blob: 0de04f17cceaeec20383d5613e9fd17152e5df10 [file] [log] [blame]
David Teiglande7fd4172006-01-18 09:30:29 +00001/******************************************************************************
2*******************************************************************************
3**
David Teiglandc36258b2007-09-27 15:53:38 -05004** Copyright (C) 2005-2007 Red Hat, Inc. All rights reserved.
David Teiglande7fd4172006-01-18 09:30:29 +00005**
6** This copyrighted material is made available to anyone wishing to use,
7** modify, copy, or redistribute it subject to the terms and conditions
8** of the GNU General Public License v.2.
9**
10*******************************************************************************
11******************************************************************************/
12
13#include "dlm_internal.h"
14#include "member.h"
15#include "lock.h"
16#include "dir.h"
17#include "config.h"
18#include "requestqueue.h"
19
20struct rq_entry {
21 struct list_head list;
22 int nodeid;
David Teiglandc36258b2007-09-27 15:53:38 -050023 char request[0];
David Teiglande7fd4172006-01-18 09:30:29 +000024};
25
26/*
27 * Requests received while the lockspace is in recovery get added to the
28 * request queue and processed when recovery is complete. This happens when
29 * the lockspace is suspended on some nodes before it is on others, or the
30 * lockspace is enabled on some while still suspended on others.
31 */
32
David Teiglandc36258b2007-09-27 15:53:38 -050033void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd)
David Teiglande7fd4172006-01-18 09:30:29 +000034{
35 struct rq_entry *e;
36 int length = hd->h_length;
37
David Teiglande7fd4172006-01-18 09:30:29 +000038 e = kmalloc(sizeof(struct rq_entry) + length, GFP_KERNEL);
39 if (!e) {
David Teiglandc36258b2007-09-27 15:53:38 -050040 log_print("dlm_add_requestqueue: out of memory len %d", length);
41 return;
David Teiglande7fd4172006-01-18 09:30:29 +000042 }
43
44 e->nodeid = nodeid;
45 memcpy(e->request, hd, length);
46
David Teigland90135922006-01-20 08:47:07 +000047 mutex_lock(&ls->ls_requestqueue_mutex);
David Teiglandc36258b2007-09-27 15:53:38 -050048 list_add_tail(&e->list, &ls->ls_requestqueue);
David Teigland90135922006-01-20 08:47:07 +000049 mutex_unlock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +000050}
51
David Teiglandc36258b2007-09-27 15:53:38 -050052/*
53 * Called by dlm_recoverd to process normal messages saved while recovery was
54 * happening. Normal locking has been enabled before this is called. dlm_recv
55 * upon receiving a message, will wait for all saved messages to be drained
56 * here before processing the message it got. If a new dlm_ls_stop() arrives
57 * while we're processing these saved messages, it may block trying to suspend
58 * dlm_recv if dlm_recv is waiting for us in dlm_wait_requestqueue. In that
59 * case, we don't abort since locking_stopped is still 0. If dlm_recv is not
60 * waiting for us, then this processing may be aborted due to locking_stopped.
61 */
62
David Teiglande7fd4172006-01-18 09:30:29 +000063int dlm_process_requestqueue(struct dlm_ls *ls)
64{
65 struct rq_entry *e;
David Teiglande7fd4172006-01-18 09:30:29 +000066 int error = 0;
67
David Teigland90135922006-01-20 08:47:07 +000068 mutex_lock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +000069
70 for (;;) {
71 if (list_empty(&ls->ls_requestqueue)) {
David Teigland90135922006-01-20 08:47:07 +000072 mutex_unlock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +000073 error = 0;
74 break;
75 }
76 e = list_entry(ls->ls_requestqueue.next, struct rq_entry, list);
David Teigland90135922006-01-20 08:47:07 +000077 mutex_unlock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +000078
David Teiglandc36258b2007-09-27 15:53:38 -050079 dlm_receive_message_saved(ls, (struct dlm_message *)e->request);
David Teiglande7fd4172006-01-18 09:30:29 +000080
David Teigland90135922006-01-20 08:47:07 +000081 mutex_lock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +000082 list_del(&e->list);
83 kfree(e);
84
85 if (dlm_locking_stopped(ls)) {
86 log_debug(ls, "process_requestqueue abort running");
David Teigland90135922006-01-20 08:47:07 +000087 mutex_unlock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +000088 error = -EINTR;
89 break;
90 }
91 schedule();
92 }
93
94 return error;
95}
96
97/*
98 * After recovery is done, locking is resumed and dlm_recoverd takes all the
David Teiglandc36258b2007-09-27 15:53:38 -050099 * saved requests and processes them as they would have been by dlm_recv. At
100 * the same time, dlm_recv will start receiving new requests from remote nodes.
101 * We want to delay dlm_recv processing new requests until dlm_recoverd has
102 * finished processing the old saved requests. We don't check for locking
103 * stopped here because dlm_ls_stop won't stop locking until it's suspended us
104 * (dlm_recv).
David Teiglande7fd4172006-01-18 09:30:29 +0000105 */
106
107void dlm_wait_requestqueue(struct dlm_ls *ls)
108{
109 for (;;) {
David Teigland90135922006-01-20 08:47:07 +0000110 mutex_lock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +0000111 if (list_empty(&ls->ls_requestqueue))
112 break;
David Teigland90135922006-01-20 08:47:07 +0000113 mutex_unlock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +0000114 schedule();
115 }
David Teigland90135922006-01-20 08:47:07 +0000116 mutex_unlock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +0000117}
118
119static int purge_request(struct dlm_ls *ls, struct dlm_message *ms, int nodeid)
120{
121 uint32_t type = ms->m_type;
122
David Teigland2896ee32006-11-27 11:31:22 -0600123 /* the ls is being cleaned up and freed by release_lockspace */
124 if (!ls->ls_count)
125 return 1;
126
David Teiglande7fd4172006-01-18 09:30:29 +0000127 if (dlm_is_removed(ls, nodeid))
128 return 1;
129
130 /* directory operations are always purged because the directory is
131 always rebuilt during recovery and the lookups resent */
132
133 if (type == DLM_MSG_REMOVE ||
134 type == DLM_MSG_LOOKUP ||
135 type == DLM_MSG_LOOKUP_REPLY)
136 return 1;
137
138 if (!dlm_no_directory(ls))
139 return 0;
140
141 /* with no directory, the master is likely to change as a part of
142 recovery; requests to/from the defunct master need to be purged */
143
144 switch (type) {
145 case DLM_MSG_REQUEST:
146 case DLM_MSG_CONVERT:
147 case DLM_MSG_UNLOCK:
148 case DLM_MSG_CANCEL:
149 /* we're no longer the master of this resource, the sender
150 will resend to the new master (see waiter_needs_recovery) */
151
152 if (dlm_hash2nodeid(ls, ms->m_hash) != dlm_our_nodeid())
153 return 1;
154 break;
155
156 case DLM_MSG_REQUEST_REPLY:
157 case DLM_MSG_CONVERT_REPLY:
158 case DLM_MSG_UNLOCK_REPLY:
159 case DLM_MSG_CANCEL_REPLY:
160 case DLM_MSG_GRANT:
161 /* this reply is from the former master of the resource,
162 we'll resend to the new master if needed */
163
164 if (dlm_hash2nodeid(ls, ms->m_hash) != nodeid)
165 return 1;
166 break;
167 }
168
169 return 0;
170}
171
172void dlm_purge_requestqueue(struct dlm_ls *ls)
173{
174 struct dlm_message *ms;
175 struct rq_entry *e, *safe;
176
David Teigland90135922006-01-20 08:47:07 +0000177 mutex_lock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +0000178 list_for_each_entry_safe(e, safe, &ls->ls_requestqueue, list) {
179 ms = (struct dlm_message *) e->request;
180
181 if (purge_request(ls, ms, e->nodeid)) {
182 list_del(&e->list);
183 kfree(e);
184 }
185 }
David Teigland90135922006-01-20 08:47:07 +0000186 mutex_unlock(&ls->ls_requestqueue_mutex);
David Teiglande7fd4172006-01-18 09:30:29 +0000187}
188