2019-05-28 19:57:20 +03:00
|
|
|
// SPDX-License-Identifier: GPL-2.0-only
|
2006-01-18 12:30:29 +03:00
|
|
|
/******************************************************************************
|
|
|
|
*******************************************************************************
|
|
|
|
**
|
2007-09-28 00:53:38 +04:00
|
|
|
** Copyright (C) 2005-2007 Red Hat, Inc. All rights reserved.
|
2006-01-18 12:30:29 +03:00
|
|
|
**
|
|
|
|
**
|
|
|
|
*******************************************************************************
|
|
|
|
******************************************************************************/
|
|
|
|
|
|
|
|
#include "dlm_internal.h"
|
|
|
|
#include "member.h"
|
|
|
|
#include "lock.h"
|
|
|
|
#include "dir.h"
|
|
|
|
#include "config.h"
|
|
|
|
#include "requestqueue.h"
|
2022-04-04 23:06:41 +03:00
|
|
|
#include "util.h"
|
2006-01-18 12:30:29 +03:00
|
|
|
|
|
|
|
struct rq_entry {
|
|
|
|
struct list_head list;
|
2012-04-24 01:36:01 +04:00
|
|
|
uint32_t recover_seq;
|
2006-01-18 12:30:29 +03:00
|
|
|
int nodeid;
|
2008-01-25 08:28:28 +03:00
|
|
|
struct dlm_message request;
|
2006-01-18 12:30:29 +03:00
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Requests received while the lockspace is in recovery get added to the
|
|
|
|
* request queue and processed when recovery is complete. This happens when
|
|
|
|
* the lockspace is suspended on some nodes before it is on others, or the
|
|
|
|
* lockspace is enabled on some while still suspended on others.
|
|
|
|
*/
|
|
|
|
|
2008-01-25 08:28:28 +03:00
|
|
|
void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_message *ms)
|
2006-01-18 12:30:29 +03:00
|
|
|
{
|
|
|
|
struct rq_entry *e;
|
2022-04-04 23:06:39 +03:00
|
|
|
int length = le16_to_cpu(ms->m_header.h_length) -
|
|
|
|
sizeof(struct dlm_message);
|
2006-01-18 12:30:29 +03:00
|
|
|
|
2009-12-01 01:34:43 +03:00
|
|
|
e = kmalloc(sizeof(struct rq_entry) + length, GFP_NOFS);
|
2006-01-18 12:30:29 +03:00
|
|
|
if (!e) {
|
2007-09-28 00:53:38 +04:00
|
|
|
log_print("dlm_add_requestqueue: out of memory len %d", length);
|
|
|
|
return;
|
2006-01-18 12:30:29 +03:00
|
|
|
}
|
|
|
|
|
2012-04-24 01:36:01 +04:00
|
|
|
e->recover_seq = ls->ls_recover_seq & 0xFFFFFFFF;
|
2006-01-18 12:30:29 +03:00
|
|
|
e->nodeid = nodeid;
|
2022-04-04 23:06:39 +03:00
|
|
|
memcpy(&e->request, ms, le16_to_cpu(ms->m_header.h_length));
|
2006-01-18 12:30:29 +03:00
|
|
|
|
2021-11-02 22:17:17 +03:00
|
|
|
atomic_inc(&ls->ls_requestqueue_cnt);
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_lock(&ls->ls_requestqueue_mutex);
|
2007-09-28 00:53:38 +04:00
|
|
|
list_add_tail(&e->list, &ls->ls_requestqueue);
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_unlock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
}
|
|
|
|
|
2007-09-28 00:53:38 +04:00
|
|
|
/*
|
|
|
|
* Called by dlm_recoverd to process normal messages saved while recovery was
|
|
|
|
* happening. Normal locking has been enabled before this is called. dlm_recv
|
|
|
|
* upon receiving a message, will wait for all saved messages to be drained
|
|
|
|
* here before processing the message it got. If a new dlm_ls_stop() arrives
|
|
|
|
* while we're processing these saved messages, it may block trying to suspend
|
|
|
|
* dlm_recv if dlm_recv is waiting for us in dlm_wait_requestqueue. In that
|
|
|
|
* case, we don't abort since locking_stopped is still 0. If dlm_recv is not
|
|
|
|
* waiting for us, then this processing may be aborted due to locking_stopped.
|
|
|
|
*/
|
|
|
|
|
2006-01-18 12:30:29 +03:00
|
|
|
int dlm_process_requestqueue(struct dlm_ls *ls)
|
|
|
|
{
|
|
|
|
struct rq_entry *e;
|
dlm: fixes for nodir mode
The "nodir" mode (statically assign master nodes instead
of using the resource directory) has always been highly
experimental, and never seriously used. This commit
fixes a number of problems, making nodir much more usable.
- Major change to recovery: recover all locks and restart
all in-progress operations after recovery. In some
cases it's not possible to know which in-progess locks
to recover, so recover all. (Most require recovery
in nodir mode anyway since rehashing changes most
master nodes.)
- Change the way nodir mode is enabled, from a command
line mount arg passed through gfs2, into a sysfs
file managed by dlm_controld, consistent with the
other config settings.
- Allow recovering MSTCPY locks on an rsb that has not
yet been turned into a master copy.
- Ignore RCOM_LOCK and RCOM_LOCK_REPLY recovery messages
from a previous, aborted recovery cycle. Base this
on the local recovery status not being in the state
where any nodes should be sending LOCK messages for the
current recovery cycle.
- Hold rsb lock around dlm_purge_mstcpy_locks() because it
may run concurrently with dlm_recover_master_copy().
- Maintain highbast on process-copy lkb's (in addition to
the master as is usual), because the lkb can switch
back and forth between being a master and being a
process copy as the master node changes in recovery.
- When recovering MSTCPY locks, flag rsb's that have
non-empty convert or waiting queues for granting
at the end of recovery. (Rename flag from LOCKS_PURGED
to RECOVER_GRANT and similar for the recovery function,
because it's not only resources with purged locks
that need grant a grant attempt.)
- Replace a couple of unnecessary assertion panics with
error messages.
Signed-off-by: David Teigland <teigland@redhat.com>
2012-04-27 00:54:29 +04:00
|
|
|
struct dlm_message *ms;
|
2006-01-18 12:30:29 +03:00
|
|
|
int error = 0;
|
|
|
|
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_lock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
if (list_empty(&ls->ls_requestqueue)) {
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_unlock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
error = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
e = list_entry(ls->ls_requestqueue.next, struct rq_entry, list);
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_unlock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
|
dlm: fixes for nodir mode
The "nodir" mode (statically assign master nodes instead
of using the resource directory) has always been highly
experimental, and never seriously used. This commit
fixes a number of problems, making nodir much more usable.
- Major change to recovery: recover all locks and restart
all in-progress operations after recovery. In some
cases it's not possible to know which in-progess locks
to recover, so recover all. (Most require recovery
in nodir mode anyway since rehashing changes most
master nodes.)
- Change the way nodir mode is enabled, from a command
line mount arg passed through gfs2, into a sysfs
file managed by dlm_controld, consistent with the
other config settings.
- Allow recovering MSTCPY locks on an rsb that has not
yet been turned into a master copy.
- Ignore RCOM_LOCK and RCOM_LOCK_REPLY recovery messages
from a previous, aborted recovery cycle. Base this
on the local recovery status not being in the state
where any nodes should be sending LOCK messages for the
current recovery cycle.
- Hold rsb lock around dlm_purge_mstcpy_locks() because it
may run concurrently with dlm_recover_master_copy().
- Maintain highbast on process-copy lkb's (in addition to
the master as is usual), because the lkb can switch
back and forth between being a master and being a
process copy as the master node changes in recovery.
- When recovering MSTCPY locks, flag rsb's that have
non-empty convert or waiting queues for granting
at the end of recovery. (Rename flag from LOCKS_PURGED
to RECOVER_GRANT and similar for the recovery function,
because it's not only resources with purged locks
that need grant a grant attempt.)
- Replace a couple of unnecessary assertion panics with
error messages.
Signed-off-by: David Teigland <teigland@redhat.com>
2012-04-27 00:54:29 +04:00
|
|
|
ms = &e->request;
|
|
|
|
|
|
|
|
log_limit(ls, "dlm_process_requestqueue msg %d from %d "
|
|
|
|
"lkid %x remid %x result %d seq %u",
|
2022-04-04 23:06:41 +03:00
|
|
|
le32_to_cpu(ms->m_type),
|
|
|
|
le32_to_cpu(ms->m_header.h_nodeid),
|
|
|
|
le32_to_cpu(ms->m_lkid), le32_to_cpu(ms->m_remid),
|
|
|
|
from_dlm_errno(le32_to_cpu(ms->m_result)),
|
dlm: fixes for nodir mode
The "nodir" mode (statically assign master nodes instead
of using the resource directory) has always been highly
experimental, and never seriously used. This commit
fixes a number of problems, making nodir much more usable.
- Major change to recovery: recover all locks and restart
all in-progress operations after recovery. In some
cases it's not possible to know which in-progess locks
to recover, so recover all. (Most require recovery
in nodir mode anyway since rehashing changes most
master nodes.)
- Change the way nodir mode is enabled, from a command
line mount arg passed through gfs2, into a sysfs
file managed by dlm_controld, consistent with the
other config settings.
- Allow recovering MSTCPY locks on an rsb that has not
yet been turned into a master copy.
- Ignore RCOM_LOCK and RCOM_LOCK_REPLY recovery messages
from a previous, aborted recovery cycle. Base this
on the local recovery status not being in the state
where any nodes should be sending LOCK messages for the
current recovery cycle.
- Hold rsb lock around dlm_purge_mstcpy_locks() because it
may run concurrently with dlm_recover_master_copy().
- Maintain highbast on process-copy lkb's (in addition to
the master as is usual), because the lkb can switch
back and forth between being a master and being a
process copy as the master node changes in recovery.
- When recovering MSTCPY locks, flag rsb's that have
non-empty convert or waiting queues for granting
at the end of recovery. (Rename flag from LOCKS_PURGED
to RECOVER_GRANT and similar for the recovery function,
because it's not only resources with purged locks
that need grant a grant attempt.)
- Replace a couple of unnecessary assertion panics with
error messages.
Signed-off-by: David Teigland <teigland@redhat.com>
2012-04-27 00:54:29 +04:00
|
|
|
e->recover_seq);
|
|
|
|
|
2012-04-24 01:36:01 +04:00
|
|
|
dlm_receive_message_saved(ls, &e->request, e->recover_seq);
|
2006-01-18 12:30:29 +03:00
|
|
|
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_lock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
list_del(&e->list);
|
2021-11-02 22:17:17 +03:00
|
|
|
if (atomic_dec_and_test(&ls->ls_requestqueue_cnt))
|
|
|
|
wake_up(&ls->ls_requestqueue_wait);
|
2006-01-18 12:30:29 +03:00
|
|
|
kfree(e);
|
|
|
|
|
|
|
|
if (dlm_locking_stopped(ls)) {
|
|
|
|
log_debug(ls, "process_requestqueue abort running");
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_unlock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
error = -EINTR;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
schedule();
|
|
|
|
}
|
|
|
|
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* After recovery is done, locking is resumed and dlm_recoverd takes all the
|
2007-09-28 00:53:38 +04:00
|
|
|
* saved requests and processes them as they would have been by dlm_recv. At
|
|
|
|
* the same time, dlm_recv will start receiving new requests from remote nodes.
|
|
|
|
* We want to delay dlm_recv processing new requests until dlm_recoverd has
|
|
|
|
* finished processing the old saved requests. We don't check for locking
|
|
|
|
* stopped here because dlm_ls_stop won't stop locking until it's suspended us
|
|
|
|
* (dlm_recv).
|
2006-01-18 12:30:29 +03:00
|
|
|
*/
|
|
|
|
|
|
|
|
void dlm_wait_requestqueue(struct dlm_ls *ls)
|
|
|
|
{
|
2021-11-02 22:17:17 +03:00
|
|
|
wait_event(ls->ls_requestqueue_wait,
|
|
|
|
atomic_read(&ls->ls_requestqueue_cnt) == 0);
|
2006-01-18 12:30:29 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
static int purge_request(struct dlm_ls *ls, struct dlm_message *ms, int nodeid)
|
|
|
|
{
|
2022-04-04 23:06:41 +03:00
|
|
|
__le32 type = ms->m_type;
|
2006-01-18 12:30:29 +03:00
|
|
|
|
2006-11-27 20:31:22 +03:00
|
|
|
/* the ls is being cleaned up and freed by release_lockspace */
|
2021-11-02 22:17:18 +03:00
|
|
|
if (!atomic_read(&ls->ls_count))
|
2006-11-27 20:31:22 +03:00
|
|
|
return 1;
|
|
|
|
|
2006-01-18 12:30:29 +03:00
|
|
|
if (dlm_is_removed(ls, nodeid))
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
/* directory operations are always purged because the directory is
|
|
|
|
always rebuilt during recovery and the lookups resent */
|
|
|
|
|
2022-04-04 23:06:41 +03:00
|
|
|
if (type == cpu_to_le32(DLM_MSG_REMOVE) ||
|
|
|
|
type == cpu_to_le32(DLM_MSG_LOOKUP) ||
|
|
|
|
type == cpu_to_le32(DLM_MSG_LOOKUP_REPLY))
|
2006-01-18 12:30:29 +03:00
|
|
|
return 1;
|
|
|
|
|
|
|
|
if (!dlm_no_directory(ls))
|
|
|
|
return 0;
|
|
|
|
|
dlm: fixes for nodir mode
The "nodir" mode (statically assign master nodes instead
of using the resource directory) has always been highly
experimental, and never seriously used. This commit
fixes a number of problems, making nodir much more usable.
- Major change to recovery: recover all locks and restart
all in-progress operations after recovery. In some
cases it's not possible to know which in-progess locks
to recover, so recover all. (Most require recovery
in nodir mode anyway since rehashing changes most
master nodes.)
- Change the way nodir mode is enabled, from a command
line mount arg passed through gfs2, into a sysfs
file managed by dlm_controld, consistent with the
other config settings.
- Allow recovering MSTCPY locks on an rsb that has not
yet been turned into a master copy.
- Ignore RCOM_LOCK and RCOM_LOCK_REPLY recovery messages
from a previous, aborted recovery cycle. Base this
on the local recovery status not being in the state
where any nodes should be sending LOCK messages for the
current recovery cycle.
- Hold rsb lock around dlm_purge_mstcpy_locks() because it
may run concurrently with dlm_recover_master_copy().
- Maintain highbast on process-copy lkb's (in addition to
the master as is usual), because the lkb can switch
back and forth between being a master and being a
process copy as the master node changes in recovery.
- When recovering MSTCPY locks, flag rsb's that have
non-empty convert or waiting queues for granting
at the end of recovery. (Rename flag from LOCKS_PURGED
to RECOVER_GRANT and similar for the recovery function,
because it's not only resources with purged locks
that need grant a grant attempt.)
- Replace a couple of unnecessary assertion panics with
error messages.
Signed-off-by: David Teigland <teigland@redhat.com>
2012-04-27 00:54:29 +04:00
|
|
|
return 1;
|
2006-01-18 12:30:29 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
void dlm_purge_requestqueue(struct dlm_ls *ls)
|
|
|
|
{
|
|
|
|
struct dlm_message *ms;
|
|
|
|
struct rq_entry *e, *safe;
|
|
|
|
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_lock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
list_for_each_entry_safe(e, safe, &ls->ls_requestqueue, list) {
|
2008-01-25 08:28:28 +03:00
|
|
|
ms = &e->request;
|
2006-01-18 12:30:29 +03:00
|
|
|
|
|
|
|
if (purge_request(ls, ms, e->nodeid)) {
|
|
|
|
list_del(&e->list);
|
2021-11-02 22:17:17 +03:00
|
|
|
if (atomic_dec_and_test(&ls->ls_requestqueue_cnt))
|
|
|
|
wake_up(&ls->ls_requestqueue_wait);
|
2006-01-18 12:30:29 +03:00
|
|
|
kfree(e);
|
|
|
|
}
|
|
|
|
}
|
2006-01-20 11:47:07 +03:00
|
|
|
mutex_unlock(&ls->ls_requestqueue_mutex);
|
2006-01-18 12:30:29 +03:00
|
|
|
}
|
|
|
|
|