On Tue, 6 May 2014, Vadim Zhukov wrote: > This fixes a bug with mtime not updated after file creation on ext2+ > file systems. Try it yourself: > > # mount -t ext2fs /dev/sd0o /mnt > # cp /tmp/qq /mnt > # ls -l /mnt/qq > -rw-r--r-- 1 root wheel 10 Jan 1 1970 qq > # touch /mnt/qq > -rw-r--r-- 1 root wheel 10 May 6 14:36 qq > > This probably does more than needed - I suspect that sometimes > the updated mtime could be already there, but I don't see any real > downside and it makes EX2FS_ITIMES() look more similar to > ufs_itimies(). Opinions? Okays?
As the person who broke this (sorry!), I think the diff should go a couple steps further and restore the IN_ACCESS handling too and mirror ufs_itimes() by bumping i_modrev on ctime change. Philip Index: ufs/ufs/inode.h =================================================================== RCS file: /cvs/src/sys/ufs/ufs/inode.h,v retrieving revision 1.44 diff -u -p -r1.44 inode.h --- ufs/ufs/inode.h 14 Apr 2014 22:25:40 -0000 1.44 +++ ufs/ufs/inode.h 6 May 2014 15:31:24 -0000 @@ -316,8 +316,14 @@ struct indir { #define EXT2FS_ITIMES(ip) do { \ if ((ip)->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) { \ (ip)->i_flag |= IN_MODIFIED; \ - if ((ip)->i_flag & IN_CHANGE) \ + if ((ip)->i_flag & IN_ACCESS) \ + (ip)->i_e2fs_atime = time_second; \ + if ((ip)->i_flag & IN_UPDATE) \ + (ip)->i_e2fs_mtime = time_second; \ + if ((ip)->i_flag & IN_CHANGE) { \ (ip)->i_e2fs_ctime = time_second; \ + (ip)->i_modrev++; \ + } \ (ip)->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); \ } \ } while (0)