From: Kirill A. Korinsky <kirill@korins.ky>
Subject: sys/tmpfs: use getnanotime() like FFS and MFS for 
To: OpenBSD tech <tech@openbsd.org>
Date: Sat, 28 Mar 2026 23:29:41 +0100

tech@,

I'd like to fix a wired race condition in tmpfs which I had discovered as as
failed lang/ruby build.

The port does:

FIX_RIPPER ?=	sed -i 's/%%CFLAGS_OVERRIDE%%/${CFLAGS_OVERRIDE}/g' \
			${WRKSRC}/ext/ripper/depend;
...

pre-configure:
	${FIX_RIPPER}
	/usr/bin/touch ${WRKSRC}/ext/ripper/ripper.c

and Makefile expects that mtime of ripper.c >= depend; and if it isn't true
it regenerates ripper.c via $(BASERUBY) ... tool/lrama/exe/lrama ... where
BASERUBY is ./tool/missing-baseruby.bat, so build fails.

Why ripper.c >= depend? Here the trick.

I run my bulk with pobj on tmpfs, and tmpfs uses nanotime() to update mtime,
and touch uses utimensat(UTIME_NOW) which uses getnanotime() which is
cached. And touch sets that cached time, which is smaller than current time.

So, it is quite possible that touch set time smaller, cached one.

I think the right fix is use getnanotime() instead nanotime() like FFS and
MFS.

Thougs? Ok?


Index: tmpfs_subr.c
===================================================================
RCS file: /home/cvs/src/sys/tmpfs/tmpfs_subr.c,v
diff -u -p -r1.27 tmpfs_subr.c
--- tmpfs_subr.c	12 Sep 2024 09:04:51 -0000	1.27
+++ tmpfs_subr.c	28 Mar 2026 22:03:32 -0000
@@ -1140,7 +1140,7 @@ tmpfs_update(tmpfs_node_t *node, int fla
 {
 	struct timespec nowtm;
 
-	nanotime(&nowtm);
+	getnanotime(&nowtm);
 
 	if (flags & TMPFS_NODE_ACCESSED) {
 		node->tn_atime = nowtm;



-- 
wbr, Kirill