From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1759271AbYAXRCH (ORCPT ); Thu, 24 Jan 2008 12:02:07 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1758313AbYAXQ6o (ORCPT ); Thu, 24 Jan 2008 11:58:44 -0500 Received: from mx1.redhat.com ([66.187.233.31]:49780 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756958AbYAXQ6j (ORCPT ); Thu, 24 Jan 2008 11:58:39 -0500 From: David Teigland To: linux-kernel@vger.kernel.org Subject: [PATCH 10/19] dlm: recover locks waiting for overlap replies Date: Thu, 24 Jan 2008 10:50:33 -0600 Message-Id: <1201193442-8260-11-git-send-email-teigland@redhat.com> X-Mailer: git-send-email 1.5.3.3 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org When recovery looks at locks waiting for replies, it fails to consider locks that have already received a reply for their first remote operation, but not received a reply for secondary, overlapping unlock/cancel. The appropriate stub reply needs to be called for these waiters. Appears when we start doing recovery in the presence of a many overlapping unlock/cancel ops. Signed-off-by: David Teigland --- fs/dlm/lock.c | 37 ++++++++++++++++++++++++++++++++----- 1 files changed, 32 insertions(+), 5 deletions(-) diff --git a/fs/dlm/lock.c b/fs/dlm/lock.c index 43ca2a3..a758f1b 100644 --- a/fs/dlm/lock.c +++ b/fs/dlm/lock.c @@ -3846,6 +3846,7 @@ static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb) void dlm_recover_waiters_pre(struct dlm_ls *ls) { struct dlm_lkb *lkb, *safe; + int wait_type, stub_unlock_result, stub_cancel_result; mutex_lock(&ls->ls_waiters_mutex); @@ -3864,7 +3865,33 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls) if (!waiter_needs_recovery(ls, lkb)) continue; - switch (lkb->lkb_wait_type) { + wait_type = lkb->lkb_wait_type; + stub_unlock_result = -DLM_EUNLOCK; + stub_cancel_result = -DLM_ECANCEL; + + /* Main reply may have been received leaving a zero wait_type, + but a reply for the overlapping op may not have been + received. In that case we need to fake the appropriate + reply for the overlap op. */ + + if (!wait_type) { + if (is_overlap_cancel(lkb)) { + wait_type = DLM_MSG_CANCEL; + if (lkb->lkb_grmode == DLM_LOCK_IV) + stub_cancel_result = 0; + } + if (is_overlap_unlock(lkb)) { + wait_type = DLM_MSG_UNLOCK; + if (lkb->lkb_grmode == DLM_LOCK_IV) + stub_unlock_result = -ENOENT; + } + + log_debug(ls, "rwpre overlap %x %x %d %d %d", + lkb->lkb_id, lkb->lkb_flags, wait_type, + stub_cancel_result, stub_unlock_result); + } + + switch (wait_type) { case DLM_MSG_REQUEST: lkb->lkb_flags |= DLM_IFL_RESEND; @@ -3877,7 +3904,7 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls) case DLM_MSG_UNLOCK: hold_lkb(lkb); ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY; - ls->ls_stub_ms.m_result = -DLM_EUNLOCK; + ls->ls_stub_ms.m_result = stub_unlock_result; ls->ls_stub_ms.m_flags = lkb->lkb_flags; _receive_unlock_reply(lkb, &ls->ls_stub_ms); dlm_put_lkb(lkb); @@ -3886,15 +3913,15 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls) case DLM_MSG_CANCEL: hold_lkb(lkb); ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY; - ls->ls_stub_ms.m_result = -DLM_ECANCEL; + ls->ls_stub_ms.m_result = stub_cancel_result; ls->ls_stub_ms.m_flags = lkb->lkb_flags; _receive_cancel_reply(lkb, &ls->ls_stub_ms); dlm_put_lkb(lkb); break; default: - log_error(ls, "invalid lkb wait_type %d", - lkb->lkb_wait_type); + log_error(ls, "invalid lkb wait_type %d %d", + lkb->lkb_wait_type, wait_type); } schedule(); } -- 1.5.3.3