Re: [PATCH] Big endian fixes for btrfs

[Anton Blanchard]

Hi Vladimir,

> object_id is left unconverted on purpose and is converted only on
> comparisons.

Could you be a bit more specific about which byteswapping of object_id
is bad?

Since we always byteswap when doing a comparison, object_ids need to be
created in little endian format:

-           key->object_id = GRUB_BTRFS_OBJECT_ID_CHUNK;
+           key->object_id = grub_cpu_to_le64 (GRUB_BTRFS_OBJECT_ID_CHUNK);

Whenever we use the contents of an object, eg as an address for reading
we need to byteswap it to native endian:

-         err = grub_btrfs_read_inode (data, &inode, cdirel->key.object_id,
+         err = grub_btrfs_read_inode (data, &inode,
+                                      grub_le_to_cpu64 (cdirel->key.object_id),
                                       tree);

and

-  data->inode = key_in.object_id;
+  data->inode = grub_le_to_cpu64 (key_in.object_id);

Comparisons need to be in native endian:

-         if (key_out.object_id != ino
+         if (grub_le_to_cpu64 (key_out.object_id) != ino


Finally we want to print the object_id in native endian or else they
will make no sense:

            grub_dprintf ("btrfs",
                          "leaf (depth %d) %" PRIxGRUB_UINT64_T
                          " %x %" PRIxGRUB_UINT64_T "\n", depth,
-                         leaf.key.object_id,leaf.key.type,leaf.key.offset);
+                         grub_le_to_cpu64(leaf.key.object_id),leaf.key.type,
+                         grub_le_to_cpu64 (leaf.key.offset));

I did test these with your fstest.sh script. Before the changes nothing
passed, after the changes everything passes.

Anton