]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/ocfs2/cluster/heartbeat.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: add a bi_error field to struct bio
[mirror_ubuntu-bionic-kernel.git] / fs / ocfs2 / cluster / heartbeat.c
CommitLineData
a7f6a5fb
MF		 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>
87d3d3f3 36#include <linux/debugfs.h>
5a0e3ad6 37#include <linux/slab.h>
a8f70de3 38#include <linux/bitmap.h>
a7f6a5fb
MF		 39
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 */
54static 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 */
34af946a 60static DEFINE_SPINLOCK(o2hb_live_lock);
a7f6a5fb
MF		 61static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
62static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
63static LIST_HEAD(o2hb_node_events);
64static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
65
536f0741
SM		 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.
e7d656ba 69 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
43182d2a
SM		 70 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
71 * heartbeat on it.
b1c5ebfb 72 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
536f0741
SM		 73 */
74static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
e7d656ba 75static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
43182d2a 76static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
b1c5ebfb 77static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
536f0741 78
8ca8b0bb 79#define O2HB_DB_TYPE_LIVENODES 0
a6de0136
SM		 80#define O2HB_DB_TYPE_LIVEREGIONS 1
81#define O2HB_DB_TYPE_QUORUMREGIONS 2
82#define O2HB_DB_TYPE_FAILEDREGIONS 3
1f285305
SM		 83#define O2HB_DB_TYPE_REGION_LIVENODES 4
84#define O2HB_DB_TYPE_REGION_NUMBER 5
43695d09 85#define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
cb0586bd 86#define O2HB_DB_TYPE_REGION_PINNED 7
8ca8b0bb
SM		 87struct o2hb_debug_buf {
88 int db_type;
89 int db_size;
90 int db_len;
91 void *db_data;
92};
93
94static struct o2hb_debug_buf *o2hb_db_livenodes;
a6de0136
SM		 95static struct o2hb_debug_buf *o2hb_db_liveregions;
96static struct o2hb_debug_buf *o2hb_db_quorumregions;
97static struct o2hb_debug_buf *o2hb_db_failedregions;
8ca8b0bb 98
87d3d3f3
SM		 99#define O2HB_DEBUG_DIR "o2hb"
100#define O2HB_DEBUG_LIVENODES "livenodes"
a6de0136
SM		 101#define O2HB_DEBUG_LIVEREGIONS "live_regions"
102#define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
103#define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
1f285305 104#define O2HB_DEBUG_REGION_NUMBER "num"
43695d09 105#define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
cb0586bd 106#define O2HB_DEBUG_REGION_PINNED "pinned"
8ca8b0bb 107
87d3d3f3
SM		 108static struct dentry *o2hb_debug_dir;
109static struct dentry *o2hb_debug_livenodes;
a6de0136
SM		 110static struct dentry *o2hb_debug_liveregions;
111static struct dentry *o2hb_debug_quorumregions;
112static struct dentry *o2hb_debug_failedregions;
87d3d3f3 113
a7f6a5fb
MF		 114static LIST_HEAD(o2hb_all_regions);
115
116static struct o2hb_callback {
117 struct list_head list;
118} o2hb_callbacks[O2HB_NUM_CB];
119
120static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
121
122#define O2HB_DEFAULT_BLOCK_BITS 9
123
54b5187b
SM		 124enum o2hb_heartbeat_modes {
125 O2HB_HEARTBEAT_LOCAL = 0,
126 O2HB_HEARTBEAT_GLOBAL,
127 O2HB_HEARTBEAT_NUM_MODES,
128};
129
130char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
131 "local", /* O2HB_HEARTBEAT_LOCAL */
132 "global", /* O2HB_HEARTBEAT_GLOBAL */
133};
134
a7f6a5fb 135unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
54b5187b 136unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
a7f6a5fb 137
58a3158a
SM		 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 */
144unsigned 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 */
162static int o2hb_region_pin(const char *region_uuid);
163static void o2hb_region_unpin(const char *region_uuid);
164
2bd63216 165/* Only sets a new threshold if there are no active regions.
a7f6a5fb
MF		 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. */
170static 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
70f651ed 180static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
54b5187b
SM		 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
a7f6a5fb
MF		 196struct 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
203struct 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 */
215struct o2hb_region {
216 struct config_item hr_item;
217
218 struct list_head hr_all_item;
58a3158a 219 unsigned hr_unclean_stop:1,
d2eece37 220 hr_aborted_start:1,
58a3158a
SM		 221 hr_item_pinned:1,
222 hr_item_dropped:1;
a7f6a5fb
MF		 223
224 /* protected by the hr_callback_sem */
225 struct task_struct *hr_task;
226
227 unsigned int hr_blocks;
228 unsigned long long hr_start_block;
229
230 unsigned int hr_block_bits;
231 unsigned int hr_block_bytes;
232
233 unsigned int hr_slots_per_page;
234 unsigned int hr_num_pages;
235
236 struct page **hr_slot_data;
237 struct block_device *hr_bdev;
238 struct o2hb_disk_slot *hr_slots;
239
823a637a
SM		 240 /* live node map of this region */
241 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
536f0741 242 unsigned int hr_region_num;
823a637a 243
1f285305
SM		 244 struct dentry *hr_debug_dir;
245 struct dentry *hr_debug_livenodes;
246 struct dentry *hr_debug_regnum;
43695d09 247 struct dentry *hr_debug_elapsed_time;
cb0586bd 248 struct dentry *hr_debug_pinned;
1f285305
SM		 249 struct o2hb_debug_buf *hr_db_livenodes;
250 struct o2hb_debug_buf *hr_db_regnum;
43695d09 251 struct o2hb_debug_buf *hr_db_elapsed_time;
cb0586bd 252 struct o2hb_debug_buf *hr_db_pinned;
1f285305 253
a7f6a5fb
MF		 254 /* let the person setting up hb wait for it to return until it
255 * has reached a 'steady' state. This will be fixed when we have
256 * a more complete api that doesn't lead to this sort of fragility. */
257 atomic_t hr_steady_iterations;
258
d2eece37
SM		 259 /* terminate o2hb thread if it does not reach steady state
260 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
261 atomic_t hr_unsteady_iterations;
262
a7f6a5fb
MF		 263 char hr_dev_name[BDEVNAME_SIZE];
264
265 unsigned int hr_timeout_ms;
266
267 /* randomized as the region goes up and down so that a node
268 * recognizes a node going up and down in one iteration */
269 u64 hr_generation;
270
c4028958 271 struct delayed_work hr_write_timeout_work;
a7f6a5fb
MF		 272 unsigned long hr_last_timeout_start;
273
274 /* Used during o2hb_check_slot to hold a copy of the block
275 * being checked because we temporarily have to zero out the
276 * crc field. */
277 struct o2hb_disk_heartbeat_block *hr_tmp_block;
278};
279
280struct o2hb_bio_wait_ctxt {
281 atomic_t wc_num_reqs;
282 struct completion wc_io_complete;
a9e2ae39 283 int wc_error;
a7f6a5fb
MF		 284};
285
c4028958 286static void o2hb_write_timeout(struct work_struct *work)
a7f6a5fb 287{
b1c5ebfb
SM		 288 int failed, quorum;
289 unsigned long flags;
c4028958
DH		 290 struct o2hb_region *reg =
291 container_of(work, struct o2hb_region,
292 hr_write_timeout_work.work);
a7f6a5fb
MF		 293
294 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
295 "milliseconds\n", reg->hr_dev_name,
2bd63216 296 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
b1c5ebfb
SM		 297
298 if (o2hb_global_heartbeat_active()) {
299 spin_lock_irqsave(&o2hb_live_lock, flags);
300 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
301 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
a8f70de3 302 failed = bitmap_weight(o2hb_failed_region_bitmap,
b1c5ebfb 303 O2NM_MAX_REGIONS);
a8f70de3 304 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
b1c5ebfb
SM		 305 O2NM_MAX_REGIONS);
306 spin_unlock_irqrestore(&o2hb_live_lock, flags);
307
308 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
309 quorum, failed);
310
311 /*
312 * Fence if the number of failed regions >= half the number
313 * of quorum regions
314 */
315 if ((failed << 1) < quorum)
316 return;
317 }
318
a7f6a5fb
MF		 319 o2quo_disk_timeout();
320}
321
322static void o2hb_arm_write_timeout(struct o2hb_region *reg)
323{
d2eece37
SM		 324 /* Arm writeout only after thread reaches steady state */
325 if (atomic_read(&reg->hr_steady_iterations) != 0)
326 return;
327
b31d308d
TM		 328 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
329 O2HB_MAX_WRITE_TIMEOUT_MS);
a7f6a5fb 330
b1c5ebfb
SM		 331 if (o2hb_global_heartbeat_active()) {
332 spin_lock(&o2hb_live_lock);
333 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
334 spin_unlock(&o2hb_live_lock);
335 }
a7f6a5fb
MF		 336 cancel_delayed_work(&reg->hr_write_timeout_work);
337 reg->hr_last_timeout_start = jiffies;
338 schedule_delayed_work(&reg->hr_write_timeout_work,
339 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
340}
341
342static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
343{
9b00a818 344 cancel_delayed_work_sync(&reg->hr_write_timeout_work);
a7f6a5fb
MF		 345}
346
b559292e 347static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
a7f6a5fb 348{
b559292e 349 atomic_set(&wc->wc_num_reqs, 1);
a7f6a5fb 350 init_completion(&wc->wc_io_complete);
a9e2ae39 351 wc->wc_error = 0;
a7f6a5fb
MF		 352}
353
354/* Used in error paths too */
355static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
356 unsigned int num)
357{
358 /* sadly atomic_sub_and_test() isn't available on all platforms. The
359 * good news is that the fast path only completes one at a time */
360 while(num--) {
361 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
362 BUG_ON(num > 0);
363 complete(&wc->wc_io_complete);
364 }
365 }
366}
367
368static void o2hb_wait_on_io(struct o2hb_region *reg,
369 struct o2hb_bio_wait_ctxt *wc)
370{
b559292e 371 o2hb_bio_wait_dec(wc, 1);
a7f6a5fb
MF		 372 wait_for_completion(&wc->wc_io_complete);
373}
374
4246a0b6 375static void o2hb_bio_end_io(struct bio *bio)
a7f6a5fb
MF		 376{
377 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
378
4246a0b6
CH		 379 if (bio->bi_error) {
380 mlog(ML_ERROR, "IO Error %d\n", bio->bi_error);
381 wc->wc_error = bio->bi_error;
a9e2ae39 382 }
a7f6a5fb 383
a7f6a5fb 384 o2hb_bio_wait_dec(wc, 1);
b559292e 385 bio_put(bio);
a7f6a5fb
MF		 386}
387
388/* Setup a Bio to cover I/O against num_slots slots starting at
389 * start_slot. */
390static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
391 struct o2hb_bio_wait_ctxt *wc,
b559292e
PR		 392 unsigned int *current_slot,
393 unsigned int max_slots)
a7f6a5fb 394{
b559292e 395 int len, current_page;
a7f6a5fb
MF		 396 unsigned int vec_len, vec_start;
397 unsigned int bits = reg->hr_block_bits;
398 unsigned int spp = reg->hr_slots_per_page;
b559292e 399 unsigned int cs = *current_slot;
a7f6a5fb
MF		 400 struct bio *bio;
401 struct page *page;
402
a7f6a5fb
MF		 403 /* Testing has shown this allocation to take long enough under
404 * GFP_KERNEL that the local node can get fenced. It would be
405 * nicest if we could pre-allocate these bios and avoid this
406 * all together. */
b559292e 407 bio = bio_alloc(GFP_ATOMIC, 16);
a7f6a5fb
MF		 408 if (!bio) {
409 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
410 bio = ERR_PTR(-ENOMEM);
411 goto bail;
412 }
413
414 /* Must put everything in 512 byte sectors for the bio... */
4f024f37 415 bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
a7f6a5fb
MF		 416 bio->bi_bdev = reg->hr_bdev;
417 bio->bi_private = wc;
418 bio->bi_end_io = o2hb_bio_end_io;
419
b559292e
PR		 420 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
421 while(cs < max_slots) {
422 current_page = cs / spp;
423 page = reg->hr_slot_data[current_page];
a7f6a5fb 424
bc7e97cb 425 vec_len = min(PAGE_CACHE_SIZE - vec_start,
b559292e 426 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
a7f6a5fb
MF		 427
428 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
b559292e 429 current_page, vec_len, vec_start);
a7f6a5fb
MF		 430
431 len = bio_add_page(bio, page, vec_len, vec_start);
b559292e 432 if (len != vec_len) break;
a7f6a5fb 433
b559292e 434 cs += vec_len / (PAGE_CACHE_SIZE/spp);
a7f6a5fb
MF		 435 vec_start = 0;
436 }
437
438bail:
b559292e 439 *current_slot = cs;
a7f6a5fb
MF		 440 return bio;
441}
442
a7f6a5fb
MF		 443static int o2hb_read_slots(struct o2hb_region *reg,
444 unsigned int max_slots)
445{
b559292e
PR		 446 unsigned int current_slot=0;
447 int status;
a7f6a5fb 448 struct o2hb_bio_wait_ctxt wc;
a7f6a5fb
MF		 449 struct bio *bio;
450
b559292e 451 o2hb_bio_wait_init(&wc);
a7f6a5fb 452
b559292e
PR		 453 while(current_slot < max_slots) {
454 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
a7f6a5fb 455 if (IS_ERR(bio)) {
a7f6a5fb
MF		 456 status = PTR_ERR(bio);
457 mlog_errno(status);
458 goto bail_and_wait;
459 }
a7f6a5fb 460
b559292e 461 atomic_inc(&wc.wc_num_reqs);
a7f6a5fb
MF		 462 submit_bio(READ, bio);
463 }
464
465 status = 0;
466
467bail_and_wait:
468 o2hb_wait_on_io(reg, &wc);
a9e2ae39
MF		 469 if (wc.wc_error && !status)
470 status = wc.wc_error;
a7f6a5fb 471
a7f6a5fb
MF		 472 return status;
473}
474
475static int o2hb_issue_node_write(struct o2hb_region *reg,
a7f6a5fb
MF		 476 struct o2hb_bio_wait_ctxt *write_wc)
477{
478 int status;
479 unsigned int slot;
480 struct bio *bio;
481
b559292e 482 o2hb_bio_wait_init(write_wc);
a7f6a5fb
MF		 483
484 slot = o2nm_this_node();
485
b559292e 486 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
a7f6a5fb
MF		 487 if (IS_ERR(bio)) {
488 status = PTR_ERR(bio);
489 mlog_errno(status);
490 goto bail;
491 }
492
b559292e 493 atomic_inc(&write_wc->wc_num_reqs);
e873fdb5 494 submit_bio(WRITE_SYNC, bio);
a7f6a5fb 495
a7f6a5fb
MF		 496 status = 0;
497bail:
498 return status;
499}
500
501static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
502 struct o2hb_disk_heartbeat_block *hb_block)
503{
504 __le32 old_cksum;
505 u32 ret;
506
507 /* We want to compute the block crc with a 0 value in the
508 * hb_cksum field. Save it off here and replace after the
509 * crc. */
510 old_cksum = hb_block->hb_cksum;
511 hb_block->hb_cksum = 0;
512
513 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
514
515 hb_block->hb_cksum = old_cksum;
516
517 return ret;
518}
519
520static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
521{
70bacbdb
MF		 522 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
523 "cksum = 0x%x, generation 0x%llx\n",
524 (long long)le64_to_cpu(hb_block->hb_seq),
525 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
526 (long long)le64_to_cpu(hb_block->hb_generation));
a7f6a5fb
MF		 527}
528
529static int o2hb_verify_crc(struct o2hb_region *reg,
530 struct o2hb_disk_heartbeat_block *hb_block)
531{
532 u32 read, computed;
533
534 read = le32_to_cpu(hb_block->hb_cksum);
535 computed = o2hb_compute_block_crc_le(reg, hb_block);
536
537 return read == computed;
538}
539
d2eece37
SM		 540/*
541 * Compare the slot data with what we wrote in the last iteration.
542 * If the match fails, print an appropriate error message. This is to
543 * detect errors like... another node hearting on the same slot,
544 * flaky device that is losing writes, etc.
545 * Returns 1 if check succeeds, 0 otherwise.
546 */
547static int o2hb_check_own_slot(struct o2hb_region *reg)
a7f6a5fb 548{
a7f6a5fb
MF		 549 struct o2hb_disk_slot *slot;
550 struct o2hb_disk_heartbeat_block *hb_block;
33c12a54 551 char *errstr;
a7f6a5fb 552
33c12a54 553 slot = &reg->hr_slots[o2nm_this_node()];
a7f6a5fb 554 /* Don't check on our 1st timestamp */
33c12a54 555 if (!slot->ds_last_time)
d2eece37 556 return 0;
a7f6a5fb 557
33c12a54
SM		 558 hb_block = slot->ds_raw_block;
559 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
560 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
561 hb_block->hb_node == slot->ds_node_num)
d2eece37 562 return 1;
a7f6a5fb 563
33c12a54
SM		 564#define ERRSTR1 "Another node is heartbeating on device"
565#define ERRSTR2 "Heartbeat generation mismatch on device"
566#define ERRSTR3 "Heartbeat sequence mismatch on device"
567
568 if (hb_block->hb_node != slot->ds_node_num)
569 errstr = ERRSTR1;
570 else if (le64_to_cpu(hb_block->hb_generation) !=
571 slot->ds_last_generation)
572 errstr = ERRSTR2;
573 else
574 errstr = ERRSTR3;
575
576 mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
577 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
578 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
579 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
580 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
581 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
d2eece37
SM		 582
583 return 0;
a7f6a5fb
MF		 584}
585
586static inline void o2hb_prepare_block(struct o2hb_region *reg,
587 u64 generation)
588{
589 int node_num;
590 u64 cputime;
591 struct o2hb_disk_slot *slot;
592 struct o2hb_disk_heartbeat_block *hb_block;
593
594 node_num = o2nm_this_node();
595 slot = &reg->hr_slots[node_num];
596
597 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
598 memset(hb_block, 0, reg->hr_block_bytes);
599 /* TODO: time stuff */
600 cputime = CURRENT_TIME.tv_sec;
601 if (!cputime)
602 cputime = 1;
603
604 hb_block->hb_seq = cpu_to_le64(cputime);
605 hb_block->hb_node = node_num;
606 hb_block->hb_generation = cpu_to_le64(generation);
0db638f4 607 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
a7f6a5fb
MF		 608
609 /* This step must always happen last! */
610 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
611 hb_block));
612
70bacbdb 613 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
5fdf1e67 614 (long long)generation,
70bacbdb 615 le32_to_cpu(hb_block->hb_cksum));
a7f6a5fb
MF		 616}
617
618static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
619 struct o2nm_node *node,
620 int idx)
621{
a7f6a5fb
MF		 622 struct o2hb_callback_func *f;
623
df53cd3b 624 list_for_each_entry(f, &hbcall->list, hc_item) {
a7f6a5fb
MF		 625 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
626 (f->hc_func)(node, idx, f->hc_data);
627 }
628}
629
630/* Will run the list in order until we process the passed event */
631static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
632{
a7f6a5fb
MF		 633 struct o2hb_callback *hbcall;
634 struct o2hb_node_event *event;
635
a7f6a5fb
MF		 636 /* Holding callback sem assures we don't alter the callback
637 * lists when doing this, and serializes ourselves with other
638 * processes wanting callbacks. */
639 down_write(&o2hb_callback_sem);
640
641 spin_lock(&o2hb_live_lock);
642 while (!list_empty(&o2hb_node_events)
643 && !list_empty(&queued_event->hn_item)) {
644 event = list_entry(o2hb_node_events.next,
645 struct o2hb_node_event,
646 hn_item);
647 list_del_init(&event->hn_item);
648 spin_unlock(&o2hb_live_lock);
649
650 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
651 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
652 event->hn_node_num);
653
654 hbcall = hbcall_from_type(event->hn_event_type);
655
656 /* We should *never* have gotten on to the list with a
657 * bad type... This isn't something that we should try
658 * to recover from. */
659 BUG_ON(IS_ERR(hbcall));
660
661 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
662
663 spin_lock(&o2hb_live_lock);
664 }
665 spin_unlock(&o2hb_live_lock);
666
667 up_write(&o2hb_callback_sem);
668}
669
670static void o2hb_queue_node_event(struct o2hb_node_event *event,
671 enum o2hb_callback_type type,
672 struct o2nm_node *node,
673 int node_num)
674{
675 assert_spin_locked(&o2hb_live_lock);
676
0e105d37
SM		 677 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
678
a7f6a5fb
MF		 679 event->hn_event_type = type;
680 event->hn_node = node;
681 event->hn_node_num = node_num;
682
683 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
684 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
685
686 list_add_tail(&event->hn_item, &o2hb_node_events);
687}
688
689static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
690{
691 struct o2hb_node_event event =
692 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
693 struct o2nm_node *node;
6f8648e8 694 int queued = 0;
a7f6a5fb
MF		 695
696 node = o2nm_get_node_by_num(slot->ds_node_num);
697 if (!node)
698 return;
699
700 spin_lock(&o2hb_live_lock);
701 if (!list_empty(&slot->ds_live_item)) {
702 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
703 slot->ds_node_num);
704
705 list_del_init(&slot->ds_live_item);
706
707 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
708 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
709
710 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
711 slot->ds_node_num);
6f8648e8 712 queued = 1;
a7f6a5fb
MF		 713 }
714 }
715 spin_unlock(&o2hb_live_lock);
716
6f8648e8
J		 717 if (queued)
718 o2hb_run_event_list(&event);
a7f6a5fb
MF		 719
720 o2nm_node_put(node);
721}
722
d2eece37 723static void o2hb_set_quorum_device(struct o2hb_region *reg)
43182d2a 724{
43182d2a
SM		 725 if (!o2hb_global_heartbeat_active())
726 return;
727
d2eece37
SM		 728 /* Prevent race with o2hb_heartbeat_group_drop_item() */
729 if (kthread_should_stop())
43182d2a
SM		 730 return;
731
d2eece37
SM		 732 /* Tag region as quorum only after thread reaches steady state */
733 if (atomic_read(&reg->hr_steady_iterations) != 0)
734 return;
735
736 spin_lock(&o2hb_live_lock);
737
738 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
739 goto unlock;
740
43182d2a
SM		 741 /*
742 * A region can be added to the quorum only when it sees all
743 * live nodes heartbeat on it. In other words, the region has been
744 * added to all nodes.
745 */
746 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
747 sizeof(o2hb_live_node_bitmap)))
d2eece37 748 goto unlock;
43182d2a 749
d2eece37
SM		 750 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
751 config_item_name(&reg->hr_item), reg->hr_dev_name);
43182d2a
SM		 752
753 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
58a3158a
SM		 754
755 /*
756 * If global heartbeat active, unpin all regions if the
757 * region count > CUT_OFF
758 */
a8f70de3 759 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 760 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
761 o2hb_region_unpin(NULL);
d2eece37
SM		 762unlock:
763 spin_unlock(&o2hb_live_lock);
43182d2a
SM		 764}
765
a7f6a5fb
MF		 766static int o2hb_check_slot(struct o2hb_region *reg,
767 struct o2hb_disk_slot *slot)
768{
769 int changed = 0, gen_changed = 0;
770 struct o2hb_node_event event =
771 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
772 struct o2nm_node *node;
773 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
774 u64 cputime;
0db638f4
MF		 775 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
776 unsigned int slot_dead_ms;
0e105d37 777 int tmp;
6f8648e8 778 int queued = 0;
a7f6a5fb
MF		 779
780 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
781
0e105d37
SM		 782 /*
783 * If a node is no longer configured but is still in the livemap, we
784 * may need to clear that bit from the livemap.
785 */
a7f6a5fb 786 node = o2nm_get_node_by_num(slot->ds_node_num);
0e105d37
SM		 787 if (!node) {
788 spin_lock(&o2hb_live_lock);
789 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
790 spin_unlock(&o2hb_live_lock);
791 if (!tmp)
792 return 0;
793 }
a7f6a5fb
MF		 794
795 if (!o2hb_verify_crc(reg, hb_block)) {
796 /* all paths from here will drop o2hb_live_lock for
797 * us. */
798 spin_lock(&o2hb_live_lock);
799
800 /* Don't print an error on the console in this case -
801 * a freshly formatted heartbeat area will not have a
802 * crc set on it. */
803 if (list_empty(&slot->ds_live_item))
804 goto out;
805
806 /* The node is live but pushed out a bad crc. We
807 * consider it a transient miss but don't populate any
808 * other values as they may be junk. */
809 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
810 slot->ds_node_num, reg->hr_dev_name);
811 o2hb_dump_slot(hb_block);
812
813 slot->ds_equal_samples++;
814 goto fire_callbacks;
815 }
816
817 /* we don't care if these wrap.. the state transitions below
818 * clear at the right places */
819 cputime = le64_to_cpu(hb_block->hb_seq);
820 if (slot->ds_last_time != cputime)
821 slot->ds_changed_samples++;
822 else
823 slot->ds_equal_samples++;
824 slot->ds_last_time = cputime;
825
826 /* The node changed heartbeat generations. We assume this to
827 * mean it dropped off but came back before we timed out. We
828 * want to consider it down for the time being but don't want
829 * to lose any changed_samples state we might build up to
830 * considering it live again. */
831 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
832 gen_changed = 1;
833 slot->ds_equal_samples = 0;
70bacbdb
MF		 834 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
835 "to 0x%llx)\n", slot->ds_node_num,
836 (long long)slot->ds_last_generation,
837 (long long)le64_to_cpu(hb_block->hb_generation));
a7f6a5fb
MF		 838 }
839
840 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
841
70bacbdb
MF		 842 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
843 "seq %llu last %llu changed %u equal %u\n",
844 slot->ds_node_num, (long long)slot->ds_last_generation,
845 le32_to_cpu(hb_block->hb_cksum),
2bd63216 846 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
70bacbdb 847 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
a7f6a5fb
MF		 848 slot->ds_equal_samples);
849
850 spin_lock(&o2hb_live_lock);
851
852fire_callbacks:
853 /* dead nodes only come to life after some number of
854 * changes at any time during their dead time */
855 if (list_empty(&slot->ds_live_item) &&
856 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
70bacbdb
MF		 857 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
858 slot->ds_node_num, (long long)slot->ds_last_generation);
a7f6a5fb 859
823a637a
SM		 860 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
861
a7f6a5fb
MF		 862 /* first on the list generates a callback */
863 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
d6aa1c7c
SM		 864 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
865 "bitmap\n", slot->ds_node_num);
a7f6a5fb
MF		 866 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
867
868 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
869 slot->ds_node_num);
870
871 changed = 1;
6f8648e8 872 queued = 1;
a7f6a5fb
MF		 873 }
874
875 list_add_tail(&slot->ds_live_item,
876 &o2hb_live_slots[slot->ds_node_num]);
877
878 slot->ds_equal_samples = 0;
0db638f4
MF		 879
880 /* We want to be sure that all nodes agree on the
881 * number of milliseconds before a node will be
882 * considered dead. The self-fencing timeout is
883 * computed from this value, and a discrepancy might
884 * result in heartbeat calling a node dead when it
885 * hasn't self-fenced yet. */
886 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
887 if (slot_dead_ms && slot_dead_ms != dead_ms) {
888 /* TODO: Perhaps we can fail the region here. */
889 mlog(ML_ERROR, "Node %d on device %s has a dead count "
890 "of %u ms, but our count is %u ms.\n"
891 "Please double check your configuration values "
892 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
893 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
894 dead_ms);
895 }
a7f6a5fb
MF		 896 goto out;
897 }
898
899 /* if the list is dead, we're done.. */
900 if (list_empty(&slot->ds_live_item))
901 goto out;
902
903 /* live nodes only go dead after enough consequtive missed
904 * samples.. reset the missed counter whenever we see
905 * activity */
906 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
907 mlog(ML_HEARTBEAT, "Node %d left my region\n",
908 slot->ds_node_num);
909
823a637a
SM		 910 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
911
a7f6a5fb
MF		 912 /* last off the live_slot generates a callback */
913 list_del_init(&slot->ds_live_item);
914 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
d6aa1c7c
SM		 915 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
916 "nodes bitmap\n", slot->ds_node_num);
a7f6a5fb
MF		 917 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
918
0e105d37
SM		 919 /* node can be null */
920 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
921 node, slot->ds_node_num);
a7f6a5fb
MF		 922
923 changed = 1;
6f8648e8 924 queued = 1;
a7f6a5fb
MF		 925 }
926
927 /* We don't clear this because the node is still
928 * actually writing new blocks. */
929 if (!gen_changed)
930 slot->ds_changed_samples = 0;
931 goto out;
932 }
933 if (slot->ds_changed_samples) {
934 slot->ds_changed_samples = 0;
935 slot->ds_equal_samples = 0;
936 }
937out:
938 spin_unlock(&o2hb_live_lock);
939
6f8648e8
J		 940 if (queued)
941 o2hb_run_event_list(&event);
a7f6a5fb 942
0e105d37
SM		 943 if (node)
944 o2nm_node_put(node);
a7f6a5fb
MF		 945 return changed;
946}
947
518df6bd 948static int o2hb_highest_node(unsigned long *nodes, int numbits)
a7f6a5fb 949{
518df6bd 950 return find_last_bit(nodes, numbits);
a7f6a5fb
MF		 951}
952
a9e2ae39 953static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
a7f6a5fb 954{
d2eece37
SM		 955 int i, ret, highest_node;
956 int membership_change = 0, own_slot_ok = 0;
a7f6a5fb 957 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
0e105d37 958 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
a7f6a5fb
MF		 959 struct o2hb_bio_wait_ctxt write_wc;
960
a9e2ae39
MF		 961 ret = o2nm_configured_node_map(configured_nodes,
962 sizeof(configured_nodes));
963 if (ret) {
964 mlog_errno(ret);
d2eece37 965 goto bail;
a9e2ae39 966 }
a7f6a5fb 967
0e105d37
SM		 968 /*
969 * If a node is not configured but is in the livemap, we still need
970 * to read the slot so as to be able to remove it from the livemap.
971 */
972 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
973 i = -1;
974 while ((i = find_next_bit(live_node_bitmap,
975 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
976 set_bit(i, configured_nodes);
977 }
978
a7f6a5fb
MF		 979 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
980 if (highest_node >= O2NM_MAX_NODES) {
d2eece37
SM		 981 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
982 ret = -EINVAL;
983 goto bail;
a7f6a5fb
MF		 984 }
985
986 /* No sense in reading the slots of nodes that don't exist
987 * yet. Of course, if the node definitions have holes in them
988 * then we're reading an empty slot anyway... Consider this
989 * best-effort. */
990 ret = o2hb_read_slots(reg, highest_node + 1);
991 if (ret < 0) {
992 mlog_errno(ret);
d2eece37 993 goto bail;
a7f6a5fb
MF		 994 }
995
996 /* With an up to date view of the slots, we can check that no
997 * other node has been improperly configured to heartbeat in
998 * our slot. */
d2eece37 999 own_slot_ok = o2hb_check_own_slot(reg);
a7f6a5fb
MF		 1000
1001 /* fill in the proper info for our next heartbeat */
1002 o2hb_prepare_block(reg, reg->hr_generation);
1003
b559292e 1004 ret = o2hb_issue_node_write(reg, &write_wc);
a7f6a5fb
MF		 1005 if (ret < 0) {
1006 mlog_errno(ret);
d2eece37 1007 goto bail;
a7f6a5fb
MF		 1008 }
1009
1010 i = -1;
33c12a54
SM		 1011 while((i = find_next_bit(configured_nodes,
1012 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
d2eece37 1013 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
a7f6a5fb
MF		 1014 }
1015
1016 /*
1017 * We have to be sure we've advertised ourselves on disk
1018 * before we can go to steady state. This ensures that
1019 * people we find in our steady state have seen us.
1020 */
1021 o2hb_wait_on_io(reg, &write_wc);
a9e2ae39
MF		 1022 if (write_wc.wc_error) {
1023 /* Do not re-arm the write timeout on I/O error - we
1024 * can't be sure that the new block ever made it to
1025 * disk */
1026 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1027 write_wc.wc_error, reg->hr_dev_name);
d2eece37
SM		 1028 ret = write_wc.wc_error;
1029 goto bail;
a9e2ae39
MF		 1030 }
1031
d2eece37
SM		 1032 /* Skip disarming the timeout if own slot has stale/bad data */
1033 if (own_slot_ok) {
1034 o2hb_set_quorum_device(reg);
1035 o2hb_arm_write_timeout(reg);
1036 }
a7f6a5fb 1037
d2eece37 1038bail:
a7f6a5fb 1039 /* let the person who launched us know when things are steady */
d2eece37
SM		 1040 if (atomic_read(&reg->hr_steady_iterations) != 0) {
1041 if (!ret && own_slot_ok && !membership_change) {
1042 if (atomic_dec_and_test(&reg->hr_steady_iterations))
1043 wake_up(&o2hb_steady_queue);
1044 }
1045 }
1046
1047 if (atomic_read(&reg->hr_steady_iterations) != 0) {
1048 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1049 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1050 "heartbeart on region %s (%s)\n",
1051 config_item_name(&reg->hr_item),
1052 reg->hr_dev_name);
1053 atomic_set(&reg->hr_steady_iterations, 0);
1054 reg->hr_aborted_start = 1;
a7f6a5fb 1055 wake_up(&o2hb_steady_queue);
d2eece37
SM		 1056 ret = -EIO;
1057 }
a7f6a5fb 1058 }
a9e2ae39 1059
d2eece37 1060 return ret;
a7f6a5fb
MF		 1061}
1062
1063/* Subtract b from a, storing the result in a. a *must* have a larger
1064 * value than b. */
1065static void o2hb_tv_subtract(struct timeval *a,
1066 struct timeval *b)
1067{
1068 /* just return 0 when a is after b */
1069 if (a->tv_sec < b->tv_sec ||
1070 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1071 a->tv_sec = 0;
1072 a->tv_usec = 0;
1073 return;
1074 }
1075
1076 a->tv_sec -= b->tv_sec;
1077 a->tv_usec -= b->tv_usec;
1078 while ( a->tv_usec < 0 ) {
1079 a->tv_sec--;
1080 a->tv_usec += 1000000;
1081 }
1082}
1083
1084static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1085 struct timeval *end)
1086{
1087 struct timeval res = *end;
1088
1089 o2hb_tv_subtract(&res, start);
1090
1091 return res.tv_sec * 1000 + res.tv_usec / 1000;
1092}
1093
1094/*
1095 * we ride the region ref that the region dir holds. before the region
1096 * dir is removed and drops it ref it will wait to tear down this
1097 * thread.
1098 */
1099static int o2hb_thread(void *data)
1100{
1101 int i, ret;
1102 struct o2hb_region *reg = data;
a7f6a5fb
MF		 1103 struct o2hb_bio_wait_ctxt write_wc;
1104 struct timeval before_hb, after_hb;
1105 unsigned int elapsed_msec;
1106
1107 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1108
8698a745 1109 set_user_nice(current, MIN_NICE);
a7f6a5fb 1110
cfc069d3
SM		 1111 /* Pin node */
1112 o2nm_depend_this_node();
1113
d2eece37
SM		 1114 while (!kthread_should_stop() &&
1115 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
a7f6a5fb 1116 /* We track the time spent inside
025dfdaf 1117 * o2hb_do_disk_heartbeat so that we avoid more than
a7f6a5fb
MF		 1118 * hr_timeout_ms between disk writes. On busy systems
1119 * this should result in a heartbeat which is less
1120 * likely to time itself out. */
1121 do_gettimeofday(&before_hb);
1122
d2eece37 1123 ret = o2hb_do_disk_heartbeat(reg);
a7f6a5fb
MF		 1124
1125 do_gettimeofday(&after_hb);
1126 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1127
b31d308d 1128 mlog(ML_HEARTBEAT,
f5425fce 1129 "start = %lu.%lu, end = %lu.%lu, msec = %u, ret = %d\n",
215c7f9f
MF		 1130 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1131 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
f5425fce 1132 elapsed_msec, ret);
a7f6a5fb 1133
d2eece37
SM		 1134 if (!kthread_should_stop() &&
1135 elapsed_msec < reg->hr_timeout_ms) {
a7f6a5fb
MF		 1136 /* the kthread api has blocked signals for us so no
1137 * need to record the return value. */
1138 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1139 }
1140 }
1141
1142 o2hb_disarm_write_timeout(reg);
1143
1144 /* unclean stop is only used in very bad situation */
1145 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1146 o2hb_shutdown_slot(&reg->hr_slots[i]);
1147
1148 /* Explicit down notification - avoid forcing the other nodes
1149 * to timeout on this region when we could just as easily
1150 * write a clear generation - thus indicating to them that
1151 * this node has left this region.
d2eece37
SM		 1152 */
1153 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1154 o2hb_prepare_block(reg, 0);
1155 ret = o2hb_issue_node_write(reg, &write_wc);
1156 if (ret == 0)
1157 o2hb_wait_on_io(reg, &write_wc);
1158 else
1159 mlog_errno(ret);
a7f6a5fb
MF		 1160 }
1161
cfc069d3
SM		 1162 /* Unpin node */
1163 o2nm_undepend_this_node();
a7f6a5fb 1164
d2eece37 1165 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
a7f6a5fb
MF		 1166
1167 return 0;
1168}
1169
87d3d3f3
SM		 1170#ifdef CONFIG_DEBUG_FS
1171static int o2hb_debug_open(struct inode *inode, struct file *file)
1172{
8ca8b0bb 1173 struct o2hb_debug_buf *db = inode->i_private;
1f285305 1174 struct o2hb_region *reg;
87d3d3f3 1175 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
bb570a5d 1176 unsigned long lts;
87d3d3f3
SM		 1177 char *buf = NULL;
1178 int i = -1;
1179 int out = 0;
1180
8ca8b0bb
SM		 1181 /* max_nodes should be the largest bitmap we pass here */
1182 BUG_ON(sizeof(map) < db->db_size);
1183
87d3d3f3
SM		 1184 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1185 if (!buf)
1186 goto bail;
1187
8ca8b0bb
SM		 1188 switch (db->db_type) {
1189 case O2HB_DB_TYPE_LIVENODES:
a6de0136
SM		 1190 case O2HB_DB_TYPE_LIVEREGIONS:
1191 case O2HB_DB_TYPE_QUORUMREGIONS:
1192 case O2HB_DB_TYPE_FAILEDREGIONS:
8ca8b0bb
SM		 1193 spin_lock(&o2hb_live_lock);
1194 memcpy(map, db->db_data, db->db_size);
1195 spin_unlock(&o2hb_live_lock);
1196 break;
87d3d3f3 1197
1f285305
SM		 1198 case O2HB_DB_TYPE_REGION_LIVENODES:
1199 spin_lock(&o2hb_live_lock);
1200 reg = (struct o2hb_region *)db->db_data;
1201 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1202 spin_unlock(&o2hb_live_lock);
1203 break;
1204
1205 case O2HB_DB_TYPE_REGION_NUMBER:
1206 reg = (struct o2hb_region *)db->db_data;
1207 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1208 reg->hr_region_num);
1209 goto done;
1210
43695d09
SM		 1211 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1212 reg = (struct o2hb_region *)db->db_data;
bb570a5d
SM		 1213 lts = reg->hr_last_timeout_start;
1214 /* If 0, it has never been set before */
1215 if (lts)
1216 lts = jiffies_to_msecs(jiffies - lts);
1217 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
43695d09
SM		 1218 goto done;
1219
cb0586bd
SM		 1220 case O2HB_DB_TYPE_REGION_PINNED:
1221 reg = (struct o2hb_region *)db->db_data;
1222 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1223 !!reg->hr_item_pinned);
1224 goto done;
1225
8ca8b0bb
SM		 1226 default:
1227 goto done;
1228 }
1229
1230 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
87d3d3f3
SM		 1231 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1232 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1233
8ca8b0bb 1234done:
87d3d3f3
SM		 1235 i_size_write(inode, out);
1236
1237 file->private_data = buf;
1238
1239 return 0;
1240bail:
1241 return -ENOMEM;
1242}
1243
1244static int o2hb_debug_release(struct inode *inode, struct file *file)
1245{
1246 kfree(file->private_data);
1247 return 0;
1248}
1249
1250static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1251 size_t nbytes, loff_t *ppos)
1252{
1253 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1254 i_size_read(file->f_mapping->host));
1255}
1256#else
1257static int o2hb_debug_open(struct inode *inode, struct file *file)
1258{
1259 return 0;
1260}
1261static int o2hb_debug_release(struct inode *inode, struct file *file)
1262{
1263 return 0;
1264}
1265static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1266 size_t nbytes, loff_t *ppos)
1267{
1268 return 0;
1269}
1270#endif /* CONFIG_DEBUG_FS */
1271
828c0950 1272static const struct file_operations o2hb_debug_fops = {
87d3d3f3
SM		 1273 .open = o2hb_debug_open,
1274 .release = o2hb_debug_release,
1275 .read = o2hb_debug_read,
1276 .llseek = generic_file_llseek,
1277};
1278
1279void o2hb_exit(void)
1280{
8ca8b0bb 1281 kfree(o2hb_db_livenodes);
a6de0136
SM		 1282 kfree(o2hb_db_liveregions);
1283 kfree(o2hb_db_quorumregions);
1284 kfree(o2hb_db_failedregions);
1285 debugfs_remove(o2hb_debug_failedregions);
1286 debugfs_remove(o2hb_debug_quorumregions);
1287 debugfs_remove(o2hb_debug_liveregions);
8ca8b0bb
SM		 1288 debugfs_remove(o2hb_debug_livenodes);
1289 debugfs_remove(o2hb_debug_dir);
1290}
1291
1292static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1293 struct o2hb_debug_buf **db, int db_len,
1294 int type, int size, int len, void *data)
1295{
1296 *db = kmalloc(db_len, GFP_KERNEL);
1297 if (!*db)
1298 return NULL;
1299
1300 (*db)->db_type = type;
1301 (*db)->db_size = size;
1302 (*db)->db_len = len;
1303 (*db)->db_data = data;
1304
1305 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1306 &o2hb_debug_fops);
1307}
1308
1309static int o2hb_debug_init(void)
1310{
1311 int ret = -ENOMEM;
1312
1313 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
8f443e23 1314 if (!o2hb_debug_dir) {
8ca8b0bb
SM		 1315 mlog_errno(ret);
1316 goto bail;
1317 }
1318
1319 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1320 o2hb_debug_dir,
1321 &o2hb_db_livenodes,
1322 sizeof(*o2hb_db_livenodes),
1323 O2HB_DB_TYPE_LIVENODES,
1324 sizeof(o2hb_live_node_bitmap),
1325 O2NM_MAX_NODES,
1326 o2hb_live_node_bitmap);
8f443e23 1327 if (!o2hb_debug_livenodes) {
8ca8b0bb
SM		 1328 mlog_errno(ret);
1329 goto bail;
1330 }
a6de0136
SM		 1331
1332 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1333 o2hb_debug_dir,
1334 &o2hb_db_liveregions,
1335 sizeof(*o2hb_db_liveregions),
1336 O2HB_DB_TYPE_LIVEREGIONS,
1337 sizeof(o2hb_live_region_bitmap),
1338 O2NM_MAX_REGIONS,
1339 o2hb_live_region_bitmap);
8f443e23 1340 if (!o2hb_debug_liveregions) {
a6de0136
SM		 1341 mlog_errno(ret);
1342 goto bail;
1343 }
1344
1345 o2hb_debug_quorumregions =
1346 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1347 o2hb_debug_dir,
1348 &o2hb_db_quorumregions,
1349 sizeof(*o2hb_db_quorumregions),
1350 O2HB_DB_TYPE_QUORUMREGIONS,
1351 sizeof(o2hb_quorum_region_bitmap),
1352 O2NM_MAX_REGIONS,
1353 o2hb_quorum_region_bitmap);
8f443e23 1354 if (!o2hb_debug_quorumregions) {
a6de0136
SM		 1355 mlog_errno(ret);
1356 goto bail;
1357 }
1358
1359 o2hb_debug_failedregions =
1360 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1361 o2hb_debug_dir,
1362 &o2hb_db_failedregions,
1363 sizeof(*o2hb_db_failedregions),
1364 O2HB_DB_TYPE_FAILEDREGIONS,
1365 sizeof(o2hb_failed_region_bitmap),
1366 O2NM_MAX_REGIONS,
1367 o2hb_failed_region_bitmap);
8f443e23 1368 if (!o2hb_debug_failedregions) {
a6de0136
SM		 1369 mlog_errno(ret);
1370 goto bail;
1371 }
1372
8ca8b0bb
SM		 1373 ret = 0;
1374bail:
1375 if (ret)
1376 o2hb_exit();
1377
1378 return ret;
87d3d3f3
SM		 1379}
1380
1381int o2hb_init(void)
a7f6a5fb
MF		 1382{
1383 int i;
1384
1385 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1386 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1387
1388 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1389 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1390
1391 INIT_LIST_HEAD(&o2hb_node_events);
1392
1393 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
536f0741 1394 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
e7d656ba 1395 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
43182d2a 1396 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
b1c5ebfb 1397 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
87d3d3f3 1398
58a3158a
SM		 1399 o2hb_dependent_users = 0;
1400
8ca8b0bb 1401 return o2hb_debug_init();
a7f6a5fb
MF		 1402}
1403
1404/* if we're already in a callback then we're already serialized by the sem */
1405static void o2hb_fill_node_map_from_callback(unsigned long *map,
1406 unsigned bytes)
1407{
1408 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1409
1410 memcpy(map, &o2hb_live_node_bitmap, bytes);
1411}
1412
1413/*
1414 * get a map of all nodes that are heartbeating in any regions
1415 */
1416void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1417{
1418 /* callers want to serialize this map and callbacks so that they
1419 * can trust that they don't miss nodes coming to the party */
1420 down_read(&o2hb_callback_sem);
1421 spin_lock(&o2hb_live_lock);
1422 o2hb_fill_node_map_from_callback(map, bytes);
1423 spin_unlock(&o2hb_live_lock);
1424 up_read(&o2hb_callback_sem);
1425}
1426EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1427
1428/*
1429 * heartbeat configfs bits. The heartbeat set is a default set under
1430 * the cluster set in nodemanager.c.
1431 */
1432
1433static struct o2hb_region *to_o2hb_region(struct config_item *item)
1434{
1435 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1436}
1437
1438/* drop_item only drops its ref after killing the thread, nothing should
1439 * be using the region anymore. this has to clean up any state that
1440 * attributes might have built up. */
1441static void o2hb_region_release(struct config_item *item)
1442{
1443 int i;
1444 struct page *page;
1445 struct o2hb_region *reg = to_o2hb_region(item);
1446
d2eece37
SM		 1447 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1448
d787ab09 1449 kfree(reg->hr_tmp_block);
a7f6a5fb
MF		 1450
1451 if (reg->hr_slot_data) {
1452 for (i = 0; i < reg->hr_num_pages; i++) {
1453 page = reg->hr_slot_data[i];
1454 if (page)
1455 __free_page(page);
1456 }
1457 kfree(reg->hr_slot_data);
1458 }
1459
1460 if (reg->hr_bdev)
9a1c3542 1461 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb 1462
d787ab09 1463 kfree(reg->hr_slots);
a7f6a5fb 1464
1f285305
SM		 1465 kfree(reg->hr_db_regnum);
1466 kfree(reg->hr_db_livenodes);
1467 debugfs_remove(reg->hr_debug_livenodes);
1468 debugfs_remove(reg->hr_debug_regnum);
d4396eaf 1469 debugfs_remove(reg->hr_debug_elapsed_time);
cb0586bd 1470 debugfs_remove(reg->hr_debug_pinned);
1f285305
SM		 1471 debugfs_remove(reg->hr_debug_dir);
1472
a7f6a5fb
MF		 1473 spin_lock(&o2hb_live_lock);
1474 list_del(&reg->hr_all_item);
1475 spin_unlock(&o2hb_live_lock);
1476
1477 kfree(reg);
1478}
1479
1480static int o2hb_read_block_input(struct o2hb_region *reg,
1481 const char *page,
1482 size_t count,
1483 unsigned long *ret_bytes,
1484 unsigned int *ret_bits)
1485{
1486 unsigned long bytes;
1487 char *p = (char *)page;
1488
1489 bytes = simple_strtoul(p, &p, 0);
1490 if (!p || (*p && (*p != '\n')))
1491 return -EINVAL;
1492
1493 /* Heartbeat and fs min / max block sizes are the same. */
1494 if (bytes > 4096 || bytes < 512)
1495 return -ERANGE;
1496 if (hweight16(bytes) != 1)
1497 return -EINVAL;
1498
1499 if (ret_bytes)
1500 *ret_bytes = bytes;
1501 if (ret_bits)
1502 *ret_bits = ffs(bytes) - 1;
1503
1504 return 0;
1505}
1506
1507static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1508 char *page)
1509{
1510 return sprintf(page, "%u\n", reg->hr_block_bytes);
1511}
1512
1513static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1514 const char *page,
1515 size_t count)
1516{
1517 int status;
1518 unsigned long block_bytes;
1519 unsigned int block_bits;
1520
1521 if (reg->hr_bdev)
1522 return -EINVAL;
1523
1524 status = o2hb_read_block_input(reg, page, count,
1525 &block_bytes, &block_bits);
1526 if (status)
1527 return status;
1528
1529 reg->hr_block_bytes = (unsigned int)block_bytes;
1530 reg->hr_block_bits = block_bits;
1531
1532 return count;
1533}
1534
1535static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1536 char *page)
1537{
1538 return sprintf(page, "%llu\n", reg->hr_start_block);
1539}
1540
1541static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1542 const char *page,
1543 size_t count)
1544{
1545 unsigned long long tmp;
1546 char *p = (char *)page;
1547
1548 if (reg->hr_bdev)
1549 return -EINVAL;
1550
1551 tmp = simple_strtoull(p, &p, 0);
1552 if (!p || (*p && (*p != '\n')))
1553 return -EINVAL;
1554
1555 reg->hr_start_block = tmp;
1556
1557 return count;
1558}
1559
1560static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1561 char *page)
1562{
1563 return sprintf(page, "%d\n", reg->hr_blocks);
1564}
1565
1566static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1567 const char *page,
1568 size_t count)
1569{
1570 unsigned long tmp;
1571 char *p = (char *)page;
1572
1573 if (reg->hr_bdev)
1574 return -EINVAL;
1575
1576 tmp = simple_strtoul(p, &p, 0);
1577 if (!p || (*p && (*p != '\n')))
1578 return -EINVAL;
1579
1580 if (tmp > O2NM_MAX_NODES || tmp == 0)
1581 return -ERANGE;
1582
1583 reg->hr_blocks = (unsigned int)tmp;
1584
1585 return count;
1586}
1587
1588static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1589 char *page)
1590{
1591 unsigned int ret = 0;
1592
1593 if (reg->hr_bdev)
1594 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1595
1596 return ret;
1597}
1598
1599static void o2hb_init_region_params(struct o2hb_region *reg)
1600{
1601 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1602 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1603
1604 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1605 reg->hr_start_block, reg->hr_blocks);
1606 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1607 reg->hr_block_bytes, reg->hr_block_bits);
1608 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1609 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1610}
1611
1612static int o2hb_map_slot_data(struct o2hb_region *reg)
1613{
1614 int i, j;
1615 unsigned int last_slot;
1616 unsigned int spp = reg->hr_slots_per_page;
1617 struct page *page;
1618 char *raw;
1619 struct o2hb_disk_slot *slot;
1620
1621 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1622 if (reg->hr_tmp_block == NULL) {
1623 mlog_errno(-ENOMEM);
1624 return -ENOMEM;
1625 }
1626
1627 reg->hr_slots = kcalloc(reg->hr_blocks,
1628 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1629 if (reg->hr_slots == NULL) {
1630 mlog_errno(-ENOMEM);
1631 return -ENOMEM;
1632 }
1633
1634 for(i = 0; i < reg->hr_blocks; i++) {
1635 slot = &reg->hr_slots[i];
1636 slot->ds_node_num = i;
1637 INIT_LIST_HEAD(&slot->ds_live_item);
1638 slot->ds_raw_block = NULL;
1639 }
1640
1641 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1642 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1643 "at %u blocks per page\n",
1644 reg->hr_num_pages, reg->hr_blocks, spp);
1645
1646 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1647 GFP_KERNEL);
1648 if (!reg->hr_slot_data) {
1649 mlog_errno(-ENOMEM);
1650 return -ENOMEM;
1651 }
1652
1653 for(i = 0; i < reg->hr_num_pages; i++) {
1654 page = alloc_page(GFP_KERNEL);
1655 if (!page) {
1656 mlog_errno(-ENOMEM);
1657 return -ENOMEM;
1658 }
1659
1660 reg->hr_slot_data[i] = page;
1661
1662 last_slot = i * spp;
1663 raw = page_address(page);
1664 for (j = 0;
1665 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1666 j++) {
1667 BUG_ON((j + last_slot) >= reg->hr_blocks);
1668
1669 slot = &reg->hr_slots[j + last_slot];
1670 slot->ds_raw_block =
1671 (struct o2hb_disk_heartbeat_block *) raw;
1672
1673 raw += reg->hr_block_bytes;
1674 }
1675 }
1676
1677 return 0;
1678}
1679
1680/* Read in all the slots available and populate the tracking
1681 * structures so that we can start with a baseline idea of what's
1682 * there. */
1683static int o2hb_populate_slot_data(struct o2hb_region *reg)
1684{
1685 int ret, i;
1686 struct o2hb_disk_slot *slot;
1687 struct o2hb_disk_heartbeat_block *hb_block;
1688
a7f6a5fb
MF		 1689 ret = o2hb_read_slots(reg, reg->hr_blocks);
1690 if (ret) {
1691 mlog_errno(ret);
1692 goto out;
1693 }
1694
1695 /* We only want to get an idea of the values initially in each
1696 * slot, so we do no verification - o2hb_check_slot will
1697 * actually determine if each configured slot is valid and
1698 * whether any values have changed. */
1699 for(i = 0; i < reg->hr_blocks; i++) {
1700 slot = &reg->hr_slots[i];
1701 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1702
1703 /* Only fill the values that o2hb_check_slot uses to
1704 * determine changing slots */
1705 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1706 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1707 }
1708
1709out:
a7f6a5fb
MF		 1710 return ret;
1711}
1712
1713/* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1714static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1715 const char *page,
1716 size_t count)
1717{
e6c352db 1718 struct task_struct *hb_task;
a7f6a5fb
MF		 1719 long fd;
1720 int sectsize;
1721 char *p = (char *)page;
2903ff01
AV		 1722 struct fd f;
1723 struct inode *inode;
a7f6a5fb 1724 ssize_t ret = -EINVAL;
76d9fc29 1725 int live_threshold;
a7f6a5fb
MF		 1726
1727 if (reg->hr_bdev)
1728 goto out;
1729
1730 /* We can't heartbeat without having had our node number
1731 * configured yet. */
1732 if (o2nm_this_node() == O2NM_MAX_NODES)
1733 goto out;
1734
1735 fd = simple_strtol(p, &p, 0);
1736 if (!p || (*p && (*p != '\n')))
1737 goto out;
1738
1739 if (fd < 0 || fd >= INT_MAX)
1740 goto out;
1741
2903ff01
AV		 1742 f = fdget(fd);
1743 if (f.file == NULL)
a7f6a5fb
MF		 1744 goto out;
1745
1746 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1747 reg->hr_block_bytes == 0)
2903ff01 1748 goto out2;
a7f6a5fb 1749
2903ff01 1750 inode = igrab(f.file->f_mapping->host);
a7f6a5fb 1751 if (inode == NULL)
2903ff01 1752 goto out2;
a7f6a5fb
MF		 1753
1754 if (!S_ISBLK(inode->i_mode))
2903ff01 1755 goto out3;
a7f6a5fb 1756
2903ff01 1757 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
e525fd89 1758 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
a7f6a5fb
MF		 1759 if (ret) {
1760 reg->hr_bdev = NULL;
2903ff01 1761 goto out3;
a7f6a5fb
MF		 1762 }
1763 inode = NULL;
1764
1765 bdevname(reg->hr_bdev, reg->hr_dev_name);
1766
e1defc4f 1767 sectsize = bdev_logical_block_size(reg->hr_bdev);
a7f6a5fb
MF		 1768 if (sectsize != reg->hr_block_bytes) {
1769 mlog(ML_ERROR,
1770 "blocksize %u incorrect for device, expected %d",
1771 reg->hr_block_bytes, sectsize);
1772 ret = -EINVAL;
2903ff01 1773 goto out3;
a7f6a5fb
MF		 1774 }
1775
1776 o2hb_init_region_params(reg);
1777
1778 /* Generation of zero is invalid */
1779 do {
1780 get_random_bytes(&reg->hr_generation,
1781 sizeof(reg->hr_generation));
1782 } while (reg->hr_generation == 0);
1783
1784 ret = o2hb_map_slot_data(reg);
1785 if (ret) {
1786 mlog_errno(ret);
2903ff01 1787 goto out3;
a7f6a5fb
MF		 1788 }
1789
1790 ret = o2hb_populate_slot_data(reg);
1791 if (ret) {
1792 mlog_errno(ret);
2903ff01 1793 goto out3;
a7f6a5fb
MF		 1794 }
1795
c4028958 1796 INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
a7f6a5fb
MF		 1797
1798 /*
1799 * A node is considered live after it has beat LIVE_THRESHOLD
1800 * times. We're not steady until we've given them a chance
1801 * _after_ our first read.
76d9fc29
SM		 1802 * The default threshold is bare minimum so as to limit the delay
1803 * during mounts. For global heartbeat, the threshold doubled for the
1804 * first region.
a7f6a5fb 1805 */
76d9fc29
SM		 1806 live_threshold = O2HB_LIVE_THRESHOLD;
1807 if (o2hb_global_heartbeat_active()) {
1808 spin_lock(&o2hb_live_lock);
a8f70de3 1809 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
76d9fc29
SM		 1810 live_threshold <<= 1;
1811 spin_unlock(&o2hb_live_lock);
1812 }
d2eece37
SM		 1813 ++live_threshold;
1814 atomic_set(&reg->hr_steady_iterations, live_threshold);
1815 /* unsteady_iterations is double the steady_iterations */
1816 atomic_set(&reg->hr_unsteady_iterations, (live_threshold << 1));
a7f6a5fb 1817
e6c352db
JB		 1818 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1819 reg->hr_item.ci_name);
1820 if (IS_ERR(hb_task)) {
1821 ret = PTR_ERR(hb_task);
a7f6a5fb 1822 mlog_errno(ret);
2903ff01 1823 goto out3;
a7f6a5fb
MF		 1824 }
1825
e6c352db
JB		 1826 spin_lock(&o2hb_live_lock);
1827 reg->hr_task = hb_task;
1828 spin_unlock(&o2hb_live_lock);
1829
a7f6a5fb
MF		 1830 ret = wait_event_interruptible(o2hb_steady_queue,
1831 atomic_read(&reg->hr_steady_iterations) == 0);
1832 if (ret) {
d2eece37
SM		 1833 atomic_set(&reg->hr_steady_iterations, 0);
1834 reg->hr_aborted_start = 1;
1835 }
e6c352db 1836
d2eece37
SM		 1837 if (reg->hr_aborted_start) {
1838 ret = -EIO;
2903ff01 1839 goto out3;
a7f6a5fb
MF		 1840 }
1841
e6df3a66
JB		 1842 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1843 spin_lock(&o2hb_live_lock);
1844 hb_task = reg->hr_task;
e7d656ba
SM		 1845 if (o2hb_global_heartbeat_active())
1846 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
e6df3a66
JB		 1847 spin_unlock(&o2hb_live_lock);
1848
1849 if (hb_task)
1850 ret = count;
1851 else
1852 ret = -EIO;
1853
18c50cb0 1854 if (hb_task && o2hb_global_heartbeat_active())
d2eece37
SM		 1855 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1856 config_item_name(&reg->hr_item), reg->hr_dev_name);
18c50cb0 1857
2903ff01
AV		 1858out3:
1859 iput(inode);
1860out2:
1861 fdput(f);
a7f6a5fb 1862out:
a7f6a5fb
MF		 1863 if (ret < 0) {
1864 if (reg->hr_bdev) {
9a1c3542 1865 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb
MF		 1866 reg->hr_bdev = NULL;
1867 }
1868 }
1869 return ret;
1870}
1871
92efc152
ZW		 1872static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1873 char *page)
1874{
e6c352db
JB		 1875 pid_t pid = 0;
1876
1877 spin_lock(&o2hb_live_lock);
1878 if (reg->hr_task)
ba25f9dc 1879 pid = task_pid_nr(reg->hr_task);
e6c352db
JB		 1880 spin_unlock(&o2hb_live_lock);
1881
1882 if (!pid)
92efc152
ZW		 1883 return 0;
1884
e6c352db 1885 return sprintf(page, "%u\n", pid);
92efc152
ZW		 1886}
1887
a7f6a5fb
MF		 1888struct o2hb_region_attribute {
1889 struct configfs_attribute attr;
1890 ssize_t (*show)(struct o2hb_region *, char *);
1891 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1892};
1893
1894static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1895 .attr = { .ca_owner = THIS_MODULE,
1896 .ca_name = "block_bytes",
1897 .ca_mode = S_IRUGO | S_IWUSR },
1898 .show = o2hb_region_block_bytes_read,
1899 .store = o2hb_region_block_bytes_write,
1900};
1901
1902static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1903 .attr = { .ca_owner = THIS_MODULE,
1904 .ca_name = "start_block",
1905 .ca_mode = S_IRUGO | S_IWUSR },
1906 .show = o2hb_region_start_block_read,
1907 .store = o2hb_region_start_block_write,
1908};
1909
1910static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1911 .attr = { .ca_owner = THIS_MODULE,
1912 .ca_name = "blocks",
1913 .ca_mode = S_IRUGO | S_IWUSR },
1914 .show = o2hb_region_blocks_read,
1915 .store = o2hb_region_blocks_write,
1916};
1917
1918static struct o2hb_region_attribute o2hb_region_attr_dev = {
1919 .attr = { .ca_owner = THIS_MODULE,
1920 .ca_name = "dev",
1921 .ca_mode = S_IRUGO | S_IWUSR },
1922 .show = o2hb_region_dev_read,
1923 .store = o2hb_region_dev_write,
1924};
1925
92efc152
ZW		 1926static struct o2hb_region_attribute o2hb_region_attr_pid = {
1927 .attr = { .ca_owner = THIS_MODULE,
1928 .ca_name = "pid",
1929 .ca_mode = S_IRUGO | S_IRUSR },
1930 .show = o2hb_region_pid_read,
1931};
1932
a7f6a5fb
MF		 1933static struct configfs_attribute *o2hb_region_attrs[] = {
1934 &o2hb_region_attr_block_bytes.attr,
1935 &o2hb_region_attr_start_block.attr,
1936 &o2hb_region_attr_blocks.attr,
1937 &o2hb_region_attr_dev.attr,
92efc152 1938 &o2hb_region_attr_pid.attr,
a7f6a5fb
MF		 1939 NULL,
1940};
1941
1942static ssize_t o2hb_region_show(struct config_item *item,
1943 struct configfs_attribute *attr,
1944 char *page)
1945{
1946 struct o2hb_region *reg = to_o2hb_region(item);
1947 struct o2hb_region_attribute *o2hb_region_attr =
1948 container_of(attr, struct o2hb_region_attribute, attr);
1949 ssize_t ret = 0;
1950
1951 if (o2hb_region_attr->show)
1952 ret = o2hb_region_attr->show(reg, page);
1953 return ret;
1954}
1955
1956static ssize_t o2hb_region_store(struct config_item *item,
1957 struct configfs_attribute *attr,
1958 const char *page, size_t count)
1959{
1960 struct o2hb_region *reg = to_o2hb_region(item);
1961 struct o2hb_region_attribute *o2hb_region_attr =
1962 container_of(attr, struct o2hb_region_attribute, attr);
1963 ssize_t ret = -EINVAL;
1964
1965 if (o2hb_region_attr->store)
1966 ret = o2hb_region_attr->store(reg, page, count);
1967 return ret;
1968}
1969
1970static struct configfs_item_operations o2hb_region_item_ops = {
1971 .release = o2hb_region_release,
1972 .show_attribute = o2hb_region_show,
1973 .store_attribute = o2hb_region_store,
1974};
1975
1976static struct config_item_type o2hb_region_type = {
1977 .ct_item_ops = &o2hb_region_item_ops,
1978 .ct_attrs = o2hb_region_attrs,
1979 .ct_owner = THIS_MODULE,
1980};
1981
1982/* heartbeat set */
1983
1984struct o2hb_heartbeat_group {
1985 struct config_group hs_group;
1986 /* some stuff? */
1987};
1988
1989static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1990{
1991 return group ?
1992 container_of(group, struct o2hb_heartbeat_group, hs_group)
1993 : NULL;
1994}
1995
1f285305
SM		 1996static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1997{
1998 int ret = -ENOMEM;
1999
2000 reg->hr_debug_dir =
2001 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
8f443e23 2002 if (!reg->hr_debug_dir) {
1f285305
SM		 2003 mlog_errno(ret);
2004 goto bail;
2005 }
2006
2007 reg->hr_debug_livenodes =
2008 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2009 reg->hr_debug_dir,
2010 &(reg->hr_db_livenodes),
2011 sizeof(*(reg->hr_db_livenodes)),
2012 O2HB_DB_TYPE_REGION_LIVENODES,
2013 sizeof(reg->hr_live_node_bitmap),
2014 O2NM_MAX_NODES, reg);
8f443e23 2015 if (!reg->hr_debug_livenodes) {
1f285305
SM		 2016 mlog_errno(ret);
2017 goto bail;
2018 }
2019
2020 reg->hr_debug_regnum =
2021 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2022 reg->hr_debug_dir,
2023 &(reg->hr_db_regnum),
2024 sizeof(*(reg->hr_db_regnum)),
2025 O2HB_DB_TYPE_REGION_NUMBER,
2026 0, O2NM_MAX_NODES, reg);
8f443e23 2027 if (!reg->hr_debug_regnum) {
1f285305
SM		 2028 mlog_errno(ret);
2029 goto bail;
2030 }
2031
43695d09
SM		 2032 reg->hr_debug_elapsed_time =
2033 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2034 reg->hr_debug_dir,
2035 &(reg->hr_db_elapsed_time),
2036 sizeof(*(reg->hr_db_elapsed_time)),
2037 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2038 0, 0, reg);
8f443e23 2039 if (!reg->hr_debug_elapsed_time) {
43695d09
SM		 2040 mlog_errno(ret);
2041 goto bail;
2042 }
2043
cb0586bd
SM		 2044 reg->hr_debug_pinned =
2045 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2046 reg->hr_debug_dir,
2047 &(reg->hr_db_pinned),
2048 sizeof(*(reg->hr_db_pinned)),
2049 O2HB_DB_TYPE_REGION_PINNED,
2050 0, 0, reg);
8f443e23 2051 if (!reg->hr_debug_pinned) {
cb0586bd
SM		 2052 mlog_errno(ret);
2053 goto bail;
2054 }
2055
8f443e23 2056 ret = 0;
1f285305
SM		 2057bail:
2058 return ret;
2059}
2060
f89ab861
JB		 2061static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2062 const char *name)
a7f6a5fb
MF		 2063{
2064 struct o2hb_region *reg = NULL;
1f285305 2065 int ret;
a7f6a5fb 2066
cd861280 2067 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
f89ab861 2068 if (reg == NULL)
a6795e9e 2069 return ERR_PTR(-ENOMEM);
a7f6a5fb 2070
1cf257f5
JS		 2071 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2072 ret = -ENAMETOOLONG;
2073 goto free;
2074 }
b3c85c4c 2075
a7f6a5fb 2076 spin_lock(&o2hb_live_lock);
536f0741
SM		 2077 reg->hr_region_num = 0;
2078 if (o2hb_global_heartbeat_active()) {
2079 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2080 O2NM_MAX_REGIONS);
2081 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2082 spin_unlock(&o2hb_live_lock);
1cf257f5
JS		 2083 ret = -EFBIG;
2084 goto free;
536f0741
SM		 2085 }
2086 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2087 }
a7f6a5fb
MF		 2088 list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2089 spin_unlock(&o2hb_live_lock);
a7f6a5fb 2090
536f0741
SM		 2091 config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2092
1f285305
SM		 2093 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2094 if (ret) {
2095 config_item_put(&reg->hr_item);
1cf257f5 2096 goto free;
1f285305
SM		 2097 }
2098
a6795e9e 2099 return &reg->hr_item;
1cf257f5
JS		 2100free:
2101 kfree(reg);
2102 return ERR_PTR(ret);
a7f6a5fb
MF		 2103}
2104
2105static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2106 struct config_item *item)
2107{
e6c352db 2108 struct task_struct *hb_task;
a7f6a5fb 2109 struct o2hb_region *reg = to_o2hb_region(item);
58a3158a 2110 int quorum_region = 0;
a7f6a5fb
MF		 2111
2112 /* stop the thread when the user removes the region dir */
e6c352db
JB		 2113 spin_lock(&o2hb_live_lock);
2114 hb_task = reg->hr_task;
2115 reg->hr_task = NULL;
58a3158a 2116 reg->hr_item_dropped = 1;
e6c352db
JB		 2117 spin_unlock(&o2hb_live_lock);
2118
2119 if (hb_task)
2120 kthread_stop(hb_task);
a7f6a5fb 2121
d2eece37
SM		 2122 if (o2hb_global_heartbeat_active()) {
2123 spin_lock(&o2hb_live_lock);
2124 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2125 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2126 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2127 quorum_region = 1;
2128 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2129 spin_unlock(&o2hb_live_lock);
2130 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2131 ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2132 "stopped" : "start aborted"), config_item_name(item),
2133 reg->hr_dev_name);
2134 }
2135
e6df3a66
JB		 2136 /*
2137 * If we're racing a dev_write(), we need to wake them. They will
2138 * check reg->hr_task
2139 */
2140 if (atomic_read(&reg->hr_steady_iterations) != 0) {
d2eece37 2141 reg->hr_aborted_start = 1;
e6df3a66
JB		 2142 atomic_set(&reg->hr_steady_iterations, 0);
2143 wake_up(&o2hb_steady_queue);
2144 }
2145
a7f6a5fb 2146 config_item_put(item);
58a3158a
SM		 2147
2148 if (!o2hb_global_heartbeat_active() || !quorum_region)
2149 return;
2150
2151 /*
2152 * If global heartbeat active and there are dependent users,
2153 * pin all regions if quorum region count <= CUT_OFF
2154 */
2155 spin_lock(&o2hb_live_lock);
2156
2157 if (!o2hb_dependent_users)
2158 goto unlock;
2159
a8f70de3 2160 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 2161 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2162 o2hb_region_pin(NULL);
2163
2164unlock:
2165 spin_unlock(&o2hb_live_lock);
a7f6a5fb
MF		 2166}
2167
2168struct o2hb_heartbeat_group_attribute {
2169 struct configfs_attribute attr;
2170 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2171 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2172};
2173
2174static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2175 struct configfs_attribute *attr,
2176 char *page)
2177{
2178 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2179 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2180 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2181 ssize_t ret = 0;
2182
2183 if (o2hb_heartbeat_group_attr->show)
2184 ret = o2hb_heartbeat_group_attr->show(reg, page);
2185 return ret;
2186}
2187
2188static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2189 struct configfs_attribute *attr,
2190 const char *page, size_t count)
2191{
2192 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2193 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2194 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2195 ssize_t ret = -EINVAL;
2196
2197 if (o2hb_heartbeat_group_attr->store)
2198 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2199 return ret;
2200}
2201
2202static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2203 char *page)
2204{
2205 return sprintf(page, "%u\n", o2hb_dead_threshold);
2206}
2207
2208static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2209 const char *page,
2210 size_t count)
2211{
2212 unsigned long tmp;
2213 char *p = (char *)page;
2214
2215 tmp = simple_strtoul(p, &p, 10);
2216 if (!p || (*p && (*p != '\n')))
2217 return -EINVAL;
2218
2219 /* this will validate ranges for us. */
2220 o2hb_dead_threshold_set((unsigned int) tmp);
2221
2222 return count;
2223}
2224
54b5187b
SM		 2225static
2226ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2227 char *page)
2228{
2229 return sprintf(page, "%s\n",
2230 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2231}
2232
2233static
2234ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2235 const char *page, size_t count)
2236{
2237 unsigned int i;
2238 int ret;
2239 size_t len;
2240
2241 len = (page[count - 1] == '\n') ? count - 1 : count;
2242 if (!len)
2243 return -EINVAL;
2244
2245 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2bd63329 2246 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
54b5187b
SM		 2247 continue;
2248
70f651ed 2249 ret = o2hb_global_heartbeat_mode_set(i);
54b5187b 2250 if (!ret)
18c50cb0 2251 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
54b5187b
SM		 2252 o2hb_heartbeat_mode_desc[i]);
2253 return count;
2254 }
2255
2256 return -EINVAL;
2257
2258}
2259
a7f6a5fb
MF		 2260static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2261 .attr = { .ca_owner = THIS_MODULE,
2262 .ca_name = "dead_threshold",
2263 .ca_mode = S_IRUGO | S_IWUSR },
2264 .show = o2hb_heartbeat_group_threshold_show,
2265 .store = o2hb_heartbeat_group_threshold_store,
2266};
2267
54b5187b
SM		 2268static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2269 .attr = { .ca_owner = THIS_MODULE,
2270 .ca_name = "mode",
2271 .ca_mode = S_IRUGO | S_IWUSR },
2272 .show = o2hb_heartbeat_group_mode_show,
2273 .store = o2hb_heartbeat_group_mode_store,
2274};
2275
a7f6a5fb
MF		 2276static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2277 &o2hb_heartbeat_group_attr_threshold.attr,
54b5187b 2278 &o2hb_heartbeat_group_attr_mode.attr,
a7f6a5fb
MF		 2279 NULL,
2280};
2281
70f651ed 2282static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
a7f6a5fb
MF		 2283 .show_attribute = o2hb_heartbeat_group_show,
2284 .store_attribute = o2hb_heartbeat_group_store,
2285};
2286
2287static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2288 .make_item = o2hb_heartbeat_group_make_item,
2289 .drop_item = o2hb_heartbeat_group_drop_item,
2290};
2291
2292static struct config_item_type o2hb_heartbeat_group_type = {
2293 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
70f651ed 2294 .ct_item_ops = &o2hb_heartbeat_group_item_ops,
a7f6a5fb
MF		 2295 .ct_attrs = o2hb_heartbeat_group_attrs,
2296 .ct_owner = THIS_MODULE,
2297};
2298
2299/* this is just here to avoid touching group in heartbeat.h which the
2300 * entire damn world #includes */
2301struct config_group *o2hb_alloc_hb_set(void)
2302{
2303 struct o2hb_heartbeat_group *hs = NULL;
2304 struct config_group *ret = NULL;
2305
cd861280 2306 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
a7f6a5fb
MF		 2307 if (hs == NULL)
2308 goto out;
2309
2310 config_group_init_type_name(&hs->hs_group, "heartbeat",
2311 &o2hb_heartbeat_group_type);
2312
2313 ret = &hs->hs_group;
2314out:
2315 if (ret == NULL)
2316 kfree(hs);
2317 return ret;
2318}
2319
2320void o2hb_free_hb_set(struct config_group *group)
2321{
2322 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2323 kfree(hs);
2324}
2325
25985edc 2326/* hb callback registration and issuing */
a7f6a5fb
MF		 2327
2328static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2329{
2330 if (type == O2HB_NUM_CB)
2331 return ERR_PTR(-EINVAL);
2332
2333 return &o2hb_callbacks[type];
2334}
2335
2336void o2hb_setup_callback(struct o2hb_callback_func *hc,
2337 enum o2hb_callback_type type,
2338 o2hb_cb_func *func,
2339 void *data,
2340 int priority)
2341{
2342 INIT_LIST_HEAD(&hc->hc_item);
2343 hc->hc_func = func;
2344 hc->hc_data = data;
2345 hc->hc_priority = priority;
2346 hc->hc_type = type;
2347 hc->hc_magic = O2HB_CB_MAGIC;
2348}
2349EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2350
58a3158a
SM		 2351/*
2352 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2353 * In global heartbeat mode, region_uuid passed is NULL.
2354 *
2355 * In local, we only pin the matching region. In global we pin all the active
2356 * regions.
2357 */
2358static int o2hb_region_pin(const char *region_uuid)
14829422 2359{
58a3158a
SM		 2360 int ret = 0, found = 0;
2361 struct o2hb_region *reg;
2362 char *uuid;
14829422
JB		 2363
2364 assert_spin_locked(&o2hb_live_lock);
2365
58a3158a 2366 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2367 if (reg->hr_item_dropped)
2368 continue;
2369
58a3158a
SM		 2370 uuid = config_item_name(&reg->hr_item);
2371
2372 /* local heartbeat */
2373 if (region_uuid) {
2374 if (strcmp(region_uuid, uuid))
2375 continue;
2376 found = 1;
14829422 2377 }
58a3158a
SM		 2378
2379 if (reg->hr_item_pinned || reg->hr_item_dropped)
2380 goto skip_pin;
2381
2382 /* Ignore ENOENT only for local hb (userdlm domain) */
2383 ret = o2nm_depend_item(&reg->hr_item);
2384 if (!ret) {
2385 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2386 reg->hr_item_pinned = 1;
2387 } else {
2388 if (ret == -ENOENT && found)
2389 ret = 0;
2390 else {
2391 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2392 uuid, ret);
2393 break;
2394 }
14829422 2395 }
58a3158a
SM		 2396skip_pin:
2397 if (found)
2398 break;
14829422
JB		 2399 }
2400
58a3158a 2401 return ret;
14829422
JB		 2402}
2403
58a3158a
SM		 2404/*
2405 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2406 * In global heartbeat mode, region_uuid passed is NULL.
2407 *
2408 * In local, we only unpin the matching region. In global we unpin all the
2409 * active regions.
2410 */
2411static void o2hb_region_unpin(const char *region_uuid)
14829422 2412{
14829422 2413 struct o2hb_region *reg;
58a3158a
SM		 2414 char *uuid;
2415 int found = 0;
14829422 2416
58a3158a 2417 assert_spin_locked(&o2hb_live_lock);
14829422 2418
58a3158a 2419 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2420 if (reg->hr_item_dropped)
2421 continue;
2422
58a3158a
SM		 2423 uuid = config_item_name(&reg->hr_item);
2424 if (region_uuid) {
2425 if (strcmp(region_uuid, uuid))
2426 continue;
2427 found = 1;
2428 }
14829422 2429
58a3158a
SM		 2430 if (reg->hr_item_pinned) {
2431 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2432 o2nm_undepend_item(&reg->hr_item);
2433 reg->hr_item_pinned = 0;
2434 }
2435 if (found)
2436 break;
2437 }
2438}
16c6a4f2 2439
58a3158a
SM		 2440static int o2hb_region_inc_user(const char *region_uuid)
2441{
2442 int ret = 0;
14829422 2443
58a3158a 2444 spin_lock(&o2hb_live_lock);
16c6a4f2 2445
58a3158a
SM		 2446 /* local heartbeat */
2447 if (!o2hb_global_heartbeat_active()) {
2448 ret = o2hb_region_pin(region_uuid);
2449 goto unlock;
2450 }
2451
2452 /*
2453 * if global heartbeat active and this is the first dependent user,
2454 * pin all regions if quorum region count <= CUT_OFF
2455 */
2456 o2hb_dependent_users++;
2457 if (o2hb_dependent_users > 1)
2458 goto unlock;
2459
a8f70de3 2460 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 2461 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2462 ret = o2hb_region_pin(NULL);
2463
2464unlock:
2465 spin_unlock(&o2hb_live_lock);
14829422
JB		 2466 return ret;
2467}
2468
58a3158a 2469void o2hb_region_dec_user(const char *region_uuid)
14829422 2470{
14829422
JB		 2471 spin_lock(&o2hb_live_lock);
2472
58a3158a
SM		 2473 /* local heartbeat */
2474 if (!o2hb_global_heartbeat_active()) {
2475 o2hb_region_unpin(region_uuid);
2476 goto unlock;
2477 }
14829422 2478
58a3158a
SM		 2479 /*
2480 * if global heartbeat active and there are no dependent users,
2481 * unpin all quorum regions
2482 */
2483 o2hb_dependent_users--;
2484 if (!o2hb_dependent_users)
2485 o2hb_region_unpin(NULL);
14829422 2486
58a3158a
SM		 2487unlock:
2488 spin_unlock(&o2hb_live_lock);
14829422
JB		 2489}
2490
2491int o2hb_register_callback(const char *region_uuid,
2492 struct o2hb_callback_func *hc)
a7f6a5fb 2493{
df53cd3b 2494 struct o2hb_callback_func *f;
a7f6a5fb
MF		 2495 struct o2hb_callback *hbcall;
2496 int ret;
2497
2498 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2499 BUG_ON(!list_empty(&hc->hc_item));
2500
2501 hbcall = hbcall_from_type(hc->hc_type);
2502 if (IS_ERR(hbcall)) {
2503 ret = PTR_ERR(hbcall);
2504 goto out;
2505 }
2506
14829422 2507 if (region_uuid) {
58a3158a
SM		 2508 ret = o2hb_region_inc_user(region_uuid);
2509 if (ret) {
2510 mlog_errno(ret);
14829422 2511 goto out;
58a3158a 2512 }
14829422
JB		 2513 }
2514
a7f6a5fb
MF		 2515 down_write(&o2hb_callback_sem);
2516
df53cd3b
DF		 2517 list_for_each_entry(f, &hbcall->list, hc_item) {
2518 if (hc->hc_priority < f->hc_priority) {
2519 list_add_tail(&hc->hc_item, &f->hc_item);
a7f6a5fb
MF		 2520 break;
2521 }
2522 }
2523 if (list_empty(&hc->hc_item))
2524 list_add_tail(&hc->hc_item, &hbcall->list);
2525
2526 up_write(&o2hb_callback_sem);
2527 ret = 0;
2528out:
58a3158a 2529 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2530 ret, __builtin_return_address(0), hc);
2531 return ret;
2532}
2533EXPORT_SYMBOL_GPL(o2hb_register_callback);
2534
14829422
JB		 2535void o2hb_unregister_callback(const char *region_uuid,
2536 struct o2hb_callback_func *hc)
a7f6a5fb
MF		 2537{
2538 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2539
58a3158a 2540 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2541 __builtin_return_address(0), hc);
2542
14829422 2543 /* XXX Can this happen _with_ a region reference? */
a7f6a5fb 2544 if (list_empty(&hc->hc_item))
c24f72cc 2545 return;
a7f6a5fb 2546
14829422 2547 if (region_uuid)
58a3158a 2548 o2hb_region_dec_user(region_uuid);
14829422 2549
a7f6a5fb
MF		 2550 down_write(&o2hb_callback_sem);
2551
2552 list_del_init(&hc->hc_item);
2553
2554 up_write(&o2hb_callback_sem);
a7f6a5fb
MF		 2555}
2556EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2557
2558int o2hb_check_node_heartbeating(u8 node_num)
2559{
2560 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2561
2562 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2563 if (!test_bit(node_num, testing_map)) {
2564 mlog(ML_HEARTBEAT,
2565 "node (%u) does not have heartbeating enabled.\n",
2566 node_num);
2567 return 0;
2568 }
2569
2570 return 1;
2571}
2572EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2573
70e82a12
JQ		 2574int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2575{
2576 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2577 unsigned long flags;
2578
2579 spin_lock_irqsave(&o2hb_live_lock, flags);
2580 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2581 spin_unlock_irqrestore(&o2hb_live_lock, flags);
2582 if (!test_bit(node_num, testing_map)) {
2583 mlog(ML_HEARTBEAT,
2584 "node (%u) does not have heartbeating enabled.\n",
2585 node_num);
2586 return 0;
2587 }
2588
2589 return 1;
2590}
2591EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2592
a7f6a5fb
MF		 2593int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2594{
2595 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2596
2597 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2598 if (!test_bit(node_num, testing_map)) {
2599 mlog(ML_HEARTBEAT,
2600 "node (%u) does not have heartbeating enabled.\n",
2601 node_num);
2602 return 0;
2603 }
2604
2605 return 1;
2606}
2607EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2608
2609/* Makes sure our local node is configured with a node number, and is
2610 * heartbeating. */
2611int o2hb_check_local_node_heartbeating(void)
2612{
2613 u8 node_num;
2614
2615 /* if this node was set then we have networking */
2616 node_num = o2nm_this_node();
2617 if (node_num == O2NM_MAX_NODES) {
2618 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2619 return 0;
2620 }
2621
2622 return o2hb_check_node_heartbeating(node_num);
2623}
2624EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2625
2626/*
2627 * this is just a hack until we get the plumbing which flips file systems
2628 * read only and drops the hb ref instead of killing the node dead.
2629 */
2630void o2hb_stop_all_regions(void)
2631{
2632 struct o2hb_region *reg;
2633
2634 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2635
2636 spin_lock(&o2hb_live_lock);
2637
2638 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2639 reg->hr_unclean_stop = 1;
2640
2641 spin_unlock(&o2hb_live_lock);
2642}
2643EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
b3c85c4c
SM		 2644
2645int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2646{
2647 struct o2hb_region *reg;
2648 int numregs = 0;
2649 char *p;
2650
2651 spin_lock(&o2hb_live_lock);
2652
2653 p = region_uuids;
2654 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2655 if (reg->hr_item_dropped)
2656 continue;
2657
b3c85c4c
SM		 2658 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2659 if (numregs < max_regions) {
2660 memcpy(p, config_item_name(&reg->hr_item),
2661 O2HB_MAX_REGION_NAME_LEN);
2662 p += O2HB_MAX_REGION_NAME_LEN;
2663 }
2664 numregs++;
2665 }
2666
2667 spin_unlock(&o2hb_live_lock);
2668
2669 return numregs;
2670}
2671EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2672
2673int o2hb_global_heartbeat_active(void)
2674{
4d94aa1b 2675 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
b3c85c4c
SM		 2676}
2677EXPORT_SYMBOL(o2hb_global_heartbeat_active);
ubuntu-bionic-kernel mirror