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. Signed-off-by: tstabrawa <[email protected]>
openzfs · Sep 25, 2024 · 202e6a4 · 202e6a4
1 parent 832f66b
commit 202e6a4
Show file tree

Hide file tree

Showing 6 changed files with 140 additions and 0 deletions.
diff --git a/config/kernel-vfs-invalidate_folio.m4 b/config/kernel-vfs-invalidate_folio.m4
@@ -0,0 +1,33 @@
+dnl #
+dnl # Linux 5.18 uses invalidate_folio in lieu of invalidate_page
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_VFS_INVALIDATE_FOLIO], [
+	ZFS_LINUX_TEST_SRC([vfs_has_invalidate_folio], [
+		#include <linux/fs.h>
+
+		static void
+		test_invalidate_folio(struct folio *folio, size_t offset,
+		                      size_t len) {
+			(void) folio; (void) offset; (void) len;
+			return;
+		}
+
+		static const struct address_space_operations
+		    aops __attribute__ ((unused)) = {
+			.invalidate_folio	= test_invalidate_folio,
+		};
+	],[])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_VFS_INVALIDATE_FOLIO], [
+	dnl #
+	dnl # Linux 5.18 uses invalidate_folio in lieu of invalidate_page
+	dnl #
+	AC_MSG_CHECKING([whether invalidate_folio exists])
+	ZFS_LINUX_TEST_RESULT([vfs_has_invalidate_folio], [
+		AC_MSG_RESULT([yes])
+		AC_DEFINE(HAVE_VFS_INVALIDATE_FOLIO, 1, [invalidate_folio exists])
+	],[
+		AC_MSG_RESULT([no])
+	])
+])
diff --git a/config/kernel-vfs-release_folio.m4 b/config/kernel-vfs-release_folio.m4
@@ -0,0 +1,32 @@
+dnl #
+dnl # Linux 5.19 uses release_folio in lieu of releasepage
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_VFS_RELEASE_FOLIO], [
+	ZFS_LINUX_TEST_SRC([vfs_has_release_folio], [
+		#include <linux/fs.h>
+
+		static bool
+		test_release_folio(struct folio *folio, gfp_t gfp) {
+			(void) folio; (void) gfp;
+			return (0);
+		}
+
+		static const struct address_space_operations
+		    aops __attribute__ ((unused)) = {
+			.release_folio	= test_release_folio,
+		};
+	],[])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_VFS_RELEASE_FOLIO], [
+	dnl #
+	dnl # Linux 5.19 uses release_folio in lieu of releasepage
+	dnl #
+	AC_MSG_CHECKING([whether release_folio exists])
+	ZFS_LINUX_TEST_RESULT([vfs_has_release_folio], [
+		AC_MSG_RESULT([yes])
+		AC_DEFINE(HAVE_VFS_RELEASE_FOLIO, 1, [release_folio exists])
+	],[
+		AC_MSG_RESULT([no])
+	])
+])
diff --git a/config/kernel.m4 b/config/kernel.m4
@@ -77,6 +77,8 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
 	ZFS_AC_KERNEL_SRC_SGET
 	ZFS_AC_KERNEL_SRC_VFS_FILEMAP_DIRTY_FOLIO
 	ZFS_AC_KERNEL_SRC_VFS_READ_FOLIO
+	ZFS_AC_KERNEL_SRC_VFS_RELEASE_FOLIO
+	ZFS_AC_KERNEL_SRC_VFS_INVALIDATE_FOLIO
 	ZFS_AC_KERNEL_SRC_VFS_FSYNC_2ARGS
 	ZFS_AC_KERNEL_SRC_VFS_DIRECT_IO
 	ZFS_AC_KERNEL_SRC_VFS_READPAGES
@@ -185,6 +187,8 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
 	ZFS_AC_KERNEL_SGET
 	ZFS_AC_KERNEL_VFS_FILEMAP_DIRTY_FOLIO
 	ZFS_AC_KERNEL_VFS_READ_FOLIO
+	ZFS_AC_KERNEL_VFS_RELEASE_FOLIO
+	ZFS_AC_KERNEL_VFS_INVALIDATE_FOLIO
 	ZFS_AC_KERNEL_VFS_FSYNC_2ARGS
 	ZFS_AC_KERNEL_VFS_DIRECT_IO
 	ZFS_AC_KERNEL_VFS_READPAGES

diff --git 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
@@ -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
@@ -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 = {