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

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

Version: ~ [ linux-4.17-rc6 ] ~ [ linux-4.16.10 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.42 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.101 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.132 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.51 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.109 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.56 ] ~ [ 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-3.2.101 ] ~ [ 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) {
574                         mlog(ML_ERROR, "Adding page[%d] to bio failed, "
575                              "page %p, len %d, vec_len %u, vec_start %u, "
576                              "bi_sector %llu\n", current_page, page, len,
577                              vec_len, vec_start,
578                              (unsigned long long)bio->bi_iter.bi_sector);
579                         bio_put(bio);
580                         bio = ERR_PTR(-EIO);
581                         return bio;
582                 }
583 
584                 cs += vec_len / (PAGE_SIZE/spp);
585                 vec_start = 0;
586         }
587 
588 bail:
589         *current_slot = cs;
590         return bio;
591 }
592 
593 static int o2hb_read_slots(struct o2hb_region *reg,
594                            unsigned int max_slots)
595 {
596         unsigned int current_slot=0;
597         int status;
598         struct o2hb_bio_wait_ctxt wc;
599         struct bio *bio;
600 
601         o2hb_bio_wait_init(&wc);
602 
603         while(current_slot < max_slots) {
604                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots,
605                                          REQ_OP_READ, 0);
606                 if (IS_ERR(bio)) {
607                         status = PTR_ERR(bio);
608                         mlog_errno(status);
609                         goto bail_and_wait;
610                 }
611 
612                 atomic_inc(&wc.wc_num_reqs);
613                 submit_bio(bio);
614         }
615 
616         status = 0;
617 
618 bail_and_wait:
619         o2hb_wait_on_io(&wc);
620         if (wc.wc_error && !status)
621                 status = wc.wc_error;
622 
623         return status;
624 }
625 
626 static int o2hb_issue_node_write(struct o2hb_region *reg,
627                                  struct o2hb_bio_wait_ctxt *write_wc)
628 {
629         int status;
630         unsigned int slot;
631         struct bio *bio;
632 
633         o2hb_bio_wait_init(write_wc);
634 
635         slot = o2nm_this_node();
636 
637         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1, REQ_OP_WRITE,
638                                  REQ_SYNC);
639         if (IS_ERR(bio)) {
640                 status = PTR_ERR(bio);
641                 mlog_errno(status);
642                 goto bail;
643         }
644 
645         atomic_inc(&write_wc->wc_num_reqs);
646         submit_bio(bio);
647 
648         status = 0;
649 bail:
650         return status;
651 }
652 
653 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
654                                      struct o2hb_disk_heartbeat_block *hb_block)
655 {
656         __le32 old_cksum;
657         u32 ret;
658 
659         /* We want to compute the block crc with a 0 value in the
660          * hb_cksum field. Save it off here and replace after the
661          * crc. */
662         old_cksum = hb_block->hb_cksum;
663         hb_block->hb_cksum = 0;
664 
665         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
666 
667         hb_block->hb_cksum = old_cksum;
668 
669         return ret;
670 }
671 
672 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
673 {
674         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
675              "cksum = 0x%x, generation 0x%llx\n",
676              (long long)le64_to_cpu(hb_block->hb_seq),
677              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
678              (long long)le64_to_cpu(hb_block->hb_generation));
679 }
680 
681 static int o2hb_verify_crc(struct o2hb_region *reg,
682                            struct o2hb_disk_heartbeat_block *hb_block)
683 {
684         u32 read, computed;
685 
686         read = le32_to_cpu(hb_block->hb_cksum);
687         computed = o2hb_compute_block_crc_le(reg, hb_block);
688 
689         return read == computed;
690 }
691 
692 /*
693  * Compare the slot data with what we wrote in the last iteration.
694  * If the match fails, print an appropriate error message. This is to
695  * detect errors like... another node hearting on the same slot,
696  * flaky device that is losing writes, etc.
697  * Returns 1 if check succeeds, 0 otherwise.
698  */
699 static int o2hb_check_own_slot(struct o2hb_region *reg)
700 {
701         struct o2hb_disk_slot *slot;
702         struct o2hb_disk_heartbeat_block *hb_block;
703         char *errstr;
704 
705         slot = &reg->hr_slots[o2nm_this_node()];
706         /* Don't check on our 1st timestamp */
707         if (!slot->ds_last_time)
708                 return 0;
709 
710         hb_block = slot->ds_raw_block;
711         if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
712             le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
713             hb_block->hb_node == slot->ds_node_num)
714                 return 1;
715 
716 #define ERRSTR1         "Another node is heartbeating on device"
717 #define ERRSTR2         "Heartbeat generation mismatch on device"
718 #define ERRSTR3         "Heartbeat sequence mismatch on device"
719 
720         if (hb_block->hb_node != slot->ds_node_num)
721                 errstr = ERRSTR1;
722         else if (le64_to_cpu(hb_block->hb_generation) !=
723                  slot->ds_last_generation)
724                 errstr = ERRSTR2;
725         else
726                 errstr = ERRSTR3;
727 
728         mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
729              "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
730              slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
731              (unsigned long long)slot->ds_last_time, hb_block->hb_node,
732              (unsigned long long)le64_to_cpu(hb_block->hb_generation),
733              (unsigned long long)le64_to_cpu(hb_block->hb_seq));
734 
735         return 0;
736 }
737 
738 static inline void o2hb_prepare_block(struct o2hb_region *reg,
739                                       u64 generation)
740 {
741         int node_num;
742         u64 cputime;
743         struct o2hb_disk_slot *slot;
744         struct o2hb_disk_heartbeat_block *hb_block;
745 
746         node_num = o2nm_this_node();
747         slot = &reg->hr_slots[node_num];
748 
749         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
750         memset(hb_block, 0, reg->hr_block_bytes);
751         /* TODO: time stuff */
752         cputime = ktime_get_real_seconds();
753         if (!cputime)
754                 cputime = 1;
755 
756         hb_block->hb_seq = cpu_to_le64(cputime);
757         hb_block->hb_node = node_num;
758         hb_block->hb_generation = cpu_to_le64(generation);
759         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
760 
761         /* This step must always happen last! */
762         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
763                                                                    hb_block));
764 
765         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
766              (long long)generation,
767              le32_to_cpu(hb_block->hb_cksum));
768 }
769 
770 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
771                                 struct o2nm_node *node,
772                                 int idx)
773 {
774         struct o2hb_callback_func *f;
775 
776         list_for_each_entry(f, &hbcall->list, hc_item) {
777                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
778                 (f->hc_func)(node, idx, f->hc_data);
779         }
780 }
781 
782 /* Will run the list in order until we process the passed event */
783 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
784 {
785         struct o2hb_callback *hbcall;
786         struct o2hb_node_event *event;
787 
788         /* Holding callback sem assures we don't alter the callback
789          * lists when doing this, and serializes ourselves with other
790          * processes wanting callbacks. */
791         down_write(&o2hb_callback_sem);
792 
793         spin_lock(&o2hb_live_lock);
794         while (!list_empty(&o2hb_node_events)
795                && !list_empty(&queued_event->hn_item)) {
796                 event = list_entry(o2hb_node_events.next,
797                                    struct o2hb_node_event,
798                                    hn_item);
799                 list_del_init(&event->hn_item);
800                 spin_unlock(&o2hb_live_lock);
801 
802                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
803                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
804                      event->hn_node_num);
805 
806                 hbcall = hbcall_from_type(event->hn_event_type);
807 
808                 /* We should *never* have gotten on to the list with a
809                  * bad type... This isn't something that we should try
810                  * to recover from. */
811                 BUG_ON(IS_ERR(hbcall));
812 
813                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
814 
815                 spin_lock(&o2hb_live_lock);
816         }
817         spin_unlock(&o2hb_live_lock);
818 
819         up_write(&o2hb_callback_sem);
820 }
821 
822 static void o2hb_queue_node_event(struct o2hb_node_event *event,
823                                   enum o2hb_callback_type type,
824                                   struct o2nm_node *node,
825                                   int node_num)
826 {
827         assert_spin_locked(&o2hb_live_lock);
828 
829         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
830 
831         event->hn_event_type = type;
832         event->hn_node = node;
833         event->hn_node_num = node_num;
834 
835         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
836              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
837 
838         list_add_tail(&event->hn_item, &o2hb_node_events);
839 }
840 
841 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
842 {
843         struct o2hb_node_event event =
844                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
845         struct o2nm_node *node;
846         int queued = 0;
847 
848         node = o2nm_get_node_by_num(slot->ds_node_num);
849         if (!node)
850                 return;
851 
852         spin_lock(&o2hb_live_lock);
853         if (!list_empty(&slot->ds_live_item)) {
854                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
855                      slot->ds_node_num);
856 
857                 list_del_init(&slot->ds_live_item);
858 
859                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
860                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
861 
862                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
863                                               slot->ds_node_num);
864                         queued = 1;
865                 }
866         }
867         spin_unlock(&o2hb_live_lock);
868 
869         if (queued)
870                 o2hb_run_event_list(&event);
871 
872         o2nm_node_put(node);
873 }
874 
875 static void o2hb_set_quorum_device(struct o2hb_region *reg)
876 {
877         if (!o2hb_global_heartbeat_active())
878                 return;
879 
880         /* Prevent race with o2hb_heartbeat_group_drop_item() */
881         if (kthread_should_stop())
882                 return;
883 
884         /* Tag region as quorum only after thread reaches steady state */
885         if (atomic_read(&reg->hr_steady_iterations) != 0)
886                 return;
887 
888         spin_lock(&o2hb_live_lock);
889 
890         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
891                 goto unlock;
892 
893         /*
894          * A region can be added to the quorum only when it sees all
895          * live nodes heartbeat on it. In other words, the region has been
896          * added to all nodes.
897          */
898         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
899                    sizeof(o2hb_live_node_bitmap)))
900                 goto unlock;
901 
902         printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
903                config_item_name(&reg->hr_item), reg->hr_dev_name);
904 
905         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
906 
907         /*
908          * If global heartbeat active, unpin all regions if the
909          * region count > CUT_OFF
910          */
911         if (bitmap_weight(o2hb_quorum_region_bitmap,
912                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
913                 o2hb_region_unpin(NULL);
914 unlock:
915         spin_unlock(&o2hb_live_lock);
916 }
917 
918 static int o2hb_check_slot(struct o2hb_region *reg,
919                            struct o2hb_disk_slot *slot)
920 {
921         int changed = 0, gen_changed = 0;
922         struct o2hb_node_event event =
923                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
924         struct o2nm_node *node;
925         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
926         u64 cputime;
927         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
928         unsigned int slot_dead_ms;
929         int tmp;
930         int queued = 0;
931 
932         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
933 
934         /*
935          * If a node is no longer configured but is still in the livemap, we
936          * may need to clear that bit from the livemap.
937          */
938         node = o2nm_get_node_by_num(slot->ds_node_num);
939         if (!node) {
940                 spin_lock(&o2hb_live_lock);
941                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
942                 spin_unlock(&o2hb_live_lock);
943                 if (!tmp)
944                         return 0;
945         }
946 
947         if (!o2hb_verify_crc(reg, hb_block)) {
948                 /* all paths from here will drop o2hb_live_lock for
949                  * us. */
950                 spin_lock(&o2hb_live_lock);
951 
952                 /* Don't print an error on the console in this case -
953                  * a freshly formatted heartbeat area will not have a
954                  * crc set on it. */
955                 if (list_empty(&slot->ds_live_item))
956                         goto out;
957 
958                 /* The node is live but pushed out a bad crc. We
959                  * consider it a transient miss but don't populate any
960                  * other values as they may be junk. */
961                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
962                      slot->ds_node_num, reg->hr_dev_name);
963                 o2hb_dump_slot(hb_block);
964 
965                 slot->ds_equal_samples++;
966                 goto fire_callbacks;
967         }
968 
969         /* we don't care if these wrap.. the state transitions below
970          * clear at the right places */
971         cputime = le64_to_cpu(hb_block->hb_seq);
972         if (slot->ds_last_time != cputime)
973                 slot->ds_changed_samples++;
974         else
975                 slot->ds_equal_samples++;
976         slot->ds_last_time = cputime;
977 
978         /* The node changed heartbeat generations. We assume this to
979          * mean it dropped off but came back before we timed out. We
980          * want to consider it down for the time being but don't want
981          * to lose any changed_samples state we might build up to
982          * considering it live again. */
983         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
984                 gen_changed = 1;
985                 slot->ds_equal_samples = 0;
986                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
987                      "to 0x%llx)\n", slot->ds_node_num,
988                      (long long)slot->ds_last_generation,
989                      (long long)le64_to_cpu(hb_block->hb_generation));
990         }
991 
992         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
993 
994         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
995              "seq %llu last %llu changed %u equal %u\n",
996              slot->ds_node_num, (long long)slot->ds_last_generation,
997              le32_to_cpu(hb_block->hb_cksum),
998              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
999              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
1000              slot->ds_equal_samples);
1001 
1002         spin_lock(&o2hb_live_lock);
1003 
1004 fire_callbacks:
1005         /* dead nodes only come to life after some number of
1006          * changes at any time during their dead time */
1007         if (list_empty(&slot->ds_live_item) &&
1008             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
1009                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
1010                      slot->ds_node_num, (long long)slot->ds_last_generation);
1011 
1012                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1013 
1014                 /* first on the list generates a callback */
1015                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1016                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
1017                              "bitmap\n", slot->ds_node_num);
1018                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1019 
1020                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
1021                                               slot->ds_node_num);
1022 
1023                         changed = 1;
1024                         queued = 1;
1025                 }
1026 
1027                 list_add_tail(&slot->ds_live_item,
1028                               &o2hb_live_slots[slot->ds_node_num]);
1029 
1030                 slot->ds_equal_samples = 0;
1031 
1032                 /* We want to be sure that all nodes agree on the
1033                  * number of milliseconds before a node will be
1034                  * considered dead. The self-fencing timeout is
1035                  * computed from this value, and a discrepancy might
1036                  * result in heartbeat calling a node dead when it
1037                  * hasn't self-fenced yet. */
1038                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
1039                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
1040                         /* TODO: Perhaps we can fail the region here. */
1041                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
1042                              "of %u ms, but our count is %u ms.\n"
1043                              "Please double check your configuration values "
1044                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1045                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
1046                              dead_ms);
1047                 }
1048                 goto out;
1049         }
1050 
1051         /* if the list is dead, we're done.. */
1052         if (list_empty(&slot->ds_live_item))
1053                 goto out;
1054 
1055         /* live nodes only go dead after enough consequtive missed
1056          * samples..  reset the missed counter whenever we see
1057          * activity */
1058         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1059                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
1060                      slot->ds_node_num);
1061 
1062                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1063 
1064                 /* last off the live_slot generates a callback */
1065                 list_del_init(&slot->ds_live_item);
1066                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1067                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
1068                              "nodes bitmap\n", slot->ds_node_num);
1069                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1070 
1071                         /* node can be null */
1072                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1073                                               node, slot->ds_node_num);
1074 
1075                         changed = 1;
1076                         queued = 1;
1077                 }
1078 
1079                 /* We don't clear this because the node is still
1080                  * actually writing new blocks. */
1081                 if (!gen_changed)
1082                         slot->ds_changed_samples = 0;
1083                 goto out;
1084         }
1085         if (slot->ds_changed_samples) {
1086                 slot->ds_changed_samples = 0;
1087                 slot->ds_equal_samples = 0;
1088         }
1089 out:
1090         spin_unlock(&o2hb_live_lock);
1091 
1092         if (queued)
1093                 o2hb_run_event_list(&event);
1094 
1095         if (node)
1096                 o2nm_node_put(node);
1097         return changed;
1098 }
1099 
1100 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1101 {
1102         return find_last_bit(nodes, numbits);
1103 }
1104 
1105 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1106 {
1107         int i, ret, highest_node;
1108         int membership_change = 0, own_slot_ok = 0;
1109         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
1110         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1111         struct o2hb_bio_wait_ctxt write_wc;
1112 
1113         ret = o2nm_configured_node_map(configured_nodes,
1114                                        sizeof(configured_nodes));
1115         if (ret) {
1116                 mlog_errno(ret);
1117                 goto bail;
1118         }
1119 
1120         /*
1121          * If a node is not configured but is in the livemap, we still need
1122          * to read the slot so as to be able to remove it from the livemap.
1123          */
1124         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
1125         i = -1;
1126         while ((i = find_next_bit(live_node_bitmap,
1127                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1128                 set_bit(i, configured_nodes);
1129         }
1130 
1131         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1132         if (highest_node >= O2NM_MAX_NODES) {
1133                 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1134                 ret = -EINVAL;
1135                 goto bail;
1136         }
1137 
1138         /* No sense in reading the slots of nodes that don't exist
1139          * yet. Of course, if the node definitions have holes in them
1140          * then we're reading an empty slot anyway... Consider this
1141          * best-effort. */
1142         ret = o2hb_read_slots(reg, highest_node + 1);
1143         if (ret < 0) {
1144                 mlog_errno(ret);
1145                 goto bail;
1146         }
1147 
1148         /* With an up to date view of the slots, we can check that no
1149          * other node has been improperly configured to heartbeat in
1150          * our slot. */
1151         own_slot_ok = o2hb_check_own_slot(reg);
1152 
1153         /* fill in the proper info for our next heartbeat */
1154         o2hb_prepare_block(reg, reg->hr_generation);
1155 
1156         ret = o2hb_issue_node_write(reg, &write_wc);
1157         if (ret < 0) {
1158                 mlog_errno(ret);
1159                 goto bail;
1160         }
1161 
1162         i = -1;
1163         while((i = find_next_bit(configured_nodes,
1164                                  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1165                 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1166         }
1167 
1168         /*
1169          * We have to be sure we've advertised ourselves on disk
1170          * before we can go to steady state.  This ensures that
1171          * people we find in our steady state have seen us.
1172          */
1173         o2hb_wait_on_io(&write_wc);
1174         if (write_wc.wc_error) {
1175                 /* Do not re-arm the write timeout on I/O error - we
1176                  * can't be sure that the new block ever made it to
1177                  * disk */
1178                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1179                      write_wc.wc_error, reg->hr_dev_name);
1180                 ret = write_wc.wc_error;
1181                 goto bail;
1182         }
1183 
1184         /* Skip disarming the timeout if own slot has stale/bad data */
1185         if (own_slot_ok) {
1186                 o2hb_set_quorum_device(reg);
1187                 o2hb_arm_timeout(reg);
1188                 reg->hr_last_timeout_start = jiffies;
1189         }
1190 
1191 bail:
1192         /* let the person who launched us know when things are steady */
1193         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1194                 if (!ret && own_slot_ok && !membership_change) {
1195                         if (atomic_dec_and_test(&reg->hr_steady_iterations))
1196                                 wake_up(&o2hb_steady_queue);
1197                 }
1198         }
1199 
1200         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1201                 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1202                         printk(KERN_NOTICE "o2hb: Unable to stabilize "
1203                                "heartbeart on region %s (%s)\n",
1204                                config_item_name(&reg->hr_item),
1205                                reg->hr_dev_name);
1206                         atomic_set(&reg->hr_steady_iterations, 0);
1207                         reg->hr_aborted_start = 1;
1208                         wake_up(&o2hb_steady_queue);
1209                         ret = -EIO;
1210                 }
1211         }
1212 
1213         return ret;
1214 }
1215 
1216 /*
1217  * we ride the region ref that the region dir holds.  before the region
1218  * dir is removed and drops it ref it will wait to tear down this
1219  * thread.
1220  */
1221 static int o2hb_thread(void *data)
1222 {
1223         int i, ret;
1224         struct o2hb_region *reg = data;
1225         struct o2hb_bio_wait_ctxt write_wc;
1226         ktime_t before_hb, after_hb;
1227         unsigned int elapsed_msec;
1228 
1229         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1230 
1231         set_user_nice(current, MIN_NICE);
1232 
1233         /* Pin node */
1234         ret = o2nm_depend_this_node();
1235         if (ret) {
1236                 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1237                 reg->hr_node_deleted = 1;
1238                 wake_up(&o2hb_steady_queue);
1239                 return 0;
1240         }
1241 
1242         while (!kthread_should_stop() &&
1243                !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1244                 /* We track the time spent inside
1245                  * o2hb_do_disk_heartbeat so that we avoid more than
1246                  * hr_timeout_ms between disk writes. On busy systems
1247                  * this should result in a heartbeat which is less
1248                  * likely to time itself out. */
1249                 before_hb = ktime_get_real();
1250 
1251                 ret = o2hb_do_disk_heartbeat(reg);
1252                 reg->hr_last_hb_status = ret;
1253 
1254                 after_hb = ktime_get_real();
1255 
1256                 elapsed_msec = (unsigned int)
1257                                 ktime_ms_delta(after_hb, before_hb);
1258 
1259                 mlog(ML_HEARTBEAT,
1260                      "start = %lld, end = %lld, msec = %u, ret = %d\n",
1261                      before_hb, after_hb, elapsed_msec, ret);
1262 
1263                 if (!kthread_should_stop() &&
1264                     elapsed_msec < reg->hr_timeout_ms) {
1265                         /* the kthread api has blocked signals for us so no
1266                          * need to record the return value. */
1267                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1268                 }
1269         }
1270 
1271         o2hb_disarm_timeout(reg);
1272 
1273         /* unclean stop is only used in very bad situation */
1274         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1275                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1276 
1277         /* Explicit down notification - avoid forcing the other nodes
1278          * to timeout on this region when we could just as easily
1279          * write a clear generation - thus indicating to them that
1280          * this node has left this region.
1281          */
1282         if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1283                 o2hb_prepare_block(reg, 0);
1284                 ret = o2hb_issue_node_write(reg, &write_wc);
1285                 if (ret == 0)
1286                         o2hb_wait_on_io(&write_wc);
1287                 else
1288                         mlog_errno(ret);
1289         }
1290 
1291         /* Unpin node */
1292         o2nm_undepend_this_node();
1293 
1294         mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1295 
1296         return 0;
1297 }
1298 
1299 #ifdef CONFIG_DEBUG_FS
1300 static int o2hb_debug_open(struct inode *inode, struct file *file)
1301 {
1302         struct o2hb_debug_buf *db = inode->i_private;
1303         struct o2hb_region *reg;
1304         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1305         unsigned long lts;
1306         char *buf = NULL;
1307         int i = -1;
1308         int out = 0;
1309 
1310         /* max_nodes should be the largest bitmap we pass here */
1311         BUG_ON(sizeof(map) < db->db_size);
1312 
1313         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1314         if (!buf)
1315                 goto bail;
1316 
1317         switch (db->db_type) {
1318         case O2HB_DB_TYPE_LIVENODES:
1319         case O2HB_DB_TYPE_LIVEREGIONS:
1320         case O2HB_DB_TYPE_QUORUMREGIONS:
1321         case O2HB_DB_TYPE_FAILEDREGIONS:
1322                 spin_lock(&o2hb_live_lock);
1323                 memcpy(map, db->db_data, db->db_size);
1324                 spin_unlock(&o2hb_live_lock);
1325                 break;
1326 
1327         case O2HB_DB_TYPE_REGION_LIVENODES:
1328                 spin_lock(&o2hb_live_lock);
1329                 reg = (struct o2hb_region *)db->db_data;
1330                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1331                 spin_unlock(&o2hb_live_lock);
1332                 break;
1333 
1334         case O2HB_DB_TYPE_REGION_NUMBER:
1335                 reg = (struct o2hb_region *)db->db_data;
1336                 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1337                                 reg->hr_region_num);
1338                 goto done;
1339 
1340         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1341                 reg = (struct o2hb_region *)db->db_data;
1342                 lts = reg->hr_last_timeout_start;
1343                 /* If 0, it has never been set before */
1344                 if (lts)
1345                         lts = jiffies_to_msecs(jiffies - lts);
1346                 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1347                 goto done;
1348 
1349         case O2HB_DB_TYPE_REGION_PINNED:
1350                 reg = (struct o2hb_region *)db->db_data;
1351                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1352                                 !!reg->hr_item_pinned);
1353                 goto done;
1354 
1355         default:
1356                 goto done;
1357         }
1358 
1359         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1360                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1361         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1362 
1363 done:
1364         i_size_write(inode, out);
1365 
1366         file->private_data = buf;
1367 
1368         return 0;
1369 bail:
1370         return -ENOMEM;
1371 }
1372 
1373 static int o2hb_debug_release(struct inode *inode, struct file *file)
1374 {
1375         kfree(file->private_data);
1376         return 0;
1377 }
1378 
1379 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1380                                  size_t nbytes, loff_t *ppos)
1381 {
1382         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1383                                        i_size_read(file->f_mapping->host));
1384 }
1385 #else
1386 static int o2hb_debug_open(struct inode *inode, struct file *file)
1387 {
1388         return 0;
1389 }
1390 static int o2hb_debug_release(struct inode *inode, struct file *file)
1391 {
1392         return 0;
1393 }
1394 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1395                                size_t nbytes, loff_t *ppos)
1396 {
1397         return 0;
1398 }
1399 #endif  /* CONFIG_DEBUG_FS */
1400 
1401 static const struct file_operations o2hb_debug_fops = {
1402         .open =         o2hb_debug_open,
1403         .release =      o2hb_debug_release,
1404         .read =         o2hb_debug_read,
1405         .llseek =       generic_file_llseek,
1406 };
1407 
1408 void o2hb_exit(void)
1409 {
1410         debugfs_remove(o2hb_debug_failedregions);
1411         debugfs_remove(o2hb_debug_quorumregions);
1412         debugfs_remove(o2hb_debug_liveregions);
1413         debugfs_remove(o2hb_debug_livenodes);
1414         debugfs_remove(o2hb_debug_dir);
1415         kfree(o2hb_db_livenodes);
1416         kfree(o2hb_db_liveregions);
1417         kfree(o2hb_db_quorumregions);
1418         kfree(o2hb_db_failedregions);
1419 }
1420 
1421 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1422                                         struct o2hb_debug_buf **db, int db_len,
1423                                         int type, int size, int len, void *data)
1424 {
1425         *db = kmalloc(db_len, GFP_KERNEL);
1426         if (!*db)
1427                 return NULL;
1428 
1429         (*db)->db_type = type;
1430         (*db)->db_size = size;
1431         (*db)->db_len = len;
1432         (*db)->db_data = data;
1433 
1434         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1435                                    &o2hb_debug_fops);
1436 }
1437 
1438 static int o2hb_debug_init(void)
1439 {
1440         int ret = -ENOMEM;
1441 
1442         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1443         if (!o2hb_debug_dir) {
1444                 mlog_errno(ret);
1445                 goto bail;
1446         }
1447 
1448         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1449                                                  o2hb_debug_dir,
1450                                                  &o2hb_db_livenodes,
1451                                                  sizeof(*o2hb_db_livenodes),
1452                                                  O2HB_DB_TYPE_LIVENODES,
1453                                                  sizeof(o2hb_live_node_bitmap),
1454                                                  O2NM_MAX_NODES,
1455                                                  o2hb_live_node_bitmap);
1456         if (!o2hb_debug_livenodes) {
1457                 mlog_errno(ret);
1458                 goto bail;
1459         }
1460 
1461         o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1462                                                    o2hb_debug_dir,
1463                                                    &o2hb_db_liveregions,
1464                                                    sizeof(*o2hb_db_liveregions),
1465                                                    O2HB_DB_TYPE_LIVEREGIONS,
1466                                                    sizeof(o2hb_live_region_bitmap),
1467                                                    O2NM_MAX_REGIONS,
1468                                                    o2hb_live_region_bitmap);
1469         if (!o2hb_debug_liveregions) {
1470                 mlog_errno(ret);
1471                 goto bail;
1472         }
1473 
1474         o2hb_debug_quorumregions =
1475                         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1476                                           o2hb_debug_dir,
1477                                           &o2hb_db_quorumregions,
1478                                           sizeof(*o2hb_db_quorumregions),
1479                                           O2HB_DB_TYPE_QUORUMREGIONS,
1480                                           sizeof(o2hb_quorum_region_bitmap),
1481                                           O2NM_MAX_REGIONS,
1482                                           o2hb_quorum_region_bitmap);
1483         if (!o2hb_debug_quorumregions) {
1484                 mlog_errno(ret);
1485                 goto bail;
1486         }
1487 
1488         o2hb_debug_failedregions =
1489                         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1490                                           o2hb_debug_dir,
1491                                           &o2hb_db_failedregions,
1492                                           sizeof(*o2hb_db_failedregions),
1493                                           O2HB_DB_TYPE_FAILEDREGIONS,
1494                                           sizeof(o2hb_failed_region_bitmap),
1495                                           O2NM_MAX_REGIONS,
1496                                           o2hb_failed_region_bitmap);
1497         if (!o2hb_debug_failedregions) {
1498                 mlog_errno(ret);
1499                 goto bail;
1500         }
1501 
1502         ret = 0;
1503 bail:
1504         if (ret)
1505                 o2hb_exit();
1506 
1507         return ret;
1508 }
1509 
1510 int o2hb_init(void)
1511 {
1512         int i;
1513 
1514         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1515                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1516 
1517         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1518                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1519 
1520         INIT_LIST_HEAD(&o2hb_node_events);
1521 
1522         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1523         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1524         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1525         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1526         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1527 
1528         o2hb_dependent_users = 0;
1529 
1530         return o2hb_debug_init();
1531 }
1532 
1533 /* if we're already in a callback then we're already serialized by the sem */
1534 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1535                                              unsigned bytes)
1536 {
1537         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1538 
1539         memcpy(map, &o2hb_live_node_bitmap, bytes);
1540 }
1541 
1542 /*
1543  * get a map of all nodes that are heartbeating in any regions
1544  */
1545 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1546 {
1547         /* callers want to serialize this map and callbacks so that they
1548          * can trust that they don't miss nodes coming to the party */
1549         down_read(&o2hb_callback_sem);
1550         spin_lock(&o2hb_live_lock);
1551         o2hb_fill_node_map_from_callback(map, bytes);
1552         spin_unlock(&o2hb_live_lock);
1553         up_read(&o2hb_callback_sem);
1554 }
1555 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1556 
1557 /*
1558  * heartbeat configfs bits.  The heartbeat set is a default set under
1559  * the cluster set in nodemanager.c.
1560  */
1561 
1562 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1563 {
1564         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1565 }
1566 
1567 /* drop_item only drops its ref after killing the thread, nothing should
1568  * be using the region anymore.  this has to clean up any state that
1569  * attributes might have built up. */
1570 static void o2hb_region_release(struct config_item *item)
1571 {
1572         int i;
1573         struct page *page;
1574         struct o2hb_region *reg = to_o2hb_region(item);
1575 
1576         mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1577 
1578         kfree(reg->hr_tmp_block);
1579 
1580         if (reg->hr_slot_data) {
1581                 for (i = 0; i < reg->hr_num_pages; i++) {
1582                         page = reg->hr_slot_data[i];
1583                         if (page)
1584                                 __free_page(page);
1585                 }
1586                 kfree(reg->hr_slot_data);
1587         }
1588 
1589         if (reg->hr_bdev)
1590                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1591 
1592         kfree(reg->hr_slots);
1593 
1594         debugfs_remove(reg->hr_debug_livenodes);
1595         debugfs_remove(reg->hr_debug_regnum);
1596         debugfs_remove(reg->hr_debug_elapsed_time);
1597         debugfs_remove(reg->hr_debug_pinned);
1598         debugfs_remove(reg->hr_debug_dir);
1599         kfree(reg->hr_db_livenodes);
1600         kfree(reg->hr_db_regnum);
1601         kfree(reg->hr_db_elapsed_time);
1602         kfree(reg->hr_db_pinned);
1603 
1604         spin_lock(&o2hb_live_lock);
1605         list_del(&reg->hr_all_item);
1606         spin_unlock(&o2hb_live_lock);
1607 
1608         o2net_unregister_handler_list(&reg->hr_handler_list);
1609         kfree(reg);
1610 }
1611 
1612 static int o2hb_read_block_input(struct o2hb_region *reg,
1613                                  const char *page,
1614                                  unsigned long *ret_bytes,
1615                                  unsigned int *ret_bits)
1616 {
1617         unsigned long bytes;
1618         char *p = (char *)page;
1619 
1620         bytes = simple_strtoul(p, &p, 0);
1621         if (!p || (*p && (*p != '\n')))
1622                 return -EINVAL;
1623 
1624         /* Heartbeat and fs min / max block sizes are the same. */
1625         if (bytes > 4096 || bytes < 512)
1626                 return -ERANGE;
1627         if (hweight16(bytes) != 1)
1628                 return -EINVAL;
1629 
1630         if (ret_bytes)
1631                 *ret_bytes = bytes;
1632         if (ret_bits)
1633                 *ret_bits = ffs(bytes) - 1;
1634 
1635         return 0;
1636 }
1637 
1638 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1639                                             char *page)
1640 {
1641         return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1642 }
1643 
1644 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1645                                              const char *page,
1646                                              size_t count)
1647 {
1648         struct o2hb_region *reg = to_o2hb_region(item);
1649         int status;
1650         unsigned long block_bytes;
1651         unsigned int block_bits;
1652 
1653         if (reg->hr_bdev)
1654                 return -EINVAL;
1655 
1656         status = o2hb_read_block_input(reg, page, &block_bytes,
1657                                        &block_bits);
1658         if (status)
1659                 return status;
1660 
1661         reg->hr_block_bytes = (unsigned int)block_bytes;
1662         reg->hr_block_bits = block_bits;
1663 
1664         return count;
1665 }
1666 
1667 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1668                                             char *page)
1669 {
1670         return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1671 }
1672 
1673 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1674                                              const char *page,
1675                                              size_t count)
1676 {
1677         struct o2hb_region *reg = to_o2hb_region(item);
1678         unsigned long long tmp;
1679         char *p = (char *)page;
1680 
1681         if (reg->hr_bdev)
1682                 return -EINVAL;
1683 
1684         tmp = simple_strtoull(p, &p, 0);
1685         if (!p || (*p && (*p != '\n')))
1686                 return -EINVAL;
1687 
1688         reg->hr_start_block = tmp;
1689 
1690         return count;
1691 }
1692 
1693 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1694 {
1695         return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1696 }
1697 
1698 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1699                                         const char *page,
1700                                         size_t count)
1701 {
1702         struct o2hb_region *reg = to_o2hb_region(item);
1703         unsigned long tmp;
1704         char *p = (char *)page;
1705 
1706         if (reg->hr_bdev)
1707                 return -EINVAL;
1708 
1709         tmp = simple_strtoul(p, &p, 0);
1710         if (!p || (*p && (*p != '\n')))
1711                 return -EINVAL;
1712 
1713         if (tmp > O2NM_MAX_NODES || tmp == 0)
1714                 return -ERANGE;
1715 
1716         reg->hr_blocks = (unsigned int)tmp;
1717 
1718         return count;
1719 }
1720 
1721 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1722 {
1723         unsigned int ret = 0;
1724 
1725         if (to_o2hb_region(item)->hr_bdev)
1726                 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1727 
1728         return ret;
1729 }
1730 
1731 static void o2hb_init_region_params(struct o2hb_region *reg)
1732 {
1733         reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1734         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1735 
1736         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1737              reg->hr_start_block, reg->hr_blocks);
1738         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1739              reg->hr_block_bytes, reg->hr_block_bits);
1740         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1741         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1742 }
1743 
1744 static int o2hb_map_slot_data(struct o2hb_region *reg)
1745 {
1746         int i, j;
1747         unsigned int last_slot;
1748         unsigned int spp = reg->hr_slots_per_page;
1749         struct page *page;
1750         char *raw;
1751         struct o2hb_disk_slot *slot;
1752 
1753         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1754         if (reg->hr_tmp_block == NULL)
1755                 return -ENOMEM;
1756 
1757         reg->hr_slots = kcalloc(reg->hr_blocks,
1758                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1759         if (reg->hr_slots == NULL)
1760                 return -ENOMEM;
1761 
1762         for(i = 0; i < reg->hr_blocks; i++) {
1763                 slot = &reg->hr_slots[i];
1764                 slot->ds_node_num = i;
1765                 INIT_LIST_HEAD(&slot->ds_live_item);
1766                 slot->ds_raw_block = NULL;
1767         }
1768 
1769         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1770         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1771                            "at %u blocks per page\n",
1772              reg->hr_num_pages, reg->hr_blocks, spp);
1773 
1774         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1775                                     GFP_KERNEL);
1776         if (!reg->hr_slot_data)
1777                 return -ENOMEM;
1778 
1779         for(i = 0; i < reg->hr_num_pages; i++) {
1780                 page = alloc_page(GFP_KERNEL);
1781                 if (!page)
1782                         return -ENOMEM;
1783 
1784                 reg->hr_slot_data[i] = page;
1785 
1786                 last_slot = i * spp;
1787                 raw = page_address(page);
1788                 for (j = 0;
1789                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1790                      j++) {
1791                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1792 
1793                         slot = &reg->hr_slots[j + last_slot];
1794                         slot->ds_raw_block =
1795                                 (struct o2hb_disk_heartbeat_block *) raw;
1796 
1797                         raw += reg->hr_block_bytes;
1798                 }
1799         }
1800 
1801         return 0;
1802 }
1803 
1804 /* Read in all the slots available and populate the tracking
1805  * structures so that we can start with a baseline idea of what's
1806  * there. */
1807 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1808 {
1809         int ret, i;
1810         struct o2hb_disk_slot *slot;
1811         struct o2hb_disk_heartbeat_block *hb_block;
1812 
1813         ret = o2hb_read_slots(reg, reg->hr_blocks);
1814         if (ret)
1815                 goto out;
1816 
1817         /* We only want to get an idea of the values initially in each
1818          * slot, so we do no verification - o2hb_check_slot will
1819          * actually determine if each configured slot is valid and
1820          * whether any values have changed. */
1821         for(i = 0; i < reg->hr_blocks; i++) {
1822                 slot = &reg->hr_slots[i];
1823                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1824 
1825                 /* Only fill the values that o2hb_check_slot uses to
1826                  * determine changing slots */
1827                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1828                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1829         }
1830 
1831 out:
1832         return ret;
1833 }
1834 
1835 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1836 static ssize_t o2hb_region_dev_store(struct config_item *item,
1837                                      const char *page,
1838                                      size_t count)
1839 {
1840         struct o2hb_region *reg = to_o2hb_region(item);
1841         struct task_struct *hb_task;
1842         long fd;
1843         int sectsize;
1844         char *p = (char *)page;
1845         struct fd f;
1846         struct inode *inode;
1847         ssize_t ret = -EINVAL;
1848         int live_threshold;
1849 
1850         if (reg->hr_bdev)
1851                 goto out;
1852 
1853         /* We can't heartbeat without having had our node number
1854          * configured yet. */
1855         if (o2nm_this_node() == O2NM_MAX_NODES)
1856                 goto out;
1857 
1858         fd = simple_strtol(p, &p, 0);
1859         if (!p || (*p && (*p != '\n')))
1860                 goto out;
1861 
1862         if (fd < 0 || fd >= INT_MAX)
1863                 goto out;
1864 
1865         f = fdget(fd);
1866         if (f.file == NULL)
1867                 goto out;
1868 
1869         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1870             reg->hr_block_bytes == 0)
1871                 goto out2;
1872 
1873         inode = igrab(f.file->f_mapping->host);
1874         if (inode == NULL)
1875                 goto out2;
1876 
1877         if (!S_ISBLK(inode->i_mode))
1878                 goto out3;
1879 
1880         reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1881         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1882         if (ret) {
1883                 reg->hr_bdev = NULL;
1884                 goto out3;
1885         }
1886         inode = NULL;
1887 
1888         bdevname(reg->hr_bdev, reg->hr_dev_name);
1889 
1890         sectsize = bdev_logical_block_size(reg->hr_bdev);
1891         if (sectsize != reg->hr_block_bytes) {
1892                 mlog(ML_ERROR,
1893                      "blocksize %u incorrect for device, expected %d",
1894                      reg->hr_block_bytes, sectsize);
1895                 ret = -EINVAL;
1896                 goto out3;
1897         }
1898 
1899         o2hb_init_region_params(reg);
1900 
1901         /* Generation of zero is invalid */
1902         do {
1903                 get_random_bytes(&reg->hr_generation,
1904                                  sizeof(reg->hr_generation));
1905         } while (reg->hr_generation == 0);
1906 
1907         ret = o2hb_map_slot_data(reg);
1908         if (ret) {
1909                 mlog_errno(ret);
1910                 goto out3;
1911         }
1912 
1913         ret = o2hb_populate_slot_data(reg);
1914         if (ret) {
1915                 mlog_errno(ret);
1916                 goto out3;
1917         }
1918 
1919         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1920         INIT_DELAYED_WORK(&reg->hr_nego_timeout_work, o2hb_nego_timeout);
1921 
1922         /*
1923          * A node is considered live after it has beat LIVE_THRESHOLD
1924          * times.  We're not steady until we've given them a chance
1925          * _after_ our first read.
1926          * The default threshold is bare minimum so as to limit the delay
1927          * during mounts. For global heartbeat, the threshold doubled for the
1928          * first region.
1929          */
1930         live_threshold = O2HB_LIVE_THRESHOLD;
1931         if (o2hb_global_heartbeat_active()) {
1932                 spin_lock(&o2hb_live_lock);
1933                 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1934                         live_threshold <<= 1;
1935                 spin_unlock(&o2hb_live_lock);
1936         }
1937         ++live_threshold;
1938         atomic_set(&reg->hr_steady_iterations, live_threshold);
1939         /* unsteady_iterations is triple the steady_iterations */
1940         atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
1941 
1942         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1943                               reg->hr_item.ci_name);
1944         if (IS_ERR(hb_task)) {
1945                 ret = PTR_ERR(hb_task);
1946                 mlog_errno(ret);
1947                 goto out3;
1948         }
1949 
1950         spin_lock(&o2hb_live_lock);
1951         reg->hr_task = hb_task;
1952         spin_unlock(&o2hb_live_lock);
1953 
1954         ret = wait_event_interruptible(o2hb_steady_queue,
1955                                 atomic_read(&reg->hr_steady_iterations) == 0 ||
1956                                 reg->hr_node_deleted);
1957         if (ret) {
1958                 atomic_set(&reg->hr_steady_iterations, 0);
1959                 reg->hr_aborted_start = 1;
1960         }
1961 
1962         if (reg->hr_aborted_start) {
1963                 ret = -EIO;
1964                 goto out3;
1965         }
1966 
1967         if (reg->hr_node_deleted) {
1968                 ret = -EINVAL;
1969                 goto out3;
1970         }
1971 
1972         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1973         spin_lock(&o2hb_live_lock);
1974         hb_task = reg->hr_task;
1975         if (o2hb_global_heartbeat_active())
1976                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1977         spin_unlock(&o2hb_live_lock);
1978 
1979         if (hb_task)
1980                 ret = count;
1981         else
1982                 ret = -EIO;
1983 
1984         if (hb_task && o2hb_global_heartbeat_active())
1985                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1986                        config_item_name(&reg->hr_item), reg->hr_dev_name);
1987 
1988 out3:
1989         iput(inode);
1990 out2:
1991         fdput(f);
1992 out:
1993         if (ret < 0) {
1994                 if (reg->hr_bdev) {
1995                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1996                         reg->hr_bdev = NULL;
1997                 }
1998         }
1999         return ret;
2000 }
2001 
2002 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
2003 {
2004         struct o2hb_region *reg = to_o2hb_region(item);
2005         pid_t pid = 0;
2006 
2007         spin_lock(&o2hb_live_lock);
2008         if (reg->hr_task)
2009                 pid = task_pid_nr(reg->hr_task);
2010         spin_unlock(&o2hb_live_lock);
2011 
2012         if (!pid)
2013                 return 0;
2014 
2015         return sprintf(page, "%u\n", pid);
2016 }
2017 
2018 CONFIGFS_ATTR(o2hb_region_, block_bytes);
2019 CONFIGFS_ATTR(o2hb_region_, start_block);
2020 CONFIGFS_ATTR(o2hb_region_, blocks);
2021 CONFIGFS_ATTR(o2hb_region_, dev);
2022 CONFIGFS_ATTR_RO(o2hb_region_, pid);
2023 
2024 static struct configfs_attribute *o2hb_region_attrs[] = {
2025         &o2hb_region_attr_block_bytes,
2026         &o2hb_region_attr_start_block,
2027         &o2hb_region_attr_blocks,
2028         &o2hb_region_attr_dev,
2029         &o2hb_region_attr_pid,
2030         NULL,
2031 };
2032 
2033 static struct configfs_item_operations o2hb_region_item_ops = {
2034         .release                = o2hb_region_release,
2035 };
2036 
2037 static const struct config_item_type o2hb_region_type = {
2038         .ct_item_ops    = &o2hb_region_item_ops,
2039         .ct_attrs       = o2hb_region_attrs,
2040         .ct_owner       = THIS_MODULE,
2041 };
2042 
2043 /* heartbeat set */
2044 
2045 struct o2hb_heartbeat_group {
2046         struct config_group hs_group;
2047         /* some stuff? */
2048 };
2049 
2050 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2051 {
2052         return group ?
2053                 container_of(group, struct o2hb_heartbeat_group, hs_group)
2054                 : NULL;
2055 }
2056 
2057 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2058 {
2059         int ret = -ENOMEM;
2060 
2061         reg->hr_debug_dir =
2062                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
2063         if (!reg->hr_debug_dir) {
2064                 mlog_errno(ret);
2065                 goto bail;
2066         }
2067 
2068         reg->hr_debug_livenodes =
2069                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2070                                           reg->hr_debug_dir,
2071                                           &(reg->hr_db_livenodes),
2072                                           sizeof(*(reg->hr_db_livenodes)),
2073                                           O2HB_DB_TYPE_REGION_LIVENODES,
2074                                           sizeof(reg->hr_live_node_bitmap),
2075                                           O2NM_MAX_NODES, reg);
2076         if (!reg->hr_debug_livenodes) {
2077                 mlog_errno(ret);
2078                 goto bail;
2079         }
2080 
2081         reg->hr_debug_regnum =
2082                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2083                                           reg->hr_debug_dir,
2084                                           &(reg->hr_db_regnum),
2085                                           sizeof(*(reg->hr_db_regnum)),
2086                                           O2HB_DB_TYPE_REGION_NUMBER,
2087                                           0, O2NM_MAX_NODES, reg);
2088         if (!reg->hr_debug_regnum) {
2089                 mlog_errno(ret);
2090                 goto bail;
2091         }
2092 
2093         reg->hr_debug_elapsed_time =
2094                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2095                                           reg->hr_debug_dir,
2096                                           &(reg->hr_db_elapsed_time),
2097                                           sizeof(*(reg->hr_db_elapsed_time)),
2098                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2099                                           0, 0, reg);
2100         if (!reg->hr_debug_elapsed_time) {
2101                 mlog_errno(ret);
2102                 goto bail;
2103         }
2104 
2105         reg->hr_debug_pinned =
2106                         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2107                                           reg->hr_debug_dir,
2108                                           &(reg->hr_db_pinned),
2109                                           sizeof(*(reg->hr_db_pinned)),
2110                                           O2HB_DB_TYPE_REGION_PINNED,
2111                                           0, 0, reg);
2112         if (!reg->hr_debug_pinned) {
2113                 mlog_errno(ret);
2114                 goto bail;
2115         }
2116 
2117         ret = 0;
2118 bail:
2119         return ret;
2120 }
2121 
2122 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2123                                                           const char *name)
2124 {
2125         struct o2hb_region *reg = NULL;
2126         int ret;
2127 
2128         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2129         if (reg == NULL)
2130                 return ERR_PTR(-ENOMEM);
2131 
2132         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2133                 ret = -ENAMETOOLONG;
2134                 goto free;
2135         }
2136 
2137         spin_lock(&o2hb_live_lock);
2138         reg->hr_region_num = 0;
2139         if (o2hb_global_heartbeat_active()) {
2140                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2141                                                          O2NM_MAX_REGIONS);
2142                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2143                         spin_unlock(&o2hb_live_lock);
2144                         ret = -EFBIG;
2145                         goto free;
2146                 }
2147                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2148         }
2149         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2150         spin_unlock(&o2hb_live_lock);
2151 
2152         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2153 
2154         /* this is the same way to generate msg key as dlm, for local heartbeat,
2155          * name is also the same, so make initial crc value different to avoid
2156          * message key conflict.
2157          */
2158         reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2159                 name, strlen(name));
2160         INIT_LIST_HEAD(&reg->hr_handler_list);
2161         ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2162                         sizeof(struct o2hb_nego_msg),
2163                         o2hb_nego_timeout_handler,
2164                         reg, NULL, &reg->hr_handler_list);
2165         if (ret)
2166                 goto free;
2167 
2168         ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2169                         sizeof(struct o2hb_nego_msg),
2170                         o2hb_nego_approve_handler,
2171                         reg, NULL, &reg->hr_handler_list);
2172         if (ret)
2173                 goto unregister_handler;
2174 
2175         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2176         if (ret) {
2177                 config_item_put(&reg->hr_item);
2178                 goto unregister_handler;
2179         }
2180 
2181         return &reg->hr_item;
2182 
2183 unregister_handler:
2184         o2net_unregister_handler_list(&reg->hr_handler_list);
2185 free:
2186         kfree(reg);
2187         return ERR_PTR(ret);
2188 }
2189 
2190 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2191                                            struct config_item *item)
2192 {
2193         struct task_struct *hb_task;
2194         struct o2hb_region *reg = to_o2hb_region(item);
2195         int quorum_region = 0;
2196 
2197         /* stop the thread when the user removes the region dir */
2198         spin_lock(&o2hb_live_lock);
2199         hb_task = reg->hr_task;
2200         reg->hr_task = NULL;
2201         reg->hr_item_dropped = 1;
2202         spin_unlock(&o2hb_live_lock);
2203 
2204         if (hb_task)
2205                 kthread_stop(hb_task);
2206 
2207         if (o2hb_global_heartbeat_active()) {
2208                 spin_lock(&o2hb_live_lock);
2209                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2210                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2211                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2212                         quorum_region = 1;
2213                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2214                 spin_unlock(&o2hb_live_lock);
2215                 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2216                        ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2217                         "stopped" : "start aborted"), config_item_name(item),
2218                        reg->hr_dev_name);
2219         }
2220 
2221         /*
2222          * If we're racing a dev_write(), we need to wake them.  They will
2223          * check reg->hr_task
2224          */
2225         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2226                 reg->hr_aborted_start = 1;
2227                 atomic_set(&reg->hr_steady_iterations, 0);
2228                 wake_up(&o2hb_steady_queue);
2229         }
2230 
2231         config_item_put(item);
2232 
2233         if (!o2hb_global_heartbeat_active() || !quorum_region)
2234                 return;
2235 
2236         /*
2237          * If global heartbeat active and there are dependent users,
2238          * pin all regions if quorum region count <= CUT_OFF
2239          */
2240         spin_lock(&o2hb_live_lock);
2241 
2242         if (!o2hb_dependent_users)
2243                 goto unlock;
2244 
2245         if (bitmap_weight(o2hb_quorum_region_bitmap,
2246                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2247                 o2hb_region_pin(NULL);
2248 
2249 unlock:
2250         spin_unlock(&o2hb_live_lock);
2251 }
2252 
2253 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2254                 char *page)
2255 {
2256         return sprintf(page, "%u\n", o2hb_dead_threshold);
2257 }
2258 
2259 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2260                 const char *page, size_t count)
2261 {
2262         unsigned long tmp;
2263         char *p = (char *)page;
2264 
2265         tmp = simple_strtoul(p, &p, 10);
2266         if (!p || (*p && (*p != '\n')))
2267                 return -EINVAL;
2268 
2269         /* this will validate ranges for us. */
2270         o2hb_dead_threshold_set((unsigned int) tmp);
2271 
2272         return count;
2273 }
2274 
2275 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2276                 char *page)
2277 {
2278         return sprintf(page, "%s\n",
2279                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2280 }
2281 
2282 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2283                 const char *page, size_t count)
2284 {
2285         unsigned int i;
2286         int ret;
2287         size_t len;
2288 
2289         len = (page[count - 1] == '\n') ? count - 1 : count;
2290         if (!len)
2291                 return -EINVAL;
2292 
2293         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2294                 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2295                         continue;
2296 
2297                 ret = o2hb_global_heartbeat_mode_set(i);
2298                 if (!ret)
2299                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2300                                o2hb_heartbeat_mode_desc[i]);
2301                 return count;
2302         }
2303 
2304         return -EINVAL;
2305 
2306 }
2307 
2308 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2309 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2310 
2311 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2312         &o2hb_heartbeat_group_attr_dead_threshold,
2313         &o2hb_heartbeat_group_attr_mode,
2314         NULL,
2315 };
2316 
2317 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2318         .make_item      = o2hb_heartbeat_group_make_item,
2319         .drop_item      = o2hb_heartbeat_group_drop_item,
2320 };
2321 
2322 static const struct config_item_type o2hb_heartbeat_group_type = {
2323         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2324         .ct_attrs       = o2hb_heartbeat_group_attrs,
2325         .ct_owner       = THIS_MODULE,
2326 };
2327 
2328 /* this is just here to avoid touching group in heartbeat.h which the
2329  * entire damn world #includes */
2330 struct config_group *o2hb_alloc_hb_set(void)
2331 {
2332         struct o2hb_heartbeat_group *hs = NULL;
2333         struct config_group *ret = NULL;
2334 
2335         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2336         if (hs == NULL)
2337                 goto out;
2338 
2339         config_group_init_type_name(&hs->hs_group, "heartbeat",
2340                                     &o2hb_heartbeat_group_type);
2341 
2342         ret = &hs->hs_group;
2343 out:
2344         if (ret == NULL)
2345                 kfree(hs);
2346         return ret;
2347 }
2348 
2349 void o2hb_free_hb_set(struct config_group *group)
2350 {
2351         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2352         kfree(hs);
2353 }
2354 
2355 /* hb callback registration and issuing */
2356 
2357 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2358 {
2359         if (type == O2HB_NUM_CB)
2360                 return ERR_PTR(-EINVAL);
2361 
2362         return &o2hb_callbacks[type];
2363 }
2364 
2365 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2366                          enum o2hb_callback_type type,
2367                          o2hb_cb_func *func,
2368                          void *data,
2369                          int priority)
2370 {
2371         INIT_LIST_HEAD(&hc->hc_item);
2372         hc->hc_func = func;
2373         hc->hc_data = data;
2374         hc->hc_priority = priority;
2375         hc->hc_type = type;
2376         hc->hc_magic = O2HB_CB_MAGIC;
2377 }
2378 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2379 
2380 /*
2381  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2382  * In global heartbeat mode, region_uuid passed is NULL.
2383  *
2384  * In local, we only pin the matching region. In global we pin all the active
2385  * regions.
2386  */
2387 static int o2hb_region_pin(const char *region_uuid)
2388 {
2389         int ret = 0, found = 0;
2390         struct o2hb_region *reg;
2391         char *uuid;
2392 
2393         assert_spin_locked(&o2hb_live_lock);
2394 
2395         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2396                 if (reg->hr_item_dropped)
2397                         continue;
2398 
2399                 uuid = config_item_name(&reg->hr_item);
2400 
2401                 /* local heartbeat */
2402                 if (region_uuid) {
2403                         if (strcmp(region_uuid, uuid))
2404                                 continue;
2405                         found = 1;
2406                 }
2407 
2408                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2409                         goto skip_pin;
2410 
2411                 /* Ignore ENOENT only for local hb (userdlm domain) */
2412                 ret = o2nm_depend_item(&reg->hr_item);
2413                 if (!ret) {
2414                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2415                         reg->hr_item_pinned = 1;
2416                 } else {
2417                         if (ret == -ENOENT && found)
2418                                 ret = 0;
2419                         else {
2420                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2421                                      uuid, ret);
2422                                 break;
2423                         }
2424                 }
2425 skip_pin:
2426                 if (found)
2427                         break;
2428         }
2429 
2430         return ret;
2431 }
2432 
2433 /*
2434  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2435  * In global heartbeat mode, region_uuid passed is NULL.
2436  *
2437  * In local, we only unpin the matching region. In global we unpin all the
2438  * active regions.
2439  */
2440 static void o2hb_region_unpin(const char *region_uuid)
2441 {
2442         struct o2hb_region *reg;
2443         char *uuid;
2444         int found = 0;
2445 
2446         assert_spin_locked(&o2hb_live_lock);
2447 
2448         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2449                 if (reg->hr_item_dropped)
2450                         continue;
2451 
2452                 uuid = config_item_name(&reg->hr_item);
2453                 if (region_uuid) {
2454                         if (strcmp(region_uuid, uuid))
2455                                 continue;
2456                         found = 1;
2457                 }
2458 
2459                 if (reg->hr_item_pinned) {
2460                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2461                         o2nm_undepend_item(&reg->hr_item);
2462                         reg->hr_item_pinned = 0;
2463                 }
2464                 if (found)
2465                         break;
2466         }
2467 }
2468 
2469 static int o2hb_region_inc_user(const char *region_uuid)
2470 {
2471         int ret = 0;
2472 
2473         spin_lock(&o2hb_live_lock);
2474 
2475         /* local heartbeat */
2476         if (!o2hb_global_heartbeat_active()) {
2477             ret = o2hb_region_pin(region_uuid);
2478             goto unlock;
2479         }
2480 
2481         /*
2482          * if global heartbeat active and this is the first dependent user,
2483          * pin all regions if quorum region count <= CUT_OFF
2484          */
2485         o2hb_dependent_users++;
2486         if (o2hb_dependent_users > 1)
2487                 goto unlock;
2488 
2489         if (bitmap_weight(o2hb_quorum_region_bitmap,
2490                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2491                 ret = o2hb_region_pin(NULL);
2492 
2493 unlock:
2494         spin_unlock(&o2hb_live_lock);
2495         return ret;
2496 }
2497 
2498 void o2hb_region_dec_user(const char *region_uuid)
2499 {
2500         spin_lock(&o2hb_live_lock);
2501 
2502         /* local heartbeat */
2503         if (!o2hb_global_heartbeat_active()) {
2504             o2hb_region_unpin(region_uuid);
2505             goto unlock;
2506         }
2507 
2508         /*
2509          * if global heartbeat active and there are no dependent users,
2510          * unpin all quorum regions
2511          */
2512         o2hb_dependent_users--;
2513         if (!o2hb_dependent_users)
2514                 o2hb_region_unpin(NULL);
2515 
2516 unlock:
2517         spin_unlock(&o2hb_live_lock);
2518 }
2519 
2520 int o2hb_register_callback(const char *region_uuid,
2521                            struct o2hb_callback_func *hc)
2522 {
2523         struct o2hb_callback_func *f;
2524         struct o2hb_callback *hbcall;
2525         int ret;
2526 
2527         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2528         BUG_ON(!list_empty(&hc->hc_item));
2529 
2530         hbcall = hbcall_from_type(hc->hc_type);
2531         if (IS_ERR(hbcall)) {
2532                 ret = PTR_ERR(hbcall);
2533                 goto out;
2534         }
2535 
2536         if (region_uuid) {
2537                 ret = o2hb_region_inc_user(region_uuid);
2538                 if (ret) {
2539                         mlog_errno(ret);
2540                         goto out;
2541                 }
2542         }
2543 
2544         down_write(&o2hb_callback_sem);
2545 
2546         list_for_each_entry(f, &hbcall->list, hc_item) {
2547                 if (hc->hc_priority < f->hc_priority) {
2548                         list_add_tail(&hc->hc_item, &f->hc_item);
2549                         break;
2550                 }
2551         }
2552         if (list_empty(&hc->hc_item))
2553                 list_add_tail(&hc->hc_item, &hbcall->list);
2554 
2555         up_write(&o2hb_callback_sem);
2556         ret = 0;
2557 out:
2558         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2559              ret, __builtin_return_address(0), hc);
2560         return ret;
2561 }
2562 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2563 
2564 void o2hb_unregister_callback(const char *region_uuid,
2565                               struct o2hb_callback_func *hc)
2566 {
2567         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2568 
2569         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2570              __builtin_return_address(0), hc);
2571 
2572         /* XXX Can this happen _with_ a region reference? */
2573         if (list_empty(&hc->hc_item))
2574                 return;
2575 
2576         if (region_uuid)
2577                 o2hb_region_dec_user(region_uuid);
2578 
2579         down_write(&o2hb_callback_sem);
2580 
2581         list_del_init(&hc->hc_item);
2582 
2583         up_write(&o2hb_callback_sem);
2584 }
2585 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2586 
2587 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2588 {
2589         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2590 
2591         spin_lock(&o2hb_live_lock);
2592         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2593         spin_unlock(&o2hb_live_lock);
2594         if (!test_bit(node_num, testing_map)) {
2595                 mlog(ML_HEARTBEAT,
2596                      "node (%u) does not have heartbeating enabled.\n",
2597                      node_num);
2598                 return 0;
2599         }
2600 
2601         return 1;
2602 }
2603 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2604 
2605 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2606 {
2607         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2608 
2609         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2610         if (!test_bit(node_num, testing_map)) {
2611                 mlog(ML_HEARTBEAT,
2612                      "node (%u) does not have heartbeating enabled.\n",
2613                      node_num);
2614                 return 0;
2615         }
2616 
2617         return 1;
2618 }
2619 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2620 
2621 /*
2622  * this is just a hack until we get the plumbing which flips file systems
2623  * read only and drops the hb ref instead of killing the node dead.
2624  */
2625 void o2hb_stop_all_regions(void)
2626 {
2627         struct o2hb_region *reg;
2628 
2629         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2630 
2631         spin_lock(&o2hb_live_lock);
2632 
2633         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2634                 reg->hr_unclean_stop = 1;
2635 
2636         spin_unlock(&o2hb_live_lock);
2637 }
2638 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2639 
2640 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2641 {
2642         struct o2hb_region *reg;
2643         int numregs = 0;
2644         char *p;
2645 
2646         spin_lock(&o2hb_live_lock);
2647 
2648         p = region_uuids;
2649         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2650                 if (reg->hr_item_dropped)
2651                         continue;
2652 
2653                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2654                 if (numregs < max_regions) {
2655                         memcpy(p, config_item_name(&reg->hr_item),
2656                                O2HB_MAX_REGION_NAME_LEN);
2657                         p += O2HB_MAX_REGION_NAME_LEN;
2658                 }
2659                 numregs++;
2660         }
2661 
2662         spin_unlock(&o2hb_live_lock);
2663 
2664         return numregs;
2665 }
2666 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2667 
2668 int o2hb_global_heartbeat_active(void)
2669 {
2670         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2671 }
2672 EXPORT_SYMBOL(o2hb_global_heartbeat_active);
2673 

~ [ 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