projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
[linux-2.6-block.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>
40476b82 39#include <linux/ktime.h>
a7f6a5fb
MF		 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 221 hr_item_pinned:1,
0986fe9b
JQ		 222 hr_item_dropped:1,
223 hr_node_deleted:1;
a7f6a5fb
MF		 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
823a637a
SM		 241 /* live node map of this region */
242 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
536f0741 243 unsigned int hr_region_num;
823a637a 244
1f285305
SM		 245 struct dentry *hr_debug_dir;
246 struct dentry *hr_debug_livenodes;
247 struct dentry *hr_debug_regnum;
43695d09 248 struct dentry *hr_debug_elapsed_time;
cb0586bd 249 struct dentry *hr_debug_pinned;
1f285305
SM		 250 struct o2hb_debug_buf *hr_db_livenodes;
251 struct o2hb_debug_buf *hr_db_regnum;
43695d09 252 struct o2hb_debug_buf *hr_db_elapsed_time;
cb0586bd 253 struct o2hb_debug_buf *hr_db_pinned;
1f285305 254
a7f6a5fb
MF		 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
d2eece37
SM		 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
a7f6a5fb
MF		 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
c4028958 272 struct delayed_work hr_write_timeout_work;
a7f6a5fb
MF		 273 unsigned long hr_last_timeout_start;
274
275 /* Used during o2hb_check_slot to hold a copy of the block
276 * being checked because we temporarily have to zero out the
277 * crc field. */
278 struct o2hb_disk_heartbeat_block *hr_tmp_block;
279};
280
281struct o2hb_bio_wait_ctxt {
282 atomic_t wc_num_reqs;
283 struct completion wc_io_complete;
a9e2ae39 284 int wc_error;
a7f6a5fb
MF		 285};
286
c4028958 287static void o2hb_write_timeout(struct work_struct *work)
a7f6a5fb 288{
b1c5ebfb 289 int failed, quorum;
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()) {
4d548f61 299 spin_lock(&o2hb_live_lock);
b1c5ebfb
SM		 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 305 O2NM_MAX_REGIONS);
4d548f61 306 spin_unlock(&o2hb_live_lock);
b1c5ebfb
SM		 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
a7f6a5fb
MF		 1063/*
1064 * we ride the region ref that the region dir holds. before the region
1065 * dir is removed and drops it ref it will wait to tear down this
1066 * thread.
1067 */
1068static int o2hb_thread(void *data)
1069{
1070 int i, ret;
1071 struct o2hb_region *reg = data;
a7f6a5fb 1072 struct o2hb_bio_wait_ctxt write_wc;
40476b82 1073 ktime_t before_hb, after_hb;
a7f6a5fb
MF		 1074 unsigned int elapsed_msec;
1075
1076 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1077
8698a745 1078 set_user_nice(current, MIN_NICE);
a7f6a5fb 1079
cfc069d3 1080 /* Pin node */
0986fe9b
JQ		 1081 ret = o2nm_depend_this_node();
1082 if (ret) {
1083 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1084 reg->hr_node_deleted = 1;
1085 wake_up(&o2hb_steady_queue);
1086 return 0;
1087 }
cfc069d3 1088
d2eece37
SM		 1089 while (!kthread_should_stop() &&
1090 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
a7f6a5fb 1091 /* We track the time spent inside
025dfdaf 1092 * o2hb_do_disk_heartbeat so that we avoid more than
a7f6a5fb
MF		 1093 * hr_timeout_ms between disk writes. On busy systems
1094 * this should result in a heartbeat which is less
1095 * likely to time itself out. */
40476b82 1096 before_hb = ktime_get_real();
a7f6a5fb 1097
d2eece37 1098 ret = o2hb_do_disk_heartbeat(reg);
a7f6a5fb 1099
40476b82
TR		 1100 after_hb = ktime_get_real();
1101
1102 elapsed_msec = (unsigned int)
1103 ktime_ms_delta(after_hb, before_hb);
a7f6a5fb 1104
b31d308d 1105 mlog(ML_HEARTBEAT,
40476b82
TR		 1106 "start = %lld, end = %lld, msec = %u, ret = %d\n",
1107 before_hb.tv64, after_hb.tv64, elapsed_msec, ret);
a7f6a5fb 1108
d2eece37
SM		 1109 if (!kthread_should_stop() &&
1110 elapsed_msec < reg->hr_timeout_ms) {
a7f6a5fb
MF		 1111 /* the kthread api has blocked signals for us so no
1112 * need to record the return value. */
1113 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1114 }
1115 }
1116
1117 o2hb_disarm_write_timeout(reg);
1118
1119 /* unclean stop is only used in very bad situation */
1120 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1121 o2hb_shutdown_slot(&reg->hr_slots[i]);
1122
1123 /* Explicit down notification - avoid forcing the other nodes
1124 * to timeout on this region when we could just as easily
1125 * write a clear generation - thus indicating to them that
1126 * this node has left this region.
d2eece37
SM		 1127 */
1128 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1129 o2hb_prepare_block(reg, 0);
1130 ret = o2hb_issue_node_write(reg, &write_wc);
1131 if (ret == 0)
1132 o2hb_wait_on_io(reg, &write_wc);
1133 else
1134 mlog_errno(ret);
a7f6a5fb
MF		 1135 }
1136
cfc069d3
SM		 1137 /* Unpin node */
1138 o2nm_undepend_this_node();
a7f6a5fb 1139
d2eece37 1140 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
a7f6a5fb
MF		 1141
1142 return 0;
1143}
1144
87d3d3f3
SM		 1145#ifdef CONFIG_DEBUG_FS
1146static int o2hb_debug_open(struct inode *inode, struct file *file)
1147{
8ca8b0bb 1148 struct o2hb_debug_buf *db = inode->i_private;
1f285305 1149 struct o2hb_region *reg;
87d3d3f3 1150 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
bb570a5d 1151 unsigned long lts;
87d3d3f3
SM		 1152 char *buf = NULL;
1153 int i = -1;
1154 int out = 0;
1155
8ca8b0bb
SM		 1156 /* max_nodes should be the largest bitmap we pass here */
1157 BUG_ON(sizeof(map) < db->db_size);
1158
87d3d3f3
SM		 1159 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1160 if (!buf)
1161 goto bail;
1162
8ca8b0bb
SM		 1163 switch (db->db_type) {
1164 case O2HB_DB_TYPE_LIVENODES:
a6de0136
SM		 1165 case O2HB_DB_TYPE_LIVEREGIONS:
1166 case O2HB_DB_TYPE_QUORUMREGIONS:
1167 case O2HB_DB_TYPE_FAILEDREGIONS:
8ca8b0bb
SM		 1168 spin_lock(&o2hb_live_lock);
1169 memcpy(map, db->db_data, db->db_size);
1170 spin_unlock(&o2hb_live_lock);
1171 break;
87d3d3f3 1172
1f285305
SM		 1173 case O2HB_DB_TYPE_REGION_LIVENODES:
1174 spin_lock(&o2hb_live_lock);
1175 reg = (struct o2hb_region *)db->db_data;
1176 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1177 spin_unlock(&o2hb_live_lock);
1178 break;
1179
1180 case O2HB_DB_TYPE_REGION_NUMBER:
1181 reg = (struct o2hb_region *)db->db_data;
1182 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1183 reg->hr_region_num);
1184 goto done;
1185
43695d09
SM		 1186 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1187 reg = (struct o2hb_region *)db->db_data;
bb570a5d
SM		 1188 lts = reg->hr_last_timeout_start;
1189 /* If 0, it has never been set before */
1190 if (lts)
1191 lts = jiffies_to_msecs(jiffies - lts);
1192 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
43695d09
SM		 1193 goto done;
1194
cb0586bd
SM		 1195 case O2HB_DB_TYPE_REGION_PINNED:
1196 reg = (struct o2hb_region *)db->db_data;
1197 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1198 !!reg->hr_item_pinned);
1199 goto done;
1200
8ca8b0bb
SM		 1201 default:
1202 goto done;
1203 }
1204
1205 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
87d3d3f3
SM		 1206 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1207 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1208
8ca8b0bb 1209done:
87d3d3f3
SM		 1210 i_size_write(inode, out);
1211
1212 file->private_data = buf;
1213
1214 return 0;
1215bail:
1216 return -ENOMEM;
1217}
1218
1219static int o2hb_debug_release(struct inode *inode, struct file *file)
1220{
1221 kfree(file->private_data);
1222 return 0;
1223}
1224
1225static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1226 size_t nbytes, loff_t *ppos)
1227{
1228 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1229 i_size_read(file->f_mapping->host));
1230}
1231#else
1232static int o2hb_debug_open(struct inode *inode, struct file *file)
1233{
1234 return 0;
1235}
1236static int o2hb_debug_release(struct inode *inode, struct file *file)
1237{
1238 return 0;
1239}
1240static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1241 size_t nbytes, loff_t *ppos)
1242{
1243 return 0;
1244}
1245#endif /* CONFIG_DEBUG_FS */
1246
828c0950 1247static const struct file_operations o2hb_debug_fops = {
87d3d3f3
SM		 1248 .open = o2hb_debug_open,
1249 .release = o2hb_debug_release,
1250 .read = o2hb_debug_read,
1251 .llseek = generic_file_llseek,
1252};
1253
1254void o2hb_exit(void)
1255{
a6de0136
SM		 1256 debugfs_remove(o2hb_debug_failedregions);
1257 debugfs_remove(o2hb_debug_quorumregions);
1258 debugfs_remove(o2hb_debug_liveregions);
8ca8b0bb
SM		 1259 debugfs_remove(o2hb_debug_livenodes);
1260 debugfs_remove(o2hb_debug_dir);
a4a1dfa4
JQ		 1261 kfree(o2hb_db_livenodes);
1262 kfree(o2hb_db_liveregions);
1263 kfree(o2hb_db_quorumregions);
1264 kfree(o2hb_db_failedregions);
8ca8b0bb
SM		 1265}
1266
1267static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1268 struct o2hb_debug_buf **db, int db_len,
1269 int type, int size, int len, void *data)
1270{
1271 *db = kmalloc(db_len, GFP_KERNEL);
1272 if (!*db)
1273 return NULL;
1274
1275 (*db)->db_type = type;
1276 (*db)->db_size = size;
1277 (*db)->db_len = len;
1278 (*db)->db_data = data;
1279
1280 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1281 &o2hb_debug_fops);
1282}
1283
1284static int o2hb_debug_init(void)
1285{
1286 int ret = -ENOMEM;
1287
1288 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
8f443e23 1289 if (!o2hb_debug_dir) {
8ca8b0bb
SM		 1290 mlog_errno(ret);
1291 goto bail;
1292 }
1293
1294 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1295 o2hb_debug_dir,
1296 &o2hb_db_livenodes,
1297 sizeof(*o2hb_db_livenodes),
1298 O2HB_DB_TYPE_LIVENODES,
1299 sizeof(o2hb_live_node_bitmap),
1300 O2NM_MAX_NODES,
1301 o2hb_live_node_bitmap);
8f443e23 1302 if (!o2hb_debug_livenodes) {
8ca8b0bb
SM		 1303 mlog_errno(ret);
1304 goto bail;
1305 }
a6de0136
SM		 1306
1307 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1308 o2hb_debug_dir,
1309 &o2hb_db_liveregions,
1310 sizeof(*o2hb_db_liveregions),
1311 O2HB_DB_TYPE_LIVEREGIONS,
1312 sizeof(o2hb_live_region_bitmap),
1313 O2NM_MAX_REGIONS,
1314 o2hb_live_region_bitmap);
8f443e23 1315 if (!o2hb_debug_liveregions) {
a6de0136
SM		 1316 mlog_errno(ret);
1317 goto bail;
1318 }
1319
1320 o2hb_debug_quorumregions =
1321 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1322 o2hb_debug_dir,
1323 &o2hb_db_quorumregions,
1324 sizeof(*o2hb_db_quorumregions),
1325 O2HB_DB_TYPE_QUORUMREGIONS,
1326 sizeof(o2hb_quorum_region_bitmap),
1327 O2NM_MAX_REGIONS,
1328 o2hb_quorum_region_bitmap);
8f443e23 1329 if (!o2hb_debug_quorumregions) {
a6de0136
SM		 1330 mlog_errno(ret);
1331 goto bail;
1332 }
1333
1334 o2hb_debug_failedregions =
1335 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1336 o2hb_debug_dir,
1337 &o2hb_db_failedregions,
1338 sizeof(*o2hb_db_failedregions),
1339 O2HB_DB_TYPE_FAILEDREGIONS,
1340 sizeof(o2hb_failed_region_bitmap),
1341 O2NM_MAX_REGIONS,
1342 o2hb_failed_region_bitmap);
8f443e23 1343 if (!o2hb_debug_failedregions) {
a6de0136
SM		 1344 mlog_errno(ret);
1345 goto bail;
1346 }
1347
8ca8b0bb
SM		 1348 ret = 0;
1349bail:
1350 if (ret)
1351 o2hb_exit();
1352
1353 return ret;
87d3d3f3
SM		 1354}
1355
1356int o2hb_init(void)
a7f6a5fb
MF		 1357{
1358 int i;
1359
1360 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1361 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1362
1363 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1364 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1365
1366 INIT_LIST_HEAD(&o2hb_node_events);
1367
1368 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
536f0741 1369 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
e7d656ba 1370 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
43182d2a 1371 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
b1c5ebfb 1372 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
87d3d3f3 1373
58a3158a
SM		 1374 o2hb_dependent_users = 0;
1375
8ca8b0bb 1376 return o2hb_debug_init();
a7f6a5fb
MF		 1377}
1378
1379/* if we're already in a callback then we're already serialized by the sem */
1380static void o2hb_fill_node_map_from_callback(unsigned long *map,
1381 unsigned bytes)
1382{
1383 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1384
1385 memcpy(map, &o2hb_live_node_bitmap, bytes);
1386}
1387
1388/*
1389 * get a map of all nodes that are heartbeating in any regions
1390 */
1391void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1392{
1393 /* callers want to serialize this map and callbacks so that they
1394 * can trust that they don't miss nodes coming to the party */
1395 down_read(&o2hb_callback_sem);
1396 spin_lock(&o2hb_live_lock);
1397 o2hb_fill_node_map_from_callback(map, bytes);
1398 spin_unlock(&o2hb_live_lock);
1399 up_read(&o2hb_callback_sem);
1400}
1401EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1402
1403/*
1404 * heartbeat configfs bits. The heartbeat set is a default set under
1405 * the cluster set in nodemanager.c.
1406 */
1407
1408static struct o2hb_region *to_o2hb_region(struct config_item *item)
1409{
1410 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1411}
1412
1413/* drop_item only drops its ref after killing the thread, nothing should
1414 * be using the region anymore. this has to clean up any state that
1415 * attributes might have built up. */
1416static void o2hb_region_release(struct config_item *item)
1417{
1418 int i;
1419 struct page *page;
1420 struct o2hb_region *reg = to_o2hb_region(item);
1421
d2eece37
SM		 1422 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1423
d787ab09 1424 kfree(reg->hr_tmp_block);
a7f6a5fb
MF		 1425
1426 if (reg->hr_slot_data) {
1427 for (i = 0; i < reg->hr_num_pages; i++) {
1428 page = reg->hr_slot_data[i];
1429 if (page)
1430 __free_page(page);
1431 }
1432 kfree(reg->hr_slot_data);
1433 }
1434
1435 if (reg->hr_bdev)
9a1c3542 1436 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb 1437
d787ab09 1438 kfree(reg->hr_slots);
a7f6a5fb 1439
1f285305
SM		 1440 debugfs_remove(reg->hr_debug_livenodes);
1441 debugfs_remove(reg->hr_debug_regnum);
d4396eaf 1442 debugfs_remove(reg->hr_debug_elapsed_time);
cb0586bd 1443 debugfs_remove(reg->hr_debug_pinned);
1f285305 1444 debugfs_remove(reg->hr_debug_dir);
a4a1dfa4
JQ		 1445 kfree(reg->hr_db_livenodes);
1446 kfree(reg->hr_db_regnum);
9e13f1f9
JB		 1447 kfree(reg->hr_db_elapsed_time);
1448 kfree(reg->hr_db_pinned);
1f285305 1449
a7f6a5fb
MF		 1450 spin_lock(&o2hb_live_lock);
1451 list_del(&reg->hr_all_item);
1452 spin_unlock(&o2hb_live_lock);
1453
1454 kfree(reg);
1455}
1456
1457static int o2hb_read_block_input(struct o2hb_region *reg,
1458 const char *page,
1459 size_t count,
1460 unsigned long *ret_bytes,
1461 unsigned int *ret_bits)
1462{
1463 unsigned long bytes;
1464 char *p = (char *)page;
1465
1466 bytes = simple_strtoul(p, &p, 0);
1467 if (!p || (*p && (*p != '\n')))
1468 return -EINVAL;
1469
1470 /* Heartbeat and fs min / max block sizes are the same. */
1471 if (bytes > 4096 || bytes < 512)
1472 return -ERANGE;
1473 if (hweight16(bytes) != 1)
1474 return -EINVAL;
1475
1476 if (ret_bytes)
1477 *ret_bytes = bytes;
1478 if (ret_bits)
1479 *ret_bits = ffs(bytes) - 1;
1480
1481 return 0;
1482}
1483
45b99773 1484static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
a7f6a5fb
MF		 1485 char *page)
1486{
45b99773 1487 return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
a7f6a5fb
MF		 1488}
1489
45b99773 1490static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
a7f6a5fb
MF		 1491 const char *page,
1492 size_t count)
1493{
45b99773 1494 struct o2hb_region *reg = to_o2hb_region(item);
a7f6a5fb
MF		 1495 int status;
1496 unsigned long block_bytes;
1497 unsigned int block_bits;
1498
1499 if (reg->hr_bdev)
1500 return -EINVAL;
1501
1502 status = o2hb_read_block_input(reg, page, count,
1503 &block_bytes, &block_bits);
1504 if (status)
1505 return status;
1506
1507 reg->hr_block_bytes = (unsigned int)block_bytes;
1508 reg->hr_block_bits = block_bits;
1509
1510 return count;
1511}
1512
45b99773 1513static ssize_t o2hb_region_start_block_show(struct config_item *item,
a7f6a5fb
MF		 1514 char *page)
1515{
45b99773 1516 return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
a7f6a5fb
MF		 1517}
1518
45b99773 1519static ssize_t o2hb_region_start_block_store(struct config_item *item,
a7f6a5fb
MF		 1520 const char *page,
1521 size_t count)
1522{
45b99773 1523 struct o2hb_region *reg = to_o2hb_region(item);
a7f6a5fb
MF		 1524 unsigned long long tmp;
1525 char *p = (char *)page;
1526
1527 if (reg->hr_bdev)
1528 return -EINVAL;
1529
1530 tmp = simple_strtoull(p, &p, 0);
1531 if (!p || (*p && (*p != '\n')))
1532 return -EINVAL;
1533
1534 reg->hr_start_block = tmp;
1535
1536 return count;
1537}
1538
45b99773 1539static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
a7f6a5fb 1540{
45b99773 1541 return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
a7f6a5fb
MF		 1542}
1543
45b99773 1544static ssize_t o2hb_region_blocks_store(struct config_item *item,
a7f6a5fb
MF		 1545 const char *page,
1546 size_t count)
1547{
45b99773 1548 struct o2hb_region *reg = to_o2hb_region(item);
a7f6a5fb
MF		 1549 unsigned long tmp;
1550 char *p = (char *)page;
1551
1552 if (reg->hr_bdev)
1553 return -EINVAL;
1554
1555 tmp = simple_strtoul(p, &p, 0);
1556 if (!p || (*p && (*p != '\n')))
1557 return -EINVAL;
1558
1559 if (tmp > O2NM_MAX_NODES || tmp == 0)
1560 return -ERANGE;
1561
1562 reg->hr_blocks = (unsigned int)tmp;
1563
1564 return count;
1565}
1566
45b99773 1567static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
a7f6a5fb
MF		 1568{
1569 unsigned int ret = 0;
1570
45b99773
CH		 1571 if (to_o2hb_region(item)->hr_bdev)
1572 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
a7f6a5fb
MF		 1573
1574 return ret;
1575}
1576
1577static void o2hb_init_region_params(struct o2hb_region *reg)
1578{
1579 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1580 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1581
1582 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1583 reg->hr_start_block, reg->hr_blocks);
1584 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1585 reg->hr_block_bytes, reg->hr_block_bits);
1586 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1587 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1588}
1589
1590static int o2hb_map_slot_data(struct o2hb_region *reg)
1591{
1592 int i, j;
1593 unsigned int last_slot;
1594 unsigned int spp = reg->hr_slots_per_page;
1595 struct page *page;
1596 char *raw;
1597 struct o2hb_disk_slot *slot;
1598
1599 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
372a447c 1600 if (reg->hr_tmp_block == NULL)
a7f6a5fb 1601 return -ENOMEM;
a7f6a5fb
MF		 1602
1603 reg->hr_slots = kcalloc(reg->hr_blocks,
1604 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
372a447c 1605 if (reg->hr_slots == NULL)
a7f6a5fb 1606 return -ENOMEM;
a7f6a5fb
MF		 1607
1608 for(i = 0; i < reg->hr_blocks; i++) {
1609 slot = &reg->hr_slots[i];
1610 slot->ds_node_num = i;
1611 INIT_LIST_HEAD(&slot->ds_live_item);
1612 slot->ds_raw_block = NULL;
1613 }
1614
1615 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1616 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1617 "at %u blocks per page\n",
1618 reg->hr_num_pages, reg->hr_blocks, spp);
1619
1620 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1621 GFP_KERNEL);
372a447c 1622 if (!reg->hr_slot_data)
a7f6a5fb 1623 return -ENOMEM;
a7f6a5fb
MF		 1624
1625 for(i = 0; i < reg->hr_num_pages; i++) {
1626 page = alloc_page(GFP_KERNEL);
372a447c 1627 if (!page)
a7f6a5fb 1628 return -ENOMEM;
a7f6a5fb
MF		 1629
1630 reg->hr_slot_data[i] = page;
1631
1632 last_slot = i * spp;
1633 raw = page_address(page);
1634 for (j = 0;
1635 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1636 j++) {
1637 BUG_ON((j + last_slot) >= reg->hr_blocks);
1638
1639 slot = &reg->hr_slots[j + last_slot];
1640 slot->ds_raw_block =
1641 (struct o2hb_disk_heartbeat_block *) raw;
1642
1643 raw += reg->hr_block_bytes;
1644 }
1645 }
1646
1647 return 0;
1648}
1649
1650/* Read in all the slots available and populate the tracking
1651 * structures so that we can start with a baseline idea of what's
1652 * there. */
1653static int o2hb_populate_slot_data(struct o2hb_region *reg)
1654{
1655 int ret, i;
1656 struct o2hb_disk_slot *slot;
1657 struct o2hb_disk_heartbeat_block *hb_block;
1658
a7f6a5fb 1659 ret = o2hb_read_slots(reg, reg->hr_blocks);
372a447c 1660 if (ret)
a7f6a5fb 1661 goto out;
a7f6a5fb
MF		 1662
1663 /* We only want to get an idea of the values initially in each
1664 * slot, so we do no verification - o2hb_check_slot will
1665 * actually determine if each configured slot is valid and
1666 * whether any values have changed. */
1667 for(i = 0; i < reg->hr_blocks; i++) {
1668 slot = &reg->hr_slots[i];
1669 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1670
1671 /* Only fill the values that o2hb_check_slot uses to
1672 * determine changing slots */
1673 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1674 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1675 }
1676
1677out:
a7f6a5fb
MF		 1678 return ret;
1679}
1680
1681/* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
45b99773 1682static ssize_t o2hb_region_dev_store(struct config_item *item,
a7f6a5fb
MF		 1683 const char *page,
1684 size_t count)
1685{
45b99773 1686 struct o2hb_region *reg = to_o2hb_region(item);
e6c352db 1687 struct task_struct *hb_task;
a7f6a5fb
MF		 1688 long fd;
1689 int sectsize;
1690 char *p = (char *)page;
2903ff01
AV		 1691 struct fd f;
1692 struct inode *inode;
a7f6a5fb 1693 ssize_t ret = -EINVAL;
76d9fc29 1694 int live_threshold;
a7f6a5fb
MF		 1695
1696 if (reg->hr_bdev)
1697 goto out;
1698
1699 /* We can't heartbeat without having had our node number
1700 * configured yet. */
1701 if (o2nm_this_node() == O2NM_MAX_NODES)
1702 goto out;
1703
1704 fd = simple_strtol(p, &p, 0);
1705 if (!p || (*p && (*p != '\n')))
1706 goto out;
1707
1708 if (fd < 0 || fd >= INT_MAX)
1709 goto out;
1710
2903ff01
AV		 1711 f = fdget(fd);
1712 if (f.file == NULL)
a7f6a5fb
MF		 1713 goto out;
1714
1715 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1716 reg->hr_block_bytes == 0)
2903ff01 1717 goto out2;
a7f6a5fb 1718
2903ff01 1719 inode = igrab(f.file->f_mapping->host);
a7f6a5fb 1720 if (inode == NULL)
2903ff01 1721 goto out2;
a7f6a5fb
MF		 1722
1723 if (!S_ISBLK(inode->i_mode))
2903ff01 1724 goto out3;
a7f6a5fb 1725
2903ff01 1726 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
e525fd89 1727 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
a7f6a5fb
MF		 1728 if (ret) {
1729 reg->hr_bdev = NULL;
2903ff01 1730 goto out3;
a7f6a5fb
MF		 1731 }
1732 inode = NULL;
1733
1734 bdevname(reg->hr_bdev, reg->hr_dev_name);
1735
e1defc4f 1736 sectsize = bdev_logical_block_size(reg->hr_bdev);
a7f6a5fb
MF		 1737 if (sectsize != reg->hr_block_bytes) {
1738 mlog(ML_ERROR,
1739 "blocksize %u incorrect for device, expected %d",
1740 reg->hr_block_bytes, sectsize);
1741 ret = -EINVAL;
2903ff01 1742 goto out3;
a7f6a5fb
MF		 1743 }
1744
1745 o2hb_init_region_params(reg);
1746
1747 /* Generation of zero is invalid */
1748 do {
1749 get_random_bytes(&reg->hr_generation,
1750 sizeof(reg->hr_generation));
1751 } while (reg->hr_generation == 0);
1752
1753 ret = o2hb_map_slot_data(reg);
1754 if (ret) {
1755 mlog_errno(ret);
2903ff01 1756 goto out3;
a7f6a5fb
MF		 1757 }
1758
1759 ret = o2hb_populate_slot_data(reg);
1760 if (ret) {
1761 mlog_errno(ret);
2903ff01 1762 goto out3;
a7f6a5fb
MF		 1763 }
1764
c4028958 1765 INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
a7f6a5fb
MF		 1766
1767 /*
1768 * A node is considered live after it has beat LIVE_THRESHOLD
1769 * times. We're not steady until we've given them a chance
1770 * _after_ our first read.
76d9fc29
SM		 1771 * The default threshold is bare minimum so as to limit the delay
1772 * during mounts. For global heartbeat, the threshold doubled for the
1773 * first region.
a7f6a5fb 1774 */
76d9fc29
SM		 1775 live_threshold = O2HB_LIVE_THRESHOLD;
1776 if (o2hb_global_heartbeat_active()) {
1777 spin_lock(&o2hb_live_lock);
a8f70de3 1778 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
76d9fc29
SM		 1779 live_threshold <<= 1;
1780 spin_unlock(&o2hb_live_lock);
1781 }
d2eece37
SM		 1782 ++live_threshold;
1783 atomic_set(&reg->hr_steady_iterations, live_threshold);
a84ac334
JB		 1784 /* unsteady_iterations is triple the steady_iterations */
1785 atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
a7f6a5fb 1786
e6c352db
JB		 1787 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1788 reg->hr_item.ci_name);
1789 if (IS_ERR(hb_task)) {
1790 ret = PTR_ERR(hb_task);
a7f6a5fb 1791 mlog_errno(ret);
2903ff01 1792 goto out3;
a7f6a5fb
MF		 1793 }
1794
e6c352db
JB		 1795 spin_lock(&o2hb_live_lock);
1796 reg->hr_task = hb_task;
1797 spin_unlock(&o2hb_live_lock);
1798
a7f6a5fb 1799 ret = wait_event_interruptible(o2hb_steady_queue,
0986fe9b
JQ		 1800 atomic_read(&reg->hr_steady_iterations) == 0 ||
1801 reg->hr_node_deleted);
a7f6a5fb 1802 if (ret) {
d2eece37
SM		 1803 atomic_set(&reg->hr_steady_iterations, 0);
1804 reg->hr_aborted_start = 1;
1805 }
e6c352db 1806
d2eece37
SM		 1807 if (reg->hr_aborted_start) {
1808 ret = -EIO;
2903ff01 1809 goto out3;
a7f6a5fb
MF		 1810 }
1811
0986fe9b
JQ		 1812 if (reg->hr_node_deleted) {
1813 ret = -EINVAL;
1814 goto out3;
1815 }
1816
e6df3a66
JB		 1817 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1818 spin_lock(&o2hb_live_lock);
1819 hb_task = reg->hr_task;
e7d656ba
SM		 1820 if (o2hb_global_heartbeat_active())
1821 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
e6df3a66
JB		 1822 spin_unlock(&o2hb_live_lock);
1823
1824 if (hb_task)
1825 ret = count;
1826 else
1827 ret = -EIO;
1828
18c50cb0 1829 if (hb_task && o2hb_global_heartbeat_active())
d2eece37
SM		 1830 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1831 config_item_name(&reg->hr_item), reg->hr_dev_name);
18c50cb0 1832
2903ff01
AV		 1833out3:
1834 iput(inode);
1835out2:
1836 fdput(f);
a7f6a5fb 1837out:
a7f6a5fb
MF		 1838 if (ret < 0) {
1839 if (reg->hr_bdev) {
9a1c3542 1840 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb
MF		 1841 reg->hr_bdev = NULL;
1842 }
1843 }
1844 return ret;
1845}
1846
45b99773 1847static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
92efc152 1848{
45b99773 1849 struct o2hb_region *reg = to_o2hb_region(item);
e6c352db
JB		 1850 pid_t pid = 0;
1851
1852 spin_lock(&o2hb_live_lock);
1853 if (reg->hr_task)
ba25f9dc 1854 pid = task_pid_nr(reg->hr_task);
e6c352db
JB		 1855 spin_unlock(&o2hb_live_lock);
1856
1857 if (!pid)
92efc152
ZW		 1858 return 0;
1859
e6c352db 1860 return sprintf(page, "%u\n", pid);
92efc152
ZW		 1861}
1862
45b99773
CH		 1863CONFIGFS_ATTR(o2hb_region_, block_bytes);
1864CONFIGFS_ATTR(o2hb_region_, start_block);
1865CONFIGFS_ATTR(o2hb_region_, blocks);
1866CONFIGFS_ATTR(o2hb_region_, dev);
1867CONFIGFS_ATTR_RO(o2hb_region_, pid);
92efc152 1868
a7f6a5fb 1869static struct configfs_attribute *o2hb_region_attrs[] = {
45b99773
CH		 1870 &o2hb_region_attr_block_bytes,
1871 &o2hb_region_attr_start_block,
1872 &o2hb_region_attr_blocks,
1873 &o2hb_region_attr_dev,
1874 &o2hb_region_attr_pid,
a7f6a5fb
MF		 1875 NULL,
1876};
1877
a7f6a5fb
MF		 1878static struct configfs_item_operations o2hb_region_item_ops = {
1879 .release = o2hb_region_release,
a7f6a5fb
MF		 1880};
1881
1882static struct config_item_type o2hb_region_type = {
1883 .ct_item_ops = &o2hb_region_item_ops,
1884 .ct_attrs = o2hb_region_attrs,
1885 .ct_owner = THIS_MODULE,
1886};
1887
1888/* heartbeat set */
1889
1890struct o2hb_heartbeat_group {
1891 struct config_group hs_group;
1892 /* some stuff? */
1893};
1894
1895static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1896{
1897 return group ?
1898 container_of(group, struct o2hb_heartbeat_group, hs_group)
1899 : NULL;
1900}
1901
1f285305
SM		 1902static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1903{
1904 int ret = -ENOMEM;
1905
1906 reg->hr_debug_dir =
1907 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
8f443e23 1908 if (!reg->hr_debug_dir) {
1f285305
SM		 1909 mlog_errno(ret);
1910 goto bail;
1911 }
1912
1913 reg->hr_debug_livenodes =
1914 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1915 reg->hr_debug_dir,
1916 &(reg->hr_db_livenodes),
1917 sizeof(*(reg->hr_db_livenodes)),
1918 O2HB_DB_TYPE_REGION_LIVENODES,
1919 sizeof(reg->hr_live_node_bitmap),
1920 O2NM_MAX_NODES, reg);
8f443e23 1921 if (!reg->hr_debug_livenodes) {
1f285305
SM		 1922 mlog_errno(ret);
1923 goto bail;
1924 }
1925
1926 reg->hr_debug_regnum =
1927 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
1928 reg->hr_debug_dir,
1929 &(reg->hr_db_regnum),
1930 sizeof(*(reg->hr_db_regnum)),
1931 O2HB_DB_TYPE_REGION_NUMBER,
1932 0, O2NM_MAX_NODES, reg);
8f443e23 1933 if (!reg->hr_debug_regnum) {
1f285305
SM		 1934 mlog_errno(ret);
1935 goto bail;
1936 }
1937
43695d09
SM		 1938 reg->hr_debug_elapsed_time =
1939 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
1940 reg->hr_debug_dir,
1941 &(reg->hr_db_elapsed_time),
1942 sizeof(*(reg->hr_db_elapsed_time)),
1943 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
1944 0, 0, reg);
8f443e23 1945 if (!reg->hr_debug_elapsed_time) {
43695d09
SM		 1946 mlog_errno(ret);
1947 goto bail;
1948 }
1949
cb0586bd
SM		 1950 reg->hr_debug_pinned =
1951 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
1952 reg->hr_debug_dir,
1953 &(reg->hr_db_pinned),
1954 sizeof(*(reg->hr_db_pinned)),
1955 O2HB_DB_TYPE_REGION_PINNED,
1956 0, 0, reg);
8f443e23 1957 if (!reg->hr_debug_pinned) {
cb0586bd
SM		 1958 mlog_errno(ret);
1959 goto bail;
1960 }
1961
8f443e23 1962 ret = 0;
1f285305
SM		 1963bail:
1964 return ret;
1965}
1966
f89ab861
JB		 1967static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1968 const char *name)
a7f6a5fb
MF		 1969{
1970 struct o2hb_region *reg = NULL;
1f285305 1971 int ret;
a7f6a5fb 1972
cd861280 1973 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
f89ab861 1974 if (reg == NULL)
a6795e9e 1975 return ERR_PTR(-ENOMEM);
a7f6a5fb 1976
1cf257f5
JS		 1977 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
1978 ret = -ENAMETOOLONG;
1979 goto free;
1980 }
b3c85c4c 1981
a7f6a5fb 1982 spin_lock(&o2hb_live_lock);
536f0741
SM		 1983 reg->hr_region_num = 0;
1984 if (o2hb_global_heartbeat_active()) {
1985 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
1986 O2NM_MAX_REGIONS);
1987 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
1988 spin_unlock(&o2hb_live_lock);
1cf257f5
JS		 1989 ret = -EFBIG;
1990 goto free;
536f0741
SM		 1991 }
1992 set_bit(reg->hr_region_num, o2hb_region_bitmap);
1993 }
a7f6a5fb
MF		 1994 list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
1995 spin_unlock(&o2hb_live_lock);
a7f6a5fb 1996
536f0741
SM		 1997 config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
1998
1f285305
SM		 1999 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2000 if (ret) {
2001 config_item_put(&reg->hr_item);
1cf257f5 2002 goto free;
1f285305
SM		 2003 }
2004
a6795e9e 2005 return &reg->hr_item;
1cf257f5
JS		 2006free:
2007 kfree(reg);
2008 return ERR_PTR(ret);
a7f6a5fb
MF		 2009}
2010
2011static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2012 struct config_item *item)
2013{
e6c352db 2014 struct task_struct *hb_task;
a7f6a5fb 2015 struct o2hb_region *reg = to_o2hb_region(item);
58a3158a 2016 int quorum_region = 0;
a7f6a5fb
MF		 2017
2018 /* stop the thread when the user removes the region dir */
e6c352db
JB		 2019 spin_lock(&o2hb_live_lock);
2020 hb_task = reg->hr_task;
2021 reg->hr_task = NULL;
58a3158a 2022 reg->hr_item_dropped = 1;
e6c352db
JB		 2023 spin_unlock(&o2hb_live_lock);
2024
2025 if (hb_task)
2026 kthread_stop(hb_task);
a7f6a5fb 2027
d2eece37
SM		 2028 if (o2hb_global_heartbeat_active()) {
2029 spin_lock(&o2hb_live_lock);
2030 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2031 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2032 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2033 quorum_region = 1;
2034 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2035 spin_unlock(&o2hb_live_lock);
2036 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2037 ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2038 "stopped" : "start aborted"), config_item_name(item),
2039 reg->hr_dev_name);
2040 }
2041
e6df3a66
JB		 2042 /*
2043 * If we're racing a dev_write(), we need to wake them. They will
2044 * check reg->hr_task
2045 */
2046 if (atomic_read(&reg->hr_steady_iterations) != 0) {
d2eece37 2047 reg->hr_aborted_start = 1;
e6df3a66
JB		 2048 atomic_set(&reg->hr_steady_iterations, 0);
2049 wake_up(&o2hb_steady_queue);
2050 }
2051
a7f6a5fb 2052 config_item_put(item);
58a3158a
SM		 2053
2054 if (!o2hb_global_heartbeat_active() || !quorum_region)
2055 return;
2056
2057 /*
2058 * If global heartbeat active and there are dependent users,
2059 * pin all regions if quorum region count <= CUT_OFF
2060 */
2061 spin_lock(&o2hb_live_lock);
2062
2063 if (!o2hb_dependent_users)
2064 goto unlock;
2065
a8f70de3 2066 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 2067 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2068 o2hb_region_pin(NULL);
2069
2070unlock:
2071 spin_unlock(&o2hb_live_lock);
a7f6a5fb
MF		 2072}
2073
45b99773
CH		 2074static ssize_t o2hb_heartbeat_group_threshold_show(struct config_item *item,
2075 char *page)
a7f6a5fb
MF		 2076{
2077 return sprintf(page, "%u\n", o2hb_dead_threshold);
2078}
2079
45b99773
CH		 2080static ssize_t o2hb_heartbeat_group_threshold_store(struct config_item *item,
2081 const char *page, size_t count)
a7f6a5fb
MF		 2082{
2083 unsigned long tmp;
2084 char *p = (char *)page;
2085
2086 tmp = simple_strtoul(p, &p, 10);
2087 if (!p || (*p && (*p != '\n')))
2088 return -EINVAL;
2089
2090 /* this will validate ranges for us. */
2091 o2hb_dead_threshold_set((unsigned int) tmp);
2092
2093 return count;
2094}
2095
45b99773
CH		 2096static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2097 char *page)
54b5187b
SM		 2098{
2099 return sprintf(page, "%s\n",
2100 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2101}
2102
45b99773
CH		 2103static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2104 const char *page, size_t count)
54b5187b
SM		 2105{
2106 unsigned int i;
2107 int ret;
2108 size_t len;
2109
2110 len = (page[count - 1] == '\n') ? count - 1 : count;
2111 if (!len)
2112 return -EINVAL;
2113
2114 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2bd63329 2115 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
54b5187b
SM		 2116 continue;
2117
70f651ed 2118 ret = o2hb_global_heartbeat_mode_set(i);
54b5187b 2119 if (!ret)
18c50cb0 2120 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
54b5187b
SM		 2121 o2hb_heartbeat_mode_desc[i]);
2122 return count;
2123 }
2124
2125 return -EINVAL;
2126
2127}
2128
45b99773
CH		 2129CONFIGFS_ATTR(o2hb_heartbeat_group_, threshold);
2130CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
54b5187b 2131
a7f6a5fb 2132static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
45b99773
CH		 2133 &o2hb_heartbeat_group_attr_threshold,
2134 &o2hb_heartbeat_group_attr_mode,
a7f6a5fb
MF		 2135 NULL,
2136};
2137
a7f6a5fb
MF		 2138static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2139 .make_item = o2hb_heartbeat_group_make_item,
2140 .drop_item = o2hb_heartbeat_group_drop_item,
2141};
2142
2143static struct config_item_type o2hb_heartbeat_group_type = {
2144 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
a7f6a5fb
MF		 2145 .ct_attrs = o2hb_heartbeat_group_attrs,
2146 .ct_owner = THIS_MODULE,
2147};
2148
2149/* this is just here to avoid touching group in heartbeat.h which the
2150 * entire damn world #includes */
2151struct config_group *o2hb_alloc_hb_set(void)
2152{
2153 struct o2hb_heartbeat_group *hs = NULL;
2154 struct config_group *ret = NULL;
2155
cd861280 2156 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
a7f6a5fb
MF		 2157 if (hs == NULL)
2158 goto out;
2159
2160 config_group_init_type_name(&hs->hs_group, "heartbeat",
2161 &o2hb_heartbeat_group_type);
2162
2163 ret = &hs->hs_group;
2164out:
2165 if (ret == NULL)
2166 kfree(hs);
2167 return ret;
2168}
2169
2170void o2hb_free_hb_set(struct config_group *group)
2171{
2172 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2173 kfree(hs);
2174}
2175
25985edc 2176/* hb callback registration and issuing */
a7f6a5fb
MF		 2177
2178static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2179{
2180 if (type == O2HB_NUM_CB)
2181 return ERR_PTR(-EINVAL);
2182
2183 return &o2hb_callbacks[type];
2184}
2185
2186void o2hb_setup_callback(struct o2hb_callback_func *hc,
2187 enum o2hb_callback_type type,
2188 o2hb_cb_func *func,
2189 void *data,
2190 int priority)
2191{
2192 INIT_LIST_HEAD(&hc->hc_item);
2193 hc->hc_func = func;
2194 hc->hc_data = data;
2195 hc->hc_priority = priority;
2196 hc->hc_type = type;
2197 hc->hc_magic = O2HB_CB_MAGIC;
2198}
2199EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2200
58a3158a
SM		 2201/*
2202 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2203 * In global heartbeat mode, region_uuid passed is NULL.
2204 *
2205 * In local, we only pin the matching region. In global we pin all the active
2206 * regions.
2207 */
2208static int o2hb_region_pin(const char *region_uuid)
14829422 2209{
58a3158a
SM		 2210 int ret = 0, found = 0;
2211 struct o2hb_region *reg;
2212 char *uuid;
14829422
JB		 2213
2214 assert_spin_locked(&o2hb_live_lock);
2215
58a3158a 2216 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2217 if (reg->hr_item_dropped)
2218 continue;
2219
58a3158a
SM		 2220 uuid = config_item_name(&reg->hr_item);
2221
2222 /* local heartbeat */
2223 if (region_uuid) {
2224 if (strcmp(region_uuid, uuid))
2225 continue;
2226 found = 1;
14829422 2227 }
58a3158a
SM		 2228
2229 if (reg->hr_item_pinned || reg->hr_item_dropped)
2230 goto skip_pin;
2231
2232 /* Ignore ENOENT only for local hb (userdlm domain) */
2233 ret = o2nm_depend_item(&reg->hr_item);
2234 if (!ret) {
2235 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2236 reg->hr_item_pinned = 1;
2237 } else {
2238 if (ret == -ENOENT && found)
2239 ret = 0;
2240 else {
2241 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2242 uuid, ret);
2243 break;
2244 }
14829422 2245 }
58a3158a
SM		 2246skip_pin:
2247 if (found)
2248 break;
14829422
JB		 2249 }
2250
58a3158a 2251 return ret;
14829422
JB		 2252}
2253
58a3158a
SM		 2254/*
2255 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2256 * In global heartbeat mode, region_uuid passed is NULL.
2257 *
2258 * In local, we only unpin the matching region. In global we unpin all the
2259 * active regions.
2260 */
2261static void o2hb_region_unpin(const char *region_uuid)
14829422 2262{
14829422 2263 struct o2hb_region *reg;
58a3158a
SM		 2264 char *uuid;
2265 int found = 0;
14829422 2266
58a3158a 2267 assert_spin_locked(&o2hb_live_lock);
14829422 2268
58a3158a 2269 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2270 if (reg->hr_item_dropped)
2271 continue;
2272
58a3158a
SM		 2273 uuid = config_item_name(&reg->hr_item);
2274 if (region_uuid) {
2275 if (strcmp(region_uuid, uuid))
2276 continue;
2277 found = 1;
2278 }
14829422 2279
58a3158a
SM		 2280 if (reg->hr_item_pinned) {
2281 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2282 o2nm_undepend_item(&reg->hr_item);
2283 reg->hr_item_pinned = 0;
2284 }
2285 if (found)
2286 break;
2287 }
2288}
16c6a4f2 2289
58a3158a
SM		 2290static int o2hb_region_inc_user(const char *region_uuid)
2291{
2292 int ret = 0;
14829422 2293
58a3158a 2294 spin_lock(&o2hb_live_lock);
16c6a4f2 2295
58a3158a
SM		 2296 /* local heartbeat */
2297 if (!o2hb_global_heartbeat_active()) {
2298 ret = o2hb_region_pin(region_uuid);
2299 goto unlock;
2300 }
2301
2302 /*
2303 * if global heartbeat active and this is the first dependent user,
2304 * pin all regions if quorum region count <= CUT_OFF
2305 */
2306 o2hb_dependent_users++;
2307 if (o2hb_dependent_users > 1)
2308 goto unlock;
2309
a8f70de3 2310 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 2311 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2312 ret = o2hb_region_pin(NULL);
2313
2314unlock:
2315 spin_unlock(&o2hb_live_lock);
14829422
JB		 2316 return ret;
2317}
2318
58a3158a 2319void o2hb_region_dec_user(const char *region_uuid)
14829422 2320{
14829422
JB		 2321 spin_lock(&o2hb_live_lock);
2322
58a3158a
SM		 2323 /* local heartbeat */
2324 if (!o2hb_global_heartbeat_active()) {
2325 o2hb_region_unpin(region_uuid);
2326 goto unlock;
2327 }
14829422 2328
58a3158a
SM		 2329 /*
2330 * if global heartbeat active and there are no dependent users,
2331 * unpin all quorum regions
2332 */
2333 o2hb_dependent_users--;
2334 if (!o2hb_dependent_users)
2335 o2hb_region_unpin(NULL);
14829422 2336
58a3158a
SM		 2337unlock:
2338 spin_unlock(&o2hb_live_lock);
14829422
JB		 2339}
2340
2341int o2hb_register_callback(const char *region_uuid,
2342 struct o2hb_callback_func *hc)
a7f6a5fb 2343{
df53cd3b 2344 struct o2hb_callback_func *f;
a7f6a5fb
MF		 2345 struct o2hb_callback *hbcall;
2346 int ret;
2347
2348 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2349 BUG_ON(!list_empty(&hc->hc_item));
2350
2351 hbcall = hbcall_from_type(hc->hc_type);
2352 if (IS_ERR(hbcall)) {
2353 ret = PTR_ERR(hbcall);
2354 goto out;
2355 }
2356
14829422 2357 if (region_uuid) {
58a3158a
SM		 2358 ret = o2hb_region_inc_user(region_uuid);
2359 if (ret) {
2360 mlog_errno(ret);
14829422 2361 goto out;
58a3158a 2362 }
14829422
JB		 2363 }
2364
a7f6a5fb
MF		 2365 down_write(&o2hb_callback_sem);
2366
df53cd3b
DF		 2367 list_for_each_entry(f, &hbcall->list, hc_item) {
2368 if (hc->hc_priority < f->hc_priority) {
2369 list_add_tail(&hc->hc_item, &f->hc_item);
a7f6a5fb
MF		 2370 break;
2371 }
2372 }
2373 if (list_empty(&hc->hc_item))
2374 list_add_tail(&hc->hc_item, &hbcall->list);
2375
2376 up_write(&o2hb_callback_sem);
2377 ret = 0;
2378out:
58a3158a 2379 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2380 ret, __builtin_return_address(0), hc);
2381 return ret;
2382}
2383EXPORT_SYMBOL_GPL(o2hb_register_callback);
2384
14829422
JB		 2385void o2hb_unregister_callback(const char *region_uuid,
2386 struct o2hb_callback_func *hc)
a7f6a5fb
MF		 2387{
2388 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2389
58a3158a 2390 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2391 __builtin_return_address(0), hc);
2392
14829422 2393 /* XXX Can this happen _with_ a region reference? */
a7f6a5fb 2394 if (list_empty(&hc->hc_item))
c24f72cc 2395 return;
a7f6a5fb 2396
14829422 2397 if (region_uuid)
58a3158a 2398 o2hb_region_dec_user(region_uuid);
14829422 2399
a7f6a5fb
MF		 2400 down_write(&o2hb_callback_sem);
2401
2402 list_del_init(&hc->hc_item);
2403
2404 up_write(&o2hb_callback_sem);
a7f6a5fb
MF		 2405}
2406EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2407
2408int o2hb_check_node_heartbeating(u8 node_num)
2409{
2410 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2411
2412 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2413 if (!test_bit(node_num, testing_map)) {
2414 mlog(ML_HEARTBEAT,
2415 "node (%u) does not have heartbeating enabled.\n",
2416 node_num);
2417 return 0;
2418 }
2419
2420 return 1;
2421}
2422EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2423
70e82a12
JQ		 2424int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2425{
2426 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
70e82a12 2427
4d548f61 2428 spin_lock(&o2hb_live_lock);
70e82a12 2429 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
4d548f61 2430 spin_unlock(&o2hb_live_lock);
70e82a12
JQ		 2431 if (!test_bit(node_num, testing_map)) {
2432 mlog(ML_HEARTBEAT,
2433 "node (%u) does not have heartbeating enabled.\n",
2434 node_num);
2435 return 0;
2436 }
2437
2438 return 1;
2439}
2440EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2441
a7f6a5fb
MF		 2442int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2443{
2444 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2445
2446 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2447 if (!test_bit(node_num, testing_map)) {
2448 mlog(ML_HEARTBEAT,
2449 "node (%u) does not have heartbeating enabled.\n",
2450 node_num);
2451 return 0;
2452 }
2453
2454 return 1;
2455}
2456EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2457
2458/* Makes sure our local node is configured with a node number, and is
2459 * heartbeating. */
2460int o2hb_check_local_node_heartbeating(void)
2461{
2462 u8 node_num;
2463
2464 /* if this node was set then we have networking */
2465 node_num = o2nm_this_node();
2466 if (node_num == O2NM_MAX_NODES) {
2467 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2468 return 0;
2469 }
2470
2471 return o2hb_check_node_heartbeating(node_num);
2472}
2473EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2474
2475/*
2476 * this is just a hack until we get the plumbing which flips file systems
2477 * read only and drops the hb ref instead of killing the node dead.
2478 */
2479void o2hb_stop_all_regions(void)
2480{
2481 struct o2hb_region *reg;
2482
2483 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2484
2485 spin_lock(&o2hb_live_lock);
2486
2487 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2488 reg->hr_unclean_stop = 1;
2489
2490 spin_unlock(&o2hb_live_lock);
2491}
2492EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
b3c85c4c
SM		 2493
2494int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2495{
2496 struct o2hb_region *reg;
2497 int numregs = 0;
2498 char *p;
2499
2500 spin_lock(&o2hb_live_lock);
2501
2502 p = region_uuids;
2503 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2504 if (reg->hr_item_dropped)
2505 continue;
2506
b3c85c4c
SM		 2507 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2508 if (numregs < max_regions) {
2509 memcpy(p, config_item_name(&reg->hr_item),
2510 O2HB_MAX_REGION_NAME_LEN);
2511 p += O2HB_MAX_REGION_NAME_LEN;
2512 }
2513 numregs++;
2514 }
2515
2516 spin_unlock(&o2hb_live_lock);
2517
2518 return numregs;
2519}
2520EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2521
2522int o2hb_global_heartbeat_active(void)
2523{
4d94aa1b 2524 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
b3c85c4c
SM		 2525}
2526EXPORT_SYMBOL(o2hb_global_heartbeat_active);
Linux 6.x block layer and io_uring tree(s)