xfsdump: zero bs_forkoff, don't fill in the value

dump/content.c copy_xfs_bstat(bstat_t *dst, xfs_bstat_t *src) · line 4931

I think I understand this now.

A legacy xfs_repair would not have bs_forkoff in

read_filehdr
xlate_filehdr

    xlate_bstat
      IXLATE(bs1, bs2, bs_forkoff);

And bs_pad is copied without any translation. This works fine as long as bs_forkoff and bs_projid_hi are 0. In the failure case bs_forkoff was nonzero
and changed the checksum because it was not translated. I could not reproduce
the checksum failure on a filesystem without xattrs because bs_forkoff was always 0.

*** Get set up with an xattr:

# touch /mnt/scratch/file
# setfattr -h -n user.testattr -v testvalue /mnt/scratch/file

*** Dump the filesystem with 3.1.2:

# xfsdump - /mnt/scratch > dump-with-attribute-and-3.1.2.out
xfsdump: using file dump (drive_simple) strategy
xfsdump: version 3.1.2 (dump format 3.0)
...
xfsdump: Dump Status: SUCCESS

*** Restore with 3.1.0, and print out bs_pad:

# xfsrestore -f dump-with-attribute-and-3.1.2.out /mnt/test 
xfsrestore: using file dump (drive_simple) strategy
xfsrestore: version 3.1.0 (dump format 3.0) - type ^C for status and control
...
xfsrestore: reading directories
read_filehdr pad is 0 0 0 0 0 0 0 0 0 0 0 0 0 0 
read_filehdr pad is 0 78 0 0 0 0 0 0 0 0 0 0 0 0 
xfsrestore: WARNING: bad file header checksum
...
xfsrestore: Restore Status: INCOMPLETE

*** Restore with 3.1.0, translating bs_pad where bs_forkoff lives:

# xfsrestore -f dump-with-attribute-and-3.1.2.out /mnt/test
xfsrestore: using file dump (drive_simple) strategy
xfsrestore: version 3.1.0 (dump format 3.0) - type ^C for status and control
...
xfsrestore: reading directories
read_filehdr pad is 0 0 0 0 0 0 0 0 0 0 0 0 0 0
read_filehdr pad is 78 0 0 0 0 0 0 0 0 0 0 0 0 0
xfsrestore: 1 directories and 1 entries processed
xfsrestore: directory post-processing
xfsrestore: restoring non-directory files
read_filehdr pad is 78 0 0 0 0 0 0 0 0 0 0 0 0 0
read_filehdr pad is 0 0 0 0 0 0 0 0 0 0 0 0 0 0
xfsrestore: restore complete: 0 seconds elapsed
xfsrestore: Restore Summary:
xfsrestore:   stream 0 /root/xfsdump/dump-with-attribute-and-3.1.2.out OK (success)
xfsrestore: Restore Status: SUCCESS

:!quilt di

Index: xfsdump/restore/content.c
===================================================================
@@ -8038,6 +8038,7 @@ read_filehdr( drive_t *drivep, filehdr_t
        intgen_t nread;
        intgen_t rval;
        filehdr_t tmpfh;
+       int i;
 
        nread = read_buf( ( char * )&tmpfh,
                          sizeof( *fhdrp ),

@@ -8078,6 +8079,16 @@ read_filehdr( drive_t *drivep, filehdr_t
                              "corrupt file header\n") );
                        return RV_CORRUPT;
                }
+               i = bstatp->bs_pad[0];
+               bstatp->bs_pad[0] = bstatp->bs_pad[1];
+               bstatp->bs_pad[1] = i;
+
+               printf("%s pad is ", __func__);
+               for (i = 0; i < 14; i++) {
+                       printf("%x ", bstatp->bs_pad[i]);
+               }
+               printf("\n");
+
                if ( !is_checksum_valid( fhdrp, FILEHDR_SZ )) {
                        mlog( MLOG_NORMAL | MLOG_WARNING, _(
                              "bad file header checksum\n") );

Now all new dumps will have zeroed bs_pad[0,1] so it doesn't matter that they aren't xlated. Your fix looks good.

I'll pull this in immediately.

Thanks,
Ben

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs