[0/3,v2] xfs: clean up buffer cache disk addressing

Message

Dave Chinner Aug. 19, 2021, midnight UTC

Over time, we've moved from using bp->b_bn as the physical disk
address for a buffer to using the address of the first buffer map
attached to the buffer. This can be seen by the implementation of
XFS_BUF_ADDR() macro:

#define XFS_BUF_ADDR(bp) ((bp)->b_maps[0].bm_bn)

The bp->b_bn value is now used as the buffer cache index for the
buffer, and it is always set to XFS_BUF_DADDR_NULL for uncached
buffers. Hence code that uses bp->b_bn for the physical address of
the buffer will not do the right thing when passed an uncached
buffer.

THis series of patches addresses this problem. It adds a helper
function xfs_buf_daddr() to extract the physical address of the
buffer from it, and replaces all the open coded bp->b_bn accesses
and the XFS_BUF_ADDR() users to use it. It then renames b_bn to
b_index and converts all the internal cache lookup code to use
b_index rather than b_bn.

The result is that all code now uses xfs_buf_daddr() where physical
addresses are required, and the cache index variable isn't named in
a manner that engenders external use of access. The changes end up
being relatively small, and the impact on userspace outside libxfs
is even smaller (only a couple of dozen references to b_bn or
XFS_BUF_ADDR).

Version 2
- rebase on 5.14-rc4 + for-next + "xfs: rework feature flags v3"
- use __be64 type for comparison in xfs_btree_new_iroot()
- rename b_index to b_rhash_key
- fix typos and wording in commit messages

Version 1
- based on 5.14-rc4 + for-next
- https://lore.kernel.org/linux-xfs/[e-mail]/