Linux/vnops: implement STATX_DIOALIGN

This statx(2) mask returns the alignment restrictions for O_DIRECT
access on the given file.

We're expected to return both memory and IO alignment. For memory, it's
always PAGE_SIZE. For IO, we return the current block size for the file,
which is the required alignment for an arbitrary block, and for the
first block we'll fall back to the ARC when necessary, so it should
always work.

Sponsored-by: https://despairlabs.com/sponsor/
Reviewed-by: Alexander Motin <[email protected]>
Signed-off-by: Rob Norris <[email protected]>
Closes #16972

Author: robn

include/sys/zfs_vnops.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, Rob Norris <[email protected]>
  */
 
 #ifndef	_SYS_FS_ZFS_VNOPS_H
@@ -42,6 +43,8 @@ extern int zfs_clone_range_replay(znode_t *, uint64_t, uint64_t, uint64_t,
 extern int zfs_getsecattr(znode_t *, vsecattr_t *, int, cred_t *);
 extern int zfs_setsecattr(znode_t *, vsecattr_t *, int, cred_t *);
 
+extern int zfs_get_direct_alignment(znode_t *, uint64_t *);
+
 extern int mappedread(znode_t *, int, zfs_uio_t *);
 extern int mappedread_sf(znode_t *, int, zfs_uio_t *);
 extern void update_pages(znode_t *, int64_t, int, objset_t *);

module/os/linux/zfs/zpl_inode.c

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2011, Lawrence Livermore National Security, LLC.
  * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
+ * Copyright (c) 2025, Rob Norris <[email protected]>
  */
 
 
@@ -30,6 +31,7 @@
 #include <sys/zfs_vnops.h>
 #include <sys/zfs_znode.h>
 #include <sys/dmu_objset.h>
+#include <sys/spa_impl.h>
 #include <sys/vfs.h>
 #include <sys/zpl.h>
 #include <sys/file.h>
@@ -490,6 +492,17 @@ zpl_getattr_impl(const struct path *path, struct kstat *stat, u32 request_mask,
 	}
 #endif
 
+#ifdef STATX_DIOALIGN
+	if (request_mask & STATX_DIOALIGN) {
+		uint64_t align;
+		if (zfs_get_direct_alignment(zp, &align) == 0) {
+			stat->dio_mem_align = PAGE_SIZE;
+			stat->dio_offset_align = align;
+			stat->result_mask |= STATX_DIOALIGN;
+		}
+	}
+#endif
+
 #ifdef STATX_ATTR_IMMUTABLE
 	if (zp->z_pflags & ZFS_IMMUTABLE)
 		stat->attributes |= STATX_ATTR_IMMUTABLE;

module/zfs/zfs_vnops.c

@@ -25,6 +25,7 @@
  * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
  * Copyright 2017 Nexenta Systems, Inc.
  * Copyright (c) 2021, 2022 by Pawel Jakub Dawidek
+ * Copyright (c) 2025, Rob Norris <[email protected]>
  */
 
 /* Portions Copyright 2007 Jeremy Teo */
@@ -1083,6 +1084,44 @@ zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr)
 	return (error);
 }
 
+/*
+ * Get the optimal alignment to ensure direct IO can be performed without
+ * incurring any RMW penalty on write. If direct IO is not enabled for this
+ * file, returns an error.
+ */
+int
+zfs_get_direct_alignment(znode_t *zp, uint64_t *alignp)
+{
+	zfsvfs_t *zfsvfs = ZTOZSB(zp);
+
+	if (!zfs_dio_enabled || zfsvfs->z_os->os_direct == ZFS_DIRECT_DISABLED)
+		return (SET_ERROR(EOPNOTSUPP));
+
+	/*
+	 * If the file has multiple blocks, then its block size is fixed
+	 * forever, and so is the ideal alignment.
+	 *
+	 * If however it only has a single block, then we want to return the
+	 * max block size it could possibly grown to (ie, the dataset
+	 * recordsize). We do this so that a program querying alignment
+	 * immediately after the file is created gets a value that won't change
+	 * once the file has grown into the second block and beyond.
+	 *
+	 * Because we don't have a count of blocks easily available here, we
+	 * check if the apparent file size is smaller than its current block
+	 * size (meaning, the file hasn't yet grown into the current block
+	 * size) and then, check if the block size is smaller than the dataset
+	 * maximum (meaning, if the file grew past the current block size, the
+	 * block size could would be increased).
+	 */
+	if (zp->z_size <= zp->z_blksz && zp->z_blksz < zfsvfs->z_max_blksz)
+		*alignp = MAX(zfsvfs->z_max_blksz, PAGE_SIZE);
+	else
+		*alignp = MAX(zp->z_blksz, PAGE_SIZE);
+
+	return (0);
+}
+
 #ifdef ZFS_DEBUG
 static int zil_fault_io = 0;
 #endif

tests/runfiles/common.run

@@ -981,7 +981,7 @@ tests = ['sparse_001_pos']
 tags = ['functional', 'sparse']
 
 [tests/functional/stat]
-tests = ['stat_001_pos']
+tests = ['stat_001_pos', 'statx_dioalign']
 tags = ['functional', 'stat']
 
 [tests/functional/suid]

tests/test-runner/bin/zts-report.py.in

@@ -128,6 +128,11 @@ idmap_reason = 'Idmapped mount needs kernel 5.12+'
 #
 cfr_reason = 'Kernel copy_file_range support required'
 
+#
+# Some statx fields are not supported by all kernels
+#
+statx_reason = 'Needed statx(2) field not supported on this kernel'
+
 if sys.platform.startswith('freebsd'):
     cfr_cross_reason = 'copy_file_range(2) cross-filesystem needs FreeBSD 14+'
 else:
@@ -293,7 +298,8 @@ if sys.platform.startswith('freebsd'):
         'block_cloning/block_cloning_cross_enc_dataset':
             ['SKIP', cfr_cross_reason],
         'block_cloning/block_cloning_copyfilerange_cross_dataset':
-            ['SKIP', cfr_cross_reason]
+            ['SKIP', cfr_cross_reason],
+        'stat/statx_dioalign': ['SKIP', 'na_reason'],
     })
 elif sys.platform.startswith('linux'):
     maybe.update({
@@ -361,6 +367,7 @@ elif sys.platform.startswith('linux'):
         'mmp/mmp_active_import': ['FAIL', known_reason],
         'mmp/mmp_exported_import': ['FAIL', known_reason],
         'mmp/mmp_inactive_import': ['FAIL', known_reason],
+        'stat/statx_dioalign': ['SKIP', 'statx_reason'],
     })
 
 

tests/zfs-tests/cmd/.gitignore

@@ -36,6 +36,7 @@
 /rename_dir
 /rm_lnkcnt_zero_file
 /send_doall
+/statx
 /stride_dd
 /threadsappend
 /user_ns_exec
@@ -54,3 +55,4 @@
 /skein_test
 /sha2_test
 /idmap_util
+/statx

tests/zfs-tests/cmd/Makefile.am

@@ -126,6 +126,7 @@ if BUILD_LINUX
 scripts_zfs_tests_bin_PROGRAMS += %D%/getversion
 scripts_zfs_tests_bin_PROGRAMS += %D%/user_ns_exec
 scripts_zfs_tests_bin_PROGRAMS += %D%/renameat2
+scripts_zfs_tests_bin_PROGRAMS += %D%/statx
 scripts_zfs_tests_bin_PROGRAMS += %D%/xattrtest
 scripts_zfs_tests_bin_PROGRAMS += %D%/zed_fd_spill-zedlet
 scripts_zfs_tests_bin_PROGRAMS += %D%/idmap_util