Add FreeBSD posix_fadvise support.

As commit 320f0c6 did for Linux, connect POSIX_FADV_WILLNEED
up to dmu_prefetch() on FreeBSD.

Reviewed-by: Mateusz Guzik <[email protected]>
Reviewed-by: Alexander Motin <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Thomas Munro <[email protected]>

Author: macdice

module/os/freebsd/zfs/zfs_vnops_os.c

@@ -6055,6 +6055,80 @@ zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap)
 	return (EOPNOTSUPP);
 }
 
+#ifndef _SYS_SYSPROTO_H_
+struct vop_advise_args {
+	struct vnode *a_vp;
+	off_t a_start;
+	off_t a_end;
+	int a_advice;
+};
+#endif
+
+static int
+zfs_freebsd_advise(struct vop_advise_args *ap)
+{
+	vnode_t *vp = ap->a_vp;
+	off_t start = ap->a_start;
+	off_t end = ap->a_end;
+	int advice = ap->a_advice;
+	off_t len;
+	znode_t *zp;
+	zfsvfs_t *zfsvfs;
+	objset_t *os;
+	int error = 0;
+
+	if (end < start)
+		return (EINVAL);
+
+	VOP_LOCK(vp, LK_SHARED);
+	if (VN_IS_DOOMED(vp)) {
+		error = EBADF;
+		goto out_unlock;
+	}
+
+	zp = VTOZ(vp);
+	zfsvfs = zp->z_zfsvfs;
+	os = zp->z_zfsvfs->z_os;
+
+	if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
+		goto out_unlock;
+
+	/* kern_posix_fadvise points to the last byte, we want one past */
+	if (end != OFF_MAX)
+		end += 1;
+	len = end - start;
+
+	switch (advice) {
+	case POSIX_FADV_WILLNEED:
+		/*
+		 * Pass on the caller's size directly, but note that
+		 * dmu_prefetch_max will effectively cap it.  If there really
+		 * is a larger sequential access pattern, perhaps dmu_zfetch
+		 * will detect it.
+		 */
+		dmu_prefetch(os, zp->z_id, 0, start, len,
+		    ZIO_PRIORITY_ASYNC_READ);
+		break;
+	case POSIX_FADV_NORMAL:
+	case POSIX_FADV_RANDOM:
+	case POSIX_FADV_SEQUENTIAL:
+	case POSIX_FADV_DONTNEED:
+	case POSIX_FADV_NOREUSE:
+		/* ignored for now */
+		break;
+	default:
+		error = EINVAL;
+		break;
+	}
+
+	zfs_exit(zfsvfs, FTAG);
+
+out_unlock:
+	VOP_UNLOCK(vp);
+
+	return (error);
+}
+
 static int
 zfs_vptocnp(struct vop_vptocnp_args *ap)
 {
@@ -6293,6 +6367,7 @@ struct vop_vector zfs_vnodeops = {
 	.vop_link =		zfs_freebsd_link,
 	.vop_symlink =		zfs_freebsd_symlink,
 	.vop_readlink =		zfs_freebsd_readlink,
+	.vop_advise =		zfs_freebsd_advise,
 	.vop_read =		zfs_freebsd_read,
 	.vop_write =		zfs_freebsd_write,
 	.vop_remove =		zfs_freebsd_remove,

module/zfs/dmu.c

@@ -730,6 +730,18 @@ dmu_prefetch_by_dnode(dnode_t *dn, int64_t level, uint64_t offset,
 	 */
 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
 	if (dn->dn_datablkshift != 0) {
+
+		/*
+		 * Limit prefetch to present blocks.
+		 */
+		uint64_t size = (dn->dn_maxblkid + 1) << dn->dn_datablkshift;
+		if (offset >= size) {
+			rw_exit(&dn->dn_struct_rwlock);
+			return;
+		}
+		if (offset + len < offset || offset + len > size)
+			len = size - offset;
+
 		/*
 		 * The object has multiple blocks.  Calculate the full range
 		 * of blocks [start, end2) and then split it into two parts,

tests/runfiles/common.run

@@ -717,6 +717,10 @@ tags = ['functional', 'direct']
 tests = ['exec_001_pos', 'exec_002_neg']
 tags = ['functional', 'exec']
 
+[tests/functional/fadvise]
+tests = ['fadvise_willneed']
+tags = ['functional', 'fadvise']
+
 [tests/functional/fallocate]
 tests = ['fallocate_punch-hole']
 tags = ['functional', 'fallocate']

tests/runfiles/linux.run

@@ -112,10 +112,6 @@ tests = ['events_001_pos', 'events_002_pos', 'zed_rc_filter', 'zed_fd_spill',
     'zed_slow_io', 'zed_slow_io_many_vdevs', 'zed_diagnose_multiple']
 tags = ['functional', 'events']
 
-[tests/functional/fadvise:Linux]
-tests = ['fadvise_willneed']
-tags = ['functional', 'fadvise']
-
 [tests/functional/fallocate:Linux]
 tests = ['fallocate_prealloc', 'fallocate_zero-range']
 tags = ['functional', 'fallocate']

tests/zfs-tests/cmd/Makefile.am

@@ -140,7 +140,7 @@ scripts_zfs_tests_bin_PROGRAMS  += %D%/read_dos_attributes %D%/write_dos_attribu
 
 scripts_zfs_tests_bin_PROGRAMS += %D%/randfree_file
 %C%_randfree_file_SOURCES       = %D%/file/randfree_file.c
+endif
 
 scripts_zfs_tests_bin_PROGRAMS += %D%/file_fadvise
 %C%_file_fadvise_SOURCES  = %D%/file/file_fadvise.c
-endif