- 2.6 fixes:

  - mds_lock_new_child() uses specific namespace to avoid deadlock
    if i_generations == 0
  - portals_debug_copy_to_user() uses kmap_atomic() instead of
    kmap() because it's called beinged locked with disabled interrupts
  - ugly hacks which prevent userspace tools to try to load modules
  - corrections in Makefiles
  - minor compatibility defines and tweaks
  - inode->c_index is used as orphans counter
  - ll_lookup_finish_locks() doesn't release intent
    because on 2.6 it may be needed several times
  - check that nd isn't NULL before using
  - ll_update_inode() decoded dev number wrong way
  - PDE define is used to retrieve procfs-related data
  - rw26.c is almost copy of rw24.c
  - super25.c registers lustre and lustre_lite

- 2.6 kernel patches:
  - 2.6.0-mm2
  - dev_read_only_2.6.0.patch stores dev number, not struct block_device *
  - ext3-init-generation-2.6.0.patch - lustre dislikes when ext3 resets
    s_next_generation to 0 upon mount
  - ext3-map_inode_page-2.6.0.patch - added fix to unmap underlaying metadata
  - ext3-start_this_handle-must-return-error.patch - jbd returned 0 always:
    posssible oops
  - vfs_nointent_2.6.0-test6.patch - small contexts changes
  - invalidate_show-2.6.0.patch - port onto 2.6
  - iopen-2.6.0-test6.patch - several fixes
  - vfs_intent_2.6.0.patch - port onto 2.5.0
  - kernel_text_address-2.6.0.patch
  - export_symbols-2.6.0.patch

- kgdb_2.6.0 series

diff --git a/lustre/llite/llite_lib.c b/lustre/llite/llite_lib.c
index df2daa7..5ca5873 100644 (file)
--- a/lustre/llite/llite_lib.c
+++ b/lustre/llite/llite_lib.c
@@ -1028,7 +1028,7 @@ void ll_update_inode(struct inode *inode, struct mds_body *body,
 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
                 inode->i_rdev = body->rdev;
 #else
-                inode->i_rdev = old_encode_dev(body->rdev);
+                inode->i_rdev = old_decode_dev(body->rdev);
 #endif
         if (body->valid & OBD_MD_FLSIZE)
                 inode->i_size = body->size;