On Thu, Jan 31, 2019 at 03:18:34PM -0800, Darrick J. Wong wrote: > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > > Add tracepoints so we can associate high level operations with low level > updates. No functional changes. > > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > --- > fs/xfs/xfs_inode.c | 4 ++++ > fs/xfs/xfs_trace.h | 3 +++ > 2 files changed, 7 insertions(+) > > > diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c > index d5b3f8fdac7e..56349497d75b 100644 > --- a/fs/xfs/xfs_inode.c > +++ b/fs/xfs/xfs_inode.c > @@ -2030,6 +2030,8 @@ xfs_iunlink( > ASSERT(VFS_I(ip)->i_mode != 0); > ASSERT(xfs_verify_ino(mp, ip->i_ino)); > > + trace_xfs_iunlink(ip); > + It's worth noting that current inode events print a global inode number (i.e., "ino 0x%llx ...") whereas the new iunlink tracepoints from previous patches print an agno/agino combination. I haven't looked at the actual trace output from an unlink with these changes yet, but I do wonder if there's value in having a more consistent format. Hm? Brian > agno = XFS_INO_TO_AGNO(mp, ip->i_ino); > agino = XFS_INO_TO_AGINO(mp, ip->i_ino); > bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; > @@ -2199,6 +2201,8 @@ xfs_iunlink_remove( > short bucket_index; > int error; > > + trace_xfs_iunlink_remove(ip); > + > if (!xfs_verify_ino(mp, ip->i_ino)) > return -EFSCORRUPTED; > > diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h > index fbec8f0e1a9a..22d4729143b5 100644 > --- a/fs/xfs/xfs_trace.h > +++ b/fs/xfs/xfs_trace.h > @@ -3423,6 +3423,9 @@ TRACE_EVENT(xfs_iunlink_update_dinode, > __entry->new_ptr) > ); > > +DEFINE_INODE_EVENT(xfs_iunlink); > +DEFINE_INODE_EVENT(xfs_iunlink_remove); > + > #endif /* _TRACE_XFS_H */ > > #undef TRACE_INCLUDE_PATH >
