Inconsistent blocking addressing in RFS

[Fan Deng]

Background
There are two ways to address a block in RFS:

1. Via a single 32bit block number (bno)
2. Via a group number(gno) and a bit offset (bit)

They should be fully convertible (1-1 mapping). In other words, the equation to convert 1 to 2 should be unique within the RFS implementation.

The bug
The RFS implementation contains two different conversions between 1 and 2.

Details

1. In rtems_rfs_group_bitmap_alloc (rtems-rfs-group.c, line 172)

   bno = gno * group_blocks + bit
   

2. In rtems_rfs_group_bitmap_alloc (rtems-rfs-group.c, line 228)

   bno = gno * group_blocks + bit + 1 (via rtems_rfs_group_block() function)
   

3. In rtems_rfs_group_bitmap_free (rtems-rfs-group.c, line 283)

   bno = gno * group_blocks + bit + 1 (RTEMS_RFS_SUPERBLOCK_SIZE)
   

4. In rtems_rfs_group_bitmap_test (rtems-rfs-group.c, line 332)

   bno = gno * group_blocks + bit
   

To summarize, the implementation contains two ways of converting a bno to a (gno, bit) pair:

Either:

bno = gno * group_blocks + bit

Or:

bno = gno * group_blocks + bit + 1

The Fix
The RFS implementation should consistently convert a bno to a (gno, bit) pair with:

bno = gno * group_blocks + bit + RTEMS_RFS_SUPERBLOCK_SIZE

This is because the superblock is not accounted for in the block bitmaps. So places to change:

1. rtems-rfs-group.c: all references to the conversion must be updated to use RTEMS_RFS_SUPERBLOCK_SIZE explicitly.

2. rtems_rfs_group_block converts the pair to bno via:

   #define rtems_rfs_group_block(_g, _b) (((_g)->base) + (_b))
   

(_g)->base is calculated via rtems-rfs-format.c from:

#define rtems_rfs_fs_block(_fs, _grp, _blk) \
  ((((_fs)->group_blocks) * (_grp)) + (_blk) + 1)

The "+ 1" part should really be "+ RTEMS_RFS_SUPERBLOCK_SIZE" to be logically correct. As RTEMS_RFS_SUPERBLOCK_SIZE itself has a comment saying:

/**
 * Number of blocks in the superblock. Yes I know it is a superblock and not
 * superblocks but if for any reason this needs to change it is handled.
 */
#define RTEMS_RFS_SUPERBLOCK_SIZE (1)

[Sebastian Huber]

Milestone renamed

[Chris Johns <chrisj@…>]

In 9e8df1fe/rtems:

fstest/fsrfsbitmap01: Update RFS bitmap tests to test fixes.

Add tests to check the patches for this ticket exist and are fixed.

Close #3089