[src/trunk]: src/sys/ufs/lfs It isn't safe to drain dirops with seglock held, ...

[maya <maya%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/a9d00d2fd5a9
branches:  trunk
changeset: 824707:a9d00d2fd5a9
user:      maya <maya%NetBSD.org@localhost>
date:      Thu Jun 15 14:37:30 2017 +0000

description:
It isn't safe to drain dirops with seglock held, it'll deadlock if there
are any dirops. drain before grabbing seglock.

lfs_dirops == 0 is always true (as we already drained dirops), so omit
that part of the comparison.

Fixes a lot of LFS deadlocks. PR kern/52301

Many thanks to dholland for help analyzing coredumps

diffstat:

 sys/ufs/lfs/lfs_segment.c |  16 ++++++----------
 1 files changed, 6 insertions(+), 10 deletions(-)

diffs (58 lines):

diff -r 05081d185b6c -r a9d00d2fd5a9 sys/ufs/lfs/lfs_segment.c
--- a/sys/ufs/lfs/lfs_segment.c Thu Jun 15 14:23:42 2017 +0000
+++ b/sys/ufs/lfs/lfs_segment.c Thu Jun 15 14:37:30 2017 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: lfs_segment.c,v 1.271 2017/06/12 15:02:32 maya Exp $   */
+/*     $NetBSD: lfs_segment.c,v 1.272 2017/06/15 14:37:30 maya Exp $   */
 
 /*-
  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
@@ -60,7 +60,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: lfs_segment.c,v 1.271 2017/06/12 15:02:32 maya Exp $");
+__KERNEL_RCSID(0, "$NetBSD: lfs_segment.c,v 1.272 2017/06/15 14:37:30 maya Exp $");
 
 #ifdef DEBUG
 # define vndebug(vp, str) do {                                         \
@@ -603,7 +603,6 @@
        SEGUSE *segusep;
        int do_ckp, did_ckp, error;
        unsigned n, segleft, maxseg, sn, i, curseg;
-       int writer_set = 0;
        int dirty;
        int redo;
        SEGSUM *ssp;
@@ -628,6 +627,8 @@
        if (do_ckp)
                flags &= ~SEGM_SINGLE;
 
+       lfs_writer_enter(fs, "lfs segwrite");
+
        lfs_seglock(fs, flags | (do_ckp ? SEGM_CKP : 0));
        sp = fs->lfs_sp;
        if (sp->seg_flags & (SEGM_CLEAN | SEGM_CKP))
@@ -653,11 +654,7 @@
                                break;
                        }
 
-                       if (do_ckp || fs->lfs_dirops == 0) {
-                               if (!writer_set) {
-                                       lfs_writer_enter(fs, "lfs writer");
-                                       writer_set = 1;
-                               }
+                       if (do_ckp) {
                                error = lfs_writevnodes(fs, mp, sp, VN_DIROP);
                                if (um_error == 0)
                                        um_error = error;
@@ -806,8 +803,7 @@
 
        /* Note Ifile no longer needs to be written */
        fs->lfs_doifile = 0;
-       if (writer_set)
-               lfs_writer_leave(fs);
+       lfs_writer_leave(fs);
 
        /*
         * If we didn't write the Ifile, we didn't really do anything.