fsck_ffs: change struct bufarea member b_index type to uint64_t

[kib]

Summary

It needs to correctly participate in the inode numbers arithmetic.
It is possible to have inode number that is negative if truncated to 32
bit, which causes OOB accesses in ginode() calculating inode location in
the inode block:
        ip->i_dp = (union dinode *)
            &ip->i_bp->b_un.b_dinode2[inumber - ip->i_bp->b_index];

[kib]

The patch is written to handle the real problem on my home machine.
I used a different change to get my (large) volume back:

diff --git a/sbin/fsck_ffs/inode.c b/sbin/fsck_ffs/inode.c
index f7b9b941fdb4..2538bf66d394 100644
--- a/sbin/fsck_ffs/inode.c
+++ b/sbin/fsck_ffs/inode.c
@@ -473,15 +473,15 @@ ginode(ino_t inumber, struct inode *ip)
                ip->i_bp = icachebp;
        }
        if (sblock.fs_magic == FS_UFS1_MAGIC) {
-               ip->i_dp = (union dinode *)
-                   &ip->i_bp->b_un.b_dinode1[inumber - ip->i_bp->b_index];
+               ip->i_dp = (union dinode *) &ip->i_bp->b_un.b_dinode1[
+                   inumber - (unsigned)ip->i_bp->b_index];
                dpp.dp1 = (struct ufs1_dinode *)ip->i_dp;
                if (ffs_oldfscompat_inode_read(&sblock, dpp, now))
                        inodirty(ip);
                return;
        }
        ip->i_dp = (union dinode *)
-           &ip->i_bp->b_un.b_dinode2[inumber - ip->i_bp->b_index];
+           &ip->i_bp->b_un.b_dinode2[inumber - (unsigned)ip->i_bp->b_index];
        dpp.dp2 = dp = (struct ufs2_dinode *)ip->i_dp;
        /* Do not check hash of inodes being created */
        if (dp->di_mode != 0 && ffs_verify_dinode_ckhash(&sblock, dp)) {

Then decided that this is really a missed place from ino64.

[pho]

I ran fsck tests with D50520.156022.patch added, without observing any issues.

[mckusick]

Agree with change.