12522fe45SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only
2e7fd4179SDavid Teigland /******************************************************************************
3e7fd4179SDavid Teigland *******************************************************************************
4e7fd4179SDavid Teigland **
5c36258b5SDavid Teigland ** Copyright (C) 2005-2007 Red Hat, Inc. All rights reserved.
6e7fd4179SDavid Teigland **
7e7fd4179SDavid Teigland **
8e7fd4179SDavid Teigland *******************************************************************************
9e7fd4179SDavid Teigland ******************************************************************************/
10e7fd4179SDavid Teigland
11e7fd4179SDavid Teigland #include "dlm_internal.h"
12e7fd4179SDavid Teigland #include "member.h"
13e7fd4179SDavid Teigland #include "lock.h"
14e7fd4179SDavid Teigland #include "dir.h"
15e7fd4179SDavid Teigland #include "config.h"
16e7fd4179SDavid Teigland #include "requestqueue.h"
1700e99ccdSAlexander Aring #include "util.h"
18e7fd4179SDavid Teigland
19e7fd4179SDavid Teigland struct rq_entry {
20e7fd4179SDavid Teigland struct list_head list;
216d40c4a7SDavid Teigland uint32_t recover_seq;
22e7fd4179SDavid Teigland int nodeid;
238b0d8e03SAl Viro struct dlm_message request;
24e7fd4179SDavid Teigland };
25e7fd4179SDavid Teigland
26e7fd4179SDavid Teigland /*
27e7fd4179SDavid Teigland * Requests received while the lockspace is in recovery get added to the
28e7fd4179SDavid Teigland * request queue and processed when recovery is complete. This happens when
29e7fd4179SDavid Teigland * the lockspace is suspended on some nodes before it is on others, or the
30e7fd4179SDavid Teigland * lockspace is enabled on some while still suspended on others.
31e7fd4179SDavid Teigland */
32e7fd4179SDavid Teigland
dlm_add_requestqueue(struct dlm_ls * ls,int nodeid,const struct dlm_message * ms)33*11519351SAlexander Aring void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid,
34*11519351SAlexander Aring const struct dlm_message *ms)
35e7fd4179SDavid Teigland {
36e7fd4179SDavid Teigland struct rq_entry *e;
373428785aSAlexander Aring int length = le16_to_cpu(ms->m_header.h_length) -
383428785aSAlexander Aring sizeof(struct dlm_message);
39e7fd4179SDavid Teigland
40573c24c4SDavid Teigland e = kmalloc(sizeof(struct rq_entry) + length, GFP_NOFS);
41e7fd4179SDavid Teigland if (!e) {
42c36258b5SDavid Teigland log_print("dlm_add_requestqueue: out of memory len %d", length);
43c36258b5SDavid Teigland return;
44e7fd4179SDavid Teigland }
45e7fd4179SDavid Teigland
466d40c4a7SDavid Teigland e->recover_seq = ls->ls_recover_seq & 0xFFFFFFFF;
47e7fd4179SDavid Teigland e->nodeid = nodeid;
48f217d7ccSAlexander Aring memcpy(&e->request, ms, sizeof(*ms));
49f217d7ccSAlexander Aring memcpy(&e->request.m_extra, ms->m_extra, length);
50e7fd4179SDavid Teigland
51164d88abSAlexander Aring atomic_inc(&ls->ls_requestqueue_cnt);
5290135925SDavid Teigland mutex_lock(&ls->ls_requestqueue_mutex);
53e7fd4179SDavid Teigland list_add_tail(&e->list, &ls->ls_requestqueue);
5490135925SDavid Teigland mutex_unlock(&ls->ls_requestqueue_mutex);
55e7fd4179SDavid Teigland }
56e7fd4179SDavid Teigland
57c36258b5SDavid Teigland /*
58c36258b5SDavid Teigland * Called by dlm_recoverd to process normal messages saved while recovery was
59c36258b5SDavid Teigland * happening. Normal locking has been enabled before this is called. dlm_recv
60c36258b5SDavid Teigland * upon receiving a message, will wait for all saved messages to be drained
61c36258b5SDavid Teigland * here before processing the message it got. If a new dlm_ls_stop() arrives
62c36258b5SDavid Teigland * while we're processing these saved messages, it may block trying to suspend
63c36258b5SDavid Teigland * dlm_recv if dlm_recv is waiting for us in dlm_wait_requestqueue. In that
64c36258b5SDavid Teigland * case, we don't abort since locking_stopped is still 0. If dlm_recv is not
65c36258b5SDavid Teigland * waiting for us, then this processing may be aborted due to locking_stopped.
66c36258b5SDavid Teigland */
67c36258b5SDavid Teigland
dlm_process_requestqueue(struct dlm_ls * ls)68e7fd4179SDavid Teigland int dlm_process_requestqueue(struct dlm_ls *ls)
69e7fd4179SDavid Teigland {
70e7fd4179SDavid Teigland struct rq_entry *e;
714875647aSDavid Teigland struct dlm_message *ms;
72e7fd4179SDavid Teigland int error = 0;
73e7fd4179SDavid Teigland
7490135925SDavid Teigland mutex_lock(&ls->ls_requestqueue_mutex);
75e7fd4179SDavid Teigland
76e7fd4179SDavid Teigland for (;;) {
77e7fd4179SDavid Teigland if (list_empty(&ls->ls_requestqueue)) {
7890135925SDavid Teigland mutex_unlock(&ls->ls_requestqueue_mutex);
79e7fd4179SDavid Teigland error = 0;
80e7fd4179SDavid Teigland break;
81e7fd4179SDavid Teigland }
82e7fd4179SDavid Teigland e = list_entry(ls->ls_requestqueue.next, struct rq_entry, list);
8390135925SDavid Teigland mutex_unlock(&ls->ls_requestqueue_mutex);
84e7fd4179SDavid Teigland
854875647aSDavid Teigland ms = &e->request;
864875647aSDavid Teigland
874875647aSDavid Teigland log_limit(ls, "dlm_process_requestqueue msg %d from %d "
884875647aSDavid Teigland "lkid %x remid %x result %d seq %u",
8900e99ccdSAlexander Aring le32_to_cpu(ms->m_type),
9000e99ccdSAlexander Aring le32_to_cpu(ms->m_header.h_nodeid),
9100e99ccdSAlexander Aring le32_to_cpu(ms->m_lkid), le32_to_cpu(ms->m_remid),
9200e99ccdSAlexander Aring from_dlm_errno(le32_to_cpu(ms->m_result)),
934875647aSDavid Teigland e->recover_seq);
944875647aSDavid Teigland
956d40c4a7SDavid Teigland dlm_receive_message_saved(ls, &e->request, e->recover_seq);
96e7fd4179SDavid Teigland
9790135925SDavid Teigland mutex_lock(&ls->ls_requestqueue_mutex);
98e7fd4179SDavid Teigland list_del(&e->list);
99164d88abSAlexander Aring if (atomic_dec_and_test(&ls->ls_requestqueue_cnt))
100164d88abSAlexander Aring wake_up(&ls->ls_requestqueue_wait);
101e7fd4179SDavid Teigland kfree(e);
102e7fd4179SDavid Teigland
103e7fd4179SDavid Teigland if (dlm_locking_stopped(ls)) {
104e7fd4179SDavid Teigland log_debug(ls, "process_requestqueue abort running");
10590135925SDavid Teigland mutex_unlock(&ls->ls_requestqueue_mutex);
106e7fd4179SDavid Teigland error = -EINTR;
107e7fd4179SDavid Teigland break;
108e7fd4179SDavid Teigland }
109e7fd4179SDavid Teigland schedule();
110e7fd4179SDavid Teigland }
111e7fd4179SDavid Teigland
112e7fd4179SDavid Teigland return error;
113e7fd4179SDavid Teigland }
114e7fd4179SDavid Teigland
115e7fd4179SDavid Teigland /*
116e7fd4179SDavid Teigland * After recovery is done, locking is resumed and dlm_recoverd takes all the
117c36258b5SDavid Teigland * saved requests and processes them as they would have been by dlm_recv. At
118c36258b5SDavid Teigland * the same time, dlm_recv will start receiving new requests from remote nodes.
119c36258b5SDavid Teigland * We want to delay dlm_recv processing new requests until dlm_recoverd has
120c36258b5SDavid Teigland * finished processing the old saved requests. We don't check for locking
121c36258b5SDavid Teigland * stopped here because dlm_ls_stop won't stop locking until it's suspended us
122c36258b5SDavid Teigland * (dlm_recv).
123e7fd4179SDavid Teigland */
124e7fd4179SDavid Teigland
dlm_wait_requestqueue(struct dlm_ls * ls)125e7fd4179SDavid Teigland void dlm_wait_requestqueue(struct dlm_ls *ls)
126e7fd4179SDavid Teigland {
127164d88abSAlexander Aring wait_event(ls->ls_requestqueue_wait,
128164d88abSAlexander Aring atomic_read(&ls->ls_requestqueue_cnt) == 0);
129e7fd4179SDavid Teigland }
130e7fd4179SDavid Teigland
purge_request(struct dlm_ls * ls,struct dlm_message * ms,int nodeid)131e7fd4179SDavid Teigland static int purge_request(struct dlm_ls *ls, struct dlm_message *ms, int nodeid)
132e7fd4179SDavid Teigland {
13300e99ccdSAlexander Aring __le32 type = ms->m_type;
134e7fd4179SDavid Teigland
1352896ee37SDavid Teigland /* the ls is being cleaned up and freed by release_lockspace */
1363cb5977cSAlexander Aring if (!atomic_read(&ls->ls_count))
1372896ee37SDavid Teigland return 1;
1382896ee37SDavid Teigland
139e7fd4179SDavid Teigland if (dlm_is_removed(ls, nodeid))
140e7fd4179SDavid Teigland return 1;
141e7fd4179SDavid Teigland
142e7fd4179SDavid Teigland /* directory operations are always purged because the directory is
143e7fd4179SDavid Teigland always rebuilt during recovery and the lookups resent */
144e7fd4179SDavid Teigland
14500e99ccdSAlexander Aring if (type == cpu_to_le32(DLM_MSG_REMOVE) ||
14600e99ccdSAlexander Aring type == cpu_to_le32(DLM_MSG_LOOKUP) ||
14700e99ccdSAlexander Aring type == cpu_to_le32(DLM_MSG_LOOKUP_REPLY))
148e7fd4179SDavid Teigland return 1;
149e7fd4179SDavid Teigland
150e7fd4179SDavid Teigland if (!dlm_no_directory(ls))
151e7fd4179SDavid Teigland return 0;
152e7fd4179SDavid Teigland
153e7fd4179SDavid Teigland return 1;
154e7fd4179SDavid Teigland }
155e7fd4179SDavid Teigland
dlm_purge_requestqueue(struct dlm_ls * ls)156e7fd4179SDavid Teigland void dlm_purge_requestqueue(struct dlm_ls *ls)
157e7fd4179SDavid Teigland {
158e7fd4179SDavid Teigland struct dlm_message *ms;
159e7fd4179SDavid Teigland struct rq_entry *e, *safe;
160e7fd4179SDavid Teigland
16190135925SDavid Teigland mutex_lock(&ls->ls_requestqueue_mutex);
162e7fd4179SDavid Teigland list_for_each_entry_safe(e, safe, &ls->ls_requestqueue, list) {
1638b0d8e03SAl Viro ms = &e->request;
164e7fd4179SDavid Teigland
165e7fd4179SDavid Teigland if (purge_request(ls, ms, e->nodeid)) {
166e7fd4179SDavid Teigland list_del(&e->list);
167164d88abSAlexander Aring if (atomic_dec_and_test(&ls->ls_requestqueue_cnt))
168164d88abSAlexander Aring wake_up(&ls->ls_requestqueue_wait);
169e7fd4179SDavid Teigland kfree(e);
170e7fd4179SDavid Teigland }
171e7fd4179SDavid Teigland }
17290135925SDavid Teigland mutex_unlock(&ls->ls_requestqueue_mutex);
173e7fd4179SDavid Teigland }
174e7fd4179SDavid Teigland
175