LU-14807 lfsck: fix race in lfsck_pos_fill

There is a race for lfsck->li_di_dir between lfsck_di_dir_put and
lfsck_pos_fill, which could cause lfsck_pos_fill to use freed
lfsck->li_di_dir (struct osd_it_ea) and trigger GPF.

Lustre-change: https://review.whamcloud.com/44130
Lustre-commit: 911f638bd6c547591e784fcec668fe9811916e21

Change-Id: Iedadf03ac15d128bb051aea8aafa24dbcd2704fb
Signed-off-by: Hongchao Zhang <hongchao@whamcloud.com>
Reviewed-by: Alex Zhuravlev <bzzz@whamcloud.com>
Reviewed-by: Lai Siyao <lai.siyao@whamcloud.com>
Signed-off-by: Minh Diep <mdiep@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-on: https://review.whamcloud.com/46020
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: John L. Hammond <jhammond@whamcloud.com>

diff --git a/lustre/lfsck/lfsck_lib.c b/lustre/lfsck/lfsck_lib.c
index 2c65436..873d494 100644 (file)
--- a/lustre/lfsck/lfsck_lib.c
+++ b/lustre/lfsck/lfsck_lib.c
@@ -1847,6 +1847,7 @@ void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
        if (unlikely(pos->lp_oit_cookie == 0))
                pos->lp_oit_cookie = 1;
 
+       spin_lock(&lfsck->li_lock);
        if (lfsck->li_di_dir != NULL) {
                struct dt_object *dto = lfsck->li_obj_dir;
 
@@ -1863,6 +1864,7 @@ void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
                fid_zero(&pos->lp_dir_parent);
                pos->lp_dir_cookie = 0;
        }
+       spin_unlock(&lfsck->li_lock);
 }
 
 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)