| 1 | |
| 2 | |
| 3 | *******************************************************************************
|
| 4 | **
|
| 5 | ** Copyright (C) 2005-2007 Red Hat, Inc. All rights reserved.
|
| 6 | **
|
| 7 | **
|
| 8 | *******************************************************************************
|
| 9 | *****************************************************************************
|
| 10 |
|
| 11 | #include "dlm_internal.h"
|
| 12 | #include "member.h"
|
| 13 | #include "lock.h"
|
| 14 | #include "dir.h"
|
| 15 | #include "config.h"
|
| 16 | #include "requestqueue.h"
|
| 17 |
|
| 18 | struct rq_entry {
|
| 19 | struct list_head list;
|
| 20 | uint32_t recover_seq;
|
| 21 | int nodeid;
|
| 22 | struct dlm_message request;
|
| 23 | };
|
| 24 |
|
| 25 | |
| 26 | * Requests received while the lockspace is in recovery get added to the
|
| 27 | * request queue and processed when recovery is complete. This happens when
|
| 28 | * the lockspace is suspended on some nodes before it is on others, or the
|
| 29 | * lockspace is enabled on some while still suspended on others.
|
| 30 |
|
| 31 |
|
| 32 | void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_message *ms)
|
| 33 | {
|
| 34 | struct rq_entry *e;
|
| 35 | int length = ms->m_header.h_length - sizeof(struct dlm_message);
|
| 36 |
|
| 37 | e = kmalloc(sizeof(struct rq_entry) + length, GFP_NOFS);
|
| 38 | if (!e) {
|
| 39 | log_print("dlm_add_requestqueue: out of memory len %d", length);
|
| 40 | return;
|
| 41 | }
|
| 42 |
|
| 43 | e->recover_seq = ls->ls_recover_seq & 0xFFFFFFFF;
|
| 44 | e->nodeid = nodeid;
|
| 45 | memcpy(&e->request, ms, ms->m_header.h_length);
|
| 46 |
|
| 47 | mutex_lock(&ls->ls_requestqueue_mutex);
|
| 48 | list_add_tail(&e->list, &ls->ls_requestqueue);
|
| 49 | mutex_unlock(&ls->ls_requestqueue_mutex);
|
| 50 | }
|
| 51 |
|
| 52 | |
| 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 |
|
| 63 | int dlm_process_requestqueue(struct dlm_ls *ls)
|
| 64 | {
|
| 65 | struct rq_entry *e;
|
| 66 | struct dlm_message *ms;
|
| 67 | int error = 0;
|
| 68 |
|
| 69 | mutex_lock(&ls->ls_requestqueue_mutex);
|
| 70 |
|
| 71 | for (;;) {
|
| 72 | if (list_empty(&ls->ls_requestqueue)) {
|
| 73 | mutex_unlock(&ls->ls_requestqueue_mutex);
|
| 74 | error = 0;
|
| 75 | break;
|
| 76 | }
|
| 77 | e = list_entry(ls->ls_requestqueue.next, struct rq_entry, list);
|
| 78 | mutex_unlock(&ls->ls_requestqueue_mutex);
|
| 79 |
|
| 80 | ms = &e->request;
|
| 81 |
|
| 82 | log_limit(ls, "dlm_process_requestqueue msg %d from %d "
|
| 83 | "lkid %x remid %x result %d seq %u",
|
| 84 | ms->m_type, ms->m_header.h_nodeid,
|
| 85 | ms->m_lkid, ms->m_remid, ms->m_result,
|
| 86 | e->recover_seq);
|
| 87 |
|
| 88 | dlm_receive_message_saved(ls, &e->request, e->recover_seq);
|
| 89 |
|
| 90 | mutex_lock(&ls->ls_requestqueue_mutex);
|
| 91 | list_del(&e->list);
|
| 92 | kfree(e);
|
| 93 |
|
| 94 | if (dlm_locking_stopped(ls)) {
|
| 95 | log_debug(ls, "process_requestqueue abort running");
|
| 96 | mutex_unlock(&ls->ls_requestqueue_mutex);
|
| 97 | error = -EINTR;
|
| 98 | break;
|
| 99 | }
|
| 100 | schedule();
|
| 101 | }
|
| 102 |
|
| 103 | return error;
|
| 104 | }
|
| 105 |
|
| 106 | |
| 107 | * After recovery is done, locking is resumed and dlm_recoverd takes all the
|
| 108 | * saved requests and processes them as they would have been by dlm_recv. At
|
| 109 | * the same time, dlm_recv will start receiving new requests from remote nodes.
|
| 110 | * We want to delay dlm_recv processing new requests until dlm_recoverd has
|
| 111 | * finished processing the old saved requests. We don't check for locking
|
| 112 | * stopped here because dlm_ls_stop won't stop locking until it's suspended us
|
| 113 | * (dlm_recv).
|
| 114 |
|
| 115 |
|
| 116 | void dlm_wait_requestqueue(struct dlm_ls *ls)
|
| 117 | {
|
| 118 | for (;;) {
|
| 119 | mutex_lock(&ls->ls_requestqueue_mutex);
|
| 120 | if (list_empty(&ls->ls_requestqueue))
|
| 121 | break;
|
| 122 | mutex_unlock(&ls->ls_requestqueue_mutex);
|
| 123 | schedule();
|
| 124 | }
|
| 125 | mutex_unlock(&ls->ls_requestqueue_mutex);
|
| 126 | }
|
| 127 |
|
| 128 | static int purge_request(struct dlm_ls *ls, struct dlm_message *ms, int nodeid)
|
| 129 | {
|
| 130 | uint32_t type = ms->m_type;
|
| 131 |
|
| 132 |
|
| 133 | if (!ls->ls_count)
|
| 134 | return 1;
|
| 135 |
|
| 136 | if (dlm_is_removed(ls, nodeid))
|
| 137 | return 1;
|
| 138 |
|
| 139 | |
| 140 | always rebuilt during recovery and the lookups resent
|
| 141 |
|
| 142 | if (type == DLM_MSG_REMOVE ||
|
| 143 | type == DLM_MSG_LOOKUP ||
|
| 144 | type == DLM_MSG_LOOKUP_REPLY)
|
| 145 | return 1;
|
| 146 |
|
| 147 | if (!dlm_no_directory(ls))
|
| 148 | return 0;
|
| 149 |
|
| 150 | return 1;
|
| 151 | }
|
| 152 |
|
| 153 | void dlm_purge_requestqueue(struct dlm_ls *ls)
|
| 154 | {
|
| 155 | struct dlm_message *ms;
|
| 156 | struct rq_entry *e, *safe;
|
| 157 |
|
| 158 | mutex_lock(&ls->ls_requestqueue_mutex);
|
| 159 | list_for_each_entry_safe(e, safe, &ls->ls_requestqueue, list) {
|
| 160 | ms = &e->request;
|
| 161 |
|
| 162 | if (purge_request(ls, ms, e->nodeid)) {
|
| 163 | list_del(&e->list);
|
| 164 | kfree(e);
|
| 165 | }
|
| 166 | }
|
| 167 | mutex_unlock(&ls->ls_requestqueue_mutex);
|
| 168 | }
|
| 169 |
|