~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ocfs2/cluster/heartbeat.c

Version: ~ [ linux-4.14 ] ~ [ linux-4.13.12 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.61 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.97 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.46 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.80 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.50 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.95 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /* -*- mode: c; c-basic-offset: 8; -*-
  2  * vim: noexpandtab sw=8 ts=8 sts=0:
  3  *
  4  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
  5  *
  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.
 10  *
 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.
 15  *
 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.
 20  */
 21 
 22 #include <linux/kernel.h>
 23 #include <linux/sched.h>
 24 #include <linux/jiffies.h>
 25 #include <linux/module.h>
 26 #include <linux/fs.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"
 41 #include "tcp.h"
 42 #include "nodemanager.h"
 43 #include "quorum.h"
 44 
 45 #include "masklog.h"
 46 
 47 
 48 /*
 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.
 53  */
 54 static DECLARE_RWSEM(o2hb_callback_sem);
 55 
 56 /*
 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.
 59  */
 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);
 65 
 66 /*
 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
 71  *              heartbeat on it.
 72  *      - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
 73  */
 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)];
 78 
 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 {
 88         int db_type;
 89         int db_size;
 90         int db_len;
 91         void *db_data;
 92 };
 93 
 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;
 98 
 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"
107 
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;
113 
114 static LIST_HEAD(o2hb_all_regions);
115 
116 static struct o2hb_callback {
117         struct list_head list;
118 } o2hb_callbacks[O2HB_NUM_CB];
119 
120 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
121 
122 #define O2HB_DEFAULT_BLOCK_BITS       9
123 
124 enum o2hb_heartbeat_modes {
125         O2HB_HEARTBEAT_LOCAL            = 0,
126         O2HB_HEARTBEAT_GLOBAL,
127         O2HB_HEARTBEAT_NUM_MODES,
128 };
129 
130 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
131                 "local",        /* O2HB_HEARTBEAT_LOCAL */
132                 "global",       /* O2HB_HEARTBEAT_GLOBAL */
133 };
134 
135 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
136 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
137 
138 /*
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.
143  */
144 unsigned int o2hb_dependent_users;
145 
146 /*
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.
151  */
152 #define O2HB_PIN_CUT_OFF                3
153 
154 /*
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.
158  *
159  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
160  * works for both file system and userdlm domains.
161  */
162 static int o2hb_region_pin(const char *region_uuid);
163 static void o2hb_region_unpin(const char *region_uuid);
164 
165 /* Only sets a new threshold if there are no active regions.
166  *
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)
171 {
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);
177         }
178 }
179 
180 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
181 {
182         int ret = -1;
183 
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;
188                         ret = 0;
189                 }
190                 spin_unlock(&o2hb_live_lock);
191         }
192 
193         return ret;
194 }
195 
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;
200         int                     hn_node_num;
201 };
202 
203 struct o2hb_disk_slot {
204         struct o2hb_disk_heartbeat_block *ds_raw_block;
205         u8                      ds_node_num;
206         u64                     ds_last_time;
207         u64                     ds_last_generation;
208         u16                     ds_equal_samples;
209         u16                     ds_changed_samples;
210         struct list_head        ds_live_item;
211 };
212 
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 */
215 struct o2hb_region {
216         struct config_item      hr_item;
217 
218         struct list_head        hr_all_item;
219         unsigned                hr_unclean_stop:1,
220                                 hr_aborted_start:1,
221                                 hr_item_pinned:1,
222                                 hr_item_dropped:1,
223                                 hr_node_deleted:1;
224 
225         /* protected by the hr_callback_sem */
226         struct task_struct      *hr_task;
227 
228         unsigned int            hr_blocks;
229         unsigned long long      hr_start_block;
230 
231         unsigned int            hr_block_bits;
232         unsigned int            hr_block_bytes;
233 
234         unsigned int            hr_slots_per_page;
235         unsigned int            hr_num_pages;
236 
237         struct page             **hr_slot_data;
238         struct block_device     *hr_bdev;
239         struct o2hb_disk_slot   *hr_slots;
240 
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;
244 
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;
254 
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;
259 
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;
263 
264         char                    hr_dev_name[BDEVNAME_SIZE];
265 
266         unsigned int            hr_timeout_ms;
267 
268         /* randomized as the region goes up and down so that a node
269          * recognizes a node going up and down in one iteration */
270         u64                     hr_generation;
271 
272         struct delayed_work     hr_write_timeout_work;
273         unsigned long           hr_last_timeout_start;
274 
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)];
278 
279         /* Used during o2hb_check_slot to hold a copy of the block
280          * being checked because we temporarily have to zero out the
281          * crc field. */
282         struct o2hb_disk_heartbeat_block *hr_tmp_block;
283 
284         /* Message key for negotiate timeout message. */
285         unsigned int            hr_key;
286         struct list_head        hr_handler_list;
287 
288         /* last hb status, 0 for success, other value for error. */
289         int                     hr_last_hb_status;
290 };
291 
292 struct o2hb_bio_wait_ctxt {
293         atomic_t          wc_num_reqs;
294         struct completion wc_io_complete;
295         int               wc_error;
296 };
297 
298 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
299 
300 enum {
301         O2HB_NEGO_TIMEOUT_MSG = 1,
302         O2HB_NEGO_APPROVE_MSG = 2,
303 };
304 
305 struct o2hb_nego_msg {
306         u8 node_num;
307 };
308 
309 static void o2hb_write_timeout(struct work_struct *work)
310 {
311         int failed, quorum;
312         struct o2hb_region *reg =
313                 container_of(work, struct o2hb_region,
314                              hr_write_timeout_work.work);
315 
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));
319 
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,
325                                         O2NM_MAX_REGIONS);
326                 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
327                                         O2NM_MAX_REGIONS);
328                 spin_unlock(&o2hb_live_lock);
329 
330                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
331                      quorum, failed);
332 
333                 /*
334                  * Fence if the number of failed regions >= half the number
335                  * of  quorum regions
336                  */
337                 if ((failed << 1) < quorum)
338                         return;
339         }
340 
341         o2quo_disk_timeout();
342 }
343 
344 static void o2hb_arm_timeout(struct o2hb_region *reg)
345 {
346         /* Arm writeout only after thread reaches steady state */
347         if (atomic_read(&reg->hr_steady_iterations) != 0)
348                 return;
349 
350         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
351              O2HB_MAX_WRITE_TIMEOUT_MS);
352 
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);
357         }
358         cancel_delayed_work(&reg->hr_write_timeout_work);
359         schedule_delayed_work(&reg->hr_write_timeout_work,
360                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
361 
362         cancel_delayed_work(&reg->hr_nego_timeout_work);
363         /* negotiate timeout must be less than write timeout. */
364         schedule_delayed_work(&reg->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));
367 }
368 
369 static void o2hb_disarm_timeout(struct o2hb_region *reg)
370 {
371         cancel_delayed_work_sync(&reg->hr_write_timeout_work);
372         cancel_delayed_work_sync(&reg->hr_nego_timeout_work);
373 }
374 
375 static int o2hb_send_nego_msg(int key, int type, u8 target)
376 {
377         struct o2hb_nego_msg msg;
378         int status, ret;
379 
380         msg.node_num = o2nm_this_node();
381 again:
382         ret = o2net_send_message(type, key, &msg, sizeof(msg),
383                         target, &status);
384 
385         if (ret == -EAGAIN || ret == -ENOMEM) {
386                 msleep(100);
387                 goto again;
388         }
389 
390         return ret;
391 }
392 
393 static void o2hb_nego_timeout(struct work_struct *work)
394 {
395         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
396         int master_node, i, ret;
397         struct o2hb_region *reg;
398 
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.
402          */
403         if (reg->hr_last_hb_status)
404                 return;
405 
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);
409 
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(&reg->hr_item), reg->hr_dev_name);
415                         set_bit(master_node, reg->hr_nego_node_bitmap);
416                 }
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
420                          * approve decision.
421                          */
422                         schedule_delayed_work(&reg->hr_nego_timeout_work,
423                                 msecs_to_jiffies(1000));
424 
425                         return;
426                 }
427 
428                 printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%s) is down.\n",
429                         config_item_name(&reg->hr_item), reg->hr_dev_name);
430                 /* approve negotiate timeout request. */
431                 o2hb_arm_timeout(reg);
432 
433                 i = -1;
434                 while ((i = find_next_bit(live_node_bitmap,
435                                 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
436                         if (i == master_node)
437                                 continue;
438 
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);
442                         if (ret)
443                                 mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
444                                         i, ret);
445                 }
446         } else {
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(&reg->hr_item),
450                         reg->hr_dev_name, master_node);
451                 ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
452                                 master_node);
453                 if (ret)
454                         mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
455                                 master_node, ret);
456         }
457 }
458 
459 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
460                                 void **ret_data)
461 {
462         struct o2hb_region *reg = data;
463         struct o2hb_nego_msg *nego_msg;
464 
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(&reg->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);
470         else
471                 mlog(ML_ERROR, "got nego timeout message from bad node.\n");
472 
473         return 0;
474 }
475 
476 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
477                                 void **ret_data)
478 {
479         struct o2hb_region *reg = data;
480 
481         printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%s).\n",
482                 config_item_name(&reg->hr_item), reg->hr_dev_name);
483         o2hb_arm_timeout(reg);
484         return 0;
485 }
486 
487 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
488 {
489         atomic_set(&wc->wc_num_reqs, 1);
490         init_completion(&wc->wc_io_complete);
491         wc->wc_error = 0;
492 }
493 
494 /* Used in error paths too */
495 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
496                                      unsigned int num)
497 {
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 */
500         while(num--) {
501                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
502                         BUG_ON(num > 0);
503                         complete(&wc->wc_io_complete);
504                 }
505         }
506 }
507 
508 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
509 {
510         o2hb_bio_wait_dec(wc, 1);
511         wait_for_completion(&wc->wc_io_complete);
512 }
513 
514 static void o2hb_bio_end_io(struct bio *bio)
515 {
516         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
517 
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);
521         }
522 
523         o2hb_bio_wait_dec(wc, 1);
524         bio_put(bio);
525 }
526 
527 /* Setup a Bio to cover I/O against num_slots slots starting at
528  * start_slot. */
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,
533                                       int op_flags)
534 {
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;
540         struct bio *bio;
541         struct page *page;
542 
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
546          * all together. */
547         bio = bio_alloc(GFP_ATOMIC, 16);
548         if (!bio) {
549                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
550                 bio = ERR_PTR(-ENOMEM);
551                 goto bail;
552         }
553 
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);
560 
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];
565 
566                 vec_len = min(PAGE_SIZE - vec_start,
567                               (max_slots-cs) * (PAGE_SIZE/spp) );
568 
569                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
570                      current_page, vec_len, vec_start);
571 
572                 len = bio_add_page(bio, page, vec_len, vec_start);
573                 if (len != vec_len) break;
574 
575                 cs += vec_len / (PAGE_SIZE/spp);
576                 vec_start = 0;
577         }
578 
579 bail:
580         *current_slot = cs;
581         return bio;
582 }
583 
584 static int o2hb_read_slots(struct o2hb_region *reg,
585                            unsigned int max_slots)
586 {
587         unsigned int current_slot=0;
588         int status;
589         struct o2hb_bio_wait_ctxt wc;
590         struct bio *bio;
591 
592         o2hb_bio_wait_init(&wc);
593 
594         while(current_slot < max_slots) {
595                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots,
596                                          REQ_OP_READ, 0);
597                 if (IS_ERR(bio)) {
598                         status = PTR_ERR(bio);
599                         mlog_errno(status);
600                         goto bail_and_wait;
601                 }
602 
603                 atomic_inc(&wc.wc_num_reqs);
604                 submit_bio(bio);
605         }
606 
607         status = 0;
608 
609 bail_and_wait:
610         o2hb_wait_on_io(&wc);
611         if (wc.wc_error && !status)
612                 status = wc.wc_error;
613 
614         return status;
615 }
616 
617 static int o2hb_issue_node_write(struct o2hb_region *reg,
618                                  struct o2hb_bio_wait_ctxt *write_wc)
619 {
620         int status;
621         unsigned int slot;
622         struct bio *bio;
623 
624         o2hb_bio_wait_init(write_wc);
625 
626         slot = o2nm_this_node();
627 
628         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1, REQ_OP_WRITE,
629                                  REQ_SYNC);
630         if (IS_ERR(bio)) {
631                 status = PTR_ERR(bio);
632                 mlog_errno(status);
633                 goto bail;
634         }
635 
636         atomic_inc(&write_wc->wc_num_reqs);
637         submit_bio(bio);
638 
639         status = 0;
640 bail:
641         return status;
642 }
643 
644 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
645                                      struct o2hb_disk_heartbeat_block *hb_block)
646 {
647         __le32 old_cksum;
648         u32 ret;
649 
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
652          * crc. */
653         old_cksum = hb_block->hb_cksum;
654         hb_block->hb_cksum = 0;
655 
656         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
657 
658         hb_block->hb_cksum = old_cksum;
659 
660         return ret;
661 }
662 
663 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
664 {
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));
670 }
671 
672 static int o2hb_verify_crc(struct o2hb_region *reg,
673                            struct o2hb_disk_heartbeat_block *hb_block)
674 {
675         u32 read, computed;
676 
677         read = le32_to_cpu(hb_block->hb_cksum);
678         computed = o2hb_compute_block_crc_le(reg, hb_block);
679 
680         return read == computed;
681 }
682 
683 /*
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.
689  */
690 static int o2hb_check_own_slot(struct o2hb_region *reg)
691 {
692         struct o2hb_disk_slot *slot;
693         struct o2hb_disk_heartbeat_block *hb_block;
694         char *errstr;
695 
696         slot = &reg->hr_slots[o2nm_this_node()];
697         /* Don't check on our 1st timestamp */
698         if (!slot->ds_last_time)
699                 return 0;
700 
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)
705                 return 1;
706 
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"
710 
711         if (hb_block->hb_node != slot->ds_node_num)
712                 errstr = ERRSTR1;
713         else if (le64_to_cpu(hb_block->hb_generation) !=
714                  slot->ds_last_generation)
715                 errstr = ERRSTR2;
716         else
717                 errstr = ERRSTR3;
718 
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));
725 
726         return 0;
727 }
728 
729 static inline void o2hb_prepare_block(struct o2hb_region *reg,
730                                       u64 generation)
731 {
732         int node_num;
733         u64 cputime;
734         struct o2hb_disk_slot *slot;
735         struct o2hb_disk_heartbeat_block *hb_block;
736 
737         node_num = o2nm_this_node();
738         slot = &reg->hr_slots[node_num];
739 
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();
744         if (!cputime)
745                 cputime = 1;
746 
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);
751 
752         /* This step must always happen last! */
753         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
754                                                                    hb_block));
755 
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));
759 }
760 
761 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
762                                 struct o2nm_node *node,
763                                 int idx)
764 {
765         struct o2hb_callback_func *f;
766 
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);
770         }
771 }
772 
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)
775 {
776         struct o2hb_callback *hbcall;
777         struct o2hb_node_event *event;
778 
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);
783 
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,
789                                    hn_item);
790                 list_del_init(&event->hn_item);
791                 spin_unlock(&o2hb_live_lock);
792 
793                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
794                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
795                      event->hn_node_num);
796 
797                 hbcall = hbcall_from_type(event->hn_event_type);
798 
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));
803 
804                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
805 
806                 spin_lock(&o2hb_live_lock);
807         }
808         spin_unlock(&o2hb_live_lock);
809 
810         up_write(&o2hb_callback_sem);
811 }
812 
813 static void o2hb_queue_node_event(struct o2hb_node_event *event,
814                                   enum o2hb_callback_type type,
815                                   struct o2nm_node *node,
816                                   int node_num)
817 {
818         assert_spin_locked(&o2hb_live_lock);
819 
820         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
821 
822         event->hn_event_type = type;
823         event->hn_node = node;
824         event->hn_node_num = node_num;
825 
826         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
827              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
828 
829         list_add_tail(&event->hn_item, &o2hb_node_events);
830 }
831 
832 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
833 {
834         struct o2hb_node_event event =
835                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
836         struct o2nm_node *node;
837         int queued = 0;
838 
839         node = o2nm_get_node_by_num(slot->ds_node_num);
840         if (!node)
841                 return;
842 
843         spin_lock(&o2hb_live_lock);
844         if (!list_empty(&slot->ds_live_item)) {
845                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
846                      slot->ds_node_num);
847 
848                 list_del_init(&slot->ds_live_item);
849 
850                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
851                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
852 
853                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
854                                               slot->ds_node_num);
855                         queued = 1;
856                 }
857         }
858         spin_unlock(&o2hb_live_lock);
859 
860         if (queued)
861                 o2hb_run_event_list(&event);
862 
863         o2nm_node_put(node);
864 }
865 
866 static void o2hb_set_quorum_device(struct o2hb_region *reg)
867 {
868         if (!o2hb_global_heartbeat_active())
869                 return;
870 
871         /* Prevent race with o2hb_heartbeat_group_drop_item() */
872         if (kthread_should_stop())
873                 return;
874 
875         /* Tag region as quorum only after thread reaches steady state */
876         if (atomic_read(&reg->hr_steady_iterations) != 0)
877                 return;
878 
879         spin_lock(&o2hb_live_lock);
880 
881         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
882                 goto unlock;
883 
884         /*
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.
888          */
889         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
890                    sizeof(o2hb_live_node_bitmap)))
891                 goto unlock;
892 
893         printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
894                config_item_name(&reg->hr_item), reg->hr_dev_name);
895 
896         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
897 
898         /*
899          * If global heartbeat active, unpin all regions if the
900          * region count > CUT_OFF
901          */
902         if (bitmap_weight(o2hb_quorum_region_bitmap,
903                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
904                 o2hb_region_unpin(NULL);
905 unlock:
906         spin_unlock(&o2hb_live_lock);
907 }
908 
909 static int o2hb_check_slot(struct o2hb_region *reg,
910                            struct o2hb_disk_slot *slot)
911 {
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;
917         u64 cputime;
918         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
919         unsigned int slot_dead_ms;
920         int tmp;
921         int queued = 0;
922 
923         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
924 
925         /*
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.
928          */
929         node = o2nm_get_node_by_num(slot->ds_node_num);
930         if (!node) {
931                 spin_lock(&o2hb_live_lock);
932                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
933                 spin_unlock(&o2hb_live_lock);
934                 if (!tmp)
935                         return 0;
936         }
937 
938         if (!o2hb_verify_crc(reg, hb_block)) {
939                 /* all paths from here will drop o2hb_live_lock for
940                  * us. */
941                 spin_lock(&o2hb_live_lock);
942 
943                 /* Don't print an error on the console in this case -
944                  * a freshly formatted heartbeat area will not have a
945                  * crc set on it. */
946                 if (list_empty(&slot->ds_live_item))
947                         goto out;
948 
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);
955 
956                 slot->ds_equal_samples++;
957                 goto fire_callbacks;
958         }
959 
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++;
965         else
966                 slot->ds_equal_samples++;
967         slot->ds_last_time = cputime;
968 
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)) {
975                 gen_changed = 1;
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));
981         }
982 
983         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
984 
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);
992 
993         spin_lock(&o2hb_live_lock);
994 
995 fire_callbacks:
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);
1002 
1003                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1004 
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);
1010 
1011                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
1012                                               slot->ds_node_num);
1013 
1014                         changed = 1;
1015                         queued = 1;
1016                 }
1017 
1018                 list_add_tail(&slot->ds_live_item,
1019                               &o2hb_live_slots[slot->ds_node_num]);
1020 
1021                 slot->ds_equal_samples = 0;
1022 
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,
1037                              dead_ms);
1038                 }
1039                 goto out;
1040         }
1041 
1042         /* if the list is dead, we're done.. */
1043         if (list_empty(&slot->ds_live_item))
1044                 goto out;
1045 
1046         /* live nodes only go dead after enough consequtive missed
1047          * samples..  reset the missed counter whenever we see
1048          * activity */
1049         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1050                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
1051                      slot->ds_node_num);
1052 
1053                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1054 
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);
1061 
1062                         /* node can be null */
1063                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1064                                               node, slot->ds_node_num);
1065 
1066                         changed = 1;
1067                         queued = 1;
1068                 }
1069 
1070                 /* We don't clear this because the node is still
1071                  * actually writing new blocks. */
1072                 if (!gen_changed)
1073                         slot->ds_changed_samples = 0;
1074                 goto out;
1075         }
1076         if (slot->ds_changed_samples) {
1077                 slot->ds_changed_samples = 0;
1078                 slot->ds_equal_samples = 0;
1079         }
1080 out:
1081         spin_unlock(&o2hb_live_lock);
1082 
1083         if (queued)
1084                 o2hb_run_event_list(&event);
1085 
1086         if (node)
1087                 o2nm_node_put(node);
1088         return changed;
1089 }
1090 
1091 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1092 {
1093         return find_last_bit(nodes, numbits);
1094 }
1095 
1096 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1097 {
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;
1103 
1104         ret = o2nm_configured_node_map(configured_nodes,
1105                                        sizeof(configured_nodes));
1106         if (ret) {
1107                 mlog_errno(ret);
1108                 goto bail;
1109         }
1110 
1111         /*
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.
1114          */
1115         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
1116         i = -1;
1117         while ((i = find_next_bit(live_node_bitmap,
1118                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1119                 set_bit(i, configured_nodes);
1120         }
1121 
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");
1125                 ret = -EINVAL;
1126                 goto bail;
1127         }
1128 
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
1132          * best-effort. */
1133         ret = o2hb_read_slots(reg, highest_node + 1);
1134         if (ret < 0) {
1135                 mlog_errno(ret);
1136                 goto bail;
1137         }
1138 
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
1141          * our slot. */
1142         own_slot_ok = o2hb_check_own_slot(reg);
1143 
1144         /* fill in the proper info for our next heartbeat */
1145         o2hb_prepare_block(reg, reg->hr_generation);
1146 
1147         ret = o2hb_issue_node_write(reg, &write_wc);
1148         if (ret < 0) {
1149                 mlog_errno(ret);
1150                 goto bail;
1151         }
1152 
1153         i = -1;
1154         while((i = find_next_bit(configured_nodes,
1155                                  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1156                 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1157         }
1158 
1159         /*
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.
1163          */
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
1168                  * disk */
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;
1172                 goto bail;
1173         }
1174 
1175         /* Skip disarming the timeout if own slot has stale/bad data */
1176         if (own_slot_ok) {
1177                 o2hb_set_quorum_device(reg);
1178                 o2hb_arm_timeout(reg);
1179                 reg->hr_last_timeout_start = jiffies;
1180         }
1181 
1182 bail:
1183         /* let the person who launched us know when things are steady */
1184         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1185                 if (!ret && own_slot_ok && !membership_change) {
1186                         if (atomic_dec_and_test(&reg->hr_steady_iterations))
1187                                 wake_up(&o2hb_steady_queue);
1188                 }
1189         }
1190 
1191         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1192                 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1193                         printk(KERN_NOTICE "o2hb: Unable to stabilize "
1194                                "heartbeart on region %s (%s)\n",
1195                                config_item_name(&reg->hr_item),
1196                                reg->hr_dev_name);
1197                         atomic_set(&reg->hr_steady_iterations, 0);
1198                         reg->hr_aborted_start = 1;
1199                         wake_up(&o2hb_steady_queue);
1200                         ret = -EIO;
1201                 }
1202         }
1203 
1204         return ret;
1205 }
1206 
1207 /*
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
1210  * thread.
1211  */
1212 static int o2hb_thread(void *data)
1213 {
1214         int i, ret;
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;
1219 
1220         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1221 
1222         set_user_nice(current, MIN_NICE);
1223 
1224         /* Pin node */
1225         ret = o2nm_depend_this_node();
1226         if (ret) {
1227                 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1228                 reg->hr_node_deleted = 1;
1229                 wake_up(&o2hb_steady_queue);
1230                 return 0;
1231         }
1232 
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();
1241 
1242                 ret = o2hb_do_disk_heartbeat(reg);
1243                 reg->hr_last_hb_status = ret;
1244 
1245                 after_hb = ktime_get_real();
1246 
1247                 elapsed_msec = (unsigned int)
1248                                 ktime_ms_delta(after_hb, before_hb);
1249 
1250                 mlog(ML_HEARTBEAT,
1251                      "start = %lld, end = %lld, msec = %u, ret = %d\n",
1252                      before_hb, after_hb, elapsed_msec, ret);
1253 
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);
1259                 }
1260         }
1261 
1262         o2hb_disarm_timeout(reg);
1263 
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(&reg->hr_slots[i]);
1267 
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.
1272          */
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);
1276                 if (ret == 0)
1277                         o2hb_wait_on_io(&write_wc);
1278                 else
1279                         mlog_errno(ret);
1280         }
1281 
1282         /* Unpin node */
1283         o2nm_undepend_this_node();
1284 
1285         mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1286 
1287         return 0;
1288 }
1289 
1290 #ifdef CONFIG_DEBUG_FS
1291 static int o2hb_debug_open(struct inode *inode, struct file *file)
1292 {
1293         struct o2hb_debug_buf *db = inode->i_private;
1294         struct o2hb_region *reg;
1295         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1296         unsigned long lts;
1297         char *buf = NULL;
1298         int i = -1;
1299         int out = 0;
1300 
1301         /* max_nodes should be the largest bitmap we pass here */
1302         BUG_ON(sizeof(map) < db->db_size);
1303 
1304         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1305         if (!buf)
1306                 goto bail;
1307 
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);
1316                 break;
1317 
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);
1323                 break;
1324 
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);
1329                 goto done;
1330 
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 */
1335                 if (lts)
1336                         lts = jiffies_to_msecs(jiffies - lts);
1337                 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1338                 goto done;
1339 
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);
1344                 goto done;
1345 
1346         default:
1347                 goto done;
1348         }
1349 
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");
1353 
1354 done:
1355         i_size_write(inode, out);
1356 
1357         file->private_data = buf;
1358 
1359         return 0;
1360 bail:
1361         return -ENOMEM;
1362 }
1363 
1364 static int o2hb_debug_release(struct inode *inode, struct file *file)
1365 {
1366         kfree(file->private_data);
1367         return 0;
1368 }
1369 
1370 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1371                                  size_t nbytes, loff_t *ppos)
1372 {
1373         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1374                                        i_size_read(file->f_mapping->host));
1375 }
1376 #else
1377 static int o2hb_debug_open(struct inode *inode, struct file *file)
1378 {
1379         return 0;
1380 }
1381 static int o2hb_debug_release(struct inode *inode, struct file *file)
1382 {
1383         return 0;
1384 }
1385 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1386                                size_t nbytes, loff_t *ppos)
1387 {
1388         return 0;
1389 }
1390 #endif  /* CONFIG_DEBUG_FS */
1391 
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,
1397 };
1398 
1399 void o2hb_exit(void)
1400 {
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);
1410 }
1411 
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)
1415 {
1416         *db = kmalloc(db_len, GFP_KERNEL);
1417         if (!*db)
1418                 return NULL;
1419 
1420         (*db)->db_type = type;
1421         (*db)->db_size = size;
1422         (*db)->db_len = len;
1423         (*db)->db_data = data;
1424 
1425         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1426                                    &o2hb_debug_fops);
1427 }
1428 
1429 static int o2hb_debug_init(void)
1430 {
1431         int ret = -ENOMEM;
1432 
1433         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1434         if (!o2hb_debug_dir) {
1435                 mlog_errno(ret);
1436                 goto bail;
1437         }
1438 
1439         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1440                                                  o2hb_debug_dir,
1441                                                  &o2hb_db_livenodes,
1442                                                  sizeof(*o2hb_db_livenodes),
1443                                                  O2HB_DB_TYPE_LIVENODES,
1444                                                  sizeof(o2hb_live_node_bitmap),
1445                                                  O2NM_MAX_NODES,
1446                                                  o2hb_live_node_bitmap);
1447         if (!o2hb_debug_livenodes) {
1448                 mlog_errno(ret);
1449                 goto bail;
1450         }
1451 
1452         o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1453                                                    o2hb_debug_dir,
1454                                                    &o2hb_db_liveregions,
1455                                                    sizeof(*o2hb_db_liveregions),
1456                                                    O2HB_DB_TYPE_LIVEREGIONS,
1457                                                    sizeof(o2hb_live_region_bitmap),
1458                                                    O2NM_MAX_REGIONS,
1459                                                    o2hb_live_region_bitmap);
1460         if (!o2hb_debug_liveregions) {
1461                 mlog_errno(ret);
1462                 goto bail;
1463         }
1464 
1465         o2hb_debug_quorumregions =
1466                         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1467                                           o2hb_debug_dir,
1468                                           &o2hb_db_quorumregions,
1469                                           sizeof(*o2hb_db_quorumregions),
1470                                           O2HB_DB_TYPE_QUORUMREGIONS,
1471                                           sizeof(o2hb_quorum_region_bitmap),
1472                                           O2NM_MAX_REGIONS,
1473                                           o2hb_quorum_region_bitmap);
1474         if (!o2hb_debug_quorumregions) {
1475                 mlog_errno(ret);
1476                 goto bail;
1477         }
1478 
1479         o2hb_debug_failedregions =
1480                         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1481                                           o2hb_debug_dir,
1482                                           &o2hb_db_failedregions,
1483                                           sizeof(*o2hb_db_failedregions),
1484                                           O2HB_DB_TYPE_FAILEDREGIONS,
1485                                           sizeof(o2hb_failed_region_bitmap),
1486                                           O2NM_MAX_REGIONS,
1487                                           o2hb_failed_region_bitmap);
1488         if (!o2hb_debug_failedregions) {
1489                 mlog_errno(ret);
1490                 goto bail;
1491         }
1492 
1493         ret = 0;
1494 bail:
1495         if (ret)
1496                 o2hb_exit();
1497 
1498         return ret;
1499 }
1500 
1501 int o2hb_init(void)
1502 {
1503         int i;
1504 
1505         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1506                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1507 
1508         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1509                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1510 
1511         INIT_LIST_HEAD(&o2hb_node_events);
1512 
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));
1518 
1519         o2hb_dependent_users = 0;
1520 
1521         return o2hb_debug_init();
1522 }
1523 
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,
1526                                              unsigned bytes)
1527 {
1528         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1529 
1530         memcpy(map, &o2hb_live_node_bitmap, bytes);
1531 }
1532 
1533 /*
1534  * get a map of all nodes that are heartbeating in any regions
1535  */
1536 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1537 {
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);
1545 }
1546 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1547 
1548 /*
1549  * heartbeat configfs bits.  The heartbeat set is a default set under
1550  * the cluster set in nodemanager.c.
1551  */
1552 
1553 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1554 {
1555         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1556 }
1557 
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)
1562 {
1563         int i;
1564         struct page *page;
1565         struct o2hb_region *reg = to_o2hb_region(item);
1566 
1567         mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1568 
1569         kfree(reg->hr_tmp_block);
1570 
1571         if (reg->hr_slot_data) {
1572                 for (i = 0; i < reg->hr_num_pages; i++) {
1573                         page = reg->hr_slot_data[i];
1574                         if (page)
1575                                 __free_page(page);
1576                 }
1577                 kfree(reg->hr_slot_data);
1578         }
1579 
1580         if (reg->hr_bdev)
1581                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1582 
1583         kfree(reg->hr_slots);
1584 
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);
1594 
1595         spin_lock(&o2hb_live_lock);
1596         list_del(&reg->hr_all_item);
1597         spin_unlock(&o2hb_live_lock);
1598 
1599         o2net_unregister_handler_list(&reg->hr_handler_list);
1600         kfree(reg);
1601 }
1602 
1603 static int o2hb_read_block_input(struct o2hb_region *reg,
1604                                  const char *page,
1605                                  unsigned long *ret_bytes,
1606                                  unsigned int *ret_bits)
1607 {
1608         unsigned long bytes;
1609         char *p = (char *)page;
1610 
1611         bytes = simple_strtoul(p, &p, 0);
1612         if (!p || (*p && (*p != '\n')))
1613                 return -EINVAL;
1614 
1615         /* Heartbeat and fs min / max block sizes are the same. */
1616         if (bytes > 4096 || bytes < 512)
1617                 return -ERANGE;
1618         if (hweight16(bytes) != 1)
1619                 return -EINVAL;
1620 
1621         if (ret_bytes)
1622                 *ret_bytes = bytes;
1623         if (ret_bits)
1624                 *ret_bits = ffs(bytes) - 1;
1625 
1626         return 0;
1627 }
1628 
1629 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1630                                             char *page)
1631 {
1632         return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1633 }
1634 
1635 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1636                                              const char *page,
1637                                              size_t count)
1638 {
1639         struct o2hb_region *reg = to_o2hb_region(item);
1640         int status;
1641         unsigned long block_bytes;
1642         unsigned int block_bits;
1643 
1644         if (reg->hr_bdev)
1645                 return -EINVAL;
1646 
1647         status = o2hb_read_block_input(reg, page, &block_bytes,
1648                                        &block_bits);
1649         if (status)
1650                 return status;
1651 
1652         reg->hr_block_bytes = (unsigned int)block_bytes;
1653         reg->hr_block_bits = block_bits;
1654 
1655         return count;
1656 }
1657 
1658 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1659                                             char *page)
1660 {
1661         return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1662 }
1663 
1664 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1665                                              const char *page,
1666                                              size_t count)
1667 {
1668         struct o2hb_region *reg = to_o2hb_region(item);
1669         unsigned long long tmp;
1670         char *p = (char *)page;
1671 
1672         if (reg->hr_bdev)
1673                 return -EINVAL;
1674 
1675         tmp = simple_strtoull(p, &p, 0);
1676         if (!p || (*p && (*p != '\n')))
1677                 return -EINVAL;
1678 
1679         reg->hr_start_block = tmp;
1680 
1681         return count;
1682 }
1683 
1684 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1685 {
1686         return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1687 }
1688 
1689 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1690                                         const char *page,
1691                                         size_t count)
1692 {
1693         struct o2hb_region *reg = to_o2hb_region(item);
1694         unsigned long tmp;
1695         char *p = (char *)page;
1696 
1697         if (reg->hr_bdev)
1698                 return -EINVAL;
1699 
1700         tmp = simple_strtoul(p, &p, 0);
1701         if (!p || (*p && (*p != '\n')))
1702                 return -EINVAL;
1703 
1704         if (tmp > O2NM_MAX_NODES || tmp == 0)
1705                 return -ERANGE;
1706 
1707         reg->hr_blocks = (unsigned int)tmp;
1708 
1709         return count;
1710 }
1711 
1712 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1713 {
1714         unsigned int ret = 0;
1715 
1716         if (to_o2hb_region(item)->hr_bdev)
1717                 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1718 
1719         return ret;
1720 }
1721 
1722 static void o2hb_init_region_params(struct o2hb_region *reg)
1723 {
1724         reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1725         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1726 
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);
1733 }
1734 
1735 static int o2hb_map_slot_data(struct o2hb_region *reg)
1736 {
1737         int i, j;
1738         unsigned int last_slot;
1739         unsigned int spp = reg->hr_slots_per_page;
1740         struct page *page;
1741         char *raw;
1742         struct o2hb_disk_slot *slot;
1743 
1744         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1745         if (reg->hr_tmp_block == NULL)
1746                 return -ENOMEM;
1747 
1748         reg->hr_slots = kcalloc(reg->hr_blocks,
1749                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1750         if (reg->hr_slots == NULL)
1751                 return -ENOMEM;
1752 
1753         for(i = 0; i < reg->hr_blocks; i++) {
1754                 slot = &reg->hr_slots[i];
1755                 slot->ds_node_num = i;
1756                 INIT_LIST_HEAD(&slot->ds_live_item);
1757                 slot->ds_raw_block = NULL;
1758         }
1759 
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);
1764 
1765         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1766                                     GFP_KERNEL);
1767         if (!reg->hr_slot_data)
1768                 return -ENOMEM;
1769 
1770         for(i = 0; i < reg->hr_num_pages; i++) {
1771                 page = alloc_page(GFP_KERNEL);
1772                 if (!page)
1773                         return -ENOMEM;
1774 
1775                 reg->hr_slot_data[i] = page;
1776 
1777                 last_slot = i * spp;
1778                 raw = page_address(page);
1779                 for (j = 0;
1780                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1781                      j++) {
1782                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1783 
1784                         slot = &reg->hr_slots[j + last_slot];
1785                         slot->ds_raw_block =
1786                                 (struct o2hb_disk_heartbeat_block *) raw;
1787 
1788                         raw += reg->hr_block_bytes;
1789                 }
1790         }
1791 
1792         return 0;
1793 }
1794 
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
1797  * there. */
1798 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1799 {
1800         int ret, i;
1801         struct o2hb_disk_slot *slot;
1802         struct o2hb_disk_heartbeat_block *hb_block;
1803 
1804         ret = o2hb_read_slots(reg, reg->hr_blocks);
1805         if (ret)
1806                 goto out;
1807 
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 = &reg->hr_slots[i];
1814                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1815 
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);
1820         }
1821 
1822 out:
1823         return ret;
1824 }
1825 
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,
1828                                      const char *page,
1829                                      size_t count)
1830 {
1831         struct o2hb_region *reg = to_o2hb_region(item);
1832         struct task_struct *hb_task;
1833         long fd;
1834         int sectsize;
1835         char *p = (char *)page;
1836         struct fd f;
1837         struct inode *inode;
1838         ssize_t ret = -EINVAL;
1839         int live_threshold;
1840 
1841         if (reg->hr_bdev)
1842                 goto out;
1843 
1844         /* We can't heartbeat without having had our node number
1845          * configured yet. */
1846         if (o2nm_this_node() == O2NM_MAX_NODES)
1847                 goto out;
1848 
1849         fd = simple_strtol(p, &p, 0);
1850         if (!p || (*p && (*p != '\n')))
1851                 goto out;
1852 
1853         if (fd < 0 || fd >= INT_MAX)
1854                 goto out;
1855 
1856         f = fdget(fd);
1857         if (f.file == NULL)
1858                 goto out;
1859 
1860         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1861             reg->hr_block_bytes == 0)
1862                 goto out2;
1863 
1864         inode = igrab(f.file->f_mapping->host);
1865         if (inode == NULL)
1866                 goto out2;
1867 
1868         if (!S_ISBLK(inode->i_mode))
1869                 goto out3;
1870 
1871         reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1872         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1873         if (ret) {
1874                 reg->hr_bdev = NULL;
1875                 goto out3;
1876         }
1877         inode = NULL;
1878 
1879         bdevname(reg->hr_bdev, reg->hr_dev_name);
1880 
1881         sectsize = bdev_logical_block_size(reg->hr_bdev);
1882         if (sectsize != reg->hr_block_bytes) {
1883                 mlog(ML_ERROR,
1884                      "blocksize %u incorrect for device, expected %d",
1885                      reg->hr_block_bytes, sectsize);
1886                 ret = -EINVAL;
1887                 goto out3;
1888         }
1889 
1890         o2hb_init_region_params(reg);
1891 
1892         /* Generation of zero is invalid */
1893         do {
1894                 get_random_bytes(&reg->hr_generation,
1895                                  sizeof(reg->hr_generation));
1896         } while (reg->hr_generation == 0);
1897 
1898         ret = o2hb_map_slot_data(reg);
1899         if (ret) {
1900                 mlog_errno(ret);
1901                 goto out3;
1902         }
1903 
1904         ret = o2hb_populate_slot_data(reg);
1905         if (ret) {
1906                 mlog_errno(ret);
1907                 goto out3;
1908         }
1909 
1910         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1911         INIT_DELAYED_WORK(&reg->hr_nego_timeout_work, o2hb_nego_timeout);
1912 
1913         /*
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
1919          * first region.
1920          */
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);
1927         }
1928         ++live_threshold;
1929         atomic_set(&reg->hr_steady_iterations, live_threshold);
1930         /* unsteady_iterations is triple the steady_iterations */
1931         atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
1932 
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);
1937                 mlog_errno(ret);
1938                 goto out3;
1939         }
1940 
1941         spin_lock(&o2hb_live_lock);
1942         reg->hr_task = hb_task;
1943         spin_unlock(&o2hb_live_lock);
1944 
1945         ret = wait_event_interruptible(o2hb_steady_queue,
1946                                 atomic_read(&reg->hr_steady_iterations) == 0 ||
1947                                 reg->hr_node_deleted);
1948         if (ret) {
1949                 atomic_set(&reg->hr_steady_iterations, 0);
1950                 reg->hr_aborted_start = 1;
1951         }
1952 
1953         if (reg->hr_aborted_start) {
1954                 ret = -EIO;
1955                 goto out3;
1956         }
1957 
1958         if (reg->hr_node_deleted) {
1959                 ret = -EINVAL;
1960                 goto out3;
1961         }
1962 
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);
1969 
1970         if (hb_task)
1971                 ret = count;
1972         else
1973                 ret = -EIO;
1974 
1975         if (hb_task && o2hb_global_heartbeat_active())
1976                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1977                        config_item_name(&reg->hr_item), reg->hr_dev_name);
1978 
1979 out3:
1980         iput(inode);
1981 out2:
1982         fdput(f);
1983 out:
1984         if (ret < 0) {
1985                 if (reg->hr_bdev) {
1986                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1987                         reg->hr_bdev = NULL;
1988                 }
1989         }
1990         return ret;
1991 }
1992 
1993 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1994 {
1995         struct o2hb_region *reg = to_o2hb_region(item);
1996         pid_t pid = 0;
1997 
1998         spin_lock(&o2hb_live_lock);
1999         if (reg->hr_task)
2000                 pid = task_pid_nr(reg->hr_task);
2001         spin_unlock(&o2hb_live_lock);
2002 
2003         if (!pid)
2004                 return 0;
2005 
2006         return sprintf(page, "%u\n", pid);
2007 }
2008 
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);
2014 
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,
2021         NULL,
2022 };
2023 
2024 static struct configfs_item_operations o2hb_region_item_ops = {
2025         .release                = o2hb_region_release,
2026 };
2027 
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,
2032 };
2033 
2034 /* heartbeat set */
2035 
2036 struct o2hb_heartbeat_group {
2037         struct config_group hs_group;
2038         /* some stuff? */
2039 };
2040 
2041 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2042 {
2043         return group ?
2044                 container_of(group, struct o2hb_heartbeat_group, hs_group)
2045                 : NULL;
2046 }
2047 
2048 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2049 {
2050         int ret = -ENOMEM;
2051 
2052         reg->hr_debug_dir =
2053                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
2054         if (!reg->hr_debug_dir) {
2055                 mlog_errno(ret);
2056                 goto bail;
2057         }
2058 
2059         reg->hr_debug_livenodes =
2060                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2061                                           reg->hr_debug_dir,
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) {
2068                 mlog_errno(ret);
2069                 goto bail;
2070         }
2071 
2072         reg->hr_debug_regnum =
2073                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2074                                           reg->hr_debug_dir,
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) {
2080                 mlog_errno(ret);
2081                 goto bail;
2082         }
2083 
2084         reg->hr_debug_elapsed_time =
2085                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2086                                           reg->hr_debug_dir,
2087                                           &(reg->hr_db_elapsed_time),
2088                                           sizeof(*(reg->hr_db_elapsed_time)),
2089                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2090                                           0, 0, reg);
2091         if (!reg->hr_debug_elapsed_time) {
2092                 mlog_errno(ret);
2093                 goto bail;
2094         }
2095 
2096         reg->hr_debug_pinned =
2097                         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2098                                           reg->hr_debug_dir,
2099                                           &(reg->hr_db_pinned),
2100                                           sizeof(*(reg->hr_db_pinned)),
2101                                           O2HB_DB_TYPE_REGION_PINNED,
2102                                           0, 0, reg);
2103         if (!reg->hr_debug_pinned) {
2104                 mlog_errno(ret);
2105                 goto bail;
2106         }
2107 
2108         ret = 0;
2109 bail:
2110         return ret;
2111 }
2112 
2113 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2114                                                           const char *name)
2115 {
2116         struct o2hb_region *reg = NULL;
2117         int ret;
2118 
2119         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2120         if (reg == NULL)
2121                 return ERR_PTR(-ENOMEM);
2122 
2123         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2124                 ret = -ENAMETOOLONG;
2125                 goto free;
2126         }
2127 
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,
2132                                                          O2NM_MAX_REGIONS);
2133                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2134                         spin_unlock(&o2hb_live_lock);
2135                         ret = -EFBIG;
2136                         goto free;
2137                 }
2138                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2139         }
2140         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2141         spin_unlock(&o2hb_live_lock);
2142 
2143         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2144 
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.
2148          */
2149         reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2150                 name, strlen(name));
2151         INIT_LIST_HEAD(&reg->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, &reg->hr_handler_list);
2156         if (ret)
2157                 goto free;
2158 
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, &reg->hr_handler_list);
2163         if (ret)
2164                 goto unregister_handler;
2165 
2166         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2167         if (ret) {
2168                 config_item_put(&reg->hr_item);
2169                 goto unregister_handler;
2170         }
2171 
2172         return &reg->hr_item;
2173 
2174 unregister_handler:
2175         o2net_unregister_handler_list(&reg->hr_handler_list);
2176 free:
2177         kfree(reg);
2178         return ERR_PTR(ret);
2179 }
2180 
2181 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2182                                            struct config_item *item)
2183 {
2184         struct task_struct *hb_task;
2185         struct o2hb_region *reg = to_o2hb_region(item);
2186         int quorum_region = 0;
2187 
2188         /* stop the thread when the user removes the region dir */
2189         spin_lock(&o2hb_live_lock);
2190         hb_task = reg->hr_task;
2191         reg->hr_task = NULL;
2192         reg->hr_item_dropped = 1;
2193         spin_unlock(&o2hb_live_lock);
2194 
2195         if (hb_task)
2196                 kthread_stop(hb_task);
2197 
2198         if (o2hb_global_heartbeat_active()) {
2199                 spin_lock(&o2hb_live_lock);
2200                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2201                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2202                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2203                         quorum_region = 1;
2204                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2205                 spin_unlock(&o2hb_live_lock);
2206                 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2207                        ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2208                         "stopped" : "start aborted"), config_item_name(item),
2209                        reg->hr_dev_name);
2210         }
2211 
2212         /*
2213          * If we're racing a dev_write(), we need to wake them.  They will
2214          * check reg->hr_task
2215          */
2216         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2217                 reg->hr_aborted_start = 1;
2218                 atomic_set(&reg->hr_steady_iterations, 0);
2219                 wake_up(&o2hb_steady_queue);
2220         }
2221 
2222         config_item_put(item);
2223 
2224         if (!o2hb_global_heartbeat_active() || !quorum_region)
2225                 return;
2226 
2227         /*
2228          * If global heartbeat active and there are dependent users,
2229          * pin all regions if quorum region count <= CUT_OFF
2230          */
2231         spin_lock(&o2hb_live_lock);
2232 
2233         if (!o2hb_dependent_users)
2234                 goto unlock;
2235 
2236         if (bitmap_weight(o2hb_quorum_region_bitmap,
2237                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2238                 o2hb_region_pin(NULL);
2239 
2240 unlock:
2241         spin_unlock(&o2hb_live_lock);
2242 }
2243 
2244 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2245                 char *page)
2246 {
2247         return sprintf(page, "%u\n", o2hb_dead_threshold);
2248 }
2249 
2250 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2251                 const char *page, size_t count)
2252 {
2253         unsigned long tmp;
2254         char *p = (char *)page;
2255 
2256         tmp = simple_strtoul(p, &p, 10);
2257         if (!p || (*p && (*p != '\n')))
2258                 return -EINVAL;
2259 
2260         /* this will validate ranges for us. */
2261         o2hb_dead_threshold_set((unsigned int) tmp);
2262 
2263         return count;
2264 }
2265 
2266 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2267                 char *page)
2268 {
2269         return sprintf(page, "%s\n",
2270                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2271 }
2272 
2273 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2274                 const char *page, size_t count)
2275 {
2276         unsigned int i;
2277         int ret;
2278         size_t len;
2279 
2280         len = (page[count - 1] == '\n') ? count - 1 : count;
2281         if (!len)
2282                 return -EINVAL;
2283 
2284         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2285                 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2286                         continue;
2287 
2288                 ret = o2hb_global_heartbeat_mode_set(i);
2289                 if (!ret)
2290                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2291                                o2hb_heartbeat_mode_desc[i]);
2292                 return count;
2293         }
2294 
2295         return -EINVAL;
2296 
2297 }
2298 
2299 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2300 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2301 
2302 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2303         &o2hb_heartbeat_group_attr_dead_threshold,
2304         &o2hb_heartbeat_group_attr_mode,
2305         NULL,
2306 };
2307 
2308 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2309         .make_item      = o2hb_heartbeat_group_make_item,
2310         .drop_item      = o2hb_heartbeat_group_drop_item,
2311 };
2312 
2313 static struct config_item_type o2hb_heartbeat_group_type = {
2314         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2315         .ct_attrs       = o2hb_heartbeat_group_attrs,
2316         .ct_owner       = THIS_MODULE,
2317 };
2318 
2319 /* this is just here to avoid touching group in heartbeat.h which the
2320  * entire damn world #includes */
2321 struct config_group *o2hb_alloc_hb_set(void)
2322 {
2323         struct o2hb_heartbeat_group *hs = NULL;
2324         struct config_group *ret = NULL;
2325 
2326         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2327         if (hs == NULL)
2328                 goto out;
2329 
2330         config_group_init_type_name(&hs->hs_group, "heartbeat",
2331                                     &o2hb_heartbeat_group_type);
2332 
2333         ret = &hs->hs_group;
2334 out:
2335         if (ret == NULL)
2336                 kfree(hs);
2337         return ret;
2338 }
2339 
2340 void o2hb_free_hb_set(struct config_group *group)
2341 {
2342         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2343         kfree(hs);
2344 }
2345 
2346 /* hb callback registration and issuing */
2347 
2348 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2349 {
2350         if (type == O2HB_NUM_CB)
2351                 return ERR_PTR(-EINVAL);
2352 
2353         return &o2hb_callbacks[type];
2354 }
2355 
2356 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2357                          enum o2hb_callback_type type,
2358                          o2hb_cb_func *func,
2359                          void *data,
2360                          int priority)
2361 {
2362         INIT_LIST_HEAD(&hc->hc_item);
2363         hc->hc_func = func;
2364         hc->hc_data = data;
2365         hc->hc_priority = priority;
2366         hc->hc_type = type;
2367         hc->hc_magic = O2HB_CB_MAGIC;
2368 }
2369 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2370 
2371 /*
2372  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2373  * In global heartbeat mode, region_uuid passed is NULL.
2374  *
2375  * In local, we only pin the matching region. In global we pin all the active
2376  * regions.
2377  */
2378 static int o2hb_region_pin(const char *region_uuid)
2379 {
2380         int ret = 0, found = 0;
2381         struct o2hb_region *reg;
2382         char *uuid;
2383 
2384         assert_spin_locked(&o2hb_live_lock);
2385 
2386         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2387                 if (reg->hr_item_dropped)
2388                         continue;
2389 
2390                 uuid = config_item_name(&reg->hr_item);
2391 
2392                 /* local heartbeat */
2393                 if (region_uuid) {
2394                         if (strcmp(region_uuid, uuid))
2395                                 continue;
2396                         found = 1;
2397                 }
2398 
2399                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2400                         goto skip_pin;
2401 
2402                 /* Ignore ENOENT only for local hb (userdlm domain) */
2403                 ret = o2nm_depend_item(&reg->hr_item);
2404                 if (!ret) {
2405                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2406                         reg->hr_item_pinned = 1;
2407                 } else {
2408                         if (ret == -ENOENT && found)
2409                                 ret = 0;
2410                         else {
2411                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2412                                      uuid, ret);
2413                                 break;
2414                         }
2415                 }
2416 skip_pin:
2417                 if (found)
2418                         break;
2419         }
2420 
2421         return ret;
2422 }
2423 
2424 /*
2425  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2426  * In global heartbeat mode, region_uuid passed is NULL.
2427  *
2428  * In local, we only unpin the matching region. In global we unpin all the
2429  * active regions.
2430  */
2431 static void o2hb_region_unpin(const char *region_uuid)
2432 {
2433         struct o2hb_region *reg;
2434         char *uuid;
2435         int found = 0;
2436 
2437         assert_spin_locked(&o2hb_live_lock);
2438 
2439         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2440                 if (reg->hr_item_dropped)
2441                         continue;
2442 
2443                 uuid = config_item_name(&reg->hr_item);
2444                 if (region_uuid) {
2445                         if (strcmp(region_uuid, uuid))
2446                                 continue;
2447                         found = 1;
2448                 }
2449 
2450                 if (reg->hr_item_pinned) {
2451                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2452                         o2nm_undepend_item(&reg->hr_item);
2453                         reg->hr_item_pinned = 0;
2454                 }
2455                 if (found)
2456                         break;
2457         }
2458 }
2459 
2460 static int o2hb_region_inc_user(const char *region_uuid)
2461 {
2462         int ret = 0;
2463 
2464         spin_lock(&o2hb_live_lock);
2465 
2466         /* local heartbeat */
2467         if (!o2hb_global_heartbeat_active()) {
2468             ret = o2hb_region_pin(region_uuid);
2469             goto unlock;
2470         }
2471 
2472         /*
2473          * if global heartbeat active and this is the first dependent user,
2474          * pin all regions if quorum region count <= CUT_OFF
2475          */
2476         o2hb_dependent_users++;
2477         if (o2hb_dependent_users > 1)
2478                 goto unlock;
2479 
2480         if (bitmap_weight(o2hb_quorum_region_bitmap,
2481                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2482                 ret = o2hb_region_pin(NULL);
2483 
2484 unlock:
2485         spin_unlock(&o2hb_live_lock);
2486         return ret;
2487 }
2488 
2489 void o2hb_region_dec_user(const char *region_uuid)
2490 {
2491         spin_lock(&o2hb_live_lock);
2492 
2493         /* local heartbeat */
2494         if (!o2hb_global_heartbeat_active()) {
2495             o2hb_region_unpin(region_uuid);
2496             goto unlock;
2497         }
2498 
2499         /*
2500          * if global heartbeat active and there are no dependent users,
2501          * unpin all quorum regions
2502          */
2503         o2hb_dependent_users--;
2504         if (!o2hb_dependent_users)
2505                 o2hb_region_unpin(NULL);
2506 
2507 unlock:
2508         spin_unlock(&o2hb_live_lock);
2509 }
2510 
2511 int o2hb_register_callback(const char *region_uuid,
2512                            struct o2hb_callback_func *hc)
2513 {
2514         struct o2hb_callback_func *f;
2515         struct o2hb_callback *hbcall;
2516         int ret;
2517 
2518         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2519         BUG_ON(!list_empty(&hc->hc_item));
2520 
2521         hbcall = hbcall_from_type(hc->hc_type);
2522         if (IS_ERR(hbcall)) {
2523                 ret = PTR_ERR(hbcall);
2524                 goto out;
2525         }
2526 
2527         if (region_uuid) {
2528                 ret = o2hb_region_inc_user(region_uuid);
2529                 if (ret) {
2530                         mlog_errno(ret);
2531                         goto out;
2532                 }
2533         }
2534 
2535         down_write(&o2hb_callback_sem);
2536 
2537         list_for_each_entry(f, &hbcall->list, hc_item) {
2538                 if (hc->hc_priority < f->hc_priority) {
2539                         list_add_tail(&hc->hc_item, &f->hc_item);
2540                         break;
2541                 }
2542         }
2543         if (list_empty(&hc->hc_item))
2544                 list_add_tail(&hc->hc_item, &hbcall->list);
2545 
2546         up_write(&o2hb_callback_sem);
2547         ret = 0;
2548 out:
2549         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2550              ret, __builtin_return_address(0), hc);
2551         return ret;
2552 }
2553 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2554 
2555 void o2hb_unregister_callback(const char *region_uuid,
2556                               struct o2hb_callback_func *hc)
2557 {
2558         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2559 
2560         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2561              __builtin_return_address(0), hc);
2562 
2563         /* XXX Can this happen _with_ a region reference? */
2564         if (list_empty(&hc->hc_item))
2565                 return;
2566 
2567         if (region_uuid)
2568                 o2hb_region_dec_user(region_uuid);
2569 
2570         down_write(&o2hb_callback_sem);
2571 
2572         list_del_init(&hc->hc_item);
2573 
2574         up_write(&o2hb_callback_sem);
2575 }
2576 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2577 
2578 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2579 {
2580         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2581 
2582         spin_lock(&o2hb_live_lock);
2583         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2584         spin_unlock(&o2hb_live_lock);
2585         if (!test_bit(node_num, testing_map)) {
2586                 mlog(ML_HEARTBEAT,
2587                      "node (%u) does not have heartbeating enabled.\n",
2588                      node_num);
2589                 return 0;
2590         }
2591 
2592         return 1;
2593 }
2594 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2595 
2596 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2597 {
2598         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2599 
2600         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2601         if (!test_bit(node_num, testing_map)) {
2602                 mlog(ML_HEARTBEAT,
2603                      "node (%u) does not have heartbeating enabled.\n",
2604                      node_num);
2605                 return 0;
2606         }
2607 
2608         return 1;
2609 }
2610 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2611 
2612 /*
2613  * this is just a hack until we get the plumbing which flips file systems
2614  * read only and drops the hb ref instead of killing the node dead.
2615  */
2616 void o2hb_stop_all_regions(void)
2617 {
2618         struct o2hb_region *reg;
2619 
2620         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2621 
2622         spin_lock(&o2hb_live_lock);
2623 
2624         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2625                 reg->hr_unclean_stop = 1;
2626 
2627         spin_unlock(&o2hb_live_lock);
2628 }
2629 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2630 
2631 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2632 {
2633         struct o2hb_region *reg;
2634         int numregs = 0;
2635         char *p;
2636 
2637         spin_lock(&o2hb_live_lock);
2638 
2639         p = region_uuids;
2640         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2641                 if (reg->hr_item_dropped)
2642                         continue;
2643 
2644                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2645                 if (numregs < max_regions) {
2646                         memcpy(p, config_item_name(&reg->hr_item),
2647                                O2HB_MAX_REGION_NAME_LEN);
2648                         p += O2HB_MAX_REGION_NAME_LEN;
2649                 }
2650                 numregs++;
2651         }
2652 
2653         spin_unlock(&o2hb_live_lock);
2654 
2655         return numregs;
2656 }
2657 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2658 
2659 int o2hb_global_heartbeat_active(void)
2660 {
2661         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2662 }
2663 EXPORT_SYMBOL(o2hb_global_heartbeat_active);
2664 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp