Re: [PATCH] biosdisk / open_device() messing up offsets

[Pavel Roskin]

On Tue, 2008-06-10 at 15:16 +0800, Bean wrote:
> Hi,
> 
> Ok, I fix some bugs, please see if the new patch works.

It doesn't work.  i sent the logs privately.  I also checked the logs
and it looks like that GRUB is reading the log properly.  Then I guess
we are not using the log correctly.

I changed fs/ext2.c to see how the results would differ if
grub_fshelp_map_block() is bypassed:

diff --git a/fs/ext2.c b/fs/ext2.c
index ffe9e33..7b4bfd6 100644
--- a/fs/ext2.c
+++ b/fs/ext2.c
@@ -257,12 +257,24 @@ inline static grub_err_t
 grub_ext2_blockgroup (struct grub_ext2_data *data, int group, 
                      struct grub_ext2_block_group *blkgrp)
 {
-  return grub_disk_read (data->disk,
-                        (grub_fshelp_map_block (data->journal,
-                                                 grub_le_to_cpu32 
(data->sblock.first_data_block) + 1)
-                         << LOG2_EXT2_BLOCK_SIZE (data)),
+  grub_err_t ret;
+  int block;
+
+  block = grub_le_to_cpu32 (data->sblock.first_data_block) + 1;
+  ret = grub_disk_read (data->disk,
+                        block << LOG2_EXT2_BLOCK_SIZE (data),
+                        group * sizeof (struct grub_ext2_block_group), 
+                        sizeof (struct grub_ext2_block_group), (char *) 
blkgrp);
+  grub_printf("direct:  ret = %d, block = %d, inode_table_id = %d\n", ret, 
block,
+             blkgrp->inode_table_id);
+  block = grub_fshelp_map_block(data->journal, block);
+  ret = grub_disk_read (data->disk,
+                        block << LOG2_EXT2_BLOCK_SIZE (data),
                         group * sizeof (struct grub_ext2_block_group), 
                         sizeof (struct grub_ext2_block_group), (char *) 
blkgrp);
+  grub_printf("journal: ret = %d, block = %d, inode_table_id = %d\n", ret, 
block,
+             blkgrp->inode_table_id);
+  return ret;
 }
 
 


That's the output:

direct:  ret = 0, block = 1, inode_table_id = 1027
journal: ret = 0, block = 1, inode_table_id = 1027
direct:  ret = 0, block = 1, inode_table_id = 1027
journal: ret = 0, block = 17707, inode_table_id = 1027
direct:  ret = 0, block = 1, inode_table_id = 1027
journal: ret = 0, block = 1, inode_table_id = 1027
direct:  ret = 0, block = 1, inode_table_id = 1027
journal: ret = 0, block = 17707, inode_table_id = 1027
direct:  ret = 0, block = 1, inode_table_id = 15237122
journal: ret = 0, block = 17707, inode_table_id = 6895416


So, originally we didn't have the journal opened, so the mapping was
trivial, then 1 became mapped to 17707.  But at some point the data in
that block became different from what it used to be and different in the
journaled and non-journaled blocks.

-- 
Regards,
Pavel Roskin