1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/kthread.h>
31 #include <cluster/heartbeat.h>
32 #include <cluster/nodemanager.h>
33 #include <cluster/tcp.h>
35 #include <dlm/dlmapi.h>
37 #define MLOG_MASK_PREFIX ML_VOTE
38 #include <cluster/masklog.h>
44 #include "extent_map.h"
45 #include "heartbeat.h"
51 #include "buffer_head_io.h"
53 #define OCFS2_MESSAGE_TYPE_VOTE (0x1)
54 #define OCFS2_MESSAGE_TYPE_RESPONSE (0x2)
57 __be32 h_response_id; /* used to lookup message handle on sending
62 __be32 h_node_num; /* node sending this particular message. */
67 struct ocfs2_msg_hdr v_hdr;
69 } __attribute__ ((packed));
71 /* Responses are given these values to maintain backwards
72 * compatibility with older ocfs2 versions */
73 #define OCFS2_RESPONSE_OK (0)
74 #define OCFS2_RESPONSE_BUSY (-16)
75 #define OCFS2_RESPONSE_BAD_MSG (-22)
77 struct ocfs2_response_msg
79 struct ocfs2_msg_hdr r_hdr;
81 } __attribute__ ((packed));
83 struct ocfs2_vote_work {
84 struct list_head w_list;
85 struct ocfs2_vote_msg w_msg;
88 enum ocfs2_vote_request {
89 OCFS2_VOTE_REQ_INVALID = 0,
91 OCFS2_VOTE_REQ_UMOUNT,
95 static inline int ocfs2_is_valid_vote_request(int request)
97 return OCFS2_VOTE_REQ_INVALID < request &&
98 request < OCFS2_VOTE_REQ_LAST;
101 typedef void (*ocfs2_net_response_callback)(void *priv,
102 struct ocfs2_response_msg *resp);
103 struct ocfs2_net_response_cb {
104 ocfs2_net_response_callback rc_cb;
108 struct ocfs2_net_wait_ctxt {
109 struct list_head n_list;
111 wait_queue_head_t n_event;
112 struct ocfs2_node_map n_node_map;
113 int n_response; /* an agreggate response. 0 if
114 * all nodes are go, < 0 on any
115 * negative response from any
116 * node or network error. */
117 struct ocfs2_net_response_cb *n_callback;
120 static void ocfs2_process_mount_request(struct ocfs2_super *osb,
121 unsigned int node_num)
123 mlog(0, "MOUNT vote from node %u\n", node_num);
124 /* The other node only sends us this message when he has an EX
125 * on the superblock, so our recovery threads (if having been
126 * launched) are waiting on it.*/
127 ocfs2_recovery_map_clear(osb, node_num);
128 ocfs2_node_map_set_bit(osb, &osb->mounted_map, node_num);
130 /* We clear the umount map here because a node may have been
131 * previously mounted, safely unmounted but never stopped
132 * heartbeating - in which case we'd have a stale entry. */
133 ocfs2_node_map_clear_bit(osb, &osb->umount_map, node_num);
136 static void ocfs2_process_umount_request(struct ocfs2_super *osb,
137 unsigned int node_num)
139 mlog(0, "UMOUNT vote from node %u\n", node_num);
140 ocfs2_node_map_clear_bit(osb, &osb->mounted_map, node_num);
141 ocfs2_node_map_set_bit(osb, &osb->umount_map, node_num);
144 static void ocfs2_process_vote(struct ocfs2_super *osb,
145 struct ocfs2_vote_msg *msg)
147 int net_status, vote_response;
148 unsigned int node_num;
150 enum ocfs2_vote_request request;
151 struct ocfs2_msg_hdr *hdr = &msg->v_hdr;
152 struct ocfs2_response_msg response;
154 /* decode the network mumbo jumbo into local variables. */
155 request = be32_to_cpu(hdr->h_request);
156 blkno = be64_to_cpu(hdr->h_blkno);
157 node_num = be32_to_cpu(hdr->h_node_num);
159 mlog(0, "processing vote: request = %u, blkno = %llu, node_num = %u\n",
160 request, (unsigned long long)blkno, node_num);
162 if (!ocfs2_is_valid_vote_request(request)) {
163 mlog(ML_ERROR, "Invalid vote request %d from node %u\n",
165 vote_response = OCFS2_RESPONSE_BAD_MSG;
169 vote_response = OCFS2_RESPONSE_OK;
172 case OCFS2_VOTE_REQ_UMOUNT:
173 ocfs2_process_umount_request(osb, node_num);
175 case OCFS2_VOTE_REQ_MOUNT:
176 ocfs2_process_mount_request(osb, node_num);
179 /* avoids a gcc warning */
184 /* Response struture is small so we just put it on the stack
185 * and stuff it inline. */
186 memset(&response, 0, sizeof(struct ocfs2_response_msg));
187 response.r_hdr.h_response_id = hdr->h_response_id;
188 response.r_hdr.h_blkno = hdr->h_blkno;
189 response.r_hdr.h_generation = hdr->h_generation;
190 response.r_hdr.h_node_num = cpu_to_be32(osb->node_num);
191 response.r_response = cpu_to_be32(vote_response);
193 net_status = o2net_send_message(OCFS2_MESSAGE_TYPE_RESPONSE,
196 sizeof(struct ocfs2_response_msg),
199 /* We still want to error print for ENOPROTOOPT here. The
200 * sending node shouldn't have unregistered his net handler
201 * without sending an unmount vote 1st */
203 && net_status != -ETIMEDOUT
204 && net_status != -ENOTCONN)
205 mlog(ML_ERROR, "message to node %u fails with error %d!\n",
206 node_num, net_status);
209 static void ocfs2_vote_thread_do_work(struct ocfs2_super *osb)
211 unsigned long processed;
212 struct ocfs2_lock_res *lockres;
213 struct ocfs2_vote_work *work;
217 spin_lock(&osb->vote_task_lock);
218 /* grab this early so we know to try again if a state change and
219 * wake happens part-way through our work */
220 osb->vote_work_sequence = osb->vote_wake_sequence;
222 processed = osb->blocked_lock_count;
224 BUG_ON(list_empty(&osb->blocked_lock_list));
226 lockres = list_entry(osb->blocked_lock_list.next,
227 struct ocfs2_lock_res, l_blocked_list);
228 list_del_init(&lockres->l_blocked_list);
229 osb->blocked_lock_count--;
230 spin_unlock(&osb->vote_task_lock);
235 ocfs2_process_blocked_lock(osb, lockres);
237 spin_lock(&osb->vote_task_lock);
240 while (osb->vote_count) {
241 BUG_ON(list_empty(&osb->vote_list));
242 work = list_entry(osb->vote_list.next,
243 struct ocfs2_vote_work, w_list);
244 list_del(&work->w_list);
246 spin_unlock(&osb->vote_task_lock);
248 ocfs2_process_vote(osb, &work->w_msg);
251 spin_lock(&osb->vote_task_lock);
253 spin_unlock(&osb->vote_task_lock);
258 static int ocfs2_vote_thread_lists_empty(struct ocfs2_super *osb)
262 spin_lock(&osb->vote_task_lock);
263 if (list_empty(&osb->blocked_lock_list) &&
264 list_empty(&osb->vote_list))
267 spin_unlock(&osb->vote_task_lock);
271 static int ocfs2_vote_thread_should_wake(struct ocfs2_super *osb)
275 spin_lock(&osb->vote_task_lock);
276 if (osb->vote_work_sequence != osb->vote_wake_sequence)
278 spin_unlock(&osb->vote_task_lock);
283 int ocfs2_vote_thread(void *arg)
286 struct ocfs2_super *osb = arg;
288 /* only quit once we've been asked to stop and there is no more
290 while (!(kthread_should_stop() &&
291 ocfs2_vote_thread_lists_empty(osb))) {
293 wait_event_interruptible(osb->vote_event,
294 ocfs2_vote_thread_should_wake(osb) ||
295 kthread_should_stop());
297 mlog(0, "vote_thread: awoken\n");
299 ocfs2_vote_thread_do_work(osb);
302 osb->vote_task = NULL;
306 static struct ocfs2_net_wait_ctxt *ocfs2_new_net_wait_ctxt(unsigned int response_id)
308 struct ocfs2_net_wait_ctxt *w;
310 w = kzalloc(sizeof(*w), GFP_NOFS);
316 INIT_LIST_HEAD(&w->n_list);
317 init_waitqueue_head(&w->n_event);
318 ocfs2_node_map_init(&w->n_node_map);
319 w->n_response_id = response_id;
320 w->n_callback = NULL;
325 static unsigned int ocfs2_new_response_id(struct ocfs2_super *osb)
329 spin_lock(&osb->net_response_lock);
330 ret = ++osb->net_response_ids;
331 spin_unlock(&osb->net_response_lock);
336 static void ocfs2_dequeue_net_wait_ctxt(struct ocfs2_super *osb,
337 struct ocfs2_net_wait_ctxt *w)
339 spin_lock(&osb->net_response_lock);
340 list_del(&w->n_list);
341 spin_unlock(&osb->net_response_lock);
344 static void ocfs2_queue_net_wait_ctxt(struct ocfs2_super *osb,
345 struct ocfs2_net_wait_ctxt *w)
347 spin_lock(&osb->net_response_lock);
348 list_add_tail(&w->n_list,
349 &osb->net_response_list);
350 spin_unlock(&osb->net_response_lock);
353 static void __ocfs2_mark_node_responded(struct ocfs2_super *osb,
354 struct ocfs2_net_wait_ctxt *w,
357 assert_spin_locked(&osb->net_response_lock);
359 ocfs2_node_map_clear_bit(osb, &w->n_node_map, node_num);
360 if (ocfs2_node_map_is_empty(osb, &w->n_node_map))
361 wake_up(&w->n_event);
364 /* Intended to be called from the node down callback, we fake remove
365 * the node from all our response contexts */
366 void ocfs2_remove_node_from_vote_queues(struct ocfs2_super *osb,
370 struct ocfs2_net_wait_ctxt *w = NULL;
372 spin_lock(&osb->net_response_lock);
374 list_for_each(p, &osb->net_response_list) {
375 w = list_entry(p, struct ocfs2_net_wait_ctxt, n_list);
377 __ocfs2_mark_node_responded(osb, w, node_num);
380 spin_unlock(&osb->net_response_lock);
383 static int ocfs2_broadcast_vote(struct ocfs2_super *osb,
384 struct ocfs2_vote_msg *request,
385 unsigned int response_id,
387 struct ocfs2_net_response_cb *callback)
389 int status, i, remote_err;
390 struct ocfs2_net_wait_ctxt *w = NULL;
395 w = ocfs2_new_net_wait_ctxt(response_id);
401 w->n_callback = callback;
403 /* we're pretty much ready to go at this point, and this fills
404 * in n_response which we need anyway... */
405 ocfs2_queue_net_wait_ctxt(osb, w);
407 i = ocfs2_node_map_iterate(osb, &osb->mounted_map, 0);
409 while (i != O2NM_INVALID_NODE_NUM) {
410 if (i != osb->node_num) {
411 mlog(0, "trying to send request to node %i\n", i);
412 ocfs2_node_map_set_bit(osb, &w->n_node_map, i);
415 status = o2net_send_message(OCFS2_MESSAGE_TYPE_VOTE,
421 if (status == -ETIMEDOUT) {
422 mlog(0, "remote node %d timed out!\n", i);
426 if (remote_err < 0) {
428 mlog(0, "remote error %d on node %d!\n",
439 i = ocfs2_node_map_iterate(osb, &osb->mounted_map, i);
440 mlog(0, "next is %d, i am %d\n", i, osb->node_num);
442 mlog(0, "done sending, now waiting on responses...\n");
444 wait_event(w->n_event, ocfs2_node_map_is_empty(osb, &w->n_node_map));
446 ocfs2_dequeue_net_wait_ctxt(osb, w);
449 *response = w->n_response;
454 ocfs2_dequeue_net_wait_ctxt(osb, w);
462 static struct ocfs2_vote_msg * ocfs2_new_vote_request(struct ocfs2_super *osb,
464 unsigned int generation,
465 enum ocfs2_vote_request type)
467 struct ocfs2_vote_msg *request;
468 struct ocfs2_msg_hdr *hdr;
470 BUG_ON(!ocfs2_is_valid_vote_request(type));
472 request = kzalloc(sizeof(*request), GFP_NOFS);
476 hdr = &request->v_hdr;
477 hdr->h_node_num = cpu_to_be32(osb->node_num);
478 hdr->h_request = cpu_to_be32(type);
479 hdr->h_blkno = cpu_to_be64(blkno);
480 hdr->h_generation = cpu_to_be32(generation);
486 /* Complete the buildup of a new vote request and process the
487 * broadcast return value. */
488 static int ocfs2_do_request_vote(struct ocfs2_super *osb,
489 struct ocfs2_vote_msg *request,
490 struct ocfs2_net_response_cb *callback)
492 int status, response = -EBUSY;
493 unsigned int response_id;
494 struct ocfs2_msg_hdr *hdr;
496 response_id = ocfs2_new_response_id(osb);
498 hdr = &request->v_hdr;
499 hdr->h_response_id = cpu_to_be32(response_id);
501 status = ocfs2_broadcast_vote(osb, request, response_id, &response,
514 int ocfs2_request_mount_vote(struct ocfs2_super *osb)
517 struct ocfs2_vote_msg *request = NULL;
519 request = ocfs2_new_vote_request(osb, 0ULL, 0, OCFS2_VOTE_REQ_MOUNT);
526 while (status == -EAGAIN) {
527 if (!(osb->s_mount_opt & OCFS2_MOUNT_NOINTR) &&
528 signal_pending(current)) {
529 status = -ERESTARTSYS;
533 if (ocfs2_node_map_is_only(osb, &osb->mounted_map,
539 status = ocfs2_do_request_vote(osb, request, NULL);
547 int ocfs2_request_umount_vote(struct ocfs2_super *osb)
550 struct ocfs2_vote_msg *request = NULL;
552 request = ocfs2_new_vote_request(osb, 0ULL, 0, OCFS2_VOTE_REQ_UMOUNT);
559 while (status == -EAGAIN) {
560 /* Do not check signals on this vote... We really want
561 * this one to go all the way through. */
563 if (ocfs2_node_map_is_only(osb, &osb->mounted_map,
569 status = ocfs2_do_request_vote(osb, request, NULL);
577 /* TODO: This should eventually be a hash table! */
578 static struct ocfs2_net_wait_ctxt * __ocfs2_find_net_wait_ctxt(struct ocfs2_super *osb,
582 struct ocfs2_net_wait_ctxt *w = NULL;
584 list_for_each(p, &osb->net_response_list) {
585 w = list_entry(p, struct ocfs2_net_wait_ctxt, n_list);
586 if (response_id == w->n_response_id)
594 /* Translate response codes into local node errno values */
595 static inline int ocfs2_translate_response(int response)
600 case OCFS2_RESPONSE_OK:
604 case OCFS2_RESPONSE_BUSY:
615 static int ocfs2_handle_response_message(struct o2net_msg *msg,
617 void *data, void **ret_data)
619 unsigned int response_id, node_num;
621 struct ocfs2_super *osb = data;
622 struct ocfs2_response_msg *resp;
623 struct ocfs2_net_wait_ctxt * w;
624 struct ocfs2_net_response_cb *resp_cb;
626 resp = (struct ocfs2_response_msg *) msg->buf;
628 response_id = be32_to_cpu(resp->r_hdr.h_response_id);
629 node_num = be32_to_cpu(resp->r_hdr.h_node_num);
631 ocfs2_translate_response(be32_to_cpu(resp->r_response));
633 mlog(0, "received response message:\n");
634 mlog(0, "h_response_id = %u\n", response_id);
635 mlog(0, "h_request = %u\n", be32_to_cpu(resp->r_hdr.h_request));
636 mlog(0, "h_blkno = %llu\n",
637 (unsigned long long)be64_to_cpu(resp->r_hdr.h_blkno));
638 mlog(0, "h_generation = %u\n", be32_to_cpu(resp->r_hdr.h_generation));
639 mlog(0, "h_node_num = %u\n", node_num);
640 mlog(0, "r_response = %d\n", response_status);
642 spin_lock(&osb->net_response_lock);
643 w = __ocfs2_find_net_wait_ctxt(osb, response_id);
645 mlog(0, "request not found!\n");
648 resp_cb = w->n_callback;
650 if (response_status && (!w->n_response)) {
651 /* we only really need one negative response so don't
653 w->n_response = response_status;
657 spin_unlock(&osb->net_response_lock);
659 resp_cb->rc_cb(resp_cb->rc_priv, resp);
661 spin_lock(&osb->net_response_lock);
664 __ocfs2_mark_node_responded(osb, w, node_num);
666 spin_unlock(&osb->net_response_lock);
671 static int ocfs2_handle_vote_message(struct o2net_msg *msg,
673 void *data, void **ret_data)
676 struct ocfs2_super *osb = data;
677 struct ocfs2_vote_work *work;
679 work = kmalloc(sizeof(struct ocfs2_vote_work), GFP_NOFS);
686 INIT_LIST_HEAD(&work->w_list);
687 memcpy(&work->w_msg, msg->buf, sizeof(struct ocfs2_vote_msg));
689 mlog(0, "scheduling vote request:\n");
690 mlog(0, "h_response_id = %u\n",
691 be32_to_cpu(work->w_msg.v_hdr.h_response_id));
692 mlog(0, "h_request = %u\n", be32_to_cpu(work->w_msg.v_hdr.h_request));
693 mlog(0, "h_blkno = %llu\n",
694 (unsigned long long)be64_to_cpu(work->w_msg.v_hdr.h_blkno));
695 mlog(0, "h_generation = %u\n",
696 be32_to_cpu(work->w_msg.v_hdr.h_generation));
697 mlog(0, "h_node_num = %u\n",
698 be32_to_cpu(work->w_msg.v_hdr.h_node_num));
700 spin_lock(&osb->vote_task_lock);
701 list_add_tail(&work->w_list, &osb->vote_list);
703 spin_unlock(&osb->vote_task_lock);
705 ocfs2_kick_vote_thread(osb);
712 void ocfs2_unregister_net_handlers(struct ocfs2_super *osb)
717 o2net_unregister_handler_list(&osb->osb_net_handlers);
719 if (!list_empty(&osb->net_response_list))
720 mlog(ML_ERROR, "net response list not empty!\n");
725 int ocfs2_register_net_handlers(struct ocfs2_super *osb)
729 if (ocfs2_mount_local(osb))
732 status = o2net_register_handler(OCFS2_MESSAGE_TYPE_RESPONSE,
734 sizeof(struct ocfs2_response_msg),
735 ocfs2_handle_response_message,
736 osb, NULL, &osb->osb_net_handlers);
742 status = o2net_register_handler(OCFS2_MESSAGE_TYPE_VOTE,
744 sizeof(struct ocfs2_vote_msg),
745 ocfs2_handle_vote_message,
746 osb, NULL, &osb->osb_net_handlers);
753 ocfs2_unregister_net_handlers(osb);