[illumos-gate merge]release-20200924

commit edd580643f2cf1434e252cd7779e83182ea84945
    12363 add O_DIRECT support

11 files changed, 209 insertions, 9 deletions

diff --git a/usr/src/cmd/truss/codes.c b/usr/src/cmd/truss/codes.c
index 88c67ea994..107b36ee87 100644
--- a/usr/src/cmd/truss/codes.c
+++ b/usr/src/cmd/truss/codes.c
@@ -23,7 +23,7 @@
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
- * Copyright 2019 Joyent, Inc.
+ * Copyright 2020 Joyent, Inc.
  * Copyright (c) 2014, OmniTI Computer Consulting, Inc. All rights reserved.
  */
 
@@ -1962,7 +1962,7 @@ pathconfname(int code)
 #define	ALL_O_FLAGS \
 	(O_NDELAY|O_APPEND|O_SYNC|O_DSYNC|O_NONBLOCK|O_CREAT|O_TRUNC\
 	|O_EXCL|O_NOCTTY|O_LARGEFILE|O_RSYNC|O_XATTR|O_NOFOLLOW|O_NOLINKS\
-	|O_CLOEXEC|O_DIRECTORY|FXATTRDIROPEN)
+	|O_CLOEXEC|O_DIRECTORY|O_DIRECT|FXATTRDIROPEN)
 
 const char *
 openarg(private_t *pri, int arg)
@@ -2024,6 +2024,8 @@ openarg(private_t *pri, int arg)
 		(void) strlcat(str, "|O_CLOEXEC", sizeof (pri->code_buf));
 	if (arg & O_DIRECTORY)
 		(void) strlcat(str, "|O_DIRECTORY", sizeof (pri->code_buf));
+	if (arg & O_DIRECT)
+		(void) strlcat(str, "|O_DIRECT", sizeof (pri->code_buf));
 	if (arg & FXATTRDIROPEN)
 		(void) strlcat(str, "|FXATTRDIROPEN", sizeof (pri->code_buf));
 
diff --git a/usr/src/compat/bhyve/sys/fcntl.h b/usr/src/compat/bhyve/sys/fcntl.h
index 062a3b84ac..9173446e8e 100644
--- a/usr/src/compat/bhyve/sys/fcntl.h
+++ b/usr/src/compat/bhyve/sys/fcntl.h
@@ -16,8 +16,6 @@
 #ifndef _COMPAT_FREEBSD_SYS_FCNTL_H_
 #define	_COMPAT_FREEBSD_SYS_FCNTL_H_
 
-#define	O_DIRECT	0x0
-
 #include_next <sys/fcntl.h>
 
 #endif	/* _COMPAT_FREEBSD_SYS_FCNTL_H_ */
diff --git a/usr/src/man/man2/open.2 b/usr/src/man/man2/open.2
index 86e00e575a..0e13f5af79 100644
--- a/usr/src/man/man2/open.2
+++ b/usr/src/man/man2/open.2
@@ -46,8 +46,9 @@
 .\" Portions Copyright (c) 2013, OmniTI Computer Consulting, Inc.
 .\" All Rights Reserved.
 .\" Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
+.\" Copyright 2020 Joyent, Inc.
 .\"
-.TH OPEN 2 "Jan 20, 2020"
+.TH OPEN 2 "Mar 10, 2020"
 .SH NAME
 open, openat \- open a file
 .SH SYNOPSIS
@@ -176,6 +177,19 @@ whether the file is open for reading, writing or for both.
 .sp
 .ne 2
 .na
+.B O_DIRECT
+.ad
+.sp .6
+.RS 4n
+Indicates that the file data is not going to be reused in the near future.
+When possible, data is read or written directly between the application's
+memory and the device when the data is accessed with \fBread\fR(2) and
+\fBwrite\fR(2) operations. See \fBdirectio\fR(3C) for more details.
+.RE
+
+.sp
+.ne 2
+.na
 .B O_DIRECTORY
 .ad
 .sp .6
@@ -619,7 +633,7 @@ The \fIpath\fR argument points to an illegal address.
 \fB\fBEINVAL\fR\fR
 .ad
 .RS 16n
-The system does not support synchronized I/O for this file, or the
+The system does not support synchronized or direct I/O for this file, or the
 \fBO_XATTR\fR flag was supplied and the underlying file system does not support
 extended file attributes.
 .RE
@@ -1018,7 +1032,8 @@ Standard	For \fBopen()\fR, see \fBstandards\fR(5).
 \fBIntro\fR(2), \fBchmod\fR(2), \fBclose\fR(2), \fBcreat\fR(2), \fBdup\fR(2),
 \fBexec\fR(2), \fBfcntl\fR(2), \fBgetmsg\fR(2), \fBgetrlimit\fR(2),
 \fBlseek\fR(2), \fBputmsg\fR(2), \fBread\fR(2), \fBstat\fR(2), \fBumask\fR(2),
-\fBwrite\fR(2), \fBattropen\fR(3C), \fBfcntl.h\fR(3HEAD), \fBstat.h\fR(3HEAD),
+\fBwrite\fR(2), \fBattropen\fR(3C), \fBdirectio\fR(3C),
+\fBfcntl.h\fR(3HEAD), \fBstat.h\fR(3HEAD),
 \fBunlockpt\fR(3C), \fBattributes\fR(5), \fBlf64\fR(5), \fBprivileges\fR(5),
 \fBstandards\fR(5), \fBconnld\fR(7M), \fBstreamio\fR(7I)
 .SH NOTES
diff --git a/usr/src/pkg/manifests/system-test-ostest.mf b/usr/src/pkg/manifests/system-test-ostest.mf
index 44dd39c58f..d132b36bf4 100644
--- a/usr/src/pkg/manifests/system-test-ostest.mf
+++ b/usr/src/pkg/manifests/system-test-ostest.mf
@@ -40,6 +40,7 @@ dir path=opt/os-tests/tests/secflags
 dir path=opt/os-tests/tests/sigqueue
 dir path=opt/os-tests/tests/sockfs
 dir path=opt/os-tests/tests/stress
+dir path=opt/os-tests/tests/syscall
 dir path=opt/os-tests/tests/timer
 dir path=opt/os-tests/tests/uccid
 file path=kernel/drv/$(ARCH64)/ksensor_test group=sys
@@ -112,6 +113,8 @@ file path=opt/os-tests/tests/sockfs/rights.64 mode=0555
 file path=opt/os-tests/tests/sockfs/sockpair mode=0555
 file path=opt/os-tests/tests/spoof-ras mode=0555
 file path=opt/os-tests/tests/stress/dladm-kstat mode=0555
+file path=opt/os-tests/tests/syscall/open.32 mode=0555
+file path=opt/os-tests/tests/syscall/open.64 mode=0555
 file path=opt/os-tests/tests/timer/timer_limit mode=0555
 file path=opt/os-tests/tests/uccid/atrparse mode=0555
 file path=opt/os-tests/tests/uccid/excl-badread mode=0555
diff --git a/usr/src/test/os-tests/runfiles/default.run b/usr/src/test/os-tests/runfiles/default.run
index 72158c8bc2..f2f66ab4d7 100644
--- a/usr/src/test/os-tests/runfiles/default.run
+++ b/usr/src/test/os-tests/runfiles/default.run
@@ -76,6 +76,9 @@ user = root
 tests = ['conn', 'dgram', 'drop_priv', 'nosignal', 'rights.32', 'rights.64',
          'sockpair']
 
+[/opt/os-tests/tests/syscall]
+tests = ['open.32', 'open.64']
+
 [/opt/os-tests/tests/pf_key]
 user = root
 timeout = 180
diff --git a/usr/src/test/os-tests/tests/Makefile b/usr/src/test/os-tests/tests/Makefile
index 7396e135c9..190c61e134 100644
--- a/usr/src/test/os-tests/tests/Makefile
+++ b/usr/src/test/os-tests/tests/Makefile
@@ -29,6 +29,7 @@ SUBDIRS =       \
 		sockfs \
 		spoof-ras \
 		stress \
+		syscall \
 		timer \
 		tmpfs \
 		uccid \
diff --git a/usr/src/test/os-tests/tests/syscall/Makefile b/usr/src/test/os-tests/tests/syscall/Makefile
new file mode 100644
index 0000000000..b4ba83a18b
--- /dev/null
+++ b/usr/src/test/os-tests/tests/syscall/Makefile
@@ -0,0 +1,55 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright 2020 Joyent, Inc.
+#
+
+include $(SRC)/cmd/Makefile.cmd
+include $(SRC)/test/Makefile.com
+
+PROGS = open
+
+CSTD = $(CSTD_GNU99)
+
+ROOTOPTPKG = $(ROOT)/opt/os-tests
+TESTDIR = $(ROOTOPTPKG)/tests/syscall
+
+PROGS32 = $(PROGS:%=%.32)
+PROGS64 = $(PROGS:%=%.64)
+CMDS = $(PROGS32) $(PROGS64)
+$(CMDS) := FILEMODE = 0555
+
+TESTDIRPROGS = $(PROGS32:%=$(TESTDIR)/%) \
+	$(PROGS64:%=$(TESTDIR)/%)
+
+all: $(PROGS32) $(PROGS64)
+
+install: all $(TESTDIR) $(TESTDIRPROGS)
+
+clobber: clean
+
+clean:
+	-$(RM) $(CMDS)
+
+$(TESTDIR):
+	$(INS.dir)
+
+$(TESTDIR)/%: %
+	$(INS.file)
+
+%.64: %.c
+	$(LINK64.c) -o $@ $< $(LDLIBS64)
+	$(POST_PROCESS)
+
+%.32: %.c
+	$(LINK.c) -o $@ $< $(LDLIBS)
+	$(POST_PROCESS)
diff --git a/usr/src/test/os-tests/tests/syscall/open.c b/usr/src/test/os-tests/tests/syscall/open.c
new file mode 100644
index 0000000000..e7f826b2a5
--- /dev/null
+++ b/usr/src/test/os-tests/tests/syscall/open.c
@@ -0,0 +1,102 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2020 Joyent, Inc.
+ */
+
+/*
+ * Test the open(2) syscall.
+ *
+ * Currently only tests O_DIRECT pass/fail based on the known support in the
+ * underlying file system.
+ *
+ * Note: there is a test for the O_DIRECTORY flag in the directory above this
+ * which could be consolidated into this code at some point.
+ */
+
+#include <stdio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/statvfs.h>
+#include <sys/param.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+
+static void
+o_direct_test(const char *dir)
+{
+	int fd;
+	struct statvfs64 buf;
+	char tmpname[MAXPATHLEN];
+	char *path;
+
+	boolean_t pass;
+
+	(void) snprintf(tmpname, sizeof (tmpname), "%s/otstXXXXXX", dir);
+	if ((path = mktemp(tmpname)) == NULL) {
+		(void) printf("FAILED: unable to create temp file name\n");
+		exit(1);
+	}
+
+	if (statvfs64(dir, &buf) == -1) {
+		perror("statvfs failed");
+		exit(1);
+	}
+
+	if (strcmp(buf.f_basetype, "zfs") == 0) {
+		pass = B_TRUE;
+	} else if (strcmp(buf.f_basetype, "tmpfs") == 0) {
+		pass = B_FALSE;
+	} else {
+		(void) printf("SKIP: expected 'zfs' or 'tmpfs'\n");
+		return;
+	}
+
+	fd = open(path, O_WRONLY | O_CREAT | O_EXCL | O_DIRECT, 0644);
+	if (fd >= 0) {
+		(void) close(fd);
+		(void) unlink(path);
+		if (!pass) {
+			(void) printf("FAILED: O_DIRECT on %s/tst_open is "
+			    "expected to fail\n", dir);
+			exit(1);
+		}
+	} else {
+		if (pass) {
+			(void) printf("FAILED: O_DIRECT on %s/tst_open is "
+			    "expected to succeed\n", dir);
+			exit(1);
+		}
+
+		if (errno != EINVAL) {
+			(void) printf("FAILED: expected EINVAL, got %d\n",
+			    errno);
+			exit(1);
+		}
+	}
+}
+
+int
+main(void)
+{
+	/* On typical illumos distros, /tmp is tmpfs, O_DIRECT should fail */
+	o_direct_test("/tmp");
+
+	/* On typical illumos distros, /var is zfs, O_DIRECT should pass */
+	o_direct_test("/var/tmp");
+
+	(void) printf("PASS\n");
+	return (0);
+}
diff --git a/usr/src/uts/common/fs/vnode.c b/usr/src/uts/common/fs/vnode.c
index c57884a940..3babdbecdb 100644
--- a/usr/src/uts/common/fs/vnode.c
+++ b/usr/src/uts/common/fs/vnode.c
@@ -51,6 +51,7 @@
 #include <sys/vfs.h>
 #include <sys/vfs_opreg.h>
 #include <sys/vnode.h>
+#include <sys/filio.h>
 #include <sys/rwstlock.h>
 #include <sys/fem.h>
 #include <sys/stat.h>
@@ -1254,6 +1255,22 @@ top:
 		if ((error = VOP_SETATTR(vp, &vattr, 0, CRED(), NULL)) != 0)
 			goto out;
 	}
+
+	/*
+	 * Turn on directio, if requested.
+	 */
+	if (filemode & FDIRECT) {
+		if ((error = VOP_IOCTL(vp, _FIODIRECTIO, DIRECTIO_ON, 0,
+		    CRED(), NULL, NULL)) != 0) {
+			/*
+			 * On Linux, O_DIRECT returns EINVAL when the file
+			 * system does not support directio, so we'll do the
+			 * same.
+			 */
+			error = EINVAL;
+			goto out;
+		}
+	}
 out:
 	ASSERT(vp->v_count > 0);
 
diff --git a/usr/src/uts/common/sys/fcntl.h b/usr/src/uts/common/sys/fcntl.h
index cf55ebbf2a..4af23583fd 100644
--- a/usr/src/uts/common/sys/fcntl.h
+++ b/usr/src/uts/common/sys/fcntl.h
@@ -37,7 +37,7 @@
  */
 
 /* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
-/* Copyright 2015, Joyent, Inc. */
+/* Copyright 2020 Joyent, Inc. */
 
 #ifndef	_SYS_FCNTL_H
 #define	_SYS_FCNTL_H
@@ -89,6 +89,9 @@ extern "C" {
 #if !defined(_STRICT_SYMBOLS) || defined(_XPG7)
 #define	O_DIRECTORY	0x1000000	/* fail if not a directory */
 #endif
+#if !defined(_STRICT_SYMBOLS)
+#define	O_DIRECT	0x2000000	/* direct disk access hint */
+#endif
 
 /*
  * fcntl(2) requests
diff --git a/usr/src/uts/common/sys/file.h b/usr/src/uts/common/sys/file.h
index f544033d2f..c7d67a5d47 100644
--- a/usr/src/uts/common/sys/file.h
+++ b/usr/src/uts/common/sys/file.h
@@ -27,7 +27,7 @@
 /*	  All Rights Reserved	*/
 
 /* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
-/* Copyright 2017 Joyent, Inc. */
+/* Copyright 2020 Joyent, Inc. */
 
 #ifndef _SYS_FILE_H
 #define	_SYS_FILE_H
@@ -119,6 +119,7 @@ typedef struct fpollinfo {
 
 #define	FCLOEXEC	0x800000	/* O_CLOEXEC = 0x800000 */
 #define	FDIRECTORY	0x1000000	/* O_DIRECTORY = 0x1000000 */
+#define	FDIRECT		0x2000000	/* O_DIRECT = 0x2000000 */
 
 #if defined(_KERNEL) || defined(_FAKE_KERNEL)