Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-3953 lustre build system improvments
  3. LU-3954

client-only "make rpms" ldiskfs detection inconsistent

    XMLWordPrintable

Details

    • Technical task
    • Resolution: Fixed
    • Minor
    • Lustre 2.5.0
    • Lustre 2.4.0
    • None
    • patchless client: 2.6.32.26-175.fc12.x86_64
    • 10524

    Description

      I tried to build a client-only RPM using in the Lustre source tree for patchless kernel 2.6.32.26-175.fc12.x86_64 using the standard:

      sh autogen
      ./configure
      make rpms
      

      In previous releases, configure would correctly detect that ldiskfs was not available for the kernel and/or the kernel sources were not available and do a client-only build.

      Unfortunately, the 2.4.0 client build died in mid-stream due to inconsistent detection of ldiskfs support. In the top-level configure.ac file it has:

      # Overridden by --without-ldiskfs/--with-zfs configure options
      AM_CONDITIONAL(LDISKFS_ENABLED, true)
      

      Which sets LDISKFS_ENABLED, but later on in the build it failed in mount_utils.c with:

      mount_lustre-mount_utils.o: In function `osd_is_lustre': 
      /usr/src/lustre-head/lustre/utils/mount_utils.c:461: undefined reference to `ldiskfs_is_lustre'          
      mount_lustre-mount_utils.o: In function `osd_fini':
      /usr/src/lustre-head/lustre/utils/mount_utils.c:650: undefined reference to `ldiskfs_fini'          
      mount_lustre-mount_utils.o: In function `osd_init':
      /usr/src/lustre-head/lustre/utils/mount_utils.c:632: undefined reference to `ldiskfs_init'
      

      This is in

      /* Was this device formatted for Lustre */
      int osd_is_lustre(char *dev, unsigned *mount_type)
      {
              vprint("checking for existing Lustre data: ");
      
      #ifdef HAVE_LDISKFS_OSD
              if (ldiskfs_is_lustre(dev, mount_type)) {
                      vprint("found\n");
                      return 1;
              }
      #endif /* HAVE_LDISKFS_OSD */
      

      Since HAVE_LDISKFS_OSD is set, but mount_utils_ldiskfs.c is not being built. The ldiskfs configure step correctly detects no ldiskfs support for this kernel, and tries to disable ldiskfs and set -disable-server, but this is "lost" in the sub-configure step? The ldiskfs configure shouldn't set -disable-server itself anyway, because it may be a ZFS-only server system.

      It would seem better to run the ldiskfs and ZFS configure stages early on in the configure process, then if no backing filesystem support is found, it should automatically set --disable-server for the main Lustre build.

      Attachments

        Issue Links

          Activity

            People

              mdiep Minh Diep
              adilger Andreas Dilger
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: