[src/trunk]: src/sys/ufs/ufs Add a comment about a matched pair of off-by-one...

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

details:   https://anonhg.NetBSD.org/src/rev/cd9b3ee9634b
branches:  trunk
changeset: 787398:cd9b3ee9634b
user:      dholland <dholland%NetBSD.org@localhost>
date:      Sun Jun 16 00:13:58 2013 +0000

description:
Add a comment about a matched pair of off-by-one tests that make the
maximum size of short symlinks one byte less than one would think it
should be. Caution against changing it; that would break compatibility
with existing disk images. Behavior noticed by qjsgkem on freenode.

If my analysis is wrong, please correct...

diffstat:

 sys/ufs/ufs/ufs_vnops.c |  17 +++++++++++++++--
 1 files changed, 15 insertions(+), 2 deletions(-)

diffs (45 lines):

diff -r bfb10baa7b2f -r cd9b3ee9634b sys/ufs/ufs/ufs_vnops.c
--- a/sys/ufs/ufs/ufs_vnops.c   Sat Jun 15 22:53:18 2013 +0000
+++ b/sys/ufs/ufs/ufs_vnops.c   Sun Jun 16 00:13:58 2013 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: ufs_vnops.c,v 1.215 2013/06/09 18:54:05 christos Exp $ */
+/*     $NetBSD: ufs_vnops.c,v 1.216 2013/06/16 00:13:58 dholland Exp $ */
 
 /*-
  * Copyright (c) 2008 The NetBSD Foundation, Inc.
@@ -66,7 +66,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ufs_vnops.c,v 1.215 2013/06/09 18:54:05 christos Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ufs_vnops.c,v 1.216 2013/06/16 00:13:58 dholland Exp $");
 
 #if defined(_KERNEL_OPT)
 #include "opt_ffs.h"
@@ -1267,6 +1267,13 @@
        vp = *vpp;
        len = strlen(ap->a_target);
        ip = VTOI(vp);
+       /*
+        * This test is off by one. um_maxsymlinklen contains the
+        * number of bytes available, and we aren't storing a \0, so
+        * the test should properly be <=. However, it cannot be
+        * changed as this would break compatibility with existing fs
+        * images -- see the way ufs_readlink() works.
+        */
        if (len < ip->i_ump->um_maxsymlinklen) {
                memcpy((char *)SHORTLINK(ip), ap->a_target, len);
                ip->i_size = len;
@@ -1453,6 +1460,12 @@
        struct ufsmount *ump = VFSTOUFS(vp->v_mount);
        int             isize;
 
+       /*
+        * The test against um_maxsymlinklen is off by one; it should
+        * theoretically be <=, not <. However, it cannot be changed
+        * as that would break compatibility with existing fs images.
+        */
+
        isize = ip->i_size;
        if (isize < ump->um_maxsymlinklen ||
            (ump->um_maxsymlinklen == 0 && DIP(ip, blocks) == 0)) {