1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2004, 2005 Oracle. All rights reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/jiffies.h>
25 #include <linux/module.h>
27 #include <linux/bio.h>
28 #include <linux/blkdev.h>
29 #include <linux/delay.h>
30 #include <linux/file.h>
31 #include <linux/kthread.h>
32 #include <linux/configfs.h>
33 #include <linux/random.h>
34 #include <linux/crc32.h>
35 #include <linux/time.h>
36 #include <linux/debugfs.h>
37 #include <linux/slab.h>
38 #include <linux/bitmap.h>
39 #include <linux/ktime.h>
40 #include "heartbeat.h"
42 #include "nodemanager.h"
49 * The first heartbeat pass had one global thread that would serialize all hb
50 * callback calls. This global serializing sem should only be removed once
51 * we've made sure that all callees can deal with being called concurrently
52 * from multiple hb region threads.
54 static DECLARE_RWSEM(o2hb_callback_sem);
57 * multiple hb threads are watching multiple regions. A node is live
58 * whenever any of the threads sees activity from the node in its region.
60 static DEFINE_SPINLOCK(o2hb_live_lock);
61 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
62 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
63 static LIST_HEAD(o2hb_node_events);
64 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
67 * In global heartbeat, we maintain a series of region bitmaps.
68 * - o2hb_region_bitmap allows us to limit the region number to max region.
69 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
70 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
72 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
74 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
75 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
76 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
77 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
79 #define O2HB_DB_TYPE_LIVENODES 0
80 #define O2HB_DB_TYPE_LIVEREGIONS 1
81 #define O2HB_DB_TYPE_QUORUMREGIONS 2
82 #define O2HB_DB_TYPE_FAILEDREGIONS 3
83 #define O2HB_DB_TYPE_REGION_LIVENODES 4
84 #define O2HB_DB_TYPE_REGION_NUMBER 5
85 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
86 #define O2HB_DB_TYPE_REGION_PINNED 7
87 struct o2hb_debug_buf {
94 static struct o2hb_debug_buf *o2hb_db_livenodes;
95 static struct o2hb_debug_buf *o2hb_db_liveregions;
96 static struct o2hb_debug_buf *o2hb_db_quorumregions;
97 static struct o2hb_debug_buf *o2hb_db_failedregions;
99 #define O2HB_DEBUG_DIR "o2hb"
100 #define O2HB_DEBUG_LIVENODES "livenodes"
101 #define O2HB_DEBUG_LIVEREGIONS "live_regions"
102 #define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
103 #define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
104 #define O2HB_DEBUG_REGION_NUMBER "num"
105 #define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
106 #define O2HB_DEBUG_REGION_PINNED "pinned"
108 static struct dentry *o2hb_debug_dir;
109 static struct dentry *o2hb_debug_livenodes;
110 static struct dentry *o2hb_debug_liveregions;
111 static struct dentry *o2hb_debug_quorumregions;
112 static struct dentry *o2hb_debug_failedregions;
114 static LIST_HEAD(o2hb_all_regions);
116 static struct o2hb_callback {
117 struct list_head list;
118 } o2hb_callbacks[O2HB_NUM_CB];
120 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
122 #define O2HB_DEFAULT_BLOCK_BITS 9
124 enum o2hb_heartbeat_modes {
125 O2HB_HEARTBEAT_LOCAL = 0,
126 O2HB_HEARTBEAT_GLOBAL,
127 O2HB_HEARTBEAT_NUM_MODES,
130 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
131 "local", /* O2HB_HEARTBEAT_LOCAL */
132 "global", /* O2HB_HEARTBEAT_GLOBAL */
135 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
136 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
139 * o2hb_dependent_users tracks the number of registered callbacks that depend
140 * on heartbeat. o2net and o2dlm are two entities that register this callback.
141 * However only o2dlm depends on the heartbeat. It does not want the heartbeat
142 * to stop while a dlm domain is still active.
144 unsigned int o2hb_dependent_users;
147 * In global heartbeat mode, all regions are pinned if there are one or more
148 * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
149 * regions are unpinned if the region count exceeds the cut off or the number
150 * of dependent users falls to zero.
152 #define O2HB_PIN_CUT_OFF 3
155 * In local heartbeat mode, we assume the dlm domain name to be the same as
156 * region uuid. This is true for domains created for the file system but not
157 * necessarily true for userdlm domains. This is a known limitation.
159 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
160 * works for both file system and userdlm domains.
162 static int o2hb_region_pin(const char *region_uuid);
163 static void o2hb_region_unpin(const char *region_uuid);
165 /* Only sets a new threshold if there are no active regions.
167 * No locking or otherwise interesting code is required for reading
168 * o2hb_dead_threshold as it can't change once regions are active and
169 * it's not interesting to anyone until then anyway. */
170 static void o2hb_dead_threshold_set(unsigned int threshold)
172 if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
173 spin_lock(&o2hb_live_lock);
174 if (list_empty(&o2hb_all_regions))
175 o2hb_dead_threshold = threshold;
176 spin_unlock(&o2hb_live_lock);
180 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
184 if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
185 spin_lock(&o2hb_live_lock);
186 if (list_empty(&o2hb_all_regions)) {
187 o2hb_heartbeat_mode = hb_mode;
190 spin_unlock(&o2hb_live_lock);
196 struct o2hb_node_event {
197 struct list_head hn_item;
198 enum o2hb_callback_type hn_event_type;
199 struct o2nm_node *hn_node;
203 struct o2hb_disk_slot {
204 struct o2hb_disk_heartbeat_block *ds_raw_block;
207 u64 ds_last_generation;
208 u16 ds_equal_samples;
209 u16 ds_changed_samples;
210 struct list_head ds_live_item;
213 /* each thread owns a region.. when we're asked to tear down the region
214 * we ask the thread to stop, who cleans up the region */
216 struct config_item hr_item;
218 struct list_head hr_all_item;
219 unsigned hr_unclean_stop:1,
225 /* protected by the hr_callback_sem */
226 struct task_struct *hr_task;
228 unsigned int hr_blocks;
229 unsigned long long hr_start_block;
231 unsigned int hr_block_bits;
232 unsigned int hr_block_bytes;
234 unsigned int hr_slots_per_page;
235 unsigned int hr_num_pages;
237 struct page **hr_slot_data;
238 struct block_device *hr_bdev;
239 struct o2hb_disk_slot *hr_slots;
241 /* live node map of this region */
242 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
243 unsigned int hr_region_num;
245 struct dentry *hr_debug_dir;
246 struct dentry *hr_debug_livenodes;
247 struct dentry *hr_debug_regnum;
248 struct dentry *hr_debug_elapsed_time;
249 struct dentry *hr_debug_pinned;
250 struct o2hb_debug_buf *hr_db_livenodes;
251 struct o2hb_debug_buf *hr_db_regnum;
252 struct o2hb_debug_buf *hr_db_elapsed_time;
253 struct o2hb_debug_buf *hr_db_pinned;
255 /* let the person setting up hb wait for it to return until it
256 * has reached a 'steady' state. This will be fixed when we have
257 * a more complete api that doesn't lead to this sort of fragility. */
258 atomic_t hr_steady_iterations;
260 /* terminate o2hb thread if it does not reach steady state
261 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
262 atomic_t hr_unsteady_iterations;
264 char hr_dev_name[BDEVNAME_SIZE];
266 unsigned int hr_timeout_ms;
268 /* randomized as the region goes up and down so that a node
269 * recognizes a node going up and down in one iteration */
272 struct delayed_work hr_write_timeout_work;
273 unsigned long hr_last_timeout_start;
275 /* negotiate timer, used to negotiate extending hb timeout. */
276 struct delayed_work hr_nego_timeout_work;
277 unsigned long hr_nego_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
279 /* Used during o2hb_check_slot to hold a copy of the block
280 * being checked because we temporarily have to zero out the
282 struct o2hb_disk_heartbeat_block *hr_tmp_block;
284 /* Message key for negotiate timeout message. */
286 struct list_head hr_handler_list;
288 /* last hb status, 0 for success, other value for error. */
289 int hr_last_hb_status;
292 struct o2hb_bio_wait_ctxt {
293 atomic_t wc_num_reqs;
294 struct completion wc_io_complete;
298 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
301 O2HB_NEGO_TIMEOUT_MSG = 1,
302 O2HB_NEGO_APPROVE_MSG = 2,
305 struct o2hb_nego_msg {
309 static void o2hb_write_timeout(struct work_struct *work)
312 struct o2hb_region *reg =
313 container_of(work, struct o2hb_region,
314 hr_write_timeout_work.work);
316 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
317 "milliseconds\n", reg->hr_dev_name,
318 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
320 if (o2hb_global_heartbeat_active()) {
321 spin_lock(&o2hb_live_lock);
322 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
323 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
324 failed = bitmap_weight(o2hb_failed_region_bitmap,
326 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
328 spin_unlock(&o2hb_live_lock);
330 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
334 * Fence if the number of failed regions >= half the number
337 if ((failed << 1) < quorum)
341 o2quo_disk_timeout();
344 static void o2hb_arm_timeout(struct o2hb_region *reg)
346 /* Arm writeout only after thread reaches steady state */
347 if (atomic_read(®->hr_steady_iterations) != 0)
350 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
351 O2HB_MAX_WRITE_TIMEOUT_MS);
353 if (o2hb_global_heartbeat_active()) {
354 spin_lock(&o2hb_live_lock);
355 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
356 spin_unlock(&o2hb_live_lock);
358 cancel_delayed_work(®->hr_write_timeout_work);
359 schedule_delayed_work(®->hr_write_timeout_work,
360 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
362 cancel_delayed_work(®->hr_nego_timeout_work);
363 /* negotiate timeout must be less than write timeout. */
364 schedule_delayed_work(®->hr_nego_timeout_work,
365 msecs_to_jiffies(O2HB_NEGO_TIMEOUT_MS));
366 memset(reg->hr_nego_node_bitmap, 0, sizeof(reg->hr_nego_node_bitmap));
369 static void o2hb_disarm_timeout(struct o2hb_region *reg)
371 cancel_delayed_work_sync(®->hr_write_timeout_work);
372 cancel_delayed_work_sync(®->hr_nego_timeout_work);
375 static int o2hb_send_nego_msg(int key, int type, u8 target)
377 struct o2hb_nego_msg msg;
380 msg.node_num = o2nm_this_node();
382 ret = o2net_send_message(type, key, &msg, sizeof(msg),
385 if (ret == -EAGAIN || ret == -ENOMEM) {
393 static void o2hb_nego_timeout(struct work_struct *work)
395 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
396 int master_node, i, ret;
397 struct o2hb_region *reg;
399 reg = container_of(work, struct o2hb_region, hr_nego_timeout_work.work);
400 /* don't negotiate timeout if last hb failed since it is very
401 * possible io failed. Should let write timeout fence self.
403 if (reg->hr_last_hb_status)
406 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
407 /* lowest node as master node to make negotiate decision. */
408 master_node = find_next_bit(live_node_bitmap, O2NM_MAX_NODES, 0);
410 if (master_node == o2nm_this_node()) {
411 if (!test_bit(master_node, reg->hr_nego_node_bitmap)) {
412 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s).\n",
413 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000,
414 config_item_name(®->hr_item), reg->hr_dev_name);
415 set_bit(master_node, reg->hr_nego_node_bitmap);
417 if (memcmp(reg->hr_nego_node_bitmap, live_node_bitmap,
418 sizeof(reg->hr_nego_node_bitmap))) {
419 /* check negotiate bitmap every second to do timeout
422 schedule_delayed_work(®->hr_nego_timeout_work,
423 msecs_to_jiffies(1000));
428 printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%s) is down.\n",
429 config_item_name(®->hr_item), reg->hr_dev_name);
430 /* approve negotiate timeout request. */
431 o2hb_arm_timeout(reg);
434 while ((i = find_next_bit(live_node_bitmap,
435 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
436 if (i == master_node)
439 mlog(ML_HEARTBEAT, "send NEGO_APPROVE msg to node %d\n", i);
440 ret = o2hb_send_nego_msg(reg->hr_key,
441 O2HB_NEGO_APPROVE_MSG, i);
443 mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
447 /* negotiate timeout with master node. */
448 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s), negotiate timeout with node %d.\n",
449 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000, config_item_name(®->hr_item),
450 reg->hr_dev_name, master_node);
451 ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
454 mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
459 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
462 struct o2hb_region *reg = data;
463 struct o2hb_nego_msg *nego_msg;
465 nego_msg = (struct o2hb_nego_msg *)msg->buf;
466 printk(KERN_NOTICE "o2hb: receive negotiate timeout message from node %d on region %s (%s).\n",
467 nego_msg->node_num, config_item_name(®->hr_item), reg->hr_dev_name);
468 if (nego_msg->node_num < O2NM_MAX_NODES)
469 set_bit(nego_msg->node_num, reg->hr_nego_node_bitmap);
471 mlog(ML_ERROR, "got nego timeout message from bad node.\n");
476 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
479 struct o2hb_region *reg = data;
481 printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%s).\n",
482 config_item_name(®->hr_item), reg->hr_dev_name);
483 o2hb_arm_timeout(reg);
487 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
489 atomic_set(&wc->wc_num_reqs, 1);
490 init_completion(&wc->wc_io_complete);
494 /* Used in error paths too */
495 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
498 /* sadly atomic_sub_and_test() isn't available on all platforms. The
499 * good news is that the fast path only completes one at a time */
501 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
503 complete(&wc->wc_io_complete);
508 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
510 o2hb_bio_wait_dec(wc, 1);
511 wait_for_completion(&wc->wc_io_complete);
514 static void o2hb_bio_end_io(struct bio *bio)
516 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
518 if (bio->bi_status) {
519 mlog(ML_ERROR, "IO Error %d\n", bio->bi_status);
520 wc->wc_error = blk_status_to_errno(bio->bi_status);
523 o2hb_bio_wait_dec(wc, 1);
527 /* Setup a Bio to cover I/O against num_slots slots starting at
529 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
530 struct o2hb_bio_wait_ctxt *wc,
531 unsigned int *current_slot,
532 unsigned int max_slots, int op,
535 int len, current_page;
536 unsigned int vec_len, vec_start;
537 unsigned int bits = reg->hr_block_bits;
538 unsigned int spp = reg->hr_slots_per_page;
539 unsigned int cs = *current_slot;
543 /* Testing has shown this allocation to take long enough under
544 * GFP_KERNEL that the local node can get fenced. It would be
545 * nicest if we could pre-allocate these bios and avoid this
547 bio = bio_alloc(GFP_ATOMIC, 16);
549 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
550 bio = ERR_PTR(-ENOMEM);
554 /* Must put everything in 512 byte sectors for the bio... */
555 bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
556 bio_set_dev(bio, reg->hr_bdev);
557 bio->bi_private = wc;
558 bio->bi_end_io = o2hb_bio_end_io;
559 bio_set_op_attrs(bio, op, op_flags);
561 vec_start = (cs << bits) % PAGE_SIZE;
562 while(cs < max_slots) {
563 current_page = cs / spp;
564 page = reg->hr_slot_data[current_page];
566 vec_len = min(PAGE_SIZE - vec_start,
567 (max_slots-cs) * (PAGE_SIZE/spp) );
569 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
570 current_page, vec_len, vec_start);
572 len = bio_add_page(bio, page, vec_len, vec_start);
573 if (len != vec_len) break;
575 cs += vec_len / (PAGE_SIZE/spp);
584 static int o2hb_read_slots(struct o2hb_region *reg,
585 unsigned int max_slots)
587 unsigned int current_slot=0;
589 struct o2hb_bio_wait_ctxt wc;
592 o2hb_bio_wait_init(&wc);
594 while(current_slot < max_slots) {
595 bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots,
598 status = PTR_ERR(bio);
603 atomic_inc(&wc.wc_num_reqs);
610 o2hb_wait_on_io(&wc);
611 if (wc.wc_error && !status)
612 status = wc.wc_error;
617 static int o2hb_issue_node_write(struct o2hb_region *reg,
618 struct o2hb_bio_wait_ctxt *write_wc)
624 o2hb_bio_wait_init(write_wc);
626 slot = o2nm_this_node();
628 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1, REQ_OP_WRITE,
631 status = PTR_ERR(bio);
636 atomic_inc(&write_wc->wc_num_reqs);
644 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
645 struct o2hb_disk_heartbeat_block *hb_block)
650 /* We want to compute the block crc with a 0 value in the
651 * hb_cksum field. Save it off here and replace after the
653 old_cksum = hb_block->hb_cksum;
654 hb_block->hb_cksum = 0;
656 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
658 hb_block->hb_cksum = old_cksum;
663 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
665 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
666 "cksum = 0x%x, generation 0x%llx\n",
667 (long long)le64_to_cpu(hb_block->hb_seq),
668 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
669 (long long)le64_to_cpu(hb_block->hb_generation));
672 static int o2hb_verify_crc(struct o2hb_region *reg,
673 struct o2hb_disk_heartbeat_block *hb_block)
677 read = le32_to_cpu(hb_block->hb_cksum);
678 computed = o2hb_compute_block_crc_le(reg, hb_block);
680 return read == computed;
684 * Compare the slot data with what we wrote in the last iteration.
685 * If the match fails, print an appropriate error message. This is to
686 * detect errors like... another node hearting on the same slot,
687 * flaky device that is losing writes, etc.
688 * Returns 1 if check succeeds, 0 otherwise.
690 static int o2hb_check_own_slot(struct o2hb_region *reg)
692 struct o2hb_disk_slot *slot;
693 struct o2hb_disk_heartbeat_block *hb_block;
696 slot = ®->hr_slots[o2nm_this_node()];
697 /* Don't check on our 1st timestamp */
698 if (!slot->ds_last_time)
701 hb_block = slot->ds_raw_block;
702 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
703 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
704 hb_block->hb_node == slot->ds_node_num)
707 #define ERRSTR1 "Another node is heartbeating on device"
708 #define ERRSTR2 "Heartbeat generation mismatch on device"
709 #define ERRSTR3 "Heartbeat sequence mismatch on device"
711 if (hb_block->hb_node != slot->ds_node_num)
713 else if (le64_to_cpu(hb_block->hb_generation) !=
714 slot->ds_last_generation)
719 mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
720 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
721 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
722 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
723 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
724 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
729 static inline void o2hb_prepare_block(struct o2hb_region *reg,
734 struct o2hb_disk_slot *slot;
735 struct o2hb_disk_heartbeat_block *hb_block;
737 node_num = o2nm_this_node();
738 slot = ®->hr_slots[node_num];
740 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
741 memset(hb_block, 0, reg->hr_block_bytes);
742 /* TODO: time stuff */
743 cputime = ktime_get_real_seconds();
747 hb_block->hb_seq = cpu_to_le64(cputime);
748 hb_block->hb_node = node_num;
749 hb_block->hb_generation = cpu_to_le64(generation);
750 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
752 /* This step must always happen last! */
753 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
756 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
757 (long long)generation,
758 le32_to_cpu(hb_block->hb_cksum));
761 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
762 struct o2nm_node *node,
765 struct o2hb_callback_func *f;
767 list_for_each_entry(f, &hbcall->list, hc_item) {
768 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
769 (f->hc_func)(node, idx, f->hc_data);
773 /* Will run the list in order until we process the passed event */
774 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
776 struct o2hb_callback *hbcall;
777 struct o2hb_node_event *event;
779 /* Holding callback sem assures we don't alter the callback
780 * lists when doing this, and serializes ourselves with other
781 * processes wanting callbacks. */
782 down_write(&o2hb_callback_sem);
784 spin_lock(&o2hb_live_lock);
785 while (!list_empty(&o2hb_node_events)
786 && !list_empty(&queued_event->hn_item)) {
787 event = list_entry(o2hb_node_events.next,
788 struct o2hb_node_event,
790 list_del_init(&event->hn_item);
791 spin_unlock(&o2hb_live_lock);
793 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
794 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
797 hbcall = hbcall_from_type(event->hn_event_type);
799 /* We should *never* have gotten on to the list with a
800 * bad type... This isn't something that we should try
801 * to recover from. */
802 BUG_ON(IS_ERR(hbcall));
804 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
806 spin_lock(&o2hb_live_lock);
808 spin_unlock(&o2hb_live_lock);
810 up_write(&o2hb_callback_sem);
813 static void o2hb_queue_node_event(struct o2hb_node_event *event,
814 enum o2hb_callback_type type,
815 struct o2nm_node *node,
818 assert_spin_locked(&o2hb_live_lock);
820 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
822 event->hn_event_type = type;
823 event->hn_node = node;
824 event->hn_node_num = node_num;
826 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
827 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
829 list_add_tail(&event->hn_item, &o2hb_node_events);
832 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
834 struct o2hb_node_event event =
835 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
836 struct o2nm_node *node;
839 node = o2nm_get_node_by_num(slot->ds_node_num);
843 spin_lock(&o2hb_live_lock);
844 if (!list_empty(&slot->ds_live_item)) {
845 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
848 list_del_init(&slot->ds_live_item);
850 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
851 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
853 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
858 spin_unlock(&o2hb_live_lock);
861 o2hb_run_event_list(&event);
866 static void o2hb_set_quorum_device(struct o2hb_region *reg)
868 if (!o2hb_global_heartbeat_active())
871 /* Prevent race with o2hb_heartbeat_group_drop_item() */
872 if (kthread_should_stop())
875 /* Tag region as quorum only after thread reaches steady state */
876 if (atomic_read(®->hr_steady_iterations) != 0)
879 spin_lock(&o2hb_live_lock);
881 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
885 * A region can be added to the quorum only when it sees all
886 * live nodes heartbeat on it. In other words, the region has been
887 * added to all nodes.
889 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
890 sizeof(o2hb_live_node_bitmap)))
893 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
894 config_item_name(®->hr_item), reg->hr_dev_name);
896 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
899 * If global heartbeat active, unpin all regions if the
900 * region count > CUT_OFF
902 if (bitmap_weight(o2hb_quorum_region_bitmap,
903 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
904 o2hb_region_unpin(NULL);
906 spin_unlock(&o2hb_live_lock);
909 static int o2hb_check_slot(struct o2hb_region *reg,
910 struct o2hb_disk_slot *slot)
912 int changed = 0, gen_changed = 0;
913 struct o2hb_node_event event =
914 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
915 struct o2nm_node *node;
916 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
918 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
919 unsigned int slot_dead_ms;
923 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
926 * If a node is no longer configured but is still in the livemap, we
927 * may need to clear that bit from the livemap.
929 node = o2nm_get_node_by_num(slot->ds_node_num);
931 spin_lock(&o2hb_live_lock);
932 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
933 spin_unlock(&o2hb_live_lock);
938 if (!o2hb_verify_crc(reg, hb_block)) {
939 /* all paths from here will drop o2hb_live_lock for
941 spin_lock(&o2hb_live_lock);
943 /* Don't print an error on the console in this case -
944 * a freshly formatted heartbeat area will not have a
946 if (list_empty(&slot->ds_live_item))
949 /* The node is live but pushed out a bad crc. We
950 * consider it a transient miss but don't populate any
951 * other values as they may be junk. */
952 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
953 slot->ds_node_num, reg->hr_dev_name);
954 o2hb_dump_slot(hb_block);
956 slot->ds_equal_samples++;
960 /* we don't care if these wrap.. the state transitions below
961 * clear at the right places */
962 cputime = le64_to_cpu(hb_block->hb_seq);
963 if (slot->ds_last_time != cputime)
964 slot->ds_changed_samples++;
966 slot->ds_equal_samples++;
967 slot->ds_last_time = cputime;
969 /* The node changed heartbeat generations. We assume this to
970 * mean it dropped off but came back before we timed out. We
971 * want to consider it down for the time being but don't want
972 * to lose any changed_samples state we might build up to
973 * considering it live again. */
974 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
976 slot->ds_equal_samples = 0;
977 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
978 "to 0x%llx)\n", slot->ds_node_num,
979 (long long)slot->ds_last_generation,
980 (long long)le64_to_cpu(hb_block->hb_generation));
983 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
985 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
986 "seq %llu last %llu changed %u equal %u\n",
987 slot->ds_node_num, (long long)slot->ds_last_generation,
988 le32_to_cpu(hb_block->hb_cksum),
989 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
990 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
991 slot->ds_equal_samples);
993 spin_lock(&o2hb_live_lock);
996 /* dead nodes only come to life after some number of
997 * changes at any time during their dead time */
998 if (list_empty(&slot->ds_live_item) &&
999 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
1000 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
1001 slot->ds_node_num, (long long)slot->ds_last_generation);
1003 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1005 /* first on the list generates a callback */
1006 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1007 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
1008 "bitmap\n", slot->ds_node_num);
1009 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1011 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
1018 list_add_tail(&slot->ds_live_item,
1019 &o2hb_live_slots[slot->ds_node_num]);
1021 slot->ds_equal_samples = 0;
1023 /* We want to be sure that all nodes agree on the
1024 * number of milliseconds before a node will be
1025 * considered dead. The self-fencing timeout is
1026 * computed from this value, and a discrepancy might
1027 * result in heartbeat calling a node dead when it
1028 * hasn't self-fenced yet. */
1029 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
1030 if (slot_dead_ms && slot_dead_ms != dead_ms) {
1031 /* TODO: Perhaps we can fail the region here. */
1032 mlog(ML_ERROR, "Node %d on device %s has a dead count "
1033 "of %u ms, but our count is %u ms.\n"
1034 "Please double check your configuration values "
1035 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1036 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
1042 /* if the list is dead, we're done.. */
1043 if (list_empty(&slot->ds_live_item))
1046 /* live nodes only go dead after enough consequtive missed
1047 * samples.. reset the missed counter whenever we see
1049 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1050 mlog(ML_HEARTBEAT, "Node %d left my region\n",
1053 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1055 /* last off the live_slot generates a callback */
1056 list_del_init(&slot->ds_live_item);
1057 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1058 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
1059 "nodes bitmap\n", slot->ds_node_num);
1060 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1062 /* node can be null */
1063 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1064 node, slot->ds_node_num);
1070 /* We don't clear this because the node is still
1071 * actually writing new blocks. */
1073 slot->ds_changed_samples = 0;
1076 if (slot->ds_changed_samples) {
1077 slot->ds_changed_samples = 0;
1078 slot->ds_equal_samples = 0;
1081 spin_unlock(&o2hb_live_lock);
1084 o2hb_run_event_list(&event);
1087 o2nm_node_put(node);
1091 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1093 return find_last_bit(nodes, numbits);
1096 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1098 int i, ret, highest_node;
1099 int membership_change = 0, own_slot_ok = 0;
1100 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
1101 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1102 struct o2hb_bio_wait_ctxt write_wc;
1104 ret = o2nm_configured_node_map(configured_nodes,
1105 sizeof(configured_nodes));
1112 * If a node is not configured but is in the livemap, we still need
1113 * to read the slot so as to be able to remove it from the livemap.
1115 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
1117 while ((i = find_next_bit(live_node_bitmap,
1118 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1119 set_bit(i, configured_nodes);
1122 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1123 if (highest_node >= O2NM_MAX_NODES) {
1124 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1129 /* No sense in reading the slots of nodes that don't exist
1130 * yet. Of course, if the node definitions have holes in them
1131 * then we're reading an empty slot anyway... Consider this
1133 ret = o2hb_read_slots(reg, highest_node + 1);
1139 /* With an up to date view of the slots, we can check that no
1140 * other node has been improperly configured to heartbeat in
1142 own_slot_ok = o2hb_check_own_slot(reg);
1144 /* fill in the proper info for our next heartbeat */
1145 o2hb_prepare_block(reg, reg->hr_generation);
1147 ret = o2hb_issue_node_write(reg, &write_wc);
1154 while((i = find_next_bit(configured_nodes,
1155 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1156 membership_change |= o2hb_check_slot(reg, ®->hr_slots[i]);
1160 * We have to be sure we've advertised ourselves on disk
1161 * before we can go to steady state. This ensures that
1162 * people we find in our steady state have seen us.
1164 o2hb_wait_on_io(&write_wc);
1165 if (write_wc.wc_error) {
1166 /* Do not re-arm the write timeout on I/O error - we
1167 * can't be sure that the new block ever made it to
1169 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1170 write_wc.wc_error, reg->hr_dev_name);
1171 ret = write_wc.wc_error;
1175 /* Skip disarming the timeout if own slot has stale/bad data */
1177 o2hb_set_quorum_device(reg);
1178 o2hb_arm_timeout(reg);
1179 reg->hr_last_timeout_start = jiffies;
1183 /* let the person who launched us know when things are steady */
1184 if (atomic_read(®->hr_steady_iterations) != 0) {
1185 if (!ret && own_slot_ok && !membership_change) {
1186 if (atomic_dec_and_test(®->hr_steady_iterations))
1187 wake_up(&o2hb_steady_queue);
1191 if (atomic_read(®->hr_steady_iterations) != 0) {
1192 if (atomic_dec_and_test(®->hr_unsteady_iterations)) {
1193 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1194 "heartbeart on region %s (%s)\n",
1195 config_item_name(®->hr_item),
1197 atomic_set(®->hr_steady_iterations, 0);
1198 reg->hr_aborted_start = 1;
1199 wake_up(&o2hb_steady_queue);
1208 * we ride the region ref that the region dir holds. before the region
1209 * dir is removed and drops it ref it will wait to tear down this
1212 static int o2hb_thread(void *data)
1215 struct o2hb_region *reg = data;
1216 struct o2hb_bio_wait_ctxt write_wc;
1217 ktime_t before_hb, after_hb;
1218 unsigned int elapsed_msec;
1220 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1222 set_user_nice(current, MIN_NICE);
1225 ret = o2nm_depend_this_node();
1227 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1228 reg->hr_node_deleted = 1;
1229 wake_up(&o2hb_steady_queue);
1233 while (!kthread_should_stop() &&
1234 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1235 /* We track the time spent inside
1236 * o2hb_do_disk_heartbeat so that we avoid more than
1237 * hr_timeout_ms between disk writes. On busy systems
1238 * this should result in a heartbeat which is less
1239 * likely to time itself out. */
1240 before_hb = ktime_get_real();
1242 ret = o2hb_do_disk_heartbeat(reg);
1243 reg->hr_last_hb_status = ret;
1245 after_hb = ktime_get_real();
1247 elapsed_msec = (unsigned int)
1248 ktime_ms_delta(after_hb, before_hb);
1251 "start = %lld, end = %lld, msec = %u, ret = %d\n",
1252 before_hb, after_hb, elapsed_msec, ret);
1254 if (!kthread_should_stop() &&
1255 elapsed_msec < reg->hr_timeout_ms) {
1256 /* the kthread api has blocked signals for us so no
1257 * need to record the return value. */
1258 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1262 o2hb_disarm_timeout(reg);
1264 /* unclean stop is only used in very bad situation */
1265 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1266 o2hb_shutdown_slot(®->hr_slots[i]);
1268 /* Explicit down notification - avoid forcing the other nodes
1269 * to timeout on this region when we could just as easily
1270 * write a clear generation - thus indicating to them that
1271 * this node has left this region.
1273 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1274 o2hb_prepare_block(reg, 0);
1275 ret = o2hb_issue_node_write(reg, &write_wc);
1277 o2hb_wait_on_io(&write_wc);
1283 o2nm_undepend_this_node();
1285 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1290 #ifdef CONFIG_DEBUG_FS
1291 static int o2hb_debug_open(struct inode *inode, struct file *file)
1293 struct o2hb_debug_buf *db = inode->i_private;
1294 struct o2hb_region *reg;
1295 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1301 /* max_nodes should be the largest bitmap we pass here */
1302 BUG_ON(sizeof(map) < db->db_size);
1304 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1308 switch (db->db_type) {
1309 case O2HB_DB_TYPE_LIVENODES:
1310 case O2HB_DB_TYPE_LIVEREGIONS:
1311 case O2HB_DB_TYPE_QUORUMREGIONS:
1312 case O2HB_DB_TYPE_FAILEDREGIONS:
1313 spin_lock(&o2hb_live_lock);
1314 memcpy(map, db->db_data, db->db_size);
1315 spin_unlock(&o2hb_live_lock);
1318 case O2HB_DB_TYPE_REGION_LIVENODES:
1319 spin_lock(&o2hb_live_lock);
1320 reg = (struct o2hb_region *)db->db_data;
1321 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1322 spin_unlock(&o2hb_live_lock);
1325 case O2HB_DB_TYPE_REGION_NUMBER:
1326 reg = (struct o2hb_region *)db->db_data;
1327 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1328 reg->hr_region_num);
1331 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1332 reg = (struct o2hb_region *)db->db_data;
1333 lts = reg->hr_last_timeout_start;
1334 /* If 0, it has never been set before */
1336 lts = jiffies_to_msecs(jiffies - lts);
1337 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1340 case O2HB_DB_TYPE_REGION_PINNED:
1341 reg = (struct o2hb_region *)db->db_data;
1342 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1343 !!reg->hr_item_pinned);
1350 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1351 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1352 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1355 i_size_write(inode, out);
1357 file->private_data = buf;
1364 static int o2hb_debug_release(struct inode *inode, struct file *file)
1366 kfree(file->private_data);
1370 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1371 size_t nbytes, loff_t *ppos)
1373 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1374 i_size_read(file->f_mapping->host));
1377 static int o2hb_debug_open(struct inode *inode, struct file *file)
1381 static int o2hb_debug_release(struct inode *inode, struct file *file)
1385 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1386 size_t nbytes, loff_t *ppos)
1390 #endif /* CONFIG_DEBUG_FS */
1392 static const struct file_operations o2hb_debug_fops = {
1393 .open = o2hb_debug_open,
1394 .release = o2hb_debug_release,
1395 .read = o2hb_debug_read,
1396 .llseek = generic_file_llseek,
1399 void o2hb_exit(void)
1401 debugfs_remove(o2hb_debug_failedregions);
1402 debugfs_remove(o2hb_debug_quorumregions);
1403 debugfs_remove(o2hb_debug_liveregions);
1404 debugfs_remove(o2hb_debug_livenodes);
1405 debugfs_remove(o2hb_debug_dir);
1406 kfree(o2hb_db_livenodes);
1407 kfree(o2hb_db_liveregions);
1408 kfree(o2hb_db_quorumregions);
1409 kfree(o2hb_db_failedregions);
1412 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1413 struct o2hb_debug_buf **db, int db_len,
1414 int type, int size, int len, void *data)
1416 *db = kmalloc(db_len, GFP_KERNEL);
1420 (*db)->db_type = type;
1421 (*db)->db_size = size;
1422 (*db)->db_len = len;
1423 (*db)->db_data = data;
1425 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1429 static int o2hb_debug_init(void)
1433 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1434 if (!o2hb_debug_dir) {
1439 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1442 sizeof(*o2hb_db_livenodes),
1443 O2HB_DB_TYPE_LIVENODES,
1444 sizeof(o2hb_live_node_bitmap),
1446 o2hb_live_node_bitmap);
1447 if (!o2hb_debug_livenodes) {
1452 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1454 &o2hb_db_liveregions,
1455 sizeof(*o2hb_db_liveregions),
1456 O2HB_DB_TYPE_LIVEREGIONS,
1457 sizeof(o2hb_live_region_bitmap),
1459 o2hb_live_region_bitmap);
1460 if (!o2hb_debug_liveregions) {
1465 o2hb_debug_quorumregions =
1466 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1468 &o2hb_db_quorumregions,
1469 sizeof(*o2hb_db_quorumregions),
1470 O2HB_DB_TYPE_QUORUMREGIONS,
1471 sizeof(o2hb_quorum_region_bitmap),
1473 o2hb_quorum_region_bitmap);
1474 if (!o2hb_debug_quorumregions) {
1479 o2hb_debug_failedregions =
1480 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1482 &o2hb_db_failedregions,
1483 sizeof(*o2hb_db_failedregions),
1484 O2HB_DB_TYPE_FAILEDREGIONS,
1485 sizeof(o2hb_failed_region_bitmap),
1487 o2hb_failed_region_bitmap);
1488 if (!o2hb_debug_failedregions) {
1505 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1506 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1508 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1509 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1511 INIT_LIST_HEAD(&o2hb_node_events);
1513 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1514 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1515 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1516 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1517 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1519 o2hb_dependent_users = 0;
1521 return o2hb_debug_init();
1524 /* if we're already in a callback then we're already serialized by the sem */
1525 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1528 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1530 memcpy(map, &o2hb_live_node_bitmap, bytes);
1534 * get a map of all nodes that are heartbeating in any regions
1536 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1538 /* callers want to serialize this map and callbacks so that they
1539 * can trust that they don't miss nodes coming to the party */
1540 down_read(&o2hb_callback_sem);
1541 spin_lock(&o2hb_live_lock);
1542 o2hb_fill_node_map_from_callback(map, bytes);
1543 spin_unlock(&o2hb_live_lock);
1544 up_read(&o2hb_callback_sem);
1546 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1549 * heartbeat configfs bits. The heartbeat set is a default set under
1550 * the cluster set in nodemanager.c.
1553 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1555 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1558 /* drop_item only drops its ref after killing the thread, nothing should
1559 * be using the region anymore. this has to clean up any state that
1560 * attributes might have built up. */
1561 static void o2hb_region_release(struct config_item *item)
1565 struct o2hb_region *reg = to_o2hb_region(item);
1567 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1569 kfree(reg->hr_tmp_block);
1571 if (reg->hr_slot_data) {
1572 for (i = 0; i < reg->hr_num_pages; i++) {
1573 page = reg->hr_slot_data[i];
1577 kfree(reg->hr_slot_data);
1581 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1583 kfree(reg->hr_slots);
1585 debugfs_remove(reg->hr_debug_livenodes);
1586 debugfs_remove(reg->hr_debug_regnum);
1587 debugfs_remove(reg->hr_debug_elapsed_time);
1588 debugfs_remove(reg->hr_debug_pinned);
1589 debugfs_remove(reg->hr_debug_dir);
1590 kfree(reg->hr_db_livenodes);
1591 kfree(reg->hr_db_regnum);
1592 kfree(reg->hr_db_elapsed_time);
1593 kfree(reg->hr_db_pinned);
1595 spin_lock(&o2hb_live_lock);
1596 list_del(®->hr_all_item);
1597 spin_unlock(&o2hb_live_lock);
1599 o2net_unregister_handler_list(®->hr_handler_list);
1603 static int o2hb_read_block_input(struct o2hb_region *reg,
1605 unsigned long *ret_bytes,
1606 unsigned int *ret_bits)
1608 unsigned long bytes;
1609 char *p = (char *)page;
1611 bytes = simple_strtoul(p, &p, 0);
1612 if (!p || (*p && (*p != '\n')))
1615 /* Heartbeat and fs min / max block sizes are the same. */
1616 if (bytes > 4096 || bytes < 512)
1618 if (hweight16(bytes) != 1)
1624 *ret_bits = ffs(bytes) - 1;
1629 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1632 return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1635 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1639 struct o2hb_region *reg = to_o2hb_region(item);
1641 unsigned long block_bytes;
1642 unsigned int block_bits;
1647 status = o2hb_read_block_input(reg, page, &block_bytes,
1652 reg->hr_block_bytes = (unsigned int)block_bytes;
1653 reg->hr_block_bits = block_bits;
1658 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1661 return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1664 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1668 struct o2hb_region *reg = to_o2hb_region(item);
1669 unsigned long long tmp;
1670 char *p = (char *)page;
1675 tmp = simple_strtoull(p, &p, 0);
1676 if (!p || (*p && (*p != '\n')))
1679 reg->hr_start_block = tmp;
1684 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1686 return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1689 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1693 struct o2hb_region *reg = to_o2hb_region(item);
1695 char *p = (char *)page;
1700 tmp = simple_strtoul(p, &p, 0);
1701 if (!p || (*p && (*p != '\n')))
1704 if (tmp > O2NM_MAX_NODES || tmp == 0)
1707 reg->hr_blocks = (unsigned int)tmp;
1712 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1714 unsigned int ret = 0;
1716 if (to_o2hb_region(item)->hr_bdev)
1717 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1722 static void o2hb_init_region_params(struct o2hb_region *reg)
1724 reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1725 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1727 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1728 reg->hr_start_block, reg->hr_blocks);
1729 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1730 reg->hr_block_bytes, reg->hr_block_bits);
1731 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1732 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1735 static int o2hb_map_slot_data(struct o2hb_region *reg)
1738 unsigned int last_slot;
1739 unsigned int spp = reg->hr_slots_per_page;
1742 struct o2hb_disk_slot *slot;
1744 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1745 if (reg->hr_tmp_block == NULL)
1748 reg->hr_slots = kcalloc(reg->hr_blocks,
1749 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1750 if (reg->hr_slots == NULL)
1753 for(i = 0; i < reg->hr_blocks; i++) {
1754 slot = ®->hr_slots[i];
1755 slot->ds_node_num = i;
1756 INIT_LIST_HEAD(&slot->ds_live_item);
1757 slot->ds_raw_block = NULL;
1760 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1761 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1762 "at %u blocks per page\n",
1763 reg->hr_num_pages, reg->hr_blocks, spp);
1765 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1767 if (!reg->hr_slot_data)
1770 for(i = 0; i < reg->hr_num_pages; i++) {
1771 page = alloc_page(GFP_KERNEL);
1775 reg->hr_slot_data[i] = page;
1777 last_slot = i * spp;
1778 raw = page_address(page);
1780 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1782 BUG_ON((j + last_slot) >= reg->hr_blocks);
1784 slot = ®->hr_slots[j + last_slot];
1785 slot->ds_raw_block =
1786 (struct o2hb_disk_heartbeat_block *) raw;
1788 raw += reg->hr_block_bytes;
1795 /* Read in all the slots available and populate the tracking
1796 * structures so that we can start with a baseline idea of what's
1798 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1801 struct o2hb_disk_slot *slot;
1802 struct o2hb_disk_heartbeat_block *hb_block;
1804 ret = o2hb_read_slots(reg, reg->hr_blocks);
1808 /* We only want to get an idea of the values initially in each
1809 * slot, so we do no verification - o2hb_check_slot will
1810 * actually determine if each configured slot is valid and
1811 * whether any values have changed. */
1812 for(i = 0; i < reg->hr_blocks; i++) {
1813 slot = ®->hr_slots[i];
1814 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1816 /* Only fill the values that o2hb_check_slot uses to
1817 * determine changing slots */
1818 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1819 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1826 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1827 static ssize_t o2hb_region_dev_store(struct config_item *item,
1831 struct o2hb_region *reg = to_o2hb_region(item);
1832 struct task_struct *hb_task;
1835 char *p = (char *)page;
1837 struct inode *inode;
1838 ssize_t ret = -EINVAL;
1844 /* We can't heartbeat without having had our node number
1845 * configured yet. */
1846 if (o2nm_this_node() == O2NM_MAX_NODES)
1849 fd = simple_strtol(p, &p, 0);
1850 if (!p || (*p && (*p != '\n')))
1853 if (fd < 0 || fd >= INT_MAX)
1860 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1861 reg->hr_block_bytes == 0)
1864 inode = igrab(f.file->f_mapping->host);
1868 if (!S_ISBLK(inode->i_mode))
1871 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1872 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1874 reg->hr_bdev = NULL;
1879 bdevname(reg->hr_bdev, reg->hr_dev_name);
1881 sectsize = bdev_logical_block_size(reg->hr_bdev);
1882 if (sectsize != reg->hr_block_bytes) {
1884 "blocksize %u incorrect for device, expected %d",
1885 reg->hr_block_bytes, sectsize);
1890 o2hb_init_region_params(reg);
1892 /* Generation of zero is invalid */
1894 get_random_bytes(®->hr_generation,
1895 sizeof(reg->hr_generation));
1896 } while (reg->hr_generation == 0);
1898 ret = o2hb_map_slot_data(reg);
1904 ret = o2hb_populate_slot_data(reg);
1910 INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout);
1911 INIT_DELAYED_WORK(®->hr_nego_timeout_work, o2hb_nego_timeout);
1914 * A node is considered live after it has beat LIVE_THRESHOLD
1915 * times. We're not steady until we've given them a chance
1916 * _after_ our first read.
1917 * The default threshold is bare minimum so as to limit the delay
1918 * during mounts. For global heartbeat, the threshold doubled for the
1921 live_threshold = O2HB_LIVE_THRESHOLD;
1922 if (o2hb_global_heartbeat_active()) {
1923 spin_lock(&o2hb_live_lock);
1924 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1925 live_threshold <<= 1;
1926 spin_unlock(&o2hb_live_lock);
1929 atomic_set(®->hr_steady_iterations, live_threshold);
1930 /* unsteady_iterations is triple the steady_iterations */
1931 atomic_set(®->hr_unsteady_iterations, (live_threshold * 3));
1933 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1934 reg->hr_item.ci_name);
1935 if (IS_ERR(hb_task)) {
1936 ret = PTR_ERR(hb_task);
1941 spin_lock(&o2hb_live_lock);
1942 reg->hr_task = hb_task;
1943 spin_unlock(&o2hb_live_lock);
1945 ret = wait_event_interruptible(o2hb_steady_queue,
1946 atomic_read(®->hr_steady_iterations) == 0 ||
1947 reg->hr_node_deleted);
1949 atomic_set(®->hr_steady_iterations, 0);
1950 reg->hr_aborted_start = 1;
1953 if (reg->hr_aborted_start) {
1958 if (reg->hr_node_deleted) {
1963 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1964 spin_lock(&o2hb_live_lock);
1965 hb_task = reg->hr_task;
1966 if (o2hb_global_heartbeat_active())
1967 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1968 spin_unlock(&o2hb_live_lock);
1975 if (hb_task && o2hb_global_heartbeat_active())
1976 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1977 config_item_name(®->hr_item), reg->hr_dev_name);
1986 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1987 reg->hr_bdev = NULL;
1993 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1995 struct o2hb_region *reg = to_o2hb_region(item);
1998 spin_lock(&o2hb_live_lock);
2000 pid = task_pid_nr(reg->hr_task);
2001 spin_unlock(&o2hb_live_lock);
2006 return sprintf(page, "%u\n", pid);
2009 CONFIGFS_ATTR(o2hb_region_, block_bytes);
2010 CONFIGFS_ATTR(o2hb_region_, start_block);
2011 CONFIGFS_ATTR(o2hb_region_, blocks);
2012 CONFIGFS_ATTR(o2hb_region_, dev);
2013 CONFIGFS_ATTR_RO(o2hb_region_, pid);
2015 static struct configfs_attribute *o2hb_region_attrs[] = {
2016 &o2hb_region_attr_block_bytes,
2017 &o2hb_region_attr_start_block,
2018 &o2hb_region_attr_blocks,
2019 &o2hb_region_attr_dev,
2020 &o2hb_region_attr_pid,
2024 static struct configfs_item_operations o2hb_region_item_ops = {
2025 .release = o2hb_region_release,
2028 static struct config_item_type o2hb_region_type = {
2029 .ct_item_ops = &o2hb_region_item_ops,
2030 .ct_attrs = o2hb_region_attrs,
2031 .ct_owner = THIS_MODULE,
2036 struct o2hb_heartbeat_group {
2037 struct config_group hs_group;
2041 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2044 container_of(group, struct o2hb_heartbeat_group, hs_group)
2048 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2053 debugfs_create_dir(config_item_name(®->hr_item), dir);
2054 if (!reg->hr_debug_dir) {
2059 reg->hr_debug_livenodes =
2060 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2062 &(reg->hr_db_livenodes),
2063 sizeof(*(reg->hr_db_livenodes)),
2064 O2HB_DB_TYPE_REGION_LIVENODES,
2065 sizeof(reg->hr_live_node_bitmap),
2066 O2NM_MAX_NODES, reg);
2067 if (!reg->hr_debug_livenodes) {
2072 reg->hr_debug_regnum =
2073 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2075 &(reg->hr_db_regnum),
2076 sizeof(*(reg->hr_db_regnum)),
2077 O2HB_DB_TYPE_REGION_NUMBER,
2078 0, O2NM_MAX_NODES, reg);
2079 if (!reg->hr_debug_regnum) {
2084 reg->hr_debug_elapsed_time =
2085 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2087 &(reg->hr_db_elapsed_time),
2088 sizeof(*(reg->hr_db_elapsed_time)),
2089 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2091 if (!reg->hr_debug_elapsed_time) {
2096 reg->hr_debug_pinned =
2097 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2099 &(reg->hr_db_pinned),
2100 sizeof(*(reg->hr_db_pinned)),
2101 O2HB_DB_TYPE_REGION_PINNED,
2103 if (!reg->hr_debug_pinned) {
2113 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2116 struct o2hb_region *reg = NULL;
2119 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2121 return ERR_PTR(-ENOMEM);
2123 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2124 ret = -ENAMETOOLONG;
2128 spin_lock(&o2hb_live_lock);
2129 reg->hr_region_num = 0;
2130 if (o2hb_global_heartbeat_active()) {
2131 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2133 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2134 spin_unlock(&o2hb_live_lock);
2138 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2140 list_add_tail(®->hr_all_item, &o2hb_all_regions);
2141 spin_unlock(&o2hb_live_lock);
2143 config_item_init_type_name(®->hr_item, name, &o2hb_region_type);
2145 /* this is the same way to generate msg key as dlm, for local heartbeat,
2146 * name is also the same, so make initial crc value different to avoid
2147 * message key conflict.
2149 reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2150 name, strlen(name));
2151 INIT_LIST_HEAD(®->hr_handler_list);
2152 ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2153 sizeof(struct o2hb_nego_msg),
2154 o2hb_nego_timeout_handler,
2155 reg, NULL, ®->hr_handler_list);
2159 ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2160 sizeof(struct o2hb_nego_msg),
2161 o2hb_nego_approve_handler,
2162 reg, NULL, ®->hr_handler_list);
2164 goto unregister_handler;
2166 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2168 config_item_put(®->hr_item);
2169 goto unregister_handler;
2172 return ®->hr_item;
2175 o2net_unregister_handler_list(®->hr_handler_list);
2177 spin_lock(&o2hb_live_lock);
2178 list_del(®->hr_all_item);
2179 if (o2hb_global_heartbeat_active())
2180 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2181 spin_unlock(&o2hb_live_lock);
2184 return ERR_PTR(ret);
2187 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2188 struct config_item *item)
2190 struct task_struct *hb_task;
2191 struct o2hb_region *reg = to_o2hb_region(item);
2192 int quorum_region = 0;
2194 /* stop the thread when the user removes the region dir */
2195 spin_lock(&o2hb_live_lock);
2196 hb_task = reg->hr_task;
2197 reg->hr_task = NULL;
2198 reg->hr_item_dropped = 1;
2199 spin_unlock(&o2hb_live_lock);
2202 kthread_stop(hb_task);
2204 if (o2hb_global_heartbeat_active()) {
2205 spin_lock(&o2hb_live_lock);
2206 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2207 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2208 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2210 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2211 spin_unlock(&o2hb_live_lock);
2212 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2213 ((atomic_read(®->hr_steady_iterations) == 0) ?
2214 "stopped" : "start aborted"), config_item_name(item),
2219 * If we're racing a dev_write(), we need to wake them. They will
2220 * check reg->hr_task
2222 if (atomic_read(®->hr_steady_iterations) != 0) {
2223 reg->hr_aborted_start = 1;
2224 atomic_set(®->hr_steady_iterations, 0);
2225 wake_up(&o2hb_steady_queue);
2228 config_item_put(item);
2230 if (!o2hb_global_heartbeat_active() || !quorum_region)
2234 * If global heartbeat active and there are dependent users,
2235 * pin all regions if quorum region count <= CUT_OFF
2237 spin_lock(&o2hb_live_lock);
2239 if (!o2hb_dependent_users)
2242 if (bitmap_weight(o2hb_quorum_region_bitmap,
2243 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2244 o2hb_region_pin(NULL);
2247 spin_unlock(&o2hb_live_lock);
2250 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2253 return sprintf(page, "%u\n", o2hb_dead_threshold);
2256 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2257 const char *page, size_t count)
2260 char *p = (char *)page;
2262 tmp = simple_strtoul(p, &p, 10);
2263 if (!p || (*p && (*p != '\n')))
2266 /* this will validate ranges for us. */
2267 o2hb_dead_threshold_set((unsigned int) tmp);
2272 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2275 return sprintf(page, "%s\n",
2276 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2279 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2280 const char *page, size_t count)
2286 len = (page[count - 1] == '\n') ? count - 1 : count;
2290 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2291 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2294 ret = o2hb_global_heartbeat_mode_set(i);
2296 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2297 o2hb_heartbeat_mode_desc[i]);
2305 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2306 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2308 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2309 &o2hb_heartbeat_group_attr_dead_threshold,
2310 &o2hb_heartbeat_group_attr_mode,
2314 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2315 .make_item = o2hb_heartbeat_group_make_item,
2316 .drop_item = o2hb_heartbeat_group_drop_item,
2319 static struct config_item_type o2hb_heartbeat_group_type = {
2320 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2321 .ct_attrs = o2hb_heartbeat_group_attrs,
2322 .ct_owner = THIS_MODULE,
2325 /* this is just here to avoid touching group in heartbeat.h which the
2326 * entire damn world #includes */
2327 struct config_group *o2hb_alloc_hb_set(void)
2329 struct o2hb_heartbeat_group *hs = NULL;
2330 struct config_group *ret = NULL;
2332 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2336 config_group_init_type_name(&hs->hs_group, "heartbeat",
2337 &o2hb_heartbeat_group_type);
2339 ret = &hs->hs_group;
2346 void o2hb_free_hb_set(struct config_group *group)
2348 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2352 /* hb callback registration and issuing */
2354 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2356 if (type == O2HB_NUM_CB)
2357 return ERR_PTR(-EINVAL);
2359 return &o2hb_callbacks[type];
2362 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2363 enum o2hb_callback_type type,
2368 INIT_LIST_HEAD(&hc->hc_item);
2371 hc->hc_priority = priority;
2373 hc->hc_magic = O2HB_CB_MAGIC;
2375 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2378 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2379 * In global heartbeat mode, region_uuid passed is NULL.
2381 * In local, we only pin the matching region. In global we pin all the active
2384 static int o2hb_region_pin(const char *region_uuid)
2386 int ret = 0, found = 0;
2387 struct o2hb_region *reg;
2390 assert_spin_locked(&o2hb_live_lock);
2392 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2393 if (reg->hr_item_dropped)
2396 uuid = config_item_name(®->hr_item);
2398 /* local heartbeat */
2400 if (strcmp(region_uuid, uuid))
2405 if (reg->hr_item_pinned || reg->hr_item_dropped)
2408 /* Ignore ENOENT only for local hb (userdlm domain) */
2409 ret = o2nm_depend_item(®->hr_item);
2411 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2412 reg->hr_item_pinned = 1;
2414 if (ret == -ENOENT && found)
2417 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2431 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2432 * In global heartbeat mode, region_uuid passed is NULL.
2434 * In local, we only unpin the matching region. In global we unpin all the
2437 static void o2hb_region_unpin(const char *region_uuid)
2439 struct o2hb_region *reg;
2443 assert_spin_locked(&o2hb_live_lock);
2445 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2446 if (reg->hr_item_dropped)
2449 uuid = config_item_name(®->hr_item);
2451 if (strcmp(region_uuid, uuid))
2456 if (reg->hr_item_pinned) {
2457 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2458 o2nm_undepend_item(®->hr_item);
2459 reg->hr_item_pinned = 0;
2466 static int o2hb_region_inc_user(const char *region_uuid)
2470 spin_lock(&o2hb_live_lock);
2472 /* local heartbeat */
2473 if (!o2hb_global_heartbeat_active()) {
2474 ret = o2hb_region_pin(region_uuid);
2479 * if global heartbeat active and this is the first dependent user,
2480 * pin all regions if quorum region count <= CUT_OFF
2482 o2hb_dependent_users++;
2483 if (o2hb_dependent_users > 1)
2486 if (bitmap_weight(o2hb_quorum_region_bitmap,
2487 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2488 ret = o2hb_region_pin(NULL);
2491 spin_unlock(&o2hb_live_lock);
2495 void o2hb_region_dec_user(const char *region_uuid)
2497 spin_lock(&o2hb_live_lock);
2499 /* local heartbeat */
2500 if (!o2hb_global_heartbeat_active()) {
2501 o2hb_region_unpin(region_uuid);
2506 * if global heartbeat active and there are no dependent users,
2507 * unpin all quorum regions
2509 o2hb_dependent_users--;
2510 if (!o2hb_dependent_users)
2511 o2hb_region_unpin(NULL);
2514 spin_unlock(&o2hb_live_lock);
2517 int o2hb_register_callback(const char *region_uuid,
2518 struct o2hb_callback_func *hc)
2520 struct o2hb_callback_func *f;
2521 struct o2hb_callback *hbcall;
2524 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2525 BUG_ON(!list_empty(&hc->hc_item));
2527 hbcall = hbcall_from_type(hc->hc_type);
2528 if (IS_ERR(hbcall)) {
2529 ret = PTR_ERR(hbcall);
2534 ret = o2hb_region_inc_user(region_uuid);
2541 down_write(&o2hb_callback_sem);
2543 list_for_each_entry(f, &hbcall->list, hc_item) {
2544 if (hc->hc_priority < f->hc_priority) {
2545 list_add_tail(&hc->hc_item, &f->hc_item);
2549 if (list_empty(&hc->hc_item))
2550 list_add_tail(&hc->hc_item, &hbcall->list);
2552 up_write(&o2hb_callback_sem);
2555 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2556 ret, __builtin_return_address(0), hc);
2559 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2561 void o2hb_unregister_callback(const char *region_uuid,
2562 struct o2hb_callback_func *hc)
2564 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2566 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2567 __builtin_return_address(0), hc);
2569 /* XXX Can this happen _with_ a region reference? */
2570 if (list_empty(&hc->hc_item))
2574 o2hb_region_dec_user(region_uuid);
2576 down_write(&o2hb_callback_sem);
2578 list_del_init(&hc->hc_item);
2580 up_write(&o2hb_callback_sem);
2582 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2584 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2586 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2588 spin_lock(&o2hb_live_lock);
2589 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2590 spin_unlock(&o2hb_live_lock);
2591 if (!test_bit(node_num, testing_map)) {
2593 "node (%u) does not have heartbeating enabled.\n",
2600 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2602 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2604 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2606 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2607 if (!test_bit(node_num, testing_map)) {
2609 "node (%u) does not have heartbeating enabled.\n",
2616 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2619 * this is just a hack until we get the plumbing which flips file systems
2620 * read only and drops the hb ref instead of killing the node dead.
2622 void o2hb_stop_all_regions(void)
2624 struct o2hb_region *reg;
2626 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2628 spin_lock(&o2hb_live_lock);
2630 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2631 reg->hr_unclean_stop = 1;
2633 spin_unlock(&o2hb_live_lock);
2635 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2637 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2639 struct o2hb_region *reg;
2643 spin_lock(&o2hb_live_lock);
2646 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2647 if (reg->hr_item_dropped)
2650 mlog(0, "Region: %s\n", config_item_name(®->hr_item));
2651 if (numregs < max_regions) {
2652 memcpy(p, config_item_name(®->hr_item),
2653 O2HB_MAX_REGION_NAME_LEN);
2654 p += O2HB_MAX_REGION_NAME_LEN;
2659 spin_unlock(&o2hb_live_lock);
2663 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2665 int o2hb_global_heartbeat_active(void)
2667 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2669 EXPORT_SYMBOL(o2hb_global_heartbeat_active);