +1,285 @@
+#pragma once
+#include <stdio.h>
+#include <windows.h>
+#include <winnt.h>
+
+// Forward declarations
+typedef struct IORING_HANDLE_REF IORING_HANDLE_REF;
+typedef struct IORING_BUFFER_REF IORING_BUFFER_REF;
+typedef void *HIORING;
+
+/* --------------------- Types declaration --------------------- */
+typedef enum IORING_CREATE_ADVISORY_FLAGS {
+  IORING_CREATE_ADVISORY_FLAGS_NONE,
+  IORING_CREATE_SKIP_BUILDER_PARAM_CHECKS
+} IORING_CREATE_ADVISORY_FLAGS;
+// Specifies advisory flags for creating an I/O ring with a call to
+// CreateIoRing.
+
+typedef enum IORING_CREATE_REQUIRED_FLAGS {
+  IORING_CREATE_REQUIRED_FLAGS_NONE
+} IORING_CREATE_REQUIRED_FLAGS;
+// Specifies required flags for creating an I/O ring with a call to
+// CreateIoRing.
+
+typedef enum IORING_REF_KIND {
+  IORING_REF_RAW = 0,
+  IORING_REF_REGISTERED = 1,
+} IORING_REF_KIND;
+// Specifies the type of an IORING_HANDLE_REF structure.
+
+typedef enum IORING_SQE_FLAGS {
+  IOSQE_FLAGS_NONE,
+  IOSQE_FLAGS_DRAIN_PRECEDING_OPS
+} IORING_SQE_FLAGS;
+// Specifies kernel behavior options for I/O ring submission queue entries
+
+// IORING_REGISTERED_BUFFER structure
+typedef struct IORING_REGISTERED_BUFFER {
+  UINT32 Index;
+  UINT32 Offset;
+} IORING_REGISTERED_BUFFER;
+
+// IORING_HANDLE_REF
+struct IORING_HANDLE_REF {
+  IORING_REF_KIND Kind;
+  union {
+    HANDLE Handle;
+    UINT32 Index;
+  } HandleUnion;
+};
+// Represents a reference to a file handle used in an I/O ring operation
+
+// IORING_BUFFER_REF
+struct IORING_BUFFER_REF {
+  IORING_REF_KIND Kind;
+  union {
+    void *Address;
+    IORING_REGISTERED_BUFFER IndexAndOffset;
+  } BufferUnion;
+};
+
+typedef struct IORING_BUFFER_INFO {
+  void *Address;
+  UINT32 Length;
+} IORING_BUFFER_INFO;
+
+// IORING_BUFFER_REF represents a reference to a buffer used in an I/O ring
+// operation
+
+// IORING_VERSION enumeration
+typedef enum IORING_VERSION {
+  IORING_VERSION_INVALID = 0,
+  IORING_VERSION_1 = 1,
+  IORING_VERSION_2 = 2,
+  IORING_VERSION_3 = 3,
+  IORING_VERSION_4 = 4,
+} IORING_VERSION;
+
+typedef enum IORING_FEATURE_FLAGS {
+  IORING_FEATURE_FLAGS_NONE = 0,
+  IORING_FEATURE_UM_EMULATION = 1
+} IORING_FEATURE_FLAGS;
+
+// IORING_CAPABILITIES structure
+typedef struct IORING_CAPABILITIES {
+  IORING_VERSION MaxVersion;
+  UINT32 MaxSubmissionQueueSize;
+  UINT32 MaxCompletionQueueSize;
+  IORING_FEATURE_FLAGS FeatureFlags;
+} IORING_CAPABILITIES;
+// Represents the IORING API capabilities.
+
+// IORING_CQE structure
+typedef struct IORING_CQE {
+  UINT_PTR UserData;
+  HRESULT ResultCode;
+  ULONG_PTR Information;
+} IORING_CQE;
+// Represents a completed I/O ring queue entry.
+
+// IORING_CREATE_FLAGS structure
+typedef struct IORING_CREATE_FLAGS {
+  IORING_CREATE_REQUIRED_FLAGS Required;
+  IORING_CREATE_ADVISORY_FLAGS Advisory;
+} IORING_CREATE_FLAGS;
+// Specifies flags for creating an I/O ring with a call to CreateIoRing.
+
+// IORING_INFO structure
+typedef struct IORING_INFO {
+  IORING_VERSION IoRingVersion;
+  IORING_CREATE_FLAGS Flags;
+  UINT32 SubmissionQueueSize;
+  UINT32 CompletionQueueSize;
+} IORING_INFO;
+// Represents the shape and version information for the specified I/O ring
+
+// IORING_OP_CODE for IsIoRingOpSupported
+typedef enum IORING_OP_CODE {
+  IORING_OP_NOP = 0,
+  IORING_OP_READ = 1,
+  IORING_OP_WRITE = 2,
+  IORING_OP_FLUSH = 3,
+  IORING_OP_REGISTER_BUFFERS = 4,
+  IORING_OP_REGISTER_FILES = 5,
+  IORING_OP_CANCEL = 6,
+} IORING_OP_CODE;
+
+/* --------------------- Dynamic loader --------------------- */
+// Function pointer types
+typedef BOOL(WINAPI *IsIoRingOpSupported_t)(HIORING, IORING_OP_CODE);
+typedef HRESULT(WINAPI *QueryIoRingCapabilities_t)(IORING_CAPABILITIES *);
+typedef HRESULT(WINAPI *GetIoRingInfo_t)(HIORING, IORING_INFO *);
+typedef HRESULT(WINAPI *CreateIoRing_t)(IORING_VERSION, IORING_CREATE_FLAGS,
+                                        UINT32, UINT32, HIORING *);
+typedef HRESULT(WINAPI *CloseIoRing_t)(HIORING);
+typedef HRESULT(WINAPI *SubmitIoRing_t)(HIORING, UINT32, UINT32, UINT32 *);
+typedef HRESULT(WINAPI *PopIoRingCompletion_t)(HIORING, IORING_CQE *);
+typedef HRESULT(WINAPI *SetIoRingCompletionEvent_t)(HIORING, HANDLE);
+typedef HRESULT(WINAPI *BuildIoRingCancelRequest_t)(HIORING, IORING_HANDLE_REF,
+                                                    UINT_PTR, UINT_PTR);
+typedef HRESULT(WINAPI *BuildIoRingReadFile_t)(HIORING, IORING_HANDLE_REF,
+                                               IORING_BUFFER_REF, UINT32,
+                                               UINT64, UINT_PTR,
+                                               IORING_SQE_FLAGS);
+typedef HRESULT(WINAPI *BuildIoRingRegisterBuffers_t)(
+    HIORING, UINT32, IORING_BUFFER_INFO const[], UINT_PTR);
+
+typedef HRESULT(WINAPI *BuildIoRingRegisterFileHandles_t)(HIORING, UINT32,
+                                                          HANDLE const[],
+                                                          UINT_PTR);
+
+// Core:
+// Queries the support of the specified operation for the specified I/O ring
+static IsIoRingOpSupported_t IsIoRingOpSupported = NULL;
+
+// Queries the OS for the supported capabilities for IORINGs
+static QueryIoRingCapabilities_t QueryIoRingCapabilities = NULL;
+
+// Gets information about the API version and queue sizes of an I/O ring
+static GetIoRingInfo_t GetIoRingInfo = NULL;
+
+// Creates a new instance of an I/O ring submission/completion queue pair and
+// returns a handle for referencing the I/O ring
+static CreateIoRing_t CreateIoRing = NULL;
+
+// Closes an HIORING handle that was previously opened with a call to
+// CreateIoRing
+static CloseIoRing_t CloseIoRing = NULL;
+
+// Submission / completion:
+// Submits all constructed but not yet submitted entries to the kernel’s queue
+// and optionally waits for a set of operations to complete
+static SubmitIoRing_t SubmitIoRing = NULL;
+
+// Pops a single entry from the completion queue, if one is available
+static PopIoRingCompletion_t PopIoRingCompletion = NULL;
+
+// Registers a completion queue event with an IORING
+static SetIoRingCompletionEvent_t SetIoRingCompletionEvent = NULL;
+
+// Operations:
+// Performs an asynchronous read from a file using an I/O ring
+static BuildIoRingReadFile_t BuildIoRingReadFile = NULL;
+
+// Attempts to cancel a previously submitted I/O ring operation
+static BuildIoRingCancelRequest_t BuildIoRingCancelRequest = NULL;
+
+// Registers an array of buffers with the system for future I/O ring operations
+static BuildIoRingRegisterBuffers_t BuildIoRingRegisterBuffers = NULL;
+
+// Registers an array of file handles with the system for future I/O ring
+// operations
+static BuildIoRingRegisterFileHandles_t BuildIoRingRegisterFileHandles = NULL;
+
+static int io_ring_loaded = 0;
+
+static int io_ring_load_functions(void) {
+  if (io_ring_loaded)
+    return 1;
+
+  HMODULE hKernel = GetModuleHandleW(L"kernel32.dll");
+  if (!hKernel)
+    return 0;
+
+  IsIoRingOpSupported =
+      (IsIoRingOpSupported_t)GetProcAddress(hKernel, "IsIoRingOpSupported");
+  QueryIoRingCapabilities = (QueryIoRingCapabilities_t)GetProcAddress(
+      hKernel, "QueryIoRingCapabilities");
+  GetIoRingInfo = (GetIoRingInfo_t)GetProcAddress(hKernel, "GetIoRingInfo");
+  CreateIoRing = (CreateIoRing_t)GetProcAddress(hKernel, "CreateIoRing");
+  CloseIoRing = (CloseIoRing_t)GetProcAddress(hKernel, "CloseIoRing");
+  SubmitIoRing = (SubmitIoRing_t)GetProcAddress(hKernel, "SubmitIoRing");
+  PopIoRingCompletion =
+      (PopIoRingCompletion_t)GetProcAddress(hKernel, "PopIoRingCompletion");
+  SetIoRingCompletionEvent = (SetIoRingCompletionEvent_t)GetProcAddress(
+      hKernel, "SetIoRingCompletionEvent");
+  BuildIoRingReadFile =
+      (BuildIoRingReadFile_t)GetProcAddress(hKernel, "BuildIoRingReadFile");
+  BuildIoRingCancelRequest = (BuildIoRingCancelRequest_t)GetProcAddress(
+      hKernel, "BuildIoRingCancelRequest");
+  BuildIoRingRegisterBuffers = (BuildIoRingRegisterBuffers_t)GetProcAddress(
+      hKernel, "BuildIoRingRegisterBuffers");
+  BuildIoRingRegisterFileHandles =
+      (BuildIoRingRegisterFileHandles_t)GetProcAddress(
+          hKernel, "BuildIoRingRegisterFileHandles");
+
+  io_ring_loaded =
+      (IsIoRingOpSupported && QueryIoRingCapabilities && CreateIoRing &&
+       CloseIoRing && SubmitIoRing && PopIoRingCompletion &&
+       SetIoRingCompletionEvent && BuildIoRingReadFile &&
+       BuildIoRingCancelRequest && BuildIoRingRegisterBuffers &&
+       BuildIoRingRegisterFileHandles);
+
+  if (io_ring_loaded)
+    printf("[I/O Ring] Functions loaded\n");
+  else
+    printf("[I/O Ring] Some functions not available\n");
+
+  return io_ring_loaded;
+}
+
+/* ------------- Standard helper functions definition ------------- */
+// Creates an instance of the IORING_BUFFER_REF structure with the provided
+// buffer index and offset
+static inline IORING_BUFFER_REF
+IoRingBufferRefFromIndexAndOffset(UINT32 index, UINT32 offset) {
+  IORING_BUFFER_REF ref;
+  ref.Kind = IORING_REF_REGISTERED;
+  ref.BufferUnion.IndexAndOffset.Index = index;
+  ref.BufferUnion.IndexAndOffset.Offset = offset;
+  return ref;
+}
+
+// Creates an instance of the IORING_BUFFER_REF structure from the provided
+// pointer
+static IORING_BUFFER_REF IoRingBufferRefFromPointer(void *addr) {
+  IORING_BUFFER_REF ref;
+  ref.Kind = IORING_REF_RAW;
+  ref.BufferUnion.Address = addr;
+  return ref;
+}
+
+// Creates an instance of the IORING_HANDLE_REF structure from the provided file
+// handle
+static IORING_HANDLE_REF IoRingHandleRefFromHandle(HANDLE h) {
+  IORING_HANDLE_REF ref;
+  ref.Kind = IORING_REF_RAW;
+  ref.HandleUnion.Handle = h;
+  return ref;
+}
+
+// Creates an instance of the IORING_HANDLE_REF structure from the provided
+// index
+static inline IORING_HANDLE_REF IoRingHandleRefFromIndex(UINT32 index) {
+  IORING_HANDLE_REF ref;
+  ref.Kind = IORING_REF_REGISTERED; // MUST be registered
+  ref.HandleUnion.Index = index;
+  return ref;
+}
+
+// NOTE: If you are using index-based buffers or handles, make sure you have
+// successfully called BuildIoRingRegisterBuffers or
+// BuildIoRingRegisterFileHandles first so the kernel has a valid table to look
+// into, otherwise the kernel will treat the index as an invalid memory
+// address/handle.