[RFC,1/9] splice: don't check for uptodate if partially uptodate is impl

Message ID 20240209142901.126894-2-da.gomez@samsung.com
State New
Headers
Series shmem: fix llseek in hugepages |

Commit Message

Daniel Gomez Feb. 9, 2024, 2:29 p.m. UTC
  From: Pankaj Raghav <p.raghav@samsung.com>

When huge_page=always is set in tmpfs, it will zero out the whole page even
if only a small part of it is written, and it updates the uptodate flag of the
whole huge page.

Once the per-block uptodate tracking is implemented for tmpfs hugepages,
pipe_buf_confirm only needs to check the range it needs to splice to be
uptodate and not the whole folio as we don't set uptodate flag for partial
writes.

Signed-off-by: Pankaj Raghav <p.raghav@samsung.com>
Signed-off-by: Daniel Gomez <da.gomez@samsung.com>
---

Other option here is to have a separate implementation of
page_cache_pipe_buf_ops  for tmpfs instead of changing the
page_cache_pipe_buf_confirm.

 fs/splice.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)
  

Patch

diff --git a/fs/splice.c b/fs/splice.c
index 218e24b1ac40..e6ac57795590 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -120,7 +120,9 @@  static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
 				       struct pipe_buffer *buf)
 {
 	struct folio *folio = page_folio(buf->page);
+	const struct address_space_operations *ops;
 	int err;
+	off_t off = folio_page_idx(folio, buf->page) * PAGE_SIZE + buf->offset;
 
 	if (!folio_test_uptodate(folio)) {
 		folio_lock(folio);
@@ -134,12 +136,21 @@  static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
 			goto error;
 		}
 
+		ops = folio->mapping->a_ops;
 		/*
 		 * Uh oh, read-error from disk.
 		 */
-		if (!folio_test_uptodate(folio)) {
-			err = -EIO;
-			goto error;
+		if (!ops->is_partially_uptodate) {
+			if (!folio_test_uptodate(folio)) {
+				err = -EIO;
+				goto error;
+			}
+		} else {
+			if (!ops->is_partially_uptodate(folio, off,
+							buf->len)) {
+				err = -EIO;
+				goto error;
+			}
 		}
 
 		/* Folio is ok after all, we are done */