Avoid BUG in migrate_folio_extra

Linux page migration code won't wait for writeback to complete unless
it needs to call release_folio.  Call SetPagePrivate wherever
PageUptodate is set and define .release_folio, to cause
fallback_migrate_folio to wait for us.

Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: tstabrawa <[email protected]>
Closes #15140
Closes #16568 

3 files changed, 71 insertions, 0 deletions

diff --git a/module/os/linux/zfs/zfs_vnops_os.c b/module/os/linux/zfs/zfs_vnops_os.c
index 7d3a7b8d9..3dce3660c 100644
--- a/module/os/linux/zfs/zfs_vnops_os.c
+++ b/module/os/linux/zfs/zfs_vnops_os.c
@@ -260,6 +260,15 @@ update_pages(znode_t *zp, int64_t start, int len, objset_t *os)
 			} else {
 				ClearPageError(pp);
 				SetPageUptodate(pp);
+				if (!PagePrivate(pp)) {
+					/*
+					 * Set private bit so page migration
+					 * will wait for us to finish writeback
+					 * before calling migrate_folio().
+					 */
+					SetPagePrivate(pp);
+					get_page(pp);
+				}
 
 				if (mapping_writably_mapped(mp))
 					flush_dcache_page(pp);
@@ -4037,6 +4046,14 @@ zfs_fillpage(struct inode *ip, struct page *pp)
 	} else {
 		ClearPageError(pp);
 		SetPageUptodate(pp);
+		if (!PagePrivate(pp)) {
+			/*
+			 * Set private bit so page migration will wait for us to
+			 * finish writeback before calling migrate_folio().
+			 */
+			SetPagePrivate(pp);
+			get_page(pp);
+		}
 	}
 
 	return (error);
diff --git a/module/os/linux/zfs/zfs_znode_os.c b/module/os/linux/zfs/zfs_znode_os.c
index e135f9044..cea15ad4b 100644
--- a/module/os/linux/zfs/zfs_znode_os.c
+++ b/module/os/linux/zfs/zfs_znode_os.c
@@ -1576,6 +1576,14 @@ zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
 		mark_page_accessed(pp);
 		SetPageUptodate(pp);
 		ClearPageError(pp);
+		if (!PagePrivate(pp)) {
+			/*
+			 * Set private bit so page migration will wait for us to
+			 * finish writeback before calling migrate_folio().
+			 */
+			SetPagePrivate(pp);
+			get_page(pp);
+		}
 		unlock_page(pp);
 		put_page(pp);
 	}
diff --git a/module/os/linux/zfs/zpl_file.c b/module/os/linux/zfs/zpl_file.c
index 4d1bf1d54..50c63695d 100644
--- a/module/os/linux/zfs/zpl_file.c
+++ b/module/os/linux/zfs/zpl_file.c
@@ -607,6 +607,42 @@ zpl_writepage(struct page *pp, struct writeback_control *wbc)
 	return (zpl_putpage(pp, wbc, &for_sync));
 }
 
+static int
+zpl_releasepage(struct page *pp, gfp_t gfp)
+{
+	if (PagePrivate(pp)) {
+		ClearPagePrivate(pp);
+		put_page(pp);
+	}
+	return (1);
+}
+
+#ifdef HAVE_VFS_RELEASE_FOLIO
+static bool
+zpl_release_folio(struct folio *folio, gfp_t gfp)
+{
+	return (zpl_releasepage(&folio->page, gfp));
+}
+#endif
+
+#ifdef HAVE_VFS_INVALIDATE_FOLIO
+static void
+zpl_invalidate_folio(struct folio *folio, size_t offset, size_t len)
+{
+	if ((offset == 0) && (len == PAGE_SIZE)) {
+		zpl_releasepage(&folio->page, 0);
+	}
+}
+#else
+static void
+zpl_invalidatepage(struct page *pp, unsigned int offset, unsigned int len)
+{
+	if ((offset == 0) && (len == PAGE_SIZE)) {
+		zpl_releasepage(pp, 0);
+	}
+}
+#endif
+
 /*
  * The flag combination which matches the behavior of zfs_space() is
  * FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
@@ -1090,6 +1126,16 @@ const struct address_space_operations zpl_address_space_operations = {
 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
 	.dirty_folio	= filemap_dirty_folio,
 #endif
+#ifdef HAVE_VFS_RELEASE_FOLIO
+	.release_folio	= zpl_release_folio,
+#else
+	.releasepage	= zpl_releasepage,
+#endif
+#ifdef HAVE_VFS_INVALIDATE_FOLIO
+	.invalidate_folio = zpl_invalidate_folio,
+#else
+	.invalidatepage = zpl_invalidatepage,
+#endif
 };
 
 const struct file_operations zpl_file_operations = {