Zero pad bytes following TX_WRITE log data

When logging a TX_WRITE record in the case where file data has to be
copied from the DMU, we pad the log record size to a multiple of 8
bytes.  In this case, any padding bytes should be zeroed, otherwise the
contents of uninitialized memory are written to the ZIL.

This was found using KMSAN.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #12383

module/zfs/zil.c

@@ -1619,7 +1619,7 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 	lr_t *lrcb, *lrc;
 	lr_write_t *lrwb, *lrw;
 	char *lr_buf;
-	uint64_t dlen, dnow, lwb_sp, reclen, txg, max_log_data;
+	uint64_t dlen, dnow, dpad, lwb_sp, reclen, txg, max_log_data;
 
 	ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
 	ASSERT3P(lwb, !=, NULL);
@@ -1653,8 +1653,9 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 	if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) {
 		dlen = P2ROUNDUP_TYPED(
 		    lrw->lr_length, sizeof (uint64_t), uint64_t);
+		dpad = dlen - lrw->lr_length;
 	} else {
-		dlen = 0;
+		dlen = dpad = 0;
 	}
 	reclen = lrc->lrc_reclen;
 	zilog->zl_cur_used += (reclen + dlen);
@@ -1748,6 +1749,9 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 			error = zilog->zl_get_data(itx->itx_private,
 			    itx->itx_gen, lrwb, dbuf, lwb,
 			    lwb->lwb_write_zio);
+			if (dbuf != NULL && error == 0 && dnow == dlen)
+				/* Zero any padding bytes in the last block. */
+				bzero((char *)dbuf + lrwb->lr_length, dpad);
 
 			if (error == EIO) {
 				txg_wait_synced(zilog->zl_dmu_pool, txg);