On Wed, 2011-09-21 at 09:05 +1000, Dave Chinner wrote:
> On Tue, Sep 20, 2011 at 01:43:13PM -0500, Alex Elder wrote:
> > This applies on top of Christoph Hellwig's recent "xfs_repair: add
> > printf format checking and fix the fallout" patch. It extends the
> > fixes for warnings beyond just xfs_repair and across everything in
> > xfsprogs.
> >
> > It builds cleanly on ia64 and x86_64, and builds without any
> > printf() format-related warnings on i386.
> >
> > Signed-off-by: Alex Elder <aelder@xxxxxxx>
> >
> > ---
> > io/parent.c | 28 ++++++++++++++++------------
> > logprint/log_misc.c | 34 ++++++++++++++++++++--------------
> > logprint/log_print_all.c | 16 ++++++++++------
> > repair/dinode.c | 20 ++++++++++++--------
> > repair/scan.c | 14 +++++++++-----
> > 5 files changed, 67 insertions(+), 45 deletions(-)
> >
> > Index: b/io/parent.c
> > ===================================================================
> > --- a/io/parent.c
> > +++ b/io/parent.c
> > @@ -52,12 +52,12 @@ check_parent_entry(xfs_bstat_t *bstatp,
> > if (sts != 0) {
> > fprintf(stderr,
> > _("inode-path for inode: %llu is incorrect - path \"%s\" non-existent\n"),
> > - bstatp->bs_ino, fullpath);
> > + (unsigned long long) bstatp->bs_ino, fullpath);
>
> Hmmm, didn't Christoph fix these inode number warnings by changing
> the format specifier to PRIU64, not by adding casts? bs_ino is
> defined as:
>
> __u64 bs_ino;
>
> So PRIu64 is the right thing to do, isn't it?
I haven't checked right now but I wrote a small thesis
in an e-mail a few months ago about it. But as I recall
uint64_t was defined based on long on some architectures
and based on long long in others. And __u64 is based
on long long everywhere in the kernel. Something like
that. OK, I went and found it.
http://oss.sgi.com/archives/xfs/2011-07/msg00399.html
The problem lies in the kernel, which defines
__u64 as a an (unsigned long) in ia64, but as
(unsigned long long) in x86_64. This would be
fine, except that the user space code uses
(unsigned long) as u_int64_t for both architectures
(and that, after all is what PRIu64 is for).
Hence for inodes (and anything else defined as
a __u64) you have to use explicit casts because
PRIu64 won't always work for you.
-Alex
>
> Either way would work, but being consistent would be good. ;)
>
> ....
>
> > ===================================================================
> > --- a/logprint/log_misc.c
> > +++ b/logprint/log_misc.c
> > @@ -307,12 +307,14 @@ xlog_print_trans_buffer(xfs_caddr_t *ptr
> > */
> > memmove(&x, *ptr, sizeof(__be64));
> > memmove(&y, *ptr+8, sizeof(__be64));
> > - printf(_("icount: %lld ifree: %lld "),
> > - be64_to_cpu(x), be64_to_cpu(y));
> > + printf(_("icount: %llu ifree: %llu "),
> > + (unsigned long long) be64_to_cpu(x),
> > + (unsigned long long) be64_to_cpu(y));
>
> Same for al the be64_to_cpu() functions - their return type is
> __u64, too.
>
> > forkname);
> > @@ -1374,7 +1376,7 @@ process_lclinode(
> > XFS_DFORK_DSIZE(dip, mp)) {
> > do_warn(
> > _("local inode %" PRIu64 " data fork is too large (size = %lld, max = %d)\n"),
> > - lino, be64_to_cpu(dip->di_size),
> > + lino, (unsigned long long) be64_to_cpu(dip->di_size),
>
> That format specifier is wrong - it is %lld. Should be %llu, or
> PRIu64 as previously mentioned without the cast.
>
> Cheers,
>
> Dave.
_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs
