LU-15776 tgt: fix transaction handling in tgt_brw_write()

Hotfix to prevent possible data loss during WRITE replay.
Since commit f0f92773ee18 from LU-14187 the obd_commitrw()
may restart write transaction in OFD and MDT. That causes
transaction number to be assigned multiple times if such
restart happens. Without flag tti_mult_trans the first
transaction number is stored only so later one could remain
not applied causing data loss after recovery.

Patch sets tti_mult_trans for tgt_brw_write() so the latest
transaction number will be used as request transno.

Fixes: f0f92773ee ("LU-14187 osd-ldiskfs: fix locking in write commit")
Signed-off-by: Mikhail Pershin <mpershin@whamcloud.com>
Change-Id: I364b478591942be5562c3e98ee6e6aa487f3e0c5
Reviewed-on: https://review.whamcloud.com/47371
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Alex Zhuravlev <bzzz@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Wang Shilong <wangshilong1991@gmail.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/target/tgt_handler.c b/lustre/target/tgt_handler.c
index 681f3d4..c42c74b 100644 (file)
--- a/lustre/target/tgt_handler.c
+++ b/lustre/target/tgt_handler.c
@@ -2604,6 +2604,7 @@ static void tgt_warn_on_cksum(struct ptlrpc_request *req,
 
 int tgt_brw_write(struct tgt_session_info *tsi)
 {
+       struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
        struct ptlrpc_request   *req = tgt_ses_req(tsi);
        struct ptlrpc_bulk_desc *desc = NULL;
        struct obd_export       *exp = req->rq_export;
@@ -2846,6 +2847,9 @@ out_commitrw:
                nob += len;
        }
 
+       /* multiple transactions can be assigned during write commit */
+       tti->tti_mult_trans = 1;
+
        /* Must commit after prep above in all cases */
        rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
                          objcount, ioo, remote_nb, npages, local_nb, rc, nob,