projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: adjust switch_case syntax at o2net_state_change()
[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>
a7f6a5fb
MF		 38
39#include "heartbeat.h"
40#include "tcp.h"
41#include "nodemanager.h"
42#include "quorum.h"
43
44#include "masklog.h"
45
46
47/*
48 * The first heartbeat pass had one global thread that would serialize all hb
49 * callback calls. This global serializing sem should only be removed once
50 * we've made sure that all callees can deal with being called concurrently
51 * from multiple hb region threads.
52 */
53static DECLARE_RWSEM(o2hb_callback_sem);
54
55/*
56 * multiple hb threads are watching multiple regions. A node is live
57 * whenever any of the threads sees activity from the node in its region.
58 */
34af946a 59static DEFINE_SPINLOCK(o2hb_live_lock);
a7f6a5fb
MF		 60static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
61static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
62static LIST_HEAD(o2hb_node_events);
63static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
64
536f0741
SM		 65/*
66 * In global heartbeat, we maintain a series of region bitmaps.
67 * - o2hb_region_bitmap allows us to limit the region number to max region.
e7d656ba 68 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
43182d2a
SM		 69 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
70 * heartbeat on it.
b1c5ebfb 71 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
536f0741
SM		 72 */
73static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
e7d656ba 74static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
43182d2a 75static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
b1c5ebfb 76static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
536f0741 77
8ca8b0bb 78#define O2HB_DB_TYPE_LIVENODES 0
a6de0136
SM		 79#define O2HB_DB_TYPE_LIVEREGIONS 1
80#define O2HB_DB_TYPE_QUORUMREGIONS 2
81#define O2HB_DB_TYPE_FAILEDREGIONS 3
1f285305
SM		 82#define O2HB_DB_TYPE_REGION_LIVENODES 4
83#define O2HB_DB_TYPE_REGION_NUMBER 5
43695d09 84#define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
cb0586bd 85#define O2HB_DB_TYPE_REGION_PINNED 7
8ca8b0bb
SM		 86struct o2hb_debug_buf {
87 int db_type;
88 int db_size;
89 int db_len;
90 void *db_data;
91};
92
93static struct o2hb_debug_buf *o2hb_db_livenodes;
a6de0136
SM		 94static struct o2hb_debug_buf *o2hb_db_liveregions;
95static struct o2hb_debug_buf *o2hb_db_quorumregions;
96static struct o2hb_debug_buf *o2hb_db_failedregions;
8ca8b0bb 97
87d3d3f3
SM		 98#define O2HB_DEBUG_DIR "o2hb"
99#define O2HB_DEBUG_LIVENODES "livenodes"
a6de0136
SM		 100#define O2HB_DEBUG_LIVEREGIONS "live_regions"
101#define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
102#define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
1f285305 103#define O2HB_DEBUG_REGION_NUMBER "num"
43695d09 104#define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
cb0586bd 105#define O2HB_DEBUG_REGION_PINNED "pinned"
8ca8b0bb 106
87d3d3f3
SM		 107static struct dentry *o2hb_debug_dir;
108static struct dentry *o2hb_debug_livenodes;
a6de0136
SM		 109static struct dentry *o2hb_debug_liveregions;
110static struct dentry *o2hb_debug_quorumregions;
111static struct dentry *o2hb_debug_failedregions;
87d3d3f3 112
a7f6a5fb
MF		 113static LIST_HEAD(o2hb_all_regions);
114
115static struct o2hb_callback {
116 struct list_head list;
117} o2hb_callbacks[O2HB_NUM_CB];
118
119static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
120
121#define O2HB_DEFAULT_BLOCK_BITS 9
122
54b5187b
SM		 123enum o2hb_heartbeat_modes {
124 O2HB_HEARTBEAT_LOCAL = 0,
125 O2HB_HEARTBEAT_GLOBAL,
126 O2HB_HEARTBEAT_NUM_MODES,
127};
128
129char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
130 "local", /* O2HB_HEARTBEAT_LOCAL */
131 "global", /* O2HB_HEARTBEAT_GLOBAL */
132};
133
a7f6a5fb 134unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
54b5187b 135unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
a7f6a5fb 136
58a3158a
SM		 137/*
138 * o2hb_dependent_users tracks the number of registered callbacks that depend
139 * on heartbeat. o2net and o2dlm are two entities that register this callback.
140 * However only o2dlm depends on the heartbeat. It does not want the heartbeat
141 * to stop while a dlm domain is still active.
142 */
143unsigned int o2hb_dependent_users;
144
145/*
146 * In global heartbeat mode, all regions are pinned if there are one or more
147 * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
148 * regions are unpinned if the region count exceeds the cut off or the number
149 * of dependent users falls to zero.
150 */
151#define O2HB_PIN_CUT_OFF 3
152
153/*
154 * In local heartbeat mode, we assume the dlm domain name to be the same as
155 * region uuid. This is true for domains created for the file system but not
156 * necessarily true for userdlm domains. This is a known limitation.
157 *
158 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
159 * works for both file system and userdlm domains.
160 */
161static int o2hb_region_pin(const char *region_uuid);
162static void o2hb_region_unpin(const char *region_uuid);
163
2bd63216 164/* Only sets a new threshold if there are no active regions.
a7f6a5fb
MF		 165 *
166 * No locking or otherwise interesting code is required for reading
167 * o2hb_dead_threshold as it can't change once regions are active and
168 * it's not interesting to anyone until then anyway. */
169static void o2hb_dead_threshold_set(unsigned int threshold)
170{
171 if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
172 spin_lock(&o2hb_live_lock);
173 if (list_empty(&o2hb_all_regions))
174 o2hb_dead_threshold = threshold;
175 spin_unlock(&o2hb_live_lock);
176 }
177}
178
70f651ed 179static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
54b5187b
SM		 180{
181 int ret = -1;
182
183 if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
184 spin_lock(&o2hb_live_lock);
185 if (list_empty(&o2hb_all_regions)) {
186 o2hb_heartbeat_mode = hb_mode;
187 ret = 0;
188 }
189 spin_unlock(&o2hb_live_lock);
190 }
191
192 return ret;
193}
194
a7f6a5fb
MF		 195struct o2hb_node_event {
196 struct list_head hn_item;
197 enum o2hb_callback_type hn_event_type;
198 struct o2nm_node *hn_node;
199 int hn_node_num;
200};
201
202struct o2hb_disk_slot {
203 struct o2hb_disk_heartbeat_block *ds_raw_block;
204 u8 ds_node_num;
205 u64 ds_last_time;
206 u64 ds_last_generation;
207 u16 ds_equal_samples;
208 u16 ds_changed_samples;
209 struct list_head ds_live_item;
210};
211
212/* each thread owns a region.. when we're asked to tear down the region
213 * we ask the thread to stop, who cleans up the region */
214struct o2hb_region {
215 struct config_item hr_item;
216
217 struct list_head hr_all_item;
58a3158a 218 unsigned hr_unclean_stop:1,
d2eece37 219 hr_aborted_start:1,
58a3158a
SM		 220 hr_item_pinned:1,
221 hr_item_dropped:1;
a7f6a5fb
MF		 222
223 /* protected by the hr_callback_sem */
224 struct task_struct *hr_task;
225
226 unsigned int hr_blocks;
227 unsigned long long hr_start_block;
228
229 unsigned int hr_block_bits;
230 unsigned int hr_block_bytes;
231
232 unsigned int hr_slots_per_page;
233 unsigned int hr_num_pages;
234
235 struct page **hr_slot_data;
236 struct block_device *hr_bdev;
237 struct o2hb_disk_slot *hr_slots;
238
823a637a
SM		 239 /* live node map of this region */
240 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
536f0741 241 unsigned int hr_region_num;
823a637a 242
1f285305
SM		 243 struct dentry *hr_debug_dir;
244 struct dentry *hr_debug_livenodes;
245 struct dentry *hr_debug_regnum;
43695d09 246 struct dentry *hr_debug_elapsed_time;
cb0586bd 247 struct dentry *hr_debug_pinned;
1f285305
SM		 248 struct o2hb_debug_buf *hr_db_livenodes;
249 struct o2hb_debug_buf *hr_db_regnum;
43695d09 250 struct o2hb_debug_buf *hr_db_elapsed_time;
cb0586bd 251 struct o2hb_debug_buf *hr_db_pinned;
1f285305 252
a7f6a5fb
MF		 253 /* let the person setting up hb wait for it to return until it
254 * has reached a 'steady' state. This will be fixed when we have
255 * a more complete api that doesn't lead to this sort of fragility. */
256 atomic_t hr_steady_iterations;
257
d2eece37
SM		 258 /* terminate o2hb thread if it does not reach steady state
259 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
260 atomic_t hr_unsteady_iterations;
261
a7f6a5fb
MF		 262 char hr_dev_name[BDEVNAME_SIZE];
263
264 unsigned int hr_timeout_ms;
265
266 /* randomized as the region goes up and down so that a node
267 * recognizes a node going up and down in one iteration */
268 u64 hr_generation;
269
c4028958 270 struct delayed_work hr_write_timeout_work;
a7f6a5fb
MF		 271 unsigned long hr_last_timeout_start;
272
273 /* Used during o2hb_check_slot to hold a copy of the block
274 * being checked because we temporarily have to zero out the
275 * crc field. */
276 struct o2hb_disk_heartbeat_block *hr_tmp_block;
277};
278
279struct o2hb_bio_wait_ctxt {
280 atomic_t wc_num_reqs;
281 struct completion wc_io_complete;
a9e2ae39 282 int wc_error;
a7f6a5fb
MF		 283};
284
b1c5ebfb
SM		 285static int o2hb_pop_count(void *map, int count)
286{
287 int i = -1, pop = 0;
288
289 while ((i = find_next_bit(map, count, i + 1)) < count)
290 pop++;
291 return pop;
292}
293
c4028958 294static void o2hb_write_timeout(struct work_struct *work)
a7f6a5fb 295{
b1c5ebfb
SM		 296 int failed, quorum;
297 unsigned long flags;
c4028958
DH		 298 struct o2hb_region *reg =
299 container_of(work, struct o2hb_region,
300 hr_write_timeout_work.work);
a7f6a5fb
MF		 301
302 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
303 "milliseconds\n", reg->hr_dev_name,
2bd63216 304 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
b1c5ebfb
SM		 305
306 if (o2hb_global_heartbeat_active()) {
307 spin_lock_irqsave(&o2hb_live_lock, flags);
308 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
309 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
310 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
311 O2NM_MAX_REGIONS);
312 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
313 O2NM_MAX_REGIONS);
314 spin_unlock_irqrestore(&o2hb_live_lock, flags);
315
316 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
317 quorum, failed);
318
319 /*
320 * Fence if the number of failed regions >= half the number
321 * of quorum regions
322 */
323 if ((failed << 1) < quorum)
324 return;
325 }
326
a7f6a5fb
MF		 327 o2quo_disk_timeout();
328}
329
330static void o2hb_arm_write_timeout(struct o2hb_region *reg)
331{
d2eece37
SM		 332 /* Arm writeout only after thread reaches steady state */
333 if (atomic_read(&reg->hr_steady_iterations) != 0)
334 return;
335
b31d308d
TM		 336 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
337 O2HB_MAX_WRITE_TIMEOUT_MS);
a7f6a5fb 338
b1c5ebfb
SM		 339 if (o2hb_global_heartbeat_active()) {
340 spin_lock(&o2hb_live_lock);
341 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
342 spin_unlock(&o2hb_live_lock);
343 }
a7f6a5fb
MF		 344 cancel_delayed_work(&reg->hr_write_timeout_work);
345 reg->hr_last_timeout_start = jiffies;
346 schedule_delayed_work(&reg->hr_write_timeout_work,
347 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
348}
349
350static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
351{
9b00a818 352 cancel_delayed_work_sync(&reg->hr_write_timeout_work);
a7f6a5fb
MF		 353}
354
b559292e 355static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
a7f6a5fb 356{
b559292e 357 atomic_set(&wc->wc_num_reqs, 1);
a7f6a5fb 358 init_completion(&wc->wc_io_complete);
a9e2ae39 359 wc->wc_error = 0;
a7f6a5fb
MF		 360}
361
362/* Used in error paths too */
363static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
364 unsigned int num)
365{
366 /* sadly atomic_sub_and_test() isn't available on all platforms. The
367 * good news is that the fast path only completes one at a time */
368 while(num--) {
369 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
370 BUG_ON(num > 0);
371 complete(&wc->wc_io_complete);
372 }
373 }
374}
375
376static void o2hb_wait_on_io(struct o2hb_region *reg,
377 struct o2hb_bio_wait_ctxt *wc)
378{
b559292e 379 o2hb_bio_wait_dec(wc, 1);
a7f6a5fb
MF		 380 wait_for_completion(&wc->wc_io_complete);
381}
382
782e3b3b 383static void o2hb_bio_end_io(struct bio *bio,
a7f6a5fb
MF		 384 int error)
385{
386 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
387
a9e2ae39 388 if (error) {
a7f6a5fb 389 mlog(ML_ERROR, "IO Error %d\n", error);
a9e2ae39
MF		 390 wc->wc_error = error;
391 }
a7f6a5fb 392
a7f6a5fb 393 o2hb_bio_wait_dec(wc, 1);
b559292e 394 bio_put(bio);
a7f6a5fb
MF		 395}
396
397/* Setup a Bio to cover I/O against num_slots slots starting at
398 * start_slot. */
399static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
400 struct o2hb_bio_wait_ctxt *wc,
b559292e
PR		 401 unsigned int *current_slot,
402 unsigned int max_slots)
a7f6a5fb 403{
b559292e 404 int len, current_page;
a7f6a5fb
MF		 405 unsigned int vec_len, vec_start;
406 unsigned int bits = reg->hr_block_bits;
407 unsigned int spp = reg->hr_slots_per_page;
b559292e 408 unsigned int cs = *current_slot;
a7f6a5fb
MF		 409 struct bio *bio;
410 struct page *page;
411
a7f6a5fb
MF		 412 /* Testing has shown this allocation to take long enough under
413 * GFP_KERNEL that the local node can get fenced. It would be
414 * nicest if we could pre-allocate these bios and avoid this
415 * all together. */
b559292e 416 bio = bio_alloc(GFP_ATOMIC, 16);
a7f6a5fb
MF		 417 if (!bio) {
418 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
419 bio = ERR_PTR(-ENOMEM);
420 goto bail;
421 }
422
423 /* Must put everything in 512 byte sectors for the bio... */
b559292e 424 bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
a7f6a5fb
MF		 425 bio->bi_bdev = reg->hr_bdev;
426 bio->bi_private = wc;
427 bio->bi_end_io = o2hb_bio_end_io;
428
b559292e
PR		 429 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
430 while(cs < max_slots) {
431 current_page = cs / spp;
432 page = reg->hr_slot_data[current_page];
a7f6a5fb 433
bc7e97cb 434 vec_len = min(PAGE_CACHE_SIZE - vec_start,
b559292e 435 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
a7f6a5fb
MF		 436
437 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
b559292e 438 current_page, vec_len, vec_start);
a7f6a5fb
MF		 439
440 len = bio_add_page(bio, page, vec_len, vec_start);
b559292e 441 if (len != vec_len) break;
a7f6a5fb 442
b559292e 443 cs += vec_len / (PAGE_CACHE_SIZE/spp);
a7f6a5fb
MF		 444 vec_start = 0;
445 }
446
447bail:
b559292e 448 *current_slot = cs;
a7f6a5fb
MF		 449 return bio;
450}
451
a7f6a5fb
MF		 452static int o2hb_read_slots(struct o2hb_region *reg,
453 unsigned int max_slots)
454{
b559292e
PR		 455 unsigned int current_slot=0;
456 int status;
a7f6a5fb 457 struct o2hb_bio_wait_ctxt wc;
a7f6a5fb
MF		 458 struct bio *bio;
459
b559292e 460 o2hb_bio_wait_init(&wc);
a7f6a5fb 461
b559292e
PR		 462 while(current_slot < max_slots) {
463 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
a7f6a5fb 464 if (IS_ERR(bio)) {
a7f6a5fb
MF		 465 status = PTR_ERR(bio);
466 mlog_errno(status);
467 goto bail_and_wait;
468 }
a7f6a5fb 469
b559292e 470 atomic_inc(&wc.wc_num_reqs);
a7f6a5fb
MF		 471 submit_bio(READ, bio);
472 }
473
474 status = 0;
475
476bail_and_wait:
477 o2hb_wait_on_io(reg, &wc);
a9e2ae39
MF		 478 if (wc.wc_error && !status)
479 status = wc.wc_error;
a7f6a5fb 480
a7f6a5fb
MF		 481 return status;
482}
483
484static int o2hb_issue_node_write(struct o2hb_region *reg,
a7f6a5fb
MF		 485 struct o2hb_bio_wait_ctxt *write_wc)
486{
487 int status;
488 unsigned int slot;
489 struct bio *bio;
490
b559292e 491 o2hb_bio_wait_init(write_wc);
a7f6a5fb
MF		 492
493 slot = o2nm_this_node();
494
b559292e 495 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
a7f6a5fb
MF		 496 if (IS_ERR(bio)) {
497 status = PTR_ERR(bio);
498 mlog_errno(status);
499 goto bail;
500 }
501
b559292e 502 atomic_inc(&write_wc->wc_num_reqs);
e873fdb5 503 submit_bio(WRITE_SYNC, bio);
a7f6a5fb 504
a7f6a5fb
MF		 505 status = 0;
506bail:
507 return status;
508}
509
510static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
511 struct o2hb_disk_heartbeat_block *hb_block)
512{
513 __le32 old_cksum;
514 u32 ret;
515
516 /* We want to compute the block crc with a 0 value in the
517 * hb_cksum field. Save it off here and replace after the
518 * crc. */
519 old_cksum = hb_block->hb_cksum;
520 hb_block->hb_cksum = 0;
521
522 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
523
524 hb_block->hb_cksum = old_cksum;
525
526 return ret;
527}
528
529static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
530{
70bacbdb
MF		 531 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
532 "cksum = 0x%x, generation 0x%llx\n",
533 (long long)le64_to_cpu(hb_block->hb_seq),
534 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
535 (long long)le64_to_cpu(hb_block->hb_generation));
a7f6a5fb
MF		 536}
537
538static int o2hb_verify_crc(struct o2hb_region *reg,
539 struct o2hb_disk_heartbeat_block *hb_block)
540{
541 u32 read, computed;
542
543 read = le32_to_cpu(hb_block->hb_cksum);
544 computed = o2hb_compute_block_crc_le(reg, hb_block);
545
546 return read == computed;
547}
548
d2eece37
SM		 549/*
550 * Compare the slot data with what we wrote in the last iteration.
551 * If the match fails, print an appropriate error message. This is to
552 * detect errors like... another node hearting on the same slot,
553 * flaky device that is losing writes, etc.
554 * Returns 1 if check succeeds, 0 otherwise.
555 */
556static int o2hb_check_own_slot(struct o2hb_region *reg)
a7f6a5fb 557{
a7f6a5fb
MF		 558 struct o2hb_disk_slot *slot;
559 struct o2hb_disk_heartbeat_block *hb_block;
33c12a54 560 char *errstr;
a7f6a5fb 561
33c12a54 562 slot = &reg->hr_slots[o2nm_this_node()];
a7f6a5fb 563 /* Don't check on our 1st timestamp */
33c12a54 564 if (!slot->ds_last_time)
d2eece37 565 return 0;
a7f6a5fb 566
33c12a54
SM		 567 hb_block = slot->ds_raw_block;
568 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
569 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
570 hb_block->hb_node == slot->ds_node_num)
d2eece37 571 return 1;
a7f6a5fb 572
33c12a54
SM		 573#define ERRSTR1 "Another node is heartbeating on device"
574#define ERRSTR2 "Heartbeat generation mismatch on device"
575#define ERRSTR3 "Heartbeat sequence mismatch on device"
576
577 if (hb_block->hb_node != slot->ds_node_num)
578 errstr = ERRSTR1;
579 else if (le64_to_cpu(hb_block->hb_generation) !=
580 slot->ds_last_generation)
581 errstr = ERRSTR2;
582 else
583 errstr = ERRSTR3;
584
585 mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
586 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
587 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
588 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
589 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
590 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
d2eece37
SM		 591
592 return 0;
a7f6a5fb
MF		 593}
594
595static inline void o2hb_prepare_block(struct o2hb_region *reg,
596 u64 generation)
597{
598 int node_num;
599 u64 cputime;
600 struct o2hb_disk_slot *slot;
601 struct o2hb_disk_heartbeat_block *hb_block;
602
603 node_num = o2nm_this_node();
604 slot = &reg->hr_slots[node_num];
605
606 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
607 memset(hb_block, 0, reg->hr_block_bytes);
608 /* TODO: time stuff */
609 cputime = CURRENT_TIME.tv_sec;
610 if (!cputime)
611 cputime = 1;
612
613 hb_block->hb_seq = cpu_to_le64(cputime);
614 hb_block->hb_node = node_num;
615 hb_block->hb_generation = cpu_to_le64(generation);
0db638f4 616 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
a7f6a5fb
MF		 617
618 /* This step must always happen last! */
619 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
620 hb_block));
621
70bacbdb 622 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
5fdf1e67 623 (long long)generation,
70bacbdb 624 le32_to_cpu(hb_block->hb_cksum));
a7f6a5fb
MF		 625}
626
627static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
628 struct o2nm_node *node,
629 int idx)
630{
631 struct list_head *iter;
632 struct o2hb_callback_func *f;
633
634 list_for_each(iter, &hbcall->list) {
635 f = list_entry(iter, struct o2hb_callback_func, hc_item);
636 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
637 (f->hc_func)(node, idx, f->hc_data);
638 }
639}
640
641/* Will run the list in order until we process the passed event */
642static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
643{
644 int empty;
645 struct o2hb_callback *hbcall;
646 struct o2hb_node_event *event;
647
648 spin_lock(&o2hb_live_lock);
649 empty = list_empty(&queued_event->hn_item);
650 spin_unlock(&o2hb_live_lock);
651 if (empty)
652 return;
653
654 /* Holding callback sem assures we don't alter the callback
655 * lists when doing this, and serializes ourselves with other
656 * processes wanting callbacks. */
657 down_write(&o2hb_callback_sem);
658
659 spin_lock(&o2hb_live_lock);
660 while (!list_empty(&o2hb_node_events)
661 && !list_empty(&queued_event->hn_item)) {
662 event = list_entry(o2hb_node_events.next,
663 struct o2hb_node_event,
664 hn_item);
665 list_del_init(&event->hn_item);
666 spin_unlock(&o2hb_live_lock);
667
668 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
669 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
670 event->hn_node_num);
671
672 hbcall = hbcall_from_type(event->hn_event_type);
673
674 /* We should *never* have gotten on to the list with a
675 * bad type... This isn't something that we should try
676 * to recover from. */
677 BUG_ON(IS_ERR(hbcall));
678
679 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
680
681 spin_lock(&o2hb_live_lock);
682 }
683 spin_unlock(&o2hb_live_lock);
684
685 up_write(&o2hb_callback_sem);
686}
687
688static void o2hb_queue_node_event(struct o2hb_node_event *event,
689 enum o2hb_callback_type type,
690 struct o2nm_node *node,
691 int node_num)
692{
693 assert_spin_locked(&o2hb_live_lock);
694
0e105d37
SM		 695 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
696
a7f6a5fb
MF		 697 event->hn_event_type = type;
698 event->hn_node = node;
699 event->hn_node_num = node_num;
700
701 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
702 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
703
704 list_add_tail(&event->hn_item, &o2hb_node_events);
705}
706
707static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
708{
709 struct o2hb_node_event event =
710 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
711 struct o2nm_node *node;
712
713 node = o2nm_get_node_by_num(slot->ds_node_num);
714 if (!node)
715 return;
716
717 spin_lock(&o2hb_live_lock);
718 if (!list_empty(&slot->ds_live_item)) {
719 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
720 slot->ds_node_num);
721
722 list_del_init(&slot->ds_live_item);
723
724 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
725 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
726
727 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
728 slot->ds_node_num);
729 }
730 }
731 spin_unlock(&o2hb_live_lock);
732
733 o2hb_run_event_list(&event);
734
735 o2nm_node_put(node);
736}
737
d2eece37 738static void o2hb_set_quorum_device(struct o2hb_region *reg)
43182d2a 739{
43182d2a
SM		 740 if (!o2hb_global_heartbeat_active())
741 return;
742
d2eece37
SM		 743 /* Prevent race with o2hb_heartbeat_group_drop_item() */
744 if (kthread_should_stop())
43182d2a
SM		 745 return;
746
d2eece37
SM		 747 /* Tag region as quorum only after thread reaches steady state */
748 if (atomic_read(&reg->hr_steady_iterations) != 0)
749 return;
750
751 spin_lock(&o2hb_live_lock);
752
753 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
754 goto unlock;
755
43182d2a
SM		 756 /*
757 * A region can be added to the quorum only when it sees all
758 * live nodes heartbeat on it. In other words, the region has been
759 * added to all nodes.
760 */
761 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
762 sizeof(o2hb_live_node_bitmap)))
d2eece37 763 goto unlock;
43182d2a 764
d2eece37
SM		 765 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
766 config_item_name(&reg->hr_item), reg->hr_dev_name);
43182d2a
SM		 767
768 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
58a3158a
SM		 769
770 /*
771 * If global heartbeat active, unpin all regions if the
772 * region count > CUT_OFF
773 */
774 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
775 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
776 o2hb_region_unpin(NULL);
d2eece37
SM		 777unlock:
778 spin_unlock(&o2hb_live_lock);
43182d2a
SM		 779}
780
a7f6a5fb
MF		 781static int o2hb_check_slot(struct o2hb_region *reg,
782 struct o2hb_disk_slot *slot)
783{
784 int changed = 0, gen_changed = 0;
785 struct o2hb_node_event event =
786 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
787 struct o2nm_node *node;
788 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
789 u64 cputime;
0db638f4
MF		 790 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
791 unsigned int slot_dead_ms;
0e105d37 792 int tmp;
a7f6a5fb
MF		 793
794 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
795
0e105d37
SM		 796 /*
797 * If a node is no longer configured but is still in the livemap, we
798 * may need to clear that bit from the livemap.
799 */
a7f6a5fb 800 node = o2nm_get_node_by_num(slot->ds_node_num);
0e105d37
SM		 801 if (!node) {
802 spin_lock(&o2hb_live_lock);
803 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
804 spin_unlock(&o2hb_live_lock);
805 if (!tmp)
806 return 0;
807 }
a7f6a5fb
MF		 808
809 if (!o2hb_verify_crc(reg, hb_block)) {
810 /* all paths from here will drop o2hb_live_lock for
811 * us. */
812 spin_lock(&o2hb_live_lock);
813
814 /* Don't print an error on the console in this case -
815 * a freshly formatted heartbeat area will not have a
816 * crc set on it. */
817 if (list_empty(&slot->ds_live_item))
818 goto out;
819
820 /* The node is live but pushed out a bad crc. We
821 * consider it a transient miss but don't populate any
822 * other values as they may be junk. */
823 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
824 slot->ds_node_num, reg->hr_dev_name);
825 o2hb_dump_slot(hb_block);
826
827 slot->ds_equal_samples++;
828 goto fire_callbacks;
829 }
830
831 /* we don't care if these wrap.. the state transitions below
832 * clear at the right places */
833 cputime = le64_to_cpu(hb_block->hb_seq);
834 if (slot->ds_last_time != cputime)
835 slot->ds_changed_samples++;
836 else
837 slot->ds_equal_samples++;
838 slot->ds_last_time = cputime;
839
840 /* The node changed heartbeat generations. We assume this to
841 * mean it dropped off but came back before we timed out. We
842 * want to consider it down for the time being but don't want
843 * to lose any changed_samples state we might build up to
844 * considering it live again. */
845 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
846 gen_changed = 1;
847 slot->ds_equal_samples = 0;
70bacbdb
MF		 848 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
849 "to 0x%llx)\n", slot->ds_node_num,
850 (long long)slot->ds_last_generation,
851 (long long)le64_to_cpu(hb_block->hb_generation));
a7f6a5fb
MF		 852 }
853
854 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
855
70bacbdb
MF		 856 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
857 "seq %llu last %llu changed %u equal %u\n",
858 slot->ds_node_num, (long long)slot->ds_last_generation,
859 le32_to_cpu(hb_block->hb_cksum),
2bd63216 860 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
70bacbdb 861 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
a7f6a5fb
MF		 862 slot->ds_equal_samples);
863
864 spin_lock(&o2hb_live_lock);
865
866fire_callbacks:
867 /* dead nodes only come to life after some number of
868 * changes at any time during their dead time */
869 if (list_empty(&slot->ds_live_item) &&
870 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
70bacbdb
MF		 871 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
872 slot->ds_node_num, (long long)slot->ds_last_generation);
a7f6a5fb 873
823a637a
SM		 874 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
875
a7f6a5fb
MF		 876 /* first on the list generates a callback */
877 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
d6aa1c7c
SM		 878 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
879 "bitmap\n", slot->ds_node_num);
a7f6a5fb
MF		 880 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
881
882 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
883 slot->ds_node_num);
884
885 changed = 1;
886 }
887
888 list_add_tail(&slot->ds_live_item,
889 &o2hb_live_slots[slot->ds_node_num]);
890
891 slot->ds_equal_samples = 0;
0db638f4
MF		 892
893 /* We want to be sure that all nodes agree on the
894 * number of milliseconds before a node will be
895 * considered dead. The self-fencing timeout is
896 * computed from this value, and a discrepancy might
897 * result in heartbeat calling a node dead when it
898 * hasn't self-fenced yet. */
899 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
900 if (slot_dead_ms && slot_dead_ms != dead_ms) {
901 /* TODO: Perhaps we can fail the region here. */
902 mlog(ML_ERROR, "Node %d on device %s has a dead count "
903 "of %u ms, but our count is %u ms.\n"
904 "Please double check your configuration values "
905 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
906 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
907 dead_ms);
908 }
a7f6a5fb
MF		 909 goto out;
910 }
911
912 /* if the list is dead, we're done.. */
913 if (list_empty(&slot->ds_live_item))
914 goto out;
915
916 /* live nodes only go dead after enough consequtive missed
917 * samples.. reset the missed counter whenever we see
918 * activity */
919 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
920 mlog(ML_HEARTBEAT, "Node %d left my region\n",
921 slot->ds_node_num);
922
823a637a
SM		 923 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
924
a7f6a5fb
MF		 925 /* last off the live_slot generates a callback */
926 list_del_init(&slot->ds_live_item);
927 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
d6aa1c7c
SM		 928 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
929 "nodes bitmap\n", slot->ds_node_num);
a7f6a5fb
MF		 930 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
931
0e105d37
SM		 932 /* node can be null */
933 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
934 node, slot->ds_node_num);
a7f6a5fb
MF		 935
936 changed = 1;
937 }
938
939 /* We don't clear this because the node is still
940 * actually writing new blocks. */
941 if (!gen_changed)
942 slot->ds_changed_samples = 0;
943 goto out;
944 }
945 if (slot->ds_changed_samples) {
946 slot->ds_changed_samples = 0;
947 slot->ds_equal_samples = 0;
948 }
949out:
950 spin_unlock(&o2hb_live_lock);
951
952 o2hb_run_event_list(&event);
953
0e105d37
SM		 954 if (node)
955 o2nm_node_put(node);
a7f6a5fb
MF		 956 return changed;
957}
958
959/* This could be faster if we just implmented a find_last_bit, but I
960 * don't think the circumstances warrant it. */
961static int o2hb_highest_node(unsigned long *nodes,
962 int numbits)
963{
964 int highest, node;
965
966 highest = numbits;
967 node = -1;
968 while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
969 if (node >= numbits)
970 break;
971
972 highest = node;
973 }
974
975 return highest;
976}
977
a9e2ae39 978static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
a7f6a5fb 979{
d2eece37
SM		 980 int i, ret, highest_node;
981 int membership_change = 0, own_slot_ok = 0;
a7f6a5fb 982 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
0e105d37 983 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
a7f6a5fb
MF		 984 struct o2hb_bio_wait_ctxt write_wc;
985
a9e2ae39
MF		 986 ret = o2nm_configured_node_map(configured_nodes,
987 sizeof(configured_nodes));
988 if (ret) {
989 mlog_errno(ret);
d2eece37 990 goto bail;
a9e2ae39 991 }
a7f6a5fb 992
0e105d37
SM		 993 /*
994 * If a node is not configured but is in the livemap, we still need
995 * to read the slot so as to be able to remove it from the livemap.
996 */
997 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
998 i = -1;
999 while ((i = find_next_bit(live_node_bitmap,
1000 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1001 set_bit(i, configured_nodes);
1002 }
1003
a7f6a5fb
MF		 1004 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1005 if (highest_node >= O2NM_MAX_NODES) {
d2eece37
SM		 1006 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1007 ret = -EINVAL;
1008 goto bail;
a7f6a5fb
MF		 1009 }
1010
1011 /* No sense in reading the slots of nodes that don't exist
1012 * yet. Of course, if the node definitions have holes in them
1013 * then we're reading an empty slot anyway... Consider this
1014 * best-effort. */
1015 ret = o2hb_read_slots(reg, highest_node + 1);
1016 if (ret < 0) {
1017 mlog_errno(ret);
d2eece37 1018 goto bail;
a7f6a5fb
MF		 1019 }
1020
1021 /* With an up to date view of the slots, we can check that no
1022 * other node has been improperly configured to heartbeat in
1023 * our slot. */
d2eece37 1024 own_slot_ok = o2hb_check_own_slot(reg);
a7f6a5fb
MF		 1025
1026 /* fill in the proper info for our next heartbeat */
1027 o2hb_prepare_block(reg, reg->hr_generation);
1028
b559292e 1029 ret = o2hb_issue_node_write(reg, &write_wc);
a7f6a5fb
MF		 1030 if (ret < 0) {
1031 mlog_errno(ret);
d2eece37 1032 goto bail;
a7f6a5fb
MF		 1033 }
1034
1035 i = -1;
33c12a54
SM		 1036 while((i = find_next_bit(configured_nodes,
1037 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
d2eece37 1038 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
a7f6a5fb
MF		 1039 }
1040
1041 /*
1042 * We have to be sure we've advertised ourselves on disk
1043 * before we can go to steady state. This ensures that
1044 * people we find in our steady state have seen us.
1045 */
1046 o2hb_wait_on_io(reg, &write_wc);
a9e2ae39
MF		 1047 if (write_wc.wc_error) {
1048 /* Do not re-arm the write timeout on I/O error - we
1049 * can't be sure that the new block ever made it to
1050 * disk */
1051 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1052 write_wc.wc_error, reg->hr_dev_name);
d2eece37
SM		 1053 ret = write_wc.wc_error;
1054 goto bail;
a9e2ae39
MF		 1055 }
1056
d2eece37
SM		 1057 /* Skip disarming the timeout if own slot has stale/bad data */
1058 if (own_slot_ok) {
1059 o2hb_set_quorum_device(reg);
1060 o2hb_arm_write_timeout(reg);
1061 }
a7f6a5fb 1062
d2eece37 1063bail:
a7f6a5fb 1064 /* let the person who launched us know when things are steady */
d2eece37
SM		 1065 if (atomic_read(&reg->hr_steady_iterations) != 0) {
1066 if (!ret && own_slot_ok && !membership_change) {
1067 if (atomic_dec_and_test(&reg->hr_steady_iterations))
1068 wake_up(&o2hb_steady_queue);
1069 }
1070 }
1071
1072 if (atomic_read(&reg->hr_steady_iterations) != 0) {
1073 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1074 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1075 "heartbeart on region %s (%s)\n",
1076 config_item_name(&reg->hr_item),
1077 reg->hr_dev_name);
1078 atomic_set(&reg->hr_steady_iterations, 0);
1079 reg->hr_aborted_start = 1;
a7f6a5fb 1080 wake_up(&o2hb_steady_queue);
d2eece37
SM		 1081 ret = -EIO;
1082 }
a7f6a5fb 1083 }
a9e2ae39 1084
d2eece37 1085 return ret;
a7f6a5fb
MF		 1086}
1087
1088/* Subtract b from a, storing the result in a. a *must* have a larger
1089 * value than b. */
1090static void o2hb_tv_subtract(struct timeval *a,
1091 struct timeval *b)
1092{
1093 /* just return 0 when a is after b */
1094 if (a->tv_sec < b->tv_sec ||
1095 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1096 a->tv_sec = 0;
1097 a->tv_usec = 0;
1098 return;
1099 }
1100
1101 a->tv_sec -= b->tv_sec;
1102 a->tv_usec -= b->tv_usec;
1103 while ( a->tv_usec < 0 ) {
1104 a->tv_sec--;
1105 a->tv_usec += 1000000;
1106 }
1107}
1108
1109static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1110 struct timeval *end)
1111{
1112 struct timeval res = *end;
1113
1114 o2hb_tv_subtract(&res, start);
1115
1116 return res.tv_sec * 1000 + res.tv_usec / 1000;
1117}
1118
1119/*
1120 * we ride the region ref that the region dir holds. before the region
1121 * dir is removed and drops it ref it will wait to tear down this
1122 * thread.
1123 */
1124static int o2hb_thread(void *data)
1125{
1126 int i, ret;
1127 struct o2hb_region *reg = data;
a7f6a5fb
MF		 1128 struct o2hb_bio_wait_ctxt write_wc;
1129 struct timeval before_hb, after_hb;
1130 unsigned int elapsed_msec;
1131
1132 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1133
1134 set_user_nice(current, -20);
1135
cfc069d3
SM		 1136 /* Pin node */
1137 o2nm_depend_this_node();
1138
d2eece37
SM		 1139 while (!kthread_should_stop() &&
1140 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
a7f6a5fb 1141 /* We track the time spent inside
025dfdaf 1142 * o2hb_do_disk_heartbeat so that we avoid more than
a7f6a5fb
MF		 1143 * hr_timeout_ms between disk writes. On busy systems
1144 * this should result in a heartbeat which is less
1145 * likely to time itself out. */
1146 do_gettimeofday(&before_hb);
1147
d2eece37 1148 ret = o2hb_do_disk_heartbeat(reg);
a7f6a5fb
MF		 1149
1150 do_gettimeofday(&after_hb);
1151 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1152
b31d308d
TM		 1153 mlog(ML_HEARTBEAT,
1154 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
215c7f9f
MF		 1155 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1156 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1157 elapsed_msec);
a7f6a5fb 1158
d2eece37
SM		 1159 if (!kthread_should_stop() &&
1160 elapsed_msec < reg->hr_timeout_ms) {
a7f6a5fb
MF		 1161 /* the kthread api has blocked signals for us so no
1162 * need to record the return value. */
1163 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1164 }
1165 }
1166
1167 o2hb_disarm_write_timeout(reg);
1168
1169 /* unclean stop is only used in very bad situation */
1170 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1171 o2hb_shutdown_slot(&reg->hr_slots[i]);
1172
1173 /* Explicit down notification - avoid forcing the other nodes
1174 * to timeout on this region when we could just as easily
1175 * write a clear generation - thus indicating to them that
1176 * this node has left this region.
d2eece37
SM		 1177 */
1178 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1179 o2hb_prepare_block(reg, 0);
1180 ret = o2hb_issue_node_write(reg, &write_wc);
1181 if (ret == 0)
1182 o2hb_wait_on_io(reg, &write_wc);
1183 else
1184 mlog_errno(ret);
a7f6a5fb
MF		 1185 }
1186
cfc069d3
SM		 1187 /* Unpin node */
1188 o2nm_undepend_this_node();
a7f6a5fb 1189
d2eece37 1190 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
a7f6a5fb
MF		 1191
1192 return 0;
1193}
1194
87d3d3f3
SM		 1195#ifdef CONFIG_DEBUG_FS
1196static int o2hb_debug_open(struct inode *inode, struct file *file)
1197{
8ca8b0bb 1198 struct o2hb_debug_buf *db = inode->i_private;
1f285305 1199 struct o2hb_region *reg;
87d3d3f3 1200 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
bb570a5d 1201 unsigned long lts;
87d3d3f3
SM		 1202 char *buf = NULL;
1203 int i = -1;
1204 int out = 0;
1205
8ca8b0bb
SM		 1206 /* max_nodes should be the largest bitmap we pass here */
1207 BUG_ON(sizeof(map) < db->db_size);
1208
87d3d3f3
SM		 1209 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1210 if (!buf)
1211 goto bail;
1212
8ca8b0bb
SM		 1213 switch (db->db_type) {
1214 case O2HB_DB_TYPE_LIVENODES:
a6de0136
SM		 1215 case O2HB_DB_TYPE_LIVEREGIONS:
1216 case O2HB_DB_TYPE_QUORUMREGIONS:
1217 case O2HB_DB_TYPE_FAILEDREGIONS:
8ca8b0bb
SM		 1218 spin_lock(&o2hb_live_lock);
1219 memcpy(map, db->db_data, db->db_size);
1220 spin_unlock(&o2hb_live_lock);
1221 break;
87d3d3f3 1222
1f285305
SM		 1223 case O2HB_DB_TYPE_REGION_LIVENODES:
1224 spin_lock(&o2hb_live_lock);
1225 reg = (struct o2hb_region *)db->db_data;
1226 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1227 spin_unlock(&o2hb_live_lock);
1228 break;
1229
1230 case O2HB_DB_TYPE_REGION_NUMBER:
1231 reg = (struct o2hb_region *)db->db_data;
1232 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1233 reg->hr_region_num);
1234 goto done;
1235
43695d09
SM		 1236 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1237 reg = (struct o2hb_region *)db->db_data;
bb570a5d
SM		 1238 lts = reg->hr_last_timeout_start;
1239 /* If 0, it has never been set before */
1240 if (lts)
1241 lts = jiffies_to_msecs(jiffies - lts);
1242 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
43695d09
SM		 1243 goto done;
1244
cb0586bd
SM		 1245 case O2HB_DB_TYPE_REGION_PINNED:
1246 reg = (struct o2hb_region *)db->db_data;
1247 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1248 !!reg->hr_item_pinned);
1249 goto done;
1250
8ca8b0bb
SM		 1251 default:
1252 goto done;
1253 }
1254
1255 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
87d3d3f3
SM		 1256 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1257 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1258
8ca8b0bb 1259done:
87d3d3f3
SM		 1260 i_size_write(inode, out);
1261
1262 file->private_data = buf;
1263
1264 return 0;
1265bail:
1266 return -ENOMEM;
1267}
1268
1269static int o2hb_debug_release(struct inode *inode, struct file *file)
1270{
1271 kfree(file->private_data);
1272 return 0;
1273}
1274
1275static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1276 size_t nbytes, loff_t *ppos)
1277{
1278 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1279 i_size_read(file->f_mapping->host));
1280}
1281#else
1282static int o2hb_debug_open(struct inode *inode, struct file *file)
1283{
1284 return 0;
1285}
1286static int o2hb_debug_release(struct inode *inode, struct file *file)
1287{
1288 return 0;
1289}
1290static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1291 size_t nbytes, loff_t *ppos)
1292{
1293 return 0;
1294}
1295#endif /* CONFIG_DEBUG_FS */
1296
828c0950 1297static const struct file_operations o2hb_debug_fops = {
87d3d3f3
SM		 1298 .open = o2hb_debug_open,
1299 .release = o2hb_debug_release,
1300 .read = o2hb_debug_read,
1301 .llseek = generic_file_llseek,
1302};
1303
1304void o2hb_exit(void)
1305{
8ca8b0bb 1306 kfree(o2hb_db_livenodes);
a6de0136
SM		 1307 kfree(o2hb_db_liveregions);
1308 kfree(o2hb_db_quorumregions);
1309 kfree(o2hb_db_failedregions);
1310 debugfs_remove(o2hb_debug_failedregions);
1311 debugfs_remove(o2hb_debug_quorumregions);
1312 debugfs_remove(o2hb_debug_liveregions);
8ca8b0bb
SM		 1313 debugfs_remove(o2hb_debug_livenodes);
1314 debugfs_remove(o2hb_debug_dir);
1315}
1316
1317static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1318 struct o2hb_debug_buf **db, int db_len,
1319 int type, int size, int len, void *data)
1320{
1321 *db = kmalloc(db_len, GFP_KERNEL);
1322 if (!*db)
1323 return NULL;
1324
1325 (*db)->db_type = type;
1326 (*db)->db_size = size;
1327 (*db)->db_len = len;
1328 (*db)->db_data = data;
1329
1330 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1331 &o2hb_debug_fops);
1332}
1333
1334static int o2hb_debug_init(void)
1335{
1336 int ret = -ENOMEM;
1337
1338 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1339 if (!o2hb_debug_dir) {
1340 mlog_errno(ret);
1341 goto bail;
1342 }
1343
1344 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1345 o2hb_debug_dir,
1346 &o2hb_db_livenodes,
1347 sizeof(*o2hb_db_livenodes),
1348 O2HB_DB_TYPE_LIVENODES,
1349 sizeof(o2hb_live_node_bitmap),
1350 O2NM_MAX_NODES,
1351 o2hb_live_node_bitmap);
1352 if (!o2hb_debug_livenodes) {
1353 mlog_errno(ret);
1354 goto bail;
1355 }
a6de0136
SM		 1356
1357 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1358 o2hb_debug_dir,
1359 &o2hb_db_liveregions,
1360 sizeof(*o2hb_db_liveregions),
1361 O2HB_DB_TYPE_LIVEREGIONS,
1362 sizeof(o2hb_live_region_bitmap),
1363 O2NM_MAX_REGIONS,
1364 o2hb_live_region_bitmap);
1365 if (!o2hb_debug_liveregions) {
1366 mlog_errno(ret);
1367 goto bail;
1368 }
1369
1370 o2hb_debug_quorumregions =
1371 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1372 o2hb_debug_dir,
1373 &o2hb_db_quorumregions,
1374 sizeof(*o2hb_db_quorumregions),
1375 O2HB_DB_TYPE_QUORUMREGIONS,
1376 sizeof(o2hb_quorum_region_bitmap),
1377 O2NM_MAX_REGIONS,
1378 o2hb_quorum_region_bitmap);
1379 if (!o2hb_debug_quorumregions) {
1380 mlog_errno(ret);
1381 goto bail;
1382 }
1383
1384 o2hb_debug_failedregions =
1385 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1386 o2hb_debug_dir,
1387 &o2hb_db_failedregions,
1388 sizeof(*o2hb_db_failedregions),
1389 O2HB_DB_TYPE_FAILEDREGIONS,
1390 sizeof(o2hb_failed_region_bitmap),
1391 O2NM_MAX_REGIONS,
1392 o2hb_failed_region_bitmap);
1393 if (!o2hb_debug_failedregions) {
1394 mlog_errno(ret);
1395 goto bail;
1396 }
1397
8ca8b0bb
SM		 1398 ret = 0;
1399bail:
1400 if (ret)
1401 o2hb_exit();
1402
1403 return ret;
87d3d3f3
SM		 1404}
1405
1406int o2hb_init(void)
a7f6a5fb
MF		 1407{
1408 int i;
1409
1410 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1411 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1412
1413 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1414 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1415
1416 INIT_LIST_HEAD(&o2hb_node_events);
1417
1418 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
536f0741 1419 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
e7d656ba 1420 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
43182d2a 1421 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
b1c5ebfb 1422 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
87d3d3f3 1423
58a3158a
SM		 1424 o2hb_dependent_users = 0;
1425
8ca8b0bb 1426 return o2hb_debug_init();
a7f6a5fb
MF		 1427}
1428
1429/* if we're already in a callback then we're already serialized by the sem */
1430static void o2hb_fill_node_map_from_callback(unsigned long *map,
1431 unsigned bytes)
1432{
1433 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1434
1435 memcpy(map, &o2hb_live_node_bitmap, bytes);
1436}
1437
1438/*
1439 * get a map of all nodes that are heartbeating in any regions
1440 */
1441void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1442{
1443 /* callers want to serialize this map and callbacks so that they
1444 * can trust that they don't miss nodes coming to the party */
1445 down_read(&o2hb_callback_sem);
1446 spin_lock(&o2hb_live_lock);
1447 o2hb_fill_node_map_from_callback(map, bytes);
1448 spin_unlock(&o2hb_live_lock);
1449 up_read(&o2hb_callback_sem);
1450}
1451EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1452
1453/*
1454 * heartbeat configfs bits. The heartbeat set is a default set under
1455 * the cluster set in nodemanager.c.
1456 */
1457
1458static struct o2hb_region *to_o2hb_region(struct config_item *item)
1459{
1460 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1461}
1462
1463/* drop_item only drops its ref after killing the thread, nothing should
1464 * be using the region anymore. this has to clean up any state that
1465 * attributes might have built up. */
1466static void o2hb_region_release(struct config_item *item)
1467{
1468 int i;
1469 struct page *page;
1470 struct o2hb_region *reg = to_o2hb_region(item);
1471
d2eece37
SM		 1472 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1473
d787ab09 1474 kfree(reg->hr_tmp_block);
a7f6a5fb
MF		 1475
1476 if (reg->hr_slot_data) {
1477 for (i = 0; i < reg->hr_num_pages; i++) {
1478 page = reg->hr_slot_data[i];
1479 if (page)
1480 __free_page(page);
1481 }
1482 kfree(reg->hr_slot_data);
1483 }
1484
1485 if (reg->hr_bdev)
9a1c3542 1486 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb 1487
d787ab09 1488 kfree(reg->hr_slots);
a7f6a5fb 1489
1f285305
SM		 1490 kfree(reg->hr_db_regnum);
1491 kfree(reg->hr_db_livenodes);
1492 debugfs_remove(reg->hr_debug_livenodes);
1493 debugfs_remove(reg->hr_debug_regnum);
d4396eaf 1494 debugfs_remove(reg->hr_debug_elapsed_time);
cb0586bd 1495 debugfs_remove(reg->hr_debug_pinned);
1f285305
SM		 1496 debugfs_remove(reg->hr_debug_dir);
1497
a7f6a5fb
MF		 1498 spin_lock(&o2hb_live_lock);
1499 list_del(&reg->hr_all_item);
1500 spin_unlock(&o2hb_live_lock);
1501
1502 kfree(reg);
1503}
1504
1505static int o2hb_read_block_input(struct o2hb_region *reg,
1506 const char *page,
1507 size_t count,
1508 unsigned long *ret_bytes,
1509 unsigned int *ret_bits)
1510{
1511 unsigned long bytes;
1512 char *p = (char *)page;
1513
1514 bytes = simple_strtoul(p, &p, 0);
1515 if (!p || (*p && (*p != '\n')))
1516 return -EINVAL;
1517
1518 /* Heartbeat and fs min / max block sizes are the same. */
1519 if (bytes > 4096 || bytes < 512)
1520 return -ERANGE;
1521 if (hweight16(bytes) != 1)
1522 return -EINVAL;
1523
1524 if (ret_bytes)
1525 *ret_bytes = bytes;
1526 if (ret_bits)
1527 *ret_bits = ffs(bytes) - 1;
1528
1529 return 0;
1530}
1531
1532static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1533 char *page)
1534{
1535 return sprintf(page, "%u\n", reg->hr_block_bytes);
1536}
1537
1538static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1539 const char *page,
1540 size_t count)
1541{
1542 int status;
1543 unsigned long block_bytes;
1544 unsigned int block_bits;
1545
1546 if (reg->hr_bdev)
1547 return -EINVAL;
1548
1549 status = o2hb_read_block_input(reg, page, count,
1550 &block_bytes, &block_bits);
1551 if (status)
1552 return status;
1553
1554 reg->hr_block_bytes = (unsigned int)block_bytes;
1555 reg->hr_block_bits = block_bits;
1556
1557 return count;
1558}
1559
1560static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1561 char *page)
1562{
1563 return sprintf(page, "%llu\n", reg->hr_start_block);
1564}
1565
1566static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1567 const char *page,
1568 size_t count)
1569{
1570 unsigned long long tmp;
1571 char *p = (char *)page;
1572
1573 if (reg->hr_bdev)
1574 return -EINVAL;
1575
1576 tmp = simple_strtoull(p, &p, 0);
1577 if (!p || (*p && (*p != '\n')))
1578 return -EINVAL;
1579
1580 reg->hr_start_block = tmp;
1581
1582 return count;
1583}
1584
1585static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1586 char *page)
1587{
1588 return sprintf(page, "%d\n", reg->hr_blocks);
1589}
1590
1591static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1592 const char *page,
1593 size_t count)
1594{
1595 unsigned long tmp;
1596 char *p = (char *)page;
1597
1598 if (reg->hr_bdev)
1599 return -EINVAL;
1600
1601 tmp = simple_strtoul(p, &p, 0);
1602 if (!p || (*p && (*p != '\n')))
1603 return -EINVAL;
1604
1605 if (tmp > O2NM_MAX_NODES || tmp == 0)
1606 return -ERANGE;
1607
1608 reg->hr_blocks = (unsigned int)tmp;
1609
1610 return count;
1611}
1612
1613static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1614 char *page)
1615{
1616 unsigned int ret = 0;
1617
1618 if (reg->hr_bdev)
1619 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1620
1621 return ret;
1622}
1623
1624static void o2hb_init_region_params(struct o2hb_region *reg)
1625{
1626 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1627 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1628
1629 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1630 reg->hr_start_block, reg->hr_blocks);
1631 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1632 reg->hr_block_bytes, reg->hr_block_bits);
1633 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1634 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1635}
1636
1637static int o2hb_map_slot_data(struct o2hb_region *reg)
1638{
1639 int i, j;
1640 unsigned int last_slot;
1641 unsigned int spp = reg->hr_slots_per_page;
1642 struct page *page;
1643 char *raw;
1644 struct o2hb_disk_slot *slot;
1645
1646 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1647 if (reg->hr_tmp_block == NULL) {
1648 mlog_errno(-ENOMEM);
1649 return -ENOMEM;
1650 }
1651
1652 reg->hr_slots = kcalloc(reg->hr_blocks,
1653 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1654 if (reg->hr_slots == NULL) {
1655 mlog_errno(-ENOMEM);
1656 return -ENOMEM;
1657 }
1658
1659 for(i = 0; i < reg->hr_blocks; i++) {
1660 slot = &reg->hr_slots[i];
1661 slot->ds_node_num = i;
1662 INIT_LIST_HEAD(&slot->ds_live_item);
1663 slot->ds_raw_block = NULL;
1664 }
1665
1666 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1667 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1668 "at %u blocks per page\n",
1669 reg->hr_num_pages, reg->hr_blocks, spp);
1670
1671 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1672 GFP_KERNEL);
1673 if (!reg->hr_slot_data) {
1674 mlog_errno(-ENOMEM);
1675 return -ENOMEM;
1676 }
1677
1678 for(i = 0; i < reg->hr_num_pages; i++) {
1679 page = alloc_page(GFP_KERNEL);
1680 if (!page) {
1681 mlog_errno(-ENOMEM);
1682 return -ENOMEM;
1683 }
1684
1685 reg->hr_slot_data[i] = page;
1686
1687 last_slot = i * spp;
1688 raw = page_address(page);
1689 for (j = 0;
1690 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1691 j++) {
1692 BUG_ON((j + last_slot) >= reg->hr_blocks);
1693
1694 slot = &reg->hr_slots[j + last_slot];
1695 slot->ds_raw_block =
1696 (struct o2hb_disk_heartbeat_block *) raw;
1697
1698 raw += reg->hr_block_bytes;
1699 }
1700 }
1701
1702 return 0;
1703}
1704
1705/* Read in all the slots available and populate the tracking
1706 * structures so that we can start with a baseline idea of what's
1707 * there. */
1708static int o2hb_populate_slot_data(struct o2hb_region *reg)
1709{
1710 int ret, i;
1711 struct o2hb_disk_slot *slot;
1712 struct o2hb_disk_heartbeat_block *hb_block;
1713
a7f6a5fb
MF		 1714 ret = o2hb_read_slots(reg, reg->hr_blocks);
1715 if (ret) {
1716 mlog_errno(ret);
1717 goto out;
1718 }
1719
1720 /* We only want to get an idea of the values initially in each
1721 * slot, so we do no verification - o2hb_check_slot will
1722 * actually determine if each configured slot is valid and
1723 * whether any values have changed. */
1724 for(i = 0; i < reg->hr_blocks; i++) {
1725 slot = &reg->hr_slots[i];
1726 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1727
1728 /* Only fill the values that o2hb_check_slot uses to
1729 * determine changing slots */
1730 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1731 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1732 }
1733
1734out:
a7f6a5fb
MF		 1735 return ret;
1736}
1737
1738/* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1739static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1740 const char *page,
1741 size_t count)
1742{
e6c352db 1743 struct task_struct *hb_task;
a7f6a5fb
MF		 1744 long fd;
1745 int sectsize;
1746 char *p = (char *)page;
2903ff01
AV		 1747 struct fd f;
1748 struct inode *inode;
a7f6a5fb 1749 ssize_t ret = -EINVAL;
76d9fc29 1750 int live_threshold;
a7f6a5fb
MF		 1751
1752 if (reg->hr_bdev)
1753 goto out;
1754
1755 /* We can't heartbeat without having had our node number
1756 * configured yet. */
1757 if (o2nm_this_node() == O2NM_MAX_NODES)
1758 goto out;
1759
1760 fd = simple_strtol(p, &p, 0);
1761 if (!p || (*p && (*p != '\n')))
1762 goto out;
1763
1764 if (fd < 0 || fd >= INT_MAX)
1765 goto out;
1766
2903ff01
AV		 1767 f = fdget(fd);
1768 if (f.file == NULL)
a7f6a5fb
MF		 1769 goto out;
1770
1771 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1772 reg->hr_block_bytes == 0)
2903ff01 1773 goto out2;
a7f6a5fb 1774
2903ff01 1775 inode = igrab(f.file->f_mapping->host);
a7f6a5fb 1776 if (inode == NULL)
2903ff01 1777 goto out2;
a7f6a5fb
MF		 1778
1779 if (!S_ISBLK(inode->i_mode))
2903ff01 1780 goto out3;
a7f6a5fb 1781
2903ff01 1782 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
e525fd89 1783 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
a7f6a5fb
MF		 1784 if (ret) {
1785 reg->hr_bdev = NULL;
2903ff01 1786 goto out3;
a7f6a5fb
MF		 1787 }
1788 inode = NULL;
1789
1790 bdevname(reg->hr_bdev, reg->hr_dev_name);
1791
e1defc4f 1792 sectsize = bdev_logical_block_size(reg->hr_bdev);
a7f6a5fb
MF		 1793 if (sectsize != reg->hr_block_bytes) {
1794 mlog(ML_ERROR,
1795 "blocksize %u incorrect for device, expected %d",
1796 reg->hr_block_bytes, sectsize);
1797 ret = -EINVAL;
2903ff01 1798 goto out3;
a7f6a5fb
MF		 1799 }
1800
1801 o2hb_init_region_params(reg);
1802
1803 /* Generation of zero is invalid */
1804 do {
1805 get_random_bytes(&reg->hr_generation,
1806 sizeof(reg->hr_generation));
1807 } while (reg->hr_generation == 0);
1808
1809 ret = o2hb_map_slot_data(reg);
1810 if (ret) {
1811 mlog_errno(ret);
2903ff01 1812 goto out3;
a7f6a5fb
MF		 1813 }
1814
1815 ret = o2hb_populate_slot_data(reg);
1816 if (ret) {
1817 mlog_errno(ret);
2903ff01 1818 goto out3;
a7f6a5fb
MF		 1819 }
1820
c4028958 1821 INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
a7f6a5fb
MF		 1822
1823 /*
1824 * A node is considered live after it has beat LIVE_THRESHOLD
1825 * times. We're not steady until we've given them a chance
1826 * _after_ our first read.
76d9fc29
SM		 1827 * The default threshold is bare minimum so as to limit the delay
1828 * during mounts. For global heartbeat, the threshold doubled for the
1829 * first region.
a7f6a5fb 1830 */
76d9fc29
SM		 1831 live_threshold = O2HB_LIVE_THRESHOLD;
1832 if (o2hb_global_heartbeat_active()) {
1833 spin_lock(&o2hb_live_lock);
1834 if (o2hb_pop_count(&o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1835 live_threshold <<= 1;
1836 spin_unlock(&o2hb_live_lock);
1837 }
d2eece37
SM		 1838 ++live_threshold;
1839 atomic_set(&reg->hr_steady_iterations, live_threshold);
1840 /* unsteady_iterations is double the steady_iterations */
1841 atomic_set(&reg->hr_unsteady_iterations, (live_threshold << 1));
a7f6a5fb 1842
e6c352db
JB		 1843 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1844 reg->hr_item.ci_name);
1845 if (IS_ERR(hb_task)) {
1846 ret = PTR_ERR(hb_task);
a7f6a5fb 1847 mlog_errno(ret);
2903ff01 1848 goto out3;
a7f6a5fb
MF		 1849 }
1850
e6c352db
JB		 1851 spin_lock(&o2hb_live_lock);
1852 reg->hr_task = hb_task;
1853 spin_unlock(&o2hb_live_lock);
1854
a7f6a5fb
MF		 1855 ret = wait_event_interruptible(o2hb_steady_queue,
1856 atomic_read(&reg->hr_steady_iterations) == 0);
1857 if (ret) {
d2eece37
SM		 1858 atomic_set(&reg->hr_steady_iterations, 0);
1859 reg->hr_aborted_start = 1;
1860 }
e6c352db 1861
d2eece37
SM		 1862 if (reg->hr_aborted_start) {
1863 ret = -EIO;
2903ff01 1864 goto out3;
a7f6a5fb
MF		 1865 }
1866
e6df3a66
JB		 1867 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1868 spin_lock(&o2hb_live_lock);
1869 hb_task = reg->hr_task;
e7d656ba
SM		 1870 if (o2hb_global_heartbeat_active())
1871 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
e6df3a66
JB		 1872 spin_unlock(&o2hb_live_lock);
1873
1874 if (hb_task)
1875 ret = count;
1876 else
1877 ret = -EIO;
1878
18c50cb0 1879 if (hb_task && o2hb_global_heartbeat_active())
d2eece37
SM		 1880 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1881 config_item_name(&reg->hr_item), reg->hr_dev_name);
18c50cb0 1882
2903ff01
AV		 1883out3:
1884 iput(inode);
1885out2:
1886 fdput(f);
a7f6a5fb 1887out:
a7f6a5fb
MF		 1888 if (ret < 0) {
1889 if (reg->hr_bdev) {
9a1c3542 1890 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
a7f6a5fb
MF		 1891 reg->hr_bdev = NULL;
1892 }
1893 }
1894 return ret;
1895}
1896
92efc152
ZW		 1897static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1898 char *page)
1899{
e6c352db
JB		 1900 pid_t pid = 0;
1901
1902 spin_lock(&o2hb_live_lock);
1903 if (reg->hr_task)
ba25f9dc 1904 pid = task_pid_nr(reg->hr_task);
e6c352db
JB		 1905 spin_unlock(&o2hb_live_lock);
1906
1907 if (!pid)
92efc152
ZW		 1908 return 0;
1909
e6c352db 1910 return sprintf(page, "%u\n", pid);
92efc152
ZW		 1911}
1912
a7f6a5fb
MF		 1913struct o2hb_region_attribute {
1914 struct configfs_attribute attr;
1915 ssize_t (*show)(struct o2hb_region *, char *);
1916 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1917};
1918
1919static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1920 .attr = { .ca_owner = THIS_MODULE,
1921 .ca_name = "block_bytes",
1922 .ca_mode = S_IRUGO | S_IWUSR },
1923 .show = o2hb_region_block_bytes_read,
1924 .store = o2hb_region_block_bytes_write,
1925};
1926
1927static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1928 .attr = { .ca_owner = THIS_MODULE,
1929 .ca_name = "start_block",
1930 .ca_mode = S_IRUGO | S_IWUSR },
1931 .show = o2hb_region_start_block_read,
1932 .store = o2hb_region_start_block_write,
1933};
1934
1935static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1936 .attr = { .ca_owner = THIS_MODULE,
1937 .ca_name = "blocks",
1938 .ca_mode = S_IRUGO | S_IWUSR },
1939 .show = o2hb_region_blocks_read,
1940 .store = o2hb_region_blocks_write,
1941};
1942
1943static struct o2hb_region_attribute o2hb_region_attr_dev = {
1944 .attr = { .ca_owner = THIS_MODULE,
1945 .ca_name = "dev",
1946 .ca_mode = S_IRUGO | S_IWUSR },
1947 .show = o2hb_region_dev_read,
1948 .store = o2hb_region_dev_write,
1949};
1950
92efc152
ZW		 1951static struct o2hb_region_attribute o2hb_region_attr_pid = {
1952 .attr = { .ca_owner = THIS_MODULE,
1953 .ca_name = "pid",
1954 .ca_mode = S_IRUGO | S_IRUSR },
1955 .show = o2hb_region_pid_read,
1956};
1957
a7f6a5fb
MF		 1958static struct configfs_attribute *o2hb_region_attrs[] = {
1959 &o2hb_region_attr_block_bytes.attr,
1960 &o2hb_region_attr_start_block.attr,
1961 &o2hb_region_attr_blocks.attr,
1962 &o2hb_region_attr_dev.attr,
92efc152 1963 &o2hb_region_attr_pid.attr,
a7f6a5fb
MF		 1964 NULL,
1965};
1966
1967static ssize_t o2hb_region_show(struct config_item *item,
1968 struct configfs_attribute *attr,
1969 char *page)
1970{
1971 struct o2hb_region *reg = to_o2hb_region(item);
1972 struct o2hb_region_attribute *o2hb_region_attr =
1973 container_of(attr, struct o2hb_region_attribute, attr);
1974 ssize_t ret = 0;
1975
1976 if (o2hb_region_attr->show)
1977 ret = o2hb_region_attr->show(reg, page);
1978 return ret;
1979}
1980
1981static ssize_t o2hb_region_store(struct config_item *item,
1982 struct configfs_attribute *attr,
1983 const char *page, size_t count)
1984{
1985 struct o2hb_region *reg = to_o2hb_region(item);
1986 struct o2hb_region_attribute *o2hb_region_attr =
1987 container_of(attr, struct o2hb_region_attribute, attr);
1988 ssize_t ret = -EINVAL;
1989
1990 if (o2hb_region_attr->store)
1991 ret = o2hb_region_attr->store(reg, page, count);
1992 return ret;
1993}
1994
1995static struct configfs_item_operations o2hb_region_item_ops = {
1996 .release = o2hb_region_release,
1997 .show_attribute = o2hb_region_show,
1998 .store_attribute = o2hb_region_store,
1999};
2000
2001static struct config_item_type o2hb_region_type = {
2002 .ct_item_ops = &o2hb_region_item_ops,
2003 .ct_attrs = o2hb_region_attrs,
2004 .ct_owner = THIS_MODULE,
2005};
2006
2007/* heartbeat set */
2008
2009struct o2hb_heartbeat_group {
2010 struct config_group hs_group;
2011 /* some stuff? */
2012};
2013
2014static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2015{
2016 return group ?
2017 container_of(group, struct o2hb_heartbeat_group, hs_group)
2018 : NULL;
2019}
2020
1f285305
SM		 2021static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2022{
2023 int ret = -ENOMEM;
2024
2025 reg->hr_debug_dir =
2026 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
2027 if (!reg->hr_debug_dir) {
2028 mlog_errno(ret);
2029 goto bail;
2030 }
2031
2032 reg->hr_debug_livenodes =
2033 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2034 reg->hr_debug_dir,
2035 &(reg->hr_db_livenodes),
2036 sizeof(*(reg->hr_db_livenodes)),
2037 O2HB_DB_TYPE_REGION_LIVENODES,
2038 sizeof(reg->hr_live_node_bitmap),
2039 O2NM_MAX_NODES, reg);
2040 if (!reg->hr_debug_livenodes) {
2041 mlog_errno(ret);
2042 goto bail;
2043 }
2044
2045 reg->hr_debug_regnum =
2046 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2047 reg->hr_debug_dir,
2048 &(reg->hr_db_regnum),
2049 sizeof(*(reg->hr_db_regnum)),
2050 O2HB_DB_TYPE_REGION_NUMBER,
2051 0, O2NM_MAX_NODES, reg);
2052 if (!reg->hr_debug_regnum) {
2053 mlog_errno(ret);
2054 goto bail;
2055 }
2056
43695d09
SM		 2057 reg->hr_debug_elapsed_time =
2058 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2059 reg->hr_debug_dir,
2060 &(reg->hr_db_elapsed_time),
2061 sizeof(*(reg->hr_db_elapsed_time)),
2062 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2063 0, 0, reg);
2064 if (!reg->hr_debug_elapsed_time) {
2065 mlog_errno(ret);
2066 goto bail;
2067 }
2068
cb0586bd
SM		 2069 reg->hr_debug_pinned =
2070 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2071 reg->hr_debug_dir,
2072 &(reg->hr_db_pinned),
2073 sizeof(*(reg->hr_db_pinned)),
2074 O2HB_DB_TYPE_REGION_PINNED,
2075 0, 0, reg);
2076 if (!reg->hr_debug_pinned) {
2077 mlog_errno(ret);
2078 goto bail;
2079 }
2080
1f285305
SM		 2081 ret = 0;
2082bail:
2083 return ret;
2084}
2085
f89ab861
JB		 2086static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2087 const char *name)
a7f6a5fb
MF		 2088{
2089 struct o2hb_region *reg = NULL;
1f285305 2090 int ret;
a7f6a5fb 2091
cd861280 2092 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
f89ab861 2093 if (reg == NULL)
a6795e9e 2094 return ERR_PTR(-ENOMEM);
a7f6a5fb 2095
1cf257f5
JS		 2096 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2097 ret = -ENAMETOOLONG;
2098 goto free;
2099 }
b3c85c4c 2100
a7f6a5fb 2101 spin_lock(&o2hb_live_lock);
536f0741
SM		 2102 reg->hr_region_num = 0;
2103 if (o2hb_global_heartbeat_active()) {
2104 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2105 O2NM_MAX_REGIONS);
2106 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2107 spin_unlock(&o2hb_live_lock);
1cf257f5
JS		 2108 ret = -EFBIG;
2109 goto free;
536f0741
SM		 2110 }
2111 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2112 }
a7f6a5fb
MF		 2113 list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2114 spin_unlock(&o2hb_live_lock);
a7f6a5fb 2115
536f0741
SM		 2116 config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2117
1f285305
SM		 2118 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2119 if (ret) {
2120 config_item_put(&reg->hr_item);
1cf257f5 2121 goto free;
1f285305
SM		 2122 }
2123
a6795e9e 2124 return &reg->hr_item;
1cf257f5
JS		 2125free:
2126 kfree(reg);
2127 return ERR_PTR(ret);
a7f6a5fb
MF		 2128}
2129
2130static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2131 struct config_item *item)
2132{
e6c352db 2133 struct task_struct *hb_task;
a7f6a5fb 2134 struct o2hb_region *reg = to_o2hb_region(item);
58a3158a 2135 int quorum_region = 0;
a7f6a5fb
MF		 2136
2137 /* stop the thread when the user removes the region dir */
e6c352db
JB		 2138 spin_lock(&o2hb_live_lock);
2139 hb_task = reg->hr_task;
2140 reg->hr_task = NULL;
58a3158a 2141 reg->hr_item_dropped = 1;
e6c352db
JB		 2142 spin_unlock(&o2hb_live_lock);
2143
2144 if (hb_task)
2145 kthread_stop(hb_task);
a7f6a5fb 2146
d2eece37
SM		 2147 if (o2hb_global_heartbeat_active()) {
2148 spin_lock(&o2hb_live_lock);
2149 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2150 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2151 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2152 quorum_region = 1;
2153 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2154 spin_unlock(&o2hb_live_lock);
2155 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2156 ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2157 "stopped" : "start aborted"), config_item_name(item),
2158 reg->hr_dev_name);
2159 }
2160
e6df3a66
JB		 2161 /*
2162 * If we're racing a dev_write(), we need to wake them. They will
2163 * check reg->hr_task
2164 */
2165 if (atomic_read(&reg->hr_steady_iterations) != 0) {
d2eece37 2166 reg->hr_aborted_start = 1;
e6df3a66
JB		 2167 atomic_set(&reg->hr_steady_iterations, 0);
2168 wake_up(&o2hb_steady_queue);
2169 }
2170
a7f6a5fb 2171 config_item_put(item);
58a3158a
SM		 2172
2173 if (!o2hb_global_heartbeat_active() || !quorum_region)
2174 return;
2175
2176 /*
2177 * If global heartbeat active and there are dependent users,
2178 * pin all regions if quorum region count <= CUT_OFF
2179 */
2180 spin_lock(&o2hb_live_lock);
2181
2182 if (!o2hb_dependent_users)
2183 goto unlock;
2184
2185 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2186 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2187 o2hb_region_pin(NULL);
2188
2189unlock:
2190 spin_unlock(&o2hb_live_lock);
a7f6a5fb
MF		 2191}
2192
2193struct o2hb_heartbeat_group_attribute {
2194 struct configfs_attribute attr;
2195 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2196 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2197};
2198
2199static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2200 struct configfs_attribute *attr,
2201 char *page)
2202{
2203 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2204 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2205 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2206 ssize_t ret = 0;
2207
2208 if (o2hb_heartbeat_group_attr->show)
2209 ret = o2hb_heartbeat_group_attr->show(reg, page);
2210 return ret;
2211}
2212
2213static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2214 struct configfs_attribute *attr,
2215 const char *page, size_t count)
2216{
2217 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2218 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2219 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2220 ssize_t ret = -EINVAL;
2221
2222 if (o2hb_heartbeat_group_attr->store)
2223 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2224 return ret;
2225}
2226
2227static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2228 char *page)
2229{
2230 return sprintf(page, "%u\n", o2hb_dead_threshold);
2231}
2232
2233static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2234 const char *page,
2235 size_t count)
2236{
2237 unsigned long tmp;
2238 char *p = (char *)page;
2239
2240 tmp = simple_strtoul(p, &p, 10);
2241 if (!p || (*p && (*p != '\n')))
2242 return -EINVAL;
2243
2244 /* this will validate ranges for us. */
2245 o2hb_dead_threshold_set((unsigned int) tmp);
2246
2247 return count;
2248}
2249
54b5187b
SM		 2250static
2251ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2252 char *page)
2253{
2254 return sprintf(page, "%s\n",
2255 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2256}
2257
2258static
2259ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2260 const char *page, size_t count)
2261{
2262 unsigned int i;
2263 int ret;
2264 size_t len;
2265
2266 len = (page[count - 1] == '\n') ? count - 1 : count;
2267 if (!len)
2268 return -EINVAL;
2269
2270 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2271 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2272 continue;
2273
70f651ed 2274 ret = o2hb_global_heartbeat_mode_set(i);
54b5187b 2275 if (!ret)
18c50cb0 2276 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
54b5187b
SM		 2277 o2hb_heartbeat_mode_desc[i]);
2278 return count;
2279 }
2280
2281 return -EINVAL;
2282
2283}
2284
a7f6a5fb
MF		 2285static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2286 .attr = { .ca_owner = THIS_MODULE,
2287 .ca_name = "dead_threshold",
2288 .ca_mode = S_IRUGO | S_IWUSR },
2289 .show = o2hb_heartbeat_group_threshold_show,
2290 .store = o2hb_heartbeat_group_threshold_store,
2291};
2292
54b5187b
SM		 2293static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2294 .attr = { .ca_owner = THIS_MODULE,
2295 .ca_name = "mode",
2296 .ca_mode = S_IRUGO | S_IWUSR },
2297 .show = o2hb_heartbeat_group_mode_show,
2298 .store = o2hb_heartbeat_group_mode_store,
2299};
2300
a7f6a5fb
MF		 2301static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2302 &o2hb_heartbeat_group_attr_threshold.attr,
54b5187b 2303 &o2hb_heartbeat_group_attr_mode.attr,
a7f6a5fb
MF		 2304 NULL,
2305};
2306
70f651ed 2307static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
a7f6a5fb
MF		 2308 .show_attribute = o2hb_heartbeat_group_show,
2309 .store_attribute = o2hb_heartbeat_group_store,
2310};
2311
2312static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2313 .make_item = o2hb_heartbeat_group_make_item,
2314 .drop_item = o2hb_heartbeat_group_drop_item,
2315};
2316
2317static struct config_item_type o2hb_heartbeat_group_type = {
2318 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
70f651ed 2319 .ct_item_ops = &o2hb_heartbeat_group_item_ops,
a7f6a5fb
MF		 2320 .ct_attrs = o2hb_heartbeat_group_attrs,
2321 .ct_owner = THIS_MODULE,
2322};
2323
2324/* this is just here to avoid touching group in heartbeat.h which the
2325 * entire damn world #includes */
2326struct config_group *o2hb_alloc_hb_set(void)
2327{
2328 struct o2hb_heartbeat_group *hs = NULL;
2329 struct config_group *ret = NULL;
2330
cd861280 2331 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
a7f6a5fb
MF		 2332 if (hs == NULL)
2333 goto out;
2334
2335 config_group_init_type_name(&hs->hs_group, "heartbeat",
2336 &o2hb_heartbeat_group_type);
2337
2338 ret = &hs->hs_group;
2339out:
2340 if (ret == NULL)
2341 kfree(hs);
2342 return ret;
2343}
2344
2345void o2hb_free_hb_set(struct config_group *group)
2346{
2347 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2348 kfree(hs);
2349}
2350
25985edc 2351/* hb callback registration and issuing */
a7f6a5fb
MF		 2352
2353static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2354{
2355 if (type == O2HB_NUM_CB)
2356 return ERR_PTR(-EINVAL);
2357
2358 return &o2hb_callbacks[type];
2359}
2360
2361void o2hb_setup_callback(struct o2hb_callback_func *hc,
2362 enum o2hb_callback_type type,
2363 o2hb_cb_func *func,
2364 void *data,
2365 int priority)
2366{
2367 INIT_LIST_HEAD(&hc->hc_item);
2368 hc->hc_func = func;
2369 hc->hc_data = data;
2370 hc->hc_priority = priority;
2371 hc->hc_type = type;
2372 hc->hc_magic = O2HB_CB_MAGIC;
2373}
2374EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2375
58a3158a
SM		 2376/*
2377 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2378 * In global heartbeat mode, region_uuid passed is NULL.
2379 *
2380 * In local, we only pin the matching region. In global we pin all the active
2381 * regions.
2382 */
2383static int o2hb_region_pin(const char *region_uuid)
14829422 2384{
58a3158a
SM		 2385 int ret = 0, found = 0;
2386 struct o2hb_region *reg;
2387 char *uuid;
14829422
JB		 2388
2389 assert_spin_locked(&o2hb_live_lock);
2390
58a3158a
SM		 2391 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2392 uuid = config_item_name(&reg->hr_item);
2393
2394 /* local heartbeat */
2395 if (region_uuid) {
2396 if (strcmp(region_uuid, uuid))
2397 continue;
2398 found = 1;
14829422 2399 }
58a3158a
SM		 2400
2401 if (reg->hr_item_pinned || reg->hr_item_dropped)
2402 goto skip_pin;
2403
2404 /* Ignore ENOENT only for local hb (userdlm domain) */
2405 ret = o2nm_depend_item(&reg->hr_item);
2406 if (!ret) {
2407 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2408 reg->hr_item_pinned = 1;
2409 } else {
2410 if (ret == -ENOENT && found)
2411 ret = 0;
2412 else {
2413 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2414 uuid, ret);
2415 break;
2416 }
14829422 2417 }
58a3158a
SM		 2418skip_pin:
2419 if (found)
2420 break;
14829422
JB		 2421 }
2422
58a3158a 2423 return ret;
14829422
JB		 2424}
2425
58a3158a
SM		 2426/*
2427 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2428 * In global heartbeat mode, region_uuid passed is NULL.
2429 *
2430 * In local, we only unpin the matching region. In global we unpin all the
2431 * active regions.
2432 */
2433static void o2hb_region_unpin(const char *region_uuid)
14829422 2434{
14829422 2435 struct o2hb_region *reg;
58a3158a
SM		 2436 char *uuid;
2437 int found = 0;
14829422 2438
58a3158a 2439 assert_spin_locked(&o2hb_live_lock);
14829422 2440
58a3158a
SM		 2441 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2442 uuid = config_item_name(&reg->hr_item);
2443 if (region_uuid) {
2444 if (strcmp(region_uuid, uuid))
2445 continue;
2446 found = 1;
2447 }
14829422 2448
58a3158a
SM		 2449 if (reg->hr_item_pinned) {
2450 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2451 o2nm_undepend_item(&reg->hr_item);
2452 reg->hr_item_pinned = 0;
2453 }
2454 if (found)
2455 break;
2456 }
2457}
16c6a4f2 2458
58a3158a
SM		 2459static int o2hb_region_inc_user(const char *region_uuid)
2460{
2461 int ret = 0;
14829422 2462
58a3158a 2463 spin_lock(&o2hb_live_lock);
16c6a4f2 2464
58a3158a
SM		 2465 /* local heartbeat */
2466 if (!o2hb_global_heartbeat_active()) {
2467 ret = o2hb_region_pin(region_uuid);
2468 goto unlock;
2469 }
2470
2471 /*
2472 * if global heartbeat active and this is the first dependent user,
2473 * pin all regions if quorum region count <= CUT_OFF
2474 */
2475 o2hb_dependent_users++;
2476 if (o2hb_dependent_users > 1)
2477 goto unlock;
2478
2479 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2480 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2481 ret = o2hb_region_pin(NULL);
2482
2483unlock:
2484 spin_unlock(&o2hb_live_lock);
14829422
JB		 2485 return ret;
2486}
2487
58a3158a 2488void o2hb_region_dec_user(const char *region_uuid)
14829422 2489{
14829422
JB		 2490 spin_lock(&o2hb_live_lock);
2491
58a3158a
SM		 2492 /* local heartbeat */
2493 if (!o2hb_global_heartbeat_active()) {
2494 o2hb_region_unpin(region_uuid);
2495 goto unlock;
2496 }
14829422 2497
58a3158a
SM		 2498 /*
2499 * if global heartbeat active and there are no dependent users,
2500 * unpin all quorum regions
2501 */
2502 o2hb_dependent_users--;
2503 if (!o2hb_dependent_users)
2504 o2hb_region_unpin(NULL);
14829422 2505
58a3158a
SM		 2506unlock:
2507 spin_unlock(&o2hb_live_lock);
14829422
JB		 2508}
2509
2510int o2hb_register_callback(const char *region_uuid,
2511 struct o2hb_callback_func *hc)
a7f6a5fb
MF		 2512{
2513 struct o2hb_callback_func *tmp;
2514 struct list_head *iter;
2515 struct o2hb_callback *hbcall;
2516 int ret;
2517
2518 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2519 BUG_ON(!list_empty(&hc->hc_item));
2520
2521 hbcall = hbcall_from_type(hc->hc_type);
2522 if (IS_ERR(hbcall)) {
2523 ret = PTR_ERR(hbcall);
2524 goto out;
2525 }
2526
14829422 2527 if (region_uuid) {
58a3158a
SM		 2528 ret = o2hb_region_inc_user(region_uuid);
2529 if (ret) {
2530 mlog_errno(ret);
14829422 2531 goto out;
58a3158a 2532 }
14829422
JB		 2533 }
2534
a7f6a5fb
MF		 2535 down_write(&o2hb_callback_sem);
2536
2537 list_for_each(iter, &hbcall->list) {
2538 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2539 if (hc->hc_priority < tmp->hc_priority) {
2540 list_add_tail(&hc->hc_item, iter);
2541 break;
2542 }
2543 }
2544 if (list_empty(&hc->hc_item))
2545 list_add_tail(&hc->hc_item, &hbcall->list);
2546
2547 up_write(&o2hb_callback_sem);
2548 ret = 0;
2549out:
58a3158a 2550 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2551 ret, __builtin_return_address(0), hc);
2552 return ret;
2553}
2554EXPORT_SYMBOL_GPL(o2hb_register_callback);
2555
14829422
JB		 2556void o2hb_unregister_callback(const char *region_uuid,
2557 struct o2hb_callback_func *hc)
a7f6a5fb
MF		 2558{
2559 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2560
58a3158a 2561 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
a7f6a5fb
MF		 2562 __builtin_return_address(0), hc);
2563
14829422 2564 /* XXX Can this happen _with_ a region reference? */
a7f6a5fb 2565 if (list_empty(&hc->hc_item))
c24f72cc 2566 return;
a7f6a5fb 2567
14829422 2568 if (region_uuid)
58a3158a 2569 o2hb_region_dec_user(region_uuid);
14829422 2570
a7f6a5fb
MF		 2571 down_write(&o2hb_callback_sem);
2572
2573 list_del_init(&hc->hc_item);
2574
2575 up_write(&o2hb_callback_sem);
a7f6a5fb
MF		 2576}
2577EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2578
2579int o2hb_check_node_heartbeating(u8 node_num)
2580{
2581 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2582
2583 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2584 if (!test_bit(node_num, testing_map)) {
2585 mlog(ML_HEARTBEAT,
2586 "node (%u) does not have heartbeating enabled.\n",
2587 node_num);
2588 return 0;
2589 }
2590
2591 return 1;
2592}
2593EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2594
2595int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2596{
2597 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2598
2599 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2600 if (!test_bit(node_num, testing_map)) {
2601 mlog(ML_HEARTBEAT,
2602 "node (%u) does not have heartbeating enabled.\n",
2603 node_num);
2604 return 0;
2605 }
2606
2607 return 1;
2608}
2609EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2610
2611/* Makes sure our local node is configured with a node number, and is
2612 * heartbeating. */
2613int o2hb_check_local_node_heartbeating(void)
2614{
2615 u8 node_num;
2616
2617 /* if this node was set then we have networking */
2618 node_num = o2nm_this_node();
2619 if (node_num == O2NM_MAX_NODES) {
2620 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2621 return 0;
2622 }
2623
2624 return o2hb_check_node_heartbeating(node_num);
2625}
2626EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2627
2628/*
2629 * this is just a hack until we get the plumbing which flips file systems
2630 * read only and drops the hb ref instead of killing the node dead.
2631 */
2632void o2hb_stop_all_regions(void)
2633{
2634 struct o2hb_region *reg;
2635
2636 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2637
2638 spin_lock(&o2hb_live_lock);
2639
2640 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2641 reg->hr_unclean_stop = 1;
2642
2643 spin_unlock(&o2hb_live_lock);
2644}
2645EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
b3c85c4c
SM		 2646
2647int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2648{
2649 struct o2hb_region *reg;
2650 int numregs = 0;
2651 char *p;
2652
2653 spin_lock(&o2hb_live_lock);
2654
2655 p = region_uuids;
2656 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2657 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2658 if (numregs < max_regions) {
2659 memcpy(p, config_item_name(&reg->hr_item),
2660 O2HB_MAX_REGION_NAME_LEN);
2661 p += O2HB_MAX_REGION_NAME_LEN;
2662 }
2663 numregs++;
2664 }
2665
2666 spin_unlock(&o2hb_live_lock);
2667
2668 return numregs;
2669}
2670EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2671
2672int o2hb_global_heartbeat_active(void)
2673{
4d94aa1b 2674 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
b3c85c4c
SM		 2675}
2676EXPORT_SYMBOL(o2hb_global_heartbeat_active);
Linux 6.x block layer and io_uring tree(s)