[PATCH v6 42/51] mm: Handle truncates that split THPs

[Matthew Wilcox <willy@infradead.org>]

From: "Matthew Wilcox (Oracle)" <willy@infradead.org>

Move shmem_punch_compound() to truncate.c and rename it to punch_thp().
Change its arguments to loff_t to make calling do_invalidatepage()
easier.  Call it when we find a THP in the cache.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
---
 mm/internal.h |  2 ++
 mm/shmem.c    | 30 ++-------------------------
 mm/truncate.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++--
 3 files changed, 59 insertions(+), 30 deletions(-)

diff --git a/mm/internal.h b/mm/internal.h
index ac3c79408045..cd7038a36354 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -613,4 +613,6 @@ static inline bool is_migrate_highatomic_page(struct page *page)
 
 void setup_zone_pageset(struct zone *zone);
 extern struct page *alloc_new_node_page(struct page *page, unsigned long node);
+
+bool punch_thp(struct page *page, loff_t start, loff_t end);
 #endif	/* __MM_INTERNAL_H */
diff --git a/mm/shmem.c b/mm/shmem.c
index 55405d811cfd..495b8684d94a 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -804,32 +804,6 @@ void shmem_unlock_mapping(struct address_space *mapping)
 	}
 }
 
-/*
- * Check whether a hole-punch or truncation needs to split a huge page,
- * returning true if no split was required, or the split has been successful.
- *
- * Eviction (or truncation to 0 size) should never need to split a huge page;
- * but in rare cases might do so, if shmem_undo_range() failed to trylock on
- * head, and then succeeded to trylock on tail.
- *
- * A split can only succeed when there are no additional references on the
- * huge page: so the split below relies upon find_get_entries() having stopped
- * when it found a subpage of the huge page, without getting further references.
- */
-static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
-{
-	if (!PageTransCompound(page))
-		return true;
-
-	/* Just proceed to delete a huge page wholly within the range punched */
-	if (PageHead(page) &&
-	    page->index >= start && page->index + HPAGE_PMD_NR <= end)
-		return true;
-
-	/* Try to split huge page, so we can truly punch the hole or truncate */
-	return split_huge_page(page) >= 0;
-}
-
 /*
  * Remove range of pages and swap entries from page cache, and free them.
  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
@@ -883,7 +857,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 			if ((!unfalloc || !PageUptodate(page)) &&
 			    page_mapping(page) == mapping) {
 				VM_BUG_ON_PAGE(PageWriteback(page), page);
-				if (shmem_punch_compound(page, start, end))
+				if (punch_thp(page, lstart, lend))
 					truncate_inode_page(mapping, page);
 			}
 			unlock_page(page);
@@ -973,7 +947,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 					break;
 				}
 				VM_BUG_ON_PAGE(PageWriteback(page), page);
-				if (shmem_punch_compound(page, start, end))
+				if (punch_thp(page, lstart, lend))
 					truncate_inode_page(mapping, page);
 				else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
 					/* Wipe the page and don't get stuck */
diff --git a/mm/truncate.c b/mm/truncate.c
index a9fde773179b..0ef2001c2f65 100644
--- a/mm/truncate.c
+++ b/mm/truncate.c
@@ -229,6 +229,55 @@ int truncate_inode_page(struct address_space *mapping, struct page *page)
 	return 0;
 }
 
+/*
+ * Check whether a hole-punch or truncation needs to split a huge page,
+ * returning true if no split was required, or the split has been
+ * successful.
+ *
+ * Eviction (or truncation to 0 size) should never need to split a huge
+ * page; but in rare cases might do so, if shmem_undo_range() failed to
+ * trylock on head, and then succeeded to trylock on tail.
+ *
+ * A split can only succeed when there are no additional references on
+ * the huge page: so the split below relies upon find_get_entries()
+ * having stopped when it found a subpage of the huge page, without
+ * getting further references.
+ */
+bool punch_thp(struct page *page, loff_t start, loff_t end)
+{
+	struct page *head = thp_head(page);
+	loff_t pos = page_offset(head);
+	unsigned int offset, length;
+
+	if (!PageTransCompound(page))
+		return true;
+
+	if (pos < start)
+		offset = start - pos;
+	else
+		offset = 0;
+	length = thp_size(head);
+	if (pos + length < end)
+		length = length - offset;
+	else
+		length = end - pos - offset;
+
+	/* Just proceed to delete a huge page wholly within the range punched */
+	if (length == thp_size(head))
+		return true;
+
+	/*
+	 * We're going to split the page into order-0 pages.  Tell the
+	 * filesystem which range of the page is going to be punched out
+	 * so it can discard unnecessary private data.
+	 */
+	if (page_has_private(head))
+		do_invalidatepage(head, offset, length);
+
+	/* Try to split huge page, so we can truly punch the hole or truncate */
+	return split_huge_page(page) >= 0;
+}
+
 /*
  * Used to get rid of pages on hardware memory corruption.
  */
@@ -359,7 +408,10 @@ void truncate_inode_pages_range(struct address_space *mapping,
 				unlock_page(page);
 				continue;
 			}
-			pagevec_add(&locked_pvec, page);
+			if (punch_thp(page, lstart, lend))
+				pagevec_add(&locked_pvec, page);
+			else
+				unlock_page(page);
 		}
 		for (i = 0; i < pagevec_count(&locked_pvec); i++)
 			truncate_cleanup_page(mapping, locked_pvec.pages[i]);
@@ -453,7 +505,8 @@ void truncate_inode_pages_range(struct address_space *mapping,
 			lock_page(page);
 			WARN_ON(page_to_index(page) != index);
 			wait_on_page_writeback(page);
-			truncate_inode_page(mapping, page);
+			if (punch_thp(page, lstart, lend))
+				truncate_inode_page(mapping, page);
 			unlock_page(page);
 		}
 		truncate_exceptional_pvec_entries(mapping, &pvec, indices, end);
-- 
2.26.2