Fix corruption caused by mmap flushing problems

1) Make mmap flushes synchronous. Linux may skip flushing dirty pages
   already in writeback unless data-integrity sync is requested.

2) Change zfs_putpage to use TXG_WAIT. Otherwise dirty pages may be
   skipped due to DMU pushing back on TX assign.

3) Add missing mmap flush when doing block cloning.

4) While here, pass errors from putpage to writepage/writepages.

This change fixes corruption edge cases, but unfortunately adds
synchronous ZIL flushes for dirty mmap pages to llseek and bclone
operations. It may be possible to avoid these sync writes later
but would need more tricky refactoring of the writeback code.

Reviewed-by: Alexander Motin <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Robert Evans <[email protected]>
Closes openzfs#15933
Closes openzfs#16019

Author: rrevans

module/os/linux/zfs/zfs_vnops_os.c

@@ -3792,11 +3792,8 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc,
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
 
-	err = dmu_tx_assign(tx, TXG_NOWAIT);
+	err = dmu_tx_assign(tx, TXG_WAIT);
 	if (err != 0) {
-		if (err == ERESTART)
-			dmu_tx_wait(tx);
-
 		dmu_tx_abort(tx);
 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO
 		filemap_dirty_folio(page_mapping(pp), page_folio(pp));

module/os/linux/zfs/zpl_file.c

@@ -720,23 +720,23 @@ zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 {
 	boolean_t *for_sync = data;
 	fstrans_cookie_t cookie;
+	int ret;
 
 	ASSERT(PageLocked(pp));
 	ASSERT(!PageWriteback(pp));
 
 	cookie = spl_fstrans_mark();
-	(void) zfs_putpage(pp->mapping->host, pp, wbc, *for_sync);
+	ret = zfs_putpage(pp->mapping->host, pp, wbc, *for_sync);
 	spl_fstrans_unmark(cookie);
 
-	return (0);
+	return (ret);
 }
 
 #ifdef HAVE_WRITEPAGE_T_FOLIO
 static int
 zpl_putfolio(struct folio *pp, struct writeback_control *wbc, void *data)
 {
-	(void) zpl_putpage(&pp->page, wbc, data);
-	return (0);
+	return (zpl_putpage(&pp->page, wbc, data));
 }
 #endif
 

module/zfs/zfs_vnops.c

@@ -130,7 +130,7 @@ zfs_holey_common(znode_t *zp, ulong_t cmd, loff_t *off)
 
 	/* Flush any mmap()'d data to disk */
 	if (zn_has_cached_data(zp, 0, file_sz - 1))
-		zn_flush_cached_data(zp, B_FALSE);
+		zn_flush_cached_data(zp, B_TRUE);
 
 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_READER);
 	error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
@@ -1193,6 +1193,10 @@ zfs_clone_range(znode_t *inzp, uint64_t *inoffp, znode_t *outzp,
 		}
 	}
 
+	/* Flush any mmap()'d data to disk */
+	if (zn_has_cached_data(inzp, inoff, inoff + len - 1))
+		zn_flush_cached_data(inzp, B_TRUE);
+
 	/*
 	 * Maintain predictable lock order.
 	 */