projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: Abstract out bvec iterator
[GitHub/exynos8895/android_kernel_samsung_universal8895.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
782e3b3b 375static void o2hb_bio_end_io(struct bio *bio,
a7f6a5fb
MF		 376 int error)
377{
378 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
379
a9e2ae39 380 if (error) {
a7f6a5fb 381 mlog(ML_ERROR, "IO Error %d\n", error);
a9e2ae39
MF		 382 wc->wc_error = error;
383 }
a7f6a5fb 384
a7f6a5fb 385 o2hb_bio_wait_dec(wc, 1);
b559292e 386 bio_put(bio);
a7f6a5fb
MF		 387}
388
389/* Setup a Bio to cover I/O against num_slots slots starting at
390 * start_slot. */
391static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
392 struct o2hb_bio_wait_ctxt *wc,
b559292e
PR		 393 unsigned int *current_slot,
394 unsigned int max_slots)
a7f6a5fb 395{
b559292e 396 int len, current_page;
a7f6a5fb
MF		 397 unsigned int vec_len, vec_start;
398 unsigned int bits = reg->hr_block_bits;
399 unsigned int spp = reg->hr_slots_per_page;
b559292e 400 unsigned int cs = *current_slot;
a7f6a5fb
MF		 401 struct bio *bio;
402 struct page *page;
403
a7f6a5fb
MF		 404 /* Testing has shown this allocation to take long enough under
405 * GFP_KERNEL that the local node can get fenced. It would be
406 * nicest if we could pre-allocate these bios and avoid this
407 * all together. */
b559292e 408 bio = bio_alloc(GFP_ATOMIC, 16);
a7f6a5fb
MF		 409 if (!bio) {
410 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
411 bio = ERR_PTR(-ENOMEM);
412 goto bail;
413 }
414
415 /* Must put everything in 512 byte sectors for the bio... */
4f024f37 416 bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
a7f6a5fb
MF		 417 bio->bi_bdev = reg->hr_bdev;
418 bio->bi_private = wc;
419 bio->bi_end_io = o2hb_bio_end_io;
420
b559292e
PR		 421 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
422 while(cs < max_slots) {
423 current_page = cs / spp;
424 page = reg->hr_slot_data[current_page];
a7f6a5fb 425
bc7e97cb 426 vec_len = min(PAGE_CACHE_SIZE - vec_start,
b559292e 427 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
a7f6a5fb
MF		 428
429 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
b559292e 430 current_page, vec_len, vec_start);
a7f6a5fb
MF		 431
432 len = bio_add_page(bio, page, vec_len, vec_start);
b559292e 433 if (len != vec_len) break;
a7f6a5fb 434
b559292e 435 cs += vec_len / (PAGE_CACHE_SIZE/spp);
a7f6a5fb
MF		 436 vec_start = 0;
437 }
438
439bail:
b559292e 440 *current_slot = cs;
a7f6a5fb
MF		 441 return bio;
442}
443
a7f6a5fb
MF		 444static int o2hb_read_slots(struct o2hb_region *reg,
445 unsigned int max_slots)
446{
b559292e
PR		 447 unsigned int current_slot=0;
448 int status;
a7f6a5fb 449 struct o2hb_bio_wait_ctxt wc;
a7f6a5fb
MF		 450 struct bio *bio;
451
b559292e 452 o2hb_bio_wait_init(&wc);
a7f6a5fb 453
b559292e
PR		 454 while(current_slot < max_slots) {
455 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
a7f6a5fb 456 if (IS_ERR(bio)) {
a7f6a5fb
MF		 457 status = PTR_ERR(bio);
458 mlog_errno(status);
459 goto bail_and_wait;
460 }
a7f6a5fb 461
b559292e 462 atomic_inc(&wc.wc_num_reqs);
a7f6a5fb
MF		 463 submit_bio(READ, bio);
464 }
465
466 status = 0;
467
468bail_and_wait:
469 o2hb_wait_on_io(reg, &wc);
a9e2ae39
MF		 470 if (wc.wc_error && !status)
471 status = wc.wc_error;
a7f6a5fb 472
a7f6a5fb
MF		 473 return status;
474}
475
476static int o2hb_issue_node_write(struct o2hb_region *reg,
a7f6a5fb
MF		 477 struct o2hb_bio_wait_ctxt *write_wc)
478{
479 int status;
480 unsigned int slot;
481 struct bio *bio;
482
b559292e 483 o2hb_bio_wait_init(write_wc);
a7f6a5fb
MF		 484
485 slot = o2nm_this_node();
486
b559292e 487 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
a7f6a5fb
MF		 488 if (IS_ERR(bio)) {
489 status = PTR_ERR(bio);
490 mlog_errno(status);
491 goto bail;
492 }
493
b559292e 494 atomic_inc(&write_wc->wc_num_reqs);
e873fdb5 495 submit_bio(WRITE_SYNC, bio);
a7f6a5fb 496
a7f6a5fb
MF		 497 status = 0;
498bail:
499 return status;
500}
501
502static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
503 struct o2hb_disk_heartbeat_block *hb_block)
504{
505 __le32 old_cksum;
506 u32 ret;
507
508 /* We want to compute the block crc with a 0 value in the
509 * hb_cksum field. Save it off here and replace after the
510 * crc. */
511 old_cksum = hb_block->hb_cksum;
512 hb_block->hb_cksum = 0;
513
514 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
515
516 hb_block->hb_cksum = old_cksum;
517
518 return ret;
519}
520
521static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
522{
70bacbdb
MF		 523 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
524 "cksum = 0x%x, generation 0x%llx\n",
525 (long long)le64_to_cpu(hb_block->hb_seq),
526 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
527 (long long)le64_to_cpu(hb_block->hb_generation));
a7f6a5fb
MF		 528}
529
530static int o2hb_verify_crc(struct o2hb_region *reg,
531 struct o2hb_disk_heartbeat_block *hb_block)
532{
533 u32 read, computed;
534
535 read = le32_to_cpu(hb_block->hb_cksum);
536 computed = o2hb_compute_block_crc_le(reg, hb_block);
537
538 return read == computed;
539}
540
d2eece37
SM		 541/*
542 * Compare the slot data with what we wrote in the last iteration.
543 * If the match fails, print an appropriate error message. This is to
544 * detect errors like... another node hearting on the same slot,
545 * flaky device that is losing writes, etc.
546 * Returns 1 if check succeeds, 0 otherwise.
547 */
548static int o2hb_check_own_slot(struct o2hb_region *reg)
a7f6a5fb 549{
a7f6a5fb
MF		 550 struct o2hb_disk_slot *slot;
551 struct o2hb_disk_heartbeat_block *hb_block;
33c12a54 552 char *errstr;
a7f6a5fb 553
33c12a54 554 slot = &reg->hr_slots[o2nm_this_node()];
a7f6a5fb 555 /* Don't check on our 1st timestamp */
33c12a54 556 if (!slot->ds_last_time)
d2eece37 557 return 0;
a7f6a5fb 558
33c12a54
SM		 559 hb_block = slot->ds_raw_block;
560 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
561 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
562 hb_block->hb_node == slot->ds_node_num)
d2eece37 563 return 1;
a7f6a5fb 564
33c12a54
SM		 565#define ERRSTR1 "Another node is heartbeating on device"
566#define ERRSTR2 "Heartbeat generation mismatch on device"
567#define ERRSTR3 "Heartbeat sequence mismatch on device"
568
569 if (hb_block->hb_node != slot->ds_node_num)
570 errstr = ERRSTR1;
571 else if (le64_to_cpu(hb_block->hb_generation) !=
572 slot->ds_last_generation)
573 errstr = ERRSTR2;
574 else
575 errstr = ERRSTR3;
576
577 mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
578 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
579 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
580 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
581 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
582 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
d2eece37
SM		 583
584 return 0;
a7f6a5fb
MF		 585}
586
587static inline void o2hb_prepare_block(struct o2hb_region *reg,
588 u64 generation)
589{
590 int node_num;
591 u64 cputime;
592 struct o2hb_disk_slot *slot;
593 struct o2hb_disk_heartbeat_block *hb_block;
594
595 node_num = o2nm_this_node();
596 slot = &reg->hr_slots[node_num];
597
598 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
599 memset(hb_block, 0, reg->hr_block_bytes);
600 /* TODO: time stuff */
601 cputime = CURRENT_TIME.tv_sec;
602 if (!cputime)
603 cputime = 1;
604
605 hb_block->hb_seq = cpu_to_le64(cputime);
606 hb_block->hb_node = node_num;
607 hb_block->hb_generation = cpu_to_le64(generation);
0db638f4 608 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
a7f6a5fb
MF		 609
610 /* This step must always happen last! */
611 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
612 hb_block));
613
70bacbdb 614 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
5fdf1e67 615 (long long)generation,
70bacbdb 616 le32_to_cpu(hb_block->hb_cksum));
a7f6a5fb
MF		 617}
618
619static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
620 struct o2nm_node *node,
621 int idx)
622{
a7f6a5fb
MF		 623 struct o2hb_callback_func *f;
624
df53cd3b 625 list_for_each_entry(f, &hbcall->list, hc_item) {
a7f6a5fb
MF		 626 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
627 (f->hc_func)(node, idx, f->hc_data);
628 }
629}
630
631/* Will run the list in order until we process the passed event */
632static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
633{
a7f6a5fb
MF		 634 struct o2hb_callback *hbcall;
635 struct o2hb_node_event *event;
636
a7f6a5fb
MF		 637 /* Holding callback sem assures we don't alter the callback
638 * lists when doing this, and serializes ourselves with other
639 * processes wanting callbacks. */
640 down_write(&o2hb_callback_sem);
641
642 spin_lock(&o2hb_live_lock);
643 while (!list_empty(&o2hb_node_events)
644 && !list_empty(&queued_event->hn_item)) {
645 event = list_entry(o2hb_node_events.next,
646 struct o2hb_node_event,
647 hn_item);
648 list_del_init(&event->hn_item);
649 spin_unlock(&o2hb_live_lock);
650
651 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
652 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
653 event->hn_node_num);
654
655 hbcall = hbcall_from_type(event->hn_event_type);
656
657 /* We should *never* have gotten on to the list with a
658 * bad type... This isn't something that we should try
659 * to recover from. */
660 BUG_ON(IS_ERR(hbcall));
661
662 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
663
664 spin_lock(&o2hb_live_lock);
665 }
666 spin_unlock(&o2hb_live_lock);
667
668 up_write(&o2hb_callback_sem);
669}
670
671static void o2hb_queue_node_event(struct o2hb_node_event *event,
672 enum o2hb_callback_type type,
673 struct o2nm_node *node,
674 int node_num)
675{
676 assert_spin_locked(&o2hb_live_lock);
677
0e105d37
SM		 678 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
679
a7f6a5fb
MF		 680 event->hn_event_type = type;
681 event->hn_node = node;
682 event->hn_node_num = node_num;
683
684 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
685 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
686
687 list_add_tail(&event->hn_item, &o2hb_node_events);
688}
689
690static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
691{
692 struct o2hb_node_event event =
693 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
694 struct o2nm_node *node;
6f8648e8 695 int queued = 0;
a7f6a5fb
MF		 696
697 node = o2nm_get_node_by_num(slot->ds_node_num);
698 if (!node)
699 return;
700
701 spin_lock(&o2hb_live_lock);
702 if (!list_empty(&slot->ds_live_item)) {
703 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
704 slot->ds_node_num);
705
706 list_del_init(&slot->ds_live_item);
707
708 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
709 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
710
711 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
712 slot->ds_node_num);
6f8648e8 713 queued = 1;
a7f6a5fb
MF		 714 }
715 }
716 spin_unlock(&o2hb_live_lock);
717
6f8648e8
J		 718 if (queued)
719 o2hb_run_event_list(&event);
a7f6a5fb
MF		 720
721 o2nm_node_put(node);
722}
723
d2eece37 724static void o2hb_set_quorum_device(struct o2hb_region *reg)
43182d2a 725{
43182d2a
SM		 726 if (!o2hb_global_heartbeat_active())
727 return;
728
d2eece37
SM		 729 /* Prevent race with o2hb_heartbeat_group_drop_item() */
730 if (kthread_should_stop())
43182d2a
SM		 731 return;
732
d2eece37
SM		 733 /* Tag region as quorum only after thread reaches steady state */
734 if (atomic_read(&reg->hr_steady_iterations) != 0)
735 return;
736
737 spin_lock(&o2hb_live_lock);
738
739 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
740 goto unlock;
741
43182d2a
SM		 742 /*
743 * A region can be added to the quorum only when it sees all
744 * live nodes heartbeat on it. In other words, the region has been
745 * added to all nodes.
746 */
747 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
748 sizeof(o2hb_live_node_bitmap)))
d2eece37 749 goto unlock;
43182d2a 750
d2eece37
SM		 751 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
752 config_item_name(&reg->hr_item), reg->hr_dev_name);
43182d2a
SM		 753
754 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
58a3158a
SM		 755
756 /*
757 * If global heartbeat active, unpin all regions if the
758 * region count > CUT_OFF
759 */
a8f70de3 760 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 761 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
762 o2hb_region_unpin(NULL);
d2eece37
SM		 763unlock:
764 spin_unlock(&o2hb_live_lock);
43182d2a
SM		 765}
766
a7f6a5fb
MF		 767static int o2hb_check_slot(struct o2hb_region *reg,
768 struct o2hb_disk_slot *slot)
769{
770 int changed = 0, gen_changed = 0;
771 struct o2hb_node_event event =
772 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
773 struct o2nm_node *node;
774 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
775 u64 cputime;
0db638f4
MF		 776 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
777 unsigned int slot_dead_ms;
0e105d37 778 int tmp;
6f8648e8 779 int queued = 0;
a7f6a5fb
MF		 780
781 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
782
0e105d37
SM		 783 /*
784 * If a node is no longer configured but is still in the livemap, we
785 * may need to clear that bit from the livemap.
786 */
a7f6a5fb 787 node = o2nm_get_node_by_num(slot->ds_node_num);
0e105d37
SM		 788 if (!node) {
789 spin_lock(&o2hb_live_lock);
790 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
791 spin_unlock(&o2hb_live_lock);
792 if (!tmp)
793 return 0;
794 }
a7f6a5fb
MF		 795
796 if (!o2hb_verify_crc(reg, hb_block)) {
797 /* all paths from here will drop o2hb_live_lock for
798 * us. */
799 spin_lock(&o2hb_live_lock);
800
801 /* Don't print an error on the console in this case -
802 * a freshly formatted heartbeat area will not have a
803 * crc set on it. */
804 if (list_empty(&slot->ds_live_item))
805 goto out;
806
807 /* The node is live but pushed out a bad crc. We
808 * consider it a transient miss but don't populate any
809 * other values as they may be junk. */
810 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
811 slot->ds_node_num, reg->hr_dev_name);
812 o2hb_dump_slot(hb_block);
813
814 slot->ds_equal_samples++;
815 goto fire_callbacks;
816 }
817
818 /* we don't care if these wrap.. the state transitions below
819 * clear at the right places */
820 cputime = le64_to_cpu(hb_block->hb_seq);
821 if (slot->ds_last_time != cputime)
822 slot->ds_changed_samples++;
823 else
824 slot->ds_equal_samples++;
825 slot->ds_last_time = cputime;
826
827 /* The node changed heartbeat generations. We assume this to
828 * mean it dropped off but came back before we timed out. We
829 * want to consider it down for the time being but don't want
830 * to lose any changed_samples state we might build up to
831 * considering it live again. */
832 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
833 gen_changed = 1;
834 slot->ds_equal_samples = 0;
70bacbdb
MF		 835 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
836 "to 0x%llx)\n", slot->ds_node_num,
837 (long long)slot->ds_last_generation,
838 (long long)le64_to_cpu(hb_block->hb_generation));
a7f6a5fb
MF		 839 }
840
841 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
842
70bacbdb
MF		 843 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
844 "seq %llu last %llu changed %u equal %u\n",
845 slot->ds_node_num, (long long)slot->ds_last_generation,
846 le32_to_cpu(hb_block->hb_cksum),
2bd63216 847 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
70bacbdb 848 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
a7f6a5fb
MF		 849 slot->ds_equal_samples);
850
851 spin_lock(&o2hb_live_lock);
852
853fire_callbacks:
854 /* dead nodes only come to life after some number of
855 * changes at any time during their dead time */
856 if (list_empty(&slot->ds_live_item) &&
857 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
70bacbdb
MF		 858 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
859 slot->ds_node_num, (long long)slot->ds_last_generation);
a7f6a5fb 860
823a637a
SM		 861 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
862
a7f6a5fb
MF		 863 /* first on the list generates a callback */
864 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
d6aa1c7c
SM		 865 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
866 "bitmap\n", slot->ds_node_num);
a7f6a5fb
MF		 867 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
868
869 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
870 slot->ds_node_num);
871
872 changed = 1;
6f8648e8 873 queued = 1;
a7f6a5fb
MF		 874 }
875
876 list_add_tail(&slot->ds_live_item,
877 &o2hb_live_slots[slot->ds_node_num]);
878
879 slot->ds_equal_samples = 0;
0db638f4
MF		 880
881 /* We want to be sure that all nodes agree on the
882 * number of milliseconds before a node will be
883 * considered dead. The self-fencing timeout is
884 * computed from this value, and a discrepancy might
885 * result in heartbeat calling a node dead when it
886 * hasn't self-fenced yet. */
887 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
888 if (slot_dead_ms && slot_dead_ms != dead_ms) {
889 /* TODO: Perhaps we can fail the region here. */
890 mlog(ML_ERROR, "Node %d on device %s has a dead count "
891 "of %u ms, but our count is %u ms.\n"
892 "Please double check your configuration values "
893 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
894 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
895 dead_ms);
896 }
a7f6a5fb
MF		 897 goto out;
898 }
899
900 /* if the list is dead, we're done.. */
901 if (list_empty(&slot->ds_live_item))
902 goto out;
903
904 /* live nodes only go dead after enough consequtive missed
905 * samples.. reset the missed counter whenever we see
906 * activity */
907 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
908 mlog(ML_HEARTBEAT, "Node %d left my region\n",
909 slot->ds_node_num);
910
823a637a
SM		 911 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
912
a7f6a5fb
MF		 913 /* last off the live_slot generates a callback */
914 list_del_init(&slot->ds_live_item);
915 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
d6aa1c7c
SM		 916 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
917 "nodes bitmap\n", slot->ds_node_num);
a7f6a5fb
MF		 918 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
919
0e105d37
SM		 920 /* node can be null */
921 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
922 node, slot->ds_node_num);
a7f6a5fb
MF		 923
924 changed = 1;
6f8648e8 925 queued = 1;
a7f6a5fb
MF		 926 }
927
928 /* We don't clear this because the node is still
929 * actually writing new blocks. */
930 if (!gen_changed)
931 slot->ds_changed_samples = 0;
932 goto out;
933 }
934 if (slot->ds_changed_samples) {
935 slot->ds_changed_samples = 0;
936 slot->ds_equal_samples = 0;
937 }
938out:
939 spin_unlock(&o2hb_live_lock);
940
6f8648e8
J		 941 if (queued)
942 o2hb_run_event_list(&event);
a7f6a5fb 943
0e105d37
SM		 944 if (node)
945 o2nm_node_put(node);
a7f6a5fb
MF		 946 return changed;
947}
948
518df6bd 949static int o2hb_highest_node(unsigned long *nodes, int numbits)
a7f6a5fb 950{
518df6bd 951 return find_last_bit(nodes, numbits);
a7f6a5fb
MF		 952}
953
a9e2ae39 954static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
a7f6a5fb 955{
d2eece37
SM		 956 int i, ret, highest_node;
957 int membership_change = 0, own_slot_ok = 0;
a7f6a5fb 958 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
0e105d37 959 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
a7f6a5fb
MF		 960 struct o2hb_bio_wait_ctxt write_wc;
961
a9e2ae39
MF		 962 ret = o2nm_configured_node_map(configured_nodes,
963 sizeof(configured_nodes));
964 if (ret) {
965 mlog_errno(ret);
d2eece37 966 goto bail;
a9e2ae39 967 }
a7f6a5fb 968
0e105d37
SM		 969 /*
970 * If a node is not configured but is in the livemap, we still need
971 * to read the slot so as to be able to remove it from the livemap.
972 */
973 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
974 i = -1;
975 while ((i = find_next_bit(live_node_bitmap,
976 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
977 set_bit(i, configured_nodes);
978 }
979
a7f6a5fb
MF		 980 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
981 if (highest_node >= O2NM_MAX_NODES) {
d2eece37
SM		 982 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
983 ret = -EINVAL;
984 goto bail;
a7f6a5fb
MF		 985 }
986
987 /* No sense in reading the slots of nodes that don't exist
988 * yet. Of course, if the node definitions have holes in them
989 * then we're reading an empty slot anyway... Consider this
990 * best-effort. */
991 ret = o2hb_read_slots(reg, highest_node + 1);
992 if (ret < 0) {
993 mlog_errno(ret);
d2eece37 994 goto bail;
a7f6a5fb
MF		 995 }
996
997 /* With an up to date view of the slots, we can check that no
998 * other node has been improperly configured to heartbeat in
999 * our slot. */
d2eece37 1000 own_slot_ok = o2hb_check_own_slot(reg);
a7f6a5fb
MF		 1001
1002 /* fill in the proper info for our next heartbeat */
1003 o2hb_prepare_block(reg, reg->hr_generation);
1004
b559292e 1005 ret = o2hb_issue_node_write(reg, &write_wc);
a7f6a5fb
MF		 1006 if (ret < 0) {
1007 mlog_errno(ret);
d2eece37 1008 goto bail;
a7f6a5fb
MF		 1009 }
1010
1011 i = -1;
33c12a54
SM		 1012 while((i = find_next_bit(configured_nodes,
1013 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
d2eece37 1014 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
a7f6a5fb
MF		 1015 }
1016
1017 /*
1018 * We have to be sure we've advertised ourselves on disk
1019 * before we can go to steady state. This ensures that
1020 * people we find in our steady state have seen us.
1021 */
1022 o2hb_wait_on_io(reg, &write_wc);
a9e2ae39
MF		 1023 if (write_wc.wc_error) {
1024 /* Do not re-arm the write timeout on I/O error - we
1025 * can't be sure that the new block ever made it to
1026 * disk */
1027 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1028 write_wc.wc_error, reg->hr_dev_name);
d2eece37
SM		 1029 ret = write_wc.wc_error;
1030 goto bail;
a9e2ae39
MF		 1031 }
1032
d2eece37
SM		 1033 /* Skip disarming the timeout if own slot has stale/bad data */
1034 if (own_slot_ok) {
1035 o2hb_set_quorum_device(reg);
1036 o2hb_arm_write_timeout(reg);
1037 }
a7f6a5fb 1038
d2eece37 1039bail:
a7f6a5fb 1040 /* let the person who launched us know when things are steady */
d2eece37
SM		 1041 if (atomic_read(&reg->hr_steady_iterations) != 0) {
1042 if (!ret && own_slot_ok && !membership_change) {
1043 if (atomic_dec_and_test(&reg->hr_steady_iterations))
1044 wake_up(&o2hb_steady_queue);
1045 }
1046 }
1047
1048 if (atomic_read(&reg->hr_steady_iterations) != 0) {
1049 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1050 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1051 "heartbeart on region %s (%s)\n",
1052 config_item_name(&reg->hr_item),
1053 reg->hr_dev_name);
1054 atomic_set(&reg->hr_steady_iterations, 0);
1055 reg->hr_aborted_start = 1;
a7f6a5fb 1056 wake_up(&o2hb_steady_queue);
d2eece37
SM		 1057 ret = -EIO;
1058 }
a7f6a5fb 1059 }
a9e2ae39 1060
d2eece37 1061 return ret;
a7f6a5fb
MF		 1062}
1063
1064/* Subtract b from a, storing the result in a. a *must* have a larger
1065 * value than b. */
1066static void o2hb_tv_subtract(struct timeval *a,
1067 struct timeval *b)
1068{
1069 /* just return 0 when a is after b */
1070 if (a->tv_sec < b->tv_sec ||
1071 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1072 a->tv_sec = 0;
1073 a->tv_usec = 0;
1074 return;
1075 }
1076
1077 a->tv_sec -= b->tv_sec;
1078 a->tv_usec -= b->tv_usec;
1079 while ( a->tv_usec < 0 ) {
1080 a->tv_sec--;
1081 a->tv_usec += 1000000;
1082 }
1083}
1084
1085static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1086 struct timeval *end)
1087{
1088 struct timeval res = *end;
1089
1090 o2hb_tv_subtract(&res, start);
1091
1092 return res.tv_sec * 1000 + res.tv_usec / 1000;
1093}
1094
1095/*
1096 * we ride the region ref that the region dir holds. before the region
1097 * dir is removed and drops it ref it will wait to tear down this
1098 * thread.
1099 */
1100static int o2hb_thread(void *data)
1101{
1102 int i, ret;
1103 struct o2hb_region *reg = data;
a7f6a5fb
MF		 1104 struct o2hb_bio_wait_ctxt write_wc;
1105 struct timeval before_hb, after_hb;
1106 unsigned int elapsed_msec;
1107
1108 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1109
1110 set_user_nice(current, -20);
1111
cfc069d3
SM		 1112 /* Pin node */
1113 o2nm_depend_this_node();
1114
d2eece37
SM		 1115 while (!kthread_should_stop() &&
1116 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
a7f6a5fb 1117 /* We track the time spent inside
025dfdaf 1118 * o2hb_do_disk_heartbeat so that we avoid more than
a7f6a5fb
MF		 1119 * hr_timeout_ms between disk writes. On busy systems
1120 * this should result in a heartbeat which is less
1121 * likely to time itself out. */
1122 do_gettimeofday(&before_hb);
1123
d2eece37 1124 ret = o2hb_do_disk_heartbeat(reg);
a7f6a5fb
MF		 1125
1126 do_gettimeofday(&after_hb);
1127 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1128
b31d308d
TM		 1129 mlog(ML_HEARTBEAT,
1130 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
215c7f9f
MF		 1131 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1132 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1133 elapsed_msec);
a7f6a5fb 1134
d2eece37
SM		 1135 if (!kthread_should_stop() &&
1136 elapsed_msec < reg->hr_timeout_ms) {
a7f6a5fb
MF		 1137 /* the kthread api has blocked signals for us so no
1138 * need to record the return value. */
1139 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1140 }
1141 }
1142
1143 o2hb_disarm_write_timeout(reg);
1144
1145 /* unclean stop is only used in very bad situation */
1146 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1147 o2hb_shutdown_slot(&reg->hr_slots[i]);
1148
1149 /* Explicit down notification - avoid forcing the other nodes
1150 * to timeout on this region when we could just as easily
1151 * write a clear generation - thus indicating to them that
1152 * this node has left this region.
d2eece37
SM		 1153 */
1154 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1155 o2hb_prepare_block(reg, 0);
1156 ret = o2hb_issue_node_write(reg, &write_wc);
1157 if (ret == 0)
1158 o2hb_wait_on_io(reg, &write_wc);
1159 else
1160 mlog_errno(ret);
a7f6a5fb
MF		 1161 }
1162
cfc069d3
SM		 1163 /* Unpin node */
1164 o2nm_undepend_this_node();
a7f6a5fb 1165
d2eece37 1166 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
a7f6a5fb
MF		 1167
1168 return 0;
1169}
1170
87d3d3f3
SM		 1171#ifdef CONFIG_DEBUG_FS
1172static int o2hb_debug_open(struct inode *inode, struct file *file)
1173{
8ca8b0bb 1174 struct o2hb_debug_buf *db = inode->i_private;
1f285305 1175 struct o2hb_region *reg;
87d3d3f3 1176 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
bb570a5d 1177 unsigned long lts;
87d3d3f3
SM		 1178 char *buf = NULL;
1179 int i = -1;
1180 int out = 0;
1181
8ca8b0bb
SM		 1182 /* max_nodes should be the largest bitmap we pass here */
1183 BUG_ON(sizeof(map) < db->db_size);
1184
87d3d3f3
SM		 1185 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1186 if (!buf)
1187 goto bail;
1188
8ca8b0bb
SM		 1189 switch (db->db_type) {
1190 case O2HB_DB_TYPE_LIVENODES:
a6de0136
SM		 1191 case O2HB_DB_TYPE_LIVEREGIONS:
1192 case O2HB_DB_TYPE_QUORUMREGIONS:
1193 case O2HB_DB_TYPE_FAILEDREGIONS:
8ca8b0bb
SM		 1194 spin_lock(&o2hb_live_lock);
1195 memcpy(map, db->db_data, db->db_size);
1196 spin_unlock(&o2hb_live_lock);
1197 break;
87d3d3f3 1198
1f285305
SM		 1199 case O2HB_DB_TYPE_REGION_LIVENODES:
1200 spin_lock(&o2hb_live_lock);
1201 reg = (struct o2hb_region *)db->db_data;
1202 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1203 spin_unlock(&o2hb_live_lock);
1204 break;
1205
1206 case O2HB_DB_TYPE_REGION_NUMBER:
1207 reg = (struct o2hb_region *)db->db_data;
1208 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1209 reg->hr_region_num);
1210 goto done;
1211
43695d09
SM		 1212 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1213 reg = (struct o2hb_region *)db->db_data;
bb570a5d
SM		 1214 lts = reg->hr_last_timeout_start;
1215 /* If 0, it has never been set before */
1216 if (lts)
1217 lts = jiffies_to_msecs(jiffies - lts);
1218 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
43695d09
SM		 1219 goto done;
1220
cb0586bd
SM		 1221 case O2HB_DB_TYPE_REGION_PINNED:
1222 reg = (struct o2hb_region *)db->db_data;
1223 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1224 !!reg->hr_item_pinned);
1225 goto done;
1226
8ca8b0bb
SM		 1227 default:
1228 goto done;
1229 }
1230
1231 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
87d3d3f3
SM		 1232 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1233 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1234
8ca8b0bb 1235done:
87d3d3f3
SM		 1236 i_size_write(inode, out);
1237
1238 file->private_data = buf;
1239
1240 return 0;
1241bail:
1242 return -ENOMEM;
1243}
1244
1245static int o2hb_debug_release(struct inode *inode, struct file *file)
1246{
1247 kfree(file->private_data);
1248 return 0;
1249}
1250
1251static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1252 size_t nbytes, loff_t *ppos)
1253{
1254 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1255 i_size_read(file->f_mapping->host));
1256}
1257#else
1258static int o2hb_debug_open(struct inode *inode, struct file *file)
1259{
1260 return 0;
1261}
1262static int o2hb_debug_release(struct inode *inode, struct file *file)
1263{
1264 return 0;
1265}
1266static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1267 size_t nbytes, loff_t *ppos)
1268{
1269 return 0;
1270}
1271#endif /* CONFIG_DEBUG_FS */
1272
828c0950 1273static const struct file_operations o2hb_debug_fops = {
87d3d3f3
SM		 1274 .open = o2hb_debug_open,
1275 .release = o2hb_debug_release,
1276 .read = o2hb_debug_read,
1277 .llseek = generic_file_llseek,
1278};
1279
1280void o2hb_exit(void)
1281{
8ca8b0bb 1282 kfree(o2hb_db_livenodes);
a6de0136
SM		 1283 kfree(o2hb_db_liveregions);
1284 kfree(o2hb_db_quorumregions);
1285 kfree(o2hb_db_failedregions);
1286 debugfs_remove(o2hb_debug_failedregions);
1287 debugfs_remove(o2hb_debug_quorumregions);
1288 debugfs_remove(o2hb_debug_liveregions);
8ca8b0bb
SM		 1289 debugfs_remove(o2hb_debug_livenodes);
1290 debugfs_remove(o2hb_debug_dir);
1291}
1292
1293static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1294 struct o2hb_debug_buf **db, int db_len,
1295 int type, int size, int len, void *data)
1296{
1297 *db = kmalloc(db_len, GFP_KERNEL);
1298 if (!*db)
1299 return NULL;
1300
1301 (*db)->db_type = type;
1302 (*db)->db_size = size;
1303 (*db)->db_len = len;
1304 (*db)->db_data = data;
1305
1306 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1307 &o2hb_debug_fops);
1308}
1309
1310static int o2hb_debug_init(void)
1311{
1312 int ret = -ENOMEM;
1313
1314 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1315 if (!o2hb_debug_dir) {
1316 mlog_errno(ret);
1317 goto bail;
1318 }
1319
1320 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1321 o2hb_debug_dir,
1322 &o2hb_db_livenodes,
1323 sizeof(*o2hb_db_livenodes),
1324 O2HB_DB_TYPE_LIVENODES,
1325 sizeof(o2hb_live_node_bitmap),
1326 O2NM_MAX_NODES,
1327 o2hb_live_node_bitmap);
1328 if (!o2hb_debug_livenodes) {
1329 mlog_errno(ret);
1330 goto bail;
1331 }
a6de0136
SM		 1332
1333 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1334 o2hb_debug_dir,
1335 &o2hb_db_liveregions,
1336 sizeof(*o2hb_db_liveregions),
1337 O2HB_DB_TYPE_LIVEREGIONS,
1338 sizeof(o2hb_live_region_bitmap),
1339 O2NM_MAX_REGIONS,
1340 o2hb_live_region_bitmap);
1341 if (!o2hb_debug_liveregions) {
1342 mlog_errno(ret);
1343 goto bail;
1344 }
1345
1346 o2hb_debug_quorumregions =
1347 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1348 o2hb_debug_dir,
1349 &o2hb_db_quorumregions,
1350 sizeof(*o2hb_db_quorumregions),
1351 O2HB_DB_TYPE_QUORUMREGIONS,
1352 sizeof(o2hb_quorum_region_bitmap),
1353 O2NM_MAX_REGIONS,
1354 o2hb_quorum_region_bitmap);
1355 if (!o2hb_debug_quorumregions) {
1356 mlog_errno(ret);
1357 goto bail;
1358 }
1359
1360 o2hb_debug_failedregions =
1361 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1362 o2hb_debug_dir,
1363 &o2hb_db_failedregions,
1364 sizeof(*o2hb_db_failedregions),
1365 O2HB_DB_TYPE_FAILEDREGIONS,
1366 sizeof(o2hb_failed_region_bitmap),
1367 O2NM_MAX_REGIONS,
1368 o2hb_failed_region_bitmap);
1369 if (!o2hb_debug_failedregions) {
1370 mlog_errno(ret);
1371 goto bail;
1372 }
1373
8ca8b0bb
SM		 1374 ret = 0;
1375bail:
1376 if (ret)
1377 o2hb_exit();
1378
1379 return ret;
87d3d3f3
SM		 1380}
1381
1382int o2hb_init(void)
a7f6a5fb
MF		 1383{
1384 int i;
1385
1386 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1387 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1388
1389 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1390 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1391
1392 INIT_LIST_HEAD(&o2hb_node_events);
1393
1394 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
536f0741 1395 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
e7d656ba 1396 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
43182d2a 1397 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
b1c5ebfb 1398 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
87d3d3f3 1399
58a3158a
SM		 1400 o2hb_dependent_users = 0;
1401
8ca8b0bb 1402 return o2hb_debug_init();
a7f6a5fb
MF		 1403}
1404
1405/* if we're already in a callback then we're already serialized by the sem */
1406static void o2hb_fill_node_map_from_callback(unsigned long *map,
1407 unsigned bytes)
1408{
1409 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1410
1411 memcpy(map, &o2hb_live_node_bitmap, bytes);
1412}
1413
1414/*
1415 * get a map of all nodes that are heartbeating in any regions
1416 */
1417void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1418{
1419 /* callers want to serialize this map and callbacks so that they
1420 * can trust that they don't miss nodes coming to the party */
1421 down_read(&o2hb_callback_sem);
1422 spin_lock(&o2hb_live_lock);
1423 o2hb_fill_node_map_from_callback(map, bytes);
1424 spin_unlock(&o2hb_live_lock);
1425 up_read(&o2hb_callback_sem);
1426}
1427EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1428
1429/*
1430 * heartbeat configfs bits. The heartbeat set is a default set under
1431 * the cluster set in nodemanager.c.
1432 */
1433
1434static struct o2hb_region *to_o2hb_region(struct config_item *item)
1435{
1436 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1437}
1438
1439/* drop_item only drops its ref after killing the thread, nothing should
1440 * be using the region anymore. this has to clean up any state that
1441 * attributes might have built up. */
1442static void o2hb_region_release(struct config_item *item)
1443{
1444 int i;
1445 struct page *page;
1446 struct o2hb_region *reg = to_o2hb_region(item);
1447
d2eece37
SM		 1448 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1449
d787ab09 1450 kfree(reg->hr_tmp_block);
a7f6a5fb
MF		 1451
1452 if (reg->hr_slot_data) {
1453 for (i = 0; i < reg->hr_num_pages; i++) {
1454 page = reg->hr_slot_data[i];
1455 if (page)
1456 __free_page(page);
1457 }
1458 kfree(reg->hr_slot_data);
1459 }
1460
1461 if (reg->hr_bdev)
9a1c3542 1462 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb 1463
d787ab09 1464 kfree(reg->hr_slots);
a7f6a5fb 1465
1f285305
SM		 1466 kfree(reg->hr_db_regnum);
1467 kfree(reg->hr_db_livenodes);
1468 debugfs_remove(reg->hr_debug_livenodes);
1469 debugfs_remove(reg->hr_debug_regnum);
d4396eaf 1470 debugfs_remove(reg->hr_debug_elapsed_time);
cb0586bd 1471 debugfs_remove(reg->hr_debug_pinned);
1f285305
SM		 1472 debugfs_remove(reg->hr_debug_dir);
1473
a7f6a5fb
MF		 1474 spin_lock(&o2hb_live_lock);
1475 list_del(&reg->hr_all_item);
1476 spin_unlock(&o2hb_live_lock);
1477
1478 kfree(reg);
1479}
1480
1481static int o2hb_read_block_input(struct o2hb_region *reg,
1482 const char *page,
1483 size_t count,
1484 unsigned long *ret_bytes,
1485 unsigned int *ret_bits)
1486{
1487 unsigned long bytes;
1488 char *p = (char *)page;
1489
1490 bytes = simple_strtoul(p, &p, 0);
1491 if (!p || (*p && (*p != '\n')))
1492 return -EINVAL;
1493
1494 /* Heartbeat and fs min / max block sizes are the same. */
1495 if (bytes > 4096 || bytes < 512)
1496 return -ERANGE;
1497 if (hweight16(bytes) != 1)
1498 return -EINVAL;
1499
1500 if (ret_bytes)
1501 *ret_bytes = bytes;
1502 if (ret_bits)
1503 *ret_bits = ffs(bytes) - 1;
1504
1505 return 0;
1506}
1507
1508static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1509 char *page)
1510{
1511 return sprintf(page, "%u\n", reg->hr_block_bytes);
1512}
1513
1514static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1515 const char *page,
1516 size_t count)
1517{
1518 int status;
1519 unsigned long block_bytes;
1520 unsigned int block_bits;
1521
1522 if (reg->hr_bdev)
1523 return -EINVAL;
1524
1525 status = o2hb_read_block_input(reg, page, count,
1526 &block_bytes, &block_bits);
1527 if (status)
1528 return status;
1529
1530 reg->hr_block_bytes = (unsigned int)block_bytes;
1531 reg->hr_block_bits = block_bits;
1532
1533 return count;
1534}
1535
1536static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1537 char *page)
1538{
1539 return sprintf(page, "%llu\n", reg->hr_start_block);
1540}
1541
1542static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1543 const char *page,
1544 size_t count)
1545{
1546 unsigned long long tmp;
1547 char *p = (char *)page;
1548
1549 if (reg->hr_bdev)
1550 return -EINVAL;
1551
1552 tmp = simple_strtoull(p, &p, 0);
1553 if (!p || (*p && (*p != '\n')))
1554 return -EINVAL;
1555
1556 reg->hr_start_block = tmp;
1557
1558 return count;
1559}
1560
1561static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1562 char *page)
1563{
1564 return sprintf(page, "%d\n", reg->hr_blocks);
1565}
1566
1567static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1568 const char *page,
1569 size_t count)
1570{
1571 unsigned long tmp;
1572 char *p = (char *)page;
1573
1574 if (reg->hr_bdev)
1575 return -EINVAL;
1576
1577 tmp = simple_strtoul(p, &p, 0);
1578 if (!p || (*p && (*p != '\n')))
1579 return -EINVAL;
1580
1581 if (tmp > O2NM_MAX_NODES || tmp == 0)
1582 return -ERANGE;
1583
1584 reg->hr_blocks = (unsigned int)tmp;
1585
1586 return count;
1587}
1588
1589static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1590 char *page)
1591{
1592 unsigned int ret = 0;
1593
1594 if (reg->hr_bdev)
1595 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1596
1597 return ret;
1598}
1599
1600static void o2hb_init_region_params(struct o2hb_region *reg)
1601{
1602 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1603 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1604
1605 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1606 reg->hr_start_block, reg->hr_blocks);
1607 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1608 reg->hr_block_bytes, reg->hr_block_bits);
1609 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1610 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1611}
1612
1613static int o2hb_map_slot_data(struct o2hb_region *reg)
1614{
1615 int i, j;
1616 unsigned int last_slot;
1617 unsigned int spp = reg->hr_slots_per_page;
1618 struct page *page;
1619 char *raw;
1620 struct o2hb_disk_slot *slot;
1621
1622 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1623 if (reg->hr_tmp_block == NULL) {
1624 mlog_errno(-ENOMEM);
1625 return -ENOMEM;
1626 }
1627
1628 reg->hr_slots = kcalloc(reg->hr_blocks,
1629 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1630 if (reg->hr_slots == NULL) {
1631 mlog_errno(-ENOMEM);
1632 return -ENOMEM;
1633 }
1634
1635 for(i = 0; i < reg->hr_blocks; i++) {
1636 slot = &reg->hr_slots[i];
1637 slot->ds_node_num = i;
1638 INIT_LIST_HEAD(&slot->ds_live_item);
1639 slot->ds_raw_block = NULL;
1640 }
1641
1642 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1643 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1644 "at %u blocks per page\n",
1645 reg->hr_num_pages, reg->hr_blocks, spp);
1646
1647 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1648 GFP_KERNEL);
1649 if (!reg->hr_slot_data) {
1650 mlog_errno(-ENOMEM);
1651 return -ENOMEM;
1652 }
1653
1654 for(i = 0; i < reg->hr_num_pages; i++) {
1655 page = alloc_page(GFP_KERNEL);
1656 if (!page) {
1657 mlog_errno(-ENOMEM);
1658 return -ENOMEM;
1659 }
1660
1661 reg->hr_slot_data[i] = page;
1662
1663 last_slot = i * spp;
1664 raw = page_address(page);
1665 for (j = 0;
1666 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1667 j++) {
1668 BUG_ON((j + last_slot) >= reg->hr_blocks);
1669
1670 slot = &reg->hr_slots[j + last_slot];
1671 slot->ds_raw_block =
1672 (struct o2hb_disk_heartbeat_block *) raw;
1673
1674 raw += reg->hr_block_bytes;
1675 }
1676 }
1677
1678 return 0;
1679}
1680
1681/* Read in all the slots available and populate the tracking
1682 * structures so that we can start with a baseline idea of what's
1683 * there. */
1684static int o2hb_populate_slot_data(struct o2hb_region *reg)
1685{
1686 int ret, i;
1687 struct o2hb_disk_slot *slot;
1688 struct o2hb_disk_heartbeat_block *hb_block;
1689
a7f6a5fb
MF		 1690 ret = o2hb_read_slots(reg, reg->hr_blocks);
1691 if (ret) {
1692 mlog_errno(ret);
1693 goto out;
1694 }
1695
1696 /* We only want to get an idea of the values initially in each
1697 * slot, so we do no verification - o2hb_check_slot will
1698 * actually determine if each configured slot is valid and
1699 * whether any values have changed. */
1700 for(i = 0; i < reg->hr_blocks; i++) {
1701 slot = &reg->hr_slots[i];
1702 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1703
1704 /* Only fill the values that o2hb_check_slot uses to
1705 * determine changing slots */
1706 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1707 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1708 }
1709
1710out:
a7f6a5fb
MF		 1711 return ret;
1712}
1713
1714/* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1715static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1716 const char *page,
1717 size_t count)
1718{
e6c352db 1719 struct task_struct *hb_task;
a7f6a5fb
MF		 1720 long fd;
1721 int sectsize;
1722 char *p = (char *)page;
2903ff01
AV		 1723 struct fd f;
1724 struct inode *inode;
a7f6a5fb 1725 ssize_t ret = -EINVAL;
76d9fc29 1726 int live_threshold;
a7f6a5fb
MF		 1727
1728 if (reg->hr_bdev)
1729 goto out;
1730
1731 /* We can't heartbeat without having had our node number
1732 * configured yet. */
1733 if (o2nm_this_node() == O2NM_MAX_NODES)
1734 goto out;
1735
1736 fd = simple_strtol(p, &p, 0);
1737 if (!p || (*p && (*p != '\n')))
1738 goto out;
1739
1740 if (fd < 0 || fd >= INT_MAX)
1741 goto out;
1742
2903ff01
AV		 1743 f = fdget(fd);
1744 if (f.file == NULL)
a7f6a5fb
MF		 1745 goto out;
1746
1747 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1748 reg->hr_block_bytes == 0)
2903ff01 1749 goto out2;
a7f6a5fb 1750
2903ff01 1751 inode = igrab(f.file->f_mapping->host);
a7f6a5fb 1752 if (inode == NULL)
2903ff01 1753 goto out2;
a7f6a5fb
MF		 1754
1755 if (!S_ISBLK(inode->i_mode))
2903ff01 1756 goto out3;
a7f6a5fb 1757
2903ff01 1758 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
e525fd89 1759 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
a7f6a5fb
MF		 1760 if (ret) {
1761 reg->hr_bdev = NULL;
2903ff01 1762 goto out3;
a7f6a5fb
MF		 1763 }
1764 inode = NULL;
1765
1766 bdevname(reg->hr_bdev, reg->hr_dev_name);
1767
e1defc4f 1768 sectsize = bdev_logical_block_size(reg->hr_bdev);
a7f6a5fb
MF		 1769 if (sectsize != reg->hr_block_bytes) {
1770 mlog(ML_ERROR,
1771 "blocksize %u incorrect for device, expected %d",
1772 reg->hr_block_bytes, sectsize);
1773 ret = -EINVAL;
2903ff01 1774 goto out3;
a7f6a5fb
MF		 1775 }
1776
1777 o2hb_init_region_params(reg);
1778
1779 /* Generation of zero is invalid */
1780 do {
1781 get_random_bytes(&reg->hr_generation,
1782 sizeof(reg->hr_generation));
1783 } while (reg->hr_generation == 0);
1784
1785 ret = o2hb_map_slot_data(reg);
1786 if (ret) {
1787 mlog_errno(ret);
2903ff01 1788 goto out3;
a7f6a5fb
MF		 1789 }
1790
1791 ret = o2hb_populate_slot_data(reg);
1792 if (ret) {
1793 mlog_errno(ret);
2903ff01 1794 goto out3;
a7f6a5fb
MF		 1795 }
1796
c4028958 1797 INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
a7f6a5fb
MF		 1798
1799 /*
1800 * A node is considered live after it has beat LIVE_THRESHOLD
1801 * times. We're not steady until we've given them a chance
1802 * _after_ our first read.
76d9fc29
SM		 1803 * The default threshold is bare minimum so as to limit the delay
1804 * during mounts. For global heartbeat, the threshold doubled for the
1805 * first region.
a7f6a5fb 1806 */
76d9fc29
SM		 1807 live_threshold = O2HB_LIVE_THRESHOLD;
1808 if (o2hb_global_heartbeat_active()) {
1809 spin_lock(&o2hb_live_lock);
a8f70de3 1810 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
76d9fc29
SM		 1811 live_threshold <<= 1;
1812 spin_unlock(&o2hb_live_lock);
1813 }
d2eece37
SM		 1814 ++live_threshold;
1815 atomic_set(&reg->hr_steady_iterations, live_threshold);
1816 /* unsteady_iterations is double the steady_iterations */
1817 atomic_set(&reg->hr_unsteady_iterations, (live_threshold << 1));
a7f6a5fb 1818
e6c352db
JB		 1819 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1820 reg->hr_item.ci_name);
1821 if (IS_ERR(hb_task)) {
1822 ret = PTR_ERR(hb_task);
a7f6a5fb 1823 mlog_errno(ret);
2903ff01 1824 goto out3;
a7f6a5fb
MF		 1825 }
1826
e6c352db
JB		 1827 spin_lock(&o2hb_live_lock);
1828 reg->hr_task = hb_task;
1829 spin_unlock(&o2hb_live_lock);
1830
a7f6a5fb
MF		 1831 ret = wait_event_interruptible(o2hb_steady_queue,
1832 atomic_read(&reg->hr_steady_iterations) == 0);
1833 if (ret) {
d2eece37
SM		 1834 atomic_set(&reg->hr_steady_iterations, 0);
1835 reg->hr_aborted_start = 1;
1836 }
e6c352db 1837
d2eece37
SM		 1838 if (reg->hr_aborted_start) {
1839 ret = -EIO;
2903ff01 1840 goto out3;
a7f6a5fb
MF		 1841 }
1842
e6df3a66
JB		 1843 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1844 spin_lock(&o2hb_live_lock);
1845 hb_task = reg->hr_task;
e7d656ba
SM		 1846 if (o2hb_global_heartbeat_active())
1847 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
e6df3a66
JB		 1848 spin_unlock(&o2hb_live_lock);
1849
1850 if (hb_task)
1851 ret = count;
1852 else
1853 ret = -EIO;
1854
18c50cb0 1855 if (hb_task && o2hb_global_heartbeat_active())
d2eece37
SM		 1856 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1857 config_item_name(&reg->hr_item), reg->hr_dev_name);
18c50cb0 1858
2903ff01
AV		 1859out3:
1860 iput(inode);
1861out2:
1862 fdput(f);
a7f6a5fb 1863out:
a7f6a5fb
MF		 1864 if (ret < 0) {
1865 if (reg->hr_bdev) {
9a1c3542 1866 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb
MF		 1867 reg->hr_bdev = NULL;
1868 }
1869 }
1870 return ret;
1871}
1872
92efc152
ZW		 1873static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1874 char *page)
1875{
e6c352db
JB		 1876 pid_t pid = 0;
1877
1878 spin_lock(&o2hb_live_lock);
1879 if (reg->hr_task)
ba25f9dc 1880 pid = task_pid_nr(reg->hr_task);
e6c352db
JB		 1881 spin_unlock(&o2hb_live_lock);
1882
1883 if (!pid)
92efc152
ZW		 1884 return 0;
1885
e6c352db 1886 return sprintf(page, "%u\n", pid);
92efc152
ZW		 1887}
1888
a7f6a5fb
MF		 1889struct o2hb_region_attribute {
1890 struct configfs_attribute attr;
1891 ssize_t (*show)(struct o2hb_region *, char *);
1892 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1893};
1894
1895static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1896 .attr = { .ca_owner = THIS_MODULE,
1897 .ca_name = "block_bytes",
1898 .ca_mode = S_IRUGO | S_IWUSR },
1899 .show = o2hb_region_block_bytes_read,
1900 .store = o2hb_region_block_bytes_write,
1901};
1902
1903static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1904 .attr = { .ca_owner = THIS_MODULE,
1905 .ca_name = "start_block",
1906 .ca_mode = S_IRUGO | S_IWUSR },
1907 .show = o2hb_region_start_block_read,
1908 .store = o2hb_region_start_block_write,
1909};
1910
1911static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1912 .attr = { .ca_owner = THIS_MODULE,
1913 .ca_name = "blocks",
1914 .ca_mode = S_IRUGO | S_IWUSR },
1915 .show = o2hb_region_blocks_read,
1916 .store = o2hb_region_blocks_write,
1917};
1918
1919static struct o2hb_region_attribute o2hb_region_attr_dev = {
1920 .attr = { .ca_owner = THIS_MODULE,
1921 .ca_name = "dev",
1922 .ca_mode = S_IRUGO | S_IWUSR },
1923 .show = o2hb_region_dev_read,
1924 .store = o2hb_region_dev_write,
1925};
1926
92efc152
ZW		 1927static struct o2hb_region_attribute o2hb_region_attr_pid = {
1928 .attr = { .ca_owner = THIS_MODULE,
1929 .ca_name = "pid",
1930 .ca_mode = S_IRUGO | S_IRUSR },
1931 .show = o2hb_region_pid_read,
1932};
1933
a7f6a5fb
MF		 1934static struct configfs_attribute *o2hb_region_attrs[] = {
1935 &o2hb_region_attr_block_bytes.attr,
1936 &o2hb_region_attr_start_block.attr,
1937 &o2hb_region_attr_blocks.attr,
1938 &o2hb_region_attr_dev.attr,
92efc152 1939 &o2hb_region_attr_pid.attr,
a7f6a5fb
MF		 1940 NULL,
1941};
1942
1943static ssize_t o2hb_region_show(struct config_item *item,
1944 struct configfs_attribute *attr,
1945 char *page)
1946{
1947 struct o2hb_region *reg = to_o2hb_region(item);
1948 struct o2hb_region_attribute *o2hb_region_attr =
1949 container_of(attr, struct o2hb_region_attribute, attr);
1950 ssize_t ret = 0;
1951
1952 if (o2hb_region_attr->show)
1953 ret = o2hb_region_attr->show(reg, page);
1954 return ret;
1955}
1956
1957static ssize_t o2hb_region_store(struct config_item *item,
1958 struct configfs_attribute *attr,
1959 const char *page, size_t count)
1960{
1961 struct o2hb_region *reg = to_o2hb_region(item);
1962 struct o2hb_region_attribute *o2hb_region_attr =
1963 container_of(attr, struct o2hb_region_attribute, attr);
1964 ssize_t ret = -EINVAL;
1965
1966 if (o2hb_region_attr->store)
1967 ret = o2hb_region_attr->store(reg, page, count);
1968 return ret;
1969}
1970
1971static struct configfs_item_operations o2hb_region_item_ops = {
1972 .release = o2hb_region_release,
1973 .show_attribute = o2hb_region_show,
1974 .store_attribute = o2hb_region_store,
1975};
1976
1977static struct config_item_type o2hb_region_type = {
1978 .ct_item_ops = &o2hb_region_item_ops,
1979 .ct_attrs = o2hb_region_attrs,
1980 .ct_owner = THIS_MODULE,
1981};
1982
1983/* heartbeat set */
1984
1985struct o2hb_heartbeat_group {
1986 struct config_group hs_group;
1987 /* some stuff? */
1988};
1989
1990static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1991{
1992 return group ?
1993 container_of(group, struct o2hb_heartbeat_group, hs_group)
1994 : NULL;
1995}
1996
1f285305
SM		 1997static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1998{
1999 int ret = -ENOMEM;
2000
2001 reg->hr_debug_dir =
2002 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
2003 if (!reg->hr_debug_dir) {
2004 mlog_errno(ret);
2005 goto bail;
2006 }
2007
2008 reg->hr_debug_livenodes =
2009 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2010 reg->hr_debug_dir,
2011 &(reg->hr_db_livenodes),
2012 sizeof(*(reg->hr_db_livenodes)),
2013 O2HB_DB_TYPE_REGION_LIVENODES,
2014 sizeof(reg->hr_live_node_bitmap),
2015 O2NM_MAX_NODES, reg);
2016 if (!reg->hr_debug_livenodes) {
2017 mlog_errno(ret);
2018 goto bail;
2019 }
2020
2021 reg->hr_debug_regnum =
2022 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2023 reg->hr_debug_dir,
2024 &(reg->hr_db_regnum),
2025 sizeof(*(reg->hr_db_regnum)),
2026 O2HB_DB_TYPE_REGION_NUMBER,
2027 0, O2NM_MAX_NODES, reg);
2028 if (!reg->hr_debug_regnum) {
2029 mlog_errno(ret);
2030 goto bail;
2031 }
2032
43695d09
SM		 2033 reg->hr_debug_elapsed_time =
2034 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2035 reg->hr_debug_dir,
2036 &(reg->hr_db_elapsed_time),
2037 sizeof(*(reg->hr_db_elapsed_time)),
2038 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2039 0, 0, reg);
2040 if (!reg->hr_debug_elapsed_time) {
2041 mlog_errno(ret);
2042 goto bail;
2043 }
2044
cb0586bd
SM		 2045 reg->hr_debug_pinned =
2046 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2047 reg->hr_debug_dir,
2048 &(reg->hr_db_pinned),
2049 sizeof(*(reg->hr_db_pinned)),
2050 O2HB_DB_TYPE_REGION_PINNED,
2051 0, 0, reg);
2052 if (!reg->hr_debug_pinned) {
2053 mlog_errno(ret);
2054 goto bail;
2055 }
2056
1f285305
SM		 2057 ret = 0;
2058bail:
2059 return ret;
2060}
2061
f89ab861
JB		 2062static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2063 const char *name)
a7f6a5fb
MF		 2064{
2065 struct o2hb_region *reg = NULL;
1f285305 2066 int ret;
a7f6a5fb 2067
cd861280 2068 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
f89ab861 2069 if (reg == NULL)
a6795e9e 2070 return ERR_PTR(-ENOMEM);
a7f6a5fb 2071
1cf257f5
JS		 2072 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2073 ret = -ENAMETOOLONG;
2074 goto free;
2075 }
b3c85c4c 2076
a7f6a5fb 2077 spin_lock(&o2hb_live_lock);
536f0741
SM		 2078 reg->hr_region_num = 0;
2079 if (o2hb_global_heartbeat_active()) {
2080 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2081 O2NM_MAX_REGIONS);
2082 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2083 spin_unlock(&o2hb_live_lock);
1cf257f5
JS		 2084 ret = -EFBIG;
2085 goto free;
536f0741
SM		 2086 }
2087 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2088 }
a7f6a5fb
MF		 2089 list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2090 spin_unlock(&o2hb_live_lock);
a7f6a5fb 2091
536f0741
SM		 2092 config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2093
1f285305
SM		 2094 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2095 if (ret) {
2096 config_item_put(&reg->hr_item);
1cf257f5 2097 goto free;
1f285305
SM		 2098 }
2099
a6795e9e 2100 return &reg->hr_item;
1cf257f5
JS		 2101free:
2102 kfree(reg);
2103 return ERR_PTR(ret);
a7f6a5fb
MF		 2104}
2105
2106static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2107 struct config_item *item)
2108{
e6c352db 2109 struct task_struct *hb_task;
a7f6a5fb 2110 struct o2hb_region *reg = to_o2hb_region(item);
58a3158a 2111 int quorum_region = 0;
a7f6a5fb
MF		 2112
2113 /* stop the thread when the user removes the region dir */
e6c352db
JB		 2114 spin_lock(&o2hb_live_lock);
2115 hb_task = reg->hr_task;
2116 reg->hr_task = NULL;
58a3158a 2117 reg->hr_item_dropped = 1;
e6c352db
JB		 2118 spin_unlock(&o2hb_live_lock);
2119
2120 if (hb_task)
2121 kthread_stop(hb_task);
a7f6a5fb 2122
d2eece37
SM		 2123 if (o2hb_global_heartbeat_active()) {
2124 spin_lock(&o2hb_live_lock);
2125 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2126 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2127 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2128 quorum_region = 1;
2129 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2130 spin_unlock(&o2hb_live_lock);
2131 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2132 ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2133 "stopped" : "start aborted"), config_item_name(item),
2134 reg->hr_dev_name);
2135 }
2136
e6df3a66
JB		 2137 /*
2138 * If we're racing a dev_write(), we need to wake them. They will
2139 * check reg->hr_task
2140 */
2141 if (atomic_read(&reg->hr_steady_iterations) != 0) {
d2eece37 2142 reg->hr_aborted_start = 1;
e6df3a66
JB		 2143 atomic_set(&reg->hr_steady_iterations, 0);
2144 wake_up(&o2hb_steady_queue);
2145 }
2146
a7f6a5fb 2147 config_item_put(item);
58a3158a
SM		 2148
2149 if (!o2hb_global_heartbeat_active() || !quorum_region)
2150 return;
2151
2152 /*
2153 * If global heartbeat active and there are dependent users,
2154 * pin all regions if quorum region count <= CUT_OFF
2155 */
2156 spin_lock(&o2hb_live_lock);
2157
2158 if (!o2hb_dependent_users)
2159 goto unlock;
2160
a8f70de3 2161 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 2162 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2163 o2hb_region_pin(NULL);
2164
2165unlock:
2166 spin_unlock(&o2hb_live_lock);
a7f6a5fb
MF		 2167}
2168
2169struct o2hb_heartbeat_group_attribute {
2170 struct configfs_attribute attr;
2171 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2172 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2173};
2174
2175static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2176 struct configfs_attribute *attr,
2177 char *page)
2178{
2179 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2180 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2181 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2182 ssize_t ret = 0;
2183
2184 if (o2hb_heartbeat_group_attr->show)
2185 ret = o2hb_heartbeat_group_attr->show(reg, page);
2186 return ret;
2187}
2188
2189static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2190 struct configfs_attribute *attr,
2191 const char *page, size_t count)
2192{
2193 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2194 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2195 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2196 ssize_t ret = -EINVAL;
2197
2198 if (o2hb_heartbeat_group_attr->store)
2199 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2200 return ret;
2201}
2202
2203static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2204 char *page)
2205{
2206 return sprintf(page, "%u\n", o2hb_dead_threshold);
2207}
2208
2209static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2210 const char *page,
2211 size_t count)
2212{
2213 unsigned long tmp;
2214 char *p = (char *)page;
2215
2216 tmp = simple_strtoul(p, &p, 10);
2217 if (!p || (*p && (*p != '\n')))
2218 return -EINVAL;
2219
2220 /* this will validate ranges for us. */
2221 o2hb_dead_threshold_set((unsigned int) tmp);
2222
2223 return count;
2224}
2225
54b5187b
SM		 2226static
2227ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2228 char *page)
2229{
2230 return sprintf(page, "%s\n",
2231 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2232}
2233
2234static
2235ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2236 const char *page, size_t count)
2237{
2238 unsigned int i;
2239 int ret;
2240 size_t len;
2241
2242 len = (page[count - 1] == '\n') ? count - 1 : count;
2243 if (!len)
2244 return -EINVAL;
2245
2246 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2247 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2248 continue;
2249
70f651ed 2250 ret = o2hb_global_heartbeat_mode_set(i);
54b5187b 2251 if (!ret)
18c50cb0 2252 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
54b5187b
SM		 2253 o2hb_heartbeat_mode_desc[i]);
2254 return count;
2255 }
2256
2257 return -EINVAL;
2258
2259}
2260
a7f6a5fb
MF		 2261static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2262 .attr = { .ca_owner = THIS_MODULE,
2263 .ca_name = "dead_threshold",
2264 .ca_mode = S_IRUGO | S_IWUSR },
2265 .show = o2hb_heartbeat_group_threshold_show,
2266 .store = o2hb_heartbeat_group_threshold_store,
2267};
2268
54b5187b
SM		 2269static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2270 .attr = { .ca_owner = THIS_MODULE,
2271 .ca_name = "mode",
2272 .ca_mode = S_IRUGO | S_IWUSR },
2273 .show = o2hb_heartbeat_group_mode_show,
2274 .store = o2hb_heartbeat_group_mode_store,
2275};
2276
a7f6a5fb
MF		 2277static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2278 &o2hb_heartbeat_group_attr_threshold.attr,
54b5187b 2279 &o2hb_heartbeat_group_attr_mode.attr,
a7f6a5fb
MF		 2280 NULL,
2281};
2282
70f651ed 2283static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
a7f6a5fb
MF		 2284 .show_attribute = o2hb_heartbeat_group_show,
2285 .store_attribute = o2hb_heartbeat_group_store,
2286};
2287
2288static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2289 .make_item = o2hb_heartbeat_group_make_item,
2290 .drop_item = o2hb_heartbeat_group_drop_item,
2291};
2292
2293static struct config_item_type o2hb_heartbeat_group_type = {
2294 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
70f651ed 2295 .ct_item_ops = &o2hb_heartbeat_group_item_ops,
a7f6a5fb
MF		 2296 .ct_attrs = o2hb_heartbeat_group_attrs,
2297 .ct_owner = THIS_MODULE,
2298};
2299
2300/* this is just here to avoid touching group in heartbeat.h which the
2301 * entire damn world #includes */
2302struct config_group *o2hb_alloc_hb_set(void)
2303{
2304 struct o2hb_heartbeat_group *hs = NULL;
2305 struct config_group *ret = NULL;
2306
cd861280 2307 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
a7f6a5fb
MF		 2308 if (hs == NULL)
2309 goto out;
2310
2311 config_group_init_type_name(&hs->hs_group, "heartbeat",
2312 &o2hb_heartbeat_group_type);
2313
2314 ret = &hs->hs_group;
2315out:
2316 if (ret == NULL)
2317 kfree(hs);
2318 return ret;
2319}
2320
2321void o2hb_free_hb_set(struct config_group *group)
2322{
2323 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2324 kfree(hs);
2325}
2326
25985edc 2327/* hb callback registration and issuing */
a7f6a5fb
MF		 2328
2329static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2330{
2331 if (type == O2HB_NUM_CB)
2332 return ERR_PTR(-EINVAL);
2333
2334 return &o2hb_callbacks[type];
2335}
2336
2337void o2hb_setup_callback(struct o2hb_callback_func *hc,
2338 enum o2hb_callback_type type,
2339 o2hb_cb_func *func,
2340 void *data,
2341 int priority)
2342{
2343 INIT_LIST_HEAD(&hc->hc_item);
2344 hc->hc_func = func;
2345 hc->hc_data = data;
2346 hc->hc_priority = priority;
2347 hc->hc_type = type;
2348 hc->hc_magic = O2HB_CB_MAGIC;
2349}
2350EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2351
58a3158a
SM		 2352/*
2353 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2354 * In global heartbeat mode, region_uuid passed is NULL.
2355 *
2356 * In local, we only pin the matching region. In global we pin all the active
2357 * regions.
2358 */
2359static int o2hb_region_pin(const char *region_uuid)
14829422 2360{
58a3158a
SM		 2361 int ret = 0, found = 0;
2362 struct o2hb_region *reg;
2363 char *uuid;
14829422
JB		 2364
2365 assert_spin_locked(&o2hb_live_lock);
2366
58a3158a 2367 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2368 if (reg->hr_item_dropped)
2369 continue;
2370
58a3158a
SM		 2371 uuid = config_item_name(&reg->hr_item);
2372
2373 /* local heartbeat */
2374 if (region_uuid) {
2375 if (strcmp(region_uuid, uuid))
2376 continue;
2377 found = 1;
14829422 2378 }
58a3158a
SM		 2379
2380 if (reg->hr_item_pinned || reg->hr_item_dropped)
2381 goto skip_pin;
2382
2383 /* Ignore ENOENT only for local hb (userdlm domain) */
2384 ret = o2nm_depend_item(&reg->hr_item);
2385 if (!ret) {
2386 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2387 reg->hr_item_pinned = 1;
2388 } else {
2389 if (ret == -ENOENT && found)
2390 ret = 0;
2391 else {
2392 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2393 uuid, ret);
2394 break;
2395 }
14829422 2396 }
58a3158a
SM		 2397skip_pin:
2398 if (found)
2399 break;
14829422
JB		 2400 }
2401
58a3158a 2402 return ret;
14829422
JB		 2403}
2404
58a3158a
SM		 2405/*
2406 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2407 * In global heartbeat mode, region_uuid passed is NULL.
2408 *
2409 * In local, we only unpin the matching region. In global we unpin all the
2410 * active regions.
2411 */
2412static void o2hb_region_unpin(const char *region_uuid)
14829422 2413{
14829422 2414 struct o2hb_region *reg;
58a3158a
SM		 2415 char *uuid;
2416 int found = 0;
14829422 2417
58a3158a 2418 assert_spin_locked(&o2hb_live_lock);
14829422 2419
58a3158a 2420 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2421 if (reg->hr_item_dropped)
2422 continue;
2423
58a3158a
SM		 2424 uuid = config_item_name(&reg->hr_item);
2425 if (region_uuid) {
2426 if (strcmp(region_uuid, uuid))
2427 continue;
2428 found = 1;
2429 }
14829422 2430
58a3158a
SM		 2431 if (reg->hr_item_pinned) {
2432 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2433 o2nm_undepend_item(&reg->hr_item);
2434 reg->hr_item_pinned = 0;
2435 }
2436 if (found)
2437 break;
2438 }
2439}
16c6a4f2 2440
58a3158a
SM		 2441static int o2hb_region_inc_user(const char *region_uuid)
2442{
2443 int ret = 0;
14829422 2444
58a3158a 2445 spin_lock(&o2hb_live_lock);
16c6a4f2 2446
58a3158a
SM		 2447 /* local heartbeat */
2448 if (!o2hb_global_heartbeat_active()) {
2449 ret = o2hb_region_pin(region_uuid);
2450 goto unlock;
2451 }
2452
2453 /*
2454 * if global heartbeat active and this is the first dependent user,
2455 * pin all regions if quorum region count <= CUT_OFF
2456 */
2457 o2hb_dependent_users++;
2458 if (o2hb_dependent_users > 1)
2459 goto unlock;
2460
a8f70de3 2461 if (bitmap_weight(o2hb_quorum_region_bitmap,
58a3158a
SM		 2462 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2463 ret = o2hb_region_pin(NULL);
2464
2465unlock:
2466 spin_unlock(&o2hb_live_lock);
14829422
JB		 2467 return ret;
2468}
2469
58a3158a 2470void o2hb_region_dec_user(const char *region_uuid)
14829422 2471{
14829422
JB		 2472 spin_lock(&o2hb_live_lock);
2473
58a3158a
SM		 2474 /* local heartbeat */
2475 if (!o2hb_global_heartbeat_active()) {
2476 o2hb_region_unpin(region_uuid);
2477 goto unlock;
2478 }
14829422 2479
58a3158a
SM		 2480 /*
2481 * if global heartbeat active and there are no dependent users,
2482 * unpin all quorum regions
2483 */
2484 o2hb_dependent_users--;
2485 if (!o2hb_dependent_users)
2486 o2hb_region_unpin(NULL);
14829422 2487
58a3158a
SM		 2488unlock:
2489 spin_unlock(&o2hb_live_lock);
14829422
JB		 2490}
2491
2492int o2hb_register_callback(const char *region_uuid,
2493 struct o2hb_callback_func *hc)
a7f6a5fb 2494{
df53cd3b 2495 struct o2hb_callback_func *f;
a7f6a5fb
MF		 2496 struct o2hb_callback *hbcall;
2497 int ret;
2498
2499 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2500 BUG_ON(!list_empty(&hc->hc_item));
2501
2502 hbcall = hbcall_from_type(hc->hc_type);
2503 if (IS_ERR(hbcall)) {
2504 ret = PTR_ERR(hbcall);
2505 goto out;
2506 }
2507
14829422 2508 if (region_uuid) {
58a3158a
SM		 2509 ret = o2hb_region_inc_user(region_uuid);
2510 if (ret) {
2511 mlog_errno(ret);
14829422 2512 goto out;
58a3158a 2513 }
14829422
JB		 2514 }
2515
a7f6a5fb
MF		 2516 down_write(&o2hb_callback_sem);
2517
df53cd3b
DF		 2518 list_for_each_entry(f, &hbcall->list, hc_item) {
2519 if (hc->hc_priority < f->hc_priority) {
2520 list_add_tail(&hc->hc_item, &f->hc_item);
a7f6a5fb
MF		 2521 break;
2522 }
2523 }
2524 if (list_empty(&hc->hc_item))
2525 list_add_tail(&hc->hc_item, &hbcall->list);
2526
2527 up_write(&o2hb_callback_sem);
2528 ret = 0;
2529out:
58a3158a 2530 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2531 ret, __builtin_return_address(0), hc);
2532 return ret;
2533}
2534EXPORT_SYMBOL_GPL(o2hb_register_callback);
2535
14829422
JB		 2536void o2hb_unregister_callback(const char *region_uuid,
2537 struct o2hb_callback_func *hc)
a7f6a5fb
MF		 2538{
2539 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2540
58a3158a 2541 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2542 __builtin_return_address(0), hc);
2543
14829422 2544 /* XXX Can this happen _with_ a region reference? */
a7f6a5fb 2545 if (list_empty(&hc->hc_item))
c24f72cc 2546 return;
a7f6a5fb 2547
14829422 2548 if (region_uuid)
58a3158a 2549 o2hb_region_dec_user(region_uuid);
14829422 2550
a7f6a5fb
MF		 2551 down_write(&o2hb_callback_sem);
2552
2553 list_del_init(&hc->hc_item);
2554
2555 up_write(&o2hb_callback_sem);
a7f6a5fb
MF		 2556}
2557EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2558
2559int o2hb_check_node_heartbeating(u8 node_num)
2560{
2561 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2562
2563 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2564 if (!test_bit(node_num, testing_map)) {
2565 mlog(ML_HEARTBEAT,
2566 "node (%u) does not have heartbeating enabled.\n",
2567 node_num);
2568 return 0;
2569 }
2570
2571 return 1;
2572}
2573EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2574
2575int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2576{
2577 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2578
2579 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2580 if (!test_bit(node_num, testing_map)) {
2581 mlog(ML_HEARTBEAT,
2582 "node (%u) does not have heartbeating enabled.\n",
2583 node_num);
2584 return 0;
2585 }
2586
2587 return 1;
2588}
2589EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2590
2591/* Makes sure our local node is configured with a node number, and is
2592 * heartbeating. */
2593int o2hb_check_local_node_heartbeating(void)
2594{
2595 u8 node_num;
2596
2597 /* if this node was set then we have networking */
2598 node_num = o2nm_this_node();
2599 if (node_num == O2NM_MAX_NODES) {
2600 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2601 return 0;
2602 }
2603
2604 return o2hb_check_node_heartbeating(node_num);
2605}
2606EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2607
2608/*
2609 * this is just a hack until we get the plumbing which flips file systems
2610 * read only and drops the hb ref instead of killing the node dead.
2611 */
2612void o2hb_stop_all_regions(void)
2613{
2614 struct o2hb_region *reg;
2615
2616 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2617
2618 spin_lock(&o2hb_live_lock);
2619
2620 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2621 reg->hr_unclean_stop = 1;
2622
2623 spin_unlock(&o2hb_live_lock);
2624}
2625EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
b3c85c4c
SM		 2626
2627int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2628{
2629 struct o2hb_region *reg;
2630 int numregs = 0;
2631 char *p;
2632
2633 spin_lock(&o2hb_live_lock);
2634
2635 p = region_uuids;
2636 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
4a184b4f
X		 2637 if (reg->hr_item_dropped)
2638 continue;
2639
b3c85c4c
SM		 2640 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2641 if (numregs < max_regions) {
2642 memcpy(p, config_item_name(&reg->hr_item),
2643 O2HB_MAX_REGION_NAME_LEN);
2644 p += O2HB_MAX_REGION_NAME_LEN;
2645 }
2646 numregs++;
2647 }
2648
2649 spin_unlock(&o2hb_live_lock);
2650
2651 return numregs;
2652}
2653EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2654
2655int o2hb_global_heartbeat_active(void)
2656{
4d94aa1b 2657 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
b3c85c4c
SM		 2658}
2659EXPORT_SYMBOL(o2hb_global_heartbeat_active);