true async io in linux

config/kernel-ki_complete.m4

@@ -0,0 +1,26 @@
+dnl # SPDX-License-Identifier: CDDL-1.0
+dnl #
+dnl # 5.19 API change,
+dnl # kiocb->ki_complete() reduced from 3 args to 2:
+dnl #   old: void (*ki_complete)(struct kiocb *, long, long)
+dnl #   new: void (*ki_complete)(struct kiocb *, long)
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_KIOCB_KI_COMPLETE], [
+	ZFS_LINUX_TEST_SRC([kiocb_ki_complete_2args], [
+		#include <linux/fs.h>
+	],[
+		struct kiocb *kiocb = NULL;
+		kiocb->ki_complete(kiocb, 0);
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_KIOCB_KI_COMPLETE], [
+	AC_MSG_CHECKING([whether kiocb->ki_complete() wants 2 args])
+	ZFS_LINUX_TEST_RESULT([kiocb_ki_complete_2args], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_2ARGS_KI_COMPLETE, 1,
+		    [kiocb->ki_complete() wants 2 args])
+	],[
+		AC_MSG_RESULT(no)
+	])
+])

config/kernel.m4

@@ -96,6 +96,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
 	ZFS_AC_KERNEL_SRC_VFS_REMAP_FILE_RANGE
 	ZFS_AC_KERNEL_SRC_VFS_CLONE_FILE_RANGE
 	ZFS_AC_KERNEL_SRC_VFS_DEDUPE_FILE_RANGE
+	ZFS_AC_KERNEL_SRC_KIOCB_KI_COMPLETE
 	ZFS_AC_KERNEL_SRC_KMAP_ATOMIC_ARGS
 	ZFS_AC_KERNEL_SRC_KMAP_LOCAL_PAGE
 	ZFS_AC_KERNEL_SRC_FOLLOW_DOWN_ONE
@@ -220,6 +221,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
 	ZFS_AC_KERNEL_VFS_REMAP_FILE_RANGE
 	ZFS_AC_KERNEL_VFS_CLONE_FILE_RANGE
 	ZFS_AC_KERNEL_VFS_DEDUPE_FILE_RANGE
+	ZFS_AC_KERNEL_KIOCB_KI_COMPLETE
 	ZFS_AC_KERNEL_KMAP_ATOMIC_ARGS
 	ZFS_AC_KERNEL_KMAP_LOCAL_PAGE
 	ZFS_AC_KERNEL_FOLLOW_DOWN_ONE

include/os/linux/Makefile.am

@@ -23,6 +23,7 @@ kernel_sysdir = $(kerneldir)/sys
 kernel_sys_HEADERS = \
 	%D%/zfs/sys/abd_os.h \
 	%D%/zfs/sys/abd_impl_os.h \
+	%D%/zfs/sys/dmu_direct_os.h \
 	%D%/zfs/sys/policy.h \
 	%D%/zfs/sys/trace_acl.h \
 	%D%/zfs/sys/trace_arc.h \

include/os/linux/zfs/sys/dmu_direct_os.h

@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: CDDL-1.0
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or https://opensource.org/licenses/CDDL-1.0.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ *
+ * Copyright 2026, tiehexue <tiehexue@hotmail.com>. All rights reserved.
+ *
+ */
+
+#ifndef	_SYS_DMU_OS_H
+#define	_SYS_DMU_OS_H
+
+#include <sys/dmu.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Async Direct I/O completion callback type (shared by read and write).
+ */
+typedef void (dmu_abd_done_func_t)(void *arg, int error);
+
+/*
+ * Async Direct I/O read.  Submits reads via the ZIO pipeline and returns
+ * immediately.  The completion callback fires from ZIO taskq context when
+ * all reads finish.  Caller retains ownership of 'data' until callback.
+ */
+int dmu_read_abd_async(dnode_t *dn, uint64_t offset, uint64_t size,
+    abd_t *data, dmu_flags_t flags,
+    dmu_abd_done_func_t *done, void *done_arg);
+
+/*
+ * Async Direct I/O write.  Submits writes via the ZIO pipeline and returns
+ * immediately.  The completion callback fires from ZIO taskq context when
+ * all writes finish.  Caller retains ownership of 'data' until callback
+ * and must commit the transaction (tx) from the callback.
+ */
+int dmu_write_abd_async(dnode_t *dn, uint64_t offset, uint64_t size,
+    abd_t *data, dmu_flags_t flags, dmu_tx_t *tx,
+    dmu_abd_done_func_t *done, void *done_arg);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_DMU_OS_H */

include/os/linux/zfs/sys/zfs_vnops_os.h

@@ -86,6 +86,15 @@ extern int zfs_map(struct inode *ip, offset_t off, caddr_t *addrp,
     size_t len, unsigned long vm_flags);
 extern void zfs_zrele_async(znode_t *zp);
 
+/* async Direct I/O */
+struct kiocb;
+extern int zfs_read_async(znode_t *zp, zfs_uio_t *uio, int ioflag, cred_t *cr,
+    struct kiocb *kiocb);
+extern int zfs_write_async(znode_t *zp, zfs_uio_t *uio, int ioflag, cred_t *cr,
+    struct kiocb *kiocb);
+extern void zfs_async_dio_init(void);
+extern void zfs_async_dio_fini(void);
+
 #ifdef	__cplusplus
 }
 #endif

include/sys/dmu_impl.h

@@ -271,6 +271,8 @@ int dmu_write_direct(zio_t *, dmu_buf_impl_t *, abd_t *, dmu_tx_t *);
 int dmu_read_abd(dnode_t *, uint64_t, uint64_t, abd_t *, dmu_flags_t);
 int dmu_write_abd(dnode_t *, uint64_t, uint64_t, abd_t *, dmu_flags_t,
     dmu_tx_t *);
+abd_t *make_abd_for_dbuf(dmu_buf_impl_t *, abd_t *, uint64_t, uint64_t);
+void dmu_read_abd_done(zio_t *);
 #if defined(_KERNEL)
 int dmu_read_uio_direct(dnode_t *, zfs_uio_t *, uint64_t, dmu_flags_t);
 int dmu_write_uio_direct(dnode_t *, zfs_uio_t *, uint64_t, dmu_flags_t,

include/sys/zfs_vnops.h

@@ -31,9 +31,33 @@
 
 extern int zfs_bclone_enabled;
 
+/*
+ * Direct I/O tunables.  zfs_dio_enabled can be set to 0 to force all
+ * I/O through the ARC; zfs_dio_strict returns EINVAL for unaligned
+ * DIO instead of falling back.
+ */
+extern int zfs_dio_enabled;
+extern int zfs_dio_strict;
+
 extern int zfs_fsync(znode_t *, int, cred_t *);
 extern int zfs_read(znode_t *, zfs_uio_t *, int, cred_t *);
 extern int zfs_write(znode_t *, zfs_uio_t *, int, cred_t *);
+
+/*
+ * Direct I/O page-pinning setup.  Pins user pages for O_DIRECT reads,
+ * enforces alignment, and skips DIO for mmap'd or encrypted ranges.
+ * Returns 0 and sets UIO_DIRECT in uio->uio_extflg on success.
+ */
+extern int zfs_setup_direct(struct znode *, zfs_uio_t *, zfs_uio_rw_t, int *);
+
+/*
+ * Clear the SUID/SGID bits after a write by non-owner.
+ * Called from the async write completion path (zfs_vnops_os.c on Linux)
+ * as well as from the synchronous zfs_write().
+ */
+extern void zfs_clear_setid_bits_if_necessary(zfsvfs_t *, znode_t *, cred_t *,
+    uint64_t *, dmu_tx_t *);
+
 extern int zfs_holey(znode_t *, ulong_t, loff_t *);
 extern int zfs_access(znode_t *, int, int, cred_t *);
 extern int zfs_clone_range(znode_t *, uint64_t *, znode_t *, uint64_t *,

module/Kbuild.in

@@ -455,6 +455,7 @@ ZFS_OBJS := \
 
 ZFS_OBJS_OS := \
 	abd_os.o \
+	dmu_direct_os.o \
 	arc_os.o \
 	kasan_compat.o \
 	mmp_os.o \