afs: Stop implementing ->writepage()

We're trying to get rid of the ->writepage() hook[1].  Stop afs from using
it by unlocking the page and calling afs_writepages_region() rather than
folio_write_one().

A flag is passed to afs_writepages_region() to indicate that it should only
write a single region so that we don't flush the entire file in
->write_begin(), but do add other dirty data to the region being written to
try and reduce the number of RPC ops.

This requires ->migrate_folio() to be implemented, so point that at
filemap_migrate_folio() for files and also for symlinks and directories.

This can be tested by turning on the afs_folio_dirty tracepoint and then
doing something like:

   xfs_io -c "w 2223 7000" -c "w 15000 22222" -c "w 23 7" /afs/my/test/foo

and then looking in the trace to see if the write at position 15000 gets
stored before page 0 gets dirtied for the write at position 23.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: Christoph Hellwig <hch@lst.de>
cc: Matthew Wilcox <willy@infradead.org>
cc: linux-afs@lists.infradead.org
Link: https://lore.kernel.org/r/20221113162902.883850-1-hch@lst.de/ [1]
Link: https://lore.kernel.org/r/166876785552.222254.4403222906022558715.stgit@warthog.procyon.org.uk/ # v1

1 files changed, 1 insertions, 0 deletions

diff --git a/fs/afs/dir.c b/fs/afs/dir.c
index 104df2964225..b7c1f8c84b38 100644
--- a/fs/afs/dir.c
+++ b/fs/afs/dir.c
@@ -77,6 +77,7 @@ const struct address_space_operations afs_dir_aops = {
 	.dirty_folio	= afs_dir_dirty_folio,
 	.release_folio	= afs_dir_release_folio,
 	.invalidate_folio = afs_dir_invalidate_folio,
+	.migrate_folio	= filemap_migrate_folio,
 };
 
 const struct dentry_operations afs_fs_dentry_operations = {