On 20.12.24 16:02, Andrey Zhadchenko wrote:
Here you are also disabling immediate metadata publish. Then you should also call ploop_attach_end_action() for all requests in ploop_locate_new_cluster_and_attach_pio(). Otherwise there we may have completed write() syscall without metadata changes (they are not delayed now but still executed in-parallel)


I missed to restore that line, yes.
When added back it causes us to "explode" on bat write complete,
i will think how to rearrange but any ideas are welcome.



On 12/5/24 22:56, Alexander Atanasov wrote:
Delayed metadata writeback results in a hang, disable
it until it is fixed. Pios end in the waiting list of
md and get stuck there never to complete.

Signed-off-by: Alexander Atanasov <alexander.atana...@virtuozzo.com>
---
  drivers/md/dm-ploop-map.c | 14 +++++++++++---
  1 file changed, 11 insertions(+), 3 deletions(-)

diff --git a/drivers/md/dm-ploop-map.c b/drivers/md/dm-ploop-map.c
index c88359677ea4..6e4d6f862135 100644
--- a/drivers/md/dm-ploop-map.c
+++ b/drivers/md/dm-ploop-map.c
@@ -817,9 +817,10 @@ static void ploop_advance_local_after_bat_wb(struct ploop *ploop,
      /* holes bit map requires bat_lock */
      spin_lock_irqsave(&ploop->bat_lock, flags);
      spin_lock(&md->md_lock);
+#ifdef PLOOP_DELAYWB
      if (piwb->type == PIWB_TYPE_ALLOC)
          goto skip_apply;
-
+#endif
      for (; i < last; i++) {
          if (piwb->type == PIWB_TYPE_DISCARD) {
              ploop_piwb_discard_completed(ploop, success, i + off, dst_clu[i]); @@ -836,7 +837,9 @@ static void ploop_advance_local_after_bat_wb(struct ploop *ploop,
              ploop_hole_set_bit(i + off, ploop);
          }
      }
+#ifdef PLOOP_DELAYWB
  skip_apply:
+#endif
      WARN_ON_ONCE(!test_bit(MD_WRITEBACK, &md->status));
      clear_bit(MD_WRITEBACK, &md->status);
@@ -1187,8 +1190,10 @@ static int ploop_alloc_cluster(struct ploop *ploop, struct ploop_index_wb *piwb,
      to = piwb->kmpage;
      WRITE_ONCE(to[clu], *dst_clu);
+#ifdef PLOOP_DELAYWB
      to = piwb->md->kmpage;
      WRITE_ONCE(to[clu], *dst_clu);
+#endif
  out:
      return ret;
  }
@@ -1481,6 +1486,7 @@ static void ploop_submit_cow_index_wb(struct ploop_cow *cow)
      WARN_ON(to[clu]);
      WRITE_ONCE(to[clu], cow->dst_clu);
+#ifdef PLOOP_DELAYWB
      spin_lock_irqsave(&md->md_lock, flags);
      to = md->kmpage;
      WRITE_ONCE(to[clu], cow->dst_clu);
@@ -1488,7 +1494,7 @@ static void ploop_submit_cow_index_wb(struct ploop_cow *cow)
      spin_unlock_irqrestore(&md->md_lock, flags);
      ploop_md_up_prio(ploop, md);
-
+#endif
      /* Prevent double clearing of holes_bitmap bit on complete_cow() */
      cow->dst_clu = BAT_ENTRY_NONE;
      spin_lock_irq(&ploop->deferred_lock);
@@ -1918,7 +1924,9 @@ static inline int ploop_submit_metadata_writeback(struct ploop *ploop, int force
       */
      llist_for_each_safe(pos, t, ll_wb_batch) {
          md = list_entry((struct list_head *)pos, typeof(*md), wb_link);
-        if (!llist_empty(&md->wait_llist) || force || test_bit(MD_HIGHPRIO, &md->status) ||
+        /* XXX: fixme delay results in a hang - TBD */
+        if (1 || !llist_empty(&md->wait_llist) || force ||
+            test_bit(MD_HIGHPRIO, &md->status) ||
              time_before(md->dirty_timeout, timeout)) {
              /* L1L2 mustn't be redirtyed, when wb in-flight! */
              WARN_ON_ONCE(!test_bit(MD_DIRTY, &md->status));

--
Regards,
Alexander Atanasov

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

Reply via email to